4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 1991, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 1990 Mentat Inc.
24 * Copyright (c) 2013 by Delphix. All rights reserved.
25 * Copyright (c) 2016, Joyent, Inc. All rights reserved.
26 * Copyright (c) 2014, OmniTI Computer Consulting, Inc. All rights reserved.
30 * This file contains the interface control functions for IP.
33 #include <sys/types.h>
34 #include <sys/stream.h>
36 #include <sys/stropts.h>
37 #include <sys/strsun.h>
38 #include <sys/sysmacros.h>
39 #include <sys/strsubr.h>
40 #include <sys/strlog.h>
42 #include <sys/sunddi.h>
43 #include <sys/cmn_err.h>
44 #include <sys/kstat.h>
45 #include <sys/debug.h>
47 #include <sys/sunldi.h>
49 #include <sys/bitmap.h>
50 #include <sys/cpuvar.h>
52 #include <sys/ctype.h>
54 #include <sys/systm.h>
55 #include <sys/param.h>
56 #include <sys/socket.h>
57 #include <sys/isa_defs.h>
59 #include <net/if_arp.h>
60 #include <net/if_types.h>
61 #include <net/if_dl.h>
62 #include <net/route.h>
63 #include <sys/sockio.h>
64 #include <netinet/in.h>
65 #include <netinet/ip6.h>
66 #include <netinet/icmp6.h>
67 #include <netinet/igmp_var.h>
68 #include <sys/policy.h>
69 #include <sys/ethernet.h>
70 #include <sys/callb.h>
73 #include <inet/common.h> /* for various inet/mi.h and inet/nd.h needs */
76 #include <inet/tunables.h>
78 #include <inet/ip_arp.h>
79 #include <inet/mib2.h>
82 #include <inet/ip6_asp.h>
84 #include <inet/ip_multi.h>
85 #include <inet/ip_ire.h>
86 #include <inet/ip_ftable.h>
87 #include <inet/ip_rts.h>
88 #include <inet/ip_ndp.h>
89 #include <inet/ip_if.h>
90 #include <inet/ip_impl.h>
91 #include <inet/sctp_ip.h>
92 #include <inet/ip_netinfo.h>
93 #include <inet/ilb_ip.h>
95 #include <netinet/igmp.h>
96 #include <inet/ip_listutils.h>
97 #include <inet/ipclassifier.h>
98 #include <sys/mac_client.h>
100 #include <sys/mac_flow.h>
102 #include <sys/systeminfo.h>
103 #include <sys/bootconf.h>
105 #include <inet/rawip_impl.h> /* needed for icmp_stack_t */
106 #include <inet/udp_impl.h> /* needed for udp_stack_t */
108 /* The character which tells where the ill_name ends */
109 #define IPIF_SEPARATOR_CHAR ':'
111 /* IP ioctl function table entry */
112 typedef struct ipft_s
{
118 #define IPFT_F_NO_REPLY 0x1 /* IP ioctl does not expect any reply */
119 #define IPFT_F_SELF_REPLY 0x2 /* ioctl callee does the ioctl reply */
121 static int nd_ill_forward_get(queue_t
*, mblk_t
*, caddr_t
, cred_t
*);
122 static int nd_ill_forward_set(queue_t
*q
, mblk_t
*mp
,
123 char *value
, caddr_t cp
, cred_t
*ioc_cr
);
125 static boolean_t
ill_is_quiescent(ill_t
*);
126 static boolean_t
ip_addr_ok_v4(ipaddr_t addr
, ipaddr_t subnet_mask
);
127 static ip_m_t
*ip_m_lookup(t_uscalar_t mac_type
);
128 static int ip_sioctl_addr_tail(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
,
129 mblk_t
*mp
, boolean_t need_up
);
130 static int ip_sioctl_dstaddr_tail(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
,
131 mblk_t
*mp
, boolean_t need_up
);
132 static int ip_sioctl_slifzone_tail(ipif_t
*ipif
, zoneid_t zoneid
,
133 queue_t
*q
, mblk_t
*mp
, boolean_t need_up
);
134 static int ip_sioctl_flags_tail(ipif_t
*ipif
, uint64_t flags
, queue_t
*q
,
136 static int ip_sioctl_netmask_tail(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
,
138 static int ip_sioctl_subnet_tail(ipif_t
*ipif
, in6_addr_t
, in6_addr_t
,
139 queue_t
*q
, mblk_t
*mp
, boolean_t need_up
);
140 static int ip_sioctl_plink_ipmod(ipsq_t
*ipsq
, queue_t
*q
, mblk_t
*mp
,
141 int ioccmd
, struct linkblk
*li
);
142 static ipaddr_t
ip_subnet_mask(ipaddr_t addr
, ipif_t
**, ip_stack_t
*);
143 static void ip_wput_ioctl(queue_t
*q
, mblk_t
*mp
);
144 static void ipsq_flush(ill_t
*ill
);
146 static int ip_sioctl_token_tail(ipif_t
*ipif
, sin6_t
*sin6
, int addrlen
,
147 queue_t
*q
, mblk_t
*mp
, boolean_t need_up
);
148 static void ipsq_delete(ipsq_t
*);
150 static ipif_t
*ipif_allocate(ill_t
*ill
, int id
, uint_t ire_type
,
151 boolean_t initialize
, boolean_t insert
, int *errorp
);
152 static ire_t
**ipif_create_bcast_ires(ipif_t
*ipif
, ire_t
**irep
);
153 static void ipif_delete_bcast_ires(ipif_t
*ipif
);
154 static int ipif_add_ires_v4(ipif_t
*, boolean_t
);
155 static boolean_t
ipif_comp_multi(ipif_t
*old_ipif
, ipif_t
*new_ipif
,
157 static int ipif_logical_down(ipif_t
*ipif
, queue_t
*q
, mblk_t
*mp
);
158 static void ipif_free(ipif_t
*ipif
);
159 static void ipif_free_tail(ipif_t
*ipif
);
160 static void ipif_set_default(ipif_t
*ipif
);
161 static int ipif_set_values(queue_t
*q
, mblk_t
*mp
,
162 char *interf_name
, uint_t
*ppa
);
163 static int ipif_set_values_tail(ill_t
*ill
, ipif_t
*ipif
, mblk_t
*mp
,
165 static ipif_t
*ipif_lookup_on_name(char *name
, size_t namelen
,
166 boolean_t do_alloc
, boolean_t
*exists
, boolean_t isv6
, zoneid_t zoneid
,
168 static ipif_t
*ipif_lookup_on_name_async(char *name
, size_t namelen
,
169 boolean_t isv6
, zoneid_t zoneid
, queue_t
*q
, mblk_t
*mp
, ipsq_func_t func
,
170 int *error
, ip_stack_t
*);
172 static int ill_alloc_ppa(ill_if_t
*, ill_t
*);
173 static void ill_delete_interface_type(ill_if_t
*);
174 static int ill_dl_up(ill_t
*ill
, ipif_t
*ipif
, mblk_t
*mp
, queue_t
*q
);
175 static void ill_dl_down(ill_t
*ill
);
176 static void ill_down(ill_t
*ill
);
177 static void ill_down_ipifs(ill_t
*, boolean_t
);
178 static void ill_free_mib(ill_t
*ill
);
179 static void ill_glist_delete(ill_t
*);
180 static void ill_phyint_reinit(ill_t
*ill
);
181 static void ill_set_nce_router_flags(ill_t
*, boolean_t
);
182 static void ill_set_phys_addr_tail(ipsq_t
*, queue_t
*, mblk_t
*, void *);
183 static void ill_replumb_tail(ipsq_t
*, queue_t
*, mblk_t
*, void *);
185 static ip_v6intfid_func_t ip_ether_v6intfid
, ip_ib_v6intfid
;
186 static ip_v6intfid_func_t ip_ipv4_v6intfid
, ip_ipv6_v6intfid
;
187 static ip_v6intfid_func_t ip_ipmp_v6intfid
, ip_nodef_v6intfid
;
188 static ip_v6intfid_func_t ip_ipv4_v6destintfid
, ip_ipv6_v6destintfid
;
189 static ip_v4mapinfo_func_t ip_ether_v4_mapping
;
190 static ip_v6mapinfo_func_t ip_ether_v6_mapping
;
191 static ip_v4mapinfo_func_t ip_ib_v4_mapping
;
192 static ip_v6mapinfo_func_t ip_ib_v6_mapping
;
193 static ip_v4mapinfo_func_t ip_mbcast_mapping
;
194 static void phyint_free(phyint_t
*);
196 static void ill_capability_dispatch(ill_t
*, mblk_t
*, dl_capability_sub_t
*);
197 static void ill_capability_id_ack(ill_t
*, mblk_t
*, dl_capability_sub_t
*);
198 static void ill_capability_vrrp_ack(ill_t
*, mblk_t
*, dl_capability_sub_t
*);
199 static void ill_capability_hcksum_ack(ill_t
*, mblk_t
*, dl_capability_sub_t
*);
200 static void ill_capability_hcksum_reset_fill(ill_t
*, mblk_t
*);
201 static void ill_capability_zerocopy_ack(ill_t
*, mblk_t
*,
202 dl_capability_sub_t
*);
203 static void ill_capability_zerocopy_reset_fill(ill_t
*, mblk_t
*);
204 static void ill_capability_dld_reset_fill(ill_t
*, mblk_t
*);
205 static void ill_capability_dld_ack(ill_t
*, mblk_t
*,
206 dl_capability_sub_t
*);
207 static void ill_capability_dld_enable(ill_t
*);
208 static void ill_capability_ack_thr(void *);
209 static void ill_capability_lso_enable(ill_t
*);
211 static ill_t
*ill_prev_usesrc(ill_t
*);
212 static int ill_relink_usesrc_ills(ill_t
*, ill_t
*, uint_t
);
213 static void ill_disband_usesrc_group(ill_t
*);
214 static void ip_sioctl_garp_reply(mblk_t
*, ill_t
*, void *, int);
217 static void ill_trace_cleanup(const ill_t
*);
218 static void ipif_trace_cleanup(const ipif_t
*);
221 static void ill_dlpi_clear_deferred(ill_t
*ill
);
223 static void phyint_flags_init(phyint_t
*, t_uscalar_t
);
226 * if we go over the memory footprint limit more than once in this msec
227 * interval, we'll start pruning aggressively.
229 int ip_min_frag_prune_time
= 0;
231 static ipft_t ip_ioctl_ftbl
[] = {
232 { IP_IOC_IRE_DELETE
, ip_ire_delete
, sizeof (ipid_t
), 0 },
233 { IP_IOC_IRE_DELETE_NO_REPLY
, ip_ire_delete
, sizeof (ipid_t
),
235 { IP_IOC_RTS_REQUEST
, ip_rts_request
, 0, IPFT_F_SELF_REPLY
},
239 /* Simple ICMP IP Header Template */
240 static ipha_t icmp_ipha
= {
241 IP_SIMPLE_HDR_VERSION
, 0, 0, 0, 0, 0, IPPROTO_ICMP
244 static uchar_t ip_six_byte_all_ones
[] = { 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF };
246 static ip_m_t ip_m_tbl
[] = {
247 { DL_ETHER
, IFT_ETHER
, ETHERTYPE_IP
, ETHERTYPE_IPV6
,
248 ip_ether_v4_mapping
, ip_ether_v6_mapping
, ip_ether_v6intfid
,
250 { DL_CSMACD
, IFT_ISO88023
, ETHERTYPE_IP
, ETHERTYPE_IPV6
,
251 ip_ether_v4_mapping
, ip_ether_v6_mapping
, ip_nodef_v6intfid
,
253 { DL_TPB
, IFT_ISO88024
, ETHERTYPE_IP
, ETHERTYPE_IPV6
,
254 ip_ether_v4_mapping
, ip_ether_v6_mapping
, ip_nodef_v6intfid
,
256 { DL_TPR
, IFT_ISO88025
, ETHERTYPE_IP
, ETHERTYPE_IPV6
,
257 ip_ether_v4_mapping
, ip_ether_v6_mapping
, ip_nodef_v6intfid
,
259 { DL_FDDI
, IFT_FDDI
, ETHERTYPE_IP
, ETHERTYPE_IPV6
,
260 ip_ether_v4_mapping
, ip_ether_v6_mapping
, ip_ether_v6intfid
,
262 { DL_IB
, IFT_IB
, ETHERTYPE_IP
, ETHERTYPE_IPV6
,
263 ip_ib_v4_mapping
, ip_ib_v6_mapping
, ip_ib_v6intfid
,
265 { DL_IPV4
, IFT_IPV4
, IPPROTO_ENCAP
, IPPROTO_IPV6
,
266 ip_mbcast_mapping
, ip_mbcast_mapping
, ip_ipv4_v6intfid
,
267 ip_ipv4_v6destintfid
},
268 { DL_IPV6
, IFT_IPV6
, IPPROTO_ENCAP
, IPPROTO_IPV6
,
269 ip_mbcast_mapping
, ip_mbcast_mapping
, ip_ipv6_v6intfid
,
270 ip_ipv6_v6destintfid
},
271 { DL_6TO4
, IFT_6TO4
, IPPROTO_ENCAP
, IPPROTO_IPV6
,
272 ip_mbcast_mapping
, ip_mbcast_mapping
, ip_ipv4_v6intfid
,
274 { SUNW_DL_VNI
, IFT_OTHER
, ETHERTYPE_IP
, ETHERTYPE_IPV6
,
275 NULL
, NULL
, ip_nodef_v6intfid
, ip_nodef_v6intfid
},
276 { SUNW_DL_IPMP
, IFT_OTHER
, ETHERTYPE_IP
, ETHERTYPE_IPV6
,
277 NULL
, NULL
, ip_ipmp_v6intfid
, ip_nodef_v6intfid
},
278 { DL_OTHER
, IFT_OTHER
, ETHERTYPE_IP
, ETHERTYPE_IPV6
,
279 ip_ether_v4_mapping
, ip_ether_v6_mapping
, ip_nodef_v6intfid
,
283 char ipif_loopback_name
[] = "lo0";
285 /* These are used by all IP network modules. */
286 sin6_t sin6_null
; /* Zero address for quick clears */
287 sin_t sin_null
; /* Zero address for quick clears */
289 /* When set search for unused ipif_seqid */
290 static ipif_t ipif_zero
;
293 * ppa arena is created after these many
294 * interfaces have been plumbed.
296 uint_t ill_no_arena
= 12; /* Setable in /etc/system */
299 * Allocate per-interface mibs.
300 * Returns true if ok. False otherwise.
301 * ipsq may not yet be allocated (loopback case ).
304 ill_allocate_mibs(ill_t
*ill
)
306 /* Already allocated? */
307 if (ill
->ill_ip_mib
!= NULL
) {
309 ASSERT(ill
->ill_icmp6_mib
!= NULL
);
313 ill
->ill_ip_mib
= kmem_zalloc(sizeof (*ill
->ill_ip_mib
),
315 if (ill
->ill_ip_mib
== NULL
) {
319 /* Setup static information */
320 SET_MIB(ill
->ill_ip_mib
->ipIfStatsEntrySize
,
321 sizeof (mib2_ipIfStatsEntry_t
));
323 ill
->ill_ip_mib
->ipIfStatsIPVersion
= MIB2_INETADDRESSTYPE_ipv6
;
324 SET_MIB(ill
->ill_ip_mib
->ipIfStatsAddrEntrySize
,
325 sizeof (mib2_ipv6AddrEntry_t
));
326 SET_MIB(ill
->ill_ip_mib
->ipIfStatsRouteEntrySize
,
327 sizeof (mib2_ipv6RouteEntry_t
));
328 SET_MIB(ill
->ill_ip_mib
->ipIfStatsNetToMediaEntrySize
,
329 sizeof (mib2_ipv6NetToMediaEntry_t
));
330 SET_MIB(ill
->ill_ip_mib
->ipIfStatsMemberEntrySize
,
331 sizeof (ipv6_member_t
));
332 SET_MIB(ill
->ill_ip_mib
->ipIfStatsGroupSourceEntrySize
,
333 sizeof (ipv6_grpsrc_t
));
335 ill
->ill_ip_mib
->ipIfStatsIPVersion
= MIB2_INETADDRESSTYPE_ipv4
;
336 SET_MIB(ill
->ill_ip_mib
->ipIfStatsAddrEntrySize
,
337 sizeof (mib2_ipAddrEntry_t
));
338 SET_MIB(ill
->ill_ip_mib
->ipIfStatsRouteEntrySize
,
339 sizeof (mib2_ipRouteEntry_t
));
340 SET_MIB(ill
->ill_ip_mib
->ipIfStatsNetToMediaEntrySize
,
341 sizeof (mib2_ipNetToMediaEntry_t
));
342 SET_MIB(ill
->ill_ip_mib
->ipIfStatsMemberEntrySize
,
343 sizeof (ip_member_t
));
344 SET_MIB(ill
->ill_ip_mib
->ipIfStatsGroupSourceEntrySize
,
345 sizeof (ip_grpsrc_t
));
348 * For a v4 ill, we are done at this point, because per ill
349 * icmp mibs are only used for v6.
354 ill
->ill_icmp6_mib
= kmem_zalloc(sizeof (*ill
->ill_icmp6_mib
),
356 if (ill
->ill_icmp6_mib
== NULL
) {
357 kmem_free(ill
->ill_ip_mib
, sizeof (*ill
->ill_ip_mib
));
358 ill
->ill_ip_mib
= NULL
;
361 /* static icmp info */
362 ill
->ill_icmp6_mib
->ipv6IfIcmpEntrySize
=
363 sizeof (mib2_ipv6IfIcmpEntry_t
);
365 * The ipIfStatsIfindex and ipv6IfIcmpIndex will be assigned later
366 * after the phyint merge occurs in ipif_set_values -> ill_glist_insert
367 * -> ill_phyint_reinit
373 * Completely vaporize a lower level tap and all associated interfaces.
374 * ill_delete is called only out of ip_close when the device control
375 * stream is being closed.
378 ill_delete(ill_t
*ill
)
382 ip_stack_t
*ipst
= ill
->ill_ipst
;
385 * ill_delete may be forcibly entering the ipsq. The previous
386 * ioctl may not have completed and may need to be aborted.
387 * ipsq_flush takes care of it. If we don't need to enter the
388 * the ipsq forcibly, the 2nd invocation of ipsq_flush in
389 * ill_delete_tail is sufficient.
394 * Nuke all interfaces. ipif_free will take down the interface,
395 * remove it from the list, and free the data structure.
396 * Walk down the ipif list and remove the logical interfaces
397 * first before removing the main ipif. We can't unplumb
398 * zeroth interface first in the case of IPv6 as update_conn_ill
399 * -> ip_ll_multireq de-references ill_ipif for checking
402 * If ill_ipif was not properly initialized (i.e low on memory),
403 * then no interfaces to clean up. In this case just clean up the
406 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
; ipif
= ipif
->ipif_next
)
410 * clean out all the nce_t entries that depend on this
411 * ill for the ill_phys_addr.
413 nce_flush(ill
, B_TRUE
);
415 /* Clean up msgs on pending upcalls for mrouted */
418 update_conn_ill(ill
, ipst
);
421 * Remove multicast references added as a result of calls to
422 * ip_join_allmulti().
424 ip_purge_allmulti(ill
);
427 * If the ill being deleted is under IPMP, boot it out of the illgrp.
429 if (IS_UNDER_IPMP(ill
))
430 ipmp_ill_leave_illgrp(ill
);
433 * ill_down will arrange to blow off any IRE's dependent on this
434 * ILL, and shut down fragmentation reassembly.
438 /* Let SCTP know, so that it can remove this from its list. */
439 sctp_update_ill(ill
, SCTP_ILL_REMOVE
);
442 * Walk all CONNs that can have a reference on an ire or nce for this
443 * ill (we actually walk all that now have stale references).
445 ipcl_walk(conn_ixa_cleanup
, (void *)B_TRUE
, ipst
);
447 /* With IPv6 we have dce_ifindex. Cleanup for neatness */
449 dce_cleanup(ill
->ill_phyint
->phyint_ifindex
, ipst
);
452 * If an address on this ILL is being used as a source address then
453 * clear out the pointers in other ILLs that point to this ILL.
455 rw_enter(&ipst
->ips_ill_g_usesrc_lock
, RW_WRITER
);
456 if (ill
->ill_usesrc_grp_next
!= NULL
) {
457 if (ill
->ill_usesrc_ifindex
== 0) { /* usesrc ILL ? */
458 ill_disband_usesrc_group(ill
);
459 } else { /* consumer of the usesrc ILL */
460 prev_ill
= ill_prev_usesrc(ill
);
461 prev_ill
->ill_usesrc_grp_next
=
462 ill
->ill_usesrc_grp_next
;
465 rw_exit(&ipst
->ips_ill_g_usesrc_lock
);
469 ipif_non_duplicate(ipif_t
*ipif
)
471 ill_t
*ill
= ipif
->ipif_ill
;
472 mutex_enter(&ill
->ill_lock
);
473 if (ipif
->ipif_flags
& IPIF_DUPLICATE
) {
474 ipif
->ipif_flags
&= ~IPIF_DUPLICATE
;
475 ASSERT(ill
->ill_ipif_dup_count
> 0);
476 ill
->ill_ipif_dup_count
--;
478 mutex_exit(&ill
->ill_lock
);
482 * ill_delete_tail is called from ip_modclose after all references
483 * to the closing ill are gone. The wait is done in ip_modclose
486 ill_delete_tail(ill_t
*ill
)
490 ip_stack_t
*ipst
= ill
->ill_ipst
;
492 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
; ipif
= ipif
->ipif_next
) {
493 ipif_non_duplicate(ipif
);
494 (void) ipif_down_tail(ipif
);
497 ASSERT(ill
->ill_ipif_dup_count
== 0);
500 * If polling capability is enabled (which signifies direct
501 * upcall into IP and driver has ill saved as a handle),
502 * we need to make sure that unbind has completed before we
503 * let the ill disappear and driver no longer has any reference
506 mutex_enter(&ill
->ill_lock
);
507 while (ill
->ill_state_flags
& ILL_DL_UNBIND_IN_PROGRESS
)
508 cv_wait(&ill
->ill_cv
, &ill
->ill_lock
);
509 mutex_exit(&ill
->ill_lock
);
510 ASSERT(!(ill
->ill_capabilities
&
511 (ILL_CAPAB_DLD
| ILL_CAPAB_DLD_POLL
| ILL_CAPAB_DLD_DIRECT
)));
513 if (ill
->ill_net_type
!= IRE_LOOPBACK
)
514 qprocsoff(ill
->ill_rq
);
517 * We do an ipsq_flush once again now. New messages could have
518 * landed up from below (M_ERROR or M_HANGUP). Similarly ioctls
519 * could also have landed up if an ioctl thread had looked up
520 * the ill before we set the ILL_CONDEMNED flag, but not yet
521 * enqueued the ioctl when we did the ipsq_flush last time.
528 if (ill
->ill_hcksum_capab
!= NULL
) {
529 kmem_free(ill
->ill_hcksum_capab
, sizeof (ill_hcksum_capab_t
));
530 ill
->ill_hcksum_capab
= NULL
;
533 if (ill
->ill_zerocopy_capab
!= NULL
) {
534 kmem_free(ill
->ill_zerocopy_capab
,
535 sizeof (ill_zerocopy_capab_t
));
536 ill
->ill_zerocopy_capab
= NULL
;
539 if (ill
->ill_lso_capab
!= NULL
) {
540 kmem_free(ill
->ill_lso_capab
, sizeof (ill_lso_capab_t
));
541 ill
->ill_lso_capab
= NULL
;
544 if (ill
->ill_dld_capab
!= NULL
) {
545 kmem_free(ill
->ill_dld_capab
, sizeof (ill_dld_capab_t
));
546 ill
->ill_dld_capab
= NULL
;
549 /* Clean up ill_allowed_ips* related state */
550 if (ill
->ill_allowed_ips
!= NULL
) {
551 ASSERT(ill
->ill_allowed_ips_cnt
> 0);
552 kmem_free(ill
->ill_allowed_ips
,
553 ill
->ill_allowed_ips_cnt
* sizeof (in6_addr_t
));
554 ill
->ill_allowed_ips
= NULL
;
555 ill
->ill_allowed_ips_cnt
= 0;
558 while (ill
->ill_ipif
!= NULL
)
559 ipif_free_tail(ill
->ill_ipif
);
562 * We have removed all references to ilm from conn and the ones joined
565 * We don't walk conns, mrts and ires because
567 * 1) update_conn_ill and reset_mrt_ill cleans up conns and mrts.
568 * 2) ill_down ->ill_downi walks all the ires and cleans up
573 * If this ill is an IPMP meta-interface, blow away the illgrp. This
574 * is safe to do because the illgrp has already been unlinked from the
575 * group by I_PUNLINK, and thus SIOCSLIFGROUPNAME cannot find it.
578 ipmp_illgrp_destroy(ill
->ill_grp
);
582 if (ill
->ill_mphysaddr_list
!= NULL
) {
583 multiphysaddr_t
*mpa
, *tmpa
;
585 mpa
= ill
->ill_mphysaddr_list
;
586 ill
->ill_mphysaddr_list
= NULL
;
588 tmpa
= mpa
->mpa_next
;
589 kmem_free(mpa
, sizeof (*mpa
));
594 * Take us out of the list of ILLs. ill_glist_delete -> phyint_free
595 * could free the phyint. No more reference to the phyint after this
598 (void) ill_glist_delete(ill
);
600 if (ill
->ill_frag_ptr
!= NULL
) {
603 for (count
= 0; count
< ILL_FRAG_HASH_TBL_COUNT
; count
++) {
604 mutex_destroy(&ill
->ill_frag_hash_tbl
[count
].ipfb_lock
);
606 mi_free(ill
->ill_frag_ptr
);
607 ill
->ill_frag_ptr
= NULL
;
608 ill
->ill_frag_hash_tbl
= NULL
;
611 freemsg(ill
->ill_nd_lla_mp
);
612 /* Free all retained control messages. */
613 mpp
= &ill
->ill_first_mp_to_free
;
621 for (mp1
= mp
; mp1
!= NULL
; mp1
= mp1
->b_cont
) {
627 } while (mpp
++ != &ill
->ill_last_mp_to_free
);
632 ill_trace_cleanup(ill
);
635 /* The default multicast interface might have changed */
636 ire_increment_multicast_generation(ipst
, ill
->ill_isv6
);
638 /* Drop refcnt here */
639 netstack_rele(ill
->ill_ipst
->ips_netstack
);
640 ill
->ill_ipst
= NULL
;
644 ill_free_mib(ill_t
*ill
)
646 ip_stack_t
*ipst
= ill
->ill_ipst
;
649 * MIB statistics must not be lost, so when an interface
650 * goes away the counter values will be added to the global
653 if (ill
->ill_ip_mib
!= NULL
) {
655 ip_mib2_add_ip_stats(&ipst
->ips_ip6_mib
,
658 ip_mib2_add_ip_stats(&ipst
->ips_ip_mib
,
662 kmem_free(ill
->ill_ip_mib
, sizeof (*ill
->ill_ip_mib
));
663 ill
->ill_ip_mib
= NULL
;
665 if (ill
->ill_icmp6_mib
!= NULL
) {
666 ip_mib2_add_icmp6_stats(&ipst
->ips_icmp6_mib
,
668 kmem_free(ill
->ill_icmp6_mib
, sizeof (*ill
->ill_icmp6_mib
));
669 ill
->ill_icmp6_mib
= NULL
;
674 * Concatenate together a physical address and a sap.
676 * Sap_lengths are interpreted as follows:
677 * sap_length == 0 ==> no sap
678 * sap_length > 0 ==> sap is at the head of the dlpi address
679 * sap_length < 0 ==> sap is at the tail of the dlpi address
682 ill_dlur_copy_address(uchar_t
*phys_src
, uint_t phys_length
,
683 t_scalar_t sap_src
, t_scalar_t sap_length
, uchar_t
*dst
)
685 uint16_t sap_addr
= (uint16_t)sap_src
;
687 if (sap_length
== 0) {
688 if (phys_src
== NULL
)
689 bzero(dst
, phys_length
);
691 bcopy(phys_src
, dst
, phys_length
);
692 } else if (sap_length
< 0) {
693 if (phys_src
== NULL
)
694 bzero(dst
, phys_length
);
696 bcopy(phys_src
, dst
, phys_length
);
697 bcopy(&sap_addr
, (char *)dst
+ phys_length
, sizeof (sap_addr
));
699 bcopy(&sap_addr
, dst
, sizeof (sap_addr
));
700 if (phys_src
== NULL
)
701 bzero((char *)dst
+ sap_length
, phys_length
);
703 bcopy(phys_src
, (char *)dst
+ sap_length
, phys_length
);
708 * Generate a dl_unitdata_req mblk for the device and address given.
709 * addr_length is the length of the physical portion of the address.
710 * If addr is NULL include an all zero address of the specified length.
711 * TRUE? In any case, addr_length is taken to be the entire length of the
712 * dlpi address, including the absolute value of sap_length.
715 ill_dlur_gen(uchar_t
*addr
, uint_t addr_length
, t_uscalar_t sap
,
716 t_scalar_t sap_length
)
718 dl_unitdata_req_t
*dlur
;
720 t_scalar_t abs_sap_length
; /* absolute value */
722 abs_sap_length
= ABS(sap_length
);
723 mp
= ip_dlpi_alloc(sizeof (*dlur
) + addr_length
+ abs_sap_length
,
727 dlur
= (dl_unitdata_req_t
*)mp
->b_rptr
;
728 /* HACK: accomodate incompatible DLPI drivers */
729 if (addr_length
== 8)
731 dlur
->dl_dest_addr_length
= addr_length
+ abs_sap_length
;
732 dlur
->dl_dest_addr_offset
= sizeof (*dlur
);
733 dlur
->dl_priority
.dl_min
= 0;
734 dlur
->dl_priority
.dl_max
= 0;
735 ill_dlur_copy_address(addr
, addr_length
, sap
, sap_length
,
736 (uchar_t
*)&dlur
[1]);
741 * Add the pending mp to the list. There can be only 1 pending mp
742 * in the list. Any exclusive ioctl that needs to wait for a response
743 * from another module or driver needs to use this function to set
744 * the ipx_pending_mp to the ioctl mblk and wait for the response from
745 * the other module/driver. This is also used while waiting for the
746 * ipif/ill/ire refcnts to drop to zero in bringing down an ipif.
749 ipsq_pending_mp_add(conn_t
*connp
, ipif_t
*ipif
, queue_t
*q
, mblk_t
*add_mp
,
752 ipxop_t
*ipx
= ipif
->ipif_ill
->ill_phyint
->phyint_ipsq
->ipsq_xop
;
754 ASSERT(IAM_WRITER_IPIF(ipif
));
755 ASSERT(MUTEX_HELD(&ipif
->ipif_ill
->ill_lock
));
756 ASSERT((add_mp
->b_next
== NULL
) && (add_mp
->b_prev
== NULL
));
757 ASSERT(ipx
->ipx_pending_mp
== NULL
);
759 * The caller may be using a different ipif than the one passed into
760 * ipsq_current_start() (e.g., suppose an ioctl that came in on the V4
761 * ill needs to wait for the V6 ill to quiesce). So we can't ASSERT
762 * that `ipx_current_ipif == ipif'.
764 ASSERT(ipx
->ipx_current_ipif
!= NULL
);
767 * M_IOCDATA from ioctls, M_ERROR/M_HANGUP/M_PROTO/M_PCPROTO from the
770 ASSERT((DB_TYPE(add_mp
) == M_IOCDATA
) || (DB_TYPE(add_mp
) == M_ERROR
) ||
771 (DB_TYPE(add_mp
) == M_HANGUP
) || (DB_TYPE(add_mp
) == M_PROTO
) ||
772 (DB_TYPE(add_mp
) == M_PCPROTO
));
775 ASSERT(MUTEX_HELD(&connp
->conn_lock
));
777 * Return error if the conn has started closing. The conn
778 * could have finished cleaning up the pending mp list,
779 * If so we should not add another mp to the list negating
782 if (connp
->conn_state_flags
& CONN_CLOSING
)
785 mutex_enter(&ipx
->ipx_lock
);
786 ipx
->ipx_pending_ipif
= ipif
;
788 * Note down the queue in b_queue. This will be returned by
789 * ipsq_pending_mp_get. Caller will then use these values to restart
792 add_mp
->b_next
= NULL
;
794 ipx
->ipx_pending_mp
= add_mp
;
795 ipx
->ipx_waitfor
= waitfor
;
796 mutex_exit(&ipx
->ipx_lock
);
799 connp
->conn_oper_pending_ill
= ipif
->ipif_ill
;
805 * Retrieve the ipx_pending_mp and return it. There can be only 1 mp
806 * queued in the list.
809 ipsq_pending_mp_get(ipsq_t
*ipsq
, conn_t
**connpp
)
812 ipxop_t
*ipx
= ipsq
->ipsq_xop
;
815 mutex_enter(&ipx
->ipx_lock
);
816 if (ipx
->ipx_pending_mp
== NULL
) {
817 mutex_exit(&ipx
->ipx_lock
);
821 /* There can be only 1 such excl message */
822 curr
= ipx
->ipx_pending_mp
;
823 ASSERT(curr
->b_next
== NULL
);
824 ipx
->ipx_pending_ipif
= NULL
;
825 ipx
->ipx_pending_mp
= NULL
;
826 ipx
->ipx_waitfor
= 0;
827 mutex_exit(&ipx
->ipx_lock
);
829 if (CONN_Q(curr
->b_queue
)) {
831 * This mp did a refhold on the conn, at the start of the ioctl.
832 * So we can safely return a pointer to the conn to the caller.
834 *connpp
= Q_TO_CONN(curr
->b_queue
);
844 * Cleanup the ioctl mp queued in ipx_pending_mp
845 * - Called in the ill_delete path
846 * - Called in the M_ERROR or M_HANGUP path on the ill.
847 * - Called in the conn close path.
849 * Returns success on finding the pending mblk associated with the ioctl or
850 * exclusive operation in progress, failure otherwise.
853 ipsq_pending_mp_cleanup(ill_t
*ill
, conn_t
*connp
)
861 ASSERT(IAM_WRITER_ILL(ill
));
862 ipx
= ill
->ill_phyint
->phyint_ipsq
->ipsq_xop
;
864 mutex_enter(&ipx
->ipx_lock
);
865 mp
= ipx
->ipx_pending_mp
;
867 if (mp
== NULL
|| mp
->b_queue
!= CONNP_TO_WQ(connp
)) {
869 * Nothing to clean since the conn that is closing
870 * does not have a matching pending mblk in
873 mutex_exit(&ipx
->ipx_lock
);
878 * A non-zero ill_error signifies we are called in the
879 * M_ERROR or M_HANGUP path and we need to unconditionally
880 * abort any current ioctl and do the corresponding cleanup.
881 * A zero ill_error means we are in the ill_delete path and
882 * we do the cleanup only if there is a pending mp.
884 if (mp
== NULL
&& ill
->ill_error
== 0) {
885 mutex_exit(&ipx
->ipx_lock
);
890 /* Now remove from the ipx_pending_mp */
891 ipx
->ipx_pending_mp
= NULL
;
892 ipif
= ipx
->ipx_pending_ipif
;
893 ipx
->ipx_pending_ipif
= NULL
;
894 ipx
->ipx_waitfor
= 0;
895 ipx
->ipx_current_ipif
= NULL
;
896 cmd
= ipx
->ipx_current_ioctl
;
897 ipx
->ipx_current_ioctl
= 0;
898 ipx
->ipx_current_done
= B_TRUE
;
899 mutex_exit(&ipx
->ipx_lock
);
909 if (DB_TYPE(mp
) == M_IOCTL
|| DB_TYPE(mp
) == M_IOCDATA
) {
910 DTRACE_PROBE4(ipif__ioctl
,
911 char *, "ipsq_pending_mp_cleanup",
912 int, cmd
, ill_t
*, ipif
== NULL
? NULL
: ipif
->ipif_ill
,
915 ip_ioctl_finish(q
, mp
, ENXIO
, NO_COPYOUT
, NULL
);
917 ip_ioctl_finish(q
, mp
, ENXIO
, CONN_CLOSE
, NULL
);
918 mutex_enter(&ipif
->ipif_ill
->ill_lock
);
919 ipif
->ipif_state_flags
&= ~IPIF_CHANGING
;
920 mutex_exit(&ipif
->ipif_ill
->ill_lock
);
929 * Called in the conn close path and ill delete path
932 ipsq_xopq_mp_cleanup(ill_t
*ill
, conn_t
*connp
)
938 queue_t
*wq
, *rq
= NULL
;
939 mblk_t
*tmp_list
= NULL
;
941 ASSERT(IAM_WRITER_ILL(ill
));
943 wq
= CONNP_TO_WQ(connp
);
948 * In the case of lo0 being unplumbed, ill_wq will be NULL. Guard
954 ipsq
= ill
->ill_phyint
->phyint_ipsq
;
956 * Cleanup the ioctl mp's queued in ipsq_xopq_pending_mp if any.
957 * In the case of ioctl from a conn, there can be only 1 mp
958 * queued on the ipsq. If an ill is being unplumbed flush all
961 mutex_enter(&ipsq
->ipsq_lock
);
962 for (prev
= NULL
, curr
= ipsq
->ipsq_xopq_mphead
; curr
!= NULL
;
966 (curr
->b_queue
== wq
|| curr
->b_queue
== rq
)) {
967 /* Unlink the mblk from the pending mp list */
969 prev
->b_next
= curr
->b_next
;
971 ASSERT(ipsq
->ipsq_xopq_mphead
== curr
);
972 ipsq
->ipsq_xopq_mphead
= curr
->b_next
;
974 if (ipsq
->ipsq_xopq_mptail
== curr
)
975 ipsq
->ipsq_xopq_mptail
= prev
;
977 * Create a temporary list and release the ipsq lock
978 * New elements are added to the head of the tmp_list
980 curr
->b_next
= tmp_list
;
986 mutex_exit(&ipsq
->ipsq_lock
);
988 while (tmp_list
!= NULL
) {
990 tmp_list
= curr
->b_next
;
994 curr
->b_queue
= NULL
;
995 if (DB_TYPE(curr
) == M_IOCTL
|| DB_TYPE(curr
) == M_IOCDATA
) {
996 DTRACE_PROBE4(ipif__ioctl
,
997 char *, "ipsq_xopq_mp_cleanup",
998 int, 0, ill_t
*, NULL
, ipif_t
*, NULL
);
999 ip_ioctl_finish(wq
, curr
, ENXIO
, connp
!= NULL
?
1000 CONN_CLOSE
: NO_COPYOUT
, NULL
);
1003 * IP-MT XXX In the case of TLI/XTI bind / optmgmt
1004 * this can't be just inet_freemsg. we have to
1005 * restart it otherwise the thread will be stuck.
1013 * This conn has started closing. Cleanup any pending ioctl from this conn.
1014 * STREAMS ensures that there can be at most 1 active ioctl on a stream.
1017 conn_ioctl_cleanup(conn_t
*connp
)
1024 * Check for a queued ioctl. If the ioctl has not yet started, the mp
1025 * is pending in the list headed by ipsq_xopq_head. If the ioctl has
1026 * started the mp could be present in ipx_pending_mp. Note that if
1027 * conn_oper_pending_ill is NULL, the ioctl may still be in flight and
1028 * not yet queued anywhere. In this case, the conn close code will wait
1029 * until the conn_ref is dropped. If the stream was a tcp stream, then
1030 * tcp_close will wait first until all ioctls have completed for this
1033 mutex_enter(&connp
->conn_lock
);
1034 ill
= connp
->conn_oper_pending_ill
;
1036 mutex_exit(&connp
->conn_lock
);
1041 * We may not be able to refhold the ill if the ill/ipif
1042 * is changing. But we need to make sure that the ill will
1043 * not vanish. So we just bump up the ill_waiter count.
1045 refheld
= ill_waiter_inc(ill
);
1046 mutex_exit(&connp
->conn_lock
);
1048 if (ipsq_enter(ill
, B_TRUE
, NEW_OP
)) {
1049 ill_waiter_dcr(ill
);
1051 * Check whether this ioctl has started and is
1052 * pending. If it is not found there then check
1053 * whether this ioctl has not even started and is in
1054 * the ipsq_xopq list.
1056 if (!ipsq_pending_mp_cleanup(ill
, connp
))
1057 ipsq_xopq_mp_cleanup(ill
, connp
);
1058 ipsq
= ill
->ill_phyint
->phyint_ipsq
;
1065 * The ill is also closing and we could not bump up the
1066 * ill_waiter_count or we could not enter the ipsq. Leave
1067 * the cleanup to ill_delete
1069 mutex_enter(&connp
->conn_lock
);
1070 while (connp
->conn_oper_pending_ill
!= NULL
)
1071 cv_wait(&connp
->conn_refcv
, &connp
->conn_lock
);
1072 mutex_exit(&connp
->conn_lock
);
1074 ill_waiter_dcr(ill
);
1078 * ipcl_walk function for cleaning up conn_*_ill fields.
1079 * Note that we leave ixa_multicast_ifindex, conn_incoming_ifindex, and
1080 * conn_bound_if in place. We prefer dropping
1081 * packets instead of sending them out the wrong interface, or accepting
1082 * packets from the wrong ifindex.
1085 conn_cleanup_ill(conn_t
*connp
, caddr_t arg
)
1087 ill_t
*ill
= (ill_t
*)arg
;
1089 mutex_enter(&connp
->conn_lock
);
1090 if (connp
->conn_dhcpinit_ill
== ill
) {
1091 connp
->conn_dhcpinit_ill
= NULL
;
1092 ASSERT(ill
->ill_dhcpinit
!= 0);
1093 atomic_dec_32(&ill
->ill_dhcpinit
);
1094 ill_set_inputfn(ill
);
1096 mutex_exit(&connp
->conn_lock
);
1100 ill_down_ipifs_tail(ill_t
*ill
)
1105 ASSERT(IAM_WRITER_ILL(ill
));
1106 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
; ipif
= ipif
->ipif_next
) {
1107 ipif_non_duplicate(ipif
);
1109 * ipif_down_tail will call arp_ll_down on the last ipif
1110 * and typically return EINPROGRESS when the DL_UNBIND is sent.
1112 if ((err
= ipif_down_tail(ipif
)) != 0)
1120 ipif_all_down_tail(ipsq_t
*ipsq
, queue_t
*q
, mblk_t
*mp
, void *dummy_arg
)
1122 ASSERT(IAM_WRITER_IPSQ(ipsq
));
1123 (void) ill_down_ipifs_tail(q
->q_ptr
);
1125 ipsq_current_finish(ipsq
);
1129 * ill_down_start is called when we want to down this ill and bring it up again
1130 * It is called when we receive an M_ERROR / M_HANGUP. In this case we shut down
1131 * all interfaces, but don't tear down any plumbing.
1134 ill_down_start(queue_t
*q
, mblk_t
*mp
)
1136 ill_t
*ill
= q
->q_ptr
;
1139 ASSERT(IAM_WRITER_ILL(ill
));
1141 * It is possible that some ioctl is already in progress while we
1142 * received the M_ERROR / M_HANGUP in which case, we need to abort
1143 * the ioctl. ill_down_start() is being processed as CUR_OP rather
1144 * than as NEW_OP since the cause of the M_ERROR / M_HANGUP may prevent
1145 * the in progress ioctl from ever completing.
1147 * The thread that started the ioctl (if any) must have returned,
1148 * since we are now executing as writer. After the 2 calls below,
1149 * the state of the ipsq and the ill would reflect no trace of any
1150 * pending operation. Subsequently if there is any response to the
1151 * original ioctl from the driver, it would be discarded as an
1152 * unsolicited message from the driver.
1154 (void) ipsq_pending_mp_cleanup(ill
, NULL
);
1155 ill_dlpi_clear_deferred(ill
);
1157 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
; ipif
= ipif
->ipif_next
)
1158 (void) ipif_down(ipif
, NULL
, NULL
);
1163 * Walk all CONNs that can have a reference on an ire or nce for this
1164 * ill (we actually walk all that now have stale references).
1166 ipcl_walk(conn_ixa_cleanup
, (void *)B_TRUE
, ill
->ill_ipst
);
1168 /* With IPv6 we have dce_ifindex. Cleanup for neatness */
1170 dce_cleanup(ill
->ill_phyint
->phyint_ifindex
, ill
->ill_ipst
);
1172 ipsq_current_start(ill
->ill_phyint
->phyint_ipsq
, ill
->ill_ipif
, 0);
1175 * Atomically test and add the pending mp if references are active.
1177 mutex_enter(&ill
->ill_lock
);
1178 if (!ill_is_quiescent(ill
)) {
1179 /* call cannot fail since `conn_t *' argument is NULL */
1180 (void) ipsq_pending_mp_add(NULL
, ill
->ill_ipif
, ill
->ill_rq
,
1182 mutex_exit(&ill
->ill_lock
);
1185 mutex_exit(&ill
->ill_lock
);
1190 ill_down(ill_t
*ill
)
1193 ip_stack_t
*ipst
= ill
->ill_ipst
;
1196 * Blow off any IREs dependent on this ILL.
1197 * The caller needs to handle conn_ixa_cleanup
1199 ill_delete_ires(ill
);
1201 ire_walk_ill(0, 0, ill_downi
, ill
, ill
);
1203 /* Remove any conn_*_ill depending on this ill */
1204 ipcl_walk(conn_cleanup_ill
, (caddr_t
)ill
, ipst
);
1207 * Free state for additional IREs.
1209 mutex_enter(&ill
->ill_saved_ire_lock
);
1210 mp
= ill
->ill_saved_ire_mp
;
1211 ill
->ill_saved_ire_mp
= NULL
;
1212 ill
->ill_saved_ire_cnt
= 0;
1213 mutex_exit(&ill
->ill_saved_ire_lock
);
1218 * ire_walk routine used to delete every IRE that depends on
1219 * 'ill'. (Always called as writer, and may only be called from ire_walk.)
1221 * Note: since the routes added by the kernel are deleted separately,
1222 * this will only be 1) IRE_IF_CLONE and 2) manually added IRE_INTERFACE.
1224 * We also remove references on ire_nce_cache entries that refer to the ill.
1227 ill_downi(ire_t
*ire
, char *ill_arg
)
1229 ill_t
*ill
= (ill_t
*)ill_arg
;
1232 mutex_enter(&ire
->ire_lock
);
1233 nce
= ire
->ire_nce_cache
;
1234 if (nce
!= NULL
&& nce
->nce_ill
== ill
)
1235 ire
->ire_nce_cache
= NULL
;
1238 mutex_exit(&ire
->ire_lock
);
1241 if (ire
->ire_ill
== ill
) {
1243 * The existing interface binding for ire must be
1244 * deleted before trying to bind the route to another
1245 * interface. However, since we are using the contents of the
1246 * ire after ire_delete, the caller has to ensure that
1247 * CONDEMNED (deleted) ire's are not removed from the list
1248 * when ire_delete() returns. Currently ill_downi() is
1249 * only called as part of ire_walk*() routines, so that
1250 * the irb_refhold() done by ire_walk*() will ensure that
1251 * ire_delete() does not lead to ire_inactive().
1253 ASSERT(ire
->ire_bucket
->irb_refcnt
> 0);
1255 if (ire
->ire_unbound
)
1260 /* Remove IRE_IF_CLONE on this ill */
1262 ill_downi_if_clone(ire_t
*ire
, char *ill_arg
)
1264 ill_t
*ill
= (ill_t
*)ill_arg
;
1266 ASSERT(ire
->ire_type
& IRE_IF_CLONE
);
1267 if (ire
->ire_ill
== ill
)
1271 /* Consume an M_IOCACK of the fastpath probe. */
1273 ill_fastpath_ack(ill_t
*ill
, mblk_t
*mp
)
1278 * If this was the first attempt turn on the fastpath probing.
1280 mutex_enter(&ill
->ill_lock
);
1281 if (ill
->ill_dlpi_fastpath_state
== IDS_INPROGRESS
)
1282 ill
->ill_dlpi_fastpath_state
= IDS_OK
;
1283 mutex_exit(&ill
->ill_lock
);
1285 /* Free the M_IOCACK mblk, hold on to the data */
1290 if (mp
->b_cont
!= NULL
)
1291 nce_fastpath_update(ill
, mp
);
1293 ip0dbg(("ill_fastpath_ack: no b_cont\n"));
1298 * Throw an M_IOCTL message downstream asking "do you know fastpath?"
1299 * The data portion of the request is a dl_unitdata_req_t template for
1300 * what we would send downstream in the absence of a fastpath confirmation.
1303 ill_fastpath_probe(ill_t
*ill
, mblk_t
*dlur_mp
)
1308 if (dlur_mp
== NULL
)
1311 mutex_enter(&ill
->ill_lock
);
1312 switch (ill
->ill_dlpi_fastpath_state
) {
1315 * Driver NAKed the first fastpath ioctl - assume it doesn't
1318 mutex_exit(&ill
->ill_lock
);
1321 /* This is the first probe */
1322 ill
->ill_dlpi_fastpath_state
= IDS_INPROGRESS
;
1327 mutex_exit(&ill
->ill_lock
);
1329 if ((mp
= mkiocb(DL_IOC_HDR_INFO
)) == NULL
)
1332 mp
->b_cont
= copyb(dlur_mp
);
1333 if (mp
->b_cont
== NULL
) {
1338 ioc
= (struct iocblk
*)mp
->b_rptr
;
1339 ioc
->ioc_count
= msgdsize(mp
->b_cont
);
1341 DTRACE_PROBE3(ill__dlpi
, char *, "ill_fastpath_probe",
1342 char *, "DL_IOC_HDR_INFO", ill_t
*, ill
);
1343 putnext(ill
->ill_wq
, mp
);
1348 ill_capability_probe(ill_t
*ill
)
1352 ASSERT(IAM_WRITER_ILL(ill
));
1354 if (ill
->ill_dlpi_capab_state
!= IDCS_UNKNOWN
&&
1355 ill
->ill_dlpi_capab_state
!= IDCS_FAILED
)
1359 * We are starting a new cycle of capability negotiation.
1360 * Free up the capab reset messages of any previous incarnation.
1361 * We will do a fresh allocation when we get the response to our probe
1363 if (ill
->ill_capab_reset_mp
!= NULL
) {
1364 freemsg(ill
->ill_capab_reset_mp
);
1365 ill
->ill_capab_reset_mp
= NULL
;
1368 ip1dbg(("ill_capability_probe: starting capability negotiation\n"));
1370 mp
= ip_dlpi_alloc(sizeof (dl_capability_req_t
), DL_CAPABILITY_REQ
);
1374 ill_capability_send(ill
, mp
);
1375 ill
->ill_dlpi_capab_state
= IDCS_PROBE_SENT
;
1379 ill_capability_reset(ill_t
*ill
, boolean_t reneg
)
1381 ASSERT(IAM_WRITER_ILL(ill
));
1383 if (ill
->ill_dlpi_capab_state
!= IDCS_OK
)
1386 ill
->ill_dlpi_capab_state
= reneg
? IDCS_RENEG
: IDCS_RESET_SENT
;
1388 ill_capability_send(ill
, ill
->ill_capab_reset_mp
);
1389 ill
->ill_capab_reset_mp
= NULL
;
1391 * We turn off all capabilities except those pertaining to
1392 * direct function call capabilities viz. ILL_CAPAB_DLD*
1393 * which will be turned off by the corresponding reset functions.
1395 ill
->ill_capabilities
&= ~(ILL_CAPAB_HCKSUM
| ILL_CAPAB_ZEROCOPY
);
1399 ill_capability_reset_alloc(ill_t
*ill
)
1404 dl_capability_req_t
*capb
;
1406 ASSERT(IAM_WRITER_ILL(ill
));
1407 ASSERT(ill
->ill_capab_reset_mp
== NULL
);
1409 if (ILL_HCKSUM_CAPABLE(ill
)) {
1410 size
+= sizeof (dl_capability_sub_t
) +
1411 sizeof (dl_capab_hcksum_t
);
1414 if (ill
->ill_capabilities
& ILL_CAPAB_ZEROCOPY
) {
1415 size
+= sizeof (dl_capability_sub_t
) +
1416 sizeof (dl_capab_zerocopy_t
);
1419 if (ill
->ill_capabilities
& ILL_CAPAB_DLD
) {
1420 size
+= sizeof (dl_capability_sub_t
) +
1421 sizeof (dl_capab_dld_t
);
1424 mp
= allocb_wait(size
+ sizeof (dl_capability_req_t
), BPRI_MED
,
1427 mp
->b_datap
->db_type
= M_PROTO
;
1428 bzero(mp
->b_rptr
, size
+ sizeof (dl_capability_req_t
));
1430 capb
= (dl_capability_req_t
*)mp
->b_rptr
;
1431 capb
->dl_primitive
= DL_CAPABILITY_REQ
;
1432 capb
->dl_sub_offset
= sizeof (dl_capability_req_t
);
1433 capb
->dl_sub_length
= size
;
1435 mp
->b_wptr
+= sizeof (dl_capability_req_t
);
1438 * Each handler fills in the corresponding dl_capability_sub_t
1441 ill_capability_hcksum_reset_fill(ill
, mp
);
1442 ill_capability_zerocopy_reset_fill(ill
, mp
);
1443 ill_capability_dld_reset_fill(ill
, mp
);
1445 ill
->ill_capab_reset_mp
= mp
;
1449 ill_capability_id_ack(ill_t
*ill
, mblk_t
*mp
, dl_capability_sub_t
*outers
)
1451 dl_capab_id_t
*id_ic
;
1452 uint_t sub_dl_cap
= outers
->dl_cap
;
1453 dl_capability_sub_t
*inners
;
1456 ASSERT(sub_dl_cap
== DL_CAPAB_ID_WRAPPER
);
1459 * Note: range checks here are not absolutely sufficient to
1460 * make us robust against malformed messages sent by drivers;
1461 * this is in keeping with the rest of IP's dlpi handling.
1462 * (Remember, it's coming from something else in the kernel
1466 capend
= (uint8_t *)(outers
+ 1) + outers
->dl_length
;
1467 if (capend
> mp
->b_wptr
) {
1468 cmn_err(CE_WARN
, "ill_capability_id_ack: "
1469 "malformed sub-capability too long for mblk");
1473 id_ic
= (dl_capab_id_t
*)(outers
+ 1);
1475 if (outers
->dl_length
< sizeof (*id_ic
) ||
1476 (inners
= &id_ic
->id_subcap
,
1477 inners
->dl_length
> (outers
->dl_length
- sizeof (*inners
)))) {
1478 cmn_err(CE_WARN
, "ill_capability_id_ack: malformed "
1479 "encapsulated capab type %d too long for mblk",
1484 if (!dlcapabcheckqid(&id_ic
->id_mid
, ill
->ill_lmod_rq
)) {
1485 ip1dbg(("ill_capability_id_ack: mid token for capab type %d "
1486 "isn't as expected; pass-thru module(s) detected, "
1487 "discarding capability\n", inners
->dl_cap
));
1491 /* Process the encapsulated sub-capability */
1492 ill_capability_dispatch(ill
, mp
, inners
);
1496 ill_capability_dld_reset_fill(ill_t
*ill
, mblk_t
*mp
)
1498 dl_capability_sub_t
*dl_subcap
;
1500 if (!(ill
->ill_capabilities
& ILL_CAPAB_DLD
))
1504 * The dl_capab_dld_t that follows the dl_capability_sub_t is not
1505 * initialized below since it is not used by DLD.
1507 dl_subcap
= (dl_capability_sub_t
*)mp
->b_wptr
;
1508 dl_subcap
->dl_cap
= DL_CAPAB_DLD
;
1509 dl_subcap
->dl_length
= sizeof (dl_capab_dld_t
);
1511 mp
->b_wptr
+= sizeof (dl_capability_sub_t
) + sizeof (dl_capab_dld_t
);
1515 ill_capability_dispatch(ill_t
*ill
, mblk_t
*mp
, dl_capability_sub_t
*subp
)
1518 * If no ipif was brought up over this ill, this DL_CAPABILITY_REQ/ACK
1519 * is only to get the VRRP capability.
1521 * Note that we cannot check ill_ipif_up_count here since
1522 * ill_ipif_up_count is only incremented when the resolver is setup.
1523 * That is done asynchronously, and can race with this function.
1525 if (!ill
->ill_dl_up
) {
1526 if (subp
->dl_cap
== DL_CAPAB_VRRP
)
1527 ill_capability_vrrp_ack(ill
, mp
, subp
);
1531 switch (subp
->dl_cap
) {
1532 case DL_CAPAB_HCKSUM
:
1533 ill_capability_hcksum_ack(ill
, mp
, subp
);
1535 case DL_CAPAB_ZEROCOPY
:
1536 ill_capability_zerocopy_ack(ill
, mp
, subp
);
1539 ill_capability_dld_ack(ill
, mp
, subp
);
1544 ip1dbg(("ill_capability_dispatch: unknown capab type %d\n",
1550 * Process the vrrp capability received from a DLS Provider. isub must point
1551 * to the sub-capability (DL_CAPAB_VRRP) of a DL_CAPABILITY_ACK message.
1554 ill_capability_vrrp_ack(ill_t
*ill
, mblk_t
*mp
, dl_capability_sub_t
*isub
)
1556 dl_capab_vrrp_t
*vrrp
;
1557 uint_t sub_dl_cap
= isub
->dl_cap
;
1560 ASSERT(IAM_WRITER_ILL(ill
));
1561 ASSERT(sub_dl_cap
== DL_CAPAB_VRRP
);
1564 * Note: range checks here are not absolutely sufficient to
1565 * make us robust against malformed messages sent by drivers;
1566 * this is in keeping with the rest of IP's dlpi handling.
1567 * (Remember, it's coming from something else in the kernel
1570 capend
= (uint8_t *)(isub
+ 1) + isub
->dl_length
;
1571 if (capend
> mp
->b_wptr
) {
1572 cmn_err(CE_WARN
, "ill_capability_vrrp_ack: "
1573 "malformed sub-capability too long for mblk");
1576 vrrp
= (dl_capab_vrrp_t
*)(isub
+ 1);
1579 * Compare the IP address family and set ILLF_VRRP for the right ill.
1581 if ((vrrp
->vrrp_af
== AF_INET6
&& ill
->ill_isv6
) ||
1582 (vrrp
->vrrp_af
== AF_INET
&& !ill
->ill_isv6
)) {
1583 ill
->ill_flags
|= ILLF_VRRP
;
1588 * Process a hardware checksum offload capability negotiation ack received
1589 * from a DLS Provider.isub must point to the sub-capability (DL_CAPAB_HCKSUM)
1590 * of a DL_CAPABILITY_ACK message.
1593 ill_capability_hcksum_ack(ill_t
*ill
, mblk_t
*mp
, dl_capability_sub_t
*isub
)
1595 dl_capability_req_t
*ocap
;
1596 dl_capab_hcksum_t
*ihck
, *ohck
;
1597 ill_hcksum_capab_t
**ill_hcksum
;
1599 uint_t sub_dl_cap
= isub
->dl_cap
;
1602 ASSERT(sub_dl_cap
== DL_CAPAB_HCKSUM
);
1604 ill_hcksum
= (ill_hcksum_capab_t
**)&ill
->ill_hcksum_capab
;
1607 * Note: range checks here are not absolutely sufficient to
1608 * make us robust against malformed messages sent by drivers;
1609 * this is in keeping with the rest of IP's dlpi handling.
1610 * (Remember, it's coming from something else in the kernel
1613 capend
= (uint8_t *)(isub
+ 1) + isub
->dl_length
;
1614 if (capend
> mp
->b_wptr
) {
1615 cmn_err(CE_WARN
, "ill_capability_hcksum_ack: "
1616 "malformed sub-capability too long for mblk");
1621 * There are two types of acks we process here:
1622 * 1. acks in reply to a (first form) generic capability req
1623 * (no ENABLE flag set)
1624 * 2. acks in reply to a ENABLE capability req.
1627 ihck
= (dl_capab_hcksum_t
*)(isub
+ 1);
1629 if (ihck
->hcksum_version
!= HCKSUM_VERSION_1
) {
1630 cmn_err(CE_CONT
, "ill_capability_hcksum_ack: "
1631 "unsupported hardware checksum "
1632 "sub-capability (version %d, expected %d)",
1633 ihck
->hcksum_version
, HCKSUM_VERSION_1
);
1637 if (!dlcapabcheckqid(&ihck
->hcksum_mid
, ill
->ill_lmod_rq
)) {
1638 ip1dbg(("ill_capability_hcksum_ack: mid token for hardware "
1639 "checksum capability isn't as expected; pass-thru "
1640 "module(s) detected, discarding capability\n"));
1644 #define CURR_HCKSUM_CAPAB \
1645 (HCKSUM_INET_PARTIAL | HCKSUM_INET_FULL_V4 | \
1646 HCKSUM_INET_FULL_V6 | HCKSUM_IPHDRCKSUM)
1648 if ((ihck
->hcksum_txflags
& HCKSUM_ENABLE
) &&
1649 (ihck
->hcksum_txflags
& CURR_HCKSUM_CAPAB
)) {
1650 /* do ENABLE processing */
1651 if (*ill_hcksum
== NULL
) {
1652 *ill_hcksum
= kmem_zalloc(sizeof (ill_hcksum_capab_t
),
1655 if (*ill_hcksum
== NULL
) {
1656 cmn_err(CE_WARN
, "ill_capability_hcksum_ack: "
1657 "could not enable hcksum version %d "
1658 "for %s (ENOMEM)\n", HCKSUM_CURRENT_VERSION
,
1664 (*ill_hcksum
)->ill_hcksum_version
= ihck
->hcksum_version
;
1665 (*ill_hcksum
)->ill_hcksum_txflags
= ihck
->hcksum_txflags
;
1666 ill
->ill_capabilities
|= ILL_CAPAB_HCKSUM
;
1667 ip1dbg(("ill_capability_hcksum_ack: interface %s "
1668 "has enabled hardware checksumming\n ",
1670 } else if (ihck
->hcksum_txflags
& CURR_HCKSUM_CAPAB
) {
1672 * Enabling hardware checksum offload
1673 * Currently IP supports {TCP,UDP}/IPv4
1674 * partial and full cksum offload and
1675 * IPv4 header checksum offload.
1676 * Allocate new mblk which will
1677 * contain a new capability request
1678 * to enable hardware checksum offload.
1683 size
= sizeof (dl_capability_req_t
) +
1684 sizeof (dl_capability_sub_t
) + isub
->dl_length
;
1686 if ((nmp
= ip_dlpi_alloc(size
, DL_CAPABILITY_REQ
)) == NULL
) {
1687 cmn_err(CE_WARN
, "ill_capability_hcksum_ack: "
1688 "could not enable hardware cksum for %s (ENOMEM)\n",
1694 /* initialize dl_capability_req_t */
1695 ocap
= (dl_capability_req_t
*)nmp
->b_rptr
;
1696 ocap
->dl_sub_offset
=
1697 sizeof (dl_capability_req_t
);
1698 ocap
->dl_sub_length
=
1699 sizeof (dl_capability_sub_t
) +
1701 nmp
->b_rptr
+= sizeof (dl_capability_req_t
);
1703 /* initialize dl_capability_sub_t */
1704 bcopy(isub
, nmp
->b_rptr
, sizeof (*isub
));
1705 nmp
->b_rptr
+= sizeof (*isub
);
1707 /* initialize dl_capab_hcksum_t */
1708 ohck
= (dl_capab_hcksum_t
*)nmp
->b_rptr
;
1709 bcopy(ihck
, ohck
, sizeof (*ihck
));
1712 ASSERT(nmp
->b_wptr
== (nmp
->b_rptr
+ size
));
1714 /* Set ENABLE flag */
1715 ohck
->hcksum_txflags
&= CURR_HCKSUM_CAPAB
;
1716 ohck
->hcksum_txflags
|= HCKSUM_ENABLE
;
1719 * nmp points to a DL_CAPABILITY_REQ message to enable
1720 * hardware checksum acceleration.
1722 ill_capability_send(ill
, nmp
);
1724 ip1dbg(("ill_capability_hcksum_ack: interface %s has "
1725 "advertised %x hardware checksum capability flags\n",
1726 ill
->ill_name
, ihck
->hcksum_txflags
));
1731 ill_capability_hcksum_reset_fill(ill_t
*ill
, mblk_t
*mp
)
1733 dl_capab_hcksum_t
*hck_subcap
;
1734 dl_capability_sub_t
*dl_subcap
;
1736 if (!ILL_HCKSUM_CAPABLE(ill
))
1739 ASSERT(ill
->ill_hcksum_capab
!= NULL
);
1741 dl_subcap
= (dl_capability_sub_t
*)mp
->b_wptr
;
1742 dl_subcap
->dl_cap
= DL_CAPAB_HCKSUM
;
1743 dl_subcap
->dl_length
= sizeof (*hck_subcap
);
1745 hck_subcap
= (dl_capab_hcksum_t
*)(dl_subcap
+ 1);
1746 hck_subcap
->hcksum_version
= ill
->ill_hcksum_capab
->ill_hcksum_version
;
1747 hck_subcap
->hcksum_txflags
= 0;
1749 mp
->b_wptr
+= sizeof (*dl_subcap
) + sizeof (*hck_subcap
);
1753 ill_capability_zerocopy_ack(ill_t
*ill
, mblk_t
*mp
, dl_capability_sub_t
*isub
)
1756 dl_capability_req_t
*oc
;
1757 dl_capab_zerocopy_t
*zc_ic
, *zc_oc
;
1758 ill_zerocopy_capab_t
**ill_zerocopy_capab
;
1759 uint_t sub_dl_cap
= isub
->dl_cap
;
1762 ASSERT(sub_dl_cap
== DL_CAPAB_ZEROCOPY
);
1764 ill_zerocopy_capab
= (ill_zerocopy_capab_t
**)&ill
->ill_zerocopy_capab
;
1767 * Note: range checks here are not absolutely sufficient to
1768 * make us robust against malformed messages sent by drivers;
1769 * this is in keeping with the rest of IP's dlpi handling.
1770 * (Remember, it's coming from something else in the kernel
1773 capend
= (uint8_t *)(isub
+ 1) + isub
->dl_length
;
1774 if (capend
> mp
->b_wptr
) {
1775 cmn_err(CE_WARN
, "ill_capability_zerocopy_ack: "
1776 "malformed sub-capability too long for mblk");
1780 zc_ic
= (dl_capab_zerocopy_t
*)(isub
+ 1);
1781 if (zc_ic
->zerocopy_version
!= ZEROCOPY_VERSION_1
) {
1782 cmn_err(CE_CONT
, "ill_capability_zerocopy_ack: "
1783 "unsupported ZEROCOPY sub-capability (version %d, "
1784 "expected %d)", zc_ic
->zerocopy_version
,
1785 ZEROCOPY_VERSION_1
);
1789 if (!dlcapabcheckqid(&zc_ic
->zerocopy_mid
, ill
->ill_lmod_rq
)) {
1790 ip1dbg(("ill_capability_zerocopy_ack: mid token for zerocopy "
1791 "capability isn't as expected; pass-thru module(s) "
1792 "detected, discarding capability\n"));
1796 if ((zc_ic
->zerocopy_flags
& DL_CAPAB_VMSAFE_MEM
) != 0) {
1797 if (*ill_zerocopy_capab
== NULL
) {
1798 *ill_zerocopy_capab
=
1799 kmem_zalloc(sizeof (ill_zerocopy_capab_t
),
1802 if (*ill_zerocopy_capab
== NULL
) {
1803 cmn_err(CE_WARN
, "ill_capability_zerocopy_ack: "
1804 "could not enable Zero-copy version %d "
1805 "for %s (ENOMEM)\n", ZEROCOPY_VERSION_1
,
1811 ip1dbg(("ill_capability_zerocopy_ack: interface %s "
1812 "supports Zero-copy version %d\n", ill
->ill_name
,
1813 ZEROCOPY_VERSION_1
));
1815 (*ill_zerocopy_capab
)->ill_zerocopy_version
=
1816 zc_ic
->zerocopy_version
;
1817 (*ill_zerocopy_capab
)->ill_zerocopy_flags
=
1818 zc_ic
->zerocopy_flags
;
1820 ill
->ill_capabilities
|= ILL_CAPAB_ZEROCOPY
;
1825 size
= sizeof (dl_capability_req_t
) +
1826 sizeof (dl_capability_sub_t
) +
1827 sizeof (dl_capab_zerocopy_t
);
1829 if ((nmp
= ip_dlpi_alloc(size
, DL_CAPABILITY_REQ
)) == NULL
) {
1830 cmn_err(CE_WARN
, "ill_capability_zerocopy_ack: "
1831 "could not enable zerocopy for %s (ENOMEM)\n",
1837 /* initialize dl_capability_req_t */
1838 oc
= (dl_capability_req_t
*)rptr
;
1839 oc
->dl_sub_offset
= sizeof (dl_capability_req_t
);
1840 oc
->dl_sub_length
= sizeof (dl_capability_sub_t
) +
1841 sizeof (dl_capab_zerocopy_t
);
1842 rptr
+= sizeof (dl_capability_req_t
);
1844 /* initialize dl_capability_sub_t */
1845 bcopy(isub
, rptr
, sizeof (*isub
));
1846 rptr
+= sizeof (*isub
);
1848 /* initialize dl_capab_zerocopy_t */
1849 zc_oc
= (dl_capab_zerocopy_t
*)rptr
;
1852 ip1dbg(("ill_capability_zerocopy_ack: asking interface %s "
1853 "to enable zero-copy version %d\n", ill
->ill_name
,
1854 ZEROCOPY_VERSION_1
));
1856 /* set VMSAFE_MEM flag */
1857 zc_oc
->zerocopy_flags
|= DL_CAPAB_VMSAFE_MEM
;
1859 /* nmp points to a DL_CAPABILITY_REQ message to enable zcopy */
1860 ill_capability_send(ill
, nmp
);
1865 ill_capability_zerocopy_reset_fill(ill_t
*ill
, mblk_t
*mp
)
1867 dl_capab_zerocopy_t
*zerocopy_subcap
;
1868 dl_capability_sub_t
*dl_subcap
;
1870 if (!(ill
->ill_capabilities
& ILL_CAPAB_ZEROCOPY
))
1873 ASSERT(ill
->ill_zerocopy_capab
!= NULL
);
1875 dl_subcap
= (dl_capability_sub_t
*)mp
->b_wptr
;
1876 dl_subcap
->dl_cap
= DL_CAPAB_ZEROCOPY
;
1877 dl_subcap
->dl_length
= sizeof (*zerocopy_subcap
);
1879 zerocopy_subcap
= (dl_capab_zerocopy_t
*)(dl_subcap
+ 1);
1880 zerocopy_subcap
->zerocopy_version
=
1881 ill
->ill_zerocopy_capab
->ill_zerocopy_version
;
1882 zerocopy_subcap
->zerocopy_flags
= 0;
1884 mp
->b_wptr
+= sizeof (*dl_subcap
) + sizeof (*zerocopy_subcap
);
1889 * Refer to dld.h for more information regarding the purpose and usage
1890 * of this capability.
1893 ill_capability_dld_ack(ill_t
*ill
, mblk_t
*mp
, dl_capability_sub_t
*isub
)
1895 dl_capab_dld_t
*dld_ic
, dld
;
1896 uint_t sub_dl_cap
= isub
->dl_cap
;
1898 ill_dld_capab_t
*idc
;
1900 ASSERT(IAM_WRITER_ILL(ill
));
1901 ASSERT(sub_dl_cap
== DL_CAPAB_DLD
);
1904 * Note: range checks here are not absolutely sufficient to
1905 * make us robust against malformed messages sent by drivers;
1906 * this is in keeping with the rest of IP's dlpi handling.
1907 * (Remember, it's coming from something else in the kernel
1910 capend
= (uint8_t *)(isub
+ 1) + isub
->dl_length
;
1911 if (capend
> mp
->b_wptr
) {
1912 cmn_err(CE_WARN
, "ill_capability_dld_ack: "
1913 "malformed sub-capability too long for mblk");
1916 dld_ic
= (dl_capab_dld_t
*)(isub
+ 1);
1917 if (dld_ic
->dld_version
!= DLD_CURRENT_VERSION
) {
1918 cmn_err(CE_CONT
, "ill_capability_dld_ack: "
1919 "unsupported DLD sub-capability (version %d, "
1920 "expected %d)", dld_ic
->dld_version
,
1921 DLD_CURRENT_VERSION
);
1924 if (!dlcapabcheckqid(&dld_ic
->dld_mid
, ill
->ill_lmod_rq
)) {
1925 ip1dbg(("ill_capability_dld_ack: mid token for dld "
1926 "capability isn't as expected; pass-thru module(s) "
1927 "detected, discarding capability\n"));
1932 * Copy locally to ensure alignment.
1934 bcopy(dld_ic
, &dld
, sizeof (dl_capab_dld_t
));
1936 if ((idc
= ill
->ill_dld_capab
) == NULL
) {
1937 idc
= kmem_zalloc(sizeof (ill_dld_capab_t
), KM_NOSLEEP
);
1939 cmn_err(CE_WARN
, "ill_capability_dld_ack: "
1940 "could not enable DLD version %d "
1941 "for %s (ENOMEM)\n", DLD_CURRENT_VERSION
,
1945 ill
->ill_dld_capab
= idc
;
1947 idc
->idc_capab_df
= (ip_capab_func_t
)dld
.dld_capab
;
1948 idc
->idc_capab_dh
= (void *)dld
.dld_capab_handle
;
1949 ip1dbg(("ill_capability_dld_ack: interface %s "
1950 "supports DLD version %d\n", ill
->ill_name
, DLD_CURRENT_VERSION
));
1952 ill_capability_dld_enable(ill
);
1956 * Typically capability negotiation between IP and the driver happens via
1957 * DLPI message exchange. However GLD also offers a direct function call
1958 * mechanism to exchange the DLD_DIRECT_CAPAB and DLD_POLL_CAPAB capabilities,
1959 * But arbitrary function calls into IP or GLD are not permitted, since both
1960 * of them are protected by their own perimeter mechanism. The perimeter can
1961 * be viewed as a coarse lock or serialization mechanism. The hierarchy of
1962 * these perimeters is IP -> MAC. Thus for example to enable the squeue
1963 * polling, IP needs to enter its perimeter, then call ill_mac_perim_enter
1964 * to enter the mac perimeter and then do the direct function calls into
1965 * GLD to enable squeue polling. The ring related callbacks from the mac into
1966 * the stack to add, bind, quiesce, restart or cleanup a ring are all
1967 * protected by the mac perimeter.
1970 ill_mac_perim_enter(ill_t
*ill
, mac_perim_handle_t
*mphp
)
1972 ill_dld_capab_t
*idc
= ill
->ill_dld_capab
;
1975 err
= idc
->idc_capab_df(idc
->idc_capab_dh
, DLD_CAPAB_PERIM
, mphp
,
1981 ill_mac_perim_exit(ill_t
*ill
, mac_perim_handle_t mph
)
1983 ill_dld_capab_t
*idc
= ill
->ill_dld_capab
;
1986 err
= idc
->idc_capab_df(idc
->idc_capab_dh
, DLD_CAPAB_PERIM
, mph
,
1992 ill_mac_perim_held(ill_t
*ill
)
1994 ill_dld_capab_t
*idc
= ill
->ill_dld_capab
;
1996 return (idc
->idc_capab_df(idc
->idc_capab_dh
, DLD_CAPAB_PERIM
, NULL
,
2001 ill_capability_direct_enable(ill_t
*ill
)
2003 ill_dld_capab_t
*idc
= ill
->ill_dld_capab
;
2004 ill_dld_direct_t
*idd
= &idc
->idc_direct
;
2005 dld_capab_direct_t direct
;
2008 ASSERT(!ill
->ill_isv6
&& IAM_WRITER_ILL(ill
));
2010 bzero(&direct
, sizeof (direct
));
2011 direct
.di_rx_cf
= (uintptr_t)ip_input
;
2012 direct
.di_rx_ch
= ill
;
2014 rc
= idc
->idc_capab_df(idc
->idc_capab_dh
, DLD_CAPAB_DIRECT
, &direct
,
2017 idd
->idd_tx_df
= (ip_dld_tx_t
)direct
.di_tx_df
;
2018 idd
->idd_tx_dh
= direct
.di_tx_dh
;
2019 idd
->idd_tx_cb_df
= (ip_dld_callb_t
)direct
.di_tx_cb_df
;
2020 idd
->idd_tx_cb_dh
= direct
.di_tx_cb_dh
;
2021 idd
->idd_tx_fctl_df
= (ip_dld_fctl_t
)direct
.di_tx_fctl_df
;
2022 idd
->idd_tx_fctl_dh
= direct
.di_tx_fctl_dh
;
2023 ASSERT(idd
->idd_tx_cb_df
!= NULL
);
2024 ASSERT(idd
->idd_tx_fctl_df
!= NULL
);
2025 ASSERT(idd
->idd_tx_df
!= NULL
);
2027 * One time registration of flow enable callback function
2029 ill
->ill_flownotify_mh
= idd
->idd_tx_cb_df(idd
->idd_tx_cb_dh
,
2030 ill_flow_enable
, ill
);
2031 ill
->ill_capabilities
|= ILL_CAPAB_DLD_DIRECT
;
2032 DTRACE_PROBE1(direct_on
, (ill_t
*), ill
);
2034 cmn_err(CE_WARN
, "warning: could not enable DIRECT "
2035 "capability, rc = %d\n", rc
);
2036 DTRACE_PROBE2(direct_off
, (ill_t
*), ill
, (int), rc
);
2041 ill_capability_poll_enable(ill_t
*ill
)
2043 ill_dld_capab_t
*idc
= ill
->ill_dld_capab
;
2044 dld_capab_poll_t poll
;
2047 ASSERT(!ill
->ill_isv6
&& IAM_WRITER_ILL(ill
));
2049 bzero(&poll
, sizeof (poll
));
2050 poll
.poll_ring_add_cf
= (uintptr_t)ip_squeue_add_ring
;
2051 poll
.poll_ring_remove_cf
= (uintptr_t)ip_squeue_clean_ring
;
2052 poll
.poll_ring_quiesce_cf
= (uintptr_t)ip_squeue_quiesce_ring
;
2053 poll
.poll_ring_restart_cf
= (uintptr_t)ip_squeue_restart_ring
;
2054 poll
.poll_ring_bind_cf
= (uintptr_t)ip_squeue_bind_ring
;
2055 poll
.poll_ring_ch
= ill
;
2056 rc
= idc
->idc_capab_df(idc
->idc_capab_dh
, DLD_CAPAB_POLL
, &poll
,
2059 ill
->ill_capabilities
|= ILL_CAPAB_DLD_POLL
;
2060 DTRACE_PROBE1(poll_on
, (ill_t
*), ill
);
2062 ip1dbg(("warning: could not enable POLL "
2063 "capability, rc = %d\n", rc
));
2064 DTRACE_PROBE2(poll_off
, (ill_t
*), ill
, (int), rc
);
2069 * Enable the LSO capability.
2072 ill_capability_lso_enable(ill_t
*ill
)
2074 ill_dld_capab_t
*idc
= ill
->ill_dld_capab
;
2075 dld_capab_lso_t lso
;
2078 ASSERT(!ill
->ill_isv6
&& IAM_WRITER_ILL(ill
));
2080 if (ill
->ill_lso_capab
== NULL
) {
2081 ill
->ill_lso_capab
= kmem_zalloc(sizeof (ill_lso_capab_t
),
2083 if (ill
->ill_lso_capab
== NULL
) {
2084 cmn_err(CE_WARN
, "ill_capability_lso_enable: "
2085 "could not enable LSO for %s (ENOMEM)\n",
2091 bzero(&lso
, sizeof (lso
));
2092 if ((rc
= idc
->idc_capab_df(idc
->idc_capab_dh
, DLD_CAPAB_LSO
, &lso
,
2093 DLD_ENABLE
)) == 0) {
2094 ill
->ill_lso_capab
->ill_lso_flags
= lso
.lso_flags
;
2095 ill
->ill_lso_capab
->ill_lso_max
= lso
.lso_max
;
2096 ill
->ill_capabilities
|= ILL_CAPAB_LSO
;
2097 ip1dbg(("ill_capability_lso_enable: interface %s "
2098 "has enabled LSO\n ", ill
->ill_name
));
2100 kmem_free(ill
->ill_lso_capab
, sizeof (ill_lso_capab_t
));
2101 ill
->ill_lso_capab
= NULL
;
2102 DTRACE_PROBE2(lso_off
, (ill_t
*), ill
, (int), rc
);
2107 ill_capability_dld_enable(ill_t
*ill
)
2109 mac_perim_handle_t mph
;
2111 ASSERT(IAM_WRITER_ILL(ill
));
2116 ill_mac_perim_enter(ill
, &mph
);
2117 if (!ill
->ill_isv6
) {
2118 ill_capability_direct_enable(ill
);
2119 ill_capability_poll_enable(ill
);
2120 ill_capability_lso_enable(ill
);
2122 ill
->ill_capabilities
|= ILL_CAPAB_DLD
;
2123 ill_mac_perim_exit(ill
, mph
);
2127 ill_capability_dld_disable(ill_t
*ill
)
2129 ill_dld_capab_t
*idc
;
2130 ill_dld_direct_t
*idd
;
2131 mac_perim_handle_t mph
;
2133 ASSERT(IAM_WRITER_ILL(ill
));
2135 if (!(ill
->ill_capabilities
& ILL_CAPAB_DLD
))
2138 ill_mac_perim_enter(ill
, &mph
);
2140 idc
= ill
->ill_dld_capab
;
2141 if ((ill
->ill_capabilities
& ILL_CAPAB_DLD_DIRECT
) != 0) {
2143 * For performance we avoid locks in the transmit data path
2144 * and don't maintain a count of the number of threads using
2145 * direct calls. Thus some threads could be using direct
2146 * transmit calls to GLD, even after the capability mechanism
2147 * turns it off. This is still safe since the handles used in
2148 * the direct calls continue to be valid until the unplumb is
2149 * completed. Remove the callback that was added (1-time) at
2150 * capab enable time.
2152 mutex_enter(&ill
->ill_lock
);
2153 ill
->ill_capabilities
&= ~ILL_CAPAB_DLD_DIRECT
;
2154 mutex_exit(&ill
->ill_lock
);
2155 if (ill
->ill_flownotify_mh
!= NULL
) {
2156 idd
= &idc
->idc_direct
;
2157 idd
->idd_tx_cb_df(idd
->idd_tx_cb_dh
, NULL
,
2158 ill
->ill_flownotify_mh
);
2159 ill
->ill_flownotify_mh
= NULL
;
2161 (void) idc
->idc_capab_df(idc
->idc_capab_dh
, DLD_CAPAB_DIRECT
,
2165 if ((ill
->ill_capabilities
& ILL_CAPAB_DLD_POLL
) != 0) {
2166 ill
->ill_capabilities
&= ~ILL_CAPAB_DLD_POLL
;
2167 ip_squeue_clean_all(ill
);
2168 (void) idc
->idc_capab_df(idc
->idc_capab_dh
, DLD_CAPAB_POLL
,
2172 if ((ill
->ill_capabilities
& ILL_CAPAB_LSO
) != 0) {
2173 ASSERT(ill
->ill_lso_capab
!= NULL
);
2175 * Clear the capability flag for LSO but retain the
2176 * ill_lso_capab structure since it's possible that another
2177 * thread is still referring to it. The structure only gets
2178 * deallocated when we destroy the ill.
2181 ill
->ill_capabilities
&= ~ILL_CAPAB_LSO
;
2182 (void) idc
->idc_capab_df(idc
->idc_capab_dh
, DLD_CAPAB_LSO
,
2186 ill
->ill_capabilities
&= ~ILL_CAPAB_DLD
;
2187 ill_mac_perim_exit(ill
, mph
);
2191 * Capability Negotiation protocol
2193 * We don't wait for DLPI capability operations to finish during interface
2194 * bringup or teardown. Doing so would introduce more asynchrony and the
2195 * interface up/down operations will need multiple return and restarts.
2196 * Instead the 'ipsq_current_ipif' of the ipsq is not cleared as long as
2197 * the 'ill_dlpi_deferred' chain is non-empty. This ensures that the next
2198 * exclusive operation won't start until the DLPI operations of the previous
2199 * exclusive operation complete.
2201 * The capability state machine is shown below.
2203 * state next state event, action
2205 * IDCS_UNKNOWN IDCS_PROBE_SENT ill_capability_probe
2206 * IDCS_PROBE_SENT IDCS_OK ill_capability_ack
2207 * IDCS_PROBE_SENT IDCS_FAILED ip_rput_dlpi_writer (nack)
2208 * IDCS_OK IDCS_RENEG Receipt of DL_NOTE_CAPAB_RENEG
2209 * IDCS_OK IDCS_RESET_SENT ill_capability_reset
2210 * IDCS_RESET_SENT IDCS_UNKNOWN ill_capability_ack_thr
2211 * IDCS_RENEG IDCS_PROBE_SENT ill_capability_ack_thr ->
2212 * ill_capability_probe.
2216 * Dedicated thread started from ip_stack_init that handles capability
2217 * disable. This thread ensures the taskq dispatch does not fail by waiting
2218 * for resources using TQ_SLEEP. The taskq mechanism is used to ensure
2219 * that direct calls to DLD are done in a cv_waitable context.
2222 ill_taskq_dispatch(ip_stack_t
*ipst
)
2224 callb_cpr_t cprinfo
;
2228 (void) snprintf(name
, sizeof (name
), "ill_taskq_dispatch_%d",
2229 ipst
->ips_netstack
->netstack_stackid
);
2230 CALLB_CPR_INIT(&cprinfo
, &ipst
->ips_capab_taskq_lock
, callb_generic_cpr
,
2232 mutex_enter(&ipst
->ips_capab_taskq_lock
);
2235 mp
= ipst
->ips_capab_taskq_head
;
2236 while (mp
!= NULL
) {
2237 ipst
->ips_capab_taskq_head
= mp
->b_next
;
2238 if (ipst
->ips_capab_taskq_head
== NULL
)
2239 ipst
->ips_capab_taskq_tail
= NULL
;
2240 mutex_exit(&ipst
->ips_capab_taskq_lock
);
2243 VERIFY(taskq_dispatch(system_taskq
,
2244 ill_capability_ack_thr
, mp
, TQ_SLEEP
) != 0);
2245 mutex_enter(&ipst
->ips_capab_taskq_lock
);
2246 mp
= ipst
->ips_capab_taskq_head
;
2249 if (ipst
->ips_capab_taskq_quit
)
2251 CALLB_CPR_SAFE_BEGIN(&cprinfo
);
2252 cv_wait(&ipst
->ips_capab_taskq_cv
, &ipst
->ips_capab_taskq_lock
);
2253 CALLB_CPR_SAFE_END(&cprinfo
, &ipst
->ips_capab_taskq_lock
);
2255 VERIFY(ipst
->ips_capab_taskq_head
== NULL
);
2256 VERIFY(ipst
->ips_capab_taskq_tail
== NULL
);
2257 CALLB_CPR_EXIT(&cprinfo
);
2262 * Consume a new-style hardware capabilities negotiation ack.
2263 * Called via taskq on receipt of DL_CAPABILITY_ACK.
2266 ill_capability_ack_thr(void *arg
)
2269 dl_capability_ack_t
*capp
;
2270 dl_capability_sub_t
*subp
, *endp
;
2274 ill
= (ill_t
*)mp
->b_prev
;
2277 VERIFY(ipsq_enter(ill
, B_FALSE
, CUR_OP
) == B_TRUE
);
2279 if (ill
->ill_dlpi_capab_state
== IDCS_RESET_SENT
||
2280 ill
->ill_dlpi_capab_state
== IDCS_RENEG
) {
2282 * We have received the ack for our DL_CAPAB reset request.
2283 * There isnt' anything in the message that needs processing.
2284 * All message based capabilities have been disabled, now
2285 * do the function call based capability disable.
2287 reneg
= ill
->ill_dlpi_capab_state
== IDCS_RENEG
;
2288 ill_capability_dld_disable(ill
);
2289 ill
->ill_dlpi_capab_state
= IDCS_UNKNOWN
;
2291 ill_capability_probe(ill
);
2295 if (ill
->ill_dlpi_capab_state
== IDCS_PROBE_SENT
)
2296 ill
->ill_dlpi_capab_state
= IDCS_OK
;
2298 capp
= (dl_capability_ack_t
*)mp
->b_rptr
;
2300 if (capp
->dl_sub_length
== 0) {
2301 /* no new-style capabilities */
2305 /* make sure the driver supplied correct dl_sub_length */
2306 if ((sizeof (*capp
) + capp
->dl_sub_length
) > MBLKL(mp
)) {
2307 ip0dbg(("ill_capability_ack: bad DL_CAPABILITY_ACK, "
2308 "invalid dl_sub_length (%d)\n", capp
->dl_sub_length
));
2312 #define SC(base, offset) (dl_capability_sub_t *)(((uchar_t *)(base))+(offset))
2314 * There are sub-capabilities. Process the ones we know about.
2315 * Loop until we don't have room for another sub-cap header..
2317 for (subp
= SC(capp
, capp
->dl_sub_offset
),
2318 endp
= SC(subp
, capp
->dl_sub_length
- sizeof (*subp
));
2320 subp
= SC(subp
, sizeof (dl_capability_sub_t
) + subp
->dl_length
)) {
2322 switch (subp
->dl_cap
) {
2323 case DL_CAPAB_ID_WRAPPER
:
2324 ill_capability_id_ack(ill
, mp
, subp
);
2327 ill_capability_dispatch(ill
, mp
, subp
);
2334 ill_capability_done(ill
);
2335 ipsq_exit(ill
->ill_phyint
->phyint_ipsq
);
2339 * This needs to be started in a taskq thread to provide a cv_waitable
2343 ill_capability_ack(ill_t
*ill
, mblk_t
*mp
)
2345 ip_stack_t
*ipst
= ill
->ill_ipst
;
2347 mp
->b_prev
= (mblk_t
*)ill
;
2348 ASSERT(mp
->b_next
== NULL
);
2350 if (taskq_dispatch(system_taskq
, ill_capability_ack_thr
, mp
,
2355 * The taskq dispatch failed. Signal the ill_taskq_dispatch thread
2356 * which will do the dispatch using TQ_SLEEP to guarantee success.
2358 mutex_enter(&ipst
->ips_capab_taskq_lock
);
2359 if (ipst
->ips_capab_taskq_head
== NULL
) {
2360 ASSERT(ipst
->ips_capab_taskq_tail
== NULL
);
2361 ipst
->ips_capab_taskq_head
= mp
;
2363 ipst
->ips_capab_taskq_tail
->b_next
= mp
;
2365 ipst
->ips_capab_taskq_tail
= mp
;
2367 cv_signal(&ipst
->ips_capab_taskq_cv
);
2368 mutex_exit(&ipst
->ips_capab_taskq_lock
);
2372 * This routine is called to scan the fragmentation reassembly table for
2373 * the specified ILL for any packets that are starting to smell.
2374 * dead_interval is the maximum time in seconds that will be tolerated. It
2375 * will either be the value specified in ip_g_frag_timeout, or zero if the
2376 * ILL is shutting down and it is time to blow everything off.
2378 * It returns the number of seconds (as a time_t) that the next frag timer
2379 * should be scheduled for, 0 meaning that the timer doesn't need to be
2380 * re-started. Note that the method of calculating next_timeout isn't
2381 * entirely accurate since time will flow between the time we grab
2382 * current_time and the time we schedule the next timeout. This isn't a
2383 * big problem since this is the timer for sending an ICMP reassembly time
2384 * exceeded messages, and it doesn't have to be exactly accurate.
2387 * sometimes called as writer, although this is not required.
2390 ill_frag_timeout(ill_t
*ill
, time_t dead_interval
)
2397 time_t current_time
= gethrestime_sec();
2398 time_t next_timeout
= 0;
2399 uint32_t hdr_length
;
2400 mblk_t
*send_icmp_head
;
2401 mblk_t
*send_icmp_head_v6
;
2402 ip_stack_t
*ipst
= ill
->ill_ipst
;
2403 ip_recv_attr_t iras
;
2405 bzero(&iras
, sizeof (iras
));
2407 iras
.ira_ill
= iras
.ira_rill
= ill
;
2408 iras
.ira_ruifindex
= ill
->ill_phyint
->phyint_ifindex
;
2409 iras
.ira_rifindex
= iras
.ira_ruifindex
;
2411 ipfb
= ill
->ill_frag_hash_tbl
;
2414 endp
= &ipfb
[ILL_FRAG_HASH_TBL_COUNT
];
2415 /* Walk the frag hash table. */
2416 for (; ipfb
< endp
; ipfb
++) {
2417 send_icmp_head
= NULL
;
2418 send_icmp_head_v6
= NULL
;
2419 mutex_enter(&ipfb
->ipfb_lock
);
2420 while ((ipf
= ipfb
->ipfb_ipf
) != 0) {
2421 time_t frag_time
= current_time
- ipf
->ipf_timestamp
;
2422 time_t frag_timeout
;
2424 if (frag_time
< dead_interval
) {
2426 * There are some outstanding fragments
2427 * that will timeout later. Make note of
2428 * the time so that we can reschedule the
2429 * next timeout appropriately.
2431 frag_timeout
= dead_interval
- frag_time
;
2432 if (next_timeout
== 0 ||
2433 frag_timeout
< next_timeout
) {
2434 next_timeout
= frag_timeout
;
2438 /* Time's up. Get it out of here. */
2439 hdr_length
= ipf
->ipf_nf_hdr_len
;
2440 ipfnext
= ipf
->ipf_hash_next
;
2442 ipfnext
->ipf_ptphn
= ipf
->ipf_ptphn
;
2443 *ipf
->ipf_ptphn
= ipfnext
;
2444 mp
= ipf
->ipf_mp
->b_cont
;
2445 for (; mp
; mp
= mp
->b_cont
) {
2446 /* Extra points for neatness. */
2447 IP_REASS_SET_START(mp
, 0);
2448 IP_REASS_SET_END(mp
, 0);
2450 mp
= ipf
->ipf_mp
->b_cont
;
2451 atomic_add_32(&ill
->ill_frag_count
, -ipf
->ipf_count
);
2452 ASSERT(ipfb
->ipfb_count
>= ipf
->ipf_count
);
2453 ipfb
->ipfb_count
-= ipf
->ipf_count
;
2454 ASSERT(ipfb
->ipfb_frag_pkts
> 0);
2455 ipfb
->ipfb_frag_pkts
--;
2457 * We do not send any icmp message from here because
2458 * we currently are holding the ipfb_lock for this
2459 * hash chain. If we try and send any icmp messages
2460 * from here we may end up via a put back into ip
2461 * trying to get the same lock, causing a recursive
2462 * mutex panic. Instead we build a list and send all
2463 * the icmp messages after we have dropped the lock.
2465 if (ill
->ill_isv6
) {
2466 if (hdr_length
!= 0) {
2467 mp
->b_next
= send_icmp_head_v6
;
2468 send_icmp_head_v6
= mp
;
2473 if (hdr_length
!= 0) {
2474 mp
->b_next
= send_icmp_head
;
2475 send_icmp_head
= mp
;
2480 BUMP_MIB(ill
->ill_ip_mib
, ipIfStatsReasmFails
);
2481 ip_drop_input("ipIfStatsReasmFails", ipf
->ipf_mp
, ill
);
2484 mutex_exit(&ipfb
->ipfb_lock
);
2486 * Now need to send any icmp messages that we delayed from
2489 while (send_icmp_head_v6
!= NULL
) {
2492 mp
= send_icmp_head_v6
;
2493 send_icmp_head_v6
= send_icmp_head_v6
->b_next
;
2495 ip6h
= (ip6_t
*)mp
->b_rptr
;
2498 * This will result in an incorrect ALL_ZONES zoneid
2499 * for multicast packets, but we
2500 * don't send ICMP errors for those in any case.
2503 ipif_lookup_addr_zoneid_v6(&ip6h
->ip6_dst
,
2505 ip_drop_input("ICMP_TIME_EXCEEDED reass", mp
, ill
);
2506 icmp_time_exceeded_v6(mp
,
2507 ICMP_REASSEMBLY_TIME_EXCEEDED
, B_FALSE
,
2509 ASSERT(!(iras
.ira_flags
& IRAF_IPSEC_SECURE
));
2511 while (send_icmp_head
!= NULL
) {
2514 mp
= send_icmp_head
;
2515 send_icmp_head
= send_icmp_head
->b_next
;
2518 dst
= ((ipha_t
*)mp
->b_rptr
)->ipha_dst
;
2520 iras
.ira_flags
= IRAF_IS_IPV4
;
2522 * This will result in an incorrect ALL_ZONES zoneid
2523 * for broadcast and multicast packets, but we
2524 * don't send ICMP errors for those in any case.
2526 iras
.ira_zoneid
= ipif_lookup_addr_zoneid(dst
,
2528 ip_drop_input("ICMP_TIME_EXCEEDED reass", mp
, ill
);
2529 icmp_time_exceeded(mp
,
2530 ICMP_REASSEMBLY_TIME_EXCEEDED
, &iras
);
2531 ASSERT(!(iras
.ira_flags
& IRAF_IPSEC_SECURE
));
2535 * A non-dying ILL will use the return value to decide whether to
2536 * restart the frag timer, and for how long.
2538 return (next_timeout
);
2542 * This routine is called when the approximate count of mblk memory used
2543 * for the specified ILL has exceeded max_count.
2546 ill_frag_prune(ill_t
*ill
, uint_t max_count
)
2554 * If we are here within ip_min_frag_prune_time msecs remove
2555 * ill_frag_free_num_pkts oldest packets from each bucket and increment
2556 * ill_frag_free_num_pkts.
2558 mutex_enter(&ill
->ill_lock
);
2559 now
= ddi_get_lbolt();
2560 if (TICK_TO_MSEC(now
- ill
->ill_last_frag_clean_time
) <=
2561 (ip_min_frag_prune_time
!= 0 ?
2562 ip_min_frag_prune_time
: msec_per_tick
)) {
2564 ill
->ill_frag_free_num_pkts
++;
2567 ill
->ill_frag_free_num_pkts
= 0;
2569 ill
->ill_last_frag_clean_time
= now
;
2570 mutex_exit(&ill
->ill_lock
);
2573 * free ill_frag_free_num_pkts oldest packets from each bucket.
2575 if (ill
->ill_frag_free_num_pkts
!= 0) {
2578 for (ix
= 0; ix
< ILL_FRAG_HASH_TBL_COUNT
; ix
++) {
2579 ipfb
= &ill
->ill_frag_hash_tbl
[ix
];
2580 mutex_enter(&ipfb
->ipfb_lock
);
2581 if (ipfb
->ipfb_ipf
!= NULL
) {
2582 ill_frag_free_pkts(ill
, ipfb
, ipfb
->ipfb_ipf
,
2583 ill
->ill_frag_free_num_pkts
);
2585 mutex_exit(&ipfb
->ipfb_lock
);
2589 * While the reassembly list for this ILL is too big, prune a fragment
2590 * queue by age, oldest first.
2592 while (ill
->ill_frag_count
> max_count
) {
2594 ipfb_t
*oipfb
= NULL
;
2595 uint_t oldest
= UINT_MAX
;
2598 for (ix
= 0; ix
< ILL_FRAG_HASH_TBL_COUNT
; ix
++) {
2599 ipfb
= &ill
->ill_frag_hash_tbl
[ix
];
2600 mutex_enter(&ipfb
->ipfb_lock
);
2601 ipf
= ipfb
->ipfb_ipf
;
2602 if (ipf
!= NULL
&& ipf
->ipf_gen
< oldest
) {
2603 oldest
= ipf
->ipf_gen
;
2606 count
+= ipfb
->ipfb_count
;
2607 mutex_exit(&ipfb
->ipfb_lock
);
2612 if (count
<= max_count
)
2613 return; /* Somebody beat us to it, nothing to do */
2614 mutex_enter(&oipfb
->ipfb_lock
);
2615 ipf
= oipfb
->ipfb_ipf
;
2617 ill_frag_free_pkts(ill
, oipfb
, ipf
, 1);
2619 mutex_exit(&oipfb
->ipfb_lock
);
2624 * free 'free_cnt' fragmented packets starting at ipf.
2627 ill_frag_free_pkts(ill_t
*ill
, ipfb_t
*ipfb
, ipf_t
*ipf
, int free_cnt
)
2632 ipf_t
**ipfp
= ipf
->ipf_ptphn
;
2634 ASSERT(MUTEX_HELD(&ipfb
->ipfb_lock
));
2635 ASSERT(ipfp
!= NULL
);
2636 ASSERT(ipf
!= NULL
);
2638 while (ipf
!= NULL
&& free_cnt
-- > 0) {
2639 count
= ipf
->ipf_count
;
2641 ipf
= ipf
->ipf_hash_next
;
2642 for (tmp
= mp
; tmp
; tmp
= tmp
->b_cont
) {
2643 IP_REASS_SET_START(tmp
, 0);
2644 IP_REASS_SET_END(tmp
, 0);
2646 atomic_add_32(&ill
->ill_frag_count
, -count
);
2647 ASSERT(ipfb
->ipfb_count
>= count
);
2648 ipfb
->ipfb_count
-= count
;
2649 ASSERT(ipfb
->ipfb_frag_pkts
> 0);
2650 ipfb
->ipfb_frag_pkts
--;
2651 BUMP_MIB(ill
->ill_ip_mib
, ipIfStatsReasmFails
);
2652 ip_drop_input("ipIfStatsReasmFails", mp
, ill
);
2657 ipf
->ipf_ptphn
= ipfp
;
2662 * Helper function for ill_forward_set().
2665 ill_forward_set_on_ill(ill_t
*ill
, boolean_t enable
)
2667 ip_stack_t
*ipst
= ill
->ill_ipst
;
2669 ASSERT(IAM_WRITER_ILL(ill
) || RW_READ_HELD(&ipst
->ips_ill_g_lock
));
2671 ip1dbg(("ill_forward_set: %s %s forwarding on %s",
2672 (enable
? "Enabling" : "Disabling"),
2673 (ill
->ill_isv6
? "IPv6" : "IPv4"), ill
->ill_name
));
2674 mutex_enter(&ill
->ill_lock
);
2676 ill
->ill_flags
|= ILLF_ROUTER
;
2678 ill
->ill_flags
&= ~ILLF_ROUTER
;
2679 mutex_exit(&ill
->ill_lock
);
2681 ill_set_nce_router_flags(ill
, enable
);
2682 /* Notify routing socket listeners of this change. */
2683 if (ill
->ill_ipif
!= NULL
)
2684 ip_rts_ifmsg(ill
->ill_ipif
, RTSQ_DEFAULT
);
2688 * Set an ill's ILLF_ROUTER flag appropriately. Send up RTS_IFINFO routing
2689 * socket messages for each interface whose flags we change.
2692 ill_forward_set(ill_t
*ill
, boolean_t enable
)
2694 ipmp_illgrp_t
*illg
;
2695 ip_stack_t
*ipst
= ill
->ill_ipst
;
2697 ASSERT(IAM_WRITER_ILL(ill
) || RW_READ_HELD(&ipst
->ips_ill_g_lock
));
2699 if ((enable
&& (ill
->ill_flags
& ILLF_ROUTER
)) ||
2700 (!enable
&& !(ill
->ill_flags
& ILLF_ROUTER
)))
2703 if (IS_LOOPBACK(ill
))
2706 if (enable
&& ill
->ill_allowed_ips_cnt
> 0)
2709 if (IS_IPMP(ill
) || IS_UNDER_IPMP(ill
)) {
2711 * Update all of the interfaces in the group.
2713 illg
= ill
->ill_grp
;
2714 ill
= list_head(&illg
->ig_if
);
2715 for (; ill
!= NULL
; ill
= list_next(&illg
->ig_if
, ill
))
2716 ill_forward_set_on_ill(ill
, enable
);
2719 * Update the IPMP meta-interface.
2721 ill_forward_set_on_ill(ipmp_illgrp_ipmp_ill(illg
), enable
);
2725 ill_forward_set_on_ill(ill
, enable
);
2730 * Based on the ILLF_ROUTER flag of an ill, make sure all local nce's for
2731 * addresses assigned to the ill have the NCE_F_ISROUTER flag appropriately
2735 ill_set_nce_router_flags(ill_t
*ill
, boolean_t enable
)
2741 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
; ipif
= ipif
->ipif_next
) {
2743 * NOTE: we match across the illgrp because nce's for
2744 * addresses on IPMP interfaces have an nce_ill that points to
2745 * the bound underlying ill.
2747 nce
= nce_lookup_v6(ill
, &ipif
->ipif_v6lcl_addr
);
2749 ncec
= nce
->nce_common
;
2750 mutex_enter(&ncec
->ncec_lock
);
2752 ncec
->ncec_flags
|= NCE_F_ISROUTER
;
2754 ncec
->ncec_flags
&= ~NCE_F_ISROUTER
;
2755 mutex_exit(&ncec
->ncec_lock
);
2762 * Intializes the context structure and returns the first ill in the list
2763 * cuurently start_list and end_list can have values:
2764 * MAX_G_HEADS Traverse both IPV4 and IPV6 lists.
2765 * IP_V4_G_HEAD Traverse IPV4 list only.
2766 * IP_V6_G_HEAD Traverse IPV6 list only.
2770 * We don't check for CONDEMNED ills here. Caller must do that if
2771 * necessary under the ill lock.
2774 ill_first(int start_list
, int end_list
, ill_walk_context_t
*ctx
,
2779 avl_tree_t
*avl_tree
;
2781 ASSERT(RW_LOCK_HELD(&ipst
->ips_ill_g_lock
));
2782 ASSERT(end_list
<= MAX_G_HEADS
&& start_list
>= 0);
2785 * setup the lists to search
2787 if (end_list
!= MAX_G_HEADS
) {
2788 ctx
->ctx_current_list
= start_list
;
2789 ctx
->ctx_last_list
= end_list
;
2791 ctx
->ctx_last_list
= MAX_G_HEADS
- 1;
2792 ctx
->ctx_current_list
= 0;
2795 while (ctx
->ctx_current_list
<= ctx
->ctx_last_list
) {
2796 ifp
= IP_VX_ILL_G_LIST(ctx
->ctx_current_list
, ipst
);
2797 if (ifp
!= (ill_if_t
*)
2798 &IP_VX_ILL_G_LIST(ctx
->ctx_current_list
, ipst
)) {
2799 avl_tree
= &ifp
->illif_avl_by_ppa
;
2800 ill
= avl_first(avl_tree
);
2802 * ill is guaranteed to be non NULL or ifp should have
2805 ASSERT(ill
!= NULL
);
2808 ctx
->ctx_current_list
++;
2815 * returns the next ill in the list. ill_first() must have been called
2816 * before calling ill_next() or bad things will happen.
2820 * We don't check for CONDEMNED ills here. Caller must do that if
2821 * necessary under the ill lock.
2824 ill_next(ill_walk_context_t
*ctx
, ill_t
*lastill
)
2828 ip_stack_t
*ipst
= lastill
->ill_ipst
;
2830 ASSERT(lastill
->ill_ifptr
!= (ill_if_t
*)
2831 &IP_VX_ILL_G_LIST(ctx
->ctx_current_list
, ipst
));
2832 if ((ill
= avl_walk(&lastill
->ill_ifptr
->illif_avl_by_ppa
, lastill
,
2833 AVL_AFTER
)) != NULL
) {
2837 /* goto next ill_ifp in the list. */
2838 ifp
= lastill
->ill_ifptr
->illif_next
;
2840 /* make sure not at end of circular list */
2842 (ill_if_t
*)&IP_VX_ILL_G_LIST(ctx
->ctx_current_list
, ipst
)) {
2843 if (++ctx
->ctx_current_list
> ctx
->ctx_last_list
)
2845 ifp
= IP_VX_ILL_G_LIST(ctx
->ctx_current_list
, ipst
);
2848 return (avl_first(&ifp
->illif_avl_by_ppa
));
2852 * Check interface name for correct format: [a-zA-Z]+[a-zA-Z0-9._]*[0-9]+
2853 * The final number (PPA) must not have any leading zeros. Upon success, a
2854 * pointer to the start of the PPA is returned; otherwise NULL is returned.
2857 ill_get_ppa_ptr(char *name
)
2859 int namelen
= strlen(name
);
2860 int end_ndx
= namelen
- 1;
2864 * Check that the first character is [a-zA-Z], and that the last
2865 * character is [0-9].
2867 if (namelen
== 0 || !isalpha(name
[0]) || !isdigit(name
[end_ndx
]))
2871 * Set `ppa_ndx' to the PPA start, and check for leading zeroes.
2873 for (ppa_ndx
= end_ndx
; ppa_ndx
> 0; ppa_ndx
--)
2874 if (!isdigit(name
[ppa_ndx
- 1]))
2877 if (name
[ppa_ndx
] == '0' && ppa_ndx
< end_ndx
)
2881 * Check that the intermediate characters are [a-z0-9.]
2883 for (i
= 1; i
< ppa_ndx
; i
++) {
2884 if (!isalpha(name
[i
]) && !isdigit(name
[i
]) &&
2885 name
[i
] != '.' && name
[i
] != '_') {
2890 return (name
+ ppa_ndx
);
2894 * use avl tree to locate the ill.
2897 ill_find_by_name(char *name
, boolean_t isv6
, ip_stack_t
*ipst
)
2899 char *ppa_ptr
= NULL
;
2910 list
= IP_V6_G_HEAD
;
2912 list
= IP_V4_G_HEAD
;
2914 if ((ppa_ptr
= ill_get_ppa_ptr(name
)) == NULL
) {
2918 len
= ppa_ptr
- name
+ 1;
2920 ppa
= stoi(&ppa_ptr
);
2922 ifp
= IP_VX_ILL_G_LIST(list
, ipst
);
2924 while (ifp
!= (ill_if_t
*)&IP_VX_ILL_G_LIST(list
, ipst
)) {
2926 * match is done on len - 1 as the name is not null
2927 * terminated it contains ppa in addition to the interface
2930 if ((ifp
->illif_name_len
== len
) &&
2931 bcmp(ifp
->illif_name
, name
, len
- 1) == 0) {
2934 ifp
= ifp
->illif_next
;
2938 if (ifp
== (ill_if_t
*)&IP_VX_ILL_G_LIST(list
, ipst
)) {
2940 * Even the interface type does not exist.
2945 ill
= avl_find(&ifp
->illif_avl_by_ppa
, (void *) &ppa
, NULL
);
2947 mutex_enter(&ill
->ill_lock
);
2948 if (ILL_CAN_LOOKUP(ill
)) {
2949 ill_refhold_locked(ill
);
2950 mutex_exit(&ill
->ill_lock
);
2953 mutex_exit(&ill
->ill_lock
);
2959 * comparison function for use with avl.
2962 ill_compare_ppa(const void *ppa_ptr
, const void *ill_ptr
)
2967 ASSERT(ppa_ptr
!= NULL
&& ill_ptr
!= NULL
);
2969 ppa
= *((uint_t
*)ppa_ptr
);
2970 ill_ppa
= ((const ill_t
*)ill_ptr
)->ill_ppa
;
2972 * We want the ill with the lowest ppa to be on the
2983 * remove an interface type from the global list.
2986 ill_delete_interface_type(ill_if_t
*interface
)
2988 ASSERT(interface
!= NULL
);
2989 ASSERT(avl_numnodes(&interface
->illif_avl_by_ppa
) == 0);
2991 avl_destroy(&interface
->illif_avl_by_ppa
);
2992 if (interface
->illif_ppa_arena
!= NULL
)
2993 vmem_destroy(interface
->illif_ppa_arena
);
3001 * remove ill from the global list.
3004 ill_glist_delete(ill_t
*ill
)
3011 ipst
= ill
->ill_ipst
;
3012 rw_enter(&ipst
->ips_ill_g_lock
, RW_WRITER
);
3015 * If the ill was never inserted into the AVL tree
3016 * we skip the if branch.
3018 if (ill
->ill_ifptr
!= NULL
) {
3020 * remove from AVL tree and free ppa number
3022 avl_remove(&ill
->ill_ifptr
->illif_avl_by_ppa
, ill
);
3024 if (ill
->ill_ifptr
->illif_ppa_arena
!= NULL
) {
3025 vmem_free(ill
->ill_ifptr
->illif_ppa_arena
,
3026 (void *)(uintptr_t)(ill
->ill_ppa
+1), 1);
3028 if (avl_numnodes(&ill
->ill_ifptr
->illif_avl_by_ppa
) == 0) {
3029 ill_delete_interface_type(ill
->ill_ifptr
);
3033 * Indicate ill is no longer in the list.
3035 ill
->ill_ifptr
= NULL
;
3036 ill
->ill_name_length
= 0;
3037 ill
->ill_name
[0] = '\0';
3038 ill
->ill_ppa
= UINT_MAX
;
3041 /* Generate one last event for this ill. */
3042 ill_nic_event_dispatch(ill
, 0, NE_UNPLUMB
, ill
->ill_name
,
3043 ill
->ill_name_length
);
3045 ASSERT(ill
->ill_phyint
!= NULL
);
3046 phyi
= ill
->ill_phyint
;
3047 ill
->ill_phyint
= NULL
;
3050 * ill_init allocates a phyint always to store the copy
3051 * of flags relevant to phyint. At that point in time, we could
3052 * not assign the name and hence phyint_illv4/v6 could not be
3053 * initialized. Later in ipif_set_values, we assign the name to
3054 * the ill, at which point in time we assign phyint_illv4/v6.
3055 * Thus we don't rely on phyint_illv6 to be initialized always.
3057 if (ill
->ill_flags
& ILLF_IPV6
)
3058 phyi
->phyint_illv6
= NULL
;
3060 phyi
->phyint_illv4
= NULL
;
3062 if (phyi
->phyint_illv4
!= NULL
|| phyi
->phyint_illv6
!= NULL
) {
3063 rw_exit(&ipst
->ips_ill_g_lock
);
3068 * There are no ills left on this phyint; pull it out of the phyint
3069 * avl trees, and free it.
3071 if (phyi
->phyint_ifindex
> 0) {
3072 avl_remove(&ipst
->ips_phyint_g_list
->phyint_list_avl_by_index
,
3074 avl_remove(&ipst
->ips_phyint_g_list
->phyint_list_avl_by_name
,
3077 rw_exit(&ipst
->ips_ill_g_lock
);
3083 * allocate a ppa, if the number of plumbed interfaces of this type are
3084 * less than ill_no_arena do a linear search to find a unused ppa.
3085 * When the number goes beyond ill_no_arena switch to using an arena.
3086 * Note: ppa value of zero cannot be allocated from vmem_arena as it
3087 * is the return value for an error condition, so allocation starts at one
3088 * and is decremented by one.
3091 ill_alloc_ppa(ill_if_t
*ifp
, ill_t
*ill
)
3097 if (ifp
->illif_ppa_arena
== NULL
&&
3098 (avl_numnodes(&ifp
->illif_avl_by_ppa
) + 1 > ill_no_arena
)) {
3102 ifp
->illif_ppa_arena
= vmem_create(ifp
->illif_name
,
3103 (void *)1, UINT_MAX
- 1, 1, NULL
, NULL
,
3104 NULL
, 0, VM_SLEEP
| VMC_IDENTIFIER
);
3105 /* allocate what has already been assigned */
3106 for (tmp_ill
= avl_first(&ifp
->illif_avl_by_ppa
);
3107 tmp_ill
!= NULL
; tmp_ill
= avl_walk(&ifp
->illif_avl_by_ppa
,
3108 tmp_ill
, AVL_AFTER
)) {
3109 ppa
= (int)(uintptr_t)vmem_xalloc(ifp
->illif_ppa_arena
,
3111 1, /* align/quantum */
3115 (void *)((uintptr_t)tmp_ill
->ill_ppa
+ 1),
3117 (void *)((uintptr_t)tmp_ill
->ill_ppa
+ 2),
3118 VM_NOSLEEP
|VM_FIRSTFIT
);
3120 ip1dbg(("ill_alloc_ppa: ppa allocation"
3121 " failed while switching"));
3122 vmem_destroy(ifp
->illif_ppa_arena
);
3123 ifp
->illif_ppa_arena
= NULL
;
3129 if (ifp
->illif_ppa_arena
!= NULL
) {
3130 if (ill
->ill_ppa
== UINT_MAX
) {
3131 ppa
= (int)(uintptr_t)vmem_alloc(ifp
->illif_ppa_arena
,
3132 1, VM_NOSLEEP
|VM_FIRSTFIT
);
3135 ill
->ill_ppa
= --ppa
;
3137 ppa
= (int)(uintptr_t)vmem_xalloc(ifp
->illif_ppa_arena
,
3139 1, /* align/quantum */
3142 (void *)(uintptr_t)(ill
->ill_ppa
+ 1), /* minaddr */
3143 (void *)(uintptr_t)(ill
->ill_ppa
+ 2), /* maxaddr */
3144 VM_NOSLEEP
|VM_FIRSTFIT
);
3146 * Most likely the allocation failed because
3147 * the requested ppa was in use.
3156 * No arena is in use and not enough (>ill_no_arena) interfaces have
3157 * been plumbed to create one. Do a linear search to get a unused ppa.
3159 if (ill
->ill_ppa
== UINT_MAX
) {
3163 end
= start
= ill
->ill_ppa
;
3166 tmp_ill
= avl_find(&ifp
->illif_avl_by_ppa
, (void *)&start
, NULL
);
3167 while (tmp_ill
!= NULL
&& tmp_ill
->ill_ppa
== start
) {
3168 if (start
++ >= end
) {
3169 if (ill
->ill_ppa
== UINT_MAX
)
3174 tmp_ill
= avl_walk(&ifp
->illif_avl_by_ppa
, tmp_ill
, AVL_AFTER
);
3176 ill
->ill_ppa
= start
;
3181 * Insert ill into the list of configured ill's. Once this function completes,
3182 * the ill is globally visible and is available through lookups. More precisely
3183 * this happens after the caller drops the ill_g_lock.
3186 ill_glist_insert(ill_t
*ill
, char *name
, boolean_t isv6
)
3188 ill_if_t
*ill_interface
;
3189 avl_index_t where
= 0;
3193 boolean_t check_length
= B_FALSE
;
3194 ip_stack_t
*ipst
= ill
->ill_ipst
;
3196 ASSERT(RW_WRITE_HELD(&ipst
->ips_ill_g_lock
));
3198 name_length
= mi_strlen(name
) + 1;
3201 index
= IP_V6_G_HEAD
;
3203 index
= IP_V4_G_HEAD
;
3205 ill_interface
= IP_VX_ILL_G_LIST(index
, ipst
);
3207 * Search for interface type based on name
3209 while (ill_interface
!= (ill_if_t
*)&IP_VX_ILL_G_LIST(index
, ipst
)) {
3210 if ((ill_interface
->illif_name_len
== name_length
) &&
3211 (strcmp(ill_interface
->illif_name
, name
) == 0)) {
3214 ill_interface
= ill_interface
->illif_next
;
3218 * Interface type not found, create one.
3220 if (ill_interface
== (ill_if_t
*)&IP_VX_ILL_G_LIST(index
, ipst
)) {
3224 * allocate ill_if_t structure
3226 ill_interface
= (ill_if_t
*)mi_zalloc(sizeof (ill_if_t
));
3227 if (ill_interface
== NULL
) {
3231 (void) strcpy(ill_interface
->illif_name
, name
);
3232 ill_interface
->illif_name_len
= name_length
;
3234 avl_create(&ill_interface
->illif_avl_by_ppa
,
3235 ill_compare_ppa
, sizeof (ill_t
),
3236 offsetof(struct ill_s
, ill_avl_byppa
));
3239 * link the structure in the back to maintain order
3240 * of configuration for ifconfig output.
3242 ghead
= ipst
->ips_ill_g_heads
[index
];
3243 insque(ill_interface
, ghead
.ill_g_list_tail
);
3246 if (ill
->ill_ppa
== UINT_MAX
)
3247 check_length
= B_TRUE
;
3249 error
= ill_alloc_ppa(ill_interface
, ill
);
3251 if (avl_numnodes(&ill_interface
->illif_avl_by_ppa
) == 0)
3252 ill_delete_interface_type(ill
->ill_ifptr
);
3257 * When the ppa is choosen by the system, check that there is
3258 * enough space to insert ppa. if a specific ppa was passed in this
3259 * check is not required as the interface name passed in will have
3260 * the right ppa in it.
3264 * UINT_MAX - 1 should fit in 10 chars, alloc 12 chars.
3266 char buf
[sizeof (uint_t
) * 3];
3269 * convert ppa to string to calculate the amount of space
3270 * required for it in the name.
3272 numtos(ill
->ill_ppa
, buf
);
3274 /* Do we have enough space to insert ppa ? */
3276 if ((mi_strlen(name
) + mi_strlen(buf
) + 1) > LIFNAMSIZ
) {
3277 /* Free ppa and interface type struct */
3278 if (ill_interface
->illif_ppa_arena
!= NULL
) {
3279 vmem_free(ill_interface
->illif_ppa_arena
,
3280 (void *)(uintptr_t)(ill
->ill_ppa
+1), 1);
3282 if (avl_numnodes(&ill_interface
->illif_avl_by_ppa
) == 0)
3283 ill_delete_interface_type(ill
->ill_ifptr
);
3289 (void) sprintf(ill
->ill_name
, "%s%u", name
, ill
->ill_ppa
);
3290 ill
->ill_name_length
= mi_strlen(ill
->ill_name
) + 1;
3292 (void) avl_find(&ill_interface
->illif_avl_by_ppa
, &ill
->ill_ppa
,
3294 ill
->ill_ifptr
= ill_interface
;
3295 avl_insert(&ill_interface
->illif_avl_by_ppa
, ill
, where
);
3297 ill_phyint_reinit(ill
);
3301 /* Initialize the per phyint ipsq used for serialization */
3303 ipsq_init(ill_t
*ill
, boolean_t enter
)
3308 if ((ipsq
= kmem_zalloc(sizeof (ipsq_t
), KM_NOSLEEP
)) == NULL
)
3311 ill
->ill_phyint
->phyint_ipsq
= ipsq
;
3312 ipx
= ipsq
->ipsq_xop
= &ipsq
->ipsq_ownxop
;
3313 ipx
->ipx_ipsq
= ipsq
;
3314 ipsq
->ipsq_next
= ipsq
;
3315 ipsq
->ipsq_phyint
= ill
->ill_phyint
;
3316 mutex_init(&ipsq
->ipsq_lock
, NULL
, MUTEX_DEFAULT
, 0);
3317 mutex_init(&ipx
->ipx_lock
, NULL
, MUTEX_DEFAULT
, 0);
3318 ipsq
->ipsq_ipst
= ill
->ill_ipst
; /* No netstack_hold */
3320 ipx
->ipx_writer
= curthread
;
3321 ipx
->ipx_forced
= B_FALSE
;
3322 ipx
->ipx_reentry_cnt
= 1;
3324 ipx
->ipx_depth
= getpcstack(ipx
->ipx_stack
, IPX_STACK_DEPTH
);
3331 * Here we perform initialisation of the ill_t common to both regular
3332 * interface ILLs and the special loopback ILL created by ill_lookup_on_name.
3335 ill_init_common(ill_t
*ill
, queue_t
*q
, boolean_t isv6
, boolean_t is_loopback
,
3336 boolean_t ipsq_enter
)
3341 mutex_init(&ill
->ill_lock
, NULL
, MUTEX_DEFAULT
, 0);
3342 mutex_init(&ill
->ill_saved_ire_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
3343 ill
->ill_saved_ire_cnt
= 0;
3346 ill
->ill_max_frag
= isv6
? ip_loopback_mtu_v6plus
:
3347 ip_loopback_mtuplus
;
3351 ill
->ill_net_type
= IRE_LOOPBACK
;
3354 ill
->ill_wq
= WR(q
);
3355 ill
->ill_ppa
= UINT_MAX
;
3358 ill
->ill_isv6
= isv6
;
3361 * Allocate sufficient space to contain our fragment hash table and
3364 frag_ptr
= (uchar_t
*)mi_zalloc(ILL_FRAG_HASH_TBL_SIZE
+ 2 * LIFNAMSIZ
);
3365 if (frag_ptr
== NULL
)
3367 ill
->ill_frag_ptr
= frag_ptr
;
3368 ill
->ill_frag_free_num_pkts
= 0;
3369 ill
->ill_last_frag_clean_time
= 0;
3370 ill
->ill_frag_hash_tbl
= (ipfb_t
*)frag_ptr
;
3371 ill
->ill_name
= (char *)(frag_ptr
+ ILL_FRAG_HASH_TBL_SIZE
);
3372 for (count
= 0; count
< ILL_FRAG_HASH_TBL_COUNT
; count
++) {
3373 mutex_init(&ill
->ill_frag_hash_tbl
[count
].ipfb_lock
,
3374 NULL
, MUTEX_DEFAULT
, NULL
);
3377 ill
->ill_phyint
= (phyint_t
*)mi_zalloc(sizeof (phyint_t
));
3378 if (ill
->ill_phyint
== NULL
) {
3383 mutex_init(&ill
->ill_phyint
->phyint_lock
, NULL
, MUTEX_DEFAULT
, 0);
3385 ill
->ill_phyint
->phyint_illv6
= ill
;
3387 ill
->ill_phyint
->phyint_illv4
= ill
;
3390 phyint_flags_init(ill
->ill_phyint
, DL_LOOP
);
3393 list_create(&ill
->ill_nce
, sizeof (nce_t
), offsetof(nce_t
, nce_node
));
3395 ill_set_inputfn(ill
);
3397 if (!ipsq_init(ill
, ipsq_enter
)) {
3399 mi_free(ill
->ill_phyint
);
3403 /* Frag queue limit stuff */
3404 ill
->ill_frag_count
= 0;
3405 ill
->ill_ipf_gen
= 0;
3407 rw_init(&ill
->ill_mcast_lock
, NULL
, RW_DEFAULT
, NULL
);
3408 mutex_init(&ill
->ill_mcast_serializer
, NULL
, MUTEX_DEFAULT
, NULL
);
3409 ill
->ill_global_timer
= INFINITY
;
3410 ill
->ill_mcast_v1_time
= ill
->ill_mcast_v2_time
= 0;
3411 ill
->ill_mcast_v1_tset
= ill
->ill_mcast_v2_tset
= 0;
3412 ill
->ill_mcast_rv
= MCAST_DEF_ROBUSTNESS
;
3413 ill
->ill_mcast_qi
= MCAST_DEF_QUERY_INTERVAL
;
3416 * Initialize IPv6 configuration variables. The IP module is always
3417 * opened as an IPv4 module. Instead tracking down the cases where
3418 * it switches to do ipv6, we'll just initialize the IPv6 configuration
3419 * here for convenience, this has no effect until the ill is set to do
3422 ill
->ill_reachable_time
= ND_REACHABLE_TIME
;
3423 ill
->ill_xmit_count
= ND_MAX_MULTICAST_SOLICIT
;
3424 ill
->ill_max_buf
= ND_MAX_Q
;
3425 ill
->ill_refcnt
= 0;
3431 * ill_init is called by ip_open when a device control stream is opened.
3432 * It does a few initializations, and shoots a DL_INFO_REQ message down
3433 * to the driver. The response is later picked up in ip_rput_dlpi and
3434 * used to set up default mechanisms for talking to the driver. (Always
3435 * called as writer.)
3437 * If this function returns error, ip_open will call ip_close which in
3438 * turn will call ill_delete to clean up any memory allocated here that
3441 * Note: ill_ipst and ill_zoneid must be set before calling ill_init.
3444 ill_init(queue_t
*q
, ill_t
*ill
)
3447 dl_info_req_t
*dlir
;
3450 info_mp
= allocb(MAX(sizeof (dl_info_req_t
), sizeof (dl_info_ack_t
)),
3452 if (info_mp
== NULL
)
3456 * For now pretend this is a v4 ill. We need to set phyint_ill*
3457 * at this point because of the following reason. If we can't
3458 * enter the ipsq at some point and cv_wait, the writer that
3459 * wakes us up tries to locate us using the list of all phyints
3460 * in an ipsq and the ills from the phyint thru the phyint_ill*.
3461 * If we don't set it now, we risk a missed wakeup.
3463 if ((ret
= ill_init_common(ill
, q
, B_FALSE
, B_FALSE
, B_TRUE
)) != 0) {
3468 ill
->ill_state_flags
|= ILL_LL_SUBNET_PENDING
;
3470 /* Send down the Info Request to the driver. */
3471 info_mp
->b_datap
->db_type
= M_PCPROTO
;
3472 dlir
= (dl_info_req_t
*)info_mp
->b_rptr
;
3473 info_mp
->b_wptr
= (uchar_t
*)&dlir
[1];
3474 dlir
->dl_primitive
= DL_INFO_REQ
;
3476 ill
->ill_dlpi_pending
= DL_PRIM_INVAL
;
3479 ill_dlpi_send(ill
, info_mp
);
3486 * creates datalink socket info from the device.
3489 ill_dls_info(struct sockaddr_dl
*sdl
, const ill_t
*ill
)
3493 sdl
->sdl_family
= AF_LINK
;
3494 sdl
->sdl_index
= ill_get_upper_ifindex(ill
);
3495 sdl
->sdl_type
= ill
->ill_type
;
3496 ill_get_name(ill
, sdl
->sdl_data
, sizeof (sdl
->sdl_data
));
3497 len
= strlen(sdl
->sdl_data
);
3499 sdl
->sdl_nlen
= (uchar_t
)len
;
3500 sdl
->sdl_alen
= ill
->ill_phys_addr_length
;
3502 if (ill
->ill_phys_addr_length
!= 0 && ill
->ill_phys_addr
!= NULL
)
3503 bcopy(ill
->ill_phys_addr
, &sdl
->sdl_data
[len
], sdl
->sdl_alen
);
3505 return (sizeof (struct sockaddr_dl
));
3510 * creates xarp info from the device.
3513 ill_xarp_info(struct sockaddr_dl
*sdl
, ill_t
*ill
)
3515 sdl
->sdl_family
= AF_LINK
;
3516 sdl
->sdl_index
= ill
->ill_phyint
->phyint_ifindex
;
3517 sdl
->sdl_type
= ill
->ill_type
;
3518 ill_get_name(ill
, sdl
->sdl_data
, sizeof (sdl
->sdl_data
));
3519 sdl
->sdl_nlen
= (uchar_t
)mi_strlen(sdl
->sdl_data
);
3520 sdl
->sdl_alen
= ill
->ill_phys_addr_length
;
3522 return (sdl
->sdl_nlen
);
3526 loopback_kstat_update(kstat_t
*ksp
, int rw
)
3529 netstackid_t stackid
;
3533 if (ksp
== NULL
|| ksp
->ks_data
== NULL
)
3536 if (rw
== KSTAT_WRITE
)
3539 kn
= KSTAT_NAMED_PTR(ksp
);
3540 stackid
= (zoneid_t
)(uintptr_t)ksp
->ks_private
;
3542 ns
= netstack_find_by_stackid(stackid
);
3546 ipst
= ns
->netstack_ip
;
3551 kn
[0].value
.ui32
= ipst
->ips_loopback_packets
;
3552 kn
[1].value
.ui32
= ipst
->ips_loopback_packets
;
3558 * Has ifindex been plumbed already?
3561 phyint_exists(uint_t index
, ip_stack_t
*ipst
)
3564 ASSERT(RW_LOCK_HELD(&ipst
->ips_ill_g_lock
));
3566 return (avl_find(&ipst
->ips_phyint_g_list
->phyint_list_avl_by_index
,
3567 &index
, NULL
) != NULL
);
3571 * Pick a unique ifindex.
3572 * When the index counter passes IF_INDEX_MAX for the first time, the wrap
3573 * flag is set so that next time time ip_assign_ifindex() is called, it
3574 * falls through and resets the index counter back to 1, the minimum value
3575 * for the interface index. The logic below assumes that ips_ill_index
3576 * can hold a value of IF_INDEX_MAX+1 without there being any loss
3577 * (i.e. reset back to 0.)
3580 ip_assign_ifindex(uint_t
*indexp
, ip_stack_t
*ipst
)
3584 if (!ipst
->ips_ill_index_wrap
) {
3585 *indexp
= ipst
->ips_ill_index
++;
3586 if (ipst
->ips_ill_index
> IF_INDEX_MAX
) {
3588 * Reached the maximum ifindex value, set the wrap
3589 * flag to indicate that it is no longer possible
3590 * to assume that a given index is unallocated.
3592 ipst
->ips_ill_index_wrap
= B_TRUE
;
3597 if (ipst
->ips_ill_index
> IF_INDEX_MAX
)
3598 ipst
->ips_ill_index
= 1;
3601 * Start reusing unused indexes. Note that we hold the ill_g_lock
3602 * at this point and don't want to call any function that attempts
3603 * to get the lock again.
3605 for (loops
= IF_INDEX_MAX
; loops
> 0; loops
--) {
3606 if (!phyint_exists(ipst
->ips_ill_index
, ipst
)) {
3607 /* found unused index - use it */
3608 *indexp
= ipst
->ips_ill_index
;
3612 ipst
->ips_ill_index
++;
3613 if (ipst
->ips_ill_index
> IF_INDEX_MAX
)
3614 ipst
->ips_ill_index
= 1;
3618 * all interface indicies are inuse.
3624 * Assign a unique interface index for the phyint.
3627 phyint_assign_ifindex(phyint_t
*phyi
, ip_stack_t
*ipst
)
3629 ASSERT(phyi
->phyint_ifindex
== 0);
3630 return (ip_assign_ifindex(&phyi
->phyint_ifindex
, ipst
));
3634 * Initialize the flags on `phyi' as per the provided mactype.
3637 phyint_flags_init(phyint_t
*phyi
, t_uscalar_t mactype
)
3642 * Initialize PHYI_RUNNING and PHYI_FAILED. For non-IPMP interfaces,
3643 * we always presume the underlying hardware is working and set
3644 * PHYI_RUNNING (if it's not, the driver will subsequently send a
3645 * DL_NOTE_LINK_DOWN message). For IPMP interfaces, at initialization
3646 * there are no active interfaces in the group so we set PHYI_FAILED.
3648 if (mactype
== SUNW_DL_IPMP
)
3649 flags
|= PHYI_FAILED
;
3651 flags
|= PHYI_RUNNING
;
3655 flags
|= PHYI_VIRTUAL
;
3661 flags
|= (PHYI_LOOPBACK
| PHYI_VIRTUAL
);
3665 mutex_enter(&phyi
->phyint_lock
);
3666 phyi
->phyint_flags
|= flags
;
3667 mutex_exit(&phyi
->phyint_lock
);
3671 * Return a pointer to the ill which matches the supplied name. Note that
3672 * the ill name length includes the null termination character. (May be
3673 * called as writer.)
3674 * If do_alloc and the interface is "lo0" it will be automatically created.
3675 * Cannot bump up reference on condemned ills. So dup detect can't be done
3679 ill_lookup_on_name(char *name
, boolean_t do_alloc
, boolean_t isv6
,
3680 boolean_t
*did_alloc
, ip_stack_t
*ipst
)
3686 boolean_t isloopback
;
3689 isloopback
= mi_strcmp(name
, ipif_loopback_name
) == 0;
3691 rw_enter(&ipst
->ips_ill_g_lock
, RW_READER
);
3692 ill
= ill_find_by_name(name
, isv6
, ipst
);
3693 rw_exit(&ipst
->ips_ill_g_lock
);
3698 * Couldn't find it. Does this happen to be a lookup for the
3699 * loopback device and are we allowed to allocate it?
3701 if (!isloopback
|| !do_alloc
)
3704 rw_enter(&ipst
->ips_ill_g_lock
, RW_WRITER
);
3705 ill
= ill_find_by_name(name
, isv6
, ipst
);
3707 rw_exit(&ipst
->ips_ill_g_lock
);
3711 /* Create the loopback device on demand */
3712 ill
= (ill_t
*)(mi_alloc(sizeof (ill_t
) +
3713 sizeof (ipif_loopback_name
), BPRI_MED
));
3717 bzero(ill
, sizeof (*ill
));
3718 ill
->ill_ipst
= ipst
;
3719 netstack_hold(ipst
->ips_netstack
);
3721 * For exclusive stacks we set the zoneid to zero
3722 * to make IP operate as if in the global zone.
3724 ill
->ill_zoneid
= GLOBAL_ZONEID
;
3726 if (ill_init_common(ill
, NULL
, isv6
, B_TRUE
, B_FALSE
) != 0)
3729 if (!ill_allocate_mibs(ill
))
3732 ill
->ill_current_frag
= ill
->ill_max_frag
;
3733 ill
->ill_mtu
= ill
->ill_max_frag
; /* Initial value */
3734 ill
->ill_mc_mtu
= ill
->ill_mtu
;
3736 * ipif_loopback_name can't be pointed at directly because its used
3737 * by both the ipv4 and ipv6 interfaces. When the ill is removed
3738 * from the glist, ill_glist_delete() sets the first character of
3741 ill
->ill_name
= (char *)ill
+ sizeof (*ill
);
3742 (void) strcpy(ill
->ill_name
, ipif_loopback_name
);
3743 ill
->ill_name_length
= sizeof (ipif_loopback_name
);
3744 /* Set ill_dlpi_pending for ipsq_current_finish() to work properly */
3745 ill
->ill_dlpi_pending
= DL_PRIM_INVAL
;
3747 ipif
= ipif_allocate(ill
, 0L, IRE_LOOPBACK
, B_TRUE
, B_TRUE
, NULL
);
3751 ill
->ill_flags
= ILLF_MULTICAST
;
3753 ov6addr
= ipif
->ipif_v6lcl_addr
;
3754 /* Set up default loopback address and mask. */
3756 ipaddr_t inaddr_loopback
= htonl(INADDR_LOOPBACK
);
3758 IN6_IPADDR_TO_V4MAPPED(inaddr_loopback
, &ipif
->ipif_v6lcl_addr
);
3759 V4MASK_TO_V6(htonl(IN_CLASSA_NET
), ipif
->ipif_v6net_mask
);
3760 V6_MASK_COPY(ipif
->ipif_v6lcl_addr
, ipif
->ipif_v6net_mask
,
3761 ipif
->ipif_v6subnet
);
3762 ill
->ill_flags
|= ILLF_IPV4
;
3764 ipif
->ipif_v6lcl_addr
= ipv6_loopback
;
3765 ipif
->ipif_v6net_mask
= ipv6_all_ones
;
3766 V6_MASK_COPY(ipif
->ipif_v6lcl_addr
, ipif
->ipif_v6net_mask
,
3767 ipif
->ipif_v6subnet
);
3768 ill
->ill_flags
|= ILLF_IPV6
;
3772 * Chain us in at the end of the ill list. hold the ill
3773 * before we make it globally visible. 1 for the lookup.
3777 ipsq
= ill
->ill_phyint
->phyint_ipsq
;
3779 if (ill_glist_insert(ill
, "lo", isv6
) != 0)
3780 cmn_err(CE_PANIC
, "cannot insert loopback interface");
3782 /* Let SCTP know so that it can add this to its list */
3783 sctp_update_ill(ill
, SCTP_ILL_INSERT
);
3786 * We have already assigned ipif_v6lcl_addr above, but we need to
3787 * call sctp_update_ipif_addr() after SCTP_ILL_INSERT, which
3788 * requires to be after ill_glist_insert() since we need the
3789 * ill_index set. Pass on ipv6_loopback as the old address.
3791 sctp_update_ipif_addr(ipif
, ov6addr
);
3793 ip_rts_newaddrmsg(RTM_CHGADDR
, 0, ipif
, RTSQ_DEFAULT
);
3796 * ill_glist_insert() -> ill_phyint_reinit() may have merged IPSQs.
3797 * If so, free our original one.
3799 if (ipsq
!= ill
->ill_phyint
->phyint_ipsq
)
3802 if (ipst
->ips_loopback_ksp
== NULL
) {
3803 /* Export loopback interface statistics */
3804 ipst
->ips_loopback_ksp
= kstat_create_netstack("lo", 0,
3805 ipif_loopback_name
, "net",
3806 KSTAT_TYPE_NAMED
, 2, 0,
3807 ipst
->ips_netstack
->netstack_stackid
);
3808 if (ipst
->ips_loopback_ksp
!= NULL
) {
3809 ipst
->ips_loopback_ksp
->ks_update
=
3810 loopback_kstat_update
;
3811 kn
= KSTAT_NAMED_PTR(ipst
->ips_loopback_ksp
);
3812 kstat_named_init(&kn
[0], "ipackets", KSTAT_DATA_UINT32
);
3813 kstat_named_init(&kn
[1], "opackets", KSTAT_DATA_UINT32
);
3814 ipst
->ips_loopback_ksp
->ks_private
=
3815 (void *)(uintptr_t)ipst
->ips_netstack
->
3817 kstat_install(ipst
->ips_loopback_ksp
);
3821 *did_alloc
= B_TRUE
;
3822 rw_exit(&ipst
->ips_ill_g_lock
);
3823 ill_nic_event_dispatch(ill
, MAP_IPIF_ID(ill
->ill_ipif
->ipif_id
),
3824 NE_PLUMB
, ill
->ill_name
, ill
->ill_name_length
);
3828 if (ill
->ill_phyint
!= NULL
) {
3829 ipsq
= ill
->ill_phyint
->phyint_ipsq
;
3831 ipsq
->ipsq_phyint
= NULL
;
3834 mi_free(ill
->ill_phyint
);
3837 if (ill
->ill_ipst
!= NULL
)
3838 netstack_rele(ill
->ill_ipst
->ips_netstack
);
3841 rw_exit(&ipst
->ips_ill_g_lock
);
3846 * For IPP calls - use the ip_stack_t for global stack.
3849 ill_lookup_on_ifindex_global_instance(uint_t index
, boolean_t isv6
)
3855 ns
= netstack_find_by_stackid(GLOBAL_NETSTACKID
);
3857 if ((ipst
= ns
->netstack_ip
) == NULL
) {
3858 cmn_err(CE_WARN
, "No ip_stack_t for zoneid zero!\n");
3863 ill
= ill_lookup_on_ifindex(index
, isv6
, ipst
);
3869 * Return a pointer to the ill which matches the index and IP version type.
3872 ill_lookup_on_ifindex(uint_t index
, boolean_t isv6
, ip_stack_t
*ipst
)
3878 * Indexes are stored in the phyint - a common structure
3879 * to both IPv4 and IPv6.
3881 rw_enter(&ipst
->ips_ill_g_lock
, RW_READER
);
3882 phyi
= avl_find(&ipst
->ips_phyint_g_list
->phyint_list_avl_by_index
,
3883 (void *) &index
, NULL
);
3885 ill
= isv6
? phyi
->phyint_illv6
: phyi
->phyint_illv4
;
3887 mutex_enter(&ill
->ill_lock
);
3888 if (!ILL_IS_CONDEMNED(ill
)) {
3889 ill_refhold_locked(ill
);
3890 mutex_exit(&ill
->ill_lock
);
3891 rw_exit(&ipst
->ips_ill_g_lock
);
3894 mutex_exit(&ill
->ill_lock
);
3897 rw_exit(&ipst
->ips_ill_g_lock
);
3902 * Verify whether or not an interface index is valid for the specified zoneid
3903 * to transmit packets.
3904 * It can be zero (meaning "reset") or an interface index assigned
3905 * to a non-VNI interface. (We don't use VNI interface to send packets.)
3908 ip_xmit_ifindex_valid(uint_t ifindex
, zoneid_t zoneid
, boolean_t isv6
,
3916 ill
= ill_lookup_on_ifindex_zoneid(ifindex
, zoneid
, isv6
, ipst
);
3928 * Return the ifindex next in sequence after the passed in ifindex.
3929 * If there is no next ifindex for the given protocol, return 0.
3932 ill_get_next_ifindex(uint_t index
, boolean_t isv6
, ip_stack_t
*ipst
)
3935 phyint_t
*phyi_initial
;
3938 rw_enter(&ipst
->ips_ill_g_lock
, RW_READER
);
3942 &ipst
->ips_phyint_g_list
->phyint_list_avl_by_index
);
3944 phyi
= phyi_initial
= avl_find(
3945 &ipst
->ips_phyint_g_list
->phyint_list_avl_by_index
,
3946 (void *) &index
, NULL
);
3949 for (; phyi
!= NULL
;
3950 phyi
= avl_walk(&ipst
->ips_phyint_g_list
->phyint_list_avl_by_index
,
3953 * If we're not returning the first interface in the tree
3954 * and we still haven't moved past the phyint_t that
3955 * corresponds to index, avl_walk needs to be called again
3957 if (!((index
!= 0) && (phyi
== phyi_initial
))) {
3959 if ((phyi
->phyint_illv6
) &&
3960 ILL_CAN_LOOKUP(phyi
->phyint_illv6
) &&
3961 (phyi
->phyint_illv6
->ill_isv6
== 1))
3964 if ((phyi
->phyint_illv4
) &&
3965 ILL_CAN_LOOKUP(phyi
->phyint_illv4
) &&
3966 (phyi
->phyint_illv4
->ill_isv6
== 0))
3972 rw_exit(&ipst
->ips_ill_g_lock
);
3975 ifindex
= phyi
->phyint_ifindex
;
3983 * Return the ifindex for the named interface.
3984 * If there is no next ifindex for the interface, return 0.
3987 ill_get_ifindex_by_name(char *name
, ip_stack_t
*ipst
)
3990 avl_index_t where
= 0;
3993 rw_enter(&ipst
->ips_ill_g_lock
, RW_READER
);
3995 if ((phyi
= avl_find(&ipst
->ips_phyint_g_list
->phyint_list_avl_by_name
,
3996 name
, &where
)) == NULL
) {
3997 rw_exit(&ipst
->ips_ill_g_lock
);
4001 ifindex
= phyi
->phyint_ifindex
;
4003 rw_exit(&ipst
->ips_ill_g_lock
);
4009 * Return the ifindex to be used by upper layer protocols for instance
4010 * for IPV6_RECVPKTINFO. If IPMP this is the one for the upper ill.
4013 ill_get_upper_ifindex(const ill_t
*ill
)
4015 if (IS_UNDER_IPMP(ill
))
4016 return (ipmp_ill_get_ipmp_ifindex(ill
));
4018 return (ill
->ill_phyint
->phyint_ifindex
);
4023 * Obtain a reference to the ill. The ill_refcnt is a dynamic refcnt
4024 * that gives a running thread a reference to the ill. This reference must be
4025 * released by the thread when it is done accessing the ill and related
4026 * objects. ill_refcnt can not be used to account for static references
4027 * such as other structures pointing to an ill. Callers must generally
4028 * check whether an ill can be refheld by using ILL_CAN_LOOKUP macros
4029 * or be sure that the ill is not being deleted or changing state before
4030 * calling the refhold functions. A non-zero ill_refcnt ensures that the
4031 * ill won't change any of its critical state such as address, netmask etc.
4034 ill_refhold(ill_t
*ill
)
4036 mutex_enter(&ill
->ill_lock
);
4039 mutex_exit(&ill
->ill_lock
);
4043 ill_refhold_locked(ill_t
*ill
)
4045 ASSERT(MUTEX_HELD(&ill
->ill_lock
));
4050 /* Returns true if we managed to get a refhold */
4052 ill_check_and_refhold(ill_t
*ill
)
4054 mutex_enter(&ill
->ill_lock
);
4055 if (!ILL_IS_CONDEMNED(ill
)) {
4056 ill_refhold_locked(ill
);
4057 mutex_exit(&ill
->ill_lock
);
4060 mutex_exit(&ill
->ill_lock
);
4065 * Must not be called while holding any locks. Otherwise if this is
4066 * the last reference to be released, there is a chance of recursive mutex
4067 * panic due to ill_refrele -> ipif_ill_refrele_tail -> qwriter_ip trying
4068 * to restart an ioctl.
4071 ill_refrele(ill_t
*ill
)
4073 mutex_enter(&ill
->ill_lock
);
4074 ASSERT(ill
->ill_refcnt
!= 0);
4076 ILL_UNTRACE_REF(ill
);
4077 if (ill
->ill_refcnt
!= 0) {
4078 /* Every ire pointing to the ill adds 1 to ill_refcnt */
4079 mutex_exit(&ill
->ill_lock
);
4083 /* Drops the ill_lock */
4084 ipif_ill_refrele_tail(ill
);
4088 * Obtain a weak reference count on the ill. This reference ensures the
4089 * ill won't be freed, but the ill may change any of its critical state
4090 * such as netmask, address etc. Returns an error if the ill has started
4094 ill_waiter_inc(ill_t
*ill
)
4096 mutex_enter(&ill
->ill_lock
);
4097 if (ill
->ill_state_flags
& ILL_CONDEMNED
) {
4098 mutex_exit(&ill
->ill_lock
);
4102 mutex_exit(&ill
->ill_lock
);
4107 ill_waiter_dcr(ill_t
*ill
)
4109 mutex_enter(&ill
->ill_lock
);
4111 if (ill
->ill_waiters
== 0)
4112 cv_broadcast(&ill
->ill_cv
);
4113 mutex_exit(&ill
->ill_lock
);
4117 * ip_ll_subnet_defaults is called when we get the DL_INFO_ACK back from the
4118 * driver. We construct best guess defaults for lower level information that
4119 * we need. If an interface is brought up without injection of any overriding
4120 * information from outside, we have to be ready to go with these defaults.
4121 * When we get the first DL_INFO_ACK (from ip_open() sending a DL_INFO_REQ)
4122 * we primarely want the dl_provider_style.
4123 * The subsequent DL_INFO_ACK is received after doing a DL_ATTACH and DL_BIND
4124 * at which point we assume the other part of the information is valid.
4127 ip_ll_subnet_defaults(ill_t
*ill
, mblk_t
*mp
)
4129 uchar_t
*brdcst_addr
;
4130 uint_t brdcst_addr_length
, phys_addr_length
;
4131 t_scalar_t sap_length
;
4132 dl_info_ack_t
*dlia
;
4134 dl_qos_cl_sel1_t
*sel1
;
4137 ASSERT(IAM_WRITER_ILL(ill
));
4140 * Till the ill is fully up the ill is not globally visible.
4141 * So no need for a lock.
4143 dlia
= (dl_info_ack_t
*)mp
->b_rptr
;
4144 ill
->ill_mactype
= dlia
->dl_mac_type
;
4146 ipm
= ip_m_lookup(dlia
->dl_mac_type
);
4148 ipm
= ip_m_lookup(DL_OTHER
);
4149 ASSERT(ipm
!= NULL
);
4151 ill
->ill_media
= ipm
;
4154 * When the new DLPI stuff is ready we'll pull lengths
4157 if (dlia
->dl_version
== DL_VERSION_2
) {
4158 brdcst_addr_length
= dlia
->dl_brdcst_addr_length
;
4159 brdcst_addr
= mi_offset_param(mp
, dlia
->dl_brdcst_addr_offset
,
4160 brdcst_addr_length
);
4161 if (brdcst_addr
== NULL
) {
4162 brdcst_addr_length
= 0;
4164 sap_length
= dlia
->dl_sap_length
;
4165 phys_addr_length
= dlia
->dl_addr_length
- ABS(sap_length
);
4166 ip1dbg(("ip: bcast_len %d, sap_len %d, phys_len %d\n",
4167 brdcst_addr_length
, sap_length
, phys_addr_length
));
4169 brdcst_addr_length
= 6;
4170 brdcst_addr
= ip_six_byte_all_ones
;
4172 phys_addr_length
= brdcst_addr_length
;
4175 ill
->ill_bcast_addr_length
= brdcst_addr_length
;
4176 ill
->ill_phys_addr_length
= phys_addr_length
;
4177 ill
->ill_sap_length
= sap_length
;
4180 * Synthetic DLPI types such as SUNW_DL_IPMP specify a zero SDU,
4181 * but we must ensure a minimum IP MTU is used since other bits of
4182 * IP will fly apart otherwise.
4184 min_mtu
= ill
->ill_isv6
? IPV6_MIN_MTU
: IP_MIN_MTU
;
4185 ill
->ill_max_frag
= MAX(min_mtu
, dlia
->dl_max_sdu
);
4186 ill
->ill_current_frag
= ill
->ill_max_frag
;
4187 ill
->ill_mtu
= ill
->ill_max_frag
;
4188 ill
->ill_mc_mtu
= ill
->ill_mtu
; /* Overridden by DL_NOTE_SDU_SIZE2 */
4190 ill
->ill_type
= ipm
->ip_m_type
;
4192 if (!ill
->ill_dlpi_style_set
) {
4193 if (dlia
->dl_provider_style
== DL_STYLE2
)
4194 ill
->ill_needs_attach
= 1;
4196 phyint_flags_init(ill
->ill_phyint
, ill
->ill_mactype
);
4199 * Allocate the first ipif on this ill. We don't delay it
4200 * further as ioctl handling assumes at least one ipif exists.
4202 * At this point we don't know whether the ill is v4 or v6.
4203 * We will know this whan the SIOCSLIFNAME happens and
4204 * the correct value for ill_isv6 will be assigned in
4205 * ipif_set_values(). We need to hold the ill lock and
4206 * clear the ILL_LL_SUBNET_PENDING flag and atomically do
4209 (void) ipif_allocate(ill
, 0, IRE_LOCAL
,
4210 dlia
->dl_provider_style
!= DL_STYLE2
, B_TRUE
, NULL
);
4211 mutex_enter(&ill
->ill_lock
);
4212 ASSERT(ill
->ill_dlpi_style_set
== 0);
4213 ill
->ill_dlpi_style_set
= 1;
4214 ill
->ill_state_flags
&= ~ILL_LL_SUBNET_PENDING
;
4215 cv_broadcast(&ill
->ill_cv
);
4216 mutex_exit(&ill
->ill_lock
);
4220 ASSERT(ill
->ill_ipif
!= NULL
);
4222 * We know whether it is IPv4 or IPv6 now, as this is the
4223 * second DL_INFO_ACK we are recieving in response to the
4224 * DL_INFO_REQ sent in ipif_set_values.
4226 ill
->ill_sap
= (ill
->ill_isv6
) ? ipm
->ip_m_ipv6sap
: ipm
->ip_m_ipv4sap
;
4228 * Clear all the flags that were set based on ill_bcast_addr_length
4229 * and ill_phys_addr_length (in ipif_set_values) as these could have
4230 * changed now and we need to re-evaluate.
4232 ill
->ill_flags
&= ~(ILLF_MULTICAST
| ILLF_NONUD
| ILLF_NOARP
);
4233 ill
->ill_ipif
->ipif_flags
&= ~(IPIF_BROADCAST
| IPIF_POINTOPOINT
);
4236 * Free ill_bcast_mp as things could have changed now.
4238 * NOTE: The IPMP meta-interface is special-cased because it starts
4239 * with no underlying interfaces (and thus an unknown broadcast
4240 * address length), but we enforce that an interface is broadcast-
4241 * capable as part of allowing it to join a group.
4243 if (ill
->ill_bcast_addr_length
== 0 && !IS_IPMP(ill
)) {
4244 if (ill
->ill_bcast_mp
!= NULL
)
4245 freemsg(ill
->ill_bcast_mp
);
4246 ill
->ill_net_type
= IRE_IF_NORESOLVER
;
4248 ill
->ill_bcast_mp
= ill_dlur_gen(NULL
,
4249 ill
->ill_phys_addr_length
,
4251 ill
->ill_sap_length
);
4255 * Note: xresolv interfaces will eventually need NOARP
4256 * set here as well, but that will require those
4257 * external resolvers to have some knowledge of
4258 * that flag and act appropriately. Not to be changed
4261 ill
->ill_flags
|= ILLF_NONUD
;
4263 ill
->ill_flags
|= ILLF_NOARP
;
4265 if (ill
->ill_mactype
== SUNW_DL_VNI
) {
4266 ill
->ill_ipif
->ipif_flags
|= IPIF_NOXMIT
;
4267 } else if (ill
->ill_phys_addr_length
== 0 ||
4268 ill
->ill_mactype
== DL_IPV4
||
4269 ill
->ill_mactype
== DL_IPV6
) {
4271 * The underying link is point-to-point, so mark the
4272 * interface as such. We can do IP multicast over
4273 * such a link since it transmits all network-layer
4274 * packets to the remote side the same way.
4276 ill
->ill_flags
|= ILLF_MULTICAST
;
4277 ill
->ill_ipif
->ipif_flags
|= IPIF_POINTOPOINT
;
4280 ill
->ill_net_type
= IRE_IF_RESOLVER
;
4281 if (ill
->ill_bcast_mp
!= NULL
)
4282 freemsg(ill
->ill_bcast_mp
);
4283 ill
->ill_bcast_mp
= ill_dlur_gen(brdcst_addr
,
4284 ill
->ill_bcast_addr_length
, ill
->ill_sap
,
4285 ill
->ill_sap_length
);
4287 * Later detect lack of DLPI driver multicast
4288 * capability by catching DL_ENABMULTI errors in
4291 ill
->ill_flags
|= ILLF_MULTICAST
;
4293 ill
->ill_ipif
->ipif_flags
|= IPIF_BROADCAST
;
4296 /* For IPMP, PHYI_IPMP should already be set by phyint_flags_init() */
4297 if (ill
->ill_mactype
== SUNW_DL_IPMP
)
4298 ASSERT(ill
->ill_phyint
->phyint_flags
& PHYI_IPMP
);
4300 /* By default an interface does not support any CoS marking */
4301 ill
->ill_flags
&= ~ILLF_COS_ENABLED
;
4304 * If we get QoS information in DL_INFO_ACK, the device supports
4305 * some form of CoS marking, set ILLF_COS_ENABLED.
4307 sel1
= (dl_qos_cl_sel1_t
*)mi_offset_param(mp
, dlia
->dl_qos_offset
,
4308 dlia
->dl_qos_length
);
4309 if ((sel1
!= NULL
) && (sel1
->dl_qos_type
== DL_QOS_CL_SEL1
)) {
4310 ill
->ill_flags
|= ILLF_COS_ENABLED
;
4313 /* Clear any previous error indication. */
4319 * Perform various checks to verify that an address would make sense as a
4320 * local, remote, or subnet interface address.
4323 ip_addr_ok_v4(ipaddr_t addr
, ipaddr_t subnet_mask
)
4328 * Don't allow all zeroes, or all ones, but allow
4331 if ((net_mask
= ip_net_mask(addr
)) == 0)
4333 /* A given netmask overrides the "guess" netmask */
4334 if (subnet_mask
!= 0)
4335 net_mask
= subnet_mask
;
4336 if ((net_mask
!= ~(ipaddr_t
)0) && ((addr
== (addr
& net_mask
)) ||
4337 (addr
== (addr
| ~net_mask
)))) {
4342 * Even if the netmask is all ones, we do not allow address to be
4345 if (addr
== INADDR_BROADCAST
)
4354 #define V6_IPIF_LINKLOCAL(p) \
4355 IN6_IS_ADDR_LINKLOCAL(&(p)->ipif_v6lcl_addr)
4358 * Compare two given ipifs and check if the second one is better than
4359 * the first one using the order of preference (not taking deprecated
4360 * into acount) specified in ipif_lookup_multicast().
4363 ipif_comp_multi(ipif_t
*old_ipif
, ipif_t
*new_ipif
, boolean_t isv6
)
4365 /* Check the least preferred first. */
4366 if (IS_LOOPBACK(old_ipif
->ipif_ill
)) {
4367 /* If both ipifs are the same, use the first one. */
4368 if (IS_LOOPBACK(new_ipif
->ipif_ill
))
4374 /* For IPv6, check for link local address. */
4375 if (isv6
&& V6_IPIF_LINKLOCAL(old_ipif
)) {
4376 if (IS_LOOPBACK(new_ipif
->ipif_ill
) ||
4377 V6_IPIF_LINKLOCAL(new_ipif
)) {
4378 /* The second one is equal or less preferred. */
4385 /* Then check for point to point interface. */
4386 if (old_ipif
->ipif_flags
& IPIF_POINTOPOINT
) {
4387 if (IS_LOOPBACK(new_ipif
->ipif_ill
) ||
4388 (isv6
&& V6_IPIF_LINKLOCAL(new_ipif
)) ||
4389 (new_ipif
->ipif_flags
& IPIF_POINTOPOINT
)) {
4396 /* old_ipif is a normal interface, so no need to use the new one. */
4401 * Find a mulitcast-capable ipif given an IP instance and zoneid.
4402 * The ipif must be up, and its ill must multicast-capable, not
4403 * condemned, not an underlying interface in an IPMP group, and
4404 * not a VNI interface. Order of preference:
4407 * 1b. normal, but deprecated
4408 * 2a. point to point
4409 * 2b. point to point, but deprecated
4411 * 3b. link local, but deprecated
4415 ipif_lookup_multicast(ip_stack_t
*ipst
, zoneid_t zoneid
, boolean_t isv6
)
4418 ill_walk_context_t ctx
;
4420 ipif_t
*saved_ipif
= NULL
;
4421 ipif_t
*dep_ipif
= NULL
;
4423 rw_enter(&ipst
->ips_ill_g_lock
, RW_READER
);
4425 ill
= ILL_START_WALK_V6(&ctx
, ipst
);
4427 ill
= ILL_START_WALK_V4(&ctx
, ipst
);
4429 for (; ill
!= NULL
; ill
= ill_next(&ctx
, ill
)) {
4430 mutex_enter(&ill
->ill_lock
);
4431 if (IS_VNI(ill
) || IS_UNDER_IPMP(ill
) ||
4432 ILL_IS_CONDEMNED(ill
) ||
4433 !(ill
->ill_flags
& ILLF_MULTICAST
)) {
4434 mutex_exit(&ill
->ill_lock
);
4437 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
;
4438 ipif
= ipif
->ipif_next
) {
4439 if (zoneid
!= ipif
->ipif_zoneid
&&
4440 zoneid
!= ALL_ZONES
&&
4441 ipif
->ipif_zoneid
!= ALL_ZONES
) {
4444 if (!(ipif
->ipif_flags
& IPIF_UP
) ||
4445 IPIF_IS_CONDEMNED(ipif
)) {
4450 * Found one candidate. If it is deprecated,
4451 * remember it in dep_ipif. If it is not deprecated,
4452 * remember it in saved_ipif.
4454 if (ipif
->ipif_flags
& IPIF_DEPRECATED
) {
4455 if (dep_ipif
== NULL
) {
4457 } else if (ipif_comp_multi(dep_ipif
, ipif
,
4460 * If the previous dep_ipif does not
4461 * belong to the same ill, we've done
4462 * a ipif_refhold() on it. So we need
4465 if (dep_ipif
->ipif_ill
!= ill
)
4466 ipif_refrele(dep_ipif
);
4471 if (saved_ipif
== NULL
) {
4474 if (ipif_comp_multi(saved_ipif
, ipif
, isv6
)) {
4475 if (saved_ipif
->ipif_ill
!= ill
)
4476 ipif_refrele(saved_ipif
);
4482 * Before going to the next ill, do a ipif_refhold() on the
4485 if (saved_ipif
!= NULL
&& saved_ipif
->ipif_ill
== ill
)
4486 ipif_refhold_locked(saved_ipif
);
4487 if (dep_ipif
!= NULL
&& dep_ipif
->ipif_ill
== ill
)
4488 ipif_refhold_locked(dep_ipif
);
4489 mutex_exit(&ill
->ill_lock
);
4491 rw_exit(&ipst
->ips_ill_g_lock
);
4494 * If we have only the saved_ipif, return it. But if we have both
4495 * saved_ipif and dep_ipif, check to see which one is better.
4497 if (saved_ipif
!= NULL
) {
4498 if (dep_ipif
!= NULL
) {
4499 if (ipif_comp_multi(saved_ipif
, dep_ipif
, isv6
)) {
4500 ipif_refrele(saved_ipif
);
4503 ipif_refrele(dep_ipif
);
4504 return (saved_ipif
);
4507 return (saved_ipif
);
4514 ill_lookup_multicast(ip_stack_t
*ipst
, zoneid_t zoneid
, boolean_t isv6
)
4519 ipif
= ipif_lookup_multicast(ipst
, zoneid
, isv6
);
4523 ill
= ipif
->ipif_ill
;
4530 * This function is called when an application does not specify an interface
4531 * to be used for multicast traffic (joining a group/sending data). It
4532 * calls ire_lookup_multi() to look for an interface route for the
4533 * specified multicast group. Doing this allows the administrator to add
4534 * prefix routes for multicast to indicate which interface to be used for
4535 * multicast traffic in the above scenario. The route could be for all
4536 * multicast (224.0/4), for a single multicast group (a /32 route) or
4537 * anything in between. If there is no such multicast route, we just find
4538 * any multicast capable interface and return it. The returned ipif
4542 ill_lookup_group_v4(ipaddr_t group
, zoneid_t zoneid
, ip_stack_t
*ipst
,
4547 ill
= ire_lookup_multi_ill_v4(group
, zoneid
, ipst
, setsrcp
);
4551 return (ill_lookup_multicast(ipst
, zoneid
, B_FALSE
));
4555 * Look for an ipif with the specified interface address and destination.
4556 * The destination address is used only for matching point-to-point interfaces.
4559 ipif_lookup_interface(ipaddr_t if_addr
, ipaddr_t dst
, ip_stack_t
*ipst
)
4563 ill_walk_context_t ctx
;
4566 * First match all the point-to-point interfaces
4567 * before looking at non-point-to-point interfaces.
4568 * This is done to avoid returning non-point-to-point
4569 * ipif instead of unnumbered point-to-point ipif.
4571 rw_enter(&ipst
->ips_ill_g_lock
, RW_READER
);
4572 ill
= ILL_START_WALK_V4(&ctx
, ipst
);
4573 for (; ill
!= NULL
; ill
= ill_next(&ctx
, ill
)) {
4574 mutex_enter(&ill
->ill_lock
);
4575 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
;
4576 ipif
= ipif
->ipif_next
) {
4577 /* Allow the ipif to be down */
4578 if ((ipif
->ipif_flags
& IPIF_POINTOPOINT
) &&
4579 (ipif
->ipif_lcl_addr
== if_addr
) &&
4580 (ipif
->ipif_pp_dst_addr
== dst
)) {
4581 if (!IPIF_IS_CONDEMNED(ipif
)) {
4582 ipif_refhold_locked(ipif
);
4583 mutex_exit(&ill
->ill_lock
);
4584 rw_exit(&ipst
->ips_ill_g_lock
);
4589 mutex_exit(&ill
->ill_lock
);
4591 rw_exit(&ipst
->ips_ill_g_lock
);
4593 /* lookup the ipif based on interface address */
4594 ipif
= ipif_lookup_addr(if_addr
, NULL
, ALL_ZONES
, ipst
);
4595 ASSERT(ipif
== NULL
|| !ipif
->ipif_isv6
);
4600 * Common function for ipif_lookup_addr() and ipif_lookup_addr_exact().
4603 ipif_lookup_addr_common(ipaddr_t addr
, ill_t
*match_ill
, uint32_t match_flags
,
4604 zoneid_t zoneid
, ip_stack_t
*ipst
)
4608 boolean_t ptp
= B_FALSE
;
4609 ill_walk_context_t ctx
;
4610 boolean_t match_illgrp
= (match_flags
& IPIF_MATCH_ILLGRP
);
4611 boolean_t no_duplicate
= (match_flags
& IPIF_MATCH_NONDUP
);
4613 rw_enter(&ipst
->ips_ill_g_lock
, RW_READER
);
4615 * Repeat twice, first based on local addresses and
4616 * next time for pointopoint.
4619 ill
= ILL_START_WALK_V4(&ctx
, ipst
);
4620 for (; ill
!= NULL
; ill
= ill_next(&ctx
, ill
)) {
4621 if (match_ill
!= NULL
&& ill
!= match_ill
&&
4622 (!match_illgrp
|| !IS_IN_SAME_ILLGRP(ill
, match_ill
))) {
4625 mutex_enter(&ill
->ill_lock
);
4626 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
;
4627 ipif
= ipif
->ipif_next
) {
4628 if (zoneid
!= ALL_ZONES
&&
4629 zoneid
!= ipif
->ipif_zoneid
&&
4630 ipif
->ipif_zoneid
!= ALL_ZONES
)
4633 if (no_duplicate
&& !(ipif
->ipif_flags
& IPIF_UP
))
4636 /* Allow the ipif to be down */
4637 if ((!ptp
&& (ipif
->ipif_lcl_addr
== addr
) &&
4638 ((ipif
->ipif_flags
& IPIF_UNNUMBERED
) == 0)) ||
4639 (ptp
&& (ipif
->ipif_flags
& IPIF_POINTOPOINT
) &&
4640 (ipif
->ipif_pp_dst_addr
== addr
))) {
4641 if (!IPIF_IS_CONDEMNED(ipif
)) {
4642 ipif_refhold_locked(ipif
);
4643 mutex_exit(&ill
->ill_lock
);
4644 rw_exit(&ipst
->ips_ill_g_lock
);
4649 mutex_exit(&ill
->ill_lock
);
4652 /* If we already did the ptp case, then we are done */
4654 rw_exit(&ipst
->ips_ill_g_lock
);
4662 * Lookup an ipif with the specified address. For point-to-point links we
4663 * look for matches on either the destination address or the local address,
4664 * but we skip the local address check if IPIF_UNNUMBERED is set. If the
4665 * `match_ill' argument is non-NULL, the lookup is restricted to that ill
4666 * (or illgrp if `match_ill' is in an IPMP group).
4669 ipif_lookup_addr(ipaddr_t addr
, ill_t
*match_ill
, zoneid_t zoneid
,
4672 return (ipif_lookup_addr_common(addr
, match_ill
, IPIF_MATCH_ILLGRP
,
4677 * Lookup an ipif with the specified address. Similar to ipif_lookup_addr,
4678 * except that we will only return an address if it is not marked as
4682 ipif_lookup_addr_nondup(ipaddr_t addr
, ill_t
*match_ill
, zoneid_t zoneid
,
4685 return (ipif_lookup_addr_common(addr
, match_ill
,
4686 (IPIF_MATCH_ILLGRP
| IPIF_MATCH_NONDUP
),
4691 * Special abbreviated version of ipif_lookup_addr() that doesn't match
4692 * `match_ill' across the IPMP group. This function is only needed in some
4693 * corner-cases; almost everything should use ipif_lookup_addr().
4696 ipif_lookup_addr_exact(ipaddr_t addr
, ill_t
*match_ill
, ip_stack_t
*ipst
)
4698 ASSERT(match_ill
!= NULL
);
4699 return (ipif_lookup_addr_common(addr
, match_ill
, 0, ALL_ZONES
,
4704 * Look for an ipif with the specified address. For point-point links
4705 * we look for matches on either the destination address and the local
4706 * address, but we ignore the check on the local address if IPIF_UNNUMBERED
4708 * If the `match_ill' argument is non-NULL, the lookup is restricted to that
4709 * ill (or illgrp if `match_ill' is in an IPMP group).
4710 * Return the zoneid for the ipif which matches. ALL_ZONES if no match.
4713 ipif_lookup_addr_zoneid(ipaddr_t addr
, ill_t
*match_ill
, ip_stack_t
*ipst
)
4718 boolean_t ptp
= B_FALSE
;
4719 ill_walk_context_t ctx
;
4721 rw_enter(&ipst
->ips_ill_g_lock
, RW_READER
);
4723 * Repeat twice, first based on local addresses and
4724 * next time for pointopoint.
4727 ill
= ILL_START_WALK_V4(&ctx
, ipst
);
4728 for (; ill
!= NULL
; ill
= ill_next(&ctx
, ill
)) {
4729 if (match_ill
!= NULL
&& ill
!= match_ill
&&
4730 !IS_IN_SAME_ILLGRP(ill
, match_ill
)) {
4733 mutex_enter(&ill
->ill_lock
);
4734 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
;
4735 ipif
= ipif
->ipif_next
) {
4736 /* Allow the ipif to be down */
4737 if ((!ptp
&& (ipif
->ipif_lcl_addr
== addr
) &&
4738 ((ipif
->ipif_flags
& IPIF_UNNUMBERED
) == 0)) ||
4739 (ptp
&& (ipif
->ipif_flags
& IPIF_POINTOPOINT
) &&
4740 (ipif
->ipif_pp_dst_addr
== addr
)) &&
4741 !(ipif
->ipif_state_flags
& IPIF_CONDEMNED
)) {
4742 zoneid
= ipif
->ipif_zoneid
;
4743 mutex_exit(&ill
->ill_lock
);
4744 rw_exit(&ipst
->ips_ill_g_lock
);
4748 mutex_exit(&ill
->ill_lock
);
4751 /* If we already did the ptp case, then we are done */
4753 rw_exit(&ipst
->ips_ill_g_lock
);
4761 * Look for an ipif that matches the specified remote address i.e. the
4762 * ipif that would receive the specified packet.
4763 * First look for directly connected interfaces and then do a recursive
4764 * IRE lookup and pick the first ipif corresponding to the source address in the
4766 * Returns: held ipif
4768 * This is only used for ICMP_ADDRESS_MASK_REQUESTs
4771 ipif_lookup_remote(ill_t
*ill
, ipaddr_t addr
, zoneid_t zoneid
)
4775 ASSERT(!ill
->ill_isv6
);
4778 * Someone could be changing this ipif currently or change it
4779 * after we return this. Thus a few packets could use the old
4780 * old values. However structure updates/creates (ire, ilg, ilm etc)
4781 * will atomically be updated or cleaned up with the new value
4782 * Thus we don't need a lock to check the flags or other attrs below.
4784 mutex_enter(&ill
->ill_lock
);
4785 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
; ipif
= ipif
->ipif_next
) {
4786 if (IPIF_IS_CONDEMNED(ipif
))
4788 if (zoneid
!= ALL_ZONES
&& zoneid
!= ipif
->ipif_zoneid
&&
4789 ipif
->ipif_zoneid
!= ALL_ZONES
)
4791 /* Allow the ipif to be down */
4792 if (ipif
->ipif_flags
& IPIF_POINTOPOINT
) {
4793 if ((ipif
->ipif_pp_dst_addr
== addr
) ||
4794 (!(ipif
->ipif_flags
& IPIF_UNNUMBERED
) &&
4795 ipif
->ipif_lcl_addr
== addr
)) {
4796 ipif_refhold_locked(ipif
);
4797 mutex_exit(&ill
->ill_lock
);
4800 } else if (ipif
->ipif_subnet
== (addr
& ipif
->ipif_net_mask
)) {
4801 ipif_refhold_locked(ipif
);
4802 mutex_exit(&ill
->ill_lock
);
4806 mutex_exit(&ill
->ill_lock
);
4808 * For a remote destination it isn't possible to nail down a particular
4812 /* Pick the first interface */
4813 ipif
= ipif_get_next_ipif(NULL
, ill
);
4818 * This func does not prevent refcnt from increasing. But if
4819 * the caller has taken steps to that effect, then this func
4820 * can be used to determine whether the ill has become quiescent
4823 ill_is_quiescent(ill_t
*ill
)
4827 ASSERT(MUTEX_HELD(&ill
->ill_lock
));
4829 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
; ipif
= ipif
->ipif_next
) {
4830 if (ipif
->ipif_refcnt
!= 0)
4833 if (!ILL_DOWN_OK(ill
) || ill
->ill_refcnt
!= 0) {
4840 ill_is_freeable(ill_t
*ill
)
4844 ASSERT(MUTEX_HELD(&ill
->ill_lock
));
4846 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
; ipif
= ipif
->ipif_next
) {
4847 if (ipif
->ipif_refcnt
!= 0) {
4851 if (!ILL_FREE_OK(ill
) || ill
->ill_refcnt
!= 0) {
4858 * This func does not prevent refcnt from increasing. But if
4859 * the caller has taken steps to that effect, then this func
4860 * can be used to determine whether the ipif has become quiescent
4863 ipif_is_quiescent(ipif_t
*ipif
)
4867 ASSERT(MUTEX_HELD(&ipif
->ipif_ill
->ill_lock
));
4869 if (ipif
->ipif_refcnt
!= 0)
4872 ill
= ipif
->ipif_ill
;
4873 if (ill
->ill_ipif_up_count
!= 0 || ill
->ill_ipif_dup_count
!= 0 ||
4874 ill
->ill_logical_down
) {
4878 /* This is the last ipif going down or being deleted on this ill */
4879 if (ill
->ill_ire_cnt
!= 0 || ill
->ill_refcnt
!= 0) {
4887 * return true if the ipif can be destroyed: the ipif has to be quiescent
4888 * with zero references from ire/ilm to it.
4891 ipif_is_freeable(ipif_t
*ipif
)
4893 ASSERT(MUTEX_HELD(&ipif
->ipif_ill
->ill_lock
));
4894 ASSERT(ipif
->ipif_id
!= 0);
4895 return (ipif
->ipif_refcnt
== 0);
4899 * The ipif/ill/ire has been refreled. Do the tail processing.
4900 * Determine if the ipif or ill in question has become quiescent and if so
4901 * wakeup close and/or restart any queued pending ioctl that is waiting
4902 * for the ipif_down (or ill_down)
4905 ipif_ill_refrele_tail(ill_t
*ill
)
4912 dl_notify_ind_t
*dlindp
;
4914 ASSERT(MUTEX_HELD(&ill
->ill_lock
));
4916 if ((ill
->ill_state_flags
& ILL_CONDEMNED
) && ill_is_freeable(ill
)) {
4917 /* ip_modclose() may be waiting */
4918 cv_broadcast(&ill
->ill_cv
);
4921 ipsq
= ill
->ill_phyint
->phyint_ipsq
;
4922 mutex_enter(&ipsq
->ipsq_lock
);
4923 ipx
= ipsq
->ipsq_xop
;
4924 mutex_enter(&ipx
->ipx_lock
);
4925 if (ipx
->ipx_waitfor
== 0) /* no one's waiting; bail */
4928 ASSERT(ipx
->ipx_pending_mp
!= NULL
&& ipx
->ipx_pending_ipif
!= NULL
);
4930 ipif
= ipx
->ipx_pending_ipif
;
4931 if (ipif
->ipif_ill
!= ill
) /* wait is for another ill; bail */
4934 switch (ipx
->ipx_waitfor
) {
4936 if (!ipif_is_quiescent(ipif
))
4940 if (!ipif_is_freeable(ipif
))
4944 if (!ill_is_quiescent(ill
))
4949 * ILL_FREE is only for loopback; normal ill teardown waits
4950 * synchronously in ip_modclose() without using ipx_waitfor,
4951 * handled by the cv_broadcast() at the top of this function.
4953 if (!ill_is_freeable(ill
))
4957 cmn_err(CE_PANIC
, "ipsq: %p unknown ipx_waitfor %d\n",
4958 (void *)ipsq
, ipx
->ipx_waitfor
);
4961 ill_refhold_locked(ill
); /* for qwriter_ip() call below */
4962 mutex_exit(&ipx
->ipx_lock
);
4963 mp
= ipsq_pending_mp_get(ipsq
, &connp
);
4964 mutex_exit(&ipsq
->ipsq_lock
);
4965 mutex_exit(&ill
->ill_lock
);
4969 * NOTE: all of the qwriter_ip() calls below use CUR_OP since
4970 * we can only get here when the current operation decides it
4971 * it needs to quiesce via ipsq_pending_mp_add().
4973 switch (mp
->b_datap
->db_type
) {
4977 * For now, only DL_NOTIFY_IND messages can use this facility.
4979 dlindp
= (dl_notify_ind_t
*)mp
->b_rptr
;
4980 ASSERT(dlindp
->dl_primitive
== DL_NOTIFY_IND
);
4982 switch (dlindp
->dl_notification
) {
4983 case DL_NOTE_PHYS_ADDR
:
4984 qwriter_ip(ill
, ill
->ill_rq
, mp
,
4985 ill_set_phys_addr_tail
, CUR_OP
, B_TRUE
);
4987 case DL_NOTE_REPLUMB
:
4988 qwriter_ip(ill
, ill
->ill_rq
, mp
,
4989 ill_replumb_tail
, CUR_OP
, B_TRUE
);
4999 qwriter_ip(ill
, ill
->ill_rq
, mp
, ipif_all_down_tail
, CUR_OP
,
5005 qwriter_ip(ill
, (connp
!= NULL
? CONNP_TO_WQ(connp
) :
5006 ill
->ill_wq
), mp
, ip_reprocess_ioctl
, CUR_OP
, B_TRUE
);
5010 cmn_err(CE_PANIC
, "ipif_ill_refrele_tail mp %p "
5011 "db_type %d\n", (void *)mp
, mp
->b_datap
->db_type
);
5015 mutex_exit(&ipsq
->ipsq_lock
);
5016 mutex_exit(&ipx
->ipx_lock
);
5017 mutex_exit(&ill
->ill_lock
);
5021 /* Reuse trace buffer from beginning (if reached the end) and record trace */
5023 th_trace_rrecord(th_trace_t
*th_trace
)
5028 lastref
= th_trace
->th_trace_lastref
;
5030 if (lastref
== TR_BUF_MAX
)
5032 th_trace
->th_trace_lastref
= lastref
;
5033 tr_buf
= &th_trace
->th_trbuf
[lastref
];
5034 tr_buf
->tr_time
= ddi_get_lbolt();
5035 tr_buf
->tr_depth
= getpcstack(tr_buf
->tr_stack
, TR_STACK_DEPTH
);
5039 th_trace_free(void *value
)
5041 th_trace_t
*th_trace
= value
;
5043 ASSERT(th_trace
->th_refcnt
== 0);
5044 kmem_free(th_trace
, sizeof (*th_trace
));
5048 * Find or create the per-thread hash table used to track object references.
5049 * The ipst argument is NULL if we shouldn't allocate.
5051 * Accesses per-thread data, so there's no need to lock here.
5054 th_trace_gethash(ip_stack_t
*ipst
)
5058 if ((thh
= tsd_get(ip_thread_data
)) == NULL
&& ipst
!= NULL
) {
5061 size_t objsize
, rshift
;
5064 if ((thh
= kmem_alloc(sizeof (*thh
), KM_NOSLEEP
)) == NULL
)
5066 (void) snprintf(name
, sizeof (name
), "th_trace_%p",
5070 * We use mod_hash_create_extended here rather than the more
5071 * obvious mod_hash_create_ptrhash because the latter has a
5072 * hard-coded KM_SLEEP, and we'd prefer to fail rather than
5075 objsize
= MAX(MAX(sizeof (ill_t
), sizeof (ipif_t
)),
5076 MAX(sizeof (ire_t
), sizeof (ncec_t
)));
5077 rshift
= highbit(objsize
);
5078 mh
= mod_hash_create_extended(name
, 64, mod_hash_null_keydtor
,
5079 th_trace_free
, mod_hash_byptr
, (void *)rshift
,
5080 mod_hash_ptrkey_cmp
, KM_NOSLEEP
);
5082 kmem_free(thh
, sizeof (*thh
));
5086 thh
->thh_ipst
= ipst
;
5088 * We trace ills, ipifs, ires, and nces. All of these are
5089 * per-IP-stack, so the lock on the thread list is as well.
5091 rw_enter(&ip_thread_rwlock
, RW_WRITER
);
5092 list_insert_tail(&ip_thread_list
, thh
);
5093 rw_exit(&ip_thread_rwlock
);
5094 retv
= tsd_set(ip_thread_data
, thh
);
5097 return (thh
!= NULL
? thh
->thh_hash
: NULL
);
5101 th_trace_ref(const void *obj
, ip_stack_t
*ipst
)
5103 th_trace_t
*th_trace
;
5107 if ((mh
= th_trace_gethash(ipst
)) == NULL
)
5111 * Attempt to locate the trace buffer for this obj and thread.
5112 * If it does not exist, then allocate a new trace buffer and
5113 * insert into the hash.
5115 if (mod_hash_find(mh
, (mod_hash_key_t
)obj
, &val
) == MH_ERR_NOTFOUND
) {
5116 th_trace
= kmem_zalloc(sizeof (th_trace_t
), KM_NOSLEEP
);
5117 if (th_trace
== NULL
)
5120 th_trace
->th_id
= curthread
;
5121 if (mod_hash_insert(mh
, (mod_hash_key_t
)obj
,
5122 (mod_hash_val_t
)th_trace
) != 0) {
5123 kmem_free(th_trace
, sizeof (th_trace_t
));
5127 th_trace
= (th_trace_t
*)val
;
5130 ASSERT(th_trace
->th_refcnt
>= 0 &&
5131 th_trace
->th_refcnt
< TR_BUF_MAX
- 1);
5133 th_trace
->th_refcnt
++;
5134 th_trace_rrecord(th_trace
);
5139 * For the purpose of tracing a reference release, we assume that global
5140 * tracing is always on and that the same thread initiated the reference hold
5144 th_trace_unref(const void *obj
)
5148 th_trace_t
*th_trace
;
5151 mh
= th_trace_gethash(NULL
);
5152 retv
= mod_hash_find(mh
, (mod_hash_key_t
)obj
, &val
);
5154 th_trace
= (th_trace_t
*)val
;
5156 ASSERT(th_trace
->th_refcnt
> 0);
5157 th_trace
->th_refcnt
--;
5158 th_trace_rrecord(th_trace
);
5162 * If tracing has been disabled, then we assume that the reference counts are
5163 * now useless, and we clear them out before destroying the entries.
5166 th_trace_cleanup(const void *obj
, boolean_t trace_disable
)
5171 th_trace_t
*th_trace
;
5174 rw_enter(&ip_thread_rwlock
, RW_READER
);
5175 for (thh
= list_head(&ip_thread_list
); thh
!= NULL
;
5176 thh
= list_next(&ip_thread_list
, thh
)) {
5177 if (mod_hash_find(mh
= thh
->thh_hash
, (mod_hash_key_t
)obj
,
5179 th_trace
= (th_trace_t
*)val
;
5181 th_trace
->th_refcnt
= 0;
5182 retv
= mod_hash_destroy(mh
, (mod_hash_key_t
)obj
);
5186 rw_exit(&ip_thread_rwlock
);
5190 ipif_trace_ref(ipif_t
*ipif
)
5192 ASSERT(MUTEX_HELD(&ipif
->ipif_ill
->ill_lock
));
5194 if (ipif
->ipif_trace_disable
)
5197 if (!th_trace_ref(ipif
, ipif
->ipif_ill
->ill_ipst
)) {
5198 ipif
->ipif_trace_disable
= B_TRUE
;
5199 ipif_trace_cleanup(ipif
);
5204 ipif_untrace_ref(ipif_t
*ipif
)
5206 ASSERT(MUTEX_HELD(&ipif
->ipif_ill
->ill_lock
));
5208 if (!ipif
->ipif_trace_disable
)
5209 th_trace_unref(ipif
);
5213 ill_trace_ref(ill_t
*ill
)
5215 ASSERT(MUTEX_HELD(&ill
->ill_lock
));
5217 if (ill
->ill_trace_disable
)
5220 if (!th_trace_ref(ill
, ill
->ill_ipst
)) {
5221 ill
->ill_trace_disable
= B_TRUE
;
5222 ill_trace_cleanup(ill
);
5227 ill_untrace_ref(ill_t
*ill
)
5229 ASSERT(MUTEX_HELD(&ill
->ill_lock
));
5231 if (!ill
->ill_trace_disable
)
5232 th_trace_unref(ill
);
5236 * Called when ipif is unplumbed or when memory alloc fails. Note that on
5237 * failure, ipif_trace_disable is set.
5240 ipif_trace_cleanup(const ipif_t
*ipif
)
5242 th_trace_cleanup(ipif
, ipif
->ipif_trace_disable
);
5246 * Called when ill is unplumbed or when memory alloc fails. Note that on
5247 * failure, ill_trace_disable is set.
5250 ill_trace_cleanup(const ill_t
*ill
)
5252 th_trace_cleanup(ill
, ill
->ill_trace_disable
);
5257 ipif_refhold_locked(ipif_t
*ipif
)
5259 ASSERT(MUTEX_HELD(&ipif
->ipif_ill
->ill_lock
));
5260 ipif
->ipif_refcnt
++;
5261 IPIF_TRACE_REF(ipif
);
5265 ipif_refhold(ipif_t
*ipif
)
5269 ill
= ipif
->ipif_ill
;
5270 mutex_enter(&ill
->ill_lock
);
5271 ipif
->ipif_refcnt
++;
5272 IPIF_TRACE_REF(ipif
);
5273 mutex_exit(&ill
->ill_lock
);
5277 * Must not be called while holding any locks. Otherwise if this is
5278 * the last reference to be released there is a chance of recursive mutex
5279 * panic due to ipif_refrele -> ipif_ill_refrele_tail -> qwriter_ip trying
5280 * to restart an ioctl.
5283 ipif_refrele(ipif_t
*ipif
)
5287 ill
= ipif
->ipif_ill
;
5289 mutex_enter(&ill
->ill_lock
);
5290 ASSERT(ipif
->ipif_refcnt
!= 0);
5291 ipif
->ipif_refcnt
--;
5292 IPIF_UNTRACE_REF(ipif
);
5293 if (ipif
->ipif_refcnt
!= 0) {
5294 mutex_exit(&ill
->ill_lock
);
5298 /* Drops the ill_lock */
5299 ipif_ill_refrele_tail(ill
);
5303 ipif_get_next_ipif(ipif_t
*curr
, ill_t
*ill
)
5307 mutex_enter(&ill
->ill_lock
);
5308 for (ipif
= (curr
== NULL
? ill
->ill_ipif
: curr
->ipif_next
);
5309 ipif
!= NULL
; ipif
= ipif
->ipif_next
) {
5310 if (IPIF_IS_CONDEMNED(ipif
))
5312 ipif_refhold_locked(ipif
);
5313 mutex_exit(&ill
->ill_lock
);
5316 mutex_exit(&ill
->ill_lock
);
5321 * TODO: make this table extendible at run time
5322 * Return a pointer to the mac type info for 'mac_type'
5325 ip_m_lookup(t_uscalar_t mac_type
)
5329 for (ipm
= ip_m_tbl
; ipm
< A_END(ip_m_tbl
); ipm
++)
5330 if (ipm
->ip_m_mac_type
== mac_type
)
5336 * Make a link layer address from the multicast IP address *addr.
5337 * To form the link layer address, invoke the ip_m_v*mapping function
5338 * associated with the link-layer type.
5341 ip_mcast_mapping(ill_t
*ill
, uchar_t
*addr
, uchar_t
*hwaddr
)
5345 if (ill
->ill_net_type
== IRE_IF_NORESOLVER
)
5348 ASSERT(addr
!= NULL
);
5350 ipm
= ip_m_lookup(ill
->ill_mactype
);
5352 (ill
->ill_isv6
&& ipm
->ip_m_v6mapping
== NULL
) ||
5353 (!ill
->ill_isv6
&& ipm
->ip_m_v4mapping
== NULL
)) {
5354 ip0dbg(("no mapping for ill %s mactype 0x%x\n",
5355 ill
->ill_name
, ill
->ill_mactype
));
5359 (*ipm
->ip_m_v6mapping
)(ill
, addr
, hwaddr
);
5361 (*ipm
->ip_m_v4mapping
)(ill
, addr
, hwaddr
);
5365 * Returns B_FALSE if the IPv4 netmask pointed by `mask' is non-contiguous.
5366 * Otherwise returns B_TRUE.
5368 * The netmask can be verified to be contiguous with 32 shifts and or
5369 * operations. Take the contiguous mask (in host byte order) and compute
5370 * mask | mask << 1 | mask << 2 | ... | mask << 31
5371 * the result will be the same as the 'mask' for contiguous mask.
5374 ip_contiguous_mask(uint32_t mask
)
5379 for (i
= 1; i
< 32; i
++)
5386 * ip_rt_add is called to add an IPv4 route to the forwarding table.
5387 * ill is passed in to associate it with the correct interface.
5388 * If ire_arg is set, then we return the held IRE in that location.
5391 ip_rt_add(ipaddr_t dst_addr
, ipaddr_t mask
, ipaddr_t gw_addr
,
5392 ipaddr_t src_addr
, int flags
, ill_t
*ill
, ire_t
**ire_arg
,
5393 boolean_t ioctl_msg
, ip_stack_t
*ipst
, zoneid_t zoneid
)
5396 ire_t
*gw_ire
= NULL
;
5397 ipif_t
*ipif
= NULL
;
5399 int match_flags
= MATCH_IRE_TYPE
;
5400 boolean_t unbound
= B_FALSE
;
5402 ip1dbg(("ip_rt_add:"));
5404 if (ire_arg
!= NULL
)
5407 /* disallow non-contiguous netmasks */
5408 if (!ip_contiguous_mask(ntohl(mask
)))
5412 * If this is the case of RTF_HOST being set, then we set the netmask
5413 * to all ones (regardless if one was supplied).
5415 if (flags
& RTF_HOST
)
5416 mask
= IP_HOST_MASK
;
5419 * Prevent routes with a zero gateway from being created (since
5420 * interfaces can currently be plumbed and brought up no assigned
5424 return (ENETUNREACH
);
5426 * Get the ipif, if any, corresponding to the gw_addr
5427 * If -ifp was specified we restrict ourselves to the ill, otherwise
5428 * we match on the gatway and destination to handle unnumbered pt-pt
5432 ipif
= ipif_lookup_addr(gw_addr
, ill
, ALL_ZONES
, ipst
);
5434 ipif
= ipif_lookup_interface(gw_addr
, dst_addr
, ipst
);
5436 if (IS_VNI(ipif
->ipif_ill
)) {
5443 * GateD will attempt to create routes with a loopback interface
5444 * address as the gateway and with RTF_GATEWAY set. We allow
5445 * these routes to be added, but create them as interface routes
5446 * since the gateway is an interface address.
5448 if ((ipif
!= NULL
) && (ipif
->ipif_ire_type
== IRE_LOOPBACK
)) {
5449 flags
&= ~RTF_GATEWAY
;
5450 if (gw_addr
== INADDR_LOOPBACK
&& dst_addr
== INADDR_LOOPBACK
&&
5451 mask
== IP_HOST_MASK
) {
5452 ire
= ire_ftable_lookup_v4(dst_addr
, 0, 0, IRE_LOOPBACK
,
5453 NULL
, ALL_ZONES
, MATCH_IRE_TYPE
, 0, ipst
, NULL
);
5459 ip1dbg(("ip_rt_add: 0x%p creating IRE 0x%x"
5460 "for 0x%x\n", (void *)ipif
,
5461 ipif
->ipif_ire_type
,
5462 ntohl(ipif
->ipif_lcl_addr
)));
5464 (uchar_t
*)&dst_addr
, /* dest address */
5465 (uchar_t
*)&mask
, /* mask */
5466 NULL
, /* no gateway */
5467 ipif
->ipif_ire_type
, /* LOOPBACK */
5470 (ipif
->ipif_flags
& IPIF_PRIVATE
) ? RTF_PRIVATE
: 0,
5477 /* src address assigned by the caller? */
5478 if ((src_addr
!= INADDR_ANY
) && (flags
& RTF_SETSRC
))
5479 ire
->ire_setsrc_addr
= src_addr
;
5481 nire
= ire_add(ire
);
5484 * In the result of failure, ire_add() will have
5485 * already deleted the ire in question, so there
5486 * is no need to do that here.
5492 * Check if it was a duplicate entry. This handles
5493 * the case of two racing route adds for the same route
5496 ASSERT(nire
->ire_identical_ref
> 1);
5508 * Traditionally, interface routes are ones where RTF_GATEWAY isn't set
5509 * and the gateway address provided is one of the system's interface
5510 * addresses. By using the routing socket interface and supplying an
5511 * RTA_IFP sockaddr with an interface index, an alternate method of
5512 * specifying an interface route to be created is available which uses
5513 * the interface index that specifies the outgoing interface rather than
5514 * the address of an outgoing interface (which may not be able to
5515 * uniquely identify an interface). When coupled with the RTF_GATEWAY
5516 * flag, routes can be specified which not only specify the next-hop to
5517 * be used when routing to a certain prefix, but also which outgoing
5518 * interface should be used.
5520 * Previously, interfaces would have unique addresses assigned to them
5521 * and so the address assigned to a particular interface could be used
5522 * to identify a particular interface. One exception to this was the
5523 * case of an unnumbered interface (where IPIF_UNNUMBERED was set).
5525 * With the advent of IPv6 and its link-local addresses, this
5526 * restriction was relaxed and interfaces could share addresses between
5527 * themselves. In fact, typically all of the link-local interfaces on
5528 * an IPv6 node or router will have the same link-local address. In
5529 * order to differentiate between these interfaces, the use of an
5530 * interface index is necessary and this index can be carried inside a
5531 * RTA_IFP sockaddr (which is actually a sockaddr_dl). One restriction
5532 * of using the interface index, however, is that all of the ipif's that
5533 * are part of an ill have the same index and so the RTA_IFP sockaddr
5534 * cannot be used to differentiate between ipif's (or logical
5535 * interfaces) that belong to the same ill (physical interface).
5537 * For example, in the following case involving IPv4 interfaces and
5538 * logical interfaces
5540 * 192.0.2.32 255.255.255.224 192.0.2.33 U if0
5541 * 192.0.2.32 255.255.255.224 192.0.2.34 U if0
5542 * 192.0.2.32 255.255.255.224 192.0.2.35 U if0
5544 * the ipif's corresponding to each of these interface routes can be
5545 * uniquely identified by the "gateway" (actually interface address).
5547 * In this case involving multiple IPv6 default routes to a particular
5548 * link-local gateway, the use of RTA_IFP is necessary to specify which
5549 * default route is of interest:
5551 * default fe80::123:4567:89ab:cdef U if0
5552 * default fe80::123:4567:89ab:cdef U if1
5555 /* RTF_GATEWAY not set */
5556 if (!(flags
& RTF_GATEWAY
)) {
5558 * Whether or not ill (RTA_IFP) is set, we require that
5559 * the gateway is one of our local addresses.
5562 return (ENETUNREACH
);
5565 * We use MATCH_IRE_ILL here. If the caller specified an
5566 * interface (from the RTA_IFP sockaddr) we use it, otherwise
5567 * we use the ill derived from the gateway address.
5568 * We can always match the gateway address since we record it
5569 * in ire_gateway_addr.
5570 * We don't allow RTA_IFP to specify a different ill than the
5571 * one matching the ipif to make sure we can delete the route.
5573 match_flags
|= MATCH_IRE_GW
| MATCH_IRE_ILL
;
5575 ill
= ipif
->ipif_ill
;
5576 } else if (ill
!= ipif
->ipif_ill
) {
5582 * We check for an existing entry at this point.
5584 * Since a netmask isn't passed in via the ioctl interface
5585 * (SIOCADDRT), we don't check for a matching netmask in that
5589 match_flags
|= MATCH_IRE_MASK
;
5590 ire
= ire_ftable_lookup_v4(dst_addr
, mask
, gw_addr
,
5591 IRE_INTERFACE
, ill
, ALL_ZONES
, match_flags
, 0, ipst
, NULL
);
5599 * Some software (for example, GateD and Sun Cluster) attempts
5600 * to create (what amount to) IRE_PREFIX routes with the
5601 * loopback address as the gateway. This is primarily done to
5602 * set up prefixes with the RTF_REJECT flag set (for example,
5603 * when generating aggregate routes.)
5605 * If the IRE type (as defined by ill->ill_net_type) would be
5606 * IRE_LOOPBACK, then we map the request into a
5607 * IRE_IF_NORESOLVER. We also OR in the RTF_BLACKHOLE flag as
5608 * these interface routes, by definition, can only be that.
5610 * Needless to say, the real IRE_LOOPBACK is NOT created by this
5611 * routine, but rather using ire_create() directly.
5614 type
= ill
->ill_net_type
;
5615 if (type
== IRE_LOOPBACK
) {
5616 type
= IRE_IF_NORESOLVER
;
5617 flags
|= RTF_BLACKHOLE
;
5621 * Create a copy of the IRE_IF_NORESOLVER or
5622 * IRE_IF_RESOLVER with the modified address, netmask, and
5626 (uchar_t
*)&dst_addr
,
5628 (uint8_t *)&gw_addr
,
5639 /* src address assigned by the caller? */
5640 if ((src_addr
!= INADDR_ANY
) && (flags
& RTF_SETSRC
))
5641 ire
->ire_setsrc_addr
= src_addr
;
5643 nire
= ire_add(ire
);
5646 * In the result of failure, ire_add() will have
5647 * already deleted the ire in question, so there
5648 * is no need to do that here.
5654 * Check if it was a duplicate entry. This handles
5655 * the case of two racing route adds for the same route
5668 * Get an interface IRE for the specified gateway.
5669 * If we don't have an IRE_IF_NORESOLVER or IRE_IF_RESOLVER for the
5670 * gateway, it is currently unreachable and we fail the request
5671 * accordingly. We reject any RTF_GATEWAY routes where the gateway
5672 * is an IRE_LOCAL or IRE_LOOPBACK.
5673 * If RTA_IFP was specified we look on that particular ill.
5676 match_flags
|= MATCH_IRE_ILL
;
5678 /* Check whether the gateway is reachable. */
5680 type
= IRE_INTERFACE
| IRE_LOCAL
| IRE_LOOPBACK
;
5681 if (flags
& RTF_INDIRECT
)
5682 type
|= IRE_OFFLINK
;
5684 gw_ire
= ire_ftable_lookup_v4(gw_addr
, 0, 0, type
, ill
,
5685 ALL_ZONES
, match_flags
, 0, ipst
, NULL
);
5686 if (gw_ire
== NULL
) {
5688 * With IPMP, we allow host routes to influence in.mpathd's
5689 * target selection. However, if the test addresses are on
5690 * their own network, the above lookup will fail since the
5691 * underlying IRE_INTERFACEs are marked hidden. So allow
5692 * hidden test IREs to be found and try again.
5694 if (!(match_flags
& MATCH_IRE_TESTHIDDEN
)) {
5695 match_flags
|= MATCH_IRE_TESTHIDDEN
;
5700 return (ENETUNREACH
);
5702 if (gw_ire
->ire_type
& (IRE_LOCAL
|IRE_LOOPBACK
)) {
5703 ire_refrele(gw_ire
);
5706 return (ENETUNREACH
);
5709 if (ill
== NULL
&& !(flags
& RTF_INDIRECT
)) {
5711 if (ipst
->ips_ip_strict_src_multihoming
> 0)
5712 ill
= gw_ire
->ire_ill
;
5716 * We create one of three types of IREs as a result of this request
5717 * based on the netmask. A netmask of all ones (which is automatically
5718 * assumed when RTF_HOST is set) results in an IRE_HOST being created.
5719 * An all zeroes netmask implies a default route so an IRE_DEFAULT is
5720 * created. Otherwise, an IRE_PREFIX route is created for the
5721 * destination prefix.
5723 if (mask
== IP_HOST_MASK
)
5730 /* check for a duplicate entry */
5731 ire
= ire_ftable_lookup_v4(dst_addr
, mask
, gw_addr
, type
, ill
,
5732 ALL_ZONES
, match_flags
| MATCH_IRE_MASK
| MATCH_IRE_GW
, 0, ipst
,
5737 ire_refrele(gw_ire
);
5742 /* Create the IRE. */
5744 (uchar_t
*)&dst_addr
, /* dest address */
5745 (uchar_t
*)&mask
, /* mask */
5746 (uchar_t
*)&gw_addr
, /* gateway address */
5747 (ushort_t
)type
, /* IRE type */
5756 ire_refrele(gw_ire
);
5760 /* src address assigned by the caller? */
5761 if ((src_addr
!= INADDR_ANY
) && (flags
& RTF_SETSRC
))
5762 ire
->ire_setsrc_addr
= src_addr
;
5764 ire
->ire_unbound
= unbound
;
5767 * POLICY: should we allow an RTF_HOST with address INADDR_ANY?
5768 * SUN/OS socket stuff does but do we really want to allow 0.0.0.0?
5771 /* Add the new IRE. */
5772 nire
= ire_add(ire
);
5775 * In the result of failure, ire_add() will have
5776 * already deleted the ire in question, so there
5777 * is no need to do that here.
5781 ire_refrele(gw_ire
);
5785 * Check if it was a duplicate entry. This handles
5786 * the case of two racing route adds for the same route
5793 ire_refrele(gw_ire
);
5799 if (gw_ire
!= NULL
) {
5800 ire_refrele(gw_ire
);
5805 * Save enough information so that we can recreate the IRE if
5806 * the interface goes down and then up. The metrics associated
5807 * with the route will be saved as well when rts_setmetrics() is
5808 * called after the IRE has been created. In the case where
5809 * memory cannot be allocated, none of this information will be
5812 ill_save_ire(ill
, ire
);
5815 ip_rts_rtmsg(RTM_OLDADD
, ire
, 0, ipst
);
5816 if (ire_arg
!= NULL
) {
5818 * Store the ire that was successfully added into where ire_arg
5819 * points to so that callers don't have to look it up
5820 * themselves (but they are responsible for ire_refrele()ing
5821 * the ire when they are finished with it).
5825 ire_refrele(ire
); /* Held in ire_add */
5833 * ip_rt_delete is called to delete an IPv4 route.
5834 * ill is passed in to associate it with the correct interface.
5838 ip_rt_delete(ipaddr_t dst_addr
, ipaddr_t mask
, ipaddr_t gw_addr
,
5839 uint_t rtm_addrs
, int flags
, ill_t
*ill
, boolean_t ioctl_msg
,
5840 ip_stack_t
*ipst
, zoneid_t zoneid
)
5845 uint_t match_flags
= MATCH_IRE_TYPE
;
5848 ip1dbg(("ip_rt_delete:"));
5850 * If this is the case of RTF_HOST being set, then we set the netmask
5851 * to all ones. Otherwise, we use the netmask if one was supplied.
5853 if (flags
& RTF_HOST
) {
5854 mask
= IP_HOST_MASK
;
5855 match_flags
|= MATCH_IRE_MASK
;
5856 } else if (rtm_addrs
& RTA_NETMASK
) {
5857 match_flags
|= MATCH_IRE_MASK
;
5861 * Note that RTF_GATEWAY is never set on a delete, therefore
5862 * we check if the gateway address is one of our interfaces first,
5863 * and fall back on RTF_GATEWAY routes.
5865 * This makes it possible to delete an original
5866 * IRE_IF_NORESOLVER/IRE_IF_RESOLVER - consistent with SunOS 4.1.
5867 * However, we have RTF_KERNEL set on the ones created by ipif_up
5868 * and those can not be deleted here.
5870 * We use MATCH_IRE_ILL if we know the interface. If the caller
5871 * specified an interface (from the RTA_IFP sockaddr) we use it,
5872 * otherwise we use the ill derived from the gateway address.
5873 * We can always match the gateway address since we record it
5874 * in ire_gateway_addr.
5876 * For more detail on specifying routes by gateway address and by
5877 * interface index, see the comments in ip_rt_add().
5879 ipif
= ipif_lookup_interface(gw_addr
, dst_addr
, ipst
);
5886 ill_match
= ipif
->ipif_ill
;
5888 match_flags
|= MATCH_IRE_ILL
;
5889 if (ipif
->ipif_ire_type
== IRE_LOOPBACK
) {
5890 ire
= ire_ftable_lookup_v4(dst_addr
, mask
, 0,
5891 IRE_LOOPBACK
, ill_match
, ALL_ZONES
, match_flags
, 0,
5895 match_flags
|= MATCH_IRE_GW
;
5896 ire
= ire_ftable_lookup_v4(dst_addr
, mask
, gw_addr
,
5897 IRE_INTERFACE
, ill_match
, ALL_ZONES
, match_flags
,
5900 /* Avoid deleting routes created by kernel from an ipif */
5901 if (ire
!= NULL
&& (ire
->ire_flags
& RTF_KERNEL
)) {
5906 /* Restore in case we didn't find a match */
5907 match_flags
&= ~(MATCH_IRE_GW
|MATCH_IRE_ILL
);
5912 * At this point, the gateway address is not one of our own
5913 * addresses or a matching interface route was not found. We
5914 * set the IRE type to lookup based on whether
5915 * this is a host route, a default route or just a prefix.
5917 * If an ill was passed in, then the lookup is based on an
5918 * interface index so MATCH_IRE_ILL is added to match_flags.
5920 match_flags
|= MATCH_IRE_GW
;
5922 match_flags
|= MATCH_IRE_ILL
;
5923 if (mask
== IP_HOST_MASK
)
5929 ire
= ire_ftable_lookup_v4(dst_addr
, mask
, gw_addr
, type
, ill
,
5930 ALL_ZONES
, match_flags
, 0, ipst
, NULL
);
5943 ill_remove_saved_ire(ill
, ire
);
5945 ip_rts_rtmsg(RTM_OLDDEL
, ire
, 0, ipst
);
5952 * ip_siocaddrt is called to complete processing of an SIOCADDRT IOCTL.
5956 ip_siocaddrt(ipif_t
*dummy_ipif
, sin_t
*dummy_sin
, queue_t
*q
, mblk_t
*mp
,
5957 ip_ioctl_cmd_t
*ipip
, void *dummy_if_req
)
5965 ipif_t
*ipif
= NULL
;
5968 ASSERT(q
->q_next
== NULL
);
5969 ipst
= CONNQ_TO_IPST(q
);
5971 ip1dbg(("ip_siocaddrt:"));
5972 /* Existence of mp1 verified in ip_wput_nondata */
5973 mp1
= mp
->b_cont
->b_cont
;
5974 rt
= (struct rtentry
*)mp1
->b_rptr
;
5976 dst_addr
= ((sin_t
*)&rt
->rt_dst
)->sin_addr
.s_addr
;
5977 gw_addr
= ((sin_t
*)&rt
->rt_gateway
)->sin_addr
.s_addr
;
5980 * If the RTF_HOST flag is on, this is a request to assign a gateway
5981 * to a particular host address. In this case, we set the netmask to
5982 * all ones for the particular destination address. Otherwise,
5983 * determine the netmask to be used based on dst_addr and the interfaces
5986 if (rt
->rt_flags
& RTF_HOST
) {
5987 mask
= IP_HOST_MASK
;
5990 * Note that ip_subnet_mask returns a zero mask in the case of
5991 * default (an all-zeroes address).
5993 mask
= ip_subnet_mask(dst_addr
, &ipif
, ipst
);
5996 error
= ip_rt_add(dst_addr
, mask
, gw_addr
, 0, rt
->rt_flags
, NULL
, NULL
,
5997 B_TRUE
, ipst
, ALL_ZONES
);
6004 * ip_siocdelrt is called to complete processing of an SIOCDELRT IOCTL.
6008 ip_siocdelrt(ipif_t
*dummy_ipif
, sin_t
*dummy_sin
, queue_t
*q
, mblk_t
*mp
,
6009 ip_ioctl_cmd_t
*ipip
, void *dummy_if_req
)
6017 ipif_t
*ipif
= NULL
;
6020 ASSERT(q
->q_next
== NULL
);
6021 ipst
= CONNQ_TO_IPST(q
);
6023 ip1dbg(("ip_siocdelrt:"));
6024 /* Existence of mp1 verified in ip_wput_nondata */
6025 mp1
= mp
->b_cont
->b_cont
;
6026 rt
= (struct rtentry
*)mp1
->b_rptr
;
6028 dst_addr
= ((sin_t
*)&rt
->rt_dst
)->sin_addr
.s_addr
;
6029 gw_addr
= ((sin_t
*)&rt
->rt_gateway
)->sin_addr
.s_addr
;
6032 * If the RTF_HOST flag is on, this is a request to delete a gateway
6033 * to a particular host address. In this case, we set the netmask to
6034 * all ones for the particular destination address. Otherwise,
6035 * determine the netmask to be used based on dst_addr and the interfaces
6038 if (rt
->rt_flags
& RTF_HOST
) {
6039 mask
= IP_HOST_MASK
;
6042 * Note that ip_subnet_mask returns a zero mask in the case of
6043 * default (an all-zeroes address).
6045 mask
= ip_subnet_mask(dst_addr
, &ipif
, ipst
);
6048 error
= ip_rt_delete(dst_addr
, mask
, gw_addr
,
6049 RTA_DST
| RTA_GATEWAY
| RTA_NETMASK
, rt
->rt_flags
, NULL
, B_TRUE
,
6057 * Enqueue the mp onto the ipsq, chained by b_next.
6058 * b_prev stores the function to be executed later, and b_queue the queue
6059 * where this mp originated.
6062 ipsq_enq(ipsq_t
*ipsq
, queue_t
*q
, mblk_t
*mp
, ipsq_func_t func
, int type
,
6066 ipxop_t
*ipx
= ipsq
->ipsq_xop
;
6068 ASSERT(MUTEX_HELD(&ipsq
->ipsq_lock
));
6069 ASSERT(MUTEX_HELD(&ipx
->ipx_lock
));
6070 ASSERT(func
!= NULL
);
6073 mp
->b_prev
= (void *)func
;
6078 if (ipx
->ipx_mptail
!= NULL
) {
6079 ASSERT(ipx
->ipx_mphead
!= NULL
);
6080 ipx
->ipx_mptail
->b_next
= mp
;
6082 ASSERT(ipx
->ipx_mphead
== NULL
);
6083 ipx
->ipx_mphead
= mp
;
6085 ipx
->ipx_mptail
= mp
;
6089 if (ipsq
->ipsq_xopq_mptail
!= NULL
) {
6090 ASSERT(ipsq
->ipsq_xopq_mphead
!= NULL
);
6091 ipsq
->ipsq_xopq_mptail
->b_next
= mp
;
6093 ASSERT(ipsq
->ipsq_xopq_mphead
== NULL
);
6094 ipsq
->ipsq_xopq_mphead
= mp
;
6096 ipsq
->ipsq_xopq_mptail
= mp
;
6097 ipx
->ipx_ipsq_queued
= B_TRUE
;
6101 ASSERT(ipsq
->ipsq_swxop
!= NULL
);
6102 /* only one switch operation is currently allowed */
6103 ASSERT(ipsq
->ipsq_switch_mp
== NULL
);
6104 ipsq
->ipsq_switch_mp
= mp
;
6105 ipx
->ipx_ipsq_queued
= B_TRUE
;
6108 cmn_err(CE_PANIC
, "ipsq_enq %d type \n", type
);
6111 if (CONN_Q(q
) && pending_ill
!= NULL
) {
6112 connp
= Q_TO_CONN(q
);
6113 ASSERT(MUTEX_HELD(&connp
->conn_lock
));
6114 connp
->conn_oper_pending_ill
= pending_ill
;
6119 * Dequeue the next message that requested exclusive access to this IPSQ's
6120 * xop. Specifically:
6122 * 1. If we're still processing the current operation on `ipsq', then
6123 * dequeue the next message for the operation (from ipx_mphead), or
6124 * return NULL if there are no queued messages for the operation.
6125 * These messages are queued via CUR_OP to qwriter_ip() and friends.
6127 * 2. If the current operation on `ipsq' has completed (ipx_current_ipif is
6128 * not set) see if the ipsq has requested an xop switch. If so, switch
6129 * `ipsq' to a different xop. Xop switches only happen when joining or
6130 * leaving IPMP groups and require a careful dance -- see the comments
6131 * in-line below for details. If we're leaving a group xop or if we're
6132 * joining a group xop and become writer on it, then we proceed to (3).
6133 * Otherwise, we return NULL and exit the xop.
6135 * 3. For each IPSQ in the xop, return any switch operation stored on
6136 * ipsq_switch_mp (set via SWITCH_OP); these must be processed before
6137 * any other messages queued on the IPSQ. Otherwise, dequeue the next
6138 * exclusive operation (queued via NEW_OP) stored on ipsq_xopq_mphead.
6139 * Note that if the phyint tied to `ipsq' is not using IPMP there will
6140 * only be one IPSQ in the xop. Otherwise, there will be one IPSQ for
6141 * each phyint in the group, including the IPMP meta-interface phyint.
6144 ipsq_dq(ipsq_t
*ipsq
)
6146 ill_t
*illv4
, *illv6
;
6149 ipsq_t
*leftipsq
= NULL
;
6151 phyint_t
*phyi
= ipsq
->ipsq_phyint
;
6152 ip_stack_t
*ipst
= ipsq
->ipsq_ipst
;
6153 boolean_t emptied
= B_FALSE
;
6156 * Grab all the locks we need in the defined order (ill_g_lock ->
6157 * ipsq_lock -> ipx_lock); ill_g_lock is needed to use ipsq_next.
6159 rw_enter(&ipst
->ips_ill_g_lock
,
6160 ipsq
->ipsq_swxop
!= NULL
? RW_WRITER
: RW_READER
);
6161 mutex_enter(&ipsq
->ipsq_lock
);
6162 ipx
= ipsq
->ipsq_xop
;
6163 mutex_enter(&ipx
->ipx_lock
);
6166 * Dequeue the next message associated with the current exclusive
6167 * operation, if any.
6169 if ((mp
= ipx
->ipx_mphead
) != NULL
) {
6170 ipx
->ipx_mphead
= mp
->b_next
;
6171 if (ipx
->ipx_mphead
== NULL
)
6172 ipx
->ipx_mptail
= NULL
;
6173 mp
->b_next
= (void *)ipsq
;
6177 if (ipx
->ipx_current_ipif
!= NULL
)
6180 if (ipsq
->ipsq_swxop
!= NULL
) {
6182 * The exclusive operation that is now being completed has
6183 * requested a switch to a different xop. This happens
6184 * when an interface joins or leaves an IPMP group. Joins
6185 * happen through SIOCSLIFGROUPNAME (ip_sioctl_groupname()).
6186 * Leaves happen via SIOCSLIFGROUPNAME, interface unplumb
6187 * (phyint_free()), or interface plumb for an ill type
6188 * not in the IPMP group (ip_rput_dlpi_writer()).
6190 * Xop switches are not allowed on the IPMP meta-interface.
6192 ASSERT(phyi
== NULL
|| !(phyi
->phyint_flags
& PHYI_IPMP
));
6193 ASSERT(RW_WRITE_HELD(&ipst
->ips_ill_g_lock
));
6194 DTRACE_PROBE1(ipsq__switch
, (ipsq_t
*), ipsq
);
6196 if (ipsq
->ipsq_swxop
== &ipsq
->ipsq_ownxop
) {
6198 * We're switching back to our own xop, so we have two
6199 * xop's to drain/exit: our own, and the group xop
6200 * that we are leaving.
6202 * First, pull ourselves out of the group ipsq list.
6203 * This is safe since we're writer on ill_g_lock.
6205 ASSERT(ipsq
->ipsq_xop
!= &ipsq
->ipsq_ownxop
);
6207 xopipsq
= ipx
->ipx_ipsq
;
6208 while (xopipsq
->ipsq_next
!= ipsq
)
6209 xopipsq
= xopipsq
->ipsq_next
;
6211 xopipsq
->ipsq_next
= ipsq
->ipsq_next
;
6212 ipsq
->ipsq_next
= ipsq
;
6213 ipsq
->ipsq_xop
= ipsq
->ipsq_swxop
;
6214 ipsq
->ipsq_swxop
= NULL
;
6217 * Second, prepare to exit the group xop. The actual
6218 * ipsq_exit() is done at the end of this function
6219 * since we cannot hold any locks across ipsq_exit().
6220 * Note that although we drop the group's ipx_lock, no
6221 * threads can proceed since we're still ipx_writer.
6224 mutex_exit(&ipx
->ipx_lock
);
6227 * Third, set ipx to point to our own xop (which was
6228 * inactive and therefore can be entered).
6230 ipx
= ipsq
->ipsq_xop
;
6231 mutex_enter(&ipx
->ipx_lock
);
6232 ASSERT(ipx
->ipx_writer
== NULL
);
6233 ASSERT(ipx
->ipx_current_ipif
== NULL
);
6236 * We're switching from our own xop to a group xop.
6237 * The requestor of the switch must ensure that the
6238 * group xop cannot go away (e.g. by ensuring the
6239 * phyint associated with the xop cannot go away).
6241 * If we can become writer on our new xop, then we'll
6242 * do the drain. Otherwise, the current writer of our
6243 * new xop will do the drain when it exits.
6245 * First, splice ourselves into the group IPSQ list.
6246 * This is safe since we're writer on ill_g_lock.
6248 ASSERT(ipsq
->ipsq_xop
== &ipsq
->ipsq_ownxop
);
6250 xopipsq
= ipsq
->ipsq_swxop
->ipx_ipsq
;
6251 while (xopipsq
->ipsq_next
!= ipsq
->ipsq_swxop
->ipx_ipsq
)
6252 xopipsq
= xopipsq
->ipsq_next
;
6254 xopipsq
->ipsq_next
= ipsq
;
6255 ipsq
->ipsq_next
= ipsq
->ipsq_swxop
->ipx_ipsq
;
6256 ipsq
->ipsq_xop
= ipsq
->ipsq_swxop
;
6257 ipsq
->ipsq_swxop
= NULL
;
6260 * Second, exit our own xop, since it's now unused.
6261 * This is safe since we've got the only reference.
6263 ASSERT(ipx
->ipx_writer
== curthread
);
6264 ipx
->ipx_writer
= NULL
;
6265 VERIFY(--ipx
->ipx_reentry_cnt
== 0);
6266 ipx
->ipx_ipsq_queued
= B_FALSE
;
6267 mutex_exit(&ipx
->ipx_lock
);
6270 * Third, set ipx to point to our new xop, and check
6271 * if we can become writer on it. If we cannot, then
6272 * the current writer will drain the IPSQ group when
6273 * it exits. Our ipsq_xop is guaranteed to be stable
6274 * because we're still holding ipsq_lock.
6276 ipx
= ipsq
->ipsq_xop
;
6277 mutex_enter(&ipx
->ipx_lock
);
6278 if (ipx
->ipx_writer
!= NULL
||
6279 ipx
->ipx_current_ipif
!= NULL
) {
6285 * Fourth, become writer on our new ipx before we continue
6286 * with the drain. Note that we never dropped ipsq_lock
6287 * above, so no other thread could've raced with us to
6288 * become writer first. Also, we're holding ipx_lock, so
6289 * no other thread can examine the ipx right now.
6291 ASSERT(ipx
->ipx_current_ipif
== NULL
);
6292 ASSERT(ipx
->ipx_mphead
== NULL
&& ipx
->ipx_mptail
== NULL
);
6293 VERIFY(ipx
->ipx_reentry_cnt
++ == 0);
6294 ipx
->ipx_writer
= curthread
;
6295 ipx
->ipx_forced
= B_FALSE
;
6297 ipx
->ipx_depth
= getpcstack(ipx
->ipx_stack
, IPX_STACK_DEPTH
);
6304 * So that other operations operate on a consistent and
6305 * complete phyint, a switch message on an IPSQ must be
6306 * handled prior to any other operations on that IPSQ.
6308 if ((mp
= xopipsq
->ipsq_switch_mp
) != NULL
) {
6309 xopipsq
->ipsq_switch_mp
= NULL
;
6310 ASSERT(mp
->b_next
== NULL
);
6311 mp
->b_next
= (void *)xopipsq
;
6315 if ((mp
= xopipsq
->ipsq_xopq_mphead
) != NULL
) {
6316 xopipsq
->ipsq_xopq_mphead
= mp
->b_next
;
6317 if (xopipsq
->ipsq_xopq_mphead
== NULL
)
6318 xopipsq
->ipsq_xopq_mptail
= NULL
;
6319 mp
->b_next
= (void *)xopipsq
;
6322 } while ((xopipsq
= xopipsq
->ipsq_next
) != ipsq
);
6325 * There are no messages. Further, we are holding ipx_lock, hence no
6326 * new messages can end up on any IPSQ in the xop.
6328 ipx
->ipx_writer
= NULL
;
6329 ipx
->ipx_forced
= B_FALSE
;
6330 VERIFY(--ipx
->ipx_reentry_cnt
== 0);
6331 ipx
->ipx_ipsq_queued
= B_FALSE
;
6337 mutex_exit(&ipx
->ipx_lock
);
6338 mutex_exit(&ipsq
->ipsq_lock
);
6341 * If we completely emptied the xop, then wake up any threads waiting
6342 * to enter any of the IPSQ's associated with it.
6347 if ((phyi
= xopipsq
->ipsq_phyint
) == NULL
)
6350 illv4
= phyi
->phyint_illv4
;
6351 illv6
= phyi
->phyint_illv6
;
6353 GRAB_ILL_LOCKS(illv4
, illv6
);
6355 cv_broadcast(&illv4
->ill_cv
);
6357 cv_broadcast(&illv6
->ill_cv
);
6358 RELEASE_ILL_LOCKS(illv4
, illv6
);
6359 } while ((xopipsq
= xopipsq
->ipsq_next
) != ipsq
);
6361 rw_exit(&ipst
->ips_ill_g_lock
);
6364 * Now that all locks are dropped, exit the IPSQ we left.
6366 if (leftipsq
!= NULL
)
6367 ipsq_exit(leftipsq
);
6373 * Return completion status of previously initiated DLPI operations on
6374 * ills in the purview of an ipsq.
6377 ipsq_dlpi_done(ipsq_t
*ipsq
)
6383 ASSERT(RW_LOCK_HELD(&ipsq
->ipsq_ipst
->ips_ill_g_lock
));
6388 * The only current users of this function are ipsq_try_enter
6389 * and ipsq_enter which have made sure that ipsq_writer is
6390 * NULL before we reach here. ill_dlpi_pending is modified
6391 * only by an ipsq writer
6393 ASSERT(ipsq
->ipsq_xop
->ipx_writer
== NULL
);
6394 phyi
= ipsq
->ipsq_phyint
;
6396 * phyi could be NULL if a phyint that is part of an
6397 * IPMP group is being unplumbed. A more detailed
6398 * comment is in ipmp_grp_update_kstats()
6401 ill
= phyi
->phyint_illv4
;
6403 (ill
->ill_dlpi_pending
!= DL_PRIM_INVAL
||
6404 ill
->ill_arl_dlpi_pending
))
6407 ill
= phyi
->phyint_illv6
;
6409 ill
->ill_dlpi_pending
!= DL_PRIM_INVAL
)
6413 } while ((ipsq
= ipsq
->ipsq_next
) != ipsq_start
);
6419 * Enter the ipsq corresponding to ill, by waiting synchronously till
6420 * we can enter the ipsq exclusively. Unless 'force' is used, the ipsq
6421 * will have to drain completely before ipsq_enter returns success.
6422 * ipx_current_ipif will be set if some exclusive op is in progress,
6423 * and the ipsq_exit logic will start the next enqueued op after
6424 * completion of the current op. If 'force' is used, we don't wait
6425 * for the enqueued ops. This is needed when a conn_close wants to
6426 * enter the ipsq and abort an ioctl that is somehow stuck. Unplumb
6427 * of an ill can also use this option. But we dont' use it currently.
6429 #define ENTER_SQ_WAIT_TICKS 100
6431 ipsq_enter(ill_t
*ill
, boolean_t force
, int type
)
6435 boolean_t waited_enough
= B_FALSE
;
6436 ip_stack_t
*ipst
= ill
->ill_ipst
;
6439 * Note that the relationship between ill and ipsq is fixed as long as
6440 * the ill is not ILL_CONDEMNED. Holding ipsq_lock ensures the
6441 * relationship between the IPSQ and xop cannot change. However,
6442 * since we cannot hold ipsq_lock across the cv_wait(), it may change
6443 * while we're waiting. We wait on ill_cv and rely on ipsq_exit()
6444 * waking up all ills in the xop when it becomes available.
6447 rw_enter(&ipst
->ips_ill_g_lock
, RW_READER
);
6448 mutex_enter(&ill
->ill_lock
);
6449 if (ill
->ill_state_flags
& ILL_CONDEMNED
) {
6450 mutex_exit(&ill
->ill_lock
);
6451 rw_exit(&ipst
->ips_ill_g_lock
);
6455 ipsq
= ill
->ill_phyint
->phyint_ipsq
;
6456 mutex_enter(&ipsq
->ipsq_lock
);
6457 ipx
= ipsq
->ipsq_xop
;
6458 mutex_enter(&ipx
->ipx_lock
);
6460 if (ipx
->ipx_writer
== NULL
&& (type
== CUR_OP
||
6461 (ipx
->ipx_current_ipif
== NULL
&& ipsq_dlpi_done(ipsq
)) ||
6465 rw_exit(&ipst
->ips_ill_g_lock
);
6467 if (!force
|| ipx
->ipx_writer
!= NULL
) {
6468 mutex_exit(&ipx
->ipx_lock
);
6469 mutex_exit(&ipsq
->ipsq_lock
);
6470 cv_wait(&ill
->ill_cv
, &ill
->ill_lock
);
6472 mutex_exit(&ipx
->ipx_lock
);
6473 mutex_exit(&ipsq
->ipsq_lock
);
6474 (void) cv_reltimedwait(&ill
->ill_cv
,
6475 &ill
->ill_lock
, ENTER_SQ_WAIT_TICKS
, TR_CLOCK_TICK
);
6476 waited_enough
= B_TRUE
;
6478 mutex_exit(&ill
->ill_lock
);
6481 ASSERT(ipx
->ipx_mphead
== NULL
&& ipx
->ipx_mptail
== NULL
);
6482 ASSERT(ipx
->ipx_reentry_cnt
== 0);
6483 ipx
->ipx_writer
= curthread
;
6484 ipx
->ipx_forced
= (ipx
->ipx_current_ipif
!= NULL
);
6485 ipx
->ipx_reentry_cnt
++;
6487 ipx
->ipx_depth
= getpcstack(ipx
->ipx_stack
, IPX_STACK_DEPTH
);
6489 mutex_exit(&ipx
->ipx_lock
);
6490 mutex_exit(&ipsq
->ipsq_lock
);
6491 mutex_exit(&ill
->ill_lock
);
6492 rw_exit(&ipst
->ips_ill_g_lock
);
6498 * ipif_set_values() has a constraint that it cannot drop the ips_ill_g_lock
6499 * across the call to the core interface ipsq_try_enter() and hence calls this
6500 * function directly. This is explained more fully in ipif_set_values().
6501 * In order to support the above constraint, ipsq_try_enter is implemented as
6502 * a wrapper that grabs the ips_ill_g_lock and calls this function subsequently
6505 ipsq_try_enter_internal(ill_t
*ill
, queue_t
*q
, mblk_t
*mp
, ipsq_func_t func
,
6506 int type
, boolean_t reentry_ok
)
6510 ip_stack_t
*ipst
= ill
->ill_ipst
;
6514 * ill_g_lock -> conn_lock -> ill_lock -> ipsq_lock -> ipx_lock.
6516 * ipx of an ipsq can't change when ipsq_lock is held.
6518 ASSERT(RW_LOCK_HELD(&ipst
->ips_ill_g_lock
));
6520 mutex_enter(&ill
->ill_lock
);
6521 ipsq
= ill
->ill_phyint
->phyint_ipsq
;
6522 mutex_enter(&ipsq
->ipsq_lock
);
6523 ipx
= ipsq
->ipsq_xop
;
6524 mutex_enter(&ipx
->ipx_lock
);
6527 * 1. Enter the ipsq if we are already writer and reentry is ok.
6528 * (Note: If the caller does not specify reentry_ok then neither
6529 * 'func' nor any of its callees must ever attempt to enter the ipsq
6530 * again. Otherwise it can lead to an infinite loop
6531 * 2. Enter the ipsq if there is no current writer and this attempted
6532 * entry is part of the current operation
6533 * 3. Enter the ipsq if there is no current writer and this is a new
6534 * operation and the operation queue is empty and there is no
6535 * operation currently in progress and if all previously initiated
6536 * DLPI operations have completed.
6538 if ((ipx
->ipx_writer
== curthread
&& reentry_ok
) ||
6539 (ipx
->ipx_writer
== NULL
&& (type
== CUR_OP
|| (type
== NEW_OP
&&
6540 !ipx
->ipx_ipsq_queued
&& ipx
->ipx_current_ipif
== NULL
&&
6541 ipsq_dlpi_done(ipsq
))))) {
6543 ipx
->ipx_reentry_cnt
++;
6544 ipx
->ipx_writer
= curthread
;
6545 ipx
->ipx_forced
= B_FALSE
;
6546 mutex_exit(&ipx
->ipx_lock
);
6547 mutex_exit(&ipsq
->ipsq_lock
);
6548 mutex_exit(&ill
->ill_lock
);
6549 RELEASE_CONN_LOCK(q
);
6551 ipx
->ipx_depth
= getpcstack(ipx
->ipx_stack
, IPX_STACK_DEPTH
);
6557 ipsq_enq(ipsq
, q
, mp
, func
, type
, ill
);
6559 mutex_exit(&ipx
->ipx_lock
);
6560 mutex_exit(&ipsq
->ipsq_lock
);
6561 mutex_exit(&ill
->ill_lock
);
6562 RELEASE_CONN_LOCK(q
);
6567 * The ipsq_t (ipsq) is the synchronization data structure used to serialize
6568 * certain critical operations like plumbing (i.e. most set ioctls), etc.
6569 * There is one ipsq per phyint. The ipsq
6570 * serializes exclusive ioctls issued by applications on a per ipsq basis in
6571 * ipsq_xopq_mphead. It also protects against multiple threads executing in
6572 * the ipsq. Responses from the driver pertain to the current ioctl (say a
6573 * DL_BIND_ACK in response to a DL_BIND_REQ initiated as part of bringing
6574 * up the interface) and are enqueued in ipx_mphead.
6576 * If a thread does not want to reenter the ipsq when it is already writer,
6577 * it must make sure that the specified reentry point to be called later
6578 * when the ipsq is empty, nor any code path starting from the specified reentry
6579 * point must never ever try to enter the ipsq again. Otherwise it can lead
6580 * to an infinite loop. The reentry point ip_rput_dlpi_writer is an example.
6581 * When the thread that is currently exclusive finishes, it (ipsq_exit)
6582 * dequeues the requests waiting to become exclusive in ipx_mphead and calls
6583 * the reentry point. When the list at ipx_mphead becomes empty ipsq_exit
6584 * proceeds to dequeue the next ioctl in ipsq_xopq_mphead and start the next
6585 * ioctl if the current ioctl has completed. If the current ioctl is still
6586 * in progress it simply returns. The current ioctl could be waiting for
6587 * a response from another module (the driver or could be waiting for
6588 * the ipif/ill/ire refcnts to drop to zero. In such a case the ipx_pending_mp
6589 * and ipx_pending_ipif are set. ipx_current_ipif is set throughout the
6590 * execution of the ioctl and ipsq_exit does not start the next ioctl unless
6591 * ipx_current_ipif is NULL which happens only once the ioctl is complete and
6592 * all associated DLPI operations have completed.
6596 * Try to enter the IPSQ corresponding to `ipif' or `ill' exclusively (`ipif'
6597 * and `ill' cannot both be specified). Returns a pointer to the entered IPSQ
6598 * on success, or NULL on failure. The caller ensures ipif/ill is valid by
6599 * refholding it as necessary. If the IPSQ cannot be entered and `func' is
6600 * non-NULL, then `func' will be called back with `q' and `mp' once the IPSQ
6601 * can be entered. If `func' is NULL, then `q' and `mp' are ignored.
6604 ipsq_try_enter(ipif_t
*ipif
, ill_t
*ill
, queue_t
*q
, mblk_t
*mp
,
6605 ipsq_func_t func
, int type
, boolean_t reentry_ok
)
6610 /* Only 1 of ipif or ill can be specified */
6611 ASSERT((ipif
!= NULL
) ^ (ill
!= NULL
));
6614 ill
= ipif
->ipif_ill
;
6615 ipst
= ill
->ill_ipst
;
6617 rw_enter(&ipst
->ips_ill_g_lock
, RW_READER
);
6618 ipsq
= ipsq_try_enter_internal(ill
, q
, mp
, func
, type
, reentry_ok
);
6619 rw_exit(&ipst
->ips_ill_g_lock
);
6625 * Try to enter the IPSQ corresponding to `ill' as writer. The caller ensures
6626 * ill is valid by refholding it if necessary; we will refrele. If the IPSQ
6627 * cannot be entered, the mp is queued for completion.
6630 qwriter_ip(ill_t
*ill
, queue_t
*q
, mblk_t
*mp
, ipsq_func_t func
, int type
,
6631 boolean_t reentry_ok
)
6635 ipsq
= ipsq_try_enter(NULL
, ill
, q
, mp
, func
, type
, reentry_ok
);
6638 * Drop the caller's refhold on the ill. This is safe since we either
6639 * entered the IPSQ (and thus are exclusive), or failed to enter the
6640 * IPSQ, in which case we return without accessing ill anymore. This
6641 * is needed because func needs to see the correct refcount.
6642 * e.g. removeif can work only then.
6646 (*func
)(ipsq
, q
, mp
, NULL
);
6652 * Exit the specified IPSQ. If this is the final exit on it then drain it
6653 * prior to exiting. Caller must be writer on the specified IPSQ.
6656 ipsq_exit(ipsq_t
*ipsq
)
6664 ASSERT(IAM_WRITER_IPSQ(ipsq
));
6666 ASSERT(ipsq
->ipsq_xop
->ipx_reentry_cnt
>= 1);
6667 if (ipsq
->ipsq_xop
->ipx_reentry_cnt
!= 1) {
6668 ipsq
->ipsq_xop
->ipx_reentry_cnt
--;
6673 phyi
= ipsq
->ipsq_phyint
;
6675 mp_ipsq
= (mp
== NULL
) ? NULL
: (ipsq_t
*)mp
->b_next
;
6678 * If we've changed to a new IPSQ, and the phyint associated
6679 * with the old one has gone away, free the old IPSQ. Note
6680 * that this cannot happen while the IPSQ is in a group.
6682 if (mp_ipsq
!= ipsq
&& phyi
== NULL
) {
6683 ASSERT(ipsq
->ipsq_next
== ipsq
);
6684 ASSERT(ipsq
->ipsq_xop
== &ipsq
->ipsq_ownxop
);
6692 func
= (ipsq_func_t
)mp
->b_prev
;
6694 mp
->b_next
= mp
->b_prev
= NULL
;
6698 * If 'q' is an conn queue, it is valid, since we did a
6699 * a refhold on the conn at the start of the ioctl.
6700 * If 'q' is an ill queue, it is valid, since close of an
6701 * ill will clean up its IPSQ.
6703 (*func
)(ipsq
, q
, mp
, NULL
);
6708 * Used to start any igmp or mld timers that could not be started
6709 * while holding ill_mcast_lock. The timers can't be started while holding
6710 * the lock, since mld/igmp_start_timers may need to call untimeout()
6711 * which can't be done while holding the lock which the timeout handler
6712 * acquires. Otherwise
6713 * there could be a deadlock since the timeout handlers
6714 * mld_timeout_handler_per_ill/igmp_timeout_handler_per_ill also acquire
6718 ill_mcast_timer_start(ip_stack_t
*ipst
)
6722 mutex_enter(&ipst
->ips_igmp_timer_lock
);
6723 next
= ipst
->ips_igmp_deferred_next
;
6724 ipst
->ips_igmp_deferred_next
= INFINITY
;
6725 mutex_exit(&ipst
->ips_igmp_timer_lock
);
6727 if (next
!= INFINITY
)
6728 igmp_start_timers(next
, ipst
);
6730 mutex_enter(&ipst
->ips_mld_timer_lock
);
6731 next
= ipst
->ips_mld_deferred_next
;
6732 ipst
->ips_mld_deferred_next
= INFINITY
;
6733 mutex_exit(&ipst
->ips_mld_timer_lock
);
6735 if (next
!= INFINITY
)
6736 mld_start_timers(next
, ipst
);
6740 * Start the current exclusive operation on `ipsq'; associate it with `ipif'
6744 ipsq_current_start(ipsq_t
*ipsq
, ipif_t
*ipif
, int ioccmd
)
6746 ill_t
*ill
= ipif
->ipif_ill
;
6747 ipxop_t
*ipx
= ipsq
->ipsq_xop
;
6749 ASSERT(IAM_WRITER_IPSQ(ipsq
));
6750 ASSERT(ipx
->ipx_current_ipif
== NULL
);
6751 ASSERT(ipx
->ipx_current_ioctl
== 0);
6753 ipx
->ipx_current_done
= B_FALSE
;
6754 ipx
->ipx_current_ioctl
= ioccmd
;
6755 mutex_enter(&ipx
->ipx_lock
);
6756 ipx
->ipx_current_ipif
= ipif
;
6757 mutex_exit(&ipx
->ipx_lock
);
6760 * Set IPIF_CHANGING on one or more ipifs associated with the
6761 * current exclusive operation. IPIF_CHANGING prevents any new
6762 * references to the ipif (so that the references will eventually
6763 * drop to zero) and also prevents any "get" operations (e.g.,
6764 * SIOCGLIFFLAGS) from being able to access the ipif until the
6765 * operation has completed and the ipif is again in a stable state.
6767 * For ioctls, IPIF_CHANGING is set on the ipif associated with the
6768 * ioctl. For internal operations (where ioccmd is zero), all ipifs
6769 * on the ill are marked with IPIF_CHANGING since it's unclear which
6770 * ipifs will be affected.
6772 * Note that SIOCLIFREMOVEIF is a special case as it sets
6773 * IPIF_CONDEMNED internally after identifying the right ipif to
6777 case SIOCLIFREMOVEIF
:
6780 mutex_enter(&ill
->ill_lock
);
6781 ipif
= ipif
->ipif_ill
->ill_ipif
;
6782 for (; ipif
!= NULL
; ipif
= ipif
->ipif_next
)
6783 ipif
->ipif_state_flags
|= IPIF_CHANGING
;
6784 mutex_exit(&ill
->ill_lock
);
6787 mutex_enter(&ill
->ill_lock
);
6788 ipif
->ipif_state_flags
|= IPIF_CHANGING
;
6789 mutex_exit(&ill
->ill_lock
);
6794 * Finish the current exclusive operation on `ipsq'. Usually, this will allow
6795 * the next exclusive operation to begin once we ipsq_exit(). However, if
6796 * pending DLPI operations remain, then we will wait for the queue to drain
6797 * before allowing the next exclusive operation to begin. This ensures that
6798 * DLPI operations from one exclusive operation are never improperly processed
6799 * as part of a subsequent exclusive operation.
6802 ipsq_current_finish(ipsq_t
*ipsq
)
6804 ipxop_t
*ipx
= ipsq
->ipsq_xop
;
6805 t_uscalar_t dlpi_pending
= DL_PRIM_INVAL
;
6806 ipif_t
*ipif
= ipx
->ipx_current_ipif
;
6808 ASSERT(IAM_WRITER_IPSQ(ipsq
));
6811 * For SIOCLIFREMOVEIF, the ipif has been already been blown away
6812 * (but in that case, IPIF_CHANGING will already be clear and no
6813 * pending DLPI messages can remain).
6815 if (ipx
->ipx_current_ioctl
!= SIOCLIFREMOVEIF
) {
6816 ill_t
*ill
= ipif
->ipif_ill
;
6818 mutex_enter(&ill
->ill_lock
);
6819 dlpi_pending
= ill
->ill_dlpi_pending
;
6820 if (ipx
->ipx_current_ioctl
== 0) {
6821 ipif
= ill
->ill_ipif
;
6822 for (; ipif
!= NULL
; ipif
= ipif
->ipif_next
)
6823 ipif
->ipif_state_flags
&= ~IPIF_CHANGING
;
6825 ipif
->ipif_state_flags
&= ~IPIF_CHANGING
;
6827 mutex_exit(&ill
->ill_lock
);
6830 ASSERT(!ipx
->ipx_current_done
);
6831 ipx
->ipx_current_done
= B_TRUE
;
6832 ipx
->ipx_current_ioctl
= 0;
6833 if (dlpi_pending
== DL_PRIM_INVAL
) {
6834 mutex_enter(&ipx
->ipx_lock
);
6835 ipx
->ipx_current_ipif
= NULL
;
6836 mutex_exit(&ipx
->ipx_lock
);
6841 * The ill is closing. Flush all messages on the ipsq that originated
6842 * from this ill. Usually there wont' be any messages on the ipsq_xopq_mphead
6843 * for this ill since ipsq_enter could not have entered until then.
6844 * New messages can't be queued since the CONDEMNED flag is set.
6847 ipsq_flush(ill_t
*ill
)
6853 ipxop_t
*ipx
= ill
->ill_phyint
->phyint_ipsq
->ipsq_xop
;
6855 ASSERT(IAM_WRITER_ILL(ill
));
6858 * Flush any messages sent up by the driver.
6860 mutex_enter(&ipx
->ipx_lock
);
6861 for (prev
= NULL
, mp
= ipx
->ipx_mphead
; mp
!= NULL
; mp
= mp_next
) {
6862 mp_next
= mp
->b_next
;
6864 if (q
== ill
->ill_rq
|| q
== ill
->ill_wq
) {
6867 ipx
->ipx_mphead
= mp
->b_next
;
6869 prev
->b_next
= mp
->b_next
;
6870 if (ipx
->ipx_mptail
== mp
) {
6871 ASSERT(mp_next
== NULL
);
6872 ipx
->ipx_mptail
= prev
;
6879 mutex_exit(&ipx
->ipx_lock
);
6880 (void) ipsq_pending_mp_cleanup(ill
, NULL
);
6881 ipsq_xopq_mp_cleanup(ill
, NULL
);
6885 * Parse an ifreq or lifreq struct coming down ioctls and refhold
6886 * and return the associated ipif.
6888 * Non zero: An error has occurred. ci may not be filled out.
6889 * zero : ci is filled out with the ioctl cmd in ci.ci_name, and
6890 * a held ipif in ci.ci_ipif.
6893 ip_extract_lifreq(queue_t
*q
, mblk_t
*mp
, const ip_ioctl_cmd_t
*ipip
,
6898 struct lifreq
*lifr
;
6899 ipif_t
*ipif
= NULL
;
6908 if (q
->q_next
!= NULL
) {
6909 ill
= (ill_t
*)q
->q_ptr
;
6910 isv6
= ill
->ill_isv6
;
6913 ipst
= ill
->ill_ipst
;
6916 connp
= Q_TO_CONN(q
);
6917 isv6
= (connp
->conn_family
== AF_INET6
);
6918 zoneid
= connp
->conn_zoneid
;
6919 if (zoneid
== GLOBAL_ZONEID
) {
6920 /* global zone can access ipifs in all zones */
6923 ipst
= connp
->conn_netstack
->netstack_ip
;
6926 /* Has been checked in ip_wput_nondata */
6927 mp1
= mp
->b_cont
->b_cont
;
6929 if (ipip
->ipi_cmd_type
== IF_CMD
) {
6930 /* This a old style SIOC[GS]IF* command */
6931 ifr
= (struct ifreq
*)mp1
->b_rptr
;
6933 * Null terminate the string to protect against buffer
6934 * overrun. String was generated by user code and may not
6937 ifr
->ifr_name
[IFNAMSIZ
- 1] = '\0';
6938 name
= ifr
->ifr_name
;
6939 ci
->ci_sin
= (sin_t
*)&ifr
->ifr_addr
;
6941 ci
->ci_lifr
= (struct lifreq
*)ifr
;
6943 /* This a new style SIOC[GS]LIF* command */
6944 ASSERT(ipip
->ipi_cmd_type
== LIF_CMD
);
6945 lifr
= (struct lifreq
*)mp1
->b_rptr
;
6947 * Null terminate the string to protect against buffer
6948 * overrun. String was generated by user code and may not
6951 lifr
->lifr_name
[LIFNAMSIZ
- 1] = '\0';
6952 name
= lifr
->lifr_name
;
6953 ci
->ci_sin
= (sin_t
*)&lifr
->lifr_addr
;
6954 ci
->ci_sin6
= (sin6_t
*)&lifr
->lifr_addr
;
6958 if (ipip
->ipi_cmd
== SIOCSLIFNAME
) {
6960 * The ioctl will be failed if the ioctl comes down
6965 * Not an ill queue, return EINVAL same as the
6970 ipif
= ill
->ill_ipif
;
6974 * Ensure that ioctls don't see any internal state changes
6975 * caused by set ioctls by deferring them if IPIF_CHANGING is
6978 ipif
= ipif_lookup_on_name_async(name
, mi_strlen(name
),
6979 isv6
, zoneid
, q
, mp
, ip_process_ioctl
, &err
, ipst
);
6981 if (err
== EINPROGRESS
)
6983 err
= 0; /* Ensure we don't use it below */
6988 * Old style [GS]IFCMD does not admit IPv6 ipif
6990 if (ipif
!= NULL
&& ipif
->ipif_isv6
&& ipip
->ipi_cmd_type
== IF_CMD
) {
6995 if (ipif
== NULL
&& ill
!= NULL
&& ill
->ill_ipif
!= NULL
&&
6998 * Handle a or a SIOC?IF* with a null name
6999 * during plumb (on the ill queue before the I_PLINK).
7001 ipif
= ill
->ill_ipif
;
7008 DTRACE_PROBE4(ipif__ioctl
, char *, "ip_extract_lifreq",
7009 int, ipip
->ipi_cmd
, ill_t
*, ipif
->ipif_ill
, ipif_t
*, ipif
);
7016 * Return the total number of ipifs.
7019 ip_get_numifs(zoneid_t zoneid
, ip_stack_t
*ipst
)
7023 ill_walk_context_t ctx
;
7026 rw_enter(&ipst
->ips_ill_g_lock
, RW_READER
);
7027 ill
= ILL_START_WALK_V4(&ctx
, ipst
);
7028 for (; ill
!= NULL
; ill
= ill_next(&ctx
, ill
)) {
7029 if (IS_UNDER_IPMP(ill
))
7031 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
;
7032 ipif
= ipif
->ipif_next
) {
7033 if (ipif
->ipif_zoneid
== zoneid
||
7034 ipif
->ipif_zoneid
== ALL_ZONES
)
7038 rw_exit(&ipst
->ips_ill_g_lock
);
7043 * Return the total number of ipifs.
7046 ip_get_numlifs(int family
, int lifn_flags
, zoneid_t zoneid
, ip_stack_t
*ipst
)
7051 ill_walk_context_t ctx
;
7053 ip1dbg(("ip_get_numlifs(%d %u %d)\n", family
, lifn_flags
, (int)zoneid
));
7055 rw_enter(&ipst
->ips_ill_g_lock
, RW_READER
);
7056 if (family
== AF_INET
)
7057 ill
= ILL_START_WALK_V4(&ctx
, ipst
);
7058 else if (family
== AF_INET6
)
7059 ill
= ILL_START_WALK_V6(&ctx
, ipst
);
7061 ill
= ILL_START_WALK_ALL(&ctx
, ipst
);
7063 for (; ill
!= NULL
; ill
= ill_next(&ctx
, ill
)) {
7064 if (IS_UNDER_IPMP(ill
) && !(lifn_flags
& LIFC_UNDER_IPMP
))
7067 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
;
7068 ipif
= ipif
->ipif_next
) {
7069 if ((ipif
->ipif_flags
& IPIF_NOXMIT
) &&
7070 !(lifn_flags
& LIFC_NOXMIT
))
7072 if ((ipif
->ipif_flags
& IPIF_TEMPORARY
) &&
7073 !(lifn_flags
& LIFC_TEMPORARY
))
7075 if (((ipif
->ipif_flags
&
7076 (IPIF_NOXMIT
|IPIF_NOLOCAL
|
7077 IPIF_DEPRECATED
)) ||
7079 !(ipif
->ipif_flags
& IPIF_UP
)) &&
7080 (lifn_flags
& LIFC_EXTERNAL_SOURCE
))
7083 if (zoneid
!= ipif
->ipif_zoneid
&&
7084 ipif
->ipif_zoneid
!= ALL_ZONES
&&
7085 (zoneid
!= GLOBAL_ZONEID
||
7086 !(lifn_flags
& LIFC_ALLZONES
)))
7092 rw_exit(&ipst
->ips_ill_g_lock
);
7097 ip_get_lifsrcofnum(ill_t
*ill
)
7100 ill_t
*ill_head
= ill
;
7101 ip_stack_t
*ipst
= ill
->ill_ipst
;
7104 * ill_g_usesrc_lock protects ill_usesrc_grp_next, for example, some
7105 * other thread may be trying to relink the ILLs in this usesrc group
7106 * and adjusting the ill_usesrc_grp_next pointers
7108 rw_enter(&ipst
->ips_ill_g_usesrc_lock
, RW_READER
);
7109 if ((ill
->ill_usesrc_ifindex
== 0) &&
7110 (ill
->ill_usesrc_grp_next
!= NULL
)) {
7111 for (; (ill
!= NULL
) && (ill
->ill_usesrc_grp_next
!= ill_head
);
7112 ill
= ill
->ill_usesrc_grp_next
)
7115 rw_exit(&ipst
->ips_ill_g_usesrc_lock
);
7120 /* Null values are passed in for ipif, sin, and ifreq */
7123 ip_sioctl_get_ifnum(ipif_t
*dummy_ipif
, sin_t
*dummy_sin
, queue_t
*q
,
7124 mblk_t
*mp
, ip_ioctl_cmd_t
*ipip
, void *ifreq
)
7127 conn_t
*connp
= Q_TO_CONN(q
);
7129 ASSERT(q
->q_next
== NULL
); /* not a valid ioctl for ip as a module */
7131 /* Existence of b_cont->b_cont checked in ip_wput_nondata */
7132 nump
= (int *)mp
->b_cont
->b_cont
->b_rptr
;
7134 *nump
= ip_get_numifs(connp
->conn_zoneid
,
7135 connp
->conn_netstack
->netstack_ip
);
7136 ip1dbg(("ip_sioctl_get_ifnum numifs %d", *nump
));
7140 /* Null values are passed in for ipif, sin, and ifreq */
7143 ip_sioctl_get_lifnum(ipif_t
*dummy_ipif
, sin_t
*dummy_sin
,
7144 queue_t
*q
, mblk_t
*mp
, ip_ioctl_cmd_t
*ipip
, void *ifreq
)
7146 struct lifnum
*lifn
;
7148 conn_t
*connp
= Q_TO_CONN(q
);
7150 ASSERT(q
->q_next
== NULL
); /* not a valid ioctl for ip as a module */
7152 /* Existence checked in ip_wput_nondata */
7153 mp1
= mp
->b_cont
->b_cont
;
7155 lifn
= (struct lifnum
*)mp1
->b_rptr
;
7156 switch (lifn
->lifn_family
) {
7162 return (EAFNOSUPPORT
);
7165 lifn
->lifn_count
= ip_get_numlifs(lifn
->lifn_family
, lifn
->lifn_flags
,
7166 connp
->conn_zoneid
, connp
->conn_netstack
->netstack_ip
);
7167 ip1dbg(("ip_sioctl_get_lifnum numifs %d", lifn
->lifn_count
));
7173 ip_sioctl_get_ifconf(ipif_t
*dummy_ipif
, sin_t
*dummy_sin
, queue_t
*q
,
7174 mblk_t
*mp
, ip_ioctl_cmd_t
*ipip
, void *ifreq
)
7176 STRUCT_HANDLE(ifconf
, ifc
);
7178 struct iocblk
*iocp
;
7180 ill_walk_context_t ctx
;
7183 struct sockaddr_in
*sin
;
7186 ip_stack_t
*ipst
= CONNQ_TO_IPST(q
);
7188 ASSERT(q
->q_next
== NULL
); /* not valid ioctls for ip as a module */
7190 ip1dbg(("ip_sioctl_get_ifconf"));
7191 /* Existence verified in ip_wput_nondata */
7192 mp1
= mp
->b_cont
->b_cont
;
7193 iocp
= (struct iocblk
*)mp
->b_rptr
;
7194 zoneid
= Q_TO_CONN(q
)->conn_zoneid
;
7197 * The original SIOCGIFCONF passed in a struct ifconf which specified
7198 * the user buffer address and length into which the list of struct
7199 * ifreqs was to be copied. Since AT&T Streams does not seem to
7200 * allow M_COPYOUT to be used in conjunction with I_STR IOCTLS,
7201 * the SIOCGIFCONF operation was redefined to simply provide
7202 * a large output buffer into which we are supposed to jam the ifreq
7203 * array. The same ioctl command code was used, despite the fact that
7204 * both the applications and the kernel code had to change, thus making
7205 * it impossible to support both interfaces.
7207 * For reasons not good enough to try to explain, the following
7208 * algorithm is used for deciding what to do with one of these:
7209 * If the IOCTL comes in as an I_STR, it is assumed to be of the new
7210 * form with the output buffer coming down as the continuation message.
7211 * If it arrives as a TRANSPARENT IOCTL, it is assumed to be old style,
7212 * and we have to copy in the ifconf structure to find out how big the
7213 * output buffer is and where to copy out to. Sure no problem...
7216 STRUCT_SET_HANDLE(ifc
, iocp
->ioc_flag
, NULL
);
7217 if ((mp1
->b_wptr
- mp1
->b_rptr
) == STRUCT_SIZE(ifc
)) {
7222 * Must be (better be!) continuation of a TRANSPARENT
7223 * IOCTL. We just copied in the ifconf structure.
7225 STRUCT_SET_HANDLE(ifc
, iocp
->ioc_flag
,
7226 (struct ifconf
*)mp1
->b_rptr
);
7229 * Allocate a buffer to hold requested information.
7231 * If ifc_len is larger than what is needed, we only
7232 * allocate what we will use.
7234 * If ifc_len is smaller than what is needed, return
7237 * XXX: the ill_t structure can hava 2 counters, for
7238 * v4 and v6 (not just ill_ipif_up_count) to store the
7239 * number of interfaces for a device, so we don't need
7240 * to count them here...
7242 numifs
= ip_get_numifs(zoneid
, ipst
);
7244 ifclen
= STRUCT_FGET(ifc
, ifc_len
);
7245 ifc_bufsize
= numifs
* sizeof (struct ifreq
);
7246 if (ifc_bufsize
> ifclen
) {
7247 if (iocp
->ioc_cmd
== O_SIOCGIFCONF
) {
7251 ifc_bufsize
= ifclen
;
7255 mp1
= mi_copyout_alloc(q
, mp
,
7256 STRUCT_FGETP(ifc
, ifc_buf
), ifc_bufsize
, B_FALSE
);
7260 mp1
->b_wptr
= mp1
->b_rptr
+ ifc_bufsize
;
7262 bzero(mp1
->b_rptr
, mp1
->b_wptr
- mp1
->b_rptr
);
7264 * the SIOCGIFCONF ioctl only knows about
7265 * IPv4 addresses, so don't try to tell
7266 * it about interfaces with IPv6-only
7267 * addresses. (Last parm 'isv6' is B_FALSE)
7270 ifr
= (struct ifreq
*)mp1
->b_rptr
;
7272 rw_enter(&ipst
->ips_ill_g_lock
, RW_READER
);
7273 ill
= ILL_START_WALK_V4(&ctx
, ipst
);
7274 for (; ill
!= NULL
; ill
= ill_next(&ctx
, ill
)) {
7275 if (IS_UNDER_IPMP(ill
))
7277 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
;
7278 ipif
= ipif
->ipif_next
) {
7279 if (zoneid
!= ipif
->ipif_zoneid
&&
7280 ipif
->ipif_zoneid
!= ALL_ZONES
)
7282 if ((uchar_t
*)&ifr
[1] > mp1
->b_wptr
) {
7283 if (iocp
->ioc_cmd
== O_SIOCGIFCONF
) {
7285 rw_exit(&ipst
->ips_ill_g_lock
);
7291 ipif_get_name(ipif
, ifr
->ifr_name
,
7292 sizeof (ifr
->ifr_name
));
7293 sin
= (sin_t
*)&ifr
->ifr_addr
;
7295 sin
->sin_family
= AF_INET
;
7296 sin
->sin_addr
.s_addr
= ipif
->ipif_lcl_addr
;
7301 rw_exit(&ipst
->ips_ill_g_lock
);
7302 mp1
->b_wptr
= (uchar_t
*)ifr
;
7304 if (STRUCT_BUF(ifc
) != NULL
) {
7305 STRUCT_FSET(ifc
, ifc_len
,
7306 (int)((uchar_t
*)ifr
- mp1
->b_rptr
));
7312 * Get the interfaces using the address hosted on the interface passed in,
7313 * as a source adddress
7317 ip_sioctl_get_lifsrcof(ipif_t
*dummy_ipif
, sin_t
*dummy_sin
, queue_t
*q
,
7318 mblk_t
*mp
, ip_ioctl_cmd_t
*ipip
, void *ifreq
)
7321 ill_t
*ill
, *ill_head
;
7322 ipif_t
*ipif
, *orig_ipif
;
7324 size_t lifs_bufsize
, lifsmaxlen
;
7325 struct lifreq
*lifr
;
7326 struct iocblk
*iocp
= (struct iocblk
*)mp
->b_rptr
;
7329 boolean_t isv6
= B_FALSE
;
7330 struct sockaddr_in
*sin
;
7331 struct sockaddr_in6
*sin6
;
7332 STRUCT_HANDLE(lifsrcof
, lifs
);
7335 ipst
= CONNQ_TO_IPST(q
);
7337 ASSERT(q
->q_next
== NULL
);
7339 zoneid
= Q_TO_CONN(q
)->conn_zoneid
;
7341 /* Existence verified in ip_wput_nondata */
7342 mp1
= mp
->b_cont
->b_cont
;
7345 * Must be (better be!) continuation of a TRANSPARENT
7346 * IOCTL. We just copied in the lifsrcof structure.
7348 STRUCT_SET_HANDLE(lifs
, iocp
->ioc_flag
,
7349 (struct lifsrcof
*)mp1
->b_rptr
);
7351 if (MBLKL(mp1
) != STRUCT_SIZE(lifs
))
7354 ifindex
= STRUCT_FGET(lifs
, lifs_ifindex
);
7355 isv6
= (Q_TO_CONN(q
))->conn_family
== AF_INET6
;
7356 ipif
= ipif_lookup_on_ifindex(ifindex
, isv6
, zoneid
, ipst
);
7358 ip1dbg(("ip_sioctl_get_lifsrcof: no ipif for ifindex %d\n",
7363 /* Allocate a buffer to hold requested information */
7364 numlifs
= ip_get_lifsrcofnum(ipif
->ipif_ill
);
7365 lifs_bufsize
= numlifs
* sizeof (struct lifreq
);
7366 lifsmaxlen
= STRUCT_FGET(lifs
, lifs_maxlen
);
7367 /* The actual size needed is always returned in lifs_len */
7368 STRUCT_FSET(lifs
, lifs_len
, lifs_bufsize
);
7370 /* If the amount we need is more than what is passed in, abort */
7371 if (lifs_bufsize
> lifsmaxlen
|| lifs_bufsize
== 0) {
7376 mp1
= mi_copyout_alloc(q
, mp
,
7377 STRUCT_FGETP(lifs
, lifs_buf
), lifs_bufsize
, B_FALSE
);
7383 mp1
->b_wptr
= mp1
->b_rptr
+ lifs_bufsize
;
7384 bzero(mp1
->b_rptr
, lifs_bufsize
);
7386 lifr
= (struct lifreq
*)mp1
->b_rptr
;
7388 ill
= ill_head
= ipif
->ipif_ill
;
7391 /* ill_g_usesrc_lock protects ill_usesrc_grp_next */
7392 rw_enter(&ipst
->ips_ill_g_usesrc_lock
, RW_READER
);
7393 rw_enter(&ipst
->ips_ill_g_lock
, RW_READER
);
7395 ill
= ill
->ill_usesrc_grp_next
; /* start from next ill */
7396 for (; (ill
!= NULL
) && (ill
!= ill_head
);
7397 ill
= ill
->ill_usesrc_grp_next
) {
7399 if ((uchar_t
*)&lifr
[1] > mp1
->b_wptr
)
7402 ipif
= ill
->ill_ipif
;
7403 ipif_get_name(ipif
, lifr
->lifr_name
, sizeof (lifr
->lifr_name
));
7404 if (ipif
->ipif_isv6
) {
7405 sin6
= (sin6_t
*)&lifr
->lifr_addr
;
7407 sin6
->sin6_family
= AF_INET6
;
7408 sin6
->sin6_addr
= ipif
->ipif_v6lcl_addr
;
7409 lifr
->lifr_addrlen
= ip_mask_to_plen_v6(
7410 &ipif
->ipif_v6net_mask
);
7412 sin
= (sin_t
*)&lifr
->lifr_addr
;
7414 sin
->sin_family
= AF_INET
;
7415 sin
->sin_addr
.s_addr
= ipif
->ipif_lcl_addr
;
7416 lifr
->lifr_addrlen
= ip_mask_to_plen(
7417 ipif
->ipif_net_mask
);
7421 rw_exit(&ipst
->ips_ill_g_lock
);
7422 rw_exit(&ipst
->ips_ill_g_usesrc_lock
);
7423 ipif_refrele(orig_ipif
);
7424 mp1
->b_wptr
= (uchar_t
*)lifr
;
7425 STRUCT_FSET(lifs
, lifs_len
, (int)((uchar_t
*)lifr
- mp1
->b_rptr
));
7432 ip_sioctl_get_lifconf(ipif_t
*dummy_ipif
, sin_t
*dummy_sin
, queue_t
*q
,
7433 mblk_t
*mp
, ip_ioctl_cmd_t
*ipip
, void *ifreq
)
7441 size_t lifc_bufsize
;
7442 struct lifreq
*lifr
;
7444 struct sockaddr_in
*sin
;
7445 struct sockaddr_in6
*sin6
;
7446 ill_walk_context_t ctx
;
7447 struct iocblk
*iocp
= (struct iocblk
*)mp
->b_rptr
;
7450 STRUCT_HANDLE(lifconf
, lifc
);
7451 ip_stack_t
*ipst
= CONNQ_TO_IPST(q
);
7453 ip1dbg(("ip_sioctl_get_lifconf"));
7455 ASSERT(q
->q_next
== NULL
);
7457 zoneid
= Q_TO_CONN(q
)->conn_zoneid
;
7459 /* Existence verified in ip_wput_nondata */
7460 mp1
= mp
->b_cont
->b_cont
;
7463 * An extended version of SIOCGIFCONF that takes an
7464 * additional address family and flags field.
7465 * AF_UNSPEC retrieve both IPv4 and IPv6.
7466 * Unless LIFC_NOXMIT is specified the IPIF_NOXMIT
7467 * interfaces are omitted.
7468 * Similarly, IPIF_TEMPORARY interfaces are omitted
7469 * unless LIFC_TEMPORARY is specified.
7470 * If LIFC_EXTERNAL_SOURCE is specified, IPIF_NOXMIT,
7471 * IPIF_NOLOCAL, PHYI_LOOPBACK, IPIF_DEPRECATED and
7472 * not IPIF_UP interfaces are omitted. LIFC_EXTERNAL_SOURCE
7473 * has priority over LIFC_NOXMIT.
7475 STRUCT_SET_HANDLE(lifc
, iocp
->ioc_flag
, NULL
);
7477 if ((mp1
->b_wptr
- mp1
->b_rptr
) != STRUCT_SIZE(lifc
))
7481 * Must be (better be!) continuation of a TRANSPARENT
7482 * IOCTL. We just copied in the lifconf structure.
7484 STRUCT_SET_HANDLE(lifc
, iocp
->ioc_flag
, (struct lifconf
*)mp1
->b_rptr
);
7486 family
= STRUCT_FGET(lifc
, lifc_family
);
7487 flags
= STRUCT_FGET(lifc
, lifc_flags
);
7498 * walk only IPV4 ILL's.
7500 list
= IP_V4_G_HEAD
;
7504 * walk only IPV6 ILL's.
7506 list
= IP_V6_G_HEAD
;
7509 return (EAFNOSUPPORT
);
7513 * Allocate a buffer to hold requested information.
7515 * If lifc_len is larger than what is needed, we only
7516 * allocate what we will use.
7518 * If lifc_len is smaller than what is needed, return
7521 numlifs
= ip_get_numlifs(family
, flags
, zoneid
, ipst
);
7522 lifc_bufsize
= numlifs
* sizeof (struct lifreq
);
7523 lifclen
= STRUCT_FGET(lifc
, lifc_len
);
7524 if (lifc_bufsize
> lifclen
) {
7525 if (iocp
->ioc_cmd
== O_SIOCGLIFCONF
)
7528 lifc_bufsize
= lifclen
;
7531 mp1
= mi_copyout_alloc(q
, mp
,
7532 STRUCT_FGETP(lifc
, lifc_buf
), lifc_bufsize
, B_FALSE
);
7536 mp1
->b_wptr
= mp1
->b_rptr
+ lifc_bufsize
;
7537 bzero(mp1
->b_rptr
, mp1
->b_wptr
- mp1
->b_rptr
);
7539 lifr
= (struct lifreq
*)mp1
->b_rptr
;
7541 rw_enter(&ipst
->ips_ill_g_lock
, RW_READER
);
7542 ill
= ill_first(list
, list
, &ctx
, ipst
);
7543 for (; ill
!= NULL
; ill
= ill_next(&ctx
, ill
)) {
7544 if (IS_UNDER_IPMP(ill
) && !(flags
& LIFC_UNDER_IPMP
))
7547 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
;
7548 ipif
= ipif
->ipif_next
) {
7549 if ((ipif
->ipif_flags
& IPIF_NOXMIT
) &&
7550 !(flags
& LIFC_NOXMIT
))
7553 if ((ipif
->ipif_flags
& IPIF_TEMPORARY
) &&
7554 !(flags
& LIFC_TEMPORARY
))
7557 if (((ipif
->ipif_flags
&
7558 (IPIF_NOXMIT
|IPIF_NOLOCAL
|
7559 IPIF_DEPRECATED
)) ||
7561 !(ipif
->ipif_flags
& IPIF_UP
)) &&
7562 (flags
& LIFC_EXTERNAL_SOURCE
))
7565 if (zoneid
!= ipif
->ipif_zoneid
&&
7566 ipif
->ipif_zoneid
!= ALL_ZONES
&&
7567 (zoneid
!= GLOBAL_ZONEID
||
7568 !(flags
& LIFC_ALLZONES
)))
7571 if ((uchar_t
*)&lifr
[1] > mp1
->b_wptr
) {
7572 if (iocp
->ioc_cmd
== O_SIOCGLIFCONF
) {
7573 rw_exit(&ipst
->ips_ill_g_lock
);
7580 ipif_get_name(ipif
, lifr
->lifr_name
,
7581 sizeof (lifr
->lifr_name
));
7582 lifr
->lifr_type
= ill
->ill_type
;
7583 if (ipif
->ipif_isv6
) {
7584 sin6
= (sin6_t
*)&lifr
->lifr_addr
;
7586 sin6
->sin6_family
= AF_INET6
;
7588 ipif
->ipif_v6lcl_addr
;
7589 lifr
->lifr_addrlen
=
7591 &ipif
->ipif_v6net_mask
);
7593 sin
= (sin_t
*)&lifr
->lifr_addr
;
7595 sin
->sin_family
= AF_INET
;
7596 sin
->sin_addr
.s_addr
=
7597 ipif
->ipif_lcl_addr
;
7598 lifr
->lifr_addrlen
=
7600 ipif
->ipif_net_mask
);
7606 rw_exit(&ipst
->ips_ill_g_lock
);
7608 mp1
->b_wptr
= (uchar_t
*)lifr
;
7609 if (STRUCT_BUF(lifc
) != NULL
) {
7610 STRUCT_FSET(lifc
, lifc_len
,
7611 (int)((uchar_t
*)lifr
- mp1
->b_rptr
));
7617 ip_sioctl_ip6addrpolicy(queue_t
*q
, mblk_t
*mp
)
7622 struct iocblk
*iocp
= (struct iocblk
*)mp
->b_rptr
;
7625 if (q
->q_next
== NULL
)
7626 ipst
= CONNQ_TO_IPST(q
);
7628 ipst
= ILLQ_TO_IPST(q
);
7630 /* These two ioctls are I_STR only */
7631 if (iocp
->ioc_count
== TRANSPARENT
) {
7632 miocnak(q
, mp
, 0, EINVAL
);
7636 data_mp
= mp
->b_cont
;
7637 if (data_mp
== NULL
) {
7638 /* The user passed us a NULL argument */
7640 table_size
= iocp
->ioc_count
;
7643 * The user provided a table. The stream head
7644 * may have copied in the user data in chunks,
7645 * so make sure everything is pulled up
7648 if (MBLKL(data_mp
) < iocp
->ioc_count
) {
7649 mblk_t
*new_data_mp
;
7650 if ((new_data_mp
= msgpullup(data_mp
, -1)) ==
7652 miocnak(q
, mp
, 0, ENOMEM
);
7656 data_mp
= new_data_mp
;
7657 mp
->b_cont
= data_mp
;
7659 table
= (ip6_asp_t
*)data_mp
->b_rptr
;
7660 table_size
= iocp
->ioc_count
;
7663 switch (iocp
->ioc_cmd
) {
7664 case SIOCGIP6ADDRPOLICY
:
7665 iocp
->ioc_rval
= ip6_asp_get(table
, table_size
, ipst
);
7666 if (iocp
->ioc_rval
== -1)
7667 iocp
->ioc_error
= EINVAL
;
7668 #if defined(_SYSCALL32_IMPL) && _LONG_LONG_ALIGNMENT_32 == 4
7669 else if (table
!= NULL
&&
7670 (iocp
->ioc_flag
& IOC_MODELS
) == IOC_ILP32
) {
7671 ip6_asp_t
*src
= table
;
7672 ip6_asp32_t
*dst
= (void *)table
;
7673 int count
= table_size
/ sizeof (ip6_asp_t
);
7677 * We need to do an in-place shrink of the array
7678 * to match the alignment attributes of the
7679 * 32-bit ABI looking at it.
7681 /* LINTED: logical expression always true: op "||" */
7682 ASSERT(sizeof (*src
) > sizeof (*dst
));
7683 for (i
= 1; i
< count
; i
++)
7684 bcopy(src
+ i
, dst
+ i
, sizeof (*dst
));
7689 case SIOCSIP6ADDRPOLICY
:
7690 ASSERT(mp
->b_prev
== NULL
);
7691 mp
->b_prev
= (void *)q
;
7692 #if defined(_SYSCALL32_IMPL) && _LONG_LONG_ALIGNMENT_32 == 4
7694 * We pass in the datamodel here so that the ip6_asp_replace()
7695 * routine can handle converting from 32-bit to native formats
7698 * A better way to handle this might be to convert the inbound
7699 * data structure here, and hang it off a new 'mp'; thus the
7700 * ip6_asp_replace() logic would always be dealing with native
7701 * format data structures..
7703 * (An even simpler way to handle these ioctls is to just
7704 * add a 32-bit trailing 'pad' field to the ip6_asp_t structure
7705 * and just recompile everything that depends on it.)
7708 ip6_asp_replace(mp
, table
, table_size
, B_FALSE
, ipst
,
7709 iocp
->ioc_flag
& IOC_MODELS
);
7713 DB_TYPE(mp
) = (iocp
->ioc_error
== 0) ? M_IOCACK
: M_IOCNAK
;
7718 ip_sioctl_dstinfo(queue_t
*q
, mblk_t
*mp
)
7721 struct dstinforeq
*dir
;
7723 in6_addr_t
*daddr
, *saddr
;
7727 in6_addr_t v6setsrc
;
7728 char *slabel
, *dlabel
;
7732 struct iocblk
*iocp
= (struct iocblk
*)mp
->b_rptr
;
7733 conn_t
*connp
= Q_TO_CONN(q
);
7734 zoneid_t zoneid
= IPCL_ZONEID(connp
);
7735 ip_stack_t
*ipst
= connp
->conn_netstack
->netstack_ip
;
7736 uint64_t ipif_flags
;
7738 ASSERT(q
->q_next
== NULL
); /* this ioctl not allowed if ip is module */
7741 * This ioctl is I_STR only, and must have a
7742 * data mblk following the M_IOCTL mblk.
7744 data_mp
= mp
->b_cont
;
7745 if (iocp
->ioc_count
== TRANSPARENT
|| data_mp
== NULL
) {
7746 miocnak(q
, mp
, 0, EINVAL
);
7750 if (MBLKL(data_mp
) < iocp
->ioc_count
) {
7751 mblk_t
*new_data_mp
;
7753 if ((new_data_mp
= msgpullup(data_mp
, -1)) == NULL
) {
7754 miocnak(q
, mp
, 0, ENOMEM
);
7758 data_mp
= new_data_mp
;
7759 mp
->b_cont
= data_mp
;
7761 match_ire
= MATCH_IRE_DSTONLY
;
7763 for (cur
= data_mp
->b_rptr
, end
= data_mp
->b_wptr
;
7764 end
- cur
>= sizeof (struct dstinforeq
);
7765 cur
+= sizeof (struct dstinforeq
)) {
7766 dir
= (struct dstinforeq
*)cur
;
7767 daddr
= &dir
->dir_daddr
;
7768 saddr
= &dir
->dir_saddr
;
7771 * ip_addr_scope_v6() and ip6_asp_lookup() handle
7772 * v4 mapped addresses; ire_ftable_lookup_v6()
7773 * and ip_select_source_v6() do not.
7775 dir
->dir_dscope
= ip_addr_scope_v6(daddr
);
7776 dlabel
= ip6_asp_lookup(daddr
, &dir
->dir_precedence
, ipst
);
7778 isipv4
= IN6_IS_ADDR_V4MAPPED(daddr
);
7780 IN6_V4MAPPED_TO_IPADDR(daddr
, v4daddr
);
7781 v4setsrc
= INADDR_ANY
;
7782 ire
= ire_route_recursive_v4(v4daddr
, 0, NULL
, zoneid
,
7783 match_ire
, IRR_ALLOCATE
, 0, ipst
, &v4setsrc
, NULL
);
7785 v6setsrc
= ipv6_all_zeros
;
7786 ire
= ire_route_recursive_v6(daddr
, 0, NULL
, zoneid
,
7787 match_ire
, IRR_ALLOCATE
, 0, ipst
, &v6setsrc
, NULL
);
7789 ASSERT(ire
!= NULL
);
7790 if (ire
->ire_flags
& (RTF_REJECT
|RTF_BLACKHOLE
)) {
7792 dir
->dir_dreachable
= 0;
7794 /* move on to next dst addr */
7797 dir
->dir_dreachable
= 1;
7799 dst_ill
= ire_nexthop_ill(ire
);
7800 if (dst_ill
== NULL
) {
7805 /* With ipmp we most likely look at the ipmp ill here */
7806 dir
->dir_dmactype
= dst_ill
->ill_mactype
;
7811 if (ip_select_source_v4(dst_ill
, v4setsrc
, v4daddr
,
7812 connp
->conn_ixa
->ixa_multicast_ifaddr
, zoneid
, ipst
,
7813 &v4saddr
, NULL
, &ipif_flags
) != 0) {
7814 v4saddr
= INADDR_ANY
;
7817 IN6_IPADDR_TO_V4MAPPED(v4saddr
, saddr
);
7819 if (ip_select_source_v6(dst_ill
, &v6setsrc
, daddr
,
7820 zoneid
, ipst
, B_FALSE
, IPV6_PREFER_SRC_DEFAULT
,
7821 saddr
, NULL
, &ipif_flags
) != 0) {
7822 *saddr
= ipv6_all_zeros
;
7827 dir
->dir_sscope
= ip_addr_scope_v6(saddr
);
7828 slabel
= ip6_asp_lookup(saddr
, NULL
, ipst
);
7829 dir
->dir_labelmatch
= ip6_asp_labelcmp(dlabel
, slabel
);
7830 dir
->dir_sdeprecated
= (ipif_flags
& IPIF_DEPRECATED
) ? 1 : 0;
7832 ill_refrele(dst_ill
);
7834 miocack(q
, mp
, iocp
->ioc_count
, 0);
7838 * Check if this is an address assigned to this machine.
7839 * Skips interfaces that are down by using ire checks.
7840 * Translates mapped addresses to v4 addresses and then
7841 * treats them as such, returning true if the v4 address
7842 * associated with this mapped address is configured.
7843 * Note: Applications will have to be careful what they do
7844 * with the response; use of mapped addresses limits
7845 * what can be done with the socket, especially with
7846 * respect to socket options and ioctls - neither IPv4
7847 * options nor IPv6 sticky options/ancillary data options
7852 ip_sioctl_tmyaddr(ipif_t
*dummy_ipif
, sin_t
*dummy_sin
, queue_t
*q
, mblk_t
*mp
,
7853 ip_ioctl_cmd_t
*ipip
, void *dummy_ifreq
)
7855 struct sioc_addrreq
*sia
;
7862 ip1dbg(("ip_sioctl_tmyaddr"));
7864 ASSERT(q
->q_next
== NULL
); /* this ioctl not allowed if ip is module */
7865 zoneid
= Q_TO_CONN(q
)->conn_zoneid
;
7866 ipst
= CONNQ_TO_IPST(q
);
7868 /* Existence verified in ip_wput_nondata */
7869 mp1
= mp
->b_cont
->b_cont
;
7870 sia
= (struct sioc_addrreq
*)mp1
->b_rptr
;
7871 sin
= (sin_t
*)&sia
->sa_addr
;
7872 switch (sin
->sin_family
) {
7874 sin6_t
*sin6
= (sin6_t
*)sin
;
7876 if (IN6_IS_ADDR_V4MAPPED(&sin6
->sin6_addr
)) {
7879 IN6_V4MAPPED_TO_IPADDR(&sin6
->sin6_addr
,
7881 ire
= ire_ftable_lookup_v4(v4_addr
, 0, 0,
7882 IRE_LOCAL
|IRE_LOOPBACK
, NULL
, zoneid
,
7883 MATCH_IRE_TYPE
| MATCH_IRE_ZONEONLY
, 0, ipst
, NULL
);
7887 v6addr
= sin6
->sin6_addr
;
7888 ire
= ire_ftable_lookup_v6(&v6addr
, 0, 0,
7889 IRE_LOCAL
|IRE_LOOPBACK
, NULL
, zoneid
,
7890 MATCH_IRE_TYPE
| MATCH_IRE_ZONEONLY
, 0, ipst
, NULL
);
7897 v4addr
= sin
->sin_addr
.s_addr
;
7898 ire
= ire_ftable_lookup_v4(v4addr
, 0, 0,
7899 IRE_LOCAL
|IRE_LOOPBACK
, NULL
, zoneid
,
7900 MATCH_IRE_TYPE
| MATCH_IRE_ZONEONLY
, 0, ipst
, NULL
);
7904 return (EAFNOSUPPORT
);
7916 * Check if this is an address assigned on-link i.e. neighbor,
7917 * and makes sure it's reachable from the current zone.
7918 * Returns true for my addresses as well.
7919 * Translates mapped addresses to v4 addresses and then
7920 * treats them as such, returning true if the v4 address
7921 * associated with this mapped address is configured.
7922 * Note: Applications will have to be careful what they do
7923 * with the response; use of mapped addresses limits
7924 * what can be done with the socket, especially with
7925 * respect to socket options and ioctls - neither IPv4
7926 * options nor IPv6 sticky options/ancillary data options
7931 ip_sioctl_tonlink(ipif_t
*dummy_ipif
, sin_t
*dummy_sin
, queue_t
*q
, mblk_t
*mp
,
7932 ip_ioctl_cmd_t
*ipip
, void *duymmy_ifreq
)
7934 struct sioc_addrreq
*sia
;
7941 ip1dbg(("ip_sioctl_tonlink"));
7943 ASSERT(q
->q_next
== NULL
); /* this ioctl not allowed if ip is module */
7944 zoneid
= Q_TO_CONN(q
)->conn_zoneid
;
7945 ipst
= CONNQ_TO_IPST(q
);
7947 /* Existence verified in ip_wput_nondata */
7948 mp1
= mp
->b_cont
->b_cont
;
7949 sia
= (struct sioc_addrreq
*)mp1
->b_rptr
;
7950 sin
= (sin_t
*)&sia
->sa_addr
;
7953 * We check for IRE_ONLINK and exclude IRE_BROADCAST|IRE_MULTICAST
7954 * to make sure we only look at on-link unicast address.
7956 switch (sin
->sin_family
) {
7958 sin6_t
*sin6
= (sin6_t
*)sin
;
7960 if (IN6_IS_ADDR_V4MAPPED(&sin6
->sin6_addr
)) {
7963 IN6_V4MAPPED_TO_IPADDR(&sin6
->sin6_addr
,
7965 if (!CLASSD(v4_addr
)) {
7966 ire
= ire_ftable_lookup_v4(v4_addr
, 0, 0, 0,
7967 NULL
, zoneid
, MATCH_IRE_DSTONLY
, 0, ipst
,
7973 v6addr
= sin6
->sin6_addr
;
7974 if (!IN6_IS_ADDR_MULTICAST(&v6addr
)) {
7975 ire
= ire_ftable_lookup_v6(&v6addr
, 0, 0, 0,
7976 NULL
, zoneid
, MATCH_IRE_DSTONLY
, 0, ipst
,
7985 v4addr
= sin
->sin_addr
.s_addr
;
7986 if (!CLASSD(v4addr
)) {
7987 ire
= ire_ftable_lookup_v4(v4addr
, 0, 0, 0, NULL
,
7988 zoneid
, MATCH_IRE_DSTONLY
, 0, ipst
, NULL
);
7993 return (EAFNOSUPPORT
);
7997 ASSERT(!(ire
->ire_type
& IRE_MULTICAST
));
7999 if ((ire
->ire_type
& IRE_ONLINK
) &&
8000 !(ire
->ire_type
& IRE_BROADCAST
))
8008 * TBD: implement when kernel maintaines a list of site prefixes.
8012 ip_sioctl_tmysite(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
8013 ip_ioctl_cmd_t
*ipip
, void *ifreq
)
8021 ip_sioctl_arp(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
8022 ip_ioctl_cmd_t
*ipip
, void *dummy_ifreq
)
8026 struct iocblk
*iocp
;
8029 struct xarpreq
*xar
;
8030 int arp_flags
, flags
, alength
;
8033 ill_t
*ill
= ipif
->ipif_ill
;
8034 ill_t
*proxy_ill
= NULL
;
8035 ipmp_arpent_t
*entp
= NULL
;
8036 boolean_t proxyarp
= B_FALSE
;
8037 boolean_t if_arp_ioctl
= B_FALSE
;
8038 ncec_t
*ncec
= NULL
;
8041 ASSERT(!(q
->q_flag
& QREADR
) && q
->q_next
== NULL
);
8042 connp
= Q_TO_CONN(q
);
8043 ipst
= connp
->conn_netstack
->netstack_ip
;
8044 iocp
= (struct iocblk
*)mp
->b_rptr
;
8046 if (ipip
->ipi_cmd_type
== XARP_CMD
) {
8047 /* We have a chain - M_IOCTL-->MI_COPY_MBLK-->XARPREQ_MBLK */
8048 xar
= (struct xarpreq
*)mp
->b_cont
->b_cont
->b_rptr
;
8051 arp_flags
= xar
->xarp_flags
;
8052 lladdr
= (uchar_t
*)LLADDR(&xar
->xarp_ha
);
8053 if_arp_ioctl
= (xar
->xarp_ha
.sdl_nlen
!= 0);
8055 * Validate against user's link layer address length
8056 * input and name and addr length limits.
8058 alength
= ill
->ill_phys_addr_length
;
8059 if (ipip
->ipi_cmd
== SIOCSXARP
) {
8060 if (alength
!= xar
->xarp_ha
.sdl_alen
||
8061 (alength
+ xar
->xarp_ha
.sdl_nlen
>
8062 sizeof (xar
->xarp_ha
.sdl_data
)))
8066 /* We have a chain - M_IOCTL-->MI_COPY_MBLK-->ARPREQ_MBLK */
8067 ar
= (struct arpreq
*)mp
->b_cont
->b_cont
->b_rptr
;
8070 arp_flags
= ar
->arp_flags
;
8071 lladdr
= (uchar_t
*)ar
->arp_ha
.sa_data
;
8073 * Theoretically, the sa_family could tell us what link
8074 * layer type this operation is trying to deal with. By
8075 * common usage AF_UNSPEC means ethernet. We'll assume
8076 * any attempt to use the SIOC?ARP ioctls is for ethernet,
8077 * for now. Our new SIOC*XARP ioctls can be used more
8080 * If the underlying media happens to have a non 6 byte
8081 * address, arp module will fail set/get, but the del
8082 * operation will succeed.
8085 if ((ipip
->ipi_cmd
!= SIOCDARP
) &&
8086 (alength
!= ill
->ill_phys_addr_length
)) {
8091 /* Translate ATF* flags to NCE* flags */
8093 if (arp_flags
& ATF_AUTHORITY
)
8094 flags
|= NCE_F_AUTHORITY
;
8095 if (arp_flags
& ATF_PERM
)
8096 flags
|= NCE_F_NONUD
; /* not subject to aging */
8097 if (arp_flags
& ATF_PUBL
)
8098 flags
|= NCE_F_PUBLISH
;
8101 * IPMP ARP special handling:
8103 * 1. Since ARP mappings must appear consistent across the group,
8104 * prohibit changing ARP mappings on the underlying interfaces.
8106 * 2. Since ARP mappings for IPMP data addresses are maintained by
8107 * IP itself, prohibit changing them.
8109 * 3. For proxy ARP, use a functioning hardware address in the group,
8110 * provided one exists. If one doesn't, just add the entry as-is;
8111 * ipmp_illgrp_refresh_arpent() will refresh it if things change.
8113 if (IS_UNDER_IPMP(ill
)) {
8114 if (ipip
->ipi_cmd
!= SIOCGARP
&& ipip
->ipi_cmd
!= SIOCGXARP
)
8118 ipmp_illgrp_t
*illg
= ill
->ill_grp
;
8120 switch (ipip
->ipi_cmd
) {
8123 proxy_ill
= ipmp_illgrp_find_ill(illg
, lladdr
, alength
);
8124 if (proxy_ill
!= NULL
) {
8126 if (!ipmp_ill_is_active(proxy_ill
))
8127 proxy_ill
= ipmp_illgrp_next_ill(illg
);
8128 if (proxy_ill
!= NULL
)
8129 lladdr
= proxy_ill
->ill_phys_addr
;
8135 ipaddr
= sin
->sin_addr
.s_addr
;
8137 * don't match across illgrp per case (1) and (2).
8138 * XXX use IS_IPMP(ill) like ndp_sioc_update?
8140 nce
= nce_lookup_v4(ill
, &ipaddr
);
8142 ncec
= nce
->nce_common
;
8144 switch (iocp
->ioc_cmd
) {
8148 * Delete the NCE if any.
8151 iocp
->ioc_error
= ENXIO
;
8154 /* Don't allow changes to arp mappings of local addresses. */
8155 if (NCE_MYADDR(ncec
)) {
8159 iocp
->ioc_error
= 0;
8162 * Delete the nce_common which has ncec_ill set to ipmp_ill.
8163 * This will delete all the nce entries on the under_ills.
8167 * Once the NCE has been deleted, then the ire_dep* consistency
8168 * mechanism will find any IRE which depended on the now
8169 * condemned NCE (as part of sending packets).
8170 * That mechanism handles redirects by deleting redirects
8171 * that refer to UNREACHABLE nces.
8178 lladdr
= ncec
->ncec_lladdr
;
8179 flags
= ncec
->ncec_flags
;
8180 iocp
->ioc_error
= 0;
8181 ip_sioctl_garp_reply(mp
, ncec
->ncec_ill
, lladdr
, flags
);
8183 iocp
->ioc_error
= ENXIO
;
8188 /* Don't allow changes to arp mappings of local addresses. */
8189 if (ncec
!= NULL
&& NCE_MYADDR(ncec
)) {
8194 /* static arp entries will undergo NUD if ATF_PERM is not set */
8195 flags
|= NCE_F_STATIC
;
8196 if (!if_arp_ioctl
) {
8197 ip_nce_lookup_and_update(&ipaddr
, NULL
, ipst
,
8198 lladdr
, alength
, flags
);
8200 ipif_t
*ipif
= ipif_get_next_ipif(NULL
, ill
);
8202 ip_nce_lookup_and_update(&ipaddr
, ipif
, ipst
,
8203 lladdr
, alength
, flags
);
8212 * NCE_F_STATIC entries will be added in state ND_REACHABLE
8213 * by nce_add_common()
8215 err
= nce_lookup_then_add_v4(ill
, lladdr
,
8216 ill
->ill_phys_addr_length
, &ipaddr
, flags
, ND_UNCHANGED
,
8218 if (err
== EEXIST
) {
8219 ncec
= nce
->nce_common
;
8220 mutex_enter(&ncec
->ncec_lock
);
8221 ncec
->ncec_state
= ND_REACHABLE
;
8222 ncec
->ncec_flags
= flags
;
8223 nce_update(ncec
, ND_UNCHANGED
, lladdr
);
8224 mutex_exit(&ncec
->ncec_lock
);
8231 if (IS_IPMP(ill
) && err
== 0) {
8232 entp
= ipmp_illgrp_create_arpent(ill
->ill_grp
,
8233 proxyarp
, ipaddr
, lladdr
, ill
->ill_phys_addr_length
,
8235 if (entp
== NULL
|| (proxyarp
&& proxy_ill
== NULL
)) {
8236 iocp
->ioc_error
= (entp
== NULL
? ENOMEM
: 0);
8240 iocp
->ioc_error
= err
;
8248 * If we created an IPMP ARP entry, mark that we've notified ARP.
8251 ipmp_illgrp_mark_arpent(ill
->ill_grp
, entp
);
8253 return (iocp
->ioc_error
);
8257 * Parse an [x]arpreq structure coming down SIOC[GSD][X]ARP ioctls, identify
8258 * the associated sin and refhold and return the associated ipif via `ci'.
8261 ip_extract_arpreq(queue_t
*q
, mblk_t
*mp
, const ip_ioctl_cmd_t
*ipip
,
8273 struct xarpreq
*xar
;
8274 struct sockaddr_dl
*sdl
;
8276 /* ioctl comes down on a conn */
8277 ASSERT(!(q
->q_flag
& QREADR
) && q
->q_next
== NULL
);
8278 connp
= Q_TO_CONN(q
);
8279 if (connp
->conn_family
== AF_INET6
)
8282 ipst
= connp
->conn_netstack
->netstack_ip
;
8284 /* Verified in ip_wput_nondata */
8285 mp1
= mp
->b_cont
->b_cont
;
8287 if (ipip
->ipi_cmd_type
== XARP_CMD
) {
8288 ASSERT(MBLKL(mp1
) >= sizeof (struct xarpreq
));
8289 xar
= (struct xarpreq
*)mp1
->b_rptr
;
8290 sin
= (sin_t
*)&xar
->xarp_pa
;
8291 sdl
= &xar
->xarp_ha
;
8293 if (sdl
->sdl_family
!= AF_LINK
|| sin
->sin_family
!= AF_INET
)
8295 if (sdl
->sdl_nlen
>= LIFNAMSIZ
)
8298 ASSERT(ipip
->ipi_cmd_type
== ARP_CMD
);
8299 ASSERT(MBLKL(mp1
) >= sizeof (struct arpreq
));
8300 ar
= (struct arpreq
*)mp1
->b_rptr
;
8301 sin
= (sin_t
*)&ar
->arp_pa
;
8304 if (ipip
->ipi_cmd_type
== XARP_CMD
&& sdl
->sdl_nlen
!= 0) {
8305 ipif
= ipif_lookup_on_name(sdl
->sdl_data
, sdl
->sdl_nlen
,
8306 B_FALSE
, &exists
, B_FALSE
, ALL_ZONES
, ipst
);
8309 if (ipif
->ipif_id
!= 0) {
8315 * Either an SIOC[DGS]ARP or an SIOC[DGS]XARP with an sdl_nlen
8316 * of 0: use the IP address to find the ipif. If the IP
8317 * address is an IPMP test address, ire_ftable_lookup() will
8318 * find the wrong ill, so we first do an ipif_lookup_addr().
8320 ipif
= ipif_lookup_addr(sin
->sin_addr
.s_addr
, NULL
, ALL_ZONES
,
8323 ire
= ire_ftable_lookup_v4(sin
->sin_addr
.s_addr
,
8324 0, 0, IRE_IF_RESOLVER
, NULL
, ALL_ZONES
,
8325 MATCH_IRE_TYPE
, 0, ipst
, NULL
);
8326 if (ire
== NULL
|| ((ill
= ire
->ire_ill
) == NULL
)) {
8331 ASSERT(ire
!= NULL
&& ill
!= NULL
);
8332 ipif
= ill
->ill_ipif
;
8338 if (ipif
->ipif_ill
->ill_net_type
!= IRE_IF_RESOLVER
) {
8349 * Link or unlink the illgrp on IPMP meta-interface `ill' depending on the
8350 * value of `ioccmd'. While an illgrp is linked to an ipmp_grp_t, it is
8351 * accessible from that ipmp_grp_t, which means SIOCSLIFGROUPNAME can look it
8352 * up and thus an ill can join that illgrp.
8354 * We use I_PLINK/I_PUNLINK to do the link/unlink operations rather than
8355 * open()/close() primarily because close() is not allowed to fail or block
8356 * forever. On the other hand, I_PUNLINK *can* fail, and there's no reason
8357 * why anyone should ever need to I_PUNLINK an in-use IPMP stream. To ensure
8358 * symmetric behavior (e.g., doing an I_PLINK after and I_PUNLINK undoes the
8359 * I_PUNLINK) we defer linking to I_PLINK. Separately, we also fail attempts
8360 * to I_LINK since I_UNLINK is optional and we'd end up in an inconsistent
8361 * state if I_UNLINK didn't occur.
8363 * Note that for each plumb/unplumb operation, we may end up here more than
8364 * once because of the way ifconfig works. However, it's OK to link the same
8365 * illgrp more than once, or unlink an illgrp that's already unlinked.
8368 ip_sioctl_plink_ipmp(ill_t
*ill
, int ioccmd
)
8371 ip_stack_t
*ipst
= ill
->ill_ipst
;
8373 ASSERT(IS_IPMP(ill
));
8374 ASSERT(IAM_WRITER_ILL(ill
));
8381 rw_enter(&ipst
->ips_ipmp_lock
, RW_WRITER
);
8382 ipmp_illgrp_link_grp(ill
->ill_grp
, ill
->ill_phyint
->phyint_grp
);
8383 rw_exit(&ipst
->ips_ipmp_lock
);
8388 * Require all UP ipifs be brought down prior to unlinking the
8389 * illgrp so any associated IREs (and other state) is torched.
8391 if (ill
->ill_ipif_up_count
+ ill
->ill_ipif_dup_count
> 0)
8395 * NOTE: We hold ipmp_lock across the unlink to prevent a race
8396 * with an SIOCSLIFGROUPNAME request from an ill trying to
8397 * join this group. Specifically: ills trying to join grab
8398 * ipmp_lock and bump a "pending join" counter checked by
8399 * ipmp_illgrp_unlink_grp(). During the unlink no new pending
8400 * joins can occur (since we have ipmp_lock). Once we drop
8401 * ipmp_lock, subsequent SIOCSLIFGROUPNAME requests will not
8402 * find the illgrp (since we unlinked it) and will return
8403 * EAFNOSUPPORT. This will then take them back through the
8404 * IPMP meta-interface plumbing logic in ifconfig, and thus
8405 * back through I_PLINK above.
8407 rw_enter(&ipst
->ips_ipmp_lock
, RW_WRITER
);
8408 err
= ipmp_illgrp_unlink_grp(ill
->ill_grp
);
8409 rw_exit(&ipst
->ips_ipmp_lock
);
8418 * Do I_PLINK/I_LINK or I_PUNLINK/I_UNLINK with consistency checks and also
8419 * atomically set/clear the muxids. Also complete the ioctl by acking or
8420 * naking it. Note that the code is structured such that the link type,
8421 * whether it's persistent or not, is treated equally. ifconfig(8) and
8422 * its clones use the persistent link, while pppd(8) and perhaps many
8423 * other daemons may use non-persistent link. When combined with some
8424 * ill_t states, linking and unlinking lower streams may be used as
8425 * indicators of dynamic re-plumbing events [see PSARC/1999/348].
8429 ip_sioctl_plink(ipsq_t
*ipsq
, queue_t
*q
, mblk_t
*mp
, void *dummy_arg
)
8433 int ioccmd
= ((struct iocblk
*)mp
->b_rptr
)->ioc_cmd
;
8436 ASSERT(ioccmd
== I_PLINK
|| ioccmd
== I_PUNLINK
||
8437 ioccmd
== I_LINK
|| ioccmd
== I_UNLINK
);
8439 mp1
= mp
->b_cont
; /* This is the linkblk info */
8440 li
= (struct linkblk
*)mp1
->b_rptr
;
8442 err
= ip_sioctl_plink_ipmod(ipsq
, q
, mp
, ioccmd
, li
);
8443 if (err
== EINPROGRESS
)
8446 miocack(q
, mp
, 0, 0);
8448 miocnak(q
, mp
, 0, err
);
8450 /* Conn was refheld in ip_sioctl_copyin_setup */
8452 CONN_DEC_IOCTLREF(Q_TO_CONN(q
));
8453 CONN_OPER_PENDING_DONE(Q_TO_CONN(q
));
8458 * Process I_{P}LINK and I_{P}UNLINK requests named by `ioccmd' and pointed to
8459 * by `mp' and `li' for the IP module stream (if li->q_bot is in fact an IP
8461 * Returns zero on success, EINPROGRESS if the operation is still pending, or
8462 * an error code on failure.
8465 ip_sioctl_plink_ipmod(ipsq_t
*ipsq
, queue_t
*q
, mblk_t
*mp
, int ioccmd
,
8470 queue_t
*ipwq
, *dwq
;
8472 struct qinit
*qinfo
;
8473 boolean_t islink
= (ioccmd
== I_PLINK
|| ioccmd
== I_LINK
);
8474 boolean_t entered_ipsq
= B_FALSE
;
8475 boolean_t is_ip
= B_FALSE
;
8479 * Walk the lower stream to verify it's the IP module stream.
8480 * The IP module is identified by its name, wput function,
8481 * and non-NULL q_next. STREAMS ensures that the lower stream
8482 * (li->l_qbot) will not vanish until this ioctl completes.
8484 for (ipwq
= li
->l_qbot
; ipwq
!= NULL
; ipwq
= ipwq
->q_next
) {
8485 qinfo
= ipwq
->q_qinfo
;
8486 name
= qinfo
->qi_minfo
->mi_idname
;
8487 if (name
!= NULL
&& strcmp(name
, ip_mod_info
.mi_idname
) == 0 &&
8488 qinfo
->qi_putp
!= ip_lwput
&& ipwq
->q_next
!= NULL
) {
8492 if (name
!= NULL
&& strcmp(name
, arp_mod_info
.mi_idname
) == 0 &&
8493 qinfo
->qi_putp
!= ip_lwput
&& ipwq
->q_next
!= NULL
) {
8499 * If this isn't an IP module stream, bail.
8505 arl
= (arl_t
*)ipwq
->q_ptr
;
8506 ill
= arl_to_ill(arl
);
8512 ASSERT(ill
!= NULL
);
8515 ipsq
= ipsq_try_enter(NULL
, ill
, q
, mp
, ip_sioctl_plink
,
8520 return (EINPROGRESS
);
8522 entered_ipsq
= B_TRUE
;
8524 ASSERT(IAM_WRITER_ILL(ill
));
8525 mutex_enter(&ill
->ill_lock
);
8527 if (islink
&& ill
->ill_muxid
== 0) {
8529 * Plumbing has to be done with IP plumbed first, arp
8530 * second, but here we have arp being plumbed first.
8532 mutex_exit(&ill
->ill_lock
);
8539 mutex_exit(&ill
->ill_lock
);
8541 arl
->arl_muxid
= islink
? li
->l_index
: 0;
8546 if (IS_IPMP(ill
) && (err
= ip_sioctl_plink_ipmp(ill
, ioccmd
)) != 0)
8550 * As part of I_{P}LINKing, stash the number of downstream modules and
8551 * the read queue of the module immediately below IP in the ill.
8552 * These are used during the capability negotiation below.
8554 ill
->ill_lmod_rq
= NULL
;
8555 ill
->ill_lmod_cnt
= 0;
8556 if (islink
&& ((dwq
= ipwq
->q_next
) != NULL
)) {
8557 ill
->ill_lmod_rq
= RD(dwq
);
8558 for (; dwq
!= NULL
; dwq
= dwq
->q_next
)
8559 ill
->ill_lmod_cnt
++;
8562 ill
->ill_muxid
= islink
? li
->l_index
: 0;
8565 * Mark the ipsq busy until the capability operations initiated below
8566 * complete. The PLINK/UNLINK ioctl itself completes when our caller
8567 * returns, but the capability operation may complete asynchronously
8570 ipsq_current_start(ipsq
, ill
->ill_ipif
, ioccmd
);
8572 * If there's at least one up ipif on this ill, then we're bound to
8573 * the underlying driver via DLPI. In that case, renegotiate
8574 * capabilities to account for any possible change in modules
8575 * interposed between IP and the driver.
8577 if (ill
->ill_ipif_up_count
> 0) {
8579 ill_capability_probe(ill
);
8581 ill_capability_reset(ill
, B_FALSE
);
8583 ipsq_current_finish(ipsq
);
8592 * Search the ioctl command in the ioctl tables and return a pointer
8593 * to the ioctl command information. The ioctl command tables are
8594 * static and fully populated at compile time.
8597 ip_sioctl_lookup(int ioc_cmd
)
8600 ip_ioctl_cmd_t
*ipip
;
8601 ip_ioctl_cmd_t
*ipip_end
;
8603 if (ioc_cmd
== IPI_DONTCARE
)
8607 * Do a 2 step search. First search the indexed table
8608 * based on the least significant byte of the ioctl cmd.
8609 * If we don't find a match, then search the misc table
8612 index
= ioc_cmd
& 0xFF;
8613 if (index
< ip_ndx_ioctl_count
) {
8614 ipip
= &ip_ndx_ioctl_table
[index
];
8615 if (ipip
->ipi_cmd
== ioc_cmd
) {
8616 /* Found a match in the ndx table */
8621 /* Search the misc table */
8622 ipip_end
= &ip_misc_ioctl_table
[ip_misc_ioctl_count
];
8623 for (ipip
= ip_misc_ioctl_table
; ipip
< ipip_end
; ipip
++) {
8624 if (ipip
->ipi_cmd
== ioc_cmd
)
8625 /* Found a match in the misc table */
8633 * helper function for ip_sioctl_getsetprop(), which does some sanity checks
8636 getset_ioctl_checks(mod_ioc_prop_t
*pioc
, int ioc_cmd
)
8638 uint_t flags
= pioc
->mpr_flags
;
8639 if (ioc_cmd
== SIOCSETPROP
) {
8641 * One can either reset the value to it's default value or
8642 * change the current value or append/remove the value from
8643 * a multi-valued properties.
8645 if ((flags
& MOD_PROP_DEFAULT
) != MOD_PROP_DEFAULT
&&
8646 flags
!= MOD_PROP_ACTIVE
&&
8647 flags
!= (MOD_PROP_ACTIVE
|MOD_PROP_APPEND
) &&
8648 flags
!= (MOD_PROP_ACTIVE
|MOD_PROP_REMOVE
))
8651 ASSERT(ioc_cmd
== SIOCGETPROP
);
8654 * One can retrieve only one kind of property information
8657 if ((flags
& MOD_PROP_ACTIVE
) != MOD_PROP_ACTIVE
&&
8658 (flags
& MOD_PROP_DEFAULT
) != MOD_PROP_DEFAULT
&&
8659 (flags
& MOD_PROP_POSSIBLE
) != MOD_PROP_POSSIBLE
&&
8660 (flags
& MOD_PROP_PERM
) != MOD_PROP_PERM
)
8668 * process the SIOC{SET|GET}PROP ioctl's
8672 ip_sioctl_getsetprop(ipif_t
*dummy_ipif
, sin_t
*dummy_sin
, queue_t
*q
,
8673 mblk_t
*mp
, ip_ioctl_cmd_t
*ipip
, void *dummy_if_req
)
8675 int ioc_cmd
= ((struct iocblk
*)mp
->b_rptr
)->ioc_cmd
;
8677 mod_ioc_prop_t
*pioc
;
8678 mod_prop_info_t
*ptbl
= NULL
, *pinfo
= NULL
;
8685 ASSERT(q
->q_next
== NULL
);
8688 mp1
= mp
->b_cont
->b_cont
;
8689 ipst
= CONNQ_TO_IPST(q
);
8690 stack
= ipst
->ips_netstack
;
8691 pioc
= (mod_ioc_prop_t
*)mp1
->b_rptr
;
8692 if (!getset_ioctl_checks(pioc
, ioc_cmd
))
8695 switch (pioc
->mpr_proto
) {
8697 case MOD_PROTO_IPV4
:
8698 case MOD_PROTO_IPV6
:
8699 ptbl
= ipst
->ips_propinfo_tbl
;
8701 case MOD_PROTO_RAWIP
:
8702 ptbl
= stack
->netstack_icmp
->is_propinfo_tbl
;
8705 ptbl
= stack
->netstack_tcp
->tcps_propinfo_tbl
;
8708 ptbl
= stack
->netstack_udp
->us_propinfo_tbl
;
8710 case MOD_PROTO_SCTP
:
8711 ptbl
= stack
->netstack_sctp
->sctps_propinfo_tbl
;
8717 pioc
->mpr_ifname
[sizeof(pioc
->mpr_ifname
)-1] = '\0';
8718 pioc
->mpr_name
[sizeof(pioc
->mpr_name
)-1] = '\0';
8719 pioc
->mpr_val
[sizeof(pioc
->mpr_val
)-1] = '\0';
8721 pinfo
= mod_prop_lookup(ptbl
, pioc
->mpr_name
, pioc
->mpr_proto
);
8725 set
= (ioc_cmd
== SIOCSETPROP
) ? B_TRUE
: B_FALSE
;
8726 if (set
&& pinfo
->mpi_setf
!= NULL
) {
8727 cr
= msg_getcred(mp
, NULL
);
8728 err
= pinfo
->mpi_setf(stack
, cr
, pinfo
, pioc
->mpr_ifname
,
8729 pioc
->mpr_val
, pioc
->mpr_flags
);
8730 } else if (!set
&& pinfo
->mpi_getf
!= NULL
) {
8731 err
= pinfo
->mpi_getf(stack
, pinfo
, pioc
->mpr_ifname
,
8732 pioc
->mpr_val
, sizeof(pioc
->mpr_val
), pioc
->mpr_flags
);
8741 * process the legacy ND_GET, ND_SET ioctl just for {ip|ip6}_forwarding
8742 * as several routing daemons have unfortunately used this 'unpublished'
8743 * but well-known ioctls.
8747 ip_process_legacy_nddprop(queue_t
*q
, mblk_t
*mp
)
8749 struct iocblk
*iocp
= (struct iocblk
*)mp
->b_rptr
;
8750 mblk_t
*mp1
= mp
->b_cont
;
8751 char *pname
, *pval
, *buf
;
8752 uint_t bufsize
, proto
;
8753 mod_prop_info_t
*pinfo
= NULL
;
8758 ipst
= CONNQ_TO_IPST(q
);
8760 if (iocp
->ioc_count
== 0 || mp1
== NULL
) {
8761 miocnak(q
, mp
, 0, EINVAL
);
8765 mp1
->b_datap
->db_lim
[-1] = '\0'; /* Force null termination */
8766 pval
= buf
= pname
= (char *)mp1
->b_rptr
;
8767 bufsize
= MBLKL(mp1
);
8769 if (strcmp(pname
, "ip_forwarding") == 0) {
8770 pname
= "forwarding";
8771 proto
= MOD_PROTO_IPV4
;
8772 } else if (strcmp(pname
, "ip6_forwarding") == 0) {
8773 pname
= "forwarding";
8774 proto
= MOD_PROTO_IPV6
;
8776 miocnak(q
, mp
, 0, EINVAL
);
8780 pinfo
= mod_prop_lookup(ipst
->ips_propinfo_tbl
, pname
, proto
);
8782 switch (iocp
->ioc_cmd
) {
8784 if ((err
= pinfo
->mpi_getf(ipst
->ips_netstack
, pinfo
, NULL
, buf
,
8785 bufsize
, 0)) == 0) {
8786 miocack(q
, mp
, iocp
->ioc_count
, 0);
8792 * buffer will have property name and value in the following
8794 * <property name>'\0'<property value>'\0', extract them;
8799 if (!*pval
|| pval
>= (char *)mp1
->b_wptr
) {
8801 } else if ((err
= pinfo
->mpi_setf(ipst
->ips_netstack
, NULL
,
8802 pinfo
, NULL
, pval
, 0)) == 0) {
8803 miocack(q
, mp
, 0, 0);
8811 miocnak(q
, mp
, 0, err
);
8815 * Wrapper function for resuming deferred ioctl processing
8816 * Used for SIOCGDSTINFO, SIOCGIP6ADDRPOLICY, SIOCGMSFILTER,
8817 * SIOCSMSFILTER, SIOCGIPMSFILTER, and SIOCSIPMSFILTER currently.
8821 ip_sioctl_copyin_resume(ipsq_t
*dummy_ipsq
, queue_t
*q
, mblk_t
*mp
,
8824 ip_sioctl_copyin_setup(q
, mp
);
8828 * ip_sioctl_copyin_setup is called by ip_wput_nondata with any M_IOCTL message
8829 * that arrives. Most of the IOCTLs are "socket" IOCTLs which we handle
8830 * in either I_STR or TRANSPARENT form, using the mi_copy facility.
8831 * We establish here the size of the block to be copied in. mi_copyin
8832 * arranges for this to happen, an processing continues in ip_wput_nondata with
8833 * an M_IOCDATA message.
8836 ip_sioctl_copyin_setup(queue_t
*q
, mblk_t
*mp
)
8839 struct iocblk
*iocp
= (struct iocblk
*)mp
->b_rptr
;
8840 ip_ioctl_cmd_t
*ipip
;
8845 ipst
= CONNQ_TO_IPST(q
);
8847 ipst
= ILLQ_TO_IPST(q
);
8849 ipip
= ip_sioctl_lookup(iocp
->ioc_cmd
);
8852 * The ioctl is not one we understand or own.
8853 * Pass it along to be processed down stream,
8854 * if this is a module instance of IP, else nak
8857 if (q
->q_next
== NULL
) {
8866 * If this is deferred, then we will do all the checks when we
8869 if ((iocp
->ioc_cmd
== SIOCGDSTINFO
||
8870 iocp
->ioc_cmd
== SIOCGIP6ADDRPOLICY
) && !ip6_asp_can_lookup(ipst
)) {
8871 ip6_asp_pending_op(q
, mp
, ip_sioctl_copyin_resume
);
8876 * Only allow a very small subset of IP ioctls on this stream if
8877 * IP is a module and not a driver. Allowing ioctls to be processed
8878 * in this case may cause assert failures or data corruption.
8879 * Typically G[L]IFFLAGS, SLIFNAME/IF_UNITSEL are the only few
8880 * ioctls allowed on an IP module stream, after which this stream
8881 * normally becomes a multiplexor (at which time the stream head
8882 * will fail all ioctls).
8884 if ((q
->q_next
!= NULL
) && !(ipip
->ipi_flags
& IPI_MODOK
)) {
8888 /* Make sure we have ioctl data to process. */
8889 if (mp
->b_cont
== NULL
&& !(ipip
->ipi_flags
& IPI_NULL_BCONT
))
8893 * Prefer dblk credential over ioctl credential; some synthesized
8894 * ioctls have kcred set because there's no way to crhold()
8895 * a credential in some contexts. (ioc_cr is not crfree() by
8896 * the framework; the caller of ioctl needs to hold the reference
8897 * for the duration of the call).
8899 cr
= msg_getcred(mp
, NULL
);
8903 /* Make sure normal users don't send down privileged ioctls */
8904 if ((ipip
->ipi_flags
& IPI_PRIV
) &&
8905 (cr
!= NULL
) && secpolicy_ip_config(cr
, B_TRUE
) != 0) {
8906 /* We checked the privilege earlier but log it here */
8907 miocnak(q
, mp
, 0, secpolicy_ip_config(cr
, B_FALSE
));
8912 * The ioctl command tables can only encode fixed length
8913 * ioctl data. If the length is variable, the table will
8914 * encode the length as zero. Such special cases are handled
8915 * below in the switch.
8917 if (ipip
->ipi_copyin_size
!= 0) {
8918 mi_copyin(q
, mp
, NULL
, ipip
->ipi_copyin_size
);
8922 switch (iocp
->ioc_cmd
) {
8926 * This IOCTL is hilarious. See comments in
8927 * ip_sioctl_get_ifconf for the story.
8929 if (iocp
->ioc_count
== TRANSPARENT
)
8930 copyin_size
= SIZEOF_STRUCT(ifconf
,
8933 copyin_size
= iocp
->ioc_count
;
8934 mi_copyin(q
, mp
, NULL
, copyin_size
);
8937 case O_SIOCGLIFCONF
:
8939 copyin_size
= SIZEOF_STRUCT(lifconf
, iocp
->ioc_flag
);
8940 mi_copyin(q
, mp
, NULL
, copyin_size
);
8944 copyin_size
= SIZEOF_STRUCT(lifsrcof
, iocp
->ioc_flag
);
8945 mi_copyin(q
, mp
, NULL
, copyin_size
);
8948 case SIOCGIP6ADDRPOLICY
:
8949 ip_sioctl_ip6addrpolicy(q
, mp
);
8950 ip6_asp_table_refrele(ipst
);
8953 case SIOCSIP6ADDRPOLICY
:
8954 ip_sioctl_ip6addrpolicy(q
, mp
);
8958 ip_sioctl_dstinfo(q
, mp
);
8959 ip6_asp_table_refrele(ipst
);
8964 ip_process_legacy_nddprop(q
, mp
);
8972 * We treat non-persistent link similarly as the persistent
8973 * link case, in terms of plumbing/unplumbing, as well as
8974 * dynamic re-plumbing events indicator. See comments
8975 * in ip_sioctl_plink() for more.
8977 * Request can be enqueued in the 'ipsq' while waiting
8978 * to become exclusive. So bump up the conn ref.
8981 CONN_INC_REF(Q_TO_CONN(q
));
8982 CONN_INC_IOCTLREF(Q_TO_CONN(q
))
8984 ip_sioctl_plink(NULL
, q
, mp
, NULL
);
8988 ip_wput_ioctl(q
, mp
);
8992 /* The ioctl length varies depending on the ILB command. */
8993 copyin_size
= iocp
->ioc_count
;
8994 if (copyin_size
< sizeof (ilb_cmd_t
))
8996 mi_copyin(q
, mp
, NULL
, copyin_size
);
9000 cmn_err(CE_WARN
, "Unknown ioctl %d/0x%x slipped through.",
9001 iocp
->ioc_cmd
, iocp
->ioc_cmd
);
9005 if (mp
->b_cont
!= NULL
) {
9006 freemsg(mp
->b_cont
);
9009 iocp
->ioc_error
= EINVAL
;
9010 mp
->b_datap
->db_type
= M_IOCNAK
;
9011 iocp
->ioc_count
= 0;
9016 ip_sioctl_garp_reply(mblk_t
*mp
, ill_t
*ill
, void *hwaddr
, int flags
)
9019 struct xarpreq
*xar
;
9021 struct iocblk
*iocp
;
9022 int x_arp_ioctl
= B_FALSE
;
9024 char *storage
= NULL
;
9026 ASSERT(ill
!= NULL
);
9028 iocp
= (struct iocblk
*)mp
->b_rptr
;
9029 ASSERT(iocp
->ioc_cmd
== SIOCGXARP
|| iocp
->ioc_cmd
== SIOCGARP
);
9031 tmp
= (mp
->b_cont
)->b_cont
; /* xarpreq/arpreq */
9032 if ((iocp
->ioc_cmd
== SIOCGXARP
) ||
9033 (iocp
->ioc_cmd
== SIOCSXARP
)) {
9034 x_arp_ioctl
= B_TRUE
;
9035 xar
= (struct xarpreq
*)tmp
->b_rptr
;
9036 flagsp
= &xar
->xarp_flags
;
9037 storage
= xar
->xarp_ha
.sdl_data
;
9039 ar
= (struct arpreq
*)tmp
->b_rptr
;
9040 flagsp
= &ar
->arp_flags
;
9041 storage
= ar
->arp_ha
.sa_data
;
9045 * We're done if this is not an SIOCG{X}ARP
9048 storage
+= ill_xarp_info(&xar
->xarp_ha
, ill
);
9049 if ((ill
->ill_phys_addr_length
+ ill
->ill_name_length
) >
9050 sizeof (xar
->xarp_ha
.sdl_data
)) {
9051 iocp
->ioc_error
= EINVAL
;
9055 *flagsp
= ATF_INUSE
;
9057 * If /sbin/arp told us we are the authority using the "permanent"
9058 * flag, or if this is one of my addresses print "permanent"
9059 * in the /sbin/arp output.
9061 if ((flags
& NCE_F_MYADDR
) || (flags
& NCE_F_AUTHORITY
))
9062 *flagsp
|= ATF_AUTHORITY
;
9063 if (flags
& NCE_F_NONUD
)
9064 *flagsp
|= ATF_PERM
; /* not subject to aging */
9065 if (flags
& NCE_F_PUBLISH
)
9066 *flagsp
|= ATF_PUBL
;
9067 if (hwaddr
!= NULL
) {
9069 bcopy((char *)hwaddr
, storage
, ill
->ill_phys_addr_length
);
9074 * Create a new logical interface. If ipif_id is zero (i.e. not a logical
9075 * interface) create the next available logical interface for this
9076 * physical interface.
9077 * If ipif is NULL (i.e. the lookup didn't find one) attempt to create an
9078 * ipif with the specified name.
9080 * If the address family is not AF_UNSPEC then set the address as well.
9082 * If ip_sioctl_addr returns EINPROGRESS then the ioctl (the copyout)
9083 * is completed when the DL_BIND_ACK arrive in ip_rput_dlpi_writer.
9085 * Executed as a writer on the ill.
9086 * So no lock is needed to traverse the ipif chain, or examine the
9091 ip_sioctl_addif(ipif_t
*dummy_ipif
, sin_t
*dummy_sin
, queue_t
*q
, mblk_t
*mp
,
9092 ip_ioctl_cmd_t
*dummy_ipip
, void *dummy_ifreq
)
9095 struct lifreq
*lifr
;
9108 boolean_t found_sep
= B_FALSE
;
9111 ip_stack_t
*ipst
= CONNQ_TO_IPST(q
);
9113 ASSERT(q
->q_next
== NULL
);
9114 ip1dbg(("ip_sioctl_addif\n"));
9115 /* Existence of mp1 has been checked in ip_wput_nondata */
9116 mp1
= mp
->b_cont
->b_cont
;
9118 * Null terminate the string to protect against buffer
9119 * overrun. String was generated by user code and may not
9122 lifr
= (struct lifreq
*)mp1
->b_rptr
;
9123 lifr
->lifr_name
[LIFNAMSIZ
- 1] = '\0';
9124 name
= lifr
->lifr_name
;
9126 connp
= Q_TO_CONN(q
);
9127 isv6
= (connp
->conn_family
== AF_INET6
);
9128 zoneid
= connp
->conn_zoneid
;
9129 namelen
= mi_strlen(name
);
9134 if ((namelen
+ 1 == sizeof (ipif_loopback_name
)) &&
9135 (mi_strcmp(name
, ipif_loopback_name
) == 0)) {
9137 * Allow creating lo0 using SIOCLIFADDIF.
9138 * can't be any other writer thread. So can pass null below
9139 * for the last 4 args to ipif_lookup_name.
9141 ipif
= ipif_lookup_on_name(lifr
->lifr_name
, namelen
, B_TRUE
,
9142 &exists
, isv6
, zoneid
, ipst
);
9143 /* Prevent any further action */
9146 } else if (!exists
) {
9147 /* We created the ipif now and as writer */
9151 ill
= ipif
->ipif_ill
;
9156 /* Look for a colon in the name. */
9157 endp
= &name
[namelen
];
9158 for (cp
= endp
; --cp
> name
; ) {
9159 if (*cp
== IPIF_SEPARATOR_CHAR
) {
9162 * Reject any non-decimal aliases for plumbing
9163 * of logical interfaces. Aliases with leading
9164 * zeroes are also rejected as they introduce
9165 * ambiguity in the naming of the interfaces.
9166 * Comparing with "0" takes care of all such
9169 if ((strncmp("0", cp
+1, 1)) == 0)
9172 if (ddi_strtol(cp
+1, &endp
, 10, &id
) != 0 ||
9173 id
<= 0 || *endp
!= '\0') {
9180 ill
= ill_lookup_on_name(name
, B_FALSE
, isv6
, NULL
, ipst
);
9182 *cp
= IPIF_SEPARATOR_CHAR
;
9187 ipsq
= ipsq_try_enter(NULL
, ill
, q
, mp
, ip_process_ioctl
, NEW_OP
,
9191 * Release the refhold due to the lookup, now that we are excl
9192 * or we are just returning
9197 return (EINPROGRESS
);
9199 /* We are now exclusive on the IPSQ */
9200 ASSERT(IAM_WRITER_ILL(ill
));
9203 /* Now see if there is an IPIF with this unit number. */
9204 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
;
9205 ipif
= ipif
->ipif_next
) {
9206 if (ipif
->ipif_id
== id
) {
9214 * We use IRE_LOCAL for lo0:1 etc. for "receive only" use
9215 * of lo0. Plumbing for lo0:0 happens in ipif_lookup_on_name()
9218 if ((ipif
= ipif_allocate(ill
, found_sep
? id
: -1, IRE_LOCAL
,
9219 B_TRUE
, B_TRUE
, &err
)) == NULL
) {
9223 /* Return created name with ioctl */
9224 (void) sprintf(lifr
->lifr_name
, "%s%c%d", ill
->ill_name
,
9225 IPIF_SEPARATOR_CHAR
, ipif
->ipif_id
);
9226 ip1dbg(("created %s\n", lifr
->lifr_name
));
9229 sin
= (sin_t
*)&lifr
->lifr_addr
;
9230 if (sin
->sin_family
!= AF_UNSPEC
) {
9231 err
= ip_sioctl_addr(ipif
, sin
, q
, mp
,
9232 &ip_ndx_ioctl_table
[SIOCLIFADDR_NDX
], lifr
);
9241 * Remove an existing logical interface. If ipif_id is zero (i.e. not a logical
9242 * interface) delete it based on the IP address (on this physical interface).
9243 * Otherwise delete it based on the ipif_id.
9244 * Also, special handling to allow a removeif of lo0.
9248 ip_sioctl_removeif(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
9249 ip_ioctl_cmd_t
*ipip
, void *dummy_if_req
)
9252 ill_t
*ill
= ipif
->ipif_ill
;
9256 ipst
= CONNQ_TO_IPST(q
);
9258 ASSERT(q
->q_next
== NULL
);
9259 ip1dbg(("ip_sioctl_remove_if(%s:%u %p)\n",
9260 ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
9261 ASSERT(IAM_WRITER_IPIF(ipif
));
9263 connp
= Q_TO_CONN(q
);
9265 * Special case for unplumbing lo0 (the loopback physical interface).
9266 * If unplumbing lo0, the incoming address structure has been
9267 * initialized to all zeros. When unplumbing lo0, all its logical
9268 * interfaces must be removed too.
9270 * Note that this interface may be called to remove a specific
9271 * loopback logical interface (eg, lo0:1). But in that case
9272 * ipif->ipif_id != 0 so that the code path for that case is the
9273 * same as any other interface (meaning it skips the code directly
9276 if (ipif
->ipif_id
== 0 && ill
->ill_net_type
== IRE_LOOPBACK
) {
9277 if (sin
->sin_family
== AF_UNSPEC
&&
9278 (IN6_IS_ADDR_UNSPECIFIED(&((sin6_t
*)sin
)->sin6_addr
))) {
9280 * Mark it condemned. No new ref. will be made to ill.
9282 mutex_enter(&ill
->ill_lock
);
9283 ill
->ill_state_flags
|= ILL_CONDEMNED
;
9284 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
;
9285 ipif
= ipif
->ipif_next
) {
9286 ipif
->ipif_state_flags
|= IPIF_CONDEMNED
;
9288 mutex_exit(&ill
->ill_lock
);
9290 ipif
= ill
->ill_ipif
;
9291 /* unplumb the loopback interface */
9293 mutex_enter(&connp
->conn_lock
);
9294 mutex_enter(&ill
->ill_lock
);
9296 /* Are any references to this ill active */
9297 if (ill_is_freeable(ill
)) {
9298 mutex_exit(&ill
->ill_lock
);
9299 mutex_exit(&connp
->conn_lock
);
9300 ill_delete_tail(ill
);
9304 success
= ipsq_pending_mp_add(connp
, ipif
,
9305 CONNP_TO_WQ(connp
), mp
, ILL_FREE
);
9306 mutex_exit(&connp
->conn_lock
);
9307 mutex_exit(&ill
->ill_lock
);
9309 return (EINPROGRESS
);
9315 if (ipif
->ipif_id
== 0) {
9318 /* Find based on address */
9319 if (ipif
->ipif_isv6
) {
9322 if (sin
->sin_family
!= AF_INET6
)
9323 return (EAFNOSUPPORT
);
9325 sin6
= (sin6_t
*)sin
;
9326 /* We are a writer, so we should be able to lookup */
9327 ipif
= ipif_lookup_addr_exact_v6(&sin6
->sin6_addr
, ill
,
9330 if (sin
->sin_family
!= AF_INET
)
9331 return (EAFNOSUPPORT
);
9333 /* We are a writer, so we should be able to lookup */
9334 ipif
= ipif_lookup_addr_exact(sin
->sin_addr
.s_addr
, ill
,
9338 return (EADDRNOTAVAIL
);
9342 * It is possible for a user to send an SIOCLIFREMOVEIF with
9343 * lifr_name of the physical interface but with an ip address
9344 * lifr_addr of a logical interface plumbed over it.
9345 * So update ipx_current_ipif now that ipif points to the
9348 ipsq
= ipif
->ipif_ill
->ill_phyint
->phyint_ipsq
;
9349 ipsq
->ipsq_xop
->ipx_current_ipif
= ipif
;
9351 /* This is a writer */
9356 * Can not delete instance zero since it is tied to the ill.
9358 if (ipif
->ipif_id
== 0)
9361 mutex_enter(&ill
->ill_lock
);
9362 ipif
->ipif_state_flags
|= IPIF_CONDEMNED
;
9363 mutex_exit(&ill
->ill_lock
);
9367 mutex_enter(&connp
->conn_lock
);
9368 mutex_enter(&ill
->ill_lock
);
9370 /* Are any references to this ipif active */
9371 if (ipif_is_freeable(ipif
)) {
9372 mutex_exit(&ill
->ill_lock
);
9373 mutex_exit(&connp
->conn_lock
);
9374 ipif_non_duplicate(ipif
);
9375 (void) ipif_down_tail(ipif
);
9376 ipif_free_tail(ipif
); /* frees ipif */
9379 success
= ipsq_pending_mp_add(connp
, ipif
, CONNP_TO_WQ(connp
), mp
,
9381 mutex_exit(&ill
->ill_lock
);
9382 mutex_exit(&connp
->conn_lock
);
9384 return (EINPROGRESS
);
9390 * Restart the removeif ioctl. The refcnt has gone down to 0.
9391 * The ipif is already condemned. So can't find it thru lookups.
9395 ip_sioctl_removeif_restart(ipif_t
*ipif
, sin_t
*dummy_sin
, queue_t
*q
,
9396 mblk_t
*mp
, ip_ioctl_cmd_t
*ipip
, void *dummy_if_req
)
9398 ill_t
*ill
= ipif
->ipif_ill
;
9400 ASSERT(IAM_WRITER_IPIF(ipif
));
9401 ASSERT(ipif
->ipif_state_flags
& IPIF_CONDEMNED
);
9403 ip1dbg(("ip_sioctl_removeif_restart(%s:%u %p)\n",
9404 ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
9406 if (ipif
->ipif_id
== 0 && ill
->ill_net_type
== IRE_LOOPBACK
) {
9407 ASSERT(ill
->ill_state_flags
& ILL_CONDEMNED
);
9408 ill_delete_tail(ill
);
9413 ipif_non_duplicate(ipif
);
9414 (void) ipif_down_tail(ipif
);
9415 ipif_free_tail(ipif
);
9421 * Set the local interface address using the given prefix and ill_token.
9425 ip_sioctl_prefix(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
9426 ip_ioctl_cmd_t
*dummy_ipip
, void *dummy_ifreq
)
9434 ip1dbg(("ip_sioctl_prefix(%s:%u %p)\n",
9435 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
9437 ASSERT(IAM_WRITER_IPIF(ipif
));
9439 if (!ipif
->ipif_isv6
)
9442 if (sin
->sin_family
!= AF_INET6
)
9443 return (EAFNOSUPPORT
);
9445 sin6
= (sin6_t
*)sin
;
9446 v6addr
= sin6
->sin6_addr
;
9447 ill
= ipif
->ipif_ill
;
9449 if (IN6_IS_ADDR_UNSPECIFIED(&v6addr
) ||
9450 IN6_IS_ADDR_UNSPECIFIED(&ill
->ill_token
))
9451 return (EADDRNOTAVAIL
);
9453 for (i
= 0; i
< 4; i
++)
9454 sin6
->sin6_addr
.s6_addr32
[i
] |= ill
->ill_token
.s6_addr32
[i
];
9456 err
= ip_sioctl_addr(ipif
, sin
, q
, mp
,
9457 &ip_ndx_ioctl_table
[SIOCLIFADDR_NDX
], dummy_ifreq
);
9462 * Restart entry point to restart the address set operation after the
9463 * refcounts have dropped to zero.
9467 ip_sioctl_prefix_restart(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
9468 ip_ioctl_cmd_t
*ipip
, void *ifreq
)
9470 ip1dbg(("ip_sioctl_prefix_restart(%s:%u %p)\n",
9471 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
9472 return (ip_sioctl_addr_restart(ipif
, sin
, q
, mp
, ipip
, ifreq
));
9476 * Set the local interface address.
9477 * Allow an address of all zero when the interface is down.
9481 ip_sioctl_addr(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
9482 ip_ioctl_cmd_t
*dummy_ipip
, void *dummy_ifreq
)
9486 boolean_t need_up
= B_FALSE
;
9490 ip1dbg(("ip_sioctl_addr(%s:%u %p)\n",
9491 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
9493 ASSERT(IAM_WRITER_IPIF(ipif
));
9495 ill
= ipif
->ipif_ill
;
9496 if (ipif
->ipif_isv6
) {
9500 if (sin
->sin_family
!= AF_INET6
)
9501 return (EAFNOSUPPORT
);
9503 sin6
= (sin6_t
*)sin
;
9504 v6addr
= sin6
->sin6_addr
;
9505 phyi
= ill
->ill_phyint
;
9508 * Enforce that true multicast interfaces have a link-local
9509 * address for logical unit 0.
9511 * However for those ipif's for which link-local address was
9512 * not created by default, also allow setting :: as the address.
9513 * This scenario would arise, when we delete an address on ipif
9514 * with logical unit 0, we would want to set :: as the address.
9516 if (ipif
->ipif_id
== 0 &&
9517 (ill
->ill_flags
& ILLF_MULTICAST
) &&
9518 !(ipif
->ipif_flags
& (IPIF_POINTOPOINT
)) &&
9519 !(phyi
->phyint_flags
& (PHYI_LOOPBACK
)) &&
9520 !IN6_IS_ADDR_LINKLOCAL(&v6addr
)) {
9523 * if default link-local was not created by kernel for
9524 * this ill, allow setting :: as the address on ipif:0.
9526 if (ill
->ill_flags
& ILLF_NOLINKLOCAL
) {
9527 if (!IN6_IS_ADDR_UNSPECIFIED(&v6addr
))
9528 return (EADDRNOTAVAIL
);
9530 return (EADDRNOTAVAIL
);
9535 * up interfaces shouldn't have the unspecified address
9536 * unless they also have the IPIF_NOLOCAL flags set and
9537 * have a subnet assigned.
9539 if ((ipif
->ipif_flags
& IPIF_UP
) &&
9540 IN6_IS_ADDR_UNSPECIFIED(&v6addr
) &&
9541 (!(ipif
->ipif_flags
& IPIF_NOLOCAL
) ||
9542 IN6_IS_ADDR_UNSPECIFIED(&ipif
->ipif_v6subnet
))) {
9543 return (EADDRNOTAVAIL
);
9546 if (!ip_local_addr_ok_v6(&v6addr
, &ipif
->ipif_v6net_mask
))
9547 return (EADDRNOTAVAIL
);
9551 if (sin
->sin_family
!= AF_INET
)
9552 return (EAFNOSUPPORT
);
9554 addr
= sin
->sin_addr
.s_addr
;
9556 /* Allow INADDR_ANY as the local address. */
9557 if (addr
!= INADDR_ANY
&&
9558 !ip_addr_ok_v4(addr
, ipif
->ipif_net_mask
))
9559 return (EADDRNOTAVAIL
);
9561 IN6_IPADDR_TO_V4MAPPED(addr
, &v6addr
);
9564 * verify that the address being configured is permitted by the
9565 * ill_allowed_ips[] for the interface.
9567 if (ill
->ill_allowed_ips_cnt
> 0) {
9568 for (i
= 0; i
< ill
->ill_allowed_ips_cnt
; i
++) {
9569 if (IN6_ARE_ADDR_EQUAL(&ill
->ill_allowed_ips
[i
],
9573 if (i
== ill
->ill_allowed_ips_cnt
) {
9574 pr_addr_dbg("!allowed addr %s\n", AF_INET6
, &v6addr
);
9579 * Even if there is no change we redo things just to rerun
9582 if (ipif
->ipif_flags
& IPIF_UP
) {
9584 * Setting a new local address, make sure
9585 * we have net and subnet bcast ire's for
9586 * the old address if we need them.
9589 * If the interface is already marked up,
9590 * we call ipif_down which will take care
9591 * of ditching any IREs that have been set
9592 * up based on the old interface address.
9594 err
= ipif_logical_down(ipif
, q
, mp
);
9595 if (err
== EINPROGRESS
)
9597 (void) ipif_down_tail(ipif
);
9601 err
= ip_sioctl_addr_tail(ipif
, sin
, q
, mp
, need_up
);
9606 ip_sioctl_addr_tail(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
9615 ill_t
*ill
= ipif
->ipif_ill
;
9616 boolean_t need_dl_down
;
9617 boolean_t need_arp_down
;
9618 struct iocblk
*iocp
;
9620 iocp
= (mp
!= NULL
) ? (struct iocblk
*)mp
->b_rptr
: NULL
;
9622 ip1dbg(("ip_sioctl_addr_tail(%s:%u %p)\n",
9623 ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
9624 ASSERT(IAM_WRITER_IPIF(ipif
));
9626 /* Must cancel any pending timer before taking the ill_lock */
9627 if (ipif
->ipif_recovery_id
!= 0)
9628 (void) untimeout(ipif
->ipif_recovery_id
);
9629 ipif
->ipif_recovery_id
= 0;
9631 if (ipif
->ipif_isv6
) {
9632 sin6
= (sin6_t
*)sin
;
9633 v6addr
= sin6
->sin6_addr
;
9634 sinlen
= sizeof (struct sockaddr_in6
);
9636 addr
= sin
->sin_addr
.s_addr
;
9637 IN6_IPADDR_TO_V4MAPPED(addr
, &v6addr
);
9638 sinlen
= sizeof (struct sockaddr_in
);
9640 mutex_enter(&ill
->ill_lock
);
9641 ov6addr
= ipif
->ipif_v6lcl_addr
;
9642 ipif
->ipif_v6lcl_addr
= v6addr
;
9643 sctp_update_ipif_addr(ipif
, ov6addr
);
9644 ipif
->ipif_addr_ready
= 0;
9646 ip_rts_newaddrmsg(RTM_CHGADDR
, 0, ipif
, RTSQ_DEFAULT
);
9649 * If the interface was previously marked as a duplicate, then since
9650 * we've now got a "new" address, it should no longer be considered a
9651 * duplicate -- even if the "new" address is the same as the old one.
9652 * Note that if all ipifs are down, we may have a pending ARP down
9653 * event to handle. This is because we want to recover from duplicates
9654 * and thus delay tearing down ARP until the duplicates have been
9655 * removed or disabled.
9657 need_dl_down
= need_arp_down
= B_FALSE
;
9658 if (ipif
->ipif_flags
& IPIF_DUPLICATE
) {
9659 need_arp_down
= !need_up
;
9660 ipif
->ipif_flags
&= ~IPIF_DUPLICATE
;
9661 if (--ill
->ill_ipif_dup_count
== 0 && !need_up
&&
9662 ill
->ill_ipif_up_count
== 0 && ill
->ill_dl_up
) {
9663 need_dl_down
= B_TRUE
;
9667 ipif_set_default(ipif
);
9670 * If we've just manually set the IPv6 link-local address (0th ipif),
9671 * tag the ill so that future updates to the interface ID don't result
9672 * in this address getting automatically reconfigured from under the
9675 if (ipif
->ipif_isv6
&& ipif
->ipif_id
== 0) {
9676 if (iocp
== NULL
|| (iocp
->ioc_cmd
== SIOCSLIFADDR
&&
9677 !IN6_IS_ADDR_UNSPECIFIED(&v6addr
)))
9678 ill
->ill_manual_linklocal
= 1;
9682 * When publishing an interface address change event, we only notify
9683 * the event listeners of the new address. It is assumed that if they
9684 * actively care about the addresses assigned that they will have
9685 * already discovered the previous address assigned (if there was one.)
9687 * Don't attach nic event message for SIOCLIFADDIF ioctl.
9689 if (iocp
!= NULL
&& iocp
->ioc_cmd
!= SIOCLIFADDIF
) {
9690 ill_nic_event_dispatch(ill
, MAP_IPIF_ID(ipif
->ipif_id
),
9691 NE_ADDRESS_CHANGE
, sin
, sinlen
);
9694 mutex_exit(&ill
->ill_lock
);
9698 * Now bring the interface back up. If this
9699 * is the only IPIF for the ILL, ipif_up
9700 * will have to re-bind to the device, so
9701 * we may get back EINPROGRESS, in which
9702 * case, this IOCTL will get completed in
9703 * ip_rput_dlpi when we see the DL_BIND_ACK.
9705 err
= ipif_up(ipif
, q
, mp
);
9707 /* Perhaps ilgs should use this ill */
9708 update_conn_ill(NULL
, ill
->ill_ipst
);
9714 if (need_arp_down
&& !ill
->ill_isv6
)
9715 (void) ipif_arp_down(ipif
);
9718 * The default multicast interface might have changed (for
9719 * instance if the IPv6 scope of the address changed)
9721 ire_increment_multicast_generation(ill
->ill_ipst
, ill
->ill_isv6
);
9727 * Restart entry point to restart the address set operation after the
9728 * refcounts have dropped to zero.
9732 ip_sioctl_addr_restart(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
9733 ip_ioctl_cmd_t
*ipip
, void *ifreq
)
9735 ip1dbg(("ip_sioctl_addr_restart(%s:%u %p)\n",
9736 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
9737 ASSERT(IAM_WRITER_IPIF(ipif
));
9738 (void) ipif_down_tail(ipif
);
9739 return (ip_sioctl_addr_tail(ipif
, sin
, q
, mp
, B_TRUE
));
9744 ip_sioctl_get_addr(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
9745 ip_ioctl_cmd_t
*ipip
, void *if_req
)
9747 sin6_t
*sin6
= (struct sockaddr_in6
*)sin
;
9748 struct lifreq
*lifr
= (struct lifreq
*)if_req
;
9750 ip1dbg(("ip_sioctl_get_addr(%s:%u %p)\n",
9751 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
9753 * The net mask and address can't change since we have a
9754 * reference to the ipif. So no lock is necessary.
9756 if (ipif
->ipif_isv6
) {
9758 sin6
->sin6_family
= AF_INET6
;
9759 sin6
->sin6_addr
= ipif
->ipif_v6lcl_addr
;
9760 if (IN6_IS_ADDR_LINKLOCAL(&sin6
->sin6_addr
)) {
9761 sin6
->sin6_scope_id
=
9762 ipif
->ipif_ill
->ill_phyint
->phyint_ifindex
;
9764 ASSERT(ipip
->ipi_cmd_type
== LIF_CMD
);
9765 lifr
->lifr_addrlen
=
9766 ip_mask_to_plen_v6(&ipif
->ipif_v6net_mask
);
9769 sin
->sin_family
= AF_INET
;
9770 sin
->sin_addr
.s_addr
= ipif
->ipif_lcl_addr
;
9771 if (ipip
->ipi_cmd_type
== LIF_CMD
) {
9772 lifr
->lifr_addrlen
=
9773 ip_mask_to_plen(ipif
->ipif_net_mask
);
9780 * Set the destination address for a pt-pt interface.
9784 ip_sioctl_dstaddr(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
9785 ip_ioctl_cmd_t
*ipip
, void *if_req
)
9789 boolean_t need_up
= B_FALSE
;
9791 ip1dbg(("ip_sioctl_dstaddr(%s:%u %p)\n",
9792 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
9793 ASSERT(IAM_WRITER_IPIF(ipif
));
9795 if (ipif
->ipif_isv6
) {
9798 if (sin
->sin_family
!= AF_INET6
)
9799 return (EAFNOSUPPORT
);
9801 sin6
= (sin6_t
*)sin
;
9802 v6addr
= sin6
->sin6_addr
;
9804 if (!ip_remote_addr_ok_v6(&v6addr
, &ipif
->ipif_v6net_mask
))
9805 return (EADDRNOTAVAIL
);
9809 if (sin
->sin_family
!= AF_INET
)
9810 return (EAFNOSUPPORT
);
9812 addr
= sin
->sin_addr
.s_addr
;
9813 if (addr
!= INADDR_ANY
&&
9814 !ip_addr_ok_v4(addr
, ipif
->ipif_net_mask
)) {
9815 return (EADDRNOTAVAIL
);
9818 IN6_IPADDR_TO_V4MAPPED(addr
, &v6addr
);
9821 if (IN6_ARE_ADDR_EQUAL(&ipif
->ipif_v6pp_dst_addr
, &v6addr
))
9822 return (0); /* No change */
9824 if (ipif
->ipif_flags
& IPIF_UP
) {
9826 * If the interface is already marked up,
9827 * we call ipif_down which will take care
9828 * of ditching any IREs that have been set
9829 * up based on the old pp dst address.
9831 err
= ipif_logical_down(ipif
, q
, mp
);
9832 if (err
== EINPROGRESS
)
9834 (void) ipif_down_tail(ipif
);
9838 * could return EINPROGRESS. If so ioctl will complete in
9839 * ip_rput_dlpi_writer
9841 err
= ip_sioctl_dstaddr_tail(ipif
, sin
, q
, mp
, need_up
);
9846 ip_sioctl_dstaddr_tail(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
9850 ill_t
*ill
= ipif
->ipif_ill
;
9852 boolean_t need_dl_down
;
9853 boolean_t need_arp_down
;
9855 ip1dbg(("ip_sioctl_dstaddr_tail(%s:%u %p)\n", ill
->ill_name
,
9856 ipif
->ipif_id
, (void *)ipif
));
9858 /* Must cancel any pending timer before taking the ill_lock */
9859 if (ipif
->ipif_recovery_id
!= 0)
9860 (void) untimeout(ipif
->ipif_recovery_id
);
9861 ipif
->ipif_recovery_id
= 0;
9863 if (ipif
->ipif_isv6
) {
9866 sin6
= (sin6_t
*)sin
;
9867 v6addr
= sin6
->sin6_addr
;
9871 addr
= sin
->sin_addr
.s_addr
;
9872 IN6_IPADDR_TO_V4MAPPED(addr
, &v6addr
);
9874 mutex_enter(&ill
->ill_lock
);
9875 /* Set point to point destination address. */
9876 if ((ipif
->ipif_flags
& IPIF_POINTOPOINT
) == 0) {
9878 * Allow this as a means of creating logical
9879 * pt-pt interfaces on top of e.g. an Ethernet.
9880 * XXX Undocumented HACK for testing.
9881 * pt-pt interfaces are created with NUD disabled.
9883 ipif
->ipif_flags
|= IPIF_POINTOPOINT
;
9884 ipif
->ipif_flags
&= ~IPIF_BROADCAST
;
9885 if (ipif
->ipif_isv6
)
9886 ill
->ill_flags
|= ILLF_NONUD
;
9890 * If the interface was previously marked as a duplicate, then since
9891 * we've now got a "new" address, it should no longer be considered a
9892 * duplicate -- even if the "new" address is the same as the old one.
9893 * Note that if all ipifs are down, we may have a pending ARP down
9896 need_dl_down
= need_arp_down
= B_FALSE
;
9897 if (ipif
->ipif_flags
& IPIF_DUPLICATE
) {
9898 need_arp_down
= !need_up
;
9899 ipif
->ipif_flags
&= ~IPIF_DUPLICATE
;
9900 if (--ill
->ill_ipif_dup_count
== 0 && !need_up
&&
9901 ill
->ill_ipif_up_count
== 0 && ill
->ill_dl_up
) {
9902 need_dl_down
= B_TRUE
;
9907 * If we've just manually set the IPv6 destination link-local address
9908 * (0th ipif), tag the ill so that future updates to the destination
9909 * interface ID (as can happen with interfaces over IP tunnels) don't
9910 * result in this address getting automatically reconfigured from
9911 * under the administrator.
9913 if (ipif
->ipif_isv6
&& ipif
->ipif_id
== 0)
9914 ill
->ill_manual_dst_linklocal
= 1;
9916 /* Set the new address. */
9917 ipif
->ipif_v6pp_dst_addr
= v6addr
;
9918 /* Make sure subnet tracks pp_dst */
9919 ipif
->ipif_v6subnet
= ipif
->ipif_v6pp_dst_addr
;
9920 mutex_exit(&ill
->ill_lock
);
9924 * Now bring the interface back up. If this
9925 * is the only IPIF for the ILL, ipif_up
9926 * will have to re-bind to the device, so
9927 * we may get back EINPROGRESS, in which
9928 * case, this IOCTL will get completed in
9929 * ip_rput_dlpi when we see the DL_BIND_ACK.
9931 err
= ipif_up(ipif
, q
, mp
);
9936 if (need_arp_down
&& !ipif
->ipif_isv6
)
9937 (void) ipif_arp_down(ipif
);
9943 * Restart entry point to restart the dstaddress set operation after the
9944 * refcounts have dropped to zero.
9948 ip_sioctl_dstaddr_restart(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
9949 ip_ioctl_cmd_t
*ipip
, void *ifreq
)
9951 ip1dbg(("ip_sioctl_dstaddr_restart(%s:%u %p)\n",
9952 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
9953 (void) ipif_down_tail(ipif
);
9954 return (ip_sioctl_dstaddr_tail(ipif
, sin
, q
, mp
, B_TRUE
));
9959 ip_sioctl_get_dstaddr(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
9960 ip_ioctl_cmd_t
*ipip
, void *if_req
)
9962 sin6_t
*sin6
= (struct sockaddr_in6
*)sin
;
9964 ip1dbg(("ip_sioctl_get_dstaddr(%s:%u %p)\n",
9965 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
9967 * Get point to point destination address. The addresses can't
9968 * change since we hold a reference to the ipif.
9970 if ((ipif
->ipif_flags
& IPIF_POINTOPOINT
) == 0)
9971 return (EADDRNOTAVAIL
);
9973 if (ipif
->ipif_isv6
) {
9974 ASSERT(ipip
->ipi_cmd_type
== LIF_CMD
);
9976 sin6
->sin6_family
= AF_INET6
;
9977 sin6
->sin6_addr
= ipif
->ipif_v6pp_dst_addr
;
9980 sin
->sin_family
= AF_INET
;
9981 sin
->sin_addr
.s_addr
= ipif
->ipif_pp_dst_addr
;
9987 * Check which flags will change by the given flags being set
9988 * silently ignore flags which userland is not allowed to control.
9989 * (Because these flags may change between SIOCGLIFFLAGS and
9990 * SIOCSLIFFLAGS, and that's outside of userland's control,
9991 * we need to silently ignore them rather than fail.)
9994 ip_sioctl_flags_onoff(ipif_t
*ipif
, uint64_t flags
, uint64_t *onp
,
9997 ill_t
*ill
= ipif
->ipif_ill
;
9998 phyint_t
*phyi
= ill
->ill_phyint
;
9999 uint64_t cantchange_flags
, intf_flags
;
10000 uint64_t turn_on
, turn_off
;
10002 intf_flags
= ipif
->ipif_flags
| ill
->ill_flags
| phyi
->phyint_flags
;
10003 cantchange_flags
= IFF_CANTCHANGE
;
10005 cantchange_flags
|= IFF_IPMP_CANTCHANGE
;
10006 turn_on
= (flags
^ intf_flags
) & ~cantchange_flags
;
10007 turn_off
= intf_flags
& turn_on
;
10008 turn_on
^= turn_off
;
10014 * Set interface flags. Many flags require special handling (e.g.,
10015 * bringing the interface down); see below for details.
10017 * NOTE : We really don't enforce that ipif_id zero should be used
10018 * for setting any flags other than IFF_LOGINT_FLAGS. This
10019 * is because applications generally does SICGLIFFLAGS and
10020 * ORs in the new flags (that affects the logical) and does a
10021 * SIOCSLIFFLAGS. Thus, "flags" below could contain bits other
10022 * than IFF_LOGINT_FLAGS. One could check whether "turn_on" - the
10023 * flags that will be turned on is correct with respect to
10024 * ipif_id 0. For backward compatibility reasons, it is not done.
10028 ip_sioctl_flags(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
10029 ip_ioctl_cmd_t
*ipip
, void *if_req
)
10037 uint64_t intf_flags
;
10038 boolean_t phyint_flags_modified
= B_FALSE
;
10041 struct lifreq
*lifr
;
10042 boolean_t set_linklocal
= B_FALSE
;
10044 ip1dbg(("ip_sioctl_flags(%s:%u %p)\n",
10045 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
10047 ASSERT(IAM_WRITER_IPIF(ipif
));
10049 ill
= ipif
->ipif_ill
;
10050 phyi
= ill
->ill_phyint
;
10052 if (ipip
->ipi_cmd_type
== IF_CMD
) {
10053 ifr
= (struct ifreq
*)if_req
;
10054 flags
= (uint64_t)(ifr
->ifr_flags
& 0x0000ffff);
10056 lifr
= (struct lifreq
*)if_req
;
10057 flags
= lifr
->lifr_flags
;
10060 intf_flags
= ipif
->ipif_flags
| ill
->ill_flags
| phyi
->phyint_flags
;
10063 * Have the flags been set correctly until now?
10065 ASSERT((phyi
->phyint_flags
& ~(IFF_PHYINT_FLAGS
)) == 0);
10066 ASSERT((ill
->ill_flags
& ~(IFF_PHYINTINST_FLAGS
)) == 0);
10067 ASSERT((ipif
->ipif_flags
& ~(IFF_LOGINT_FLAGS
)) == 0);
10069 * Compare the new flags to the old, and partition
10070 * into those coming on and those going off.
10071 * For the 16 bit command keep the bits above bit 16 unchanged.
10073 if (ipip
->ipi_cmd
== SIOCSIFFLAGS
)
10074 flags
|= intf_flags
& ~0xFFFF;
10077 * Explicitly fail attempts to change flags that are always invalid on
10078 * an IPMP meta-interface.
10080 if (IS_IPMP(ill
) && ((flags
^ intf_flags
) & IFF_IPMP_INVALID
))
10083 ip_sioctl_flags_onoff(ipif
, flags
, &turn_on
, &turn_off
);
10084 if ((turn_on
|turn_off
) == 0)
10085 return (0); /* No change */
10088 * All test addresses must be IFF_DEPRECATED (to ensure source address
10089 * selection avoids them) -- so force IFF_DEPRECATED on, and do not
10090 * allow it to be turned off.
10092 if ((turn_off
& (IFF_DEPRECATED
|IFF_NOFAILOVER
)) == IFF_DEPRECATED
&&
10093 (turn_on
|intf_flags
) & IFF_NOFAILOVER
)
10096 if ((connp
= Q_TO_CONN(q
)) == NULL
)
10100 * Only vrrp control socket is allowed to change IFF_UP and
10101 * IFF_NOACCEPT flags when IFF_VRRP is set.
10103 if ((intf_flags
& IFF_VRRP
) && ((turn_off
| turn_on
) & IFF_UP
)) {
10104 if (!connp
->conn_isvrrp
)
10109 * The IFF_NOACCEPT flag can only be set on an IFF_VRRP IP address by
10110 * VRRP control socket.
10112 if ((turn_off
| turn_on
) & IFF_NOACCEPT
) {
10113 if (!connp
->conn_isvrrp
|| !(intf_flags
& IFF_VRRP
))
10117 if (turn_on
& IFF_NOFAILOVER
) {
10118 turn_on
|= IFF_DEPRECATED
;
10119 flags
|= IFF_DEPRECATED
;
10123 * On underlying interfaces, only allow applications to manage test
10124 * addresses -- otherwise, they may get confused when the address
10125 * moves as part of being brought up. Likewise, prevent an
10126 * application-managed test address from being converted to a data
10127 * address. To prevent migration of administratively up addresses in
10128 * the kernel, we don't allow them to be converted either.
10130 if (IS_UNDER_IPMP(ill
)) {
10131 const uint64_t appflags
= IFF_DHCPRUNNING
| IFF_ADDRCONF
;
10133 if ((turn_on
& appflags
) && !(flags
& IFF_NOFAILOVER
))
10136 if ((turn_off
& IFF_NOFAILOVER
) &&
10137 (flags
& (appflags
| IFF_UP
| IFF_DUPLICATE
)))
10142 * Only allow IFF_TEMPORARY flag to be set on
10145 if ((turn_on
& IFF_TEMPORARY
) && !(ipif
->ipif_isv6
))
10149 * cannot turn off IFF_NOXMIT on VNI interfaces.
10151 if ((turn_off
& IFF_NOXMIT
) && IS_VNI(ipif
->ipif_ill
))
10155 * Don't allow the IFF_ROUTER flag to be turned on on loopback
10156 * interfaces. It makes no sense in that context.
10158 if ((turn_on
& IFF_ROUTER
) && (phyi
->phyint_flags
& PHYI_LOOPBACK
))
10162 * For IPv6 ipif_id 0, don't allow the interface to be up without
10163 * a link local address if IFF_NOLOCAL or IFF_ANYCAST are not set.
10164 * If the link local address isn't set, and can be set, it will get
10165 * set later on in this function.
10167 if (ipif
->ipif_id
== 0 && ipif
->ipif_isv6
&&
10168 (flags
& IFF_UP
) && !(flags
& (IFF_NOLOCAL
|IFF_ANYCAST
)) &&
10169 IN6_IS_ADDR_UNSPECIFIED(&ipif
->ipif_v6lcl_addr
)) {
10170 if (ipif_cant_setlinklocal(ipif
))
10172 set_linklocal
= B_TRUE
;
10176 * If we modify physical interface flags, we'll potentially need to
10177 * send up two routing socket messages for the changes (one for the
10178 * IPv4 ill, and another for the IPv6 ill). Note that here.
10180 if ((turn_on
|turn_off
) & IFF_PHYINT_FLAGS
)
10181 phyint_flags_modified
= B_TRUE
;
10184 * All functioning PHYI_STANDBY interfaces start life PHYI_INACTIVE
10185 * (otherwise, we'd immediately use them, defeating standby). Also,
10186 * since PHYI_INACTIVE has a separate meaning when PHYI_STANDBY is not
10187 * set, don't allow PHYI_STANDBY to be set if PHYI_INACTIVE is already
10188 * set, and clear PHYI_INACTIVE if PHYI_STANDBY is being cleared. We
10189 * also don't allow PHYI_STANDBY if VNI is enabled since its semantics
10190 * will not be honored.
10192 if (turn_on
& PHYI_STANDBY
) {
10194 * No need to grab ill_g_usesrc_lock here; see the
10195 * synchronization notes in ip.c.
10197 if (ill
->ill_usesrc_grp_next
!= NULL
||
10198 intf_flags
& PHYI_INACTIVE
)
10200 if (!(flags
& PHYI_FAILED
)) {
10201 flags
|= PHYI_INACTIVE
;
10202 turn_on
|= PHYI_INACTIVE
;
10206 if (turn_off
& PHYI_STANDBY
) {
10207 flags
&= ~PHYI_INACTIVE
;
10208 turn_off
|= PHYI_INACTIVE
;
10212 * PHYI_FAILED and PHYI_INACTIVE are mutually exclusive; fail if both
10215 if ((flags
& (PHYI_FAILED
| PHYI_INACTIVE
)) ==
10216 (PHYI_FAILED
| PHYI_INACTIVE
))
10220 * If ILLF_ROUTER changes, we need to change the ip forwarding
10221 * status of the interface.
10223 if ((turn_on
| turn_off
) & ILLF_ROUTER
) {
10224 err
= ill_forward_set(ill
, ((turn_on
& ILLF_ROUTER
) != 0));
10230 * If the interface is not UP and we are not going to
10231 * bring it UP, record the flags and return. When the
10232 * interface comes UP later, the right actions will be
10235 if (!(ipif
->ipif_flags
& IPIF_UP
) &&
10236 !(turn_on
& IPIF_UP
)) {
10237 /* Record new flags in their respective places. */
10238 mutex_enter(&ill
->ill_lock
);
10239 mutex_enter(&ill
->ill_phyint
->phyint_lock
);
10240 ipif
->ipif_flags
|= (turn_on
& IFF_LOGINT_FLAGS
);
10241 ipif
->ipif_flags
&= (~turn_off
& IFF_LOGINT_FLAGS
);
10242 ill
->ill_flags
|= (turn_on
& IFF_PHYINTINST_FLAGS
);
10243 ill
->ill_flags
&= (~turn_off
& IFF_PHYINTINST_FLAGS
);
10244 phyi
->phyint_flags
|= (turn_on
& IFF_PHYINT_FLAGS
);
10245 phyi
->phyint_flags
&= (~turn_off
& IFF_PHYINT_FLAGS
);
10246 mutex_exit(&ill
->ill_lock
);
10247 mutex_exit(&ill
->ill_phyint
->phyint_lock
);
10250 * PHYI_FAILED, PHYI_INACTIVE, and PHYI_OFFLINE are all the
10251 * same to the kernel: if any of them has been set by
10252 * userland, the interface cannot be used for data traffic.
10254 if ((turn_on
|turn_off
) &
10255 (PHYI_FAILED
| PHYI_INACTIVE
| PHYI_OFFLINE
)) {
10256 ASSERT(!IS_IPMP(ill
));
10258 * It's possible the ill is part of an "anonymous"
10259 * IPMP group rather than a real group. In that case,
10260 * there are no other interfaces in the group and thus
10261 * no need to call ipmp_phyint_refresh_active().
10263 if (IS_UNDER_IPMP(ill
))
10264 ipmp_phyint_refresh_active(phyi
);
10267 if (phyint_flags_modified
) {
10268 if (phyi
->phyint_illv4
!= NULL
) {
10269 ip_rts_ifmsg(phyi
->phyint_illv4
->
10270 ill_ipif
, RTSQ_DEFAULT
);
10272 if (phyi
->phyint_illv6
!= NULL
) {
10273 ip_rts_ifmsg(phyi
->phyint_illv6
->
10274 ill_ipif
, RTSQ_DEFAULT
);
10277 /* The default multicast interface might have changed */
10278 ire_increment_multicast_generation(ill
->ill_ipst
,
10282 } else if (set_linklocal
) {
10283 mutex_enter(&ill
->ill_lock
);
10285 ipif
->ipif_state_flags
|= IPIF_SET_LINKLOCAL
;
10286 mutex_exit(&ill
->ill_lock
);
10290 * Disallow IPv6 interfaces coming up that have the unspecified address,
10291 * or point-to-point interfaces with an unspecified destination. We do
10292 * allow the address to be unspecified for IPIF_NOLOCAL interfaces that
10293 * have a subnet assigned, which is how in.ndpd currently manages its
10294 * onlink prefix list when no addresses are configured with those
10297 if (ipif
->ipif_isv6
&&
10298 ((IN6_IS_ADDR_UNSPECIFIED(&ipif
->ipif_v6lcl_addr
) &&
10299 (!(ipif
->ipif_flags
& IPIF_NOLOCAL
) && !(turn_on
& IPIF_NOLOCAL
) ||
10300 IN6_IS_ADDR_UNSPECIFIED(&ipif
->ipif_v6subnet
))) ||
10301 ((ipif
->ipif_flags
& IPIF_POINTOPOINT
) &&
10302 IN6_IS_ADDR_UNSPECIFIED(&ipif
->ipif_v6pp_dst_addr
)))) {
10307 * Prevent IPv4 point-to-point interfaces with a 0.0.0.0 destination
10308 * from being brought up.
10310 if (!ipif
->ipif_isv6
&&
10311 ((ipif
->ipif_flags
& IPIF_POINTOPOINT
) &&
10312 ipif
->ipif_pp_dst_addr
== INADDR_ANY
)) {
10317 * If we are going to change one or more of the flags that are
10318 * IPIF_UP, IPIF_DEPRECATED, IPIF_NOXMIT, IPIF_NOLOCAL, ILLF_NOARP,
10319 * ILLF_NONUD, IPIF_PRIVATE, IPIF_ANYCAST, IPIF_PREFERRED, and
10320 * IPIF_NOFAILOVER, we will take special action. This is
10321 * done by bring the ipif down, changing the flags and bringing
10322 * it back up again. For IPIF_NOFAILOVER, the act of bringing it
10323 * back up will trigger the address to be moved.
10325 * If we are going to change IFF_NOACCEPT, we need to bring
10326 * all the ipifs down then bring them up again. The act of
10327 * bringing all the ipifs back up will trigger the local
10328 * ires being recreated with "no_accept" set/cleared.
10330 * Note that ILLF_NOACCEPT is always set separately from the
10333 if ((turn_on
|turn_off
) &
10334 (IPIF_UP
|IPIF_DEPRECATED
|IPIF_NOXMIT
|IPIF_NOLOCAL
|ILLF_NOARP
|
10335 ILLF_NONUD
|IPIF_PRIVATE
|IPIF_ANYCAST
|IPIF_PREFERRED
|
10336 IPIF_NOFAILOVER
)) {
10338 * ipif_down() will ire_delete bcast ire's for the subnet,
10339 * while the ire_identical_ref tracks the case of IRE_BROADCAST
10340 * entries shared between multiple ipifs on the same subnet.
10342 if (((ipif
->ipif_flags
| turn_on
) & IPIF_UP
) &&
10343 !(turn_off
& IPIF_UP
)) {
10344 if (ipif
->ipif_flags
& IPIF_UP
)
10345 ill
->ill_logical_down
= 1;
10346 turn_on
&= ~IPIF_UP
;
10348 err
= ipif_down(ipif
, q
, mp
);
10349 ip1dbg(("ipif_down returns %d err ", err
));
10350 if (err
== EINPROGRESS
)
10352 (void) ipif_down_tail(ipif
);
10353 } else if ((turn_on
|turn_off
) & ILLF_NOACCEPT
) {
10355 * If we can quiesce the ill, then continue. If not, then
10356 * ip_sioctl_flags_tail() will be called from
10357 * ipif_ill_refrele_tail().
10359 ill_down_ipifs(ill
, B_TRUE
);
10361 mutex_enter(&connp
->conn_lock
);
10362 mutex_enter(&ill
->ill_lock
);
10363 if (!ill_is_quiescent(ill
)) {
10366 success
= ipsq_pending_mp_add(connp
, ill
->ill_ipif
,
10368 mutex_exit(&ill
->ill_lock
);
10369 mutex_exit(&connp
->conn_lock
);
10370 return (success
? EINPROGRESS
: EINTR
);
10372 mutex_exit(&ill
->ill_lock
);
10373 mutex_exit(&connp
->conn_lock
);
10375 return (ip_sioctl_flags_tail(ipif
, flags
, q
, mp
));
10379 ip_sioctl_flags_tail(ipif_t
*ipif
, uint64_t flags
, queue_t
*q
, mblk_t
*mp
)
10383 uint64_t turn_on
, turn_off
;
10384 boolean_t phyint_flags_modified
= B_FALSE
;
10386 boolean_t set_linklocal
= B_FALSE
;
10388 ip1dbg(("ip_sioctl_flags_tail(%s:%u)\n",
10389 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
));
10391 ASSERT(IAM_WRITER_IPIF(ipif
));
10393 ill
= ipif
->ipif_ill
;
10394 phyi
= ill
->ill_phyint
;
10396 ip_sioctl_flags_onoff(ipif
, flags
, &turn_on
, &turn_off
);
10399 * IFF_UP is handled separately.
10401 turn_on
&= ~IFF_UP
;
10402 turn_off
&= ~IFF_UP
;
10404 if ((turn_on
|turn_off
) & IFF_PHYINT_FLAGS
)
10405 phyint_flags_modified
= B_TRUE
;
10408 * Now we change the flags. Track current value of
10409 * other flags in their respective places.
10411 mutex_enter(&ill
->ill_lock
);
10412 mutex_enter(&phyi
->phyint_lock
);
10413 ipif
->ipif_flags
|= (turn_on
& IFF_LOGINT_FLAGS
);
10414 ipif
->ipif_flags
&= (~turn_off
& IFF_LOGINT_FLAGS
);
10415 ill
->ill_flags
|= (turn_on
& IFF_PHYINTINST_FLAGS
);
10416 ill
->ill_flags
&= (~turn_off
& IFF_PHYINTINST_FLAGS
);
10417 phyi
->phyint_flags
|= (turn_on
& IFF_PHYINT_FLAGS
);
10418 phyi
->phyint_flags
&= (~turn_off
& IFF_PHYINT_FLAGS
);
10419 if (ipif
->ipif_state_flags
& IPIF_SET_LINKLOCAL
) {
10420 set_linklocal
= B_TRUE
;
10421 ipif
->ipif_state_flags
&= ~IPIF_SET_LINKLOCAL
;
10424 mutex_exit(&ill
->ill_lock
);
10425 mutex_exit(&phyi
->phyint_lock
);
10428 (void) ipif_setlinklocal(ipif
);
10431 * PHYI_FAILED, PHYI_INACTIVE, and PHYI_OFFLINE are all the same to
10432 * the kernel: if any of them has been set by userland, the interface
10433 * cannot be used for data traffic.
10435 if ((turn_on
|turn_off
) & (PHYI_FAILED
| PHYI_INACTIVE
| PHYI_OFFLINE
)) {
10436 ASSERT(!IS_IPMP(ill
));
10438 * It's possible the ill is part of an "anonymous" IPMP group
10439 * rather than a real group. In that case, there are no other
10440 * interfaces in the group and thus no need for us to call
10441 * ipmp_phyint_refresh_active().
10443 if (IS_UNDER_IPMP(ill
))
10444 ipmp_phyint_refresh_active(phyi
);
10447 if ((turn_on
|turn_off
) & ILLF_NOACCEPT
) {
10449 * If the ILLF_NOACCEPT flag is changed, bring up all the
10450 * ipifs that were brought down.
10452 * The routing sockets messages are sent as the result
10453 * of ill_up_ipifs(), further, SCTP's IPIF list was updated
10456 err
= ill_up_ipifs(ill
, q
, mp
);
10457 } else if ((flags
& IFF_UP
) && !(ipif
->ipif_flags
& IPIF_UP
)) {
10459 * XXX ipif_up really does not know whether a phyint flags
10460 * was modified or not. So, it sends up information on
10461 * only one routing sockets message. As we don't bring up
10462 * the interface and also set PHYI_ flags simultaneously
10463 * it should be okay.
10465 err
= ipif_up(ipif
, q
, mp
);
10468 * Make sure routing socket sees all changes to the flags.
10469 * ipif_up_done* handles this when we use ipif_up.
10471 if (phyint_flags_modified
) {
10472 if (phyi
->phyint_illv4
!= NULL
) {
10473 ip_rts_ifmsg(phyi
->phyint_illv4
->
10474 ill_ipif
, RTSQ_DEFAULT
);
10476 if (phyi
->phyint_illv6
!= NULL
) {
10477 ip_rts_ifmsg(phyi
->phyint_illv6
->
10478 ill_ipif
, RTSQ_DEFAULT
);
10481 ip_rts_ifmsg(ipif
, RTSQ_DEFAULT
);
10484 * Update the flags in SCTP's IPIF list, ipif_up() will do
10485 * this in need_up case.
10487 sctp_update_ipif(ipif
, SCTP_IPIF_UPDATE
);
10490 /* The default multicast interface might have changed */
10491 ire_increment_multicast_generation(ill
->ill_ipst
, ill
->ill_isv6
);
10496 * Restart the flags operation now that the refcounts have dropped to zero.
10500 ip_sioctl_flags_restart(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
10501 ip_ioctl_cmd_t
*ipip
, void *if_req
)
10504 struct ifreq
*ifr
= if_req
;
10505 struct lifreq
*lifr
= if_req
;
10506 uint64_t turn_on
, turn_off
;
10508 ip1dbg(("ip_sioctl_flags_restart(%s:%u %p)\n",
10509 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
10511 if (ipip
->ipi_cmd_type
== IF_CMD
) {
10512 /* cast to uint16_t prevents unwanted sign extension */
10513 flags
= (uint16_t)ifr
->ifr_flags
;
10515 flags
= lifr
->lifr_flags
;
10519 * If this function call is a result of the ILLF_NOACCEPT flag
10520 * change, do not call ipif_down_tail(). See ip_sioctl_flags().
10522 ip_sioctl_flags_onoff(ipif
, flags
, &turn_on
, &turn_off
);
10523 if (!((turn_on
|turn_off
) & ILLF_NOACCEPT
))
10524 (void) ipif_down_tail(ipif
);
10526 return (ip_sioctl_flags_tail(ipif
, flags
, q
, mp
));
10530 * Can operate on either a module or a driver queue.
10534 ip_sioctl_get_flags(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
10535 ip_ioctl_cmd_t
*ipip
, void *if_req
)
10538 * Has the flags been set correctly till now ?
10540 ill_t
*ill
= ipif
->ipif_ill
;
10541 phyint_t
*phyi
= ill
->ill_phyint
;
10543 ip1dbg(("ip_sioctl_get_flags(%s:%u %p)\n",
10544 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
10545 ASSERT((phyi
->phyint_flags
& ~(IFF_PHYINT_FLAGS
)) == 0);
10546 ASSERT((ill
->ill_flags
& ~(IFF_PHYINTINST_FLAGS
)) == 0);
10547 ASSERT((ipif
->ipif_flags
& ~(IFF_LOGINT_FLAGS
)) == 0);
10550 * Need a lock since some flags can be set even when there are
10551 * references to the ipif.
10553 mutex_enter(&ill
->ill_lock
);
10554 if (ipip
->ipi_cmd_type
== IF_CMD
) {
10555 struct ifreq
*ifr
= (struct ifreq
*)if_req
;
10557 /* Get interface flags (low 16 only). */
10558 ifr
->ifr_flags
= ((ipif
->ipif_flags
|
10559 ill
->ill_flags
| phyi
->phyint_flags
) & 0xffff);
10561 struct lifreq
*lifr
= (struct lifreq
*)if_req
;
10563 /* Get interface flags. */
10564 lifr
->lifr_flags
= ipif
->ipif_flags
|
10565 ill
->ill_flags
| phyi
->phyint_flags
;
10567 mutex_exit(&ill
->ill_lock
);
10572 * We allow the MTU to be set on an ILL, but not have it be different
10573 * for different IPIFs since we don't actually send packets on IPIFs.
10577 ip_sioctl_mtu(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
10578 ip_ioctl_cmd_t
*ipip
, void *if_req
)
10583 struct lifreq
*lifr
;
10586 ip1dbg(("ip_sioctl_mtu(%s:%u %p)\n", ipif
->ipif_ill
->ill_name
,
10587 ipif
->ipif_id
, (void *)ipif
));
10588 if (ipip
->ipi_cmd_type
== IF_CMD
) {
10589 ifr
= (struct ifreq
*)if_req
;
10590 mtu
= ifr
->ifr_metric
;
10592 lifr
= (struct lifreq
*)if_req
;
10593 mtu
= lifr
->lifr_mtu
;
10595 /* Only allow for logical unit zero i.e. not on "bge0:17" */
10596 if (ipif
->ipif_id
!= 0)
10599 ill
= ipif
->ipif_ill
;
10600 if (ipif
->ipif_isv6
)
10601 ip_min_mtu
= IPV6_MIN_MTU
;
10603 ip_min_mtu
= IP_MIN_MTU
;
10605 mutex_enter(&ill
->ill_lock
);
10606 if (mtu
> ill
->ill_max_frag
|| mtu
< ip_min_mtu
) {
10607 mutex_exit(&ill
->ill_lock
);
10610 /* Avoid increasing ill_mc_mtu */
10611 if (ill
->ill_mc_mtu
> mtu
)
10612 ill
->ill_mc_mtu
= mtu
;
10615 * The dce and fragmentation code can handle changes to ill_mtu
10616 * concurrent with sending/fragmenting packets.
10618 ill
->ill_mtu
= mtu
;
10619 ill
->ill_flags
|= ILLF_FIXEDMTU
;
10620 mutex_exit(&ill
->ill_lock
);
10623 * Make sure all dce_generation checks find out
10624 * that ill_mtu/ill_mc_mtu has changed.
10626 dce_increment_all_generations(ill
->ill_isv6
, ill
->ill_ipst
);
10629 * Refresh IPMP meta-interface MTU if necessary.
10631 if (IS_UNDER_IPMP(ill
))
10632 ipmp_illgrp_refresh_mtu(ill
->ill_grp
);
10634 /* Update the MTU in SCTP's list */
10635 sctp_update_ipif(ipif
, SCTP_IPIF_UPDATE
);
10639 /* Get interface MTU. */
10642 ip_sioctl_get_mtu(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
10643 ip_ioctl_cmd_t
*ipip
, void *if_req
)
10646 struct lifreq
*lifr
;
10648 ip1dbg(("ip_sioctl_get_mtu(%s:%u %p)\n",
10649 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
10652 * We allow a get on any logical interface even though the set
10653 * can only be done on logical unit 0.
10655 if (ipip
->ipi_cmd_type
== IF_CMD
) {
10656 ifr
= (struct ifreq
*)if_req
;
10657 ifr
->ifr_metric
= ipif
->ipif_ill
->ill_mtu
;
10659 lifr
= (struct lifreq
*)if_req
;
10660 lifr
->lifr_mtu
= ipif
->ipif_ill
->ill_mtu
;
10665 /* Set interface broadcast address. */
10668 ip_sioctl_brdaddr(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
10669 ip_ioctl_cmd_t
*ipip
, void *if_req
)
10673 ill_t
*ill
= ipif
->ipif_ill
;
10674 ip_stack_t
*ipst
= ill
->ill_ipst
;
10676 ip1dbg(("ip_sioctl_brdaddr(%s:%u)\n", ill
->ill_name
,
10679 ASSERT(IAM_WRITER_IPIF(ipif
));
10680 if (!(ipif
->ipif_flags
& IPIF_BROADCAST
))
10681 return (EADDRNOTAVAIL
);
10683 ASSERT(!(ipif
->ipif_isv6
)); /* No IPv6 broadcast */
10685 if (sin
->sin_family
!= AF_INET
)
10686 return (EAFNOSUPPORT
);
10688 addr
= sin
->sin_addr
.s_addr
;
10690 if (ipif
->ipif_flags
& IPIF_UP
) {
10692 * If we are already up, make sure the new
10693 * broadcast address makes sense. If it does,
10694 * there should be an IRE for it already.
10696 ire
= ire_ftable_lookup_v4(addr
, 0, 0, IRE_BROADCAST
,
10697 ill
, ipif
->ipif_zoneid
,
10698 (MATCH_IRE_ILL
| MATCH_IRE_TYPE
), 0, ipst
, NULL
);
10706 * Changing the broadcast addr for this ipif. Since the IRE_BROADCAST
10707 * needs to already exist we never need to change the set of
10708 * IRE_BROADCASTs when we are UP.
10710 if (addr
!= ipif
->ipif_brd_addr
)
10711 IN6_IPADDR_TO_V4MAPPED(addr
, &ipif
->ipif_v6brd_addr
);
10716 /* Get interface broadcast address. */
10719 ip_sioctl_get_brdaddr(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
10720 ip_ioctl_cmd_t
*ipip
, void *if_req
)
10722 ip1dbg(("ip_sioctl_get_brdaddr(%s:%u %p)\n",
10723 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
10724 if (!(ipif
->ipif_flags
& IPIF_BROADCAST
))
10725 return (EADDRNOTAVAIL
);
10727 /* IPIF_BROADCAST not possible with IPv6 */
10728 ASSERT(!ipif
->ipif_isv6
);
10730 sin
->sin_family
= AF_INET
;
10731 sin
->sin_addr
.s_addr
= ipif
->ipif_brd_addr
;
10736 * This routine is called to handle the SIOCS*IFNETMASK IOCTL.
10740 ip_sioctl_netmask(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
10741 ip_ioctl_cmd_t
*ipip
, void *if_req
)
10746 ip1dbg(("ip_sioctl_netmask(%s:%u %p)\n",
10747 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
10749 ASSERT(IAM_WRITER_IPIF(ipif
));
10751 if (ipif
->ipif_isv6
) {
10754 if (sin
->sin_family
!= AF_INET6
)
10755 return (EAFNOSUPPORT
);
10757 sin6
= (sin6_t
*)sin
;
10758 v6mask
= sin6
->sin6_addr
;
10762 if (sin
->sin_family
!= AF_INET
)
10763 return (EAFNOSUPPORT
);
10765 mask
= sin
->sin_addr
.s_addr
;
10766 if (!ip_contiguous_mask(ntohl(mask
)))
10768 V4MASK_TO_V6(mask
, v6mask
);
10772 * No big deal if the interface isn't already up, or the mask
10773 * isn't really changing, or this is pt-pt.
10775 if (!(ipif
->ipif_flags
& IPIF_UP
) ||
10776 IN6_ARE_ADDR_EQUAL(&v6mask
, &ipif
->ipif_v6net_mask
) ||
10777 (ipif
->ipif_flags
& IPIF_POINTOPOINT
)) {
10778 ipif
->ipif_v6net_mask
= v6mask
;
10779 if ((ipif
->ipif_flags
& IPIF_POINTOPOINT
) == 0) {
10780 V6_MASK_COPY(ipif
->ipif_v6lcl_addr
,
10781 ipif
->ipif_v6net_mask
,
10782 ipif
->ipif_v6subnet
);
10787 * Make sure we have valid net and subnet broadcast ire's
10788 * for the old netmask, if needed by other logical interfaces.
10790 err
= ipif_logical_down(ipif
, q
, mp
);
10791 if (err
== EINPROGRESS
)
10793 (void) ipif_down_tail(ipif
);
10794 err
= ip_sioctl_netmask_tail(ipif
, sin
, q
, mp
);
10799 ip_sioctl_netmask_tail(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
)
10804 ip1dbg(("ip_sioctl_netmask_tail(%s:%u %p)\n",
10805 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
10807 if (ipif
->ipif_isv6
) {
10810 sin6
= (sin6_t
*)sin
;
10811 v6mask
= sin6
->sin6_addr
;
10815 mask
= sin
->sin_addr
.s_addr
;
10816 V4MASK_TO_V6(mask
, v6mask
);
10819 ipif
->ipif_v6net_mask
= v6mask
;
10820 if ((ipif
->ipif_flags
& IPIF_POINTOPOINT
) == 0) {
10821 V6_MASK_COPY(ipif
->ipif_v6lcl_addr
, ipif
->ipif_v6net_mask
,
10822 ipif
->ipif_v6subnet
);
10824 err
= ipif_up(ipif
, q
, mp
);
10826 if (err
== 0 || err
== EINPROGRESS
) {
10828 * The interface must be DL_BOUND if this packet has to
10829 * go out on the wire. Since we only go through a logical
10830 * down and are bound with the driver during an internal
10831 * down/up that is satisfied.
10833 if (!ipif
->ipif_isv6
&& ipif
->ipif_ill
->ill_wq
!= NULL
) {
10834 /* Potentially broadcast an address mask reply. */
10835 ipif_mask_reply(ipif
);
10843 ip_sioctl_netmask_restart(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
10844 ip_ioctl_cmd_t
*ipip
, void *if_req
)
10846 ip1dbg(("ip_sioctl_netmask_restart(%s:%u %p)\n",
10847 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
10848 (void) ipif_down_tail(ipif
);
10849 return (ip_sioctl_netmask_tail(ipif
, sin
, q
, mp
));
10852 /* Get interface net mask. */
10855 ip_sioctl_get_netmask(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
10856 ip_ioctl_cmd_t
*ipip
, void *if_req
)
10858 struct lifreq
*lifr
= (struct lifreq
*)if_req
;
10859 struct sockaddr_in6
*sin6
= (sin6_t
*)sin
;
10861 ip1dbg(("ip_sioctl_get_netmask(%s:%u %p)\n",
10862 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
10865 * net mask can't change since we have a reference to the ipif.
10867 if (ipif
->ipif_isv6
) {
10868 ASSERT(ipip
->ipi_cmd_type
== LIF_CMD
);
10870 sin6
->sin6_family
= AF_INET6
;
10871 sin6
->sin6_addr
= ipif
->ipif_v6net_mask
;
10872 lifr
->lifr_addrlen
=
10873 ip_mask_to_plen_v6(&ipif
->ipif_v6net_mask
);
10876 sin
->sin_family
= AF_INET
;
10877 sin
->sin_addr
.s_addr
= ipif
->ipif_net_mask
;
10878 if (ipip
->ipi_cmd_type
== LIF_CMD
) {
10879 lifr
->lifr_addrlen
=
10880 ip_mask_to_plen(ipif
->ipif_net_mask
);
10888 ip_sioctl_metric(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
10889 ip_ioctl_cmd_t
*ipip
, void *if_req
)
10891 ip1dbg(("ip_sioctl_metric(%s:%u %p)\n",
10892 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
10895 * Since no applications should ever be setting metrics on underlying
10896 * interfaces, we explicitly fail to smoke 'em out.
10898 if (IS_UNDER_IPMP(ipif
->ipif_ill
))
10902 * Set interface metric. We don't use this for
10903 * anything but we keep track of it in case it is
10904 * important to routing applications or such.
10906 if (ipip
->ipi_cmd_type
== IF_CMD
) {
10909 ifr
= (struct ifreq
*)if_req
;
10910 ipif
->ipif_ill
->ill_metric
= ifr
->ifr_metric
;
10912 struct lifreq
*lifr
;
10914 lifr
= (struct lifreq
*)if_req
;
10915 ipif
->ipif_ill
->ill_metric
= lifr
->lifr_metric
;
10922 ip_sioctl_get_metric(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
10923 ip_ioctl_cmd_t
*ipip
, void *if_req
)
10925 /* Get interface metric. */
10926 ip1dbg(("ip_sioctl_get_metric(%s:%u %p)\n",
10927 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
10929 if (ipip
->ipi_cmd_type
== IF_CMD
) {
10932 ifr
= (struct ifreq
*)if_req
;
10933 ifr
->ifr_metric
= ipif
->ipif_ill
->ill_metric
;
10935 struct lifreq
*lifr
;
10937 lifr
= (struct lifreq
*)if_req
;
10938 lifr
->lifr_metric
= ipif
->ipif_ill
->ill_metric
;
10946 ip_sioctl_muxid(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
10947 ip_ioctl_cmd_t
*ipip
, void *if_req
)
10951 ip1dbg(("ip_sioctl_muxid(%s:%u %p)\n",
10952 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
10954 * Set the muxid returned from I_PLINK.
10956 if (ipip
->ipi_cmd_type
== IF_CMD
) {
10957 struct ifreq
*ifr
= (struct ifreq
*)if_req
;
10959 ipif
->ipif_ill
->ill_muxid
= ifr
->ifr_ip_muxid
;
10960 arp_muxid
= ifr
->ifr_arp_muxid
;
10962 struct lifreq
*lifr
= (struct lifreq
*)if_req
;
10964 ipif
->ipif_ill
->ill_muxid
= lifr
->lifr_ip_muxid
;
10965 arp_muxid
= lifr
->lifr_arp_muxid
;
10967 arl_set_muxid(ipif
->ipif_ill
, arp_muxid
);
10973 ip_sioctl_get_muxid(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
10974 ip_ioctl_cmd_t
*ipip
, void *if_req
)
10978 ip1dbg(("ip_sioctl_get_muxid(%s:%u %p)\n",
10979 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
10981 * Get the muxid saved in ill for I_PUNLINK.
10983 arp_muxid
= arl_get_muxid(ipif
->ipif_ill
);
10984 if (ipip
->ipi_cmd_type
== IF_CMD
) {
10985 struct ifreq
*ifr
= (struct ifreq
*)if_req
;
10987 ifr
->ifr_ip_muxid
= ipif
->ipif_ill
->ill_muxid
;
10988 ifr
->ifr_arp_muxid
= arp_muxid
;
10990 struct lifreq
*lifr
= (struct lifreq
*)if_req
;
10992 lifr
->lifr_ip_muxid
= ipif
->ipif_ill
->ill_muxid
;
10993 lifr
->lifr_arp_muxid
= arp_muxid
;
10999 * Set the subnet prefix. Does not modify the broadcast address.
11003 ip_sioctl_subnet(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
11004 ip_ioctl_cmd_t
*ipip
, void *if_req
)
11009 boolean_t need_up
= B_FALSE
;
11012 ip1dbg(("ip_sioctl_subnet(%s:%u %p)\n",
11013 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
11015 ASSERT(IAM_WRITER_IPIF(ipif
));
11016 addrlen
= ((struct lifreq
*)if_req
)->lifr_addrlen
;
11018 if (ipif
->ipif_isv6
) {
11021 if (sin
->sin_family
!= AF_INET6
)
11022 return (EAFNOSUPPORT
);
11024 sin6
= (sin6_t
*)sin
;
11025 v6addr
= sin6
->sin6_addr
;
11026 if (!ip_remote_addr_ok_v6(&v6addr
, &ipv6_all_ones
))
11027 return (EADDRNOTAVAIL
);
11031 if (sin
->sin_family
!= AF_INET
)
11032 return (EAFNOSUPPORT
);
11034 addr
= sin
->sin_addr
.s_addr
;
11035 if (!ip_addr_ok_v4(addr
, 0xFFFFFFFF))
11036 return (EADDRNOTAVAIL
);
11037 IN6_IPADDR_TO_V4MAPPED(addr
, &v6addr
);
11039 addrlen
+= IPV6_ABITS
- IP_ABITS
;
11042 if (ip_plen_to_mask_v6(addrlen
, &v6mask
) == NULL
)
11045 /* Check if bits in the address is set past the mask */
11046 if (!V6_MASK_EQ(v6addr
, v6mask
, v6addr
))
11049 if (IN6_ARE_ADDR_EQUAL(&ipif
->ipif_v6subnet
, &v6addr
) &&
11050 IN6_ARE_ADDR_EQUAL(&ipif
->ipif_v6net_mask
, &v6mask
))
11051 return (0); /* No change */
11053 if (ipif
->ipif_flags
& IPIF_UP
) {
11055 * If the interface is already marked up,
11056 * we call ipif_down which will take care
11057 * of ditching any IREs that have been set
11058 * up based on the old interface address.
11060 err
= ipif_logical_down(ipif
, q
, mp
);
11061 if (err
== EINPROGRESS
)
11063 (void) ipif_down_tail(ipif
);
11067 err
= ip_sioctl_subnet_tail(ipif
, v6addr
, v6mask
, q
, mp
, need_up
);
11072 ip_sioctl_subnet_tail(ipif_t
*ipif
, in6_addr_t v6addr
, in6_addr_t v6mask
,
11073 queue_t
*q
, mblk_t
*mp
, boolean_t need_up
)
11075 ill_t
*ill
= ipif
->ipif_ill
;
11078 ip1dbg(("ip_sioctl_subnet_tail(%s:%u %p)\n",
11079 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
11081 /* Set the new address. */
11082 mutex_enter(&ill
->ill_lock
);
11083 ipif
->ipif_v6net_mask
= v6mask
;
11084 if ((ipif
->ipif_flags
& IPIF_POINTOPOINT
) == 0) {
11085 V6_MASK_COPY(v6addr
, ipif
->ipif_v6net_mask
,
11086 ipif
->ipif_v6subnet
);
11088 mutex_exit(&ill
->ill_lock
);
11092 * Now bring the interface back up. If this
11093 * is the only IPIF for the ILL, ipif_up
11094 * will have to re-bind to the device, so
11095 * we may get back EINPROGRESS, in which
11096 * case, this IOCTL will get completed in
11097 * ip_rput_dlpi when we see the DL_BIND_ACK.
11099 err
= ipif_up(ipif
, q
, mp
);
11100 if (err
== EINPROGRESS
)
11108 ip_sioctl_subnet_restart(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
11109 ip_ioctl_cmd_t
*ipip
, void *if_req
)
11114 struct lifreq
*lifr
= (struct lifreq
*)if_req
;
11116 ip1dbg(("ip_sioctl_subnet_restart(%s:%u %p)\n",
11117 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
11118 (void) ipif_down_tail(ipif
);
11120 addrlen
= lifr
->lifr_addrlen
;
11121 if (ipif
->ipif_isv6
) {
11124 sin6
= (sin6_t
*)sin
;
11125 v6addr
= sin6
->sin6_addr
;
11129 addr
= sin
->sin_addr
.s_addr
;
11130 IN6_IPADDR_TO_V4MAPPED(addr
, &v6addr
);
11131 addrlen
+= IPV6_ABITS
- IP_ABITS
;
11133 (void) ip_plen_to_mask_v6(addrlen
, &v6mask
);
11135 return (ip_sioctl_subnet_tail(ipif
, v6addr
, v6mask
, q
, mp
, B_TRUE
));
11140 ip_sioctl_get_subnet(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
11141 ip_ioctl_cmd_t
*ipip
, void *if_req
)
11143 struct lifreq
*lifr
= (struct lifreq
*)if_req
;
11144 struct sockaddr_in6
*sin6
= (struct sockaddr_in6
*)sin
;
11146 ip1dbg(("ip_sioctl_get_subnet(%s:%u %p)\n",
11147 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
11148 ASSERT(ipip
->ipi_cmd_type
== LIF_CMD
);
11150 if (ipif
->ipif_isv6
) {
11152 sin6
->sin6_family
= AF_INET6
;
11153 sin6
->sin6_addr
= ipif
->ipif_v6subnet
;
11154 lifr
->lifr_addrlen
=
11155 ip_mask_to_plen_v6(&ipif
->ipif_v6net_mask
);
11158 sin
->sin_family
= AF_INET
;
11159 sin
->sin_addr
.s_addr
= ipif
->ipif_subnet
;
11160 lifr
->lifr_addrlen
= ip_mask_to_plen(ipif
->ipif_net_mask
);
11166 * Set the IPv6 address token.
11170 ip_sioctl_token(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
11171 ip_ioctl_cmd_t
*ipi
, void *if_req
)
11173 ill_t
*ill
= ipif
->ipif_ill
;
11177 boolean_t need_up
= B_FALSE
;
11179 sin6_t
*sin6
= (sin6_t
*)sin
;
11180 struct lifreq
*lifr
= (struct lifreq
*)if_req
;
11183 ip1dbg(("ip_sioctl_token(%s:%u %p)\n",
11184 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
11185 ASSERT(IAM_WRITER_IPIF(ipif
));
11187 addrlen
= lifr
->lifr_addrlen
;
11188 /* Only allow for logical unit zero i.e. not on "le0:17" */
11189 if (ipif
->ipif_id
!= 0)
11192 if (!ipif
->ipif_isv6
)
11195 if (addrlen
> IPV6_ABITS
)
11198 v6addr
= sin6
->sin6_addr
;
11201 * The length of the token is the length from the end. To get
11202 * the proper mask for this, compute the mask of the bits not
11203 * in the token; ie. the prefix, and then xor to get the mask.
11205 if (ip_plen_to_mask_v6(IPV6_ABITS
- addrlen
, &v6mask
) == NULL
)
11207 for (i
= 0; i
< 4; i
++) {
11208 v6mask
.s6_addr32
[i
] ^= (uint32_t)0xffffffff;
11211 if (V6_MASK_EQ(v6addr
, v6mask
, ill
->ill_token
) &&
11212 ill
->ill_token_length
== addrlen
)
11213 return (0); /* No change */
11215 if (ipif
->ipif_flags
& IPIF_UP
) {
11216 err
= ipif_logical_down(ipif
, q
, mp
);
11217 if (err
== EINPROGRESS
)
11219 (void) ipif_down_tail(ipif
);
11222 err
= ip_sioctl_token_tail(ipif
, sin6
, addrlen
, q
, mp
, need_up
);
11227 ip_sioctl_token_tail(ipif_t
*ipif
, sin6_t
*sin6
, int addrlen
, queue_t
*q
,
11228 mblk_t
*mp
, boolean_t need_up
)
11232 ill_t
*ill
= ipif
->ipif_ill
;
11236 ip1dbg(("ip_sioctl_token_tail(%s:%u %p)\n",
11237 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
11238 v6addr
= sin6
->sin6_addr
;
11240 * The length of the token is the length from the end. To get
11241 * the proper mask for this, compute the mask of the bits not
11242 * in the token; ie. the prefix, and then xor to get the mask.
11244 (void) ip_plen_to_mask_v6(IPV6_ABITS
- addrlen
, &v6mask
);
11245 for (i
= 0; i
< 4; i
++)
11246 v6mask
.s6_addr32
[i
] ^= (uint32_t)0xffffffff;
11248 mutex_enter(&ill
->ill_lock
);
11249 V6_MASK_COPY(v6addr
, v6mask
, ill
->ill_token
);
11250 ill
->ill_token_length
= addrlen
;
11251 ill
->ill_manual_token
= 1;
11253 /* Reconfigure the link-local address based on this new token */
11254 ipif_setlinklocal(ill
->ill_ipif
);
11256 mutex_exit(&ill
->ill_lock
);
11260 * Now bring the interface back up. If this
11261 * is the only IPIF for the ILL, ipif_up
11262 * will have to re-bind to the device, so
11263 * we may get back EINPROGRESS, in which
11264 * case, this IOCTL will get completed in
11265 * ip_rput_dlpi when we see the DL_BIND_ACK.
11267 err
= ipif_up(ipif
, q
, mp
);
11268 if (err
== EINPROGRESS
)
11276 ip_sioctl_get_token(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
11277 ip_ioctl_cmd_t
*ipi
, void *if_req
)
11280 sin6_t
*sin6
= (sin6_t
*)sin
;
11281 struct lifreq
*lifr
= (struct lifreq
*)if_req
;
11283 ip1dbg(("ip_sioctl_get_token(%s:%u %p)\n",
11284 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
11285 if (ipif
->ipif_id
!= 0)
11288 ill
= ipif
->ipif_ill
;
11289 if (!ill
->ill_isv6
)
11293 sin6
->sin6_family
= AF_INET6
;
11294 ASSERT(!IN6_IS_ADDR_V4MAPPED(&ill
->ill_token
));
11295 sin6
->sin6_addr
= ill
->ill_token
;
11296 lifr
->lifr_addrlen
= ill
->ill_token_length
;
11301 * Set (hardware) link specific information that might override
11302 * what was acquired through the DL_INFO_ACK.
11306 ip_sioctl_lnkinfo(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
11307 ip_ioctl_cmd_t
*ipi
, void *if_req
)
11309 ill_t
*ill
= ipif
->ipif_ill
;
11311 struct lifreq
*lifr
= (struct lifreq
*)if_req
;
11312 lif_ifinfo_req_t
*lir
;
11314 ip1dbg(("ip_sioctl_lnkinfo(%s:%u %p)\n",
11315 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
11316 lir
= &lifr
->lifr_ifinfo
;
11317 ASSERT(IAM_WRITER_IPIF(ipif
));
11319 /* Only allow for logical unit zero i.e. not on "bge0:17" */
11320 if (ipif
->ipif_id
!= 0)
11323 /* Set interface MTU. */
11324 if (ipif
->ipif_isv6
)
11325 ip_min_mtu
= IPV6_MIN_MTU
;
11327 ip_min_mtu
= IP_MIN_MTU
;
11330 * Verify values before we set anything. Allow zero to
11331 * mean unspecified.
11333 * XXX We should be able to set the user-defined lir_mtu to some value
11334 * that is greater than ill_current_frag but less than ill_max_frag- the
11335 * ill_max_frag value tells us the max MTU that can be handled by the
11336 * datalink, whereas the ill_current_frag is dynamically computed for
11337 * some link-types like tunnels, based on the tunnel PMTU. However,
11338 * since there is currently no way of distinguishing between
11339 * administratively fixed link mtu values (e.g., those set via
11340 * /sbin/dladm) and dynamically discovered MTUs (e.g., those discovered
11341 * for tunnels) we conservatively choose the ill_current_frag as the
11344 if (lir
->lir_maxmtu
!= 0 &&
11345 (lir
->lir_maxmtu
> ill
->ill_current_frag
||
11346 lir
->lir_maxmtu
< ip_min_mtu
))
11348 if (lir
->lir_reachtime
!= 0 &&
11349 lir
->lir_reachtime
> ND_MAX_REACHTIME
)
11351 if (lir
->lir_reachretrans
!= 0 &&
11352 lir
->lir_reachretrans
> ND_MAX_REACHRETRANSTIME
)
11355 mutex_enter(&ill
->ill_lock
);
11357 * The dce and fragmentation code can handle changes to ill_mtu
11358 * concurrent with sending/fragmenting packets.
11360 if (lir
->lir_maxmtu
!= 0)
11361 ill
->ill_user_mtu
= lir
->lir_maxmtu
;
11363 if (lir
->lir_reachtime
!= 0)
11364 ill
->ill_reachable_time
= lir
->lir_reachtime
;
11366 if (lir
->lir_reachretrans
!= 0)
11367 ill
->ill_reachable_retrans_time
= lir
->lir_reachretrans
;
11369 ill
->ill_max_hops
= lir
->lir_maxhops
;
11370 ill
->ill_max_buf
= ND_MAX_Q
;
11371 if (!(ill
->ill_flags
& ILLF_FIXEDMTU
) && ill
->ill_user_mtu
!= 0) {
11373 * ill_mtu is the actual interface MTU, obtained as the min
11374 * of user-configured mtu and the value announced by the
11375 * driver (via DL_NOTE_SDU_SIZE/DL_INFO_ACK). Note that since
11376 * we have already made the choice of requiring
11377 * ill_user_mtu < ill_current_frag by the time we get here,
11378 * the ill_mtu effectively gets assigned to the ill_user_mtu
11381 ill
->ill_mtu
= MIN(ill
->ill_current_frag
, ill
->ill_user_mtu
);
11382 ill
->ill_mc_mtu
= MIN(ill
->ill_mc_mtu
, ill
->ill_user_mtu
);
11384 mutex_exit(&ill
->ill_lock
);
11387 * Make sure all dce_generation checks find out
11388 * that ill_mtu/ill_mc_mtu has changed.
11390 if (!(ill
->ill_flags
& ILLF_FIXEDMTU
) && (lir
->lir_maxmtu
!= 0))
11391 dce_increment_all_generations(ill
->ill_isv6
, ill
->ill_ipst
);
11394 * Refresh IPMP meta-interface MTU if necessary.
11396 if (IS_UNDER_IPMP(ill
))
11397 ipmp_illgrp_refresh_mtu(ill
->ill_grp
);
11404 ip_sioctl_get_lnkinfo(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
11405 ip_ioctl_cmd_t
*ipi
, void *if_req
)
11407 struct lif_ifinfo_req
*lir
;
11408 ill_t
*ill
= ipif
->ipif_ill
;
11410 ip1dbg(("ip_sioctl_get_lnkinfo(%s:%u %p)\n",
11411 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
11412 if (ipif
->ipif_id
!= 0)
11415 lir
= &((struct lifreq
*)if_req
)->lifr_ifinfo
;
11416 lir
->lir_maxhops
= ill
->ill_max_hops
;
11417 lir
->lir_reachtime
= ill
->ill_reachable_time
;
11418 lir
->lir_reachretrans
= ill
->ill_reachable_retrans_time
;
11419 lir
->lir_maxmtu
= ill
->ill_mtu
;
11425 * Return best guess as to the subnet mask for the specified address.
11426 * Based on the subnet masks for all the configured interfaces.
11428 * We end up returning a zero mask in the case of default, multicast or
11432 ip_subnet_mask(ipaddr_t addr
, ipif_t
**ipifp
, ip_stack_t
*ipst
)
11437 ill_walk_context_t ctx
;
11438 ipif_t
*fallback_ipif
= NULL
;
11440 net_mask
= ip_net_mask(addr
);
11441 if (net_mask
== 0) {
11446 /* Let's check to see if this is maybe a local subnet route. */
11447 /* this function only applies to IPv4 interfaces */
11448 rw_enter(&ipst
->ips_ill_g_lock
, RW_READER
);
11449 ill
= ILL_START_WALK_V4(&ctx
, ipst
);
11450 for (; ill
!= NULL
; ill
= ill_next(&ctx
, ill
)) {
11451 mutex_enter(&ill
->ill_lock
);
11452 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
;
11453 ipif
= ipif
->ipif_next
) {
11454 if (IPIF_IS_CONDEMNED(ipif
))
11456 if (!(ipif
->ipif_flags
& IPIF_UP
))
11458 if ((ipif
->ipif_subnet
& net_mask
) ==
11459 (addr
& net_mask
)) {
11461 * Don't trust pt-pt interfaces if there are
11462 * other interfaces.
11464 if (ipif
->ipif_flags
& IPIF_POINTOPOINT
) {
11465 if (fallback_ipif
== NULL
) {
11466 ipif_refhold_locked(ipif
);
11467 fallback_ipif
= ipif
;
11473 * Fine. Just assume the same net mask as the
11474 * directly attached subnet interface is using.
11476 ipif_refhold_locked(ipif
);
11477 mutex_exit(&ill
->ill_lock
);
11478 rw_exit(&ipst
->ips_ill_g_lock
);
11479 if (fallback_ipif
!= NULL
)
11480 ipif_refrele(fallback_ipif
);
11482 return (ipif
->ipif_net_mask
);
11485 mutex_exit(&ill
->ill_lock
);
11487 rw_exit(&ipst
->ips_ill_g_lock
);
11489 *ipifp
= fallback_ipif
;
11490 return ((fallback_ipif
!= NULL
) ?
11491 fallback_ipif
->ipif_net_mask
: net_mask
);
11495 * ip_sioctl_copyin_setup calls ip_wput_ioctl to process the IP_IOCTL ioctl.
11498 ip_wput_ioctl(queue_t
*q
, mblk_t
*mp
)
11508 ip1dbg(("ip_wput_ioctl"));
11509 iocp
= (IOCP
)mp
->b_rptr
;
11512 iocp
->ioc_error
= EINVAL
;
11513 mp
->b_datap
->db_type
= M_IOCNAK
;
11514 iocp
->ioc_count
= 0;
11520 * These IOCTLs provide various control capabilities to
11521 * upstream agents such as ULPs and processes. There
11522 * are currently two such IOCTLs implemented. They
11523 * are used by TCP to provide update information for
11524 * existing IREs and to forcibly delete an IRE for a
11525 * host that is not responding, thereby forcing an
11526 * attempt at a new route.
11528 iocp
->ioc_error
= EINVAL
;
11529 if (!pullupmsg(mp1
, sizeof (ipllc
->ipllc_cmd
)))
11532 ipllc
= (ipllc_t
*)mp1
->b_rptr
;
11533 for (ipft
= ip_ioctl_ftbl
; ipft
->ipft_pfi
; ipft
++) {
11534 if (ipllc
->ipllc_cmd
== ipft
->ipft_cmd
)
11538 * prefer credential from mblk over ioctl;
11539 * see ip_sioctl_copyin_setup
11541 cr
= msg_getcred(mp
, NULL
);
11546 * Refhold the conn in case the request gets queued up in some lookup
11549 connp
= Q_TO_CONN(q
);
11550 CONN_INC_REF(connp
);
11551 CONN_INC_IOCTLREF(connp
);
11552 if (ipft
->ipft_pfi
&&
11553 ((mp1
->b_wptr
- mp1
->b_rptr
) >= ipft
->ipft_min_size
||
11554 pullupmsg(mp1
, ipft
->ipft_min_size
))) {
11555 error
= (*ipft
->ipft_pfi
)(q
,
11556 (ipft
->ipft_flags
& IPFT_F_SELF_REPLY
) ? mp
: mp1
, cr
);
11558 if (ipft
->ipft_flags
& IPFT_F_SELF_REPLY
) {
11560 * CONN_OPER_PENDING_DONE happens in the function called
11561 * through ipft_pfi above.
11566 CONN_DEC_IOCTLREF(connp
);
11567 CONN_OPER_PENDING_DONE(connp
);
11568 if (ipft
->ipft_flags
& IPFT_F_NO_REPLY
) {
11572 iocp
->ioc_error
= error
;
11575 mp
->b_datap
->db_type
= M_IOCACK
;
11576 if (iocp
->ioc_error
)
11577 iocp
->ioc_count
= 0;
11582 * Assign a unique id for the ipif. This is used by sctp_addr.c
11583 * Note: remove if sctp_addr.c is redone to not shadow ill/ipif data structures.
11586 ipif_assign_seqid(ipif_t
*ipif
)
11588 ip_stack_t
*ipst
= ipif
->ipif_ill
->ill_ipst
;
11590 ipif
->ipif_seqid
= atomic_inc_64_nv(&ipst
->ips_ipif_g_seqid
);
11594 * Clone the contents of `sipif' to `dipif'. Requires that both ipifs are
11595 * administratively down (i.e., no DAD), of the same type, and locked. Note
11596 * that the clone is complete -- including the seqid -- and the expectation is
11597 * that the caller will either free or overwrite `sipif' before it's unlocked.
11600 ipif_clone(const ipif_t
*sipif
, ipif_t
*dipif
)
11602 ASSERT(MUTEX_HELD(&sipif
->ipif_ill
->ill_lock
));
11603 ASSERT(MUTEX_HELD(&dipif
->ipif_ill
->ill_lock
));
11604 ASSERT(!(sipif
->ipif_flags
& (IPIF_UP
|IPIF_DUPLICATE
)));
11605 ASSERT(!(dipif
->ipif_flags
& (IPIF_UP
|IPIF_DUPLICATE
)));
11606 ASSERT(sipif
->ipif_ire_type
== dipif
->ipif_ire_type
);
11608 dipif
->ipif_flags
= sipif
->ipif_flags
;
11609 dipif
->ipif_zoneid
= sipif
->ipif_zoneid
;
11610 dipif
->ipif_v6subnet
= sipif
->ipif_v6subnet
;
11611 dipif
->ipif_v6lcl_addr
= sipif
->ipif_v6lcl_addr
;
11612 dipif
->ipif_v6net_mask
= sipif
->ipif_v6net_mask
;
11613 dipif
->ipif_v6brd_addr
= sipif
->ipif_v6brd_addr
;
11614 dipif
->ipif_v6pp_dst_addr
= sipif
->ipif_v6pp_dst_addr
;
11617 * As per the comment atop the function, we assume that these sipif
11618 * fields will be changed before sipif is unlocked.
11620 dipif
->ipif_seqid
= sipif
->ipif_seqid
;
11621 dipif
->ipif_state_flags
= sipif
->ipif_state_flags
;
11625 * Transfer the contents of `sipif' to `dipif', and then free (if `virgipif'
11626 * is NULL) or overwrite `sipif' with `virgipif', which must be a virgin
11627 * (unreferenced) ipif. Also, if `sipif' is used by the current xop, then
11628 * transfer the xop to `dipif'. Requires that all ipifs are administratively
11629 * down (i.e., no DAD), of the same type, and unlocked.
11632 ipif_transfer(ipif_t
*sipif
, ipif_t
*dipif
, ipif_t
*virgipif
)
11634 ipsq_t
*ipsq
= sipif
->ipif_ill
->ill_phyint
->phyint_ipsq
;
11635 ipxop_t
*ipx
= ipsq
->ipsq_xop
;
11637 ASSERT(sipif
!= dipif
);
11638 ASSERT(sipif
!= virgipif
);
11641 * Grab all of the locks that protect the ipif in a defined order.
11643 GRAB_ILL_LOCKS(sipif
->ipif_ill
, dipif
->ipif_ill
);
11645 ipif_clone(sipif
, dipif
);
11646 if (virgipif
!= NULL
) {
11647 ipif_clone(virgipif
, sipif
);
11651 RELEASE_ILL_LOCKS(sipif
->ipif_ill
, dipif
->ipif_ill
);
11654 * Transfer ownership of the current xop, if necessary.
11656 if (ipx
->ipx_current_ipif
== sipif
) {
11657 ASSERT(ipx
->ipx_pending_ipif
== NULL
);
11658 mutex_enter(&ipx
->ipx_lock
);
11659 ipx
->ipx_current_ipif
= dipif
;
11660 mutex_exit(&ipx
->ipx_lock
);
11663 if (virgipif
== NULL
)
11669 * - <ill_name>:<ipif_id> is at most LIFNAMSIZ - 1 and
11670 * - logical interface is within the allowed range
11673 is_lifname_valid(ill_t
*ill
, unsigned int ipif_id
)
11675 if (snprintf(NULL
, 0, "%s:%d", ill
->ill_name
, ipif_id
) >= LIFNAMSIZ
)
11676 return (ENAMETOOLONG
);
11678 if (ipif_id
>= ill
->ill_ipst
->ips_ip_addrs_per_if
)
11684 * Insert the ipif, so that the list of ipifs on the ill will be sorted
11685 * with respect to ipif_id. Note that an ipif with an ipif_id of -1 will
11686 * be inserted into the first space available in the list. The value of
11687 * ipif_id will then be set to the appropriate value for its position.
11690 ipif_insert(ipif_t
*ipif
, boolean_t acquire_g_lock
)
11698 ASSERT(ipif
->ipif_ill
->ill_net_type
== IRE_LOOPBACK
||
11699 IAM_WRITER_IPIF(ipif
));
11701 ill
= ipif
->ipif_ill
;
11702 ASSERT(ill
!= NULL
);
11703 ipst
= ill
->ill_ipst
;
11706 * In the case of lo0:0 we already hold the ill_g_lock.
11707 * ill_lookup_on_name (acquires ill_g_lock) -> ipif_allocate ->
11710 if (acquire_g_lock
)
11711 rw_enter(&ipst
->ips_ill_g_lock
, RW_WRITER
);
11712 mutex_enter(&ill
->ill_lock
);
11713 id
= ipif
->ipif_id
;
11714 tipifp
= &(ill
->ill_ipif
);
11715 if (id
== -1) { /* need to find a real id */
11717 while ((tipif
= *tipifp
) != NULL
) {
11718 ASSERT(tipif
->ipif_id
>= id
);
11719 if (tipif
->ipif_id
!= id
)
11720 break; /* non-consecutive id */
11722 tipifp
= &(tipif
->ipif_next
);
11724 if ((err
= is_lifname_valid(ill
, id
)) != 0) {
11725 mutex_exit(&ill
->ill_lock
);
11726 if (acquire_g_lock
)
11727 rw_exit(&ipst
->ips_ill_g_lock
);
11730 ipif
->ipif_id
= id
; /* assign new id */
11731 } else if ((err
= is_lifname_valid(ill
, id
)) == 0) {
11732 /* we have a real id; insert ipif in the right place */
11733 while ((tipif
= *tipifp
) != NULL
) {
11734 ASSERT(tipif
->ipif_id
!= id
);
11735 if (tipif
->ipif_id
> id
)
11736 break; /* found correct location */
11737 tipifp
= &(tipif
->ipif_next
);
11740 mutex_exit(&ill
->ill_lock
);
11741 if (acquire_g_lock
)
11742 rw_exit(&ipst
->ips_ill_g_lock
);
11746 ASSERT(tipifp
!= &(ill
->ill_ipif
) || id
== 0);
11748 ipif
->ipif_next
= tipif
;
11750 mutex_exit(&ill
->ill_lock
);
11751 if (acquire_g_lock
)
11752 rw_exit(&ipst
->ips_ill_g_lock
);
11758 ipif_remove(ipif_t
*ipif
)
11761 ill_t
*ill
= ipif
->ipif_ill
;
11763 ASSERT(RW_WRITE_HELD(&ill
->ill_ipst
->ips_ill_g_lock
));
11765 mutex_enter(&ill
->ill_lock
);
11766 ipifp
= &ill
->ill_ipif
;
11767 for (; *ipifp
!= NULL
; ipifp
= &ipifp
[0]->ipif_next
) {
11768 if (*ipifp
== ipif
) {
11769 *ipifp
= ipif
->ipif_next
;
11773 mutex_exit(&ill
->ill_lock
);
11777 * Allocate and initialize a new interface control structure. (Always
11778 * called as writer.)
11779 * When ipif_allocate() is called from ip_ll_subnet_defaults, the ill
11780 * is not part of the global linked list of ills. ipif_seqid is unique
11781 * in the system and to preserve the uniqueness, it is assigned only
11782 * when ill becomes part of the global list. At that point ill will
11783 * have a name. If it doesn't get assigned here, it will get assigned
11784 * in ipif_set_values() as part of SIOCSLIFNAME processing.
11785 * Aditionally, if we come here from ip_ll_subnet_defaults, we don't set
11786 * the interface flags or any other information from the DL_INFO_ACK for
11787 * DL_STYLE2 drivers (initialize == B_FALSE), since we won't have them at
11788 * this point. The flags etc. will be set in ip_ll_subnet_defaults when the
11789 * second DL_INFO_ACK comes in from the driver.
11792 ipif_allocate(ill_t
*ill
, int id
, uint_t ire_type
, boolean_t initialize
,
11793 boolean_t insert
, int *errorp
)
11797 ip_stack_t
*ipst
= ill
->ill_ipst
;
11799 ip1dbg(("ipif_allocate(%s:%d ill %p)\n",
11800 ill
->ill_name
, id
, (void *)ill
));
11801 ASSERT(ire_type
== IRE_LOOPBACK
|| IAM_WRITER_ILL(ill
));
11803 if (errorp
!= NULL
)
11806 if ((ipif
= mi_alloc(sizeof (ipif_t
), BPRI_MED
)) == NULL
) {
11807 if (errorp
!= NULL
)
11811 *ipif
= ipif_zero
; /* start clean */
11813 ipif
->ipif_ill
= ill
;
11814 ipif
->ipif_id
= id
; /* could be -1 */
11816 * Inherit the zoneid from the ill; for the shared stack instance
11817 * this is always the global zone
11819 ipif
->ipif_zoneid
= ill
->ill_zoneid
;
11821 ipif
->ipif_refcnt
= 0;
11824 if ((err
= ipif_insert(ipif
, ire_type
!= IRE_LOOPBACK
)) != 0) {
11826 if (errorp
!= NULL
)
11830 /* -1 id should have been replaced by real id */
11831 id
= ipif
->ipif_id
;
11835 if (ill
->ill_name
[0] != '\0')
11836 ipif_assign_seqid(ipif
);
11839 * If this is the zeroth ipif on the IPMP ill, create the illgrp
11840 * (which must not exist yet because the zeroth ipif is created once
11841 * per ill). However, do not not link it to the ipmp_grp_t until
11842 * I_PLINK is called; see ip_sioctl_plink_ipmp() for details.
11844 if (id
== 0 && IS_IPMP(ill
)) {
11845 if (ipmp_illgrp_create(ill
) == NULL
) {
11847 rw_enter(&ipst
->ips_ill_g_lock
, RW_WRITER
);
11849 rw_exit(&ipst
->ips_ill_g_lock
);
11852 if (errorp
!= NULL
)
11859 * We grab ill_lock to protect the flag changes. The ipif is still
11860 * not up and can't be looked up until the ioctl completes and the
11861 * IPIF_CHANGING flag is cleared.
11863 mutex_enter(&ill
->ill_lock
);
11865 ipif
->ipif_ire_type
= ire_type
;
11867 if (ipif
->ipif_isv6
) {
11868 ill
->ill_flags
|= ILLF_IPV6
;
11870 ipaddr_t inaddr_any
= INADDR_ANY
;
11872 ill
->ill_flags
|= ILLF_IPV4
;
11874 /* Keep the IN6_IS_ADDR_V4MAPPED assertions happy */
11875 IN6_IPADDR_TO_V4MAPPED(inaddr_any
,
11876 &ipif
->ipif_v6lcl_addr
);
11877 IN6_IPADDR_TO_V4MAPPED(inaddr_any
,
11878 &ipif
->ipif_v6subnet
);
11879 IN6_IPADDR_TO_V4MAPPED(inaddr_any
,
11880 &ipif
->ipif_v6net_mask
);
11881 IN6_IPADDR_TO_V4MAPPED(inaddr_any
,
11882 &ipif
->ipif_v6brd_addr
);
11883 IN6_IPADDR_TO_V4MAPPED(inaddr_any
,
11884 &ipif
->ipif_v6pp_dst_addr
);
11888 * Don't set the interface flags etc. now, will do it in
11889 * ip_ll_subnet_defaults.
11895 * NOTE: The IPMP meta-interface is special-cased because it starts
11896 * with no underlying interfaces (and thus an unknown broadcast
11897 * address length), but all interfaces that can be placed into an IPMP
11898 * group are required to be broadcast-capable.
11900 if (ill
->ill_bcast_addr_length
!= 0 || IS_IPMP(ill
)) {
11902 * Later detect lack of DLPI driver multicast capability by
11903 * catching DL_ENABMULTI_REQ errors in ip_rput_dlpi().
11905 ill
->ill_flags
|= ILLF_MULTICAST
;
11906 if (!ipif
->ipif_isv6
)
11907 ipif
->ipif_flags
|= IPIF_BROADCAST
;
11909 if (ill
->ill_net_type
!= IRE_LOOPBACK
) {
11910 if (ipif
->ipif_isv6
)
11912 * Note: xresolv interfaces will eventually need
11913 * NOARP set here as well, but that will require
11914 * those external resolvers to have some
11915 * knowledge of that flag and act appropriately.
11916 * Not to be changed at present.
11918 ill
->ill_flags
|= ILLF_NONUD
;
11920 ill
->ill_flags
|= ILLF_NOARP
;
11922 if (ill
->ill_phys_addr_length
== 0) {
11924 ipif
->ipif_flags
|= IPIF_NOXMIT
;
11926 /* pt-pt supports multicast. */
11927 ill
->ill_flags
|= ILLF_MULTICAST
;
11928 if (ill
->ill_net_type
!= IRE_LOOPBACK
)
11929 ipif
->ipif_flags
|= IPIF_POINTOPOINT
;
11934 mutex_exit(&ill
->ill_lock
);
11939 * Remove the neighbor cache entries associated with this logical
11943 ipif_arp_down(ipif_t
*ipif
)
11945 ill_t
*ill
= ipif
->ipif_ill
;
11948 ip1dbg(("ipif_arp_down(%s:%u)\n", ill
->ill_name
, ipif
->ipif_id
));
11949 ASSERT(IAM_WRITER_IPIF(ipif
));
11951 DTRACE_PROBE3(ipif__downup
, char *, "ipif_arp_down",
11952 ill_t
*, ill
, ipif_t
*, ipif
);
11953 ipif_nce_down(ipif
);
11956 * If this is the last ipif that is going down and there are no
11957 * duplicate addresses we may yet attempt to re-probe, then we need to
11958 * clean up ARP completely.
11960 if (ill
->ill_ipif_up_count
== 0 && ill
->ill_ipif_dup_count
== 0 &&
11961 !ill
->ill_logical_down
&& ill
->ill_net_type
== IRE_IF_RESOLVER
) {
11963 * If this was the last ipif on an IPMP interface, purge any
11964 * static ARP entries associated with it.
11967 ipmp_illgrp_refresh_arpent(ill
->ill_grp
);
11969 /* UNBIND, DETACH */
11970 err
= arp_ll_down(ill
);
11977 * Get the resolver set up for a new IP address. (Always called as writer.)
11978 * Called both for IPv4 and IPv6 interfaces, though it only does some
11979 * basic DAD related initialization for IPv6. Honors ILLF_NOARP.
11981 * The enumerated value res_act tunes the behavior:
11982 * * Res_act_initial: set up all the resolver structures for a new
11984 * * Res_act_defend: tell ARP that it needs to send a single gratuitous
11985 * ARP message in defense of the address.
11986 * * Res_act_rebind: tell ARP to change the hardware address for an IP
11987 * address (and issue gratuitous ARPs). Used by ipmp_ill_bind_ipif().
11989 * Returns zero on success, or an errno upon failure.
11992 ipif_resolver_up(ipif_t
*ipif
, enum ip_resolver_action res_act
)
11994 ill_t
*ill
= ipif
->ipif_ill
;
11998 ip1dbg(("ipif_resolver_up(%s:%u) flags 0x%x\n",
11999 ill
->ill_name
, ipif
->ipif_id
, (uint_t
)ipif
->ipif_flags
));
12000 ASSERT(IAM_WRITER_IPIF(ipif
));
12003 if (res_act
== Res_act_initial
) {
12004 ipif
->ipif_addr_ready
= 0;
12006 * We're bringing an interface up here. There's no way that we
12007 * should need to shut down ARP now.
12009 mutex_enter(&ill
->ill_lock
);
12010 if (ipif
->ipif_flags
& IPIF_DUPLICATE
) {
12011 ipif
->ipif_flags
&= ~IPIF_DUPLICATE
;
12012 ill
->ill_ipif_dup_count
--;
12015 mutex_exit(&ill
->ill_lock
);
12017 if (ipif
->ipif_recovery_id
!= 0)
12018 (void) untimeout(ipif
->ipif_recovery_id
);
12019 ipif
->ipif_recovery_id
= 0;
12020 if (ill
->ill_net_type
!= IRE_IF_RESOLVER
) {
12021 ipif
->ipif_addr_ready
= 1;
12024 /* NDP will set the ipif_addr_ready flag when it's ready */
12028 err
= ipif_arp_up(ipif
, res_act
, was_dup
);
12033 * This routine restarts IPv4/IPv6 duplicate address detection (DAD)
12034 * when a link has just gone back up.
12037 ipif_nce_start_dad(ipif_t
*ipif
)
12040 ill_t
*ill
= ipif
->ipif_ill
;
12041 boolean_t isv6
= ill
->ill_isv6
;
12044 ncec
= ncec_lookup_illgrp_v6(ipif
->ipif_ill
,
12045 &ipif
->ipif_v6lcl_addr
);
12049 if (ill
->ill_net_type
!= IRE_IF_RESOLVER
||
12050 (ipif
->ipif_flags
& IPIF_UNNUMBERED
) ||
12051 ipif
->ipif_lcl_addr
== INADDR_ANY
) {
12053 * If we can't contact ARP for some reason,
12054 * that's not really a problem. Just send
12055 * out the routing socket notification that
12056 * DAD completion would have done, and continue.
12058 ipif_mask_reply(ipif
);
12059 ipif_up_notify(ipif
);
12060 ipif
->ipif_addr_ready
= 1;
12064 IN6_V4MAPPED_TO_IPADDR(&ipif
->ipif_v6lcl_addr
, v4addr
);
12065 ncec
= ncec_lookup_illgrp_v4(ipif
->ipif_ill
, &v4addr
);
12068 if (ncec
== NULL
) {
12069 ip1dbg(("couldn't find ncec for ipif %p leaving !ready\n",
12073 if (!nce_restart_dad(ncec
)) {
12075 * If we can't restart DAD for some reason, that's not really a
12076 * problem. Just send out the routing socket notification that
12077 * DAD completion would have done, and continue.
12079 ipif_up_notify(ipif
);
12080 ipif
->ipif_addr_ready
= 1;
12082 ncec_refrele(ncec
);
12086 * Restart duplicate address detection on all interfaces on the given ill.
12088 * This is called when an interface transitions from down to up
12089 * (DL_NOTE_LINK_UP) or up to down (DL_NOTE_LINK_DOWN).
12091 * Note that since the underlying physical link has transitioned, we must cause
12092 * at least one routing socket message to be sent here, either via DAD
12093 * completion or just by default on the first ipif. (If we don't do this, then
12094 * in.mpathd will see long delays when doing link-based failure recovery.)
12097 ill_restart_dad(ill_t
*ill
, boolean_t went_up
)
12105 * If layer two doesn't support duplicate address detection, then just
12106 * send the routing socket message now and be done with it.
12108 if (!ill
->ill_isv6
&& arp_no_defense
) {
12109 ip_rts_ifmsg(ill
->ill_ipif
, RTSQ_DEFAULT
);
12113 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
; ipif
= ipif
->ipif_next
) {
12116 if (ipif
->ipif_flags
& IPIF_UP
) {
12117 ipif_nce_start_dad(ipif
);
12118 } else if (ipif
->ipif_flags
& IPIF_DUPLICATE
) {
12120 * kick off the bring-up process now.
12122 ipif_do_recovery(ipif
);
12125 * Unfortunately, the first ipif is "special"
12126 * and represents the underlying ill in the
12127 * routing socket messages. Thus, when this
12128 * one ipif is down, we must still notify so
12129 * that the user knows the IFF_RUNNING status
12130 * change. (If the first ipif is up, then
12131 * we'll handle eventual routing socket
12132 * notification via DAD completion.)
12134 if (ipif
== ill
->ill_ipif
) {
12135 ip_rts_ifmsg(ill
->ill_ipif
,
12141 * After link down, we'll need to send a new routing
12142 * message when the link comes back, so clear
12145 ipif
->ipif_addr_ready
= 0;
12150 * If we've torn down links, then notify the user right away.
12153 ip_rts_ifmsg(ill
->ill_ipif
, RTSQ_DEFAULT
);
12157 ipsq_delete(ipsq_t
*ipsq
)
12159 ipxop_t
*ipx
= ipsq
->ipsq_xop
;
12161 ipsq
->ipsq_ipst
= NULL
;
12162 ASSERT(ipsq
->ipsq_phyint
== NULL
);
12163 ASSERT(ipsq
->ipsq_xop
!= NULL
);
12164 ASSERT(ipsq
->ipsq_xopq_mphead
== NULL
&& ipx
->ipx_mphead
== NULL
);
12165 ASSERT(ipx
->ipx_pending_mp
== NULL
);
12166 kmem_free(ipsq
, sizeof (ipsq_t
));
12170 ill_up_ipifs_on_ill(ill_t
*ill
, queue_t
*q
, mblk_t
*mp
)
12178 ASSERT(IAM_WRITER_ILL(ill
));
12179 ill
->ill_up_ipifs
= B_TRUE
;
12180 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
; ipif
= ipif
->ipif_next
) {
12181 if (ipif
->ipif_was_up
) {
12182 if (!(ipif
->ipif_flags
& IPIF_UP
))
12183 err
= ipif_up(ipif
, q
, mp
);
12184 ipif
->ipif_was_up
= B_FALSE
;
12186 ASSERT(err
== EINPROGRESS
);
12191 ill
->ill_up_ipifs
= B_FALSE
;
12196 * This function is called to bring up all the ipifs that were up before
12197 * bringing the ill down via ill_down_ipifs().
12200 ill_up_ipifs(ill_t
*ill
, queue_t
*q
, mblk_t
*mp
)
12204 ASSERT(IAM_WRITER_ILL(ill
));
12206 if (ill
->ill_replumbing
) {
12207 ill
->ill_replumbing
= 0;
12209 * Send down REPLUMB_DONE notification followed by the
12210 * BIND_REQ on the arp stream.
12212 if (!ill
->ill_isv6
)
12213 arp_send_replumb_conf(ill
);
12215 err
= ill_up_ipifs_on_ill(ill
->ill_phyint
->phyint_illv4
, q
, mp
);
12219 return (ill_up_ipifs_on_ill(ill
->ill_phyint
->phyint_illv6
, q
, mp
));
12223 * Bring down any IPIF_UP ipifs on ill. If "logical" is B_TRUE, we bring
12224 * down the ipifs without sending DL_UNBIND_REQ to the driver.
12227 ill_down_ipifs(ill_t
*ill
, boolean_t logical
)
12231 ASSERT(IAM_WRITER_ILL(ill
));
12233 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
; ipif
= ipif
->ipif_next
) {
12235 * We go through the ipif_down logic even if the ipif
12236 * is already down, since routes can be added based
12237 * on down ipifs. Going through ipif_down once again
12238 * will delete any IREs created based on these routes.
12240 if (ipif
->ipif_flags
& IPIF_UP
)
12241 ipif
->ipif_was_up
= B_TRUE
;
12244 (void) ipif_logical_down(ipif
, NULL
, NULL
);
12245 ipif_non_duplicate(ipif
);
12246 (void) ipif_down_tail(ipif
);
12248 (void) ipif_down(ipif
, NULL
, NULL
);
12254 * Redo source address selection. This makes IXAF_VERIFY_SOURCE take
12255 * a look again at valid source addresses.
12256 * This should be called each time after the set of source addresses has been
12260 ip_update_source_selection(ip_stack_t
*ipst
)
12262 /* We skip past SRC_GENERATION_VERIFY */
12263 if (atomic_inc_32_nv(&ipst
->ips_src_generation
) ==
12264 SRC_GENERATION_VERIFY
)
12265 atomic_inc_32(&ipst
->ips_src_generation
);
12269 * Finish the group join started in ip_sioctl_groupname().
12273 ip_join_illgrps(ipsq_t
*ipsq
, queue_t
*q
, mblk_t
*mp
, void *dummy
)
12275 ill_t
*ill
= q
->q_ptr
;
12276 phyint_t
*phyi
= ill
->ill_phyint
;
12277 ipmp_grp_t
*grp
= phyi
->phyint_grp
;
12278 ip_stack_t
*ipst
= ill
->ill_ipst
;
12280 /* IS_UNDER_IPMP() won't work until ipmp_ill_join_illgrp() is called */
12281 ASSERT(!IS_IPMP(ill
) && grp
!= NULL
);
12282 ASSERT(IAM_WRITER_IPSQ(ipsq
));
12284 if (phyi
->phyint_illv4
!= NULL
) {
12285 rw_enter(&ipst
->ips_ipmp_lock
, RW_WRITER
);
12286 VERIFY(grp
->gr_pendv4
-- > 0);
12287 rw_exit(&ipst
->ips_ipmp_lock
);
12288 ipmp_ill_join_illgrp(phyi
->phyint_illv4
, grp
->gr_v4
);
12290 if (phyi
->phyint_illv6
!= NULL
) {
12291 rw_enter(&ipst
->ips_ipmp_lock
, RW_WRITER
);
12292 VERIFY(grp
->gr_pendv6
-- > 0);
12293 rw_exit(&ipst
->ips_ipmp_lock
);
12294 ipmp_ill_join_illgrp(phyi
->phyint_illv6
, grp
->gr_v6
);
12300 * Process an SIOCSLIFGROUPNAME request.
12304 ip_sioctl_groupname(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
12305 ip_ioctl_cmd_t
*ipip
, void *ifreq
)
12307 struct lifreq
*lifr
= ifreq
;
12308 ill_t
*ill
= ipif
->ipif_ill
;
12309 ip_stack_t
*ipst
= ill
->ill_ipst
;
12310 phyint_t
*phyi
= ill
->ill_phyint
;
12311 ipmp_grp_t
*grp
= phyi
->phyint_grp
;
12316 * Note that phyint_grp can only change here, where we're exclusive.
12318 ASSERT(IAM_WRITER_ILL(ill
));
12320 if (ipif
->ipif_id
!= 0 || ill
->ill_usesrc_grp_next
!= NULL
||
12321 (phyi
->phyint_flags
& PHYI_VIRTUAL
))
12324 lifr
->lifr_groupname
[LIFGRNAMSIZ
- 1] = '\0';
12326 rw_enter(&ipst
->ips_ipmp_lock
, RW_WRITER
);
12329 * If the name hasn't changed, there's nothing to do.
12331 if (grp
!= NULL
&& strcmp(grp
->gr_name
, lifr
->lifr_groupname
) == 0)
12335 * Handle requests to rename an IPMP meta-interface.
12337 * Note that creation of the IPMP meta-interface is handled in
12338 * userland through the standard plumbing sequence. As part of the
12339 * plumbing the IPMP meta-interface, its initial groupname is set to
12340 * the name of the interface (see ipif_set_values_tail()).
12342 if (IS_IPMP(ill
)) {
12343 err
= ipmp_grp_rename(grp
, lifr
->lifr_groupname
);
12348 * Handle requests to add or remove an IP interface from a group.
12350 if (lifr
->lifr_groupname
[0] != '\0') { /* add */
12352 * Moves are handled by first removing the interface from
12353 * its existing group, and then adding it to another group.
12354 * So, fail if it's already in a group.
12356 if (IS_UNDER_IPMP(ill
)) {
12361 grp
= ipmp_grp_lookup(lifr
->lifr_groupname
, ipst
);
12368 * Check if the phyint and its ills are suitable for
12369 * inclusion into the group.
12371 if ((err
= ipmp_grp_vet_phyint(grp
, phyi
)) != 0)
12375 * Checks pass; join the group, and enqueue the remaining
12376 * illgrp joins for when we've become part of the group xop
12377 * and are exclusive across its IPSQs. Since qwriter_ip()
12378 * requires an mblk_t to scribble on, and since `mp' will be
12379 * freed as part of completing the ioctl, allocate another.
12381 if ((ipsq_mp
= allocb(0, BPRI_MED
)) == NULL
) {
12387 * Before we drop ipmp_lock, bump gr_pend* to ensure that the
12388 * IPMP meta-interface ills needed by `phyi' cannot go away
12389 * before ip_join_illgrps() is called back. See the comments
12390 * in ip_sioctl_plink_ipmp() for more.
12392 if (phyi
->phyint_illv4
!= NULL
)
12394 if (phyi
->phyint_illv6
!= NULL
)
12397 rw_exit(&ipst
->ips_ipmp_lock
);
12399 ipmp_phyint_join_grp(phyi
, grp
);
12401 qwriter_ip(ill
, ill
->ill_rq
, ipsq_mp
, ip_join_illgrps
,
12402 SWITCH_OP
, B_FALSE
);
12406 * Request to remove the interface from a group. If the
12407 * interface is not in a group, this trivially succeeds.
12409 rw_exit(&ipst
->ips_ipmp_lock
);
12410 if (IS_UNDER_IPMP(ill
))
12411 ipmp_phyint_leave_grp(phyi
);
12415 rw_exit(&ipst
->ips_ipmp_lock
);
12420 * Process an SIOCGLIFBINDING request.
12424 ip_sioctl_get_binding(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
12425 ip_ioctl_cmd_t
*ipip
, void *ifreq
)
12428 struct lifreq
*lifr
= ifreq
;
12429 ip_stack_t
*ipst
= ipif
->ipif_ill
->ill_ipst
;
12431 if (!IS_IPMP(ipif
->ipif_ill
))
12434 rw_enter(&ipst
->ips_ipmp_lock
, RW_READER
);
12435 if ((ill
= ipif
->ipif_bound_ill
) == NULL
)
12436 lifr
->lifr_binding
[0] = '\0';
12438 (void) strlcpy(lifr
->lifr_binding
, ill
->ill_name
, LIFNAMSIZ
);
12439 rw_exit(&ipst
->ips_ipmp_lock
);
12444 * Process an SIOCGLIFGROUPNAME request.
12448 ip_sioctl_get_groupname(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
12449 ip_ioctl_cmd_t
*ipip
, void *ifreq
)
12452 struct lifreq
*lifr
= ifreq
;
12453 ip_stack_t
*ipst
= ipif
->ipif_ill
->ill_ipst
;
12455 rw_enter(&ipst
->ips_ipmp_lock
, RW_READER
);
12456 if ((grp
= ipif
->ipif_ill
->ill_phyint
->phyint_grp
) == NULL
)
12457 lifr
->lifr_groupname
[0] = '\0';
12459 (void) strlcpy(lifr
->lifr_groupname
, grp
->gr_name
, LIFGRNAMSIZ
);
12460 rw_exit(&ipst
->ips_ipmp_lock
);
12465 * Process an SIOCGLIFGROUPINFO request.
12469 ip_sioctl_groupinfo(ipif_t
*dummy_ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
12470 ip_ioctl_cmd_t
*ipip
, void *dummy
)
12473 lifgroupinfo_t
*lifgr
;
12474 ip_stack_t
*ipst
= CONNQ_TO_IPST(q
);
12476 /* ip_wput_nondata() verified mp->b_cont->b_cont */
12477 lifgr
= (lifgroupinfo_t
*)mp
->b_cont
->b_cont
->b_rptr
;
12478 lifgr
->gi_grname
[LIFGRNAMSIZ
- 1] = '\0';
12480 rw_enter(&ipst
->ips_ipmp_lock
, RW_READER
);
12481 if ((grp
= ipmp_grp_lookup(lifgr
->gi_grname
, ipst
)) == NULL
) {
12482 rw_exit(&ipst
->ips_ipmp_lock
);
12485 ipmp_grp_info(grp
, lifgr
);
12486 rw_exit(&ipst
->ips_ipmp_lock
);
12491 ill_dl_down(ill_t
*ill
)
12493 DTRACE_PROBE2(ill__downup
, char *, "ill_dl_down", ill_t
*, ill
);
12496 * The ill is down; unbind but stay attached since we're still
12497 * associated with a PPA. If we have negotiated DLPI capabilites
12498 * with the data link service provider (IDS_OK) then reset them.
12499 * The interval between unbinding and rebinding is potentially
12500 * unbounded hence we cannot assume things will be the same.
12501 * The DLPI capabilities will be probed again when the data link
12504 mblk_t
*mp
= ill
->ill_unbind_mp
;
12506 ip1dbg(("ill_dl_down(%s)\n", ill
->ill_name
));
12508 if (!ill
->ill_replumbing
) {
12509 /* Free all ilms for this ill */
12510 update_conn_ill(ill
, ill
->ill_ipst
);
12512 ill_leave_multicast(ill
);
12515 ill
->ill_unbind_mp
= NULL
;
12517 ip1dbg(("ill_dl_down: %s (%u) for %s\n",
12518 dl_primstr(*(int *)mp
->b_rptr
), *(int *)mp
->b_rptr
,
12520 mutex_enter(&ill
->ill_lock
);
12521 ill
->ill_state_flags
|= ILL_DL_UNBIND_IN_PROGRESS
;
12522 mutex_exit(&ill
->ill_lock
);
12524 * ip_rput does not pass up normal (M_PROTO) DLPI messages
12525 * after ILL_CONDEMNED is set. So in the unplumb case, we call
12526 * ill_capability_dld_disable disable rightaway. If this is not
12527 * an unplumb operation then the disable happens on receipt of
12528 * the capab ack via ip_rput_dlpi_writer ->
12529 * ill_capability_ack_thr. In both cases the order of
12530 * the operations seen by DLD is capability disable followed
12531 * by DL_UNBIND. Also the DLD capability disable needs a
12532 * cv_wait'able context.
12534 if (ill
->ill_state_flags
& ILL_CONDEMNED
)
12535 ill_capability_dld_disable(ill
);
12536 ill_capability_reset(ill
, B_FALSE
);
12537 ill_dlpi_send(ill
, mp
);
12539 mutex_enter(&ill
->ill_lock
);
12540 ill
->ill_dl_up
= 0;
12541 ill_nic_event_dispatch(ill
, 0, NE_DOWN
, NULL
, 0);
12542 mutex_exit(&ill
->ill_lock
);
12546 ill_dlpi_dispatch(ill_t
*ill
, mblk_t
*mp
)
12548 union DL_primitives
*dlp
;
12550 boolean_t waitack
= B_FALSE
;
12552 ASSERT(DB_TYPE(mp
) == M_PROTO
|| DB_TYPE(mp
) == M_PCPROTO
);
12554 dlp
= (union DL_primitives
*)mp
->b_rptr
;
12555 prim
= dlp
->dl_primitive
;
12557 ip1dbg(("ill_dlpi_dispatch: sending %s (%u) to %s\n",
12558 dl_primstr(prim
), prim
, ill
->ill_name
));
12561 case DL_PHYS_ADDR_REQ
:
12563 dl_phys_addr_req_t
*dlpap
= (dl_phys_addr_req_t
*)mp
->b_rptr
;
12564 ill
->ill_phys_addr_pend
= dlpap
->dl_addr_type
;
12568 mutex_enter(&ill
->ill_lock
);
12569 ill
->ill_state_flags
&= ~ILL_DL_UNBIND_IN_PROGRESS
;
12570 mutex_exit(&ill
->ill_lock
);
12575 * Except for the ACKs for the M_PCPROTO messages, all other ACKs
12576 * are dropped by ip_rput() if ILL_CONDEMNED is set. Therefore
12577 * we only wait for the ACK of the DL_UNBIND_REQ.
12579 mutex_enter(&ill
->ill_lock
);
12580 if (!(ill
->ill_state_flags
& ILL_CONDEMNED
) ||
12581 (prim
== DL_UNBIND_REQ
)) {
12582 ill
->ill_dlpi_pending
= prim
;
12586 mutex_exit(&ill
->ill_lock
);
12587 DTRACE_PROBE3(ill__dlpi
, char *, "ill_dlpi_dispatch",
12588 char *, dl_primstr(prim
), ill_t
*, ill
);
12589 putnext(ill
->ill_wq
, mp
);
12592 * There is no ack for DL_NOTIFY_CONF messages
12594 if (waitack
&& prim
== DL_NOTIFY_CONF
)
12595 ill_dlpi_done(ill
, prim
);
12599 * Helper function for ill_dlpi_send().
12603 ill_dlpi_send_writer(ipsq_t
*ipsq
, queue_t
*q
, mblk_t
*mp
, void *arg
)
12605 ill_dlpi_send(q
->q_ptr
, mp
);
12609 * Send a DLPI control message to the driver but make sure there
12610 * is only one outstanding message. Uses ill_dlpi_pending to tell
12611 * when it must queue. ip_rput_dlpi_writer calls ill_dlpi_done()
12612 * when an ACK or a NAK is received to process the next queued message.
12615 ill_dlpi_send(ill_t
*ill
, mblk_t
*mp
)
12619 ASSERT(DB_TYPE(mp
) == M_PROTO
|| DB_TYPE(mp
) == M_PCPROTO
);
12622 * To ensure that any DLPI requests for current exclusive operation
12623 * are always completely sent before any DLPI messages for other
12624 * operations, require writer access before enqueuing.
12626 if (!IAM_WRITER_ILL(ill
)) {
12628 /* qwriter_ip() does the ill_refrele() */
12629 qwriter_ip(ill
, ill
->ill_wq
, mp
, ill_dlpi_send_writer
,
12634 mutex_enter(&ill
->ill_lock
);
12635 if (ill
->ill_dlpi_pending
!= DL_PRIM_INVAL
) {
12636 /* Must queue message. Tail insertion */
12637 mpp
= &ill
->ill_dlpi_deferred
;
12638 while (*mpp
!= NULL
)
12639 mpp
= &((*mpp
)->b_next
);
12641 ip1dbg(("ill_dlpi_send: deferring request for %s "
12642 "while %s pending\n", ill
->ill_name
,
12643 dl_primstr(ill
->ill_dlpi_pending
)));
12646 mutex_exit(&ill
->ill_lock
);
12649 mutex_exit(&ill
->ill_lock
);
12650 ill_dlpi_dispatch(ill
, mp
);
12654 ill_capability_send(ill_t
*ill
, mblk_t
*mp
)
12656 ill
->ill_capab_pending_cnt
++;
12657 ill_dlpi_send(ill
, mp
);
12661 ill_capability_done(ill_t
*ill
)
12663 ASSERT(ill
->ill_capab_pending_cnt
!= 0);
12665 ill_dlpi_done(ill
, DL_CAPABILITY_REQ
);
12667 ill
->ill_capab_pending_cnt
--;
12668 if (ill
->ill_capab_pending_cnt
== 0 &&
12669 ill
->ill_dlpi_capab_state
== IDCS_OK
)
12670 ill_capability_reset_alloc(ill
);
12674 * Send all deferred DLPI messages without waiting for their ACKs.
12677 ill_dlpi_send_deferred(ill_t
*ill
)
12679 mblk_t
*mp
, *nextmp
;
12682 * Clear ill_dlpi_pending so that the message is not queued in
12685 mutex_enter(&ill
->ill_lock
);
12686 ill
->ill_dlpi_pending
= DL_PRIM_INVAL
;
12687 mp
= ill
->ill_dlpi_deferred
;
12688 ill
->ill_dlpi_deferred
= NULL
;
12689 mutex_exit(&ill
->ill_lock
);
12691 for (; mp
!= NULL
; mp
= nextmp
) {
12692 nextmp
= mp
->b_next
;
12694 ill_dlpi_send(ill
, mp
);
12699 * Clear all the deferred DLPI messages. Called on receiving an M_ERROR
12703 ill_dlpi_clear_deferred(ill_t
*ill
)
12705 mblk_t
*mp
, *nextmp
;
12707 mutex_enter(&ill
->ill_lock
);
12708 ill
->ill_dlpi_pending
= DL_PRIM_INVAL
;
12709 mp
= ill
->ill_dlpi_deferred
;
12710 ill
->ill_dlpi_deferred
= NULL
;
12711 mutex_exit(&ill
->ill_lock
);
12713 for (; mp
!= NULL
; mp
= nextmp
) {
12714 nextmp
= mp
->b_next
;
12720 * Check if the DLPI primitive `prim' is pending; print a warning if not.
12723 ill_dlpi_pending(ill_t
*ill
, t_uscalar_t prim
)
12725 t_uscalar_t pending
;
12727 mutex_enter(&ill
->ill_lock
);
12728 if (ill
->ill_dlpi_pending
== prim
) {
12729 mutex_exit(&ill
->ill_lock
);
12734 * During teardown, ill_dlpi_dispatch() will send DLPI requests
12735 * without waiting, so don't print any warnings in that case.
12737 if (ill
->ill_state_flags
& ILL_CONDEMNED
) {
12738 mutex_exit(&ill
->ill_lock
);
12741 pending
= ill
->ill_dlpi_pending
;
12742 mutex_exit(&ill
->ill_lock
);
12744 if (pending
== DL_PRIM_INVAL
) {
12745 (void) mi_strlog(ill
->ill_rq
, 1, SL_CONSOLE
|SL_ERROR
|SL_TRACE
,
12746 "received unsolicited ack for %s on %s\n",
12747 dl_primstr(prim
), ill
->ill_name
);
12749 (void) mi_strlog(ill
->ill_rq
, 1, SL_CONSOLE
|SL_ERROR
|SL_TRACE
,
12750 "received unexpected ack for %s on %s (expecting %s)\n",
12751 dl_primstr(prim
), ill
->ill_name
, dl_primstr(pending
));
12757 * Complete the current DLPI operation associated with `prim' on `ill' and
12758 * start the next queued DLPI operation (if any). If there are no queued DLPI
12759 * operations and the ill's current exclusive IPSQ operation has finished
12760 * (i.e., ipsq_current_finish() was called), then clear ipsq_current_ipif to
12761 * allow the next exclusive IPSQ operation to begin upon ipsq_exit(). See
12762 * the comments above ipsq_current_finish() for details.
12765 ill_dlpi_done(ill_t
*ill
, t_uscalar_t prim
)
12768 ipsq_t
*ipsq
= ill
->ill_phyint
->phyint_ipsq
;
12769 ipxop_t
*ipx
= ipsq
->ipsq_xop
;
12771 ASSERT(IAM_WRITER_IPSQ(ipsq
));
12772 mutex_enter(&ill
->ill_lock
);
12774 ASSERT(prim
!= DL_PRIM_INVAL
);
12775 ASSERT(ill
->ill_dlpi_pending
== prim
);
12777 ip1dbg(("ill_dlpi_done: %s has completed %s (%u)\n", ill
->ill_name
,
12778 dl_primstr(ill
->ill_dlpi_pending
), ill
->ill_dlpi_pending
));
12780 if ((mp
= ill
->ill_dlpi_deferred
) == NULL
) {
12781 ill
->ill_dlpi_pending
= DL_PRIM_INVAL
;
12782 if (ipx
->ipx_current_done
) {
12783 mutex_enter(&ipx
->ipx_lock
);
12784 ipx
->ipx_current_ipif
= NULL
;
12785 mutex_exit(&ipx
->ipx_lock
);
12787 cv_signal(&ill
->ill_cv
);
12788 mutex_exit(&ill
->ill_lock
);
12792 ill
->ill_dlpi_deferred
= mp
->b_next
;
12794 mutex_exit(&ill
->ill_lock
);
12796 ill_dlpi_dispatch(ill
, mp
);
12800 * Queue a (multicast) DLPI control message to be sent to the driver by
12801 * later calling ill_dlpi_send_queued.
12802 * We queue them while holding a lock (ill_mcast_lock) to ensure that they
12803 * are sent in order i.e., prevent a DL_DISABMULTI_REQ and DL_ENABMULTI_REQ
12804 * for the same group to race.
12805 * We send DLPI control messages in order using ill_lock.
12806 * For IPMP we should be called on the cast_ill.
12809 ill_dlpi_queue(ill_t
*ill
, mblk_t
*mp
)
12813 ASSERT(DB_TYPE(mp
) == M_PROTO
|| DB_TYPE(mp
) == M_PCPROTO
);
12815 mutex_enter(&ill
->ill_lock
);
12816 /* Must queue message. Tail insertion */
12817 mpp
= &ill
->ill_dlpi_deferred
;
12818 while (*mpp
!= NULL
)
12819 mpp
= &((*mpp
)->b_next
);
12822 mutex_exit(&ill
->ill_lock
);
12826 * Send the messages that were queued. Make sure there is only
12827 * one outstanding message. ip_rput_dlpi_writer calls ill_dlpi_done()
12828 * when an ACK or a NAK is received to process the next queued message.
12829 * For IPMP we are called on the upper ill, but when send what is queued
12833 ill_dlpi_send_queued(ill_t
*ill
)
12836 union DL_primitives
*dlp
;
12838 ill_t
*release_ill
= NULL
;
12840 if (IS_IPMP(ill
)) {
12841 /* On the upper IPMP ill. */
12842 release_ill
= ipmp_illgrp_hold_cast_ill(ill
->ill_grp
);
12843 if (release_ill
== NULL
) {
12844 /* Avoid ever sending anything down to the ipmpstub */
12849 mutex_enter(&ill
->ill_lock
);
12850 while ((mp
= ill
->ill_dlpi_deferred
) != NULL
) {
12851 if (ill
->ill_dlpi_pending
!= DL_PRIM_INVAL
) {
12852 /* Can't send. Somebody else will send it */
12853 mutex_exit(&ill
->ill_lock
);
12856 ill
->ill_dlpi_deferred
= mp
->b_next
;
12858 if (!ill
->ill_dl_up
) {
12860 * Nobody there. All multicast addresses will be
12861 * re-joined when we get the DL_BIND_ACK bringing the
12867 dlp
= (union DL_primitives
*)mp
->b_rptr
;
12868 prim
= dlp
->dl_primitive
;
12870 if (!(ill
->ill_state_flags
& ILL_CONDEMNED
) ||
12871 (prim
== DL_UNBIND_REQ
)) {
12872 ill
->ill_dlpi_pending
= prim
;
12874 mutex_exit(&ill
->ill_lock
);
12876 DTRACE_PROBE3(ill__dlpi
, char *, "ill_dlpi_send_queued",
12877 char *, dl_primstr(prim
), ill_t
*, ill
);
12878 putnext(ill
->ill_wq
, mp
);
12879 mutex_enter(&ill
->ill_lock
);
12881 mutex_exit(&ill
->ill_lock
);
12883 if (release_ill
!= NULL
)
12884 ill_refrele(release_ill
);
12888 * Queue an IP (IGMP/MLD) message to be sent by IP from
12889 * ill_mcast_send_queued
12890 * We queue them while holding a lock (ill_mcast_lock) to ensure that they
12891 * are sent in order i.e., prevent a IGMP leave and IGMP join for the same
12893 * We send them in order using ill_lock.
12894 * For IPMP we are called on the upper ill, but we queue on the cast_ill.
12897 ill_mcast_queue(ill_t
*ill
, mblk_t
*mp
)
12900 ill_t
*release_ill
= NULL
;
12902 ASSERT(RW_LOCK_HELD(&ill
->ill_mcast_lock
));
12904 if (IS_IPMP(ill
)) {
12905 /* On the upper IPMP ill. */
12906 release_ill
= ipmp_illgrp_hold_cast_ill(ill
->ill_grp
);
12907 if (release_ill
== NULL
) {
12908 /* Discard instead of queuing for the ipmp interface */
12909 BUMP_MIB(ill
->ill_ip_mib
, ipIfStatsOutDiscards
);
12910 ip_drop_output("ipIfStatsOutDiscards - no cast_ill",
12918 mutex_enter(&ill
->ill_lock
);
12919 /* Must queue message. Tail insertion */
12920 mpp
= &ill
->ill_mcast_deferred
;
12921 while (*mpp
!= NULL
)
12922 mpp
= &((*mpp
)->b_next
);
12925 mutex_exit(&ill
->ill_lock
);
12926 if (release_ill
!= NULL
)
12927 ill_refrele(release_ill
);
12931 * Send the IP packets that were queued by ill_mcast_queue.
12932 * These are IGMP/MLD packets.
12934 * For IPMP we are called on the upper ill, but when send what is queued
12937 * Request loopback of the report if we are acting as a multicast
12938 * router, so that the process-level routing demon can hear it.
12939 * This will run multiple times for the same group if there are members
12940 * on the same group for multiple ipif's on the same ill. The
12941 * igmp_input/mld_input code will suppress this due to the loopback thus we
12942 * always loopback membership report.
12944 * We also need to make sure that this does not get load balanced
12945 * by IPMP. We do this by passing an ill to ip_output_simple.
12948 ill_mcast_send_queued(ill_t
*ill
)
12951 ip_xmit_attr_t ixas
;
12952 ill_t
*release_ill
= NULL
;
12954 if (IS_IPMP(ill
)) {
12955 /* On the upper IPMP ill. */
12956 release_ill
= ipmp_illgrp_hold_cast_ill(ill
->ill_grp
);
12957 if (release_ill
== NULL
) {
12959 * We should have no messages on the ipmp interface
12960 * but no point in trying to send them.
12966 bzero(&ixas
, sizeof (ixas
));
12967 ixas
.ixa_zoneid
= ALL_ZONES
;
12968 ixas
.ixa_cred
= kcred
;
12969 ixas
.ixa_cpid
= NOPID
;
12971 * Here we set ixa_ifindex. If IPMP it will be the lower ill which
12972 * makes ip_select_route pick the IRE_MULTICAST for the cast_ill.
12973 * That is necessary to handle IGMP/MLD snooping switches.
12975 ixas
.ixa_ifindex
= ill
->ill_phyint
->phyint_ifindex
;
12976 ixas
.ixa_ipst
= ill
->ill_ipst
;
12978 mutex_enter(&ill
->ill_lock
);
12979 while ((mp
= ill
->ill_mcast_deferred
) != NULL
) {
12980 ill
->ill_mcast_deferred
= mp
->b_next
;
12982 if (!ill
->ill_dl_up
) {
12984 * Nobody there. Just drop the ip packets.
12985 * IGMP/MLD will resend later, if this is a replumb.
12990 mutex_enter(&ill
->ill_phyint
->phyint_lock
);
12991 if (IS_UNDER_IPMP(ill
) && !ipmp_ill_is_active(ill
)) {
12993 * When the ill is getting deactivated, we only want to
12994 * send the DLPI messages, so drop IGMP/MLD packets.
12995 * DLPI messages are handled by ill_dlpi_send_queued()
12997 mutex_exit(&ill
->ill_phyint
->phyint_lock
);
13001 mutex_exit(&ill
->ill_phyint
->phyint_lock
);
13002 mutex_exit(&ill
->ill_lock
);
13004 /* Check whether we are sending IPv4 or IPv6. */
13005 if (ill
->ill_isv6
) {
13006 ip6_t
*ip6h
= (ip6_t
*)mp
->b_rptr
;
13008 ixas
.ixa_multicast_ttl
= ip6h
->ip6_hops
;
13009 ixas
.ixa_flags
= IXAF_BASIC_SIMPLE_V6
;
13011 ipha_t
*ipha
= (ipha_t
*)mp
->b_rptr
;
13013 ixas
.ixa_multicast_ttl
= ipha
->ipha_ttl
;
13014 ixas
.ixa_flags
= IXAF_BASIC_SIMPLE_V4
;
13015 ixas
.ixa_flags
&= ~IXAF_SET_ULP_CKSUM
;
13017 ixas
.ixa_flags
&= ~IXAF_VERIFY_SOURCE
;
13018 ixas
.ixa_flags
|= IXAF_MULTICAST_LOOP
| IXAF_SET_SOURCE
;
13019 (void) ip_output_simple(mp
, &ixas
);
13020 ixa_cleanup(&ixas
);
13022 mutex_enter(&ill
->ill_lock
);
13024 mutex_exit(&ill
->ill_lock
);
13027 if (release_ill
!= NULL
)
13028 ill_refrele(release_ill
);
13032 * Take down a specific interface, but don't lose any information about it.
13033 * (Always called as writer.)
13034 * This function goes through the down sequence even if the interface is
13035 * already down. There are 2 reasons.
13036 * a. Currently we permit interface routes that depend on down interfaces
13037 * to be added. This behaviour itself is questionable. However it appears
13038 * that both Solaris and 4.3 BSD have exhibited this behaviour for a long
13039 * time. We go thru the cleanup in order to remove these routes.
13040 * b. The bringup of the interface could fail in ill_dl_up i.e. we get
13041 * DL_ERROR_ACK in response to the DL_BIND request. The interface is
13042 * down, but we need to cleanup i.e. do ill_dl_down and
13043 * ip_rput_dlpi_writer (DL_ERROR_ACK) -> ipif_down.
13047 * Model of reference to interfaces.
13049 * The following members in ipif_t track references to the ipif.
13050 * int ipif_refcnt; Active reference count
13052 * The following members in ill_t track references to the ill.
13053 * int ill_refcnt; active refcnt
13054 * uint_t ill_ire_cnt; Number of ires referencing ill
13055 * uint_t ill_ncec_cnt; Number of ncecs referencing ill
13056 * uint_t ill_nce_cnt; Number of nces referencing ill
13057 * uint_t ill_ilm_cnt; Number of ilms referencing ill
13059 * Reference to an ipif or ill can be obtained in any of the following ways.
13061 * Through the lookup functions ipif_lookup_* / ill_lookup_* functions
13062 * Pointers to ipif / ill from other data structures viz ire and conn.
13063 * Implicit reference to the ipif / ill by holding a reference to the ire.
13065 * The ipif/ill lookup functions return a reference held ipif / ill.
13066 * ipif_refcnt and ill_refcnt track the reference counts respectively.
13067 * This is a purely dynamic reference count associated with threads holding
13068 * references to the ipif / ill. Pointers from other structures do not
13069 * count towards this reference count.
13071 * ill_ire_cnt is the number of ire's associated with the
13072 * ill. This is incremented whenever a new ire is created referencing the
13073 * ill. This is done atomically inside ire_add_v[46] where the ire is
13074 * actually added to the ire hash table. The count is decremented in
13075 * ire_inactive where the ire is destroyed.
13077 * ill_ncec_cnt is the number of ncec's referencing the ill thru ncec_ill.
13078 * This is incremented atomically in
13079 * ndp_add_v4()/ndp_add_v6() where the nce is actually added to the
13080 * table. Similarly it is decremented in ncec_inactive() where the ncec
13083 * ill_nce_cnt is the number of nce's referencing the ill thru nce_ill. This is
13084 * incremented atomically in nce_add() where the nce is actually added to the
13085 * ill_nce. Similarly it is decremented in nce_inactive() where the nce
13088 * ill_ilm_cnt is the ilm's reference to the ill. It is incremented in
13089 * ilm_add() and decremented before the ilm is freed in ilm_delete().
13091 * Flow of ioctls involving interface down/up
13093 * The following is the sequence of an attempt to set some critical flags on an
13097 * wait for ipif to be quiescent
13099 * ip_sioctl_flags_tail
13101 * All set ioctls that involve down/up sequence would have a skeleton similar
13102 * to the above. All the *tail functions are called after the refcounts have
13103 * dropped to the appropriate values.
13105 * SIOC ioctls during the IPIF_CHANGING interval.
13107 * Threads handling SIOC set ioctls serialize on the squeue, but this
13108 * is not done for SIOC get ioctls. Since a set ioctl can cause several
13109 * steps of internal changes to the state, some of which are visible in
13110 * ipif_flags (such as IFF_UP being cleared and later set), and we want
13111 * the set ioctl to be atomic related to the get ioctls, the SIOC get code
13112 * will wait and restart ioctls if IPIF_CHANGING is set. The mblk is then
13113 * enqueued in the ipsq and the operation is restarted by ipsq_exit() when
13114 * the current exclusive operation completes. The IPIF_CHANGING check
13115 * and enqueue is atomic using the ill_lock and ipsq_lock. The
13116 * lookup is done holding the ill_lock. Hence the ill/ipif state flags can't
13117 * change while the ill_lock is held. Before dropping the ill_lock we acquire
13118 * the ipsq_lock and call ipsq_enq. This ensures that ipsq_exit can't finish
13119 * until we release the ipsq_lock, even though the ill/ipif state flags
13120 * can change after we drop the ill_lock.
13123 ipif_down(ipif_t
*ipif
, queue_t
*q
, mblk_t
*mp
)
13125 ill_t
*ill
= ipif
->ipif_ill
;
13128 boolean_t ipif_was_up
= B_FALSE
;
13129 ip_stack_t
*ipst
= ill
->ill_ipst
;
13131 ASSERT(IAM_WRITER_IPIF(ipif
));
13133 ip1dbg(("ipif_down(%s:%u)\n", ill
->ill_name
, ipif
->ipif_id
));
13135 DTRACE_PROBE3(ipif__downup
, char *, "ipif_down",
13136 ill_t
*, ill
, ipif_t
*, ipif
);
13138 if (ipif
->ipif_flags
& IPIF_UP
) {
13139 mutex_enter(&ill
->ill_lock
);
13140 ipif
->ipif_flags
&= ~IPIF_UP
;
13141 ASSERT(ill
->ill_ipif_up_count
> 0);
13142 --ill
->ill_ipif_up_count
;
13143 mutex_exit(&ill
->ill_lock
);
13144 ipif_was_up
= B_TRUE
;
13145 /* Update status in SCTP's list */
13146 sctp_update_ipif(ipif
, SCTP_IPIF_DOWN
);
13147 ill_nic_event_dispatch(ipif
->ipif_ill
,
13148 MAP_IPIF_ID(ipif
->ipif_id
), NE_LIF_DOWN
, NULL
, 0);
13152 * Removal of the last ipif from an ill may result in a DL_UNBIND
13153 * being sent to the driver, and we must not send any data packets to
13154 * the driver after the DL_UNBIND_REQ. To ensure this, all the
13155 * ire and nce entries used in the data path will be cleaned
13156 * up, and we also set the ILL_DOWN_IN_PROGRESS bit to make
13157 * sure on new entries will be added until the ill is bound
13158 * again. The ILL_DOWN_IN_PROGRESS bit is turned off upon
13159 * receipt of a DL_BIND_ACK.
13161 if (ill
->ill_wq
!= NULL
&& !ill
->ill_logical_down
&&
13162 ill
->ill_ipif_up_count
== 0 && ill
->ill_ipif_dup_count
== 0 &&
13164 ill
->ill_state_flags
|= ILL_DOWN_IN_PROGRESS
;
13168 * Blow away memberships we established in ipif_multicast_up().
13170 ipif_multicast_down(ipif
);
13173 * Remove from the mapping for __sin6_src_id. We insert only
13174 * when the address is not INADDR_ANY. As IPv4 addresses are
13175 * stored as mapped addresses, we need to check for mapped
13178 if (ipif_was_up
&& !IN6_IS_ADDR_UNSPECIFIED(&ipif
->ipif_v6lcl_addr
) &&
13179 !IN6_IS_ADDR_V4MAPPED_ANY(&ipif
->ipif_v6lcl_addr
) &&
13180 !(ipif
->ipif_flags
& IPIF_NOLOCAL
)) {
13183 err
= ip_srcid_remove(&ipif
->ipif_v6lcl_addr
,
13184 ipif
->ipif_zoneid
, ipst
);
13186 ip0dbg(("ipif_down: srcid_remove %d\n", err
));
13191 /* only delete if we'd added ire's before */
13192 if (ipif
->ipif_isv6
)
13193 ipif_delete_ires_v6(ipif
);
13195 ipif_delete_ires_v4(ipif
);
13198 if (ipif_was_up
&& ill
->ill_ipif_up_count
== 0) {
13200 * Since the interface is now down, it may have just become
13201 * inactive. Note that this needs to be done even for a
13202 * lll_logical_down(), or ARP entries will not get correctly
13203 * restored when the interface comes back up.
13205 if (IS_UNDER_IPMP(ill
))
13206 ipmp_ill_refresh_active(ill
);
13210 * neighbor-discovery or arp entries for this interface. The ipif
13211 * has to be quiesced, so we walk all the nce's and delete those
13212 * that point at the ipif->ipif_ill. At the same time, we also
13213 * update IPMP so that ipifs for data addresses are unbound. We dont
13214 * call ipif_arp_down to DL_UNBIND the arp stream itself here, but defer
13215 * that for ipif_down_tail()
13217 ipif_nce_down(ipif
);
13220 * If this is the last ipif on the ill, we also need to remove
13221 * any IREs with ire_ill set. Otherwise ipif_is_quiescent() will
13224 if (ill
->ill_ipif_up_count
== 0 && ill
->ill_ipif_dup_count
== 0)
13225 ire_walk_ill(0, 0, ill_downi
, ill
, ill
);
13228 * Walk all CONNs that can have a reference on an ire for this
13229 * ipif (we actually walk all that now have stale references).
13231 ipcl_walk(conn_ixa_cleanup
, (void *)B_TRUE
, ipst
);
13234 * If mp is NULL the caller will wait for the appropriate refcnt.
13235 * Eg. ip_sioctl_removeif -> ipif_free -> ipif_down
13236 * and ill_delete -> ipif_free -> ipif_down
13244 connp
= Q_TO_CONN(q
);
13245 mutex_enter(&connp
->conn_lock
);
13249 mutex_enter(&ill
->ill_lock
);
13251 * Are there any ire's pointing to this ipif that are still active ?
13252 * If this is the last ipif going down, are there any ire's pointing
13253 * to this ill that are still active ?
13255 if (ipif_is_quiescent(ipif
)) {
13256 mutex_exit(&ill
->ill_lock
);
13258 mutex_exit(&connp
->conn_lock
);
13262 ip1dbg(("ipif_down: need to wait, adding pending mp %s ill %p",
13263 ill
->ill_name
, (void *)ill
));
13265 * Enqueue the mp atomically in ipsq_pending_mp. When the refcount
13266 * drops down, the operation will be restarted by ipif_ill_refrele_tail
13267 * which in turn is called by the last refrele on the ipif/ill/ire.
13269 success
= ipsq_pending_mp_add(connp
, ipif
, q
, mp
, IPIF_DOWN
);
13271 /* The conn is closing. So just return */
13272 ASSERT(connp
!= NULL
);
13273 mutex_exit(&ill
->ill_lock
);
13274 mutex_exit(&connp
->conn_lock
);
13278 mutex_exit(&ill
->ill_lock
);
13280 mutex_exit(&connp
->conn_lock
);
13281 return (EINPROGRESS
);
13285 ipif_down_tail(ipif_t
*ipif
)
13287 ill_t
*ill
= ipif
->ipif_ill
;
13290 DTRACE_PROBE3(ipif__downup
, char *, "ipif_down_tail",
13291 ill_t
*, ill
, ipif_t
*, ipif
);
13294 * Skip any loopback interface (null wq).
13295 * If this is the last logical interface on the ill
13296 * have ill_dl_down tell the driver we are gone (unbind)
13297 * Note that lun 0 can ipif_down even though
13298 * there are other logical units that are up.
13299 * This occurs e.g. when we change a "significant" IFF_ flag.
13301 if (ill
->ill_wq
!= NULL
&& !ill
->ill_logical_down
&&
13302 ill
->ill_ipif_up_count
== 0 && ill
->ill_ipif_dup_count
== 0 &&
13306 if (!ipif
->ipif_isv6
)
13307 err
= ipif_arp_down(ipif
);
13309 ill
->ill_logical_down
= 0;
13311 ip_rts_ifmsg(ipif
, RTSQ_DEFAULT
);
13312 ip_rts_newaddrmsg(RTM_DELETE
, 0, ipif
, RTSQ_DEFAULT
);
13317 * Bring interface logically down without bringing the physical interface
13318 * down e.g. when the netmask is changed. This avoids long lasting link
13319 * negotiations between an ethernet interface and a certain switches.
13322 ipif_logical_down(ipif_t
*ipif
, queue_t
*q
, mblk_t
*mp
)
13324 DTRACE_PROBE3(ipif__downup
, char *, "ipif_logical_down",
13325 ill_t
*, ipif
->ipif_ill
, ipif_t
*, ipif
);
13328 * The ill_logical_down flag is a transient flag. It is set here
13329 * and is cleared once the down has completed in ipif_down_tail.
13330 * This flag does not indicate whether the ill stream is in the
13331 * DL_BOUND state with the driver. Instead this flag is used by
13332 * ipif_down_tail to determine whether to DL_UNBIND the stream with
13333 * the driver. The state of the ill stream i.e. whether it is
13334 * DL_BOUND with the driver or not is indicated by the ill_dl_up flag.
13336 ipif
->ipif_ill
->ill_logical_down
= 1;
13337 return (ipif_down(ipif
, q
, mp
));
13341 * Initiate deallocate of an IPIF. Always called as writer. Called by
13342 * ill_delete or ip_sioctl_removeif.
13345 ipif_free(ipif_t
*ipif
)
13347 ip_stack_t
*ipst
= ipif
->ipif_ill
->ill_ipst
;
13349 ASSERT(IAM_WRITER_IPIF(ipif
));
13351 if (ipif
->ipif_recovery_id
!= 0)
13352 (void) untimeout(ipif
->ipif_recovery_id
);
13353 ipif
->ipif_recovery_id
= 0;
13356 * Take down the interface. We can be called either from ill_delete
13357 * or from ip_sioctl_removeif.
13359 (void) ipif_down(ipif
, NULL
, NULL
);
13362 * Now that the interface is down, there's no chance it can still
13363 * become a duplicate. Cancel any timer that may have been set while
13366 if (ipif
->ipif_recovery_id
!= 0)
13367 (void) untimeout(ipif
->ipif_recovery_id
);
13368 ipif
->ipif_recovery_id
= 0;
13370 rw_enter(&ipst
->ips_ill_g_lock
, RW_WRITER
);
13371 /* Remove pointers to this ill in the multicast routing tables */
13372 reset_mrt_vif_ipif(ipif
);
13373 /* If necessary, clear the cached source ipif rotor. */
13374 if (ipif
->ipif_ill
->ill_src_ipif
== ipif
)
13375 ipif
->ipif_ill
->ill_src_ipif
= NULL
;
13376 rw_exit(&ipst
->ips_ill_g_lock
);
13380 ipif_free_tail(ipif_t
*ipif
)
13382 ip_stack_t
*ipst
= ipif
->ipif_ill
->ill_ipst
;
13385 * Need to hold both ill_g_lock and ill_lock while
13386 * inserting or removing an ipif from the linked list
13387 * of ipifs hanging off the ill.
13389 rw_enter(&ipst
->ips_ill_g_lock
, RW_WRITER
);
13392 ipif_trace_cleanup(ipif
);
13395 /* Ask SCTP to take it out of it list */
13396 sctp_update_ipif(ipif
, SCTP_IPIF_REMOVE
);
13397 ip_rts_newaddrmsg(RTM_FREEADDR
, 0, ipif
, RTSQ_DEFAULT
);
13399 /* Get it out of the ILL interface list. */
13401 rw_exit(&ipst
->ips_ill_g_lock
);
13403 ASSERT(!(ipif
->ipif_flags
& (IPIF_UP
| IPIF_DUPLICATE
)));
13404 ASSERT(ipif
->ipif_recovery_id
== 0);
13405 ASSERT(ipif
->ipif_ire_local
== NULL
);
13406 ASSERT(ipif
->ipif_ire_if
== NULL
);
13408 /* Free the memory. */
13413 * Sets `buf' to an ipif name of the form "ill_name:id", or "ill_name" if "id"
13417 ipif_get_name(const ipif_t
*ipif
, char *buf
, int len
)
13419 char lbuf
[LIFNAMSIZ
];
13424 name
= ipif
->ipif_ill
->ill_name
;
13425 name_len
= ipif
->ipif_ill
->ill_name_length
;
13426 if (ipif
->ipif_id
!= 0) {
13427 (void) sprintf(lbuf
, "%s%c%d", name
, IPIF_SEPARATOR_CHAR
,
13430 name_len
= mi_strlen(name
) + 1;
13434 len
= MIN(len
, name_len
);
13435 bcopy(name
, buf
, len
);
13439 * Sets `buf' to an ill name.
13442 ill_get_name(const ill_t
*ill
, char *buf
, int len
)
13447 name
= ill
->ill_name
;
13448 name_len
= ill
->ill_name_length
;
13451 len
= MIN(len
, name_len
);
13452 bcopy(name
, buf
, len
);
13456 * Find an IPIF based on the name passed in. Names can be of the form <phys>
13457 * (e.g., le0) or <phys>:<#> (e.g., le0:1). When there is no colon, the
13458 * implied unit id is zero. <phys> must correspond to the name of an ILL.
13459 * (May be called as writer.)
13462 ipif_lookup_on_name(char *name
, size_t namelen
, boolean_t do_alloc
,
13463 boolean_t
*exists
, boolean_t isv6
, zoneid_t zoneid
, ip_stack_t
*ipst
)
13471 boolean_t did_alloc
= B_FALSE
;
13475 * If the caller wants to us to create the ipif, make sure we have a
13478 ASSERT(!do_alloc
|| zoneid
!= ALL_ZONES
);
13480 if (namelen
== 0) {
13485 /* Look for a colon in the name. */
13486 endp
= &name
[namelen
];
13487 for (cp
= endp
; --cp
> name
; ) {
13488 if (*cp
== IPIF_SEPARATOR_CHAR
)
13492 if (*cp
== IPIF_SEPARATOR_CHAR
) {
13494 * Reject any non-decimal aliases for logical
13495 * interfaces. Aliases with leading zeroes
13496 * are also rejected as they introduce ambiguity
13497 * in the naming of the interfaces.
13498 * In order to confirm with existing semantics,
13499 * and to not break any programs/script relying
13500 * on that behaviour, if<0>:0 is considered to be
13501 * a valid interface.
13503 * If alias has two or more digits and the first
13506 if (&cp
[2] < endp
&& cp
[1] == '0') {
13518 * Look up the ILL, based on the portion of the name
13519 * before the slash. ill_lookup_on_name returns a held ill.
13520 * Temporary to check whether ill exists already. If so
13521 * ill_lookup_on_name will clear it.
13523 ill
= ill_lookup_on_name(name
, do_alloc
, isv6
,
13529 /* Establish the unit number in the name. */
13531 if (cp
< endp
&& *endp
== '\0') {
13532 /* If there was a colon, the unit number follows. */
13534 if (ddi_strtol(cp
, NULL
, 0, &id
) != 0) {
13540 mutex_enter(&ill
->ill_lock
);
13541 /* Now see if there is an IPIF with this unit number. */
13542 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
; ipif
= ipif
->ipif_next
) {
13543 if (ipif
->ipif_id
== id
) {
13544 if (zoneid
!= ALL_ZONES
&&
13545 zoneid
!= ipif
->ipif_zoneid
&&
13546 ipif
->ipif_zoneid
!= ALL_ZONES
) {
13547 mutex_exit(&ill
->ill_lock
);
13551 if (IPIF_CAN_LOOKUP(ipif
)) {
13552 ipif_refhold_locked(ipif
);
13553 mutex_exit(&ill
->ill_lock
);
13557 * Drop locks before calling ill_refrele
13558 * since it can potentially call into
13559 * ipif_ill_refrele_tail which can end up
13560 * in trying to acquire any lock.
13569 mutex_exit(&ill
->ill_lock
);
13575 * If none found, atomically allocate and return a new one.
13576 * Historically, we used IRE_LOOPBACK only for lun 0, and IRE_LOCAL
13577 * to support "receive only" use of lo0:1 etc. as is still done
13578 * below as an initial guess.
13579 * However, this is now likely to be overriden later in ipif_up_done()
13580 * when we know for sure what address has been configured on the
13581 * interface, since we might have more than one loopback interface
13582 * with a loopback address, e.g. in the case of zones, and all the
13583 * interfaces with loopback addresses need to be marked IRE_LOOPBACK.
13585 if (ill
->ill_net_type
== IRE_LOOPBACK
&& id
== 0)
13586 ire_type
= IRE_LOOPBACK
;
13588 ire_type
= IRE_LOCAL
;
13589 ipif
= ipif_allocate(ill
, id
, ire_type
, B_TRUE
, B_TRUE
, NULL
);
13591 ipif_refhold_locked(ipif
);
13592 mutex_exit(&ill
->ill_lock
);
13598 * Variant of the above that queues the request on the ipsq when
13599 * IPIF_CHANGING is set.
13602 ipif_lookup_on_name_async(char *name
, size_t namelen
, boolean_t isv6
,
13603 zoneid_t zoneid
, queue_t
*q
, mblk_t
*mp
, ipsq_func_t func
, int *error
,
13611 boolean_t did_alloc
= B_FALSE
;
13617 if (namelen
== 0) {
13623 /* Look for a colon in the name. */
13624 endp
= &name
[namelen
];
13625 for (cp
= endp
; --cp
> name
; ) {
13626 if (*cp
== IPIF_SEPARATOR_CHAR
)
13630 if (*cp
== IPIF_SEPARATOR_CHAR
) {
13632 * Reject any non-decimal aliases for logical
13633 * interfaces. Aliases with leading zeroes
13634 * are also rejected as they introduce ambiguity
13635 * in the naming of the interfaces.
13636 * In order to confirm with existing semantics,
13637 * and to not break any programs/script relying
13638 * on that behaviour, if<0>:0 is considered to be
13639 * a valid interface.
13641 * If alias has two or more digits and the first
13644 if (&cp
[2] < endp
&& cp
[1] == '0') {
13658 * Look up the ILL, based on the portion of the name
13659 * before the slash. ill_lookup_on_name returns a held ill.
13660 * Temporary to check whether ill exists already. If so
13661 * ill_lookup_on_name will clear it.
13663 ill
= ill_lookup_on_name(name
, B_FALSE
, isv6
, &did_alloc
, ipst
);
13665 *cp
= IPIF_SEPARATOR_CHAR
;
13669 /* Establish the unit number in the name. */
13671 if (cp
< endp
&& *endp
== '\0') {
13672 /* If there was a colon, the unit number follows. */
13674 if (ddi_strtol(cp
, NULL
, 0, &id
) != 0) {
13683 mutex_enter(&ill
->ill_lock
);
13684 /* Now see if there is an IPIF with this unit number. */
13685 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
; ipif
= ipif
->ipif_next
) {
13686 if (ipif
->ipif_id
== id
) {
13687 if (zoneid
!= ALL_ZONES
&&
13688 zoneid
!= ipif
->ipif_zoneid
&&
13689 ipif
->ipif_zoneid
!= ALL_ZONES
) {
13690 mutex_exit(&ill
->ill_lock
);
13691 RELEASE_CONN_LOCK(q
);
13698 if (!(IPIF_IS_CHANGING(ipif
) ||
13699 IPIF_IS_CONDEMNED(ipif
)) ||
13700 IAM_WRITER_IPIF(ipif
)) {
13701 ipif_refhold_locked(ipif
);
13702 mutex_exit(&ill
->ill_lock
);
13704 * Drop locks before calling ill_refrele
13705 * since it can potentially call into
13706 * ipif_ill_refrele_tail which can end up
13707 * in trying to acquire any lock.
13709 RELEASE_CONN_LOCK(q
);
13712 } else if (q
!= NULL
&& !IPIF_IS_CONDEMNED(ipif
)) {
13713 ipsq
= ill
->ill_phyint
->phyint_ipsq
;
13714 mutex_enter(&ipsq
->ipsq_lock
);
13715 mutex_enter(&ipsq
->ipsq_xop
->ipx_lock
);
13716 mutex_exit(&ill
->ill_lock
);
13717 ipsq_enq(ipsq
, q
, mp
, func
, NEW_OP
, ill
);
13718 mutex_exit(&ipsq
->ipsq_xop
->ipx_lock
);
13719 mutex_exit(&ipsq
->ipsq_lock
);
13720 RELEASE_CONN_LOCK(q
);
13723 *error
= EINPROGRESS
;
13728 RELEASE_CONN_LOCK(q
);
13729 mutex_exit(&ill
->ill_lock
);
13737 * This routine is called whenever a new address comes up on an ipif. If
13738 * we are configured to respond to address mask requests, then we are supposed
13739 * to broadcast an address mask reply at this time. This routine is also
13740 * called if we are already up, but a netmask change is made. This is legal
13741 * but might not make the system manager very popular. (May be called
13745 ipif_mask_reply(ipif_t
*ipif
)
13750 ip_stack_t
*ipst
= ipif
->ipif_ill
->ill_ipst
;
13751 ip_xmit_attr_t ixas
;
13753 #define REPLY_LEN (sizeof (icmp_ipha) + sizeof (icmph_t) + IP_ADDR_LEN)
13755 if (!ipst
->ips_ip_respond_to_address_mask_broadcast
)
13758 /* ICMP mask reply is IPv4 only */
13759 ASSERT(!ipif
->ipif_isv6
);
13760 /* ICMP mask reply is not for a loopback interface */
13761 ASSERT(ipif
->ipif_ill
->ill_wq
!= NULL
);
13763 if (ipif
->ipif_lcl_addr
== INADDR_ANY
)
13766 mp
= allocb(REPLY_LEN
, BPRI_HI
);
13769 mp
->b_wptr
= mp
->b_rptr
+ REPLY_LEN
;
13771 ipha
= (ipha_t
*)mp
->b_rptr
;
13772 bzero(ipha
, REPLY_LEN
);
13774 ipha
->ipha_ttl
= ipst
->ips_ip_broadcast_ttl
;
13775 ipha
->ipha_src
= ipif
->ipif_lcl_addr
;
13776 ipha
->ipha_dst
= ipif
->ipif_brd_addr
;
13777 ipha
->ipha_length
= htons(REPLY_LEN
);
13778 ipha
->ipha_ident
= 0;
13780 icmph
= (icmph_t
*)&ipha
[1];
13781 icmph
->icmph_type
= ICMP_ADDRESS_MASK_REPLY
;
13782 bcopy(&ipif
->ipif_net_mask
, &icmph
[1], IP_ADDR_LEN
);
13783 icmph
->icmph_checksum
= IP_CSUM(mp
, sizeof (ipha_t
), 0);
13785 bzero(&ixas
, sizeof (ixas
));
13786 ixas
.ixa_flags
= IXAF_BASIC_SIMPLE_V4
;
13787 ixas
.ixa_zoneid
= ALL_ZONES
;
13788 ixas
.ixa_ifindex
= 0;
13789 ixas
.ixa_ipst
= ipst
;
13790 ixas
.ixa_multicast_ttl
= IP_DEFAULT_MULTICAST_TTL
;
13791 (void) ip_output_simple(mp
, &ixas
);
13792 ixa_cleanup(&ixas
);
13797 * Join the ipif specific multicast groups.
13798 * Must be called after a mapping has been set up in the resolver. (Always
13799 * called as writer.)
13802 ipif_multicast_up(ipif_t
*ipif
)
13808 ASSERT(IAM_WRITER_IPIF(ipif
));
13810 ill
= ipif
->ipif_ill
;
13812 ip1dbg(("ipif_multicast_up\n"));
13813 if (!(ill
->ill_flags
& ILLF_MULTICAST
) ||
13814 ipif
->ipif_allhosts_ilm
!= NULL
)
13817 if (ipif
->ipif_isv6
) {
13818 in6_addr_t v6allmc
= ipv6_all_hosts_mcast
;
13819 in6_addr_t v6solmc
= ipv6_solicited_node_mcast
;
13821 v6solmc
.s6_addr32
[3] |= ipif
->ipif_v6lcl_addr
.s6_addr32
[3];
13823 if (IN6_IS_ADDR_UNSPECIFIED(&ipif
->ipif_v6lcl_addr
))
13826 ip1dbg(("ipif_multicast_up - addmulti\n"));
13829 * Join the all hosts multicast address. We skip this for
13830 * underlying IPMP interfaces since they should be invisible.
13832 if (!IS_UNDER_IPMP(ill
)) {
13833 ilm
= ip_addmulti(&v6allmc
, ill
, ipif
->ipif_zoneid
,
13837 ip0dbg(("ipif_multicast_up: "
13838 "all_hosts_mcast failed %d\n", err
));
13841 ipif
->ipif_allhosts_ilm
= ilm
;
13845 * Enable multicast for the solicited node multicast address.
13846 * If IPMP we need to put the membership on the upper ill.
13848 if (!(ipif
->ipif_flags
& IPIF_NOLOCAL
)) {
13849 ill_t
*mcast_ill
= NULL
;
13850 boolean_t need_refrele
;
13852 if (IS_UNDER_IPMP(ill
) &&
13853 (mcast_ill
= ipmp_ill_hold_ipmp_ill(ill
)) != NULL
) {
13854 need_refrele
= B_TRUE
;
13857 need_refrele
= B_FALSE
;
13860 ilm
= ip_addmulti(&v6solmc
, mcast_ill
,
13861 ipif
->ipif_zoneid
, &err
);
13863 ill_refrele(mcast_ill
);
13867 ip0dbg(("ipif_multicast_up: solicited MC"
13868 " failed %d\n", err
));
13869 if ((ilm
= ipif
->ipif_allhosts_ilm
) != NULL
) {
13870 ipif
->ipif_allhosts_ilm
= NULL
;
13871 (void) ip_delmulti(ilm
);
13875 ipif
->ipif_solmulti_ilm
= ilm
;
13878 in6_addr_t v6group
;
13880 if (ipif
->ipif_lcl_addr
== INADDR_ANY
|| IS_UNDER_IPMP(ill
))
13883 /* Join the all hosts multicast address */
13884 ip1dbg(("ipif_multicast_up - addmulti\n"));
13885 IN6_IPADDR_TO_V4MAPPED(htonl(INADDR_ALLHOSTS_GROUP
), &v6group
);
13887 ilm
= ip_addmulti(&v6group
, ill
, ipif
->ipif_zoneid
, &err
);
13890 ip0dbg(("ipif_multicast_up: failed %d\n", err
));
13893 ipif
->ipif_allhosts_ilm
= ilm
;
13898 * Blow away any multicast groups that we joined in ipif_multicast_up().
13899 * (ilms from explicit memberships are handled in conn_update_ill.)
13902 ipif_multicast_down(ipif_t
*ipif
)
13904 ASSERT(IAM_WRITER_IPIF(ipif
));
13906 ip1dbg(("ipif_multicast_down\n"));
13908 if (ipif
->ipif_allhosts_ilm
!= NULL
) {
13909 (void) ip_delmulti(ipif
->ipif_allhosts_ilm
);
13910 ipif
->ipif_allhosts_ilm
= NULL
;
13912 if (ipif
->ipif_solmulti_ilm
!= NULL
) {
13913 (void) ip_delmulti(ipif
->ipif_solmulti_ilm
);
13914 ipif
->ipif_solmulti_ilm
= NULL
;
13919 * Used when an interface comes up to recreate any extra routes on this
13923 ill_recover_saved_ire(ill_t
*ill
)
13926 ip_stack_t
*ipst
= ill
->ill_ipst
;
13928 ip1dbg(("ill_recover_saved_ire(%s)", ill
->ill_name
));
13930 mutex_enter(&ill
->ill_saved_ire_lock
);
13931 for (mp
= ill
->ill_saved_ire_mp
; mp
!= NULL
; mp
= mp
->b_cont
) {
13935 ifrt
= (ifrt_t
*)mp
->b_rptr
;
13937 * Create a copy of the IRE with the saved address and netmask.
13939 if (ill
->ill_isv6
) {
13940 ire
= ire_create_v6(
13941 &ifrt
->ifrt_v6addr
,
13942 &ifrt
->ifrt_v6mask
,
13943 &ifrt
->ifrt_v6gateway_addr
,
13951 (uint8_t *)&ifrt
->ifrt_addr
,
13952 (uint8_t *)&ifrt
->ifrt_mask
,
13953 (uint8_t *)&ifrt
->ifrt_gateway_addr
,
13961 mutex_exit(&ill
->ill_saved_ire_lock
);
13965 if (ifrt
->ifrt_flags
& RTF_SETSRC
) {
13966 if (ill
->ill_isv6
) {
13967 ire
->ire_setsrc_addr_v6
=
13968 ifrt
->ifrt_v6setsrc_addr
;
13970 ire
->ire_setsrc_addr
= ifrt
->ifrt_setsrc_addr
;
13975 * Some software (for example, GateD and Sun Cluster) attempts
13976 * to create (what amount to) IRE_PREFIX routes with the
13977 * loopback address as the gateway. This is primarily done to
13978 * set up prefixes with the RTF_REJECT flag set (for example,
13979 * when generating aggregate routes.)
13981 * If the IRE type (as defined by ill->ill_net_type) is
13982 * IRE_LOOPBACK, then we map the request into a
13983 * IRE_IF_NORESOLVER.
13985 if (ill
->ill_net_type
== IRE_LOOPBACK
)
13986 ire
->ire_type
= IRE_IF_NORESOLVER
;
13989 * ire held by ire_add, will be refreled' towards the
13990 * the end of ipif_up_done
13992 nire
= ire_add(ire
);
13994 * Check if it was a duplicate entry. This handles
13995 * the case of two racing route adds for the same route
13997 if (nire
== NULL
) {
13998 ip1dbg(("ill_recover_saved_ire: FAILED\n"));
13999 } else if (nire
!= ire
) {
14000 ip1dbg(("ill_recover_saved_ire: duplicate ire %p\n",
14004 ip1dbg(("ill_recover_saved_ire: added ire %p\n",
14010 mutex_exit(&ill
->ill_saved_ire_lock
);
14015 * Used to set the netmask and broadcast address to default values when the
14016 * interface is brought up. (Always called as writer.)
14019 ipif_set_default(ipif_t
*ipif
)
14021 ASSERT(MUTEX_HELD(&ipif
->ipif_ill
->ill_lock
));
14023 if (!ipif
->ipif_isv6
) {
14025 * Interface holds an IPv4 address. Default
14026 * mask is the natural netmask.
14028 if (!ipif
->ipif_net_mask
) {
14031 v4mask
= ip_net_mask(ipif
->ipif_lcl_addr
);
14032 V4MASK_TO_V6(v4mask
, ipif
->ipif_v6net_mask
);
14034 if (ipif
->ipif_flags
& IPIF_POINTOPOINT
) {
14035 /* ipif_subnet is ipif_pp_dst_addr for pt-pt */
14036 ipif
->ipif_v6subnet
= ipif
->ipif_v6pp_dst_addr
;
14038 V6_MASK_COPY(ipif
->ipif_v6lcl_addr
,
14039 ipif
->ipif_v6net_mask
, ipif
->ipif_v6subnet
);
14042 * NOTE: SunOS 4.X does this even if the broadcast address
14043 * has been already set thus we do the same here.
14045 if (ipif
->ipif_flags
& IPIF_BROADCAST
) {
14048 v4addr
= ipif
->ipif_subnet
| ~ipif
->ipif_net_mask
;
14049 IN6_IPADDR_TO_V4MAPPED(v4addr
, &ipif
->ipif_v6brd_addr
);
14053 * Interface holds an IPv6-only address. Default
14054 * mask is all-ones.
14056 if (IN6_IS_ADDR_UNSPECIFIED(&ipif
->ipif_v6net_mask
))
14057 ipif
->ipif_v6net_mask
= ipv6_all_ones
;
14058 if (ipif
->ipif_flags
& IPIF_POINTOPOINT
) {
14059 /* ipif_subnet is ipif_pp_dst_addr for pt-pt */
14060 ipif
->ipif_v6subnet
= ipif
->ipif_v6pp_dst_addr
;
14062 V6_MASK_COPY(ipif
->ipif_v6lcl_addr
,
14063 ipif
->ipif_v6net_mask
, ipif
->ipif_v6subnet
);
14069 * Return 0 if this address can be used as local address without causing
14070 * duplicate address problems. Otherwise, return EADDRNOTAVAIL if the address
14071 * is already up on a different ill, and EADDRINUSE if it's up on the same ill.
14072 * Note that the same IPv6 link-local address is allowed as long as the ills
14073 * are not on the same link.
14076 ip_addr_availability_check(ipif_t
*new_ipif
)
14078 in6_addr_t our_v6addr
;
14081 ill_walk_context_t ctx
;
14082 ip_stack_t
*ipst
= new_ipif
->ipif_ill
->ill_ipst
;
14084 ASSERT(IAM_WRITER_IPIF(new_ipif
));
14085 ASSERT(MUTEX_HELD(&ipst
->ips_ip_addr_avail_lock
));
14086 ASSERT(RW_READ_HELD(&ipst
->ips_ill_g_lock
));
14088 new_ipif
->ipif_flags
&= ~IPIF_UNNUMBERED
;
14089 if (IN6_IS_ADDR_UNSPECIFIED(&new_ipif
->ipif_v6lcl_addr
) ||
14090 IN6_IS_ADDR_V4MAPPED_ANY(&new_ipif
->ipif_v6lcl_addr
))
14093 our_v6addr
= new_ipif
->ipif_v6lcl_addr
;
14095 if (new_ipif
->ipif_isv6
)
14096 ill
= ILL_START_WALK_V6(&ctx
, ipst
);
14098 ill
= ILL_START_WALK_V4(&ctx
, ipst
);
14100 for (; ill
!= NULL
; ill
= ill_next(&ctx
, ill
)) {
14101 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
;
14102 ipif
= ipif
->ipif_next
) {
14103 if ((ipif
== new_ipif
) ||
14104 !(ipif
->ipif_flags
& IPIF_UP
) ||
14105 (ipif
->ipif_flags
& IPIF_UNNUMBERED
) ||
14106 !IN6_ARE_ADDR_EQUAL(&ipif
->ipif_v6lcl_addr
,
14110 if (new_ipif
->ipif_flags
& IPIF_POINTOPOINT
)
14111 new_ipif
->ipif_flags
|= IPIF_UNNUMBERED
;
14112 else if (ipif
->ipif_flags
& IPIF_POINTOPOINT
)
14113 ipif
->ipif_flags
|= IPIF_UNNUMBERED
;
14114 else if ((IN6_IS_ADDR_LINKLOCAL(&our_v6addr
) ||
14115 IN6_IS_ADDR_SITELOCAL(&our_v6addr
)) &&
14116 !IS_ON_SAME_LAN(ill
, new_ipif
->ipif_ill
))
14118 else if (new_ipif
->ipif_zoneid
!= ipif
->ipif_zoneid
&&
14119 ipif
->ipif_zoneid
!= ALL_ZONES
&& IS_LOOPBACK(ill
))
14121 else if (new_ipif
->ipif_ill
== ill
)
14122 return (EADDRINUSE
);
14124 return (EADDRNOTAVAIL
);
14132 * Bring up an ipif: bring up arp/ndp, bring up the DLPI stream, and add
14133 * IREs for the ipif.
14134 * When the routine returns EINPROGRESS then mp has been consumed and
14135 * the ioctl will be acked from ip_rput_dlpi.
14138 ipif_up(ipif_t
*ipif
, queue_t
*q
, mblk_t
*mp
)
14140 ill_t
*ill
= ipif
->ipif_ill
;
14141 boolean_t isv6
= ipif
->ipif_isv6
;
14144 uint_t ipif_orig_id
;
14145 ip_stack_t
*ipst
= ill
->ill_ipst
;
14147 ASSERT(IAM_WRITER_IPIF(ipif
));
14149 ip1dbg(("ipif_up(%s:%u)\n", ill
->ill_name
, ipif
->ipif_id
));
14150 DTRACE_PROBE3(ipif__downup
, char *, "ipif_up",
14151 ill_t
*, ill
, ipif_t
*, ipif
);
14153 /* Shouldn't get here if it is already up. */
14154 if (ipif
->ipif_flags
& IPIF_UP
)
14158 * If this is a request to bring up a data address on an interface
14159 * under IPMP, then move the address to its IPMP meta-interface and
14160 * try to bring it up. One complication is that the zeroth ipif for
14161 * an ill is special, in that every ill always has one, and that code
14162 * throughout IP deferences ill->ill_ipif without holding any locks.
14164 if (IS_UNDER_IPMP(ill
) && ipmp_ipif_is_dataaddr(ipif
) &&
14165 (!ipif
->ipif_isv6
|| !V6_IPIF_LINKLOCAL(ipif
))) {
14166 ipif_t
*stubipif
= NULL
, *moveipif
= NULL
;
14167 ill_t
*ipmp_ill
= ipmp_illgrp_ipmp_ill(ill
->ill_grp
);
14170 * The ipif being brought up should be quiesced. If it's not,
14171 * something has gone amiss and we need to bail out. (If it's
14172 * quiesced, we know it will remain so via IPIF_CONDEMNED.)
14174 mutex_enter(&ill
->ill_lock
);
14175 if (!ipif_is_quiescent(ipif
)) {
14176 mutex_exit(&ill
->ill_lock
);
14179 mutex_exit(&ill
->ill_lock
);
14182 * If we're going to need to allocate ipifs, do it prior
14183 * to starting the move (and grabbing locks).
14185 if (ipif
->ipif_id
== 0) {
14186 if ((moveipif
= ipif_allocate(ill
, 0, IRE_LOCAL
, B_TRUE
,
14187 B_FALSE
, &err
)) == NULL
) {
14190 if ((stubipif
= ipif_allocate(ill
, 0, IRE_LOCAL
, B_TRUE
,
14191 B_FALSE
, &err
)) == NULL
) {
14198 * Grab or transfer the ipif to move. During the move, keep
14199 * ill_g_lock held to prevent any ill walker threads from
14200 * seeing things in an inconsistent state.
14202 rw_enter(&ipst
->ips_ill_g_lock
, RW_WRITER
);
14203 if (ipif
->ipif_id
!= 0) {
14206 ipif_transfer(ipif
, moveipif
, stubipif
);
14211 * Place the ipif on the IPMP ill. If the zeroth ipif on
14212 * the IPMP ill is a stub (0.0.0.0 down address) then we
14213 * replace that one. Otherwise, pick the next available slot.
14215 ipif
->ipif_ill
= ipmp_ill
;
14216 ipif_orig_id
= ipif
->ipif_id
;
14218 if (ipmp_ipif_is_stubaddr(ipmp_ill
->ill_ipif
)) {
14219 ipif_transfer(ipif
, ipmp_ill
->ill_ipif
, NULL
);
14220 ipif
= ipmp_ill
->ill_ipif
;
14222 ipif
->ipif_id
= -1;
14223 if ((err
= ipif_insert(ipif
, B_FALSE
)) != 0) {
14225 * No more available ipif_id's -- put it back
14226 * on the original ill and fail the operation.
14227 * Since we're writer on the ill, we can be
14228 * sure our old slot is still available.
14230 ipif
->ipif_id
= ipif_orig_id
;
14231 ipif
->ipif_ill
= ill
;
14232 if (ipif_orig_id
== 0) {
14233 ipif_transfer(ipif
, ill
->ill_ipif
,
14236 VERIFY(ipif_insert(ipif
, B_FALSE
) == 0);
14238 rw_exit(&ipst
->ips_ill_g_lock
);
14242 rw_exit(&ipst
->ips_ill_g_lock
);
14245 * Tell SCTP that the ipif has moved. Note that even if we
14246 * had to allocate a new ipif, the original sequence id was
14247 * preserved and therefore SCTP won't know.
14249 sctp_move_ipif(ipif
, ill
, ipmp_ill
);
14252 * If the ipif being brought up was on slot zero, then we
14253 * first need to bring up the placeholder we stuck there. In
14254 * ip_rput_dlpi_writer(), arp_bringup_done(), or the recursive
14255 * call to ipif_up() itself, if we successfully bring up the
14256 * placeholder, we'll check ill_move_ipif and bring it up too.
14258 if (ipif_orig_id
== 0) {
14259 ASSERT(ill
->ill_move_ipif
== NULL
);
14260 ill
->ill_move_ipif
= ipif
;
14261 if ((err
= ipif_up(ill
->ill_ipif
, q
, mp
)) == 0)
14262 ASSERT(ill
->ill_move_ipif
== NULL
);
14263 if (err
!= EINPROGRESS
)
14264 ill
->ill_move_ipif
= NULL
;
14269 * Bring it up on the IPMP ill.
14271 return (ipif_up(ipif
, q
, mp
));
14274 /* Skip arp/ndp for any loopback interface. */
14275 if (ill
->ill_wq
!= NULL
) {
14276 conn_t
*connp
= CONN_Q(q
) ? Q_TO_CONN(q
) : NULL
;
14277 ipsq_t
*ipsq
= ill
->ill_phyint
->phyint_ipsq
;
14279 if (!ill
->ill_dl_up
) {
14281 * ill_dl_up is not yet set. i.e. we are yet to
14282 * DL_BIND with the driver and this is the first
14283 * logical interface on the ill to become "up".
14284 * Tell the driver to get going (via DL_BIND_REQ).
14285 * Note that changing "significant" IFF_ flags
14286 * address/netmask etc cause a down/up dance, but
14287 * does not cause an unbind (DL_UNBIND) with the driver
14289 return (ill_dl_up(ill
, ipif
, mp
, q
));
14293 * ipif_resolver_up may end up needeing to bind/attach
14294 * the ARP stream, which in turn necessitates a
14295 * DLPI message exchange with the driver. ioctls are
14296 * serialized and so we cannot send more than one
14297 * interface up message at a time. If ipif_resolver_up
14298 * does need to wait for the DLPI handshake for the ARP stream,
14299 * we get EINPROGRESS and we will complete in arp_bringup_done.
14302 ASSERT(connp
!= NULL
|| !CONN_Q(q
));
14304 mutex_enter(&connp
->conn_lock
);
14305 mutex_enter(&ill
->ill_lock
);
14306 success
= ipsq_pending_mp_add(connp
, ipif
, q
, mp
, 0);
14307 mutex_exit(&ill
->ill_lock
);
14309 mutex_exit(&connp
->conn_lock
);
14314 * Crank up IPv6 neighbor discovery. Unlike ARP, this should
14315 * complete when ipif_ndp_up returns.
14317 err
= ipif_resolver_up(ipif
, Res_act_initial
);
14318 if (err
== EINPROGRESS
) {
14319 /* We will complete it in arp_bringup_done() */
14323 if (isv6
&& err
== 0)
14324 err
= ipif_ndp_up(ipif
, B_TRUE
);
14326 ASSERT(err
!= EINPROGRESS
);
14327 mp
= ipsq_pending_mp_get(ipsq
, &connp
);
14328 ASSERT(mp
!= NULL
);
14333 * Interfaces without underlying hardware don't do duplicate
14334 * address detection.
14336 ASSERT(!(ipif
->ipif_flags
& IPIF_DUPLICATE
));
14337 ipif
->ipif_addr_ready
= 1;
14338 err
= ill_add_ires(ill
);
14339 /* allocation failure? */
14344 err
= (isv6
? ipif_up_done_v6(ipif
) : ipif_up_done(ipif
));
14345 if (err
== 0 && ill
->ill_move_ipif
!= NULL
) {
14346 ipif
= ill
->ill_move_ipif
;
14347 ill
->ill_move_ipif
= NULL
;
14348 return (ipif_up(ipif
, q
, mp
));
14354 * Add any IREs tied to the ill. For now this is just an IRE_MULTICAST.
14355 * The identical set of IREs need to be removed in ill_delete_ires().
14358 ill_add_ires(ill_t
*ill
)
14361 in6_addr_t dummy6
= IN6ADDR_INITIALIZER(V6_MCAST
, 0, 0, 1);
14362 in_addr_t dummy4
= htonl(INADDR_ALLHOSTS_GROUP
);
14364 if (ill
->ill_ire_multicast
!= NULL
)
14368 * provide some dummy ire_addr for creating the ire.
14370 if (ill
->ill_isv6
) {
14371 ire
= ire_create_v6(&dummy6
, 0, 0, IRE_MULTICAST
, ill
,
14372 ALL_ZONES
, RTF_UP
, ill
->ill_ipst
);
14374 ire
= ire_create((uchar_t
*)&dummy4
, 0, 0, IRE_MULTICAST
, ill
,
14375 ALL_ZONES
, RTF_UP
, ill
->ill_ipst
);
14380 ill
->ill_ire_multicast
= ire
;
14385 ill_delete_ires(ill_t
*ill
)
14387 if (ill
->ill_ire_multicast
!= NULL
) {
14389 * BIND/ATTACH completed; Release the ref for ill_ire_multicast
14390 * which was taken without any th_tracing enabled.
14391 * We also mark it as condemned (note that it was never added)
14392 * so that caching conn's can move off of it.
14394 ire_make_condemned(ill
->ill_ire_multicast
);
14395 ire_refrele_notr(ill
->ill_ire_multicast
);
14396 ill
->ill_ire_multicast
= NULL
;
14401 * Perform a bind for the physical device.
14402 * When the routine returns EINPROGRESS then mp has been consumed and
14403 * the ioctl will be acked from ip_rput_dlpi.
14404 * Allocate an unbind message and save it until ipif_down.
14407 ill_dl_up(ill_t
*ill
, ipif_t
*ipif
, mblk_t
*mp
, queue_t
*q
)
14409 mblk_t
*bind_mp
= NULL
;
14410 mblk_t
*unbind_mp
= NULL
;
14415 DTRACE_PROBE2(ill__downup
, char *, "ill_dl_up", ill_t
*, ill
);
14417 ip1dbg(("ill_dl_up(%s)\n", ill
->ill_name
));
14418 ASSERT(IAM_WRITER_ILL(ill
));
14419 ASSERT(mp
!= NULL
);
14422 * Make sure we have an IRE_MULTICAST in case we immediately
14423 * start receiving packets.
14425 err
= ill_add_ires(ill
);
14429 bind_mp
= ip_dlpi_alloc(sizeof (dl_bind_req_t
) + sizeof (long),
14431 if (bind_mp
== NULL
)
14433 ((dl_bind_req_t
*)bind_mp
->b_rptr
)->dl_sap
= ill
->ill_sap
;
14434 ((dl_bind_req_t
*)bind_mp
->b_rptr
)->dl_service_mode
= DL_CLDLS
;
14437 * ill_unbind_mp would be non-null if the following sequence had
14439 * - send DL_BIND_REQ to driver, wait for response
14440 * - multiple ioctls that need to bring the ipif up are encountered,
14441 * but they cannot enter the ipsq due to the outstanding DL_BIND_REQ.
14442 * These ioctls will then be enqueued on the ipsq
14443 * - a DL_ERROR_ACK is returned for the DL_BIND_REQ
14444 * At this point, the pending ioctls in the ipsq will be drained, and
14445 * since ill->ill_dl_up was not set, ill_dl_up would be invoked with
14446 * a non-null ill->ill_unbind_mp
14448 if (ill
->ill_unbind_mp
== NULL
) {
14449 unbind_mp
= ip_dlpi_alloc(sizeof (dl_unbind_req_t
),
14451 if (unbind_mp
== NULL
)
14455 * Record state needed to complete this operation when the
14456 * DL_BIND_ACK shows up. Also remember the pre-allocated mblks.
14458 connp
= CONN_Q(q
) ? Q_TO_CONN(q
) : NULL
;
14459 ASSERT(connp
!= NULL
|| !CONN_Q(q
));
14461 mutex_enter(&ipif
->ipif_ill
->ill_lock
);
14462 success
= ipsq_pending_mp_add(connp
, ipif
, q
, mp
, 0);
14463 mutex_exit(&ipif
->ipif_ill
->ill_lock
);
14464 RELEASE_CONN_LOCK(q
);
14469 * Save the unbind message for ill_dl_down(); it will be consumed when
14470 * the interface goes down.
14472 if (ill
->ill_unbind_mp
== NULL
)
14473 ill
->ill_unbind_mp
= unbind_mp
;
14475 ill_dlpi_send(ill
, bind_mp
);
14476 /* Send down link-layer capabilities probe if not already done. */
14477 ill_capability_probe(ill
);
14480 * Sysid used to rely on the fact that netboots set domainname
14481 * and the like. Now that miniroot boots aren't strictly netboots
14482 * and miniroot network configuration is driven from userland
14483 * these things still need to be set. This situation can be detected
14484 * by comparing the interface being configured here to the one
14485 * dhcifname was set to reference by the boot loader. Once sysid is
14486 * converted to use dhcp_ipc_getinfo() this call can go away.
14488 if ((ipif
->ipif_flags
& IPIF_DHCPRUNNING
) &&
14489 (strcmp(ill
->ill_name
, dhcifname
) == 0) &&
14490 (strlen(srpc_domain
) == 0)) {
14491 if (dhcpinit() != 0)
14492 cmn_err(CE_WARN
, "no cached dhcp response");
14496 * This operation will complete in ip_rput_dlpi with either
14497 * a DL_BIND_ACK or DL_ERROR_ACK.
14499 return (EINPROGRESS
);
14501 ip1dbg(("ill_dl_up(%s) FAILED\n", ill
->ill_name
));
14504 freemsg(unbind_mp
);
14508 /* Add room for tcp+ip headers */
14509 uint_t ip_loopback_mtuplus
= IP_LOOPBACK_MTU
+ IP_SIMPLE_HDR_LENGTH
+ 20;
14512 * DLPI and ARP is up.
14513 * Create all the IREs associated with an interface. Bring up multicast.
14514 * Set the interface flag and finish other initialization
14515 * that potentially had to be deferred to after DL_BIND_ACK.
14518 ipif_up_done(ipif_t
*ipif
)
14520 ill_t
*ill
= ipif
->ipif_ill
;
14522 boolean_t loopback
= B_FALSE
;
14523 boolean_t update_src_selection
= B_TRUE
;
14526 ip1dbg(("ipif_up_done(%s:%u)\n",
14527 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
));
14528 DTRACE_PROBE3(ipif__downup
, char *, "ipif_up_done",
14529 ill_t
*, ill
, ipif_t
*, ipif
);
14531 /* Check if this is a loopback interface */
14532 if (ipif
->ipif_ill
->ill_wq
== NULL
)
14535 ASSERT(!MUTEX_HELD(&ipif
->ipif_ill
->ill_lock
));
14538 * If all other interfaces for this ill are down or DEPRECATED,
14539 * or otherwise unsuitable for source address selection,
14540 * reset the src generation numbers to make sure source
14541 * address selection gets to take this new ipif into account.
14542 * No need to hold ill_lock while traversing the ipif list since
14545 for (tmp_ipif
= ill
->ill_ipif
; tmp_ipif
;
14546 tmp_ipif
= tmp_ipif
->ipif_next
) {
14547 if (((tmp_ipif
->ipif_flags
&
14548 (IPIF_NOXMIT
|IPIF_ANYCAST
|IPIF_NOLOCAL
|IPIF_DEPRECATED
)) ||
14549 !(tmp_ipif
->ipif_flags
& IPIF_UP
)) ||
14550 (tmp_ipif
== ipif
))
14552 /* first useable pre-existing interface */
14553 update_src_selection
= B_FALSE
;
14556 if (update_src_selection
)
14557 ip_update_source_selection(ill
->ill_ipst
);
14559 if (IS_LOOPBACK(ill
) || ill
->ill_net_type
== IRE_IF_NORESOLVER
) {
14560 nce_t
*loop_nce
= NULL
;
14561 uint16_t flags
= (NCE_F_MYADDR
| NCE_F_AUTHORITY
| NCE_F_NONUD
);
14564 * lo0:1 and subsequent ipifs were marked IRE_LOCAL in
14565 * ipif_lookup_on_name(), but in the case of zones we can have
14566 * several loopback addresses on lo0. So all the interfaces with
14567 * loopback addresses need to be marked IRE_LOOPBACK.
14569 if (V4_PART_OF_V6(ipif
->ipif_v6lcl_addr
) ==
14570 htonl(INADDR_LOOPBACK
))
14571 ipif
->ipif_ire_type
= IRE_LOOPBACK
;
14573 ipif
->ipif_ire_type
= IRE_LOCAL
;
14574 if (ill
->ill_net_type
!= IRE_LOOPBACK
)
14575 flags
|= NCE_F_PUBLISH
;
14577 /* add unicast nce for the local addr */
14578 err
= nce_lookup_then_add_v4(ill
, NULL
,
14579 ill
->ill_phys_addr_length
, &ipif
->ipif_lcl_addr
, flags
,
14580 ND_REACHABLE
, &loop_nce
);
14581 /* A shared-IP zone sees EEXIST for lo0:N */
14582 if (err
== 0 || err
== EEXIST
) {
14583 ipif
->ipif_added_nce
= 1;
14584 loop_nce
->nce_ipif_cnt
++;
14585 nce_refrele(loop_nce
);
14588 ASSERT(loop_nce
== NULL
);
14593 /* Create all the IREs associated with this interface */
14594 err
= ipif_add_ires_v4(ipif
, loopback
);
14597 * see comments about return value from
14598 * ip_addr_availability_check() in ipif_add_ires_v4().
14600 if (err
!= EADDRINUSE
) {
14601 (void) ipif_arp_down(ipif
);
14604 * Make IPMP aware of the deleted ipif so that
14605 * the needed ipmp cleanup (e.g., of ipif_bound_ill)
14606 * can be completed. Note that we do not want to
14607 * destroy the nce that was created on the ipmp_ill
14608 * for the active copy of the duplicate address in
14612 ipmp_illgrp_del_ipif(ill
->ill_grp
, ipif
);
14613 err
= EADDRNOTAVAIL
;
14618 if (ill
->ill_ipif_up_count
== 1 && !loopback
) {
14619 /* Recover any additional IREs entries for this ill */
14620 (void) ill_recover_saved_ire(ill
);
14623 if (ill
->ill_need_recover_multicast
) {
14625 * Need to recover all multicast memberships in the driver.
14626 * This had to be deferred until we had attached. The same
14627 * code exists in ipif_up_done_v6() to recover IPv6
14630 * Note that it would be preferable to unconditionally do the
14631 * ill_recover_multicast() in ill_dl_up(), but we cannot do
14632 * that since ill_join_allmulti() depends on ill_dl_up being
14633 * set, and it is not set until we receive a DL_BIND_ACK after
14634 * having called ill_dl_up().
14636 ill_recover_multicast(ill
);
14639 if (ill
->ill_ipif_up_count
== 1) {
14641 * Since the interface is now up, it may now be active.
14643 if (IS_UNDER_IPMP(ill
))
14644 ipmp_ill_refresh_active(ill
);
14647 * If this is an IPMP interface, we may now be able to
14648 * establish ARP entries.
14651 ipmp_illgrp_refresh_arpent(ill
->ill_grp
);
14654 /* Join the allhosts multicast address */
14655 ipif_multicast_up(ipif
);
14657 if (!loopback
&& !update_src_selection
&&
14658 !(ipif
->ipif_flags
& (IPIF_NOLOCAL
|IPIF_ANYCAST
|IPIF_DEPRECATED
)))
14659 ip_update_source_selection(ill
->ill_ipst
);
14661 if (!loopback
&& ipif
->ipif_addr_ready
) {
14662 /* Broadcast an address mask reply. */
14663 ipif_mask_reply(ipif
);
14665 /* Perhaps ilgs should use this ill */
14666 update_conn_ill(NULL
, ill
->ill_ipst
);
14669 * This had to be deferred until we had bound. Tell routing sockets and
14670 * others that this interface is up if it looks like the address has
14671 * been validated. Otherwise, if it isn't ready yet, wait for
14672 * duplicate address detection to do its thing.
14674 if (ipif
->ipif_addr_ready
)
14675 ipif_up_notify(ipif
);
14680 * Add the IREs associated with the ipif.
14681 * Those MUST be explicitly removed in ipif_delete_ires_v4.
14684 ipif_add_ires_v4(ipif_t
*ipif
, boolean_t loopback
)
14686 ill_t
*ill
= ipif
->ipif_ill
;
14687 ip_stack_t
*ipst
= ill
->ill_ipst
;
14688 ire_t
*ire_array
[20];
14689 ire_t
**irep
= ire_array
;
14691 ipaddr_t net_mask
= 0;
14692 ipaddr_t subnet_mask
, route_mask
;
14694 ire_t
*ire_local
= NULL
; /* LOCAL or LOOPBACK */
14695 ire_t
*ire_if
= NULL
;
14698 if ((ipif
->ipif_lcl_addr
!= INADDR_ANY
) &&
14699 !(ipif
->ipif_flags
& IPIF_NOLOCAL
)) {
14700 /* Register the source address for __sin6_src_id */
14701 err
= ip_srcid_insert(&ipif
->ipif_v6lcl_addr
,
14702 ipif
->ipif_zoneid
, ipst
);
14704 ip0dbg(("ipif_add_ires: srcid_insert %d\n", err
));
14709 gw
= (uchar_t
*)&ipif
->ipif_lcl_addr
;
14713 /* If the interface address is set, create the local IRE. */
14714 ire_local
= ire_create(
14715 (uchar_t
*)&ipif
->ipif_lcl_addr
, /* dest address */
14716 (uchar_t
*)&ip_g_all_ones
, /* mask */
14718 ipif
->ipif_ire_type
, /* LOCAL or LOOPBACK */
14721 ((ipif
->ipif_flags
& IPIF_PRIVATE
) ?
14722 RTF_PRIVATE
: 0) | RTF_KERNEL
,
14724 ip1dbg(("ipif_add_ires: 0x%p creating IRE %p type 0x%x"
14725 " for 0x%x\n", (void *)ipif
, (void *)ire_local
,
14726 ipif
->ipif_ire_type
,
14727 ntohl(ipif
->ipif_lcl_addr
)));
14728 if (ire_local
== NULL
) {
14729 ip1dbg(("ipif_up_done: NULL ire_local\n"));
14735 "ipif_add_ires: not creating IRE %d for 0x%x: flags 0x%x\n",
14736 ipif
->ipif_ire_type
,
14737 ntohl(ipif
->ipif_lcl_addr
),
14738 (uint_t
)ipif
->ipif_flags
));
14740 if ((ipif
->ipif_lcl_addr
!= INADDR_ANY
) &&
14741 !(ipif
->ipif_flags
& IPIF_NOLOCAL
)) {
14742 net_mask
= ip_net_mask(ipif
->ipif_lcl_addr
);
14744 net_mask
= htonl(IN_CLASSA_NET
); /* fallback */
14747 subnet_mask
= ipif
->ipif_net_mask
;
14750 * If mask was not specified, use natural netmask of
14751 * interface address. Also, store this mask back into the
14754 if (subnet_mask
== 0) {
14755 subnet_mask
= net_mask
;
14756 V4MASK_TO_V6(subnet_mask
, ipif
->ipif_v6net_mask
);
14757 V6_MASK_COPY(ipif
->ipif_v6lcl_addr
, ipif
->ipif_v6net_mask
,
14758 ipif
->ipif_v6subnet
);
14761 /* Set up the IRE_IF_RESOLVER or IRE_IF_NORESOLVER, as appropriate. */
14762 if (!loopback
&& !(ipif
->ipif_flags
& IPIF_NOXMIT
) &&
14763 ipif
->ipif_subnet
!= INADDR_ANY
) {
14764 /* ipif_subnet is ipif_pp_dst_addr for pt-pt */
14766 if (ipif
->ipif_flags
& IPIF_POINTOPOINT
) {
14767 route_mask
= IP_HOST_MASK
;
14769 route_mask
= subnet_mask
;
14772 ip1dbg(("ipif_add_ires: ipif 0x%p ill 0x%p "
14773 "creating if IRE ill_net_type 0x%x for 0x%x\n",
14774 (void *)ipif
, (void *)ill
, ill
->ill_net_type
,
14775 ntohl(ipif
->ipif_subnet
)));
14776 ire_if
= ire_create(
14777 (uchar_t
*)&ipif
->ipif_subnet
,
14778 (uchar_t
*)&route_mask
,
14779 (uchar_t
*)&ipif
->ipif_lcl_addr
,
14783 ((ipif
->ipif_flags
& IPIF_PRIVATE
) ?
14784 RTF_PRIVATE
: 0) | RTF_KERNEL
,
14786 if (ire_if
== NULL
) {
14787 ip1dbg(("ipif_up_done: NULL ire_if\n"));
14794 * Create any necessary broadcast IREs.
14796 if ((ipif
->ipif_flags
& IPIF_BROADCAST
) &&
14797 !(ipif
->ipif_flags
& IPIF_NOXMIT
))
14798 irep
= ipif_create_bcast_ires(ipif
, irep
);
14800 /* If an earlier ire_create failed, get out now */
14801 for (irep1
= irep
; irep1
> ire_array
; ) {
14803 if (*irep1
== NULL
) {
14804 ip1dbg(("ipif_up_done: NULL ire found in ire_array\n"));
14811 * Need to atomically check for IP address availability under
14812 * ip_addr_avail_lock. ill_g_lock is held as reader to ensure no new
14813 * ills or new ipifs can be added while we are checking availability.
14815 rw_enter(&ipst
->ips_ill_g_lock
, RW_READER
);
14816 mutex_enter(&ipst
->ips_ip_addr_avail_lock
);
14817 /* Mark it up, and increment counters. */
14818 ipif
->ipif_flags
|= IPIF_UP
;
14819 ill
->ill_ipif_up_count
++;
14820 err
= ip_addr_availability_check(ipif
);
14821 mutex_exit(&ipst
->ips_ip_addr_avail_lock
);
14822 rw_exit(&ipst
->ips_ill_g_lock
);
14826 * Our address may already be up on the same ill. In this case,
14827 * the ARP entry for our ipif replaced the one for the other
14828 * ipif. So we don't want to delete it (otherwise the other ipif
14829 * would be unable to send packets).
14830 * ip_addr_availability_check() identifies this case for us and
14831 * returns EADDRINUSE; Caller should turn it into EADDRNOTAVAIL
14832 * which is the expected error code.
14834 ill
->ill_ipif_up_count
--;
14835 ipif
->ipif_flags
&= ~IPIF_UP
;
14840 * Add in all newly created IREs. ire_create_bcast() has
14841 * already checked for duplicates of the IRE_BROADCAST type.
14842 * We add the IRE_INTERFACE before the IRE_LOCAL to ensure
14843 * that lookups find the IRE_LOCAL even if the IRE_INTERFACE is
14846 if (ire_if
!= NULL
) {
14847 ire_if
= ire_add(ire_if
);
14848 if (ire_if
== NULL
) {
14853 ire_refhold_notr(ire_if
);
14854 ire_refrele(ire_if
);
14857 if (ire_local
!= NULL
) {
14858 ire_local
= ire_add(ire_local
);
14859 if (ire_local
== NULL
) {
14864 ire_refhold_notr(ire_local
);
14865 ire_refrele(ire_local
);
14868 rw_enter(&ipst
->ips_ill_g_lock
, RW_WRITER
);
14869 if (ire_local
!= NULL
)
14870 ipif
->ipif_ire_local
= ire_local
;
14871 if (ire_if
!= NULL
)
14872 ipif
->ipif_ire_if
= ire_if
;
14873 rw_exit(&ipst
->ips_ill_g_lock
);
14878 * We first add all of them, and if that succeeds we refrele the
14879 * bunch. That enables us to delete all of them should any of the
14882 for (irep1
= irep
; irep1
> ire_array
; ) {
14884 ASSERT(!MUTEX_HELD(&((*irep1
)->ire_ill
->ill_lock
)));
14885 *irep1
= ire_add(*irep1
);
14886 if (*irep1
== NULL
) {
14892 for (irep1
= irep
; irep1
> ire_array
; ) {
14894 /* refheld by ire_add. */
14895 if (*irep1
!= NULL
) {
14896 ire_refrele(*irep1
);
14903 * If the broadcast address has been set, make sure it makes
14904 * sense based on the interface address.
14905 * Only match on ill since we are sharing broadcast addresses.
14907 if ((ipif
->ipif_brd_addr
!= INADDR_ANY
) &&
14908 (ipif
->ipif_flags
& IPIF_BROADCAST
)) {
14911 ire
= ire_ftable_lookup_v4(ipif
->ipif_brd_addr
, 0, 0,
14912 IRE_BROADCAST
, ipif
->ipif_ill
, ALL_ZONES
,
14913 (MATCH_IRE_TYPE
| MATCH_IRE_ILL
), 0, ipst
, NULL
);
14917 * If there isn't a matching broadcast IRE,
14918 * revert to the default for this netmask.
14920 ipif
->ipif_v6brd_addr
= ipv6_all_zeros
;
14921 mutex_enter(&ipif
->ipif_ill
->ill_lock
);
14922 ipif_set_default(ipif
);
14923 mutex_exit(&ipif
->ipif_ill
->ill_lock
);
14933 ill
->ill_ipif_up_count
--;
14934 ipif
->ipif_flags
&= ~IPIF_UP
;
14937 ip1dbg(("ipif_add_ires: FAILED \n"));
14938 if (ire_local
!= NULL
)
14939 ire_delete(ire_local
);
14940 if (ire_if
!= NULL
)
14941 ire_delete(ire_if
);
14943 rw_enter(&ipst
->ips_ill_g_lock
, RW_WRITER
);
14944 ire_local
= ipif
->ipif_ire_local
;
14945 ipif
->ipif_ire_local
= NULL
;
14946 ire_if
= ipif
->ipif_ire_if
;
14947 ipif
->ipif_ire_if
= NULL
;
14948 rw_exit(&ipst
->ips_ill_g_lock
);
14949 if (ire_local
!= NULL
) {
14950 ire_delete(ire_local
);
14951 ire_refrele_notr(ire_local
);
14953 if (ire_if
!= NULL
) {
14954 ire_delete(ire_if
);
14955 ire_refrele_notr(ire_if
);
14958 while (irep
> ire_array
) {
14960 if (*irep
!= NULL
) {
14964 (void) ip_srcid_remove(&ipif
->ipif_v6lcl_addr
, ipif
->ipif_zoneid
, ipst
);
14969 /* Remove all the IREs created by ipif_add_ires_v4 */
14971 ipif_delete_ires_v4(ipif_t
*ipif
)
14973 ill_t
*ill
= ipif
->ipif_ill
;
14974 ip_stack_t
*ipst
= ill
->ill_ipst
;
14977 rw_enter(&ipst
->ips_ill_g_lock
, RW_WRITER
);
14978 ire
= ipif
->ipif_ire_local
;
14979 ipif
->ipif_ire_local
= NULL
;
14980 rw_exit(&ipst
->ips_ill_g_lock
);
14983 * Move count to ipif so we don't loose the count due to
14986 atomic_add_32(&ipif
->ipif_ib_pkt_count
, ire
->ire_ib_pkt_count
);
14989 ire_refrele_notr(ire
);
14991 rw_enter(&ipst
->ips_ill_g_lock
, RW_WRITER
);
14992 ire
= ipif
->ipif_ire_if
;
14993 ipif
->ipif_ire_if
= NULL
;
14994 rw_exit(&ipst
->ips_ill_g_lock
);
14997 ire_refrele_notr(ire
);
15001 * Delete the broadcast IREs.
15003 if ((ipif
->ipif_flags
& IPIF_BROADCAST
) &&
15004 !(ipif
->ipif_flags
& IPIF_NOXMIT
))
15005 ipif_delete_bcast_ires(ipif
);
15009 * Checks for availbility of a usable source address (if there is one) when the
15010 * destination ILL has the ill_usesrc_ifindex pointing to another ILL. Note
15011 * this selection is done regardless of the destination.
15014 ipif_zone_avail(uint_t ifindex
, boolean_t isv6
, zoneid_t zoneid
,
15017 ipif_t
*ipif
= NULL
;
15020 ASSERT(ifindex
!= 0);
15022 uill
= ill_lookup_on_ifindex(ifindex
, isv6
, ipst
);
15026 mutex_enter(&uill
->ill_lock
);
15027 for (ipif
= uill
->ill_ipif
; ipif
!= NULL
; ipif
= ipif
->ipif_next
) {
15028 if (IPIF_IS_CONDEMNED(ipif
))
15030 if (ipif
->ipif_flags
& (IPIF_NOLOCAL
|IPIF_ANYCAST
))
15032 if (!(ipif
->ipif_flags
& IPIF_UP
))
15034 if (ipif
->ipif_zoneid
!= zoneid
)
15036 if (isv6
? IN6_IS_ADDR_UNSPECIFIED(&ipif
->ipif_v6lcl_addr
) :
15037 ipif
->ipif_lcl_addr
== INADDR_ANY
)
15039 mutex_exit(&uill
->ill_lock
);
15043 mutex_exit(&uill
->ill_lock
);
15049 * Find an ipif with a good local address on the ill+zoneid.
15052 ipif_good_addr(ill_t
*ill
, zoneid_t zoneid
)
15056 mutex_enter(&ill
->ill_lock
);
15057 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
; ipif
= ipif
->ipif_next
) {
15058 if (IPIF_IS_CONDEMNED(ipif
))
15060 if (ipif
->ipif_flags
& (IPIF_NOLOCAL
|IPIF_ANYCAST
))
15062 if (!(ipif
->ipif_flags
& IPIF_UP
))
15064 if (ipif
->ipif_zoneid
!= zoneid
&&
15065 ipif
->ipif_zoneid
!= ALL_ZONES
&& zoneid
!= ALL_ZONES
)
15067 if (ill
->ill_isv6
?
15068 IN6_IS_ADDR_UNSPECIFIED(&ipif
->ipif_v6lcl_addr
) :
15069 ipif
->ipif_lcl_addr
== INADDR_ANY
)
15071 ipif_refhold_locked(ipif
);
15072 mutex_exit(&ill
->ill_lock
);
15075 mutex_exit(&ill
->ill_lock
);
15080 * IP source address type, sorted from worst to best. For a given type,
15081 * always prefer IP addresses on the same subnet. All-zones addresses are
15082 * suboptimal because they pose problems with unlabeled destinations.
15086 IPIF_DIFFNET_DEPRECATED
, /* deprecated and different subnet */
15087 IPIF_SAMENET_DEPRECATED
, /* deprecated and same subnet */
15088 IPIF_DIFFNET_ALLZONES
, /* allzones and different subnet */
15089 IPIF_SAMENET_ALLZONES
, /* allzones and same subnet */
15090 IPIF_DIFFNET
, /* normal and different subnet */
15091 IPIF_SAMENET
, /* normal and same subnet */
15092 IPIF_LOCALADDR
/* local loopback */
15096 * Pick the optimal ipif on `ill' for sending to destination `dst' from zone
15097 * `zoneid'. We rate usable ipifs from low -> high as per the ipif_type_t
15098 * enumeration, and return the highest-rated ipif. If there's a tie, we pick
15099 * the first one, unless IPMP is used in which case we round-robin among them;
15100 * see below for more.
15102 * Returns NULL if there is no suitable source address for the ill.
15103 * This only occurs when there is no valid source address for the ill.
15106 ipif_select_source_v4(ill_t
*ill
, ipaddr_t dst
, zoneid_t zoneid
,
15107 boolean_t allow_usesrc
, boolean_t
*notreadyp
)
15109 ill_t
*usill
= NULL
;
15110 ill_t
*ipmp_ill
= NULL
;
15111 ipif_t
*start_ipif
, *next_ipif
, *ipif
, *best_ipif
;
15112 ipif_type_t type
, best_type
;
15113 ip_stack_t
*ipst
= ill
->ill_ipst
;
15116 if (ill
->ill_usesrc_ifindex
!= 0 && allow_usesrc
) {
15117 usill
= ill_lookup_on_ifindex(ill
->ill_usesrc_ifindex
,
15120 ill
= usill
; /* Select source from usesrc ILL */
15126 * Test addresses should never be used for source address selection,
15127 * so if we were passed one, switch to the IPMP meta-interface.
15129 if (IS_UNDER_IPMP(ill
)) {
15130 if ((ipmp_ill
= ipmp_ill_hold_ipmp_ill(ill
)) != NULL
)
15131 ill
= ipmp_ill
; /* Select source from IPMP ill */
15137 * Hold the ill_g_lock as reader. This makes sure that no ipif/ill
15138 * can be deleted. But an ipif/ill can get CONDEMNED any time.
15139 * After selecting the right ipif, under ill_lock make sure ipif is
15140 * not condemned, and increment refcnt. If ipif is CONDEMNED,
15141 * we retry. Inside the loop we still need to check for CONDEMNED,
15142 * but not under a lock.
15144 rw_enter(&ipst
->ips_ill_g_lock
, RW_READER
);
15147 * For source address selection, we treat the ipif list as circular
15148 * and continue until we get back to where we started. This allows
15149 * IPMP to vary source address selection (which improves inbound load
15150 * spreading) by caching its last ending point and starting from
15151 * there. NOTE: we don't have to worry about ill_src_ipif changing
15152 * ills since that can't happen on the IPMP ill.
15154 start_ipif
= ill
->ill_ipif
;
15155 if (IS_IPMP(ill
) && ill
->ill_src_ipif
!= NULL
)
15156 start_ipif
= ill
->ill_src_ipif
;
15160 best_type
= IPIF_NONE
;
15162 if ((next_ipif
= ipif
->ipif_next
) == NULL
)
15163 next_ipif
= ill
->ill_ipif
;
15165 if (IPIF_IS_CONDEMNED(ipif
))
15167 /* Always skip NOLOCAL and ANYCAST interfaces */
15168 if (ipif
->ipif_flags
& (IPIF_NOLOCAL
|IPIF_ANYCAST
))
15170 /* Always skip NOACCEPT interfaces */
15171 if (ipif
->ipif_ill
->ill_flags
& ILLF_NOACCEPT
)
15173 if (!(ipif
->ipif_flags
& IPIF_UP
))
15176 if (!ipif
->ipif_addr_ready
) {
15177 if (notreadyp
!= NULL
)
15178 *notreadyp
= B_TRUE
;
15182 if (zoneid
!= ALL_ZONES
&&
15183 ipif
->ipif_zoneid
!= zoneid
&&
15184 ipif
->ipif_zoneid
!= ALL_ZONES
)
15188 * Interfaces with 0.0.0.0 address are allowed to be UP, but
15189 * are not valid as source addresses.
15191 if (ipif
->ipif_lcl_addr
== INADDR_ANY
)
15194 samenet
= ((ipif
->ipif_net_mask
& dst
) == ipif
->ipif_subnet
);
15196 if (ipif
->ipif_lcl_addr
== dst
) {
15197 type
= IPIF_LOCALADDR
;
15198 } else if (ipif
->ipif_flags
& IPIF_DEPRECATED
) {
15199 type
= samenet
? IPIF_SAMENET_DEPRECATED
:
15200 IPIF_DIFFNET_DEPRECATED
;
15201 } else if (ipif
->ipif_zoneid
== ALL_ZONES
) {
15202 type
= samenet
? IPIF_SAMENET_ALLZONES
:
15203 IPIF_DIFFNET_ALLZONES
;
15205 type
= samenet
? IPIF_SAMENET
: IPIF_DIFFNET
;
15208 if (type
> best_type
) {
15211 if (best_type
== IPIF_LOCALADDR
)
15212 break; /* can't get better */
15214 } while ((ipif
= next_ipif
) != start_ipif
);
15216 if ((ipif
= best_ipif
) != NULL
) {
15217 mutex_enter(&ipif
->ipif_ill
->ill_lock
);
15218 if (IPIF_IS_CONDEMNED(ipif
)) {
15219 mutex_exit(&ipif
->ipif_ill
->ill_lock
);
15222 ipif_refhold_locked(ipif
);
15225 * For IPMP, update the source ipif rotor to the next ipif,
15226 * provided we can look it up. (We must not use it if it's
15227 * IPIF_CONDEMNED since we may have grabbed ill_g_lock after
15228 * ipif_free() checked ill_src_ipif.)
15230 if (IS_IPMP(ill
) && ipif
!= NULL
) {
15231 next_ipif
= ipif
->ipif_next
;
15232 if (next_ipif
!= NULL
&& !IPIF_IS_CONDEMNED(next_ipif
))
15233 ill
->ill_src_ipif
= next_ipif
;
15235 ill
->ill_src_ipif
= NULL
;
15237 mutex_exit(&ipif
->ipif_ill
->ill_lock
);
15240 rw_exit(&ipst
->ips_ill_g_lock
);
15242 ill_refrele(usill
);
15243 if (ipmp_ill
!= NULL
)
15244 ill_refrele(ipmp_ill
);
15247 if (ipif
== NULL
) {
15248 char buf1
[INET6_ADDRSTRLEN
];
15250 ip1dbg(("ipif_select_source_v4(%s, %s) -> NULL\n",
15252 inet_ntop(AF_INET
, &dst
, buf1
, sizeof (buf1
))));
15254 char buf1
[INET6_ADDRSTRLEN
];
15255 char buf2
[INET6_ADDRSTRLEN
];
15257 ip1dbg(("ipif_select_source_v4(%s, %s) -> %s\n",
15258 ipif
->ipif_ill
->ill_name
,
15259 inet_ntop(AF_INET
, &dst
, buf1
, sizeof (buf1
)),
15260 inet_ntop(AF_INET
, &ipif
->ipif_lcl_addr
,
15261 buf2
, sizeof (buf2
))));
15268 * Pick a source address based on the destination ill and an optional setsrc
15270 * The result is stored in srcp. If generation is set, then put the source
15271 * generation number there before we look for the source address (to avoid
15272 * missing changes in the set of source addresses.
15273 * If flagsp is set, then us it to pass back ipif_flags.
15275 * If the caller wants to cache the returned source address and detect when
15276 * that might be stale, the caller should pass in a generation argument,
15277 * which the caller can later compare against ips_src_generation
15279 * The precedence order for selecting an IPv4 source address is:
15280 * - RTF_SETSRC on the offlink ire always wins.
15281 * - If usrsrc is set, swap the ill to be the usesrc one.
15282 * - If IPMP is used on the ill, select a random address from the most
15283 * preferred ones below:
15284 * 1. If onlink destination, same subnet and not deprecated, not ALL_ZONES
15285 * 2. Not deprecated, not ALL_ZONES
15286 * 3. If onlink destination, same subnet and not deprecated, ALL_ZONES
15287 * 4. Not deprecated, ALL_ZONES
15288 * 5. If onlink destination, same subnet and deprecated
15291 * We have lower preference for ALL_ZONES IP addresses,
15292 * as they pose problems with unlabeled destinations.
15294 * Note that when multiple IP addresses match e.g., #1 we pick
15295 * the first one if IPMP is not in use. With IPMP we randomize.
15298 ip_select_source_v4(ill_t
*ill
, ipaddr_t setsrc
, ipaddr_t dst
,
15299 ipaddr_t multicast_ifaddr
,
15300 zoneid_t zoneid
, ip_stack_t
*ipst
, ipaddr_t
*srcp
,
15301 uint32_t *generation
, uint64_t *flagsp
)
15304 boolean_t notready
= B_FALSE
; /* Set if !ipif_addr_ready found */
15306 if (flagsp
!= NULL
)
15310 * Need to grab the generation number before we check to
15311 * avoid a race with a change to the set of local addresses.
15312 * No lock needed since the thread which updates the set of local
15313 * addresses use ipif/ill locks and exit those (hence a store memory
15314 * barrier) before doing the atomic increase of ips_src_generation.
15316 if (generation
!= NULL
) {
15317 *generation
= ipst
->ips_src_generation
;
15320 if (CLASSD(dst
) && multicast_ifaddr
!= INADDR_ANY
) {
15321 *srcp
= multicast_ifaddr
;
15325 /* Was RTF_SETSRC set on the first IRE in the recursive lookup? */
15326 if (setsrc
!= INADDR_ANY
) {
15330 ipif
= ipif_select_source_v4(ill
, dst
, zoneid
, B_TRUE
, ¬ready
);
15331 if (ipif
== NULL
) {
15335 return (EADDRNOTAVAIL
);
15337 *srcp
= ipif
->ipif_lcl_addr
;
15338 if (flagsp
!= NULL
)
15339 *flagsp
= ipif
->ipif_flags
;
15340 ipif_refrele(ipif
);
15346 if_unitsel_restart(ipif_t
*ipif
, sin_t
*dummy_sin
, queue_t
*q
, mblk_t
*mp
,
15347 ip_ioctl_cmd_t
*ipip
, void *dummy_ifreq
)
15350 * ill_phyint_reinit merged the v4 and v6 into a single
15351 * ipsq. We might not have been able to complete the
15352 * operation in ipif_set_values, if we could not become
15353 * exclusive. If so restart it here.
15355 return (ipif_set_values_tail(ipif
->ipif_ill
, ipif
, mp
, q
));
15359 * Can operate on either a module or a driver queue.
15360 * Returns an error if not a module queue.
15364 if_unitsel(ipif_t
*dummy_ipif
, sin_t
*dummy_sin
, queue_t
*q
, mblk_t
*mp
,
15365 ip_ioctl_cmd_t
*ipip
, void *dummy_ifreq
)
15369 char interf_name
[LIFNAMSIZ
];
15370 uint_t ppa
= *(uint_t
*)mp
->b_cont
->b_cont
->b_rptr
;
15372 if (q
->q_next
== NULL
) {
15374 "if_unitsel: IF_UNITSEL: no q_next\n"));
15378 if (((ill_t
*)(q
->q_ptr
))->ill_name
[0] != '\0')
15383 } while (q1
->q_next
);
15384 cp
= q1
->q_qinfo
->qi_minfo
->mi_idname
;
15385 (void) sprintf(interf_name
, "%s%d", cp
, ppa
);
15388 * Here we are not going to delay the ioack until after
15389 * ACKs from DL_ATTACH_REQ/DL_BIND_REQ. So no need to save the
15390 * original ioctl message before sending the requests.
15392 return (ipif_set_values(q
, mp
, interf_name
, &ppa
));
15397 ip_sioctl_sifname(ipif_t
*dummy_ipif
, sin_t
*dummy_sin
, queue_t
*q
, mblk_t
*mp
,
15398 ip_ioctl_cmd_t
*ipip
, void *dummy_ifreq
)
15404 * Create any IRE_BROADCAST entries for `ipif', and store those entries in
15405 * `irep'. Returns a pointer to the next free `irep' entry
15406 * A mirror exists in ipif_delete_bcast_ires().
15408 * The management of any "extra" or seemingly duplicate IRE_BROADCASTs is
15412 ipif_create_bcast_ires(ipif_t
*ipif
, ire_t
**irep
)
15415 ipaddr_t netmask
= ip_net_mask(ipif
->ipif_lcl_addr
);
15416 ipaddr_t subnetmask
= ipif
->ipif_net_mask
;
15417 ill_t
*ill
= ipif
->ipif_ill
;
15418 zoneid_t zoneid
= ipif
->ipif_zoneid
;
15420 ip1dbg(("ipif_create_bcast_ires: creating broadcast IREs\n"));
15422 ASSERT(ipif
->ipif_flags
& IPIF_BROADCAST
);
15423 ASSERT(!(ipif
->ipif_flags
& IPIF_NOXMIT
));
15425 if (ipif
->ipif_lcl_addr
== INADDR_ANY
||
15426 (ipif
->ipif_flags
& IPIF_NOLOCAL
))
15427 netmask
= htonl(IN_CLASSA_NET
); /* fallback */
15429 irep
= ire_create_bcast(ill
, 0, zoneid
, irep
);
15430 irep
= ire_create_bcast(ill
, INADDR_BROADCAST
, zoneid
, irep
);
15433 * For backward compatibility, we create net broadcast IREs based on
15434 * the old "IP address class system", since some old machines only
15435 * respond to these class derived net broadcast. However, we must not
15436 * create these net broadcast IREs if the subnetmask is shorter than
15437 * the IP address class based derived netmask. Otherwise, we may
15438 * create a net broadcast address which is the same as an IP address
15439 * on the subnet -- and then TCP will refuse to talk to that address.
15441 if (netmask
< subnetmask
) {
15442 addr
= netmask
& ipif
->ipif_subnet
;
15443 irep
= ire_create_bcast(ill
, addr
, zoneid
, irep
);
15444 irep
= ire_create_bcast(ill
, ~netmask
| addr
, zoneid
, irep
);
15448 * Don't create IRE_BROADCAST IREs for the interface if the subnetmask
15449 * is 0xFFFFFFFF, as an IRE_LOCAL for that interface is already
15450 * created. Creating these broadcast IREs will only create confusion
15451 * as `addr' will be the same as the IP address.
15453 if (subnetmask
!= 0xFFFFFFFF) {
15454 addr
= ipif
->ipif_subnet
;
15455 irep
= ire_create_bcast(ill
, addr
, zoneid
, irep
);
15456 irep
= ire_create_bcast(ill
, ~subnetmask
| addr
, zoneid
, irep
);
15463 * Mirror of ipif_create_bcast_ires()
15466 ipif_delete_bcast_ires(ipif_t
*ipif
)
15469 ipaddr_t netmask
= ip_net_mask(ipif
->ipif_lcl_addr
);
15470 ipaddr_t subnetmask
= ipif
->ipif_net_mask
;
15471 ill_t
*ill
= ipif
->ipif_ill
;
15472 zoneid_t zoneid
= ipif
->ipif_zoneid
;
15475 ASSERT(ipif
->ipif_flags
& IPIF_BROADCAST
);
15476 ASSERT(!(ipif
->ipif_flags
& IPIF_NOXMIT
));
15478 if (ipif
->ipif_lcl_addr
== INADDR_ANY
||
15479 (ipif
->ipif_flags
& IPIF_NOLOCAL
))
15480 netmask
= htonl(IN_CLASSA_NET
); /* fallback */
15482 ire
= ire_lookup_bcast(ill
, 0, zoneid
);
15483 ASSERT(ire
!= NULL
);
15484 ire_delete(ire
); ire_refrele(ire
);
15485 ire
= ire_lookup_bcast(ill
, INADDR_BROADCAST
, zoneid
);
15486 ASSERT(ire
!= NULL
);
15487 ire_delete(ire
); ire_refrele(ire
);
15490 * For backward compatibility, we create net broadcast IREs based on
15491 * the old "IP address class system", since some old machines only
15492 * respond to these class derived net broadcast. However, we must not
15493 * create these net broadcast IREs if the subnetmask is shorter than
15494 * the IP address class based derived netmask. Otherwise, we may
15495 * create a net broadcast address which is the same as an IP address
15496 * on the subnet -- and then TCP will refuse to talk to that address.
15498 if (netmask
< subnetmask
) {
15499 addr
= netmask
& ipif
->ipif_subnet
;
15500 ire
= ire_lookup_bcast(ill
, addr
, zoneid
);
15501 ASSERT(ire
!= NULL
);
15502 ire_delete(ire
); ire_refrele(ire
);
15503 ire
= ire_lookup_bcast(ill
, ~netmask
| addr
, zoneid
);
15504 ASSERT(ire
!= NULL
);
15505 ire_delete(ire
); ire_refrele(ire
);
15509 * Don't create IRE_BROADCAST IREs for the interface if the subnetmask
15510 * is 0xFFFFFFFF, as an IRE_LOCAL for that interface is already
15511 * created. Creating these broadcast IREs will only create confusion
15512 * as `addr' will be the same as the IP address.
15514 if (subnetmask
!= 0xFFFFFFFF) {
15515 addr
= ipif
->ipif_subnet
;
15516 ire
= ire_lookup_bcast(ill
, addr
, zoneid
);
15517 ASSERT(ire
!= NULL
);
15518 ire_delete(ire
); ire_refrele(ire
);
15519 ire
= ire_lookup_bcast(ill
, ~subnetmask
| addr
, zoneid
);
15520 ASSERT(ire
!= NULL
);
15521 ire_delete(ire
); ire_refrele(ire
);
15526 * Extract both the flags (including IFF_CANTCHANGE) such as IFF_IPV*
15527 * from lifr_flags and the name from lifr_name.
15528 * Set IFF_IPV* and ill_isv6 prior to doing the lookup
15529 * since ipif_lookup_on_name uses the _isv6 flags when matching.
15530 * Returns EINPROGRESS when mp has been consumed by queueing it on
15531 * ipx_pending_mp and the ioctl will complete in ip_rput.
15533 * Can operate on either a module or a driver queue.
15534 * Returns an error if not a module queue.
15538 ip_sioctl_slifname(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
15539 ip_ioctl_cmd_t
*ipip
, void *if_req
)
15541 ill_t
*ill
= q
->q_ptr
;
15544 struct lifreq
*lifr
= if_req
;
15545 uint64_t new_flags
;
15547 ASSERT(ipif
!= NULL
);
15548 ip1dbg(("ip_sioctl_slifname %s\n", lifr
->lifr_name
));
15550 if (q
->q_next
== NULL
) {
15551 ip1dbg(("if_sioctl_slifname: SIOCSLIFNAME: no q_next\n"));
15556 * If we are not writer on 'q' then this interface exists already
15557 * and previous lookups (ip_extract_lifreq()) found this ipif --
15558 * so return EALREADY.
15560 if (ill
!= ipif
->ipif_ill
)
15563 if (ill
->ill_name
[0] != '\0')
15567 * If there's another ill already with the requested name, ensure
15568 * that it's of the same type. Otherwise, ill_phyint_reinit() will
15569 * fuse together two unrelated ills, which will cause chaos.
15571 ipst
= ill
->ill_ipst
;
15572 phyi
= avl_find(&ipst
->ips_phyint_g_list
->phyint_list_avl_by_name
,
15573 lifr
->lifr_name
, NULL
);
15574 if (phyi
!= NULL
) {
15575 ill_t
*ill_mate
= phyi
->phyint_illv4
;
15577 if (ill_mate
== NULL
)
15578 ill_mate
= phyi
->phyint_illv6
;
15579 ASSERT(ill_mate
!= NULL
);
15581 if (ill_mate
->ill_media
->ip_m_mac_type
!=
15582 ill
->ill_media
->ip_m_mac_type
) {
15583 ip1dbg(("if_sioctl_slifname: SIOCSLIFNAME: attempt to "
15584 "use the same ill name on differing media\n"));
15590 * We start off as IFF_IPV4 in ipif_allocate and become
15591 * IFF_IPV4 or IFF_IPV6 here depending on lifr_flags value.
15592 * The only flags that we read from user space are IFF_IPV4,
15593 * IFF_IPV6, and IFF_BROADCAST.
15595 * This ill has not been inserted into the global list.
15596 * So we are still single threaded and don't need any lock
15598 * Saniy check the flags.
15601 if ((lifr
->lifr_flags
& IFF_BROADCAST
) &&
15602 ((lifr
->lifr_flags
& IFF_IPV6
) ||
15603 (!ill
->ill_needs_attach
&& ill
->ill_bcast_addr_length
== 0))) {
15604 ip1dbg(("ip_sioctl_slifname: link not broadcast capable "
15605 "or IPv6 i.e., no broadcast \n"));
15610 lifr
->lifr_flags
& (IFF_IPV6
|IFF_IPV4
|IFF_BROADCAST
);
15612 if ((new_flags
^ (IFF_IPV6
|IFF_IPV4
)) == 0) {
15613 ip1dbg(("ip_sioctl_slifname: flags must be exactly one of "
15614 "IFF_IPV4 or IFF_IPV6\n"));
15619 * We always start off as IPv4, so only need to check for IPv6.
15621 if ((new_flags
& IFF_IPV6
) != 0) {
15622 ill
->ill_flags
|= ILLF_IPV6
;
15623 ill
->ill_flags
&= ~ILLF_IPV4
;
15625 if (lifr
->lifr_flags
& IFF_NOLINKLOCAL
)
15626 ill
->ill_flags
|= ILLF_NOLINKLOCAL
;
15629 if ((new_flags
& IFF_BROADCAST
) != 0)
15630 ipif
->ipif_flags
|= IPIF_BROADCAST
;
15632 ipif
->ipif_flags
&= ~IPIF_BROADCAST
;
15634 /* We started off as V4. */
15635 if (ill
->ill_flags
& ILLF_IPV6
) {
15636 ill
->ill_phyint
->phyint_illv6
= ill
;
15637 ill
->ill_phyint
->phyint_illv4
= NULL
;
15640 return (ipif_set_values(q
, mp
, lifr
->lifr_name
, &lifr
->lifr_ppa
));
15645 ip_sioctl_slifname_restart(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
15646 ip_ioctl_cmd_t
*ipip
, void *if_req
)
15649 * ill_phyint_reinit merged the v4 and v6 into a single
15650 * ipsq. We might not have been able to complete the
15651 * slifname in ipif_set_values, if we could not become
15652 * exclusive. If so restart it here
15654 return (ipif_set_values_tail(ipif
->ipif_ill
, ipif
, mp
, q
));
15658 * Return a pointer to the ipif which matches the index, IP version type and
15662 ipif_lookup_on_ifindex(uint_t index
, boolean_t isv6
, zoneid_t zoneid
,
15666 ipif_t
*ipif
= NULL
;
15668 ill
= ill_lookup_on_ifindex(index
, isv6
, ipst
);
15670 mutex_enter(&ill
->ill_lock
);
15671 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
;
15672 ipif
= ipif
->ipif_next
) {
15673 if (!IPIF_IS_CONDEMNED(ipif
) && (zoneid
== ALL_ZONES
||
15674 zoneid
== ipif
->ipif_zoneid
||
15675 ipif
->ipif_zoneid
== ALL_ZONES
)) {
15676 ipif_refhold_locked(ipif
);
15680 mutex_exit(&ill
->ill_lock
);
15687 * Change an existing physical interface's index. If the new index
15688 * is acceptable we update the index and the phyint_list_avl_by_index tree.
15689 * Finally, we update other systems which may have a dependence on the
15694 ip_sioctl_slifindex(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
15695 ip_ioctl_cmd_t
*ipip
, void *ifreq
)
15699 struct ifreq
*ifr
= (struct ifreq
*)ifreq
;
15700 struct lifreq
*lifr
= (struct lifreq
*)ifreq
;
15701 uint_t old_index
, index
;
15702 ip_stack_t
*ipst
= ipif
->ipif_ill
->ill_ipst
;
15705 if (ipip
->ipi_cmd_type
== IF_CMD
)
15706 index
= ifr
->ifr_index
;
15708 index
= lifr
->lifr_index
;
15711 * Only allow on physical interface. Also, index zero is illegal.
15713 ill
= ipif
->ipif_ill
;
15714 phyi
= ill
->ill_phyint
;
15715 if (ipif
->ipif_id
!= 0 || index
== 0 || index
> IF_INDEX_MAX
) {
15719 /* If the index is not changing, no work to do */
15720 if (phyi
->phyint_ifindex
== index
)
15724 * Use phyint_exists() to determine if the new interface index
15725 * is already in use. If the index is unused then we need to
15726 * change the phyint's position in the phyint_list_avl_by_index
15727 * tree. If we do not do this, subsequent lookups (using the new
15728 * index value) will not find the phyint.
15730 rw_enter(&ipst
->ips_ill_g_lock
, RW_WRITER
);
15731 if (phyint_exists(index
, ipst
)) {
15732 rw_exit(&ipst
->ips_ill_g_lock
);
15737 * The new index is unused. Set it in the phyint. However we must not
15738 * forget to trigger NE_IFINDEX_CHANGE event before the ifindex
15739 * changes. The event must be bound to old ifindex value.
15741 ill_nic_event_dispatch(ill
, 0, NE_IFINDEX_CHANGE
,
15742 &index
, sizeof (index
));
15744 old_index
= phyi
->phyint_ifindex
;
15745 phyi
->phyint_ifindex
= index
;
15747 avl_remove(&ipst
->ips_phyint_g_list
->phyint_list_avl_by_index
, phyi
);
15748 (void) avl_find(&ipst
->ips_phyint_g_list
->phyint_list_avl_by_index
,
15750 avl_insert(&ipst
->ips_phyint_g_list
->phyint_list_avl_by_index
,
15752 rw_exit(&ipst
->ips_ill_g_lock
);
15754 /* Update SCTP's ILL list */
15755 sctp_ill_reindex(ill
, old_index
);
15757 /* Send the routing sockets message */
15758 ip_rts_ifmsg(ipif
, RTSQ_DEFAULT
);
15759 if (ILL_OTHER(ill
))
15760 ip_rts_ifmsg(ILL_OTHER(ill
)->ill_ipif
, RTSQ_DEFAULT
);
15762 /* Perhaps ilgs should use this ill */
15763 update_conn_ill(NULL
, ill
->ill_ipst
);
15769 ip_sioctl_get_lifindex(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
15770 ip_ioctl_cmd_t
*ipip
, void *ifreq
)
15772 struct ifreq
*ifr
= (struct ifreq
*)ifreq
;
15773 struct lifreq
*lifr
= (struct lifreq
*)ifreq
;
15775 ip1dbg(("ip_sioctl_get_lifindex(%s:%u %p)\n",
15776 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
15777 /* Get the interface index */
15778 if (ipip
->ipi_cmd_type
== IF_CMD
) {
15779 ifr
->ifr_index
= ipif
->ipif_ill
->ill_phyint
->phyint_ifindex
;
15781 lifr
->lifr_index
= ipif
->ipif_ill
->ill_phyint
->phyint_ifindex
;
15788 ip_sioctl_get_lifzone(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
15789 ip_ioctl_cmd_t
*ipip
, void *ifreq
)
15791 struct lifreq
*lifr
= (struct lifreq
*)ifreq
;
15793 ip1dbg(("ip_sioctl_get_lifzone(%s:%u %p)\n",
15794 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
15795 /* Get the interface zone */
15796 ASSERT(ipip
->ipi_cmd_type
== LIF_CMD
);
15797 lifr
->lifr_zoneid
= ipif
->ipif_zoneid
;
15802 * Set the zoneid of an interface.
15806 ip_sioctl_slifzone(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
15807 ip_ioctl_cmd_t
*ipip
, void *ifreq
)
15809 struct lifreq
*lifr
= (struct lifreq
*)ifreq
;
15811 boolean_t need_up
= B_FALSE
;
15813 zone_status_t status
;
15816 ASSERT(ipip
->ipi_cmd_type
== LIF_CMD
);
15817 if ((zoneid
= lifr
->lifr_zoneid
) == ALL_ZONES
)
15820 /* cannot assign instance zero to a non-global zone */
15821 if (ipif
->ipif_id
== 0 && zoneid
!= GLOBAL_ZONEID
)
15825 * Cannot assign to a zone that doesn't exist or is shutting down. In
15826 * the event of a race with the zone shutdown processing, since IP
15827 * serializes this ioctl and SIOCGLIFCONF/SIOCLIFREMOVEIF, we know the
15828 * interface will be cleaned up even if the zone is shut down
15829 * immediately after the status check. If the interface can't be brought
15830 * down right away, and the zone is shut down before the restart
15831 * function is called, we resolve the possible races by rechecking the
15832 * zone status in the restart function.
15834 if ((zptr
= zone_find_by_id(zoneid
)) == NULL
)
15836 status
= zone_status_get(zptr
);
15839 if (status
!= ZONE_IS_READY
&& status
!= ZONE_IS_RUNNING
)
15842 if (ipif
->ipif_flags
& IPIF_UP
) {
15844 * If the interface is already marked up,
15845 * we call ipif_down which will take care
15846 * of ditching any IREs that have been set
15847 * up based on the old interface address.
15849 err
= ipif_logical_down(ipif
, q
, mp
);
15850 if (err
== EINPROGRESS
)
15852 (void) ipif_down_tail(ipif
);
15856 err
= ip_sioctl_slifzone_tail(ipif
, lifr
->lifr_zoneid
, q
, mp
, need_up
);
15861 ip_sioctl_slifzone_tail(ipif_t
*ipif
, zoneid_t zoneid
,
15862 queue_t
*q
, mblk_t
*mp
, boolean_t need_up
)
15867 ip1dbg(("ip_sioctl_zoneid_tail(%s:%u %p)\n",
15868 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
15871 ipst
= CONNQ_TO_IPST(q
);
15873 ipst
= ILLQ_TO_IPST(q
);
15876 * For exclusive stacks we don't allow a different zoneid than
15879 if (ipst
->ips_netstack
->netstack_stackid
!= GLOBAL_NETSTACKID
&&
15880 zoneid
!= GLOBAL_ZONEID
)
15883 /* Set the new zone id. */
15884 ipif
->ipif_zoneid
= zoneid
;
15886 /* Update sctp list */
15887 sctp_update_ipif(ipif
, SCTP_IPIF_UPDATE
);
15889 /* The default multicast interface might have changed */
15890 ire_increment_multicast_generation(ipst
, ipif
->ipif_ill
->ill_isv6
);
15894 * Now bring the interface back up. If this
15895 * is the only IPIF for the ILL, ipif_up
15896 * will have to re-bind to the device, so
15897 * we may get back EINPROGRESS, in which
15898 * case, this IOCTL will get completed in
15899 * ip_rput_dlpi when we see the DL_BIND_ACK.
15901 err
= ipif_up(ipif
, q
, mp
);
15908 ip_sioctl_slifzone_restart(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
15909 ip_ioctl_cmd_t
*ipip
, void *if_req
)
15911 struct lifreq
*lifr
= (struct lifreq
*)if_req
;
15914 zone_status_t status
;
15916 ASSERT(ipip
->ipi_cmd_type
== LIF_CMD
);
15917 if ((zoneid
= lifr
->lifr_zoneid
) == ALL_ZONES
)
15918 zoneid
= GLOBAL_ZONEID
;
15920 ip1dbg(("ip_sioctl_slifzone_restart(%s:%u %p)\n",
15921 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
15924 * We recheck the zone status to resolve the following race condition:
15925 * 1) process sends SIOCSLIFZONE to put hme0:1 in zone "myzone";
15926 * 2) hme0:1 is up and can't be brought down right away;
15927 * ip_sioctl_slifzone() returns EINPROGRESS and the request is queued;
15928 * 3) zone "myzone" is halted; the zone status switches to
15929 * 'shutting_down' and the zones framework sends SIOCGLIFCONF to list
15930 * the interfaces to remove - hme0:1 is not returned because it's not
15931 * yet in "myzone", so it won't be removed;
15932 * 4) the restart function for SIOCSLIFZONE is called; without the
15933 * status check here, we would have hme0:1 in "myzone" after it's been
15935 * Note that if the status check fails, we need to bring the interface
15936 * back to its state prior to ip_sioctl_slifzone(), hence the call to
15937 * ipif_up_done[_v6]().
15939 status
= ZONE_IS_UNINITIALIZED
;
15940 if ((zptr
= zone_find_by_id(zoneid
)) != NULL
) {
15941 status
= zone_status_get(zptr
);
15944 if (status
!= ZONE_IS_READY
&& status
!= ZONE_IS_RUNNING
) {
15945 if (ipif
->ipif_isv6
) {
15946 (void) ipif_up_done_v6(ipif
);
15948 (void) ipif_up_done(ipif
);
15953 (void) ipif_down_tail(ipif
);
15955 return (ip_sioctl_slifzone_tail(ipif
, lifr
->lifr_zoneid
, q
, mp
,
15960 * Return the number of addresses on `ill' with one or more of the values
15961 * in `set' set and all of the values in `clear' clear.
15964 ill_flagaddr_cnt(const ill_t
*ill
, uint64_t set
, uint64_t clear
)
15969 ASSERT(IAM_WRITER_ILL(ill
));
15971 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
; ipif
= ipif
->ipif_next
)
15972 if ((ipif
->ipif_flags
& set
) && !(ipif
->ipif_flags
& clear
))
15979 * Return the number of migratable addresses on `ill' that are under
15980 * application control.
15983 ill_appaddr_cnt(const ill_t
*ill
)
15985 return (ill_flagaddr_cnt(ill
, IPIF_DHCPRUNNING
| IPIF_ADDRCONF
,
15990 * Return the number of point-to-point addresses on `ill'.
15993 ill_ptpaddr_cnt(const ill_t
*ill
)
15995 return (ill_flagaddr_cnt(ill
, IPIF_POINTOPOINT
, 0));
16000 ip_sioctl_get_lifusesrc(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
16001 ip_ioctl_cmd_t
*ipip
, void *ifreq
)
16003 struct lifreq
*lifr
= ifreq
;
16005 ASSERT(q
->q_next
== NULL
);
16008 ip1dbg(("ip_sioctl_get_lifusesrc(%s:%u %p)\n",
16009 ipif
->ipif_ill
->ill_name
, ipif
->ipif_id
, (void *)ipif
));
16010 lifr
->lifr_index
= ipif
->ipif_ill
->ill_usesrc_ifindex
;
16011 ip1dbg(("ip_sioctl_get_lifusesrc:lifr_index = %d\n", lifr
->lifr_index
));
16016 /* Find the previous ILL in this usesrc group */
16018 ill_prev_usesrc(ill_t
*uill
)
16022 for (ill
= uill
->ill_usesrc_grp_next
;
16023 ASSERT(ill
), ill
->ill_usesrc_grp_next
!= uill
;
16024 ill
= ill
->ill_usesrc_grp_next
)
16030 * Release all members of the usesrc group. This routine is called
16031 * from ill_delete when the interface being unplumbed is the
16034 * This silently clears the usesrc that ifconfig setup.
16035 * An alternative would be to keep that ifindex, and drop packets on the floor
16036 * since no source address can be selected.
16037 * Even if we keep the current semantics, don't need a lock and a linked list.
16038 * Can walk all the ills checking if they have a ill_usesrc_ifindex matching
16039 * the one that is being removed. Issue is how we return the usesrc users
16040 * (SIOCGLIFSRCOF). We want to be able to find the ills which have an
16041 * ill_usesrc_ifindex matching a target ill. We could also do that with an
16042 * ill walk, but the walker would need to insert in the ioctl response.
16045 ill_disband_usesrc_group(ill_t
*uill
)
16047 ill_t
*next_ill
, *tmp_ill
;
16048 ip_stack_t
*ipst
= uill
->ill_ipst
;
16050 ASSERT(RW_WRITE_HELD(&ipst
->ips_ill_g_usesrc_lock
));
16051 next_ill
= uill
->ill_usesrc_grp_next
;
16054 ASSERT(next_ill
!= NULL
);
16055 tmp_ill
= next_ill
->ill_usesrc_grp_next
;
16056 ASSERT(tmp_ill
!= NULL
);
16057 next_ill
->ill_usesrc_grp_next
= NULL
;
16058 next_ill
->ill_usesrc_ifindex
= 0;
16059 next_ill
= tmp_ill
;
16060 } while (next_ill
->ill_usesrc_ifindex
!= 0);
16061 uill
->ill_usesrc_grp_next
= NULL
;
16065 * Remove the client usesrc ILL from the list and relink to a new list
16068 ill_relink_usesrc_ills(ill_t
*ucill
, ill_t
*uill
, uint_t ifindex
)
16070 ill_t
*ill
, *tmp_ill
;
16071 ip_stack_t
*ipst
= ucill
->ill_ipst
;
16073 ASSERT((ucill
!= NULL
) && (ucill
->ill_usesrc_grp_next
!= NULL
) &&
16074 (uill
!= NULL
) && RW_WRITE_HELD(&ipst
->ips_ill_g_usesrc_lock
));
16077 * Check if the usesrc client ILL passed in is not already
16078 * in use as a usesrc ILL i.e one whose source address is
16079 * in use OR a usesrc ILL is not already in use as a usesrc
16082 if ((ucill
->ill_usesrc_ifindex
== 0) ||
16083 (uill
->ill_usesrc_ifindex
!= 0)) {
16087 ill
= ill_prev_usesrc(ucill
);
16088 ASSERT(ill
->ill_usesrc_grp_next
!= NULL
);
16090 /* Remove from the current list */
16091 if (ill
->ill_usesrc_grp_next
->ill_usesrc_grp_next
== ill
) {
16092 /* Only two elements in the list */
16093 ASSERT(ill
->ill_usesrc_ifindex
== 0);
16094 ill
->ill_usesrc_grp_next
= NULL
;
16096 ill
->ill_usesrc_grp_next
= ucill
->ill_usesrc_grp_next
;
16099 if (ifindex
== 0) {
16100 ucill
->ill_usesrc_ifindex
= 0;
16101 ucill
->ill_usesrc_grp_next
= NULL
;
16105 ucill
->ill_usesrc_ifindex
= ifindex
;
16106 tmp_ill
= uill
->ill_usesrc_grp_next
;
16107 uill
->ill_usesrc_grp_next
= ucill
;
16108 ucill
->ill_usesrc_grp_next
=
16109 (tmp_ill
!= NULL
) ? tmp_ill
: uill
;
16114 * Set the ill_usesrc and ill_usesrc_head fields. See synchronization notes in
16115 * ip.c for locking details.
16119 ip_sioctl_slifusesrc(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
16120 ip_ioctl_cmd_t
*ipip
, void *ifreq
)
16122 struct lifreq
*lifr
= (struct lifreq
*)ifreq
;
16123 boolean_t isv6
= B_FALSE
, reset_flg
= B_FALSE
;
16124 ill_t
*usesrc_ill
, *usesrc_cli_ill
= ipif
->ipif_ill
;
16127 ipsq_t
*ipsq
= NULL
;
16128 ip_stack_t
*ipst
= ipif
->ipif_ill
->ill_ipst
;
16130 ASSERT(IAM_WRITER_IPIF(ipif
));
16131 ASSERT(q
->q_next
== NULL
);
16134 isv6
= (Q_TO_CONN(q
))->conn_family
== AF_INET6
;
16136 ifindex
= lifr
->lifr_index
;
16137 if (ifindex
== 0) {
16138 if (usesrc_cli_ill
->ill_usesrc_grp_next
== NULL
) {
16139 /* non usesrc group interface, nothing to reset */
16142 ifindex
= usesrc_cli_ill
->ill_usesrc_ifindex
;
16143 /* valid reset request */
16144 reset_flg
= B_TRUE
;
16147 usesrc_ill
= ill_lookup_on_ifindex(ifindex
, isv6
, ipst
);
16148 if (usesrc_ill
== NULL
)
16150 if (usesrc_ill
== ipif
->ipif_ill
) {
16151 ill_refrele(usesrc_ill
);
16155 ipsq
= ipsq_try_enter(NULL
, usesrc_ill
, q
, mp
, ip_process_ioctl
,
16157 if (ipsq
== NULL
) {
16159 /* Operation enqueued on the ipsq of the usesrc ILL */
16163 /* USESRC isn't currently supported with IPMP */
16164 if (IS_IPMP(usesrc_ill
) || IS_UNDER_IPMP(usesrc_ill
)) {
16170 * USESRC isn't compatible with the STANDBY flag. (STANDBY is only
16171 * used by IPMP underlying interfaces, but someone might think it's
16172 * more general and try to use it independently with VNI.)
16174 if (usesrc_ill
->ill_phyint
->phyint_flags
& PHYI_STANDBY
) {
16180 * If the client is already in use as a usesrc_ill or a usesrc_ill is
16181 * already a client then return EINVAL
16183 if (IS_USESRC_ILL(usesrc_cli_ill
) || IS_USESRC_CLI_ILL(usesrc_ill
)) {
16189 * If the ill_usesrc_ifindex field is already set to what it needs to
16190 * be then this is a duplicate operation.
16192 if (!reset_flg
&& usesrc_cli_ill
->ill_usesrc_ifindex
== ifindex
) {
16197 ip1dbg(("ip_sioctl_slifusesrc: usesrc_cli_ill %s, usesrc_ill %s,"
16198 " v6 = %d", usesrc_cli_ill
->ill_name
, usesrc_ill
->ill_name
,
16199 usesrc_ill
->ill_isv6
));
16202 * ill_g_usesrc_lock global lock protects the ill_usesrc_grp_next
16203 * and the ill_usesrc_ifindex fields
16205 rw_enter(&ipst
->ips_ill_g_usesrc_lock
, RW_WRITER
);
16208 ret
= ill_relink_usesrc_ills(usesrc_cli_ill
, usesrc_ill
, 0);
16212 rw_exit(&ipst
->ips_ill_g_usesrc_lock
);
16217 * Four possibilities to consider:
16218 * 1. Both usesrc_ill and usesrc_cli_ill are not part of any usesrc grp
16219 * 2. usesrc_ill is part of a group but usesrc_cli_ill isn't
16220 * 3. usesrc_cli_ill is part of a group but usesrc_ill isn't
16221 * 4. Both are part of their respective usesrc groups
16223 if ((usesrc_ill
->ill_usesrc_grp_next
== NULL
) &&
16224 (usesrc_cli_ill
->ill_usesrc_grp_next
== NULL
)) {
16225 ASSERT(usesrc_ill
->ill_usesrc_ifindex
== 0);
16226 usesrc_cli_ill
->ill_usesrc_ifindex
= ifindex
;
16227 usesrc_ill
->ill_usesrc_grp_next
= usesrc_cli_ill
;
16228 usesrc_cli_ill
->ill_usesrc_grp_next
= usesrc_ill
;
16229 } else if ((usesrc_ill
->ill_usesrc_grp_next
!= NULL
) &&
16230 (usesrc_cli_ill
->ill_usesrc_grp_next
== NULL
)) {
16231 usesrc_cli_ill
->ill_usesrc_ifindex
= ifindex
;
16232 /* Insert at head of list */
16233 usesrc_cli_ill
->ill_usesrc_grp_next
=
16234 usesrc_ill
->ill_usesrc_grp_next
;
16235 usesrc_ill
->ill_usesrc_grp_next
= usesrc_cli_ill
;
16237 ret
= ill_relink_usesrc_ills(usesrc_cli_ill
, usesrc_ill
,
16242 rw_exit(&ipst
->ips_ill_g_usesrc_lock
);
16247 /* The refrele on the lifr_name ipif is done by ip_process_ioctl */
16248 ill_refrele(usesrc_ill
);
16250 /* Let conn_ixa caching know that source address selection changed */
16251 ip_update_source_selection(ipst
);
16258 ip_sioctl_get_dadstate(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
16259 ip_ioctl_cmd_t
*ipip
, void *if_req
)
16261 struct lifreq
*lifr
= (struct lifreq
*)if_req
;
16262 ill_t
*ill
= ipif
->ipif_ill
;
16265 * Need a lock since IFF_UP can be set even when there are
16266 * references to the ipif.
16268 mutex_enter(&ill
->ill_lock
);
16269 if ((ipif
->ipif_flags
& IPIF_UP
) && ipif
->ipif_addr_ready
== 0)
16270 lifr
->lifr_dadstate
= DAD_IN_PROGRESS
;
16272 lifr
->lifr_dadstate
= DAD_DONE
;
16273 mutex_exit(&ill
->ill_lock
);
16278 * comparison function used by avl.
16281 ill_phyint_compare_index(const void *index_ptr
, const void *phyip
)
16286 ASSERT(phyip
!= NULL
&& index_ptr
!= NULL
);
16288 index
= *((uint_t
*)index_ptr
);
16290 * let the phyint with the lowest index be on top.
16292 if (((phyint_t
*)phyip
)->phyint_ifindex
< index
)
16294 if (((phyint_t
*)phyip
)->phyint_ifindex
> index
)
16300 * comparison function used by avl.
16303 ill_phyint_compare_name(const void *name_ptr
, const void *phyip
)
16308 ASSERT(phyip
!= NULL
&& name_ptr
!= NULL
);
16310 if (((phyint_t
*)phyip
)->phyint_illv4
)
16311 ill
= ((phyint_t
*)phyip
)->phyint_illv4
;
16313 ill
= ((phyint_t
*)phyip
)->phyint_illv6
;
16314 ASSERT(ill
!= NULL
);
16316 res
= strcmp(ill
->ill_name
, (char *)name_ptr
);
16325 * This function is called on the unplumb path via ill_glist_delete() when
16326 * there are no ills left on the phyint and thus the phyint can be freed.
16329 phyint_free(phyint_t
*phyi
)
16331 ip_stack_t
*ipst
= PHYINT_TO_IPST(phyi
);
16333 ASSERT(phyi
->phyint_illv4
== NULL
&& phyi
->phyint_illv6
== NULL
);
16336 * If this phyint was an IPMP meta-interface, blow away the group.
16337 * This is safe to do because all of the illgrps have already been
16338 * removed by I_PUNLINK, and thus SIOCSLIFGROUPNAME cannot find us.
16339 * If we're cleaning up as a result of failed initialization,
16340 * phyint_grp may be NULL.
16342 if ((phyi
->phyint_flags
& PHYI_IPMP
) && (phyi
->phyint_grp
!= NULL
)) {
16343 rw_enter(&ipst
->ips_ipmp_lock
, RW_WRITER
);
16344 ipmp_grp_destroy(phyi
->phyint_grp
);
16345 phyi
->phyint_grp
= NULL
;
16346 rw_exit(&ipst
->ips_ipmp_lock
);
16350 * If this interface was under IPMP, take it out of the group.
16352 if (phyi
->phyint_grp
!= NULL
)
16353 ipmp_phyint_leave_grp(phyi
);
16356 * Delete the phyint and disassociate its ipsq. The ipsq itself
16357 * will be freed in ipsq_exit().
16359 phyi
->phyint_ipsq
->ipsq_phyint
= NULL
;
16360 phyi
->phyint_name
[0] = '\0';
16366 * Attach the ill to the phyint structure which can be shared by both
16367 * IPv4 and IPv6 ill. ill_init allocates a phyint to just hold flags. This
16368 * function is called from ipif_set_values and ill_lookup_on_name (for
16369 * loopback) where we know the name of the ill. We lookup the ill and if
16370 * there is one present already with the name use that phyint. Otherwise
16371 * reuse the one allocated by ill_init.
16374 ill_phyint_reinit(ill_t
*ill
)
16376 boolean_t isv6
= ill
->ill_isv6
;
16377 phyint_t
*phyi_old
;
16379 avl_index_t where
= 0;
16380 ill_t
*ill_other
= NULL
;
16381 ip_stack_t
*ipst
= ill
->ill_ipst
;
16383 ASSERT(RW_WRITE_HELD(&ipst
->ips_ill_g_lock
));
16385 phyi_old
= ill
->ill_phyint
;
16386 ASSERT(isv6
|| (phyi_old
->phyint_illv4
== ill
&&
16387 phyi_old
->phyint_illv6
== NULL
));
16388 ASSERT(!isv6
|| (phyi_old
->phyint_illv6
== ill
&&
16389 phyi_old
->phyint_illv4
== NULL
));
16390 ASSERT(phyi_old
->phyint_ifindex
== 0);
16393 * Now that our ill has a name, set it in the phyint.
16395 (void) strlcpy(ill
->ill_phyint
->phyint_name
, ill
->ill_name
, LIFNAMSIZ
);
16397 phyi
= avl_find(&ipst
->ips_phyint_g_list
->phyint_list_avl_by_name
,
16398 ill
->ill_name
, &where
);
16401 * 1. We grabbed the ill_g_lock before inserting this ill into
16402 * the global list of ills. So no other thread could have located
16403 * this ill and hence the ipsq of this ill is guaranteed to be empty.
16404 * 2. Now locate the other protocol instance of this ill.
16405 * 3. Now grab both ill locks in the right order, and the phyint lock of
16406 * the new ipsq. Holding ill locks + ill_g_lock ensures that the ipsq
16407 * of neither ill can change.
16408 * 4. Merge the phyint and thus the ipsq as well of this ill onto the
16410 * 5. Release all locks.
16414 * Look for IPv4 if we are initializing IPv6 or look for IPv6 if
16415 * we are initializing IPv4.
16417 if (phyi
!= NULL
) {
16418 ill_other
= (isv6
) ? phyi
->phyint_illv4
: phyi
->phyint_illv6
;
16419 ASSERT(ill_other
->ill_phyint
!= NULL
);
16420 ASSERT((isv6
&& !ill_other
->ill_isv6
) ||
16421 (!isv6
&& ill_other
->ill_isv6
));
16422 GRAB_ILL_LOCKS(ill
, ill_other
);
16424 * We are potentially throwing away phyint_flags which
16425 * could be different from the one that we obtain from
16426 * ill_other->ill_phyint. But it is okay as we are assuming
16427 * that the state maintained within IP is correct.
16429 mutex_enter(&phyi
->phyint_lock
);
16431 ASSERT(phyi
->phyint_illv6
== NULL
);
16432 phyi
->phyint_illv6
= ill
;
16434 ASSERT(phyi
->phyint_illv4
== NULL
);
16435 phyi
->phyint_illv4
= ill
;
16439 * Delete the old phyint and make its ipsq eligible
16440 * to be freed in ipsq_exit().
16442 phyi_old
->phyint_illv4
= NULL
;
16443 phyi_old
->phyint_illv6
= NULL
;
16444 phyi_old
->phyint_ipsq
->ipsq_phyint
= NULL
;
16445 phyi_old
->phyint_name
[0] = '\0';
16448 mutex_enter(&ill
->ill_lock
);
16450 * We don't need to acquire any lock, since
16451 * the ill is not yet visible globally and we
16452 * have not yet released the ill_g_lock.
16455 mutex_enter(&phyi
->phyint_lock
);
16456 /* XXX We need a recovery strategy here. */
16457 if (!phyint_assign_ifindex(phyi
, ipst
))
16458 cmn_err(CE_PANIC
, "phyint_assign_ifindex() failed");
16460 avl_insert(&ipst
->ips_phyint_g_list
->phyint_list_avl_by_name
,
16461 (void *)phyi
, where
);
16463 (void) avl_find(&ipst
->ips_phyint_g_list
->
16464 phyint_list_avl_by_index
,
16465 &phyi
->phyint_ifindex
, &where
);
16466 avl_insert(&ipst
->ips_phyint_g_list
->phyint_list_avl_by_index
,
16467 (void *)phyi
, where
);
16471 * Reassigning ill_phyint automatically reassigns the ipsq also.
16472 * pending mp is not affected because that is per ill basis.
16474 ill
->ill_phyint
= phyi
;
16477 * Now that the phyint's ifindex has been assigned, complete the
16480 ill
->ill_ip_mib
->ipIfStatsIfIndex
= ill
->ill_phyint
->phyint_ifindex
;
16481 if (ill
->ill_isv6
) {
16482 ill
->ill_icmp6_mib
->ipv6IfIcmpIfIndex
=
16483 ill
->ill_phyint
->phyint_ifindex
;
16484 ill
->ill_mcast_type
= ipst
->ips_mld_max_version
;
16486 ill
->ill_mcast_type
= ipst
->ips_igmp_max_version
;
16490 * Generate an event within the hooks framework to indicate that
16491 * a new interface has just been added to IP. For this event to
16492 * be generated, the network interface must, at least, have an
16493 * ifindex assigned to it. (We don't generate the event for
16494 * loopback since ill_lookup_on_name() has its own NE_PLUMB event.)
16496 * This needs to be run inside the ill_g_lock perimeter to ensure
16497 * that the ordering of delivered events to listeners matches the
16498 * order of them in the kernel.
16500 if (!IS_LOOPBACK(ill
)) {
16501 ill_nic_event_dispatch(ill
, 0, NE_PLUMB
, ill
->ill_name
,
16502 ill
->ill_name_length
);
16504 RELEASE_ILL_LOCKS(ill
, ill_other
);
16505 mutex_exit(&phyi
->phyint_lock
);
16509 * Notify any downstream modules of the name of this interface.
16510 * An M_IOCTL is used even though we don't expect a successful reply.
16511 * Any reply message from the driver (presumably an M_IOCNAK) will
16512 * eventually get discarded somewhere upstream. The message format is
16513 * simply an SIOCSLIFNAME ioctl just as might be sent from ifconfig
16517 ip_ifname_notify(ill_t
*ill
, queue_t
*q
)
16520 struct iocblk
*iocp
;
16521 struct lifreq
*lifr
;
16523 mp1
= mkiocb(SIOCSLIFNAME
);
16526 mp2
= allocb(sizeof (struct lifreq
), BPRI_HI
);
16533 iocp
= (struct iocblk
*)mp1
->b_rptr
;
16534 iocp
->ioc_count
= sizeof (struct lifreq
);
16536 lifr
= (struct lifreq
*)mp2
->b_rptr
;
16537 mp2
->b_wptr
+= sizeof (struct lifreq
);
16538 bzero(lifr
, sizeof (struct lifreq
));
16540 (void) strncpy(lifr
->lifr_name
, ill
->ill_name
, LIFNAMSIZ
);
16541 lifr
->lifr_ppa
= ill
->ill_ppa
;
16542 lifr
->lifr_flags
= (ill
->ill_flags
& (ILLF_IPV4
|ILLF_IPV6
));
16544 DTRACE_PROBE3(ill__dlpi
, char *, "ip_ifname_notify",
16545 char *, "SIOCSLIFNAME", ill_t
*, ill
);
16550 ipif_set_values_tail(ill_t
*ill
, ipif_t
*ipif
, mblk_t
*mp
, queue_t
*q
)
16553 ip_stack_t
*ipst
= ill
->ill_ipst
;
16554 phyint_t
*phyi
= ill
->ill_phyint
;
16557 * Now that ill_name is set, the configuration for the IPMP
16558 * meta-interface can be performed.
16560 if (IS_IPMP(ill
)) {
16561 rw_enter(&ipst
->ips_ipmp_lock
, RW_WRITER
);
16563 * If phyi->phyint_grp is NULL, then this is the first IPMP
16564 * meta-interface and we need to create the IPMP group.
16566 if (phyi
->phyint_grp
== NULL
) {
16568 * If someone has renamed another IPMP group to have
16569 * the same name as our interface, bail.
16571 if (ipmp_grp_lookup(ill
->ill_name
, ipst
) != NULL
) {
16572 rw_exit(&ipst
->ips_ipmp_lock
);
16575 phyi
->phyint_grp
= ipmp_grp_create(ill
->ill_name
, phyi
);
16576 if (phyi
->phyint_grp
== NULL
) {
16577 rw_exit(&ipst
->ips_ipmp_lock
);
16581 rw_exit(&ipst
->ips_ipmp_lock
);
16584 /* Tell downstream modules where they are. */
16585 ip_ifname_notify(ill
, q
);
16588 * ill_dl_phys returns EINPROGRESS in the usual case.
16589 * Error cases are ENOMEM ...
16591 err
= ill_dl_phys(ill
, ipif
, mp
, q
);
16593 if (ill
->ill_isv6
) {
16594 mutex_enter(&ipst
->ips_mld_slowtimeout_lock
);
16595 if (ipst
->ips_mld_slowtimeout_id
== 0) {
16596 ipst
->ips_mld_slowtimeout_id
= timeout(mld_slowtimo
,
16598 MSEC_TO_TICK(MCAST_SLOWTIMO_INTERVAL
));
16600 mutex_exit(&ipst
->ips_mld_slowtimeout_lock
);
16602 mutex_enter(&ipst
->ips_igmp_slowtimeout_lock
);
16603 if (ipst
->ips_igmp_slowtimeout_id
== 0) {
16604 ipst
->ips_igmp_slowtimeout_id
= timeout(igmp_slowtimo
,
16606 MSEC_TO_TICK(MCAST_SLOWTIMO_INTERVAL
));
16608 mutex_exit(&ipst
->ips_igmp_slowtimeout_lock
);
16615 * Common routine for ppa and ifname setting. Should be called exclusive.
16617 * Returns EINPROGRESS when mp has been consumed by queueing it on
16618 * ipx_pending_mp and the ioctl will complete in ip_rput.
16620 * NOTE : If ppa is UNIT_MAX, we assign the next valid ppa and return
16621 * the new name and new ppa in lifr_name and lifr_ppa respectively.
16622 * For SLIFNAME, we pass these values back to the userland.
16625 ipif_set_values(queue_t
*q
, mblk_t
*mp
, char *interf_name
, uint_t
*new_ppa_ptr
)
16636 ip1dbg(("ipif_set_values: interface %s\n", interf_name
));
16637 ASSERT(q
->q_next
!= NULL
);
16638 ASSERT(interf_name
!= NULL
);
16640 ill
= (ill_t
*)q
->q_ptr
;
16641 ipst
= ill
->ill_ipst
;
16643 ASSERT(ill
->ill_ipst
!= NULL
);
16644 ASSERT(ill
->ill_name
[0] == '\0');
16645 ASSERT(IAM_WRITER_ILL(ill
));
16646 ASSERT((mi_strlen(interf_name
) + 1) <= LIFNAMSIZ
);
16647 ASSERT(ill
->ill_ppa
== UINT_MAX
);
16649 ill
->ill_defend_start
= ill
->ill_defend_count
= 0;
16650 /* The ppa is sent down by ifconfig or is chosen */
16651 if ((ppa_ptr
= ill_get_ppa_ptr(interf_name
)) == NULL
) {
16656 * make sure ppa passed in is same as ppa in the name.
16657 * This check is not made when ppa == UINT_MAX in that case ppa
16658 * in the name could be anything. System will choose a ppa and
16659 * update new_ppa_ptr and inter_name to contain the choosen ppa.
16661 if (*new_ppa_ptr
!= UINT_MAX
) {
16662 /* stoi changes the pointer */
16665 * ifconfig passed in 0 for the ppa for DLPI 1 style devices
16666 * (they don't have an externally visible ppa). We assign one
16667 * here so that we can manage the interface. Note that in
16668 * the past this value was always 0 for DLPI 1 drivers.
16670 if (*new_ppa_ptr
== 0)
16671 *new_ppa_ptr
= stoi(&old_ptr
);
16672 else if (*new_ppa_ptr
!= (uint_t
)stoi(&old_ptr
))
16676 * terminate string before ppa
16677 * save char at that location.
16679 old_char
= ppa_ptr
[0];
16682 ill
->ill_ppa
= *new_ppa_ptr
;
16684 * Finish as much work now as possible before calling ill_glist_insert
16685 * which makes the ill globally visible and also merges it with the
16686 * other protocol instance of this phyint. The remaining work is
16687 * done after entering the ipsq which may happen sometime later.
16689 ipif
= ill
->ill_ipif
;
16691 /* We didn't do this when we allocated ipif in ip_ll_subnet_defaults */
16692 ipif_assign_seqid(ipif
);
16694 if (!(ill
->ill_flags
& (ILLF_IPV4
|ILLF_IPV6
)))
16695 ill
->ill_flags
|= ILLF_IPV4
;
16697 ASSERT(ipif
->ipif_next
== NULL
); /* Only one ipif on ill */
16698 ASSERT((ipif
->ipif_flags
& IPIF_UP
) == 0);
16700 if (ill
->ill_flags
& ILLF_IPV6
) {
16702 ill
->ill_isv6
= B_TRUE
;
16703 ill_set_inputfn(ill
);
16704 if (ill
->ill_rq
!= NULL
) {
16705 ill
->ill_rq
->q_qinfo
= &iprinitv6
;
16708 /* Keep the !IN6_IS_ADDR_V4MAPPED assertions happy */
16709 ipif
->ipif_v6lcl_addr
= ipv6_all_zeros
;
16710 ipif
->ipif_v6subnet
= ipv6_all_zeros
;
16711 ipif
->ipif_v6net_mask
= ipv6_all_zeros
;
16712 ipif
->ipif_v6brd_addr
= ipv6_all_zeros
;
16713 ipif
->ipif_v6pp_dst_addr
= ipv6_all_zeros
;
16714 ill
->ill_reachable_retrans_time
= ND_RETRANS_TIMER
;
16716 * point-to-point or Non-mulicast capable
16717 * interfaces won't do NUD unless explicitly
16718 * configured to do so.
16720 if (ipif
->ipif_flags
& IPIF_POINTOPOINT
||
16721 !(ill
->ill_flags
& ILLF_MULTICAST
)) {
16722 ill
->ill_flags
|= ILLF_NONUD
;
16724 /* Make sure IPv4 specific flag is not set on IPv6 if */
16725 if (ill
->ill_flags
& ILLF_NOARP
) {
16727 * Note: xresolv interfaces will eventually need
16728 * NOARP set here as well, but that will require
16729 * those external resolvers to have some
16730 * knowledge of that flag and act appropriately.
16731 * Not to be changed at present.
16733 ill
->ill_flags
&= ~ILLF_NOARP
;
16736 * Set the ILLF_ROUTER flag according to the global
16737 * IPv6 forwarding policy.
16739 if (ipst
->ips_ipv6_forwarding
!= 0)
16740 ill
->ill_flags
|= ILLF_ROUTER
;
16741 } else if (ill
->ill_flags
& ILLF_IPV4
) {
16742 ill
->ill_isv6
= B_FALSE
;
16743 ill_set_inputfn(ill
);
16744 ill
->ill_reachable_retrans_time
= ARP_RETRANS_TIMER
;
16745 IN6_IPADDR_TO_V4MAPPED(INADDR_ANY
, &ipif
->ipif_v6lcl_addr
);
16746 IN6_IPADDR_TO_V4MAPPED(INADDR_ANY
, &ipif
->ipif_v6subnet
);
16747 IN6_IPADDR_TO_V4MAPPED(INADDR_ANY
, &ipif
->ipif_v6net_mask
);
16748 IN6_IPADDR_TO_V4MAPPED(INADDR_ANY
, &ipif
->ipif_v6brd_addr
);
16749 IN6_IPADDR_TO_V4MAPPED(INADDR_ANY
, &ipif
->ipif_v6pp_dst_addr
);
16751 * Set the ILLF_ROUTER flag according to the global
16752 * IPv4 forwarding policy.
16754 if (ipst
->ips_ip_forwarding
!= 0)
16755 ill
->ill_flags
|= ILLF_ROUTER
;
16758 ASSERT(ill
->ill_phyint
!= NULL
);
16761 * The ipIfStatsIfindex and ipv6IfIcmpIfIndex assignments will
16762 * be completed in ill_glist_insert -> ill_phyint_reinit
16764 if (!ill_allocate_mibs(ill
))
16768 * Pick a default sap until we get the DL_INFO_ACK back from
16771 ill
->ill_sap
= (ill
->ill_isv6
) ? ill
->ill_media
->ip_m_ipv6sap
:
16772 ill
->ill_media
->ip_m_ipv4sap
;
16774 ill
->ill_ifname_pending
= 1;
16775 ill
->ill_ifname_pending_err
= 0;
16778 * When the first ipif comes up in ipif_up_done(), multicast groups
16779 * that were joined while this ill was not bound to the DLPI link need
16780 * to be recovered by ill_recover_multicast().
16782 ill
->ill_need_recover_multicast
= 1;
16785 rw_enter(&ipst
->ips_ill_g_lock
, RW_WRITER
);
16786 if ((error
= ill_glist_insert(ill
, interf_name
,
16787 (ill
->ill_flags
& ILLF_IPV6
) == ILLF_IPV6
)) > 0) {
16788 ill
->ill_ppa
= UINT_MAX
;
16789 ill
->ill_name
[0] = '\0';
16791 * undo null termination done above.
16793 ppa_ptr
[0] = old_char
;
16794 rw_exit(&ipst
->ips_ill_g_lock
);
16799 ASSERT(ill
->ill_name_length
<= LIFNAMSIZ
);
16802 * When we return the buffer pointed to by interf_name should contain
16803 * the same name as in ill_name.
16804 * If a ppa was choosen by the system (ppa passed in was UINT_MAX)
16805 * the buffer pointed to by new_ppa_ptr would not contain the right ppa
16806 * so copy full name and update the ppa ptr.
16807 * When ppa passed in != UINT_MAX all values are correct just undo
16808 * null termination, this saves a bcopy.
16810 if (*new_ppa_ptr
== UINT_MAX
) {
16811 bcopy(ill
->ill_name
, interf_name
, ill
->ill_name_length
);
16812 *new_ppa_ptr
= ill
->ill_ppa
;
16815 * undo null termination done above.
16817 ppa_ptr
[0] = old_char
;
16820 /* Let SCTP know about this ILL */
16821 sctp_update_ill(ill
, SCTP_ILL_INSERT
);
16824 * ill_glist_insert has made the ill visible globally, and
16825 * ill_phyint_reinit could have changed the ipsq. At this point,
16826 * we need to hold the ips_ill_g_lock across the call to enter the
16827 * ipsq to enforce atomicity and prevent reordering. In the event
16828 * the ipsq has changed, and if the new ipsq is currently busy,
16829 * we need to make sure that this half-completed ioctl is ahead of
16830 * any subsequent ioctl. We achieve this by not dropping the
16831 * ips_ill_g_lock which prevents any ill lookup itself thereby
16832 * ensuring that new ioctls can't start.
16834 ipsq
= ipsq_try_enter_internal(ill
, q
, mp
, ip_reprocess_ioctl
, NEW_OP
,
16837 rw_exit(&ipst
->ips_ill_g_lock
);
16840 return (EINPROGRESS
);
16843 * If ill_phyint_reinit() changed our ipsq, then start on the new ipsq.
16845 if (ipsq
->ipsq_xop
->ipx_current_ipif
== NULL
)
16846 ipsq_current_start(ipsq
, ipif
, SIOCSLIFNAME
);
16848 ASSERT(ipsq
->ipsq_xop
->ipx_current_ipif
== ipif
);
16850 error
= ipif_set_values_tail(ill
, ipif
, mp
, q
);
16852 if (error
!= 0 && error
!= EINPROGRESS
) {
16854 * restore previous values
16856 ill
->ill_isv6
= B_FALSE
;
16857 ill_set_inputfn(ill
);
16863 ipif_init(ip_stack_t
*ipst
)
16867 for (i
= 0; i
< MAX_G_HEADS
; i
++) {
16868 ipst
->ips_ill_g_heads
[i
].ill_g_list_head
=
16869 (ill_if_t
*)&ipst
->ips_ill_g_heads
[i
];
16870 ipst
->ips_ill_g_heads
[i
].ill_g_list_tail
=
16871 (ill_if_t
*)&ipst
->ips_ill_g_heads
[i
];
16874 avl_create(&ipst
->ips_phyint_g_list
->phyint_list_avl_by_index
,
16875 ill_phyint_compare_index
,
16877 offsetof(struct phyint
, phyint_avl_by_index
));
16878 avl_create(&ipst
->ips_phyint_g_list
->phyint_list_avl_by_name
,
16879 ill_phyint_compare_name
,
16881 offsetof(struct phyint
, phyint_avl_by_name
));
16885 * Save enough information so that we can recreate the IRE if
16886 * the interface goes down and then up.
16889 ill_save_ire(ill_t
*ill
, ire_t
*ire
)
16893 save_mp
= allocb(sizeof (ifrt_t
), BPRI_MED
);
16894 if (save_mp
!= NULL
) {
16897 save_mp
->b_wptr
+= sizeof (ifrt_t
);
16898 ifrt
= (ifrt_t
*)save_mp
->b_rptr
;
16899 bzero(ifrt
, sizeof (ifrt_t
));
16900 ifrt
->ifrt_type
= ire
->ire_type
;
16901 if (ire
->ire_ipversion
== IPV4_VERSION
) {
16902 ASSERT(!ill
->ill_isv6
);
16903 ifrt
->ifrt_addr
= ire
->ire_addr
;
16904 ifrt
->ifrt_gateway_addr
= ire
->ire_gateway_addr
;
16905 ifrt
->ifrt_setsrc_addr
= ire
->ire_setsrc_addr
;
16906 ifrt
->ifrt_mask
= ire
->ire_mask
;
16908 ASSERT(ill
->ill_isv6
);
16909 ifrt
->ifrt_v6addr
= ire
->ire_addr_v6
;
16910 /* ire_gateway_addr_v6 can change due to RTM_CHANGE */
16911 mutex_enter(&ire
->ire_lock
);
16912 ifrt
->ifrt_v6gateway_addr
= ire
->ire_gateway_addr_v6
;
16913 mutex_exit(&ire
->ire_lock
);
16914 ifrt
->ifrt_v6setsrc_addr
= ire
->ire_setsrc_addr_v6
;
16915 ifrt
->ifrt_v6mask
= ire
->ire_mask_v6
;
16917 ifrt
->ifrt_flags
= ire
->ire_flags
;
16918 ifrt
->ifrt_zoneid
= ire
->ire_zoneid
;
16919 mutex_enter(&ill
->ill_saved_ire_lock
);
16920 save_mp
->b_cont
= ill
->ill_saved_ire_mp
;
16921 ill
->ill_saved_ire_mp
= save_mp
;
16922 ill
->ill_saved_ire_cnt
++;
16923 mutex_exit(&ill
->ill_saved_ire_lock
);
16928 * Remove one entry from ill_saved_ire_mp.
16931 ill_remove_saved_ire(ill_t
*ill
, ire_t
*ire
)
16937 /* Remove from ill_saved_ire_mp list if it is there */
16938 mutex_enter(&ill
->ill_saved_ire_lock
);
16939 for (mpp
= &ill
->ill_saved_ire_mp
; *mpp
!= NULL
;
16940 mpp
= &(*mpp
)->b_cont
) {
16941 in6_addr_t gw_addr_v6
;
16944 * On a given ill, the tuple of address, gateway, mask,
16945 * ire_type, and zoneid is unique for each saved IRE.
16948 ifrt
= (ifrt_t
*)mp
->b_rptr
;
16949 /* ire_gateway_addr_v6 can change - need lock */
16950 mutex_enter(&ire
->ire_lock
);
16951 gw_addr_v6
= ire
->ire_gateway_addr_v6
;
16952 mutex_exit(&ire
->ire_lock
);
16954 if (ifrt
->ifrt_zoneid
!= ire
->ire_zoneid
||
16955 ifrt
->ifrt_type
!= ire
->ire_type
)
16958 if (ill
->ill_isv6
?
16959 (IN6_ARE_ADDR_EQUAL(&ifrt
->ifrt_v6addr
,
16960 &ire
->ire_addr_v6
) &&
16961 IN6_ARE_ADDR_EQUAL(&ifrt
->ifrt_v6gateway_addr
,
16963 IN6_ARE_ADDR_EQUAL(&ifrt
->ifrt_v6mask
,
16964 &ire
->ire_mask_v6
)) :
16965 (ifrt
->ifrt_addr
== ire
->ire_addr
&&
16966 ifrt
->ifrt_gateway_addr
== ire
->ire_gateway_addr
&&
16967 ifrt
->ifrt_mask
== ire
->ire_mask
)) {
16969 ill
->ill_saved_ire_cnt
--;
16974 mutex_exit(&ill
->ill_saved_ire_lock
);
16978 * Derive an interface id from the link layer address.
16979 * Knows about IEEE 802 and IEEE EUI-64 mappings.
16982 ip_ether_v6intfid(ill_t
*ill
, in6_addr_t
*v6addr
)
16987 * Note that some IPv6 interfaces get plumbed over links that claim to
16988 * be DL_ETHER, but don't actually have Ethernet MAC addresses (e.g.
16989 * PPP links). The ETHERADDRL check here ensures that we only set the
16990 * interface ID on IPv6 interfaces above links that actually have real
16991 * Ethernet addresses.
16993 if (ill
->ill_phys_addr_length
== ETHERADDRL
) {
16994 /* Form EUI-64 like address */
16995 addr
= (char *)&v6addr
->s6_addr32
[2];
16996 bcopy(ill
->ill_phys_addr
, addr
, 3);
16997 addr
[0] ^= 0x2; /* Toggle Universal/Local bit */
16998 addr
[3] = (char)0xff;
16999 addr
[4] = (char)0xfe;
17000 bcopy(ill
->ill_phys_addr
+ 3, addr
+ 5, 3);
17006 ip_nodef_v6intfid(ill_t
*ill
, in6_addr_t
*v6addr
)
17010 typedef struct ipmp_ifcookie
{
17011 uint32_t ic_hostid
;
17012 char ic_ifname
[LIFNAMSIZ
];
17013 char ic_zonename
[ZONENAME_MAX
];
17017 * Construct a pseudo-random interface ID for the IPMP interface that's both
17018 * predictable and (almost) guaranteed to be unique.
17021 ip_ipmp_v6intfid(ill_t
*ill
, in6_addr_t
*v6addr
)
17028 ipmp_ifcookie_t ic
= { 0 };
17030 ASSERT(IS_IPMP(ill
));
17032 (void) ddi_strtoul(hw_serial
, NULL
, 10, &hostid
);
17033 ic
.ic_hostid
= htonl((uint32_t)hostid
);
17035 (void) strlcpy(ic
.ic_ifname
, ill
->ill_name
, LIFNAMSIZ
);
17037 if ((zp
= zone_find_by_id(ill
->ill_zoneid
)) != NULL
) {
17038 (void) strlcpy(ic
.ic_zonename
, zp
->zone_name
, ZONENAME_MAX
);
17043 MD5Update(&ctx
, &ic
, sizeof (ic
));
17044 MD5Final(hash
, &ctx
);
17047 * Map the hash to an interface ID per the basic approach in RFC3041.
17049 addr
= &v6addr
->s6_addr8
[8];
17050 bcopy(hash
+ 8, addr
, sizeof (uint64_t));
17051 addr
[0] &= ~0x2; /* set local bit */
17055 * Map the multicast in6_addr_t in m_ip6addr to the physaddr for ethernet.
17058 ip_ether_v6_mapping(ill_t
*ill
, uchar_t
*m_ip6addr
, uchar_t
*m_physaddr
)
17060 phyint_t
*phyi
= ill
->ill_phyint
;
17063 * Check PHYI_MULTI_BCAST and length of physical
17064 * address to determine if we use the mapping or the
17065 * broadcast address.
17067 if ((phyi
->phyint_flags
& PHYI_MULTI_BCAST
) != 0 ||
17068 ill
->ill_phys_addr_length
!= ETHERADDRL
) {
17069 ip_mbcast_mapping(ill
, m_ip6addr
, m_physaddr
);
17072 m_physaddr
[0] = 0x33;
17073 m_physaddr
[1] = 0x33;
17074 m_physaddr
[2] = m_ip6addr
[12];
17075 m_physaddr
[3] = m_ip6addr
[13];
17076 m_physaddr
[4] = m_ip6addr
[14];
17077 m_physaddr
[5] = m_ip6addr
[15];
17081 * Map the multicast ipaddr_t in m_ipaddr to the physaddr for ethernet.
17084 ip_ether_v4_mapping(ill_t
*ill
, uchar_t
*m_ipaddr
, uchar_t
*m_physaddr
)
17086 phyint_t
*phyi
= ill
->ill_phyint
;
17089 * Check PHYI_MULTI_BCAST and length of physical
17090 * address to determine if we use the mapping or the
17091 * broadcast address.
17093 if ((phyi
->phyint_flags
& PHYI_MULTI_BCAST
) != 0 ||
17094 ill
->ill_phys_addr_length
!= ETHERADDRL
) {
17095 ip_mbcast_mapping(ill
, m_ipaddr
, m_physaddr
);
17098 m_physaddr
[0] = 0x01;
17099 m_physaddr
[1] = 0x00;
17100 m_physaddr
[2] = 0x5e;
17101 m_physaddr
[3] = m_ipaddr
[1] & 0x7f;
17102 m_physaddr
[4] = m_ipaddr
[2];
17103 m_physaddr
[5] = m_ipaddr
[3];
17108 ip_mbcast_mapping(ill_t
*ill
, uchar_t
*m_ipaddr
, uchar_t
*m_physaddr
)
17111 * for the MULTI_BCAST case and other cases when we want to
17112 * use the link-layer broadcast address for multicast.
17114 uint8_t *bphys_addr
;
17115 dl_unitdata_req_t
*dlur
;
17117 dlur
= (dl_unitdata_req_t
*)ill
->ill_bcast_mp
->b_rptr
;
17118 if (ill
->ill_sap_length
< 0) {
17119 bphys_addr
= (uchar_t
*)dlur
+
17120 dlur
->dl_dest_addr_offset
;
17122 bphys_addr
= (uchar_t
*)dlur
+
17123 dlur
->dl_dest_addr_offset
+ ill
->ill_sap_length
;
17126 bcopy(bphys_addr
, m_physaddr
, ill
->ill_phys_addr_length
);
17130 * Derive IPoIB interface id from the link layer address.
17133 ip_ib_v6intfid(ill_t
*ill
, in6_addr_t
*v6addr
)
17137 ASSERT(ill
->ill_phys_addr_length
== 20);
17138 addr
= (char *)&v6addr
->s6_addr32
[2];
17139 bcopy(ill
->ill_phys_addr
+ 12, addr
, 8);
17141 * In IBA 1.1 timeframe, some vendors erroneously set the u/l bit
17142 * in the globally assigned EUI-64 GUID to 1, in violation of IEEE
17143 * rules. In these cases, the IBA considers these GUIDs to be in
17144 * "Modified EUI-64" format, and thus toggling the u/l bit is not
17145 * required; vendors are required not to assign global EUI-64's
17146 * that differ only in u/l bit values, thus guaranteeing uniqueness
17147 * of the interface identifier. Whether the GUID is in modified
17148 * or proper EUI-64 format, the ipv6 identifier must have the u/l
17151 addr
[0] |= 2; /* Set Universal/Local bit to 1 */
17155 * Map the multicast ipaddr_t in m_ipaddr to the physaddr for InfiniBand.
17156 * Note on mapping from multicast IP addresses to IPoIB multicast link
17157 * addresses. IPoIB multicast link addresses are based on IBA link addresses.
17158 * The format of an IPoIB multicast address is:
17160 * 4 byte QPN Scope Sign. Pkey
17161 * +--------------------------------------------+
17162 * | 00FFFFFF | FF | 1X | X01B | Pkey | GroupID |
17163 * +--------------------------------------------+
17165 * The Scope and Pkey components are properties of the IBA port and
17166 * network interface. They can be ascertained from the broadcast address.
17167 * The Sign. part is the signature, and is 401B for IPv4 and 601B for IPv6.
17170 ip_ib_v4_mapping(ill_t
*ill
, uchar_t
*m_ipaddr
, uchar_t
*m_physaddr
)
17172 static uint8_t ipv4_g_phys_ibmulti_addr
[] = { 0x00, 0xff, 0xff, 0xff,
17173 0xff, 0x10, 0x40, 0x1b, 0x00, 0x00, 0x00, 0x00,
17174 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 };
17175 uint8_t *bphys_addr
;
17176 dl_unitdata_req_t
*dlur
;
17178 bcopy(ipv4_g_phys_ibmulti_addr
, m_physaddr
, ill
->ill_phys_addr_length
);
17181 * RFC 4391: IPv4 MGID is 28-bit long.
17183 m_physaddr
[16] = m_ipaddr
[0] & 0x0f;
17184 m_physaddr
[17] = m_ipaddr
[1];
17185 m_physaddr
[18] = m_ipaddr
[2];
17186 m_physaddr
[19] = m_ipaddr
[3];
17189 dlur
= (dl_unitdata_req_t
*)ill
->ill_bcast_mp
->b_rptr
;
17190 if (ill
->ill_sap_length
< 0) {
17191 bphys_addr
= (uchar_t
*)dlur
+ dlur
->dl_dest_addr_offset
;
17193 bphys_addr
= (uchar_t
*)dlur
+ dlur
->dl_dest_addr_offset
+
17194 ill
->ill_sap_length
;
17197 * Now fill in the IBA scope/Pkey values from the broadcast address.
17199 m_physaddr
[5] = bphys_addr
[5];
17200 m_physaddr
[8] = bphys_addr
[8];
17201 m_physaddr
[9] = bphys_addr
[9];
17205 ip_ib_v6_mapping(ill_t
*ill
, uchar_t
*m_ipaddr
, uchar_t
*m_physaddr
)
17207 static uint8_t ipv4_g_phys_ibmulti_addr
[] = { 0x00, 0xff, 0xff, 0xff,
17208 0xff, 0x10, 0x60, 0x1b, 0x00, 0x00, 0x00, 0x00,
17209 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 };
17210 uint8_t *bphys_addr
;
17211 dl_unitdata_req_t
*dlur
;
17213 bcopy(ipv4_g_phys_ibmulti_addr
, m_physaddr
, ill
->ill_phys_addr_length
);
17216 * RFC 4391: IPv4 MGID is 80-bit long.
17218 bcopy(&m_ipaddr
[6], &m_physaddr
[10], 10);
17220 dlur
= (dl_unitdata_req_t
*)ill
->ill_bcast_mp
->b_rptr
;
17221 if (ill
->ill_sap_length
< 0) {
17222 bphys_addr
= (uchar_t
*)dlur
+ dlur
->dl_dest_addr_offset
;
17224 bphys_addr
= (uchar_t
*)dlur
+ dlur
->dl_dest_addr_offset
+
17225 ill
->ill_sap_length
;
17228 * Now fill in the IBA scope/Pkey values from the broadcast address.
17230 m_physaddr
[5] = bphys_addr
[5];
17231 m_physaddr
[8] = bphys_addr
[8];
17232 m_physaddr
[9] = bphys_addr
[9];
17236 * Derive IPv6 interface id from an IPv4 link-layer address (e.g. from an IPv4
17237 * tunnel). The IPv4 address simply get placed in the lower 4 bytes of the
17238 * IPv6 interface id. This is a suggested mechanism described in section 3.7
17242 ip_ipv4_genv6intfid(ill_t
*ill
, uint8_t *physaddr
, in6_addr_t
*v6addr
)
17244 ASSERT(ill
->ill_phys_addr_length
== sizeof (ipaddr_t
));
17245 v6addr
->s6_addr32
[2] = 0;
17246 bcopy(physaddr
, &v6addr
->s6_addr32
[3], sizeof (ipaddr_t
));
17250 * Derive IPv6 interface id from an IPv6 link-layer address (e.g. from an IPv6
17251 * tunnel). The lower 8 bytes of the IPv6 address simply become the interface
17255 ip_ipv6_genv6intfid(ill_t
*ill
, uint8_t *physaddr
, in6_addr_t
*v6addr
)
17257 in6_addr_t
*v6lladdr
= (in6_addr_t
*)physaddr
;
17259 ASSERT(ill
->ill_phys_addr_length
== sizeof (in6_addr_t
));
17260 bcopy(&v6lladdr
->s6_addr32
[2], &v6addr
->s6_addr32
[2], 8);
17264 ip_ipv6_v6intfid(ill_t
*ill
, in6_addr_t
*v6addr
)
17266 ip_ipv6_genv6intfid(ill
, ill
->ill_phys_addr
, v6addr
);
17270 ip_ipv6_v6destintfid(ill_t
*ill
, in6_addr_t
*v6addr
)
17272 ip_ipv6_genv6intfid(ill
, ill
->ill_dest_addr
, v6addr
);
17276 ip_ipv4_v6intfid(ill_t
*ill
, in6_addr_t
*v6addr
)
17278 ip_ipv4_genv6intfid(ill
, ill
->ill_phys_addr
, v6addr
);
17282 ip_ipv4_v6destintfid(ill_t
*ill
, in6_addr_t
*v6addr
)
17284 ip_ipv4_genv6intfid(ill
, ill
->ill_dest_addr
, v6addr
);
17288 * Lookup an ill and verify that the zoneid has an ipif on that ill.
17289 * Returns an held ill, or NULL.
17292 ill_lookup_on_ifindex_zoneid(uint_t index
, zoneid_t zoneid
, boolean_t isv6
,
17298 ill
= ill_lookup_on_ifindex(index
, isv6
, ipst
);
17302 mutex_enter(&ill
->ill_lock
);
17303 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
; ipif
= ipif
->ipif_next
) {
17304 if (IPIF_IS_CONDEMNED(ipif
))
17306 if (zoneid
!= ALL_ZONES
&& ipif
->ipif_zoneid
!= zoneid
&&
17307 ipif
->ipif_zoneid
!= ALL_ZONES
)
17310 mutex_exit(&ill
->ill_lock
);
17313 mutex_exit(&ill
->ill_lock
);
17319 * Return a pointer to an ipif_t given a combination of (ill_idx,ipif_id)
17320 * If a pointer to an ipif_t is returned then the caller will need to do
17321 * an ill_refrele().
17324 ipif_getby_indexes(uint_t ifindex
, uint_t lifidx
, boolean_t isv6
,
17330 ill
= ill_lookup_on_ifindex(ifindex
, isv6
, ipst
);
17334 mutex_enter(&ill
->ill_lock
);
17335 if (ill
->ill_state_flags
& ILL_CONDEMNED
) {
17336 mutex_exit(&ill
->ill_lock
);
17341 for (ipif
= ill
->ill_ipif
; ipif
!= NULL
; ipif
= ipif
->ipif_next
) {
17342 if (!IPIF_CAN_LOOKUP(ipif
))
17344 if (lifidx
== ipif
->ipif_id
) {
17345 ipif_refhold_locked(ipif
);
17350 mutex_exit(&ill
->ill_lock
);
17356 * Set ill_inputfn based on the current know state.
17357 * This needs to be called when any of the factors taken into
17361 ill_set_inputfn(ill_t
*ill
)
17363 ip_stack_t
*ipst
= ill
->ill_ipst
;
17365 if (ill
->ill_isv6
) {
17366 ill
->ill_inputfn
= ill_input_short_v6
;
17368 if (ill
->ill_dhcpinit
!= 0)
17369 ill
->ill_inputfn
= ill_input_full_v4
;
17370 else if (ipst
->ips_ipcl_proto_fanout_v4
[IPPROTO_RSVP
].connf_head
17372 ill
->ill_inputfn
= ill_input_full_v4
;
17374 ill
->ill_inputfn
= ill_input_short_v4
;
17379 * Re-evaluate ill_inputfn for all the IPv4 ills.
17380 * Used when RSVP comes and goes.
17383 ill_set_inputfn_all(ip_stack_t
*ipst
)
17385 ill_walk_context_t ctx
;
17388 rw_enter(&ipst
->ips_ill_g_lock
, RW_READER
);
17389 ill
= ILL_START_WALK_V4(&ctx
, ipst
);
17390 for (; ill
!= NULL
; ill
= ill_next(&ctx
, ill
))
17391 ill_set_inputfn(ill
);
17393 rw_exit(&ipst
->ips_ill_g_lock
);
17397 * Set the physical address information for `ill' to the contents of the
17398 * dl_notify_ind_t pointed to by `mp'. Must be called as writer, and will be
17399 * asynchronous if `ill' cannot immediately be quiesced -- in which case
17400 * EINPROGRESS will be returned.
17403 ill_set_phys_addr(ill_t
*ill
, mblk_t
*mp
)
17405 ipsq_t
*ipsq
= ill
->ill_phyint
->phyint_ipsq
;
17406 dl_notify_ind_t
*dlindp
= (dl_notify_ind_t
*)mp
->b_rptr
;
17408 ASSERT(IAM_WRITER_IPSQ(ipsq
));
17410 if (dlindp
->dl_data
!= DL_IPV6_LINK_LAYER_ADDR
&&
17411 dlindp
->dl_data
!= DL_CURR_DEST_ADDR
&&
17412 dlindp
->dl_data
!= DL_CURR_PHYS_ADDR
) {
17413 /* Changing DL_IPV6_TOKEN is not yet supported */
17418 * We need to store up to two copies of `mp' in `ill'. Due to the
17419 * design of ipsq_pending_mp_add(), we can't pass them as separate
17420 * arguments to ill_set_phys_addr_tail(). Instead, chain them
17421 * together here, then pull 'em apart in ill_set_phys_addr_tail().
17423 if ((mp
= copyb(mp
)) == NULL
|| (mp
->b_cont
= copyb(mp
)) == NULL
) {
17428 ipsq_current_start(ipsq
, ill
->ill_ipif
, 0);
17431 * Since we'll only do a logical down, we can't rely on ipif_down
17432 * to turn on ILL_DOWN_IN_PROGRESS, or for the DL_BIND_ACK to reset
17433 * ILL_DOWN_IN_PROGRESS. We instead manage this separately for this
17434 * case, to quiesce ire's and nce's for ill_is_quiescent.
17436 mutex_enter(&ill
->ill_lock
);
17437 ill
->ill_state_flags
|= ILL_DOWN_IN_PROGRESS
;
17438 /* no more ire/nce addition allowed */
17439 mutex_exit(&ill
->ill_lock
);
17442 * If we can quiesce the ill, then set the address. If not, then
17443 * ill_set_phys_addr_tail() will be called from ipif_ill_refrele_tail().
17445 ill_down_ipifs(ill
, B_TRUE
);
17446 mutex_enter(&ill
->ill_lock
);
17447 if (!ill_is_quiescent(ill
)) {
17448 /* call cannot fail since `conn_t *' argument is NULL */
17449 (void) ipsq_pending_mp_add(NULL
, ill
->ill_ipif
, ill
->ill_rq
,
17451 mutex_exit(&ill
->ill_lock
);
17452 return (EINPROGRESS
);
17454 mutex_exit(&ill
->ill_lock
);
17456 ill_set_phys_addr_tail(ipsq
, ill
->ill_rq
, mp
, NULL
);
17461 * When the allowed-ips link property is set on the datalink, IP receives a
17462 * DL_NOTE_ALLOWED_IPS notification that is processed in ill_set_allowed_ips()
17463 * to initialize the ill_allowed_ips[] array in the ill_t. This array is then
17464 * used to vet addresses passed to ip_sioctl_addr() and to ensure that the
17465 * only IP addresses configured on the ill_t are those in the ill_allowed_ips[]
17469 ill_set_allowed_ips(ill_t
*ill
, mblk_t
*mp
)
17471 ipsq_t
*ipsq
= ill
->ill_phyint
->phyint_ipsq
;
17472 dl_notify_ind_t
*dlip
= (dl_notify_ind_t
*)mp
->b_rptr
;
17473 mac_protect_t
*mrp
;
17476 ASSERT(IAM_WRITER_IPSQ(ipsq
));
17477 mrp
= (mac_protect_t
*)&dlip
[1];
17479 if (mrp
->mp_ipaddrcnt
== 0) { /* reset allowed-ips */
17480 kmem_free(ill
->ill_allowed_ips
,
17481 ill
->ill_allowed_ips_cnt
* sizeof (in6_addr_t
));
17482 ill
->ill_allowed_ips_cnt
= 0;
17483 ill
->ill_allowed_ips
= NULL
;
17484 mutex_enter(&ill
->ill_phyint
->phyint_lock
);
17485 ill
->ill_phyint
->phyint_flags
&= ~PHYI_L3PROTECT
;
17486 mutex_exit(&ill
->ill_phyint
->phyint_lock
);
17490 if (ill
->ill_allowed_ips
!= NULL
) {
17491 kmem_free(ill
->ill_allowed_ips
,
17492 ill
->ill_allowed_ips_cnt
* sizeof (in6_addr_t
));
17494 ill
->ill_allowed_ips_cnt
= mrp
->mp_ipaddrcnt
;
17495 ill
->ill_allowed_ips
= kmem_alloc(
17496 ill
->ill_allowed_ips_cnt
* sizeof (in6_addr_t
), KM_SLEEP
);
17497 for (i
= 0; i
< mrp
->mp_ipaddrcnt
; i
++)
17498 ill
->ill_allowed_ips
[i
] = mrp
->mp_ipaddrs
[i
].ip_addr
;
17500 mutex_enter(&ill
->ill_phyint
->phyint_lock
);
17501 ill
->ill_phyint
->phyint_flags
|= PHYI_L3PROTECT
;
17502 mutex_exit(&ill
->ill_phyint
->phyint_lock
);
17506 * Once the ill associated with `q' has quiesced, set its physical address
17507 * information to the values in `addrmp'. Note that two copies of `addrmp'
17508 * are passed (linked by b_cont), since we sometimes need to save two distinct
17509 * copies in the ill_t, and our context doesn't permit sleeping or allocation
17510 * failure (we'll free the other copy if it's not needed). Since the ill_t
17511 * is quiesced, we know any stale nce's with the old address information have
17512 * already been removed, so we don't need to call nce_flush().
17516 ill_set_phys_addr_tail(ipsq_t
*ipsq
, queue_t
*q
, mblk_t
*addrmp
, void *dummy
)
17518 ill_t
*ill
= q
->q_ptr
;
17519 mblk_t
*addrmp2
= unlinkb(addrmp
);
17520 dl_notify_ind_t
*dlindp
= (dl_notify_ind_t
*)addrmp
->b_rptr
;
17521 uint_t addrlen
, addroff
;
17524 ASSERT(IAM_WRITER_IPSQ(ipsq
));
17526 addroff
= dlindp
->dl_addr_offset
;
17527 addrlen
= dlindp
->dl_addr_length
- ABS(ill
->ill_sap_length
);
17529 switch (dlindp
->dl_data
) {
17530 case DL_IPV6_LINK_LAYER_ADDR
:
17531 ill_set_ndmp(ill
, addrmp
, addroff
, addrlen
);
17535 case DL_CURR_DEST_ADDR
:
17536 freemsg(ill
->ill_dest_addr_mp
);
17537 ill
->ill_dest_addr
= addrmp
->b_rptr
+ addroff
;
17538 ill
->ill_dest_addr_mp
= addrmp
;
17539 if (ill
->ill_isv6
) {
17540 ill_setdesttoken(ill
);
17541 ipif_setdestlinklocal(ill
->ill_ipif
);
17546 case DL_CURR_PHYS_ADDR
:
17547 freemsg(ill
->ill_phys_addr_mp
);
17548 ill
->ill_phys_addr
= addrmp
->b_rptr
+ addroff
;
17549 ill
->ill_phys_addr_mp
= addrmp
;
17550 ill
->ill_phys_addr_length
= addrlen
;
17552 ill_set_ndmp(ill
, addrmp2
, addroff
, addrlen
);
17555 if (ill
->ill_isv6
) {
17556 ill_setdefaulttoken(ill
);
17557 ipif_setlinklocal(ill
->ill_ipif
);
17565 * reset ILL_DOWN_IN_PROGRESS so that we can successfully add ires
17566 * as we bring the ipifs up again.
17568 mutex_enter(&ill
->ill_lock
);
17569 ill
->ill_state_flags
&= ~ILL_DOWN_IN_PROGRESS
;
17570 mutex_exit(&ill
->ill_lock
);
17572 * If there are ipifs to bring up, ill_up_ipifs() will return
17573 * EINPROGRESS, and ipsq_current_finish() will be called by
17574 * ip_rput_dlpi_writer() or arp_bringup_done() when the last ipif is
17577 status
= ill_up_ipifs(ill
, q
, addrmp
);
17578 if (status
!= EINPROGRESS
)
17579 ipsq_current_finish(ipsq
);
17583 * Helper routine for setting the ill_nd_lla fields.
17586 ill_set_ndmp(ill_t
*ill
, mblk_t
*ndmp
, uint_t addroff
, uint_t addrlen
)
17588 freemsg(ill
->ill_nd_lla_mp
);
17589 ill
->ill_nd_lla
= ndmp
->b_rptr
+ addroff
;
17590 ill
->ill_nd_lla_mp
= ndmp
;
17591 ill
->ill_nd_lla_len
= addrlen
;
17598 ill_replumb(ill_t
*ill
, mblk_t
*mp
)
17600 ipsq_t
*ipsq
= ill
->ill_phyint
->phyint_ipsq
;
17602 ASSERT(IAM_WRITER_IPSQ(ipsq
));
17604 ipsq_current_start(ipsq
, ill
->ill_ipif
, 0);
17607 * If we can quiesce the ill, then continue. If not, then
17608 * ill_replumb_tail() will be called from ipif_ill_refrele_tail().
17610 ill_down_ipifs(ill
, B_FALSE
);
17612 mutex_enter(&ill
->ill_lock
);
17613 if (!ill_is_quiescent(ill
)) {
17614 /* call cannot fail since `conn_t *' argument is NULL */
17615 (void) ipsq_pending_mp_add(NULL
, ill
->ill_ipif
, ill
->ill_rq
,
17617 mutex_exit(&ill
->ill_lock
);
17618 return (EINPROGRESS
);
17620 mutex_exit(&ill
->ill_lock
);
17622 ill_replumb_tail(ipsq
, ill
->ill_rq
, mp
, NULL
);
17628 ill_replumb_tail(ipsq_t
*ipsq
, queue_t
*q
, mblk_t
*mp
, void *dummy
)
17630 ill_t
*ill
= q
->q_ptr
;
17632 conn_t
*connp
= NULL
;
17634 ASSERT(IAM_WRITER_IPSQ(ipsq
));
17635 freemsg(ill
->ill_replumb_mp
);
17636 ill
->ill_replumb_mp
= copyb(mp
);
17638 if (ill
->ill_replumb_mp
== NULL
) {
17639 /* out of memory */
17640 ipsq_current_finish(ipsq
);
17644 mutex_enter(&ill
->ill_lock
);
17645 ill
->ill_up_ipifs
= ipsq_pending_mp_add(NULL
, ill
->ill_ipif
,
17646 ill
->ill_rq
, ill
->ill_replumb_mp
, 0);
17647 mutex_exit(&ill
->ill_lock
);
17649 if (!ill
->ill_up_ipifs
) {
17650 /* already closing */
17651 ipsq_current_finish(ipsq
);
17654 ill
->ill_replumbing
= 1;
17655 err
= ill_down_ipifs_tail(ill
);
17658 * Successfully quiesced and brought down the interface, now we send
17659 * the DL_NOTE_REPLUMB_DONE message down to the driver. Reuse the
17660 * DL_NOTE_REPLUMB message.
17662 mp
= mexchange(NULL
, mp
, sizeof (dl_notify_conf_t
), M_PROTO
,
17664 ASSERT(mp
!= NULL
);
17665 ((dl_notify_conf_t
*)mp
->b_rptr
)->dl_notification
=
17666 DL_NOTE_REPLUMB_DONE
;
17667 ill_dlpi_send(ill
, mp
);
17670 * For IPv4, we would usually get EINPROGRESS because the ETHERTYPE_ARP
17671 * streams have to be unbound. When all the DLPI exchanges are done,
17672 * ipsq_current_finish() will be called by arp_bringup_done(). The
17673 * remainder of ipif bringup via ill_up_ipifs() will also be done in
17674 * arp_bringup_done().
17676 ASSERT(ill
->ill_replumb_mp
!= NULL
);
17677 if (err
== EINPROGRESS
)
17680 ill
->ill_replumb_mp
= ipsq_pending_mp_get(ipsq
, &connp
);
17681 ASSERT(connp
== NULL
);
17682 if (err
== 0 && ill
->ill_replumb_mp
!= NULL
&&
17683 ill_up_ipifs(ill
, q
, ill
->ill_replumb_mp
) == EINPROGRESS
) {
17686 ipsq_current_finish(ipsq
);
17690 * Issue ioctl `cmd' on `lh'; caller provides the initial payload in `buf'
17691 * which is `bufsize' bytes. On success, zero is returned and `buf' updated
17692 * as per the ioctl. On failure, an errno is returned.
17695 ip_ioctl(ldi_handle_t lh
, int cmd
, void *buf
, uint_t bufsize
, cred_t
*cr
)
17698 struct strioctl iocb
;
17701 iocb
.ic_timout
= 15;
17702 iocb
.ic_len
= bufsize
;
17705 return (ldi_ioctl(lh
, I_STR
, (intptr_t)&iocb
, FKIOCTL
, cr
, &rval
));
17709 * Issue an SIOCGLIFCONF for address family `af' and store the result into a
17710 * dynamically-allocated `lifcp' that will be `bufsizep' bytes on success.
17713 ip_lifconf_ioctl(ldi_handle_t lh
, int af
, struct lifconf
*lifcp
,
17714 uint_t
*bufsizep
, cred_t
*cr
)
17717 struct lifnum lifn
;
17719 bzero(&lifn
, sizeof (lifn
));
17720 lifn
.lifn_family
= af
;
17721 lifn
.lifn_flags
= LIFC_UNDER_IPMP
;
17723 if ((err
= ip_ioctl(lh
, SIOCGLIFNUM
, &lifn
, sizeof (lifn
), cr
)) != 0)
17727 * Pad the interface count to account for additional interfaces that
17728 * may have been configured between the SIOCGLIFNUM and SIOCGLIFCONF.
17730 lifn
.lifn_count
+= 4;
17731 bzero(lifcp
, sizeof (*lifcp
));
17732 lifcp
->lifc_flags
= LIFC_UNDER_IPMP
;
17733 lifcp
->lifc_family
= af
;
17734 lifcp
->lifc_len
= *bufsizep
= lifn
.lifn_count
* sizeof (struct lifreq
);
17735 lifcp
->lifc_buf
= kmem_zalloc(*bufsizep
, KM_SLEEP
);
17737 err
= ip_ioctl(lh
, SIOCGLIFCONF
, lifcp
, sizeof (*lifcp
), cr
);
17739 kmem_free(lifcp
->lifc_buf
, *bufsizep
);
17747 * Helper for ip_interface_cleanup() that removes the loopback interface.
17750 ip_loopback_removeif(ldi_handle_t lh
, boolean_t isv6
, cred_t
*cr
)
17753 struct lifreq lifr
;
17755 bzero(&lifr
, sizeof (lifr
));
17756 (void) strcpy(lifr
.lifr_name
, ipif_loopback_name
);
17759 * Attempt to remove the interface. It may legitimately not exist
17760 * (e.g. the zone administrator unplumbed it), so ignore ENXIO.
17762 err
= ip_ioctl(lh
, SIOCLIFREMOVEIF
, &lifr
, sizeof (lifr
), cr
);
17763 if (err
!= 0 && err
!= ENXIO
) {
17764 ip0dbg(("ip_loopback_removeif: IP%s SIOCLIFREMOVEIF failed: "
17765 "error %d\n", isv6
? "v6" : "v4", err
));
17770 * Helper for ip_interface_cleanup() that ensures no IP interfaces are in IPMP
17771 * groups and that IPMP data addresses are down. These conditions must be met
17772 * so that IPMP interfaces can be I_PUNLINK'd, as per ip_sioctl_plink_ipmp().
17775 ip_ipmp_cleanup(ldi_handle_t lh
, boolean_t isv6
, cred_t
*cr
)
17777 int af
= isv6
? AF_INET6
: AF_INET
;
17781 uint_t lifrsize
= sizeof (struct lifreq
);
17782 struct lifconf lifc
;
17783 struct lifreq
*lifrp
;
17785 if ((err
= ip_lifconf_ioctl(lh
, af
, &lifc
, &bufsize
, cr
)) != 0) {
17786 cmn_err(CE_WARN
, "ip_ipmp_cleanup: cannot get interface list "
17787 "(error %d); any IPMP interfaces cannot be shutdown", err
);
17791 nifs
= lifc
.lifc_len
/ lifrsize
;
17792 for (lifrp
= lifc
.lifc_req
, i
= 0; i
< nifs
; i
++, lifrp
++) {
17793 err
= ip_ioctl(lh
, SIOCGLIFFLAGS
, lifrp
, lifrsize
, cr
);
17795 cmn_err(CE_WARN
, "ip_ipmp_cleanup: %s: cannot get "
17796 "flags: error %d", lifrp
->lifr_name
, err
);
17800 if (lifrp
->lifr_flags
& IFF_IPMP
) {
17801 if ((lifrp
->lifr_flags
& (IFF_UP
|IFF_DUPLICATE
)) == 0)
17804 lifrp
->lifr_flags
&= ~IFF_UP
;
17805 err
= ip_ioctl(lh
, SIOCSLIFFLAGS
, lifrp
, lifrsize
, cr
);
17807 cmn_err(CE_WARN
, "ip_ipmp_cleanup: %s: cannot "
17808 "bring down (error %d); IPMP interface may "
17809 "not be shutdown", lifrp
->lifr_name
, err
);
17813 * Check if IFF_DUPLICATE is still set -- and if so,
17814 * reset the address to clear it.
17816 err
= ip_ioctl(lh
, SIOCGLIFFLAGS
, lifrp
, lifrsize
, cr
);
17817 if (err
!= 0 || !(lifrp
->lifr_flags
& IFF_DUPLICATE
))
17820 err
= ip_ioctl(lh
, SIOCGLIFADDR
, lifrp
, lifrsize
, cr
);
17821 if (err
!= 0 || (err
= ip_ioctl(lh
, SIOCGLIFADDR
,
17822 lifrp
, lifrsize
, cr
)) != 0) {
17823 cmn_err(CE_WARN
, "ip_ipmp_cleanup: %s: cannot "
17824 "reset DAD (error %d); IPMP interface may "
17825 "not be shutdown", lifrp
->lifr_name
, err
);
17830 if (strchr(lifrp
->lifr_name
, IPIF_SEPARATOR_CHAR
) == 0) {
17831 lifrp
->lifr_groupname
[0] = '\0';
17832 if ((err
= ip_ioctl(lh
, SIOCSLIFGROUPNAME
, lifrp
,
17833 lifrsize
, cr
)) != 0) {
17834 cmn_err(CE_WARN
, "ip_ipmp_cleanup: %s: cannot "
17835 "leave IPMP group (error %d); associated "
17836 "IPMP interface may not be shutdown",
17837 lifrp
->lifr_name
, err
);
17843 kmem_free(lifc
.lifc_buf
, bufsize
);
17846 #define UDPDEV "/devices/pseudo/udp@0:udp"
17847 #define UDP6DEV "/devices/pseudo/udp6@0:udp6"
17850 * Remove the loopback interfaces and prep the IPMP interfaces to be torn down.
17851 * Non-loopback interfaces are either I_LINK'd or I_PLINK'd; the former go away
17852 * when the user-level processes in the zone are killed and the latter are
17853 * cleaned up by str_stack_shutdown().
17856 ip_interface_cleanup(ip_stack_t
*ipst
)
17863 char *devs
[] = { UDP6DEV
, UDPDEV
};
17864 netstackid_t stackid
= ipst
->ips_netstack
->netstack_stackid
;
17866 if ((err
= ldi_ident_from_major(ddi_name_to_major("ip"), &li
)) != 0) {
17867 cmn_err(CE_WARN
, "ip_interface_cleanup: cannot get ldi ident:"
17872 cr
= zone_get_kcred(netstackid_to_zoneid(stackid
));
17873 ASSERT(cr
!= NULL
);
17876 * NOTE: loop executes exactly twice and is hardcoded to know that the
17877 * first iteration is IPv6. (Unrolling yields repetitious code, hence
17880 for (i
= 0; i
< 2; i
++) {
17881 err
= ldi_open_by_name(devs
[i
], FREAD
|FWRITE
, cr
, &lh
, li
);
17883 cmn_err(CE_WARN
, "ip_interface_cleanup: cannot open %s:"
17884 " error %d", devs
[i
], err
);
17888 ip_loopback_removeif(lh
, i
== 0, cr
);
17889 ip_ipmp_cleanup(lh
, i
== 0, cr
);
17891 (void) ldi_close(lh
, FREAD
|FWRITE
, cr
);
17894 ldi_ident_release(li
);
17899 * This needs to be in-sync with nic_event_t definition
17901 static const char *
17902 ill_hook_event2str(nic_event_t event
)
17908 return ("UNPLUMB");
17913 case NE_ADDRESS_CHANGE
:
17914 return ("ADDRESS_CHANGE");
17918 return ("LIF_DOWN");
17919 case NE_IFINDEX_CHANGE
:
17920 return ("IFINDEX_CHANGE");
17922 return ("UNKNOWN");
17927 ill_nic_event_dispatch(ill_t
*ill
, lif_if_t lif
, nic_event_t event
,
17928 nic_event_data_t data
, size_t datalen
)
17930 ip_stack_t
*ipst
= ill
->ill_ipst
;
17931 hook_nic_event_int_t
*info
;
17932 const char *str
= NULL
;
17934 /* create a new nic event info */
17935 if ((info
= kmem_alloc(sizeof (*info
), KM_NOSLEEP
)) == NULL
)
17938 info
->hnei_event
.hne_nic
= ill
->ill_phyint
->phyint_ifindex
;
17939 info
->hnei_event
.hne_lif
= lif
;
17940 info
->hnei_event
.hne_event
= event
;
17941 info
->hnei_event
.hne_protocol
= ill
->ill_isv6
?
17942 ipst
->ips_ipv6_net_data
: ipst
->ips_ipv4_net_data
;
17943 info
->hnei_event
.hne_data
= NULL
;
17944 info
->hnei_event
.hne_datalen
= 0;
17945 info
->hnei_stackid
= ipst
->ips_netstack
->netstack_stackid
;
17947 if (data
!= NULL
&& datalen
!= 0) {
17948 info
->hnei_event
.hne_data
= kmem_alloc(datalen
, KM_NOSLEEP
);
17949 if (info
->hnei_event
.hne_data
== NULL
)
17951 bcopy(data
, info
->hnei_event
.hne_data
, datalen
);
17952 info
->hnei_event
.hne_datalen
= datalen
;
17955 if (ddi_taskq_dispatch(eventq_queue_nic
, ip_ne_queue_func
, info
,
17956 DDI_NOSLEEP
) == DDI_SUCCESS
)
17960 if (info
!= NULL
) {
17961 if (info
->hnei_event
.hne_data
!= NULL
) {
17962 kmem_free(info
->hnei_event
.hne_data
,
17963 info
->hnei_event
.hne_datalen
);
17965 kmem_free(info
, sizeof (hook_nic_event_t
));
17967 str
= ill_hook_event2str(event
);
17968 ip2dbg(("ill_nic_event_dispatch: could not dispatch %s nic event "
17969 "information for %s (ENOMEM)\n", str
, ill
->ill_name
));
17973 ipif_arp_up_done_tail(ipif_t
*ipif
, enum ip_resolver_action res_act
)
17976 const in_addr_t
*addr
= NULL
;
17978 ill_t
*ill
= ipif
->ipif_ill
;
17980 boolean_t added_ipif
= B_FALSE
;
17984 DTRACE_PROBE3(ipif__downup
, char *, "ipif_arp_up_done_tail",
17985 ill_t
*, ill
, ipif_t
*, ipif
);
17986 if (ipif
->ipif_lcl_addr
!= INADDR_ANY
) {
17987 addr
= &ipif
->ipif_lcl_addr
;
17990 if ((ipif
->ipif_flags
& IPIF_UNNUMBERED
) || addr
== NULL
) {
17991 if (res_act
!= Res_act_initial
)
17995 if (addr
!= NULL
) {
17996 ipmp_illgrp_t
*illg
= ill
->ill_grp
;
17998 /* add unicast nce for the local addr */
18000 if (IS_IPMP(ill
)) {
18002 * If we're here via ipif_up(), then the ipif
18003 * won't be bound yet -- add it to the group,
18004 * which will bind it if possible. (We would
18005 * add it in ipif_up(), but deleting on failure
18006 * there is gruesome.) If we're here via
18007 * ipmp_ill_bind_ipif(), then the ipif has
18008 * already been added to the group and we
18009 * just need to use the binding.
18011 if ((bound_ill
= ipmp_ipif_bound_ill(ipif
)) == NULL
) {
18012 bound_ill
= ipmp_illgrp_add_ipif(illg
, ipif
);
18013 if (bound_ill
== NULL
) {
18015 * We couldn't bind the ipif to an ill
18016 * yet, so we have nothing to publish.
18017 * Mark the address as ready and return.
18019 ipif
->ipif_addr_ready
= 1;
18022 added_ipif
= B_TRUE
;
18028 flags
= (NCE_F_MYADDR
| NCE_F_PUBLISH
| NCE_F_AUTHORITY
|
18031 * If this is an initial bring-up (or the ipif was never
18032 * completely brought up), do DAD. Otherwise, we're here
18033 * because IPMP has rebound an address to this ill: send
18034 * unsolicited advertisements (ARP announcements) to
18037 if (res_act
== Res_act_initial
|| !ipif
->ipif_addr_ready
) {
18038 state
= ND_UNCHANGED
; /* compute in nce_add_common() */
18040 state
= ND_REACHABLE
;
18041 flags
|= NCE_F_UNSOL_ADV
;
18045 err
= nce_lookup_then_add_v4(ill
,
18046 bound_ill
->ill_phys_addr
, bound_ill
->ill_phys_addr_length
,
18047 addr
, flags
, state
, &nce
);
18050 * note that we may encounter EEXIST if we are moving
18051 * the nce as a result of a rebind operation.
18055 ipif
->ipif_added_nce
= 1;
18056 nce
->nce_ipif_cnt
++;
18059 ip1dbg(("ipif_arp_up: NCE already exists for %s\n",
18061 if (!NCE_MYADDR(nce
->nce_common
)) {
18063 * A leftover nce from before this address
18066 ncec_delete(nce
->nce_common
);
18071 if ((ipif
->ipif_flags
& IPIF_POINTOPOINT
) == 0) {
18074 ip1dbg(("ipif_arp_up: NCE already exists "
18075 "for %s:%u\n", ill
->ill_name
,
18080 * Duplicate local addresses are permissible for
18081 * IPIF_POINTOPOINT interfaces which will get marked
18082 * IPIF_UNNUMBERED later in
18083 * ip_addr_availability_check().
18085 * The nce_ipif_cnt field tracks the number of
18086 * ipifs that have nce_addr as their local address.
18088 ipif
->ipif_addr_ready
= 1;
18089 ipif
->ipif_added_nce
= 1;
18090 nce
->nce_ipif_cnt
++;
18094 ASSERT(nce
== NULL
);
18097 if (arp_no_defense
) {
18098 if ((ipif
->ipif_flags
& IPIF_UP
) &&
18099 !ipif
->ipif_addr_ready
)
18100 ipif_up_notify(ipif
);
18101 ipif
->ipif_addr_ready
= 1;
18104 /* zero address. nothing to publish */
18105 ipif
->ipif_addr_ready
= 1;
18110 if (added_ipif
&& err
!= 0)
18111 ipmp_illgrp_del_ipif(ill
->ill_grp
, ipif
);
18116 ipif_arp_up(ipif_t
*ipif
, enum ip_resolver_action res_act
, boolean_t was_dup
)
18119 ill_t
*ill
= ipif
->ipif_ill
;
18120 boolean_t first_interface
, wait_for_dlpi
= B_FALSE
;
18122 DTRACE_PROBE3(ipif__downup
, char *, "ipif_arp_up",
18123 ill_t
*, ill
, ipif_t
*, ipif
);
18126 * need to bring up ARP or setup mcast mapping only
18127 * when the first interface is coming UP.
18129 first_interface
= (ill
->ill_ipif_up_count
== 0 &&
18130 ill
->ill_ipif_dup_count
== 0 && !was_dup
);
18132 if (res_act
== Res_act_initial
&& first_interface
) {
18134 * Send ATTACH + BIND
18136 err
= arp_ll_up(ill
);
18137 if (err
!= EINPROGRESS
&& err
!= 0)
18141 * Add NCE for local address. Start DAD.
18142 * we'll wait to hear that DAD has finished
18143 * before using the interface.
18145 if (err
== EINPROGRESS
)
18146 wait_for_dlpi
= B_TRUE
;
18149 if (!wait_for_dlpi
)
18150 (void) ipif_arp_up_done_tail(ipif
, res_act
);
18152 return (!wait_for_dlpi
? 0 : EINPROGRESS
);
18156 * Finish processing of "arp_up" after all the DLPI message
18157 * exchanges have completed between arp and the driver.
18160 arp_bringup_done(ill_t
*ill
, int err
)
18164 conn_t
*connp
= NULL
;
18168 ip1dbg(("arp_bringup_done(%s)\n", ill
->ill_name
));
18170 ASSERT(IAM_WRITER_ILL(ill
));
18172 ipsq
= ill
->ill_phyint
->phyint_ipsq
;
18173 ipif
= ipsq
->ipsq_xop
->ipx_pending_ipif
;
18174 mp1
= ipsq_pending_mp_get(ipsq
, &connp
);
18175 ASSERT(!((mp1
!= NULL
) ^ (ipif
!= NULL
)));
18176 if (mp1
== NULL
) /* bringup was aborted by the user */
18180 * If an IOCTL is waiting on this (ipsq_current_ioctl != 0), then we
18181 * must have an associated conn_t. Otherwise, we're bringing this
18182 * interface back up as part of handling an asynchronous event (e.g.,
18183 * physical address change).
18185 if (ipsq
->ipsq_xop
->ipx_current_ioctl
!= 0) {
18186 ASSERT(connp
!= NULL
);
18187 q
= CONNP_TO_WQ(connp
);
18189 ASSERT(connp
== NULL
);
18193 if (ipif
->ipif_isv6
) {
18194 if ((err
= ipif_up_done_v6(ipif
)) != 0)
18195 ip0dbg(("arp_bringup_done: init failed\n"));
18197 err
= ipif_arp_up_done_tail(ipif
, Res_act_initial
);
18199 (err
= ipif_up_done(ipif
)) != 0) {
18200 ip0dbg(("arp_bringup_done: "
18201 "init failed err %x\n", err
));
18202 (void) ipif_arp_down(ipif
);
18207 ip0dbg(("arp_bringup_done: DL_BIND_REQ failed\n"));
18210 if ((err
== 0) && (ill
->ill_up_ipifs
)) {
18211 err
= ill_up_ipifs(ill
, q
, mp1
);
18212 if (err
== EINPROGRESS
)
18217 * If we have a moved ipif to bring up, and everything has succeeded
18218 * to this point, bring it up on the IPMP ill. Otherwise, leave it
18219 * down -- the admin can try to bring it up by hand if need be.
18221 if (ill
->ill_move_ipif
!= NULL
) {
18222 ipif
= ill
->ill_move_ipif
;
18223 ip1dbg(("bringing up ipif %p on ill %s\n", (void *)ipif
,
18224 ipif
->ipif_ill
->ill_name
));
18225 ill
->ill_move_ipif
= NULL
;
18227 err
= ipif_up(ipif
, q
, mp1
);
18228 if (err
== EINPROGRESS
)
18234 * The operation must complete without EINPROGRESS since
18235 * ipsq_pending_mp_get() has removed the mblk from ipsq_pending_mp.
18236 * Otherwise, the operation will be stuck forever in the ipsq.
18238 ASSERT(err
!= EINPROGRESS
);
18239 if (ipsq
->ipsq_xop
->ipx_current_ioctl
!= 0) {
18240 DTRACE_PROBE4(ipif__ioctl
, char *, "arp_bringup_done finish",
18241 int, ipsq
->ipsq_xop
->ipx_current_ioctl
,
18242 ill_t
*, ill
, ipif_t
*, ipif
);
18243 ip_ioctl_finish(q
, mp1
, err
, NO_COPYOUT
, ipsq
);
18245 ipsq_current_finish(ipsq
);
18250 * Finish processing of arp replumb after all the DLPI message
18251 * exchanges have completed between arp and the driver.
18254 arp_replumb_done(ill_t
*ill
, int err
)
18258 conn_t
*connp
= NULL
;
18262 ASSERT(IAM_WRITER_ILL(ill
));
18264 ipsq
= ill
->ill_phyint
->phyint_ipsq
;
18265 ipif
= ipsq
->ipsq_xop
->ipx_pending_ipif
;
18266 mp1
= ipsq_pending_mp_get(ipsq
, &connp
);
18267 ASSERT(!((mp1
!= NULL
) ^ (ipif
!= NULL
)));
18269 ip0dbg(("arp_replumb_done: bringup aborted ioctl %x\n",
18270 ipsq
->ipsq_xop
->ipx_current_ioctl
));
18271 /* bringup was aborted by the user */
18275 * If an IOCTL is waiting on this (ipsq_current_ioctl != 0), then we
18276 * must have an associated conn_t. Otherwise, we're bringing this
18277 * interface back up as part of handling an asynchronous event (e.g.,
18278 * physical address change).
18280 if (ipsq
->ipsq_xop
->ipx_current_ioctl
!= 0) {
18281 ASSERT(connp
!= NULL
);
18282 q
= CONNP_TO_WQ(connp
);
18284 ASSERT(connp
== NULL
);
18287 if ((err
== 0) && (ill
->ill_up_ipifs
)) {
18288 err
= ill_up_ipifs(ill
, q
, mp1
);
18289 if (err
== EINPROGRESS
)
18293 * The operation must complete without EINPROGRESS since
18294 * ipsq_pending_mp_get() has removed the mblk from ipsq_pending_mp.
18295 * Otherwise, the operation will be stuck forever in the ipsq.
18297 ASSERT(err
!= EINPROGRESS
);
18298 if (ipsq
->ipsq_xop
->ipx_current_ioctl
!= 0) {
18299 DTRACE_PROBE4(ipif__ioctl
, char *,
18300 "arp_replumb_done finish",
18301 int, ipsq
->ipsq_xop
->ipx_current_ioctl
,
18302 ill_t
*, ill
, ipif_t
*, ipif
);
18303 ip_ioctl_finish(q
, mp1
, err
, NO_COPYOUT
, ipsq
);
18305 ipsq_current_finish(ipsq
);
18310 ipif_up_notify(ipif_t
*ipif
)
18312 ip_rts_ifmsg(ipif
, RTSQ_DEFAULT
);
18313 ip_rts_newaddrmsg(RTM_ADD
, 0, ipif
, RTSQ_DEFAULT
);
18314 sctp_update_ipif(ipif
, SCTP_IPIF_UP
);
18315 ill_nic_event_dispatch(ipif
->ipif_ill
, MAP_IPIF_ID(ipif
->ipif_id
),
18316 NE_LIF_UP
, NULL
, 0);
18320 * ILB ioctl uses cv_wait (such as deleting a rule or adding a server) and
18321 * this assumes the context is cv_wait'able. Hence it shouldnt' be used on
18322 * TPI end points with STREAMS modules pushed above. This is assured by not
18323 * having the IPI_MODOK flag for the ioctl. And IP ensures the ILB ioctl
18324 * never ends up on an ipsq, otherwise we may end up processing the ioctl
18325 * while unwinding from the ispq and that could be a thread from the bottom.
18329 ip_sioctl_ilb_cmd(ipif_t
*ipif
, sin_t
*sin
, queue_t
*q
, mblk_t
*mp
,
18330 ip_ioctl_cmd_t
*ipip
, void *arg
)
18332 mblk_t
*cmd_mp
= mp
->b_cont
->b_cont
;
18333 ilb_cmd_t command
= *((ilb_cmd_t
*)cmd_mp
->b_rptr
);
18341 ipst
= CONNQ_TO_IPST(q
);
18342 ilbs
= ipst
->ips_netstack
->netstack_ilb
;
18343 zoneid
= Q_TO_CONN(q
)->conn_zoneid
;
18346 case ILB_CREATE_RULE
: {
18347 ilb_rule_cmd_t
*cmd
= (ilb_rule_cmd_t
*)cmd_mp
->b_rptr
;
18349 if (MBLKL(cmd_mp
) != sizeof (ilb_rule_cmd_t
)) {
18354 ret
= ilb_rule_add(ilbs
, zoneid
, cmd
);
18357 case ILB_DESTROY_RULE
:
18358 case ILB_ENABLE_RULE
:
18359 case ILB_DISABLE_RULE
: {
18360 ilb_name_cmd_t
*cmd
= (ilb_name_cmd_t
*)cmd_mp
->b_rptr
;
18362 if (MBLKL(cmd_mp
) != sizeof (ilb_name_cmd_t
)) {
18367 if (cmd
->flags
& ILB_RULE_ALLRULES
) {
18368 if (command
== ILB_DESTROY_RULE
) {
18369 ilb_rule_del_all(ilbs
, zoneid
);
18371 } else if (command
== ILB_ENABLE_RULE
) {
18372 ilb_rule_enable_all(ilbs
, zoneid
);
18374 } else if (command
== ILB_DISABLE_RULE
) {
18375 ilb_rule_disable_all(ilbs
, zoneid
);
18379 if (command
== ILB_DESTROY_RULE
) {
18380 ret
= ilb_rule_del(ilbs
, zoneid
, cmd
->name
);
18381 } else if (command
== ILB_ENABLE_RULE
) {
18382 ret
= ilb_rule_enable(ilbs
, zoneid
, cmd
->name
,
18384 } else if (command
== ILB_DISABLE_RULE
) {
18385 ret
= ilb_rule_disable(ilbs
, zoneid
, cmd
->name
,
18391 case ILB_NUM_RULES
: {
18392 ilb_num_rules_cmd_t
*cmd
;
18394 if (MBLKL(cmd_mp
) != sizeof (ilb_num_rules_cmd_t
)) {
18398 cmd
= (ilb_num_rules_cmd_t
*)cmd_mp
->b_rptr
;
18399 ilb_get_num_rules(ilbs
, zoneid
, &(cmd
->num
));
18402 case ILB_RULE_NAMES
: {
18403 ilb_rule_names_cmd_t
*cmd
;
18405 cmd
= (ilb_rule_names_cmd_t
*)cmd_mp
->b_rptr
;
18406 if (MBLKL(cmd_mp
) < sizeof (ilb_rule_names_cmd_t
) ||
18407 cmd
->num_names
== 0) {
18411 size
= cmd
->num_names
* ILB_RULE_NAMESZ
;
18412 if (cmd_mp
->b_rptr
+ offsetof(ilb_rule_names_cmd_t
, buf
) +
18413 size
!= cmd_mp
->b_wptr
) {
18417 ilb_get_rulenames(ilbs
, zoneid
, &cmd
->num_names
, cmd
->buf
);
18420 case ILB_NUM_SERVERS
: {
18421 ilb_num_servers_cmd_t
*cmd
;
18423 if (MBLKL(cmd_mp
) != sizeof (ilb_num_servers_cmd_t
)) {
18427 cmd
= (ilb_num_servers_cmd_t
*)cmd_mp
->b_rptr
;
18428 ret
= ilb_get_num_servers(ilbs
, zoneid
, cmd
->name
,
18432 case ILB_LIST_RULE
: {
18433 ilb_rule_cmd_t
*cmd
= (ilb_rule_cmd_t
*)cmd_mp
->b_rptr
;
18435 if (MBLKL(cmd_mp
) != sizeof (ilb_rule_cmd_t
)) {
18439 ret
= ilb_rule_list(ilbs
, zoneid
, cmd
);
18442 case ILB_LIST_SERVERS
: {
18443 ilb_servers_info_cmd_t
*cmd
;
18445 cmd
= (ilb_servers_info_cmd_t
*)cmd_mp
->b_rptr
;
18446 if (MBLKL(cmd_mp
) < sizeof (ilb_servers_info_cmd_t
) ||
18447 cmd
->num_servers
== 0) {
18451 size
= cmd
->num_servers
* sizeof (ilb_server_info_t
);
18452 if (cmd_mp
->b_rptr
+ offsetof(ilb_servers_info_cmd_t
, servers
) +
18453 size
!= cmd_mp
->b_wptr
) {
18458 ret
= ilb_get_servers(ilbs
, zoneid
, cmd
->name
, cmd
->servers
,
18459 &cmd
->num_servers
);
18462 case ILB_ADD_SERVERS
: {
18463 ilb_servers_info_cmd_t
*cmd
;
18466 cmd
= (ilb_servers_info_cmd_t
*)cmd_mp
->b_rptr
;
18467 if (MBLKL(cmd_mp
) < sizeof (ilb_servers_info_cmd_t
)) {
18471 size
= cmd
->num_servers
* sizeof (ilb_server_info_t
);
18472 if (cmd_mp
->b_rptr
+ offsetof(ilb_servers_info_cmd_t
, servers
) +
18473 size
!= cmd_mp
->b_wptr
) {
18477 rule
= ilb_find_rule(ilbs
, zoneid
, cmd
->name
, &ret
);
18478 if (rule
== NULL
) {
18482 for (i
= 0; i
< cmd
->num_servers
; i
++) {
18483 ilb_server_info_t
*s
;
18485 s
= &cmd
->servers
[i
];
18486 s
->err
= ilb_server_add(ilbs
, rule
, s
);
18488 ILB_RULE_REFRELE(rule
);
18491 case ILB_DEL_SERVERS
:
18492 case ILB_ENABLE_SERVERS
:
18493 case ILB_DISABLE_SERVERS
: {
18494 ilb_servers_cmd_t
*cmd
;
18498 cmd
= (ilb_servers_cmd_t
*)cmd_mp
->b_rptr
;
18499 if (MBLKL(cmd_mp
) < sizeof (ilb_servers_cmd_t
)) {
18503 size
= cmd
->num_servers
* sizeof (ilb_server_arg_t
);
18504 if (cmd_mp
->b_rptr
+ offsetof(ilb_servers_cmd_t
, servers
) +
18505 size
!= cmd_mp
->b_wptr
) {
18510 if (command
== ILB_DEL_SERVERS
)
18511 f
= ilb_server_del
;
18512 else if (command
== ILB_ENABLE_SERVERS
)
18513 f
= ilb_server_enable
;
18514 else if (command
== ILB_DISABLE_SERVERS
)
18515 f
= ilb_server_disable
;
18517 rule
= ilb_find_rule(ilbs
, zoneid
, cmd
->name
, &ret
);
18518 if (rule
== NULL
) {
18523 for (i
= 0; i
< cmd
->num_servers
; i
++) {
18524 ilb_server_arg_t
*s
;
18526 s
= &cmd
->servers
[i
];
18527 s
->err
= f(ilbs
, zoneid
, NULL
, rule
, &s
->addr
);
18529 ILB_RULE_REFRELE(rule
);
18532 case ILB_LIST_NAT_TABLE
: {
18533 ilb_list_nat_cmd_t
*cmd
;
18535 cmd
= (ilb_list_nat_cmd_t
*)cmd_mp
->b_rptr
;
18536 if (MBLKL(cmd_mp
) < sizeof (ilb_list_nat_cmd_t
)) {
18540 size
= cmd
->num_nat
* sizeof (ilb_nat_entry_t
);
18541 if (cmd_mp
->b_rptr
+ offsetof(ilb_list_nat_cmd_t
, entries
) +
18542 size
!= cmd_mp
->b_wptr
) {
18547 ret
= ilb_list_nat(ilbs
, zoneid
, cmd
->entries
, &cmd
->num_nat
,
18551 case ILB_LIST_STICKY_TABLE
: {
18552 ilb_list_sticky_cmd_t
*cmd
;
18554 cmd
= (ilb_list_sticky_cmd_t
*)cmd_mp
->b_rptr
;
18555 if (MBLKL(cmd_mp
) < sizeof (ilb_list_sticky_cmd_t
)) {
18559 size
= cmd
->num_sticky
* sizeof (ilb_sticky_entry_t
);
18560 if (cmd_mp
->b_rptr
+ offsetof(ilb_list_sticky_cmd_t
, entries
) +
18561 size
!= cmd_mp
->b_wptr
) {
18566 ret
= ilb_list_sticky(ilbs
, zoneid
, cmd
->entries
,
18567 &cmd
->num_sticky
, &cmd
->flags
);
18578 /* Remove all cache entries for this logical interface */
18580 ipif_nce_down(ipif_t
*ipif
)
18582 ill_t
*ill
= ipif
->ipif_ill
;
18585 DTRACE_PROBE3(ipif__downup
, char *, "ipif_nce_down",
18586 ill_t
*, ill
, ipif_t
*, ipif
);
18587 if (ipif
->ipif_added_nce
) {
18588 if (ipif
->ipif_isv6
)
18589 nce
= nce_lookup_v6(ill
, &ipif
->ipif_v6lcl_addr
);
18591 nce
= nce_lookup_v4(ill
, &ipif
->ipif_lcl_addr
);
18593 if (--nce
->nce_ipif_cnt
== 0)
18594 ncec_delete(nce
->nce_common
);
18595 ipif
->ipif_added_nce
= 0;
18599 * nce may already be NULL because it was already
18600 * flushed, e.g., due to a call to nce_flush
18602 ipif
->ipif_added_nce
= 0;
18606 * Make IPMP aware of the deleted data address.
18609 ipmp_illgrp_del_ipif(ill
->ill_grp
, ipif
);
18612 * Remove all other nces dependent on this ill when the last ipif
18615 if (ill
->ill_ipif_up_count
== 0) {
18616 ncec_walk(ill
, ncec_delete_per_ill
, ill
, ill
->ill_ipst
);
18617 if (IS_UNDER_IPMP(ill
))
18618 nce_flush(ill
, B_TRUE
);
18623 * find the first interface that uses usill for its source address.
18626 ill_lookup_usesrc(ill_t
*usill
)
18628 ip_stack_t
*ipst
= usill
->ill_ipst
;
18631 ASSERT(usill
!= NULL
);
18633 /* ill_g_usesrc_lock protects ill_usesrc_grp_next */
18634 rw_enter(&ipst
->ips_ill_g_usesrc_lock
, RW_WRITER
);
18635 rw_enter(&ipst
->ips_ill_g_lock
, RW_READER
);
18636 for (ill
= usill
->ill_usesrc_grp_next
; ill
!= NULL
&& ill
!= usill
;
18637 ill
= ill
->ill_usesrc_grp_next
) {
18638 if (!IS_UNDER_IPMP(ill
) && (ill
->ill_flags
& ILLF_MULTICAST
) &&
18639 !ILL_IS_CONDEMNED(ill
)) {
18644 rw_exit(&ipst
->ips_ill_g_lock
);
18645 rw_exit(&ipst
->ips_ill_g_usesrc_lock
);
18650 * This comment applies to both ip_sioctl_get_ifhwaddr and
18651 * ip_sioctl_get_lifhwaddr as the basic function of these two functions
18654 * The goal here is to find an IP interface that corresponds to the name
18655 * provided by the caller in the ifreq/lifreq structure held in the mblk_t
18656 * chain and to fill out a sockaddr/sockaddr_storage structure with the
18659 * The SIOCGIFHWADDR/SIOCGLIFHWADDR ioctl may return an error for a number
18660 * of different reasons:
18661 * ENXIO - the device name is not known to IP.
18662 * EADDRNOTAVAIL - the device has no hardware address. This is indicated
18663 * by ill_phys_addr not pointing to an actual address.
18664 * EPFNOSUPPORT - this will indicate that a request is being made for a
18665 * mac address that will not fit in the data structure supplier (struct
18671 ip_sioctl_get_ifhwaddr(ipif_t
*ipif
, sin_t
*dummy_sin
, queue_t
*q
, mblk_t
*mp
,
18672 ip_ioctl_cmd_t
*ipip
, void *if_req
)
18674 struct sockaddr
*sock
;
18679 ASSERT(ipif
!= NULL
);
18680 ill
= ipif
->ipif_ill
;
18682 if (ill
->ill_phys_addr
== NULL
) {
18683 return (EADDRNOTAVAIL
);
18685 if (ill
->ill_phys_addr_length
> sizeof (sock
->sa_data
)) {
18686 return (EPFNOSUPPORT
);
18689 ip1dbg(("ip_sioctl_get_hwaddr(%s)\n", ill
->ill_name
));
18691 /* Existence of mp1 has been checked in ip_wput_nondata */
18692 mp1
= mp
->b_cont
->b_cont
;
18693 ifr
= (struct ifreq
*)mp1
->b_rptr
;
18695 sock
= &ifr
->ifr_addr
;
18697 * The "family" field in the returned structure is set to a value
18698 * that represents the type of device to which the address belongs.
18699 * The value returned may differ to that on Linux but it will still
18700 * represent the correct symbol on Solaris.
18702 sock
->sa_family
= arp_hw_type(ill
->ill_mactype
);
18703 bcopy(ill
->ill_phys_addr
, &sock
->sa_data
, ill
->ill_phys_addr_length
);
18709 * The expection of applications using SIOCGIFHWADDR is that data will
18710 * be returned in the sa_data field of the sockaddr structure. With
18711 * SIOCGLIFHWADDR, we're breaking new ground as there is no Linux
18712 * equivalent. In light of this, struct sockaddr_dl is used as it
18713 * offers more space for address storage in sll_data.
18717 ip_sioctl_get_lifhwaddr(ipif_t
*ipif
, sin_t
*dummy_sin
, queue_t
*q
, mblk_t
*mp
,
18718 ip_ioctl_cmd_t
*ipip
, void *if_req
)
18720 struct sockaddr_dl
*sock
;
18721 struct lifreq
*lifr
;
18725 ASSERT(ipif
!= NULL
);
18726 ill
= ipif
->ipif_ill
;
18728 if (ill
->ill_phys_addr
== NULL
) {
18729 return (EADDRNOTAVAIL
);
18731 if (ill
->ill_phys_addr_length
> sizeof (sock
->sdl_data
)) {
18732 return (EPFNOSUPPORT
);
18735 ip1dbg(("ip_sioctl_get_lifhwaddr(%s)\n", ill
->ill_name
));
18737 /* Existence of mp1 has been checked in ip_wput_nondata */
18738 mp1
= mp
->b_cont
->b_cont
;
18739 lifr
= (struct lifreq
*)mp1
->b_rptr
;
18742 * sockaddr_ll is used here because it is also the structure used in
18743 * responding to the same ioctl in sockpfp. The only other choice is
18744 * sockaddr_dl which contains fields that are not required here
18745 * because its purpose is different.
18747 lifr
->lifr_type
= ill
->ill_type
;
18748 sock
= (struct sockaddr_dl
*)&lifr
->lifr_addr
;
18749 sock
->sdl_family
= AF_LINK
;
18750 sock
->sdl_index
= ill
->ill_phyint
->phyint_ifindex
;
18751 sock
->sdl_type
= ill
->ill_mactype
;
18752 sock
->sdl_nlen
= 0;
18753 sock
->sdl_slen
= 0;
18754 sock
->sdl_alen
= ill
->ill_phys_addr_length
;
18755 bcopy(ill
->ill_phys_addr
, sock
->sdl_data
, ill
->ill_phys_addr_length
);