2 * Copyright (c) 2003, 2004 Matthew Dillon. All rights reserved.
3 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved.
4 * Copyright (c) 2003 Jonathan Lemon. All rights reserved.
5 * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved.
7 * This code is derived from software contributed to The DragonFly Project
8 * by Jonathan Lemon, Jeffrey M. Hsu, and Matthew Dillon.
10 * Jonathan Lemon gave Jeffrey Hsu permission to combine his copyright
11 * into this one around July 8 2004.
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. Neither the name of The DragonFly Project nor the names of its
22 * contributors may be used to endorse or promote products derived
23 * from this software without specific, prior written permission.
25 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
26 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
27 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
28 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
29 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
30 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
31 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
32 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
33 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
34 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
35 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * $DragonFly: src/sys/net/netisr.c,v 1.27 2006/09/05 00:55:46 dillon Exp $
42 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved.
44 * License terms: all terms for the DragonFly license above plus the following:
46 * 4. All advertising materials mentioning features or use of this software
47 * must display the following acknowledgement:
49 * This product includes software developed by Jeffrey M. Hsu
50 * for the DragonFly Project.
52 * This requirement may be waived with permission from Jeffrey Hsu.
53 * This requirement will sunset and may be removed on July 8 2005,
54 * after which the standard DragonFly license (as shown above) will
58 #include <sys/param.h>
59 #include <sys/systm.h>
60 #include <sys/kernel.h>
61 #include <sys/malloc.h>
62 #include <sys/msgport.h>
64 #include <sys/interrupt.h>
65 #include <sys/socket.h>
66 #include <sys/sysctl.h>
68 #include <net/if_var.h>
69 #include <net/netisr.h>
70 #include <machine/cpufunc.h>
71 #include <machine/ipl.h>
73 #include <sys/thread2.h>
74 #include <sys/msgport2.h>
76 static int netmsg_sync_func(struct netmsg
*msg
);
78 struct netmsg_port_registration
{
79 TAILQ_ENTRY(netmsg_port_registration
) npr_entry
;
83 static struct netisr netisrs
[NETISR_MAX
];
84 static TAILQ_HEAD(,netmsg_port_registration
) netreglist
;
86 /* Per-CPU thread to handle any protocol. */
87 struct thread netisr_cpu
[MAXCPU
];
88 lwkt_port netisr_afree_rport
;
89 lwkt_port netisr_adone_rport
;
90 lwkt_port netisr_apanic_rport
;
91 lwkt_port netisr_sync_port
;
94 * netisr_afree_rport replymsg function, only used to handle async
95 * messages which the sender has abandoned to their fate.
98 netisr_autofree_reply(lwkt_port_t port
, lwkt_msg_t msg
)
100 kfree(msg
, M_LWKTMSG
);
104 netisr_autopanic_reply(lwkt_port_t port
, lwkt_msg_t msg
)
106 panic("unreplyable msg %p was replied!", msg
);
110 * We must construct a custom putport function (which runs in the context
111 * of the message originator)
113 * Our custom putport must check for self-referential messages, which can
114 * occur when the so_upcall routine is called (e.g. nfs). Self referential
115 * messages are executed synchronously. However, we must panic if the message
116 * is not marked DONE on completion because the self-referential case cannot
117 * block without deadlocking.
119 * note: ms_target_port does not need to be set when returning a synchronous
123 netmsg_put_port(lwkt_port_t port
, lwkt_msg_t lmsg
)
127 if ((lmsg
->ms_flags
& MSGF_ASYNC
) == 0 && port
->mp_td
== curthread
) {
128 error
= lmsg
->ms_cmd
.cm_func(lmsg
);
129 if (error
== EASYNC
&& (lmsg
->ms_flags
& MSGF_DONE
) == 0)
130 panic("netmsg_put_port: self-referential deadlock on netport");
133 return(lwkt_default_putport(port
, lmsg
));
138 * UNIX DOMAIN sockets still have to run their uipc functions synchronously,
139 * because they depend on the user proc context for a number of things
140 * (like creds) which we have not yet incorporated into the message structure.
142 * However, we maintain or message/port abstraction. Having a special
143 * synchronous port which runs the commands synchronously gives us the
144 * ability to serialize operations in one place later on when we start
147 * We clear MSGF_DONE prior to executing the message in order to close
148 * any potential replymsg races with the flags field. If a synchronous
149 * result code is returned we set MSGF_DONE again. MSGF_DONE's flag state
150 * must be correct or the caller will be confused.
153 netmsg_sync_putport(lwkt_port_t port
, lwkt_msg_t lmsg
)
157 lmsg
->ms_flags
&= ~MSGF_DONE
;
158 lmsg
->ms_target_port
= port
; /* required for abort */
159 error
= lmsg
->ms_cmd
.cm_func(lmsg
);
161 error
= lwkt_waitmsg(lmsg
);
163 lmsg
->ms_flags
|= MSGF_DONE
;
168 netmsg_sync_abortport(lwkt_port_t port
, lwkt_msg_t lmsg
)
170 lmsg
->ms_abort_port
= lmsg
->ms_reply_port
;
171 lmsg
->ms_flags
|= MSGF_ABORTED
;
172 lmsg
->ms_abort
.cm_func(lmsg
);
180 TAILQ_INIT(&netreglist
);
183 * Create default per-cpu threads for generic protocol handling.
185 for (i
= 0; i
< ncpus
; ++i
) {
186 lwkt_create(netmsg_service_loop
, NULL
, NULL
, &netisr_cpu
[i
], 0, i
,
188 netmsg_service_port_init(&netisr_cpu
[i
].td_msgport
);
192 * The netisr_afree_rport is a special reply port which automatically
193 * frees the replied message. The netisr_adone_rport simply marks
194 * the message as being done. The netisr_apanic_rport panics if
195 * the message is replied to.
197 lwkt_initport(&netisr_afree_rport
, NULL
);
198 netisr_afree_rport
.mp_replyport
= netisr_autofree_reply
;
199 lwkt_initport_null_rport(&netisr_adone_rport
, NULL
);
200 lwkt_initport(&netisr_apanic_rport
, NULL
);
201 netisr_apanic_rport
.mp_replyport
= netisr_autopanic_reply
;
204 * The netisr_syncport is a special port which executes the message
205 * synchronously and waits for it if EASYNC is returned.
207 lwkt_initport(&netisr_sync_port
, NULL
);
208 netisr_sync_port
.mp_putport
= netmsg_sync_putport
;
209 netisr_sync_port
.mp_abortport
= netmsg_sync_abortport
;
212 SYSINIT(netisr
, SI_SUB_PROTO_BEGIN
, SI_ORDER_FIRST
, netisr_init
, NULL
);
215 * Finish initializing the message port for a netmsg service. This also
216 * registers the port for synchronous cleanup operations such as when an
217 * ifnet is being destroyed. There is no deregistration API yet.
220 netmsg_service_port_init(lwkt_port_t port
)
222 struct netmsg_port_registration
*reg
;
225 * Override the putport function. Our custom function checks for
226 * self-references and executes such commands synchronously.
228 port
->mp_putport
= netmsg_put_port
;
231 * Keep track of ports using the netmsg API so we can synchronize
232 * certain operations (such as freeing an ifnet structure) across all
235 reg
= kmalloc(sizeof(*reg
), M_TEMP
, M_WAITOK
|M_ZERO
);
236 reg
->npr_port
= port
;
237 TAILQ_INSERT_TAIL(&netreglist
, reg
, npr_entry
);
241 * This function synchronizes the caller with all netmsg services. For
242 * example, if an interface is being removed we must make sure that all
243 * packets related to that interface complete processing before the structure
244 * can actually be freed. This sort of synchronization is an alternative to
245 * ref-counting the netif, removing the ref counting overhead in favor of
246 * placing additional overhead in the netif freeing sequence (where it is
250 netmsg_service_sync(void)
252 struct netmsg_port_registration
*reg
;
255 lwkt_initmsg(&smsg
.nm_lmsg
, &curthread
->td_msgport
, 0,
256 lwkt_cmd_func((void *)netmsg_sync_func
), lwkt_cmd_op_none
);
258 TAILQ_FOREACH(reg
, &netreglist
, npr_entry
) {
259 lwkt_domsg(reg
->npr_port
, &smsg
.nm_lmsg
);
264 * The netmsg function simply replies the message. API semantics require
265 * EASYNC to be returned if the netmsg function disposes of the message.
268 netmsg_sync_func(struct netmsg
*msg
)
270 lwkt_replymsg(&msg
->nm_lmsg
, 0);
275 * Generic netmsg service loop. Some protocols may roll their own but all
276 * must do the basic command dispatch function call done here.
279 netmsg_service_loop(void *arg
)
283 while ((msg
= lwkt_waitport(&curthread
->td_msgport
, NULL
))) {
284 msg
->nm_lmsg
.ms_cmd
.cm_func(&msg
->nm_lmsg
);
289 * Call the netisr directly.
290 * Queueing may be done in the msg port layer at its discretion.
293 netisr_dispatch(int num
, struct mbuf
*m
)
295 /* just queue it for now XXX JH */
296 netisr_queue(num
, m
);
300 * Same as netisr_dispatch(), but always queue.
301 * This is either used in places where we are not confident that
302 * direct dispatch is possible, or where queueing is required.
305 netisr_queue(int num
, struct mbuf
*m
)
308 struct netmsg_packet
*pmsg
;
311 KASSERT((num
> 0 && num
<= (sizeof(netisrs
)/sizeof(netisrs
[0]))),
312 ("netisr_queue: bad isr %d", num
));
315 if (ni
->ni_handler
== NULL
) {
316 printf("netisr_queue: unregistered isr %d\n", num
);
320 if ((port
= ni
->ni_mport(&m
)) == NULL
)
323 pmsg
= &m
->m_hdr
.mh_netmsg
;
325 lwkt_initmsg(&pmsg
->nm_lmsg
, &netisr_apanic_rport
, 0,
326 lwkt_cmd_func((void *)ni
->ni_handler
), lwkt_cmd_op_none
);
328 pmsg
->nm_lmsg
.u
.ms_result
= num
;
329 lwkt_sendmsg(port
, &pmsg
->nm_lmsg
);
334 netisr_register(int num
, lwkt_portfn_t mportfn
, netisr_fn_t handler
)
336 KASSERT((num
> 0 && num
<= (sizeof(netisrs
)/sizeof(netisrs
[0]))),
337 ("netisr_register: bad isr %d", num
));
338 lwkt_initmsg(&netisrs
[num
].ni_netmsg
.nm_lmsg
, &netisr_adone_rport
, 0,
339 lwkt_cmd_op_none
, lwkt_cmd_op_none
);
340 netisrs
[num
].ni_mport
= mportfn
;
341 netisrs
[num
].ni_handler
= handler
;
345 netisr_unregister(int num
)
347 KASSERT((num
> 0 && num
<= (sizeof(netisrs
)/sizeof(netisrs
[0]))),
348 ("unregister_netisr: bad isr number: %d\n", num
));
355 * Return message port for default handler thread on CPU 0.
358 cpu0_portfn(struct mbuf
**mptr
)
360 return (&netisr_cpu
[0].td_msgport
);
366 return (&netisr_cpu
[cpu
].td_msgport
);
371 cpu0_soport(struct socket
*so __unused
, struct sockaddr
*nam __unused
,
374 return (&netisr_cpu
[0].td_msgport
);
378 sync_soport(struct socket
*so __unused
, struct sockaddr
*nam __unused
,
381 return (&netisr_sync_port
);
385 * schednetisr() is used to call the netisr handler from the appropriate
386 * netisr thread for polling and other purposes.
388 * This function may be called from a hard interrupt or IPI and must be
389 * MP SAFE and non-blocking. We use a fixed per-cpu message instead of
390 * trying to allocate one. We must get ourselves onto the target cpu
391 * to safely check the MSGF_DONE bit on the message but since the message
392 * will be sent to that cpu anyway this does not add any extra work beyond
393 * what lwkt_sendmsg() would have already had to do to schedule the target
397 schednetisr_remote(void *data
)
400 struct netisr
*ni
= &netisrs
[num
];
401 lwkt_port_t port
= &netisr_cpu
[0].td_msgport
;
404 pmsg
= &netisrs
[num
].ni_netmsg
;
406 if (pmsg
->nm_lmsg
.ms_flags
& MSGF_DONE
) {
407 lwkt_initmsg(&pmsg
->nm_lmsg
, &netisr_adone_rport
, 0,
408 lwkt_cmd_func((void *)ni
->ni_handler
), lwkt_cmd_op_none
);
409 pmsg
->nm_lmsg
.u
.ms_result
= num
;
410 lwkt_sendmsg(port
, &pmsg
->nm_lmsg
);
418 KASSERT((num
> 0 && num
<= (sizeof(netisrs
)/sizeof(netisrs
[0]))),
419 ("schednetisr: bad isr %d", num
));
421 if (mycpu
->gd_cpuid
!= 0)
422 lwkt_send_ipiq(globaldata_find(0), schednetisr_remote
, (void *)num
);
424 schednetisr_remote((void *)num
);
426 schednetisr_remote((void *)num
);