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 1.33 2007/05/24 20:51:21 dillon Exp $
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/kernel.h>
44 #include <sys/malloc.h>
45 #include <sys/msgport.h>
47 #include <sys/interrupt.h>
48 #include <sys/socket.h>
49 #include <sys/sysctl.h>
51 #include <net/if_var.h>
52 #include <net/netisr.h>
53 #include <machine/cpufunc.h>
55 #include <sys/thread2.h>
56 #include <sys/msgport2.h>
57 #include <net/netmsg2.h>
59 static void netmsg_sync_func(struct netmsg
*msg
);
61 struct netmsg_port_registration
{
62 TAILQ_ENTRY(netmsg_port_registration
) npr_entry
;
66 static struct netisr netisrs
[NETISR_MAX
];
67 static TAILQ_HEAD(,netmsg_port_registration
) netreglist
;
69 /* Per-CPU thread to handle any protocol. */
70 struct thread netisr_cpu
[MAXCPU
];
71 lwkt_port netisr_afree_rport
;
72 lwkt_port netisr_adone_rport
;
73 lwkt_port netisr_apanic_rport
;
74 lwkt_port netisr_sync_port
;
76 static int (*netmsg_fwd_port_fn
)(lwkt_port_t
, lwkt_msg_t
);
79 * netisr_afree_rport replymsg function, only used to handle async
80 * messages which the sender has abandoned to their fate.
83 netisr_autofree_reply(lwkt_port_t port
, lwkt_msg_t msg
)
85 kfree(msg
, M_LWKTMSG
);
89 * We need a custom putport function to handle the case where the
90 * message target is the current thread's message port. This case
91 * can occur when the TCP or UDP stack does a direct callback to NFS and NFS
92 * then turns around and executes a network operation synchronously.
94 * To prevent deadlocking, we must execute these self-referential messages
95 * synchronously, effectively turning the message into a glorified direct
96 * procedure call back into the protocol stack. The operation must be
97 * complete on return or we will deadlock, so panic if it isn't.
100 netmsg_put_port(lwkt_port_t port
, lwkt_msg_t lmsg
)
102 netmsg_t netmsg
= (void *)lmsg
;
104 if ((lmsg
->ms_flags
& MSGF_SYNC
) && port
== &curthread
->td_msgport
) {
105 netmsg
->nm_dispatch(netmsg
);
106 if ((lmsg
->ms_flags
& MSGF_DONE
) == 0)
107 panic("netmsg_put_port: self-referential deadlock on netport");
110 return(netmsg_fwd_port_fn(port
, lmsg
));
115 * UNIX DOMAIN sockets still have to run their uipc functions synchronously,
116 * because they depend on the user proc context for a number of things
117 * (like creds) which we have not yet incorporated into the message structure.
119 * However, we maintain or message/port abstraction. Having a special
120 * synchronous port which runs the commands synchronously gives us the
121 * ability to serialize operations in one place later on when we start
124 * We clear MSGF_DONE prior to executing the message in order to close
125 * any potential replymsg races with the flags field. If a synchronous
126 * result code is returned we set MSGF_DONE again. MSGF_DONE's flag state
127 * must be correct or the caller will be confused.
130 netmsg_sync_putport(lwkt_port_t port
, lwkt_msg_t lmsg
)
132 netmsg_t netmsg
= (void *)lmsg
;
135 lmsg
->ms_flags
&= ~MSGF_DONE
;
136 lmsg
->ms_target_port
= port
; /* required for abort */
137 netmsg
->nm_dispatch(netmsg
);
138 error
= lwkt_waitmsg(lmsg
, 0);
147 TAILQ_INIT(&netreglist
);
150 * Create default per-cpu threads for generic protocol handling.
152 for (i
= 0; i
< ncpus
; ++i
) {
153 lwkt_create(netmsg_service_loop
, NULL
, NULL
, &netisr_cpu
[i
], 0, i
,
155 netmsg_service_port_init(&netisr_cpu
[i
].td_msgport
);
159 * The netisr_afree_rport is a special reply port which automatically
160 * frees the replied message. The netisr_adone_rport simply marks
161 * the message as being done. The netisr_apanic_rport panics if
162 * the message is replied to.
164 lwkt_initport_replyonly(&netisr_afree_rport
, netisr_autofree_reply
);
165 lwkt_initport_replyonly_null(&netisr_adone_rport
);
166 lwkt_initport_panic(&netisr_apanic_rport
);
169 * The netisr_syncport is a special port which executes the message
170 * synchronously and waits for it if EASYNC is returned.
172 lwkt_initport_putonly(&netisr_sync_port
, netmsg_sync_putport
);
173 netisr_sync_port
.mp_putport
= netmsg_sync_putport
;
176 SYSINIT(netisr
, SI_SUB_PROTO_BEGIN
, SI_ORDER_FIRST
, netisr_init
, NULL
);
179 * Finish initializing the message port for a netmsg service. This also
180 * registers the port for synchronous cleanup operations such as when an
181 * ifnet is being destroyed. There is no deregistration API yet.
184 netmsg_service_port_init(lwkt_port_t port
)
186 struct netmsg_port_registration
*reg
;
189 * Override the putport function. Our custom function checks for
190 * self-references and executes such commands synchronously.
192 if (netmsg_fwd_port_fn
== NULL
)
193 netmsg_fwd_port_fn
= port
->mp_putport
;
194 KKASSERT(netmsg_fwd_port_fn
== port
->mp_putport
);
195 port
->mp_putport
= netmsg_put_port
;
198 * Keep track of ports using the netmsg API so we can synchronize
199 * certain operations (such as freeing an ifnet structure) across all
202 reg
= kmalloc(sizeof(*reg
), M_TEMP
, M_WAITOK
|M_ZERO
);
203 reg
->npr_port
= port
;
204 TAILQ_INSERT_TAIL(&netreglist
, reg
, npr_entry
);
208 * This function synchronizes the caller with all netmsg services. For
209 * example, if an interface is being removed we must make sure that all
210 * packets related to that interface complete processing before the structure
211 * can actually be freed. This sort of synchronization is an alternative to
212 * ref-counting the netif, removing the ref counting overhead in favor of
213 * placing additional overhead in the netif freeing sequence (where it is
217 netmsg_service_sync(void)
219 struct netmsg_port_registration
*reg
;
222 netmsg_init(&smsg
, &curthread
->td_msgport
, 0, netmsg_sync_func
);
224 TAILQ_FOREACH(reg
, &netreglist
, npr_entry
) {
225 lwkt_domsg(reg
->npr_port
, &smsg
.nm_lmsg
, 0);
230 * The netmsg function simply replies the message. API semantics require
231 * EASYNC to be returned if the netmsg function disposes of the message.
234 netmsg_sync_func(struct netmsg
*msg
)
236 lwkt_replymsg(&msg
->nm_lmsg
, 0);
240 * Generic netmsg service loop. Some protocols may roll their own but all
241 * must do the basic command dispatch function call done here.
244 netmsg_service_loop(void *arg
)
248 while ((msg
= lwkt_waitport(&curthread
->td_msgport
, NULL
))) {
249 msg
->nm_dispatch(msg
);
254 * Call the netisr directly.
255 * Queueing may be done in the msg port layer at its discretion.
258 netisr_dispatch(int num
, struct mbuf
*m
)
260 /* just queue it for now XXX JH */
261 netisr_queue(num
, m
);
265 * Same as netisr_dispatch(), but always queue.
266 * This is either used in places where we are not confident that
267 * direct dispatch is possible, or where queueing is required.
270 netisr_queue(int num
, struct mbuf
*m
)
273 struct netmsg_packet
*pmsg
;
276 KASSERT((num
> 0 && num
<= (sizeof(netisrs
)/sizeof(netisrs
[0]))),
277 ("netisr_queue: bad isr %d", num
));
280 if (ni
->ni_handler
== NULL
) {
281 kprintf("netisr_queue: unregistered isr %d\n", num
);
285 if ((port
= ni
->ni_mport(&m
)) == NULL
)
288 pmsg
= &m
->m_hdr
.mh_netmsg
;
290 netmsg_init(&pmsg
->nm_netmsg
, &netisr_apanic_rport
, 0, ni
->ni_handler
);
292 pmsg
->nm_netmsg
.nm_lmsg
.u
.ms_result
= num
;
293 lwkt_sendmsg(port
, &pmsg
->nm_netmsg
.nm_lmsg
);
298 netisr_register(int num
, lwkt_portfn_t mportfn
, netisr_fn_t handler
)
300 KASSERT((num
> 0 && num
<= (sizeof(netisrs
)/sizeof(netisrs
[0]))),
301 ("netisr_register: bad isr %d", num
));
302 netmsg_init(&netisrs
[num
].ni_netmsg
, &netisr_adone_rport
, 0, NULL
);
303 netisrs
[num
].ni_mport
= mportfn
;
304 netisrs
[num
].ni_handler
= handler
;
308 netisr_unregister(int num
)
310 KASSERT((num
> 0 && num
<= (sizeof(netisrs
)/sizeof(netisrs
[0]))),
311 ("unregister_netisr: bad isr number: %d\n", num
));
318 * Return message port for default handler thread on CPU 0.
321 cpu0_portfn(struct mbuf
**mptr
)
323 return (&netisr_cpu
[0].td_msgport
);
329 return (&netisr_cpu
[cpu
].td_msgport
);
334 cpu0_soport(struct socket
*so __unused
, struct sockaddr
*nam __unused
,
337 return (&netisr_cpu
[0].td_msgport
);
341 sync_soport(struct socket
*so __unused
, struct sockaddr
*nam __unused
,
344 return (&netisr_sync_port
);
348 * schednetisr() is used to call the netisr handler from the appropriate
349 * netisr thread for polling and other purposes.
351 * This function may be called from a hard interrupt or IPI and must be
352 * MP SAFE and non-blocking. We use a fixed per-cpu message instead of
353 * trying to allocate one. We must get ourselves onto the target cpu
354 * to safely check the MSGF_DONE bit on the message but since the message
355 * will be sent to that cpu anyway this does not add any extra work beyond
356 * what lwkt_sendmsg() would have already had to do to schedule the target
360 schednetisr_remote(void *data
)
363 struct netisr
*ni
= &netisrs
[num
];
364 lwkt_port_t port
= &netisr_cpu
[0].td_msgport
;
367 pmsg
= &netisrs
[num
].ni_netmsg
;
369 if (pmsg
->nm_lmsg
.ms_flags
& MSGF_DONE
) {
370 netmsg_init(pmsg
, &netisr_adone_rport
, 0, ni
->ni_handler
);
371 pmsg
->nm_lmsg
.u
.ms_result
= num
;
372 lwkt_sendmsg(port
, &pmsg
->nm_lmsg
);
380 KASSERT((num
> 0 && num
<= (sizeof(netisrs
)/sizeof(netisrs
[0]))),
381 ("schednetisr: bad isr %d", num
));
383 if (mycpu
->gd_cpuid
!= 0)
384 lwkt_send_ipiq(globaldata_find(0), schednetisr_remote
, (void *)num
);
386 schednetisr_remote((void *)num
);
388 schednetisr_remote((void *)num
);