2 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved.
3 * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved.
5 * This code is derived from software contributed to The DragonFly Project
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of The DragonFly Project nor the names of its
17 * contributors may be used to endorse or promote products derived
18 * from this software without specific, prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * $DragonFly: src/sys/kern/uipc_msg.c,v 1.26 2008/10/27 02:56:30 sephe Exp $
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/kernel.h>
39 #include <sys/msgport.h>
40 #include <sys/protosw.h>
41 #include <sys/socket.h>
42 #include <sys/socketvar.h>
43 #include <sys/socketops.h>
44 #include <sys/thread.h>
45 #include <sys/thread2.h>
46 #include <sys/msgport2.h>
49 #include <net/netmsg2.h>
51 #include <net/netisr.h>
52 #include <net/netmsg.h>
55 * Abort a socket and free it. Called from soabort() only. soabort()
56 * got a ref on the socket which we must free on reply.
59 so_pru_abort(struct socket *so)
61 struct netmsg_pru_abort msg;
63 netmsg_init(&msg.base, so, &curthread->td_msgport,
64 0, so->so_proto->pr_usrreqs->pru_abort);
65 (void)lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
66 sofree(msg.base.nm_so);
70 * Abort a socket and free it, asynchronously. Called from
71 * soaborta() only. soaborta() got a ref on the socket which we must
75 so_pru_aborta(struct socket *so)
77 struct netmsg_pru_abort *msg;
79 msg = kmalloc(sizeof(*msg), M_LWKTMSG, M_WAITOK | M_ZERO);
80 netmsg_init(&msg->base, so, &netisr_afree_free_so_rport,
81 0, so->so_proto->pr_usrreqs->pru_abort);
82 lwkt_sendmsg(so->so_port, &msg->base.lmsg);
86 * Abort a socket and free it. Called from soabort_oncpu() only.
87 * Caller must make sure that the current CPU is inpcb's owner CPU.
90 so_pru_abort_oncpu(struct socket *so)
92 struct netmsg_pru_abort msg;
93 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_abort;
95 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func);
96 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE);
97 msg.base.lmsg.ms_flags |= MSGF_SYNC;
99 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE);
100 sofree(msg.base.nm_so);
104 * WARNING! Synchronous call from user context
107 so_pru_accept_direct(struct socket *so, struct sockaddr **nam)
109 struct netmsg_pru_accept msg;
110 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_accept;
112 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func);
113 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE);
114 msg.base.lmsg.ms_flags |= MSGF_SYNC;
116 func((netmsg_t)&msg);
117 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE);
118 return(msg.base.lmsg.ms_error);
122 so_pru_attach(struct socket *so, int proto, struct pru_attach_info *ai)
124 struct netmsg_pru_attach msg;
127 netmsg_init(&msg.base, so, &curthread->td_msgport,
128 0, so->so_proto->pr_usrreqs->pru_attach);
129 msg.nm_proto = proto;
131 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
136 so_pru_attach_direct(struct socket *so, int proto, struct pru_attach_info *ai)
138 struct netmsg_pru_attach msg;
139 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_attach;
141 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func);
142 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE);
143 msg.base.lmsg.ms_flags |= MSGF_SYNC;
144 msg.nm_proto = proto;
146 func((netmsg_t)&msg);
147 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE);
148 return(msg.base.lmsg.ms_error);
152 * NOTE: If the target port changes the bind operation will deal with it.
155 so_pru_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
157 struct netmsg_pru_bind msg;
160 netmsg_init(&msg.base, so, &curthread->td_msgport,
161 0, so->so_proto->pr_usrreqs->pru_bind);
163 msg.nm_td = td; /* used only for prison_ip() */
164 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
169 so_pru_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
171 struct netmsg_pru_connect msg;
174 netmsg_init(&msg.base, so, &curthread->td_msgport,
175 0, so->so_proto->pr_usrreqs->pru_connect);
180 msg.nm_reconnect = 0;
181 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
186 so_pru_connect2(struct socket *so1, struct socket *so2)
188 struct netmsg_pru_connect2 msg;
191 netmsg_init(&msg.base, so1, &curthread->td_msgport,
192 0, so1->so_proto->pr_usrreqs->pru_connect2);
195 error = lwkt_domsg(so1->so_port, &msg.base.lmsg, 0);
200 * WARNING! Synchronous call from user context. Control function may do
204 so_pru_control_direct(struct socket *so, u_long cmd, caddr_t data,
207 struct netmsg_pru_control msg;
208 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_control;
210 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func);
211 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE);
212 msg.base.lmsg.ms_flags |= MSGF_SYNC;
216 msg.nm_td = curthread;
217 func((netmsg_t)&msg);
218 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE);
219 return(msg.base.lmsg.ms_error);
223 so_pru_detach(struct socket *so)
225 struct netmsg_pru_detach msg;
228 netmsg_init(&msg.base, so, &curthread->td_msgport,
229 0, so->so_proto->pr_usrreqs->pru_detach);
230 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
235 so_pru_disconnect(struct socket *so)
237 struct netmsg_pru_disconnect msg;
240 netmsg_init(&msg.base, so, &curthread->td_msgport,
241 0, so->so_proto->pr_usrreqs->pru_disconnect);
242 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
247 so_pru_listen(struct socket *so, struct thread *td)
249 struct netmsg_pru_listen msg;
252 netmsg_init(&msg.base, so, &curthread->td_msgport,
253 0, so->so_proto->pr_usrreqs->pru_listen);
254 msg.nm_td = td; /* used only for prison_ip() XXX JH */
255 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
260 so_pru_peeraddr(struct socket *so, struct sockaddr **nam)
262 struct netmsg_pru_peeraddr msg;
265 netmsg_init(&msg.base, so, &curthread->td_msgport,
266 0, so->so_proto->pr_usrreqs->pru_peeraddr);
268 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
273 so_pru_rcvd(struct socket *so, int flags)
275 struct netmsg_pru_rcvd msg;
278 netmsg_init(&msg.base, so, &curthread->td_msgport,
279 0, so->so_proto->pr_usrreqs->pru_rcvd);
280 msg.nm_flags = flags;
281 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
286 so_pru_rcvoob(struct socket *so, struct mbuf *m, int flags)
288 struct netmsg_pru_rcvoob msg;
291 netmsg_init(&msg.base, so, &curthread->td_msgport,
292 0, so->so_proto->pr_usrreqs->pru_rcvoob);
294 msg.nm_flags = flags;
295 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
300 * NOTE: If the target port changes the implied connect will deal with it.
303 so_pru_send(struct socket *so, int flags, struct mbuf *m,
304 struct sockaddr *addr, struct mbuf *control, struct thread *td)
306 struct netmsg_pru_send msg;
309 netmsg_init(&msg.base, so, &curthread->td_msgport,
310 0, so->so_proto->pr_usrreqs->pru_send);
311 msg.nm_flags = flags;
314 msg.nm_control = control;
316 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
321 so_pru_sync(struct socket *so)
323 struct netmsg_base msg;
325 netmsg_init(&msg, so, &curthread->td_msgport, 0,
326 netmsg_sync_handler);
327 lwkt_domsg(so->so_port, &msg.lmsg, 0);
331 so_pru_send_async(struct socket *so, int flags, struct mbuf *m,
332 struct sockaddr *addr, struct mbuf *control, struct thread *td)
334 struct netmsg_pru_send *msg;
336 KASSERT(so->so_proto->pr_flags & PR_ASYNC_SEND,
337 ("async pru_send is not supported\n"));
339 msg = &m->m_hdr.mh_sndmsg;
340 netmsg_init(&msg->base, so, &netisr_apanic_rport,
341 0, so->so_proto->pr_usrreqs->pru_send);
342 msg->nm_flags = flags | PRUS_NOREPLY;
345 msg->nm_control = control;
347 lwkt_sendmsg(so->so_port, &msg->base.lmsg);
351 so_pru_sense(struct socket *so, struct stat *sb)
353 struct netmsg_pru_sense msg;
356 netmsg_init(&msg.base, so, &curthread->td_msgport,
357 0, so->so_proto->pr_usrreqs->pru_sense);
359 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
364 so_pru_shutdown(struct socket *so)
366 struct netmsg_pru_shutdown msg;
369 netmsg_init(&msg.base, so, &curthread->td_msgport,
370 0, so->so_proto->pr_usrreqs->pru_shutdown);
371 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
376 so_pru_sockaddr(struct socket *so, struct sockaddr **nam)
378 struct netmsg_pru_sockaddr msg;
381 netmsg_init(&msg.base, so, &curthread->td_msgport,
382 0, so->so_proto->pr_usrreqs->pru_sockaddr);
384 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
389 so_pr_ctloutput(struct socket *so, struct sockopt *sopt)
391 struct netmsg_pr_ctloutput msg;
394 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
395 netmsg_init(&msg.base, so, &curthread->td_msgport,
396 0, so->so_proto->pr_ctloutput);
398 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
403 * Protocol control input, typically via icmp.
405 * If the protocol pr_ctlport is not NULL we call it to figure out the
406 * protocol port. If NULL is returned we can just return, otherwise
407 * we issue a netmsg to call pr_ctlinput in the proper thread.
409 * This must be done synchronously as arg and/or extra may point to
413 so_pru_ctlinput(struct protosw *pr, int cmd, struct sockaddr *arg, void *extra)
415 struct netmsg_pru_ctlinput msg;
418 if (pr->pr_ctlport == NULL)
420 KKASSERT(pr->pr_ctlinput != NULL);
421 port = pr->pr_ctlport(cmd, arg, extra);
424 netmsg_init(&msg.base, NULL, &curthread->td_msgport,
428 msg.nm_extra = extra;
429 lwkt_domsg(port, &msg.base.lmsg, 0);
433 * If we convert all the protosw pr_ functions for all the protocols
434 * to take a message directly, this layer can go away. For the moment
435 * our dispatcher ignores the return value, but since we are handling
436 * the replymsg ourselves we return EASYNC by convention.
440 * Handle a predicate event request. This function is only called once
441 * when the predicate message queueing request is received.
444 netmsg_so_notify(netmsg_t msg)
446 struct lwkt_token *tok;
447 struct signalsockbuf *ssb;
449 ssb = (msg->notify.nm_etype & NM_REVENT) ?
450 &msg->base.nm_so->so_rcv :
451 &msg->base.nm_so->so_snd;
454 * Reply immediately if the event has occured, otherwise queue the
457 * NOTE: Socket can change if this is an accept predicate so cache
460 tok = lwkt_token_pool_lookup(msg->base.nm_so);
462 if (msg->notify.nm_predicate(&msg->notify)) {
464 lwkt_replymsg(&msg->base.lmsg,
465 msg->base.lmsg.ms_error);
467 TAILQ_INSERT_TAIL(&ssb->ssb_kq.ki_mlist, &msg->notify, nm_list);
468 atomic_set_int(&ssb->ssb_flags, SSB_MEVENT);
474 * Called by doio when trying to abort a netmsg_so_notify message.
475 * Unlike the other functions this one is dispatched directly by
476 * the LWKT subsystem, so it takes a lwkt_msg_t as an argument.
478 * The original message, lmsg, is under the control of the caller and
479 * will not be destroyed until we return so we can safely reference it
480 * in our synchronous abort request.
482 * This part of the abort request occurs on the originating cpu which
483 * means we may race the message flags and the original message may
484 * not even have been processed by the target cpu yet.
487 netmsg_so_notify_doabort(lwkt_msg_t lmsg)
489 struct netmsg_so_notify_abort msg;
491 if ((lmsg->ms_flags & (MSGF_DONE | MSGF_REPLY)) == 0) {
492 netmsg_init(&msg.base, NULL, &curthread->td_msgport,
493 0, netmsg_so_notify_abort);
494 msg.nm_notifymsg = (void *)lmsg;
495 lwkt_domsg(lmsg->ms_target_port, &msg.base.lmsg, 0);
500 * Predicate requests can be aborted. This function is only called once
501 * and will interlock against processing/reply races (since such races
502 * occur on the same thread that controls the port where the abort is
505 * This part of the abort request occurs on the target cpu. The message
506 * flags must be tested again in case the test that we did on the
507 * originating cpu raced. Since messages are handled in sequence, the
508 * original message will have already been handled by the loop and either
509 * replied to or queued.
511 * We really only need to interlock with MSGF_REPLY (a bit that is set on
512 * our cpu when we reply). Note that MSGF_DONE is not set until the
513 * reply reaches the originating cpu. Test both bits anyway.
516 netmsg_so_notify_abort(netmsg_t msg)
518 struct netmsg_so_notify_abort *abrtmsg = &msg->notify_abort;
519 struct netmsg_so_notify *nmsg = abrtmsg->nm_notifymsg;
520 struct signalsockbuf *ssb;
523 * The original notify message is not destroyed until after the
524 * abort request is returned, so we can check its state.
526 lwkt_getpooltoken(nmsg->base.nm_so);
527 if ((nmsg->base.lmsg.ms_flags & (MSGF_DONE | MSGF_REPLY)) == 0) {
528 ssb = (nmsg->nm_etype & NM_REVENT) ?
529 &nmsg->base.nm_so->so_rcv :
530 &nmsg->base.nm_so->so_snd;
531 TAILQ_REMOVE(&ssb->ssb_kq.ki_mlist, nmsg, nm_list);
532 lwkt_relpooltoken(nmsg->base.nm_so);
533 lwkt_replymsg(&nmsg->base.lmsg, EINTR);
535 lwkt_relpooltoken(nmsg->base.nm_so);
539 * Reply to the abort message
541 lwkt_replymsg(&abrtmsg->base.lmsg, 0);