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
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/kernel.h>
37 #include <sys/msgport.h>
38 #include <sys/protosw.h>
39 #include <sys/socket.h>
40 #include <sys/socketvar.h>
41 #include <sys/socketops.h>
42 #include <sys/thread.h>
43 #include <sys/thread2.h>
44 #include <sys/msgport2.h>
45 #include <sys/spinlock2.h>
49 #include <net/netmsg2.h>
50 #include <sys/socketvar2.h>
52 #include <net/netisr.h>
53 #include <net/netmsg.h>
56 * Abort a socket and free it. Called from soabort() only. soabort()
57 * got a ref on the socket which we must free on reply.
60 so_pru_abort(struct socket *so)
62 struct netmsg_pru_abort msg;
64 netmsg_init(&msg.base, so, &curthread->td_msgport,
65 0, so->so_proto->pr_usrreqs->pru_abort);
66 (void)lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
67 sofree(msg.base.nm_so);
71 * Abort a socket and free it, asynchronously. Called from
72 * soaborta() only. soaborta() got a ref on the socket which we must
76 so_pru_aborta(struct socket *so)
78 struct netmsg_pru_abort *msg;
80 msg = kmalloc(sizeof(*msg), M_LWKTMSG, M_WAITOK | M_ZERO);
81 netmsg_init(&msg->base, so, &netisr_afree_free_so_rport,
82 0, so->so_proto->pr_usrreqs->pru_abort);
83 lwkt_sendmsg(so->so_port, &msg->base.lmsg);
87 * Abort a socket and free it. Called from soabort_oncpu() only.
88 * Caller must make sure that the current CPU is inpcb's owner CPU.
91 so_pru_abort_oncpu(struct socket *so)
93 struct netmsg_pru_abort msg;
94 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_abort;
96 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func);
97 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE);
98 msg.base.lmsg.ms_flags |= MSGF_SYNC;
100 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE);
101 sofree(msg.base.nm_so);
105 so_pru_accept(struct socket *so, struct sockaddr **nam)
107 struct netmsg_pru_accept msg;
109 netmsg_init(&msg.base, so, &curthread->td_msgport,
110 0, so->so_proto->pr_usrreqs->pru_accept);
113 return lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
117 so_pru_attach(struct socket *so, int proto, struct pru_attach_info *ai)
119 struct netmsg_pru_attach msg;
122 netmsg_init(&msg.base, so, &curthread->td_msgport,
123 0, so->so_proto->pr_usrreqs->pru_attach);
124 msg.nm_proto = proto;
126 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
131 so_pru_attach_direct(struct socket *so, int proto, struct pru_attach_info *ai)
133 struct netmsg_pru_attach msg;
134 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_attach;
136 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func);
137 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE);
138 msg.base.lmsg.ms_flags |= MSGF_SYNC;
139 msg.nm_proto = proto;
141 func((netmsg_t)&msg);
142 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE);
143 return(msg.base.lmsg.ms_error);
147 * NOTE: If the target port changes the bind operation will deal with it.
150 so_pru_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
152 struct netmsg_pru_bind msg;
155 netmsg_init(&msg.base, so, &curthread->td_msgport,
156 0, so->so_proto->pr_usrreqs->pru_bind);
158 msg.nm_td = td; /* used only for prison_ip() */
159 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
164 so_pru_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
166 struct netmsg_pru_connect msg;
169 netmsg_init(&msg.base, so, &curthread->td_msgport,
170 0, so->so_proto->pr_usrreqs->pru_connect);
175 msg.nm_reconnect = 0;
176 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
181 so_pru_connect2(struct socket *so1, struct socket *so2)
183 struct netmsg_pru_connect2 msg;
186 netmsg_init(&msg.base, so1, &curthread->td_msgport,
187 0, so1->so_proto->pr_usrreqs->pru_connect2);
190 error = lwkt_domsg(so1->so_port, &msg.base.lmsg, 0);
195 * WARNING! Synchronous call from user context. Control function may do
199 so_pru_control_direct(struct socket *so, u_long cmd, caddr_t data,
202 struct netmsg_pru_control msg;
203 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_control;
205 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func);
206 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE);
207 msg.base.lmsg.ms_flags |= MSGF_SYNC;
211 msg.nm_td = curthread;
212 func((netmsg_t)&msg);
213 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE);
214 return(msg.base.lmsg.ms_error);
218 so_pru_detach(struct socket *so)
220 struct netmsg_pru_detach msg;
223 netmsg_init(&msg.base, so, &curthread->td_msgport,
224 0, so->so_proto->pr_usrreqs->pru_detach);
225 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
230 so_pru_detach_direct(struct socket *so)
232 struct netmsg_pru_detach msg;
233 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_detach;
235 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func);
236 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE);
237 msg.base.lmsg.ms_flags |= MSGF_SYNC;
238 func((netmsg_t)&msg);
239 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE);
243 so_pru_disconnect(struct socket *so)
245 struct netmsg_pru_disconnect msg;
248 netmsg_init(&msg.base, so, &curthread->td_msgport,
249 0, so->so_proto->pr_usrreqs->pru_disconnect);
250 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
255 so_pru_disconnect_direct(struct socket *so)
257 struct netmsg_pru_disconnect msg;
258 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_disconnect;
260 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func);
261 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE);
262 msg.base.lmsg.ms_flags |= MSGF_SYNC;
263 func((netmsg_t)&msg);
264 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE);
268 so_pru_listen(struct socket *so, struct thread *td)
270 struct netmsg_pru_listen msg;
273 netmsg_init(&msg.base, so, &curthread->td_msgport,
274 0, so->so_proto->pr_usrreqs->pru_listen);
275 msg.nm_td = td; /* used only for prison_ip() XXX JH */
276 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
281 so_pru_peeraddr(struct socket *so, struct sockaddr **nam)
283 struct netmsg_pru_peeraddr msg;
286 netmsg_init(&msg.base, so, &curthread->td_msgport,
287 0, so->so_proto->pr_usrreqs->pru_peeraddr);
289 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
294 so_pru_rcvd(struct socket *so, int flags)
296 struct netmsg_pru_rcvd msg;
299 netmsg_init(&msg.base, so, &curthread->td_msgport,
300 0, so->so_proto->pr_usrreqs->pru_rcvd);
301 msg.nm_flags = flags;
302 msg.nm_pru_flags = 0;
303 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
308 so_pru_rcvd_async(struct socket *so)
310 lwkt_msg_t lmsg = &so->so_rcvd_msg.base.lmsg;
312 KASSERT(so->so_proto->pr_flags & PR_ASYNC_RCVD,
313 ("async pru_rcvd is not supported"));
316 * WARNING! Spinlock is a bit dodgy, use hacked up sendmsg
317 * to avoid deadlocking.
319 spin_lock(&so->so_rcvd_spin);
320 if ((so->so_rcvd_msg.nm_pru_flags & PRUR_DEAD) == 0) {
321 if (lmsg->ms_flags & MSGF_DONE) {
323 lwkt_sendmsg_stage1(so->so_port, lmsg);
324 spin_unlock(&so->so_rcvd_spin);
325 lwkt_sendmsg_stage2(so->so_port, lmsg);
327 spin_unlock(&so->so_rcvd_spin);
330 static int deadlog = 0;
333 kprintf("async rcvd is dead\n");
336 spin_unlock(&so->so_rcvd_spin);
341 so_pru_rcvoob(struct socket *so, struct mbuf *m, int flags)
343 struct netmsg_pru_rcvoob msg;
346 netmsg_init(&msg.base, so, &curthread->td_msgport,
347 0, so->so_proto->pr_usrreqs->pru_rcvoob);
349 msg.nm_flags = flags;
350 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
355 * NOTE: If the target port changes the implied connect will deal with it.
358 so_pru_send(struct socket *so, int flags, struct mbuf *m,
359 struct sockaddr *addr, struct mbuf *control, struct thread *td)
361 struct netmsg_pru_send msg;
364 netmsg_init(&msg.base, so, &curthread->td_msgport,
365 0, so->so_proto->pr_usrreqs->pru_send);
366 msg.nm_flags = flags;
369 msg.nm_control = control;
371 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
376 so_pru_sync(struct socket *so)
378 struct netmsg_base msg;
380 netmsg_init(&msg, so, &curthread->td_msgport, 0,
381 netmsg_sync_handler);
382 lwkt_domsg(so->so_port, &msg.lmsg, 0);
386 so_pru_send_async(struct socket *so, int flags, struct mbuf *m,
387 struct sockaddr *addr0, struct mbuf *control, struct thread *td)
389 struct netmsg_pru_send *msg;
390 struct sockaddr *addr = NULL;
392 KASSERT(so->so_proto->pr_flags & PR_ASYNC_SEND,
393 ("async pru_send is not supported"));
395 flags |= PRUS_NOREPLY;
397 addr = kmalloc(addr0->sa_len, M_SONAME, M_WAITOK);
398 memcpy(addr, addr0, addr0->sa_len);
399 flags |= PRUS_FREEADDR;
402 msg = &m->m_hdr.mh_sndmsg;
403 netmsg_init(&msg->base, so, &netisr_apanic_rport,
404 0, so->so_proto->pr_usrreqs->pru_send);
405 msg->nm_flags = flags;
408 msg->nm_control = control;
410 lwkt_sendmsg(so->so_port, &msg->base.lmsg);
414 so_pru_sense(struct socket *so, struct stat *sb)
416 struct netmsg_pru_sense msg;
419 netmsg_init(&msg.base, so, &curthread->td_msgport,
420 0, so->so_proto->pr_usrreqs->pru_sense);
422 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
427 so_pru_shutdown(struct socket *so)
429 struct netmsg_pru_shutdown msg;
432 netmsg_init(&msg.base, so, &curthread->td_msgport,
433 0, so->so_proto->pr_usrreqs->pru_shutdown);
434 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
439 so_pru_sockaddr(struct socket *so, struct sockaddr **nam)
441 struct netmsg_pru_sockaddr msg;
444 netmsg_init(&msg.base, so, &curthread->td_msgport,
445 0, so->so_proto->pr_usrreqs->pru_sockaddr);
447 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
452 so_pr_ctloutput(struct socket *so, struct sockopt *sopt)
454 struct netmsg_pr_ctloutput msg;
457 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
458 netmsg_init(&msg.base, so, &curthread->td_msgport,
459 0, so->so_proto->pr_ctloutput);
461 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
466 * Protocol control input, typically via icmp.
468 * If the protocol pr_ctlport is not NULL we call it to figure out the
469 * protocol port. If NULL is returned we can just return, otherwise
470 * we issue a netmsg to call pr_ctlinput in the proper thread.
472 * This must be done synchronously as arg and/or extra may point to
476 so_pru_ctlinput(struct protosw *pr, int cmd, struct sockaddr *arg, void *extra)
478 struct netmsg_pru_ctlinput msg;
481 if (pr->pr_ctlport == NULL)
483 KKASSERT(pr->pr_ctlinput != NULL);
484 port = pr->pr_ctlport(cmd, arg, extra);
487 netmsg_init(&msg.base, NULL, &curthread->td_msgport,
491 msg.nm_extra = extra;
492 lwkt_domsg(port, &msg.base.lmsg, 0);
496 * If we convert all the protosw pr_ functions for all the protocols
497 * to take a message directly, this layer can go away. For the moment
498 * our dispatcher ignores the return value, but since we are handling
499 * the replymsg ourselves we return EASYNC by convention.
503 * Handle a predicate event request. This function is only called once
504 * when the predicate message queueing request is received.
507 netmsg_so_notify(netmsg_t msg)
509 struct lwkt_token *tok;
510 struct signalsockbuf *ssb;
512 ssb = (msg->notify.nm_etype & NM_REVENT) ?
513 &msg->base.nm_so->so_rcv :
514 &msg->base.nm_so->so_snd;
517 * Reply immediately if the event has occured, otherwise queue the
520 * NOTE: Socket can change if this is an accept predicate so cache
523 tok = lwkt_token_pool_lookup(msg->base.nm_so);
525 if (msg->notify.nm_predicate(&msg->notify)) {
527 lwkt_replymsg(&msg->base.lmsg,
528 msg->base.lmsg.ms_error);
530 TAILQ_INSERT_TAIL(&ssb->ssb_kq.ki_mlist, &msg->notify, nm_list);
531 atomic_set_int(&ssb->ssb_flags, SSB_MEVENT);
537 * Called by doio when trying to abort a netmsg_so_notify message.
538 * Unlike the other functions this one is dispatched directly by
539 * the LWKT subsystem, so it takes a lwkt_msg_t as an argument.
541 * The original message, lmsg, is under the control of the caller and
542 * will not be destroyed until we return so we can safely reference it
543 * in our synchronous abort request.
545 * This part of the abort request occurs on the originating cpu which
546 * means we may race the message flags and the original message may
547 * not even have been processed by the target cpu yet.
550 netmsg_so_notify_doabort(lwkt_msg_t lmsg)
552 struct netmsg_so_notify_abort msg;
554 if ((lmsg->ms_flags & (MSGF_DONE | MSGF_REPLY)) == 0) {
555 netmsg_init(&msg.base, NULL, &curthread->td_msgport,
556 0, netmsg_so_notify_abort);
557 msg.nm_notifymsg = (void *)lmsg;
558 lwkt_domsg(lmsg->ms_target_port, &msg.base.lmsg, 0);
563 * Predicate requests can be aborted. This function is only called once
564 * and will interlock against processing/reply races (since such races
565 * occur on the same thread that controls the port where the abort is
568 * This part of the abort request occurs on the target cpu. The message
569 * flags must be tested again in case the test that we did on the
570 * originating cpu raced. Since messages are handled in sequence, the
571 * original message will have already been handled by the loop and either
572 * replied to or queued.
574 * We really only need to interlock with MSGF_REPLY (a bit that is set on
575 * our cpu when we reply). Note that MSGF_DONE is not set until the
576 * reply reaches the originating cpu. Test both bits anyway.
579 netmsg_so_notify_abort(netmsg_t msg)
581 struct netmsg_so_notify_abort *abrtmsg = &msg->notify_abort;
582 struct netmsg_so_notify *nmsg = abrtmsg->nm_notifymsg;
583 struct signalsockbuf *ssb;
586 * The original notify message is not destroyed until after the
587 * abort request is returned, so we can check its state.
589 lwkt_getpooltoken(nmsg->base.nm_so);
590 if ((nmsg->base.lmsg.ms_flags & (MSGF_DONE | MSGF_REPLY)) == 0) {
591 ssb = (nmsg->nm_etype & NM_REVENT) ?
592 &nmsg->base.nm_so->so_rcv :
593 &nmsg->base.nm_so->so_snd;
594 TAILQ_REMOVE(&ssb->ssb_kq.ki_mlist, nmsg, nm_list);
595 lwkt_relpooltoken(nmsg->base.nm_so);
596 lwkt_replymsg(&nmsg->base.lmsg, EINTR);
598 lwkt_relpooltoken(nmsg->base.nm_so);
602 * Reply to the abort message
604 lwkt_replymsg(&abrtmsg->base.lmsg, 0);
608 so_async_rcvd_reply(struct socket *so)
611 * Spinlock safe, reply runs to degenerate lwkt_null_replyport()
613 spin_lock(&so->so_rcvd_spin);
614 lwkt_replymsg(&so->so_rcvd_msg.base.lmsg, 0);
615 spin_unlock(&so->so_rcvd_spin);
620 so_async_rcvd_drop(struct socket *so)
622 lwkt_msg_t lmsg = &so->so_rcvd_msg.base.lmsg;
626 * Spinlock safe, reply runs to degenerate lwkt_spin_dropmsg()
628 spin_lock(&so->so_rcvd_spin);
629 if (lwkt_dropmsg(lmsg) == 0)
631 so->so_rcvd_msg.nm_pru_flags |= PRUR_DEAD;
632 spin_unlock(&so->so_rcvd_spin);
633 if ((lmsg->ms_flags & MSGF_DONE) == 0) {
634 kprintf("Warning: tcp: so_async_rcvd_drop() raced message\n");
635 tsleep(so, 0, "soadrop", 1);