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>
46 #include <sys/sysctl.h>
50 #include <net/netmsg2.h>
51 #include <sys/socketvar2.h>
53 #include <net/netisr.h>
54 #include <net/netmsg.h>
56 static int async_rcvd_drop_race = 0;
57 SYSCTL_INT(_kern_ipc, OID_AUTO, async_rcvd_drop_race, CTLFLAG_RW,
58 &async_rcvd_drop_race, 0, "# of asynchronized pru_rcvd msg drop races");
61 * Abort a socket and free it, asynchronously. Called from
62 * soabort_async() only. soabort_async() got a ref on the
63 * socket which we must free on reply.
66 so_pru_abort_async(struct socket *so)
68 struct netmsg_pru_abort *msg;
70 msg = kmalloc(sizeof(*msg), M_LWKTMSG, M_WAITOK | M_ZERO);
71 netmsg_init(&msg->base, so, &netisr_afree_free_so_rport,
72 0, so->so_proto->pr_usrreqs->pru_abort);
73 lwkt_sendmsg(so->so_port, &msg->base.lmsg);
77 * Abort a socket and free it. Called from soabort_oncpu() only.
78 * Caller must make sure that the current CPU is inpcb's owner CPU.
79 * soabort_oncpu() got a ref on the socket which we must free.
82 so_pru_abort_direct(struct socket *so)
84 struct netmsg_pru_abort msg;
85 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_abort;
87 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func);
88 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE);
89 msg.base.lmsg.ms_flags |= MSGF_SYNC;
91 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE);
92 sofree(msg.base.nm_so);
96 so_pru_accept(struct socket *so, struct sockaddr **nam)
98 struct netmsg_pru_accept msg;
100 netmsg_init(&msg.base, so, &curthread->td_msgport,
101 0, so->so_proto->pr_usrreqs->pru_accept);
104 return lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
108 so_pru_attach(struct socket *so, int proto, struct pru_attach_info *ai)
110 struct netmsg_pru_attach msg;
113 netmsg_init(&msg.base, so, &curthread->td_msgport,
114 0, so->so_proto->pr_usrreqs->pru_attach);
115 msg.nm_proto = proto;
117 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
122 so_pru_attach_direct(struct socket *so, int proto, struct pru_attach_info *ai)
124 struct netmsg_pru_attach msg;
125 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_attach;
127 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func);
128 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE);
129 msg.base.lmsg.ms_flags |= MSGF_SYNC;
130 msg.nm_proto = proto;
132 func((netmsg_t)&msg);
133 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE);
134 return(msg.base.lmsg.ms_error);
138 * NOTE: If the target port changes the bind operation will deal with it.
141 so_pru_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
143 struct netmsg_pru_bind msg;
146 netmsg_init(&msg.base, so, &curthread->td_msgport,
147 0, so->so_proto->pr_usrreqs->pru_bind);
149 msg.nm_td = td; /* used only for prison_ip() */
150 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
155 so_pru_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
157 struct netmsg_pru_connect msg;
160 netmsg_init(&msg.base, so, &curthread->td_msgport,
161 0, so->so_proto->pr_usrreqs->pru_connect);
167 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
172 so_pru_connect_async(struct socket *so, struct sockaddr *nam, struct thread *td)
174 struct netmsg_pru_connect *msg;
177 KASSERT(so->so_proto->pr_usrreqs->pru_preconnect != NULL,
178 ("async pru_connect is not supported"));
180 /* NOTE: sockaddr immediately follows netmsg */
181 msg = kmalloc(sizeof(*msg) + nam->sa_len, M_LWKTMSG, M_NOWAIT);
184 * Fail to allocate message w/o waiting;
185 * fallback to synchronized pru_connect.
187 return so_pru_connect(so, nam, td);
190 error = so->so_proto->pr_usrreqs->pru_preconnect(so, nam, td);
192 kfree(msg, M_LWKTMSG);
197 if (td != NULL && (so->so_proto->pr_flags & PR_ACONN_HOLDTD)) {
199 flags |= PRUC_HELDTD;
202 netmsg_init(&msg->base, so, &netisr_afree_rport, 0,
203 so->so_proto->pr_usrreqs->pru_connect);
204 msg->nm_nam = (struct sockaddr *)(msg + 1);
205 memcpy(msg->nm_nam, nam, nam->sa_len);
208 msg->nm_sndflags = 0;
209 msg->nm_flags = flags;
210 lwkt_sendmsg(so->so_port, &msg->base.lmsg);
215 so_pru_connect2(struct socket *so1, struct socket *so2)
217 struct netmsg_pru_connect2 msg;
220 netmsg_init(&msg.base, so1, &curthread->td_msgport,
221 0, so1->so_proto->pr_usrreqs->pru_connect2);
224 error = lwkt_domsg(so1->so_port, &msg.base.lmsg, 0);
229 * WARNING! Synchronous call from user context. Control function may do
233 so_pru_control_direct(struct socket *so, u_long cmd, caddr_t data,
236 struct netmsg_pru_control msg;
237 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_control;
239 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func);
240 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE);
241 msg.base.lmsg.ms_flags |= MSGF_SYNC;
245 msg.nm_td = curthread;
246 func((netmsg_t)&msg);
247 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE);
248 return(msg.base.lmsg.ms_error);
252 so_pru_detach(struct socket *so)
254 struct netmsg_pru_detach msg;
257 netmsg_init(&msg.base, so, &curthread->td_msgport,
258 0, so->so_proto->pr_usrreqs->pru_detach);
259 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
264 so_pru_detach_direct(struct socket *so)
266 struct netmsg_pru_detach msg;
267 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_detach;
269 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func);
270 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE);
271 msg.base.lmsg.ms_flags |= MSGF_SYNC;
272 func((netmsg_t)&msg);
273 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE);
274 return(msg.base.lmsg.ms_error);
278 so_pru_disconnect(struct socket *so)
280 struct netmsg_pru_disconnect msg;
283 netmsg_init(&msg.base, so, &curthread->td_msgport,
284 0, so->so_proto->pr_usrreqs->pru_disconnect);
285 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
290 so_pru_disconnect_direct(struct socket *so)
292 struct netmsg_pru_disconnect msg;
293 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_disconnect;
295 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func);
296 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE);
297 msg.base.lmsg.ms_flags |= MSGF_SYNC;
298 func((netmsg_t)&msg);
299 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE);
303 so_pru_listen(struct socket *so, struct thread *td)
305 struct netmsg_pru_listen msg;
308 netmsg_init(&msg.base, so, &curthread->td_msgport,
309 0, so->so_proto->pr_usrreqs->pru_listen);
310 msg.nm_td = td; /* used only for prison_ip() XXX JH */
312 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
317 so_pru_peeraddr(struct socket *so, struct sockaddr **nam)
319 struct netmsg_pru_peeraddr msg;
322 netmsg_init(&msg.base, so, &curthread->td_msgport,
323 0, so->so_proto->pr_usrreqs->pru_peeraddr);
325 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
330 so_pru_rcvd(struct socket *so, int flags)
332 struct netmsg_pru_rcvd msg;
335 netmsg_init(&msg.base, so, &curthread->td_msgport,
336 0, so->so_proto->pr_usrreqs->pru_rcvd);
337 msg.nm_flags = flags;
338 msg.nm_pru_flags = 0;
339 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
344 so_pru_rcvd_async(struct socket *so)
346 lwkt_msg_t lmsg = &so->so_rcvd_msg.base.lmsg;
348 KASSERT(so->so_proto->pr_flags & PR_ASYNC_RCVD,
349 ("async pru_rcvd is not supported"));
352 * WARNING! Spinlock is a bit dodgy, use hacked up sendmsg
353 * to avoid deadlocking.
355 spin_lock(&so->so_rcvd_spin);
356 if ((so->so_rcvd_msg.nm_pru_flags & PRUR_DEAD) == 0) {
357 if (lmsg->ms_flags & MSGF_DONE) {
358 lwkt_sendmsg_prepare(so->so_port, lmsg);
359 spin_unlock(&so->so_rcvd_spin);
360 lwkt_sendmsg_start(so->so_port, lmsg);
362 spin_unlock(&so->so_rcvd_spin);
365 spin_unlock(&so->so_rcvd_spin);
370 so_pru_rcvoob(struct socket *so, struct mbuf *m, int flags)
372 struct netmsg_pru_rcvoob msg;
375 netmsg_init(&msg.base, so, &curthread->td_msgport,
376 0, so->so_proto->pr_usrreqs->pru_rcvoob);
378 msg.nm_flags = flags;
379 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
384 * NOTE: If the target port changes the implied connect will deal with it.
387 so_pru_send(struct socket *so, int flags, struct mbuf *m,
388 struct sockaddr *addr, struct mbuf *control, struct thread *td)
390 struct netmsg_pru_send msg;
393 netmsg_init(&msg.base, so, &curthread->td_msgport,
394 0, so->so_proto->pr_usrreqs->pru_send);
395 msg.nm_flags = flags;
398 msg.nm_control = control;
400 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
405 so_pru_sync(struct socket *so)
407 struct netmsg_base msg;
409 netmsg_init(&msg, so, &curthread->td_msgport, 0,
410 netmsg_sync_handler);
411 lwkt_domsg(so->so_port, &msg.lmsg, 0);
415 so_pru_send_async(struct socket *so, int flags, struct mbuf *m,
416 struct sockaddr *addr0, struct mbuf *control, struct thread *td)
418 struct netmsg_pru_send *msg;
419 struct sockaddr *addr = NULL;
421 KASSERT(so->so_proto->pr_flags & PR_ASYNC_SEND,
422 ("async pru_send is not supported"));
425 addr = kmalloc(addr0->sa_len, M_SONAME, M_WAITOK | M_NULLOK);
428 * Fail to allocate address; fallback to
429 * synchronized pru_send.
431 so_pru_send(so, flags, m, addr0, control, td);
434 memcpy(addr, addr0, addr0->sa_len);
435 flags |= PRUS_FREEADDR;
437 flags |= PRUS_NOREPLY;
439 if (td != NULL && (so->so_proto->pr_flags & PR_ASEND_HOLDTD)) {
441 flags |= PRUS_HELDTD;
444 msg = &m->m_hdr.mh_sndmsg;
445 netmsg_init(&msg->base, so, &netisr_apanic_rport,
446 0, so->so_proto->pr_usrreqs->pru_send);
447 msg->nm_flags = flags;
450 msg->nm_control = control;
452 lwkt_sendmsg(so->so_port, &msg->base.lmsg);
456 so_pru_sense(struct socket *so, struct stat *sb)
458 struct netmsg_pru_sense msg;
461 netmsg_init(&msg.base, so, &curthread->td_msgport,
462 0, so->so_proto->pr_usrreqs->pru_sense);
464 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
469 so_pru_shutdown(struct socket *so)
471 struct netmsg_pru_shutdown msg;
474 netmsg_init(&msg.base, so, &curthread->td_msgport,
475 0, so->so_proto->pr_usrreqs->pru_shutdown);
476 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
481 so_pru_sockaddr(struct socket *so, struct sockaddr **nam)
483 struct netmsg_pru_sockaddr msg;
486 netmsg_init(&msg.base, so, &curthread->td_msgport,
487 0, so->so_proto->pr_usrreqs->pru_sockaddr);
489 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
494 so_pr_ctloutput(struct socket *so, struct sockopt *sopt)
496 struct netmsg_pr_ctloutput msg;
499 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
500 netmsg_init(&msg.base, so, &curthread->td_msgport,
501 0, so->so_proto->pr_ctloutput);
503 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
508 so_pr_ctlport(struct protosw *pr, int cmd, struct sockaddr *arg,
509 void *extra, int *cpuid)
511 if (pr->pr_ctlport == NULL)
513 KKASSERT(pr->pr_ctlinput != NULL);
515 return pr->pr_ctlport(cmd, arg, extra, cpuid);
519 * Protocol control input, typically via icmp.
521 * If the protocol pr_ctlport is not NULL we call it to figure out the
522 * protocol port. If NULL is returned we can just return, otherwise
523 * we issue a netmsg to call pr_ctlinput in the proper thread.
525 * This must be done synchronously as arg and/or extra may point to
529 so_pr_ctlinput(struct protosw *pr, int cmd, struct sockaddr *arg, void *extra)
531 struct netmsg_pr_ctlinput msg;
535 port = so_pr_ctlport(pr, cmd, arg, extra, &cpuid);
538 netmsg_init(&msg.base, NULL, &curthread->td_msgport,
543 msg.nm_extra = extra;
544 lwkt_domsg(port, &msg.base.lmsg, 0);
548 so_pr_ctlinput_direct(struct protosw *pr, int cmd, struct sockaddr *arg,
551 struct netmsg_pr_ctlinput msg;
556 port = so_pr_ctlport(pr, cmd, arg, extra, &cpuid);
559 if (cpuid != ncpus && cpuid != mycpuid)
562 func = pr->pr_ctlinput;
563 netmsg_init(&msg.base, NULL, &netisr_adone_rport, 0, func);
564 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE);
565 msg.base.lmsg.ms_flags |= MSGF_SYNC;
569 msg.nm_extra = extra;
570 func((netmsg_t)&msg);
571 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE);
575 * If we convert all the protosw pr_ functions for all the protocols
576 * to take a message directly, this layer can go away. For the moment
577 * our dispatcher ignores the return value, but since we are handling
578 * the replymsg ourselves we return EASYNC by convention.
582 * Handle a predicate event request. This function is only called once
583 * when the predicate message queueing request is received.
586 netmsg_so_notify(netmsg_t msg)
588 struct lwkt_token *tok;
589 struct signalsockbuf *ssb;
591 ssb = (msg->notify.nm_etype & NM_REVENT) ?
592 &msg->base.nm_so->so_rcv :
593 &msg->base.nm_so->so_snd;
596 * Reply immediately if the event has occured, otherwise queue the
599 * NOTE: Socket can change if this is an accept predicate so cache
602 tok = lwkt_token_pool_lookup(msg->base.nm_so);
604 atomic_set_int(&ssb->ssb_flags, SSB_MEVENT);
605 if (msg->notify.nm_predicate(&msg->notify)) {
606 if (TAILQ_EMPTY(&ssb->ssb_kq.ki_mlist))
607 atomic_clear_int(&ssb->ssb_flags, SSB_MEVENT);
609 lwkt_replymsg(&msg->base.lmsg,
610 msg->base.lmsg.ms_error);
612 TAILQ_INSERT_TAIL(&ssb->ssb_kq.ki_mlist, &msg->notify, nm_list);
615 * If predict ever blocks, 'tok' will be released, so
616 * SSB_MEVENT set beforehand could have been cleared
617 * when we reach here. In case that happens, we set
618 * SSB_MEVENT again, after the notify has been queued.
620 atomic_set_int(&ssb->ssb_flags, SSB_MEVENT);
626 * Called by doio when trying to abort a netmsg_so_notify message.
627 * Unlike the other functions this one is dispatched directly by
628 * the LWKT subsystem, so it takes a lwkt_msg_t as an argument.
630 * The original message, lmsg, is under the control of the caller and
631 * will not be destroyed until we return so we can safely reference it
632 * in our synchronous abort request.
634 * This part of the abort request occurs on the originating cpu which
635 * means we may race the message flags and the original message may
636 * not even have been processed by the target cpu yet.
639 netmsg_so_notify_doabort(lwkt_msg_t lmsg)
641 struct netmsg_so_notify_abort msg;
643 if ((lmsg->ms_flags & (MSGF_DONE | MSGF_REPLY)) == 0) {
644 const struct netmsg_base *nmsg =
645 (const struct netmsg_base *)lmsg;
647 netmsg_init(&msg.base, nmsg->nm_so, &curthread->td_msgport,
648 0, netmsg_so_notify_abort);
649 msg.nm_notifymsg = (void *)lmsg;
650 lwkt_domsg(lmsg->ms_target_port, &msg.base.lmsg, 0);
655 * Predicate requests can be aborted. This function is only called once
656 * and will interlock against processing/reply races (since such races
657 * occur on the same thread that controls the port where the abort is
660 * This part of the abort request occurs on the target cpu. The message
661 * flags must be tested again in case the test that we did on the
662 * originating cpu raced. Since messages are handled in sequence, the
663 * original message will have already been handled by the loop and either
664 * replied to or queued.
666 * We really only need to interlock with MSGF_REPLY (a bit that is set on
667 * our cpu when we reply). Note that MSGF_DONE is not set until the
668 * reply reaches the originating cpu. Test both bits anyway.
671 netmsg_so_notify_abort(netmsg_t msg)
673 struct netmsg_so_notify_abort *abrtmsg = &msg->notify_abort;
674 struct netmsg_so_notify *nmsg = abrtmsg->nm_notifymsg;
675 struct signalsockbuf *ssb;
678 * The original notify message is not destroyed until after the
679 * abort request is returned, so we can check its state.
681 lwkt_getpooltoken(nmsg->base.nm_so);
682 if ((nmsg->base.lmsg.ms_flags & (MSGF_DONE | MSGF_REPLY)) == 0) {
683 ssb = (nmsg->nm_etype & NM_REVENT) ?
684 &nmsg->base.nm_so->so_rcv :
685 &nmsg->base.nm_so->so_snd;
686 TAILQ_REMOVE(&ssb->ssb_kq.ki_mlist, nmsg, nm_list);
687 lwkt_relpooltoken(nmsg->base.nm_so);
688 lwkt_replymsg(&nmsg->base.lmsg, EINTR);
690 lwkt_relpooltoken(nmsg->base.nm_so);
694 * Reply to the abort message
696 lwkt_replymsg(&abrtmsg->base.lmsg, 0);
700 so_async_rcvd_reply(struct socket *so)
703 * Spinlock safe, reply runs to degenerate lwkt_null_replyport()
705 spin_lock(&so->so_rcvd_spin);
706 lwkt_replymsg(&so->so_rcvd_msg.base.lmsg, 0);
707 spin_unlock(&so->so_rcvd_spin);
711 so_async_rcvd_drop(struct socket *so)
713 lwkt_msg_t lmsg = &so->so_rcvd_msg.base.lmsg;
716 * Spinlock safe, drop runs to degenerate lwkt_spin_dropmsg()
718 spin_lock(&so->so_rcvd_spin);
719 so->so_rcvd_msg.nm_pru_flags |= PRUR_DEAD;
722 if ((lmsg->ms_flags & MSGF_DONE) == 0) {
723 ++async_rcvd_drop_race;
724 ssleep(so, &so->so_rcvd_spin, 0, "soadrop", 1);
727 spin_unlock(&so->so_rcvd_spin);