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. Called from soabort() only. soabort()
62 * got a ref on the socket which we must free on reply.
65 so_pru_abort(struct socket *so)
67 struct netmsg_pru_abort msg;
69 netmsg_init(&msg.base, so, &curthread->td_msgport,
70 0, so->so_proto->pr_usrreqs->pru_abort);
71 (void)lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
72 sofree(msg.base.nm_so);
76 * Abort a socket and free it, asynchronously. Called from
77 * soaborta() only. soaborta() got a ref on the socket which we must
81 so_pru_aborta(struct socket *so)
83 struct netmsg_pru_abort *msg;
85 msg = kmalloc(sizeof(*msg), M_LWKTMSG, M_WAITOK | M_ZERO);
86 netmsg_init(&msg->base, so, &netisr_afree_free_so_rport,
87 0, so->so_proto->pr_usrreqs->pru_abort);
88 lwkt_sendmsg(so->so_port, &msg->base.lmsg);
92 * Abort a socket and free it. Called from soabort_oncpu() only.
93 * Caller must make sure that the current CPU is inpcb's owner CPU.
96 so_pru_abort_oncpu(struct socket *so)
98 struct netmsg_pru_abort msg;
99 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_abort;
101 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func);
102 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE);
103 msg.base.lmsg.ms_flags |= MSGF_SYNC;
104 func((netmsg_t)&msg);
105 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE);
106 sofree(msg.base.nm_so);
110 so_pru_accept(struct socket *so, struct sockaddr **nam)
112 struct netmsg_pru_accept msg;
114 netmsg_init(&msg.base, so, &curthread->td_msgport,
115 0, so->so_proto->pr_usrreqs->pru_accept);
118 return lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
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_detach_direct(struct socket *so)
237 struct netmsg_pru_detach msg;
238 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_detach;
240 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func);
241 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE);
242 msg.base.lmsg.ms_flags |= MSGF_SYNC;
243 func((netmsg_t)&msg);
244 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE);
248 so_pru_disconnect(struct socket *so)
250 struct netmsg_pru_disconnect msg;
253 netmsg_init(&msg.base, so, &curthread->td_msgport,
254 0, so->so_proto->pr_usrreqs->pru_disconnect);
255 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
260 so_pru_disconnect_direct(struct socket *so)
262 struct netmsg_pru_disconnect msg;
263 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_disconnect;
265 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func);
266 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE);
267 msg.base.lmsg.ms_flags |= MSGF_SYNC;
268 func((netmsg_t)&msg);
269 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE);
273 so_pru_listen(struct socket *so, struct thread *td)
275 struct netmsg_pru_listen msg;
278 netmsg_init(&msg.base, so, &curthread->td_msgport,
279 0, so->so_proto->pr_usrreqs->pru_listen);
280 msg.nm_td = td; /* used only for prison_ip() XXX JH */
281 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
286 so_pru_peeraddr(struct socket *so, struct sockaddr **nam)
288 struct netmsg_pru_peeraddr msg;
291 netmsg_init(&msg.base, so, &curthread->td_msgport,
292 0, so->so_proto->pr_usrreqs->pru_peeraddr);
294 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
299 so_pru_rcvd(struct socket *so, int flags)
301 struct netmsg_pru_rcvd msg;
304 netmsg_init(&msg.base, so, &curthread->td_msgport,
305 0, so->so_proto->pr_usrreqs->pru_rcvd);
306 msg.nm_flags = flags;
307 msg.nm_pru_flags = 0;
308 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
313 so_pru_rcvd_async(struct socket *so)
315 lwkt_msg_t lmsg = &so->so_rcvd_msg.base.lmsg;
317 KASSERT(so->so_proto->pr_flags & PR_ASYNC_RCVD,
318 ("async pru_rcvd is not supported"));
321 * WARNING! Spinlock is a bit dodgy, use hacked up sendmsg
322 * to avoid deadlocking.
324 spin_lock(&so->so_rcvd_spin);
325 if ((so->so_rcvd_msg.nm_pru_flags & PRUR_DEAD) == 0) {
326 if (lmsg->ms_flags & MSGF_DONE) {
327 lwkt_sendmsg_stage1(so->so_port, lmsg);
328 spin_unlock(&so->so_rcvd_spin);
329 lwkt_sendmsg_stage2(so->so_port, lmsg);
331 spin_unlock(&so->so_rcvd_spin);
334 spin_unlock(&so->so_rcvd_spin);
339 so_pru_rcvoob(struct socket *so, struct mbuf *m, int flags)
341 struct netmsg_pru_rcvoob msg;
344 netmsg_init(&msg.base, so, &curthread->td_msgport,
345 0, so->so_proto->pr_usrreqs->pru_rcvoob);
347 msg.nm_flags = flags;
348 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
353 * NOTE: If the target port changes the implied connect will deal with it.
356 so_pru_send(struct socket *so, int flags, struct mbuf *m,
357 struct sockaddr *addr, struct mbuf *control, struct thread *td)
359 struct netmsg_pru_send msg;
362 netmsg_init(&msg.base, so, &curthread->td_msgport,
363 0, so->so_proto->pr_usrreqs->pru_send);
364 msg.nm_flags = flags;
367 msg.nm_control = control;
369 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
374 so_pru_sync(struct socket *so)
376 struct netmsg_base msg;
378 netmsg_init(&msg, so, &curthread->td_msgport, 0,
379 netmsg_sync_handler);
380 lwkt_domsg(so->so_port, &msg.lmsg, 0);
384 so_pru_send_async(struct socket *so, int flags, struct mbuf *m,
385 struct sockaddr *addr0, struct mbuf *control, struct thread *td)
387 struct netmsg_pru_send *msg;
388 struct sockaddr *addr = NULL;
390 KASSERT(so->so_proto->pr_flags & PR_ASYNC_SEND,
391 ("async pru_send is not supported"));
393 flags |= PRUS_NOREPLY;
395 addr = kmalloc(addr0->sa_len, M_SONAME, M_WAITOK);
396 memcpy(addr, addr0, addr0->sa_len);
397 flags |= PRUS_FREEADDR;
400 msg = &m->m_hdr.mh_sndmsg;
401 netmsg_init(&msg->base, so, &netisr_apanic_rport,
402 0, so->so_proto->pr_usrreqs->pru_send);
403 msg->nm_flags = flags;
406 msg->nm_control = control;
408 lwkt_sendmsg(so->so_port, &msg->base.lmsg);
412 so_pru_sense(struct socket *so, struct stat *sb)
414 struct netmsg_pru_sense msg;
417 netmsg_init(&msg.base, so, &curthread->td_msgport,
418 0, so->so_proto->pr_usrreqs->pru_sense);
420 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
425 so_pru_shutdown(struct socket *so)
427 struct netmsg_pru_shutdown msg;
430 netmsg_init(&msg.base, so, &curthread->td_msgport,
431 0, so->so_proto->pr_usrreqs->pru_shutdown);
432 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
437 so_pru_sockaddr(struct socket *so, struct sockaddr **nam)
439 struct netmsg_pru_sockaddr msg;
442 netmsg_init(&msg.base, so, &curthread->td_msgport,
443 0, so->so_proto->pr_usrreqs->pru_sockaddr);
445 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
450 so_pr_ctloutput(struct socket *so, struct sockopt *sopt)
452 struct netmsg_pr_ctloutput msg;
455 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
456 netmsg_init(&msg.base, so, &curthread->td_msgport,
457 0, so->so_proto->pr_ctloutput);
459 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
464 * Protocol control input, typically via icmp.
466 * If the protocol pr_ctlport is not NULL we call it to figure out the
467 * protocol port. If NULL is returned we can just return, otherwise
468 * we issue a netmsg to call pr_ctlinput in the proper thread.
470 * This must be done synchronously as arg and/or extra may point to
474 so_pru_ctlinput(struct protosw *pr, int cmd, struct sockaddr *arg, void *extra)
476 struct netmsg_pru_ctlinput msg;
479 if (pr->pr_ctlport == NULL)
481 KKASSERT(pr->pr_ctlinput != NULL);
482 port = pr->pr_ctlport(cmd, arg, extra);
485 netmsg_init(&msg.base, NULL, &curthread->td_msgport,
489 msg.nm_extra = extra;
490 lwkt_domsg(port, &msg.base.lmsg, 0);
494 * If we convert all the protosw pr_ functions for all the protocols
495 * to take a message directly, this layer can go away. For the moment
496 * our dispatcher ignores the return value, but since we are handling
497 * the replymsg ourselves we return EASYNC by convention.
501 * Handle a predicate event request. This function is only called once
502 * when the predicate message queueing request is received.
505 netmsg_so_notify(netmsg_t msg)
507 struct lwkt_token *tok;
508 struct signalsockbuf *ssb;
510 ssb = (msg->notify.nm_etype & NM_REVENT) ?
511 &msg->base.nm_so->so_rcv :
512 &msg->base.nm_so->so_snd;
515 * Reply immediately if the event has occured, otherwise queue the
518 * NOTE: Socket can change if this is an accept predicate so cache
521 tok = lwkt_token_pool_lookup(msg->base.nm_so);
523 if (msg->notify.nm_predicate(&msg->notify)) {
525 lwkt_replymsg(&msg->base.lmsg,
526 msg->base.lmsg.ms_error);
528 TAILQ_INSERT_TAIL(&ssb->ssb_kq.ki_mlist, &msg->notify, nm_list);
529 atomic_set_int(&ssb->ssb_flags, SSB_MEVENT);
535 * Called by doio when trying to abort a netmsg_so_notify message.
536 * Unlike the other functions this one is dispatched directly by
537 * the LWKT subsystem, so it takes a lwkt_msg_t as an argument.
539 * The original message, lmsg, is under the control of the caller and
540 * will not be destroyed until we return so we can safely reference it
541 * in our synchronous abort request.
543 * This part of the abort request occurs on the originating cpu which
544 * means we may race the message flags and the original message may
545 * not even have been processed by the target cpu yet.
548 netmsg_so_notify_doabort(lwkt_msg_t lmsg)
550 struct netmsg_so_notify_abort msg;
552 if ((lmsg->ms_flags & (MSGF_DONE | MSGF_REPLY)) == 0) {
553 netmsg_init(&msg.base, NULL, &curthread->td_msgport,
554 0, netmsg_so_notify_abort);
555 msg.nm_notifymsg = (void *)lmsg;
556 lwkt_domsg(lmsg->ms_target_port, &msg.base.lmsg, 0);
561 * Predicate requests can be aborted. This function is only called once
562 * and will interlock against processing/reply races (since such races
563 * occur on the same thread that controls the port where the abort is
566 * This part of the abort request occurs on the target cpu. The message
567 * flags must be tested again in case the test that we did on the
568 * originating cpu raced. Since messages are handled in sequence, the
569 * original message will have already been handled by the loop and either
570 * replied to or queued.
572 * We really only need to interlock with MSGF_REPLY (a bit that is set on
573 * our cpu when we reply). Note that MSGF_DONE is not set until the
574 * reply reaches the originating cpu. Test both bits anyway.
577 netmsg_so_notify_abort(netmsg_t msg)
579 struct netmsg_so_notify_abort *abrtmsg = &msg->notify_abort;
580 struct netmsg_so_notify *nmsg = abrtmsg->nm_notifymsg;
581 struct signalsockbuf *ssb;
584 * The original notify message is not destroyed until after the
585 * abort request is returned, so we can check its state.
587 lwkt_getpooltoken(nmsg->base.nm_so);
588 if ((nmsg->base.lmsg.ms_flags & (MSGF_DONE | MSGF_REPLY)) == 0) {
589 ssb = (nmsg->nm_etype & NM_REVENT) ?
590 &nmsg->base.nm_so->so_rcv :
591 &nmsg->base.nm_so->so_snd;
592 TAILQ_REMOVE(&ssb->ssb_kq.ki_mlist, nmsg, nm_list);
593 lwkt_relpooltoken(nmsg->base.nm_so);
594 lwkt_replymsg(&nmsg->base.lmsg, EINTR);
596 lwkt_relpooltoken(nmsg->base.nm_so);
600 * Reply to the abort message
602 lwkt_replymsg(&abrtmsg->base.lmsg, 0);
606 so_async_rcvd_reply(struct socket *so)
609 * Spinlock safe, reply runs to degenerate lwkt_null_replyport()
611 spin_lock(&so->so_rcvd_spin);
612 lwkt_replymsg(&so->so_rcvd_msg.base.lmsg, 0);
613 spin_unlock(&so->so_rcvd_spin);
617 so_async_rcvd_drop(struct socket *so)
619 lwkt_msg_t lmsg = &so->so_rcvd_msg.base.lmsg;
622 * Spinlock safe, drop runs to degenerate lwkt_spin_dropmsg()
624 spin_lock(&so->so_rcvd_spin);
625 so->so_rcvd_msg.nm_pru_flags |= PRUR_DEAD;
628 if ((lmsg->ms_flags & MSGF_DONE) == 0) {
629 ++async_rcvd_drop_race;
630 ssleep(so, &so->so_rcvd_spin, 0, "soadrop", 1);
633 spin_unlock(&so->so_rcvd_spin);