2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1990, 1993
5 * The Regents of the University of California. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * @(#)sys_socket.c 8.1 (Berkeley) 6/10/93
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
37 #include <sys/param.h>
38 #include <sys/systm.h>
40 #include <sys/domain.h>
42 #include <sys/filedesc.h>
43 #include <sys/kernel.h>
44 #include <sys/kthread.h>
45 #include <sys/malloc.h>
47 #include <sys/protosw.h>
48 #include <sys/sigio.h>
49 #include <sys/signal.h>
50 #include <sys/signalvar.h>
51 #include <sys/socket.h>
52 #include <sys/socketvar.h>
53 #include <sys/filio.h> /* XXX */
54 #include <sys/sockio.h>
56 #include <sys/sysctl.h>
57 #include <sys/sysproto.h>
58 #include <sys/taskqueue.h>
60 #include <sys/ucred.h>
62 #include <sys/unpcb.h>
66 #include <net/if_var.h>
67 #include <net/route.h>
70 #include <netinet/in.h>
71 #include <netinet/in_pcb.h>
73 #include <security/mac/mac_framework.h>
77 #include <vm/vm_extern.h>
78 #include <vm/vm_map.h>
80 static SYSCTL_NODE(_kern_ipc, OID_AUTO, aio, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
83 static int empty_results;
84 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, empty_results, CTLFLAG_RD, &empty_results,
85 0, "socket operation returned EAGAIN");
87 static int empty_retries;
88 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, empty_retries, CTLFLAG_RD, &empty_retries,
89 0, "socket operation retries");
91 static fo_rdwr_t soo_read;
92 static fo_rdwr_t soo_write;
93 static fo_ioctl_t soo_ioctl;
94 static fo_poll_t soo_poll;
95 extern fo_kqfilter_t soo_kqfilter;
96 static fo_stat_t soo_stat;
97 static fo_close_t soo_close;
98 static fo_fill_kinfo_t soo_fill_kinfo;
99 static fo_aio_queue_t soo_aio_queue;
101 static void soo_aio_cancel(struct kaiocb *job);
103 struct fileops socketops = {
105 .fo_write = soo_write,
106 .fo_truncate = invfo_truncate,
107 .fo_ioctl = soo_ioctl,
109 .fo_kqfilter = soo_kqfilter,
111 .fo_close = soo_close,
112 .fo_chmod = invfo_chmod,
113 .fo_chown = invfo_chown,
114 .fo_sendfile = invfo_sendfile,
115 .fo_fill_kinfo = soo_fill_kinfo,
116 .fo_aio_queue = soo_aio_queue,
117 .fo_flags = DFLAG_PASSABLE
121 soo_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
122 int flags, struct thread *td)
124 struct socket *so = fp->f_data;
128 error = mac_socket_check_receive(active_cred, so);
132 error = soreceive(so, 0, uio, 0, 0, 0);
137 soo_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
138 int flags, struct thread *td)
140 struct socket *so = fp->f_data;
144 error = mac_socket_check_send(active_cred, so);
148 error = sousrsend(so, NULL, uio, NULL, 0, NULL);
153 soo_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *active_cred,
156 struct socket *so = fp->f_data;
163 so->so_state |= SS_NBIO;
165 so->so_state &= ~SS_NBIO;
172 so->so_state |= SS_ASYNC;
173 if (SOLISTENING(so)) {
174 so->sol_sbrcv_flags |= SB_ASYNC;
175 so->sol_sbsnd_flags |= SB_ASYNC;
177 SOCK_RECVBUF_LOCK(so);
178 so->so_rcv.sb_flags |= SB_ASYNC;
179 SOCK_RECVBUF_UNLOCK(so);
180 SOCK_SENDBUF_LOCK(so);
181 so->so_snd.sb_flags |= SB_ASYNC;
182 SOCK_SENDBUF_UNLOCK(so);
187 so->so_state &= ~SS_ASYNC;
188 if (SOLISTENING(so)) {
189 so->sol_sbrcv_flags &= ~SB_ASYNC;
190 so->sol_sbsnd_flags &= ~SB_ASYNC;
192 SOCK_RECVBUF_LOCK(so);
193 so->so_rcv.sb_flags &= ~SB_ASYNC;
194 SOCK_RECVBUF_UNLOCK(so);
195 SOCK_SENDBUF_LOCK(so);
196 so->so_snd.sb_flags &= ~SB_ASYNC;
197 SOCK_SENDBUF_UNLOCK(so);
204 SOCK_RECVBUF_LOCK(so);
205 if (SOLISTENING(so)) {
208 *(int *)data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
210 SOCK_RECVBUF_UNLOCK(so);
215 if (SOLISTENING(so)) {
218 *(int *)data = sbavail(&so->so_snd);
224 if (SOLISTENING(so)) {
227 if ((so->so_snd.sb_hiwat < sbused(&so->so_snd)) ||
228 (so->so_snd.sb_mbmax < so->so_snd.sb_mbcnt)) {
231 *(int *)data = sbspace(&so->so_snd);
237 error = fsetown(*(int *)data, &so->so_sigio);
241 *(int *)data = fgetown(&so->so_sigio);
245 error = fsetown(-(*(int *)data), &so->so_sigio);
249 *(int *)data = -fgetown(&so->so_sigio);
254 if (SOLISTENING(so)) {
257 *(int *)data = (so->so_rcv.sb_state & SBS_RCVATMARK) != 0;
262 * Interface/routing/protocol specific ioctls: interface and
263 * routing ioctls should have a different entry since a
264 * socket is unnecessary.
266 if (IOCGROUP(cmd) == 'i')
267 error = ifioctl(so, cmd, data, td);
268 else if (IOCGROUP(cmd) == 'r') {
269 CURVNET_SET(so->so_vnet);
270 error = rtioctl_fib(cmd, data, so->so_fibnum);
273 CURVNET_SET(so->so_vnet);
274 error = so->so_proto->pr_control(so, cmd, data, 0, td);
283 soo_poll(struct file *fp, int events, struct ucred *active_cred,
286 struct socket *so = fp->f_data;
290 error = mac_socket_check_poll(active_cred, so);
294 return (sopoll(so, events, fp->f_cred, td));
298 soo_stat(struct file *fp, struct stat *ub, struct ucred *active_cred)
300 struct socket *so = fp->f_data;
303 bzero((caddr_t)ub, sizeof (*ub));
304 ub->st_mode = S_IFSOCK;
306 error = mac_socket_check_stat(active_cred, so);
311 if (!SOLISTENING(so)) {
315 * If SBS_CANTRCVMORE is set, but there's still data left
316 * in the receive buffer, the socket is still readable.
319 SOCK_RECVBUF_LOCK(so);
320 if ((sb->sb_state & SBS_CANTRCVMORE) == 0 || sbavail(sb))
321 ub->st_mode |= S_IRUSR | S_IRGRP | S_IROTH;
322 ub->st_size = sbavail(sb) - sb->sb_ctl;
323 SOCK_RECVBUF_UNLOCK(so);
326 SOCK_SENDBUF_LOCK(so);
327 if ((sb->sb_state & SBS_CANTSENDMORE) == 0)
328 ub->st_mode |= S_IWUSR | S_IWGRP | S_IWOTH;
329 SOCK_SENDBUF_UNLOCK(so);
331 ub->st_uid = so->so_cred->cr_uid;
332 ub->st_gid = so->so_cred->cr_gid;
333 if (so->so_proto->pr_sense)
334 error = so->so_proto->pr_sense(so, ub);
340 * API socket close on file pointer. We call soclose() to close the socket
341 * (including initiating closing protocols). soclose() will sorele() the
342 * file reference but the actual socket will not go away until the socket's
346 soo_close(struct file *fp, struct thread *td)
352 fp->f_ops = &badfileops;
361 soo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
369 kif->kf_type = KF_TYPE_SOCKET;
371 CURVNET_SET(so->so_vnet);
372 kif->kf_un.kf_sock.kf_sock_domain0 =
373 so->so_proto->pr_domain->dom_family;
374 kif->kf_un.kf_sock.kf_sock_type0 = so->so_type;
375 kif->kf_un.kf_sock.kf_sock_protocol0 = so->so_proto->pr_protocol;
376 kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb;
377 switch (kif->kf_un.kf_sock.kf_sock_domain0) {
380 if (so->so_pcb != NULL) {
381 inpcb = (struct inpcb *)(so->so_pcb);
382 kif->kf_un.kf_sock.kf_sock_inpcb =
383 (uintptr_t)inpcb->inp_ppcb;
385 kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
387 kif->kf_un.kf_sock.kf_sock_snd_sb_state =
389 kif->kf_un.kf_sock.kf_sock_sendq =
391 kif->kf_un.kf_sock.kf_sock_recvq =
395 if (so->so_pcb != NULL) {
396 unpcb = (struct unpcb *)(so->so_pcb);
397 if (unpcb->unp_conn) {
398 kif->kf_un.kf_sock.kf_sock_unpconn =
399 (uintptr_t)unpcb->unp_conn;
400 kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
402 kif->kf_un.kf_sock.kf_sock_snd_sb_state =
404 kif->kf_un.kf_sock.kf_sock_sendq =
406 kif->kf_un.kf_sock.kf_sock_recvq =
412 error = so->so_proto->pr_sockaddr(so, &sa);
414 sa->sa_len <= sizeof(kif->kf_un.kf_sock.kf_sa_local)) {
415 bcopy(sa, &kif->kf_un.kf_sock.kf_sa_local, sa->sa_len);
418 error = so->so_proto->pr_peeraddr(so, &sa);
420 sa->sa_len <= sizeof(kif->kf_un.kf_sock.kf_sa_peer)) {
421 bcopy(sa, &kif->kf_un.kf_sock.kf_sa_peer, sa->sa_len);
424 strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name,
425 sizeof(kif->kf_path));
431 * Use the 'backend3' field in AIO jobs to store the amount of data
432 * completed by the AIO job so far.
434 #define aio_done backend3
436 static STAILQ_HEAD(, task) soaio_jobs;
437 static struct mtx soaio_jobs_lock;
438 static struct task soaio_kproc_task;
439 static int soaio_starting, soaio_idle, soaio_queued;
440 static struct unrhdr *soaio_kproc_unr;
442 static int soaio_max_procs = MAX_AIO_PROCS;
443 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, max_procs, CTLFLAG_RW, &soaio_max_procs, 0,
444 "Maximum number of kernel processes to use for async socket IO");
446 static int soaio_num_procs;
447 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, num_procs, CTLFLAG_RD, &soaio_num_procs, 0,
448 "Number of active kernel processes for async socket IO");
450 static int soaio_target_procs = TARGET_AIO_PROCS;
451 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, target_procs, CTLFLAG_RD,
452 &soaio_target_procs, 0,
453 "Preferred number of ready kernel processes for async socket IO");
455 static int soaio_lifetime;
456 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, lifetime, CTLFLAG_RW, &soaio_lifetime, 0,
457 "Maximum lifetime for idle aiod");
460 soaio_kproc_loop(void *arg)
463 struct vmspace *myvm;
465 int error, id, pending;
470 * Grab an extra reference on the daemon's vmspace so that it
471 * doesn't get freed by jobs that switch to a different
475 myvm = vmspace_acquire_ref(p);
477 mtx_lock(&soaio_jobs_lock);
478 MPASS(soaio_starting > 0);
481 while (!STAILQ_EMPTY(&soaio_jobs)) {
482 task = STAILQ_FIRST(&soaio_jobs);
483 STAILQ_REMOVE_HEAD(&soaio_jobs, ta_link);
485 pending = task->ta_pending;
486 task->ta_pending = 0;
487 mtx_unlock(&soaio_jobs_lock);
489 task->ta_func(task->ta_context, pending);
491 mtx_lock(&soaio_jobs_lock);
493 MPASS(soaio_queued == 0);
495 if (p->p_vmspace != myvm) {
496 mtx_unlock(&soaio_jobs_lock);
497 vmspace_switch_aio(myvm);
498 mtx_lock(&soaio_jobs_lock);
503 error = mtx_sleep(&soaio_idle, &soaio_jobs_lock, 0, "-",
506 if (error == EWOULDBLOCK && STAILQ_EMPTY(&soaio_jobs) &&
507 soaio_num_procs > soaio_target_procs)
511 mtx_unlock(&soaio_jobs_lock);
512 free_unr(soaio_kproc_unr, id);
517 soaio_kproc_create(void *context, int pending)
522 mtx_lock(&soaio_jobs_lock);
524 if (soaio_num_procs < soaio_target_procs) {
526 } else if (soaio_num_procs >= soaio_max_procs) {
528 * Hit the limit on kernel processes, don't
529 * create another one.
532 } else if (soaio_queued <= soaio_idle + soaio_starting) {
534 * No more AIO jobs waiting for a process to be
540 mtx_unlock(&soaio_jobs_lock);
542 id = alloc_unr(soaio_kproc_unr);
543 error = kproc_create(soaio_kproc_loop, (void *)(intptr_t)id,
544 &p, 0, 0, "soaiod%d", id);
546 free_unr(soaio_kproc_unr, id);
547 mtx_lock(&soaio_jobs_lock);
552 mtx_lock(&soaio_jobs_lock);
555 mtx_unlock(&soaio_jobs_lock);
559 soaio_enqueue(struct task *task)
562 mtx_lock(&soaio_jobs_lock);
563 MPASS(task->ta_pending == 0);
565 STAILQ_INSERT_TAIL(&soaio_jobs, task, ta_link);
567 if (soaio_queued <= soaio_idle)
568 wakeup_one(&soaio_idle);
569 else if (soaio_num_procs < soaio_max_procs)
570 taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task);
571 mtx_unlock(&soaio_jobs_lock);
578 soaio_lifetime = AIOD_LIFETIME_DEFAULT;
579 STAILQ_INIT(&soaio_jobs);
580 mtx_init(&soaio_jobs_lock, "soaio jobs", NULL, MTX_DEF);
581 soaio_kproc_unr = new_unrhdr(1, INT_MAX, NULL);
582 TASK_INIT(&soaio_kproc_task, 0, soaio_kproc_create, NULL);
584 SYSINIT(soaio, SI_SUB_VFS, SI_ORDER_ANY, soaio_init, NULL);
587 soaio_ready(struct socket *so, struct sockbuf *sb)
589 return (sb == &so->so_rcv ? soreadable(so) : sowriteable(so));
593 soaio_process_job(struct socket *so, sb_which which, struct kaiocb *job)
595 struct ucred *td_savedcred;
597 struct sockbuf *sb = sobuf(so, which);
599 struct file *fp = job->fd_file;
601 size_t cnt, done, job_total_nbytes __diagused;
605 SOCK_BUF_UNLOCK(so, which);
606 aio_switch_vmspace(job);
609 td_savedcred = td->td_ucred;
610 td->td_ucred = job->cred;
612 job_total_nbytes = job->uiop->uio_resid + job->aio_done;
613 done = job->aio_done;
614 cnt = job->uiop->uio_resid;
615 job->uiop->uio_offset = 0;
616 job->uiop->uio_td = td;
620 * For resource usage accounting, only count a completed request
621 * as a single message to avoid counting multiple calls to
622 * sosend/soreceive on a blocking socket.
625 if (sb == &so->so_rcv) {
626 ru_before = td->td_ru.ru_msgrcv;
628 error = mac_socket_check_receive(fp->f_cred, so);
632 error = soreceive(so, NULL, job->uiop, NULL, NULL,
634 if (td->td_ru.ru_msgrcv != ru_before)
637 if (!TAILQ_EMPTY(&sb->sb_aiojobq))
638 flags |= MSG_MORETOCOME;
639 ru_before = td->td_ru.ru_msgsnd;
641 error = mac_socket_check_send(fp->f_cred, so);
644 error = sousrsend(so, NULL, job->uiop, NULL, flags,
646 if (td->td_ru.ru_msgsnd != ru_before)
650 done += cnt - job->uiop->uio_resid;
651 job->aio_done = done;
652 td->td_ucred = td_savedcred;
654 if (error == EWOULDBLOCK) {
656 * The request was either partially completed or not
657 * completed at all due to racing with a read() or
658 * write() on the socket. If the socket is
659 * non-blocking, return with any partial completion.
660 * If the socket is blocking or if no progress has
661 * been made, requeue this request at the head of the
662 * queue to try again when the socket is ready.
664 MPASS(done != job_total_nbytes);
665 SOCK_BUF_LOCK(so, which);
666 if (done == 0 || !(so->so_state & SS_NBIO)) {
668 if (soaio_ready(so, sb)) {
670 SOCK_BUF_UNLOCK(so, which);
674 if (!aio_set_cancel_function(job, soo_aio_cancel)) {
675 SOCK_BUF_UNLOCK(so, which);
677 aio_complete(job, done, 0);
680 SOCK_BUF_LOCK(so, which);
682 TAILQ_INSERT_HEAD(&sb->sb_aiojobq, job, list);
686 SOCK_BUF_UNLOCK(so, which);
688 if (done != 0 && (error == ERESTART || error == EINTR ||
689 error == EWOULDBLOCK))
692 aio_complete(job, -1, error);
694 aio_complete(job, done, 0);
695 SOCK_BUF_LOCK(so, which);
699 soaio_process_sb(struct socket *so, sb_which which)
702 struct sockbuf *sb = sobuf(so, which);
704 CURVNET_SET(so->so_vnet);
705 SOCK_BUF_LOCK(so, which);
706 while (!TAILQ_EMPTY(&sb->sb_aiojobq) && soaio_ready(so, sb)) {
707 job = TAILQ_FIRST(&sb->sb_aiojobq);
708 TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
709 if (!aio_clear_cancel_function(job))
712 soaio_process_job(so, which, job);
716 * If there are still pending requests, the socket must not be
717 * ready so set SB_AIO to request a wakeup when the socket
720 if (!TAILQ_EMPTY(&sb->sb_aiojobq))
721 sb->sb_flags |= SB_AIO;
722 sb->sb_flags &= ~SB_AIO_RUNNING;
723 SOCK_BUF_UNLOCK(so, which);
730 soaio_rcv(void *context, int pending)
735 soaio_process_sb(so, SO_RCV);
739 soaio_snd(void *context, int pending)
744 soaio_process_sb(so, SO_SND);
748 sowakeup_aio(struct socket *so, sb_which which)
750 struct sockbuf *sb = sobuf(so, which);
752 SOCK_BUF_LOCK_ASSERT(so, which);
754 sb->sb_flags &= ~SB_AIO;
755 if (sb->sb_flags & SB_AIO_RUNNING)
757 sb->sb_flags |= SB_AIO_RUNNING;
759 soaio_enqueue(&sb->sb_aiotask);
763 soo_aio_cancel(struct kaiocb *job)
771 so = job->fd_file->f_data;
772 opcode = job->uaiocb.aio_lio_opcode;
773 if (opcode & LIO_READ) {
777 MPASS(opcode & LIO_WRITE);
782 SOCK_BUF_LOCK(so, which);
783 if (!aio_cancel_cleared(job))
784 TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
785 if (TAILQ_EMPTY(&sb->sb_aiojobq))
786 sb->sb_flags &= ~SB_AIO;
787 SOCK_BUF_UNLOCK(so, which);
789 done = job->aio_done;
791 aio_complete(job, done, 0);
797 soo_aio_queue(struct file *fp, struct kaiocb *job)
805 error = so->so_proto->pr_aio_queue(so, job);
809 /* Lock through the socket, since this may be a listening socket. */
810 switch (job->uaiocb.aio_lio_opcode & (LIO_WRITE | LIO_READ)) {
812 SOCK_RECVBUF_LOCK(so);
817 SOCK_SENDBUF_LOCK(so);
825 if (SOLISTENING(so)) {
826 SOCK_BUF_UNLOCK(so, which);
830 if (!aio_set_cancel_function(job, soo_aio_cancel))
831 panic("new job was cancelled");
832 TAILQ_INSERT_TAIL(&sb->sb_aiojobq, job, list);
833 if (!(sb->sb_flags & SB_AIO_RUNNING)) {
834 if (soaio_ready(so, sb))
835 sowakeup_aio(so, which);
837 sb->sb_flags |= SB_AIO;
839 SOCK_BUF_UNLOCK(so, which);