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 #include <sys/param.h>
36 #include <sys/systm.h>
38 #include <sys/domain.h>
40 #include <sys/filedesc.h>
41 #include <sys/kernel.h>
42 #include <sys/kthread.h>
43 #include <sys/malloc.h>
45 #include <sys/protosw.h>
46 #include <sys/sigio.h>
47 #include <sys/signal.h>
48 #include <sys/signalvar.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/filio.h> /* XXX */
52 #include <sys/sockio.h>
54 #include <sys/sysctl.h>
55 #include <sys/sysproto.h>
56 #include <sys/taskqueue.h>
58 #include <sys/ucred.h>
60 #include <sys/unpcb.h>
64 #include <net/if_var.h>
65 #include <net/route.h>
68 #include <netinet/in.h>
69 #include <netinet/in_pcb.h>
71 #include <security/mac/mac_framework.h>
75 #include <vm/vm_extern.h>
76 #include <vm/vm_map.h>
78 static SYSCTL_NODE(_kern_ipc, OID_AUTO, aio, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
81 static int empty_results;
82 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, empty_results, CTLFLAG_RD, &empty_results,
83 0, "socket operation returned EAGAIN");
85 static int empty_retries;
86 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, empty_retries, CTLFLAG_RD, &empty_retries,
87 0, "socket operation retries");
89 static fo_rdwr_t soo_read;
90 static fo_rdwr_t soo_write;
91 static fo_ioctl_t soo_ioctl;
92 static fo_poll_t soo_poll;
93 extern fo_kqfilter_t soo_kqfilter;
94 static fo_stat_t soo_stat;
95 static fo_close_t soo_close;
96 static fo_fill_kinfo_t soo_fill_kinfo;
97 static fo_aio_queue_t soo_aio_queue;
99 static void soo_aio_cancel(struct kaiocb *job);
101 struct fileops socketops = {
103 .fo_write = soo_write,
104 .fo_truncate = invfo_truncate,
105 .fo_ioctl = soo_ioctl,
107 .fo_kqfilter = soo_kqfilter,
109 .fo_close = soo_close,
110 .fo_chmod = invfo_chmod,
111 .fo_chown = invfo_chown,
112 .fo_sendfile = invfo_sendfile,
113 .fo_fill_kinfo = soo_fill_kinfo,
114 .fo_aio_queue = soo_aio_queue,
115 .fo_flags = DFLAG_PASSABLE
119 soo_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
120 int flags, struct thread *td)
122 struct socket *so = fp->f_data;
126 error = mac_socket_check_receive(active_cred, so);
130 error = soreceive(so, 0, uio, 0, 0, 0);
135 soo_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
136 int flags, struct thread *td)
138 struct socket *so = fp->f_data;
142 error = mac_socket_check_send(active_cred, so);
146 error = sosend(so, 0, uio, 0, 0, 0, uio->uio_td);
147 if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0) {
148 PROC_LOCK(uio->uio_td->td_proc);
149 tdsignal(uio->uio_td, SIGPIPE);
150 PROC_UNLOCK(uio->uio_td->td_proc);
156 soo_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *active_cred,
159 struct socket *so = fp->f_data;
166 so->so_state |= SS_NBIO;
168 so->so_state &= ~SS_NBIO;
175 so->so_state |= SS_ASYNC;
176 if (SOLISTENING(so)) {
177 so->sol_sbrcv_flags |= SB_ASYNC;
178 so->sol_sbsnd_flags |= SB_ASYNC;
180 SOCKBUF_LOCK(&so->so_rcv);
181 so->so_rcv.sb_flags |= SB_ASYNC;
182 SOCKBUF_UNLOCK(&so->so_rcv);
183 SOCKBUF_LOCK(&so->so_snd);
184 so->so_snd.sb_flags |= SB_ASYNC;
185 SOCKBUF_UNLOCK(&so->so_snd);
190 so->so_state &= ~SS_ASYNC;
191 if (SOLISTENING(so)) {
192 so->sol_sbrcv_flags &= ~SB_ASYNC;
193 so->sol_sbsnd_flags &= ~SB_ASYNC;
195 SOCKBUF_LOCK(&so->so_rcv);
196 so->so_rcv.sb_flags &= ~SB_ASYNC;
197 SOCKBUF_UNLOCK(&so->so_rcv);
198 SOCKBUF_LOCK(&so->so_snd);
199 so->so_snd.sb_flags &= ~SB_ASYNC;
200 SOCKBUF_UNLOCK(&so->so_snd);
208 if (SOLISTENING(so)) {
211 *(int *)data = sbavail(&so->so_rcv);
217 if (SOLISTENING(so)) {
220 *(int *)data = sbavail(&so->so_snd);
226 if (SOLISTENING(so)) {
229 if ((so->so_snd.sb_hiwat < sbused(&so->so_snd)) ||
230 (so->so_snd.sb_mbmax < so->so_snd.sb_mbcnt)) {
233 *(int *)data = sbspace(&so->so_snd);
239 error = fsetown(*(int *)data, &so->so_sigio);
243 *(int *)data = fgetown(&so->so_sigio);
247 error = fsetown(-(*(int *)data), &so->so_sigio);
251 *(int *)data = -fgetown(&so->so_sigio);
256 if (SOLISTENING(so)) {
259 *(int *)data = (so->so_rcv.sb_state & SBS_RCVATMARK) != 0;
264 * Interface/routing/protocol specific ioctls: interface and
265 * routing ioctls should have a different entry since a
266 * socket is unnecessary.
268 if (IOCGROUP(cmd) == 'i')
269 error = ifioctl(so, cmd, data, td);
270 else if (IOCGROUP(cmd) == 'r') {
271 CURVNET_SET(so->so_vnet);
272 error = rtioctl_fib(cmd, data, so->so_fibnum);
275 CURVNET_SET(so->so_vnet);
276 error = ((*so->so_proto->pr_usrreqs->pru_control)
277 (so, cmd, data, 0, td));
286 soo_poll(struct file *fp, int events, struct ucred *active_cred,
289 struct socket *so = fp->f_data;
293 error = mac_socket_check_poll(active_cred, so);
297 return (sopoll(so, events, fp->f_cred, td));
301 soo_stat(struct file *fp, struct stat *ub, struct ucred *active_cred,
304 struct socket *so = fp->f_data;
307 bzero((caddr_t)ub, sizeof (*ub));
308 ub->st_mode = S_IFSOCK;
310 error = mac_socket_check_stat(active_cred, so);
315 if (!SOLISTENING(so)) {
319 * If SBS_CANTRCVMORE is set, but there's still data left
320 * in the receive buffer, the socket is still readable.
324 if ((sb->sb_state & SBS_CANTRCVMORE) == 0 || sbavail(sb))
325 ub->st_mode |= S_IRUSR | S_IRGRP | S_IROTH;
326 ub->st_size = sbavail(sb) - sb->sb_ctl;
331 if ((sb->sb_state & SBS_CANTSENDMORE) == 0)
332 ub->st_mode |= S_IWUSR | S_IWGRP | S_IWOTH;
335 ub->st_uid = so->so_cred->cr_uid;
336 ub->st_gid = so->so_cred->cr_gid;
337 error = so->so_proto->pr_usrreqs->pru_sense(so, ub);
343 * API socket close on file pointer. We call soclose() to close the socket
344 * (including initiating closing protocols). soclose() will sorele() the
345 * file reference but the actual socket will not go away until the socket's
349 soo_close(struct file *fp, struct thread *td)
355 fp->f_ops = &badfileops;
364 soo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
372 kif->kf_type = KF_TYPE_SOCKET;
374 CURVNET_SET(so->so_vnet);
375 kif->kf_un.kf_sock.kf_sock_domain0 =
376 so->so_proto->pr_domain->dom_family;
377 kif->kf_un.kf_sock.kf_sock_type0 = so->so_type;
378 kif->kf_un.kf_sock.kf_sock_protocol0 = so->so_proto->pr_protocol;
379 kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb;
380 switch (kif->kf_un.kf_sock.kf_sock_domain0) {
383 if (so->so_pcb != NULL) {
384 inpcb = (struct inpcb *)(so->so_pcb);
385 kif->kf_un.kf_sock.kf_sock_inpcb =
386 (uintptr_t)inpcb->inp_ppcb;
388 kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
390 kif->kf_un.kf_sock.kf_sock_snd_sb_state =
392 kif->kf_un.kf_sock.kf_sock_sendq =
394 kif->kf_un.kf_sock.kf_sock_recvq =
398 if (so->so_pcb != NULL) {
399 unpcb = (struct unpcb *)(so->so_pcb);
400 if (unpcb->unp_conn) {
401 kif->kf_un.kf_sock.kf_sock_unpconn =
402 (uintptr_t)unpcb->unp_conn;
403 kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
405 kif->kf_un.kf_sock.kf_sock_snd_sb_state =
407 kif->kf_un.kf_sock.kf_sock_sendq =
409 kif->kf_un.kf_sock.kf_sock_recvq =
415 error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa);
417 sa->sa_len <= sizeof(kif->kf_un.kf_sock.kf_sa_local)) {
418 bcopy(sa, &kif->kf_un.kf_sock.kf_sa_local, sa->sa_len);
421 error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa);
423 sa->sa_len <= sizeof(kif->kf_un.kf_sock.kf_sa_peer)) {
424 bcopy(sa, &kif->kf_un.kf_sock.kf_sa_peer, sa->sa_len);
427 strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name,
428 sizeof(kif->kf_path));
434 * Use the 'backend3' field in AIO jobs to store the amount of data
435 * completed by the AIO job so far.
437 #define aio_done backend3
439 static STAILQ_HEAD(, task) soaio_jobs;
440 static struct mtx soaio_jobs_lock;
441 static struct task soaio_kproc_task;
442 static int soaio_starting, soaio_idle, soaio_queued;
443 static struct unrhdr *soaio_kproc_unr;
445 static int soaio_max_procs = MAX_AIO_PROCS;
446 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, max_procs, CTLFLAG_RW, &soaio_max_procs, 0,
447 "Maximum number of kernel processes to use for async socket IO");
449 static int soaio_num_procs;
450 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, num_procs, CTLFLAG_RD, &soaio_num_procs, 0,
451 "Number of active kernel processes for async socket IO");
453 static int soaio_target_procs = TARGET_AIO_PROCS;
454 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, target_procs, CTLFLAG_RD,
455 &soaio_target_procs, 0,
456 "Preferred number of ready kernel processes for async socket IO");
458 static int soaio_lifetime;
459 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, lifetime, CTLFLAG_RW, &soaio_lifetime, 0,
460 "Maximum lifetime for idle aiod");
463 soaio_kproc_loop(void *arg)
466 struct vmspace *myvm;
468 int error, id, pending;
473 * Grab an extra reference on the daemon's vmspace so that it
474 * doesn't get freed by jobs that switch to a different
478 myvm = vmspace_acquire_ref(p);
480 mtx_lock(&soaio_jobs_lock);
481 MPASS(soaio_starting > 0);
484 while (!STAILQ_EMPTY(&soaio_jobs)) {
485 task = STAILQ_FIRST(&soaio_jobs);
486 STAILQ_REMOVE_HEAD(&soaio_jobs, ta_link);
488 pending = task->ta_pending;
489 task->ta_pending = 0;
490 mtx_unlock(&soaio_jobs_lock);
492 task->ta_func(task->ta_context, pending);
494 mtx_lock(&soaio_jobs_lock);
496 MPASS(soaio_queued == 0);
498 if (p->p_vmspace != myvm) {
499 mtx_unlock(&soaio_jobs_lock);
500 vmspace_switch_aio(myvm);
501 mtx_lock(&soaio_jobs_lock);
506 error = mtx_sleep(&soaio_idle, &soaio_jobs_lock, 0, "-",
509 if (error == EWOULDBLOCK && STAILQ_EMPTY(&soaio_jobs) &&
510 soaio_num_procs > soaio_target_procs)
514 mtx_unlock(&soaio_jobs_lock);
515 free_unr(soaio_kproc_unr, id);
520 soaio_kproc_create(void *context, int pending)
525 mtx_lock(&soaio_jobs_lock);
527 if (soaio_num_procs < soaio_target_procs) {
529 } else if (soaio_num_procs >= soaio_max_procs) {
531 * Hit the limit on kernel processes, don't
532 * create another one.
535 } else if (soaio_queued <= soaio_idle + soaio_starting) {
537 * No more AIO jobs waiting for a process to be
543 mtx_unlock(&soaio_jobs_lock);
545 id = alloc_unr(soaio_kproc_unr);
546 error = kproc_create(soaio_kproc_loop, (void *)(intptr_t)id,
547 &p, 0, 0, "soaiod%d", id);
549 free_unr(soaio_kproc_unr, id);
550 mtx_lock(&soaio_jobs_lock);
555 mtx_lock(&soaio_jobs_lock);
558 mtx_unlock(&soaio_jobs_lock);
562 soaio_enqueue(struct task *task)
565 mtx_lock(&soaio_jobs_lock);
566 MPASS(task->ta_pending == 0);
568 STAILQ_INSERT_TAIL(&soaio_jobs, task, ta_link);
570 if (soaio_queued <= soaio_idle)
571 wakeup_one(&soaio_idle);
572 else if (soaio_num_procs < soaio_max_procs)
573 taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task);
574 mtx_unlock(&soaio_jobs_lock);
581 soaio_lifetime = AIOD_LIFETIME_DEFAULT;
582 STAILQ_INIT(&soaio_jobs);
583 mtx_init(&soaio_jobs_lock, "soaio jobs", NULL, MTX_DEF);
584 soaio_kproc_unr = new_unrhdr(1, INT_MAX, NULL);
585 TASK_INIT(&soaio_kproc_task, 0, soaio_kproc_create, NULL);
587 SYSINIT(soaio, SI_SUB_VFS, SI_ORDER_ANY, soaio_init, NULL);
590 soaio_ready(struct socket *so, struct sockbuf *sb)
592 return (sb == &so->so_rcv ? soreadable(so) : sowriteable(so));
596 soaio_process_job(struct socket *so, struct sockbuf *sb, struct kaiocb *job)
598 struct ucred *td_savedcred;
601 size_t cnt, done, job_total_nbytes;
606 aio_switch_vmspace(job);
610 td_savedcred = td->td_ucred;
611 td->td_ucred = job->cred;
613 job_total_nbytes = job->uiop->uio_resid + job->aio_done;
614 done = job->aio_done;
615 cnt = job->uiop->uio_resid;
616 job->uiop->uio_offset = 0;
617 job->uiop->uio_td = td;
621 * For resource usage accounting, only count a completed request
622 * as a single message to avoid counting multiple calls to
623 * sosend/soreceive on a blocking socket.
626 if (sb == &so->so_rcv) {
627 ru_before = td->td_ru.ru_msgrcv;
629 error = mac_socket_check_receive(fp->f_cred, so);
633 error = soreceive(so, NULL, job->uiop, NULL, NULL,
635 if (td->td_ru.ru_msgrcv != ru_before)
638 if (!TAILQ_EMPTY(&sb->sb_aiojobq))
639 flags |= MSG_MORETOCOME;
640 ru_before = td->td_ru.ru_msgsnd;
642 error = mac_socket_check_send(fp->f_cred, so);
645 error = sosend(so, NULL, job->uiop, NULL, NULL, flags,
647 if (td->td_ru.ru_msgsnd != ru_before)
649 if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0) {
650 PROC_LOCK(job->userproc);
651 kern_psignal(job->userproc, SIGPIPE);
652 PROC_UNLOCK(job->userproc);
656 done += cnt - job->uiop->uio_resid;
657 job->aio_done = done;
658 td->td_ucred = td_savedcred;
660 if (error == EWOULDBLOCK) {
662 * The request was either partially completed or not
663 * completed at all due to racing with a read() or
664 * write() on the socket. If the socket is
665 * non-blocking, return with any partial completion.
666 * If the socket is blocking or if no progress has
667 * been made, requeue this request at the head of the
668 * queue to try again when the socket is ready.
670 MPASS(done != job_total_nbytes);
672 if (done == 0 || !(so->so_state & SS_NBIO)) {
674 if (soaio_ready(so, sb)) {
680 if (!aio_set_cancel_function(job, soo_aio_cancel)) {
683 aio_complete(job, done, 0);
688 TAILQ_INSERT_HEAD(&sb->sb_aiojobq, job, list);
694 if (done != 0 && (error == ERESTART || error == EINTR ||
695 error == EWOULDBLOCK))
698 aio_complete(job, -1, error);
700 aio_complete(job, done, 0);
705 soaio_process_sb(struct socket *so, struct sockbuf *sb)
709 CURVNET_SET(so->so_vnet);
711 while (!TAILQ_EMPTY(&sb->sb_aiojobq) && soaio_ready(so, sb)) {
712 job = TAILQ_FIRST(&sb->sb_aiojobq);
713 TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
714 if (!aio_clear_cancel_function(job))
717 soaio_process_job(so, sb, job);
721 * If there are still pending requests, the socket must not be
722 * ready so set SB_AIO to request a wakeup when the socket
725 if (!TAILQ_EMPTY(&sb->sb_aiojobq))
726 sb->sb_flags |= SB_AIO;
727 sb->sb_flags &= ~SB_AIO_RUNNING;
736 soaio_rcv(void *context, int pending)
741 soaio_process_sb(so, &so->so_rcv);
745 soaio_snd(void *context, int pending)
750 soaio_process_sb(so, &so->so_snd);
754 sowakeup_aio(struct socket *so, struct sockbuf *sb)
757 SOCKBUF_LOCK_ASSERT(sb);
758 sb->sb_flags &= ~SB_AIO;
759 if (sb->sb_flags & SB_AIO_RUNNING)
761 sb->sb_flags |= SB_AIO_RUNNING;
763 soaio_enqueue(&sb->sb_aiotask);
767 soo_aio_cancel(struct kaiocb *job)
774 so = job->fd_file->f_data;
775 opcode = job->uaiocb.aio_lio_opcode;
776 if (opcode & LIO_READ)
779 MPASS(opcode & LIO_WRITE);
784 if (!aio_cancel_cleared(job))
785 TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
786 if (TAILQ_EMPTY(&sb->sb_aiojobq))
787 sb->sb_flags &= ~SB_AIO;
790 done = job->aio_done;
792 aio_complete(job, done, 0);
798 soo_aio_queue(struct file *fp, struct kaiocb *job)
805 error = (*so->so_proto->pr_usrreqs->pru_aio_queue)(so, job);
809 switch (job->uaiocb.aio_lio_opcode & (LIO_WRITE | LIO_READ)) {
821 if (!aio_set_cancel_function(job, soo_aio_cancel))
822 panic("new job was cancelled");
823 TAILQ_INSERT_TAIL(&sb->sb_aiojobq, job, list);
824 if (!(sb->sb_flags & SB_AIO_RUNNING)) {
825 if (soaio_ready(so, sb))
826 sowakeup_aio(so, sb);
828 sb->sb_flags |= SB_AIO;