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
32 #include <sys/param.h>
33 #include <sys/systm.h>
35 #include <sys/domain.h>
37 #include <sys/filedesc.h>
38 #include <sys/kernel.h>
39 #include <sys/kthread.h>
40 #include <sys/malloc.h>
42 #include <sys/protosw.h>
43 #include <sys/sigio.h>
44 #include <sys/signal.h>
45 #include <sys/signalvar.h>
46 #include <sys/socket.h>
47 #include <sys/socketvar.h>
48 #include <sys/filio.h> /* XXX */
49 #include <sys/sockio.h>
51 #include <sys/sysctl.h>
52 #include <sys/sysproto.h>
53 #include <sys/taskqueue.h>
55 #include <sys/ucred.h>
57 #include <sys/unpcb.h>
61 #include <net/if_var.h>
62 #include <net/route.h>
65 #include <netinet/in.h>
66 #include <netinet/in_pcb.h>
68 #include <security/mac/mac_framework.h>
72 #include <vm/vm_extern.h>
73 #include <vm/vm_map.h>
75 static SYSCTL_NODE(_kern_ipc, OID_AUTO, aio, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
78 static int empty_results;
79 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, empty_results, CTLFLAG_RD, &empty_results,
80 0, "socket operation returned EAGAIN");
82 static int empty_retries;
83 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, empty_retries, CTLFLAG_RD, &empty_retries,
84 0, "socket operation retries");
86 static fo_rdwr_t soo_read;
87 static fo_rdwr_t soo_write;
88 static fo_ioctl_t soo_ioctl;
89 static fo_poll_t soo_poll;
90 extern fo_kqfilter_t soo_kqfilter;
91 static fo_stat_t soo_stat;
92 static fo_close_t soo_close;
93 static fo_fill_kinfo_t soo_fill_kinfo;
94 static fo_aio_queue_t soo_aio_queue;
96 static void soo_aio_cancel(struct kaiocb *job);
98 struct fileops socketops = {
100 .fo_write = soo_write,
101 .fo_truncate = invfo_truncate,
102 .fo_ioctl = soo_ioctl,
104 .fo_kqfilter = soo_kqfilter,
106 .fo_close = soo_close,
107 .fo_chmod = invfo_chmod,
108 .fo_chown = invfo_chown,
109 .fo_sendfile = invfo_sendfile,
110 .fo_fill_kinfo = soo_fill_kinfo,
111 .fo_aio_queue = soo_aio_queue,
112 .fo_cmp = file_kcmp_generic,
113 .fo_flags = DFLAG_PASSABLE
117 soo_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
118 int flags, struct thread *td)
120 struct socket *so = fp->f_data;
124 error = mac_socket_check_receive(active_cred, so);
128 error = soreceive(so, 0, uio, 0, 0, 0);
133 soo_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
134 int flags, struct thread *td)
136 struct socket *so = fp->f_data;
140 error = mac_socket_check_send(active_cred, so);
144 error = sousrsend(so, NULL, uio, NULL, 0, NULL);
149 soo_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *active_cred,
152 struct socket *so = fp->f_data;
159 so->so_state |= SS_NBIO;
161 so->so_state &= ~SS_NBIO;
168 so->so_state |= SS_ASYNC;
169 if (SOLISTENING(so)) {
170 so->sol_sbrcv_flags |= SB_ASYNC;
171 so->sol_sbsnd_flags |= SB_ASYNC;
173 SOCK_RECVBUF_LOCK(so);
174 so->so_rcv.sb_flags |= SB_ASYNC;
175 SOCK_RECVBUF_UNLOCK(so);
176 SOCK_SENDBUF_LOCK(so);
177 so->so_snd.sb_flags |= SB_ASYNC;
178 SOCK_SENDBUF_UNLOCK(so);
183 so->so_state &= ~SS_ASYNC;
184 if (SOLISTENING(so)) {
185 so->sol_sbrcv_flags &= ~SB_ASYNC;
186 so->sol_sbsnd_flags &= ~SB_ASYNC;
188 SOCK_RECVBUF_LOCK(so);
189 so->so_rcv.sb_flags &= ~SB_ASYNC;
190 SOCK_RECVBUF_UNLOCK(so);
191 SOCK_SENDBUF_LOCK(so);
192 so->so_snd.sb_flags &= ~SB_ASYNC;
193 SOCK_SENDBUF_UNLOCK(so);
200 SOCK_RECVBUF_LOCK(so);
201 if (SOLISTENING(so)) {
204 *(int *)data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
206 SOCK_RECVBUF_UNLOCK(so);
211 if (SOLISTENING(so)) {
214 *(int *)data = sbavail(&so->so_snd);
220 if (SOLISTENING(so)) {
223 if ((so->so_snd.sb_hiwat < sbused(&so->so_snd)) ||
224 (so->so_snd.sb_mbmax < so->so_snd.sb_mbcnt)) {
227 *(int *)data = sbspace(&so->so_snd);
233 error = fsetown(*(int *)data, &so->so_sigio);
237 *(int *)data = fgetown(&so->so_sigio);
241 error = fsetown(-(*(int *)data), &so->so_sigio);
245 *(int *)data = -fgetown(&so->so_sigio);
250 if (SOLISTENING(so)) {
253 *(int *)data = (so->so_rcv.sb_state & SBS_RCVATMARK) != 0;
258 * Interface/routing/protocol specific ioctls: interface and
259 * routing ioctls should have a different entry since a
260 * socket is unnecessary.
262 if (IOCGROUP(cmd) == 'i')
263 error = ifioctl(so, cmd, data, td);
264 else if (IOCGROUP(cmd) == 'r') {
265 CURVNET_SET(so->so_vnet);
266 error = rtioctl_fib(cmd, data, so->so_fibnum);
269 CURVNET_SET(so->so_vnet);
270 error = so->so_proto->pr_control(so, cmd, data, 0, td);
279 soo_poll(struct file *fp, int events, struct ucred *active_cred,
282 struct socket *so = fp->f_data;
286 error = mac_socket_check_poll(active_cred, so);
290 return (sopoll(so, events, fp->f_cred, td));
294 soo_stat(struct file *fp, struct stat *ub, struct ucred *active_cred)
296 struct socket *so = fp->f_data;
299 bzero((caddr_t)ub, sizeof (*ub));
300 ub->st_mode = S_IFSOCK;
302 error = mac_socket_check_stat(active_cred, so);
307 if (!SOLISTENING(so)) {
311 * If SBS_CANTRCVMORE is set, but there's still data left
312 * in the receive buffer, the socket is still readable.
315 SOCK_RECVBUF_LOCK(so);
316 if ((sb->sb_state & SBS_CANTRCVMORE) == 0 || sbavail(sb))
317 ub->st_mode |= S_IRUSR | S_IRGRP | S_IROTH;
318 ub->st_size = sbavail(sb) - sb->sb_ctl;
319 SOCK_RECVBUF_UNLOCK(so);
322 SOCK_SENDBUF_LOCK(so);
323 if ((sb->sb_state & SBS_CANTSENDMORE) == 0)
324 ub->st_mode |= S_IWUSR | S_IWGRP | S_IWOTH;
325 SOCK_SENDBUF_UNLOCK(so);
327 ub->st_uid = so->so_cred->cr_uid;
328 ub->st_gid = so->so_cred->cr_gid;
329 if (so->so_proto->pr_sense)
330 error = so->so_proto->pr_sense(so, ub);
336 * API socket close on file pointer. We call soclose() to close the socket
337 * (including initiating closing protocols). soclose() will sorele() the
338 * file reference but the actual socket will not go away until the socket's
342 soo_close(struct file *fp, struct thread *td)
348 fp->f_ops = &badfileops;
357 soo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
359 struct sockaddr_storage ss = { .ss_len = sizeof(ss) };
365 kif->kf_type = KF_TYPE_SOCKET;
367 CURVNET_SET(so->so_vnet);
368 kif->kf_un.kf_sock.kf_sock_domain0 =
369 so->so_proto->pr_domain->dom_family;
370 kif->kf_un.kf_sock.kf_sock_type0 = so->so_type;
371 kif->kf_un.kf_sock.kf_sock_protocol0 = so->so_proto->pr_protocol;
372 kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb;
373 switch (kif->kf_un.kf_sock.kf_sock_domain0) {
376 if (so->so_pcb != NULL) {
377 inpcb = (struct inpcb *)(so->so_pcb);
378 kif->kf_un.kf_sock.kf_sock_inpcb =
379 (uintptr_t)inpcb->inp_ppcb;
381 kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
383 kif->kf_un.kf_sock.kf_sock_snd_sb_state =
385 kif->kf_un.kf_sock.kf_sock_sendq =
387 kif->kf_un.kf_sock.kf_sock_recvq =
391 if (so->so_pcb != NULL) {
392 unpcb = (struct unpcb *)(so->so_pcb);
393 if (unpcb->unp_conn) {
394 kif->kf_un.kf_sock.kf_sock_unpconn =
395 (uintptr_t)unpcb->unp_conn;
396 kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
398 kif->kf_un.kf_sock.kf_sock_snd_sb_state =
400 kif->kf_un.kf_sock.kf_sock_sendq =
402 kif->kf_un.kf_sock.kf_sock_recvq =
408 error = sosockaddr(so, (struct sockaddr *)&ss);
410 ss.ss_len <= sizeof(kif->kf_un.kf_sock.kf_sa_local)) {
411 bcopy(&ss, &kif->kf_un.kf_sock.kf_sa_local, ss.ss_len);
413 ss.ss_len = sizeof(ss);
414 error = sopeeraddr(so, (struct sockaddr *)&ss);
416 ss.ss_len <= sizeof(kif->kf_un.kf_sock.kf_sa_peer)) {
417 bcopy(&ss, &kif->kf_un.kf_sock.kf_sa_peer, ss.ss_len);
419 strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name,
420 sizeof(kif->kf_path));
426 * Use the 'backend3' field in AIO jobs to store the amount of data
427 * completed by the AIO job so far.
429 #define aio_done backend3
431 static STAILQ_HEAD(, task) soaio_jobs;
432 static struct mtx soaio_jobs_lock;
433 static struct task soaio_kproc_task;
434 static int soaio_starting, soaio_idle, soaio_queued;
435 static struct unrhdr *soaio_kproc_unr;
437 static int soaio_max_procs = MAX_AIO_PROCS;
438 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, max_procs, CTLFLAG_RW, &soaio_max_procs, 0,
439 "Maximum number of kernel processes to use for async socket IO");
441 static int soaio_num_procs;
442 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, num_procs, CTLFLAG_RD, &soaio_num_procs, 0,
443 "Number of active kernel processes for async socket IO");
445 static int soaio_target_procs = TARGET_AIO_PROCS;
446 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, target_procs, CTLFLAG_RD,
447 &soaio_target_procs, 0,
448 "Preferred number of ready kernel processes for async socket IO");
450 static int soaio_lifetime;
451 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, lifetime, CTLFLAG_RW, &soaio_lifetime, 0,
452 "Maximum lifetime for idle aiod");
455 soaio_kproc_loop(void *arg)
458 struct vmspace *myvm;
460 int error, id, pending;
465 * Grab an extra reference on the daemon's vmspace so that it
466 * doesn't get freed by jobs that switch to a different
470 myvm = vmspace_acquire_ref(p);
472 mtx_lock(&soaio_jobs_lock);
473 MPASS(soaio_starting > 0);
476 while (!STAILQ_EMPTY(&soaio_jobs)) {
477 task = STAILQ_FIRST(&soaio_jobs);
478 STAILQ_REMOVE_HEAD(&soaio_jobs, ta_link);
480 pending = task->ta_pending;
481 task->ta_pending = 0;
482 mtx_unlock(&soaio_jobs_lock);
484 task->ta_func(task->ta_context, pending);
486 mtx_lock(&soaio_jobs_lock);
488 MPASS(soaio_queued == 0);
490 if (p->p_vmspace != myvm) {
491 mtx_unlock(&soaio_jobs_lock);
492 vmspace_switch_aio(myvm);
493 mtx_lock(&soaio_jobs_lock);
498 error = mtx_sleep(&soaio_idle, &soaio_jobs_lock, 0, "-",
501 if (error == EWOULDBLOCK && STAILQ_EMPTY(&soaio_jobs) &&
502 soaio_num_procs > soaio_target_procs)
506 mtx_unlock(&soaio_jobs_lock);
507 free_unr(soaio_kproc_unr, id);
512 soaio_kproc_create(void *context, int pending)
517 mtx_lock(&soaio_jobs_lock);
519 if (soaio_num_procs < soaio_target_procs) {
521 } else if (soaio_num_procs >= soaio_max_procs) {
523 * Hit the limit on kernel processes, don't
524 * create another one.
527 } else if (soaio_queued <= soaio_idle + soaio_starting) {
529 * No more AIO jobs waiting for a process to be
535 mtx_unlock(&soaio_jobs_lock);
537 id = alloc_unr(soaio_kproc_unr);
538 error = kproc_create(soaio_kproc_loop, (void *)(intptr_t)id,
539 &p, 0, 0, "soaiod%d", id);
541 free_unr(soaio_kproc_unr, id);
542 mtx_lock(&soaio_jobs_lock);
547 mtx_lock(&soaio_jobs_lock);
550 mtx_unlock(&soaio_jobs_lock);
554 soaio_enqueue(struct task *task)
557 mtx_lock(&soaio_jobs_lock);
558 MPASS(task->ta_pending == 0);
560 STAILQ_INSERT_TAIL(&soaio_jobs, task, ta_link);
562 if (soaio_queued <= soaio_idle)
563 wakeup_one(&soaio_idle);
564 else if (soaio_num_procs < soaio_max_procs)
565 taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task);
566 mtx_unlock(&soaio_jobs_lock);
573 soaio_lifetime = AIOD_LIFETIME_DEFAULT;
574 STAILQ_INIT(&soaio_jobs);
575 mtx_init(&soaio_jobs_lock, "soaio jobs", NULL, MTX_DEF);
576 soaio_kproc_unr = new_unrhdr(1, INT_MAX, NULL);
577 TASK_INIT(&soaio_kproc_task, 0, soaio_kproc_create, NULL);
579 SYSINIT(soaio, SI_SUB_VFS, SI_ORDER_ANY, soaio_init, NULL);
582 soaio_ready(struct socket *so, struct sockbuf *sb)
584 return (sb == &so->so_rcv ? soreadable(so) : sowriteable(so));
588 soaio_process_job(struct socket *so, sb_which which, struct kaiocb *job)
590 struct ucred *td_savedcred;
592 struct sockbuf *sb = sobuf(so, which);
594 struct file *fp = job->fd_file;
596 size_t cnt, done, job_total_nbytes __diagused;
600 SOCK_BUF_UNLOCK(so, which);
601 aio_switch_vmspace(job);
604 td_savedcred = td->td_ucred;
605 td->td_ucred = job->cred;
607 job_total_nbytes = job->uiop->uio_resid + job->aio_done;
608 done = job->aio_done;
609 cnt = job->uiop->uio_resid;
610 job->uiop->uio_offset = 0;
611 job->uiop->uio_td = td;
615 * For resource usage accounting, only count a completed request
616 * as a single message to avoid counting multiple calls to
617 * sosend/soreceive on a blocking socket.
620 if (sb == &so->so_rcv) {
621 ru_before = td->td_ru.ru_msgrcv;
623 error = mac_socket_check_receive(fp->f_cred, so);
627 error = soreceive(so, NULL, job->uiop, NULL, NULL,
629 if (td->td_ru.ru_msgrcv != ru_before)
632 if (!TAILQ_EMPTY(&sb->sb_aiojobq))
633 flags |= MSG_MORETOCOME;
634 ru_before = td->td_ru.ru_msgsnd;
636 error = mac_socket_check_send(fp->f_cred, so);
639 error = sousrsend(so, NULL, job->uiop, NULL, flags,
641 if (td->td_ru.ru_msgsnd != ru_before)
645 done += cnt - job->uiop->uio_resid;
646 job->aio_done = done;
647 td->td_ucred = td_savedcred;
649 if (error == EWOULDBLOCK) {
651 * The request was either partially completed or not
652 * completed at all due to racing with a read() or
653 * write() on the socket. If the socket is
654 * non-blocking, return with any partial completion.
655 * If the socket is blocking or if no progress has
656 * been made, requeue this request at the head of the
657 * queue to try again when the socket is ready.
659 MPASS(done != job_total_nbytes);
660 SOCK_BUF_LOCK(so, which);
661 if (done == 0 || !(so->so_state & SS_NBIO)) {
663 if (soaio_ready(so, sb)) {
665 SOCK_BUF_UNLOCK(so, which);
669 if (!aio_set_cancel_function(job, soo_aio_cancel)) {
670 SOCK_BUF_UNLOCK(so, which);
672 aio_complete(job, done, 0);
675 SOCK_BUF_LOCK(so, which);
677 TAILQ_INSERT_HEAD(&sb->sb_aiojobq, job, list);
681 SOCK_BUF_UNLOCK(so, which);
683 if (done != 0 && (error == ERESTART || error == EINTR ||
684 error == EWOULDBLOCK))
687 aio_complete(job, -1, error);
689 aio_complete(job, done, 0);
690 SOCK_BUF_LOCK(so, which);
694 soaio_process_sb(struct socket *so, sb_which which)
697 struct sockbuf *sb = sobuf(so, which);
699 CURVNET_SET(so->so_vnet);
700 SOCK_BUF_LOCK(so, which);
701 while (!TAILQ_EMPTY(&sb->sb_aiojobq) && soaio_ready(so, sb)) {
702 job = TAILQ_FIRST(&sb->sb_aiojobq);
703 TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
704 if (!aio_clear_cancel_function(job))
707 soaio_process_job(so, which, job);
711 * If there are still pending requests, the socket must not be
712 * ready so set SB_AIO to request a wakeup when the socket
715 if (!TAILQ_EMPTY(&sb->sb_aiojobq))
716 sb->sb_flags |= SB_AIO;
717 sb->sb_flags &= ~SB_AIO_RUNNING;
718 SOCK_BUF_UNLOCK(so, which);
725 soaio_rcv(void *context, int pending)
730 soaio_process_sb(so, SO_RCV);
734 soaio_snd(void *context, int pending)
739 soaio_process_sb(so, SO_SND);
743 sowakeup_aio(struct socket *so, sb_which which)
745 struct sockbuf *sb = sobuf(so, which);
747 SOCK_BUF_LOCK_ASSERT(so, which);
749 sb->sb_flags &= ~SB_AIO;
750 if (sb->sb_flags & SB_AIO_RUNNING)
752 sb->sb_flags |= SB_AIO_RUNNING;
754 soaio_enqueue(&sb->sb_aiotask);
758 soo_aio_cancel(struct kaiocb *job)
766 so = job->fd_file->f_data;
767 opcode = job->uaiocb.aio_lio_opcode;
768 if (opcode & LIO_READ) {
772 MPASS(opcode & LIO_WRITE);
777 SOCK_BUF_LOCK(so, which);
778 if (!aio_cancel_cleared(job))
779 TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
780 if (TAILQ_EMPTY(&sb->sb_aiojobq))
781 sb->sb_flags &= ~SB_AIO;
782 SOCK_BUF_UNLOCK(so, which);
784 done = job->aio_done;
786 aio_complete(job, done, 0);
792 soo_aio_queue(struct file *fp, struct kaiocb *job)
800 error = so->so_proto->pr_aio_queue(so, job);
804 /* Lock through the socket, since this may be a listening socket. */
805 switch (job->uaiocb.aio_lio_opcode & (LIO_WRITE | LIO_READ)) {
807 SOCK_RECVBUF_LOCK(so);
812 SOCK_SENDBUF_LOCK(so);
820 if (SOLISTENING(so)) {
821 SOCK_BUF_UNLOCK(so, which);
825 if (!aio_set_cancel_function(job, soo_aio_cancel))
826 panic("new job was cancelled");
827 TAILQ_INSERT_TAIL(&sb->sb_aiojobq, job, list);
828 if (!(sb->sb_flags & SB_AIO_RUNNING)) {
829 if (soaio_ready(so, sb))
830 sowakeup_aio(so, which);
832 sb->sb_flags |= SB_AIO;
834 SOCK_BUF_UNLOCK(so, which);