2 * Copyright (c) 1982, 1986, 1990, 1993
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * @(#)sys_socket.c 8.1 (Berkeley) 6/10/93
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
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, 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 *(int *)data = sbavail(&so->so_rcv);
213 *(int *)data = sbavail(&so->so_snd);
218 if ((so->so_snd.sb_hiwat < sbused(&so->so_snd)) ||
219 (so->so_snd.sb_mbmax < so->so_snd.sb_mbcnt))
222 *(int *)data = sbspace(&so->so_snd);
226 error = fsetown(*(int *)data, &so->so_sigio);
230 *(int *)data = fgetown(&so->so_sigio);
234 error = fsetown(-(*(int *)data), &so->so_sigio);
238 *(int *)data = -fgetown(&so->so_sigio);
243 *(int *)data = (so->so_rcv.sb_state & SBS_RCVATMARK) != 0;
247 * Interface/routing/protocol specific ioctls: interface and
248 * routing ioctls should have a different entry since a
249 * socket is unnecessary.
251 if (IOCGROUP(cmd) == 'i')
252 error = ifioctl(so, cmd, data, td);
253 else if (IOCGROUP(cmd) == 'r') {
254 CURVNET_SET(so->so_vnet);
255 error = rtioctl_fib(cmd, data, so->so_fibnum);
258 CURVNET_SET(so->so_vnet);
259 error = ((*so->so_proto->pr_usrreqs->pru_control)
260 (so, cmd, data, 0, td));
269 soo_poll(struct file *fp, int events, struct ucred *active_cred,
272 struct socket *so = fp->f_data;
276 error = mac_socket_check_poll(active_cred, so);
280 return (sopoll(so, events, fp->f_cred, td));
284 soo_stat(struct file *fp, struct stat *ub, struct ucred *active_cred,
287 struct socket *so = fp->f_data;
292 bzero((caddr_t)ub, sizeof (*ub));
293 ub->st_mode = S_IFSOCK;
295 error = mac_socket_check_stat(active_cred, so);
299 if (!SOLISTENING(so)) {
303 * If SBS_CANTRCVMORE is set, but there's still data left
304 * in the receive buffer, the socket is still readable.
308 if ((sb->sb_state & SBS_CANTRCVMORE) == 0 || sbavail(sb))
309 ub->st_mode |= S_IRUSR | S_IRGRP | S_IROTH;
310 ub->st_size = sbavail(sb) - sb->sb_ctl;
315 if ((sb->sb_state & SBS_CANTSENDMORE) == 0)
316 ub->st_mode |= S_IWUSR | S_IWGRP | S_IWOTH;
319 ub->st_uid = so->so_cred->cr_uid;
320 ub->st_gid = so->so_cred->cr_gid;
321 return (*so->so_proto->pr_usrreqs->pru_sense)(so, ub);
325 * API socket close on file pointer. We call soclose() to close the socket
326 * (including initiating closing protocols). soclose() will sorele() the
327 * file reference but the actual socket will not go away until the socket's
331 soo_close(struct file *fp, struct thread *td)
337 fp->f_ops = &badfileops;
346 soo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
354 kif->kf_type = KF_TYPE_SOCKET;
356 kif->kf_un.kf_sock.kf_sock_domain0 =
357 so->so_proto->pr_domain->dom_family;
358 kif->kf_un.kf_sock.kf_sock_type0 = so->so_type;
359 kif->kf_un.kf_sock.kf_sock_protocol0 = so->so_proto->pr_protocol;
360 kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb;
361 switch (kif->kf_un.kf_sock.kf_sock_domain0) {
364 if (kif->kf_un.kf_sock.kf_sock_protocol0 == IPPROTO_TCP) {
365 if (so->so_pcb != NULL) {
366 inpcb = (struct inpcb *)(so->so_pcb);
367 kif->kf_un.kf_sock.kf_sock_inpcb =
368 (uintptr_t)inpcb->inp_ppcb;
369 kif->kf_un.kf_sock.kf_sock_sendq =
371 kif->kf_un.kf_sock.kf_sock_recvq =
377 if (so->so_pcb != NULL) {
378 unpcb = (struct unpcb *)(so->so_pcb);
379 if (unpcb->unp_conn) {
380 kif->kf_un.kf_sock.kf_sock_unpconn =
381 (uintptr_t)unpcb->unp_conn;
382 kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
384 kif->kf_un.kf_sock.kf_sock_snd_sb_state =
386 kif->kf_un.kf_sock.kf_sock_sendq =
388 kif->kf_un.kf_sock.kf_sock_recvq =
394 error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa);
396 sa->sa_len <= sizeof(kif->kf_un.kf_sock.kf_sa_local)) {
397 bcopy(sa, &kif->kf_un.kf_sock.kf_sa_local, sa->sa_len);
400 error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa);
402 sa->sa_len <= sizeof(kif->kf_un.kf_sock.kf_sa_peer)) {
403 bcopy(sa, &kif->kf_un.kf_sock.kf_sa_peer, sa->sa_len);
406 strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name,
407 sizeof(kif->kf_path));
412 * Use the 'backend3' field in AIO jobs to store the amount of data
413 * completed by the AIO job so far.
415 #define aio_done backend3
417 static STAILQ_HEAD(, task) soaio_jobs;
418 static struct mtx soaio_jobs_lock;
419 static struct task soaio_kproc_task;
420 static int soaio_starting, soaio_idle, soaio_queued;
421 static struct unrhdr *soaio_kproc_unr;
423 static int soaio_max_procs = MAX_AIO_PROCS;
424 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, max_procs, CTLFLAG_RW, &soaio_max_procs, 0,
425 "Maximum number of kernel processes to use for async socket IO");
427 static int soaio_num_procs;
428 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, num_procs, CTLFLAG_RD, &soaio_num_procs, 0,
429 "Number of active kernel processes for async socket IO");
431 static int soaio_target_procs = TARGET_AIO_PROCS;
432 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, target_procs, CTLFLAG_RD,
433 &soaio_target_procs, 0,
434 "Preferred number of ready kernel processes for async socket IO");
436 static int soaio_lifetime;
437 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, lifetime, CTLFLAG_RW, &soaio_lifetime, 0,
438 "Maximum lifetime for idle aiod");
441 soaio_kproc_loop(void *arg)
444 struct vmspace *myvm;
446 int error, id, pending;
451 * Grab an extra reference on the daemon's vmspace so that it
452 * doesn't get freed by jobs that switch to a different
456 myvm = vmspace_acquire_ref(p);
458 mtx_lock(&soaio_jobs_lock);
459 MPASS(soaio_starting > 0);
462 while (!STAILQ_EMPTY(&soaio_jobs)) {
463 task = STAILQ_FIRST(&soaio_jobs);
464 STAILQ_REMOVE_HEAD(&soaio_jobs, ta_link);
466 pending = task->ta_pending;
467 task->ta_pending = 0;
468 mtx_unlock(&soaio_jobs_lock);
470 task->ta_func(task->ta_context, pending);
472 mtx_lock(&soaio_jobs_lock);
474 MPASS(soaio_queued == 0);
476 if (p->p_vmspace != myvm) {
477 mtx_unlock(&soaio_jobs_lock);
478 vmspace_switch_aio(myvm);
479 mtx_lock(&soaio_jobs_lock);
484 error = mtx_sleep(&soaio_idle, &soaio_jobs_lock, 0, "-",
487 if (error == EWOULDBLOCK && STAILQ_EMPTY(&soaio_jobs) &&
488 soaio_num_procs > soaio_target_procs)
492 mtx_unlock(&soaio_jobs_lock);
493 free_unr(soaio_kproc_unr, id);
498 soaio_kproc_create(void *context, int pending)
503 mtx_lock(&soaio_jobs_lock);
505 if (soaio_num_procs < soaio_target_procs) {
507 } else if (soaio_num_procs >= soaio_max_procs) {
509 * Hit the limit on kernel processes, don't
510 * create another one.
513 } else if (soaio_queued <= soaio_idle + soaio_starting) {
515 * No more AIO jobs waiting for a process to be
521 mtx_unlock(&soaio_jobs_lock);
523 id = alloc_unr(soaio_kproc_unr);
524 error = kproc_create(soaio_kproc_loop, (void *)(intptr_t)id,
525 &p, 0, 0, "soaiod%d", id);
527 free_unr(soaio_kproc_unr, id);
528 mtx_lock(&soaio_jobs_lock);
533 mtx_lock(&soaio_jobs_lock);
536 mtx_unlock(&soaio_jobs_lock);
540 soaio_enqueue(struct task *task)
543 mtx_lock(&soaio_jobs_lock);
544 MPASS(task->ta_pending == 0);
546 STAILQ_INSERT_TAIL(&soaio_jobs, task, ta_link);
548 if (soaio_queued <= soaio_idle)
549 wakeup_one(&soaio_idle);
550 else if (soaio_num_procs < soaio_max_procs)
551 taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task);
552 mtx_unlock(&soaio_jobs_lock);
559 soaio_lifetime = AIOD_LIFETIME_DEFAULT;
560 STAILQ_INIT(&soaio_jobs);
561 mtx_init(&soaio_jobs_lock, "soaio jobs", NULL, MTX_DEF);
562 soaio_kproc_unr = new_unrhdr(1, INT_MAX, NULL);
563 TASK_INIT(&soaio_kproc_task, 0, soaio_kproc_create, NULL);
564 if (soaio_target_procs > 0)
565 taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task);
567 SYSINIT(soaio, SI_SUB_VFS, SI_ORDER_ANY, soaio_init, NULL);
570 soaio_ready(struct socket *so, struct sockbuf *sb)
572 return (sb == &so->so_rcv ? soreadable(so) : sowriteable(so));
576 soaio_process_job(struct socket *so, struct sockbuf *sb, struct kaiocb *job)
578 struct ucred *td_savedcred;
588 aio_switch_vmspace(job);
592 td_savedcred = td->td_ucred;
593 td->td_ucred = job->cred;
595 done = job->aio_done;
596 cnt = job->uaiocb.aio_nbytes - done;
597 iov.iov_base = (void *)((uintptr_t)job->uaiocb.aio_buf + done);
603 uio.uio_segflg = UIO_USERSPACE;
608 * For resource usage accounting, only count a completed request
609 * as a single message to avoid counting multiple calls to
610 * sosend/soreceive on a blocking socket.
613 if (sb == &so->so_rcv) {
614 uio.uio_rw = UIO_READ;
615 ru_before = td->td_ru.ru_msgrcv;
617 error = mac_socket_check_receive(fp->f_cred, so);
621 error = soreceive(so, NULL, &uio, NULL, NULL, &flags);
622 if (td->td_ru.ru_msgrcv != ru_before)
625 if (!TAILQ_EMPTY(&sb->sb_aiojobq))
626 flags |= MSG_MORETOCOME;
627 uio.uio_rw = UIO_WRITE;
628 ru_before = td->td_ru.ru_msgsnd;
630 error = mac_socket_check_send(fp->f_cred, so);
633 error = sosend(so, NULL, &uio, NULL, NULL, flags, td);
634 if (td->td_ru.ru_msgsnd != ru_before)
636 if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0) {
637 PROC_LOCK(job->userproc);
638 kern_psignal(job->userproc, SIGPIPE);
639 PROC_UNLOCK(job->userproc);
643 done += cnt - uio.uio_resid;
644 job->aio_done = done;
645 td->td_ucred = td_savedcred;
647 if (error == EWOULDBLOCK) {
649 * The request was either partially completed or not
650 * completed at all due to racing with a read() or
651 * write() on the socket. If the socket is
652 * non-blocking, return with any partial completion.
653 * If the socket is blocking or if no progress has
654 * been made, requeue this request at the head of the
655 * queue to try again when the socket is ready.
657 MPASS(done != job->uaiocb.aio_nbytes);
659 if (done == 0 || !(so->so_state & SS_NBIO)) {
661 if (soaio_ready(so, sb)) {
667 if (!aio_set_cancel_function(job, soo_aio_cancel)) {
670 aio_complete(job, done, 0);
675 TAILQ_INSERT_HEAD(&sb->sb_aiojobq, job, list);
681 if (done != 0 && (error == ERESTART || error == EINTR ||
682 error == EWOULDBLOCK))
685 aio_complete(job, -1, error);
687 aio_complete(job, done, 0);
692 soaio_process_sb(struct socket *so, struct sockbuf *sb)
696 CURVNET_SET(so->so_vnet);
698 while (!TAILQ_EMPTY(&sb->sb_aiojobq) && soaio_ready(so, sb)) {
699 job = TAILQ_FIRST(&sb->sb_aiojobq);
700 TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
701 if (!aio_clear_cancel_function(job))
704 soaio_process_job(so, sb, job);
708 * If there are still pending requests, the socket must not be
709 * ready so set SB_AIO to request a wakeup when the socket
712 if (!TAILQ_EMPTY(&sb->sb_aiojobq))
713 sb->sb_flags |= SB_AIO;
714 sb->sb_flags &= ~SB_AIO_RUNNING;
723 soaio_rcv(void *context, int pending)
728 soaio_process_sb(so, &so->so_rcv);
732 soaio_snd(void *context, int pending)
737 soaio_process_sb(so, &so->so_snd);
741 sowakeup_aio(struct socket *so, struct sockbuf *sb)
744 SOCKBUF_LOCK_ASSERT(sb);
745 sb->sb_flags &= ~SB_AIO;
746 if (sb->sb_flags & SB_AIO_RUNNING)
748 sb->sb_flags |= SB_AIO_RUNNING;
749 if (sb == &so->so_snd)
752 if (sb == &so->so_snd)
754 soaio_enqueue(&sb->sb_aiotask);
758 soo_aio_cancel(struct kaiocb *job)
765 so = job->fd_file->f_data;
766 opcode = job->uaiocb.aio_lio_opcode;
767 if (opcode == LIO_READ)
770 MPASS(opcode == LIO_WRITE);
775 if (!aio_cancel_cleared(job))
776 TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
777 if (TAILQ_EMPTY(&sb->sb_aiojobq))
778 sb->sb_flags &= ~SB_AIO;
781 done = job->aio_done;
783 aio_complete(job, done, 0);
789 soo_aio_queue(struct file *fp, struct kaiocb *job)
796 error = (*so->so_proto->pr_usrreqs->pru_aio_queue)(so, job);
800 switch (job->uaiocb.aio_lio_opcode) {
812 if (!aio_set_cancel_function(job, soo_aio_cancel))
813 panic("new job was cancelled");
814 TAILQ_INSERT_TAIL(&sb->sb_aiojobq, job, list);
815 if (!(sb->sb_flags & SB_AIO_RUNNING)) {
816 if (soaio_ready(so, sb))
817 sowakeup_aio(so, sb);
819 sb->sb_flags |= SB_AIO;