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 * 4. 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;
174 * XXXRW: This code separately acquires SOCK_LOCK(so) and
175 * SOCKBUF_LOCK(&so->so_rcv) even though they are the same
176 * mutex to avoid introducing the assumption that they are
181 so->so_state |= SS_ASYNC;
183 SOCKBUF_LOCK(&so->so_rcv);
184 so->so_rcv.sb_flags |= SB_ASYNC;
185 SOCKBUF_UNLOCK(&so->so_rcv);
186 SOCKBUF_LOCK(&so->so_snd);
187 so->so_snd.sb_flags |= SB_ASYNC;
188 SOCKBUF_UNLOCK(&so->so_snd);
191 so->so_state &= ~SS_ASYNC;
193 SOCKBUF_LOCK(&so->so_rcv);
194 so->so_rcv.sb_flags &= ~SB_ASYNC;
195 SOCKBUF_UNLOCK(&so->so_rcv);
196 SOCKBUF_LOCK(&so->so_snd);
197 so->so_snd.sb_flags &= ~SB_ASYNC;
198 SOCKBUF_UNLOCK(&so->so_snd);
204 *(int *)data = sbavail(&so->so_rcv);
209 *(int *)data = sbavail(&so->so_snd);
214 if ((so->so_snd.sb_hiwat < sbused(&so->so_snd)) ||
215 (so->so_snd.sb_mbmax < so->so_snd.sb_mbcnt))
218 *(int *)data = sbspace(&so->so_snd);
222 error = fsetown(*(int *)data, &so->so_sigio);
226 *(int *)data = fgetown(&so->so_sigio);
230 error = fsetown(-(*(int *)data), &so->so_sigio);
234 *(int *)data = -fgetown(&so->so_sigio);
239 *(int *)data = (so->so_rcv.sb_state & SBS_RCVATMARK) != 0;
243 * Interface/routing/protocol specific ioctls: interface and
244 * routing ioctls should have a different entry since a
245 * socket is unnecessary.
247 if (IOCGROUP(cmd) == 'i')
248 error = ifioctl(so, cmd, data, td);
249 else if (IOCGROUP(cmd) == 'r') {
250 CURVNET_SET(so->so_vnet);
251 error = rtioctl_fib(cmd, data, so->so_fibnum);
254 CURVNET_SET(so->so_vnet);
255 error = ((*so->so_proto->pr_usrreqs->pru_control)
256 (so, cmd, data, 0, td));
265 soo_poll(struct file *fp, int events, struct ucred *active_cred,
268 struct socket *so = fp->f_data;
272 error = mac_socket_check_poll(active_cred, so);
276 return (sopoll(so, events, fp->f_cred, td));
280 soo_stat(struct file *fp, struct stat *ub, struct ucred *active_cred,
283 struct socket *so = fp->f_data;
289 bzero((caddr_t)ub, sizeof (*ub));
290 ub->st_mode = S_IFSOCK;
292 error = mac_socket_check_stat(active_cred, so);
297 * If SBS_CANTRCVMORE is set, but there's still data left in the
298 * receive buffer, the socket is still readable.
302 if ((sb->sb_state & SBS_CANTRCVMORE) == 0 || sbavail(sb))
303 ub->st_mode |= S_IRUSR | S_IRGRP | S_IROTH;
304 ub->st_size = sbavail(sb) - sb->sb_ctl;
309 if ((sb->sb_state & SBS_CANTSENDMORE) == 0)
310 ub->st_mode |= S_IWUSR | S_IWGRP | S_IWOTH;
312 ub->st_uid = so->so_cred->cr_uid;
313 ub->st_gid = so->so_cred->cr_gid;
314 return (*so->so_proto->pr_usrreqs->pru_sense)(so, ub);
318 * API socket close on file pointer. We call soclose() to close the socket
319 * (including initiating closing protocols). soclose() will sorele() the
320 * file reference but the actual socket will not go away until the socket's
324 soo_close(struct file *fp, struct thread *td)
330 fp->f_ops = &badfileops;
339 soo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
347 kif->kf_type = KF_TYPE_SOCKET;
349 CURVNET_SET(so->so_vnet);
350 kif->kf_sock_domain = so->so_proto->pr_domain->dom_family;
351 kif->kf_sock_type = so->so_type;
352 kif->kf_sock_protocol = so->so_proto->pr_protocol;
353 kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb;
354 switch (kif->kf_sock_domain) {
357 if (kif->kf_sock_protocol == IPPROTO_TCP) {
358 if (so->so_pcb != NULL) {
359 inpcb = (struct inpcb *)(so->so_pcb);
360 kif->kf_un.kf_sock.kf_sock_inpcb =
361 (uintptr_t)inpcb->inp_ppcb;
366 if (so->so_pcb != NULL) {
367 unpcb = (struct unpcb *)(so->so_pcb);
368 if (unpcb->unp_conn) {
369 kif->kf_un.kf_sock.kf_sock_unpconn =
370 (uintptr_t)unpcb->unp_conn;
371 kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
373 kif->kf_un.kf_sock.kf_sock_snd_sb_state =
379 error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa);
380 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
381 bcopy(sa, &kif->kf_sa_local, sa->sa_len);
384 error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa);
385 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
386 bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
389 strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name,
390 sizeof(kif->kf_path));
396 * Use the 'backend3' field in AIO jobs to store the amount of data
397 * completed by the AIO job so far.
399 #define aio_done backend3
401 static STAILQ_HEAD(, task) soaio_jobs;
402 static struct mtx soaio_jobs_lock;
403 static struct task soaio_kproc_task;
404 static int soaio_starting, soaio_idle, soaio_queued;
405 static struct unrhdr *soaio_kproc_unr;
407 static int soaio_max_procs = MAX_AIO_PROCS;
408 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, max_procs, CTLFLAG_RW, &soaio_max_procs, 0,
409 "Maximum number of kernel processes to use for async socket IO");
411 static int soaio_num_procs;
412 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, num_procs, CTLFLAG_RD, &soaio_num_procs, 0,
413 "Number of active kernel processes for async socket IO");
415 static int soaio_target_procs = TARGET_AIO_PROCS;
416 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, target_procs, CTLFLAG_RD,
417 &soaio_target_procs, 0,
418 "Preferred number of ready kernel processes for async socket IO");
420 static int soaio_lifetime;
421 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, lifetime, CTLFLAG_RW, &soaio_lifetime, 0,
422 "Maximum lifetime for idle aiod");
425 soaio_kproc_loop(void *arg)
428 struct vmspace *myvm;
430 int error, id, pending;
435 * Grab an extra reference on the daemon's vmspace so that it
436 * doesn't get freed by jobs that switch to a different
440 myvm = vmspace_acquire_ref(p);
442 mtx_lock(&soaio_jobs_lock);
443 MPASS(soaio_starting > 0);
446 while (!STAILQ_EMPTY(&soaio_jobs)) {
447 task = STAILQ_FIRST(&soaio_jobs);
448 STAILQ_REMOVE_HEAD(&soaio_jobs, ta_link);
450 pending = task->ta_pending;
451 task->ta_pending = 0;
452 mtx_unlock(&soaio_jobs_lock);
454 task->ta_func(task->ta_context, pending);
456 mtx_lock(&soaio_jobs_lock);
458 MPASS(soaio_queued == 0);
460 if (p->p_vmspace != myvm) {
461 mtx_unlock(&soaio_jobs_lock);
462 vmspace_switch_aio(myvm);
463 mtx_lock(&soaio_jobs_lock);
468 error = mtx_sleep(&soaio_idle, &soaio_jobs_lock, 0, "-",
471 if (error == EWOULDBLOCK && STAILQ_EMPTY(&soaio_jobs) &&
472 soaio_num_procs > soaio_target_procs)
476 mtx_unlock(&soaio_jobs_lock);
477 free_unr(soaio_kproc_unr, id);
482 soaio_kproc_create(void *context, int pending)
487 mtx_lock(&soaio_jobs_lock);
489 if (soaio_num_procs < soaio_target_procs) {
491 } else if (soaio_num_procs >= soaio_max_procs) {
493 * Hit the limit on kernel processes, don't
494 * create another one.
497 } else if (soaio_queued <= soaio_idle + soaio_starting) {
499 * No more AIO jobs waiting for a process to be
505 mtx_unlock(&soaio_jobs_lock);
507 id = alloc_unr(soaio_kproc_unr);
508 error = kproc_create(soaio_kproc_loop, (void *)(intptr_t)id,
509 &p, 0, 0, "soaiod%d", id);
511 free_unr(soaio_kproc_unr, id);
512 mtx_lock(&soaio_jobs_lock);
517 mtx_lock(&soaio_jobs_lock);
520 mtx_unlock(&soaio_jobs_lock);
524 soaio_enqueue(struct task *task)
527 mtx_lock(&soaio_jobs_lock);
528 MPASS(task->ta_pending == 0);
530 STAILQ_INSERT_TAIL(&soaio_jobs, task, ta_link);
532 if (soaio_queued <= soaio_idle)
533 wakeup_one(&soaio_idle);
534 else if (soaio_num_procs < soaio_max_procs)
535 taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task);
536 mtx_unlock(&soaio_jobs_lock);
543 soaio_lifetime = AIOD_LIFETIME_DEFAULT;
544 STAILQ_INIT(&soaio_jobs);
545 mtx_init(&soaio_jobs_lock, "soaio jobs", NULL, MTX_DEF);
546 soaio_kproc_unr = new_unrhdr(1, INT_MAX, NULL);
547 TASK_INIT(&soaio_kproc_task, 0, soaio_kproc_create, NULL);
548 if (soaio_target_procs > 0)
549 taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task);
551 SYSINIT(soaio, SI_SUB_VFS, SI_ORDER_ANY, soaio_init, NULL);
554 soaio_ready(struct socket *so, struct sockbuf *sb)
556 return (sb == &so->so_rcv ? soreadable(so) : sowriteable(so));
560 soaio_process_job(struct socket *so, struct sockbuf *sb, struct kaiocb *job)
562 struct ucred *td_savedcred;
572 aio_switch_vmspace(job);
576 td_savedcred = td->td_ucred;
577 td->td_ucred = job->cred;
579 done = job->aio_done;
580 cnt = job->uaiocb.aio_nbytes - done;
581 iov.iov_base = (void *)((uintptr_t)job->uaiocb.aio_buf + done);
587 uio.uio_segflg = UIO_USERSPACE;
592 * For resource usage accounting, only count a completed request
593 * as a single message to avoid counting multiple calls to
594 * sosend/soreceive on a blocking socket.
597 if (sb == &so->so_rcv) {
598 uio.uio_rw = UIO_READ;
599 ru_before = td->td_ru.ru_msgrcv;
601 error = mac_socket_check_receive(fp->f_cred, so);
605 error = soreceive(so, NULL, &uio, NULL, NULL, &flags);
606 if (td->td_ru.ru_msgrcv != ru_before)
609 if (!TAILQ_EMPTY(&sb->sb_aiojobq))
610 flags |= MSG_MORETOCOME;
611 uio.uio_rw = UIO_WRITE;
612 ru_before = td->td_ru.ru_msgsnd;
614 error = mac_socket_check_send(fp->f_cred, so);
617 error = sosend(so, NULL, &uio, NULL, NULL, flags, td);
618 if (td->td_ru.ru_msgsnd != ru_before)
620 if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0) {
621 PROC_LOCK(job->userproc);
622 kern_psignal(job->userproc, SIGPIPE);
623 PROC_UNLOCK(job->userproc);
627 done += cnt - uio.uio_resid;
628 job->aio_done = done;
629 td->td_ucred = td_savedcred;
631 if (error == EWOULDBLOCK) {
633 * The request was either partially completed or not
634 * completed at all due to racing with a read() or
635 * write() on the socket. If the socket is
636 * non-blocking, return with any partial completion.
637 * If the socket is blocking or if no progress has
638 * been made, requeue this request at the head of the
639 * queue to try again when the socket is ready.
641 MPASS(done != job->uaiocb.aio_nbytes);
643 if (done == 0 || !(so->so_state & SS_NBIO)) {
645 if (soaio_ready(so, sb)) {
651 if (!aio_set_cancel_function(job, soo_aio_cancel)) {
654 aio_complete(job, done, 0);
659 TAILQ_INSERT_HEAD(&sb->sb_aiojobq, job, list);
665 if (done != 0 && (error == ERESTART || error == EINTR ||
666 error == EWOULDBLOCK))
669 aio_complete(job, -1, error);
671 aio_complete(job, done, 0);
676 soaio_process_sb(struct socket *so, struct sockbuf *sb)
680 CURVNET_SET(so->so_vnet);
682 while (!TAILQ_EMPTY(&sb->sb_aiojobq) && soaio_ready(so, sb)) {
683 job = TAILQ_FIRST(&sb->sb_aiojobq);
684 TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
685 if (!aio_clear_cancel_function(job))
688 soaio_process_job(so, sb, job);
692 * If there are still pending requests, the socket must not be
693 * ready so set SB_AIO to request a wakeup when the socket
696 if (!TAILQ_EMPTY(&sb->sb_aiojobq))
697 sb->sb_flags |= SB_AIO;
698 sb->sb_flags &= ~SB_AIO_RUNNING;
708 soaio_rcv(void *context, int pending)
713 soaio_process_sb(so, &so->so_rcv);
717 soaio_snd(void *context, int pending)
722 soaio_process_sb(so, &so->so_snd);
726 sowakeup_aio(struct socket *so, struct sockbuf *sb)
729 SOCKBUF_LOCK_ASSERT(sb);
730 sb->sb_flags &= ~SB_AIO;
731 if (sb->sb_flags & SB_AIO_RUNNING)
733 sb->sb_flags |= SB_AIO_RUNNING;
734 if (sb == &so->so_snd)
737 if (sb == &so->so_snd)
739 soaio_enqueue(&sb->sb_aiotask);
743 soo_aio_cancel(struct kaiocb *job)
750 so = job->fd_file->f_data;
751 opcode = job->uaiocb.aio_lio_opcode;
752 if (opcode == LIO_READ)
755 MPASS(opcode == LIO_WRITE);
760 if (!aio_cancel_cleared(job))
761 TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
762 if (TAILQ_EMPTY(&sb->sb_aiojobq))
763 sb->sb_flags &= ~SB_AIO;
766 done = job->aio_done;
768 aio_complete(job, done, 0);
774 soo_aio_queue(struct file *fp, struct kaiocb *job)
781 error = (*so->so_proto->pr_usrreqs->pru_aio_queue)(so, job);
785 switch (job->uaiocb.aio_lio_opcode) {
797 if (!aio_set_cancel_function(job, soo_aio_cancel))
798 panic("new job was cancelled");
799 TAILQ_INSERT_TAIL(&sb->sb_aiojobq, job, list);
800 if (!(sb->sb_flags & SB_AIO_RUNNING)) {
801 if (soaio_ready(so, sb))
802 sowakeup_aio(so, sb);
804 sb->sb_flags |= SB_AIO;