FreeBSD/Linux Kernel Cross Reference
sys/kern/sys_socket.c
1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1982, 1986, 1990, 1993
5 * The Regents of the University of California. All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
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.
18 *
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
29 * SUCH DAMAGE.
30 *
31 * @(#)sys_socket.c 8.1 (Berkeley) 6/10/93
32 */
33
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/aio.h>
40 #include <sys/domain.h>
41 #include <sys/file.h>
42 #include <sys/filedesc.h>
43 #include <sys/kernel.h>
44 #include <sys/kthread.h>
45 #include <sys/malloc.h>
46 #include <sys/proc.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>
55 #include <sys/stat.h>
56 #include <sys/sysctl.h>
57 #include <sys/sysproto.h>
58 #include <sys/taskqueue.h>
59 #include <sys/uio.h>
60 #include <sys/ucred.h>
61 #include <sys/un.h>
62 #include <sys/unpcb.h>
63 #include <sys/user.h>
64
65 #include <net/if.h>
66 #include <net/if_var.h>
67 #include <net/route.h>
68 #include <net/vnet.h>
69
70 #include <netinet/in.h>
71 #include <netinet/in_pcb.h>
72
73 #include <security/mac/mac_framework.h>
74
75 #include <vm/vm.h>
76 #include <vm/pmap.h>
77 #include <vm/vm_extern.h>
78 #include <vm/vm_map.h>
79
80 static SYSCTL_NODE(_kern_ipc, OID_AUTO, aio, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
81 "socket AIO stats");
82
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");
86
87 static int empty_retries;
88 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, empty_retries, CTLFLAG_RD, &empty_retries,
89 0, "socket operation retries");
90
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;
100
101 static void soo_aio_cancel(struct kaiocb *job);
102
103 struct fileops socketops = {
104 .fo_read = soo_read,
105 .fo_write = soo_write,
106 .fo_truncate = invfo_truncate,
107 .fo_ioctl = soo_ioctl,
108 .fo_poll = soo_poll,
109 .fo_kqfilter = soo_kqfilter,
110 .fo_stat = soo_stat,
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
118 };
119
120 static int
121 soo_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
122 int flags, struct thread *td)
123 {
124 struct socket *so = fp->f_data;
125 int error;
126
127 #ifdef MAC
128 error = mac_socket_check_receive(active_cred, so);
129 if (error)
130 return (error);
131 #endif
132 error = soreceive(so, 0, uio, 0, 0, 0);
133 return (error);
134 }
135
136 static int
137 soo_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
138 int flags, struct thread *td)
139 {
140 struct socket *so = fp->f_data;
141 int error;
142
143 #ifdef MAC
144 error = mac_socket_check_send(active_cred, so);
145 if (error)
146 return (error);
147 #endif
148 error = sosend(so, 0, uio, 0, 0, 0, uio->uio_td);
149 if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0) {
150 PROC_LOCK(uio->uio_td->td_proc);
151 tdsignal(uio->uio_td, SIGPIPE);
152 PROC_UNLOCK(uio->uio_td->td_proc);
153 }
154 return (error);
155 }
156
157 static int
158 soo_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *active_cred,
159 struct thread *td)
160 {
161 struct socket *so = fp->f_data;
162 int error = 0;
163
164 switch (cmd) {
165 case FIONBIO:
166 SOCK_LOCK(so);
167 if (*(int *)data)
168 so->so_state |= SS_NBIO;
169 else
170 so->so_state &= ~SS_NBIO;
171 SOCK_UNLOCK(so);
172 break;
173
174 case FIOASYNC:
175 if (*(int *)data) {
176 SOCK_LOCK(so);
177 so->so_state |= SS_ASYNC;
178 if (SOLISTENING(so)) {
179 so->sol_sbrcv_flags |= SB_ASYNC;
180 so->sol_sbsnd_flags |= SB_ASYNC;
181 } else {
182 SOCKBUF_LOCK(&so->so_rcv);
183 so->so_rcv.sb_flags |= SB_ASYNC;
184 SOCKBUF_UNLOCK(&so->so_rcv);
185 SOCKBUF_LOCK(&so->so_snd);
186 so->so_snd.sb_flags |= SB_ASYNC;
187 SOCKBUF_UNLOCK(&so->so_snd);
188 }
189 SOCK_UNLOCK(so);
190 } else {
191 SOCK_LOCK(so);
192 so->so_state &= ~SS_ASYNC;
193 if (SOLISTENING(so)) {
194 so->sol_sbrcv_flags &= ~SB_ASYNC;
195 so->sol_sbsnd_flags &= ~SB_ASYNC;
196 } else {
197 SOCKBUF_LOCK(&so->so_rcv);
198 so->so_rcv.sb_flags &= ~SB_ASYNC;
199 SOCKBUF_UNLOCK(&so->so_rcv);
200 SOCKBUF_LOCK(&so->so_snd);
201 so->so_snd.sb_flags &= ~SB_ASYNC;
202 SOCKBUF_UNLOCK(&so->so_snd);
203 }
204 SOCK_UNLOCK(so);
205 }
206 break;
207
208 case FIONREAD:
209 /* Unlocked read. */
210 if (SOLISTENING(so)) {
211 error = EINVAL;
212 } else {
213 *(int *)data = sbavail(&so->so_rcv);
214 }
215 break;
216
217 case FIONWRITE:
218 /* Unlocked read. */
219 if (SOLISTENING(so)) {
220 error = EINVAL;
221 } else {
222 *(int *)data = sbavail(&so->so_snd);
223 }
224 break;
225
226 case FIONSPACE:
227 /* Unlocked read. */
228 if (SOLISTENING(so)) {
229 error = EINVAL;
230 } else {
231 if ((so->so_snd.sb_hiwat < sbused(&so->so_snd)) ||
232 (so->so_snd.sb_mbmax < so->so_snd.sb_mbcnt)) {
233 *(int *)data = 0;
234 } else {
235 *(int *)data = sbspace(&so->so_snd);
236 }
237 }
238 break;
239
240 case FIOSETOWN:
241 error = fsetown(*(int *)data, &so->so_sigio);
242 break;
243
244 case FIOGETOWN:
245 *(int *)data = fgetown(&so->so_sigio);
246 break;
247
248 case SIOCSPGRP:
249 error = fsetown(-(*(int *)data), &so->so_sigio);
250 break;
251
252 case SIOCGPGRP:
253 *(int *)data = -fgetown(&so->so_sigio);
254 break;
255
256 case SIOCATMARK:
257 /* Unlocked read. */
258 if (SOLISTENING(so)) {
259 error = EINVAL;
260 } else {
261 *(int *)data = (so->so_rcv.sb_state & SBS_RCVATMARK) != 0;
262 }
263 break;
264 default:
265 /*
266 * Interface/routing/protocol specific ioctls: interface and
267 * routing ioctls should have a different entry since a
268 * socket is unnecessary.
269 */
270 if (IOCGROUP(cmd) == 'i')
271 error = ifioctl(so, cmd, data, td);
272 else if (IOCGROUP(cmd) == 'r') {
273 CURVNET_SET(so->so_vnet);
274 error = rtioctl_fib(cmd, data, so->so_fibnum);
275 CURVNET_RESTORE();
276 } else {
277 CURVNET_SET(so->so_vnet);
278 error = ((*so->so_proto->pr_usrreqs->pru_control)
279 (so, cmd, data, 0, td));
280 CURVNET_RESTORE();
281 }
282 break;
283 }
284 return (error);
285 }
286
287 static int
288 soo_poll(struct file *fp, int events, struct ucred *active_cred,
289 struct thread *td)
290 {
291 struct socket *so = fp->f_data;
292 #ifdef MAC
293 int error;
294
295 error = mac_socket_check_poll(active_cred, so);
296 if (error)
297 return (error);
298 #endif
299 return (sopoll(so, events, fp->f_cred, td));
300 }
301
302 static int
303 soo_stat(struct file *fp, struct stat *ub, struct ucred *active_cred,
304 struct thread *td)
305 {
306 struct socket *so = fp->f_data;
307 int error;
308
309 bzero((caddr_t)ub, sizeof (*ub));
310 ub->st_mode = S_IFSOCK;
311 #ifdef MAC
312 error = mac_socket_check_stat(active_cred, so);
313 if (error)
314 return (error);
315 #endif
316 SOCK_LOCK(so);
317 if (!SOLISTENING(so)) {
318 struct sockbuf *sb;
319
320 /*
321 * If SBS_CANTRCVMORE is set, but there's still data left
322 * in the receive buffer, the socket is still readable.
323 */
324 sb = &so->so_rcv;
325 SOCKBUF_LOCK(sb);
326 if ((sb->sb_state & SBS_CANTRCVMORE) == 0 || sbavail(sb))
327 ub->st_mode |= S_IRUSR | S_IRGRP | S_IROTH;
328 ub->st_size = sbavail(sb) - sb->sb_ctl;
329 SOCKBUF_UNLOCK(sb);
330
331 sb = &so->so_snd;
332 SOCKBUF_LOCK(sb);
333 if ((sb->sb_state & SBS_CANTSENDMORE) == 0)
334 ub->st_mode |= S_IWUSR | S_IWGRP | S_IWOTH;
335 SOCKBUF_UNLOCK(sb);
336 }
337 ub->st_uid = so->so_cred->cr_uid;
338 ub->st_gid = so->so_cred->cr_gid;
339 error = so->so_proto->pr_usrreqs->pru_sense(so, ub);
340 SOCK_UNLOCK(so);
341 return (error);
342 }
343
344 /*
345 * API socket close on file pointer. We call soclose() to close the socket
346 * (including initiating closing protocols). soclose() will sorele() the
347 * file reference but the actual socket will not go away until the socket's
348 * ref count hits 0.
349 */
350 static int
351 soo_close(struct file *fp, struct thread *td)
352 {
353 int error = 0;
354 struct socket *so;
355
356 so = fp->f_data;
357 fp->f_ops = &badfileops;
358 fp->f_data = NULL;
359
360 if (so)
361 error = soclose(so);
362 return (error);
363 }
364
365 static int
366 soo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
367 {
368 struct sockaddr *sa;
369 struct inpcb *inpcb;
370 struct unpcb *unpcb;
371 struct socket *so;
372 int error;
373
374 kif->kf_type = KF_TYPE_SOCKET;
375 so = fp->f_data;
376 CURVNET_SET(so->so_vnet);
377 kif->kf_un.kf_sock.kf_sock_domain0 =
378 so->so_proto->pr_domain->dom_family;
379 kif->kf_un.kf_sock.kf_sock_type0 = so->so_type;
380 kif->kf_un.kf_sock.kf_sock_protocol0 = so->so_proto->pr_protocol;
381 kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb;
382 switch (kif->kf_un.kf_sock.kf_sock_domain0) {
383 case AF_INET:
384 case AF_INET6:
385 if (so->so_pcb != NULL) {
386 inpcb = (struct inpcb *)(so->so_pcb);
387 kif->kf_un.kf_sock.kf_sock_inpcb =
388 (uintptr_t)inpcb->inp_ppcb;
389 }
390 kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
391 so->so_rcv.sb_state;
392 kif->kf_un.kf_sock.kf_sock_snd_sb_state =
393 so->so_snd.sb_state;
394 kif->kf_un.kf_sock.kf_sock_sendq =
395 sbused(&so->so_snd);
396 kif->kf_un.kf_sock.kf_sock_recvq =
397 sbused(&so->so_rcv);
398 break;
399 case AF_UNIX:
400 if (so->so_pcb != NULL) {
401 unpcb = (struct unpcb *)(so->so_pcb);
402 if (unpcb->unp_conn) {
403 kif->kf_un.kf_sock.kf_sock_unpconn =
404 (uintptr_t)unpcb->unp_conn;
405 kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
406 so->so_rcv.sb_state;
407 kif->kf_un.kf_sock.kf_sock_snd_sb_state =
408 so->so_snd.sb_state;
409 kif->kf_un.kf_sock.kf_sock_sendq =
410 sbused(&so->so_snd);
411 kif->kf_un.kf_sock.kf_sock_recvq =
412 sbused(&so->so_rcv);
413 }
414 }
415 break;
416 }
417 error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa);
418 if (error == 0 &&
419 sa->sa_len <= sizeof(kif->kf_un.kf_sock.kf_sa_local)) {
420 bcopy(sa, &kif->kf_un.kf_sock.kf_sa_local, sa->sa_len);
421 free(sa, M_SONAME);
422 }
423 error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa);
424 if (error == 0 &&
425 sa->sa_len <= sizeof(kif->kf_un.kf_sock.kf_sa_peer)) {
426 bcopy(sa, &kif->kf_un.kf_sock.kf_sa_peer, sa->sa_len);
427 free(sa, M_SONAME);
428 }
429 strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name,
430 sizeof(kif->kf_path));
431 CURVNET_RESTORE();
432 return (0);
433 }
434
435 /*
436 * Use the 'backend3' field in AIO jobs to store the amount of data
437 * completed by the AIO job so far.
438 */
439 #define aio_done backend3
440
441 static STAILQ_HEAD(, task) soaio_jobs;
442 static struct mtx soaio_jobs_lock;
443 static struct task soaio_kproc_task;
444 static int soaio_starting, soaio_idle, soaio_queued;
445 static struct unrhdr *soaio_kproc_unr;
446
447 static int soaio_max_procs = MAX_AIO_PROCS;
448 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, max_procs, CTLFLAG_RW, &soaio_max_procs, 0,
449 "Maximum number of kernel processes to use for async socket IO");
450
451 static int soaio_num_procs;
452 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, num_procs, CTLFLAG_RD, &soaio_num_procs, 0,
453 "Number of active kernel processes for async socket IO");
454
455 static int soaio_target_procs = TARGET_AIO_PROCS;
456 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, target_procs, CTLFLAG_RD,
457 &soaio_target_procs, 0,
458 "Preferred number of ready kernel processes for async socket IO");
459
460 static int soaio_lifetime;
461 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, lifetime, CTLFLAG_RW, &soaio_lifetime, 0,
462 "Maximum lifetime for idle aiod");
463
464 static void
465 soaio_kproc_loop(void *arg)
466 {
467 struct proc *p;
468 struct vmspace *myvm;
469 struct task *task;
470 int error, id, pending;
471
472 id = (intptr_t)arg;
473
474 /*
475 * Grab an extra reference on the daemon's vmspace so that it
476 * doesn't get freed by jobs that switch to a different
477 * vmspace.
478 */
479 p = curproc;
480 myvm = vmspace_acquire_ref(p);
481
482 mtx_lock(&soaio_jobs_lock);
483 MPASS(soaio_starting > 0);
484 soaio_starting--;
485 for (;;) {
486 while (!STAILQ_EMPTY(&soaio_jobs)) {
487 task = STAILQ_FIRST(&soaio_jobs);
488 STAILQ_REMOVE_HEAD(&soaio_jobs, ta_link);
489 soaio_queued--;
490 pending = task->ta_pending;
491 task->ta_pending = 0;
492 mtx_unlock(&soaio_jobs_lock);
493
494 task->ta_func(task->ta_context, pending);
495
496 mtx_lock(&soaio_jobs_lock);
497 }
498 MPASS(soaio_queued == 0);
499
500 if (p->p_vmspace != myvm) {
501 mtx_unlock(&soaio_jobs_lock);
502 vmspace_switch_aio(myvm);
503 mtx_lock(&soaio_jobs_lock);
504 continue;
505 }
506
507 soaio_idle++;
508 error = mtx_sleep(&soaio_idle, &soaio_jobs_lock, 0, "-",
509 soaio_lifetime);
510 soaio_idle--;
511 if (error == EWOULDBLOCK && STAILQ_EMPTY(&soaio_jobs) &&
512 soaio_num_procs > soaio_target_procs)
513 break;
514 }
515 soaio_num_procs--;
516 mtx_unlock(&soaio_jobs_lock);
517 free_unr(soaio_kproc_unr, id);
518 kproc_exit(0);
519 }
520
521 static void
522 soaio_kproc_create(void *context, int pending)
523 {
524 struct proc *p;
525 int error, id;
526
527 mtx_lock(&soaio_jobs_lock);
528 for (;;) {
529 if (soaio_num_procs < soaio_target_procs) {
530 /* Must create */
531 } else if (soaio_num_procs >= soaio_max_procs) {
532 /*
533 * Hit the limit on kernel processes, don't
534 * create another one.
535 */
536 break;
537 } else if (soaio_queued <= soaio_idle + soaio_starting) {
538 /*
539 * No more AIO jobs waiting for a process to be
540 * created, so stop.
541 */
542 break;
543 }
544 soaio_starting++;
545 mtx_unlock(&soaio_jobs_lock);
546
547 id = alloc_unr(soaio_kproc_unr);
548 error = kproc_create(soaio_kproc_loop, (void *)(intptr_t)id,
549 &p, 0, 0, "soaiod%d", id);
550 if (error != 0) {
551 free_unr(soaio_kproc_unr, id);
552 mtx_lock(&soaio_jobs_lock);
553 soaio_starting--;
554 break;
555 }
556
557 mtx_lock(&soaio_jobs_lock);
558 soaio_num_procs++;
559 }
560 mtx_unlock(&soaio_jobs_lock);
561 }
562
563 void
564 soaio_enqueue(struct task *task)
565 {
566
567 mtx_lock(&soaio_jobs_lock);
568 MPASS(task->ta_pending == 0);
569 task->ta_pending++;
570 STAILQ_INSERT_TAIL(&soaio_jobs, task, ta_link);
571 soaio_queued++;
572 if (soaio_queued <= soaio_idle)
573 wakeup_one(&soaio_idle);
574 else if (soaio_num_procs < soaio_max_procs)
575 taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task);
576 mtx_unlock(&soaio_jobs_lock);
577 }
578
579 static void
580 soaio_init(void)
581 {
582
583 soaio_lifetime = AIOD_LIFETIME_DEFAULT;
584 STAILQ_INIT(&soaio_jobs);
585 mtx_init(&soaio_jobs_lock, "soaio jobs", NULL, MTX_DEF);
586 soaio_kproc_unr = new_unrhdr(1, INT_MAX, NULL);
587 TASK_INIT(&soaio_kproc_task, 0, soaio_kproc_create, NULL);
588 }
589 SYSINIT(soaio, SI_SUB_VFS, SI_ORDER_ANY, soaio_init, NULL);
590
591 static __inline int
592 soaio_ready(struct socket *so, struct sockbuf *sb)
593 {
594 return (sb == &so->so_rcv ? soreadable(so) : sowriteable(so));
595 }
596
597 static void
598 soaio_process_job(struct socket *so, struct sockbuf *sb, struct kaiocb *job)
599 {
600 struct ucred *td_savedcred;
601 struct thread *td;
602 struct file *fp;
603 size_t cnt, done, job_total_nbytes;
604 long ru_before;
605 int error, flags;
606
607 SOCKBUF_UNLOCK(sb);
608 aio_switch_vmspace(job);
609 td = curthread;
610 fp = job->fd_file;
611 retry:
612 td_savedcred = td->td_ucred;
613 td->td_ucred = job->cred;
614
615 job_total_nbytes = job->uiop->uio_resid + job->aio_done;
616 done = job->aio_done;
617 cnt = job->uiop->uio_resid;
618 job->uiop->uio_offset = 0;
619 job->uiop->uio_td = td;
620 flags = MSG_NBIO;
621
622 /*
623 * For resource usage accounting, only count a completed request
624 * as a single message to avoid counting multiple calls to
625 * sosend/soreceive on a blocking socket.
626 */
627
628 if (sb == &so->so_rcv) {
629 ru_before = td->td_ru.ru_msgrcv;
630 #ifdef MAC
631 error = mac_socket_check_receive(fp->f_cred, so);
632 if (error == 0)
633
634 #endif
635 error = soreceive(so, NULL, job->uiop, NULL, NULL,
636 &flags);
637 if (td->td_ru.ru_msgrcv != ru_before)
638 job->msgrcv = 1;
639 } else {
640 if (!TAILQ_EMPTY(&sb->sb_aiojobq))
641 flags |= MSG_MORETOCOME;
642 ru_before = td->td_ru.ru_msgsnd;
643 #ifdef MAC
644 error = mac_socket_check_send(fp->f_cred, so);
645 if (error == 0)
646 #endif
647 error = sosend(so, NULL, job->uiop, NULL, NULL, flags,
648 td);
649 if (td->td_ru.ru_msgsnd != ru_before)
650 job->msgsnd = 1;
651 if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0) {
652 PROC_LOCK(job->userproc);
653 kern_psignal(job->userproc, SIGPIPE);
654 PROC_UNLOCK(job->userproc);
655 }
656 }
657
658 done += cnt - job->uiop->uio_resid;
659 job->aio_done = done;
660 td->td_ucred = td_savedcred;
661
662 if (error == EWOULDBLOCK) {
663 /*
664 * The request was either partially completed or not
665 * completed at all due to racing with a read() or
666 * write() on the socket. If the socket is
667 * non-blocking, return with any partial completion.
668 * If the socket is blocking or if no progress has
669 * been made, requeue this request at the head of the
670 * queue to try again when the socket is ready.
671 */
672 MPASS(done != job_total_nbytes);
673 SOCKBUF_LOCK(sb);
674 if (done == 0 || !(so->so_state & SS_NBIO)) {
675 empty_results++;
676 if (soaio_ready(so, sb)) {
677 empty_retries++;
678 SOCKBUF_UNLOCK(sb);
679 goto retry;
680 }
681
682 if (!aio_set_cancel_function(job, soo_aio_cancel)) {
683 SOCKBUF_UNLOCK(sb);
684 if (done != 0)
685 aio_complete(job, done, 0);
686 else
687 aio_cancel(job);
688 SOCKBUF_LOCK(sb);
689 } else {
690 TAILQ_INSERT_HEAD(&sb->sb_aiojobq, job, list);
691 }
692 return;
693 }
694 SOCKBUF_UNLOCK(sb);
695 }
696 if (done != 0 && (error == ERESTART || error == EINTR ||
697 error == EWOULDBLOCK))
698 error = 0;
699 if (error)
700 aio_complete(job, -1, error);
701 else
702 aio_complete(job, done, 0);
703 SOCKBUF_LOCK(sb);
704 }
705
706 static void
707 soaio_process_sb(struct socket *so, struct sockbuf *sb)
708 {
709 struct kaiocb *job;
710
711 CURVNET_SET(so->so_vnet);
712 SOCKBUF_LOCK(sb);
713 while (!TAILQ_EMPTY(&sb->sb_aiojobq) && soaio_ready(so, sb)) {
714 job = TAILQ_FIRST(&sb->sb_aiojobq);
715 TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
716 if (!aio_clear_cancel_function(job))
717 continue;
718
719 soaio_process_job(so, sb, job);
720 }
721
722 /*
723 * If there are still pending requests, the socket must not be
724 * ready so set SB_AIO to request a wakeup when the socket
725 * becomes ready.
726 */
727 if (!TAILQ_EMPTY(&sb->sb_aiojobq))
728 sb->sb_flags |= SB_AIO;
729 sb->sb_flags &= ~SB_AIO_RUNNING;
730 SOCKBUF_UNLOCK(sb);
731
732 SOCK_LOCK(so);
733 sorele(so);
734 CURVNET_RESTORE();
735 }
736
737 void
738 soaio_rcv(void *context, int pending)
739 {
740 struct socket *so;
741
742 so = context;
743 soaio_process_sb(so, &so->so_rcv);
744 }
745
746 void
747 soaio_snd(void *context, int pending)
748 {
749 struct socket *so;
750
751 so = context;
752 soaio_process_sb(so, &so->so_snd);
753 }
754
755 void
756 sowakeup_aio(struct socket *so, struct sockbuf *sb)
757 {
758
759 SOCKBUF_LOCK_ASSERT(sb);
760 sb->sb_flags &= ~SB_AIO;
761 if (sb->sb_flags & SB_AIO_RUNNING)
762 return;
763 sb->sb_flags |= SB_AIO_RUNNING;
764 soref(so);
765 soaio_enqueue(&sb->sb_aiotask);
766 }
767
768 static void
769 soo_aio_cancel(struct kaiocb *job)
770 {
771 struct socket *so;
772 struct sockbuf *sb;
773 long done;
774 int opcode;
775
776 so = job->fd_file->f_data;
777 opcode = job->uaiocb.aio_lio_opcode;
778 if (opcode & LIO_READ)
779 sb = &so->so_rcv;
780 else {
781 MPASS(opcode & LIO_WRITE);
782 sb = &so->so_snd;
783 }
784
785 SOCKBUF_LOCK(sb);
786 if (!aio_cancel_cleared(job))
787 TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
788 if (TAILQ_EMPTY(&sb->sb_aiojobq))
789 sb->sb_flags &= ~SB_AIO;
790 SOCKBUF_UNLOCK(sb);
791
792 done = job->aio_done;
793 if (done != 0)
794 aio_complete(job, done, 0);
795 else
796 aio_cancel(job);
797 }
798
799 static int
800 soo_aio_queue(struct file *fp, struct kaiocb *job)
801 {
802 struct socket *so;
803 struct sockbuf *sb;
804 int error;
805
806 so = fp->f_data;
807 error = (*so->so_proto->pr_usrreqs->pru_aio_queue)(so, job);
808 if (error == 0)
809 return (0);
810
811 switch (job->uaiocb.aio_lio_opcode & (LIO_WRITE | LIO_READ)) {
812 case LIO_READ:
813 sb = &so->so_rcv;
814 break;
815 case LIO_WRITE:
816 sb = &so->so_snd;
817 break;
818 default:
819 return (EINVAL);
820 }
821
822 SOCKBUF_LOCK(sb);
823 if (!aio_set_cancel_function(job, soo_aio_cancel))
824 panic("new job was cancelled");
825 TAILQ_INSERT_TAIL(&sb->sb_aiojobq, job, list);
826 if (!(sb->sb_flags & SB_AIO_RUNNING)) {
827 if (soaio_ready(so, sb))
828 sowakeup_aio(so, sb);
829 else
830 sb->sb_flags |= SB_AIO;
831 }
832 SOCKBUF_UNLOCK(sb);
833 return (0);
834 }
Cache object: e3cfcf9c77c7a06f7482c9b984a818c8
|