FreeBSD/Linux Kernel Cross Reference
sys/kern/sys_socket.c
1 /*-
2 * Copyright (c) 1982, 1986, 1990, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
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.
16 *
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
27 * SUCH DAMAGE.
28 *
29 * @(#)sys_socket.c 8.1 (Berkeley) 6/10/93
30 */
31
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD: releng/11.0/sys/kern/sys_socket.c 322343 2017-08-10 06:59:26Z delphij $");
34
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/aio.h>
38 #include <sys/domain.h>
39 #include <sys/file.h>
40 #include <sys/filedesc.h>
41 #include <sys/kernel.h>
42 #include <sys/kthread.h>
43 #include <sys/malloc.h>
44 #include <sys/proc.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>
53 #include <sys/stat.h>
54 #include <sys/sysctl.h>
55 #include <sys/sysproto.h>
56 #include <sys/taskqueue.h>
57 #include <sys/uio.h>
58 #include <sys/ucred.h>
59 #include <sys/un.h>
60 #include <sys/unpcb.h>
61 #include <sys/user.h>
62
63 #include <net/if.h>
64 #include <net/if_var.h>
65 #include <net/route.h>
66 #include <net/vnet.h>
67
68 #include <netinet/in.h>
69 #include <netinet/in_pcb.h>
70
71 #include <security/mac/mac_framework.h>
72
73 #include <vm/vm.h>
74 #include <vm/pmap.h>
75 #include <vm/vm_extern.h>
76 #include <vm/vm_map.h>
77
78 static SYSCTL_NODE(_kern_ipc, OID_AUTO, aio, CTLFLAG_RD, NULL,
79 "socket AIO stats");
80
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");
84
85 static int empty_retries;
86 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, empty_retries, CTLFLAG_RD, &empty_retries,
87 0, "socket operation retries");
88
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;
98
99 static void soo_aio_cancel(struct kaiocb *job);
100
101 struct fileops socketops = {
102 .fo_read = soo_read,
103 .fo_write = soo_write,
104 .fo_truncate = invfo_truncate,
105 .fo_ioctl = soo_ioctl,
106 .fo_poll = soo_poll,
107 .fo_kqfilter = soo_kqfilter,
108 .fo_stat = soo_stat,
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
116 };
117
118 static int
119 soo_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
120 int flags, struct thread *td)
121 {
122 struct socket *so = fp->f_data;
123 int error;
124
125 #ifdef MAC
126 error = mac_socket_check_receive(active_cred, so);
127 if (error)
128 return (error);
129 #endif
130 error = soreceive(so, 0, uio, 0, 0, 0);
131 return (error);
132 }
133
134 static int
135 soo_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
136 int flags, struct thread *td)
137 {
138 struct socket *so = fp->f_data;
139 int error;
140
141 #ifdef MAC
142 error = mac_socket_check_send(active_cred, so);
143 if (error)
144 return (error);
145 #endif
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);
151 }
152 return (error);
153 }
154
155 static int
156 soo_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *active_cred,
157 struct thread *td)
158 {
159 struct socket *so = fp->f_data;
160 int error = 0;
161
162 switch (cmd) {
163 case FIONBIO:
164 SOCK_LOCK(so);
165 if (*(int *)data)
166 so->so_state |= SS_NBIO;
167 else
168 so->so_state &= ~SS_NBIO;
169 SOCK_UNLOCK(so);
170 break;
171
172 case FIOASYNC:
173 /*
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
177 * the same.
178 */
179 if (*(int *)data) {
180 SOCK_LOCK(so);
181 so->so_state |= SS_ASYNC;
182 SOCK_UNLOCK(so);
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);
189 } else {
190 SOCK_LOCK(so);
191 so->so_state &= ~SS_ASYNC;
192 SOCK_UNLOCK(so);
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);
199 }
200 break;
201
202 case FIONREAD:
203 /* Unlocked read. */
204 *(int *)data = sbavail(&so->so_rcv);
205 break;
206
207 case FIONWRITE:
208 /* Unlocked read. */
209 *(int *)data = sbavail(&so->so_snd);
210 break;
211
212 case FIONSPACE:
213 /* Unlocked read. */
214 if ((so->so_snd.sb_hiwat < sbused(&so->so_snd)) ||
215 (so->so_snd.sb_mbmax < so->so_snd.sb_mbcnt))
216 *(int *)data = 0;
217 else
218 *(int *)data = sbspace(&so->so_snd);
219 break;
220
221 case FIOSETOWN:
222 error = fsetown(*(int *)data, &so->so_sigio);
223 break;
224
225 case FIOGETOWN:
226 *(int *)data = fgetown(&so->so_sigio);
227 break;
228
229 case SIOCSPGRP:
230 error = fsetown(-(*(int *)data), &so->so_sigio);
231 break;
232
233 case SIOCGPGRP:
234 *(int *)data = -fgetown(&so->so_sigio);
235 break;
236
237 case SIOCATMARK:
238 /* Unlocked read. */
239 *(int *)data = (so->so_rcv.sb_state & SBS_RCVATMARK) != 0;
240 break;
241 default:
242 /*
243 * Interface/routing/protocol specific ioctls: interface and
244 * routing ioctls should have a different entry since a
245 * socket is unnecessary.
246 */
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);
252 CURVNET_RESTORE();
253 } else {
254 CURVNET_SET(so->so_vnet);
255 error = ((*so->so_proto->pr_usrreqs->pru_control)
256 (so, cmd, data, 0, td));
257 CURVNET_RESTORE();
258 }
259 break;
260 }
261 return (error);
262 }
263
264 static int
265 soo_poll(struct file *fp, int events, struct ucred *active_cred,
266 struct thread *td)
267 {
268 struct socket *so = fp->f_data;
269 #ifdef MAC
270 int error;
271
272 error = mac_socket_check_poll(active_cred, so);
273 if (error)
274 return (error);
275 #endif
276 return (sopoll(so, events, fp->f_cred, td));
277 }
278
279 static int
280 soo_stat(struct file *fp, struct stat *ub, struct ucred *active_cred,
281 struct thread *td)
282 {
283 struct socket *so = fp->f_data;
284 struct sockbuf *sb;
285 #ifdef MAC
286 int error;
287 #endif
288
289 bzero((caddr_t)ub, sizeof (*ub));
290 ub->st_mode = S_IFSOCK;
291 #ifdef MAC
292 error = mac_socket_check_stat(active_cred, so);
293 if (error)
294 return (error);
295 #endif
296 /*
297 * If SBS_CANTRCVMORE is set, but there's still data left in the
298 * receive buffer, the socket is still readable.
299 */
300 sb = &so->so_rcv;
301 SOCKBUF_LOCK(sb);
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;
305 SOCKBUF_UNLOCK(sb);
306
307 sb = &so->so_snd;
308 SOCKBUF_LOCK(sb);
309 if ((sb->sb_state & SBS_CANTSENDMORE) == 0)
310 ub->st_mode |= S_IWUSR | S_IWGRP | S_IWOTH;
311 SOCKBUF_UNLOCK(sb);
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);
315 }
316
317 /*
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
321 * ref count hits 0.
322 */
323 static int
324 soo_close(struct file *fp, struct thread *td)
325 {
326 int error = 0;
327 struct socket *so;
328
329 so = fp->f_data;
330 fp->f_ops = &badfileops;
331 fp->f_data = NULL;
332
333 if (so)
334 error = soclose(so);
335 return (error);
336 }
337
338 static int
339 soo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
340 {
341 struct sockaddr *sa;
342 struct inpcb *inpcb;
343 struct unpcb *unpcb;
344 struct socket *so;
345 int error;
346
347 kif->kf_type = KF_TYPE_SOCKET;
348 so = fp->f_data;
349 kif->kf_sock_domain = so->so_proto->pr_domain->dom_family;
350 kif->kf_sock_type = so->so_type;
351 kif->kf_sock_protocol = so->so_proto->pr_protocol;
352 kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb;
353 switch (kif->kf_sock_domain) {
354 case AF_INET:
355 case AF_INET6:
356 if (kif->kf_sock_protocol == IPPROTO_TCP) {
357 if (so->so_pcb != NULL) {
358 inpcb = (struct inpcb *)(so->so_pcb);
359 kif->kf_un.kf_sock.kf_sock_inpcb =
360 (uintptr_t)inpcb->inp_ppcb;
361 }
362 }
363 break;
364 case AF_UNIX:
365 if (so->so_pcb != NULL) {
366 unpcb = (struct unpcb *)(so->so_pcb);
367 if (unpcb->unp_conn) {
368 kif->kf_un.kf_sock.kf_sock_unpconn =
369 (uintptr_t)unpcb->unp_conn;
370 kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
371 so->so_rcv.sb_state;
372 kif->kf_un.kf_sock.kf_sock_snd_sb_state =
373 so->so_snd.sb_state;
374 }
375 }
376 break;
377 }
378 error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa);
379 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
380 bcopy(sa, &kif->kf_sa_local, sa->sa_len);
381 free(sa, M_SONAME);
382 }
383 error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa);
384 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
385 bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
386 free(sa, M_SONAME);
387 }
388 strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name,
389 sizeof(kif->kf_path));
390 return (0);
391 }
392
393 /*
394 * Use the 'backend3' field in AIO jobs to store the amount of data
395 * completed by the AIO job so far.
396 */
397 #define aio_done backend3
398
399 static STAILQ_HEAD(, task) soaio_jobs;
400 static struct mtx soaio_jobs_lock;
401 static struct task soaio_kproc_task;
402 static int soaio_starting, soaio_idle, soaio_queued;
403 static struct unrhdr *soaio_kproc_unr;
404
405 static int soaio_max_procs = MAX_AIO_PROCS;
406 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, max_procs, CTLFLAG_RW, &soaio_max_procs, 0,
407 "Maximum number of kernel processes to use for async socket IO");
408
409 static int soaio_num_procs;
410 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, num_procs, CTLFLAG_RD, &soaio_num_procs, 0,
411 "Number of active kernel processes for async socket IO");
412
413 static int soaio_target_procs = TARGET_AIO_PROCS;
414 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, target_procs, CTLFLAG_RD,
415 &soaio_target_procs, 0,
416 "Preferred number of ready kernel processes for async socket IO");
417
418 static int soaio_lifetime;
419 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, lifetime, CTLFLAG_RW, &soaio_lifetime, 0,
420 "Maximum lifetime for idle aiod");
421
422 static void
423 soaio_kproc_loop(void *arg)
424 {
425 struct proc *p;
426 struct vmspace *myvm;
427 struct task *task;
428 int error, id, pending;
429
430 id = (intptr_t)arg;
431
432 /*
433 * Grab an extra reference on the daemon's vmspace so that it
434 * doesn't get freed by jobs that switch to a different
435 * vmspace.
436 */
437 p = curproc;
438 myvm = vmspace_acquire_ref(p);
439
440 mtx_lock(&soaio_jobs_lock);
441 MPASS(soaio_starting > 0);
442 soaio_starting--;
443 for (;;) {
444 while (!STAILQ_EMPTY(&soaio_jobs)) {
445 task = STAILQ_FIRST(&soaio_jobs);
446 STAILQ_REMOVE_HEAD(&soaio_jobs, ta_link);
447 soaio_queued--;
448 pending = task->ta_pending;
449 task->ta_pending = 0;
450 mtx_unlock(&soaio_jobs_lock);
451
452 task->ta_func(task->ta_context, pending);
453
454 mtx_lock(&soaio_jobs_lock);
455 }
456 MPASS(soaio_queued == 0);
457
458 if (p->p_vmspace != myvm) {
459 mtx_unlock(&soaio_jobs_lock);
460 vmspace_switch_aio(myvm);
461 mtx_lock(&soaio_jobs_lock);
462 continue;
463 }
464
465 soaio_idle++;
466 error = mtx_sleep(&soaio_idle, &soaio_jobs_lock, 0, "-",
467 soaio_lifetime);
468 soaio_idle--;
469 if (error == EWOULDBLOCK && STAILQ_EMPTY(&soaio_jobs) &&
470 soaio_num_procs > soaio_target_procs)
471 break;
472 }
473 soaio_num_procs--;
474 mtx_unlock(&soaio_jobs_lock);
475 free_unr(soaio_kproc_unr, id);
476 kproc_exit(0);
477 }
478
479 static void
480 soaio_kproc_create(void *context, int pending)
481 {
482 struct proc *p;
483 int error, id;
484
485 mtx_lock(&soaio_jobs_lock);
486 for (;;) {
487 if (soaio_num_procs < soaio_target_procs) {
488 /* Must create */
489 } else if (soaio_num_procs >= soaio_max_procs) {
490 /*
491 * Hit the limit on kernel processes, don't
492 * create another one.
493 */
494 break;
495 } else if (soaio_queued <= soaio_idle + soaio_starting) {
496 /*
497 * No more AIO jobs waiting for a process to be
498 * created, so stop.
499 */
500 break;
501 }
502 soaio_starting++;
503 mtx_unlock(&soaio_jobs_lock);
504
505 id = alloc_unr(soaio_kproc_unr);
506 error = kproc_create(soaio_kproc_loop, (void *)(intptr_t)id,
507 &p, 0, 0, "soaiod%d", id);
508 if (error != 0) {
509 free_unr(soaio_kproc_unr, id);
510 mtx_lock(&soaio_jobs_lock);
511 soaio_starting--;
512 break;
513 }
514
515 mtx_lock(&soaio_jobs_lock);
516 soaio_num_procs++;
517 }
518 mtx_unlock(&soaio_jobs_lock);
519 }
520
521 void
522 soaio_enqueue(struct task *task)
523 {
524
525 mtx_lock(&soaio_jobs_lock);
526 MPASS(task->ta_pending == 0);
527 task->ta_pending++;
528 STAILQ_INSERT_TAIL(&soaio_jobs, task, ta_link);
529 soaio_queued++;
530 if (soaio_queued <= soaio_idle)
531 wakeup_one(&soaio_idle);
532 else if (soaio_num_procs < soaio_max_procs)
533 taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task);
534 mtx_unlock(&soaio_jobs_lock);
535 }
536
537 static void
538 soaio_init(void)
539 {
540
541 soaio_lifetime = AIOD_LIFETIME_DEFAULT;
542 STAILQ_INIT(&soaio_jobs);
543 mtx_init(&soaio_jobs_lock, "soaio jobs", NULL, MTX_DEF);
544 soaio_kproc_unr = new_unrhdr(1, INT_MAX, NULL);
545 TASK_INIT(&soaio_kproc_task, 0, soaio_kproc_create, NULL);
546 if (soaio_target_procs > 0)
547 taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task);
548 }
549 SYSINIT(soaio, SI_SUB_VFS, SI_ORDER_ANY, soaio_init, NULL);
550
551 static __inline int
552 soaio_ready(struct socket *so, struct sockbuf *sb)
553 {
554 return (sb == &so->so_rcv ? soreadable(so) : sowriteable(so));
555 }
556
557 static void
558 soaio_process_job(struct socket *so, struct sockbuf *sb, struct kaiocb *job)
559 {
560 struct ucred *td_savedcred;
561 struct thread *td;
562 struct file *fp;
563 struct uio uio;
564 struct iovec iov;
565 size_t cnt, done;
566 long ru_before;
567 int error, flags;
568
569 SOCKBUF_UNLOCK(sb);
570 aio_switch_vmspace(job);
571 td = curthread;
572 fp = job->fd_file;
573 retry:
574 td_savedcred = td->td_ucred;
575 td->td_ucred = job->cred;
576
577 done = job->aio_done;
578 cnt = job->uaiocb.aio_nbytes - done;
579 iov.iov_base = (void *)((uintptr_t)job->uaiocb.aio_buf + done);
580 iov.iov_len = cnt;
581 uio.uio_iov = &iov;
582 uio.uio_iovcnt = 1;
583 uio.uio_offset = 0;
584 uio.uio_resid = cnt;
585 uio.uio_segflg = UIO_USERSPACE;
586 uio.uio_td = td;
587 flags = MSG_NBIO;
588
589 /*
590 * For resource usage accounting, only count a completed request
591 * as a single message to avoid counting multiple calls to
592 * sosend/soreceive on a blocking socket.
593 */
594
595 if (sb == &so->so_rcv) {
596 uio.uio_rw = UIO_READ;
597 ru_before = td->td_ru.ru_msgrcv;
598 #ifdef MAC
599 error = mac_socket_check_receive(fp->f_cred, so);
600 if (error == 0)
601
602 #endif
603 error = soreceive(so, NULL, &uio, NULL, NULL, &flags);
604 if (td->td_ru.ru_msgrcv != ru_before)
605 job->msgrcv = 1;
606 } else {
607 uio.uio_rw = UIO_WRITE;
608 ru_before = td->td_ru.ru_msgsnd;
609 #ifdef MAC
610 error = mac_socket_check_send(fp->f_cred, so);
611 if (error == 0)
612 #endif
613 error = sosend(so, NULL, &uio, NULL, NULL, flags, td);
614 if (td->td_ru.ru_msgsnd != ru_before)
615 job->msgsnd = 1;
616 if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0) {
617 PROC_LOCK(job->userproc);
618 kern_psignal(job->userproc, SIGPIPE);
619 PROC_UNLOCK(job->userproc);
620 }
621 }
622
623 done += cnt - uio.uio_resid;
624 job->aio_done = done;
625 td->td_ucred = td_savedcred;
626
627 if (error == EWOULDBLOCK) {
628 /*
629 * The request was either partially completed or not
630 * completed at all due to racing with a read() or
631 * write() on the socket. If the socket is
632 * non-blocking, return with any partial completion.
633 * If the socket is blocking or if no progress has
634 * been made, requeue this request at the head of the
635 * queue to try again when the socket is ready.
636 */
637 MPASS(done != job->uaiocb.aio_nbytes);
638 SOCKBUF_LOCK(sb);
639 if (done == 0 || !(so->so_state & SS_NBIO)) {
640 empty_results++;
641 if (soaio_ready(so, sb)) {
642 empty_retries++;
643 SOCKBUF_UNLOCK(sb);
644 goto retry;
645 }
646
647 if (!aio_set_cancel_function(job, soo_aio_cancel)) {
648 SOCKBUF_UNLOCK(sb);
649 if (done != 0)
650 aio_complete(job, done, 0);
651 else
652 aio_cancel(job);
653 SOCKBUF_LOCK(sb);
654 } else {
655 TAILQ_INSERT_HEAD(&sb->sb_aiojobq, job, list);
656 }
657 return;
658 }
659 SOCKBUF_UNLOCK(sb);
660 }
661 if (done != 0 && (error == ERESTART || error == EINTR ||
662 error == EWOULDBLOCK))
663 error = 0;
664 if (error)
665 aio_complete(job, -1, error);
666 else
667 aio_complete(job, done, 0);
668 SOCKBUF_LOCK(sb);
669 }
670
671 static void
672 soaio_process_sb(struct socket *so, struct sockbuf *sb)
673 {
674 struct kaiocb *job;
675
676 CURVNET_SET(so->so_vnet);
677 SOCKBUF_LOCK(sb);
678 while (!TAILQ_EMPTY(&sb->sb_aiojobq) && soaio_ready(so, sb)) {
679 job = TAILQ_FIRST(&sb->sb_aiojobq);
680 TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
681 if (!aio_clear_cancel_function(job))
682 continue;
683
684 soaio_process_job(so, sb, job);
685 }
686
687 /*
688 * If there are still pending requests, the socket must not be
689 * ready so set SB_AIO to request a wakeup when the socket
690 * becomes ready.
691 */
692 if (!TAILQ_EMPTY(&sb->sb_aiojobq))
693 sb->sb_flags |= SB_AIO;
694 sb->sb_flags &= ~SB_AIO_RUNNING;
695 SOCKBUF_UNLOCK(sb);
696
697 ACCEPT_LOCK();
698 SOCK_LOCK(so);
699 sorele(so);
700 CURVNET_RESTORE();
701 }
702
703 void
704 soaio_rcv(void *context, int pending)
705 {
706 struct socket *so;
707
708 so = context;
709 soaio_process_sb(so, &so->so_rcv);
710 }
711
712 void
713 soaio_snd(void *context, int pending)
714 {
715 struct socket *so;
716
717 so = context;
718 soaio_process_sb(so, &so->so_snd);
719 }
720
721 void
722 sowakeup_aio(struct socket *so, struct sockbuf *sb)
723 {
724
725 SOCKBUF_LOCK_ASSERT(sb);
726 sb->sb_flags &= ~SB_AIO;
727 if (sb->sb_flags & SB_AIO_RUNNING)
728 return;
729 sb->sb_flags |= SB_AIO_RUNNING;
730 if (sb == &so->so_snd)
731 SOCK_LOCK(so);
732 soref(so);
733 if (sb == &so->so_snd)
734 SOCK_UNLOCK(so);
735 soaio_enqueue(&sb->sb_aiotask);
736 }
737
738 static void
739 soo_aio_cancel(struct kaiocb *job)
740 {
741 struct socket *so;
742 struct sockbuf *sb;
743 long done;
744 int opcode;
745
746 so = job->fd_file->f_data;
747 opcode = job->uaiocb.aio_lio_opcode;
748 if (opcode == LIO_READ)
749 sb = &so->so_rcv;
750 else {
751 MPASS(opcode == LIO_WRITE);
752 sb = &so->so_snd;
753 }
754
755 SOCKBUF_LOCK(sb);
756 if (!aio_cancel_cleared(job))
757 TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
758 if (TAILQ_EMPTY(&sb->sb_aiojobq))
759 sb->sb_flags &= ~SB_AIO;
760 SOCKBUF_UNLOCK(sb);
761
762 done = job->aio_done;
763 if (done != 0)
764 aio_complete(job, done, 0);
765 else
766 aio_cancel(job);
767 }
768
769 static int
770 soo_aio_queue(struct file *fp, struct kaiocb *job)
771 {
772 struct socket *so;
773 struct sockbuf *sb;
774 int error;
775
776 so = fp->f_data;
777 error = (*so->so_proto->pr_usrreqs->pru_aio_queue)(so, job);
778 if (error == 0)
779 return (0);
780
781 switch (job->uaiocb.aio_lio_opcode) {
782 case LIO_READ:
783 sb = &so->so_rcv;
784 break;
785 case LIO_WRITE:
786 sb = &so->so_snd;
787 break;
788 default:
789 return (EINVAL);
790 }
791
792 SOCKBUF_LOCK(sb);
793 if (!aio_set_cancel_function(job, soo_aio_cancel))
794 panic("new job was cancelled");
795 TAILQ_INSERT_TAIL(&sb->sb_aiojobq, job, list);
796 if (!(sb->sb_flags & SB_AIO_RUNNING)) {
797 if (soaio_ready(so, sb))
798 sowakeup_aio(so, sb);
799 else
800 sb->sb_flags |= SB_AIO;
801 }
802 SOCKBUF_UNLOCK(sb);
803 return (0);
804 }
Cache object: d9c37e25df28fefb9572ce3ee7233e9c
|