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 #ifndef FSTACK
99 static fo_fill_kinfo_t soo_fill_kinfo;
100 static fo_aio_queue_t soo_aio_queue;
101
102 static void soo_aio_cancel(struct kaiocb *job);
103 #endif
104
105 struct fileops socketops = {
106 .fo_read = soo_read,
107 .fo_write = soo_write,
108 .fo_truncate = invfo_truncate,
109 .fo_ioctl = soo_ioctl,
110 .fo_poll = soo_poll,
111 .fo_kqfilter = soo_kqfilter,
112 .fo_stat = soo_stat,
113 .fo_close = soo_close,
114 .fo_chmod = invfo_chmod,
115 .fo_chown = invfo_chown,
116 .fo_sendfile = invfo_sendfile,
117 #ifndef FSTACK
118 .fo_fill_kinfo = soo_fill_kinfo,
119 .fo_aio_queue = soo_aio_queue,
120 #endif
121 .fo_flags = DFLAG_PASSABLE
122 };
123
124 static int
soo_read(struct file * fp,struct uio * uio,struct ucred * active_cred,int flags,struct thread * td)125 soo_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
126 int flags, struct thread *td)
127 {
128 struct socket *so = fp->f_data;
129 int error;
130
131 #ifdef MAC
132 error = mac_socket_check_receive(active_cred, so);
133 if (error)
134 return (error);
135 #endif
136 error = soreceive(so, 0, uio, 0, 0, 0);
137 return (error);
138 }
139
140 static int
soo_write(struct file * fp,struct uio * uio,struct ucred * active_cred,int flags,struct thread * td)141 soo_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
142 int flags, struct thread *td)
143 {
144 struct socket *so = fp->f_data;
145 int error;
146
147 #ifdef MAC
148 error = mac_socket_check_send(active_cred, so);
149 if (error)
150 return (error);
151 #endif
152 error = sosend(so, 0, uio, 0, 0, 0, uio->uio_td);
153 if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0) {
154 PROC_LOCK(uio->uio_td->td_proc);
155 tdsignal(uio->uio_td, SIGPIPE);
156 PROC_UNLOCK(uio->uio_td->td_proc);
157 }
158 return (error);
159 }
160
161 static int
soo_ioctl(struct file * fp,u_long cmd,void * data,struct ucred * active_cred,struct thread * td)162 soo_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *active_cred,
163 struct thread *td)
164 {
165 struct socket *so = fp->f_data;
166 int error = 0;
167
168 switch (cmd) {
169 case FIONBIO:
170 SOCK_LOCK(so);
171 if (*(int *)data)
172 so->so_state |= SS_NBIO;
173 else
174 so->so_state &= ~SS_NBIO;
175 SOCK_UNLOCK(so);
176 break;
177
178 case FIOASYNC:
179 if (*(int *)data) {
180 SOCK_LOCK(so);
181 so->so_state |= SS_ASYNC;
182 if (SOLISTENING(so)) {
183 so->sol_sbrcv_flags |= SB_ASYNC;
184 so->sol_sbsnd_flags |= SB_ASYNC;
185 } else {
186 SOCKBUF_LOCK(&so->so_rcv);
187 so->so_rcv.sb_flags |= SB_ASYNC;
188 SOCKBUF_UNLOCK(&so->so_rcv);
189 SOCKBUF_LOCK(&so->so_snd);
190 so->so_snd.sb_flags |= SB_ASYNC;
191 SOCKBUF_UNLOCK(&so->so_snd);
192 }
193 SOCK_UNLOCK(so);
194 } else {
195 SOCK_LOCK(so);
196 so->so_state &= ~SS_ASYNC;
197 if (SOLISTENING(so)) {
198 so->sol_sbrcv_flags &= ~SB_ASYNC;
199 so->sol_sbsnd_flags &= ~SB_ASYNC;
200 } else {
201 SOCKBUF_LOCK(&so->so_rcv);
202 so->so_rcv.sb_flags &= ~SB_ASYNC;
203 SOCKBUF_UNLOCK(&so->so_rcv);
204 SOCKBUF_LOCK(&so->so_snd);
205 so->so_snd.sb_flags &= ~SB_ASYNC;
206 SOCKBUF_UNLOCK(&so->so_snd);
207 }
208 SOCK_UNLOCK(so);
209 }
210 break;
211
212 case FIONREAD:
213 /* Unlocked read. */
214 if (SOLISTENING(so)) {
215 error = EINVAL;
216 } else {
217 *(int *)data = sbavail(&so->so_rcv);
218 }
219 break;
220
221 case FIONWRITE:
222 /* Unlocked read. */
223 if (SOLISTENING(so)) {
224 error = EINVAL;
225 } else {
226 *(int *)data = sbavail(&so->so_snd);
227 }
228 break;
229
230 case FIONSPACE:
231 /* Unlocked read. */
232 if (SOLISTENING(so)) {
233 error = EINVAL;
234 } else {
235 if ((so->so_snd.sb_hiwat < sbused(&so->so_snd)) ||
236 (so->so_snd.sb_mbmax < so->so_snd.sb_mbcnt)) {
237 *(int *)data = 0;
238 } else {
239 *(int *)data = sbspace(&so->so_snd);
240 }
241 }
242 break;
243
244 case FIOSETOWN:
245 error = fsetown(*(int *)data, &so->so_sigio);
246 break;
247
248 case FIOGETOWN:
249 *(int *)data = fgetown(&so->so_sigio);
250 break;
251
252 case SIOCSPGRP:
253 error = fsetown(-(*(int *)data), &so->so_sigio);
254 break;
255
256 case SIOCGPGRP:
257 *(int *)data = -fgetown(&so->so_sigio);
258 break;
259
260 case SIOCATMARK:
261 /* Unlocked read. */
262 if (SOLISTENING(so)) {
263 error = EINVAL;
264 } else {
265 *(int *)data = (so->so_rcv.sb_state & SBS_RCVATMARK) != 0;
266 }
267 break;
268 default:
269 /*
270 * Interface/routing/protocol specific ioctls: interface and
271 * routing ioctls should have a different entry since a
272 * socket is unnecessary.
273 */
274 if (IOCGROUP(cmd) == 'i')
275 error = ifioctl(so, cmd, data, td);
276 else if (IOCGROUP(cmd) == 'r') {
277 CURVNET_SET(so->so_vnet);
278 error = rtioctl_fib(cmd, data, so->so_fibnum);
279 CURVNET_RESTORE();
280 } else {
281 CURVNET_SET(so->so_vnet);
282 error = ((*so->so_proto->pr_usrreqs->pru_control)
283 (so, cmd, data, 0, td));
284 CURVNET_RESTORE();
285 }
286 break;
287 }
288 return (error);
289 }
290
291 static int
soo_poll(struct file * fp,int events,struct ucred * active_cred,struct thread * td)292 soo_poll(struct file *fp, int events, struct ucred *active_cred,
293 struct thread *td)
294 {
295 struct socket *so = fp->f_data;
296 #ifdef MAC
297 int error;
298
299 error = mac_socket_check_poll(active_cred, so);
300 if (error)
301 return (error);
302 #endif
303 return (sopoll(so, events, fp->f_cred, td));
304 }
305
306 static int
soo_stat(struct file * fp,struct stat * ub,struct ucred * active_cred,struct thread * td)307 soo_stat(struct file *fp, struct stat *ub, struct ucred *active_cred,
308 struct thread *td)
309 {
310 struct socket *so = fp->f_data;
311 int error;
312
313 bzero((caddr_t)ub, sizeof (*ub));
314 ub->st_mode = S_IFSOCK;
315 #ifdef MAC
316 error = mac_socket_check_stat(active_cred, so);
317 if (error)
318 return (error);
319 #endif
320 SOCK_LOCK(so);
321 if (!SOLISTENING(so)) {
322 struct sockbuf *sb;
323
324 /*
325 * If SBS_CANTRCVMORE is set, but there's still data left
326 * in the receive buffer, the socket is still readable.
327 */
328 sb = &so->so_rcv;
329 SOCKBUF_LOCK(sb);
330 if ((sb->sb_state & SBS_CANTRCVMORE) == 0 || sbavail(sb))
331 ub->st_mode |= S_IRUSR | S_IRGRP | S_IROTH;
332 ub->st_size = sbavail(sb) - sb->sb_ctl;
333 SOCKBUF_UNLOCK(sb);
334
335 sb = &so->so_snd;
336 SOCKBUF_LOCK(sb);
337 if ((sb->sb_state & SBS_CANTSENDMORE) == 0)
338 ub->st_mode |= S_IWUSR | S_IWGRP | S_IWOTH;
339 SOCKBUF_UNLOCK(sb);
340 }
341 ub->st_uid = so->so_cred->cr_uid;
342 ub->st_gid = so->so_cred->cr_gid;
343 error = so->so_proto->pr_usrreqs->pru_sense(so, ub);
344 SOCK_UNLOCK(so);
345 return (error);
346 }
347
348 /*
349 * API socket close on file pointer. We call soclose() to close the socket
350 * (including initiating closing protocols). soclose() will sorele() the
351 * file reference but the actual socket will not go away until the socket's
352 * ref count hits 0.
353 */
354 static int
soo_close(struct file * fp,struct thread * td)355 soo_close(struct file *fp, struct thread *td)
356 {
357 int error = 0;
358 struct socket *so;
359
360 so = fp->f_data;
361 fp->f_ops = &badfileops;
362 fp->f_data = NULL;
363
364 if (so)
365 error = soclose(so);
366 return (error);
367 }
368
369 #ifndef FSTACK
370 static int
soo_fill_kinfo(struct file * fp,struct kinfo_file * kif,struct filedesc * fdp)371 soo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
372 {
373 struct sockaddr *sa;
374 struct inpcb *inpcb;
375 struct unpcb *unpcb;
376 struct socket *so;
377 int error;
378
379 kif->kf_type = KF_TYPE_SOCKET;
380 so = fp->f_data;
381 CURVNET_SET(so->so_vnet);
382 kif->kf_un.kf_sock.kf_sock_domain0 =
383 so->so_proto->pr_domain->dom_family;
384 kif->kf_un.kf_sock.kf_sock_type0 = so->so_type;
385 kif->kf_un.kf_sock.kf_sock_protocol0 = so->so_proto->pr_protocol;
386 kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb;
387 switch (kif->kf_un.kf_sock.kf_sock_domain0) {
388 case AF_INET:
389 case AF_INET6:
390 if (kif->kf_un.kf_sock.kf_sock_protocol0 == IPPROTO_TCP) {
391 if (so->so_pcb != NULL) {
392 inpcb = (struct inpcb *)(so->so_pcb);
393 kif->kf_un.kf_sock.kf_sock_inpcb =
394 (uintptr_t)inpcb->inp_ppcb;
395 kif->kf_un.kf_sock.kf_sock_sendq =
396 sbused(&so->so_snd);
397 kif->kf_un.kf_sock.kf_sock_recvq =
398 sbused(&so->so_rcv);
399 }
400 }
401 break;
402 case AF_UNIX:
403 if (so->so_pcb != NULL) {
404 unpcb = (struct unpcb *)(so->so_pcb);
405 if (unpcb->unp_conn) {
406 kif->kf_un.kf_sock.kf_sock_unpconn =
407 (uintptr_t)unpcb->unp_conn;
408 kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
409 so->so_rcv.sb_state;
410 kif->kf_un.kf_sock.kf_sock_snd_sb_state =
411 so->so_snd.sb_state;
412 kif->kf_un.kf_sock.kf_sock_sendq =
413 sbused(&so->so_snd);
414 kif->kf_un.kf_sock.kf_sock_recvq =
415 sbused(&so->so_rcv);
416 }
417 }
418 break;
419 }
420 error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa);
421 if (error == 0 &&
422 sa->sa_len <= sizeof(kif->kf_un.kf_sock.kf_sa_local)) {
423 bcopy(sa, &kif->kf_un.kf_sock.kf_sa_local, sa->sa_len);
424 free(sa, M_SONAME);
425 }
426 error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa);
427 if (error == 0 &&
428 sa->sa_len <= sizeof(kif->kf_un.kf_sock.kf_sa_peer)) {
429 bcopy(sa, &kif->kf_un.kf_sock.kf_sa_peer, sa->sa_len);
430 free(sa, M_SONAME);
431 }
432 strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name,
433 sizeof(kif->kf_path));
434 CURVNET_RESTORE();
435 return (0);
436 }
437
438 /*
439 * Use the 'backend3' field in AIO jobs to store the amount of data
440 * completed by the AIO job so far.
441 */
442 #define aio_done backend3
443
444 static STAILQ_HEAD(, task) soaio_jobs;
445 static struct mtx soaio_jobs_lock;
446 static struct task soaio_kproc_task;
447 static int soaio_starting, soaio_idle, soaio_queued;
448 static struct unrhdr *soaio_kproc_unr;
449
450 static int soaio_max_procs = MAX_AIO_PROCS;
451 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, max_procs, CTLFLAG_RW, &soaio_max_procs, 0,
452 "Maximum number of kernel processes to use for async socket IO");
453
454 static int soaio_num_procs;
455 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, num_procs, CTLFLAG_RD, &soaio_num_procs, 0,
456 "Number of active kernel processes for async socket IO");
457
458 static int soaio_target_procs = TARGET_AIO_PROCS;
459 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, target_procs, CTLFLAG_RD,
460 &soaio_target_procs, 0,
461 "Preferred number of ready kernel processes for async socket IO");
462
463 static int soaio_lifetime;
464 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, lifetime, CTLFLAG_RW, &soaio_lifetime, 0,
465 "Maximum lifetime for idle aiod");
466
467 static void
soaio_kproc_loop(void * arg)468 soaio_kproc_loop(void *arg)
469 {
470 struct proc *p;
471 struct vmspace *myvm;
472 struct task *task;
473 int error, id, pending;
474
475 id = (intptr_t)arg;
476
477 /*
478 * Grab an extra reference on the daemon's vmspace so that it
479 * doesn't get freed by jobs that switch to a different
480 * vmspace.
481 */
482 p = curproc;
483 myvm = vmspace_acquire_ref(p);
484
485 mtx_lock(&soaio_jobs_lock);
486 MPASS(soaio_starting > 0);
487 soaio_starting--;
488 for (;;) {
489 while (!STAILQ_EMPTY(&soaio_jobs)) {
490 task = STAILQ_FIRST(&soaio_jobs);
491 STAILQ_REMOVE_HEAD(&soaio_jobs, ta_link);
492 soaio_queued--;
493 pending = task->ta_pending;
494 task->ta_pending = 0;
495 mtx_unlock(&soaio_jobs_lock);
496
497 task->ta_func(task->ta_context, pending);
498
499 mtx_lock(&soaio_jobs_lock);
500 }
501 MPASS(soaio_queued == 0);
502
503 if (p->p_vmspace != myvm) {
504 mtx_unlock(&soaio_jobs_lock);
505 vmspace_switch_aio(myvm);
506 mtx_lock(&soaio_jobs_lock);
507 continue;
508 }
509
510 soaio_idle++;
511 error = mtx_sleep(&soaio_idle, &soaio_jobs_lock, 0, "-",
512 soaio_lifetime);
513 soaio_idle--;
514 if (error == EWOULDBLOCK && STAILQ_EMPTY(&soaio_jobs) &&
515 soaio_num_procs > soaio_target_procs)
516 break;
517 }
518 soaio_num_procs--;
519 mtx_unlock(&soaio_jobs_lock);
520 free_unr(soaio_kproc_unr, id);
521 kproc_exit(0);
522 }
523
524 static void
soaio_kproc_create(void * context,int pending)525 soaio_kproc_create(void *context, int pending)
526 {
527 struct proc *p;
528 int error, id;
529
530 mtx_lock(&soaio_jobs_lock);
531 for (;;) {
532 if (soaio_num_procs < soaio_target_procs) {
533 /* Must create */
534 } else if (soaio_num_procs >= soaio_max_procs) {
535 /*
536 * Hit the limit on kernel processes, don't
537 * create another one.
538 */
539 break;
540 } else if (soaio_queued <= soaio_idle + soaio_starting) {
541 /*
542 * No more AIO jobs waiting for a process to be
543 * created, so stop.
544 */
545 break;
546 }
547 soaio_starting++;
548 mtx_unlock(&soaio_jobs_lock);
549
550 id = alloc_unr(soaio_kproc_unr);
551 error = kproc_create(soaio_kproc_loop, (void *)(intptr_t)id,
552 &p, 0, 0, "soaiod%d", id);
553 if (error != 0) {
554 free_unr(soaio_kproc_unr, id);
555 mtx_lock(&soaio_jobs_lock);
556 soaio_starting--;
557 break;
558 }
559
560 mtx_lock(&soaio_jobs_lock);
561 soaio_num_procs++;
562 }
563 mtx_unlock(&soaio_jobs_lock);
564 }
565
566 void
soaio_enqueue(struct task * task)567 soaio_enqueue(struct task *task)
568 {
569
570 mtx_lock(&soaio_jobs_lock);
571 MPASS(task->ta_pending == 0);
572 task->ta_pending++;
573 STAILQ_INSERT_TAIL(&soaio_jobs, task, ta_link);
574 soaio_queued++;
575 if (soaio_queued <= soaio_idle)
576 wakeup_one(&soaio_idle);
577 else if (soaio_num_procs < soaio_max_procs)
578 taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task);
579 mtx_unlock(&soaio_jobs_lock);
580 }
581
582 static void
soaio_init(void)583 soaio_init(void)
584 {
585
586 soaio_lifetime = AIOD_LIFETIME_DEFAULT;
587 STAILQ_INIT(&soaio_jobs);
588 mtx_init(&soaio_jobs_lock, "soaio jobs", NULL, MTX_DEF);
589 soaio_kproc_unr = new_unrhdr(1, INT_MAX, NULL);
590 TASK_INIT(&soaio_kproc_task, 0, soaio_kproc_create, NULL);
591 if (soaio_target_procs > 0)
592 taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task);
593 }
594 SYSINIT(soaio, SI_SUB_VFS, SI_ORDER_ANY, soaio_init, NULL);
595
596 static __inline int
soaio_ready(struct socket * so,struct sockbuf * sb)597 soaio_ready(struct socket *so, struct sockbuf *sb)
598 {
599 return (sb == &so->so_rcv ? soreadable(so) : sowriteable(so));
600 }
601
602 static void
soaio_process_job(struct socket * so,struct sockbuf * sb,struct kaiocb * job)603 soaio_process_job(struct socket *so, struct sockbuf *sb, struct kaiocb *job)
604 {
605 struct ucred *td_savedcred;
606 struct thread *td;
607 struct file *fp;
608 size_t cnt, done, job_total_nbytes;
609 long ru_before;
610 int error, flags;
611
612 SOCKBUF_UNLOCK(sb);
613 aio_switch_vmspace(job);
614 td = curthread;
615 fp = job->fd_file;
616 retry:
617 td_savedcred = td->td_ucred;
618 td->td_ucred = job->cred;
619
620 job_total_nbytes = job->uiop->uio_resid + job->aio_done;
621 done = job->aio_done;
622 cnt = job->uiop->uio_resid;
623 job->uiop->uio_offset = 0;
624 job->uiop->uio_td = td;
625 flags = MSG_NBIO;
626
627 /*
628 * For resource usage accounting, only count a completed request
629 * as a single message to avoid counting multiple calls to
630 * sosend/soreceive on a blocking socket.
631 */
632
633 if (sb == &so->so_rcv) {
634 ru_before = td->td_ru.ru_msgrcv;
635 #ifdef MAC
636 error = mac_socket_check_receive(fp->f_cred, so);
637 if (error == 0)
638
639 #endif
640 error = soreceive(so, NULL, job->uiop, NULL, NULL,
641 &flags);
642 if (td->td_ru.ru_msgrcv != ru_before)
643 job->msgrcv = 1;
644 } else {
645 if (!TAILQ_EMPTY(&sb->sb_aiojobq))
646 flags |= MSG_MORETOCOME;
647 ru_before = td->td_ru.ru_msgsnd;
648 #ifdef MAC
649 error = mac_socket_check_send(fp->f_cred, so);
650 if (error == 0)
651 #endif
652 error = sosend(so, NULL, job->uiop, NULL, NULL, flags,
653 td);
654 if (td->td_ru.ru_msgsnd != ru_before)
655 job->msgsnd = 1;
656 if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0) {
657 PROC_LOCK(job->userproc);
658 kern_psignal(job->userproc, SIGPIPE);
659 PROC_UNLOCK(job->userproc);
660 }
661 }
662
663 done += cnt - job->uiop->uio_resid;
664 job->aio_done = done;
665 td->td_ucred = td_savedcred;
666
667 if (error == EWOULDBLOCK) {
668 /*
669 * The request was either partially completed or not
670 * completed at all due to racing with a read() or
671 * write() on the socket. If the socket is
672 * non-blocking, return with any partial completion.
673 * If the socket is blocking or if no progress has
674 * been made, requeue this request at the head of the
675 * queue to try again when the socket is ready.
676 */
677 MPASS(done != job_total_nbytes);
678 SOCKBUF_LOCK(sb);
679 if (done == 0 || !(so->so_state & SS_NBIO)) {
680 empty_results++;
681 if (soaio_ready(so, sb)) {
682 empty_retries++;
683 SOCKBUF_UNLOCK(sb);
684 goto retry;
685 }
686
687 if (!aio_set_cancel_function(job, soo_aio_cancel)) {
688 SOCKBUF_UNLOCK(sb);
689 if (done != 0)
690 aio_complete(job, done, 0);
691 else
692 aio_cancel(job);
693 SOCKBUF_LOCK(sb);
694 } else {
695 TAILQ_INSERT_HEAD(&sb->sb_aiojobq, job, list);
696 }
697 return;
698 }
699 SOCKBUF_UNLOCK(sb);
700 }
701 if (done != 0 && (error == ERESTART || error == EINTR ||
702 error == EWOULDBLOCK))
703 error = 0;
704 if (error)
705 aio_complete(job, -1, error);
706 else
707 aio_complete(job, done, 0);
708 SOCKBUF_LOCK(sb);
709 }
710
711 static void
soaio_process_sb(struct socket * so,struct sockbuf * sb)712 soaio_process_sb(struct socket *so, struct sockbuf *sb)
713 {
714 struct kaiocb *job;
715
716 CURVNET_SET(so->so_vnet);
717 SOCKBUF_LOCK(sb);
718 while (!TAILQ_EMPTY(&sb->sb_aiojobq) && soaio_ready(so, sb)) {
719 job = TAILQ_FIRST(&sb->sb_aiojobq);
720 TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
721 if (!aio_clear_cancel_function(job))
722 continue;
723
724 soaio_process_job(so, sb, job);
725 }
726
727 /*
728 * If there are still pending requests, the socket must not be
729 * ready so set SB_AIO to request a wakeup when the socket
730 * becomes ready.
731 */
732 if (!TAILQ_EMPTY(&sb->sb_aiojobq))
733 sb->sb_flags |= SB_AIO;
734 sb->sb_flags &= ~SB_AIO_RUNNING;
735 SOCKBUF_UNLOCK(sb);
736
737 SOCK_LOCK(so);
738 sorele(so);
739 CURVNET_RESTORE();
740 }
741
742 void
soaio_rcv(void * context,int pending)743 soaio_rcv(void *context, int pending)
744 {
745 struct socket *so;
746
747 so = context;
748 soaio_process_sb(so, &so->so_rcv);
749 }
750
751 void
soaio_snd(void * context,int pending)752 soaio_snd(void *context, int pending)
753 {
754 struct socket *so;
755
756 so = context;
757 soaio_process_sb(so, &so->so_snd);
758 }
759
760 void
sowakeup_aio(struct socket * so,struct sockbuf * sb)761 sowakeup_aio(struct socket *so, struct sockbuf *sb)
762 {
763
764 SOCKBUF_LOCK_ASSERT(sb);
765 sb->sb_flags &= ~SB_AIO;
766 if (sb->sb_flags & SB_AIO_RUNNING)
767 return;
768 sb->sb_flags |= SB_AIO_RUNNING;
769 soref(so);
770 soaio_enqueue(&sb->sb_aiotask);
771 }
772
773 static void
soo_aio_cancel(struct kaiocb * job)774 soo_aio_cancel(struct kaiocb *job)
775 {
776 struct socket *so;
777 struct sockbuf *sb;
778 long done;
779 int opcode;
780
781 so = job->fd_file->f_data;
782 opcode = job->uaiocb.aio_lio_opcode;
783 if (opcode & LIO_READ)
784 sb = &so->so_rcv;
785 else {
786 MPASS(opcode & LIO_WRITE);
787 sb = &so->so_snd;
788 }
789
790 SOCKBUF_LOCK(sb);
791 if (!aio_cancel_cleared(job))
792 TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
793 if (TAILQ_EMPTY(&sb->sb_aiojobq))
794 sb->sb_flags &= ~SB_AIO;
795 SOCKBUF_UNLOCK(sb);
796
797 done = job->aio_done;
798 if (done != 0)
799 aio_complete(job, done, 0);
800 else
801 aio_cancel(job);
802 }
803
804 static int
soo_aio_queue(struct file * fp,struct kaiocb * job)805 soo_aio_queue(struct file *fp, struct kaiocb *job)
806 {
807 struct socket *so;
808 struct sockbuf *sb;
809 int error;
810
811 so = fp->f_data;
812 error = (*so->so_proto->pr_usrreqs->pru_aio_queue)(so, job);
813 if (error == 0)
814 return (0);
815
816 switch (job->uaiocb.aio_lio_opcode & (LIO_WRITE | LIO_READ)) {
817 case LIO_READ:
818 sb = &so->so_rcv;
819 break;
820 case LIO_WRITE:
821 sb = &so->so_snd;
822 break;
823 default:
824 return (EINVAL);
825 }
826
827 SOCKBUF_LOCK(sb);
828 if (!aio_set_cancel_function(job, soo_aio_cancel))
829 panic("new job was cancelled");
830 TAILQ_INSERT_TAIL(&sb->sb_aiojobq, job, list);
831 if (!(sb->sb_flags & SB_AIO_RUNNING)) {
832 if (soaio_ready(so, sb))
833 sowakeup_aio(so, sb);
834 else
835 sb->sb_flags |= SB_AIO;
836 }
837 SOCKBUF_UNLOCK(sb);
838 return (0);
839 }
840 #endif
841