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
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/aio.h>
35 #include <sys/domain.h>
36 #include <sys/file.h>
37 #include <sys/filedesc.h>
38 #include <sys/kernel.h>
39 #include <sys/kthread.h>
40 #include <sys/malloc.h>
41 #include <sys/proc.h>
42 #include <sys/protosw.h>
43 #include <sys/sigio.h>
44 #include <sys/signal.h>
45 #include <sys/signalvar.h>
46 #include <sys/socket.h>
47 #include <sys/socketvar.h>
48 #include <sys/filio.h> /* XXX */
49 #include <sys/sockio.h>
50 #include <sys/stat.h>
51 #include <sys/sysctl.h>
52 #include <sys/sysproto.h>
53 #include <sys/taskqueue.h>
54 #include <sys/uio.h>
55 #include <sys/ucred.h>
56 #include <sys/un.h>
57 #include <sys/unpcb.h>
58 #include <sys/user.h>
59
60 #include <net/if.h>
61 #include <net/if_var.h>
62 #include <net/route.h>
63 #include <net/vnet.h>
64
65 #include <netinet/in.h>
66 #include <netinet/in_pcb.h>
67
68 #include <security/mac/mac_framework.h>
69
70 #include <vm/vm.h>
71 #include <vm/pmap.h>
72 #include <vm/vm_extern.h>
73 #include <vm/vm_map.h>
74
75 static SYSCTL_NODE(_kern_ipc, OID_AUTO, aio, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
76 "socket AIO stats");
77
78 static int empty_results;
79 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, empty_results, CTLFLAG_RD, &empty_results,
80 0, "socket operation returned EAGAIN");
81
82 static int empty_retries;
83 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, empty_retries, CTLFLAG_RD, &empty_retries,
84 0, "socket operation retries");
85
86 static fo_rdwr_t soo_read;
87 static fo_rdwr_t soo_write;
88 static fo_ioctl_t soo_ioctl;
89 static fo_poll_t soo_poll;
90 extern fo_kqfilter_t soo_kqfilter;
91 static fo_stat_t soo_stat;
92 static fo_close_t soo_close;
93 static fo_chmod_t soo_chmod;
94 static fo_fill_kinfo_t soo_fill_kinfo;
95 static fo_aio_queue_t soo_aio_queue;
96
97 static void soo_aio_cancel(struct kaiocb *job);
98
99 struct fileops socketops = {
100 .fo_read = soo_read,
101 .fo_write = soo_write,
102 .fo_truncate = invfo_truncate,
103 .fo_ioctl = soo_ioctl,
104 .fo_poll = soo_poll,
105 .fo_kqfilter = soo_kqfilter,
106 .fo_stat = soo_stat,
107 .fo_close = soo_close,
108 .fo_chmod = soo_chmod,
109 .fo_chown = invfo_chown,
110 .fo_sendfile = invfo_sendfile,
111 .fo_fill_kinfo = soo_fill_kinfo,
112 .fo_aio_queue = soo_aio_queue,
113 .fo_cmp = file_kcmp_generic,
114 .fo_flags = DFLAG_PASSABLE
115 };
116
117 static int
soo_read(struct file * fp,struct uio * uio,struct ucred * active_cred,int flags,struct thread * td)118 soo_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
119 int flags, struct thread *td)
120 {
121 struct socket *so = fp->f_data;
122 int error;
123
124 #ifdef MAC
125 error = mac_socket_check_receive(active_cred, so);
126 if (error)
127 return (error);
128 #endif
129 error = soreceive(so, 0, uio, 0, 0, 0);
130 return (error);
131 }
132
133 static int
soo_write(struct file * fp,struct uio * uio,struct ucred * active_cred,int flags,struct thread * td)134 soo_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
135 int flags, struct thread *td)
136 {
137 struct socket *so = fp->f_data;
138 int error;
139
140 #ifdef MAC
141 error = mac_socket_check_send(active_cred, so);
142 if (error)
143 return (error);
144 #endif
145 error = sousrsend(so, NULL, uio, NULL, 0, NULL);
146 return (error);
147 }
148
149 static int
soo_ioctl(struct file * fp,u_long cmd,void * data,struct ucred * active_cred,struct thread * td)150 soo_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *active_cred,
151 struct thread *td)
152 {
153 struct socket *so = fp->f_data;
154 int error = 0;
155
156 switch (cmd) {
157 case FIONBIO:
158 SOCK_LOCK(so);
159 if (*(int *)data)
160 so->so_state |= SS_NBIO;
161 else
162 so->so_state &= ~SS_NBIO;
163 SOCK_UNLOCK(so);
164 break;
165
166 case FIOASYNC:
167 if (*(int *)data) {
168 SOCK_LOCK(so);
169 so->so_state |= SS_ASYNC;
170 if (SOLISTENING(so)) {
171 so->sol_sbrcv_flags |= SB_ASYNC;
172 so->sol_sbsnd_flags |= SB_ASYNC;
173 } else {
174 SOCK_RECVBUF_LOCK(so);
175 so->so_rcv.sb_flags |= SB_ASYNC;
176 SOCK_RECVBUF_UNLOCK(so);
177 SOCK_SENDBUF_LOCK(so);
178 so->so_snd.sb_flags |= SB_ASYNC;
179 SOCK_SENDBUF_UNLOCK(so);
180 }
181 SOCK_UNLOCK(so);
182 } else {
183 SOCK_LOCK(so);
184 so->so_state &= ~SS_ASYNC;
185 if (SOLISTENING(so)) {
186 so->sol_sbrcv_flags &= ~SB_ASYNC;
187 so->sol_sbsnd_flags &= ~SB_ASYNC;
188 } else {
189 SOCK_RECVBUF_LOCK(so);
190 so->so_rcv.sb_flags &= ~SB_ASYNC;
191 SOCK_RECVBUF_UNLOCK(so);
192 SOCK_SENDBUF_LOCK(so);
193 so->so_snd.sb_flags &= ~SB_ASYNC;
194 SOCK_SENDBUF_UNLOCK(so);
195 }
196 SOCK_UNLOCK(so);
197 }
198 break;
199
200 case FIONREAD:
201 SOCK_RECVBUF_LOCK(so);
202 if (SOLISTENING(so)) {
203 error = EINVAL;
204 } else {
205 *(int *)data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
206 }
207 SOCK_RECVBUF_UNLOCK(so);
208 break;
209
210 case FIONWRITE:
211 /* Unlocked read. */
212 if (SOLISTENING(so)) {
213 error = EINVAL;
214 } else {
215 *(int *)data = sbavail(&so->so_snd);
216 }
217 break;
218
219 case FIONSPACE:
220 /* Unlocked read. */
221 if (SOLISTENING(so)) {
222 error = EINVAL;
223 } else {
224 if ((so->so_snd.sb_hiwat < sbused(&so->so_snd)) ||
225 (so->so_snd.sb_mbmax < so->so_snd.sb_mbcnt)) {
226 *(int *)data = 0;
227 } else {
228 *(int *)data = sbspace(&so->so_snd);
229 }
230 }
231 break;
232
233 case FIOSETOWN:
234 error = fsetown(*(int *)data, &so->so_sigio);
235 break;
236
237 case FIOGETOWN:
238 *(int *)data = fgetown(&so->so_sigio);
239 break;
240
241 case SIOCSPGRP:
242 error = fsetown(-(*(int *)data), &so->so_sigio);
243 break;
244
245 case SIOCGPGRP:
246 *(int *)data = -fgetown(&so->so_sigio);
247 break;
248
249 case SIOCATMARK:
250 /* Unlocked read. */
251 if (SOLISTENING(so)) {
252 error = EINVAL;
253 } else {
254 *(int *)data = (so->so_rcv.sb_state & SBS_RCVATMARK) != 0;
255 }
256 break;
257 default:
258 /*
259 * Interface/routing/protocol specific ioctls: interface and
260 * routing ioctls should have a different entry since a
261 * socket is unnecessary.
262 */
263 if (IOCGROUP(cmd) == 'i')
264 error = ifioctl(so, cmd, data, td);
265 else if (IOCGROUP(cmd) == 'r') {
266 CURVNET_SET(so->so_vnet);
267 error = rtioctl_fib(cmd, data, so->so_fibnum);
268 CURVNET_RESTORE();
269 } else {
270 CURVNET_SET(so->so_vnet);
271 error = so->so_proto->pr_control(so, cmd, data, 0, td);
272 CURVNET_RESTORE();
273 }
274 break;
275 }
276 return (error);
277 }
278
279 static int
soo_poll(struct file * fp,int events,struct ucred * active_cred,struct thread * td)280 soo_poll(struct file *fp, int events, struct ucred *active_cred,
281 struct thread *td)
282 {
283 struct socket *so = fp->f_data;
284 #ifdef MAC
285 int error;
286
287 error = mac_socket_check_poll(active_cred, so);
288 if (error)
289 return (error);
290 #endif
291 return (sopoll(so, events, fp->f_cred, td));
292 }
293
294 static int
soo_stat(struct file * fp,struct stat * ub,struct ucred * active_cred)295 soo_stat(struct file *fp, struct stat *ub, struct ucred *active_cred)
296 {
297 struct socket *so = fp->f_data;
298 int error = 0;
299
300 bzero((caddr_t)ub, sizeof (*ub));
301 ub->st_mode = S_IFSOCK;
302 #ifdef MAC
303 error = mac_socket_check_stat(active_cred, so);
304 if (error)
305 return (error);
306 #endif
307 SOCK_LOCK(so);
308 if (!SOLISTENING(so)) {
309 struct sockbuf *sb;
310
311 /*
312 * If SBS_CANTRCVMORE is set, but there's still data left
313 * in the receive buffer, the socket is still readable.
314 */
315 sb = &so->so_rcv;
316 SOCK_RECVBUF_LOCK(so);
317 if ((sb->sb_state & SBS_CANTRCVMORE) == 0 || sbavail(sb))
318 ub->st_mode |= S_IRUSR | S_IRGRP | S_IROTH;
319 ub->st_size = sbavail(sb) - sb->sb_ctl;
320 SOCK_RECVBUF_UNLOCK(so);
321
322 sb = &so->so_snd;
323 SOCK_SENDBUF_LOCK(so);
324 if ((sb->sb_state & SBS_CANTSENDMORE) == 0)
325 ub->st_mode |= S_IWUSR | S_IWGRP | S_IWOTH;
326 SOCK_SENDBUF_UNLOCK(so);
327 }
328 ub->st_uid = so->so_cred->cr_uid;
329 ub->st_gid = so->so_cred->cr_gid;
330 if (so->so_proto->pr_sense)
331 error = so->so_proto->pr_sense(so, ub);
332 SOCK_UNLOCK(so);
333 return (error);
334 }
335
336 /*
337 * API socket close on file pointer. We call soclose() to close the socket
338 * (including initiating closing protocols). soclose() will sorele() the
339 * file reference but the actual socket will not go away until the socket's
340 * ref count hits 0.
341 */
342 static int
soo_close(struct file * fp,struct thread * td)343 soo_close(struct file *fp, struct thread *td)
344 {
345 int error = 0;
346 struct socket *so;
347
348 so = fp->f_data;
349 fp->f_ops = &badfileops;
350 fp->f_data = NULL;
351
352 if (so)
353 error = soclose(so);
354 return (error);
355 }
356
357 static int
soo_chmod(struct file * fp,mode_t mode,struct ucred * cred,struct thread * td)358 soo_chmod(struct file *fp, mode_t mode, struct ucred *cred, struct thread *td)
359 {
360 struct socket *so;
361 int error;
362
363 so = fp->f_data;
364 if (so->so_proto->pr_chmod != NULL)
365 error = so->so_proto->pr_chmod(so, mode, cred, td);
366 else
367 error = EINVAL;
368 return (error);
369 }
370
371 static int
soo_fill_kinfo(struct file * fp,struct kinfo_file * kif,struct filedesc * fdp)372 soo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
373 {
374 struct sockaddr_storage ss = { .ss_len = sizeof(ss) };
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 /* XXX: kf_sock_inpcb is obsolete. It may be removed. */
391 kif->kf_un.kf_sock.kf_sock_inpcb = (uintptr_t)so->so_pcb;
392 kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
393 so->so_rcv.sb_state;
394 kif->kf_un.kf_sock.kf_sock_snd_sb_state =
395 so->so_snd.sb_state;
396 kif->kf_un.kf_sock.kf_sock_sendq =
397 sbused(&so->so_snd);
398 kif->kf_un.kf_sock.kf_sock_recvq =
399 sbused(&so->so_rcv);
400 break;
401 case AF_UNIX:
402 if (so->so_pcb != NULL) {
403 unpcb = (struct unpcb *)(so->so_pcb);
404 if (unpcb->unp_conn) {
405 kif->kf_un.kf_sock.kf_sock_unpconn =
406 (uintptr_t)unpcb->unp_conn;
407 kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
408 so->so_rcv.sb_state;
409 kif->kf_un.kf_sock.kf_sock_snd_sb_state =
410 so->so_snd.sb_state;
411 kif->kf_un.kf_sock.kf_sock_sendq =
412 sbused(&so->so_snd);
413 kif->kf_un.kf_sock.kf_sock_recvq =
414 sbused(&so->so_rcv);
415 }
416 }
417 break;
418 }
419 error = sosockaddr(so, (struct sockaddr *)&ss);
420 if (error == 0 &&
421 ss.ss_len <= sizeof(kif->kf_un.kf_sock.kf_sa_local)) {
422 bcopy(&ss, &kif->kf_un.kf_sock.kf_sa_local, ss.ss_len);
423 }
424 ss.ss_len = sizeof(ss);
425 error = sopeeraddr(so, (struct sockaddr *)&ss);
426 if (error == 0 &&
427 ss.ss_len <= sizeof(kif->kf_un.kf_sock.kf_sa_peer)) {
428 bcopy(&ss, &kif->kf_un.kf_sock.kf_sa_peer, ss.ss_len);
429 }
430 strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name,
431 sizeof(kif->kf_path));
432 CURVNET_RESTORE();
433 return (0);
434 }
435
436 /*
437 * Use the 'backend3' field in AIO jobs to store the amount of data
438 * completed by the AIO job so far.
439 */
440 #define aio_done backend3
441
442 static STAILQ_HEAD(, task) soaio_jobs;
443 static struct mtx soaio_jobs_lock;
444 static struct task soaio_kproc_task;
445 static int soaio_starting, soaio_idle, soaio_queued;
446 static struct unrhdr *soaio_kproc_unr;
447
448 static int soaio_max_procs = MAX_AIO_PROCS;
449 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, max_procs, CTLFLAG_RW, &soaio_max_procs, 0,
450 "Maximum number of kernel processes to use for async socket IO");
451
452 static int soaio_num_procs;
453 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, num_procs, CTLFLAG_RD, &soaio_num_procs, 0,
454 "Number of active kernel processes for async socket IO");
455
456 static int soaio_target_procs = TARGET_AIO_PROCS;
457 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, target_procs, CTLFLAG_RD,
458 &soaio_target_procs, 0,
459 "Preferred number of ready kernel processes for async socket IO");
460
461 static int soaio_lifetime;
462 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, lifetime, CTLFLAG_RW, &soaio_lifetime, 0,
463 "Maximum lifetime for idle aiod");
464
465 static void
soaio_kproc_loop(void * arg)466 soaio_kproc_loop(void *arg)
467 {
468 struct proc *p;
469 struct vmspace *myvm;
470 struct task *task;
471 int error, id, pending;
472
473 id = (intptr_t)arg;
474
475 /*
476 * Grab an extra reference on the daemon's vmspace so that it
477 * doesn't get freed by jobs that switch to a different
478 * vmspace.
479 */
480 p = curproc;
481 myvm = vmspace_acquire_ref(p);
482
483 mtx_lock(&soaio_jobs_lock);
484 MPASS(soaio_starting > 0);
485 soaio_starting--;
486 for (;;) {
487 while (!STAILQ_EMPTY(&soaio_jobs)) {
488 task = STAILQ_FIRST(&soaio_jobs);
489 STAILQ_REMOVE_HEAD(&soaio_jobs, ta_link);
490 soaio_queued--;
491 pending = task->ta_pending;
492 task->ta_pending = 0;
493 mtx_unlock(&soaio_jobs_lock);
494
495 task->ta_func(task->ta_context, pending);
496
497 mtx_lock(&soaio_jobs_lock);
498 }
499 MPASS(soaio_queued == 0);
500
501 if (p->p_vmspace != myvm) {
502 mtx_unlock(&soaio_jobs_lock);
503 vmspace_switch_aio(myvm);
504 mtx_lock(&soaio_jobs_lock);
505 continue;
506 }
507
508 soaio_idle++;
509 error = mtx_sleep(&soaio_idle, &soaio_jobs_lock, 0, "-",
510 soaio_lifetime);
511 soaio_idle--;
512 if (error == EWOULDBLOCK && STAILQ_EMPTY(&soaio_jobs) &&
513 soaio_num_procs > soaio_target_procs)
514 break;
515 }
516 soaio_num_procs--;
517 mtx_unlock(&soaio_jobs_lock);
518 free_unr(soaio_kproc_unr, id);
519 kproc_exit(0);
520 }
521
522 static void
soaio_kproc_create(void * context,int pending)523 soaio_kproc_create(void *context, int pending)
524 {
525 struct proc *p;
526 int error, id;
527
528 mtx_lock(&soaio_jobs_lock);
529 for (;;) {
530 if (soaio_num_procs < soaio_target_procs) {
531 /* Must create */
532 } else if (soaio_num_procs >= soaio_max_procs) {
533 /*
534 * Hit the limit on kernel processes, don't
535 * create another one.
536 */
537 break;
538 } else if (soaio_queued <= soaio_idle + soaio_starting) {
539 /*
540 * No more AIO jobs waiting for a process to be
541 * created, so stop.
542 */
543 break;
544 }
545 soaio_starting++;
546 mtx_unlock(&soaio_jobs_lock);
547
548 id = alloc_unr(soaio_kproc_unr);
549 error = kproc_create(soaio_kproc_loop, (void *)(intptr_t)id,
550 &p, 0, 0, "soaiod%d", id);
551 if (error != 0) {
552 free_unr(soaio_kproc_unr, id);
553 mtx_lock(&soaio_jobs_lock);
554 soaio_starting--;
555 break;
556 }
557
558 mtx_lock(&soaio_jobs_lock);
559 soaio_num_procs++;
560 }
561 mtx_unlock(&soaio_jobs_lock);
562 }
563
564 void
soaio_enqueue(struct task * task)565 soaio_enqueue(struct task *task)
566 {
567
568 mtx_lock(&soaio_jobs_lock);
569 MPASS(task->ta_pending == 0);
570 task->ta_pending++;
571 STAILQ_INSERT_TAIL(&soaio_jobs, task, ta_link);
572 soaio_queued++;
573 if (soaio_queued <= soaio_idle)
574 wakeup_one(&soaio_idle);
575 else if (soaio_num_procs < soaio_max_procs)
576 taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task);
577 mtx_unlock(&soaio_jobs_lock);
578 }
579
580 static void
soaio_init(void)581 soaio_init(void)
582 {
583
584 soaio_lifetime = AIOD_LIFETIME_DEFAULT;
585 STAILQ_INIT(&soaio_jobs);
586 mtx_init(&soaio_jobs_lock, "soaio jobs", NULL, MTX_DEF);
587 soaio_kproc_unr = new_unrhdr(1, INT_MAX, NULL);
588 TASK_INIT(&soaio_kproc_task, 0, soaio_kproc_create, NULL);
589 }
590 SYSINIT(soaio, SI_SUB_VFS, SI_ORDER_ANY, soaio_init, NULL);
591
592 static __inline int
soaio_ready(struct socket * so,struct sockbuf * sb)593 soaio_ready(struct socket *so, struct sockbuf *sb)
594 {
595 return (sb == &so->so_rcv ? soreadable(so) : sowriteable(so));
596 }
597
598 static void
soaio_process_job(struct socket * so,sb_which which,struct kaiocb * job)599 soaio_process_job(struct socket *so, sb_which which, struct kaiocb *job)
600 {
601 struct ucred *td_savedcred;
602 struct thread *td;
603 struct sockbuf *sb = sobuf(so, which);
604 #ifdef MAC
605 struct file *fp = job->fd_file;
606 #endif
607 size_t cnt, done, job_total_nbytes __diagused;
608 long ru_before;
609 int error, flags;
610
611 SOCK_BUF_UNLOCK(so, which);
612 aio_switch_vmspace(job);
613 td = curthread;
614 retry:
615 td_savedcred = td->td_ucred;
616 td->td_ucred = job->cred;
617
618 job_total_nbytes = job->uiop->uio_resid + job->aio_done;
619 done = job->aio_done;
620 cnt = job->uiop->uio_resid;
621 job->uiop->uio_offset = 0;
622 job->uiop->uio_td = td;
623 flags = MSG_NBIO;
624
625 /*
626 * For resource usage accounting, only count a completed request
627 * as a single message to avoid counting multiple calls to
628 * sosend/soreceive on a blocking socket.
629 */
630
631 if (sb == &so->so_rcv) {
632 ru_before = td->td_ru.ru_msgrcv;
633 #ifdef MAC
634 error = mac_socket_check_receive(fp->f_cred, so);
635 if (error == 0)
636
637 #endif
638 error = soreceive(so, NULL, job->uiop, NULL, NULL,
639 &flags);
640 if (td->td_ru.ru_msgrcv != ru_before)
641 job->msgrcv = 1;
642 } else {
643 if (!TAILQ_EMPTY(&sb->sb_aiojobq))
644 flags |= MSG_MORETOCOME;
645 ru_before = td->td_ru.ru_msgsnd;
646 #ifdef MAC
647 error = mac_socket_check_send(fp->f_cred, so);
648 if (error == 0)
649 #endif
650 error = sousrsend(so, NULL, job->uiop, NULL, flags,
651 job->userproc);
652 if (td->td_ru.ru_msgsnd != ru_before)
653 job->msgsnd = 1;
654 }
655
656 done += cnt - job->uiop->uio_resid;
657 job->aio_done = done;
658 td->td_ucred = td_savedcred;
659
660 if (error == EWOULDBLOCK) {
661 /*
662 * The request was either partially completed or not
663 * completed at all due to racing with a read() or
664 * write() on the socket. If the socket is
665 * non-blocking, return with any partial completion.
666 * If the socket is blocking or if no progress has
667 * been made, requeue this request at the head of the
668 * queue to try again when the socket is ready.
669 */
670 MPASS(done != job_total_nbytes);
671 SOCK_BUF_LOCK(so, which);
672 if (done == 0 || !(so->so_state & SS_NBIO)) {
673 empty_results++;
674 if (soaio_ready(so, sb)) {
675 empty_retries++;
676 SOCK_BUF_UNLOCK(so, which);
677 goto retry;
678 }
679
680 if (!aio_set_cancel_function(job, soo_aio_cancel)) {
681 SOCK_BUF_UNLOCK(so, which);
682 if (done != 0)
683 aio_complete(job, done, 0);
684 else
685 aio_cancel(job);
686 SOCK_BUF_LOCK(so, which);
687 } else {
688 TAILQ_INSERT_HEAD(&sb->sb_aiojobq, job, list);
689 }
690 return;
691 }
692 SOCK_BUF_UNLOCK(so, which);
693 }
694 if (done != 0 && (error == ERESTART || error == EINTR ||
695 error == EWOULDBLOCK))
696 error = 0;
697 if (error)
698 aio_complete(job, -1, error);
699 else
700 aio_complete(job, done, 0);
701 SOCK_BUF_LOCK(so, which);
702 }
703
704 static void
soaio_process_sb(struct socket * so,sb_which which)705 soaio_process_sb(struct socket *so, sb_which which)
706 {
707 struct kaiocb *job;
708 struct sockbuf *sb = sobuf(so, which);
709
710 CURVNET_SET(so->so_vnet);
711 SOCK_BUF_LOCK(so, which);
712 while (!TAILQ_EMPTY(&sb->sb_aiojobq) && soaio_ready(so, sb)) {
713 job = TAILQ_FIRST(&sb->sb_aiojobq);
714 TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
715 if (!aio_clear_cancel_function(job))
716 continue;
717
718 soaio_process_job(so, which, job);
719 }
720
721 /*
722 * If there are still pending requests, the socket must not be
723 * ready so set SB_AIO to request a wakeup when the socket
724 * becomes ready.
725 */
726 if (!TAILQ_EMPTY(&sb->sb_aiojobq))
727 sb->sb_flags |= SB_AIO;
728 sb->sb_flags &= ~SB_AIO_RUNNING;
729 SOCK_BUF_UNLOCK(so, which);
730
731 sorele(so);
732 CURVNET_RESTORE();
733 }
734
735 void
soaio_rcv(void * context,int pending)736 soaio_rcv(void *context, int pending)
737 {
738 struct socket *so;
739
740 so = context;
741 soaio_process_sb(so, SO_RCV);
742 }
743
744 void
soaio_snd(void * context,int pending)745 soaio_snd(void *context, int pending)
746 {
747 struct socket *so;
748
749 so = context;
750 soaio_process_sb(so, SO_SND);
751 }
752
753 void
sowakeup_aio(struct socket * so,sb_which which)754 sowakeup_aio(struct socket *so, sb_which which)
755 {
756 struct sockbuf *sb = sobuf(so, which);
757
758 SOCK_BUF_LOCK_ASSERT(so, which);
759
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
soo_aio_cancel(struct kaiocb * job)769 soo_aio_cancel(struct kaiocb *job)
770 {
771 struct socket *so;
772 struct sockbuf *sb;
773 long done;
774 int opcode;
775 sb_which which;
776
777 so = job->fd_file->f_data;
778 opcode = job->uaiocb.aio_lio_opcode;
779 if (opcode & LIO_READ) {
780 sb = &so->so_rcv;
781 which = SO_RCV;
782 } else {
783 MPASS(opcode & LIO_WRITE);
784 sb = &so->so_snd;
785 which = SO_SND;
786 }
787
788 SOCK_BUF_LOCK(so, which);
789 if (!aio_cancel_cleared(job))
790 TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
791 if (TAILQ_EMPTY(&sb->sb_aiojobq))
792 sb->sb_flags &= ~SB_AIO;
793 SOCK_BUF_UNLOCK(so, which);
794
795 done = job->aio_done;
796 if (done != 0)
797 aio_complete(job, done, 0);
798 else
799 aio_cancel(job);
800 }
801
802 static int
soo_aio_queue(struct file * fp,struct kaiocb * job)803 soo_aio_queue(struct file *fp, struct kaiocb *job)
804 {
805 struct socket *so;
806 struct sockbuf *sb;
807 sb_which which;
808 int error;
809
810 so = fp->f_data;
811 error = so->so_proto->pr_aio_queue(so, job);
812 if (error == 0)
813 return (0);
814
815 /* Lock through the socket, since this may be a listening socket. */
816 switch (job->uaiocb.aio_lio_opcode & (LIO_WRITE | LIO_READ)) {
817 case LIO_READ:
818 SOCK_RECVBUF_LOCK(so);
819 sb = &so->so_rcv;
820 which = SO_RCV;
821 break;
822 case LIO_WRITE:
823 SOCK_SENDBUF_LOCK(so);
824 sb = &so->so_snd;
825 which = SO_SND;
826 break;
827 default:
828 return (EINVAL);
829 }
830
831 if (SOLISTENING(so)) {
832 SOCK_BUF_UNLOCK(so, which);
833 return (EINVAL);
834 }
835
836 if (!aio_set_cancel_function(job, soo_aio_cancel))
837 panic("new job was cancelled");
838 TAILQ_INSERT_TAIL(&sb->sb_aiojobq, job, list);
839 if (!(sb->sb_flags & SB_AIO_RUNNING)) {
840 if (soaio_ready(so, sb))
841 sowakeup_aio(so, which);
842 else
843 sb->sb_flags |= SB_AIO;
844 }
845 SOCK_BUF_UNLOCK(so, which);
846 return (0);
847 }
848