xref: /freebsd/sys/kern/sys_socket.c (revision 2f78413036dbd2525302e09a9457b32353a8e1ae)
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  * 3. 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("$FreeBSD$");
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 		if (*(int *)data) {
174 			SOCK_LOCK(so);
175 			so->so_state |= SS_ASYNC;
176 			if (SOLISTENING(so)) {
177 				so->sol_sbrcv_flags |= SB_ASYNC;
178 				so->sol_sbsnd_flags |= SB_ASYNC;
179 			} else {
180 				SOCKBUF_LOCK(&so->so_rcv);
181 				so->so_rcv.sb_flags |= SB_ASYNC;
182 				SOCKBUF_UNLOCK(&so->so_rcv);
183 				SOCKBUF_LOCK(&so->so_snd);
184 				so->so_snd.sb_flags |= SB_ASYNC;
185 				SOCKBUF_UNLOCK(&so->so_snd);
186 			}
187 			SOCK_UNLOCK(so);
188 		} else {
189 			SOCK_LOCK(so);
190 			so->so_state &= ~SS_ASYNC;
191 			if (SOLISTENING(so)) {
192 				so->sol_sbrcv_flags &= ~SB_ASYNC;
193 				so->sol_sbsnd_flags &= ~SB_ASYNC;
194 			} else {
195 				SOCKBUF_LOCK(&so->so_rcv);
196 				so->so_rcv.sb_flags &= ~SB_ASYNC;
197 				SOCKBUF_UNLOCK(&so->so_rcv);
198 				SOCKBUF_LOCK(&so->so_snd);
199 				so->so_snd.sb_flags &= ~SB_ASYNC;
200 				SOCKBUF_UNLOCK(&so->so_snd);
201 			}
202 			SOCK_UNLOCK(so);
203 		}
204 		break;
205 
206 	case FIONREAD:
207 		/* Unlocked read. */
208 		*(int *)data = sbavail(&so->so_rcv);
209 		break;
210 
211 	case FIONWRITE:
212 		/* Unlocked read. */
213 		*(int *)data = sbavail(&so->so_snd);
214 		break;
215 
216 	case FIONSPACE:
217 		/* Unlocked read. */
218 		if ((so->so_snd.sb_hiwat < sbused(&so->so_snd)) ||
219 		    (so->so_snd.sb_mbmax < so->so_snd.sb_mbcnt))
220 			*(int *)data = 0;
221 		else
222 			*(int *)data = sbspace(&so->so_snd);
223 		break;
224 
225 	case FIOSETOWN:
226 		error = fsetown(*(int *)data, &so->so_sigio);
227 		break;
228 
229 	case FIOGETOWN:
230 		*(int *)data = fgetown(&so->so_sigio);
231 		break;
232 
233 	case SIOCSPGRP:
234 		error = fsetown(-(*(int *)data), &so->so_sigio);
235 		break;
236 
237 	case SIOCGPGRP:
238 		*(int *)data = -fgetown(&so->so_sigio);
239 		break;
240 
241 	case SIOCATMARK:
242 		/* Unlocked read. */
243 		*(int *)data = (so->so_rcv.sb_state & SBS_RCVATMARK) != 0;
244 		break;
245 	default:
246 		/*
247 		 * Interface/routing/protocol specific ioctls: interface and
248 		 * routing ioctls should have a different entry since a
249 		 * socket is unnecessary.
250 		 */
251 		if (IOCGROUP(cmd) == 'i')
252 			error = ifioctl(so, cmd, data, td);
253 		else if (IOCGROUP(cmd) == 'r') {
254 			CURVNET_SET(so->so_vnet);
255 			error = rtioctl_fib(cmd, data, so->so_fibnum);
256 			CURVNET_RESTORE();
257 		} else {
258 			CURVNET_SET(so->so_vnet);
259 			error = ((*so->so_proto->pr_usrreqs->pru_control)
260 			    (so, cmd, data, 0, td));
261 			CURVNET_RESTORE();
262 		}
263 		break;
264 	}
265 	return (error);
266 }
267 
268 static int
269 soo_poll(struct file *fp, int events, struct ucred *active_cred,
270     struct thread *td)
271 {
272 	struct socket *so = fp->f_data;
273 #ifdef MAC
274 	int error;
275 
276 	error = mac_socket_check_poll(active_cred, so);
277 	if (error)
278 		return (error);
279 #endif
280 	return (sopoll(so, events, fp->f_cred, td));
281 }
282 
283 static int
284 soo_stat(struct file *fp, struct stat *ub, struct ucred *active_cred,
285     struct thread *td)
286 {
287 	struct socket *so = fp->f_data;
288 	struct sockbuf *sb;
289 #ifdef MAC
290 	int error;
291 #endif
292 
293 	bzero((caddr_t)ub, sizeof (*ub));
294 	ub->st_mode = S_IFSOCK;
295 #ifdef MAC
296 	error = mac_socket_check_stat(active_cred, so);
297 	if (error)
298 		return (error);
299 #endif
300 	/*
301 	 * If SBS_CANTRCVMORE is set, but there's still data left in the
302 	 * receive buffer, the socket is still readable.
303 	 */
304 	sb = &so->so_rcv;
305 	SOCKBUF_LOCK(sb);
306 	if ((sb->sb_state & SBS_CANTRCVMORE) == 0 || sbavail(sb))
307 		ub->st_mode |= S_IRUSR | S_IRGRP | S_IROTH;
308 	ub->st_size = sbavail(sb) - sb->sb_ctl;
309 	SOCKBUF_UNLOCK(sb);
310 
311 	sb = &so->so_snd;
312 	SOCKBUF_LOCK(sb);
313 	if ((sb->sb_state & SBS_CANTSENDMORE) == 0)
314 		ub->st_mode |= S_IWUSR | S_IWGRP | S_IWOTH;
315 	SOCKBUF_UNLOCK(sb);
316 	ub->st_uid = so->so_cred->cr_uid;
317 	ub->st_gid = so->so_cred->cr_gid;
318 	return (*so->so_proto->pr_usrreqs->pru_sense)(so, ub);
319 }
320 
321 /*
322  * API socket close on file pointer.  We call soclose() to close the socket
323  * (including initiating closing protocols).  soclose() will sorele() the
324  * file reference but the actual socket will not go away until the socket's
325  * ref count hits 0.
326  */
327 static int
328 soo_close(struct file *fp, struct thread *td)
329 {
330 	int error = 0;
331 	struct socket *so;
332 
333 	so = fp->f_data;
334 	fp->f_ops = &badfileops;
335 	fp->f_data = NULL;
336 
337 	if (so)
338 		error = soclose(so);
339 	return (error);
340 }
341 
342 static int
343 soo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
344 {
345 	struct sockaddr *sa;
346 	struct inpcb *inpcb;
347 	struct unpcb *unpcb;
348 	struct socket *so;
349 	int error;
350 
351 	kif->kf_type = KF_TYPE_SOCKET;
352 	so = fp->f_data;
353 	kif->kf_un.kf_sock.kf_sock_domain0 =
354 	    so->so_proto->pr_domain->dom_family;
355 	kif->kf_un.kf_sock.kf_sock_type0 = so->so_type;
356 	kif->kf_un.kf_sock.kf_sock_protocol0 = so->so_proto->pr_protocol;
357 	kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb;
358 	switch (kif->kf_un.kf_sock.kf_sock_domain0) {
359 	case AF_INET:
360 	case AF_INET6:
361 		if (kif->kf_un.kf_sock.kf_sock_protocol0 == IPPROTO_TCP) {
362 			if (so->so_pcb != NULL) {
363 				inpcb = (struct inpcb *)(so->so_pcb);
364 				kif->kf_un.kf_sock.kf_sock_inpcb =
365 				    (uintptr_t)inpcb->inp_ppcb;
366 				kif->kf_un.kf_sock.kf_sock_sendq =
367 				    sbused(&so->so_snd);
368 				kif->kf_un.kf_sock.kf_sock_recvq =
369 				    sbused(&so->so_rcv);
370 			}
371 		}
372 		break;
373 	case AF_UNIX:
374 		if (so->so_pcb != NULL) {
375 			unpcb = (struct unpcb *)(so->so_pcb);
376 			if (unpcb->unp_conn) {
377 				kif->kf_un.kf_sock.kf_sock_unpconn =
378 				    (uintptr_t)unpcb->unp_conn;
379 				kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
380 				    so->so_rcv.sb_state;
381 				kif->kf_un.kf_sock.kf_sock_snd_sb_state =
382 				    so->so_snd.sb_state;
383 				kif->kf_un.kf_sock.kf_sock_sendq =
384 				    sbused(&so->so_snd);
385 				kif->kf_un.kf_sock.kf_sock_recvq =
386 				    sbused(&so->so_rcv);
387 			}
388 		}
389 		break;
390 	}
391 	error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa);
392 	if (error == 0 &&
393 	    sa->sa_len <= sizeof(kif->kf_un.kf_sock.kf_sa_local)) {
394 		bcopy(sa, &kif->kf_un.kf_sock.kf_sa_local, sa->sa_len);
395 		free(sa, M_SONAME);
396 	}
397 	error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa);
398 	if (error == 0 &&
399 	    sa->sa_len <= sizeof(kif->kf_un.kf_sock.kf_sa_peer)) {
400 		bcopy(sa, &kif->kf_un.kf_sock.kf_sa_peer, sa->sa_len);
401 		free(sa, M_SONAME);
402 	}
403 	strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name,
404 	    sizeof(kif->kf_path));
405 	return (0);
406 }
407 
408 /*
409  * Use the 'backend3' field in AIO jobs to store the amount of data
410  * completed by the AIO job so far.
411  */
412 #define	aio_done	backend3
413 
414 static STAILQ_HEAD(, task) soaio_jobs;
415 static struct mtx soaio_jobs_lock;
416 static struct task soaio_kproc_task;
417 static int soaio_starting, soaio_idle, soaio_queued;
418 static struct unrhdr *soaio_kproc_unr;
419 
420 static int soaio_max_procs = MAX_AIO_PROCS;
421 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, max_procs, CTLFLAG_RW, &soaio_max_procs, 0,
422     "Maximum number of kernel processes to use for async socket IO");
423 
424 static int soaio_num_procs;
425 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, num_procs, CTLFLAG_RD, &soaio_num_procs, 0,
426     "Number of active kernel processes for async socket IO");
427 
428 static int soaio_target_procs = TARGET_AIO_PROCS;
429 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, target_procs, CTLFLAG_RD,
430     &soaio_target_procs, 0,
431     "Preferred number of ready kernel processes for async socket IO");
432 
433 static int soaio_lifetime;
434 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, lifetime, CTLFLAG_RW, &soaio_lifetime, 0,
435     "Maximum lifetime for idle aiod");
436 
437 static void
438 soaio_kproc_loop(void *arg)
439 {
440 	struct proc *p;
441 	struct vmspace *myvm;
442 	struct task *task;
443 	int error, id, pending;
444 
445 	id = (intptr_t)arg;
446 
447 	/*
448 	 * Grab an extra reference on the daemon's vmspace so that it
449 	 * doesn't get freed by jobs that switch to a different
450 	 * vmspace.
451 	 */
452 	p = curproc;
453 	myvm = vmspace_acquire_ref(p);
454 
455 	mtx_lock(&soaio_jobs_lock);
456 	MPASS(soaio_starting > 0);
457 	soaio_starting--;
458 	for (;;) {
459 		while (!STAILQ_EMPTY(&soaio_jobs)) {
460 			task = STAILQ_FIRST(&soaio_jobs);
461 			STAILQ_REMOVE_HEAD(&soaio_jobs, ta_link);
462 			soaio_queued--;
463 			pending = task->ta_pending;
464 			task->ta_pending = 0;
465 			mtx_unlock(&soaio_jobs_lock);
466 
467 			task->ta_func(task->ta_context, pending);
468 
469 			mtx_lock(&soaio_jobs_lock);
470 		}
471 		MPASS(soaio_queued == 0);
472 
473 		if (p->p_vmspace != myvm) {
474 			mtx_unlock(&soaio_jobs_lock);
475 			vmspace_switch_aio(myvm);
476 			mtx_lock(&soaio_jobs_lock);
477 			continue;
478 		}
479 
480 		soaio_idle++;
481 		error = mtx_sleep(&soaio_idle, &soaio_jobs_lock, 0, "-",
482 		    soaio_lifetime);
483 		soaio_idle--;
484 		if (error == EWOULDBLOCK && STAILQ_EMPTY(&soaio_jobs) &&
485 		    soaio_num_procs > soaio_target_procs)
486 			break;
487 	}
488 	soaio_num_procs--;
489 	mtx_unlock(&soaio_jobs_lock);
490 	free_unr(soaio_kproc_unr, id);
491 	kproc_exit(0);
492 }
493 
494 static void
495 soaio_kproc_create(void *context, int pending)
496 {
497 	struct proc *p;
498 	int error, id;
499 
500 	mtx_lock(&soaio_jobs_lock);
501 	for (;;) {
502 		if (soaio_num_procs < soaio_target_procs) {
503 			/* Must create */
504 		} else if (soaio_num_procs >= soaio_max_procs) {
505 			/*
506 			 * Hit the limit on kernel processes, don't
507 			 * create another one.
508 			 */
509 			break;
510 		} else if (soaio_queued <= soaio_idle + soaio_starting) {
511 			/*
512 			 * No more AIO jobs waiting for a process to be
513 			 * created, so stop.
514 			 */
515 			break;
516 		}
517 		soaio_starting++;
518 		mtx_unlock(&soaio_jobs_lock);
519 
520 		id = alloc_unr(soaio_kproc_unr);
521 		error = kproc_create(soaio_kproc_loop, (void *)(intptr_t)id,
522 		    &p, 0, 0, "soaiod%d", id);
523 		if (error != 0) {
524 			free_unr(soaio_kproc_unr, id);
525 			mtx_lock(&soaio_jobs_lock);
526 			soaio_starting--;
527 			break;
528 		}
529 
530 		mtx_lock(&soaio_jobs_lock);
531 		soaio_num_procs++;
532 	}
533 	mtx_unlock(&soaio_jobs_lock);
534 }
535 
536 void
537 soaio_enqueue(struct task *task)
538 {
539 
540 	mtx_lock(&soaio_jobs_lock);
541 	MPASS(task->ta_pending == 0);
542 	task->ta_pending++;
543 	STAILQ_INSERT_TAIL(&soaio_jobs, task, ta_link);
544 	soaio_queued++;
545 	if (soaio_queued <= soaio_idle)
546 		wakeup_one(&soaio_idle);
547 	else if (soaio_num_procs < soaio_max_procs)
548 		taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task);
549 	mtx_unlock(&soaio_jobs_lock);
550 }
551 
552 static void
553 soaio_init(void)
554 {
555 
556 	soaio_lifetime = AIOD_LIFETIME_DEFAULT;
557 	STAILQ_INIT(&soaio_jobs);
558 	mtx_init(&soaio_jobs_lock, "soaio jobs", NULL, MTX_DEF);
559 	soaio_kproc_unr = new_unrhdr(1, INT_MAX, NULL);
560 	TASK_INIT(&soaio_kproc_task, 0, soaio_kproc_create, NULL);
561 	if (soaio_target_procs > 0)
562 		taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task);
563 }
564 SYSINIT(soaio, SI_SUB_VFS, SI_ORDER_ANY, soaio_init, NULL);
565 
566 static __inline int
567 soaio_ready(struct socket *so, struct sockbuf *sb)
568 {
569 	return (sb == &so->so_rcv ? soreadable(so) : sowriteable(so));
570 }
571 
572 static void
573 soaio_process_job(struct socket *so, struct sockbuf *sb, struct kaiocb *job)
574 {
575 	struct ucred *td_savedcred;
576 	struct thread *td;
577 	struct file *fp;
578 	struct uio uio;
579 	struct iovec iov;
580 	size_t cnt, done;
581 	long ru_before;
582 	int error, flags;
583 
584 	SOCKBUF_UNLOCK(sb);
585 	aio_switch_vmspace(job);
586 	td = curthread;
587 	fp = job->fd_file;
588 retry:
589 	td_savedcred = td->td_ucred;
590 	td->td_ucred = job->cred;
591 
592 	done = job->aio_done;
593 	cnt = job->uaiocb.aio_nbytes - done;
594 	iov.iov_base = (void *)((uintptr_t)job->uaiocb.aio_buf + done);
595 	iov.iov_len = cnt;
596 	uio.uio_iov = &iov;
597 	uio.uio_iovcnt = 1;
598 	uio.uio_offset = 0;
599 	uio.uio_resid = cnt;
600 	uio.uio_segflg = UIO_USERSPACE;
601 	uio.uio_td = td;
602 	flags = MSG_NBIO;
603 
604 	/*
605 	 * For resource usage accounting, only count a completed request
606 	 * as a single message to avoid counting multiple calls to
607 	 * sosend/soreceive on a blocking socket.
608 	 */
609 
610 	if (sb == &so->so_rcv) {
611 		uio.uio_rw = UIO_READ;
612 		ru_before = td->td_ru.ru_msgrcv;
613 #ifdef MAC
614 		error = mac_socket_check_receive(fp->f_cred, so);
615 		if (error == 0)
616 
617 #endif
618 			error = soreceive(so, NULL, &uio, NULL, NULL, &flags);
619 		if (td->td_ru.ru_msgrcv != ru_before)
620 			job->msgrcv = 1;
621 	} else {
622 		if (!TAILQ_EMPTY(&sb->sb_aiojobq))
623 			flags |= MSG_MORETOCOME;
624 		uio.uio_rw = UIO_WRITE;
625 		ru_before = td->td_ru.ru_msgsnd;
626 #ifdef MAC
627 		error = mac_socket_check_send(fp->f_cred, so);
628 		if (error == 0)
629 #endif
630 			error = sosend(so, NULL, &uio, NULL, NULL, flags, td);
631 		if (td->td_ru.ru_msgsnd != ru_before)
632 			job->msgsnd = 1;
633 		if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0) {
634 			PROC_LOCK(job->userproc);
635 			kern_psignal(job->userproc, SIGPIPE);
636 			PROC_UNLOCK(job->userproc);
637 		}
638 	}
639 
640 	done += cnt - uio.uio_resid;
641 	job->aio_done = done;
642 	td->td_ucred = td_savedcred;
643 
644 	if (error == EWOULDBLOCK) {
645 		/*
646 		 * The request was either partially completed or not
647 		 * completed at all due to racing with a read() or
648 		 * write() on the socket.  If the socket is
649 		 * non-blocking, return with any partial completion.
650 		 * If the socket is blocking or if no progress has
651 		 * been made, requeue this request at the head of the
652 		 * queue to try again when the socket is ready.
653 		 */
654 		MPASS(done != job->uaiocb.aio_nbytes);
655 		SOCKBUF_LOCK(sb);
656 		if (done == 0 || !(so->so_state & SS_NBIO)) {
657 			empty_results++;
658 			if (soaio_ready(so, sb)) {
659 				empty_retries++;
660 				SOCKBUF_UNLOCK(sb);
661 				goto retry;
662 			}
663 
664 			if (!aio_set_cancel_function(job, soo_aio_cancel)) {
665 				SOCKBUF_UNLOCK(sb);
666 				if (done != 0)
667 					aio_complete(job, done, 0);
668 				else
669 					aio_cancel(job);
670 				SOCKBUF_LOCK(sb);
671 			} else {
672 				TAILQ_INSERT_HEAD(&sb->sb_aiojobq, job, list);
673 			}
674 			return;
675 		}
676 		SOCKBUF_UNLOCK(sb);
677 	}
678 	if (done != 0 && (error == ERESTART || error == EINTR ||
679 	    error == EWOULDBLOCK))
680 		error = 0;
681 	if (error)
682 		aio_complete(job, -1, error);
683 	else
684 		aio_complete(job, done, 0);
685 	SOCKBUF_LOCK(sb);
686 }
687 
688 static void
689 soaio_process_sb(struct socket *so, struct sockbuf *sb)
690 {
691 	struct kaiocb *job;
692 
693 	SOCKBUF_LOCK(sb);
694 	while (!TAILQ_EMPTY(&sb->sb_aiojobq) && soaio_ready(so, sb)) {
695 		job = TAILQ_FIRST(&sb->sb_aiojobq);
696 		TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
697 		if (!aio_clear_cancel_function(job))
698 			continue;
699 
700 		soaio_process_job(so, sb, job);
701 	}
702 
703 	/*
704 	 * If there are still pending requests, the socket must not be
705 	 * ready so set SB_AIO to request a wakeup when the socket
706 	 * becomes ready.
707 	 */
708 	if (!TAILQ_EMPTY(&sb->sb_aiojobq))
709 		sb->sb_flags |= SB_AIO;
710 	sb->sb_flags &= ~SB_AIO_RUNNING;
711 	SOCKBUF_UNLOCK(sb);
712 
713 	SOCK_LOCK(so);
714 	sorele(so);
715 }
716 
717 void
718 soaio_rcv(void *context, int pending)
719 {
720 	struct socket *so;
721 
722 	so = context;
723 	soaio_process_sb(so, &so->so_rcv);
724 }
725 
726 void
727 soaio_snd(void *context, int pending)
728 {
729 	struct socket *so;
730 
731 	so = context;
732 	soaio_process_sb(so, &so->so_snd);
733 }
734 
735 void
736 sowakeup_aio(struct socket *so, struct sockbuf *sb)
737 {
738 
739 	SOCKBUF_LOCK_ASSERT(sb);
740 	sb->sb_flags &= ~SB_AIO;
741 	if (sb->sb_flags & SB_AIO_RUNNING)
742 		return;
743 	sb->sb_flags |= SB_AIO_RUNNING;
744 	if (sb == &so->so_snd)
745 		SOCK_LOCK(so);
746 	soref(so);
747 	if (sb == &so->so_snd)
748 		SOCK_UNLOCK(so);
749 	soaio_enqueue(&sb->sb_aiotask);
750 }
751 
752 static void
753 soo_aio_cancel(struct kaiocb *job)
754 {
755 	struct socket *so;
756 	struct sockbuf *sb;
757 	long done;
758 	int opcode;
759 
760 	so = job->fd_file->f_data;
761 	opcode = job->uaiocb.aio_lio_opcode;
762 	if (opcode == LIO_READ)
763 		sb = &so->so_rcv;
764 	else {
765 		MPASS(opcode == LIO_WRITE);
766 		sb = &so->so_snd;
767 	}
768 
769 	SOCKBUF_LOCK(sb);
770 	if (!aio_cancel_cleared(job))
771 		TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
772 	if (TAILQ_EMPTY(&sb->sb_aiojobq))
773 		sb->sb_flags &= ~SB_AIO;
774 	SOCKBUF_UNLOCK(sb);
775 
776 	done = job->aio_done;
777 	if (done != 0)
778 		aio_complete(job, done, 0);
779 	else
780 		aio_cancel(job);
781 }
782 
783 static int
784 soo_aio_queue(struct file *fp, struct kaiocb *job)
785 {
786 	struct socket *so;
787 	struct sockbuf *sb;
788 	int error;
789 
790 	so = fp->f_data;
791 	error = (*so->so_proto->pr_usrreqs->pru_aio_queue)(so, job);
792 	if (error == 0)
793 		return (0);
794 
795 	switch (job->uaiocb.aio_lio_opcode) {
796 	case LIO_READ:
797 		sb = &so->so_rcv;
798 		break;
799 	case LIO_WRITE:
800 		sb = &so->so_snd;
801 		break;
802 	default:
803 		return (EINVAL);
804 	}
805 
806 	SOCKBUF_LOCK(sb);
807 	if (!aio_set_cancel_function(job, soo_aio_cancel))
808 		panic("new job was cancelled");
809 	TAILQ_INSERT_TAIL(&sb->sb_aiojobq, job, list);
810 	if (!(sb->sb_flags & SB_AIO_RUNNING)) {
811 		if (soaio_ready(so, sb))
812 			sowakeup_aio(so, sb);
813 		else
814 			sb->sb_flags |= SB_AIO;
815 	}
816 	SOCKBUF_UNLOCK(sb);
817 	return (0);
818 }
819