xref: /freebsd/sys/kern/sys_socket.c (revision 953111c9c536daf5304062914c0083cea5efb46b)
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 		/*
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_un.kf_sock.kf_sock_domain0 =
350 	    so->so_proto->pr_domain->dom_family;
351 	kif->kf_un.kf_sock.kf_sock_type0 = so->so_type;
352 	kif->kf_un.kf_sock.kf_sock_protocol0 = so->so_proto->pr_protocol;
353 	kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb;
354 	switch (kif->kf_un.kf_sock.kf_sock_domain0) {
355 	case AF_INET:
356 	case AF_INET6:
357 		if (kif->kf_un.kf_sock.kf_sock_protocol0 == IPPROTO_TCP) {
358 			if (so->so_pcb != NULL) {
359 				inpcb = (struct inpcb *)(so->so_pcb);
360 				kif->kf_un.kf_sock.kf_sock_inpcb =
361 				    (uintptr_t)inpcb->inp_ppcb;
362 				kif->kf_un.kf_sock.kf_sock_sendq =
363 				    sbused(&so->so_snd);
364 				kif->kf_un.kf_sock.kf_sock_recvq =
365 				    sbused(&so->so_rcv);
366 			}
367 		}
368 		break;
369 	case AF_UNIX:
370 		if (so->so_pcb != NULL) {
371 			unpcb = (struct unpcb *)(so->so_pcb);
372 			if (unpcb->unp_conn) {
373 				kif->kf_un.kf_sock.kf_sock_unpconn =
374 				    (uintptr_t)unpcb->unp_conn;
375 				kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
376 				    so->so_rcv.sb_state;
377 				kif->kf_un.kf_sock.kf_sock_snd_sb_state =
378 				    so->so_snd.sb_state;
379 				kif->kf_un.kf_sock.kf_sock_sendq =
380 				    sbused(&so->so_snd);
381 				kif->kf_un.kf_sock.kf_sock_recvq =
382 				    sbused(&so->so_rcv);
383 			}
384 		}
385 		break;
386 	}
387 	error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa);
388 	if (error == 0 &&
389 	    sa->sa_len <= sizeof(kif->kf_un.kf_sock.kf_sa_local)) {
390 		bcopy(sa, &kif->kf_un.kf_sock.kf_sa_local, sa->sa_len);
391 		free(sa, M_SONAME);
392 	}
393 	error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa);
394 	if (error == 0 &&
395 	    sa->sa_len <= sizeof(kif->kf_un.kf_sock.kf_sa_peer)) {
396 		bcopy(sa, &kif->kf_un.kf_sock.kf_sa_peer, sa->sa_len);
397 		free(sa, M_SONAME);
398 	}
399 	strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name,
400 	    sizeof(kif->kf_path));
401 	return (0);
402 }
403 
404 /*
405  * Use the 'backend3' field in AIO jobs to store the amount of data
406  * completed by the AIO job so far.
407  */
408 #define	aio_done	backend3
409 
410 static STAILQ_HEAD(, task) soaio_jobs;
411 static struct mtx soaio_jobs_lock;
412 static struct task soaio_kproc_task;
413 static int soaio_starting, soaio_idle, soaio_queued;
414 static struct unrhdr *soaio_kproc_unr;
415 
416 static int soaio_max_procs = MAX_AIO_PROCS;
417 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, max_procs, CTLFLAG_RW, &soaio_max_procs, 0,
418     "Maximum number of kernel processes to use for async socket IO");
419 
420 static int soaio_num_procs;
421 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, num_procs, CTLFLAG_RD, &soaio_num_procs, 0,
422     "Number of active kernel processes for async socket IO");
423 
424 static int soaio_target_procs = TARGET_AIO_PROCS;
425 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, target_procs, CTLFLAG_RD,
426     &soaio_target_procs, 0,
427     "Preferred number of ready kernel processes for async socket IO");
428 
429 static int soaio_lifetime;
430 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, lifetime, CTLFLAG_RW, &soaio_lifetime, 0,
431     "Maximum lifetime for idle aiod");
432 
433 static void
434 soaio_kproc_loop(void *arg)
435 {
436 	struct proc *p;
437 	struct vmspace *myvm;
438 	struct task *task;
439 	int error, id, pending;
440 
441 	id = (intptr_t)arg;
442 
443 	/*
444 	 * Grab an extra reference on the daemon's vmspace so that it
445 	 * doesn't get freed by jobs that switch to a different
446 	 * vmspace.
447 	 */
448 	p = curproc;
449 	myvm = vmspace_acquire_ref(p);
450 
451 	mtx_lock(&soaio_jobs_lock);
452 	MPASS(soaio_starting > 0);
453 	soaio_starting--;
454 	for (;;) {
455 		while (!STAILQ_EMPTY(&soaio_jobs)) {
456 			task = STAILQ_FIRST(&soaio_jobs);
457 			STAILQ_REMOVE_HEAD(&soaio_jobs, ta_link);
458 			soaio_queued--;
459 			pending = task->ta_pending;
460 			task->ta_pending = 0;
461 			mtx_unlock(&soaio_jobs_lock);
462 
463 			task->ta_func(task->ta_context, pending);
464 
465 			mtx_lock(&soaio_jobs_lock);
466 		}
467 		MPASS(soaio_queued == 0);
468 
469 		if (p->p_vmspace != myvm) {
470 			mtx_unlock(&soaio_jobs_lock);
471 			vmspace_switch_aio(myvm);
472 			mtx_lock(&soaio_jobs_lock);
473 			continue;
474 		}
475 
476 		soaio_idle++;
477 		error = mtx_sleep(&soaio_idle, &soaio_jobs_lock, 0, "-",
478 		    soaio_lifetime);
479 		soaio_idle--;
480 		if (error == EWOULDBLOCK && STAILQ_EMPTY(&soaio_jobs) &&
481 		    soaio_num_procs > soaio_target_procs)
482 			break;
483 	}
484 	soaio_num_procs--;
485 	mtx_unlock(&soaio_jobs_lock);
486 	free_unr(soaio_kproc_unr, id);
487 	kproc_exit(0);
488 }
489 
490 static void
491 soaio_kproc_create(void *context, int pending)
492 {
493 	struct proc *p;
494 	int error, id;
495 
496 	mtx_lock(&soaio_jobs_lock);
497 	for (;;) {
498 		if (soaio_num_procs < soaio_target_procs) {
499 			/* Must create */
500 		} else if (soaio_num_procs >= soaio_max_procs) {
501 			/*
502 			 * Hit the limit on kernel processes, don't
503 			 * create another one.
504 			 */
505 			break;
506 		} else if (soaio_queued <= soaio_idle + soaio_starting) {
507 			/*
508 			 * No more AIO jobs waiting for a process to be
509 			 * created, so stop.
510 			 */
511 			break;
512 		}
513 		soaio_starting++;
514 		mtx_unlock(&soaio_jobs_lock);
515 
516 		id = alloc_unr(soaio_kproc_unr);
517 		error = kproc_create(soaio_kproc_loop, (void *)(intptr_t)id,
518 		    &p, 0, 0, "soaiod%d", id);
519 		if (error != 0) {
520 			free_unr(soaio_kproc_unr, id);
521 			mtx_lock(&soaio_jobs_lock);
522 			soaio_starting--;
523 			break;
524 		}
525 
526 		mtx_lock(&soaio_jobs_lock);
527 		soaio_num_procs++;
528 	}
529 	mtx_unlock(&soaio_jobs_lock);
530 }
531 
532 void
533 soaio_enqueue(struct task *task)
534 {
535 
536 	mtx_lock(&soaio_jobs_lock);
537 	MPASS(task->ta_pending == 0);
538 	task->ta_pending++;
539 	STAILQ_INSERT_TAIL(&soaio_jobs, task, ta_link);
540 	soaio_queued++;
541 	if (soaio_queued <= soaio_idle)
542 		wakeup_one(&soaio_idle);
543 	else if (soaio_num_procs < soaio_max_procs)
544 		taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task);
545 	mtx_unlock(&soaio_jobs_lock);
546 }
547 
548 static void
549 soaio_init(void)
550 {
551 
552 	soaio_lifetime = AIOD_LIFETIME_DEFAULT;
553 	STAILQ_INIT(&soaio_jobs);
554 	mtx_init(&soaio_jobs_lock, "soaio jobs", NULL, MTX_DEF);
555 	soaio_kproc_unr = new_unrhdr(1, INT_MAX, NULL);
556 	TASK_INIT(&soaio_kproc_task, 0, soaio_kproc_create, NULL);
557 	if (soaio_target_procs > 0)
558 		taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task);
559 }
560 SYSINIT(soaio, SI_SUB_VFS, SI_ORDER_ANY, soaio_init, NULL);
561 
562 static __inline int
563 soaio_ready(struct socket *so, struct sockbuf *sb)
564 {
565 	return (sb == &so->so_rcv ? soreadable(so) : sowriteable(so));
566 }
567 
568 static void
569 soaio_process_job(struct socket *so, struct sockbuf *sb, struct kaiocb *job)
570 {
571 	struct ucred *td_savedcred;
572 	struct thread *td;
573 	struct file *fp;
574 	struct uio uio;
575 	struct iovec iov;
576 	size_t cnt, done;
577 	long ru_before;
578 	int error, flags;
579 
580 	SOCKBUF_UNLOCK(sb);
581 	aio_switch_vmspace(job);
582 	td = curthread;
583 	fp = job->fd_file;
584 retry:
585 	td_savedcred = td->td_ucred;
586 	td->td_ucred = job->cred;
587 
588 	done = job->aio_done;
589 	cnt = job->uaiocb.aio_nbytes - done;
590 	iov.iov_base = (void *)((uintptr_t)job->uaiocb.aio_buf + done);
591 	iov.iov_len = cnt;
592 	uio.uio_iov = &iov;
593 	uio.uio_iovcnt = 1;
594 	uio.uio_offset = 0;
595 	uio.uio_resid = cnt;
596 	uio.uio_segflg = UIO_USERSPACE;
597 	uio.uio_td = td;
598 	flags = MSG_NBIO;
599 
600 	/*
601 	 * For resource usage accounting, only count a completed request
602 	 * as a single message to avoid counting multiple calls to
603 	 * sosend/soreceive on a blocking socket.
604 	 */
605 
606 	if (sb == &so->so_rcv) {
607 		uio.uio_rw = UIO_READ;
608 		ru_before = td->td_ru.ru_msgrcv;
609 #ifdef MAC
610 		error = mac_socket_check_receive(fp->f_cred, so);
611 		if (error == 0)
612 
613 #endif
614 			error = soreceive(so, NULL, &uio, NULL, NULL, &flags);
615 		if (td->td_ru.ru_msgrcv != ru_before)
616 			job->msgrcv = 1;
617 	} else {
618 		if (!TAILQ_EMPTY(&sb->sb_aiojobq))
619 			flags |= MSG_MORETOCOME;
620 		uio.uio_rw = UIO_WRITE;
621 		ru_before = td->td_ru.ru_msgsnd;
622 #ifdef MAC
623 		error = mac_socket_check_send(fp->f_cred, so);
624 		if (error == 0)
625 #endif
626 			error = sosend(so, NULL, &uio, NULL, NULL, flags, td);
627 		if (td->td_ru.ru_msgsnd != ru_before)
628 			job->msgsnd = 1;
629 		if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0) {
630 			PROC_LOCK(job->userproc);
631 			kern_psignal(job->userproc, SIGPIPE);
632 			PROC_UNLOCK(job->userproc);
633 		}
634 	}
635 
636 	done += cnt - uio.uio_resid;
637 	job->aio_done = done;
638 	td->td_ucred = td_savedcred;
639 
640 	if (error == EWOULDBLOCK) {
641 		/*
642 		 * The request was either partially completed or not
643 		 * completed at all due to racing with a read() or
644 		 * write() on the socket.  If the socket is
645 		 * non-blocking, return with any partial completion.
646 		 * If the socket is blocking or if no progress has
647 		 * been made, requeue this request at the head of the
648 		 * queue to try again when the socket is ready.
649 		 */
650 		MPASS(done != job->uaiocb.aio_nbytes);
651 		SOCKBUF_LOCK(sb);
652 		if (done == 0 || !(so->so_state & SS_NBIO)) {
653 			empty_results++;
654 			if (soaio_ready(so, sb)) {
655 				empty_retries++;
656 				SOCKBUF_UNLOCK(sb);
657 				goto retry;
658 			}
659 
660 			if (!aio_set_cancel_function(job, soo_aio_cancel)) {
661 				SOCKBUF_UNLOCK(sb);
662 				if (done != 0)
663 					aio_complete(job, done, 0);
664 				else
665 					aio_cancel(job);
666 				SOCKBUF_LOCK(sb);
667 			} else {
668 				TAILQ_INSERT_HEAD(&sb->sb_aiojobq, job, list);
669 			}
670 			return;
671 		}
672 		SOCKBUF_UNLOCK(sb);
673 	}
674 	if (done != 0 && (error == ERESTART || error == EINTR ||
675 	    error == EWOULDBLOCK))
676 		error = 0;
677 	if (error)
678 		aio_complete(job, -1, error);
679 	else
680 		aio_complete(job, done, 0);
681 	SOCKBUF_LOCK(sb);
682 }
683 
684 static void
685 soaio_process_sb(struct socket *so, struct sockbuf *sb)
686 {
687 	struct kaiocb *job;
688 
689 	SOCKBUF_LOCK(sb);
690 	while (!TAILQ_EMPTY(&sb->sb_aiojobq) && soaio_ready(so, sb)) {
691 		job = TAILQ_FIRST(&sb->sb_aiojobq);
692 		TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
693 		if (!aio_clear_cancel_function(job))
694 			continue;
695 
696 		soaio_process_job(so, sb, job);
697 	}
698 
699 	/*
700 	 * If there are still pending requests, the socket must not be
701 	 * ready so set SB_AIO to request a wakeup when the socket
702 	 * becomes ready.
703 	 */
704 	if (!TAILQ_EMPTY(&sb->sb_aiojobq))
705 		sb->sb_flags |= SB_AIO;
706 	sb->sb_flags &= ~SB_AIO_RUNNING;
707 	SOCKBUF_UNLOCK(sb);
708 
709 	ACCEPT_LOCK();
710 	SOCK_LOCK(so);
711 	sorele(so);
712 }
713 
714 void
715 soaio_rcv(void *context, int pending)
716 {
717 	struct socket *so;
718 
719 	so = context;
720 	soaio_process_sb(so, &so->so_rcv);
721 }
722 
723 void
724 soaio_snd(void *context, int pending)
725 {
726 	struct socket *so;
727 
728 	so = context;
729 	soaio_process_sb(so, &so->so_snd);
730 }
731 
732 void
733 sowakeup_aio(struct socket *so, struct sockbuf *sb)
734 {
735 
736 	SOCKBUF_LOCK_ASSERT(sb);
737 	sb->sb_flags &= ~SB_AIO;
738 	if (sb->sb_flags & SB_AIO_RUNNING)
739 		return;
740 	sb->sb_flags |= SB_AIO_RUNNING;
741 	if (sb == &so->so_snd)
742 		SOCK_LOCK(so);
743 	soref(so);
744 	if (sb == &so->so_snd)
745 		SOCK_UNLOCK(so);
746 	soaio_enqueue(&sb->sb_aiotask);
747 }
748 
749 static void
750 soo_aio_cancel(struct kaiocb *job)
751 {
752 	struct socket *so;
753 	struct sockbuf *sb;
754 	long done;
755 	int opcode;
756 
757 	so = job->fd_file->f_data;
758 	opcode = job->uaiocb.aio_lio_opcode;
759 	if (opcode == LIO_READ)
760 		sb = &so->so_rcv;
761 	else {
762 		MPASS(opcode == LIO_WRITE);
763 		sb = &so->so_snd;
764 	}
765 
766 	SOCKBUF_LOCK(sb);
767 	if (!aio_cancel_cleared(job))
768 		TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
769 	if (TAILQ_EMPTY(&sb->sb_aiojobq))
770 		sb->sb_flags &= ~SB_AIO;
771 	SOCKBUF_UNLOCK(sb);
772 
773 	done = job->aio_done;
774 	if (done != 0)
775 		aio_complete(job, done, 0);
776 	else
777 		aio_cancel(job);
778 }
779 
780 static int
781 soo_aio_queue(struct file *fp, struct kaiocb *job)
782 {
783 	struct socket *so;
784 	struct sockbuf *sb;
785 	int error;
786 
787 	so = fp->f_data;
788 	error = (*so->so_proto->pr_usrreqs->pru_aio_queue)(so, job);
789 	if (error == 0)
790 		return (0);
791 
792 	switch (job->uaiocb.aio_lio_opcode) {
793 	case LIO_READ:
794 		sb = &so->so_rcv;
795 		break;
796 	case LIO_WRITE:
797 		sb = &so->so_snd;
798 		break;
799 	default:
800 		return (EINVAL);
801 	}
802 
803 	SOCKBUF_LOCK(sb);
804 	if (!aio_set_cancel_function(job, soo_aio_cancel))
805 		panic("new job was cancelled");
806 	TAILQ_INSERT_TAIL(&sb->sb_aiojobq, job, list);
807 	if (!(sb->sb_flags & SB_AIO_RUNNING)) {
808 		if (soaio_ready(so, sb))
809 			sowakeup_aio(so, sb);
810 		else
811 			sb->sb_flags |= SB_AIO;
812 	}
813 	SOCKBUF_UNLOCK(sb);
814 	return (0);
815 }
816