xref: /freebsd/sys/kern/sys_socket.c (revision a8e286bb5d076686c85f10b73ec33acc133ec1ab)
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 static fo_fill_kinfo_t soo_fill_kinfo;
99 static fo_aio_queue_t soo_aio_queue;
100 
101 static void	soo_aio_cancel(struct kaiocb *job);
102 
103 struct fileops	socketops = {
104 	.fo_read = soo_read,
105 	.fo_write = soo_write,
106 	.fo_truncate = invfo_truncate,
107 	.fo_ioctl = soo_ioctl,
108 	.fo_poll = soo_poll,
109 	.fo_kqfilter = soo_kqfilter,
110 	.fo_stat = soo_stat,
111 	.fo_close = soo_close,
112 	.fo_chmod = invfo_chmod,
113 	.fo_chown = invfo_chown,
114 	.fo_sendfile = invfo_sendfile,
115 	.fo_fill_kinfo = soo_fill_kinfo,
116 	.fo_aio_queue = soo_aio_queue,
117 	.fo_flags = DFLAG_PASSABLE
118 };
119 
120 static int
121 soo_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
122     int flags, struct thread *td)
123 {
124 	struct socket *so = fp->f_data;
125 	int error;
126 
127 #ifdef MAC
128 	error = mac_socket_check_receive(active_cred, so);
129 	if (error)
130 		return (error);
131 #endif
132 	error = soreceive(so, 0, uio, 0, 0, 0);
133 	return (error);
134 }
135 
136 static int
137 soo_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
138     int flags, struct thread *td)
139 {
140 	struct socket *so = fp->f_data;
141 	int error;
142 
143 #ifdef MAC
144 	error = mac_socket_check_send(active_cred, so);
145 	if (error)
146 		return (error);
147 #endif
148 	error = sosend(so, 0, uio, 0, 0, 0, uio->uio_td);
149 	if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0) {
150 		PROC_LOCK(uio->uio_td->td_proc);
151 		tdsignal(uio->uio_td, SIGPIPE);
152 		PROC_UNLOCK(uio->uio_td->td_proc);
153 	}
154 	return (error);
155 }
156 
157 static int
158 soo_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *active_cred,
159     struct thread *td)
160 {
161 	struct socket *so = fp->f_data;
162 	int error = 0;
163 
164 	switch (cmd) {
165 	case FIONBIO:
166 		SOCK_LOCK(so);
167 		if (*(int *)data)
168 			so->so_state |= SS_NBIO;
169 		else
170 			so->so_state &= ~SS_NBIO;
171 		SOCK_UNLOCK(so);
172 		break;
173 
174 	case FIOASYNC:
175 		if (*(int *)data) {
176 			SOCK_LOCK(so);
177 			so->so_state |= SS_ASYNC;
178 			if (SOLISTENING(so)) {
179 				so->sol_sbrcv_flags |= SB_ASYNC;
180 				so->sol_sbsnd_flags |= SB_ASYNC;
181 			} else {
182 				SOCK_RECVBUF_LOCK(so);
183 				so->so_rcv.sb_flags |= SB_ASYNC;
184 				SOCK_RECVBUF_UNLOCK(so);
185 				SOCK_SENDBUF_LOCK(so);
186 				so->so_snd.sb_flags |= SB_ASYNC;
187 				SOCK_SENDBUF_UNLOCK(so);
188 			}
189 			SOCK_UNLOCK(so);
190 		} else {
191 			SOCK_LOCK(so);
192 			so->so_state &= ~SS_ASYNC;
193 			if (SOLISTENING(so)) {
194 				so->sol_sbrcv_flags &= ~SB_ASYNC;
195 				so->sol_sbsnd_flags &= ~SB_ASYNC;
196 			} else {
197 				SOCK_RECVBUF_LOCK(so);
198 				so->so_rcv.sb_flags &= ~SB_ASYNC;
199 				SOCK_RECVBUF_UNLOCK(so);
200 				SOCK_SENDBUF_LOCK(so);
201 				so->so_snd.sb_flags &= ~SB_ASYNC;
202 				SOCK_SENDBUF_UNLOCK(so);
203 			}
204 			SOCK_UNLOCK(so);
205 		}
206 		break;
207 
208 	case FIONREAD:
209 		SOCK_RECVBUF_LOCK(so);
210 		if (SOLISTENING(so)) {
211 			error = EINVAL;
212 		} else {
213 			*(int *)data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
214 		}
215 		SOCK_RECVBUF_UNLOCK(so);
216 		break;
217 
218 	case FIONWRITE:
219 		/* Unlocked read. */
220 		if (SOLISTENING(so)) {
221 			error = EINVAL;
222 		} else {
223 			*(int *)data = sbavail(&so->so_snd);
224 		}
225 		break;
226 
227 	case FIONSPACE:
228 		/* Unlocked read. */
229 		if (SOLISTENING(so)) {
230 			error = EINVAL;
231 		} else {
232 			if ((so->so_snd.sb_hiwat < sbused(&so->so_snd)) ||
233 			    (so->so_snd.sb_mbmax < so->so_snd.sb_mbcnt)) {
234 				*(int *)data = 0;
235 			} else {
236 				*(int *)data = sbspace(&so->so_snd);
237 			}
238 		}
239 		break;
240 
241 	case FIOSETOWN:
242 		error = fsetown(*(int *)data, &so->so_sigio);
243 		break;
244 
245 	case FIOGETOWN:
246 		*(int *)data = fgetown(&so->so_sigio);
247 		break;
248 
249 	case SIOCSPGRP:
250 		error = fsetown(-(*(int *)data), &so->so_sigio);
251 		break;
252 
253 	case SIOCGPGRP:
254 		*(int *)data = -fgetown(&so->so_sigio);
255 		break;
256 
257 	case SIOCATMARK:
258 		/* Unlocked read. */
259 		if (SOLISTENING(so)) {
260 			error = EINVAL;
261 		} else {
262 			*(int *)data = (so->so_rcv.sb_state & SBS_RCVATMARK) != 0;
263 		}
264 		break;
265 	default:
266 		/*
267 		 * Interface/routing/protocol specific ioctls: interface and
268 		 * routing ioctls should have a different entry since a
269 		 * socket is unnecessary.
270 		 */
271 		if (IOCGROUP(cmd) == 'i')
272 			error = ifioctl(so, cmd, data, td);
273 		else if (IOCGROUP(cmd) == 'r') {
274 			CURVNET_SET(so->so_vnet);
275 			error = rtioctl_fib(cmd, data, so->so_fibnum);
276 			CURVNET_RESTORE();
277 		} else {
278 			CURVNET_SET(so->so_vnet);
279 			error = ((*so->so_proto->pr_usrreqs->pru_control)
280 			    (so, cmd, data, 0, td));
281 			CURVNET_RESTORE();
282 		}
283 		break;
284 	}
285 	return (error);
286 }
287 
288 static int
289 soo_poll(struct file *fp, int events, struct ucred *active_cred,
290     struct thread *td)
291 {
292 	struct socket *so = fp->f_data;
293 #ifdef MAC
294 	int error;
295 
296 	error = mac_socket_check_poll(active_cred, so);
297 	if (error)
298 		return (error);
299 #endif
300 	return (sopoll(so, events, fp->f_cred, td));
301 }
302 
303 static int
304 soo_stat(struct file *fp, struct stat *ub, struct ucred *active_cred)
305 {
306 	struct socket *so = fp->f_data;
307 	int error;
308 
309 	bzero((caddr_t)ub, sizeof (*ub));
310 	ub->st_mode = S_IFSOCK;
311 #ifdef MAC
312 	error = mac_socket_check_stat(active_cred, so);
313 	if (error)
314 		return (error);
315 #endif
316 	SOCK_LOCK(so);
317 	if (!SOLISTENING(so)) {
318 		struct sockbuf *sb;
319 
320 		/*
321 		 * If SBS_CANTRCVMORE is set, but there's still data left
322 		 * in the receive buffer, the socket is still readable.
323 		 */
324 		sb = &so->so_rcv;
325 		SOCK_RECVBUF_LOCK(so);
326 		if ((sb->sb_state & SBS_CANTRCVMORE) == 0 || sbavail(sb))
327 			ub->st_mode |= S_IRUSR | S_IRGRP | S_IROTH;
328 		ub->st_size = sbavail(sb) - sb->sb_ctl;
329 		SOCK_RECVBUF_UNLOCK(so);
330 
331 		sb = &so->so_snd;
332 		SOCK_SENDBUF_LOCK(so);
333 		if ((sb->sb_state & SBS_CANTSENDMORE) == 0)
334 			ub->st_mode |= S_IWUSR | S_IWGRP | S_IWOTH;
335 		SOCK_SENDBUF_UNLOCK(so);
336 	}
337 	ub->st_uid = so->so_cred->cr_uid;
338 	ub->st_gid = so->so_cred->cr_gid;
339 	error = so->so_proto->pr_usrreqs->pru_sense(so, ub);
340 	SOCK_UNLOCK(so);
341 	return (error);
342 }
343 
344 /*
345  * API socket close on file pointer.  We call soclose() to close the socket
346  * (including initiating closing protocols).  soclose() will sorele() the
347  * file reference but the actual socket will not go away until the socket's
348  * ref count hits 0.
349  */
350 static int
351 soo_close(struct file *fp, struct thread *td)
352 {
353 	int error = 0;
354 	struct socket *so;
355 
356 	so = fp->f_data;
357 	fp->f_ops = &badfileops;
358 	fp->f_data = NULL;
359 
360 	if (so)
361 		error = soclose(so);
362 	return (error);
363 }
364 
365 static int
366 soo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
367 {
368 	struct sockaddr *sa;
369 	struct inpcb *inpcb;
370 	struct unpcb *unpcb;
371 	struct socket *so;
372 	int error;
373 
374 	kif->kf_type = KF_TYPE_SOCKET;
375 	so = fp->f_data;
376 	CURVNET_SET(so->so_vnet);
377 	kif->kf_un.kf_sock.kf_sock_domain0 =
378 	    so->so_proto->pr_domain->dom_family;
379 	kif->kf_un.kf_sock.kf_sock_type0 = so->so_type;
380 	kif->kf_un.kf_sock.kf_sock_protocol0 = so->so_proto->pr_protocol;
381 	kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb;
382 	switch (kif->kf_un.kf_sock.kf_sock_domain0) {
383 	case AF_INET:
384 	case AF_INET6:
385 		if (kif->kf_un.kf_sock.kf_sock_protocol0 == IPPROTO_TCP) {
386 			if (so->so_pcb != NULL) {
387 				inpcb = (struct inpcb *)(so->so_pcb);
388 				kif->kf_un.kf_sock.kf_sock_inpcb =
389 				    (uintptr_t)inpcb->inp_ppcb;
390 				kif->kf_un.kf_sock.kf_sock_sendq =
391 				    sbused(&so->so_snd);
392 				kif->kf_un.kf_sock.kf_sock_recvq =
393 				    sbused(&so->so_rcv);
394 			}
395 		}
396 		break;
397 	case AF_UNIX:
398 		if (so->so_pcb != NULL) {
399 			unpcb = (struct unpcb *)(so->so_pcb);
400 			if (unpcb->unp_conn) {
401 				kif->kf_un.kf_sock.kf_sock_unpconn =
402 				    (uintptr_t)unpcb->unp_conn;
403 				kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
404 				    so->so_rcv.sb_state;
405 				kif->kf_un.kf_sock.kf_sock_snd_sb_state =
406 				    so->so_snd.sb_state;
407 				kif->kf_un.kf_sock.kf_sock_sendq =
408 				    sbused(&so->so_snd);
409 				kif->kf_un.kf_sock.kf_sock_recvq =
410 				    sbused(&so->so_rcv);
411 			}
412 		}
413 		break;
414 	}
415 	error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa);
416 	if (error == 0 &&
417 	    sa->sa_len <= sizeof(kif->kf_un.kf_sock.kf_sa_local)) {
418 		bcopy(sa, &kif->kf_un.kf_sock.kf_sa_local, sa->sa_len);
419 		free(sa, M_SONAME);
420 	}
421 	error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa);
422 	if (error == 0 &&
423 	    sa->sa_len <= sizeof(kif->kf_un.kf_sock.kf_sa_peer)) {
424 		bcopy(sa, &kif->kf_un.kf_sock.kf_sa_peer, sa->sa_len);
425 		free(sa, M_SONAME);
426 	}
427 	strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name,
428 	    sizeof(kif->kf_path));
429 	CURVNET_RESTORE();
430 	return (0);
431 }
432 
433 /*
434  * Use the 'backend3' field in AIO jobs to store the amount of data
435  * completed by the AIO job so far.
436  */
437 #define	aio_done	backend3
438 
439 static STAILQ_HEAD(, task) soaio_jobs;
440 static struct mtx soaio_jobs_lock;
441 static struct task soaio_kproc_task;
442 static int soaio_starting, soaio_idle, soaio_queued;
443 static struct unrhdr *soaio_kproc_unr;
444 
445 static int soaio_max_procs = MAX_AIO_PROCS;
446 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, max_procs, CTLFLAG_RW, &soaio_max_procs, 0,
447     "Maximum number of kernel processes to use for async socket IO");
448 
449 static int soaio_num_procs;
450 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, num_procs, CTLFLAG_RD, &soaio_num_procs, 0,
451     "Number of active kernel processes for async socket IO");
452 
453 static int soaio_target_procs = TARGET_AIO_PROCS;
454 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, target_procs, CTLFLAG_RD,
455     &soaio_target_procs, 0,
456     "Preferred number of ready kernel processes for async socket IO");
457 
458 static int soaio_lifetime;
459 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, lifetime, CTLFLAG_RW, &soaio_lifetime, 0,
460     "Maximum lifetime for idle aiod");
461 
462 static void
463 soaio_kproc_loop(void *arg)
464 {
465 	struct proc *p;
466 	struct vmspace *myvm;
467 	struct task *task;
468 	int error, id, pending;
469 
470 	id = (intptr_t)arg;
471 
472 	/*
473 	 * Grab an extra reference on the daemon's vmspace so that it
474 	 * doesn't get freed by jobs that switch to a different
475 	 * vmspace.
476 	 */
477 	p = curproc;
478 	myvm = vmspace_acquire_ref(p);
479 
480 	mtx_lock(&soaio_jobs_lock);
481 	MPASS(soaio_starting > 0);
482 	soaio_starting--;
483 	for (;;) {
484 		while (!STAILQ_EMPTY(&soaio_jobs)) {
485 			task = STAILQ_FIRST(&soaio_jobs);
486 			STAILQ_REMOVE_HEAD(&soaio_jobs, ta_link);
487 			soaio_queued--;
488 			pending = task->ta_pending;
489 			task->ta_pending = 0;
490 			mtx_unlock(&soaio_jobs_lock);
491 
492 			task->ta_func(task->ta_context, pending);
493 
494 			mtx_lock(&soaio_jobs_lock);
495 		}
496 		MPASS(soaio_queued == 0);
497 
498 		if (p->p_vmspace != myvm) {
499 			mtx_unlock(&soaio_jobs_lock);
500 			vmspace_switch_aio(myvm);
501 			mtx_lock(&soaio_jobs_lock);
502 			continue;
503 		}
504 
505 		soaio_idle++;
506 		error = mtx_sleep(&soaio_idle, &soaio_jobs_lock, 0, "-",
507 		    soaio_lifetime);
508 		soaio_idle--;
509 		if (error == EWOULDBLOCK && STAILQ_EMPTY(&soaio_jobs) &&
510 		    soaio_num_procs > soaio_target_procs)
511 			break;
512 	}
513 	soaio_num_procs--;
514 	mtx_unlock(&soaio_jobs_lock);
515 	free_unr(soaio_kproc_unr, id);
516 	kproc_exit(0);
517 }
518 
519 static void
520 soaio_kproc_create(void *context, int pending)
521 {
522 	struct proc *p;
523 	int error, id;
524 
525 	mtx_lock(&soaio_jobs_lock);
526 	for (;;) {
527 		if (soaio_num_procs < soaio_target_procs) {
528 			/* Must create */
529 		} else if (soaio_num_procs >= soaio_max_procs) {
530 			/*
531 			 * Hit the limit on kernel processes, don't
532 			 * create another one.
533 			 */
534 			break;
535 		} else if (soaio_queued <= soaio_idle + soaio_starting) {
536 			/*
537 			 * No more AIO jobs waiting for a process to be
538 			 * created, so stop.
539 			 */
540 			break;
541 		}
542 		soaio_starting++;
543 		mtx_unlock(&soaio_jobs_lock);
544 
545 		id = alloc_unr(soaio_kproc_unr);
546 		error = kproc_create(soaio_kproc_loop, (void *)(intptr_t)id,
547 		    &p, 0, 0, "soaiod%d", id);
548 		if (error != 0) {
549 			free_unr(soaio_kproc_unr, id);
550 			mtx_lock(&soaio_jobs_lock);
551 			soaio_starting--;
552 			break;
553 		}
554 
555 		mtx_lock(&soaio_jobs_lock);
556 		soaio_num_procs++;
557 	}
558 	mtx_unlock(&soaio_jobs_lock);
559 }
560 
561 void
562 soaio_enqueue(struct task *task)
563 {
564 
565 	mtx_lock(&soaio_jobs_lock);
566 	MPASS(task->ta_pending == 0);
567 	task->ta_pending++;
568 	STAILQ_INSERT_TAIL(&soaio_jobs, task, ta_link);
569 	soaio_queued++;
570 	if (soaio_queued <= soaio_idle)
571 		wakeup_one(&soaio_idle);
572 	else if (soaio_num_procs < soaio_max_procs)
573 		taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task);
574 	mtx_unlock(&soaio_jobs_lock);
575 }
576 
577 static void
578 soaio_init(void)
579 {
580 
581 	soaio_lifetime = AIOD_LIFETIME_DEFAULT;
582 	STAILQ_INIT(&soaio_jobs);
583 	mtx_init(&soaio_jobs_lock, "soaio jobs", NULL, MTX_DEF);
584 	soaio_kproc_unr = new_unrhdr(1, INT_MAX, NULL);
585 	TASK_INIT(&soaio_kproc_task, 0, soaio_kproc_create, NULL);
586 }
587 SYSINIT(soaio, SI_SUB_VFS, SI_ORDER_ANY, soaio_init, NULL);
588 
589 static __inline int
590 soaio_ready(struct socket *so, struct sockbuf *sb)
591 {
592 	return (sb == &so->so_rcv ? soreadable(so) : sowriteable(so));
593 }
594 
595 static void
596 soaio_process_job(struct socket *so, sb_which which, struct kaiocb *job)
597 {
598 	struct ucred *td_savedcred;
599 	struct thread *td;
600 	struct sockbuf *sb = sobuf(so, which);
601 #ifdef MAC
602 	struct file *fp = job->fd_file;
603 #endif
604 	size_t cnt, done, job_total_nbytes __diagused;
605 	long ru_before;
606 	int error, flags;
607 
608 	SOCK_BUF_UNLOCK(so, which);
609 	aio_switch_vmspace(job);
610 	td = curthread;
611 retry:
612 	td_savedcred = td->td_ucred;
613 	td->td_ucred = job->cred;
614 
615 	job_total_nbytes = job->uiop->uio_resid + job->aio_done;
616 	done = job->aio_done;
617 	cnt = job->uiop->uio_resid;
618 	job->uiop->uio_offset = 0;
619 	job->uiop->uio_td = td;
620 	flags = MSG_NBIO;
621 
622 	/*
623 	 * For resource usage accounting, only count a completed request
624 	 * as a single message to avoid counting multiple calls to
625 	 * sosend/soreceive on a blocking socket.
626 	 */
627 
628 	if (sb == &so->so_rcv) {
629 		ru_before = td->td_ru.ru_msgrcv;
630 #ifdef MAC
631 		error = mac_socket_check_receive(fp->f_cred, so);
632 		if (error == 0)
633 
634 #endif
635 			error = soreceive(so, NULL, job->uiop, NULL, NULL,
636 			    &flags);
637 		if (td->td_ru.ru_msgrcv != ru_before)
638 			job->msgrcv = 1;
639 	} else {
640 		if (!TAILQ_EMPTY(&sb->sb_aiojobq))
641 			flags |= MSG_MORETOCOME;
642 		ru_before = td->td_ru.ru_msgsnd;
643 #ifdef MAC
644 		error = mac_socket_check_send(fp->f_cred, so);
645 		if (error == 0)
646 #endif
647 			error = sosend(so, NULL, job->uiop, NULL, NULL, flags,
648 			    td);
649 		if (td->td_ru.ru_msgsnd != ru_before)
650 			job->msgsnd = 1;
651 		if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0) {
652 			PROC_LOCK(job->userproc);
653 			kern_psignal(job->userproc, SIGPIPE);
654 			PROC_UNLOCK(job->userproc);
655 		}
656 	}
657 
658 	done += cnt - job->uiop->uio_resid;
659 	job->aio_done = done;
660 	td->td_ucred = td_savedcred;
661 
662 	if (error == EWOULDBLOCK) {
663 		/*
664 		 * The request was either partially completed or not
665 		 * completed at all due to racing with a read() or
666 		 * write() on the socket.  If the socket is
667 		 * non-blocking, return with any partial completion.
668 		 * If the socket is blocking or if no progress has
669 		 * been made, requeue this request at the head of the
670 		 * queue to try again when the socket is ready.
671 		 */
672 		MPASS(done != job_total_nbytes);
673 		SOCK_BUF_LOCK(so, which);
674 		if (done == 0 || !(so->so_state & SS_NBIO)) {
675 			empty_results++;
676 			if (soaio_ready(so, sb)) {
677 				empty_retries++;
678 				SOCK_BUF_UNLOCK(so, which);
679 				goto retry;
680 			}
681 
682 			if (!aio_set_cancel_function(job, soo_aio_cancel)) {
683 				SOCK_BUF_UNLOCK(so, which);
684 				if (done != 0)
685 					aio_complete(job, done, 0);
686 				else
687 					aio_cancel(job);
688 				SOCK_BUF_LOCK(so, which);
689 			} else {
690 				TAILQ_INSERT_HEAD(&sb->sb_aiojobq, job, list);
691 			}
692 			return;
693 		}
694 		SOCK_BUF_UNLOCK(so, which);
695 	}
696 	if (done != 0 && (error == ERESTART || error == EINTR ||
697 	    error == EWOULDBLOCK))
698 		error = 0;
699 	if (error)
700 		aio_complete(job, -1, error);
701 	else
702 		aio_complete(job, done, 0);
703 	SOCK_BUF_LOCK(so, which);
704 }
705 
706 static void
707 soaio_process_sb(struct socket *so, sb_which which)
708 {
709 	struct kaiocb *job;
710 	struct sockbuf *sb = sobuf(so, which);
711 
712 	CURVNET_SET(so->so_vnet);
713 	SOCK_BUF_LOCK(so, which);
714 	while (!TAILQ_EMPTY(&sb->sb_aiojobq) && soaio_ready(so, sb)) {
715 		job = TAILQ_FIRST(&sb->sb_aiojobq);
716 		TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
717 		if (!aio_clear_cancel_function(job))
718 			continue;
719 
720 		soaio_process_job(so, which, job);
721 	}
722 
723 	/*
724 	 * If there are still pending requests, the socket must not be
725 	 * ready so set SB_AIO to request a wakeup when the socket
726 	 * becomes ready.
727 	 */
728 	if (!TAILQ_EMPTY(&sb->sb_aiojobq))
729 		sb->sb_flags |= SB_AIO;
730 	sb->sb_flags &= ~SB_AIO_RUNNING;
731 	SOCK_BUF_UNLOCK(so, which);
732 
733 	sorele(so);
734 	CURVNET_RESTORE();
735 }
736 
737 void
738 soaio_rcv(void *context, int pending)
739 {
740 	struct socket *so;
741 
742 	so = context;
743 	soaio_process_sb(so, SO_RCV);
744 }
745 
746 void
747 soaio_snd(void *context, int pending)
748 {
749 	struct socket *so;
750 
751 	so = context;
752 	soaio_process_sb(so, SO_SND);
753 }
754 
755 void
756 sowakeup_aio(struct socket *so, sb_which which)
757 {
758 	struct sockbuf *sb = sobuf(so, which);
759 
760 	SOCK_BUF_LOCK_ASSERT(so, which);
761 
762 	sb->sb_flags &= ~SB_AIO;
763 	if (sb->sb_flags & SB_AIO_RUNNING)
764 		return;
765 	sb->sb_flags |= SB_AIO_RUNNING;
766 	soref(so);
767 	soaio_enqueue(&sb->sb_aiotask);
768 }
769 
770 static void
771 soo_aio_cancel(struct kaiocb *job)
772 {
773 	struct socket *so;
774 	struct sockbuf *sb;
775 	long done;
776 	int opcode;
777 	sb_which which;
778 
779 	so = job->fd_file->f_data;
780 	opcode = job->uaiocb.aio_lio_opcode;
781 	if (opcode & LIO_READ) {
782 		sb = &so->so_rcv;
783 		which = SO_RCV;
784 	} else {
785 		MPASS(opcode & LIO_WRITE);
786 		sb = &so->so_snd;
787 		which = SO_SND;
788 	}
789 
790 	SOCK_BUF_LOCK(so, which);
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 	SOCK_BUF_UNLOCK(so, which);
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
805 soo_aio_queue(struct file *fp, struct kaiocb *job)
806 {
807 	struct socket *so;
808 	struct sockbuf *sb;
809 	sb_which which;
810 	int error;
811 
812 	so = fp->f_data;
813 	error = (*so->so_proto->pr_usrreqs->pru_aio_queue)(so, job);
814 	if (error == 0)
815 		return (0);
816 
817 	/* Lock through the socket, since this may be a listening socket. */
818 	switch (job->uaiocb.aio_lio_opcode & (LIO_WRITE | LIO_READ)) {
819 	case LIO_READ:
820 		SOCK_RECVBUF_LOCK(so);
821 		sb = &so->so_rcv;
822 		which = SO_RCV;
823 		break;
824 	case LIO_WRITE:
825 		SOCK_SENDBUF_LOCK(so);
826 		sb = &so->so_snd;
827 		which = SO_SND;
828 		break;
829 	default:
830 		return (EINVAL);
831 	}
832 
833 	if (SOLISTENING(so)) {
834 		SOCK_BUF_UNLOCK(so, which);
835 		return (EINVAL);
836 	}
837 
838 	if (!aio_set_cancel_function(job, soo_aio_cancel))
839 		panic("new job was cancelled");
840 	TAILQ_INSERT_TAIL(&sb->sb_aiojobq, job, list);
841 	if (!(sb->sb_flags & SB_AIO_RUNNING)) {
842 		if (soaio_ready(so, sb))
843 			sowakeup_aio(so, which);
844 		else
845 			sb->sb_flags |= SB_AIO;
846 	}
847 	SOCK_BUF_UNLOCK(so, which);
848 	return (0);
849 }
850