xref: /freebsd/sys/kern/sys_socket.c (revision f81cdf24ba5436367377f7c8e8f51f6df2a75ca7)
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_fill_kinfo_t soo_fill_kinfo;
94 static fo_aio_queue_t soo_aio_queue;
95 
96 static void	soo_aio_cancel(struct kaiocb *job);
97 
98 struct fileops	socketops = {
99 	.fo_read = soo_read,
100 	.fo_write = soo_write,
101 	.fo_truncate = invfo_truncate,
102 	.fo_ioctl = soo_ioctl,
103 	.fo_poll = soo_poll,
104 	.fo_kqfilter = soo_kqfilter,
105 	.fo_stat = soo_stat,
106 	.fo_close = soo_close,
107 	.fo_chmod = invfo_chmod,
108 	.fo_chown = invfo_chown,
109 	.fo_sendfile = invfo_sendfile,
110 	.fo_fill_kinfo = soo_fill_kinfo,
111 	.fo_aio_queue = soo_aio_queue,
112 	.fo_cmp = file_kcmp_generic,
113 	.fo_flags = DFLAG_PASSABLE
114 };
115 
116 static int
117 soo_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
118     int flags, struct thread *td)
119 {
120 	struct socket *so = fp->f_data;
121 	int error;
122 
123 #ifdef MAC
124 	error = mac_socket_check_receive(active_cred, so);
125 	if (error)
126 		return (error);
127 #endif
128 	error = soreceive(so, 0, uio, 0, 0, 0);
129 	return (error);
130 }
131 
132 static int
133 soo_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
134     int flags, struct thread *td)
135 {
136 	struct socket *so = fp->f_data;
137 	int error;
138 
139 #ifdef MAC
140 	error = mac_socket_check_send(active_cred, so);
141 	if (error)
142 		return (error);
143 #endif
144 	error = sousrsend(so, NULL, uio, NULL, 0, NULL);
145 	return (error);
146 }
147 
148 static int
149 soo_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *active_cred,
150     struct thread *td)
151 {
152 	struct socket *so = fp->f_data;
153 	int error = 0;
154 
155 	switch (cmd) {
156 	case FIONBIO:
157 		SOCK_LOCK(so);
158 		if (*(int *)data)
159 			so->so_state |= SS_NBIO;
160 		else
161 			so->so_state &= ~SS_NBIO;
162 		SOCK_UNLOCK(so);
163 		break;
164 
165 	case FIOASYNC:
166 		if (*(int *)data) {
167 			SOCK_LOCK(so);
168 			so->so_state |= SS_ASYNC;
169 			if (SOLISTENING(so)) {
170 				so->sol_sbrcv_flags |= SB_ASYNC;
171 				so->sol_sbsnd_flags |= SB_ASYNC;
172 			} else {
173 				SOCK_RECVBUF_LOCK(so);
174 				so->so_rcv.sb_flags |= SB_ASYNC;
175 				SOCK_RECVBUF_UNLOCK(so);
176 				SOCK_SENDBUF_LOCK(so);
177 				so->so_snd.sb_flags |= SB_ASYNC;
178 				SOCK_SENDBUF_UNLOCK(so);
179 			}
180 			SOCK_UNLOCK(so);
181 		} else {
182 			SOCK_LOCK(so);
183 			so->so_state &= ~SS_ASYNC;
184 			if (SOLISTENING(so)) {
185 				so->sol_sbrcv_flags &= ~SB_ASYNC;
186 				so->sol_sbsnd_flags &= ~SB_ASYNC;
187 			} else {
188 				SOCK_RECVBUF_LOCK(so);
189 				so->so_rcv.sb_flags &= ~SB_ASYNC;
190 				SOCK_RECVBUF_UNLOCK(so);
191 				SOCK_SENDBUF_LOCK(so);
192 				so->so_snd.sb_flags &= ~SB_ASYNC;
193 				SOCK_SENDBUF_UNLOCK(so);
194 			}
195 			SOCK_UNLOCK(so);
196 		}
197 		break;
198 
199 	case FIONREAD:
200 		SOCK_RECVBUF_LOCK(so);
201 		if (SOLISTENING(so)) {
202 			error = EINVAL;
203 		} else {
204 			*(int *)data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
205 		}
206 		SOCK_RECVBUF_UNLOCK(so);
207 		break;
208 
209 	case FIONWRITE:
210 		/* Unlocked read. */
211 		if (SOLISTENING(so)) {
212 			error = EINVAL;
213 		} else {
214 			*(int *)data = sbavail(&so->so_snd);
215 		}
216 		break;
217 
218 	case FIONSPACE:
219 		/* Unlocked read. */
220 		if (SOLISTENING(so)) {
221 			error = EINVAL;
222 		} else {
223 			if ((so->so_snd.sb_hiwat < sbused(&so->so_snd)) ||
224 			    (so->so_snd.sb_mbmax < so->so_snd.sb_mbcnt)) {
225 				*(int *)data = 0;
226 			} else {
227 				*(int *)data = sbspace(&so->so_snd);
228 			}
229 		}
230 		break;
231 
232 	case FIOSETOWN:
233 		error = fsetown(*(int *)data, &so->so_sigio);
234 		break;
235 
236 	case FIOGETOWN:
237 		*(int *)data = fgetown(&so->so_sigio);
238 		break;
239 
240 	case SIOCSPGRP:
241 		error = fsetown(-(*(int *)data), &so->so_sigio);
242 		break;
243 
244 	case SIOCGPGRP:
245 		*(int *)data = -fgetown(&so->so_sigio);
246 		break;
247 
248 	case SIOCATMARK:
249 		/* Unlocked read. */
250 		if (SOLISTENING(so)) {
251 			error = EINVAL;
252 		} else {
253 			*(int *)data = (so->so_rcv.sb_state & SBS_RCVATMARK) != 0;
254 		}
255 		break;
256 	default:
257 		/*
258 		 * Interface/routing/protocol specific ioctls: interface and
259 		 * routing ioctls should have a different entry since a
260 		 * socket is unnecessary.
261 		 */
262 		if (IOCGROUP(cmd) == 'i')
263 			error = ifioctl(so, cmd, data, td);
264 		else if (IOCGROUP(cmd) == 'r') {
265 			CURVNET_SET(so->so_vnet);
266 			error = rtioctl_fib(cmd, data, so->so_fibnum);
267 			CURVNET_RESTORE();
268 		} else {
269 			CURVNET_SET(so->so_vnet);
270 			error = so->so_proto->pr_control(so, cmd, data, 0, td);
271 			CURVNET_RESTORE();
272 		}
273 		break;
274 	}
275 	return (error);
276 }
277 
278 static int
279 soo_poll(struct file *fp, int events, struct ucred *active_cred,
280     struct thread *td)
281 {
282 	struct socket *so = fp->f_data;
283 #ifdef MAC
284 	int error;
285 
286 	error = mac_socket_check_poll(active_cred, so);
287 	if (error)
288 		return (error);
289 #endif
290 	return (sopoll(so, events, fp->f_cred, td));
291 }
292 
293 static int
294 soo_stat(struct file *fp, struct stat *ub, struct ucred *active_cred)
295 {
296 	struct socket *so = fp->f_data;
297 	int error = 0;
298 
299 	bzero((caddr_t)ub, sizeof (*ub));
300 	ub->st_mode = S_IFSOCK;
301 #ifdef MAC
302 	error = mac_socket_check_stat(active_cred, so);
303 	if (error)
304 		return (error);
305 #endif
306 	SOCK_LOCK(so);
307 	if (!SOLISTENING(so)) {
308 		struct sockbuf *sb;
309 
310 		/*
311 		 * If SBS_CANTRCVMORE is set, but there's still data left
312 		 * in the receive buffer, the socket is still readable.
313 		 */
314 		sb = &so->so_rcv;
315 		SOCK_RECVBUF_LOCK(so);
316 		if ((sb->sb_state & SBS_CANTRCVMORE) == 0 || sbavail(sb))
317 			ub->st_mode |= S_IRUSR | S_IRGRP | S_IROTH;
318 		ub->st_size = sbavail(sb) - sb->sb_ctl;
319 		SOCK_RECVBUF_UNLOCK(so);
320 
321 		sb = &so->so_snd;
322 		SOCK_SENDBUF_LOCK(so);
323 		if ((sb->sb_state & SBS_CANTSENDMORE) == 0)
324 			ub->st_mode |= S_IWUSR | S_IWGRP | S_IWOTH;
325 		SOCK_SENDBUF_UNLOCK(so);
326 	}
327 	ub->st_uid = so->so_cred->cr_uid;
328 	ub->st_gid = so->so_cred->cr_gid;
329 	if (so->so_proto->pr_sense)
330 		error = so->so_proto->pr_sense(so, ub);
331 	SOCK_UNLOCK(so);
332 	return (error);
333 }
334 
335 /*
336  * API socket close on file pointer.  We call soclose() to close the socket
337  * (including initiating closing protocols).  soclose() will sorele() the
338  * file reference but the actual socket will not go away until the socket's
339  * ref count hits 0.
340  */
341 static int
342 soo_close(struct file *fp, struct thread *td)
343 {
344 	int error = 0;
345 	struct socket *so;
346 
347 	so = fp->f_data;
348 	fp->f_ops = &badfileops;
349 	fp->f_data = NULL;
350 
351 	if (so)
352 		error = soclose(so);
353 	return (error);
354 }
355 
356 static int
357 soo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
358 {
359 	struct sockaddr_storage ss = { .ss_len = sizeof(ss) };
360 	struct inpcb *inpcb;
361 	struct unpcb *unpcb;
362 	struct socket *so;
363 	int error;
364 
365 	kif->kf_type = KF_TYPE_SOCKET;
366 	so = fp->f_data;
367 	CURVNET_SET(so->so_vnet);
368 	kif->kf_un.kf_sock.kf_sock_domain0 =
369 	    so->so_proto->pr_domain->dom_family;
370 	kif->kf_un.kf_sock.kf_sock_type0 = so->so_type;
371 	kif->kf_un.kf_sock.kf_sock_protocol0 = so->so_proto->pr_protocol;
372 	kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb;
373 	switch (kif->kf_un.kf_sock.kf_sock_domain0) {
374 	case AF_INET:
375 	case AF_INET6:
376 		if (so->so_pcb != NULL) {
377 			inpcb = (struct inpcb *)(so->so_pcb);
378 			kif->kf_un.kf_sock.kf_sock_inpcb =
379 			    (uintptr_t)inpcb->inp_ppcb;
380 		}
381 		kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
382 		    so->so_rcv.sb_state;
383 		kif->kf_un.kf_sock.kf_sock_snd_sb_state =
384 		    so->so_snd.sb_state;
385 		kif->kf_un.kf_sock.kf_sock_sendq =
386 		    sbused(&so->so_snd);
387 		kif->kf_un.kf_sock.kf_sock_recvq =
388 		    sbused(&so->so_rcv);
389 		break;
390 	case AF_UNIX:
391 		if (so->so_pcb != NULL) {
392 			unpcb = (struct unpcb *)(so->so_pcb);
393 			if (unpcb->unp_conn) {
394 				kif->kf_un.kf_sock.kf_sock_unpconn =
395 				    (uintptr_t)unpcb->unp_conn;
396 				kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
397 				    so->so_rcv.sb_state;
398 				kif->kf_un.kf_sock.kf_sock_snd_sb_state =
399 				    so->so_snd.sb_state;
400 				kif->kf_un.kf_sock.kf_sock_sendq =
401 				    sbused(&so->so_snd);
402 				kif->kf_un.kf_sock.kf_sock_recvq =
403 				    sbused(&so->so_rcv);
404 			}
405 		}
406 		break;
407 	}
408 	error = sosockaddr(so, (struct sockaddr *)&ss);
409 	if (error == 0 &&
410 	    ss.ss_len <= sizeof(kif->kf_un.kf_sock.kf_sa_local)) {
411 		bcopy(&ss, &kif->kf_un.kf_sock.kf_sa_local, ss.ss_len);
412 	}
413 	ss.ss_len = sizeof(ss);
414 	error = sopeeraddr(so, (struct sockaddr *)&ss);
415 	if (error == 0 &&
416 	    ss.ss_len <= sizeof(kif->kf_un.kf_sock.kf_sa_peer)) {
417 		bcopy(&ss, &kif->kf_un.kf_sock.kf_sa_peer, ss.ss_len);
418 	}
419 	strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name,
420 	    sizeof(kif->kf_path));
421 	CURVNET_RESTORE();
422 	return (0);
423 }
424 
425 /*
426  * Use the 'backend3' field in AIO jobs to store the amount of data
427  * completed by the AIO job so far.
428  */
429 #define	aio_done	backend3
430 
431 static STAILQ_HEAD(, task) soaio_jobs;
432 static struct mtx soaio_jobs_lock;
433 static struct task soaio_kproc_task;
434 static int soaio_starting, soaio_idle, soaio_queued;
435 static struct unrhdr *soaio_kproc_unr;
436 
437 static int soaio_max_procs = MAX_AIO_PROCS;
438 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, max_procs, CTLFLAG_RW, &soaio_max_procs, 0,
439     "Maximum number of kernel processes to use for async socket IO");
440 
441 static int soaio_num_procs;
442 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, num_procs, CTLFLAG_RD, &soaio_num_procs, 0,
443     "Number of active kernel processes for async socket IO");
444 
445 static int soaio_target_procs = TARGET_AIO_PROCS;
446 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, target_procs, CTLFLAG_RD,
447     &soaio_target_procs, 0,
448     "Preferred number of ready kernel processes for async socket IO");
449 
450 static int soaio_lifetime;
451 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, lifetime, CTLFLAG_RW, &soaio_lifetime, 0,
452     "Maximum lifetime for idle aiod");
453 
454 static void
455 soaio_kproc_loop(void *arg)
456 {
457 	struct proc *p;
458 	struct vmspace *myvm;
459 	struct task *task;
460 	int error, id, pending;
461 
462 	id = (intptr_t)arg;
463 
464 	/*
465 	 * Grab an extra reference on the daemon's vmspace so that it
466 	 * doesn't get freed by jobs that switch to a different
467 	 * vmspace.
468 	 */
469 	p = curproc;
470 	myvm = vmspace_acquire_ref(p);
471 
472 	mtx_lock(&soaio_jobs_lock);
473 	MPASS(soaio_starting > 0);
474 	soaio_starting--;
475 	for (;;) {
476 		while (!STAILQ_EMPTY(&soaio_jobs)) {
477 			task = STAILQ_FIRST(&soaio_jobs);
478 			STAILQ_REMOVE_HEAD(&soaio_jobs, ta_link);
479 			soaio_queued--;
480 			pending = task->ta_pending;
481 			task->ta_pending = 0;
482 			mtx_unlock(&soaio_jobs_lock);
483 
484 			task->ta_func(task->ta_context, pending);
485 
486 			mtx_lock(&soaio_jobs_lock);
487 		}
488 		MPASS(soaio_queued == 0);
489 
490 		if (p->p_vmspace != myvm) {
491 			mtx_unlock(&soaio_jobs_lock);
492 			vmspace_switch_aio(myvm);
493 			mtx_lock(&soaio_jobs_lock);
494 			continue;
495 		}
496 
497 		soaio_idle++;
498 		error = mtx_sleep(&soaio_idle, &soaio_jobs_lock, 0, "-",
499 		    soaio_lifetime);
500 		soaio_idle--;
501 		if (error == EWOULDBLOCK && STAILQ_EMPTY(&soaio_jobs) &&
502 		    soaio_num_procs > soaio_target_procs)
503 			break;
504 	}
505 	soaio_num_procs--;
506 	mtx_unlock(&soaio_jobs_lock);
507 	free_unr(soaio_kproc_unr, id);
508 	kproc_exit(0);
509 }
510 
511 static void
512 soaio_kproc_create(void *context, int pending)
513 {
514 	struct proc *p;
515 	int error, id;
516 
517 	mtx_lock(&soaio_jobs_lock);
518 	for (;;) {
519 		if (soaio_num_procs < soaio_target_procs) {
520 			/* Must create */
521 		} else if (soaio_num_procs >= soaio_max_procs) {
522 			/*
523 			 * Hit the limit on kernel processes, don't
524 			 * create another one.
525 			 */
526 			break;
527 		} else if (soaio_queued <= soaio_idle + soaio_starting) {
528 			/*
529 			 * No more AIO jobs waiting for a process to be
530 			 * created, so stop.
531 			 */
532 			break;
533 		}
534 		soaio_starting++;
535 		mtx_unlock(&soaio_jobs_lock);
536 
537 		id = alloc_unr(soaio_kproc_unr);
538 		error = kproc_create(soaio_kproc_loop, (void *)(intptr_t)id,
539 		    &p, 0, 0, "soaiod%d", id);
540 		if (error != 0) {
541 			free_unr(soaio_kproc_unr, id);
542 			mtx_lock(&soaio_jobs_lock);
543 			soaio_starting--;
544 			break;
545 		}
546 
547 		mtx_lock(&soaio_jobs_lock);
548 		soaio_num_procs++;
549 	}
550 	mtx_unlock(&soaio_jobs_lock);
551 }
552 
553 void
554 soaio_enqueue(struct task *task)
555 {
556 
557 	mtx_lock(&soaio_jobs_lock);
558 	MPASS(task->ta_pending == 0);
559 	task->ta_pending++;
560 	STAILQ_INSERT_TAIL(&soaio_jobs, task, ta_link);
561 	soaio_queued++;
562 	if (soaio_queued <= soaio_idle)
563 		wakeup_one(&soaio_idle);
564 	else if (soaio_num_procs < soaio_max_procs)
565 		taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task);
566 	mtx_unlock(&soaio_jobs_lock);
567 }
568 
569 static void
570 soaio_init(void)
571 {
572 
573 	soaio_lifetime = AIOD_LIFETIME_DEFAULT;
574 	STAILQ_INIT(&soaio_jobs);
575 	mtx_init(&soaio_jobs_lock, "soaio jobs", NULL, MTX_DEF);
576 	soaio_kproc_unr = new_unrhdr(1, INT_MAX, NULL);
577 	TASK_INIT(&soaio_kproc_task, 0, soaio_kproc_create, NULL);
578 }
579 SYSINIT(soaio, SI_SUB_VFS, SI_ORDER_ANY, soaio_init, NULL);
580 
581 static __inline int
582 soaio_ready(struct socket *so, struct sockbuf *sb)
583 {
584 	return (sb == &so->so_rcv ? soreadable(so) : sowriteable(so));
585 }
586 
587 static void
588 soaio_process_job(struct socket *so, sb_which which, struct kaiocb *job)
589 {
590 	struct ucred *td_savedcred;
591 	struct thread *td;
592 	struct sockbuf *sb = sobuf(so, which);
593 #ifdef MAC
594 	struct file *fp = job->fd_file;
595 #endif
596 	size_t cnt, done, job_total_nbytes __diagused;
597 	long ru_before;
598 	int error, flags;
599 
600 	SOCK_BUF_UNLOCK(so, which);
601 	aio_switch_vmspace(job);
602 	td = curthread;
603 retry:
604 	td_savedcred = td->td_ucred;
605 	td->td_ucred = job->cred;
606 
607 	job_total_nbytes = job->uiop->uio_resid + job->aio_done;
608 	done = job->aio_done;
609 	cnt = job->uiop->uio_resid;
610 	job->uiop->uio_offset = 0;
611 	job->uiop->uio_td = td;
612 	flags = MSG_NBIO;
613 
614 	/*
615 	 * For resource usage accounting, only count a completed request
616 	 * as a single message to avoid counting multiple calls to
617 	 * sosend/soreceive on a blocking socket.
618 	 */
619 
620 	if (sb == &so->so_rcv) {
621 		ru_before = td->td_ru.ru_msgrcv;
622 #ifdef MAC
623 		error = mac_socket_check_receive(fp->f_cred, so);
624 		if (error == 0)
625 
626 #endif
627 			error = soreceive(so, NULL, job->uiop, NULL, NULL,
628 			    &flags);
629 		if (td->td_ru.ru_msgrcv != ru_before)
630 			job->msgrcv = 1;
631 	} else {
632 		if (!TAILQ_EMPTY(&sb->sb_aiojobq))
633 			flags |= MSG_MORETOCOME;
634 		ru_before = td->td_ru.ru_msgsnd;
635 #ifdef MAC
636 		error = mac_socket_check_send(fp->f_cred, so);
637 		if (error == 0)
638 #endif
639 			error = sousrsend(so, NULL, job->uiop, NULL, flags,
640 			    job->userproc);
641 		if (td->td_ru.ru_msgsnd != ru_before)
642 			job->msgsnd = 1;
643 	}
644 
645 	done += cnt - job->uiop->uio_resid;
646 	job->aio_done = done;
647 	td->td_ucred = td_savedcred;
648 
649 	if (error == EWOULDBLOCK) {
650 		/*
651 		 * The request was either partially completed or not
652 		 * completed at all due to racing with a read() or
653 		 * write() on the socket.  If the socket is
654 		 * non-blocking, return with any partial completion.
655 		 * If the socket is blocking or if no progress has
656 		 * been made, requeue this request at the head of the
657 		 * queue to try again when the socket is ready.
658 		 */
659 		MPASS(done != job_total_nbytes);
660 		SOCK_BUF_LOCK(so, which);
661 		if (done == 0 || !(so->so_state & SS_NBIO)) {
662 			empty_results++;
663 			if (soaio_ready(so, sb)) {
664 				empty_retries++;
665 				SOCK_BUF_UNLOCK(so, which);
666 				goto retry;
667 			}
668 
669 			if (!aio_set_cancel_function(job, soo_aio_cancel)) {
670 				SOCK_BUF_UNLOCK(so, which);
671 				if (done != 0)
672 					aio_complete(job, done, 0);
673 				else
674 					aio_cancel(job);
675 				SOCK_BUF_LOCK(so, which);
676 			} else {
677 				TAILQ_INSERT_HEAD(&sb->sb_aiojobq, job, list);
678 			}
679 			return;
680 		}
681 		SOCK_BUF_UNLOCK(so, which);
682 	}
683 	if (done != 0 && (error == ERESTART || error == EINTR ||
684 	    error == EWOULDBLOCK))
685 		error = 0;
686 	if (error)
687 		aio_complete(job, -1, error);
688 	else
689 		aio_complete(job, done, 0);
690 	SOCK_BUF_LOCK(so, which);
691 }
692 
693 static void
694 soaio_process_sb(struct socket *so, sb_which which)
695 {
696 	struct kaiocb *job;
697 	struct sockbuf *sb = sobuf(so, which);
698 
699 	CURVNET_SET(so->so_vnet);
700 	SOCK_BUF_LOCK(so, which);
701 	while (!TAILQ_EMPTY(&sb->sb_aiojobq) && soaio_ready(so, sb)) {
702 		job = TAILQ_FIRST(&sb->sb_aiojobq);
703 		TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
704 		if (!aio_clear_cancel_function(job))
705 			continue;
706 
707 		soaio_process_job(so, which, job);
708 	}
709 
710 	/*
711 	 * If there are still pending requests, the socket must not be
712 	 * ready so set SB_AIO to request a wakeup when the socket
713 	 * becomes ready.
714 	 */
715 	if (!TAILQ_EMPTY(&sb->sb_aiojobq))
716 		sb->sb_flags |= SB_AIO;
717 	sb->sb_flags &= ~SB_AIO_RUNNING;
718 	SOCK_BUF_UNLOCK(so, which);
719 
720 	sorele(so);
721 	CURVNET_RESTORE();
722 }
723 
724 void
725 soaio_rcv(void *context, int pending)
726 {
727 	struct socket *so;
728 
729 	so = context;
730 	soaio_process_sb(so, SO_RCV);
731 }
732 
733 void
734 soaio_snd(void *context, int pending)
735 {
736 	struct socket *so;
737 
738 	so = context;
739 	soaio_process_sb(so, SO_SND);
740 }
741 
742 void
743 sowakeup_aio(struct socket *so, sb_which which)
744 {
745 	struct sockbuf *sb = sobuf(so, which);
746 
747 	SOCK_BUF_LOCK_ASSERT(so, which);
748 
749 	sb->sb_flags &= ~SB_AIO;
750 	if (sb->sb_flags & SB_AIO_RUNNING)
751 		return;
752 	sb->sb_flags |= SB_AIO_RUNNING;
753 	soref(so);
754 	soaio_enqueue(&sb->sb_aiotask);
755 }
756 
757 static void
758 soo_aio_cancel(struct kaiocb *job)
759 {
760 	struct socket *so;
761 	struct sockbuf *sb;
762 	long done;
763 	int opcode;
764 	sb_which which;
765 
766 	so = job->fd_file->f_data;
767 	opcode = job->uaiocb.aio_lio_opcode;
768 	if (opcode & LIO_READ) {
769 		sb = &so->so_rcv;
770 		which = SO_RCV;
771 	} else {
772 		MPASS(opcode & LIO_WRITE);
773 		sb = &so->so_snd;
774 		which = SO_SND;
775 	}
776 
777 	SOCK_BUF_LOCK(so, which);
778 	if (!aio_cancel_cleared(job))
779 		TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
780 	if (TAILQ_EMPTY(&sb->sb_aiojobq))
781 		sb->sb_flags &= ~SB_AIO;
782 	SOCK_BUF_UNLOCK(so, which);
783 
784 	done = job->aio_done;
785 	if (done != 0)
786 		aio_complete(job, done, 0);
787 	else
788 		aio_cancel(job);
789 }
790 
791 static int
792 soo_aio_queue(struct file *fp, struct kaiocb *job)
793 {
794 	struct socket *so;
795 	struct sockbuf *sb;
796 	sb_which which;
797 	int error;
798 
799 	so = fp->f_data;
800 	error = so->so_proto->pr_aio_queue(so, job);
801 	if (error == 0)
802 		return (0);
803 
804 	/* Lock through the socket, since this may be a listening socket. */
805 	switch (job->uaiocb.aio_lio_opcode & (LIO_WRITE | LIO_READ)) {
806 	case LIO_READ:
807 		SOCK_RECVBUF_LOCK(so);
808 		sb = &so->so_rcv;
809 		which = SO_RCV;
810 		break;
811 	case LIO_WRITE:
812 		SOCK_SENDBUF_LOCK(so);
813 		sb = &so->so_snd;
814 		which = SO_SND;
815 		break;
816 	default:
817 		return (EINVAL);
818 	}
819 
820 	if (SOLISTENING(so)) {
821 		SOCK_BUF_UNLOCK(so, which);
822 		return (EINVAL);
823 	}
824 
825 	if (!aio_set_cancel_function(job, soo_aio_cancel))
826 		panic("new job was cancelled");
827 	TAILQ_INSERT_TAIL(&sb->sb_aiojobq, job, list);
828 	if (!(sb->sb_flags & SB_AIO_RUNNING)) {
829 		if (soaio_ready(so, sb))
830 			sowakeup_aio(so, which);
831 		else
832 			sb->sb_flags |= SB_AIO;
833 	}
834 	SOCK_BUF_UNLOCK(so, which);
835 	return (0);
836 }
837