xref: /freebsd/sys/kern/sys_socket.c (revision 389e4940069316fe667ffa263fa7d6390d0a960f)
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, 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 				SOCKBUF_LOCK(&so->so_rcv);
183 				so->so_rcv.sb_flags |= SB_ASYNC;
184 				SOCKBUF_UNLOCK(&so->so_rcv);
185 				SOCKBUF_LOCK(&so->so_snd);
186 				so->so_snd.sb_flags |= SB_ASYNC;
187 				SOCKBUF_UNLOCK(&so->so_snd);
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 				SOCKBUF_LOCK(&so->so_rcv);
198 				so->so_rcv.sb_flags &= ~SB_ASYNC;
199 				SOCKBUF_UNLOCK(&so->so_rcv);
200 				SOCKBUF_LOCK(&so->so_snd);
201 				so->so_snd.sb_flags &= ~SB_ASYNC;
202 				SOCKBUF_UNLOCK(&so->so_snd);
203 			}
204 			SOCK_UNLOCK(so);
205 		}
206 		break;
207 
208 	case FIONREAD:
209 		/* Unlocked read. */
210 		*(int *)data = sbavail(&so->so_rcv);
211 		break;
212 
213 	case FIONWRITE:
214 		/* Unlocked read. */
215 		*(int *)data = sbavail(&so->so_snd);
216 		break;
217 
218 	case FIONSPACE:
219 		/* Unlocked read. */
220 		if ((so->so_snd.sb_hiwat < sbused(&so->so_snd)) ||
221 		    (so->so_snd.sb_mbmax < so->so_snd.sb_mbcnt))
222 			*(int *)data = 0;
223 		else
224 			*(int *)data = sbspace(&so->so_snd);
225 		break;
226 
227 	case FIOSETOWN:
228 		error = fsetown(*(int *)data, &so->so_sigio);
229 		break;
230 
231 	case FIOGETOWN:
232 		*(int *)data = fgetown(&so->so_sigio);
233 		break;
234 
235 	case SIOCSPGRP:
236 		error = fsetown(-(*(int *)data), &so->so_sigio);
237 		break;
238 
239 	case SIOCGPGRP:
240 		*(int *)data = -fgetown(&so->so_sigio);
241 		break;
242 
243 	case SIOCATMARK:
244 		/* Unlocked read. */
245 		*(int *)data = (so->so_rcv.sb_state & SBS_RCVATMARK) != 0;
246 		break;
247 	default:
248 		/*
249 		 * Interface/routing/protocol specific ioctls: interface and
250 		 * routing ioctls should have a different entry since a
251 		 * socket is unnecessary.
252 		 */
253 		if (IOCGROUP(cmd) == 'i')
254 			error = ifioctl(so, cmd, data, td);
255 		else if (IOCGROUP(cmd) == 'r') {
256 			CURVNET_SET(so->so_vnet);
257 			error = rtioctl_fib(cmd, data, so->so_fibnum);
258 			CURVNET_RESTORE();
259 		} else {
260 			CURVNET_SET(so->so_vnet);
261 			error = ((*so->so_proto->pr_usrreqs->pru_control)
262 			    (so, cmd, data, 0, td));
263 			CURVNET_RESTORE();
264 		}
265 		break;
266 	}
267 	return (error);
268 }
269 
270 static int
271 soo_poll(struct file *fp, int events, struct ucred *active_cred,
272     struct thread *td)
273 {
274 	struct socket *so = fp->f_data;
275 #ifdef MAC
276 	int error;
277 
278 	error = mac_socket_check_poll(active_cred, so);
279 	if (error)
280 		return (error);
281 #endif
282 	return (sopoll(so, events, fp->f_cred, td));
283 }
284 
285 static int
286 soo_stat(struct file *fp, struct stat *ub, struct ucred *active_cred,
287     struct thread *td)
288 {
289 	struct socket *so = fp->f_data;
290 #ifdef MAC
291 	int error;
292 #endif
293 
294 	bzero((caddr_t)ub, sizeof (*ub));
295 	ub->st_mode = S_IFSOCK;
296 #ifdef MAC
297 	error = mac_socket_check_stat(active_cred, so);
298 	if (error)
299 		return (error);
300 #endif
301 	if (!SOLISTENING(so)) {
302 		struct sockbuf *sb;
303 
304 		/*
305 		 * If SBS_CANTRCVMORE is set, but there's still data left
306 		 * in the receive buffer, the socket is still readable.
307 		 */
308 		sb = &so->so_rcv;
309 		SOCKBUF_LOCK(sb);
310 		if ((sb->sb_state & SBS_CANTRCVMORE) == 0 || sbavail(sb))
311 			ub->st_mode |= S_IRUSR | S_IRGRP | S_IROTH;
312 		ub->st_size = sbavail(sb) - sb->sb_ctl;
313 		SOCKBUF_UNLOCK(sb);
314 
315 		sb = &so->so_snd;
316 		SOCKBUF_LOCK(sb);
317 		if ((sb->sb_state & SBS_CANTSENDMORE) == 0)
318 			ub->st_mode |= S_IWUSR | S_IWGRP | S_IWOTH;
319 		SOCKBUF_UNLOCK(sb);
320 	}
321 	ub->st_uid = so->so_cred->cr_uid;
322 	ub->st_gid = so->so_cred->cr_gid;
323 	return (*so->so_proto->pr_usrreqs->pru_sense)(so, ub);
324 }
325 
326 /*
327  * API socket close on file pointer.  We call soclose() to close the socket
328  * (including initiating closing protocols).  soclose() will sorele() the
329  * file reference but the actual socket will not go away until the socket's
330  * ref count hits 0.
331  */
332 static int
333 soo_close(struct file *fp, struct thread *td)
334 {
335 	int error = 0;
336 	struct socket *so;
337 
338 	so = fp->f_data;
339 	fp->f_ops = &badfileops;
340 	fp->f_data = NULL;
341 
342 	if (so)
343 		error = soclose(so);
344 	return (error);
345 }
346 
347 static int
348 soo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
349 {
350 	struct sockaddr *sa;
351 	struct inpcb *inpcb;
352 	struct unpcb *unpcb;
353 	struct socket *so;
354 	int error;
355 
356 	kif->kf_type = KF_TYPE_SOCKET;
357 	so = fp->f_data;
358 	CURVNET_SET(so->so_vnet);
359 	kif->kf_un.kf_sock.kf_sock_domain0 =
360 	    so->so_proto->pr_domain->dom_family;
361 	kif->kf_un.kf_sock.kf_sock_type0 = so->so_type;
362 	kif->kf_un.kf_sock.kf_sock_protocol0 = so->so_proto->pr_protocol;
363 	kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb;
364 	switch (kif->kf_un.kf_sock.kf_sock_domain0) {
365 	case AF_INET:
366 	case AF_INET6:
367 		if (kif->kf_un.kf_sock.kf_sock_protocol0 == IPPROTO_TCP) {
368 			if (so->so_pcb != NULL) {
369 				inpcb = (struct inpcb *)(so->so_pcb);
370 				kif->kf_un.kf_sock.kf_sock_inpcb =
371 				    (uintptr_t)inpcb->inp_ppcb;
372 				kif->kf_un.kf_sock.kf_sock_sendq =
373 				    sbused(&so->so_snd);
374 				kif->kf_un.kf_sock.kf_sock_recvq =
375 				    sbused(&so->so_rcv);
376 			}
377 		}
378 		break;
379 	case AF_UNIX:
380 		if (so->so_pcb != NULL) {
381 			unpcb = (struct unpcb *)(so->so_pcb);
382 			if (unpcb->unp_conn) {
383 				kif->kf_un.kf_sock.kf_sock_unpconn =
384 				    (uintptr_t)unpcb->unp_conn;
385 				kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
386 				    so->so_rcv.sb_state;
387 				kif->kf_un.kf_sock.kf_sock_snd_sb_state =
388 				    so->so_snd.sb_state;
389 				kif->kf_un.kf_sock.kf_sock_sendq =
390 				    sbused(&so->so_snd);
391 				kif->kf_un.kf_sock.kf_sock_recvq =
392 				    sbused(&so->so_rcv);
393 			}
394 		}
395 		break;
396 	}
397 	error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa);
398 	if (error == 0 &&
399 	    sa->sa_len <= sizeof(kif->kf_un.kf_sock.kf_sa_local)) {
400 		bcopy(sa, &kif->kf_un.kf_sock.kf_sa_local, sa->sa_len);
401 		free(sa, M_SONAME);
402 	}
403 	error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa);
404 	if (error == 0 &&
405 	    sa->sa_len <= sizeof(kif->kf_un.kf_sock.kf_sa_peer)) {
406 		bcopy(sa, &kif->kf_un.kf_sock.kf_sa_peer, sa->sa_len);
407 		free(sa, M_SONAME);
408 	}
409 	strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name,
410 	    sizeof(kif->kf_path));
411 	CURVNET_RESTORE();
412 	return (0);
413 }
414 
415 /*
416  * Use the 'backend3' field in AIO jobs to store the amount of data
417  * completed by the AIO job so far.
418  */
419 #define	aio_done	backend3
420 
421 static STAILQ_HEAD(, task) soaio_jobs;
422 static struct mtx soaio_jobs_lock;
423 static struct task soaio_kproc_task;
424 static int soaio_starting, soaio_idle, soaio_queued;
425 static struct unrhdr *soaio_kproc_unr;
426 
427 static int soaio_max_procs = MAX_AIO_PROCS;
428 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, max_procs, CTLFLAG_RW, &soaio_max_procs, 0,
429     "Maximum number of kernel processes to use for async socket IO");
430 
431 static int soaio_num_procs;
432 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, num_procs, CTLFLAG_RD, &soaio_num_procs, 0,
433     "Number of active kernel processes for async socket IO");
434 
435 static int soaio_target_procs = TARGET_AIO_PROCS;
436 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, target_procs, CTLFLAG_RD,
437     &soaio_target_procs, 0,
438     "Preferred number of ready kernel processes for async socket IO");
439 
440 static int soaio_lifetime;
441 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, lifetime, CTLFLAG_RW, &soaio_lifetime, 0,
442     "Maximum lifetime for idle aiod");
443 
444 static void
445 soaio_kproc_loop(void *arg)
446 {
447 	struct proc *p;
448 	struct vmspace *myvm;
449 	struct task *task;
450 	int error, id, pending;
451 
452 	id = (intptr_t)arg;
453 
454 	/*
455 	 * Grab an extra reference on the daemon's vmspace so that it
456 	 * doesn't get freed by jobs that switch to a different
457 	 * vmspace.
458 	 */
459 	p = curproc;
460 	myvm = vmspace_acquire_ref(p);
461 
462 	mtx_lock(&soaio_jobs_lock);
463 	MPASS(soaio_starting > 0);
464 	soaio_starting--;
465 	for (;;) {
466 		while (!STAILQ_EMPTY(&soaio_jobs)) {
467 			task = STAILQ_FIRST(&soaio_jobs);
468 			STAILQ_REMOVE_HEAD(&soaio_jobs, ta_link);
469 			soaio_queued--;
470 			pending = task->ta_pending;
471 			task->ta_pending = 0;
472 			mtx_unlock(&soaio_jobs_lock);
473 
474 			task->ta_func(task->ta_context, pending);
475 
476 			mtx_lock(&soaio_jobs_lock);
477 		}
478 		MPASS(soaio_queued == 0);
479 
480 		if (p->p_vmspace != myvm) {
481 			mtx_unlock(&soaio_jobs_lock);
482 			vmspace_switch_aio(myvm);
483 			mtx_lock(&soaio_jobs_lock);
484 			continue;
485 		}
486 
487 		soaio_idle++;
488 		error = mtx_sleep(&soaio_idle, &soaio_jobs_lock, 0, "-",
489 		    soaio_lifetime);
490 		soaio_idle--;
491 		if (error == EWOULDBLOCK && STAILQ_EMPTY(&soaio_jobs) &&
492 		    soaio_num_procs > soaio_target_procs)
493 			break;
494 	}
495 	soaio_num_procs--;
496 	mtx_unlock(&soaio_jobs_lock);
497 	free_unr(soaio_kproc_unr, id);
498 	kproc_exit(0);
499 }
500 
501 static void
502 soaio_kproc_create(void *context, int pending)
503 {
504 	struct proc *p;
505 	int error, id;
506 
507 	mtx_lock(&soaio_jobs_lock);
508 	for (;;) {
509 		if (soaio_num_procs < soaio_target_procs) {
510 			/* Must create */
511 		} else if (soaio_num_procs >= soaio_max_procs) {
512 			/*
513 			 * Hit the limit on kernel processes, don't
514 			 * create another one.
515 			 */
516 			break;
517 		} else if (soaio_queued <= soaio_idle + soaio_starting) {
518 			/*
519 			 * No more AIO jobs waiting for a process to be
520 			 * created, so stop.
521 			 */
522 			break;
523 		}
524 		soaio_starting++;
525 		mtx_unlock(&soaio_jobs_lock);
526 
527 		id = alloc_unr(soaio_kproc_unr);
528 		error = kproc_create(soaio_kproc_loop, (void *)(intptr_t)id,
529 		    &p, 0, 0, "soaiod%d", id);
530 		if (error != 0) {
531 			free_unr(soaio_kproc_unr, id);
532 			mtx_lock(&soaio_jobs_lock);
533 			soaio_starting--;
534 			break;
535 		}
536 
537 		mtx_lock(&soaio_jobs_lock);
538 		soaio_num_procs++;
539 	}
540 	mtx_unlock(&soaio_jobs_lock);
541 }
542 
543 void
544 soaio_enqueue(struct task *task)
545 {
546 
547 	mtx_lock(&soaio_jobs_lock);
548 	MPASS(task->ta_pending == 0);
549 	task->ta_pending++;
550 	STAILQ_INSERT_TAIL(&soaio_jobs, task, ta_link);
551 	soaio_queued++;
552 	if (soaio_queued <= soaio_idle)
553 		wakeup_one(&soaio_idle);
554 	else if (soaio_num_procs < soaio_max_procs)
555 		taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task);
556 	mtx_unlock(&soaio_jobs_lock);
557 }
558 
559 static void
560 soaio_init(void)
561 {
562 
563 	soaio_lifetime = AIOD_LIFETIME_DEFAULT;
564 	STAILQ_INIT(&soaio_jobs);
565 	mtx_init(&soaio_jobs_lock, "soaio jobs", NULL, MTX_DEF);
566 	soaio_kproc_unr = new_unrhdr(1, INT_MAX, NULL);
567 	TASK_INIT(&soaio_kproc_task, 0, soaio_kproc_create, NULL);
568 	if (soaio_target_procs > 0)
569 		taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task);
570 }
571 SYSINIT(soaio, SI_SUB_VFS, SI_ORDER_ANY, soaio_init, NULL);
572 
573 static __inline int
574 soaio_ready(struct socket *so, struct sockbuf *sb)
575 {
576 	return (sb == &so->so_rcv ? soreadable(so) : sowriteable(so));
577 }
578 
579 static void
580 soaio_process_job(struct socket *so, struct sockbuf *sb, struct kaiocb *job)
581 {
582 	struct ucred *td_savedcred;
583 	struct thread *td;
584 	struct file *fp;
585 	struct uio uio;
586 	struct iovec iov;
587 	size_t cnt, done;
588 	long ru_before;
589 	int error, flags;
590 
591 	SOCKBUF_UNLOCK(sb);
592 	aio_switch_vmspace(job);
593 	td = curthread;
594 	fp = job->fd_file;
595 retry:
596 	td_savedcred = td->td_ucred;
597 	td->td_ucred = job->cred;
598 
599 	done = job->aio_done;
600 	cnt = job->uaiocb.aio_nbytes - done;
601 	iov.iov_base = (void *)((uintptr_t)job->uaiocb.aio_buf + done);
602 	iov.iov_len = cnt;
603 	uio.uio_iov = &iov;
604 	uio.uio_iovcnt = 1;
605 	uio.uio_offset = 0;
606 	uio.uio_resid = cnt;
607 	uio.uio_segflg = UIO_USERSPACE;
608 	uio.uio_td = td;
609 	flags = MSG_NBIO;
610 
611 	/*
612 	 * For resource usage accounting, only count a completed request
613 	 * as a single message to avoid counting multiple calls to
614 	 * sosend/soreceive on a blocking socket.
615 	 */
616 
617 	if (sb == &so->so_rcv) {
618 		uio.uio_rw = UIO_READ;
619 		ru_before = td->td_ru.ru_msgrcv;
620 #ifdef MAC
621 		error = mac_socket_check_receive(fp->f_cred, so);
622 		if (error == 0)
623 
624 #endif
625 			error = soreceive(so, NULL, &uio, NULL, NULL, &flags);
626 		if (td->td_ru.ru_msgrcv != ru_before)
627 			job->msgrcv = 1;
628 	} else {
629 		if (!TAILQ_EMPTY(&sb->sb_aiojobq))
630 			flags |= MSG_MORETOCOME;
631 		uio.uio_rw = UIO_WRITE;
632 		ru_before = td->td_ru.ru_msgsnd;
633 #ifdef MAC
634 		error = mac_socket_check_send(fp->f_cred, so);
635 		if (error == 0)
636 #endif
637 			error = sosend(so, NULL, &uio, NULL, NULL, flags, td);
638 		if (td->td_ru.ru_msgsnd != ru_before)
639 			job->msgsnd = 1;
640 		if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0) {
641 			PROC_LOCK(job->userproc);
642 			kern_psignal(job->userproc, SIGPIPE);
643 			PROC_UNLOCK(job->userproc);
644 		}
645 	}
646 
647 	done += cnt - uio.uio_resid;
648 	job->aio_done = done;
649 	td->td_ucred = td_savedcred;
650 
651 	if (error == EWOULDBLOCK) {
652 		/*
653 		 * The request was either partially completed or not
654 		 * completed at all due to racing with a read() or
655 		 * write() on the socket.  If the socket is
656 		 * non-blocking, return with any partial completion.
657 		 * If the socket is blocking or if no progress has
658 		 * been made, requeue this request at the head of the
659 		 * queue to try again when the socket is ready.
660 		 */
661 		MPASS(done != job->uaiocb.aio_nbytes);
662 		SOCKBUF_LOCK(sb);
663 		if (done == 0 || !(so->so_state & SS_NBIO)) {
664 			empty_results++;
665 			if (soaio_ready(so, sb)) {
666 				empty_retries++;
667 				SOCKBUF_UNLOCK(sb);
668 				goto retry;
669 			}
670 
671 			if (!aio_set_cancel_function(job, soo_aio_cancel)) {
672 				SOCKBUF_UNLOCK(sb);
673 				if (done != 0)
674 					aio_complete(job, done, 0);
675 				else
676 					aio_cancel(job);
677 				SOCKBUF_LOCK(sb);
678 			} else {
679 				TAILQ_INSERT_HEAD(&sb->sb_aiojobq, job, list);
680 			}
681 			return;
682 		}
683 		SOCKBUF_UNLOCK(sb);
684 	}
685 	if (done != 0 && (error == ERESTART || error == EINTR ||
686 	    error == EWOULDBLOCK))
687 		error = 0;
688 	if (error)
689 		aio_complete(job, -1, error);
690 	else
691 		aio_complete(job, done, 0);
692 	SOCKBUF_LOCK(sb);
693 }
694 
695 static void
696 soaio_process_sb(struct socket *so, struct sockbuf *sb)
697 {
698 	struct kaiocb *job;
699 
700 	CURVNET_SET(so->so_vnet);
701 	SOCKBUF_LOCK(sb);
702 	while (!TAILQ_EMPTY(&sb->sb_aiojobq) && soaio_ready(so, sb)) {
703 		job = TAILQ_FIRST(&sb->sb_aiojobq);
704 		TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
705 		if (!aio_clear_cancel_function(job))
706 			continue;
707 
708 		soaio_process_job(so, sb, job);
709 	}
710 
711 	/*
712 	 * If there are still pending requests, the socket must not be
713 	 * ready so set SB_AIO to request a wakeup when the socket
714 	 * becomes ready.
715 	 */
716 	if (!TAILQ_EMPTY(&sb->sb_aiojobq))
717 		sb->sb_flags |= SB_AIO;
718 	sb->sb_flags &= ~SB_AIO_RUNNING;
719 	SOCKBUF_UNLOCK(sb);
720 
721 	SOCK_LOCK(so);
722 	sorele(so);
723 	CURVNET_RESTORE();
724 }
725 
726 void
727 soaio_rcv(void *context, int pending)
728 {
729 	struct socket *so;
730 
731 	so = context;
732 	soaio_process_sb(so, &so->so_rcv);
733 }
734 
735 void
736 soaio_snd(void *context, int pending)
737 {
738 	struct socket *so;
739 
740 	so = context;
741 	soaio_process_sb(so, &so->so_snd);
742 }
743 
744 void
745 sowakeup_aio(struct socket *so, struct sockbuf *sb)
746 {
747 
748 	SOCKBUF_LOCK_ASSERT(sb);
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 
765 	so = job->fd_file->f_data;
766 	opcode = job->uaiocb.aio_lio_opcode;
767 	if (opcode == LIO_READ)
768 		sb = &so->so_rcv;
769 	else {
770 		MPASS(opcode == LIO_WRITE);
771 		sb = &so->so_snd;
772 	}
773 
774 	SOCKBUF_LOCK(sb);
775 	if (!aio_cancel_cleared(job))
776 		TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
777 	if (TAILQ_EMPTY(&sb->sb_aiojobq))
778 		sb->sb_flags &= ~SB_AIO;
779 	SOCKBUF_UNLOCK(sb);
780 
781 	done = job->aio_done;
782 	if (done != 0)
783 		aio_complete(job, done, 0);
784 	else
785 		aio_cancel(job);
786 }
787 
788 static int
789 soo_aio_queue(struct file *fp, struct kaiocb *job)
790 {
791 	struct socket *so;
792 	struct sockbuf *sb;
793 	int error;
794 
795 	so = fp->f_data;
796 	error = (*so->so_proto->pr_usrreqs->pru_aio_queue)(so, job);
797 	if (error == 0)
798 		return (0);
799 
800 	switch (job->uaiocb.aio_lio_opcode) {
801 	case LIO_READ:
802 		sb = &so->so_rcv;
803 		break;
804 	case LIO_WRITE:
805 		sb = &so->so_snd;
806 		break;
807 	default:
808 		return (EINVAL);
809 	}
810 
811 	SOCKBUF_LOCK(sb);
812 	if (!aio_set_cancel_function(job, soo_aio_cancel))
813 		panic("new job was cancelled");
814 	TAILQ_INSERT_TAIL(&sb->sb_aiojobq, job, list);
815 	if (!(sb->sb_flags & SB_AIO_RUNNING)) {
816 		if (soaio_ready(so, sb))
817 			sowakeup_aio(so, sb);
818 		else
819 			sb->sb_flags |= SB_AIO;
820 	}
821 	SOCKBUF_UNLOCK(sb);
822 	return (0);
823 }
824