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
75static SYSCTL_NODE(_kern_ipc, OID_AUTO, aio, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
76    "socket AIO stats");
77
78static int empty_results;
79SYSCTL_INT(_kern_ipc_aio, OID_AUTO, empty_results, CTLFLAG_RD, &empty_results,
80    0, "socket operation returned EAGAIN");
81
82static int empty_retries;
83SYSCTL_INT(_kern_ipc_aio, OID_AUTO, empty_retries, CTLFLAG_RD, &empty_retries,
84    0, "socket operation retries");
85
86static fo_rdwr_t soo_read;
87static fo_rdwr_t soo_write;
88static fo_ioctl_t soo_ioctl;
89static fo_poll_t soo_poll;
90extern fo_kqfilter_t soo_kqfilter;
91static fo_stat_t soo_stat;
92static fo_close_t soo_close;
93static fo_fill_kinfo_t soo_fill_kinfo;
94static fo_aio_queue_t soo_aio_queue;
95
96static void	soo_aio_cancel(struct kaiocb *job);
97
98struct 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
116static int
117soo_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
132static int
133soo_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
148static int
149soo_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
278static int
279soo_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
293static int
294soo_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 */
341static int
342soo_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
356static int
357soo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
358{
359	struct sockaddr_storage ss = { .ss_len = sizeof(ss) };
360	struct unpcb *unpcb;
361	struct socket *so;
362	int error;
363
364	kif->kf_type = KF_TYPE_SOCKET;
365	so = fp->f_data;
366	CURVNET_SET(so->so_vnet);
367	kif->kf_un.kf_sock.kf_sock_domain0 =
368	    so->so_proto->pr_domain->dom_family;
369	kif->kf_un.kf_sock.kf_sock_type0 = so->so_type;
370	kif->kf_un.kf_sock.kf_sock_protocol0 = so->so_proto->pr_protocol;
371	kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb;
372	switch (kif->kf_un.kf_sock.kf_sock_domain0) {
373	case AF_INET:
374	case AF_INET6:
375		/* XXX: kf_sock_inpcb is obsolete.  It may be removed. */
376		kif->kf_un.kf_sock.kf_sock_inpcb = (uintptr_t)so->so_pcb;
377		kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
378		    so->so_rcv.sb_state;
379		kif->kf_un.kf_sock.kf_sock_snd_sb_state =
380		    so->so_snd.sb_state;
381		kif->kf_un.kf_sock.kf_sock_sendq =
382		    sbused(&so->so_snd);
383		kif->kf_un.kf_sock.kf_sock_recvq =
384		    sbused(&so->so_rcv);
385		break;
386	case AF_UNIX:
387		if (so->so_pcb != NULL) {
388			unpcb = (struct unpcb *)(so->so_pcb);
389			if (unpcb->unp_conn) {
390				kif->kf_un.kf_sock.kf_sock_unpconn =
391				    (uintptr_t)unpcb->unp_conn;
392				kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
393				    so->so_rcv.sb_state;
394				kif->kf_un.kf_sock.kf_sock_snd_sb_state =
395				    so->so_snd.sb_state;
396				kif->kf_un.kf_sock.kf_sock_sendq =
397				    sbused(&so->so_snd);
398				kif->kf_un.kf_sock.kf_sock_recvq =
399				    sbused(&so->so_rcv);
400			}
401		}
402		break;
403	}
404	error = sosockaddr(so, (struct sockaddr *)&ss);
405	if (error == 0 &&
406	    ss.ss_len <= sizeof(kif->kf_un.kf_sock.kf_sa_local)) {
407		bcopy(&ss, &kif->kf_un.kf_sock.kf_sa_local, ss.ss_len);
408	}
409	ss.ss_len = sizeof(ss);
410	error = sopeeraddr(so, (struct sockaddr *)&ss);
411	if (error == 0 &&
412	    ss.ss_len <= sizeof(kif->kf_un.kf_sock.kf_sa_peer)) {
413		bcopy(&ss, &kif->kf_un.kf_sock.kf_sa_peer, ss.ss_len);
414	}
415	strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name,
416	    sizeof(kif->kf_path));
417	CURVNET_RESTORE();
418	return (0);
419}
420
421/*
422 * Use the 'backend3' field in AIO jobs to store the amount of data
423 * completed by the AIO job so far.
424 */
425#define	aio_done	backend3
426
427static STAILQ_HEAD(, task) soaio_jobs;
428static struct mtx soaio_jobs_lock;
429static struct task soaio_kproc_task;
430static int soaio_starting, soaio_idle, soaio_queued;
431static struct unrhdr *soaio_kproc_unr;
432
433static int soaio_max_procs = MAX_AIO_PROCS;
434SYSCTL_INT(_kern_ipc_aio, OID_AUTO, max_procs, CTLFLAG_RW, &soaio_max_procs, 0,
435    "Maximum number of kernel processes to use for async socket IO");
436
437static int soaio_num_procs;
438SYSCTL_INT(_kern_ipc_aio, OID_AUTO, num_procs, CTLFLAG_RD, &soaio_num_procs, 0,
439    "Number of active kernel processes for async socket IO");
440
441static int soaio_target_procs = TARGET_AIO_PROCS;
442SYSCTL_INT(_kern_ipc_aio, OID_AUTO, target_procs, CTLFLAG_RD,
443    &soaio_target_procs, 0,
444    "Preferred number of ready kernel processes for async socket IO");
445
446static int soaio_lifetime;
447SYSCTL_INT(_kern_ipc_aio, OID_AUTO, lifetime, CTLFLAG_RW, &soaio_lifetime, 0,
448    "Maximum lifetime for idle aiod");
449
450static void
451soaio_kproc_loop(void *arg)
452{
453	struct proc *p;
454	struct vmspace *myvm;
455	struct task *task;
456	int error, id, pending;
457
458	id = (intptr_t)arg;
459
460	/*
461	 * Grab an extra reference on the daemon's vmspace so that it
462	 * doesn't get freed by jobs that switch to a different
463	 * vmspace.
464	 */
465	p = curproc;
466	myvm = vmspace_acquire_ref(p);
467
468	mtx_lock(&soaio_jobs_lock);
469	MPASS(soaio_starting > 0);
470	soaio_starting--;
471	for (;;) {
472		while (!STAILQ_EMPTY(&soaio_jobs)) {
473			task = STAILQ_FIRST(&soaio_jobs);
474			STAILQ_REMOVE_HEAD(&soaio_jobs, ta_link);
475			soaio_queued--;
476			pending = task->ta_pending;
477			task->ta_pending = 0;
478			mtx_unlock(&soaio_jobs_lock);
479
480			task->ta_func(task->ta_context, pending);
481
482			mtx_lock(&soaio_jobs_lock);
483		}
484		MPASS(soaio_queued == 0);
485
486		if (p->p_vmspace != myvm) {
487			mtx_unlock(&soaio_jobs_lock);
488			vmspace_switch_aio(myvm);
489			mtx_lock(&soaio_jobs_lock);
490			continue;
491		}
492
493		soaio_idle++;
494		error = mtx_sleep(&soaio_idle, &soaio_jobs_lock, 0, "-",
495		    soaio_lifetime);
496		soaio_idle--;
497		if (error == EWOULDBLOCK && STAILQ_EMPTY(&soaio_jobs) &&
498		    soaio_num_procs > soaio_target_procs)
499			break;
500	}
501	soaio_num_procs--;
502	mtx_unlock(&soaio_jobs_lock);
503	free_unr(soaio_kproc_unr, id);
504	kproc_exit(0);
505}
506
507static void
508soaio_kproc_create(void *context, int pending)
509{
510	struct proc *p;
511	int error, id;
512
513	mtx_lock(&soaio_jobs_lock);
514	for (;;) {
515		if (soaio_num_procs < soaio_target_procs) {
516			/* Must create */
517		} else if (soaio_num_procs >= soaio_max_procs) {
518			/*
519			 * Hit the limit on kernel processes, don't
520			 * create another one.
521			 */
522			break;
523		} else if (soaio_queued <= soaio_idle + soaio_starting) {
524			/*
525			 * No more AIO jobs waiting for a process to be
526			 * created, so stop.
527			 */
528			break;
529		}
530		soaio_starting++;
531		mtx_unlock(&soaio_jobs_lock);
532
533		id = alloc_unr(soaio_kproc_unr);
534		error = kproc_create(soaio_kproc_loop, (void *)(intptr_t)id,
535		    &p, 0, 0, "soaiod%d", id);
536		if (error != 0) {
537			free_unr(soaio_kproc_unr, id);
538			mtx_lock(&soaio_jobs_lock);
539			soaio_starting--;
540			break;
541		}
542
543		mtx_lock(&soaio_jobs_lock);
544		soaio_num_procs++;
545	}
546	mtx_unlock(&soaio_jobs_lock);
547}
548
549void
550soaio_enqueue(struct task *task)
551{
552
553	mtx_lock(&soaio_jobs_lock);
554	MPASS(task->ta_pending == 0);
555	task->ta_pending++;
556	STAILQ_INSERT_TAIL(&soaio_jobs, task, ta_link);
557	soaio_queued++;
558	if (soaio_queued <= soaio_idle)
559		wakeup_one(&soaio_idle);
560	else if (soaio_num_procs < soaio_max_procs)
561		taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task);
562	mtx_unlock(&soaio_jobs_lock);
563}
564
565static void
566soaio_init(void)
567{
568
569	soaio_lifetime = AIOD_LIFETIME_DEFAULT;
570	STAILQ_INIT(&soaio_jobs);
571	mtx_init(&soaio_jobs_lock, "soaio jobs", NULL, MTX_DEF);
572	soaio_kproc_unr = new_unrhdr(1, INT_MAX, NULL);
573	TASK_INIT(&soaio_kproc_task, 0, soaio_kproc_create, NULL);
574}
575SYSINIT(soaio, SI_SUB_VFS, SI_ORDER_ANY, soaio_init, NULL);
576
577static __inline int
578soaio_ready(struct socket *so, struct sockbuf *sb)
579{
580	return (sb == &so->so_rcv ? soreadable(so) : sowriteable(so));
581}
582
583static void
584soaio_process_job(struct socket *so, sb_which which, struct kaiocb *job)
585{
586	struct ucred *td_savedcred;
587	struct thread *td;
588	struct sockbuf *sb = sobuf(so, which);
589#ifdef MAC
590	struct file *fp = job->fd_file;
591#endif
592	size_t cnt, done, job_total_nbytes __diagused;
593	long ru_before;
594	int error, flags;
595
596	SOCK_BUF_UNLOCK(so, which);
597	aio_switch_vmspace(job);
598	td = curthread;
599retry:
600	td_savedcred = td->td_ucred;
601	td->td_ucred = job->cred;
602
603	job_total_nbytes = job->uiop->uio_resid + job->aio_done;
604	done = job->aio_done;
605	cnt = job->uiop->uio_resid;
606	job->uiop->uio_offset = 0;
607	job->uiop->uio_td = td;
608	flags = MSG_NBIO;
609
610	/*
611	 * For resource usage accounting, only count a completed request
612	 * as a single message to avoid counting multiple calls to
613	 * sosend/soreceive on a blocking socket.
614	 */
615
616	if (sb == &so->so_rcv) {
617		ru_before = td->td_ru.ru_msgrcv;
618#ifdef MAC
619		error = mac_socket_check_receive(fp->f_cred, so);
620		if (error == 0)
621
622#endif
623			error = soreceive(so, NULL, job->uiop, NULL, NULL,
624			    &flags);
625		if (td->td_ru.ru_msgrcv != ru_before)
626			job->msgrcv = 1;
627	} else {
628		if (!TAILQ_EMPTY(&sb->sb_aiojobq))
629			flags |= MSG_MORETOCOME;
630		ru_before = td->td_ru.ru_msgsnd;
631#ifdef MAC
632		error = mac_socket_check_send(fp->f_cred, so);
633		if (error == 0)
634#endif
635			error = sousrsend(so, NULL, job->uiop, NULL, flags,
636			    job->userproc);
637		if (td->td_ru.ru_msgsnd != ru_before)
638			job->msgsnd = 1;
639	}
640
641	done += cnt - job->uiop->uio_resid;
642	job->aio_done = done;
643	td->td_ucred = td_savedcred;
644
645	if (error == EWOULDBLOCK) {
646		/*
647		 * The request was either partially completed or not
648		 * completed at all due to racing with a read() or
649		 * write() on the socket.  If the socket is
650		 * non-blocking, return with any partial completion.
651		 * If the socket is blocking or if no progress has
652		 * been made, requeue this request at the head of the
653		 * queue to try again when the socket is ready.
654		 */
655		MPASS(done != job_total_nbytes);
656		SOCK_BUF_LOCK(so, which);
657		if (done == 0 || !(so->so_state & SS_NBIO)) {
658			empty_results++;
659			if (soaio_ready(so, sb)) {
660				empty_retries++;
661				SOCK_BUF_UNLOCK(so, which);
662				goto retry;
663			}
664
665			if (!aio_set_cancel_function(job, soo_aio_cancel)) {
666				SOCK_BUF_UNLOCK(so, which);
667				if (done != 0)
668					aio_complete(job, done, 0);
669				else
670					aio_cancel(job);
671				SOCK_BUF_LOCK(so, which);
672			} else {
673				TAILQ_INSERT_HEAD(&sb->sb_aiojobq, job, list);
674			}
675			return;
676		}
677		SOCK_BUF_UNLOCK(so, which);
678	}
679	if (done != 0 && (error == ERESTART || error == EINTR ||
680	    error == EWOULDBLOCK))
681		error = 0;
682	if (error)
683		aio_complete(job, -1, error);
684	else
685		aio_complete(job, done, 0);
686	SOCK_BUF_LOCK(so, which);
687}
688
689static void
690soaio_process_sb(struct socket *so, sb_which which)
691{
692	struct kaiocb *job;
693	struct sockbuf *sb = sobuf(so, which);
694
695	CURVNET_SET(so->so_vnet);
696	SOCK_BUF_LOCK(so, which);
697	while (!TAILQ_EMPTY(&sb->sb_aiojobq) && soaio_ready(so, sb)) {
698		job = TAILQ_FIRST(&sb->sb_aiojobq);
699		TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
700		if (!aio_clear_cancel_function(job))
701			continue;
702
703		soaio_process_job(so, which, job);
704	}
705
706	/*
707	 * If there are still pending requests, the socket must not be
708	 * ready so set SB_AIO to request a wakeup when the socket
709	 * becomes ready.
710	 */
711	if (!TAILQ_EMPTY(&sb->sb_aiojobq))
712		sb->sb_flags |= SB_AIO;
713	sb->sb_flags &= ~SB_AIO_RUNNING;
714	SOCK_BUF_UNLOCK(so, which);
715
716	sorele(so);
717	CURVNET_RESTORE();
718}
719
720void
721soaio_rcv(void *context, int pending)
722{
723	struct socket *so;
724
725	so = context;
726	soaio_process_sb(so, SO_RCV);
727}
728
729void
730soaio_snd(void *context, int pending)
731{
732	struct socket *so;
733
734	so = context;
735	soaio_process_sb(so, SO_SND);
736}
737
738void
739sowakeup_aio(struct socket *so, sb_which which)
740{
741	struct sockbuf *sb = sobuf(so, which);
742
743	SOCK_BUF_LOCK_ASSERT(so, which);
744
745	sb->sb_flags &= ~SB_AIO;
746	if (sb->sb_flags & SB_AIO_RUNNING)
747		return;
748	sb->sb_flags |= SB_AIO_RUNNING;
749	soref(so);
750	soaio_enqueue(&sb->sb_aiotask);
751}
752
753static void
754soo_aio_cancel(struct kaiocb *job)
755{
756	struct socket *so;
757	struct sockbuf *sb;
758	long done;
759	int opcode;
760	sb_which which;
761
762	so = job->fd_file->f_data;
763	opcode = job->uaiocb.aio_lio_opcode;
764	if (opcode & LIO_READ) {
765		sb = &so->so_rcv;
766		which = SO_RCV;
767	} else {
768		MPASS(opcode & LIO_WRITE);
769		sb = &so->so_snd;
770		which = SO_SND;
771	}
772
773	SOCK_BUF_LOCK(so, which);
774	if (!aio_cancel_cleared(job))
775		TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
776	if (TAILQ_EMPTY(&sb->sb_aiojobq))
777		sb->sb_flags &= ~SB_AIO;
778	SOCK_BUF_UNLOCK(so, which);
779
780	done = job->aio_done;
781	if (done != 0)
782		aio_complete(job, done, 0);
783	else
784		aio_cancel(job);
785}
786
787static int
788soo_aio_queue(struct file *fp, struct kaiocb *job)
789{
790	struct socket *so;
791	struct sockbuf *sb;
792	sb_which which;
793	int error;
794
795	so = fp->f_data;
796	error = so->so_proto->pr_aio_queue(so, job);
797	if (error == 0)
798		return (0);
799
800	/* Lock through the socket, since this may be a listening socket. */
801	switch (job->uaiocb.aio_lio_opcode & (LIO_WRITE | LIO_READ)) {
802	case LIO_READ:
803		SOCK_RECVBUF_LOCK(so);
804		sb = &so->so_rcv;
805		which = SO_RCV;
806		break;
807	case LIO_WRITE:
808		SOCK_SENDBUF_LOCK(so);
809		sb = &so->so_snd;
810		which = SO_SND;
811		break;
812	default:
813		return (EINVAL);
814	}
815
816	if (SOLISTENING(so)) {
817		SOCK_BUF_UNLOCK(so, which);
818		return (EINVAL);
819	}
820
821	if (!aio_set_cancel_function(job, soo_aio_cancel))
822		panic("new job was cancelled");
823	TAILQ_INSERT_TAIL(&sb->sb_aiojobq, job, list);
824	if (!(sb->sb_flags & SB_AIO_RUNNING)) {
825		if (soaio_ready(so, sb))
826			sowakeup_aio(so, which);
827		else
828			sb->sb_flags |= SB_AIO;
829	}
830	SOCK_BUF_UNLOCK(so, which);
831	return (0);
832}
833