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
80static SYSCTL_NODE(_kern_ipc, OID_AUTO, aio, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
81    "socket AIO stats");
82
83static int empty_results;
84SYSCTL_INT(_kern_ipc_aio, OID_AUTO, empty_results, CTLFLAG_RD, &empty_results,
85    0, "socket operation returned EAGAIN");
86
87static int empty_retries;
88SYSCTL_INT(_kern_ipc_aio, OID_AUTO, empty_retries, CTLFLAG_RD, &empty_retries,
89    0, "socket operation retries");
90
91static fo_rdwr_t soo_read;
92static fo_rdwr_t soo_write;
93static fo_ioctl_t soo_ioctl;
94static fo_poll_t soo_poll;
95extern fo_kqfilter_t soo_kqfilter;
96static fo_stat_t soo_stat;
97static fo_close_t soo_close;
98static fo_fill_kinfo_t soo_fill_kinfo;
99static fo_aio_queue_t soo_aio_queue;
100
101static void	soo_aio_cancel(struct kaiocb *job);
102
103struct 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
120static int
121soo_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
136static int
137soo_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
157static int
158soo_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
270static int
271soo_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
285static int
286soo_stat(struct file *fp, struct stat *ub, struct ucred *active_cred,
287    struct thread *td)
288{
289	struct socket *so = fp->f_data;
290	int error;
291
292	bzero((caddr_t)ub, sizeof (*ub));
293	ub->st_mode = S_IFSOCK;
294#ifdef MAC
295	error = mac_socket_check_stat(active_cred, so);
296	if (error)
297		return (error);
298#endif
299	SOCK_LOCK(so);
300	if (!SOLISTENING(so)) {
301		struct sockbuf *sb;
302
303		/*
304		 * If SBS_CANTRCVMORE is set, but there's still data left
305		 * in the receive buffer, the socket is still readable.
306		 */
307		sb = &so->so_rcv;
308		SOCKBUF_LOCK(sb);
309		if ((sb->sb_state & SBS_CANTRCVMORE) == 0 || sbavail(sb))
310			ub->st_mode |= S_IRUSR | S_IRGRP | S_IROTH;
311		ub->st_size = sbavail(sb) - sb->sb_ctl;
312		SOCKBUF_UNLOCK(sb);
313
314		sb = &so->so_snd;
315		SOCKBUF_LOCK(sb);
316		if ((sb->sb_state & SBS_CANTSENDMORE) == 0)
317			ub->st_mode |= S_IWUSR | S_IWGRP | S_IWOTH;
318		SOCKBUF_UNLOCK(sb);
319	}
320	ub->st_uid = so->so_cred->cr_uid;
321	ub->st_gid = so->so_cred->cr_gid;
322	error = so->so_proto->pr_usrreqs->pru_sense(so, ub);
323	SOCK_UNLOCK(so);
324	return (error);
325}
326
327/*
328 * API socket close on file pointer.  We call soclose() to close the socket
329 * (including initiating closing protocols).  soclose() will sorele() the
330 * file reference but the actual socket will not go away until the socket's
331 * ref count hits 0.
332 */
333static int
334soo_close(struct file *fp, struct thread *td)
335{
336	int error = 0;
337	struct socket *so;
338
339	so = fp->f_data;
340	fp->f_ops = &badfileops;
341	fp->f_data = NULL;
342
343	if (so)
344		error = soclose(so);
345	return (error);
346}
347
348static int
349soo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
350{
351	struct sockaddr *sa;
352	struct inpcb *inpcb;
353	struct unpcb *unpcb;
354	struct socket *so;
355	int error;
356
357	kif->kf_type = KF_TYPE_SOCKET;
358	so = fp->f_data;
359	CURVNET_SET(so->so_vnet);
360	kif->kf_un.kf_sock.kf_sock_domain0 =
361	    so->so_proto->pr_domain->dom_family;
362	kif->kf_un.kf_sock.kf_sock_type0 = so->so_type;
363	kif->kf_un.kf_sock.kf_sock_protocol0 = so->so_proto->pr_protocol;
364	kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb;
365	switch (kif->kf_un.kf_sock.kf_sock_domain0) {
366	case AF_INET:
367	case AF_INET6:
368		if (kif->kf_un.kf_sock.kf_sock_protocol0 == IPPROTO_TCP) {
369			if (so->so_pcb != NULL) {
370				inpcb = (struct inpcb *)(so->so_pcb);
371				kif->kf_un.kf_sock.kf_sock_inpcb =
372				    (uintptr_t)inpcb->inp_ppcb;
373				kif->kf_un.kf_sock.kf_sock_sendq =
374				    sbused(&so->so_snd);
375				kif->kf_un.kf_sock.kf_sock_recvq =
376				    sbused(&so->so_rcv);
377			}
378		}
379		break;
380	case AF_UNIX:
381		if (so->so_pcb != NULL) {
382			unpcb = (struct unpcb *)(so->so_pcb);
383			if (unpcb->unp_conn) {
384				kif->kf_un.kf_sock.kf_sock_unpconn =
385				    (uintptr_t)unpcb->unp_conn;
386				kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
387				    so->so_rcv.sb_state;
388				kif->kf_un.kf_sock.kf_sock_snd_sb_state =
389				    so->so_snd.sb_state;
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	}
398	error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa);
399	if (error == 0 &&
400	    sa->sa_len <= sizeof(kif->kf_un.kf_sock.kf_sa_local)) {
401		bcopy(sa, &kif->kf_un.kf_sock.kf_sa_local, sa->sa_len);
402		free(sa, M_SONAME);
403	}
404	error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa);
405	if (error == 0 &&
406	    sa->sa_len <= sizeof(kif->kf_un.kf_sock.kf_sa_peer)) {
407		bcopy(sa, &kif->kf_un.kf_sock.kf_sa_peer, sa->sa_len);
408		free(sa, M_SONAME);
409	}
410	strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name,
411	    sizeof(kif->kf_path));
412	CURVNET_RESTORE();
413	return (0);
414}
415
416/*
417 * Use the 'backend3' field in AIO jobs to store the amount of data
418 * completed by the AIO job so far.
419 */
420#define	aio_done	backend3
421
422static STAILQ_HEAD(, task) soaio_jobs;
423static struct mtx soaio_jobs_lock;
424static struct task soaio_kproc_task;
425static int soaio_starting, soaio_idle, soaio_queued;
426static struct unrhdr *soaio_kproc_unr;
427
428static int soaio_max_procs = MAX_AIO_PROCS;
429SYSCTL_INT(_kern_ipc_aio, OID_AUTO, max_procs, CTLFLAG_RW, &soaio_max_procs, 0,
430    "Maximum number of kernel processes to use for async socket IO");
431
432static int soaio_num_procs;
433SYSCTL_INT(_kern_ipc_aio, OID_AUTO, num_procs, CTLFLAG_RD, &soaio_num_procs, 0,
434    "Number of active kernel processes for async socket IO");
435
436static int soaio_target_procs = TARGET_AIO_PROCS;
437SYSCTL_INT(_kern_ipc_aio, OID_AUTO, target_procs, CTLFLAG_RD,
438    &soaio_target_procs, 0,
439    "Preferred number of ready kernel processes for async socket IO");
440
441static int soaio_lifetime;
442SYSCTL_INT(_kern_ipc_aio, OID_AUTO, lifetime, CTLFLAG_RW, &soaio_lifetime, 0,
443    "Maximum lifetime for idle aiod");
444
445static void
446soaio_kproc_loop(void *arg)
447{
448	struct proc *p;
449	struct vmspace *myvm;
450	struct task *task;
451	int error, id, pending;
452
453	id = (intptr_t)arg;
454
455	/*
456	 * Grab an extra reference on the daemon's vmspace so that it
457	 * doesn't get freed by jobs that switch to a different
458	 * vmspace.
459	 */
460	p = curproc;
461	myvm = vmspace_acquire_ref(p);
462
463	mtx_lock(&soaio_jobs_lock);
464	MPASS(soaio_starting > 0);
465	soaio_starting--;
466	for (;;) {
467		while (!STAILQ_EMPTY(&soaio_jobs)) {
468			task = STAILQ_FIRST(&soaio_jobs);
469			STAILQ_REMOVE_HEAD(&soaio_jobs, ta_link);
470			soaio_queued--;
471			pending = task->ta_pending;
472			task->ta_pending = 0;
473			mtx_unlock(&soaio_jobs_lock);
474
475			task->ta_func(task->ta_context, pending);
476
477			mtx_lock(&soaio_jobs_lock);
478		}
479		MPASS(soaio_queued == 0);
480
481		if (p->p_vmspace != myvm) {
482			mtx_unlock(&soaio_jobs_lock);
483			vmspace_switch_aio(myvm);
484			mtx_lock(&soaio_jobs_lock);
485			continue;
486		}
487
488		soaio_idle++;
489		error = mtx_sleep(&soaio_idle, &soaio_jobs_lock, 0, "-",
490		    soaio_lifetime);
491		soaio_idle--;
492		if (error == EWOULDBLOCK && STAILQ_EMPTY(&soaio_jobs) &&
493		    soaio_num_procs > soaio_target_procs)
494			break;
495	}
496	soaio_num_procs--;
497	mtx_unlock(&soaio_jobs_lock);
498	free_unr(soaio_kproc_unr, id);
499	kproc_exit(0);
500}
501
502static void
503soaio_kproc_create(void *context, int pending)
504{
505	struct proc *p;
506	int error, id;
507
508	mtx_lock(&soaio_jobs_lock);
509	for (;;) {
510		if (soaio_num_procs < soaio_target_procs) {
511			/* Must create */
512		} else if (soaio_num_procs >= soaio_max_procs) {
513			/*
514			 * Hit the limit on kernel processes, don't
515			 * create another one.
516			 */
517			break;
518		} else if (soaio_queued <= soaio_idle + soaio_starting) {
519			/*
520			 * No more AIO jobs waiting for a process to be
521			 * created, so stop.
522			 */
523			break;
524		}
525		soaio_starting++;
526		mtx_unlock(&soaio_jobs_lock);
527
528		id = alloc_unr(soaio_kproc_unr);
529		error = kproc_create(soaio_kproc_loop, (void *)(intptr_t)id,
530		    &p, 0, 0, "soaiod%d", id);
531		if (error != 0) {
532			free_unr(soaio_kproc_unr, id);
533			mtx_lock(&soaio_jobs_lock);
534			soaio_starting--;
535			break;
536		}
537
538		mtx_lock(&soaio_jobs_lock);
539		soaio_num_procs++;
540	}
541	mtx_unlock(&soaio_jobs_lock);
542}
543
544void
545soaio_enqueue(struct task *task)
546{
547
548	mtx_lock(&soaio_jobs_lock);
549	MPASS(task->ta_pending == 0);
550	task->ta_pending++;
551	STAILQ_INSERT_TAIL(&soaio_jobs, task, ta_link);
552	soaio_queued++;
553	if (soaio_queued <= soaio_idle)
554		wakeup_one(&soaio_idle);
555	else if (soaio_num_procs < soaio_max_procs)
556		taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task);
557	mtx_unlock(&soaio_jobs_lock);
558}
559
560static void
561soaio_init(void)
562{
563
564	soaio_lifetime = AIOD_LIFETIME_DEFAULT;
565	STAILQ_INIT(&soaio_jobs);
566	mtx_init(&soaio_jobs_lock, "soaio jobs", NULL, MTX_DEF);
567	soaio_kproc_unr = new_unrhdr(1, INT_MAX, NULL);
568	TASK_INIT(&soaio_kproc_task, 0, soaio_kproc_create, NULL);
569	if (soaio_target_procs > 0)
570		taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task);
571}
572SYSINIT(soaio, SI_SUB_VFS, SI_ORDER_ANY, soaio_init, NULL);
573
574static __inline int
575soaio_ready(struct socket *so, struct sockbuf *sb)
576{
577	return (sb == &so->so_rcv ? soreadable(so) : sowriteable(so));
578}
579
580static void
581soaio_process_job(struct socket *so, struct sockbuf *sb, struct kaiocb *job)
582{
583	struct ucred *td_savedcred;
584	struct thread *td;
585	struct file *fp;
586	struct uio uio;
587	struct iovec iov;
588	size_t cnt, done;
589	long ru_before;
590	int error, flags;
591
592	SOCKBUF_UNLOCK(sb);
593	aio_switch_vmspace(job);
594	td = curthread;
595	fp = job->fd_file;
596retry:
597	td_savedcred = td->td_ucred;
598	td->td_ucred = job->cred;
599
600	done = job->aio_done;
601	cnt = job->uaiocb.aio_nbytes - done;
602	iov.iov_base = (void *)((uintptr_t)job->uaiocb.aio_buf + done);
603	iov.iov_len = cnt;
604	uio.uio_iov = &iov;
605	uio.uio_iovcnt = 1;
606	uio.uio_offset = 0;
607	uio.uio_resid = cnt;
608	uio.uio_segflg = UIO_USERSPACE;
609	uio.uio_td = td;
610	flags = MSG_NBIO;
611
612	/*
613	 * For resource usage accounting, only count a completed request
614	 * as a single message to avoid counting multiple calls to
615	 * sosend/soreceive on a blocking socket.
616	 */
617
618	if (sb == &so->so_rcv) {
619		uio.uio_rw = UIO_READ;
620		ru_before = td->td_ru.ru_msgrcv;
621#ifdef MAC
622		error = mac_socket_check_receive(fp->f_cred, so);
623		if (error == 0)
624
625#endif
626			error = soreceive(so, NULL, &uio, NULL, NULL, &flags);
627		if (td->td_ru.ru_msgrcv != ru_before)
628			job->msgrcv = 1;
629	} else {
630		if (!TAILQ_EMPTY(&sb->sb_aiojobq))
631			flags |= MSG_MORETOCOME;
632		uio.uio_rw = UIO_WRITE;
633		ru_before = td->td_ru.ru_msgsnd;
634#ifdef MAC
635		error = mac_socket_check_send(fp->f_cred, so);
636		if (error == 0)
637#endif
638			error = sosend(so, NULL, &uio, NULL, NULL, flags, td);
639		if (td->td_ru.ru_msgsnd != ru_before)
640			job->msgsnd = 1;
641		if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0) {
642			PROC_LOCK(job->userproc);
643			kern_psignal(job->userproc, SIGPIPE);
644			PROC_UNLOCK(job->userproc);
645		}
646	}
647
648	done += cnt - uio.uio_resid;
649	job->aio_done = done;
650	td->td_ucred = td_savedcred;
651
652	if (error == EWOULDBLOCK) {
653		/*
654		 * The request was either partially completed or not
655		 * completed at all due to racing with a read() or
656		 * write() on the socket.  If the socket is
657		 * non-blocking, return with any partial completion.
658		 * If the socket is blocking or if no progress has
659		 * been made, requeue this request at the head of the
660		 * queue to try again when the socket is ready.
661		 */
662		MPASS(done != job->uaiocb.aio_nbytes);
663		SOCKBUF_LOCK(sb);
664		if (done == 0 || !(so->so_state & SS_NBIO)) {
665			empty_results++;
666			if (soaio_ready(so, sb)) {
667				empty_retries++;
668				SOCKBUF_UNLOCK(sb);
669				goto retry;
670			}
671
672			if (!aio_set_cancel_function(job, soo_aio_cancel)) {
673				SOCKBUF_UNLOCK(sb);
674				if (done != 0)
675					aio_complete(job, done, 0);
676				else
677					aio_cancel(job);
678				SOCKBUF_LOCK(sb);
679			} else {
680				TAILQ_INSERT_HEAD(&sb->sb_aiojobq, job, list);
681			}
682			return;
683		}
684		SOCKBUF_UNLOCK(sb);
685	}
686	if (done != 0 && (error == ERESTART || error == EINTR ||
687	    error == EWOULDBLOCK))
688		error = 0;
689	if (error)
690		aio_complete(job, -1, error);
691	else
692		aio_complete(job, done, 0);
693	SOCKBUF_LOCK(sb);
694}
695
696static void
697soaio_process_sb(struct socket *so, struct sockbuf *sb)
698{
699	struct kaiocb *job;
700
701	CURVNET_SET(so->so_vnet);
702	SOCKBUF_LOCK(sb);
703	while (!TAILQ_EMPTY(&sb->sb_aiojobq) && soaio_ready(so, sb)) {
704		job = TAILQ_FIRST(&sb->sb_aiojobq);
705		TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
706		if (!aio_clear_cancel_function(job))
707			continue;
708
709		soaio_process_job(so, sb, job);
710	}
711
712	/*
713	 * If there are still pending requests, the socket must not be
714	 * ready so set SB_AIO to request a wakeup when the socket
715	 * becomes ready.
716	 */
717	if (!TAILQ_EMPTY(&sb->sb_aiojobq))
718		sb->sb_flags |= SB_AIO;
719	sb->sb_flags &= ~SB_AIO_RUNNING;
720	SOCKBUF_UNLOCK(sb);
721
722	SOCK_LOCK(so);
723	sorele(so);
724	CURVNET_RESTORE();
725}
726
727void
728soaio_rcv(void *context, int pending)
729{
730	struct socket *so;
731
732	so = context;
733	soaio_process_sb(so, &so->so_rcv);
734}
735
736void
737soaio_snd(void *context, int pending)
738{
739	struct socket *so;
740
741	so = context;
742	soaio_process_sb(so, &so->so_snd);
743}
744
745void
746sowakeup_aio(struct socket *so, struct sockbuf *sb)
747{
748
749	SOCKBUF_LOCK_ASSERT(sb);
750	sb->sb_flags &= ~SB_AIO;
751	if (sb->sb_flags & SB_AIO_RUNNING)
752		return;
753	sb->sb_flags |= SB_AIO_RUNNING;
754	soref(so);
755	soaio_enqueue(&sb->sb_aiotask);
756}
757
758static void
759soo_aio_cancel(struct kaiocb *job)
760{
761	struct socket *so;
762	struct sockbuf *sb;
763	long done;
764	int opcode;
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	else {
771		MPASS(opcode == LIO_WRITE);
772		sb = &so->so_snd;
773	}
774
775	SOCKBUF_LOCK(sb);
776	if (!aio_cancel_cleared(job))
777		TAILQ_REMOVE(&sb->sb_aiojobq, job, list);
778	if (TAILQ_EMPTY(&sb->sb_aiojobq))
779		sb->sb_flags &= ~SB_AIO;
780	SOCKBUF_UNLOCK(sb);
781
782	done = job->aio_done;
783	if (done != 0)
784		aio_complete(job, done, 0);
785	else
786		aio_cancel(job);
787}
788
789static int
790soo_aio_queue(struct file *fp, struct kaiocb *job)
791{
792	struct socket *so;
793	struct sockbuf *sb;
794	int error;
795
796	so = fp->f_data;
797	error = (*so->so_proto->pr_usrreqs->pru_aio_queue)(so, job);
798	if (error == 0)
799		return (0);
800
801	switch (job->uaiocb.aio_lio_opcode) {
802	case LIO_READ:
803		sb = &so->so_rcv;
804		break;
805	case LIO_WRITE:
806		sb = &so->so_snd;
807		break;
808	default:
809		return (EINVAL);
810	}
811
812	SOCKBUF_LOCK(sb);
813	if (!aio_set_cancel_function(job, soo_aio_cancel))
814		panic("new job was cancelled");
815	TAILQ_INSERT_TAIL(&sb->sb_aiojobq, job, list);
816	if (!(sb->sb_flags & SB_AIO_RUNNING)) {
817		if (soaio_ready(so, sb))
818			sowakeup_aio(so, sb);
819		else
820			sb->sb_flags |= SB_AIO;
821	}
822	SOCKBUF_UNLOCK(sb);
823	return (0);
824}
825