xref: /openbsd-src/sys/kern/uipc_socket.c (revision 7c0ec4b8992567abb1e1536622dc789a9a39d9f1)
1 /*	$OpenBSD: uipc_socket.c,v 1.343 2024/08/11 00:19:00 jsg Exp $	*/
2 /*	$NetBSD: uipc_socket.c,v 1.21 1996/02/04 02:17:52 christos Exp $	*/
3 
4 /*
5  * Copyright (c) 1982, 1986, 1988, 1990, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	@(#)uipc_socket.c	8.3 (Berkeley) 4/15/94
33  */
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/proc.h>
38 #include <sys/file.h>
39 #include <sys/filedesc.h>
40 #include <sys/malloc.h>
41 #include <sys/mbuf.h>
42 #include <sys/domain.h>
43 #include <sys/event.h>
44 #include <sys/protosw.h>
45 #include <sys/socket.h>
46 #include <sys/unpcb.h>
47 #include <sys/socketvar.h>
48 #include <sys/signalvar.h>
49 #include <sys/pool.h>
50 #include <sys/atomic.h>
51 #include <sys/rwlock.h>
52 #include <sys/time.h>
53 #include <sys/refcnt.h>
54 
55 #ifdef DDB
56 #include <machine/db_machdep.h>
57 #endif
58 
59 void	sbsync(struct sockbuf *, struct mbuf *);
60 
61 int	sosplice(struct socket *, int, off_t, struct timeval *);
62 void	sounsplice(struct socket *, struct socket *, int);
63 void	soidle(void *);
64 void	sotask(void *);
65 void	soreaper(void *);
66 void	soput(void *);
67 int	somove(struct socket *, int);
68 void	sorflush(struct socket *);
69 
70 void	filt_sordetach(struct knote *kn);
71 int	filt_soread(struct knote *kn, long hint);
72 void	filt_sowdetach(struct knote *kn);
73 int	filt_sowrite(struct knote *kn, long hint);
74 int	filt_soexcept(struct knote *kn, long hint);
75 
76 int	filt_sowmodify(struct kevent *kev, struct knote *kn);
77 int	filt_sowprocess(struct knote *kn, struct kevent *kev);
78 
79 int	filt_sormodify(struct kevent *kev, struct knote *kn);
80 int	filt_sorprocess(struct knote *kn, struct kevent *kev);
81 
82 const struct filterops soread_filtops = {
83 	.f_flags	= FILTEROP_ISFD | FILTEROP_MPSAFE,
84 	.f_attach	= NULL,
85 	.f_detach	= filt_sordetach,
86 	.f_event	= filt_soread,
87 	.f_modify	= filt_sormodify,
88 	.f_process	= filt_sorprocess,
89 };
90 
91 const struct filterops sowrite_filtops = {
92 	.f_flags	= FILTEROP_ISFD | FILTEROP_MPSAFE,
93 	.f_attach	= NULL,
94 	.f_detach	= filt_sowdetach,
95 	.f_event	= filt_sowrite,
96 	.f_modify	= filt_sowmodify,
97 	.f_process	= filt_sowprocess,
98 };
99 
100 const struct filterops soexcept_filtops = {
101 	.f_flags	= FILTEROP_ISFD | FILTEROP_MPSAFE,
102 	.f_attach	= NULL,
103 	.f_detach	= filt_sordetach,
104 	.f_event	= filt_soexcept,
105 	.f_modify	= filt_sormodify,
106 	.f_process	= filt_sorprocess,
107 };
108 
109 #ifndef SOMINCONN
110 #define SOMINCONN 80
111 #endif /* SOMINCONN */
112 
113 int	somaxconn = SOMAXCONN;
114 int	sominconn = SOMINCONN;
115 
116 struct pool socket_pool;
117 #ifdef SOCKET_SPLICE
118 struct pool sosplice_pool;
119 struct taskq *sosplice_taskq;
120 struct rwlock sosplice_lock = RWLOCK_INITIALIZER("sosplicelk");
121 #endif
122 
123 void
124 soinit(void)
125 {
126 	pool_init(&socket_pool, sizeof(struct socket), 0, IPL_SOFTNET, 0,
127 	    "sockpl", NULL);
128 #ifdef SOCKET_SPLICE
129 	pool_init(&sosplice_pool, sizeof(struct sosplice), 0, IPL_SOFTNET, 0,
130 	    "sosppl", NULL);
131 #endif
132 }
133 
134 struct socket *
135 soalloc(const struct protosw *prp, int wait)
136 {
137 	const struct domain *dp = prp->pr_domain;
138 	struct socket *so;
139 
140 	so = pool_get(&socket_pool, (wait == M_WAIT ? PR_WAITOK : PR_NOWAIT) |
141 	    PR_ZERO);
142 	if (so == NULL)
143 		return (NULL);
144 	rw_init_flags(&so->so_lock, dp->dom_name, RWL_DUPOK);
145 	refcnt_init(&so->so_refcnt);
146 	rw_init(&so->so_rcv.sb_lock, "sbufrcv");
147 	rw_init(&so->so_snd.sb_lock, "sbufsnd");
148 	mtx_init_flags(&so->so_rcv.sb_mtx, IPL_MPFLOOR, "sbrcv", 0);
149 	mtx_init_flags(&so->so_snd.sb_mtx, IPL_MPFLOOR, "sbsnd", 0);
150 	klist_init_mutex(&so->so_rcv.sb_klist, &so->so_rcv.sb_mtx);
151 	klist_init_mutex(&so->so_snd.sb_klist, &so->so_snd.sb_mtx);
152 	sigio_init(&so->so_sigio);
153 	TAILQ_INIT(&so->so_q0);
154 	TAILQ_INIT(&so->so_q);
155 
156 	switch (dp->dom_family) {
157 	case AF_INET:
158 	case AF_INET6:
159 		switch (prp->pr_type) {
160 		case SOCK_RAW:
161 		case SOCK_DGRAM:
162 			so->so_snd.sb_flags |= SB_MTXLOCK;
163 			so->so_rcv.sb_flags |= SB_MTXLOCK;
164 			break;
165 		}
166 		break;
167 	case AF_KEY:
168 	case AF_ROUTE:
169 	case AF_UNIX:
170 		so->so_snd.sb_flags |= SB_MTXLOCK;
171 		so->so_rcv.sb_flags |= SB_MTXLOCK;
172 		break;
173 	}
174 
175 	return (so);
176 }
177 
178 /*
179  * Socket operation routines.
180  * These routines are called by the routines in
181  * sys_socket.c or from a system process, and
182  * implement the semantics of socket operations by
183  * switching out to the protocol specific routines.
184  */
185 int
186 socreate(int dom, struct socket **aso, int type, int proto)
187 {
188 	struct proc *p = curproc;		/* XXX */
189 	const struct protosw *prp;
190 	struct socket *so;
191 	int error;
192 
193 	if (proto)
194 		prp = pffindproto(dom, proto, type);
195 	else
196 		prp = pffindtype(dom, type);
197 	if (prp == NULL || prp->pr_usrreqs == NULL)
198 		return (EPROTONOSUPPORT);
199 	if (prp->pr_type != type)
200 		return (EPROTOTYPE);
201 	so = soalloc(prp, M_WAIT);
202 	so->so_type = type;
203 	if (suser(p) == 0)
204 		so->so_state = SS_PRIV;
205 	so->so_ruid = p->p_ucred->cr_ruid;
206 	so->so_euid = p->p_ucred->cr_uid;
207 	so->so_rgid = p->p_ucred->cr_rgid;
208 	so->so_egid = p->p_ucred->cr_gid;
209 	so->so_cpid = p->p_p->ps_pid;
210 	so->so_proto = prp;
211 	so->so_snd.sb_timeo_nsecs = INFSLP;
212 	so->so_rcv.sb_timeo_nsecs = INFSLP;
213 
214 	solock(so);
215 	error = pru_attach(so, proto, M_WAIT);
216 	if (error) {
217 		so->so_state |= SS_NOFDREF;
218 		/* sofree() calls sounlock(). */
219 		sofree(so, 0);
220 		return (error);
221 	}
222 	sounlock(so);
223 	*aso = so;
224 	return (0);
225 }
226 
227 int
228 sobind(struct socket *so, struct mbuf *nam, struct proc *p)
229 {
230 	soassertlocked(so);
231 	return pru_bind(so, nam, p);
232 }
233 
234 int
235 solisten(struct socket *so, int backlog)
236 {
237 	int somaxconn_local = atomic_load_int(&somaxconn);
238 	int sominconn_local = atomic_load_int(&sominconn);
239 	int error;
240 
241 	switch (so->so_type) {
242 	case SOCK_STREAM:
243 	case SOCK_SEQPACKET:
244 		break;
245 	default:
246 		return (EOPNOTSUPP);
247 	}
248 
249 	soassertlocked(so);
250 
251 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING|SS_ISDISCONNECTING))
252 		return (EINVAL);
253 #ifdef SOCKET_SPLICE
254 	if (isspliced(so) || issplicedback(so))
255 		return (EOPNOTSUPP);
256 #endif /* SOCKET_SPLICE */
257 	error = pru_listen(so);
258 	if (error)
259 		return (error);
260 	if (TAILQ_FIRST(&so->so_q) == NULL)
261 		so->so_options |= SO_ACCEPTCONN;
262 	if (backlog < 0 || backlog > somaxconn_local)
263 		backlog = somaxconn_local;
264 	if (backlog < sominconn_local)
265 		backlog = sominconn_local;
266 	so->so_qlimit = backlog;
267 	return (0);
268 }
269 
270 #define SOSP_FREEING_READ	1
271 #define SOSP_FREEING_WRITE	2
272 void
273 sofree(struct socket *so, int keep_lock)
274 {
275 	int persocket = solock_persocket(so);
276 
277 	soassertlocked(so);
278 
279 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
280 		if (!keep_lock)
281 			sounlock(so);
282 		return;
283 	}
284 	if (so->so_head) {
285 		struct socket *head = so->so_head;
286 
287 		/*
288 		 * We must not decommission a socket that's on the accept(2)
289 		 * queue.  If we do, then accept(2) may hang after select(2)
290 		 * indicated that the listening socket was ready.
291 		 */
292 		if (so->so_onq == &head->so_q) {
293 			if (!keep_lock)
294 				sounlock(so);
295 			return;
296 		}
297 
298 		if (persocket) {
299 			/*
300 			 * Concurrent close of `head' could
301 			 * abort `so' due to re-lock.
302 			 */
303 			soref(so);
304 			soref(head);
305 			sounlock(so);
306 			solock(head);
307 			solock(so);
308 
309 			if (so->so_onq != &head->so_q0) {
310 				sounlock(head);
311 				sounlock(so);
312 				sorele(head);
313 				sorele(so);
314 				return;
315 			}
316 
317 			sorele(head);
318 			sorele(so);
319 		}
320 
321 		soqremque(so, 0);
322 
323 		if (persocket)
324 			sounlock(head);
325 	}
326 
327 	switch (so->so_proto->pr_domain->dom_family) {
328 	case AF_INET:
329 	case AF_INET6:
330 		if (so->so_proto->pr_type == SOCK_STREAM)
331 			break;
332 		/* FALLTHROUGH */
333 	default:
334 		sounlock(so);
335 		refcnt_finalize(&so->so_refcnt, "sofinal");
336 		solock(so);
337 		break;
338 	}
339 
340 	sigio_free(&so->so_sigio);
341 	klist_free(&so->so_rcv.sb_klist);
342 	klist_free(&so->so_snd.sb_klist);
343 
344 	mtx_enter(&so->so_snd.sb_mtx);
345 	sbrelease(so, &so->so_snd);
346 	mtx_leave(&so->so_snd.sb_mtx);
347 
348 	/*
349 	 * Unlocked dispose and cleanup is safe. Socket is unlinked
350 	 * from everywhere. Even concurrent sotask() thread will not
351 	 * call somove().
352 	 */
353 	if (so->so_proto->pr_flags & PR_RIGHTS &&
354 	    so->so_proto->pr_domain->dom_dispose)
355 		(*so->so_proto->pr_domain->dom_dispose)(so->so_rcv.sb_mb);
356 	m_purge(so->so_rcv.sb_mb);
357 
358 	if (!keep_lock)
359 		sounlock(so);
360 
361 #ifdef SOCKET_SPLICE
362 	if (so->so_sp) {
363 		/* Reuse splice idle, sounsplice() has been called before. */
364 		timeout_set_flags(&so->so_sp->ssp_idleto, soreaper, so,
365 		    KCLOCK_NONE, TIMEOUT_PROC | TIMEOUT_MPSAFE);
366 		timeout_add(&so->so_sp->ssp_idleto, 0);
367 	} else
368 #endif /* SOCKET_SPLICE */
369 	{
370 		pool_put(&socket_pool, so);
371 	}
372 }
373 
374 static inline uint64_t
375 solinger_nsec(struct socket *so)
376 {
377 	if (so->so_linger == 0)
378 		return INFSLP;
379 
380 	return SEC_TO_NSEC(so->so_linger);
381 }
382 
383 /*
384  * Close a socket on last file table reference removal.
385  * Initiate disconnect if connected.
386  * Free socket when disconnect complete.
387  */
388 int
389 soclose(struct socket *so, int flags)
390 {
391 	struct socket *so2;
392 	int error = 0;
393 
394 	solock(so);
395 	/* Revoke async IO early. There is a final revocation in sofree(). */
396 	sigio_free(&so->so_sigio);
397 	if (so->so_state & SS_ISCONNECTED) {
398 		if (so->so_pcb == NULL)
399 			goto discard;
400 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
401 			error = sodisconnect(so);
402 			if (error)
403 				goto drop;
404 		}
405 		if (so->so_options & SO_LINGER) {
406 			if ((so->so_state & SS_ISDISCONNECTING) &&
407 			    (flags & MSG_DONTWAIT))
408 				goto drop;
409 			while (so->so_state & SS_ISCONNECTED) {
410 				error = sosleep_nsec(so, &so->so_timeo,
411 				    PSOCK | PCATCH, "netcls",
412 				    solinger_nsec(so));
413 				if (error)
414 					break;
415 			}
416 		}
417 	}
418 drop:
419 	if (so->so_pcb) {
420 		int error2;
421 		error2 = pru_detach(so);
422 		if (error == 0)
423 			error = error2;
424 	}
425 	if (so->so_options & SO_ACCEPTCONN) {
426 		int persocket = solock_persocket(so);
427 
428 		while ((so2 = TAILQ_FIRST(&so->so_q0)) != NULL) {
429 			if (persocket)
430 				solock(so2);
431 			(void) soqremque(so2, 0);
432 			if (persocket)
433 				sounlock(so);
434 			soabort(so2);
435 			if (persocket)
436 				solock(so);
437 		}
438 		while ((so2 = TAILQ_FIRST(&so->so_q)) != NULL) {
439 			if (persocket)
440 				solock(so2);
441 			(void) soqremque(so2, 1);
442 			if (persocket)
443 				sounlock(so);
444 			soabort(so2);
445 			if (persocket)
446 				solock(so);
447 		}
448 	}
449 discard:
450 	if (so->so_state & SS_NOFDREF)
451 		panic("soclose NOFDREF: so %p, so_type %d", so, so->so_type);
452 	so->so_state |= SS_NOFDREF;
453 
454 #ifdef SOCKET_SPLICE
455 	if (so->so_sp) {
456 		struct socket *soback;
457 
458 		if (so->so_proto->pr_flags & PR_WANTRCVD) {
459 			/*
460 			 * Copy - Paste, but can't relock and sleep in
461 			 * sofree() in tcp(4) case. That's why tcp(4)
462 			 * still rely on solock() for splicing and
463 			 * unsplicing.
464 			 */
465 
466 			if (issplicedback(so)) {
467 				int freeing = SOSP_FREEING_WRITE;
468 
469 				if (so->so_sp->ssp_soback == so)
470 					freeing |= SOSP_FREEING_READ;
471 				sounsplice(so->so_sp->ssp_soback, so, freeing);
472 			}
473 			if (isspliced(so)) {
474 				int freeing = SOSP_FREEING_READ;
475 
476 				if (so == so->so_sp->ssp_socket)
477 					freeing |= SOSP_FREEING_WRITE;
478 				sounsplice(so, so->so_sp->ssp_socket, freeing);
479 			}
480 			goto free;
481 		}
482 
483 		sounlock(so);
484 		mtx_enter(&so->so_snd.sb_mtx);
485 		/*
486 		 * Concurrent sounsplice() locks `sb_mtx' mutexes on
487 		 * both `so_snd' and `so_rcv' before unsplice sockets.
488 		 */
489 		if ((soback = so->so_sp->ssp_soback) == NULL) {
490 			mtx_leave(&so->so_snd.sb_mtx);
491 			goto notsplicedback;
492 		}
493 		soref(soback);
494 		mtx_leave(&so->so_snd.sb_mtx);
495 
496 		/*
497 		 * `so' can be only unspliced, and never spliced again.
498 		 * Thus if issplicedback(so) check is positive, socket is
499 		 * still spliced and `ssp_soback' points to the same
500 		 * socket that `soback'.
501 		 */
502 		sblock(&soback->so_rcv, SBL_WAIT | SBL_NOINTR);
503 		if (issplicedback(so)) {
504 			int freeing = SOSP_FREEING_WRITE;
505 
506 			if (so->so_sp->ssp_soback == so)
507 				freeing |= SOSP_FREEING_READ;
508 			solock(soback);
509 			sounsplice(so->so_sp->ssp_soback, so, freeing);
510 			sounlock(soback);
511 		}
512 		sbunlock(&soback->so_rcv);
513 		sorele(soback);
514 
515 notsplicedback:
516 		sblock(&so->so_rcv, SBL_WAIT | SBL_NOINTR);
517 		if (isspliced(so)) {
518 			int freeing = SOSP_FREEING_READ;
519 
520 			if (so == so->so_sp->ssp_socket)
521 				freeing |= SOSP_FREEING_WRITE;
522 			solock(so);
523 			sounsplice(so, so->so_sp->ssp_socket, freeing);
524 			sounlock(so);
525 		}
526 		sbunlock(&so->so_rcv);
527 
528 		solock(so);
529 	}
530 free:
531 #endif /* SOCKET_SPLICE */
532 	/* sofree() calls sounlock(). */
533 	sofree(so, 0);
534 	return (error);
535 }
536 
537 void
538 soabort(struct socket *so)
539 {
540 	soassertlocked(so);
541 	pru_abort(so);
542 }
543 
544 int
545 soaccept(struct socket *so, struct mbuf *nam)
546 {
547 	int error = 0;
548 
549 	soassertlocked(so);
550 
551 	if ((so->so_state & SS_NOFDREF) == 0)
552 		panic("soaccept !NOFDREF: so %p, so_type %d", so, so->so_type);
553 	so->so_state &= ~SS_NOFDREF;
554 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
555 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
556 		error = pru_accept(so, nam);
557 	else
558 		error = ECONNABORTED;
559 	return (error);
560 }
561 
562 int
563 soconnect(struct socket *so, struct mbuf *nam)
564 {
565 	int error;
566 
567 	soassertlocked(so);
568 
569 	if (so->so_options & SO_ACCEPTCONN)
570 		return (EOPNOTSUPP);
571 	/*
572 	 * If protocol is connection-based, can only connect once.
573 	 * Otherwise, if connected, try to disconnect first.
574 	 * This allows user to disconnect by connecting to, e.g.,
575 	 * a null address.
576 	 */
577 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
578 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
579 	    (error = sodisconnect(so))))
580 		error = EISCONN;
581 	else
582 		error = pru_connect(so, nam);
583 	return (error);
584 }
585 
586 int
587 soconnect2(struct socket *so1, struct socket *so2)
588 {
589 	int persocket, error;
590 
591 	if ((persocket = solock_persocket(so1)))
592 		solock_pair(so1, so2);
593 	else
594 		solock(so1);
595 
596 	error = pru_connect2(so1, so2);
597 
598 	if (persocket)
599 		sounlock(so2);
600 	sounlock(so1);
601 	return (error);
602 }
603 
604 int
605 sodisconnect(struct socket *so)
606 {
607 	int error;
608 
609 	soassertlocked(so);
610 
611 	if ((so->so_state & SS_ISCONNECTED) == 0)
612 		return (ENOTCONN);
613 	if (so->so_state & SS_ISDISCONNECTING)
614 		return (EALREADY);
615 	error = pru_disconnect(so);
616 	return (error);
617 }
618 
619 int m_getuio(struct mbuf **, int, long, struct uio *);
620 
621 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
622 /*
623  * Send on a socket.
624  * If send must go all at once and message is larger than
625  * send buffering, then hard error.
626  * Lock against other senders.
627  * If must go all at once and not enough room now, then
628  * inform user that this would block and do nothing.
629  * Otherwise, if nonblocking, send as much as possible.
630  * The data to be sent is described by "uio" if nonzero,
631  * otherwise by the mbuf chain "top" (which must be null
632  * if uio is not).  Data provided in mbuf chain must be small
633  * enough to send all at once.
634  *
635  * Returns nonzero on error, timeout or signal; callers
636  * must check for short counts if EINTR/ERESTART are returned.
637  * Data and control buffers are freed on return.
638  */
639 int
640 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
641     struct mbuf *control, int flags)
642 {
643 	long space, clen = 0;
644 	size_t resid;
645 	int error;
646 	int atomic = sosendallatonce(so) || top;
647 	int dosolock = ((so->so_snd.sb_flags & SB_MTXLOCK) == 0);
648 
649 	if (uio)
650 		resid = uio->uio_resid;
651 	else
652 		resid = top->m_pkthdr.len;
653 	/* MSG_EOR on a SOCK_STREAM socket is invalid. */
654 	if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
655 		m_freem(top);
656 		m_freem(control);
657 		return (EINVAL);
658 	}
659 	if (uio && uio->uio_procp)
660 		uio->uio_procp->p_ru.ru_msgsnd++;
661 	if (control) {
662 		/*
663 		 * In theory clen should be unsigned (since control->m_len is).
664 		 * However, space must be signed, as it might be less than 0
665 		 * if we over-committed, and we must use a signed comparison
666 		 * of space and clen.
667 		 */
668 		clen = control->m_len;
669 		/* reserve extra space for AF_UNIX's internalize */
670 		if (so->so_proto->pr_domain->dom_family == AF_UNIX &&
671 		    clen >= CMSG_ALIGN(sizeof(struct cmsghdr)) &&
672 		    mtod(control, struct cmsghdr *)->cmsg_type == SCM_RIGHTS)
673 			clen = CMSG_SPACE(
674 			    (clen - CMSG_ALIGN(sizeof(struct cmsghdr))) *
675 			    (sizeof(struct fdpass) / sizeof(int)));
676 	}
677 
678 #define	snderr(errno)	{ error = errno; goto release; }
679 
680 restart:
681 	if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
682 		goto out;
683 	if (dosolock)
684 		solock_shared(so);
685 	sb_mtx_lock(&so->so_snd);
686 	so->so_snd.sb_state |= SS_ISSENDING;
687 	do {
688 		if (so->so_snd.sb_state & SS_CANTSENDMORE)
689 			snderr(EPIPE);
690 		if ((error = READ_ONCE(so->so_error))) {
691 			so->so_error = 0;
692 			snderr(error);
693 		}
694 		if ((so->so_state & SS_ISCONNECTED) == 0) {
695 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
696 				if (!(resid == 0 && clen != 0))
697 					snderr(ENOTCONN);
698 			} else if (addr == NULL)
699 				snderr(EDESTADDRREQ);
700 		}
701 		space = sbspace_locked(so, &so->so_snd);
702 		if (flags & MSG_OOB)
703 			space += 1024;
704 		if (so->so_proto->pr_domain->dom_family == AF_UNIX) {
705 			if (atomic && resid > so->so_snd.sb_hiwat)
706 				snderr(EMSGSIZE);
707 		} else {
708 			if (clen > so->so_snd.sb_hiwat ||
709 			    (atomic && resid > so->so_snd.sb_hiwat - clen))
710 				snderr(EMSGSIZE);
711 		}
712 		if (space < clen ||
713 		    (space - clen < resid &&
714 		    (atomic || space < so->so_snd.sb_lowat))) {
715 			if (flags & MSG_DONTWAIT)
716 				snderr(EWOULDBLOCK);
717 			sbunlock(&so->so_snd);
718 			error = sbwait(so, &so->so_snd);
719 			so->so_snd.sb_state &= ~SS_ISSENDING;
720 			sb_mtx_unlock(&so->so_snd);
721 			if (dosolock)
722 				sounlock_shared(so);
723 			if (error)
724 				goto out;
725 			goto restart;
726 		}
727 		space -= clen;
728 		do {
729 			if (uio == NULL) {
730 				/*
731 				 * Data is prepackaged in "top".
732 				 */
733 				resid = 0;
734 				if (flags & MSG_EOR)
735 					top->m_flags |= M_EOR;
736 			} else {
737 				sb_mtx_unlock(&so->so_snd);
738 				if (dosolock)
739 					sounlock_shared(so);
740 				error = m_getuio(&top, atomic, space, uio);
741 				if (dosolock)
742 					solock_shared(so);
743 				sb_mtx_lock(&so->so_snd);
744 				if (error)
745 					goto release;
746 				space -= top->m_pkthdr.len;
747 				resid = uio->uio_resid;
748 				if (flags & MSG_EOR)
749 					top->m_flags |= M_EOR;
750 			}
751 			if (resid == 0)
752 				so->so_snd.sb_state &= ~SS_ISSENDING;
753 			if (top && so->so_options & SO_ZEROIZE)
754 				top->m_flags |= M_ZEROIZE;
755 			sb_mtx_unlock(&so->so_snd);
756 			if (!dosolock)
757 				solock_shared(so);
758 			if (flags & MSG_OOB)
759 				error = pru_sendoob(so, top, addr, control);
760 			else
761 				error = pru_send(so, top, addr, control);
762 			if (!dosolock)
763 				sounlock_shared(so);
764 			sb_mtx_lock(&so->so_snd);
765 			clen = 0;
766 			control = NULL;
767 			top = NULL;
768 			if (error)
769 				goto release;
770 		} while (resid && space > 0);
771 	} while (resid);
772 
773 release:
774 	so->so_snd.sb_state &= ~SS_ISSENDING;
775 	sb_mtx_unlock(&so->so_snd);
776 	if (dosolock)
777 		sounlock_shared(so);
778 	sbunlock(&so->so_snd);
779 out:
780 	m_freem(top);
781 	m_freem(control);
782 	return (error);
783 }
784 
785 int
786 m_getuio(struct mbuf **mp, int atomic, long space, struct uio *uio)
787 {
788 	struct mbuf *m, *top = NULL;
789 	struct mbuf **nextp = &top;
790 	u_long len, mlen;
791 	size_t resid = uio->uio_resid;
792 	int error;
793 
794 	do {
795 		if (top == NULL) {
796 			MGETHDR(m, M_WAIT, MT_DATA);
797 			mlen = MHLEN;
798 			m->m_pkthdr.len = 0;
799 			m->m_pkthdr.ph_ifidx = 0;
800 		} else {
801 			MGET(m, M_WAIT, MT_DATA);
802 			mlen = MLEN;
803 		}
804 		/* chain mbuf together */
805 		*nextp = m;
806 		nextp = &m->m_next;
807 
808 		resid = ulmin(resid, space);
809 		if (resid >= MINCLSIZE) {
810 			MCLGETL(m, M_NOWAIT, ulmin(resid, MAXMCLBYTES));
811 			if ((m->m_flags & M_EXT) == 0)
812 				MCLGETL(m, M_NOWAIT, MCLBYTES);
813 			if ((m->m_flags & M_EXT) == 0)
814 				goto nopages;
815 			mlen = m->m_ext.ext_size;
816 			len = ulmin(mlen, resid);
817 			/*
818 			 * For datagram protocols, leave room
819 			 * for protocol headers in first mbuf.
820 			 */
821 			if (atomic && m == top && len < mlen - max_hdr)
822 				m->m_data += max_hdr;
823 		} else {
824 nopages:
825 			len = ulmin(mlen, resid);
826 			/*
827 			 * For datagram protocols, leave room
828 			 * for protocol headers in first mbuf.
829 			 */
830 			if (atomic && m == top && len < mlen - max_hdr)
831 				m_align(m, len);
832 		}
833 
834 		error = uiomove(mtod(m, caddr_t), len, uio);
835 		if (error) {
836 			m_freem(top);
837 			return (error);
838 		}
839 
840 		/* adjust counters */
841 		resid = uio->uio_resid;
842 		space -= len;
843 		m->m_len = len;
844 		top->m_pkthdr.len += len;
845 
846 		/* Is there more space and more data? */
847 	} while (space > 0 && resid > 0);
848 
849 	*mp = top;
850 	return 0;
851 }
852 
853 /*
854  * Following replacement or removal of the first mbuf on the first
855  * mbuf chain of a socket buffer, push necessary state changes back
856  * into the socket buffer so that other consumers see the values
857  * consistently.  'nextrecord' is the callers locally stored value of
858  * the original value of sb->sb_mb->m_nextpkt which must be restored
859  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
860  */
861 void
862 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
863 {
864 
865 	/*
866 	 * First, update for the new value of nextrecord.  If necessary,
867 	 * make it the first record.
868 	 */
869 	if (sb->sb_mb != NULL)
870 		sb->sb_mb->m_nextpkt = nextrecord;
871 	else
872 		sb->sb_mb = nextrecord;
873 
874 	/*
875 	 * Now update any dependent socket buffer fields to reflect
876 	 * the new state.  This is an inline of SB_EMPTY_FIXUP, with
877 	 * the addition of a second clause that takes care of the
878 	 * case where sb_mb has been updated, but remains the last
879 	 * record.
880 	 */
881 	if (sb->sb_mb == NULL) {
882 		sb->sb_mbtail = NULL;
883 		sb->sb_lastrecord = NULL;
884 	} else if (sb->sb_mb->m_nextpkt == NULL)
885 		sb->sb_lastrecord = sb->sb_mb;
886 }
887 
888 /*
889  * Implement receive operations on a socket.
890  * We depend on the way that records are added to the sockbuf
891  * by sbappend*.  In particular, each record (mbufs linked through m_next)
892  * must begin with an address if the protocol so specifies,
893  * followed by an optional mbuf or mbufs containing ancillary data,
894  * and then zero or more mbufs of data.
895  * In order to avoid blocking network for the entire time here, we release
896  * the solock() while doing the actual copy to user space.
897  * Although the sockbuf is locked, new data may still be appended,
898  * and thus we must maintain consistency of the sockbuf during that time.
899  *
900  * The caller may receive the data as a single mbuf chain by supplying
901  * an mbuf **mp0 for use in returning the chain.  The uio is then used
902  * only for the count in uio_resid.
903  */
904 int
905 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
906     struct mbuf **mp0, struct mbuf **controlp, int *flagsp,
907     socklen_t controllen)
908 {
909 	struct mbuf *m, **mp;
910 	struct mbuf *cm;
911 	u_long len, offset, moff;
912 	int flags, error, error2, type, uio_error = 0;
913 	const struct protosw *pr = so->so_proto;
914 	struct mbuf *nextrecord;
915 	size_t resid, orig_resid = uio->uio_resid;
916 	int dosolock = ((so->so_rcv.sb_flags & SB_MTXLOCK) == 0);
917 
918 	mp = mp0;
919 	if (paddr)
920 		*paddr = NULL;
921 	if (controlp)
922 		*controlp = NULL;
923 	if (flagsp)
924 		flags = *flagsp &~ MSG_EOR;
925 	else
926 		flags = 0;
927 	if (flags & MSG_OOB) {
928 		m = m_get(M_WAIT, MT_DATA);
929 		solock(so);
930 		error = pru_rcvoob(so, m, flags & MSG_PEEK);
931 		sounlock(so);
932 		if (error)
933 			goto bad;
934 		do {
935 			error = uiomove(mtod(m, caddr_t),
936 			    ulmin(uio->uio_resid, m->m_len), uio);
937 			m = m_free(m);
938 		} while (uio->uio_resid && error == 0 && m);
939 bad:
940 		m_freem(m);
941 		return (error);
942 	}
943 	if (mp)
944 		*mp = NULL;
945 
946 restart:
947 	if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0)
948 		return (error);
949 	if (dosolock)
950 		solock_shared(so);
951 	sb_mtx_lock(&so->so_rcv);
952 
953 	m = so->so_rcv.sb_mb;
954 #ifdef SOCKET_SPLICE
955 	if (isspliced(so))
956 		m = NULL;
957 #endif /* SOCKET_SPLICE */
958 	/*
959 	 * If we have less data than requested, block awaiting more
960 	 * (subject to any timeout) if:
961 	 *   1. the current count is less than the low water mark,
962 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
963 	 *	receive operation at once if we block (resid <= hiwat), or
964 	 *   3. MSG_DONTWAIT is not set.
965 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
966 	 * we have to do the receive in sections, and thus risk returning
967 	 * a short count if a timeout or signal occurs after we start.
968 	 */
969 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
970 	    so->so_rcv.sb_cc < uio->uio_resid) &&
971 	    (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
972 	    ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
973 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
974 #ifdef DIAGNOSTIC
975 		if (m == NULL && so->so_rcv.sb_cc)
976 #ifdef SOCKET_SPLICE
977 		    if (!isspliced(so))
978 #endif /* SOCKET_SPLICE */
979 			panic("receive 1: so %p, so_type %d, sb_cc %lu",
980 			    so, so->so_type, so->so_rcv.sb_cc);
981 #endif
982 		if ((error2 = READ_ONCE(so->so_error))) {
983 			if (m)
984 				goto dontblock;
985 			error = error2;
986 			if ((flags & MSG_PEEK) == 0)
987 				so->so_error = 0;
988 			goto release;
989 		}
990 		if (so->so_rcv.sb_state & SS_CANTRCVMORE) {
991 			if (m)
992 				goto dontblock;
993 			else if (so->so_rcv.sb_cc == 0)
994 				goto release;
995 		}
996 		for (; m; m = m->m_next)
997 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
998 				m = so->so_rcv.sb_mb;
999 				goto dontblock;
1000 			}
1001 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1002 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1003 			error = ENOTCONN;
1004 			goto release;
1005 		}
1006 		if (uio->uio_resid == 0 && controlp == NULL)
1007 			goto release;
1008 		if (flags & MSG_DONTWAIT) {
1009 			error = EWOULDBLOCK;
1010 			goto release;
1011 		}
1012 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
1013 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
1014 
1015 		sbunlock(&so->so_rcv);
1016 		error = sbwait(so, &so->so_rcv);
1017 		sb_mtx_unlock(&so->so_rcv);
1018 		if (dosolock)
1019 			sounlock_shared(so);
1020 		if (error)
1021 			return (error);
1022 		goto restart;
1023 	}
1024 dontblock:
1025 	/*
1026 	 * On entry here, m points to the first record of the socket buffer.
1027 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1028 	 * pointer to the next record in the socket buffer.  We must keep the
1029 	 * various socket buffer pointers and local stack versions of the
1030 	 * pointers in sync, pushing out modifications before operations that
1031 	 * may sleep, and re-reading them afterwards.
1032 	 *
1033 	 * Otherwise, we will race with the network stack appending new data
1034 	 * or records onto the socket buffer by using inconsistent/stale
1035 	 * versions of the field, possibly resulting in socket buffer
1036 	 * corruption.
1037 	 */
1038 	if (uio->uio_procp)
1039 		uio->uio_procp->p_ru.ru_msgrcv++;
1040 	KASSERT(m == so->so_rcv.sb_mb);
1041 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
1042 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
1043 	nextrecord = m->m_nextpkt;
1044 	if (pr->pr_flags & PR_ADDR) {
1045 #ifdef DIAGNOSTIC
1046 		if (m->m_type != MT_SONAME)
1047 			panic("receive 1a: so %p, so_type %d, m %p, m_type %d",
1048 			    so, so->so_type, m, m->m_type);
1049 #endif
1050 		orig_resid = 0;
1051 		if (flags & MSG_PEEK) {
1052 			if (paddr)
1053 				*paddr = m_copym(m, 0, m->m_len, M_NOWAIT);
1054 			m = m->m_next;
1055 		} else {
1056 			sbfree(so, &so->so_rcv, m);
1057 			if (paddr) {
1058 				*paddr = m;
1059 				so->so_rcv.sb_mb = m->m_next;
1060 				m->m_next = NULL;
1061 				m = so->so_rcv.sb_mb;
1062 			} else {
1063 				so->so_rcv.sb_mb = m_free(m);
1064 				m = so->so_rcv.sb_mb;
1065 			}
1066 			sbsync(&so->so_rcv, nextrecord);
1067 		}
1068 	}
1069 	while (m && m->m_type == MT_CONTROL && error == 0) {
1070 		int skip = 0;
1071 		if (flags & MSG_PEEK) {
1072 			if (mtod(m, struct cmsghdr *)->cmsg_type ==
1073 			    SCM_RIGHTS) {
1074 				/* don't leak internalized SCM_RIGHTS msgs */
1075 				skip = 1;
1076 			} else if (controlp)
1077 				*controlp = m_copym(m, 0, m->m_len, M_NOWAIT);
1078 			m = m->m_next;
1079 		} else {
1080 			sbfree(so, &so->so_rcv, m);
1081 			so->so_rcv.sb_mb = m->m_next;
1082 			m->m_nextpkt = m->m_next = NULL;
1083 			cm = m;
1084 			m = so->so_rcv.sb_mb;
1085 			sbsync(&so->so_rcv, nextrecord);
1086 			if (controlp) {
1087 				if (pr->pr_domain->dom_externalize) {
1088 					sb_mtx_unlock(&so->so_rcv);
1089 					if (dosolock)
1090 						sounlock_shared(so);
1091 					error =
1092 					    (*pr->pr_domain->dom_externalize)
1093 					    (cm, controllen, flags);
1094 					if (dosolock)
1095 						solock_shared(so);
1096 					sb_mtx_lock(&so->so_rcv);
1097 				}
1098 				*controlp = cm;
1099 			} else {
1100 				/*
1101 				 * Dispose of any SCM_RIGHTS message that went
1102 				 * through the read path rather than recv.
1103 				 */
1104 				if (pr->pr_domain->dom_dispose) {
1105 					sb_mtx_unlock(&so->so_rcv);
1106 					pr->pr_domain->dom_dispose(cm);
1107 					sb_mtx_lock(&so->so_rcv);
1108 				}
1109 				m_free(cm);
1110 			}
1111 		}
1112 		if (m != NULL)
1113 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1114 		else
1115 			nextrecord = so->so_rcv.sb_mb;
1116 		if (controlp && !skip)
1117 			controlp = &(*controlp)->m_next;
1118 		orig_resid = 0;
1119 	}
1120 
1121 	/* If m is non-NULL, we have some data to read. */
1122 	if (m) {
1123 		type = m->m_type;
1124 		if (type == MT_OOBDATA)
1125 			flags |= MSG_OOB;
1126 		if (m->m_flags & M_BCAST)
1127 			flags |= MSG_BCAST;
1128 		if (m->m_flags & M_MCAST)
1129 			flags |= MSG_MCAST;
1130 	}
1131 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1132 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1133 
1134 	moff = 0;
1135 	offset = 0;
1136 	while (m && uio->uio_resid > 0 && error == 0) {
1137 		if (m->m_type == MT_OOBDATA) {
1138 			if (type != MT_OOBDATA)
1139 				break;
1140 		} else if (type == MT_OOBDATA) {
1141 			break;
1142 		} else if (m->m_type == MT_CONTROL) {
1143 			/*
1144 			 * If there is more than one control message in the
1145 			 * stream, we do a short read.  Next can be received
1146 			 * or disposed by another system call.
1147 			 */
1148 			break;
1149 #ifdef DIAGNOSTIC
1150 		} else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) {
1151 			panic("receive 3: so %p, so_type %d, m %p, m_type %d",
1152 			    so, so->so_type, m, m->m_type);
1153 #endif
1154 		}
1155 		so->so_rcv.sb_state &= ~SS_RCVATMARK;
1156 		len = uio->uio_resid;
1157 		if (so->so_oobmark && len > so->so_oobmark - offset)
1158 			len = so->so_oobmark - offset;
1159 		if (len > m->m_len - moff)
1160 			len = m->m_len - moff;
1161 		/*
1162 		 * If mp is set, just pass back the mbufs.
1163 		 * Otherwise copy them out via the uio, then free.
1164 		 * Sockbuf must be consistent here (points to current mbuf,
1165 		 * it points to next record) when we drop priority;
1166 		 * we must note any additions to the sockbuf when we
1167 		 * block interrupts again.
1168 		 */
1169 		if (mp == NULL && uio_error == 0) {
1170 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1171 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1172 			resid = uio->uio_resid;
1173 			sb_mtx_unlock(&so->so_rcv);
1174 			if (dosolock)
1175 				sounlock_shared(so);
1176 			uio_error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1177 			if (dosolock)
1178 				solock_shared(so);
1179 			sb_mtx_lock(&so->so_rcv);
1180 			if (uio_error)
1181 				uio->uio_resid = resid - len;
1182 		} else
1183 			uio->uio_resid -= len;
1184 		if (len == m->m_len - moff) {
1185 			if (m->m_flags & M_EOR)
1186 				flags |= MSG_EOR;
1187 			if (flags & MSG_PEEK) {
1188 				m = m->m_next;
1189 				moff = 0;
1190 				orig_resid = 0;
1191 			} else {
1192 				nextrecord = m->m_nextpkt;
1193 				sbfree(so, &so->so_rcv, m);
1194 				if (mp) {
1195 					*mp = m;
1196 					mp = &m->m_next;
1197 					so->so_rcv.sb_mb = m = m->m_next;
1198 					*mp = NULL;
1199 				} else {
1200 					so->so_rcv.sb_mb = m_free(m);
1201 					m = so->so_rcv.sb_mb;
1202 				}
1203 				/*
1204 				 * If m != NULL, we also know that
1205 				 * so->so_rcv.sb_mb != NULL.
1206 				 */
1207 				KASSERT(so->so_rcv.sb_mb == m);
1208 				if (m) {
1209 					m->m_nextpkt = nextrecord;
1210 					if (nextrecord == NULL)
1211 						so->so_rcv.sb_lastrecord = m;
1212 				} else {
1213 					so->so_rcv.sb_mb = nextrecord;
1214 					SB_EMPTY_FIXUP(&so->so_rcv);
1215 				}
1216 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1217 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1218 			}
1219 		} else {
1220 			if (flags & MSG_PEEK) {
1221 				moff += len;
1222 				orig_resid = 0;
1223 			} else {
1224 				if (mp)
1225 					*mp = m_copym(m, 0, len, M_WAIT);
1226 				m->m_data += len;
1227 				m->m_len -= len;
1228 				so->so_rcv.sb_cc -= len;
1229 				so->so_rcv.sb_datacc -= len;
1230 			}
1231 		}
1232 		if (so->so_oobmark) {
1233 			if ((flags & MSG_PEEK) == 0) {
1234 				so->so_oobmark -= len;
1235 				if (so->so_oobmark == 0) {
1236 					so->so_rcv.sb_state |= SS_RCVATMARK;
1237 					break;
1238 				}
1239 			} else {
1240 				offset += len;
1241 				if (offset == so->so_oobmark)
1242 					break;
1243 			}
1244 		}
1245 		if (flags & MSG_EOR)
1246 			break;
1247 		/*
1248 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1249 		 * we must not quit until "uio->uio_resid == 0" or an error
1250 		 * termination.  If a signal/timeout occurs, return
1251 		 * with a short count but without error.
1252 		 * Keep sockbuf locked against other readers.
1253 		 */
1254 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1255 		    !sosendallatonce(so) && !nextrecord) {
1256 			if (so->so_rcv.sb_state & SS_CANTRCVMORE ||
1257 			    so->so_error)
1258 				break;
1259 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1260 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1261 			if (sbwait(so, &so->so_rcv)) {
1262 				sb_mtx_unlock(&so->so_rcv);
1263 				if (dosolock)
1264 					sounlock_shared(so);
1265 				sbunlock(&so->so_rcv);
1266 				return (0);
1267 			}
1268 			if ((m = so->so_rcv.sb_mb) != NULL)
1269 				nextrecord = m->m_nextpkt;
1270 		}
1271 	}
1272 
1273 	if (m && pr->pr_flags & PR_ATOMIC) {
1274 		flags |= MSG_TRUNC;
1275 		if ((flags & MSG_PEEK) == 0)
1276 			(void) sbdroprecord(so, &so->so_rcv);
1277 	}
1278 	if ((flags & MSG_PEEK) == 0) {
1279 		if (m == NULL) {
1280 			/*
1281 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1282 			 * part makes sure sb_lastrecord is up-to-date if
1283 			 * there is still data in the socket buffer.
1284 			 */
1285 			so->so_rcv.sb_mb = nextrecord;
1286 			if (so->so_rcv.sb_mb == NULL) {
1287 				so->so_rcv.sb_mbtail = NULL;
1288 				so->so_rcv.sb_lastrecord = NULL;
1289 			} else if (nextrecord->m_nextpkt == NULL)
1290 				so->so_rcv.sb_lastrecord = nextrecord;
1291 		}
1292 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1293 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1294 		if (pr->pr_flags & PR_WANTRCVD) {
1295 			sb_mtx_unlock(&so->so_rcv);
1296 			if (!dosolock)
1297 				solock_shared(so);
1298 			pru_rcvd(so);
1299 			if (!dosolock)
1300 				sounlock_shared(so);
1301 			sb_mtx_lock(&so->so_rcv);
1302 		}
1303 	}
1304 	if (orig_resid == uio->uio_resid && orig_resid &&
1305 	    (flags & MSG_EOR) == 0 &&
1306 	    (so->so_rcv.sb_state & SS_CANTRCVMORE) == 0) {
1307 		sb_mtx_unlock(&so->so_rcv);
1308 		sbunlock(&so->so_rcv);
1309 		goto restart;
1310 	}
1311 
1312 	if (uio_error)
1313 		error = uio_error;
1314 
1315 	if (flagsp)
1316 		*flagsp |= flags;
1317 release:
1318 	sb_mtx_unlock(&so->so_rcv);
1319 	if (dosolock)
1320 		sounlock_shared(so);
1321 	sbunlock(&so->so_rcv);
1322 	return (error);
1323 }
1324 
1325 int
1326 soshutdown(struct socket *so, int how)
1327 {
1328 	int error = 0;
1329 
1330 	switch (how) {
1331 	case SHUT_RD:
1332 		sorflush(so);
1333 		break;
1334 	case SHUT_RDWR:
1335 		sorflush(so);
1336 		/* FALLTHROUGH */
1337 	case SHUT_WR:
1338 		solock(so);
1339 		error = pru_shutdown(so);
1340 		sounlock(so);
1341 		break;
1342 	default:
1343 		error = EINVAL;
1344 		break;
1345 	}
1346 
1347 	return (error);
1348 }
1349 
1350 void
1351 sorflush(struct socket *so)
1352 {
1353 	struct sockbuf *sb = &so->so_rcv;
1354 	struct mbuf *m;
1355 	const struct protosw *pr = so->so_proto;
1356 	int error;
1357 
1358 	error = sblock(sb, SBL_WAIT | SBL_NOINTR);
1359 	/* with SBL_WAIT and SLB_NOINTR sblock() must not fail */
1360 	KASSERT(error == 0);
1361 
1362 	solock_shared(so);
1363 	socantrcvmore(so);
1364 	mtx_enter(&sb->sb_mtx);
1365 	m = sb->sb_mb;
1366 	memset(&sb->sb_startzero, 0,
1367 	     (caddr_t)&sb->sb_endzero - (caddr_t)&sb->sb_startzero);
1368 	sb->sb_timeo_nsecs = INFSLP;
1369 	mtx_leave(&sb->sb_mtx);
1370 	sounlock_shared(so);
1371 	sbunlock(sb);
1372 
1373 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose)
1374 		(*pr->pr_domain->dom_dispose)(m);
1375 	m_purge(m);
1376 }
1377 
1378 #ifdef SOCKET_SPLICE
1379 
1380 #define so_splicelen	so_sp->ssp_len
1381 #define so_splicemax	so_sp->ssp_max
1382 #define so_idletv	so_sp->ssp_idletv
1383 #define so_idleto	so_sp->ssp_idleto
1384 #define so_splicetask	so_sp->ssp_task
1385 
1386 int
1387 sosplice(struct socket *so, int fd, off_t max, struct timeval *tv)
1388 {
1389 	struct file	*fp;
1390 	struct socket	*sosp;
1391 	struct taskq	*tq;
1392 	int		 error = 0;
1393 
1394 	if ((so->so_proto->pr_flags & PR_SPLICE) == 0)
1395 		return (EPROTONOSUPPORT);
1396 	if (max && max < 0)
1397 		return (EINVAL);
1398 	if (tv && (tv->tv_sec < 0 || !timerisvalid(tv)))
1399 		return (EINVAL);
1400 
1401 	/* If no fd is given, unsplice by removing existing link. */
1402 	if (fd < 0) {
1403 		if ((error = sblock(&so->so_rcv, SBL_WAIT)) != 0)
1404 			return (error);
1405 		solock(so);
1406 		if (so->so_options & SO_ACCEPTCONN) {
1407 			error = EOPNOTSUPP;
1408 			goto out;
1409 		}
1410 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1411 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1412 			error = ENOTCONN;
1413 			goto out;
1414 		}
1415 
1416 		if (so->so_sp && so->so_sp->ssp_socket)
1417 			sounsplice(so, so->so_sp->ssp_socket, 0);
1418  out:
1419 		sounlock(so);
1420 		sbunlock(&so->so_rcv);
1421 		return (error);
1422 	}
1423 
1424 	if (sosplice_taskq == NULL) {
1425 		rw_enter_write(&sosplice_lock);
1426 		if (sosplice_taskq == NULL) {
1427 			tq = taskq_create("sosplice", 1, IPL_SOFTNET,
1428 			    TASKQ_MPSAFE);
1429 			if (tq == NULL) {
1430 				rw_exit_write(&sosplice_lock);
1431 				return (ENOMEM);
1432 			}
1433 			/* Ensure the taskq is fully visible to other CPUs. */
1434 			membar_producer();
1435 			sosplice_taskq = tq;
1436 		}
1437 		rw_exit_write(&sosplice_lock);
1438 	} else {
1439 		/* Ensure the taskq is fully visible on this CPU. */
1440 		membar_consumer();
1441 	}
1442 
1443 	/* Find sosp, the drain socket where data will be spliced into. */
1444 	if ((error = getsock(curproc, fd, &fp)) != 0)
1445 		return (error);
1446 	sosp = fp->f_data;
1447 
1448 	if (sosp->so_proto->pr_usrreqs->pru_send !=
1449 	    so->so_proto->pr_usrreqs->pru_send) {
1450 		error = EPROTONOSUPPORT;
1451 		goto frele;
1452 	}
1453 
1454 	if ((error = sblock(&so->so_rcv, SBL_WAIT)) != 0)
1455 		goto frele;
1456 	if ((error = sblock(&sosp->so_snd, SBL_WAIT)) != 0) {
1457 		sbunlock(&so->so_rcv);
1458 		goto frele;
1459 	}
1460 	solock(so);
1461 
1462 	if ((so->so_options & SO_ACCEPTCONN) ||
1463 	    (sosp->so_options & SO_ACCEPTCONN)) {
1464 		error = EOPNOTSUPP;
1465 		goto release;
1466 	}
1467 	if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1468 	    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1469 		error = ENOTCONN;
1470 		goto release;
1471 	}
1472 	if ((sosp->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0) {
1473 		error = ENOTCONN;
1474 		goto release;
1475 	}
1476 	if (so->so_sp == NULL)
1477 		so->so_sp = pool_get(&sosplice_pool, PR_WAITOK | PR_ZERO);
1478 	if (sosp->so_sp == NULL)
1479 		sosp->so_sp = pool_get(&sosplice_pool, PR_WAITOK | PR_ZERO);
1480 	if (so->so_sp->ssp_socket || sosp->so_sp->ssp_soback) {
1481 		error = EBUSY;
1482 		goto release;
1483 	}
1484 
1485 	so->so_splicelen = 0;
1486 	so->so_splicemax = max;
1487 	if (tv)
1488 		so->so_idletv = *tv;
1489 	else
1490 		timerclear(&so->so_idletv);
1491 	timeout_set_flags(&so->so_idleto, soidle, so,
1492 	    KCLOCK_NONE, TIMEOUT_PROC | TIMEOUT_MPSAFE);
1493 	task_set(&so->so_splicetask, sotask, so);
1494 
1495 	/*
1496 	 * To prevent sorwakeup() calling somove() before this somove()
1497 	 * has finished, the socket buffers are not marked as spliced yet.
1498 	 */
1499 
1500 	/* Splice so and sosp together. */
1501 	mtx_enter(&so->so_rcv.sb_mtx);
1502 	mtx_enter(&sosp->so_snd.sb_mtx);
1503 	so->so_sp->ssp_socket = sosp;
1504 	sosp->so_sp->ssp_soback = so;
1505 	mtx_leave(&sosp->so_snd.sb_mtx);
1506 	mtx_leave(&so->so_rcv.sb_mtx);
1507 
1508 	if ((so->so_proto->pr_flags & PR_WANTRCVD) == 0)
1509 		sounlock(so);
1510 	if (somove(so, M_WAIT)) {
1511 		mtx_enter(&so->so_rcv.sb_mtx);
1512 		mtx_enter(&sosp->so_snd.sb_mtx);
1513 		so->so_rcv.sb_flags |= SB_SPLICE;
1514 		sosp->so_snd.sb_flags |= SB_SPLICE;
1515 		mtx_leave(&sosp->so_snd.sb_mtx);
1516 		mtx_leave(&so->so_rcv.sb_mtx);
1517 	}
1518 	if ((so->so_proto->pr_flags & PR_WANTRCVD) == 0)
1519 		solock(so);
1520 
1521  release:
1522 	sounlock(so);
1523 	sbunlock(&sosp->so_snd);
1524 	sbunlock(&so->so_rcv);
1525  frele:
1526 	FRELE(fp, curproc);
1527 
1528 	return (error);
1529 }
1530 
1531 void
1532 sounsplice(struct socket *so, struct socket *sosp, int freeing)
1533 {
1534 	if ((so->so_proto->pr_flags & PR_WANTRCVD) == 0)
1535 		sbassertlocked(&so->so_rcv);
1536 	soassertlocked(so);
1537 
1538 	task_del(sosplice_taskq, &so->so_splicetask);
1539 	timeout_del(&so->so_idleto);
1540 
1541 	mtx_enter(&so->so_rcv.sb_mtx);
1542 	mtx_enter(&sosp->so_snd.sb_mtx);
1543 	so->so_rcv.sb_flags &= ~SB_SPLICE;
1544 	sosp->so_snd.sb_flags &= ~SB_SPLICE;
1545 	so->so_sp->ssp_socket = sosp->so_sp->ssp_soback = NULL;
1546 	mtx_leave(&sosp->so_snd.sb_mtx);
1547 	mtx_leave(&so->so_rcv.sb_mtx);
1548 
1549 	/* Do not wakeup a socket that is about to be freed. */
1550 	if ((freeing & SOSP_FREEING_READ) == 0 && soreadable(so))
1551 		sorwakeup(so);
1552 	if ((freeing & SOSP_FREEING_WRITE) == 0 && sowriteable(sosp))
1553 		sowwakeup(sosp);
1554 }
1555 
1556 void
1557 soidle(void *arg)
1558 {
1559 	struct socket *so = arg;
1560 
1561 	sblock(&so->so_rcv, SBL_WAIT | SBL_NOINTR);
1562 	solock(so);
1563 	/*
1564 	 * Depending on socket type, sblock(&so->so_rcv) or solock()
1565 	 * is always held while modifying SB_SPLICE and
1566 	 * so->so_sp->ssp_socket.
1567 	 */
1568 	if (so->so_rcv.sb_flags & SB_SPLICE) {
1569 		so->so_error = ETIMEDOUT;
1570 		sounsplice(so, so->so_sp->ssp_socket, 0);
1571 	}
1572 	sounlock(so);
1573 	sbunlock(&so->so_rcv);
1574 }
1575 
1576 void
1577 sotask(void *arg)
1578 {
1579 	struct socket *so = arg;
1580 	int doyield = 0;
1581 	int sockstream = (so->so_proto->pr_flags & PR_WANTRCVD);
1582 
1583 	/*
1584 	 * sblock() on `so_rcv' protects sockets from being unspliced
1585 	 * for UDP case. TCP sockets still rely on solock().
1586 	 */
1587 
1588 	sblock(&so->so_rcv, SBL_WAIT | SBL_NOINTR);
1589 	if (sockstream)
1590 		solock(so);
1591 
1592 	if (so->so_rcv.sb_flags & SB_SPLICE) {
1593 		if (sockstream)
1594 			doyield = 1;
1595 		somove(so, M_DONTWAIT);
1596 	}
1597 
1598 	if (sockstream)
1599 		sounlock(so);
1600 	sbunlock(&so->so_rcv);
1601 
1602 	if (doyield) {
1603 		/* Avoid user land starvation. */
1604 		yield();
1605 	}
1606 }
1607 
1608 /*
1609  * The socket splicing task or idle timeout may sleep while grabbing the net
1610  * lock.  As sofree() can be called anytime, sotask() or soidle() could access
1611  * the socket memory of a freed socket after wakeup.  So delay the pool_put()
1612  * after all pending socket splicing tasks or timeouts have finished.  Do this
1613  * by scheduling it on the same threads.
1614  */
1615 void
1616 soreaper(void *arg)
1617 {
1618 	struct socket *so = arg;
1619 
1620 	/* Reuse splice task, sounsplice() has been called before. */
1621 	task_set(&so->so_sp->ssp_task, soput, so);
1622 	task_add(sosplice_taskq, &so->so_sp->ssp_task);
1623 }
1624 
1625 void
1626 soput(void *arg)
1627 {
1628 	struct socket *so = arg;
1629 
1630 	pool_put(&sosplice_pool, so->so_sp);
1631 	pool_put(&socket_pool, so);
1632 }
1633 
1634 /*
1635  * Move data from receive buffer of spliced source socket to send
1636  * buffer of drain socket.  Try to move as much as possible in one
1637  * big chunk.  It is a TCP only implementation.
1638  * Return value 0 means splicing has been finished, 1 continue.
1639  */
1640 int
1641 somove(struct socket *so, int wait)
1642 {
1643 	struct socket	*sosp = so->so_sp->ssp_socket;
1644 	struct mbuf	*m, **mp, *nextrecord;
1645 	u_long		 len, off, oobmark;
1646 	long		 space;
1647 	int		 error = 0, maxreached = 0, unsplice = 0;
1648 	unsigned int	 rcvstate;
1649 	int		 sockdgram = ((so->so_proto->pr_flags &
1650 			     PR_WANTRCVD) == 0);
1651 
1652 	if (sockdgram)
1653 		sbassertlocked(&so->so_rcv);
1654 	else
1655 		soassertlocked(so);
1656 
1657 	mtx_enter(&so->so_rcv.sb_mtx);
1658 	mtx_enter(&sosp->so_snd.sb_mtx);
1659 
1660  nextpkt:
1661 	if ((error = READ_ONCE(so->so_error)))
1662 		goto release;
1663 	if (sosp->so_snd.sb_state & SS_CANTSENDMORE) {
1664 		error = EPIPE;
1665 		goto release;
1666 	}
1667 
1668 	error = READ_ONCE(sosp->so_error);
1669 	if (error) {
1670 		if (error != ETIMEDOUT && error != EFBIG && error != ELOOP)
1671 			goto release;
1672 		error = 0;
1673 	}
1674 	if ((sosp->so_state & SS_ISCONNECTED) == 0)
1675 		goto release;
1676 
1677 	/* Calculate how many bytes can be copied now. */
1678 	len = so->so_rcv.sb_datacc;
1679 	if (so->so_splicemax) {
1680 		KASSERT(so->so_splicelen < so->so_splicemax);
1681 		if (so->so_splicemax <= so->so_splicelen + len) {
1682 			len = so->so_splicemax - so->so_splicelen;
1683 			maxreached = 1;
1684 		}
1685 	}
1686 	space = sbspace_locked(sosp, &sosp->so_snd);
1687 	if (so->so_oobmark && so->so_oobmark < len &&
1688 	    so->so_oobmark < space + 1024)
1689 		space += 1024;
1690 	if (space <= 0) {
1691 		maxreached = 0;
1692 		goto release;
1693 	}
1694 	if (space < len) {
1695 		maxreached = 0;
1696 		if (space < sosp->so_snd.sb_lowat)
1697 			goto release;
1698 		len = space;
1699 	}
1700 	sosp->so_snd.sb_state |= SS_ISSENDING;
1701 
1702 	SBLASTRECORDCHK(&so->so_rcv, "somove 1");
1703 	SBLASTMBUFCHK(&so->so_rcv, "somove 1");
1704 	m = so->so_rcv.sb_mb;
1705 	if (m == NULL)
1706 		goto release;
1707 	nextrecord = m->m_nextpkt;
1708 
1709 	/* Drop address and control information not used with splicing. */
1710 	if (so->so_proto->pr_flags & PR_ADDR) {
1711 #ifdef DIAGNOSTIC
1712 		if (m->m_type != MT_SONAME)
1713 			panic("somove soname: so %p, so_type %d, m %p, "
1714 			    "m_type %d", so, so->so_type, m, m->m_type);
1715 #endif
1716 		m = m->m_next;
1717 	}
1718 	while (m && m->m_type == MT_CONTROL)
1719 		m = m->m_next;
1720 	if (m == NULL) {
1721 		sbdroprecord(so, &so->so_rcv);
1722 		if (so->so_proto->pr_flags & PR_WANTRCVD) {
1723 			mtx_leave(&sosp->so_snd.sb_mtx);
1724 			mtx_leave(&so->so_rcv.sb_mtx);
1725 			pru_rcvd(so);
1726 			mtx_enter(&so->so_rcv.sb_mtx);
1727 			mtx_enter(&sosp->so_snd.sb_mtx);
1728 		}
1729 		goto nextpkt;
1730 	}
1731 
1732 	/*
1733 	 * By splicing sockets connected to localhost, userland might create a
1734 	 * loop.  Dissolve splicing with error if loop is detected by counter.
1735 	 *
1736 	 * If we deal with looped broadcast/multicast packet we bail out with
1737 	 * no error to suppress splice termination.
1738 	 */
1739 	if ((m->m_flags & M_PKTHDR) &&
1740 	    ((m->m_pkthdr.ph_loopcnt++ >= M_MAXLOOP) ||
1741 	    ((m->m_flags & M_LOOP) && (m->m_flags & (M_BCAST|M_MCAST))))) {
1742 		error = ELOOP;
1743 		goto release;
1744 	}
1745 
1746 	if (so->so_proto->pr_flags & PR_ATOMIC) {
1747 		if ((m->m_flags & M_PKTHDR) == 0)
1748 			panic("somove !PKTHDR: so %p, so_type %d, m %p, "
1749 			    "m_type %d", so, so->so_type, m, m->m_type);
1750 		if (sosp->so_snd.sb_hiwat < m->m_pkthdr.len) {
1751 			error = EMSGSIZE;
1752 			goto release;
1753 		}
1754 		if (len < m->m_pkthdr.len)
1755 			goto release;
1756 		if (m->m_pkthdr.len < len) {
1757 			maxreached = 0;
1758 			len = m->m_pkthdr.len;
1759 		}
1760 		/*
1761 		 * Throw away the name mbuf after it has been assured
1762 		 * that the whole first record can be processed.
1763 		 */
1764 		m = so->so_rcv.sb_mb;
1765 		sbfree(so, &so->so_rcv, m);
1766 		so->so_rcv.sb_mb = m_free(m);
1767 		sbsync(&so->so_rcv, nextrecord);
1768 	}
1769 	/*
1770 	 * Throw away the control mbufs after it has been assured
1771 	 * that the whole first record can be processed.
1772 	 */
1773 	m = so->so_rcv.sb_mb;
1774 	while (m && m->m_type == MT_CONTROL) {
1775 		sbfree(so, &so->so_rcv, m);
1776 		so->so_rcv.sb_mb = m_free(m);
1777 		m = so->so_rcv.sb_mb;
1778 		sbsync(&so->so_rcv, nextrecord);
1779 	}
1780 
1781 	SBLASTRECORDCHK(&so->so_rcv, "somove 2");
1782 	SBLASTMBUFCHK(&so->so_rcv, "somove 2");
1783 
1784 	/* Take at most len mbufs out of receive buffer. */
1785 	for (off = 0, mp = &m; off <= len && *mp;
1786 	    off += (*mp)->m_len, mp = &(*mp)->m_next) {
1787 		u_long size = len - off;
1788 
1789 #ifdef DIAGNOSTIC
1790 		if ((*mp)->m_type != MT_DATA && (*mp)->m_type != MT_HEADER)
1791 			panic("somove type: so %p, so_type %d, m %p, "
1792 			    "m_type %d", so, so->so_type, *mp, (*mp)->m_type);
1793 #endif
1794 		if ((*mp)->m_len > size) {
1795 			/*
1796 			 * Move only a partial mbuf at maximum splice length or
1797 			 * if the drain buffer is too small for this large mbuf.
1798 			 */
1799 			if (!maxreached && sosp->so_snd.sb_datacc > 0) {
1800 				len -= size;
1801 				break;
1802 			}
1803 			*mp = m_copym(so->so_rcv.sb_mb, 0, size, wait);
1804 			if (*mp == NULL) {
1805 				len -= size;
1806 				break;
1807 			}
1808 			so->so_rcv.sb_mb->m_data += size;
1809 			so->so_rcv.sb_mb->m_len -= size;
1810 			so->so_rcv.sb_cc -= size;
1811 			so->so_rcv.sb_datacc -= size;
1812 		} else {
1813 			*mp = so->so_rcv.sb_mb;
1814 			sbfree(so, &so->so_rcv, *mp);
1815 			so->so_rcv.sb_mb = (*mp)->m_next;
1816 			sbsync(&so->so_rcv, nextrecord);
1817 		}
1818 	}
1819 	*mp = NULL;
1820 
1821 	SBLASTRECORDCHK(&so->so_rcv, "somove 3");
1822 	SBLASTMBUFCHK(&so->so_rcv, "somove 3");
1823 	SBCHECK(so, &so->so_rcv);
1824 	if (m == NULL)
1825 		goto release;
1826 	m->m_nextpkt = NULL;
1827 	if (m->m_flags & M_PKTHDR) {
1828 		m_resethdr(m);
1829 		m->m_pkthdr.len = len;
1830 	}
1831 
1832 	/* Send window update to source peer as receive buffer has changed. */
1833 	if (so->so_proto->pr_flags & PR_WANTRCVD) {
1834 		mtx_leave(&sosp->so_snd.sb_mtx);
1835 		mtx_leave(&so->so_rcv.sb_mtx);
1836 		pru_rcvd(so);
1837 		mtx_enter(&so->so_rcv.sb_mtx);
1838 		mtx_enter(&sosp->so_snd.sb_mtx);
1839 	}
1840 
1841 	/* Receive buffer did shrink by len bytes, adjust oob. */
1842 	rcvstate = so->so_rcv.sb_state;
1843 	so->so_rcv.sb_state &= ~SS_RCVATMARK;
1844 	oobmark = so->so_oobmark;
1845 	so->so_oobmark = oobmark > len ? oobmark - len : 0;
1846 	if (oobmark) {
1847 		if (oobmark == len)
1848 			so->so_rcv.sb_state |= SS_RCVATMARK;
1849 		if (oobmark >= len)
1850 			oobmark = 0;
1851 	}
1852 
1853 	/*
1854 	 * Handle oob data.  If any malloc fails, ignore error.
1855 	 * TCP urgent data is not very reliable anyway.
1856 	 */
1857 	while (((rcvstate & SS_RCVATMARK) || oobmark) &&
1858 	    (so->so_options & SO_OOBINLINE)) {
1859 		struct mbuf *o = NULL;
1860 
1861 		if (rcvstate & SS_RCVATMARK) {
1862 			o = m_get(wait, MT_DATA);
1863 			rcvstate &= ~SS_RCVATMARK;
1864 		} else if (oobmark) {
1865 			o = m_split(m, oobmark, wait);
1866 			if (o) {
1867 				mtx_leave(&sosp->so_snd.sb_mtx);
1868 				mtx_leave(&so->so_rcv.sb_mtx);
1869 				error = pru_send(sosp, m, NULL, NULL);
1870 				mtx_enter(&so->so_rcv.sb_mtx);
1871 				mtx_enter(&sosp->so_snd.sb_mtx);
1872 
1873 				if (error) {
1874 					if (sosp->so_snd.sb_state &
1875 					    SS_CANTSENDMORE)
1876 						error = EPIPE;
1877 					m_freem(o);
1878 					goto release;
1879 				}
1880 				len -= oobmark;
1881 				so->so_splicelen += oobmark;
1882 				m = o;
1883 				o = m_get(wait, MT_DATA);
1884 			}
1885 			oobmark = 0;
1886 		}
1887 		if (o) {
1888 			o->m_len = 1;
1889 			*mtod(o, caddr_t) = *mtod(m, caddr_t);
1890 
1891 			mtx_leave(&sosp->so_snd.sb_mtx);
1892 			mtx_leave(&so->so_rcv.sb_mtx);
1893 			error = pru_sendoob(sosp, o, NULL, NULL);
1894 			mtx_enter(&so->so_rcv.sb_mtx);
1895 			mtx_enter(&sosp->so_snd.sb_mtx);
1896 
1897 			if (error) {
1898 				if (sosp->so_snd.sb_state & SS_CANTSENDMORE)
1899 					error = EPIPE;
1900 				m_freem(m);
1901 				goto release;
1902 			}
1903 			len -= 1;
1904 			so->so_splicelen += 1;
1905 			if (oobmark) {
1906 				oobmark -= 1;
1907 				if (oobmark == 0)
1908 					rcvstate |= SS_RCVATMARK;
1909 			}
1910 			m_adj(m, 1);
1911 		}
1912 	}
1913 
1914 	/* Append all remaining data to drain socket. */
1915 	if (so->so_rcv.sb_cc == 0 || maxreached)
1916 		sosp->so_snd.sb_state &= ~SS_ISSENDING;
1917 
1918 	mtx_leave(&sosp->so_snd.sb_mtx);
1919 	mtx_leave(&so->so_rcv.sb_mtx);
1920 
1921 	if (sockdgram)
1922 		solock_shared(sosp);
1923 	error = pru_send(sosp, m, NULL, NULL);
1924 	if (sockdgram)
1925 		sounlock_shared(sosp);
1926 
1927 	mtx_enter(&so->so_rcv.sb_mtx);
1928 	mtx_enter(&sosp->so_snd.sb_mtx);
1929 
1930 	if (error) {
1931 		if (sosp->so_snd.sb_state & SS_CANTSENDMORE ||
1932 		    sosp->so_pcb == NULL)
1933 			error = EPIPE;
1934 		goto release;
1935 	}
1936 	so->so_splicelen += len;
1937 
1938 	/* Move several packets if possible. */
1939 	if (!maxreached && nextrecord)
1940 		goto nextpkt;
1941 
1942  release:
1943 	sosp->so_snd.sb_state &= ~SS_ISSENDING;
1944 
1945 	if (!error && maxreached && so->so_splicemax == so->so_splicelen)
1946 		error = EFBIG;
1947 	if (error)
1948 		WRITE_ONCE(so->so_error, error);
1949 
1950 	if (((so->so_rcv.sb_state & SS_CANTRCVMORE) &&
1951 	    so->so_rcv.sb_cc == 0) ||
1952 	    (sosp->so_snd.sb_state & SS_CANTSENDMORE) ||
1953 	    maxreached || error)
1954 		unsplice = 1;
1955 
1956 	mtx_leave(&sosp->so_snd.sb_mtx);
1957 	mtx_leave(&so->so_rcv.sb_mtx);
1958 
1959 	if (unsplice) {
1960 		if (sockdgram)
1961 			solock(so);
1962 		sounsplice(so, sosp, 0);
1963 		if (sockdgram)
1964 			sounlock(so);
1965 
1966 		return (0);
1967 	}
1968 	if (timerisset(&so->so_idletv))
1969 		timeout_add_tv(&so->so_idleto, &so->so_idletv);
1970 	return (1);
1971 }
1972 #endif /* SOCKET_SPLICE */
1973 
1974 void
1975 sorwakeup(struct socket *so)
1976 {
1977 	if ((so->so_rcv.sb_flags & SB_MTXLOCK) == 0)
1978 		soassertlocked_readonly(so);
1979 
1980 #ifdef SOCKET_SPLICE
1981 	if (so->so_proto->pr_flags & PR_SPLICE) {
1982 		sb_mtx_lock(&so->so_rcv);
1983 		if (so->so_rcv.sb_flags & SB_SPLICE)
1984 			task_add(sosplice_taskq, &so->so_splicetask);
1985 		if (isspliced(so)) {
1986 			sb_mtx_unlock(&so->so_rcv);
1987 			return;
1988 		}
1989 		sb_mtx_unlock(&so->so_rcv);
1990 	}
1991 #endif
1992 	sowakeup(so, &so->so_rcv);
1993 	if (so->so_upcall)
1994 		(*(so->so_upcall))(so, so->so_upcallarg, M_DONTWAIT);
1995 }
1996 
1997 void
1998 sowwakeup(struct socket *so)
1999 {
2000 	if ((so->so_snd.sb_flags & SB_MTXLOCK) == 0)
2001 		soassertlocked_readonly(so);
2002 
2003 #ifdef SOCKET_SPLICE
2004 	if (so->so_proto->pr_flags & PR_SPLICE) {
2005 		sb_mtx_lock(&so->so_snd);
2006 		if (so->so_snd.sb_flags & SB_SPLICE)
2007 			task_add(sosplice_taskq,
2008 			    &so->so_sp->ssp_soback->so_splicetask);
2009 		if (issplicedback(so)) {
2010 			sb_mtx_unlock(&so->so_snd);
2011 			return;
2012 		}
2013 		sb_mtx_unlock(&so->so_snd);
2014 	}
2015 #endif
2016 	sowakeup(so, &so->so_snd);
2017 }
2018 
2019 int
2020 sosetopt(struct socket *so, int level, int optname, struct mbuf *m)
2021 {
2022 	int error = 0;
2023 
2024 	if (level != SOL_SOCKET) {
2025 		if (so->so_proto->pr_ctloutput) {
2026 			solock(so);
2027 			error = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so,
2028 			    level, optname, m);
2029 			sounlock(so);
2030 			return (error);
2031 		}
2032 		error = ENOPROTOOPT;
2033 	} else {
2034 		switch (optname) {
2035 
2036 		case SO_LINGER:
2037 			if (m == NULL || m->m_len != sizeof (struct linger) ||
2038 			    mtod(m, struct linger *)->l_linger < 0 ||
2039 			    mtod(m, struct linger *)->l_linger > SHRT_MAX)
2040 				return (EINVAL);
2041 
2042 			solock(so);
2043 			so->so_linger = mtod(m, struct linger *)->l_linger;
2044 			if (*mtod(m, int *))
2045 				so->so_options |= optname;
2046 			else
2047 				so->so_options &= ~optname;
2048 			sounlock(so);
2049 
2050 			break;
2051 		case SO_BINDANY:
2052 			if ((error = suser(curproc)) != 0)	/* XXX */
2053 				return (error);
2054 			/* FALLTHROUGH */
2055 
2056 		case SO_DEBUG:
2057 		case SO_KEEPALIVE:
2058 		case SO_USELOOPBACK:
2059 		case SO_BROADCAST:
2060 		case SO_REUSEADDR:
2061 		case SO_REUSEPORT:
2062 		case SO_OOBINLINE:
2063 		case SO_TIMESTAMP:
2064 		case SO_ZEROIZE:
2065 			if (m == NULL || m->m_len < sizeof (int))
2066 				return (EINVAL);
2067 
2068 			solock(so);
2069 			if (*mtod(m, int *))
2070 				so->so_options |= optname;
2071 			else
2072 				so->so_options &= ~optname;
2073 			sounlock(so);
2074 
2075 			break;
2076 		case SO_DONTROUTE:
2077 			if (m == NULL || m->m_len < sizeof (int))
2078 				return (EINVAL);
2079 			if (*mtod(m, int *))
2080 				error = EOPNOTSUPP;
2081 			break;
2082 
2083 		case SO_SNDBUF:
2084 		case SO_RCVBUF:
2085 		case SO_SNDLOWAT:
2086 		case SO_RCVLOWAT:
2087 		    {
2088 			struct sockbuf *sb = (optname == SO_SNDBUF ||
2089 			    optname == SO_SNDLOWAT ?
2090 			    &so->so_snd : &so->so_rcv);
2091 			u_long cnt;
2092 
2093 			if (m == NULL || m->m_len < sizeof (int))
2094 				return (EINVAL);
2095 			cnt = *mtod(m, int *);
2096 			if ((long)cnt <= 0)
2097 				cnt = 1;
2098 
2099 			if (((sb->sb_flags & SB_MTXLOCK) == 0))
2100 				solock(so);
2101 			mtx_enter(&sb->sb_mtx);
2102 
2103 			switch (optname) {
2104 			case SO_SNDBUF:
2105 			case SO_RCVBUF:
2106 				if (sb->sb_state &
2107 				    (SS_CANTSENDMORE | SS_CANTRCVMORE)) {
2108 					error = EINVAL;
2109 					break;
2110 				}
2111 				if (sbcheckreserve(cnt, sb->sb_wat) ||
2112 				    sbreserve(so, sb, cnt)) {
2113 					error = ENOBUFS;
2114 					break;
2115 				}
2116 				sb->sb_wat = cnt;
2117 				break;
2118 			case SO_SNDLOWAT:
2119 			case SO_RCVLOWAT:
2120 				sb->sb_lowat = (cnt > sb->sb_hiwat) ?
2121 				    sb->sb_hiwat : cnt;
2122 				break;
2123 			}
2124 
2125 			mtx_leave(&sb->sb_mtx);
2126 			if (((sb->sb_flags & SB_MTXLOCK) == 0))
2127 				sounlock(so);
2128 
2129 			break;
2130 		    }
2131 
2132 		case SO_SNDTIMEO:
2133 		case SO_RCVTIMEO:
2134 		    {
2135 			struct sockbuf *sb = (optname == SO_SNDTIMEO ?
2136 			    &so->so_snd : &so->so_rcv);
2137 			struct timeval tv;
2138 			uint64_t nsecs;
2139 
2140 			if (m == NULL || m->m_len < sizeof (tv))
2141 				return (EINVAL);
2142 			memcpy(&tv, mtod(m, struct timeval *), sizeof tv);
2143 			if (!timerisvalid(&tv))
2144 				return (EINVAL);
2145 			nsecs = TIMEVAL_TO_NSEC(&tv);
2146 			if (nsecs == UINT64_MAX)
2147 				return (EDOM);
2148 			if (nsecs == 0)
2149 				nsecs = INFSLP;
2150 
2151 			mtx_enter(&sb->sb_mtx);
2152 			sb->sb_timeo_nsecs = nsecs;
2153 			mtx_leave(&sb->sb_mtx);
2154 			break;
2155 		    }
2156 
2157 		case SO_RTABLE:
2158 			if (so->so_proto->pr_domain &&
2159 			    so->so_proto->pr_domain->dom_protosw &&
2160 			    so->so_proto->pr_ctloutput) {
2161 				const struct domain *dom =
2162 				    so->so_proto->pr_domain;
2163 
2164 				level = dom->dom_protosw->pr_protocol;
2165 				solock(so);
2166 				error = (*so->so_proto->pr_ctloutput)
2167 				    (PRCO_SETOPT, so, level, optname, m);
2168 				sounlock(so);
2169 			} else
2170 				error = ENOPROTOOPT;
2171 			break;
2172 #ifdef SOCKET_SPLICE
2173 		case SO_SPLICE:
2174 			if (m == NULL) {
2175 				error = sosplice(so, -1, 0, NULL);
2176 			} else if (m->m_len < sizeof(int)) {
2177 				error = EINVAL;
2178 			} else if (m->m_len < sizeof(struct splice)) {
2179 				error = sosplice(so, *mtod(m, int *), 0, NULL);
2180 			} else {
2181 				error = sosplice(so,
2182 				    mtod(m, struct splice *)->sp_fd,
2183 				    mtod(m, struct splice *)->sp_max,
2184 				   &mtod(m, struct splice *)->sp_idle);
2185 			}
2186 			break;
2187 #endif /* SOCKET_SPLICE */
2188 
2189 		default:
2190 			error = ENOPROTOOPT;
2191 			break;
2192 		}
2193 	}
2194 
2195 	return (error);
2196 }
2197 
2198 int
2199 sogetopt(struct socket *so, int level, int optname, struct mbuf *m)
2200 {
2201 	int error = 0;
2202 
2203 	if (level != SOL_SOCKET) {
2204 		if (so->so_proto->pr_ctloutput) {
2205 			m->m_len = 0;
2206 
2207 			solock(so);
2208 			error = (*so->so_proto->pr_ctloutput)(PRCO_GETOPT, so,
2209 			    level, optname, m);
2210 			sounlock(so);
2211 			return (error);
2212 		} else
2213 			return (ENOPROTOOPT);
2214 	} else {
2215 		m->m_len = sizeof (int);
2216 
2217 		switch (optname) {
2218 
2219 		case SO_LINGER:
2220 			m->m_len = sizeof (struct linger);
2221 			solock_shared(so);
2222 			mtod(m, struct linger *)->l_onoff =
2223 				so->so_options & SO_LINGER;
2224 			mtod(m, struct linger *)->l_linger = so->so_linger;
2225 			sounlock_shared(so);
2226 			break;
2227 
2228 		case SO_BINDANY:
2229 		case SO_USELOOPBACK:
2230 		case SO_DEBUG:
2231 		case SO_KEEPALIVE:
2232 		case SO_REUSEADDR:
2233 		case SO_REUSEPORT:
2234 		case SO_BROADCAST:
2235 		case SO_OOBINLINE:
2236 		case SO_ACCEPTCONN:
2237 		case SO_TIMESTAMP:
2238 		case SO_ZEROIZE:
2239 			*mtod(m, int *) = so->so_options & optname;
2240 			break;
2241 
2242 		case SO_DONTROUTE:
2243 			*mtod(m, int *) = 0;
2244 			break;
2245 
2246 		case SO_TYPE:
2247 			*mtod(m, int *) = so->so_type;
2248 			break;
2249 
2250 		case SO_ERROR:
2251 			solock(so);
2252 			*mtod(m, int *) = so->so_error;
2253 			so->so_error = 0;
2254 			sounlock(so);
2255 
2256 			break;
2257 
2258 		case SO_DOMAIN:
2259 			*mtod(m, int *) = so->so_proto->pr_domain->dom_family;
2260 			break;
2261 
2262 		case SO_PROTOCOL:
2263 			*mtod(m, int *) = so->so_proto->pr_protocol;
2264 			break;
2265 
2266 		case SO_SNDBUF:
2267 			*mtod(m, int *) = so->so_snd.sb_hiwat;
2268 			break;
2269 
2270 		case SO_RCVBUF:
2271 			*mtod(m, int *) = so->so_rcv.sb_hiwat;
2272 			break;
2273 
2274 		case SO_SNDLOWAT:
2275 			*mtod(m, int *) = so->so_snd.sb_lowat;
2276 			break;
2277 
2278 		case SO_RCVLOWAT:
2279 			*mtod(m, int *) = so->so_rcv.sb_lowat;
2280 			break;
2281 
2282 		case SO_SNDTIMEO:
2283 		case SO_RCVTIMEO:
2284 		    {
2285 			struct sockbuf *sb = (optname == SO_SNDTIMEO ?
2286 			    &so->so_snd : &so->so_rcv);
2287 			struct timeval tv;
2288 			uint64_t nsecs;
2289 
2290 			mtx_enter(&sb->sb_mtx);
2291 			nsecs = sb->sb_timeo_nsecs;
2292 			mtx_leave(&sb->sb_mtx);
2293 
2294 			m->m_len = sizeof(struct timeval);
2295 			memset(&tv, 0, sizeof(tv));
2296 			if (nsecs != INFSLP)
2297 				NSEC_TO_TIMEVAL(nsecs, &tv);
2298 			memcpy(mtod(m, struct timeval *), &tv, sizeof tv);
2299 			break;
2300 		    }
2301 
2302 		case SO_RTABLE:
2303 			if (so->so_proto->pr_domain &&
2304 			    so->so_proto->pr_domain->dom_protosw &&
2305 			    so->so_proto->pr_ctloutput) {
2306 				const struct domain *dom =
2307 				    so->so_proto->pr_domain;
2308 
2309 				level = dom->dom_protosw->pr_protocol;
2310 				solock(so);
2311 				error = (*so->so_proto->pr_ctloutput)
2312 				    (PRCO_GETOPT, so, level, optname, m);
2313 				sounlock(so);
2314 				if (error)
2315 					return (error);
2316 				break;
2317 			}
2318 			return (ENOPROTOOPT);
2319 
2320 #ifdef SOCKET_SPLICE
2321 		case SO_SPLICE:
2322 		    {
2323 			off_t len;
2324 
2325 			m->m_len = sizeof(off_t);
2326 			solock_shared(so);
2327 			len = so->so_sp ? so->so_sp->ssp_len : 0;
2328 			sounlock_shared(so);
2329 			memcpy(mtod(m, off_t *), &len, sizeof(off_t));
2330 			break;
2331 		    }
2332 #endif /* SOCKET_SPLICE */
2333 
2334 		case SO_PEERCRED:
2335 			if (so->so_proto->pr_protocol == AF_UNIX) {
2336 				struct unpcb *unp = sotounpcb(so);
2337 
2338 				solock(so);
2339 				if (unp->unp_flags & UNP_FEIDS) {
2340 					m->m_len = sizeof(unp->unp_connid);
2341 					memcpy(mtod(m, caddr_t),
2342 					    &(unp->unp_connid), m->m_len);
2343 					sounlock(so);
2344 					break;
2345 				}
2346 				sounlock(so);
2347 
2348 				return (ENOTCONN);
2349 			}
2350 			return (EOPNOTSUPP);
2351 
2352 		default:
2353 			return (ENOPROTOOPT);
2354 		}
2355 		return (0);
2356 	}
2357 }
2358 
2359 void
2360 sohasoutofband(struct socket *so)
2361 {
2362 	pgsigio(&so->so_sigio, SIGURG, 0);
2363 	knote(&so->so_rcv.sb_klist, 0);
2364 }
2365 
2366 void
2367 sofilt_lock(struct socket *so, struct sockbuf *sb)
2368 {
2369 	switch (so->so_proto->pr_domain->dom_family) {
2370 	case PF_INET:
2371 	case PF_INET6:
2372 		NET_LOCK_SHARED();
2373 		break;
2374 	default:
2375 		rw_enter_write(&so->so_lock);
2376 		break;
2377 	}
2378 
2379 	mtx_enter(&sb->sb_mtx);
2380 }
2381 
2382 void
2383 sofilt_unlock(struct socket *so, struct sockbuf *sb)
2384 {
2385 	mtx_leave(&sb->sb_mtx);
2386 
2387 	switch (so->so_proto->pr_domain->dom_family) {
2388 	case PF_INET:
2389 	case PF_INET6:
2390 		NET_UNLOCK_SHARED();
2391 		break;
2392 	default:
2393 		rw_exit_write(&so->so_lock);
2394 		break;
2395 	}
2396 }
2397 
2398 int
2399 soo_kqfilter(struct file *fp, struct knote *kn)
2400 {
2401 	struct socket *so = kn->kn_fp->f_data;
2402 	struct sockbuf *sb;
2403 
2404 	switch (kn->kn_filter) {
2405 	case EVFILT_READ:
2406 		kn->kn_fop = &soread_filtops;
2407 		sb = &so->so_rcv;
2408 		break;
2409 	case EVFILT_WRITE:
2410 		kn->kn_fop = &sowrite_filtops;
2411 		sb = &so->so_snd;
2412 		break;
2413 	case EVFILT_EXCEPT:
2414 		kn->kn_fop = &soexcept_filtops;
2415 		sb = &so->so_rcv;
2416 		break;
2417 	default:
2418 		return (EINVAL);
2419 	}
2420 
2421 	klist_insert(&sb->sb_klist, kn);
2422 
2423 	return (0);
2424 }
2425 
2426 void
2427 filt_sordetach(struct knote *kn)
2428 {
2429 	struct socket *so = kn->kn_fp->f_data;
2430 
2431 	klist_remove(&so->so_rcv.sb_klist, kn);
2432 }
2433 
2434 int
2435 filt_soread(struct knote *kn, long hint)
2436 {
2437 	struct socket *so = kn->kn_fp->f_data;
2438 	int rv = 0;
2439 
2440 	MUTEX_ASSERT_LOCKED(&so->so_rcv.sb_mtx);
2441 	if ((so->so_rcv.sb_flags & SB_MTXLOCK) == 0)
2442 		soassertlocked_readonly(so);
2443 
2444 	if (so->so_options & SO_ACCEPTCONN) {
2445 		if (so->so_rcv.sb_flags & SB_MTXLOCK)
2446 			soassertlocked_readonly(so);
2447 
2448 		kn->kn_data = so->so_qlen;
2449 		rv = (kn->kn_data != 0);
2450 
2451 		if (kn->kn_flags & (__EV_POLL | __EV_SELECT)) {
2452 			if (so->so_state & SS_ISDISCONNECTED) {
2453 				kn->kn_flags |= __EV_HUP;
2454 				rv = 1;
2455 			} else {
2456 				rv = soreadable(so);
2457 			}
2458 		}
2459 
2460 		return rv;
2461 	}
2462 
2463 	kn->kn_data = so->so_rcv.sb_cc;
2464 #ifdef SOCKET_SPLICE
2465 	if (isspliced(so)) {
2466 		rv = 0;
2467 	} else
2468 #endif /* SOCKET_SPLICE */
2469 	if (so->so_rcv.sb_state & SS_CANTRCVMORE) {
2470 		kn->kn_flags |= EV_EOF;
2471 		if (kn->kn_flags & __EV_POLL) {
2472 			if (so->so_state & SS_ISDISCONNECTED)
2473 				kn->kn_flags |= __EV_HUP;
2474 		}
2475 		kn->kn_fflags = so->so_error;
2476 		rv = 1;
2477 	} else if (so->so_error) {
2478 		rv = 1;
2479 	} else if (kn->kn_sfflags & NOTE_LOWAT) {
2480 		rv = (kn->kn_data >= kn->kn_sdata);
2481 	} else {
2482 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
2483 	}
2484 
2485 	return rv;
2486 }
2487 
2488 void
2489 filt_sowdetach(struct knote *kn)
2490 {
2491 	struct socket *so = kn->kn_fp->f_data;
2492 
2493 	klist_remove(&so->so_snd.sb_klist, kn);
2494 }
2495 
2496 int
2497 filt_sowrite(struct knote *kn, long hint)
2498 {
2499 	struct socket *so = kn->kn_fp->f_data;
2500 	int rv;
2501 
2502 	MUTEX_ASSERT_LOCKED(&so->so_snd.sb_mtx);
2503 	if ((so->so_snd.sb_flags & SB_MTXLOCK) == 0)
2504 		soassertlocked_readonly(so);
2505 
2506 	kn->kn_data = sbspace_locked(so, &so->so_snd);
2507 	if (so->so_snd.sb_state & SS_CANTSENDMORE) {
2508 		kn->kn_flags |= EV_EOF;
2509 		if (kn->kn_flags & __EV_POLL) {
2510 			if (so->so_state & SS_ISDISCONNECTED)
2511 				kn->kn_flags |= __EV_HUP;
2512 		}
2513 		kn->kn_fflags = so->so_error;
2514 		rv = 1;
2515 	} else if (so->so_error) {
2516 		rv = 1;
2517 	} else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2518 	    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
2519 		rv = 0;
2520 	} else if (kn->kn_sfflags & NOTE_LOWAT) {
2521 		rv = (kn->kn_data >= kn->kn_sdata);
2522 	} else {
2523 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
2524 	}
2525 
2526 	return (rv);
2527 }
2528 
2529 int
2530 filt_soexcept(struct knote *kn, long hint)
2531 {
2532 	struct socket *so = kn->kn_fp->f_data;
2533 	int rv = 0;
2534 
2535 	MUTEX_ASSERT_LOCKED(&so->so_rcv.sb_mtx);
2536 	if ((so->so_rcv.sb_flags & SB_MTXLOCK) == 0)
2537 		soassertlocked_readonly(so);
2538 
2539 #ifdef SOCKET_SPLICE
2540 	if (isspliced(so)) {
2541 		rv = 0;
2542 	} else
2543 #endif /* SOCKET_SPLICE */
2544 	if (kn->kn_sfflags & NOTE_OOB) {
2545 		if (so->so_oobmark || (so->so_rcv.sb_state & SS_RCVATMARK)) {
2546 			kn->kn_fflags |= NOTE_OOB;
2547 			kn->kn_data -= so->so_oobmark;
2548 			rv = 1;
2549 		}
2550 	}
2551 
2552 	if (kn->kn_flags & __EV_POLL) {
2553 		if (so->so_state & SS_ISDISCONNECTED) {
2554 			kn->kn_flags |= __EV_HUP;
2555 			rv = 1;
2556 		}
2557 	}
2558 
2559 	return rv;
2560 }
2561 
2562 int
2563 filt_sowmodify(struct kevent *kev, struct knote *kn)
2564 {
2565 	struct socket *so = kn->kn_fp->f_data;
2566 	int rv;
2567 
2568 	sofilt_lock(so, &so->so_snd);
2569 	rv = knote_modify(kev, kn);
2570 	sofilt_unlock(so, &so->so_snd);
2571 
2572 	return (rv);
2573 }
2574 
2575 int
2576 filt_sowprocess(struct knote *kn, struct kevent *kev)
2577 {
2578 	struct socket *so = kn->kn_fp->f_data;
2579 	int rv;
2580 
2581 	sofilt_lock(so, &so->so_snd);
2582 	rv = knote_process(kn, kev);
2583 	sofilt_unlock(so, &so->so_snd);
2584 
2585 	return (rv);
2586 }
2587 
2588 int
2589 filt_sormodify(struct kevent *kev, struct knote *kn)
2590 {
2591 	struct socket *so = kn->kn_fp->f_data;
2592 	int rv;
2593 
2594 	sofilt_lock(so, &so->so_rcv);
2595 	rv = knote_modify(kev, kn);
2596 	sofilt_unlock(so, &so->so_rcv);
2597 
2598 	return (rv);
2599 }
2600 
2601 int
2602 filt_sorprocess(struct knote *kn, struct kevent *kev)
2603 {
2604 	struct socket *so = kn->kn_fp->f_data;
2605 	int rv;
2606 
2607 	sofilt_lock(so, &so->so_rcv);
2608 	rv = knote_process(kn, kev);
2609 	sofilt_unlock(so, &so->so_rcv);
2610 
2611 	return (rv);
2612 }
2613 
2614 #ifdef DDB
2615 void
2616 sobuf_print(struct sockbuf *,
2617     int (*)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))));
2618 
2619 void
2620 sobuf_print(struct sockbuf *sb,
2621     int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2622 {
2623 	(*pr)("\tsb_cc: %lu\n", sb->sb_cc);
2624 	(*pr)("\tsb_datacc: %lu\n", sb->sb_datacc);
2625 	(*pr)("\tsb_hiwat: %lu\n", sb->sb_hiwat);
2626 	(*pr)("\tsb_wat: %lu\n", sb->sb_wat);
2627 	(*pr)("\tsb_mbcnt: %lu\n", sb->sb_mbcnt);
2628 	(*pr)("\tsb_mbmax: %lu\n", sb->sb_mbmax);
2629 	(*pr)("\tsb_lowat: %ld\n", sb->sb_lowat);
2630 	(*pr)("\tsb_mb: %p\n", sb->sb_mb);
2631 	(*pr)("\tsb_mbtail: %p\n", sb->sb_mbtail);
2632 	(*pr)("\tsb_lastrecord: %p\n", sb->sb_lastrecord);
2633 	(*pr)("\tsb_flags: %04x\n", sb->sb_flags);
2634 	(*pr)("\tsb_state: %04x\n", sb->sb_state);
2635 	(*pr)("\tsb_timeo_nsecs: %llu\n", sb->sb_timeo_nsecs);
2636 }
2637 
2638 void
2639 so_print(void *v,
2640     int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2641 {
2642 	struct socket *so = v;
2643 
2644 	(*pr)("socket %p\n", so);
2645 	(*pr)("so_type: %i\n", so->so_type);
2646 	(*pr)("so_options: 0x%04x\n", so->so_options); /* %b */
2647 	(*pr)("so_linger: %i\n", so->so_linger);
2648 	(*pr)("so_state: 0x%04x\n", so->so_state);
2649 	(*pr)("so_pcb: %p\n", so->so_pcb);
2650 	(*pr)("so_proto: %p\n", so->so_proto);
2651 	(*pr)("so_sigio: %p\n", so->so_sigio.sir_sigio);
2652 
2653 	(*pr)("so_head: %p\n", so->so_head);
2654 	(*pr)("so_onq: %p\n", so->so_onq);
2655 	(*pr)("so_q0: @%p first: %p\n", &so->so_q0, TAILQ_FIRST(&so->so_q0));
2656 	(*pr)("so_q: @%p first: %p\n", &so->so_q, TAILQ_FIRST(&so->so_q));
2657 	(*pr)("so_eq: next: %p\n", TAILQ_NEXT(so, so_qe));
2658 	(*pr)("so_q0len: %i\n", so->so_q0len);
2659 	(*pr)("so_qlen: %i\n", so->so_qlen);
2660 	(*pr)("so_qlimit: %i\n", so->so_qlimit);
2661 	(*pr)("so_timeo: %i\n", so->so_timeo);
2662 	(*pr)("so_obmark: %lu\n", so->so_oobmark);
2663 
2664 	(*pr)("so_sp: %p\n", so->so_sp);
2665 	if (so->so_sp != NULL) {
2666 		(*pr)("\tssp_socket: %p\n", so->so_sp->ssp_socket);
2667 		(*pr)("\tssp_soback: %p\n", so->so_sp->ssp_soback);
2668 		(*pr)("\tssp_len: %lld\n",
2669 		    (unsigned long long)so->so_sp->ssp_len);
2670 		(*pr)("\tssp_max: %lld\n",
2671 		    (unsigned long long)so->so_sp->ssp_max);
2672 		(*pr)("\tssp_idletv: %lld %ld\n", so->so_sp->ssp_idletv.tv_sec,
2673 		    so->so_sp->ssp_idletv.tv_usec);
2674 		(*pr)("\tssp_idleto: %spending (@%i)\n",
2675 		    timeout_pending(&so->so_sp->ssp_idleto) ? "" : "not ",
2676 		    so->so_sp->ssp_idleto.to_time);
2677 	}
2678 
2679 	(*pr)("so_rcv:\n");
2680 	sobuf_print(&so->so_rcv, pr);
2681 	(*pr)("so_snd:\n");
2682 	sobuf_print(&so->so_snd, pr);
2683 
2684 	(*pr)("so_upcall: %p so_upcallarg: %p\n",
2685 	    so->so_upcall, so->so_upcallarg);
2686 
2687 	(*pr)("so_euid: %d so_ruid: %d\n", so->so_euid, so->so_ruid);
2688 	(*pr)("so_egid: %d so_rgid: %d\n", so->so_egid, so->so_rgid);
2689 	(*pr)("so_cpid: %d\n", so->so_cpid);
2690 }
2691 #endif
2692