xref: /openbsd-src/sys/kern/uipc_socket.c (revision a089aaec06fed3a1e33b342766ffe977cfbb4c76)
1 /*	$OpenBSD: uipc_socket.c,v 1.350 2024/12/27 13:08:11 mvs 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 			/* FALLTHROUGH */
164 		case SOCK_STREAM:
165 			so->so_rcv.sb_flags |= SB_MTXLOCK;
166 			break;
167 		}
168 		break;
169 	case AF_KEY:
170 	case AF_ROUTE:
171 	case AF_UNIX:
172 	case AF_FRAME:
173 		so->so_snd.sb_flags |= SB_MTXLOCK;
174 		so->so_rcv.sb_flags |= SB_MTXLOCK;
175 		break;
176 	}
177 
178 	return (so);
179 }
180 
181 /*
182  * Socket operation routines.
183  * These routines are called by the routines in
184  * sys_socket.c or from a system process, and
185  * implement the semantics of socket operations by
186  * switching out to the protocol specific routines.
187  */
188 int
189 socreate(int dom, struct socket **aso, int type, int proto)
190 {
191 	struct proc *p = curproc;		/* XXX */
192 	const struct protosw *prp;
193 	struct socket *so;
194 	int error;
195 
196 	if (proto)
197 		prp = pffindproto(dom, proto, type);
198 	else
199 		prp = pffindtype(dom, type);
200 	if (prp == NULL || prp->pr_usrreqs == NULL)
201 		return (EPROTONOSUPPORT);
202 	if (prp->pr_type != type)
203 		return (EPROTOTYPE);
204 	so = soalloc(prp, M_WAIT);
205 	so->so_type = type;
206 	if (suser(p) == 0)
207 		so->so_state = SS_PRIV;
208 	so->so_ruid = p->p_ucred->cr_ruid;
209 	so->so_euid = p->p_ucred->cr_uid;
210 	so->so_rgid = p->p_ucred->cr_rgid;
211 	so->so_egid = p->p_ucred->cr_gid;
212 	so->so_cpid = p->p_p->ps_pid;
213 	so->so_proto = prp;
214 	so->so_snd.sb_timeo_nsecs = INFSLP;
215 	so->so_rcv.sb_timeo_nsecs = INFSLP;
216 
217 	solock(so);
218 	error = pru_attach(so, proto, M_WAIT);
219 	if (error) {
220 		so->so_state |= SS_NOFDREF;
221 		/* sofree() calls sounlock(). */
222 		sofree(so, 0);
223 		return (error);
224 	}
225 	sounlock(so);
226 	*aso = so;
227 	return (0);
228 }
229 
230 int
231 sobind(struct socket *so, struct mbuf *nam, struct proc *p)
232 {
233 	soassertlocked(so);
234 	return pru_bind(so, nam, p);
235 }
236 
237 int
238 solisten(struct socket *so, int backlog)
239 {
240 	int somaxconn_local = atomic_load_int(&somaxconn);
241 	int sominconn_local = atomic_load_int(&sominconn);
242 	int error;
243 
244 	switch (so->so_type) {
245 	case SOCK_STREAM:
246 	case SOCK_SEQPACKET:
247 		break;
248 	default:
249 		return (EOPNOTSUPP);
250 	}
251 
252 	soassertlocked(so);
253 
254 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING|SS_ISDISCONNECTING))
255 		return (EINVAL);
256 #ifdef SOCKET_SPLICE
257 	if (isspliced(so) || issplicedback(so))
258 		return (EOPNOTSUPP);
259 #endif /* SOCKET_SPLICE */
260 	error = pru_listen(so);
261 	if (error)
262 		return (error);
263 	if (TAILQ_FIRST(&so->so_q) == NULL)
264 		so->so_options |= SO_ACCEPTCONN;
265 	if (backlog < 0 || backlog > somaxconn_local)
266 		backlog = somaxconn_local;
267 	if (backlog < sominconn_local)
268 		backlog = sominconn_local;
269 	so->so_qlimit = backlog;
270 	return (0);
271 }
272 
273 #define SOSP_FREEING_READ	1
274 #define SOSP_FREEING_WRITE	2
275 void
276 sofree(struct socket *so, int keep_lock)
277 {
278 	int persocket = solock_persocket(so);
279 
280 	soassertlocked(so);
281 
282 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
283 		if (!keep_lock)
284 			sounlock(so);
285 		return;
286 	}
287 	if (so->so_head) {
288 		struct socket *head = so->so_head;
289 
290 		/*
291 		 * We must not decommission a socket that's on the accept(2)
292 		 * queue.  If we do, then accept(2) may hang after select(2)
293 		 * indicated that the listening socket was ready.
294 		 */
295 		if (so->so_onq == &head->so_q) {
296 			if (!keep_lock)
297 				sounlock(so);
298 			return;
299 		}
300 
301 		if (persocket) {
302 			/*
303 			 * Concurrent close of `head' could
304 			 * abort `so' due to re-lock.
305 			 */
306 			soref(so);
307 			soref(head);
308 			sounlock(so);
309 			solock(head);
310 			solock(so);
311 
312 			if (so->so_onq != &head->so_q0) {
313 				sounlock(head);
314 				sounlock(so);
315 				sorele(head);
316 				sorele(so);
317 				return;
318 			}
319 
320 			sorele(head);
321 			sorele(so);
322 		}
323 
324 		soqremque(so, 0);
325 
326 		if (persocket)
327 			sounlock(head);
328 	}
329 
330 	switch (so->so_proto->pr_domain->dom_family) {
331 	case AF_INET:
332 	case AF_INET6:
333 		if (so->so_proto->pr_type == SOCK_STREAM)
334 			break;
335 		/* FALLTHROUGH */
336 	default:
337 		sounlock(so);
338 		refcnt_finalize(&so->so_refcnt, "sofinal");
339 		solock(so);
340 		break;
341 	}
342 
343 	sigio_free(&so->so_sigio);
344 	klist_free(&so->so_rcv.sb_klist);
345 	klist_free(&so->so_snd.sb_klist);
346 
347 	mtx_enter(&so->so_snd.sb_mtx);
348 	sbrelease(so, &so->so_snd);
349 	mtx_leave(&so->so_snd.sb_mtx);
350 
351 	/*
352 	 * Unlocked dispose and cleanup is safe. Socket is unlinked
353 	 * from everywhere. Even concurrent sotask() thread will not
354 	 * call somove().
355 	 */
356 	if (so->so_proto->pr_flags & PR_RIGHTS &&
357 	    so->so_proto->pr_domain->dom_dispose)
358 		(*so->so_proto->pr_domain->dom_dispose)(so->so_rcv.sb_mb);
359 	m_purge(so->so_rcv.sb_mb);
360 
361 	if (!keep_lock)
362 		sounlock(so);
363 
364 #ifdef SOCKET_SPLICE
365 	if (so->so_sp) {
366 		/* Reuse splice idle, sounsplice() has been called before. */
367 		timeout_set_flags(&so->so_sp->ssp_idleto, soreaper, so,
368 		    KCLOCK_NONE, TIMEOUT_PROC | TIMEOUT_MPSAFE);
369 		timeout_add(&so->so_sp->ssp_idleto, 0);
370 	} else
371 #endif /* SOCKET_SPLICE */
372 	{
373 		pool_put(&socket_pool, so);
374 	}
375 }
376 
377 static inline uint64_t
378 solinger_nsec(struct socket *so)
379 {
380 	if (so->so_linger == 0)
381 		return INFSLP;
382 
383 	return SEC_TO_NSEC(so->so_linger);
384 }
385 
386 /*
387  * Close a socket on last file table reference removal.
388  * Initiate disconnect if connected.
389  * Free socket when disconnect complete.
390  */
391 int
392 soclose(struct socket *so, int flags)
393 {
394 	struct socket *so2;
395 	int error = 0;
396 
397 	solock(so);
398 	/* Revoke async IO early. There is a final revocation in sofree(). */
399 	sigio_free(&so->so_sigio);
400 	if (so->so_state & SS_ISCONNECTED) {
401 		if (so->so_pcb == NULL)
402 			goto discard;
403 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
404 			error = sodisconnect(so);
405 			if (error)
406 				goto drop;
407 		}
408 		if (so->so_options & SO_LINGER) {
409 			if ((so->so_state & SS_ISDISCONNECTING) &&
410 			    (flags & MSG_DONTWAIT))
411 				goto drop;
412 			while (so->so_state & SS_ISCONNECTED) {
413 				error = sosleep_nsec(so, &so->so_timeo,
414 				    PSOCK | PCATCH, "netcls",
415 				    solinger_nsec(so));
416 				if (error)
417 					break;
418 			}
419 		}
420 	}
421 drop:
422 	if (so->so_pcb) {
423 		int error2;
424 		error2 = pru_detach(so);
425 		if (error == 0)
426 			error = error2;
427 	}
428 	if (so->so_options & SO_ACCEPTCONN) {
429 		int persocket = solock_persocket(so);
430 
431 		while ((so2 = TAILQ_FIRST(&so->so_q0)) != NULL) {
432 			if (persocket)
433 				solock(so2);
434 			(void) soqremque(so2, 0);
435 			if (persocket)
436 				sounlock(so);
437 			soabort(so2);
438 			if (persocket)
439 				solock(so);
440 		}
441 		while ((so2 = TAILQ_FIRST(&so->so_q)) != NULL) {
442 			if (persocket)
443 				solock(so2);
444 			(void) soqremque(so2, 1);
445 			if (persocket)
446 				sounlock(so);
447 			soabort(so2);
448 			if (persocket)
449 				solock(so);
450 		}
451 	}
452 discard:
453 	if (so->so_state & SS_NOFDREF)
454 		panic("soclose NOFDREF: so %p, so_type %d", so, so->so_type);
455 	so->so_state |= SS_NOFDREF;
456 
457 #ifdef SOCKET_SPLICE
458 	if (so->so_sp) {
459 		struct socket *soback;
460 
461 		if (so->so_proto->pr_flags & PR_WANTRCVD) {
462 			/*
463 			 * Copy - Paste, but can't relock and sleep in
464 			 * sofree() in tcp(4) case. That's why tcp(4)
465 			 * still rely on solock() for splicing and
466 			 * unsplicing.
467 			 */
468 
469 			if (issplicedback(so)) {
470 				int freeing = SOSP_FREEING_WRITE;
471 
472 				if (so->so_sp->ssp_soback == so)
473 					freeing |= SOSP_FREEING_READ;
474 				sounsplice(so->so_sp->ssp_soback, so, freeing);
475 			}
476 			if (isspliced(so)) {
477 				int freeing = SOSP_FREEING_READ;
478 
479 				if (so == so->so_sp->ssp_socket)
480 					freeing |= SOSP_FREEING_WRITE;
481 				sounsplice(so, so->so_sp->ssp_socket, freeing);
482 			}
483 			goto free;
484 		}
485 
486 		sounlock(so);
487 		mtx_enter(&so->so_snd.sb_mtx);
488 		/*
489 		 * Concurrent sounsplice() locks `sb_mtx' mutexes on
490 		 * both `so_snd' and `so_rcv' before unsplice sockets.
491 		 */
492 		if ((soback = so->so_sp->ssp_soback) == NULL) {
493 			mtx_leave(&so->so_snd.sb_mtx);
494 			goto notsplicedback;
495 		}
496 		soref(soback);
497 		mtx_leave(&so->so_snd.sb_mtx);
498 
499 		/*
500 		 * `so' can be only unspliced, and never spliced again.
501 		 * Thus if issplicedback(so) check is positive, socket is
502 		 * still spliced and `ssp_soback' points to the same
503 		 * socket that `soback'.
504 		 */
505 		sblock(&soback->so_rcv, SBL_WAIT | SBL_NOINTR);
506 		if (issplicedback(so)) {
507 			int freeing = SOSP_FREEING_WRITE;
508 
509 			if (so->so_sp->ssp_soback == so)
510 				freeing |= SOSP_FREEING_READ;
511 			solock(soback);
512 			sounsplice(so->so_sp->ssp_soback, so, freeing);
513 			sounlock(soback);
514 		}
515 		sbunlock(&soback->so_rcv);
516 		sorele(soback);
517 
518 notsplicedback:
519 		sblock(&so->so_rcv, SBL_WAIT | SBL_NOINTR);
520 		if (isspliced(so)) {
521 			int freeing = SOSP_FREEING_READ;
522 
523 			if (so == so->so_sp->ssp_socket)
524 				freeing |= SOSP_FREEING_WRITE;
525 			solock(so);
526 			sounsplice(so, so->so_sp->ssp_socket, freeing);
527 			sounlock(so);
528 		}
529 		sbunlock(&so->so_rcv);
530 
531 		solock(so);
532 	}
533 free:
534 #endif /* SOCKET_SPLICE */
535 	/* sofree() calls sounlock(). */
536 	sofree(so, 0);
537 	return (error);
538 }
539 
540 void
541 soabort(struct socket *so)
542 {
543 	soassertlocked(so);
544 	pru_abort(so);
545 }
546 
547 int
548 soaccept(struct socket *so, struct mbuf *nam)
549 {
550 	int error = 0;
551 
552 	soassertlocked(so);
553 
554 	if ((so->so_state & SS_NOFDREF) == 0)
555 		panic("soaccept !NOFDREF: so %p, so_type %d", so, so->so_type);
556 	so->so_state &= ~SS_NOFDREF;
557 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
558 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
559 		error = pru_accept(so, nam);
560 	else
561 		error = ECONNABORTED;
562 	return (error);
563 }
564 
565 int
566 soconnect(struct socket *so, struct mbuf *nam)
567 {
568 	int error;
569 
570 	soassertlocked(so);
571 
572 	if (so->so_options & SO_ACCEPTCONN)
573 		return (EOPNOTSUPP);
574 	/*
575 	 * If protocol is connection-based, can only connect once.
576 	 * Otherwise, if connected, try to disconnect first.
577 	 * This allows user to disconnect by connecting to, e.g.,
578 	 * a null address.
579 	 */
580 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
581 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
582 	    (error = sodisconnect(so))))
583 		error = EISCONN;
584 	else
585 		error = pru_connect(so, nam);
586 	return (error);
587 }
588 
589 int
590 soconnect2(struct socket *so1, struct socket *so2)
591 {
592 	int persocket, error;
593 
594 	if ((persocket = solock_persocket(so1)))
595 		solock_pair(so1, so2);
596 	else
597 		solock(so1);
598 
599 	error = pru_connect2(so1, so2);
600 
601 	if (persocket)
602 		sounlock(so2);
603 	sounlock(so1);
604 	return (error);
605 }
606 
607 int
608 sodisconnect(struct socket *so)
609 {
610 	int error;
611 
612 	soassertlocked(so);
613 
614 	if ((so->so_state & SS_ISCONNECTED) == 0)
615 		return (ENOTCONN);
616 	if (so->so_state & SS_ISDISCONNECTING)
617 		return (EALREADY);
618 	error = pru_disconnect(so);
619 	return (error);
620 }
621 
622 int m_getuio(struct mbuf **, int, long, struct uio *);
623 
624 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
625 /*
626  * Send on a socket.
627  * If send must go all at once and message is larger than
628  * send buffering, then hard error.
629  * Lock against other senders.
630  * If must go all at once and not enough room now, then
631  * inform user that this would block and do nothing.
632  * Otherwise, if nonblocking, send as much as possible.
633  * The data to be sent is described by "uio" if nonzero,
634  * otherwise by the mbuf chain "top" (which must be null
635  * if uio is not).  Data provided in mbuf chain must be small
636  * enough to send all at once.
637  *
638  * Returns nonzero on error, timeout or signal; callers
639  * must check for short counts if EINTR/ERESTART are returned.
640  * Data and control buffers are freed on return.
641  */
642 int
643 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
644     struct mbuf *control, int flags)
645 {
646 	long space, clen = 0;
647 	size_t resid;
648 	int error;
649 	int atomic = sosendallatonce(so) || top;
650 	int dosolock = ((so->so_snd.sb_flags & SB_MTXLOCK) == 0);
651 
652 	if (uio)
653 		resid = uio->uio_resid;
654 	else
655 		resid = top->m_pkthdr.len;
656 	/* MSG_EOR on a SOCK_STREAM socket is invalid. */
657 	if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
658 		m_freem(top);
659 		m_freem(control);
660 		return (EINVAL);
661 	}
662 	if (uio && uio->uio_procp)
663 		uio->uio_procp->p_ru.ru_msgsnd++;
664 	if (control) {
665 		/*
666 		 * In theory clen should be unsigned (since control->m_len is).
667 		 * However, space must be signed, as it might be less than 0
668 		 * if we over-committed, and we must use a signed comparison
669 		 * of space and clen.
670 		 */
671 		clen = control->m_len;
672 		/* reserve extra space for AF_UNIX's internalize */
673 		if (so->so_proto->pr_domain->dom_family == AF_UNIX &&
674 		    clen >= CMSG_ALIGN(sizeof(struct cmsghdr)) &&
675 		    mtod(control, struct cmsghdr *)->cmsg_type == SCM_RIGHTS)
676 			clen = CMSG_SPACE(
677 			    (clen - CMSG_ALIGN(sizeof(struct cmsghdr))) *
678 			    (sizeof(struct fdpass) / sizeof(int)));
679 	}
680 
681 #define	snderr(errno)	{ error = errno; goto release; }
682 
683 restart:
684 	if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
685 		goto out;
686 	if (dosolock)
687 		solock_shared(so);
688 	sb_mtx_lock(&so->so_snd);
689 	so->so_snd.sb_state |= SS_ISSENDING;
690 	do {
691 		if (so->so_snd.sb_state & SS_CANTSENDMORE)
692 			snderr(EPIPE);
693 		if ((error = READ_ONCE(so->so_error))) {
694 			so->so_error = 0;
695 			snderr(error);
696 		}
697 		if ((so->so_state & SS_ISCONNECTED) == 0) {
698 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
699 				if (!(resid == 0 && clen != 0))
700 					snderr(ENOTCONN);
701 			} else if (addr == NULL)
702 				snderr(EDESTADDRREQ);
703 		}
704 		space = sbspace_locked(so, &so->so_snd);
705 		if (flags & MSG_OOB)
706 			space += 1024;
707 		if (so->so_proto->pr_domain->dom_family == AF_UNIX) {
708 			if (atomic && resid > so->so_snd.sb_hiwat)
709 				snderr(EMSGSIZE);
710 		} else {
711 			if (clen > so->so_snd.sb_hiwat ||
712 			    (atomic && resid > so->so_snd.sb_hiwat - clen))
713 				snderr(EMSGSIZE);
714 		}
715 		if (space < clen ||
716 		    (space - clen < resid &&
717 		    (atomic || space < so->so_snd.sb_lowat))) {
718 			if (flags & MSG_DONTWAIT)
719 				snderr(EWOULDBLOCK);
720 			sbunlock(&so->so_snd);
721 			error = sbwait(so, &so->so_snd);
722 			so->so_snd.sb_state &= ~SS_ISSENDING;
723 			sb_mtx_unlock(&so->so_snd);
724 			if (dosolock)
725 				sounlock_shared(so);
726 			if (error)
727 				goto out;
728 			goto restart;
729 		}
730 		space -= clen;
731 		do {
732 			if (uio == NULL) {
733 				/*
734 				 * Data is prepackaged in "top".
735 				 */
736 				resid = 0;
737 				if (flags & MSG_EOR)
738 					top->m_flags |= M_EOR;
739 			} else {
740 				sb_mtx_unlock(&so->so_snd);
741 				if (dosolock)
742 					sounlock_shared(so);
743 				error = m_getuio(&top, atomic, space, uio);
744 				if (dosolock)
745 					solock_shared(so);
746 				sb_mtx_lock(&so->so_snd);
747 				if (error)
748 					goto release;
749 				space -= top->m_pkthdr.len;
750 				resid = uio->uio_resid;
751 				if (flags & MSG_EOR)
752 					top->m_flags |= M_EOR;
753 			}
754 			if (resid == 0)
755 				so->so_snd.sb_state &= ~SS_ISSENDING;
756 			if (top && so->so_options & SO_ZEROIZE)
757 				top->m_flags |= M_ZEROIZE;
758 			sb_mtx_unlock(&so->so_snd);
759 			if (!dosolock)
760 				solock_shared(so);
761 			if (flags & MSG_OOB)
762 				error = pru_sendoob(so, top, addr, control);
763 			else
764 				error = pru_send(so, top, addr, control);
765 			if (!dosolock)
766 				sounlock_shared(so);
767 			sb_mtx_lock(&so->so_snd);
768 			clen = 0;
769 			control = NULL;
770 			top = NULL;
771 			if (error)
772 				goto release;
773 		} while (resid && space > 0);
774 	} while (resid);
775 
776 release:
777 	so->so_snd.sb_state &= ~SS_ISSENDING;
778 	sb_mtx_unlock(&so->so_snd);
779 	if (dosolock)
780 		sounlock_shared(so);
781 	sbunlock(&so->so_snd);
782 out:
783 	m_freem(top);
784 	m_freem(control);
785 	return (error);
786 }
787 
788 int
789 m_getuio(struct mbuf **mp, int atomic, long space, struct uio *uio)
790 {
791 	struct mbuf *m, *top = NULL;
792 	struct mbuf **nextp = &top;
793 	u_long len, mlen;
794 	size_t resid = uio->uio_resid;
795 	int error;
796 
797 	do {
798 		if (top == NULL) {
799 			MGETHDR(m, M_WAIT, MT_DATA);
800 			mlen = MHLEN;
801 		} else {
802 			MGET(m, M_WAIT, MT_DATA);
803 			mlen = MLEN;
804 		}
805 		/* chain mbuf together */
806 		*nextp = m;
807 		nextp = &m->m_next;
808 
809 		resid = ulmin(resid, space);
810 		if (resid >= MINCLSIZE) {
811 			MCLGETL(m, M_NOWAIT, ulmin(resid, MAXMCLBYTES));
812 			if ((m->m_flags & M_EXT) == 0)
813 				MCLGETL(m, M_NOWAIT, MCLBYTES);
814 			if ((m->m_flags & M_EXT) == 0)
815 				goto nopages;
816 			mlen = m->m_ext.ext_size;
817 			len = ulmin(mlen, resid);
818 			/*
819 			 * For datagram protocols, leave room
820 			 * for protocol headers in first mbuf.
821 			 */
822 			if (atomic && m == top && len < mlen - max_hdr)
823 				m->m_data += max_hdr;
824 		} else {
825 nopages:
826 			len = ulmin(mlen, resid);
827 			/*
828 			 * For datagram protocols, leave room
829 			 * for protocol headers in first mbuf.
830 			 */
831 			if (atomic && m == top && len < mlen - max_hdr)
832 				m_align(m, len);
833 		}
834 
835 		error = uiomove(mtod(m, caddr_t), len, uio);
836 		if (error) {
837 			m_freem(top);
838 			return (error);
839 		}
840 
841 		/* adjust counters */
842 		resid = uio->uio_resid;
843 		space -= len;
844 		m->m_len = len;
845 		top->m_pkthdr.len += len;
846 
847 		/* Is there more space and more data? */
848 	} while (space > 0 && resid > 0);
849 
850 	*mp = top;
851 	return 0;
852 }
853 
854 /*
855  * Following replacement or removal of the first mbuf on the first
856  * mbuf chain of a socket buffer, push necessary state changes back
857  * into the socket buffer so that other consumers see the values
858  * consistently.  'nextrecord' is the callers locally stored value of
859  * the original value of sb->sb_mb->m_nextpkt which must be restored
860  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
861  */
862 void
863 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
864 {
865 
866 	/*
867 	 * First, update for the new value of nextrecord.  If necessary,
868 	 * make it the first record.
869 	 */
870 	if (sb->sb_mb != NULL)
871 		sb->sb_mb->m_nextpkt = nextrecord;
872 	else
873 		sb->sb_mb = nextrecord;
874 
875 	/*
876 	 * Now update any dependent socket buffer fields to reflect
877 	 * the new state.  This is an inline of SB_EMPTY_FIXUP, with
878 	 * the addition of a second clause that takes care of the
879 	 * case where sb_mb has been updated, but remains the last
880 	 * record.
881 	 */
882 	if (sb->sb_mb == NULL) {
883 		sb->sb_mbtail = NULL;
884 		sb->sb_lastrecord = NULL;
885 	} else if (sb->sb_mb->m_nextpkt == NULL)
886 		sb->sb_lastrecord = sb->sb_mb;
887 }
888 
889 /*
890  * Implement receive operations on a socket.
891  * We depend on the way that records are added to the sockbuf
892  * by sbappend*.  In particular, each record (mbufs linked through m_next)
893  * must begin with an address if the protocol so specifies,
894  * followed by an optional mbuf or mbufs containing ancillary data,
895  * and then zero or more mbufs of data.
896  * In order to avoid blocking network for the entire time here, we release
897  * the solock() while doing the actual copy to user space.
898  * Although the sockbuf is locked, new data may still be appended,
899  * and thus we must maintain consistency of the sockbuf during that time.
900  *
901  * The caller may receive the data as a single mbuf chain by supplying
902  * an mbuf **mp0 for use in returning the chain.  The uio is then used
903  * only for the count in uio_resid.
904  */
905 int
906 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
907     struct mbuf **mp0, struct mbuf **controlp, int *flagsp,
908     socklen_t controllen)
909 {
910 	struct mbuf *m, **mp;
911 	struct mbuf *cm;
912 	u_long len, offset, moff;
913 	int flags, error, error2, type, uio_error = 0;
914 	const struct protosw *pr = so->so_proto;
915 	struct mbuf *nextrecord;
916 	size_t resid, orig_resid = uio->uio_resid;
917 	int dosolock = ((so->so_rcv.sb_flags & SB_MTXLOCK) == 0);
918 
919 	mp = mp0;
920 	if (paddr)
921 		*paddr = NULL;
922 	if (controlp)
923 		*controlp = NULL;
924 	if (flagsp)
925 		flags = *flagsp &~ MSG_EOR;
926 	else
927 		flags = 0;
928 	if (flags & MSG_OOB) {
929 		m = m_get(M_WAIT, MT_DATA);
930 		solock_shared(so);
931 		error = pru_rcvoob(so, m, flags & MSG_PEEK);
932 		sounlock_shared(so);
933 		if (error)
934 			goto bad;
935 		do {
936 			error = uiomove(mtod(m, caddr_t),
937 			    ulmin(uio->uio_resid, m->m_len), uio);
938 			m = m_free(m);
939 		} while (uio->uio_resid && error == 0 && m);
940 bad:
941 		m_freem(m);
942 		return (error);
943 	}
944 	if (mp)
945 		*mp = NULL;
946 
947 restart:
948 	if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0)
949 		return (error);
950 	if (dosolock)
951 		solock_shared(so);
952 	sb_mtx_lock(&so->so_rcv);
953 
954 	m = so->so_rcv.sb_mb;
955 #ifdef SOCKET_SPLICE
956 	if (isspliced(so))
957 		m = NULL;
958 #endif /* SOCKET_SPLICE */
959 	/*
960 	 * If we have less data than requested, block awaiting more
961 	 * (subject to any timeout) if:
962 	 *   1. the current count is less than the low water mark,
963 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
964 	 *	receive operation at once if we block (resid <= hiwat), or
965 	 *   3. MSG_DONTWAIT is not set.
966 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
967 	 * we have to do the receive in sections, and thus risk returning
968 	 * a short count if a timeout or signal occurs after we start.
969 	 */
970 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
971 	    so->so_rcv.sb_cc < uio->uio_resid) &&
972 	    (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
973 	    ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
974 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
975 #ifdef DIAGNOSTIC
976 		if (m == NULL && so->so_rcv.sb_cc)
977 #ifdef SOCKET_SPLICE
978 		    if (!isspliced(so))
979 #endif /* SOCKET_SPLICE */
980 			panic("receive 1: so %p, so_type %d, sb_cc %lu",
981 			    so, so->so_type, so->so_rcv.sb_cc);
982 #endif
983 		if ((error2 = READ_ONCE(so->so_error))) {
984 			if (m)
985 				goto dontblock;
986 			error = error2;
987 			if ((flags & MSG_PEEK) == 0)
988 				so->so_error = 0;
989 			goto release;
990 		}
991 		if (so->so_rcv.sb_state & SS_CANTRCVMORE) {
992 			if (m)
993 				goto dontblock;
994 			else if (so->so_rcv.sb_cc == 0)
995 				goto release;
996 		}
997 		for (; m; m = m->m_next)
998 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
999 				m = so->so_rcv.sb_mb;
1000 				goto dontblock;
1001 			}
1002 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1003 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1004 			error = ENOTCONN;
1005 			goto release;
1006 		}
1007 		if (uio->uio_resid == 0 && controlp == NULL)
1008 			goto release;
1009 		if (flags & MSG_DONTWAIT) {
1010 			error = EWOULDBLOCK;
1011 			goto release;
1012 		}
1013 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
1014 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
1015 
1016 		sbunlock(&so->so_rcv);
1017 		error = sbwait(so, &so->so_rcv);
1018 		sb_mtx_unlock(&so->so_rcv);
1019 		if (dosolock)
1020 			sounlock_shared(so);
1021 		if (error)
1022 			return (error);
1023 		goto restart;
1024 	}
1025 dontblock:
1026 	/*
1027 	 * On entry here, m points to the first record of the socket buffer.
1028 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1029 	 * pointer to the next record in the socket buffer.  We must keep the
1030 	 * various socket buffer pointers and local stack versions of the
1031 	 * pointers in sync, pushing out modifications before operations that
1032 	 * may sleep, and re-reading them afterwards.
1033 	 *
1034 	 * Otherwise, we will race with the network stack appending new data
1035 	 * or records onto the socket buffer by using inconsistent/stale
1036 	 * versions of the field, possibly resulting in socket buffer
1037 	 * corruption.
1038 	 */
1039 	if (uio->uio_procp)
1040 		uio->uio_procp->p_ru.ru_msgrcv++;
1041 	KASSERT(m == so->so_rcv.sb_mb);
1042 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
1043 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
1044 	nextrecord = m->m_nextpkt;
1045 	if (pr->pr_flags & PR_ADDR) {
1046 #ifdef DIAGNOSTIC
1047 		if (m->m_type != MT_SONAME)
1048 			panic("receive 1a: so %p, so_type %d, m %p, m_type %d",
1049 			    so, so->so_type, m, m->m_type);
1050 #endif
1051 		orig_resid = 0;
1052 		if (flags & MSG_PEEK) {
1053 			if (paddr)
1054 				*paddr = m_copym(m, 0, m->m_len, M_NOWAIT);
1055 			m = m->m_next;
1056 		} else {
1057 			sbfree(so, &so->so_rcv, m);
1058 			if (paddr) {
1059 				*paddr = m;
1060 				so->so_rcv.sb_mb = m->m_next;
1061 				m->m_next = NULL;
1062 				m = so->so_rcv.sb_mb;
1063 			} else {
1064 				so->so_rcv.sb_mb = m_free(m);
1065 				m = so->so_rcv.sb_mb;
1066 			}
1067 			sbsync(&so->so_rcv, nextrecord);
1068 		}
1069 	}
1070 	while (m && m->m_type == MT_CONTROL && error == 0) {
1071 		int skip = 0;
1072 		if (flags & MSG_PEEK) {
1073 			if (mtod(m, struct cmsghdr *)->cmsg_type ==
1074 			    SCM_RIGHTS) {
1075 				/* don't leak internalized SCM_RIGHTS msgs */
1076 				skip = 1;
1077 			} else if (controlp)
1078 				*controlp = m_copym(m, 0, m->m_len, M_NOWAIT);
1079 			m = m->m_next;
1080 		} else {
1081 			sbfree(so, &so->so_rcv, m);
1082 			so->so_rcv.sb_mb = m->m_next;
1083 			m->m_nextpkt = m->m_next = NULL;
1084 			cm = m;
1085 			m = so->so_rcv.sb_mb;
1086 			sbsync(&so->so_rcv, nextrecord);
1087 			if (controlp) {
1088 				if (pr->pr_domain->dom_externalize) {
1089 					sb_mtx_unlock(&so->so_rcv);
1090 					if (dosolock)
1091 						sounlock_shared(so);
1092 					error =
1093 					    (*pr->pr_domain->dom_externalize)
1094 					    (cm, controllen, flags);
1095 					if (dosolock)
1096 						solock_shared(so);
1097 					sb_mtx_lock(&so->so_rcv);
1098 				}
1099 				*controlp = cm;
1100 			} else {
1101 				/*
1102 				 * Dispose of any SCM_RIGHTS message that went
1103 				 * through the read path rather than recv.
1104 				 */
1105 				if (pr->pr_domain->dom_dispose) {
1106 					sb_mtx_unlock(&so->so_rcv);
1107 					pr->pr_domain->dom_dispose(cm);
1108 					sb_mtx_lock(&so->so_rcv);
1109 				}
1110 				m_free(cm);
1111 			}
1112 		}
1113 		if (m != NULL)
1114 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1115 		else
1116 			nextrecord = so->so_rcv.sb_mb;
1117 		if (controlp && !skip)
1118 			controlp = &(*controlp)->m_next;
1119 		orig_resid = 0;
1120 	}
1121 
1122 	/* If m is non-NULL, we have some data to read. */
1123 	if (m) {
1124 		type = m->m_type;
1125 		if (type == MT_OOBDATA)
1126 			flags |= MSG_OOB;
1127 		if (m->m_flags & M_BCAST)
1128 			flags |= MSG_BCAST;
1129 		if (m->m_flags & M_MCAST)
1130 			flags |= MSG_MCAST;
1131 	}
1132 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1133 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1134 
1135 	moff = 0;
1136 	offset = 0;
1137 	while (m && uio->uio_resid > 0 && error == 0) {
1138 		if (m->m_type == MT_OOBDATA) {
1139 			if (type != MT_OOBDATA)
1140 				break;
1141 		} else if (type == MT_OOBDATA) {
1142 			break;
1143 		} else if (m->m_type == MT_CONTROL) {
1144 			/*
1145 			 * If there is more than one control message in the
1146 			 * stream, we do a short read.  Next can be received
1147 			 * or disposed by another system call.
1148 			 */
1149 			break;
1150 #ifdef DIAGNOSTIC
1151 		} else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) {
1152 			panic("receive 3: so %p, so_type %d, m %p, m_type %d",
1153 			    so, so->so_type, m, m->m_type);
1154 #endif
1155 		}
1156 		so->so_rcv.sb_state &= ~SS_RCVATMARK;
1157 		len = uio->uio_resid;
1158 		if (so->so_oobmark && len > so->so_oobmark - offset)
1159 			len = so->so_oobmark - offset;
1160 		if (len > m->m_len - moff)
1161 			len = m->m_len - moff;
1162 		/*
1163 		 * If mp is set, just pass back the mbufs.
1164 		 * Otherwise copy them out via the uio, then free.
1165 		 * Sockbuf must be consistent here (points to current mbuf,
1166 		 * it points to next record) when we drop priority;
1167 		 * we must note any additions to the sockbuf when we
1168 		 * block interrupts again.
1169 		 */
1170 		if (mp == NULL && uio_error == 0) {
1171 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1172 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1173 			resid = uio->uio_resid;
1174 			sb_mtx_unlock(&so->so_rcv);
1175 			if (dosolock)
1176 				sounlock_shared(so);
1177 			uio_error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1178 			if (dosolock)
1179 				solock_shared(so);
1180 			sb_mtx_lock(&so->so_rcv);
1181 			if (uio_error)
1182 				uio->uio_resid = resid - len;
1183 		} else
1184 			uio->uio_resid -= len;
1185 		if (len == m->m_len - moff) {
1186 			if (m->m_flags & M_EOR)
1187 				flags |= MSG_EOR;
1188 			if (flags & MSG_PEEK) {
1189 				m = m->m_next;
1190 				moff = 0;
1191 				orig_resid = 0;
1192 			} else {
1193 				nextrecord = m->m_nextpkt;
1194 				sbfree(so, &so->so_rcv, m);
1195 				if (mp) {
1196 					*mp = m;
1197 					mp = &m->m_next;
1198 					so->so_rcv.sb_mb = m = m->m_next;
1199 					*mp = NULL;
1200 				} else {
1201 					so->so_rcv.sb_mb = m_free(m);
1202 					m = so->so_rcv.sb_mb;
1203 				}
1204 				/*
1205 				 * If m != NULL, we also know that
1206 				 * so->so_rcv.sb_mb != NULL.
1207 				 */
1208 				KASSERT(so->so_rcv.sb_mb == m);
1209 				if (m) {
1210 					m->m_nextpkt = nextrecord;
1211 					if (nextrecord == NULL)
1212 						so->so_rcv.sb_lastrecord = m;
1213 				} else {
1214 					so->so_rcv.sb_mb = nextrecord;
1215 					SB_EMPTY_FIXUP(&so->so_rcv);
1216 				}
1217 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1218 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1219 			}
1220 		} else {
1221 			if (flags & MSG_PEEK) {
1222 				moff += len;
1223 				orig_resid = 0;
1224 			} else {
1225 				if (mp)
1226 					*mp = m_copym(m, 0, len, M_WAIT);
1227 				m->m_data += len;
1228 				m->m_len -= len;
1229 				so->so_rcv.sb_cc -= len;
1230 				so->so_rcv.sb_datacc -= len;
1231 			}
1232 		}
1233 		if (so->so_oobmark) {
1234 			if ((flags & MSG_PEEK) == 0) {
1235 				so->so_oobmark -= len;
1236 				if (so->so_oobmark == 0) {
1237 					so->so_rcv.sb_state |= SS_RCVATMARK;
1238 					break;
1239 				}
1240 			} else {
1241 				offset += len;
1242 				if (offset == so->so_oobmark)
1243 					break;
1244 			}
1245 		}
1246 		if (flags & MSG_EOR)
1247 			break;
1248 		/*
1249 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1250 		 * we must not quit until "uio->uio_resid == 0" or an error
1251 		 * termination.  If a signal/timeout occurs, return
1252 		 * with a short count but without error.
1253 		 * Keep sockbuf locked against other readers.
1254 		 */
1255 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1256 		    !sosendallatonce(so) && !nextrecord) {
1257 			if (so->so_rcv.sb_state & SS_CANTRCVMORE ||
1258 			    so->so_error)
1259 				break;
1260 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1261 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1262 			if (sbwait(so, &so->so_rcv)) {
1263 				sb_mtx_unlock(&so->so_rcv);
1264 				if (dosolock)
1265 					sounlock_shared(so);
1266 				sbunlock(&so->so_rcv);
1267 				return (0);
1268 			}
1269 			if ((m = so->so_rcv.sb_mb) != NULL)
1270 				nextrecord = m->m_nextpkt;
1271 		}
1272 	}
1273 
1274 	if (m && pr->pr_flags & PR_ATOMIC) {
1275 		flags |= MSG_TRUNC;
1276 		if ((flags & MSG_PEEK) == 0)
1277 			(void) sbdroprecord(so, &so->so_rcv);
1278 	}
1279 	if ((flags & MSG_PEEK) == 0) {
1280 		if (m == NULL) {
1281 			/*
1282 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1283 			 * part makes sure sb_lastrecord is up-to-date if
1284 			 * there is still data in the socket buffer.
1285 			 */
1286 			so->so_rcv.sb_mb = nextrecord;
1287 			if (so->so_rcv.sb_mb == NULL) {
1288 				so->so_rcv.sb_mbtail = NULL;
1289 				so->so_rcv.sb_lastrecord = NULL;
1290 			} else if (nextrecord->m_nextpkt == NULL)
1291 				so->so_rcv.sb_lastrecord = nextrecord;
1292 		}
1293 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1294 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1295 		if (pr->pr_flags & PR_WANTRCVD) {
1296 			sb_mtx_unlock(&so->so_rcv);
1297 			if (!dosolock)
1298 				solock_shared(so);
1299 			pru_rcvd(so);
1300 			if (!dosolock)
1301 				sounlock_shared(so);
1302 			sb_mtx_lock(&so->so_rcv);
1303 		}
1304 	}
1305 	if (orig_resid == uio->uio_resid && orig_resid &&
1306 	    (flags & MSG_EOR) == 0 &&
1307 	    (so->so_rcv.sb_state & SS_CANTRCVMORE) == 0) {
1308 		sb_mtx_unlock(&so->so_rcv);
1309 		sbunlock(&so->so_rcv);
1310 		goto restart;
1311 	}
1312 
1313 	if (uio_error)
1314 		error = uio_error;
1315 
1316 	if (flagsp)
1317 		*flagsp |= flags;
1318 release:
1319 	sb_mtx_unlock(&so->so_rcv);
1320 	if (dosolock)
1321 		sounlock_shared(so);
1322 	sbunlock(&so->so_rcv);
1323 	return (error);
1324 }
1325 
1326 int
1327 soshutdown(struct socket *so, int how)
1328 {
1329 	int error = 0;
1330 
1331 	switch (how) {
1332 	case SHUT_RD:
1333 		sorflush(so);
1334 		break;
1335 	case SHUT_RDWR:
1336 		sorflush(so);
1337 		/* FALLTHROUGH */
1338 	case SHUT_WR:
1339 		solock(so);
1340 		error = pru_shutdown(so);
1341 		sounlock(so);
1342 		break;
1343 	default:
1344 		error = EINVAL;
1345 		break;
1346 	}
1347 
1348 	return (error);
1349 }
1350 
1351 void
1352 sorflush(struct socket *so)
1353 {
1354 	struct sockbuf *sb = &so->so_rcv;
1355 	struct mbuf *m;
1356 	const struct protosw *pr = so->so_proto;
1357 	int error;
1358 
1359 	error = sblock(sb, SBL_WAIT | SBL_NOINTR);
1360 	/* with SBL_WAIT and SLB_NOINTR sblock() must not fail */
1361 	KASSERT(error == 0);
1362 
1363 	solock_shared(so);
1364 	socantrcvmore(so);
1365 	mtx_enter(&sb->sb_mtx);
1366 	m = sb->sb_mb;
1367 	memset(&sb->sb_startzero, 0,
1368 	     (caddr_t)&sb->sb_endzero - (caddr_t)&sb->sb_startzero);
1369 	sb->sb_timeo_nsecs = INFSLP;
1370 	mtx_leave(&sb->sb_mtx);
1371 	sounlock_shared(so);
1372 	sbunlock(sb);
1373 
1374 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose)
1375 		(*pr->pr_domain->dom_dispose)(m);
1376 	m_purge(m);
1377 }
1378 
1379 #ifdef SOCKET_SPLICE
1380 
1381 #define so_splicelen	so_sp->ssp_len
1382 #define so_splicemax	so_sp->ssp_max
1383 #define so_idletv	so_sp->ssp_idletv
1384 #define so_idleto	so_sp->ssp_idleto
1385 #define so_splicetask	so_sp->ssp_task
1386 
1387 int
1388 sosplice(struct socket *so, int fd, off_t max, struct timeval *tv)
1389 {
1390 	struct file	*fp;
1391 	struct socket	*sosp;
1392 	struct taskq	*tq;
1393 	int		 error = 0;
1394 
1395 	if ((so->so_proto->pr_flags & PR_SPLICE) == 0)
1396 		return (EPROTONOSUPPORT);
1397 	if (max && max < 0)
1398 		return (EINVAL);
1399 	if (tv && (tv->tv_sec < 0 || !timerisvalid(tv)))
1400 		return (EINVAL);
1401 
1402 	/* If no fd is given, unsplice by removing existing link. */
1403 	if (fd < 0) {
1404 		if ((error = sblock(&so->so_rcv, SBL_WAIT)) != 0)
1405 			return (error);
1406 		solock(so);
1407 		if (so->so_options & SO_ACCEPTCONN) {
1408 			error = EOPNOTSUPP;
1409 			goto out;
1410 		}
1411 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1412 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1413 			error = ENOTCONN;
1414 			goto out;
1415 		}
1416 
1417 		if (so->so_sp && so->so_sp->ssp_socket)
1418 			sounsplice(so, so->so_sp->ssp_socket, 0);
1419  out:
1420 		sounlock(so);
1421 		sbunlock(&so->so_rcv);
1422 		return (error);
1423 	}
1424 
1425 	if (sosplice_taskq == NULL) {
1426 		rw_enter_write(&sosplice_lock);
1427 		if (sosplice_taskq == NULL) {
1428 			tq = taskq_create("sosplice", 1, IPL_SOFTNET,
1429 			    TASKQ_MPSAFE);
1430 			if (tq == NULL) {
1431 				rw_exit_write(&sosplice_lock);
1432 				return (ENOMEM);
1433 			}
1434 			/* Ensure the taskq is fully visible to other CPUs. */
1435 			membar_producer();
1436 			sosplice_taskq = tq;
1437 		}
1438 		rw_exit_write(&sosplice_lock);
1439 	} else {
1440 		/* Ensure the taskq is fully visible on this CPU. */
1441 		membar_consumer();
1442 	}
1443 
1444 	/* Find sosp, the drain socket where data will be spliced into. */
1445 	if ((error = getsock(curproc, fd, &fp)) != 0)
1446 		return (error);
1447 	sosp = fp->f_data;
1448 
1449 	if (sosp->so_proto->pr_usrreqs->pru_send !=
1450 	    so->so_proto->pr_usrreqs->pru_send) {
1451 		error = EPROTONOSUPPORT;
1452 		goto frele;
1453 	}
1454 
1455 	if ((error = sblock(&so->so_rcv, SBL_WAIT)) != 0)
1456 		goto frele;
1457 	if ((error = sblock(&sosp->so_snd, SBL_WAIT)) != 0) {
1458 		sbunlock(&so->so_rcv);
1459 		goto frele;
1460 	}
1461 	solock(so);
1462 
1463 	if ((so->so_options & SO_ACCEPTCONN) ||
1464 	    (sosp->so_options & SO_ACCEPTCONN)) {
1465 		error = EOPNOTSUPP;
1466 		goto release;
1467 	}
1468 	if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1469 	    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1470 		error = ENOTCONN;
1471 		goto release;
1472 	}
1473 	if ((sosp->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0) {
1474 		error = ENOTCONN;
1475 		goto release;
1476 	}
1477 	if (so->so_sp == NULL)
1478 		so->so_sp = pool_get(&sosplice_pool, PR_WAITOK | PR_ZERO);
1479 	if (sosp->so_sp == NULL)
1480 		sosp->so_sp = pool_get(&sosplice_pool, PR_WAITOK | PR_ZERO);
1481 	if (so->so_sp->ssp_socket || sosp->so_sp->ssp_soback) {
1482 		error = EBUSY;
1483 		goto release;
1484 	}
1485 
1486 	so->so_splicelen = 0;
1487 	so->so_splicemax = max;
1488 	if (tv)
1489 		so->so_idletv = *tv;
1490 	else
1491 		timerclear(&so->so_idletv);
1492 	timeout_set_flags(&so->so_idleto, soidle, so,
1493 	    KCLOCK_NONE, TIMEOUT_PROC | TIMEOUT_MPSAFE);
1494 	task_set(&so->so_splicetask, sotask, so);
1495 
1496 	/*
1497 	 * To prevent sorwakeup() calling somove() before this somove()
1498 	 * has finished, the socket buffers are not marked as spliced yet.
1499 	 */
1500 
1501 	/* Splice so and sosp together. */
1502 	mtx_enter(&so->so_rcv.sb_mtx);
1503 	mtx_enter(&sosp->so_snd.sb_mtx);
1504 	so->so_sp->ssp_socket = sosp;
1505 	sosp->so_sp->ssp_soback = so;
1506 	mtx_leave(&sosp->so_snd.sb_mtx);
1507 	mtx_leave(&so->so_rcv.sb_mtx);
1508 
1509 	if ((so->so_proto->pr_flags & PR_WANTRCVD) == 0)
1510 		sounlock(so);
1511 	if (somove(so, M_WAIT)) {
1512 		mtx_enter(&so->so_rcv.sb_mtx);
1513 		mtx_enter(&sosp->so_snd.sb_mtx);
1514 		so->so_rcv.sb_flags |= SB_SPLICE;
1515 		sosp->so_snd.sb_flags |= SB_SPLICE;
1516 		mtx_leave(&sosp->so_snd.sb_mtx);
1517 		mtx_leave(&so->so_rcv.sb_mtx);
1518 	}
1519 	if ((so->so_proto->pr_flags & PR_WANTRCVD) == 0)
1520 		solock(so);
1521 
1522  release:
1523 	sounlock(so);
1524 	sbunlock(&sosp->so_snd);
1525 	sbunlock(&so->so_rcv);
1526  frele:
1527 	FRELE(fp, curproc);
1528 
1529 	return (error);
1530 }
1531 
1532 void
1533 sounsplice(struct socket *so, struct socket *sosp, int freeing)
1534 {
1535 	if ((so->so_proto->pr_flags & PR_WANTRCVD) == 0)
1536 		sbassertlocked(&so->so_rcv);
1537 	soassertlocked(so);
1538 
1539 	task_del(sosplice_taskq, &so->so_splicetask);
1540 	timeout_del(&so->so_idleto);
1541 
1542 	mtx_enter(&so->so_rcv.sb_mtx);
1543 	mtx_enter(&sosp->so_snd.sb_mtx);
1544 	so->so_rcv.sb_flags &= ~SB_SPLICE;
1545 	sosp->so_snd.sb_flags &= ~SB_SPLICE;
1546 	so->so_sp->ssp_socket = sosp->so_sp->ssp_soback = NULL;
1547 	mtx_leave(&sosp->so_snd.sb_mtx);
1548 	mtx_leave(&so->so_rcv.sb_mtx);
1549 
1550 	/* Do not wakeup a socket that is about to be freed. */
1551 	if ((freeing & SOSP_FREEING_READ) == 0 && soreadable(so))
1552 		sorwakeup(so);
1553 	if ((freeing & SOSP_FREEING_WRITE) == 0 && sowriteable(sosp))
1554 		sowwakeup(sosp);
1555 }
1556 
1557 void
1558 soidle(void *arg)
1559 {
1560 	struct socket *so = arg;
1561 
1562 	sblock(&so->so_rcv, SBL_WAIT | SBL_NOINTR);
1563 	solock(so);
1564 	/*
1565 	 * Depending on socket type, sblock(&so->so_rcv) or solock()
1566 	 * is always held while modifying SB_SPLICE and
1567 	 * so->so_sp->ssp_socket.
1568 	 */
1569 	if (so->so_rcv.sb_flags & SB_SPLICE) {
1570 		so->so_error = ETIMEDOUT;
1571 		sounsplice(so, so->so_sp->ssp_socket, 0);
1572 	}
1573 	sounlock(so);
1574 	sbunlock(&so->so_rcv);
1575 }
1576 
1577 void
1578 sotask(void *arg)
1579 {
1580 	struct socket *so = arg;
1581 	int doyield = 0;
1582 	int sockstream = (so->so_proto->pr_flags & PR_WANTRCVD);
1583 
1584 	/*
1585 	 * sblock() on `so_rcv' protects sockets from being unspliced
1586 	 * for UDP case. TCP sockets still rely on solock().
1587 	 */
1588 
1589 	sblock(&so->so_rcv, SBL_WAIT | SBL_NOINTR);
1590 	if (sockstream)
1591 		solock(so);
1592 
1593 	if (so->so_rcv.sb_flags & SB_SPLICE) {
1594 		if (sockstream)
1595 			doyield = 1;
1596 		somove(so, M_DONTWAIT);
1597 	}
1598 
1599 	if (sockstream)
1600 		sounlock(so);
1601 	sbunlock(&so->so_rcv);
1602 
1603 	if (doyield) {
1604 		/* Avoid user land starvation. */
1605 		yield();
1606 	}
1607 }
1608 
1609 /*
1610  * The socket splicing task or idle timeout may sleep while grabbing the net
1611  * lock.  As sofree() can be called anytime, sotask() or soidle() could access
1612  * the socket memory of a freed socket after wakeup.  So delay the pool_put()
1613  * after all pending socket splicing tasks or timeouts have finished.  Do this
1614  * by scheduling it on the same threads.
1615  */
1616 void
1617 soreaper(void *arg)
1618 {
1619 	struct socket *so = arg;
1620 
1621 	/* Reuse splice task, sounsplice() has been called before. */
1622 	task_set(&so->so_sp->ssp_task, soput, so);
1623 	task_add(sosplice_taskq, &so->so_sp->ssp_task);
1624 }
1625 
1626 void
1627 soput(void *arg)
1628 {
1629 	struct socket *so = arg;
1630 
1631 	pool_put(&sosplice_pool, so->so_sp);
1632 	pool_put(&socket_pool, so);
1633 }
1634 
1635 /*
1636  * Move data from receive buffer of spliced source socket to send
1637  * buffer of drain socket.  Try to move as much as possible in one
1638  * big chunk.  It is a TCP only implementation.
1639  * Return value 0 means splicing has been finished, 1 continue.
1640  */
1641 int
1642 somove(struct socket *so, int wait)
1643 {
1644 	struct socket	*sosp = so->so_sp->ssp_socket;
1645 	struct mbuf	*m, **mp, *nextrecord;
1646 	u_long		 len, off, oobmark;
1647 	long		 space;
1648 	int		 error = 0, maxreached = 0, unsplice = 0;
1649 	unsigned int	 rcvstate;
1650 	int		 sockdgram = ((so->so_proto->pr_flags &
1651 			     PR_WANTRCVD) == 0);
1652 
1653 	if (sockdgram)
1654 		sbassertlocked(&so->so_rcv);
1655 	else
1656 		soassertlocked(so);
1657 
1658 	mtx_enter(&so->so_rcv.sb_mtx);
1659 	mtx_enter(&sosp->so_snd.sb_mtx);
1660 
1661  nextpkt:
1662 	if ((error = READ_ONCE(so->so_error)))
1663 		goto release;
1664 	if (sosp->so_snd.sb_state & SS_CANTSENDMORE) {
1665 		error = EPIPE;
1666 		goto release;
1667 	}
1668 
1669 	error = READ_ONCE(sosp->so_error);
1670 	if (error) {
1671 		if (error != ETIMEDOUT && error != EFBIG && error != ELOOP)
1672 			goto release;
1673 		error = 0;
1674 	}
1675 	if ((sosp->so_state & SS_ISCONNECTED) == 0)
1676 		goto release;
1677 
1678 	/* Calculate how many bytes can be copied now. */
1679 	len = so->so_rcv.sb_datacc;
1680 	if (so->so_splicemax) {
1681 		KASSERT(so->so_splicelen < so->so_splicemax);
1682 		if (so->so_splicemax <= so->so_splicelen + len) {
1683 			len = so->so_splicemax - so->so_splicelen;
1684 			maxreached = 1;
1685 		}
1686 	}
1687 	space = sbspace_locked(sosp, &sosp->so_snd);
1688 	if (so->so_oobmark && so->so_oobmark < len &&
1689 	    so->so_oobmark < space + 1024)
1690 		space += 1024;
1691 	if (space <= 0) {
1692 		maxreached = 0;
1693 		goto release;
1694 	}
1695 	if (space < len) {
1696 		maxreached = 0;
1697 		if (space < sosp->so_snd.sb_lowat)
1698 			goto release;
1699 		len = space;
1700 	}
1701 	sosp->so_snd.sb_state |= SS_ISSENDING;
1702 
1703 	SBLASTRECORDCHK(&so->so_rcv, "somove 1");
1704 	SBLASTMBUFCHK(&so->so_rcv, "somove 1");
1705 	m = so->so_rcv.sb_mb;
1706 	if (m == NULL)
1707 		goto release;
1708 	nextrecord = m->m_nextpkt;
1709 
1710 	/* Drop address and control information not used with splicing. */
1711 	if (so->so_proto->pr_flags & PR_ADDR) {
1712 #ifdef DIAGNOSTIC
1713 		if (m->m_type != MT_SONAME)
1714 			panic("somove soname: so %p, so_type %d, m %p, "
1715 			    "m_type %d", so, so->so_type, m, m->m_type);
1716 #endif
1717 		m = m->m_next;
1718 	}
1719 	while (m && m->m_type == MT_CONTROL)
1720 		m = m->m_next;
1721 	if (m == NULL) {
1722 		sbdroprecord(so, &so->so_rcv);
1723 		if (so->so_proto->pr_flags & PR_WANTRCVD) {
1724 			mtx_leave(&sosp->so_snd.sb_mtx);
1725 			mtx_leave(&so->so_rcv.sb_mtx);
1726 			pru_rcvd(so);
1727 			mtx_enter(&so->so_rcv.sb_mtx);
1728 			mtx_enter(&sosp->so_snd.sb_mtx);
1729 		}
1730 		goto nextpkt;
1731 	}
1732 
1733 	/*
1734 	 * By splicing sockets connected to localhost, userland might create a
1735 	 * loop.  Dissolve splicing with error if loop is detected by counter.
1736 	 *
1737 	 * If we deal with looped broadcast/multicast packet we bail out with
1738 	 * no error to suppress splice termination.
1739 	 */
1740 	if ((m->m_flags & M_PKTHDR) &&
1741 	    ((m->m_pkthdr.ph_loopcnt++ >= M_MAXLOOP) ||
1742 	    ((m->m_flags & M_LOOP) && (m->m_flags & (M_BCAST|M_MCAST))))) {
1743 		error = ELOOP;
1744 		goto release;
1745 	}
1746 
1747 	if (so->so_proto->pr_flags & PR_ATOMIC) {
1748 		if ((m->m_flags & M_PKTHDR) == 0)
1749 			panic("somove !PKTHDR: so %p, so_type %d, m %p, "
1750 			    "m_type %d", so, so->so_type, m, m->m_type);
1751 		if (sosp->so_snd.sb_hiwat < m->m_pkthdr.len) {
1752 			error = EMSGSIZE;
1753 			goto release;
1754 		}
1755 		if (len < m->m_pkthdr.len)
1756 			goto release;
1757 		if (m->m_pkthdr.len < len) {
1758 			maxreached = 0;
1759 			len = m->m_pkthdr.len;
1760 		}
1761 		/*
1762 		 * Throw away the name mbuf after it has been assured
1763 		 * that the whole first record can be processed.
1764 		 */
1765 		m = so->so_rcv.sb_mb;
1766 		sbfree(so, &so->so_rcv, m);
1767 		so->so_rcv.sb_mb = m_free(m);
1768 		sbsync(&so->so_rcv, nextrecord);
1769 	}
1770 	/*
1771 	 * Throw away the control mbufs after it has been assured
1772 	 * that the whole first record can be processed.
1773 	 */
1774 	m = so->so_rcv.sb_mb;
1775 	while (m && m->m_type == MT_CONTROL) {
1776 		sbfree(so, &so->so_rcv, m);
1777 		so->so_rcv.sb_mb = m_free(m);
1778 		m = so->so_rcv.sb_mb;
1779 		sbsync(&so->so_rcv, nextrecord);
1780 	}
1781 
1782 	SBLASTRECORDCHK(&so->so_rcv, "somove 2");
1783 	SBLASTMBUFCHK(&so->so_rcv, "somove 2");
1784 
1785 	/* Take at most len mbufs out of receive buffer. */
1786 	for (off = 0, mp = &m; off <= len && *mp;
1787 	    off += (*mp)->m_len, mp = &(*mp)->m_next) {
1788 		u_long size = len - off;
1789 
1790 #ifdef DIAGNOSTIC
1791 		if ((*mp)->m_type != MT_DATA && (*mp)->m_type != MT_HEADER)
1792 			panic("somove type: so %p, so_type %d, m %p, "
1793 			    "m_type %d", so, so->so_type, *mp, (*mp)->m_type);
1794 #endif
1795 		if ((*mp)->m_len > size) {
1796 			/*
1797 			 * Move only a partial mbuf at maximum splice length or
1798 			 * if the drain buffer is too small for this large mbuf.
1799 			 */
1800 			if (!maxreached && sosp->so_snd.sb_datacc > 0) {
1801 				len -= size;
1802 				break;
1803 			}
1804 			*mp = m_copym(so->so_rcv.sb_mb, 0, size, wait);
1805 			if (*mp == NULL) {
1806 				len -= size;
1807 				break;
1808 			}
1809 			so->so_rcv.sb_mb->m_data += size;
1810 			so->so_rcv.sb_mb->m_len -= size;
1811 			so->so_rcv.sb_cc -= size;
1812 			so->so_rcv.sb_datacc -= size;
1813 		} else {
1814 			*mp = so->so_rcv.sb_mb;
1815 			sbfree(so, &so->so_rcv, *mp);
1816 			so->so_rcv.sb_mb = (*mp)->m_next;
1817 			sbsync(&so->so_rcv, nextrecord);
1818 		}
1819 	}
1820 	*mp = NULL;
1821 
1822 	SBLASTRECORDCHK(&so->so_rcv, "somove 3");
1823 	SBLASTMBUFCHK(&so->so_rcv, "somove 3");
1824 	SBCHECK(so, &so->so_rcv);
1825 	if (m == NULL)
1826 		goto release;
1827 	m->m_nextpkt = NULL;
1828 	if (m->m_flags & M_PKTHDR) {
1829 		m_resethdr(m);
1830 		m->m_pkthdr.len = len;
1831 	}
1832 
1833 	/* Send window update to source peer as receive buffer has changed. */
1834 	if (so->so_proto->pr_flags & PR_WANTRCVD) {
1835 		mtx_leave(&sosp->so_snd.sb_mtx);
1836 		mtx_leave(&so->so_rcv.sb_mtx);
1837 		pru_rcvd(so);
1838 		mtx_enter(&so->so_rcv.sb_mtx);
1839 		mtx_enter(&sosp->so_snd.sb_mtx);
1840 	}
1841 
1842 	/* Receive buffer did shrink by len bytes, adjust oob. */
1843 	rcvstate = so->so_rcv.sb_state;
1844 	so->so_rcv.sb_state &= ~SS_RCVATMARK;
1845 	oobmark = so->so_oobmark;
1846 	so->so_oobmark = oobmark > len ? oobmark - len : 0;
1847 	if (oobmark) {
1848 		if (oobmark == len)
1849 			so->so_rcv.sb_state |= SS_RCVATMARK;
1850 		if (oobmark >= len)
1851 			oobmark = 0;
1852 	}
1853 
1854 	/*
1855 	 * Handle oob data.  If any malloc fails, ignore error.
1856 	 * TCP urgent data is not very reliable anyway.
1857 	 */
1858 	while (((rcvstate & SS_RCVATMARK) || oobmark) &&
1859 	    (so->so_options & SO_OOBINLINE)) {
1860 		struct mbuf *o = NULL;
1861 
1862 		if (rcvstate & SS_RCVATMARK) {
1863 			o = m_get(wait, MT_DATA);
1864 			rcvstate &= ~SS_RCVATMARK;
1865 		} else if (oobmark) {
1866 			o = m_split(m, oobmark, wait);
1867 			if (o) {
1868 				mtx_leave(&sosp->so_snd.sb_mtx);
1869 				mtx_leave(&so->so_rcv.sb_mtx);
1870 				error = pru_send(sosp, m, NULL, NULL);
1871 				mtx_enter(&so->so_rcv.sb_mtx);
1872 				mtx_enter(&sosp->so_snd.sb_mtx);
1873 
1874 				if (error) {
1875 					if (sosp->so_snd.sb_state &
1876 					    SS_CANTSENDMORE)
1877 						error = EPIPE;
1878 					m_freem(o);
1879 					goto release;
1880 				}
1881 				len -= oobmark;
1882 				so->so_splicelen += oobmark;
1883 				m = o;
1884 				o = m_get(wait, MT_DATA);
1885 			}
1886 			oobmark = 0;
1887 		}
1888 		if (o) {
1889 			o->m_len = 1;
1890 			*mtod(o, caddr_t) = *mtod(m, caddr_t);
1891 
1892 			mtx_leave(&sosp->so_snd.sb_mtx);
1893 			mtx_leave(&so->so_rcv.sb_mtx);
1894 			error = pru_sendoob(sosp, o, NULL, NULL);
1895 			mtx_enter(&so->so_rcv.sb_mtx);
1896 			mtx_enter(&sosp->so_snd.sb_mtx);
1897 
1898 			if (error) {
1899 				if (sosp->so_snd.sb_state & SS_CANTSENDMORE)
1900 					error = EPIPE;
1901 				m_freem(m);
1902 				goto release;
1903 			}
1904 			len -= 1;
1905 			so->so_splicelen += 1;
1906 			if (oobmark) {
1907 				oobmark -= 1;
1908 				if (oobmark == 0)
1909 					rcvstate |= SS_RCVATMARK;
1910 			}
1911 			m_adj(m, 1);
1912 		}
1913 	}
1914 
1915 	/* Append all remaining data to drain socket. */
1916 	if (so->so_rcv.sb_cc == 0 || maxreached)
1917 		sosp->so_snd.sb_state &= ~SS_ISSENDING;
1918 
1919 	mtx_leave(&sosp->so_snd.sb_mtx);
1920 	mtx_leave(&so->so_rcv.sb_mtx);
1921 
1922 	if (sockdgram)
1923 		solock_shared(sosp);
1924 	error = pru_send(sosp, m, NULL, NULL);
1925 	if (sockdgram)
1926 		sounlock_shared(sosp);
1927 
1928 	mtx_enter(&so->so_rcv.sb_mtx);
1929 	mtx_enter(&sosp->so_snd.sb_mtx);
1930 
1931 	if (error) {
1932 		if (sosp->so_snd.sb_state & SS_CANTSENDMORE ||
1933 		    sosp->so_pcb == NULL)
1934 			error = EPIPE;
1935 		goto release;
1936 	}
1937 	so->so_splicelen += len;
1938 
1939 	/* Move several packets if possible. */
1940 	if (!maxreached && nextrecord)
1941 		goto nextpkt;
1942 
1943  release:
1944 	sosp->so_snd.sb_state &= ~SS_ISSENDING;
1945 
1946 	if (!error && maxreached && so->so_splicemax == so->so_splicelen)
1947 		error = EFBIG;
1948 	if (error)
1949 		WRITE_ONCE(so->so_error, error);
1950 
1951 	if (((so->so_rcv.sb_state & SS_CANTRCVMORE) &&
1952 	    so->so_rcv.sb_cc == 0) ||
1953 	    (sosp->so_snd.sb_state & SS_CANTSENDMORE) ||
1954 	    maxreached || error)
1955 		unsplice = 1;
1956 
1957 	mtx_leave(&sosp->so_snd.sb_mtx);
1958 	mtx_leave(&so->so_rcv.sb_mtx);
1959 
1960 	if (unsplice) {
1961 		if (sockdgram)
1962 			solock(so);
1963 		sounsplice(so, sosp, 0);
1964 		if (sockdgram)
1965 			sounlock(so);
1966 
1967 		return (0);
1968 	}
1969 	if (timerisset(&so->so_idletv))
1970 		timeout_add_tv(&so->so_idleto, &so->so_idletv);
1971 	return (1);
1972 }
1973 #endif /* SOCKET_SPLICE */
1974 
1975 void
1976 sorwakeup(struct socket *so)
1977 {
1978 	if ((so->so_rcv.sb_flags & SB_MTXLOCK) == 0)
1979 		soassertlocked_readonly(so);
1980 
1981 #ifdef SOCKET_SPLICE
1982 	if (so->so_proto->pr_flags & PR_SPLICE) {
1983 		sb_mtx_lock(&so->so_rcv);
1984 		if (so->so_rcv.sb_flags & SB_SPLICE)
1985 			task_add(sosplice_taskq, &so->so_splicetask);
1986 		if (isspliced(so)) {
1987 			sb_mtx_unlock(&so->so_rcv);
1988 			return;
1989 		}
1990 		sb_mtx_unlock(&so->so_rcv);
1991 	}
1992 #endif
1993 	sowakeup(so, &so->so_rcv);
1994 	if (so->so_upcall)
1995 		(*(so->so_upcall))(so, so->so_upcallarg, M_DONTWAIT);
1996 }
1997 
1998 void
1999 sowwakeup(struct socket *so)
2000 {
2001 	if ((so->so_snd.sb_flags & SB_MTXLOCK) == 0)
2002 		soassertlocked_readonly(so);
2003 
2004 #ifdef SOCKET_SPLICE
2005 	if (so->so_proto->pr_flags & PR_SPLICE) {
2006 		sb_mtx_lock(&so->so_snd);
2007 		if (so->so_snd.sb_flags & SB_SPLICE)
2008 			task_add(sosplice_taskq,
2009 			    &so->so_sp->ssp_soback->so_splicetask);
2010 		if (issplicedback(so)) {
2011 			sb_mtx_unlock(&so->so_snd);
2012 			return;
2013 		}
2014 		sb_mtx_unlock(&so->so_snd);
2015 	}
2016 #endif
2017 	sowakeup(so, &so->so_snd);
2018 }
2019 
2020 int
2021 sosetopt(struct socket *so, int level, int optname, struct mbuf *m)
2022 {
2023 	int error = 0;
2024 
2025 	if (level != SOL_SOCKET) {
2026 		if (so->so_proto->pr_ctloutput) {
2027 			solock(so);
2028 			error = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so,
2029 			    level, optname, m);
2030 			sounlock(so);
2031 			return (error);
2032 		}
2033 		error = ENOPROTOOPT;
2034 	} else {
2035 		switch (optname) {
2036 
2037 		case SO_LINGER:
2038 			if (m == NULL || m->m_len != sizeof (struct linger) ||
2039 			    mtod(m, struct linger *)->l_linger < 0 ||
2040 			    mtod(m, struct linger *)->l_linger > SHRT_MAX)
2041 				return (EINVAL);
2042 
2043 			solock(so);
2044 			so->so_linger = mtod(m, struct linger *)->l_linger;
2045 			if (*mtod(m, int *))
2046 				so->so_options |= optname;
2047 			else
2048 				so->so_options &= ~optname;
2049 			sounlock(so);
2050 
2051 			break;
2052 		case SO_BINDANY:
2053 			if ((error = suser(curproc)) != 0)	/* XXX */
2054 				return (error);
2055 			/* FALLTHROUGH */
2056 
2057 		case SO_DEBUG:
2058 		case SO_KEEPALIVE:
2059 		case SO_USELOOPBACK:
2060 		case SO_BROADCAST:
2061 		case SO_REUSEADDR:
2062 		case SO_REUSEPORT:
2063 		case SO_OOBINLINE:
2064 		case SO_TIMESTAMP:
2065 		case SO_ZEROIZE:
2066 			if (m == NULL || m->m_len < sizeof (int))
2067 				return (EINVAL);
2068 
2069 			solock(so);
2070 			if (*mtod(m, int *))
2071 				so->so_options |= optname;
2072 			else
2073 				so->so_options &= ~optname;
2074 			sounlock(so);
2075 
2076 			break;
2077 		case SO_DONTROUTE:
2078 			if (m == NULL || m->m_len < sizeof (int))
2079 				return (EINVAL);
2080 			if (*mtod(m, int *))
2081 				error = EOPNOTSUPP;
2082 			break;
2083 
2084 		case SO_SNDBUF:
2085 		case SO_RCVBUF:
2086 		case SO_SNDLOWAT:
2087 		case SO_RCVLOWAT:
2088 		    {
2089 			struct sockbuf *sb = (optname == SO_SNDBUF ||
2090 			    optname == SO_SNDLOWAT ?
2091 			    &so->so_snd : &so->so_rcv);
2092 			u_long cnt;
2093 
2094 			if (m == NULL || m->m_len < sizeof (int))
2095 				return (EINVAL);
2096 			cnt = *mtod(m, int *);
2097 			if ((long)cnt <= 0)
2098 				cnt = 1;
2099 
2100 			if (((sb->sb_flags & SB_MTXLOCK) == 0))
2101 				solock(so);
2102 			mtx_enter(&sb->sb_mtx);
2103 
2104 			switch (optname) {
2105 			case SO_SNDBUF:
2106 			case SO_RCVBUF:
2107 				if (sb->sb_state &
2108 				    (SS_CANTSENDMORE | SS_CANTRCVMORE)) {
2109 					error = EINVAL;
2110 					break;
2111 				}
2112 				if (sbcheckreserve(cnt, sb->sb_wat) ||
2113 				    sbreserve(so, sb, cnt)) {
2114 					error = ENOBUFS;
2115 					break;
2116 				}
2117 				sb->sb_wat = cnt;
2118 				break;
2119 			case SO_SNDLOWAT:
2120 			case SO_RCVLOWAT:
2121 				sb->sb_lowat = (cnt > sb->sb_hiwat) ?
2122 				    sb->sb_hiwat : cnt;
2123 				break;
2124 			}
2125 
2126 			mtx_leave(&sb->sb_mtx);
2127 			if (((sb->sb_flags & SB_MTXLOCK) == 0))
2128 				sounlock(so);
2129 
2130 			break;
2131 		    }
2132 
2133 		case SO_SNDTIMEO:
2134 		case SO_RCVTIMEO:
2135 		    {
2136 			struct sockbuf *sb = (optname == SO_SNDTIMEO ?
2137 			    &so->so_snd : &so->so_rcv);
2138 			struct timeval tv;
2139 			uint64_t nsecs;
2140 
2141 			if (m == NULL || m->m_len < sizeof (tv))
2142 				return (EINVAL);
2143 			memcpy(&tv, mtod(m, struct timeval *), sizeof tv);
2144 			if (!timerisvalid(&tv))
2145 				return (EINVAL);
2146 			nsecs = TIMEVAL_TO_NSEC(&tv);
2147 			if (nsecs == UINT64_MAX)
2148 				return (EDOM);
2149 			if (nsecs == 0)
2150 				nsecs = INFSLP;
2151 
2152 			mtx_enter(&sb->sb_mtx);
2153 			sb->sb_timeo_nsecs = nsecs;
2154 			mtx_leave(&sb->sb_mtx);
2155 			break;
2156 		    }
2157 
2158 		case SO_RTABLE:
2159 			if (so->so_proto->pr_domain &&
2160 			    so->so_proto->pr_domain->dom_protosw &&
2161 			    so->so_proto->pr_ctloutput) {
2162 				const struct domain *dom =
2163 				    so->so_proto->pr_domain;
2164 
2165 				level = dom->dom_protosw->pr_protocol;
2166 				solock(so);
2167 				error = (*so->so_proto->pr_ctloutput)
2168 				    (PRCO_SETOPT, so, level, optname, m);
2169 				sounlock(so);
2170 			} else
2171 				error = ENOPROTOOPT;
2172 			break;
2173 #ifdef SOCKET_SPLICE
2174 		case SO_SPLICE:
2175 			if (m == NULL) {
2176 				error = sosplice(so, -1, 0, NULL);
2177 			} else if (m->m_len < sizeof(int)) {
2178 				error = EINVAL;
2179 			} else if (m->m_len < sizeof(struct splice)) {
2180 				error = sosplice(so, *mtod(m, int *), 0, NULL);
2181 			} else {
2182 				error = sosplice(so,
2183 				    mtod(m, struct splice *)->sp_fd,
2184 				    mtod(m, struct splice *)->sp_max,
2185 				   &mtod(m, struct splice *)->sp_idle);
2186 			}
2187 			break;
2188 #endif /* SOCKET_SPLICE */
2189 
2190 		default:
2191 			error = ENOPROTOOPT;
2192 			break;
2193 		}
2194 	}
2195 
2196 	return (error);
2197 }
2198 
2199 int
2200 sogetopt(struct socket *so, int level, int optname, struct mbuf *m)
2201 {
2202 	int error = 0;
2203 
2204 	if (level != SOL_SOCKET) {
2205 		if (so->so_proto->pr_ctloutput) {
2206 			m->m_len = 0;
2207 
2208 			solock(so);
2209 			error = (*so->so_proto->pr_ctloutput)(PRCO_GETOPT, so,
2210 			    level, optname, m);
2211 			sounlock(so);
2212 			return (error);
2213 		} else
2214 			return (ENOPROTOOPT);
2215 	} else {
2216 		m->m_len = sizeof (int);
2217 
2218 		switch (optname) {
2219 
2220 		case SO_LINGER:
2221 			m->m_len = sizeof (struct linger);
2222 			solock_shared(so);
2223 			mtod(m, struct linger *)->l_onoff =
2224 				so->so_options & SO_LINGER;
2225 			mtod(m, struct linger *)->l_linger = so->so_linger;
2226 			sounlock_shared(so);
2227 			break;
2228 
2229 		case SO_BINDANY:
2230 		case SO_USELOOPBACK:
2231 		case SO_DEBUG:
2232 		case SO_KEEPALIVE:
2233 		case SO_REUSEADDR:
2234 		case SO_REUSEPORT:
2235 		case SO_BROADCAST:
2236 		case SO_OOBINLINE:
2237 		case SO_ACCEPTCONN:
2238 		case SO_TIMESTAMP:
2239 		case SO_ZEROIZE:
2240 			*mtod(m, int *) = so->so_options & optname;
2241 			break;
2242 
2243 		case SO_DONTROUTE:
2244 			*mtod(m, int *) = 0;
2245 			break;
2246 
2247 		case SO_TYPE:
2248 			*mtod(m, int *) = so->so_type;
2249 			break;
2250 
2251 		case SO_ERROR:
2252 			solock(so);
2253 			*mtod(m, int *) = so->so_error;
2254 			so->so_error = 0;
2255 			sounlock(so);
2256 
2257 			break;
2258 
2259 		case SO_DOMAIN:
2260 			*mtod(m, int *) = so->so_proto->pr_domain->dom_family;
2261 			break;
2262 
2263 		case SO_PROTOCOL:
2264 			*mtod(m, int *) = so->so_proto->pr_protocol;
2265 			break;
2266 
2267 		case SO_SNDBUF:
2268 			*mtod(m, int *) = so->so_snd.sb_hiwat;
2269 			break;
2270 
2271 		case SO_RCVBUF:
2272 			*mtod(m, int *) = so->so_rcv.sb_hiwat;
2273 			break;
2274 
2275 		case SO_SNDLOWAT:
2276 			*mtod(m, int *) = so->so_snd.sb_lowat;
2277 			break;
2278 
2279 		case SO_RCVLOWAT:
2280 			*mtod(m, int *) = so->so_rcv.sb_lowat;
2281 			break;
2282 
2283 		case SO_SNDTIMEO:
2284 		case SO_RCVTIMEO:
2285 		    {
2286 			struct sockbuf *sb = (optname == SO_SNDTIMEO ?
2287 			    &so->so_snd : &so->so_rcv);
2288 			struct timeval tv;
2289 			uint64_t nsecs;
2290 
2291 			mtx_enter(&sb->sb_mtx);
2292 			nsecs = sb->sb_timeo_nsecs;
2293 			mtx_leave(&sb->sb_mtx);
2294 
2295 			m->m_len = sizeof(struct timeval);
2296 			memset(&tv, 0, sizeof(tv));
2297 			if (nsecs != INFSLP)
2298 				NSEC_TO_TIMEVAL(nsecs, &tv);
2299 			memcpy(mtod(m, struct timeval *), &tv, sizeof tv);
2300 			break;
2301 		    }
2302 
2303 		case SO_RTABLE:
2304 			if (so->so_proto->pr_domain &&
2305 			    so->so_proto->pr_domain->dom_protosw &&
2306 			    so->so_proto->pr_ctloutput) {
2307 				const struct domain *dom =
2308 				    so->so_proto->pr_domain;
2309 
2310 				level = dom->dom_protosw->pr_protocol;
2311 				solock(so);
2312 				error = (*so->so_proto->pr_ctloutput)
2313 				    (PRCO_GETOPT, so, level, optname, m);
2314 				sounlock(so);
2315 				if (error)
2316 					return (error);
2317 				break;
2318 			}
2319 			return (ENOPROTOOPT);
2320 
2321 #ifdef SOCKET_SPLICE
2322 		case SO_SPLICE:
2323 		    {
2324 			off_t len;
2325 
2326 			m->m_len = sizeof(off_t);
2327 			solock_shared(so);
2328 			len = so->so_sp ? so->so_sp->ssp_len : 0;
2329 			sounlock_shared(so);
2330 			memcpy(mtod(m, off_t *), &len, sizeof(off_t));
2331 			break;
2332 		    }
2333 #endif /* SOCKET_SPLICE */
2334 
2335 		case SO_PEERCRED:
2336 			if (so->so_proto->pr_protocol == AF_UNIX) {
2337 				struct unpcb *unp = sotounpcb(so);
2338 
2339 				solock(so);
2340 				if (unp->unp_flags & UNP_FEIDS) {
2341 					m->m_len = sizeof(unp->unp_connid);
2342 					memcpy(mtod(m, caddr_t),
2343 					    &(unp->unp_connid), m->m_len);
2344 					sounlock(so);
2345 					break;
2346 				}
2347 				sounlock(so);
2348 
2349 				return (ENOTCONN);
2350 			}
2351 			return (EOPNOTSUPP);
2352 
2353 		default:
2354 			return (ENOPROTOOPT);
2355 		}
2356 		return (0);
2357 	}
2358 }
2359 
2360 void
2361 sohasoutofband(struct socket *so)
2362 {
2363 	pgsigio(&so->so_sigio, SIGURG, 0);
2364 	knote(&so->so_rcv.sb_klist, 0);
2365 }
2366 
2367 void
2368 sofilt_lock(struct socket *so, struct sockbuf *sb)
2369 {
2370 	switch (so->so_proto->pr_domain->dom_family) {
2371 	case PF_INET:
2372 	case PF_INET6:
2373 		NET_LOCK_SHARED();
2374 		break;
2375 	default:
2376 		rw_enter_write(&so->so_lock);
2377 		break;
2378 	}
2379 
2380 	mtx_enter(&sb->sb_mtx);
2381 }
2382 
2383 void
2384 sofilt_unlock(struct socket *so, struct sockbuf *sb)
2385 {
2386 	mtx_leave(&sb->sb_mtx);
2387 
2388 	switch (so->so_proto->pr_domain->dom_family) {
2389 	case PF_INET:
2390 	case PF_INET6:
2391 		NET_UNLOCK_SHARED();
2392 		break;
2393 	default:
2394 		rw_exit_write(&so->so_lock);
2395 		break;
2396 	}
2397 }
2398 
2399 int
2400 soo_kqfilter(struct file *fp, struct knote *kn)
2401 {
2402 	struct socket *so = kn->kn_fp->f_data;
2403 	struct sockbuf *sb;
2404 
2405 	switch (kn->kn_filter) {
2406 	case EVFILT_READ:
2407 		kn->kn_fop = &soread_filtops;
2408 		sb = &so->so_rcv;
2409 		break;
2410 	case EVFILT_WRITE:
2411 		kn->kn_fop = &sowrite_filtops;
2412 		sb = &so->so_snd;
2413 		break;
2414 	case EVFILT_EXCEPT:
2415 		kn->kn_fop = &soexcept_filtops;
2416 		sb = &so->so_rcv;
2417 		break;
2418 	default:
2419 		return (EINVAL);
2420 	}
2421 
2422 	klist_insert(&sb->sb_klist, kn);
2423 
2424 	return (0);
2425 }
2426 
2427 void
2428 filt_sordetach(struct knote *kn)
2429 {
2430 	struct socket *so = kn->kn_fp->f_data;
2431 
2432 	klist_remove(&so->so_rcv.sb_klist, kn);
2433 }
2434 
2435 int
2436 filt_soread(struct knote *kn, long hint)
2437 {
2438 	struct socket *so = kn->kn_fp->f_data;
2439 	u_int state = READ_ONCE(so->so_state);
2440 	u_int error = READ_ONCE(so->so_error);
2441 	int rv = 0;
2442 
2443 	MUTEX_ASSERT_LOCKED(&so->so_rcv.sb_mtx);
2444 	if ((so->so_rcv.sb_flags & SB_MTXLOCK) == 0)
2445 		soassertlocked_readonly(so);
2446 
2447 	if (so->so_options & SO_ACCEPTCONN) {
2448 		short qlen = READ_ONCE(so->so_qlen);
2449 
2450 		if (so->so_rcv.sb_flags & SB_MTXLOCK)
2451 			soassertlocked_readonly(so);
2452 
2453 		kn->kn_data = qlen;
2454 		rv = (kn->kn_data != 0);
2455 
2456 		if (kn->kn_flags & (__EV_POLL | __EV_SELECT)) {
2457 			if (state & SS_ISDISCONNECTED) {
2458 				kn->kn_flags |= __EV_HUP;
2459 				rv = 1;
2460 			} else {
2461 				rv = qlen || soreadable(so);
2462 			}
2463 		}
2464 
2465 		return rv;
2466 	}
2467 
2468 	kn->kn_data = so->so_rcv.sb_cc;
2469 #ifdef SOCKET_SPLICE
2470 	if (isspliced(so)) {
2471 		rv = 0;
2472 	} else
2473 #endif /* SOCKET_SPLICE */
2474 	if (so->so_rcv.sb_state & SS_CANTRCVMORE) {
2475 		kn->kn_flags |= EV_EOF;
2476 		if (kn->kn_flags & __EV_POLL) {
2477 			if (state & SS_ISDISCONNECTED)
2478 				kn->kn_flags |= __EV_HUP;
2479 		}
2480 		kn->kn_fflags = error;
2481 		rv = 1;
2482 	} else if (error) {
2483 		rv = 1;
2484 	} else if (kn->kn_sfflags & NOTE_LOWAT) {
2485 		rv = (kn->kn_data >= kn->kn_sdata);
2486 	} else {
2487 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
2488 	}
2489 
2490 	return rv;
2491 }
2492 
2493 void
2494 filt_sowdetach(struct knote *kn)
2495 {
2496 	struct socket *so = kn->kn_fp->f_data;
2497 
2498 	klist_remove(&so->so_snd.sb_klist, kn);
2499 }
2500 
2501 int
2502 filt_sowrite(struct knote *kn, long hint)
2503 {
2504 	struct socket *so = kn->kn_fp->f_data;
2505 	u_int state = READ_ONCE(so->so_state);
2506 	u_int error = READ_ONCE(so->so_error);
2507 	int rv;
2508 
2509 	MUTEX_ASSERT_LOCKED(&so->so_snd.sb_mtx);
2510 	if ((so->so_snd.sb_flags & SB_MTXLOCK) == 0)
2511 		soassertlocked_readonly(so);
2512 
2513 	kn->kn_data = sbspace_locked(so, &so->so_snd);
2514 	if (so->so_snd.sb_state & SS_CANTSENDMORE) {
2515 		kn->kn_flags |= EV_EOF;
2516 		if (kn->kn_flags & __EV_POLL) {
2517 			if (state & SS_ISDISCONNECTED)
2518 				kn->kn_flags |= __EV_HUP;
2519 		}
2520 		kn->kn_fflags = error;
2521 		rv = 1;
2522 	} else if (error) {
2523 		rv = 1;
2524 	} else if (((state & SS_ISCONNECTED) == 0) &&
2525 	    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
2526 		rv = 0;
2527 	} else if (kn->kn_sfflags & NOTE_LOWAT) {
2528 		rv = (kn->kn_data >= kn->kn_sdata);
2529 	} else {
2530 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
2531 	}
2532 
2533 	return (rv);
2534 }
2535 
2536 int
2537 filt_soexcept(struct knote *kn, long hint)
2538 {
2539 	struct socket *so = kn->kn_fp->f_data;
2540 	int rv = 0;
2541 
2542 	MUTEX_ASSERT_LOCKED(&so->so_rcv.sb_mtx);
2543 	if ((so->so_rcv.sb_flags & SB_MTXLOCK) == 0)
2544 		soassertlocked_readonly(so);
2545 
2546 #ifdef SOCKET_SPLICE
2547 	if (isspliced(so)) {
2548 		rv = 0;
2549 	} else
2550 #endif /* SOCKET_SPLICE */
2551 	if (kn->kn_sfflags & NOTE_OOB) {
2552 		if (so->so_oobmark || (so->so_rcv.sb_state & SS_RCVATMARK)) {
2553 			kn->kn_fflags |= NOTE_OOB;
2554 			kn->kn_data -= so->so_oobmark;
2555 			rv = 1;
2556 		}
2557 	}
2558 
2559 	if (kn->kn_flags & __EV_POLL) {
2560 		u_int state = READ_ONCE(so->so_state);
2561 
2562 		if (state & SS_ISDISCONNECTED) {
2563 			kn->kn_flags |= __EV_HUP;
2564 			rv = 1;
2565 		}
2566 	}
2567 
2568 	return rv;
2569 }
2570 
2571 int
2572 filt_sowmodify(struct kevent *kev, struct knote *kn)
2573 {
2574 	struct socket *so = kn->kn_fp->f_data;
2575 	int rv;
2576 
2577 	sofilt_lock(so, &so->so_snd);
2578 	rv = knote_modify(kev, kn);
2579 	sofilt_unlock(so, &so->so_snd);
2580 
2581 	return (rv);
2582 }
2583 
2584 int
2585 filt_sowprocess(struct knote *kn, struct kevent *kev)
2586 {
2587 	struct socket *so = kn->kn_fp->f_data;
2588 	int rv;
2589 
2590 	sofilt_lock(so, &so->so_snd);
2591 	rv = knote_process(kn, kev);
2592 	sofilt_unlock(so, &so->so_snd);
2593 
2594 	return (rv);
2595 }
2596 
2597 int
2598 filt_sormodify(struct kevent *kev, struct knote *kn)
2599 {
2600 	struct socket *so = kn->kn_fp->f_data;
2601 	int rv;
2602 
2603 	sofilt_lock(so, &so->so_rcv);
2604 	rv = knote_modify(kev, kn);
2605 	sofilt_unlock(so, &so->so_rcv);
2606 
2607 	return (rv);
2608 }
2609 
2610 int
2611 filt_sorprocess(struct knote *kn, struct kevent *kev)
2612 {
2613 	struct socket *so = kn->kn_fp->f_data;
2614 	int rv;
2615 
2616 	sofilt_lock(so, &so->so_rcv);
2617 	rv = knote_process(kn, kev);
2618 	sofilt_unlock(so, &so->so_rcv);
2619 
2620 	return (rv);
2621 }
2622 
2623 #ifdef DDB
2624 void
2625 sobuf_print(struct sockbuf *,
2626     int (*)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))));
2627 
2628 void
2629 sobuf_print(struct sockbuf *sb,
2630     int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2631 {
2632 	(*pr)("\tsb_cc: %lu\n", sb->sb_cc);
2633 	(*pr)("\tsb_datacc: %lu\n", sb->sb_datacc);
2634 	(*pr)("\tsb_hiwat: %lu\n", sb->sb_hiwat);
2635 	(*pr)("\tsb_wat: %lu\n", sb->sb_wat);
2636 	(*pr)("\tsb_mbcnt: %lu\n", sb->sb_mbcnt);
2637 	(*pr)("\tsb_mbmax: %lu\n", sb->sb_mbmax);
2638 	(*pr)("\tsb_lowat: %ld\n", sb->sb_lowat);
2639 	(*pr)("\tsb_mb: %p\n", sb->sb_mb);
2640 	(*pr)("\tsb_mbtail: %p\n", sb->sb_mbtail);
2641 	(*pr)("\tsb_lastrecord: %p\n", sb->sb_lastrecord);
2642 	(*pr)("\tsb_flags: %04x\n", sb->sb_flags);
2643 	(*pr)("\tsb_state: %04x\n", sb->sb_state);
2644 	(*pr)("\tsb_timeo_nsecs: %llu\n", sb->sb_timeo_nsecs);
2645 }
2646 
2647 void
2648 so_print(void *v,
2649     int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2650 {
2651 	struct socket *so = v;
2652 
2653 	(*pr)("socket %p\n", so);
2654 	(*pr)("so_type: %i\n", so->so_type);
2655 	(*pr)("so_options: 0x%04x\n", so->so_options); /* %b */
2656 	(*pr)("so_linger: %i\n", so->so_linger);
2657 	(*pr)("so_state: 0x%04x\n", so->so_state);
2658 	(*pr)("so_pcb: %p\n", so->so_pcb);
2659 	(*pr)("so_proto: %p\n", so->so_proto);
2660 	(*pr)("so_sigio: %p\n", so->so_sigio.sir_sigio);
2661 
2662 	(*pr)("so_head: %p\n", so->so_head);
2663 	(*pr)("so_onq: %p\n", so->so_onq);
2664 	(*pr)("so_q0: @%p first: %p\n", &so->so_q0, TAILQ_FIRST(&so->so_q0));
2665 	(*pr)("so_q: @%p first: %p\n", &so->so_q, TAILQ_FIRST(&so->so_q));
2666 	(*pr)("so_eq: next: %p\n", TAILQ_NEXT(so, so_qe));
2667 	(*pr)("so_q0len: %i\n", so->so_q0len);
2668 	(*pr)("so_qlen: %i\n", so->so_qlen);
2669 	(*pr)("so_qlimit: %i\n", so->so_qlimit);
2670 	(*pr)("so_timeo: %i\n", so->so_timeo);
2671 	(*pr)("so_obmark: %lu\n", so->so_oobmark);
2672 
2673 	(*pr)("so_sp: %p\n", so->so_sp);
2674 	if (so->so_sp != NULL) {
2675 		(*pr)("\tssp_socket: %p\n", so->so_sp->ssp_socket);
2676 		(*pr)("\tssp_soback: %p\n", so->so_sp->ssp_soback);
2677 		(*pr)("\tssp_len: %lld\n",
2678 		    (unsigned long long)so->so_sp->ssp_len);
2679 		(*pr)("\tssp_max: %lld\n",
2680 		    (unsigned long long)so->so_sp->ssp_max);
2681 		(*pr)("\tssp_idletv: %lld %ld\n", so->so_sp->ssp_idletv.tv_sec,
2682 		    so->so_sp->ssp_idletv.tv_usec);
2683 		(*pr)("\tssp_idleto: %spending (@%i)\n",
2684 		    timeout_pending(&so->so_sp->ssp_idleto) ? "" : "not ",
2685 		    so->so_sp->ssp_idleto.to_time);
2686 	}
2687 
2688 	(*pr)("so_rcv:\n");
2689 	sobuf_print(&so->so_rcv, pr);
2690 	(*pr)("so_snd:\n");
2691 	sobuf_print(&so->so_snd, pr);
2692 
2693 	(*pr)("so_upcall: %p so_upcallarg: %p\n",
2694 	    so->so_upcall, so->so_upcallarg);
2695 
2696 	(*pr)("so_euid: %d so_ruid: %d\n", so->so_euid, so->so_ruid);
2697 	(*pr)("so_egid: %d so_rgid: %d\n", so->so_egid, so->so_rgid);
2698 	(*pr)("so_cpid: %d\n", so->so_cpid);
2699 }
2700 #endif
2701