xref: /openbsd-src/sys/kern/uipc_socket.c (revision c7e8ea31cd41a963f06f0a8ba93948b06aa6b4a4)
1 /*	$OpenBSD: uipc_socket.c,v 1.202 2017/08/22 09:13:36 mpi 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/kernel.h>
44 #include <sys/event.h>
45 #include <sys/protosw.h>
46 #include <sys/socket.h>
47 #include <sys/unpcb.h>
48 #include <sys/socketvar.h>
49 #include <sys/signalvar.h>
50 #include <net/if.h>
51 #include <sys/pool.h>
52 
53 #ifdef DDB
54 #include <machine/db_machdep.h>
55 #endif
56 
57 void	sbsync(struct sockbuf *, struct mbuf *);
58 
59 int	sosplice(struct socket *, int, off_t, struct timeval *);
60 void	sounsplice(struct socket *, struct socket *, int);
61 void	soidle(void *);
62 void	sotask(void *);
63 int	somove(struct socket *, int);
64 
65 void	filt_sordetach(struct knote *kn);
66 int	filt_soread(struct knote *kn, long hint);
67 void	filt_sowdetach(struct knote *kn);
68 int	filt_sowrite(struct knote *kn, long hint);
69 int	filt_solisten(struct knote *kn, long hint);
70 
71 struct filterops solisten_filtops =
72 	{ 1, NULL, filt_sordetach, filt_solisten };
73 struct filterops soread_filtops =
74 	{ 1, NULL, filt_sordetach, filt_soread };
75 struct filterops sowrite_filtops =
76 	{ 1, NULL, filt_sowdetach, filt_sowrite };
77 
78 
79 #ifndef SOMINCONN
80 #define SOMINCONN 80
81 #endif /* SOMINCONN */
82 
83 int	somaxconn = SOMAXCONN;
84 int	sominconn = SOMINCONN;
85 
86 struct pool socket_pool;
87 #ifdef SOCKET_SPLICE
88 struct pool sosplice_pool;
89 struct taskq *sosplice_taskq;
90 #endif
91 
92 void
93 soinit(void)
94 {
95 	pool_init(&socket_pool, sizeof(struct socket), 0, IPL_SOFTNET, 0,
96 	    "sockpl", NULL);
97 #ifdef SOCKET_SPLICE
98 	pool_init(&sosplice_pool, sizeof(struct sosplice), 0, IPL_SOFTNET, 0,
99 	    "sosppl", NULL);
100 #endif
101 }
102 
103 /*
104  * Socket operation routines.
105  * These routines are called by the routines in
106  * sys_socket.c or from a system process, and
107  * implement the semantics of socket operations by
108  * switching out to the protocol specific routines.
109  */
110 int
111 socreate(int dom, struct socket **aso, int type, int proto)
112 {
113 	struct proc *p = curproc;		/* XXX */
114 	struct protosw *prp;
115 	struct socket *so;
116 	int error, s;
117 
118 	if (proto)
119 		prp = pffindproto(dom, proto, type);
120 	else
121 		prp = pffindtype(dom, type);
122 	if (prp == NULL || prp->pr_attach == NULL)
123 		return (EPROTONOSUPPORT);
124 	if (prp->pr_type != type)
125 		return (EPROTOTYPE);
126 	so = pool_get(&socket_pool, PR_WAITOK | PR_ZERO);
127 	TAILQ_INIT(&so->so_q0);
128 	TAILQ_INIT(&so->so_q);
129 	so->so_type = type;
130 	if (suser(p, 0) == 0)
131 		so->so_state = SS_PRIV;
132 	so->so_ruid = p->p_ucred->cr_ruid;
133 	so->so_euid = p->p_ucred->cr_uid;
134 	so->so_rgid = p->p_ucred->cr_rgid;
135 	so->so_egid = p->p_ucred->cr_gid;
136 	so->so_cpid = p->p_p->ps_pid;
137 	so->so_proto = prp;
138 
139 	s = solock(so);
140 	error = (*prp->pr_attach)(so, proto);
141 	if (error) {
142 		so->so_state |= SS_NOFDREF;
143 		sofree(so);
144 		sounlock(s);
145 		return (error);
146 	}
147 	sounlock(s);
148 	*aso = so;
149 	return (0);
150 }
151 
152 int
153 sobind(struct socket *so, struct mbuf *nam, struct proc *p)
154 {
155 	int error;
156 
157 	soassertlocked(so);
158 
159 	error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, NULL, nam, NULL, p);
160 	return (error);
161 }
162 
163 int
164 solisten(struct socket *so, int backlog)
165 {
166 	int s, error;
167 
168 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING|SS_ISDISCONNECTING))
169 		return (EOPNOTSUPP);
170 #ifdef SOCKET_SPLICE
171 	if (isspliced(so) || issplicedback(so))
172 		return (EOPNOTSUPP);
173 #endif /* SOCKET_SPLICE */
174 	s = solock(so);
175 	error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, NULL, NULL, NULL,
176 	    curproc);
177 	if (error) {
178 		sounlock(s);
179 		return (error);
180 	}
181 	if (TAILQ_FIRST(&so->so_q) == NULL)
182 		so->so_options |= SO_ACCEPTCONN;
183 	if (backlog < 0 || backlog > somaxconn)
184 		backlog = somaxconn;
185 	if (backlog < sominconn)
186 		backlog = sominconn;
187 	so->so_qlimit = backlog;
188 	sounlock(s);
189 	return (0);
190 }
191 
192 void
193 sofree(struct socket *so)
194 {
195 	soassertlocked(so);
196 
197 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0)
198 		return;
199 	if (so->so_head) {
200 		/*
201 		 * We must not decommission a socket that's on the accept(2)
202 		 * queue.  If we do, then accept(2) may hang after select(2)
203 		 * indicated that the listening socket was ready.
204 		 */
205 		if (!soqremque(so, 0))
206 			return;
207 	}
208 #ifdef SOCKET_SPLICE
209 	if (so->so_sp) {
210 		if (issplicedback(so))
211 			sounsplice(so->so_sp->ssp_soback, so,
212 			    so->so_sp->ssp_soback != so);
213 		if (isspliced(so))
214 			sounsplice(so, so->so_sp->ssp_socket, 0);
215 		pool_put(&sosplice_pool, so->so_sp);
216 		so->so_sp = NULL;
217 	}
218 #endif /* SOCKET_SPLICE */
219 	sbrelease(so, &so->so_snd);
220 	sorflush(so);
221 	pool_put(&socket_pool, so);
222 }
223 
224 /*
225  * Close a socket on last file table reference removal.
226  * Initiate disconnect if connected.
227  * Free socket when disconnect complete.
228  */
229 int
230 soclose(struct socket *so)
231 {
232 	struct socket *so2;
233 	int s, error = 0;
234 
235 	s = solock(so);
236 	if (so->so_options & SO_ACCEPTCONN) {
237 		while ((so2 = TAILQ_FIRST(&so->so_q0)) != NULL) {
238 			(void) soqremque(so2, 0);
239 			(void) soabort(so2);
240 		}
241 		while ((so2 = TAILQ_FIRST(&so->so_q)) != NULL) {
242 			(void) soqremque(so2, 1);
243 			(void) soabort(so2);
244 		}
245 	}
246 	if (so->so_pcb == 0)
247 		goto discard;
248 	if (so->so_state & SS_ISCONNECTED) {
249 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
250 			error = sodisconnect(so);
251 			if (error)
252 				goto drop;
253 		}
254 		if (so->so_options & SO_LINGER) {
255 			if ((so->so_state & SS_ISDISCONNECTING) &&
256 			    (so->so_state & SS_NBIO))
257 				goto drop;
258 			while (so->so_state & SS_ISCONNECTED) {
259 				error = sosleep(so, &so->so_timeo,
260 				    PSOCK | PCATCH, "netcls",
261 				    so->so_linger * hz);
262 				if (error)
263 					break;
264 			}
265 		}
266 	}
267 drop:
268 	if (so->so_pcb) {
269 		int error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH, NULL,
270 		    NULL, NULL, curproc);
271 		if (error == 0)
272 			error = error2;
273 	}
274 discard:
275 	if (so->so_state & SS_NOFDREF)
276 		panic("soclose NOFDREF: so %p, so_type %d", so, so->so_type);
277 	so->so_state |= SS_NOFDREF;
278 	sofree(so);
279 	sounlock(s);
280 	return (error);
281 }
282 
283 int
284 soabort(struct socket *so)
285 {
286 	soassertlocked(so);
287 
288 	return (*so->so_proto->pr_usrreq)(so, PRU_ABORT, NULL, NULL, NULL,
289 	   curproc);
290 }
291 
292 int
293 soaccept(struct socket *so, struct mbuf *nam)
294 {
295 	int error = 0;
296 
297 	soassertlocked(so);
298 
299 	if ((so->so_state & SS_NOFDREF) == 0)
300 		panic("soaccept !NOFDREF: so %p, so_type %d", so, so->so_type);
301 	so->so_state &= ~SS_NOFDREF;
302 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
303 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
304 		error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT, NULL,
305 		    nam, NULL, curproc);
306 	else
307 		error = ECONNABORTED;
308 	return (error);
309 }
310 
311 int
312 soconnect(struct socket *so, struct mbuf *nam)
313 {
314 	int error;
315 
316 	soassertlocked(so);
317 
318 	if (so->so_options & SO_ACCEPTCONN)
319 		return (EOPNOTSUPP);
320 	/*
321 	 * If protocol is connection-based, can only connect once.
322 	 * Otherwise, if connected, try to disconnect first.
323 	 * This allows user to disconnect by connecting to, e.g.,
324 	 * a null address.
325 	 */
326 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
327 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
328 	    (error = sodisconnect(so))))
329 		error = EISCONN;
330 	else
331 		error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT,
332 		    NULL, nam, NULL, curproc);
333 	return (error);
334 }
335 
336 int
337 soconnect2(struct socket *so1, struct socket *so2)
338 {
339 	int error;
340 
341 	soassertlocked(so1);
342 	error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2, NULL,
343 	    (struct mbuf *)so2, NULL, curproc);
344 	return (error);
345 }
346 
347 int
348 sodisconnect(struct socket *so)
349 {
350 	int error;
351 
352 	soassertlocked(so);
353 
354 	if ((so->so_state & SS_ISCONNECTED) == 0)
355 		return (ENOTCONN);
356 	if (so->so_state & SS_ISDISCONNECTING)
357 		return (EALREADY);
358 	error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT, NULL, NULL,
359 	    NULL, curproc);
360 	return (error);
361 }
362 
363 int m_getuio(struct mbuf **, int, long, struct uio *);
364 
365 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
366 /*
367  * Send on a socket.
368  * If send must go all at once and message is larger than
369  * send buffering, then hard error.
370  * Lock against other senders.
371  * If must go all at once and not enough room now, then
372  * inform user that this would block and do nothing.
373  * Otherwise, if nonblocking, send as much as possible.
374  * The data to be sent is described by "uio" if nonzero,
375  * otherwise by the mbuf chain "top" (which must be null
376  * if uio is not).  Data provided in mbuf chain must be small
377  * enough to send all at once.
378  *
379  * Returns nonzero on error, timeout or signal; callers
380  * must check for short counts if EINTR/ERESTART are returned.
381  * Data and control buffers are freed on return.
382  */
383 int
384 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
385     struct mbuf *control, int flags)
386 {
387 	long space, clen = 0;
388 	size_t resid;
389 	int error, s;
390 	int atomic = sosendallatonce(so) || top;
391 
392 	if (uio)
393 		resid = uio->uio_resid;
394 	else
395 		resid = top->m_pkthdr.len;
396 	/* MSG_EOR on a SOCK_STREAM socket is invalid. */
397 	if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
398 		m_freem(top);
399 		m_freem(control);
400 		return (EINVAL);
401 	}
402 	if (uio && uio->uio_procp)
403 		uio->uio_procp->p_ru.ru_msgsnd++;
404 	if (control) {
405 		/*
406 		 * In theory clen should be unsigned (since control->m_len is).
407 		 * However, space must be signed, as it might be less than 0
408 		 * if we over-committed, and we must use a signed comparison
409 		 * of space and clen.
410 		 */
411 		clen = control->m_len;
412 		/* reserve extra space for AF_LOCAL's internalize */
413 		if (so->so_proto->pr_domain->dom_family == AF_LOCAL &&
414 		    clen >= CMSG_ALIGN(sizeof(struct cmsghdr)) &&
415 		    mtod(control, struct cmsghdr *)->cmsg_type == SCM_RIGHTS)
416 			clen = CMSG_SPACE(
417 			    (clen - CMSG_ALIGN(sizeof(struct cmsghdr))) *
418 			    (sizeof(struct fdpass) / sizeof(int)));
419 	}
420 
421 #define	snderr(errno)	{ error = errno; goto release; }
422 
423 	s = solock(so);
424 restart:
425 	if ((error = sblock(so, &so->so_snd, SBLOCKWAIT(flags))) != 0)
426 		goto out;
427 	so->so_state |= SS_ISSENDING;
428 	do {
429 		if (so->so_state & SS_CANTSENDMORE)
430 			snderr(EPIPE);
431 		if (so->so_error) {
432 			error = so->so_error;
433 			so->so_error = 0;
434 			snderr(error);
435 		}
436 		if ((so->so_state & SS_ISCONNECTED) == 0) {
437 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
438 				if (!(resid == 0 && clen != 0))
439 					snderr(ENOTCONN);
440 			} else if (addr == 0)
441 				snderr(EDESTADDRREQ);
442 		}
443 		space = sbspace(so, &so->so_snd);
444 		if (flags & MSG_OOB)
445 			space += 1024;
446 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
447 		    (so->so_proto->pr_domain->dom_family != AF_LOCAL &&
448 		    clen > so->so_snd.sb_hiwat))
449 			snderr(EMSGSIZE);
450 		if (space < clen ||
451 		    (space - clen < resid &&
452 		    (atomic || space < so->so_snd.sb_lowat))) {
453 			if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT))
454 				snderr(EWOULDBLOCK);
455 			sbunlock(&so->so_snd);
456 			error = sbwait(so, &so->so_snd);
457 			so->so_state &= ~SS_ISSENDING;
458 			if (error)
459 				goto out;
460 			goto restart;
461 		}
462 		space -= clen;
463 		do {
464 			if (uio == NULL) {
465 				/*
466 				 * Data is prepackaged in "top".
467 				 */
468 				resid = 0;
469 				if (flags & MSG_EOR)
470 					top->m_flags |= M_EOR;
471 			} else {
472 				sounlock(s);
473 				error = m_getuio(&top, atomic, space, uio);
474 				s = solock(so);
475 				if (error)
476 					goto release;
477 				space -= top->m_pkthdr.len;
478 				resid = uio->uio_resid;
479 				if (flags & MSG_EOR)
480 					top->m_flags |= M_EOR;
481 			}
482 			if (resid == 0)
483 				so->so_state &= ~SS_ISSENDING;
484 			if (top && so->so_options & SO_ZEROIZE)
485 				top->m_flags |= M_ZEROIZE;
486 			error = (*so->so_proto->pr_usrreq)(so,
487 			    (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND,
488 			    top, addr, control, curproc);
489 			clen = 0;
490 			control = NULL;
491 			top = NULL;
492 			if (error)
493 				goto release;
494 		} while (resid && space > 0);
495 	} while (resid);
496 
497 release:
498 	so->so_state &= ~SS_ISSENDING;
499 	sbunlock(&so->so_snd);
500 out:
501 	sounlock(s);
502 	m_freem(top);
503 	m_freem(control);
504 	return (error);
505 }
506 
507 int
508 m_getuio(struct mbuf **mp, int atomic, long space, struct uio *uio)
509 {
510 	struct mbuf *m, *top = NULL;
511 	struct mbuf **nextp = &top;
512 	u_long len, mlen;
513 	size_t resid = uio->uio_resid;
514 	int error;
515 
516 	do {
517 		if (top == NULL) {
518 			MGETHDR(m, M_WAIT, MT_DATA);
519 			mlen = MHLEN;
520 			m->m_pkthdr.len = 0;
521 			m->m_pkthdr.ph_ifidx = 0;
522 		} else {
523 			MGET(m, M_WAIT, MT_DATA);
524 			mlen = MLEN;
525 		}
526 		/* chain mbuf together */
527 		*nextp = m;
528 		nextp = &m->m_next;
529 
530 		resid = ulmin(resid, space);
531 		if (resid >= MINCLSIZE) {
532 			MCLGETI(m, M_NOWAIT, NULL, ulmin(resid, MAXMCLBYTES));
533 			if ((m->m_flags & M_EXT) == 0)
534 				MCLGETI(m, M_NOWAIT, NULL, MCLBYTES);
535 			if ((m->m_flags & M_EXT) == 0)
536 				goto nopages;
537 			mlen = m->m_ext.ext_size;
538 			len = ulmin(mlen, resid);
539 			/*
540 			 * For datagram protocols, leave room
541 			 * for protocol headers in first mbuf.
542 			 */
543 			if (atomic && top == NULL && len < mlen - max_hdr)
544 				m->m_data += max_hdr;
545 		} else {
546 nopages:
547 			len = ulmin(mlen, resid);
548 			/*
549 			 * For datagram protocols, leave room
550 			 * for protocol headers in first mbuf.
551 			 */
552 			if (atomic && top == NULL && len < mlen - max_hdr)
553 				MH_ALIGN(m, len);
554 		}
555 
556 		error = uiomove(mtod(m, caddr_t), len, uio);
557 		if (error) {
558 			m_freem(top);
559 			return (error);
560 		}
561 
562 		/* adjust counters */
563 		resid = uio->uio_resid;
564 		space -= len;
565 		m->m_len = len;
566 		top->m_pkthdr.len += len;
567 
568 		/* Is there more space and more data? */
569 	} while (space > 0 && resid > 0);
570 
571 	*mp = top;
572 	return 0;
573 }
574 
575 /*
576  * Following replacement or removal of the first mbuf on the first
577  * mbuf chain of a socket buffer, push necessary state changes back
578  * into the socket buffer so that other consumers see the values
579  * consistently.  'nextrecord' is the callers locally stored value of
580  * the original value of sb->sb_mb->m_nextpkt which must be restored
581  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
582  */
583 void
584 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
585 {
586 
587 	/*
588 	 * First, update for the new value of nextrecord.  If necessary,
589 	 * make it the first record.
590 	 */
591 	if (sb->sb_mb != NULL)
592 		sb->sb_mb->m_nextpkt = nextrecord;
593 	else
594 		sb->sb_mb = nextrecord;
595 
596 	/*
597 	 * Now update any dependent socket buffer fields to reflect
598 	 * the new state.  This is an inline of SB_EMPTY_FIXUP, with
599 	 * the addition of a second clause that takes care of the
600 	 * case where sb_mb has been updated, but remains the last
601 	 * record.
602 	 */
603 	if (sb->sb_mb == NULL) {
604 		sb->sb_mbtail = NULL;
605 		sb->sb_lastrecord = NULL;
606 	} else if (sb->sb_mb->m_nextpkt == NULL)
607 		sb->sb_lastrecord = sb->sb_mb;
608 }
609 
610 /*
611  * Implement receive operations on a socket.
612  * We depend on the way that records are added to the sockbuf
613  * by sbappend*.  In particular, each record (mbufs linked through m_next)
614  * must begin with an address if the protocol so specifies,
615  * followed by an optional mbuf or mbufs containing ancillary data,
616  * and then zero or more mbufs of data.
617  * In order to avoid blocking network for the entire time here, we release
618  * the solock() while doing the actual copy to user space.
619  * Although the sockbuf is locked, new data may still be appended,
620  * and thus we must maintain consistency of the sockbuf during that time.
621  *
622  * The caller may receive the data as a single mbuf chain by supplying
623  * an mbuf **mp0 for use in returning the chain.  The uio is then used
624  * only for the count in uio_resid.
625  */
626 int
627 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
628     struct mbuf **mp0, struct mbuf **controlp, int *flagsp,
629     socklen_t controllen)
630 {
631 	struct mbuf *m, **mp;
632 	struct mbuf *cm;
633 	u_long len, offset, moff;
634 	int flags, error, s, type, uio_error = 0;
635 	struct protosw *pr = so->so_proto;
636 	struct mbuf *nextrecord;
637 	size_t resid, orig_resid = uio->uio_resid;
638 
639 	mp = mp0;
640 	if (paddr)
641 		*paddr = 0;
642 	if (controlp)
643 		*controlp = 0;
644 	if (flagsp)
645 		flags = *flagsp &~ MSG_EOR;
646 	else
647 		flags = 0;
648 	if (so->so_state & SS_NBIO)
649 		flags |= MSG_DONTWAIT;
650 	if (flags & MSG_OOB) {
651 		m = m_get(M_WAIT, MT_DATA);
652 		s = solock(so);
653 		error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m,
654 		    (struct mbuf *)(long)(flags & MSG_PEEK), NULL, curproc);
655 		sounlock(s);
656 		if (error)
657 			goto bad;
658 		do {
659 			error = uiomove(mtod(m, caddr_t),
660 			    ulmin(uio->uio_resid, m->m_len), uio);
661 			m = m_free(m);
662 		} while (uio->uio_resid && error == 0 && m);
663 bad:
664 		m_freem(m);
665 		return (error);
666 	}
667 	if (mp)
668 		*mp = NULL;
669 
670 restart:
671 	s = solock(so);
672 	if ((error = sblock(so, &so->so_rcv, SBLOCKWAIT(flags))) != 0) {
673 		sounlock(s);
674 		return (error);
675 	}
676 
677 	m = so->so_rcv.sb_mb;
678 #ifdef SOCKET_SPLICE
679 	if (isspliced(so))
680 		m = NULL;
681 #endif /* SOCKET_SPLICE */
682 	/*
683 	 * If we have less data than requested, block awaiting more
684 	 * (subject to any timeout) if:
685 	 *   1. the current count is less than the low water mark,
686 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
687 	 *	receive operation at once if we block (resid <= hiwat), or
688 	 *   3. MSG_DONTWAIT is not set.
689 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
690 	 * we have to do the receive in sections, and thus risk returning
691 	 * a short count if a timeout or signal occurs after we start.
692 	 */
693 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
694 	    so->so_rcv.sb_cc < uio->uio_resid) &&
695 	    (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
696 	    ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
697 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
698 #ifdef DIAGNOSTIC
699 		if (m == NULL && so->so_rcv.sb_cc)
700 #ifdef SOCKET_SPLICE
701 		    if (!isspliced(so))
702 #endif /* SOCKET_SPLICE */
703 			panic("receive 1: so %p, so_type %d, sb_cc %lu",
704 			    so, so->so_type, so->so_rcv.sb_cc);
705 #endif
706 		if (so->so_error) {
707 			if (m)
708 				goto dontblock;
709 			error = so->so_error;
710 			if ((flags & MSG_PEEK) == 0)
711 				so->so_error = 0;
712 			goto release;
713 		}
714 		if (so->so_state & SS_CANTRCVMORE) {
715 			if (m)
716 				goto dontblock;
717 			else if (so->so_rcv.sb_cc == 0)
718 				goto release;
719 		}
720 		for (; m; m = m->m_next)
721 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
722 				m = so->so_rcv.sb_mb;
723 				goto dontblock;
724 			}
725 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
726 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
727 			error = ENOTCONN;
728 			goto release;
729 		}
730 		if (uio->uio_resid == 0 && controlp == NULL)
731 			goto release;
732 		if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT)) {
733 			error = EWOULDBLOCK;
734 			goto release;
735 		}
736 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
737 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
738 		sbunlock(&so->so_rcv);
739 		error = sbwait(so, &so->so_rcv);
740 		sounlock(s);
741 		if (error)
742 			return (error);
743 		goto restart;
744 	}
745 dontblock:
746 	/*
747 	 * On entry here, m points to the first record of the socket buffer.
748 	 * From this point onward, we maintain 'nextrecord' as a cache of the
749 	 * pointer to the next record in the socket buffer.  We must keep the
750 	 * various socket buffer pointers and local stack versions of the
751 	 * pointers in sync, pushing out modifications before operations that
752 	 * may sleep, and re-reading them afterwards.
753 	 *
754 	 * Otherwise, we will race with the network stack appending new data
755 	 * or records onto the socket buffer by using inconsistent/stale
756 	 * versions of the field, possibly resulting in socket buffer
757 	 * corruption.
758 	 */
759 	if (uio->uio_procp)
760 		uio->uio_procp->p_ru.ru_msgrcv++;
761 	KASSERT(m == so->so_rcv.sb_mb);
762 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
763 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
764 	nextrecord = m->m_nextpkt;
765 	if (pr->pr_flags & PR_ADDR) {
766 #ifdef DIAGNOSTIC
767 		if (m->m_type != MT_SONAME)
768 			panic("receive 1a: so %p, so_type %d, m %p, m_type %d",
769 			    so, so->so_type, m, m->m_type);
770 #endif
771 		orig_resid = 0;
772 		if (flags & MSG_PEEK) {
773 			if (paddr)
774 				*paddr = m_copym(m, 0, m->m_len, M_NOWAIT);
775 			m = m->m_next;
776 		} else {
777 			sbfree(&so->so_rcv, m);
778 			if (paddr) {
779 				*paddr = m;
780 				so->so_rcv.sb_mb = m->m_next;
781 				m->m_next = 0;
782 				m = so->so_rcv.sb_mb;
783 			} else {
784 				so->so_rcv.sb_mb = m_free(m);
785 				m = so->so_rcv.sb_mb;
786 			}
787 			sbsync(&so->so_rcv, nextrecord);
788 		}
789 	}
790 	while (m && m->m_type == MT_CONTROL && error == 0) {
791 		if (flags & MSG_PEEK) {
792 			if (controlp)
793 				*controlp = m_copym(m, 0, m->m_len, M_NOWAIT);
794 			m = m->m_next;
795 		} else {
796 			sbfree(&so->so_rcv, m);
797 			so->so_rcv.sb_mb = m->m_next;
798 			m->m_nextpkt = m->m_next = NULL;
799 			cm = m;
800 			m = so->so_rcv.sb_mb;
801 			sbsync(&so->so_rcv, nextrecord);
802 			if (controlp) {
803 				if (pr->pr_domain->dom_externalize &&
804 				    mtod(cm, struct cmsghdr *)->cmsg_type ==
805 				    SCM_RIGHTS) {
806 					error =
807 					    (*pr->pr_domain->dom_externalize)
808 					    (cm, controllen, flags);
809 				}
810 				*controlp = cm;
811 			} else {
812 				/*
813 				 * Dispose of any SCM_RIGHTS message that went
814 				 * through the read path rather than recv.
815 				 */
816 				if (pr->pr_domain->dom_dispose &&
817 				    mtod(cm, struct cmsghdr *)->cmsg_type == SCM_RIGHTS)
818 					pr->pr_domain->dom_dispose(cm);
819 				m_free(cm);
820 			}
821 		}
822 		if (m != NULL)
823 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
824 		else
825 			nextrecord = so->so_rcv.sb_mb;
826 		if (controlp) {
827 			orig_resid = 0;
828 			controlp = &(*controlp)->m_next;
829 		}
830 	}
831 
832 	/* If m is non-NULL, we have some data to read. */
833 	if (m) {
834 		type = m->m_type;
835 		if (type == MT_OOBDATA)
836 			flags |= MSG_OOB;
837 		if (m->m_flags & M_BCAST)
838 			flags |= MSG_BCAST;
839 		if (m->m_flags & M_MCAST)
840 			flags |= MSG_MCAST;
841 	}
842 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
843 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
844 
845 	moff = 0;
846 	offset = 0;
847 	while (m && uio->uio_resid > 0 && error == 0) {
848 		if (m->m_type == MT_OOBDATA) {
849 			if (type != MT_OOBDATA)
850 				break;
851 		} else if (type == MT_OOBDATA)
852 			break;
853 #ifdef DIAGNOSTIC
854 		else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
855 			panic("receive 3: so %p, so_type %d, m %p, m_type %d",
856 			    so, so->so_type, m, m->m_type);
857 #endif
858 		so->so_state &= ~SS_RCVATMARK;
859 		len = uio->uio_resid;
860 		if (so->so_oobmark && len > so->so_oobmark - offset)
861 			len = so->so_oobmark - offset;
862 		if (len > m->m_len - moff)
863 			len = m->m_len - moff;
864 		/*
865 		 * If mp is set, just pass back the mbufs.
866 		 * Otherwise copy them out via the uio, then free.
867 		 * Sockbuf must be consistent here (points to current mbuf,
868 		 * it points to next record) when we drop priority;
869 		 * we must note any additions to the sockbuf when we
870 		 * block interrupts again.
871 		 */
872 		if (mp == NULL && uio_error == 0) {
873 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
874 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
875 			resid = uio->uio_resid;
876 			sounlock(s);
877 			uio_error = uiomove(mtod(m, caddr_t) + moff, len, uio);
878 			s = solock(so);
879 			if (uio_error)
880 				uio->uio_resid = resid - len;
881 		} else
882 			uio->uio_resid -= len;
883 		if (len == m->m_len - moff) {
884 			if (m->m_flags & M_EOR)
885 				flags |= MSG_EOR;
886 			if (flags & MSG_PEEK) {
887 				m = m->m_next;
888 				moff = 0;
889 			} else {
890 				nextrecord = m->m_nextpkt;
891 				sbfree(&so->so_rcv, m);
892 				if (mp) {
893 					*mp = m;
894 					mp = &m->m_next;
895 					so->so_rcv.sb_mb = m = m->m_next;
896 					*mp = NULL;
897 				} else {
898 					so->so_rcv.sb_mb = m_free(m);
899 					m = so->so_rcv.sb_mb;
900 				}
901 				/*
902 				 * If m != NULL, we also know that
903 				 * so->so_rcv.sb_mb != NULL.
904 				 */
905 				KASSERT(so->so_rcv.sb_mb == m);
906 				if (m) {
907 					m->m_nextpkt = nextrecord;
908 					if (nextrecord == NULL)
909 						so->so_rcv.sb_lastrecord = m;
910 				} else {
911 					so->so_rcv.sb_mb = nextrecord;
912 					SB_EMPTY_FIXUP(&so->so_rcv);
913 				}
914 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
915 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
916 			}
917 		} else {
918 			if (flags & MSG_PEEK)
919 				moff += len;
920 			else {
921 				if (mp)
922 					*mp = m_copym(m, 0, len, M_WAIT);
923 				m->m_data += len;
924 				m->m_len -= len;
925 				so->so_rcv.sb_cc -= len;
926 				so->so_rcv.sb_datacc -= len;
927 			}
928 		}
929 		if (so->so_oobmark) {
930 			if ((flags & MSG_PEEK) == 0) {
931 				so->so_oobmark -= len;
932 				if (so->so_oobmark == 0) {
933 					so->so_state |= SS_RCVATMARK;
934 					break;
935 				}
936 			} else {
937 				offset += len;
938 				if (offset == so->so_oobmark)
939 					break;
940 			}
941 		}
942 		if (flags & MSG_EOR)
943 			break;
944 		/*
945 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
946 		 * we must not quit until "uio->uio_resid == 0" or an error
947 		 * termination.  If a signal/timeout occurs, return
948 		 * with a short count but without error.
949 		 * Keep sockbuf locked against other readers.
950 		 */
951 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
952 		    !sosendallatonce(so) && !nextrecord) {
953 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
954 				break;
955 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
956 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
957 			error = sbwait(so, &so->so_rcv);
958 			if (error) {
959 				sbunlock(&so->so_rcv);
960 				sounlock(s);
961 				return (0);
962 			}
963 			if ((m = so->so_rcv.sb_mb) != NULL)
964 				nextrecord = m->m_nextpkt;
965 		}
966 	}
967 
968 	if (m && pr->pr_flags & PR_ATOMIC) {
969 		flags |= MSG_TRUNC;
970 		if ((flags & MSG_PEEK) == 0)
971 			(void) sbdroprecord(&so->so_rcv);
972 	}
973 	if ((flags & MSG_PEEK) == 0) {
974 		if (m == NULL) {
975 			/*
976 			 * First part is an inline SB_EMPTY_FIXUP().  Second
977 			 * part makes sure sb_lastrecord is up-to-date if
978 			 * there is still data in the socket buffer.
979 			 */
980 			so->so_rcv.sb_mb = nextrecord;
981 			if (so->so_rcv.sb_mb == NULL) {
982 				so->so_rcv.sb_mbtail = NULL;
983 				so->so_rcv.sb_lastrecord = NULL;
984 			} else if (nextrecord->m_nextpkt == NULL)
985 				so->so_rcv.sb_lastrecord = nextrecord;
986 		}
987 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
988 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
989 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
990 			(*pr->pr_usrreq)(so, PRU_RCVD, NULL,
991 			    (struct mbuf *)(long)flags, NULL, curproc);
992 	}
993 	if (orig_resid == uio->uio_resid && orig_resid &&
994 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
995 		sbunlock(&so->so_rcv);
996 		sounlock(s);
997 		goto restart;
998 	}
999 
1000 	if (uio_error)
1001 		error = uio_error;
1002 
1003 	if (flagsp)
1004 		*flagsp |= flags;
1005 release:
1006 	sbunlock(&so->so_rcv);
1007 	sounlock(s);
1008 	return (error);
1009 }
1010 
1011 int
1012 soshutdown(struct socket *so, int how)
1013 {
1014 	struct protosw *pr = so->so_proto;
1015 	int s, error = 0;
1016 
1017 	s = solock(so);
1018 	switch (how) {
1019 	case SHUT_RD:
1020 		sorflush(so);
1021 		break;
1022 	case SHUT_RDWR:
1023 		sorflush(so);
1024 		/* FALLTHROUGH */
1025 	case SHUT_WR:
1026 		error = (*pr->pr_usrreq)(so, PRU_SHUTDOWN, NULL, NULL, NULL,
1027 		    curproc);
1028 		break;
1029 	default:
1030 		error = EINVAL;
1031 		break;
1032 	}
1033 	sounlock(s);
1034 
1035 	return (error);
1036 }
1037 
1038 void
1039 sorflush(struct socket *so)
1040 {
1041 	struct sockbuf *sb = &so->so_rcv;
1042 	struct protosw *pr = so->so_proto;
1043 	struct socket aso;
1044 
1045 	sb->sb_flags |= SB_NOINTR;
1046 	sblock(so, sb, M_WAITOK);
1047 	socantrcvmore(so);
1048 	sbunlock(sb);
1049 	aso.so_proto = pr;
1050 	aso.so_rcv = *sb;
1051 	memset(sb, 0, sizeof (*sb));
1052 	/* XXX - the memset stomps all over so_rcv */
1053 	if (aso.so_rcv.sb_flags & SB_KNOTE) {
1054 		sb->sb_sel.si_note = aso.so_rcv.sb_sel.si_note;
1055 		sb->sb_flags = SB_KNOTE;
1056 	}
1057 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose)
1058 		(*pr->pr_domain->dom_dispose)(aso.so_rcv.sb_mb);
1059 	sbrelease(&aso, &aso.so_rcv);
1060 }
1061 
1062 #ifdef SOCKET_SPLICE
1063 
1064 #define so_splicelen	so_sp->ssp_len
1065 #define so_splicemax	so_sp->ssp_max
1066 #define so_idletv	so_sp->ssp_idletv
1067 #define so_idleto	so_sp->ssp_idleto
1068 #define so_splicetask	so_sp->ssp_task
1069 
1070 int
1071 sosplice(struct socket *so, int fd, off_t max, struct timeval *tv)
1072 {
1073 	struct file	*fp;
1074 	struct socket	*sosp;
1075 	struct sosplice	*sp;
1076 	int		 error = 0;
1077 
1078 	soassertlocked(so);
1079 
1080 	if (sosplice_taskq == NULL)
1081 		sosplice_taskq = taskq_create("sosplice", 1, IPL_SOFTNET, 0);
1082 	if (sosplice_taskq == NULL)
1083 		return (ENOMEM);
1084 
1085 	if ((so->so_proto->pr_flags & PR_SPLICE) == 0)
1086 		return (EPROTONOSUPPORT);
1087 	if (so->so_options & SO_ACCEPTCONN)
1088 		return (EOPNOTSUPP);
1089 	if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1090 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
1091 		return (ENOTCONN);
1092 	if (so->so_sp == NULL) {
1093 		sp = pool_get(&sosplice_pool, PR_WAITOK | PR_ZERO);
1094 		if (so->so_sp == NULL)
1095 			so->so_sp = sp;
1096 		else
1097 			pool_put(&sosplice_pool, sp);
1098 	}
1099 
1100 	/* If no fd is given, unsplice by removing existing link. */
1101 	if (fd < 0) {
1102 		/* Lock receive buffer. */
1103 		if ((error = sblock(so, &so->so_rcv,
1104 		    (so->so_state & SS_NBIO) ? M_NOWAIT : M_WAITOK)) != 0) {
1105 			return (error);
1106 		}
1107 		if (so->so_sp->ssp_socket)
1108 			sounsplice(so, so->so_sp->ssp_socket, 1);
1109 		sbunlock(&so->so_rcv);
1110 		return (0);
1111 	}
1112 
1113 	if (max && max < 0)
1114 		return (EINVAL);
1115 
1116 	if (tv && (tv->tv_sec < 0 || tv->tv_usec < 0))
1117 		return (EINVAL);
1118 
1119 	/* Find sosp, the drain socket where data will be spliced into. */
1120 	if ((error = getsock(curproc, fd, &fp)) != 0)
1121 		return (error);
1122 	sosp = fp->f_data;
1123 	if (sosp->so_sp == NULL) {
1124 		sp = pool_get(&sosplice_pool, PR_WAITOK | PR_ZERO);
1125 		if (sosp->so_sp == NULL)
1126 			sosp->so_sp = sp;
1127 		else
1128 			pool_put(&sosplice_pool, sp);
1129 	}
1130 
1131 	/* Lock both receive and send buffer. */
1132 	if ((error = sblock(so, &so->so_rcv,
1133 	    (so->so_state & SS_NBIO) ? M_NOWAIT : M_WAITOK)) != 0) {
1134 		FRELE(fp, curproc);
1135 		return (error);
1136 	}
1137 	if ((error = sblock(so, &sosp->so_snd, M_WAITOK)) != 0) {
1138 		sbunlock(&so->so_rcv);
1139 		FRELE(fp, curproc);
1140 		return (error);
1141 	}
1142 
1143 	if (so->so_sp->ssp_socket || sosp->so_sp->ssp_soback) {
1144 		error = EBUSY;
1145 		goto release;
1146 	}
1147 	if (sosp->so_proto->pr_usrreq != so->so_proto->pr_usrreq) {
1148 		error = EPROTONOSUPPORT;
1149 		goto release;
1150 	}
1151 	if (sosp->so_options & SO_ACCEPTCONN) {
1152 		error = EOPNOTSUPP;
1153 		goto release;
1154 	}
1155 	if ((sosp->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0) {
1156 		error = ENOTCONN;
1157 		goto release;
1158 	}
1159 
1160 	/* Splice so and sosp together. */
1161 	so->so_sp->ssp_socket = sosp;
1162 	sosp->so_sp->ssp_soback = so;
1163 	so->so_splicelen = 0;
1164 	so->so_splicemax = max;
1165 	if (tv)
1166 		so->so_idletv = *tv;
1167 	else
1168 		timerclear(&so->so_idletv);
1169 	timeout_set_proc(&so->so_idleto, soidle, so);
1170 	task_set(&so->so_splicetask, sotask, so);
1171 
1172 	/*
1173 	 * To prevent softnet interrupt from calling somove() while
1174 	 * we sleep, the socket buffers are not marked as spliced yet.
1175 	 */
1176 	if (somove(so, M_WAIT)) {
1177 		so->so_rcv.sb_flagsintr |= SB_SPLICE;
1178 		sosp->so_snd.sb_flagsintr |= SB_SPLICE;
1179 	}
1180 
1181  release:
1182 	sbunlock(&sosp->so_snd);
1183 	sbunlock(&so->so_rcv);
1184 	FRELE(fp, curproc);
1185 	return (error);
1186 }
1187 
1188 void
1189 sounsplice(struct socket *so, struct socket *sosp, int wakeup)
1190 {
1191 	soassertlocked(so);
1192 
1193 	task_del(sosplice_taskq, &so->so_splicetask);
1194 	timeout_del(&so->so_idleto);
1195 	sosp->so_snd.sb_flagsintr &= ~SB_SPLICE;
1196 	so->so_rcv.sb_flagsintr &= ~SB_SPLICE;
1197 	so->so_sp->ssp_socket = sosp->so_sp->ssp_soback = NULL;
1198 	if (wakeup && soreadable(so))
1199 		sorwakeup(so);
1200 }
1201 
1202 void
1203 soidle(void *arg)
1204 {
1205 	struct socket *so = arg;
1206 	int s;
1207 
1208 	s = solock(so);
1209 	if (so->so_rcv.sb_flagsintr & SB_SPLICE) {
1210 		so->so_error = ETIMEDOUT;
1211 		sounsplice(so, so->so_sp->ssp_socket, 1);
1212 	}
1213 	sounlock(s);
1214 }
1215 
1216 void
1217 sotask(void *arg)
1218 {
1219 	struct socket *so = arg;
1220 	int s;
1221 
1222 	s = solock(so);
1223 	if (so->so_rcv.sb_flagsintr & SB_SPLICE) {
1224 		/*
1225 		 * We may not sleep here as sofree() and unsplice() may be
1226 		 * called from softnet interrupt context.  This would remove
1227 		 * the socket during somove().
1228 		 */
1229 		somove(so, M_DONTWAIT);
1230 	}
1231 	sounlock(s);
1232 
1233 	/* Avoid user land starvation. */
1234 	yield();
1235 }
1236 
1237 /*
1238  * Move data from receive buffer of spliced source socket to send
1239  * buffer of drain socket.  Try to move as much as possible in one
1240  * big chunk.  It is a TCP only implementation.
1241  * Return value 0 means splicing has been finished, 1 continue.
1242  */
1243 int
1244 somove(struct socket *so, int wait)
1245 {
1246 	struct socket	*sosp = so->so_sp->ssp_socket;
1247 	struct mbuf	*m, **mp, *nextrecord;
1248 	u_long		 len, off, oobmark;
1249 	long		 space;
1250 	int		 error = 0, maxreached = 0;
1251 	short		 state;
1252 
1253 	soassertlocked(so);
1254 
1255  nextpkt:
1256 	if (so->so_error) {
1257 		error = so->so_error;
1258 		goto release;
1259 	}
1260 	if (sosp->so_state & SS_CANTSENDMORE) {
1261 		error = EPIPE;
1262 		goto release;
1263 	}
1264 	if (sosp->so_error && sosp->so_error != ETIMEDOUT &&
1265 	    sosp->so_error != EFBIG && sosp->so_error != ELOOP) {
1266 		error = sosp->so_error;
1267 		goto release;
1268 	}
1269 	if ((sosp->so_state & SS_ISCONNECTED) == 0)
1270 		goto release;
1271 
1272 	/* Calculate how many bytes can be copied now. */
1273 	len = so->so_rcv.sb_datacc;
1274 	if (so->so_splicemax) {
1275 		KASSERT(so->so_splicelen < so->so_splicemax);
1276 		if (so->so_splicemax <= so->so_splicelen + len) {
1277 			len = so->so_splicemax - so->so_splicelen;
1278 			maxreached = 1;
1279 		}
1280 	}
1281 	space = sbspace(sosp, &sosp->so_snd);
1282 	if (so->so_oobmark && so->so_oobmark < len &&
1283 	    so->so_oobmark < space + 1024)
1284 		space += 1024;
1285 	if (space <= 0) {
1286 		maxreached = 0;
1287 		goto release;
1288 	}
1289 	if (space < len) {
1290 		maxreached = 0;
1291 		if (space < sosp->so_snd.sb_lowat)
1292 			goto release;
1293 		len = space;
1294 	}
1295 	sosp->so_state |= SS_ISSENDING;
1296 
1297 	SBLASTRECORDCHK(&so->so_rcv, "somove 1");
1298 	SBLASTMBUFCHK(&so->so_rcv, "somove 1");
1299 	m = so->so_rcv.sb_mb;
1300 	if (m == NULL)
1301 		goto release;
1302 	nextrecord = m->m_nextpkt;
1303 
1304 	/* Drop address and control information not used with splicing. */
1305 	if (so->so_proto->pr_flags & PR_ADDR) {
1306 #ifdef DIAGNOSTIC
1307 		if (m->m_type != MT_SONAME)
1308 			panic("somove soname: so %p, so_type %d, m %p, "
1309 			    "m_type %d", so, so->so_type, m, m->m_type);
1310 #endif
1311 		m = m->m_next;
1312 	}
1313 	while (m && m->m_type == MT_CONTROL)
1314 		m = m->m_next;
1315 	if (m == NULL) {
1316 		sbdroprecord(&so->so_rcv);
1317 		if (so->so_proto->pr_flags & PR_WANTRCVD && so->so_pcb)
1318 			(so->so_proto->pr_usrreq)(so, PRU_RCVD, NULL,
1319 			    NULL, NULL, NULL);
1320 		goto nextpkt;
1321 	}
1322 
1323 	/*
1324 	 * By splicing sockets connected to localhost, userland might create a
1325 	 * loop.  Dissolve splicing with error if loop is detected by counter.
1326 	 */
1327 	if ((m->m_flags & M_PKTHDR) && m->m_pkthdr.ph_loopcnt++ >= M_MAXLOOP) {
1328 		error = ELOOP;
1329 		goto release;
1330 	}
1331 
1332 	if (so->so_proto->pr_flags & PR_ATOMIC) {
1333 		if ((m->m_flags & M_PKTHDR) == 0)
1334 			panic("somove !PKTHDR: so %p, so_type %d, m %p, "
1335 			    "m_type %d", so, so->so_type, m, m->m_type);
1336 		if (sosp->so_snd.sb_hiwat < m->m_pkthdr.len) {
1337 			error = EMSGSIZE;
1338 			goto release;
1339 		}
1340 		if (len < m->m_pkthdr.len)
1341 			goto release;
1342 		if (m->m_pkthdr.len < len) {
1343 			maxreached = 0;
1344 			len = m->m_pkthdr.len;
1345 		}
1346 		/*
1347 		 * Throw away the name mbuf after it has been assured
1348 		 * that the whole first record can be processed.
1349 		 */
1350 		m = so->so_rcv.sb_mb;
1351 		sbfree(&so->so_rcv, m);
1352 		so->so_rcv.sb_mb = m_free(m);
1353 		sbsync(&so->so_rcv, nextrecord);
1354 	}
1355 	/*
1356 	 * Throw away the control mbufs after it has been assured
1357 	 * that the whole first record can be processed.
1358 	 */
1359 	m = so->so_rcv.sb_mb;
1360 	while (m && m->m_type == MT_CONTROL) {
1361 		sbfree(&so->so_rcv, m);
1362 		so->so_rcv.sb_mb = m_free(m);
1363 		m = so->so_rcv.sb_mb;
1364 		sbsync(&so->so_rcv, nextrecord);
1365 	}
1366 
1367 	SBLASTRECORDCHK(&so->so_rcv, "somove 2");
1368 	SBLASTMBUFCHK(&so->so_rcv, "somove 2");
1369 
1370 	/* Take at most len mbufs out of receive buffer. */
1371 	for (off = 0, mp = &m; off <= len && *mp;
1372 	    off += (*mp)->m_len, mp = &(*mp)->m_next) {
1373 		u_long size = len - off;
1374 
1375 #ifdef DIAGNOSTIC
1376 		if ((*mp)->m_type != MT_DATA && (*mp)->m_type != MT_HEADER)
1377 			panic("somove type: so %p, so_type %d, m %p, "
1378 			    "m_type %d", so, so->so_type, *mp, (*mp)->m_type);
1379 #endif
1380 		if ((*mp)->m_len > size) {
1381 			/*
1382 			 * Move only a partial mbuf at maximum splice length or
1383 			 * if the drain buffer is too small for this large mbuf.
1384 			 */
1385 			if (!maxreached && so->so_snd.sb_datacc > 0) {
1386 				len -= size;
1387 				break;
1388 			}
1389 			*mp = m_copym(so->so_rcv.sb_mb, 0, size, wait);
1390 			if (*mp == NULL) {
1391 				len -= size;
1392 				break;
1393 			}
1394 			so->so_rcv.sb_mb->m_data += size;
1395 			so->so_rcv.sb_mb->m_len -= size;
1396 			so->so_rcv.sb_cc -= size;
1397 			so->so_rcv.sb_datacc -= size;
1398 		} else {
1399 			*mp = so->so_rcv.sb_mb;
1400 			sbfree(&so->so_rcv, *mp);
1401 			so->so_rcv.sb_mb = (*mp)->m_next;
1402 			sbsync(&so->so_rcv, nextrecord);
1403 		}
1404 	}
1405 	*mp = NULL;
1406 
1407 	SBLASTRECORDCHK(&so->so_rcv, "somove 3");
1408 	SBLASTMBUFCHK(&so->so_rcv, "somove 3");
1409 	SBCHECK(&so->so_rcv);
1410 	if (m == NULL)
1411 		goto release;
1412 	m->m_nextpkt = NULL;
1413 	if (m->m_flags & M_PKTHDR) {
1414 		m_resethdr(m);
1415 		m->m_pkthdr.len = len;
1416 	}
1417 
1418 	/* Send window update to source peer as receive buffer has changed. */
1419 	if (so->so_proto->pr_flags & PR_WANTRCVD && so->so_pcb)
1420 		(so->so_proto->pr_usrreq)(so, PRU_RCVD, NULL,
1421 		    NULL, NULL, NULL);
1422 
1423 	/* Receive buffer did shrink by len bytes, adjust oob. */
1424 	state = so->so_state;
1425 	so->so_state &= ~SS_RCVATMARK;
1426 	oobmark = so->so_oobmark;
1427 	so->so_oobmark = oobmark > len ? oobmark - len : 0;
1428 	if (oobmark) {
1429 		if (oobmark == len)
1430 			so->so_state |= SS_RCVATMARK;
1431 		if (oobmark >= len)
1432 			oobmark = 0;
1433 	}
1434 
1435 	/*
1436 	 * Handle oob data.  If any malloc fails, ignore error.
1437 	 * TCP urgent data is not very reliable anyway.
1438 	 */
1439 	while (((state & SS_RCVATMARK) || oobmark) &&
1440 	    (so->so_options & SO_OOBINLINE)) {
1441 		struct mbuf *o = NULL;
1442 
1443 		if (state & SS_RCVATMARK) {
1444 			o = m_get(wait, MT_DATA);
1445 			state &= ~SS_RCVATMARK;
1446 		} else if (oobmark) {
1447 			o = m_split(m, oobmark, wait);
1448 			if (o) {
1449 				error = (*sosp->so_proto->pr_usrreq)(sosp,
1450 				    PRU_SEND, m, NULL, NULL, NULL);
1451 				if (error) {
1452 					if (sosp->so_state & SS_CANTSENDMORE)
1453 						error = EPIPE;
1454 					m_freem(o);
1455 					goto release;
1456 				}
1457 				len -= oobmark;
1458 				so->so_splicelen += oobmark;
1459 				m = o;
1460 				o = m_get(wait, MT_DATA);
1461 			}
1462 			oobmark = 0;
1463 		}
1464 		if (o) {
1465 			o->m_len = 1;
1466 			*mtod(o, caddr_t) = *mtod(m, caddr_t);
1467 			error = (*sosp->so_proto->pr_usrreq)(sosp, PRU_SENDOOB,
1468 			    o, NULL, NULL, NULL);
1469 			if (error) {
1470 				if (sosp->so_state & SS_CANTSENDMORE)
1471 					error = EPIPE;
1472 				m_freem(m);
1473 				goto release;
1474 			}
1475 			len -= 1;
1476 			so->so_splicelen += 1;
1477 			if (oobmark) {
1478 				oobmark -= 1;
1479 				if (oobmark == 0)
1480 					state |= SS_RCVATMARK;
1481 			}
1482 			m_adj(m, 1);
1483 		}
1484 	}
1485 
1486 	/* Append all remaining data to drain socket. */
1487 	if (so->so_rcv.sb_cc == 0 || maxreached)
1488 		sosp->so_state &= ~SS_ISSENDING;
1489 	error = (*sosp->so_proto->pr_usrreq)(sosp, PRU_SEND, m, NULL, NULL,
1490 	    NULL);
1491 	if (error) {
1492 		if (sosp->so_state & SS_CANTSENDMORE)
1493 			error = EPIPE;
1494 		goto release;
1495 	}
1496 	so->so_splicelen += len;
1497 
1498 	/* Move several packets if possible. */
1499 	if (!maxreached && nextrecord)
1500 		goto nextpkt;
1501 
1502  release:
1503 	sosp->so_state &= ~SS_ISSENDING;
1504 	if (!error && maxreached && so->so_splicemax == so->so_splicelen)
1505 		error = EFBIG;
1506 	if (error)
1507 		so->so_error = error;
1508 	if (((so->so_state & SS_CANTRCVMORE) && so->so_rcv.sb_cc == 0) ||
1509 	    (sosp->so_state & SS_CANTSENDMORE) || maxreached || error) {
1510 		sounsplice(so, sosp, 1);
1511 		return (0);
1512 	}
1513 	if (timerisset(&so->so_idletv))
1514 		timeout_add_tv(&so->so_idleto, &so->so_idletv);
1515 	return (1);
1516 }
1517 
1518 #endif /* SOCKET_SPLICE */
1519 
1520 void
1521 sorwakeup(struct socket *so)
1522 {
1523 	soassertlocked(so);
1524 
1525 #ifdef SOCKET_SPLICE
1526 	if (so->so_rcv.sb_flagsintr & SB_SPLICE) {
1527 		/*
1528 		 * TCP has a sendbuffer that can handle multiple packets
1529 		 * at once.  So queue the stream a bit to accumulate data.
1530 		 * The sosplice thread will call somove() later and send
1531 		 * the packets calling tcp_output() only once.
1532 		 * In the UDP case, send out the packets immediately.
1533 		 * Using a thread would make things slower.
1534 		 */
1535 		if (so->so_proto->pr_flags & PR_WANTRCVD)
1536 			task_add(sosplice_taskq, &so->so_splicetask);
1537 		else
1538 			somove(so, M_DONTWAIT);
1539 	}
1540 	if (isspliced(so))
1541 		return;
1542 #endif
1543 	sowakeup(so, &so->so_rcv);
1544 	if (so->so_upcall)
1545 		(*(so->so_upcall))(so, so->so_upcallarg, M_DONTWAIT);
1546 }
1547 
1548 void
1549 sowwakeup(struct socket *so)
1550 {
1551 	soassertlocked(so);
1552 
1553 #ifdef SOCKET_SPLICE
1554 	if (so->so_snd.sb_flagsintr & SB_SPLICE)
1555 		task_add(sosplice_taskq, &so->so_sp->ssp_soback->so_splicetask);
1556 #endif
1557 	sowakeup(so, &so->so_snd);
1558 }
1559 
1560 int
1561 sosetopt(struct socket *so, int level, int optname, struct mbuf *m0)
1562 {
1563 	int error = 0;
1564 	struct mbuf *m = m0;
1565 
1566 	soassertlocked(so);
1567 
1568 	if (level != SOL_SOCKET) {
1569 		if (so->so_proto->pr_ctloutput) {
1570 			error = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so,
1571 			    level, optname, m0);
1572 			return (error);
1573 		}
1574 		error = ENOPROTOOPT;
1575 	} else {
1576 		switch (optname) {
1577 		case SO_BINDANY:
1578 			if ((error = suser(curproc, 0)) != 0)	/* XXX */
1579 				goto bad;
1580 			break;
1581 		}
1582 
1583 		switch (optname) {
1584 
1585 		case SO_LINGER:
1586 			if (m == NULL || m->m_len != sizeof (struct linger) ||
1587 			    mtod(m, struct linger *)->l_linger < 0 ||
1588 			    mtod(m, struct linger *)->l_linger > SHRT_MAX) {
1589 				error = EINVAL;
1590 				goto bad;
1591 			}
1592 			so->so_linger = mtod(m, struct linger *)->l_linger;
1593 			/* FALLTHROUGH */
1594 
1595 		case SO_BINDANY:
1596 		case SO_DEBUG:
1597 		case SO_KEEPALIVE:
1598 		case SO_USELOOPBACK:
1599 		case SO_BROADCAST:
1600 		case SO_REUSEADDR:
1601 		case SO_REUSEPORT:
1602 		case SO_OOBINLINE:
1603 		case SO_TIMESTAMP:
1604 		case SO_ZEROIZE:
1605 			if (m == NULL || m->m_len < sizeof (int)) {
1606 				error = EINVAL;
1607 				goto bad;
1608 			}
1609 			if (*mtod(m, int *))
1610 				so->so_options |= optname;
1611 			else
1612 				so->so_options &= ~optname;
1613 			break;
1614 
1615 		case SO_DONTROUTE:
1616 			if (m == NULL || m->m_len < sizeof (int)) {
1617 				error = EINVAL;
1618 				goto bad;
1619 			}
1620 			if (*mtod(m, int *))
1621 				error = EOPNOTSUPP;
1622 			break;
1623 
1624 		case SO_SNDBUF:
1625 		case SO_RCVBUF:
1626 		case SO_SNDLOWAT:
1627 		case SO_RCVLOWAT:
1628 		    {
1629 			u_long cnt;
1630 
1631 			if (m == NULL || m->m_len < sizeof (int)) {
1632 				error = EINVAL;
1633 				goto bad;
1634 			}
1635 			cnt = *mtod(m, int *);
1636 			if ((long)cnt <= 0)
1637 				cnt = 1;
1638 			switch (optname) {
1639 
1640 			case SO_SNDBUF:
1641 				if (so->so_state & SS_CANTSENDMORE) {
1642 					error = EINVAL;
1643 					goto bad;
1644 				}
1645 				if (sbcheckreserve(cnt, so->so_snd.sb_wat) ||
1646 				    sbreserve(so, &so->so_snd, cnt)) {
1647 					error = ENOBUFS;
1648 					goto bad;
1649 				}
1650 				so->so_snd.sb_wat = cnt;
1651 				break;
1652 
1653 			case SO_RCVBUF:
1654 				if (so->so_state & SS_CANTRCVMORE) {
1655 					error = EINVAL;
1656 					goto bad;
1657 				}
1658 				if (sbcheckreserve(cnt, so->so_rcv.sb_wat) ||
1659 				    sbreserve(so, &so->so_rcv, cnt)) {
1660 					error = ENOBUFS;
1661 					goto bad;
1662 				}
1663 				so->so_rcv.sb_wat = cnt;
1664 				break;
1665 
1666 			case SO_SNDLOWAT:
1667 				so->so_snd.sb_lowat =
1668 				    (cnt > so->so_snd.sb_hiwat) ?
1669 				    so->so_snd.sb_hiwat : cnt;
1670 				break;
1671 			case SO_RCVLOWAT:
1672 				so->so_rcv.sb_lowat =
1673 				    (cnt > so->so_rcv.sb_hiwat) ?
1674 				    so->so_rcv.sb_hiwat : cnt;
1675 				break;
1676 			}
1677 			break;
1678 		    }
1679 
1680 		case SO_SNDTIMEO:
1681 		case SO_RCVTIMEO:
1682 		    {
1683 			struct timeval tv;
1684 			int val;
1685 
1686 			if (m == NULL || m->m_len < sizeof (tv)) {
1687 				error = EINVAL;
1688 				goto bad;
1689 			}
1690 			memcpy(&tv, mtod(m, struct timeval *), sizeof tv);
1691 			val = tvtohz(&tv);
1692 			if (val > USHRT_MAX) {
1693 				error = EDOM;
1694 				goto bad;
1695 			}
1696 
1697 			switch (optname) {
1698 
1699 			case SO_SNDTIMEO:
1700 				so->so_snd.sb_timeo = val;
1701 				break;
1702 			case SO_RCVTIMEO:
1703 				so->so_rcv.sb_timeo = val;
1704 				break;
1705 			}
1706 			break;
1707 		    }
1708 
1709 		case SO_RTABLE:
1710 			if (so->so_proto->pr_domain &&
1711 			    so->so_proto->pr_domain->dom_protosw &&
1712 			    so->so_proto->pr_ctloutput) {
1713 				struct domain *dom = so->so_proto->pr_domain;
1714 
1715 				level = dom->dom_protosw->pr_protocol;
1716 				error = (*so->so_proto->pr_ctloutput)
1717 				    (PRCO_SETOPT, so, level, optname, m0);
1718 				return (error);
1719 			}
1720 			error = ENOPROTOOPT;
1721 			break;
1722 
1723 #ifdef SOCKET_SPLICE
1724 		case SO_SPLICE:
1725 			if (m == NULL) {
1726 				error = sosplice(so, -1, 0, NULL);
1727 			} else if (m->m_len < sizeof(int)) {
1728 				error = EINVAL;
1729 				goto bad;
1730 			} else if (m->m_len < sizeof(struct splice)) {
1731 				error = sosplice(so, *mtod(m, int *), 0, NULL);
1732 			} else {
1733 				error = sosplice(so,
1734 				    mtod(m, struct splice *)->sp_fd,
1735 				    mtod(m, struct splice *)->sp_max,
1736 				   &mtod(m, struct splice *)->sp_idle);
1737 			}
1738 			break;
1739 #endif /* SOCKET_SPLICE */
1740 
1741 		default:
1742 			error = ENOPROTOOPT;
1743 			break;
1744 		}
1745 		if (error == 0 && so->so_proto->pr_ctloutput) {
1746 			(*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so,
1747 			    level, optname, m0);
1748 			m = NULL;	/* freed by protocol */
1749 		}
1750 	}
1751 bad:
1752 	if (m)
1753 		(void) m_free(m);
1754 	return (error);
1755 }
1756 
1757 int
1758 sogetopt(struct socket *so, int level, int optname, struct mbuf *m)
1759 {
1760 	int error = 0;
1761 
1762 	soassertlocked(so);
1763 
1764 	if (level != SOL_SOCKET) {
1765 		if (so->so_proto->pr_ctloutput) {
1766 			m->m_len = 0;
1767 
1768 			error = (*so->so_proto->pr_ctloutput)(PRCO_GETOPT, so,
1769 			    level, optname, m);
1770 			if (error)
1771 				return (error);
1772 			return (0);
1773 		} else
1774 			return (ENOPROTOOPT);
1775 	} else {
1776 		m->m_len = sizeof (int);
1777 
1778 		switch (optname) {
1779 
1780 		case SO_LINGER:
1781 			m->m_len = sizeof (struct linger);
1782 			mtod(m, struct linger *)->l_onoff =
1783 				so->so_options & SO_LINGER;
1784 			mtod(m, struct linger *)->l_linger = so->so_linger;
1785 			break;
1786 
1787 		case SO_BINDANY:
1788 		case SO_USELOOPBACK:
1789 		case SO_DEBUG:
1790 		case SO_KEEPALIVE:
1791 		case SO_REUSEADDR:
1792 		case SO_REUSEPORT:
1793 		case SO_BROADCAST:
1794 		case SO_OOBINLINE:
1795 		case SO_TIMESTAMP:
1796 		case SO_ZEROIZE:
1797 			*mtod(m, int *) = so->so_options & optname;
1798 			break;
1799 
1800 		case SO_DONTROUTE:
1801 			*mtod(m, int *) = 0;
1802 			break;
1803 
1804 		case SO_TYPE:
1805 			*mtod(m, int *) = so->so_type;
1806 			break;
1807 
1808 		case SO_ERROR:
1809 			*mtod(m, int *) = so->so_error;
1810 			so->so_error = 0;
1811 			break;
1812 
1813 		case SO_SNDBUF:
1814 			*mtod(m, int *) = so->so_snd.sb_hiwat;
1815 			break;
1816 
1817 		case SO_RCVBUF:
1818 			*mtod(m, int *) = so->so_rcv.sb_hiwat;
1819 			break;
1820 
1821 		case SO_SNDLOWAT:
1822 			*mtod(m, int *) = so->so_snd.sb_lowat;
1823 			break;
1824 
1825 		case SO_RCVLOWAT:
1826 			*mtod(m, int *) = so->so_rcv.sb_lowat;
1827 			break;
1828 
1829 		case SO_SNDTIMEO:
1830 		case SO_RCVTIMEO:
1831 		    {
1832 			struct timeval tv;
1833 			int val = (optname == SO_SNDTIMEO ?
1834 			    so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1835 
1836 			m->m_len = sizeof(struct timeval);
1837 			memset(&tv, 0, sizeof(tv));
1838 			tv.tv_sec = val / hz;
1839 			tv.tv_usec = (val % hz) * tick;
1840 			memcpy(mtod(m, struct timeval *), &tv, sizeof tv);
1841 			break;
1842 		    }
1843 
1844 		case SO_RTABLE:
1845 			if (so->so_proto->pr_domain &&
1846 			    so->so_proto->pr_domain->dom_protosw &&
1847 			    so->so_proto->pr_ctloutput) {
1848 				struct domain *dom = so->so_proto->pr_domain;
1849 
1850 				level = dom->dom_protosw->pr_protocol;
1851 				error = (*so->so_proto->pr_ctloutput)
1852 				    (PRCO_GETOPT, so, level, optname, m);
1853 				if (error)
1854 					return (error);
1855 				break;
1856 			}
1857 			return (ENOPROTOOPT);
1858 
1859 #ifdef SOCKET_SPLICE
1860 		case SO_SPLICE:
1861 		    {
1862 			off_t len;
1863 
1864 			m->m_len = sizeof(off_t);
1865 			len = so->so_sp ? so->so_sp->ssp_len : 0;
1866 			memcpy(mtod(m, off_t *), &len, sizeof(off_t));
1867 			break;
1868 		    }
1869 #endif /* SOCKET_SPLICE */
1870 
1871 		case SO_PEERCRED:
1872 			if (so->so_proto->pr_protocol == AF_UNIX) {
1873 				struct unpcb *unp = sotounpcb(so);
1874 
1875 				if (unp->unp_flags & UNP_FEIDS) {
1876 					m->m_len = sizeof(unp->unp_connid);
1877 					memcpy(mtod(m, caddr_t),
1878 					    &(unp->unp_connid), m->m_len);
1879 					break;
1880 				}
1881 				return (ENOTCONN);
1882 			}
1883 			return (EOPNOTSUPP);
1884 
1885 		default:
1886 			return (ENOPROTOOPT);
1887 		}
1888 		return (0);
1889 	}
1890 }
1891 
1892 void
1893 sohasoutofband(struct socket *so)
1894 {
1895 	KERNEL_ASSERT_LOCKED();
1896 	csignal(so->so_pgid, SIGURG, so->so_siguid, so->so_sigeuid);
1897 	selwakeup(&so->so_rcv.sb_sel);
1898 }
1899 
1900 int
1901 soo_kqfilter(struct file *fp, struct knote *kn)
1902 {
1903 	struct socket *so = kn->kn_fp->f_data;
1904 	struct sockbuf *sb;
1905 
1906 	KERNEL_ASSERT_LOCKED();
1907 
1908 	switch (kn->kn_filter) {
1909 	case EVFILT_READ:
1910 		if (so->so_options & SO_ACCEPTCONN)
1911 			kn->kn_fop = &solisten_filtops;
1912 		else
1913 			kn->kn_fop = &soread_filtops;
1914 		sb = &so->so_rcv;
1915 		break;
1916 	case EVFILT_WRITE:
1917 		kn->kn_fop = &sowrite_filtops;
1918 		sb = &so->so_snd;
1919 		break;
1920 	default:
1921 		return (EINVAL);
1922 	}
1923 
1924 	SLIST_INSERT_HEAD(&sb->sb_sel.si_note, kn, kn_selnext);
1925 	sb->sb_flags |= SB_KNOTE;
1926 
1927 	return (0);
1928 }
1929 
1930 void
1931 filt_sordetach(struct knote *kn)
1932 {
1933 	struct socket *so = kn->kn_fp->f_data;
1934 
1935 	KERNEL_ASSERT_LOCKED();
1936 
1937 	SLIST_REMOVE(&so->so_rcv.sb_sel.si_note, kn, knote, kn_selnext);
1938 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.si_note))
1939 		so->so_rcv.sb_flags &= ~SB_KNOTE;
1940 }
1941 
1942 int
1943 filt_soread(struct knote *kn, long hint)
1944 {
1945 	struct socket *so = kn->kn_fp->f_data;
1946 	int rv;
1947 
1948 	kn->kn_data = so->so_rcv.sb_cc;
1949 #ifdef SOCKET_SPLICE
1950 	if (isspliced(so)) {
1951 		rv = 0;
1952 	} else
1953 #endif /* SOCKET_SPLICE */
1954 	if (so->so_state & SS_CANTRCVMORE) {
1955 		kn->kn_flags |= EV_EOF;
1956 		kn->kn_fflags = so->so_error;
1957 		rv = 1;
1958 	} else if (so->so_error) {	/* temporary udp error */
1959 		rv = 1;
1960 	} else if (kn->kn_sfflags & NOTE_LOWAT) {
1961 		rv = (kn->kn_data >= kn->kn_sdata);
1962 	} else {
1963 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
1964 	}
1965 
1966 	return rv;
1967 }
1968 
1969 void
1970 filt_sowdetach(struct knote *kn)
1971 {
1972 	struct socket *so = kn->kn_fp->f_data;
1973 
1974 	KERNEL_ASSERT_LOCKED();
1975 
1976 	SLIST_REMOVE(&so->so_snd.sb_sel.si_note, kn, knote, kn_selnext);
1977 	if (SLIST_EMPTY(&so->so_snd.sb_sel.si_note))
1978 		so->so_snd.sb_flags &= ~SB_KNOTE;
1979 }
1980 
1981 int
1982 filt_sowrite(struct knote *kn, long hint)
1983 {
1984 	struct socket *so = kn->kn_fp->f_data;
1985 	int rv;
1986 
1987 	kn->kn_data = sbspace(so, &so->so_snd);
1988 	if (so->so_state & SS_CANTSENDMORE) {
1989 		kn->kn_flags |= EV_EOF;
1990 		kn->kn_fflags = so->so_error;
1991 		rv = 1;
1992 	} else if (so->so_error) {	/* temporary udp error */
1993 		rv = 1;
1994 	} else if (((so->so_state & SS_ISCONNECTED) == 0) &&
1995 	    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1996 		rv = 0;
1997 	} else if (kn->kn_sfflags & NOTE_LOWAT) {
1998 		rv = (kn->kn_data >= kn->kn_sdata);
1999 	} else {
2000 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
2001 	}
2002 
2003 	return (rv);
2004 }
2005 
2006 int
2007 filt_solisten(struct knote *kn, long hint)
2008 {
2009 	struct socket *so = kn->kn_fp->f_data;
2010 
2011 	kn->kn_data = so->so_qlen;
2012 
2013 	return (kn->kn_data != 0);
2014 }
2015 
2016 #ifdef DDB
2017 void
2018 sobuf_print(struct sockbuf *,
2019     int (*)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))));
2020 
2021 void
2022 sobuf_print(struct sockbuf *sb,
2023     int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2024 {
2025 	(*pr)("\tsb_cc: %lu\n", sb->sb_cc);
2026 	(*pr)("\tsb_datacc: %lu\n", sb->sb_datacc);
2027 	(*pr)("\tsb_hiwat: %lu\n", sb->sb_hiwat);
2028 	(*pr)("\tsb_wat: %lu\n", sb->sb_wat);
2029 	(*pr)("\tsb_mbcnt: %lu\n", sb->sb_mbcnt);
2030 	(*pr)("\tsb_mbmax: %lu\n", sb->sb_mbmax);
2031 	(*pr)("\tsb_lowat: %ld\n", sb->sb_lowat);
2032 	(*pr)("\tsb_mb: %p\n", sb->sb_mb);
2033 	(*pr)("\tsb_mbtail: %p\n", sb->sb_mbtail);
2034 	(*pr)("\tsb_lastrecord: %p\n", sb->sb_lastrecord);
2035 	(*pr)("\tsb_sel: ...\n");
2036 	(*pr)("\tsb_flagsintr: %d\n", sb->sb_flagsintr);
2037 	(*pr)("\tsb_flags: %i\n", sb->sb_flags);
2038 	(*pr)("\tsb_timeo: %i\n", sb->sb_timeo);
2039 }
2040 
2041 void
2042 so_print(void *v,
2043     int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2044 {
2045 	struct socket *so = v;
2046 
2047 	(*pr)("socket %p\n", so);
2048 	(*pr)("so_type: %i\n", so->so_type);
2049 	(*pr)("so_options: 0x%04x\n", so->so_options); /* %b */
2050 	(*pr)("so_linger: %i\n", so->so_linger);
2051 	(*pr)("so_state: 0x%04x\n", so->so_state);
2052 	(*pr)("so_pcb: %p\n", so->so_pcb);
2053 	(*pr)("so_proto: %p\n", so->so_proto);
2054 
2055 	(*pr)("so_head: %p\n", so->so_head);
2056 	(*pr)("so_onq: %p\n", so->so_onq);
2057 	(*pr)("so_q0: @%p first: %p\n", &so->so_q0, TAILQ_FIRST(&so->so_q0));
2058 	(*pr)("so_q: @%p first: %p\n", &so->so_q, TAILQ_FIRST(&so->so_q));
2059 	(*pr)("so_eq: next: %p\n", TAILQ_NEXT(so, so_qe));
2060 	(*pr)("so_q0len: %i\n", so->so_q0len);
2061 	(*pr)("so_qlen: %i\n", so->so_qlen);
2062 	(*pr)("so_qlimit: %i\n", so->so_qlimit);
2063 	(*pr)("so_timeo: %i\n", so->so_timeo);
2064 	(*pr)("so_pgid: %i\n", so->so_pgid);
2065 	(*pr)("so_siguid: %i\n", so->so_siguid);
2066 	(*pr)("so_sigeuid: %i\n", so->so_sigeuid);
2067 	(*pr)("so_obmark: %lu\n", so->so_oobmark);
2068 
2069 	(*pr)("so_sp: %p\n", so->so_sp);
2070 	if (so->so_sp != NULL) {
2071 		(*pr)("\tssp_socket: %p\n", so->so_sp->ssp_socket);
2072 		(*pr)("\tssp_soback: %p\n", so->so_sp->ssp_soback);
2073 		(*pr)("\tssp_len: %lld\n",
2074 		    (unsigned long long)so->so_sp->ssp_len);
2075 		(*pr)("\tssp_max: %lld\n",
2076 		    (unsigned long long)so->so_sp->ssp_max);
2077 		(*pr)("\tssp_idletv: %lld %ld\n", so->so_sp->ssp_idletv.tv_sec,
2078 		    so->so_sp->ssp_idletv.tv_usec);
2079 		(*pr)("\tssp_idleto: %spending (@%i)\n",
2080 		    timeout_pending(&so->so_sp->ssp_idleto) ? "" : "not ",
2081 		    so->so_sp->ssp_idleto.to_time);
2082 	}
2083 
2084 	(*pr)("so_rcv:\n");
2085 	sobuf_print(&so->so_rcv, pr);
2086 	(*pr)("so_snd:\n");
2087 	sobuf_print(&so->so_snd, pr);
2088 
2089 	(*pr)("so_upcall: %p so_upcallarg: %p\n",
2090 	    so->so_upcall, so->so_upcallarg);
2091 
2092 	(*pr)("so_euid: %d so_ruid: %d\n", so->so_euid, so->so_ruid);
2093 	(*pr)("so_egid: %d so_rgid: %d\n", so->so_egid, so->so_rgid);
2094 	(*pr)("so_cpid: %d\n", so->so_cpid);
2095 }
2096 #endif
2097 
2098