xref: /openbsd-src/sys/kern/uipc_socket.c (revision f2da64fbbbf1b03f09f390ab01267c93dfd77c4c)
1 /*	$OpenBSD: uipc_socket.c,v 1.161 2016/09/20 14:27:43 bluhm 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_usrreq == 0)
123 		return (EPROTONOSUPPORT);
124 	if (prp->pr_type != type)
125 		return (EPROTOTYPE);
126 	s = splsoftnet();
127 	so = pool_get(&socket_pool, PR_WAITOK | PR_ZERO);
128 	TAILQ_INIT(&so->so_q0);
129 	TAILQ_INIT(&so->so_q);
130 	so->so_type = type;
131 	if (suser(p, 0) == 0)
132 		so->so_state = SS_PRIV;
133 	so->so_ruid = p->p_ucred->cr_ruid;
134 	so->so_euid = p->p_ucred->cr_uid;
135 	so->so_rgid = p->p_ucred->cr_rgid;
136 	so->so_egid = p->p_ucred->cr_gid;
137 	so->so_cpid = p->p_p->ps_pid;
138 	so->so_proto = prp;
139 	error = (*prp->pr_usrreq)(so, PRU_ATTACH, NULL,
140 	    (struct mbuf *)(long)proto, NULL, p);
141 	if (error) {
142 		so->so_state |= SS_NOFDREF;
143 		sofree(so);
144 		splx(s);
145 		return (error);
146 	}
147 	splx(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 s = splsoftnet();
156 	int error;
157 
158 	error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, NULL, nam, NULL, p);
159 	splx(s);
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 = splsoftnet();
175 	error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, NULL, NULL, NULL,
176 	    curproc);
177 	if (error) {
178 		splx(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 	splx(s);
189 	return (0);
190 }
191 
192 /*
193  *  Must be called at splsoftnet()
194  */
195 
196 void
197 sofree(struct socket *so)
198 {
199 	splsoftassert(IPL_SOFTNET);
200 
201 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0)
202 		return;
203 	if (so->so_head) {
204 		/*
205 		 * We must not decommission a socket that's on the accept(2)
206 		 * queue.  If we do, then accept(2) may hang after select(2)
207 		 * indicated that the listening socket was ready.
208 		 */
209 		if (!soqremque(so, 0))
210 			return;
211 	}
212 #ifdef SOCKET_SPLICE
213 	if (so->so_sp) {
214 		if (issplicedback(so))
215 			sounsplice(so->so_sp->ssp_soback, so,
216 			    so->so_sp->ssp_soback != so);
217 		if (isspliced(so))
218 			sounsplice(so, so->so_sp->ssp_socket, 0);
219 		pool_put(&sosplice_pool, so->so_sp);
220 		so->so_sp = NULL;
221 	}
222 #endif /* SOCKET_SPLICE */
223 	sbrelease(&so->so_snd);
224 	sorflush(so);
225 	pool_put(&socket_pool, so);
226 }
227 
228 /*
229  * Close a socket on last file table reference removal.
230  * Initiate disconnect if connected.
231  * Free socket when disconnect complete.
232  */
233 int
234 soclose(struct socket *so)
235 {
236 	struct socket *so2;
237 	int s = splsoftnet();		/* conservative */
238 	int error = 0;
239 
240 	if (so->so_options & SO_ACCEPTCONN) {
241 		while ((so2 = TAILQ_FIRST(&so->so_q0)) != NULL) {
242 			(void) soqremque(so2, 0);
243 			(void) soabort(so2);
244 		}
245 		while ((so2 = TAILQ_FIRST(&so->so_q)) != NULL) {
246 			(void) soqremque(so2, 1);
247 			(void) soabort(so2);
248 		}
249 	}
250 	if (so->so_pcb == 0)
251 		goto discard;
252 	if (so->so_state & SS_ISCONNECTED) {
253 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
254 			error = sodisconnect(so);
255 			if (error)
256 				goto drop;
257 		}
258 		if (so->so_options & SO_LINGER) {
259 			if ((so->so_state & SS_ISDISCONNECTING) &&
260 			    (so->so_state & SS_NBIO))
261 				goto drop;
262 			while (so->so_state & SS_ISCONNECTED) {
263 				error = tsleep(&so->so_timeo,
264 				    PSOCK | PCATCH, "netcls",
265 				    so->so_linger * hz);
266 				if (error)
267 					break;
268 			}
269 		}
270 	}
271 drop:
272 	if (so->so_pcb) {
273 		int error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH, NULL,
274 		    NULL, NULL, curproc);
275 		if (error == 0)
276 			error = error2;
277 	}
278 discard:
279 	if (so->so_state & SS_NOFDREF)
280 		panic("soclose NOFDREF: so %p, so_type %d", so, so->so_type);
281 	so->so_state |= SS_NOFDREF;
282 	sofree(so);
283 	splx(s);
284 	return (error);
285 }
286 
287 /*
288  * Must be called at splsoftnet.
289  */
290 int
291 soabort(struct socket *so)
292 {
293 	splsoftassert(IPL_SOFTNET);
294 
295 	return (*so->so_proto->pr_usrreq)(so, PRU_ABORT, NULL, NULL, NULL,
296 	   curproc);
297 }
298 
299 int
300 soaccept(struct socket *so, struct mbuf *nam)
301 {
302 	int error = 0;
303 
304 	splsoftassert(IPL_SOFTNET);
305 
306 	if ((so->so_state & SS_NOFDREF) == 0)
307 		panic("soaccept !NOFDREF: so %p, so_type %d", so, so->so_type);
308 	so->so_state &= ~SS_NOFDREF;
309 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
310 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
311 		error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT, NULL,
312 		    nam, NULL, curproc);
313 	else
314 		error = ECONNABORTED;
315 	return (error);
316 }
317 
318 int
319 soconnect(struct socket *so, struct mbuf *nam)
320 {
321 	int s;
322 	int error;
323 
324 	if (so->so_options & SO_ACCEPTCONN)
325 		return (EOPNOTSUPP);
326 	s = splsoftnet();
327 	/*
328 	 * If protocol is connection-based, can only connect once.
329 	 * Otherwise, if connected, try to disconnect first.
330 	 * This allows user to disconnect by connecting to, e.g.,
331 	 * a null address.
332 	 */
333 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
334 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
335 	    (error = sodisconnect(so))))
336 		error = EISCONN;
337 	else
338 		error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT,
339 		    NULL, nam, NULL, curproc);
340 	splx(s);
341 	return (error);
342 }
343 
344 int
345 soconnect2(struct socket *so1, struct socket *so2)
346 {
347 	int s = splsoftnet();
348 	int error;
349 
350 	error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2, NULL,
351 	    (struct mbuf *)so2, NULL, curproc);
352 	splx(s);
353 	return (error);
354 }
355 
356 int
357 sodisconnect(struct socket *so)
358 {
359 	int error;
360 
361 	splsoftassert(IPL_SOFTNET);
362 
363 	if ((so->so_state & SS_ISCONNECTED) == 0)
364 		return (ENOTCONN);
365 	if (so->so_state & SS_ISDISCONNECTING)
366 		return (EALREADY);
367 	error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT, NULL, NULL,
368 	    NULL, curproc);
369 	return (error);
370 }
371 
372 int m_getuio(struct mbuf **, int, long, struct uio *);
373 
374 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
375 /*
376  * Send on a socket.
377  * If send must go all at once and message is larger than
378  * send buffering, then hard error.
379  * Lock against other senders.
380  * If must go all at once and not enough room now, then
381  * inform user that this would block and do nothing.
382  * Otherwise, if nonblocking, send as much as possible.
383  * The data to be sent is described by "uio" if nonzero,
384  * otherwise by the mbuf chain "top" (which must be null
385  * if uio is not).  Data provided in mbuf chain must be small
386  * enough to send all at once.
387  *
388  * Returns nonzero on error, timeout or signal; callers
389  * must check for short counts if EINTR/ERESTART are returned.
390  * Data and control buffers are freed on return.
391  */
392 int
393 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
394     struct mbuf *control, int flags)
395 {
396 	long space, clen = 0;
397 	size_t resid;
398 	int error, s;
399 	int atomic = sosendallatonce(so) || top;
400 
401 	if (uio)
402 		resid = uio->uio_resid;
403 	else
404 		resid = top->m_pkthdr.len;
405 	/* MSG_EOR on a SOCK_STREAM socket is invalid. */
406 	if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
407 		error = EINVAL;
408 		goto out;
409 	}
410 	if (uio && uio->uio_procp)
411 		uio->uio_procp->p_ru.ru_msgsnd++;
412 	if (control) {
413 		/*
414 		 * In theory clen should be unsigned (since control->m_len is).
415 		 * However, space must be signed, as it might be less than 0
416 		 * if we over-committed, and we must use a signed comparison
417 		 * of space and clen.
418 		 */
419 		clen = control->m_len;
420 		/* reserve extra space for AF_LOCAL's internalize */
421 		if (so->so_proto->pr_domain->dom_family == AF_LOCAL &&
422 		    clen >= CMSG_ALIGN(sizeof(struct cmsghdr)) &&
423 		    mtod(control, struct cmsghdr *)->cmsg_type == SCM_RIGHTS)
424 			clen = CMSG_SPACE(
425 			    (clen - CMSG_ALIGN(sizeof(struct cmsghdr))) *
426 			    (sizeof(struct file *) / sizeof(int)));
427 	}
428 
429 #define	snderr(errno)	{ error = errno; splx(s); goto release; }
430 
431 restart:
432 	if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
433 		goto out;
434 	so->so_state |= SS_ISSENDING;
435 	do {
436 		s = splsoftnet();
437 		if (so->so_state & SS_CANTSENDMORE)
438 			snderr(EPIPE);
439 		if (so->so_error) {
440 			error = so->so_error;
441 			so->so_error = 0;
442 			splx(s);
443 			goto release;
444 		}
445 		if ((so->so_state & SS_ISCONNECTED) == 0) {
446 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
447 				if (!(resid == 0 && clen != 0))
448 					snderr(ENOTCONN);
449 			} else if (addr == 0)
450 				snderr(EDESTADDRREQ);
451 		}
452 		space = sbspace(&so->so_snd);
453 		if (flags & MSG_OOB)
454 			space += 1024;
455 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
456 		    (so->so_proto->pr_domain->dom_family != AF_LOCAL &&
457 		    clen > so->so_snd.sb_hiwat))
458 			snderr(EMSGSIZE);
459 		if (space < clen ||
460 		    (space - clen < resid &&
461 		    (atomic || space < so->so_snd.sb_lowat))) {
462 			if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT))
463 				snderr(EWOULDBLOCK);
464 			sbunlock(&so->so_snd);
465 			error = sbwait(&so->so_snd);
466 			so->so_state &= ~SS_ISSENDING;
467 			splx(s);
468 			if (error)
469 				goto out;
470 			goto restart;
471 		}
472 		splx(s);
473 		space -= clen;
474 		do {
475 			if (uio == NULL) {
476 				/*
477 				 * Data is prepackaged in "top".
478 				 */
479 				resid = 0;
480 				if (flags & MSG_EOR)
481 					top->m_flags |= M_EOR;
482 			} else {
483 				error = m_getuio(&top, atomic,
484 				    space, uio);
485 				if (error)
486 					goto release;
487 				space -= top->m_pkthdr.len;
488 				resid = uio->uio_resid;
489 				if (flags & MSG_EOR)
490 					top->m_flags |= M_EOR;
491 			}
492 			s = splsoftnet();		/* XXX */
493 			if (resid == 0)
494 				so->so_state &= ~SS_ISSENDING;
495 			error = (*so->so_proto->pr_usrreq)(so,
496 			    (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND,
497 			    top, addr, control, curproc);
498 			splx(s);
499 			clen = 0;
500 			control = NULL;
501 			top = NULL;
502 			if (error)
503 				goto release;
504 		} while (resid && space > 0);
505 	} while (resid);
506 
507 release:
508 	so->so_state &= ~SS_ISSENDING;
509 	sbunlock(&so->so_snd);
510 out:
511 	if (top)
512 		m_freem(top);
513 	if (control)
514 		m_freem(control);
515 	return (error);
516 }
517 
518 int
519 m_getuio(struct mbuf **mp, int atomic, long space, struct uio *uio)
520 {
521 	struct mbuf *m, *top = NULL;
522 	struct mbuf **nextp = &top;
523 	u_long len, mlen;
524 	size_t resid = uio->uio_resid;
525 	int error;
526 
527 	do {
528 		if (top == NULL) {
529 			MGETHDR(m, M_WAIT, MT_DATA);
530 			mlen = MHLEN;
531 			m->m_pkthdr.len = 0;
532 			m->m_pkthdr.ph_ifidx = 0;
533 		} else {
534 			MGET(m, M_WAIT, MT_DATA);
535 			mlen = MLEN;
536 		}
537 		/* chain mbuf together */
538 		*nextp = m;
539 		nextp = &m->m_next;
540 
541 		resid = ulmin(resid, space);
542 		if (resid >= MINCLSIZE) {
543 			MCLGETI(m, M_NOWAIT, NULL, ulmin(resid, MAXMCLBYTES));
544 			if ((m->m_flags & M_EXT) == 0)
545 				MCLGETI(m, M_NOWAIT, NULL, MCLBYTES);
546 			if ((m->m_flags & M_EXT) == 0)
547 				goto nopages;
548 			mlen = m->m_ext.ext_size;
549 			len = ulmin(mlen, resid);
550 			/*
551 			 * For datagram protocols, leave room
552 			 * for protocol headers in first mbuf.
553 			 */
554 			if (atomic && top == NULL && len < mlen - max_hdr)
555 				m->m_data += max_hdr;
556 		} else {
557 nopages:
558 			len = ulmin(mlen, resid);
559 			/*
560 			 * For datagram protocols, leave room
561 			 * for protocol headers in first mbuf.
562 			 */
563 			if (atomic && top == NULL && len < mlen - max_hdr)
564 				MH_ALIGN(m, len);
565 		}
566 
567 		error = uiomove(mtod(m, caddr_t), len, uio);
568 		if (error) {
569 			m_freem(top);
570 			return (error);
571 		}
572 
573 		/* adjust counters */
574 		resid = uio->uio_resid;
575 		space -= len;
576 		m->m_len = len;
577 		top->m_pkthdr.len += len;
578 
579 		/* Is there more space and more data? */
580 	} while (space > 0 && resid > 0);
581 
582 	*mp = top;
583 	return 0;
584 }
585 
586 /*
587  * Following replacement or removal of the first mbuf on the first
588  * mbuf chain of a socket buffer, push necessary state changes back
589  * into the socket buffer so that other consumers see the values
590  * consistently.  'nextrecord' is the callers locally stored value of
591  * the original value of sb->sb_mb->m_nextpkt which must be restored
592  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
593  */
594 void
595 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
596 {
597 
598 	/*
599 	 * First, update for the new value of nextrecord.  If necessary,
600 	 * make it the first record.
601 	 */
602 	if (sb->sb_mb != NULL)
603 		sb->sb_mb->m_nextpkt = nextrecord;
604 	else
605 		sb->sb_mb = nextrecord;
606 
607 	/*
608 	 * Now update any dependent socket buffer fields to reflect
609 	 * the new state.  This is an inline of SB_EMPTY_FIXUP, with
610 	 * the addition of a second clause that takes care of the
611 	 * case where sb_mb has been updated, but remains the last
612 	 * record.
613 	 */
614 	if (sb->sb_mb == NULL) {
615 		sb->sb_mbtail = NULL;
616 		sb->sb_lastrecord = NULL;
617 	} else if (sb->sb_mb->m_nextpkt == NULL)
618 		sb->sb_lastrecord = sb->sb_mb;
619 }
620 
621 /*
622  * Implement receive operations on a socket.
623  * We depend on the way that records are added to the sockbuf
624  * by sbappend*.  In particular, each record (mbufs linked through m_next)
625  * must begin with an address if the protocol so specifies,
626  * followed by an optional mbuf or mbufs containing ancillary data,
627  * and then zero or more mbufs of data.
628  * In order to avoid blocking network interrupts for the entire time here,
629  * we splx() while doing the actual copy to user space.
630  * Although the sockbuf is locked, new data may still be appended,
631  * and thus we must maintain consistency of the sockbuf during that time.
632  *
633  * The caller may receive the data as a single mbuf chain by supplying
634  * an mbuf **mp0 for use in returning the chain.  The uio is then used
635  * only for the count in uio_resid.
636  */
637 int
638 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
639     struct mbuf **mp0, struct mbuf **controlp, int *flagsp,
640     socklen_t controllen)
641 {
642 	struct mbuf *m, **mp;
643 	struct mbuf *cm;
644 	u_long len, offset, moff;
645 	int flags, error, s, type, uio_error = 0;
646 	struct protosw *pr = so->so_proto;
647 	struct mbuf *nextrecord;
648 	size_t resid, orig_resid = uio->uio_resid;
649 
650 	mp = mp0;
651 	if (paddr)
652 		*paddr = 0;
653 	if (controlp)
654 		*controlp = 0;
655 	if (flagsp)
656 		flags = *flagsp &~ MSG_EOR;
657 	else
658 		flags = 0;
659 	if (so->so_state & SS_NBIO)
660 		flags |= MSG_DONTWAIT;
661 	if (flags & MSG_OOB) {
662 		m = m_get(M_WAIT, MT_DATA);
663 		error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m,
664 		    (struct mbuf *)(long)(flags & MSG_PEEK), NULL, curproc);
665 		if (error)
666 			goto bad;
667 		do {
668 			error = uiomove(mtod(m, caddr_t),
669 			    ulmin(uio->uio_resid, m->m_len), uio);
670 			m = m_free(m);
671 		} while (uio->uio_resid && error == 0 && m);
672 bad:
673 		if (m)
674 			m_freem(m);
675 		return (error);
676 	}
677 	if (mp)
678 		*mp = NULL;
679 
680 restart:
681 	if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0)
682 		return (error);
683 	s = splsoftnet();
684 
685 	m = so->so_rcv.sb_mb;
686 #ifdef SOCKET_SPLICE
687 	if (isspliced(so))
688 		m = NULL;
689 #endif /* SOCKET_SPLICE */
690 	/*
691 	 * If we have less data than requested, block awaiting more
692 	 * (subject to any timeout) if:
693 	 *   1. the current count is less than the low water mark,
694 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
695 	 *	receive operation at once if we block (resid <= hiwat), or
696 	 *   3. MSG_DONTWAIT is not set.
697 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
698 	 * we have to do the receive in sections, and thus risk returning
699 	 * a short count if a timeout or signal occurs after we start.
700 	 */
701 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
702 	    so->so_rcv.sb_cc < uio->uio_resid) &&
703 	    (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
704 	    ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
705 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
706 #ifdef DIAGNOSTIC
707 		if (m == NULL && so->so_rcv.sb_cc)
708 #ifdef SOCKET_SPLICE
709 		    if (!isspliced(so))
710 #endif /* SOCKET_SPLICE */
711 			panic("receive 1: so %p, so_type %d, sb_cc %lu",
712 			    so, so->so_type, so->so_rcv.sb_cc);
713 #endif
714 		if (so->so_error) {
715 			if (m)
716 				goto dontblock;
717 			error = so->so_error;
718 			if ((flags & MSG_PEEK) == 0)
719 				so->so_error = 0;
720 			goto release;
721 		}
722 		if (so->so_state & SS_CANTRCVMORE) {
723 			if (m)
724 				goto dontblock;
725 			else if (so->so_rcv.sb_cc == 0)
726 				goto release;
727 		}
728 		for (; m; m = m->m_next)
729 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
730 				m = so->so_rcv.sb_mb;
731 				goto dontblock;
732 			}
733 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
734 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
735 			error = ENOTCONN;
736 			goto release;
737 		}
738 		if (uio->uio_resid == 0 && controlp == NULL)
739 			goto release;
740 		if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT)) {
741 			error = EWOULDBLOCK;
742 			goto release;
743 		}
744 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
745 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
746 		sbunlock(&so->so_rcv);
747 		error = sbwait(&so->so_rcv);
748 		splx(s);
749 		if (error)
750 			return (error);
751 		goto restart;
752 	}
753 dontblock:
754 	/*
755 	 * On entry here, m points to the first record of the socket buffer.
756 	 * From this point onward, we maintain 'nextrecord' as a cache of the
757 	 * pointer to the next record in the socket buffer.  We must keep the
758 	 * various socket buffer pointers and local stack versions of the
759 	 * pointers in sync, pushing out modifications before operations that
760 	 * may sleep, and re-reading them afterwards.
761 	 *
762 	 * Otherwise, we will race with the network stack appending new data
763 	 * or records onto the socket buffer by using inconsistent/stale
764 	 * versions of the field, possibly resulting in socket buffer
765 	 * corruption.
766 	 */
767 	if (uio->uio_procp)
768 		uio->uio_procp->p_ru.ru_msgrcv++;
769 	KASSERT(m == so->so_rcv.sb_mb);
770 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
771 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
772 	nextrecord = m->m_nextpkt;
773 	if (pr->pr_flags & PR_ADDR) {
774 #ifdef DIAGNOSTIC
775 		if (m->m_type != MT_SONAME)
776 			panic("receive 1a: so %p, so_type %d, m %p, m_type %d",
777 			    so, so->so_type, m, m->m_type);
778 #endif
779 		orig_resid = 0;
780 		if (flags & MSG_PEEK) {
781 			if (paddr)
782 				*paddr = m_copym(m, 0, m->m_len, M_NOWAIT);
783 			m = m->m_next;
784 		} else {
785 			sbfree(&so->so_rcv, m);
786 			if (paddr) {
787 				*paddr = m;
788 				so->so_rcv.sb_mb = m->m_next;
789 				m->m_next = 0;
790 				m = so->so_rcv.sb_mb;
791 			} else {
792 				so->so_rcv.sb_mb = m_free(m);
793 				m = so->so_rcv.sb_mb;
794 			}
795 			sbsync(&so->so_rcv, nextrecord);
796 		}
797 	}
798 	while (m && m->m_type == MT_CONTROL && error == 0) {
799 		if (flags & MSG_PEEK) {
800 			if (controlp)
801 				*controlp = m_copym(m, 0, m->m_len, M_NOWAIT);
802 			m = m->m_next;
803 		} else {
804 			sbfree(&so->so_rcv, m);
805 			so->so_rcv.sb_mb = m->m_next;
806 			m->m_nextpkt = m->m_next = NULL;
807 			cm = m;
808 			m = so->so_rcv.sb_mb;
809 			sbsync(&so->so_rcv, nextrecord);
810 			if (controlp) {
811 				if (pr->pr_domain->dom_externalize &&
812 				    mtod(cm, struct cmsghdr *)->cmsg_type ==
813 				    SCM_RIGHTS)
814 				   error = (*pr->pr_domain->dom_externalize)(cm,
815 				       controllen, flags);
816 				*controlp = cm;
817 			} else {
818 				/*
819 				 * Dispose of any SCM_RIGHTS message that went
820 				 * through the read path rather than recv.
821 				 */
822 				if (pr->pr_domain->dom_dispose &&
823 				    mtod(cm, struct cmsghdr *)->cmsg_type == SCM_RIGHTS)
824 					pr->pr_domain->dom_dispose(cm);
825 				m_free(cm);
826 			}
827 		}
828 		if (m != NULL)
829 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
830 		else
831 			nextrecord = so->so_rcv.sb_mb;
832 		if (controlp) {
833 			orig_resid = 0;
834 			controlp = &(*controlp)->m_next;
835 		}
836 	}
837 
838 	/* If m is non-NULL, we have some data to read. */
839 	if (m) {
840 		type = m->m_type;
841 		if (type == MT_OOBDATA)
842 			flags |= MSG_OOB;
843 		if (m->m_flags & M_BCAST)
844 			flags |= MSG_BCAST;
845 		if (m->m_flags & M_MCAST)
846 			flags |= MSG_MCAST;
847 	}
848 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
849 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
850 
851 	moff = 0;
852 	offset = 0;
853 	while (m && uio->uio_resid > 0 && error == 0) {
854 		if (m->m_type == MT_OOBDATA) {
855 			if (type != MT_OOBDATA)
856 				break;
857 		} else if (type == MT_OOBDATA)
858 			break;
859 #ifdef DIAGNOSTIC
860 		else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
861 			panic("receive 3: so %p, so_type %d, m %p, m_type %d",
862 			    so, so->so_type, m, m->m_type);
863 #endif
864 		so->so_state &= ~SS_RCVATMARK;
865 		len = uio->uio_resid;
866 		if (so->so_oobmark && len > so->so_oobmark - offset)
867 			len = so->so_oobmark - offset;
868 		if (len > m->m_len - moff)
869 			len = m->m_len - moff;
870 		/*
871 		 * If mp is set, just pass back the mbufs.
872 		 * Otherwise copy them out via the uio, then free.
873 		 * Sockbuf must be consistent here (points to current mbuf,
874 		 * it points to next record) when we drop priority;
875 		 * we must note any additions to the sockbuf when we
876 		 * block interrupts again.
877 		 */
878 		if (mp == NULL && uio_error == 0) {
879 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
880 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
881 			resid = uio->uio_resid;
882 			splx(s);
883 			uio_error = uiomove(mtod(m, caddr_t) + moff, len, uio);
884 			s = splsoftnet();
885 			if (uio_error)
886 				uio->uio_resid = resid - len;
887 		} else
888 			uio->uio_resid -= len;
889 		if (len == m->m_len - moff) {
890 			if (m->m_flags & M_EOR)
891 				flags |= MSG_EOR;
892 			if (flags & MSG_PEEK) {
893 				m = m->m_next;
894 				moff = 0;
895 			} else {
896 				nextrecord = m->m_nextpkt;
897 				sbfree(&so->so_rcv, m);
898 				if (mp) {
899 					*mp = m;
900 					mp = &m->m_next;
901 					so->so_rcv.sb_mb = m = m->m_next;
902 					*mp = NULL;
903 				} else {
904 					so->so_rcv.sb_mb = m_free(m);
905 					m = so->so_rcv.sb_mb;
906 				}
907 				/*
908 				 * If m != NULL, we also know that
909 				 * so->so_rcv.sb_mb != NULL.
910 				 */
911 				KASSERT(so->so_rcv.sb_mb == m);
912 				if (m) {
913 					m->m_nextpkt = nextrecord;
914 					if (nextrecord == NULL)
915 						so->so_rcv.sb_lastrecord = m;
916 				} else {
917 					so->so_rcv.sb_mb = nextrecord;
918 					SB_EMPTY_FIXUP(&so->so_rcv);
919 				}
920 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
921 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
922 			}
923 		} else {
924 			if (flags & MSG_PEEK)
925 				moff += len;
926 			else {
927 				if (mp)
928 					*mp = m_copym(m, 0, len, M_WAIT);
929 				m->m_data += len;
930 				m->m_len -= len;
931 				so->so_rcv.sb_cc -= len;
932 				so->so_rcv.sb_datacc -= len;
933 			}
934 		}
935 		if (so->so_oobmark) {
936 			if ((flags & MSG_PEEK) == 0) {
937 				so->so_oobmark -= len;
938 				if (so->so_oobmark == 0) {
939 					so->so_state |= SS_RCVATMARK;
940 					break;
941 				}
942 			} else {
943 				offset += len;
944 				if (offset == so->so_oobmark)
945 					break;
946 			}
947 		}
948 		if (flags & MSG_EOR)
949 			break;
950 		/*
951 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
952 		 * we must not quit until "uio->uio_resid == 0" or an error
953 		 * termination.  If a signal/timeout occurs, return
954 		 * with a short count but without error.
955 		 * Keep sockbuf locked against other readers.
956 		 */
957 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
958 		    !sosendallatonce(so) && !nextrecord) {
959 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
960 				break;
961 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
962 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
963 			error = sbwait(&so->so_rcv);
964 			if (error) {
965 				sbunlock(&so->so_rcv);
966 				splx(s);
967 				return (0);
968 			}
969 			if ((m = so->so_rcv.sb_mb) != NULL)
970 				nextrecord = m->m_nextpkt;
971 		}
972 	}
973 
974 	if (m && pr->pr_flags & PR_ATOMIC) {
975 		flags |= MSG_TRUNC;
976 		if ((flags & MSG_PEEK) == 0)
977 			(void) sbdroprecord(&so->so_rcv);
978 	}
979 	if ((flags & MSG_PEEK) == 0) {
980 		if (m == NULL) {
981 			/*
982 			 * First part is an inline SB_EMPTY_FIXUP().  Second
983 			 * part makes sure sb_lastrecord is up-to-date if
984 			 * there is still data in the socket buffer.
985 			 */
986 			so->so_rcv.sb_mb = nextrecord;
987 			if (so->so_rcv.sb_mb == NULL) {
988 				so->so_rcv.sb_mbtail = NULL;
989 				so->so_rcv.sb_lastrecord = NULL;
990 			} else if (nextrecord->m_nextpkt == NULL)
991 				so->so_rcv.sb_lastrecord = nextrecord;
992 		}
993 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
994 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
995 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
996 			(*pr->pr_usrreq)(so, PRU_RCVD, NULL,
997 			    (struct mbuf *)(long)flags, NULL, curproc);
998 	}
999 	if (orig_resid == uio->uio_resid && orig_resid &&
1000 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1001 		sbunlock(&so->so_rcv);
1002 		splx(s);
1003 		goto restart;
1004 	}
1005 
1006 	if (uio_error)
1007 		error = uio_error;
1008 
1009 	if (flagsp)
1010 		*flagsp |= flags;
1011 release:
1012 	sbunlock(&so->so_rcv);
1013 	splx(s);
1014 	return (error);
1015 }
1016 
1017 int
1018 soshutdown(struct socket *so, int how)
1019 {
1020 	struct protosw *pr = so->so_proto;
1021 	int s, error = 0;
1022 
1023 	s = splsoftnet();
1024 	switch (how) {
1025 	case SHUT_RD:
1026 	case SHUT_RDWR:
1027 		sorflush(so);
1028 		if (how == SHUT_RD)
1029 			break;
1030 		/* FALLTHROUGH */
1031 	case SHUT_WR:
1032 		error = (*pr->pr_usrreq)(so, PRU_SHUTDOWN, NULL, NULL, NULL,
1033 		    curproc);
1034 		break;
1035 	default:
1036 		error = EINVAL;
1037 		break;
1038 	}
1039 	splx(s);
1040 	return (error);
1041 }
1042 
1043 void
1044 sorflush(struct socket *so)
1045 {
1046 	struct sockbuf *sb = &so->so_rcv;
1047 	struct protosw *pr = so->so_proto;
1048 	int s;
1049 	struct sockbuf asb;
1050 
1051 	sb->sb_flags |= SB_NOINTR;
1052 	(void) sblock(sb, M_WAITOK);
1053 	s = splnet();
1054 	socantrcvmore(so);
1055 	sbunlock(sb);
1056 	asb = *sb;
1057 	memset(sb, 0, sizeof (*sb));
1058 	/* XXX - the memset stomps all over so_rcv */
1059 	if (asb.sb_flags & SB_KNOTE) {
1060 		sb->sb_sel.si_note = asb.sb_sel.si_note;
1061 		sb->sb_flags = SB_KNOTE;
1062 	}
1063 	splx(s);
1064 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose)
1065 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1066 	sbrelease(&asb);
1067 }
1068 
1069 #ifdef SOCKET_SPLICE
1070 
1071 #define so_splicelen	so_sp->ssp_len
1072 #define so_splicemax	so_sp->ssp_max
1073 #define so_idletv	so_sp->ssp_idletv
1074 #define so_idleto	so_sp->ssp_idleto
1075 #define so_splicetask	so_sp->ssp_task
1076 
1077 int
1078 sosplice(struct socket *so, int fd, off_t max, struct timeval *tv)
1079 {
1080 	struct file	*fp;
1081 	struct socket	*sosp;
1082 	int		 s, error = 0;
1083 
1084 	if (sosplice_taskq == NULL)
1085 		sosplice_taskq = taskq_create("sosplice", 1, IPL_SOFTNET,
1086 		    TASKQ_CANTSLEEP);
1087 	if (sosplice_taskq == NULL)
1088 		return (ENOMEM);
1089 
1090 	if ((so->so_proto->pr_flags & PR_SPLICE) == 0)
1091 		return (EPROTONOSUPPORT);
1092 	if (so->so_options & SO_ACCEPTCONN)
1093 		return (EOPNOTSUPP);
1094 	if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1095 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
1096 		return (ENOTCONN);
1097 	if (so->so_sp == NULL)
1098 		so->so_sp = pool_get(&sosplice_pool, PR_WAITOK | PR_ZERO);
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_rcv,
1104 		    (so->so_state & SS_NBIO) ? M_NOWAIT : M_WAITOK)) != 0)
1105 			return (error);
1106 		s = splsoftnet();
1107 		if (so->so_sp->ssp_socket)
1108 			sounsplice(so, so->so_sp->ssp_socket, 1);
1109 		splx(s);
1110 		sbunlock(&so->so_rcv);
1111 		return (0);
1112 	}
1113 
1114 	if (max && max < 0)
1115 		return (EINVAL);
1116 
1117 	if (tv && (tv->tv_sec < 0 || tv->tv_usec < 0))
1118 		return (EINVAL);
1119 
1120 	/* Find sosp, the drain socket where data will be spliced into. */
1121 	if ((error = getsock(curproc, fd, &fp)) != 0)
1122 		return (error);
1123 	sosp = fp->f_data;
1124 	if (sosp->so_sp == NULL)
1125 		sosp->so_sp = pool_get(&sosplice_pool, PR_WAITOK | PR_ZERO);
1126 
1127 	/* Lock both receive and send buffer. */
1128 	if ((error = sblock(&so->so_rcv,
1129 	    (so->so_state & SS_NBIO) ? M_NOWAIT : M_WAITOK)) != 0) {
1130 		FRELE(fp, curproc);
1131 		return (error);
1132 	}
1133 	if ((error = sblock(&sosp->so_snd, M_WAITOK)) != 0) {
1134 		sbunlock(&so->so_rcv);
1135 		FRELE(fp, curproc);
1136 		return (error);
1137 	}
1138 	s = splsoftnet();
1139 
1140 	if (so->so_sp->ssp_socket || sosp->so_sp->ssp_soback) {
1141 		error = EBUSY;
1142 		goto release;
1143 	}
1144 	if (sosp->so_proto->pr_usrreq != so->so_proto->pr_usrreq) {
1145 		error = EPROTONOSUPPORT;
1146 		goto release;
1147 	}
1148 	if (sosp->so_options & SO_ACCEPTCONN) {
1149 		error = EOPNOTSUPP;
1150 		goto release;
1151 	}
1152 	if ((sosp->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0) {
1153 		error = ENOTCONN;
1154 		goto release;
1155 	}
1156 
1157 	/* Splice so and sosp together. */
1158 	so->so_sp->ssp_socket = sosp;
1159 	sosp->so_sp->ssp_soback = so;
1160 	so->so_splicelen = 0;
1161 	so->so_splicemax = max;
1162 	if (tv)
1163 		so->so_idletv = *tv;
1164 	else
1165 		timerclear(&so->so_idletv);
1166 	timeout_set(&so->so_idleto, soidle, so);
1167 	task_set(&so->so_splicetask, sotask, so);
1168 
1169 	/*
1170 	 * To prevent softnet interrupt from calling somove() while
1171 	 * we sleep, the socket buffers are not marked as spliced yet.
1172 	 */
1173 	if (somove(so, M_WAIT)) {
1174 		so->so_rcv.sb_flagsintr |= SB_SPLICE;
1175 		sosp->so_snd.sb_flagsintr |= SB_SPLICE;
1176 	}
1177 
1178  release:
1179 	splx(s);
1180 	sbunlock(&sosp->so_snd);
1181 	sbunlock(&so->so_rcv);
1182 	FRELE(fp, curproc);
1183 	return (error);
1184 }
1185 
1186 void
1187 sounsplice(struct socket *so, struct socket *sosp, int wakeup)
1188 {
1189 	splsoftassert(IPL_SOFTNET);
1190 
1191 	task_del(sosplice_taskq, &so->so_splicetask);
1192 	timeout_del(&so->so_idleto);
1193 	sosp->so_snd.sb_flagsintr &= ~SB_SPLICE;
1194 	so->so_rcv.sb_flagsintr &= ~SB_SPLICE;
1195 	so->so_sp->ssp_socket = sosp->so_sp->ssp_soback = NULL;
1196 	if (wakeup && soreadable(so))
1197 		sorwakeup(so);
1198 }
1199 
1200 void
1201 soidle(void *arg)
1202 {
1203 	struct socket *so = arg;
1204 	int s;
1205 
1206 	s = splsoftnet();
1207 	if (so->so_rcv.sb_flagsintr & SB_SPLICE) {
1208 		so->so_error = ETIMEDOUT;
1209 		sounsplice(so, so->so_sp->ssp_socket, 1);
1210 	}
1211 	splx(s);
1212 }
1213 
1214 void
1215 sotask(void *arg)
1216 {
1217 	struct socket *so = arg;
1218 	int s;
1219 
1220 	s = splsoftnet();
1221 	if (so->so_rcv.sb_flagsintr & SB_SPLICE) {
1222 		/*
1223 		 * We may not sleep here as sofree() and unsplice() may be
1224 		 * called from softnet interrupt context.  This would remove
1225 		 * the socket during somove().
1226 		 */
1227 		somove(so, M_DONTWAIT);
1228 	}
1229 	splx(s);
1230 
1231 	/* Avoid user land starvation. */
1232 	yield();
1233 }
1234 
1235 /*
1236  * Move data from receive buffer of spliced source socket to send
1237  * buffer of drain socket.  Try to move as much as possible in one
1238  * big chunk.  It is a TCP only implementation.
1239  * Return value 0 means splicing has been finished, 1 continue.
1240  */
1241 int
1242 somove(struct socket *so, int wait)
1243 {
1244 	struct socket	*sosp = so->so_sp->ssp_socket;
1245 	struct mbuf	*m, **mp, *nextrecord;
1246 	u_long		 len, off, oobmark;
1247 	long		 space;
1248 	int		 error = 0, maxreached = 0;
1249 	short		 state;
1250 
1251 	splsoftassert(IPL_SOFTNET);
1252 
1253  nextpkt:
1254 	if (so->so_error) {
1255 		error = so->so_error;
1256 		goto release;
1257 	}
1258 	if (sosp->so_state & SS_CANTSENDMORE) {
1259 		error = EPIPE;
1260 		goto release;
1261 	}
1262 	if (sosp->so_error && sosp->so_error != ETIMEDOUT &&
1263 	    sosp->so_error != EFBIG && sosp->so_error != ELOOP) {
1264 		error = sosp->so_error;
1265 		goto release;
1266 	}
1267 	if ((sosp->so_state & SS_ISCONNECTED) == 0)
1268 		goto release;
1269 
1270 	/* Calculate how many bytes can be copied now. */
1271 	len = so->so_rcv.sb_datacc;
1272 	if (so->so_splicemax) {
1273 		KASSERT(so->so_splicelen < so->so_splicemax);
1274 		if (so->so_splicemax <= so->so_splicelen + len) {
1275 			len = so->so_splicemax - so->so_splicelen;
1276 			maxreached = 1;
1277 		}
1278 	}
1279 	space = sbspace(&sosp->so_snd);
1280 	if (so->so_oobmark && so->so_oobmark < len &&
1281 	    so->so_oobmark < space + 1024)
1282 		space += 1024;
1283 	if (space <= 0) {
1284 		maxreached = 0;
1285 		goto release;
1286 	}
1287 	if (space < len) {
1288 		maxreached = 0;
1289 		if (space < sosp->so_snd.sb_lowat)
1290 			goto release;
1291 		len = space;
1292 	}
1293 	sosp->so_state |= SS_ISSENDING;
1294 
1295 	SBLASTRECORDCHK(&so->so_rcv, "somove 1");
1296 	SBLASTMBUFCHK(&so->so_rcv, "somove 1");
1297 	m = so->so_rcv.sb_mb;
1298 	if (m == NULL)
1299 		goto release;
1300 	nextrecord = m->m_nextpkt;
1301 
1302 	/* Drop address and control information not used with splicing. */
1303 	if (so->so_proto->pr_flags & PR_ADDR) {
1304 #ifdef DIAGNOSTIC
1305 		if (m->m_type != MT_SONAME)
1306 			panic("somove soname: so %p, so_type %d, m %p, "
1307 			    "m_type %d", so, so->so_type, m, m->m_type);
1308 #endif
1309 		m = m->m_next;
1310 	}
1311 	while (m && m->m_type == MT_CONTROL)
1312 		m = m->m_next;
1313 	if (m == NULL) {
1314 		sbdroprecord(&so->so_rcv);
1315 		if (so->so_proto->pr_flags & PR_WANTRCVD && so->so_pcb)
1316 			(so->so_proto->pr_usrreq)(so, PRU_RCVD, NULL,
1317 			    NULL, NULL, NULL);
1318 		goto nextpkt;
1319 	}
1320 
1321 	/*
1322 	 * By splicing sockets connected to localhost, userland might create a
1323 	 * loop.  Dissolve splicing with error if loop is detected by counter.
1324 	 */
1325 	if ((m->m_flags & M_PKTHDR) && m->m_pkthdr.ph_loopcnt++ >= M_MAXLOOP) {
1326 		error = ELOOP;
1327 		goto release;
1328 	}
1329 
1330 	if (so->so_proto->pr_flags & PR_ATOMIC) {
1331 		if ((m->m_flags & M_PKTHDR) == 0)
1332 			panic("somove !PKTHDR: so %p, so_type %d, m %p, "
1333 			    "m_type %d", so, so->so_type, m, m->m_type);
1334 		if (sosp->so_snd.sb_hiwat < m->m_pkthdr.len) {
1335 			error = EMSGSIZE;
1336 			goto release;
1337 		}
1338 		if (len < m->m_pkthdr.len)
1339 			goto release;
1340 		if (m->m_pkthdr.len < len) {
1341 			maxreached = 0;
1342 			len = m->m_pkthdr.len;
1343 		}
1344 		/*
1345 		 * Throw away the name mbuf after it has been assured
1346 		 * that the whole first record can be processed.
1347 		 */
1348 		m = so->so_rcv.sb_mb;
1349 		sbfree(&so->so_rcv, m);
1350 		so->so_rcv.sb_mb = m_free(m);
1351 		sbsync(&so->so_rcv, nextrecord);
1352 	}
1353 	/*
1354 	 * Throw away the control mbufs after it has been assured
1355 	 * that the whole first record can be processed.
1356 	 */
1357 	m = so->so_rcv.sb_mb;
1358 	while (m && m->m_type == MT_CONTROL) {
1359 		sbfree(&so->so_rcv, m);
1360 		so->so_rcv.sb_mb = m_free(m);
1361 		m = so->so_rcv.sb_mb;
1362 		sbsync(&so->so_rcv, nextrecord);
1363 	}
1364 
1365 	SBLASTRECORDCHK(&so->so_rcv, "somove 2");
1366 	SBLASTMBUFCHK(&so->so_rcv, "somove 2");
1367 
1368 	/* Take at most len mbufs out of receive buffer. */
1369 	for (off = 0, mp = &m; off <= len && *mp;
1370 	    off += (*mp)->m_len, mp = &(*mp)->m_next) {
1371 		u_long size = len - off;
1372 
1373 #ifdef DIAGNOSTIC
1374 		if ((*mp)->m_type != MT_DATA && (*mp)->m_type != MT_HEADER)
1375 			panic("somove type: so %p, so_type %d, m %p, "
1376 			    "m_type %d", so, so->so_type, *mp, (*mp)->m_type);
1377 #endif
1378 		if ((*mp)->m_len > size) {
1379 			if (!maxreached || (*mp = m_copym(
1380 			    so->so_rcv.sb_mb, 0, size, wait)) == NULL) {
1381 				len -= size;
1382 				break;
1383 			}
1384 			so->so_rcv.sb_mb->m_data += size;
1385 			so->so_rcv.sb_mb->m_len -= size;
1386 			so->so_rcv.sb_cc -= size;
1387 			so->so_rcv.sb_datacc -= size;
1388 		} else {
1389 			*mp = so->so_rcv.sb_mb;
1390 			sbfree(&so->so_rcv, *mp);
1391 			so->so_rcv.sb_mb = (*mp)->m_next;
1392 			sbsync(&so->so_rcv, nextrecord);
1393 		}
1394 	}
1395 	*mp = NULL;
1396 
1397 	SBLASTRECORDCHK(&so->so_rcv, "somove 3");
1398 	SBLASTMBUFCHK(&so->so_rcv, "somove 3");
1399 	SBCHECK(&so->so_rcv);
1400 	if (m == NULL)
1401 		goto release;
1402 	m->m_nextpkt = NULL;
1403 	if (m->m_flags & M_PKTHDR) {
1404 		m_resethdr(m);
1405 		m->m_pkthdr.len = len;
1406 	}
1407 
1408 	/* Send window update to source peer as receive buffer has changed. */
1409 	if (so->so_proto->pr_flags & PR_WANTRCVD && so->so_pcb)
1410 		(so->so_proto->pr_usrreq)(so, PRU_RCVD, NULL,
1411 		    NULL, NULL, NULL);
1412 
1413 	/* Receive buffer did shrink by len bytes, adjust oob. */
1414 	state = so->so_state;
1415 	so->so_state &= ~SS_RCVATMARK;
1416 	oobmark = so->so_oobmark;
1417 	so->so_oobmark = oobmark > len ? oobmark - len : 0;
1418 	if (oobmark) {
1419 		if (oobmark == len)
1420 			so->so_state |= SS_RCVATMARK;
1421 		if (oobmark >= len)
1422 			oobmark = 0;
1423 	}
1424 
1425 	/*
1426 	 * Handle oob data.  If any malloc fails, ignore error.
1427 	 * TCP urgent data is not very reliable anyway.
1428 	 */
1429 	while (((state & SS_RCVATMARK) || oobmark) &&
1430 	    (so->so_options & SO_OOBINLINE)) {
1431 		struct mbuf *o = NULL;
1432 
1433 		if (state & SS_RCVATMARK) {
1434 			o = m_get(wait, MT_DATA);
1435 			state &= ~SS_RCVATMARK;
1436 		} else if (oobmark) {
1437 			o = m_split(m, oobmark, wait);
1438 			if (o) {
1439 				error = (*sosp->so_proto->pr_usrreq)(sosp,
1440 				    PRU_SEND, m, NULL, NULL, NULL);
1441 				if (error) {
1442 					if (sosp->so_state & SS_CANTSENDMORE)
1443 						error = EPIPE;
1444 					m_freem(o);
1445 					goto release;
1446 				}
1447 				len -= oobmark;
1448 				so->so_splicelen += oobmark;
1449 				m = o;
1450 				o = m_get(wait, MT_DATA);
1451 			}
1452 			oobmark = 0;
1453 		}
1454 		if (o) {
1455 			o->m_len = 1;
1456 			*mtod(o, caddr_t) = *mtod(m, caddr_t);
1457 			error = (*sosp->so_proto->pr_usrreq)(sosp, PRU_SENDOOB,
1458 			    o, NULL, NULL, NULL);
1459 			if (error) {
1460 				if (sosp->so_state & SS_CANTSENDMORE)
1461 					error = EPIPE;
1462 				m_freem(m);
1463 				goto release;
1464 			}
1465 			len -= 1;
1466 			so->so_splicelen += 1;
1467 			if (oobmark) {
1468 				oobmark -= 1;
1469 				if (oobmark == 0)
1470 					state |= SS_RCVATMARK;
1471 			}
1472 			m_adj(m, 1);
1473 		}
1474 	}
1475 
1476 	/* Append all remaining data to drain socket. */
1477 	if (so->so_rcv.sb_cc == 0 || maxreached)
1478 		sosp->so_state &= ~SS_ISSENDING;
1479 	error = (*sosp->so_proto->pr_usrreq)(sosp, PRU_SEND, m, NULL, NULL,
1480 	    NULL);
1481 	if (error) {
1482 		if (sosp->so_state & SS_CANTSENDMORE)
1483 			error = EPIPE;
1484 		goto release;
1485 	}
1486 	so->so_splicelen += len;
1487 
1488 	/* Move several packets if possible. */
1489 	if (!maxreached && nextrecord)
1490 		goto nextpkt;
1491 
1492  release:
1493 	sosp->so_state &= ~SS_ISSENDING;
1494 	if (!error && maxreached && so->so_splicemax == so->so_splicelen)
1495 		error = EFBIG;
1496 	if (error)
1497 		so->so_error = error;
1498 	if (((so->so_state & SS_CANTRCVMORE) && so->so_rcv.sb_cc == 0) ||
1499 	    (sosp->so_state & SS_CANTSENDMORE) || maxreached || error) {
1500 		sounsplice(so, sosp, 1);
1501 		return (0);
1502 	}
1503 	if (timerisset(&so->so_idletv))
1504 		timeout_add_tv(&so->so_idleto, &so->so_idletv);
1505 	return (1);
1506 }
1507 
1508 #endif /* SOCKET_SPLICE */
1509 
1510 void
1511 sorwakeup(struct socket *so)
1512 {
1513 	splsoftassert(IPL_SOFTNET);
1514 
1515 #ifdef SOCKET_SPLICE
1516 	if (so->so_rcv.sb_flagsintr & SB_SPLICE) {
1517 		/*
1518 		 * TCP has a sendbuffer that can handle multiple packets
1519 		 * at once.  So queue the stream a bit to accumulate data.
1520 		 * The sosplice thread will call somove() later and send
1521 		 * the packets calling tcp_output() only once.
1522 		 * In the UDP case, send out the packets immediately.
1523 		 * Using a thread would make things slower.
1524 		 */
1525 		if (so->so_proto->pr_flags & PR_WANTRCVD)
1526 			task_add(sosplice_taskq, &so->so_splicetask);
1527 		else
1528 			somove(so, M_DONTWAIT);
1529 	}
1530 	if (isspliced(so))
1531 		return;
1532 #endif
1533 	sowakeup(so, &so->so_rcv);
1534 	if (so->so_upcall)
1535 		(*(so->so_upcall))(so, so->so_upcallarg, M_DONTWAIT);
1536 }
1537 
1538 void
1539 sowwakeup(struct socket *so)
1540 {
1541 	splsoftassert(IPL_SOFTNET);
1542 
1543 #ifdef SOCKET_SPLICE
1544 	if (so->so_snd.sb_flagsintr & SB_SPLICE)
1545 		task_add(sosplice_taskq, &so->so_sp->ssp_soback->so_splicetask);
1546 #endif
1547 	sowakeup(so, &so->so_snd);
1548 }
1549 
1550 int
1551 sosetopt(struct socket *so, int level, int optname, struct mbuf *m0)
1552 {
1553 	int error = 0;
1554 	struct mbuf *m = m0;
1555 
1556 	if (level != SOL_SOCKET) {
1557 		if (so->so_proto && so->so_proto->pr_ctloutput)
1558 			return ((*so->so_proto->pr_ctloutput)
1559 				  (PRCO_SETOPT, so, level, optname, &m0));
1560 		error = ENOPROTOOPT;
1561 	} else {
1562 		switch (optname) {
1563 		case SO_BINDANY:
1564 			if ((error = suser(curproc, 0)) != 0)	/* XXX */
1565 				goto bad;
1566 			break;
1567 		}
1568 
1569 		switch (optname) {
1570 
1571 		case SO_LINGER:
1572 			if (m == NULL || m->m_len != sizeof (struct linger) ||
1573 			    mtod(m, struct linger *)->l_linger < 0 ||
1574 			    mtod(m, struct linger *)->l_linger > SHRT_MAX) {
1575 				error = EINVAL;
1576 				goto bad;
1577 			}
1578 			so->so_linger = mtod(m, struct linger *)->l_linger;
1579 			/* FALLTHROUGH */
1580 
1581 		case SO_BINDANY:
1582 		case SO_DEBUG:
1583 		case SO_KEEPALIVE:
1584 		case SO_USELOOPBACK:
1585 		case SO_BROADCAST:
1586 		case SO_REUSEADDR:
1587 		case SO_REUSEPORT:
1588 		case SO_OOBINLINE:
1589 		case SO_TIMESTAMP:
1590 			if (m == NULL || m->m_len < sizeof (int)) {
1591 				error = EINVAL;
1592 				goto bad;
1593 			}
1594 			if (*mtod(m, int *))
1595 				so->so_options |= optname;
1596 			else
1597 				so->so_options &= ~optname;
1598 			break;
1599 
1600 		case SO_DONTROUTE:
1601 			if (m == NULL || m->m_len < sizeof (int)) {
1602 				error = EINVAL;
1603 				goto bad;
1604 			}
1605 			if (*mtod(m, int *))
1606 				error = EOPNOTSUPP;
1607 			break;
1608 
1609 		case SO_SNDBUF:
1610 		case SO_RCVBUF:
1611 		case SO_SNDLOWAT:
1612 		case SO_RCVLOWAT:
1613 		    {
1614 			u_long cnt;
1615 
1616 			if (m == NULL || m->m_len < sizeof (int)) {
1617 				error = EINVAL;
1618 				goto bad;
1619 			}
1620 			cnt = *mtod(m, int *);
1621 			if ((long)cnt <= 0)
1622 				cnt = 1;
1623 			switch (optname) {
1624 
1625 			case SO_SNDBUF:
1626 				if (so->so_state & SS_CANTSENDMORE) {
1627 					error = EINVAL;
1628 					goto bad;
1629 				}
1630 				if (sbcheckreserve(cnt, so->so_snd.sb_wat) ||
1631 				    sbreserve(&so->so_snd, cnt)) {
1632 					error = ENOBUFS;
1633 					goto bad;
1634 				}
1635 				so->so_snd.sb_wat = cnt;
1636 				break;
1637 
1638 			case SO_RCVBUF:
1639 				if (so->so_state & SS_CANTRCVMORE) {
1640 					error = EINVAL;
1641 					goto bad;
1642 				}
1643 				if (sbcheckreserve(cnt, so->so_rcv.sb_wat) ||
1644 				    sbreserve(&so->so_rcv, cnt)) {
1645 					error = ENOBUFS;
1646 					goto bad;
1647 				}
1648 				so->so_rcv.sb_wat = cnt;
1649 				break;
1650 
1651 			case SO_SNDLOWAT:
1652 				so->so_snd.sb_lowat =
1653 				    (cnt > so->so_snd.sb_hiwat) ?
1654 				    so->so_snd.sb_hiwat : cnt;
1655 				break;
1656 			case SO_RCVLOWAT:
1657 				so->so_rcv.sb_lowat =
1658 				    (cnt > so->so_rcv.sb_hiwat) ?
1659 				    so->so_rcv.sb_hiwat : cnt;
1660 				break;
1661 			}
1662 			break;
1663 		    }
1664 
1665 		case SO_SNDTIMEO:
1666 		case SO_RCVTIMEO:
1667 		    {
1668 			struct timeval tv;
1669 			int val;
1670 
1671 			if (m == NULL || m->m_len < sizeof (tv)) {
1672 				error = EINVAL;
1673 				goto bad;
1674 			}
1675 			memcpy(&tv, mtod(m, struct timeval *), sizeof tv);
1676 			val = tvtohz(&tv);
1677 			if (val > USHRT_MAX) {
1678 				error = EDOM;
1679 				goto bad;
1680 			}
1681 
1682 			switch (optname) {
1683 
1684 			case SO_SNDTIMEO:
1685 				so->so_snd.sb_timeo = val;
1686 				break;
1687 			case SO_RCVTIMEO:
1688 				so->so_rcv.sb_timeo = val;
1689 				break;
1690 			}
1691 			break;
1692 		    }
1693 
1694 		case SO_RTABLE:
1695 			if (so->so_proto && so->so_proto->pr_domain &&
1696 			    so->so_proto->pr_domain->dom_protosw &&
1697 			    so->so_proto->pr_ctloutput) {
1698 				struct domain *dom = so->so_proto->pr_domain;
1699 
1700 				level = dom->dom_protosw->pr_protocol;
1701 				return ((*so->so_proto->pr_ctloutput)
1702 				    (PRCO_SETOPT, so, level, optname, &m0));
1703 			}
1704 			error = ENOPROTOOPT;
1705 			break;
1706 
1707 #ifdef SOCKET_SPLICE
1708 		case SO_SPLICE:
1709 			if (m == NULL) {
1710 				error = sosplice(so, -1, 0, NULL);
1711 			} else if (m->m_len < sizeof(int)) {
1712 				error = EINVAL;
1713 				goto bad;
1714 			} else if (m->m_len < sizeof(struct splice)) {
1715 				error = sosplice(so, *mtod(m, int *), 0, NULL);
1716 			} else {
1717 				error = sosplice(so,
1718 				    mtod(m, struct splice *)->sp_fd,
1719 				    mtod(m, struct splice *)->sp_max,
1720 				   &mtod(m, struct splice *)->sp_idle);
1721 			}
1722 			break;
1723 #endif /* SOCKET_SPLICE */
1724 
1725 		default:
1726 			error = ENOPROTOOPT;
1727 			break;
1728 		}
1729 		if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
1730 			(void) ((*so->so_proto->pr_ctloutput)
1731 				  (PRCO_SETOPT, so, level, optname, &m0));
1732 			m = NULL;	/* freed by protocol */
1733 		}
1734 	}
1735 bad:
1736 	if (m)
1737 		(void) m_free(m);
1738 	return (error);
1739 }
1740 
1741 int
1742 sogetopt(struct socket *so, int level, int optname, struct mbuf **mp)
1743 {
1744 	struct mbuf *m;
1745 
1746 	if (level != SOL_SOCKET) {
1747 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1748 			return ((*so->so_proto->pr_ctloutput)
1749 				  (PRCO_GETOPT, so, level, optname, mp));
1750 		} else
1751 			return (ENOPROTOOPT);
1752 	} else {
1753 		m = m_get(M_WAIT, MT_SOOPTS);
1754 		m->m_len = sizeof (int);
1755 
1756 		switch (optname) {
1757 
1758 		case SO_LINGER:
1759 			m->m_len = sizeof (struct linger);
1760 			mtod(m, struct linger *)->l_onoff =
1761 				so->so_options & SO_LINGER;
1762 			mtod(m, struct linger *)->l_linger = so->so_linger;
1763 			break;
1764 
1765 		case SO_BINDANY:
1766 		case SO_USELOOPBACK:
1767 		case SO_DEBUG:
1768 		case SO_KEEPALIVE:
1769 		case SO_REUSEADDR:
1770 		case SO_REUSEPORT:
1771 		case SO_BROADCAST:
1772 		case SO_OOBINLINE:
1773 		case SO_TIMESTAMP:
1774 			*mtod(m, int *) = so->so_options & optname;
1775 			break;
1776 
1777 		case SO_DONTROUTE:
1778 			*mtod(m, int *) = 0;
1779 			break;
1780 
1781 		case SO_TYPE:
1782 			*mtod(m, int *) = so->so_type;
1783 			break;
1784 
1785 		case SO_ERROR:
1786 			*mtod(m, int *) = so->so_error;
1787 			so->so_error = 0;
1788 			break;
1789 
1790 		case SO_SNDBUF:
1791 			*mtod(m, int *) = so->so_snd.sb_hiwat;
1792 			break;
1793 
1794 		case SO_RCVBUF:
1795 			*mtod(m, int *) = so->so_rcv.sb_hiwat;
1796 			break;
1797 
1798 		case SO_SNDLOWAT:
1799 			*mtod(m, int *) = so->so_snd.sb_lowat;
1800 			break;
1801 
1802 		case SO_RCVLOWAT:
1803 			*mtod(m, int *) = so->so_rcv.sb_lowat;
1804 			break;
1805 
1806 		case SO_SNDTIMEO:
1807 		case SO_RCVTIMEO:
1808 		    {
1809 			struct timeval tv;
1810 			int val = (optname == SO_SNDTIMEO ?
1811 			    so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1812 
1813 			m->m_len = sizeof(struct timeval);
1814 			memset(&tv, 0, sizeof(tv));
1815 			tv.tv_sec = val / hz;
1816 			tv.tv_usec = (val % hz) * tick;
1817 			memcpy(mtod(m, struct timeval *), &tv, sizeof tv);
1818 			break;
1819 		    }
1820 
1821 		case SO_RTABLE:
1822 			(void)m_free(m);
1823 			if (so->so_proto && so->so_proto->pr_domain &&
1824 			    so->so_proto->pr_domain->dom_protosw &&
1825 			    so->so_proto->pr_ctloutput) {
1826 				struct domain *dom = so->so_proto->pr_domain;
1827 
1828 				level = dom->dom_protosw->pr_protocol;
1829 				return ((*so->so_proto->pr_ctloutput)
1830 				    (PRCO_GETOPT, so, level, optname, mp));
1831 			}
1832 			return (ENOPROTOOPT);
1833 			break;
1834 
1835 #ifdef SOCKET_SPLICE
1836 		case SO_SPLICE:
1837 		    {
1838 			off_t len;
1839 			int s = splsoftnet();
1840 
1841 			m->m_len = sizeof(off_t);
1842 			len = so->so_sp ? so->so_sp->ssp_len : 0;
1843 			memcpy(mtod(m, off_t *), &len, sizeof(off_t));
1844 			splx(s);
1845 			break;
1846 		    }
1847 #endif /* SOCKET_SPLICE */
1848 
1849 		case SO_PEERCRED:
1850 			if (so->so_proto->pr_protocol == AF_UNIX) {
1851 				struct unpcb *unp = sotounpcb(so);
1852 
1853 				if (unp->unp_flags & UNP_FEIDS) {
1854 					m->m_len = sizeof(unp->unp_connid);
1855 					memcpy(mtod(m, caddr_t),
1856 					    &(unp->unp_connid), m->m_len);
1857 					break;
1858 				}
1859 				(void)m_free(m);
1860 				return (ENOTCONN);
1861 			}
1862 			(void)m_free(m);
1863 			return (EOPNOTSUPP);
1864 			break;
1865 
1866 		default:
1867 			(void)m_free(m);
1868 			return (ENOPROTOOPT);
1869 		}
1870 		*mp = m;
1871 		return (0);
1872 	}
1873 }
1874 
1875 void
1876 sohasoutofband(struct socket *so)
1877 {
1878 	csignal(so->so_pgid, SIGURG, so->so_siguid, so->so_sigeuid);
1879 	selwakeup(&so->so_rcv.sb_sel);
1880 }
1881 
1882 int
1883 soo_kqfilter(struct file *fp, struct knote *kn)
1884 {
1885 	struct socket *so = kn->kn_fp->f_data;
1886 	struct sockbuf *sb;
1887 	int s;
1888 
1889 	switch (kn->kn_filter) {
1890 	case EVFILT_READ:
1891 		if (so->so_options & SO_ACCEPTCONN)
1892 			kn->kn_fop = &solisten_filtops;
1893 		else
1894 			kn->kn_fop = &soread_filtops;
1895 		sb = &so->so_rcv;
1896 		break;
1897 	case EVFILT_WRITE:
1898 		kn->kn_fop = &sowrite_filtops;
1899 		sb = &so->so_snd;
1900 		break;
1901 	default:
1902 		return (EINVAL);
1903 	}
1904 
1905 	s = splnet();
1906 	SLIST_INSERT_HEAD(&sb->sb_sel.si_note, kn, kn_selnext);
1907 	sb->sb_flags |= SB_KNOTE;
1908 	splx(s);
1909 	return (0);
1910 }
1911 
1912 void
1913 filt_sordetach(struct knote *kn)
1914 {
1915 	struct socket *so = kn->kn_fp->f_data;
1916 	int s = splnet();
1917 
1918 	SLIST_REMOVE(&so->so_rcv.sb_sel.si_note, kn, knote, kn_selnext);
1919 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.si_note))
1920 		so->so_rcv.sb_flags &= ~SB_KNOTE;
1921 	splx(s);
1922 }
1923 
1924 int
1925 filt_soread(struct knote *kn, long hint)
1926 {
1927 	struct socket *so = kn->kn_fp->f_data;
1928 
1929 	kn->kn_data = so->so_rcv.sb_cc;
1930 #ifdef SOCKET_SPLICE
1931 	if (isspliced(so))
1932 		return (0);
1933 #endif /* SOCKET_SPLICE */
1934 	if (so->so_state & SS_CANTRCVMORE) {
1935 		kn->kn_flags |= EV_EOF;
1936 		kn->kn_fflags = so->so_error;
1937 		return (1);
1938 	}
1939 	if (so->so_error)	/* temporary udp error */
1940 		return (1);
1941 	if (kn->kn_sfflags & NOTE_LOWAT)
1942 		return (kn->kn_data >= kn->kn_sdata);
1943 	return (kn->kn_data >= so->so_rcv.sb_lowat);
1944 }
1945 
1946 void
1947 filt_sowdetach(struct knote *kn)
1948 {
1949 	struct socket *so = kn->kn_fp->f_data;
1950 	int s = splnet();
1951 
1952 	SLIST_REMOVE(&so->so_snd.sb_sel.si_note, kn, knote, kn_selnext);
1953 	if (SLIST_EMPTY(&so->so_snd.sb_sel.si_note))
1954 		so->so_snd.sb_flags &= ~SB_KNOTE;
1955 	splx(s);
1956 }
1957 
1958 int
1959 filt_sowrite(struct knote *kn, long hint)
1960 {
1961 	struct socket *so = kn->kn_fp->f_data;
1962 
1963 	kn->kn_data = sbspace(&so->so_snd);
1964 	if (so->so_state & SS_CANTSENDMORE) {
1965 		kn->kn_flags |= EV_EOF;
1966 		kn->kn_fflags = so->so_error;
1967 		return (1);
1968 	}
1969 	if (so->so_error)	/* temporary udp error */
1970 		return (1);
1971 	if (((so->so_state & SS_ISCONNECTED) == 0) &&
1972 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
1973 		return (0);
1974 	if (kn->kn_sfflags & NOTE_LOWAT)
1975 		return (kn->kn_data >= kn->kn_sdata);
1976 	return (kn->kn_data >= so->so_snd.sb_lowat);
1977 }
1978 
1979 int
1980 filt_solisten(struct knote *kn, long hint)
1981 {
1982 	struct socket *so = kn->kn_fp->f_data;
1983 
1984 	kn->kn_data = so->so_qlen;
1985 	return (so->so_qlen != 0);
1986 }
1987 
1988 #ifdef DDB
1989 void
1990 sobuf_print(struct sockbuf *,
1991     int (*)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))));
1992 
1993 void
1994 sobuf_print(struct sockbuf *sb,
1995     int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
1996 {
1997 	(*pr)("\tsb_cc: %lu\n", sb->sb_cc);
1998 	(*pr)("\tsb_datacc: %lu\n", sb->sb_datacc);
1999 	(*pr)("\tsb_hiwat: %lu\n", sb->sb_hiwat);
2000 	(*pr)("\tsb_wat: %lu\n", sb->sb_wat);
2001 	(*pr)("\tsb_mbcnt: %lu\n", sb->sb_mbcnt);
2002 	(*pr)("\tsb_mbmax: %lu\n", sb->sb_mbmax);
2003 	(*pr)("\tsb_lowat: %ld\n", sb->sb_lowat);
2004 	(*pr)("\tsb_mb: %p\n", sb->sb_mb);
2005 	(*pr)("\tsb_mbtail: %p\n", sb->sb_mbtail);
2006 	(*pr)("\tsb_lastrecord: %p\n", sb->sb_lastrecord);
2007 	(*pr)("\tsb_sel: ...\n");
2008 	(*pr)("\tsb_flagsintr: %d\n", sb->sb_flagsintr);
2009 	(*pr)("\tsb_flags: %i\n", sb->sb_flags);
2010 	(*pr)("\tsb_timeo: %i\n", sb->sb_timeo);
2011 }
2012 
2013 void
2014 so_print(void *v,
2015     int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2016 {
2017 	struct socket *so = v;
2018 
2019 	(*pr)("socket %p\n", so);
2020 	(*pr)("so_type: %i\n", so->so_type);
2021 	(*pr)("so_options: 0x%04x\n", so->so_options); /* %b */
2022 	(*pr)("so_linger: %i\n", so->so_linger);
2023 	(*pr)("so_state: %i\n", so->so_state);
2024 	(*pr)("so_pcb: %p\n", so->so_pcb);
2025 	(*pr)("so_proto: %p\n", so->so_proto);
2026 
2027 	(*pr)("so_head: %p\n", so->so_head);
2028 	(*pr)("so_onq: %p\n", so->so_onq);
2029 	(*pr)("so_q0: @%p first: %p\n", &so->so_q0, TAILQ_FIRST(&so->so_q0));
2030 	(*pr)("so_q: @%p first: %p\n", &so->so_q, TAILQ_FIRST(&so->so_q));
2031 	(*pr)("so_eq: next: %p\n", TAILQ_NEXT(so, so_qe));
2032 	(*pr)("so_q0len: %i\n", so->so_q0len);
2033 	(*pr)("so_qlen: %i\n", so->so_qlen);
2034 	(*pr)("so_qlimit: %i\n", so->so_qlimit);
2035 	(*pr)("so_timeo: %i\n", so->so_timeo);
2036 	(*pr)("so_pgid: %i\n", so->so_pgid);
2037 	(*pr)("so_siguid: %i\n", so->so_siguid);
2038 	(*pr)("so_sigeuid: %i\n", so->so_sigeuid);
2039 	(*pr)("so_obmark: %lu\n", so->so_oobmark);
2040 
2041 	(*pr)("so_sp: %p\n", so->so_sp);
2042 	if (so->so_sp != NULL) {
2043 		(*pr)("\tssp_socket: %p\n", so->so_sp->ssp_socket);
2044 		(*pr)("\tssp_soback: %p\n", so->so_sp->ssp_soback);
2045 		(*pr)("\tssp_len: %lld\n",
2046 		    (unsigned long long)so->so_sp->ssp_len);
2047 		(*pr)("\tssp_max: %lld\n",
2048 		    (unsigned long long)so->so_sp->ssp_max);
2049 		(*pr)("\tssp_idletv: %lld %ld\n", so->so_sp->ssp_idletv.tv_sec,
2050 		    so->so_sp->ssp_idletv.tv_usec);
2051 		(*pr)("\tssp_idleto: %spending (@%i)\n",
2052 		    timeout_pending(&so->so_sp->ssp_idleto) ? "" : "not ",
2053 		    so->so_sp->ssp_idleto.to_time);
2054 	}
2055 
2056 	(*pr)("so_rcv:\n");
2057 	sobuf_print(&so->so_rcv, pr);
2058 	(*pr)("so_snd:\n");
2059 	sobuf_print(&so->so_snd, pr);
2060 
2061 	(*pr)("so_upcall: %p so_upcallarg: %p\n",
2062 	    so->so_upcall, so->so_upcallarg);
2063 
2064 	(*pr)("so_euid: %d so_ruid: %d\n", so->so_euid, so->so_ruid);
2065 	(*pr)("so_egid: %d so_rgid: %d\n", so->so_egid, so->so_rgid);
2066 	(*pr)("so_cpid: %d\n", so->so_cpid);
2067 }
2068 #endif
2069 
2070