xref: /openbsd-src/sys/kern/uipc_socket.c (revision f763167468dba5339ed4b14b7ecaca2a397ab0f6)
1 /*	$OpenBSD: uipc_socket.c,v 1.205 2017/09/15 19:29:28 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_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 && m == top && 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 && m == top && 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 	int error;
1045 
1046 	sb->sb_flags |= SB_NOINTR;
1047 	error = sblock(so, sb, M_WAITOK);
1048 	/* with SB_NOINTR and M_WAITOK sblock() must not fail */
1049 	KASSERT(error == 0);
1050 	socantrcvmore(so);
1051 	sbunlock(sb);
1052 	aso.so_proto = pr;
1053 	aso.so_rcv = *sb;
1054 	memset(sb, 0, sizeof (*sb));
1055 	/* XXX - the memset stomps all over so_rcv */
1056 	if (aso.so_rcv.sb_flags & SB_KNOTE) {
1057 		sb->sb_sel.si_note = aso.so_rcv.sb_sel.si_note;
1058 		sb->sb_flags = SB_KNOTE;
1059 	}
1060 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose)
1061 		(*pr->pr_domain->dom_dispose)(aso.so_rcv.sb_mb);
1062 	sbrelease(&aso, &aso.so_rcv);
1063 }
1064 
1065 #ifdef SOCKET_SPLICE
1066 
1067 #define so_splicelen	so_sp->ssp_len
1068 #define so_splicemax	so_sp->ssp_max
1069 #define so_idletv	so_sp->ssp_idletv
1070 #define so_idleto	so_sp->ssp_idleto
1071 #define so_splicetask	so_sp->ssp_task
1072 
1073 int
1074 sosplice(struct socket *so, int fd, off_t max, struct timeval *tv)
1075 {
1076 	struct file	*fp;
1077 	struct socket	*sosp;
1078 	struct sosplice	*sp;
1079 	int		 error = 0;
1080 
1081 	soassertlocked(so);
1082 
1083 	if (sosplice_taskq == NULL)
1084 		sosplice_taskq = taskq_create("sosplice", 1, IPL_SOFTNET, 0);
1085 	if (sosplice_taskq == NULL)
1086 		return (ENOMEM);
1087 
1088 	if ((so->so_proto->pr_flags & PR_SPLICE) == 0)
1089 		return (EPROTONOSUPPORT);
1090 	if (so->so_options & SO_ACCEPTCONN)
1091 		return (EOPNOTSUPP);
1092 	if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1093 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
1094 		return (ENOTCONN);
1095 	if (so->so_sp == NULL) {
1096 		sp = pool_get(&sosplice_pool, PR_WAITOK | PR_ZERO);
1097 		if (so->so_sp == NULL)
1098 			so->so_sp = sp;
1099 		else
1100 			pool_put(&sosplice_pool, sp);
1101 	}
1102 
1103 	/* If no fd is given, unsplice by removing existing link. */
1104 	if (fd < 0) {
1105 		/* Lock receive buffer. */
1106 		if ((error = sblock(so, &so->so_rcv,
1107 		    (so->so_state & SS_NBIO) ? M_NOWAIT : M_WAITOK)) != 0) {
1108 			return (error);
1109 		}
1110 		if (so->so_sp->ssp_socket)
1111 			sounsplice(so, so->so_sp->ssp_socket, 1);
1112 		sbunlock(&so->so_rcv);
1113 		return (0);
1114 	}
1115 
1116 	if (max && max < 0)
1117 		return (EINVAL);
1118 
1119 	if (tv && (tv->tv_sec < 0 || tv->tv_usec < 0))
1120 		return (EINVAL);
1121 
1122 	/* Find sosp, the drain socket where data will be spliced into. */
1123 	if ((error = getsock(curproc, fd, &fp)) != 0)
1124 		return (error);
1125 	sosp = fp->f_data;
1126 	if (sosp->so_sp == NULL) {
1127 		sp = pool_get(&sosplice_pool, PR_WAITOK | PR_ZERO);
1128 		if (sosp->so_sp == NULL)
1129 			sosp->so_sp = sp;
1130 		else
1131 			pool_put(&sosplice_pool, sp);
1132 	}
1133 
1134 	/* Lock both receive and send buffer. */
1135 	if ((error = sblock(so, &so->so_rcv,
1136 	    (so->so_state & SS_NBIO) ? M_NOWAIT : M_WAITOK)) != 0) {
1137 		FRELE(fp, curproc);
1138 		return (error);
1139 	}
1140 	if ((error = sblock(so, &sosp->so_snd, M_WAITOK)) != 0) {
1141 		sbunlock(&so->so_rcv);
1142 		FRELE(fp, curproc);
1143 		return (error);
1144 	}
1145 
1146 	if (so->so_sp->ssp_socket || sosp->so_sp->ssp_soback) {
1147 		error = EBUSY;
1148 		goto release;
1149 	}
1150 	if (sosp->so_proto->pr_usrreq != so->so_proto->pr_usrreq) {
1151 		error = EPROTONOSUPPORT;
1152 		goto release;
1153 	}
1154 	if (sosp->so_options & SO_ACCEPTCONN) {
1155 		error = EOPNOTSUPP;
1156 		goto release;
1157 	}
1158 	if ((sosp->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0) {
1159 		error = ENOTCONN;
1160 		goto release;
1161 	}
1162 
1163 	/* Splice so and sosp together. */
1164 	so->so_sp->ssp_socket = sosp;
1165 	sosp->so_sp->ssp_soback = so;
1166 	so->so_splicelen = 0;
1167 	so->so_splicemax = max;
1168 	if (tv)
1169 		so->so_idletv = *tv;
1170 	else
1171 		timerclear(&so->so_idletv);
1172 	timeout_set_proc(&so->so_idleto, soidle, so);
1173 	task_set(&so->so_splicetask, sotask, so);
1174 
1175 	/*
1176 	 * To prevent softnet interrupt from calling somove() while
1177 	 * we sleep, the socket buffers are not marked as spliced yet.
1178 	 */
1179 	if (somove(so, M_WAIT)) {
1180 		so->so_rcv.sb_flagsintr |= SB_SPLICE;
1181 		sosp->so_snd.sb_flagsintr |= SB_SPLICE;
1182 	}
1183 
1184  release:
1185 	sbunlock(&sosp->so_snd);
1186 	sbunlock(&so->so_rcv);
1187 	FRELE(fp, curproc);
1188 	return (error);
1189 }
1190 
1191 void
1192 sounsplice(struct socket *so, struct socket *sosp, int wakeup)
1193 {
1194 	soassertlocked(so);
1195 
1196 	task_del(sosplice_taskq, &so->so_splicetask);
1197 	timeout_del(&so->so_idleto);
1198 	sosp->so_snd.sb_flagsintr &= ~SB_SPLICE;
1199 	so->so_rcv.sb_flagsintr &= ~SB_SPLICE;
1200 	so->so_sp->ssp_socket = sosp->so_sp->ssp_soback = NULL;
1201 	if (wakeup && soreadable(so))
1202 		sorwakeup(so);
1203 }
1204 
1205 void
1206 soidle(void *arg)
1207 {
1208 	struct socket *so = arg;
1209 	int s;
1210 
1211 	s = solock(so);
1212 	if (so->so_rcv.sb_flagsintr & SB_SPLICE) {
1213 		so->so_error = ETIMEDOUT;
1214 		sounsplice(so, so->so_sp->ssp_socket, 1);
1215 	}
1216 	sounlock(s);
1217 }
1218 
1219 void
1220 sotask(void *arg)
1221 {
1222 	struct socket *so = arg;
1223 	int s;
1224 
1225 	s = solock(so);
1226 	if (so->so_rcv.sb_flagsintr & SB_SPLICE) {
1227 		/*
1228 		 * We may not sleep here as sofree() and unsplice() may be
1229 		 * called from softnet interrupt context.  This would remove
1230 		 * the socket during somove().
1231 		 */
1232 		somove(so, M_DONTWAIT);
1233 	}
1234 	sounlock(s);
1235 
1236 	/* Avoid user land starvation. */
1237 	yield();
1238 }
1239 
1240 /*
1241  * Move data from receive buffer of spliced source socket to send
1242  * buffer of drain socket.  Try to move as much as possible in one
1243  * big chunk.  It is a TCP only implementation.
1244  * Return value 0 means splicing has been finished, 1 continue.
1245  */
1246 int
1247 somove(struct socket *so, int wait)
1248 {
1249 	struct socket	*sosp = so->so_sp->ssp_socket;
1250 	struct mbuf	*m, **mp, *nextrecord;
1251 	u_long		 len, off, oobmark;
1252 	long		 space;
1253 	int		 error = 0, maxreached = 0;
1254 	short		 state;
1255 
1256 	soassertlocked(so);
1257 
1258  nextpkt:
1259 	if (so->so_error) {
1260 		error = so->so_error;
1261 		goto release;
1262 	}
1263 	if (sosp->so_state & SS_CANTSENDMORE) {
1264 		error = EPIPE;
1265 		goto release;
1266 	}
1267 	if (sosp->so_error && sosp->so_error != ETIMEDOUT &&
1268 	    sosp->so_error != EFBIG && sosp->so_error != ELOOP) {
1269 		error = sosp->so_error;
1270 		goto release;
1271 	}
1272 	if ((sosp->so_state & SS_ISCONNECTED) == 0)
1273 		goto release;
1274 
1275 	/* Calculate how many bytes can be copied now. */
1276 	len = so->so_rcv.sb_datacc;
1277 	if (so->so_splicemax) {
1278 		KASSERT(so->so_splicelen < so->so_splicemax);
1279 		if (so->so_splicemax <= so->so_splicelen + len) {
1280 			len = so->so_splicemax - so->so_splicelen;
1281 			maxreached = 1;
1282 		}
1283 	}
1284 	space = sbspace(sosp, &sosp->so_snd);
1285 	if (so->so_oobmark && so->so_oobmark < len &&
1286 	    so->so_oobmark < space + 1024)
1287 		space += 1024;
1288 	if (space <= 0) {
1289 		maxreached = 0;
1290 		goto release;
1291 	}
1292 	if (space < len) {
1293 		maxreached = 0;
1294 		if (space < sosp->so_snd.sb_lowat)
1295 			goto release;
1296 		len = space;
1297 	}
1298 	sosp->so_state |= SS_ISSENDING;
1299 
1300 	SBLASTRECORDCHK(&so->so_rcv, "somove 1");
1301 	SBLASTMBUFCHK(&so->so_rcv, "somove 1");
1302 	m = so->so_rcv.sb_mb;
1303 	if (m == NULL)
1304 		goto release;
1305 	nextrecord = m->m_nextpkt;
1306 
1307 	/* Drop address and control information not used with splicing. */
1308 	if (so->so_proto->pr_flags & PR_ADDR) {
1309 #ifdef DIAGNOSTIC
1310 		if (m->m_type != MT_SONAME)
1311 			panic("somove soname: so %p, so_type %d, m %p, "
1312 			    "m_type %d", so, so->so_type, m, m->m_type);
1313 #endif
1314 		m = m->m_next;
1315 	}
1316 	while (m && m->m_type == MT_CONTROL)
1317 		m = m->m_next;
1318 	if (m == NULL) {
1319 		sbdroprecord(&so->so_rcv);
1320 		if (so->so_proto->pr_flags & PR_WANTRCVD && so->so_pcb)
1321 			(so->so_proto->pr_usrreq)(so, PRU_RCVD, NULL,
1322 			    NULL, NULL, NULL);
1323 		goto nextpkt;
1324 	}
1325 
1326 	/*
1327 	 * By splicing sockets connected to localhost, userland might create a
1328 	 * loop.  Dissolve splicing with error if loop is detected by counter.
1329 	 */
1330 	if ((m->m_flags & M_PKTHDR) && m->m_pkthdr.ph_loopcnt++ >= M_MAXLOOP) {
1331 		error = ELOOP;
1332 		goto release;
1333 	}
1334 
1335 	if (so->so_proto->pr_flags & PR_ATOMIC) {
1336 		if ((m->m_flags & M_PKTHDR) == 0)
1337 			panic("somove !PKTHDR: so %p, so_type %d, m %p, "
1338 			    "m_type %d", so, so->so_type, m, m->m_type);
1339 		if (sosp->so_snd.sb_hiwat < m->m_pkthdr.len) {
1340 			error = EMSGSIZE;
1341 			goto release;
1342 		}
1343 		if (len < m->m_pkthdr.len)
1344 			goto release;
1345 		if (m->m_pkthdr.len < len) {
1346 			maxreached = 0;
1347 			len = m->m_pkthdr.len;
1348 		}
1349 		/*
1350 		 * Throw away the name mbuf after it has been assured
1351 		 * that the whole first record can be processed.
1352 		 */
1353 		m = so->so_rcv.sb_mb;
1354 		sbfree(&so->so_rcv, m);
1355 		so->so_rcv.sb_mb = m_free(m);
1356 		sbsync(&so->so_rcv, nextrecord);
1357 	}
1358 	/*
1359 	 * Throw away the control mbufs after it has been assured
1360 	 * that the whole first record can be processed.
1361 	 */
1362 	m = so->so_rcv.sb_mb;
1363 	while (m && m->m_type == MT_CONTROL) {
1364 		sbfree(&so->so_rcv, m);
1365 		so->so_rcv.sb_mb = m_free(m);
1366 		m = so->so_rcv.sb_mb;
1367 		sbsync(&so->so_rcv, nextrecord);
1368 	}
1369 
1370 	SBLASTRECORDCHK(&so->so_rcv, "somove 2");
1371 	SBLASTMBUFCHK(&so->so_rcv, "somove 2");
1372 
1373 	/* Take at most len mbufs out of receive buffer. */
1374 	for (off = 0, mp = &m; off <= len && *mp;
1375 	    off += (*mp)->m_len, mp = &(*mp)->m_next) {
1376 		u_long size = len - off;
1377 
1378 #ifdef DIAGNOSTIC
1379 		if ((*mp)->m_type != MT_DATA && (*mp)->m_type != MT_HEADER)
1380 			panic("somove type: so %p, so_type %d, m %p, "
1381 			    "m_type %d", so, so->so_type, *mp, (*mp)->m_type);
1382 #endif
1383 		if ((*mp)->m_len > size) {
1384 			/*
1385 			 * Move only a partial mbuf at maximum splice length or
1386 			 * if the drain buffer is too small for this large mbuf.
1387 			 */
1388 			if (!maxreached && so->so_snd.sb_datacc > 0) {
1389 				len -= size;
1390 				break;
1391 			}
1392 			*mp = m_copym(so->so_rcv.sb_mb, 0, size, wait);
1393 			if (*mp == NULL) {
1394 				len -= size;
1395 				break;
1396 			}
1397 			so->so_rcv.sb_mb->m_data += size;
1398 			so->so_rcv.sb_mb->m_len -= size;
1399 			so->so_rcv.sb_cc -= size;
1400 			so->so_rcv.sb_datacc -= size;
1401 		} else {
1402 			*mp = so->so_rcv.sb_mb;
1403 			sbfree(&so->so_rcv, *mp);
1404 			so->so_rcv.sb_mb = (*mp)->m_next;
1405 			sbsync(&so->so_rcv, nextrecord);
1406 		}
1407 	}
1408 	*mp = NULL;
1409 
1410 	SBLASTRECORDCHK(&so->so_rcv, "somove 3");
1411 	SBLASTMBUFCHK(&so->so_rcv, "somove 3");
1412 	SBCHECK(&so->so_rcv);
1413 	if (m == NULL)
1414 		goto release;
1415 	m->m_nextpkt = NULL;
1416 	if (m->m_flags & M_PKTHDR) {
1417 		m_resethdr(m);
1418 		m->m_pkthdr.len = len;
1419 	}
1420 
1421 	/* Send window update to source peer as receive buffer has changed. */
1422 	if (so->so_proto->pr_flags & PR_WANTRCVD && so->so_pcb)
1423 		(so->so_proto->pr_usrreq)(so, PRU_RCVD, NULL,
1424 		    NULL, NULL, NULL);
1425 
1426 	/* Receive buffer did shrink by len bytes, adjust oob. */
1427 	state = so->so_state;
1428 	so->so_state &= ~SS_RCVATMARK;
1429 	oobmark = so->so_oobmark;
1430 	so->so_oobmark = oobmark > len ? oobmark - len : 0;
1431 	if (oobmark) {
1432 		if (oobmark == len)
1433 			so->so_state |= SS_RCVATMARK;
1434 		if (oobmark >= len)
1435 			oobmark = 0;
1436 	}
1437 
1438 	/*
1439 	 * Handle oob data.  If any malloc fails, ignore error.
1440 	 * TCP urgent data is not very reliable anyway.
1441 	 */
1442 	while (((state & SS_RCVATMARK) || oobmark) &&
1443 	    (so->so_options & SO_OOBINLINE)) {
1444 		struct mbuf *o = NULL;
1445 
1446 		if (state & SS_RCVATMARK) {
1447 			o = m_get(wait, MT_DATA);
1448 			state &= ~SS_RCVATMARK;
1449 		} else if (oobmark) {
1450 			o = m_split(m, oobmark, wait);
1451 			if (o) {
1452 				error = (*sosp->so_proto->pr_usrreq)(sosp,
1453 				    PRU_SEND, m, NULL, NULL, NULL);
1454 				if (error) {
1455 					if (sosp->so_state & SS_CANTSENDMORE)
1456 						error = EPIPE;
1457 					m_freem(o);
1458 					goto release;
1459 				}
1460 				len -= oobmark;
1461 				so->so_splicelen += oobmark;
1462 				m = o;
1463 				o = m_get(wait, MT_DATA);
1464 			}
1465 			oobmark = 0;
1466 		}
1467 		if (o) {
1468 			o->m_len = 1;
1469 			*mtod(o, caddr_t) = *mtod(m, caddr_t);
1470 			error = (*sosp->so_proto->pr_usrreq)(sosp, PRU_SENDOOB,
1471 			    o, NULL, NULL, NULL);
1472 			if (error) {
1473 				if (sosp->so_state & SS_CANTSENDMORE)
1474 					error = EPIPE;
1475 				m_freem(m);
1476 				goto release;
1477 			}
1478 			len -= 1;
1479 			so->so_splicelen += 1;
1480 			if (oobmark) {
1481 				oobmark -= 1;
1482 				if (oobmark == 0)
1483 					state |= SS_RCVATMARK;
1484 			}
1485 			m_adj(m, 1);
1486 		}
1487 	}
1488 
1489 	/* Append all remaining data to drain socket. */
1490 	if (so->so_rcv.sb_cc == 0 || maxreached)
1491 		sosp->so_state &= ~SS_ISSENDING;
1492 	error = (*sosp->so_proto->pr_usrreq)(sosp, PRU_SEND, m, NULL, NULL,
1493 	    NULL);
1494 	if (error) {
1495 		if (sosp->so_state & SS_CANTSENDMORE)
1496 			error = EPIPE;
1497 		goto release;
1498 	}
1499 	so->so_splicelen += len;
1500 
1501 	/* Move several packets if possible. */
1502 	if (!maxreached && nextrecord)
1503 		goto nextpkt;
1504 
1505  release:
1506 	sosp->so_state &= ~SS_ISSENDING;
1507 	if (!error && maxreached && so->so_splicemax == so->so_splicelen)
1508 		error = EFBIG;
1509 	if (error)
1510 		so->so_error = error;
1511 	if (((so->so_state & SS_CANTRCVMORE) && so->so_rcv.sb_cc == 0) ||
1512 	    (sosp->so_state & SS_CANTSENDMORE) || maxreached || error) {
1513 		sounsplice(so, sosp, 1);
1514 		return (0);
1515 	}
1516 	if (timerisset(&so->so_idletv))
1517 		timeout_add_tv(&so->so_idleto, &so->so_idletv);
1518 	return (1);
1519 }
1520 
1521 #endif /* SOCKET_SPLICE */
1522 
1523 void
1524 sorwakeup(struct socket *so)
1525 {
1526 	soassertlocked(so);
1527 
1528 #ifdef SOCKET_SPLICE
1529 	if (so->so_rcv.sb_flagsintr & SB_SPLICE) {
1530 		/*
1531 		 * TCP has a sendbuffer that can handle multiple packets
1532 		 * at once.  So queue the stream a bit to accumulate data.
1533 		 * The sosplice thread will call somove() later and send
1534 		 * the packets calling tcp_output() only once.
1535 		 * In the UDP case, send out the packets immediately.
1536 		 * Using a thread would make things slower.
1537 		 */
1538 		if (so->so_proto->pr_flags & PR_WANTRCVD)
1539 			task_add(sosplice_taskq, &so->so_splicetask);
1540 		else
1541 			somove(so, M_DONTWAIT);
1542 	}
1543 	if (isspliced(so))
1544 		return;
1545 #endif
1546 	sowakeup(so, &so->so_rcv);
1547 	if (so->so_upcall)
1548 		(*(so->so_upcall))(so, so->so_upcallarg, M_DONTWAIT);
1549 }
1550 
1551 void
1552 sowwakeup(struct socket *so)
1553 {
1554 	soassertlocked(so);
1555 
1556 #ifdef SOCKET_SPLICE
1557 	if (so->so_snd.sb_flagsintr & SB_SPLICE)
1558 		task_add(sosplice_taskq, &so->so_sp->ssp_soback->so_splicetask);
1559 #endif
1560 	sowakeup(so, &so->so_snd);
1561 }
1562 
1563 int
1564 sosetopt(struct socket *so, int level, int optname, struct mbuf *m)
1565 {
1566 	int error = 0;
1567 
1568 	soassertlocked(so);
1569 
1570 	if (level != SOL_SOCKET) {
1571 		if (so->so_proto->pr_ctloutput) {
1572 			error = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so,
1573 			    level, optname, m);
1574 			return (error);
1575 		}
1576 		error = ENOPROTOOPT;
1577 	} else {
1578 		switch (optname) {
1579 		case SO_BINDANY:
1580 			if ((error = suser(curproc, 0)) != 0)	/* XXX */
1581 				return (error);
1582 			break;
1583 		}
1584 
1585 		switch (optname) {
1586 
1587 		case SO_LINGER:
1588 			if (m == NULL || m->m_len != sizeof (struct linger) ||
1589 			    mtod(m, struct linger *)->l_linger < 0 ||
1590 			    mtod(m, struct linger *)->l_linger > SHRT_MAX)
1591 				return (EINVAL);
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 				return (EINVAL);
1607 			if (*mtod(m, int *))
1608 				so->so_options |= optname;
1609 			else
1610 				so->so_options &= ~optname;
1611 			break;
1612 
1613 		case SO_DONTROUTE:
1614 			if (m == NULL || m->m_len < sizeof (int))
1615 				return (EINVAL);
1616 			if (*mtod(m, int *))
1617 				error = EOPNOTSUPP;
1618 			break;
1619 
1620 		case SO_SNDBUF:
1621 		case SO_RCVBUF:
1622 		case SO_SNDLOWAT:
1623 		case SO_RCVLOWAT:
1624 		    {
1625 			u_long cnt;
1626 
1627 			if (m == NULL || m->m_len < sizeof (int))
1628 				return (EINVAL);
1629 			cnt = *mtod(m, int *);
1630 			if ((long)cnt <= 0)
1631 				cnt = 1;
1632 			switch (optname) {
1633 
1634 			case SO_SNDBUF:
1635 				if (so->so_state & SS_CANTSENDMORE)
1636 					return (EINVAL);
1637 				if (sbcheckreserve(cnt, so->so_snd.sb_wat) ||
1638 				    sbreserve(so, &so->so_snd, cnt))
1639 					return (ENOBUFS);
1640 				so->so_snd.sb_wat = cnt;
1641 				break;
1642 
1643 			case SO_RCVBUF:
1644 				if (so->so_state & SS_CANTRCVMORE)
1645 					return (EINVAL);
1646 				if (sbcheckreserve(cnt, so->so_rcv.sb_wat) ||
1647 				    sbreserve(so, &so->so_rcv, cnt))
1648 					return (ENOBUFS);
1649 				so->so_rcv.sb_wat = cnt;
1650 				break;
1651 
1652 			case SO_SNDLOWAT:
1653 				so->so_snd.sb_lowat =
1654 				    (cnt > so->so_snd.sb_hiwat) ?
1655 				    so->so_snd.sb_hiwat : cnt;
1656 				break;
1657 			case SO_RCVLOWAT:
1658 				so->so_rcv.sb_lowat =
1659 				    (cnt > so->so_rcv.sb_hiwat) ?
1660 				    so->so_rcv.sb_hiwat : cnt;
1661 				break;
1662 			}
1663 			break;
1664 		    }
1665 
1666 		case SO_SNDTIMEO:
1667 		case SO_RCVTIMEO:
1668 		    {
1669 			struct timeval tv;
1670 			int val;
1671 
1672 			if (m == NULL || m->m_len < sizeof (tv))
1673 				return (EINVAL);
1674 			memcpy(&tv, mtod(m, struct timeval *), sizeof tv);
1675 			val = tvtohz(&tv);
1676 			if (val > USHRT_MAX)
1677 				return (EDOM);
1678 
1679 			switch (optname) {
1680 
1681 			case SO_SNDTIMEO:
1682 				so->so_snd.sb_timeo = val;
1683 				break;
1684 			case SO_RCVTIMEO:
1685 				so->so_rcv.sb_timeo = val;
1686 				break;
1687 			}
1688 			break;
1689 		    }
1690 
1691 		case SO_RTABLE:
1692 			if (so->so_proto->pr_domain &&
1693 			    so->so_proto->pr_domain->dom_protosw &&
1694 			    so->so_proto->pr_ctloutput) {
1695 				struct domain *dom = so->so_proto->pr_domain;
1696 
1697 				level = dom->dom_protosw->pr_protocol;
1698 				error = (*so->so_proto->pr_ctloutput)
1699 				    (PRCO_SETOPT, so, level, optname, m);
1700 				return (error);
1701 			}
1702 			error = ENOPROTOOPT;
1703 			break;
1704 
1705 #ifdef SOCKET_SPLICE
1706 		case SO_SPLICE:
1707 			if (m == NULL) {
1708 				error = sosplice(so, -1, 0, NULL);
1709 			} else if (m->m_len < sizeof(int)) {
1710 				return (EINVAL);
1711 			} else if (m->m_len < sizeof(struct splice)) {
1712 				error = sosplice(so, *mtod(m, int *), 0, NULL);
1713 			} else {
1714 				error = sosplice(so,
1715 				    mtod(m, struct splice *)->sp_fd,
1716 				    mtod(m, struct splice *)->sp_max,
1717 				   &mtod(m, struct splice *)->sp_idle);
1718 			}
1719 			break;
1720 #endif /* SOCKET_SPLICE */
1721 
1722 		default:
1723 			error = ENOPROTOOPT;
1724 			break;
1725 		}
1726 		if (error == 0 && so->so_proto->pr_ctloutput) {
1727 			(*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so,
1728 			    level, optname, m);
1729 		}
1730 	}
1731 
1732 	return (error);
1733 }
1734 
1735 int
1736 sogetopt(struct socket *so, int level, int optname, struct mbuf *m)
1737 {
1738 	int error = 0;
1739 
1740 	soassertlocked(so);
1741 
1742 	if (level != SOL_SOCKET) {
1743 		if (so->so_proto->pr_ctloutput) {
1744 			m->m_len = 0;
1745 
1746 			error = (*so->so_proto->pr_ctloutput)(PRCO_GETOPT, so,
1747 			    level, optname, m);
1748 			if (error)
1749 				return (error);
1750 			return (0);
1751 		} else
1752 			return (ENOPROTOOPT);
1753 	} else {
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 		case SO_ZEROIZE:
1775 			*mtod(m, int *) = so->so_options & optname;
1776 			break;
1777 
1778 		case SO_DONTROUTE:
1779 			*mtod(m, int *) = 0;
1780 			break;
1781 
1782 		case SO_TYPE:
1783 			*mtod(m, int *) = so->so_type;
1784 			break;
1785 
1786 		case SO_ERROR:
1787 			*mtod(m, int *) = so->so_error;
1788 			so->so_error = 0;
1789 			break;
1790 
1791 		case SO_SNDBUF:
1792 			*mtod(m, int *) = so->so_snd.sb_hiwat;
1793 			break;
1794 
1795 		case SO_RCVBUF:
1796 			*mtod(m, int *) = so->so_rcv.sb_hiwat;
1797 			break;
1798 
1799 		case SO_SNDLOWAT:
1800 			*mtod(m, int *) = so->so_snd.sb_lowat;
1801 			break;
1802 
1803 		case SO_RCVLOWAT:
1804 			*mtod(m, int *) = so->so_rcv.sb_lowat;
1805 			break;
1806 
1807 		case SO_SNDTIMEO:
1808 		case SO_RCVTIMEO:
1809 		    {
1810 			struct timeval tv;
1811 			int val = (optname == SO_SNDTIMEO ?
1812 			    so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1813 
1814 			m->m_len = sizeof(struct timeval);
1815 			memset(&tv, 0, sizeof(tv));
1816 			tv.tv_sec = val / hz;
1817 			tv.tv_usec = (val % hz) * tick;
1818 			memcpy(mtod(m, struct timeval *), &tv, sizeof tv);
1819 			break;
1820 		    }
1821 
1822 		case SO_RTABLE:
1823 			if (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 				error = (*so->so_proto->pr_ctloutput)
1830 				    (PRCO_GETOPT, so, level, optname, m);
1831 				if (error)
1832 					return (error);
1833 				break;
1834 			}
1835 			return (ENOPROTOOPT);
1836 
1837 #ifdef SOCKET_SPLICE
1838 		case SO_SPLICE:
1839 		    {
1840 			off_t len;
1841 
1842 			m->m_len = sizeof(off_t);
1843 			len = so->so_sp ? so->so_sp->ssp_len : 0;
1844 			memcpy(mtod(m, off_t *), &len, sizeof(off_t));
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 				return (ENOTCONN);
1860 			}
1861 			return (EOPNOTSUPP);
1862 
1863 		default:
1864 			return (ENOPROTOOPT);
1865 		}
1866 		return (0);
1867 	}
1868 }
1869 
1870 void
1871 sohasoutofband(struct socket *so)
1872 {
1873 	KERNEL_ASSERT_LOCKED();
1874 	csignal(so->so_pgid, SIGURG, so->so_siguid, so->so_sigeuid);
1875 	selwakeup(&so->so_rcv.sb_sel);
1876 }
1877 
1878 int
1879 soo_kqfilter(struct file *fp, struct knote *kn)
1880 {
1881 	struct socket *so = kn->kn_fp->f_data;
1882 	struct sockbuf *sb;
1883 
1884 	KERNEL_ASSERT_LOCKED();
1885 
1886 	switch (kn->kn_filter) {
1887 	case EVFILT_READ:
1888 		if (so->so_options & SO_ACCEPTCONN)
1889 			kn->kn_fop = &solisten_filtops;
1890 		else
1891 			kn->kn_fop = &soread_filtops;
1892 		sb = &so->so_rcv;
1893 		break;
1894 	case EVFILT_WRITE:
1895 		kn->kn_fop = &sowrite_filtops;
1896 		sb = &so->so_snd;
1897 		break;
1898 	default:
1899 		return (EINVAL);
1900 	}
1901 
1902 	SLIST_INSERT_HEAD(&sb->sb_sel.si_note, kn, kn_selnext);
1903 	sb->sb_flags |= SB_KNOTE;
1904 
1905 	return (0);
1906 }
1907 
1908 void
1909 filt_sordetach(struct knote *kn)
1910 {
1911 	struct socket *so = kn->kn_fp->f_data;
1912 
1913 	KERNEL_ASSERT_LOCKED();
1914 
1915 	SLIST_REMOVE(&so->so_rcv.sb_sel.si_note, kn, knote, kn_selnext);
1916 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.si_note))
1917 		so->so_rcv.sb_flags &= ~SB_KNOTE;
1918 }
1919 
1920 int
1921 filt_soread(struct knote *kn, long hint)
1922 {
1923 	struct socket *so = kn->kn_fp->f_data;
1924 	int rv;
1925 
1926 	kn->kn_data = so->so_rcv.sb_cc;
1927 #ifdef SOCKET_SPLICE
1928 	if (isspliced(so)) {
1929 		rv = 0;
1930 	} else
1931 #endif /* SOCKET_SPLICE */
1932 	if (so->so_state & SS_CANTRCVMORE) {
1933 		kn->kn_flags |= EV_EOF;
1934 		kn->kn_fflags = so->so_error;
1935 		rv = 1;
1936 	} else if (so->so_error) {	/* temporary udp error */
1937 		rv = 1;
1938 	} else if (kn->kn_sfflags & NOTE_LOWAT) {
1939 		rv = (kn->kn_data >= kn->kn_sdata);
1940 	} else {
1941 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
1942 	}
1943 
1944 	return rv;
1945 }
1946 
1947 void
1948 filt_sowdetach(struct knote *kn)
1949 {
1950 	struct socket *so = kn->kn_fp->f_data;
1951 
1952 	KERNEL_ASSERT_LOCKED();
1953 
1954 	SLIST_REMOVE(&so->so_snd.sb_sel.si_note, kn, knote, kn_selnext);
1955 	if (SLIST_EMPTY(&so->so_snd.sb_sel.si_note))
1956 		so->so_snd.sb_flags &= ~SB_KNOTE;
1957 }
1958 
1959 int
1960 filt_sowrite(struct knote *kn, long hint)
1961 {
1962 	struct socket *so = kn->kn_fp->f_data;
1963 	int rv;
1964 
1965 	kn->kn_data = sbspace(so, &so->so_snd);
1966 	if (so->so_state & SS_CANTSENDMORE) {
1967 		kn->kn_flags |= EV_EOF;
1968 		kn->kn_fflags = so->so_error;
1969 		rv = 1;
1970 	} else if (so->so_error) {	/* temporary udp error */
1971 		rv = 1;
1972 	} else if (((so->so_state & SS_ISCONNECTED) == 0) &&
1973 	    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1974 		rv = 0;
1975 	} else if (kn->kn_sfflags & NOTE_LOWAT) {
1976 		rv = (kn->kn_data >= kn->kn_sdata);
1977 	} else {
1978 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
1979 	}
1980 
1981 	return (rv);
1982 }
1983 
1984 int
1985 filt_solisten(struct knote *kn, long hint)
1986 {
1987 	struct socket *so = kn->kn_fp->f_data;
1988 
1989 	kn->kn_data = so->so_qlen;
1990 
1991 	return (kn->kn_data != 0);
1992 }
1993 
1994 #ifdef DDB
1995 void
1996 sobuf_print(struct sockbuf *,
1997     int (*)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))));
1998 
1999 void
2000 sobuf_print(struct sockbuf *sb,
2001     int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2002 {
2003 	(*pr)("\tsb_cc: %lu\n", sb->sb_cc);
2004 	(*pr)("\tsb_datacc: %lu\n", sb->sb_datacc);
2005 	(*pr)("\tsb_hiwat: %lu\n", sb->sb_hiwat);
2006 	(*pr)("\tsb_wat: %lu\n", sb->sb_wat);
2007 	(*pr)("\tsb_mbcnt: %lu\n", sb->sb_mbcnt);
2008 	(*pr)("\tsb_mbmax: %lu\n", sb->sb_mbmax);
2009 	(*pr)("\tsb_lowat: %ld\n", sb->sb_lowat);
2010 	(*pr)("\tsb_mb: %p\n", sb->sb_mb);
2011 	(*pr)("\tsb_mbtail: %p\n", sb->sb_mbtail);
2012 	(*pr)("\tsb_lastrecord: %p\n", sb->sb_lastrecord);
2013 	(*pr)("\tsb_sel: ...\n");
2014 	(*pr)("\tsb_flagsintr: %d\n", sb->sb_flagsintr);
2015 	(*pr)("\tsb_flags: %i\n", sb->sb_flags);
2016 	(*pr)("\tsb_timeo: %i\n", sb->sb_timeo);
2017 }
2018 
2019 void
2020 so_print(void *v,
2021     int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2022 {
2023 	struct socket *so = v;
2024 
2025 	(*pr)("socket %p\n", so);
2026 	(*pr)("so_type: %i\n", so->so_type);
2027 	(*pr)("so_options: 0x%04x\n", so->so_options); /* %b */
2028 	(*pr)("so_linger: %i\n", so->so_linger);
2029 	(*pr)("so_state: 0x%04x\n", so->so_state);
2030 	(*pr)("so_pcb: %p\n", so->so_pcb);
2031 	(*pr)("so_proto: %p\n", so->so_proto);
2032 
2033 	(*pr)("so_head: %p\n", so->so_head);
2034 	(*pr)("so_onq: %p\n", so->so_onq);
2035 	(*pr)("so_q0: @%p first: %p\n", &so->so_q0, TAILQ_FIRST(&so->so_q0));
2036 	(*pr)("so_q: @%p first: %p\n", &so->so_q, TAILQ_FIRST(&so->so_q));
2037 	(*pr)("so_eq: next: %p\n", TAILQ_NEXT(so, so_qe));
2038 	(*pr)("so_q0len: %i\n", so->so_q0len);
2039 	(*pr)("so_qlen: %i\n", so->so_qlen);
2040 	(*pr)("so_qlimit: %i\n", so->so_qlimit);
2041 	(*pr)("so_timeo: %i\n", so->so_timeo);
2042 	(*pr)("so_pgid: %i\n", so->so_pgid);
2043 	(*pr)("so_siguid: %i\n", so->so_siguid);
2044 	(*pr)("so_sigeuid: %i\n", so->so_sigeuid);
2045 	(*pr)("so_obmark: %lu\n", so->so_oobmark);
2046 
2047 	(*pr)("so_sp: %p\n", so->so_sp);
2048 	if (so->so_sp != NULL) {
2049 		(*pr)("\tssp_socket: %p\n", so->so_sp->ssp_socket);
2050 		(*pr)("\tssp_soback: %p\n", so->so_sp->ssp_soback);
2051 		(*pr)("\tssp_len: %lld\n",
2052 		    (unsigned long long)so->so_sp->ssp_len);
2053 		(*pr)("\tssp_max: %lld\n",
2054 		    (unsigned long long)so->so_sp->ssp_max);
2055 		(*pr)("\tssp_idletv: %lld %ld\n", so->so_sp->ssp_idletv.tv_sec,
2056 		    so->so_sp->ssp_idletv.tv_usec);
2057 		(*pr)("\tssp_idleto: %spending (@%i)\n",
2058 		    timeout_pending(&so->so_sp->ssp_idleto) ? "" : "not ",
2059 		    so->so_sp->ssp_idleto.to_time);
2060 	}
2061 
2062 	(*pr)("so_rcv:\n");
2063 	sobuf_print(&so->so_rcv, pr);
2064 	(*pr)("so_snd:\n");
2065 	sobuf_print(&so->so_snd, pr);
2066 
2067 	(*pr)("so_upcall: %p so_upcallarg: %p\n",
2068 	    so->so_upcall, so->so_upcallarg);
2069 
2070 	(*pr)("so_euid: %d so_ruid: %d\n", so->so_euid, so->so_ruid);
2071 	(*pr)("so_egid: %d so_rgid: %d\n", so->so_egid, so->so_rgid);
2072 	(*pr)("so_cpid: %d\n", so->so_cpid);
2073 }
2074 #endif
2075 
2076