xref: /openbsd-src/sys/kern/uipc_socket.c (revision 897fc685943471cf985a0fe38ba076ea6fe74fa5)
1 /*	$OpenBSD: uipc_socket.c,v 1.220 2018/04/08 18:57:39 guenther 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 void	soput(void *);
64 int	somove(struct socket *, int);
65 
66 void	filt_sordetach(struct knote *kn);
67 int	filt_soread(struct knote *kn, long hint);
68 void	filt_sowdetach(struct knote *kn);
69 int	filt_sowrite(struct knote *kn, long hint);
70 int	filt_solisten(struct knote *kn, long hint);
71 
72 struct filterops solisten_filtops =
73 	{ 1, NULL, filt_sordetach, filt_solisten };
74 struct filterops soread_filtops =
75 	{ 1, NULL, filt_sordetach, filt_soread };
76 struct filterops sowrite_filtops =
77 	{ 1, NULL, filt_sowdetach, filt_sowrite };
78 
79 
80 #ifndef SOMINCONN
81 #define SOMINCONN 80
82 #endif /* SOMINCONN */
83 
84 int	somaxconn = SOMAXCONN;
85 int	sominconn = SOMINCONN;
86 
87 struct pool socket_pool;
88 #ifdef SOCKET_SPLICE
89 struct pool sosplice_pool;
90 struct taskq *sosplice_taskq;
91 #endif
92 
93 void
94 soinit(void)
95 {
96 	pool_init(&socket_pool, sizeof(struct socket), 0, IPL_SOFTNET, 0,
97 	    "sockpl", NULL);
98 #ifdef SOCKET_SPLICE
99 	pool_init(&sosplice_pool, sizeof(struct sosplice), 0, IPL_SOFTNET, 0,
100 	    "sosppl", NULL);
101 #endif
102 }
103 
104 /*
105  * Socket operation routines.
106  * These routines are called by the routines in
107  * sys_socket.c or from a system process, and
108  * implement the semantics of socket operations by
109  * switching out to the protocol specific routines.
110  */
111 int
112 socreate(int dom, struct socket **aso, int type, int proto)
113 {
114 	struct proc *p = curproc;		/* XXX */
115 	const struct protosw *prp;
116 	struct socket *so;
117 	int error, s;
118 
119 	if (proto)
120 		prp = pffindproto(dom, proto, type);
121 	else
122 		prp = pffindtype(dom, type);
123 	if (prp == NULL || prp->pr_attach == NULL)
124 		return (EPROTONOSUPPORT);
125 	if (prp->pr_type != type)
126 		return (EPROTOTYPE);
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)
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 
140 	s = solock(so);
141 	error = (*prp->pr_attach)(so, proto);
142 	if (error) {
143 		so->so_state |= SS_NOFDREF;
144 		sofree(so);
145 		sounlock(s);
146 		return (error);
147 	}
148 	sounlock(s);
149 	*aso = so;
150 	return (0);
151 }
152 
153 int
154 sobind(struct socket *so, struct mbuf *nam, struct proc *p)
155 {
156 	int error;
157 
158 	soassertlocked(so);
159 
160 	error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, NULL, nam, NULL, p);
161 	return (error);
162 }
163 
164 int
165 solisten(struct socket *so, int backlog)
166 {
167 	int s, error;
168 
169 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING|SS_ISDISCONNECTING))
170 		return (EOPNOTSUPP);
171 #ifdef SOCKET_SPLICE
172 	if (isspliced(so) || issplicedback(so))
173 		return (EOPNOTSUPP);
174 #endif /* SOCKET_SPLICE */
175 	s = solock(so);
176 	error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, NULL, NULL, NULL,
177 	    curproc);
178 	if (error) {
179 		sounlock(s);
180 		return (error);
181 	}
182 	if (TAILQ_FIRST(&so->so_q) == NULL)
183 		so->so_options |= SO_ACCEPTCONN;
184 	if (backlog < 0 || backlog > somaxconn)
185 		backlog = somaxconn;
186 	if (backlog < sominconn)
187 		backlog = sominconn;
188 	so->so_qlimit = backlog;
189 	sounlock(s);
190 	return (0);
191 }
192 
193 void
194 sofree(struct socket *so)
195 {
196 	soassertlocked(so);
197 
198 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0)
199 		return;
200 	if (so->so_head) {
201 		/*
202 		 * We must not decommission a socket that's on the accept(2)
203 		 * queue.  If we do, then accept(2) may hang after select(2)
204 		 * indicated that the listening socket was ready.
205 		 */
206 		if (!soqremque(so, 0))
207 			return;
208 	}
209 #ifdef SOCKET_SPLICE
210 	if (so->so_sp) {
211 		if (issplicedback(so))
212 			sounsplice(so->so_sp->ssp_soback, so,
213 			    so->so_sp->ssp_soback != so);
214 		if (isspliced(so))
215 			sounsplice(so, so->so_sp->ssp_socket, 0);
216 	}
217 #endif /* SOCKET_SPLICE */
218 	sbrelease(so, &so->so_snd);
219 	sorflush(so);
220 #ifdef SOCKET_SPLICE
221 	if (so->so_sp) {
222 		/* Reuse splice task, sounsplice() has been called before. */
223 		task_set(&so->so_sp->ssp_task, soput, so);
224 		task_add(sosplice_taskq, &so->so_sp->ssp_task);
225 	} else
226 #endif /* SOCKET_SPLICE */
227 	{
228 		pool_put(&socket_pool, so);
229 	}
230 }
231 
232 /*
233  * Close a socket on last file table reference removal.
234  * Initiate disconnect if connected.
235  * Free socket when disconnect complete.
236  */
237 int
238 soclose(struct socket *so)
239 {
240 	struct socket *so2;
241 	int s, error = 0;
242 
243 	s = solock(so);
244 	if (so->so_options & SO_ACCEPTCONN) {
245 		while ((so2 = TAILQ_FIRST(&so->so_q0)) != NULL) {
246 			(void) soqremque(so2, 0);
247 			(void) soabort(so2);
248 		}
249 		while ((so2 = TAILQ_FIRST(&so->so_q)) != NULL) {
250 			(void) soqremque(so2, 1);
251 			(void) soabort(so2);
252 		}
253 	}
254 	if (so->so_pcb == 0)
255 		goto discard;
256 	if (so->so_state & SS_ISCONNECTED) {
257 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
258 			error = sodisconnect(so);
259 			if (error)
260 				goto drop;
261 		}
262 		if (so->so_options & SO_LINGER) {
263 			if ((so->so_state & SS_ISDISCONNECTING) &&
264 			    (so->so_state & SS_NBIO))
265 				goto drop;
266 			while (so->so_state & SS_ISCONNECTED) {
267 				error = sosleep(so, &so->so_timeo,
268 				    PSOCK | PCATCH, "netcls",
269 				    so->so_linger * hz);
270 				if (error)
271 					break;
272 			}
273 		}
274 	}
275 drop:
276 	if (so->so_pcb) {
277 		int error2;
278 		KASSERT(so->so_proto->pr_detach);
279 		error2 = (*so->so_proto->pr_detach)(so);
280 		if (error == 0)
281 			error = error2;
282 	}
283 discard:
284 	if (so->so_state & SS_NOFDREF)
285 		panic("soclose NOFDREF: so %p, so_type %d", so, so->so_type);
286 	so->so_state |= SS_NOFDREF;
287 	sofree(so);
288 	sounlock(s);
289 	return (error);
290 }
291 
292 int
293 soabort(struct socket *so)
294 {
295 	soassertlocked(so);
296 
297 	return (*so->so_proto->pr_usrreq)(so, PRU_ABORT, NULL, NULL, NULL,
298 	   curproc);
299 }
300 
301 int
302 soaccept(struct socket *so, struct mbuf *nam)
303 {
304 	int error = 0;
305 
306 	soassertlocked(so);
307 
308 	if ((so->so_state & SS_NOFDREF) == 0)
309 		panic("soaccept !NOFDREF: so %p, so_type %d", so, so->so_type);
310 	so->so_state &= ~SS_NOFDREF;
311 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
312 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
313 		error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT, NULL,
314 		    nam, NULL, curproc);
315 	else
316 		error = ECONNABORTED;
317 	return (error);
318 }
319 
320 int
321 soconnect(struct socket *so, struct mbuf *nam)
322 {
323 	int error;
324 
325 	soassertlocked(so);
326 
327 	if (so->so_options & SO_ACCEPTCONN)
328 		return (EOPNOTSUPP);
329 	/*
330 	 * If protocol is connection-based, can only connect once.
331 	 * Otherwise, if connected, try to disconnect first.
332 	 * This allows user to disconnect by connecting to, e.g.,
333 	 * a null address.
334 	 */
335 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
336 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
337 	    (error = sodisconnect(so))))
338 		error = EISCONN;
339 	else
340 		error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT,
341 		    NULL, nam, NULL, curproc);
342 	return (error);
343 }
344 
345 int
346 soconnect2(struct socket *so1, struct socket *so2)
347 {
348 	int s, error;
349 
350 	s = solock(so1);
351 	error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2, NULL,
352 	    (struct mbuf *)so2, NULL, curproc);
353 	sounlock(s);
354 	return (error);
355 }
356 
357 int
358 sodisconnect(struct socket *so)
359 {
360 	int error;
361 
362 	soassertlocked(so);
363 
364 	if ((so->so_state & SS_ISCONNECTED) == 0)
365 		return (ENOTCONN);
366 	if (so->so_state & SS_ISDISCONNECTING)
367 		return (EALREADY);
368 	error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT, NULL, NULL,
369 	    NULL, curproc);
370 	return (error);
371 }
372 
373 int m_getuio(struct mbuf **, int, long, struct uio *);
374 
375 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
376 /*
377  * Send on a socket.
378  * If send must go all at once and message is larger than
379  * send buffering, then hard error.
380  * Lock against other senders.
381  * If must go all at once and not enough room now, then
382  * inform user that this would block and do nothing.
383  * Otherwise, if nonblocking, send as much as possible.
384  * The data to be sent is described by "uio" if nonzero,
385  * otherwise by the mbuf chain "top" (which must be null
386  * if uio is not).  Data provided in mbuf chain must be small
387  * enough to send all at once.
388  *
389  * Returns nonzero on error, timeout or signal; callers
390  * must check for short counts if EINTR/ERESTART are returned.
391  * Data and control buffers are freed on return.
392  */
393 int
394 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
395     struct mbuf *control, int flags)
396 {
397 	long space, clen = 0;
398 	size_t resid;
399 	int error, s;
400 	int atomic = sosendallatonce(so) || top;
401 
402 	if (uio)
403 		resid = uio->uio_resid;
404 	else
405 		resid = top->m_pkthdr.len;
406 	/* MSG_EOR on a SOCK_STREAM socket is invalid. */
407 	if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
408 		m_freem(top);
409 		m_freem(control);
410 		return (EINVAL);
411 	}
412 	if (uio && uio->uio_procp)
413 		uio->uio_procp->p_ru.ru_msgsnd++;
414 	if (control) {
415 		/*
416 		 * In theory clen should be unsigned (since control->m_len is).
417 		 * However, space must be signed, as it might be less than 0
418 		 * if we over-committed, and we must use a signed comparison
419 		 * of space and clen.
420 		 */
421 		clen = control->m_len;
422 		/* reserve extra space for AF_UNIX's internalize */
423 		if (so->so_proto->pr_domain->dom_family == AF_UNIX &&
424 		    clen >= CMSG_ALIGN(sizeof(struct cmsghdr)) &&
425 		    mtod(control, struct cmsghdr *)->cmsg_type == SCM_RIGHTS)
426 			clen = CMSG_SPACE(
427 			    (clen - CMSG_ALIGN(sizeof(struct cmsghdr))) *
428 			    (sizeof(struct fdpass) / sizeof(int)));
429 	}
430 
431 #define	snderr(errno)	{ error = errno; goto release; }
432 
433 	s = solock(so);
434 restart:
435 	if ((error = sblock(so, &so->so_snd, SBLOCKWAIT(flags))) != 0)
436 		goto out;
437 	so->so_state |= SS_ISSENDING;
438 	do {
439 		if (so->so_state & SS_CANTSENDMORE)
440 			snderr(EPIPE);
441 		if (so->so_error) {
442 			error = so->so_error;
443 			so->so_error = 0;
444 			snderr(error);
445 		}
446 		if ((so->so_state & SS_ISCONNECTED) == 0) {
447 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
448 				if (!(resid == 0 && clen != 0))
449 					snderr(ENOTCONN);
450 			} else if (addr == 0)
451 				snderr(EDESTADDRREQ);
452 		}
453 		space = sbspace(so, &so->so_snd);
454 		if (flags & MSG_OOB)
455 			space += 1024;
456 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
457 		    (so->so_proto->pr_domain->dom_family != AF_UNIX &&
458 		    clen > so->so_snd.sb_hiwat))
459 			snderr(EMSGSIZE);
460 		if (space < clen ||
461 		    (space - clen < resid &&
462 		    (atomic || space < so->so_snd.sb_lowat))) {
463 			if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT))
464 				snderr(EWOULDBLOCK);
465 			sbunlock(so, &so->so_snd);
466 			error = sbwait(so, &so->so_snd);
467 			so->so_state &= ~SS_ISSENDING;
468 			if (error)
469 				goto out;
470 			goto restart;
471 		}
472 		space -= clen;
473 		do {
474 			if (uio == NULL) {
475 				/*
476 				 * Data is prepackaged in "top".
477 				 */
478 				resid = 0;
479 				if (flags & MSG_EOR)
480 					top->m_flags |= M_EOR;
481 			} else {
482 				sounlock(s);
483 				error = m_getuio(&top, atomic, space, uio);
484 				s = solock(so);
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 			if (resid == 0)
493 				so->so_state &= ~SS_ISSENDING;
494 			if (top && so->so_options & SO_ZEROIZE)
495 				top->m_flags |= M_ZEROIZE;
496 			error = (*so->so_proto->pr_usrreq)(so,
497 			    (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND,
498 			    top, addr, control, curproc);
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->so_snd);
510 out:
511 	sounlock(s);
512 	m_freem(top);
513 	m_freem(control);
514 	return (error);
515 }
516 
517 int
518 m_getuio(struct mbuf **mp, int atomic, long space, struct uio *uio)
519 {
520 	struct mbuf *m, *top = NULL;
521 	struct mbuf **nextp = &top;
522 	u_long len, mlen;
523 	size_t resid = uio->uio_resid;
524 	int error;
525 
526 	do {
527 		if (top == NULL) {
528 			MGETHDR(m, M_WAIT, MT_DATA);
529 			mlen = MHLEN;
530 			m->m_pkthdr.len = 0;
531 			m->m_pkthdr.ph_ifidx = 0;
532 		} else {
533 			MGET(m, M_WAIT, MT_DATA);
534 			mlen = MLEN;
535 		}
536 		/* chain mbuf together */
537 		*nextp = m;
538 		nextp = &m->m_next;
539 
540 		resid = ulmin(resid, space);
541 		if (resid >= MINCLSIZE) {
542 			MCLGETI(m, M_NOWAIT, NULL, ulmin(resid, MAXMCLBYTES));
543 			if ((m->m_flags & M_EXT) == 0)
544 				MCLGETI(m, M_NOWAIT, NULL, MCLBYTES);
545 			if ((m->m_flags & M_EXT) == 0)
546 				goto nopages;
547 			mlen = m->m_ext.ext_size;
548 			len = ulmin(mlen, resid);
549 			/*
550 			 * For datagram protocols, leave room
551 			 * for protocol headers in first mbuf.
552 			 */
553 			if (atomic && m == top && len < mlen - max_hdr)
554 				m->m_data += max_hdr;
555 		} else {
556 nopages:
557 			len = ulmin(mlen, resid);
558 			/*
559 			 * For datagram protocols, leave room
560 			 * for protocol headers in first mbuf.
561 			 */
562 			if (atomic && m == top && len < mlen - max_hdr)
563 				MH_ALIGN(m, len);
564 		}
565 
566 		error = uiomove(mtod(m, caddr_t), len, uio);
567 		if (error) {
568 			m_freem(top);
569 			return (error);
570 		}
571 
572 		/* adjust counters */
573 		resid = uio->uio_resid;
574 		space -= len;
575 		m->m_len = len;
576 		top->m_pkthdr.len += len;
577 
578 		/* Is there more space and more data? */
579 	} while (space > 0 && resid > 0);
580 
581 	*mp = top;
582 	return 0;
583 }
584 
585 /*
586  * Following replacement or removal of the first mbuf on the first
587  * mbuf chain of a socket buffer, push necessary state changes back
588  * into the socket buffer so that other consumers see the values
589  * consistently.  'nextrecord' is the callers locally stored value of
590  * the original value of sb->sb_mb->m_nextpkt which must be restored
591  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
592  */
593 void
594 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
595 {
596 
597 	/*
598 	 * First, update for the new value of nextrecord.  If necessary,
599 	 * make it the first record.
600 	 */
601 	if (sb->sb_mb != NULL)
602 		sb->sb_mb->m_nextpkt = nextrecord;
603 	else
604 		sb->sb_mb = nextrecord;
605 
606 	/*
607 	 * Now update any dependent socket buffer fields to reflect
608 	 * the new state.  This is an inline of SB_EMPTY_FIXUP, with
609 	 * the addition of a second clause that takes care of the
610 	 * case where sb_mb has been updated, but remains the last
611 	 * record.
612 	 */
613 	if (sb->sb_mb == NULL) {
614 		sb->sb_mbtail = NULL;
615 		sb->sb_lastrecord = NULL;
616 	} else if (sb->sb_mb->m_nextpkt == NULL)
617 		sb->sb_lastrecord = sb->sb_mb;
618 }
619 
620 /*
621  * Implement receive operations on a socket.
622  * We depend on the way that records are added to the sockbuf
623  * by sbappend*.  In particular, each record (mbufs linked through m_next)
624  * must begin with an address if the protocol so specifies,
625  * followed by an optional mbuf or mbufs containing ancillary data,
626  * and then zero or more mbufs of data.
627  * In order to avoid blocking network for the entire time here, we release
628  * the solock() while doing the actual copy to user space.
629  * Although the sockbuf is locked, new data may still be appended,
630  * and thus we must maintain consistency of the sockbuf during that time.
631  *
632  * The caller may receive the data as a single mbuf chain by supplying
633  * an mbuf **mp0 for use in returning the chain.  The uio is then used
634  * only for the count in uio_resid.
635  */
636 int
637 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
638     struct mbuf **mp0, struct mbuf **controlp, int *flagsp,
639     socklen_t controllen)
640 {
641 	struct mbuf *m, **mp;
642 	struct mbuf *cm;
643 	u_long len, offset, moff;
644 	int flags, error, s, type, uio_error = 0;
645 	const struct protosw *pr = so->so_proto;
646 	struct mbuf *nextrecord;
647 	size_t resid, orig_resid = uio->uio_resid;
648 
649 	mp = mp0;
650 	if (paddr)
651 		*paddr = 0;
652 	if (controlp)
653 		*controlp = 0;
654 	if (flagsp)
655 		flags = *flagsp &~ MSG_EOR;
656 	else
657 		flags = 0;
658 	if (so->so_state & SS_NBIO)
659 		flags |= MSG_DONTWAIT;
660 	if (flags & MSG_OOB) {
661 		m = m_get(M_WAIT, MT_DATA);
662 		s = solock(so);
663 		error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m,
664 		    (struct mbuf *)(long)(flags & MSG_PEEK), NULL, curproc);
665 		sounlock(s);
666 		if (error)
667 			goto bad;
668 		do {
669 			error = uiomove(mtod(m, caddr_t),
670 			    ulmin(uio->uio_resid, m->m_len), uio);
671 			m = m_free(m);
672 		} while (uio->uio_resid && error == 0 && m);
673 bad:
674 		m_freem(m);
675 		return (error);
676 	}
677 	if (mp)
678 		*mp = NULL;
679 
680 	s = solock(so);
681 restart:
682 	if ((error = sblock(so, &so->so_rcv, SBLOCKWAIT(flags))) != 0) {
683 		sounlock(s);
684 		return (error);
685 	}
686 
687 	m = so->so_rcv.sb_mb;
688 #ifdef SOCKET_SPLICE
689 	if (isspliced(so))
690 		m = NULL;
691 #endif /* SOCKET_SPLICE */
692 	/*
693 	 * If we have less data than requested, block awaiting more
694 	 * (subject to any timeout) if:
695 	 *   1. the current count is less than the low water mark,
696 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
697 	 *	receive operation at once if we block (resid <= hiwat), or
698 	 *   3. MSG_DONTWAIT is not set.
699 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
700 	 * we have to do the receive in sections, and thus risk returning
701 	 * a short count if a timeout or signal occurs after we start.
702 	 */
703 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
704 	    so->so_rcv.sb_cc < uio->uio_resid) &&
705 	    (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
706 	    ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
707 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
708 #ifdef DIAGNOSTIC
709 		if (m == NULL && so->so_rcv.sb_cc)
710 #ifdef SOCKET_SPLICE
711 		    if (!isspliced(so))
712 #endif /* SOCKET_SPLICE */
713 			panic("receive 1: so %p, so_type %d, sb_cc %lu",
714 			    so, so->so_type, so->so_rcv.sb_cc);
715 #endif
716 		if (so->so_error) {
717 			if (m)
718 				goto dontblock;
719 			error = so->so_error;
720 			if ((flags & MSG_PEEK) == 0)
721 				so->so_error = 0;
722 			goto release;
723 		}
724 		if (so->so_state & SS_CANTRCVMORE) {
725 			if (m)
726 				goto dontblock;
727 			else if (so->so_rcv.sb_cc == 0)
728 				goto release;
729 		}
730 		for (; m; m = m->m_next)
731 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
732 				m = so->so_rcv.sb_mb;
733 				goto dontblock;
734 			}
735 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
736 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
737 			error = ENOTCONN;
738 			goto release;
739 		}
740 		if (uio->uio_resid == 0 && controlp == NULL)
741 			goto release;
742 		if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT)) {
743 			error = EWOULDBLOCK;
744 			goto release;
745 		}
746 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
747 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
748 		sbunlock(so, &so->so_rcv);
749 		error = sbwait(so, &so->so_rcv);
750 		if (error) {
751 			sounlock(s);
752 			return (error);
753 		}
754 		goto restart;
755 	}
756 dontblock:
757 	/*
758 	 * On entry here, m points to the first record of the socket buffer.
759 	 * From this point onward, we maintain 'nextrecord' as a cache of the
760 	 * pointer to the next record in the socket buffer.  We must keep the
761 	 * various socket buffer pointers and local stack versions of the
762 	 * pointers in sync, pushing out modifications before operations that
763 	 * may sleep, and re-reading them afterwards.
764 	 *
765 	 * Otherwise, we will race with the network stack appending new data
766 	 * or records onto the socket buffer by using inconsistent/stale
767 	 * versions of the field, possibly resulting in socket buffer
768 	 * corruption.
769 	 */
770 	if (uio->uio_procp)
771 		uio->uio_procp->p_ru.ru_msgrcv++;
772 	KASSERT(m == so->so_rcv.sb_mb);
773 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
774 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
775 	nextrecord = m->m_nextpkt;
776 	if (pr->pr_flags & PR_ADDR) {
777 #ifdef DIAGNOSTIC
778 		if (m->m_type != MT_SONAME)
779 			panic("receive 1a: so %p, so_type %d, m %p, m_type %d",
780 			    so, so->so_type, m, m->m_type);
781 #endif
782 		orig_resid = 0;
783 		if (flags & MSG_PEEK) {
784 			if (paddr)
785 				*paddr = m_copym(m, 0, m->m_len, M_NOWAIT);
786 			m = m->m_next;
787 		} else {
788 			sbfree(&so->so_rcv, m);
789 			if (paddr) {
790 				*paddr = m;
791 				so->so_rcv.sb_mb = m->m_next;
792 				m->m_next = 0;
793 				m = so->so_rcv.sb_mb;
794 			} else {
795 				so->so_rcv.sb_mb = m_free(m);
796 				m = so->so_rcv.sb_mb;
797 			}
798 			sbsync(&so->so_rcv, nextrecord);
799 		}
800 	}
801 	while (m && m->m_type == MT_CONTROL && error == 0) {
802 		if (flags & MSG_PEEK) {
803 			if (controlp)
804 				*controlp = m_copym(m, 0, m->m_len, M_NOWAIT);
805 			m = m->m_next;
806 		} else {
807 			sbfree(&so->so_rcv, m);
808 			so->so_rcv.sb_mb = m->m_next;
809 			m->m_nextpkt = m->m_next = NULL;
810 			cm = m;
811 			m = so->so_rcv.sb_mb;
812 			sbsync(&so->so_rcv, nextrecord);
813 			if (controlp) {
814 				if (pr->pr_domain->dom_externalize &&
815 				    mtod(cm, struct cmsghdr *)->cmsg_type ==
816 				    SCM_RIGHTS) {
817 					error =
818 					    (*pr->pr_domain->dom_externalize)
819 					    (cm, controllen, flags);
820 				}
821 				*controlp = cm;
822 			} else {
823 				/*
824 				 * Dispose of any SCM_RIGHTS message that went
825 				 * through the read path rather than recv.
826 				 */
827 				if (pr->pr_domain->dom_dispose &&
828 				    mtod(cm, struct cmsghdr *)->cmsg_type == SCM_RIGHTS)
829 					pr->pr_domain->dom_dispose(cm);
830 				m_free(cm);
831 			}
832 		}
833 		if (m != NULL)
834 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
835 		else
836 			nextrecord = so->so_rcv.sb_mb;
837 		if (controlp) {
838 			orig_resid = 0;
839 			controlp = &(*controlp)->m_next;
840 		}
841 	}
842 
843 	/* If m is non-NULL, we have some data to read. */
844 	if (m) {
845 		type = m->m_type;
846 		if (type == MT_OOBDATA)
847 			flags |= MSG_OOB;
848 		if (m->m_flags & M_BCAST)
849 			flags |= MSG_BCAST;
850 		if (m->m_flags & M_MCAST)
851 			flags |= MSG_MCAST;
852 	}
853 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
854 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
855 
856 	moff = 0;
857 	offset = 0;
858 	while (m && uio->uio_resid > 0 && error == 0) {
859 		if (m->m_type == MT_OOBDATA) {
860 			if (type != MT_OOBDATA)
861 				break;
862 		} else if (type == MT_OOBDATA)
863 			break;
864 #ifdef DIAGNOSTIC
865 		else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
866 			panic("receive 3: so %p, so_type %d, m %p, m_type %d",
867 			    so, so->so_type, m, m->m_type);
868 #endif
869 		so->so_state &= ~SS_RCVATMARK;
870 		len = uio->uio_resid;
871 		if (so->so_oobmark && len > so->so_oobmark - offset)
872 			len = so->so_oobmark - offset;
873 		if (len > m->m_len - moff)
874 			len = m->m_len - moff;
875 		/*
876 		 * If mp is set, just pass back the mbufs.
877 		 * Otherwise copy them out via the uio, then free.
878 		 * Sockbuf must be consistent here (points to current mbuf,
879 		 * it points to next record) when we drop priority;
880 		 * we must note any additions to the sockbuf when we
881 		 * block interrupts again.
882 		 */
883 		if (mp == NULL && uio_error == 0) {
884 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
885 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
886 			resid = uio->uio_resid;
887 			sounlock(s);
888 			uio_error = uiomove(mtod(m, caddr_t) + moff, len, uio);
889 			s = solock(so);
890 			if (uio_error)
891 				uio->uio_resid = resid - len;
892 		} else
893 			uio->uio_resid -= len;
894 		if (len == m->m_len - moff) {
895 			if (m->m_flags & M_EOR)
896 				flags |= MSG_EOR;
897 			if (flags & MSG_PEEK) {
898 				m = m->m_next;
899 				moff = 0;
900 			} else {
901 				nextrecord = m->m_nextpkt;
902 				sbfree(&so->so_rcv, m);
903 				if (mp) {
904 					*mp = m;
905 					mp = &m->m_next;
906 					so->so_rcv.sb_mb = m = m->m_next;
907 					*mp = NULL;
908 				} else {
909 					so->so_rcv.sb_mb = m_free(m);
910 					m = so->so_rcv.sb_mb;
911 				}
912 				/*
913 				 * If m != NULL, we also know that
914 				 * so->so_rcv.sb_mb != NULL.
915 				 */
916 				KASSERT(so->so_rcv.sb_mb == m);
917 				if (m) {
918 					m->m_nextpkt = nextrecord;
919 					if (nextrecord == NULL)
920 						so->so_rcv.sb_lastrecord = m;
921 				} else {
922 					so->so_rcv.sb_mb = nextrecord;
923 					SB_EMPTY_FIXUP(&so->so_rcv);
924 				}
925 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
926 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
927 			}
928 		} else {
929 			if (flags & MSG_PEEK)
930 				moff += len;
931 			else {
932 				if (mp)
933 					*mp = m_copym(m, 0, len, M_WAIT);
934 				m->m_data += len;
935 				m->m_len -= len;
936 				so->so_rcv.sb_cc -= len;
937 				so->so_rcv.sb_datacc -= len;
938 			}
939 		}
940 		if (so->so_oobmark) {
941 			if ((flags & MSG_PEEK) == 0) {
942 				so->so_oobmark -= len;
943 				if (so->so_oobmark == 0) {
944 					so->so_state |= SS_RCVATMARK;
945 					break;
946 				}
947 			} else {
948 				offset += len;
949 				if (offset == so->so_oobmark)
950 					break;
951 			}
952 		}
953 		if (flags & MSG_EOR)
954 			break;
955 		/*
956 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
957 		 * we must not quit until "uio->uio_resid == 0" or an error
958 		 * termination.  If a signal/timeout occurs, return
959 		 * with a short count but without error.
960 		 * Keep sockbuf locked against other readers.
961 		 */
962 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
963 		    !sosendallatonce(so) && !nextrecord) {
964 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
965 				break;
966 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
967 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
968 			error = sbwait(so, &so->so_rcv);
969 			if (error) {
970 				sbunlock(so, &so->so_rcv);
971 				sounlock(s);
972 				return (0);
973 			}
974 			if ((m = so->so_rcv.sb_mb) != NULL)
975 				nextrecord = m->m_nextpkt;
976 		}
977 	}
978 
979 	if (m && pr->pr_flags & PR_ATOMIC) {
980 		flags |= MSG_TRUNC;
981 		if ((flags & MSG_PEEK) == 0)
982 			(void) sbdroprecord(&so->so_rcv);
983 	}
984 	if ((flags & MSG_PEEK) == 0) {
985 		if (m == NULL) {
986 			/*
987 			 * First part is an inline SB_EMPTY_FIXUP().  Second
988 			 * part makes sure sb_lastrecord is up-to-date if
989 			 * there is still data in the socket buffer.
990 			 */
991 			so->so_rcv.sb_mb = nextrecord;
992 			if (so->so_rcv.sb_mb == NULL) {
993 				so->so_rcv.sb_mbtail = NULL;
994 				so->so_rcv.sb_lastrecord = NULL;
995 			} else if (nextrecord->m_nextpkt == NULL)
996 				so->so_rcv.sb_lastrecord = nextrecord;
997 		}
998 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
999 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1000 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1001 			(*pr->pr_usrreq)(so, PRU_RCVD, NULL,
1002 			    (struct mbuf *)(long)flags, NULL, curproc);
1003 	}
1004 	if (orig_resid == uio->uio_resid && orig_resid &&
1005 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1006 		sbunlock(so, &so->so_rcv);
1007 		goto restart;
1008 	}
1009 
1010 	if (uio_error)
1011 		error = uio_error;
1012 
1013 	if (flagsp)
1014 		*flagsp |= flags;
1015 release:
1016 	sbunlock(so, &so->so_rcv);
1017 	sounlock(s);
1018 	return (error);
1019 }
1020 
1021 int
1022 soshutdown(struct socket *so, int how)
1023 {
1024 	const struct protosw *pr = so->so_proto;
1025 	int s, error = 0;
1026 
1027 	s = solock(so);
1028 	switch (how) {
1029 	case SHUT_RD:
1030 		sorflush(so);
1031 		break;
1032 	case SHUT_RDWR:
1033 		sorflush(so);
1034 		/* FALLTHROUGH */
1035 	case SHUT_WR:
1036 		error = (*pr->pr_usrreq)(so, PRU_SHUTDOWN, NULL, NULL, NULL,
1037 		    curproc);
1038 		break;
1039 	default:
1040 		error = EINVAL;
1041 		break;
1042 	}
1043 	sounlock(s);
1044 
1045 	return (error);
1046 }
1047 
1048 void
1049 sorflush(struct socket *so)
1050 {
1051 	struct sockbuf *sb = &so->so_rcv;
1052 	const struct protosw *pr = so->so_proto;
1053 	struct socket aso;
1054 	int error;
1055 
1056 	sb->sb_flags |= SB_NOINTR;
1057 	error = sblock(so, sb, M_WAITOK);
1058 	/* with SB_NOINTR and M_WAITOK sblock() must not fail */
1059 	KASSERT(error == 0);
1060 	socantrcvmore(so);
1061 	sbunlock(so, sb);
1062 	aso.so_proto = pr;
1063 	aso.so_rcv = *sb;
1064 	memset(&sb->sb_startzero, 0,
1065 	     (caddr_t)&sb->sb_endzero - (caddr_t)&sb->sb_startzero);
1066 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose)
1067 		(*pr->pr_domain->dom_dispose)(aso.so_rcv.sb_mb);
1068 	sbrelease(&aso, &aso.so_rcv);
1069 }
1070 
1071 #ifdef SOCKET_SPLICE
1072 
1073 #define so_splicelen	so_sp->ssp_len
1074 #define so_splicemax	so_sp->ssp_max
1075 #define so_idletv	so_sp->ssp_idletv
1076 #define so_idleto	so_sp->ssp_idleto
1077 #define so_splicetask	so_sp->ssp_task
1078 
1079 int
1080 sosplice(struct socket *so, int fd, off_t max, struct timeval *tv)
1081 {
1082 	struct file	*fp;
1083 	struct socket	*sosp;
1084 	struct sosplice	*sp;
1085 	int		 error = 0;
1086 
1087 	soassertlocked(so);
1088 
1089 	if (sosplice_taskq == NULL)
1090 		sosplice_taskq = taskq_create("sosplice", 1, IPL_SOFTNET,
1091 		    TASKQ_MPSAFE);
1092 	if (sosplice_taskq == NULL)
1093 		return (ENOMEM);
1094 
1095 	if ((so->so_proto->pr_flags & PR_SPLICE) == 0)
1096 		return (EPROTONOSUPPORT);
1097 	if (so->so_options & SO_ACCEPTCONN)
1098 		return (EOPNOTSUPP);
1099 	if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1100 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
1101 		return (ENOTCONN);
1102 	if (so->so_sp == NULL) {
1103 		sp = pool_get(&sosplice_pool, PR_WAITOK | PR_ZERO);
1104 		if (so->so_sp == NULL)
1105 			so->so_sp = sp;
1106 		else
1107 			pool_put(&sosplice_pool, sp);
1108 	}
1109 
1110 	/* If no fd is given, unsplice by removing existing link. */
1111 	if (fd < 0) {
1112 		/* Lock receive buffer. */
1113 		if ((error = sblock(so, &so->so_rcv,
1114 		    (so->so_state & SS_NBIO) ? M_NOWAIT : M_WAITOK)) != 0) {
1115 			return (error);
1116 		}
1117 		if (so->so_sp->ssp_socket)
1118 			sounsplice(so, so->so_sp->ssp_socket, 1);
1119 		sbunlock(so, &so->so_rcv);
1120 		return (0);
1121 	}
1122 
1123 	if (max && max < 0)
1124 		return (EINVAL);
1125 
1126 	if (tv && (tv->tv_sec < 0 || tv->tv_usec < 0))
1127 		return (EINVAL);
1128 
1129 	/* Find sosp, the drain socket where data will be spliced into. */
1130 	if ((error = getsock(curproc, fd, &fp)) != 0)
1131 		return (error);
1132 	sosp = fp->f_data;
1133 	if (sosp->so_sp == NULL) {
1134 		sp = pool_get(&sosplice_pool, PR_WAITOK | PR_ZERO);
1135 		if (sosp->so_sp == NULL)
1136 			sosp->so_sp = sp;
1137 		else
1138 			pool_put(&sosplice_pool, sp);
1139 	}
1140 
1141 	/* Lock both receive and send buffer. */
1142 	if ((error = sblock(so, &so->so_rcv,
1143 	    (so->so_state & SS_NBIO) ? M_NOWAIT : M_WAITOK)) != 0) {
1144 		goto frele;
1145 	}
1146 	if ((error = sblock(so, &sosp->so_snd, M_WAITOK)) != 0) {
1147 		sbunlock(so, &so->so_rcv);
1148 		goto frele;
1149 	}
1150 
1151 	if (so->so_sp->ssp_socket || sosp->so_sp->ssp_soback) {
1152 		error = EBUSY;
1153 		goto release;
1154 	}
1155 	if (sosp->so_proto->pr_usrreq != so->so_proto->pr_usrreq) {
1156 		error = EPROTONOSUPPORT;
1157 		goto release;
1158 	}
1159 	if (sosp->so_options & SO_ACCEPTCONN) {
1160 		error = EOPNOTSUPP;
1161 		goto release;
1162 	}
1163 	if ((sosp->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0) {
1164 		error = ENOTCONN;
1165 		goto release;
1166 	}
1167 
1168 	/* Splice so and sosp together. */
1169 	so->so_sp->ssp_socket = sosp;
1170 	sosp->so_sp->ssp_soback = so;
1171 	so->so_splicelen = 0;
1172 	so->so_splicemax = max;
1173 	if (tv)
1174 		so->so_idletv = *tv;
1175 	else
1176 		timerclear(&so->so_idletv);
1177 	timeout_set_proc(&so->so_idleto, soidle, so);
1178 	task_set(&so->so_splicetask, sotask, so);
1179 
1180 	/*
1181 	 * To prevent softnet interrupt from calling somove() while
1182 	 * we sleep, the socket buffers are not marked as spliced yet.
1183 	 */
1184 	if (somove(so, M_WAIT)) {
1185 		so->so_rcv.sb_flags |= SB_SPLICE;
1186 		sosp->so_snd.sb_flags |= SB_SPLICE;
1187 	}
1188 
1189  release:
1190 	sbunlock(sosp, &sosp->so_snd);
1191 	sbunlock(so, &so->so_rcv);
1192  frele:
1193 	FRELE(fp, curproc);
1194 	return (error);
1195 }
1196 
1197 void
1198 sounsplice(struct socket *so, struct socket *sosp, int wakeup)
1199 {
1200 	soassertlocked(so);
1201 
1202 	task_del(sosplice_taskq, &so->so_splicetask);
1203 	timeout_del(&so->so_idleto);
1204 	sosp->so_snd.sb_flags &= ~SB_SPLICE;
1205 	so->so_rcv.sb_flags &= ~SB_SPLICE;
1206 	so->so_sp->ssp_socket = sosp->so_sp->ssp_soback = NULL;
1207 	if (wakeup && soreadable(so))
1208 		sorwakeup(so);
1209 }
1210 
1211 void
1212 soidle(void *arg)
1213 {
1214 	struct socket *so = arg;
1215 	int s;
1216 
1217 	s = solock(so);
1218 	if (so->so_rcv.sb_flags & SB_SPLICE) {
1219 		so->so_error = ETIMEDOUT;
1220 		sounsplice(so, so->so_sp->ssp_socket, 1);
1221 	}
1222 	sounlock(s);
1223 }
1224 
1225 void
1226 sotask(void *arg)
1227 {
1228 	struct socket *so = arg;
1229 	int s;
1230 
1231 	s = solock(so);
1232 	if (so->so_rcv.sb_flags & SB_SPLICE) {
1233 		/*
1234 		 * We may not sleep here as sofree() and unsplice() may be
1235 		 * called from softnet interrupt context.  This would remove
1236 		 * the socket during somove().
1237 		 */
1238 		somove(so, M_DONTWAIT);
1239 	}
1240 	sounlock(s);
1241 
1242 	/* Avoid user land starvation. */
1243 	yield();
1244 }
1245 
1246 /*
1247  * The socket splicing task may sleep while grabbing the net lock.  As sofree()
1248  * can be called anytime, sotask() can access the socket memory of a freed
1249  * socket after wakeup.  So delay the pool_put() after all pending socket
1250  * splicing tasks have finished.  Do this by scheduling it on the same thread.
1251  */
1252 void
1253 soput(void *arg)
1254 {
1255 	struct socket *so = arg;
1256 
1257 	pool_put(&sosplice_pool, so->so_sp);
1258 	pool_put(&socket_pool, so);
1259 }
1260 
1261 /*
1262  * Move data from receive buffer of spliced source socket to send
1263  * buffer of drain socket.  Try to move as much as possible in one
1264  * big chunk.  It is a TCP only implementation.
1265  * Return value 0 means splicing has been finished, 1 continue.
1266  */
1267 int
1268 somove(struct socket *so, int wait)
1269 {
1270 	struct socket	*sosp = so->so_sp->ssp_socket;
1271 	struct mbuf	*m, **mp, *nextrecord;
1272 	u_long		 len, off, oobmark;
1273 	long		 space;
1274 	int		 error = 0, maxreached = 0;
1275 	unsigned int	 state;
1276 
1277 	soassertlocked(so);
1278 
1279  nextpkt:
1280 	if (so->so_error) {
1281 		error = so->so_error;
1282 		goto release;
1283 	}
1284 	if (sosp->so_state & SS_CANTSENDMORE) {
1285 		error = EPIPE;
1286 		goto release;
1287 	}
1288 	if (sosp->so_error && sosp->so_error != ETIMEDOUT &&
1289 	    sosp->so_error != EFBIG && sosp->so_error != ELOOP) {
1290 		error = sosp->so_error;
1291 		goto release;
1292 	}
1293 	if ((sosp->so_state & SS_ISCONNECTED) == 0)
1294 		goto release;
1295 
1296 	/* Calculate how many bytes can be copied now. */
1297 	len = so->so_rcv.sb_datacc;
1298 	if (so->so_splicemax) {
1299 		KASSERT(so->so_splicelen < so->so_splicemax);
1300 		if (so->so_splicemax <= so->so_splicelen + len) {
1301 			len = so->so_splicemax - so->so_splicelen;
1302 			maxreached = 1;
1303 		}
1304 	}
1305 	space = sbspace(sosp, &sosp->so_snd);
1306 	if (so->so_oobmark && so->so_oobmark < len &&
1307 	    so->so_oobmark < space + 1024)
1308 		space += 1024;
1309 	if (space <= 0) {
1310 		maxreached = 0;
1311 		goto release;
1312 	}
1313 	if (space < len) {
1314 		maxreached = 0;
1315 		if (space < sosp->so_snd.sb_lowat)
1316 			goto release;
1317 		len = space;
1318 	}
1319 	sosp->so_state |= SS_ISSENDING;
1320 
1321 	SBLASTRECORDCHK(&so->so_rcv, "somove 1");
1322 	SBLASTMBUFCHK(&so->so_rcv, "somove 1");
1323 	m = so->so_rcv.sb_mb;
1324 	if (m == NULL)
1325 		goto release;
1326 	nextrecord = m->m_nextpkt;
1327 
1328 	/* Drop address and control information not used with splicing. */
1329 	if (so->so_proto->pr_flags & PR_ADDR) {
1330 #ifdef DIAGNOSTIC
1331 		if (m->m_type != MT_SONAME)
1332 			panic("somove soname: so %p, so_type %d, m %p, "
1333 			    "m_type %d", so, so->so_type, m, m->m_type);
1334 #endif
1335 		m = m->m_next;
1336 	}
1337 	while (m && m->m_type == MT_CONTROL)
1338 		m = m->m_next;
1339 	if (m == NULL) {
1340 		sbdroprecord(&so->so_rcv);
1341 		if (so->so_proto->pr_flags & PR_WANTRCVD && so->so_pcb)
1342 			(so->so_proto->pr_usrreq)(so, PRU_RCVD, NULL,
1343 			    NULL, NULL, NULL);
1344 		goto nextpkt;
1345 	}
1346 
1347 	/*
1348 	 * By splicing sockets connected to localhost, userland might create a
1349 	 * loop.  Dissolve splicing with error if loop is detected by counter.
1350 	 */
1351 	if ((m->m_flags & M_PKTHDR) && m->m_pkthdr.ph_loopcnt++ >= M_MAXLOOP) {
1352 		error = ELOOP;
1353 		goto release;
1354 	}
1355 
1356 	if (so->so_proto->pr_flags & PR_ATOMIC) {
1357 		if ((m->m_flags & M_PKTHDR) == 0)
1358 			panic("somove !PKTHDR: so %p, so_type %d, m %p, "
1359 			    "m_type %d", so, so->so_type, m, m->m_type);
1360 		if (sosp->so_snd.sb_hiwat < m->m_pkthdr.len) {
1361 			error = EMSGSIZE;
1362 			goto release;
1363 		}
1364 		if (len < m->m_pkthdr.len)
1365 			goto release;
1366 		if (m->m_pkthdr.len < len) {
1367 			maxreached = 0;
1368 			len = m->m_pkthdr.len;
1369 		}
1370 		/*
1371 		 * Throw away the name mbuf after it has been assured
1372 		 * that the whole first record can be processed.
1373 		 */
1374 		m = so->so_rcv.sb_mb;
1375 		sbfree(&so->so_rcv, m);
1376 		so->so_rcv.sb_mb = m_free(m);
1377 		sbsync(&so->so_rcv, nextrecord);
1378 	}
1379 	/*
1380 	 * Throw away the control mbufs after it has been assured
1381 	 * that the whole first record can be processed.
1382 	 */
1383 	m = so->so_rcv.sb_mb;
1384 	while (m && m->m_type == MT_CONTROL) {
1385 		sbfree(&so->so_rcv, m);
1386 		so->so_rcv.sb_mb = m_free(m);
1387 		m = so->so_rcv.sb_mb;
1388 		sbsync(&so->so_rcv, nextrecord);
1389 	}
1390 
1391 	SBLASTRECORDCHK(&so->so_rcv, "somove 2");
1392 	SBLASTMBUFCHK(&so->so_rcv, "somove 2");
1393 
1394 	/* Take at most len mbufs out of receive buffer. */
1395 	for (off = 0, mp = &m; off <= len && *mp;
1396 	    off += (*mp)->m_len, mp = &(*mp)->m_next) {
1397 		u_long size = len - off;
1398 
1399 #ifdef DIAGNOSTIC
1400 		if ((*mp)->m_type != MT_DATA && (*mp)->m_type != MT_HEADER)
1401 			panic("somove type: so %p, so_type %d, m %p, "
1402 			    "m_type %d", so, so->so_type, *mp, (*mp)->m_type);
1403 #endif
1404 		if ((*mp)->m_len > size) {
1405 			/*
1406 			 * Move only a partial mbuf at maximum splice length or
1407 			 * if the drain buffer is too small for this large mbuf.
1408 			 */
1409 			if (!maxreached && so->so_snd.sb_datacc > 0) {
1410 				len -= size;
1411 				break;
1412 			}
1413 			*mp = m_copym(so->so_rcv.sb_mb, 0, size, wait);
1414 			if (*mp == NULL) {
1415 				len -= size;
1416 				break;
1417 			}
1418 			so->so_rcv.sb_mb->m_data += size;
1419 			so->so_rcv.sb_mb->m_len -= size;
1420 			so->so_rcv.sb_cc -= size;
1421 			so->so_rcv.sb_datacc -= size;
1422 		} else {
1423 			*mp = so->so_rcv.sb_mb;
1424 			sbfree(&so->so_rcv, *mp);
1425 			so->so_rcv.sb_mb = (*mp)->m_next;
1426 			sbsync(&so->so_rcv, nextrecord);
1427 		}
1428 	}
1429 	*mp = NULL;
1430 
1431 	SBLASTRECORDCHK(&so->so_rcv, "somove 3");
1432 	SBLASTMBUFCHK(&so->so_rcv, "somove 3");
1433 	SBCHECK(&so->so_rcv);
1434 	if (m == NULL)
1435 		goto release;
1436 	m->m_nextpkt = NULL;
1437 	if (m->m_flags & M_PKTHDR) {
1438 		m_resethdr(m);
1439 		m->m_pkthdr.len = len;
1440 	}
1441 
1442 	/* Send window update to source peer as receive buffer has changed. */
1443 	if (so->so_proto->pr_flags & PR_WANTRCVD && so->so_pcb)
1444 		(so->so_proto->pr_usrreq)(so, PRU_RCVD, NULL,
1445 		    NULL, NULL, NULL);
1446 
1447 	/* Receive buffer did shrink by len bytes, adjust oob. */
1448 	state = so->so_state;
1449 	so->so_state &= ~SS_RCVATMARK;
1450 	oobmark = so->so_oobmark;
1451 	so->so_oobmark = oobmark > len ? oobmark - len : 0;
1452 	if (oobmark) {
1453 		if (oobmark == len)
1454 			so->so_state |= SS_RCVATMARK;
1455 		if (oobmark >= len)
1456 			oobmark = 0;
1457 	}
1458 
1459 	/*
1460 	 * Handle oob data.  If any malloc fails, ignore error.
1461 	 * TCP urgent data is not very reliable anyway.
1462 	 */
1463 	while (((state & SS_RCVATMARK) || oobmark) &&
1464 	    (so->so_options & SO_OOBINLINE)) {
1465 		struct mbuf *o = NULL;
1466 
1467 		if (state & SS_RCVATMARK) {
1468 			o = m_get(wait, MT_DATA);
1469 			state &= ~SS_RCVATMARK;
1470 		} else if (oobmark) {
1471 			o = m_split(m, oobmark, wait);
1472 			if (o) {
1473 				error = (*sosp->so_proto->pr_usrreq)(sosp,
1474 				    PRU_SEND, m, NULL, NULL, NULL);
1475 				if (error) {
1476 					if (sosp->so_state & SS_CANTSENDMORE)
1477 						error = EPIPE;
1478 					m_freem(o);
1479 					goto release;
1480 				}
1481 				len -= oobmark;
1482 				so->so_splicelen += oobmark;
1483 				m = o;
1484 				o = m_get(wait, MT_DATA);
1485 			}
1486 			oobmark = 0;
1487 		}
1488 		if (o) {
1489 			o->m_len = 1;
1490 			*mtod(o, caddr_t) = *mtod(m, caddr_t);
1491 			error = (*sosp->so_proto->pr_usrreq)(sosp, PRU_SENDOOB,
1492 			    o, NULL, NULL, NULL);
1493 			if (error) {
1494 				if (sosp->so_state & SS_CANTSENDMORE)
1495 					error = EPIPE;
1496 				m_freem(m);
1497 				goto release;
1498 			}
1499 			len -= 1;
1500 			so->so_splicelen += 1;
1501 			if (oobmark) {
1502 				oobmark -= 1;
1503 				if (oobmark == 0)
1504 					state |= SS_RCVATMARK;
1505 			}
1506 			m_adj(m, 1);
1507 		}
1508 	}
1509 
1510 	/* Append all remaining data to drain socket. */
1511 	if (so->so_rcv.sb_cc == 0 || maxreached)
1512 		sosp->so_state &= ~SS_ISSENDING;
1513 	error = (*sosp->so_proto->pr_usrreq)(sosp, PRU_SEND, m, NULL, NULL,
1514 	    NULL);
1515 	if (error) {
1516 		if (sosp->so_state & SS_CANTSENDMORE)
1517 			error = EPIPE;
1518 		goto release;
1519 	}
1520 	so->so_splicelen += len;
1521 
1522 	/* Move several packets if possible. */
1523 	if (!maxreached && nextrecord)
1524 		goto nextpkt;
1525 
1526  release:
1527 	sosp->so_state &= ~SS_ISSENDING;
1528 	if (!error && maxreached && so->so_splicemax == so->so_splicelen)
1529 		error = EFBIG;
1530 	if (error)
1531 		so->so_error = error;
1532 	if (((so->so_state & SS_CANTRCVMORE) && so->so_rcv.sb_cc == 0) ||
1533 	    (sosp->so_state & SS_CANTSENDMORE) || maxreached || error) {
1534 		sounsplice(so, sosp, 1);
1535 		return (0);
1536 	}
1537 	if (timerisset(&so->so_idletv))
1538 		timeout_add_tv(&so->so_idleto, &so->so_idletv);
1539 	return (1);
1540 }
1541 
1542 #endif /* SOCKET_SPLICE */
1543 
1544 void
1545 sorwakeup(struct socket *so)
1546 {
1547 	soassertlocked(so);
1548 
1549 #ifdef SOCKET_SPLICE
1550 	if (so->so_rcv.sb_flags & SB_SPLICE) {
1551 		/*
1552 		 * TCP has a sendbuffer that can handle multiple packets
1553 		 * at once.  So queue the stream a bit to accumulate data.
1554 		 * The sosplice thread will call somove() later and send
1555 		 * the packets calling tcp_output() only once.
1556 		 * In the UDP case, send out the packets immediately.
1557 		 * Using a thread would make things slower.
1558 		 */
1559 		if (so->so_proto->pr_flags & PR_WANTRCVD)
1560 			task_add(sosplice_taskq, &so->so_splicetask);
1561 		else
1562 			somove(so, M_DONTWAIT);
1563 	}
1564 	if (isspliced(so))
1565 		return;
1566 #endif
1567 	sowakeup(so, &so->so_rcv);
1568 	if (so->so_upcall)
1569 		(*(so->so_upcall))(so, so->so_upcallarg, M_DONTWAIT);
1570 }
1571 
1572 void
1573 sowwakeup(struct socket *so)
1574 {
1575 	soassertlocked(so);
1576 
1577 #ifdef SOCKET_SPLICE
1578 	if (so->so_snd.sb_flags & SB_SPLICE)
1579 		task_add(sosplice_taskq, &so->so_sp->ssp_soback->so_splicetask);
1580 #endif
1581 	sowakeup(so, &so->so_snd);
1582 }
1583 
1584 int
1585 sosetopt(struct socket *so, int level, int optname, struct mbuf *m)
1586 {
1587 	int error = 0;
1588 
1589 	soassertlocked(so);
1590 
1591 	if (level != SOL_SOCKET) {
1592 		if (so->so_proto->pr_ctloutput) {
1593 			error = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so,
1594 			    level, optname, m);
1595 			return (error);
1596 		}
1597 		error = ENOPROTOOPT;
1598 	} else {
1599 		switch (optname) {
1600 		case SO_BINDANY:
1601 			if ((error = suser(curproc)) != 0)	/* XXX */
1602 				return (error);
1603 			break;
1604 		}
1605 
1606 		switch (optname) {
1607 
1608 		case SO_LINGER:
1609 			if (m == NULL || m->m_len != sizeof (struct linger) ||
1610 			    mtod(m, struct linger *)->l_linger < 0 ||
1611 			    mtod(m, struct linger *)->l_linger > SHRT_MAX)
1612 				return (EINVAL);
1613 			so->so_linger = mtod(m, struct linger *)->l_linger;
1614 			/* FALLTHROUGH */
1615 
1616 		case SO_BINDANY:
1617 		case SO_DEBUG:
1618 		case SO_KEEPALIVE:
1619 		case SO_USELOOPBACK:
1620 		case SO_BROADCAST:
1621 		case SO_REUSEADDR:
1622 		case SO_REUSEPORT:
1623 		case SO_OOBINLINE:
1624 		case SO_TIMESTAMP:
1625 		case SO_ZEROIZE:
1626 			if (m == NULL || m->m_len < sizeof (int))
1627 				return (EINVAL);
1628 			if (*mtod(m, int *))
1629 				so->so_options |= optname;
1630 			else
1631 				so->so_options &= ~optname;
1632 			break;
1633 
1634 		case SO_DONTROUTE:
1635 			if (m == NULL || m->m_len < sizeof (int))
1636 				return (EINVAL);
1637 			if (*mtod(m, int *))
1638 				error = EOPNOTSUPP;
1639 			break;
1640 
1641 		case SO_SNDBUF:
1642 		case SO_RCVBUF:
1643 		case SO_SNDLOWAT:
1644 		case SO_RCVLOWAT:
1645 		    {
1646 			u_long cnt;
1647 
1648 			if (m == NULL || m->m_len < sizeof (int))
1649 				return (EINVAL);
1650 			cnt = *mtod(m, int *);
1651 			if ((long)cnt <= 0)
1652 				cnt = 1;
1653 			switch (optname) {
1654 
1655 			case SO_SNDBUF:
1656 				if (so->so_state & SS_CANTSENDMORE)
1657 					return (EINVAL);
1658 				if (sbcheckreserve(cnt, so->so_snd.sb_wat) ||
1659 				    sbreserve(so, &so->so_snd, cnt))
1660 					return (ENOBUFS);
1661 				so->so_snd.sb_wat = cnt;
1662 				break;
1663 
1664 			case SO_RCVBUF:
1665 				if (so->so_state & SS_CANTRCVMORE)
1666 					return (EINVAL);
1667 				if (sbcheckreserve(cnt, so->so_rcv.sb_wat) ||
1668 				    sbreserve(so, &so->so_rcv, cnt))
1669 					return (ENOBUFS);
1670 				so->so_rcv.sb_wat = cnt;
1671 				break;
1672 
1673 			case SO_SNDLOWAT:
1674 				so->so_snd.sb_lowat =
1675 				    (cnt > so->so_snd.sb_hiwat) ?
1676 				    so->so_snd.sb_hiwat : cnt;
1677 				break;
1678 			case SO_RCVLOWAT:
1679 				so->so_rcv.sb_lowat =
1680 				    (cnt > so->so_rcv.sb_hiwat) ?
1681 				    so->so_rcv.sb_hiwat : cnt;
1682 				break;
1683 			}
1684 			break;
1685 		    }
1686 
1687 		case SO_SNDTIMEO:
1688 		case SO_RCVTIMEO:
1689 		    {
1690 			struct timeval tv;
1691 			int val;
1692 
1693 			if (m == NULL || m->m_len < sizeof (tv))
1694 				return (EINVAL);
1695 			memcpy(&tv, mtod(m, struct timeval *), sizeof tv);
1696 			val = tvtohz(&tv);
1697 			if (val > USHRT_MAX)
1698 				return (EDOM);
1699 
1700 			switch (optname) {
1701 
1702 			case SO_SNDTIMEO:
1703 				so->so_snd.sb_timeo = val;
1704 				break;
1705 			case SO_RCVTIMEO:
1706 				so->so_rcv.sb_timeo = val;
1707 				break;
1708 			}
1709 			break;
1710 		    }
1711 
1712 		case SO_RTABLE:
1713 			if (so->so_proto->pr_domain &&
1714 			    so->so_proto->pr_domain->dom_protosw &&
1715 			    so->so_proto->pr_ctloutput) {
1716 				struct domain *dom = so->so_proto->pr_domain;
1717 
1718 				level = dom->dom_protosw->pr_protocol;
1719 				error = (*so->so_proto->pr_ctloutput)
1720 				    (PRCO_SETOPT, so, level, optname, m);
1721 				return (error);
1722 			}
1723 			error = ENOPROTOOPT;
1724 			break;
1725 
1726 #ifdef SOCKET_SPLICE
1727 		case SO_SPLICE:
1728 			if (m == NULL) {
1729 				error = sosplice(so, -1, 0, NULL);
1730 			} else if (m->m_len < sizeof(int)) {
1731 				return (EINVAL);
1732 			} else if (m->m_len < sizeof(struct splice)) {
1733 				error = sosplice(so, *mtod(m, int *), 0, NULL);
1734 			} else {
1735 				error = sosplice(so,
1736 				    mtod(m, struct splice *)->sp_fd,
1737 				    mtod(m, struct splice *)->sp_max,
1738 				   &mtod(m, struct splice *)->sp_idle);
1739 			}
1740 			break;
1741 #endif /* SOCKET_SPLICE */
1742 
1743 		default:
1744 			error = ENOPROTOOPT;
1745 			break;
1746 		}
1747 		if (error == 0 && so->so_proto->pr_ctloutput) {
1748 			(*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so,
1749 			    level, optname, m);
1750 		}
1751 	}
1752 
1753 	return (error);
1754 }
1755 
1756 int
1757 sogetopt(struct socket *so, int level, int optname, struct mbuf *m)
1758 {
1759 	int error = 0;
1760 
1761 	soassertlocked(so);
1762 
1763 	if (level != SOL_SOCKET) {
1764 		if (so->so_proto->pr_ctloutput) {
1765 			m->m_len = 0;
1766 
1767 			error = (*so->so_proto->pr_ctloutput)(PRCO_GETOPT, so,
1768 			    level, optname, m);
1769 			if (error)
1770 				return (error);
1771 			return (0);
1772 		} else
1773 			return (ENOPROTOOPT);
1774 	} else {
1775 		m->m_len = sizeof (int);
1776 
1777 		switch (optname) {
1778 
1779 		case SO_LINGER:
1780 			m->m_len = sizeof (struct linger);
1781 			mtod(m, struct linger *)->l_onoff =
1782 				so->so_options & SO_LINGER;
1783 			mtod(m, struct linger *)->l_linger = so->so_linger;
1784 			break;
1785 
1786 		case SO_BINDANY:
1787 		case SO_USELOOPBACK:
1788 		case SO_DEBUG:
1789 		case SO_KEEPALIVE:
1790 		case SO_REUSEADDR:
1791 		case SO_REUSEPORT:
1792 		case SO_BROADCAST:
1793 		case SO_OOBINLINE:
1794 		case SO_TIMESTAMP:
1795 		case SO_ZEROIZE:
1796 			*mtod(m, int *) = so->so_options & optname;
1797 			break;
1798 
1799 		case SO_DONTROUTE:
1800 			*mtod(m, int *) = 0;
1801 			break;
1802 
1803 		case SO_TYPE:
1804 			*mtod(m, int *) = so->so_type;
1805 			break;
1806 
1807 		case SO_ERROR:
1808 			*mtod(m, int *) = so->so_error;
1809 			so->so_error = 0;
1810 			break;
1811 
1812 		case SO_SNDBUF:
1813 			*mtod(m, int *) = so->so_snd.sb_hiwat;
1814 			break;
1815 
1816 		case SO_RCVBUF:
1817 			*mtod(m, int *) = so->so_rcv.sb_hiwat;
1818 			break;
1819 
1820 		case SO_SNDLOWAT:
1821 			*mtod(m, int *) = so->so_snd.sb_lowat;
1822 			break;
1823 
1824 		case SO_RCVLOWAT:
1825 			*mtod(m, int *) = so->so_rcv.sb_lowat;
1826 			break;
1827 
1828 		case SO_SNDTIMEO:
1829 		case SO_RCVTIMEO:
1830 		    {
1831 			struct timeval tv;
1832 			int val = (optname == SO_SNDTIMEO ?
1833 			    so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1834 
1835 			m->m_len = sizeof(struct timeval);
1836 			memset(&tv, 0, sizeof(tv));
1837 			tv.tv_sec = val / hz;
1838 			tv.tv_usec = (val % hz) * tick;
1839 			memcpy(mtod(m, struct timeval *), &tv, sizeof tv);
1840 			break;
1841 		    }
1842 
1843 		case SO_RTABLE:
1844 			if (so->so_proto->pr_domain &&
1845 			    so->so_proto->pr_domain->dom_protosw &&
1846 			    so->so_proto->pr_ctloutput) {
1847 				struct domain *dom = so->so_proto->pr_domain;
1848 
1849 				level = dom->dom_protosw->pr_protocol;
1850 				error = (*so->so_proto->pr_ctloutput)
1851 				    (PRCO_GETOPT, so, level, optname, m);
1852 				if (error)
1853 					return (error);
1854 				break;
1855 			}
1856 			return (ENOPROTOOPT);
1857 
1858 #ifdef SOCKET_SPLICE
1859 		case SO_SPLICE:
1860 		    {
1861 			off_t len;
1862 
1863 			m->m_len = sizeof(off_t);
1864 			len = so->so_sp ? so->so_sp->ssp_len : 0;
1865 			memcpy(mtod(m, off_t *), &len, sizeof(off_t));
1866 			break;
1867 		    }
1868 #endif /* SOCKET_SPLICE */
1869 
1870 		case SO_PEERCRED:
1871 			if (so->so_proto->pr_protocol == AF_UNIX) {
1872 				struct unpcb *unp = sotounpcb(so);
1873 
1874 				if (unp->unp_flags & UNP_FEIDS) {
1875 					m->m_len = sizeof(unp->unp_connid);
1876 					memcpy(mtod(m, caddr_t),
1877 					    &(unp->unp_connid), m->m_len);
1878 					break;
1879 				}
1880 				return (ENOTCONN);
1881 			}
1882 			return (EOPNOTSUPP);
1883 
1884 		default:
1885 			return (ENOPROTOOPT);
1886 		}
1887 		return (0);
1888 	}
1889 }
1890 
1891 void
1892 sohasoutofband(struct socket *so)
1893 {
1894 	KERNEL_LOCK();
1895 	csignal(so->so_pgid, SIGURG, so->so_siguid, so->so_sigeuid);
1896 	selwakeup(&so->so_rcv.sb_sel);
1897 	KERNEL_UNLOCK();
1898 }
1899 
1900 int
1901 soo_kqfilter(struct file *fp, struct knote *kn)
1902 {
1903 	struct socket *so = kn->kn_fp->f_data;
1904 	struct sockbuf *sb;
1905 
1906 	KERNEL_ASSERT_LOCKED();
1907 
1908 	switch (kn->kn_filter) {
1909 	case EVFILT_READ:
1910 		if (so->so_options & SO_ACCEPTCONN)
1911 			kn->kn_fop = &solisten_filtops;
1912 		else
1913 			kn->kn_fop = &soread_filtops;
1914 		sb = &so->so_rcv;
1915 		break;
1916 	case EVFILT_WRITE:
1917 		kn->kn_fop = &sowrite_filtops;
1918 		sb = &so->so_snd;
1919 		break;
1920 	default:
1921 		return (EINVAL);
1922 	}
1923 
1924 	SLIST_INSERT_HEAD(&sb->sb_sel.si_note, kn, kn_selnext);
1925 	sb->sb_flagsintr |= SB_KNOTE;
1926 
1927 	return (0);
1928 }
1929 
1930 void
1931 filt_sordetach(struct knote *kn)
1932 {
1933 	struct socket *so = kn->kn_fp->f_data;
1934 
1935 	KERNEL_ASSERT_LOCKED();
1936 
1937 	SLIST_REMOVE(&so->so_rcv.sb_sel.si_note, kn, knote, kn_selnext);
1938 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.si_note))
1939 		so->so_rcv.sb_flagsintr &= ~SB_KNOTE;
1940 }
1941 
1942 int
1943 filt_soread(struct knote *kn, long hint)
1944 {
1945 	struct socket *so = kn->kn_fp->f_data;
1946 	int rv;
1947 
1948 	kn->kn_data = so->so_rcv.sb_cc;
1949 #ifdef SOCKET_SPLICE
1950 	if (isspliced(so)) {
1951 		rv = 0;
1952 	} else
1953 #endif /* SOCKET_SPLICE */
1954 	if (so->so_state & SS_CANTRCVMORE) {
1955 		kn->kn_flags |= EV_EOF;
1956 		kn->kn_fflags = so->so_error;
1957 		rv = 1;
1958 	} else if (so->so_error) {	/* temporary udp error */
1959 		rv = 1;
1960 	} else if (kn->kn_sfflags & NOTE_LOWAT) {
1961 		rv = (kn->kn_data >= kn->kn_sdata);
1962 	} else {
1963 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
1964 	}
1965 
1966 	return rv;
1967 }
1968 
1969 void
1970 filt_sowdetach(struct knote *kn)
1971 {
1972 	struct socket *so = kn->kn_fp->f_data;
1973 
1974 	KERNEL_ASSERT_LOCKED();
1975 
1976 	SLIST_REMOVE(&so->so_snd.sb_sel.si_note, kn, knote, kn_selnext);
1977 	if (SLIST_EMPTY(&so->so_snd.sb_sel.si_note))
1978 		so->so_snd.sb_flagsintr &= ~SB_KNOTE;
1979 }
1980 
1981 int
1982 filt_sowrite(struct knote *kn, long hint)
1983 {
1984 	struct socket *so = kn->kn_fp->f_data;
1985 	int rv;
1986 
1987 	kn->kn_data = sbspace(so, &so->so_snd);
1988 	if (so->so_state & SS_CANTSENDMORE) {
1989 		kn->kn_flags |= EV_EOF;
1990 		kn->kn_fflags = so->so_error;
1991 		rv = 1;
1992 	} else if (so->so_error) {	/* temporary udp error */
1993 		rv = 1;
1994 	} else if (((so->so_state & SS_ISCONNECTED) == 0) &&
1995 	    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1996 		rv = 0;
1997 	} else if (kn->kn_sfflags & NOTE_LOWAT) {
1998 		rv = (kn->kn_data >= kn->kn_sdata);
1999 	} else {
2000 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
2001 	}
2002 
2003 	return (rv);
2004 }
2005 
2006 int
2007 filt_solisten(struct knote *kn, long hint)
2008 {
2009 	struct socket *so = kn->kn_fp->f_data;
2010 
2011 	kn->kn_data = so->so_qlen;
2012 
2013 	return (kn->kn_data != 0);
2014 }
2015 
2016 #ifdef DDB
2017 void
2018 sobuf_print(struct sockbuf *,
2019     int (*)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))));
2020 
2021 void
2022 sobuf_print(struct sockbuf *sb,
2023     int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2024 {
2025 	(*pr)("\tsb_cc: %lu\n", sb->sb_cc);
2026 	(*pr)("\tsb_datacc: %lu\n", sb->sb_datacc);
2027 	(*pr)("\tsb_hiwat: %lu\n", sb->sb_hiwat);
2028 	(*pr)("\tsb_wat: %lu\n", sb->sb_wat);
2029 	(*pr)("\tsb_mbcnt: %lu\n", sb->sb_mbcnt);
2030 	(*pr)("\tsb_mbmax: %lu\n", sb->sb_mbmax);
2031 	(*pr)("\tsb_lowat: %ld\n", sb->sb_lowat);
2032 	(*pr)("\tsb_mb: %p\n", sb->sb_mb);
2033 	(*pr)("\tsb_mbtail: %p\n", sb->sb_mbtail);
2034 	(*pr)("\tsb_lastrecord: %p\n", sb->sb_lastrecord);
2035 	(*pr)("\tsb_sel: ...\n");
2036 	(*pr)("\tsb_flagsintr: %d\n", sb->sb_flagsintr);
2037 	(*pr)("\tsb_flags: %i\n", sb->sb_flags);
2038 	(*pr)("\tsb_timeo: %i\n", sb->sb_timeo);
2039 }
2040 
2041 void
2042 so_print(void *v,
2043     int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2044 {
2045 	struct socket *so = v;
2046 
2047 	(*pr)("socket %p\n", so);
2048 	(*pr)("so_type: %i\n", so->so_type);
2049 	(*pr)("so_options: 0x%04x\n", so->so_options); /* %b */
2050 	(*pr)("so_linger: %i\n", so->so_linger);
2051 	(*pr)("so_state: 0x%04x\n", so->so_state);
2052 	(*pr)("so_pcb: %p\n", so->so_pcb);
2053 	(*pr)("so_proto: %p\n", so->so_proto);
2054 
2055 	(*pr)("so_head: %p\n", so->so_head);
2056 	(*pr)("so_onq: %p\n", so->so_onq);
2057 	(*pr)("so_q0: @%p first: %p\n", &so->so_q0, TAILQ_FIRST(&so->so_q0));
2058 	(*pr)("so_q: @%p first: %p\n", &so->so_q, TAILQ_FIRST(&so->so_q));
2059 	(*pr)("so_eq: next: %p\n", TAILQ_NEXT(so, so_qe));
2060 	(*pr)("so_q0len: %i\n", so->so_q0len);
2061 	(*pr)("so_qlen: %i\n", so->so_qlen);
2062 	(*pr)("so_qlimit: %i\n", so->so_qlimit);
2063 	(*pr)("so_timeo: %i\n", so->so_timeo);
2064 	(*pr)("so_pgid: %i\n", so->so_pgid);
2065 	(*pr)("so_siguid: %i\n", so->so_siguid);
2066 	(*pr)("so_sigeuid: %i\n", so->so_sigeuid);
2067 	(*pr)("so_obmark: %lu\n", so->so_oobmark);
2068 
2069 	(*pr)("so_sp: %p\n", so->so_sp);
2070 	if (so->so_sp != NULL) {
2071 		(*pr)("\tssp_socket: %p\n", so->so_sp->ssp_socket);
2072 		(*pr)("\tssp_soback: %p\n", so->so_sp->ssp_soback);
2073 		(*pr)("\tssp_len: %lld\n",
2074 		    (unsigned long long)so->so_sp->ssp_len);
2075 		(*pr)("\tssp_max: %lld\n",
2076 		    (unsigned long long)so->so_sp->ssp_max);
2077 		(*pr)("\tssp_idletv: %lld %ld\n", so->so_sp->ssp_idletv.tv_sec,
2078 		    so->so_sp->ssp_idletv.tv_usec);
2079 		(*pr)("\tssp_idleto: %spending (@%i)\n",
2080 		    timeout_pending(&so->so_sp->ssp_idleto) ? "" : "not ",
2081 		    so->so_sp->ssp_idleto.to_time);
2082 	}
2083 
2084 	(*pr)("so_rcv:\n");
2085 	sobuf_print(&so->so_rcv, pr);
2086 	(*pr)("so_snd:\n");
2087 	sobuf_print(&so->so_snd, pr);
2088 
2089 	(*pr)("so_upcall: %p so_upcallarg: %p\n",
2090 	    so->so_upcall, so->so_upcallarg);
2091 
2092 	(*pr)("so_euid: %d so_ruid: %d\n", so->so_euid, so->so_ruid);
2093 	(*pr)("so_egid: %d so_rgid: %d\n", so->so_egid, so->so_rgid);
2094 	(*pr)("so_cpid: %d\n", so->so_cpid);
2095 }
2096 #endif
2097 
2098