xref: /openbsd-src/sys/kern/uipc_socket.c (revision 99fd087599a8791921855f21bd7e36130f39aadc)
1 /*	$OpenBSD: uipc_socket.c,v 1.241 2020/02/20 16:56:52 visa 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 #include <sys/atomic.h>
53 #include <sys/rwlock.h>
54 
55 #ifdef DDB
56 #include <machine/db_machdep.h>
57 #endif
58 
59 void	sbsync(struct sockbuf *, struct mbuf *);
60 
61 int	sosplice(struct socket *, int, off_t, struct timeval *);
62 void	sounsplice(struct socket *, struct socket *, int);
63 void	soidle(void *);
64 void	sotask(void *);
65 void	soreaper(void *);
66 void	soput(void *);
67 int	somove(struct socket *, int);
68 
69 void	filt_sordetach(struct knote *kn);
70 int	filt_soread(struct knote *kn, long hint);
71 void	filt_sowdetach(struct knote *kn);
72 int	filt_sowrite(struct knote *kn, long hint);
73 int	filt_solisten(struct knote *kn, long hint);
74 
75 const struct filterops solisten_filtops = {
76 	.f_flags	= FILTEROP_ISFD,
77 	.f_attach	= NULL,
78 	.f_detach	= filt_sordetach,
79 	.f_event	= filt_solisten,
80 };
81 
82 const struct filterops soread_filtops = {
83 	.f_flags	= FILTEROP_ISFD,
84 	.f_attach	= NULL,
85 	.f_detach	= filt_sordetach,
86 	.f_event	= filt_soread,
87 };
88 
89 const struct filterops sowrite_filtops = {
90 	.f_flags	= FILTEROP_ISFD,
91 	.f_attach	= NULL,
92 	.f_detach	= filt_sowdetach,
93 	.f_event	= filt_sowrite,
94 };
95 
96 
97 #ifndef SOMINCONN
98 #define SOMINCONN 80
99 #endif /* SOMINCONN */
100 
101 int	somaxconn = SOMAXCONN;
102 int	sominconn = SOMINCONN;
103 
104 struct pool socket_pool;
105 #ifdef SOCKET_SPLICE
106 struct pool sosplice_pool;
107 struct taskq *sosplice_taskq;
108 struct rwlock sosplice_lock = RWLOCK_INITIALIZER("sosplicelk");
109 #endif
110 
111 void
112 soinit(void)
113 {
114 	pool_init(&socket_pool, sizeof(struct socket), 0, IPL_SOFTNET, 0,
115 	    "sockpl", NULL);
116 #ifdef SOCKET_SPLICE
117 	pool_init(&sosplice_pool, sizeof(struct sosplice), 0, IPL_SOFTNET, 0,
118 	    "sosppl", NULL);
119 #endif
120 }
121 
122 /*
123  * Socket operation routines.
124  * These routines are called by the routines in
125  * sys_socket.c or from a system process, and
126  * implement the semantics of socket operations by
127  * switching out to the protocol specific routines.
128  */
129 int
130 socreate(int dom, struct socket **aso, int type, int proto)
131 {
132 	struct proc *p = curproc;		/* XXX */
133 	const struct protosw *prp;
134 	struct socket *so;
135 	int error, s;
136 
137 	if (proto)
138 		prp = pffindproto(dom, proto, type);
139 	else
140 		prp = pffindtype(dom, type);
141 	if (prp == NULL || prp->pr_attach == NULL)
142 		return (EPROTONOSUPPORT);
143 	if (prp->pr_type != type)
144 		return (EPROTOTYPE);
145 	so = pool_get(&socket_pool, PR_WAITOK | PR_ZERO);
146 	sigio_init(&so->so_sigio);
147 	TAILQ_INIT(&so->so_q0);
148 	TAILQ_INIT(&so->so_q);
149 	so->so_type = type;
150 	if (suser(p) == 0)
151 		so->so_state = SS_PRIV;
152 	so->so_ruid = p->p_ucred->cr_ruid;
153 	so->so_euid = p->p_ucred->cr_uid;
154 	so->so_rgid = p->p_ucred->cr_rgid;
155 	so->so_egid = p->p_ucred->cr_gid;
156 	so->so_cpid = p->p_p->ps_pid;
157 	so->so_proto = prp;
158 	so->so_snd.sb_timeo_nsecs = INFSLP;
159 	so->so_rcv.sb_timeo_nsecs = INFSLP;
160 
161 	s = solock(so);
162 	error = (*prp->pr_attach)(so, proto);
163 	if (error) {
164 		so->so_state |= SS_NOFDREF;
165 		/* sofree() calls sounlock(). */
166 		sofree(so, s);
167 		return (error);
168 	}
169 	sounlock(so, s);
170 	*aso = so;
171 	return (0);
172 }
173 
174 int
175 sobind(struct socket *so, struct mbuf *nam, struct proc *p)
176 {
177 	int error;
178 
179 	soassertlocked(so);
180 
181 	error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, NULL, nam, NULL, p);
182 	return (error);
183 }
184 
185 int
186 solisten(struct socket *so, int backlog)
187 {
188 	int s, error;
189 
190 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING|SS_ISDISCONNECTING))
191 		return (EINVAL);
192 #ifdef SOCKET_SPLICE
193 	if (isspliced(so) || issplicedback(so))
194 		return (EOPNOTSUPP);
195 #endif /* SOCKET_SPLICE */
196 	s = solock(so);
197 	error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, NULL, NULL, NULL,
198 	    curproc);
199 	if (error) {
200 		sounlock(so, s);
201 		return (error);
202 	}
203 	if (TAILQ_FIRST(&so->so_q) == NULL)
204 		so->so_options |= SO_ACCEPTCONN;
205 	if (backlog < 0 || backlog > somaxconn)
206 		backlog = somaxconn;
207 	if (backlog < sominconn)
208 		backlog = sominconn;
209 	so->so_qlimit = backlog;
210 	sounlock(so, s);
211 	return (0);
212 }
213 
214 #define SOSP_FREEING_READ	1
215 #define SOSP_FREEING_WRITE	2
216 void
217 sofree(struct socket *so, int s)
218 {
219 	soassertlocked(so);
220 
221 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
222 		sounlock(so, s);
223 		return;
224 	}
225 	if (so->so_head) {
226 		/*
227 		 * We must not decommission a socket that's on the accept(2)
228 		 * queue.  If we do, then accept(2) may hang after select(2)
229 		 * indicated that the listening socket was ready.
230 		 */
231 		if (!soqremque(so, 0)) {
232 			sounlock(so, s);
233 			return;
234 		}
235 	}
236 	sigio_free(&so->so_sigio);
237 #ifdef SOCKET_SPLICE
238 	if (so->so_sp) {
239 		if (issplicedback(so)) {
240 			int freeing = SOSP_FREEING_WRITE;
241 
242 			if (so->so_sp->ssp_soback == so)
243 				freeing |= SOSP_FREEING_READ;
244 			sounsplice(so->so_sp->ssp_soback, so, freeing);
245 		}
246 		if (isspliced(so)) {
247 			int freeing = SOSP_FREEING_READ;
248 
249 			if (so == so->so_sp->ssp_socket)
250 				freeing |= SOSP_FREEING_WRITE;
251 			sounsplice(so, so->so_sp->ssp_socket, freeing);
252 		}
253 	}
254 #endif /* SOCKET_SPLICE */
255 	sbrelease(so, &so->so_snd);
256 	sorflush(so);
257 	sounlock(so, s);
258 #ifdef SOCKET_SPLICE
259 	if (so->so_sp) {
260 		/* Reuse splice idle, sounsplice() has been called before. */
261 		timeout_set_proc(&so->so_sp->ssp_idleto, soreaper, so);
262 		timeout_add(&so->so_sp->ssp_idleto, 0);
263 	} else
264 #endif /* SOCKET_SPLICE */
265 	{
266 		pool_put(&socket_pool, so);
267 	}
268 }
269 
270 static inline uint64_t
271 solinger_nsec(struct socket *so)
272 {
273 	if (so->so_linger == 0)
274 		return INFSLP;
275 
276 	return SEC_TO_NSEC(so->so_linger);
277 }
278 
279 /*
280  * Close a socket on last file table reference removal.
281  * Initiate disconnect if connected.
282  * Free socket when disconnect complete.
283  */
284 int
285 soclose(struct socket *so, int flags)
286 {
287 	struct socket *so2;
288 	int s, error = 0;
289 
290 	s = solock(so);
291 	/* Revoke async IO early. There is a final revocation in sofree(). */
292 	sigio_free(&so->so_sigio);
293 	if (so->so_options & SO_ACCEPTCONN) {
294 		while ((so2 = TAILQ_FIRST(&so->so_q0)) != NULL) {
295 			(void) soqremque(so2, 0);
296 			(void) soabort(so2);
297 		}
298 		while ((so2 = TAILQ_FIRST(&so->so_q)) != NULL) {
299 			(void) soqremque(so2, 1);
300 			(void) soabort(so2);
301 		}
302 	}
303 	if (so->so_pcb == NULL)
304 		goto discard;
305 	if (so->so_state & SS_ISCONNECTED) {
306 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
307 			error = sodisconnect(so);
308 			if (error)
309 				goto drop;
310 		}
311 		if (so->so_options & SO_LINGER) {
312 			if ((so->so_state & SS_ISDISCONNECTING) &&
313 			    (flags & MSG_DONTWAIT))
314 				goto drop;
315 			while (so->so_state & SS_ISCONNECTED) {
316 				error = sosleep_nsec(so, &so->so_timeo,
317 				    PSOCK | PCATCH, "netcls",
318 				    solinger_nsec(so));
319 				if (error)
320 					break;
321 			}
322 		}
323 	}
324 drop:
325 	if (so->so_pcb) {
326 		int error2;
327 		KASSERT(so->so_proto->pr_detach);
328 		error2 = (*so->so_proto->pr_detach)(so);
329 		if (error == 0)
330 			error = error2;
331 	}
332 discard:
333 	if (so->so_state & SS_NOFDREF)
334 		panic("soclose NOFDREF: so %p, so_type %d", so, so->so_type);
335 	so->so_state |= SS_NOFDREF;
336 	/* sofree() calls sounlock(). */
337 	sofree(so, s);
338 	return (error);
339 }
340 
341 int
342 soabort(struct socket *so)
343 {
344 	soassertlocked(so);
345 
346 	return (*so->so_proto->pr_usrreq)(so, PRU_ABORT, NULL, NULL, NULL,
347 	   curproc);
348 }
349 
350 int
351 soaccept(struct socket *so, struct mbuf *nam)
352 {
353 	int error = 0;
354 
355 	soassertlocked(so);
356 
357 	if ((so->so_state & SS_NOFDREF) == 0)
358 		panic("soaccept !NOFDREF: so %p, so_type %d", so, so->so_type);
359 	so->so_state &= ~SS_NOFDREF;
360 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
361 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
362 		error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT, NULL,
363 		    nam, NULL, curproc);
364 	else
365 		error = ECONNABORTED;
366 	return (error);
367 }
368 
369 int
370 soconnect(struct socket *so, struct mbuf *nam)
371 {
372 	int error;
373 
374 	soassertlocked(so);
375 
376 	if (so->so_options & SO_ACCEPTCONN)
377 		return (EOPNOTSUPP);
378 	/*
379 	 * If protocol is connection-based, can only connect once.
380 	 * Otherwise, if connected, try to disconnect first.
381 	 * This allows user to disconnect by connecting to, e.g.,
382 	 * a null address.
383 	 */
384 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
385 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
386 	    (error = sodisconnect(so))))
387 		error = EISCONN;
388 	else
389 		error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT,
390 		    NULL, nam, NULL, curproc);
391 	return (error);
392 }
393 
394 int
395 soconnect2(struct socket *so1, struct socket *so2)
396 {
397 	int s, error;
398 
399 	s = solock(so1);
400 	error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2, NULL,
401 	    (struct mbuf *)so2, NULL, curproc);
402 	sounlock(so1, s);
403 	return (error);
404 }
405 
406 int
407 sodisconnect(struct socket *so)
408 {
409 	int error;
410 
411 	soassertlocked(so);
412 
413 	if ((so->so_state & SS_ISCONNECTED) == 0)
414 		return (ENOTCONN);
415 	if (so->so_state & SS_ISDISCONNECTING)
416 		return (EALREADY);
417 	error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT, NULL, NULL,
418 	    NULL, curproc);
419 	return (error);
420 }
421 
422 int m_getuio(struct mbuf **, int, long, struct uio *);
423 
424 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
425 /*
426  * Send on a socket.
427  * If send must go all at once and message is larger than
428  * send buffering, then hard error.
429  * Lock against other senders.
430  * If must go all at once and not enough room now, then
431  * inform user that this would block and do nothing.
432  * Otherwise, if nonblocking, send as much as possible.
433  * The data to be sent is described by "uio" if nonzero,
434  * otherwise by the mbuf chain "top" (which must be null
435  * if uio is not).  Data provided in mbuf chain must be small
436  * enough to send all at once.
437  *
438  * Returns nonzero on error, timeout or signal; callers
439  * must check for short counts if EINTR/ERESTART are returned.
440  * Data and control buffers are freed on return.
441  */
442 int
443 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
444     struct mbuf *control, int flags)
445 {
446 	long space, clen = 0;
447 	size_t resid;
448 	int error, s;
449 	int atomic = sosendallatonce(so) || top;
450 
451 	if (uio)
452 		resid = uio->uio_resid;
453 	else
454 		resid = top->m_pkthdr.len;
455 	/* MSG_EOR on a SOCK_STREAM socket is invalid. */
456 	if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
457 		m_freem(top);
458 		m_freem(control);
459 		return (EINVAL);
460 	}
461 	if (uio && uio->uio_procp)
462 		uio->uio_procp->p_ru.ru_msgsnd++;
463 	if (control) {
464 		/*
465 		 * In theory clen should be unsigned (since control->m_len is).
466 		 * However, space must be signed, as it might be less than 0
467 		 * if we over-committed, and we must use a signed comparison
468 		 * of space and clen.
469 		 */
470 		clen = control->m_len;
471 		/* reserve extra space for AF_UNIX's internalize */
472 		if (so->so_proto->pr_domain->dom_family == AF_UNIX &&
473 		    clen >= CMSG_ALIGN(sizeof(struct cmsghdr)) &&
474 		    mtod(control, struct cmsghdr *)->cmsg_type == SCM_RIGHTS)
475 			clen = CMSG_SPACE(
476 			    (clen - CMSG_ALIGN(sizeof(struct cmsghdr))) *
477 			    (sizeof(struct fdpass) / sizeof(int)));
478 	}
479 
480 #define	snderr(errno)	{ error = errno; goto release; }
481 
482 	s = solock(so);
483 restart:
484 	if ((error = sblock(so, &so->so_snd, SBLOCKWAIT(flags))) != 0)
485 		goto out;
486 	so->so_state |= SS_ISSENDING;
487 	do {
488 		if (so->so_state & SS_CANTSENDMORE)
489 			snderr(EPIPE);
490 		if (so->so_error) {
491 			error = so->so_error;
492 			so->so_error = 0;
493 			snderr(error);
494 		}
495 		if ((so->so_state & SS_ISCONNECTED) == 0) {
496 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
497 				if (!(resid == 0 && clen != 0))
498 					snderr(ENOTCONN);
499 			} else if (addr == 0)
500 				snderr(EDESTADDRREQ);
501 		}
502 		space = sbspace(so, &so->so_snd);
503 		if (flags & MSG_OOB)
504 			space += 1024;
505 		if (so->so_proto->pr_domain->dom_family == AF_UNIX) {
506 			if (atomic && resid > so->so_snd.sb_hiwat)
507 				snderr(EMSGSIZE);
508 		} else {
509 			if (clen > so->so_snd.sb_hiwat ||
510 			    (atomic && resid > so->so_snd.sb_hiwat - clen))
511 				snderr(EMSGSIZE);
512 		}
513 		if (space < clen ||
514 		    (space - clen < resid &&
515 		    (atomic || space < so->so_snd.sb_lowat))) {
516 			if (flags & MSG_DONTWAIT)
517 				snderr(EWOULDBLOCK);
518 			sbunlock(so, &so->so_snd);
519 			error = sbwait(so, &so->so_snd);
520 			so->so_state &= ~SS_ISSENDING;
521 			if (error)
522 				goto out;
523 			goto restart;
524 		}
525 		space -= clen;
526 		do {
527 			if (uio == NULL) {
528 				/*
529 				 * Data is prepackaged in "top".
530 				 */
531 				resid = 0;
532 				if (flags & MSG_EOR)
533 					top->m_flags |= M_EOR;
534 			} else {
535 				sounlock(so, s);
536 				error = m_getuio(&top, atomic, space, uio);
537 				s = solock(so);
538 				if (error)
539 					goto release;
540 				space -= top->m_pkthdr.len;
541 				resid = uio->uio_resid;
542 				if (flags & MSG_EOR)
543 					top->m_flags |= M_EOR;
544 			}
545 			if (resid == 0)
546 				so->so_state &= ~SS_ISSENDING;
547 			if (top && so->so_options & SO_ZEROIZE)
548 				top->m_flags |= M_ZEROIZE;
549 			error = (*so->so_proto->pr_usrreq)(so,
550 			    (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND,
551 			    top, addr, control, curproc);
552 			clen = 0;
553 			control = NULL;
554 			top = NULL;
555 			if (error)
556 				goto release;
557 		} while (resid && space > 0);
558 	} while (resid);
559 
560 release:
561 	so->so_state &= ~SS_ISSENDING;
562 	sbunlock(so, &so->so_snd);
563 out:
564 	sounlock(so, s);
565 	m_freem(top);
566 	m_freem(control);
567 	return (error);
568 }
569 
570 int
571 m_getuio(struct mbuf **mp, int atomic, long space, struct uio *uio)
572 {
573 	struct mbuf *m, *top = NULL;
574 	struct mbuf **nextp = &top;
575 	u_long len, mlen;
576 	size_t resid = uio->uio_resid;
577 	int error;
578 
579 	do {
580 		if (top == NULL) {
581 			MGETHDR(m, M_WAIT, MT_DATA);
582 			mlen = MHLEN;
583 			m->m_pkthdr.len = 0;
584 			m->m_pkthdr.ph_ifidx = 0;
585 		} else {
586 			MGET(m, M_WAIT, MT_DATA);
587 			mlen = MLEN;
588 		}
589 		/* chain mbuf together */
590 		*nextp = m;
591 		nextp = &m->m_next;
592 
593 		resid = ulmin(resid, space);
594 		if (resid >= MINCLSIZE) {
595 			MCLGETI(m, M_NOWAIT, NULL, ulmin(resid, MAXMCLBYTES));
596 			if ((m->m_flags & M_EXT) == 0)
597 				MCLGETI(m, M_NOWAIT, NULL, MCLBYTES);
598 			if ((m->m_flags & M_EXT) == 0)
599 				goto nopages;
600 			mlen = m->m_ext.ext_size;
601 			len = ulmin(mlen, resid);
602 			/*
603 			 * For datagram protocols, leave room
604 			 * for protocol headers in first mbuf.
605 			 */
606 			if (atomic && m == top && len < mlen - max_hdr)
607 				m->m_data += max_hdr;
608 		} else {
609 nopages:
610 			len = ulmin(mlen, resid);
611 			/*
612 			 * For datagram protocols, leave room
613 			 * for protocol headers in first mbuf.
614 			 */
615 			if (atomic && m == top && len < mlen - max_hdr)
616 				m_align(m, len);
617 		}
618 
619 		error = uiomove(mtod(m, caddr_t), len, uio);
620 		if (error) {
621 			m_freem(top);
622 			return (error);
623 		}
624 
625 		/* adjust counters */
626 		resid = uio->uio_resid;
627 		space -= len;
628 		m->m_len = len;
629 		top->m_pkthdr.len += len;
630 
631 		/* Is there more space and more data? */
632 	} while (space > 0 && resid > 0);
633 
634 	*mp = top;
635 	return 0;
636 }
637 
638 /*
639  * Following replacement or removal of the first mbuf on the first
640  * mbuf chain of a socket buffer, push necessary state changes back
641  * into the socket buffer so that other consumers see the values
642  * consistently.  'nextrecord' is the callers locally stored value of
643  * the original value of sb->sb_mb->m_nextpkt which must be restored
644  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
645  */
646 void
647 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
648 {
649 
650 	/*
651 	 * First, update for the new value of nextrecord.  If necessary,
652 	 * make it the first record.
653 	 */
654 	if (sb->sb_mb != NULL)
655 		sb->sb_mb->m_nextpkt = nextrecord;
656 	else
657 		sb->sb_mb = nextrecord;
658 
659 	/*
660 	 * Now update any dependent socket buffer fields to reflect
661 	 * the new state.  This is an inline of SB_EMPTY_FIXUP, with
662 	 * the addition of a second clause that takes care of the
663 	 * case where sb_mb has been updated, but remains the last
664 	 * record.
665 	 */
666 	if (sb->sb_mb == NULL) {
667 		sb->sb_mbtail = NULL;
668 		sb->sb_lastrecord = NULL;
669 	} else if (sb->sb_mb->m_nextpkt == NULL)
670 		sb->sb_lastrecord = sb->sb_mb;
671 }
672 
673 /*
674  * Implement receive operations on a socket.
675  * We depend on the way that records are added to the sockbuf
676  * by sbappend*.  In particular, each record (mbufs linked through m_next)
677  * must begin with an address if the protocol so specifies,
678  * followed by an optional mbuf or mbufs containing ancillary data,
679  * and then zero or more mbufs of data.
680  * In order to avoid blocking network for the entire time here, we release
681  * the solock() while doing the actual copy to user space.
682  * Although the sockbuf is locked, new data may still be appended,
683  * and thus we must maintain consistency of the sockbuf during that time.
684  *
685  * The caller may receive the data as a single mbuf chain by supplying
686  * an mbuf **mp0 for use in returning the chain.  The uio is then used
687  * only for the count in uio_resid.
688  */
689 int
690 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
691     struct mbuf **mp0, struct mbuf **controlp, int *flagsp,
692     socklen_t controllen)
693 {
694 	struct mbuf *m, **mp;
695 	struct mbuf *cm;
696 	u_long len, offset, moff;
697 	int flags, error, s, type, uio_error = 0;
698 	const struct protosw *pr = so->so_proto;
699 	struct mbuf *nextrecord;
700 	size_t resid, orig_resid = uio->uio_resid;
701 
702 	mp = mp0;
703 	if (paddr)
704 		*paddr = NULL;
705 	if (controlp)
706 		*controlp = NULL;
707 	if (flagsp)
708 		flags = *flagsp &~ MSG_EOR;
709 	else
710 		flags = 0;
711 	if (flags & MSG_OOB) {
712 		m = m_get(M_WAIT, MT_DATA);
713 		s = solock(so);
714 		error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m,
715 		    (struct mbuf *)(long)(flags & MSG_PEEK), NULL, curproc);
716 		sounlock(so, s);
717 		if (error)
718 			goto bad;
719 		do {
720 			error = uiomove(mtod(m, caddr_t),
721 			    ulmin(uio->uio_resid, m->m_len), uio);
722 			m = m_free(m);
723 		} while (uio->uio_resid && error == 0 && m);
724 bad:
725 		m_freem(m);
726 		return (error);
727 	}
728 	if (mp)
729 		*mp = NULL;
730 
731 	s = solock(so);
732 restart:
733 	if ((error = sblock(so, &so->so_rcv, SBLOCKWAIT(flags))) != 0) {
734 		sounlock(so, s);
735 		return (error);
736 	}
737 
738 	m = so->so_rcv.sb_mb;
739 #ifdef SOCKET_SPLICE
740 	if (isspliced(so))
741 		m = NULL;
742 #endif /* SOCKET_SPLICE */
743 	/*
744 	 * If we have less data than requested, block awaiting more
745 	 * (subject to any timeout) if:
746 	 *   1. the current count is less than the low water mark,
747 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
748 	 *	receive operation at once if we block (resid <= hiwat), or
749 	 *   3. MSG_DONTWAIT is not set.
750 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
751 	 * we have to do the receive in sections, and thus risk returning
752 	 * a short count if a timeout or signal occurs after we start.
753 	 */
754 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
755 	    so->so_rcv.sb_cc < uio->uio_resid) &&
756 	    (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
757 	    ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
758 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
759 #ifdef DIAGNOSTIC
760 		if (m == NULL && so->so_rcv.sb_cc)
761 #ifdef SOCKET_SPLICE
762 		    if (!isspliced(so))
763 #endif /* SOCKET_SPLICE */
764 			panic("receive 1: so %p, so_type %d, sb_cc %lu",
765 			    so, so->so_type, so->so_rcv.sb_cc);
766 #endif
767 		if (so->so_error) {
768 			if (m)
769 				goto dontblock;
770 			error = so->so_error;
771 			if ((flags & MSG_PEEK) == 0)
772 				so->so_error = 0;
773 			goto release;
774 		}
775 		if (so->so_state & SS_CANTRCVMORE) {
776 			if (m)
777 				goto dontblock;
778 			else if (so->so_rcv.sb_cc == 0)
779 				goto release;
780 		}
781 		for (; m; m = m->m_next)
782 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
783 				m = so->so_rcv.sb_mb;
784 				goto dontblock;
785 			}
786 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
787 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
788 			error = ENOTCONN;
789 			goto release;
790 		}
791 		if (uio->uio_resid == 0 && controlp == NULL)
792 			goto release;
793 		if (flags & MSG_DONTWAIT) {
794 			error = EWOULDBLOCK;
795 			goto release;
796 		}
797 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
798 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
799 		sbunlock(so, &so->so_rcv);
800 		error = sbwait(so, &so->so_rcv);
801 		if (error) {
802 			sounlock(so, s);
803 			return (error);
804 		}
805 		goto restart;
806 	}
807 dontblock:
808 	/*
809 	 * On entry here, m points to the first record of the socket buffer.
810 	 * From this point onward, we maintain 'nextrecord' as a cache of the
811 	 * pointer to the next record in the socket buffer.  We must keep the
812 	 * various socket buffer pointers and local stack versions of the
813 	 * pointers in sync, pushing out modifications before operations that
814 	 * may sleep, and re-reading them afterwards.
815 	 *
816 	 * Otherwise, we will race with the network stack appending new data
817 	 * or records onto the socket buffer by using inconsistent/stale
818 	 * versions of the field, possibly resulting in socket buffer
819 	 * corruption.
820 	 */
821 	if (uio->uio_procp)
822 		uio->uio_procp->p_ru.ru_msgrcv++;
823 	KASSERT(m == so->so_rcv.sb_mb);
824 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
825 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
826 	nextrecord = m->m_nextpkt;
827 	if (pr->pr_flags & PR_ADDR) {
828 #ifdef DIAGNOSTIC
829 		if (m->m_type != MT_SONAME)
830 			panic("receive 1a: so %p, so_type %d, m %p, m_type %d",
831 			    so, so->so_type, m, m->m_type);
832 #endif
833 		orig_resid = 0;
834 		if (flags & MSG_PEEK) {
835 			if (paddr)
836 				*paddr = m_copym(m, 0, m->m_len, M_NOWAIT);
837 			m = m->m_next;
838 		} else {
839 			sbfree(&so->so_rcv, m);
840 			if (paddr) {
841 				*paddr = m;
842 				so->so_rcv.sb_mb = m->m_next;
843 				m->m_next = 0;
844 				m = so->so_rcv.sb_mb;
845 			} else {
846 				so->so_rcv.sb_mb = m_free(m);
847 				m = so->so_rcv.sb_mb;
848 			}
849 			sbsync(&so->so_rcv, nextrecord);
850 		}
851 	}
852 	while (m && m->m_type == MT_CONTROL && error == 0) {
853 		int skip = 0;
854 		if (flags & MSG_PEEK) {
855 			if (mtod(m, struct cmsghdr *)->cmsg_type ==
856 			    SCM_RIGHTS) {
857 				/* don't leak internalized SCM_RIGHTS msgs */
858 				skip = 1;
859 			} else if (controlp)
860 				*controlp = m_copym(m, 0, m->m_len, M_NOWAIT);
861 			m = m->m_next;
862 		} else {
863 			sbfree(&so->so_rcv, m);
864 			so->so_rcv.sb_mb = m->m_next;
865 			m->m_nextpkt = m->m_next = NULL;
866 			cm = m;
867 			m = so->so_rcv.sb_mb;
868 			sbsync(&so->so_rcv, nextrecord);
869 			if (controlp) {
870 				if (pr->pr_domain->dom_externalize) {
871 					error =
872 					    (*pr->pr_domain->dom_externalize)
873 					    (cm, controllen, flags);
874 				}
875 				*controlp = cm;
876 			} else {
877 				/*
878 				 * Dispose of any SCM_RIGHTS message that went
879 				 * through the read path rather than recv.
880 				 */
881 				if (pr->pr_domain->dom_dispose)
882 					pr->pr_domain->dom_dispose(cm);
883 				m_free(cm);
884 			}
885 		}
886 		if (m != NULL)
887 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
888 		else
889 			nextrecord = so->so_rcv.sb_mb;
890 		if (controlp && !skip) {
891 			orig_resid = 0;
892 			controlp = &(*controlp)->m_next;
893 		}
894 	}
895 
896 	/* If m is non-NULL, we have some data to read. */
897 	if (m) {
898 		type = m->m_type;
899 		if (type == MT_OOBDATA)
900 			flags |= MSG_OOB;
901 		if (m->m_flags & M_BCAST)
902 			flags |= MSG_BCAST;
903 		if (m->m_flags & M_MCAST)
904 			flags |= MSG_MCAST;
905 	}
906 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
907 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
908 
909 	moff = 0;
910 	offset = 0;
911 	while (m && uio->uio_resid > 0 && error == 0) {
912 		if (m->m_type == MT_OOBDATA) {
913 			if (type != MT_OOBDATA)
914 				break;
915 		} else if (type == MT_OOBDATA) {
916 			break;
917 		} else if (m->m_type == MT_CONTROL) {
918 			/*
919 			 * If there is more than one control message in the
920 			 * stream, we do a short read.  Next can be received
921 			 * or disposed by another system call.
922 			 */
923 			break;
924 #ifdef DIAGNOSTIC
925 		} else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) {
926 			panic("receive 3: so %p, so_type %d, m %p, m_type %d",
927 			    so, so->so_type, m, m->m_type);
928 #endif
929 		}
930 		so->so_state &= ~SS_RCVATMARK;
931 		len = uio->uio_resid;
932 		if (so->so_oobmark && len > so->so_oobmark - offset)
933 			len = so->so_oobmark - offset;
934 		if (len > m->m_len - moff)
935 			len = m->m_len - moff;
936 		/*
937 		 * If mp is set, just pass back the mbufs.
938 		 * Otherwise copy them out via the uio, then free.
939 		 * Sockbuf must be consistent here (points to current mbuf,
940 		 * it points to next record) when we drop priority;
941 		 * we must note any additions to the sockbuf when we
942 		 * block interrupts again.
943 		 */
944 		if (mp == NULL && uio_error == 0) {
945 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
946 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
947 			resid = uio->uio_resid;
948 			sounlock(so, s);
949 			uio_error = uiomove(mtod(m, caddr_t) + moff, len, uio);
950 			s = solock(so);
951 			if (uio_error)
952 				uio->uio_resid = resid - len;
953 		} else
954 			uio->uio_resid -= len;
955 		if (len == m->m_len - moff) {
956 			if (m->m_flags & M_EOR)
957 				flags |= MSG_EOR;
958 			if (flags & MSG_PEEK) {
959 				m = m->m_next;
960 				moff = 0;
961 			} else {
962 				nextrecord = m->m_nextpkt;
963 				sbfree(&so->so_rcv, m);
964 				if (mp) {
965 					*mp = m;
966 					mp = &m->m_next;
967 					so->so_rcv.sb_mb = m = m->m_next;
968 					*mp = NULL;
969 				} else {
970 					so->so_rcv.sb_mb = m_free(m);
971 					m = so->so_rcv.sb_mb;
972 				}
973 				/*
974 				 * If m != NULL, we also know that
975 				 * so->so_rcv.sb_mb != NULL.
976 				 */
977 				KASSERT(so->so_rcv.sb_mb == m);
978 				if (m) {
979 					m->m_nextpkt = nextrecord;
980 					if (nextrecord == NULL)
981 						so->so_rcv.sb_lastrecord = m;
982 				} else {
983 					so->so_rcv.sb_mb = nextrecord;
984 					SB_EMPTY_FIXUP(&so->so_rcv);
985 				}
986 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
987 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
988 			}
989 		} else {
990 			if (flags & MSG_PEEK)
991 				moff += len;
992 			else {
993 				if (mp)
994 					*mp = m_copym(m, 0, len, M_WAIT);
995 				m->m_data += len;
996 				m->m_len -= len;
997 				so->so_rcv.sb_cc -= len;
998 				so->so_rcv.sb_datacc -= len;
999 			}
1000 		}
1001 		if (so->so_oobmark) {
1002 			if ((flags & MSG_PEEK) == 0) {
1003 				so->so_oobmark -= len;
1004 				if (so->so_oobmark == 0) {
1005 					so->so_state |= SS_RCVATMARK;
1006 					break;
1007 				}
1008 			} else {
1009 				offset += len;
1010 				if (offset == so->so_oobmark)
1011 					break;
1012 			}
1013 		}
1014 		if (flags & MSG_EOR)
1015 			break;
1016 		/*
1017 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1018 		 * we must not quit until "uio->uio_resid == 0" or an error
1019 		 * termination.  If a signal/timeout occurs, return
1020 		 * with a short count but without error.
1021 		 * Keep sockbuf locked against other readers.
1022 		 */
1023 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1024 		    !sosendallatonce(so) && !nextrecord) {
1025 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1026 				break;
1027 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1028 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1029 			error = sbwait(so, &so->so_rcv);
1030 			if (error) {
1031 				sbunlock(so, &so->so_rcv);
1032 				sounlock(so, s);
1033 				return (0);
1034 			}
1035 			if ((m = so->so_rcv.sb_mb) != NULL)
1036 				nextrecord = m->m_nextpkt;
1037 		}
1038 	}
1039 
1040 	if (m && pr->pr_flags & PR_ATOMIC) {
1041 		flags |= MSG_TRUNC;
1042 		if ((flags & MSG_PEEK) == 0)
1043 			(void) sbdroprecord(&so->so_rcv);
1044 	}
1045 	if ((flags & MSG_PEEK) == 0) {
1046 		if (m == NULL) {
1047 			/*
1048 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1049 			 * part makes sure sb_lastrecord is up-to-date if
1050 			 * there is still data in the socket buffer.
1051 			 */
1052 			so->so_rcv.sb_mb = nextrecord;
1053 			if (so->so_rcv.sb_mb == NULL) {
1054 				so->so_rcv.sb_mbtail = NULL;
1055 				so->so_rcv.sb_lastrecord = NULL;
1056 			} else if (nextrecord->m_nextpkt == NULL)
1057 				so->so_rcv.sb_lastrecord = nextrecord;
1058 		}
1059 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1060 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1061 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1062 			(*pr->pr_usrreq)(so, PRU_RCVD, NULL,
1063 			    (struct mbuf *)(long)flags, NULL, curproc);
1064 	}
1065 	if (orig_resid == uio->uio_resid && orig_resid &&
1066 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1067 		sbunlock(so, &so->so_rcv);
1068 		goto restart;
1069 	}
1070 
1071 	if (uio_error)
1072 		error = uio_error;
1073 
1074 	if (flagsp)
1075 		*flagsp |= flags;
1076 release:
1077 	sbunlock(so, &so->so_rcv);
1078 	sounlock(so, s);
1079 	return (error);
1080 }
1081 
1082 int
1083 soshutdown(struct socket *so, int how)
1084 {
1085 	const struct protosw *pr = so->so_proto;
1086 	int s, error = 0;
1087 
1088 	s = solock(so);
1089 	switch (how) {
1090 	case SHUT_RD:
1091 		sorflush(so);
1092 		break;
1093 	case SHUT_RDWR:
1094 		sorflush(so);
1095 		/* FALLTHROUGH */
1096 	case SHUT_WR:
1097 		error = (*pr->pr_usrreq)(so, PRU_SHUTDOWN, NULL, NULL, NULL,
1098 		    curproc);
1099 		break;
1100 	default:
1101 		error = EINVAL;
1102 		break;
1103 	}
1104 	sounlock(so, s);
1105 
1106 	return (error);
1107 }
1108 
1109 void
1110 sorflush(struct socket *so)
1111 {
1112 	struct sockbuf *sb = &so->so_rcv;
1113 	const struct protosw *pr = so->so_proto;
1114 	struct socket aso;
1115 	int error;
1116 
1117 	sb->sb_flags |= SB_NOINTR;
1118 	error = sblock(so, sb, M_WAITOK);
1119 	/* with SB_NOINTR and M_WAITOK sblock() must not fail */
1120 	KASSERT(error == 0);
1121 	socantrcvmore(so);
1122 	sbunlock(so, sb);
1123 	aso.so_proto = pr;
1124 	aso.so_rcv = *sb;
1125 	memset(&sb->sb_startzero, 0,
1126 	     (caddr_t)&sb->sb_endzero - (caddr_t)&sb->sb_startzero);
1127 	sb->sb_timeo_nsecs = INFSLP;
1128 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose)
1129 		(*pr->pr_domain->dom_dispose)(aso.so_rcv.sb_mb);
1130 	sbrelease(&aso, &aso.so_rcv);
1131 }
1132 
1133 #ifdef SOCKET_SPLICE
1134 
1135 #define so_splicelen	so_sp->ssp_len
1136 #define so_splicemax	so_sp->ssp_max
1137 #define so_idletv	so_sp->ssp_idletv
1138 #define so_idleto	so_sp->ssp_idleto
1139 #define so_splicetask	so_sp->ssp_task
1140 
1141 int
1142 sosplice(struct socket *so, int fd, off_t max, struct timeval *tv)
1143 {
1144 	struct file	*fp;
1145 	struct socket	*sosp;
1146 	struct sosplice	*sp;
1147 	struct taskq	*tq;
1148 	int		 error = 0;
1149 
1150 	soassertlocked(so);
1151 
1152 	if (sosplice_taskq == NULL) {
1153 		rw_enter_write(&sosplice_lock);
1154 		if (sosplice_taskq == NULL) {
1155 			tq = taskq_create("sosplice", 1, IPL_SOFTNET,
1156 			    TASKQ_MPSAFE);
1157 			/* Ensure the taskq is fully visible to other CPUs. */
1158 			membar_producer();
1159 			sosplice_taskq = tq;
1160 		}
1161 		rw_exit_write(&sosplice_lock);
1162 	}
1163 	if (sosplice_taskq == NULL)
1164 		return (ENOMEM);
1165 
1166 	if ((so->so_proto->pr_flags & PR_SPLICE) == 0)
1167 		return (EPROTONOSUPPORT);
1168 	if (so->so_options & SO_ACCEPTCONN)
1169 		return (EOPNOTSUPP);
1170 	if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1171 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
1172 		return (ENOTCONN);
1173 	if (so->so_sp == NULL) {
1174 		sp = pool_get(&sosplice_pool, PR_WAITOK | PR_ZERO);
1175 		if (so->so_sp == NULL)
1176 			so->so_sp = sp;
1177 		else
1178 			pool_put(&sosplice_pool, sp);
1179 	}
1180 
1181 	/* If no fd is given, unsplice by removing existing link. */
1182 	if (fd < 0) {
1183 		/* Lock receive buffer. */
1184 		if ((error = sblock(so, &so->so_rcv, M_WAITOK)) != 0) {
1185 			return (error);
1186 		}
1187 		if (so->so_sp->ssp_socket)
1188 			sounsplice(so, so->so_sp->ssp_socket, 0);
1189 		sbunlock(so, &so->so_rcv);
1190 		return (0);
1191 	}
1192 
1193 	if (max && max < 0)
1194 		return (EINVAL);
1195 
1196 	if (tv && (tv->tv_sec < 0 || tv->tv_usec < 0))
1197 		return (EINVAL);
1198 
1199 	/* Find sosp, the drain socket where data will be spliced into. */
1200 	if ((error = getsock(curproc, fd, &fp)) != 0)
1201 		return (error);
1202 	sosp = fp->f_data;
1203 	if (sosp->so_sp == NULL) {
1204 		sp = pool_get(&sosplice_pool, PR_WAITOK | PR_ZERO);
1205 		if (sosp->so_sp == NULL)
1206 			sosp->so_sp = sp;
1207 		else
1208 			pool_put(&sosplice_pool, sp);
1209 	}
1210 
1211 	/* Lock both receive and send buffer. */
1212 	if ((error = sblock(so, &so->so_rcv, M_WAITOK)) != 0) {
1213 		goto frele;
1214 	}
1215 	if ((error = sblock(so, &sosp->so_snd, M_WAITOK)) != 0) {
1216 		sbunlock(so, &so->so_rcv);
1217 		goto frele;
1218 	}
1219 
1220 	if (so->so_sp->ssp_socket || sosp->so_sp->ssp_soback) {
1221 		error = EBUSY;
1222 		goto release;
1223 	}
1224 	if (sosp->so_proto->pr_usrreq != so->so_proto->pr_usrreq) {
1225 		error = EPROTONOSUPPORT;
1226 		goto release;
1227 	}
1228 	if (sosp->so_options & SO_ACCEPTCONN) {
1229 		error = EOPNOTSUPP;
1230 		goto release;
1231 	}
1232 	if ((sosp->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0) {
1233 		error = ENOTCONN;
1234 		goto release;
1235 	}
1236 
1237 	/* Splice so and sosp together. */
1238 	so->so_sp->ssp_socket = sosp;
1239 	sosp->so_sp->ssp_soback = so;
1240 	so->so_splicelen = 0;
1241 	so->so_splicemax = max;
1242 	if (tv)
1243 		so->so_idletv = *tv;
1244 	else
1245 		timerclear(&so->so_idletv);
1246 	timeout_set_proc(&so->so_idleto, soidle, so);
1247 	task_set(&so->so_splicetask, sotask, so);
1248 
1249 	/*
1250 	 * To prevent softnet interrupt from calling somove() while
1251 	 * we sleep, the socket buffers are not marked as spliced yet.
1252 	 */
1253 	if (somove(so, M_WAIT)) {
1254 		so->so_rcv.sb_flags |= SB_SPLICE;
1255 		sosp->so_snd.sb_flags |= SB_SPLICE;
1256 	}
1257 
1258  release:
1259 	sbunlock(sosp, &sosp->so_snd);
1260 	sbunlock(so, &so->so_rcv);
1261  frele:
1262 	FRELE(fp, curproc);
1263 	return (error);
1264 }
1265 
1266 void
1267 sounsplice(struct socket *so, struct socket *sosp, int freeing)
1268 {
1269 	soassertlocked(so);
1270 
1271 	task_del(sosplice_taskq, &so->so_splicetask);
1272 	timeout_del(&so->so_idleto);
1273 	sosp->so_snd.sb_flags &= ~SB_SPLICE;
1274 	so->so_rcv.sb_flags &= ~SB_SPLICE;
1275 	so->so_sp->ssp_socket = sosp->so_sp->ssp_soback = NULL;
1276 	/* Do not wakeup a socket that is about to be freed. */
1277 	if ((freeing & SOSP_FREEING_READ) == 0 && soreadable(so))
1278 		sorwakeup(so);
1279 	if ((freeing & SOSP_FREEING_WRITE) == 0 && sowriteable(sosp))
1280 		sowwakeup(sosp);
1281 }
1282 
1283 void
1284 soidle(void *arg)
1285 {
1286 	struct socket *so = arg;
1287 	int s;
1288 
1289 	s = solock(so);
1290 	if (so->so_rcv.sb_flags & SB_SPLICE) {
1291 		so->so_error = ETIMEDOUT;
1292 		sounsplice(so, so->so_sp->ssp_socket, 0);
1293 	}
1294 	sounlock(so, s);
1295 }
1296 
1297 void
1298 sotask(void *arg)
1299 {
1300 	struct socket *so = arg;
1301 	int s;
1302 
1303 	s = solock(so);
1304 	if (so->so_rcv.sb_flags & SB_SPLICE) {
1305 		/*
1306 		 * We may not sleep here as sofree() and unsplice() may be
1307 		 * called from softnet interrupt context.  This would remove
1308 		 * the socket during somove().
1309 		 */
1310 		somove(so, M_DONTWAIT);
1311 	}
1312 	sounlock(so, s);
1313 
1314 	/* Avoid user land starvation. */
1315 	yield();
1316 }
1317 
1318 /*
1319  * The socket splicing task or idle timeout may sleep while grabbing the net
1320  * lock.  As sofree() can be called anytime, sotask() or soidle() could access
1321  * the socket memory of a freed socket after wakeup.  So delay the pool_put()
1322  * after all pending socket splicing tasks or timeouts have finished.  Do this
1323  * by scheduling it on the same threads.
1324  */
1325 void
1326 soreaper(void *arg)
1327 {
1328 	struct socket *so = arg;
1329 
1330 	/* Reuse splice task, sounsplice() has been called before. */
1331 	task_set(&so->so_sp->ssp_task, soput, so);
1332 	task_add(sosplice_taskq, &so->so_sp->ssp_task);
1333 }
1334 
1335 void
1336 soput(void *arg)
1337 {
1338 	struct socket *so = arg;
1339 
1340 	pool_put(&sosplice_pool, so->so_sp);
1341 	pool_put(&socket_pool, so);
1342 }
1343 
1344 /*
1345  * Move data from receive buffer of spliced source socket to send
1346  * buffer of drain socket.  Try to move as much as possible in one
1347  * big chunk.  It is a TCP only implementation.
1348  * Return value 0 means splicing has been finished, 1 continue.
1349  */
1350 int
1351 somove(struct socket *so, int wait)
1352 {
1353 	struct socket	*sosp = so->so_sp->ssp_socket;
1354 	struct mbuf	*m, **mp, *nextrecord;
1355 	u_long		 len, off, oobmark;
1356 	long		 space;
1357 	int		 error = 0, maxreached = 0;
1358 	unsigned int	 state;
1359 
1360 	soassertlocked(so);
1361 
1362  nextpkt:
1363 	if (so->so_error) {
1364 		error = so->so_error;
1365 		goto release;
1366 	}
1367 	if (sosp->so_state & SS_CANTSENDMORE) {
1368 		error = EPIPE;
1369 		goto release;
1370 	}
1371 	if (sosp->so_error && sosp->so_error != ETIMEDOUT &&
1372 	    sosp->so_error != EFBIG && sosp->so_error != ELOOP) {
1373 		error = sosp->so_error;
1374 		goto release;
1375 	}
1376 	if ((sosp->so_state & SS_ISCONNECTED) == 0)
1377 		goto release;
1378 
1379 	/* Calculate how many bytes can be copied now. */
1380 	len = so->so_rcv.sb_datacc;
1381 	if (so->so_splicemax) {
1382 		KASSERT(so->so_splicelen < so->so_splicemax);
1383 		if (so->so_splicemax <= so->so_splicelen + len) {
1384 			len = so->so_splicemax - so->so_splicelen;
1385 			maxreached = 1;
1386 		}
1387 	}
1388 	space = sbspace(sosp, &sosp->so_snd);
1389 	if (so->so_oobmark && so->so_oobmark < len &&
1390 	    so->so_oobmark < space + 1024)
1391 		space += 1024;
1392 	if (space <= 0) {
1393 		maxreached = 0;
1394 		goto release;
1395 	}
1396 	if (space < len) {
1397 		maxreached = 0;
1398 		if (space < sosp->so_snd.sb_lowat)
1399 			goto release;
1400 		len = space;
1401 	}
1402 	sosp->so_state |= SS_ISSENDING;
1403 
1404 	SBLASTRECORDCHK(&so->so_rcv, "somove 1");
1405 	SBLASTMBUFCHK(&so->so_rcv, "somove 1");
1406 	m = so->so_rcv.sb_mb;
1407 	if (m == NULL)
1408 		goto release;
1409 	nextrecord = m->m_nextpkt;
1410 
1411 	/* Drop address and control information not used with splicing. */
1412 	if (so->so_proto->pr_flags & PR_ADDR) {
1413 #ifdef DIAGNOSTIC
1414 		if (m->m_type != MT_SONAME)
1415 			panic("somove soname: so %p, so_type %d, m %p, "
1416 			    "m_type %d", so, so->so_type, m, m->m_type);
1417 #endif
1418 		m = m->m_next;
1419 	}
1420 	while (m && m->m_type == MT_CONTROL)
1421 		m = m->m_next;
1422 	if (m == NULL) {
1423 		sbdroprecord(&so->so_rcv);
1424 		if (so->so_proto->pr_flags & PR_WANTRCVD && so->so_pcb)
1425 			(so->so_proto->pr_usrreq)(so, PRU_RCVD, NULL,
1426 			    NULL, NULL, NULL);
1427 		goto nextpkt;
1428 	}
1429 
1430 	/*
1431 	 * By splicing sockets connected to localhost, userland might create a
1432 	 * loop.  Dissolve splicing with error if loop is detected by counter.
1433 	 */
1434 	if ((m->m_flags & M_PKTHDR) && m->m_pkthdr.ph_loopcnt++ >= M_MAXLOOP) {
1435 		error = ELOOP;
1436 		goto release;
1437 	}
1438 
1439 	if (so->so_proto->pr_flags & PR_ATOMIC) {
1440 		if ((m->m_flags & M_PKTHDR) == 0)
1441 			panic("somove !PKTHDR: so %p, so_type %d, m %p, "
1442 			    "m_type %d", so, so->so_type, m, m->m_type);
1443 		if (sosp->so_snd.sb_hiwat < m->m_pkthdr.len) {
1444 			error = EMSGSIZE;
1445 			goto release;
1446 		}
1447 		if (len < m->m_pkthdr.len)
1448 			goto release;
1449 		if (m->m_pkthdr.len < len) {
1450 			maxreached = 0;
1451 			len = m->m_pkthdr.len;
1452 		}
1453 		/*
1454 		 * Throw away the name mbuf after it has been assured
1455 		 * that the whole first record can be processed.
1456 		 */
1457 		m = so->so_rcv.sb_mb;
1458 		sbfree(&so->so_rcv, m);
1459 		so->so_rcv.sb_mb = m_free(m);
1460 		sbsync(&so->so_rcv, nextrecord);
1461 	}
1462 	/*
1463 	 * Throw away the control mbufs after it has been assured
1464 	 * that the whole first record can be processed.
1465 	 */
1466 	m = so->so_rcv.sb_mb;
1467 	while (m && m->m_type == MT_CONTROL) {
1468 		sbfree(&so->so_rcv, m);
1469 		so->so_rcv.sb_mb = m_free(m);
1470 		m = so->so_rcv.sb_mb;
1471 		sbsync(&so->so_rcv, nextrecord);
1472 	}
1473 
1474 	SBLASTRECORDCHK(&so->so_rcv, "somove 2");
1475 	SBLASTMBUFCHK(&so->so_rcv, "somove 2");
1476 
1477 	/* Take at most len mbufs out of receive buffer. */
1478 	for (off = 0, mp = &m; off <= len && *mp;
1479 	    off += (*mp)->m_len, mp = &(*mp)->m_next) {
1480 		u_long size = len - off;
1481 
1482 #ifdef DIAGNOSTIC
1483 		if ((*mp)->m_type != MT_DATA && (*mp)->m_type != MT_HEADER)
1484 			panic("somove type: so %p, so_type %d, m %p, "
1485 			    "m_type %d", so, so->so_type, *mp, (*mp)->m_type);
1486 #endif
1487 		if ((*mp)->m_len > size) {
1488 			/*
1489 			 * Move only a partial mbuf at maximum splice length or
1490 			 * if the drain buffer is too small for this large mbuf.
1491 			 */
1492 			if (!maxreached && so->so_snd.sb_datacc > 0) {
1493 				len -= size;
1494 				break;
1495 			}
1496 			*mp = m_copym(so->so_rcv.sb_mb, 0, size, wait);
1497 			if (*mp == NULL) {
1498 				len -= size;
1499 				break;
1500 			}
1501 			so->so_rcv.sb_mb->m_data += size;
1502 			so->so_rcv.sb_mb->m_len -= size;
1503 			so->so_rcv.sb_cc -= size;
1504 			so->so_rcv.sb_datacc -= size;
1505 		} else {
1506 			*mp = so->so_rcv.sb_mb;
1507 			sbfree(&so->so_rcv, *mp);
1508 			so->so_rcv.sb_mb = (*mp)->m_next;
1509 			sbsync(&so->so_rcv, nextrecord);
1510 		}
1511 	}
1512 	*mp = NULL;
1513 
1514 	SBLASTRECORDCHK(&so->so_rcv, "somove 3");
1515 	SBLASTMBUFCHK(&so->so_rcv, "somove 3");
1516 	SBCHECK(&so->so_rcv);
1517 	if (m == NULL)
1518 		goto release;
1519 	m->m_nextpkt = NULL;
1520 	if (m->m_flags & M_PKTHDR) {
1521 		m_resethdr(m);
1522 		m->m_pkthdr.len = len;
1523 	}
1524 
1525 	/* Send window update to source peer as receive buffer has changed. */
1526 	if (so->so_proto->pr_flags & PR_WANTRCVD && so->so_pcb)
1527 		(so->so_proto->pr_usrreq)(so, PRU_RCVD, NULL,
1528 		    NULL, NULL, NULL);
1529 
1530 	/* Receive buffer did shrink by len bytes, adjust oob. */
1531 	state = so->so_state;
1532 	so->so_state &= ~SS_RCVATMARK;
1533 	oobmark = so->so_oobmark;
1534 	so->so_oobmark = oobmark > len ? oobmark - len : 0;
1535 	if (oobmark) {
1536 		if (oobmark == len)
1537 			so->so_state |= SS_RCVATMARK;
1538 		if (oobmark >= len)
1539 			oobmark = 0;
1540 	}
1541 
1542 	/*
1543 	 * Handle oob data.  If any malloc fails, ignore error.
1544 	 * TCP urgent data is not very reliable anyway.
1545 	 */
1546 	while (((state & SS_RCVATMARK) || oobmark) &&
1547 	    (so->so_options & SO_OOBINLINE)) {
1548 		struct mbuf *o = NULL;
1549 
1550 		if (state & SS_RCVATMARK) {
1551 			o = m_get(wait, MT_DATA);
1552 			state &= ~SS_RCVATMARK;
1553 		} else if (oobmark) {
1554 			o = m_split(m, oobmark, wait);
1555 			if (o) {
1556 				error = (*sosp->so_proto->pr_usrreq)(sosp,
1557 				    PRU_SEND, m, NULL, NULL, NULL);
1558 				if (error) {
1559 					if (sosp->so_state & SS_CANTSENDMORE)
1560 						error = EPIPE;
1561 					m_freem(o);
1562 					goto release;
1563 				}
1564 				len -= oobmark;
1565 				so->so_splicelen += oobmark;
1566 				m = o;
1567 				o = m_get(wait, MT_DATA);
1568 			}
1569 			oobmark = 0;
1570 		}
1571 		if (o) {
1572 			o->m_len = 1;
1573 			*mtod(o, caddr_t) = *mtod(m, caddr_t);
1574 			error = (*sosp->so_proto->pr_usrreq)(sosp, PRU_SENDOOB,
1575 			    o, NULL, NULL, NULL);
1576 			if (error) {
1577 				if (sosp->so_state & SS_CANTSENDMORE)
1578 					error = EPIPE;
1579 				m_freem(m);
1580 				goto release;
1581 			}
1582 			len -= 1;
1583 			so->so_splicelen += 1;
1584 			if (oobmark) {
1585 				oobmark -= 1;
1586 				if (oobmark == 0)
1587 					state |= SS_RCVATMARK;
1588 			}
1589 			m_adj(m, 1);
1590 		}
1591 	}
1592 
1593 	/* Append all remaining data to drain socket. */
1594 	if (so->so_rcv.sb_cc == 0 || maxreached)
1595 		sosp->so_state &= ~SS_ISSENDING;
1596 	error = (*sosp->so_proto->pr_usrreq)(sosp, PRU_SEND, m, NULL, NULL,
1597 	    NULL);
1598 	if (error) {
1599 		if (sosp->so_state & SS_CANTSENDMORE)
1600 			error = EPIPE;
1601 		goto release;
1602 	}
1603 	so->so_splicelen += len;
1604 
1605 	/* Move several packets if possible. */
1606 	if (!maxreached && nextrecord)
1607 		goto nextpkt;
1608 
1609  release:
1610 	sosp->so_state &= ~SS_ISSENDING;
1611 	if (!error && maxreached && so->so_splicemax == so->so_splicelen)
1612 		error = EFBIG;
1613 	if (error)
1614 		so->so_error = error;
1615 	if (((so->so_state & SS_CANTRCVMORE) && so->so_rcv.sb_cc == 0) ||
1616 	    (sosp->so_state & SS_CANTSENDMORE) || maxreached || error) {
1617 		sounsplice(so, sosp, 0);
1618 		return (0);
1619 	}
1620 	if (timerisset(&so->so_idletv))
1621 		timeout_add_tv(&so->so_idleto, &so->so_idletv);
1622 	return (1);
1623 }
1624 
1625 #endif /* SOCKET_SPLICE */
1626 
1627 void
1628 sorwakeup(struct socket *so)
1629 {
1630 	soassertlocked(so);
1631 
1632 #ifdef SOCKET_SPLICE
1633 	if (so->so_rcv.sb_flags & SB_SPLICE) {
1634 		/*
1635 		 * TCP has a sendbuffer that can handle multiple packets
1636 		 * at once.  So queue the stream a bit to accumulate data.
1637 		 * The sosplice thread will call somove() later and send
1638 		 * the packets calling tcp_output() only once.
1639 		 * In the UDP case, send out the packets immediately.
1640 		 * Using a thread would make things slower.
1641 		 */
1642 		if (so->so_proto->pr_flags & PR_WANTRCVD)
1643 			task_add(sosplice_taskq, &so->so_splicetask);
1644 		else
1645 			somove(so, M_DONTWAIT);
1646 	}
1647 	if (isspliced(so))
1648 		return;
1649 #endif
1650 	sowakeup(so, &so->so_rcv);
1651 	if (so->so_upcall)
1652 		(*(so->so_upcall))(so, so->so_upcallarg, M_DONTWAIT);
1653 }
1654 
1655 void
1656 sowwakeup(struct socket *so)
1657 {
1658 	soassertlocked(so);
1659 
1660 #ifdef SOCKET_SPLICE
1661 	if (so->so_snd.sb_flags & SB_SPLICE)
1662 		task_add(sosplice_taskq, &so->so_sp->ssp_soback->so_splicetask);
1663 	if (issplicedback(so))
1664 		return;
1665 #endif
1666 	sowakeup(so, &so->so_snd);
1667 }
1668 
1669 int
1670 sosetopt(struct socket *so, int level, int optname, struct mbuf *m)
1671 {
1672 	int error = 0;
1673 
1674 	soassertlocked(so);
1675 
1676 	if (level != SOL_SOCKET) {
1677 		if (so->so_proto->pr_ctloutput) {
1678 			error = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so,
1679 			    level, optname, m);
1680 			return (error);
1681 		}
1682 		error = ENOPROTOOPT;
1683 	} else {
1684 		switch (optname) {
1685 		case SO_BINDANY:
1686 			if ((error = suser(curproc)) != 0)	/* XXX */
1687 				return (error);
1688 			break;
1689 		}
1690 
1691 		switch (optname) {
1692 
1693 		case SO_LINGER:
1694 			if (m == NULL || m->m_len != sizeof (struct linger) ||
1695 			    mtod(m, struct linger *)->l_linger < 0 ||
1696 			    mtod(m, struct linger *)->l_linger > SHRT_MAX)
1697 				return (EINVAL);
1698 			so->so_linger = mtod(m, struct linger *)->l_linger;
1699 			/* FALLTHROUGH */
1700 
1701 		case SO_BINDANY:
1702 		case SO_DEBUG:
1703 		case SO_KEEPALIVE:
1704 		case SO_USELOOPBACK:
1705 		case SO_BROADCAST:
1706 		case SO_REUSEADDR:
1707 		case SO_REUSEPORT:
1708 		case SO_OOBINLINE:
1709 		case SO_TIMESTAMP:
1710 		case SO_ZEROIZE:
1711 			if (m == NULL || m->m_len < sizeof (int))
1712 				return (EINVAL);
1713 			if (*mtod(m, int *))
1714 				so->so_options |= optname;
1715 			else
1716 				so->so_options &= ~optname;
1717 			break;
1718 
1719 		case SO_DONTROUTE:
1720 			if (m == NULL || m->m_len < sizeof (int))
1721 				return (EINVAL);
1722 			if (*mtod(m, int *))
1723 				error = EOPNOTSUPP;
1724 			break;
1725 
1726 		case SO_SNDBUF:
1727 		case SO_RCVBUF:
1728 		case SO_SNDLOWAT:
1729 		case SO_RCVLOWAT:
1730 		    {
1731 			u_long cnt;
1732 
1733 			if (m == NULL || m->m_len < sizeof (int))
1734 				return (EINVAL);
1735 			cnt = *mtod(m, int *);
1736 			if ((long)cnt <= 0)
1737 				cnt = 1;
1738 			switch (optname) {
1739 
1740 			case SO_SNDBUF:
1741 				if (so->so_state & SS_CANTSENDMORE)
1742 					return (EINVAL);
1743 				if (sbcheckreserve(cnt, so->so_snd.sb_wat) ||
1744 				    sbreserve(so, &so->so_snd, cnt))
1745 					return (ENOBUFS);
1746 				so->so_snd.sb_wat = cnt;
1747 				break;
1748 
1749 			case SO_RCVBUF:
1750 				if (so->so_state & SS_CANTRCVMORE)
1751 					return (EINVAL);
1752 				if (sbcheckreserve(cnt, so->so_rcv.sb_wat) ||
1753 				    sbreserve(so, &so->so_rcv, cnt))
1754 					return (ENOBUFS);
1755 				so->so_rcv.sb_wat = cnt;
1756 				break;
1757 
1758 			case SO_SNDLOWAT:
1759 				so->so_snd.sb_lowat =
1760 				    (cnt > so->so_snd.sb_hiwat) ?
1761 				    so->so_snd.sb_hiwat : cnt;
1762 				break;
1763 			case SO_RCVLOWAT:
1764 				so->so_rcv.sb_lowat =
1765 				    (cnt > so->so_rcv.sb_hiwat) ?
1766 				    so->so_rcv.sb_hiwat : cnt;
1767 				break;
1768 			}
1769 			break;
1770 		    }
1771 
1772 		case SO_SNDTIMEO:
1773 		case SO_RCVTIMEO:
1774 		    {
1775 			struct timeval tv;
1776 			uint64_t nsecs;
1777 
1778 			if (m == NULL || m->m_len < sizeof (tv))
1779 				return (EINVAL);
1780 			memcpy(&tv, mtod(m, struct timeval *), sizeof tv);
1781 			if (!timerisvalid(&tv))
1782 				return (EINVAL);
1783 			nsecs = TIMEVAL_TO_NSEC(&tv);
1784 			if (nsecs == UINT64_MAX)
1785 				return (EDOM);
1786 			if (nsecs == 0)
1787 				nsecs = INFSLP;
1788 			switch (optname) {
1789 
1790 			case SO_SNDTIMEO:
1791 				so->so_snd.sb_timeo_nsecs = nsecs;
1792 				break;
1793 			case SO_RCVTIMEO:
1794 				so->so_rcv.sb_timeo_nsecs = nsecs;
1795 				break;
1796 			}
1797 			break;
1798 		    }
1799 
1800 		case SO_RTABLE:
1801 			if (so->so_proto->pr_domain &&
1802 			    so->so_proto->pr_domain->dom_protosw &&
1803 			    so->so_proto->pr_ctloutput) {
1804 				struct domain *dom = so->so_proto->pr_domain;
1805 
1806 				level = dom->dom_protosw->pr_protocol;
1807 				error = (*so->so_proto->pr_ctloutput)
1808 				    (PRCO_SETOPT, so, level, optname, m);
1809 				return (error);
1810 			}
1811 			error = ENOPROTOOPT;
1812 			break;
1813 
1814 #ifdef SOCKET_SPLICE
1815 		case SO_SPLICE:
1816 			if (m == NULL) {
1817 				error = sosplice(so, -1, 0, NULL);
1818 			} else if (m->m_len < sizeof(int)) {
1819 				return (EINVAL);
1820 			} else if (m->m_len < sizeof(struct splice)) {
1821 				error = sosplice(so, *mtod(m, int *), 0, NULL);
1822 			} else {
1823 				error = sosplice(so,
1824 				    mtod(m, struct splice *)->sp_fd,
1825 				    mtod(m, struct splice *)->sp_max,
1826 				   &mtod(m, struct splice *)->sp_idle);
1827 			}
1828 			break;
1829 #endif /* SOCKET_SPLICE */
1830 
1831 		default:
1832 			error = ENOPROTOOPT;
1833 			break;
1834 		}
1835 		if (error == 0 && so->so_proto->pr_ctloutput) {
1836 			(*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so,
1837 			    level, optname, m);
1838 		}
1839 	}
1840 
1841 	return (error);
1842 }
1843 
1844 int
1845 sogetopt(struct socket *so, int level, int optname, struct mbuf *m)
1846 {
1847 	int error = 0;
1848 
1849 	soassertlocked(so);
1850 
1851 	if (level != SOL_SOCKET) {
1852 		if (so->so_proto->pr_ctloutput) {
1853 			m->m_len = 0;
1854 
1855 			error = (*so->so_proto->pr_ctloutput)(PRCO_GETOPT, so,
1856 			    level, optname, m);
1857 			if (error)
1858 				return (error);
1859 			return (0);
1860 		} else
1861 			return (ENOPROTOOPT);
1862 	} else {
1863 		m->m_len = sizeof (int);
1864 
1865 		switch (optname) {
1866 
1867 		case SO_LINGER:
1868 			m->m_len = sizeof (struct linger);
1869 			mtod(m, struct linger *)->l_onoff =
1870 				so->so_options & SO_LINGER;
1871 			mtod(m, struct linger *)->l_linger = so->so_linger;
1872 			break;
1873 
1874 		case SO_BINDANY:
1875 		case SO_USELOOPBACK:
1876 		case SO_DEBUG:
1877 		case SO_KEEPALIVE:
1878 		case SO_REUSEADDR:
1879 		case SO_REUSEPORT:
1880 		case SO_BROADCAST:
1881 		case SO_OOBINLINE:
1882 		case SO_TIMESTAMP:
1883 		case SO_ZEROIZE:
1884 			*mtod(m, int *) = so->so_options & optname;
1885 			break;
1886 
1887 		case SO_DONTROUTE:
1888 			*mtod(m, int *) = 0;
1889 			break;
1890 
1891 		case SO_TYPE:
1892 			*mtod(m, int *) = so->so_type;
1893 			break;
1894 
1895 		case SO_ERROR:
1896 			*mtod(m, int *) = so->so_error;
1897 			so->so_error = 0;
1898 			break;
1899 
1900 		case SO_DOMAIN:
1901 			*mtod(m, int *) = so->so_proto->pr_domain->dom_family;
1902 			break;
1903 
1904 		case SO_PROTOCOL:
1905 			*mtod(m, int *) = so->so_proto->pr_protocol;
1906 			break;
1907 
1908 		case SO_SNDBUF:
1909 			*mtod(m, int *) = so->so_snd.sb_hiwat;
1910 			break;
1911 
1912 		case SO_RCVBUF:
1913 			*mtod(m, int *) = so->so_rcv.sb_hiwat;
1914 			break;
1915 
1916 		case SO_SNDLOWAT:
1917 			*mtod(m, int *) = so->so_snd.sb_lowat;
1918 			break;
1919 
1920 		case SO_RCVLOWAT:
1921 			*mtod(m, int *) = so->so_rcv.sb_lowat;
1922 			break;
1923 
1924 		case SO_SNDTIMEO:
1925 		case SO_RCVTIMEO:
1926 		    {
1927 			struct timeval tv;
1928 			uint64_t nsecs = (optname == SO_SNDTIMEO ?
1929 			    so->so_snd.sb_timeo_nsecs :
1930 			    so->so_rcv.sb_timeo_nsecs);
1931 
1932 			m->m_len = sizeof(struct timeval);
1933 			memset(&tv, 0, sizeof(tv));
1934 			if (nsecs != INFSLP)
1935 				NSEC_TO_TIMEVAL(nsecs, &tv);
1936 			memcpy(mtod(m, struct timeval *), &tv, sizeof tv);
1937 			break;
1938 		    }
1939 
1940 		case SO_RTABLE:
1941 			if (so->so_proto->pr_domain &&
1942 			    so->so_proto->pr_domain->dom_protosw &&
1943 			    so->so_proto->pr_ctloutput) {
1944 				struct domain *dom = so->so_proto->pr_domain;
1945 
1946 				level = dom->dom_protosw->pr_protocol;
1947 				error = (*so->so_proto->pr_ctloutput)
1948 				    (PRCO_GETOPT, so, level, optname, m);
1949 				if (error)
1950 					return (error);
1951 				break;
1952 			}
1953 			return (ENOPROTOOPT);
1954 
1955 #ifdef SOCKET_SPLICE
1956 		case SO_SPLICE:
1957 		    {
1958 			off_t len;
1959 
1960 			m->m_len = sizeof(off_t);
1961 			len = so->so_sp ? so->so_sp->ssp_len : 0;
1962 			memcpy(mtod(m, off_t *), &len, sizeof(off_t));
1963 			break;
1964 		    }
1965 #endif /* SOCKET_SPLICE */
1966 
1967 		case SO_PEERCRED:
1968 			if (so->so_proto->pr_protocol == AF_UNIX) {
1969 				struct unpcb *unp = sotounpcb(so);
1970 
1971 				if (unp->unp_flags & UNP_FEIDS) {
1972 					m->m_len = sizeof(unp->unp_connid);
1973 					memcpy(mtod(m, caddr_t),
1974 					    &(unp->unp_connid), m->m_len);
1975 					break;
1976 				}
1977 				return (ENOTCONN);
1978 			}
1979 			return (EOPNOTSUPP);
1980 
1981 		default:
1982 			return (ENOPROTOOPT);
1983 		}
1984 		return (0);
1985 	}
1986 }
1987 
1988 void
1989 sohasoutofband(struct socket *so)
1990 {
1991 	pgsigio(&so->so_sigio, SIGURG, 0);
1992 	selwakeup(&so->so_rcv.sb_sel);
1993 }
1994 
1995 int
1996 soo_kqfilter(struct file *fp, struct knote *kn)
1997 {
1998 	struct socket *so = kn->kn_fp->f_data;
1999 	struct sockbuf *sb;
2000 
2001 	KERNEL_ASSERT_LOCKED();
2002 
2003 	switch (kn->kn_filter) {
2004 	case EVFILT_READ:
2005 		if (so->so_options & SO_ACCEPTCONN)
2006 			kn->kn_fop = &solisten_filtops;
2007 		else
2008 			kn->kn_fop = &soread_filtops;
2009 		sb = &so->so_rcv;
2010 		break;
2011 	case EVFILT_WRITE:
2012 		kn->kn_fop = &sowrite_filtops;
2013 		sb = &so->so_snd;
2014 		break;
2015 	default:
2016 		return (EINVAL);
2017 	}
2018 
2019 	SLIST_INSERT_HEAD(&sb->sb_sel.si_note, kn, kn_selnext);
2020 	sb->sb_flagsintr |= SB_KNOTE;
2021 
2022 	return (0);
2023 }
2024 
2025 void
2026 filt_sordetach(struct knote *kn)
2027 {
2028 	struct socket *so = kn->kn_fp->f_data;
2029 
2030 	KERNEL_ASSERT_LOCKED();
2031 
2032 	SLIST_REMOVE(&so->so_rcv.sb_sel.si_note, kn, knote, kn_selnext);
2033 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.si_note))
2034 		so->so_rcv.sb_flagsintr &= ~SB_KNOTE;
2035 }
2036 
2037 int
2038 filt_soread(struct knote *kn, long hint)
2039 {
2040 	struct socket *so = kn->kn_fp->f_data;
2041 	int s, rv;
2042 
2043 	if ((hint & NOTE_SUBMIT) == 0)
2044 		s = solock(so);
2045 	kn->kn_data = so->so_rcv.sb_cc;
2046 #ifdef SOCKET_SPLICE
2047 	if (isspliced(so)) {
2048 		rv = 0;
2049 	} else
2050 #endif /* SOCKET_SPLICE */
2051 	if (so->so_state & SS_CANTRCVMORE) {
2052 		kn->kn_flags |= EV_EOF;
2053 		kn->kn_fflags = so->so_error;
2054 		rv = 1;
2055 	} else if (so->so_error) {	/* temporary udp error */
2056 		rv = 1;
2057 	} else if (kn->kn_sfflags & NOTE_LOWAT) {
2058 		rv = (kn->kn_data >= kn->kn_sdata);
2059 	} else {
2060 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
2061 	}
2062 	if ((hint & NOTE_SUBMIT) == 0)
2063 		sounlock(so, s);
2064 
2065 	return rv;
2066 }
2067 
2068 void
2069 filt_sowdetach(struct knote *kn)
2070 {
2071 	struct socket *so = kn->kn_fp->f_data;
2072 
2073 	KERNEL_ASSERT_LOCKED();
2074 
2075 	SLIST_REMOVE(&so->so_snd.sb_sel.si_note, kn, knote, kn_selnext);
2076 	if (SLIST_EMPTY(&so->so_snd.sb_sel.si_note))
2077 		so->so_snd.sb_flagsintr &= ~SB_KNOTE;
2078 }
2079 
2080 int
2081 filt_sowrite(struct knote *kn, long hint)
2082 {
2083 	struct socket *so = kn->kn_fp->f_data;
2084 	int s, rv;
2085 
2086 	if ((hint & NOTE_SUBMIT) == 0)
2087 		s = solock(so);
2088 	kn->kn_data = sbspace(so, &so->so_snd);
2089 	if (so->so_state & SS_CANTSENDMORE) {
2090 		kn->kn_flags |= EV_EOF;
2091 		kn->kn_fflags = so->so_error;
2092 		rv = 1;
2093 	} else if (so->so_error) {	/* temporary udp error */
2094 		rv = 1;
2095 	} else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2096 	    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
2097 		rv = 0;
2098 	} else if (kn->kn_sfflags & NOTE_LOWAT) {
2099 		rv = (kn->kn_data >= kn->kn_sdata);
2100 	} else {
2101 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
2102 	}
2103 	if ((hint & NOTE_SUBMIT) == 0)
2104 		sounlock(so, s);
2105 
2106 	return (rv);
2107 }
2108 
2109 int
2110 filt_solisten(struct knote *kn, long hint)
2111 {
2112 	struct socket *so = kn->kn_fp->f_data;
2113 	int s;
2114 
2115 	if ((hint & NOTE_SUBMIT) == 0)
2116 		s = solock(so);
2117 	kn->kn_data = so->so_qlen;
2118 	if ((hint & NOTE_SUBMIT) == 0)
2119 		sounlock(so, s);
2120 
2121 	return (kn->kn_data != 0);
2122 }
2123 
2124 #ifdef DDB
2125 void
2126 sobuf_print(struct sockbuf *,
2127     int (*)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))));
2128 
2129 void
2130 sobuf_print(struct sockbuf *sb,
2131     int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2132 {
2133 	(*pr)("\tsb_cc: %lu\n", sb->sb_cc);
2134 	(*pr)("\tsb_datacc: %lu\n", sb->sb_datacc);
2135 	(*pr)("\tsb_hiwat: %lu\n", sb->sb_hiwat);
2136 	(*pr)("\tsb_wat: %lu\n", sb->sb_wat);
2137 	(*pr)("\tsb_mbcnt: %lu\n", sb->sb_mbcnt);
2138 	(*pr)("\tsb_mbmax: %lu\n", sb->sb_mbmax);
2139 	(*pr)("\tsb_lowat: %ld\n", sb->sb_lowat);
2140 	(*pr)("\tsb_mb: %p\n", sb->sb_mb);
2141 	(*pr)("\tsb_mbtail: %p\n", sb->sb_mbtail);
2142 	(*pr)("\tsb_lastrecord: %p\n", sb->sb_lastrecord);
2143 	(*pr)("\tsb_sel: ...\n");
2144 	(*pr)("\tsb_flagsintr: %d\n", sb->sb_flagsintr);
2145 	(*pr)("\tsb_flags: %i\n", sb->sb_flags);
2146 	(*pr)("\tsb_timeo_nsecs: %llu\n", sb->sb_timeo_nsecs);
2147 }
2148 
2149 void
2150 so_print(void *v,
2151     int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2152 {
2153 	struct socket *so = v;
2154 
2155 	(*pr)("socket %p\n", so);
2156 	(*pr)("so_type: %i\n", so->so_type);
2157 	(*pr)("so_options: 0x%04x\n", so->so_options); /* %b */
2158 	(*pr)("so_linger: %i\n", so->so_linger);
2159 	(*pr)("so_state: 0x%04x\n", so->so_state);
2160 	(*pr)("so_pcb: %p\n", so->so_pcb);
2161 	(*pr)("so_proto: %p\n", so->so_proto);
2162 	(*pr)("so_sigio: %p\n", so->so_sigio.sir_sigio);
2163 
2164 	(*pr)("so_head: %p\n", so->so_head);
2165 	(*pr)("so_onq: %p\n", so->so_onq);
2166 	(*pr)("so_q0: @%p first: %p\n", &so->so_q0, TAILQ_FIRST(&so->so_q0));
2167 	(*pr)("so_q: @%p first: %p\n", &so->so_q, TAILQ_FIRST(&so->so_q));
2168 	(*pr)("so_eq: next: %p\n", TAILQ_NEXT(so, so_qe));
2169 	(*pr)("so_q0len: %i\n", so->so_q0len);
2170 	(*pr)("so_qlen: %i\n", so->so_qlen);
2171 	(*pr)("so_qlimit: %i\n", so->so_qlimit);
2172 	(*pr)("so_timeo: %i\n", so->so_timeo);
2173 	(*pr)("so_obmark: %lu\n", so->so_oobmark);
2174 
2175 	(*pr)("so_sp: %p\n", so->so_sp);
2176 	if (so->so_sp != NULL) {
2177 		(*pr)("\tssp_socket: %p\n", so->so_sp->ssp_socket);
2178 		(*pr)("\tssp_soback: %p\n", so->so_sp->ssp_soback);
2179 		(*pr)("\tssp_len: %lld\n",
2180 		    (unsigned long long)so->so_sp->ssp_len);
2181 		(*pr)("\tssp_max: %lld\n",
2182 		    (unsigned long long)so->so_sp->ssp_max);
2183 		(*pr)("\tssp_idletv: %lld %ld\n", so->so_sp->ssp_idletv.tv_sec,
2184 		    so->so_sp->ssp_idletv.tv_usec);
2185 		(*pr)("\tssp_idleto: %spending (@%i)\n",
2186 		    timeout_pending(&so->so_sp->ssp_idleto) ? "" : "not ",
2187 		    so->so_sp->ssp_idleto.to_time);
2188 	}
2189 
2190 	(*pr)("so_rcv:\n");
2191 	sobuf_print(&so->so_rcv, pr);
2192 	(*pr)("so_snd:\n");
2193 	sobuf_print(&so->so_snd, pr);
2194 
2195 	(*pr)("so_upcall: %p so_upcallarg: %p\n",
2196 	    so->so_upcall, so->so_upcallarg);
2197 
2198 	(*pr)("so_euid: %d so_ruid: %d\n", so->so_euid, so->so_ruid);
2199 	(*pr)("so_egid: %d so_rgid: %d\n", so->so_egid, so->so_rgid);
2200 	(*pr)("so_cpid: %d\n", so->so_cpid);
2201 }
2202 #endif
2203