xref: /openbsd-src/sys/kern/uipc_socket.c (revision 46035553bfdd96e63c94e32da0210227ec2e3cf1)
1 /*	$OpenBSD: uipc_socket.c,v 1.252 2020/12/25 12:59:52 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 #include <sys/time.h>
55 
56 #ifdef DDB
57 #include <machine/db_machdep.h>
58 #endif
59 
60 void	sbsync(struct sockbuf *, struct mbuf *);
61 
62 int	sosplice(struct socket *, int, off_t, struct timeval *);
63 void	sounsplice(struct socket *, struct socket *, int);
64 void	soidle(void *);
65 void	sotask(void *);
66 void	soreaper(void *);
67 void	soput(void *);
68 int	somove(struct socket *, int);
69 
70 void	filt_sordetach(struct knote *kn);
71 int	filt_soread(struct knote *kn, long hint);
72 void	filt_sowdetach(struct knote *kn);
73 int	filt_sowrite(struct knote *kn, long hint);
74 int	filt_solisten(struct knote *kn, long hint);
75 
76 const struct filterops solisten_filtops = {
77 	.f_flags	= FILTEROP_ISFD,
78 	.f_attach	= NULL,
79 	.f_detach	= filt_sordetach,
80 	.f_event	= filt_solisten,
81 };
82 
83 const struct filterops soread_filtops = {
84 	.f_flags	= FILTEROP_ISFD,
85 	.f_attach	= NULL,
86 	.f_detach	= filt_sordetach,
87 	.f_event	= filt_soread,
88 };
89 
90 const struct filterops sowrite_filtops = {
91 	.f_flags	= FILTEROP_ISFD,
92 	.f_attach	= NULL,
93 	.f_detach	= filt_sowdetach,
94 	.f_event	= filt_sowrite,
95 };
96 
97 const struct filterops soexcept_filtops = {
98 	.f_flags	= FILTEROP_ISFD,
99 	.f_attach	= NULL,
100 	.f_detach	= filt_sordetach,
101 	.f_event	= filt_soread,
102 };
103 
104 #ifndef SOMINCONN
105 #define SOMINCONN 80
106 #endif /* SOMINCONN */
107 
108 int	somaxconn = SOMAXCONN;
109 int	sominconn = SOMINCONN;
110 
111 struct pool socket_pool;
112 #ifdef SOCKET_SPLICE
113 struct pool sosplice_pool;
114 struct taskq *sosplice_taskq;
115 struct rwlock sosplice_lock = RWLOCK_INITIALIZER("sosplicelk");
116 #endif
117 
118 void
119 soinit(void)
120 {
121 	pool_init(&socket_pool, sizeof(struct socket), 0, IPL_SOFTNET, 0,
122 	    "sockpl", NULL);
123 #ifdef SOCKET_SPLICE
124 	pool_init(&sosplice_pool, sizeof(struct sosplice), 0, IPL_SOFTNET, 0,
125 	    "sosppl", NULL);
126 #endif
127 }
128 
129 /*
130  * Socket operation routines.
131  * These routines are called by the routines in
132  * sys_socket.c or from a system process, and
133  * implement the semantics of socket operations by
134  * switching out to the protocol specific routines.
135  */
136 int
137 socreate(int dom, struct socket **aso, int type, int proto)
138 {
139 	struct proc *p = curproc;		/* XXX */
140 	const struct protosw *prp;
141 	struct socket *so;
142 	int error, s;
143 
144 	if (proto)
145 		prp = pffindproto(dom, proto, type);
146 	else
147 		prp = pffindtype(dom, type);
148 	if (prp == NULL || prp->pr_attach == NULL)
149 		return (EPROTONOSUPPORT);
150 	if (prp->pr_type != type)
151 		return (EPROTOTYPE);
152 	so = pool_get(&socket_pool, PR_WAITOK | PR_ZERO);
153 	sigio_init(&so->so_sigio);
154 	TAILQ_INIT(&so->so_q0);
155 	TAILQ_INIT(&so->so_q);
156 	so->so_type = type;
157 	if (suser(p) == 0)
158 		so->so_state = SS_PRIV;
159 	so->so_ruid = p->p_ucred->cr_ruid;
160 	so->so_euid = p->p_ucred->cr_uid;
161 	so->so_rgid = p->p_ucred->cr_rgid;
162 	so->so_egid = p->p_ucred->cr_gid;
163 	so->so_cpid = p->p_p->ps_pid;
164 	so->so_proto = prp;
165 	so->so_snd.sb_timeo_nsecs = INFSLP;
166 	so->so_rcv.sb_timeo_nsecs = INFSLP;
167 
168 	s = solock(so);
169 	error = (*prp->pr_attach)(so, proto);
170 	if (error) {
171 		so->so_state |= SS_NOFDREF;
172 		/* sofree() calls sounlock(). */
173 		sofree(so, s);
174 		return (error);
175 	}
176 	sounlock(so, s);
177 	*aso = so;
178 	return (0);
179 }
180 
181 int
182 sobind(struct socket *so, struct mbuf *nam, struct proc *p)
183 {
184 	int error;
185 
186 	soassertlocked(so);
187 
188 	error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, NULL, nam, NULL, p);
189 	return (error);
190 }
191 
192 int
193 solisten(struct socket *so, int backlog)
194 {
195 	int error;
196 
197 	soassertlocked(so);
198 
199 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING|SS_ISDISCONNECTING))
200 		return (EINVAL);
201 #ifdef SOCKET_SPLICE
202 	if (isspliced(so) || issplicedback(so))
203 		return (EOPNOTSUPP);
204 #endif /* SOCKET_SPLICE */
205 	error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, NULL, NULL, NULL,
206 	    curproc);
207 	if (error)
208 		return (error);
209 	if (TAILQ_FIRST(&so->so_q) == NULL)
210 		so->so_options |= SO_ACCEPTCONN;
211 	if (backlog < 0 || backlog > somaxconn)
212 		backlog = somaxconn;
213 	if (backlog < sominconn)
214 		backlog = sominconn;
215 	so->so_qlimit = backlog;
216 	return (0);
217 }
218 
219 #define SOSP_FREEING_READ	1
220 #define SOSP_FREEING_WRITE	2
221 void
222 sofree(struct socket *so, int s)
223 {
224 	soassertlocked(so);
225 
226 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
227 		sounlock(so, s);
228 		return;
229 	}
230 	if (so->so_head) {
231 		/*
232 		 * We must not decommission a socket that's on the accept(2)
233 		 * queue.  If we do, then accept(2) may hang after select(2)
234 		 * indicated that the listening socket was ready.
235 		 */
236 		if (!soqremque(so, 0)) {
237 			sounlock(so, s);
238 			return;
239 		}
240 	}
241 	sigio_free(&so->so_sigio);
242 #ifdef SOCKET_SPLICE
243 	if (so->so_sp) {
244 		if (issplicedback(so)) {
245 			int freeing = SOSP_FREEING_WRITE;
246 
247 			if (so->so_sp->ssp_soback == so)
248 				freeing |= SOSP_FREEING_READ;
249 			sounsplice(so->so_sp->ssp_soback, so, freeing);
250 		}
251 		if (isspliced(so)) {
252 			int freeing = SOSP_FREEING_READ;
253 
254 			if (so == so->so_sp->ssp_socket)
255 				freeing |= SOSP_FREEING_WRITE;
256 			sounsplice(so, so->so_sp->ssp_socket, freeing);
257 		}
258 	}
259 #endif /* SOCKET_SPLICE */
260 	sbrelease(so, &so->so_snd);
261 	sorflush(so);
262 	sounlock(so, s);
263 #ifdef SOCKET_SPLICE
264 	if (so->so_sp) {
265 		/* Reuse splice idle, sounsplice() has been called before. */
266 		timeout_set_proc(&so->so_sp->ssp_idleto, soreaper, so);
267 		timeout_add(&so->so_sp->ssp_idleto, 0);
268 	} else
269 #endif /* SOCKET_SPLICE */
270 	{
271 		pool_put(&socket_pool, so);
272 	}
273 }
274 
275 static inline uint64_t
276 solinger_nsec(struct socket *so)
277 {
278 	if (so->so_linger == 0)
279 		return INFSLP;
280 
281 	return SEC_TO_NSEC(so->so_linger);
282 }
283 
284 /*
285  * Close a socket on last file table reference removal.
286  * Initiate disconnect if connected.
287  * Free socket when disconnect complete.
288  */
289 int
290 soclose(struct socket *so, int flags)
291 {
292 	struct socket *so2;
293 	int s, error = 0;
294 
295 	s = solock(so);
296 	/* Revoke async IO early. There is a final revocation in sofree(). */
297 	sigio_free(&so->so_sigio);
298 	if (so->so_options & SO_ACCEPTCONN) {
299 		while ((so2 = TAILQ_FIRST(&so->so_q0)) != NULL) {
300 			(void) soqremque(so2, 0);
301 			(void) soabort(so2);
302 		}
303 		while ((so2 = TAILQ_FIRST(&so->so_q)) != NULL) {
304 			(void) soqremque(so2, 1);
305 			(void) soabort(so2);
306 		}
307 	}
308 	if (so->so_pcb == NULL)
309 		goto discard;
310 	if (so->so_state & SS_ISCONNECTED) {
311 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
312 			error = sodisconnect(so);
313 			if (error)
314 				goto drop;
315 		}
316 		if (so->so_options & SO_LINGER) {
317 			if ((so->so_state & SS_ISDISCONNECTING) &&
318 			    (flags & MSG_DONTWAIT))
319 				goto drop;
320 			while (so->so_state & SS_ISCONNECTED) {
321 				error = sosleep_nsec(so, &so->so_timeo,
322 				    PSOCK | PCATCH, "netcls",
323 				    solinger_nsec(so));
324 				if (error)
325 					break;
326 			}
327 		}
328 	}
329 drop:
330 	if (so->so_pcb) {
331 		int error2;
332 		KASSERT(so->so_proto->pr_detach);
333 		error2 = (*so->so_proto->pr_detach)(so);
334 		if (error == 0)
335 			error = error2;
336 	}
337 discard:
338 	if (so->so_state & SS_NOFDREF)
339 		panic("soclose NOFDREF: so %p, so_type %d", so, so->so_type);
340 	so->so_state |= SS_NOFDREF;
341 	/* sofree() calls sounlock(). */
342 	sofree(so, s);
343 	return (error);
344 }
345 
346 int
347 soabort(struct socket *so)
348 {
349 	soassertlocked(so);
350 
351 	return (*so->so_proto->pr_usrreq)(so, PRU_ABORT, NULL, NULL, NULL,
352 	   curproc);
353 }
354 
355 int
356 soaccept(struct socket *so, struct mbuf *nam)
357 {
358 	int error = 0;
359 
360 	soassertlocked(so);
361 
362 	if ((so->so_state & SS_NOFDREF) == 0)
363 		panic("soaccept !NOFDREF: so %p, so_type %d", so, so->so_type);
364 	so->so_state &= ~SS_NOFDREF;
365 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
366 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
367 		error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT, NULL,
368 		    nam, NULL, curproc);
369 	else
370 		error = ECONNABORTED;
371 	return (error);
372 }
373 
374 int
375 soconnect(struct socket *so, struct mbuf *nam)
376 {
377 	int error;
378 
379 	soassertlocked(so);
380 
381 	if (so->so_options & SO_ACCEPTCONN)
382 		return (EOPNOTSUPP);
383 	/*
384 	 * If protocol is connection-based, can only connect once.
385 	 * Otherwise, if connected, try to disconnect first.
386 	 * This allows user to disconnect by connecting to, e.g.,
387 	 * a null address.
388 	 */
389 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
390 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
391 	    (error = sodisconnect(so))))
392 		error = EISCONN;
393 	else
394 		error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT,
395 		    NULL, nam, NULL, curproc);
396 	return (error);
397 }
398 
399 int
400 soconnect2(struct socket *so1, struct socket *so2)
401 {
402 	int s, error;
403 
404 	s = solock(so1);
405 	error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2, NULL,
406 	    (struct mbuf *)so2, NULL, curproc);
407 	sounlock(so1, s);
408 	return (error);
409 }
410 
411 int
412 sodisconnect(struct socket *so)
413 {
414 	int error;
415 
416 	soassertlocked(so);
417 
418 	if ((so->so_state & SS_ISCONNECTED) == 0)
419 		return (ENOTCONN);
420 	if (so->so_state & SS_ISDISCONNECTING)
421 		return (EALREADY);
422 	error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT, NULL, NULL,
423 	    NULL, curproc);
424 	return (error);
425 }
426 
427 int m_getuio(struct mbuf **, int, long, struct uio *);
428 
429 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
430 /*
431  * Send on a socket.
432  * If send must go all at once and message is larger than
433  * send buffering, then hard error.
434  * Lock against other senders.
435  * If must go all at once and not enough room now, then
436  * inform user that this would block and do nothing.
437  * Otherwise, if nonblocking, send as much as possible.
438  * The data to be sent is described by "uio" if nonzero,
439  * otherwise by the mbuf chain "top" (which must be null
440  * if uio is not).  Data provided in mbuf chain must be small
441  * enough to send all at once.
442  *
443  * Returns nonzero on error, timeout or signal; callers
444  * must check for short counts if EINTR/ERESTART are returned.
445  * Data and control buffers are freed on return.
446  */
447 int
448 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
449     struct mbuf *control, int flags)
450 {
451 	long space, clen = 0;
452 	size_t resid;
453 	int error, s;
454 	int atomic = sosendallatonce(so) || top;
455 
456 	if (uio)
457 		resid = uio->uio_resid;
458 	else
459 		resid = top->m_pkthdr.len;
460 	/* MSG_EOR on a SOCK_STREAM socket is invalid. */
461 	if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
462 		m_freem(top);
463 		m_freem(control);
464 		return (EINVAL);
465 	}
466 	if (uio && uio->uio_procp)
467 		uio->uio_procp->p_ru.ru_msgsnd++;
468 	if (control) {
469 		/*
470 		 * In theory clen should be unsigned (since control->m_len is).
471 		 * However, space must be signed, as it might be less than 0
472 		 * if we over-committed, and we must use a signed comparison
473 		 * of space and clen.
474 		 */
475 		clen = control->m_len;
476 		/* reserve extra space for AF_UNIX's internalize */
477 		if (so->so_proto->pr_domain->dom_family == AF_UNIX &&
478 		    clen >= CMSG_ALIGN(sizeof(struct cmsghdr)) &&
479 		    mtod(control, struct cmsghdr *)->cmsg_type == SCM_RIGHTS)
480 			clen = CMSG_SPACE(
481 			    (clen - CMSG_ALIGN(sizeof(struct cmsghdr))) *
482 			    (sizeof(struct fdpass) / sizeof(int)));
483 	}
484 
485 #define	snderr(errno)	{ error = errno; goto release; }
486 
487 	s = solock(so);
488 restart:
489 	if ((error = sblock(so, &so->so_snd, SBLOCKWAIT(flags))) != 0)
490 		goto out;
491 	so->so_state |= SS_ISSENDING;
492 	do {
493 		if (so->so_state & SS_CANTSENDMORE)
494 			snderr(EPIPE);
495 		if (so->so_error) {
496 			error = so->so_error;
497 			so->so_error = 0;
498 			snderr(error);
499 		}
500 		if ((so->so_state & SS_ISCONNECTED) == 0) {
501 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
502 				if (!(resid == 0 && clen != 0))
503 					snderr(ENOTCONN);
504 			} else if (addr == 0)
505 				snderr(EDESTADDRREQ);
506 		}
507 		space = sbspace(so, &so->so_snd);
508 		if (flags & MSG_OOB)
509 			space += 1024;
510 		if (so->so_proto->pr_domain->dom_family == AF_UNIX) {
511 			if (atomic && resid > so->so_snd.sb_hiwat)
512 				snderr(EMSGSIZE);
513 		} else {
514 			if (clen > so->so_snd.sb_hiwat ||
515 			    (atomic && resid > so->so_snd.sb_hiwat - clen))
516 				snderr(EMSGSIZE);
517 		}
518 		if (space < clen ||
519 		    (space - clen < resid &&
520 		    (atomic || space < so->so_snd.sb_lowat))) {
521 			if (flags & MSG_DONTWAIT)
522 				snderr(EWOULDBLOCK);
523 			sbunlock(so, &so->so_snd);
524 			error = sbwait(so, &so->so_snd);
525 			so->so_state &= ~SS_ISSENDING;
526 			if (error)
527 				goto out;
528 			goto restart;
529 		}
530 		space -= clen;
531 		do {
532 			if (uio == NULL) {
533 				/*
534 				 * Data is prepackaged in "top".
535 				 */
536 				resid = 0;
537 				if (flags & MSG_EOR)
538 					top->m_flags |= M_EOR;
539 			} else {
540 				sounlock(so, s);
541 				error = m_getuio(&top, atomic, space, uio);
542 				s = solock(so);
543 				if (error)
544 					goto release;
545 				space -= top->m_pkthdr.len;
546 				resid = uio->uio_resid;
547 				if (flags & MSG_EOR)
548 					top->m_flags |= M_EOR;
549 			}
550 			if (resid == 0)
551 				so->so_state &= ~SS_ISSENDING;
552 			if (top && so->so_options & SO_ZEROIZE)
553 				top->m_flags |= M_ZEROIZE;
554 			error = (*so->so_proto->pr_usrreq)(so,
555 			    (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND,
556 			    top, addr, control, curproc);
557 			clen = 0;
558 			control = NULL;
559 			top = NULL;
560 			if (error)
561 				goto release;
562 		} while (resid && space > 0);
563 	} while (resid);
564 
565 release:
566 	so->so_state &= ~SS_ISSENDING;
567 	sbunlock(so, &so->so_snd);
568 out:
569 	sounlock(so, s);
570 	m_freem(top);
571 	m_freem(control);
572 	return (error);
573 }
574 
575 int
576 m_getuio(struct mbuf **mp, int atomic, long space, struct uio *uio)
577 {
578 	struct mbuf *m, *top = NULL;
579 	struct mbuf **nextp = &top;
580 	u_long len, mlen;
581 	size_t resid = uio->uio_resid;
582 	int error;
583 
584 	do {
585 		if (top == NULL) {
586 			MGETHDR(m, M_WAIT, MT_DATA);
587 			mlen = MHLEN;
588 			m->m_pkthdr.len = 0;
589 			m->m_pkthdr.ph_ifidx = 0;
590 		} else {
591 			MGET(m, M_WAIT, MT_DATA);
592 			mlen = MLEN;
593 		}
594 		/* chain mbuf together */
595 		*nextp = m;
596 		nextp = &m->m_next;
597 
598 		resid = ulmin(resid, space);
599 		if (resid >= MINCLSIZE) {
600 			MCLGETL(m, M_NOWAIT, ulmin(resid, MAXMCLBYTES));
601 			if ((m->m_flags & M_EXT) == 0)
602 				MCLGETL(m, M_NOWAIT, MCLBYTES);
603 			if ((m->m_flags & M_EXT) == 0)
604 				goto nopages;
605 			mlen = m->m_ext.ext_size;
606 			len = ulmin(mlen, resid);
607 			/*
608 			 * For datagram protocols, leave room
609 			 * for protocol headers in first mbuf.
610 			 */
611 			if (atomic && m == top && len < mlen - max_hdr)
612 				m->m_data += max_hdr;
613 		} else {
614 nopages:
615 			len = ulmin(mlen, resid);
616 			/*
617 			 * For datagram protocols, leave room
618 			 * for protocol headers in first mbuf.
619 			 */
620 			if (atomic && m == top && len < mlen - max_hdr)
621 				m_align(m, len);
622 		}
623 
624 		error = uiomove(mtod(m, caddr_t), len, uio);
625 		if (error) {
626 			m_freem(top);
627 			return (error);
628 		}
629 
630 		/* adjust counters */
631 		resid = uio->uio_resid;
632 		space -= len;
633 		m->m_len = len;
634 		top->m_pkthdr.len += len;
635 
636 		/* Is there more space and more data? */
637 	} while (space > 0 && resid > 0);
638 
639 	*mp = top;
640 	return 0;
641 }
642 
643 /*
644  * Following replacement or removal of the first mbuf on the first
645  * mbuf chain of a socket buffer, push necessary state changes back
646  * into the socket buffer so that other consumers see the values
647  * consistently.  'nextrecord' is the callers locally stored value of
648  * the original value of sb->sb_mb->m_nextpkt which must be restored
649  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
650  */
651 void
652 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
653 {
654 
655 	/*
656 	 * First, update for the new value of nextrecord.  If necessary,
657 	 * make it the first record.
658 	 */
659 	if (sb->sb_mb != NULL)
660 		sb->sb_mb->m_nextpkt = nextrecord;
661 	else
662 		sb->sb_mb = nextrecord;
663 
664 	/*
665 	 * Now update any dependent socket buffer fields to reflect
666 	 * the new state.  This is an inline of SB_EMPTY_FIXUP, with
667 	 * the addition of a second clause that takes care of the
668 	 * case where sb_mb has been updated, but remains the last
669 	 * record.
670 	 */
671 	if (sb->sb_mb == NULL) {
672 		sb->sb_mbtail = NULL;
673 		sb->sb_lastrecord = NULL;
674 	} else if (sb->sb_mb->m_nextpkt == NULL)
675 		sb->sb_lastrecord = sb->sb_mb;
676 }
677 
678 /*
679  * Implement receive operations on a socket.
680  * We depend on the way that records are added to the sockbuf
681  * by sbappend*.  In particular, each record (mbufs linked through m_next)
682  * must begin with an address if the protocol so specifies,
683  * followed by an optional mbuf or mbufs containing ancillary data,
684  * and then zero or more mbufs of data.
685  * In order to avoid blocking network for the entire time here, we release
686  * the solock() while doing the actual copy to user space.
687  * Although the sockbuf is locked, new data may still be appended,
688  * and thus we must maintain consistency of the sockbuf during that time.
689  *
690  * The caller may receive the data as a single mbuf chain by supplying
691  * an mbuf **mp0 for use in returning the chain.  The uio is then used
692  * only for the count in uio_resid.
693  */
694 int
695 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
696     struct mbuf **mp0, struct mbuf **controlp, int *flagsp,
697     socklen_t controllen)
698 {
699 	struct mbuf *m, **mp;
700 	struct mbuf *cm;
701 	u_long len, offset, moff;
702 	int flags, error, s, type, uio_error = 0;
703 	const struct protosw *pr = so->so_proto;
704 	struct mbuf *nextrecord;
705 	size_t resid, orig_resid = uio->uio_resid;
706 
707 	mp = mp0;
708 	if (paddr)
709 		*paddr = NULL;
710 	if (controlp)
711 		*controlp = NULL;
712 	if (flagsp)
713 		flags = *flagsp &~ MSG_EOR;
714 	else
715 		flags = 0;
716 	if (flags & MSG_OOB) {
717 		m = m_get(M_WAIT, MT_DATA);
718 		s = solock(so);
719 		error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m,
720 		    (struct mbuf *)(long)(flags & MSG_PEEK), NULL, curproc);
721 		sounlock(so, s);
722 		if (error)
723 			goto bad;
724 		do {
725 			error = uiomove(mtod(m, caddr_t),
726 			    ulmin(uio->uio_resid, m->m_len), uio);
727 			m = m_free(m);
728 		} while (uio->uio_resid && error == 0 && m);
729 bad:
730 		m_freem(m);
731 		return (error);
732 	}
733 	if (mp)
734 		*mp = NULL;
735 
736 	s = solock(so);
737 restart:
738 	if ((error = sblock(so, &so->so_rcv, SBLOCKWAIT(flags))) != 0) {
739 		sounlock(so, s);
740 		return (error);
741 	}
742 
743 	m = so->so_rcv.sb_mb;
744 #ifdef SOCKET_SPLICE
745 	if (isspliced(so))
746 		m = NULL;
747 #endif /* SOCKET_SPLICE */
748 	/*
749 	 * If we have less data than requested, block awaiting more
750 	 * (subject to any timeout) if:
751 	 *   1. the current count is less than the low water mark,
752 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
753 	 *	receive operation at once if we block (resid <= hiwat), or
754 	 *   3. MSG_DONTWAIT is not set.
755 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
756 	 * we have to do the receive in sections, and thus risk returning
757 	 * a short count if a timeout or signal occurs after we start.
758 	 */
759 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
760 	    so->so_rcv.sb_cc < uio->uio_resid) &&
761 	    (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
762 	    ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
763 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
764 #ifdef DIAGNOSTIC
765 		if (m == NULL && so->so_rcv.sb_cc)
766 #ifdef SOCKET_SPLICE
767 		    if (!isspliced(so))
768 #endif /* SOCKET_SPLICE */
769 			panic("receive 1: so %p, so_type %d, sb_cc %lu",
770 			    so, so->so_type, so->so_rcv.sb_cc);
771 #endif
772 		if (so->so_error) {
773 			if (m)
774 				goto dontblock;
775 			error = so->so_error;
776 			if ((flags & MSG_PEEK) == 0)
777 				so->so_error = 0;
778 			goto release;
779 		}
780 		if (so->so_state & SS_CANTRCVMORE) {
781 			if (m)
782 				goto dontblock;
783 			else if (so->so_rcv.sb_cc == 0)
784 				goto release;
785 		}
786 		for (; m; m = m->m_next)
787 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
788 				m = so->so_rcv.sb_mb;
789 				goto dontblock;
790 			}
791 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
792 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
793 			error = ENOTCONN;
794 			goto release;
795 		}
796 		if (uio->uio_resid == 0 && controlp == NULL)
797 			goto release;
798 		if (flags & MSG_DONTWAIT) {
799 			error = EWOULDBLOCK;
800 			goto release;
801 		}
802 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
803 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
804 		sbunlock(so, &so->so_rcv);
805 		error = sbwait(so, &so->so_rcv);
806 		if (error) {
807 			sounlock(so, s);
808 			return (error);
809 		}
810 		goto restart;
811 	}
812 dontblock:
813 	/*
814 	 * On entry here, m points to the first record of the socket buffer.
815 	 * From this point onward, we maintain 'nextrecord' as a cache of the
816 	 * pointer to the next record in the socket buffer.  We must keep the
817 	 * various socket buffer pointers and local stack versions of the
818 	 * pointers in sync, pushing out modifications before operations that
819 	 * may sleep, and re-reading them afterwards.
820 	 *
821 	 * Otherwise, we will race with the network stack appending new data
822 	 * or records onto the socket buffer by using inconsistent/stale
823 	 * versions of the field, possibly resulting in socket buffer
824 	 * corruption.
825 	 */
826 	if (uio->uio_procp)
827 		uio->uio_procp->p_ru.ru_msgrcv++;
828 	KASSERT(m == so->so_rcv.sb_mb);
829 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
830 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
831 	nextrecord = m->m_nextpkt;
832 	if (pr->pr_flags & PR_ADDR) {
833 #ifdef DIAGNOSTIC
834 		if (m->m_type != MT_SONAME)
835 			panic("receive 1a: so %p, so_type %d, m %p, m_type %d",
836 			    so, so->so_type, m, m->m_type);
837 #endif
838 		orig_resid = 0;
839 		if (flags & MSG_PEEK) {
840 			if (paddr)
841 				*paddr = m_copym(m, 0, m->m_len, M_NOWAIT);
842 			m = m->m_next;
843 		} else {
844 			sbfree(&so->so_rcv, m);
845 			if (paddr) {
846 				*paddr = m;
847 				so->so_rcv.sb_mb = m->m_next;
848 				m->m_next = 0;
849 				m = so->so_rcv.sb_mb;
850 			} else {
851 				so->so_rcv.sb_mb = m_free(m);
852 				m = so->so_rcv.sb_mb;
853 			}
854 			sbsync(&so->so_rcv, nextrecord);
855 		}
856 	}
857 	while (m && m->m_type == MT_CONTROL && error == 0) {
858 		int skip = 0;
859 		if (flags & MSG_PEEK) {
860 			if (mtod(m, struct cmsghdr *)->cmsg_type ==
861 			    SCM_RIGHTS) {
862 				/* don't leak internalized SCM_RIGHTS msgs */
863 				skip = 1;
864 			} else if (controlp)
865 				*controlp = m_copym(m, 0, m->m_len, M_NOWAIT);
866 			m = m->m_next;
867 		} else {
868 			sbfree(&so->so_rcv, m);
869 			so->so_rcv.sb_mb = m->m_next;
870 			m->m_nextpkt = m->m_next = NULL;
871 			cm = m;
872 			m = so->so_rcv.sb_mb;
873 			sbsync(&so->so_rcv, nextrecord);
874 			if (controlp) {
875 				if (pr->pr_domain->dom_externalize) {
876 					error =
877 					    (*pr->pr_domain->dom_externalize)
878 					    (cm, controllen, flags);
879 				}
880 				*controlp = cm;
881 			} else {
882 				/*
883 				 * Dispose of any SCM_RIGHTS message that went
884 				 * through the read path rather than recv.
885 				 */
886 				if (pr->pr_domain->dom_dispose)
887 					pr->pr_domain->dom_dispose(cm);
888 				m_free(cm);
889 			}
890 		}
891 		if (m != NULL)
892 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
893 		else
894 			nextrecord = so->so_rcv.sb_mb;
895 		if (controlp && !skip)
896 			controlp = &(*controlp)->m_next;
897 		orig_resid = 0;
898 	}
899 
900 	/* If m is non-NULL, we have some data to read. */
901 	if (m) {
902 		type = m->m_type;
903 		if (type == MT_OOBDATA)
904 			flags |= MSG_OOB;
905 		if (m->m_flags & M_BCAST)
906 			flags |= MSG_BCAST;
907 		if (m->m_flags & M_MCAST)
908 			flags |= MSG_MCAST;
909 	}
910 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
911 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
912 
913 	moff = 0;
914 	offset = 0;
915 	while (m && uio->uio_resid > 0 && error == 0) {
916 		if (m->m_type == MT_OOBDATA) {
917 			if (type != MT_OOBDATA)
918 				break;
919 		} else if (type == MT_OOBDATA) {
920 			break;
921 		} else if (m->m_type == MT_CONTROL) {
922 			/*
923 			 * If there is more than one control message in the
924 			 * stream, we do a short read.  Next can be received
925 			 * or disposed by another system call.
926 			 */
927 			break;
928 #ifdef DIAGNOSTIC
929 		} else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) {
930 			panic("receive 3: so %p, so_type %d, m %p, m_type %d",
931 			    so, so->so_type, m, m->m_type);
932 #endif
933 		}
934 		so->so_state &= ~SS_RCVATMARK;
935 		len = uio->uio_resid;
936 		if (so->so_oobmark && len > so->so_oobmark - offset)
937 			len = so->so_oobmark - offset;
938 		if (len > m->m_len - moff)
939 			len = m->m_len - moff;
940 		/*
941 		 * If mp is set, just pass back the mbufs.
942 		 * Otherwise copy them out via the uio, then free.
943 		 * Sockbuf must be consistent here (points to current mbuf,
944 		 * it points to next record) when we drop priority;
945 		 * we must note any additions to the sockbuf when we
946 		 * block interrupts again.
947 		 */
948 		if (mp == NULL && uio_error == 0) {
949 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
950 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
951 			resid = uio->uio_resid;
952 			sounlock(so, s);
953 			uio_error = uiomove(mtod(m, caddr_t) + moff, len, uio);
954 			s = solock(so);
955 			if (uio_error)
956 				uio->uio_resid = resid - len;
957 		} else
958 			uio->uio_resid -= len;
959 		if (len == m->m_len - moff) {
960 			if (m->m_flags & M_EOR)
961 				flags |= MSG_EOR;
962 			if (flags & MSG_PEEK) {
963 				m = m->m_next;
964 				moff = 0;
965 				orig_resid = 0;
966 			} else {
967 				nextrecord = m->m_nextpkt;
968 				sbfree(&so->so_rcv, m);
969 				if (mp) {
970 					*mp = m;
971 					mp = &m->m_next;
972 					so->so_rcv.sb_mb = m = m->m_next;
973 					*mp = NULL;
974 				} else {
975 					so->so_rcv.sb_mb = m_free(m);
976 					m = so->so_rcv.sb_mb;
977 				}
978 				/*
979 				 * If m != NULL, we also know that
980 				 * so->so_rcv.sb_mb != NULL.
981 				 */
982 				KASSERT(so->so_rcv.sb_mb == m);
983 				if (m) {
984 					m->m_nextpkt = nextrecord;
985 					if (nextrecord == NULL)
986 						so->so_rcv.sb_lastrecord = m;
987 				} else {
988 					so->so_rcv.sb_mb = nextrecord;
989 					SB_EMPTY_FIXUP(&so->so_rcv);
990 				}
991 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
992 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
993 			}
994 		} else {
995 			if (flags & MSG_PEEK) {
996 				moff += len;
997 				orig_resid = 0;
998 			} else {
999 				if (mp)
1000 					*mp = m_copym(m, 0, len, M_WAIT);
1001 				m->m_data += len;
1002 				m->m_len -= len;
1003 				so->so_rcv.sb_cc -= len;
1004 				so->so_rcv.sb_datacc -= len;
1005 			}
1006 		}
1007 		if (so->so_oobmark) {
1008 			if ((flags & MSG_PEEK) == 0) {
1009 				so->so_oobmark -= len;
1010 				if (so->so_oobmark == 0) {
1011 					so->so_state |= SS_RCVATMARK;
1012 					break;
1013 				}
1014 			} else {
1015 				offset += len;
1016 				if (offset == so->so_oobmark)
1017 					break;
1018 			}
1019 		}
1020 		if (flags & MSG_EOR)
1021 			break;
1022 		/*
1023 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1024 		 * we must not quit until "uio->uio_resid == 0" or an error
1025 		 * termination.  If a signal/timeout occurs, return
1026 		 * with a short count but without error.
1027 		 * Keep sockbuf locked against other readers.
1028 		 */
1029 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1030 		    !sosendallatonce(so) && !nextrecord) {
1031 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1032 				break;
1033 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1034 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1035 			error = sbwait(so, &so->so_rcv);
1036 			if (error) {
1037 				sbunlock(so, &so->so_rcv);
1038 				sounlock(so, s);
1039 				return (0);
1040 			}
1041 			if ((m = so->so_rcv.sb_mb) != NULL)
1042 				nextrecord = m->m_nextpkt;
1043 		}
1044 	}
1045 
1046 	if (m && pr->pr_flags & PR_ATOMIC) {
1047 		flags |= MSG_TRUNC;
1048 		if ((flags & MSG_PEEK) == 0)
1049 			(void) sbdroprecord(&so->so_rcv);
1050 	}
1051 	if ((flags & MSG_PEEK) == 0) {
1052 		if (m == NULL) {
1053 			/*
1054 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1055 			 * part makes sure sb_lastrecord is up-to-date if
1056 			 * there is still data in the socket buffer.
1057 			 */
1058 			so->so_rcv.sb_mb = nextrecord;
1059 			if (so->so_rcv.sb_mb == NULL) {
1060 				so->so_rcv.sb_mbtail = NULL;
1061 				so->so_rcv.sb_lastrecord = NULL;
1062 			} else if (nextrecord->m_nextpkt == NULL)
1063 				so->so_rcv.sb_lastrecord = nextrecord;
1064 		}
1065 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1066 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1067 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1068 			(*pr->pr_usrreq)(so, PRU_RCVD, NULL,
1069 			    (struct mbuf *)(long)flags, NULL, curproc);
1070 	}
1071 	if (orig_resid == uio->uio_resid && orig_resid &&
1072 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1073 		sbunlock(so, &so->so_rcv);
1074 		goto restart;
1075 	}
1076 
1077 	if (uio_error)
1078 		error = uio_error;
1079 
1080 	if (flagsp)
1081 		*flagsp |= flags;
1082 release:
1083 	sbunlock(so, &so->so_rcv);
1084 	sounlock(so, s);
1085 	return (error);
1086 }
1087 
1088 int
1089 soshutdown(struct socket *so, int how)
1090 {
1091 	const struct protosw *pr = so->so_proto;
1092 	int s, error = 0;
1093 
1094 	s = solock(so);
1095 	switch (how) {
1096 	case SHUT_RD:
1097 		sorflush(so);
1098 		break;
1099 	case SHUT_RDWR:
1100 		sorflush(so);
1101 		/* FALLTHROUGH */
1102 	case SHUT_WR:
1103 		error = (*pr->pr_usrreq)(so, PRU_SHUTDOWN, NULL, NULL, NULL,
1104 		    curproc);
1105 		break;
1106 	default:
1107 		error = EINVAL;
1108 		break;
1109 	}
1110 	sounlock(so, s);
1111 
1112 	return (error);
1113 }
1114 
1115 void
1116 sorflush(struct socket *so)
1117 {
1118 	struct sockbuf *sb = &so->so_rcv;
1119 	const struct protosw *pr = so->so_proto;
1120 	struct socket aso;
1121 	int error;
1122 
1123 	sb->sb_flags |= SB_NOINTR;
1124 	error = sblock(so, sb, M_WAITOK);
1125 	/* with SB_NOINTR and M_WAITOK sblock() must not fail */
1126 	KASSERT(error == 0);
1127 	socantrcvmore(so);
1128 	sbunlock(so, sb);
1129 	aso.so_proto = pr;
1130 	aso.so_rcv = *sb;
1131 	memset(&sb->sb_startzero, 0,
1132 	     (caddr_t)&sb->sb_endzero - (caddr_t)&sb->sb_startzero);
1133 	sb->sb_timeo_nsecs = INFSLP;
1134 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose)
1135 		(*pr->pr_domain->dom_dispose)(aso.so_rcv.sb_mb);
1136 	sbrelease(&aso, &aso.so_rcv);
1137 }
1138 
1139 #ifdef SOCKET_SPLICE
1140 
1141 #define so_splicelen	so_sp->ssp_len
1142 #define so_splicemax	so_sp->ssp_max
1143 #define so_idletv	so_sp->ssp_idletv
1144 #define so_idleto	so_sp->ssp_idleto
1145 #define so_splicetask	so_sp->ssp_task
1146 
1147 int
1148 sosplice(struct socket *so, int fd, off_t max, struct timeval *tv)
1149 {
1150 	struct file	*fp;
1151 	struct socket	*sosp;
1152 	struct sosplice	*sp;
1153 	struct taskq	*tq;
1154 	int		 error = 0;
1155 
1156 	soassertlocked(so);
1157 
1158 	if (sosplice_taskq == NULL) {
1159 		rw_enter_write(&sosplice_lock);
1160 		if (sosplice_taskq == NULL) {
1161 			tq = taskq_create("sosplice", 1, IPL_SOFTNET,
1162 			    TASKQ_MPSAFE);
1163 			/* Ensure the taskq is fully visible to other CPUs. */
1164 			membar_producer();
1165 			sosplice_taskq = tq;
1166 		}
1167 		rw_exit_write(&sosplice_lock);
1168 	}
1169 	if (sosplice_taskq == NULL)
1170 		return (ENOMEM);
1171 
1172 	if ((so->so_proto->pr_flags & PR_SPLICE) == 0)
1173 		return (EPROTONOSUPPORT);
1174 	if (so->so_options & SO_ACCEPTCONN)
1175 		return (EOPNOTSUPP);
1176 	if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1177 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
1178 		return (ENOTCONN);
1179 	if (so->so_sp == NULL) {
1180 		sp = pool_get(&sosplice_pool, PR_WAITOK | PR_ZERO);
1181 		if (so->so_sp == NULL)
1182 			so->so_sp = sp;
1183 		else
1184 			pool_put(&sosplice_pool, sp);
1185 	}
1186 
1187 	/* If no fd is given, unsplice by removing existing link. */
1188 	if (fd < 0) {
1189 		/* Lock receive buffer. */
1190 		if ((error = sblock(so, &so->so_rcv, M_WAITOK)) != 0) {
1191 			return (error);
1192 		}
1193 		if (so->so_sp->ssp_socket)
1194 			sounsplice(so, so->so_sp->ssp_socket, 0);
1195 		sbunlock(so, &so->so_rcv);
1196 		return (0);
1197 	}
1198 
1199 	if (max && max < 0)
1200 		return (EINVAL);
1201 
1202 	if (tv && (tv->tv_sec < 0 || !timerisvalid(tv)))
1203 		return (EINVAL);
1204 
1205 	/* Find sosp, the drain socket where data will be spliced into. */
1206 	if ((error = getsock(curproc, fd, &fp)) != 0)
1207 		return (error);
1208 	sosp = fp->f_data;
1209 	if (sosp->so_proto->pr_usrreq != so->so_proto->pr_usrreq) {
1210 		error = EPROTONOSUPPORT;
1211 		goto frele;
1212 	}
1213 	if (sosp->so_sp == NULL) {
1214 		sp = pool_get(&sosplice_pool, PR_WAITOK | PR_ZERO);
1215 		if (sosp->so_sp == NULL)
1216 			sosp->so_sp = sp;
1217 		else
1218 			pool_put(&sosplice_pool, sp);
1219 	}
1220 
1221 	/* Lock both receive and send buffer. */
1222 	if ((error = sblock(so, &so->so_rcv, M_WAITOK)) != 0) {
1223 		goto frele;
1224 	}
1225 	if ((error = sblock(so, &sosp->so_snd, M_WAITOK)) != 0) {
1226 		sbunlock(so, &so->so_rcv);
1227 		goto frele;
1228 	}
1229 
1230 	if (so->so_sp->ssp_socket || sosp->so_sp->ssp_soback) {
1231 		error = EBUSY;
1232 		goto release;
1233 	}
1234 	if (sosp->so_options & SO_ACCEPTCONN) {
1235 		error = EOPNOTSUPP;
1236 		goto release;
1237 	}
1238 	if ((sosp->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0) {
1239 		error = ENOTCONN;
1240 		goto release;
1241 	}
1242 
1243 	/* Splice so and sosp together. */
1244 	so->so_sp->ssp_socket = sosp;
1245 	sosp->so_sp->ssp_soback = so;
1246 	so->so_splicelen = 0;
1247 	so->so_splicemax = max;
1248 	if (tv)
1249 		so->so_idletv = *tv;
1250 	else
1251 		timerclear(&so->so_idletv);
1252 	timeout_set_proc(&so->so_idleto, soidle, so);
1253 	task_set(&so->so_splicetask, sotask, so);
1254 
1255 	/*
1256 	 * To prevent softnet interrupt from calling somove() while
1257 	 * we sleep, the socket buffers are not marked as spliced yet.
1258 	 */
1259 	if (somove(so, M_WAIT)) {
1260 		so->so_rcv.sb_flags |= SB_SPLICE;
1261 		sosp->so_snd.sb_flags |= SB_SPLICE;
1262 	}
1263 
1264  release:
1265 	sbunlock(sosp, &sosp->so_snd);
1266 	sbunlock(so, &so->so_rcv);
1267  frele:
1268 	/*
1269 	 * FRELE() must not be called with the socket lock held. It is safe to
1270 	 * release the lock here as long as no other operation happen on the
1271 	 * socket when sosplice() returns. The dance could be avoided by
1272 	 * grabbing the socket lock inside this function.
1273 	 */
1274 	sounlock(so, SL_LOCKED);
1275 	FRELE(fp, curproc);
1276 	solock(so);
1277 	return (error);
1278 }
1279 
1280 void
1281 sounsplice(struct socket *so, struct socket *sosp, int freeing)
1282 {
1283 	soassertlocked(so);
1284 
1285 	task_del(sosplice_taskq, &so->so_splicetask);
1286 	timeout_del(&so->so_idleto);
1287 	sosp->so_snd.sb_flags &= ~SB_SPLICE;
1288 	so->so_rcv.sb_flags &= ~SB_SPLICE;
1289 	so->so_sp->ssp_socket = sosp->so_sp->ssp_soback = NULL;
1290 	/* Do not wakeup a socket that is about to be freed. */
1291 	if ((freeing & SOSP_FREEING_READ) == 0 && soreadable(so))
1292 		sorwakeup(so);
1293 	if ((freeing & SOSP_FREEING_WRITE) == 0 && sowriteable(sosp))
1294 		sowwakeup(sosp);
1295 }
1296 
1297 void
1298 soidle(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 		so->so_error = ETIMEDOUT;
1306 		sounsplice(so, so->so_sp->ssp_socket, 0);
1307 	}
1308 	sounlock(so, s);
1309 }
1310 
1311 void
1312 sotask(void *arg)
1313 {
1314 	struct socket *so = arg;
1315 	int s;
1316 
1317 	s = solock(so);
1318 	if (so->so_rcv.sb_flags & SB_SPLICE) {
1319 		/*
1320 		 * We may not sleep here as sofree() and unsplice() may be
1321 		 * called from softnet interrupt context.  This would remove
1322 		 * the socket during somove().
1323 		 */
1324 		somove(so, M_DONTWAIT);
1325 	}
1326 	sounlock(so, s);
1327 
1328 	/* Avoid user land starvation. */
1329 	yield();
1330 }
1331 
1332 /*
1333  * The socket splicing task or idle timeout may sleep while grabbing the net
1334  * lock.  As sofree() can be called anytime, sotask() or soidle() could access
1335  * the socket memory of a freed socket after wakeup.  So delay the pool_put()
1336  * after all pending socket splicing tasks or timeouts have finished.  Do this
1337  * by scheduling it on the same threads.
1338  */
1339 void
1340 soreaper(void *arg)
1341 {
1342 	struct socket *so = arg;
1343 
1344 	/* Reuse splice task, sounsplice() has been called before. */
1345 	task_set(&so->so_sp->ssp_task, soput, so);
1346 	task_add(sosplice_taskq, &so->so_sp->ssp_task);
1347 }
1348 
1349 void
1350 soput(void *arg)
1351 {
1352 	struct socket *so = arg;
1353 
1354 	pool_put(&sosplice_pool, so->so_sp);
1355 	pool_put(&socket_pool, so);
1356 }
1357 
1358 /*
1359  * Move data from receive buffer of spliced source socket to send
1360  * buffer of drain socket.  Try to move as much as possible in one
1361  * big chunk.  It is a TCP only implementation.
1362  * Return value 0 means splicing has been finished, 1 continue.
1363  */
1364 int
1365 somove(struct socket *so, int wait)
1366 {
1367 	struct socket	*sosp = so->so_sp->ssp_socket;
1368 	struct mbuf	*m, **mp, *nextrecord;
1369 	u_long		 len, off, oobmark;
1370 	long		 space;
1371 	int		 error = 0, maxreached = 0;
1372 	unsigned int	 state;
1373 
1374 	soassertlocked(so);
1375 
1376  nextpkt:
1377 	if (so->so_error) {
1378 		error = so->so_error;
1379 		goto release;
1380 	}
1381 	if (sosp->so_state & SS_CANTSENDMORE) {
1382 		error = EPIPE;
1383 		goto release;
1384 	}
1385 	if (sosp->so_error && sosp->so_error != ETIMEDOUT &&
1386 	    sosp->so_error != EFBIG && sosp->so_error != ELOOP) {
1387 		error = sosp->so_error;
1388 		goto release;
1389 	}
1390 	if ((sosp->so_state & SS_ISCONNECTED) == 0)
1391 		goto release;
1392 
1393 	/* Calculate how many bytes can be copied now. */
1394 	len = so->so_rcv.sb_datacc;
1395 	if (so->so_splicemax) {
1396 		KASSERT(so->so_splicelen < so->so_splicemax);
1397 		if (so->so_splicemax <= so->so_splicelen + len) {
1398 			len = so->so_splicemax - so->so_splicelen;
1399 			maxreached = 1;
1400 		}
1401 	}
1402 	space = sbspace(sosp, &sosp->so_snd);
1403 	if (so->so_oobmark && so->so_oobmark < len &&
1404 	    so->so_oobmark < space + 1024)
1405 		space += 1024;
1406 	if (space <= 0) {
1407 		maxreached = 0;
1408 		goto release;
1409 	}
1410 	if (space < len) {
1411 		maxreached = 0;
1412 		if (space < sosp->so_snd.sb_lowat)
1413 			goto release;
1414 		len = space;
1415 	}
1416 	sosp->so_state |= SS_ISSENDING;
1417 
1418 	SBLASTRECORDCHK(&so->so_rcv, "somove 1");
1419 	SBLASTMBUFCHK(&so->so_rcv, "somove 1");
1420 	m = so->so_rcv.sb_mb;
1421 	if (m == NULL)
1422 		goto release;
1423 	nextrecord = m->m_nextpkt;
1424 
1425 	/* Drop address and control information not used with splicing. */
1426 	if (so->so_proto->pr_flags & PR_ADDR) {
1427 #ifdef DIAGNOSTIC
1428 		if (m->m_type != MT_SONAME)
1429 			panic("somove soname: so %p, so_type %d, m %p, "
1430 			    "m_type %d", so, so->so_type, m, m->m_type);
1431 #endif
1432 		m = m->m_next;
1433 	}
1434 	while (m && m->m_type == MT_CONTROL)
1435 		m = m->m_next;
1436 	if (m == NULL) {
1437 		sbdroprecord(&so->so_rcv);
1438 		if (so->so_proto->pr_flags & PR_WANTRCVD && so->so_pcb)
1439 			(so->so_proto->pr_usrreq)(so, PRU_RCVD, NULL,
1440 			    NULL, NULL, NULL);
1441 		goto nextpkt;
1442 	}
1443 
1444 	/*
1445 	 * By splicing sockets connected to localhost, userland might create a
1446 	 * loop.  Dissolve splicing with error if loop is detected by counter.
1447 	 *
1448 	 * If we deal with looped broadcast/multicast packet we bail out with
1449 	 * no error to suppress splice termination.
1450 	 */
1451 	if ((m->m_flags & M_PKTHDR) &&
1452 	    ((m->m_pkthdr.ph_loopcnt++ >= M_MAXLOOP) ||
1453 	    ((m->m_flags & M_LOOP) && (m->m_flags & (M_BCAST|M_MCAST))))) {
1454 		if (m->m_pkthdr.ph_loopcnt >= M_MAXLOOP)
1455 			error = ELOOP;
1456 		goto release;
1457 	}
1458 
1459 	if (so->so_proto->pr_flags & PR_ATOMIC) {
1460 		if ((m->m_flags & M_PKTHDR) == 0)
1461 			panic("somove !PKTHDR: so %p, so_type %d, m %p, "
1462 			    "m_type %d", so, so->so_type, m, m->m_type);
1463 		if (sosp->so_snd.sb_hiwat < m->m_pkthdr.len) {
1464 			error = EMSGSIZE;
1465 			goto release;
1466 		}
1467 		if (len < m->m_pkthdr.len)
1468 			goto release;
1469 		if (m->m_pkthdr.len < len) {
1470 			maxreached = 0;
1471 			len = m->m_pkthdr.len;
1472 		}
1473 		/*
1474 		 * Throw away the name mbuf after it has been assured
1475 		 * that the whole first record can be processed.
1476 		 */
1477 		m = so->so_rcv.sb_mb;
1478 		sbfree(&so->so_rcv, m);
1479 		so->so_rcv.sb_mb = m_free(m);
1480 		sbsync(&so->so_rcv, nextrecord);
1481 	}
1482 	/*
1483 	 * Throw away the control mbufs after it has been assured
1484 	 * that the whole first record can be processed.
1485 	 */
1486 	m = so->so_rcv.sb_mb;
1487 	while (m && m->m_type == MT_CONTROL) {
1488 		sbfree(&so->so_rcv, m);
1489 		so->so_rcv.sb_mb = m_free(m);
1490 		m = so->so_rcv.sb_mb;
1491 		sbsync(&so->so_rcv, nextrecord);
1492 	}
1493 
1494 	SBLASTRECORDCHK(&so->so_rcv, "somove 2");
1495 	SBLASTMBUFCHK(&so->so_rcv, "somove 2");
1496 
1497 	/* Take at most len mbufs out of receive buffer. */
1498 	for (off = 0, mp = &m; off <= len && *mp;
1499 	    off += (*mp)->m_len, mp = &(*mp)->m_next) {
1500 		u_long size = len - off;
1501 
1502 #ifdef DIAGNOSTIC
1503 		if ((*mp)->m_type != MT_DATA && (*mp)->m_type != MT_HEADER)
1504 			panic("somove type: so %p, so_type %d, m %p, "
1505 			    "m_type %d", so, so->so_type, *mp, (*mp)->m_type);
1506 #endif
1507 		if ((*mp)->m_len > size) {
1508 			/*
1509 			 * Move only a partial mbuf at maximum splice length or
1510 			 * if the drain buffer is too small for this large mbuf.
1511 			 */
1512 			if (!maxreached && so->so_snd.sb_datacc > 0) {
1513 				len -= size;
1514 				break;
1515 			}
1516 			*mp = m_copym(so->so_rcv.sb_mb, 0, size, wait);
1517 			if (*mp == NULL) {
1518 				len -= size;
1519 				break;
1520 			}
1521 			so->so_rcv.sb_mb->m_data += size;
1522 			so->so_rcv.sb_mb->m_len -= size;
1523 			so->so_rcv.sb_cc -= size;
1524 			so->so_rcv.sb_datacc -= size;
1525 		} else {
1526 			*mp = so->so_rcv.sb_mb;
1527 			sbfree(&so->so_rcv, *mp);
1528 			so->so_rcv.sb_mb = (*mp)->m_next;
1529 			sbsync(&so->so_rcv, nextrecord);
1530 		}
1531 	}
1532 	*mp = NULL;
1533 
1534 	SBLASTRECORDCHK(&so->so_rcv, "somove 3");
1535 	SBLASTMBUFCHK(&so->so_rcv, "somove 3");
1536 	SBCHECK(&so->so_rcv);
1537 	if (m == NULL)
1538 		goto release;
1539 	m->m_nextpkt = NULL;
1540 	if (m->m_flags & M_PKTHDR) {
1541 		m_resethdr(m);
1542 		m->m_pkthdr.len = len;
1543 	}
1544 
1545 	/* Send window update to source peer as receive buffer has changed. */
1546 	if (so->so_proto->pr_flags & PR_WANTRCVD && so->so_pcb)
1547 		(so->so_proto->pr_usrreq)(so, PRU_RCVD, NULL,
1548 		    NULL, NULL, NULL);
1549 
1550 	/* Receive buffer did shrink by len bytes, adjust oob. */
1551 	state = so->so_state;
1552 	so->so_state &= ~SS_RCVATMARK;
1553 	oobmark = so->so_oobmark;
1554 	so->so_oobmark = oobmark > len ? oobmark - len : 0;
1555 	if (oobmark) {
1556 		if (oobmark == len)
1557 			so->so_state |= SS_RCVATMARK;
1558 		if (oobmark >= len)
1559 			oobmark = 0;
1560 	}
1561 
1562 	/*
1563 	 * Handle oob data.  If any malloc fails, ignore error.
1564 	 * TCP urgent data is not very reliable anyway.
1565 	 */
1566 	while (((state & SS_RCVATMARK) || oobmark) &&
1567 	    (so->so_options & SO_OOBINLINE)) {
1568 		struct mbuf *o = NULL;
1569 
1570 		if (state & SS_RCVATMARK) {
1571 			o = m_get(wait, MT_DATA);
1572 			state &= ~SS_RCVATMARK;
1573 		} else if (oobmark) {
1574 			o = m_split(m, oobmark, wait);
1575 			if (o) {
1576 				error = (*sosp->so_proto->pr_usrreq)(sosp,
1577 				    PRU_SEND, m, NULL, NULL, NULL);
1578 				if (error) {
1579 					if (sosp->so_state & SS_CANTSENDMORE)
1580 						error = EPIPE;
1581 					m_freem(o);
1582 					goto release;
1583 				}
1584 				len -= oobmark;
1585 				so->so_splicelen += oobmark;
1586 				m = o;
1587 				o = m_get(wait, MT_DATA);
1588 			}
1589 			oobmark = 0;
1590 		}
1591 		if (o) {
1592 			o->m_len = 1;
1593 			*mtod(o, caddr_t) = *mtod(m, caddr_t);
1594 			error = (*sosp->so_proto->pr_usrreq)(sosp, PRU_SENDOOB,
1595 			    o, NULL, NULL, NULL);
1596 			if (error) {
1597 				if (sosp->so_state & SS_CANTSENDMORE)
1598 					error = EPIPE;
1599 				m_freem(m);
1600 				goto release;
1601 			}
1602 			len -= 1;
1603 			so->so_splicelen += 1;
1604 			if (oobmark) {
1605 				oobmark -= 1;
1606 				if (oobmark == 0)
1607 					state |= SS_RCVATMARK;
1608 			}
1609 			m_adj(m, 1);
1610 		}
1611 	}
1612 
1613 	/* Append all remaining data to drain socket. */
1614 	if (so->so_rcv.sb_cc == 0 || maxreached)
1615 		sosp->so_state &= ~SS_ISSENDING;
1616 	error = (*sosp->so_proto->pr_usrreq)(sosp, PRU_SEND, m, NULL, NULL,
1617 	    NULL);
1618 	if (error) {
1619 		if (sosp->so_state & SS_CANTSENDMORE)
1620 			error = EPIPE;
1621 		goto release;
1622 	}
1623 	so->so_splicelen += len;
1624 
1625 	/* Move several packets if possible. */
1626 	if (!maxreached && nextrecord)
1627 		goto nextpkt;
1628 
1629  release:
1630 	sosp->so_state &= ~SS_ISSENDING;
1631 	if (!error && maxreached && so->so_splicemax == so->so_splicelen)
1632 		error = EFBIG;
1633 	if (error)
1634 		so->so_error = error;
1635 	if (((so->so_state & SS_CANTRCVMORE) && so->so_rcv.sb_cc == 0) ||
1636 	    (sosp->so_state & SS_CANTSENDMORE) || maxreached || error) {
1637 		sounsplice(so, sosp, 0);
1638 		return (0);
1639 	}
1640 	if (timerisset(&so->so_idletv))
1641 		timeout_add_tv(&so->so_idleto, &so->so_idletv);
1642 	return (1);
1643 }
1644 
1645 #endif /* SOCKET_SPLICE */
1646 
1647 void
1648 sorwakeup(struct socket *so)
1649 {
1650 	soassertlocked(so);
1651 
1652 #ifdef SOCKET_SPLICE
1653 	if (so->so_rcv.sb_flags & SB_SPLICE) {
1654 		/*
1655 		 * TCP has a sendbuffer that can handle multiple packets
1656 		 * at once.  So queue the stream a bit to accumulate data.
1657 		 * The sosplice thread will call somove() later and send
1658 		 * the packets calling tcp_output() only once.
1659 		 * In the UDP case, send out the packets immediately.
1660 		 * Using a thread would make things slower.
1661 		 */
1662 		if (so->so_proto->pr_flags & PR_WANTRCVD)
1663 			task_add(sosplice_taskq, &so->so_splicetask);
1664 		else
1665 			somove(so, M_DONTWAIT);
1666 	}
1667 	if (isspliced(so))
1668 		return;
1669 #endif
1670 	sowakeup(so, &so->so_rcv);
1671 	if (so->so_upcall)
1672 		(*(so->so_upcall))(so, so->so_upcallarg, M_DONTWAIT);
1673 }
1674 
1675 void
1676 sowwakeup(struct socket *so)
1677 {
1678 	soassertlocked(so);
1679 
1680 #ifdef SOCKET_SPLICE
1681 	if (so->so_snd.sb_flags & SB_SPLICE)
1682 		task_add(sosplice_taskq, &so->so_sp->ssp_soback->so_splicetask);
1683 	if (issplicedback(so))
1684 		return;
1685 #endif
1686 	sowakeup(so, &so->so_snd);
1687 }
1688 
1689 int
1690 sosetopt(struct socket *so, int level, int optname, struct mbuf *m)
1691 {
1692 	int error = 0;
1693 
1694 	soassertlocked(so);
1695 
1696 	if (level != SOL_SOCKET) {
1697 		if (so->so_proto->pr_ctloutput) {
1698 			error = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so,
1699 			    level, optname, m);
1700 			return (error);
1701 		}
1702 		error = ENOPROTOOPT;
1703 	} else {
1704 		switch (optname) {
1705 		case SO_BINDANY:
1706 			if ((error = suser(curproc)) != 0)	/* XXX */
1707 				return (error);
1708 			break;
1709 		}
1710 
1711 		switch (optname) {
1712 
1713 		case SO_LINGER:
1714 			if (m == NULL || m->m_len != sizeof (struct linger) ||
1715 			    mtod(m, struct linger *)->l_linger < 0 ||
1716 			    mtod(m, struct linger *)->l_linger > SHRT_MAX)
1717 				return (EINVAL);
1718 			so->so_linger = mtod(m, struct linger *)->l_linger;
1719 			/* FALLTHROUGH */
1720 
1721 		case SO_BINDANY:
1722 		case SO_DEBUG:
1723 		case SO_KEEPALIVE:
1724 		case SO_USELOOPBACK:
1725 		case SO_BROADCAST:
1726 		case SO_REUSEADDR:
1727 		case SO_REUSEPORT:
1728 		case SO_OOBINLINE:
1729 		case SO_TIMESTAMP:
1730 		case SO_ZEROIZE:
1731 			if (m == NULL || m->m_len < sizeof (int))
1732 				return (EINVAL);
1733 			if (*mtod(m, int *))
1734 				so->so_options |= optname;
1735 			else
1736 				so->so_options &= ~optname;
1737 			break;
1738 
1739 		case SO_DONTROUTE:
1740 			if (m == NULL || m->m_len < sizeof (int))
1741 				return (EINVAL);
1742 			if (*mtod(m, int *))
1743 				error = EOPNOTSUPP;
1744 			break;
1745 
1746 		case SO_SNDBUF:
1747 		case SO_RCVBUF:
1748 		case SO_SNDLOWAT:
1749 		case SO_RCVLOWAT:
1750 		    {
1751 			u_long cnt;
1752 
1753 			if (m == NULL || m->m_len < sizeof (int))
1754 				return (EINVAL);
1755 			cnt = *mtod(m, int *);
1756 			if ((long)cnt <= 0)
1757 				cnt = 1;
1758 			switch (optname) {
1759 
1760 			case SO_SNDBUF:
1761 				if (so->so_state & SS_CANTSENDMORE)
1762 					return (EINVAL);
1763 				if (sbcheckreserve(cnt, so->so_snd.sb_wat) ||
1764 				    sbreserve(so, &so->so_snd, cnt))
1765 					return (ENOBUFS);
1766 				so->so_snd.sb_wat = cnt;
1767 				break;
1768 
1769 			case SO_RCVBUF:
1770 				if (so->so_state & SS_CANTRCVMORE)
1771 					return (EINVAL);
1772 				if (sbcheckreserve(cnt, so->so_rcv.sb_wat) ||
1773 				    sbreserve(so, &so->so_rcv, cnt))
1774 					return (ENOBUFS);
1775 				so->so_rcv.sb_wat = cnt;
1776 				break;
1777 
1778 			case SO_SNDLOWAT:
1779 				so->so_snd.sb_lowat =
1780 				    (cnt > so->so_snd.sb_hiwat) ?
1781 				    so->so_snd.sb_hiwat : cnt;
1782 				break;
1783 			case SO_RCVLOWAT:
1784 				so->so_rcv.sb_lowat =
1785 				    (cnt > so->so_rcv.sb_hiwat) ?
1786 				    so->so_rcv.sb_hiwat : cnt;
1787 				break;
1788 			}
1789 			break;
1790 		    }
1791 
1792 		case SO_SNDTIMEO:
1793 		case SO_RCVTIMEO:
1794 		    {
1795 			struct timeval tv;
1796 			uint64_t nsecs;
1797 
1798 			if (m == NULL || m->m_len < sizeof (tv))
1799 				return (EINVAL);
1800 			memcpy(&tv, mtod(m, struct timeval *), sizeof tv);
1801 			if (!timerisvalid(&tv))
1802 				return (EINVAL);
1803 			nsecs = TIMEVAL_TO_NSEC(&tv);
1804 			if (nsecs == UINT64_MAX)
1805 				return (EDOM);
1806 			if (nsecs == 0)
1807 				nsecs = INFSLP;
1808 			switch (optname) {
1809 
1810 			case SO_SNDTIMEO:
1811 				so->so_snd.sb_timeo_nsecs = nsecs;
1812 				break;
1813 			case SO_RCVTIMEO:
1814 				so->so_rcv.sb_timeo_nsecs = nsecs;
1815 				break;
1816 			}
1817 			break;
1818 		    }
1819 
1820 		case SO_RTABLE:
1821 			if (so->so_proto->pr_domain &&
1822 			    so->so_proto->pr_domain->dom_protosw &&
1823 			    so->so_proto->pr_ctloutput) {
1824 				struct domain *dom = so->so_proto->pr_domain;
1825 
1826 				level = dom->dom_protosw->pr_protocol;
1827 				error = (*so->so_proto->pr_ctloutput)
1828 				    (PRCO_SETOPT, so, level, optname, m);
1829 				return (error);
1830 			}
1831 			error = ENOPROTOOPT;
1832 			break;
1833 
1834 #ifdef SOCKET_SPLICE
1835 		case SO_SPLICE:
1836 			if (m == NULL) {
1837 				error = sosplice(so, -1, 0, NULL);
1838 			} else if (m->m_len < sizeof(int)) {
1839 				return (EINVAL);
1840 			} else if (m->m_len < sizeof(struct splice)) {
1841 				error = sosplice(so, *mtod(m, int *), 0, NULL);
1842 			} else {
1843 				error = sosplice(so,
1844 				    mtod(m, struct splice *)->sp_fd,
1845 				    mtod(m, struct splice *)->sp_max,
1846 				   &mtod(m, struct splice *)->sp_idle);
1847 			}
1848 			break;
1849 #endif /* SOCKET_SPLICE */
1850 
1851 		default:
1852 			error = ENOPROTOOPT;
1853 			break;
1854 		}
1855 		if (error == 0 && so->so_proto->pr_ctloutput) {
1856 			(*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so,
1857 			    level, optname, m);
1858 		}
1859 	}
1860 
1861 	return (error);
1862 }
1863 
1864 int
1865 sogetopt(struct socket *so, int level, int optname, struct mbuf *m)
1866 {
1867 	int error = 0;
1868 
1869 	soassertlocked(so);
1870 
1871 	if (level != SOL_SOCKET) {
1872 		if (so->so_proto->pr_ctloutput) {
1873 			m->m_len = 0;
1874 
1875 			error = (*so->so_proto->pr_ctloutput)(PRCO_GETOPT, so,
1876 			    level, optname, m);
1877 			if (error)
1878 				return (error);
1879 			return (0);
1880 		} else
1881 			return (ENOPROTOOPT);
1882 	} else {
1883 		m->m_len = sizeof (int);
1884 
1885 		switch (optname) {
1886 
1887 		case SO_LINGER:
1888 			m->m_len = sizeof (struct linger);
1889 			mtod(m, struct linger *)->l_onoff =
1890 				so->so_options & SO_LINGER;
1891 			mtod(m, struct linger *)->l_linger = so->so_linger;
1892 			break;
1893 
1894 		case SO_BINDANY:
1895 		case SO_USELOOPBACK:
1896 		case SO_DEBUG:
1897 		case SO_KEEPALIVE:
1898 		case SO_REUSEADDR:
1899 		case SO_REUSEPORT:
1900 		case SO_BROADCAST:
1901 		case SO_OOBINLINE:
1902 		case SO_TIMESTAMP:
1903 		case SO_ZEROIZE:
1904 			*mtod(m, int *) = so->so_options & optname;
1905 			break;
1906 
1907 		case SO_DONTROUTE:
1908 			*mtod(m, int *) = 0;
1909 			break;
1910 
1911 		case SO_TYPE:
1912 			*mtod(m, int *) = so->so_type;
1913 			break;
1914 
1915 		case SO_ERROR:
1916 			*mtod(m, int *) = so->so_error;
1917 			so->so_error = 0;
1918 			break;
1919 
1920 		case SO_DOMAIN:
1921 			*mtod(m, int *) = so->so_proto->pr_domain->dom_family;
1922 			break;
1923 
1924 		case SO_PROTOCOL:
1925 			*mtod(m, int *) = so->so_proto->pr_protocol;
1926 			break;
1927 
1928 		case SO_SNDBUF:
1929 			*mtod(m, int *) = so->so_snd.sb_hiwat;
1930 			break;
1931 
1932 		case SO_RCVBUF:
1933 			*mtod(m, int *) = so->so_rcv.sb_hiwat;
1934 			break;
1935 
1936 		case SO_SNDLOWAT:
1937 			*mtod(m, int *) = so->so_snd.sb_lowat;
1938 			break;
1939 
1940 		case SO_RCVLOWAT:
1941 			*mtod(m, int *) = so->so_rcv.sb_lowat;
1942 			break;
1943 
1944 		case SO_SNDTIMEO:
1945 		case SO_RCVTIMEO:
1946 		    {
1947 			struct timeval tv;
1948 			uint64_t nsecs = (optname == SO_SNDTIMEO ?
1949 			    so->so_snd.sb_timeo_nsecs :
1950 			    so->so_rcv.sb_timeo_nsecs);
1951 
1952 			m->m_len = sizeof(struct timeval);
1953 			memset(&tv, 0, sizeof(tv));
1954 			if (nsecs != INFSLP)
1955 				NSEC_TO_TIMEVAL(nsecs, &tv);
1956 			memcpy(mtod(m, struct timeval *), &tv, sizeof tv);
1957 			break;
1958 		    }
1959 
1960 		case SO_RTABLE:
1961 			if (so->so_proto->pr_domain &&
1962 			    so->so_proto->pr_domain->dom_protosw &&
1963 			    so->so_proto->pr_ctloutput) {
1964 				struct domain *dom = so->so_proto->pr_domain;
1965 
1966 				level = dom->dom_protosw->pr_protocol;
1967 				error = (*so->so_proto->pr_ctloutput)
1968 				    (PRCO_GETOPT, so, level, optname, m);
1969 				if (error)
1970 					return (error);
1971 				break;
1972 			}
1973 			return (ENOPROTOOPT);
1974 
1975 #ifdef SOCKET_SPLICE
1976 		case SO_SPLICE:
1977 		    {
1978 			off_t len;
1979 
1980 			m->m_len = sizeof(off_t);
1981 			len = so->so_sp ? so->so_sp->ssp_len : 0;
1982 			memcpy(mtod(m, off_t *), &len, sizeof(off_t));
1983 			break;
1984 		    }
1985 #endif /* SOCKET_SPLICE */
1986 
1987 		case SO_PEERCRED:
1988 			if (so->so_proto->pr_protocol == AF_UNIX) {
1989 				struct unpcb *unp = sotounpcb(so);
1990 
1991 				if (unp->unp_flags & UNP_FEIDS) {
1992 					m->m_len = sizeof(unp->unp_connid);
1993 					memcpy(mtod(m, caddr_t),
1994 					    &(unp->unp_connid), m->m_len);
1995 					break;
1996 				}
1997 				return (ENOTCONN);
1998 			}
1999 			return (EOPNOTSUPP);
2000 
2001 		default:
2002 			return (ENOPROTOOPT);
2003 		}
2004 		return (0);
2005 	}
2006 }
2007 
2008 void
2009 sohasoutofband(struct socket *so)
2010 {
2011 	pgsigio(&so->so_sigio, SIGURG, 0);
2012 	selwakeup(&so->so_rcv.sb_sel);
2013 }
2014 
2015 int
2016 soo_kqfilter(struct file *fp, struct knote *kn)
2017 {
2018 	struct socket *so = kn->kn_fp->f_data;
2019 	struct sockbuf *sb;
2020 
2021 	KERNEL_ASSERT_LOCKED();
2022 
2023 	switch (kn->kn_filter) {
2024 	case EVFILT_READ:
2025 		if (so->so_options & SO_ACCEPTCONN)
2026 			kn->kn_fop = &solisten_filtops;
2027 		else
2028 			kn->kn_fop = &soread_filtops;
2029 		sb = &so->so_rcv;
2030 		break;
2031 	case EVFILT_WRITE:
2032 		kn->kn_fop = &sowrite_filtops;
2033 		sb = &so->so_snd;
2034 		break;
2035 	case EVFILT_EXCEPT:
2036 		kn->kn_fop = &soexcept_filtops;
2037 		sb = &so->so_rcv;
2038 		break;
2039 	default:
2040 		return (EINVAL);
2041 	}
2042 
2043 	klist_insert_locked(&sb->sb_sel.si_note, kn);
2044 	sb->sb_flagsintr |= SB_KNOTE;
2045 
2046 	return (0);
2047 }
2048 
2049 void
2050 filt_sordetach(struct knote *kn)
2051 {
2052 	struct socket *so = kn->kn_fp->f_data;
2053 
2054 	KERNEL_ASSERT_LOCKED();
2055 
2056 	klist_remove_locked(&so->so_rcv.sb_sel.si_note, kn);
2057 	if (klist_empty(&so->so_rcv.sb_sel.si_note))
2058 		so->so_rcv.sb_flagsintr &= ~SB_KNOTE;
2059 }
2060 
2061 int
2062 filt_soread(struct knote *kn, long hint)
2063 {
2064 	struct socket *so = kn->kn_fp->f_data;
2065 	int s, rv = 0;
2066 
2067 	if ((hint & NOTE_SUBMIT) == 0)
2068 		s = solock(so);
2069 	kn->kn_data = so->so_rcv.sb_cc;
2070 #ifdef SOCKET_SPLICE
2071 	if (isspliced(so)) {
2072 		rv = 0;
2073 	} else
2074 #endif /* SOCKET_SPLICE */
2075 	if (kn->kn_sfflags & NOTE_OOB) {
2076 		if (so->so_oobmark || (so->so_state & SS_RCVATMARK)) {
2077 			kn->kn_fflags |= NOTE_OOB;
2078 			kn->kn_data -= so->so_oobmark;
2079 			rv = 1;
2080 		}
2081 	} else if (so->so_state & SS_CANTRCVMORE) {
2082 		kn->kn_flags |= EV_EOF;
2083 		if (kn->kn_flags & __EV_POLL) {
2084 			if (so->so_state & SS_ISDISCONNECTED)
2085 				kn->kn_flags |= __EV_HUP;
2086 		}
2087 		kn->kn_fflags = so->so_error;
2088 		rv = 1;
2089 	} else if (so->so_error) {	/* temporary udp error */
2090 		rv = 1;
2091 	} else if (kn->kn_sfflags & NOTE_LOWAT) {
2092 		rv = (kn->kn_data >= kn->kn_sdata);
2093 	} else {
2094 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
2095 	}
2096 	if ((hint & NOTE_SUBMIT) == 0)
2097 		sounlock(so, s);
2098 
2099 	return rv;
2100 }
2101 
2102 void
2103 filt_sowdetach(struct knote *kn)
2104 {
2105 	struct socket *so = kn->kn_fp->f_data;
2106 
2107 	KERNEL_ASSERT_LOCKED();
2108 
2109 	klist_remove_locked(&so->so_snd.sb_sel.si_note, kn);
2110 	if (klist_empty(&so->so_snd.sb_sel.si_note))
2111 		so->so_snd.sb_flagsintr &= ~SB_KNOTE;
2112 }
2113 
2114 int
2115 filt_sowrite(struct knote *kn, long hint)
2116 {
2117 	struct socket *so = kn->kn_fp->f_data;
2118 	int s, rv;
2119 
2120 	if ((hint & NOTE_SUBMIT) == 0)
2121 		s = solock(so);
2122 	kn->kn_data = sbspace(so, &so->so_snd);
2123 	if (so->so_state & SS_CANTSENDMORE) {
2124 		kn->kn_flags |= EV_EOF;
2125 		if (kn->kn_flags & __EV_POLL) {
2126 			if (so->so_state & SS_ISDISCONNECTED)
2127 				kn->kn_flags |= __EV_HUP;
2128 		}
2129 		kn->kn_fflags = so->so_error;
2130 		rv = 1;
2131 	} else if (so->so_error) {	/* temporary udp error */
2132 		rv = 1;
2133 	} else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2134 	    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
2135 		rv = 0;
2136 	} else if (kn->kn_sfflags & NOTE_LOWAT) {
2137 		rv = (kn->kn_data >= kn->kn_sdata);
2138 	} else {
2139 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
2140 	}
2141 	if ((hint & NOTE_SUBMIT) == 0)
2142 		sounlock(so, s);
2143 
2144 	return (rv);
2145 }
2146 
2147 int
2148 filt_solisten(struct knote *kn, long hint)
2149 {
2150 	struct socket *so = kn->kn_fp->f_data;
2151 	int s;
2152 
2153 	if ((hint & NOTE_SUBMIT) == 0)
2154 		s = solock(so);
2155 	kn->kn_data = so->so_qlen;
2156 	if ((hint & NOTE_SUBMIT) == 0)
2157 		sounlock(so, s);
2158 
2159 	return (kn->kn_data != 0);
2160 }
2161 
2162 #ifdef DDB
2163 void
2164 sobuf_print(struct sockbuf *,
2165     int (*)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))));
2166 
2167 void
2168 sobuf_print(struct sockbuf *sb,
2169     int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2170 {
2171 	(*pr)("\tsb_cc: %lu\n", sb->sb_cc);
2172 	(*pr)("\tsb_datacc: %lu\n", sb->sb_datacc);
2173 	(*pr)("\tsb_hiwat: %lu\n", sb->sb_hiwat);
2174 	(*pr)("\tsb_wat: %lu\n", sb->sb_wat);
2175 	(*pr)("\tsb_mbcnt: %lu\n", sb->sb_mbcnt);
2176 	(*pr)("\tsb_mbmax: %lu\n", sb->sb_mbmax);
2177 	(*pr)("\tsb_lowat: %ld\n", sb->sb_lowat);
2178 	(*pr)("\tsb_mb: %p\n", sb->sb_mb);
2179 	(*pr)("\tsb_mbtail: %p\n", sb->sb_mbtail);
2180 	(*pr)("\tsb_lastrecord: %p\n", sb->sb_lastrecord);
2181 	(*pr)("\tsb_sel: ...\n");
2182 	(*pr)("\tsb_flagsintr: %d\n", sb->sb_flagsintr);
2183 	(*pr)("\tsb_flags: %i\n", sb->sb_flags);
2184 	(*pr)("\tsb_timeo_nsecs: %llu\n", sb->sb_timeo_nsecs);
2185 }
2186 
2187 void
2188 so_print(void *v,
2189     int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2190 {
2191 	struct socket *so = v;
2192 
2193 	(*pr)("socket %p\n", so);
2194 	(*pr)("so_type: %i\n", so->so_type);
2195 	(*pr)("so_options: 0x%04x\n", so->so_options); /* %b */
2196 	(*pr)("so_linger: %i\n", so->so_linger);
2197 	(*pr)("so_state: 0x%04x\n", so->so_state);
2198 	(*pr)("so_pcb: %p\n", so->so_pcb);
2199 	(*pr)("so_proto: %p\n", so->so_proto);
2200 	(*pr)("so_sigio: %p\n", so->so_sigio.sir_sigio);
2201 
2202 	(*pr)("so_head: %p\n", so->so_head);
2203 	(*pr)("so_onq: %p\n", so->so_onq);
2204 	(*pr)("so_q0: @%p first: %p\n", &so->so_q0, TAILQ_FIRST(&so->so_q0));
2205 	(*pr)("so_q: @%p first: %p\n", &so->so_q, TAILQ_FIRST(&so->so_q));
2206 	(*pr)("so_eq: next: %p\n", TAILQ_NEXT(so, so_qe));
2207 	(*pr)("so_q0len: %i\n", so->so_q0len);
2208 	(*pr)("so_qlen: %i\n", so->so_qlen);
2209 	(*pr)("so_qlimit: %i\n", so->so_qlimit);
2210 	(*pr)("so_timeo: %i\n", so->so_timeo);
2211 	(*pr)("so_obmark: %lu\n", so->so_oobmark);
2212 
2213 	(*pr)("so_sp: %p\n", so->so_sp);
2214 	if (so->so_sp != NULL) {
2215 		(*pr)("\tssp_socket: %p\n", so->so_sp->ssp_socket);
2216 		(*pr)("\tssp_soback: %p\n", so->so_sp->ssp_soback);
2217 		(*pr)("\tssp_len: %lld\n",
2218 		    (unsigned long long)so->so_sp->ssp_len);
2219 		(*pr)("\tssp_max: %lld\n",
2220 		    (unsigned long long)so->so_sp->ssp_max);
2221 		(*pr)("\tssp_idletv: %lld %ld\n", so->so_sp->ssp_idletv.tv_sec,
2222 		    so->so_sp->ssp_idletv.tv_usec);
2223 		(*pr)("\tssp_idleto: %spending (@%i)\n",
2224 		    timeout_pending(&so->so_sp->ssp_idleto) ? "" : "not ",
2225 		    so->so_sp->ssp_idleto.to_time);
2226 	}
2227 
2228 	(*pr)("so_rcv:\n");
2229 	sobuf_print(&so->so_rcv, pr);
2230 	(*pr)("so_snd:\n");
2231 	sobuf_print(&so->so_snd, pr);
2232 
2233 	(*pr)("so_upcall: %p so_upcallarg: %p\n",
2234 	    so->so_upcall, so->so_upcallarg);
2235 
2236 	(*pr)("so_euid: %d so_ruid: %d\n", so->so_euid, so->so_ruid);
2237 	(*pr)("so_egid: %d so_rgid: %d\n", so->so_egid, so->so_rgid);
2238 	(*pr)("so_cpid: %d\n", so->so_cpid);
2239 }
2240 #endif
2241