xref: /openbsd-src/sys/kern/uipc_socket.c (revision f1dd7b858388b4a23f4f67a4957ec5ff656ebbe8)
1 /*	$OpenBSD: uipc_socket.c,v 1.261 2021/05/13 19:43:11 mvs 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 void	sorflush(struct socket *);
70 
71 void	filt_sordetach(struct knote *kn);
72 int	filt_soread(struct knote *kn, long hint);
73 void	filt_sowdetach(struct knote *kn);
74 int	filt_sowrite(struct knote *kn, long hint);
75 int	filt_solisten(struct knote *kn, long hint);
76 
77 const struct filterops solisten_filtops = {
78 	.f_flags	= FILTEROP_ISFD,
79 	.f_attach	= NULL,
80 	.f_detach	= filt_sordetach,
81 	.f_event	= filt_solisten,
82 };
83 
84 const struct filterops soread_filtops = {
85 	.f_flags	= FILTEROP_ISFD,
86 	.f_attach	= NULL,
87 	.f_detach	= filt_sordetach,
88 	.f_event	= filt_soread,
89 };
90 
91 const struct filterops sowrite_filtops = {
92 	.f_flags	= FILTEROP_ISFD,
93 	.f_attach	= NULL,
94 	.f_detach	= filt_sowdetach,
95 	.f_event	= filt_sowrite,
96 };
97 
98 const struct filterops soexcept_filtops = {
99 	.f_flags	= FILTEROP_ISFD,
100 	.f_attach	= NULL,
101 	.f_detach	= filt_sordetach,
102 	.f_event	= filt_soread,
103 };
104 
105 #ifndef SOMINCONN
106 #define SOMINCONN 80
107 #endif /* SOMINCONN */
108 
109 int	somaxconn = SOMAXCONN;
110 int	sominconn = SOMINCONN;
111 
112 struct pool socket_pool;
113 #ifdef SOCKET_SPLICE
114 struct pool sosplice_pool;
115 struct taskq *sosplice_taskq;
116 struct rwlock sosplice_lock = RWLOCK_INITIALIZER("sosplicelk");
117 #endif
118 
119 void
120 soinit(void)
121 {
122 	pool_init(&socket_pool, sizeof(struct socket), 0, IPL_SOFTNET, 0,
123 	    "sockpl", NULL);
124 #ifdef SOCKET_SPLICE
125 	pool_init(&sosplice_pool, sizeof(struct sosplice), 0, IPL_SOFTNET, 0,
126 	    "sosppl", NULL);
127 #endif
128 }
129 
130 /*
131  * Socket operation routines.
132  * These routines are called by the routines in
133  * sys_socket.c or from a system process, and
134  * implement the semantics of socket operations by
135  * switching out to the protocol specific routines.
136  */
137 int
138 socreate(int dom, struct socket **aso, int type, int proto)
139 {
140 	struct proc *p = curproc;		/* XXX */
141 	const struct protosw *prp;
142 	struct socket *so;
143 	int error, s;
144 
145 	if (proto)
146 		prp = pffindproto(dom, proto, type);
147 	else
148 		prp = pffindtype(dom, type);
149 	if (prp == NULL || prp->pr_attach == NULL)
150 		return (EPROTONOSUPPORT);
151 	if (prp->pr_type != type)
152 		return (EPROTOTYPE);
153 	so = pool_get(&socket_pool, PR_WAITOK | PR_ZERO);
154 	rw_init(&so->so_lock, "solock");
155 	sigio_init(&so->so_sigio);
156 	TAILQ_INIT(&so->so_q0);
157 	TAILQ_INIT(&so->so_q);
158 	so->so_type = type;
159 	if (suser(p) == 0)
160 		so->so_state = SS_PRIV;
161 	so->so_ruid = p->p_ucred->cr_ruid;
162 	so->so_euid = p->p_ucred->cr_uid;
163 	so->so_rgid = p->p_ucred->cr_rgid;
164 	so->so_egid = p->p_ucred->cr_gid;
165 	so->so_cpid = p->p_p->ps_pid;
166 	so->so_proto = prp;
167 	so->so_snd.sb_timeo_nsecs = INFSLP;
168 	so->so_rcv.sb_timeo_nsecs = INFSLP;
169 
170 	s = solock(so);
171 	error = (*prp->pr_attach)(so, proto);
172 	if (error) {
173 		so->so_state |= SS_NOFDREF;
174 		/* sofree() calls sounlock(). */
175 		sofree(so, s);
176 		return (error);
177 	}
178 	sounlock(so, s);
179 	*aso = so;
180 	return (0);
181 }
182 
183 int
184 sobind(struct socket *so, struct mbuf *nam, struct proc *p)
185 {
186 	int error;
187 
188 	soassertlocked(so);
189 
190 	error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, NULL, nam, NULL, p);
191 	return (error);
192 }
193 
194 int
195 solisten(struct socket *so, int backlog)
196 {
197 	int error;
198 
199 	soassertlocked(so);
200 
201 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING|SS_ISDISCONNECTING))
202 		return (EINVAL);
203 #ifdef SOCKET_SPLICE
204 	if (isspliced(so) || issplicedback(so))
205 		return (EOPNOTSUPP);
206 #endif /* SOCKET_SPLICE */
207 	error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, NULL, NULL, NULL,
208 	    curproc);
209 	if (error)
210 		return (error);
211 	if (TAILQ_FIRST(&so->so_q) == NULL)
212 		so->so_options |= SO_ACCEPTCONN;
213 	if (backlog < 0 || backlog > somaxconn)
214 		backlog = somaxconn;
215 	if (backlog < sominconn)
216 		backlog = sominconn;
217 	so->so_qlimit = backlog;
218 	return (0);
219 }
220 
221 #define SOSP_FREEING_READ	1
222 #define SOSP_FREEING_WRITE	2
223 void
224 sofree(struct socket *so, int s)
225 {
226 	soassertlocked(so);
227 
228 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
229 		sounlock(so, s);
230 		return;
231 	}
232 	if (so->so_head) {
233 		/*
234 		 * We must not decommission a socket that's on the accept(2)
235 		 * queue.  If we do, then accept(2) may hang after select(2)
236 		 * indicated that the listening socket was ready.
237 		 */
238 		if (!soqremque(so, 0)) {
239 			sounlock(so, s);
240 			return;
241 		}
242 	}
243 	sigio_free(&so->so_sigio);
244 #ifdef SOCKET_SPLICE
245 	if (so->so_sp) {
246 		if (issplicedback(so)) {
247 			int freeing = SOSP_FREEING_WRITE;
248 
249 			if (so->so_sp->ssp_soback == so)
250 				freeing |= SOSP_FREEING_READ;
251 			sounsplice(so->so_sp->ssp_soback, so, freeing);
252 		}
253 		if (isspliced(so)) {
254 			int freeing = SOSP_FREEING_READ;
255 
256 			if (so == so->so_sp->ssp_socket)
257 				freeing |= SOSP_FREEING_WRITE;
258 			sounsplice(so, so->so_sp->ssp_socket, freeing);
259 		}
260 	}
261 #endif /* SOCKET_SPLICE */
262 	sbrelease(so, &so->so_snd);
263 	sorflush(so);
264 	sounlock(so, s);
265 #ifdef SOCKET_SPLICE
266 	if (so->so_sp) {
267 		/* Reuse splice idle, sounsplice() has been called before. */
268 		timeout_set_proc(&so->so_sp->ssp_idleto, soreaper, so);
269 		timeout_add(&so->so_sp->ssp_idleto, 0);
270 	} else
271 #endif /* SOCKET_SPLICE */
272 	{
273 		pool_put(&socket_pool, so);
274 	}
275 }
276 
277 static inline uint64_t
278 solinger_nsec(struct socket *so)
279 {
280 	if (so->so_linger == 0)
281 		return INFSLP;
282 
283 	return SEC_TO_NSEC(so->so_linger);
284 }
285 
286 /*
287  * Close a socket on last file table reference removal.
288  * Initiate disconnect if connected.
289  * Free socket when disconnect complete.
290  */
291 int
292 soclose(struct socket *so, int flags)
293 {
294 	struct socket *so2;
295 	int s, error = 0;
296 
297 	s = solock(so);
298 	/* Revoke async IO early. There is a final revocation in sofree(). */
299 	sigio_free(&so->so_sigio);
300 	if (so->so_options & SO_ACCEPTCONN) {
301 		while ((so2 = TAILQ_FIRST(&so->so_q0)) != NULL) {
302 			(void) soqremque(so2, 0);
303 			(void) soabort(so2);
304 		}
305 		while ((so2 = TAILQ_FIRST(&so->so_q)) != NULL) {
306 			(void) soqremque(so2, 1);
307 			(void) soabort(so2);
308 		}
309 	}
310 	if (so->so_pcb == NULL)
311 		goto discard;
312 	if (so->so_state & SS_ISCONNECTED) {
313 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
314 			error = sodisconnect(so);
315 			if (error)
316 				goto drop;
317 		}
318 		if (so->so_options & SO_LINGER) {
319 			if ((so->so_state & SS_ISDISCONNECTING) &&
320 			    (flags & MSG_DONTWAIT))
321 				goto drop;
322 			while (so->so_state & SS_ISCONNECTED) {
323 				error = sosleep_nsec(so, &so->so_timeo,
324 				    PSOCK | PCATCH, "netcls",
325 				    solinger_nsec(so));
326 				if (error)
327 					break;
328 			}
329 		}
330 	}
331 drop:
332 	if (so->so_pcb) {
333 		int error2;
334 		KASSERT(so->so_proto->pr_detach);
335 		error2 = (*so->so_proto->pr_detach)(so);
336 		if (error == 0)
337 			error = error2;
338 	}
339 discard:
340 	if (so->so_state & SS_NOFDREF)
341 		panic("soclose NOFDREF: so %p, so_type %d", so, so->so_type);
342 	so->so_state |= SS_NOFDREF;
343 	/* sofree() calls sounlock(). */
344 	sofree(so, s);
345 	return (error);
346 }
347 
348 int
349 soabort(struct socket *so)
350 {
351 	soassertlocked(so);
352 
353 	return (*so->so_proto->pr_usrreq)(so, PRU_ABORT, NULL, NULL, NULL,
354 	   curproc);
355 }
356 
357 int
358 soaccept(struct socket *so, struct mbuf *nam)
359 {
360 	int error = 0;
361 
362 	soassertlocked(so);
363 
364 	if ((so->so_state & SS_NOFDREF) == 0)
365 		panic("soaccept !NOFDREF: so %p, so_type %d", so, so->so_type);
366 	so->so_state &= ~SS_NOFDREF;
367 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
368 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
369 		error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT, NULL,
370 		    nam, NULL, curproc);
371 	else
372 		error = ECONNABORTED;
373 	return (error);
374 }
375 
376 int
377 soconnect(struct socket *so, struct mbuf *nam)
378 {
379 	int error;
380 
381 	soassertlocked(so);
382 
383 	if (so->so_options & SO_ACCEPTCONN)
384 		return (EOPNOTSUPP);
385 	/*
386 	 * If protocol is connection-based, can only connect once.
387 	 * Otherwise, if connected, try to disconnect first.
388 	 * This allows user to disconnect by connecting to, e.g.,
389 	 * a null address.
390 	 */
391 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
392 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
393 	    (error = sodisconnect(so))))
394 		error = EISCONN;
395 	else
396 		error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT,
397 		    NULL, nam, NULL, curproc);
398 	return (error);
399 }
400 
401 int
402 soconnect2(struct socket *so1, struct socket *so2)
403 {
404 	int s, error;
405 
406 	s = solock(so1);
407 	error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2, NULL,
408 	    (struct mbuf *)so2, NULL, curproc);
409 	sounlock(so1, s);
410 	return (error);
411 }
412 
413 int
414 sodisconnect(struct socket *so)
415 {
416 	int error;
417 
418 	soassertlocked(so);
419 
420 	if ((so->so_state & SS_ISCONNECTED) == 0)
421 		return (ENOTCONN);
422 	if (so->so_state & SS_ISDISCONNECTING)
423 		return (EALREADY);
424 	error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT, NULL, NULL,
425 	    NULL, curproc);
426 	return (error);
427 }
428 
429 int m_getuio(struct mbuf **, int, long, struct uio *);
430 
431 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
432 /*
433  * Send on a socket.
434  * If send must go all at once and message is larger than
435  * send buffering, then hard error.
436  * Lock against other senders.
437  * If must go all at once and not enough room now, then
438  * inform user that this would block and do nothing.
439  * Otherwise, if nonblocking, send as much as possible.
440  * The data to be sent is described by "uio" if nonzero,
441  * otherwise by the mbuf chain "top" (which must be null
442  * if uio is not).  Data provided in mbuf chain must be small
443  * enough to send all at once.
444  *
445  * Returns nonzero on error, timeout or signal; callers
446  * must check for short counts if EINTR/ERESTART are returned.
447  * Data and control buffers are freed on return.
448  */
449 int
450 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
451     struct mbuf *control, int flags)
452 {
453 	long space, clen = 0;
454 	size_t resid;
455 	int error, s;
456 	int atomic = sosendallatonce(so) || top;
457 
458 	if (uio)
459 		resid = uio->uio_resid;
460 	else
461 		resid = top->m_pkthdr.len;
462 	/* MSG_EOR on a SOCK_STREAM socket is invalid. */
463 	if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
464 		m_freem(top);
465 		m_freem(control);
466 		return (EINVAL);
467 	}
468 	if (uio && uio->uio_procp)
469 		uio->uio_procp->p_ru.ru_msgsnd++;
470 	if (control) {
471 		/*
472 		 * In theory clen should be unsigned (since control->m_len is).
473 		 * However, space must be signed, as it might be less than 0
474 		 * if we over-committed, and we must use a signed comparison
475 		 * of space and clen.
476 		 */
477 		clen = control->m_len;
478 		/* reserve extra space for AF_UNIX's internalize */
479 		if (so->so_proto->pr_domain->dom_family == AF_UNIX &&
480 		    clen >= CMSG_ALIGN(sizeof(struct cmsghdr)) &&
481 		    mtod(control, struct cmsghdr *)->cmsg_type == SCM_RIGHTS)
482 			clen = CMSG_SPACE(
483 			    (clen - CMSG_ALIGN(sizeof(struct cmsghdr))) *
484 			    (sizeof(struct fdpass) / sizeof(int)));
485 	}
486 
487 #define	snderr(errno)	{ error = errno; goto release; }
488 
489 	s = solock(so);
490 restart:
491 	if ((error = sblock(so, &so->so_snd, SBLOCKWAIT(flags))) != 0)
492 		goto out;
493 	so->so_state |= SS_ISSENDING;
494 	do {
495 		if (so->so_state & SS_CANTSENDMORE)
496 			snderr(EPIPE);
497 		if (so->so_error) {
498 			error = so->so_error;
499 			so->so_error = 0;
500 			snderr(error);
501 		}
502 		if ((so->so_state & SS_ISCONNECTED) == 0) {
503 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
504 				if (!(resid == 0 && clen != 0))
505 					snderr(ENOTCONN);
506 			} else if (addr == NULL)
507 				snderr(EDESTADDRREQ);
508 		}
509 		space = sbspace(so, &so->so_snd);
510 		if (flags & MSG_OOB)
511 			space += 1024;
512 		if (so->so_proto->pr_domain->dom_family == AF_UNIX) {
513 			if (atomic && resid > so->so_snd.sb_hiwat)
514 				snderr(EMSGSIZE);
515 		} else {
516 			if (clen > so->so_snd.sb_hiwat ||
517 			    (atomic && resid > so->so_snd.sb_hiwat - clen))
518 				snderr(EMSGSIZE);
519 		}
520 		if (space < clen ||
521 		    (space - clen < resid &&
522 		    (atomic || space < so->so_snd.sb_lowat))) {
523 			if (flags & MSG_DONTWAIT)
524 				snderr(EWOULDBLOCK);
525 			sbunlock(so, &so->so_snd);
526 			error = sbwait(so, &so->so_snd);
527 			so->so_state &= ~SS_ISSENDING;
528 			if (error)
529 				goto out;
530 			goto restart;
531 		}
532 		space -= clen;
533 		do {
534 			if (uio == NULL) {
535 				/*
536 				 * Data is prepackaged in "top".
537 				 */
538 				resid = 0;
539 				if (flags & MSG_EOR)
540 					top->m_flags |= M_EOR;
541 			} else {
542 				sounlock(so, s);
543 				error = m_getuio(&top, atomic, space, uio);
544 				s = solock(so);
545 				if (error)
546 					goto release;
547 				space -= top->m_pkthdr.len;
548 				resid = uio->uio_resid;
549 				if (flags & MSG_EOR)
550 					top->m_flags |= M_EOR;
551 			}
552 			if (resid == 0)
553 				so->so_state &= ~SS_ISSENDING;
554 			if (top && so->so_options & SO_ZEROIZE)
555 				top->m_flags |= M_ZEROIZE;
556 			error = (*so->so_proto->pr_usrreq)(so,
557 			    (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND,
558 			    top, addr, control, curproc);
559 			clen = 0;
560 			control = NULL;
561 			top = NULL;
562 			if (error)
563 				goto release;
564 		} while (resid && space > 0);
565 	} while (resid);
566 
567 release:
568 	so->so_state &= ~SS_ISSENDING;
569 	sbunlock(so, &so->so_snd);
570 out:
571 	sounlock(so, s);
572 	m_freem(top);
573 	m_freem(control);
574 	return (error);
575 }
576 
577 int
578 m_getuio(struct mbuf **mp, int atomic, long space, struct uio *uio)
579 {
580 	struct mbuf *m, *top = NULL;
581 	struct mbuf **nextp = &top;
582 	u_long len, mlen;
583 	size_t resid = uio->uio_resid;
584 	int error;
585 
586 	do {
587 		if (top == NULL) {
588 			MGETHDR(m, M_WAIT, MT_DATA);
589 			mlen = MHLEN;
590 			m->m_pkthdr.len = 0;
591 			m->m_pkthdr.ph_ifidx = 0;
592 		} else {
593 			MGET(m, M_WAIT, MT_DATA);
594 			mlen = MLEN;
595 		}
596 		/* chain mbuf together */
597 		*nextp = m;
598 		nextp = &m->m_next;
599 
600 		resid = ulmin(resid, space);
601 		if (resid >= MINCLSIZE) {
602 			MCLGETL(m, M_NOWAIT, ulmin(resid, MAXMCLBYTES));
603 			if ((m->m_flags & M_EXT) == 0)
604 				MCLGETL(m, M_NOWAIT, MCLBYTES);
605 			if ((m->m_flags & M_EXT) == 0)
606 				goto nopages;
607 			mlen = m->m_ext.ext_size;
608 			len = ulmin(mlen, resid);
609 			/*
610 			 * For datagram protocols, leave room
611 			 * for protocol headers in first mbuf.
612 			 */
613 			if (atomic && m == top && len < mlen - max_hdr)
614 				m->m_data += max_hdr;
615 		} else {
616 nopages:
617 			len = ulmin(mlen, resid);
618 			/*
619 			 * For datagram protocols, leave room
620 			 * for protocol headers in first mbuf.
621 			 */
622 			if (atomic && m == top && len < mlen - max_hdr)
623 				m_align(m, len);
624 		}
625 
626 		error = uiomove(mtod(m, caddr_t), len, uio);
627 		if (error) {
628 			m_freem(top);
629 			return (error);
630 		}
631 
632 		/* adjust counters */
633 		resid = uio->uio_resid;
634 		space -= len;
635 		m->m_len = len;
636 		top->m_pkthdr.len += len;
637 
638 		/* Is there more space and more data? */
639 	} while (space > 0 && resid > 0);
640 
641 	*mp = top;
642 	return 0;
643 }
644 
645 /*
646  * Following replacement or removal of the first mbuf on the first
647  * mbuf chain of a socket buffer, push necessary state changes back
648  * into the socket buffer so that other consumers see the values
649  * consistently.  'nextrecord' is the callers locally stored value of
650  * the original value of sb->sb_mb->m_nextpkt which must be restored
651  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
652  */
653 void
654 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
655 {
656 
657 	/*
658 	 * First, update for the new value of nextrecord.  If necessary,
659 	 * make it the first record.
660 	 */
661 	if (sb->sb_mb != NULL)
662 		sb->sb_mb->m_nextpkt = nextrecord;
663 	else
664 		sb->sb_mb = nextrecord;
665 
666 	/*
667 	 * Now update any dependent socket buffer fields to reflect
668 	 * the new state.  This is an inline of SB_EMPTY_FIXUP, with
669 	 * the addition of a second clause that takes care of the
670 	 * case where sb_mb has been updated, but remains the last
671 	 * record.
672 	 */
673 	if (sb->sb_mb == NULL) {
674 		sb->sb_mbtail = NULL;
675 		sb->sb_lastrecord = NULL;
676 	} else if (sb->sb_mb->m_nextpkt == NULL)
677 		sb->sb_lastrecord = sb->sb_mb;
678 }
679 
680 /*
681  * Implement receive operations on a socket.
682  * We depend on the way that records are added to the sockbuf
683  * by sbappend*.  In particular, each record (mbufs linked through m_next)
684  * must begin with an address if the protocol so specifies,
685  * followed by an optional mbuf or mbufs containing ancillary data,
686  * and then zero or more mbufs of data.
687  * In order to avoid blocking network for the entire time here, we release
688  * the solock() while doing the actual copy to user space.
689  * Although the sockbuf is locked, new data may still be appended,
690  * and thus we must maintain consistency of the sockbuf during that time.
691  *
692  * The caller may receive the data as a single mbuf chain by supplying
693  * an mbuf **mp0 for use in returning the chain.  The uio is then used
694  * only for the count in uio_resid.
695  */
696 int
697 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
698     struct mbuf **mp0, struct mbuf **controlp, int *flagsp,
699     socklen_t controllen)
700 {
701 	struct mbuf *m, **mp;
702 	struct mbuf *cm;
703 	u_long len, offset, moff;
704 	int flags, error, s, type, uio_error = 0;
705 	const struct protosw *pr = so->so_proto;
706 	struct mbuf *nextrecord;
707 	size_t resid, orig_resid = uio->uio_resid;
708 
709 	mp = mp0;
710 	if (paddr)
711 		*paddr = NULL;
712 	if (controlp)
713 		*controlp = NULL;
714 	if (flagsp)
715 		flags = *flagsp &~ MSG_EOR;
716 	else
717 		flags = 0;
718 	if (flags & MSG_OOB) {
719 		m = m_get(M_WAIT, MT_DATA);
720 		s = solock(so);
721 		error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m,
722 		    (struct mbuf *)(long)(flags & MSG_PEEK), NULL, curproc);
723 		sounlock(so, s);
724 		if (error)
725 			goto bad;
726 		do {
727 			error = uiomove(mtod(m, caddr_t),
728 			    ulmin(uio->uio_resid, m->m_len), uio);
729 			m = m_free(m);
730 		} while (uio->uio_resid && error == 0 && m);
731 bad:
732 		m_freem(m);
733 		return (error);
734 	}
735 	if (mp)
736 		*mp = NULL;
737 
738 	s = solock(so);
739 restart:
740 	if ((error = sblock(so, &so->so_rcv, SBLOCKWAIT(flags))) != 0) {
741 		sounlock(so, s);
742 		return (error);
743 	}
744 
745 	m = so->so_rcv.sb_mb;
746 #ifdef SOCKET_SPLICE
747 	if (isspliced(so))
748 		m = NULL;
749 #endif /* SOCKET_SPLICE */
750 	/*
751 	 * If we have less data than requested, block awaiting more
752 	 * (subject to any timeout) if:
753 	 *   1. the current count is less than the low water mark,
754 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
755 	 *	receive operation at once if we block (resid <= hiwat), or
756 	 *   3. MSG_DONTWAIT is not set.
757 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
758 	 * we have to do the receive in sections, and thus risk returning
759 	 * a short count if a timeout or signal occurs after we start.
760 	 */
761 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
762 	    so->so_rcv.sb_cc < uio->uio_resid) &&
763 	    (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
764 	    ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
765 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
766 #ifdef DIAGNOSTIC
767 		if (m == NULL && so->so_rcv.sb_cc)
768 #ifdef SOCKET_SPLICE
769 		    if (!isspliced(so))
770 #endif /* SOCKET_SPLICE */
771 			panic("receive 1: so %p, so_type %d, sb_cc %lu",
772 			    so, so->so_type, so->so_rcv.sb_cc);
773 #endif
774 		if (so->so_error) {
775 			if (m)
776 				goto dontblock;
777 			error = so->so_error;
778 			if ((flags & MSG_PEEK) == 0)
779 				so->so_error = 0;
780 			goto release;
781 		}
782 		if (so->so_state & SS_CANTRCVMORE) {
783 			if (m)
784 				goto dontblock;
785 			else if (so->so_rcv.sb_cc == 0)
786 				goto release;
787 		}
788 		for (; m; m = m->m_next)
789 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
790 				m = so->so_rcv.sb_mb;
791 				goto dontblock;
792 			}
793 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
794 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
795 			error = ENOTCONN;
796 			goto release;
797 		}
798 		if (uio->uio_resid == 0 && controlp == NULL)
799 			goto release;
800 		if (flags & MSG_DONTWAIT) {
801 			error = EWOULDBLOCK;
802 			goto release;
803 		}
804 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
805 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
806 		sbunlock(so, &so->so_rcv);
807 		error = sbwait(so, &so->so_rcv);
808 		if (error) {
809 			sounlock(so, s);
810 			return (error);
811 		}
812 		goto restart;
813 	}
814 dontblock:
815 	/*
816 	 * On entry here, m points to the first record of the socket buffer.
817 	 * From this point onward, we maintain 'nextrecord' as a cache of the
818 	 * pointer to the next record in the socket buffer.  We must keep the
819 	 * various socket buffer pointers and local stack versions of the
820 	 * pointers in sync, pushing out modifications before operations that
821 	 * may sleep, and re-reading them afterwards.
822 	 *
823 	 * Otherwise, we will race with the network stack appending new data
824 	 * or records onto the socket buffer by using inconsistent/stale
825 	 * versions of the field, possibly resulting in socket buffer
826 	 * corruption.
827 	 */
828 	if (uio->uio_procp)
829 		uio->uio_procp->p_ru.ru_msgrcv++;
830 	KASSERT(m == so->so_rcv.sb_mb);
831 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
832 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
833 	nextrecord = m->m_nextpkt;
834 	if (pr->pr_flags & PR_ADDR) {
835 #ifdef DIAGNOSTIC
836 		if (m->m_type != MT_SONAME)
837 			panic("receive 1a: so %p, so_type %d, m %p, m_type %d",
838 			    so, so->so_type, m, m->m_type);
839 #endif
840 		orig_resid = 0;
841 		if (flags & MSG_PEEK) {
842 			if (paddr)
843 				*paddr = m_copym(m, 0, m->m_len, M_NOWAIT);
844 			m = m->m_next;
845 		} else {
846 			sbfree(&so->so_rcv, m);
847 			if (paddr) {
848 				*paddr = m;
849 				so->so_rcv.sb_mb = m->m_next;
850 				m->m_next = NULL;
851 				m = so->so_rcv.sb_mb;
852 			} else {
853 				so->so_rcv.sb_mb = m_free(m);
854 				m = so->so_rcv.sb_mb;
855 			}
856 			sbsync(&so->so_rcv, nextrecord);
857 		}
858 	}
859 	while (m && m->m_type == MT_CONTROL && error == 0) {
860 		int skip = 0;
861 		if (flags & MSG_PEEK) {
862 			if (mtod(m, struct cmsghdr *)->cmsg_type ==
863 			    SCM_RIGHTS) {
864 				/* don't leak internalized SCM_RIGHTS msgs */
865 				skip = 1;
866 			} else if (controlp)
867 				*controlp = m_copym(m, 0, m->m_len, M_NOWAIT);
868 			m = m->m_next;
869 		} else {
870 			sbfree(&so->so_rcv, m);
871 			so->so_rcv.sb_mb = m->m_next;
872 			m->m_nextpkt = m->m_next = NULL;
873 			cm = m;
874 			m = so->so_rcv.sb_mb;
875 			sbsync(&so->so_rcv, nextrecord);
876 			if (controlp) {
877 				if (pr->pr_domain->dom_externalize) {
878 					error =
879 					    (*pr->pr_domain->dom_externalize)
880 					    (cm, controllen, flags);
881 				}
882 				*controlp = cm;
883 			} else {
884 				/*
885 				 * Dispose of any SCM_RIGHTS message that went
886 				 * through the read path rather than recv.
887 				 */
888 				if (pr->pr_domain->dom_dispose)
889 					pr->pr_domain->dom_dispose(cm);
890 				m_free(cm);
891 			}
892 		}
893 		if (m != NULL)
894 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
895 		else
896 			nextrecord = so->so_rcv.sb_mb;
897 		if (controlp && !skip)
898 			controlp = &(*controlp)->m_next;
899 		orig_resid = 0;
900 	}
901 
902 	/* If m is non-NULL, we have some data to read. */
903 	if (m) {
904 		type = m->m_type;
905 		if (type == MT_OOBDATA)
906 			flags |= MSG_OOB;
907 		if (m->m_flags & M_BCAST)
908 			flags |= MSG_BCAST;
909 		if (m->m_flags & M_MCAST)
910 			flags |= MSG_MCAST;
911 	}
912 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
913 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
914 
915 	moff = 0;
916 	offset = 0;
917 	while (m && uio->uio_resid > 0 && error == 0) {
918 		if (m->m_type == MT_OOBDATA) {
919 			if (type != MT_OOBDATA)
920 				break;
921 		} else if (type == MT_OOBDATA) {
922 			break;
923 		} else if (m->m_type == MT_CONTROL) {
924 			/*
925 			 * If there is more than one control message in the
926 			 * stream, we do a short read.  Next can be received
927 			 * or disposed by another system call.
928 			 */
929 			break;
930 #ifdef DIAGNOSTIC
931 		} else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) {
932 			panic("receive 3: so %p, so_type %d, m %p, m_type %d",
933 			    so, so->so_type, m, m->m_type);
934 #endif
935 		}
936 		so->so_state &= ~SS_RCVATMARK;
937 		len = uio->uio_resid;
938 		if (so->so_oobmark && len > so->so_oobmark - offset)
939 			len = so->so_oobmark - offset;
940 		if (len > m->m_len - moff)
941 			len = m->m_len - moff;
942 		/*
943 		 * If mp is set, just pass back the mbufs.
944 		 * Otherwise copy them out via the uio, then free.
945 		 * Sockbuf must be consistent here (points to current mbuf,
946 		 * it points to next record) when we drop priority;
947 		 * we must note any additions to the sockbuf when we
948 		 * block interrupts again.
949 		 */
950 		if (mp == NULL && uio_error == 0) {
951 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
952 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
953 			resid = uio->uio_resid;
954 			sounlock(so, s);
955 			uio_error = uiomove(mtod(m, caddr_t) + moff, len, uio);
956 			s = solock(so);
957 			if (uio_error)
958 				uio->uio_resid = resid - len;
959 		} else
960 			uio->uio_resid -= len;
961 		if (len == m->m_len - moff) {
962 			if (m->m_flags & M_EOR)
963 				flags |= MSG_EOR;
964 			if (flags & MSG_PEEK) {
965 				m = m->m_next;
966 				moff = 0;
967 				orig_resid = 0;
968 			} else {
969 				nextrecord = m->m_nextpkt;
970 				sbfree(&so->so_rcv, m);
971 				if (mp) {
972 					*mp = m;
973 					mp = &m->m_next;
974 					so->so_rcv.sb_mb = m = m->m_next;
975 					*mp = NULL;
976 				} else {
977 					so->so_rcv.sb_mb = m_free(m);
978 					m = so->so_rcv.sb_mb;
979 				}
980 				/*
981 				 * If m != NULL, we also know that
982 				 * so->so_rcv.sb_mb != NULL.
983 				 */
984 				KASSERT(so->so_rcv.sb_mb == m);
985 				if (m) {
986 					m->m_nextpkt = nextrecord;
987 					if (nextrecord == NULL)
988 						so->so_rcv.sb_lastrecord = m;
989 				} else {
990 					so->so_rcv.sb_mb = nextrecord;
991 					SB_EMPTY_FIXUP(&so->so_rcv);
992 				}
993 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
994 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
995 			}
996 		} else {
997 			if (flags & MSG_PEEK) {
998 				moff += len;
999 				orig_resid = 0;
1000 			} else {
1001 				if (mp)
1002 					*mp = m_copym(m, 0, len, M_WAIT);
1003 				m->m_data += len;
1004 				m->m_len -= len;
1005 				so->so_rcv.sb_cc -= len;
1006 				so->so_rcv.sb_datacc -= len;
1007 			}
1008 		}
1009 		if (so->so_oobmark) {
1010 			if ((flags & MSG_PEEK) == 0) {
1011 				so->so_oobmark -= len;
1012 				if (so->so_oobmark == 0) {
1013 					so->so_state |= SS_RCVATMARK;
1014 					break;
1015 				}
1016 			} else {
1017 				offset += len;
1018 				if (offset == so->so_oobmark)
1019 					break;
1020 			}
1021 		}
1022 		if (flags & MSG_EOR)
1023 			break;
1024 		/*
1025 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1026 		 * we must not quit until "uio->uio_resid == 0" or an error
1027 		 * termination.  If a signal/timeout occurs, return
1028 		 * with a short count but without error.
1029 		 * Keep sockbuf locked against other readers.
1030 		 */
1031 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1032 		    !sosendallatonce(so) && !nextrecord) {
1033 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1034 				break;
1035 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1036 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1037 			error = sbwait(so, &so->so_rcv);
1038 			if (error) {
1039 				sbunlock(so, &so->so_rcv);
1040 				sounlock(so, s);
1041 				return (0);
1042 			}
1043 			if ((m = so->so_rcv.sb_mb) != NULL)
1044 				nextrecord = m->m_nextpkt;
1045 		}
1046 	}
1047 
1048 	if (m && pr->pr_flags & PR_ATOMIC) {
1049 		flags |= MSG_TRUNC;
1050 		if ((flags & MSG_PEEK) == 0)
1051 			(void) sbdroprecord(&so->so_rcv);
1052 	}
1053 	if ((flags & MSG_PEEK) == 0) {
1054 		if (m == NULL) {
1055 			/*
1056 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1057 			 * part makes sure sb_lastrecord is up-to-date if
1058 			 * there is still data in the socket buffer.
1059 			 */
1060 			so->so_rcv.sb_mb = nextrecord;
1061 			if (so->so_rcv.sb_mb == NULL) {
1062 				so->so_rcv.sb_mbtail = NULL;
1063 				so->so_rcv.sb_lastrecord = NULL;
1064 			} else if (nextrecord->m_nextpkt == NULL)
1065 				so->so_rcv.sb_lastrecord = nextrecord;
1066 		}
1067 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1068 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1069 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1070 			(*pr->pr_usrreq)(so, PRU_RCVD, NULL,
1071 			    (struct mbuf *)(long)flags, NULL, curproc);
1072 	}
1073 	if (orig_resid == uio->uio_resid && orig_resid &&
1074 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1075 		sbunlock(so, &so->so_rcv);
1076 		goto restart;
1077 	}
1078 
1079 	if (uio_error)
1080 		error = uio_error;
1081 
1082 	if (flagsp)
1083 		*flagsp |= flags;
1084 release:
1085 	sbunlock(so, &so->so_rcv);
1086 	sounlock(so, s);
1087 	return (error);
1088 }
1089 
1090 int
1091 soshutdown(struct socket *so, int how)
1092 {
1093 	const struct protosw *pr = so->so_proto;
1094 	int s, error = 0;
1095 
1096 	s = solock(so);
1097 	switch (how) {
1098 	case SHUT_RD:
1099 		sorflush(so);
1100 		break;
1101 	case SHUT_RDWR:
1102 		sorflush(so);
1103 		/* FALLTHROUGH */
1104 	case SHUT_WR:
1105 		error = (*pr->pr_usrreq)(so, PRU_SHUTDOWN, NULL, NULL, NULL,
1106 		    curproc);
1107 		break;
1108 	default:
1109 		error = EINVAL;
1110 		break;
1111 	}
1112 	sounlock(so, s);
1113 
1114 	return (error);
1115 }
1116 
1117 void
1118 sorflush(struct socket *so)
1119 {
1120 	struct sockbuf *sb = &so->so_rcv;
1121 	struct mbuf *m;
1122 	const struct protosw *pr = so->so_proto;
1123 	int error;
1124 
1125 	sb->sb_flags |= SB_NOINTR;
1126 	error = sblock(so, sb, M_WAITOK);
1127 	/* with SB_NOINTR and M_WAITOK sblock() must not fail */
1128 	KASSERT(error == 0);
1129 	socantrcvmore(so);
1130 	m = sb->sb_mb;
1131 	memset(&sb->sb_startzero, 0,
1132 	     (caddr_t)&sb->sb_endzero - (caddr_t)&sb->sb_startzero);
1133 	sb->sb_timeo_nsecs = INFSLP;
1134 	sbunlock(so, sb);
1135 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose)
1136 		(*pr->pr_domain->dom_dispose)(m);
1137 	m_purge(m);
1138 }
1139 
1140 #ifdef SOCKET_SPLICE
1141 
1142 #define so_splicelen	so_sp->ssp_len
1143 #define so_splicemax	so_sp->ssp_max
1144 #define so_idletv	so_sp->ssp_idletv
1145 #define so_idleto	so_sp->ssp_idleto
1146 #define so_splicetask	so_sp->ssp_task
1147 
1148 int
1149 sosplice(struct socket *so, int fd, off_t max, struct timeval *tv)
1150 {
1151 	struct file	*fp;
1152 	struct socket	*sosp;
1153 	struct sosplice	*sp;
1154 	struct taskq	*tq;
1155 	int		 error = 0;
1156 
1157 	soassertlocked(so);
1158 
1159 	if (sosplice_taskq == NULL) {
1160 		rw_enter_write(&sosplice_lock);
1161 		if (sosplice_taskq == NULL) {
1162 			tq = taskq_create("sosplice", 1, IPL_SOFTNET,
1163 			    TASKQ_MPSAFE);
1164 			/* Ensure the taskq is fully visible to other CPUs. */
1165 			membar_producer();
1166 			sosplice_taskq = tq;
1167 		}
1168 		rw_exit_write(&sosplice_lock);
1169 	}
1170 	if (sosplice_taskq == NULL)
1171 		return (ENOMEM);
1172 
1173 	if ((so->so_proto->pr_flags & PR_SPLICE) == 0)
1174 		return (EPROTONOSUPPORT);
1175 	if (so->so_options & SO_ACCEPTCONN)
1176 		return (EOPNOTSUPP);
1177 	if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1178 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
1179 		return (ENOTCONN);
1180 	if (so->so_sp == NULL) {
1181 		sp = pool_get(&sosplice_pool, PR_WAITOK | PR_ZERO);
1182 		if (so->so_sp == NULL)
1183 			so->so_sp = sp;
1184 		else
1185 			pool_put(&sosplice_pool, sp);
1186 	}
1187 
1188 	/* If no fd is given, unsplice by removing existing link. */
1189 	if (fd < 0) {
1190 		/* Lock receive buffer. */
1191 		if ((error = sblock(so, &so->so_rcv, M_WAITOK)) != 0) {
1192 			return (error);
1193 		}
1194 		if (so->so_sp->ssp_socket)
1195 			sounsplice(so, so->so_sp->ssp_socket, 0);
1196 		sbunlock(so, &so->so_rcv);
1197 		return (0);
1198 	}
1199 
1200 	if (max && max < 0)
1201 		return (EINVAL);
1202 
1203 	if (tv && (tv->tv_sec < 0 || !timerisvalid(tv)))
1204 		return (EINVAL);
1205 
1206 	/* Find sosp, the drain socket where data will be spliced into. */
1207 	if ((error = getsock(curproc, fd, &fp)) != 0)
1208 		return (error);
1209 	sosp = fp->f_data;
1210 	if (sosp->so_proto->pr_usrreq != so->so_proto->pr_usrreq) {
1211 		error = EPROTONOSUPPORT;
1212 		goto frele;
1213 	}
1214 	if (sosp->so_sp == NULL) {
1215 		sp = pool_get(&sosplice_pool, PR_WAITOK | PR_ZERO);
1216 		if (sosp->so_sp == NULL)
1217 			sosp->so_sp = sp;
1218 		else
1219 			pool_put(&sosplice_pool, sp);
1220 	}
1221 
1222 	/* Lock both receive and send buffer. */
1223 	if ((error = sblock(so, &so->so_rcv, M_WAITOK)) != 0) {
1224 		goto frele;
1225 	}
1226 	if ((error = sblock(so, &sosp->so_snd, M_WAITOK)) != 0) {
1227 		sbunlock(so, &so->so_rcv);
1228 		goto frele;
1229 	}
1230 
1231 	if (so->so_sp->ssp_socket || sosp->so_sp->ssp_soback) {
1232 		error = EBUSY;
1233 		goto release;
1234 	}
1235 	if (sosp->so_options & SO_ACCEPTCONN) {
1236 		error = EOPNOTSUPP;
1237 		goto release;
1238 	}
1239 	if ((sosp->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0) {
1240 		error = ENOTCONN;
1241 		goto release;
1242 	}
1243 
1244 	/* Splice so and sosp together. */
1245 	so->so_sp->ssp_socket = sosp;
1246 	sosp->so_sp->ssp_soback = so;
1247 	so->so_splicelen = 0;
1248 	so->so_splicemax = max;
1249 	if (tv)
1250 		so->so_idletv = *tv;
1251 	else
1252 		timerclear(&so->so_idletv);
1253 	timeout_set_proc(&so->so_idleto, soidle, so);
1254 	task_set(&so->so_splicetask, sotask, so);
1255 
1256 	/*
1257 	 * To prevent softnet interrupt from calling somove() while
1258 	 * we sleep, the socket buffers are not marked as spliced yet.
1259 	 */
1260 	if (somove(so, M_WAIT)) {
1261 		so->so_rcv.sb_flags |= SB_SPLICE;
1262 		sosp->so_snd.sb_flags |= SB_SPLICE;
1263 	}
1264 
1265  release:
1266 	sbunlock(sosp, &sosp->so_snd);
1267 	sbunlock(so, &so->so_rcv);
1268  frele:
1269 	/*
1270 	 * FRELE() must not be called with the socket lock held. It is safe to
1271 	 * release the lock here as long as no other operation happen on the
1272 	 * socket when sosplice() returns. The dance could be avoided by
1273 	 * grabbing the socket lock inside this function.
1274 	 */
1275 	sounlock(so, SL_LOCKED);
1276 	FRELE(fp, curproc);
1277 	solock(so);
1278 	return (error);
1279 }
1280 
1281 void
1282 sounsplice(struct socket *so, struct socket *sosp, int freeing)
1283 {
1284 	soassertlocked(so);
1285 
1286 	task_del(sosplice_taskq, &so->so_splicetask);
1287 	timeout_del(&so->so_idleto);
1288 	sosp->so_snd.sb_flags &= ~SB_SPLICE;
1289 	so->so_rcv.sb_flags &= ~SB_SPLICE;
1290 	so->so_sp->ssp_socket = sosp->so_sp->ssp_soback = NULL;
1291 	/* Do not wakeup a socket that is about to be freed. */
1292 	if ((freeing & SOSP_FREEING_READ) == 0 && soreadable(so))
1293 		sorwakeup(so);
1294 	if ((freeing & SOSP_FREEING_WRITE) == 0 && sowriteable(sosp))
1295 		sowwakeup(sosp);
1296 }
1297 
1298 void
1299 soidle(void *arg)
1300 {
1301 	struct socket *so = arg;
1302 	int s;
1303 
1304 	s = solock(so);
1305 	if (so->so_rcv.sb_flags & SB_SPLICE) {
1306 		so->so_error = ETIMEDOUT;
1307 		sounsplice(so, so->so_sp->ssp_socket, 0);
1308 	}
1309 	sounlock(so, s);
1310 }
1311 
1312 void
1313 sotask(void *arg)
1314 {
1315 	struct socket *so = arg;
1316 	int s;
1317 
1318 	s = solock(so);
1319 	if (so->so_rcv.sb_flags & SB_SPLICE) {
1320 		/*
1321 		 * We may not sleep here as sofree() and unsplice() may be
1322 		 * called from softnet interrupt context.  This would remove
1323 		 * the socket during somove().
1324 		 */
1325 		somove(so, M_DONTWAIT);
1326 	}
1327 	sounlock(so, s);
1328 
1329 	/* Avoid user land starvation. */
1330 	yield();
1331 }
1332 
1333 /*
1334  * The socket splicing task or idle timeout may sleep while grabbing the net
1335  * lock.  As sofree() can be called anytime, sotask() or soidle() could access
1336  * the socket memory of a freed socket after wakeup.  So delay the pool_put()
1337  * after all pending socket splicing tasks or timeouts have finished.  Do this
1338  * by scheduling it on the same threads.
1339  */
1340 void
1341 soreaper(void *arg)
1342 {
1343 	struct socket *so = arg;
1344 
1345 	/* Reuse splice task, sounsplice() has been called before. */
1346 	task_set(&so->so_sp->ssp_task, soput, so);
1347 	task_add(sosplice_taskq, &so->so_sp->ssp_task);
1348 }
1349 
1350 void
1351 soput(void *arg)
1352 {
1353 	struct socket *so = arg;
1354 
1355 	pool_put(&sosplice_pool, so->so_sp);
1356 	pool_put(&socket_pool, so);
1357 }
1358 
1359 /*
1360  * Move data from receive buffer of spliced source socket to send
1361  * buffer of drain socket.  Try to move as much as possible in one
1362  * big chunk.  It is a TCP only implementation.
1363  * Return value 0 means splicing has been finished, 1 continue.
1364  */
1365 int
1366 somove(struct socket *so, int wait)
1367 {
1368 	struct socket	*sosp = so->so_sp->ssp_socket;
1369 	struct mbuf	*m, **mp, *nextrecord;
1370 	u_long		 len, off, oobmark;
1371 	long		 space;
1372 	int		 error = 0, maxreached = 0;
1373 	unsigned int	 state;
1374 
1375 	soassertlocked(so);
1376 
1377  nextpkt:
1378 	if (so->so_error) {
1379 		error = so->so_error;
1380 		goto release;
1381 	}
1382 	if (sosp->so_state & SS_CANTSENDMORE) {
1383 		error = EPIPE;
1384 		goto release;
1385 	}
1386 	if (sosp->so_error && sosp->so_error != ETIMEDOUT &&
1387 	    sosp->so_error != EFBIG && sosp->so_error != ELOOP) {
1388 		error = sosp->so_error;
1389 		goto release;
1390 	}
1391 	if ((sosp->so_state & SS_ISCONNECTED) == 0)
1392 		goto release;
1393 
1394 	/* Calculate how many bytes can be copied now. */
1395 	len = so->so_rcv.sb_datacc;
1396 	if (so->so_splicemax) {
1397 		KASSERT(so->so_splicelen < so->so_splicemax);
1398 		if (so->so_splicemax <= so->so_splicelen + len) {
1399 			len = so->so_splicemax - so->so_splicelen;
1400 			maxreached = 1;
1401 		}
1402 	}
1403 	space = sbspace(sosp, &sosp->so_snd);
1404 	if (so->so_oobmark && so->so_oobmark < len &&
1405 	    so->so_oobmark < space + 1024)
1406 		space += 1024;
1407 	if (space <= 0) {
1408 		maxreached = 0;
1409 		goto release;
1410 	}
1411 	if (space < len) {
1412 		maxreached = 0;
1413 		if (space < sosp->so_snd.sb_lowat)
1414 			goto release;
1415 		len = space;
1416 	}
1417 	sosp->so_state |= SS_ISSENDING;
1418 
1419 	SBLASTRECORDCHK(&so->so_rcv, "somove 1");
1420 	SBLASTMBUFCHK(&so->so_rcv, "somove 1");
1421 	m = so->so_rcv.sb_mb;
1422 	if (m == NULL)
1423 		goto release;
1424 	nextrecord = m->m_nextpkt;
1425 
1426 	/* Drop address and control information not used with splicing. */
1427 	if (so->so_proto->pr_flags & PR_ADDR) {
1428 #ifdef DIAGNOSTIC
1429 		if (m->m_type != MT_SONAME)
1430 			panic("somove soname: so %p, so_type %d, m %p, "
1431 			    "m_type %d", so, so->so_type, m, m->m_type);
1432 #endif
1433 		m = m->m_next;
1434 	}
1435 	while (m && m->m_type == MT_CONTROL)
1436 		m = m->m_next;
1437 	if (m == NULL) {
1438 		sbdroprecord(&so->so_rcv);
1439 		if (so->so_proto->pr_flags & PR_WANTRCVD && so->so_pcb)
1440 			(so->so_proto->pr_usrreq)(so, PRU_RCVD, NULL,
1441 			    NULL, NULL, NULL);
1442 		goto nextpkt;
1443 	}
1444 
1445 	/*
1446 	 * By splicing sockets connected to localhost, userland might create a
1447 	 * loop.  Dissolve splicing with error if loop is detected by counter.
1448 	 *
1449 	 * If we deal with looped broadcast/multicast packet we bail out with
1450 	 * no error to suppress splice termination.
1451 	 */
1452 	if ((m->m_flags & M_PKTHDR) &&
1453 	    ((m->m_pkthdr.ph_loopcnt++ >= M_MAXLOOP) ||
1454 	    ((m->m_flags & M_LOOP) && (m->m_flags & (M_BCAST|M_MCAST))))) {
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 
2045 	return (0);
2046 }
2047 
2048 void
2049 filt_sordetach(struct knote *kn)
2050 {
2051 	struct socket *so = kn->kn_fp->f_data;
2052 
2053 	KERNEL_ASSERT_LOCKED();
2054 
2055 	klist_remove_locked(&so->so_rcv.sb_sel.si_note, kn);
2056 }
2057 
2058 int
2059 filt_soread(struct knote *kn, long hint)
2060 {
2061 	struct socket *so = kn->kn_fp->f_data;
2062 	int s, rv = 0;
2063 
2064 	if ((hint & NOTE_SUBMIT) == 0)
2065 		s = solock(so);
2066 	kn->kn_data = so->so_rcv.sb_cc;
2067 #ifdef SOCKET_SPLICE
2068 	if (isspliced(so)) {
2069 		rv = 0;
2070 	} else
2071 #endif /* SOCKET_SPLICE */
2072 	if (kn->kn_sfflags & NOTE_OOB) {
2073 		if (so->so_oobmark || (so->so_state & SS_RCVATMARK)) {
2074 			kn->kn_fflags |= NOTE_OOB;
2075 			kn->kn_data -= so->so_oobmark;
2076 			rv = 1;
2077 		}
2078 	} else if (so->so_state & SS_CANTRCVMORE) {
2079 		kn->kn_flags |= EV_EOF;
2080 		if (kn->kn_flags & __EV_POLL) {
2081 			if (so->so_state & SS_ISDISCONNECTED)
2082 				kn->kn_flags |= __EV_HUP;
2083 		}
2084 		kn->kn_fflags = so->so_error;
2085 		rv = 1;
2086 	} else if (so->so_error) {	/* temporary udp error */
2087 		rv = 1;
2088 	} else if (kn->kn_sfflags & NOTE_LOWAT) {
2089 		rv = (kn->kn_data >= kn->kn_sdata);
2090 	} else {
2091 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
2092 	}
2093 	if ((hint & NOTE_SUBMIT) == 0)
2094 		sounlock(so, s);
2095 
2096 	return rv;
2097 }
2098 
2099 void
2100 filt_sowdetach(struct knote *kn)
2101 {
2102 	struct socket *so = kn->kn_fp->f_data;
2103 
2104 	KERNEL_ASSERT_LOCKED();
2105 
2106 	klist_remove_locked(&so->so_snd.sb_sel.si_note, kn);
2107 }
2108 
2109 int
2110 filt_sowrite(struct knote *kn, long hint)
2111 {
2112 	struct socket *so = kn->kn_fp->f_data;
2113 	int s, rv;
2114 
2115 	if ((hint & NOTE_SUBMIT) == 0)
2116 		s = solock(so);
2117 	kn->kn_data = sbspace(so, &so->so_snd);
2118 	if (so->so_state & SS_CANTSENDMORE) {
2119 		kn->kn_flags |= EV_EOF;
2120 		if (kn->kn_flags & __EV_POLL) {
2121 			if (so->so_state & SS_ISDISCONNECTED)
2122 				kn->kn_flags |= __EV_HUP;
2123 		}
2124 		kn->kn_fflags = so->so_error;
2125 		rv = 1;
2126 	} else if (so->so_error) {	/* temporary udp error */
2127 		rv = 1;
2128 	} else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2129 	    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
2130 		rv = 0;
2131 	} else if (kn->kn_sfflags & NOTE_LOWAT) {
2132 		rv = (kn->kn_data >= kn->kn_sdata);
2133 	} else {
2134 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
2135 	}
2136 	if ((hint & NOTE_SUBMIT) == 0)
2137 		sounlock(so, s);
2138 
2139 	return (rv);
2140 }
2141 
2142 int
2143 filt_solisten(struct knote *kn, long hint)
2144 {
2145 	struct socket *so = kn->kn_fp->f_data;
2146 	int s;
2147 
2148 	if ((hint & NOTE_SUBMIT) == 0)
2149 		s = solock(so);
2150 	kn->kn_data = so->so_qlen;
2151 	if ((hint & NOTE_SUBMIT) == 0)
2152 		sounlock(so, s);
2153 
2154 	return (kn->kn_data != 0);
2155 }
2156 
2157 #ifdef DDB
2158 void
2159 sobuf_print(struct sockbuf *,
2160     int (*)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))));
2161 
2162 void
2163 sobuf_print(struct sockbuf *sb,
2164     int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2165 {
2166 	(*pr)("\tsb_cc: %lu\n", sb->sb_cc);
2167 	(*pr)("\tsb_datacc: %lu\n", sb->sb_datacc);
2168 	(*pr)("\tsb_hiwat: %lu\n", sb->sb_hiwat);
2169 	(*pr)("\tsb_wat: %lu\n", sb->sb_wat);
2170 	(*pr)("\tsb_mbcnt: %lu\n", sb->sb_mbcnt);
2171 	(*pr)("\tsb_mbmax: %lu\n", sb->sb_mbmax);
2172 	(*pr)("\tsb_lowat: %ld\n", sb->sb_lowat);
2173 	(*pr)("\tsb_mb: %p\n", sb->sb_mb);
2174 	(*pr)("\tsb_mbtail: %p\n", sb->sb_mbtail);
2175 	(*pr)("\tsb_lastrecord: %p\n", sb->sb_lastrecord);
2176 	(*pr)("\tsb_sel: ...\n");
2177 	(*pr)("\tsb_flags: %i\n", sb->sb_flags);
2178 	(*pr)("\tsb_timeo_nsecs: %llu\n", sb->sb_timeo_nsecs);
2179 }
2180 
2181 void
2182 so_print(void *v,
2183     int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2184 {
2185 	struct socket *so = v;
2186 
2187 	(*pr)("socket %p\n", so);
2188 	(*pr)("so_type: %i\n", so->so_type);
2189 	(*pr)("so_options: 0x%04x\n", so->so_options); /* %b */
2190 	(*pr)("so_linger: %i\n", so->so_linger);
2191 	(*pr)("so_state: 0x%04x\n", so->so_state);
2192 	(*pr)("so_pcb: %p\n", so->so_pcb);
2193 	(*pr)("so_proto: %p\n", so->so_proto);
2194 	(*pr)("so_sigio: %p\n", so->so_sigio.sir_sigio);
2195 
2196 	(*pr)("so_head: %p\n", so->so_head);
2197 	(*pr)("so_onq: %p\n", so->so_onq);
2198 	(*pr)("so_q0: @%p first: %p\n", &so->so_q0, TAILQ_FIRST(&so->so_q0));
2199 	(*pr)("so_q: @%p first: %p\n", &so->so_q, TAILQ_FIRST(&so->so_q));
2200 	(*pr)("so_eq: next: %p\n", TAILQ_NEXT(so, so_qe));
2201 	(*pr)("so_q0len: %i\n", so->so_q0len);
2202 	(*pr)("so_qlen: %i\n", so->so_qlen);
2203 	(*pr)("so_qlimit: %i\n", so->so_qlimit);
2204 	(*pr)("so_timeo: %i\n", so->so_timeo);
2205 	(*pr)("so_obmark: %lu\n", so->so_oobmark);
2206 
2207 	(*pr)("so_sp: %p\n", so->so_sp);
2208 	if (so->so_sp != NULL) {
2209 		(*pr)("\tssp_socket: %p\n", so->so_sp->ssp_socket);
2210 		(*pr)("\tssp_soback: %p\n", so->so_sp->ssp_soback);
2211 		(*pr)("\tssp_len: %lld\n",
2212 		    (unsigned long long)so->so_sp->ssp_len);
2213 		(*pr)("\tssp_max: %lld\n",
2214 		    (unsigned long long)so->so_sp->ssp_max);
2215 		(*pr)("\tssp_idletv: %lld %ld\n", so->so_sp->ssp_idletv.tv_sec,
2216 		    so->so_sp->ssp_idletv.tv_usec);
2217 		(*pr)("\tssp_idleto: %spending (@%i)\n",
2218 		    timeout_pending(&so->so_sp->ssp_idleto) ? "" : "not ",
2219 		    so->so_sp->ssp_idleto.to_time);
2220 	}
2221 
2222 	(*pr)("so_rcv:\n");
2223 	sobuf_print(&so->so_rcv, pr);
2224 	(*pr)("so_snd:\n");
2225 	sobuf_print(&so->so_snd, pr);
2226 
2227 	(*pr)("so_upcall: %p so_upcallarg: %p\n",
2228 	    so->so_upcall, so->so_upcallarg);
2229 
2230 	(*pr)("so_euid: %d so_ruid: %d\n", so->so_euid, so->so_ruid);
2231 	(*pr)("so_egid: %d so_rgid: %d\n", so->so_egid, so->so_rgid);
2232 	(*pr)("so_cpid: %d\n", so->so_cpid);
2233 }
2234 #endif
2235