xref: /netbsd-src/sys/kern/uipc_socket.c (revision 8ac07aec990b9d2e483062509d0a9fa5b4f57cf2)
1 /*	$NetBSD: uipc_socket.c,v 1.160 2008/04/24 11:38:36 ad Exp $	*/
2 
3 /*-
4  * Copyright (c) 2002, 2007, 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe of Wasabi Systems, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the NetBSD
21  *	Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 
39 /*
40  * Copyright (c) 2004 The FreeBSD Foundation
41  * Copyright (c) 2004 Robert Watson
42  * Copyright (c) 1982, 1986, 1988, 1990, 1993
43  *	The Regents of the University of California.  All rights reserved.
44  *
45  * Redistribution and use in source and binary forms, with or without
46  * modification, are permitted provided that the following conditions
47  * are met:
48  * 1. Redistributions of source code must retain the above copyright
49  *    notice, this list of conditions and the following disclaimer.
50  * 2. Redistributions in binary form must reproduce the above copyright
51  *    notice, this list of conditions and the following disclaimer in the
52  *    documentation and/or other materials provided with the distribution.
53  * 3. Neither the name of the University nor the names of its contributors
54  *    may be used to endorse or promote products derived from this software
55  *    without specific prior written permission.
56  *
57  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
58  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
59  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
60  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
61  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
62  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
63  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
64  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
65  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
66  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
67  * SUCH DAMAGE.
68  *
69  *	@(#)uipc_socket.c	8.6 (Berkeley) 5/2/95
70  */
71 
72 #include <sys/cdefs.h>
73 __KERNEL_RCSID(0, "$NetBSD: uipc_socket.c,v 1.160 2008/04/24 11:38:36 ad Exp $");
74 
75 #include "opt_sock_counters.h"
76 #include "opt_sosend_loan.h"
77 #include "opt_mbuftrace.h"
78 #include "opt_somaxkva.h"
79 
80 #include <sys/param.h>
81 #include <sys/systm.h>
82 #include <sys/proc.h>
83 #include <sys/file.h>
84 #include <sys/filedesc.h>
85 #include <sys/malloc.h>
86 #include <sys/mbuf.h>
87 #include <sys/domain.h>
88 #include <sys/kernel.h>
89 #include <sys/protosw.h>
90 #include <sys/socket.h>
91 #include <sys/socketvar.h>
92 #include <sys/signalvar.h>
93 #include <sys/resourcevar.h>
94 #include <sys/event.h>
95 #include <sys/poll.h>
96 #include <sys/kauth.h>
97 #include <sys/mutex.h>
98 #include <sys/condvar.h>
99 
100 #include <uvm/uvm.h>
101 
102 MALLOC_DEFINE(M_SOOPTS, "soopts", "socket options");
103 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
104 
105 extern const struct fileops socketops;
106 
107 extern int	somaxconn;			/* patchable (XXX sysctl) */
108 int		somaxconn = SOMAXCONN;
109 kmutex_t	*softnet_lock;
110 
111 #ifdef SOSEND_COUNTERS
112 #include <sys/device.h>
113 
114 static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
115     NULL, "sosend", "loan big");
116 static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
117     NULL, "sosend", "copy big");
118 static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
119     NULL, "sosend", "copy small");
120 static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
121     NULL, "sosend", "kva limit");
122 
123 #define	SOSEND_COUNTER_INCR(ev)		(ev)->ev_count++
124 
125 EVCNT_ATTACH_STATIC(sosend_loan_big);
126 EVCNT_ATTACH_STATIC(sosend_copy_big);
127 EVCNT_ATTACH_STATIC(sosend_copy_small);
128 EVCNT_ATTACH_STATIC(sosend_kvalimit);
129 #else
130 
131 #define	SOSEND_COUNTER_INCR(ev)		/* nothing */
132 
133 #endif /* SOSEND_COUNTERS */
134 
135 static struct callback_entry sokva_reclaimerentry;
136 
137 #ifdef SOSEND_NO_LOAN
138 int sock_loan_thresh = -1;
139 #else
140 int sock_loan_thresh = 4096;
141 #endif
142 
143 static kmutex_t so_pendfree_lock;
144 static struct mbuf *so_pendfree;
145 
146 #ifndef SOMAXKVA
147 #define	SOMAXKVA (16 * 1024 * 1024)
148 #endif
149 int somaxkva = SOMAXKVA;
150 static int socurkva;
151 static kcondvar_t socurkva_cv;
152 
153 #define	SOCK_LOAN_CHUNK		65536
154 
155 static size_t sodopendfree(void);
156 static size_t sodopendfreel(void);
157 
158 static vsize_t
159 sokvareserve(struct socket *so, vsize_t len)
160 {
161 	int error;
162 
163 	mutex_enter(&so_pendfree_lock);
164 	while (socurkva + len > somaxkva) {
165 		size_t freed;
166 
167 		/*
168 		 * try to do pendfree.
169 		 */
170 
171 		freed = sodopendfreel();
172 
173 		/*
174 		 * if some kva was freed, try again.
175 		 */
176 
177 		if (freed)
178 			continue;
179 
180 		SOSEND_COUNTER_INCR(&sosend_kvalimit);
181 		error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock);
182 		if (error) {
183 			len = 0;
184 			break;
185 		}
186 	}
187 	socurkva += len;
188 	mutex_exit(&so_pendfree_lock);
189 	return len;
190 }
191 
192 static void
193 sokvaunreserve(vsize_t len)
194 {
195 
196 	mutex_enter(&so_pendfree_lock);
197 	socurkva -= len;
198 	cv_broadcast(&socurkva_cv);
199 	mutex_exit(&so_pendfree_lock);
200 }
201 
202 /*
203  * sokvaalloc: allocate kva for loan.
204  */
205 
206 vaddr_t
207 sokvaalloc(vsize_t len, struct socket *so)
208 {
209 	vaddr_t lva;
210 
211 	/*
212 	 * reserve kva.
213 	 */
214 
215 	if (sokvareserve(so, len) == 0)
216 		return 0;
217 
218 	/*
219 	 * allocate kva.
220 	 */
221 
222 	lva = uvm_km_alloc(kernel_map, len, 0, UVM_KMF_VAONLY | UVM_KMF_WAITVA);
223 	if (lva == 0) {
224 		sokvaunreserve(len);
225 		return (0);
226 	}
227 
228 	return lva;
229 }
230 
231 /*
232  * sokvafree: free kva for loan.
233  */
234 
235 void
236 sokvafree(vaddr_t sva, vsize_t len)
237 {
238 
239 	/*
240 	 * free kva.
241 	 */
242 
243 	uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY);
244 
245 	/*
246 	 * unreserve kva.
247 	 */
248 
249 	sokvaunreserve(len);
250 }
251 
252 static void
253 sodoloanfree(struct vm_page **pgs, void *buf, size_t size)
254 {
255 	vaddr_t sva, eva;
256 	vsize_t len;
257 	int npgs;
258 
259 	KASSERT(pgs != NULL);
260 
261 	eva = round_page((vaddr_t) buf + size);
262 	sva = trunc_page((vaddr_t) buf);
263 	len = eva - sva;
264 	npgs = len >> PAGE_SHIFT;
265 
266 	pmap_kremove(sva, len);
267 	pmap_update(pmap_kernel());
268 	uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE);
269 	sokvafree(sva, len);
270 }
271 
272 static size_t
273 sodopendfree(void)
274 {
275 	size_t rv;
276 
277 	if (__predict_true(so_pendfree == NULL))
278 		return 0;
279 
280 	mutex_enter(&so_pendfree_lock);
281 	rv = sodopendfreel();
282 	mutex_exit(&so_pendfree_lock);
283 
284 	return rv;
285 }
286 
287 /*
288  * sodopendfreel: free mbufs on "pendfree" list.
289  * unlock and relock so_pendfree_lock when freeing mbufs.
290  *
291  * => called with so_pendfree_lock held.
292  */
293 
294 static size_t
295 sodopendfreel(void)
296 {
297 	struct mbuf *m, *next;
298 	size_t rv = 0;
299 
300 	KASSERT(mutex_owned(&so_pendfree_lock));
301 
302 	while (so_pendfree != NULL) {
303 		m = so_pendfree;
304 		so_pendfree = NULL;
305 		mutex_exit(&so_pendfree_lock);
306 
307 		for (; m != NULL; m = next) {
308 			next = m->m_next;
309 			KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0);
310 			KASSERT(m->m_ext.ext_refcnt == 0);
311 
312 			rv += m->m_ext.ext_size;
313 			sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf,
314 			    m->m_ext.ext_size);
315 			pool_cache_put(mb_cache, m);
316 		}
317 
318 		mutex_enter(&so_pendfree_lock);
319 	}
320 
321 	return (rv);
322 }
323 
324 void
325 soloanfree(struct mbuf *m, void *buf, size_t size, void *arg)
326 {
327 
328 	KASSERT(m != NULL);
329 
330 	/*
331 	 * postpone freeing mbuf.
332 	 *
333 	 * we can't do it in interrupt context
334 	 * because we need to put kva back to kernel_map.
335 	 */
336 
337 	mutex_enter(&so_pendfree_lock);
338 	m->m_next = so_pendfree;
339 	so_pendfree = m;
340 	cv_broadcast(&socurkva_cv);
341 	mutex_exit(&so_pendfree_lock);
342 }
343 
344 static long
345 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space)
346 {
347 	struct iovec *iov = uio->uio_iov;
348 	vaddr_t sva, eva;
349 	vsize_t len;
350 	vaddr_t lva;
351 	int npgs, error;
352 	vaddr_t va;
353 	int i;
354 
355 	if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace))
356 		return (0);
357 
358 	if (iov->iov_len < (size_t) space)
359 		space = iov->iov_len;
360 	if (space > SOCK_LOAN_CHUNK)
361 		space = SOCK_LOAN_CHUNK;
362 
363 	eva = round_page((vaddr_t) iov->iov_base + space);
364 	sva = trunc_page((vaddr_t) iov->iov_base);
365 	len = eva - sva;
366 	npgs = len >> PAGE_SHIFT;
367 
368 	KASSERT(npgs <= M_EXT_MAXPAGES);
369 
370 	lva = sokvaalloc(len, so);
371 	if (lva == 0)
372 		return 0;
373 
374 	error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len,
375 	    m->m_ext.ext_pgs, UVM_LOAN_TOPAGE);
376 	if (error) {
377 		sokvafree(lva, len);
378 		return (0);
379 	}
380 
381 	for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE)
382 		pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
383 		    VM_PROT_READ);
384 	pmap_update(pmap_kernel());
385 
386 	lva += (vaddr_t) iov->iov_base & PAGE_MASK;
387 
388 	MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so);
389 	m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
390 
391 	uio->uio_resid -= space;
392 	/* uio_offset not updated, not set/used for write(2) */
393 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space;
394 	uio->uio_iov->iov_len -= space;
395 	if (uio->uio_iov->iov_len == 0) {
396 		uio->uio_iov++;
397 		uio->uio_iovcnt--;
398 	}
399 
400 	return (space);
401 }
402 
403 static int
404 sokva_reclaim_callback(struct callback_entry *ce, void *obj, void *arg)
405 {
406 
407 	KASSERT(ce == &sokva_reclaimerentry);
408 	KASSERT(obj == NULL);
409 
410 	sodopendfree();
411 	if (!vm_map_starved_p(kernel_map)) {
412 		return CALLBACK_CHAIN_ABORT;
413 	}
414 	return CALLBACK_CHAIN_CONTINUE;
415 }
416 
417 struct mbuf *
418 getsombuf(struct socket *so, int type)
419 {
420 	struct mbuf *m;
421 
422 	m = m_get(M_WAIT, type);
423 	MCLAIM(m, so->so_mowner);
424 	return m;
425 }
426 
427 struct mbuf *
428 m_intopt(struct socket *so, int val)
429 {
430 	struct mbuf *m;
431 
432 	m = getsombuf(so, MT_SOOPTS);
433 	m->m_len = sizeof(int);
434 	*mtod(m, int *) = val;
435 	return m;
436 }
437 
438 void
439 soinit(void)
440 {
441 
442 	mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM);
443 	softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
444 	cv_init(&socurkva_cv, "sokva");
445 
446 	/* Set the initial adjusted socket buffer size. */
447 	if (sb_max_set(sb_max))
448 		panic("bad initial sb_max value: %lu", sb_max);
449 
450 	callback_register(&vm_map_to_kernel(kernel_map)->vmk_reclaim_callback,
451 	    &sokva_reclaimerentry, NULL, sokva_reclaim_callback);
452 }
453 
454 /*
455  * Socket operation routines.
456  * These routines are called by the routines in
457  * sys_socket.c or from a system process, and
458  * implement the semantics of socket operations by
459  * switching out to the protocol specific routines.
460  */
461 /*ARGSUSED*/
462 int
463 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l,
464 	 struct socket *lockso)
465 {
466 	const struct protosw	*prp;
467 	struct socket	*so;
468 	uid_t		uid;
469 	int		error;
470 	kmutex_t	*lock;
471 
472 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET,
473 	    KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type),
474 	    KAUTH_ARG(proto));
475 	if (error != 0)
476 		return error;
477 
478 	if (proto)
479 		prp = pffindproto(dom, proto, type);
480 	else
481 		prp = pffindtype(dom, type);
482 	if (prp == NULL) {
483 		/* no support for domain */
484 		if (pffinddomain(dom) == 0)
485 			return EAFNOSUPPORT;
486 		/* no support for socket type */
487 		if (proto == 0 && type != 0)
488 			return EPROTOTYPE;
489 		return EPROTONOSUPPORT;
490 	}
491 	if (prp->pr_usrreq == NULL)
492 		return EPROTONOSUPPORT;
493 	if (prp->pr_type != type)
494 		return EPROTOTYPE;
495 
496 	so = soget(true);
497 	so->so_type = type;
498 	so->so_proto = prp;
499 	so->so_send = sosend;
500 	so->so_receive = soreceive;
501 #ifdef MBUFTRACE
502 	so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner;
503 	so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner;
504 	so->so_mowner = &prp->pr_domain->dom_mowner;
505 #endif
506 	uid = kauth_cred_geteuid(l->l_cred);
507 	so->so_uidinfo = uid_find(uid);
508 	if (lockso != NULL) {
509 		/* Caller wants us to share a lock. */
510 		lock = lockso->so_lock;
511 		so->so_lock = lock;
512 		mutex_obj_hold(lock);
513 		mutex_enter(lock);
514 	} else {
515 		/* Lock assigned and taken during PRU_ATTACH. */
516 	}
517 	error = (*prp->pr_usrreq)(so, PRU_ATTACH, NULL,
518 	    (struct mbuf *)(long)proto, NULL, l);
519 	KASSERT(solocked(so));
520 	if (error != 0) {
521 		so->so_state |= SS_NOFDREF;
522 		sofree(so);
523 		return error;
524 	}
525 	sounlock(so);
526 	*aso = so;
527 	return 0;
528 }
529 
530 /* On success, write file descriptor to fdout and return zero.  On
531  * failure, return non-zero; *fdout will be undefined.
532  */
533 int
534 fsocreate(int domain, struct socket **sop, int type, int protocol,
535     struct lwp *l, int *fdout)
536 {
537 	struct socket	*so;
538 	struct file	*fp;
539 	int		fd, error;
540 
541 	if ((error = fd_allocfile(&fp, &fd)) != 0)
542 		return (error);
543 	fp->f_flag = FREAD|FWRITE;
544 	fp->f_type = DTYPE_SOCKET;
545 	fp->f_ops = &socketops;
546 	error = socreate(domain, &so, type, protocol, l, NULL);
547 	if (error != 0) {
548 		fd_abort(curproc, fp, fd);
549 	} else {
550 		if (sop != NULL)
551 			*sop = so;
552 		fp->f_data = so;
553 		fd_affix(curproc, fp, fd);
554 		*fdout = fd;
555 	}
556 	return error;
557 }
558 
559 int
560 sobind(struct socket *so, struct mbuf *nam, struct lwp *l)
561 {
562 	int	error;
563 
564 	solock(so);
565 	error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, NULL, nam, NULL, l);
566 	sounlock(so);
567 	return error;
568 }
569 
570 int
571 solisten(struct socket *so, int backlog, struct lwp *l)
572 {
573 	int	error;
574 
575 	solock(so);
576 	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
577 	    SS_ISDISCONNECTING)) != 0)
578 		return (EOPNOTSUPP);
579 	error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, NULL,
580 	    NULL, NULL, l);
581 	if (error != 0) {
582 		sounlock(so);
583 		return error;
584 	}
585 	if (TAILQ_EMPTY(&so->so_q))
586 		so->so_options |= SO_ACCEPTCONN;
587 	if (backlog < 0)
588 		backlog = 0;
589 	so->so_qlimit = min(backlog, somaxconn);
590 	sounlock(so);
591 	return 0;
592 }
593 
594 void
595 sofree(struct socket *so)
596 {
597 
598 	KASSERT(solocked(so));
599 
600 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
601 		sounlock(so);
602 		return;
603 	}
604 	if (so->so_head) {
605 		/*
606 		 * We must not decommission a socket that's on the accept(2)
607 		 * queue.  If we do, then accept(2) may hang after select(2)
608 		 * indicated that the listening socket was ready.
609 		 */
610 		if (!soqremque(so, 0)) {
611 			sounlock(so);
612 			return;
613 		}
614 	}
615 	if (so->so_rcv.sb_hiwat)
616 		(void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0,
617 		    RLIM_INFINITY);
618 	if (so->so_snd.sb_hiwat)
619 		(void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0,
620 		    RLIM_INFINITY);
621 	sbrelease(&so->so_snd, so);
622 	KASSERT(!cv_has_waiters(&so->so_cv));
623 	KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv));
624 	KASSERT(!cv_has_waiters(&so->so_snd.sb_cv));
625 	sorflush(so);
626 	sounlock(so);
627 	soput(so);
628 }
629 
630 /*
631  * Close a socket on last file table reference removal.
632  * Initiate disconnect if connected.
633  * Free socket when disconnect complete.
634  */
635 int
636 soclose(struct socket *so)
637 {
638 	struct socket	*so2;
639 	int		error;
640 	int		error2;
641 
642 	error = 0;
643 	solock(so);
644 	if (so->so_options & SO_ACCEPTCONN) {
645 		do {
646 			if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) {
647 				KASSERT(solocked2(so, so2));
648 				(void) soqremque(so2, 0);
649 				/* soabort drops the lock. */
650 				(void) soabort(so2);
651 				solock(so);
652 				continue;
653 			}
654 			if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) {
655 				KASSERT(solocked2(so, so2));
656 				(void) soqremque(so2, 1);
657 				/* soabort drops the lock. */
658 				(void) soabort(so2);
659 				solock(so);
660 				continue;
661 			}
662 		} while (0);
663 	}
664 	if (so->so_pcb == 0)
665 		goto discard;
666 	if (so->so_state & SS_ISCONNECTED) {
667 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
668 			error = sodisconnect(so);
669 			if (error)
670 				goto drop;
671 		}
672 		if (so->so_options & SO_LINGER) {
673 			if ((so->so_state & SS_ISDISCONNECTING) && so->so_nbio)
674 				goto drop;
675 			while (so->so_state & SS_ISCONNECTED) {
676 				error = sowait(so, so->so_linger * hz);
677 				if (error)
678 					break;
679 			}
680 		}
681 	}
682  drop:
683 	if (so->so_pcb) {
684 		error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH,
685 		    NULL, NULL, NULL, NULL);
686 		if (error == 0)
687 			error = error2;
688 	}
689  discard:
690 	if (so->so_state & SS_NOFDREF)
691 		panic("soclose: NOFDREF");
692 	so->so_state |= SS_NOFDREF;
693 	sofree(so);
694 	return (error);
695 }
696 
697 /*
698  * Must be called with the socket locked..  Will return with it unlocked.
699  */
700 int
701 soabort(struct socket *so)
702 {
703 	int error;
704 
705 	KASSERT(solocked(so));
706 	KASSERT(so->so_head == NULL);
707 
708 	error = (*so->so_proto->pr_usrreq)(so, PRU_ABORT, NULL,
709 	    NULL, NULL, NULL);
710 	if (error) {
711 		sofree(so);
712 	} else {
713 		sounlock(so);
714 	}
715 	return error;
716 }
717 
718 int
719 soaccept(struct socket *so, struct mbuf *nam)
720 {
721 	int	error;
722 
723 	KASSERT(solocked(so));
724 
725 	error = 0;
726 	if ((so->so_state & SS_NOFDREF) == 0)
727 		panic("soaccept: !NOFDREF");
728 	so->so_state &= ~SS_NOFDREF;
729 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
730 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
731 		error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT,
732 		    NULL, nam, NULL, NULL);
733 	else
734 		error = ECONNABORTED;
735 
736 	return (error);
737 }
738 
739 int
740 soconnect(struct socket *so, struct mbuf *nam, struct lwp *l)
741 {
742 	int		error;
743 
744 	KASSERT(solocked(so));
745 
746 	if (so->so_options & SO_ACCEPTCONN)
747 		return (EOPNOTSUPP);
748 	/*
749 	 * If protocol is connection-based, can only connect once.
750 	 * Otherwise, if connected, try to disconnect first.
751 	 * This allows user to disconnect by connecting to, e.g.,
752 	 * a null address.
753 	 */
754 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
755 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
756 	    (error = sodisconnect(so))))
757 		error = EISCONN;
758 	else
759 		error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT,
760 		    NULL, nam, NULL, l);
761 	return (error);
762 }
763 
764 int
765 soconnect2(struct socket *so1, struct socket *so2)
766 {
767 	int	error;
768 
769 	KASSERT(solocked2(so1, so2));
770 
771 	error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2,
772 	    NULL, (struct mbuf *)so2, NULL, NULL);
773 	return (error);
774 }
775 
776 int
777 sodisconnect(struct socket *so)
778 {
779 	int	error;
780 
781 	KASSERT(solocked(so));
782 
783 	if ((so->so_state & SS_ISCONNECTED) == 0) {
784 		error = ENOTCONN;
785 	} else if (so->so_state & SS_ISDISCONNECTING) {
786 		error = EALREADY;
787 	} else {
788 		error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT,
789 		    NULL, NULL, NULL, NULL);
790 	}
791 	sodopendfree();
792 	return (error);
793 }
794 
795 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
796 /*
797  * Send on a socket.
798  * If send must go all at once and message is larger than
799  * send buffering, then hard error.
800  * Lock against other senders.
801  * If must go all at once and not enough room now, then
802  * inform user that this would block and do nothing.
803  * Otherwise, if nonblocking, send as much as possible.
804  * The data to be sent is described by "uio" if nonzero,
805  * otherwise by the mbuf chain "top" (which must be null
806  * if uio is not).  Data provided in mbuf chain must be small
807  * enough to send all at once.
808  *
809  * Returns nonzero on error, timeout or signal; callers
810  * must check for short counts if EINTR/ERESTART are returned.
811  * Data and control buffers are freed on return.
812  */
813 int
814 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
815 	struct mbuf *control, int flags, struct lwp *l)
816 {
817 	struct mbuf	**mp, *m;
818 	struct proc	*p;
819 	long		space, len, resid, clen, mlen;
820 	int		error, s, dontroute, atomic;
821 
822 	p = l->l_proc;
823 	sodopendfree();
824 	clen = 0;
825 
826 	/*
827 	 * solock() provides atomicity of access.  splsoftnet() prevents
828 	 * protocol processing soft interrupts from interrupting us and
829 	 * blocking (expensive).
830 	 */
831 	s = splsoftnet();
832 	solock(so);
833 	atomic = sosendallatonce(so) || top;
834 	if (uio)
835 		resid = uio->uio_resid;
836 	else
837 		resid = top->m_pkthdr.len;
838 	/*
839 	 * In theory resid should be unsigned.
840 	 * However, space must be signed, as it might be less than 0
841 	 * if we over-committed, and we must use a signed comparison
842 	 * of space and resid.  On the other hand, a negative resid
843 	 * causes us to loop sending 0-length segments to the protocol.
844 	 */
845 	if (resid < 0) {
846 		error = EINVAL;
847 		goto out;
848 	}
849 	dontroute =
850 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
851 	    (so->so_proto->pr_flags & PR_ATOMIC);
852 	if (l)
853 		l->l_ru.ru_msgsnd++;
854 	if (control)
855 		clen = control->m_len;
856  restart:
857 	if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
858 		goto out;
859 	do {
860 		if (so->so_state & SS_CANTSENDMORE) {
861 			error = EPIPE;
862 			goto release;
863 		}
864 		if (so->so_error) {
865 			error = so->so_error;
866 			so->so_error = 0;
867 			goto release;
868 		}
869 		if ((so->so_state & SS_ISCONNECTED) == 0) {
870 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
871 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
872 				    !(resid == 0 && clen != 0)) {
873 					error = ENOTCONN;
874 					goto release;
875 				}
876 			} else if (addr == 0) {
877 				error = EDESTADDRREQ;
878 				goto release;
879 			}
880 		}
881 		space = sbspace(&so->so_snd);
882 		if (flags & MSG_OOB)
883 			space += 1024;
884 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
885 		    clen > so->so_snd.sb_hiwat) {
886 			error = EMSGSIZE;
887 			goto release;
888 		}
889 		if (space < resid + clen &&
890 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
891 			if (so->so_nbio) {
892 				error = EWOULDBLOCK;
893 				goto release;
894 			}
895 			sbunlock(&so->so_snd);
896 			error = sbwait(&so->so_snd);
897 			if (error)
898 				goto out;
899 			goto restart;
900 		}
901 		mp = &top;
902 		space -= clen;
903 		do {
904 			if (uio == NULL) {
905 				/*
906 				 * Data is prepackaged in "top".
907 				 */
908 				resid = 0;
909 				if (flags & MSG_EOR)
910 					top->m_flags |= M_EOR;
911 			} else do {
912 				sounlock(so);
913 				splx(s);
914 				if (top == NULL) {
915 					m = m_gethdr(M_WAIT, MT_DATA);
916 					mlen = MHLEN;
917 					m->m_pkthdr.len = 0;
918 					m->m_pkthdr.rcvif = NULL;
919 				} else {
920 					m = m_get(M_WAIT, MT_DATA);
921 					mlen = MLEN;
922 				}
923 				MCLAIM(m, so->so_snd.sb_mowner);
924 				if (sock_loan_thresh >= 0 &&
925 				    uio->uio_iov->iov_len >= sock_loan_thresh &&
926 				    space >= sock_loan_thresh &&
927 				    (len = sosend_loan(so, uio, m,
928 						       space)) != 0) {
929 					SOSEND_COUNTER_INCR(&sosend_loan_big);
930 					space -= len;
931 					goto have_data;
932 				}
933 				if (resid >= MINCLSIZE && space >= MCLBYTES) {
934 					SOSEND_COUNTER_INCR(&sosend_copy_big);
935 					m_clget(m, M_WAIT);
936 					if ((m->m_flags & M_EXT) == 0)
937 						goto nopages;
938 					mlen = MCLBYTES;
939 					if (atomic && top == 0) {
940 						len = lmin(MCLBYTES - max_hdr,
941 						    resid);
942 						m->m_data += max_hdr;
943 					} else
944 						len = lmin(MCLBYTES, resid);
945 					space -= len;
946 				} else {
947  nopages:
948 					SOSEND_COUNTER_INCR(&sosend_copy_small);
949 					len = lmin(lmin(mlen, resid), space);
950 					space -= len;
951 					/*
952 					 * For datagram protocols, leave room
953 					 * for protocol headers in first mbuf.
954 					 */
955 					if (atomic && top == 0 && len < mlen)
956 						MH_ALIGN(m, len);
957 				}
958 				error = uiomove(mtod(m, void *), (int)len, uio);
959  have_data:
960 				resid = uio->uio_resid;
961 				m->m_len = len;
962 				*mp = m;
963 				top->m_pkthdr.len += len;
964 				s = splsoftnet();
965 				solock(so);
966 				if (error != 0)
967 					goto release;
968 				mp = &m->m_next;
969 				if (resid <= 0) {
970 					if (flags & MSG_EOR)
971 						top->m_flags |= M_EOR;
972 					break;
973 				}
974 			} while (space > 0 && atomic);
975 
976 			if (so->so_state & SS_CANTSENDMORE) {
977 				error = EPIPE;
978 				goto release;
979 			}
980 			if (dontroute)
981 				so->so_options |= SO_DONTROUTE;
982 			if (resid > 0)
983 				so->so_state |= SS_MORETOCOME;
984 			error = (*so->so_proto->pr_usrreq)(so,
985 			    (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND,
986 			    top, addr, control, curlwp);
987 			if (dontroute)
988 				so->so_options &= ~SO_DONTROUTE;
989 			if (resid > 0)
990 				so->so_state &= ~SS_MORETOCOME;
991 			clen = 0;
992 			control = NULL;
993 			top = NULL;
994 			mp = &top;
995 			if (error != 0)
996 				goto release;
997 		} while (resid && space > 0);
998 	} while (resid);
999 
1000  release:
1001 	sbunlock(&so->so_snd);
1002  out:
1003 	sounlock(so);
1004 	splx(s);
1005 	if (top)
1006 		m_freem(top);
1007 	if (control)
1008 		m_freem(control);
1009 	return (error);
1010 }
1011 
1012 /*
1013  * Following replacement or removal of the first mbuf on the first
1014  * mbuf chain of a socket buffer, push necessary state changes back
1015  * into the socket buffer so that other consumers see the values
1016  * consistently.  'nextrecord' is the callers locally stored value of
1017  * the original value of sb->sb_mb->m_nextpkt which must be restored
1018  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
1019  */
1020 static void
1021 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
1022 {
1023 
1024 	KASSERT(solocked(sb->sb_so));
1025 
1026 	/*
1027 	 * First, update for the new value of nextrecord.  If necessary,
1028 	 * make it the first record.
1029 	 */
1030 	if (sb->sb_mb != NULL)
1031 		sb->sb_mb->m_nextpkt = nextrecord;
1032 	else
1033 		sb->sb_mb = nextrecord;
1034 
1035         /*
1036          * Now update any dependent socket buffer fields to reflect
1037          * the new state.  This is an inline of SB_EMPTY_FIXUP, with
1038          * the addition of a second clause that takes care of the
1039          * case where sb_mb has been updated, but remains the last
1040          * record.
1041          */
1042         if (sb->sb_mb == NULL) {
1043                 sb->sb_mbtail = NULL;
1044                 sb->sb_lastrecord = NULL;
1045         } else if (sb->sb_mb->m_nextpkt == NULL)
1046                 sb->sb_lastrecord = sb->sb_mb;
1047 }
1048 
1049 /*
1050  * Implement receive operations on a socket.
1051  * We depend on the way that records are added to the sockbuf
1052  * by sbappend*.  In particular, each record (mbufs linked through m_next)
1053  * must begin with an address if the protocol so specifies,
1054  * followed by an optional mbuf or mbufs containing ancillary data,
1055  * and then zero or more mbufs of data.
1056  * In order to avoid blocking network interrupts for the entire time here,
1057  * we splx() while doing the actual copy to user space.
1058  * Although the sockbuf is locked, new data may still be appended,
1059  * and thus we must maintain consistency of the sockbuf during that time.
1060  *
1061  * The caller may receive the data as a single mbuf chain by supplying
1062  * an mbuf **mp0 for use in returning the chain.  The uio is then used
1063  * only for the count in uio_resid.
1064  */
1065 int
1066 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
1067 	struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1068 {
1069 	struct lwp *l = curlwp;
1070 	struct mbuf	*m, **mp, *mt;
1071 	int atomic, flags, len, error, s, offset, moff, type, orig_resid;
1072 	const struct protosw	*pr;
1073 	struct mbuf	*nextrecord;
1074 	int		mbuf_removed = 0;
1075 	const struct domain *dom;
1076 
1077 	pr = so->so_proto;
1078 	atomic = pr->pr_flags & PR_ATOMIC;
1079 	dom = pr->pr_domain;
1080 	mp = mp0;
1081 	type = 0;
1082 	orig_resid = uio->uio_resid;
1083 
1084 	if (paddr != NULL)
1085 		*paddr = NULL;
1086 	if (controlp != NULL)
1087 		*controlp = NULL;
1088 	if (flagsp != NULL)
1089 		flags = *flagsp &~ MSG_EOR;
1090 	else
1091 		flags = 0;
1092 
1093 	if ((flags & MSG_DONTWAIT) == 0)
1094 		sodopendfree();
1095 
1096 	if (flags & MSG_OOB) {
1097 		m = m_get(M_WAIT, MT_DATA);
1098 		solock(so);
1099 		error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m,
1100 		    (struct mbuf *)(long)(flags & MSG_PEEK), NULL, l);
1101 		sounlock(so);
1102 		if (error)
1103 			goto bad;
1104 		do {
1105 			error = uiomove(mtod(m, void *),
1106 			    (int) min(uio->uio_resid, m->m_len), uio);
1107 			m = m_free(m);
1108 		} while (uio->uio_resid > 0 && error == 0 && m);
1109  bad:
1110 		if (m != NULL)
1111 			m_freem(m);
1112 		return error;
1113 	}
1114 	if (mp != NULL)
1115 		*mp = NULL;
1116 
1117 	/*
1118 	 * solock() provides atomicity of access.  splsoftnet() prevents
1119 	 * protocol processing soft interrupts from interrupting us and
1120 	 * blocking (expensive).
1121 	 */
1122 	s = splsoftnet();
1123 	solock(so);
1124 	if (so->so_state & SS_ISCONFIRMING && uio->uio_resid)
1125 		(*pr->pr_usrreq)(so, PRU_RCVD, NULL, NULL, NULL, l);
1126 
1127  restart:
1128 	if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) {
1129 		sounlock(so);
1130 		splx(s);
1131 		return error;
1132 	}
1133 
1134 	m = so->so_rcv.sb_mb;
1135 	/*
1136 	 * If we have less data than requested, block awaiting more
1137 	 * (subject to any timeout) if:
1138 	 *   1. the current count is less than the low water mark,
1139 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1140 	 *	receive operation at once if we block (resid <= hiwat), or
1141 	 *   3. MSG_DONTWAIT is not set.
1142 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1143 	 * we have to do the receive in sections, and thus risk returning
1144 	 * a short count if a timeout or signal occurs after we start.
1145 	 */
1146 	if (m == NULL ||
1147 	    ((flags & MSG_DONTWAIT) == 0 &&
1148 	     so->so_rcv.sb_cc < uio->uio_resid &&
1149 	     (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1150 	      ((flags & MSG_WAITALL) &&
1151 	       uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1152 	     m->m_nextpkt == NULL && !atomic)) {
1153 #ifdef DIAGNOSTIC
1154 		if (m == NULL && so->so_rcv.sb_cc)
1155 			panic("receive 1");
1156 #endif
1157 		if (so->so_error) {
1158 			if (m != NULL)
1159 				goto dontblock;
1160 			error = so->so_error;
1161 			if ((flags & MSG_PEEK) == 0)
1162 				so->so_error = 0;
1163 			goto release;
1164 		}
1165 		if (so->so_state & SS_CANTRCVMORE) {
1166 			if (m != NULL)
1167 				goto dontblock;
1168 			else
1169 				goto release;
1170 		}
1171 		for (; m != NULL; m = m->m_next)
1172 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1173 				m = so->so_rcv.sb_mb;
1174 				goto dontblock;
1175 			}
1176 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1177 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1178 			error = ENOTCONN;
1179 			goto release;
1180 		}
1181 		if (uio->uio_resid == 0)
1182 			goto release;
1183 		if (so->so_nbio || (flags & MSG_DONTWAIT)) {
1184 			error = EWOULDBLOCK;
1185 			goto release;
1186 		}
1187 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
1188 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
1189 		sbunlock(&so->so_rcv);
1190 		error = sbwait(&so->so_rcv);
1191 		if (error != 0) {
1192 			sounlock(so);
1193 			splx(s);
1194 			return error;
1195 		}
1196 		goto restart;
1197 	}
1198  dontblock:
1199 	/*
1200 	 * On entry here, m points to the first record of the socket buffer.
1201 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1202 	 * pointer to the next record in the socket buffer.  We must keep the
1203 	 * various socket buffer pointers and local stack versions of the
1204 	 * pointers in sync, pushing out modifications before dropping the
1205 	 * socket lock, and re-reading them when picking it up.
1206 	 *
1207 	 * Otherwise, we will race with the network stack appending new data
1208 	 * or records onto the socket buffer by using inconsistent/stale
1209 	 * versions of the field, possibly resulting in socket buffer
1210 	 * corruption.
1211 	 *
1212 	 * By holding the high-level sblock(), we prevent simultaneous
1213 	 * readers from pulling off the front of the socket buffer.
1214 	 */
1215 	if (l != NULL)
1216 		l->l_ru.ru_msgrcv++;
1217 	KASSERT(m == so->so_rcv.sb_mb);
1218 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
1219 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
1220 	nextrecord = m->m_nextpkt;
1221 	if (pr->pr_flags & PR_ADDR) {
1222 #ifdef DIAGNOSTIC
1223 		if (m->m_type != MT_SONAME)
1224 			panic("receive 1a");
1225 #endif
1226 		orig_resid = 0;
1227 		if (flags & MSG_PEEK) {
1228 			if (paddr)
1229 				*paddr = m_copy(m, 0, m->m_len);
1230 			m = m->m_next;
1231 		} else {
1232 			sbfree(&so->so_rcv, m);
1233 			mbuf_removed = 1;
1234 			if (paddr != NULL) {
1235 				*paddr = m;
1236 				so->so_rcv.sb_mb = m->m_next;
1237 				m->m_next = NULL;
1238 				m = so->so_rcv.sb_mb;
1239 			} else {
1240 				MFREE(m, so->so_rcv.sb_mb);
1241 				m = so->so_rcv.sb_mb;
1242 			}
1243 			sbsync(&so->so_rcv, nextrecord);
1244 		}
1245 	}
1246 
1247 	/*
1248 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1249 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1250 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1251 	 * perform externalization (or freeing if controlp == NULL).
1252 	 */
1253 	if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) {
1254 		struct mbuf *cm = NULL, *cmn;
1255 		struct mbuf **cme = &cm;
1256 
1257 		do {
1258 			if (flags & MSG_PEEK) {
1259 				if (controlp != NULL) {
1260 					*controlp = m_copy(m, 0, m->m_len);
1261 					controlp = &(*controlp)->m_next;
1262 				}
1263 				m = m->m_next;
1264 			} else {
1265 				sbfree(&so->so_rcv, m);
1266 				so->so_rcv.sb_mb = m->m_next;
1267 				m->m_next = NULL;
1268 				*cme = m;
1269 				cme = &(*cme)->m_next;
1270 				m = so->so_rcv.sb_mb;
1271 			}
1272 		} while (m != NULL && m->m_type == MT_CONTROL);
1273 		if ((flags & MSG_PEEK) == 0)
1274 			sbsync(&so->so_rcv, nextrecord);
1275 		for (; cm != NULL; cm = cmn) {
1276 			cmn = cm->m_next;
1277 			cm->m_next = NULL;
1278 			type = mtod(cm, struct cmsghdr *)->cmsg_type;
1279 			if (controlp != NULL) {
1280 				if (dom->dom_externalize != NULL &&
1281 				    type == SCM_RIGHTS) {
1282 					sounlock(so);
1283 					splx(s);
1284 					error = (*dom->dom_externalize)(cm, l);
1285 					s = splsoftnet();
1286 					solock(so);
1287 				}
1288 				*controlp = cm;
1289 				while (*controlp != NULL)
1290 					controlp = &(*controlp)->m_next;
1291 			} else {
1292 				/*
1293 				 * Dispose of any SCM_RIGHTS message that went
1294 				 * through the read path rather than recv.
1295 				 */
1296 				if (dom->dom_dispose != NULL &&
1297 				    type == SCM_RIGHTS) {
1298 				    	sounlock(so);
1299 					(*dom->dom_dispose)(cm);
1300 					solock(so);
1301 				}
1302 				m_freem(cm);
1303 			}
1304 		}
1305 		if (m != NULL)
1306 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1307 		else
1308 			nextrecord = so->so_rcv.sb_mb;
1309 		orig_resid = 0;
1310 	}
1311 
1312 	/* If m is non-NULL, we have some data to read. */
1313 	if (__predict_true(m != NULL)) {
1314 		type = m->m_type;
1315 		if (type == MT_OOBDATA)
1316 			flags |= MSG_OOB;
1317 	}
1318 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1319 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1320 
1321 	moff = 0;
1322 	offset = 0;
1323 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1324 		if (m->m_type == MT_OOBDATA) {
1325 			if (type != MT_OOBDATA)
1326 				break;
1327 		} else if (type == MT_OOBDATA)
1328 			break;
1329 #ifdef DIAGNOSTIC
1330 		else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
1331 			panic("receive 3");
1332 #endif
1333 		so->so_state &= ~SS_RCVATMARK;
1334 		len = uio->uio_resid;
1335 		if (so->so_oobmark && len > so->so_oobmark - offset)
1336 			len = so->so_oobmark - offset;
1337 		if (len > m->m_len - moff)
1338 			len = m->m_len - moff;
1339 		/*
1340 		 * If mp is set, just pass back the mbufs.
1341 		 * Otherwise copy them out via the uio, then free.
1342 		 * Sockbuf must be consistent here (points to current mbuf,
1343 		 * it points to next record) when we drop priority;
1344 		 * we must note any additions to the sockbuf when we
1345 		 * block interrupts again.
1346 		 */
1347 		if (mp == NULL) {
1348 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1349 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1350 			sounlock(so);
1351 			splx(s);
1352 			error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1353 			s = splsoftnet();
1354 			solock(so);
1355 			if (error != 0) {
1356 				/*
1357 				 * If any part of the record has been removed
1358 				 * (such as the MT_SONAME mbuf, which will
1359 				 * happen when PR_ADDR, and thus also
1360 				 * PR_ATOMIC, is set), then drop the entire
1361 				 * record to maintain the atomicity of the
1362 				 * receive operation.
1363 				 *
1364 				 * This avoids a later panic("receive 1a")
1365 				 * when compiled with DIAGNOSTIC.
1366 				 */
1367 				if (m && mbuf_removed && atomic)
1368 					(void) sbdroprecord(&so->so_rcv);
1369 
1370 				goto release;
1371 			}
1372 		} else
1373 			uio->uio_resid -= len;
1374 		if (len == m->m_len - moff) {
1375 			if (m->m_flags & M_EOR)
1376 				flags |= MSG_EOR;
1377 			if (flags & MSG_PEEK) {
1378 				m = m->m_next;
1379 				moff = 0;
1380 			} else {
1381 				nextrecord = m->m_nextpkt;
1382 				sbfree(&so->so_rcv, m);
1383 				if (mp) {
1384 					*mp = m;
1385 					mp = &m->m_next;
1386 					so->so_rcv.sb_mb = m = m->m_next;
1387 					*mp = NULL;
1388 				} else {
1389 					MFREE(m, so->so_rcv.sb_mb);
1390 					m = so->so_rcv.sb_mb;
1391 				}
1392 				/*
1393 				 * If m != NULL, we also know that
1394 				 * so->so_rcv.sb_mb != NULL.
1395 				 */
1396 				KASSERT(so->so_rcv.sb_mb == m);
1397 				if (m) {
1398 					m->m_nextpkt = nextrecord;
1399 					if (nextrecord == NULL)
1400 						so->so_rcv.sb_lastrecord = m;
1401 				} else {
1402 					so->so_rcv.sb_mb = nextrecord;
1403 					SB_EMPTY_FIXUP(&so->so_rcv);
1404 				}
1405 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1406 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1407 			}
1408 		} else if (flags & MSG_PEEK)
1409 			moff += len;
1410 		else {
1411 			if (mp != NULL) {
1412 				mt = m_copym(m, 0, len, M_NOWAIT);
1413 				if (__predict_false(mt == NULL)) {
1414 					sounlock(so);
1415 					mt = m_copym(m, 0, len, M_WAIT);
1416 					solock(so);
1417 				}
1418 				*mp = mt;
1419 			}
1420 			m->m_data += len;
1421 			m->m_len -= len;
1422 			so->so_rcv.sb_cc -= len;
1423 		}
1424 		if (so->so_oobmark) {
1425 			if ((flags & MSG_PEEK) == 0) {
1426 				so->so_oobmark -= len;
1427 				if (so->so_oobmark == 0) {
1428 					so->so_state |= SS_RCVATMARK;
1429 					break;
1430 				}
1431 			} else {
1432 				offset += len;
1433 				if (offset == so->so_oobmark)
1434 					break;
1435 			}
1436 		}
1437 		if (flags & MSG_EOR)
1438 			break;
1439 		/*
1440 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1441 		 * we must not quit until "uio->uio_resid == 0" or an error
1442 		 * termination.  If a signal/timeout occurs, return
1443 		 * with a short count but without error.
1444 		 * Keep sockbuf locked against other readers.
1445 		 */
1446 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1447 		    !sosendallatonce(so) && !nextrecord) {
1448 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1449 				break;
1450 			/*
1451 			 * If we are peeking and the socket receive buffer is
1452 			 * full, stop since we can't get more data to peek at.
1453 			 */
1454 			if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
1455 				break;
1456 			/*
1457 			 * If we've drained the socket buffer, tell the
1458 			 * protocol in case it needs to do something to
1459 			 * get it filled again.
1460 			 */
1461 			if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1462 				(*pr->pr_usrreq)(so, PRU_RCVD,
1463 				    NULL, (struct mbuf *)(long)flags, NULL, l);
1464 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1465 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1466 			error = sbwait(&so->so_rcv);
1467 			if (error != 0) {
1468 				sbunlock(&so->so_rcv);
1469 				sounlock(so);
1470 				splx(s);
1471 				return 0;
1472 			}
1473 			if ((m = so->so_rcv.sb_mb) != NULL)
1474 				nextrecord = m->m_nextpkt;
1475 		}
1476 	}
1477 
1478 	if (m && atomic) {
1479 		flags |= MSG_TRUNC;
1480 		if ((flags & MSG_PEEK) == 0)
1481 			(void) sbdroprecord(&so->so_rcv);
1482 	}
1483 	if ((flags & MSG_PEEK) == 0) {
1484 		if (m == NULL) {
1485 			/*
1486 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1487 			 * part makes sure sb_lastrecord is up-to-date if
1488 			 * there is still data in the socket buffer.
1489 			 */
1490 			so->so_rcv.sb_mb = nextrecord;
1491 			if (so->so_rcv.sb_mb == NULL) {
1492 				so->so_rcv.sb_mbtail = NULL;
1493 				so->so_rcv.sb_lastrecord = NULL;
1494 			} else if (nextrecord->m_nextpkt == NULL)
1495 				so->so_rcv.sb_lastrecord = nextrecord;
1496 		}
1497 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1498 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1499 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1500 			(*pr->pr_usrreq)(so, PRU_RCVD, NULL,
1501 			    (struct mbuf *)(long)flags, NULL, l);
1502 	}
1503 	if (orig_resid == uio->uio_resid && orig_resid &&
1504 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1505 		sbunlock(&so->so_rcv);
1506 		goto restart;
1507 	}
1508 
1509 	if (flagsp != NULL)
1510 		*flagsp |= flags;
1511  release:
1512 	sbunlock(&so->so_rcv);
1513 	sounlock(so);
1514 	splx(s);
1515 	return error;
1516 }
1517 
1518 int
1519 soshutdown(struct socket *so, int how)
1520 {
1521 	const struct protosw	*pr;
1522 	int	error;
1523 
1524 	KASSERT(solocked(so));
1525 
1526 	pr = so->so_proto;
1527 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1528 		return (EINVAL);
1529 
1530 	if (how == SHUT_RD || how == SHUT_RDWR) {
1531 		sorflush(so);
1532 		error = 0;
1533 	}
1534 	if (how == SHUT_WR || how == SHUT_RDWR)
1535 		error = (*pr->pr_usrreq)(so, PRU_SHUTDOWN, NULL,
1536 		    NULL, NULL, NULL);
1537 
1538 	return error;
1539 }
1540 
1541 void
1542 sorflush(struct socket *so)
1543 {
1544 	struct sockbuf	*sb, asb;
1545 	const struct protosw	*pr;
1546 
1547 	KASSERT(solocked(so));
1548 
1549 	sb = &so->so_rcv;
1550 	pr = so->so_proto;
1551 	socantrcvmore(so);
1552 	sb->sb_flags |= SB_NOINTR;
1553 	(void )sblock(sb, M_WAITOK);
1554 	sbunlock(sb);
1555 	asb = *sb;
1556 	/*
1557 	 * Clear most of the sockbuf structure, but leave some of the
1558 	 * fields valid.
1559 	 */
1560 	memset(&sb->sb_startzero, 0,
1561 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1562 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) {
1563 		sounlock(so);
1564 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1565 		solock(so);
1566 	}
1567 	sbrelease(&asb, so);
1568 }
1569 
1570 static int
1571 sosetopt1(struct socket *so, int level, int optname, struct mbuf *m)
1572 {
1573 	int optval, val;
1574 	struct linger	*l;
1575 	struct sockbuf	*sb;
1576 	struct timeval *tv;
1577 
1578 	switch (optname) {
1579 
1580 	case SO_LINGER:
1581 		if (m == NULL || m->m_len != sizeof(struct linger))
1582 			return EINVAL;
1583 		l = mtod(m, struct linger *);
1584 		if (l->l_linger < 0 || l->l_linger > USHRT_MAX ||
1585 		    l->l_linger > (INT_MAX / hz))
1586 			return EDOM;
1587 		so->so_linger = l->l_linger;
1588 		if (l->l_onoff)
1589 			so->so_options |= SO_LINGER;
1590 		else
1591 			so->so_options &= ~SO_LINGER;
1592 		break;
1593 
1594 	case SO_DEBUG:
1595 	case SO_KEEPALIVE:
1596 	case SO_DONTROUTE:
1597 	case SO_USELOOPBACK:
1598 	case SO_BROADCAST:
1599 	case SO_REUSEADDR:
1600 	case SO_REUSEPORT:
1601 	case SO_OOBINLINE:
1602 	case SO_TIMESTAMP:
1603 		if (m == NULL || m->m_len < sizeof(int))
1604 			return EINVAL;
1605 		if (*mtod(m, int *))
1606 			so->so_options |= optname;
1607 		else
1608 			so->so_options &= ~optname;
1609 		break;
1610 
1611 	case SO_SNDBUF:
1612 	case SO_RCVBUF:
1613 	case SO_SNDLOWAT:
1614 	case SO_RCVLOWAT:
1615 		if (m == NULL || m->m_len < sizeof(int))
1616 			return EINVAL;
1617 
1618 		/*
1619 		 * Values < 1 make no sense for any of these
1620 		 * options, so disallow them.
1621 		 */
1622 		optval = *mtod(m, int *);
1623 		if (optval < 1)
1624 			return EINVAL;
1625 
1626 		switch (optname) {
1627 
1628 		case SO_SNDBUF:
1629 		case SO_RCVBUF:
1630 			sb = (optname == SO_SNDBUF) ?
1631 			    &so->so_snd : &so->so_rcv;
1632 			if (sbreserve(sb, (u_long)optval, so) == 0)
1633 				return ENOBUFS;
1634 			sb->sb_flags &= ~SB_AUTOSIZE;
1635 			break;
1636 
1637 		/*
1638 		 * Make sure the low-water is never greater than
1639 		 * the high-water.
1640 		 */
1641 		case SO_SNDLOWAT:
1642 			so->so_snd.sb_lowat =
1643 			    (optval > so->so_snd.sb_hiwat) ?
1644 			    so->so_snd.sb_hiwat : optval;
1645 			break;
1646 		case SO_RCVLOWAT:
1647 			so->so_rcv.sb_lowat =
1648 			    (optval > so->so_rcv.sb_hiwat) ?
1649 			    so->so_rcv.sb_hiwat : optval;
1650 			break;
1651 		}
1652 		break;
1653 
1654 	case SO_SNDTIMEO:
1655 	case SO_RCVTIMEO:
1656 		if (m == NULL || m->m_len < sizeof(*tv))
1657 			return EINVAL;
1658 		tv = mtod(m, struct timeval *);
1659 		if (tv->tv_sec > (INT_MAX - tv->tv_usec / tick) / hz)
1660 			return EDOM;
1661 		val = tv->tv_sec * hz + tv->tv_usec / tick;
1662 		if (val == 0 && tv->tv_usec != 0)
1663 			val = 1;
1664 
1665 		switch (optname) {
1666 
1667 		case SO_SNDTIMEO:
1668 			so->so_snd.sb_timeo = val;
1669 			break;
1670 		case SO_RCVTIMEO:
1671 			so->so_rcv.sb_timeo = val;
1672 			break;
1673 		}
1674 		break;
1675 
1676 	default:
1677 		return ENOPROTOOPT;
1678 	}
1679 	return 0;
1680 }
1681 
1682 int
1683 sosetopt(struct socket *so, int level, int optname, struct mbuf *m)
1684 {
1685 	int error, prerr;
1686 
1687 	solock(so);
1688 	if (level == SOL_SOCKET)
1689 		error = sosetopt1(so, level, optname, m);
1690 	else
1691 		error = ENOPROTOOPT;
1692 
1693 	if ((error == 0 || error == ENOPROTOOPT) &&
1694 	    so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) {
1695 		/* give the protocol stack a shot */
1696 		prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, level,
1697 		    optname, &m);
1698 		if (prerr == 0)
1699 			error = 0;
1700 		else if (prerr != ENOPROTOOPT)
1701 			error = prerr;
1702 	} else if (m != NULL)
1703 		(void)m_free(m);
1704 	sounlock(so);
1705 	return error;
1706 }
1707 
1708 int
1709 sogetopt(struct socket *so, int level, int optname, struct mbuf **mp)
1710 {
1711 	struct mbuf	*m;
1712 	int		error;
1713 
1714 	solock(so);
1715 	if (level != SOL_SOCKET) {
1716 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1717 			error = ((*so->so_proto->pr_ctloutput)
1718 				  (PRCO_GETOPT, so, level, optname, mp));
1719 		} else
1720 			error = (ENOPROTOOPT);
1721 	} else {
1722 		m = m_get(M_WAIT, MT_SOOPTS);
1723 		m->m_len = sizeof(int);
1724 
1725 		switch (optname) {
1726 
1727 		case SO_LINGER:
1728 			m->m_len = sizeof(struct linger);
1729 			mtod(m, struct linger *)->l_onoff =
1730 			    (so->so_options & SO_LINGER) ? 1 : 0;
1731 			mtod(m, struct linger *)->l_linger = so->so_linger;
1732 			break;
1733 
1734 		case SO_USELOOPBACK:
1735 		case SO_DONTROUTE:
1736 		case SO_DEBUG:
1737 		case SO_KEEPALIVE:
1738 		case SO_REUSEADDR:
1739 		case SO_REUSEPORT:
1740 		case SO_BROADCAST:
1741 		case SO_OOBINLINE:
1742 		case SO_TIMESTAMP:
1743 			*mtod(m, int *) = (so->so_options & optname) ? 1 : 0;
1744 			break;
1745 
1746 		case SO_TYPE:
1747 			*mtod(m, int *) = so->so_type;
1748 			break;
1749 
1750 		case SO_ERROR:
1751 			*mtod(m, int *) = so->so_error;
1752 			so->so_error = 0;
1753 			break;
1754 
1755 		case SO_SNDBUF:
1756 			*mtod(m, int *) = so->so_snd.sb_hiwat;
1757 			break;
1758 
1759 		case SO_RCVBUF:
1760 			*mtod(m, int *) = so->so_rcv.sb_hiwat;
1761 			break;
1762 
1763 		case SO_SNDLOWAT:
1764 			*mtod(m, int *) = so->so_snd.sb_lowat;
1765 			break;
1766 
1767 		case SO_RCVLOWAT:
1768 			*mtod(m, int *) = so->so_rcv.sb_lowat;
1769 			break;
1770 
1771 		case SO_SNDTIMEO:
1772 		case SO_RCVTIMEO:
1773 		    {
1774 			int val = (optname == SO_SNDTIMEO ?
1775 			     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1776 
1777 			m->m_len = sizeof(struct timeval);
1778 			mtod(m, struct timeval *)->tv_sec = val / hz;
1779 			mtod(m, struct timeval *)->tv_usec =
1780 			    (val % hz) * tick;
1781 			break;
1782 		    }
1783 
1784 		case SO_OVERFLOWED:
1785 			*mtod(m, int *) = so->so_rcv.sb_overflowed;
1786 			break;
1787 
1788 		default:
1789 			sounlock(so);
1790 			(void)m_free(m);
1791 			return (ENOPROTOOPT);
1792 		}
1793 		*mp = m;
1794 		error = 0;
1795 	}
1796 
1797 	sounlock(so);
1798 	return (error);
1799 }
1800 
1801 void
1802 sohasoutofband(struct socket *so)
1803 {
1804 
1805 	fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
1806 	selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, 0);
1807 }
1808 
1809 static void
1810 filt_sordetach(struct knote *kn)
1811 {
1812 	struct socket	*so;
1813 
1814 	so = ((file_t *)kn->kn_obj)->f_data;
1815 	solock(so);
1816 	SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
1817 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
1818 		so->so_rcv.sb_flags &= ~SB_KNOTE;
1819 	sounlock(so);
1820 }
1821 
1822 /*ARGSUSED*/
1823 static int
1824 filt_soread(struct knote *kn, long hint)
1825 {
1826 	struct socket	*so;
1827 	int rv;
1828 
1829 	so = ((file_t *)kn->kn_obj)->f_data;
1830 	if (hint != NOTE_SUBMIT)
1831 		solock(so);
1832 	kn->kn_data = so->so_rcv.sb_cc;
1833 	if (so->so_state & SS_CANTRCVMORE) {
1834 		kn->kn_flags |= EV_EOF;
1835 		kn->kn_fflags = so->so_error;
1836 		rv = 1;
1837 	} else if (so->so_error)	/* temporary udp error */
1838 		rv = 1;
1839 	else if (kn->kn_sfflags & NOTE_LOWAT)
1840 		rv = (kn->kn_data >= kn->kn_sdata);
1841 	else
1842 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
1843 	if (hint != NOTE_SUBMIT)
1844 		sounlock(so);
1845 	return rv;
1846 }
1847 
1848 static void
1849 filt_sowdetach(struct knote *kn)
1850 {
1851 	struct socket	*so;
1852 
1853 	so = ((file_t *)kn->kn_obj)->f_data;
1854 	solock(so);
1855 	SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
1856 	if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
1857 		so->so_snd.sb_flags &= ~SB_KNOTE;
1858 	sounlock(so);
1859 }
1860 
1861 /*ARGSUSED*/
1862 static int
1863 filt_sowrite(struct knote *kn, long hint)
1864 {
1865 	struct socket	*so;
1866 	int rv;
1867 
1868 	so = ((file_t *)kn->kn_obj)->f_data;
1869 	if (hint != NOTE_SUBMIT)
1870 		solock(so);
1871 	kn->kn_data = sbspace(&so->so_snd);
1872 	if (so->so_state & SS_CANTSENDMORE) {
1873 		kn->kn_flags |= EV_EOF;
1874 		kn->kn_fflags = so->so_error;
1875 		rv = 1;
1876 	} else if (so->so_error)	/* temporary udp error */
1877 		rv = 1;
1878 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
1879 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
1880 		rv = 0;
1881 	else if (kn->kn_sfflags & NOTE_LOWAT)
1882 		rv = (kn->kn_data >= kn->kn_sdata);
1883 	else
1884 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
1885 	if (hint != NOTE_SUBMIT)
1886 		sounlock(so);
1887 	return rv;
1888 }
1889 
1890 /*ARGSUSED*/
1891 static int
1892 filt_solisten(struct knote *kn, long hint)
1893 {
1894 	struct socket	*so;
1895 	int rv;
1896 
1897 	so = ((file_t *)kn->kn_obj)->f_data;
1898 
1899 	/*
1900 	 * Set kn_data to number of incoming connections, not
1901 	 * counting partial (incomplete) connections.
1902 	 */
1903 	if (hint != NOTE_SUBMIT)
1904 		solock(so);
1905 	kn->kn_data = so->so_qlen;
1906 	rv = (kn->kn_data > 0);
1907 	if (hint != NOTE_SUBMIT)
1908 		sounlock(so);
1909 	return rv;
1910 }
1911 
1912 static const struct filterops solisten_filtops =
1913 	{ 1, NULL, filt_sordetach, filt_solisten };
1914 static const struct filterops soread_filtops =
1915 	{ 1, NULL, filt_sordetach, filt_soread };
1916 static const struct filterops sowrite_filtops =
1917 	{ 1, NULL, filt_sowdetach, filt_sowrite };
1918 
1919 int
1920 soo_kqfilter(struct file *fp, struct knote *kn)
1921 {
1922 	struct socket	*so;
1923 	struct sockbuf	*sb;
1924 
1925 	so = ((file_t *)kn->kn_obj)->f_data;
1926 	solock(so);
1927 	switch (kn->kn_filter) {
1928 	case EVFILT_READ:
1929 		if (so->so_options & SO_ACCEPTCONN)
1930 			kn->kn_fop = &solisten_filtops;
1931 		else
1932 			kn->kn_fop = &soread_filtops;
1933 		sb = &so->so_rcv;
1934 		break;
1935 	case EVFILT_WRITE:
1936 		kn->kn_fop = &sowrite_filtops;
1937 		sb = &so->so_snd;
1938 		break;
1939 	default:
1940 		sounlock(so);
1941 		return (EINVAL);
1942 	}
1943 	SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
1944 	sb->sb_flags |= SB_KNOTE;
1945 	sounlock(so);
1946 	return (0);
1947 }
1948 
1949 static int
1950 sodopoll(struct socket *so, int events)
1951 {
1952 	int revents;
1953 
1954 	revents = 0;
1955 
1956 	if (events & (POLLIN | POLLRDNORM))
1957 		if (soreadable(so))
1958 			revents |= events & (POLLIN | POLLRDNORM);
1959 
1960 	if (events & (POLLOUT | POLLWRNORM))
1961 		if (sowritable(so))
1962 			revents |= events & (POLLOUT | POLLWRNORM);
1963 
1964 	if (events & (POLLPRI | POLLRDBAND))
1965 		if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
1966 			revents |= events & (POLLPRI | POLLRDBAND);
1967 
1968 	return revents;
1969 }
1970 
1971 int
1972 sopoll(struct socket *so, int events)
1973 {
1974 	int revents = 0;
1975 
1976 #ifndef DIAGNOSTIC
1977 	/*
1978 	 * Do a quick, unlocked check in expectation that the socket
1979 	 * will be ready for I/O.  Don't do this check if DIAGNOSTIC,
1980 	 * as the solocked() assertions will fail.
1981 	 */
1982 	if ((revents = sodopoll(so, events)) != 0)
1983 		return revents;
1984 #endif
1985 
1986 	solock(so);
1987 	if ((revents = sodopoll(so, events)) == 0) {
1988 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
1989 			selrecord(curlwp, &so->so_rcv.sb_sel);
1990 			so->so_rcv.sb_flags |= SB_NOTIFY;
1991 		}
1992 
1993 		if (events & (POLLOUT | POLLWRNORM)) {
1994 			selrecord(curlwp, &so->so_snd.sb_sel);
1995 			so->so_snd.sb_flags |= SB_NOTIFY;
1996 		}
1997 	}
1998 	sounlock(so);
1999 
2000 	return revents;
2001 }
2002 
2003 
2004 #include <sys/sysctl.h>
2005 
2006 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO);
2007 
2008 /*
2009  * sysctl helper routine for kern.somaxkva.  ensures that the given
2010  * value is not too small.
2011  * (XXX should we maybe make sure it's not too large as well?)
2012  */
2013 static int
2014 sysctl_kern_somaxkva(SYSCTLFN_ARGS)
2015 {
2016 	int error, new_somaxkva;
2017 	struct sysctlnode node;
2018 
2019 	new_somaxkva = somaxkva;
2020 	node = *rnode;
2021 	node.sysctl_data = &new_somaxkva;
2022 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2023 	if (error || newp == NULL)
2024 		return (error);
2025 
2026 	if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */
2027 		return (EINVAL);
2028 
2029 	mutex_enter(&so_pendfree_lock);
2030 	somaxkva = new_somaxkva;
2031 	cv_broadcast(&socurkva_cv);
2032 	mutex_exit(&so_pendfree_lock);
2033 
2034 	return (error);
2035 }
2036 
2037 SYSCTL_SETUP(sysctl_kern_somaxkva_setup, "sysctl kern.somaxkva setup")
2038 {
2039 
2040 	sysctl_createv(clog, 0, NULL, NULL,
2041 		       CTLFLAG_PERMANENT,
2042 		       CTLTYPE_NODE, "kern", NULL,
2043 		       NULL, 0, NULL, 0,
2044 		       CTL_KERN, CTL_EOL);
2045 
2046 	sysctl_createv(clog, 0, NULL, NULL,
2047 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2048 		       CTLTYPE_INT, "somaxkva",
2049 		       SYSCTL_DESCR("Maximum amount of kernel memory to be "
2050 				    "used for socket buffers"),
2051 		       sysctl_kern_somaxkva, 0, NULL, 0,
2052 		       CTL_KERN, KERN_SOMAXKVA, CTL_EOL);
2053 }
2054