xref: /netbsd-src/sys/kern/uipc_socket.c (revision 4b71a66d0f279143147d63ebfcfd8a59499a3684)
1 /*	$NetBSD: uipc_socket.c,v 1.167 2008/05/28 21:01:42 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  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Copyright (c) 2004 The FreeBSD Foundation
34  * Copyright (c) 2004 Robert Watson
35  * Copyright (c) 1982, 1986, 1988, 1990, 1993
36  *	The Regents of the University of California.  All rights reserved.
37  *
38  * Redistribution and use in source and binary forms, with or without
39  * modification, are permitted provided that the following conditions
40  * are met:
41  * 1. Redistributions of source code must retain the above copyright
42  *    notice, this list of conditions and the following disclaimer.
43  * 2. Redistributions in binary form must reproduce the above copyright
44  *    notice, this list of conditions and the following disclaimer in the
45  *    documentation and/or other materials provided with the distribution.
46  * 3. Neither the name of the University nor the names of its contributors
47  *    may be used to endorse or promote products derived from this software
48  *    without specific prior written permission.
49  *
50  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
52  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
53  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
54  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
55  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
56  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
57  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
58  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
59  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
60  * SUCH DAMAGE.
61  *
62  *	@(#)uipc_socket.c	8.6 (Berkeley) 5/2/95
63  */
64 
65 #include <sys/cdefs.h>
66 __KERNEL_RCSID(0, "$NetBSD: uipc_socket.c,v 1.167 2008/05/28 21:01:42 ad Exp $");
67 
68 #include "opt_sock_counters.h"
69 #include "opt_sosend_loan.h"
70 #include "opt_mbuftrace.h"
71 #include "opt_somaxkva.h"
72 #include "opt_multiprocessor.h"	/* XXX */
73 
74 #include <sys/param.h>
75 #include <sys/systm.h>
76 #include <sys/proc.h>
77 #include <sys/file.h>
78 #include <sys/filedesc.h>
79 #include <sys/malloc.h>
80 #include <sys/mbuf.h>
81 #include <sys/domain.h>
82 #include <sys/kernel.h>
83 #include <sys/protosw.h>
84 #include <sys/socket.h>
85 #include <sys/socketvar.h>
86 #include <sys/signalvar.h>
87 #include <sys/resourcevar.h>
88 #include <sys/event.h>
89 #include <sys/poll.h>
90 #include <sys/kauth.h>
91 #include <sys/mutex.h>
92 #include <sys/condvar.h>
93 
94 #include <uvm/uvm.h>
95 
96 MALLOC_DEFINE(M_SOOPTS, "soopts", "socket options");
97 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
98 
99 extern const struct fileops socketops;
100 
101 extern int	somaxconn;			/* patchable (XXX sysctl) */
102 int		somaxconn = SOMAXCONN;
103 kmutex_t	*softnet_lock;
104 
105 #ifdef SOSEND_COUNTERS
106 #include <sys/device.h>
107 
108 static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
109     NULL, "sosend", "loan big");
110 static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
111     NULL, "sosend", "copy big");
112 static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
113     NULL, "sosend", "copy small");
114 static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
115     NULL, "sosend", "kva limit");
116 
117 #define	SOSEND_COUNTER_INCR(ev)		(ev)->ev_count++
118 
119 EVCNT_ATTACH_STATIC(sosend_loan_big);
120 EVCNT_ATTACH_STATIC(sosend_copy_big);
121 EVCNT_ATTACH_STATIC(sosend_copy_small);
122 EVCNT_ATTACH_STATIC(sosend_kvalimit);
123 #else
124 
125 #define	SOSEND_COUNTER_INCR(ev)		/* nothing */
126 
127 #endif /* SOSEND_COUNTERS */
128 
129 static struct callback_entry sokva_reclaimerentry;
130 
131 #if defined(SOSEND_NO_LOAN) || defined(MULTIPROCESSOR)
132 int sock_loan_thresh = -1;
133 #else
134 int sock_loan_thresh = 4096;
135 #endif
136 
137 static kmutex_t so_pendfree_lock;
138 static struct mbuf *so_pendfree;
139 
140 #ifndef SOMAXKVA
141 #define	SOMAXKVA (16 * 1024 * 1024)
142 #endif
143 int somaxkva = SOMAXKVA;
144 static int socurkva;
145 static kcondvar_t socurkva_cv;
146 
147 #define	SOCK_LOAN_CHUNK		65536
148 
149 static size_t sodopendfree(void);
150 static size_t sodopendfreel(void);
151 
152 static vsize_t
153 sokvareserve(struct socket *so, vsize_t len)
154 {
155 	int error;
156 
157 	mutex_enter(&so_pendfree_lock);
158 	while (socurkva + len > somaxkva) {
159 		size_t freed;
160 
161 		/*
162 		 * try to do pendfree.
163 		 */
164 
165 		freed = sodopendfreel();
166 
167 		/*
168 		 * if some kva was freed, try again.
169 		 */
170 
171 		if (freed)
172 			continue;
173 
174 		SOSEND_COUNTER_INCR(&sosend_kvalimit);
175 		error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock);
176 		if (error) {
177 			len = 0;
178 			break;
179 		}
180 	}
181 	socurkva += len;
182 	mutex_exit(&so_pendfree_lock);
183 	return len;
184 }
185 
186 static void
187 sokvaunreserve(vsize_t len)
188 {
189 
190 	mutex_enter(&so_pendfree_lock);
191 	socurkva -= len;
192 	cv_broadcast(&socurkva_cv);
193 	mutex_exit(&so_pendfree_lock);
194 }
195 
196 /*
197  * sokvaalloc: allocate kva for loan.
198  */
199 
200 vaddr_t
201 sokvaalloc(vsize_t len, struct socket *so)
202 {
203 	vaddr_t lva;
204 
205 	/*
206 	 * reserve kva.
207 	 */
208 
209 	if (sokvareserve(so, len) == 0)
210 		return 0;
211 
212 	/*
213 	 * allocate kva.
214 	 */
215 
216 	lva = uvm_km_alloc(kernel_map, len, 0, UVM_KMF_VAONLY | UVM_KMF_WAITVA);
217 	if (lva == 0) {
218 		sokvaunreserve(len);
219 		return (0);
220 	}
221 
222 	return lva;
223 }
224 
225 /*
226  * sokvafree: free kva for loan.
227  */
228 
229 void
230 sokvafree(vaddr_t sva, vsize_t len)
231 {
232 
233 	/*
234 	 * free kva.
235 	 */
236 
237 	uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY);
238 
239 	/*
240 	 * unreserve kva.
241 	 */
242 
243 	sokvaunreserve(len);
244 }
245 
246 static void
247 sodoloanfree(struct vm_page **pgs, void *buf, size_t size)
248 {
249 	vaddr_t sva, eva;
250 	vsize_t len;
251 	int npgs;
252 
253 	KASSERT(pgs != NULL);
254 
255 	eva = round_page((vaddr_t) buf + size);
256 	sva = trunc_page((vaddr_t) buf);
257 	len = eva - sva;
258 	npgs = len >> PAGE_SHIFT;
259 
260 	pmap_kremove(sva, len);
261 	pmap_update(pmap_kernel());
262 	uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE);
263 	sokvafree(sva, len);
264 }
265 
266 static size_t
267 sodopendfree(void)
268 {
269 	size_t rv;
270 
271 	if (__predict_true(so_pendfree == NULL))
272 		return 0;
273 
274 	mutex_enter(&so_pendfree_lock);
275 	rv = sodopendfreel();
276 	mutex_exit(&so_pendfree_lock);
277 
278 	return rv;
279 }
280 
281 /*
282  * sodopendfreel: free mbufs on "pendfree" list.
283  * unlock and relock so_pendfree_lock when freeing mbufs.
284  *
285  * => called with so_pendfree_lock held.
286  */
287 
288 static size_t
289 sodopendfreel(void)
290 {
291 	struct mbuf *m, *next;
292 	size_t rv = 0;
293 
294 	KASSERT(mutex_owned(&so_pendfree_lock));
295 
296 	while (so_pendfree != NULL) {
297 		m = so_pendfree;
298 		so_pendfree = NULL;
299 		mutex_exit(&so_pendfree_lock);
300 
301 		for (; m != NULL; m = next) {
302 			next = m->m_next;
303 			KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0);
304 			KASSERT(m->m_ext.ext_refcnt == 0);
305 
306 			rv += m->m_ext.ext_size;
307 			sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf,
308 			    m->m_ext.ext_size);
309 			pool_cache_put(mb_cache, m);
310 		}
311 
312 		mutex_enter(&so_pendfree_lock);
313 	}
314 
315 	return (rv);
316 }
317 
318 void
319 soloanfree(struct mbuf *m, void *buf, size_t size, void *arg)
320 {
321 
322 	KASSERT(m != NULL);
323 
324 	/*
325 	 * postpone freeing mbuf.
326 	 *
327 	 * we can't do it in interrupt context
328 	 * because we need to put kva back to kernel_map.
329 	 */
330 
331 	mutex_enter(&so_pendfree_lock);
332 	m->m_next = so_pendfree;
333 	so_pendfree = m;
334 	cv_broadcast(&socurkva_cv);
335 	mutex_exit(&so_pendfree_lock);
336 }
337 
338 static long
339 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space)
340 {
341 	struct iovec *iov = uio->uio_iov;
342 	vaddr_t sva, eva;
343 	vsize_t len;
344 	vaddr_t lva;
345 	int npgs, error;
346 	vaddr_t va;
347 	int i;
348 
349 	if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace))
350 		return (0);
351 
352 	if (iov->iov_len < (size_t) space)
353 		space = iov->iov_len;
354 	if (space > SOCK_LOAN_CHUNK)
355 		space = SOCK_LOAN_CHUNK;
356 
357 	eva = round_page((vaddr_t) iov->iov_base + space);
358 	sva = trunc_page((vaddr_t) iov->iov_base);
359 	len = eva - sva;
360 	npgs = len >> PAGE_SHIFT;
361 
362 	KASSERT(npgs <= M_EXT_MAXPAGES);
363 
364 	lva = sokvaalloc(len, so);
365 	if (lva == 0)
366 		return 0;
367 
368 	error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len,
369 	    m->m_ext.ext_pgs, UVM_LOAN_TOPAGE);
370 	if (error) {
371 		sokvafree(lva, len);
372 		return (0);
373 	}
374 
375 	for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE)
376 		pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
377 		    VM_PROT_READ);
378 	pmap_update(pmap_kernel());
379 
380 	lva += (vaddr_t) iov->iov_base & PAGE_MASK;
381 
382 	MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so);
383 	m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
384 
385 	uio->uio_resid -= space;
386 	/* uio_offset not updated, not set/used for write(2) */
387 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space;
388 	uio->uio_iov->iov_len -= space;
389 	if (uio->uio_iov->iov_len == 0) {
390 		uio->uio_iov++;
391 		uio->uio_iovcnt--;
392 	}
393 
394 	return (space);
395 }
396 
397 static int
398 sokva_reclaim_callback(struct callback_entry *ce, void *obj, void *arg)
399 {
400 
401 	KASSERT(ce == &sokva_reclaimerentry);
402 	KASSERT(obj == NULL);
403 
404 	sodopendfree();
405 	if (!vm_map_starved_p(kernel_map)) {
406 		return CALLBACK_CHAIN_ABORT;
407 	}
408 	return CALLBACK_CHAIN_CONTINUE;
409 }
410 
411 struct mbuf *
412 getsombuf(struct socket *so, int type)
413 {
414 	struct mbuf *m;
415 
416 	m = m_get(M_WAIT, type);
417 	MCLAIM(m, so->so_mowner);
418 	return m;
419 }
420 
421 struct mbuf *
422 m_intopt(struct socket *so, int val)
423 {
424 	struct mbuf *m;
425 
426 	m = getsombuf(so, MT_SOOPTS);
427 	m->m_len = sizeof(int);
428 	*mtod(m, int *) = val;
429 	return m;
430 }
431 
432 void
433 soinit(void)
434 {
435 
436 	mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM);
437 	softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
438 	cv_init(&socurkva_cv, "sokva");
439 	soinit2();
440 
441 	/* Set the initial adjusted socket buffer size. */
442 	if (sb_max_set(sb_max))
443 		panic("bad initial sb_max value: %lu", sb_max);
444 
445 	callback_register(&vm_map_to_kernel(kernel_map)->vmk_reclaim_callback,
446 	    &sokva_reclaimerentry, NULL, sokva_reclaim_callback);
447 }
448 
449 /*
450  * Socket operation routines.
451  * These routines are called by the routines in
452  * sys_socket.c or from a system process, and
453  * implement the semantics of socket operations by
454  * switching out to the protocol specific routines.
455  */
456 /*ARGSUSED*/
457 int
458 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l,
459 	 struct socket *lockso)
460 {
461 	const struct protosw	*prp;
462 	struct socket	*so;
463 	uid_t		uid;
464 	int		error;
465 	kmutex_t	*lock;
466 
467 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET,
468 	    KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type),
469 	    KAUTH_ARG(proto));
470 	if (error != 0)
471 		return error;
472 
473 	if (proto)
474 		prp = pffindproto(dom, proto, type);
475 	else
476 		prp = pffindtype(dom, type);
477 	if (prp == NULL) {
478 		/* no support for domain */
479 		if (pffinddomain(dom) == 0)
480 			return EAFNOSUPPORT;
481 		/* no support for socket type */
482 		if (proto == 0 && type != 0)
483 			return EPROTOTYPE;
484 		return EPROTONOSUPPORT;
485 	}
486 	if (prp->pr_usrreq == NULL)
487 		return EPROTONOSUPPORT;
488 	if (prp->pr_type != type)
489 		return EPROTOTYPE;
490 
491 	so = soget(true);
492 	so->so_type = type;
493 	so->so_proto = prp;
494 	so->so_send = sosend;
495 	so->so_receive = soreceive;
496 #ifdef MBUFTRACE
497 	so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner;
498 	so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner;
499 	so->so_mowner = &prp->pr_domain->dom_mowner;
500 #endif
501 	uid = kauth_cred_geteuid(l->l_cred);
502 	so->so_uidinfo = uid_find(uid);
503 	if (lockso != NULL) {
504 		/* Caller wants us to share a lock. */
505 		lock = lockso->so_lock;
506 		so->so_lock = lock;
507 		mutex_obj_hold(lock);
508 		mutex_enter(lock);
509 	} else {
510 		/* Lock assigned and taken during PRU_ATTACH. */
511 	}
512 	error = (*prp->pr_usrreq)(so, PRU_ATTACH, NULL,
513 	    (struct mbuf *)(long)proto, NULL, l);
514 	KASSERT(solocked(so));
515 	if (error != 0) {
516 		so->so_state |= SS_NOFDREF;
517 		sofree(so);
518 		return error;
519 	}
520 	sounlock(so);
521 	*aso = so;
522 	return 0;
523 }
524 
525 /* On success, write file descriptor to fdout and return zero.  On
526  * failure, return non-zero; *fdout will be undefined.
527  */
528 int
529 fsocreate(int domain, struct socket **sop, int type, int protocol,
530     struct lwp *l, int *fdout)
531 {
532 	struct socket	*so;
533 	struct file	*fp;
534 	int		fd, error;
535 
536 	if ((error = fd_allocfile(&fp, &fd)) != 0)
537 		return (error);
538 	fp->f_flag = FREAD|FWRITE;
539 	fp->f_type = DTYPE_SOCKET;
540 	fp->f_ops = &socketops;
541 	error = socreate(domain, &so, type, protocol, l, NULL);
542 	if (error != 0) {
543 		fd_abort(curproc, fp, fd);
544 	} else {
545 		if (sop != NULL)
546 			*sop = so;
547 		fp->f_data = so;
548 		fd_affix(curproc, fp, fd);
549 		*fdout = fd;
550 	}
551 	return error;
552 }
553 
554 int
555 sobind(struct socket *so, struct mbuf *nam, struct lwp *l)
556 {
557 	int	error;
558 
559 	solock(so);
560 	error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, NULL, nam, NULL, l);
561 	sounlock(so);
562 	return error;
563 }
564 
565 int
566 solisten(struct socket *so, int backlog, struct lwp *l)
567 {
568 	int	error;
569 
570 	solock(so);
571 	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
572 	    SS_ISDISCONNECTING)) != 0) {
573 	    	sounlock(so);
574 		return (EOPNOTSUPP);
575 	}
576 	error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, NULL,
577 	    NULL, NULL, l);
578 	if (error != 0) {
579 		sounlock(so);
580 		return error;
581 	}
582 	if (TAILQ_EMPTY(&so->so_q))
583 		so->so_options |= SO_ACCEPTCONN;
584 	if (backlog < 0)
585 		backlog = 0;
586 	so->so_qlimit = min(backlog, somaxconn);
587 	sounlock(so);
588 	return 0;
589 }
590 
591 void
592 sofree(struct socket *so)
593 {
594 	u_int refs;
595 
596 	KASSERT(solocked(so));
597 
598 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
599 		sounlock(so);
600 		return;
601 	}
602 	if (so->so_head) {
603 		/*
604 		 * We must not decommission a socket that's on the accept(2)
605 		 * queue.  If we do, then accept(2) may hang after select(2)
606 		 * indicated that the listening socket was ready.
607 		 */
608 		if (!soqremque(so, 0)) {
609 			sounlock(so);
610 			return;
611 		}
612 	}
613 	if (so->so_rcv.sb_hiwat)
614 		(void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0,
615 		    RLIM_INFINITY);
616 	if (so->so_snd.sb_hiwat)
617 		(void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0,
618 		    RLIM_INFINITY);
619 	sbrelease(&so->so_snd, so);
620 	KASSERT(!cv_has_waiters(&so->so_cv));
621 	KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv));
622 	KASSERT(!cv_has_waiters(&so->so_snd.sb_cv));
623 	sorflush(so);
624 	refs = so->so_aborting;	/* XXX */
625 	sounlock(so);
626 	if (refs == 0)		/* XXX */
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 	u_int refs;
704 	int error;
705 
706 	KASSERT(solocked(so));
707 	KASSERT(so->so_head == NULL);
708 
709 	so->so_aborting++;		/* XXX */
710 	error = (*so->so_proto->pr_usrreq)(so, PRU_ABORT, NULL,
711 	    NULL, NULL, NULL);
712 	refs = --so->so_aborting;	/* XXX */
713 	if (error || (refs == 0)) {
714 		sofree(so);
715 	} else {
716 		sounlock(so);
717 	}
718 	return error;
719 }
720 
721 int
722 soaccept(struct socket *so, struct mbuf *nam)
723 {
724 	int	error;
725 
726 	KASSERT(solocked(so));
727 
728 	error = 0;
729 	if ((so->so_state & SS_NOFDREF) == 0)
730 		panic("soaccept: !NOFDREF");
731 	so->so_state &= ~SS_NOFDREF;
732 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
733 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
734 		error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT,
735 		    NULL, nam, NULL, NULL);
736 	else
737 		error = ECONNABORTED;
738 
739 	return (error);
740 }
741 
742 int
743 soconnect(struct socket *so, struct mbuf *nam, struct lwp *l)
744 {
745 	int		error;
746 
747 	KASSERT(solocked(so));
748 
749 	if (so->so_options & SO_ACCEPTCONN)
750 		return (EOPNOTSUPP);
751 	/*
752 	 * If protocol is connection-based, can only connect once.
753 	 * Otherwise, if connected, try to disconnect first.
754 	 * This allows user to disconnect by connecting to, e.g.,
755 	 * a null address.
756 	 */
757 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
758 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
759 	    (error = sodisconnect(so))))
760 		error = EISCONN;
761 	else
762 		error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT,
763 		    NULL, nam, NULL, l);
764 	return (error);
765 }
766 
767 int
768 soconnect2(struct socket *so1, struct socket *so2)
769 {
770 	int	error;
771 
772 	KASSERT(solocked2(so1, so2));
773 
774 	error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2,
775 	    NULL, (struct mbuf *)so2, NULL, NULL);
776 	return (error);
777 }
778 
779 int
780 sodisconnect(struct socket *so)
781 {
782 	int	error;
783 
784 	KASSERT(solocked(so));
785 
786 	if ((so->so_state & SS_ISCONNECTED) == 0) {
787 		error = ENOTCONN;
788 	} else if (so->so_state & SS_ISDISCONNECTING) {
789 		error = EALREADY;
790 	} else {
791 		error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT,
792 		    NULL, NULL, NULL, NULL);
793 	}
794 	sodopendfree();
795 	return (error);
796 }
797 
798 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
799 /*
800  * Send on a socket.
801  * If send must go all at once and message is larger than
802  * send buffering, then hard error.
803  * Lock against other senders.
804  * If must go all at once and not enough room now, then
805  * inform user that this would block and do nothing.
806  * Otherwise, if nonblocking, send as much as possible.
807  * The data to be sent is described by "uio" if nonzero,
808  * otherwise by the mbuf chain "top" (which must be null
809  * if uio is not).  Data provided in mbuf chain must be small
810  * enough to send all at once.
811  *
812  * Returns nonzero on error, timeout or signal; callers
813  * must check for short counts if EINTR/ERESTART are returned.
814  * Data and control buffers are freed on return.
815  */
816 int
817 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
818 	struct mbuf *control, int flags, struct lwp *l)
819 {
820 	struct mbuf	**mp, *m;
821 	struct proc	*p;
822 	long		space, len, resid, clen, mlen;
823 	int		error, s, dontroute, atomic;
824 
825 	p = l->l_proc;
826 	sodopendfree();
827 	clen = 0;
828 
829 	/*
830 	 * solock() provides atomicity of access.  splsoftnet() prevents
831 	 * protocol processing soft interrupts from interrupting us and
832 	 * blocking (expensive).
833 	 */
834 	s = splsoftnet();
835 	solock(so);
836 	atomic = sosendallatonce(so) || top;
837 	if (uio)
838 		resid = uio->uio_resid;
839 	else
840 		resid = top->m_pkthdr.len;
841 	/*
842 	 * In theory resid should be unsigned.
843 	 * However, space must be signed, as it might be less than 0
844 	 * if we over-committed, and we must use a signed comparison
845 	 * of space and resid.  On the other hand, a negative resid
846 	 * causes us to loop sending 0-length segments to the protocol.
847 	 */
848 	if (resid < 0) {
849 		error = EINVAL;
850 		goto out;
851 	}
852 	dontroute =
853 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
854 	    (so->so_proto->pr_flags & PR_ATOMIC);
855 	l->l_ru.ru_msgsnd++;
856 	if (control)
857 		clen = control->m_len;
858  restart:
859 	if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
860 		goto out;
861 	do {
862 		if (so->so_state & SS_CANTSENDMORE) {
863 			error = EPIPE;
864 			goto release;
865 		}
866 		if (so->so_error) {
867 			error = so->so_error;
868 			so->so_error = 0;
869 			goto release;
870 		}
871 		if ((so->so_state & SS_ISCONNECTED) == 0) {
872 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
873 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
874 				    !(resid == 0 && clen != 0)) {
875 					error = ENOTCONN;
876 					goto release;
877 				}
878 			} else if (addr == 0) {
879 				error = EDESTADDRREQ;
880 				goto release;
881 			}
882 		}
883 		space = sbspace(&so->so_snd);
884 		if (flags & MSG_OOB)
885 			space += 1024;
886 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
887 		    clen > so->so_snd.sb_hiwat) {
888 			error = EMSGSIZE;
889 			goto release;
890 		}
891 		if (space < resid + clen &&
892 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
893 			if (so->so_nbio) {
894 				error = EWOULDBLOCK;
895 				goto release;
896 			}
897 			sbunlock(&so->so_snd);
898 			error = sbwait(&so->so_snd);
899 			if (error)
900 				goto out;
901 			goto restart;
902 		}
903 		mp = &top;
904 		space -= clen;
905 		do {
906 			if (uio == NULL) {
907 				/*
908 				 * Data is prepackaged in "top".
909 				 */
910 				resid = 0;
911 				if (flags & MSG_EOR)
912 					top->m_flags |= M_EOR;
913 			} else do {
914 				sounlock(so);
915 				splx(s);
916 				if (top == NULL) {
917 					m = m_gethdr(M_WAIT, MT_DATA);
918 					mlen = MHLEN;
919 					m->m_pkthdr.len = 0;
920 					m->m_pkthdr.rcvif = NULL;
921 				} else {
922 					m = m_get(M_WAIT, MT_DATA);
923 					mlen = MLEN;
924 				}
925 				MCLAIM(m, so->so_snd.sb_mowner);
926 				if (sock_loan_thresh >= 0 &&
927 				    uio->uio_iov->iov_len >= sock_loan_thresh &&
928 				    space >= sock_loan_thresh &&
929 				    (len = sosend_loan(so, uio, m,
930 						       space)) != 0) {
931 					SOSEND_COUNTER_INCR(&sosend_loan_big);
932 					space -= len;
933 					goto have_data;
934 				}
935 				if (resid >= MINCLSIZE && space >= MCLBYTES) {
936 					SOSEND_COUNTER_INCR(&sosend_copy_big);
937 					m_clget(m, M_WAIT);
938 					if ((m->m_flags & M_EXT) == 0)
939 						goto nopages;
940 					mlen = MCLBYTES;
941 					if (atomic && top == 0) {
942 						len = lmin(MCLBYTES - max_hdr,
943 						    resid);
944 						m->m_data += max_hdr;
945 					} else
946 						len = lmin(MCLBYTES, resid);
947 					space -= len;
948 				} else {
949  nopages:
950 					SOSEND_COUNTER_INCR(&sosend_copy_small);
951 					len = lmin(lmin(mlen, resid), space);
952 					space -= len;
953 					/*
954 					 * For datagram protocols, leave room
955 					 * for protocol headers in first mbuf.
956 					 */
957 					if (atomic && top == 0 && len < mlen)
958 						MH_ALIGN(m, len);
959 				}
960 				error = uiomove(mtod(m, void *), (int)len, uio);
961  have_data:
962 				resid = uio->uio_resid;
963 				m->m_len = len;
964 				*mp = m;
965 				top->m_pkthdr.len += len;
966 				s = splsoftnet();
967 				solock(so);
968 				if (error != 0)
969 					goto release;
970 				mp = &m->m_next;
971 				if (resid <= 0) {
972 					if (flags & MSG_EOR)
973 						top->m_flags |= M_EOR;
974 					break;
975 				}
976 			} while (space > 0 && atomic);
977 
978 			if (so->so_state & SS_CANTSENDMORE) {
979 				error = EPIPE;
980 				goto release;
981 			}
982 			if (dontroute)
983 				so->so_options |= SO_DONTROUTE;
984 			if (resid > 0)
985 				so->so_state |= SS_MORETOCOME;
986 			error = (*so->so_proto->pr_usrreq)(so,
987 			    (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND,
988 			    top, addr, control, curlwp);
989 			if (dontroute)
990 				so->so_options &= ~SO_DONTROUTE;
991 			if (resid > 0)
992 				so->so_state &= ~SS_MORETOCOME;
993 			clen = 0;
994 			control = NULL;
995 			top = NULL;
996 			mp = &top;
997 			if (error != 0)
998 				goto release;
999 		} while (resid && space > 0);
1000 	} while (resid);
1001 
1002  release:
1003 	sbunlock(&so->so_snd);
1004  out:
1005 	sounlock(so);
1006 	splx(s);
1007 	if (top)
1008 		m_freem(top);
1009 	if (control)
1010 		m_freem(control);
1011 	return (error);
1012 }
1013 
1014 /*
1015  * Following replacement or removal of the first mbuf on the first
1016  * mbuf chain of a socket buffer, push necessary state changes back
1017  * into the socket buffer so that other consumers see the values
1018  * consistently.  'nextrecord' is the callers locally stored value of
1019  * the original value of sb->sb_mb->m_nextpkt which must be restored
1020  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
1021  */
1022 static void
1023 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
1024 {
1025 
1026 	KASSERT(solocked(sb->sb_so));
1027 
1028 	/*
1029 	 * First, update for the new value of nextrecord.  If necessary,
1030 	 * make it the first record.
1031 	 */
1032 	if (sb->sb_mb != NULL)
1033 		sb->sb_mb->m_nextpkt = nextrecord;
1034 	else
1035 		sb->sb_mb = nextrecord;
1036 
1037         /*
1038          * Now update any dependent socket buffer fields to reflect
1039          * the new state.  This is an inline of SB_EMPTY_FIXUP, with
1040          * the addition of a second clause that takes care of the
1041          * case where sb_mb has been updated, but remains the last
1042          * record.
1043          */
1044         if (sb->sb_mb == NULL) {
1045                 sb->sb_mbtail = NULL;
1046                 sb->sb_lastrecord = NULL;
1047         } else if (sb->sb_mb->m_nextpkt == NULL)
1048                 sb->sb_lastrecord = sb->sb_mb;
1049 }
1050 
1051 /*
1052  * Implement receive operations on a socket.
1053  * We depend on the way that records are added to the sockbuf
1054  * by sbappend*.  In particular, each record (mbufs linked through m_next)
1055  * must begin with an address if the protocol so specifies,
1056  * followed by an optional mbuf or mbufs containing ancillary data,
1057  * and then zero or more mbufs of data.
1058  * In order to avoid blocking network interrupts for the entire time here,
1059  * we splx() while doing the actual copy to user space.
1060  * Although the sockbuf is locked, new data may still be appended,
1061  * and thus we must maintain consistency of the sockbuf during that time.
1062  *
1063  * The caller may receive the data as a single mbuf chain by supplying
1064  * an mbuf **mp0 for use in returning the chain.  The uio is then used
1065  * only for the count in uio_resid.
1066  */
1067 int
1068 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
1069 	struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1070 {
1071 	struct lwp *l = curlwp;
1072 	struct mbuf	*m, **mp, *mt;
1073 	int atomic, flags, len, error, s, offset, moff, type, orig_resid;
1074 	const struct protosw	*pr;
1075 	struct mbuf	*nextrecord;
1076 	int		mbuf_removed = 0;
1077 	const struct domain *dom;
1078 
1079 	pr = so->so_proto;
1080 	atomic = pr->pr_flags & PR_ATOMIC;
1081 	dom = pr->pr_domain;
1082 	mp = mp0;
1083 	type = 0;
1084 	orig_resid = uio->uio_resid;
1085 
1086 	if (paddr != NULL)
1087 		*paddr = NULL;
1088 	if (controlp != NULL)
1089 		*controlp = NULL;
1090 	if (flagsp != NULL)
1091 		flags = *flagsp &~ MSG_EOR;
1092 	else
1093 		flags = 0;
1094 
1095 	if ((flags & MSG_DONTWAIT) == 0)
1096 		sodopendfree();
1097 
1098 	if (flags & MSG_OOB) {
1099 		m = m_get(M_WAIT, MT_DATA);
1100 		solock(so);
1101 		error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m,
1102 		    (struct mbuf *)(long)(flags & MSG_PEEK), NULL, l);
1103 		sounlock(so);
1104 		if (error)
1105 			goto bad;
1106 		do {
1107 			error = uiomove(mtod(m, void *),
1108 			    (int) min(uio->uio_resid, m->m_len), uio);
1109 			m = m_free(m);
1110 		} while (uio->uio_resid > 0 && error == 0 && m);
1111  bad:
1112 		if (m != NULL)
1113 			m_freem(m);
1114 		return error;
1115 	}
1116 	if (mp != NULL)
1117 		*mp = NULL;
1118 
1119 	/*
1120 	 * solock() provides atomicity of access.  splsoftnet() prevents
1121 	 * protocol processing soft interrupts from interrupting us and
1122 	 * blocking (expensive).
1123 	 */
1124 	s = splsoftnet();
1125 	solock(so);
1126 	if (so->so_state & SS_ISCONFIRMING && uio->uio_resid)
1127 		(*pr->pr_usrreq)(so, PRU_RCVD, NULL, NULL, NULL, l);
1128 
1129  restart:
1130 	if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) {
1131 		sounlock(so);
1132 		splx(s);
1133 		return error;
1134 	}
1135 
1136 	m = so->so_rcv.sb_mb;
1137 	/*
1138 	 * If we have less data than requested, block awaiting more
1139 	 * (subject to any timeout) if:
1140 	 *   1. the current count is less than the low water mark,
1141 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1142 	 *	receive operation at once if we block (resid <= hiwat), or
1143 	 *   3. MSG_DONTWAIT is not set.
1144 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1145 	 * we have to do the receive in sections, and thus risk returning
1146 	 * a short count if a timeout or signal occurs after we start.
1147 	 */
1148 	if (m == NULL ||
1149 	    ((flags & MSG_DONTWAIT) == 0 &&
1150 	     so->so_rcv.sb_cc < uio->uio_resid &&
1151 	     (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1152 	      ((flags & MSG_WAITALL) &&
1153 	       uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1154 	     m->m_nextpkt == NULL && !atomic)) {
1155 #ifdef DIAGNOSTIC
1156 		if (m == NULL && so->so_rcv.sb_cc)
1157 			panic("receive 1");
1158 #endif
1159 		if (so->so_error) {
1160 			if (m != NULL)
1161 				goto dontblock;
1162 			error = so->so_error;
1163 			if ((flags & MSG_PEEK) == 0)
1164 				so->so_error = 0;
1165 			goto release;
1166 		}
1167 		if (so->so_state & SS_CANTRCVMORE) {
1168 			if (m != NULL)
1169 				goto dontblock;
1170 			else
1171 				goto release;
1172 		}
1173 		for (; m != NULL; m = m->m_next)
1174 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1175 				m = so->so_rcv.sb_mb;
1176 				goto dontblock;
1177 			}
1178 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1179 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1180 			error = ENOTCONN;
1181 			goto release;
1182 		}
1183 		if (uio->uio_resid == 0)
1184 			goto release;
1185 		if (so->so_nbio || (flags & MSG_DONTWAIT)) {
1186 			error = EWOULDBLOCK;
1187 			goto release;
1188 		}
1189 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
1190 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
1191 		sbunlock(&so->so_rcv);
1192 		error = sbwait(&so->so_rcv);
1193 		if (error != 0) {
1194 			sounlock(so);
1195 			splx(s);
1196 			return error;
1197 		}
1198 		goto restart;
1199 	}
1200  dontblock:
1201 	/*
1202 	 * On entry here, m points to the first record of the socket buffer.
1203 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1204 	 * pointer to the next record in the socket buffer.  We must keep the
1205 	 * various socket buffer pointers and local stack versions of the
1206 	 * pointers in sync, pushing out modifications before dropping the
1207 	 * socket lock, and re-reading them when picking it up.
1208 	 *
1209 	 * Otherwise, we will race with the network stack appending new data
1210 	 * or records onto the socket buffer by using inconsistent/stale
1211 	 * versions of the field, possibly resulting in socket buffer
1212 	 * corruption.
1213 	 *
1214 	 * By holding the high-level sblock(), we prevent simultaneous
1215 	 * readers from pulling off the front of the socket buffer.
1216 	 */
1217 	if (l != NULL)
1218 		l->l_ru.ru_msgrcv++;
1219 	KASSERT(m == so->so_rcv.sb_mb);
1220 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
1221 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
1222 	nextrecord = m->m_nextpkt;
1223 	if (pr->pr_flags & PR_ADDR) {
1224 #ifdef DIAGNOSTIC
1225 		if (m->m_type != MT_SONAME)
1226 			panic("receive 1a");
1227 #endif
1228 		orig_resid = 0;
1229 		if (flags & MSG_PEEK) {
1230 			if (paddr)
1231 				*paddr = m_copy(m, 0, m->m_len);
1232 			m = m->m_next;
1233 		} else {
1234 			sbfree(&so->so_rcv, m);
1235 			mbuf_removed = 1;
1236 			if (paddr != NULL) {
1237 				*paddr = m;
1238 				so->so_rcv.sb_mb = m->m_next;
1239 				m->m_next = NULL;
1240 				m = so->so_rcv.sb_mb;
1241 			} else {
1242 				MFREE(m, so->so_rcv.sb_mb);
1243 				m = so->so_rcv.sb_mb;
1244 			}
1245 			sbsync(&so->so_rcv, nextrecord);
1246 		}
1247 	}
1248 
1249 	/*
1250 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1251 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1252 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1253 	 * perform externalization (or freeing if controlp == NULL).
1254 	 */
1255 	if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) {
1256 		struct mbuf *cm = NULL, *cmn;
1257 		struct mbuf **cme = &cm;
1258 
1259 		do {
1260 			if (flags & MSG_PEEK) {
1261 				if (controlp != NULL) {
1262 					*controlp = m_copy(m, 0, m->m_len);
1263 					controlp = &(*controlp)->m_next;
1264 				}
1265 				m = m->m_next;
1266 			} else {
1267 				sbfree(&so->so_rcv, m);
1268 				so->so_rcv.sb_mb = m->m_next;
1269 				m->m_next = NULL;
1270 				*cme = m;
1271 				cme = &(*cme)->m_next;
1272 				m = so->so_rcv.sb_mb;
1273 			}
1274 		} while (m != NULL && m->m_type == MT_CONTROL);
1275 		if ((flags & MSG_PEEK) == 0)
1276 			sbsync(&so->so_rcv, nextrecord);
1277 		for (; cm != NULL; cm = cmn) {
1278 			cmn = cm->m_next;
1279 			cm->m_next = NULL;
1280 			type = mtod(cm, struct cmsghdr *)->cmsg_type;
1281 			if (controlp != NULL) {
1282 				if (dom->dom_externalize != NULL &&
1283 				    type == SCM_RIGHTS) {
1284 					sounlock(so);
1285 					splx(s);
1286 					error = (*dom->dom_externalize)(cm, l);
1287 					s = splsoftnet();
1288 					solock(so);
1289 				}
1290 				*controlp = cm;
1291 				while (*controlp != NULL)
1292 					controlp = &(*controlp)->m_next;
1293 			} else {
1294 				/*
1295 				 * Dispose of any SCM_RIGHTS message that went
1296 				 * through the read path rather than recv.
1297 				 */
1298 				if (dom->dom_dispose != NULL &&
1299 				    type == SCM_RIGHTS) {
1300 				    	sounlock(so);
1301 					(*dom->dom_dispose)(cm);
1302 					solock(so);
1303 				}
1304 				m_freem(cm);
1305 			}
1306 		}
1307 		if (m != NULL)
1308 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1309 		else
1310 			nextrecord = so->so_rcv.sb_mb;
1311 		orig_resid = 0;
1312 	}
1313 
1314 	/* If m is non-NULL, we have some data to read. */
1315 	if (__predict_true(m != NULL)) {
1316 		type = m->m_type;
1317 		if (type == MT_OOBDATA)
1318 			flags |= MSG_OOB;
1319 	}
1320 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1321 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1322 
1323 	moff = 0;
1324 	offset = 0;
1325 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1326 		if (m->m_type == MT_OOBDATA) {
1327 			if (type != MT_OOBDATA)
1328 				break;
1329 		} else if (type == MT_OOBDATA)
1330 			break;
1331 #ifdef DIAGNOSTIC
1332 		else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
1333 			panic("receive 3");
1334 #endif
1335 		so->so_state &= ~SS_RCVATMARK;
1336 		len = uio->uio_resid;
1337 		if (so->so_oobmark && len > so->so_oobmark - offset)
1338 			len = so->so_oobmark - offset;
1339 		if (len > m->m_len - moff)
1340 			len = m->m_len - moff;
1341 		/*
1342 		 * If mp is set, just pass back the mbufs.
1343 		 * Otherwise copy them out via the uio, then free.
1344 		 * Sockbuf must be consistent here (points to current mbuf,
1345 		 * it points to next record) when we drop priority;
1346 		 * we must note any additions to the sockbuf when we
1347 		 * block interrupts again.
1348 		 */
1349 		if (mp == NULL) {
1350 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1351 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1352 			sounlock(so);
1353 			splx(s);
1354 			error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1355 			s = splsoftnet();
1356 			solock(so);
1357 			if (error != 0) {
1358 				/*
1359 				 * If any part of the record has been removed
1360 				 * (such as the MT_SONAME mbuf, which will
1361 				 * happen when PR_ADDR, and thus also
1362 				 * PR_ATOMIC, is set), then drop the entire
1363 				 * record to maintain the atomicity of the
1364 				 * receive operation.
1365 				 *
1366 				 * This avoids a later panic("receive 1a")
1367 				 * when compiled with DIAGNOSTIC.
1368 				 */
1369 				if (m && mbuf_removed && atomic)
1370 					(void) sbdroprecord(&so->so_rcv);
1371 
1372 				goto release;
1373 			}
1374 		} else
1375 			uio->uio_resid -= len;
1376 		if (len == m->m_len - moff) {
1377 			if (m->m_flags & M_EOR)
1378 				flags |= MSG_EOR;
1379 			if (flags & MSG_PEEK) {
1380 				m = m->m_next;
1381 				moff = 0;
1382 			} else {
1383 				nextrecord = m->m_nextpkt;
1384 				sbfree(&so->so_rcv, m);
1385 				if (mp) {
1386 					*mp = m;
1387 					mp = &m->m_next;
1388 					so->so_rcv.sb_mb = m = m->m_next;
1389 					*mp = NULL;
1390 				} else {
1391 					MFREE(m, so->so_rcv.sb_mb);
1392 					m = so->so_rcv.sb_mb;
1393 				}
1394 				/*
1395 				 * If m != NULL, we also know that
1396 				 * so->so_rcv.sb_mb != NULL.
1397 				 */
1398 				KASSERT(so->so_rcv.sb_mb == m);
1399 				if (m) {
1400 					m->m_nextpkt = nextrecord;
1401 					if (nextrecord == NULL)
1402 						so->so_rcv.sb_lastrecord = m;
1403 				} else {
1404 					so->so_rcv.sb_mb = nextrecord;
1405 					SB_EMPTY_FIXUP(&so->so_rcv);
1406 				}
1407 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1408 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1409 			}
1410 		} else if (flags & MSG_PEEK)
1411 			moff += len;
1412 		else {
1413 			if (mp != NULL) {
1414 				mt = m_copym(m, 0, len, M_NOWAIT);
1415 				if (__predict_false(mt == NULL)) {
1416 					sounlock(so);
1417 					mt = m_copym(m, 0, len, M_WAIT);
1418 					solock(so);
1419 				}
1420 				*mp = mt;
1421 			}
1422 			m->m_data += len;
1423 			m->m_len -= len;
1424 			so->so_rcv.sb_cc -= len;
1425 		}
1426 		if (so->so_oobmark) {
1427 			if ((flags & MSG_PEEK) == 0) {
1428 				so->so_oobmark -= len;
1429 				if (so->so_oobmark == 0) {
1430 					so->so_state |= SS_RCVATMARK;
1431 					break;
1432 				}
1433 			} else {
1434 				offset += len;
1435 				if (offset == so->so_oobmark)
1436 					break;
1437 			}
1438 		}
1439 		if (flags & MSG_EOR)
1440 			break;
1441 		/*
1442 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1443 		 * we must not quit until "uio->uio_resid == 0" or an error
1444 		 * termination.  If a signal/timeout occurs, return
1445 		 * with a short count but without error.
1446 		 * Keep sockbuf locked against other readers.
1447 		 */
1448 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1449 		    !sosendallatonce(so) && !nextrecord) {
1450 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1451 				break;
1452 			/*
1453 			 * If we are peeking and the socket receive buffer is
1454 			 * full, stop since we can't get more data to peek at.
1455 			 */
1456 			if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
1457 				break;
1458 			/*
1459 			 * If we've drained the socket buffer, tell the
1460 			 * protocol in case it needs to do something to
1461 			 * get it filled again.
1462 			 */
1463 			if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1464 				(*pr->pr_usrreq)(so, PRU_RCVD,
1465 				    NULL, (struct mbuf *)(long)flags, NULL, l);
1466 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1467 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1468 			error = sbwait(&so->so_rcv);
1469 			if (error != 0) {
1470 				sbunlock(&so->so_rcv);
1471 				sounlock(so);
1472 				splx(s);
1473 				return 0;
1474 			}
1475 			if ((m = so->so_rcv.sb_mb) != NULL)
1476 				nextrecord = m->m_nextpkt;
1477 		}
1478 	}
1479 
1480 	if (m && atomic) {
1481 		flags |= MSG_TRUNC;
1482 		if ((flags & MSG_PEEK) == 0)
1483 			(void) sbdroprecord(&so->so_rcv);
1484 	}
1485 	if ((flags & MSG_PEEK) == 0) {
1486 		if (m == NULL) {
1487 			/*
1488 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1489 			 * part makes sure sb_lastrecord is up-to-date if
1490 			 * there is still data in the socket buffer.
1491 			 */
1492 			so->so_rcv.sb_mb = nextrecord;
1493 			if (so->so_rcv.sb_mb == NULL) {
1494 				so->so_rcv.sb_mbtail = NULL;
1495 				so->so_rcv.sb_lastrecord = NULL;
1496 			} else if (nextrecord->m_nextpkt == NULL)
1497 				so->so_rcv.sb_lastrecord = nextrecord;
1498 		}
1499 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1500 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1501 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1502 			(*pr->pr_usrreq)(so, PRU_RCVD, NULL,
1503 			    (struct mbuf *)(long)flags, NULL, l);
1504 	}
1505 	if (orig_resid == uio->uio_resid && orig_resid &&
1506 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1507 		sbunlock(&so->so_rcv);
1508 		goto restart;
1509 	}
1510 
1511 	if (flagsp != NULL)
1512 		*flagsp |= flags;
1513  release:
1514 	sbunlock(&so->so_rcv);
1515 	sounlock(so);
1516 	splx(s);
1517 	return error;
1518 }
1519 
1520 int
1521 soshutdown(struct socket *so, int how)
1522 {
1523 	const struct protosw	*pr;
1524 	int	error;
1525 
1526 	KASSERT(solocked(so));
1527 
1528 	pr = so->so_proto;
1529 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1530 		return (EINVAL);
1531 
1532 	if (how == SHUT_RD || how == SHUT_RDWR) {
1533 		sorflush(so);
1534 		error = 0;
1535 	}
1536 	if (how == SHUT_WR || how == SHUT_RDWR)
1537 		error = (*pr->pr_usrreq)(so, PRU_SHUTDOWN, NULL,
1538 		    NULL, NULL, NULL);
1539 
1540 	return error;
1541 }
1542 
1543 void
1544 sorflush(struct socket *so)
1545 {
1546 	struct sockbuf	*sb, asb;
1547 	const struct protosw	*pr;
1548 
1549 	KASSERT(solocked(so));
1550 
1551 	sb = &so->so_rcv;
1552 	pr = so->so_proto;
1553 	socantrcvmore(so);
1554 	sb->sb_flags |= SB_NOINTR;
1555 	(void )sblock(sb, M_WAITOK);
1556 	sbunlock(sb);
1557 	asb = *sb;
1558 	/*
1559 	 * Clear most of the sockbuf structure, but leave some of the
1560 	 * fields valid.
1561 	 */
1562 	memset(&sb->sb_startzero, 0,
1563 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1564 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) {
1565 		sounlock(so);
1566 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1567 		solock(so);
1568 	}
1569 	sbrelease(&asb, so);
1570 }
1571 
1572 static int
1573 sosetopt1(struct socket *so, int level, int optname, struct mbuf *m)
1574 {
1575 	int optval, val;
1576 	struct linger	*l;
1577 	struct sockbuf	*sb;
1578 	struct timeval *tv;
1579 
1580 	switch (optname) {
1581 
1582 	case SO_LINGER:
1583 		if (m == NULL || m->m_len != sizeof(struct linger))
1584 			return EINVAL;
1585 		l = mtod(m, struct linger *);
1586 		if (l->l_linger < 0 || l->l_linger > USHRT_MAX ||
1587 		    l->l_linger > (INT_MAX / hz))
1588 			return EDOM;
1589 		so->so_linger = l->l_linger;
1590 		if (l->l_onoff)
1591 			so->so_options |= SO_LINGER;
1592 		else
1593 			so->so_options &= ~SO_LINGER;
1594 		break;
1595 
1596 	case SO_DEBUG:
1597 	case SO_KEEPALIVE:
1598 	case SO_DONTROUTE:
1599 	case SO_USELOOPBACK:
1600 	case SO_BROADCAST:
1601 	case SO_REUSEADDR:
1602 	case SO_REUSEPORT:
1603 	case SO_OOBINLINE:
1604 	case SO_TIMESTAMP:
1605 		if (m == NULL || m->m_len < sizeof(int))
1606 			return EINVAL;
1607 		if (*mtod(m, int *))
1608 			so->so_options |= optname;
1609 		else
1610 			so->so_options &= ~optname;
1611 		break;
1612 
1613 	case SO_SNDBUF:
1614 	case SO_RCVBUF:
1615 	case SO_SNDLOWAT:
1616 	case SO_RCVLOWAT:
1617 		if (m == NULL || m->m_len < sizeof(int))
1618 			return EINVAL;
1619 
1620 		/*
1621 		 * Values < 1 make no sense for any of these
1622 		 * options, so disallow them.
1623 		 */
1624 		optval = *mtod(m, int *);
1625 		if (optval < 1)
1626 			return EINVAL;
1627 
1628 		switch (optname) {
1629 
1630 		case SO_SNDBUF:
1631 		case SO_RCVBUF:
1632 			sb = (optname == SO_SNDBUF) ?
1633 			    &so->so_snd : &so->so_rcv;
1634 			if (sbreserve(sb, (u_long)optval, so) == 0)
1635 				return ENOBUFS;
1636 			sb->sb_flags &= ~SB_AUTOSIZE;
1637 			break;
1638 
1639 		/*
1640 		 * Make sure the low-water is never greater than
1641 		 * the high-water.
1642 		 */
1643 		case SO_SNDLOWAT:
1644 			so->so_snd.sb_lowat =
1645 			    (optval > so->so_snd.sb_hiwat) ?
1646 			    so->so_snd.sb_hiwat : optval;
1647 			break;
1648 		case SO_RCVLOWAT:
1649 			so->so_rcv.sb_lowat =
1650 			    (optval > so->so_rcv.sb_hiwat) ?
1651 			    so->so_rcv.sb_hiwat : optval;
1652 			break;
1653 		}
1654 		break;
1655 
1656 	case SO_SNDTIMEO:
1657 	case SO_RCVTIMEO:
1658 		if (m == NULL || m->m_len < sizeof(*tv))
1659 			return EINVAL;
1660 		tv = mtod(m, struct timeval *);
1661 		if (tv->tv_sec > (INT_MAX - tv->tv_usec / tick) / hz)
1662 			return EDOM;
1663 		val = tv->tv_sec * hz + tv->tv_usec / tick;
1664 		if (val == 0 && tv->tv_usec != 0)
1665 			val = 1;
1666 
1667 		switch (optname) {
1668 
1669 		case SO_SNDTIMEO:
1670 			so->so_snd.sb_timeo = val;
1671 			break;
1672 		case SO_RCVTIMEO:
1673 			so->so_rcv.sb_timeo = val;
1674 			break;
1675 		}
1676 		break;
1677 
1678 	default:
1679 		return ENOPROTOOPT;
1680 	}
1681 	return 0;
1682 }
1683 
1684 int
1685 sosetopt(struct socket *so, int level, int optname, struct mbuf *m)
1686 {
1687 	int error, prerr;
1688 
1689 	solock(so);
1690 	if (level == SOL_SOCKET)
1691 		error = sosetopt1(so, level, optname, m);
1692 	else
1693 		error = ENOPROTOOPT;
1694 
1695 	if ((error == 0 || error == ENOPROTOOPT) &&
1696 	    so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) {
1697 		/* give the protocol stack a shot */
1698 		prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, level,
1699 		    optname, &m);
1700 		if (prerr == 0)
1701 			error = 0;
1702 		else if (prerr != ENOPROTOOPT)
1703 			error = prerr;
1704 	} else if (m != NULL)
1705 		(void)m_free(m);
1706 	sounlock(so);
1707 	return error;
1708 }
1709 
1710 int
1711 sogetopt(struct socket *so, int level, int optname, struct mbuf **mp)
1712 {
1713 	struct mbuf	*m;
1714 	int		error;
1715 
1716 	solock(so);
1717 	if (level != SOL_SOCKET) {
1718 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1719 			error = ((*so->so_proto->pr_ctloutput)
1720 				  (PRCO_GETOPT, so, level, optname, mp));
1721 		} else
1722 			error = (ENOPROTOOPT);
1723 	} else {
1724 		m = m_get(M_WAIT, MT_SOOPTS);
1725 		m->m_len = sizeof(int);
1726 
1727 		switch (optname) {
1728 
1729 		case SO_LINGER:
1730 			m->m_len = sizeof(struct linger);
1731 			mtod(m, struct linger *)->l_onoff =
1732 			    (so->so_options & SO_LINGER) ? 1 : 0;
1733 			mtod(m, struct linger *)->l_linger = so->so_linger;
1734 			break;
1735 
1736 		case SO_USELOOPBACK:
1737 		case SO_DONTROUTE:
1738 		case SO_DEBUG:
1739 		case SO_KEEPALIVE:
1740 		case SO_REUSEADDR:
1741 		case SO_REUSEPORT:
1742 		case SO_BROADCAST:
1743 		case SO_OOBINLINE:
1744 		case SO_TIMESTAMP:
1745 			*mtod(m, int *) = (so->so_options & optname) ? 1 : 0;
1746 			break;
1747 
1748 		case SO_TYPE:
1749 			*mtod(m, int *) = so->so_type;
1750 			break;
1751 
1752 		case SO_ERROR:
1753 			*mtod(m, int *) = so->so_error;
1754 			so->so_error = 0;
1755 			break;
1756 
1757 		case SO_SNDBUF:
1758 			*mtod(m, int *) = so->so_snd.sb_hiwat;
1759 			break;
1760 
1761 		case SO_RCVBUF:
1762 			*mtod(m, int *) = so->so_rcv.sb_hiwat;
1763 			break;
1764 
1765 		case SO_SNDLOWAT:
1766 			*mtod(m, int *) = so->so_snd.sb_lowat;
1767 			break;
1768 
1769 		case SO_RCVLOWAT:
1770 			*mtod(m, int *) = so->so_rcv.sb_lowat;
1771 			break;
1772 
1773 		case SO_SNDTIMEO:
1774 		case SO_RCVTIMEO:
1775 		    {
1776 			int val = (optname == SO_SNDTIMEO ?
1777 			     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1778 
1779 			m->m_len = sizeof(struct timeval);
1780 			mtod(m, struct timeval *)->tv_sec = val / hz;
1781 			mtod(m, struct timeval *)->tv_usec =
1782 			    (val % hz) * tick;
1783 			break;
1784 		    }
1785 
1786 		case SO_OVERFLOWED:
1787 			*mtod(m, int *) = so->so_rcv.sb_overflowed;
1788 			break;
1789 
1790 		default:
1791 			sounlock(so);
1792 			(void)m_free(m);
1793 			return (ENOPROTOOPT);
1794 		}
1795 		*mp = m;
1796 		error = 0;
1797 	}
1798 
1799 	sounlock(so);
1800 	return (error);
1801 }
1802 
1803 void
1804 sohasoutofband(struct socket *so)
1805 {
1806 
1807 	fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
1808 	selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, 0);
1809 }
1810 
1811 static void
1812 filt_sordetach(struct knote *kn)
1813 {
1814 	struct socket	*so;
1815 
1816 	so = ((file_t *)kn->kn_obj)->f_data;
1817 	solock(so);
1818 	SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
1819 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
1820 		so->so_rcv.sb_flags &= ~SB_KNOTE;
1821 	sounlock(so);
1822 }
1823 
1824 /*ARGSUSED*/
1825 static int
1826 filt_soread(struct knote *kn, long hint)
1827 {
1828 	struct socket	*so;
1829 	int rv;
1830 
1831 	so = ((file_t *)kn->kn_obj)->f_data;
1832 	if (hint != NOTE_SUBMIT)
1833 		solock(so);
1834 	kn->kn_data = so->so_rcv.sb_cc;
1835 	if (so->so_state & SS_CANTRCVMORE) {
1836 		kn->kn_flags |= EV_EOF;
1837 		kn->kn_fflags = so->so_error;
1838 		rv = 1;
1839 	} else if (so->so_error)	/* temporary udp error */
1840 		rv = 1;
1841 	else if (kn->kn_sfflags & NOTE_LOWAT)
1842 		rv = (kn->kn_data >= kn->kn_sdata);
1843 	else
1844 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
1845 	if (hint != NOTE_SUBMIT)
1846 		sounlock(so);
1847 	return rv;
1848 }
1849 
1850 static void
1851 filt_sowdetach(struct knote *kn)
1852 {
1853 	struct socket	*so;
1854 
1855 	so = ((file_t *)kn->kn_obj)->f_data;
1856 	solock(so);
1857 	SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
1858 	if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
1859 		so->so_snd.sb_flags &= ~SB_KNOTE;
1860 	sounlock(so);
1861 }
1862 
1863 /*ARGSUSED*/
1864 static int
1865 filt_sowrite(struct knote *kn, long hint)
1866 {
1867 	struct socket	*so;
1868 	int rv;
1869 
1870 	so = ((file_t *)kn->kn_obj)->f_data;
1871 	if (hint != NOTE_SUBMIT)
1872 		solock(so);
1873 	kn->kn_data = sbspace(&so->so_snd);
1874 	if (so->so_state & SS_CANTSENDMORE) {
1875 		kn->kn_flags |= EV_EOF;
1876 		kn->kn_fflags = so->so_error;
1877 		rv = 1;
1878 	} else if (so->so_error)	/* temporary udp error */
1879 		rv = 1;
1880 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
1881 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
1882 		rv = 0;
1883 	else if (kn->kn_sfflags & NOTE_LOWAT)
1884 		rv = (kn->kn_data >= kn->kn_sdata);
1885 	else
1886 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
1887 	if (hint != NOTE_SUBMIT)
1888 		sounlock(so);
1889 	return rv;
1890 }
1891 
1892 /*ARGSUSED*/
1893 static int
1894 filt_solisten(struct knote *kn, long hint)
1895 {
1896 	struct socket	*so;
1897 	int rv;
1898 
1899 	so = ((file_t *)kn->kn_obj)->f_data;
1900 
1901 	/*
1902 	 * Set kn_data to number of incoming connections, not
1903 	 * counting partial (incomplete) connections.
1904 	 */
1905 	if (hint != NOTE_SUBMIT)
1906 		solock(so);
1907 	kn->kn_data = so->so_qlen;
1908 	rv = (kn->kn_data > 0);
1909 	if (hint != NOTE_SUBMIT)
1910 		sounlock(so);
1911 	return rv;
1912 }
1913 
1914 static const struct filterops solisten_filtops =
1915 	{ 1, NULL, filt_sordetach, filt_solisten };
1916 static const struct filterops soread_filtops =
1917 	{ 1, NULL, filt_sordetach, filt_soread };
1918 static const struct filterops sowrite_filtops =
1919 	{ 1, NULL, filt_sowdetach, filt_sowrite };
1920 
1921 int
1922 soo_kqfilter(struct file *fp, struct knote *kn)
1923 {
1924 	struct socket	*so;
1925 	struct sockbuf	*sb;
1926 
1927 	so = ((file_t *)kn->kn_obj)->f_data;
1928 	solock(so);
1929 	switch (kn->kn_filter) {
1930 	case EVFILT_READ:
1931 		if (so->so_options & SO_ACCEPTCONN)
1932 			kn->kn_fop = &solisten_filtops;
1933 		else
1934 			kn->kn_fop = &soread_filtops;
1935 		sb = &so->so_rcv;
1936 		break;
1937 	case EVFILT_WRITE:
1938 		kn->kn_fop = &sowrite_filtops;
1939 		sb = &so->so_snd;
1940 		break;
1941 	default:
1942 		sounlock(so);
1943 		return (EINVAL);
1944 	}
1945 	SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
1946 	sb->sb_flags |= SB_KNOTE;
1947 	sounlock(so);
1948 	return (0);
1949 }
1950 
1951 static int
1952 sodopoll(struct socket *so, int events)
1953 {
1954 	int revents;
1955 
1956 	revents = 0;
1957 
1958 	if (events & (POLLIN | POLLRDNORM))
1959 		if (soreadable(so))
1960 			revents |= events & (POLLIN | POLLRDNORM);
1961 
1962 	if (events & (POLLOUT | POLLWRNORM))
1963 		if (sowritable(so))
1964 			revents |= events & (POLLOUT | POLLWRNORM);
1965 
1966 	if (events & (POLLPRI | POLLRDBAND))
1967 		if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
1968 			revents |= events & (POLLPRI | POLLRDBAND);
1969 
1970 	return revents;
1971 }
1972 
1973 int
1974 sopoll(struct socket *so, int events)
1975 {
1976 	int revents = 0;
1977 
1978 #ifndef DIAGNOSTIC
1979 	/*
1980 	 * Do a quick, unlocked check in expectation that the socket
1981 	 * will be ready for I/O.  Don't do this check if DIAGNOSTIC,
1982 	 * as the solocked() assertions will fail.
1983 	 */
1984 	if ((revents = sodopoll(so, events)) != 0)
1985 		return revents;
1986 #endif
1987 
1988 	solock(so);
1989 	if ((revents = sodopoll(so, events)) == 0) {
1990 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
1991 			selrecord(curlwp, &so->so_rcv.sb_sel);
1992 			so->so_rcv.sb_flags |= SB_NOTIFY;
1993 		}
1994 
1995 		if (events & (POLLOUT | POLLWRNORM)) {
1996 			selrecord(curlwp, &so->so_snd.sb_sel);
1997 			so->so_snd.sb_flags |= SB_NOTIFY;
1998 		}
1999 	}
2000 	sounlock(so);
2001 
2002 	return revents;
2003 }
2004 
2005 
2006 #include <sys/sysctl.h>
2007 
2008 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO);
2009 
2010 /*
2011  * sysctl helper routine for kern.somaxkva.  ensures that the given
2012  * value is not too small.
2013  * (XXX should we maybe make sure it's not too large as well?)
2014  */
2015 static int
2016 sysctl_kern_somaxkva(SYSCTLFN_ARGS)
2017 {
2018 	int error, new_somaxkva;
2019 	struct sysctlnode node;
2020 
2021 	new_somaxkva = somaxkva;
2022 	node = *rnode;
2023 	node.sysctl_data = &new_somaxkva;
2024 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2025 	if (error || newp == NULL)
2026 		return (error);
2027 
2028 	if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */
2029 		return (EINVAL);
2030 
2031 	mutex_enter(&so_pendfree_lock);
2032 	somaxkva = new_somaxkva;
2033 	cv_broadcast(&socurkva_cv);
2034 	mutex_exit(&so_pendfree_lock);
2035 
2036 	return (error);
2037 }
2038 
2039 SYSCTL_SETUP(sysctl_kern_somaxkva_setup, "sysctl kern.somaxkva setup")
2040 {
2041 
2042 	sysctl_createv(clog, 0, NULL, NULL,
2043 		       CTLFLAG_PERMANENT,
2044 		       CTLTYPE_NODE, "kern", NULL,
2045 		       NULL, 0, NULL, 0,
2046 		       CTL_KERN, CTL_EOL);
2047 
2048 	sysctl_createv(clog, 0, NULL, NULL,
2049 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2050 		       CTLTYPE_INT, "somaxkva",
2051 		       SYSCTL_DESCR("Maximum amount of kernel memory to be "
2052 				    "used for socket buffers"),
2053 		       sysctl_kern_somaxkva, 0, NULL, 0,
2054 		       CTL_KERN, KERN_SOMAXKVA, CTL_EOL);
2055 }
2056