xref: /netbsd-src/sys/kern/uipc_socket.c (revision 11a6dbe72840351315e0652b2fc6663628c84cad)
1 /*	$NetBSD: uipc_socket.c,v 1.163 2008/04/29 17:35:31 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 1.163 2008/04/29 17:35:31 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 
73 #include <sys/param.h>
74 #include <sys/systm.h>
75 #include <sys/proc.h>
76 #include <sys/file.h>
77 #include <sys/filedesc.h>
78 #include <sys/malloc.h>
79 #include <sys/mbuf.h>
80 #include <sys/domain.h>
81 #include <sys/kernel.h>
82 #include <sys/protosw.h>
83 #include <sys/socket.h>
84 #include <sys/socketvar.h>
85 #include <sys/signalvar.h>
86 #include <sys/resourcevar.h>
87 #include <sys/event.h>
88 #include <sys/poll.h>
89 #include <sys/kauth.h>
90 #include <sys/mutex.h>
91 #include <sys/condvar.h>
92 
93 #include <uvm/uvm.h>
94 
95 MALLOC_DEFINE(M_SOOPTS, "soopts", "socket options");
96 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
97 
98 extern const struct fileops socketops;
99 
100 extern int	somaxconn;			/* patchable (XXX sysctl) */
101 int		somaxconn = SOMAXCONN;
102 kmutex_t	*softnet_lock;
103 
104 #ifdef SOSEND_COUNTERS
105 #include <sys/device.h>
106 
107 static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
108     NULL, "sosend", "loan big");
109 static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
110     NULL, "sosend", "copy big");
111 static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
112     NULL, "sosend", "copy small");
113 static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
114     NULL, "sosend", "kva limit");
115 
116 #define	SOSEND_COUNTER_INCR(ev)		(ev)->ev_count++
117 
118 EVCNT_ATTACH_STATIC(sosend_loan_big);
119 EVCNT_ATTACH_STATIC(sosend_copy_big);
120 EVCNT_ATTACH_STATIC(sosend_copy_small);
121 EVCNT_ATTACH_STATIC(sosend_kvalimit);
122 #else
123 
124 #define	SOSEND_COUNTER_INCR(ev)		/* nothing */
125 
126 #endif /* SOSEND_COUNTERS */
127 
128 static struct callback_entry sokva_reclaimerentry;
129 
130 #ifdef SOSEND_NO_LOAN
131 int sock_loan_thresh = -1;
132 #else
133 int sock_loan_thresh = 4096;
134 #endif
135 
136 static kmutex_t so_pendfree_lock;
137 static struct mbuf *so_pendfree;
138 
139 #ifndef SOMAXKVA
140 #define	SOMAXKVA (16 * 1024 * 1024)
141 #endif
142 int somaxkva = SOMAXKVA;
143 static int socurkva;
144 static kcondvar_t socurkva_cv;
145 
146 #define	SOCK_LOAN_CHUNK		65536
147 
148 static size_t sodopendfree(void);
149 static size_t sodopendfreel(void);
150 
151 static vsize_t
152 sokvareserve(struct socket *so, vsize_t len)
153 {
154 	int error;
155 
156 	mutex_enter(&so_pendfree_lock);
157 	while (socurkva + len > somaxkva) {
158 		size_t freed;
159 
160 		/*
161 		 * try to do pendfree.
162 		 */
163 
164 		freed = sodopendfreel();
165 
166 		/*
167 		 * if some kva was freed, try again.
168 		 */
169 
170 		if (freed)
171 			continue;
172 
173 		SOSEND_COUNTER_INCR(&sosend_kvalimit);
174 		error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock);
175 		if (error) {
176 			len = 0;
177 			break;
178 		}
179 	}
180 	socurkva += len;
181 	mutex_exit(&so_pendfree_lock);
182 	return len;
183 }
184 
185 static void
186 sokvaunreserve(vsize_t len)
187 {
188 
189 	mutex_enter(&so_pendfree_lock);
190 	socurkva -= len;
191 	cv_broadcast(&socurkva_cv);
192 	mutex_exit(&so_pendfree_lock);
193 }
194 
195 /*
196  * sokvaalloc: allocate kva for loan.
197  */
198 
199 vaddr_t
200 sokvaalloc(vsize_t len, struct socket *so)
201 {
202 	vaddr_t lva;
203 
204 	/*
205 	 * reserve kva.
206 	 */
207 
208 	if (sokvareserve(so, len) == 0)
209 		return 0;
210 
211 	/*
212 	 * allocate kva.
213 	 */
214 
215 	lva = uvm_km_alloc(kernel_map, len, 0, UVM_KMF_VAONLY | UVM_KMF_WAITVA);
216 	if (lva == 0) {
217 		sokvaunreserve(len);
218 		return (0);
219 	}
220 
221 	return lva;
222 }
223 
224 /*
225  * sokvafree: free kva for loan.
226  */
227 
228 void
229 sokvafree(vaddr_t sva, vsize_t len)
230 {
231 
232 	/*
233 	 * free kva.
234 	 */
235 
236 	uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY);
237 
238 	/*
239 	 * unreserve kva.
240 	 */
241 
242 	sokvaunreserve(len);
243 }
244 
245 static void
246 sodoloanfree(struct vm_page **pgs, void *buf, size_t size)
247 {
248 	vaddr_t sva, eva;
249 	vsize_t len;
250 	int npgs;
251 
252 	KASSERT(pgs != NULL);
253 
254 	eva = round_page((vaddr_t) buf + size);
255 	sva = trunc_page((vaddr_t) buf);
256 	len = eva - sva;
257 	npgs = len >> PAGE_SHIFT;
258 
259 	pmap_kremove(sva, len);
260 	pmap_update(pmap_kernel());
261 	uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE);
262 	sokvafree(sva, len);
263 }
264 
265 static size_t
266 sodopendfree(void)
267 {
268 	size_t rv;
269 
270 	if (__predict_true(so_pendfree == NULL))
271 		return 0;
272 
273 	mutex_enter(&so_pendfree_lock);
274 	rv = sodopendfreel();
275 	mutex_exit(&so_pendfree_lock);
276 
277 	return rv;
278 }
279 
280 /*
281  * sodopendfreel: free mbufs on "pendfree" list.
282  * unlock and relock so_pendfree_lock when freeing mbufs.
283  *
284  * => called with so_pendfree_lock held.
285  */
286 
287 static size_t
288 sodopendfreel(void)
289 {
290 	struct mbuf *m, *next;
291 	size_t rv = 0;
292 
293 	KASSERT(mutex_owned(&so_pendfree_lock));
294 
295 	while (so_pendfree != NULL) {
296 		m = so_pendfree;
297 		so_pendfree = NULL;
298 		mutex_exit(&so_pendfree_lock);
299 
300 		for (; m != NULL; m = next) {
301 			next = m->m_next;
302 			KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0);
303 			KASSERT(m->m_ext.ext_refcnt == 0);
304 
305 			rv += m->m_ext.ext_size;
306 			sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf,
307 			    m->m_ext.ext_size);
308 			pool_cache_put(mb_cache, m);
309 		}
310 
311 		mutex_enter(&so_pendfree_lock);
312 	}
313 
314 	return (rv);
315 }
316 
317 void
318 soloanfree(struct mbuf *m, void *buf, size_t size, void *arg)
319 {
320 
321 	KASSERT(m != NULL);
322 
323 	/*
324 	 * postpone freeing mbuf.
325 	 *
326 	 * we can't do it in interrupt context
327 	 * because we need to put kva back to kernel_map.
328 	 */
329 
330 	mutex_enter(&so_pendfree_lock);
331 	m->m_next = so_pendfree;
332 	so_pendfree = m;
333 	cv_broadcast(&socurkva_cv);
334 	mutex_exit(&so_pendfree_lock);
335 }
336 
337 static long
338 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space)
339 {
340 	struct iovec *iov = uio->uio_iov;
341 	vaddr_t sva, eva;
342 	vsize_t len;
343 	vaddr_t lva;
344 	int npgs, error;
345 	vaddr_t va;
346 	int i;
347 
348 	if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace))
349 		return (0);
350 
351 	if (iov->iov_len < (size_t) space)
352 		space = iov->iov_len;
353 	if (space > SOCK_LOAN_CHUNK)
354 		space = SOCK_LOAN_CHUNK;
355 
356 	eva = round_page((vaddr_t) iov->iov_base + space);
357 	sva = trunc_page((vaddr_t) iov->iov_base);
358 	len = eva - sva;
359 	npgs = len >> PAGE_SHIFT;
360 
361 	KASSERT(npgs <= M_EXT_MAXPAGES);
362 
363 	lva = sokvaalloc(len, so);
364 	if (lva == 0)
365 		return 0;
366 
367 	error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len,
368 	    m->m_ext.ext_pgs, UVM_LOAN_TOPAGE);
369 	if (error) {
370 		sokvafree(lva, len);
371 		return (0);
372 	}
373 
374 	for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE)
375 		pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
376 		    VM_PROT_READ);
377 	pmap_update(pmap_kernel());
378 
379 	lva += (vaddr_t) iov->iov_base & PAGE_MASK;
380 
381 	MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so);
382 	m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
383 
384 	uio->uio_resid -= space;
385 	/* uio_offset not updated, not set/used for write(2) */
386 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space;
387 	uio->uio_iov->iov_len -= space;
388 	if (uio->uio_iov->iov_len == 0) {
389 		uio->uio_iov++;
390 		uio->uio_iovcnt--;
391 	}
392 
393 	return (space);
394 }
395 
396 static int
397 sokva_reclaim_callback(struct callback_entry *ce, void *obj, void *arg)
398 {
399 
400 	KASSERT(ce == &sokva_reclaimerentry);
401 	KASSERT(obj == NULL);
402 
403 	sodopendfree();
404 	if (!vm_map_starved_p(kernel_map)) {
405 		return CALLBACK_CHAIN_ABORT;
406 	}
407 	return CALLBACK_CHAIN_CONTINUE;
408 }
409 
410 struct mbuf *
411 getsombuf(struct socket *so, int type)
412 {
413 	struct mbuf *m;
414 
415 	m = m_get(M_WAIT, type);
416 	MCLAIM(m, so->so_mowner);
417 	return m;
418 }
419 
420 struct mbuf *
421 m_intopt(struct socket *so, int val)
422 {
423 	struct mbuf *m;
424 
425 	m = getsombuf(so, MT_SOOPTS);
426 	m->m_len = sizeof(int);
427 	*mtod(m, int *) = val;
428 	return m;
429 }
430 
431 void
432 soinit(void)
433 {
434 
435 	mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM);
436 	softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
437 	cv_init(&socurkva_cv, "sokva");
438 
439 	/* Set the initial adjusted socket buffer size. */
440 	if (sb_max_set(sb_max))
441 		panic("bad initial sb_max value: %lu", sb_max);
442 
443 	callback_register(&vm_map_to_kernel(kernel_map)->vmk_reclaim_callback,
444 	    &sokva_reclaimerentry, NULL, sokva_reclaim_callback);
445 }
446 
447 /*
448  * Socket operation routines.
449  * These routines are called by the routines in
450  * sys_socket.c or from a system process, and
451  * implement the semantics of socket operations by
452  * switching out to the protocol specific routines.
453  */
454 /*ARGSUSED*/
455 int
456 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l,
457 	 struct socket *lockso)
458 {
459 	const struct protosw	*prp;
460 	struct socket	*so;
461 	uid_t		uid;
462 	int		error;
463 	kmutex_t	*lock;
464 
465 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET,
466 	    KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type),
467 	    KAUTH_ARG(proto));
468 	if (error != 0)
469 		return error;
470 
471 	if (proto)
472 		prp = pffindproto(dom, proto, type);
473 	else
474 		prp = pffindtype(dom, type);
475 	if (prp == NULL) {
476 		/* no support for domain */
477 		if (pffinddomain(dom) == 0)
478 			return EAFNOSUPPORT;
479 		/* no support for socket type */
480 		if (proto == 0 && type != 0)
481 			return EPROTOTYPE;
482 		return EPROTONOSUPPORT;
483 	}
484 	if (prp->pr_usrreq == NULL)
485 		return EPROTONOSUPPORT;
486 	if (prp->pr_type != type)
487 		return EPROTOTYPE;
488 
489 	so = soget(true);
490 	so->so_type = type;
491 	so->so_proto = prp;
492 	so->so_send = sosend;
493 	so->so_receive = soreceive;
494 #ifdef MBUFTRACE
495 	so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner;
496 	so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner;
497 	so->so_mowner = &prp->pr_domain->dom_mowner;
498 #endif
499 	uid = kauth_cred_geteuid(l->l_cred);
500 	so->so_uidinfo = uid_find(uid);
501 	if (lockso != NULL) {
502 		/* Caller wants us to share a lock. */
503 		lock = lockso->so_lock;
504 		so->so_lock = lock;
505 		mutex_obj_hold(lock);
506 		mutex_enter(lock);
507 	} else {
508 		/* Lock assigned and taken during PRU_ATTACH. */
509 	}
510 	error = (*prp->pr_usrreq)(so, PRU_ATTACH, NULL,
511 	    (struct mbuf *)(long)proto, NULL, l);
512 	KASSERT(solocked(so));
513 	if (error != 0) {
514 		so->so_state |= SS_NOFDREF;
515 		sofree(so);
516 		return error;
517 	}
518 	sounlock(so);
519 	*aso = so;
520 	return 0;
521 }
522 
523 /* On success, write file descriptor to fdout and return zero.  On
524  * failure, return non-zero; *fdout will be undefined.
525  */
526 int
527 fsocreate(int domain, struct socket **sop, int type, int protocol,
528     struct lwp *l, int *fdout)
529 {
530 	struct socket	*so;
531 	struct file	*fp;
532 	int		fd, error;
533 
534 	if ((error = fd_allocfile(&fp, &fd)) != 0)
535 		return (error);
536 	fp->f_flag = FREAD|FWRITE;
537 	fp->f_type = DTYPE_SOCKET;
538 	fp->f_ops = &socketops;
539 	error = socreate(domain, &so, type, protocol, l, NULL);
540 	if (error != 0) {
541 		fd_abort(curproc, fp, fd);
542 	} else {
543 		if (sop != NULL)
544 			*sop = so;
545 		fp->f_data = so;
546 		fd_affix(curproc, fp, fd);
547 		*fdout = fd;
548 	}
549 	return error;
550 }
551 
552 int
553 sobind(struct socket *so, struct mbuf *nam, struct lwp *l)
554 {
555 	int	error;
556 
557 	solock(so);
558 	error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, NULL, nam, NULL, l);
559 	sounlock(so);
560 	return error;
561 }
562 
563 int
564 solisten(struct socket *so, int backlog, struct lwp *l)
565 {
566 	int	error;
567 
568 	solock(so);
569 	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
570 	    SS_ISDISCONNECTING)) != 0) {
571 	    	sounlock(so);
572 		return (EOPNOTSUPP);
573 	}
574 	error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, NULL,
575 	    NULL, NULL, l);
576 	if (error != 0) {
577 		sounlock(so);
578 		return error;
579 	}
580 	if (TAILQ_EMPTY(&so->so_q))
581 		so->so_options |= SO_ACCEPTCONN;
582 	if (backlog < 0)
583 		backlog = 0;
584 	so->so_qlimit = min(backlog, somaxconn);
585 	sounlock(so);
586 	return 0;
587 }
588 
589 void
590 sofree(struct socket *so)
591 {
592 	u_int refs;
593 
594 	KASSERT(solocked(so));
595 
596 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
597 		sounlock(so);
598 		return;
599 	}
600 	if (so->so_head) {
601 		/*
602 		 * We must not decommission a socket that's on the accept(2)
603 		 * queue.  If we do, then accept(2) may hang after select(2)
604 		 * indicated that the listening socket was ready.
605 		 */
606 		if (!soqremque(so, 0)) {
607 			sounlock(so);
608 			return;
609 		}
610 	}
611 	if (so->so_rcv.sb_hiwat)
612 		(void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0,
613 		    RLIM_INFINITY);
614 	if (so->so_snd.sb_hiwat)
615 		(void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0,
616 		    RLIM_INFINITY);
617 	sbrelease(&so->so_snd, so);
618 	KASSERT(!cv_has_waiters(&so->so_cv));
619 	KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv));
620 	KASSERT(!cv_has_waiters(&so->so_snd.sb_cv));
621 	sorflush(so);
622 	refs = so->so_aborting;	/* XXX */
623 	sounlock(so);
624 	if (refs == 0)		/* XXX */
625 		soput(so);
626 }
627 
628 /*
629  * Close a socket on last file table reference removal.
630  * Initiate disconnect if connected.
631  * Free socket when disconnect complete.
632  */
633 int
634 soclose(struct socket *so)
635 {
636 	struct socket	*so2;
637 	int		error;
638 	int		error2;
639 
640 	error = 0;
641 	solock(so);
642 	if (so->so_options & SO_ACCEPTCONN) {
643 		do {
644 			if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) {
645 				KASSERT(solocked2(so, so2));
646 				(void) soqremque(so2, 0);
647 				/* soabort drops the lock. */
648 				(void) soabort(so2);
649 				solock(so);
650 				continue;
651 			}
652 			if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) {
653 				KASSERT(solocked2(so, so2));
654 				(void) soqremque(so2, 1);
655 				/* soabort drops the lock. */
656 				(void) soabort(so2);
657 				solock(so);
658 				continue;
659 			}
660 		} while (0);
661 	}
662 	if (so->so_pcb == 0)
663 		goto discard;
664 	if (so->so_state & SS_ISCONNECTED) {
665 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
666 			error = sodisconnect(so);
667 			if (error)
668 				goto drop;
669 		}
670 		if (so->so_options & SO_LINGER) {
671 			if ((so->so_state & SS_ISDISCONNECTING) && so->so_nbio)
672 				goto drop;
673 			while (so->so_state & SS_ISCONNECTED) {
674 				error = sowait(so, so->so_linger * hz);
675 				if (error)
676 					break;
677 			}
678 		}
679 	}
680  drop:
681 	if (so->so_pcb) {
682 		error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH,
683 		    NULL, NULL, NULL, NULL);
684 		if (error == 0)
685 			error = error2;
686 	}
687  discard:
688 	if (so->so_state & SS_NOFDREF)
689 		panic("soclose: NOFDREF");
690 	so->so_state |= SS_NOFDREF;
691 	sofree(so);
692 	return (error);
693 }
694 
695 /*
696  * Must be called with the socket locked..  Will return with it unlocked.
697  */
698 int
699 soabort(struct socket *so)
700 {
701 	u_int refs;
702 	int error;
703 
704 	KASSERT(solocked(so));
705 	KASSERT(so->so_head == NULL);
706 
707 	so->so_aborting++;		/* XXX */
708 	error = (*so->so_proto->pr_usrreq)(so, PRU_ABORT, NULL,
709 	    NULL, NULL, NULL);
710 	refs = --so->so_aborting;	/* XXX */
711 	if (error) {
712 		sofree(so);
713 	} else {
714 		sounlock(so);
715 		if (refs == 0)
716 			sofree(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 	if (l)
856 		l->l_ru.ru_msgsnd++;
857 	if (control)
858 		clen = control->m_len;
859  restart:
860 	if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
861 		goto out;
862 	do {
863 		if (so->so_state & SS_CANTSENDMORE) {
864 			error = EPIPE;
865 			goto release;
866 		}
867 		if (so->so_error) {
868 			error = so->so_error;
869 			so->so_error = 0;
870 			goto release;
871 		}
872 		if ((so->so_state & SS_ISCONNECTED) == 0) {
873 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
874 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
875 				    !(resid == 0 && clen != 0)) {
876 					error = ENOTCONN;
877 					goto release;
878 				}
879 			} else if (addr == 0) {
880 				error = EDESTADDRREQ;
881 				goto release;
882 			}
883 		}
884 		space = sbspace(&so->so_snd);
885 		if (flags & MSG_OOB)
886 			space += 1024;
887 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
888 		    clen > so->so_snd.sb_hiwat) {
889 			error = EMSGSIZE;
890 			goto release;
891 		}
892 		if (space < resid + clen &&
893 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
894 			if (so->so_nbio) {
895 				error = EWOULDBLOCK;
896 				goto release;
897 			}
898 			sbunlock(&so->so_snd);
899 			error = sbwait(&so->so_snd);
900 			if (error)
901 				goto out;
902 			goto restart;
903 		}
904 		mp = &top;
905 		space -= clen;
906 		do {
907 			if (uio == NULL) {
908 				/*
909 				 * Data is prepackaged in "top".
910 				 */
911 				resid = 0;
912 				if (flags & MSG_EOR)
913 					top->m_flags |= M_EOR;
914 			} else do {
915 				sounlock(so);
916 				splx(s);
917 				if (top == NULL) {
918 					m = m_gethdr(M_WAIT, MT_DATA);
919 					mlen = MHLEN;
920 					m->m_pkthdr.len = 0;
921 					m->m_pkthdr.rcvif = NULL;
922 				} else {
923 					m = m_get(M_WAIT, MT_DATA);
924 					mlen = MLEN;
925 				}
926 				MCLAIM(m, so->so_snd.sb_mowner);
927 				if (sock_loan_thresh >= 0 &&
928 				    uio->uio_iov->iov_len >= sock_loan_thresh &&
929 				    space >= sock_loan_thresh &&
930 				    (len = sosend_loan(so, uio, m,
931 						       space)) != 0) {
932 					SOSEND_COUNTER_INCR(&sosend_loan_big);
933 					space -= len;
934 					goto have_data;
935 				}
936 				if (resid >= MINCLSIZE && space >= MCLBYTES) {
937 					SOSEND_COUNTER_INCR(&sosend_copy_big);
938 					m_clget(m, M_WAIT);
939 					if ((m->m_flags & M_EXT) == 0)
940 						goto nopages;
941 					mlen = MCLBYTES;
942 					if (atomic && top == 0) {
943 						len = lmin(MCLBYTES - max_hdr,
944 						    resid);
945 						m->m_data += max_hdr;
946 					} else
947 						len = lmin(MCLBYTES, resid);
948 					space -= len;
949 				} else {
950  nopages:
951 					SOSEND_COUNTER_INCR(&sosend_copy_small);
952 					len = lmin(lmin(mlen, resid), space);
953 					space -= len;
954 					/*
955 					 * For datagram protocols, leave room
956 					 * for protocol headers in first mbuf.
957 					 */
958 					if (atomic && top == 0 && len < mlen)
959 						MH_ALIGN(m, len);
960 				}
961 				error = uiomove(mtod(m, void *), (int)len, uio);
962  have_data:
963 				resid = uio->uio_resid;
964 				m->m_len = len;
965 				*mp = m;
966 				top->m_pkthdr.len += len;
967 				s = splsoftnet();
968 				solock(so);
969 				if (error != 0)
970 					goto release;
971 				mp = &m->m_next;
972 				if (resid <= 0) {
973 					if (flags & MSG_EOR)
974 						top->m_flags |= M_EOR;
975 					break;
976 				}
977 			} while (space > 0 && atomic);
978 
979 			if (so->so_state & SS_CANTSENDMORE) {
980 				error = EPIPE;
981 				goto release;
982 			}
983 			if (dontroute)
984 				so->so_options |= SO_DONTROUTE;
985 			if (resid > 0)
986 				so->so_state |= SS_MORETOCOME;
987 			error = (*so->so_proto->pr_usrreq)(so,
988 			    (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND,
989 			    top, addr, control, curlwp);
990 			if (dontroute)
991 				so->so_options &= ~SO_DONTROUTE;
992 			if (resid > 0)
993 				so->so_state &= ~SS_MORETOCOME;
994 			clen = 0;
995 			control = NULL;
996 			top = NULL;
997 			mp = &top;
998 			if (error != 0)
999 				goto release;
1000 		} while (resid && space > 0);
1001 	} while (resid);
1002 
1003  release:
1004 	sbunlock(&so->so_snd);
1005  out:
1006 	sounlock(so);
1007 	splx(s);
1008 	if (top)
1009 		m_freem(top);
1010 	if (control)
1011 		m_freem(control);
1012 	return (error);
1013 }
1014 
1015 /*
1016  * Following replacement or removal of the first mbuf on the first
1017  * mbuf chain of a socket buffer, push necessary state changes back
1018  * into the socket buffer so that other consumers see the values
1019  * consistently.  'nextrecord' is the callers locally stored value of
1020  * the original value of sb->sb_mb->m_nextpkt which must be restored
1021  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
1022  */
1023 static void
1024 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
1025 {
1026 
1027 	KASSERT(solocked(sb->sb_so));
1028 
1029 	/*
1030 	 * First, update for the new value of nextrecord.  If necessary,
1031 	 * make it the first record.
1032 	 */
1033 	if (sb->sb_mb != NULL)
1034 		sb->sb_mb->m_nextpkt = nextrecord;
1035 	else
1036 		sb->sb_mb = nextrecord;
1037 
1038         /*
1039          * Now update any dependent socket buffer fields to reflect
1040          * the new state.  This is an inline of SB_EMPTY_FIXUP, with
1041          * the addition of a second clause that takes care of the
1042          * case where sb_mb has been updated, but remains the last
1043          * record.
1044          */
1045         if (sb->sb_mb == NULL) {
1046                 sb->sb_mbtail = NULL;
1047                 sb->sb_lastrecord = NULL;
1048         } else if (sb->sb_mb->m_nextpkt == NULL)
1049                 sb->sb_lastrecord = sb->sb_mb;
1050 }
1051 
1052 /*
1053  * Implement receive operations on a socket.
1054  * We depend on the way that records are added to the sockbuf
1055  * by sbappend*.  In particular, each record (mbufs linked through m_next)
1056  * must begin with an address if the protocol so specifies,
1057  * followed by an optional mbuf or mbufs containing ancillary data,
1058  * and then zero or more mbufs of data.
1059  * In order to avoid blocking network interrupts for the entire time here,
1060  * we splx() while doing the actual copy to user space.
1061  * Although the sockbuf is locked, new data may still be appended,
1062  * and thus we must maintain consistency of the sockbuf during that time.
1063  *
1064  * The caller may receive the data as a single mbuf chain by supplying
1065  * an mbuf **mp0 for use in returning the chain.  The uio is then used
1066  * only for the count in uio_resid.
1067  */
1068 int
1069 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
1070 	struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1071 {
1072 	struct lwp *l = curlwp;
1073 	struct mbuf	*m, **mp, *mt;
1074 	int atomic, flags, len, error, s, offset, moff, type, orig_resid;
1075 	const struct protosw	*pr;
1076 	struct mbuf	*nextrecord;
1077 	int		mbuf_removed = 0;
1078 	const struct domain *dom;
1079 
1080 	pr = so->so_proto;
1081 	atomic = pr->pr_flags & PR_ATOMIC;
1082 	dom = pr->pr_domain;
1083 	mp = mp0;
1084 	type = 0;
1085 	orig_resid = uio->uio_resid;
1086 
1087 	if (paddr != NULL)
1088 		*paddr = NULL;
1089 	if (controlp != NULL)
1090 		*controlp = NULL;
1091 	if (flagsp != NULL)
1092 		flags = *flagsp &~ MSG_EOR;
1093 	else
1094 		flags = 0;
1095 
1096 	if ((flags & MSG_DONTWAIT) == 0)
1097 		sodopendfree();
1098 
1099 	if (flags & MSG_OOB) {
1100 		m = m_get(M_WAIT, MT_DATA);
1101 		solock(so);
1102 		error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m,
1103 		    (struct mbuf *)(long)(flags & MSG_PEEK), NULL, l);
1104 		sounlock(so);
1105 		if (error)
1106 			goto bad;
1107 		do {
1108 			error = uiomove(mtod(m, void *),
1109 			    (int) min(uio->uio_resid, m->m_len), uio);
1110 			m = m_free(m);
1111 		} while (uio->uio_resid > 0 && error == 0 && m);
1112  bad:
1113 		if (m != NULL)
1114 			m_freem(m);
1115 		return error;
1116 	}
1117 	if (mp != NULL)
1118 		*mp = NULL;
1119 
1120 	/*
1121 	 * solock() provides atomicity of access.  splsoftnet() prevents
1122 	 * protocol processing soft interrupts from interrupting us and
1123 	 * blocking (expensive).
1124 	 */
1125 	s = splsoftnet();
1126 	solock(so);
1127 	if (so->so_state & SS_ISCONFIRMING && uio->uio_resid)
1128 		(*pr->pr_usrreq)(so, PRU_RCVD, NULL, NULL, NULL, l);
1129 
1130  restart:
1131 	if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) {
1132 		sounlock(so);
1133 		splx(s);
1134 		return error;
1135 	}
1136 
1137 	m = so->so_rcv.sb_mb;
1138 	/*
1139 	 * If we have less data than requested, block awaiting more
1140 	 * (subject to any timeout) if:
1141 	 *   1. the current count is less than the low water mark,
1142 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1143 	 *	receive operation at once if we block (resid <= hiwat), or
1144 	 *   3. MSG_DONTWAIT is not set.
1145 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1146 	 * we have to do the receive in sections, and thus risk returning
1147 	 * a short count if a timeout or signal occurs after we start.
1148 	 */
1149 	if (m == NULL ||
1150 	    ((flags & MSG_DONTWAIT) == 0 &&
1151 	     so->so_rcv.sb_cc < uio->uio_resid &&
1152 	     (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1153 	      ((flags & MSG_WAITALL) &&
1154 	       uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1155 	     m->m_nextpkt == NULL && !atomic)) {
1156 #ifdef DIAGNOSTIC
1157 		if (m == NULL && so->so_rcv.sb_cc)
1158 			panic("receive 1");
1159 #endif
1160 		if (so->so_error) {
1161 			if (m != NULL)
1162 				goto dontblock;
1163 			error = so->so_error;
1164 			if ((flags & MSG_PEEK) == 0)
1165 				so->so_error = 0;
1166 			goto release;
1167 		}
1168 		if (so->so_state & SS_CANTRCVMORE) {
1169 			if (m != NULL)
1170 				goto dontblock;
1171 			else
1172 				goto release;
1173 		}
1174 		for (; m != NULL; m = m->m_next)
1175 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1176 				m = so->so_rcv.sb_mb;
1177 				goto dontblock;
1178 			}
1179 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1180 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1181 			error = ENOTCONN;
1182 			goto release;
1183 		}
1184 		if (uio->uio_resid == 0)
1185 			goto release;
1186 		if (so->so_nbio || (flags & MSG_DONTWAIT)) {
1187 			error = EWOULDBLOCK;
1188 			goto release;
1189 		}
1190 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
1191 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
1192 		sbunlock(&so->so_rcv);
1193 		error = sbwait(&so->so_rcv);
1194 		if (error != 0) {
1195 			sounlock(so);
1196 			splx(s);
1197 			return error;
1198 		}
1199 		goto restart;
1200 	}
1201  dontblock:
1202 	/*
1203 	 * On entry here, m points to the first record of the socket buffer.
1204 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1205 	 * pointer to the next record in the socket buffer.  We must keep the
1206 	 * various socket buffer pointers and local stack versions of the
1207 	 * pointers in sync, pushing out modifications before dropping the
1208 	 * socket lock, and re-reading them when picking it up.
1209 	 *
1210 	 * Otherwise, we will race with the network stack appending new data
1211 	 * or records onto the socket buffer by using inconsistent/stale
1212 	 * versions of the field, possibly resulting in socket buffer
1213 	 * corruption.
1214 	 *
1215 	 * By holding the high-level sblock(), we prevent simultaneous
1216 	 * readers from pulling off the front of the socket buffer.
1217 	 */
1218 	if (l != NULL)
1219 		l->l_ru.ru_msgrcv++;
1220 	KASSERT(m == so->so_rcv.sb_mb);
1221 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
1222 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
1223 	nextrecord = m->m_nextpkt;
1224 	if (pr->pr_flags & PR_ADDR) {
1225 #ifdef DIAGNOSTIC
1226 		if (m->m_type != MT_SONAME)
1227 			panic("receive 1a");
1228 #endif
1229 		orig_resid = 0;
1230 		if (flags & MSG_PEEK) {
1231 			if (paddr)
1232 				*paddr = m_copy(m, 0, m->m_len);
1233 			m = m->m_next;
1234 		} else {
1235 			sbfree(&so->so_rcv, m);
1236 			mbuf_removed = 1;
1237 			if (paddr != NULL) {
1238 				*paddr = m;
1239 				so->so_rcv.sb_mb = m->m_next;
1240 				m->m_next = NULL;
1241 				m = so->so_rcv.sb_mb;
1242 			} else {
1243 				MFREE(m, so->so_rcv.sb_mb);
1244 				m = so->so_rcv.sb_mb;
1245 			}
1246 			sbsync(&so->so_rcv, nextrecord);
1247 		}
1248 	}
1249 
1250 	/*
1251 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1252 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1253 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1254 	 * perform externalization (or freeing if controlp == NULL).
1255 	 */
1256 	if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) {
1257 		struct mbuf *cm = NULL, *cmn;
1258 		struct mbuf **cme = &cm;
1259 
1260 		do {
1261 			if (flags & MSG_PEEK) {
1262 				if (controlp != NULL) {
1263 					*controlp = m_copy(m, 0, m->m_len);
1264 					controlp = &(*controlp)->m_next;
1265 				}
1266 				m = m->m_next;
1267 			} else {
1268 				sbfree(&so->so_rcv, m);
1269 				so->so_rcv.sb_mb = m->m_next;
1270 				m->m_next = NULL;
1271 				*cme = m;
1272 				cme = &(*cme)->m_next;
1273 				m = so->so_rcv.sb_mb;
1274 			}
1275 		} while (m != NULL && m->m_type == MT_CONTROL);
1276 		if ((flags & MSG_PEEK) == 0)
1277 			sbsync(&so->so_rcv, nextrecord);
1278 		for (; cm != NULL; cm = cmn) {
1279 			cmn = cm->m_next;
1280 			cm->m_next = NULL;
1281 			type = mtod(cm, struct cmsghdr *)->cmsg_type;
1282 			if (controlp != NULL) {
1283 				if (dom->dom_externalize != NULL &&
1284 				    type == SCM_RIGHTS) {
1285 					sounlock(so);
1286 					splx(s);
1287 					error = (*dom->dom_externalize)(cm, l);
1288 					s = splsoftnet();
1289 					solock(so);
1290 				}
1291 				*controlp = cm;
1292 				while (*controlp != NULL)
1293 					controlp = &(*controlp)->m_next;
1294 			} else {
1295 				/*
1296 				 * Dispose of any SCM_RIGHTS message that went
1297 				 * through the read path rather than recv.
1298 				 */
1299 				if (dom->dom_dispose != NULL &&
1300 				    type == SCM_RIGHTS) {
1301 				    	sounlock(so);
1302 					(*dom->dom_dispose)(cm);
1303 					solock(so);
1304 				}
1305 				m_freem(cm);
1306 			}
1307 		}
1308 		if (m != NULL)
1309 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1310 		else
1311 			nextrecord = so->so_rcv.sb_mb;
1312 		orig_resid = 0;
1313 	}
1314 
1315 	/* If m is non-NULL, we have some data to read. */
1316 	if (__predict_true(m != NULL)) {
1317 		type = m->m_type;
1318 		if (type == MT_OOBDATA)
1319 			flags |= MSG_OOB;
1320 	}
1321 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1322 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1323 
1324 	moff = 0;
1325 	offset = 0;
1326 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1327 		if (m->m_type == MT_OOBDATA) {
1328 			if (type != MT_OOBDATA)
1329 				break;
1330 		} else if (type == MT_OOBDATA)
1331 			break;
1332 #ifdef DIAGNOSTIC
1333 		else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
1334 			panic("receive 3");
1335 #endif
1336 		so->so_state &= ~SS_RCVATMARK;
1337 		len = uio->uio_resid;
1338 		if (so->so_oobmark && len > so->so_oobmark - offset)
1339 			len = so->so_oobmark - offset;
1340 		if (len > m->m_len - moff)
1341 			len = m->m_len - moff;
1342 		/*
1343 		 * If mp is set, just pass back the mbufs.
1344 		 * Otherwise copy them out via the uio, then free.
1345 		 * Sockbuf must be consistent here (points to current mbuf,
1346 		 * it points to next record) when we drop priority;
1347 		 * we must note any additions to the sockbuf when we
1348 		 * block interrupts again.
1349 		 */
1350 		if (mp == NULL) {
1351 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1352 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1353 			sounlock(so);
1354 			splx(s);
1355 			error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1356 			s = splsoftnet();
1357 			solock(so);
1358 			if (error != 0) {
1359 				/*
1360 				 * If any part of the record has been removed
1361 				 * (such as the MT_SONAME mbuf, which will
1362 				 * happen when PR_ADDR, and thus also
1363 				 * PR_ATOMIC, is set), then drop the entire
1364 				 * record to maintain the atomicity of the
1365 				 * receive operation.
1366 				 *
1367 				 * This avoids a later panic("receive 1a")
1368 				 * when compiled with DIAGNOSTIC.
1369 				 */
1370 				if (m && mbuf_removed && atomic)
1371 					(void) sbdroprecord(&so->so_rcv);
1372 
1373 				goto release;
1374 			}
1375 		} else
1376 			uio->uio_resid -= len;
1377 		if (len == m->m_len - moff) {
1378 			if (m->m_flags & M_EOR)
1379 				flags |= MSG_EOR;
1380 			if (flags & MSG_PEEK) {
1381 				m = m->m_next;
1382 				moff = 0;
1383 			} else {
1384 				nextrecord = m->m_nextpkt;
1385 				sbfree(&so->so_rcv, m);
1386 				if (mp) {
1387 					*mp = m;
1388 					mp = &m->m_next;
1389 					so->so_rcv.sb_mb = m = m->m_next;
1390 					*mp = NULL;
1391 				} else {
1392 					MFREE(m, so->so_rcv.sb_mb);
1393 					m = so->so_rcv.sb_mb;
1394 				}
1395 				/*
1396 				 * If m != NULL, we also know that
1397 				 * so->so_rcv.sb_mb != NULL.
1398 				 */
1399 				KASSERT(so->so_rcv.sb_mb == m);
1400 				if (m) {
1401 					m->m_nextpkt = nextrecord;
1402 					if (nextrecord == NULL)
1403 						so->so_rcv.sb_lastrecord = m;
1404 				} else {
1405 					so->so_rcv.sb_mb = nextrecord;
1406 					SB_EMPTY_FIXUP(&so->so_rcv);
1407 				}
1408 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1409 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1410 			}
1411 		} else if (flags & MSG_PEEK)
1412 			moff += len;
1413 		else {
1414 			if (mp != NULL) {
1415 				mt = m_copym(m, 0, len, M_NOWAIT);
1416 				if (__predict_false(mt == NULL)) {
1417 					sounlock(so);
1418 					mt = m_copym(m, 0, len, M_WAIT);
1419 					solock(so);
1420 				}
1421 				*mp = mt;
1422 			}
1423 			m->m_data += len;
1424 			m->m_len -= len;
1425 			so->so_rcv.sb_cc -= len;
1426 		}
1427 		if (so->so_oobmark) {
1428 			if ((flags & MSG_PEEK) == 0) {
1429 				so->so_oobmark -= len;
1430 				if (so->so_oobmark == 0) {
1431 					so->so_state |= SS_RCVATMARK;
1432 					break;
1433 				}
1434 			} else {
1435 				offset += len;
1436 				if (offset == so->so_oobmark)
1437 					break;
1438 			}
1439 		}
1440 		if (flags & MSG_EOR)
1441 			break;
1442 		/*
1443 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1444 		 * we must not quit until "uio->uio_resid == 0" or an error
1445 		 * termination.  If a signal/timeout occurs, return
1446 		 * with a short count but without error.
1447 		 * Keep sockbuf locked against other readers.
1448 		 */
1449 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1450 		    !sosendallatonce(so) && !nextrecord) {
1451 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1452 				break;
1453 			/*
1454 			 * If we are peeking and the socket receive buffer is
1455 			 * full, stop since we can't get more data to peek at.
1456 			 */
1457 			if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
1458 				break;
1459 			/*
1460 			 * If we've drained the socket buffer, tell the
1461 			 * protocol in case it needs to do something to
1462 			 * get it filled again.
1463 			 */
1464 			if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1465 				(*pr->pr_usrreq)(so, PRU_RCVD,
1466 				    NULL, (struct mbuf *)(long)flags, NULL, l);
1467 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1468 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1469 			error = sbwait(&so->so_rcv);
1470 			if (error != 0) {
1471 				sbunlock(&so->so_rcv);
1472 				sounlock(so);
1473 				splx(s);
1474 				return 0;
1475 			}
1476 			if ((m = so->so_rcv.sb_mb) != NULL)
1477 				nextrecord = m->m_nextpkt;
1478 		}
1479 	}
1480 
1481 	if (m && atomic) {
1482 		flags |= MSG_TRUNC;
1483 		if ((flags & MSG_PEEK) == 0)
1484 			(void) sbdroprecord(&so->so_rcv);
1485 	}
1486 	if ((flags & MSG_PEEK) == 0) {
1487 		if (m == NULL) {
1488 			/*
1489 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1490 			 * part makes sure sb_lastrecord is up-to-date if
1491 			 * there is still data in the socket buffer.
1492 			 */
1493 			so->so_rcv.sb_mb = nextrecord;
1494 			if (so->so_rcv.sb_mb == NULL) {
1495 				so->so_rcv.sb_mbtail = NULL;
1496 				so->so_rcv.sb_lastrecord = NULL;
1497 			} else if (nextrecord->m_nextpkt == NULL)
1498 				so->so_rcv.sb_lastrecord = nextrecord;
1499 		}
1500 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1501 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1502 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1503 			(*pr->pr_usrreq)(so, PRU_RCVD, NULL,
1504 			    (struct mbuf *)(long)flags, NULL, l);
1505 	}
1506 	if (orig_resid == uio->uio_resid && orig_resid &&
1507 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1508 		sbunlock(&so->so_rcv);
1509 		goto restart;
1510 	}
1511 
1512 	if (flagsp != NULL)
1513 		*flagsp |= flags;
1514  release:
1515 	sbunlock(&so->so_rcv);
1516 	sounlock(so);
1517 	splx(s);
1518 	return error;
1519 }
1520 
1521 int
1522 soshutdown(struct socket *so, int how)
1523 {
1524 	const struct protosw	*pr;
1525 	int	error;
1526 
1527 	KASSERT(solocked(so));
1528 
1529 	pr = so->so_proto;
1530 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1531 		return (EINVAL);
1532 
1533 	if (how == SHUT_RD || how == SHUT_RDWR) {
1534 		sorflush(so);
1535 		error = 0;
1536 	}
1537 	if (how == SHUT_WR || how == SHUT_RDWR)
1538 		error = (*pr->pr_usrreq)(so, PRU_SHUTDOWN, NULL,
1539 		    NULL, NULL, NULL);
1540 
1541 	return error;
1542 }
1543 
1544 void
1545 sorflush(struct socket *so)
1546 {
1547 	struct sockbuf	*sb, asb;
1548 	const struct protosw	*pr;
1549 
1550 	KASSERT(solocked(so));
1551 
1552 	sb = &so->so_rcv;
1553 	pr = so->so_proto;
1554 	socantrcvmore(so);
1555 	sb->sb_flags |= SB_NOINTR;
1556 	(void )sblock(sb, M_WAITOK);
1557 	sbunlock(sb);
1558 	asb = *sb;
1559 	/*
1560 	 * Clear most of the sockbuf structure, but leave some of the
1561 	 * fields valid.
1562 	 */
1563 	memset(&sb->sb_startzero, 0,
1564 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1565 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) {
1566 		sounlock(so);
1567 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1568 		solock(so);
1569 	}
1570 	sbrelease(&asb, so);
1571 }
1572 
1573 static int
1574 sosetopt1(struct socket *so, int level, int optname, struct mbuf *m)
1575 {
1576 	int optval, val;
1577 	struct linger	*l;
1578 	struct sockbuf	*sb;
1579 	struct timeval *tv;
1580 
1581 	switch (optname) {
1582 
1583 	case SO_LINGER:
1584 		if (m == NULL || m->m_len != sizeof(struct linger))
1585 			return EINVAL;
1586 		l = mtod(m, struct linger *);
1587 		if (l->l_linger < 0 || l->l_linger > USHRT_MAX ||
1588 		    l->l_linger > (INT_MAX / hz))
1589 			return EDOM;
1590 		so->so_linger = l->l_linger;
1591 		if (l->l_onoff)
1592 			so->so_options |= SO_LINGER;
1593 		else
1594 			so->so_options &= ~SO_LINGER;
1595 		break;
1596 
1597 	case SO_DEBUG:
1598 	case SO_KEEPALIVE:
1599 	case SO_DONTROUTE:
1600 	case SO_USELOOPBACK:
1601 	case SO_BROADCAST:
1602 	case SO_REUSEADDR:
1603 	case SO_REUSEPORT:
1604 	case SO_OOBINLINE:
1605 	case SO_TIMESTAMP:
1606 		if (m == NULL || m->m_len < sizeof(int))
1607 			return EINVAL;
1608 		if (*mtod(m, int *))
1609 			so->so_options |= optname;
1610 		else
1611 			so->so_options &= ~optname;
1612 		break;
1613 
1614 	case SO_SNDBUF:
1615 	case SO_RCVBUF:
1616 	case SO_SNDLOWAT:
1617 	case SO_RCVLOWAT:
1618 		if (m == NULL || m->m_len < sizeof(int))
1619 			return EINVAL;
1620 
1621 		/*
1622 		 * Values < 1 make no sense for any of these
1623 		 * options, so disallow them.
1624 		 */
1625 		optval = *mtod(m, int *);
1626 		if (optval < 1)
1627 			return EINVAL;
1628 
1629 		switch (optname) {
1630 
1631 		case SO_SNDBUF:
1632 		case SO_RCVBUF:
1633 			sb = (optname == SO_SNDBUF) ?
1634 			    &so->so_snd : &so->so_rcv;
1635 			if (sbreserve(sb, (u_long)optval, so) == 0)
1636 				return ENOBUFS;
1637 			sb->sb_flags &= ~SB_AUTOSIZE;
1638 			break;
1639 
1640 		/*
1641 		 * Make sure the low-water is never greater than
1642 		 * the high-water.
1643 		 */
1644 		case SO_SNDLOWAT:
1645 			so->so_snd.sb_lowat =
1646 			    (optval > so->so_snd.sb_hiwat) ?
1647 			    so->so_snd.sb_hiwat : optval;
1648 			break;
1649 		case SO_RCVLOWAT:
1650 			so->so_rcv.sb_lowat =
1651 			    (optval > so->so_rcv.sb_hiwat) ?
1652 			    so->so_rcv.sb_hiwat : optval;
1653 			break;
1654 		}
1655 		break;
1656 
1657 	case SO_SNDTIMEO:
1658 	case SO_RCVTIMEO:
1659 		if (m == NULL || m->m_len < sizeof(*tv))
1660 			return EINVAL;
1661 		tv = mtod(m, struct timeval *);
1662 		if (tv->tv_sec > (INT_MAX - tv->tv_usec / tick) / hz)
1663 			return EDOM;
1664 		val = tv->tv_sec * hz + tv->tv_usec / tick;
1665 		if (val == 0 && tv->tv_usec != 0)
1666 			val = 1;
1667 
1668 		switch (optname) {
1669 
1670 		case SO_SNDTIMEO:
1671 			so->so_snd.sb_timeo = val;
1672 			break;
1673 		case SO_RCVTIMEO:
1674 			so->so_rcv.sb_timeo = val;
1675 			break;
1676 		}
1677 		break;
1678 
1679 	default:
1680 		return ENOPROTOOPT;
1681 	}
1682 	return 0;
1683 }
1684 
1685 int
1686 sosetopt(struct socket *so, int level, int optname, struct mbuf *m)
1687 {
1688 	int error, prerr;
1689 
1690 	solock(so);
1691 	if (level == SOL_SOCKET)
1692 		error = sosetopt1(so, level, optname, m);
1693 	else
1694 		error = ENOPROTOOPT;
1695 
1696 	if ((error == 0 || error == ENOPROTOOPT) &&
1697 	    so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) {
1698 		/* give the protocol stack a shot */
1699 		prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, level,
1700 		    optname, &m);
1701 		if (prerr == 0)
1702 			error = 0;
1703 		else if (prerr != ENOPROTOOPT)
1704 			error = prerr;
1705 	} else if (m != NULL)
1706 		(void)m_free(m);
1707 	sounlock(so);
1708 	return error;
1709 }
1710 
1711 int
1712 sogetopt(struct socket *so, int level, int optname, struct mbuf **mp)
1713 {
1714 	struct mbuf	*m;
1715 	int		error;
1716 
1717 	solock(so);
1718 	if (level != SOL_SOCKET) {
1719 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1720 			error = ((*so->so_proto->pr_ctloutput)
1721 				  (PRCO_GETOPT, so, level, optname, mp));
1722 		} else
1723 			error = (ENOPROTOOPT);
1724 	} else {
1725 		m = m_get(M_WAIT, MT_SOOPTS);
1726 		m->m_len = sizeof(int);
1727 
1728 		switch (optname) {
1729 
1730 		case SO_LINGER:
1731 			m->m_len = sizeof(struct linger);
1732 			mtod(m, struct linger *)->l_onoff =
1733 			    (so->so_options & SO_LINGER) ? 1 : 0;
1734 			mtod(m, struct linger *)->l_linger = so->so_linger;
1735 			break;
1736 
1737 		case SO_USELOOPBACK:
1738 		case SO_DONTROUTE:
1739 		case SO_DEBUG:
1740 		case SO_KEEPALIVE:
1741 		case SO_REUSEADDR:
1742 		case SO_REUSEPORT:
1743 		case SO_BROADCAST:
1744 		case SO_OOBINLINE:
1745 		case SO_TIMESTAMP:
1746 			*mtod(m, int *) = (so->so_options & optname) ? 1 : 0;
1747 			break;
1748 
1749 		case SO_TYPE:
1750 			*mtod(m, int *) = so->so_type;
1751 			break;
1752 
1753 		case SO_ERROR:
1754 			*mtod(m, int *) = so->so_error;
1755 			so->so_error = 0;
1756 			break;
1757 
1758 		case SO_SNDBUF:
1759 			*mtod(m, int *) = so->so_snd.sb_hiwat;
1760 			break;
1761 
1762 		case SO_RCVBUF:
1763 			*mtod(m, int *) = so->so_rcv.sb_hiwat;
1764 			break;
1765 
1766 		case SO_SNDLOWAT:
1767 			*mtod(m, int *) = so->so_snd.sb_lowat;
1768 			break;
1769 
1770 		case SO_RCVLOWAT:
1771 			*mtod(m, int *) = so->so_rcv.sb_lowat;
1772 			break;
1773 
1774 		case SO_SNDTIMEO:
1775 		case SO_RCVTIMEO:
1776 		    {
1777 			int val = (optname == SO_SNDTIMEO ?
1778 			     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1779 
1780 			m->m_len = sizeof(struct timeval);
1781 			mtod(m, struct timeval *)->tv_sec = val / hz;
1782 			mtod(m, struct timeval *)->tv_usec =
1783 			    (val % hz) * tick;
1784 			break;
1785 		    }
1786 
1787 		case SO_OVERFLOWED:
1788 			*mtod(m, int *) = so->so_rcv.sb_overflowed;
1789 			break;
1790 
1791 		default:
1792 			sounlock(so);
1793 			(void)m_free(m);
1794 			return (ENOPROTOOPT);
1795 		}
1796 		*mp = m;
1797 		error = 0;
1798 	}
1799 
1800 	sounlock(so);
1801 	return (error);
1802 }
1803 
1804 void
1805 sohasoutofband(struct socket *so)
1806 {
1807 
1808 	fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
1809 	selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, 0);
1810 }
1811 
1812 static void
1813 filt_sordetach(struct knote *kn)
1814 {
1815 	struct socket	*so;
1816 
1817 	so = ((file_t *)kn->kn_obj)->f_data;
1818 	solock(so);
1819 	SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
1820 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
1821 		so->so_rcv.sb_flags &= ~SB_KNOTE;
1822 	sounlock(so);
1823 }
1824 
1825 /*ARGSUSED*/
1826 static int
1827 filt_soread(struct knote *kn, long hint)
1828 {
1829 	struct socket	*so;
1830 	int rv;
1831 
1832 	so = ((file_t *)kn->kn_obj)->f_data;
1833 	if (hint != NOTE_SUBMIT)
1834 		solock(so);
1835 	kn->kn_data = so->so_rcv.sb_cc;
1836 	if (so->so_state & SS_CANTRCVMORE) {
1837 		kn->kn_flags |= EV_EOF;
1838 		kn->kn_fflags = so->so_error;
1839 		rv = 1;
1840 	} else if (so->so_error)	/* temporary udp error */
1841 		rv = 1;
1842 	else if (kn->kn_sfflags & NOTE_LOWAT)
1843 		rv = (kn->kn_data >= kn->kn_sdata);
1844 	else
1845 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
1846 	if (hint != NOTE_SUBMIT)
1847 		sounlock(so);
1848 	return rv;
1849 }
1850 
1851 static void
1852 filt_sowdetach(struct knote *kn)
1853 {
1854 	struct socket	*so;
1855 
1856 	so = ((file_t *)kn->kn_obj)->f_data;
1857 	solock(so);
1858 	SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
1859 	if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
1860 		so->so_snd.sb_flags &= ~SB_KNOTE;
1861 	sounlock(so);
1862 }
1863 
1864 /*ARGSUSED*/
1865 static int
1866 filt_sowrite(struct knote *kn, long hint)
1867 {
1868 	struct socket	*so;
1869 	int rv;
1870 
1871 	so = ((file_t *)kn->kn_obj)->f_data;
1872 	if (hint != NOTE_SUBMIT)
1873 		solock(so);
1874 	kn->kn_data = sbspace(&so->so_snd);
1875 	if (so->so_state & SS_CANTSENDMORE) {
1876 		kn->kn_flags |= EV_EOF;
1877 		kn->kn_fflags = so->so_error;
1878 		rv = 1;
1879 	} else if (so->so_error)	/* temporary udp error */
1880 		rv = 1;
1881 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
1882 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
1883 		rv = 0;
1884 	else if (kn->kn_sfflags & NOTE_LOWAT)
1885 		rv = (kn->kn_data >= kn->kn_sdata);
1886 	else
1887 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
1888 	if (hint != NOTE_SUBMIT)
1889 		sounlock(so);
1890 	return rv;
1891 }
1892 
1893 /*ARGSUSED*/
1894 static int
1895 filt_solisten(struct knote *kn, long hint)
1896 {
1897 	struct socket	*so;
1898 	int rv;
1899 
1900 	so = ((file_t *)kn->kn_obj)->f_data;
1901 
1902 	/*
1903 	 * Set kn_data to number of incoming connections, not
1904 	 * counting partial (incomplete) connections.
1905 	 */
1906 	if (hint != NOTE_SUBMIT)
1907 		solock(so);
1908 	kn->kn_data = so->so_qlen;
1909 	rv = (kn->kn_data > 0);
1910 	if (hint != NOTE_SUBMIT)
1911 		sounlock(so);
1912 	return rv;
1913 }
1914 
1915 static const struct filterops solisten_filtops =
1916 	{ 1, NULL, filt_sordetach, filt_solisten };
1917 static const struct filterops soread_filtops =
1918 	{ 1, NULL, filt_sordetach, filt_soread };
1919 static const struct filterops sowrite_filtops =
1920 	{ 1, NULL, filt_sowdetach, filt_sowrite };
1921 
1922 int
1923 soo_kqfilter(struct file *fp, struct knote *kn)
1924 {
1925 	struct socket	*so;
1926 	struct sockbuf	*sb;
1927 
1928 	so = ((file_t *)kn->kn_obj)->f_data;
1929 	solock(so);
1930 	switch (kn->kn_filter) {
1931 	case EVFILT_READ:
1932 		if (so->so_options & SO_ACCEPTCONN)
1933 			kn->kn_fop = &solisten_filtops;
1934 		else
1935 			kn->kn_fop = &soread_filtops;
1936 		sb = &so->so_rcv;
1937 		break;
1938 	case EVFILT_WRITE:
1939 		kn->kn_fop = &sowrite_filtops;
1940 		sb = &so->so_snd;
1941 		break;
1942 	default:
1943 		sounlock(so);
1944 		return (EINVAL);
1945 	}
1946 	SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
1947 	sb->sb_flags |= SB_KNOTE;
1948 	sounlock(so);
1949 	return (0);
1950 }
1951 
1952 static int
1953 sodopoll(struct socket *so, int events)
1954 {
1955 	int revents;
1956 
1957 	revents = 0;
1958 
1959 	if (events & (POLLIN | POLLRDNORM))
1960 		if (soreadable(so))
1961 			revents |= events & (POLLIN | POLLRDNORM);
1962 
1963 	if (events & (POLLOUT | POLLWRNORM))
1964 		if (sowritable(so))
1965 			revents |= events & (POLLOUT | POLLWRNORM);
1966 
1967 	if (events & (POLLPRI | POLLRDBAND))
1968 		if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
1969 			revents |= events & (POLLPRI | POLLRDBAND);
1970 
1971 	return revents;
1972 }
1973 
1974 int
1975 sopoll(struct socket *so, int events)
1976 {
1977 	int revents = 0;
1978 
1979 #ifndef DIAGNOSTIC
1980 	/*
1981 	 * Do a quick, unlocked check in expectation that the socket
1982 	 * will be ready for I/O.  Don't do this check if DIAGNOSTIC,
1983 	 * as the solocked() assertions will fail.
1984 	 */
1985 	if ((revents = sodopoll(so, events)) != 0)
1986 		return revents;
1987 #endif
1988 
1989 	solock(so);
1990 	if ((revents = sodopoll(so, events)) == 0) {
1991 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
1992 			selrecord(curlwp, &so->so_rcv.sb_sel);
1993 			so->so_rcv.sb_flags |= SB_NOTIFY;
1994 		}
1995 
1996 		if (events & (POLLOUT | POLLWRNORM)) {
1997 			selrecord(curlwp, &so->so_snd.sb_sel);
1998 			so->so_snd.sb_flags |= SB_NOTIFY;
1999 		}
2000 	}
2001 	sounlock(so);
2002 
2003 	return revents;
2004 }
2005 
2006 
2007 #include <sys/sysctl.h>
2008 
2009 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO);
2010 
2011 /*
2012  * sysctl helper routine for kern.somaxkva.  ensures that the given
2013  * value is not too small.
2014  * (XXX should we maybe make sure it's not too large as well?)
2015  */
2016 static int
2017 sysctl_kern_somaxkva(SYSCTLFN_ARGS)
2018 {
2019 	int error, new_somaxkva;
2020 	struct sysctlnode node;
2021 
2022 	new_somaxkva = somaxkva;
2023 	node = *rnode;
2024 	node.sysctl_data = &new_somaxkva;
2025 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2026 	if (error || newp == NULL)
2027 		return (error);
2028 
2029 	if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */
2030 		return (EINVAL);
2031 
2032 	mutex_enter(&so_pendfree_lock);
2033 	somaxkva = new_somaxkva;
2034 	cv_broadcast(&socurkva_cv);
2035 	mutex_exit(&so_pendfree_lock);
2036 
2037 	return (error);
2038 }
2039 
2040 SYSCTL_SETUP(sysctl_kern_somaxkva_setup, "sysctl kern.somaxkva setup")
2041 {
2042 
2043 	sysctl_createv(clog, 0, NULL, NULL,
2044 		       CTLFLAG_PERMANENT,
2045 		       CTLTYPE_NODE, "kern", NULL,
2046 		       NULL, 0, NULL, 0,
2047 		       CTL_KERN, CTL_EOL);
2048 
2049 	sysctl_createv(clog, 0, NULL, NULL,
2050 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2051 		       CTLTYPE_INT, "somaxkva",
2052 		       SYSCTL_DESCR("Maximum amount of kernel memory to be "
2053 				    "used for socket buffers"),
2054 		       sysctl_kern_somaxkva, 0, NULL, 0,
2055 		       CTL_KERN, KERN_SOMAXKVA, CTL_EOL);
2056 }
2057