xref: /netbsd-src/sys/kern/uipc_socket.c (revision cac8e449158efc7261bebc8657cbb0125a2cfdde)
1 /*	$NetBSD: uipc_socket.c,v 1.170 2008/08/04 03:55:47 tls 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 1.170 2008/08/04 03:55:47 tls Exp $");
67 
68 #include "opt_inet.h"
69 #include "opt_sock_counters.h"
70 #include "opt_sosend_loan.h"
71 #include "opt_mbuftrace.h"
72 #include "opt_somaxkva.h"
73 #include "opt_multiprocessor.h"	/* XXX */
74 
75 #include <sys/param.h>
76 #include <sys/systm.h>
77 #include <sys/proc.h>
78 #include <sys/file.h>
79 #include <sys/filedesc.h>
80 #include <sys/malloc.h>
81 #include <sys/mbuf.h>
82 #include <sys/domain.h>
83 #include <sys/kernel.h>
84 #include <sys/protosw.h>
85 #include <sys/socket.h>
86 #include <sys/socketvar.h>
87 #include <sys/signalvar.h>
88 #include <sys/resourcevar.h>
89 #include <sys/event.h>
90 #include <sys/poll.h>
91 #include <sys/kauth.h>
92 #include <sys/mutex.h>
93 #include <sys/condvar.h>
94 
95 #include <uvm/uvm.h>
96 
97 MALLOC_DEFINE(M_SOOPTS, "soopts", "socket options");
98 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
99 
100 extern const struct fileops socketops;
101 
102 extern int	somaxconn;			/* patchable (XXX sysctl) */
103 int		somaxconn = SOMAXCONN;
104 kmutex_t	*softnet_lock;
105 
106 #ifdef SOSEND_COUNTERS
107 #include <sys/device.h>
108 
109 static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
110     NULL, "sosend", "loan big");
111 static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
112     NULL, "sosend", "copy big");
113 static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
114     NULL, "sosend", "copy small");
115 static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
116     NULL, "sosend", "kva limit");
117 
118 #define	SOSEND_COUNTER_INCR(ev)		(ev)->ev_count++
119 
120 EVCNT_ATTACH_STATIC(sosend_loan_big);
121 EVCNT_ATTACH_STATIC(sosend_copy_big);
122 EVCNT_ATTACH_STATIC(sosend_copy_small);
123 EVCNT_ATTACH_STATIC(sosend_kvalimit);
124 #else
125 
126 #define	SOSEND_COUNTER_INCR(ev)		/* nothing */
127 
128 #endif /* SOSEND_COUNTERS */
129 
130 static struct callback_entry sokva_reclaimerentry;
131 
132 #if defined(SOSEND_NO_LOAN) || defined(MULTIPROCESSOR)
133 int sock_loan_thresh = -1;
134 #else
135 int sock_loan_thresh = 4096;
136 #endif
137 
138 static kmutex_t so_pendfree_lock;
139 static struct mbuf *so_pendfree;
140 
141 #ifndef SOMAXKVA
142 #define	SOMAXKVA (16 * 1024 * 1024)
143 #endif
144 int somaxkva = SOMAXKVA;
145 static int socurkva;
146 static kcondvar_t socurkva_cv;
147 
148 #define	SOCK_LOAN_CHUNK		65536
149 
150 static size_t sodopendfree(void);
151 static size_t sodopendfreel(void);
152 
153 static vsize_t
154 sokvareserve(struct socket *so, vsize_t len)
155 {
156 	int error;
157 
158 	mutex_enter(&so_pendfree_lock);
159 	while (socurkva + len > somaxkva) {
160 		size_t freed;
161 
162 		/*
163 		 * try to do pendfree.
164 		 */
165 
166 		freed = sodopendfreel();
167 
168 		/*
169 		 * if some kva was freed, try again.
170 		 */
171 
172 		if (freed)
173 			continue;
174 
175 		SOSEND_COUNTER_INCR(&sosend_kvalimit);
176 		error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock);
177 		if (error) {
178 			len = 0;
179 			break;
180 		}
181 	}
182 	socurkva += len;
183 	mutex_exit(&so_pendfree_lock);
184 	return len;
185 }
186 
187 static void
188 sokvaunreserve(vsize_t len)
189 {
190 
191 	mutex_enter(&so_pendfree_lock);
192 	socurkva -= len;
193 	cv_broadcast(&socurkva_cv);
194 	mutex_exit(&so_pendfree_lock);
195 }
196 
197 /*
198  * sokvaalloc: allocate kva for loan.
199  */
200 
201 vaddr_t
202 sokvaalloc(vsize_t len, struct socket *so)
203 {
204 	vaddr_t lva;
205 
206 	/*
207 	 * reserve kva.
208 	 */
209 
210 	if (sokvareserve(so, len) == 0)
211 		return 0;
212 
213 	/*
214 	 * allocate kva.
215 	 */
216 
217 	lva = uvm_km_alloc(kernel_map, len, 0, UVM_KMF_VAONLY | UVM_KMF_WAITVA);
218 	if (lva == 0) {
219 		sokvaunreserve(len);
220 		return (0);
221 	}
222 
223 	return lva;
224 }
225 
226 /*
227  * sokvafree: free kva for loan.
228  */
229 
230 void
231 sokvafree(vaddr_t sva, vsize_t len)
232 {
233 
234 	/*
235 	 * free kva.
236 	 */
237 
238 	uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY);
239 
240 	/*
241 	 * unreserve kva.
242 	 */
243 
244 	sokvaunreserve(len);
245 }
246 
247 static void
248 sodoloanfree(struct vm_page **pgs, void *buf, size_t size)
249 {
250 	vaddr_t sva, eva;
251 	vsize_t len;
252 	int npgs;
253 
254 	KASSERT(pgs != NULL);
255 
256 	eva = round_page((vaddr_t) buf + size);
257 	sva = trunc_page((vaddr_t) buf);
258 	len = eva - sva;
259 	npgs = len >> PAGE_SHIFT;
260 
261 	pmap_kremove(sva, len);
262 	pmap_update(pmap_kernel());
263 	uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE);
264 	sokvafree(sva, len);
265 }
266 
267 static size_t
268 sodopendfree(void)
269 {
270 	size_t rv;
271 
272 	if (__predict_true(so_pendfree == NULL))
273 		return 0;
274 
275 	mutex_enter(&so_pendfree_lock);
276 	rv = sodopendfreel();
277 	mutex_exit(&so_pendfree_lock);
278 
279 	return rv;
280 }
281 
282 /*
283  * sodopendfreel: free mbufs on "pendfree" list.
284  * unlock and relock so_pendfree_lock when freeing mbufs.
285  *
286  * => called with so_pendfree_lock held.
287  */
288 
289 static size_t
290 sodopendfreel(void)
291 {
292 	struct mbuf *m, *next;
293 	size_t rv = 0;
294 
295 	KASSERT(mutex_owned(&so_pendfree_lock));
296 
297 	while (so_pendfree != NULL) {
298 		m = so_pendfree;
299 		so_pendfree = NULL;
300 		mutex_exit(&so_pendfree_lock);
301 
302 		for (; m != NULL; m = next) {
303 			next = m->m_next;
304 			KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0);
305 			KASSERT(m->m_ext.ext_refcnt == 0);
306 
307 			rv += m->m_ext.ext_size;
308 			sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf,
309 			    m->m_ext.ext_size);
310 			pool_cache_put(mb_cache, m);
311 		}
312 
313 		mutex_enter(&so_pendfree_lock);
314 	}
315 
316 	return (rv);
317 }
318 
319 void
320 soloanfree(struct mbuf *m, void *buf, size_t size, void *arg)
321 {
322 
323 	KASSERT(m != NULL);
324 
325 	/*
326 	 * postpone freeing mbuf.
327 	 *
328 	 * we can't do it in interrupt context
329 	 * because we need to put kva back to kernel_map.
330 	 */
331 
332 	mutex_enter(&so_pendfree_lock);
333 	m->m_next = so_pendfree;
334 	so_pendfree = m;
335 	cv_broadcast(&socurkva_cv);
336 	mutex_exit(&so_pendfree_lock);
337 }
338 
339 static long
340 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space)
341 {
342 	struct iovec *iov = uio->uio_iov;
343 	vaddr_t sva, eva;
344 	vsize_t len;
345 	vaddr_t lva;
346 	int npgs, error;
347 	vaddr_t va;
348 	int i;
349 
350 	if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace))
351 		return (0);
352 
353 	if (iov->iov_len < (size_t) space)
354 		space = iov->iov_len;
355 	if (space > SOCK_LOAN_CHUNK)
356 		space = SOCK_LOAN_CHUNK;
357 
358 	eva = round_page((vaddr_t) iov->iov_base + space);
359 	sva = trunc_page((vaddr_t) iov->iov_base);
360 	len = eva - sva;
361 	npgs = len >> PAGE_SHIFT;
362 
363 	KASSERT(npgs <= M_EXT_MAXPAGES);
364 
365 	lva = sokvaalloc(len, so);
366 	if (lva == 0)
367 		return 0;
368 
369 	error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len,
370 	    m->m_ext.ext_pgs, UVM_LOAN_TOPAGE);
371 	if (error) {
372 		sokvafree(lva, len);
373 		return (0);
374 	}
375 
376 	for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE)
377 		pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
378 		    VM_PROT_READ);
379 	pmap_update(pmap_kernel());
380 
381 	lva += (vaddr_t) iov->iov_base & PAGE_MASK;
382 
383 	MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so);
384 	m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
385 
386 	uio->uio_resid -= space;
387 	/* uio_offset not updated, not set/used for write(2) */
388 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space;
389 	uio->uio_iov->iov_len -= space;
390 	if (uio->uio_iov->iov_len == 0) {
391 		uio->uio_iov++;
392 		uio->uio_iovcnt--;
393 	}
394 
395 	return (space);
396 }
397 
398 static int
399 sokva_reclaim_callback(struct callback_entry *ce, void *obj, void *arg)
400 {
401 
402 	KASSERT(ce == &sokva_reclaimerentry);
403 	KASSERT(obj == NULL);
404 
405 	sodopendfree();
406 	if (!vm_map_starved_p(kernel_map)) {
407 		return CALLBACK_CHAIN_ABORT;
408 	}
409 	return CALLBACK_CHAIN_CONTINUE;
410 }
411 
412 struct mbuf *
413 getsombuf(struct socket *so, int type)
414 {
415 	struct mbuf *m;
416 
417 	m = m_get(M_WAIT, type);
418 	MCLAIM(m, so->so_mowner);
419 	return m;
420 }
421 
422 struct mbuf *
423 m_intopt(struct socket *so, int val)
424 {
425 	struct mbuf *m;
426 
427 	m = getsombuf(so, MT_SOOPTS);
428 	m->m_len = sizeof(int);
429 	*mtod(m, int *) = val;
430 	return m;
431 }
432 
433 void
434 soinit(void)
435 {
436 
437 	mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM);
438 	softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
439 	cv_init(&socurkva_cv, "sokva");
440 	soinit2();
441 
442 	/* Set the initial adjusted socket buffer size. */
443 	if (sb_max_set(sb_max))
444 		panic("bad initial sb_max value: %lu", sb_max);
445 
446 	callback_register(&vm_map_to_kernel(kernel_map)->vmk_reclaim_callback,
447 	    &sokva_reclaimerentry, NULL, sokva_reclaim_callback);
448 }
449 
450 /*
451  * Socket operation routines.
452  * These routines are called by the routines in
453  * sys_socket.c or from a system process, and
454  * implement the semantics of socket operations by
455  * switching out to the protocol specific routines.
456  */
457 /*ARGSUSED*/
458 int
459 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l,
460 	 struct socket *lockso)
461 {
462 	const struct protosw	*prp;
463 	struct socket	*so;
464 	uid_t		uid;
465 	int		error;
466 	kmutex_t	*lock;
467 
468 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET,
469 	    KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type),
470 	    KAUTH_ARG(proto));
471 	if (error != 0)
472 		return error;
473 
474 	if (proto)
475 		prp = pffindproto(dom, proto, type);
476 	else
477 		prp = pffindtype(dom, type);
478 	if (prp == NULL) {
479 		/* no support for domain */
480 		if (pffinddomain(dom) == 0)
481 			return EAFNOSUPPORT;
482 		/* no support for socket type */
483 		if (proto == 0 && type != 0)
484 			return EPROTOTYPE;
485 		return EPROTONOSUPPORT;
486 	}
487 	if (prp->pr_usrreq == NULL)
488 		return EPROTONOSUPPORT;
489 	if (prp->pr_type != type)
490 		return EPROTOTYPE;
491 
492 	so = soget(true);
493 	so->so_type = type;
494 	so->so_proto = prp;
495 	so->so_send = sosend;
496 	so->so_receive = soreceive;
497 #ifdef MBUFTRACE
498 	so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner;
499 	so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner;
500 	so->so_mowner = &prp->pr_domain->dom_mowner;
501 #endif
502 	uid = kauth_cred_geteuid(l->l_cred);
503 	so->so_uidinfo = uid_find(uid);
504 	so->so_egid = kauth_cred_getegid(l->l_cred);
505 	so->so_cpid = l->l_proc->p_pid;
506 	if (lockso != NULL) {
507 		/* Caller wants us to share a lock. */
508 		lock = lockso->so_lock;
509 		so->so_lock = lock;
510 		mutex_obj_hold(lock);
511 		mutex_enter(lock);
512 	} else {
513 		/* Lock assigned and taken during PRU_ATTACH. */
514 	}
515 	error = (*prp->pr_usrreq)(so, PRU_ATTACH, NULL,
516 	    (struct mbuf *)(long)proto, NULL, l);
517 	KASSERT(solocked(so));
518 	if (error != 0) {
519 		so->so_state |= SS_NOFDREF;
520 		sofree(so);
521 		return error;
522 	}
523 	sounlock(so);
524 	*aso = so;
525 	return 0;
526 }
527 
528 /* On success, write file descriptor to fdout and return zero.  On
529  * failure, return non-zero; *fdout will be undefined.
530  */
531 int
532 fsocreate(int domain, struct socket **sop, int type, int protocol,
533     struct lwp *l, int *fdout)
534 {
535 	struct socket	*so;
536 	struct file	*fp;
537 	int		fd, error;
538 
539 	if ((error = fd_allocfile(&fp, &fd)) != 0)
540 		return (error);
541 	fp->f_flag = FREAD|FWRITE;
542 	fp->f_type = DTYPE_SOCKET;
543 	fp->f_ops = &socketops;
544 	error = socreate(domain, &so, type, protocol, l, NULL);
545 	if (error != 0) {
546 		fd_abort(curproc, fp, fd);
547 	} else {
548 		if (sop != NULL)
549 			*sop = so;
550 		fp->f_data = so;
551 		fd_affix(curproc, fp, fd);
552 		*fdout = fd;
553 	}
554 	return error;
555 }
556 
557 int
558 sobind(struct socket *so, struct mbuf *nam, struct lwp *l)
559 {
560 	int	error;
561 
562 	solock(so);
563 	error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, NULL, nam, NULL, l);
564 	sounlock(so);
565 	return error;
566 }
567 
568 int
569 solisten(struct socket *so, int backlog, struct lwp *l)
570 {
571 	int	error;
572 
573 	solock(so);
574 	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
575 	    SS_ISDISCONNECTING)) != 0) {
576 	    	sounlock(so);
577 		return (EOPNOTSUPP);
578 	}
579 	error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, NULL,
580 	    NULL, NULL, l);
581 	if (error != 0) {
582 		sounlock(so);
583 		return error;
584 	}
585 	if (TAILQ_EMPTY(&so->so_q))
586 		so->so_options |= SO_ACCEPTCONN;
587 	if (backlog < 0)
588 		backlog = 0;
589 	so->so_qlimit = min(backlog, somaxconn);
590 	sounlock(so);
591 	return 0;
592 }
593 
594 void
595 sofree(struct socket *so)
596 {
597 	u_int refs;
598 
599 	KASSERT(solocked(so));
600 
601 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
602 		sounlock(so);
603 		return;
604 	}
605 	if (so->so_head) {
606 		/*
607 		 * We must not decommission a socket that's on the accept(2)
608 		 * queue.  If we do, then accept(2) may hang after select(2)
609 		 * indicated that the listening socket was ready.
610 		 */
611 		if (!soqremque(so, 0)) {
612 			sounlock(so);
613 			return;
614 		}
615 	}
616 	if (so->so_rcv.sb_hiwat)
617 		(void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0,
618 		    RLIM_INFINITY);
619 	if (so->so_snd.sb_hiwat)
620 		(void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0,
621 		    RLIM_INFINITY);
622 	sbrelease(&so->so_snd, so);
623 	KASSERT(!cv_has_waiters(&so->so_cv));
624 	KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv));
625 	KASSERT(!cv_has_waiters(&so->so_snd.sb_cv));
626 	sorflush(so);
627 	refs = so->so_aborting;	/* XXX */
628 #ifdef INET
629 	/* remove acccept filter if one is present. */
630 	if (so->so_accf != NULL)
631 		do_setopt_accept_filter(so, NULL);
632 #endif
633 	sounlock(so);
634 	if (refs == 0)		/* XXX */
635 		soput(so);
636 }
637 
638 /*
639  * Close a socket on last file table reference removal.
640  * Initiate disconnect if connected.
641  * Free socket when disconnect complete.
642  */
643 int
644 soclose(struct socket *so)
645 {
646 	struct socket	*so2;
647 	int		error;
648 	int		error2;
649 
650 	error = 0;
651 	solock(so);
652 	if (so->so_options & SO_ACCEPTCONN) {
653 		do {
654 			while ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) {
655 				KASSERT(solocked2(so, so2));
656 				(void) soqremque(so2, 0);
657 				/* soabort drops the lock. */
658 				(void) soabort(so2);
659 				solock(so);
660 			}
661 			while ((so2 = TAILQ_FIRST(&so->so_q)) != 0) {
662 				KASSERT(solocked2(so, so2));
663 				(void) soqremque(so2, 1);
664 				/* soabort drops the lock. */
665 				(void) soabort(so2);
666 				solock(so);
667 			}
668 		} while (!TAILQ_EMPTY(&so->so_q0));
669 	}
670 	if (so->so_pcb == 0)
671 		goto discard;
672 	if (so->so_state & SS_ISCONNECTED) {
673 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
674 			error = sodisconnect(so);
675 			if (error)
676 				goto drop;
677 		}
678 		if (so->so_options & SO_LINGER) {
679 			if ((so->so_state & SS_ISDISCONNECTING) && so->so_nbio)
680 				goto drop;
681 			while (so->so_state & SS_ISCONNECTED) {
682 				error = sowait(so, so->so_linger * hz);
683 				if (error)
684 					break;
685 			}
686 		}
687 	}
688  drop:
689 	if (so->so_pcb) {
690 		error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH,
691 		    NULL, NULL, NULL, NULL);
692 		if (error == 0)
693 			error = error2;
694 	}
695  discard:
696 	if (so->so_state & SS_NOFDREF)
697 		panic("soclose: NOFDREF");
698 	so->so_state |= SS_NOFDREF;
699 	sofree(so);
700 	return (error);
701 }
702 
703 /*
704  * Must be called with the socket locked..  Will return with it unlocked.
705  */
706 int
707 soabort(struct socket *so)
708 {
709 	u_int refs;
710 	int error;
711 
712 	KASSERT(solocked(so));
713 	KASSERT(so->so_head == NULL);
714 
715 	so->so_aborting++;		/* XXX */
716 	error = (*so->so_proto->pr_usrreq)(so, PRU_ABORT, NULL,
717 	    NULL, NULL, NULL);
718 	refs = --so->so_aborting;	/* XXX */
719 	if (error || (refs == 0)) {
720 		sofree(so);
721 	} else {
722 		sounlock(so);
723 	}
724 	return error;
725 }
726 
727 int
728 soaccept(struct socket *so, struct mbuf *nam)
729 {
730 	int	error;
731 
732 	KASSERT(solocked(so));
733 
734 	error = 0;
735 	if ((so->so_state & SS_NOFDREF) == 0)
736 		panic("soaccept: !NOFDREF");
737 	so->so_state &= ~SS_NOFDREF;
738 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
739 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
740 		error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT,
741 		    NULL, nam, NULL, NULL);
742 	else
743 		error = ECONNABORTED;
744 
745 	return (error);
746 }
747 
748 int
749 soconnect(struct socket *so, struct mbuf *nam, struct lwp *l)
750 {
751 	int		error;
752 
753 	KASSERT(solocked(so));
754 
755 	if (so->so_options & SO_ACCEPTCONN)
756 		return (EOPNOTSUPP);
757 	/*
758 	 * If protocol is connection-based, can only connect once.
759 	 * Otherwise, if connected, try to disconnect first.
760 	 * This allows user to disconnect by connecting to, e.g.,
761 	 * a null address.
762 	 */
763 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
764 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
765 	    (error = sodisconnect(so))))
766 		error = EISCONN;
767 	else
768 		error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT,
769 		    NULL, nam, NULL, l);
770 	return (error);
771 }
772 
773 int
774 soconnect2(struct socket *so1, struct socket *so2)
775 {
776 	int	error;
777 
778 	KASSERT(solocked2(so1, so2));
779 
780 	error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2,
781 	    NULL, (struct mbuf *)so2, NULL, NULL);
782 	return (error);
783 }
784 
785 int
786 sodisconnect(struct socket *so)
787 {
788 	int	error;
789 
790 	KASSERT(solocked(so));
791 
792 	if ((so->so_state & SS_ISCONNECTED) == 0) {
793 		error = ENOTCONN;
794 	} else if (so->so_state & SS_ISDISCONNECTING) {
795 		error = EALREADY;
796 	} else {
797 		error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT,
798 		    NULL, NULL, NULL, NULL);
799 	}
800 	sodopendfree();
801 	return (error);
802 }
803 
804 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
805 /*
806  * Send on a socket.
807  * If send must go all at once and message is larger than
808  * send buffering, then hard error.
809  * Lock against other senders.
810  * If must go all at once and not enough room now, then
811  * inform user that this would block and do nothing.
812  * Otherwise, if nonblocking, send as much as possible.
813  * The data to be sent is described by "uio" if nonzero,
814  * otherwise by the mbuf chain "top" (which must be null
815  * if uio is not).  Data provided in mbuf chain must be small
816  * enough to send all at once.
817  *
818  * Returns nonzero on error, timeout or signal; callers
819  * must check for short counts if EINTR/ERESTART are returned.
820  * Data and control buffers are freed on return.
821  */
822 int
823 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
824 	struct mbuf *control, int flags, struct lwp *l)
825 {
826 	struct mbuf	**mp, *m;
827 	struct proc	*p;
828 	long		space, len, resid, clen, mlen;
829 	int		error, s, dontroute, atomic;
830 
831 	p = l->l_proc;
832 	sodopendfree();
833 	clen = 0;
834 
835 	/*
836 	 * solock() provides atomicity of access.  splsoftnet() prevents
837 	 * protocol processing soft interrupts from interrupting us and
838 	 * blocking (expensive).
839 	 */
840 	s = splsoftnet();
841 	solock(so);
842 	atomic = sosendallatonce(so) || top;
843 	if (uio)
844 		resid = uio->uio_resid;
845 	else
846 		resid = top->m_pkthdr.len;
847 	/*
848 	 * In theory resid should be unsigned.
849 	 * However, space must be signed, as it might be less than 0
850 	 * if we over-committed, and we must use a signed comparison
851 	 * of space and resid.  On the other hand, a negative resid
852 	 * causes us to loop sending 0-length segments to the protocol.
853 	 */
854 	if (resid < 0) {
855 		error = EINVAL;
856 		goto out;
857 	}
858 	dontroute =
859 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
860 	    (so->so_proto->pr_flags & PR_ATOMIC);
861 	l->l_ru.ru_msgsnd++;
862 	if (control)
863 		clen = control->m_len;
864  restart:
865 	if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
866 		goto out;
867 	do {
868 		if (so->so_state & SS_CANTSENDMORE) {
869 			error = EPIPE;
870 			goto release;
871 		}
872 		if (so->so_error) {
873 			error = so->so_error;
874 			so->so_error = 0;
875 			goto release;
876 		}
877 		if ((so->so_state & SS_ISCONNECTED) == 0) {
878 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
879 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
880 				    !(resid == 0 && clen != 0)) {
881 					error = ENOTCONN;
882 					goto release;
883 				}
884 			} else if (addr == 0) {
885 				error = EDESTADDRREQ;
886 				goto release;
887 			}
888 		}
889 		space = sbspace(&so->so_snd);
890 		if (flags & MSG_OOB)
891 			space += 1024;
892 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
893 		    clen > so->so_snd.sb_hiwat) {
894 			error = EMSGSIZE;
895 			goto release;
896 		}
897 		if (space < resid + clen &&
898 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
899 			if (so->so_nbio) {
900 				error = EWOULDBLOCK;
901 				goto release;
902 			}
903 			sbunlock(&so->so_snd);
904 			error = sbwait(&so->so_snd);
905 			if (error)
906 				goto out;
907 			goto restart;
908 		}
909 		mp = &top;
910 		space -= clen;
911 		do {
912 			if (uio == NULL) {
913 				/*
914 				 * Data is prepackaged in "top".
915 				 */
916 				resid = 0;
917 				if (flags & MSG_EOR)
918 					top->m_flags |= M_EOR;
919 			} else do {
920 				sounlock(so);
921 				splx(s);
922 				if (top == NULL) {
923 					m = m_gethdr(M_WAIT, MT_DATA);
924 					mlen = MHLEN;
925 					m->m_pkthdr.len = 0;
926 					m->m_pkthdr.rcvif = NULL;
927 				} else {
928 					m = m_get(M_WAIT, MT_DATA);
929 					mlen = MLEN;
930 				}
931 				MCLAIM(m, so->so_snd.sb_mowner);
932 				if (sock_loan_thresh >= 0 &&
933 				    uio->uio_iov->iov_len >= sock_loan_thresh &&
934 				    space >= sock_loan_thresh &&
935 				    (len = sosend_loan(so, uio, m,
936 						       space)) != 0) {
937 					SOSEND_COUNTER_INCR(&sosend_loan_big);
938 					space -= len;
939 					goto have_data;
940 				}
941 				if (resid >= MINCLSIZE && space >= MCLBYTES) {
942 					SOSEND_COUNTER_INCR(&sosend_copy_big);
943 					m_clget(m, M_WAIT);
944 					if ((m->m_flags & M_EXT) == 0)
945 						goto nopages;
946 					mlen = MCLBYTES;
947 					if (atomic && top == 0) {
948 						len = lmin(MCLBYTES - max_hdr,
949 						    resid);
950 						m->m_data += max_hdr;
951 					} else
952 						len = lmin(MCLBYTES, resid);
953 					space -= len;
954 				} else {
955  nopages:
956 					SOSEND_COUNTER_INCR(&sosend_copy_small);
957 					len = lmin(lmin(mlen, resid), space);
958 					space -= len;
959 					/*
960 					 * For datagram protocols, leave room
961 					 * for protocol headers in first mbuf.
962 					 */
963 					if (atomic && top == 0 && len < mlen)
964 						MH_ALIGN(m, len);
965 				}
966 				error = uiomove(mtod(m, void *), (int)len, uio);
967  have_data:
968 				resid = uio->uio_resid;
969 				m->m_len = len;
970 				*mp = m;
971 				top->m_pkthdr.len += len;
972 				s = splsoftnet();
973 				solock(so);
974 				if (error != 0)
975 					goto release;
976 				mp = &m->m_next;
977 				if (resid <= 0) {
978 					if (flags & MSG_EOR)
979 						top->m_flags |= M_EOR;
980 					break;
981 				}
982 			} while (space > 0 && atomic);
983 
984 			if (so->so_state & SS_CANTSENDMORE) {
985 				error = EPIPE;
986 				goto release;
987 			}
988 			if (dontroute)
989 				so->so_options |= SO_DONTROUTE;
990 			if (resid > 0)
991 				so->so_state |= SS_MORETOCOME;
992 			error = (*so->so_proto->pr_usrreq)(so,
993 			    (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND,
994 			    top, addr, control, curlwp);
995 			if (dontroute)
996 				so->so_options &= ~SO_DONTROUTE;
997 			if (resid > 0)
998 				so->so_state &= ~SS_MORETOCOME;
999 			clen = 0;
1000 			control = NULL;
1001 			top = NULL;
1002 			mp = &top;
1003 			if (error != 0)
1004 				goto release;
1005 		} while (resid && space > 0);
1006 	} while (resid);
1007 
1008  release:
1009 	sbunlock(&so->so_snd);
1010  out:
1011 	sounlock(so);
1012 	splx(s);
1013 	if (top)
1014 		m_freem(top);
1015 	if (control)
1016 		m_freem(control);
1017 	return (error);
1018 }
1019 
1020 /*
1021  * Following replacement or removal of the first mbuf on the first
1022  * mbuf chain of a socket buffer, push necessary state changes back
1023  * into the socket buffer so that other consumers see the values
1024  * consistently.  'nextrecord' is the callers locally stored value of
1025  * the original value of sb->sb_mb->m_nextpkt which must be restored
1026  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
1027  */
1028 static void
1029 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
1030 {
1031 
1032 	KASSERT(solocked(sb->sb_so));
1033 
1034 	/*
1035 	 * First, update for the new value of nextrecord.  If necessary,
1036 	 * make it the first record.
1037 	 */
1038 	if (sb->sb_mb != NULL)
1039 		sb->sb_mb->m_nextpkt = nextrecord;
1040 	else
1041 		sb->sb_mb = nextrecord;
1042 
1043         /*
1044          * Now update any dependent socket buffer fields to reflect
1045          * the new state.  This is an inline of SB_EMPTY_FIXUP, with
1046          * the addition of a second clause that takes care of the
1047          * case where sb_mb has been updated, but remains the last
1048          * record.
1049          */
1050         if (sb->sb_mb == NULL) {
1051                 sb->sb_mbtail = NULL;
1052                 sb->sb_lastrecord = NULL;
1053         } else if (sb->sb_mb->m_nextpkt == NULL)
1054                 sb->sb_lastrecord = sb->sb_mb;
1055 }
1056 
1057 /*
1058  * Implement receive operations on a socket.
1059  * We depend on the way that records are added to the sockbuf
1060  * by sbappend*.  In particular, each record (mbufs linked through m_next)
1061  * must begin with an address if the protocol so specifies,
1062  * followed by an optional mbuf or mbufs containing ancillary data,
1063  * and then zero or more mbufs of data.
1064  * In order to avoid blocking network interrupts for the entire time here,
1065  * we splx() while doing the actual copy to user space.
1066  * Although the sockbuf is locked, new data may still be appended,
1067  * and thus we must maintain consistency of the sockbuf during that time.
1068  *
1069  * The caller may receive the data as a single mbuf chain by supplying
1070  * an mbuf **mp0 for use in returning the chain.  The uio is then used
1071  * only for the count in uio_resid.
1072  */
1073 int
1074 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
1075 	struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1076 {
1077 	struct lwp *l = curlwp;
1078 	struct mbuf	*m, **mp, *mt;
1079 	int atomic, flags, len, error, s, offset, moff, type, orig_resid;
1080 	const struct protosw	*pr;
1081 	struct mbuf	*nextrecord;
1082 	int		mbuf_removed = 0;
1083 	const struct domain *dom;
1084 
1085 	pr = so->so_proto;
1086 	atomic = pr->pr_flags & PR_ATOMIC;
1087 	dom = pr->pr_domain;
1088 	mp = mp0;
1089 	type = 0;
1090 	orig_resid = uio->uio_resid;
1091 
1092 	if (paddr != NULL)
1093 		*paddr = NULL;
1094 	if (controlp != NULL)
1095 		*controlp = NULL;
1096 	if (flagsp != NULL)
1097 		flags = *flagsp &~ MSG_EOR;
1098 	else
1099 		flags = 0;
1100 
1101 	if ((flags & MSG_DONTWAIT) == 0)
1102 		sodopendfree();
1103 
1104 	if (flags & MSG_OOB) {
1105 		m = m_get(M_WAIT, MT_DATA);
1106 		solock(so);
1107 		error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m,
1108 		    (struct mbuf *)(long)(flags & MSG_PEEK), NULL, l);
1109 		sounlock(so);
1110 		if (error)
1111 			goto bad;
1112 		do {
1113 			error = uiomove(mtod(m, void *),
1114 			    (int) min(uio->uio_resid, m->m_len), uio);
1115 			m = m_free(m);
1116 		} while (uio->uio_resid > 0 && error == 0 && m);
1117  bad:
1118 		if (m != NULL)
1119 			m_freem(m);
1120 		return error;
1121 	}
1122 	if (mp != NULL)
1123 		*mp = NULL;
1124 
1125 	/*
1126 	 * solock() provides atomicity of access.  splsoftnet() prevents
1127 	 * protocol processing soft interrupts from interrupting us and
1128 	 * blocking (expensive).
1129 	 */
1130 	s = splsoftnet();
1131 	solock(so);
1132 	if (so->so_state & SS_ISCONFIRMING && uio->uio_resid)
1133 		(*pr->pr_usrreq)(so, PRU_RCVD, NULL, NULL, NULL, l);
1134 
1135  restart:
1136 	if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) {
1137 		sounlock(so);
1138 		splx(s);
1139 		return error;
1140 	}
1141 
1142 	m = so->so_rcv.sb_mb;
1143 	/*
1144 	 * If we have less data than requested, block awaiting more
1145 	 * (subject to any timeout) if:
1146 	 *   1. the current count is less than the low water mark,
1147 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1148 	 *	receive operation at once if we block (resid <= hiwat), or
1149 	 *   3. MSG_DONTWAIT is not set.
1150 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1151 	 * we have to do the receive in sections, and thus risk returning
1152 	 * a short count if a timeout or signal occurs after we start.
1153 	 */
1154 	if (m == NULL ||
1155 	    ((flags & MSG_DONTWAIT) == 0 &&
1156 	     so->so_rcv.sb_cc < uio->uio_resid &&
1157 	     (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1158 	      ((flags & MSG_WAITALL) &&
1159 	       uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1160 	     m->m_nextpkt == NULL && !atomic)) {
1161 #ifdef DIAGNOSTIC
1162 		if (m == NULL && so->so_rcv.sb_cc)
1163 			panic("receive 1");
1164 #endif
1165 		if (so->so_error) {
1166 			if (m != NULL)
1167 				goto dontblock;
1168 			error = so->so_error;
1169 			if ((flags & MSG_PEEK) == 0)
1170 				so->so_error = 0;
1171 			goto release;
1172 		}
1173 		if (so->so_state & SS_CANTRCVMORE) {
1174 			if (m != NULL)
1175 				goto dontblock;
1176 			else
1177 				goto release;
1178 		}
1179 		for (; m != NULL; m = m->m_next)
1180 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1181 				m = so->so_rcv.sb_mb;
1182 				goto dontblock;
1183 			}
1184 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1185 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1186 			error = ENOTCONN;
1187 			goto release;
1188 		}
1189 		if (uio->uio_resid == 0)
1190 			goto release;
1191 		if (so->so_nbio || (flags & MSG_DONTWAIT)) {
1192 			error = EWOULDBLOCK;
1193 			goto release;
1194 		}
1195 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
1196 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
1197 		sbunlock(&so->so_rcv);
1198 		error = sbwait(&so->so_rcv);
1199 		if (error != 0) {
1200 			sounlock(so);
1201 			splx(s);
1202 			return error;
1203 		}
1204 		goto restart;
1205 	}
1206  dontblock:
1207 	/*
1208 	 * On entry here, m points to the first record of the socket buffer.
1209 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1210 	 * pointer to the next record in the socket buffer.  We must keep the
1211 	 * various socket buffer pointers and local stack versions of the
1212 	 * pointers in sync, pushing out modifications before dropping the
1213 	 * socket lock, and re-reading them when picking it up.
1214 	 *
1215 	 * Otherwise, we will race with the network stack appending new data
1216 	 * or records onto the socket buffer by using inconsistent/stale
1217 	 * versions of the field, possibly resulting in socket buffer
1218 	 * corruption.
1219 	 *
1220 	 * By holding the high-level sblock(), we prevent simultaneous
1221 	 * readers from pulling off the front of the socket buffer.
1222 	 */
1223 	if (l != NULL)
1224 		l->l_ru.ru_msgrcv++;
1225 	KASSERT(m == so->so_rcv.sb_mb);
1226 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
1227 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
1228 	nextrecord = m->m_nextpkt;
1229 	if (pr->pr_flags & PR_ADDR) {
1230 #ifdef DIAGNOSTIC
1231 		if (m->m_type != MT_SONAME)
1232 			panic("receive 1a");
1233 #endif
1234 		orig_resid = 0;
1235 		if (flags & MSG_PEEK) {
1236 			if (paddr)
1237 				*paddr = m_copy(m, 0, m->m_len);
1238 			m = m->m_next;
1239 		} else {
1240 			sbfree(&so->so_rcv, m);
1241 			mbuf_removed = 1;
1242 			if (paddr != NULL) {
1243 				*paddr = m;
1244 				so->so_rcv.sb_mb = m->m_next;
1245 				m->m_next = NULL;
1246 				m = so->so_rcv.sb_mb;
1247 			} else {
1248 				MFREE(m, so->so_rcv.sb_mb);
1249 				m = so->so_rcv.sb_mb;
1250 			}
1251 			sbsync(&so->so_rcv, nextrecord);
1252 		}
1253 	}
1254 
1255 	/*
1256 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1257 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1258 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1259 	 * perform externalization (or freeing if controlp == NULL).
1260 	 */
1261 	if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) {
1262 		struct mbuf *cm = NULL, *cmn;
1263 		struct mbuf **cme = &cm;
1264 
1265 		do {
1266 			if (flags & MSG_PEEK) {
1267 				if (controlp != NULL) {
1268 					*controlp = m_copy(m, 0, m->m_len);
1269 					controlp = &(*controlp)->m_next;
1270 				}
1271 				m = m->m_next;
1272 			} else {
1273 				sbfree(&so->so_rcv, m);
1274 				so->so_rcv.sb_mb = m->m_next;
1275 				m->m_next = NULL;
1276 				*cme = m;
1277 				cme = &(*cme)->m_next;
1278 				m = so->so_rcv.sb_mb;
1279 			}
1280 		} while (m != NULL && m->m_type == MT_CONTROL);
1281 		if ((flags & MSG_PEEK) == 0)
1282 			sbsync(&so->so_rcv, nextrecord);
1283 		for (; cm != NULL; cm = cmn) {
1284 			cmn = cm->m_next;
1285 			cm->m_next = NULL;
1286 			type = mtod(cm, struct cmsghdr *)->cmsg_type;
1287 			if (controlp != NULL) {
1288 				if (dom->dom_externalize != NULL &&
1289 				    type == SCM_RIGHTS) {
1290 					sounlock(so);
1291 					splx(s);
1292 					error = (*dom->dom_externalize)(cm, l);
1293 					s = splsoftnet();
1294 					solock(so);
1295 				}
1296 				*controlp = cm;
1297 				while (*controlp != NULL)
1298 					controlp = &(*controlp)->m_next;
1299 			} else {
1300 				/*
1301 				 * Dispose of any SCM_RIGHTS message that went
1302 				 * through the read path rather than recv.
1303 				 */
1304 				if (dom->dom_dispose != NULL &&
1305 				    type == SCM_RIGHTS) {
1306 				    	sounlock(so);
1307 					(*dom->dom_dispose)(cm);
1308 					solock(so);
1309 				}
1310 				m_freem(cm);
1311 			}
1312 		}
1313 		if (m != NULL)
1314 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1315 		else
1316 			nextrecord = so->so_rcv.sb_mb;
1317 		orig_resid = 0;
1318 	}
1319 
1320 	/* If m is non-NULL, we have some data to read. */
1321 	if (__predict_true(m != NULL)) {
1322 		type = m->m_type;
1323 		if (type == MT_OOBDATA)
1324 			flags |= MSG_OOB;
1325 	}
1326 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1327 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1328 
1329 	moff = 0;
1330 	offset = 0;
1331 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1332 		if (m->m_type == MT_OOBDATA) {
1333 			if (type != MT_OOBDATA)
1334 				break;
1335 		} else if (type == MT_OOBDATA)
1336 			break;
1337 #ifdef DIAGNOSTIC
1338 		else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
1339 			panic("receive 3");
1340 #endif
1341 		so->so_state &= ~SS_RCVATMARK;
1342 		len = uio->uio_resid;
1343 		if (so->so_oobmark && len > so->so_oobmark - offset)
1344 			len = so->so_oobmark - offset;
1345 		if (len > m->m_len - moff)
1346 			len = m->m_len - moff;
1347 		/*
1348 		 * If mp is set, just pass back the mbufs.
1349 		 * Otherwise copy them out via the uio, then free.
1350 		 * Sockbuf must be consistent here (points to current mbuf,
1351 		 * it points to next record) when we drop priority;
1352 		 * we must note any additions to the sockbuf when we
1353 		 * block interrupts again.
1354 		 */
1355 		if (mp == NULL) {
1356 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1357 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1358 			sounlock(so);
1359 			splx(s);
1360 			error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1361 			s = splsoftnet();
1362 			solock(so);
1363 			if (error != 0) {
1364 				/*
1365 				 * If any part of the record has been removed
1366 				 * (such as the MT_SONAME mbuf, which will
1367 				 * happen when PR_ADDR, and thus also
1368 				 * PR_ATOMIC, is set), then drop the entire
1369 				 * record to maintain the atomicity of the
1370 				 * receive operation.
1371 				 *
1372 				 * This avoids a later panic("receive 1a")
1373 				 * when compiled with DIAGNOSTIC.
1374 				 */
1375 				if (m && mbuf_removed && atomic)
1376 					(void) sbdroprecord(&so->so_rcv);
1377 
1378 				goto release;
1379 			}
1380 		} else
1381 			uio->uio_resid -= len;
1382 		if (len == m->m_len - moff) {
1383 			if (m->m_flags & M_EOR)
1384 				flags |= MSG_EOR;
1385 			if (flags & MSG_PEEK) {
1386 				m = m->m_next;
1387 				moff = 0;
1388 			} else {
1389 				nextrecord = m->m_nextpkt;
1390 				sbfree(&so->so_rcv, m);
1391 				if (mp) {
1392 					*mp = m;
1393 					mp = &m->m_next;
1394 					so->so_rcv.sb_mb = m = m->m_next;
1395 					*mp = NULL;
1396 				} else {
1397 					MFREE(m, so->so_rcv.sb_mb);
1398 					m = so->so_rcv.sb_mb;
1399 				}
1400 				/*
1401 				 * If m != NULL, we also know that
1402 				 * so->so_rcv.sb_mb != NULL.
1403 				 */
1404 				KASSERT(so->so_rcv.sb_mb == m);
1405 				if (m) {
1406 					m->m_nextpkt = nextrecord;
1407 					if (nextrecord == NULL)
1408 						so->so_rcv.sb_lastrecord = m;
1409 				} else {
1410 					so->so_rcv.sb_mb = nextrecord;
1411 					SB_EMPTY_FIXUP(&so->so_rcv);
1412 				}
1413 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1414 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1415 			}
1416 		} else if (flags & MSG_PEEK)
1417 			moff += len;
1418 		else {
1419 			if (mp != NULL) {
1420 				mt = m_copym(m, 0, len, M_NOWAIT);
1421 				if (__predict_false(mt == NULL)) {
1422 					sounlock(so);
1423 					mt = m_copym(m, 0, len, M_WAIT);
1424 					solock(so);
1425 				}
1426 				*mp = mt;
1427 			}
1428 			m->m_data += len;
1429 			m->m_len -= len;
1430 			so->so_rcv.sb_cc -= len;
1431 		}
1432 		if (so->so_oobmark) {
1433 			if ((flags & MSG_PEEK) == 0) {
1434 				so->so_oobmark -= len;
1435 				if (so->so_oobmark == 0) {
1436 					so->so_state |= SS_RCVATMARK;
1437 					break;
1438 				}
1439 			} else {
1440 				offset += len;
1441 				if (offset == so->so_oobmark)
1442 					break;
1443 			}
1444 		}
1445 		if (flags & MSG_EOR)
1446 			break;
1447 		/*
1448 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1449 		 * we must not quit until "uio->uio_resid == 0" or an error
1450 		 * termination.  If a signal/timeout occurs, return
1451 		 * with a short count but without error.
1452 		 * Keep sockbuf locked against other readers.
1453 		 */
1454 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1455 		    !sosendallatonce(so) && !nextrecord) {
1456 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1457 				break;
1458 			/*
1459 			 * If we are peeking and the socket receive buffer is
1460 			 * full, stop since we can't get more data to peek at.
1461 			 */
1462 			if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
1463 				break;
1464 			/*
1465 			 * If we've drained the socket buffer, tell the
1466 			 * protocol in case it needs to do something to
1467 			 * get it filled again.
1468 			 */
1469 			if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1470 				(*pr->pr_usrreq)(so, PRU_RCVD,
1471 				    NULL, (struct mbuf *)(long)flags, NULL, l);
1472 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1473 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1474 			error = sbwait(&so->so_rcv);
1475 			if (error != 0) {
1476 				sbunlock(&so->so_rcv);
1477 				sounlock(so);
1478 				splx(s);
1479 				return 0;
1480 			}
1481 			if ((m = so->so_rcv.sb_mb) != NULL)
1482 				nextrecord = m->m_nextpkt;
1483 		}
1484 	}
1485 
1486 	if (m && atomic) {
1487 		flags |= MSG_TRUNC;
1488 		if ((flags & MSG_PEEK) == 0)
1489 			(void) sbdroprecord(&so->so_rcv);
1490 	}
1491 	if ((flags & MSG_PEEK) == 0) {
1492 		if (m == NULL) {
1493 			/*
1494 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1495 			 * part makes sure sb_lastrecord is up-to-date if
1496 			 * there is still data in the socket buffer.
1497 			 */
1498 			so->so_rcv.sb_mb = nextrecord;
1499 			if (so->so_rcv.sb_mb == NULL) {
1500 				so->so_rcv.sb_mbtail = NULL;
1501 				so->so_rcv.sb_lastrecord = NULL;
1502 			} else if (nextrecord->m_nextpkt == NULL)
1503 				so->so_rcv.sb_lastrecord = nextrecord;
1504 		}
1505 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1506 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1507 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1508 			(*pr->pr_usrreq)(so, PRU_RCVD, NULL,
1509 			    (struct mbuf *)(long)flags, NULL, l);
1510 	}
1511 	if (orig_resid == uio->uio_resid && orig_resid &&
1512 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1513 		sbunlock(&so->so_rcv);
1514 		goto restart;
1515 	}
1516 
1517 	if (flagsp != NULL)
1518 		*flagsp |= flags;
1519  release:
1520 	sbunlock(&so->so_rcv);
1521 	sounlock(so);
1522 	splx(s);
1523 	return error;
1524 }
1525 
1526 int
1527 soshutdown(struct socket *so, int how)
1528 {
1529 	const struct protosw	*pr;
1530 	int	error;
1531 
1532 	KASSERT(solocked(so));
1533 
1534 	pr = so->so_proto;
1535 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1536 		return (EINVAL);
1537 
1538 	if (how == SHUT_RD || how == SHUT_RDWR) {
1539 		sorflush(so);
1540 		error = 0;
1541 	}
1542 	if (how == SHUT_WR || how == SHUT_RDWR)
1543 		error = (*pr->pr_usrreq)(so, PRU_SHUTDOWN, NULL,
1544 		    NULL, NULL, NULL);
1545 
1546 	return error;
1547 }
1548 
1549 void
1550 sorflush(struct socket *so)
1551 {
1552 	struct sockbuf	*sb, asb;
1553 	const struct protosw	*pr;
1554 
1555 	KASSERT(solocked(so));
1556 
1557 	sb = &so->so_rcv;
1558 	pr = so->so_proto;
1559 	socantrcvmore(so);
1560 	sb->sb_flags |= SB_NOINTR;
1561 	(void )sblock(sb, M_WAITOK);
1562 	sbunlock(sb);
1563 	asb = *sb;
1564 	/*
1565 	 * Clear most of the sockbuf structure, but leave some of the
1566 	 * fields valid.
1567 	 */
1568 	memset(&sb->sb_startzero, 0,
1569 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1570 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) {
1571 		sounlock(so);
1572 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1573 		solock(so);
1574 	}
1575 	sbrelease(&asb, so);
1576 }
1577 
1578 static int
1579 sosetopt1(struct socket *so, int level, int optname, struct mbuf *m)
1580 {
1581 #ifdef INET
1582 	int error, optval, val;
1583 #else
1584 	int optval, val;
1585 #endif
1586 	struct linger	*l;
1587 	struct sockbuf	*sb;
1588 	struct timeval *tv;
1589 
1590 	switch (optname) {
1591 
1592 #ifdef INET
1593 	case SO_ACCEPTFILTER:
1594 		error = do_setopt_accept_filter(so, m);
1595 		if (error)
1596 			return error;
1597 		break;
1598 #endif
1599 
1600 	case SO_LINGER:
1601 		if (m == NULL || m->m_len != sizeof(struct linger))
1602 			return EINVAL;
1603 		l = mtod(m, struct linger *);
1604 		if (l->l_linger < 0 || l->l_linger > USHRT_MAX ||
1605 		    l->l_linger > (INT_MAX / hz))
1606 			return EDOM;
1607 		so->so_linger = l->l_linger;
1608 		if (l->l_onoff)
1609 			so->so_options |= SO_LINGER;
1610 		else
1611 			so->so_options &= ~SO_LINGER;
1612 		break;
1613 
1614 	case SO_DEBUG:
1615 	case SO_KEEPALIVE:
1616 	case SO_DONTROUTE:
1617 	case SO_USELOOPBACK:
1618 	case SO_BROADCAST:
1619 	case SO_REUSEADDR:
1620 	case SO_REUSEPORT:
1621 	case SO_OOBINLINE:
1622 	case SO_TIMESTAMP:
1623 		if (m == NULL || m->m_len < sizeof(int))
1624 			return EINVAL;
1625 		if (*mtod(m, int *))
1626 			so->so_options |= optname;
1627 		else
1628 			so->so_options &= ~optname;
1629 		break;
1630 
1631 	case SO_SNDBUF:
1632 	case SO_RCVBUF:
1633 	case SO_SNDLOWAT:
1634 	case SO_RCVLOWAT:
1635 		if (m == NULL || m->m_len < sizeof(int))
1636 			return EINVAL;
1637 
1638 		/*
1639 		 * Values < 1 make no sense for any of these
1640 		 * options, so disallow them.
1641 		 */
1642 		optval = *mtod(m, int *);
1643 		if (optval < 1)
1644 			return EINVAL;
1645 
1646 		switch (optname) {
1647 
1648 		case SO_SNDBUF:
1649 		case SO_RCVBUF:
1650 			sb = (optname == SO_SNDBUF) ?
1651 			    &so->so_snd : &so->so_rcv;
1652 			if (sbreserve(sb, (u_long)optval, so) == 0)
1653 				return ENOBUFS;
1654 			sb->sb_flags &= ~SB_AUTOSIZE;
1655 			break;
1656 
1657 		/*
1658 		 * Make sure the low-water is never greater than
1659 		 * the high-water.
1660 		 */
1661 		case SO_SNDLOWAT:
1662 			so->so_snd.sb_lowat =
1663 			    (optval > so->so_snd.sb_hiwat) ?
1664 			    so->so_snd.sb_hiwat : optval;
1665 			break;
1666 		case SO_RCVLOWAT:
1667 			so->so_rcv.sb_lowat =
1668 			    (optval > so->so_rcv.sb_hiwat) ?
1669 			    so->so_rcv.sb_hiwat : optval;
1670 			break;
1671 		}
1672 		break;
1673 
1674 	case SO_SNDTIMEO:
1675 	case SO_RCVTIMEO:
1676 		if (m == NULL || m->m_len < sizeof(*tv))
1677 			return EINVAL;
1678 		tv = mtod(m, struct timeval *);
1679 		if (tv->tv_sec > (INT_MAX - tv->tv_usec / tick) / hz)
1680 			return EDOM;
1681 		val = tv->tv_sec * hz + tv->tv_usec / tick;
1682 		if (val == 0 && tv->tv_usec != 0)
1683 			val = 1;
1684 
1685 		switch (optname) {
1686 
1687 		case SO_SNDTIMEO:
1688 			so->so_snd.sb_timeo = val;
1689 			break;
1690 		case SO_RCVTIMEO:
1691 			so->so_rcv.sb_timeo = val;
1692 			break;
1693 		}
1694 		break;
1695 
1696 	default:
1697 		return ENOPROTOOPT;
1698 	}
1699 	return 0;
1700 }
1701 
1702 int
1703 sosetopt(struct socket *so, int level, int optname, struct mbuf *m)
1704 {
1705 	int error, prerr;
1706 
1707 	solock(so);
1708 	if (level == SOL_SOCKET)
1709 		error = sosetopt1(so, level, optname, m);
1710 	else
1711 		error = ENOPROTOOPT;
1712 
1713 	if ((error == 0 || error == ENOPROTOOPT) &&
1714 	    so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) {
1715 		/* give the protocol stack a shot */
1716 		prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, level,
1717 		    optname, &m);
1718 		if (prerr == 0)
1719 			error = 0;
1720 		else if (prerr != ENOPROTOOPT)
1721 			error = prerr;
1722 	} else if (m != NULL)
1723 		(void)m_free(m);
1724 	sounlock(so);
1725 	return error;
1726 }
1727 
1728 int
1729 sogetopt(struct socket *so, int level, int optname, struct mbuf **mp)
1730 {
1731 	struct mbuf	*m;
1732 	int		error;
1733 
1734 	solock(so);
1735 	if (level != SOL_SOCKET) {
1736 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1737 			error = ((*so->so_proto->pr_ctloutput)
1738 				  (PRCO_GETOPT, so, level, optname, mp));
1739 		} else
1740 			error = (ENOPROTOOPT);
1741 	} else {
1742 		m = m_get(M_WAIT, MT_SOOPTS);
1743 		m->m_len = sizeof(int);
1744 
1745 		switch (optname) {
1746 
1747 #ifdef INET
1748 		case SO_ACCEPTFILTER:
1749 			error = do_getopt_accept_filter(so, m);
1750 			break;
1751 #endif
1752 
1753 		case SO_LINGER:
1754 			m->m_len = sizeof(struct linger);
1755 			mtod(m, struct linger *)->l_onoff =
1756 			    (so->so_options & SO_LINGER) ? 1 : 0;
1757 			mtod(m, struct linger *)->l_linger = so->so_linger;
1758 			break;
1759 
1760 		case SO_USELOOPBACK:
1761 		case SO_DONTROUTE:
1762 		case SO_DEBUG:
1763 		case SO_KEEPALIVE:
1764 		case SO_REUSEADDR:
1765 		case SO_REUSEPORT:
1766 		case SO_BROADCAST:
1767 		case SO_OOBINLINE:
1768 		case SO_TIMESTAMP:
1769 			*mtod(m, int *) = (so->so_options & optname) ? 1 : 0;
1770 			break;
1771 
1772 		case SO_TYPE:
1773 			*mtod(m, int *) = so->so_type;
1774 			break;
1775 
1776 		case SO_ERROR:
1777 			*mtod(m, int *) = so->so_error;
1778 			so->so_error = 0;
1779 			break;
1780 
1781 		case SO_SNDBUF:
1782 			*mtod(m, int *) = so->so_snd.sb_hiwat;
1783 			break;
1784 
1785 		case SO_RCVBUF:
1786 			*mtod(m, int *) = so->so_rcv.sb_hiwat;
1787 			break;
1788 
1789 		case SO_SNDLOWAT:
1790 			*mtod(m, int *) = so->so_snd.sb_lowat;
1791 			break;
1792 
1793 		case SO_RCVLOWAT:
1794 			*mtod(m, int *) = so->so_rcv.sb_lowat;
1795 			break;
1796 
1797 		case SO_SNDTIMEO:
1798 		case SO_RCVTIMEO:
1799 		    {
1800 			int val = (optname == SO_SNDTIMEO ?
1801 			     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1802 
1803 			m->m_len = sizeof(struct timeval);
1804 			mtod(m, struct timeval *)->tv_sec = val / hz;
1805 			mtod(m, struct timeval *)->tv_usec =
1806 			    (val % hz) * tick;
1807 			break;
1808 		    }
1809 
1810 		case SO_OVERFLOWED:
1811 			*mtod(m, int *) = so->so_rcv.sb_overflowed;
1812 			break;
1813 
1814 		default:
1815 			sounlock(so);
1816 			(void)m_free(m);
1817 			return (ENOPROTOOPT);
1818 		}
1819 		*mp = m;
1820 		error = 0;
1821 	}
1822 
1823 	sounlock(so);
1824 	return (error);
1825 }
1826 
1827 void
1828 sohasoutofband(struct socket *so)
1829 {
1830 
1831 	fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
1832 	selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, 0);
1833 }
1834 
1835 static void
1836 filt_sordetach(struct knote *kn)
1837 {
1838 	struct socket	*so;
1839 
1840 	so = ((file_t *)kn->kn_obj)->f_data;
1841 	solock(so);
1842 	SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
1843 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
1844 		so->so_rcv.sb_flags &= ~SB_KNOTE;
1845 	sounlock(so);
1846 }
1847 
1848 /*ARGSUSED*/
1849 static int
1850 filt_soread(struct knote *kn, long hint)
1851 {
1852 	struct socket	*so;
1853 	int rv;
1854 
1855 	so = ((file_t *)kn->kn_obj)->f_data;
1856 	if (hint != NOTE_SUBMIT)
1857 		solock(so);
1858 	kn->kn_data = so->so_rcv.sb_cc;
1859 	if (so->so_state & SS_CANTRCVMORE) {
1860 		kn->kn_flags |= EV_EOF;
1861 		kn->kn_fflags = so->so_error;
1862 		rv = 1;
1863 	} else if (so->so_error)	/* temporary udp error */
1864 		rv = 1;
1865 	else if (kn->kn_sfflags & NOTE_LOWAT)
1866 		rv = (kn->kn_data >= kn->kn_sdata);
1867 	else
1868 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
1869 	if (hint != NOTE_SUBMIT)
1870 		sounlock(so);
1871 	return rv;
1872 }
1873 
1874 static void
1875 filt_sowdetach(struct knote *kn)
1876 {
1877 	struct socket	*so;
1878 
1879 	so = ((file_t *)kn->kn_obj)->f_data;
1880 	solock(so);
1881 	SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
1882 	if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
1883 		so->so_snd.sb_flags &= ~SB_KNOTE;
1884 	sounlock(so);
1885 }
1886 
1887 /*ARGSUSED*/
1888 static int
1889 filt_sowrite(struct knote *kn, long hint)
1890 {
1891 	struct socket	*so;
1892 	int rv;
1893 
1894 	so = ((file_t *)kn->kn_obj)->f_data;
1895 	if (hint != NOTE_SUBMIT)
1896 		solock(so);
1897 	kn->kn_data = sbspace(&so->so_snd);
1898 	if (so->so_state & SS_CANTSENDMORE) {
1899 		kn->kn_flags |= EV_EOF;
1900 		kn->kn_fflags = so->so_error;
1901 		rv = 1;
1902 	} else if (so->so_error)	/* temporary udp error */
1903 		rv = 1;
1904 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
1905 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
1906 		rv = 0;
1907 	else if (kn->kn_sfflags & NOTE_LOWAT)
1908 		rv = (kn->kn_data >= kn->kn_sdata);
1909 	else
1910 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
1911 	if (hint != NOTE_SUBMIT)
1912 		sounlock(so);
1913 	return rv;
1914 }
1915 
1916 /*ARGSUSED*/
1917 static int
1918 filt_solisten(struct knote *kn, long hint)
1919 {
1920 	struct socket	*so;
1921 	int rv;
1922 
1923 	so = ((file_t *)kn->kn_obj)->f_data;
1924 
1925 	/*
1926 	 * Set kn_data to number of incoming connections, not
1927 	 * counting partial (incomplete) connections.
1928 	 */
1929 	if (hint != NOTE_SUBMIT)
1930 		solock(so);
1931 	kn->kn_data = so->so_qlen;
1932 	rv = (kn->kn_data > 0);
1933 	if (hint != NOTE_SUBMIT)
1934 		sounlock(so);
1935 	return rv;
1936 }
1937 
1938 static const struct filterops solisten_filtops =
1939 	{ 1, NULL, filt_sordetach, filt_solisten };
1940 static const struct filterops soread_filtops =
1941 	{ 1, NULL, filt_sordetach, filt_soread };
1942 static const struct filterops sowrite_filtops =
1943 	{ 1, NULL, filt_sowdetach, filt_sowrite };
1944 
1945 int
1946 soo_kqfilter(struct file *fp, struct knote *kn)
1947 {
1948 	struct socket	*so;
1949 	struct sockbuf	*sb;
1950 
1951 	so = ((file_t *)kn->kn_obj)->f_data;
1952 	solock(so);
1953 	switch (kn->kn_filter) {
1954 	case EVFILT_READ:
1955 		if (so->so_options & SO_ACCEPTCONN)
1956 			kn->kn_fop = &solisten_filtops;
1957 		else
1958 			kn->kn_fop = &soread_filtops;
1959 		sb = &so->so_rcv;
1960 		break;
1961 	case EVFILT_WRITE:
1962 		kn->kn_fop = &sowrite_filtops;
1963 		sb = &so->so_snd;
1964 		break;
1965 	default:
1966 		sounlock(so);
1967 		return (EINVAL);
1968 	}
1969 	SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
1970 	sb->sb_flags |= SB_KNOTE;
1971 	sounlock(so);
1972 	return (0);
1973 }
1974 
1975 static int
1976 sodopoll(struct socket *so, int events)
1977 {
1978 	int revents;
1979 
1980 	revents = 0;
1981 
1982 	if (events & (POLLIN | POLLRDNORM))
1983 		if (soreadable(so))
1984 			revents |= events & (POLLIN | POLLRDNORM);
1985 
1986 	if (events & (POLLOUT | POLLWRNORM))
1987 		if (sowritable(so))
1988 			revents |= events & (POLLOUT | POLLWRNORM);
1989 
1990 	if (events & (POLLPRI | POLLRDBAND))
1991 		if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
1992 			revents |= events & (POLLPRI | POLLRDBAND);
1993 
1994 	return revents;
1995 }
1996 
1997 int
1998 sopoll(struct socket *so, int events)
1999 {
2000 	int revents = 0;
2001 
2002 #ifndef DIAGNOSTIC
2003 	/*
2004 	 * Do a quick, unlocked check in expectation that the socket
2005 	 * will be ready for I/O.  Don't do this check if DIAGNOSTIC,
2006 	 * as the solocked() assertions will fail.
2007 	 */
2008 	if ((revents = sodopoll(so, events)) != 0)
2009 		return revents;
2010 #endif
2011 
2012 	solock(so);
2013 	if ((revents = sodopoll(so, events)) == 0) {
2014 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2015 			selrecord(curlwp, &so->so_rcv.sb_sel);
2016 			so->so_rcv.sb_flags |= SB_NOTIFY;
2017 		}
2018 
2019 		if (events & (POLLOUT | POLLWRNORM)) {
2020 			selrecord(curlwp, &so->so_snd.sb_sel);
2021 			so->so_snd.sb_flags |= SB_NOTIFY;
2022 		}
2023 	}
2024 	sounlock(so);
2025 
2026 	return revents;
2027 }
2028 
2029 
2030 #include <sys/sysctl.h>
2031 
2032 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO);
2033 
2034 /*
2035  * sysctl helper routine for kern.somaxkva.  ensures that the given
2036  * value is not too small.
2037  * (XXX should we maybe make sure it's not too large as well?)
2038  */
2039 static int
2040 sysctl_kern_somaxkva(SYSCTLFN_ARGS)
2041 {
2042 	int error, new_somaxkva;
2043 	struct sysctlnode node;
2044 
2045 	new_somaxkva = somaxkva;
2046 	node = *rnode;
2047 	node.sysctl_data = &new_somaxkva;
2048 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2049 	if (error || newp == NULL)
2050 		return (error);
2051 
2052 	if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */
2053 		return (EINVAL);
2054 
2055 	mutex_enter(&so_pendfree_lock);
2056 	somaxkva = new_somaxkva;
2057 	cv_broadcast(&socurkva_cv);
2058 	mutex_exit(&so_pendfree_lock);
2059 
2060 	return (error);
2061 }
2062 
2063 SYSCTL_SETUP(sysctl_kern_somaxkva_setup, "sysctl kern.somaxkva setup")
2064 {
2065 
2066 	sysctl_createv(clog, 0, NULL, NULL,
2067 		       CTLFLAG_PERMANENT,
2068 		       CTLTYPE_NODE, "kern", NULL,
2069 		       NULL, 0, NULL, 0,
2070 		       CTL_KERN, CTL_EOL);
2071 
2072 	sysctl_createv(clog, 0, NULL, NULL,
2073 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2074 		       CTLTYPE_INT, "somaxkva",
2075 		       SYSCTL_DESCR("Maximum amount of kernel memory to be "
2076 				    "used for socket buffers"),
2077 		       sysctl_kern_somaxkva, 0, NULL, 0,
2078 		       CTL_KERN, KERN_SOMAXKVA, CTL_EOL);
2079 }
2080