xref: /netbsd-src/sys/kern/uipc_socket.c (revision 0c4ddb1599a0bea866fde8522a74cfbd2f68cd1b)
1 /*	$NetBSD: uipc_socket.c,v 1.168 2008/06/18 09:06:27 yamt 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 1.168 2008/06/18 09:06:27 yamt Exp $");
67 
68 #include "opt_sock_counters.h"
69 #include "opt_sosend_loan.h"
70 #include "opt_mbuftrace.h"
71 #include "opt_somaxkva.h"
72 #include "opt_multiprocessor.h"	/* XXX */
73 
74 #include <sys/param.h>
75 #include <sys/systm.h>
76 #include <sys/proc.h>
77 #include <sys/file.h>
78 #include <sys/filedesc.h>
79 #include <sys/malloc.h>
80 #include <sys/mbuf.h>
81 #include <sys/domain.h>
82 #include <sys/kernel.h>
83 #include <sys/protosw.h>
84 #include <sys/socket.h>
85 #include <sys/socketvar.h>
86 #include <sys/signalvar.h>
87 #include <sys/resourcevar.h>
88 #include <sys/event.h>
89 #include <sys/poll.h>
90 #include <sys/kauth.h>
91 #include <sys/mutex.h>
92 #include <sys/condvar.h>
93 
94 #include <uvm/uvm.h>
95 
96 MALLOC_DEFINE(M_SOOPTS, "soopts", "socket options");
97 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
98 
99 extern const struct fileops socketops;
100 
101 extern int	somaxconn;			/* patchable (XXX sysctl) */
102 int		somaxconn = SOMAXCONN;
103 kmutex_t	*softnet_lock;
104 
105 #ifdef SOSEND_COUNTERS
106 #include <sys/device.h>
107 
108 static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
109     NULL, "sosend", "loan big");
110 static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
111     NULL, "sosend", "copy big");
112 static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
113     NULL, "sosend", "copy small");
114 static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
115     NULL, "sosend", "kva limit");
116 
117 #define	SOSEND_COUNTER_INCR(ev)		(ev)->ev_count++
118 
119 EVCNT_ATTACH_STATIC(sosend_loan_big);
120 EVCNT_ATTACH_STATIC(sosend_copy_big);
121 EVCNT_ATTACH_STATIC(sosend_copy_small);
122 EVCNT_ATTACH_STATIC(sosend_kvalimit);
123 #else
124 
125 #define	SOSEND_COUNTER_INCR(ev)		/* nothing */
126 
127 #endif /* SOSEND_COUNTERS */
128 
129 static struct callback_entry sokva_reclaimerentry;
130 
131 #if defined(SOSEND_NO_LOAN) || defined(MULTIPROCESSOR)
132 int sock_loan_thresh = -1;
133 #else
134 int sock_loan_thresh = 4096;
135 #endif
136 
137 static kmutex_t so_pendfree_lock;
138 static struct mbuf *so_pendfree;
139 
140 #ifndef SOMAXKVA
141 #define	SOMAXKVA (16 * 1024 * 1024)
142 #endif
143 int somaxkva = SOMAXKVA;
144 static int socurkva;
145 static kcondvar_t socurkva_cv;
146 
147 #define	SOCK_LOAN_CHUNK		65536
148 
149 static size_t sodopendfree(void);
150 static size_t sodopendfreel(void);
151 
152 static vsize_t
153 sokvareserve(struct socket *so, vsize_t len)
154 {
155 	int error;
156 
157 	mutex_enter(&so_pendfree_lock);
158 	while (socurkva + len > somaxkva) {
159 		size_t freed;
160 
161 		/*
162 		 * try to do pendfree.
163 		 */
164 
165 		freed = sodopendfreel();
166 
167 		/*
168 		 * if some kva was freed, try again.
169 		 */
170 
171 		if (freed)
172 			continue;
173 
174 		SOSEND_COUNTER_INCR(&sosend_kvalimit);
175 		error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock);
176 		if (error) {
177 			len = 0;
178 			break;
179 		}
180 	}
181 	socurkva += len;
182 	mutex_exit(&so_pendfree_lock);
183 	return len;
184 }
185 
186 static void
187 sokvaunreserve(vsize_t len)
188 {
189 
190 	mutex_enter(&so_pendfree_lock);
191 	socurkva -= len;
192 	cv_broadcast(&socurkva_cv);
193 	mutex_exit(&so_pendfree_lock);
194 }
195 
196 /*
197  * sokvaalloc: allocate kva for loan.
198  */
199 
200 vaddr_t
201 sokvaalloc(vsize_t len, struct socket *so)
202 {
203 	vaddr_t lva;
204 
205 	/*
206 	 * reserve kva.
207 	 */
208 
209 	if (sokvareserve(so, len) == 0)
210 		return 0;
211 
212 	/*
213 	 * allocate kva.
214 	 */
215 
216 	lva = uvm_km_alloc(kernel_map, len, 0, UVM_KMF_VAONLY | UVM_KMF_WAITVA);
217 	if (lva == 0) {
218 		sokvaunreserve(len);
219 		return (0);
220 	}
221 
222 	return lva;
223 }
224 
225 /*
226  * sokvafree: free kva for loan.
227  */
228 
229 void
230 sokvafree(vaddr_t sva, vsize_t len)
231 {
232 
233 	/*
234 	 * free kva.
235 	 */
236 
237 	uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY);
238 
239 	/*
240 	 * unreserve kva.
241 	 */
242 
243 	sokvaunreserve(len);
244 }
245 
246 static void
247 sodoloanfree(struct vm_page **pgs, void *buf, size_t size)
248 {
249 	vaddr_t sva, eva;
250 	vsize_t len;
251 	int npgs;
252 
253 	KASSERT(pgs != NULL);
254 
255 	eva = round_page((vaddr_t) buf + size);
256 	sva = trunc_page((vaddr_t) buf);
257 	len = eva - sva;
258 	npgs = len >> PAGE_SHIFT;
259 
260 	pmap_kremove(sva, len);
261 	pmap_update(pmap_kernel());
262 	uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE);
263 	sokvafree(sva, len);
264 }
265 
266 static size_t
267 sodopendfree(void)
268 {
269 	size_t rv;
270 
271 	if (__predict_true(so_pendfree == NULL))
272 		return 0;
273 
274 	mutex_enter(&so_pendfree_lock);
275 	rv = sodopendfreel();
276 	mutex_exit(&so_pendfree_lock);
277 
278 	return rv;
279 }
280 
281 /*
282  * sodopendfreel: free mbufs on "pendfree" list.
283  * unlock and relock so_pendfree_lock when freeing mbufs.
284  *
285  * => called with so_pendfree_lock held.
286  */
287 
288 static size_t
289 sodopendfreel(void)
290 {
291 	struct mbuf *m, *next;
292 	size_t rv = 0;
293 
294 	KASSERT(mutex_owned(&so_pendfree_lock));
295 
296 	while (so_pendfree != NULL) {
297 		m = so_pendfree;
298 		so_pendfree = NULL;
299 		mutex_exit(&so_pendfree_lock);
300 
301 		for (; m != NULL; m = next) {
302 			next = m->m_next;
303 			KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0);
304 			KASSERT(m->m_ext.ext_refcnt == 0);
305 
306 			rv += m->m_ext.ext_size;
307 			sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf,
308 			    m->m_ext.ext_size);
309 			pool_cache_put(mb_cache, m);
310 		}
311 
312 		mutex_enter(&so_pendfree_lock);
313 	}
314 
315 	return (rv);
316 }
317 
318 void
319 soloanfree(struct mbuf *m, void *buf, size_t size, void *arg)
320 {
321 
322 	KASSERT(m != NULL);
323 
324 	/*
325 	 * postpone freeing mbuf.
326 	 *
327 	 * we can't do it in interrupt context
328 	 * because we need to put kva back to kernel_map.
329 	 */
330 
331 	mutex_enter(&so_pendfree_lock);
332 	m->m_next = so_pendfree;
333 	so_pendfree = m;
334 	cv_broadcast(&socurkva_cv);
335 	mutex_exit(&so_pendfree_lock);
336 }
337 
338 static long
339 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space)
340 {
341 	struct iovec *iov = uio->uio_iov;
342 	vaddr_t sva, eva;
343 	vsize_t len;
344 	vaddr_t lva;
345 	int npgs, error;
346 	vaddr_t va;
347 	int i;
348 
349 	if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace))
350 		return (0);
351 
352 	if (iov->iov_len < (size_t) space)
353 		space = iov->iov_len;
354 	if (space > SOCK_LOAN_CHUNK)
355 		space = SOCK_LOAN_CHUNK;
356 
357 	eva = round_page((vaddr_t) iov->iov_base + space);
358 	sva = trunc_page((vaddr_t) iov->iov_base);
359 	len = eva - sva;
360 	npgs = len >> PAGE_SHIFT;
361 
362 	KASSERT(npgs <= M_EXT_MAXPAGES);
363 
364 	lva = sokvaalloc(len, so);
365 	if (lva == 0)
366 		return 0;
367 
368 	error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len,
369 	    m->m_ext.ext_pgs, UVM_LOAN_TOPAGE);
370 	if (error) {
371 		sokvafree(lva, len);
372 		return (0);
373 	}
374 
375 	for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE)
376 		pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
377 		    VM_PROT_READ);
378 	pmap_update(pmap_kernel());
379 
380 	lva += (vaddr_t) iov->iov_base & PAGE_MASK;
381 
382 	MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so);
383 	m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
384 
385 	uio->uio_resid -= space;
386 	/* uio_offset not updated, not set/used for write(2) */
387 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space;
388 	uio->uio_iov->iov_len -= space;
389 	if (uio->uio_iov->iov_len == 0) {
390 		uio->uio_iov++;
391 		uio->uio_iovcnt--;
392 	}
393 
394 	return (space);
395 }
396 
397 static int
398 sokva_reclaim_callback(struct callback_entry *ce, void *obj, void *arg)
399 {
400 
401 	KASSERT(ce == &sokva_reclaimerentry);
402 	KASSERT(obj == NULL);
403 
404 	sodopendfree();
405 	if (!vm_map_starved_p(kernel_map)) {
406 		return CALLBACK_CHAIN_ABORT;
407 	}
408 	return CALLBACK_CHAIN_CONTINUE;
409 }
410 
411 struct mbuf *
412 getsombuf(struct socket *so, int type)
413 {
414 	struct mbuf *m;
415 
416 	m = m_get(M_WAIT, type);
417 	MCLAIM(m, so->so_mowner);
418 	return m;
419 }
420 
421 struct mbuf *
422 m_intopt(struct socket *so, int val)
423 {
424 	struct mbuf *m;
425 
426 	m = getsombuf(so, MT_SOOPTS);
427 	m->m_len = sizeof(int);
428 	*mtod(m, int *) = val;
429 	return m;
430 }
431 
432 void
433 soinit(void)
434 {
435 
436 	mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM);
437 	softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
438 	cv_init(&socurkva_cv, "sokva");
439 	soinit2();
440 
441 	/* Set the initial adjusted socket buffer size. */
442 	if (sb_max_set(sb_max))
443 		panic("bad initial sb_max value: %lu", sb_max);
444 
445 	callback_register(&vm_map_to_kernel(kernel_map)->vmk_reclaim_callback,
446 	    &sokva_reclaimerentry, NULL, sokva_reclaim_callback);
447 }
448 
449 /*
450  * Socket operation routines.
451  * These routines are called by the routines in
452  * sys_socket.c or from a system process, and
453  * implement the semantics of socket operations by
454  * switching out to the protocol specific routines.
455  */
456 /*ARGSUSED*/
457 int
458 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l,
459 	 struct socket *lockso)
460 {
461 	const struct protosw	*prp;
462 	struct socket	*so;
463 	uid_t		uid;
464 	int		error;
465 	kmutex_t	*lock;
466 
467 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET,
468 	    KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type),
469 	    KAUTH_ARG(proto));
470 	if (error != 0)
471 		return error;
472 
473 	if (proto)
474 		prp = pffindproto(dom, proto, type);
475 	else
476 		prp = pffindtype(dom, type);
477 	if (prp == NULL) {
478 		/* no support for domain */
479 		if (pffinddomain(dom) == 0)
480 			return EAFNOSUPPORT;
481 		/* no support for socket type */
482 		if (proto == 0 && type != 0)
483 			return EPROTOTYPE;
484 		return EPROTONOSUPPORT;
485 	}
486 	if (prp->pr_usrreq == NULL)
487 		return EPROTONOSUPPORT;
488 	if (prp->pr_type != type)
489 		return EPROTOTYPE;
490 
491 	so = soget(true);
492 	so->so_type = type;
493 	so->so_proto = prp;
494 	so->so_send = sosend;
495 	so->so_receive = soreceive;
496 #ifdef MBUFTRACE
497 	so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner;
498 	so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner;
499 	so->so_mowner = &prp->pr_domain->dom_mowner;
500 #endif
501 	uid = kauth_cred_geteuid(l->l_cred);
502 	so->so_uidinfo = uid_find(uid);
503 	so->so_egid = kauth_cred_getegid(l->l_cred);
504 	so->so_cpid = l->l_proc->p_pid;
505 	if (lockso != NULL) {
506 		/* Caller wants us to share a lock. */
507 		lock = lockso->so_lock;
508 		so->so_lock = lock;
509 		mutex_obj_hold(lock);
510 		mutex_enter(lock);
511 	} else {
512 		/* Lock assigned and taken during PRU_ATTACH. */
513 	}
514 	error = (*prp->pr_usrreq)(so, PRU_ATTACH, NULL,
515 	    (struct mbuf *)(long)proto, NULL, l);
516 	KASSERT(solocked(so));
517 	if (error != 0) {
518 		so->so_state |= SS_NOFDREF;
519 		sofree(so);
520 		return error;
521 	}
522 	sounlock(so);
523 	*aso = so;
524 	return 0;
525 }
526 
527 /* On success, write file descriptor to fdout and return zero.  On
528  * failure, return non-zero; *fdout will be undefined.
529  */
530 int
531 fsocreate(int domain, struct socket **sop, int type, int protocol,
532     struct lwp *l, int *fdout)
533 {
534 	struct socket	*so;
535 	struct file	*fp;
536 	int		fd, error;
537 
538 	if ((error = fd_allocfile(&fp, &fd)) != 0)
539 		return (error);
540 	fp->f_flag = FREAD|FWRITE;
541 	fp->f_type = DTYPE_SOCKET;
542 	fp->f_ops = &socketops;
543 	error = socreate(domain, &so, type, protocol, l, NULL);
544 	if (error != 0) {
545 		fd_abort(curproc, fp, fd);
546 	} else {
547 		if (sop != NULL)
548 			*sop = so;
549 		fp->f_data = so;
550 		fd_affix(curproc, fp, fd);
551 		*fdout = fd;
552 	}
553 	return error;
554 }
555 
556 int
557 sobind(struct socket *so, struct mbuf *nam, struct lwp *l)
558 {
559 	int	error;
560 
561 	solock(so);
562 	error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, NULL, nam, NULL, l);
563 	sounlock(so);
564 	return error;
565 }
566 
567 int
568 solisten(struct socket *so, int backlog, struct lwp *l)
569 {
570 	int	error;
571 
572 	solock(so);
573 	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
574 	    SS_ISDISCONNECTING)) != 0) {
575 	    	sounlock(so);
576 		return (EOPNOTSUPP);
577 	}
578 	error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, NULL,
579 	    NULL, NULL, l);
580 	if (error != 0) {
581 		sounlock(so);
582 		return error;
583 	}
584 	if (TAILQ_EMPTY(&so->so_q))
585 		so->so_options |= SO_ACCEPTCONN;
586 	if (backlog < 0)
587 		backlog = 0;
588 	so->so_qlimit = min(backlog, somaxconn);
589 	sounlock(so);
590 	return 0;
591 }
592 
593 void
594 sofree(struct socket *so)
595 {
596 	u_int refs;
597 
598 	KASSERT(solocked(so));
599 
600 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
601 		sounlock(so);
602 		return;
603 	}
604 	if (so->so_head) {
605 		/*
606 		 * We must not decommission a socket that's on the accept(2)
607 		 * queue.  If we do, then accept(2) may hang after select(2)
608 		 * indicated that the listening socket was ready.
609 		 */
610 		if (!soqremque(so, 0)) {
611 			sounlock(so);
612 			return;
613 		}
614 	}
615 	if (so->so_rcv.sb_hiwat)
616 		(void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0,
617 		    RLIM_INFINITY);
618 	if (so->so_snd.sb_hiwat)
619 		(void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0,
620 		    RLIM_INFINITY);
621 	sbrelease(&so->so_snd, so);
622 	KASSERT(!cv_has_waiters(&so->so_cv));
623 	KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv));
624 	KASSERT(!cv_has_waiters(&so->so_snd.sb_cv));
625 	sorflush(so);
626 	refs = so->so_aborting;	/* XXX */
627 	sounlock(so);
628 	if (refs == 0)		/* XXX */
629 		soput(so);
630 }
631 
632 /*
633  * Close a socket on last file table reference removal.
634  * Initiate disconnect if connected.
635  * Free socket when disconnect complete.
636  */
637 int
638 soclose(struct socket *so)
639 {
640 	struct socket	*so2;
641 	int		error;
642 	int		error2;
643 
644 	error = 0;
645 	solock(so);
646 	if (so->so_options & SO_ACCEPTCONN) {
647 		do {
648 			if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) {
649 				KASSERT(solocked2(so, so2));
650 				(void) soqremque(so2, 0);
651 				/* soabort drops the lock. */
652 				(void) soabort(so2);
653 				solock(so);
654 				continue;
655 			}
656 			if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) {
657 				KASSERT(solocked2(so, so2));
658 				(void) soqremque(so2, 1);
659 				/* soabort drops the lock. */
660 				(void) soabort(so2);
661 				solock(so);
662 				continue;
663 			}
664 		} while (0);
665 	}
666 	if (so->so_pcb == 0)
667 		goto discard;
668 	if (so->so_state & SS_ISCONNECTED) {
669 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
670 			error = sodisconnect(so);
671 			if (error)
672 				goto drop;
673 		}
674 		if (so->so_options & SO_LINGER) {
675 			if ((so->so_state & SS_ISDISCONNECTING) && so->so_nbio)
676 				goto drop;
677 			while (so->so_state & SS_ISCONNECTED) {
678 				error = sowait(so, so->so_linger * hz);
679 				if (error)
680 					break;
681 			}
682 		}
683 	}
684  drop:
685 	if (so->so_pcb) {
686 		error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH,
687 		    NULL, NULL, NULL, NULL);
688 		if (error == 0)
689 			error = error2;
690 	}
691  discard:
692 	if (so->so_state & SS_NOFDREF)
693 		panic("soclose: NOFDREF");
694 	so->so_state |= SS_NOFDREF;
695 	sofree(so);
696 	return (error);
697 }
698 
699 /*
700  * Must be called with the socket locked..  Will return with it unlocked.
701  */
702 int
703 soabort(struct socket *so)
704 {
705 	u_int refs;
706 	int error;
707 
708 	KASSERT(solocked(so));
709 	KASSERT(so->so_head == NULL);
710 
711 	so->so_aborting++;		/* XXX */
712 	error = (*so->so_proto->pr_usrreq)(so, PRU_ABORT, NULL,
713 	    NULL, NULL, NULL);
714 	refs = --so->so_aborting;	/* XXX */
715 	if (error || (refs == 0)) {
716 		sofree(so);
717 	} else {
718 		sounlock(so);
719 	}
720 	return error;
721 }
722 
723 int
724 soaccept(struct socket *so, struct mbuf *nam)
725 {
726 	int	error;
727 
728 	KASSERT(solocked(so));
729 
730 	error = 0;
731 	if ((so->so_state & SS_NOFDREF) == 0)
732 		panic("soaccept: !NOFDREF");
733 	so->so_state &= ~SS_NOFDREF;
734 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
735 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
736 		error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT,
737 		    NULL, nam, NULL, NULL);
738 	else
739 		error = ECONNABORTED;
740 
741 	return (error);
742 }
743 
744 int
745 soconnect(struct socket *so, struct mbuf *nam, struct lwp *l)
746 {
747 	int		error;
748 
749 	KASSERT(solocked(so));
750 
751 	if (so->so_options & SO_ACCEPTCONN)
752 		return (EOPNOTSUPP);
753 	/*
754 	 * If protocol is connection-based, can only connect once.
755 	 * Otherwise, if connected, try to disconnect first.
756 	 * This allows user to disconnect by connecting to, e.g.,
757 	 * a null address.
758 	 */
759 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
760 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
761 	    (error = sodisconnect(so))))
762 		error = EISCONN;
763 	else
764 		error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT,
765 		    NULL, nam, NULL, l);
766 	return (error);
767 }
768 
769 int
770 soconnect2(struct socket *so1, struct socket *so2)
771 {
772 	int	error;
773 
774 	KASSERT(solocked2(so1, so2));
775 
776 	error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2,
777 	    NULL, (struct mbuf *)so2, NULL, NULL);
778 	return (error);
779 }
780 
781 int
782 sodisconnect(struct socket *so)
783 {
784 	int	error;
785 
786 	KASSERT(solocked(so));
787 
788 	if ((so->so_state & SS_ISCONNECTED) == 0) {
789 		error = ENOTCONN;
790 	} else if (so->so_state & SS_ISDISCONNECTING) {
791 		error = EALREADY;
792 	} else {
793 		error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT,
794 		    NULL, NULL, NULL, NULL);
795 	}
796 	sodopendfree();
797 	return (error);
798 }
799 
800 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
801 /*
802  * Send on a socket.
803  * If send must go all at once and message is larger than
804  * send buffering, then hard error.
805  * Lock against other senders.
806  * If must go all at once and not enough room now, then
807  * inform user that this would block and do nothing.
808  * Otherwise, if nonblocking, send as much as possible.
809  * The data to be sent is described by "uio" if nonzero,
810  * otherwise by the mbuf chain "top" (which must be null
811  * if uio is not).  Data provided in mbuf chain must be small
812  * enough to send all at once.
813  *
814  * Returns nonzero on error, timeout or signal; callers
815  * must check for short counts if EINTR/ERESTART are returned.
816  * Data and control buffers are freed on return.
817  */
818 int
819 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
820 	struct mbuf *control, int flags, struct lwp *l)
821 {
822 	struct mbuf	**mp, *m;
823 	struct proc	*p;
824 	long		space, len, resid, clen, mlen;
825 	int		error, s, dontroute, atomic;
826 
827 	p = l->l_proc;
828 	sodopendfree();
829 	clen = 0;
830 
831 	/*
832 	 * solock() provides atomicity of access.  splsoftnet() prevents
833 	 * protocol processing soft interrupts from interrupting us and
834 	 * blocking (expensive).
835 	 */
836 	s = splsoftnet();
837 	solock(so);
838 	atomic = sosendallatonce(so) || top;
839 	if (uio)
840 		resid = uio->uio_resid;
841 	else
842 		resid = top->m_pkthdr.len;
843 	/*
844 	 * In theory resid should be unsigned.
845 	 * However, space must be signed, as it might be less than 0
846 	 * if we over-committed, and we must use a signed comparison
847 	 * of space and resid.  On the other hand, a negative resid
848 	 * causes us to loop sending 0-length segments to the protocol.
849 	 */
850 	if (resid < 0) {
851 		error = EINVAL;
852 		goto out;
853 	}
854 	dontroute =
855 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
856 	    (so->so_proto->pr_flags & PR_ATOMIC);
857 	l->l_ru.ru_msgsnd++;
858 	if (control)
859 		clen = control->m_len;
860  restart:
861 	if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
862 		goto out;
863 	do {
864 		if (so->so_state & SS_CANTSENDMORE) {
865 			error = EPIPE;
866 			goto release;
867 		}
868 		if (so->so_error) {
869 			error = so->so_error;
870 			so->so_error = 0;
871 			goto release;
872 		}
873 		if ((so->so_state & SS_ISCONNECTED) == 0) {
874 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
875 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
876 				    !(resid == 0 && clen != 0)) {
877 					error = ENOTCONN;
878 					goto release;
879 				}
880 			} else if (addr == 0) {
881 				error = EDESTADDRREQ;
882 				goto release;
883 			}
884 		}
885 		space = sbspace(&so->so_snd);
886 		if (flags & MSG_OOB)
887 			space += 1024;
888 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
889 		    clen > so->so_snd.sb_hiwat) {
890 			error = EMSGSIZE;
891 			goto release;
892 		}
893 		if (space < resid + clen &&
894 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
895 			if (so->so_nbio) {
896 				error = EWOULDBLOCK;
897 				goto release;
898 			}
899 			sbunlock(&so->so_snd);
900 			error = sbwait(&so->so_snd);
901 			if (error)
902 				goto out;
903 			goto restart;
904 		}
905 		mp = &top;
906 		space -= clen;
907 		do {
908 			if (uio == NULL) {
909 				/*
910 				 * Data is prepackaged in "top".
911 				 */
912 				resid = 0;
913 				if (flags & MSG_EOR)
914 					top->m_flags |= M_EOR;
915 			} else do {
916 				sounlock(so);
917 				splx(s);
918 				if (top == NULL) {
919 					m = m_gethdr(M_WAIT, MT_DATA);
920 					mlen = MHLEN;
921 					m->m_pkthdr.len = 0;
922 					m->m_pkthdr.rcvif = NULL;
923 				} else {
924 					m = m_get(M_WAIT, MT_DATA);
925 					mlen = MLEN;
926 				}
927 				MCLAIM(m, so->so_snd.sb_mowner);
928 				if (sock_loan_thresh >= 0 &&
929 				    uio->uio_iov->iov_len >= sock_loan_thresh &&
930 				    space >= sock_loan_thresh &&
931 				    (len = sosend_loan(so, uio, m,
932 						       space)) != 0) {
933 					SOSEND_COUNTER_INCR(&sosend_loan_big);
934 					space -= len;
935 					goto have_data;
936 				}
937 				if (resid >= MINCLSIZE && space >= MCLBYTES) {
938 					SOSEND_COUNTER_INCR(&sosend_copy_big);
939 					m_clget(m, M_WAIT);
940 					if ((m->m_flags & M_EXT) == 0)
941 						goto nopages;
942 					mlen = MCLBYTES;
943 					if (atomic && top == 0) {
944 						len = lmin(MCLBYTES - max_hdr,
945 						    resid);
946 						m->m_data += max_hdr;
947 					} else
948 						len = lmin(MCLBYTES, resid);
949 					space -= len;
950 				} else {
951  nopages:
952 					SOSEND_COUNTER_INCR(&sosend_copy_small);
953 					len = lmin(lmin(mlen, resid), space);
954 					space -= len;
955 					/*
956 					 * For datagram protocols, leave room
957 					 * for protocol headers in first mbuf.
958 					 */
959 					if (atomic && top == 0 && len < mlen)
960 						MH_ALIGN(m, len);
961 				}
962 				error = uiomove(mtod(m, void *), (int)len, uio);
963  have_data:
964 				resid = uio->uio_resid;
965 				m->m_len = len;
966 				*mp = m;
967 				top->m_pkthdr.len += len;
968 				s = splsoftnet();
969 				solock(so);
970 				if (error != 0)
971 					goto release;
972 				mp = &m->m_next;
973 				if (resid <= 0) {
974 					if (flags & MSG_EOR)
975 						top->m_flags |= M_EOR;
976 					break;
977 				}
978 			} while (space > 0 && atomic);
979 
980 			if (so->so_state & SS_CANTSENDMORE) {
981 				error = EPIPE;
982 				goto release;
983 			}
984 			if (dontroute)
985 				so->so_options |= SO_DONTROUTE;
986 			if (resid > 0)
987 				so->so_state |= SS_MORETOCOME;
988 			error = (*so->so_proto->pr_usrreq)(so,
989 			    (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND,
990 			    top, addr, control, curlwp);
991 			if (dontroute)
992 				so->so_options &= ~SO_DONTROUTE;
993 			if (resid > 0)
994 				so->so_state &= ~SS_MORETOCOME;
995 			clen = 0;
996 			control = NULL;
997 			top = NULL;
998 			mp = &top;
999 			if (error != 0)
1000 				goto release;
1001 		} while (resid && space > 0);
1002 	} while (resid);
1003 
1004  release:
1005 	sbunlock(&so->so_snd);
1006  out:
1007 	sounlock(so);
1008 	splx(s);
1009 	if (top)
1010 		m_freem(top);
1011 	if (control)
1012 		m_freem(control);
1013 	return (error);
1014 }
1015 
1016 /*
1017  * Following replacement or removal of the first mbuf on the first
1018  * mbuf chain of a socket buffer, push necessary state changes back
1019  * into the socket buffer so that other consumers see the values
1020  * consistently.  'nextrecord' is the callers locally stored value of
1021  * the original value of sb->sb_mb->m_nextpkt which must be restored
1022  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
1023  */
1024 static void
1025 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
1026 {
1027 
1028 	KASSERT(solocked(sb->sb_so));
1029 
1030 	/*
1031 	 * First, update for the new value of nextrecord.  If necessary,
1032 	 * make it the first record.
1033 	 */
1034 	if (sb->sb_mb != NULL)
1035 		sb->sb_mb->m_nextpkt = nextrecord;
1036 	else
1037 		sb->sb_mb = nextrecord;
1038 
1039         /*
1040          * Now update any dependent socket buffer fields to reflect
1041          * the new state.  This is an inline of SB_EMPTY_FIXUP, with
1042          * the addition of a second clause that takes care of the
1043          * case where sb_mb has been updated, but remains the last
1044          * record.
1045          */
1046         if (sb->sb_mb == NULL) {
1047                 sb->sb_mbtail = NULL;
1048                 sb->sb_lastrecord = NULL;
1049         } else if (sb->sb_mb->m_nextpkt == NULL)
1050                 sb->sb_lastrecord = sb->sb_mb;
1051 }
1052 
1053 /*
1054  * Implement receive operations on a socket.
1055  * We depend on the way that records are added to the sockbuf
1056  * by sbappend*.  In particular, each record (mbufs linked through m_next)
1057  * must begin with an address if the protocol so specifies,
1058  * followed by an optional mbuf or mbufs containing ancillary data,
1059  * and then zero or more mbufs of data.
1060  * In order to avoid blocking network interrupts for the entire time here,
1061  * we splx() while doing the actual copy to user space.
1062  * Although the sockbuf is locked, new data may still be appended,
1063  * and thus we must maintain consistency of the sockbuf during that time.
1064  *
1065  * The caller may receive the data as a single mbuf chain by supplying
1066  * an mbuf **mp0 for use in returning the chain.  The uio is then used
1067  * only for the count in uio_resid.
1068  */
1069 int
1070 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
1071 	struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1072 {
1073 	struct lwp *l = curlwp;
1074 	struct mbuf	*m, **mp, *mt;
1075 	int atomic, flags, len, error, s, offset, moff, type, orig_resid;
1076 	const struct protosw	*pr;
1077 	struct mbuf	*nextrecord;
1078 	int		mbuf_removed = 0;
1079 	const struct domain *dom;
1080 
1081 	pr = so->so_proto;
1082 	atomic = pr->pr_flags & PR_ATOMIC;
1083 	dom = pr->pr_domain;
1084 	mp = mp0;
1085 	type = 0;
1086 	orig_resid = uio->uio_resid;
1087 
1088 	if (paddr != NULL)
1089 		*paddr = NULL;
1090 	if (controlp != NULL)
1091 		*controlp = NULL;
1092 	if (flagsp != NULL)
1093 		flags = *flagsp &~ MSG_EOR;
1094 	else
1095 		flags = 0;
1096 
1097 	if ((flags & MSG_DONTWAIT) == 0)
1098 		sodopendfree();
1099 
1100 	if (flags & MSG_OOB) {
1101 		m = m_get(M_WAIT, MT_DATA);
1102 		solock(so);
1103 		error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m,
1104 		    (struct mbuf *)(long)(flags & MSG_PEEK), NULL, l);
1105 		sounlock(so);
1106 		if (error)
1107 			goto bad;
1108 		do {
1109 			error = uiomove(mtod(m, void *),
1110 			    (int) min(uio->uio_resid, m->m_len), uio);
1111 			m = m_free(m);
1112 		} while (uio->uio_resid > 0 && error == 0 && m);
1113  bad:
1114 		if (m != NULL)
1115 			m_freem(m);
1116 		return error;
1117 	}
1118 	if (mp != NULL)
1119 		*mp = NULL;
1120 
1121 	/*
1122 	 * solock() provides atomicity of access.  splsoftnet() prevents
1123 	 * protocol processing soft interrupts from interrupting us and
1124 	 * blocking (expensive).
1125 	 */
1126 	s = splsoftnet();
1127 	solock(so);
1128 	if (so->so_state & SS_ISCONFIRMING && uio->uio_resid)
1129 		(*pr->pr_usrreq)(so, PRU_RCVD, NULL, NULL, NULL, l);
1130 
1131  restart:
1132 	if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) {
1133 		sounlock(so);
1134 		splx(s);
1135 		return error;
1136 	}
1137 
1138 	m = so->so_rcv.sb_mb;
1139 	/*
1140 	 * If we have less data than requested, block awaiting more
1141 	 * (subject to any timeout) if:
1142 	 *   1. the current count is less than the low water mark,
1143 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1144 	 *	receive operation at once if we block (resid <= hiwat), or
1145 	 *   3. MSG_DONTWAIT is not set.
1146 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1147 	 * we have to do the receive in sections, and thus risk returning
1148 	 * a short count if a timeout or signal occurs after we start.
1149 	 */
1150 	if (m == NULL ||
1151 	    ((flags & MSG_DONTWAIT) == 0 &&
1152 	     so->so_rcv.sb_cc < uio->uio_resid &&
1153 	     (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1154 	      ((flags & MSG_WAITALL) &&
1155 	       uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1156 	     m->m_nextpkt == NULL && !atomic)) {
1157 #ifdef DIAGNOSTIC
1158 		if (m == NULL && so->so_rcv.sb_cc)
1159 			panic("receive 1");
1160 #endif
1161 		if (so->so_error) {
1162 			if (m != NULL)
1163 				goto dontblock;
1164 			error = so->so_error;
1165 			if ((flags & MSG_PEEK) == 0)
1166 				so->so_error = 0;
1167 			goto release;
1168 		}
1169 		if (so->so_state & SS_CANTRCVMORE) {
1170 			if (m != NULL)
1171 				goto dontblock;
1172 			else
1173 				goto release;
1174 		}
1175 		for (; m != NULL; m = m->m_next)
1176 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1177 				m = so->so_rcv.sb_mb;
1178 				goto dontblock;
1179 			}
1180 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1181 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1182 			error = ENOTCONN;
1183 			goto release;
1184 		}
1185 		if (uio->uio_resid == 0)
1186 			goto release;
1187 		if (so->so_nbio || (flags & MSG_DONTWAIT)) {
1188 			error = EWOULDBLOCK;
1189 			goto release;
1190 		}
1191 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
1192 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
1193 		sbunlock(&so->so_rcv);
1194 		error = sbwait(&so->so_rcv);
1195 		if (error != 0) {
1196 			sounlock(so);
1197 			splx(s);
1198 			return error;
1199 		}
1200 		goto restart;
1201 	}
1202  dontblock:
1203 	/*
1204 	 * On entry here, m points to the first record of the socket buffer.
1205 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1206 	 * pointer to the next record in the socket buffer.  We must keep the
1207 	 * various socket buffer pointers and local stack versions of the
1208 	 * pointers in sync, pushing out modifications before dropping the
1209 	 * socket lock, and re-reading them when picking it up.
1210 	 *
1211 	 * Otherwise, we will race with the network stack appending new data
1212 	 * or records onto the socket buffer by using inconsistent/stale
1213 	 * versions of the field, possibly resulting in socket buffer
1214 	 * corruption.
1215 	 *
1216 	 * By holding the high-level sblock(), we prevent simultaneous
1217 	 * readers from pulling off the front of the socket buffer.
1218 	 */
1219 	if (l != NULL)
1220 		l->l_ru.ru_msgrcv++;
1221 	KASSERT(m == so->so_rcv.sb_mb);
1222 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
1223 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
1224 	nextrecord = m->m_nextpkt;
1225 	if (pr->pr_flags & PR_ADDR) {
1226 #ifdef DIAGNOSTIC
1227 		if (m->m_type != MT_SONAME)
1228 			panic("receive 1a");
1229 #endif
1230 		orig_resid = 0;
1231 		if (flags & MSG_PEEK) {
1232 			if (paddr)
1233 				*paddr = m_copy(m, 0, m->m_len);
1234 			m = m->m_next;
1235 		} else {
1236 			sbfree(&so->so_rcv, m);
1237 			mbuf_removed = 1;
1238 			if (paddr != NULL) {
1239 				*paddr = m;
1240 				so->so_rcv.sb_mb = m->m_next;
1241 				m->m_next = NULL;
1242 				m = so->so_rcv.sb_mb;
1243 			} else {
1244 				MFREE(m, so->so_rcv.sb_mb);
1245 				m = so->so_rcv.sb_mb;
1246 			}
1247 			sbsync(&so->so_rcv, nextrecord);
1248 		}
1249 	}
1250 
1251 	/*
1252 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1253 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1254 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1255 	 * perform externalization (or freeing if controlp == NULL).
1256 	 */
1257 	if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) {
1258 		struct mbuf *cm = NULL, *cmn;
1259 		struct mbuf **cme = &cm;
1260 
1261 		do {
1262 			if (flags & MSG_PEEK) {
1263 				if (controlp != NULL) {
1264 					*controlp = m_copy(m, 0, m->m_len);
1265 					controlp = &(*controlp)->m_next;
1266 				}
1267 				m = m->m_next;
1268 			} else {
1269 				sbfree(&so->so_rcv, m);
1270 				so->so_rcv.sb_mb = m->m_next;
1271 				m->m_next = NULL;
1272 				*cme = m;
1273 				cme = &(*cme)->m_next;
1274 				m = so->so_rcv.sb_mb;
1275 			}
1276 		} while (m != NULL && m->m_type == MT_CONTROL);
1277 		if ((flags & MSG_PEEK) == 0)
1278 			sbsync(&so->so_rcv, nextrecord);
1279 		for (; cm != NULL; cm = cmn) {
1280 			cmn = cm->m_next;
1281 			cm->m_next = NULL;
1282 			type = mtod(cm, struct cmsghdr *)->cmsg_type;
1283 			if (controlp != NULL) {
1284 				if (dom->dom_externalize != NULL &&
1285 				    type == SCM_RIGHTS) {
1286 					sounlock(so);
1287 					splx(s);
1288 					error = (*dom->dom_externalize)(cm, l);
1289 					s = splsoftnet();
1290 					solock(so);
1291 				}
1292 				*controlp = cm;
1293 				while (*controlp != NULL)
1294 					controlp = &(*controlp)->m_next;
1295 			} else {
1296 				/*
1297 				 * Dispose of any SCM_RIGHTS message that went
1298 				 * through the read path rather than recv.
1299 				 */
1300 				if (dom->dom_dispose != NULL &&
1301 				    type == SCM_RIGHTS) {
1302 				    	sounlock(so);
1303 					(*dom->dom_dispose)(cm);
1304 					solock(so);
1305 				}
1306 				m_freem(cm);
1307 			}
1308 		}
1309 		if (m != NULL)
1310 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1311 		else
1312 			nextrecord = so->so_rcv.sb_mb;
1313 		orig_resid = 0;
1314 	}
1315 
1316 	/* If m is non-NULL, we have some data to read. */
1317 	if (__predict_true(m != NULL)) {
1318 		type = m->m_type;
1319 		if (type == MT_OOBDATA)
1320 			flags |= MSG_OOB;
1321 	}
1322 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1323 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1324 
1325 	moff = 0;
1326 	offset = 0;
1327 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1328 		if (m->m_type == MT_OOBDATA) {
1329 			if (type != MT_OOBDATA)
1330 				break;
1331 		} else if (type == MT_OOBDATA)
1332 			break;
1333 #ifdef DIAGNOSTIC
1334 		else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
1335 			panic("receive 3");
1336 #endif
1337 		so->so_state &= ~SS_RCVATMARK;
1338 		len = uio->uio_resid;
1339 		if (so->so_oobmark && len > so->so_oobmark - offset)
1340 			len = so->so_oobmark - offset;
1341 		if (len > m->m_len - moff)
1342 			len = m->m_len - moff;
1343 		/*
1344 		 * If mp is set, just pass back the mbufs.
1345 		 * Otherwise copy them out via the uio, then free.
1346 		 * Sockbuf must be consistent here (points to current mbuf,
1347 		 * it points to next record) when we drop priority;
1348 		 * we must note any additions to the sockbuf when we
1349 		 * block interrupts again.
1350 		 */
1351 		if (mp == NULL) {
1352 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1353 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1354 			sounlock(so);
1355 			splx(s);
1356 			error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1357 			s = splsoftnet();
1358 			solock(so);
1359 			if (error != 0) {
1360 				/*
1361 				 * If any part of the record has been removed
1362 				 * (such as the MT_SONAME mbuf, which will
1363 				 * happen when PR_ADDR, and thus also
1364 				 * PR_ATOMIC, is set), then drop the entire
1365 				 * record to maintain the atomicity of the
1366 				 * receive operation.
1367 				 *
1368 				 * This avoids a later panic("receive 1a")
1369 				 * when compiled with DIAGNOSTIC.
1370 				 */
1371 				if (m && mbuf_removed && atomic)
1372 					(void) sbdroprecord(&so->so_rcv);
1373 
1374 				goto release;
1375 			}
1376 		} else
1377 			uio->uio_resid -= len;
1378 		if (len == m->m_len - moff) {
1379 			if (m->m_flags & M_EOR)
1380 				flags |= MSG_EOR;
1381 			if (flags & MSG_PEEK) {
1382 				m = m->m_next;
1383 				moff = 0;
1384 			} else {
1385 				nextrecord = m->m_nextpkt;
1386 				sbfree(&so->so_rcv, m);
1387 				if (mp) {
1388 					*mp = m;
1389 					mp = &m->m_next;
1390 					so->so_rcv.sb_mb = m = m->m_next;
1391 					*mp = NULL;
1392 				} else {
1393 					MFREE(m, so->so_rcv.sb_mb);
1394 					m = so->so_rcv.sb_mb;
1395 				}
1396 				/*
1397 				 * If m != NULL, we also know that
1398 				 * so->so_rcv.sb_mb != NULL.
1399 				 */
1400 				KASSERT(so->so_rcv.sb_mb == m);
1401 				if (m) {
1402 					m->m_nextpkt = nextrecord;
1403 					if (nextrecord == NULL)
1404 						so->so_rcv.sb_lastrecord = m;
1405 				} else {
1406 					so->so_rcv.sb_mb = nextrecord;
1407 					SB_EMPTY_FIXUP(&so->so_rcv);
1408 				}
1409 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1410 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1411 			}
1412 		} else if (flags & MSG_PEEK)
1413 			moff += len;
1414 		else {
1415 			if (mp != NULL) {
1416 				mt = m_copym(m, 0, len, M_NOWAIT);
1417 				if (__predict_false(mt == NULL)) {
1418 					sounlock(so);
1419 					mt = m_copym(m, 0, len, M_WAIT);
1420 					solock(so);
1421 				}
1422 				*mp = mt;
1423 			}
1424 			m->m_data += len;
1425 			m->m_len -= len;
1426 			so->so_rcv.sb_cc -= len;
1427 		}
1428 		if (so->so_oobmark) {
1429 			if ((flags & MSG_PEEK) == 0) {
1430 				so->so_oobmark -= len;
1431 				if (so->so_oobmark == 0) {
1432 					so->so_state |= SS_RCVATMARK;
1433 					break;
1434 				}
1435 			} else {
1436 				offset += len;
1437 				if (offset == so->so_oobmark)
1438 					break;
1439 			}
1440 		}
1441 		if (flags & MSG_EOR)
1442 			break;
1443 		/*
1444 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1445 		 * we must not quit until "uio->uio_resid == 0" or an error
1446 		 * termination.  If a signal/timeout occurs, return
1447 		 * with a short count but without error.
1448 		 * Keep sockbuf locked against other readers.
1449 		 */
1450 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1451 		    !sosendallatonce(so) && !nextrecord) {
1452 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1453 				break;
1454 			/*
1455 			 * If we are peeking and the socket receive buffer is
1456 			 * full, stop since we can't get more data to peek at.
1457 			 */
1458 			if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
1459 				break;
1460 			/*
1461 			 * If we've drained the socket buffer, tell the
1462 			 * protocol in case it needs to do something to
1463 			 * get it filled again.
1464 			 */
1465 			if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1466 				(*pr->pr_usrreq)(so, PRU_RCVD,
1467 				    NULL, (struct mbuf *)(long)flags, NULL, l);
1468 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1469 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1470 			error = sbwait(&so->so_rcv);
1471 			if (error != 0) {
1472 				sbunlock(&so->so_rcv);
1473 				sounlock(so);
1474 				splx(s);
1475 				return 0;
1476 			}
1477 			if ((m = so->so_rcv.sb_mb) != NULL)
1478 				nextrecord = m->m_nextpkt;
1479 		}
1480 	}
1481 
1482 	if (m && atomic) {
1483 		flags |= MSG_TRUNC;
1484 		if ((flags & MSG_PEEK) == 0)
1485 			(void) sbdroprecord(&so->so_rcv);
1486 	}
1487 	if ((flags & MSG_PEEK) == 0) {
1488 		if (m == NULL) {
1489 			/*
1490 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1491 			 * part makes sure sb_lastrecord is up-to-date if
1492 			 * there is still data in the socket buffer.
1493 			 */
1494 			so->so_rcv.sb_mb = nextrecord;
1495 			if (so->so_rcv.sb_mb == NULL) {
1496 				so->so_rcv.sb_mbtail = NULL;
1497 				so->so_rcv.sb_lastrecord = NULL;
1498 			} else if (nextrecord->m_nextpkt == NULL)
1499 				so->so_rcv.sb_lastrecord = nextrecord;
1500 		}
1501 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1502 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1503 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1504 			(*pr->pr_usrreq)(so, PRU_RCVD, NULL,
1505 			    (struct mbuf *)(long)flags, NULL, l);
1506 	}
1507 	if (orig_resid == uio->uio_resid && orig_resid &&
1508 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1509 		sbunlock(&so->so_rcv);
1510 		goto restart;
1511 	}
1512 
1513 	if (flagsp != NULL)
1514 		*flagsp |= flags;
1515  release:
1516 	sbunlock(&so->so_rcv);
1517 	sounlock(so);
1518 	splx(s);
1519 	return error;
1520 }
1521 
1522 int
1523 soshutdown(struct socket *so, int how)
1524 {
1525 	const struct protosw	*pr;
1526 	int	error;
1527 
1528 	KASSERT(solocked(so));
1529 
1530 	pr = so->so_proto;
1531 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1532 		return (EINVAL);
1533 
1534 	if (how == SHUT_RD || how == SHUT_RDWR) {
1535 		sorflush(so);
1536 		error = 0;
1537 	}
1538 	if (how == SHUT_WR || how == SHUT_RDWR)
1539 		error = (*pr->pr_usrreq)(so, PRU_SHUTDOWN, NULL,
1540 		    NULL, NULL, NULL);
1541 
1542 	return error;
1543 }
1544 
1545 void
1546 sorflush(struct socket *so)
1547 {
1548 	struct sockbuf	*sb, asb;
1549 	const struct protosw	*pr;
1550 
1551 	KASSERT(solocked(so));
1552 
1553 	sb = &so->so_rcv;
1554 	pr = so->so_proto;
1555 	socantrcvmore(so);
1556 	sb->sb_flags |= SB_NOINTR;
1557 	(void )sblock(sb, M_WAITOK);
1558 	sbunlock(sb);
1559 	asb = *sb;
1560 	/*
1561 	 * Clear most of the sockbuf structure, but leave some of the
1562 	 * fields valid.
1563 	 */
1564 	memset(&sb->sb_startzero, 0,
1565 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1566 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) {
1567 		sounlock(so);
1568 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1569 		solock(so);
1570 	}
1571 	sbrelease(&asb, so);
1572 }
1573 
1574 static int
1575 sosetopt1(struct socket *so, int level, int optname, struct mbuf *m)
1576 {
1577 	int optval, val;
1578 	struct linger	*l;
1579 	struct sockbuf	*sb;
1580 	struct timeval *tv;
1581 
1582 	switch (optname) {
1583 
1584 	case SO_LINGER:
1585 		if (m == NULL || m->m_len != sizeof(struct linger))
1586 			return EINVAL;
1587 		l = mtod(m, struct linger *);
1588 		if (l->l_linger < 0 || l->l_linger > USHRT_MAX ||
1589 		    l->l_linger > (INT_MAX / hz))
1590 			return EDOM;
1591 		so->so_linger = l->l_linger;
1592 		if (l->l_onoff)
1593 			so->so_options |= SO_LINGER;
1594 		else
1595 			so->so_options &= ~SO_LINGER;
1596 		break;
1597 
1598 	case SO_DEBUG:
1599 	case SO_KEEPALIVE:
1600 	case SO_DONTROUTE:
1601 	case SO_USELOOPBACK:
1602 	case SO_BROADCAST:
1603 	case SO_REUSEADDR:
1604 	case SO_REUSEPORT:
1605 	case SO_OOBINLINE:
1606 	case SO_TIMESTAMP:
1607 		if (m == NULL || m->m_len < sizeof(int))
1608 			return EINVAL;
1609 		if (*mtod(m, int *))
1610 			so->so_options |= optname;
1611 		else
1612 			so->so_options &= ~optname;
1613 		break;
1614 
1615 	case SO_SNDBUF:
1616 	case SO_RCVBUF:
1617 	case SO_SNDLOWAT:
1618 	case SO_RCVLOWAT:
1619 		if (m == NULL || m->m_len < sizeof(int))
1620 			return EINVAL;
1621 
1622 		/*
1623 		 * Values < 1 make no sense for any of these
1624 		 * options, so disallow them.
1625 		 */
1626 		optval = *mtod(m, int *);
1627 		if (optval < 1)
1628 			return EINVAL;
1629 
1630 		switch (optname) {
1631 
1632 		case SO_SNDBUF:
1633 		case SO_RCVBUF:
1634 			sb = (optname == SO_SNDBUF) ?
1635 			    &so->so_snd : &so->so_rcv;
1636 			if (sbreserve(sb, (u_long)optval, so) == 0)
1637 				return ENOBUFS;
1638 			sb->sb_flags &= ~SB_AUTOSIZE;
1639 			break;
1640 
1641 		/*
1642 		 * Make sure the low-water is never greater than
1643 		 * the high-water.
1644 		 */
1645 		case SO_SNDLOWAT:
1646 			so->so_snd.sb_lowat =
1647 			    (optval > so->so_snd.sb_hiwat) ?
1648 			    so->so_snd.sb_hiwat : optval;
1649 			break;
1650 		case SO_RCVLOWAT:
1651 			so->so_rcv.sb_lowat =
1652 			    (optval > so->so_rcv.sb_hiwat) ?
1653 			    so->so_rcv.sb_hiwat : optval;
1654 			break;
1655 		}
1656 		break;
1657 
1658 	case SO_SNDTIMEO:
1659 	case SO_RCVTIMEO:
1660 		if (m == NULL || m->m_len < sizeof(*tv))
1661 			return EINVAL;
1662 		tv = mtod(m, struct timeval *);
1663 		if (tv->tv_sec > (INT_MAX - tv->tv_usec / tick) / hz)
1664 			return EDOM;
1665 		val = tv->tv_sec * hz + tv->tv_usec / tick;
1666 		if (val == 0 && tv->tv_usec != 0)
1667 			val = 1;
1668 
1669 		switch (optname) {
1670 
1671 		case SO_SNDTIMEO:
1672 			so->so_snd.sb_timeo = val;
1673 			break;
1674 		case SO_RCVTIMEO:
1675 			so->so_rcv.sb_timeo = val;
1676 			break;
1677 		}
1678 		break;
1679 
1680 	default:
1681 		return ENOPROTOOPT;
1682 	}
1683 	return 0;
1684 }
1685 
1686 int
1687 sosetopt(struct socket *so, int level, int optname, struct mbuf *m)
1688 {
1689 	int error, prerr;
1690 
1691 	solock(so);
1692 	if (level == SOL_SOCKET)
1693 		error = sosetopt1(so, level, optname, m);
1694 	else
1695 		error = ENOPROTOOPT;
1696 
1697 	if ((error == 0 || error == ENOPROTOOPT) &&
1698 	    so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) {
1699 		/* give the protocol stack a shot */
1700 		prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, level,
1701 		    optname, &m);
1702 		if (prerr == 0)
1703 			error = 0;
1704 		else if (prerr != ENOPROTOOPT)
1705 			error = prerr;
1706 	} else if (m != NULL)
1707 		(void)m_free(m);
1708 	sounlock(so);
1709 	return error;
1710 }
1711 
1712 int
1713 sogetopt(struct socket *so, int level, int optname, struct mbuf **mp)
1714 {
1715 	struct mbuf	*m;
1716 	int		error;
1717 
1718 	solock(so);
1719 	if (level != SOL_SOCKET) {
1720 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1721 			error = ((*so->so_proto->pr_ctloutput)
1722 				  (PRCO_GETOPT, so, level, optname, mp));
1723 		} else
1724 			error = (ENOPROTOOPT);
1725 	} else {
1726 		m = m_get(M_WAIT, MT_SOOPTS);
1727 		m->m_len = sizeof(int);
1728 
1729 		switch (optname) {
1730 
1731 		case SO_LINGER:
1732 			m->m_len = sizeof(struct linger);
1733 			mtod(m, struct linger *)->l_onoff =
1734 			    (so->so_options & SO_LINGER) ? 1 : 0;
1735 			mtod(m, struct linger *)->l_linger = so->so_linger;
1736 			break;
1737 
1738 		case SO_USELOOPBACK:
1739 		case SO_DONTROUTE:
1740 		case SO_DEBUG:
1741 		case SO_KEEPALIVE:
1742 		case SO_REUSEADDR:
1743 		case SO_REUSEPORT:
1744 		case SO_BROADCAST:
1745 		case SO_OOBINLINE:
1746 		case SO_TIMESTAMP:
1747 			*mtod(m, int *) = (so->so_options & optname) ? 1 : 0;
1748 			break;
1749 
1750 		case SO_TYPE:
1751 			*mtod(m, int *) = so->so_type;
1752 			break;
1753 
1754 		case SO_ERROR:
1755 			*mtod(m, int *) = so->so_error;
1756 			so->so_error = 0;
1757 			break;
1758 
1759 		case SO_SNDBUF:
1760 			*mtod(m, int *) = so->so_snd.sb_hiwat;
1761 			break;
1762 
1763 		case SO_RCVBUF:
1764 			*mtod(m, int *) = so->so_rcv.sb_hiwat;
1765 			break;
1766 
1767 		case SO_SNDLOWAT:
1768 			*mtod(m, int *) = so->so_snd.sb_lowat;
1769 			break;
1770 
1771 		case SO_RCVLOWAT:
1772 			*mtod(m, int *) = so->so_rcv.sb_lowat;
1773 			break;
1774 
1775 		case SO_SNDTIMEO:
1776 		case SO_RCVTIMEO:
1777 		    {
1778 			int val = (optname == SO_SNDTIMEO ?
1779 			     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1780 
1781 			m->m_len = sizeof(struct timeval);
1782 			mtod(m, struct timeval *)->tv_sec = val / hz;
1783 			mtod(m, struct timeval *)->tv_usec =
1784 			    (val % hz) * tick;
1785 			break;
1786 		    }
1787 
1788 		case SO_OVERFLOWED:
1789 			*mtod(m, int *) = so->so_rcv.sb_overflowed;
1790 			break;
1791 
1792 		default:
1793 			sounlock(so);
1794 			(void)m_free(m);
1795 			return (ENOPROTOOPT);
1796 		}
1797 		*mp = m;
1798 		error = 0;
1799 	}
1800 
1801 	sounlock(so);
1802 	return (error);
1803 }
1804 
1805 void
1806 sohasoutofband(struct socket *so)
1807 {
1808 
1809 	fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
1810 	selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, 0);
1811 }
1812 
1813 static void
1814 filt_sordetach(struct knote *kn)
1815 {
1816 	struct socket	*so;
1817 
1818 	so = ((file_t *)kn->kn_obj)->f_data;
1819 	solock(so);
1820 	SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
1821 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
1822 		so->so_rcv.sb_flags &= ~SB_KNOTE;
1823 	sounlock(so);
1824 }
1825 
1826 /*ARGSUSED*/
1827 static int
1828 filt_soread(struct knote *kn, long hint)
1829 {
1830 	struct socket	*so;
1831 	int rv;
1832 
1833 	so = ((file_t *)kn->kn_obj)->f_data;
1834 	if (hint != NOTE_SUBMIT)
1835 		solock(so);
1836 	kn->kn_data = so->so_rcv.sb_cc;
1837 	if (so->so_state & SS_CANTRCVMORE) {
1838 		kn->kn_flags |= EV_EOF;
1839 		kn->kn_fflags = so->so_error;
1840 		rv = 1;
1841 	} else if (so->so_error)	/* temporary udp error */
1842 		rv = 1;
1843 	else if (kn->kn_sfflags & NOTE_LOWAT)
1844 		rv = (kn->kn_data >= kn->kn_sdata);
1845 	else
1846 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
1847 	if (hint != NOTE_SUBMIT)
1848 		sounlock(so);
1849 	return rv;
1850 }
1851 
1852 static void
1853 filt_sowdetach(struct knote *kn)
1854 {
1855 	struct socket	*so;
1856 
1857 	so = ((file_t *)kn->kn_obj)->f_data;
1858 	solock(so);
1859 	SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
1860 	if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
1861 		so->so_snd.sb_flags &= ~SB_KNOTE;
1862 	sounlock(so);
1863 }
1864 
1865 /*ARGSUSED*/
1866 static int
1867 filt_sowrite(struct knote *kn, long hint)
1868 {
1869 	struct socket	*so;
1870 	int rv;
1871 
1872 	so = ((file_t *)kn->kn_obj)->f_data;
1873 	if (hint != NOTE_SUBMIT)
1874 		solock(so);
1875 	kn->kn_data = sbspace(&so->so_snd);
1876 	if (so->so_state & SS_CANTSENDMORE) {
1877 		kn->kn_flags |= EV_EOF;
1878 		kn->kn_fflags = so->so_error;
1879 		rv = 1;
1880 	} else if (so->so_error)	/* temporary udp error */
1881 		rv = 1;
1882 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
1883 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
1884 		rv = 0;
1885 	else if (kn->kn_sfflags & NOTE_LOWAT)
1886 		rv = (kn->kn_data >= kn->kn_sdata);
1887 	else
1888 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
1889 	if (hint != NOTE_SUBMIT)
1890 		sounlock(so);
1891 	return rv;
1892 }
1893 
1894 /*ARGSUSED*/
1895 static int
1896 filt_solisten(struct knote *kn, long hint)
1897 {
1898 	struct socket	*so;
1899 	int rv;
1900 
1901 	so = ((file_t *)kn->kn_obj)->f_data;
1902 
1903 	/*
1904 	 * Set kn_data to number of incoming connections, not
1905 	 * counting partial (incomplete) connections.
1906 	 */
1907 	if (hint != NOTE_SUBMIT)
1908 		solock(so);
1909 	kn->kn_data = so->so_qlen;
1910 	rv = (kn->kn_data > 0);
1911 	if (hint != NOTE_SUBMIT)
1912 		sounlock(so);
1913 	return rv;
1914 }
1915 
1916 static const struct filterops solisten_filtops =
1917 	{ 1, NULL, filt_sordetach, filt_solisten };
1918 static const struct filterops soread_filtops =
1919 	{ 1, NULL, filt_sordetach, filt_soread };
1920 static const struct filterops sowrite_filtops =
1921 	{ 1, NULL, filt_sowdetach, filt_sowrite };
1922 
1923 int
1924 soo_kqfilter(struct file *fp, struct knote *kn)
1925 {
1926 	struct socket	*so;
1927 	struct sockbuf	*sb;
1928 
1929 	so = ((file_t *)kn->kn_obj)->f_data;
1930 	solock(so);
1931 	switch (kn->kn_filter) {
1932 	case EVFILT_READ:
1933 		if (so->so_options & SO_ACCEPTCONN)
1934 			kn->kn_fop = &solisten_filtops;
1935 		else
1936 			kn->kn_fop = &soread_filtops;
1937 		sb = &so->so_rcv;
1938 		break;
1939 	case EVFILT_WRITE:
1940 		kn->kn_fop = &sowrite_filtops;
1941 		sb = &so->so_snd;
1942 		break;
1943 	default:
1944 		sounlock(so);
1945 		return (EINVAL);
1946 	}
1947 	SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
1948 	sb->sb_flags |= SB_KNOTE;
1949 	sounlock(so);
1950 	return (0);
1951 }
1952 
1953 static int
1954 sodopoll(struct socket *so, int events)
1955 {
1956 	int revents;
1957 
1958 	revents = 0;
1959 
1960 	if (events & (POLLIN | POLLRDNORM))
1961 		if (soreadable(so))
1962 			revents |= events & (POLLIN | POLLRDNORM);
1963 
1964 	if (events & (POLLOUT | POLLWRNORM))
1965 		if (sowritable(so))
1966 			revents |= events & (POLLOUT | POLLWRNORM);
1967 
1968 	if (events & (POLLPRI | POLLRDBAND))
1969 		if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
1970 			revents |= events & (POLLPRI | POLLRDBAND);
1971 
1972 	return revents;
1973 }
1974 
1975 int
1976 sopoll(struct socket *so, int events)
1977 {
1978 	int revents = 0;
1979 
1980 #ifndef DIAGNOSTIC
1981 	/*
1982 	 * Do a quick, unlocked check in expectation that the socket
1983 	 * will be ready for I/O.  Don't do this check if DIAGNOSTIC,
1984 	 * as the solocked() assertions will fail.
1985 	 */
1986 	if ((revents = sodopoll(so, events)) != 0)
1987 		return revents;
1988 #endif
1989 
1990 	solock(so);
1991 	if ((revents = sodopoll(so, events)) == 0) {
1992 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
1993 			selrecord(curlwp, &so->so_rcv.sb_sel);
1994 			so->so_rcv.sb_flags |= SB_NOTIFY;
1995 		}
1996 
1997 		if (events & (POLLOUT | POLLWRNORM)) {
1998 			selrecord(curlwp, &so->so_snd.sb_sel);
1999 			so->so_snd.sb_flags |= SB_NOTIFY;
2000 		}
2001 	}
2002 	sounlock(so);
2003 
2004 	return revents;
2005 }
2006 
2007 
2008 #include <sys/sysctl.h>
2009 
2010 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO);
2011 
2012 /*
2013  * sysctl helper routine for kern.somaxkva.  ensures that the given
2014  * value is not too small.
2015  * (XXX should we maybe make sure it's not too large as well?)
2016  */
2017 static int
2018 sysctl_kern_somaxkva(SYSCTLFN_ARGS)
2019 {
2020 	int error, new_somaxkva;
2021 	struct sysctlnode node;
2022 
2023 	new_somaxkva = somaxkva;
2024 	node = *rnode;
2025 	node.sysctl_data = &new_somaxkva;
2026 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2027 	if (error || newp == NULL)
2028 		return (error);
2029 
2030 	if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */
2031 		return (EINVAL);
2032 
2033 	mutex_enter(&so_pendfree_lock);
2034 	somaxkva = new_somaxkva;
2035 	cv_broadcast(&socurkva_cv);
2036 	mutex_exit(&so_pendfree_lock);
2037 
2038 	return (error);
2039 }
2040 
2041 SYSCTL_SETUP(sysctl_kern_somaxkva_setup, "sysctl kern.somaxkva setup")
2042 {
2043 
2044 	sysctl_createv(clog, 0, NULL, NULL,
2045 		       CTLFLAG_PERMANENT,
2046 		       CTLTYPE_NODE, "kern", NULL,
2047 		       NULL, 0, NULL, 0,
2048 		       CTL_KERN, CTL_EOL);
2049 
2050 	sysctl_createv(clog, 0, NULL, NULL,
2051 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2052 		       CTLTYPE_INT, "somaxkva",
2053 		       SYSCTL_DESCR("Maximum amount of kernel memory to be "
2054 				    "used for socket buffers"),
2055 		       sysctl_kern_somaxkva, 0, NULL, 0,
2056 		       CTL_KERN, KERN_SOMAXKVA, CTL_EOL);
2057 }
2058