xref: /netbsd-src/sys/kern/uipc_socket.c (revision c2f76ff004a2cb67efe5b12d97bd3ef7fe89e18d)
1 /*	$NetBSD: uipc_socket.c,v 1.202 2011/01/17 07:13:32 uebayasi Exp $	*/
2 
3 /*-
4  * Copyright (c) 2002, 2007, 2008, 2009 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, and by Andrew Doran.
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 1.202 2011/01/17 07:13:32 uebayasi Exp $");
67 
68 #include "opt_compat_netbsd.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/kmem.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/uidinfo.h>
90 #include <sys/event.h>
91 #include <sys/poll.h>
92 #include <sys/kauth.h>
93 #include <sys/mutex.h>
94 #include <sys/condvar.h>
95 
96 #ifdef COMPAT_50
97 #include <compat/sys/time.h>
98 #include <compat/sys/socket.h>
99 #endif
100 
101 #include <uvm/uvm_extern.h>
102 #include <uvm/uvm_loan.h>
103 #include <uvm/uvm_page.h>
104 
105 MALLOC_DEFINE(M_SOOPTS, "soopts", "socket options");
106 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
107 
108 extern const struct fileops socketops;
109 
110 extern int	somaxconn;			/* patchable (XXX sysctl) */
111 int		somaxconn = SOMAXCONN;
112 kmutex_t	*softnet_lock;
113 
114 #ifdef SOSEND_COUNTERS
115 #include <sys/device.h>
116 
117 static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
118     NULL, "sosend", "loan big");
119 static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
120     NULL, "sosend", "copy big");
121 static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
122     NULL, "sosend", "copy small");
123 static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
124     NULL, "sosend", "kva limit");
125 
126 #define	SOSEND_COUNTER_INCR(ev)		(ev)->ev_count++
127 
128 EVCNT_ATTACH_STATIC(sosend_loan_big);
129 EVCNT_ATTACH_STATIC(sosend_copy_big);
130 EVCNT_ATTACH_STATIC(sosend_copy_small);
131 EVCNT_ATTACH_STATIC(sosend_kvalimit);
132 #else
133 
134 #define	SOSEND_COUNTER_INCR(ev)		/* nothing */
135 
136 #endif /* SOSEND_COUNTERS */
137 
138 static struct callback_entry sokva_reclaimerentry;
139 
140 #if defined(SOSEND_NO_LOAN) || defined(MULTIPROCESSOR)
141 int sock_loan_thresh = -1;
142 #else
143 int sock_loan_thresh = 4096;
144 #endif
145 
146 static kmutex_t so_pendfree_lock;
147 static struct mbuf *so_pendfree;
148 
149 #ifndef SOMAXKVA
150 #define	SOMAXKVA (16 * 1024 * 1024)
151 #endif
152 int somaxkva = SOMAXKVA;
153 static int socurkva;
154 static kcondvar_t socurkva_cv;
155 
156 static kauth_listener_t socket_listener;
157 
158 #define	SOCK_LOAN_CHUNK		65536
159 
160 static size_t sodopendfree(void);
161 static size_t sodopendfreel(void);
162 
163 static void sysctl_kern_somaxkva_setup(void);
164 static struct sysctllog *socket_sysctllog;
165 
166 static vsize_t
167 sokvareserve(struct socket *so, vsize_t len)
168 {
169 	int error;
170 
171 	mutex_enter(&so_pendfree_lock);
172 	while (socurkva + len > somaxkva) {
173 		size_t freed;
174 
175 		/*
176 		 * try to do pendfree.
177 		 */
178 
179 		freed = sodopendfreel();
180 
181 		/*
182 		 * if some kva was freed, try again.
183 		 */
184 
185 		if (freed)
186 			continue;
187 
188 		SOSEND_COUNTER_INCR(&sosend_kvalimit);
189 		error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock);
190 		if (error) {
191 			len = 0;
192 			break;
193 		}
194 	}
195 	socurkva += len;
196 	mutex_exit(&so_pendfree_lock);
197 	return len;
198 }
199 
200 static void
201 sokvaunreserve(vsize_t len)
202 {
203 
204 	mutex_enter(&so_pendfree_lock);
205 	socurkva -= len;
206 	cv_broadcast(&socurkva_cv);
207 	mutex_exit(&so_pendfree_lock);
208 }
209 
210 /*
211  * sokvaalloc: allocate kva for loan.
212  */
213 
214 vaddr_t
215 sokvaalloc(vsize_t len, struct socket *so)
216 {
217 	vaddr_t lva;
218 
219 	/*
220 	 * reserve kva.
221 	 */
222 
223 	if (sokvareserve(so, len) == 0)
224 		return 0;
225 
226 	/*
227 	 * allocate kva.
228 	 */
229 
230 	lva = uvm_km_alloc(kernel_map, len, 0, UVM_KMF_VAONLY | UVM_KMF_WAITVA);
231 	if (lva == 0) {
232 		sokvaunreserve(len);
233 		return (0);
234 	}
235 
236 	return lva;
237 }
238 
239 /*
240  * sokvafree: free kva for loan.
241  */
242 
243 void
244 sokvafree(vaddr_t sva, vsize_t len)
245 {
246 
247 	/*
248 	 * free kva.
249 	 */
250 
251 	uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY);
252 
253 	/*
254 	 * unreserve kva.
255 	 */
256 
257 	sokvaunreserve(len);
258 }
259 
260 static void
261 sodoloanfree(struct vm_page **pgs, void *buf, size_t size)
262 {
263 	vaddr_t sva, eva;
264 	vsize_t len;
265 	int npgs;
266 
267 	KASSERT(pgs != NULL);
268 
269 	eva = round_page((vaddr_t) buf + size);
270 	sva = trunc_page((vaddr_t) buf);
271 	len = eva - sva;
272 	npgs = len >> PAGE_SHIFT;
273 
274 	pmap_kremove(sva, len);
275 	pmap_update(pmap_kernel());
276 	uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE);
277 	sokvafree(sva, len);
278 }
279 
280 static size_t
281 sodopendfree(void)
282 {
283 	size_t rv;
284 
285 	if (__predict_true(so_pendfree == NULL))
286 		return 0;
287 
288 	mutex_enter(&so_pendfree_lock);
289 	rv = sodopendfreel();
290 	mutex_exit(&so_pendfree_lock);
291 
292 	return rv;
293 }
294 
295 /*
296  * sodopendfreel: free mbufs on "pendfree" list.
297  * unlock and relock so_pendfree_lock when freeing mbufs.
298  *
299  * => called with so_pendfree_lock held.
300  */
301 
302 static size_t
303 sodopendfreel(void)
304 {
305 	struct mbuf *m, *next;
306 	size_t rv = 0;
307 
308 	KASSERT(mutex_owned(&so_pendfree_lock));
309 
310 	while (so_pendfree != NULL) {
311 		m = so_pendfree;
312 		so_pendfree = NULL;
313 		mutex_exit(&so_pendfree_lock);
314 
315 		for (; m != NULL; m = next) {
316 			next = m->m_next;
317 			KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0);
318 			KASSERT(m->m_ext.ext_refcnt == 0);
319 
320 			rv += m->m_ext.ext_size;
321 			sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf,
322 			    m->m_ext.ext_size);
323 			pool_cache_put(mb_cache, m);
324 		}
325 
326 		mutex_enter(&so_pendfree_lock);
327 	}
328 
329 	return (rv);
330 }
331 
332 void
333 soloanfree(struct mbuf *m, void *buf, size_t size, void *arg)
334 {
335 
336 	KASSERT(m != NULL);
337 
338 	/*
339 	 * postpone freeing mbuf.
340 	 *
341 	 * we can't do it in interrupt context
342 	 * because we need to put kva back to kernel_map.
343 	 */
344 
345 	mutex_enter(&so_pendfree_lock);
346 	m->m_next = so_pendfree;
347 	so_pendfree = m;
348 	cv_broadcast(&socurkva_cv);
349 	mutex_exit(&so_pendfree_lock);
350 }
351 
352 static long
353 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space)
354 {
355 	struct iovec *iov = uio->uio_iov;
356 	vaddr_t sva, eva;
357 	vsize_t len;
358 	vaddr_t lva;
359 	int npgs, error;
360 	vaddr_t va;
361 	int i;
362 
363 	if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace))
364 		return (0);
365 
366 	if (iov->iov_len < (size_t) space)
367 		space = iov->iov_len;
368 	if (space > SOCK_LOAN_CHUNK)
369 		space = SOCK_LOAN_CHUNK;
370 
371 	eva = round_page((vaddr_t) iov->iov_base + space);
372 	sva = trunc_page((vaddr_t) iov->iov_base);
373 	len = eva - sva;
374 	npgs = len >> PAGE_SHIFT;
375 
376 	KASSERT(npgs <= M_EXT_MAXPAGES);
377 
378 	lva = sokvaalloc(len, so);
379 	if (lva == 0)
380 		return 0;
381 
382 	error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len,
383 	    m->m_ext.ext_pgs, UVM_LOAN_TOPAGE);
384 	if (error) {
385 		sokvafree(lva, len);
386 		return (0);
387 	}
388 
389 	for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE)
390 		pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
391 		    VM_PROT_READ, 0);
392 	pmap_update(pmap_kernel());
393 
394 	lva += (vaddr_t) iov->iov_base & PAGE_MASK;
395 
396 	MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so);
397 	m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
398 
399 	uio->uio_resid -= space;
400 	/* uio_offset not updated, not set/used for write(2) */
401 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space;
402 	uio->uio_iov->iov_len -= space;
403 	if (uio->uio_iov->iov_len == 0) {
404 		uio->uio_iov++;
405 		uio->uio_iovcnt--;
406 	}
407 
408 	return (space);
409 }
410 
411 static int
412 sokva_reclaim_callback(struct callback_entry *ce, void *obj, void *arg)
413 {
414 
415 	KASSERT(ce == &sokva_reclaimerentry);
416 	KASSERT(obj == NULL);
417 
418 	sodopendfree();
419 	if (!vm_map_starved_p(kernel_map)) {
420 		return CALLBACK_CHAIN_ABORT;
421 	}
422 	return CALLBACK_CHAIN_CONTINUE;
423 }
424 
425 struct mbuf *
426 getsombuf(struct socket *so, int type)
427 {
428 	struct mbuf *m;
429 
430 	m = m_get(M_WAIT, type);
431 	MCLAIM(m, so->so_mowner);
432 	return m;
433 }
434 
435 static int
436 socket_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
437     void *arg0, void *arg1, void *arg2, void *arg3)
438 {
439 	int result;
440 	enum kauth_network_req req;
441 
442 	result = KAUTH_RESULT_DEFER;
443 	req = (enum kauth_network_req)arg0;
444 
445 	if ((action != KAUTH_NETWORK_SOCKET) &&
446 	    (action != KAUTH_NETWORK_BIND))
447 		return result;
448 
449 	switch (req) {
450 	case KAUTH_REQ_NETWORK_BIND_PORT:
451 		result = KAUTH_RESULT_ALLOW;
452 		break;
453 
454 	case KAUTH_REQ_NETWORK_SOCKET_DROP: {
455 		/* Normal users can only drop their own connections. */
456 		struct socket *so = (struct socket *)arg1;
457 
458 		if (proc_uidmatch(cred, so->so_cred))
459 			result = KAUTH_RESULT_ALLOW;
460 
461 		break;
462 		}
463 
464 	case KAUTH_REQ_NETWORK_SOCKET_OPEN:
465 		/* We allow "raw" routing/bluetooth sockets to anyone. */
466 		if ((u_long)arg1 == PF_ROUTE || (u_long)arg1 == PF_BLUETOOTH)
467 			result = KAUTH_RESULT_ALLOW;
468 		else {
469 			/* Privileged, let secmodel handle this. */
470 			if ((u_long)arg2 == SOCK_RAW)
471 				break;
472 		}
473 
474 		result = KAUTH_RESULT_ALLOW;
475 
476 		break;
477 
478 	case KAUTH_REQ_NETWORK_SOCKET_CANSEE:
479 		result = KAUTH_RESULT_ALLOW;
480 
481 		break;
482 
483 	default:
484 		break;
485 	}
486 
487 	return result;
488 }
489 
490 void
491 soinit(void)
492 {
493 
494 	sysctl_kern_somaxkva_setup();
495 
496 	mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM);
497 	softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
498 	cv_init(&socurkva_cv, "sokva");
499 	soinit2();
500 
501 	/* Set the initial adjusted socket buffer size. */
502 	if (sb_max_set(sb_max))
503 		panic("bad initial sb_max value: %lu", sb_max);
504 
505 	callback_register(&vm_map_to_kernel(kernel_map)->vmk_reclaim_callback,
506 	    &sokva_reclaimerentry, NULL, sokva_reclaim_callback);
507 
508 	socket_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK,
509 	    socket_listener_cb, NULL);
510 }
511 
512 /*
513  * Socket operation routines.
514  * These routines are called by the routines in
515  * sys_socket.c or from a system process, and
516  * implement the semantics of socket operations by
517  * switching out to the protocol specific routines.
518  */
519 /*ARGSUSED*/
520 int
521 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l,
522 	 struct socket *lockso)
523 {
524 	const struct protosw	*prp;
525 	struct socket	*so;
526 	uid_t		uid;
527 	int		error;
528 	kmutex_t	*lock;
529 
530 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET,
531 	    KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type),
532 	    KAUTH_ARG(proto));
533 	if (error != 0)
534 		return error;
535 
536 	if (proto)
537 		prp = pffindproto(dom, proto, type);
538 	else
539 		prp = pffindtype(dom, type);
540 	if (prp == NULL) {
541 		/* no support for domain */
542 		if (pffinddomain(dom) == 0)
543 			return EAFNOSUPPORT;
544 		/* no support for socket type */
545 		if (proto == 0 && type != 0)
546 			return EPROTOTYPE;
547 		return EPROTONOSUPPORT;
548 	}
549 	if (prp->pr_usrreq == NULL)
550 		return EPROTONOSUPPORT;
551 	if (prp->pr_type != type)
552 		return EPROTOTYPE;
553 
554 	so = soget(true);
555 	so->so_type = type;
556 	so->so_proto = prp;
557 	so->so_send = sosend;
558 	so->so_receive = soreceive;
559 #ifdef MBUFTRACE
560 	so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner;
561 	so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner;
562 	so->so_mowner = &prp->pr_domain->dom_mowner;
563 #endif
564 	uid = kauth_cred_geteuid(l->l_cred);
565 	so->so_uidinfo = uid_find(uid);
566 	so->so_cpid = l->l_proc->p_pid;
567 	if (lockso != NULL) {
568 		/* Caller wants us to share a lock. */
569 		lock = lockso->so_lock;
570 		so->so_lock = lock;
571 		mutex_obj_hold(lock);
572 		mutex_enter(lock);
573 	} else {
574 		/* Lock assigned and taken during PRU_ATTACH. */
575 	}
576 	error = (*prp->pr_usrreq)(so, PRU_ATTACH, NULL,
577 	    (struct mbuf *)(long)proto, NULL, l);
578 	KASSERT(solocked(so));
579 	if (error != 0) {
580 		so->so_state |= SS_NOFDREF;
581 		sofree(so);
582 		return error;
583 	}
584 	so->so_cred = kauth_cred_dup(l->l_cred);
585 	sounlock(so);
586 	*aso = so;
587 	return 0;
588 }
589 
590 /* On success, write file descriptor to fdout and return zero.  On
591  * failure, return non-zero; *fdout will be undefined.
592  */
593 int
594 fsocreate(int domain, struct socket **sop, int type, int protocol,
595     struct lwp *l, int *fdout)
596 {
597 	struct socket	*so;
598 	struct file	*fp;
599 	int		fd, error;
600 
601 	if ((error = fd_allocfile(&fp, &fd)) != 0)
602 		return (error);
603 	fp->f_flag = FREAD|FWRITE;
604 	fp->f_type = DTYPE_SOCKET;
605 	fp->f_ops = &socketops;
606 	error = socreate(domain, &so, type, protocol, l, NULL);
607 	if (error != 0) {
608 		fd_abort(curproc, fp, fd);
609 	} else {
610 		if (sop != NULL)
611 			*sop = so;
612 		fp->f_data = so;
613 		fd_affix(curproc, fp, fd);
614 		*fdout = fd;
615 	}
616 	return error;
617 }
618 
619 int
620 sofamily(const struct socket *so)
621 {
622 	const struct protosw *pr;
623 	const struct domain *dom;
624 
625 	if ((pr = so->so_proto) == NULL)
626 		return AF_UNSPEC;
627 	if ((dom = pr->pr_domain) == NULL)
628 		return AF_UNSPEC;
629 	return dom->dom_family;
630 }
631 
632 int
633 sobind(struct socket *so, struct mbuf *nam, struct lwp *l)
634 {
635 	int	error;
636 
637 	solock(so);
638 	error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, NULL, nam, NULL, l);
639 	sounlock(so);
640 	return error;
641 }
642 
643 int
644 solisten(struct socket *so, int backlog, struct lwp *l)
645 {
646 	int	error;
647 
648 	solock(so);
649 	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
650 	    SS_ISDISCONNECTING)) != 0) {
651 	    	sounlock(so);
652 		return (EOPNOTSUPP);
653 	}
654 	error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, NULL,
655 	    NULL, NULL, l);
656 	if (error != 0) {
657 		sounlock(so);
658 		return error;
659 	}
660 	if (TAILQ_EMPTY(&so->so_q))
661 		so->so_options |= SO_ACCEPTCONN;
662 	if (backlog < 0)
663 		backlog = 0;
664 	so->so_qlimit = min(backlog, somaxconn);
665 	sounlock(so);
666 	return 0;
667 }
668 
669 void
670 sofree(struct socket *so)
671 {
672 	u_int refs;
673 
674 	KASSERT(solocked(so));
675 
676 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
677 		sounlock(so);
678 		return;
679 	}
680 	if (so->so_head) {
681 		/*
682 		 * We must not decommission a socket that's on the accept(2)
683 		 * queue.  If we do, then accept(2) may hang after select(2)
684 		 * indicated that the listening socket was ready.
685 		 */
686 		if (!soqremque(so, 0)) {
687 			sounlock(so);
688 			return;
689 		}
690 	}
691 	if (so->so_rcv.sb_hiwat)
692 		(void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0,
693 		    RLIM_INFINITY);
694 	if (so->so_snd.sb_hiwat)
695 		(void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0,
696 		    RLIM_INFINITY);
697 	sbrelease(&so->so_snd, so);
698 	KASSERT(!cv_has_waiters(&so->so_cv));
699 	KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv));
700 	KASSERT(!cv_has_waiters(&so->so_snd.sb_cv));
701 	sorflush(so);
702 	refs = so->so_aborting;	/* XXX */
703 	/* Remove acccept filter if one is present. */
704 	if (so->so_accf != NULL)
705 		(void)accept_filt_clear(so);
706 	sounlock(so);
707 	if (refs == 0)		/* XXX */
708 		soput(so);
709 }
710 
711 /*
712  * Close a socket on last file table reference removal.
713  * Initiate disconnect if connected.
714  * Free socket when disconnect complete.
715  */
716 int
717 soclose(struct socket *so)
718 {
719 	struct socket	*so2;
720 	int		error;
721 	int		error2;
722 
723 	error = 0;
724 	solock(so);
725 	if (so->so_options & SO_ACCEPTCONN) {
726 		for (;;) {
727 			if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) {
728 				KASSERT(solocked2(so, so2));
729 				(void) soqremque(so2, 0);
730 				/* soabort drops the lock. */
731 				(void) soabort(so2);
732 				solock(so);
733 				continue;
734 			}
735 			if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) {
736 				KASSERT(solocked2(so, so2));
737 				(void) soqremque(so2, 1);
738 				/* soabort drops the lock. */
739 				(void) soabort(so2);
740 				solock(so);
741 				continue;
742 			}
743 			break;
744 		}
745 	}
746 	if (so->so_pcb == 0)
747 		goto discard;
748 	if (so->so_state & SS_ISCONNECTED) {
749 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
750 			error = sodisconnect(so);
751 			if (error)
752 				goto drop;
753 		}
754 		if (so->so_options & SO_LINGER) {
755 			if ((so->so_state & SS_ISDISCONNECTING) && so->so_nbio)
756 				goto drop;
757 			while (so->so_state & SS_ISCONNECTED) {
758 				error = sowait(so, true, so->so_linger * hz);
759 				if (error)
760 					break;
761 			}
762 		}
763 	}
764  drop:
765 	if (so->so_pcb) {
766 		error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH,
767 		    NULL, NULL, NULL, NULL);
768 		if (error == 0)
769 			error = error2;
770 	}
771  discard:
772 	if (so->so_state & SS_NOFDREF)
773 		panic("soclose: NOFDREF");
774 	kauth_cred_free(so->so_cred);
775 	so->so_state |= SS_NOFDREF;
776 	sofree(so);
777 	return (error);
778 }
779 
780 /*
781  * Must be called with the socket locked..  Will return with it unlocked.
782  */
783 int
784 soabort(struct socket *so)
785 {
786 	u_int refs;
787 	int error;
788 
789 	KASSERT(solocked(so));
790 	KASSERT(so->so_head == NULL);
791 
792 	so->so_aborting++;		/* XXX */
793 	error = (*so->so_proto->pr_usrreq)(so, PRU_ABORT, NULL,
794 	    NULL, NULL, NULL);
795 	refs = --so->so_aborting;	/* XXX */
796 	if (error || (refs == 0)) {
797 		sofree(so);
798 	} else {
799 		sounlock(so);
800 	}
801 	return error;
802 }
803 
804 int
805 soaccept(struct socket *so, struct mbuf *nam)
806 {
807 	int	error;
808 
809 	KASSERT(solocked(so));
810 
811 	error = 0;
812 	if ((so->so_state & SS_NOFDREF) == 0)
813 		panic("soaccept: !NOFDREF");
814 	so->so_state &= ~SS_NOFDREF;
815 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
816 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
817 		error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT,
818 		    NULL, nam, NULL, NULL);
819 	else
820 		error = ECONNABORTED;
821 
822 	return (error);
823 }
824 
825 int
826 soconnect(struct socket *so, struct mbuf *nam, struct lwp *l)
827 {
828 	int		error;
829 
830 	KASSERT(solocked(so));
831 
832 	if (so->so_options & SO_ACCEPTCONN)
833 		return (EOPNOTSUPP);
834 	/*
835 	 * If protocol is connection-based, can only connect once.
836 	 * Otherwise, if connected, try to disconnect first.
837 	 * This allows user to disconnect by connecting to, e.g.,
838 	 * a null address.
839 	 */
840 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
841 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
842 	    (error = sodisconnect(so))))
843 		error = EISCONN;
844 	else
845 		error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT,
846 		    NULL, nam, NULL, l);
847 	return (error);
848 }
849 
850 int
851 soconnect2(struct socket *so1, struct socket *so2)
852 {
853 	int	error;
854 
855 	KASSERT(solocked2(so1, so2));
856 
857 	error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2,
858 	    NULL, (struct mbuf *)so2, NULL, NULL);
859 	return (error);
860 }
861 
862 int
863 sodisconnect(struct socket *so)
864 {
865 	int	error;
866 
867 	KASSERT(solocked(so));
868 
869 	if ((so->so_state & SS_ISCONNECTED) == 0) {
870 		error = ENOTCONN;
871 	} else if (so->so_state & SS_ISDISCONNECTING) {
872 		error = EALREADY;
873 	} else {
874 		error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT,
875 		    NULL, NULL, NULL, NULL);
876 	}
877 	sodopendfree();
878 	return (error);
879 }
880 
881 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
882 /*
883  * Send on a socket.
884  * If send must go all at once and message is larger than
885  * send buffering, then hard error.
886  * Lock against other senders.
887  * If must go all at once and not enough room now, then
888  * inform user that this would block and do nothing.
889  * Otherwise, if nonblocking, send as much as possible.
890  * The data to be sent is described by "uio" if nonzero,
891  * otherwise by the mbuf chain "top" (which must be null
892  * if uio is not).  Data provided in mbuf chain must be small
893  * enough to send all at once.
894  *
895  * Returns nonzero on error, timeout or signal; callers
896  * must check for short counts if EINTR/ERESTART are returned.
897  * Data and control buffers are freed on return.
898  */
899 int
900 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
901 	struct mbuf *control, int flags, struct lwp *l)
902 {
903 	struct mbuf	**mp, *m;
904 	struct proc	*p;
905 	long		space, len, resid, clen, mlen;
906 	int		error, s, dontroute, atomic;
907 	short		wakeup_state = 0;
908 
909 	p = l->l_proc;
910 	sodopendfree();
911 	clen = 0;
912 
913 	/*
914 	 * solock() provides atomicity of access.  splsoftnet() prevents
915 	 * protocol processing soft interrupts from interrupting us and
916 	 * blocking (expensive).
917 	 */
918 	s = splsoftnet();
919 	solock(so);
920 	atomic = sosendallatonce(so) || top;
921 	if (uio)
922 		resid = uio->uio_resid;
923 	else
924 		resid = top->m_pkthdr.len;
925 	/*
926 	 * In theory resid should be unsigned.
927 	 * However, space must be signed, as it might be less than 0
928 	 * if we over-committed, and we must use a signed comparison
929 	 * of space and resid.  On the other hand, a negative resid
930 	 * causes us to loop sending 0-length segments to the protocol.
931 	 */
932 	if (resid < 0) {
933 		error = EINVAL;
934 		goto out;
935 	}
936 	dontroute =
937 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
938 	    (so->so_proto->pr_flags & PR_ATOMIC);
939 	l->l_ru.ru_msgsnd++;
940 	if (control)
941 		clen = control->m_len;
942  restart:
943 	if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
944 		goto out;
945 	do {
946 		if (so->so_state & SS_CANTSENDMORE) {
947 			error = EPIPE;
948 			goto release;
949 		}
950 		if (so->so_error) {
951 			error = so->so_error;
952 			so->so_error = 0;
953 			goto release;
954 		}
955 		if ((so->so_state & SS_ISCONNECTED) == 0) {
956 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
957 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
958 				    !(resid == 0 && clen != 0)) {
959 					error = ENOTCONN;
960 					goto release;
961 				}
962 			} else if (addr == 0) {
963 				error = EDESTADDRREQ;
964 				goto release;
965 			}
966 		}
967 		space = sbspace(&so->so_snd);
968 		if (flags & MSG_OOB)
969 			space += 1024;
970 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
971 		    clen > so->so_snd.sb_hiwat) {
972 			error = EMSGSIZE;
973 			goto release;
974 		}
975 		if (space < resid + clen &&
976 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
977 			if (so->so_nbio) {
978 				error = EWOULDBLOCK;
979 				goto release;
980 			}
981 			sbunlock(&so->so_snd);
982 			if (wakeup_state & SS_RESTARTSYS) {
983 				error = ERESTART;
984 				goto out;
985 			}
986 			error = sbwait(&so->so_snd);
987 			if (error)
988 				goto out;
989 			wakeup_state = so->so_state;
990 			goto restart;
991 		}
992 		wakeup_state = 0;
993 		mp = &top;
994 		space -= clen;
995 		do {
996 			if (uio == NULL) {
997 				/*
998 				 * Data is prepackaged in "top".
999 				 */
1000 				resid = 0;
1001 				if (flags & MSG_EOR)
1002 					top->m_flags |= M_EOR;
1003 			} else do {
1004 				sounlock(so);
1005 				splx(s);
1006 				if (top == NULL) {
1007 					m = m_gethdr(M_WAIT, MT_DATA);
1008 					mlen = MHLEN;
1009 					m->m_pkthdr.len = 0;
1010 					m->m_pkthdr.rcvif = NULL;
1011 				} else {
1012 					m = m_get(M_WAIT, MT_DATA);
1013 					mlen = MLEN;
1014 				}
1015 				MCLAIM(m, so->so_snd.sb_mowner);
1016 				if (sock_loan_thresh >= 0 &&
1017 				    uio->uio_iov->iov_len >= sock_loan_thresh &&
1018 				    space >= sock_loan_thresh &&
1019 				    (len = sosend_loan(so, uio, m,
1020 						       space)) != 0) {
1021 					SOSEND_COUNTER_INCR(&sosend_loan_big);
1022 					space -= len;
1023 					goto have_data;
1024 				}
1025 				if (resid >= MINCLSIZE && space >= MCLBYTES) {
1026 					SOSEND_COUNTER_INCR(&sosend_copy_big);
1027 					m_clget(m, M_DONTWAIT);
1028 					if ((m->m_flags & M_EXT) == 0)
1029 						goto nopages;
1030 					mlen = MCLBYTES;
1031 					if (atomic && top == 0) {
1032 						len = lmin(MCLBYTES - max_hdr,
1033 						    resid);
1034 						m->m_data += max_hdr;
1035 					} else
1036 						len = lmin(MCLBYTES, resid);
1037 					space -= len;
1038 				} else {
1039  nopages:
1040 					SOSEND_COUNTER_INCR(&sosend_copy_small);
1041 					len = lmin(lmin(mlen, resid), space);
1042 					space -= len;
1043 					/*
1044 					 * For datagram protocols, leave room
1045 					 * for protocol headers in first mbuf.
1046 					 */
1047 					if (atomic && top == 0 && len < mlen)
1048 						MH_ALIGN(m, len);
1049 				}
1050 				error = uiomove(mtod(m, void *), (int)len, uio);
1051  have_data:
1052 				resid = uio->uio_resid;
1053 				m->m_len = len;
1054 				*mp = m;
1055 				top->m_pkthdr.len += len;
1056 				s = splsoftnet();
1057 				solock(so);
1058 				if (error != 0)
1059 					goto release;
1060 				mp = &m->m_next;
1061 				if (resid <= 0) {
1062 					if (flags & MSG_EOR)
1063 						top->m_flags |= M_EOR;
1064 					break;
1065 				}
1066 			} while (space > 0 && atomic);
1067 
1068 			if (so->so_state & SS_CANTSENDMORE) {
1069 				error = EPIPE;
1070 				goto release;
1071 			}
1072 			if (dontroute)
1073 				so->so_options |= SO_DONTROUTE;
1074 			if (resid > 0)
1075 				so->so_state |= SS_MORETOCOME;
1076 			error = (*so->so_proto->pr_usrreq)(so,
1077 			    (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND,
1078 			    top, addr, control, curlwp);
1079 			if (dontroute)
1080 				so->so_options &= ~SO_DONTROUTE;
1081 			if (resid > 0)
1082 				so->so_state &= ~SS_MORETOCOME;
1083 			clen = 0;
1084 			control = NULL;
1085 			top = NULL;
1086 			mp = &top;
1087 			if (error != 0)
1088 				goto release;
1089 		} while (resid && space > 0);
1090 	} while (resid);
1091 
1092  release:
1093 	sbunlock(&so->so_snd);
1094  out:
1095 	sounlock(so);
1096 	splx(s);
1097 	if (top)
1098 		m_freem(top);
1099 	if (control)
1100 		m_freem(control);
1101 	return (error);
1102 }
1103 
1104 /*
1105  * Following replacement or removal of the first mbuf on the first
1106  * mbuf chain of a socket buffer, push necessary state changes back
1107  * into the socket buffer so that other consumers see the values
1108  * consistently.  'nextrecord' is the callers locally stored value of
1109  * the original value of sb->sb_mb->m_nextpkt which must be restored
1110  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
1111  */
1112 static void
1113 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
1114 {
1115 
1116 	KASSERT(solocked(sb->sb_so));
1117 
1118 	/*
1119 	 * First, update for the new value of nextrecord.  If necessary,
1120 	 * make it the first record.
1121 	 */
1122 	if (sb->sb_mb != NULL)
1123 		sb->sb_mb->m_nextpkt = nextrecord;
1124 	else
1125 		sb->sb_mb = nextrecord;
1126 
1127         /*
1128          * Now update any dependent socket buffer fields to reflect
1129          * the new state.  This is an inline of SB_EMPTY_FIXUP, with
1130          * the addition of a second clause that takes care of the
1131          * case where sb_mb has been updated, but remains the last
1132          * record.
1133          */
1134         if (sb->sb_mb == NULL) {
1135                 sb->sb_mbtail = NULL;
1136                 sb->sb_lastrecord = NULL;
1137         } else if (sb->sb_mb->m_nextpkt == NULL)
1138                 sb->sb_lastrecord = sb->sb_mb;
1139 }
1140 
1141 /*
1142  * Implement receive operations on a socket.
1143  * We depend on the way that records are added to the sockbuf
1144  * by sbappend*.  In particular, each record (mbufs linked through m_next)
1145  * must begin with an address if the protocol so specifies,
1146  * followed by an optional mbuf or mbufs containing ancillary data,
1147  * and then zero or more mbufs of data.
1148  * In order to avoid blocking network interrupts for the entire time here,
1149  * we splx() while doing the actual copy to user space.
1150  * Although the sockbuf is locked, new data may still be appended,
1151  * and thus we must maintain consistency of the sockbuf during that time.
1152  *
1153  * The caller may receive the data as a single mbuf chain by supplying
1154  * an mbuf **mp0 for use in returning the chain.  The uio is then used
1155  * only for the count in uio_resid.
1156  */
1157 int
1158 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
1159 	struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1160 {
1161 	struct lwp *l = curlwp;
1162 	struct mbuf	*m, **mp, *mt;
1163 	int atomic, flags, len, error, s, offset, moff, type, orig_resid;
1164 	const struct protosw	*pr;
1165 	struct mbuf	*nextrecord;
1166 	int		mbuf_removed = 0;
1167 	const struct domain *dom;
1168 	short		wakeup_state = 0;
1169 
1170 	pr = so->so_proto;
1171 	atomic = pr->pr_flags & PR_ATOMIC;
1172 	dom = pr->pr_domain;
1173 	mp = mp0;
1174 	type = 0;
1175 	orig_resid = uio->uio_resid;
1176 
1177 	if (paddr != NULL)
1178 		*paddr = NULL;
1179 	if (controlp != NULL)
1180 		*controlp = NULL;
1181 	if (flagsp != NULL)
1182 		flags = *flagsp &~ MSG_EOR;
1183 	else
1184 		flags = 0;
1185 
1186 	if ((flags & MSG_DONTWAIT) == 0)
1187 		sodopendfree();
1188 
1189 	if (flags & MSG_OOB) {
1190 		m = m_get(M_WAIT, MT_DATA);
1191 		solock(so);
1192 		error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m,
1193 		    (struct mbuf *)(long)(flags & MSG_PEEK), NULL, l);
1194 		sounlock(so);
1195 		if (error)
1196 			goto bad;
1197 		do {
1198 			error = uiomove(mtod(m, void *),
1199 			    (int) min(uio->uio_resid, m->m_len), uio);
1200 			m = m_free(m);
1201 		} while (uio->uio_resid > 0 && error == 0 && m);
1202  bad:
1203 		if (m != NULL)
1204 			m_freem(m);
1205 		return error;
1206 	}
1207 	if (mp != NULL)
1208 		*mp = NULL;
1209 
1210 	/*
1211 	 * solock() provides atomicity of access.  splsoftnet() prevents
1212 	 * protocol processing soft interrupts from interrupting us and
1213 	 * blocking (expensive).
1214 	 */
1215 	s = splsoftnet();
1216 	solock(so);
1217 	if (so->so_state & SS_ISCONFIRMING && uio->uio_resid)
1218 		(*pr->pr_usrreq)(so, PRU_RCVD, NULL, NULL, NULL, l);
1219 
1220  restart:
1221 	if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) {
1222 		sounlock(so);
1223 		splx(s);
1224 		return error;
1225 	}
1226 
1227 	m = so->so_rcv.sb_mb;
1228 	/*
1229 	 * If we have less data than requested, block awaiting more
1230 	 * (subject to any timeout) if:
1231 	 *   1. the current count is less than the low water mark,
1232 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1233 	 *	receive operation at once if we block (resid <= hiwat), or
1234 	 *   3. MSG_DONTWAIT is not set.
1235 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1236 	 * we have to do the receive in sections, and thus risk returning
1237 	 * a short count if a timeout or signal occurs after we start.
1238 	 */
1239 	if (m == NULL ||
1240 	    ((flags & MSG_DONTWAIT) == 0 &&
1241 	     so->so_rcv.sb_cc < uio->uio_resid &&
1242 	     (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1243 	      ((flags & MSG_WAITALL) &&
1244 	       uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1245 	     m->m_nextpkt == NULL && !atomic)) {
1246 #ifdef DIAGNOSTIC
1247 		if (m == NULL && so->so_rcv.sb_cc)
1248 			panic("receive 1");
1249 #endif
1250 		if (so->so_error) {
1251 			if (m != NULL)
1252 				goto dontblock;
1253 			error = so->so_error;
1254 			if ((flags & MSG_PEEK) == 0)
1255 				so->so_error = 0;
1256 			goto release;
1257 		}
1258 		if (so->so_state & SS_CANTRCVMORE) {
1259 			if (m != NULL)
1260 				goto dontblock;
1261 			else
1262 				goto release;
1263 		}
1264 		for (; m != NULL; m = m->m_next)
1265 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1266 				m = so->so_rcv.sb_mb;
1267 				goto dontblock;
1268 			}
1269 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1270 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1271 			error = ENOTCONN;
1272 			goto release;
1273 		}
1274 		if (uio->uio_resid == 0)
1275 			goto release;
1276 		if (so->so_nbio || (flags & MSG_DONTWAIT)) {
1277 			error = EWOULDBLOCK;
1278 			goto release;
1279 		}
1280 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
1281 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
1282 		sbunlock(&so->so_rcv);
1283 		if (wakeup_state & SS_RESTARTSYS)
1284 			error = ERESTART;
1285 		else
1286 			error = sbwait(&so->so_rcv);
1287 		if (error != 0) {
1288 			sounlock(so);
1289 			splx(s);
1290 			return error;
1291 		}
1292 		wakeup_state = so->so_state;
1293 		goto restart;
1294 	}
1295  dontblock:
1296 	/*
1297 	 * On entry here, m points to the first record of the socket buffer.
1298 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1299 	 * pointer to the next record in the socket buffer.  We must keep the
1300 	 * various socket buffer pointers and local stack versions of the
1301 	 * pointers in sync, pushing out modifications before dropping the
1302 	 * socket lock, and re-reading them when picking it up.
1303 	 *
1304 	 * Otherwise, we will race with the network stack appending new data
1305 	 * or records onto the socket buffer by using inconsistent/stale
1306 	 * versions of the field, possibly resulting in socket buffer
1307 	 * corruption.
1308 	 *
1309 	 * By holding the high-level sblock(), we prevent simultaneous
1310 	 * readers from pulling off the front of the socket buffer.
1311 	 */
1312 	if (l != NULL)
1313 		l->l_ru.ru_msgrcv++;
1314 	KASSERT(m == so->so_rcv.sb_mb);
1315 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
1316 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
1317 	nextrecord = m->m_nextpkt;
1318 	if (pr->pr_flags & PR_ADDR) {
1319 #ifdef DIAGNOSTIC
1320 		if (m->m_type != MT_SONAME)
1321 			panic("receive 1a");
1322 #endif
1323 		orig_resid = 0;
1324 		if (flags & MSG_PEEK) {
1325 			if (paddr)
1326 				*paddr = m_copy(m, 0, m->m_len);
1327 			m = m->m_next;
1328 		} else {
1329 			sbfree(&so->so_rcv, m);
1330 			mbuf_removed = 1;
1331 			if (paddr != NULL) {
1332 				*paddr = m;
1333 				so->so_rcv.sb_mb = m->m_next;
1334 				m->m_next = NULL;
1335 				m = so->so_rcv.sb_mb;
1336 			} else {
1337 				MFREE(m, so->so_rcv.sb_mb);
1338 				m = so->so_rcv.sb_mb;
1339 			}
1340 			sbsync(&so->so_rcv, nextrecord);
1341 		}
1342 	}
1343 
1344 	/*
1345 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1346 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1347 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1348 	 * perform externalization (or freeing if controlp == NULL).
1349 	 */
1350 	if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) {
1351 		struct mbuf *cm = NULL, *cmn;
1352 		struct mbuf **cme = &cm;
1353 
1354 		do {
1355 			if (flags & MSG_PEEK) {
1356 				if (controlp != NULL) {
1357 					*controlp = m_copy(m, 0, m->m_len);
1358 					controlp = &(*controlp)->m_next;
1359 				}
1360 				m = m->m_next;
1361 			} else {
1362 				sbfree(&so->so_rcv, m);
1363 				so->so_rcv.sb_mb = m->m_next;
1364 				m->m_next = NULL;
1365 				*cme = m;
1366 				cme = &(*cme)->m_next;
1367 				m = so->so_rcv.sb_mb;
1368 			}
1369 		} while (m != NULL && m->m_type == MT_CONTROL);
1370 		if ((flags & MSG_PEEK) == 0)
1371 			sbsync(&so->so_rcv, nextrecord);
1372 		for (; cm != NULL; cm = cmn) {
1373 			cmn = cm->m_next;
1374 			cm->m_next = NULL;
1375 			type = mtod(cm, struct cmsghdr *)->cmsg_type;
1376 			if (controlp != NULL) {
1377 				if (dom->dom_externalize != NULL &&
1378 				    type == SCM_RIGHTS) {
1379 					sounlock(so);
1380 					splx(s);
1381 					error = (*dom->dom_externalize)(cm, l);
1382 					s = splsoftnet();
1383 					solock(so);
1384 				}
1385 				*controlp = cm;
1386 				while (*controlp != NULL)
1387 					controlp = &(*controlp)->m_next;
1388 			} else {
1389 				/*
1390 				 * Dispose of any SCM_RIGHTS message that went
1391 				 * through the read path rather than recv.
1392 				 */
1393 				if (dom->dom_dispose != NULL &&
1394 				    type == SCM_RIGHTS) {
1395 				    	sounlock(so);
1396 					(*dom->dom_dispose)(cm);
1397 					solock(so);
1398 				}
1399 				m_freem(cm);
1400 			}
1401 		}
1402 		if (m != NULL)
1403 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1404 		else
1405 			nextrecord = so->so_rcv.sb_mb;
1406 		orig_resid = 0;
1407 	}
1408 
1409 	/* If m is non-NULL, we have some data to read. */
1410 	if (__predict_true(m != NULL)) {
1411 		type = m->m_type;
1412 		if (type == MT_OOBDATA)
1413 			flags |= MSG_OOB;
1414 	}
1415 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1416 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1417 
1418 	moff = 0;
1419 	offset = 0;
1420 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1421 		if (m->m_type == MT_OOBDATA) {
1422 			if (type != MT_OOBDATA)
1423 				break;
1424 		} else if (type == MT_OOBDATA)
1425 			break;
1426 #ifdef DIAGNOSTIC
1427 		else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
1428 			panic("receive 3");
1429 #endif
1430 		so->so_state &= ~SS_RCVATMARK;
1431 		wakeup_state = 0;
1432 		len = uio->uio_resid;
1433 		if (so->so_oobmark && len > so->so_oobmark - offset)
1434 			len = so->so_oobmark - offset;
1435 		if (len > m->m_len - moff)
1436 			len = m->m_len - moff;
1437 		/*
1438 		 * If mp is set, just pass back the mbufs.
1439 		 * Otherwise copy them out via the uio, then free.
1440 		 * Sockbuf must be consistent here (points to current mbuf,
1441 		 * it points to next record) when we drop priority;
1442 		 * we must note any additions to the sockbuf when we
1443 		 * block interrupts again.
1444 		 */
1445 		if (mp == NULL) {
1446 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1447 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1448 			sounlock(so);
1449 			splx(s);
1450 			error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1451 			s = splsoftnet();
1452 			solock(so);
1453 			if (error != 0) {
1454 				/*
1455 				 * If any part of the record has been removed
1456 				 * (such as the MT_SONAME mbuf, which will
1457 				 * happen when PR_ADDR, and thus also
1458 				 * PR_ATOMIC, is set), then drop the entire
1459 				 * record to maintain the atomicity of the
1460 				 * receive operation.
1461 				 *
1462 				 * This avoids a later panic("receive 1a")
1463 				 * when compiled with DIAGNOSTIC.
1464 				 */
1465 				if (m && mbuf_removed && atomic)
1466 					(void) sbdroprecord(&so->so_rcv);
1467 
1468 				goto release;
1469 			}
1470 		} else
1471 			uio->uio_resid -= len;
1472 		if (len == m->m_len - moff) {
1473 			if (m->m_flags & M_EOR)
1474 				flags |= MSG_EOR;
1475 			if (flags & MSG_PEEK) {
1476 				m = m->m_next;
1477 				moff = 0;
1478 			} else {
1479 				nextrecord = m->m_nextpkt;
1480 				sbfree(&so->so_rcv, m);
1481 				if (mp) {
1482 					*mp = m;
1483 					mp = &m->m_next;
1484 					so->so_rcv.sb_mb = m = m->m_next;
1485 					*mp = NULL;
1486 				} else {
1487 					MFREE(m, so->so_rcv.sb_mb);
1488 					m = so->so_rcv.sb_mb;
1489 				}
1490 				/*
1491 				 * If m != NULL, we also know that
1492 				 * so->so_rcv.sb_mb != NULL.
1493 				 */
1494 				KASSERT(so->so_rcv.sb_mb == m);
1495 				if (m) {
1496 					m->m_nextpkt = nextrecord;
1497 					if (nextrecord == NULL)
1498 						so->so_rcv.sb_lastrecord = m;
1499 				} else {
1500 					so->so_rcv.sb_mb = nextrecord;
1501 					SB_EMPTY_FIXUP(&so->so_rcv);
1502 				}
1503 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1504 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1505 			}
1506 		} else if (flags & MSG_PEEK)
1507 			moff += len;
1508 		else {
1509 			if (mp != NULL) {
1510 				mt = m_copym(m, 0, len, M_NOWAIT);
1511 				if (__predict_false(mt == NULL)) {
1512 					sounlock(so);
1513 					mt = m_copym(m, 0, len, M_WAIT);
1514 					solock(so);
1515 				}
1516 				*mp = mt;
1517 			}
1518 			m->m_data += len;
1519 			m->m_len -= len;
1520 			so->so_rcv.sb_cc -= len;
1521 		}
1522 		if (so->so_oobmark) {
1523 			if ((flags & MSG_PEEK) == 0) {
1524 				so->so_oobmark -= len;
1525 				if (so->so_oobmark == 0) {
1526 					so->so_state |= SS_RCVATMARK;
1527 					break;
1528 				}
1529 			} else {
1530 				offset += len;
1531 				if (offset == so->so_oobmark)
1532 					break;
1533 			}
1534 		}
1535 		if (flags & MSG_EOR)
1536 			break;
1537 		/*
1538 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1539 		 * we must not quit until "uio->uio_resid == 0" or an error
1540 		 * termination.  If a signal/timeout occurs, return
1541 		 * with a short count but without error.
1542 		 * Keep sockbuf locked against other readers.
1543 		 */
1544 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1545 		    !sosendallatonce(so) && !nextrecord) {
1546 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1547 				break;
1548 			/*
1549 			 * If we are peeking and the socket receive buffer is
1550 			 * full, stop since we can't get more data to peek at.
1551 			 */
1552 			if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
1553 				break;
1554 			/*
1555 			 * If we've drained the socket buffer, tell the
1556 			 * protocol in case it needs to do something to
1557 			 * get it filled again.
1558 			 */
1559 			if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1560 				(*pr->pr_usrreq)(so, PRU_RCVD,
1561 				    NULL, (struct mbuf *)(long)flags, NULL, l);
1562 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1563 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1564 			if (wakeup_state & SS_RESTARTSYS)
1565 				error = ERESTART;
1566 			else
1567 				error = sbwait(&so->so_rcv);
1568 			if (error != 0) {
1569 				sbunlock(&so->so_rcv);
1570 				sounlock(so);
1571 				splx(s);
1572 				return 0;
1573 			}
1574 			if ((m = so->so_rcv.sb_mb) != NULL)
1575 				nextrecord = m->m_nextpkt;
1576 			wakeup_state = so->so_state;
1577 		}
1578 	}
1579 
1580 	if (m && atomic) {
1581 		flags |= MSG_TRUNC;
1582 		if ((flags & MSG_PEEK) == 0)
1583 			(void) sbdroprecord(&so->so_rcv);
1584 	}
1585 	if ((flags & MSG_PEEK) == 0) {
1586 		if (m == NULL) {
1587 			/*
1588 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1589 			 * part makes sure sb_lastrecord is up-to-date if
1590 			 * there is still data in the socket buffer.
1591 			 */
1592 			so->so_rcv.sb_mb = nextrecord;
1593 			if (so->so_rcv.sb_mb == NULL) {
1594 				so->so_rcv.sb_mbtail = NULL;
1595 				so->so_rcv.sb_lastrecord = NULL;
1596 			} else if (nextrecord->m_nextpkt == NULL)
1597 				so->so_rcv.sb_lastrecord = nextrecord;
1598 		}
1599 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1600 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1601 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1602 			(*pr->pr_usrreq)(so, PRU_RCVD, NULL,
1603 			    (struct mbuf *)(long)flags, NULL, l);
1604 	}
1605 	if (orig_resid == uio->uio_resid && orig_resid &&
1606 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1607 		sbunlock(&so->so_rcv);
1608 		goto restart;
1609 	}
1610 
1611 	if (flagsp != NULL)
1612 		*flagsp |= flags;
1613  release:
1614 	sbunlock(&so->so_rcv);
1615 	sounlock(so);
1616 	splx(s);
1617 	return error;
1618 }
1619 
1620 int
1621 soshutdown(struct socket *so, int how)
1622 {
1623 	const struct protosw	*pr;
1624 	int	error;
1625 
1626 	KASSERT(solocked(so));
1627 
1628 	pr = so->so_proto;
1629 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1630 		return (EINVAL);
1631 
1632 	if (how == SHUT_RD || how == SHUT_RDWR) {
1633 		sorflush(so);
1634 		error = 0;
1635 	}
1636 	if (how == SHUT_WR || how == SHUT_RDWR)
1637 		error = (*pr->pr_usrreq)(so, PRU_SHUTDOWN, NULL,
1638 		    NULL, NULL, NULL);
1639 
1640 	return error;
1641 }
1642 
1643 void
1644 sorestart(struct socket *so)
1645 {
1646 	/*
1647 	 * An application has called close() on an fd on which another
1648 	 * of its threads has called a socket system call.
1649 	 * Mark this and wake everyone up, and code that would block again
1650 	 * instead returns ERESTART.
1651 	 * On system call re-entry the fd is validated and EBADF returned.
1652 	 * Any other fd will block again on the 2nd syscall.
1653 	 */
1654 	solock(so);
1655 	so->so_state |= SS_RESTARTSYS;
1656 	cv_broadcast(&so->so_cv);
1657 	cv_broadcast(&so->so_snd.sb_cv);
1658 	cv_broadcast(&so->so_rcv.sb_cv);
1659 	sounlock(so);
1660 }
1661 
1662 void
1663 sorflush(struct socket *so)
1664 {
1665 	struct sockbuf	*sb, asb;
1666 	const struct protosw	*pr;
1667 
1668 	KASSERT(solocked(so));
1669 
1670 	sb = &so->so_rcv;
1671 	pr = so->so_proto;
1672 	socantrcvmore(so);
1673 	sb->sb_flags |= SB_NOINTR;
1674 	(void )sblock(sb, M_WAITOK);
1675 	sbunlock(sb);
1676 	asb = *sb;
1677 	/*
1678 	 * Clear most of the sockbuf structure, but leave some of the
1679 	 * fields valid.
1680 	 */
1681 	memset(&sb->sb_startzero, 0,
1682 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1683 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) {
1684 		sounlock(so);
1685 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1686 		solock(so);
1687 	}
1688 	sbrelease(&asb, so);
1689 }
1690 
1691 /*
1692  * internal set SOL_SOCKET options
1693  */
1694 static int
1695 sosetopt1(struct socket *so, const struct sockopt *sopt)
1696 {
1697 	int error = EINVAL, optval, opt;
1698 	struct linger l;
1699 	struct timeval tv;
1700 
1701 	switch ((opt = sopt->sopt_name)) {
1702 
1703 	case SO_ACCEPTFILTER:
1704 		error = accept_filt_setopt(so, sopt);
1705 		KASSERT(solocked(so));
1706 		break;
1707 
1708   	case SO_LINGER:
1709  		error = sockopt_get(sopt, &l, sizeof(l));
1710 		solock(so);
1711  		if (error)
1712  			break;
1713  		if (l.l_linger < 0 || l.l_linger > USHRT_MAX ||
1714  		    l.l_linger > (INT_MAX / hz)) {
1715 			error = EDOM;
1716 			break;
1717 		}
1718  		so->so_linger = l.l_linger;
1719  		if (l.l_onoff)
1720  			so->so_options |= SO_LINGER;
1721  		else
1722  			so->so_options &= ~SO_LINGER;
1723    		break;
1724 
1725 	case SO_DEBUG:
1726 	case SO_KEEPALIVE:
1727 	case SO_DONTROUTE:
1728 	case SO_USELOOPBACK:
1729 	case SO_BROADCAST:
1730 	case SO_REUSEADDR:
1731 	case SO_REUSEPORT:
1732 	case SO_OOBINLINE:
1733 	case SO_TIMESTAMP:
1734 #ifdef SO_OTIMESTAMP
1735 	case SO_OTIMESTAMP:
1736 #endif
1737 		error = sockopt_getint(sopt, &optval);
1738 		solock(so);
1739 		if (error)
1740 			break;
1741 		if (optval)
1742 			so->so_options |= opt;
1743 		else
1744 			so->so_options &= ~opt;
1745 		break;
1746 
1747 	case SO_SNDBUF:
1748 	case SO_RCVBUF:
1749 	case SO_SNDLOWAT:
1750 	case SO_RCVLOWAT:
1751 		error = sockopt_getint(sopt, &optval);
1752 		solock(so);
1753 		if (error)
1754 			break;
1755 
1756 		/*
1757 		 * Values < 1 make no sense for any of these
1758 		 * options, so disallow them.
1759 		 */
1760 		if (optval < 1) {
1761 			error = EINVAL;
1762 			break;
1763 		}
1764 
1765 		switch (opt) {
1766 		case SO_SNDBUF:
1767 			if (sbreserve(&so->so_snd, (u_long)optval, so) == 0) {
1768 				error = ENOBUFS;
1769 				break;
1770 			}
1771 			so->so_snd.sb_flags &= ~SB_AUTOSIZE;
1772 			break;
1773 
1774 		case SO_RCVBUF:
1775 			if (sbreserve(&so->so_rcv, (u_long)optval, so) == 0) {
1776 				error = ENOBUFS;
1777 				break;
1778 			}
1779 			so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1780 			break;
1781 
1782 		/*
1783 		 * Make sure the low-water is never greater than
1784 		 * the high-water.
1785 		 */
1786 		case SO_SNDLOWAT:
1787 			if (optval > so->so_snd.sb_hiwat)
1788 				optval = so->so_snd.sb_hiwat;
1789 
1790 			so->so_snd.sb_lowat = optval;
1791 			break;
1792 
1793 		case SO_RCVLOWAT:
1794 			if (optval > so->so_rcv.sb_hiwat)
1795 				optval = so->so_rcv.sb_hiwat;
1796 
1797 			so->so_rcv.sb_lowat = optval;
1798 			break;
1799 		}
1800 		break;
1801 
1802 #ifdef COMPAT_50
1803 	case SO_OSNDTIMEO:
1804 	case SO_ORCVTIMEO: {
1805 		struct timeval50 otv;
1806 		error = sockopt_get(sopt, &otv, sizeof(otv));
1807 		if (error) {
1808 			solock(so);
1809 			break;
1810 		}
1811 		timeval50_to_timeval(&otv, &tv);
1812 		opt = opt == SO_OSNDTIMEO ? SO_SNDTIMEO : SO_RCVTIMEO;
1813 		error = 0;
1814 		/*FALLTHROUGH*/
1815 	}
1816 #endif /* COMPAT_50 */
1817 
1818 	case SO_SNDTIMEO:
1819 	case SO_RCVTIMEO:
1820 		if (error)
1821 			error = sockopt_get(sopt, &tv, sizeof(tv));
1822 		solock(so);
1823 		if (error)
1824 			break;
1825 
1826 		if (tv.tv_sec > (INT_MAX - tv.tv_usec / tick) / hz) {
1827 			error = EDOM;
1828 			break;
1829 		}
1830 
1831 		optval = tv.tv_sec * hz + tv.tv_usec / tick;
1832 		if (optval == 0 && tv.tv_usec != 0)
1833 			optval = 1;
1834 
1835 		switch (opt) {
1836 		case SO_SNDTIMEO:
1837 			so->so_snd.sb_timeo = optval;
1838 			break;
1839 		case SO_RCVTIMEO:
1840 			so->so_rcv.sb_timeo = optval;
1841 			break;
1842 		}
1843 		break;
1844 
1845 	default:
1846 		solock(so);
1847 		error = ENOPROTOOPT;
1848 		break;
1849 	}
1850 	KASSERT(solocked(so));
1851 	return error;
1852 }
1853 
1854 int
1855 sosetopt(struct socket *so, struct sockopt *sopt)
1856 {
1857 	int error, prerr;
1858 
1859 	if (sopt->sopt_level == SOL_SOCKET) {
1860 		error = sosetopt1(so, sopt);
1861 		KASSERT(solocked(so));
1862 	} else {
1863 		error = ENOPROTOOPT;
1864 		solock(so);
1865 	}
1866 
1867 	if ((error == 0 || error == ENOPROTOOPT) &&
1868 	    so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) {
1869 		/* give the protocol stack a shot */
1870 		prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, sopt);
1871 		if (prerr == 0)
1872 			error = 0;
1873 		else if (prerr != ENOPROTOOPT)
1874 			error = prerr;
1875 	}
1876 	sounlock(so);
1877 	return error;
1878 }
1879 
1880 /*
1881  * so_setsockopt() is a wrapper providing a sockopt structure for sosetopt()
1882  */
1883 int
1884 so_setsockopt(struct lwp *l, struct socket *so, int level, int name,
1885     const void *val, size_t valsize)
1886 {
1887 	struct sockopt sopt;
1888 	int error;
1889 
1890 	KASSERT(valsize == 0 || val != NULL);
1891 
1892 	sockopt_init(&sopt, level, name, valsize);
1893 	sockopt_set(&sopt, val, valsize);
1894 
1895 	error = sosetopt(so, &sopt);
1896 
1897 	sockopt_destroy(&sopt);
1898 
1899 	return error;
1900 }
1901 
1902 /*
1903  * internal get SOL_SOCKET options
1904  */
1905 static int
1906 sogetopt1(struct socket *so, struct sockopt *sopt)
1907 {
1908 	int error, optval, opt;
1909 	struct linger l;
1910 	struct timeval tv;
1911 
1912 	switch ((opt = sopt->sopt_name)) {
1913 
1914 	case SO_ACCEPTFILTER:
1915 		error = accept_filt_getopt(so, sopt);
1916 		break;
1917 
1918 	case SO_LINGER:
1919 		l.l_onoff = (so->so_options & SO_LINGER) ? 1 : 0;
1920 		l.l_linger = so->so_linger;
1921 
1922 		error = sockopt_set(sopt, &l, sizeof(l));
1923 		break;
1924 
1925 	case SO_USELOOPBACK:
1926 	case SO_DONTROUTE:
1927 	case SO_DEBUG:
1928 	case SO_KEEPALIVE:
1929 	case SO_REUSEADDR:
1930 	case SO_REUSEPORT:
1931 	case SO_BROADCAST:
1932 	case SO_OOBINLINE:
1933 	case SO_TIMESTAMP:
1934 #ifdef SO_OTIMESTAMP
1935 	case SO_OTIMESTAMP:
1936 #endif
1937 		error = sockopt_setint(sopt, (so->so_options & opt) ? 1 : 0);
1938 		break;
1939 
1940 	case SO_TYPE:
1941 		error = sockopt_setint(sopt, so->so_type);
1942 		break;
1943 
1944 	case SO_ERROR:
1945 		error = sockopt_setint(sopt, so->so_error);
1946 		so->so_error = 0;
1947 		break;
1948 
1949 	case SO_SNDBUF:
1950 		error = sockopt_setint(sopt, so->so_snd.sb_hiwat);
1951 		break;
1952 
1953 	case SO_RCVBUF:
1954 		error = sockopt_setint(sopt, so->so_rcv.sb_hiwat);
1955 		break;
1956 
1957 	case SO_SNDLOWAT:
1958 		error = sockopt_setint(sopt, so->so_snd.sb_lowat);
1959 		break;
1960 
1961 	case SO_RCVLOWAT:
1962 		error = sockopt_setint(sopt, so->so_rcv.sb_lowat);
1963 		break;
1964 
1965 #ifdef COMPAT_50
1966 	case SO_OSNDTIMEO:
1967 	case SO_ORCVTIMEO: {
1968 		struct timeval50 otv;
1969 
1970 		optval = (opt == SO_OSNDTIMEO ?
1971 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1972 
1973 		otv.tv_sec = optval / hz;
1974 		otv.tv_usec = (optval % hz) * tick;
1975 
1976 		error = sockopt_set(sopt, &otv, sizeof(otv));
1977 		break;
1978 	}
1979 #endif /* COMPAT_50 */
1980 
1981 	case SO_SNDTIMEO:
1982 	case SO_RCVTIMEO:
1983 		optval = (opt == SO_SNDTIMEO ?
1984 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1985 
1986 		tv.tv_sec = optval / hz;
1987 		tv.tv_usec = (optval % hz) * tick;
1988 
1989 		error = sockopt_set(sopt, &tv, sizeof(tv));
1990 		break;
1991 
1992 	case SO_OVERFLOWED:
1993 		error = sockopt_setint(sopt, so->so_rcv.sb_overflowed);
1994 		break;
1995 
1996 	default:
1997 		error = ENOPROTOOPT;
1998 		break;
1999 	}
2000 
2001 	return (error);
2002 }
2003 
2004 int
2005 sogetopt(struct socket *so, struct sockopt *sopt)
2006 {
2007 	int		error;
2008 
2009 	solock(so);
2010 	if (sopt->sopt_level != SOL_SOCKET) {
2011 		if (so->so_proto && so->so_proto->pr_ctloutput) {
2012 			error = ((*so->so_proto->pr_ctloutput)
2013 			    (PRCO_GETOPT, so, sopt));
2014 		} else
2015 			error = (ENOPROTOOPT);
2016 	} else {
2017 		error = sogetopt1(so, sopt);
2018 	}
2019 	sounlock(so);
2020 	return (error);
2021 }
2022 
2023 /*
2024  * alloc sockopt data buffer buffer
2025  *	- will be released at destroy
2026  */
2027 static int
2028 sockopt_alloc(struct sockopt *sopt, size_t len, km_flag_t kmflag)
2029 {
2030 
2031 	KASSERT(sopt->sopt_size == 0);
2032 
2033 	if (len > sizeof(sopt->sopt_buf)) {
2034 		sopt->sopt_data = kmem_zalloc(len, kmflag);
2035 		if (sopt->sopt_data == NULL)
2036 			return ENOMEM;
2037 	} else
2038 		sopt->sopt_data = sopt->sopt_buf;
2039 
2040 	sopt->sopt_size = len;
2041 	return 0;
2042 }
2043 
2044 /*
2045  * initialise sockopt storage
2046  *	- MAY sleep during allocation
2047  */
2048 void
2049 sockopt_init(struct sockopt *sopt, int level, int name, size_t size)
2050 {
2051 
2052 	memset(sopt, 0, sizeof(*sopt));
2053 
2054 	sopt->sopt_level = level;
2055 	sopt->sopt_name = name;
2056 	(void)sockopt_alloc(sopt, size, KM_SLEEP);
2057 }
2058 
2059 /*
2060  * destroy sockopt storage
2061  *	- will release any held memory references
2062  */
2063 void
2064 sockopt_destroy(struct sockopt *sopt)
2065 {
2066 
2067 	if (sopt->sopt_data != sopt->sopt_buf)
2068 		kmem_free(sopt->sopt_data, sopt->sopt_size);
2069 
2070 	memset(sopt, 0, sizeof(*sopt));
2071 }
2072 
2073 /*
2074  * set sockopt value
2075  *	- value is copied into sockopt
2076  * 	- memory is allocated when necessary, will not sleep
2077  */
2078 int
2079 sockopt_set(struct sockopt *sopt, const void *buf, size_t len)
2080 {
2081 	int error;
2082 
2083 	if (sopt->sopt_size == 0) {
2084 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2085 		if (error)
2086 			return error;
2087 	}
2088 
2089 	KASSERT(sopt->sopt_size == len);
2090 	memcpy(sopt->sopt_data, buf, len);
2091 	return 0;
2092 }
2093 
2094 /*
2095  * common case of set sockopt integer value
2096  */
2097 int
2098 sockopt_setint(struct sockopt *sopt, int val)
2099 {
2100 
2101 	return sockopt_set(sopt, &val, sizeof(int));
2102 }
2103 
2104 /*
2105  * get sockopt value
2106  *	- correct size must be given
2107  */
2108 int
2109 sockopt_get(const struct sockopt *sopt, void *buf, size_t len)
2110 {
2111 
2112 	if (sopt->sopt_size != len)
2113 		return EINVAL;
2114 
2115 	memcpy(buf, sopt->sopt_data, len);
2116 	return 0;
2117 }
2118 
2119 /*
2120  * common case of get sockopt integer value
2121  */
2122 int
2123 sockopt_getint(const struct sockopt *sopt, int *valp)
2124 {
2125 
2126 	return sockopt_get(sopt, valp, sizeof(int));
2127 }
2128 
2129 /*
2130  * set sockopt value from mbuf
2131  *	- ONLY for legacy code
2132  *	- mbuf is released by sockopt
2133  *	- will not sleep
2134  */
2135 int
2136 sockopt_setmbuf(struct sockopt *sopt, struct mbuf *m)
2137 {
2138 	size_t len;
2139 	int error;
2140 
2141 	len = m_length(m);
2142 
2143 	if (sopt->sopt_size == 0) {
2144 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2145 		if (error)
2146 			return error;
2147 	}
2148 
2149 	KASSERT(sopt->sopt_size == len);
2150 	m_copydata(m, 0, len, sopt->sopt_data);
2151 	m_freem(m);
2152 
2153 	return 0;
2154 }
2155 
2156 /*
2157  * get sockopt value into mbuf
2158  *	- ONLY for legacy code
2159  *	- mbuf to be released by the caller
2160  *	- will not sleep
2161  */
2162 struct mbuf *
2163 sockopt_getmbuf(const struct sockopt *sopt)
2164 {
2165 	struct mbuf *m;
2166 
2167 	if (sopt->sopt_size > MCLBYTES)
2168 		return NULL;
2169 
2170 	m = m_get(M_DONTWAIT, MT_SOOPTS);
2171 	if (m == NULL)
2172 		return NULL;
2173 
2174 	if (sopt->sopt_size > MLEN) {
2175 		MCLGET(m, M_DONTWAIT);
2176 		if ((m->m_flags & M_EXT) == 0) {
2177 			m_free(m);
2178 			return NULL;
2179 		}
2180 	}
2181 
2182 	memcpy(mtod(m, void *), sopt->sopt_data, sopt->sopt_size);
2183 	m->m_len = sopt->sopt_size;
2184 
2185 	return m;
2186 }
2187 
2188 void
2189 sohasoutofband(struct socket *so)
2190 {
2191 
2192 	fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
2193 	selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, NOTE_SUBMIT);
2194 }
2195 
2196 static void
2197 filt_sordetach(struct knote *kn)
2198 {
2199 	struct socket	*so;
2200 
2201 	so = ((file_t *)kn->kn_obj)->f_data;
2202 	solock(so);
2203 	SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
2204 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
2205 		so->so_rcv.sb_flags &= ~SB_KNOTE;
2206 	sounlock(so);
2207 }
2208 
2209 /*ARGSUSED*/
2210 static int
2211 filt_soread(struct knote *kn, long hint)
2212 {
2213 	struct socket	*so;
2214 	int rv;
2215 
2216 	so = ((file_t *)kn->kn_obj)->f_data;
2217 	if (hint != NOTE_SUBMIT)
2218 		solock(so);
2219 	kn->kn_data = so->so_rcv.sb_cc;
2220 	if (so->so_state & SS_CANTRCVMORE) {
2221 		kn->kn_flags |= EV_EOF;
2222 		kn->kn_fflags = so->so_error;
2223 		rv = 1;
2224 	} else if (so->so_error)	/* temporary udp error */
2225 		rv = 1;
2226 	else if (kn->kn_sfflags & NOTE_LOWAT)
2227 		rv = (kn->kn_data >= kn->kn_sdata);
2228 	else
2229 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
2230 	if (hint != NOTE_SUBMIT)
2231 		sounlock(so);
2232 	return rv;
2233 }
2234 
2235 static void
2236 filt_sowdetach(struct knote *kn)
2237 {
2238 	struct socket	*so;
2239 
2240 	so = ((file_t *)kn->kn_obj)->f_data;
2241 	solock(so);
2242 	SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
2243 	if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
2244 		so->so_snd.sb_flags &= ~SB_KNOTE;
2245 	sounlock(so);
2246 }
2247 
2248 /*ARGSUSED*/
2249 static int
2250 filt_sowrite(struct knote *kn, long hint)
2251 {
2252 	struct socket	*so;
2253 	int rv;
2254 
2255 	so = ((file_t *)kn->kn_obj)->f_data;
2256 	if (hint != NOTE_SUBMIT)
2257 		solock(so);
2258 	kn->kn_data = sbspace(&so->so_snd);
2259 	if (so->so_state & SS_CANTSENDMORE) {
2260 		kn->kn_flags |= EV_EOF;
2261 		kn->kn_fflags = so->so_error;
2262 		rv = 1;
2263 	} else if (so->so_error)	/* temporary udp error */
2264 		rv = 1;
2265 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2266 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
2267 		rv = 0;
2268 	else if (kn->kn_sfflags & NOTE_LOWAT)
2269 		rv = (kn->kn_data >= kn->kn_sdata);
2270 	else
2271 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
2272 	if (hint != NOTE_SUBMIT)
2273 		sounlock(so);
2274 	return rv;
2275 }
2276 
2277 /*ARGSUSED*/
2278 static int
2279 filt_solisten(struct knote *kn, long hint)
2280 {
2281 	struct socket	*so;
2282 	int rv;
2283 
2284 	so = ((file_t *)kn->kn_obj)->f_data;
2285 
2286 	/*
2287 	 * Set kn_data to number of incoming connections, not
2288 	 * counting partial (incomplete) connections.
2289 	 */
2290 	if (hint != NOTE_SUBMIT)
2291 		solock(so);
2292 	kn->kn_data = so->so_qlen;
2293 	rv = (kn->kn_data > 0);
2294 	if (hint != NOTE_SUBMIT)
2295 		sounlock(so);
2296 	return rv;
2297 }
2298 
2299 static const struct filterops solisten_filtops =
2300 	{ 1, NULL, filt_sordetach, filt_solisten };
2301 static const struct filterops soread_filtops =
2302 	{ 1, NULL, filt_sordetach, filt_soread };
2303 static const struct filterops sowrite_filtops =
2304 	{ 1, NULL, filt_sowdetach, filt_sowrite };
2305 
2306 int
2307 soo_kqfilter(struct file *fp, struct knote *kn)
2308 {
2309 	struct socket	*so;
2310 	struct sockbuf	*sb;
2311 
2312 	so = ((file_t *)kn->kn_obj)->f_data;
2313 	solock(so);
2314 	switch (kn->kn_filter) {
2315 	case EVFILT_READ:
2316 		if (so->so_options & SO_ACCEPTCONN)
2317 			kn->kn_fop = &solisten_filtops;
2318 		else
2319 			kn->kn_fop = &soread_filtops;
2320 		sb = &so->so_rcv;
2321 		break;
2322 	case EVFILT_WRITE:
2323 		kn->kn_fop = &sowrite_filtops;
2324 		sb = &so->so_snd;
2325 		break;
2326 	default:
2327 		sounlock(so);
2328 		return (EINVAL);
2329 	}
2330 	SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
2331 	sb->sb_flags |= SB_KNOTE;
2332 	sounlock(so);
2333 	return (0);
2334 }
2335 
2336 static int
2337 sodopoll(struct socket *so, int events)
2338 {
2339 	int revents;
2340 
2341 	revents = 0;
2342 
2343 	if (events & (POLLIN | POLLRDNORM))
2344 		if (soreadable(so))
2345 			revents |= events & (POLLIN | POLLRDNORM);
2346 
2347 	if (events & (POLLOUT | POLLWRNORM))
2348 		if (sowritable(so))
2349 			revents |= events & (POLLOUT | POLLWRNORM);
2350 
2351 	if (events & (POLLPRI | POLLRDBAND))
2352 		if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
2353 			revents |= events & (POLLPRI | POLLRDBAND);
2354 
2355 	return revents;
2356 }
2357 
2358 int
2359 sopoll(struct socket *so, int events)
2360 {
2361 	int revents = 0;
2362 
2363 #ifndef DIAGNOSTIC
2364 	/*
2365 	 * Do a quick, unlocked check in expectation that the socket
2366 	 * will be ready for I/O.  Don't do this check if DIAGNOSTIC,
2367 	 * as the solocked() assertions will fail.
2368 	 */
2369 	if ((revents = sodopoll(so, events)) != 0)
2370 		return revents;
2371 #endif
2372 
2373 	solock(so);
2374 	if ((revents = sodopoll(so, events)) == 0) {
2375 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2376 			selrecord(curlwp, &so->so_rcv.sb_sel);
2377 			so->so_rcv.sb_flags |= SB_NOTIFY;
2378 		}
2379 
2380 		if (events & (POLLOUT | POLLWRNORM)) {
2381 			selrecord(curlwp, &so->so_snd.sb_sel);
2382 			so->so_snd.sb_flags |= SB_NOTIFY;
2383 		}
2384 	}
2385 	sounlock(so);
2386 
2387 	return revents;
2388 }
2389 
2390 
2391 #include <sys/sysctl.h>
2392 
2393 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO);
2394 
2395 /*
2396  * sysctl helper routine for kern.somaxkva.  ensures that the given
2397  * value is not too small.
2398  * (XXX should we maybe make sure it's not too large as well?)
2399  */
2400 static int
2401 sysctl_kern_somaxkva(SYSCTLFN_ARGS)
2402 {
2403 	int error, new_somaxkva;
2404 	struct sysctlnode node;
2405 
2406 	new_somaxkva = somaxkva;
2407 	node = *rnode;
2408 	node.sysctl_data = &new_somaxkva;
2409 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2410 	if (error || newp == NULL)
2411 		return (error);
2412 
2413 	if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */
2414 		return (EINVAL);
2415 
2416 	mutex_enter(&so_pendfree_lock);
2417 	somaxkva = new_somaxkva;
2418 	cv_broadcast(&socurkva_cv);
2419 	mutex_exit(&so_pendfree_lock);
2420 
2421 	return (error);
2422 }
2423 
2424 static void
2425 sysctl_kern_somaxkva_setup(void)
2426 {
2427 
2428 	KASSERT(socket_sysctllog == NULL);
2429 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2430 		       CTLFLAG_PERMANENT,
2431 		       CTLTYPE_NODE, "kern", NULL,
2432 		       NULL, 0, NULL, 0,
2433 		       CTL_KERN, CTL_EOL);
2434 
2435 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2436 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2437 		       CTLTYPE_INT, "somaxkva",
2438 		       SYSCTL_DESCR("Maximum amount of kernel memory to be "
2439 				    "used for socket buffers"),
2440 		       sysctl_kern_somaxkva, 0, NULL, 0,
2441 		       CTL_KERN, KERN_SOMAXKVA, CTL_EOL);
2442 }
2443