xref: /netbsd-src/sys/kern/uipc_socket.c (revision b1c86f5f087524e68db12794ee9c3e3da1ab17a0)
1 /*	$NetBSD: uipc_socket.c,v 1.200 2009/12/30 22:12:12 elad 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 1.200 2009/12/30 22:12:12 elad 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.h>
102 
103 MALLOC_DEFINE(M_SOOPTS, "soopts", "socket options");
104 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
105 
106 extern const struct fileops socketops;
107 
108 extern int	somaxconn;			/* patchable (XXX sysctl) */
109 int		somaxconn = SOMAXCONN;
110 kmutex_t	*softnet_lock;
111 
112 #ifdef SOSEND_COUNTERS
113 #include <sys/device.h>
114 
115 static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
116     NULL, "sosend", "loan big");
117 static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
118     NULL, "sosend", "copy big");
119 static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
120     NULL, "sosend", "copy small");
121 static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
122     NULL, "sosend", "kva limit");
123 
124 #define	SOSEND_COUNTER_INCR(ev)		(ev)->ev_count++
125 
126 EVCNT_ATTACH_STATIC(sosend_loan_big);
127 EVCNT_ATTACH_STATIC(sosend_copy_big);
128 EVCNT_ATTACH_STATIC(sosend_copy_small);
129 EVCNT_ATTACH_STATIC(sosend_kvalimit);
130 #else
131 
132 #define	SOSEND_COUNTER_INCR(ev)		/* nothing */
133 
134 #endif /* SOSEND_COUNTERS */
135 
136 static struct callback_entry sokva_reclaimerentry;
137 
138 #if defined(SOSEND_NO_LOAN) || defined(MULTIPROCESSOR)
139 int sock_loan_thresh = -1;
140 #else
141 int sock_loan_thresh = 4096;
142 #endif
143 
144 static kmutex_t so_pendfree_lock;
145 static struct mbuf *so_pendfree;
146 
147 #ifndef SOMAXKVA
148 #define	SOMAXKVA (16 * 1024 * 1024)
149 #endif
150 int somaxkva = SOMAXKVA;
151 static int socurkva;
152 static kcondvar_t socurkva_cv;
153 
154 static kauth_listener_t socket_listener;
155 
156 #define	SOCK_LOAN_CHUNK		65536
157 
158 static size_t sodopendfree(void);
159 static size_t sodopendfreel(void);
160 
161 static void sysctl_kern_somaxkva_setup(void);
162 static struct sysctllog *socket_sysctllog;
163 
164 static vsize_t
165 sokvareserve(struct socket *so, vsize_t len)
166 {
167 	int error;
168 
169 	mutex_enter(&so_pendfree_lock);
170 	while (socurkva + len > somaxkva) {
171 		size_t freed;
172 
173 		/*
174 		 * try to do pendfree.
175 		 */
176 
177 		freed = sodopendfreel();
178 
179 		/*
180 		 * if some kva was freed, try again.
181 		 */
182 
183 		if (freed)
184 			continue;
185 
186 		SOSEND_COUNTER_INCR(&sosend_kvalimit);
187 		error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock);
188 		if (error) {
189 			len = 0;
190 			break;
191 		}
192 	}
193 	socurkva += len;
194 	mutex_exit(&so_pendfree_lock);
195 	return len;
196 }
197 
198 static void
199 sokvaunreserve(vsize_t len)
200 {
201 
202 	mutex_enter(&so_pendfree_lock);
203 	socurkva -= len;
204 	cv_broadcast(&socurkva_cv);
205 	mutex_exit(&so_pendfree_lock);
206 }
207 
208 /*
209  * sokvaalloc: allocate kva for loan.
210  */
211 
212 vaddr_t
213 sokvaalloc(vsize_t len, struct socket *so)
214 {
215 	vaddr_t lva;
216 
217 	/*
218 	 * reserve kva.
219 	 */
220 
221 	if (sokvareserve(so, len) == 0)
222 		return 0;
223 
224 	/*
225 	 * allocate kva.
226 	 */
227 
228 	lva = uvm_km_alloc(kernel_map, len, 0, UVM_KMF_VAONLY | UVM_KMF_WAITVA);
229 	if (lva == 0) {
230 		sokvaunreserve(len);
231 		return (0);
232 	}
233 
234 	return lva;
235 }
236 
237 /*
238  * sokvafree: free kva for loan.
239  */
240 
241 void
242 sokvafree(vaddr_t sva, vsize_t len)
243 {
244 
245 	/*
246 	 * free kva.
247 	 */
248 
249 	uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY);
250 
251 	/*
252 	 * unreserve kva.
253 	 */
254 
255 	sokvaunreserve(len);
256 }
257 
258 static void
259 sodoloanfree(struct vm_page **pgs, void *buf, size_t size)
260 {
261 	vaddr_t sva, eva;
262 	vsize_t len;
263 	int npgs;
264 
265 	KASSERT(pgs != NULL);
266 
267 	eva = round_page((vaddr_t) buf + size);
268 	sva = trunc_page((vaddr_t) buf);
269 	len = eva - sva;
270 	npgs = len >> PAGE_SHIFT;
271 
272 	pmap_kremove(sva, len);
273 	pmap_update(pmap_kernel());
274 	uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE);
275 	sokvafree(sva, len);
276 }
277 
278 static size_t
279 sodopendfree(void)
280 {
281 	size_t rv;
282 
283 	if (__predict_true(so_pendfree == NULL))
284 		return 0;
285 
286 	mutex_enter(&so_pendfree_lock);
287 	rv = sodopendfreel();
288 	mutex_exit(&so_pendfree_lock);
289 
290 	return rv;
291 }
292 
293 /*
294  * sodopendfreel: free mbufs on "pendfree" list.
295  * unlock and relock so_pendfree_lock when freeing mbufs.
296  *
297  * => called with so_pendfree_lock held.
298  */
299 
300 static size_t
301 sodopendfreel(void)
302 {
303 	struct mbuf *m, *next;
304 	size_t rv = 0;
305 
306 	KASSERT(mutex_owned(&so_pendfree_lock));
307 
308 	while (so_pendfree != NULL) {
309 		m = so_pendfree;
310 		so_pendfree = NULL;
311 		mutex_exit(&so_pendfree_lock);
312 
313 		for (; m != NULL; m = next) {
314 			next = m->m_next;
315 			KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0);
316 			KASSERT(m->m_ext.ext_refcnt == 0);
317 
318 			rv += m->m_ext.ext_size;
319 			sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf,
320 			    m->m_ext.ext_size);
321 			pool_cache_put(mb_cache, m);
322 		}
323 
324 		mutex_enter(&so_pendfree_lock);
325 	}
326 
327 	return (rv);
328 }
329 
330 void
331 soloanfree(struct mbuf *m, void *buf, size_t size, void *arg)
332 {
333 
334 	KASSERT(m != NULL);
335 
336 	/*
337 	 * postpone freeing mbuf.
338 	 *
339 	 * we can't do it in interrupt context
340 	 * because we need to put kva back to kernel_map.
341 	 */
342 
343 	mutex_enter(&so_pendfree_lock);
344 	m->m_next = so_pendfree;
345 	so_pendfree = m;
346 	cv_broadcast(&socurkva_cv);
347 	mutex_exit(&so_pendfree_lock);
348 }
349 
350 static long
351 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space)
352 {
353 	struct iovec *iov = uio->uio_iov;
354 	vaddr_t sva, eva;
355 	vsize_t len;
356 	vaddr_t lva;
357 	int npgs, error;
358 	vaddr_t va;
359 	int i;
360 
361 	if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace))
362 		return (0);
363 
364 	if (iov->iov_len < (size_t) space)
365 		space = iov->iov_len;
366 	if (space > SOCK_LOAN_CHUNK)
367 		space = SOCK_LOAN_CHUNK;
368 
369 	eva = round_page((vaddr_t) iov->iov_base + space);
370 	sva = trunc_page((vaddr_t) iov->iov_base);
371 	len = eva - sva;
372 	npgs = len >> PAGE_SHIFT;
373 
374 	KASSERT(npgs <= M_EXT_MAXPAGES);
375 
376 	lva = sokvaalloc(len, so);
377 	if (lva == 0)
378 		return 0;
379 
380 	error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len,
381 	    m->m_ext.ext_pgs, UVM_LOAN_TOPAGE);
382 	if (error) {
383 		sokvafree(lva, len);
384 		return (0);
385 	}
386 
387 	for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE)
388 		pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
389 		    VM_PROT_READ, 0);
390 	pmap_update(pmap_kernel());
391 
392 	lva += (vaddr_t) iov->iov_base & PAGE_MASK;
393 
394 	MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so);
395 	m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
396 
397 	uio->uio_resid -= space;
398 	/* uio_offset not updated, not set/used for write(2) */
399 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space;
400 	uio->uio_iov->iov_len -= space;
401 	if (uio->uio_iov->iov_len == 0) {
402 		uio->uio_iov++;
403 		uio->uio_iovcnt--;
404 	}
405 
406 	return (space);
407 }
408 
409 static int
410 sokva_reclaim_callback(struct callback_entry *ce, void *obj, void *arg)
411 {
412 
413 	KASSERT(ce == &sokva_reclaimerentry);
414 	KASSERT(obj == NULL);
415 
416 	sodopendfree();
417 	if (!vm_map_starved_p(kernel_map)) {
418 		return CALLBACK_CHAIN_ABORT;
419 	}
420 	return CALLBACK_CHAIN_CONTINUE;
421 }
422 
423 struct mbuf *
424 getsombuf(struct socket *so, int type)
425 {
426 	struct mbuf *m;
427 
428 	m = m_get(M_WAIT, type);
429 	MCLAIM(m, so->so_mowner);
430 	return m;
431 }
432 
433 static int
434 socket_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
435     void *arg0, void *arg1, void *arg2, void *arg3)
436 {
437 	int result;
438 	enum kauth_network_req req;
439 
440 	result = KAUTH_RESULT_DEFER;
441 	req = (enum kauth_network_req)arg0;
442 
443 	if ((action != KAUTH_NETWORK_SOCKET) &&
444 	    (action != KAUTH_NETWORK_BIND))
445 		return result;
446 
447 	switch (req) {
448 	case KAUTH_REQ_NETWORK_BIND_PORT:
449 		result = KAUTH_RESULT_ALLOW;
450 		break;
451 
452 	case KAUTH_REQ_NETWORK_SOCKET_DROP: {
453 		/* Normal users can only drop their own connections. */
454 		struct socket *so = (struct socket *)arg1;
455 
456 		if (proc_uidmatch(cred, so->so_cred))
457 			result = KAUTH_RESULT_ALLOW;
458 
459 		break;
460 		}
461 
462 	case KAUTH_REQ_NETWORK_SOCKET_OPEN:
463 		/* We allow "raw" routing/bluetooth sockets to anyone. */
464 		if ((u_long)arg1 == PF_ROUTE || (u_long)arg1 == PF_BLUETOOTH)
465 			result = KAUTH_RESULT_ALLOW;
466 		else {
467 			/* Privileged, let secmodel handle this. */
468 			if ((u_long)arg2 == SOCK_RAW)
469 				break;
470 		}
471 
472 		result = KAUTH_RESULT_ALLOW;
473 
474 		break;
475 
476 	case KAUTH_REQ_NETWORK_SOCKET_CANSEE:
477 		result = KAUTH_RESULT_ALLOW;
478 
479 		break;
480 
481 	default:
482 		break;
483 	}
484 
485 	return result;
486 }
487 
488 void
489 soinit(void)
490 {
491 
492 	sysctl_kern_somaxkva_setup();
493 
494 	mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM);
495 	softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
496 	cv_init(&socurkva_cv, "sokva");
497 	soinit2();
498 
499 	/* Set the initial adjusted socket buffer size. */
500 	if (sb_max_set(sb_max))
501 		panic("bad initial sb_max value: %lu", sb_max);
502 
503 	callback_register(&vm_map_to_kernel(kernel_map)->vmk_reclaim_callback,
504 	    &sokva_reclaimerentry, NULL, sokva_reclaim_callback);
505 
506 	socket_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK,
507 	    socket_listener_cb, NULL);
508 }
509 
510 /*
511  * Socket operation routines.
512  * These routines are called by the routines in
513  * sys_socket.c or from a system process, and
514  * implement the semantics of socket operations by
515  * switching out to the protocol specific routines.
516  */
517 /*ARGSUSED*/
518 int
519 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l,
520 	 struct socket *lockso)
521 {
522 	const struct protosw	*prp;
523 	struct socket	*so;
524 	uid_t		uid;
525 	int		error;
526 	kmutex_t	*lock;
527 
528 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET,
529 	    KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type),
530 	    KAUTH_ARG(proto));
531 	if (error != 0)
532 		return error;
533 
534 	if (proto)
535 		prp = pffindproto(dom, proto, type);
536 	else
537 		prp = pffindtype(dom, type);
538 	if (prp == NULL) {
539 		/* no support for domain */
540 		if (pffinddomain(dom) == 0)
541 			return EAFNOSUPPORT;
542 		/* no support for socket type */
543 		if (proto == 0 && type != 0)
544 			return EPROTOTYPE;
545 		return EPROTONOSUPPORT;
546 	}
547 	if (prp->pr_usrreq == NULL)
548 		return EPROTONOSUPPORT;
549 	if (prp->pr_type != type)
550 		return EPROTOTYPE;
551 
552 	so = soget(true);
553 	so->so_type = type;
554 	so->so_proto = prp;
555 	so->so_send = sosend;
556 	so->so_receive = soreceive;
557 #ifdef MBUFTRACE
558 	so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner;
559 	so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner;
560 	so->so_mowner = &prp->pr_domain->dom_mowner;
561 #endif
562 	uid = kauth_cred_geteuid(l->l_cred);
563 	so->so_uidinfo = uid_find(uid);
564 	so->so_cpid = l->l_proc->p_pid;
565 	if (lockso != NULL) {
566 		/* Caller wants us to share a lock. */
567 		lock = lockso->so_lock;
568 		so->so_lock = lock;
569 		mutex_obj_hold(lock);
570 		mutex_enter(lock);
571 	} else {
572 		/* Lock assigned and taken during PRU_ATTACH. */
573 	}
574 	error = (*prp->pr_usrreq)(so, PRU_ATTACH, NULL,
575 	    (struct mbuf *)(long)proto, NULL, l);
576 	KASSERT(solocked(so));
577 	if (error != 0) {
578 		so->so_state |= SS_NOFDREF;
579 		sofree(so);
580 		return error;
581 	}
582 	so->so_cred = kauth_cred_dup(l->l_cred);
583 	sounlock(so);
584 	*aso = so;
585 	return 0;
586 }
587 
588 /* On success, write file descriptor to fdout and return zero.  On
589  * failure, return non-zero; *fdout will be undefined.
590  */
591 int
592 fsocreate(int domain, struct socket **sop, int type, int protocol,
593     struct lwp *l, int *fdout)
594 {
595 	struct socket	*so;
596 	struct file	*fp;
597 	int		fd, error;
598 
599 	if ((error = fd_allocfile(&fp, &fd)) != 0)
600 		return (error);
601 	fp->f_flag = FREAD|FWRITE;
602 	fp->f_type = DTYPE_SOCKET;
603 	fp->f_ops = &socketops;
604 	error = socreate(domain, &so, type, protocol, l, NULL);
605 	if (error != 0) {
606 		fd_abort(curproc, fp, fd);
607 	} else {
608 		if (sop != NULL)
609 			*sop = so;
610 		fp->f_data = so;
611 		fd_affix(curproc, fp, fd);
612 		*fdout = fd;
613 	}
614 	return error;
615 }
616 
617 int
618 sofamily(const struct socket *so)
619 {
620 	const struct protosw *pr;
621 	const struct domain *dom;
622 
623 	if ((pr = so->so_proto) == NULL)
624 		return AF_UNSPEC;
625 	if ((dom = pr->pr_domain) == NULL)
626 		return AF_UNSPEC;
627 	return dom->dom_family;
628 }
629 
630 int
631 sobind(struct socket *so, struct mbuf *nam, struct lwp *l)
632 {
633 	int	error;
634 
635 	solock(so);
636 	error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, NULL, nam, NULL, l);
637 	sounlock(so);
638 	return error;
639 }
640 
641 int
642 solisten(struct socket *so, int backlog, struct lwp *l)
643 {
644 	int	error;
645 
646 	solock(so);
647 	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
648 	    SS_ISDISCONNECTING)) != 0) {
649 	    	sounlock(so);
650 		return (EOPNOTSUPP);
651 	}
652 	error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, NULL,
653 	    NULL, NULL, l);
654 	if (error != 0) {
655 		sounlock(so);
656 		return error;
657 	}
658 	if (TAILQ_EMPTY(&so->so_q))
659 		so->so_options |= SO_ACCEPTCONN;
660 	if (backlog < 0)
661 		backlog = 0;
662 	so->so_qlimit = min(backlog, somaxconn);
663 	sounlock(so);
664 	return 0;
665 }
666 
667 void
668 sofree(struct socket *so)
669 {
670 	u_int refs;
671 
672 	KASSERT(solocked(so));
673 
674 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
675 		sounlock(so);
676 		return;
677 	}
678 	if (so->so_head) {
679 		/*
680 		 * We must not decommission a socket that's on the accept(2)
681 		 * queue.  If we do, then accept(2) may hang after select(2)
682 		 * indicated that the listening socket was ready.
683 		 */
684 		if (!soqremque(so, 0)) {
685 			sounlock(so);
686 			return;
687 		}
688 	}
689 	if (so->so_rcv.sb_hiwat)
690 		(void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0,
691 		    RLIM_INFINITY);
692 	if (so->so_snd.sb_hiwat)
693 		(void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0,
694 		    RLIM_INFINITY);
695 	sbrelease(&so->so_snd, so);
696 	KASSERT(!cv_has_waiters(&so->so_cv));
697 	KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv));
698 	KASSERT(!cv_has_waiters(&so->so_snd.sb_cv));
699 	sorflush(so);
700 	refs = so->so_aborting;	/* XXX */
701 	/* Remove acccept filter if one is present. */
702 	if (so->so_accf != NULL)
703 		(void)accept_filt_clear(so);
704 	sounlock(so);
705 	if (refs == 0)		/* XXX */
706 		soput(so);
707 }
708 
709 /*
710  * Close a socket on last file table reference removal.
711  * Initiate disconnect if connected.
712  * Free socket when disconnect complete.
713  */
714 int
715 soclose(struct socket *so)
716 {
717 	struct socket	*so2;
718 	int		error;
719 	int		error2;
720 
721 	error = 0;
722 	solock(so);
723 	if (so->so_options & SO_ACCEPTCONN) {
724 		for (;;) {
725 			if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) {
726 				KASSERT(solocked2(so, so2));
727 				(void) soqremque(so2, 0);
728 				/* soabort drops the lock. */
729 				(void) soabort(so2);
730 				solock(so);
731 				continue;
732 			}
733 			if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) {
734 				KASSERT(solocked2(so, so2));
735 				(void) soqremque(so2, 1);
736 				/* soabort drops the lock. */
737 				(void) soabort(so2);
738 				solock(so);
739 				continue;
740 			}
741 			break;
742 		}
743 	}
744 	if (so->so_pcb == 0)
745 		goto discard;
746 	if (so->so_state & SS_ISCONNECTED) {
747 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
748 			error = sodisconnect(so);
749 			if (error)
750 				goto drop;
751 		}
752 		if (so->so_options & SO_LINGER) {
753 			if ((so->so_state & SS_ISDISCONNECTING) && so->so_nbio)
754 				goto drop;
755 			while (so->so_state & SS_ISCONNECTED) {
756 				error = sowait(so, true, so->so_linger * hz);
757 				if (error)
758 					break;
759 			}
760 		}
761 	}
762  drop:
763 	if (so->so_pcb) {
764 		error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH,
765 		    NULL, NULL, NULL, NULL);
766 		if (error == 0)
767 			error = error2;
768 	}
769  discard:
770 	if (so->so_state & SS_NOFDREF)
771 		panic("soclose: NOFDREF");
772 	kauth_cred_free(so->so_cred);
773 	so->so_state |= SS_NOFDREF;
774 	sofree(so);
775 	return (error);
776 }
777 
778 /*
779  * Must be called with the socket locked..  Will return with it unlocked.
780  */
781 int
782 soabort(struct socket *so)
783 {
784 	u_int refs;
785 	int error;
786 
787 	KASSERT(solocked(so));
788 	KASSERT(so->so_head == NULL);
789 
790 	so->so_aborting++;		/* XXX */
791 	error = (*so->so_proto->pr_usrreq)(so, PRU_ABORT, NULL,
792 	    NULL, NULL, NULL);
793 	refs = --so->so_aborting;	/* XXX */
794 	if (error || (refs == 0)) {
795 		sofree(so);
796 	} else {
797 		sounlock(so);
798 	}
799 	return error;
800 }
801 
802 int
803 soaccept(struct socket *so, struct mbuf *nam)
804 {
805 	int	error;
806 
807 	KASSERT(solocked(so));
808 
809 	error = 0;
810 	if ((so->so_state & SS_NOFDREF) == 0)
811 		panic("soaccept: !NOFDREF");
812 	so->so_state &= ~SS_NOFDREF;
813 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
814 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
815 		error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT,
816 		    NULL, nam, NULL, NULL);
817 	else
818 		error = ECONNABORTED;
819 
820 	return (error);
821 }
822 
823 int
824 soconnect(struct socket *so, struct mbuf *nam, struct lwp *l)
825 {
826 	int		error;
827 
828 	KASSERT(solocked(so));
829 
830 	if (so->so_options & SO_ACCEPTCONN)
831 		return (EOPNOTSUPP);
832 	/*
833 	 * If protocol is connection-based, can only connect once.
834 	 * Otherwise, if connected, try to disconnect first.
835 	 * This allows user to disconnect by connecting to, e.g.,
836 	 * a null address.
837 	 */
838 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
839 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
840 	    (error = sodisconnect(so))))
841 		error = EISCONN;
842 	else
843 		error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT,
844 		    NULL, nam, NULL, l);
845 	return (error);
846 }
847 
848 int
849 soconnect2(struct socket *so1, struct socket *so2)
850 {
851 	int	error;
852 
853 	KASSERT(solocked2(so1, so2));
854 
855 	error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2,
856 	    NULL, (struct mbuf *)so2, NULL, NULL);
857 	return (error);
858 }
859 
860 int
861 sodisconnect(struct socket *so)
862 {
863 	int	error;
864 
865 	KASSERT(solocked(so));
866 
867 	if ((so->so_state & SS_ISCONNECTED) == 0) {
868 		error = ENOTCONN;
869 	} else if (so->so_state & SS_ISDISCONNECTING) {
870 		error = EALREADY;
871 	} else {
872 		error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT,
873 		    NULL, NULL, NULL, NULL);
874 	}
875 	sodopendfree();
876 	return (error);
877 }
878 
879 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
880 /*
881  * Send on a socket.
882  * If send must go all at once and message is larger than
883  * send buffering, then hard error.
884  * Lock against other senders.
885  * If must go all at once and not enough room now, then
886  * inform user that this would block and do nothing.
887  * Otherwise, if nonblocking, send as much as possible.
888  * The data to be sent is described by "uio" if nonzero,
889  * otherwise by the mbuf chain "top" (which must be null
890  * if uio is not).  Data provided in mbuf chain must be small
891  * enough to send all at once.
892  *
893  * Returns nonzero on error, timeout or signal; callers
894  * must check for short counts if EINTR/ERESTART are returned.
895  * Data and control buffers are freed on return.
896  */
897 int
898 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
899 	struct mbuf *control, int flags, struct lwp *l)
900 {
901 	struct mbuf	**mp, *m;
902 	struct proc	*p;
903 	long		space, len, resid, clen, mlen;
904 	int		error, s, dontroute, atomic;
905 	short		wakeup_state = 0;
906 
907 	p = l->l_proc;
908 	sodopendfree();
909 	clen = 0;
910 
911 	/*
912 	 * solock() provides atomicity of access.  splsoftnet() prevents
913 	 * protocol processing soft interrupts from interrupting us and
914 	 * blocking (expensive).
915 	 */
916 	s = splsoftnet();
917 	solock(so);
918 	atomic = sosendallatonce(so) || top;
919 	if (uio)
920 		resid = uio->uio_resid;
921 	else
922 		resid = top->m_pkthdr.len;
923 	/*
924 	 * In theory resid should be unsigned.
925 	 * However, space must be signed, as it might be less than 0
926 	 * if we over-committed, and we must use a signed comparison
927 	 * of space and resid.  On the other hand, a negative resid
928 	 * causes us to loop sending 0-length segments to the protocol.
929 	 */
930 	if (resid < 0) {
931 		error = EINVAL;
932 		goto out;
933 	}
934 	dontroute =
935 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
936 	    (so->so_proto->pr_flags & PR_ATOMIC);
937 	l->l_ru.ru_msgsnd++;
938 	if (control)
939 		clen = control->m_len;
940  restart:
941 	if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
942 		goto out;
943 	do {
944 		if (so->so_state & SS_CANTSENDMORE) {
945 			error = EPIPE;
946 			goto release;
947 		}
948 		if (so->so_error) {
949 			error = so->so_error;
950 			so->so_error = 0;
951 			goto release;
952 		}
953 		if ((so->so_state & SS_ISCONNECTED) == 0) {
954 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
955 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
956 				    !(resid == 0 && clen != 0)) {
957 					error = ENOTCONN;
958 					goto release;
959 				}
960 			} else if (addr == 0) {
961 				error = EDESTADDRREQ;
962 				goto release;
963 			}
964 		}
965 		space = sbspace(&so->so_snd);
966 		if (flags & MSG_OOB)
967 			space += 1024;
968 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
969 		    clen > so->so_snd.sb_hiwat) {
970 			error = EMSGSIZE;
971 			goto release;
972 		}
973 		if (space < resid + clen &&
974 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
975 			if (so->so_nbio) {
976 				error = EWOULDBLOCK;
977 				goto release;
978 			}
979 			sbunlock(&so->so_snd);
980 			if (wakeup_state & SS_RESTARTSYS) {
981 				error = ERESTART;
982 				goto out;
983 			}
984 			error = sbwait(&so->so_snd);
985 			if (error)
986 				goto out;
987 			wakeup_state = so->so_state;
988 			goto restart;
989 		}
990 		wakeup_state = 0;
991 		mp = &top;
992 		space -= clen;
993 		do {
994 			if (uio == NULL) {
995 				/*
996 				 * Data is prepackaged in "top".
997 				 */
998 				resid = 0;
999 				if (flags & MSG_EOR)
1000 					top->m_flags |= M_EOR;
1001 			} else do {
1002 				sounlock(so);
1003 				splx(s);
1004 				if (top == NULL) {
1005 					m = m_gethdr(M_WAIT, MT_DATA);
1006 					mlen = MHLEN;
1007 					m->m_pkthdr.len = 0;
1008 					m->m_pkthdr.rcvif = NULL;
1009 				} else {
1010 					m = m_get(M_WAIT, MT_DATA);
1011 					mlen = MLEN;
1012 				}
1013 				MCLAIM(m, so->so_snd.sb_mowner);
1014 				if (sock_loan_thresh >= 0 &&
1015 				    uio->uio_iov->iov_len >= sock_loan_thresh &&
1016 				    space >= sock_loan_thresh &&
1017 				    (len = sosend_loan(so, uio, m,
1018 						       space)) != 0) {
1019 					SOSEND_COUNTER_INCR(&sosend_loan_big);
1020 					space -= len;
1021 					goto have_data;
1022 				}
1023 				if (resid >= MINCLSIZE && space >= MCLBYTES) {
1024 					SOSEND_COUNTER_INCR(&sosend_copy_big);
1025 					m_clget(m, M_WAIT);
1026 					if ((m->m_flags & M_EXT) == 0)
1027 						goto nopages;
1028 					mlen = MCLBYTES;
1029 					if (atomic && top == 0) {
1030 						len = lmin(MCLBYTES - max_hdr,
1031 						    resid);
1032 						m->m_data += max_hdr;
1033 					} else
1034 						len = lmin(MCLBYTES, resid);
1035 					space -= len;
1036 				} else {
1037  nopages:
1038 					SOSEND_COUNTER_INCR(&sosend_copy_small);
1039 					len = lmin(lmin(mlen, resid), space);
1040 					space -= len;
1041 					/*
1042 					 * For datagram protocols, leave room
1043 					 * for protocol headers in first mbuf.
1044 					 */
1045 					if (atomic && top == 0 && len < mlen)
1046 						MH_ALIGN(m, len);
1047 				}
1048 				error = uiomove(mtod(m, void *), (int)len, uio);
1049  have_data:
1050 				resid = uio->uio_resid;
1051 				m->m_len = len;
1052 				*mp = m;
1053 				top->m_pkthdr.len += len;
1054 				s = splsoftnet();
1055 				solock(so);
1056 				if (error != 0)
1057 					goto release;
1058 				mp = &m->m_next;
1059 				if (resid <= 0) {
1060 					if (flags & MSG_EOR)
1061 						top->m_flags |= M_EOR;
1062 					break;
1063 				}
1064 			} while (space > 0 && atomic);
1065 
1066 			if (so->so_state & SS_CANTSENDMORE) {
1067 				error = EPIPE;
1068 				goto release;
1069 			}
1070 			if (dontroute)
1071 				so->so_options |= SO_DONTROUTE;
1072 			if (resid > 0)
1073 				so->so_state |= SS_MORETOCOME;
1074 			error = (*so->so_proto->pr_usrreq)(so,
1075 			    (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND,
1076 			    top, addr, control, curlwp);
1077 			if (dontroute)
1078 				so->so_options &= ~SO_DONTROUTE;
1079 			if (resid > 0)
1080 				so->so_state &= ~SS_MORETOCOME;
1081 			clen = 0;
1082 			control = NULL;
1083 			top = NULL;
1084 			mp = &top;
1085 			if (error != 0)
1086 				goto release;
1087 		} while (resid && space > 0);
1088 	} while (resid);
1089 
1090  release:
1091 	sbunlock(&so->so_snd);
1092  out:
1093 	sounlock(so);
1094 	splx(s);
1095 	if (top)
1096 		m_freem(top);
1097 	if (control)
1098 		m_freem(control);
1099 	return (error);
1100 }
1101 
1102 /*
1103  * Following replacement or removal of the first mbuf on the first
1104  * mbuf chain of a socket buffer, push necessary state changes back
1105  * into the socket buffer so that other consumers see the values
1106  * consistently.  'nextrecord' is the callers locally stored value of
1107  * the original value of sb->sb_mb->m_nextpkt which must be restored
1108  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
1109  */
1110 static void
1111 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
1112 {
1113 
1114 	KASSERT(solocked(sb->sb_so));
1115 
1116 	/*
1117 	 * First, update for the new value of nextrecord.  If necessary,
1118 	 * make it the first record.
1119 	 */
1120 	if (sb->sb_mb != NULL)
1121 		sb->sb_mb->m_nextpkt = nextrecord;
1122 	else
1123 		sb->sb_mb = nextrecord;
1124 
1125         /*
1126          * Now update any dependent socket buffer fields to reflect
1127          * the new state.  This is an inline of SB_EMPTY_FIXUP, with
1128          * the addition of a second clause that takes care of the
1129          * case where sb_mb has been updated, but remains the last
1130          * record.
1131          */
1132         if (sb->sb_mb == NULL) {
1133                 sb->sb_mbtail = NULL;
1134                 sb->sb_lastrecord = NULL;
1135         } else if (sb->sb_mb->m_nextpkt == NULL)
1136                 sb->sb_lastrecord = sb->sb_mb;
1137 }
1138 
1139 /*
1140  * Implement receive operations on a socket.
1141  * We depend on the way that records are added to the sockbuf
1142  * by sbappend*.  In particular, each record (mbufs linked through m_next)
1143  * must begin with an address if the protocol so specifies,
1144  * followed by an optional mbuf or mbufs containing ancillary data,
1145  * and then zero or more mbufs of data.
1146  * In order to avoid blocking network interrupts for the entire time here,
1147  * we splx() while doing the actual copy to user space.
1148  * Although the sockbuf is locked, new data may still be appended,
1149  * and thus we must maintain consistency of the sockbuf during that time.
1150  *
1151  * The caller may receive the data as a single mbuf chain by supplying
1152  * an mbuf **mp0 for use in returning the chain.  The uio is then used
1153  * only for the count in uio_resid.
1154  */
1155 int
1156 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
1157 	struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1158 {
1159 	struct lwp *l = curlwp;
1160 	struct mbuf	*m, **mp, *mt;
1161 	int atomic, flags, len, error, s, offset, moff, type, orig_resid;
1162 	const struct protosw	*pr;
1163 	struct mbuf	*nextrecord;
1164 	int		mbuf_removed = 0;
1165 	const struct domain *dom;
1166 	short		wakeup_state = 0;
1167 
1168 	pr = so->so_proto;
1169 	atomic = pr->pr_flags & PR_ATOMIC;
1170 	dom = pr->pr_domain;
1171 	mp = mp0;
1172 	type = 0;
1173 	orig_resid = uio->uio_resid;
1174 
1175 	if (paddr != NULL)
1176 		*paddr = NULL;
1177 	if (controlp != NULL)
1178 		*controlp = NULL;
1179 	if (flagsp != NULL)
1180 		flags = *flagsp &~ MSG_EOR;
1181 	else
1182 		flags = 0;
1183 
1184 	if ((flags & MSG_DONTWAIT) == 0)
1185 		sodopendfree();
1186 
1187 	if (flags & MSG_OOB) {
1188 		m = m_get(M_WAIT, MT_DATA);
1189 		solock(so);
1190 		error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m,
1191 		    (struct mbuf *)(long)(flags & MSG_PEEK), NULL, l);
1192 		sounlock(so);
1193 		if (error)
1194 			goto bad;
1195 		do {
1196 			error = uiomove(mtod(m, void *),
1197 			    (int) min(uio->uio_resid, m->m_len), uio);
1198 			m = m_free(m);
1199 		} while (uio->uio_resid > 0 && error == 0 && m);
1200  bad:
1201 		if (m != NULL)
1202 			m_freem(m);
1203 		return error;
1204 	}
1205 	if (mp != NULL)
1206 		*mp = NULL;
1207 
1208 	/*
1209 	 * solock() provides atomicity of access.  splsoftnet() prevents
1210 	 * protocol processing soft interrupts from interrupting us and
1211 	 * blocking (expensive).
1212 	 */
1213 	s = splsoftnet();
1214 	solock(so);
1215 	if (so->so_state & SS_ISCONFIRMING && uio->uio_resid)
1216 		(*pr->pr_usrreq)(so, PRU_RCVD, NULL, NULL, NULL, l);
1217 
1218  restart:
1219 	if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) {
1220 		sounlock(so);
1221 		splx(s);
1222 		return error;
1223 	}
1224 
1225 	m = so->so_rcv.sb_mb;
1226 	/*
1227 	 * If we have less data than requested, block awaiting more
1228 	 * (subject to any timeout) if:
1229 	 *   1. the current count is less than the low water mark,
1230 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1231 	 *	receive operation at once if we block (resid <= hiwat), or
1232 	 *   3. MSG_DONTWAIT is not set.
1233 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1234 	 * we have to do the receive in sections, and thus risk returning
1235 	 * a short count if a timeout or signal occurs after we start.
1236 	 */
1237 	if (m == NULL ||
1238 	    ((flags & MSG_DONTWAIT) == 0 &&
1239 	     so->so_rcv.sb_cc < uio->uio_resid &&
1240 	     (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1241 	      ((flags & MSG_WAITALL) &&
1242 	       uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1243 	     m->m_nextpkt == NULL && !atomic)) {
1244 #ifdef DIAGNOSTIC
1245 		if (m == NULL && so->so_rcv.sb_cc)
1246 			panic("receive 1");
1247 #endif
1248 		if (so->so_error) {
1249 			if (m != NULL)
1250 				goto dontblock;
1251 			error = so->so_error;
1252 			if ((flags & MSG_PEEK) == 0)
1253 				so->so_error = 0;
1254 			goto release;
1255 		}
1256 		if (so->so_state & SS_CANTRCVMORE) {
1257 			if (m != NULL)
1258 				goto dontblock;
1259 			else
1260 				goto release;
1261 		}
1262 		for (; m != NULL; m = m->m_next)
1263 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1264 				m = so->so_rcv.sb_mb;
1265 				goto dontblock;
1266 			}
1267 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1268 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1269 			error = ENOTCONN;
1270 			goto release;
1271 		}
1272 		if (uio->uio_resid == 0)
1273 			goto release;
1274 		if (so->so_nbio || (flags & MSG_DONTWAIT)) {
1275 			error = EWOULDBLOCK;
1276 			goto release;
1277 		}
1278 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
1279 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
1280 		sbunlock(&so->so_rcv);
1281 		if (wakeup_state & SS_RESTARTSYS)
1282 			error = ERESTART;
1283 		else
1284 			error = sbwait(&so->so_rcv);
1285 		if (error != 0) {
1286 			sounlock(so);
1287 			splx(s);
1288 			return error;
1289 		}
1290 		wakeup_state = so->so_state;
1291 		goto restart;
1292 	}
1293  dontblock:
1294 	/*
1295 	 * On entry here, m points to the first record of the socket buffer.
1296 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1297 	 * pointer to the next record in the socket buffer.  We must keep the
1298 	 * various socket buffer pointers and local stack versions of the
1299 	 * pointers in sync, pushing out modifications before dropping the
1300 	 * socket lock, and re-reading them when picking it up.
1301 	 *
1302 	 * Otherwise, we will race with the network stack appending new data
1303 	 * or records onto the socket buffer by using inconsistent/stale
1304 	 * versions of the field, possibly resulting in socket buffer
1305 	 * corruption.
1306 	 *
1307 	 * By holding the high-level sblock(), we prevent simultaneous
1308 	 * readers from pulling off the front of the socket buffer.
1309 	 */
1310 	if (l != NULL)
1311 		l->l_ru.ru_msgrcv++;
1312 	KASSERT(m == so->so_rcv.sb_mb);
1313 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
1314 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
1315 	nextrecord = m->m_nextpkt;
1316 	if (pr->pr_flags & PR_ADDR) {
1317 #ifdef DIAGNOSTIC
1318 		if (m->m_type != MT_SONAME)
1319 			panic("receive 1a");
1320 #endif
1321 		orig_resid = 0;
1322 		if (flags & MSG_PEEK) {
1323 			if (paddr)
1324 				*paddr = m_copy(m, 0, m->m_len);
1325 			m = m->m_next;
1326 		} else {
1327 			sbfree(&so->so_rcv, m);
1328 			mbuf_removed = 1;
1329 			if (paddr != NULL) {
1330 				*paddr = m;
1331 				so->so_rcv.sb_mb = m->m_next;
1332 				m->m_next = NULL;
1333 				m = so->so_rcv.sb_mb;
1334 			} else {
1335 				MFREE(m, so->so_rcv.sb_mb);
1336 				m = so->so_rcv.sb_mb;
1337 			}
1338 			sbsync(&so->so_rcv, nextrecord);
1339 		}
1340 	}
1341 
1342 	/*
1343 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1344 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1345 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1346 	 * perform externalization (or freeing if controlp == NULL).
1347 	 */
1348 	if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) {
1349 		struct mbuf *cm = NULL, *cmn;
1350 		struct mbuf **cme = &cm;
1351 
1352 		do {
1353 			if (flags & MSG_PEEK) {
1354 				if (controlp != NULL) {
1355 					*controlp = m_copy(m, 0, m->m_len);
1356 					controlp = &(*controlp)->m_next;
1357 				}
1358 				m = m->m_next;
1359 			} else {
1360 				sbfree(&so->so_rcv, m);
1361 				so->so_rcv.sb_mb = m->m_next;
1362 				m->m_next = NULL;
1363 				*cme = m;
1364 				cme = &(*cme)->m_next;
1365 				m = so->so_rcv.sb_mb;
1366 			}
1367 		} while (m != NULL && m->m_type == MT_CONTROL);
1368 		if ((flags & MSG_PEEK) == 0)
1369 			sbsync(&so->so_rcv, nextrecord);
1370 		for (; cm != NULL; cm = cmn) {
1371 			cmn = cm->m_next;
1372 			cm->m_next = NULL;
1373 			type = mtod(cm, struct cmsghdr *)->cmsg_type;
1374 			if (controlp != NULL) {
1375 				if (dom->dom_externalize != NULL &&
1376 				    type == SCM_RIGHTS) {
1377 					sounlock(so);
1378 					splx(s);
1379 					error = (*dom->dom_externalize)(cm, l);
1380 					s = splsoftnet();
1381 					solock(so);
1382 				}
1383 				*controlp = cm;
1384 				while (*controlp != NULL)
1385 					controlp = &(*controlp)->m_next;
1386 			} else {
1387 				/*
1388 				 * Dispose of any SCM_RIGHTS message that went
1389 				 * through the read path rather than recv.
1390 				 */
1391 				if (dom->dom_dispose != NULL &&
1392 				    type == SCM_RIGHTS) {
1393 				    	sounlock(so);
1394 					(*dom->dom_dispose)(cm);
1395 					solock(so);
1396 				}
1397 				m_freem(cm);
1398 			}
1399 		}
1400 		if (m != NULL)
1401 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1402 		else
1403 			nextrecord = so->so_rcv.sb_mb;
1404 		orig_resid = 0;
1405 	}
1406 
1407 	/* If m is non-NULL, we have some data to read. */
1408 	if (__predict_true(m != NULL)) {
1409 		type = m->m_type;
1410 		if (type == MT_OOBDATA)
1411 			flags |= MSG_OOB;
1412 	}
1413 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1414 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1415 
1416 	moff = 0;
1417 	offset = 0;
1418 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1419 		if (m->m_type == MT_OOBDATA) {
1420 			if (type != MT_OOBDATA)
1421 				break;
1422 		} else if (type == MT_OOBDATA)
1423 			break;
1424 #ifdef DIAGNOSTIC
1425 		else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
1426 			panic("receive 3");
1427 #endif
1428 		so->so_state &= ~SS_RCVATMARK;
1429 		wakeup_state = 0;
1430 		len = uio->uio_resid;
1431 		if (so->so_oobmark && len > so->so_oobmark - offset)
1432 			len = so->so_oobmark - offset;
1433 		if (len > m->m_len - moff)
1434 			len = m->m_len - moff;
1435 		/*
1436 		 * If mp is set, just pass back the mbufs.
1437 		 * Otherwise copy them out via the uio, then free.
1438 		 * Sockbuf must be consistent here (points to current mbuf,
1439 		 * it points to next record) when we drop priority;
1440 		 * we must note any additions to the sockbuf when we
1441 		 * block interrupts again.
1442 		 */
1443 		if (mp == NULL) {
1444 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1445 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1446 			sounlock(so);
1447 			splx(s);
1448 			error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1449 			s = splsoftnet();
1450 			solock(so);
1451 			if (error != 0) {
1452 				/*
1453 				 * If any part of the record has been removed
1454 				 * (such as the MT_SONAME mbuf, which will
1455 				 * happen when PR_ADDR, and thus also
1456 				 * PR_ATOMIC, is set), then drop the entire
1457 				 * record to maintain the atomicity of the
1458 				 * receive operation.
1459 				 *
1460 				 * This avoids a later panic("receive 1a")
1461 				 * when compiled with DIAGNOSTIC.
1462 				 */
1463 				if (m && mbuf_removed && atomic)
1464 					(void) sbdroprecord(&so->so_rcv);
1465 
1466 				goto release;
1467 			}
1468 		} else
1469 			uio->uio_resid -= len;
1470 		if (len == m->m_len - moff) {
1471 			if (m->m_flags & M_EOR)
1472 				flags |= MSG_EOR;
1473 			if (flags & MSG_PEEK) {
1474 				m = m->m_next;
1475 				moff = 0;
1476 			} else {
1477 				nextrecord = m->m_nextpkt;
1478 				sbfree(&so->so_rcv, m);
1479 				if (mp) {
1480 					*mp = m;
1481 					mp = &m->m_next;
1482 					so->so_rcv.sb_mb = m = m->m_next;
1483 					*mp = NULL;
1484 				} else {
1485 					MFREE(m, so->so_rcv.sb_mb);
1486 					m = so->so_rcv.sb_mb;
1487 				}
1488 				/*
1489 				 * If m != NULL, we also know that
1490 				 * so->so_rcv.sb_mb != NULL.
1491 				 */
1492 				KASSERT(so->so_rcv.sb_mb == m);
1493 				if (m) {
1494 					m->m_nextpkt = nextrecord;
1495 					if (nextrecord == NULL)
1496 						so->so_rcv.sb_lastrecord = m;
1497 				} else {
1498 					so->so_rcv.sb_mb = nextrecord;
1499 					SB_EMPTY_FIXUP(&so->so_rcv);
1500 				}
1501 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1502 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1503 			}
1504 		} else if (flags & MSG_PEEK)
1505 			moff += len;
1506 		else {
1507 			if (mp != NULL) {
1508 				mt = m_copym(m, 0, len, M_NOWAIT);
1509 				if (__predict_false(mt == NULL)) {
1510 					sounlock(so);
1511 					mt = m_copym(m, 0, len, M_WAIT);
1512 					solock(so);
1513 				}
1514 				*mp = mt;
1515 			}
1516 			m->m_data += len;
1517 			m->m_len -= len;
1518 			so->so_rcv.sb_cc -= len;
1519 		}
1520 		if (so->so_oobmark) {
1521 			if ((flags & MSG_PEEK) == 0) {
1522 				so->so_oobmark -= len;
1523 				if (so->so_oobmark == 0) {
1524 					so->so_state |= SS_RCVATMARK;
1525 					break;
1526 				}
1527 			} else {
1528 				offset += len;
1529 				if (offset == so->so_oobmark)
1530 					break;
1531 			}
1532 		}
1533 		if (flags & MSG_EOR)
1534 			break;
1535 		/*
1536 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1537 		 * we must not quit until "uio->uio_resid == 0" or an error
1538 		 * termination.  If a signal/timeout occurs, return
1539 		 * with a short count but without error.
1540 		 * Keep sockbuf locked against other readers.
1541 		 */
1542 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1543 		    !sosendallatonce(so) && !nextrecord) {
1544 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1545 				break;
1546 			/*
1547 			 * If we are peeking and the socket receive buffer is
1548 			 * full, stop since we can't get more data to peek at.
1549 			 */
1550 			if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
1551 				break;
1552 			/*
1553 			 * If we've drained the socket buffer, tell the
1554 			 * protocol in case it needs to do something to
1555 			 * get it filled again.
1556 			 */
1557 			if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1558 				(*pr->pr_usrreq)(so, PRU_RCVD,
1559 				    NULL, (struct mbuf *)(long)flags, NULL, l);
1560 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1561 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1562 			if (wakeup_state & SS_RESTARTSYS)
1563 				error = ERESTART;
1564 			else
1565 				error = sbwait(&so->so_rcv);
1566 			if (error != 0) {
1567 				sbunlock(&so->so_rcv);
1568 				sounlock(so);
1569 				splx(s);
1570 				return 0;
1571 			}
1572 			if ((m = so->so_rcv.sb_mb) != NULL)
1573 				nextrecord = m->m_nextpkt;
1574 			wakeup_state = so->so_state;
1575 		}
1576 	}
1577 
1578 	if (m && atomic) {
1579 		flags |= MSG_TRUNC;
1580 		if ((flags & MSG_PEEK) == 0)
1581 			(void) sbdroprecord(&so->so_rcv);
1582 	}
1583 	if ((flags & MSG_PEEK) == 0) {
1584 		if (m == NULL) {
1585 			/*
1586 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1587 			 * part makes sure sb_lastrecord is up-to-date if
1588 			 * there is still data in the socket buffer.
1589 			 */
1590 			so->so_rcv.sb_mb = nextrecord;
1591 			if (so->so_rcv.sb_mb == NULL) {
1592 				so->so_rcv.sb_mbtail = NULL;
1593 				so->so_rcv.sb_lastrecord = NULL;
1594 			} else if (nextrecord->m_nextpkt == NULL)
1595 				so->so_rcv.sb_lastrecord = nextrecord;
1596 		}
1597 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1598 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1599 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1600 			(*pr->pr_usrreq)(so, PRU_RCVD, NULL,
1601 			    (struct mbuf *)(long)flags, NULL, l);
1602 	}
1603 	if (orig_resid == uio->uio_resid && orig_resid &&
1604 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1605 		sbunlock(&so->so_rcv);
1606 		goto restart;
1607 	}
1608 
1609 	if (flagsp != NULL)
1610 		*flagsp |= flags;
1611  release:
1612 	sbunlock(&so->so_rcv);
1613 	sounlock(so);
1614 	splx(s);
1615 	return error;
1616 }
1617 
1618 int
1619 soshutdown(struct socket *so, int how)
1620 {
1621 	const struct protosw	*pr;
1622 	int	error;
1623 
1624 	KASSERT(solocked(so));
1625 
1626 	pr = so->so_proto;
1627 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1628 		return (EINVAL);
1629 
1630 	if (how == SHUT_RD || how == SHUT_RDWR) {
1631 		sorflush(so);
1632 		error = 0;
1633 	}
1634 	if (how == SHUT_WR || how == SHUT_RDWR)
1635 		error = (*pr->pr_usrreq)(so, PRU_SHUTDOWN, NULL,
1636 		    NULL, NULL, NULL);
1637 
1638 	return error;
1639 }
1640 
1641 void
1642 sorestart(struct socket *so)
1643 {
1644 	/*
1645 	 * An application has called close() on an fd on which another
1646 	 * of its threads has called a socket system call.
1647 	 * Mark this and wake everyone up, and code that would block again
1648 	 * instead returns ERESTART.
1649 	 * On system call re-entry the fd is validated and EBADF returned.
1650 	 * Any other fd will block again on the 2nd syscall.
1651 	 */
1652 	solock(so);
1653 	so->so_state |= SS_RESTARTSYS;
1654 	cv_broadcast(&so->so_cv);
1655 	cv_broadcast(&so->so_snd.sb_cv);
1656 	cv_broadcast(&so->so_rcv.sb_cv);
1657 	sounlock(so);
1658 }
1659 
1660 void
1661 sorflush(struct socket *so)
1662 {
1663 	struct sockbuf	*sb, asb;
1664 	const struct protosw	*pr;
1665 
1666 	KASSERT(solocked(so));
1667 
1668 	sb = &so->so_rcv;
1669 	pr = so->so_proto;
1670 	socantrcvmore(so);
1671 	sb->sb_flags |= SB_NOINTR;
1672 	(void )sblock(sb, M_WAITOK);
1673 	sbunlock(sb);
1674 	asb = *sb;
1675 	/*
1676 	 * Clear most of the sockbuf structure, but leave some of the
1677 	 * fields valid.
1678 	 */
1679 	memset(&sb->sb_startzero, 0,
1680 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1681 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) {
1682 		sounlock(so);
1683 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1684 		solock(so);
1685 	}
1686 	sbrelease(&asb, so);
1687 }
1688 
1689 /*
1690  * internal set SOL_SOCKET options
1691  */
1692 static int
1693 sosetopt1(struct socket *so, const struct sockopt *sopt)
1694 {
1695 	int error = EINVAL, optval, opt;
1696 	struct linger l;
1697 	struct timeval tv;
1698 
1699 	switch ((opt = sopt->sopt_name)) {
1700 
1701 	case SO_ACCEPTFILTER:
1702 		error = accept_filt_setopt(so, sopt);
1703 		KASSERT(solocked(so));
1704 		break;
1705 
1706   	case SO_LINGER:
1707  		error = sockopt_get(sopt, &l, sizeof(l));
1708 		solock(so);
1709  		if (error)
1710  			break;
1711  		if (l.l_linger < 0 || l.l_linger > USHRT_MAX ||
1712  		    l.l_linger > (INT_MAX / hz)) {
1713 			error = EDOM;
1714 			break;
1715 		}
1716  		so->so_linger = l.l_linger;
1717  		if (l.l_onoff)
1718  			so->so_options |= SO_LINGER;
1719  		else
1720  			so->so_options &= ~SO_LINGER;
1721    		break;
1722 
1723 	case SO_DEBUG:
1724 	case SO_KEEPALIVE:
1725 	case SO_DONTROUTE:
1726 	case SO_USELOOPBACK:
1727 	case SO_BROADCAST:
1728 	case SO_REUSEADDR:
1729 	case SO_REUSEPORT:
1730 	case SO_OOBINLINE:
1731 	case SO_TIMESTAMP:
1732 #ifdef SO_OTIMESTAMP
1733 	case SO_OTIMESTAMP:
1734 #endif
1735 		error = sockopt_getint(sopt, &optval);
1736 		solock(so);
1737 		if (error)
1738 			break;
1739 		if (optval)
1740 			so->so_options |= opt;
1741 		else
1742 			so->so_options &= ~opt;
1743 		break;
1744 
1745 	case SO_SNDBUF:
1746 	case SO_RCVBUF:
1747 	case SO_SNDLOWAT:
1748 	case SO_RCVLOWAT:
1749 		error = sockopt_getint(sopt, &optval);
1750 		solock(so);
1751 		if (error)
1752 			break;
1753 
1754 		/*
1755 		 * Values < 1 make no sense for any of these
1756 		 * options, so disallow them.
1757 		 */
1758 		if (optval < 1) {
1759 			error = EINVAL;
1760 			break;
1761 		}
1762 
1763 		switch (opt) {
1764 		case SO_SNDBUF:
1765 			if (sbreserve(&so->so_snd, (u_long)optval, so) == 0) {
1766 				error = ENOBUFS;
1767 				break;
1768 			}
1769 			so->so_snd.sb_flags &= ~SB_AUTOSIZE;
1770 			break;
1771 
1772 		case SO_RCVBUF:
1773 			if (sbreserve(&so->so_rcv, (u_long)optval, so) == 0) {
1774 				error = ENOBUFS;
1775 				break;
1776 			}
1777 			so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1778 			break;
1779 
1780 		/*
1781 		 * Make sure the low-water is never greater than
1782 		 * the high-water.
1783 		 */
1784 		case SO_SNDLOWAT:
1785 			if (optval > so->so_snd.sb_hiwat)
1786 				optval = so->so_snd.sb_hiwat;
1787 
1788 			so->so_snd.sb_lowat = optval;
1789 			break;
1790 
1791 		case SO_RCVLOWAT:
1792 			if (optval > so->so_rcv.sb_hiwat)
1793 				optval = so->so_rcv.sb_hiwat;
1794 
1795 			so->so_rcv.sb_lowat = optval;
1796 			break;
1797 		}
1798 		break;
1799 
1800 #ifdef COMPAT_50
1801 	case SO_OSNDTIMEO:
1802 	case SO_ORCVTIMEO: {
1803 		struct timeval50 otv;
1804 		error = sockopt_get(sopt, &otv, sizeof(otv));
1805 		if (error) {
1806 			solock(so);
1807 			break;
1808 		}
1809 		timeval50_to_timeval(&otv, &tv);
1810 		opt = opt == SO_OSNDTIMEO ? SO_SNDTIMEO : SO_RCVTIMEO;
1811 		error = 0;
1812 		/*FALLTHROUGH*/
1813 	}
1814 #endif /* COMPAT_50 */
1815 
1816 	case SO_SNDTIMEO:
1817 	case SO_RCVTIMEO:
1818 		if (error)
1819 			error = sockopt_get(sopt, &tv, sizeof(tv));
1820 		solock(so);
1821 		if (error)
1822 			break;
1823 
1824 		if (tv.tv_sec > (INT_MAX - tv.tv_usec / tick) / hz) {
1825 			error = EDOM;
1826 			break;
1827 		}
1828 
1829 		optval = tv.tv_sec * hz + tv.tv_usec / tick;
1830 		if (optval == 0 && tv.tv_usec != 0)
1831 			optval = 1;
1832 
1833 		switch (opt) {
1834 		case SO_SNDTIMEO:
1835 			so->so_snd.sb_timeo = optval;
1836 			break;
1837 		case SO_RCVTIMEO:
1838 			so->so_rcv.sb_timeo = optval;
1839 			break;
1840 		}
1841 		break;
1842 
1843 	default:
1844 		solock(so);
1845 		error = ENOPROTOOPT;
1846 		break;
1847 	}
1848 	KASSERT(solocked(so));
1849 	return error;
1850 }
1851 
1852 int
1853 sosetopt(struct socket *so, struct sockopt *sopt)
1854 {
1855 	int error, prerr;
1856 
1857 	if (sopt->sopt_level == SOL_SOCKET) {
1858 		error = sosetopt1(so, sopt);
1859 		KASSERT(solocked(so));
1860 	} else {
1861 		error = ENOPROTOOPT;
1862 		solock(so);
1863 	}
1864 
1865 	if ((error == 0 || error == ENOPROTOOPT) &&
1866 	    so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) {
1867 		/* give the protocol stack a shot */
1868 		prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, sopt);
1869 		if (prerr == 0)
1870 			error = 0;
1871 		else if (prerr != ENOPROTOOPT)
1872 			error = prerr;
1873 	}
1874 	sounlock(so);
1875 	return error;
1876 }
1877 
1878 /*
1879  * so_setsockopt() is a wrapper providing a sockopt structure for sosetopt()
1880  */
1881 int
1882 so_setsockopt(struct lwp *l, struct socket *so, int level, int name,
1883     const void *val, size_t valsize)
1884 {
1885 	struct sockopt sopt;
1886 	int error;
1887 
1888 	KASSERT(valsize == 0 || val != NULL);
1889 
1890 	sockopt_init(&sopt, level, name, valsize);
1891 	sockopt_set(&sopt, val, valsize);
1892 
1893 	error = sosetopt(so, &sopt);
1894 
1895 	sockopt_destroy(&sopt);
1896 
1897 	return error;
1898 }
1899 
1900 /*
1901  * internal get SOL_SOCKET options
1902  */
1903 static int
1904 sogetopt1(struct socket *so, struct sockopt *sopt)
1905 {
1906 	int error, optval, opt;
1907 	struct linger l;
1908 	struct timeval tv;
1909 
1910 	switch ((opt = sopt->sopt_name)) {
1911 
1912 	case SO_ACCEPTFILTER:
1913 		error = accept_filt_getopt(so, sopt);
1914 		break;
1915 
1916 	case SO_LINGER:
1917 		l.l_onoff = (so->so_options & SO_LINGER) ? 1 : 0;
1918 		l.l_linger = so->so_linger;
1919 
1920 		error = sockopt_set(sopt, &l, sizeof(l));
1921 		break;
1922 
1923 	case SO_USELOOPBACK:
1924 	case SO_DONTROUTE:
1925 	case SO_DEBUG:
1926 	case SO_KEEPALIVE:
1927 	case SO_REUSEADDR:
1928 	case SO_REUSEPORT:
1929 	case SO_BROADCAST:
1930 	case SO_OOBINLINE:
1931 	case SO_TIMESTAMP:
1932 #ifdef SO_OTIMESTAMP
1933 	case SO_OTIMESTAMP:
1934 #endif
1935 		error = sockopt_setint(sopt, (so->so_options & opt) ? 1 : 0);
1936 		break;
1937 
1938 	case SO_TYPE:
1939 		error = sockopt_setint(sopt, so->so_type);
1940 		break;
1941 
1942 	case SO_ERROR:
1943 		error = sockopt_setint(sopt, so->so_error);
1944 		so->so_error = 0;
1945 		break;
1946 
1947 	case SO_SNDBUF:
1948 		error = sockopt_setint(sopt, so->so_snd.sb_hiwat);
1949 		break;
1950 
1951 	case SO_RCVBUF:
1952 		error = sockopt_setint(sopt, so->so_rcv.sb_hiwat);
1953 		break;
1954 
1955 	case SO_SNDLOWAT:
1956 		error = sockopt_setint(sopt, so->so_snd.sb_lowat);
1957 		break;
1958 
1959 	case SO_RCVLOWAT:
1960 		error = sockopt_setint(sopt, so->so_rcv.sb_lowat);
1961 		break;
1962 
1963 #ifdef COMPAT_50
1964 	case SO_OSNDTIMEO:
1965 	case SO_ORCVTIMEO: {
1966 		struct timeval50 otv;
1967 
1968 		optval = (opt == SO_OSNDTIMEO ?
1969 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1970 
1971 		otv.tv_sec = optval / hz;
1972 		otv.tv_usec = (optval % hz) * tick;
1973 
1974 		error = sockopt_set(sopt, &otv, sizeof(otv));
1975 		break;
1976 	}
1977 #endif /* COMPAT_50 */
1978 
1979 	case SO_SNDTIMEO:
1980 	case SO_RCVTIMEO:
1981 		optval = (opt == SO_SNDTIMEO ?
1982 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1983 
1984 		tv.tv_sec = optval / hz;
1985 		tv.tv_usec = (optval % hz) * tick;
1986 
1987 		error = sockopt_set(sopt, &tv, sizeof(tv));
1988 		break;
1989 
1990 	case SO_OVERFLOWED:
1991 		error = sockopt_setint(sopt, so->so_rcv.sb_overflowed);
1992 		break;
1993 
1994 	default:
1995 		error = ENOPROTOOPT;
1996 		break;
1997 	}
1998 
1999 	return (error);
2000 }
2001 
2002 int
2003 sogetopt(struct socket *so, struct sockopt *sopt)
2004 {
2005 	int		error;
2006 
2007 	solock(so);
2008 	if (sopt->sopt_level != SOL_SOCKET) {
2009 		if (so->so_proto && so->so_proto->pr_ctloutput) {
2010 			error = ((*so->so_proto->pr_ctloutput)
2011 			    (PRCO_GETOPT, so, sopt));
2012 		} else
2013 			error = (ENOPROTOOPT);
2014 	} else {
2015 		error = sogetopt1(so, sopt);
2016 	}
2017 	sounlock(so);
2018 	return (error);
2019 }
2020 
2021 /*
2022  * alloc sockopt data buffer buffer
2023  *	- will be released at destroy
2024  */
2025 static int
2026 sockopt_alloc(struct sockopt *sopt, size_t len, km_flag_t kmflag)
2027 {
2028 
2029 	KASSERT(sopt->sopt_size == 0);
2030 
2031 	if (len > sizeof(sopt->sopt_buf)) {
2032 		sopt->sopt_data = kmem_zalloc(len, kmflag);
2033 		if (sopt->sopt_data == NULL)
2034 			return ENOMEM;
2035 	} else
2036 		sopt->sopt_data = sopt->sopt_buf;
2037 
2038 	sopt->sopt_size = len;
2039 	return 0;
2040 }
2041 
2042 /*
2043  * initialise sockopt storage
2044  *	- MAY sleep during allocation
2045  */
2046 void
2047 sockopt_init(struct sockopt *sopt, int level, int name, size_t size)
2048 {
2049 
2050 	memset(sopt, 0, sizeof(*sopt));
2051 
2052 	sopt->sopt_level = level;
2053 	sopt->sopt_name = name;
2054 	(void)sockopt_alloc(sopt, size, KM_SLEEP);
2055 }
2056 
2057 /*
2058  * destroy sockopt storage
2059  *	- will release any held memory references
2060  */
2061 void
2062 sockopt_destroy(struct sockopt *sopt)
2063 {
2064 
2065 	if (sopt->sopt_data != sopt->sopt_buf)
2066 		kmem_free(sopt->sopt_data, sopt->sopt_size);
2067 
2068 	memset(sopt, 0, sizeof(*sopt));
2069 }
2070 
2071 /*
2072  * set sockopt value
2073  *	- value is copied into sockopt
2074  * 	- memory is allocated when necessary, will not sleep
2075  */
2076 int
2077 sockopt_set(struct sockopt *sopt, const void *buf, size_t len)
2078 {
2079 	int error;
2080 
2081 	if (sopt->sopt_size == 0) {
2082 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2083 		if (error)
2084 			return error;
2085 	}
2086 
2087 	KASSERT(sopt->sopt_size == len);
2088 	memcpy(sopt->sopt_data, buf, len);
2089 	return 0;
2090 }
2091 
2092 /*
2093  * common case of set sockopt integer value
2094  */
2095 int
2096 sockopt_setint(struct sockopt *sopt, int val)
2097 {
2098 
2099 	return sockopt_set(sopt, &val, sizeof(int));
2100 }
2101 
2102 /*
2103  * get sockopt value
2104  *	- correct size must be given
2105  */
2106 int
2107 sockopt_get(const struct sockopt *sopt, void *buf, size_t len)
2108 {
2109 
2110 	if (sopt->sopt_size != len)
2111 		return EINVAL;
2112 
2113 	memcpy(buf, sopt->sopt_data, len);
2114 	return 0;
2115 }
2116 
2117 /*
2118  * common case of get sockopt integer value
2119  */
2120 int
2121 sockopt_getint(const struct sockopt *sopt, int *valp)
2122 {
2123 
2124 	return sockopt_get(sopt, valp, sizeof(int));
2125 }
2126 
2127 /*
2128  * set sockopt value from mbuf
2129  *	- ONLY for legacy code
2130  *	- mbuf is released by sockopt
2131  *	- will not sleep
2132  */
2133 int
2134 sockopt_setmbuf(struct sockopt *sopt, struct mbuf *m)
2135 {
2136 	size_t len;
2137 	int error;
2138 
2139 	len = m_length(m);
2140 
2141 	if (sopt->sopt_size == 0) {
2142 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2143 		if (error)
2144 			return error;
2145 	}
2146 
2147 	KASSERT(sopt->sopt_size == len);
2148 	m_copydata(m, 0, len, sopt->sopt_data);
2149 	m_freem(m);
2150 
2151 	return 0;
2152 }
2153 
2154 /*
2155  * get sockopt value into mbuf
2156  *	- ONLY for legacy code
2157  *	- mbuf to be released by the caller
2158  *	- will not sleep
2159  */
2160 struct mbuf *
2161 sockopt_getmbuf(const struct sockopt *sopt)
2162 {
2163 	struct mbuf *m;
2164 
2165 	if (sopt->sopt_size > MCLBYTES)
2166 		return NULL;
2167 
2168 	m = m_get(M_DONTWAIT, MT_SOOPTS);
2169 	if (m == NULL)
2170 		return NULL;
2171 
2172 	if (sopt->sopt_size > MLEN) {
2173 		MCLGET(m, M_DONTWAIT);
2174 		if ((m->m_flags & M_EXT) == 0) {
2175 			m_free(m);
2176 			return NULL;
2177 		}
2178 	}
2179 
2180 	memcpy(mtod(m, void *), sopt->sopt_data, sopt->sopt_size);
2181 	m->m_len = sopt->sopt_size;
2182 
2183 	return m;
2184 }
2185 
2186 void
2187 sohasoutofband(struct socket *so)
2188 {
2189 
2190 	fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
2191 	selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, NOTE_SUBMIT);
2192 }
2193 
2194 static void
2195 filt_sordetach(struct knote *kn)
2196 {
2197 	struct socket	*so;
2198 
2199 	so = ((file_t *)kn->kn_obj)->f_data;
2200 	solock(so);
2201 	SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
2202 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
2203 		so->so_rcv.sb_flags &= ~SB_KNOTE;
2204 	sounlock(so);
2205 }
2206 
2207 /*ARGSUSED*/
2208 static int
2209 filt_soread(struct knote *kn, long hint)
2210 {
2211 	struct socket	*so;
2212 	int rv;
2213 
2214 	so = ((file_t *)kn->kn_obj)->f_data;
2215 	if (hint != NOTE_SUBMIT)
2216 		solock(so);
2217 	kn->kn_data = so->so_rcv.sb_cc;
2218 	if (so->so_state & SS_CANTRCVMORE) {
2219 		kn->kn_flags |= EV_EOF;
2220 		kn->kn_fflags = so->so_error;
2221 		rv = 1;
2222 	} else if (so->so_error)	/* temporary udp error */
2223 		rv = 1;
2224 	else if (kn->kn_sfflags & NOTE_LOWAT)
2225 		rv = (kn->kn_data >= kn->kn_sdata);
2226 	else
2227 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
2228 	if (hint != NOTE_SUBMIT)
2229 		sounlock(so);
2230 	return rv;
2231 }
2232 
2233 static void
2234 filt_sowdetach(struct knote *kn)
2235 {
2236 	struct socket	*so;
2237 
2238 	so = ((file_t *)kn->kn_obj)->f_data;
2239 	solock(so);
2240 	SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
2241 	if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
2242 		so->so_snd.sb_flags &= ~SB_KNOTE;
2243 	sounlock(so);
2244 }
2245 
2246 /*ARGSUSED*/
2247 static int
2248 filt_sowrite(struct knote *kn, long hint)
2249 {
2250 	struct socket	*so;
2251 	int rv;
2252 
2253 	so = ((file_t *)kn->kn_obj)->f_data;
2254 	if (hint != NOTE_SUBMIT)
2255 		solock(so);
2256 	kn->kn_data = sbspace(&so->so_snd);
2257 	if (so->so_state & SS_CANTSENDMORE) {
2258 		kn->kn_flags |= EV_EOF;
2259 		kn->kn_fflags = so->so_error;
2260 		rv = 1;
2261 	} else if (so->so_error)	/* temporary udp error */
2262 		rv = 1;
2263 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2264 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
2265 		rv = 0;
2266 	else if (kn->kn_sfflags & NOTE_LOWAT)
2267 		rv = (kn->kn_data >= kn->kn_sdata);
2268 	else
2269 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
2270 	if (hint != NOTE_SUBMIT)
2271 		sounlock(so);
2272 	return rv;
2273 }
2274 
2275 /*ARGSUSED*/
2276 static int
2277 filt_solisten(struct knote *kn, long hint)
2278 {
2279 	struct socket	*so;
2280 	int rv;
2281 
2282 	so = ((file_t *)kn->kn_obj)->f_data;
2283 
2284 	/*
2285 	 * Set kn_data to number of incoming connections, not
2286 	 * counting partial (incomplete) connections.
2287 	 */
2288 	if (hint != NOTE_SUBMIT)
2289 		solock(so);
2290 	kn->kn_data = so->so_qlen;
2291 	rv = (kn->kn_data > 0);
2292 	if (hint != NOTE_SUBMIT)
2293 		sounlock(so);
2294 	return rv;
2295 }
2296 
2297 static const struct filterops solisten_filtops =
2298 	{ 1, NULL, filt_sordetach, filt_solisten };
2299 static const struct filterops soread_filtops =
2300 	{ 1, NULL, filt_sordetach, filt_soread };
2301 static const struct filterops sowrite_filtops =
2302 	{ 1, NULL, filt_sowdetach, filt_sowrite };
2303 
2304 int
2305 soo_kqfilter(struct file *fp, struct knote *kn)
2306 {
2307 	struct socket	*so;
2308 	struct sockbuf	*sb;
2309 
2310 	so = ((file_t *)kn->kn_obj)->f_data;
2311 	solock(so);
2312 	switch (kn->kn_filter) {
2313 	case EVFILT_READ:
2314 		if (so->so_options & SO_ACCEPTCONN)
2315 			kn->kn_fop = &solisten_filtops;
2316 		else
2317 			kn->kn_fop = &soread_filtops;
2318 		sb = &so->so_rcv;
2319 		break;
2320 	case EVFILT_WRITE:
2321 		kn->kn_fop = &sowrite_filtops;
2322 		sb = &so->so_snd;
2323 		break;
2324 	default:
2325 		sounlock(so);
2326 		return (EINVAL);
2327 	}
2328 	SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
2329 	sb->sb_flags |= SB_KNOTE;
2330 	sounlock(so);
2331 	return (0);
2332 }
2333 
2334 static int
2335 sodopoll(struct socket *so, int events)
2336 {
2337 	int revents;
2338 
2339 	revents = 0;
2340 
2341 	if (events & (POLLIN | POLLRDNORM))
2342 		if (soreadable(so))
2343 			revents |= events & (POLLIN | POLLRDNORM);
2344 
2345 	if (events & (POLLOUT | POLLWRNORM))
2346 		if (sowritable(so))
2347 			revents |= events & (POLLOUT | POLLWRNORM);
2348 
2349 	if (events & (POLLPRI | POLLRDBAND))
2350 		if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
2351 			revents |= events & (POLLPRI | POLLRDBAND);
2352 
2353 	return revents;
2354 }
2355 
2356 int
2357 sopoll(struct socket *so, int events)
2358 {
2359 	int revents = 0;
2360 
2361 #ifndef DIAGNOSTIC
2362 	/*
2363 	 * Do a quick, unlocked check in expectation that the socket
2364 	 * will be ready for I/O.  Don't do this check if DIAGNOSTIC,
2365 	 * as the solocked() assertions will fail.
2366 	 */
2367 	if ((revents = sodopoll(so, events)) != 0)
2368 		return revents;
2369 #endif
2370 
2371 	solock(so);
2372 	if ((revents = sodopoll(so, events)) == 0) {
2373 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2374 			selrecord(curlwp, &so->so_rcv.sb_sel);
2375 			so->so_rcv.sb_flags |= SB_NOTIFY;
2376 		}
2377 
2378 		if (events & (POLLOUT | POLLWRNORM)) {
2379 			selrecord(curlwp, &so->so_snd.sb_sel);
2380 			so->so_snd.sb_flags |= SB_NOTIFY;
2381 		}
2382 	}
2383 	sounlock(so);
2384 
2385 	return revents;
2386 }
2387 
2388 
2389 #include <sys/sysctl.h>
2390 
2391 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO);
2392 
2393 /*
2394  * sysctl helper routine for kern.somaxkva.  ensures that the given
2395  * value is not too small.
2396  * (XXX should we maybe make sure it's not too large as well?)
2397  */
2398 static int
2399 sysctl_kern_somaxkva(SYSCTLFN_ARGS)
2400 {
2401 	int error, new_somaxkva;
2402 	struct sysctlnode node;
2403 
2404 	new_somaxkva = somaxkva;
2405 	node = *rnode;
2406 	node.sysctl_data = &new_somaxkva;
2407 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2408 	if (error || newp == NULL)
2409 		return (error);
2410 
2411 	if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */
2412 		return (EINVAL);
2413 
2414 	mutex_enter(&so_pendfree_lock);
2415 	somaxkva = new_somaxkva;
2416 	cv_broadcast(&socurkva_cv);
2417 	mutex_exit(&so_pendfree_lock);
2418 
2419 	return (error);
2420 }
2421 
2422 static void
2423 sysctl_kern_somaxkva_setup(void)
2424 {
2425 
2426 	KASSERT(socket_sysctllog == NULL);
2427 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2428 		       CTLFLAG_PERMANENT,
2429 		       CTLTYPE_NODE, "kern", NULL,
2430 		       NULL, 0, NULL, 0,
2431 		       CTL_KERN, CTL_EOL);
2432 
2433 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2434 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2435 		       CTLTYPE_INT, "somaxkva",
2436 		       SYSCTL_DESCR("Maximum amount of kernel memory to be "
2437 				    "used for socket buffers"),
2438 		       sysctl_kern_somaxkva, 0, NULL, 0,
2439 		       CTL_KERN, KERN_SOMAXKVA, CTL_EOL);
2440 }
2441