xref: /netbsd-src/sys/kern/uipc_socket.c (revision 6deb2c22d20de1d75d538e8a5c57b573926fd157)
1 /*	$NetBSD: uipc_socket.c,v 1.193 2009/10/03 03:59:39 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 1.193 2009/10/03 03:59:39 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);
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 		uid_t sockuid = so->so_uidinfo->ui_uid;
456 
457 		if (sockuid == kauth_cred_getuid(cred) ||
458 		    sockuid == kauth_cred_geteuid(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 	/* so->so_cred = kauth_cred_dup(l->l_cred); */
565 	uid = kauth_cred_geteuid(l->l_cred);
566 	so->so_uidinfo = uid_find(uid);
567 	so->so_egid = kauth_cred_getegid(l->l_cred);
568 	so->so_cpid = l->l_proc->p_pid;
569 	if (lockso != NULL) {
570 		/* Caller wants us to share a lock. */
571 		lock = lockso->so_lock;
572 		so->so_lock = lock;
573 		mutex_obj_hold(lock);
574 		mutex_enter(lock);
575 	} else {
576 		/* Lock assigned and taken during PRU_ATTACH. */
577 	}
578 	error = (*prp->pr_usrreq)(so, PRU_ATTACH, NULL,
579 	    (struct mbuf *)(long)proto, NULL, l);
580 	KASSERT(solocked(so));
581 	if (error != 0) {
582 		so->so_state |= SS_NOFDREF;
583 		sofree(so);
584 		return error;
585 	}
586 	sounlock(so);
587 	*aso = so;
588 	return 0;
589 }
590 
591 /* On success, write file descriptor to fdout and return zero.  On
592  * failure, return non-zero; *fdout will be undefined.
593  */
594 int
595 fsocreate(int domain, struct socket **sop, int type, int protocol,
596     struct lwp *l, int *fdout)
597 {
598 	struct socket	*so;
599 	struct file	*fp;
600 	int		fd, error;
601 
602 	if ((error = fd_allocfile(&fp, &fd)) != 0)
603 		return (error);
604 	fp->f_flag = FREAD|FWRITE;
605 	fp->f_type = DTYPE_SOCKET;
606 	fp->f_ops = &socketops;
607 	error = socreate(domain, &so, type, protocol, l, NULL);
608 	if (error != 0) {
609 		fd_abort(curproc, fp, fd);
610 	} else {
611 		if (sop != NULL)
612 			*sop = so;
613 		fp->f_data = so;
614 		fd_affix(curproc, fp, fd);
615 		*fdout = fd;
616 	}
617 	return error;
618 }
619 
620 int
621 sofamily(const struct socket *so)
622 {
623 	const struct protosw *pr;
624 	const struct domain *dom;
625 
626 	if ((pr = so->so_proto) == NULL)
627 		return AF_UNSPEC;
628 	if ((dom = pr->pr_domain) == NULL)
629 		return AF_UNSPEC;
630 	return dom->dom_family;
631 }
632 
633 int
634 sobind(struct socket *so, struct mbuf *nam, struct lwp *l)
635 {
636 	int	error;
637 
638 	solock(so);
639 	error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, NULL, nam, NULL, l);
640 	sounlock(so);
641 	return error;
642 }
643 
644 int
645 solisten(struct socket *so, int backlog, struct lwp *l)
646 {
647 	int	error;
648 
649 	solock(so);
650 	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
651 	    SS_ISDISCONNECTING)) != 0) {
652 	    	sounlock(so);
653 		return (EOPNOTSUPP);
654 	}
655 	error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, NULL,
656 	    NULL, NULL, l);
657 	if (error != 0) {
658 		sounlock(so);
659 		return error;
660 	}
661 	if (TAILQ_EMPTY(&so->so_q))
662 		so->so_options |= SO_ACCEPTCONN;
663 	if (backlog < 0)
664 		backlog = 0;
665 	so->so_qlimit = min(backlog, somaxconn);
666 	sounlock(so);
667 	return 0;
668 }
669 
670 void
671 sofree(struct socket *so)
672 {
673 	u_int refs;
674 
675 	KASSERT(solocked(so));
676 
677 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
678 		sounlock(so);
679 		return;
680 	}
681 	if (so->so_head) {
682 		/*
683 		 * We must not decommission a socket that's on the accept(2)
684 		 * queue.  If we do, then accept(2) may hang after select(2)
685 		 * indicated that the listening socket was ready.
686 		 */
687 		if (!soqremque(so, 0)) {
688 			sounlock(so);
689 			return;
690 		}
691 	}
692 	if (so->so_rcv.sb_hiwat)
693 		(void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0,
694 		    RLIM_INFINITY);
695 	if (so->so_snd.sb_hiwat)
696 		(void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0,
697 		    RLIM_INFINITY);
698 	sbrelease(&so->so_snd, so);
699 	KASSERT(!cv_has_waiters(&so->so_cv));
700 	KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv));
701 	KASSERT(!cv_has_waiters(&so->so_snd.sb_cv));
702 	sorflush(so);
703 	refs = so->so_aborting;	/* XXX */
704 	/* Remove acccept filter if one is present. */
705 	if (so->so_accf != NULL)
706 		(void)accept_filt_clear(so);
707 	/* kauth_cred_free(so->so_cred); */
708 	sounlock(so);
709 	if (refs == 0)		/* XXX */
710 		soput(so);
711 }
712 
713 /*
714  * Close a socket on last file table reference removal.
715  * Initiate disconnect if connected.
716  * Free socket when disconnect complete.
717  */
718 int
719 soclose(struct socket *so)
720 {
721 	struct socket	*so2;
722 	int		error;
723 	int		error2;
724 
725 	error = 0;
726 	solock(so);
727 	if (so->so_options & SO_ACCEPTCONN) {
728 		for (;;) {
729 			if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) {
730 				KASSERT(solocked2(so, so2));
731 				(void) soqremque(so2, 0);
732 				/* soabort drops the lock. */
733 				(void) soabort(so2);
734 				solock(so);
735 				continue;
736 			}
737 			if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) {
738 				KASSERT(solocked2(so, so2));
739 				(void) soqremque(so2, 1);
740 				/* soabort drops the lock. */
741 				(void) soabort(so2);
742 				solock(so);
743 				continue;
744 			}
745 			break;
746 		}
747 	}
748 	if (so->so_pcb == 0)
749 		goto discard;
750 	if (so->so_state & SS_ISCONNECTED) {
751 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
752 			error = sodisconnect(so);
753 			if (error)
754 				goto drop;
755 		}
756 		if (so->so_options & SO_LINGER) {
757 			if ((so->so_state & SS_ISDISCONNECTING) && so->so_nbio)
758 				goto drop;
759 			while (so->so_state & SS_ISCONNECTED) {
760 				error = sowait(so, true, so->so_linger * hz);
761 				if (error)
762 					break;
763 			}
764 		}
765 	}
766  drop:
767 	if (so->so_pcb) {
768 		error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH,
769 		    NULL, NULL, NULL, NULL);
770 		if (error == 0)
771 			error = error2;
772 	}
773  discard:
774 	if (so->so_state & SS_NOFDREF)
775 		panic("soclose: NOFDREF");
776 	so->so_state |= SS_NOFDREF;
777 	sofree(so);
778 	return (error);
779 }
780 
781 /*
782  * Must be called with the socket locked..  Will return with it unlocked.
783  */
784 int
785 soabort(struct socket *so)
786 {
787 	u_int refs;
788 	int error;
789 
790 	KASSERT(solocked(so));
791 	KASSERT(so->so_head == NULL);
792 
793 	so->so_aborting++;		/* XXX */
794 	error = (*so->so_proto->pr_usrreq)(so, PRU_ABORT, NULL,
795 	    NULL, NULL, NULL);
796 	refs = --so->so_aborting;	/* XXX */
797 	if (error || (refs == 0)) {
798 		sofree(so);
799 	} else {
800 		sounlock(so);
801 	}
802 	return error;
803 }
804 
805 int
806 soaccept(struct socket *so, struct mbuf *nam)
807 {
808 	int	error;
809 
810 	KASSERT(solocked(so));
811 
812 	error = 0;
813 	if ((so->so_state & SS_NOFDREF) == 0)
814 		panic("soaccept: !NOFDREF");
815 	so->so_state &= ~SS_NOFDREF;
816 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
817 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
818 		error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT,
819 		    NULL, nam, NULL, NULL);
820 	else
821 		error = ECONNABORTED;
822 
823 	return (error);
824 }
825 
826 int
827 soconnect(struct socket *so, struct mbuf *nam, struct lwp *l)
828 {
829 	int		error;
830 
831 	KASSERT(solocked(so));
832 
833 	if (so->so_options & SO_ACCEPTCONN)
834 		return (EOPNOTSUPP);
835 	/*
836 	 * If protocol is connection-based, can only connect once.
837 	 * Otherwise, if connected, try to disconnect first.
838 	 * This allows user to disconnect by connecting to, e.g.,
839 	 * a null address.
840 	 */
841 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
842 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
843 	    (error = sodisconnect(so))))
844 		error = EISCONN;
845 	else
846 		error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT,
847 		    NULL, nam, NULL, l);
848 	return (error);
849 }
850 
851 int
852 soconnect2(struct socket *so1, struct socket *so2)
853 {
854 	int	error;
855 
856 	KASSERT(solocked2(so1, so2));
857 
858 	error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2,
859 	    NULL, (struct mbuf *)so2, NULL, NULL);
860 	return (error);
861 }
862 
863 int
864 sodisconnect(struct socket *so)
865 {
866 	int	error;
867 
868 	KASSERT(solocked(so));
869 
870 	if ((so->so_state & SS_ISCONNECTED) == 0) {
871 		error = ENOTCONN;
872 	} else if (so->so_state & SS_ISDISCONNECTING) {
873 		error = EALREADY;
874 	} else {
875 		error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT,
876 		    NULL, NULL, NULL, NULL);
877 	}
878 	sodopendfree();
879 	return (error);
880 }
881 
882 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
883 /*
884  * Send on a socket.
885  * If send must go all at once and message is larger than
886  * send buffering, then hard error.
887  * Lock against other senders.
888  * If must go all at once and not enough room now, then
889  * inform user that this would block and do nothing.
890  * Otherwise, if nonblocking, send as much as possible.
891  * The data to be sent is described by "uio" if nonzero,
892  * otherwise by the mbuf chain "top" (which must be null
893  * if uio is not).  Data provided in mbuf chain must be small
894  * enough to send all at once.
895  *
896  * Returns nonzero on error, timeout or signal; callers
897  * must check for short counts if EINTR/ERESTART are returned.
898  * Data and control buffers are freed on return.
899  */
900 int
901 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
902 	struct mbuf *control, int flags, struct lwp *l)
903 {
904 	struct mbuf	**mp, *m;
905 	struct proc	*p;
906 	long		space, len, resid, clen, mlen;
907 	int		error, s, dontroute, atomic;
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 			error = sbwait(&so->so_snd);
983 			if (error)
984 				goto out;
985 			goto restart;
986 		}
987 		mp = &top;
988 		space -= clen;
989 		do {
990 			if (uio == NULL) {
991 				/*
992 				 * Data is prepackaged in "top".
993 				 */
994 				resid = 0;
995 				if (flags & MSG_EOR)
996 					top->m_flags |= M_EOR;
997 			} else do {
998 				sounlock(so);
999 				splx(s);
1000 				if (top == NULL) {
1001 					m = m_gethdr(M_WAIT, MT_DATA);
1002 					mlen = MHLEN;
1003 					m->m_pkthdr.len = 0;
1004 					m->m_pkthdr.rcvif = NULL;
1005 				} else {
1006 					m = m_get(M_WAIT, MT_DATA);
1007 					mlen = MLEN;
1008 				}
1009 				MCLAIM(m, so->so_snd.sb_mowner);
1010 				if (sock_loan_thresh >= 0 &&
1011 				    uio->uio_iov->iov_len >= sock_loan_thresh &&
1012 				    space >= sock_loan_thresh &&
1013 				    (len = sosend_loan(so, uio, m,
1014 						       space)) != 0) {
1015 					SOSEND_COUNTER_INCR(&sosend_loan_big);
1016 					space -= len;
1017 					goto have_data;
1018 				}
1019 				if (resid >= MINCLSIZE && space >= MCLBYTES) {
1020 					SOSEND_COUNTER_INCR(&sosend_copy_big);
1021 					m_clget(m, M_WAIT);
1022 					if ((m->m_flags & M_EXT) == 0)
1023 						goto nopages;
1024 					mlen = MCLBYTES;
1025 					if (atomic && top == 0) {
1026 						len = lmin(MCLBYTES - max_hdr,
1027 						    resid);
1028 						m->m_data += max_hdr;
1029 					} else
1030 						len = lmin(MCLBYTES, resid);
1031 					space -= len;
1032 				} else {
1033  nopages:
1034 					SOSEND_COUNTER_INCR(&sosend_copy_small);
1035 					len = lmin(lmin(mlen, resid), space);
1036 					space -= len;
1037 					/*
1038 					 * For datagram protocols, leave room
1039 					 * for protocol headers in first mbuf.
1040 					 */
1041 					if (atomic && top == 0 && len < mlen)
1042 						MH_ALIGN(m, len);
1043 				}
1044 				error = uiomove(mtod(m, void *), (int)len, uio);
1045  have_data:
1046 				resid = uio->uio_resid;
1047 				m->m_len = len;
1048 				*mp = m;
1049 				top->m_pkthdr.len += len;
1050 				s = splsoftnet();
1051 				solock(so);
1052 				if (error != 0)
1053 					goto release;
1054 				mp = &m->m_next;
1055 				if (resid <= 0) {
1056 					if (flags & MSG_EOR)
1057 						top->m_flags |= M_EOR;
1058 					break;
1059 				}
1060 			} while (space > 0 && atomic);
1061 
1062 			if (so->so_state & SS_CANTSENDMORE) {
1063 				error = EPIPE;
1064 				goto release;
1065 			}
1066 			if (dontroute)
1067 				so->so_options |= SO_DONTROUTE;
1068 			if (resid > 0)
1069 				so->so_state |= SS_MORETOCOME;
1070 			error = (*so->so_proto->pr_usrreq)(so,
1071 			    (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND,
1072 			    top, addr, control, curlwp);
1073 			if (dontroute)
1074 				so->so_options &= ~SO_DONTROUTE;
1075 			if (resid > 0)
1076 				so->so_state &= ~SS_MORETOCOME;
1077 			clen = 0;
1078 			control = NULL;
1079 			top = NULL;
1080 			mp = &top;
1081 			if (error != 0)
1082 				goto release;
1083 		} while (resid && space > 0);
1084 	} while (resid);
1085 
1086  release:
1087 	sbunlock(&so->so_snd);
1088  out:
1089 	sounlock(so);
1090 	splx(s);
1091 	if (top)
1092 		m_freem(top);
1093 	if (control)
1094 		m_freem(control);
1095 	return (error);
1096 }
1097 
1098 /*
1099  * Following replacement or removal of the first mbuf on the first
1100  * mbuf chain of a socket buffer, push necessary state changes back
1101  * into the socket buffer so that other consumers see the values
1102  * consistently.  'nextrecord' is the callers locally stored value of
1103  * the original value of sb->sb_mb->m_nextpkt which must be restored
1104  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
1105  */
1106 static void
1107 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
1108 {
1109 
1110 	KASSERT(solocked(sb->sb_so));
1111 
1112 	/*
1113 	 * First, update for the new value of nextrecord.  If necessary,
1114 	 * make it the first record.
1115 	 */
1116 	if (sb->sb_mb != NULL)
1117 		sb->sb_mb->m_nextpkt = nextrecord;
1118 	else
1119 		sb->sb_mb = nextrecord;
1120 
1121         /*
1122          * Now update any dependent socket buffer fields to reflect
1123          * the new state.  This is an inline of SB_EMPTY_FIXUP, with
1124          * the addition of a second clause that takes care of the
1125          * case where sb_mb has been updated, but remains the last
1126          * record.
1127          */
1128         if (sb->sb_mb == NULL) {
1129                 sb->sb_mbtail = NULL;
1130                 sb->sb_lastrecord = NULL;
1131         } else if (sb->sb_mb->m_nextpkt == NULL)
1132                 sb->sb_lastrecord = sb->sb_mb;
1133 }
1134 
1135 /*
1136  * Implement receive operations on a socket.
1137  * We depend on the way that records are added to the sockbuf
1138  * by sbappend*.  In particular, each record (mbufs linked through m_next)
1139  * must begin with an address if the protocol so specifies,
1140  * followed by an optional mbuf or mbufs containing ancillary data,
1141  * and then zero or more mbufs of data.
1142  * In order to avoid blocking network interrupts for the entire time here,
1143  * we splx() while doing the actual copy to user space.
1144  * Although the sockbuf is locked, new data may still be appended,
1145  * and thus we must maintain consistency of the sockbuf during that time.
1146  *
1147  * The caller may receive the data as a single mbuf chain by supplying
1148  * an mbuf **mp0 for use in returning the chain.  The uio is then used
1149  * only for the count in uio_resid.
1150  */
1151 int
1152 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
1153 	struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1154 {
1155 	struct lwp *l = curlwp;
1156 	struct mbuf	*m, **mp, *mt;
1157 	int atomic, flags, len, error, s, offset, moff, type, orig_resid;
1158 	const struct protosw	*pr;
1159 	struct mbuf	*nextrecord;
1160 	int		mbuf_removed = 0;
1161 	const struct domain *dom;
1162 
1163 	pr = so->so_proto;
1164 	atomic = pr->pr_flags & PR_ATOMIC;
1165 	dom = pr->pr_domain;
1166 	mp = mp0;
1167 	type = 0;
1168 	orig_resid = uio->uio_resid;
1169 
1170 	if (paddr != NULL)
1171 		*paddr = NULL;
1172 	if (controlp != NULL)
1173 		*controlp = NULL;
1174 	if (flagsp != NULL)
1175 		flags = *flagsp &~ MSG_EOR;
1176 	else
1177 		flags = 0;
1178 
1179 	if ((flags & MSG_DONTWAIT) == 0)
1180 		sodopendfree();
1181 
1182 	if (flags & MSG_OOB) {
1183 		m = m_get(M_WAIT, MT_DATA);
1184 		solock(so);
1185 		error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m,
1186 		    (struct mbuf *)(long)(flags & MSG_PEEK), NULL, l);
1187 		sounlock(so);
1188 		if (error)
1189 			goto bad;
1190 		do {
1191 			error = uiomove(mtod(m, void *),
1192 			    (int) min(uio->uio_resid, m->m_len), uio);
1193 			m = m_free(m);
1194 		} while (uio->uio_resid > 0 && error == 0 && m);
1195  bad:
1196 		if (m != NULL)
1197 			m_freem(m);
1198 		return error;
1199 	}
1200 	if (mp != NULL)
1201 		*mp = NULL;
1202 
1203 	/*
1204 	 * solock() provides atomicity of access.  splsoftnet() prevents
1205 	 * protocol processing soft interrupts from interrupting us and
1206 	 * blocking (expensive).
1207 	 */
1208 	s = splsoftnet();
1209 	solock(so);
1210 	if (so->so_state & SS_ISCONFIRMING && uio->uio_resid)
1211 		(*pr->pr_usrreq)(so, PRU_RCVD, NULL, NULL, NULL, l);
1212 
1213  restart:
1214 	if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) {
1215 		sounlock(so);
1216 		splx(s);
1217 		return error;
1218 	}
1219 
1220 	m = so->so_rcv.sb_mb;
1221 	/*
1222 	 * If we have less data than requested, block awaiting more
1223 	 * (subject to any timeout) if:
1224 	 *   1. the current count is less than the low water mark,
1225 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1226 	 *	receive operation at once if we block (resid <= hiwat), or
1227 	 *   3. MSG_DONTWAIT is not set.
1228 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1229 	 * we have to do the receive in sections, and thus risk returning
1230 	 * a short count if a timeout or signal occurs after we start.
1231 	 */
1232 	if (m == NULL ||
1233 	    ((flags & MSG_DONTWAIT) == 0 &&
1234 	     so->so_rcv.sb_cc < uio->uio_resid &&
1235 	     (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1236 	      ((flags & MSG_WAITALL) &&
1237 	       uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1238 	     m->m_nextpkt == NULL && !atomic)) {
1239 #ifdef DIAGNOSTIC
1240 		if (m == NULL && so->so_rcv.sb_cc)
1241 			panic("receive 1");
1242 #endif
1243 		if (so->so_error) {
1244 			if (m != NULL)
1245 				goto dontblock;
1246 			error = so->so_error;
1247 			if ((flags & MSG_PEEK) == 0)
1248 				so->so_error = 0;
1249 			goto release;
1250 		}
1251 		if (so->so_state & SS_CANTRCVMORE) {
1252 			if (m != NULL)
1253 				goto dontblock;
1254 			else
1255 				goto release;
1256 		}
1257 		for (; m != NULL; m = m->m_next)
1258 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1259 				m = so->so_rcv.sb_mb;
1260 				goto dontblock;
1261 			}
1262 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1263 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1264 			error = ENOTCONN;
1265 			goto release;
1266 		}
1267 		if (uio->uio_resid == 0)
1268 			goto release;
1269 		if (so->so_nbio || (flags & MSG_DONTWAIT)) {
1270 			error = EWOULDBLOCK;
1271 			goto release;
1272 		}
1273 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
1274 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
1275 		sbunlock(&so->so_rcv);
1276 		error = sbwait(&so->so_rcv);
1277 		if (error != 0) {
1278 			sounlock(so);
1279 			splx(s);
1280 			return error;
1281 		}
1282 		goto restart;
1283 	}
1284  dontblock:
1285 	/*
1286 	 * On entry here, m points to the first record of the socket buffer.
1287 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1288 	 * pointer to the next record in the socket buffer.  We must keep the
1289 	 * various socket buffer pointers and local stack versions of the
1290 	 * pointers in sync, pushing out modifications before dropping the
1291 	 * socket lock, and re-reading them when picking it up.
1292 	 *
1293 	 * Otherwise, we will race with the network stack appending new data
1294 	 * or records onto the socket buffer by using inconsistent/stale
1295 	 * versions of the field, possibly resulting in socket buffer
1296 	 * corruption.
1297 	 *
1298 	 * By holding the high-level sblock(), we prevent simultaneous
1299 	 * readers from pulling off the front of the socket buffer.
1300 	 */
1301 	if (l != NULL)
1302 		l->l_ru.ru_msgrcv++;
1303 	KASSERT(m == so->so_rcv.sb_mb);
1304 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
1305 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
1306 	nextrecord = m->m_nextpkt;
1307 	if (pr->pr_flags & PR_ADDR) {
1308 #ifdef DIAGNOSTIC
1309 		if (m->m_type != MT_SONAME)
1310 			panic("receive 1a");
1311 #endif
1312 		orig_resid = 0;
1313 		if (flags & MSG_PEEK) {
1314 			if (paddr)
1315 				*paddr = m_copy(m, 0, m->m_len);
1316 			m = m->m_next;
1317 		} else {
1318 			sbfree(&so->so_rcv, m);
1319 			mbuf_removed = 1;
1320 			if (paddr != NULL) {
1321 				*paddr = m;
1322 				so->so_rcv.sb_mb = m->m_next;
1323 				m->m_next = NULL;
1324 				m = so->so_rcv.sb_mb;
1325 			} else {
1326 				MFREE(m, so->so_rcv.sb_mb);
1327 				m = so->so_rcv.sb_mb;
1328 			}
1329 			sbsync(&so->so_rcv, nextrecord);
1330 		}
1331 	}
1332 
1333 	/*
1334 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1335 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1336 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1337 	 * perform externalization (or freeing if controlp == NULL).
1338 	 */
1339 	if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) {
1340 		struct mbuf *cm = NULL, *cmn;
1341 		struct mbuf **cme = &cm;
1342 
1343 		do {
1344 			if (flags & MSG_PEEK) {
1345 				if (controlp != NULL) {
1346 					*controlp = m_copy(m, 0, m->m_len);
1347 					controlp = &(*controlp)->m_next;
1348 				}
1349 				m = m->m_next;
1350 			} else {
1351 				sbfree(&so->so_rcv, m);
1352 				so->so_rcv.sb_mb = m->m_next;
1353 				m->m_next = NULL;
1354 				*cme = m;
1355 				cme = &(*cme)->m_next;
1356 				m = so->so_rcv.sb_mb;
1357 			}
1358 		} while (m != NULL && m->m_type == MT_CONTROL);
1359 		if ((flags & MSG_PEEK) == 0)
1360 			sbsync(&so->so_rcv, nextrecord);
1361 		for (; cm != NULL; cm = cmn) {
1362 			cmn = cm->m_next;
1363 			cm->m_next = NULL;
1364 			type = mtod(cm, struct cmsghdr *)->cmsg_type;
1365 			if (controlp != NULL) {
1366 				if (dom->dom_externalize != NULL &&
1367 				    type == SCM_RIGHTS) {
1368 					sounlock(so);
1369 					splx(s);
1370 					error = (*dom->dom_externalize)(cm, l);
1371 					s = splsoftnet();
1372 					solock(so);
1373 				}
1374 				*controlp = cm;
1375 				while (*controlp != NULL)
1376 					controlp = &(*controlp)->m_next;
1377 			} else {
1378 				/*
1379 				 * Dispose of any SCM_RIGHTS message that went
1380 				 * through the read path rather than recv.
1381 				 */
1382 				if (dom->dom_dispose != NULL &&
1383 				    type == SCM_RIGHTS) {
1384 				    	sounlock(so);
1385 					(*dom->dom_dispose)(cm);
1386 					solock(so);
1387 				}
1388 				m_freem(cm);
1389 			}
1390 		}
1391 		if (m != NULL)
1392 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1393 		else
1394 			nextrecord = so->so_rcv.sb_mb;
1395 		orig_resid = 0;
1396 	}
1397 
1398 	/* If m is non-NULL, we have some data to read. */
1399 	if (__predict_true(m != NULL)) {
1400 		type = m->m_type;
1401 		if (type == MT_OOBDATA)
1402 			flags |= MSG_OOB;
1403 	}
1404 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1405 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1406 
1407 	moff = 0;
1408 	offset = 0;
1409 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1410 		if (m->m_type == MT_OOBDATA) {
1411 			if (type != MT_OOBDATA)
1412 				break;
1413 		} else if (type == MT_OOBDATA)
1414 			break;
1415 #ifdef DIAGNOSTIC
1416 		else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
1417 			panic("receive 3");
1418 #endif
1419 		so->so_state &= ~SS_RCVATMARK;
1420 		len = uio->uio_resid;
1421 		if (so->so_oobmark && len > so->so_oobmark - offset)
1422 			len = so->so_oobmark - offset;
1423 		if (len > m->m_len - moff)
1424 			len = m->m_len - moff;
1425 		/*
1426 		 * If mp is set, just pass back the mbufs.
1427 		 * Otherwise copy them out via the uio, then free.
1428 		 * Sockbuf must be consistent here (points to current mbuf,
1429 		 * it points to next record) when we drop priority;
1430 		 * we must note any additions to the sockbuf when we
1431 		 * block interrupts again.
1432 		 */
1433 		if (mp == NULL) {
1434 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1435 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1436 			sounlock(so);
1437 			splx(s);
1438 			error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1439 			s = splsoftnet();
1440 			solock(so);
1441 			if (error != 0) {
1442 				/*
1443 				 * If any part of the record has been removed
1444 				 * (such as the MT_SONAME mbuf, which will
1445 				 * happen when PR_ADDR, and thus also
1446 				 * PR_ATOMIC, is set), then drop the entire
1447 				 * record to maintain the atomicity of the
1448 				 * receive operation.
1449 				 *
1450 				 * This avoids a later panic("receive 1a")
1451 				 * when compiled with DIAGNOSTIC.
1452 				 */
1453 				if (m && mbuf_removed && atomic)
1454 					(void) sbdroprecord(&so->so_rcv);
1455 
1456 				goto release;
1457 			}
1458 		} else
1459 			uio->uio_resid -= len;
1460 		if (len == m->m_len - moff) {
1461 			if (m->m_flags & M_EOR)
1462 				flags |= MSG_EOR;
1463 			if (flags & MSG_PEEK) {
1464 				m = m->m_next;
1465 				moff = 0;
1466 			} else {
1467 				nextrecord = m->m_nextpkt;
1468 				sbfree(&so->so_rcv, m);
1469 				if (mp) {
1470 					*mp = m;
1471 					mp = &m->m_next;
1472 					so->so_rcv.sb_mb = m = m->m_next;
1473 					*mp = NULL;
1474 				} else {
1475 					MFREE(m, so->so_rcv.sb_mb);
1476 					m = so->so_rcv.sb_mb;
1477 				}
1478 				/*
1479 				 * If m != NULL, we also know that
1480 				 * so->so_rcv.sb_mb != NULL.
1481 				 */
1482 				KASSERT(so->so_rcv.sb_mb == m);
1483 				if (m) {
1484 					m->m_nextpkt = nextrecord;
1485 					if (nextrecord == NULL)
1486 						so->so_rcv.sb_lastrecord = m;
1487 				} else {
1488 					so->so_rcv.sb_mb = nextrecord;
1489 					SB_EMPTY_FIXUP(&so->so_rcv);
1490 				}
1491 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1492 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1493 			}
1494 		} else if (flags & MSG_PEEK)
1495 			moff += len;
1496 		else {
1497 			if (mp != NULL) {
1498 				mt = m_copym(m, 0, len, M_NOWAIT);
1499 				if (__predict_false(mt == NULL)) {
1500 					sounlock(so);
1501 					mt = m_copym(m, 0, len, M_WAIT);
1502 					solock(so);
1503 				}
1504 				*mp = mt;
1505 			}
1506 			m->m_data += len;
1507 			m->m_len -= len;
1508 			so->so_rcv.sb_cc -= len;
1509 		}
1510 		if (so->so_oobmark) {
1511 			if ((flags & MSG_PEEK) == 0) {
1512 				so->so_oobmark -= len;
1513 				if (so->so_oobmark == 0) {
1514 					so->so_state |= SS_RCVATMARK;
1515 					break;
1516 				}
1517 			} else {
1518 				offset += len;
1519 				if (offset == so->so_oobmark)
1520 					break;
1521 			}
1522 		}
1523 		if (flags & MSG_EOR)
1524 			break;
1525 		/*
1526 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1527 		 * we must not quit until "uio->uio_resid == 0" or an error
1528 		 * termination.  If a signal/timeout occurs, return
1529 		 * with a short count but without error.
1530 		 * Keep sockbuf locked against other readers.
1531 		 */
1532 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1533 		    !sosendallatonce(so) && !nextrecord) {
1534 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1535 				break;
1536 			/*
1537 			 * If we are peeking and the socket receive buffer is
1538 			 * full, stop since we can't get more data to peek at.
1539 			 */
1540 			if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
1541 				break;
1542 			/*
1543 			 * If we've drained the socket buffer, tell the
1544 			 * protocol in case it needs to do something to
1545 			 * get it filled again.
1546 			 */
1547 			if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1548 				(*pr->pr_usrreq)(so, PRU_RCVD,
1549 				    NULL, (struct mbuf *)(long)flags, NULL, l);
1550 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1551 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1552 			error = sbwait(&so->so_rcv);
1553 			if (error != 0) {
1554 				sbunlock(&so->so_rcv);
1555 				sounlock(so);
1556 				splx(s);
1557 				return 0;
1558 			}
1559 			if ((m = so->so_rcv.sb_mb) != NULL)
1560 				nextrecord = m->m_nextpkt;
1561 		}
1562 	}
1563 
1564 	if (m && atomic) {
1565 		flags |= MSG_TRUNC;
1566 		if ((flags & MSG_PEEK) == 0)
1567 			(void) sbdroprecord(&so->so_rcv);
1568 	}
1569 	if ((flags & MSG_PEEK) == 0) {
1570 		if (m == NULL) {
1571 			/*
1572 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1573 			 * part makes sure sb_lastrecord is up-to-date if
1574 			 * there is still data in the socket buffer.
1575 			 */
1576 			so->so_rcv.sb_mb = nextrecord;
1577 			if (so->so_rcv.sb_mb == NULL) {
1578 				so->so_rcv.sb_mbtail = NULL;
1579 				so->so_rcv.sb_lastrecord = NULL;
1580 			} else if (nextrecord->m_nextpkt == NULL)
1581 				so->so_rcv.sb_lastrecord = nextrecord;
1582 		}
1583 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1584 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1585 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1586 			(*pr->pr_usrreq)(so, PRU_RCVD, NULL,
1587 			    (struct mbuf *)(long)flags, NULL, l);
1588 	}
1589 	if (orig_resid == uio->uio_resid && orig_resid &&
1590 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1591 		sbunlock(&so->so_rcv);
1592 		goto restart;
1593 	}
1594 
1595 	if (flagsp != NULL)
1596 		*flagsp |= flags;
1597  release:
1598 	sbunlock(&so->so_rcv);
1599 	sounlock(so);
1600 	splx(s);
1601 	return error;
1602 }
1603 
1604 int
1605 soshutdown(struct socket *so, int how)
1606 {
1607 	const struct protosw	*pr;
1608 	int	error;
1609 
1610 	KASSERT(solocked(so));
1611 
1612 	pr = so->so_proto;
1613 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1614 		return (EINVAL);
1615 
1616 	if (how == SHUT_RD || how == SHUT_RDWR) {
1617 		sorflush(so);
1618 		error = 0;
1619 	}
1620 	if (how == SHUT_WR || how == SHUT_RDWR)
1621 		error = (*pr->pr_usrreq)(so, PRU_SHUTDOWN, NULL,
1622 		    NULL, NULL, NULL);
1623 
1624 	return error;
1625 }
1626 
1627 int
1628 sodrain(struct socket *so)
1629 {
1630 	int error;
1631 
1632 	solock(so);
1633 	so->so_state |= SS_ISDRAINING;
1634 	cv_broadcast(&so->so_cv);
1635 	error = soshutdown(so, SHUT_RDWR);
1636 	sounlock(so);
1637 
1638 	return error;
1639 }
1640 
1641 void
1642 sorflush(struct socket *so)
1643 {
1644 	struct sockbuf	*sb, asb;
1645 	const struct protosw	*pr;
1646 
1647 	KASSERT(solocked(so));
1648 
1649 	sb = &so->so_rcv;
1650 	pr = so->so_proto;
1651 	socantrcvmore(so);
1652 	sb->sb_flags |= SB_NOINTR;
1653 	(void )sblock(sb, M_WAITOK);
1654 	sbunlock(sb);
1655 	asb = *sb;
1656 	/*
1657 	 * Clear most of the sockbuf structure, but leave some of the
1658 	 * fields valid.
1659 	 */
1660 	memset(&sb->sb_startzero, 0,
1661 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1662 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) {
1663 		sounlock(so);
1664 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1665 		solock(so);
1666 	}
1667 	sbrelease(&asb, so);
1668 }
1669 
1670 /*
1671  * internal set SOL_SOCKET options
1672  */
1673 static int
1674 sosetopt1(struct socket *so, const struct sockopt *sopt)
1675 {
1676 	int error = EINVAL, optval, opt;
1677 	struct linger l;
1678 	struct timeval tv;
1679 
1680 	switch ((opt = sopt->sopt_name)) {
1681 
1682 	case SO_ACCEPTFILTER:
1683 		error = accept_filt_setopt(so, sopt);
1684 		KASSERT(solocked(so));
1685 		break;
1686 
1687   	case SO_LINGER:
1688  		error = sockopt_get(sopt, &l, sizeof(l));
1689 		solock(so);
1690  		if (error)
1691  			break;
1692  		if (l.l_linger < 0 || l.l_linger > USHRT_MAX ||
1693  		    l.l_linger > (INT_MAX / hz)) {
1694 			error = EDOM;
1695 			break;
1696 		}
1697  		so->so_linger = l.l_linger;
1698  		if (l.l_onoff)
1699  			so->so_options |= SO_LINGER;
1700  		else
1701  			so->so_options &= ~SO_LINGER;
1702    		break;
1703 
1704 	case SO_DEBUG:
1705 	case SO_KEEPALIVE:
1706 	case SO_DONTROUTE:
1707 	case SO_USELOOPBACK:
1708 	case SO_BROADCAST:
1709 	case SO_REUSEADDR:
1710 	case SO_REUSEPORT:
1711 	case SO_OOBINLINE:
1712 	case SO_TIMESTAMP:
1713 #ifdef SO_OTIMESTAMP
1714 	case SO_OTIMESTAMP:
1715 #endif
1716 		error = sockopt_getint(sopt, &optval);
1717 		solock(so);
1718 		if (error)
1719 			break;
1720 		if (optval)
1721 			so->so_options |= opt;
1722 		else
1723 			so->so_options &= ~opt;
1724 		break;
1725 
1726 	case SO_SNDBUF:
1727 	case SO_RCVBUF:
1728 	case SO_SNDLOWAT:
1729 	case SO_RCVLOWAT:
1730 		error = sockopt_getint(sopt, &optval);
1731 		solock(so);
1732 		if (error)
1733 			break;
1734 
1735 		/*
1736 		 * Values < 1 make no sense for any of these
1737 		 * options, so disallow them.
1738 		 */
1739 		if (optval < 1) {
1740 			error = EINVAL;
1741 			break;
1742 		}
1743 
1744 		switch (opt) {
1745 		case SO_SNDBUF:
1746 			if (sbreserve(&so->so_snd, (u_long)optval, so) == 0) {
1747 				error = ENOBUFS;
1748 				break;
1749 			}
1750 			so->so_snd.sb_flags &= ~SB_AUTOSIZE;
1751 			break;
1752 
1753 		case SO_RCVBUF:
1754 			if (sbreserve(&so->so_rcv, (u_long)optval, so) == 0) {
1755 				error = ENOBUFS;
1756 				break;
1757 			}
1758 			so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1759 			break;
1760 
1761 		/*
1762 		 * Make sure the low-water is never greater than
1763 		 * the high-water.
1764 		 */
1765 		case SO_SNDLOWAT:
1766 			if (optval > so->so_snd.sb_hiwat)
1767 				optval = so->so_snd.sb_hiwat;
1768 
1769 			so->so_snd.sb_lowat = optval;
1770 			break;
1771 
1772 		case SO_RCVLOWAT:
1773 			if (optval > so->so_rcv.sb_hiwat)
1774 				optval = so->so_rcv.sb_hiwat;
1775 
1776 			so->so_rcv.sb_lowat = optval;
1777 			break;
1778 		}
1779 		break;
1780 
1781 #ifdef COMPAT_50
1782 	case SO_OSNDTIMEO:
1783 	case SO_ORCVTIMEO: {
1784 		struct timeval50 otv;
1785 		error = sockopt_get(sopt, &otv, sizeof(otv));
1786 		if (error) {
1787 			solock(so);
1788 			break;
1789 		}
1790 		timeval50_to_timeval(&otv, &tv);
1791 		opt = opt == SO_OSNDTIMEO ? SO_SNDTIMEO : SO_RCVTIMEO;
1792 		error = 0;
1793 		/*FALLTHROUGH*/
1794 	}
1795 #endif /* COMPAT_50 */
1796 
1797 	case SO_SNDTIMEO:
1798 	case SO_RCVTIMEO:
1799 		if (error)
1800 			error = sockopt_get(sopt, &tv, sizeof(tv));
1801 		solock(so);
1802 		if (error)
1803 			break;
1804 
1805 		if (tv.tv_sec > (INT_MAX - tv.tv_usec / tick) / hz) {
1806 			error = EDOM;
1807 			break;
1808 		}
1809 
1810 		optval = tv.tv_sec * hz + tv.tv_usec / tick;
1811 		if (optval == 0 && tv.tv_usec != 0)
1812 			optval = 1;
1813 
1814 		switch (opt) {
1815 		case SO_SNDTIMEO:
1816 			so->so_snd.sb_timeo = optval;
1817 			break;
1818 		case SO_RCVTIMEO:
1819 			so->so_rcv.sb_timeo = optval;
1820 			break;
1821 		}
1822 		break;
1823 
1824 	default:
1825 		solock(so);
1826 		error = ENOPROTOOPT;
1827 		break;
1828 	}
1829 	KASSERT(solocked(so));
1830 	return error;
1831 }
1832 
1833 int
1834 sosetopt(struct socket *so, struct sockopt *sopt)
1835 {
1836 	int error, prerr;
1837 
1838 	if (sopt->sopt_level == SOL_SOCKET) {
1839 		error = sosetopt1(so, sopt);
1840 		KASSERT(solocked(so));
1841 	} else {
1842 		error = ENOPROTOOPT;
1843 		solock(so);
1844 	}
1845 
1846 	if ((error == 0 || error == ENOPROTOOPT) &&
1847 	    so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) {
1848 		/* give the protocol stack a shot */
1849 		prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, sopt);
1850 		if (prerr == 0)
1851 			error = 0;
1852 		else if (prerr != ENOPROTOOPT)
1853 			error = prerr;
1854 	}
1855 	sounlock(so);
1856 	return error;
1857 }
1858 
1859 /*
1860  * so_setsockopt() is a wrapper providing a sockopt structure for sosetopt()
1861  */
1862 int
1863 so_setsockopt(struct lwp *l, struct socket *so, int level, int name,
1864     const void *val, size_t valsize)
1865 {
1866 	struct sockopt sopt;
1867 	int error;
1868 
1869 	KASSERT(valsize == 0 || val != NULL);
1870 
1871 	sockopt_init(&sopt, level, name, valsize);
1872 	sockopt_set(&sopt, val, valsize);
1873 
1874 	error = sosetopt(so, &sopt);
1875 
1876 	sockopt_destroy(&sopt);
1877 
1878 	return error;
1879 }
1880 
1881 /*
1882  * internal get SOL_SOCKET options
1883  */
1884 static int
1885 sogetopt1(struct socket *so, struct sockopt *sopt)
1886 {
1887 	int error, optval, opt;
1888 	struct linger l;
1889 	struct timeval tv;
1890 
1891 	switch ((opt = sopt->sopt_name)) {
1892 
1893 	case SO_ACCEPTFILTER:
1894 		error = accept_filt_getopt(so, sopt);
1895 		break;
1896 
1897 	case SO_LINGER:
1898 		l.l_onoff = (so->so_options & SO_LINGER) ? 1 : 0;
1899 		l.l_linger = so->so_linger;
1900 
1901 		error = sockopt_set(sopt, &l, sizeof(l));
1902 		break;
1903 
1904 	case SO_USELOOPBACK:
1905 	case SO_DONTROUTE:
1906 	case SO_DEBUG:
1907 	case SO_KEEPALIVE:
1908 	case SO_REUSEADDR:
1909 	case SO_REUSEPORT:
1910 	case SO_BROADCAST:
1911 	case SO_OOBINLINE:
1912 	case SO_TIMESTAMP:
1913 #ifdef SO_OTIMESTAMP
1914 	case SO_OTIMESTAMP:
1915 #endif
1916 		error = sockopt_setint(sopt, (so->so_options & opt) ? 1 : 0);
1917 		break;
1918 
1919 	case SO_TYPE:
1920 		error = sockopt_setint(sopt, so->so_type);
1921 		break;
1922 
1923 	case SO_ERROR:
1924 		error = sockopt_setint(sopt, so->so_error);
1925 		so->so_error = 0;
1926 		break;
1927 
1928 	case SO_SNDBUF:
1929 		error = sockopt_setint(sopt, so->so_snd.sb_hiwat);
1930 		break;
1931 
1932 	case SO_RCVBUF:
1933 		error = sockopt_setint(sopt, so->so_rcv.sb_hiwat);
1934 		break;
1935 
1936 	case SO_SNDLOWAT:
1937 		error = sockopt_setint(sopt, so->so_snd.sb_lowat);
1938 		break;
1939 
1940 	case SO_RCVLOWAT:
1941 		error = sockopt_setint(sopt, so->so_rcv.sb_lowat);
1942 		break;
1943 
1944 #ifdef COMPAT_50
1945 	case SO_OSNDTIMEO:
1946 	case SO_ORCVTIMEO: {
1947 		struct timeval50 otv;
1948 
1949 		optval = (opt == SO_OSNDTIMEO ?
1950 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1951 
1952 		otv.tv_sec = optval / hz;
1953 		otv.tv_usec = (optval % hz) * tick;
1954 
1955 		error = sockopt_set(sopt, &otv, sizeof(otv));
1956 		break;
1957 	}
1958 #endif /* COMPAT_50 */
1959 
1960 	case SO_SNDTIMEO:
1961 	case SO_RCVTIMEO:
1962 		optval = (opt == SO_SNDTIMEO ?
1963 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1964 
1965 		tv.tv_sec = optval / hz;
1966 		tv.tv_usec = (optval % hz) * tick;
1967 
1968 		error = sockopt_set(sopt, &tv, sizeof(tv));
1969 		break;
1970 
1971 	case SO_OVERFLOWED:
1972 		error = sockopt_setint(sopt, so->so_rcv.sb_overflowed);
1973 		break;
1974 
1975 	default:
1976 		error = ENOPROTOOPT;
1977 		break;
1978 	}
1979 
1980 	return (error);
1981 }
1982 
1983 int
1984 sogetopt(struct socket *so, struct sockopt *sopt)
1985 {
1986 	int		error;
1987 
1988 	solock(so);
1989 	if (sopt->sopt_level != SOL_SOCKET) {
1990 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1991 			error = ((*so->so_proto->pr_ctloutput)
1992 			    (PRCO_GETOPT, so, sopt));
1993 		} else
1994 			error = (ENOPROTOOPT);
1995 	} else {
1996 		error = sogetopt1(so, sopt);
1997 	}
1998 	sounlock(so);
1999 	return (error);
2000 }
2001 
2002 /*
2003  * alloc sockopt data buffer buffer
2004  *	- will be released at destroy
2005  */
2006 static int
2007 sockopt_alloc(struct sockopt *sopt, size_t len, km_flag_t kmflag)
2008 {
2009 
2010 	KASSERT(sopt->sopt_size == 0);
2011 
2012 	if (len > sizeof(sopt->sopt_buf)) {
2013 		sopt->sopt_data = kmem_zalloc(len, kmflag);
2014 		if (sopt->sopt_data == NULL)
2015 			return ENOMEM;
2016 	} else
2017 		sopt->sopt_data = sopt->sopt_buf;
2018 
2019 	sopt->sopt_size = len;
2020 	return 0;
2021 }
2022 
2023 /*
2024  * initialise sockopt storage
2025  *	- MAY sleep during allocation
2026  */
2027 void
2028 sockopt_init(struct sockopt *sopt, int level, int name, size_t size)
2029 {
2030 
2031 	memset(sopt, 0, sizeof(*sopt));
2032 
2033 	sopt->sopt_level = level;
2034 	sopt->sopt_name = name;
2035 	(void)sockopt_alloc(sopt, size, KM_SLEEP);
2036 }
2037 
2038 /*
2039  * destroy sockopt storage
2040  *	- will release any held memory references
2041  */
2042 void
2043 sockopt_destroy(struct sockopt *sopt)
2044 {
2045 
2046 	if (sopt->sopt_data != sopt->sopt_buf)
2047 		kmem_free(sopt->sopt_data, sopt->sopt_size);
2048 
2049 	memset(sopt, 0, sizeof(*sopt));
2050 }
2051 
2052 /*
2053  * set sockopt value
2054  *	- value is copied into sockopt
2055  * 	- memory is allocated when necessary, will not sleep
2056  */
2057 int
2058 sockopt_set(struct sockopt *sopt, const void *buf, size_t len)
2059 {
2060 	int error;
2061 
2062 	if (sopt->sopt_size == 0) {
2063 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2064 		if (error)
2065 			return error;
2066 	}
2067 
2068 	KASSERT(sopt->sopt_size == len);
2069 	memcpy(sopt->sopt_data, buf, len);
2070 	return 0;
2071 }
2072 
2073 /*
2074  * common case of set sockopt integer value
2075  */
2076 int
2077 sockopt_setint(struct sockopt *sopt, int val)
2078 {
2079 
2080 	return sockopt_set(sopt, &val, sizeof(int));
2081 }
2082 
2083 /*
2084  * get sockopt value
2085  *	- correct size must be given
2086  */
2087 int
2088 sockopt_get(const struct sockopt *sopt, void *buf, size_t len)
2089 {
2090 
2091 	if (sopt->sopt_size != len)
2092 		return EINVAL;
2093 
2094 	memcpy(buf, sopt->sopt_data, len);
2095 	return 0;
2096 }
2097 
2098 /*
2099  * common case of get sockopt integer value
2100  */
2101 int
2102 sockopt_getint(const struct sockopt *sopt, int *valp)
2103 {
2104 
2105 	return sockopt_get(sopt, valp, sizeof(int));
2106 }
2107 
2108 /*
2109  * set sockopt value from mbuf
2110  *	- ONLY for legacy code
2111  *	- mbuf is released by sockopt
2112  *	- will not sleep
2113  */
2114 int
2115 sockopt_setmbuf(struct sockopt *sopt, struct mbuf *m)
2116 {
2117 	size_t len;
2118 	int error;
2119 
2120 	len = m_length(m);
2121 
2122 	if (sopt->sopt_size == 0) {
2123 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2124 		if (error)
2125 			return error;
2126 	}
2127 
2128 	KASSERT(sopt->sopt_size == len);
2129 	m_copydata(m, 0, len, sopt->sopt_data);
2130 	m_freem(m);
2131 
2132 	return 0;
2133 }
2134 
2135 /*
2136  * get sockopt value into mbuf
2137  *	- ONLY for legacy code
2138  *	- mbuf to be released by the caller
2139  *	- will not sleep
2140  */
2141 struct mbuf *
2142 sockopt_getmbuf(const struct sockopt *sopt)
2143 {
2144 	struct mbuf *m;
2145 
2146 	if (sopt->sopt_size > MCLBYTES)
2147 		return NULL;
2148 
2149 	m = m_get(M_DONTWAIT, MT_SOOPTS);
2150 	if (m == NULL)
2151 		return NULL;
2152 
2153 	if (sopt->sopt_size > MLEN) {
2154 		MCLGET(m, M_DONTWAIT);
2155 		if ((m->m_flags & M_EXT) == 0) {
2156 			m_free(m);
2157 			return NULL;
2158 		}
2159 	}
2160 
2161 	memcpy(mtod(m, void *), sopt->sopt_data, sopt->sopt_size);
2162 	m->m_len = sopt->sopt_size;
2163 
2164 	return m;
2165 }
2166 
2167 void
2168 sohasoutofband(struct socket *so)
2169 {
2170 
2171 	fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
2172 	selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, NOTE_SUBMIT);
2173 }
2174 
2175 static void
2176 filt_sordetach(struct knote *kn)
2177 {
2178 	struct socket	*so;
2179 
2180 	so = ((file_t *)kn->kn_obj)->f_data;
2181 	solock(so);
2182 	SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
2183 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
2184 		so->so_rcv.sb_flags &= ~SB_KNOTE;
2185 	sounlock(so);
2186 }
2187 
2188 /*ARGSUSED*/
2189 static int
2190 filt_soread(struct knote *kn, long hint)
2191 {
2192 	struct socket	*so;
2193 	int rv;
2194 
2195 	so = ((file_t *)kn->kn_obj)->f_data;
2196 	if (hint != NOTE_SUBMIT)
2197 		solock(so);
2198 	kn->kn_data = so->so_rcv.sb_cc;
2199 	if (so->so_state & SS_CANTRCVMORE) {
2200 		kn->kn_flags |= EV_EOF;
2201 		kn->kn_fflags = so->so_error;
2202 		rv = 1;
2203 	} else if (so->so_error)	/* temporary udp error */
2204 		rv = 1;
2205 	else if (kn->kn_sfflags & NOTE_LOWAT)
2206 		rv = (kn->kn_data >= kn->kn_sdata);
2207 	else
2208 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
2209 	if (hint != NOTE_SUBMIT)
2210 		sounlock(so);
2211 	return rv;
2212 }
2213 
2214 static void
2215 filt_sowdetach(struct knote *kn)
2216 {
2217 	struct socket	*so;
2218 
2219 	so = ((file_t *)kn->kn_obj)->f_data;
2220 	solock(so);
2221 	SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
2222 	if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
2223 		so->so_snd.sb_flags &= ~SB_KNOTE;
2224 	sounlock(so);
2225 }
2226 
2227 /*ARGSUSED*/
2228 static int
2229 filt_sowrite(struct knote *kn, long hint)
2230 {
2231 	struct socket	*so;
2232 	int rv;
2233 
2234 	so = ((file_t *)kn->kn_obj)->f_data;
2235 	if (hint != NOTE_SUBMIT)
2236 		solock(so);
2237 	kn->kn_data = sbspace(&so->so_snd);
2238 	if (so->so_state & SS_CANTSENDMORE) {
2239 		kn->kn_flags |= EV_EOF;
2240 		kn->kn_fflags = so->so_error;
2241 		rv = 1;
2242 	} else if (so->so_error)	/* temporary udp error */
2243 		rv = 1;
2244 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2245 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
2246 		rv = 0;
2247 	else if (kn->kn_sfflags & NOTE_LOWAT)
2248 		rv = (kn->kn_data >= kn->kn_sdata);
2249 	else
2250 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
2251 	if (hint != NOTE_SUBMIT)
2252 		sounlock(so);
2253 	return rv;
2254 }
2255 
2256 /*ARGSUSED*/
2257 static int
2258 filt_solisten(struct knote *kn, long hint)
2259 {
2260 	struct socket	*so;
2261 	int rv;
2262 
2263 	so = ((file_t *)kn->kn_obj)->f_data;
2264 
2265 	/*
2266 	 * Set kn_data to number of incoming connections, not
2267 	 * counting partial (incomplete) connections.
2268 	 */
2269 	if (hint != NOTE_SUBMIT)
2270 		solock(so);
2271 	kn->kn_data = so->so_qlen;
2272 	rv = (kn->kn_data > 0);
2273 	if (hint != NOTE_SUBMIT)
2274 		sounlock(so);
2275 	return rv;
2276 }
2277 
2278 static const struct filterops solisten_filtops =
2279 	{ 1, NULL, filt_sordetach, filt_solisten };
2280 static const struct filterops soread_filtops =
2281 	{ 1, NULL, filt_sordetach, filt_soread };
2282 static const struct filterops sowrite_filtops =
2283 	{ 1, NULL, filt_sowdetach, filt_sowrite };
2284 
2285 int
2286 soo_kqfilter(struct file *fp, struct knote *kn)
2287 {
2288 	struct socket	*so;
2289 	struct sockbuf	*sb;
2290 
2291 	so = ((file_t *)kn->kn_obj)->f_data;
2292 	solock(so);
2293 	switch (kn->kn_filter) {
2294 	case EVFILT_READ:
2295 		if (so->so_options & SO_ACCEPTCONN)
2296 			kn->kn_fop = &solisten_filtops;
2297 		else
2298 			kn->kn_fop = &soread_filtops;
2299 		sb = &so->so_rcv;
2300 		break;
2301 	case EVFILT_WRITE:
2302 		kn->kn_fop = &sowrite_filtops;
2303 		sb = &so->so_snd;
2304 		break;
2305 	default:
2306 		sounlock(so);
2307 		return (EINVAL);
2308 	}
2309 	SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
2310 	sb->sb_flags |= SB_KNOTE;
2311 	sounlock(so);
2312 	return (0);
2313 }
2314 
2315 static int
2316 sodopoll(struct socket *so, int events)
2317 {
2318 	int revents;
2319 
2320 	revents = 0;
2321 
2322 	if (events & (POLLIN | POLLRDNORM))
2323 		if (soreadable(so))
2324 			revents |= events & (POLLIN | POLLRDNORM);
2325 
2326 	if (events & (POLLOUT | POLLWRNORM))
2327 		if (sowritable(so))
2328 			revents |= events & (POLLOUT | POLLWRNORM);
2329 
2330 	if (events & (POLLPRI | POLLRDBAND))
2331 		if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
2332 			revents |= events & (POLLPRI | POLLRDBAND);
2333 
2334 	return revents;
2335 }
2336 
2337 int
2338 sopoll(struct socket *so, int events)
2339 {
2340 	int revents = 0;
2341 
2342 #ifndef DIAGNOSTIC
2343 	/*
2344 	 * Do a quick, unlocked check in expectation that the socket
2345 	 * will be ready for I/O.  Don't do this check if DIAGNOSTIC,
2346 	 * as the solocked() assertions will fail.
2347 	 */
2348 	if ((revents = sodopoll(so, events)) != 0)
2349 		return revents;
2350 #endif
2351 
2352 	solock(so);
2353 	if ((revents = sodopoll(so, events)) == 0) {
2354 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2355 			selrecord(curlwp, &so->so_rcv.sb_sel);
2356 			so->so_rcv.sb_flags |= SB_NOTIFY;
2357 		}
2358 
2359 		if (events & (POLLOUT | POLLWRNORM)) {
2360 			selrecord(curlwp, &so->so_snd.sb_sel);
2361 			so->so_snd.sb_flags |= SB_NOTIFY;
2362 		}
2363 	}
2364 	sounlock(so);
2365 
2366 	return revents;
2367 }
2368 
2369 
2370 #include <sys/sysctl.h>
2371 
2372 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO);
2373 
2374 /*
2375  * sysctl helper routine for kern.somaxkva.  ensures that the given
2376  * value is not too small.
2377  * (XXX should we maybe make sure it's not too large as well?)
2378  */
2379 static int
2380 sysctl_kern_somaxkva(SYSCTLFN_ARGS)
2381 {
2382 	int error, new_somaxkva;
2383 	struct sysctlnode node;
2384 
2385 	new_somaxkva = somaxkva;
2386 	node = *rnode;
2387 	node.sysctl_data = &new_somaxkva;
2388 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2389 	if (error || newp == NULL)
2390 		return (error);
2391 
2392 	if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */
2393 		return (EINVAL);
2394 
2395 	mutex_enter(&so_pendfree_lock);
2396 	somaxkva = new_somaxkva;
2397 	cv_broadcast(&socurkva_cv);
2398 	mutex_exit(&so_pendfree_lock);
2399 
2400 	return (error);
2401 }
2402 
2403 static void
2404 sysctl_kern_somaxkva_setup(void)
2405 {
2406 
2407 	KASSERT(socket_sysctllog == NULL);
2408 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2409 		       CTLFLAG_PERMANENT,
2410 		       CTLTYPE_NODE, "kern", NULL,
2411 		       NULL, 0, NULL, 0,
2412 		       CTL_KERN, CTL_EOL);
2413 
2414 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2415 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2416 		       CTLTYPE_INT, "somaxkva",
2417 		       SYSCTL_DESCR("Maximum amount of kernel memory to be "
2418 				    "used for socket buffers"),
2419 		       sysctl_kern_somaxkva, 0, NULL, 0,
2420 		       CTL_KERN, KERN_SOMAXKVA, CTL_EOL);
2421 }
2422