xref: /netbsd-src/sys/kern/uipc_socket.c (revision fc4f42693f9b1c31f39f9cf50af1bf2010325808)
1 /*	$NetBSD: uipc_socket.c,v 1.261 2018/03/19 16:32:30 roy 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 /*
66  * Socket operation routines.
67  *
68  * These routines are called by the routines in sys_socket.c or from a
69  * system process, and implement the semantics of socket operations by
70  * switching out to the protocol specific routines.
71  */
72 
73 #include <sys/cdefs.h>
74 __KERNEL_RCSID(0, "$NetBSD: uipc_socket.c,v 1.261 2018/03/19 16:32:30 roy Exp $");
75 
76 #ifdef _KERNEL_OPT
77 #include "opt_compat_netbsd.h"
78 #include "opt_sock_counters.h"
79 #include "opt_sosend_loan.h"
80 #include "opt_mbuftrace.h"
81 #include "opt_somaxkva.h"
82 #include "opt_multiprocessor.h"	/* XXX */
83 #include "opt_sctp.h"
84 #endif
85 
86 #include <sys/param.h>
87 #include <sys/systm.h>
88 #include <sys/proc.h>
89 #include <sys/file.h>
90 #include <sys/filedesc.h>
91 #include <sys/kmem.h>
92 #include <sys/mbuf.h>
93 #include <sys/domain.h>
94 #include <sys/kernel.h>
95 #include <sys/protosw.h>
96 #include <sys/socket.h>
97 #include <sys/socketvar.h>
98 #include <sys/signalvar.h>
99 #include <sys/resourcevar.h>
100 #include <sys/uidinfo.h>
101 #include <sys/event.h>
102 #include <sys/poll.h>
103 #include <sys/kauth.h>
104 #include <sys/mutex.h>
105 #include <sys/condvar.h>
106 #include <sys/kthread.h>
107 
108 #ifdef COMPAT_50
109 #include <compat/sys/time.h>
110 #include <compat/sys/socket.h>
111 #endif
112 
113 #include <uvm/uvm_extern.h>
114 #include <uvm/uvm_loan.h>
115 #include <uvm/uvm_page.h>
116 
117 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
118 
119 extern const struct fileops socketops;
120 
121 extern int	somaxconn;			/* patchable (XXX sysctl) */
122 int		somaxconn = SOMAXCONN;
123 kmutex_t	*softnet_lock;
124 
125 #ifdef SOSEND_COUNTERS
126 #include <sys/device.h>
127 
128 static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
129     NULL, "sosend", "loan big");
130 static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
131     NULL, "sosend", "copy big");
132 static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
133     NULL, "sosend", "copy small");
134 static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
135     NULL, "sosend", "kva limit");
136 
137 #define	SOSEND_COUNTER_INCR(ev)		(ev)->ev_count++
138 
139 EVCNT_ATTACH_STATIC(sosend_loan_big);
140 EVCNT_ATTACH_STATIC(sosend_copy_big);
141 EVCNT_ATTACH_STATIC(sosend_copy_small);
142 EVCNT_ATTACH_STATIC(sosend_kvalimit);
143 #else
144 
145 #define	SOSEND_COUNTER_INCR(ev)		/* nothing */
146 
147 #endif /* SOSEND_COUNTERS */
148 
149 #if defined(SOSEND_NO_LOAN) || defined(MULTIPROCESSOR)
150 int sock_loan_thresh = -1;
151 #else
152 int sock_loan_thresh = 4096;
153 #endif
154 
155 static kmutex_t so_pendfree_lock;
156 static struct mbuf *so_pendfree = NULL;
157 
158 #ifndef SOMAXKVA
159 #define	SOMAXKVA (16 * 1024 * 1024)
160 #endif
161 int somaxkva = SOMAXKVA;
162 static int socurkva;
163 static kcondvar_t socurkva_cv;
164 
165 static kauth_listener_t socket_listener;
166 
167 #define	SOCK_LOAN_CHUNK		65536
168 
169 static void sopendfree_thread(void *);
170 static kcondvar_t pendfree_thread_cv;
171 static lwp_t *sopendfree_lwp;
172 
173 static void sysctl_kern_socket_setup(void);
174 static struct sysctllog *socket_sysctllog;
175 
176 static vsize_t
177 sokvareserve(struct socket *so, vsize_t len)
178 {
179 	int error;
180 
181 	mutex_enter(&so_pendfree_lock);
182 	while (socurkva + len > somaxkva) {
183 		SOSEND_COUNTER_INCR(&sosend_kvalimit);
184 		error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock);
185 		if (error) {
186 			len = 0;
187 			break;
188 		}
189 	}
190 	socurkva += len;
191 	mutex_exit(&so_pendfree_lock);
192 	return len;
193 }
194 
195 static void
196 sokvaunreserve(vsize_t len)
197 {
198 
199 	mutex_enter(&so_pendfree_lock);
200 	socurkva -= len;
201 	cv_broadcast(&socurkva_cv);
202 	mutex_exit(&so_pendfree_lock);
203 }
204 
205 /*
206  * sokvaalloc: allocate kva for loan.
207  */
208 
209 vaddr_t
210 sokvaalloc(vaddr_t sva, vsize_t len, struct socket *so)
211 {
212 	vaddr_t lva;
213 
214 	/*
215 	 * reserve kva.
216 	 */
217 
218 	if (sokvareserve(so, len) == 0)
219 		return 0;
220 
221 	/*
222 	 * allocate kva.
223 	 */
224 
225 	lva = uvm_km_alloc(kernel_map, len, atop(sva) & uvmexp.colormask,
226 	    UVM_KMF_COLORMATCH | UVM_KMF_VAONLY | UVM_KMF_WAITVA);
227 	if (lva == 0) {
228 		sokvaunreserve(len);
229 		return (0);
230 	}
231 
232 	return lva;
233 }
234 
235 /*
236  * sokvafree: free kva for loan.
237  */
238 
239 void
240 sokvafree(vaddr_t sva, vsize_t len)
241 {
242 
243 	/*
244 	 * free kva.
245 	 */
246 
247 	uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY);
248 
249 	/*
250 	 * unreserve kva.
251 	 */
252 
253 	sokvaunreserve(len);
254 }
255 
256 static void
257 sodoloanfree(struct vm_page **pgs, void *buf, size_t size)
258 {
259 	vaddr_t sva, eva;
260 	vsize_t len;
261 	int npgs;
262 
263 	KASSERT(pgs != NULL);
264 
265 	eva = round_page((vaddr_t) buf + size);
266 	sva = trunc_page((vaddr_t) buf);
267 	len = eva - sva;
268 	npgs = len >> PAGE_SHIFT;
269 
270 	pmap_kremove(sva, len);
271 	pmap_update(pmap_kernel());
272 	uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE);
273 	sokvafree(sva, len);
274 }
275 
276 /*
277  * sopendfree_thread: free mbufs on "pendfree" list.
278  * unlock and relock so_pendfree_lock when freeing mbufs.
279  */
280 
281 static void
282 sopendfree_thread(void *v)
283 {
284 	struct mbuf *m, *next;
285 	size_t rv;
286 
287 	mutex_enter(&so_pendfree_lock);
288 
289 	for (;;) {
290 		rv = 0;
291 		while (so_pendfree != NULL) {
292 			m = so_pendfree;
293 			so_pendfree = NULL;
294 			mutex_exit(&so_pendfree_lock);
295 
296 			for (; m != NULL; m = next) {
297 				next = m->m_next;
298 				KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) ==
299 				    0);
300 				KASSERT(m->m_ext.ext_refcnt == 0);
301 
302 				rv += m->m_ext.ext_size;
303 				sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf,
304 				    m->m_ext.ext_size);
305 				pool_cache_put(mb_cache, m);
306 			}
307 
308 			mutex_enter(&so_pendfree_lock);
309 		}
310 		if (rv)
311 			cv_broadcast(&socurkva_cv);
312 		cv_wait(&pendfree_thread_cv, &so_pendfree_lock);
313 	}
314 	panic("sopendfree_thread");
315 	/* NOTREACHED */
316 }
317 
318 void
319 soloanfree(struct mbuf *m, void *buf, size_t size, void *arg)
320 {
321 
322 	KASSERT(m != NULL);
323 
324 	/*
325 	 * postpone freeing mbuf.
326 	 *
327 	 * we can't do it in interrupt context
328 	 * because we need to put kva back to kernel_map.
329 	 */
330 
331 	mutex_enter(&so_pendfree_lock);
332 	m->m_next = so_pendfree;
333 	so_pendfree = m;
334 	cv_signal(&pendfree_thread_cv);
335 	mutex_exit(&so_pendfree_lock);
336 }
337 
338 static long
339 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space)
340 {
341 	struct iovec *iov = uio->uio_iov;
342 	vaddr_t sva, eva;
343 	vsize_t len;
344 	vaddr_t lva;
345 	int npgs, error;
346 	vaddr_t va;
347 	int i;
348 
349 	if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace))
350 		return (0);
351 
352 	if (iov->iov_len < (size_t) space)
353 		space = iov->iov_len;
354 	if (space > SOCK_LOAN_CHUNK)
355 		space = SOCK_LOAN_CHUNK;
356 
357 	eva = round_page((vaddr_t) iov->iov_base + space);
358 	sva = trunc_page((vaddr_t) iov->iov_base);
359 	len = eva - sva;
360 	npgs = len >> PAGE_SHIFT;
361 
362 	KASSERT(npgs <= M_EXT_MAXPAGES);
363 
364 	lva = sokvaalloc(sva, len, so);
365 	if (lva == 0)
366 		return 0;
367 
368 	error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len,
369 	    m->m_ext.ext_pgs, UVM_LOAN_TOPAGE);
370 	if (error) {
371 		sokvafree(lva, len);
372 		return (0);
373 	}
374 
375 	for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE)
376 		pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
377 		    VM_PROT_READ, 0);
378 	pmap_update(pmap_kernel());
379 
380 	lva += (vaddr_t) iov->iov_base & PAGE_MASK;
381 
382 	MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so);
383 	m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
384 
385 	uio->uio_resid -= space;
386 	/* uio_offset not updated, not set/used for write(2) */
387 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space;
388 	uio->uio_iov->iov_len -= space;
389 	if (uio->uio_iov->iov_len == 0) {
390 		uio->uio_iov++;
391 		uio->uio_iovcnt--;
392 	}
393 
394 	return (space);
395 }
396 
397 struct mbuf *
398 getsombuf(struct socket *so, int type)
399 {
400 	struct mbuf *m;
401 
402 	m = m_get(M_WAIT, type);
403 	MCLAIM(m, so->so_mowner);
404 	return m;
405 }
406 
407 static int
408 socket_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
409     void *arg0, void *arg1, void *arg2, void *arg3)
410 {
411 	int result;
412 	enum kauth_network_req req;
413 
414 	result = KAUTH_RESULT_DEFER;
415 	req = (enum kauth_network_req)arg0;
416 
417 	if ((action != KAUTH_NETWORK_SOCKET) &&
418 	    (action != KAUTH_NETWORK_BIND))
419 		return result;
420 
421 	switch (req) {
422 	case KAUTH_REQ_NETWORK_BIND_PORT:
423 		result = KAUTH_RESULT_ALLOW;
424 		break;
425 
426 	case KAUTH_REQ_NETWORK_SOCKET_DROP: {
427 		/* Normal users can only drop their own connections. */
428 		struct socket *so = (struct socket *)arg1;
429 
430 		if (so->so_cred && proc_uidmatch(cred, so->so_cred) == 0)
431 			result = KAUTH_RESULT_ALLOW;
432 
433 		break;
434 		}
435 
436 	case KAUTH_REQ_NETWORK_SOCKET_OPEN:
437 		/* We allow "raw" routing/bluetooth sockets to anyone. */
438 		switch ((u_long)arg1) {
439 		case PF_ROUTE:
440 		case PF_OROUTE:
441 		case PF_BLUETOOTH:
442 		case PF_CAN:
443 			result = KAUTH_RESULT_ALLOW;
444 			break;
445 		default:
446 			/* Privileged, let secmodel handle this. */
447 			if ((u_long)arg2 == SOCK_RAW)
448 				break;
449 			result = KAUTH_RESULT_ALLOW;
450 			break;
451 		}
452 		break;
453 
454 	case KAUTH_REQ_NETWORK_SOCKET_CANSEE:
455 		result = KAUTH_RESULT_ALLOW;
456 
457 		break;
458 
459 	default:
460 		break;
461 	}
462 
463 	return result;
464 }
465 
466 void
467 soinit(void)
468 {
469 
470 	sysctl_kern_socket_setup();
471 
472 	mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM);
473 	softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
474 	cv_init(&socurkva_cv, "sokva");
475 	cv_init(&pendfree_thread_cv, "sopendfr");
476 	soinit2();
477 
478 	/* Set the initial adjusted socket buffer size. */
479 	if (sb_max_set(sb_max))
480 		panic("bad initial sb_max value: %lu", sb_max);
481 
482 	socket_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK,
483 	    socket_listener_cb, NULL);
484 }
485 
486 void
487 soinit1(void)
488 {
489 	int error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL,
490 	    sopendfree_thread, NULL, &sopendfree_lwp, "sopendfree");
491 	if (error)
492 		panic("soinit1 %d", error);
493 }
494 
495 /*
496  * socreate: create a new socket of the specified type and the protocol.
497  *
498  * => Caller may specify another socket for lock sharing (must not be held).
499  * => Returns the new socket without lock held.
500  */
501 int
502 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l,
503 	 struct socket *lockso)
504 {
505 	const struct protosw	*prp;
506 	struct socket	*so;
507 	uid_t		uid;
508 	int		error;
509 	kmutex_t	*lock;
510 
511 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET,
512 	    KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type),
513 	    KAUTH_ARG(proto));
514 	if (error != 0)
515 		return error;
516 
517 	if (proto)
518 		prp = pffindproto(dom, proto, type);
519 	else
520 		prp = pffindtype(dom, type);
521 	if (prp == NULL) {
522 		/* no support for domain */
523 		if (pffinddomain(dom) == 0)
524 			return EAFNOSUPPORT;
525 		/* no support for socket type */
526 		if (proto == 0 && type != 0)
527 			return EPROTOTYPE;
528 		return EPROTONOSUPPORT;
529 	}
530 	if (prp->pr_usrreqs == NULL)
531 		return EPROTONOSUPPORT;
532 	if (prp->pr_type != type)
533 		return EPROTOTYPE;
534 
535 	so = soget(true);
536 	so->so_type = type;
537 	so->so_proto = prp;
538 	so->so_send = sosend;
539 	so->so_receive = soreceive;
540 #ifdef MBUFTRACE
541 	so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner;
542 	so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner;
543 	so->so_mowner = &prp->pr_domain->dom_mowner;
544 #endif
545 	uid = kauth_cred_geteuid(l->l_cred);
546 	so->so_uidinfo = uid_find(uid);
547 	so->so_cpid = l->l_proc->p_pid;
548 
549 	/*
550 	 * Lock assigned and taken during PCB attach, unless we share
551 	 * the lock with another socket, e.g. socketpair(2) case.
552 	 */
553 	if (lockso) {
554 		lock = lockso->so_lock;
555 		so->so_lock = lock;
556 		mutex_obj_hold(lock);
557 		mutex_enter(lock);
558 	}
559 
560 	/* Attach the PCB (returns with the socket lock held). */
561 	error = (*prp->pr_usrreqs->pr_attach)(so, proto);
562 	KASSERT(solocked(so));
563 
564 	if (error) {
565 		KASSERT(so->so_pcb == NULL);
566 		so->so_state |= SS_NOFDREF;
567 		sofree(so);
568 		return error;
569 	}
570 	so->so_cred = kauth_cred_dup(l->l_cred);
571 	sounlock(so);
572 
573 	*aso = so;
574 	return 0;
575 }
576 
577 /*
578  * fsocreate: create a socket and a file descriptor associated with it.
579  *
580  * => On success, write file descriptor to fdout and return zero.
581  * => On failure, return non-zero; *fdout will be undefined.
582  */
583 int
584 fsocreate(int domain, struct socket **sop, int type, int proto, int *fdout)
585 {
586 	lwp_t *l = curlwp;
587 	int error, fd, flags;
588 	struct socket *so;
589 	struct file *fp;
590 
591 	if ((error = fd_allocfile(&fp, &fd)) != 0) {
592 		return error;
593 	}
594 	flags = type & SOCK_FLAGS_MASK;
595 	fd_set_exclose(l, fd, (flags & SOCK_CLOEXEC) != 0);
596 	fp->f_flag = FREAD|FWRITE|((flags & SOCK_NONBLOCK) ? FNONBLOCK : 0)|
597 	    ((flags & SOCK_NOSIGPIPE) ? FNOSIGPIPE : 0);
598 	fp->f_type = DTYPE_SOCKET;
599 	fp->f_ops = &socketops;
600 
601 	type &= ~SOCK_FLAGS_MASK;
602 	error = socreate(domain, &so, type, proto, l, NULL);
603 	if (error) {
604 		fd_abort(curproc, fp, fd);
605 		return error;
606 	}
607 	if (flags & SOCK_NONBLOCK) {
608 		so->so_state |= SS_NBIO;
609 	}
610 	fp->f_socket = so;
611 	fd_affix(curproc, fp, fd);
612 
613 	if (sop != NULL) {
614 		*sop = so;
615 	}
616 	*fdout = fd;
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 sockaddr *nam, struct lwp *l)
635 {
636 	int	error;
637 
638 	solock(so);
639 	if (nam->sa_family != so->so_proto->pr_domain->dom_family) {
640 		sounlock(so);
641 		return EAFNOSUPPORT;
642 	}
643 	error = (*so->so_proto->pr_usrreqs->pr_bind)(so, nam, l);
644 	sounlock(so);
645 	return error;
646 }
647 
648 int
649 solisten(struct socket *so, int backlog, struct lwp *l)
650 {
651 	int	error;
652 	short	oldopt, oldqlimit;
653 
654 	solock(so);
655 	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
656 	    SS_ISDISCONNECTING)) != 0) {
657 		sounlock(so);
658 		return EINVAL;
659 	}
660 	oldopt = so->so_options;
661 	oldqlimit = so->so_qlimit;
662 	if (TAILQ_EMPTY(&so->so_q))
663 		so->so_options |= SO_ACCEPTCONN;
664 	if (backlog < 0)
665 		backlog = 0;
666 	so->so_qlimit = min(backlog, somaxconn);
667 
668 	error = (*so->so_proto->pr_usrreqs->pr_listen)(so, l);
669 	if (error != 0) {
670 		so->so_options = oldopt;
671 		so->so_qlimit = oldqlimit;
672 		sounlock(so);
673 		return error;
674 	}
675 	sounlock(so);
676 	return 0;
677 }
678 
679 void
680 sofree(struct socket *so)
681 {
682 	u_int refs;
683 
684 	KASSERT(solocked(so));
685 
686 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
687 		sounlock(so);
688 		return;
689 	}
690 	if (so->so_head) {
691 		/*
692 		 * We must not decommission a socket that's on the accept(2)
693 		 * queue.  If we do, then accept(2) may hang after select(2)
694 		 * indicated that the listening socket was ready.
695 		 */
696 		if (!soqremque(so, 0)) {
697 			sounlock(so);
698 			return;
699 		}
700 	}
701 	if (so->so_rcv.sb_hiwat)
702 		(void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0,
703 		    RLIM_INFINITY);
704 	if (so->so_snd.sb_hiwat)
705 		(void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0,
706 		    RLIM_INFINITY);
707 	sbrelease(&so->so_snd, so);
708 	KASSERT(!cv_has_waiters(&so->so_cv));
709 	KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv));
710 	KASSERT(!cv_has_waiters(&so->so_snd.sb_cv));
711 	sorflush(so);
712 	refs = so->so_aborting;	/* XXX */
713 	/* Remove acccept filter if one is present. */
714 	if (so->so_accf != NULL)
715 		(void)accept_filt_clear(so);
716 	sounlock(so);
717 	if (refs == 0)		/* XXX */
718 		soput(so);
719 }
720 
721 /*
722  * soclose: close a socket on last file table reference removal.
723  * Initiate disconnect if connected.  Free socket when disconnect complete.
724  */
725 int
726 soclose(struct socket *so)
727 {
728 	struct socket *so2;
729 	int error = 0;
730 
731 	solock(so);
732 	if (so->so_options & SO_ACCEPTCONN) {
733 		for (;;) {
734 			if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) {
735 				KASSERT(solocked2(so, so2));
736 				(void) soqremque(so2, 0);
737 				/* soabort drops the lock. */
738 				(void) soabort(so2);
739 				solock(so);
740 				continue;
741 			}
742 			if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) {
743 				KASSERT(solocked2(so, so2));
744 				(void) soqremque(so2, 1);
745 				/* soabort drops the lock. */
746 				(void) soabort(so2);
747 				solock(so);
748 				continue;
749 			}
750 			break;
751 		}
752 	}
753 	if (so->so_pcb == NULL)
754 		goto discard;
755 	if (so->so_state & SS_ISCONNECTED) {
756 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
757 			error = sodisconnect(so);
758 			if (error)
759 				goto drop;
760 		}
761 		if (so->so_options & SO_LINGER) {
762 			if ((so->so_state & (SS_ISDISCONNECTING|SS_NBIO)) ==
763 			    (SS_ISDISCONNECTING|SS_NBIO))
764 				goto drop;
765 			while (so->so_state & SS_ISCONNECTED) {
766 				error = sowait(so, true, so->so_linger * hz);
767 				if (error)
768 					break;
769 			}
770 		}
771 	}
772  drop:
773 	if (so->so_pcb) {
774 		KASSERT(solocked(so));
775 		(*so->so_proto->pr_usrreqs->pr_detach)(so);
776 	}
777  discard:
778 	KASSERT((so->so_state & SS_NOFDREF) == 0);
779 	kauth_cred_free(so->so_cred);
780 	so->so_state |= SS_NOFDREF;
781 	sofree(so);
782 	return error;
783 }
784 
785 /*
786  * Must be called with the socket locked..  Will return with it unlocked.
787  */
788 int
789 soabort(struct socket *so)
790 {
791 	u_int refs;
792 	int error;
793 
794 	KASSERT(solocked(so));
795 	KASSERT(so->so_head == NULL);
796 
797 	so->so_aborting++;		/* XXX */
798 	error = (*so->so_proto->pr_usrreqs->pr_abort)(so);
799 	refs = --so->so_aborting;	/* XXX */
800 	if (error || (refs == 0)) {
801 		sofree(so);
802 	} else {
803 		sounlock(so);
804 	}
805 	return error;
806 }
807 
808 int
809 soaccept(struct socket *so, struct sockaddr *nam)
810 {
811 	int error;
812 
813 	KASSERT(solocked(so));
814 	KASSERT((so->so_state & SS_NOFDREF) != 0);
815 
816 	so->so_state &= ~SS_NOFDREF;
817 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
818 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
819 		error = (*so->so_proto->pr_usrreqs->pr_accept)(so, nam);
820 	else
821 		error = ECONNABORTED;
822 
823 	return error;
824 }
825 
826 int
827 soconnect(struct socket *so, struct sockaddr *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 		if (nam->sa_family != so->so_proto->pr_domain->dom_family) {
847 			return EAFNOSUPPORT;
848 		}
849 		error = (*so->so_proto->pr_usrreqs->pr_connect)(so, nam, l);
850 	}
851 
852 	return error;
853 }
854 
855 int
856 soconnect2(struct socket *so1, struct socket *so2)
857 {
858 	KASSERT(solocked2(so1, so2));
859 
860 	return (*so1->so_proto->pr_usrreqs->pr_connect2)(so1, so2);
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_usrreqs->pr_disconnect)(so);
876 	}
877 	return (error);
878 }
879 
880 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
881 /*
882  * Send on a socket.
883  * If send must go all at once and message is larger than
884  * send buffering, then hard error.
885  * Lock against other senders.
886  * If must go all at once and not enough room now, then
887  * inform user that this would block and do nothing.
888  * Otherwise, if nonblocking, send as much as possible.
889  * The data to be sent is described by "uio" if nonzero,
890  * otherwise by the mbuf chain "top" (which must be null
891  * if uio is not).  Data provided in mbuf chain must be small
892  * enough to send all at once.
893  *
894  * Returns nonzero on error, timeout or signal; callers
895  * must check for short counts if EINTR/ERESTART are returned.
896  * Data and control buffers are freed on return.
897  */
898 int
899 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
900 	struct mbuf *top, struct mbuf *control, int flags, struct lwp *l)
901 {
902 	struct mbuf	**mp, *m;
903 	long		space, len, resid, clen, mlen;
904 	int		error, s, dontroute, atomic;
905 	short		wakeup_state = 0;
906 
907 	clen = 0;
908 
909 	/*
910 	 * solock() provides atomicity of access.  splsoftnet() prevents
911 	 * protocol processing soft interrupts from interrupting us and
912 	 * blocking (expensive).
913 	 */
914 	s = splsoftnet();
915 	solock(so);
916 	atomic = sosendallatonce(so) || top;
917 	if (uio)
918 		resid = uio->uio_resid;
919 	else
920 		resid = top->m_pkthdr.len;
921 	/*
922 	 * In theory resid should be unsigned.
923 	 * However, space must be signed, as it might be less than 0
924 	 * if we over-committed, and we must use a signed comparison
925 	 * of space and resid.  On the other hand, a negative resid
926 	 * causes us to loop sending 0-length segments to the protocol.
927 	 */
928 	if (resid < 0) {
929 		error = EINVAL;
930 		goto out;
931 	}
932 	dontroute =
933 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
934 	    (so->so_proto->pr_flags & PR_ATOMIC);
935 	l->l_ru.ru_msgsnd++;
936 	if (control)
937 		clen = control->m_len;
938  restart:
939 	if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
940 		goto out;
941 	do {
942 		if (so->so_state & SS_CANTSENDMORE) {
943 			error = EPIPE;
944 			goto release;
945 		}
946 		if (so->so_error) {
947 			error = so->so_error;
948 			so->so_error = 0;
949 			goto release;
950 		}
951 		if ((so->so_state & SS_ISCONNECTED) == 0) {
952 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
953 				if (resid || clen == 0) {
954 					error = ENOTCONN;
955 					goto release;
956 				}
957 			} else if (addr == NULL) {
958 				error = EDESTADDRREQ;
959 				goto release;
960 			}
961 		}
962 		space = sbspace(&so->so_snd);
963 		if (flags & MSG_OOB)
964 			space += 1024;
965 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
966 		    clen > so->so_snd.sb_hiwat) {
967 			error = EMSGSIZE;
968 			goto release;
969 		}
970 		if (space < resid + clen &&
971 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
972 			if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
973 				error = EWOULDBLOCK;
974 				goto release;
975 			}
976 			sbunlock(&so->so_snd);
977 			if (wakeup_state & SS_RESTARTSYS) {
978 				error = ERESTART;
979 				goto out;
980 			}
981 			error = sbwait(&so->so_snd);
982 			if (error)
983 				goto out;
984 			wakeup_state = so->so_state;
985 			goto restart;
986 		}
987 		wakeup_state = 0;
988 		mp = &top;
989 		space -= clen;
990 		do {
991 			if (uio == NULL) {
992 				/*
993 				 * Data is prepackaged in "top".
994 				 */
995 				resid = 0;
996 				if (flags & MSG_EOR)
997 					top->m_flags |= M_EOR;
998 			} else do {
999 				sounlock(so);
1000 				splx(s);
1001 				if (top == NULL) {
1002 					m = m_gethdr(M_WAIT, MT_DATA);
1003 					mlen = MHLEN;
1004 					m->m_pkthdr.len = 0;
1005 					m_reset_rcvif(m);
1006 				} else {
1007 					m = m_get(M_WAIT, MT_DATA);
1008 					mlen = MLEN;
1009 				}
1010 				MCLAIM(m, so->so_snd.sb_mowner);
1011 				if (sock_loan_thresh >= 0 &&
1012 				    uio->uio_iov->iov_len >= sock_loan_thresh &&
1013 				    space >= sock_loan_thresh &&
1014 				    (len = sosend_loan(so, uio, m,
1015 						       space)) != 0) {
1016 					SOSEND_COUNTER_INCR(&sosend_loan_big);
1017 					space -= len;
1018 					goto have_data;
1019 				}
1020 				if (resid >= MINCLSIZE && space >= MCLBYTES) {
1021 					SOSEND_COUNTER_INCR(&sosend_copy_big);
1022 					m_clget(m, M_DONTWAIT);
1023 					if ((m->m_flags & M_EXT) == 0)
1024 						goto nopages;
1025 					mlen = MCLBYTES;
1026 					if (atomic && top == 0) {
1027 						len = lmin(MCLBYTES - max_hdr,
1028 						    resid);
1029 						m->m_data += max_hdr;
1030 					} else
1031 						len = lmin(MCLBYTES, resid);
1032 					space -= len;
1033 				} else {
1034  nopages:
1035 					SOSEND_COUNTER_INCR(&sosend_copy_small);
1036 					len = lmin(lmin(mlen, resid), space);
1037 					space -= len;
1038 					/*
1039 					 * For datagram protocols, leave room
1040 					 * for protocol headers in first mbuf.
1041 					 */
1042 					if (atomic && top == 0 && len < mlen)
1043 						MH_ALIGN(m, len);
1044 				}
1045 				error = uiomove(mtod(m, void *), (int)len, uio);
1046  have_data:
1047 				resid = uio->uio_resid;
1048 				m->m_len = len;
1049 				*mp = m;
1050 				top->m_pkthdr.len += len;
1051 				s = splsoftnet();
1052 				solock(so);
1053 				if (error != 0)
1054 					goto release;
1055 				mp = &m->m_next;
1056 				if (resid <= 0) {
1057 					if (flags & MSG_EOR)
1058 						top->m_flags |= M_EOR;
1059 					break;
1060 				}
1061 			} while (space > 0 && atomic);
1062 
1063 			if (so->so_state & SS_CANTSENDMORE) {
1064 				error = EPIPE;
1065 				goto release;
1066 			}
1067 			if (dontroute)
1068 				so->so_options |= SO_DONTROUTE;
1069 			if (resid > 0)
1070 				so->so_state |= SS_MORETOCOME;
1071 			if (flags & MSG_OOB) {
1072 				error = (*so->so_proto->pr_usrreqs->pr_sendoob)(
1073 				    so, top, control);
1074 			} else {
1075 				error = (*so->so_proto->pr_usrreqs->pr_send)(so,
1076 				    top, addr, control, l);
1077 			}
1078 			if (dontroute)
1079 				so->so_options &= ~SO_DONTROUTE;
1080 			if (resid > 0)
1081 				so->so_state &= ~SS_MORETOCOME;
1082 			clen = 0;
1083 			control = NULL;
1084 			top = NULL;
1085 			mp = &top;
1086 			if (error != 0)
1087 				goto release;
1088 		} while (resid && space > 0);
1089 	} while (resid);
1090 
1091  release:
1092 	sbunlock(&so->so_snd);
1093  out:
1094 	sounlock(so);
1095 	splx(s);
1096 	if (top)
1097 		m_freem(top);
1098 	if (control)
1099 		m_freem(control);
1100 	return (error);
1101 }
1102 
1103 /*
1104  * Following replacement or removal of the first mbuf on the first
1105  * mbuf chain of a socket buffer, push necessary state changes back
1106  * into the socket buffer so that other consumers see the values
1107  * consistently.  'nextrecord' is the callers locally stored value of
1108  * the original value of sb->sb_mb->m_nextpkt which must be restored
1109  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
1110  */
1111 static void
1112 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
1113 {
1114 
1115 	KASSERT(solocked(sb->sb_so));
1116 
1117 	/*
1118 	 * First, update for the new value of nextrecord.  If necessary,
1119 	 * make it the first record.
1120 	 */
1121 	if (sb->sb_mb != NULL)
1122 		sb->sb_mb->m_nextpkt = nextrecord;
1123 	else
1124 		sb->sb_mb = nextrecord;
1125 
1126         /*
1127          * Now update any dependent socket buffer fields to reflect
1128          * the new state.  This is an inline of SB_EMPTY_FIXUP, with
1129          * the addition of a second clause that takes care of the
1130          * case where sb_mb has been updated, but remains the last
1131          * record.
1132          */
1133         if (sb->sb_mb == NULL) {
1134                 sb->sb_mbtail = NULL;
1135                 sb->sb_lastrecord = NULL;
1136         } else if (sb->sb_mb->m_nextpkt == NULL)
1137                 sb->sb_lastrecord = sb->sb_mb;
1138 }
1139 
1140 /*
1141  * Implement receive operations on a socket.
1142  * We depend on the way that records are added to the sockbuf
1143  * by sbappend*.  In particular, each record (mbufs linked through m_next)
1144  * must begin with an address if the protocol so specifies,
1145  * followed by an optional mbuf or mbufs containing ancillary data,
1146  * and then zero or more mbufs of data.
1147  * In order to avoid blocking network interrupts for the entire time here,
1148  * we splx() while doing the actual copy to user space.
1149  * Although the sockbuf is locked, new data may still be appended,
1150  * and thus we must maintain consistency of the sockbuf during that time.
1151  *
1152  * The caller may receive the data as a single mbuf chain by supplying
1153  * an mbuf **mp0 for use in returning the chain.  The uio is then used
1154  * only for the count in uio_resid.
1155  */
1156 int
1157 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
1158 	struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1159 {
1160 	struct lwp *l = curlwp;
1161 	struct mbuf	*m, **mp, *mt;
1162 	size_t len, offset, moff, orig_resid;
1163 	int atomic, flags, error, s, type;
1164 	const struct protosw	*pr;
1165 	struct mbuf	*nextrecord;
1166 	int		mbuf_removed = 0;
1167 	const struct domain *dom;
1168 	short		wakeup_state = 0;
1169 
1170 	pr = so->so_proto;
1171 	atomic = pr->pr_flags & PR_ATOMIC;
1172 	dom = pr->pr_domain;
1173 	mp = mp0;
1174 	type = 0;
1175 	orig_resid = uio->uio_resid;
1176 
1177 	if (paddr != NULL)
1178 		*paddr = NULL;
1179 	if (controlp != NULL)
1180 		*controlp = NULL;
1181 	if (flagsp != NULL)
1182 		flags = *flagsp &~ MSG_EOR;
1183 	else
1184 		flags = 0;
1185 
1186 	if (flags & MSG_OOB) {
1187 		m = m_get(M_WAIT, MT_DATA);
1188 		solock(so);
1189 		error = (*pr->pr_usrreqs->pr_recvoob)(so, m, flags & MSG_PEEK);
1190 		sounlock(so);
1191 		if (error)
1192 			goto bad;
1193 		do {
1194 			error = uiomove(mtod(m, void *),
1195 			    MIN(uio->uio_resid, m->m_len), uio);
1196 			m = m_free(m);
1197 		} while (uio->uio_resid > 0 && error == 0 && m);
1198  bad:
1199 		if (m != NULL)
1200 			m_freem(m);
1201 		return error;
1202 	}
1203 	if (mp != NULL)
1204 		*mp = NULL;
1205 
1206 	/*
1207 	 * solock() provides atomicity of access.  splsoftnet() prevents
1208 	 * protocol processing soft interrupts from interrupting us and
1209 	 * blocking (expensive).
1210 	 */
1211 	s = splsoftnet();
1212 	solock(so);
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 			so->so_error = 0;
1248 			goto release;
1249 		}
1250 		if (so->so_state & SS_CANTRCVMORE) {
1251 			if (m != NULL)
1252 				goto dontblock;
1253 			else
1254 				goto release;
1255 		}
1256 		for (; m != NULL; m = m->m_next)
1257 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1258 				m = so->so_rcv.sb_mb;
1259 				goto dontblock;
1260 			}
1261 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1262 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1263 			error = ENOTCONN;
1264 			goto release;
1265 		}
1266 		if (uio->uio_resid == 0)
1267 			goto release;
1268 		if ((so->so_state & SS_NBIO) ||
1269 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
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 		if (wakeup_state & SS_RESTARTSYS)
1277 			error = ERESTART;
1278 		else
1279 			error = sbwait(&so->so_rcv);
1280 		if (error != 0) {
1281 			sounlock(so);
1282 			splx(s);
1283 			return error;
1284 		}
1285 		wakeup_state = so->so_state;
1286 		goto restart;
1287 	}
1288  dontblock:
1289 	/*
1290 	 * On entry here, m points to the first record of the socket buffer.
1291 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1292 	 * pointer to the next record in the socket buffer.  We must keep the
1293 	 * various socket buffer pointers and local stack versions of the
1294 	 * pointers in sync, pushing out modifications before dropping the
1295 	 * socket lock, and re-reading them when picking it up.
1296 	 *
1297 	 * Otherwise, we will race with the network stack appending new data
1298 	 * or records onto the socket buffer by using inconsistent/stale
1299 	 * versions of the field, possibly resulting in socket buffer
1300 	 * corruption.
1301 	 *
1302 	 * By holding the high-level sblock(), we prevent simultaneous
1303 	 * readers from pulling off the front of the socket buffer.
1304 	 */
1305 	if (l != NULL)
1306 		l->l_ru.ru_msgrcv++;
1307 	KASSERT(m == so->so_rcv.sb_mb);
1308 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
1309 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
1310 	nextrecord = m->m_nextpkt;
1311 	if (pr->pr_flags & PR_ADDR) {
1312 #ifdef DIAGNOSTIC
1313 		if (m->m_type != MT_SONAME)
1314 			panic("receive 1a");
1315 #endif
1316 		orig_resid = 0;
1317 		if (flags & MSG_PEEK) {
1318 			if (paddr)
1319 				*paddr = m_copy(m, 0, m->m_len);
1320 			m = m->m_next;
1321 		} else {
1322 			sbfree(&so->so_rcv, m);
1323 			mbuf_removed = 1;
1324 			if (paddr != NULL) {
1325 				*paddr = m;
1326 				so->so_rcv.sb_mb = m->m_next;
1327 				m->m_next = NULL;
1328 				m = so->so_rcv.sb_mb;
1329 			} else {
1330 				m = so->so_rcv.sb_mb = m_free(m);
1331 			}
1332 			sbsync(&so->so_rcv, nextrecord);
1333 		}
1334 	}
1335 	if (pr->pr_flags & PR_ADDR_OPT) {
1336 		/*
1337 		 * For SCTP we may be getting a
1338 		 * whole message OR a partial delivery.
1339 		 */
1340 		if (m->m_type == MT_SONAME) {
1341 			orig_resid = 0;
1342 			if (flags & MSG_PEEK) {
1343 				if (paddr)
1344 					*paddr = m_copy(m, 0, m->m_len);
1345 				m = m->m_next;
1346 			} else {
1347 				sbfree(&so->so_rcv, m);
1348 				if (paddr) {
1349 					*paddr = m;
1350 					so->so_rcv.sb_mb = m->m_next;
1351 					m->m_next = 0;
1352 					m = so->so_rcv.sb_mb;
1353 				} else {
1354 					m = so->so_rcv.sb_mb = m_free(m);
1355 				}
1356 			}
1357 		}
1358 	}
1359 
1360 	/*
1361 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1362 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1363 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1364 	 * perform externalization (or freeing if controlp == NULL).
1365 	 */
1366 	if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) {
1367 		struct mbuf *cm = NULL, *cmn;
1368 		struct mbuf **cme = &cm;
1369 
1370 		do {
1371 			if (flags & MSG_PEEK) {
1372 				if (controlp != NULL) {
1373 					*controlp = m_copy(m, 0, m->m_len);
1374 					controlp = &(*controlp)->m_next;
1375 				}
1376 				m = m->m_next;
1377 			} else {
1378 				sbfree(&so->so_rcv, m);
1379 				so->so_rcv.sb_mb = m->m_next;
1380 				m->m_next = NULL;
1381 				*cme = m;
1382 				cme = &(*cme)->m_next;
1383 				m = so->so_rcv.sb_mb;
1384 			}
1385 		} while (m != NULL && m->m_type == MT_CONTROL);
1386 		if ((flags & MSG_PEEK) == 0)
1387 			sbsync(&so->so_rcv, nextrecord);
1388 		for (; cm != NULL; cm = cmn) {
1389 			cmn = cm->m_next;
1390 			cm->m_next = NULL;
1391 			type = mtod(cm, struct cmsghdr *)->cmsg_type;
1392 			if (controlp != NULL) {
1393 				if (dom->dom_externalize != NULL &&
1394 				    type == SCM_RIGHTS) {
1395 					sounlock(so);
1396 					splx(s);
1397 					error = (*dom->dom_externalize)(cm, l,
1398 					    (flags & MSG_CMSG_CLOEXEC) ?
1399 					    O_CLOEXEC : 0);
1400 					s = splsoftnet();
1401 					solock(so);
1402 				}
1403 				*controlp = cm;
1404 				while (*controlp != NULL)
1405 					controlp = &(*controlp)->m_next;
1406 			} else {
1407 				/*
1408 				 * Dispose of any SCM_RIGHTS message that went
1409 				 * through the read path rather than recv.
1410 				 */
1411 				if (dom->dom_dispose != NULL &&
1412 				    type == SCM_RIGHTS) {
1413 					sounlock(so);
1414 					(*dom->dom_dispose)(cm);
1415 					solock(so);
1416 				}
1417 				m_freem(cm);
1418 			}
1419 		}
1420 		if (m != NULL)
1421 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1422 		else
1423 			nextrecord = so->so_rcv.sb_mb;
1424 		orig_resid = 0;
1425 	}
1426 
1427 	/* If m is non-NULL, we have some data to read. */
1428 	if (__predict_true(m != NULL)) {
1429 		type = m->m_type;
1430 		if (type == MT_OOBDATA)
1431 			flags |= MSG_OOB;
1432 	}
1433 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1434 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1435 
1436 	moff = 0;
1437 	offset = 0;
1438 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1439 		if (m->m_type == MT_OOBDATA) {
1440 			if (type != MT_OOBDATA)
1441 				break;
1442 		} else if (type == MT_OOBDATA)
1443 			break;
1444 #ifdef DIAGNOSTIC
1445 		else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
1446 			panic("receive 3");
1447 #endif
1448 		so->so_state &= ~SS_RCVATMARK;
1449 		wakeup_state = 0;
1450 		len = uio->uio_resid;
1451 		if (so->so_oobmark && len > so->so_oobmark - offset)
1452 			len = so->so_oobmark - offset;
1453 		if (len > m->m_len - moff)
1454 			len = m->m_len - moff;
1455 		/*
1456 		 * If mp is set, just pass back the mbufs.
1457 		 * Otherwise copy them out via the uio, then free.
1458 		 * Sockbuf must be consistent here (points to current mbuf,
1459 		 * it points to next record) when we drop priority;
1460 		 * we must note any additions to the sockbuf when we
1461 		 * block interrupts again.
1462 		 */
1463 		if (mp == NULL) {
1464 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1465 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1466 			sounlock(so);
1467 			splx(s);
1468 			error = uiomove(mtod(m, char *) + moff, len, uio);
1469 			s = splsoftnet();
1470 			solock(so);
1471 			if (error != 0) {
1472 				/*
1473 				 * If any part of the record has been removed
1474 				 * (such as the MT_SONAME mbuf, which will
1475 				 * happen when PR_ADDR, and thus also
1476 				 * PR_ATOMIC, is set), then drop the entire
1477 				 * record to maintain the atomicity of the
1478 				 * receive operation.
1479 				 *
1480 				 * This avoids a later panic("receive 1a")
1481 				 * when compiled with DIAGNOSTIC.
1482 				 */
1483 				if (m && mbuf_removed && atomic)
1484 					(void) sbdroprecord(&so->so_rcv);
1485 
1486 				goto release;
1487 			}
1488 		} else
1489 			uio->uio_resid -= len;
1490 		if (len == m->m_len - moff) {
1491 			if (m->m_flags & M_EOR)
1492 				flags |= MSG_EOR;
1493 #ifdef SCTP
1494 			if (m->m_flags & M_NOTIFICATION)
1495 				flags |= MSG_NOTIFICATION;
1496 #endif /* SCTP */
1497 			if (flags & MSG_PEEK) {
1498 				m = m->m_next;
1499 				moff = 0;
1500 			} else {
1501 				nextrecord = m->m_nextpkt;
1502 				sbfree(&so->so_rcv, m);
1503 				if (mp) {
1504 					*mp = m;
1505 					mp = &m->m_next;
1506 					so->so_rcv.sb_mb = m = m->m_next;
1507 					*mp = NULL;
1508 				} else {
1509 					m = so->so_rcv.sb_mb = m_free(m);
1510 				}
1511 				/*
1512 				 * If m != NULL, we also know that
1513 				 * so->so_rcv.sb_mb != NULL.
1514 				 */
1515 				KASSERT(so->so_rcv.sb_mb == m);
1516 				if (m) {
1517 					m->m_nextpkt = nextrecord;
1518 					if (nextrecord == NULL)
1519 						so->so_rcv.sb_lastrecord = m;
1520 				} else {
1521 					so->so_rcv.sb_mb = nextrecord;
1522 					SB_EMPTY_FIXUP(&so->so_rcv);
1523 				}
1524 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1525 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1526 			}
1527 		} else if (flags & MSG_PEEK)
1528 			moff += len;
1529 		else {
1530 			if (mp != NULL) {
1531 				mt = m_copym(m, 0, len, M_NOWAIT);
1532 				if (__predict_false(mt == NULL)) {
1533 					sounlock(so);
1534 					mt = m_copym(m, 0, len, M_WAIT);
1535 					solock(so);
1536 				}
1537 				*mp = mt;
1538 			}
1539 			m->m_data += len;
1540 			m->m_len -= len;
1541 			so->so_rcv.sb_cc -= len;
1542 		}
1543 		if (so->so_oobmark) {
1544 			if ((flags & MSG_PEEK) == 0) {
1545 				so->so_oobmark -= len;
1546 				if (so->so_oobmark == 0) {
1547 					so->so_state |= SS_RCVATMARK;
1548 					break;
1549 				}
1550 			} else {
1551 				offset += len;
1552 				if (offset == so->so_oobmark)
1553 					break;
1554 			}
1555 		}
1556 		if (flags & MSG_EOR)
1557 			break;
1558 		/*
1559 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1560 		 * we must not quit until "uio->uio_resid == 0" or an error
1561 		 * termination.  If a signal/timeout occurs, return
1562 		 * with a short count but without error.
1563 		 * Keep sockbuf locked against other readers.
1564 		 */
1565 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1566 		    !sosendallatonce(so) && !nextrecord) {
1567 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1568 				break;
1569 			/*
1570 			 * If we are peeking and the socket receive buffer is
1571 			 * full, stop since we can't get more data to peek at.
1572 			 */
1573 			if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
1574 				break;
1575 			/*
1576 			 * If we've drained the socket buffer, tell the
1577 			 * protocol in case it needs to do something to
1578 			 * get it filled again.
1579 			 */
1580 			if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1581 				(*pr->pr_usrreqs->pr_rcvd)(so, flags, l);
1582 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1583 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1584 			if (wakeup_state & SS_RESTARTSYS)
1585 				error = ERESTART;
1586 			else
1587 				error = sbwait(&so->so_rcv);
1588 			if (error != 0) {
1589 				sbunlock(&so->so_rcv);
1590 				sounlock(so);
1591 				splx(s);
1592 				return 0;
1593 			}
1594 			if ((m = so->so_rcv.sb_mb) != NULL)
1595 				nextrecord = m->m_nextpkt;
1596 			wakeup_state = so->so_state;
1597 		}
1598 	}
1599 
1600 	if (m && atomic) {
1601 		flags |= MSG_TRUNC;
1602 		if ((flags & MSG_PEEK) == 0)
1603 			(void) sbdroprecord(&so->so_rcv);
1604 	}
1605 	if ((flags & MSG_PEEK) == 0) {
1606 		if (m == NULL) {
1607 			/*
1608 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1609 			 * part makes sure sb_lastrecord is up-to-date if
1610 			 * there is still data in the socket buffer.
1611 			 */
1612 			so->so_rcv.sb_mb = nextrecord;
1613 			if (so->so_rcv.sb_mb == NULL) {
1614 				so->so_rcv.sb_mbtail = NULL;
1615 				so->so_rcv.sb_lastrecord = NULL;
1616 			} else if (nextrecord->m_nextpkt == NULL)
1617 				so->so_rcv.sb_lastrecord = nextrecord;
1618 		}
1619 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1620 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1621 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1622 			(*pr->pr_usrreqs->pr_rcvd)(so, flags, l);
1623 	}
1624 	if (orig_resid == uio->uio_resid && orig_resid &&
1625 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1626 		sbunlock(&so->so_rcv);
1627 		goto restart;
1628 	}
1629 
1630 	if (flagsp != NULL)
1631 		*flagsp |= flags;
1632  release:
1633 	sbunlock(&so->so_rcv);
1634 	sounlock(so);
1635 	splx(s);
1636 	return error;
1637 }
1638 
1639 int
1640 soshutdown(struct socket *so, int how)
1641 {
1642 	const struct protosw	*pr;
1643 	int	error;
1644 
1645 	KASSERT(solocked(so));
1646 
1647 	pr = so->so_proto;
1648 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1649 		return (EINVAL);
1650 
1651 	if (how == SHUT_RD || how == SHUT_RDWR) {
1652 		sorflush(so);
1653 		error = 0;
1654 	}
1655 	if (how == SHUT_WR || how == SHUT_RDWR)
1656 		error = (*pr->pr_usrreqs->pr_shutdown)(so);
1657 
1658 	return error;
1659 }
1660 
1661 void
1662 sorestart(struct socket *so)
1663 {
1664 	/*
1665 	 * An application has called close() on an fd on which another
1666 	 * of its threads has called a socket system call.
1667 	 * Mark this and wake everyone up, and code that would block again
1668 	 * instead returns ERESTART.
1669 	 * On system call re-entry the fd is validated and EBADF returned.
1670 	 * Any other fd will block again on the 2nd syscall.
1671 	 */
1672 	solock(so);
1673 	so->so_state |= SS_RESTARTSYS;
1674 	cv_broadcast(&so->so_cv);
1675 	cv_broadcast(&so->so_snd.sb_cv);
1676 	cv_broadcast(&so->so_rcv.sb_cv);
1677 	sounlock(so);
1678 }
1679 
1680 void
1681 sorflush(struct socket *so)
1682 {
1683 	struct sockbuf	*sb, asb;
1684 	const struct protosw	*pr;
1685 
1686 	KASSERT(solocked(so));
1687 
1688 	sb = &so->so_rcv;
1689 	pr = so->so_proto;
1690 	socantrcvmore(so);
1691 	sb->sb_flags |= SB_NOINTR;
1692 	(void )sblock(sb, M_WAITOK);
1693 	sbunlock(sb);
1694 	asb = *sb;
1695 	/*
1696 	 * Clear most of the sockbuf structure, but leave some of the
1697 	 * fields valid.
1698 	 */
1699 	memset(&sb->sb_startzero, 0,
1700 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1701 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) {
1702 		sounlock(so);
1703 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1704 		solock(so);
1705 	}
1706 	sbrelease(&asb, so);
1707 }
1708 
1709 /*
1710  * internal set SOL_SOCKET options
1711  */
1712 static int
1713 sosetopt1(struct socket *so, const struct sockopt *sopt)
1714 {
1715 	int error = EINVAL, opt;
1716 	int optval = 0; /* XXX: gcc */
1717 	struct linger l;
1718 	struct timeval tv;
1719 
1720 	switch ((opt = sopt->sopt_name)) {
1721 
1722 	case SO_ACCEPTFILTER:
1723 		error = accept_filt_setopt(so, sopt);
1724 		KASSERT(solocked(so));
1725 		break;
1726 
1727 	case SO_LINGER:
1728 		error = sockopt_get(sopt, &l, sizeof(l));
1729 		solock(so);
1730 		if (error)
1731 			break;
1732 		if (l.l_linger < 0 || l.l_linger > USHRT_MAX ||
1733 		    l.l_linger > (INT_MAX / hz)) {
1734 			error = EDOM;
1735 			break;
1736 		}
1737 		so->so_linger = l.l_linger;
1738 		if (l.l_onoff)
1739 			so->so_options |= SO_LINGER;
1740 		else
1741 			so->so_options &= ~SO_LINGER;
1742 		break;
1743 
1744 	case SO_DEBUG:
1745 	case SO_KEEPALIVE:
1746 	case SO_DONTROUTE:
1747 	case SO_USELOOPBACK:
1748 	case SO_BROADCAST:
1749 	case SO_REUSEADDR:
1750 	case SO_REUSEPORT:
1751 	case SO_OOBINLINE:
1752 	case SO_TIMESTAMP:
1753 	case SO_NOSIGPIPE:
1754 #ifdef SO_OTIMESTAMP
1755 	case SO_OTIMESTAMP:
1756 #endif
1757 		error = sockopt_getint(sopt, &optval);
1758 		solock(so);
1759 		if (error)
1760 			break;
1761 		if (optval)
1762 			so->so_options |= opt;
1763 		else
1764 			so->so_options &= ~opt;
1765 		break;
1766 
1767 	case SO_SNDBUF:
1768 	case SO_RCVBUF:
1769 	case SO_SNDLOWAT:
1770 	case SO_RCVLOWAT:
1771 		error = sockopt_getint(sopt, &optval);
1772 		solock(so);
1773 		if (error)
1774 			break;
1775 
1776 		/*
1777 		 * Values < 1 make no sense for any of these
1778 		 * options, so disallow them.
1779 		 */
1780 		if (optval < 1) {
1781 			error = EINVAL;
1782 			break;
1783 		}
1784 
1785 		switch (opt) {
1786 		case SO_SNDBUF:
1787 			if (sbreserve(&so->so_snd, (u_long)optval, so) == 0) {
1788 				error = ENOBUFS;
1789 				break;
1790 			}
1791 			so->so_snd.sb_flags &= ~SB_AUTOSIZE;
1792 			break;
1793 
1794 		case SO_RCVBUF:
1795 			if (sbreserve(&so->so_rcv, (u_long)optval, so) == 0) {
1796 				error = ENOBUFS;
1797 				break;
1798 			}
1799 			so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1800 			break;
1801 
1802 		/*
1803 		 * Make sure the low-water is never greater than
1804 		 * the high-water.
1805 		 */
1806 		case SO_SNDLOWAT:
1807 			if (optval > so->so_snd.sb_hiwat)
1808 				optval = so->so_snd.sb_hiwat;
1809 
1810 			so->so_snd.sb_lowat = optval;
1811 			break;
1812 
1813 		case SO_RCVLOWAT:
1814 			if (optval > so->so_rcv.sb_hiwat)
1815 				optval = so->so_rcv.sb_hiwat;
1816 
1817 			so->so_rcv.sb_lowat = optval;
1818 			break;
1819 		}
1820 		break;
1821 
1822 #ifdef COMPAT_50
1823 	case SO_OSNDTIMEO:
1824 	case SO_ORCVTIMEO: {
1825 		struct timeval50 otv;
1826 		error = sockopt_get(sopt, &otv, sizeof(otv));
1827 		if (error) {
1828 			solock(so);
1829 			break;
1830 		}
1831 		timeval50_to_timeval(&otv, &tv);
1832 		opt = opt == SO_OSNDTIMEO ? SO_SNDTIMEO : SO_RCVTIMEO;
1833 		error = 0;
1834 		/*FALLTHROUGH*/
1835 	}
1836 #endif /* COMPAT_50 */
1837 
1838 	case SO_SNDTIMEO:
1839 	case SO_RCVTIMEO:
1840 		if (error)
1841 			error = sockopt_get(sopt, &tv, sizeof(tv));
1842 		solock(so);
1843 		if (error)
1844 			break;
1845 
1846 		if (tv.tv_sec > (INT_MAX - tv.tv_usec / tick) / hz) {
1847 			error = EDOM;
1848 			break;
1849 		}
1850 
1851 		optval = tv.tv_sec * hz + tv.tv_usec / tick;
1852 		if (optval == 0 && tv.tv_usec != 0)
1853 			optval = 1;
1854 
1855 		switch (opt) {
1856 		case SO_SNDTIMEO:
1857 			so->so_snd.sb_timeo = optval;
1858 			break;
1859 		case SO_RCVTIMEO:
1860 			so->so_rcv.sb_timeo = optval;
1861 			break;
1862 		}
1863 		break;
1864 
1865 	default:
1866 		solock(so);
1867 		error = ENOPROTOOPT;
1868 		break;
1869 	}
1870 	KASSERT(solocked(so));
1871 	return error;
1872 }
1873 
1874 int
1875 sosetopt(struct socket *so, struct sockopt *sopt)
1876 {
1877 	int error, prerr;
1878 
1879 	if (sopt->sopt_level == SOL_SOCKET) {
1880 		error = sosetopt1(so, sopt);
1881 		KASSERT(solocked(so));
1882 	} else {
1883 		error = ENOPROTOOPT;
1884 		solock(so);
1885 	}
1886 
1887 	if ((error == 0 || error == ENOPROTOOPT) &&
1888 	    so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) {
1889 		/* give the protocol stack a shot */
1890 		prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, sopt);
1891 		if (prerr == 0)
1892 			error = 0;
1893 		else if (prerr != ENOPROTOOPT)
1894 			error = prerr;
1895 	}
1896 	sounlock(so);
1897 	return error;
1898 }
1899 
1900 /*
1901  * so_setsockopt() is a wrapper providing a sockopt structure for sosetopt()
1902  */
1903 int
1904 so_setsockopt(struct lwp *l, struct socket *so, int level, int name,
1905     const void *val, size_t valsize)
1906 {
1907 	struct sockopt sopt;
1908 	int error;
1909 
1910 	KASSERT(valsize == 0 || val != NULL);
1911 
1912 	sockopt_init(&sopt, level, name, valsize);
1913 	sockopt_set(&sopt, val, valsize);
1914 
1915 	error = sosetopt(so, &sopt);
1916 
1917 	sockopt_destroy(&sopt);
1918 
1919 	return error;
1920 }
1921 
1922 /*
1923  * internal get SOL_SOCKET options
1924  */
1925 static int
1926 sogetopt1(struct socket *so, struct sockopt *sopt)
1927 {
1928 	int error, optval, opt;
1929 	struct linger l;
1930 	struct timeval tv;
1931 
1932 	switch ((opt = sopt->sopt_name)) {
1933 
1934 	case SO_ACCEPTFILTER:
1935 		error = accept_filt_getopt(so, sopt);
1936 		break;
1937 
1938 	case SO_LINGER:
1939 		l.l_onoff = (so->so_options & SO_LINGER) ? 1 : 0;
1940 		l.l_linger = so->so_linger;
1941 
1942 		error = sockopt_set(sopt, &l, sizeof(l));
1943 		break;
1944 
1945 	case SO_USELOOPBACK:
1946 	case SO_DONTROUTE:
1947 	case SO_DEBUG:
1948 	case SO_KEEPALIVE:
1949 	case SO_REUSEADDR:
1950 	case SO_REUSEPORT:
1951 	case SO_BROADCAST:
1952 	case SO_OOBINLINE:
1953 	case SO_TIMESTAMP:
1954 	case SO_NOSIGPIPE:
1955 #ifdef SO_OTIMESTAMP
1956 	case SO_OTIMESTAMP:
1957 #endif
1958 	case SO_ACCEPTCONN:
1959 		error = sockopt_setint(sopt, (so->so_options & opt) ? 1 : 0);
1960 		break;
1961 
1962 	case SO_TYPE:
1963 		error = sockopt_setint(sopt, so->so_type);
1964 		break;
1965 
1966 	case SO_ERROR:
1967 		error = sockopt_setint(sopt, so->so_error);
1968 		so->so_error = 0;
1969 		break;
1970 
1971 	case SO_SNDBUF:
1972 		error = sockopt_setint(sopt, so->so_snd.sb_hiwat);
1973 		break;
1974 
1975 	case SO_RCVBUF:
1976 		error = sockopt_setint(sopt, so->so_rcv.sb_hiwat);
1977 		break;
1978 
1979 	case SO_SNDLOWAT:
1980 		error = sockopt_setint(sopt, so->so_snd.sb_lowat);
1981 		break;
1982 
1983 	case SO_RCVLOWAT:
1984 		error = sockopt_setint(sopt, so->so_rcv.sb_lowat);
1985 		break;
1986 
1987 #ifdef COMPAT_50
1988 	case SO_OSNDTIMEO:
1989 	case SO_ORCVTIMEO: {
1990 		struct timeval50 otv;
1991 
1992 		optval = (opt == SO_OSNDTIMEO ?
1993 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1994 
1995 		otv.tv_sec = optval / hz;
1996 		otv.tv_usec = (optval % hz) * tick;
1997 
1998 		error = sockopt_set(sopt, &otv, sizeof(otv));
1999 		break;
2000 	}
2001 #endif /* COMPAT_50 */
2002 
2003 	case SO_SNDTIMEO:
2004 	case SO_RCVTIMEO:
2005 		optval = (opt == SO_SNDTIMEO ?
2006 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2007 
2008 		tv.tv_sec = optval / hz;
2009 		tv.tv_usec = (optval % hz) * tick;
2010 
2011 		error = sockopt_set(sopt, &tv, sizeof(tv));
2012 		break;
2013 
2014 	case SO_OVERFLOWED:
2015 		error = sockopt_setint(sopt, so->so_rcv.sb_overflowed);
2016 		break;
2017 
2018 	default:
2019 		error = ENOPROTOOPT;
2020 		break;
2021 	}
2022 
2023 	return (error);
2024 }
2025 
2026 int
2027 sogetopt(struct socket *so, struct sockopt *sopt)
2028 {
2029 	int		error;
2030 
2031 	solock(so);
2032 	if (sopt->sopt_level != SOL_SOCKET) {
2033 		if (so->so_proto && so->so_proto->pr_ctloutput) {
2034 			error = ((*so->so_proto->pr_ctloutput)
2035 			    (PRCO_GETOPT, so, sopt));
2036 		} else
2037 			error = (ENOPROTOOPT);
2038 	} else {
2039 		error = sogetopt1(so, sopt);
2040 	}
2041 	sounlock(so);
2042 	return (error);
2043 }
2044 
2045 /*
2046  * alloc sockopt data buffer buffer
2047  *	- will be released at destroy
2048  */
2049 static int
2050 sockopt_alloc(struct sockopt *sopt, size_t len, km_flag_t kmflag)
2051 {
2052 
2053 	KASSERT(sopt->sopt_size == 0);
2054 
2055 	if (len > sizeof(sopt->sopt_buf)) {
2056 		sopt->sopt_data = kmem_zalloc(len, kmflag);
2057 		if (sopt->sopt_data == NULL)
2058 			return ENOMEM;
2059 	} else
2060 		sopt->sopt_data = sopt->sopt_buf;
2061 
2062 	sopt->sopt_size = len;
2063 	return 0;
2064 }
2065 
2066 /*
2067  * initialise sockopt storage
2068  *	- MAY sleep during allocation
2069  */
2070 void
2071 sockopt_init(struct sockopt *sopt, int level, int name, size_t size)
2072 {
2073 
2074 	memset(sopt, 0, sizeof(*sopt));
2075 
2076 	sopt->sopt_level = level;
2077 	sopt->sopt_name = name;
2078 	(void)sockopt_alloc(sopt, size, KM_SLEEP);
2079 }
2080 
2081 /*
2082  * destroy sockopt storage
2083  *	- will release any held memory references
2084  */
2085 void
2086 sockopt_destroy(struct sockopt *sopt)
2087 {
2088 
2089 	if (sopt->sopt_data != sopt->sopt_buf)
2090 		kmem_free(sopt->sopt_data, sopt->sopt_size);
2091 
2092 	memset(sopt, 0, sizeof(*sopt));
2093 }
2094 
2095 /*
2096  * set sockopt value
2097  *	- value is copied into sockopt
2098  *	- memory is allocated when necessary, will not sleep
2099  */
2100 int
2101 sockopt_set(struct sockopt *sopt, const void *buf, size_t len)
2102 {
2103 	int error;
2104 
2105 	if (sopt->sopt_size == 0) {
2106 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2107 		if (error)
2108 			return error;
2109 	}
2110 
2111 	if (sopt->sopt_size < len)
2112 		return EINVAL;
2113 
2114 	memcpy(sopt->sopt_data, buf, len);
2115 	sopt->sopt_retsize = len;
2116 
2117 	return 0;
2118 }
2119 
2120 /*
2121  * common case of set sockopt integer value
2122  */
2123 int
2124 sockopt_setint(struct sockopt *sopt, int val)
2125 {
2126 
2127 	return sockopt_set(sopt, &val, sizeof(int));
2128 }
2129 
2130 /*
2131  * get sockopt value
2132  *	- correct size must be given
2133  */
2134 int
2135 sockopt_get(const struct sockopt *sopt, void *buf, size_t len)
2136 {
2137 
2138 	if (sopt->sopt_size != len)
2139 		return EINVAL;
2140 
2141 	memcpy(buf, sopt->sopt_data, len);
2142 	return 0;
2143 }
2144 
2145 /*
2146  * common case of get sockopt integer value
2147  */
2148 int
2149 sockopt_getint(const struct sockopt *sopt, int *valp)
2150 {
2151 
2152 	return sockopt_get(sopt, valp, sizeof(int));
2153 }
2154 
2155 /*
2156  * set sockopt value from mbuf
2157  *	- ONLY for legacy code
2158  *	- mbuf is released by sockopt
2159  *	- will not sleep
2160  */
2161 int
2162 sockopt_setmbuf(struct sockopt *sopt, struct mbuf *m)
2163 {
2164 	size_t len;
2165 	int error;
2166 
2167 	len = m_length(m);
2168 
2169 	if (sopt->sopt_size == 0) {
2170 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2171 		if (error)
2172 			return error;
2173 	}
2174 
2175 	if (sopt->sopt_size < len)
2176 		return EINVAL;
2177 
2178 	m_copydata(m, 0, len, sopt->sopt_data);
2179 	m_freem(m);
2180 	sopt->sopt_retsize = len;
2181 
2182 	return 0;
2183 }
2184 
2185 /*
2186  * get sockopt value into mbuf
2187  *	- ONLY for legacy code
2188  *	- mbuf to be released by the caller
2189  *	- will not sleep
2190  */
2191 struct mbuf *
2192 sockopt_getmbuf(const struct sockopt *sopt)
2193 {
2194 	struct mbuf *m;
2195 
2196 	if (sopt->sopt_size > MCLBYTES)
2197 		return NULL;
2198 
2199 	m = m_get(M_DONTWAIT, MT_SOOPTS);
2200 	if (m == NULL)
2201 		return NULL;
2202 
2203 	if (sopt->sopt_size > MLEN) {
2204 		MCLGET(m, M_DONTWAIT);
2205 		if ((m->m_flags & M_EXT) == 0) {
2206 			m_free(m);
2207 			return NULL;
2208 		}
2209 	}
2210 
2211 	memcpy(mtod(m, void *), sopt->sopt_data, sopt->sopt_size);
2212 	m->m_len = sopt->sopt_size;
2213 
2214 	return m;
2215 }
2216 
2217 void
2218 sohasoutofband(struct socket *so)
2219 {
2220 
2221 	fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
2222 	selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, NOTE_SUBMIT);
2223 }
2224 
2225 static void
2226 filt_sordetach(struct knote *kn)
2227 {
2228 	struct socket	*so;
2229 
2230 	so = ((file_t *)kn->kn_obj)->f_socket;
2231 	solock(so);
2232 	SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
2233 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
2234 		so->so_rcv.sb_flags &= ~SB_KNOTE;
2235 	sounlock(so);
2236 }
2237 
2238 /*ARGSUSED*/
2239 static int
2240 filt_soread(struct knote *kn, long hint)
2241 {
2242 	struct socket	*so;
2243 	int rv;
2244 
2245 	so = ((file_t *)kn->kn_obj)->f_socket;
2246 	if (hint != NOTE_SUBMIT)
2247 		solock(so);
2248 	kn->kn_data = so->so_rcv.sb_cc;
2249 	if (so->so_state & SS_CANTRCVMORE) {
2250 		kn->kn_flags |= EV_EOF;
2251 		kn->kn_fflags = so->so_error;
2252 		rv = 1;
2253 	} else if (so->so_error)
2254 		rv = 1;
2255 	else if (kn->kn_sfflags & NOTE_LOWAT)
2256 		rv = (kn->kn_data >= kn->kn_sdata);
2257 	else
2258 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
2259 	if (hint != NOTE_SUBMIT)
2260 		sounlock(so);
2261 	return rv;
2262 }
2263 
2264 static void
2265 filt_sowdetach(struct knote *kn)
2266 {
2267 	struct socket	*so;
2268 
2269 	so = ((file_t *)kn->kn_obj)->f_socket;
2270 	solock(so);
2271 	SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
2272 	if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
2273 		so->so_snd.sb_flags &= ~SB_KNOTE;
2274 	sounlock(so);
2275 }
2276 
2277 /*ARGSUSED*/
2278 static int
2279 filt_sowrite(struct knote *kn, long hint)
2280 {
2281 	struct socket	*so;
2282 	int rv;
2283 
2284 	so = ((file_t *)kn->kn_obj)->f_socket;
2285 	if (hint != NOTE_SUBMIT)
2286 		solock(so);
2287 	kn->kn_data = sbspace(&so->so_snd);
2288 	if (so->so_state & SS_CANTSENDMORE) {
2289 		kn->kn_flags |= EV_EOF;
2290 		kn->kn_fflags = so->so_error;
2291 		rv = 1;
2292 	} else if (so->so_error)
2293 		rv = 1;
2294 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2295 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
2296 		rv = 0;
2297 	else if (kn->kn_sfflags & NOTE_LOWAT)
2298 		rv = (kn->kn_data >= kn->kn_sdata);
2299 	else
2300 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
2301 	if (hint != NOTE_SUBMIT)
2302 		sounlock(so);
2303 	return rv;
2304 }
2305 
2306 /*ARGSUSED*/
2307 static int
2308 filt_solisten(struct knote *kn, long hint)
2309 {
2310 	struct socket	*so;
2311 	int rv;
2312 
2313 	so = ((file_t *)kn->kn_obj)->f_socket;
2314 
2315 	/*
2316 	 * Set kn_data to number of incoming connections, not
2317 	 * counting partial (incomplete) connections.
2318 	 */
2319 	if (hint != NOTE_SUBMIT)
2320 		solock(so);
2321 	kn->kn_data = so->so_qlen;
2322 	rv = (kn->kn_data > 0);
2323 	if (hint != NOTE_SUBMIT)
2324 		sounlock(so);
2325 	return rv;
2326 }
2327 
2328 static const struct filterops solisten_filtops = {
2329 	.f_isfd = 1,
2330 	.f_attach = NULL,
2331 	.f_detach = filt_sordetach,
2332 	.f_event = filt_solisten,
2333 };
2334 
2335 static const struct filterops soread_filtops = {
2336 	.f_isfd = 1,
2337 	.f_attach = NULL,
2338 	.f_detach = filt_sordetach,
2339 	.f_event = filt_soread,
2340 };
2341 
2342 static const struct filterops sowrite_filtops = {
2343 	.f_isfd = 1,
2344 	.f_attach = NULL,
2345 	.f_detach = filt_sowdetach,
2346 	.f_event = filt_sowrite,
2347 };
2348 
2349 int
2350 soo_kqfilter(struct file *fp, struct knote *kn)
2351 {
2352 	struct socket	*so;
2353 	struct sockbuf	*sb;
2354 
2355 	so = ((file_t *)kn->kn_obj)->f_socket;
2356 	solock(so);
2357 	switch (kn->kn_filter) {
2358 	case EVFILT_READ:
2359 		if (so->so_options & SO_ACCEPTCONN)
2360 			kn->kn_fop = &solisten_filtops;
2361 		else
2362 			kn->kn_fop = &soread_filtops;
2363 		sb = &so->so_rcv;
2364 		break;
2365 	case EVFILT_WRITE:
2366 		kn->kn_fop = &sowrite_filtops;
2367 		sb = &so->so_snd;
2368 		break;
2369 	default:
2370 		sounlock(so);
2371 		return (EINVAL);
2372 	}
2373 	SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
2374 	sb->sb_flags |= SB_KNOTE;
2375 	sounlock(so);
2376 	return (0);
2377 }
2378 
2379 static int
2380 sodopoll(struct socket *so, int events)
2381 {
2382 	int revents;
2383 
2384 	revents = 0;
2385 
2386 	if (events & (POLLIN | POLLRDNORM))
2387 		if (soreadable(so))
2388 			revents |= events & (POLLIN | POLLRDNORM);
2389 
2390 	if (events & (POLLOUT | POLLWRNORM))
2391 		if (sowritable(so))
2392 			revents |= events & (POLLOUT | POLLWRNORM);
2393 
2394 	if (events & (POLLPRI | POLLRDBAND))
2395 		if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
2396 			revents |= events & (POLLPRI | POLLRDBAND);
2397 
2398 	return revents;
2399 }
2400 
2401 int
2402 sopoll(struct socket *so, int events)
2403 {
2404 	int revents = 0;
2405 
2406 #ifndef DIAGNOSTIC
2407 	/*
2408 	 * Do a quick, unlocked check in expectation that the socket
2409 	 * will be ready for I/O.  Don't do this check if DIAGNOSTIC,
2410 	 * as the solocked() assertions will fail.
2411 	 */
2412 	if ((revents = sodopoll(so, events)) != 0)
2413 		return revents;
2414 #endif
2415 
2416 	solock(so);
2417 	if ((revents = sodopoll(so, events)) == 0) {
2418 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2419 			selrecord(curlwp, &so->so_rcv.sb_sel);
2420 			so->so_rcv.sb_flags |= SB_NOTIFY;
2421 		}
2422 
2423 		if (events & (POLLOUT | POLLWRNORM)) {
2424 			selrecord(curlwp, &so->so_snd.sb_sel);
2425 			so->so_snd.sb_flags |= SB_NOTIFY;
2426 		}
2427 	}
2428 	sounlock(so);
2429 
2430 	return revents;
2431 }
2432 
2433 struct mbuf **
2434 sbsavetimestamp(int opt, struct mbuf *m, struct mbuf **mp)
2435 {
2436 	struct timeval tv;
2437 	microtime(&tv);
2438 
2439 #ifdef SO_OTIMESTAMP
2440 	if (opt & SO_OTIMESTAMP) {
2441 		struct timeval50 tv50;
2442 
2443 		timeval_to_timeval50(&tv, &tv50);
2444 		*mp = sbcreatecontrol(&tv50, sizeof(tv50),
2445 		    SCM_OTIMESTAMP, SOL_SOCKET);
2446 		if (*mp)
2447 			mp = &(*mp)->m_next;
2448 	} else
2449 #endif
2450 
2451 	if (opt & SO_TIMESTAMP) {
2452 		*mp = sbcreatecontrol(&tv, sizeof(tv),
2453 		    SCM_TIMESTAMP, SOL_SOCKET);
2454 		if (*mp)
2455 			mp = &(*mp)->m_next;
2456 	}
2457 	return mp;
2458 }
2459 
2460 
2461 #include <sys/sysctl.h>
2462 
2463 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO);
2464 static int sysctl_kern_sbmax(SYSCTLFN_PROTO);
2465 
2466 /*
2467  * sysctl helper routine for kern.somaxkva.  ensures that the given
2468  * value is not too small.
2469  * (XXX should we maybe make sure it's not too large as well?)
2470  */
2471 static int
2472 sysctl_kern_somaxkva(SYSCTLFN_ARGS)
2473 {
2474 	int error, new_somaxkva;
2475 	struct sysctlnode node;
2476 
2477 	new_somaxkva = somaxkva;
2478 	node = *rnode;
2479 	node.sysctl_data = &new_somaxkva;
2480 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2481 	if (error || newp == NULL)
2482 		return (error);
2483 
2484 	if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */
2485 		return (EINVAL);
2486 
2487 	mutex_enter(&so_pendfree_lock);
2488 	somaxkva = new_somaxkva;
2489 	cv_broadcast(&socurkva_cv);
2490 	mutex_exit(&so_pendfree_lock);
2491 
2492 	return (error);
2493 }
2494 
2495 /*
2496  * sysctl helper routine for kern.sbmax. Basically just ensures that
2497  * any new value is not too small.
2498  */
2499 static int
2500 sysctl_kern_sbmax(SYSCTLFN_ARGS)
2501 {
2502 	int error, new_sbmax;
2503 	struct sysctlnode node;
2504 
2505 	new_sbmax = sb_max;
2506 	node = *rnode;
2507 	node.sysctl_data = &new_sbmax;
2508 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2509 	if (error || newp == NULL)
2510 		return (error);
2511 
2512 	KERNEL_LOCK(1, NULL);
2513 	error = sb_max_set(new_sbmax);
2514 	KERNEL_UNLOCK_ONE(NULL);
2515 
2516 	return (error);
2517 }
2518 
2519 static void
2520 sysctl_kern_socket_setup(void)
2521 {
2522 
2523 	KASSERT(socket_sysctllog == NULL);
2524 
2525 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2526 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2527 		       CTLTYPE_INT, "somaxkva",
2528 		       SYSCTL_DESCR("Maximum amount of kernel memory to be "
2529 				    "used for socket buffers"),
2530 		       sysctl_kern_somaxkva, 0, NULL, 0,
2531 		       CTL_KERN, KERN_SOMAXKVA, CTL_EOL);
2532 
2533 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2534 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2535 		       CTLTYPE_INT, "sbmax",
2536 		       SYSCTL_DESCR("Maximum socket buffer size"),
2537 		       sysctl_kern_sbmax, 0, NULL, 0,
2538 		       CTL_KERN, KERN_SBMAX, CTL_EOL);
2539 }
2540