xref: /netbsd-src/sys/kern/uipc_socket.c (revision aad9773e38ed2370a628a6416e098f9008fc10a7)
1 /*	$NetBSD: uipc_socket.c,v 1.235 2014/09/05 09:20:59 matt 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 1.235 2014/09/05 09:20:59 matt Exp $");
75 
76 #include "opt_compat_netbsd.h"
77 #include "opt_sock_counters.h"
78 #include "opt_sosend_loan.h"
79 #include "opt_mbuftrace.h"
80 #include "opt_somaxkva.h"
81 #include "opt_multiprocessor.h"	/* XXX */
82 
83 #include <sys/param.h>
84 #include <sys/systm.h>
85 #include <sys/proc.h>
86 #include <sys/file.h>
87 #include <sys/filedesc.h>
88 #include <sys/kmem.h>
89 #include <sys/mbuf.h>
90 #include <sys/domain.h>
91 #include <sys/kernel.h>
92 #include <sys/protosw.h>
93 #include <sys/socket.h>
94 #include <sys/socketvar.h>
95 #include <sys/signalvar.h>
96 #include <sys/resourcevar.h>
97 #include <sys/uidinfo.h>
98 #include <sys/event.h>
99 #include <sys/poll.h>
100 #include <sys/kauth.h>
101 #include <sys/mutex.h>
102 #include <sys/condvar.h>
103 #include <sys/kthread.h>
104 
105 #ifdef COMPAT_50
106 #include <compat/sys/time.h>
107 #include <compat/sys/socket.h>
108 #endif
109 
110 #include <uvm/uvm_extern.h>
111 #include <uvm/uvm_loan.h>
112 #include <uvm/uvm_page.h>
113 
114 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
115 
116 extern const struct fileops socketops;
117 
118 extern int	somaxconn;			/* patchable (XXX sysctl) */
119 int		somaxconn = SOMAXCONN;
120 kmutex_t	*softnet_lock;
121 
122 #ifdef SOSEND_COUNTERS
123 #include <sys/device.h>
124 
125 static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
126     NULL, "sosend", "loan big");
127 static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
128     NULL, "sosend", "copy big");
129 static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
130     NULL, "sosend", "copy small");
131 static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
132     NULL, "sosend", "kva limit");
133 
134 #define	SOSEND_COUNTER_INCR(ev)		(ev)->ev_count++
135 
136 EVCNT_ATTACH_STATIC(sosend_loan_big);
137 EVCNT_ATTACH_STATIC(sosend_copy_big);
138 EVCNT_ATTACH_STATIC(sosend_copy_small);
139 EVCNT_ATTACH_STATIC(sosend_kvalimit);
140 #else
141 
142 #define	SOSEND_COUNTER_INCR(ev)		/* nothing */
143 
144 #endif /* SOSEND_COUNTERS */
145 
146 #if defined(SOSEND_NO_LOAN) || defined(MULTIPROCESSOR)
147 int sock_loan_thresh = -1;
148 #else
149 int sock_loan_thresh = 4096;
150 #endif
151 
152 static kmutex_t so_pendfree_lock;
153 static struct mbuf *so_pendfree = NULL;
154 
155 #ifndef SOMAXKVA
156 #define	SOMAXKVA (16 * 1024 * 1024)
157 #endif
158 int somaxkva = SOMAXKVA;
159 static int socurkva;
160 static kcondvar_t socurkva_cv;
161 
162 static kauth_listener_t socket_listener;
163 
164 #define	SOCK_LOAN_CHUNK		65536
165 
166 static void sopendfree_thread(void *);
167 static kcondvar_t pendfree_thread_cv;
168 static lwp_t *sopendfree_lwp;
169 
170 static void sysctl_kern_socket_setup(void);
171 static struct sysctllog *socket_sysctllog;
172 
173 static vsize_t
174 sokvareserve(struct socket *so, vsize_t len)
175 {
176 	int error;
177 
178 	mutex_enter(&so_pendfree_lock);
179 	while (socurkva + len > somaxkva) {
180 		SOSEND_COUNTER_INCR(&sosend_kvalimit);
181 		error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock);
182 		if (error) {
183 			len = 0;
184 			break;
185 		}
186 	}
187 	socurkva += len;
188 	mutex_exit(&so_pendfree_lock);
189 	return len;
190 }
191 
192 static void
193 sokvaunreserve(vsize_t len)
194 {
195 
196 	mutex_enter(&so_pendfree_lock);
197 	socurkva -= len;
198 	cv_broadcast(&socurkva_cv);
199 	mutex_exit(&so_pendfree_lock);
200 }
201 
202 /*
203  * sokvaalloc: allocate kva for loan.
204  */
205 
206 vaddr_t
207 sokvaalloc(vaddr_t sva, vsize_t len, struct socket *so)
208 {
209 	vaddr_t lva;
210 
211 	/*
212 	 * reserve kva.
213 	 */
214 
215 	if (sokvareserve(so, len) == 0)
216 		return 0;
217 
218 	/*
219 	 * allocate kva.
220 	 */
221 
222 	lva = uvm_km_alloc(kernel_map, len, atop(sva) & uvmexp.colormask,
223 	    UVM_KMF_COLORMATCH | UVM_KMF_VAONLY | UVM_KMF_WAITVA);
224 	if (lva == 0) {
225 		sokvaunreserve(len);
226 		return (0);
227 	}
228 
229 	return lva;
230 }
231 
232 /*
233  * sokvafree: free kva for loan.
234  */
235 
236 void
237 sokvafree(vaddr_t sva, vsize_t len)
238 {
239 
240 	/*
241 	 * free kva.
242 	 */
243 
244 	uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY);
245 
246 	/*
247 	 * unreserve kva.
248 	 */
249 
250 	sokvaunreserve(len);
251 }
252 
253 static void
254 sodoloanfree(struct vm_page **pgs, void *buf, size_t size)
255 {
256 	vaddr_t sva, eva;
257 	vsize_t len;
258 	int npgs;
259 
260 	KASSERT(pgs != NULL);
261 
262 	eva = round_page((vaddr_t) buf + size);
263 	sva = trunc_page((vaddr_t) buf);
264 	len = eva - sva;
265 	npgs = len >> PAGE_SHIFT;
266 
267 	pmap_kremove(sva, len);
268 	pmap_update(pmap_kernel());
269 	uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE);
270 	sokvafree(sva, len);
271 }
272 
273 /*
274  * sopendfree_thread: free mbufs on "pendfree" list.
275  * unlock and relock so_pendfree_lock when freeing mbufs.
276  */
277 
278 static void
279 sopendfree_thread(void *v)
280 {
281 	struct mbuf *m, *next;
282 	size_t rv;
283 
284 	mutex_enter(&so_pendfree_lock);
285 
286 	for (;;) {
287 		rv = 0;
288 		while (so_pendfree != NULL) {
289 			m = so_pendfree;
290 			so_pendfree = NULL;
291 			mutex_exit(&so_pendfree_lock);
292 
293 			for (; m != NULL; m = next) {
294 				next = m->m_next;
295 				KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0);
296 				KASSERT(m->m_ext.ext_refcnt == 0);
297 
298 				rv += m->m_ext.ext_size;
299 				sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf,
300 				    m->m_ext.ext_size);
301 				pool_cache_put(mb_cache, m);
302 			}
303 
304 			mutex_enter(&so_pendfree_lock);
305 		}
306 		if (rv)
307 			cv_broadcast(&socurkva_cv);
308 		cv_wait(&pendfree_thread_cv, &so_pendfree_lock);
309 	}
310 	panic("sopendfree_thread");
311 	/* NOTREACHED */
312 }
313 
314 void
315 soloanfree(struct mbuf *m, void *buf, size_t size, void *arg)
316 {
317 
318 	KASSERT(m != NULL);
319 
320 	/*
321 	 * postpone freeing mbuf.
322 	 *
323 	 * we can't do it in interrupt context
324 	 * because we need to put kva back to kernel_map.
325 	 */
326 
327 	mutex_enter(&so_pendfree_lock);
328 	m->m_next = so_pendfree;
329 	so_pendfree = m;
330 	cv_signal(&pendfree_thread_cv);
331 	mutex_exit(&so_pendfree_lock);
332 }
333 
334 static long
335 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space)
336 {
337 	struct iovec *iov = uio->uio_iov;
338 	vaddr_t sva, eva;
339 	vsize_t len;
340 	vaddr_t lva;
341 	int npgs, error;
342 	vaddr_t va;
343 	int i;
344 
345 	if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace))
346 		return (0);
347 
348 	if (iov->iov_len < (size_t) space)
349 		space = iov->iov_len;
350 	if (space > SOCK_LOAN_CHUNK)
351 		space = SOCK_LOAN_CHUNK;
352 
353 	eva = round_page((vaddr_t) iov->iov_base + space);
354 	sva = trunc_page((vaddr_t) iov->iov_base);
355 	len = eva - sva;
356 	npgs = len >> PAGE_SHIFT;
357 
358 	KASSERT(npgs <= M_EXT_MAXPAGES);
359 
360 	lva = sokvaalloc(sva, len, so);
361 	if (lva == 0)
362 		return 0;
363 
364 	error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len,
365 	    m->m_ext.ext_pgs, UVM_LOAN_TOPAGE);
366 	if (error) {
367 		sokvafree(lva, len);
368 		return (0);
369 	}
370 
371 	for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE)
372 		pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
373 		    VM_PROT_READ, 0);
374 	pmap_update(pmap_kernel());
375 
376 	lva += (vaddr_t) iov->iov_base & PAGE_MASK;
377 
378 	MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so);
379 	m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
380 
381 	uio->uio_resid -= space;
382 	/* uio_offset not updated, not set/used for write(2) */
383 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space;
384 	uio->uio_iov->iov_len -= space;
385 	if (uio->uio_iov->iov_len == 0) {
386 		uio->uio_iov++;
387 		uio->uio_iovcnt--;
388 	}
389 
390 	return (space);
391 }
392 
393 struct mbuf *
394 getsombuf(struct socket *so, int type)
395 {
396 	struct mbuf *m;
397 
398 	m = m_get(M_WAIT, type);
399 	MCLAIM(m, so->so_mowner);
400 	return m;
401 }
402 
403 static int
404 socket_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
405     void *arg0, void *arg1, void *arg2, void *arg3)
406 {
407 	int result;
408 	enum kauth_network_req req;
409 
410 	result = KAUTH_RESULT_DEFER;
411 	req = (enum kauth_network_req)arg0;
412 
413 	if ((action != KAUTH_NETWORK_SOCKET) &&
414 	    (action != KAUTH_NETWORK_BIND))
415 		return result;
416 
417 	switch (req) {
418 	case KAUTH_REQ_NETWORK_BIND_PORT:
419 		result = KAUTH_RESULT_ALLOW;
420 		break;
421 
422 	case KAUTH_REQ_NETWORK_SOCKET_DROP: {
423 		/* Normal users can only drop their own connections. */
424 		struct socket *so = (struct socket *)arg1;
425 
426 		if (so->so_cred && proc_uidmatch(cred, so->so_cred) == 0)
427 			result = KAUTH_RESULT_ALLOW;
428 
429 		break;
430 		}
431 
432 	case KAUTH_REQ_NETWORK_SOCKET_OPEN:
433 		/* We allow "raw" routing/bluetooth sockets to anyone. */
434 		if ((u_long)arg1 == PF_ROUTE || (u_long)arg1 == PF_OROUTE
435 		    || (u_long)arg1 == PF_BLUETOOTH) {
436 			result = KAUTH_RESULT_ALLOW;
437 		} else {
438 			/* Privileged, let secmodel handle this. */
439 			if ((u_long)arg2 == SOCK_RAW)
440 				break;
441 		}
442 
443 		result = KAUTH_RESULT_ALLOW;
444 
445 		break;
446 
447 	case KAUTH_REQ_NETWORK_SOCKET_CANSEE:
448 		result = KAUTH_RESULT_ALLOW;
449 
450 		break;
451 
452 	default:
453 		break;
454 	}
455 
456 	return result;
457 }
458 
459 void
460 soinit(void)
461 {
462 
463 	sysctl_kern_socket_setup();
464 
465 	mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM);
466 	softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
467 	cv_init(&socurkva_cv, "sokva");
468 	cv_init(&pendfree_thread_cv, "sopendfr");
469 	soinit2();
470 
471 	/* Set the initial adjusted socket buffer size. */
472 	if (sb_max_set(sb_max))
473 		panic("bad initial sb_max value: %lu", sb_max);
474 
475 	socket_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK,
476 	    socket_listener_cb, NULL);
477 }
478 
479 void
480 soinit1(void)
481 {
482 	int error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL,
483 	    sopendfree_thread, NULL, &sopendfree_lwp, "sopendfree");
484 	if (error)
485 		panic("soinit1 %d", error);
486 }
487 
488 /*
489  * socreate: create a new socket of the specified type and the protocol.
490  *
491  * => Caller may specify another socket for lock sharing (must not be held).
492  * => Returns the new socket without lock held.
493  */
494 int
495 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l,
496 	 struct socket *lockso)
497 {
498 	const struct protosw	*prp;
499 	struct socket	*so;
500 	uid_t		uid;
501 	int		error;
502 	kmutex_t	*lock;
503 
504 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET,
505 	    KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type),
506 	    KAUTH_ARG(proto));
507 	if (error != 0)
508 		return error;
509 
510 	if (proto)
511 		prp = pffindproto(dom, proto, type);
512 	else
513 		prp = pffindtype(dom, type);
514 	if (prp == NULL) {
515 		/* no support for domain */
516 		if (pffinddomain(dom) == 0)
517 			return EAFNOSUPPORT;
518 		/* no support for socket type */
519 		if (proto == 0 && type != 0)
520 			return EPROTOTYPE;
521 		return EPROTONOSUPPORT;
522 	}
523 	if (prp->pr_usrreqs == NULL)
524 		return EPROTONOSUPPORT;
525 	if (prp->pr_type != type)
526 		return EPROTOTYPE;
527 
528 	so = soget(true);
529 	so->so_type = type;
530 	so->so_proto = prp;
531 	so->so_send = sosend;
532 	so->so_receive = soreceive;
533 #ifdef MBUFTRACE
534 	so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner;
535 	so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner;
536 	so->so_mowner = &prp->pr_domain->dom_mowner;
537 #endif
538 	uid = kauth_cred_geteuid(l->l_cred);
539 	so->so_uidinfo = uid_find(uid);
540 	so->so_cpid = l->l_proc->p_pid;
541 
542 	/*
543 	 * Lock assigned and taken during PCB attach, unless we share
544 	 * the lock with another socket, e.g. socketpair(2) case.
545 	 */
546 	if (lockso) {
547 		lock = lockso->so_lock;
548 		so->so_lock = lock;
549 		mutex_obj_hold(lock);
550 		mutex_enter(lock);
551 	}
552 
553 	/* Attach the PCB (returns with the socket lock held). */
554 	error = (*prp->pr_usrreqs->pr_attach)(so, proto);
555 	KASSERT(solocked(so));
556 
557 	if (error) {
558 		KASSERT(so->so_pcb == NULL);
559 		so->so_state |= SS_NOFDREF;
560 		sofree(so);
561 		return error;
562 	}
563 	so->so_cred = kauth_cred_dup(l->l_cred);
564 	sounlock(so);
565 
566 	*aso = so;
567 	return 0;
568 }
569 
570 /*
571  * fsocreate: create a socket and a file descriptor associated with it.
572  *
573  * => On success, write file descriptor to fdout and return zero.
574  * => On failure, return non-zero; *fdout will be undefined.
575  */
576 int
577 fsocreate(int domain, struct socket **sop, int type, int proto, int *fdout)
578 {
579 	lwp_t *l = curlwp;
580 	int error, fd, flags;
581 	struct socket *so;
582 	struct file *fp;
583 
584 	if ((error = fd_allocfile(&fp, &fd)) != 0) {
585 		return error;
586 	}
587 	flags = type & SOCK_FLAGS_MASK;
588 	fd_set_exclose(l, fd, (flags & SOCK_CLOEXEC) != 0);
589 	fp->f_flag = FREAD|FWRITE|((flags & SOCK_NONBLOCK) ? FNONBLOCK : 0)|
590 	    ((flags & SOCK_NOSIGPIPE) ? FNOSIGPIPE : 0);
591 	fp->f_type = DTYPE_SOCKET;
592 	fp->f_ops = &socketops;
593 
594 	type &= ~SOCK_FLAGS_MASK;
595 	error = socreate(domain, &so, type, proto, l, NULL);
596 	if (error) {
597 		fd_abort(curproc, fp, fd);
598 		return error;
599 	}
600 	if (flags & SOCK_NONBLOCK) {
601 		so->so_state |= SS_NBIO;
602 	}
603 	fp->f_socket = so;
604 	fd_affix(curproc, fp, fd);
605 
606 	if (sop != NULL) {
607 		*sop = so;
608 	}
609 	*fdout = fd;
610 	return error;
611 }
612 
613 int
614 sofamily(const struct socket *so)
615 {
616 	const struct protosw *pr;
617 	const struct domain *dom;
618 
619 	if ((pr = so->so_proto) == NULL)
620 		return AF_UNSPEC;
621 	if ((dom = pr->pr_domain) == NULL)
622 		return AF_UNSPEC;
623 	return dom->dom_family;
624 }
625 
626 int
627 sobind(struct socket *so, struct mbuf *nam, struct lwp *l)
628 {
629 	int	error;
630 
631 	solock(so);
632 	error = (*so->so_proto->pr_usrreqs->pr_bind)(so, nam, l);
633 	sounlock(so);
634 	return error;
635 }
636 
637 int
638 solisten(struct socket *so, int backlog, struct lwp *l)
639 {
640 	int	error;
641 
642 	solock(so);
643 	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
644 	    SS_ISDISCONNECTING)) != 0) {
645 		sounlock(so);
646 		return EINVAL;
647 	}
648 	error = (*so->so_proto->pr_usrreqs->pr_listen)(so, l);
649 	if (error != 0) {
650 		sounlock(so);
651 		return error;
652 	}
653 	if (TAILQ_EMPTY(&so->so_q))
654 		so->so_options |= SO_ACCEPTCONN;
655 	if (backlog < 0)
656 		backlog = 0;
657 	so->so_qlimit = min(backlog, somaxconn);
658 	sounlock(so);
659 	return 0;
660 }
661 
662 void
663 sofree(struct socket *so)
664 {
665 	u_int refs;
666 
667 	KASSERT(solocked(so));
668 
669 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
670 		sounlock(so);
671 		return;
672 	}
673 	if (so->so_head) {
674 		/*
675 		 * We must not decommission a socket that's on the accept(2)
676 		 * queue.  If we do, then accept(2) may hang after select(2)
677 		 * indicated that the listening socket was ready.
678 		 */
679 		if (!soqremque(so, 0)) {
680 			sounlock(so);
681 			return;
682 		}
683 	}
684 	if (so->so_rcv.sb_hiwat)
685 		(void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0,
686 		    RLIM_INFINITY);
687 	if (so->so_snd.sb_hiwat)
688 		(void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0,
689 		    RLIM_INFINITY);
690 	sbrelease(&so->so_snd, so);
691 	KASSERT(!cv_has_waiters(&so->so_cv));
692 	KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv));
693 	KASSERT(!cv_has_waiters(&so->so_snd.sb_cv));
694 	sorflush(so);
695 	refs = so->so_aborting;	/* XXX */
696 	/* Remove acccept filter if one is present. */
697 	if (so->so_accf != NULL)
698 		(void)accept_filt_clear(so);
699 	sounlock(so);
700 	if (refs == 0)		/* XXX */
701 		soput(so);
702 }
703 
704 /*
705  * soclose: close a socket on last file table reference removal.
706  * Initiate disconnect if connected.  Free socket when disconnect complete.
707  */
708 int
709 soclose(struct socket *so)
710 {
711 	struct socket *so2;
712 	int error = 0;
713 
714 	solock(so);
715 	if (so->so_options & SO_ACCEPTCONN) {
716 		for (;;) {
717 			if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) {
718 				KASSERT(solocked2(so, so2));
719 				(void) soqremque(so2, 0);
720 				/* soabort drops the lock. */
721 				(void) soabort(so2);
722 				solock(so);
723 				continue;
724 			}
725 			if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) {
726 				KASSERT(solocked2(so, so2));
727 				(void) soqremque(so2, 1);
728 				/* soabort drops the lock. */
729 				(void) soabort(so2);
730 				solock(so);
731 				continue;
732 			}
733 			break;
734 		}
735 	}
736 	if (so->so_pcb == NULL)
737 		goto discard;
738 	if (so->so_state & SS_ISCONNECTED) {
739 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
740 			error = sodisconnect(so);
741 			if (error)
742 				goto drop;
743 		}
744 		if (so->so_options & SO_LINGER) {
745 			if ((so->so_state & (SS_ISDISCONNECTING|SS_NBIO)) ==
746 			    (SS_ISDISCONNECTING|SS_NBIO))
747 				goto drop;
748 			while (so->so_state & SS_ISCONNECTED) {
749 				error = sowait(so, true, so->so_linger * hz);
750 				if (error)
751 					break;
752 			}
753 		}
754 	}
755  drop:
756 	if (so->so_pcb) {
757 		KASSERT(solocked(so));
758 		(*so->so_proto->pr_usrreqs->pr_detach)(so);
759 	}
760  discard:
761 	KASSERT((so->so_state & SS_NOFDREF) == 0);
762 	kauth_cred_free(so->so_cred);
763 	so->so_state |= SS_NOFDREF;
764 	sofree(so);
765 	return error;
766 }
767 
768 /*
769  * Must be called with the socket locked..  Will return with it unlocked.
770  */
771 int
772 soabort(struct socket *so)
773 {
774 	u_int refs;
775 	int error;
776 
777 	KASSERT(solocked(so));
778 	KASSERT(so->so_head == NULL);
779 
780 	so->so_aborting++;		/* XXX */
781 	error = (*so->so_proto->pr_usrreqs->pr_abort)(so);
782 	refs = --so->so_aborting;	/* XXX */
783 	if (error || (refs == 0)) {
784 		sofree(so);
785 	} else {
786 		sounlock(so);
787 	}
788 	return error;
789 }
790 
791 int
792 soaccept(struct socket *so, struct mbuf *nam)
793 {
794 	int error;
795 
796 	KASSERT(solocked(so));
797 	KASSERT((so->so_state & SS_NOFDREF) != 0);
798 
799 	so->so_state &= ~SS_NOFDREF;
800 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
801 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
802 		error = (*so->so_proto->pr_usrreqs->pr_accept)(so, nam);
803 	else
804 		error = ECONNABORTED;
805 
806 	return error;
807 }
808 
809 int
810 soconnect(struct socket *so, struct mbuf *nam, struct lwp *l)
811 {
812 	int error;
813 
814 	KASSERT(solocked(so));
815 
816 	if (so->so_options & SO_ACCEPTCONN)
817 		return EOPNOTSUPP;
818 	/*
819 	 * If protocol is connection-based, can only connect once.
820 	 * Otherwise, if connected, try to disconnect first.
821 	 * This allows user to disconnect by connecting to, e.g.,
822 	 * a null address.
823 	 */
824 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
825 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
826 	    (error = sodisconnect(so))))
827 		error = EISCONN;
828 	else
829 		error = (*so->so_proto->pr_usrreqs->pr_connect)(so, nam, l);
830 
831 	return error;
832 }
833 
834 int
835 soconnect2(struct socket *so1, struct socket *so2)
836 {
837 	KASSERT(solocked2(so1, so2));
838 
839 	return (*so1->so_proto->pr_usrreqs->pr_connect2)(so1, so2);
840 }
841 
842 int
843 sodisconnect(struct socket *so)
844 {
845 	int	error;
846 
847 	KASSERT(solocked(so));
848 
849 	if ((so->so_state & SS_ISCONNECTED) == 0) {
850 		error = ENOTCONN;
851 	} else if (so->so_state & SS_ISDISCONNECTING) {
852 		error = EALREADY;
853 	} else {
854 		error = (*so->so_proto->pr_usrreqs->pr_disconnect)(so);
855 	}
856 	return (error);
857 }
858 
859 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
860 /*
861  * Send on a socket.
862  * If send must go all at once and message is larger than
863  * send buffering, then hard error.
864  * Lock against other senders.
865  * If must go all at once and not enough room now, then
866  * inform user that this would block and do nothing.
867  * Otherwise, if nonblocking, send as much as possible.
868  * The data to be sent is described by "uio" if nonzero,
869  * otherwise by the mbuf chain "top" (which must be null
870  * if uio is not).  Data provided in mbuf chain must be small
871  * enough to send all at once.
872  *
873  * Returns nonzero on error, timeout or signal; callers
874  * must check for short counts if EINTR/ERESTART are returned.
875  * Data and control buffers are freed on return.
876  */
877 int
878 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
879 	struct mbuf *control, int flags, struct lwp *l)
880 {
881 	struct mbuf	**mp, *m;
882 	long		space, len, resid, clen, mlen;
883 	int		error, s, dontroute, atomic;
884 	short		wakeup_state = 0;
885 
886 	clen = 0;
887 
888 	/*
889 	 * solock() provides atomicity of access.  splsoftnet() prevents
890 	 * protocol processing soft interrupts from interrupting us and
891 	 * blocking (expensive).
892 	 */
893 	s = splsoftnet();
894 	solock(so);
895 	atomic = sosendallatonce(so) || top;
896 	if (uio)
897 		resid = uio->uio_resid;
898 	else
899 		resid = top->m_pkthdr.len;
900 	/*
901 	 * In theory resid should be unsigned.
902 	 * However, space must be signed, as it might be less than 0
903 	 * if we over-committed, and we must use a signed comparison
904 	 * of space and resid.  On the other hand, a negative resid
905 	 * causes us to loop sending 0-length segments to the protocol.
906 	 */
907 	if (resid < 0) {
908 		error = EINVAL;
909 		goto out;
910 	}
911 	dontroute =
912 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
913 	    (so->so_proto->pr_flags & PR_ATOMIC);
914 	l->l_ru.ru_msgsnd++;
915 	if (control)
916 		clen = control->m_len;
917  restart:
918 	if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
919 		goto out;
920 	do {
921 		if (so->so_state & SS_CANTSENDMORE) {
922 			error = EPIPE;
923 			goto release;
924 		}
925 		if (so->so_error) {
926 			error = so->so_error;
927 			so->so_error = 0;
928 			goto release;
929 		}
930 		if ((so->so_state & SS_ISCONNECTED) == 0) {
931 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
932 				if (resid || clen == 0) {
933 					error = ENOTCONN;
934 					goto release;
935 				}
936 			} else if (addr == 0) {
937 				error = EDESTADDRREQ;
938 				goto release;
939 			}
940 		}
941 		space = sbspace(&so->so_snd);
942 		if (flags & MSG_OOB)
943 			space += 1024;
944 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
945 		    clen > so->so_snd.sb_hiwat) {
946 			error = EMSGSIZE;
947 			goto release;
948 		}
949 		if (space < resid + clen &&
950 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
951 			if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
952 				error = EWOULDBLOCK;
953 				goto release;
954 			}
955 			sbunlock(&so->so_snd);
956 			if (wakeup_state & SS_RESTARTSYS) {
957 				error = ERESTART;
958 				goto out;
959 			}
960 			error = sbwait(&so->so_snd);
961 			if (error)
962 				goto out;
963 			wakeup_state = so->so_state;
964 			goto restart;
965 		}
966 		wakeup_state = 0;
967 		mp = &top;
968 		space -= clen;
969 		do {
970 			if (uio == NULL) {
971 				/*
972 				 * Data is prepackaged in "top".
973 				 */
974 				resid = 0;
975 				if (flags & MSG_EOR)
976 					top->m_flags |= M_EOR;
977 			} else do {
978 				sounlock(so);
979 				splx(s);
980 				if (top == NULL) {
981 					m = m_gethdr(M_WAIT, MT_DATA);
982 					mlen = MHLEN;
983 					m->m_pkthdr.len = 0;
984 					m->m_pkthdr.rcvif = NULL;
985 				} else {
986 					m = m_get(M_WAIT, MT_DATA);
987 					mlen = MLEN;
988 				}
989 				MCLAIM(m, so->so_snd.sb_mowner);
990 				if (sock_loan_thresh >= 0 &&
991 				    uio->uio_iov->iov_len >= sock_loan_thresh &&
992 				    space >= sock_loan_thresh &&
993 				    (len = sosend_loan(so, uio, m,
994 						       space)) != 0) {
995 					SOSEND_COUNTER_INCR(&sosend_loan_big);
996 					space -= len;
997 					goto have_data;
998 				}
999 				if (resid >= MINCLSIZE && space >= MCLBYTES) {
1000 					SOSEND_COUNTER_INCR(&sosend_copy_big);
1001 					m_clget(m, M_DONTWAIT);
1002 					if ((m->m_flags & M_EXT) == 0)
1003 						goto nopages;
1004 					mlen = MCLBYTES;
1005 					if (atomic && top == 0) {
1006 						len = lmin(MCLBYTES - max_hdr,
1007 						    resid);
1008 						m->m_data += max_hdr;
1009 					} else
1010 						len = lmin(MCLBYTES, resid);
1011 					space -= len;
1012 				} else {
1013  nopages:
1014 					SOSEND_COUNTER_INCR(&sosend_copy_small);
1015 					len = lmin(lmin(mlen, resid), space);
1016 					space -= len;
1017 					/*
1018 					 * For datagram protocols, leave room
1019 					 * for protocol headers in first mbuf.
1020 					 */
1021 					if (atomic && top == 0 && len < mlen)
1022 						MH_ALIGN(m, len);
1023 				}
1024 				error = uiomove(mtod(m, void *), (int)len, uio);
1025  have_data:
1026 				resid = uio->uio_resid;
1027 				m->m_len = len;
1028 				*mp = m;
1029 				top->m_pkthdr.len += len;
1030 				s = splsoftnet();
1031 				solock(so);
1032 				if (error != 0)
1033 					goto release;
1034 				mp = &m->m_next;
1035 				if (resid <= 0) {
1036 					if (flags & MSG_EOR)
1037 						top->m_flags |= M_EOR;
1038 					break;
1039 				}
1040 			} while (space > 0 && atomic);
1041 
1042 			if (so->so_state & SS_CANTSENDMORE) {
1043 				error = EPIPE;
1044 				goto release;
1045 			}
1046 			if (dontroute)
1047 				so->so_options |= SO_DONTROUTE;
1048 			if (resid > 0)
1049 				so->so_state |= SS_MORETOCOME;
1050 			if (flags & MSG_OOB)
1051 				error = (*so->so_proto->pr_usrreqs->pr_sendoob)(so,
1052 				    top, control);
1053 			else
1054 				error = (*so->so_proto->pr_usrreqs->pr_send)(so,
1055 				    top, addr, control, l);
1056 			if (dontroute)
1057 				so->so_options &= ~SO_DONTROUTE;
1058 			if (resid > 0)
1059 				so->so_state &= ~SS_MORETOCOME;
1060 			clen = 0;
1061 			control = NULL;
1062 			top = NULL;
1063 			mp = &top;
1064 			if (error != 0)
1065 				goto release;
1066 		} while (resid && space > 0);
1067 	} while (resid);
1068 
1069  release:
1070 	sbunlock(&so->so_snd);
1071  out:
1072 	sounlock(so);
1073 	splx(s);
1074 	if (top)
1075 		m_freem(top);
1076 	if (control)
1077 		m_freem(control);
1078 	return (error);
1079 }
1080 
1081 /*
1082  * Following replacement or removal of the first mbuf on the first
1083  * mbuf chain of a socket buffer, push necessary state changes back
1084  * into the socket buffer so that other consumers see the values
1085  * consistently.  'nextrecord' is the callers locally stored value of
1086  * the original value of sb->sb_mb->m_nextpkt which must be restored
1087  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
1088  */
1089 static void
1090 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
1091 {
1092 
1093 	KASSERT(solocked(sb->sb_so));
1094 
1095 	/*
1096 	 * First, update for the new value of nextrecord.  If necessary,
1097 	 * make it the first record.
1098 	 */
1099 	if (sb->sb_mb != NULL)
1100 		sb->sb_mb->m_nextpkt = nextrecord;
1101 	else
1102 		sb->sb_mb = nextrecord;
1103 
1104         /*
1105          * Now update any dependent socket buffer fields to reflect
1106          * the new state.  This is an inline of SB_EMPTY_FIXUP, with
1107          * the addition of a second clause that takes care of the
1108          * case where sb_mb has been updated, but remains the last
1109          * record.
1110          */
1111         if (sb->sb_mb == NULL) {
1112                 sb->sb_mbtail = NULL;
1113                 sb->sb_lastrecord = NULL;
1114         } else if (sb->sb_mb->m_nextpkt == NULL)
1115                 sb->sb_lastrecord = sb->sb_mb;
1116 }
1117 
1118 /*
1119  * Implement receive operations on a socket.
1120  * We depend on the way that records are added to the sockbuf
1121  * by sbappend*.  In particular, each record (mbufs linked through m_next)
1122  * must begin with an address if the protocol so specifies,
1123  * followed by an optional mbuf or mbufs containing ancillary data,
1124  * and then zero or more mbufs of data.
1125  * In order to avoid blocking network interrupts for the entire time here,
1126  * we splx() while doing the actual copy to user space.
1127  * Although the sockbuf is locked, new data may still be appended,
1128  * and thus we must maintain consistency of the sockbuf during that time.
1129  *
1130  * The caller may receive the data as a single mbuf chain by supplying
1131  * an mbuf **mp0 for use in returning the chain.  The uio is then used
1132  * only for the count in uio_resid.
1133  */
1134 int
1135 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
1136 	struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1137 {
1138 	struct lwp *l = curlwp;
1139 	struct mbuf	*m, **mp, *mt;
1140 	size_t len, offset, moff, orig_resid;
1141 	int atomic, flags, error, s, type;
1142 	const struct protosw	*pr;
1143 	struct mbuf	*nextrecord;
1144 	int		mbuf_removed = 0;
1145 	const struct domain *dom;
1146 	short		wakeup_state = 0;
1147 
1148 	pr = so->so_proto;
1149 	atomic = pr->pr_flags & PR_ATOMIC;
1150 	dom = pr->pr_domain;
1151 	mp = mp0;
1152 	type = 0;
1153 	orig_resid = uio->uio_resid;
1154 
1155 	if (paddr != NULL)
1156 		*paddr = NULL;
1157 	if (controlp != NULL)
1158 		*controlp = NULL;
1159 	if (flagsp != NULL)
1160 		flags = *flagsp &~ MSG_EOR;
1161 	else
1162 		flags = 0;
1163 
1164 	if (flags & MSG_OOB) {
1165 		m = m_get(M_WAIT, MT_DATA);
1166 		solock(so);
1167 		error = (*pr->pr_usrreqs->pr_recvoob)(so, m, flags & MSG_PEEK);
1168 		sounlock(so);
1169 		if (error)
1170 			goto bad;
1171 		do {
1172 			error = uiomove(mtod(m, void *),
1173 			    MIN(uio->uio_resid, m->m_len), uio);
1174 			m = m_free(m);
1175 		} while (uio->uio_resid > 0 && error == 0 && m);
1176  bad:
1177 		if (m != NULL)
1178 			m_freem(m);
1179 		return error;
1180 	}
1181 	if (mp != NULL)
1182 		*mp = NULL;
1183 
1184 	/*
1185 	 * solock() provides atomicity of access.  splsoftnet() prevents
1186 	 * protocol processing soft interrupts from interrupting us and
1187 	 * blocking (expensive).
1188 	 */
1189 	s = splsoftnet();
1190 	solock(so);
1191  restart:
1192 	if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) {
1193 		sounlock(so);
1194 		splx(s);
1195 		return error;
1196 	}
1197 
1198 	m = so->so_rcv.sb_mb;
1199 	/*
1200 	 * If we have less data than requested, block awaiting more
1201 	 * (subject to any timeout) if:
1202 	 *   1. the current count is less than the low water mark,
1203 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1204 	 *	receive operation at once if we block (resid <= hiwat), or
1205 	 *   3. MSG_DONTWAIT is not set.
1206 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1207 	 * we have to do the receive in sections, and thus risk returning
1208 	 * a short count if a timeout or signal occurs after we start.
1209 	 */
1210 	if (m == NULL ||
1211 	    ((flags & MSG_DONTWAIT) == 0 &&
1212 	     so->so_rcv.sb_cc < uio->uio_resid &&
1213 	     (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1214 	      ((flags & MSG_WAITALL) &&
1215 	       uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1216 	     m->m_nextpkt == NULL && !atomic)) {
1217 #ifdef DIAGNOSTIC
1218 		if (m == NULL && so->so_rcv.sb_cc)
1219 			panic("receive 1");
1220 #endif
1221 		if (so->so_error) {
1222 			if (m != NULL)
1223 				goto dontblock;
1224 			error = so->so_error;
1225 			if ((flags & MSG_PEEK) == 0)
1226 				so->so_error = 0;
1227 			goto release;
1228 		}
1229 		if (so->so_state & SS_CANTRCVMORE) {
1230 			if (m != NULL)
1231 				goto dontblock;
1232 			else
1233 				goto release;
1234 		}
1235 		for (; m != NULL; m = m->m_next)
1236 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1237 				m = so->so_rcv.sb_mb;
1238 				goto dontblock;
1239 			}
1240 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1241 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1242 			error = ENOTCONN;
1243 			goto release;
1244 		}
1245 		if (uio->uio_resid == 0)
1246 			goto release;
1247 		if ((so->so_state & SS_NBIO) ||
1248 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1249 			error = EWOULDBLOCK;
1250 			goto release;
1251 		}
1252 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
1253 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
1254 		sbunlock(&so->so_rcv);
1255 		if (wakeup_state & SS_RESTARTSYS)
1256 			error = ERESTART;
1257 		else
1258 			error = sbwait(&so->so_rcv);
1259 		if (error != 0) {
1260 			sounlock(so);
1261 			splx(s);
1262 			return error;
1263 		}
1264 		wakeup_state = so->so_state;
1265 		goto restart;
1266 	}
1267  dontblock:
1268 	/*
1269 	 * On entry here, m points to the first record of the socket buffer.
1270 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1271 	 * pointer to the next record in the socket buffer.  We must keep the
1272 	 * various socket buffer pointers and local stack versions of the
1273 	 * pointers in sync, pushing out modifications before dropping the
1274 	 * socket lock, and re-reading them when picking it up.
1275 	 *
1276 	 * Otherwise, we will race with the network stack appending new data
1277 	 * or records onto the socket buffer by using inconsistent/stale
1278 	 * versions of the field, possibly resulting in socket buffer
1279 	 * corruption.
1280 	 *
1281 	 * By holding the high-level sblock(), we prevent simultaneous
1282 	 * readers from pulling off the front of the socket buffer.
1283 	 */
1284 	if (l != NULL)
1285 		l->l_ru.ru_msgrcv++;
1286 	KASSERT(m == so->so_rcv.sb_mb);
1287 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
1288 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
1289 	nextrecord = m->m_nextpkt;
1290 	if (pr->pr_flags & PR_ADDR) {
1291 #ifdef DIAGNOSTIC
1292 		if (m->m_type != MT_SONAME)
1293 			panic("receive 1a");
1294 #endif
1295 		orig_resid = 0;
1296 		if (flags & MSG_PEEK) {
1297 			if (paddr)
1298 				*paddr = m_copy(m, 0, m->m_len);
1299 			m = m->m_next;
1300 		} else {
1301 			sbfree(&so->so_rcv, m);
1302 			mbuf_removed = 1;
1303 			if (paddr != NULL) {
1304 				*paddr = m;
1305 				so->so_rcv.sb_mb = m->m_next;
1306 				m->m_next = NULL;
1307 				m = so->so_rcv.sb_mb;
1308 			} else {
1309 				MFREE(m, so->so_rcv.sb_mb);
1310 				m = so->so_rcv.sb_mb;
1311 			}
1312 			sbsync(&so->so_rcv, nextrecord);
1313 		}
1314 	}
1315 
1316 	/*
1317 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1318 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1319 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1320 	 * perform externalization (or freeing if controlp == NULL).
1321 	 */
1322 	if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) {
1323 		struct mbuf *cm = NULL, *cmn;
1324 		struct mbuf **cme = &cm;
1325 
1326 		do {
1327 			if (flags & MSG_PEEK) {
1328 				if (controlp != NULL) {
1329 					*controlp = m_copy(m, 0, m->m_len);
1330 					controlp = &(*controlp)->m_next;
1331 				}
1332 				m = m->m_next;
1333 			} else {
1334 				sbfree(&so->so_rcv, m);
1335 				so->so_rcv.sb_mb = m->m_next;
1336 				m->m_next = NULL;
1337 				*cme = m;
1338 				cme = &(*cme)->m_next;
1339 				m = so->so_rcv.sb_mb;
1340 			}
1341 		} while (m != NULL && m->m_type == MT_CONTROL);
1342 		if ((flags & MSG_PEEK) == 0)
1343 			sbsync(&so->so_rcv, nextrecord);
1344 		for (; cm != NULL; cm = cmn) {
1345 			cmn = cm->m_next;
1346 			cm->m_next = NULL;
1347 			type = mtod(cm, struct cmsghdr *)->cmsg_type;
1348 			if (controlp != NULL) {
1349 				if (dom->dom_externalize != NULL &&
1350 				    type == SCM_RIGHTS) {
1351 					sounlock(so);
1352 					splx(s);
1353 					error = (*dom->dom_externalize)(cm, l,
1354 					    (flags & MSG_CMSG_CLOEXEC) ?
1355 					    O_CLOEXEC : 0);
1356 					s = splsoftnet();
1357 					solock(so);
1358 				}
1359 				*controlp = cm;
1360 				while (*controlp != NULL)
1361 					controlp = &(*controlp)->m_next;
1362 			} else {
1363 				/*
1364 				 * Dispose of any SCM_RIGHTS message that went
1365 				 * through the read path rather than recv.
1366 				 */
1367 				if (dom->dom_dispose != NULL &&
1368 				    type == SCM_RIGHTS) {
1369 				    	sounlock(so);
1370 					(*dom->dom_dispose)(cm);
1371 					solock(so);
1372 				}
1373 				m_freem(cm);
1374 			}
1375 		}
1376 		if (m != NULL)
1377 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1378 		else
1379 			nextrecord = so->so_rcv.sb_mb;
1380 		orig_resid = 0;
1381 	}
1382 
1383 	/* If m is non-NULL, we have some data to read. */
1384 	if (__predict_true(m != NULL)) {
1385 		type = m->m_type;
1386 		if (type == MT_OOBDATA)
1387 			flags |= MSG_OOB;
1388 	}
1389 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1390 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1391 
1392 	moff = 0;
1393 	offset = 0;
1394 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1395 		if (m->m_type == MT_OOBDATA) {
1396 			if (type != MT_OOBDATA)
1397 				break;
1398 		} else if (type == MT_OOBDATA)
1399 			break;
1400 #ifdef DIAGNOSTIC
1401 		else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
1402 			panic("receive 3");
1403 #endif
1404 		so->so_state &= ~SS_RCVATMARK;
1405 		wakeup_state = 0;
1406 		len = uio->uio_resid;
1407 		if (so->so_oobmark && len > so->so_oobmark - offset)
1408 			len = so->so_oobmark - offset;
1409 		if (len > m->m_len - moff)
1410 			len = m->m_len - moff;
1411 		/*
1412 		 * If mp is set, just pass back the mbufs.
1413 		 * Otherwise copy them out via the uio, then free.
1414 		 * Sockbuf must be consistent here (points to current mbuf,
1415 		 * it points to next record) when we drop priority;
1416 		 * we must note any additions to the sockbuf when we
1417 		 * block interrupts again.
1418 		 */
1419 		if (mp == NULL) {
1420 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1421 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1422 			sounlock(so);
1423 			splx(s);
1424 			error = uiomove(mtod(m, char *) + moff, len, uio);
1425 			s = splsoftnet();
1426 			solock(so);
1427 			if (error != 0) {
1428 				/*
1429 				 * If any part of the record has been removed
1430 				 * (such as the MT_SONAME mbuf, which will
1431 				 * happen when PR_ADDR, and thus also
1432 				 * PR_ATOMIC, is set), then drop the entire
1433 				 * record to maintain the atomicity of the
1434 				 * receive operation.
1435 				 *
1436 				 * This avoids a later panic("receive 1a")
1437 				 * when compiled with DIAGNOSTIC.
1438 				 */
1439 				if (m && mbuf_removed && atomic)
1440 					(void) sbdroprecord(&so->so_rcv);
1441 
1442 				goto release;
1443 			}
1444 		} else
1445 			uio->uio_resid -= len;
1446 		if (len == m->m_len - moff) {
1447 			if (m->m_flags & M_EOR)
1448 				flags |= MSG_EOR;
1449 			if (flags & MSG_PEEK) {
1450 				m = m->m_next;
1451 				moff = 0;
1452 			} else {
1453 				nextrecord = m->m_nextpkt;
1454 				sbfree(&so->so_rcv, m);
1455 				if (mp) {
1456 					*mp = m;
1457 					mp = &m->m_next;
1458 					so->so_rcv.sb_mb = m = m->m_next;
1459 					*mp = NULL;
1460 				} else {
1461 					MFREE(m, so->so_rcv.sb_mb);
1462 					m = so->so_rcv.sb_mb;
1463 				}
1464 				/*
1465 				 * If m != NULL, we also know that
1466 				 * so->so_rcv.sb_mb != NULL.
1467 				 */
1468 				KASSERT(so->so_rcv.sb_mb == m);
1469 				if (m) {
1470 					m->m_nextpkt = nextrecord;
1471 					if (nextrecord == NULL)
1472 						so->so_rcv.sb_lastrecord = m;
1473 				} else {
1474 					so->so_rcv.sb_mb = nextrecord;
1475 					SB_EMPTY_FIXUP(&so->so_rcv);
1476 				}
1477 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1478 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1479 			}
1480 		} else if (flags & MSG_PEEK)
1481 			moff += len;
1482 		else {
1483 			if (mp != NULL) {
1484 				mt = m_copym(m, 0, len, M_NOWAIT);
1485 				if (__predict_false(mt == NULL)) {
1486 					sounlock(so);
1487 					mt = m_copym(m, 0, len, M_WAIT);
1488 					solock(so);
1489 				}
1490 				*mp = mt;
1491 			}
1492 			m->m_data += len;
1493 			m->m_len -= len;
1494 			so->so_rcv.sb_cc -= len;
1495 		}
1496 		if (so->so_oobmark) {
1497 			if ((flags & MSG_PEEK) == 0) {
1498 				so->so_oobmark -= len;
1499 				if (so->so_oobmark == 0) {
1500 					so->so_state |= SS_RCVATMARK;
1501 					break;
1502 				}
1503 			} else {
1504 				offset += len;
1505 				if (offset == so->so_oobmark)
1506 					break;
1507 			}
1508 		}
1509 		if (flags & MSG_EOR)
1510 			break;
1511 		/*
1512 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1513 		 * we must not quit until "uio->uio_resid == 0" or an error
1514 		 * termination.  If a signal/timeout occurs, return
1515 		 * with a short count but without error.
1516 		 * Keep sockbuf locked against other readers.
1517 		 */
1518 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1519 		    !sosendallatonce(so) && !nextrecord) {
1520 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1521 				break;
1522 			/*
1523 			 * If we are peeking and the socket receive buffer is
1524 			 * full, stop since we can't get more data to peek at.
1525 			 */
1526 			if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
1527 				break;
1528 			/*
1529 			 * If we've drained the socket buffer, tell the
1530 			 * protocol in case it needs to do something to
1531 			 * get it filled again.
1532 			 */
1533 			if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1534 				(*pr->pr_usrreqs->pr_rcvd)(so, flags, l);
1535 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1536 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1537 			if (wakeup_state & SS_RESTARTSYS)
1538 				error = ERESTART;
1539 			else
1540 				error = sbwait(&so->so_rcv);
1541 			if (error != 0) {
1542 				sbunlock(&so->so_rcv);
1543 				sounlock(so);
1544 				splx(s);
1545 				return 0;
1546 			}
1547 			if ((m = so->so_rcv.sb_mb) != NULL)
1548 				nextrecord = m->m_nextpkt;
1549 			wakeup_state = so->so_state;
1550 		}
1551 	}
1552 
1553 	if (m && atomic) {
1554 		flags |= MSG_TRUNC;
1555 		if ((flags & MSG_PEEK) == 0)
1556 			(void) sbdroprecord(&so->so_rcv);
1557 	}
1558 	if ((flags & MSG_PEEK) == 0) {
1559 		if (m == NULL) {
1560 			/*
1561 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1562 			 * part makes sure sb_lastrecord is up-to-date if
1563 			 * there is still data in the socket buffer.
1564 			 */
1565 			so->so_rcv.sb_mb = nextrecord;
1566 			if (so->so_rcv.sb_mb == NULL) {
1567 				so->so_rcv.sb_mbtail = NULL;
1568 				so->so_rcv.sb_lastrecord = NULL;
1569 			} else if (nextrecord->m_nextpkt == NULL)
1570 				so->so_rcv.sb_lastrecord = nextrecord;
1571 		}
1572 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1573 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1574 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1575 			(*pr->pr_usrreqs->pr_rcvd)(so, flags, l);
1576 	}
1577 	if (orig_resid == uio->uio_resid && orig_resid &&
1578 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1579 		sbunlock(&so->so_rcv);
1580 		goto restart;
1581 	}
1582 
1583 	if (flagsp != NULL)
1584 		*flagsp |= flags;
1585  release:
1586 	sbunlock(&so->so_rcv);
1587 	sounlock(so);
1588 	splx(s);
1589 	return error;
1590 }
1591 
1592 int
1593 soshutdown(struct socket *so, int how)
1594 {
1595 	const struct protosw	*pr;
1596 	int	error;
1597 
1598 	KASSERT(solocked(so));
1599 
1600 	pr = so->so_proto;
1601 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1602 		return (EINVAL);
1603 
1604 	if (how == SHUT_RD || how == SHUT_RDWR) {
1605 		sorflush(so);
1606 		error = 0;
1607 	}
1608 	if (how == SHUT_WR || how == SHUT_RDWR)
1609 		error = (*pr->pr_usrreqs->pr_shutdown)(so);
1610 
1611 	return error;
1612 }
1613 
1614 void
1615 sorestart(struct socket *so)
1616 {
1617 	/*
1618 	 * An application has called close() on an fd on which another
1619 	 * of its threads has called a socket system call.
1620 	 * Mark this and wake everyone up, and code that would block again
1621 	 * instead returns ERESTART.
1622 	 * On system call re-entry the fd is validated and EBADF returned.
1623 	 * Any other fd will block again on the 2nd syscall.
1624 	 */
1625 	solock(so);
1626 	so->so_state |= SS_RESTARTSYS;
1627 	cv_broadcast(&so->so_cv);
1628 	cv_broadcast(&so->so_snd.sb_cv);
1629 	cv_broadcast(&so->so_rcv.sb_cv);
1630 	sounlock(so);
1631 }
1632 
1633 void
1634 sorflush(struct socket *so)
1635 {
1636 	struct sockbuf	*sb, asb;
1637 	const struct protosw	*pr;
1638 
1639 	KASSERT(solocked(so));
1640 
1641 	sb = &so->so_rcv;
1642 	pr = so->so_proto;
1643 	socantrcvmore(so);
1644 	sb->sb_flags |= SB_NOINTR;
1645 	(void )sblock(sb, M_WAITOK);
1646 	sbunlock(sb);
1647 	asb = *sb;
1648 	/*
1649 	 * Clear most of the sockbuf structure, but leave some of the
1650 	 * fields valid.
1651 	 */
1652 	memset(&sb->sb_startzero, 0,
1653 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1654 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) {
1655 		sounlock(so);
1656 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1657 		solock(so);
1658 	}
1659 	sbrelease(&asb, so);
1660 }
1661 
1662 /*
1663  * internal set SOL_SOCKET options
1664  */
1665 static int
1666 sosetopt1(struct socket *so, const struct sockopt *sopt)
1667 {
1668 	int error = EINVAL, opt;
1669 	int optval = 0; /* XXX: gcc */
1670 	struct linger l;
1671 	struct timeval tv;
1672 
1673 	switch ((opt = sopt->sopt_name)) {
1674 
1675 	case SO_ACCEPTFILTER:
1676 		error = accept_filt_setopt(so, sopt);
1677 		KASSERT(solocked(so));
1678 		break;
1679 
1680   	case SO_LINGER:
1681  		error = sockopt_get(sopt, &l, sizeof(l));
1682 		solock(so);
1683  		if (error)
1684  			break;
1685  		if (l.l_linger < 0 || l.l_linger > USHRT_MAX ||
1686  		    l.l_linger > (INT_MAX / hz)) {
1687 			error = EDOM;
1688 			break;
1689 		}
1690  		so->so_linger = l.l_linger;
1691  		if (l.l_onoff)
1692  			so->so_options |= SO_LINGER;
1693  		else
1694  			so->so_options &= ~SO_LINGER;
1695    		break;
1696 
1697 	case SO_DEBUG:
1698 	case SO_KEEPALIVE:
1699 	case SO_DONTROUTE:
1700 	case SO_USELOOPBACK:
1701 	case SO_BROADCAST:
1702 	case SO_REUSEADDR:
1703 	case SO_REUSEPORT:
1704 	case SO_OOBINLINE:
1705 	case SO_TIMESTAMP:
1706 	case SO_NOSIGPIPE:
1707 #ifdef SO_OTIMESTAMP
1708 	case SO_OTIMESTAMP:
1709 #endif
1710 		error = sockopt_getint(sopt, &optval);
1711 		solock(so);
1712 		if (error)
1713 			break;
1714 		if (optval)
1715 			so->so_options |= opt;
1716 		else
1717 			so->so_options &= ~opt;
1718 		break;
1719 
1720 	case SO_SNDBUF:
1721 	case SO_RCVBUF:
1722 	case SO_SNDLOWAT:
1723 	case SO_RCVLOWAT:
1724 		error = sockopt_getint(sopt, &optval);
1725 		solock(so);
1726 		if (error)
1727 			break;
1728 
1729 		/*
1730 		 * Values < 1 make no sense for any of these
1731 		 * options, so disallow them.
1732 		 */
1733 		if (optval < 1) {
1734 			error = EINVAL;
1735 			break;
1736 		}
1737 
1738 		switch (opt) {
1739 		case SO_SNDBUF:
1740 			if (sbreserve(&so->so_snd, (u_long)optval, so) == 0) {
1741 				error = ENOBUFS;
1742 				break;
1743 			}
1744 			so->so_snd.sb_flags &= ~SB_AUTOSIZE;
1745 			break;
1746 
1747 		case SO_RCVBUF:
1748 			if (sbreserve(&so->so_rcv, (u_long)optval, so) == 0) {
1749 				error = ENOBUFS;
1750 				break;
1751 			}
1752 			so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1753 			break;
1754 
1755 		/*
1756 		 * Make sure the low-water is never greater than
1757 		 * the high-water.
1758 		 */
1759 		case SO_SNDLOWAT:
1760 			if (optval > so->so_snd.sb_hiwat)
1761 				optval = so->so_snd.sb_hiwat;
1762 
1763 			so->so_snd.sb_lowat = optval;
1764 			break;
1765 
1766 		case SO_RCVLOWAT:
1767 			if (optval > so->so_rcv.sb_hiwat)
1768 				optval = so->so_rcv.sb_hiwat;
1769 
1770 			so->so_rcv.sb_lowat = optval;
1771 			break;
1772 		}
1773 		break;
1774 
1775 #ifdef COMPAT_50
1776 	case SO_OSNDTIMEO:
1777 	case SO_ORCVTIMEO: {
1778 		struct timeval50 otv;
1779 		error = sockopt_get(sopt, &otv, sizeof(otv));
1780 		if (error) {
1781 			solock(so);
1782 			break;
1783 		}
1784 		timeval50_to_timeval(&otv, &tv);
1785 		opt = opt == SO_OSNDTIMEO ? SO_SNDTIMEO : SO_RCVTIMEO;
1786 		error = 0;
1787 		/*FALLTHROUGH*/
1788 	}
1789 #endif /* COMPAT_50 */
1790 
1791 	case SO_SNDTIMEO:
1792 	case SO_RCVTIMEO:
1793 		if (error)
1794 			error = sockopt_get(sopt, &tv, sizeof(tv));
1795 		solock(so);
1796 		if (error)
1797 			break;
1798 
1799 		if (tv.tv_sec > (INT_MAX - tv.tv_usec / tick) / hz) {
1800 			error = EDOM;
1801 			break;
1802 		}
1803 
1804 		optval = tv.tv_sec * hz + tv.tv_usec / tick;
1805 		if (optval == 0 && tv.tv_usec != 0)
1806 			optval = 1;
1807 
1808 		switch (opt) {
1809 		case SO_SNDTIMEO:
1810 			so->so_snd.sb_timeo = optval;
1811 			break;
1812 		case SO_RCVTIMEO:
1813 			so->so_rcv.sb_timeo = optval;
1814 			break;
1815 		}
1816 		break;
1817 
1818 	default:
1819 		solock(so);
1820 		error = ENOPROTOOPT;
1821 		break;
1822 	}
1823 	KASSERT(solocked(so));
1824 	return error;
1825 }
1826 
1827 int
1828 sosetopt(struct socket *so, struct sockopt *sopt)
1829 {
1830 	int error, prerr;
1831 
1832 	if (sopt->sopt_level == SOL_SOCKET) {
1833 		error = sosetopt1(so, sopt);
1834 		KASSERT(solocked(so));
1835 	} else {
1836 		error = ENOPROTOOPT;
1837 		solock(so);
1838 	}
1839 
1840 	if ((error == 0 || error == ENOPROTOOPT) &&
1841 	    so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) {
1842 		/* give the protocol stack a shot */
1843 		prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, sopt);
1844 		if (prerr == 0)
1845 			error = 0;
1846 		else if (prerr != ENOPROTOOPT)
1847 			error = prerr;
1848 	}
1849 	sounlock(so);
1850 	return error;
1851 }
1852 
1853 /*
1854  * so_setsockopt() is a wrapper providing a sockopt structure for sosetopt()
1855  */
1856 int
1857 so_setsockopt(struct lwp *l, struct socket *so, int level, int name,
1858     const void *val, size_t valsize)
1859 {
1860 	struct sockopt sopt;
1861 	int error;
1862 
1863 	KASSERT(valsize == 0 || val != NULL);
1864 
1865 	sockopt_init(&sopt, level, name, valsize);
1866 	sockopt_set(&sopt, val, valsize);
1867 
1868 	error = sosetopt(so, &sopt);
1869 
1870 	sockopt_destroy(&sopt);
1871 
1872 	return error;
1873 }
1874 
1875 /*
1876  * internal get SOL_SOCKET options
1877  */
1878 static int
1879 sogetopt1(struct socket *so, struct sockopt *sopt)
1880 {
1881 	int error, optval, opt;
1882 	struct linger l;
1883 	struct timeval tv;
1884 
1885 	switch ((opt = sopt->sopt_name)) {
1886 
1887 	case SO_ACCEPTFILTER:
1888 		error = accept_filt_getopt(so, sopt);
1889 		break;
1890 
1891 	case SO_LINGER:
1892 		l.l_onoff = (so->so_options & SO_LINGER) ? 1 : 0;
1893 		l.l_linger = so->so_linger;
1894 
1895 		error = sockopt_set(sopt, &l, sizeof(l));
1896 		break;
1897 
1898 	case SO_USELOOPBACK:
1899 	case SO_DONTROUTE:
1900 	case SO_DEBUG:
1901 	case SO_KEEPALIVE:
1902 	case SO_REUSEADDR:
1903 	case SO_REUSEPORT:
1904 	case SO_BROADCAST:
1905 	case SO_OOBINLINE:
1906 	case SO_TIMESTAMP:
1907 	case SO_NOSIGPIPE:
1908 #ifdef SO_OTIMESTAMP
1909 	case SO_OTIMESTAMP:
1910 #endif
1911 	case SO_ACCEPTCONN:
1912 		error = sockopt_setint(sopt, (so->so_options & opt) ? 1 : 0);
1913 		break;
1914 
1915 	case SO_TYPE:
1916 		error = sockopt_setint(sopt, so->so_type);
1917 		break;
1918 
1919 	case SO_ERROR:
1920 		error = sockopt_setint(sopt, so->so_error);
1921 		so->so_error = 0;
1922 		break;
1923 
1924 	case SO_SNDBUF:
1925 		error = sockopt_setint(sopt, so->so_snd.sb_hiwat);
1926 		break;
1927 
1928 	case SO_RCVBUF:
1929 		error = sockopt_setint(sopt, so->so_rcv.sb_hiwat);
1930 		break;
1931 
1932 	case SO_SNDLOWAT:
1933 		error = sockopt_setint(sopt, so->so_snd.sb_lowat);
1934 		break;
1935 
1936 	case SO_RCVLOWAT:
1937 		error = sockopt_setint(sopt, so->so_rcv.sb_lowat);
1938 		break;
1939 
1940 #ifdef COMPAT_50
1941 	case SO_OSNDTIMEO:
1942 	case SO_ORCVTIMEO: {
1943 		struct timeval50 otv;
1944 
1945 		optval = (opt == SO_OSNDTIMEO ?
1946 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1947 
1948 		otv.tv_sec = optval / hz;
1949 		otv.tv_usec = (optval % hz) * tick;
1950 
1951 		error = sockopt_set(sopt, &otv, sizeof(otv));
1952 		break;
1953 	}
1954 #endif /* COMPAT_50 */
1955 
1956 	case SO_SNDTIMEO:
1957 	case SO_RCVTIMEO:
1958 		optval = (opt == SO_SNDTIMEO ?
1959 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1960 
1961 		tv.tv_sec = optval / hz;
1962 		tv.tv_usec = (optval % hz) * tick;
1963 
1964 		error = sockopt_set(sopt, &tv, sizeof(tv));
1965 		break;
1966 
1967 	case SO_OVERFLOWED:
1968 		error = sockopt_setint(sopt, so->so_rcv.sb_overflowed);
1969 		break;
1970 
1971 	default:
1972 		error = ENOPROTOOPT;
1973 		break;
1974 	}
1975 
1976 	return (error);
1977 }
1978 
1979 int
1980 sogetopt(struct socket *so, struct sockopt *sopt)
1981 {
1982 	int		error;
1983 
1984 	solock(so);
1985 	if (sopt->sopt_level != SOL_SOCKET) {
1986 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1987 			error = ((*so->so_proto->pr_ctloutput)
1988 			    (PRCO_GETOPT, so, sopt));
1989 		} else
1990 			error = (ENOPROTOOPT);
1991 	} else {
1992 		error = sogetopt1(so, sopt);
1993 	}
1994 	sounlock(so);
1995 	return (error);
1996 }
1997 
1998 /*
1999  * alloc sockopt data buffer buffer
2000  *	- will be released at destroy
2001  */
2002 static int
2003 sockopt_alloc(struct sockopt *sopt, size_t len, km_flag_t kmflag)
2004 {
2005 
2006 	KASSERT(sopt->sopt_size == 0);
2007 
2008 	if (len > sizeof(sopt->sopt_buf)) {
2009 		sopt->sopt_data = kmem_zalloc(len, kmflag);
2010 		if (sopt->sopt_data == NULL)
2011 			return ENOMEM;
2012 	} else
2013 		sopt->sopt_data = sopt->sopt_buf;
2014 
2015 	sopt->sopt_size = len;
2016 	return 0;
2017 }
2018 
2019 /*
2020  * initialise sockopt storage
2021  *	- MAY sleep during allocation
2022  */
2023 void
2024 sockopt_init(struct sockopt *sopt, int level, int name, size_t size)
2025 {
2026 
2027 	memset(sopt, 0, sizeof(*sopt));
2028 
2029 	sopt->sopt_level = level;
2030 	sopt->sopt_name = name;
2031 	(void)sockopt_alloc(sopt, size, KM_SLEEP);
2032 }
2033 
2034 /*
2035  * destroy sockopt storage
2036  *	- will release any held memory references
2037  */
2038 void
2039 sockopt_destroy(struct sockopt *sopt)
2040 {
2041 
2042 	if (sopt->sopt_data != sopt->sopt_buf)
2043 		kmem_free(sopt->sopt_data, sopt->sopt_size);
2044 
2045 	memset(sopt, 0, sizeof(*sopt));
2046 }
2047 
2048 /*
2049  * set sockopt value
2050  *	- value is copied into sockopt
2051  * 	- memory is allocated when necessary, will not sleep
2052  */
2053 int
2054 sockopt_set(struct sockopt *sopt, const void *buf, size_t len)
2055 {
2056 	int error;
2057 
2058 	if (sopt->sopt_size == 0) {
2059 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2060 		if (error)
2061 			return error;
2062 	}
2063 
2064 	KASSERT(sopt->sopt_size == len);
2065 	memcpy(sopt->sopt_data, buf, len);
2066 	return 0;
2067 }
2068 
2069 /*
2070  * common case of set sockopt integer value
2071  */
2072 int
2073 sockopt_setint(struct sockopt *sopt, int val)
2074 {
2075 
2076 	return sockopt_set(sopt, &val, sizeof(int));
2077 }
2078 
2079 /*
2080  * get sockopt value
2081  *	- correct size must be given
2082  */
2083 int
2084 sockopt_get(const struct sockopt *sopt, void *buf, size_t len)
2085 {
2086 
2087 	if (sopt->sopt_size != len)
2088 		return EINVAL;
2089 
2090 	memcpy(buf, sopt->sopt_data, len);
2091 	return 0;
2092 }
2093 
2094 /*
2095  * common case of get sockopt integer value
2096  */
2097 int
2098 sockopt_getint(const struct sockopt *sopt, int *valp)
2099 {
2100 
2101 	return sockopt_get(sopt, valp, sizeof(int));
2102 }
2103 
2104 /*
2105  * set sockopt value from mbuf
2106  *	- ONLY for legacy code
2107  *	- mbuf is released by sockopt
2108  *	- will not sleep
2109  */
2110 int
2111 sockopt_setmbuf(struct sockopt *sopt, struct mbuf *m)
2112 {
2113 	size_t len;
2114 	int error;
2115 
2116 	len = m_length(m);
2117 
2118 	if (sopt->sopt_size == 0) {
2119 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2120 		if (error)
2121 			return error;
2122 	}
2123 
2124 	KASSERT(sopt->sopt_size == len);
2125 	m_copydata(m, 0, len, sopt->sopt_data);
2126 	m_freem(m);
2127 
2128 	return 0;
2129 }
2130 
2131 /*
2132  * get sockopt value into mbuf
2133  *	- ONLY for legacy code
2134  *	- mbuf to be released by the caller
2135  *	- will not sleep
2136  */
2137 struct mbuf *
2138 sockopt_getmbuf(const struct sockopt *sopt)
2139 {
2140 	struct mbuf *m;
2141 
2142 	if (sopt->sopt_size > MCLBYTES)
2143 		return NULL;
2144 
2145 	m = m_get(M_DONTWAIT, MT_SOOPTS);
2146 	if (m == NULL)
2147 		return NULL;
2148 
2149 	if (sopt->sopt_size > MLEN) {
2150 		MCLGET(m, M_DONTWAIT);
2151 		if ((m->m_flags & M_EXT) == 0) {
2152 			m_free(m);
2153 			return NULL;
2154 		}
2155 	}
2156 
2157 	memcpy(mtod(m, void *), sopt->sopt_data, sopt->sopt_size);
2158 	m->m_len = sopt->sopt_size;
2159 
2160 	return m;
2161 }
2162 
2163 void
2164 sohasoutofband(struct socket *so)
2165 {
2166 
2167 	fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
2168 	selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, NOTE_SUBMIT);
2169 }
2170 
2171 static void
2172 filt_sordetach(struct knote *kn)
2173 {
2174 	struct socket	*so;
2175 
2176 	so = ((file_t *)kn->kn_obj)->f_socket;
2177 	solock(so);
2178 	SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
2179 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
2180 		so->so_rcv.sb_flags &= ~SB_KNOTE;
2181 	sounlock(so);
2182 }
2183 
2184 /*ARGSUSED*/
2185 static int
2186 filt_soread(struct knote *kn, long hint)
2187 {
2188 	struct socket	*so;
2189 	int rv;
2190 
2191 	so = ((file_t *)kn->kn_obj)->f_socket;
2192 	if (hint != NOTE_SUBMIT)
2193 		solock(so);
2194 	kn->kn_data = so->so_rcv.sb_cc;
2195 	if (so->so_state & SS_CANTRCVMORE) {
2196 		kn->kn_flags |= EV_EOF;
2197 		kn->kn_fflags = so->so_error;
2198 		rv = 1;
2199 	} else if (so->so_error)	/* temporary udp error */
2200 		rv = 1;
2201 	else if (kn->kn_sfflags & NOTE_LOWAT)
2202 		rv = (kn->kn_data >= kn->kn_sdata);
2203 	else
2204 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
2205 	if (hint != NOTE_SUBMIT)
2206 		sounlock(so);
2207 	return rv;
2208 }
2209 
2210 static void
2211 filt_sowdetach(struct knote *kn)
2212 {
2213 	struct socket	*so;
2214 
2215 	so = ((file_t *)kn->kn_obj)->f_socket;
2216 	solock(so);
2217 	SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
2218 	if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
2219 		so->so_snd.sb_flags &= ~SB_KNOTE;
2220 	sounlock(so);
2221 }
2222 
2223 /*ARGSUSED*/
2224 static int
2225 filt_sowrite(struct knote *kn, long hint)
2226 {
2227 	struct socket	*so;
2228 	int rv;
2229 
2230 	so = ((file_t *)kn->kn_obj)->f_socket;
2231 	if (hint != NOTE_SUBMIT)
2232 		solock(so);
2233 	kn->kn_data = sbspace(&so->so_snd);
2234 	if (so->so_state & SS_CANTSENDMORE) {
2235 		kn->kn_flags |= EV_EOF;
2236 		kn->kn_fflags = so->so_error;
2237 		rv = 1;
2238 	} else if (so->so_error)	/* temporary udp error */
2239 		rv = 1;
2240 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2241 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
2242 		rv = 0;
2243 	else if (kn->kn_sfflags & NOTE_LOWAT)
2244 		rv = (kn->kn_data >= kn->kn_sdata);
2245 	else
2246 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
2247 	if (hint != NOTE_SUBMIT)
2248 		sounlock(so);
2249 	return rv;
2250 }
2251 
2252 /*ARGSUSED*/
2253 static int
2254 filt_solisten(struct knote *kn, long hint)
2255 {
2256 	struct socket	*so;
2257 	int rv;
2258 
2259 	so = ((file_t *)kn->kn_obj)->f_socket;
2260 
2261 	/*
2262 	 * Set kn_data to number of incoming connections, not
2263 	 * counting partial (incomplete) connections.
2264 	 */
2265 	if (hint != NOTE_SUBMIT)
2266 		solock(so);
2267 	kn->kn_data = so->so_qlen;
2268 	rv = (kn->kn_data > 0);
2269 	if (hint != NOTE_SUBMIT)
2270 		sounlock(so);
2271 	return rv;
2272 }
2273 
2274 static const struct filterops solisten_filtops =
2275 	{ 1, NULL, filt_sordetach, filt_solisten };
2276 static const struct filterops soread_filtops =
2277 	{ 1, NULL, filt_sordetach, filt_soread };
2278 static const struct filterops sowrite_filtops =
2279 	{ 1, NULL, filt_sowdetach, filt_sowrite };
2280 
2281 int
2282 soo_kqfilter(struct file *fp, struct knote *kn)
2283 {
2284 	struct socket	*so;
2285 	struct sockbuf	*sb;
2286 
2287 	so = ((file_t *)kn->kn_obj)->f_socket;
2288 	solock(so);
2289 	switch (kn->kn_filter) {
2290 	case EVFILT_READ:
2291 		if (so->so_options & SO_ACCEPTCONN)
2292 			kn->kn_fop = &solisten_filtops;
2293 		else
2294 			kn->kn_fop = &soread_filtops;
2295 		sb = &so->so_rcv;
2296 		break;
2297 	case EVFILT_WRITE:
2298 		kn->kn_fop = &sowrite_filtops;
2299 		sb = &so->so_snd;
2300 		break;
2301 	default:
2302 		sounlock(so);
2303 		return (EINVAL);
2304 	}
2305 	SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
2306 	sb->sb_flags |= SB_KNOTE;
2307 	sounlock(so);
2308 	return (0);
2309 }
2310 
2311 static int
2312 sodopoll(struct socket *so, int events)
2313 {
2314 	int revents;
2315 
2316 	revents = 0;
2317 
2318 	if (events & (POLLIN | POLLRDNORM))
2319 		if (soreadable(so))
2320 			revents |= events & (POLLIN | POLLRDNORM);
2321 
2322 	if (events & (POLLOUT | POLLWRNORM))
2323 		if (sowritable(so))
2324 			revents |= events & (POLLOUT | POLLWRNORM);
2325 
2326 	if (events & (POLLPRI | POLLRDBAND))
2327 		if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
2328 			revents |= events & (POLLPRI | POLLRDBAND);
2329 
2330 	return revents;
2331 }
2332 
2333 int
2334 sopoll(struct socket *so, int events)
2335 {
2336 	int revents = 0;
2337 
2338 #ifndef DIAGNOSTIC
2339 	/*
2340 	 * Do a quick, unlocked check in expectation that the socket
2341 	 * will be ready for I/O.  Don't do this check if DIAGNOSTIC,
2342 	 * as the solocked() assertions will fail.
2343 	 */
2344 	if ((revents = sodopoll(so, events)) != 0)
2345 		return revents;
2346 #endif
2347 
2348 	solock(so);
2349 	if ((revents = sodopoll(so, events)) == 0) {
2350 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2351 			selrecord(curlwp, &so->so_rcv.sb_sel);
2352 			so->so_rcv.sb_flags |= SB_NOTIFY;
2353 		}
2354 
2355 		if (events & (POLLOUT | POLLWRNORM)) {
2356 			selrecord(curlwp, &so->so_snd.sb_sel);
2357 			so->so_snd.sb_flags |= SB_NOTIFY;
2358 		}
2359 	}
2360 	sounlock(so);
2361 
2362 	return revents;
2363 }
2364 
2365 
2366 #include <sys/sysctl.h>
2367 
2368 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO);
2369 static int sysctl_kern_sbmax(SYSCTLFN_PROTO);
2370 
2371 /*
2372  * sysctl helper routine for kern.somaxkva.  ensures that the given
2373  * value is not too small.
2374  * (XXX should we maybe make sure it's not too large as well?)
2375  */
2376 static int
2377 sysctl_kern_somaxkva(SYSCTLFN_ARGS)
2378 {
2379 	int error, new_somaxkva;
2380 	struct sysctlnode node;
2381 
2382 	new_somaxkva = somaxkva;
2383 	node = *rnode;
2384 	node.sysctl_data = &new_somaxkva;
2385 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2386 	if (error || newp == NULL)
2387 		return (error);
2388 
2389 	if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */
2390 		return (EINVAL);
2391 
2392 	mutex_enter(&so_pendfree_lock);
2393 	somaxkva = new_somaxkva;
2394 	cv_broadcast(&socurkva_cv);
2395 	mutex_exit(&so_pendfree_lock);
2396 
2397 	return (error);
2398 }
2399 
2400 /*
2401  * sysctl helper routine for kern.sbmax. Basically just ensures that
2402  * any new value is not too small.
2403  */
2404 static int
2405 sysctl_kern_sbmax(SYSCTLFN_ARGS)
2406 {
2407 	int error, new_sbmax;
2408 	struct sysctlnode node;
2409 
2410 	new_sbmax = sb_max;
2411 	node = *rnode;
2412 	node.sysctl_data = &new_sbmax;
2413 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2414 	if (error || newp == NULL)
2415 		return (error);
2416 
2417 	KERNEL_LOCK(1, NULL);
2418 	error = sb_max_set(new_sbmax);
2419 	KERNEL_UNLOCK_ONE(NULL);
2420 
2421 	return (error);
2422 }
2423 
2424 static void
2425 sysctl_kern_socket_setup(void)
2426 {
2427 
2428 	KASSERT(socket_sysctllog == NULL);
2429 
2430 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2431 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2432 		       CTLTYPE_INT, "somaxkva",
2433 		       SYSCTL_DESCR("Maximum amount of kernel memory to be "
2434 				    "used for socket buffers"),
2435 		       sysctl_kern_somaxkva, 0, NULL, 0,
2436 		       CTL_KERN, KERN_SOMAXKVA, CTL_EOL);
2437 
2438 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2439 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2440 		       CTLTYPE_INT, "sbmax",
2441 		       SYSCTL_DESCR("Maximum socket buffer size"),
2442 		       sysctl_kern_sbmax, 0, NULL, 0,
2443 		       CTL_KERN, KERN_SBMAX, CTL_EOL);
2444 }
2445