xref: /netbsd-src/sys/kern/uipc_socket.c (revision 62f324d0121177eaf2e0384f92fd9ca2a751c795)
1 /*	$NetBSD: uipc_socket.c,v 1.215 2013/04/08 21:12:33 skrll Exp $	*/
2 
3 /*-
4  * Copyright (c) 2002, 2007, 2008, 2009 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe of Wasabi Systems, Inc, and by Andrew Doran.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Copyright (c) 2004 The FreeBSD Foundation
34  * Copyright (c) 2004 Robert Watson
35  * Copyright (c) 1982, 1986, 1988, 1990, 1993
36  *	The Regents of the University of California.  All rights reserved.
37  *
38  * Redistribution and use in source and binary forms, with or without
39  * modification, are permitted provided that the following conditions
40  * are met:
41  * 1. Redistributions of source code must retain the above copyright
42  *    notice, this list of conditions and the following disclaimer.
43  * 2. Redistributions in binary form must reproduce the above copyright
44  *    notice, this list of conditions and the following disclaimer in the
45  *    documentation and/or other materials provided with the distribution.
46  * 3. Neither the name of the University nor the names of its contributors
47  *    may be used to endorse or promote products derived from this software
48  *    without specific prior written permission.
49  *
50  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
52  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
53  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
54  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
55  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
56  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
57  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
58  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
59  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
60  * SUCH DAMAGE.
61  *
62  *	@(#)uipc_socket.c	8.6 (Berkeley) 5/2/95
63  */
64 
65 #include <sys/cdefs.h>
66 __KERNEL_RCSID(0, "$NetBSD: uipc_socket.c,v 1.215 2013/04/08 21:12:33 skrll Exp $");
67 
68 #include "opt_compat_netbsd.h"
69 #include "opt_sock_counters.h"
70 #include "opt_sosend_loan.h"
71 #include "opt_mbuftrace.h"
72 #include "opt_somaxkva.h"
73 #include "opt_multiprocessor.h"	/* XXX */
74 
75 #include <sys/param.h>
76 #include <sys/systm.h>
77 #include <sys/proc.h>
78 #include <sys/file.h>
79 #include <sys/filedesc.h>
80 #include <sys/kmem.h>
81 #include <sys/mbuf.h>
82 #include <sys/domain.h>
83 #include <sys/kernel.h>
84 #include <sys/protosw.h>
85 #include <sys/socket.h>
86 #include <sys/socketvar.h>
87 #include <sys/signalvar.h>
88 #include <sys/resourcevar.h>
89 #include <sys/uidinfo.h>
90 #include <sys/event.h>
91 #include <sys/poll.h>
92 #include <sys/kauth.h>
93 #include <sys/mutex.h>
94 #include <sys/condvar.h>
95 #include <sys/kthread.h>
96 
97 #ifdef COMPAT_50
98 #include <compat/sys/time.h>
99 #include <compat/sys/socket.h>
100 #endif
101 
102 #include <uvm/uvm_extern.h>
103 #include <uvm/uvm_loan.h>
104 #include <uvm/uvm_page.h>
105 
106 MALLOC_DEFINE(M_SOOPTS, "soopts", "socket options");
107 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
108 
109 extern const struct fileops socketops;
110 
111 extern int	somaxconn;			/* patchable (XXX sysctl) */
112 int		somaxconn = SOMAXCONN;
113 kmutex_t	*softnet_lock;
114 
115 #ifdef SOSEND_COUNTERS
116 #include <sys/device.h>
117 
118 static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
119     NULL, "sosend", "loan big");
120 static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
121     NULL, "sosend", "copy big");
122 static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
123     NULL, "sosend", "copy small");
124 static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
125     NULL, "sosend", "kva limit");
126 
127 #define	SOSEND_COUNTER_INCR(ev)		(ev)->ev_count++
128 
129 EVCNT_ATTACH_STATIC(sosend_loan_big);
130 EVCNT_ATTACH_STATIC(sosend_copy_big);
131 EVCNT_ATTACH_STATIC(sosend_copy_small);
132 EVCNT_ATTACH_STATIC(sosend_kvalimit);
133 #else
134 
135 #define	SOSEND_COUNTER_INCR(ev)		/* nothing */
136 
137 #endif /* SOSEND_COUNTERS */
138 
139 #if defined(SOSEND_NO_LOAN) || defined(MULTIPROCESSOR)
140 int sock_loan_thresh = -1;
141 #else
142 int sock_loan_thresh = 4096;
143 #endif
144 
145 static kmutex_t so_pendfree_lock;
146 static struct mbuf *so_pendfree = NULL;
147 
148 #ifndef SOMAXKVA
149 #define	SOMAXKVA (16 * 1024 * 1024)
150 #endif
151 int somaxkva = SOMAXKVA;
152 static int socurkva;
153 static kcondvar_t socurkva_cv;
154 
155 static kauth_listener_t socket_listener;
156 
157 #define	SOCK_LOAN_CHUNK		65536
158 
159 static void sopendfree_thread(void *);
160 static kcondvar_t pendfree_thread_cv;
161 static lwp_t *sopendfree_lwp;
162 
163 static void sysctl_kern_socket_setup(void);
164 static struct sysctllog *socket_sysctllog;
165 
166 static vsize_t
167 sokvareserve(struct socket *so, vsize_t len)
168 {
169 	int error;
170 
171 	mutex_enter(&so_pendfree_lock);
172 	while (socurkva + len > somaxkva) {
173 		SOSEND_COUNTER_INCR(&sosend_kvalimit);
174 		error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock);
175 		if (error) {
176 			len = 0;
177 			break;
178 		}
179 	}
180 	socurkva += len;
181 	mutex_exit(&so_pendfree_lock);
182 	return len;
183 }
184 
185 static void
186 sokvaunreserve(vsize_t len)
187 {
188 
189 	mutex_enter(&so_pendfree_lock);
190 	socurkva -= len;
191 	cv_broadcast(&socurkva_cv);
192 	mutex_exit(&so_pendfree_lock);
193 }
194 
195 /*
196  * sokvaalloc: allocate kva for loan.
197  */
198 
199 vaddr_t
200 sokvaalloc(vaddr_t sva, vsize_t len, struct socket *so)
201 {
202 	vaddr_t lva;
203 
204 	/*
205 	 * reserve kva.
206 	 */
207 
208 	if (sokvareserve(so, len) == 0)
209 		return 0;
210 
211 	/*
212 	 * allocate kva.
213 	 */
214 
215 	lva = uvm_km_alloc(kernel_map, len, atop(sva) & uvmexp.colormask,
216 	    UVM_KMF_COLORMATCH | UVM_KMF_VAONLY | UVM_KMF_WAITVA);
217 	if (lva == 0) {
218 		sokvaunreserve(len);
219 		return (0);
220 	}
221 
222 	return lva;
223 }
224 
225 /*
226  * sokvafree: free kva for loan.
227  */
228 
229 void
230 sokvafree(vaddr_t sva, vsize_t len)
231 {
232 
233 	/*
234 	 * free kva.
235 	 */
236 
237 	uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY);
238 
239 	/*
240 	 * unreserve kva.
241 	 */
242 
243 	sokvaunreserve(len);
244 }
245 
246 static void
247 sodoloanfree(struct vm_page **pgs, void *buf, size_t size)
248 {
249 	vaddr_t sva, eva;
250 	vsize_t len;
251 	int npgs;
252 
253 	KASSERT(pgs != NULL);
254 
255 	eva = round_page((vaddr_t) buf + size);
256 	sva = trunc_page((vaddr_t) buf);
257 	len = eva - sva;
258 	npgs = len >> PAGE_SHIFT;
259 
260 	pmap_kremove(sva, len);
261 	pmap_update(pmap_kernel());
262 	uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE);
263 	sokvafree(sva, len);
264 }
265 
266 /*
267  * sopendfree_thread: free mbufs on "pendfree" list.
268  * unlock and relock so_pendfree_lock when freeing mbufs.
269  */
270 
271 static void
272 sopendfree_thread(void *v)
273 {
274 	struct mbuf *m, *next;
275 	size_t rv;
276 
277 	mutex_enter(&so_pendfree_lock);
278 
279 	for (;;) {
280 		rv = 0;
281 		while (so_pendfree != NULL) {
282 			m = so_pendfree;
283 			so_pendfree = NULL;
284 			mutex_exit(&so_pendfree_lock);
285 
286 			for (; m != NULL; m = next) {
287 				next = m->m_next;
288 				KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0);
289 				KASSERT(m->m_ext.ext_refcnt == 0);
290 
291 				rv += m->m_ext.ext_size;
292 				sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf,
293 				    m->m_ext.ext_size);
294 				pool_cache_put(mb_cache, m);
295 			}
296 
297 			mutex_enter(&so_pendfree_lock);
298 		}
299 		if (rv)
300 			cv_broadcast(&socurkva_cv);
301 		cv_wait(&pendfree_thread_cv, &so_pendfree_lock);
302 	}
303 	panic("sopendfree_thread");
304 	/* NOTREACHED */
305 }
306 
307 void
308 soloanfree(struct mbuf *m, void *buf, size_t size, void *arg)
309 {
310 
311 	KASSERT(m != NULL);
312 
313 	/*
314 	 * postpone freeing mbuf.
315 	 *
316 	 * we can't do it in interrupt context
317 	 * because we need to put kva back to kernel_map.
318 	 */
319 
320 	mutex_enter(&so_pendfree_lock);
321 	m->m_next = so_pendfree;
322 	so_pendfree = m;
323 	cv_signal(&pendfree_thread_cv);
324 	mutex_exit(&so_pendfree_lock);
325 }
326 
327 static long
328 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space)
329 {
330 	struct iovec *iov = uio->uio_iov;
331 	vaddr_t sva, eva;
332 	vsize_t len;
333 	vaddr_t lva;
334 	int npgs, error;
335 	vaddr_t va;
336 	int i;
337 
338 	if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace))
339 		return (0);
340 
341 	if (iov->iov_len < (size_t) space)
342 		space = iov->iov_len;
343 	if (space > SOCK_LOAN_CHUNK)
344 		space = SOCK_LOAN_CHUNK;
345 
346 	eva = round_page((vaddr_t) iov->iov_base + space);
347 	sva = trunc_page((vaddr_t) iov->iov_base);
348 	len = eva - sva;
349 	npgs = len >> PAGE_SHIFT;
350 
351 	KASSERT(npgs <= M_EXT_MAXPAGES);
352 
353 	lva = sokvaalloc(sva, len, so);
354 	if (lva == 0)
355 		return 0;
356 
357 	error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len,
358 	    m->m_ext.ext_pgs, UVM_LOAN_TOPAGE);
359 	if (error) {
360 		sokvafree(lva, len);
361 		return (0);
362 	}
363 
364 	for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE)
365 		pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
366 		    VM_PROT_READ, 0);
367 	pmap_update(pmap_kernel());
368 
369 	lva += (vaddr_t) iov->iov_base & PAGE_MASK;
370 
371 	MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so);
372 	m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
373 
374 	uio->uio_resid -= space;
375 	/* uio_offset not updated, not set/used for write(2) */
376 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space;
377 	uio->uio_iov->iov_len -= space;
378 	if (uio->uio_iov->iov_len == 0) {
379 		uio->uio_iov++;
380 		uio->uio_iovcnt--;
381 	}
382 
383 	return (space);
384 }
385 
386 struct mbuf *
387 getsombuf(struct socket *so, int type)
388 {
389 	struct mbuf *m;
390 
391 	m = m_get(M_WAIT, type);
392 	MCLAIM(m, so->so_mowner);
393 	return m;
394 }
395 
396 static int
397 socket_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
398     void *arg0, void *arg1, void *arg2, void *arg3)
399 {
400 	int result;
401 	enum kauth_network_req req;
402 
403 	result = KAUTH_RESULT_DEFER;
404 	req = (enum kauth_network_req)arg0;
405 
406 	if ((action != KAUTH_NETWORK_SOCKET) &&
407 	    (action != KAUTH_NETWORK_BIND))
408 		return result;
409 
410 	switch (req) {
411 	case KAUTH_REQ_NETWORK_BIND_PORT:
412 		result = KAUTH_RESULT_ALLOW;
413 		break;
414 
415 	case KAUTH_REQ_NETWORK_SOCKET_DROP: {
416 		/* Normal users can only drop their own connections. */
417 		struct socket *so = (struct socket *)arg1;
418 
419 		if (proc_uidmatch(cred, so->so_cred))
420 			result = KAUTH_RESULT_ALLOW;
421 
422 		break;
423 		}
424 
425 	case KAUTH_REQ_NETWORK_SOCKET_OPEN:
426 		/* We allow "raw" routing/bluetooth sockets to anyone. */
427 		if ((u_long)arg1 == PF_ROUTE || (u_long)arg1 == PF_OROUTE
428 		    || (u_long)arg1 == PF_BLUETOOTH) {
429 			result = KAUTH_RESULT_ALLOW;
430 		} else {
431 			/* Privileged, let secmodel handle this. */
432 			if ((u_long)arg2 == SOCK_RAW)
433 				break;
434 		}
435 
436 		result = KAUTH_RESULT_ALLOW;
437 
438 		break;
439 
440 	case KAUTH_REQ_NETWORK_SOCKET_CANSEE:
441 		result = KAUTH_RESULT_ALLOW;
442 
443 		break;
444 
445 	default:
446 		break;
447 	}
448 
449 	return result;
450 }
451 
452 void
453 soinit(void)
454 {
455 
456 	sysctl_kern_socket_setup();
457 
458 	mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM);
459 	softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
460 	cv_init(&socurkva_cv, "sokva");
461 	cv_init(&pendfree_thread_cv, "sopendfr");
462 	soinit2();
463 
464 	/* Set the initial adjusted socket buffer size. */
465 	if (sb_max_set(sb_max))
466 		panic("bad initial sb_max value: %lu", sb_max);
467 
468 	socket_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK,
469 	    socket_listener_cb, NULL);
470 }
471 
472 void
473 soinit1(void)
474 {
475 	int error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL,
476 	    sopendfree_thread, NULL, &sopendfree_lwp, "sopendfree");
477 	if (error)
478 		panic("soinit1 %d", error);
479 }
480 
481 /*
482  * Socket operation routines.
483  * These routines are called by the routines in
484  * sys_socket.c or from a system process, and
485  * implement the semantics of socket operations by
486  * switching out to the protocol specific routines.
487  */
488 /*ARGSUSED*/
489 int
490 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l,
491 	 struct socket *lockso)
492 {
493 	const struct protosw	*prp;
494 	struct socket	*so;
495 	uid_t		uid;
496 	int		error;
497 	kmutex_t	*lock;
498 
499 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET,
500 	    KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type),
501 	    KAUTH_ARG(proto));
502 	if (error != 0)
503 		return error;
504 
505 	if (proto)
506 		prp = pffindproto(dom, proto, type);
507 	else
508 		prp = pffindtype(dom, type);
509 	if (prp == NULL) {
510 		/* no support for domain */
511 		if (pffinddomain(dom) == 0)
512 			return EAFNOSUPPORT;
513 		/* no support for socket type */
514 		if (proto == 0 && type != 0)
515 			return EPROTOTYPE;
516 		return EPROTONOSUPPORT;
517 	}
518 	if (prp->pr_usrreq == NULL)
519 		return EPROTONOSUPPORT;
520 	if (prp->pr_type != type)
521 		return EPROTOTYPE;
522 
523 	so = soget(true);
524 	so->so_type = type;
525 	so->so_proto = prp;
526 	so->so_send = sosend;
527 	so->so_receive = soreceive;
528 #ifdef MBUFTRACE
529 	so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner;
530 	so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner;
531 	so->so_mowner = &prp->pr_domain->dom_mowner;
532 #endif
533 	uid = kauth_cred_geteuid(l->l_cred);
534 	so->so_uidinfo = uid_find(uid);
535 	so->so_cpid = l->l_proc->p_pid;
536 	if (lockso != NULL) {
537 		/* Caller wants us to share a lock. */
538 		lock = lockso->so_lock;
539 		so->so_lock = lock;
540 		mutex_obj_hold(lock);
541 		/* XXX Why is this not solock, to match sounlock? */
542 		mutex_enter(lock);
543 	} else {
544 		/* Lock assigned and taken during PRU_ATTACH. */
545 	}
546 	error = (*prp->pr_usrreq)(so, PRU_ATTACH, NULL,
547 	    (struct mbuf *)(long)proto, NULL, l);
548 	KASSERT(solocked(so));
549 	if (error != 0) {
550 		so->so_state |= SS_NOFDREF;
551 		sofree(so);
552 		return error;
553 	}
554 	so->so_cred = kauth_cred_dup(l->l_cred);
555 	sounlock(so);
556 	*aso = so;
557 	return 0;
558 }
559 
560 /* On success, write file descriptor to fdout and return zero.  On
561  * failure, return non-zero; *fdout will be undefined.
562  */
563 int
564 fsocreate(int domain, struct socket **sop, int type, int protocol,
565     struct lwp *l, int *fdout)
566 {
567 	struct socket	*so;
568 	struct file	*fp;
569 	int		fd, error;
570 	int		flags = type & SOCK_FLAGS_MASK;
571 
572 	type &= ~SOCK_FLAGS_MASK;
573 	if ((error = fd_allocfile(&fp, &fd)) != 0)
574 		return error;
575 	fd_set_exclose(l, fd, (flags & SOCK_CLOEXEC) != 0);
576 	fp->f_flag = FREAD|FWRITE|((flags & SOCK_NONBLOCK) ? FNONBLOCK : 0)|
577 	    ((flags & SOCK_NOSIGPIPE) ? FNOSIGPIPE : 0);
578 	fp->f_type = DTYPE_SOCKET;
579 	fp->f_ops = &socketops;
580 	error = socreate(domain, &so, type, protocol, l, NULL);
581 	if (error != 0) {
582 		fd_abort(curproc, fp, fd);
583 	} else {
584 		if (sop != NULL)
585 			*sop = so;
586 		fp->f_data = so;
587 		fd_affix(curproc, fp, fd);
588 		*fdout = fd;
589 		if (flags & SOCK_NONBLOCK)
590 			so->so_state |= SS_NBIO;
591 	}
592 	return error;
593 }
594 
595 int
596 sofamily(const struct socket *so)
597 {
598 	const struct protosw *pr;
599 	const struct domain *dom;
600 
601 	if ((pr = so->so_proto) == NULL)
602 		return AF_UNSPEC;
603 	if ((dom = pr->pr_domain) == NULL)
604 		return AF_UNSPEC;
605 	return dom->dom_family;
606 }
607 
608 int
609 sobind(struct socket *so, struct mbuf *nam, struct lwp *l)
610 {
611 	int	error;
612 
613 	solock(so);
614 	error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, NULL, nam, NULL, l);
615 	sounlock(so);
616 	return error;
617 }
618 
619 int
620 solisten(struct socket *so, int backlog, struct lwp *l)
621 {
622 	int	error;
623 
624 	solock(so);
625 	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
626 	    SS_ISDISCONNECTING)) != 0) {
627 	    	sounlock(so);
628 		return (EINVAL);
629 	}
630 	error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, NULL,
631 	    NULL, NULL, l);
632 	if (error != 0) {
633 		sounlock(so);
634 		return error;
635 	}
636 	if (TAILQ_EMPTY(&so->so_q))
637 		so->so_options |= SO_ACCEPTCONN;
638 	if (backlog < 0)
639 		backlog = 0;
640 	so->so_qlimit = min(backlog, somaxconn);
641 	sounlock(so);
642 	return 0;
643 }
644 
645 void
646 sofree(struct socket *so)
647 {
648 	u_int refs;
649 
650 	KASSERT(solocked(so));
651 
652 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
653 		sounlock(so);
654 		return;
655 	}
656 	if (so->so_head) {
657 		/*
658 		 * We must not decommission a socket that's on the accept(2)
659 		 * queue.  If we do, then accept(2) may hang after select(2)
660 		 * indicated that the listening socket was ready.
661 		 */
662 		if (!soqremque(so, 0)) {
663 			sounlock(so);
664 			return;
665 		}
666 	}
667 	if (so->so_rcv.sb_hiwat)
668 		(void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0,
669 		    RLIM_INFINITY);
670 	if (so->so_snd.sb_hiwat)
671 		(void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0,
672 		    RLIM_INFINITY);
673 	sbrelease(&so->so_snd, so);
674 	KASSERT(!cv_has_waiters(&so->so_cv));
675 	KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv));
676 	KASSERT(!cv_has_waiters(&so->so_snd.sb_cv));
677 	sorflush(so);
678 	refs = so->so_aborting;	/* XXX */
679 	/* Remove acccept filter if one is present. */
680 	if (so->so_accf != NULL)
681 		(void)accept_filt_clear(so);
682 	sounlock(so);
683 	if (refs == 0)		/* XXX */
684 		soput(so);
685 }
686 
687 /*
688  * Close a socket on last file table reference removal.
689  * Initiate disconnect if connected.
690  * Free socket when disconnect complete.
691  */
692 int
693 soclose(struct socket *so)
694 {
695 	struct socket	*so2;
696 	int		error;
697 	int		error2;
698 
699 	error = 0;
700 	solock(so);
701 	if (so->so_options & SO_ACCEPTCONN) {
702 		for (;;) {
703 			if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) {
704 				KASSERT(solocked2(so, so2));
705 				(void) soqremque(so2, 0);
706 				/* soabort drops the lock. */
707 				(void) soabort(so2);
708 				solock(so);
709 				continue;
710 			}
711 			if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) {
712 				KASSERT(solocked2(so, so2));
713 				(void) soqremque(so2, 1);
714 				/* soabort drops the lock. */
715 				(void) soabort(so2);
716 				solock(so);
717 				continue;
718 			}
719 			break;
720 		}
721 	}
722 	if (so->so_pcb == 0)
723 		goto discard;
724 	if (so->so_state & SS_ISCONNECTED) {
725 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
726 			error = sodisconnect(so);
727 			if (error)
728 				goto drop;
729 		}
730 		if (so->so_options & SO_LINGER) {
731 			if ((so->so_state & (SS_ISDISCONNECTING|SS_NBIO)) ==
732 			    (SS_ISDISCONNECTING|SS_NBIO))
733 				goto drop;
734 			while (so->so_state & SS_ISCONNECTED) {
735 				error = sowait(so, true, so->so_linger * hz);
736 				if (error)
737 					break;
738 			}
739 		}
740 	}
741  drop:
742 	if (so->so_pcb) {
743 		error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH,
744 		    NULL, NULL, NULL, NULL);
745 		if (error == 0)
746 			error = error2;
747 	}
748  discard:
749 	if (so->so_state & SS_NOFDREF)
750 		panic("soclose: NOFDREF");
751 	kauth_cred_free(so->so_cred);
752 	so->so_state |= SS_NOFDREF;
753 	sofree(so);
754 	return (error);
755 }
756 
757 /*
758  * Must be called with the socket locked..  Will return with it unlocked.
759  */
760 int
761 soabort(struct socket *so)
762 {
763 	u_int refs;
764 	int error;
765 
766 	KASSERT(solocked(so));
767 	KASSERT(so->so_head == NULL);
768 
769 	so->so_aborting++;		/* XXX */
770 	error = (*so->so_proto->pr_usrreq)(so, PRU_ABORT, NULL,
771 	    NULL, NULL, NULL);
772 	refs = --so->so_aborting;	/* XXX */
773 	if (error || (refs == 0)) {
774 		sofree(so);
775 	} else {
776 		sounlock(so);
777 	}
778 	return error;
779 }
780 
781 int
782 soaccept(struct socket *so, struct mbuf *nam)
783 {
784 	int	error;
785 
786 	KASSERT(solocked(so));
787 
788 	error = 0;
789 	if ((so->so_state & SS_NOFDREF) == 0)
790 		panic("soaccept: !NOFDREF");
791 	so->so_state &= ~SS_NOFDREF;
792 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
793 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
794 		error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT,
795 		    NULL, nam, NULL, NULL);
796 	else
797 		error = ECONNABORTED;
798 
799 	return (error);
800 }
801 
802 int
803 soconnect(struct socket *so, struct mbuf *nam, struct lwp *l)
804 {
805 	int		error;
806 
807 	KASSERT(solocked(so));
808 
809 	if (so->so_options & SO_ACCEPTCONN)
810 		return (EOPNOTSUPP);
811 	/*
812 	 * If protocol is connection-based, can only connect once.
813 	 * Otherwise, if connected, try to disconnect first.
814 	 * This allows user to disconnect by connecting to, e.g.,
815 	 * a null address.
816 	 */
817 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
818 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
819 	    (error = sodisconnect(so))))
820 		error = EISCONN;
821 	else
822 		error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT,
823 		    NULL, nam, NULL, l);
824 	return (error);
825 }
826 
827 int
828 soconnect2(struct socket *so1, struct socket *so2)
829 {
830 	int	error;
831 
832 	KASSERT(solocked2(so1, so2));
833 
834 	error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2,
835 	    NULL, (struct mbuf *)so2, NULL, NULL);
836 	return (error);
837 }
838 
839 int
840 sodisconnect(struct socket *so)
841 {
842 	int	error;
843 
844 	KASSERT(solocked(so));
845 
846 	if ((so->so_state & SS_ISCONNECTED) == 0) {
847 		error = ENOTCONN;
848 	} else if (so->so_state & SS_ISDISCONNECTING) {
849 		error = EALREADY;
850 	} else {
851 		error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT,
852 		    NULL, NULL, NULL, NULL);
853 	}
854 	return (error);
855 }
856 
857 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
858 /*
859  * Send on a socket.
860  * If send must go all at once and message is larger than
861  * send buffering, then hard error.
862  * Lock against other senders.
863  * If must go all at once and not enough room now, then
864  * inform user that this would block and do nothing.
865  * Otherwise, if nonblocking, send as much as possible.
866  * The data to be sent is described by "uio" if nonzero,
867  * otherwise by the mbuf chain "top" (which must be null
868  * if uio is not).  Data provided in mbuf chain must be small
869  * enough to send all at once.
870  *
871  * Returns nonzero on error, timeout or signal; callers
872  * must check for short counts if EINTR/ERESTART are returned.
873  * Data and control buffers are freed on return.
874  */
875 int
876 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
877 	struct mbuf *control, int flags, struct lwp *l)
878 {
879 	struct mbuf	**mp, *m;
880 	long		space, len, resid, clen, mlen;
881 	int		error, s, dontroute, atomic;
882 	short		wakeup_state = 0;
883 
884 	clen = 0;
885 
886 	/*
887 	 * solock() provides atomicity of access.  splsoftnet() prevents
888 	 * protocol processing soft interrupts from interrupting us and
889 	 * blocking (expensive).
890 	 */
891 	s = splsoftnet();
892 	solock(so);
893 	atomic = sosendallatonce(so) || top;
894 	if (uio)
895 		resid = uio->uio_resid;
896 	else
897 		resid = top->m_pkthdr.len;
898 	/*
899 	 * In theory resid should be unsigned.
900 	 * However, space must be signed, as it might be less than 0
901 	 * if we over-committed, and we must use a signed comparison
902 	 * of space and resid.  On the other hand, a negative resid
903 	 * causes us to loop sending 0-length segments to the protocol.
904 	 */
905 	if (resid < 0) {
906 		error = EINVAL;
907 		goto out;
908 	}
909 	dontroute =
910 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
911 	    (so->so_proto->pr_flags & PR_ATOMIC);
912 	l->l_ru.ru_msgsnd++;
913 	if (control)
914 		clen = control->m_len;
915  restart:
916 	if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
917 		goto out;
918 	do {
919 		if (so->so_state & SS_CANTSENDMORE) {
920 			error = EPIPE;
921 			goto release;
922 		}
923 		if (so->so_error) {
924 			error = so->so_error;
925 			so->so_error = 0;
926 			goto release;
927 		}
928 		if ((so->so_state & SS_ISCONNECTED) == 0) {
929 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
930 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
931 				    !(resid == 0 && clen != 0)) {
932 					error = ENOTCONN;
933 					goto release;
934 				}
935 			} else if (addr == 0) {
936 				error = EDESTADDRREQ;
937 				goto release;
938 			}
939 		}
940 		space = sbspace(&so->so_snd);
941 		if (flags & MSG_OOB)
942 			space += 1024;
943 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
944 		    clen > so->so_snd.sb_hiwat) {
945 			error = EMSGSIZE;
946 			goto release;
947 		}
948 		if (space < resid + clen &&
949 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
950 			if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
951 				error = EWOULDBLOCK;
952 				goto release;
953 			}
954 			sbunlock(&so->so_snd);
955 			if (wakeup_state & SS_RESTARTSYS) {
956 				error = ERESTART;
957 				goto out;
958 			}
959 			error = sbwait(&so->so_snd);
960 			if (error)
961 				goto out;
962 			wakeup_state = so->so_state;
963 			goto restart;
964 		}
965 		wakeup_state = 0;
966 		mp = &top;
967 		space -= clen;
968 		do {
969 			if (uio == NULL) {
970 				/*
971 				 * Data is prepackaged in "top".
972 				 */
973 				resid = 0;
974 				if (flags & MSG_EOR)
975 					top->m_flags |= M_EOR;
976 			} else do {
977 				sounlock(so);
978 				splx(s);
979 				if (top == NULL) {
980 					m = m_gethdr(M_WAIT, MT_DATA);
981 					mlen = MHLEN;
982 					m->m_pkthdr.len = 0;
983 					m->m_pkthdr.rcvif = NULL;
984 				} else {
985 					m = m_get(M_WAIT, MT_DATA);
986 					mlen = MLEN;
987 				}
988 				MCLAIM(m, so->so_snd.sb_mowner);
989 				if (sock_loan_thresh >= 0 &&
990 				    uio->uio_iov->iov_len >= sock_loan_thresh &&
991 				    space >= sock_loan_thresh &&
992 				    (len = sosend_loan(so, uio, m,
993 						       space)) != 0) {
994 					SOSEND_COUNTER_INCR(&sosend_loan_big);
995 					space -= len;
996 					goto have_data;
997 				}
998 				if (resid >= MINCLSIZE && space >= MCLBYTES) {
999 					SOSEND_COUNTER_INCR(&sosend_copy_big);
1000 					m_clget(m, M_DONTWAIT);
1001 					if ((m->m_flags & M_EXT) == 0)
1002 						goto nopages;
1003 					mlen = MCLBYTES;
1004 					if (atomic && top == 0) {
1005 						len = lmin(MCLBYTES - max_hdr,
1006 						    resid);
1007 						m->m_data += max_hdr;
1008 					} else
1009 						len = lmin(MCLBYTES, resid);
1010 					space -= len;
1011 				} else {
1012  nopages:
1013 					SOSEND_COUNTER_INCR(&sosend_copy_small);
1014 					len = lmin(lmin(mlen, resid), space);
1015 					space -= len;
1016 					/*
1017 					 * For datagram protocols, leave room
1018 					 * for protocol headers in first mbuf.
1019 					 */
1020 					if (atomic && top == 0 && len < mlen)
1021 						MH_ALIGN(m, len);
1022 				}
1023 				error = uiomove(mtod(m, void *), (int)len, uio);
1024  have_data:
1025 				resid = uio->uio_resid;
1026 				m->m_len = len;
1027 				*mp = m;
1028 				top->m_pkthdr.len += len;
1029 				s = splsoftnet();
1030 				solock(so);
1031 				if (error != 0)
1032 					goto release;
1033 				mp = &m->m_next;
1034 				if (resid <= 0) {
1035 					if (flags & MSG_EOR)
1036 						top->m_flags |= M_EOR;
1037 					break;
1038 				}
1039 			} while (space > 0 && atomic);
1040 
1041 			if (so->so_state & SS_CANTSENDMORE) {
1042 				error = EPIPE;
1043 				goto release;
1044 			}
1045 			if (dontroute)
1046 				so->so_options |= SO_DONTROUTE;
1047 			if (resid > 0)
1048 				so->so_state |= SS_MORETOCOME;
1049 			error = (*so->so_proto->pr_usrreq)(so,
1050 			    (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND,
1051 			    top, addr, control, curlwp);
1052 			if (dontroute)
1053 				so->so_options &= ~SO_DONTROUTE;
1054 			if (resid > 0)
1055 				so->so_state &= ~SS_MORETOCOME;
1056 			clen = 0;
1057 			control = NULL;
1058 			top = NULL;
1059 			mp = &top;
1060 			if (error != 0)
1061 				goto release;
1062 		} while (resid && space > 0);
1063 	} while (resid);
1064 
1065  release:
1066 	sbunlock(&so->so_snd);
1067  out:
1068 	sounlock(so);
1069 	splx(s);
1070 	if (top)
1071 		m_freem(top);
1072 	if (control)
1073 		m_freem(control);
1074 	return (error);
1075 }
1076 
1077 /*
1078  * Following replacement or removal of the first mbuf on the first
1079  * mbuf chain of a socket buffer, push necessary state changes back
1080  * into the socket buffer so that other consumers see the values
1081  * consistently.  'nextrecord' is the callers locally stored value of
1082  * the original value of sb->sb_mb->m_nextpkt which must be restored
1083  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
1084  */
1085 static void
1086 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
1087 {
1088 
1089 	KASSERT(solocked(sb->sb_so));
1090 
1091 	/*
1092 	 * First, update for the new value of nextrecord.  If necessary,
1093 	 * make it the first record.
1094 	 */
1095 	if (sb->sb_mb != NULL)
1096 		sb->sb_mb->m_nextpkt = nextrecord;
1097 	else
1098 		sb->sb_mb = nextrecord;
1099 
1100         /*
1101          * Now update any dependent socket buffer fields to reflect
1102          * the new state.  This is an inline of SB_EMPTY_FIXUP, with
1103          * the addition of a second clause that takes care of the
1104          * case where sb_mb has been updated, but remains the last
1105          * record.
1106          */
1107         if (sb->sb_mb == NULL) {
1108                 sb->sb_mbtail = NULL;
1109                 sb->sb_lastrecord = NULL;
1110         } else if (sb->sb_mb->m_nextpkt == NULL)
1111                 sb->sb_lastrecord = sb->sb_mb;
1112 }
1113 
1114 /*
1115  * Implement receive operations on a socket.
1116  * We depend on the way that records are added to the sockbuf
1117  * by sbappend*.  In particular, each record (mbufs linked through m_next)
1118  * must begin with an address if the protocol so specifies,
1119  * followed by an optional mbuf or mbufs containing ancillary data,
1120  * and then zero or more mbufs of data.
1121  * In order to avoid blocking network interrupts for the entire time here,
1122  * we splx() while doing the actual copy to user space.
1123  * Although the sockbuf is locked, new data may still be appended,
1124  * and thus we must maintain consistency of the sockbuf during that time.
1125  *
1126  * The caller may receive the data as a single mbuf chain by supplying
1127  * an mbuf **mp0 for use in returning the chain.  The uio is then used
1128  * only for the count in uio_resid.
1129  */
1130 int
1131 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
1132 	struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1133 {
1134 	struct lwp *l = curlwp;
1135 	struct mbuf	*m, **mp, *mt;
1136 	size_t len, offset, moff, orig_resid;
1137 	int atomic, flags, error, s, type;
1138 	const struct protosw	*pr;
1139 	struct mbuf	*nextrecord;
1140 	int		mbuf_removed = 0;
1141 	const struct domain *dom;
1142 	short		wakeup_state = 0;
1143 
1144 	pr = so->so_proto;
1145 	atomic = pr->pr_flags & PR_ATOMIC;
1146 	dom = pr->pr_domain;
1147 	mp = mp0;
1148 	type = 0;
1149 	orig_resid = uio->uio_resid;
1150 
1151 	if (paddr != NULL)
1152 		*paddr = NULL;
1153 	if (controlp != NULL)
1154 		*controlp = NULL;
1155 	if (flagsp != NULL)
1156 		flags = *flagsp &~ MSG_EOR;
1157 	else
1158 		flags = 0;
1159 
1160 	if (flags & MSG_OOB) {
1161 		m = m_get(M_WAIT, MT_DATA);
1162 		solock(so);
1163 		error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m,
1164 		    (struct mbuf *)(long)(flags & MSG_PEEK), NULL, l);
1165 		sounlock(so);
1166 		if (error)
1167 			goto bad;
1168 		do {
1169 			error = uiomove(mtod(m, void *),
1170 			    MIN(uio->uio_resid, m->m_len), uio);
1171 			m = m_free(m);
1172 		} while (uio->uio_resid > 0 && error == 0 && m);
1173  bad:
1174 		if (m != NULL)
1175 			m_freem(m);
1176 		return error;
1177 	}
1178 	if (mp != NULL)
1179 		*mp = NULL;
1180 
1181 	/*
1182 	 * solock() provides atomicity of access.  splsoftnet() prevents
1183 	 * protocol processing soft interrupts from interrupting us and
1184 	 * blocking (expensive).
1185 	 */
1186 	s = splsoftnet();
1187 	solock(so);
1188 	if (so->so_state & SS_ISCONFIRMING && uio->uio_resid)
1189 		(*pr->pr_usrreq)(so, PRU_RCVD, NULL, NULL, NULL, l);
1190 
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_usrreq)(so, PRU_RCVD,
1535 				    NULL, (struct mbuf *)(long)flags, NULL, l);
1536 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1537 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1538 			if (wakeup_state & SS_RESTARTSYS)
1539 				error = ERESTART;
1540 			else
1541 				error = sbwait(&so->so_rcv);
1542 			if (error != 0) {
1543 				sbunlock(&so->so_rcv);
1544 				sounlock(so);
1545 				splx(s);
1546 				return 0;
1547 			}
1548 			if ((m = so->so_rcv.sb_mb) != NULL)
1549 				nextrecord = m->m_nextpkt;
1550 			wakeup_state = so->so_state;
1551 		}
1552 	}
1553 
1554 	if (m && atomic) {
1555 		flags |= MSG_TRUNC;
1556 		if ((flags & MSG_PEEK) == 0)
1557 			(void) sbdroprecord(&so->so_rcv);
1558 	}
1559 	if ((flags & MSG_PEEK) == 0) {
1560 		if (m == NULL) {
1561 			/*
1562 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1563 			 * part makes sure sb_lastrecord is up-to-date if
1564 			 * there is still data in the socket buffer.
1565 			 */
1566 			so->so_rcv.sb_mb = nextrecord;
1567 			if (so->so_rcv.sb_mb == NULL) {
1568 				so->so_rcv.sb_mbtail = NULL;
1569 				so->so_rcv.sb_lastrecord = NULL;
1570 			} else if (nextrecord->m_nextpkt == NULL)
1571 				so->so_rcv.sb_lastrecord = nextrecord;
1572 		}
1573 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1574 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1575 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1576 			(*pr->pr_usrreq)(so, PRU_RCVD, NULL,
1577 			    (struct mbuf *)(long)flags, NULL, l);
1578 	}
1579 	if (orig_resid == uio->uio_resid && orig_resid &&
1580 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1581 		sbunlock(&so->so_rcv);
1582 		goto restart;
1583 	}
1584 
1585 	if (flagsp != NULL)
1586 		*flagsp |= flags;
1587  release:
1588 	sbunlock(&so->so_rcv);
1589 	sounlock(so);
1590 	splx(s);
1591 	return error;
1592 }
1593 
1594 int
1595 soshutdown(struct socket *so, int how)
1596 {
1597 	const struct protosw	*pr;
1598 	int	error;
1599 
1600 	KASSERT(solocked(so));
1601 
1602 	pr = so->so_proto;
1603 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1604 		return (EINVAL);
1605 
1606 	if (how == SHUT_RD || how == SHUT_RDWR) {
1607 		sorflush(so);
1608 		error = 0;
1609 	}
1610 	if (how == SHUT_WR || how == SHUT_RDWR)
1611 		error = (*pr->pr_usrreq)(so, PRU_SHUTDOWN, NULL,
1612 		    NULL, NULL, NULL);
1613 
1614 	return error;
1615 }
1616 
1617 void
1618 sorestart(struct socket *so)
1619 {
1620 	/*
1621 	 * An application has called close() on an fd on which another
1622 	 * of its threads has called a socket system call.
1623 	 * Mark this and wake everyone up, and code that would block again
1624 	 * instead returns ERESTART.
1625 	 * On system call re-entry the fd is validated and EBADF returned.
1626 	 * Any other fd will block again on the 2nd syscall.
1627 	 */
1628 	solock(so);
1629 	so->so_state |= SS_RESTARTSYS;
1630 	cv_broadcast(&so->so_cv);
1631 	cv_broadcast(&so->so_snd.sb_cv);
1632 	cv_broadcast(&so->so_rcv.sb_cv);
1633 	sounlock(so);
1634 }
1635 
1636 void
1637 sorflush(struct socket *so)
1638 {
1639 	struct sockbuf	*sb, asb;
1640 	const struct protosw	*pr;
1641 
1642 	KASSERT(solocked(so));
1643 
1644 	sb = &so->so_rcv;
1645 	pr = so->so_proto;
1646 	socantrcvmore(so);
1647 	sb->sb_flags |= SB_NOINTR;
1648 	(void )sblock(sb, M_WAITOK);
1649 	sbunlock(sb);
1650 	asb = *sb;
1651 	/*
1652 	 * Clear most of the sockbuf structure, but leave some of the
1653 	 * fields valid.
1654 	 */
1655 	memset(&sb->sb_startzero, 0,
1656 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1657 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) {
1658 		sounlock(so);
1659 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1660 		solock(so);
1661 	}
1662 	sbrelease(&asb, so);
1663 }
1664 
1665 /*
1666  * internal set SOL_SOCKET options
1667  */
1668 static int
1669 sosetopt1(struct socket *so, const struct sockopt *sopt)
1670 {
1671 	int error = EINVAL, optval, opt;
1672 	struct linger l;
1673 	struct timeval tv;
1674 
1675 	switch ((opt = sopt->sopt_name)) {
1676 
1677 	case SO_ACCEPTFILTER:
1678 		error = accept_filt_setopt(so, sopt);
1679 		KASSERT(solocked(so));
1680 		break;
1681 
1682   	case SO_LINGER:
1683  		error = sockopt_get(sopt, &l, sizeof(l));
1684 		solock(so);
1685  		if (error)
1686  			break;
1687  		if (l.l_linger < 0 || l.l_linger > USHRT_MAX ||
1688  		    l.l_linger > (INT_MAX / hz)) {
1689 			error = EDOM;
1690 			break;
1691 		}
1692  		so->so_linger = l.l_linger;
1693  		if (l.l_onoff)
1694  			so->so_options |= SO_LINGER;
1695  		else
1696  			so->so_options &= ~SO_LINGER;
1697    		break;
1698 
1699 	case SO_DEBUG:
1700 	case SO_KEEPALIVE:
1701 	case SO_DONTROUTE:
1702 	case SO_USELOOPBACK:
1703 	case SO_BROADCAST:
1704 	case SO_REUSEADDR:
1705 	case SO_REUSEPORT:
1706 	case SO_OOBINLINE:
1707 	case SO_TIMESTAMP:
1708 	case SO_NOSIGPIPE:
1709 #ifdef SO_OTIMESTAMP
1710 	case SO_OTIMESTAMP:
1711 #endif
1712 		error = sockopt_getint(sopt, &optval);
1713 		solock(so);
1714 		if (error)
1715 			break;
1716 		if (optval)
1717 			so->so_options |= opt;
1718 		else
1719 			so->so_options &= ~opt;
1720 		break;
1721 
1722 	case SO_SNDBUF:
1723 	case SO_RCVBUF:
1724 	case SO_SNDLOWAT:
1725 	case SO_RCVLOWAT:
1726 		error = sockopt_getint(sopt, &optval);
1727 		solock(so);
1728 		if (error)
1729 			break;
1730 
1731 		/*
1732 		 * Values < 1 make no sense for any of these
1733 		 * options, so disallow them.
1734 		 */
1735 		if (optval < 1) {
1736 			error = EINVAL;
1737 			break;
1738 		}
1739 
1740 		switch (opt) {
1741 		case SO_SNDBUF:
1742 			if (sbreserve(&so->so_snd, (u_long)optval, so) == 0) {
1743 				error = ENOBUFS;
1744 				break;
1745 			}
1746 			so->so_snd.sb_flags &= ~SB_AUTOSIZE;
1747 			break;
1748 
1749 		case SO_RCVBUF:
1750 			if (sbreserve(&so->so_rcv, (u_long)optval, so) == 0) {
1751 				error = ENOBUFS;
1752 				break;
1753 			}
1754 			so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1755 			break;
1756 
1757 		/*
1758 		 * Make sure the low-water is never greater than
1759 		 * the high-water.
1760 		 */
1761 		case SO_SNDLOWAT:
1762 			if (optval > so->so_snd.sb_hiwat)
1763 				optval = so->so_snd.sb_hiwat;
1764 
1765 			so->so_snd.sb_lowat = optval;
1766 			break;
1767 
1768 		case SO_RCVLOWAT:
1769 			if (optval > so->so_rcv.sb_hiwat)
1770 				optval = so->so_rcv.sb_hiwat;
1771 
1772 			so->so_rcv.sb_lowat = optval;
1773 			break;
1774 		}
1775 		break;
1776 
1777 #ifdef COMPAT_50
1778 	case SO_OSNDTIMEO:
1779 	case SO_ORCVTIMEO: {
1780 		struct timeval50 otv;
1781 		error = sockopt_get(sopt, &otv, sizeof(otv));
1782 		if (error) {
1783 			solock(so);
1784 			break;
1785 		}
1786 		timeval50_to_timeval(&otv, &tv);
1787 		opt = opt == SO_OSNDTIMEO ? SO_SNDTIMEO : SO_RCVTIMEO;
1788 		error = 0;
1789 		/*FALLTHROUGH*/
1790 	}
1791 #endif /* COMPAT_50 */
1792 
1793 	case SO_SNDTIMEO:
1794 	case SO_RCVTIMEO:
1795 		if (error)
1796 			error = sockopt_get(sopt, &tv, sizeof(tv));
1797 		solock(so);
1798 		if (error)
1799 			break;
1800 
1801 		if (tv.tv_sec > (INT_MAX - tv.tv_usec / tick) / hz) {
1802 			error = EDOM;
1803 			break;
1804 		}
1805 
1806 		optval = tv.tv_sec * hz + tv.tv_usec / tick;
1807 		if (optval == 0 && tv.tv_usec != 0)
1808 			optval = 1;
1809 
1810 		switch (opt) {
1811 		case SO_SNDTIMEO:
1812 			so->so_snd.sb_timeo = optval;
1813 			break;
1814 		case SO_RCVTIMEO:
1815 			so->so_rcv.sb_timeo = optval;
1816 			break;
1817 		}
1818 		break;
1819 
1820 	default:
1821 		solock(so);
1822 		error = ENOPROTOOPT;
1823 		break;
1824 	}
1825 	KASSERT(solocked(so));
1826 	return error;
1827 }
1828 
1829 int
1830 sosetopt(struct socket *so, struct sockopt *sopt)
1831 {
1832 	int error, prerr;
1833 
1834 	if (sopt->sopt_level == SOL_SOCKET) {
1835 		error = sosetopt1(so, sopt);
1836 		KASSERT(solocked(so));
1837 	} else {
1838 		error = ENOPROTOOPT;
1839 		solock(so);
1840 	}
1841 
1842 	if ((error == 0 || error == ENOPROTOOPT) &&
1843 	    so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) {
1844 		/* give the protocol stack a shot */
1845 		prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, sopt);
1846 		if (prerr == 0)
1847 			error = 0;
1848 		else if (prerr != ENOPROTOOPT)
1849 			error = prerr;
1850 	}
1851 	sounlock(so);
1852 	return error;
1853 }
1854 
1855 /*
1856  * so_setsockopt() is a wrapper providing a sockopt structure for sosetopt()
1857  */
1858 int
1859 so_setsockopt(struct lwp *l, struct socket *so, int level, int name,
1860     const void *val, size_t valsize)
1861 {
1862 	struct sockopt sopt;
1863 	int error;
1864 
1865 	KASSERT(valsize == 0 || val != NULL);
1866 
1867 	sockopt_init(&sopt, level, name, valsize);
1868 	sockopt_set(&sopt, val, valsize);
1869 
1870 	error = sosetopt(so, &sopt);
1871 
1872 	sockopt_destroy(&sopt);
1873 
1874 	return error;
1875 }
1876 
1877 /*
1878  * internal get SOL_SOCKET options
1879  */
1880 static int
1881 sogetopt1(struct socket *so, struct sockopt *sopt)
1882 {
1883 	int error, optval, opt;
1884 	struct linger l;
1885 	struct timeval tv;
1886 
1887 	switch ((opt = sopt->sopt_name)) {
1888 
1889 	case SO_ACCEPTFILTER:
1890 		error = accept_filt_getopt(so, sopt);
1891 		break;
1892 
1893 	case SO_LINGER:
1894 		l.l_onoff = (so->so_options & SO_LINGER) ? 1 : 0;
1895 		l.l_linger = so->so_linger;
1896 
1897 		error = sockopt_set(sopt, &l, sizeof(l));
1898 		break;
1899 
1900 	case SO_USELOOPBACK:
1901 	case SO_DONTROUTE:
1902 	case SO_DEBUG:
1903 	case SO_KEEPALIVE:
1904 	case SO_REUSEADDR:
1905 	case SO_REUSEPORT:
1906 	case SO_BROADCAST:
1907 	case SO_OOBINLINE:
1908 	case SO_TIMESTAMP:
1909 	case SO_NOSIGPIPE:
1910 #ifdef SO_OTIMESTAMP
1911 	case SO_OTIMESTAMP:
1912 #endif
1913 		error = sockopt_setint(sopt, (so->so_options & opt) ? 1 : 0);
1914 		break;
1915 
1916 	case SO_TYPE:
1917 		error = sockopt_setint(sopt, so->so_type);
1918 		break;
1919 
1920 	case SO_ERROR:
1921 		error = sockopt_setint(sopt, so->so_error);
1922 		so->so_error = 0;
1923 		break;
1924 
1925 	case SO_SNDBUF:
1926 		error = sockopt_setint(sopt, so->so_snd.sb_hiwat);
1927 		break;
1928 
1929 	case SO_RCVBUF:
1930 		error = sockopt_setint(sopt, so->so_rcv.sb_hiwat);
1931 		break;
1932 
1933 	case SO_SNDLOWAT:
1934 		error = sockopt_setint(sopt, so->so_snd.sb_lowat);
1935 		break;
1936 
1937 	case SO_RCVLOWAT:
1938 		error = sockopt_setint(sopt, so->so_rcv.sb_lowat);
1939 		break;
1940 
1941 #ifdef COMPAT_50
1942 	case SO_OSNDTIMEO:
1943 	case SO_ORCVTIMEO: {
1944 		struct timeval50 otv;
1945 
1946 		optval = (opt == SO_OSNDTIMEO ?
1947 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1948 
1949 		otv.tv_sec = optval / hz;
1950 		otv.tv_usec = (optval % hz) * tick;
1951 
1952 		error = sockopt_set(sopt, &otv, sizeof(otv));
1953 		break;
1954 	}
1955 #endif /* COMPAT_50 */
1956 
1957 	case SO_SNDTIMEO:
1958 	case SO_RCVTIMEO:
1959 		optval = (opt == SO_SNDTIMEO ?
1960 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1961 
1962 		tv.tv_sec = optval / hz;
1963 		tv.tv_usec = (optval % hz) * tick;
1964 
1965 		error = sockopt_set(sopt, &tv, sizeof(tv));
1966 		break;
1967 
1968 	case SO_OVERFLOWED:
1969 		error = sockopt_setint(sopt, so->so_rcv.sb_overflowed);
1970 		break;
1971 
1972 	default:
1973 		error = ENOPROTOOPT;
1974 		break;
1975 	}
1976 
1977 	return (error);
1978 }
1979 
1980 int
1981 sogetopt(struct socket *so, struct sockopt *sopt)
1982 {
1983 	int		error;
1984 
1985 	solock(so);
1986 	if (sopt->sopt_level != SOL_SOCKET) {
1987 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1988 			error = ((*so->so_proto->pr_ctloutput)
1989 			    (PRCO_GETOPT, so, sopt));
1990 		} else
1991 			error = (ENOPROTOOPT);
1992 	} else {
1993 		error = sogetopt1(so, sopt);
1994 	}
1995 	sounlock(so);
1996 	return (error);
1997 }
1998 
1999 /*
2000  * alloc sockopt data buffer buffer
2001  *	- will be released at destroy
2002  */
2003 static int
2004 sockopt_alloc(struct sockopt *sopt, size_t len, km_flag_t kmflag)
2005 {
2006 
2007 	KASSERT(sopt->sopt_size == 0);
2008 
2009 	if (len > sizeof(sopt->sopt_buf)) {
2010 		sopt->sopt_data = kmem_zalloc(len, kmflag);
2011 		if (sopt->sopt_data == NULL)
2012 			return ENOMEM;
2013 	} else
2014 		sopt->sopt_data = sopt->sopt_buf;
2015 
2016 	sopt->sopt_size = len;
2017 	return 0;
2018 }
2019 
2020 /*
2021  * initialise sockopt storage
2022  *	- MAY sleep during allocation
2023  */
2024 void
2025 sockopt_init(struct sockopt *sopt, int level, int name, size_t size)
2026 {
2027 
2028 	memset(sopt, 0, sizeof(*sopt));
2029 
2030 	sopt->sopt_level = level;
2031 	sopt->sopt_name = name;
2032 	(void)sockopt_alloc(sopt, size, KM_SLEEP);
2033 }
2034 
2035 /*
2036  * destroy sockopt storage
2037  *	- will release any held memory references
2038  */
2039 void
2040 sockopt_destroy(struct sockopt *sopt)
2041 {
2042 
2043 	if (sopt->sopt_data != sopt->sopt_buf)
2044 		kmem_free(sopt->sopt_data, sopt->sopt_size);
2045 
2046 	memset(sopt, 0, sizeof(*sopt));
2047 }
2048 
2049 /*
2050  * set sockopt value
2051  *	- value is copied into sockopt
2052  * 	- memory is allocated when necessary, will not sleep
2053  */
2054 int
2055 sockopt_set(struct sockopt *sopt, const void *buf, size_t len)
2056 {
2057 	int error;
2058 
2059 	if (sopt->sopt_size == 0) {
2060 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2061 		if (error)
2062 			return error;
2063 	}
2064 
2065 	KASSERT(sopt->sopt_size == len);
2066 	memcpy(sopt->sopt_data, buf, len);
2067 	return 0;
2068 }
2069 
2070 /*
2071  * common case of set sockopt integer value
2072  */
2073 int
2074 sockopt_setint(struct sockopt *sopt, int val)
2075 {
2076 
2077 	return sockopt_set(sopt, &val, sizeof(int));
2078 }
2079 
2080 /*
2081  * get sockopt value
2082  *	- correct size must be given
2083  */
2084 int
2085 sockopt_get(const struct sockopt *sopt, void *buf, size_t len)
2086 {
2087 
2088 	if (sopt->sopt_size != len)
2089 		return EINVAL;
2090 
2091 	memcpy(buf, sopt->sopt_data, len);
2092 	return 0;
2093 }
2094 
2095 /*
2096  * common case of get sockopt integer value
2097  */
2098 int
2099 sockopt_getint(const struct sockopt *sopt, int *valp)
2100 {
2101 
2102 	return sockopt_get(sopt, valp, sizeof(int));
2103 }
2104 
2105 /*
2106  * set sockopt value from mbuf
2107  *	- ONLY for legacy code
2108  *	- mbuf is released by sockopt
2109  *	- will not sleep
2110  */
2111 int
2112 sockopt_setmbuf(struct sockopt *sopt, struct mbuf *m)
2113 {
2114 	size_t len;
2115 	int error;
2116 
2117 	len = m_length(m);
2118 
2119 	if (sopt->sopt_size == 0) {
2120 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2121 		if (error)
2122 			return error;
2123 	}
2124 
2125 	KASSERT(sopt->sopt_size == len);
2126 	m_copydata(m, 0, len, sopt->sopt_data);
2127 	m_freem(m);
2128 
2129 	return 0;
2130 }
2131 
2132 /*
2133  * get sockopt value into mbuf
2134  *	- ONLY for legacy code
2135  *	- mbuf to be released by the caller
2136  *	- will not sleep
2137  */
2138 struct mbuf *
2139 sockopt_getmbuf(const struct sockopt *sopt)
2140 {
2141 	struct mbuf *m;
2142 
2143 	if (sopt->sopt_size > MCLBYTES)
2144 		return NULL;
2145 
2146 	m = m_get(M_DONTWAIT, MT_SOOPTS);
2147 	if (m == NULL)
2148 		return NULL;
2149 
2150 	if (sopt->sopt_size > MLEN) {
2151 		MCLGET(m, M_DONTWAIT);
2152 		if ((m->m_flags & M_EXT) == 0) {
2153 			m_free(m);
2154 			return NULL;
2155 		}
2156 	}
2157 
2158 	memcpy(mtod(m, void *), sopt->sopt_data, sopt->sopt_size);
2159 	m->m_len = sopt->sopt_size;
2160 
2161 	return m;
2162 }
2163 
2164 void
2165 sohasoutofband(struct socket *so)
2166 {
2167 
2168 	fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
2169 	selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, NOTE_SUBMIT);
2170 }
2171 
2172 static void
2173 filt_sordetach(struct knote *kn)
2174 {
2175 	struct socket	*so;
2176 
2177 	so = ((file_t *)kn->kn_obj)->f_data;
2178 	solock(so);
2179 	SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
2180 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
2181 		so->so_rcv.sb_flags &= ~SB_KNOTE;
2182 	sounlock(so);
2183 }
2184 
2185 /*ARGSUSED*/
2186 static int
2187 filt_soread(struct knote *kn, long hint)
2188 {
2189 	struct socket	*so;
2190 	int rv;
2191 
2192 	so = ((file_t *)kn->kn_obj)->f_data;
2193 	if (hint != NOTE_SUBMIT)
2194 		solock(so);
2195 	kn->kn_data = so->so_rcv.sb_cc;
2196 	if (so->so_state & SS_CANTRCVMORE) {
2197 		kn->kn_flags |= EV_EOF;
2198 		kn->kn_fflags = so->so_error;
2199 		rv = 1;
2200 	} else if (so->so_error)	/* temporary udp error */
2201 		rv = 1;
2202 	else if (kn->kn_sfflags & NOTE_LOWAT)
2203 		rv = (kn->kn_data >= kn->kn_sdata);
2204 	else
2205 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
2206 	if (hint != NOTE_SUBMIT)
2207 		sounlock(so);
2208 	return rv;
2209 }
2210 
2211 static void
2212 filt_sowdetach(struct knote *kn)
2213 {
2214 	struct socket	*so;
2215 
2216 	so = ((file_t *)kn->kn_obj)->f_data;
2217 	solock(so);
2218 	SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
2219 	if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
2220 		so->so_snd.sb_flags &= ~SB_KNOTE;
2221 	sounlock(so);
2222 }
2223 
2224 /*ARGSUSED*/
2225 static int
2226 filt_sowrite(struct knote *kn, long hint)
2227 {
2228 	struct socket	*so;
2229 	int rv;
2230 
2231 	so = ((file_t *)kn->kn_obj)->f_data;
2232 	if (hint != NOTE_SUBMIT)
2233 		solock(so);
2234 	kn->kn_data = sbspace(&so->so_snd);
2235 	if (so->so_state & SS_CANTSENDMORE) {
2236 		kn->kn_flags |= EV_EOF;
2237 		kn->kn_fflags = so->so_error;
2238 		rv = 1;
2239 	} else if (so->so_error)	/* temporary udp error */
2240 		rv = 1;
2241 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2242 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
2243 		rv = 0;
2244 	else if (kn->kn_sfflags & NOTE_LOWAT)
2245 		rv = (kn->kn_data >= kn->kn_sdata);
2246 	else
2247 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
2248 	if (hint != NOTE_SUBMIT)
2249 		sounlock(so);
2250 	return rv;
2251 }
2252 
2253 /*ARGSUSED*/
2254 static int
2255 filt_solisten(struct knote *kn, long hint)
2256 {
2257 	struct socket	*so;
2258 	int rv;
2259 
2260 	so = ((file_t *)kn->kn_obj)->f_data;
2261 
2262 	/*
2263 	 * Set kn_data to number of incoming connections, not
2264 	 * counting partial (incomplete) connections.
2265 	 */
2266 	if (hint != NOTE_SUBMIT)
2267 		solock(so);
2268 	kn->kn_data = so->so_qlen;
2269 	rv = (kn->kn_data > 0);
2270 	if (hint != NOTE_SUBMIT)
2271 		sounlock(so);
2272 	return rv;
2273 }
2274 
2275 static const struct filterops solisten_filtops =
2276 	{ 1, NULL, filt_sordetach, filt_solisten };
2277 static const struct filterops soread_filtops =
2278 	{ 1, NULL, filt_sordetach, filt_soread };
2279 static const struct filterops sowrite_filtops =
2280 	{ 1, NULL, filt_sowdetach, filt_sowrite };
2281 
2282 int
2283 soo_kqfilter(struct file *fp, struct knote *kn)
2284 {
2285 	struct socket	*so;
2286 	struct sockbuf	*sb;
2287 
2288 	so = ((file_t *)kn->kn_obj)->f_data;
2289 	solock(so);
2290 	switch (kn->kn_filter) {
2291 	case EVFILT_READ:
2292 		if (so->so_options & SO_ACCEPTCONN)
2293 			kn->kn_fop = &solisten_filtops;
2294 		else
2295 			kn->kn_fop = &soread_filtops;
2296 		sb = &so->so_rcv;
2297 		break;
2298 	case EVFILT_WRITE:
2299 		kn->kn_fop = &sowrite_filtops;
2300 		sb = &so->so_snd;
2301 		break;
2302 	default:
2303 		sounlock(so);
2304 		return (EINVAL);
2305 	}
2306 	SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
2307 	sb->sb_flags |= SB_KNOTE;
2308 	sounlock(so);
2309 	return (0);
2310 }
2311 
2312 static int
2313 sodopoll(struct socket *so, int events)
2314 {
2315 	int revents;
2316 
2317 	revents = 0;
2318 
2319 	if (events & (POLLIN | POLLRDNORM))
2320 		if (soreadable(so))
2321 			revents |= events & (POLLIN | POLLRDNORM);
2322 
2323 	if (events & (POLLOUT | POLLWRNORM))
2324 		if (sowritable(so))
2325 			revents |= events & (POLLOUT | POLLWRNORM);
2326 
2327 	if (events & (POLLPRI | POLLRDBAND))
2328 		if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
2329 			revents |= events & (POLLPRI | POLLRDBAND);
2330 
2331 	return revents;
2332 }
2333 
2334 int
2335 sopoll(struct socket *so, int events)
2336 {
2337 	int revents = 0;
2338 
2339 #ifndef DIAGNOSTIC
2340 	/*
2341 	 * Do a quick, unlocked check in expectation that the socket
2342 	 * will be ready for I/O.  Don't do this check if DIAGNOSTIC,
2343 	 * as the solocked() assertions will fail.
2344 	 */
2345 	if ((revents = sodopoll(so, events)) != 0)
2346 		return revents;
2347 #endif
2348 
2349 	solock(so);
2350 	if ((revents = sodopoll(so, events)) == 0) {
2351 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2352 			selrecord(curlwp, &so->so_rcv.sb_sel);
2353 			so->so_rcv.sb_flags |= SB_NOTIFY;
2354 		}
2355 
2356 		if (events & (POLLOUT | POLLWRNORM)) {
2357 			selrecord(curlwp, &so->so_snd.sb_sel);
2358 			so->so_snd.sb_flags |= SB_NOTIFY;
2359 		}
2360 	}
2361 	sounlock(so);
2362 
2363 	return revents;
2364 }
2365 
2366 
2367 #include <sys/sysctl.h>
2368 
2369 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO);
2370 static int sysctl_kern_sbmax(SYSCTLFN_PROTO);
2371 
2372 /*
2373  * sysctl helper routine for kern.somaxkva.  ensures that the given
2374  * value is not too small.
2375  * (XXX should we maybe make sure it's not too large as well?)
2376  */
2377 static int
2378 sysctl_kern_somaxkva(SYSCTLFN_ARGS)
2379 {
2380 	int error, new_somaxkva;
2381 	struct sysctlnode node;
2382 
2383 	new_somaxkva = somaxkva;
2384 	node = *rnode;
2385 	node.sysctl_data = &new_somaxkva;
2386 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2387 	if (error || newp == NULL)
2388 		return (error);
2389 
2390 	if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */
2391 		return (EINVAL);
2392 
2393 	mutex_enter(&so_pendfree_lock);
2394 	somaxkva = new_somaxkva;
2395 	cv_broadcast(&socurkva_cv);
2396 	mutex_exit(&so_pendfree_lock);
2397 
2398 	return (error);
2399 }
2400 
2401 /*
2402  * sysctl helper routine for kern.sbmax. Basically just ensures that
2403  * any new value is not too small.
2404  */
2405 static int
2406 sysctl_kern_sbmax(SYSCTLFN_ARGS)
2407 {
2408 	int error, new_sbmax;
2409 	struct sysctlnode node;
2410 
2411 	new_sbmax = sb_max;
2412 	node = *rnode;
2413 	node.sysctl_data = &new_sbmax;
2414 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2415 	if (error || newp == NULL)
2416 		return (error);
2417 
2418 	KERNEL_LOCK(1, NULL);
2419 	error = sb_max_set(new_sbmax);
2420 	KERNEL_UNLOCK_ONE(NULL);
2421 
2422 	return (error);
2423 }
2424 
2425 static void
2426 sysctl_kern_socket_setup(void)
2427 {
2428 
2429 	KASSERT(socket_sysctllog == NULL);
2430 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2431 		       CTLFLAG_PERMANENT,
2432 		       CTLTYPE_NODE, "kern", NULL,
2433 		       NULL, 0, NULL, 0,
2434 		       CTL_KERN, CTL_EOL);
2435 
2436 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2437 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2438 		       CTLTYPE_INT, "somaxkva",
2439 		       SYSCTL_DESCR("Maximum amount of kernel memory to be "
2440 				    "used for socket buffers"),
2441 		       sysctl_kern_somaxkva, 0, NULL, 0,
2442 		       CTL_KERN, KERN_SOMAXKVA, CTL_EOL);
2443 
2444 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2445 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2446 		       CTLTYPE_INT, "sbmax",
2447 		       SYSCTL_DESCR("Maximum socket buffer size"),
2448 		       sysctl_kern_sbmax, 0, NULL, 0,
2449 		       CTL_KERN, KERN_SBMAX, CTL_EOL);
2450 }
2451