xref: /netbsd-src/sys/kern/uipc_socket.c (revision 4d12bfcd155352508213ace5ccc59ce930ea2974)
1 /*	$NetBSD: uipc_socket.c,v 1.217 2013/08/29 17:49:21 rmind 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 1.217 2013/08/29 17:49:21 rmind 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) == 0)
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 (resid || clen == 0) {
931 					error = ENOTCONN;
932 					goto release;
933 				}
934 			} else if (addr == 0) {
935 				error = EDESTADDRREQ;
936 				goto release;
937 			}
938 		}
939 		space = sbspace(&so->so_snd);
940 		if (flags & MSG_OOB)
941 			space += 1024;
942 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
943 		    clen > so->so_snd.sb_hiwat) {
944 			error = EMSGSIZE;
945 			goto release;
946 		}
947 		if (space < resid + clen &&
948 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
949 			if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
950 				error = EWOULDBLOCK;
951 				goto release;
952 			}
953 			sbunlock(&so->so_snd);
954 			if (wakeup_state & SS_RESTARTSYS) {
955 				error = ERESTART;
956 				goto out;
957 			}
958 			error = sbwait(&so->so_snd);
959 			if (error)
960 				goto out;
961 			wakeup_state = so->so_state;
962 			goto restart;
963 		}
964 		wakeup_state = 0;
965 		mp = &top;
966 		space -= clen;
967 		do {
968 			if (uio == NULL) {
969 				/*
970 				 * Data is prepackaged in "top".
971 				 */
972 				resid = 0;
973 				if (flags & MSG_EOR)
974 					top->m_flags |= M_EOR;
975 			} else do {
976 				sounlock(so);
977 				splx(s);
978 				if (top == NULL) {
979 					m = m_gethdr(M_WAIT, MT_DATA);
980 					mlen = MHLEN;
981 					m->m_pkthdr.len = 0;
982 					m->m_pkthdr.rcvif = NULL;
983 				} else {
984 					m = m_get(M_WAIT, MT_DATA);
985 					mlen = MLEN;
986 				}
987 				MCLAIM(m, so->so_snd.sb_mowner);
988 				if (sock_loan_thresh >= 0 &&
989 				    uio->uio_iov->iov_len >= sock_loan_thresh &&
990 				    space >= sock_loan_thresh &&
991 				    (len = sosend_loan(so, uio, m,
992 						       space)) != 0) {
993 					SOSEND_COUNTER_INCR(&sosend_loan_big);
994 					space -= len;
995 					goto have_data;
996 				}
997 				if (resid >= MINCLSIZE && space >= MCLBYTES) {
998 					SOSEND_COUNTER_INCR(&sosend_copy_big);
999 					m_clget(m, M_DONTWAIT);
1000 					if ((m->m_flags & M_EXT) == 0)
1001 						goto nopages;
1002 					mlen = MCLBYTES;
1003 					if (atomic && top == 0) {
1004 						len = lmin(MCLBYTES - max_hdr,
1005 						    resid);
1006 						m->m_data += max_hdr;
1007 					} else
1008 						len = lmin(MCLBYTES, resid);
1009 					space -= len;
1010 				} else {
1011  nopages:
1012 					SOSEND_COUNTER_INCR(&sosend_copy_small);
1013 					len = lmin(lmin(mlen, resid), space);
1014 					space -= len;
1015 					/*
1016 					 * For datagram protocols, leave room
1017 					 * for protocol headers in first mbuf.
1018 					 */
1019 					if (atomic && top == 0 && len < mlen)
1020 						MH_ALIGN(m, len);
1021 				}
1022 				error = uiomove(mtod(m, void *), (int)len, uio);
1023  have_data:
1024 				resid = uio->uio_resid;
1025 				m->m_len = len;
1026 				*mp = m;
1027 				top->m_pkthdr.len += len;
1028 				s = splsoftnet();
1029 				solock(so);
1030 				if (error != 0)
1031 					goto release;
1032 				mp = &m->m_next;
1033 				if (resid <= 0) {
1034 					if (flags & MSG_EOR)
1035 						top->m_flags |= M_EOR;
1036 					break;
1037 				}
1038 			} while (space > 0 && atomic);
1039 
1040 			if (so->so_state & SS_CANTSENDMORE) {
1041 				error = EPIPE;
1042 				goto release;
1043 			}
1044 			if (dontroute)
1045 				so->so_options |= SO_DONTROUTE;
1046 			if (resid > 0)
1047 				so->so_state |= SS_MORETOCOME;
1048 			error = (*so->so_proto->pr_usrreq)(so,
1049 			    (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND,
1050 			    top, addr, control, curlwp);
1051 			if (dontroute)
1052 				so->so_options &= ~SO_DONTROUTE;
1053 			if (resid > 0)
1054 				so->so_state &= ~SS_MORETOCOME;
1055 			clen = 0;
1056 			control = NULL;
1057 			top = NULL;
1058 			mp = &top;
1059 			if (error != 0)
1060 				goto release;
1061 		} while (resid && space > 0);
1062 	} while (resid);
1063 
1064  release:
1065 	sbunlock(&so->so_snd);
1066  out:
1067 	sounlock(so);
1068 	splx(s);
1069 	if (top)
1070 		m_freem(top);
1071 	if (control)
1072 		m_freem(control);
1073 	return (error);
1074 }
1075 
1076 /*
1077  * Following replacement or removal of the first mbuf on the first
1078  * mbuf chain of a socket buffer, push necessary state changes back
1079  * into the socket buffer so that other consumers see the values
1080  * consistently.  'nextrecord' is the callers locally stored value of
1081  * the original value of sb->sb_mb->m_nextpkt which must be restored
1082  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
1083  */
1084 static void
1085 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
1086 {
1087 
1088 	KASSERT(solocked(sb->sb_so));
1089 
1090 	/*
1091 	 * First, update for the new value of nextrecord.  If necessary,
1092 	 * make it the first record.
1093 	 */
1094 	if (sb->sb_mb != NULL)
1095 		sb->sb_mb->m_nextpkt = nextrecord;
1096 	else
1097 		sb->sb_mb = nextrecord;
1098 
1099         /*
1100          * Now update any dependent socket buffer fields to reflect
1101          * the new state.  This is an inline of SB_EMPTY_FIXUP, with
1102          * the addition of a second clause that takes care of the
1103          * case where sb_mb has been updated, but remains the last
1104          * record.
1105          */
1106         if (sb->sb_mb == NULL) {
1107                 sb->sb_mbtail = NULL;
1108                 sb->sb_lastrecord = NULL;
1109         } else if (sb->sb_mb->m_nextpkt == NULL)
1110                 sb->sb_lastrecord = sb->sb_mb;
1111 }
1112 
1113 /*
1114  * Implement receive operations on a socket.
1115  * We depend on the way that records are added to the sockbuf
1116  * by sbappend*.  In particular, each record (mbufs linked through m_next)
1117  * must begin with an address if the protocol so specifies,
1118  * followed by an optional mbuf or mbufs containing ancillary data,
1119  * and then zero or more mbufs of data.
1120  * In order to avoid blocking network interrupts for the entire time here,
1121  * we splx() while doing the actual copy to user space.
1122  * Although the sockbuf is locked, new data may still be appended,
1123  * and thus we must maintain consistency of the sockbuf during that time.
1124  *
1125  * The caller may receive the data as a single mbuf chain by supplying
1126  * an mbuf **mp0 for use in returning the chain.  The uio is then used
1127  * only for the count in uio_resid.
1128  */
1129 int
1130 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
1131 	struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1132 {
1133 	struct lwp *l = curlwp;
1134 	struct mbuf	*m, **mp, *mt;
1135 	size_t len, offset, moff, orig_resid;
1136 	int atomic, flags, error, s, type;
1137 	const struct protosw	*pr;
1138 	struct mbuf	*nextrecord;
1139 	int		mbuf_removed = 0;
1140 	const struct domain *dom;
1141 	short		wakeup_state = 0;
1142 
1143 	pr = so->so_proto;
1144 	atomic = pr->pr_flags & PR_ATOMIC;
1145 	dom = pr->pr_domain;
1146 	mp = mp0;
1147 	type = 0;
1148 	orig_resid = uio->uio_resid;
1149 
1150 	if (paddr != NULL)
1151 		*paddr = NULL;
1152 	if (controlp != NULL)
1153 		*controlp = NULL;
1154 	if (flagsp != NULL)
1155 		flags = *flagsp &~ MSG_EOR;
1156 	else
1157 		flags = 0;
1158 
1159 	if (flags & MSG_OOB) {
1160 		m = m_get(M_WAIT, MT_DATA);
1161 		solock(so);
1162 		error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m,
1163 		    (struct mbuf *)(long)(flags & MSG_PEEK), NULL, l);
1164 		sounlock(so);
1165 		if (error)
1166 			goto bad;
1167 		do {
1168 			error = uiomove(mtod(m, void *),
1169 			    MIN(uio->uio_resid, m->m_len), uio);
1170 			m = m_free(m);
1171 		} while (uio->uio_resid > 0 && error == 0 && m);
1172  bad:
1173 		if (m != NULL)
1174 			m_freem(m);
1175 		return error;
1176 	}
1177 	if (mp != NULL)
1178 		*mp = NULL;
1179 
1180 	/*
1181 	 * solock() provides atomicity of access.  splsoftnet() prevents
1182 	 * protocol processing soft interrupts from interrupting us and
1183 	 * blocking (expensive).
1184 	 */
1185 	s = splsoftnet();
1186 	solock(so);
1187  restart:
1188 	if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) {
1189 		sounlock(so);
1190 		splx(s);
1191 		return error;
1192 	}
1193 
1194 	m = so->so_rcv.sb_mb;
1195 	/*
1196 	 * If we have less data than requested, block awaiting more
1197 	 * (subject to any timeout) if:
1198 	 *   1. the current count is less than the low water mark,
1199 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1200 	 *	receive operation at once if we block (resid <= hiwat), or
1201 	 *   3. MSG_DONTWAIT is not set.
1202 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1203 	 * we have to do the receive in sections, and thus risk returning
1204 	 * a short count if a timeout or signal occurs after we start.
1205 	 */
1206 	if (m == NULL ||
1207 	    ((flags & MSG_DONTWAIT) == 0 &&
1208 	     so->so_rcv.sb_cc < uio->uio_resid &&
1209 	     (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1210 	      ((flags & MSG_WAITALL) &&
1211 	       uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1212 	     m->m_nextpkt == NULL && !atomic)) {
1213 #ifdef DIAGNOSTIC
1214 		if (m == NULL && so->so_rcv.sb_cc)
1215 			panic("receive 1");
1216 #endif
1217 		if (so->so_error) {
1218 			if (m != NULL)
1219 				goto dontblock;
1220 			error = so->so_error;
1221 			if ((flags & MSG_PEEK) == 0)
1222 				so->so_error = 0;
1223 			goto release;
1224 		}
1225 		if (so->so_state & SS_CANTRCVMORE) {
1226 			if (m != NULL)
1227 				goto dontblock;
1228 			else
1229 				goto release;
1230 		}
1231 		for (; m != NULL; m = m->m_next)
1232 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1233 				m = so->so_rcv.sb_mb;
1234 				goto dontblock;
1235 			}
1236 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1237 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1238 			error = ENOTCONN;
1239 			goto release;
1240 		}
1241 		if (uio->uio_resid == 0)
1242 			goto release;
1243 		if ((so->so_state & SS_NBIO) ||
1244 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1245 			error = EWOULDBLOCK;
1246 			goto release;
1247 		}
1248 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
1249 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
1250 		sbunlock(&so->so_rcv);
1251 		if (wakeup_state & SS_RESTARTSYS)
1252 			error = ERESTART;
1253 		else
1254 			error = sbwait(&so->so_rcv);
1255 		if (error != 0) {
1256 			sounlock(so);
1257 			splx(s);
1258 			return error;
1259 		}
1260 		wakeup_state = so->so_state;
1261 		goto restart;
1262 	}
1263  dontblock:
1264 	/*
1265 	 * On entry here, m points to the first record of the socket buffer.
1266 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1267 	 * pointer to the next record in the socket buffer.  We must keep the
1268 	 * various socket buffer pointers and local stack versions of the
1269 	 * pointers in sync, pushing out modifications before dropping the
1270 	 * socket lock, and re-reading them when picking it up.
1271 	 *
1272 	 * Otherwise, we will race with the network stack appending new data
1273 	 * or records onto the socket buffer by using inconsistent/stale
1274 	 * versions of the field, possibly resulting in socket buffer
1275 	 * corruption.
1276 	 *
1277 	 * By holding the high-level sblock(), we prevent simultaneous
1278 	 * readers from pulling off the front of the socket buffer.
1279 	 */
1280 	if (l != NULL)
1281 		l->l_ru.ru_msgrcv++;
1282 	KASSERT(m == so->so_rcv.sb_mb);
1283 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
1284 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
1285 	nextrecord = m->m_nextpkt;
1286 	if (pr->pr_flags & PR_ADDR) {
1287 #ifdef DIAGNOSTIC
1288 		if (m->m_type != MT_SONAME)
1289 			panic("receive 1a");
1290 #endif
1291 		orig_resid = 0;
1292 		if (flags & MSG_PEEK) {
1293 			if (paddr)
1294 				*paddr = m_copy(m, 0, m->m_len);
1295 			m = m->m_next;
1296 		} else {
1297 			sbfree(&so->so_rcv, m);
1298 			mbuf_removed = 1;
1299 			if (paddr != NULL) {
1300 				*paddr = m;
1301 				so->so_rcv.sb_mb = m->m_next;
1302 				m->m_next = NULL;
1303 				m = so->so_rcv.sb_mb;
1304 			} else {
1305 				MFREE(m, so->so_rcv.sb_mb);
1306 				m = so->so_rcv.sb_mb;
1307 			}
1308 			sbsync(&so->so_rcv, nextrecord);
1309 		}
1310 	}
1311 
1312 	/*
1313 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1314 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1315 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1316 	 * perform externalization (or freeing if controlp == NULL).
1317 	 */
1318 	if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) {
1319 		struct mbuf *cm = NULL, *cmn;
1320 		struct mbuf **cme = &cm;
1321 
1322 		do {
1323 			if (flags & MSG_PEEK) {
1324 				if (controlp != NULL) {
1325 					*controlp = m_copy(m, 0, m->m_len);
1326 					controlp = &(*controlp)->m_next;
1327 				}
1328 				m = m->m_next;
1329 			} else {
1330 				sbfree(&so->so_rcv, m);
1331 				so->so_rcv.sb_mb = m->m_next;
1332 				m->m_next = NULL;
1333 				*cme = m;
1334 				cme = &(*cme)->m_next;
1335 				m = so->so_rcv.sb_mb;
1336 			}
1337 		} while (m != NULL && m->m_type == MT_CONTROL);
1338 		if ((flags & MSG_PEEK) == 0)
1339 			sbsync(&so->so_rcv, nextrecord);
1340 		for (; cm != NULL; cm = cmn) {
1341 			cmn = cm->m_next;
1342 			cm->m_next = NULL;
1343 			type = mtod(cm, struct cmsghdr *)->cmsg_type;
1344 			if (controlp != NULL) {
1345 				if (dom->dom_externalize != NULL &&
1346 				    type == SCM_RIGHTS) {
1347 					sounlock(so);
1348 					splx(s);
1349 					error = (*dom->dom_externalize)(cm, l,
1350 					    (flags & MSG_CMSG_CLOEXEC) ?
1351 					    O_CLOEXEC : 0);
1352 					s = splsoftnet();
1353 					solock(so);
1354 				}
1355 				*controlp = cm;
1356 				while (*controlp != NULL)
1357 					controlp = &(*controlp)->m_next;
1358 			} else {
1359 				/*
1360 				 * Dispose of any SCM_RIGHTS message that went
1361 				 * through the read path rather than recv.
1362 				 */
1363 				if (dom->dom_dispose != NULL &&
1364 				    type == SCM_RIGHTS) {
1365 				    	sounlock(so);
1366 					(*dom->dom_dispose)(cm);
1367 					solock(so);
1368 				}
1369 				m_freem(cm);
1370 			}
1371 		}
1372 		if (m != NULL)
1373 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1374 		else
1375 			nextrecord = so->so_rcv.sb_mb;
1376 		orig_resid = 0;
1377 	}
1378 
1379 	/* If m is non-NULL, we have some data to read. */
1380 	if (__predict_true(m != NULL)) {
1381 		type = m->m_type;
1382 		if (type == MT_OOBDATA)
1383 			flags |= MSG_OOB;
1384 	}
1385 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1386 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1387 
1388 	moff = 0;
1389 	offset = 0;
1390 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1391 		if (m->m_type == MT_OOBDATA) {
1392 			if (type != MT_OOBDATA)
1393 				break;
1394 		} else if (type == MT_OOBDATA)
1395 			break;
1396 #ifdef DIAGNOSTIC
1397 		else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
1398 			panic("receive 3");
1399 #endif
1400 		so->so_state &= ~SS_RCVATMARK;
1401 		wakeup_state = 0;
1402 		len = uio->uio_resid;
1403 		if (so->so_oobmark && len > so->so_oobmark - offset)
1404 			len = so->so_oobmark - offset;
1405 		if (len > m->m_len - moff)
1406 			len = m->m_len - moff;
1407 		/*
1408 		 * If mp is set, just pass back the mbufs.
1409 		 * Otherwise copy them out via the uio, then free.
1410 		 * Sockbuf must be consistent here (points to current mbuf,
1411 		 * it points to next record) when we drop priority;
1412 		 * we must note any additions to the sockbuf when we
1413 		 * block interrupts again.
1414 		 */
1415 		if (mp == NULL) {
1416 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1417 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1418 			sounlock(so);
1419 			splx(s);
1420 			error = uiomove(mtod(m, char *) + moff, len, uio);
1421 			s = splsoftnet();
1422 			solock(so);
1423 			if (error != 0) {
1424 				/*
1425 				 * If any part of the record has been removed
1426 				 * (such as the MT_SONAME mbuf, which will
1427 				 * happen when PR_ADDR, and thus also
1428 				 * PR_ATOMIC, is set), then drop the entire
1429 				 * record to maintain the atomicity of the
1430 				 * receive operation.
1431 				 *
1432 				 * This avoids a later panic("receive 1a")
1433 				 * when compiled with DIAGNOSTIC.
1434 				 */
1435 				if (m && mbuf_removed && atomic)
1436 					(void) sbdroprecord(&so->so_rcv);
1437 
1438 				goto release;
1439 			}
1440 		} else
1441 			uio->uio_resid -= len;
1442 		if (len == m->m_len - moff) {
1443 			if (m->m_flags & M_EOR)
1444 				flags |= MSG_EOR;
1445 			if (flags & MSG_PEEK) {
1446 				m = m->m_next;
1447 				moff = 0;
1448 			} else {
1449 				nextrecord = m->m_nextpkt;
1450 				sbfree(&so->so_rcv, m);
1451 				if (mp) {
1452 					*mp = m;
1453 					mp = &m->m_next;
1454 					so->so_rcv.sb_mb = m = m->m_next;
1455 					*mp = NULL;
1456 				} else {
1457 					MFREE(m, so->so_rcv.sb_mb);
1458 					m = so->so_rcv.sb_mb;
1459 				}
1460 				/*
1461 				 * If m != NULL, we also know that
1462 				 * so->so_rcv.sb_mb != NULL.
1463 				 */
1464 				KASSERT(so->so_rcv.sb_mb == m);
1465 				if (m) {
1466 					m->m_nextpkt = nextrecord;
1467 					if (nextrecord == NULL)
1468 						so->so_rcv.sb_lastrecord = m;
1469 				} else {
1470 					so->so_rcv.sb_mb = nextrecord;
1471 					SB_EMPTY_FIXUP(&so->so_rcv);
1472 				}
1473 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1474 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1475 			}
1476 		} else if (flags & MSG_PEEK)
1477 			moff += len;
1478 		else {
1479 			if (mp != NULL) {
1480 				mt = m_copym(m, 0, len, M_NOWAIT);
1481 				if (__predict_false(mt == NULL)) {
1482 					sounlock(so);
1483 					mt = m_copym(m, 0, len, M_WAIT);
1484 					solock(so);
1485 				}
1486 				*mp = mt;
1487 			}
1488 			m->m_data += len;
1489 			m->m_len -= len;
1490 			so->so_rcv.sb_cc -= len;
1491 		}
1492 		if (so->so_oobmark) {
1493 			if ((flags & MSG_PEEK) == 0) {
1494 				so->so_oobmark -= len;
1495 				if (so->so_oobmark == 0) {
1496 					so->so_state |= SS_RCVATMARK;
1497 					break;
1498 				}
1499 			} else {
1500 				offset += len;
1501 				if (offset == so->so_oobmark)
1502 					break;
1503 			}
1504 		}
1505 		if (flags & MSG_EOR)
1506 			break;
1507 		/*
1508 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1509 		 * we must not quit until "uio->uio_resid == 0" or an error
1510 		 * termination.  If a signal/timeout occurs, return
1511 		 * with a short count but without error.
1512 		 * Keep sockbuf locked against other readers.
1513 		 */
1514 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1515 		    !sosendallatonce(so) && !nextrecord) {
1516 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1517 				break;
1518 			/*
1519 			 * If we are peeking and the socket receive buffer is
1520 			 * full, stop since we can't get more data to peek at.
1521 			 */
1522 			if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
1523 				break;
1524 			/*
1525 			 * If we've drained the socket buffer, tell the
1526 			 * protocol in case it needs to do something to
1527 			 * get it filled again.
1528 			 */
1529 			if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1530 				(*pr->pr_usrreq)(so, PRU_RCVD,
1531 				    NULL, (struct mbuf *)(long)flags, NULL, l);
1532 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1533 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1534 			if (wakeup_state & SS_RESTARTSYS)
1535 				error = ERESTART;
1536 			else
1537 				error = sbwait(&so->so_rcv);
1538 			if (error != 0) {
1539 				sbunlock(&so->so_rcv);
1540 				sounlock(so);
1541 				splx(s);
1542 				return 0;
1543 			}
1544 			if ((m = so->so_rcv.sb_mb) != NULL)
1545 				nextrecord = m->m_nextpkt;
1546 			wakeup_state = so->so_state;
1547 		}
1548 	}
1549 
1550 	if (m && atomic) {
1551 		flags |= MSG_TRUNC;
1552 		if ((flags & MSG_PEEK) == 0)
1553 			(void) sbdroprecord(&so->so_rcv);
1554 	}
1555 	if ((flags & MSG_PEEK) == 0) {
1556 		if (m == NULL) {
1557 			/*
1558 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1559 			 * part makes sure sb_lastrecord is up-to-date if
1560 			 * there is still data in the socket buffer.
1561 			 */
1562 			so->so_rcv.sb_mb = nextrecord;
1563 			if (so->so_rcv.sb_mb == NULL) {
1564 				so->so_rcv.sb_mbtail = NULL;
1565 				so->so_rcv.sb_lastrecord = NULL;
1566 			} else if (nextrecord->m_nextpkt == NULL)
1567 				so->so_rcv.sb_lastrecord = nextrecord;
1568 		}
1569 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1570 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1571 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1572 			(*pr->pr_usrreq)(so, PRU_RCVD, NULL,
1573 			    (struct mbuf *)(long)flags, NULL, l);
1574 	}
1575 	if (orig_resid == uio->uio_resid && orig_resid &&
1576 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1577 		sbunlock(&so->so_rcv);
1578 		goto restart;
1579 	}
1580 
1581 	if (flagsp != NULL)
1582 		*flagsp |= flags;
1583  release:
1584 	sbunlock(&so->so_rcv);
1585 	sounlock(so);
1586 	splx(s);
1587 	return error;
1588 }
1589 
1590 int
1591 soshutdown(struct socket *so, int how)
1592 {
1593 	const struct protosw	*pr;
1594 	int	error;
1595 
1596 	KASSERT(solocked(so));
1597 
1598 	pr = so->so_proto;
1599 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1600 		return (EINVAL);
1601 
1602 	if (how == SHUT_RD || how == SHUT_RDWR) {
1603 		sorflush(so);
1604 		error = 0;
1605 	}
1606 	if (how == SHUT_WR || how == SHUT_RDWR)
1607 		error = (*pr->pr_usrreq)(so, PRU_SHUTDOWN, NULL,
1608 		    NULL, NULL, NULL);
1609 
1610 	return error;
1611 }
1612 
1613 void
1614 sorestart(struct socket *so)
1615 {
1616 	/*
1617 	 * An application has called close() on an fd on which another
1618 	 * of its threads has called a socket system call.
1619 	 * Mark this and wake everyone up, and code that would block again
1620 	 * instead returns ERESTART.
1621 	 * On system call re-entry the fd is validated and EBADF returned.
1622 	 * Any other fd will block again on the 2nd syscall.
1623 	 */
1624 	solock(so);
1625 	so->so_state |= SS_RESTARTSYS;
1626 	cv_broadcast(&so->so_cv);
1627 	cv_broadcast(&so->so_snd.sb_cv);
1628 	cv_broadcast(&so->so_rcv.sb_cv);
1629 	sounlock(so);
1630 }
1631 
1632 void
1633 sorflush(struct socket *so)
1634 {
1635 	struct sockbuf	*sb, asb;
1636 	const struct protosw	*pr;
1637 
1638 	KASSERT(solocked(so));
1639 
1640 	sb = &so->so_rcv;
1641 	pr = so->so_proto;
1642 	socantrcvmore(so);
1643 	sb->sb_flags |= SB_NOINTR;
1644 	(void )sblock(sb, M_WAITOK);
1645 	sbunlock(sb);
1646 	asb = *sb;
1647 	/*
1648 	 * Clear most of the sockbuf structure, but leave some of the
1649 	 * fields valid.
1650 	 */
1651 	memset(&sb->sb_startzero, 0,
1652 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1653 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) {
1654 		sounlock(so);
1655 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1656 		solock(so);
1657 	}
1658 	sbrelease(&asb, so);
1659 }
1660 
1661 /*
1662  * internal set SOL_SOCKET options
1663  */
1664 static int
1665 sosetopt1(struct socket *so, const struct sockopt *sopt)
1666 {
1667 	int error = EINVAL, optval, opt;
1668 	struct linger l;
1669 	struct timeval tv;
1670 
1671 	switch ((opt = sopt->sopt_name)) {
1672 
1673 	case SO_ACCEPTFILTER:
1674 		error = accept_filt_setopt(so, sopt);
1675 		KASSERT(solocked(so));
1676 		break;
1677 
1678   	case SO_LINGER:
1679  		error = sockopt_get(sopt, &l, sizeof(l));
1680 		solock(so);
1681  		if (error)
1682  			break;
1683  		if (l.l_linger < 0 || l.l_linger > USHRT_MAX ||
1684  		    l.l_linger > (INT_MAX / hz)) {
1685 			error = EDOM;
1686 			break;
1687 		}
1688  		so->so_linger = l.l_linger;
1689  		if (l.l_onoff)
1690  			so->so_options |= SO_LINGER;
1691  		else
1692  			so->so_options &= ~SO_LINGER;
1693    		break;
1694 
1695 	case SO_DEBUG:
1696 	case SO_KEEPALIVE:
1697 	case SO_DONTROUTE:
1698 	case SO_USELOOPBACK:
1699 	case SO_BROADCAST:
1700 	case SO_REUSEADDR:
1701 	case SO_REUSEPORT:
1702 	case SO_OOBINLINE:
1703 	case SO_TIMESTAMP:
1704 	case SO_NOSIGPIPE:
1705 #ifdef SO_OTIMESTAMP
1706 	case SO_OTIMESTAMP:
1707 #endif
1708 		error = sockopt_getint(sopt, &optval);
1709 		solock(so);
1710 		if (error)
1711 			break;
1712 		if (optval)
1713 			so->so_options |= opt;
1714 		else
1715 			so->so_options &= ~opt;
1716 		break;
1717 
1718 	case SO_SNDBUF:
1719 	case SO_RCVBUF:
1720 	case SO_SNDLOWAT:
1721 	case SO_RCVLOWAT:
1722 		error = sockopt_getint(sopt, &optval);
1723 		solock(so);
1724 		if (error)
1725 			break;
1726 
1727 		/*
1728 		 * Values < 1 make no sense for any of these
1729 		 * options, so disallow them.
1730 		 */
1731 		if (optval < 1) {
1732 			error = EINVAL;
1733 			break;
1734 		}
1735 
1736 		switch (opt) {
1737 		case SO_SNDBUF:
1738 			if (sbreserve(&so->so_snd, (u_long)optval, so) == 0) {
1739 				error = ENOBUFS;
1740 				break;
1741 			}
1742 			so->so_snd.sb_flags &= ~SB_AUTOSIZE;
1743 			break;
1744 
1745 		case SO_RCVBUF:
1746 			if (sbreserve(&so->so_rcv, (u_long)optval, so) == 0) {
1747 				error = ENOBUFS;
1748 				break;
1749 			}
1750 			so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1751 			break;
1752 
1753 		/*
1754 		 * Make sure the low-water is never greater than
1755 		 * the high-water.
1756 		 */
1757 		case SO_SNDLOWAT:
1758 			if (optval > so->so_snd.sb_hiwat)
1759 				optval = so->so_snd.sb_hiwat;
1760 
1761 			so->so_snd.sb_lowat = optval;
1762 			break;
1763 
1764 		case SO_RCVLOWAT:
1765 			if (optval > so->so_rcv.sb_hiwat)
1766 				optval = so->so_rcv.sb_hiwat;
1767 
1768 			so->so_rcv.sb_lowat = optval;
1769 			break;
1770 		}
1771 		break;
1772 
1773 #ifdef COMPAT_50
1774 	case SO_OSNDTIMEO:
1775 	case SO_ORCVTIMEO: {
1776 		struct timeval50 otv;
1777 		error = sockopt_get(sopt, &otv, sizeof(otv));
1778 		if (error) {
1779 			solock(so);
1780 			break;
1781 		}
1782 		timeval50_to_timeval(&otv, &tv);
1783 		opt = opt == SO_OSNDTIMEO ? SO_SNDTIMEO : SO_RCVTIMEO;
1784 		error = 0;
1785 		/*FALLTHROUGH*/
1786 	}
1787 #endif /* COMPAT_50 */
1788 
1789 	case SO_SNDTIMEO:
1790 	case SO_RCVTIMEO:
1791 		if (error)
1792 			error = sockopt_get(sopt, &tv, sizeof(tv));
1793 		solock(so);
1794 		if (error)
1795 			break;
1796 
1797 		if (tv.tv_sec > (INT_MAX - tv.tv_usec / tick) / hz) {
1798 			error = EDOM;
1799 			break;
1800 		}
1801 
1802 		optval = tv.tv_sec * hz + tv.tv_usec / tick;
1803 		if (optval == 0 && tv.tv_usec != 0)
1804 			optval = 1;
1805 
1806 		switch (opt) {
1807 		case SO_SNDTIMEO:
1808 			so->so_snd.sb_timeo = optval;
1809 			break;
1810 		case SO_RCVTIMEO:
1811 			so->so_rcv.sb_timeo = optval;
1812 			break;
1813 		}
1814 		break;
1815 
1816 	default:
1817 		solock(so);
1818 		error = ENOPROTOOPT;
1819 		break;
1820 	}
1821 	KASSERT(solocked(so));
1822 	return error;
1823 }
1824 
1825 int
1826 sosetopt(struct socket *so, struct sockopt *sopt)
1827 {
1828 	int error, prerr;
1829 
1830 	if (sopt->sopt_level == SOL_SOCKET) {
1831 		error = sosetopt1(so, sopt);
1832 		KASSERT(solocked(so));
1833 	} else {
1834 		error = ENOPROTOOPT;
1835 		solock(so);
1836 	}
1837 
1838 	if ((error == 0 || error == ENOPROTOOPT) &&
1839 	    so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) {
1840 		/* give the protocol stack a shot */
1841 		prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, sopt);
1842 		if (prerr == 0)
1843 			error = 0;
1844 		else if (prerr != ENOPROTOOPT)
1845 			error = prerr;
1846 	}
1847 	sounlock(so);
1848 	return error;
1849 }
1850 
1851 /*
1852  * so_setsockopt() is a wrapper providing a sockopt structure for sosetopt()
1853  */
1854 int
1855 so_setsockopt(struct lwp *l, struct socket *so, int level, int name,
1856     const void *val, size_t valsize)
1857 {
1858 	struct sockopt sopt;
1859 	int error;
1860 
1861 	KASSERT(valsize == 0 || val != NULL);
1862 
1863 	sockopt_init(&sopt, level, name, valsize);
1864 	sockopt_set(&sopt, val, valsize);
1865 
1866 	error = sosetopt(so, &sopt);
1867 
1868 	sockopt_destroy(&sopt);
1869 
1870 	return error;
1871 }
1872 
1873 /*
1874  * internal get SOL_SOCKET options
1875  */
1876 static int
1877 sogetopt1(struct socket *so, struct sockopt *sopt)
1878 {
1879 	int error, optval, opt;
1880 	struct linger l;
1881 	struct timeval tv;
1882 
1883 	switch ((opt = sopt->sopt_name)) {
1884 
1885 	case SO_ACCEPTFILTER:
1886 		error = accept_filt_getopt(so, sopt);
1887 		break;
1888 
1889 	case SO_LINGER:
1890 		l.l_onoff = (so->so_options & SO_LINGER) ? 1 : 0;
1891 		l.l_linger = so->so_linger;
1892 
1893 		error = sockopt_set(sopt, &l, sizeof(l));
1894 		break;
1895 
1896 	case SO_USELOOPBACK:
1897 	case SO_DONTROUTE:
1898 	case SO_DEBUG:
1899 	case SO_KEEPALIVE:
1900 	case SO_REUSEADDR:
1901 	case SO_REUSEPORT:
1902 	case SO_BROADCAST:
1903 	case SO_OOBINLINE:
1904 	case SO_TIMESTAMP:
1905 	case SO_NOSIGPIPE:
1906 #ifdef SO_OTIMESTAMP
1907 	case SO_OTIMESTAMP:
1908 #endif
1909 		error = sockopt_setint(sopt, (so->so_options & opt) ? 1 : 0);
1910 		break;
1911 
1912 	case SO_TYPE:
1913 		error = sockopt_setint(sopt, so->so_type);
1914 		break;
1915 
1916 	case SO_ERROR:
1917 		error = sockopt_setint(sopt, so->so_error);
1918 		so->so_error = 0;
1919 		break;
1920 
1921 	case SO_SNDBUF:
1922 		error = sockopt_setint(sopt, so->so_snd.sb_hiwat);
1923 		break;
1924 
1925 	case SO_RCVBUF:
1926 		error = sockopt_setint(sopt, so->so_rcv.sb_hiwat);
1927 		break;
1928 
1929 	case SO_SNDLOWAT:
1930 		error = sockopt_setint(sopt, so->so_snd.sb_lowat);
1931 		break;
1932 
1933 	case SO_RCVLOWAT:
1934 		error = sockopt_setint(sopt, so->so_rcv.sb_lowat);
1935 		break;
1936 
1937 #ifdef COMPAT_50
1938 	case SO_OSNDTIMEO:
1939 	case SO_ORCVTIMEO: {
1940 		struct timeval50 otv;
1941 
1942 		optval = (opt == SO_OSNDTIMEO ?
1943 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1944 
1945 		otv.tv_sec = optval / hz;
1946 		otv.tv_usec = (optval % hz) * tick;
1947 
1948 		error = sockopt_set(sopt, &otv, sizeof(otv));
1949 		break;
1950 	}
1951 #endif /* COMPAT_50 */
1952 
1953 	case SO_SNDTIMEO:
1954 	case SO_RCVTIMEO:
1955 		optval = (opt == SO_SNDTIMEO ?
1956 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1957 
1958 		tv.tv_sec = optval / hz;
1959 		tv.tv_usec = (optval % hz) * tick;
1960 
1961 		error = sockopt_set(sopt, &tv, sizeof(tv));
1962 		break;
1963 
1964 	case SO_OVERFLOWED:
1965 		error = sockopt_setint(sopt, so->so_rcv.sb_overflowed);
1966 		break;
1967 
1968 	default:
1969 		error = ENOPROTOOPT;
1970 		break;
1971 	}
1972 
1973 	return (error);
1974 }
1975 
1976 int
1977 sogetopt(struct socket *so, struct sockopt *sopt)
1978 {
1979 	int		error;
1980 
1981 	solock(so);
1982 	if (sopt->sopt_level != SOL_SOCKET) {
1983 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1984 			error = ((*so->so_proto->pr_ctloutput)
1985 			    (PRCO_GETOPT, so, sopt));
1986 		} else
1987 			error = (ENOPROTOOPT);
1988 	} else {
1989 		error = sogetopt1(so, sopt);
1990 	}
1991 	sounlock(so);
1992 	return (error);
1993 }
1994 
1995 /*
1996  * alloc sockopt data buffer buffer
1997  *	- will be released at destroy
1998  */
1999 static int
2000 sockopt_alloc(struct sockopt *sopt, size_t len, km_flag_t kmflag)
2001 {
2002 
2003 	KASSERT(sopt->sopt_size == 0);
2004 
2005 	if (len > sizeof(sopt->sopt_buf)) {
2006 		sopt->sopt_data = kmem_zalloc(len, kmflag);
2007 		if (sopt->sopt_data == NULL)
2008 			return ENOMEM;
2009 	} else
2010 		sopt->sopt_data = sopt->sopt_buf;
2011 
2012 	sopt->sopt_size = len;
2013 	return 0;
2014 }
2015 
2016 /*
2017  * initialise sockopt storage
2018  *	- MAY sleep during allocation
2019  */
2020 void
2021 sockopt_init(struct sockopt *sopt, int level, int name, size_t size)
2022 {
2023 
2024 	memset(sopt, 0, sizeof(*sopt));
2025 
2026 	sopt->sopt_level = level;
2027 	sopt->sopt_name = name;
2028 	(void)sockopt_alloc(sopt, size, KM_SLEEP);
2029 }
2030 
2031 /*
2032  * destroy sockopt storage
2033  *	- will release any held memory references
2034  */
2035 void
2036 sockopt_destroy(struct sockopt *sopt)
2037 {
2038 
2039 	if (sopt->sopt_data != sopt->sopt_buf)
2040 		kmem_free(sopt->sopt_data, sopt->sopt_size);
2041 
2042 	memset(sopt, 0, sizeof(*sopt));
2043 }
2044 
2045 /*
2046  * set sockopt value
2047  *	- value is copied into sockopt
2048  * 	- memory is allocated when necessary, will not sleep
2049  */
2050 int
2051 sockopt_set(struct sockopt *sopt, const void *buf, size_t len)
2052 {
2053 	int error;
2054 
2055 	if (sopt->sopt_size == 0) {
2056 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2057 		if (error)
2058 			return error;
2059 	}
2060 
2061 	KASSERT(sopt->sopt_size == len);
2062 	memcpy(sopt->sopt_data, buf, len);
2063 	return 0;
2064 }
2065 
2066 /*
2067  * common case of set sockopt integer value
2068  */
2069 int
2070 sockopt_setint(struct sockopt *sopt, int val)
2071 {
2072 
2073 	return sockopt_set(sopt, &val, sizeof(int));
2074 }
2075 
2076 /*
2077  * get sockopt value
2078  *	- correct size must be given
2079  */
2080 int
2081 sockopt_get(const struct sockopt *sopt, void *buf, size_t len)
2082 {
2083 
2084 	if (sopt->sopt_size != len)
2085 		return EINVAL;
2086 
2087 	memcpy(buf, sopt->sopt_data, len);
2088 	return 0;
2089 }
2090 
2091 /*
2092  * common case of get sockopt integer value
2093  */
2094 int
2095 sockopt_getint(const struct sockopt *sopt, int *valp)
2096 {
2097 
2098 	return sockopt_get(sopt, valp, sizeof(int));
2099 }
2100 
2101 /*
2102  * set sockopt value from mbuf
2103  *	- ONLY for legacy code
2104  *	- mbuf is released by sockopt
2105  *	- will not sleep
2106  */
2107 int
2108 sockopt_setmbuf(struct sockopt *sopt, struct mbuf *m)
2109 {
2110 	size_t len;
2111 	int error;
2112 
2113 	len = m_length(m);
2114 
2115 	if (sopt->sopt_size == 0) {
2116 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2117 		if (error)
2118 			return error;
2119 	}
2120 
2121 	KASSERT(sopt->sopt_size == len);
2122 	m_copydata(m, 0, len, sopt->sopt_data);
2123 	m_freem(m);
2124 
2125 	return 0;
2126 }
2127 
2128 /*
2129  * get sockopt value into mbuf
2130  *	- ONLY for legacy code
2131  *	- mbuf to be released by the caller
2132  *	- will not sleep
2133  */
2134 struct mbuf *
2135 sockopt_getmbuf(const struct sockopt *sopt)
2136 {
2137 	struct mbuf *m;
2138 
2139 	if (sopt->sopt_size > MCLBYTES)
2140 		return NULL;
2141 
2142 	m = m_get(M_DONTWAIT, MT_SOOPTS);
2143 	if (m == NULL)
2144 		return NULL;
2145 
2146 	if (sopt->sopt_size > MLEN) {
2147 		MCLGET(m, M_DONTWAIT);
2148 		if ((m->m_flags & M_EXT) == 0) {
2149 			m_free(m);
2150 			return NULL;
2151 		}
2152 	}
2153 
2154 	memcpy(mtod(m, void *), sopt->sopt_data, sopt->sopt_size);
2155 	m->m_len = sopt->sopt_size;
2156 
2157 	return m;
2158 }
2159 
2160 void
2161 sohasoutofband(struct socket *so)
2162 {
2163 
2164 	fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
2165 	selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, NOTE_SUBMIT);
2166 }
2167 
2168 static void
2169 filt_sordetach(struct knote *kn)
2170 {
2171 	struct socket	*so;
2172 
2173 	so = ((file_t *)kn->kn_obj)->f_data;
2174 	solock(so);
2175 	SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
2176 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
2177 		so->so_rcv.sb_flags &= ~SB_KNOTE;
2178 	sounlock(so);
2179 }
2180 
2181 /*ARGSUSED*/
2182 static int
2183 filt_soread(struct knote *kn, long hint)
2184 {
2185 	struct socket	*so;
2186 	int rv;
2187 
2188 	so = ((file_t *)kn->kn_obj)->f_data;
2189 	if (hint != NOTE_SUBMIT)
2190 		solock(so);
2191 	kn->kn_data = so->so_rcv.sb_cc;
2192 	if (so->so_state & SS_CANTRCVMORE) {
2193 		kn->kn_flags |= EV_EOF;
2194 		kn->kn_fflags = so->so_error;
2195 		rv = 1;
2196 	} else if (so->so_error)	/* temporary udp error */
2197 		rv = 1;
2198 	else if (kn->kn_sfflags & NOTE_LOWAT)
2199 		rv = (kn->kn_data >= kn->kn_sdata);
2200 	else
2201 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
2202 	if (hint != NOTE_SUBMIT)
2203 		sounlock(so);
2204 	return rv;
2205 }
2206 
2207 static void
2208 filt_sowdetach(struct knote *kn)
2209 {
2210 	struct socket	*so;
2211 
2212 	so = ((file_t *)kn->kn_obj)->f_data;
2213 	solock(so);
2214 	SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
2215 	if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
2216 		so->so_snd.sb_flags &= ~SB_KNOTE;
2217 	sounlock(so);
2218 }
2219 
2220 /*ARGSUSED*/
2221 static int
2222 filt_sowrite(struct knote *kn, long hint)
2223 {
2224 	struct socket	*so;
2225 	int rv;
2226 
2227 	so = ((file_t *)kn->kn_obj)->f_data;
2228 	if (hint != NOTE_SUBMIT)
2229 		solock(so);
2230 	kn->kn_data = sbspace(&so->so_snd);
2231 	if (so->so_state & SS_CANTSENDMORE) {
2232 		kn->kn_flags |= EV_EOF;
2233 		kn->kn_fflags = so->so_error;
2234 		rv = 1;
2235 	} else if (so->so_error)	/* temporary udp error */
2236 		rv = 1;
2237 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2238 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
2239 		rv = 0;
2240 	else if (kn->kn_sfflags & NOTE_LOWAT)
2241 		rv = (kn->kn_data >= kn->kn_sdata);
2242 	else
2243 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
2244 	if (hint != NOTE_SUBMIT)
2245 		sounlock(so);
2246 	return rv;
2247 }
2248 
2249 /*ARGSUSED*/
2250 static int
2251 filt_solisten(struct knote *kn, long hint)
2252 {
2253 	struct socket	*so;
2254 	int rv;
2255 
2256 	so = ((file_t *)kn->kn_obj)->f_data;
2257 
2258 	/*
2259 	 * Set kn_data to number of incoming connections, not
2260 	 * counting partial (incomplete) connections.
2261 	 */
2262 	if (hint != NOTE_SUBMIT)
2263 		solock(so);
2264 	kn->kn_data = so->so_qlen;
2265 	rv = (kn->kn_data > 0);
2266 	if (hint != NOTE_SUBMIT)
2267 		sounlock(so);
2268 	return rv;
2269 }
2270 
2271 static const struct filterops solisten_filtops =
2272 	{ 1, NULL, filt_sordetach, filt_solisten };
2273 static const struct filterops soread_filtops =
2274 	{ 1, NULL, filt_sordetach, filt_soread };
2275 static const struct filterops sowrite_filtops =
2276 	{ 1, NULL, filt_sowdetach, filt_sowrite };
2277 
2278 int
2279 soo_kqfilter(struct file *fp, struct knote *kn)
2280 {
2281 	struct socket	*so;
2282 	struct sockbuf	*sb;
2283 
2284 	so = ((file_t *)kn->kn_obj)->f_data;
2285 	solock(so);
2286 	switch (kn->kn_filter) {
2287 	case EVFILT_READ:
2288 		if (so->so_options & SO_ACCEPTCONN)
2289 			kn->kn_fop = &solisten_filtops;
2290 		else
2291 			kn->kn_fop = &soread_filtops;
2292 		sb = &so->so_rcv;
2293 		break;
2294 	case EVFILT_WRITE:
2295 		kn->kn_fop = &sowrite_filtops;
2296 		sb = &so->so_snd;
2297 		break;
2298 	default:
2299 		sounlock(so);
2300 		return (EINVAL);
2301 	}
2302 	SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
2303 	sb->sb_flags |= SB_KNOTE;
2304 	sounlock(so);
2305 	return (0);
2306 }
2307 
2308 static int
2309 sodopoll(struct socket *so, int events)
2310 {
2311 	int revents;
2312 
2313 	revents = 0;
2314 
2315 	if (events & (POLLIN | POLLRDNORM))
2316 		if (soreadable(so))
2317 			revents |= events & (POLLIN | POLLRDNORM);
2318 
2319 	if (events & (POLLOUT | POLLWRNORM))
2320 		if (sowritable(so))
2321 			revents |= events & (POLLOUT | POLLWRNORM);
2322 
2323 	if (events & (POLLPRI | POLLRDBAND))
2324 		if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
2325 			revents |= events & (POLLPRI | POLLRDBAND);
2326 
2327 	return revents;
2328 }
2329 
2330 int
2331 sopoll(struct socket *so, int events)
2332 {
2333 	int revents = 0;
2334 
2335 #ifndef DIAGNOSTIC
2336 	/*
2337 	 * Do a quick, unlocked check in expectation that the socket
2338 	 * will be ready for I/O.  Don't do this check if DIAGNOSTIC,
2339 	 * as the solocked() assertions will fail.
2340 	 */
2341 	if ((revents = sodopoll(so, events)) != 0)
2342 		return revents;
2343 #endif
2344 
2345 	solock(so);
2346 	if ((revents = sodopoll(so, events)) == 0) {
2347 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2348 			selrecord(curlwp, &so->so_rcv.sb_sel);
2349 			so->so_rcv.sb_flags |= SB_NOTIFY;
2350 		}
2351 
2352 		if (events & (POLLOUT | POLLWRNORM)) {
2353 			selrecord(curlwp, &so->so_snd.sb_sel);
2354 			so->so_snd.sb_flags |= SB_NOTIFY;
2355 		}
2356 	}
2357 	sounlock(so);
2358 
2359 	return revents;
2360 }
2361 
2362 
2363 #include <sys/sysctl.h>
2364 
2365 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO);
2366 static int sysctl_kern_sbmax(SYSCTLFN_PROTO);
2367 
2368 /*
2369  * sysctl helper routine for kern.somaxkva.  ensures that the given
2370  * value is not too small.
2371  * (XXX should we maybe make sure it's not too large as well?)
2372  */
2373 static int
2374 sysctl_kern_somaxkva(SYSCTLFN_ARGS)
2375 {
2376 	int error, new_somaxkva;
2377 	struct sysctlnode node;
2378 
2379 	new_somaxkva = somaxkva;
2380 	node = *rnode;
2381 	node.sysctl_data = &new_somaxkva;
2382 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2383 	if (error || newp == NULL)
2384 		return (error);
2385 
2386 	if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */
2387 		return (EINVAL);
2388 
2389 	mutex_enter(&so_pendfree_lock);
2390 	somaxkva = new_somaxkva;
2391 	cv_broadcast(&socurkva_cv);
2392 	mutex_exit(&so_pendfree_lock);
2393 
2394 	return (error);
2395 }
2396 
2397 /*
2398  * sysctl helper routine for kern.sbmax. Basically just ensures that
2399  * any new value is not too small.
2400  */
2401 static int
2402 sysctl_kern_sbmax(SYSCTLFN_ARGS)
2403 {
2404 	int error, new_sbmax;
2405 	struct sysctlnode node;
2406 
2407 	new_sbmax = sb_max;
2408 	node = *rnode;
2409 	node.sysctl_data = &new_sbmax;
2410 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2411 	if (error || newp == NULL)
2412 		return (error);
2413 
2414 	KERNEL_LOCK(1, NULL);
2415 	error = sb_max_set(new_sbmax);
2416 	KERNEL_UNLOCK_ONE(NULL);
2417 
2418 	return (error);
2419 }
2420 
2421 static void
2422 sysctl_kern_socket_setup(void)
2423 {
2424 
2425 	KASSERT(socket_sysctllog == NULL);
2426 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2427 		       CTLFLAG_PERMANENT,
2428 		       CTLTYPE_NODE, "kern", NULL,
2429 		       NULL, 0, NULL, 0,
2430 		       CTL_KERN, CTL_EOL);
2431 
2432 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2433 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2434 		       CTLTYPE_INT, "somaxkva",
2435 		       SYSCTL_DESCR("Maximum amount of kernel memory to be "
2436 				    "used for socket buffers"),
2437 		       sysctl_kern_somaxkva, 0, NULL, 0,
2438 		       CTL_KERN, KERN_SOMAXKVA, CTL_EOL);
2439 
2440 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2441 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2442 		       CTLTYPE_INT, "sbmax",
2443 		       SYSCTL_DESCR("Maximum socket buffer size"),
2444 		       sysctl_kern_sbmax, 0, NULL, 0,
2445 		       CTL_KERN, KERN_SBMAX, CTL_EOL);
2446 }
2447