xref: /netbsd-src/sys/kern/uipc_socket.c (revision f89f6560d453f5e37386cc7938c072d2f528b9fa)
1 /*	$NetBSD: uipc_socket.c,v 1.238 2015/04/05 23:19:56 rtr Exp $	*/
2 
3 /*-
4  * Copyright (c) 2002, 2007, 2008, 2009 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe of Wasabi Systems, Inc, and by Andrew Doran.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Copyright (c) 2004 The FreeBSD Foundation
34  * Copyright (c) 2004 Robert Watson
35  * Copyright (c) 1982, 1986, 1988, 1990, 1993
36  *	The Regents of the University of California.  All rights reserved.
37  *
38  * Redistribution and use in source and binary forms, with or without
39  * modification, are permitted provided that the following conditions
40  * are met:
41  * 1. Redistributions of source code must retain the above copyright
42  *    notice, this list of conditions and the following disclaimer.
43  * 2. Redistributions in binary form must reproduce the above copyright
44  *    notice, this list of conditions and the following disclaimer in the
45  *    documentation and/or other materials provided with the distribution.
46  * 3. Neither the name of the University nor the names of its contributors
47  *    may be used to endorse or promote products derived from this software
48  *    without specific prior written permission.
49  *
50  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
52  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
53  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
54  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
55  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
56  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
57  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
58  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
59  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
60  * SUCH DAMAGE.
61  *
62  *	@(#)uipc_socket.c	8.6 (Berkeley) 5/2/95
63  */
64 
65 /*
66  * Socket operation routines.
67  *
68  * These routines are called by the routines in sys_socket.c or from a
69  * system process, and implement the semantics of socket operations by
70  * switching out to the protocol specific routines.
71  */
72 
73 #include <sys/cdefs.h>
74 __KERNEL_RCSID(0, "$NetBSD: uipc_socket.c,v 1.238 2015/04/05 23:19:56 rtr Exp $");
75 
76 #include "opt_compat_netbsd.h"
77 #include "opt_sock_counters.h"
78 #include "opt_sosend_loan.h"
79 #include "opt_mbuftrace.h"
80 #include "opt_somaxkva.h"
81 #include "opt_multiprocessor.h"	/* XXX */
82 
83 #include <sys/param.h>
84 #include <sys/systm.h>
85 #include <sys/proc.h>
86 #include <sys/file.h>
87 #include <sys/filedesc.h>
88 #include <sys/kmem.h>
89 #include <sys/mbuf.h>
90 #include <sys/domain.h>
91 #include <sys/kernel.h>
92 #include <sys/protosw.h>
93 #include <sys/socket.h>
94 #include <sys/socketvar.h>
95 #include <sys/signalvar.h>
96 #include <sys/resourcevar.h>
97 #include <sys/uidinfo.h>
98 #include <sys/event.h>
99 #include <sys/poll.h>
100 #include <sys/kauth.h>
101 #include <sys/mutex.h>
102 #include <sys/condvar.h>
103 #include <sys/kthread.h>
104 
105 #ifdef COMPAT_50
106 #include <compat/sys/time.h>
107 #include <compat/sys/socket.h>
108 #endif
109 
110 #include <uvm/uvm_extern.h>
111 #include <uvm/uvm_loan.h>
112 #include <uvm/uvm_page.h>
113 
114 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
115 
116 extern const struct fileops socketops;
117 
118 extern int	somaxconn;			/* patchable (XXX sysctl) */
119 int		somaxconn = SOMAXCONN;
120 kmutex_t	*softnet_lock;
121 
122 #ifdef SOSEND_COUNTERS
123 #include <sys/device.h>
124 
125 static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
126     NULL, "sosend", "loan big");
127 static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
128     NULL, "sosend", "copy big");
129 static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
130     NULL, "sosend", "copy small");
131 static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
132     NULL, "sosend", "kva limit");
133 
134 #define	SOSEND_COUNTER_INCR(ev)		(ev)->ev_count++
135 
136 EVCNT_ATTACH_STATIC(sosend_loan_big);
137 EVCNT_ATTACH_STATIC(sosend_copy_big);
138 EVCNT_ATTACH_STATIC(sosend_copy_small);
139 EVCNT_ATTACH_STATIC(sosend_kvalimit);
140 #else
141 
142 #define	SOSEND_COUNTER_INCR(ev)		/* nothing */
143 
144 #endif /* SOSEND_COUNTERS */
145 
146 #if defined(SOSEND_NO_LOAN) || defined(MULTIPROCESSOR)
147 int sock_loan_thresh = -1;
148 #else
149 int sock_loan_thresh = 4096;
150 #endif
151 
152 static kmutex_t so_pendfree_lock;
153 static struct mbuf *so_pendfree = NULL;
154 
155 #ifndef SOMAXKVA
156 #define	SOMAXKVA (16 * 1024 * 1024)
157 #endif
158 int somaxkva = SOMAXKVA;
159 static int socurkva;
160 static kcondvar_t socurkva_cv;
161 
162 static kauth_listener_t socket_listener;
163 
164 #define	SOCK_LOAN_CHUNK		65536
165 
166 static void sopendfree_thread(void *);
167 static kcondvar_t pendfree_thread_cv;
168 static lwp_t *sopendfree_lwp;
169 
170 static void sysctl_kern_socket_setup(void);
171 static struct sysctllog *socket_sysctllog;
172 
173 static vsize_t
174 sokvareserve(struct socket *so, vsize_t len)
175 {
176 	int error;
177 
178 	mutex_enter(&so_pendfree_lock);
179 	while (socurkva + len > somaxkva) {
180 		SOSEND_COUNTER_INCR(&sosend_kvalimit);
181 		error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock);
182 		if (error) {
183 			len = 0;
184 			break;
185 		}
186 	}
187 	socurkva += len;
188 	mutex_exit(&so_pendfree_lock);
189 	return len;
190 }
191 
192 static void
193 sokvaunreserve(vsize_t len)
194 {
195 
196 	mutex_enter(&so_pendfree_lock);
197 	socurkva -= len;
198 	cv_broadcast(&socurkva_cv);
199 	mutex_exit(&so_pendfree_lock);
200 }
201 
202 /*
203  * sokvaalloc: allocate kva for loan.
204  */
205 
206 vaddr_t
207 sokvaalloc(vaddr_t sva, vsize_t len, struct socket *so)
208 {
209 	vaddr_t lva;
210 
211 	/*
212 	 * reserve kva.
213 	 */
214 
215 	if (sokvareserve(so, len) == 0)
216 		return 0;
217 
218 	/*
219 	 * allocate kva.
220 	 */
221 
222 	lva = uvm_km_alloc(kernel_map, len, atop(sva) & uvmexp.colormask,
223 	    UVM_KMF_COLORMATCH | UVM_KMF_VAONLY | UVM_KMF_WAITVA);
224 	if (lva == 0) {
225 		sokvaunreserve(len);
226 		return (0);
227 	}
228 
229 	return lva;
230 }
231 
232 /*
233  * sokvafree: free kva for loan.
234  */
235 
236 void
237 sokvafree(vaddr_t sva, vsize_t len)
238 {
239 
240 	/*
241 	 * free kva.
242 	 */
243 
244 	uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY);
245 
246 	/*
247 	 * unreserve kva.
248 	 */
249 
250 	sokvaunreserve(len);
251 }
252 
253 static void
254 sodoloanfree(struct vm_page **pgs, void *buf, size_t size)
255 {
256 	vaddr_t sva, eva;
257 	vsize_t len;
258 	int npgs;
259 
260 	KASSERT(pgs != NULL);
261 
262 	eva = round_page((vaddr_t) buf + size);
263 	sva = trunc_page((vaddr_t) buf);
264 	len = eva - sva;
265 	npgs = len >> PAGE_SHIFT;
266 
267 	pmap_kremove(sva, len);
268 	pmap_update(pmap_kernel());
269 	uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE);
270 	sokvafree(sva, len);
271 }
272 
273 /*
274  * sopendfree_thread: free mbufs on "pendfree" list.
275  * unlock and relock so_pendfree_lock when freeing mbufs.
276  */
277 
278 static void
279 sopendfree_thread(void *v)
280 {
281 	struct mbuf *m, *next;
282 	size_t rv;
283 
284 	mutex_enter(&so_pendfree_lock);
285 
286 	for (;;) {
287 		rv = 0;
288 		while (so_pendfree != NULL) {
289 			m = so_pendfree;
290 			so_pendfree = NULL;
291 			mutex_exit(&so_pendfree_lock);
292 
293 			for (; m != NULL; m = next) {
294 				next = m->m_next;
295 				KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0);
296 				KASSERT(m->m_ext.ext_refcnt == 0);
297 
298 				rv += m->m_ext.ext_size;
299 				sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf,
300 				    m->m_ext.ext_size);
301 				pool_cache_put(mb_cache, m);
302 			}
303 
304 			mutex_enter(&so_pendfree_lock);
305 		}
306 		if (rv)
307 			cv_broadcast(&socurkva_cv);
308 		cv_wait(&pendfree_thread_cv, &so_pendfree_lock);
309 	}
310 	panic("sopendfree_thread");
311 	/* NOTREACHED */
312 }
313 
314 void
315 soloanfree(struct mbuf *m, void *buf, size_t size, void *arg)
316 {
317 
318 	KASSERT(m != NULL);
319 
320 	/*
321 	 * postpone freeing mbuf.
322 	 *
323 	 * we can't do it in interrupt context
324 	 * because we need to put kva back to kernel_map.
325 	 */
326 
327 	mutex_enter(&so_pendfree_lock);
328 	m->m_next = so_pendfree;
329 	so_pendfree = m;
330 	cv_signal(&pendfree_thread_cv);
331 	mutex_exit(&so_pendfree_lock);
332 }
333 
334 static long
335 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space)
336 {
337 	struct iovec *iov = uio->uio_iov;
338 	vaddr_t sva, eva;
339 	vsize_t len;
340 	vaddr_t lva;
341 	int npgs, error;
342 	vaddr_t va;
343 	int i;
344 
345 	if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace))
346 		return (0);
347 
348 	if (iov->iov_len < (size_t) space)
349 		space = iov->iov_len;
350 	if (space > SOCK_LOAN_CHUNK)
351 		space = SOCK_LOAN_CHUNK;
352 
353 	eva = round_page((vaddr_t) iov->iov_base + space);
354 	sva = trunc_page((vaddr_t) iov->iov_base);
355 	len = eva - sva;
356 	npgs = len >> PAGE_SHIFT;
357 
358 	KASSERT(npgs <= M_EXT_MAXPAGES);
359 
360 	lva = sokvaalloc(sva, len, so);
361 	if (lva == 0)
362 		return 0;
363 
364 	error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len,
365 	    m->m_ext.ext_pgs, UVM_LOAN_TOPAGE);
366 	if (error) {
367 		sokvafree(lva, len);
368 		return (0);
369 	}
370 
371 	for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE)
372 		pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
373 		    VM_PROT_READ, 0);
374 	pmap_update(pmap_kernel());
375 
376 	lva += (vaddr_t) iov->iov_base & PAGE_MASK;
377 
378 	MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so);
379 	m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
380 
381 	uio->uio_resid -= space;
382 	/* uio_offset not updated, not set/used for write(2) */
383 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space;
384 	uio->uio_iov->iov_len -= space;
385 	if (uio->uio_iov->iov_len == 0) {
386 		uio->uio_iov++;
387 		uio->uio_iovcnt--;
388 	}
389 
390 	return (space);
391 }
392 
393 struct mbuf *
394 getsombuf(struct socket *so, int type)
395 {
396 	struct mbuf *m;
397 
398 	m = m_get(M_WAIT, type);
399 	MCLAIM(m, so->so_mowner);
400 	return m;
401 }
402 
403 static int
404 socket_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
405     void *arg0, void *arg1, void *arg2, void *arg3)
406 {
407 	int result;
408 	enum kauth_network_req req;
409 
410 	result = KAUTH_RESULT_DEFER;
411 	req = (enum kauth_network_req)arg0;
412 
413 	if ((action != KAUTH_NETWORK_SOCKET) &&
414 	    (action != KAUTH_NETWORK_BIND))
415 		return result;
416 
417 	switch (req) {
418 	case KAUTH_REQ_NETWORK_BIND_PORT:
419 		result = KAUTH_RESULT_ALLOW;
420 		break;
421 
422 	case KAUTH_REQ_NETWORK_SOCKET_DROP: {
423 		/* Normal users can only drop their own connections. */
424 		struct socket *so = (struct socket *)arg1;
425 
426 		if (so->so_cred && proc_uidmatch(cred, so->so_cred) == 0)
427 			result = KAUTH_RESULT_ALLOW;
428 
429 		break;
430 		}
431 
432 	case KAUTH_REQ_NETWORK_SOCKET_OPEN:
433 		/* We allow "raw" routing/bluetooth sockets to anyone. */
434 		if ((u_long)arg1 == PF_ROUTE || (u_long)arg1 == PF_OROUTE
435 		    || (u_long)arg1 == PF_BLUETOOTH) {
436 			result = KAUTH_RESULT_ALLOW;
437 		} else {
438 			/* Privileged, let secmodel handle this. */
439 			if ((u_long)arg2 == SOCK_RAW)
440 				break;
441 		}
442 
443 		result = KAUTH_RESULT_ALLOW;
444 
445 		break;
446 
447 	case KAUTH_REQ_NETWORK_SOCKET_CANSEE:
448 		result = KAUTH_RESULT_ALLOW;
449 
450 		break;
451 
452 	default:
453 		break;
454 	}
455 
456 	return result;
457 }
458 
459 void
460 soinit(void)
461 {
462 
463 	sysctl_kern_socket_setup();
464 
465 	mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM);
466 	softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
467 	cv_init(&socurkva_cv, "sokva");
468 	cv_init(&pendfree_thread_cv, "sopendfr");
469 	soinit2();
470 
471 	/* Set the initial adjusted socket buffer size. */
472 	if (sb_max_set(sb_max))
473 		panic("bad initial sb_max value: %lu", sb_max);
474 
475 	socket_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK,
476 	    socket_listener_cb, NULL);
477 }
478 
479 void
480 soinit1(void)
481 {
482 	int error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL,
483 	    sopendfree_thread, NULL, &sopendfree_lwp, "sopendfree");
484 	if (error)
485 		panic("soinit1 %d", error);
486 }
487 
488 /*
489  * socreate: create a new socket of the specified type and the protocol.
490  *
491  * => Caller may specify another socket for lock sharing (must not be held).
492  * => Returns the new socket without lock held.
493  */
494 int
495 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l,
496 	 struct socket *lockso)
497 {
498 	const struct protosw	*prp;
499 	struct socket	*so;
500 	uid_t		uid;
501 	int		error;
502 	kmutex_t	*lock;
503 
504 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET,
505 	    KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type),
506 	    KAUTH_ARG(proto));
507 	if (error != 0)
508 		return error;
509 
510 	if (proto)
511 		prp = pffindproto(dom, proto, type);
512 	else
513 		prp = pffindtype(dom, type);
514 	if (prp == NULL) {
515 		/* no support for domain */
516 		if (pffinddomain(dom) == 0)
517 			return EAFNOSUPPORT;
518 		/* no support for socket type */
519 		if (proto == 0 && type != 0)
520 			return EPROTOTYPE;
521 		return EPROTONOSUPPORT;
522 	}
523 	if (prp->pr_usrreqs == NULL)
524 		return EPROTONOSUPPORT;
525 	if (prp->pr_type != type)
526 		return EPROTOTYPE;
527 
528 	so = soget(true);
529 	so->so_type = type;
530 	so->so_proto = prp;
531 	so->so_send = sosend;
532 	so->so_receive = soreceive;
533 #ifdef MBUFTRACE
534 	so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner;
535 	so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner;
536 	so->so_mowner = &prp->pr_domain->dom_mowner;
537 #endif
538 	uid = kauth_cred_geteuid(l->l_cred);
539 	so->so_uidinfo = uid_find(uid);
540 	so->so_cpid = l->l_proc->p_pid;
541 
542 	/*
543 	 * Lock assigned and taken during PCB attach, unless we share
544 	 * the lock with another socket, e.g. socketpair(2) case.
545 	 */
546 	if (lockso) {
547 		lock = lockso->so_lock;
548 		so->so_lock = lock;
549 		mutex_obj_hold(lock);
550 		mutex_enter(lock);
551 	}
552 
553 	/* Attach the PCB (returns with the socket lock held). */
554 	error = (*prp->pr_usrreqs->pr_attach)(so, proto);
555 	KASSERT(solocked(so));
556 
557 	if (error) {
558 		KASSERT(so->so_pcb == NULL);
559 		so->so_state |= SS_NOFDREF;
560 		sofree(so);
561 		return error;
562 	}
563 	so->so_cred = kauth_cred_dup(l->l_cred);
564 	sounlock(so);
565 
566 	*aso = so;
567 	return 0;
568 }
569 
570 /*
571  * fsocreate: create a socket and a file descriptor associated with it.
572  *
573  * => On success, write file descriptor to fdout and return zero.
574  * => On failure, return non-zero; *fdout will be undefined.
575  */
576 int
577 fsocreate(int domain, struct socket **sop, int type, int proto, int *fdout)
578 {
579 	lwp_t *l = curlwp;
580 	int error, fd, flags;
581 	struct socket *so;
582 	struct file *fp;
583 
584 	if ((error = fd_allocfile(&fp, &fd)) != 0) {
585 		return error;
586 	}
587 	flags = type & SOCK_FLAGS_MASK;
588 	fd_set_exclose(l, fd, (flags & SOCK_CLOEXEC) != 0);
589 	fp->f_flag = FREAD|FWRITE|((flags & SOCK_NONBLOCK) ? FNONBLOCK : 0)|
590 	    ((flags & SOCK_NOSIGPIPE) ? FNOSIGPIPE : 0);
591 	fp->f_type = DTYPE_SOCKET;
592 	fp->f_ops = &socketops;
593 
594 	type &= ~SOCK_FLAGS_MASK;
595 	error = socreate(domain, &so, type, proto, l, NULL);
596 	if (error) {
597 		fd_abort(curproc, fp, fd);
598 		return error;
599 	}
600 	if (flags & SOCK_NONBLOCK) {
601 		so->so_state |= SS_NBIO;
602 	}
603 	fp->f_socket = so;
604 	fd_affix(curproc, fp, fd);
605 
606 	if (sop != NULL) {
607 		*sop = so;
608 	}
609 	*fdout = fd;
610 	return error;
611 }
612 
613 int
614 sofamily(const struct socket *so)
615 {
616 	const struct protosw *pr;
617 	const struct domain *dom;
618 
619 	if ((pr = so->so_proto) == NULL)
620 		return AF_UNSPEC;
621 	if ((dom = pr->pr_domain) == NULL)
622 		return AF_UNSPEC;
623 	return dom->dom_family;
624 }
625 
626 int
627 sobind(struct socket *so, struct sockaddr *nam, struct lwp *l)
628 {
629 	int	error;
630 
631 	solock(so);
632 	if (nam->sa_family != so->so_proto->pr_domain->dom_family) {
633 		sounlock(so);
634 		return EAFNOSUPPORT;
635 	}
636 	error = (*so->so_proto->pr_usrreqs->pr_bind)(so, nam, l);
637 	sounlock(so);
638 	return error;
639 }
640 
641 int
642 solisten(struct socket *so, int backlog, struct lwp *l)
643 {
644 	int	error;
645 
646 	solock(so);
647 	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
648 	    SS_ISDISCONNECTING)) != 0) {
649 		sounlock(so);
650 		return EINVAL;
651 	}
652 	error = (*so->so_proto->pr_usrreqs->pr_listen)(so, l);
653 	if (error != 0) {
654 		sounlock(so);
655 		return error;
656 	}
657 	if (TAILQ_EMPTY(&so->so_q))
658 		so->so_options |= SO_ACCEPTCONN;
659 	if (backlog < 0)
660 		backlog = 0;
661 	so->so_qlimit = min(backlog, somaxconn);
662 	sounlock(so);
663 	return 0;
664 }
665 
666 void
667 sofree(struct socket *so)
668 {
669 	u_int refs;
670 
671 	KASSERT(solocked(so));
672 
673 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
674 		sounlock(so);
675 		return;
676 	}
677 	if (so->so_head) {
678 		/*
679 		 * We must not decommission a socket that's on the accept(2)
680 		 * queue.  If we do, then accept(2) may hang after select(2)
681 		 * indicated that the listening socket was ready.
682 		 */
683 		if (!soqremque(so, 0)) {
684 			sounlock(so);
685 			return;
686 		}
687 	}
688 	if (so->so_rcv.sb_hiwat)
689 		(void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0,
690 		    RLIM_INFINITY);
691 	if (so->so_snd.sb_hiwat)
692 		(void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0,
693 		    RLIM_INFINITY);
694 	sbrelease(&so->so_snd, so);
695 	KASSERT(!cv_has_waiters(&so->so_cv));
696 	KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv));
697 	KASSERT(!cv_has_waiters(&so->so_snd.sb_cv));
698 	sorflush(so);
699 	refs = so->so_aborting;	/* XXX */
700 	/* Remove acccept filter if one is present. */
701 	if (so->so_accf != NULL)
702 		(void)accept_filt_clear(so);
703 	sounlock(so);
704 	if (refs == 0)		/* XXX */
705 		soput(so);
706 }
707 
708 /*
709  * soclose: close a socket on last file table reference removal.
710  * Initiate disconnect if connected.  Free socket when disconnect complete.
711  */
712 int
713 soclose(struct socket *so)
714 {
715 	struct socket *so2;
716 	int error = 0;
717 
718 	solock(so);
719 	if (so->so_options & SO_ACCEPTCONN) {
720 		for (;;) {
721 			if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) {
722 				KASSERT(solocked2(so, so2));
723 				(void) soqremque(so2, 0);
724 				/* soabort drops the lock. */
725 				(void) soabort(so2);
726 				solock(so);
727 				continue;
728 			}
729 			if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) {
730 				KASSERT(solocked2(so, so2));
731 				(void) soqremque(so2, 1);
732 				/* soabort drops the lock. */
733 				(void) soabort(so2);
734 				solock(so);
735 				continue;
736 			}
737 			break;
738 		}
739 	}
740 	if (so->so_pcb == NULL)
741 		goto discard;
742 	if (so->so_state & SS_ISCONNECTED) {
743 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
744 			error = sodisconnect(so);
745 			if (error)
746 				goto drop;
747 		}
748 		if (so->so_options & SO_LINGER) {
749 			if ((so->so_state & (SS_ISDISCONNECTING|SS_NBIO)) ==
750 			    (SS_ISDISCONNECTING|SS_NBIO))
751 				goto drop;
752 			while (so->so_state & SS_ISCONNECTED) {
753 				error = sowait(so, true, so->so_linger * hz);
754 				if (error)
755 					break;
756 			}
757 		}
758 	}
759  drop:
760 	if (so->so_pcb) {
761 		KASSERT(solocked(so));
762 		(*so->so_proto->pr_usrreqs->pr_detach)(so);
763 	}
764  discard:
765 	KASSERT((so->so_state & SS_NOFDREF) == 0);
766 	kauth_cred_free(so->so_cred);
767 	so->so_state |= SS_NOFDREF;
768 	sofree(so);
769 	return error;
770 }
771 
772 /*
773  * Must be called with the socket locked..  Will return with it unlocked.
774  */
775 int
776 soabort(struct socket *so)
777 {
778 	u_int refs;
779 	int error;
780 
781 	KASSERT(solocked(so));
782 	KASSERT(so->so_head == NULL);
783 
784 	so->so_aborting++;		/* XXX */
785 	error = (*so->so_proto->pr_usrreqs->pr_abort)(so);
786 	refs = --so->so_aborting;	/* XXX */
787 	if (error || (refs == 0)) {
788 		sofree(so);
789 	} else {
790 		sounlock(so);
791 	}
792 	return error;
793 }
794 
795 int
796 soaccept(struct socket *so, struct mbuf *nam)
797 {
798 	int error;
799 
800 	KASSERT(solocked(so));
801 	KASSERT((so->so_state & SS_NOFDREF) != 0);
802 
803 	so->so_state &= ~SS_NOFDREF;
804 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
805 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
806 		error = (*so->so_proto->pr_usrreqs->pr_accept)(so, nam);
807 	else
808 		error = ECONNABORTED;
809 
810 	return error;
811 }
812 
813 int
814 soconnect(struct socket *so, struct mbuf *nam, struct lwp *l)
815 {
816 	int error;
817 
818 	KASSERT(solocked(so));
819 
820 	if (so->so_options & SO_ACCEPTCONN)
821 		return EOPNOTSUPP;
822 	/*
823 	 * If protocol is connection-based, can only connect once.
824 	 * Otherwise, if connected, try to disconnect first.
825 	 * This allows user to disconnect by connecting to, e.g.,
826 	 * a null address.
827 	 */
828 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
829 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
830 	    (error = sodisconnect(so))))
831 		error = EISCONN;
832 	else
833 		error = (*so->so_proto->pr_usrreqs->pr_connect)(so, nam, l);
834 
835 	return error;
836 }
837 
838 int
839 soconnect2(struct socket *so1, struct socket *so2)
840 {
841 	KASSERT(solocked2(so1, so2));
842 
843 	return (*so1->so_proto->pr_usrreqs->pr_connect2)(so1, so2);
844 }
845 
846 int
847 sodisconnect(struct socket *so)
848 {
849 	int	error;
850 
851 	KASSERT(solocked(so));
852 
853 	if ((so->so_state & SS_ISCONNECTED) == 0) {
854 		error = ENOTCONN;
855 	} else if (so->so_state & SS_ISDISCONNECTING) {
856 		error = EALREADY;
857 	} else {
858 		error = (*so->so_proto->pr_usrreqs->pr_disconnect)(so);
859 	}
860 	return (error);
861 }
862 
863 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
864 /*
865  * Send on a socket.
866  * If send must go all at once and message is larger than
867  * send buffering, then hard error.
868  * Lock against other senders.
869  * If must go all at once and not enough room now, then
870  * inform user that this would block and do nothing.
871  * Otherwise, if nonblocking, send as much as possible.
872  * The data to be sent is described by "uio" if nonzero,
873  * otherwise by the mbuf chain "top" (which must be null
874  * if uio is not).  Data provided in mbuf chain must be small
875  * enough to send all at once.
876  *
877  * Returns nonzero on error, timeout or signal; callers
878  * must check for short counts if EINTR/ERESTART are returned.
879  * Data and control buffers are freed on return.
880  */
881 int
882 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
883 	struct mbuf *control, int flags, struct lwp *l)
884 {
885 	struct mbuf	**mp, *m;
886 	long		space, len, resid, clen, mlen;
887 	int		error, s, dontroute, atomic;
888 	short		wakeup_state = 0;
889 
890 	clen = 0;
891 
892 	/*
893 	 * solock() provides atomicity of access.  splsoftnet() prevents
894 	 * protocol processing soft interrupts from interrupting us and
895 	 * blocking (expensive).
896 	 */
897 	s = splsoftnet();
898 	solock(so);
899 	atomic = sosendallatonce(so) || top;
900 	if (uio)
901 		resid = uio->uio_resid;
902 	else
903 		resid = top->m_pkthdr.len;
904 	/*
905 	 * In theory resid should be unsigned.
906 	 * However, space must be signed, as it might be less than 0
907 	 * if we over-committed, and we must use a signed comparison
908 	 * of space and resid.  On the other hand, a negative resid
909 	 * causes us to loop sending 0-length segments to the protocol.
910 	 */
911 	if (resid < 0) {
912 		error = EINVAL;
913 		goto out;
914 	}
915 	dontroute =
916 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
917 	    (so->so_proto->pr_flags & PR_ATOMIC);
918 	l->l_ru.ru_msgsnd++;
919 	if (control)
920 		clen = control->m_len;
921  restart:
922 	if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
923 		goto out;
924 	do {
925 		if (so->so_state & SS_CANTSENDMORE) {
926 			error = EPIPE;
927 			goto release;
928 		}
929 		if (so->so_error) {
930 			error = so->so_error;
931 			so->so_error = 0;
932 			goto release;
933 		}
934 		if ((so->so_state & SS_ISCONNECTED) == 0) {
935 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
936 				if (resid || clen == 0) {
937 					error = ENOTCONN;
938 					goto release;
939 				}
940 			} else if (addr == 0) {
941 				error = EDESTADDRREQ;
942 				goto release;
943 			}
944 		}
945 		space = sbspace(&so->so_snd);
946 		if (flags & MSG_OOB)
947 			space += 1024;
948 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
949 		    clen > so->so_snd.sb_hiwat) {
950 			error = EMSGSIZE;
951 			goto release;
952 		}
953 		if (space < resid + clen &&
954 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
955 			if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
956 				error = EWOULDBLOCK;
957 				goto release;
958 			}
959 			sbunlock(&so->so_snd);
960 			if (wakeup_state & SS_RESTARTSYS) {
961 				error = ERESTART;
962 				goto out;
963 			}
964 			error = sbwait(&so->so_snd);
965 			if (error)
966 				goto out;
967 			wakeup_state = so->so_state;
968 			goto restart;
969 		}
970 		wakeup_state = 0;
971 		mp = &top;
972 		space -= clen;
973 		do {
974 			if (uio == NULL) {
975 				/*
976 				 * Data is prepackaged in "top".
977 				 */
978 				resid = 0;
979 				if (flags & MSG_EOR)
980 					top->m_flags |= M_EOR;
981 			} else do {
982 				sounlock(so);
983 				splx(s);
984 				if (top == NULL) {
985 					m = m_gethdr(M_WAIT, MT_DATA);
986 					mlen = MHLEN;
987 					m->m_pkthdr.len = 0;
988 					m->m_pkthdr.rcvif = NULL;
989 				} else {
990 					m = m_get(M_WAIT, MT_DATA);
991 					mlen = MLEN;
992 				}
993 				MCLAIM(m, so->so_snd.sb_mowner);
994 				if (sock_loan_thresh >= 0 &&
995 				    uio->uio_iov->iov_len >= sock_loan_thresh &&
996 				    space >= sock_loan_thresh &&
997 				    (len = sosend_loan(so, uio, m,
998 						       space)) != 0) {
999 					SOSEND_COUNTER_INCR(&sosend_loan_big);
1000 					space -= len;
1001 					goto have_data;
1002 				}
1003 				if (resid >= MINCLSIZE && space >= MCLBYTES) {
1004 					SOSEND_COUNTER_INCR(&sosend_copy_big);
1005 					m_clget(m, M_DONTWAIT);
1006 					if ((m->m_flags & M_EXT) == 0)
1007 						goto nopages;
1008 					mlen = MCLBYTES;
1009 					if (atomic && top == 0) {
1010 						len = lmin(MCLBYTES - max_hdr,
1011 						    resid);
1012 						m->m_data += max_hdr;
1013 					} else
1014 						len = lmin(MCLBYTES, resid);
1015 					space -= len;
1016 				} else {
1017  nopages:
1018 					SOSEND_COUNTER_INCR(&sosend_copy_small);
1019 					len = lmin(lmin(mlen, resid), space);
1020 					space -= len;
1021 					/*
1022 					 * For datagram protocols, leave room
1023 					 * for protocol headers in first mbuf.
1024 					 */
1025 					if (atomic && top == 0 && len < mlen)
1026 						MH_ALIGN(m, len);
1027 				}
1028 				error = uiomove(mtod(m, void *), (int)len, uio);
1029  have_data:
1030 				resid = uio->uio_resid;
1031 				m->m_len = len;
1032 				*mp = m;
1033 				top->m_pkthdr.len += len;
1034 				s = splsoftnet();
1035 				solock(so);
1036 				if (error != 0)
1037 					goto release;
1038 				mp = &m->m_next;
1039 				if (resid <= 0) {
1040 					if (flags & MSG_EOR)
1041 						top->m_flags |= M_EOR;
1042 					break;
1043 				}
1044 			} while (space > 0 && atomic);
1045 
1046 			if (so->so_state & SS_CANTSENDMORE) {
1047 				error = EPIPE;
1048 				goto release;
1049 			}
1050 			if (dontroute)
1051 				so->so_options |= SO_DONTROUTE;
1052 			if (resid > 0)
1053 				so->so_state |= SS_MORETOCOME;
1054 			if (flags & MSG_OOB)
1055 				error = (*so->so_proto->pr_usrreqs->pr_sendoob)(so,
1056 				    top, control);
1057 			else
1058 				error = (*so->so_proto->pr_usrreqs->pr_send)(so,
1059 				    top, addr, control, l);
1060 			if (dontroute)
1061 				so->so_options &= ~SO_DONTROUTE;
1062 			if (resid > 0)
1063 				so->so_state &= ~SS_MORETOCOME;
1064 			clen = 0;
1065 			control = NULL;
1066 			top = NULL;
1067 			mp = &top;
1068 			if (error != 0)
1069 				goto release;
1070 		} while (resid && space > 0);
1071 	} while (resid);
1072 
1073  release:
1074 	sbunlock(&so->so_snd);
1075  out:
1076 	sounlock(so);
1077 	splx(s);
1078 	if (top)
1079 		m_freem(top);
1080 	if (control)
1081 		m_freem(control);
1082 	return (error);
1083 }
1084 
1085 /*
1086  * Following replacement or removal of the first mbuf on the first
1087  * mbuf chain of a socket buffer, push necessary state changes back
1088  * into the socket buffer so that other consumers see the values
1089  * consistently.  'nextrecord' is the callers locally stored value of
1090  * the original value of sb->sb_mb->m_nextpkt which must be restored
1091  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
1092  */
1093 static void
1094 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
1095 {
1096 
1097 	KASSERT(solocked(sb->sb_so));
1098 
1099 	/*
1100 	 * First, update for the new value of nextrecord.  If necessary,
1101 	 * make it the first record.
1102 	 */
1103 	if (sb->sb_mb != NULL)
1104 		sb->sb_mb->m_nextpkt = nextrecord;
1105 	else
1106 		sb->sb_mb = nextrecord;
1107 
1108         /*
1109          * Now update any dependent socket buffer fields to reflect
1110          * the new state.  This is an inline of SB_EMPTY_FIXUP, with
1111          * the addition of a second clause that takes care of the
1112          * case where sb_mb has been updated, but remains the last
1113          * record.
1114          */
1115         if (sb->sb_mb == NULL) {
1116                 sb->sb_mbtail = NULL;
1117                 sb->sb_lastrecord = NULL;
1118         } else if (sb->sb_mb->m_nextpkt == NULL)
1119                 sb->sb_lastrecord = sb->sb_mb;
1120 }
1121 
1122 /*
1123  * Implement receive operations on a socket.
1124  * We depend on the way that records are added to the sockbuf
1125  * by sbappend*.  In particular, each record (mbufs linked through m_next)
1126  * must begin with an address if the protocol so specifies,
1127  * followed by an optional mbuf or mbufs containing ancillary data,
1128  * and then zero or more mbufs of data.
1129  * In order to avoid blocking network interrupts for the entire time here,
1130  * we splx() while doing the actual copy to user space.
1131  * Although the sockbuf is locked, new data may still be appended,
1132  * and thus we must maintain consistency of the sockbuf during that time.
1133  *
1134  * The caller may receive the data as a single mbuf chain by supplying
1135  * an mbuf **mp0 for use in returning the chain.  The uio is then used
1136  * only for the count in uio_resid.
1137  */
1138 int
1139 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
1140 	struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1141 {
1142 	struct lwp *l = curlwp;
1143 	struct mbuf	*m, **mp, *mt;
1144 	size_t len, offset, moff, orig_resid;
1145 	int atomic, flags, error, s, type;
1146 	const struct protosw	*pr;
1147 	struct mbuf	*nextrecord;
1148 	int		mbuf_removed = 0;
1149 	const struct domain *dom;
1150 	short		wakeup_state = 0;
1151 
1152 	pr = so->so_proto;
1153 	atomic = pr->pr_flags & PR_ATOMIC;
1154 	dom = pr->pr_domain;
1155 	mp = mp0;
1156 	type = 0;
1157 	orig_resid = uio->uio_resid;
1158 
1159 	if (paddr != NULL)
1160 		*paddr = NULL;
1161 	if (controlp != NULL)
1162 		*controlp = NULL;
1163 	if (flagsp != NULL)
1164 		flags = *flagsp &~ MSG_EOR;
1165 	else
1166 		flags = 0;
1167 
1168 	if (flags & MSG_OOB) {
1169 		m = m_get(M_WAIT, MT_DATA);
1170 		solock(so);
1171 		error = (*pr->pr_usrreqs->pr_recvoob)(so, m, flags & MSG_PEEK);
1172 		sounlock(so);
1173 		if (error)
1174 			goto bad;
1175 		do {
1176 			error = uiomove(mtod(m, void *),
1177 			    MIN(uio->uio_resid, m->m_len), uio);
1178 			m = m_free(m);
1179 		} while (uio->uio_resid > 0 && error == 0 && m);
1180  bad:
1181 		if (m != NULL)
1182 			m_freem(m);
1183 		return error;
1184 	}
1185 	if (mp != NULL)
1186 		*mp = NULL;
1187 
1188 	/*
1189 	 * solock() provides atomicity of access.  splsoftnet() prevents
1190 	 * protocol processing soft interrupts from interrupting us and
1191 	 * blocking (expensive).
1192 	 */
1193 	s = splsoftnet();
1194 	solock(so);
1195  restart:
1196 	if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) {
1197 		sounlock(so);
1198 		splx(s);
1199 		return error;
1200 	}
1201 
1202 	m = so->so_rcv.sb_mb;
1203 	/*
1204 	 * If we have less data than requested, block awaiting more
1205 	 * (subject to any timeout) if:
1206 	 *   1. the current count is less than the low water mark,
1207 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1208 	 *	receive operation at once if we block (resid <= hiwat), or
1209 	 *   3. MSG_DONTWAIT is not set.
1210 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1211 	 * we have to do the receive in sections, and thus risk returning
1212 	 * a short count if a timeout or signal occurs after we start.
1213 	 */
1214 	if (m == NULL ||
1215 	    ((flags & MSG_DONTWAIT) == 0 &&
1216 	     so->so_rcv.sb_cc < uio->uio_resid &&
1217 	     (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1218 	      ((flags & MSG_WAITALL) &&
1219 	       uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1220 	     m->m_nextpkt == NULL && !atomic)) {
1221 #ifdef DIAGNOSTIC
1222 		if (m == NULL && so->so_rcv.sb_cc)
1223 			panic("receive 1");
1224 #endif
1225 		if (so->so_error) {
1226 			if (m != NULL)
1227 				goto dontblock;
1228 			error = so->so_error;
1229 			if ((flags & MSG_PEEK) == 0)
1230 				so->so_error = 0;
1231 			goto release;
1232 		}
1233 		if (so->so_state & SS_CANTRCVMORE) {
1234 			if (m != NULL)
1235 				goto dontblock;
1236 			else
1237 				goto release;
1238 		}
1239 		for (; m != NULL; m = m->m_next)
1240 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1241 				m = so->so_rcv.sb_mb;
1242 				goto dontblock;
1243 			}
1244 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1245 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1246 			error = ENOTCONN;
1247 			goto release;
1248 		}
1249 		if (uio->uio_resid == 0)
1250 			goto release;
1251 		if ((so->so_state & SS_NBIO) ||
1252 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1253 			error = EWOULDBLOCK;
1254 			goto release;
1255 		}
1256 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
1257 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
1258 		sbunlock(&so->so_rcv);
1259 		if (wakeup_state & SS_RESTARTSYS)
1260 			error = ERESTART;
1261 		else
1262 			error = sbwait(&so->so_rcv);
1263 		if (error != 0) {
1264 			sounlock(so);
1265 			splx(s);
1266 			return error;
1267 		}
1268 		wakeup_state = so->so_state;
1269 		goto restart;
1270 	}
1271  dontblock:
1272 	/*
1273 	 * On entry here, m points to the first record of the socket buffer.
1274 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1275 	 * pointer to the next record in the socket buffer.  We must keep the
1276 	 * various socket buffer pointers and local stack versions of the
1277 	 * pointers in sync, pushing out modifications before dropping the
1278 	 * socket lock, and re-reading them when picking it up.
1279 	 *
1280 	 * Otherwise, we will race with the network stack appending new data
1281 	 * or records onto the socket buffer by using inconsistent/stale
1282 	 * versions of the field, possibly resulting in socket buffer
1283 	 * corruption.
1284 	 *
1285 	 * By holding the high-level sblock(), we prevent simultaneous
1286 	 * readers from pulling off the front of the socket buffer.
1287 	 */
1288 	if (l != NULL)
1289 		l->l_ru.ru_msgrcv++;
1290 	KASSERT(m == so->so_rcv.sb_mb);
1291 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
1292 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
1293 	nextrecord = m->m_nextpkt;
1294 	if (pr->pr_flags & PR_ADDR) {
1295 #ifdef DIAGNOSTIC
1296 		if (m->m_type != MT_SONAME)
1297 			panic("receive 1a");
1298 #endif
1299 		orig_resid = 0;
1300 		if (flags & MSG_PEEK) {
1301 			if (paddr)
1302 				*paddr = m_copy(m, 0, m->m_len);
1303 			m = m->m_next;
1304 		} else {
1305 			sbfree(&so->so_rcv, m);
1306 			mbuf_removed = 1;
1307 			if (paddr != NULL) {
1308 				*paddr = m;
1309 				so->so_rcv.sb_mb = m->m_next;
1310 				m->m_next = NULL;
1311 				m = so->so_rcv.sb_mb;
1312 			} else {
1313 				MFREE(m, so->so_rcv.sb_mb);
1314 				m = so->so_rcv.sb_mb;
1315 			}
1316 			sbsync(&so->so_rcv, nextrecord);
1317 		}
1318 	}
1319 
1320 	/*
1321 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1322 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1323 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1324 	 * perform externalization (or freeing if controlp == NULL).
1325 	 */
1326 	if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) {
1327 		struct mbuf *cm = NULL, *cmn;
1328 		struct mbuf **cme = &cm;
1329 
1330 		do {
1331 			if (flags & MSG_PEEK) {
1332 				if (controlp != NULL) {
1333 					*controlp = m_copy(m, 0, m->m_len);
1334 					controlp = &(*controlp)->m_next;
1335 				}
1336 				m = m->m_next;
1337 			} else {
1338 				sbfree(&so->so_rcv, m);
1339 				so->so_rcv.sb_mb = m->m_next;
1340 				m->m_next = NULL;
1341 				*cme = m;
1342 				cme = &(*cme)->m_next;
1343 				m = so->so_rcv.sb_mb;
1344 			}
1345 		} while (m != NULL && m->m_type == MT_CONTROL);
1346 		if ((flags & MSG_PEEK) == 0)
1347 			sbsync(&so->so_rcv, nextrecord);
1348 		for (; cm != NULL; cm = cmn) {
1349 			cmn = cm->m_next;
1350 			cm->m_next = NULL;
1351 			type = mtod(cm, struct cmsghdr *)->cmsg_type;
1352 			if (controlp != NULL) {
1353 				if (dom->dom_externalize != NULL &&
1354 				    type == SCM_RIGHTS) {
1355 					sounlock(so);
1356 					splx(s);
1357 					error = (*dom->dom_externalize)(cm, l,
1358 					    (flags & MSG_CMSG_CLOEXEC) ?
1359 					    O_CLOEXEC : 0);
1360 					s = splsoftnet();
1361 					solock(so);
1362 				}
1363 				*controlp = cm;
1364 				while (*controlp != NULL)
1365 					controlp = &(*controlp)->m_next;
1366 			} else {
1367 				/*
1368 				 * Dispose of any SCM_RIGHTS message that went
1369 				 * through the read path rather than recv.
1370 				 */
1371 				if (dom->dom_dispose != NULL &&
1372 				    type == SCM_RIGHTS) {
1373 				    	sounlock(so);
1374 					(*dom->dom_dispose)(cm);
1375 					solock(so);
1376 				}
1377 				m_freem(cm);
1378 			}
1379 		}
1380 		if (m != NULL)
1381 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1382 		else
1383 			nextrecord = so->so_rcv.sb_mb;
1384 		orig_resid = 0;
1385 	}
1386 
1387 	/* If m is non-NULL, we have some data to read. */
1388 	if (__predict_true(m != NULL)) {
1389 		type = m->m_type;
1390 		if (type == MT_OOBDATA)
1391 			flags |= MSG_OOB;
1392 	}
1393 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1394 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1395 
1396 	moff = 0;
1397 	offset = 0;
1398 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1399 		if (m->m_type == MT_OOBDATA) {
1400 			if (type != MT_OOBDATA)
1401 				break;
1402 		} else if (type == MT_OOBDATA)
1403 			break;
1404 #ifdef DIAGNOSTIC
1405 		else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
1406 			panic("receive 3");
1407 #endif
1408 		so->so_state &= ~SS_RCVATMARK;
1409 		wakeup_state = 0;
1410 		len = uio->uio_resid;
1411 		if (so->so_oobmark && len > so->so_oobmark - offset)
1412 			len = so->so_oobmark - offset;
1413 		if (len > m->m_len - moff)
1414 			len = m->m_len - moff;
1415 		/*
1416 		 * If mp is set, just pass back the mbufs.
1417 		 * Otherwise copy them out via the uio, then free.
1418 		 * Sockbuf must be consistent here (points to current mbuf,
1419 		 * it points to next record) when we drop priority;
1420 		 * we must note any additions to the sockbuf when we
1421 		 * block interrupts again.
1422 		 */
1423 		if (mp == NULL) {
1424 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1425 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1426 			sounlock(so);
1427 			splx(s);
1428 			error = uiomove(mtod(m, char *) + moff, len, uio);
1429 			s = splsoftnet();
1430 			solock(so);
1431 			if (error != 0) {
1432 				/*
1433 				 * If any part of the record has been removed
1434 				 * (such as the MT_SONAME mbuf, which will
1435 				 * happen when PR_ADDR, and thus also
1436 				 * PR_ATOMIC, is set), then drop the entire
1437 				 * record to maintain the atomicity of the
1438 				 * receive operation.
1439 				 *
1440 				 * This avoids a later panic("receive 1a")
1441 				 * when compiled with DIAGNOSTIC.
1442 				 */
1443 				if (m && mbuf_removed && atomic)
1444 					(void) sbdroprecord(&so->so_rcv);
1445 
1446 				goto release;
1447 			}
1448 		} else
1449 			uio->uio_resid -= len;
1450 		if (len == m->m_len - moff) {
1451 			if (m->m_flags & M_EOR)
1452 				flags |= MSG_EOR;
1453 			if (flags & MSG_PEEK) {
1454 				m = m->m_next;
1455 				moff = 0;
1456 			} else {
1457 				nextrecord = m->m_nextpkt;
1458 				sbfree(&so->so_rcv, m);
1459 				if (mp) {
1460 					*mp = m;
1461 					mp = &m->m_next;
1462 					so->so_rcv.sb_mb = m = m->m_next;
1463 					*mp = NULL;
1464 				} else {
1465 					MFREE(m, so->so_rcv.sb_mb);
1466 					m = so->so_rcv.sb_mb;
1467 				}
1468 				/*
1469 				 * If m != NULL, we also know that
1470 				 * so->so_rcv.sb_mb != NULL.
1471 				 */
1472 				KASSERT(so->so_rcv.sb_mb == m);
1473 				if (m) {
1474 					m->m_nextpkt = nextrecord;
1475 					if (nextrecord == NULL)
1476 						so->so_rcv.sb_lastrecord = m;
1477 				} else {
1478 					so->so_rcv.sb_mb = nextrecord;
1479 					SB_EMPTY_FIXUP(&so->so_rcv);
1480 				}
1481 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1482 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1483 			}
1484 		} else if (flags & MSG_PEEK)
1485 			moff += len;
1486 		else {
1487 			if (mp != NULL) {
1488 				mt = m_copym(m, 0, len, M_NOWAIT);
1489 				if (__predict_false(mt == NULL)) {
1490 					sounlock(so);
1491 					mt = m_copym(m, 0, len, M_WAIT);
1492 					solock(so);
1493 				}
1494 				*mp = mt;
1495 			}
1496 			m->m_data += len;
1497 			m->m_len -= len;
1498 			so->so_rcv.sb_cc -= len;
1499 		}
1500 		if (so->so_oobmark) {
1501 			if ((flags & MSG_PEEK) == 0) {
1502 				so->so_oobmark -= len;
1503 				if (so->so_oobmark == 0) {
1504 					so->so_state |= SS_RCVATMARK;
1505 					break;
1506 				}
1507 			} else {
1508 				offset += len;
1509 				if (offset == so->so_oobmark)
1510 					break;
1511 			}
1512 		}
1513 		if (flags & MSG_EOR)
1514 			break;
1515 		/*
1516 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1517 		 * we must not quit until "uio->uio_resid == 0" or an error
1518 		 * termination.  If a signal/timeout occurs, return
1519 		 * with a short count but without error.
1520 		 * Keep sockbuf locked against other readers.
1521 		 */
1522 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1523 		    !sosendallatonce(so) && !nextrecord) {
1524 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1525 				break;
1526 			/*
1527 			 * If we are peeking and the socket receive buffer is
1528 			 * full, stop since we can't get more data to peek at.
1529 			 */
1530 			if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
1531 				break;
1532 			/*
1533 			 * If we've drained the socket buffer, tell the
1534 			 * protocol in case it needs to do something to
1535 			 * get it filled again.
1536 			 */
1537 			if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1538 				(*pr->pr_usrreqs->pr_rcvd)(so, flags, l);
1539 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1540 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1541 			if (wakeup_state & SS_RESTARTSYS)
1542 				error = ERESTART;
1543 			else
1544 				error = sbwait(&so->so_rcv);
1545 			if (error != 0) {
1546 				sbunlock(&so->so_rcv);
1547 				sounlock(so);
1548 				splx(s);
1549 				return 0;
1550 			}
1551 			if ((m = so->so_rcv.sb_mb) != NULL)
1552 				nextrecord = m->m_nextpkt;
1553 			wakeup_state = so->so_state;
1554 		}
1555 	}
1556 
1557 	if (m && atomic) {
1558 		flags |= MSG_TRUNC;
1559 		if ((flags & MSG_PEEK) == 0)
1560 			(void) sbdroprecord(&so->so_rcv);
1561 	}
1562 	if ((flags & MSG_PEEK) == 0) {
1563 		if (m == NULL) {
1564 			/*
1565 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1566 			 * part makes sure sb_lastrecord is up-to-date if
1567 			 * there is still data in the socket buffer.
1568 			 */
1569 			so->so_rcv.sb_mb = nextrecord;
1570 			if (so->so_rcv.sb_mb == NULL) {
1571 				so->so_rcv.sb_mbtail = NULL;
1572 				so->so_rcv.sb_lastrecord = NULL;
1573 			} else if (nextrecord->m_nextpkt == NULL)
1574 				so->so_rcv.sb_lastrecord = nextrecord;
1575 		}
1576 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1577 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1578 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1579 			(*pr->pr_usrreqs->pr_rcvd)(so, flags, l);
1580 	}
1581 	if (orig_resid == uio->uio_resid && orig_resid &&
1582 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1583 		sbunlock(&so->so_rcv);
1584 		goto restart;
1585 	}
1586 
1587 	if (flagsp != NULL)
1588 		*flagsp |= flags;
1589  release:
1590 	sbunlock(&so->so_rcv);
1591 	sounlock(so);
1592 	splx(s);
1593 	return error;
1594 }
1595 
1596 int
1597 soshutdown(struct socket *so, int how)
1598 {
1599 	const struct protosw	*pr;
1600 	int	error;
1601 
1602 	KASSERT(solocked(so));
1603 
1604 	pr = so->so_proto;
1605 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1606 		return (EINVAL);
1607 
1608 	if (how == SHUT_RD || how == SHUT_RDWR) {
1609 		sorflush(so);
1610 		error = 0;
1611 	}
1612 	if (how == SHUT_WR || how == SHUT_RDWR)
1613 		error = (*pr->pr_usrreqs->pr_shutdown)(so);
1614 
1615 	return error;
1616 }
1617 
1618 void
1619 sorestart(struct socket *so)
1620 {
1621 	/*
1622 	 * An application has called close() on an fd on which another
1623 	 * of its threads has called a socket system call.
1624 	 * Mark this and wake everyone up, and code that would block again
1625 	 * instead returns ERESTART.
1626 	 * On system call re-entry the fd is validated and EBADF returned.
1627 	 * Any other fd will block again on the 2nd syscall.
1628 	 */
1629 	solock(so);
1630 	so->so_state |= SS_RESTARTSYS;
1631 	cv_broadcast(&so->so_cv);
1632 	cv_broadcast(&so->so_snd.sb_cv);
1633 	cv_broadcast(&so->so_rcv.sb_cv);
1634 	sounlock(so);
1635 }
1636 
1637 void
1638 sorflush(struct socket *so)
1639 {
1640 	struct sockbuf	*sb, asb;
1641 	const struct protosw	*pr;
1642 
1643 	KASSERT(solocked(so));
1644 
1645 	sb = &so->so_rcv;
1646 	pr = so->so_proto;
1647 	socantrcvmore(so);
1648 	sb->sb_flags |= SB_NOINTR;
1649 	(void )sblock(sb, M_WAITOK);
1650 	sbunlock(sb);
1651 	asb = *sb;
1652 	/*
1653 	 * Clear most of the sockbuf structure, but leave some of the
1654 	 * fields valid.
1655 	 */
1656 	memset(&sb->sb_startzero, 0,
1657 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1658 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) {
1659 		sounlock(so);
1660 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1661 		solock(so);
1662 	}
1663 	sbrelease(&asb, so);
1664 }
1665 
1666 /*
1667  * internal set SOL_SOCKET options
1668  */
1669 static int
1670 sosetopt1(struct socket *so, const struct sockopt *sopt)
1671 {
1672 	int error = EINVAL, opt;
1673 	int optval = 0; /* XXX: gcc */
1674 	struct linger l;
1675 	struct timeval tv;
1676 
1677 	switch ((opt = sopt->sopt_name)) {
1678 
1679 	case SO_ACCEPTFILTER:
1680 		error = accept_filt_setopt(so, sopt);
1681 		KASSERT(solocked(so));
1682 		break;
1683 
1684   	case SO_LINGER:
1685  		error = sockopt_get(sopt, &l, sizeof(l));
1686 		solock(so);
1687  		if (error)
1688  			break;
1689  		if (l.l_linger < 0 || l.l_linger > USHRT_MAX ||
1690  		    l.l_linger > (INT_MAX / hz)) {
1691 			error = EDOM;
1692 			break;
1693 		}
1694  		so->so_linger = l.l_linger;
1695  		if (l.l_onoff)
1696  			so->so_options |= SO_LINGER;
1697  		else
1698  			so->so_options &= ~SO_LINGER;
1699    		break;
1700 
1701 	case SO_DEBUG:
1702 	case SO_KEEPALIVE:
1703 	case SO_DONTROUTE:
1704 	case SO_USELOOPBACK:
1705 	case SO_BROADCAST:
1706 	case SO_REUSEADDR:
1707 	case SO_REUSEPORT:
1708 	case SO_OOBINLINE:
1709 	case SO_TIMESTAMP:
1710 	case SO_NOSIGPIPE:
1711 #ifdef SO_OTIMESTAMP
1712 	case SO_OTIMESTAMP:
1713 #endif
1714 		error = sockopt_getint(sopt, &optval);
1715 		solock(so);
1716 		if (error)
1717 			break;
1718 		if (optval)
1719 			so->so_options |= opt;
1720 		else
1721 			so->so_options &= ~opt;
1722 		break;
1723 
1724 	case SO_SNDBUF:
1725 	case SO_RCVBUF:
1726 	case SO_SNDLOWAT:
1727 	case SO_RCVLOWAT:
1728 		error = sockopt_getint(sopt, &optval);
1729 		solock(so);
1730 		if (error)
1731 			break;
1732 
1733 		/*
1734 		 * Values < 1 make no sense for any of these
1735 		 * options, so disallow them.
1736 		 */
1737 		if (optval < 1) {
1738 			error = EINVAL;
1739 			break;
1740 		}
1741 
1742 		switch (opt) {
1743 		case SO_SNDBUF:
1744 			if (sbreserve(&so->so_snd, (u_long)optval, so) == 0) {
1745 				error = ENOBUFS;
1746 				break;
1747 			}
1748 			so->so_snd.sb_flags &= ~SB_AUTOSIZE;
1749 			break;
1750 
1751 		case SO_RCVBUF:
1752 			if (sbreserve(&so->so_rcv, (u_long)optval, so) == 0) {
1753 				error = ENOBUFS;
1754 				break;
1755 			}
1756 			so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1757 			break;
1758 
1759 		/*
1760 		 * Make sure the low-water is never greater than
1761 		 * the high-water.
1762 		 */
1763 		case SO_SNDLOWAT:
1764 			if (optval > so->so_snd.sb_hiwat)
1765 				optval = so->so_snd.sb_hiwat;
1766 
1767 			so->so_snd.sb_lowat = optval;
1768 			break;
1769 
1770 		case SO_RCVLOWAT:
1771 			if (optval > so->so_rcv.sb_hiwat)
1772 				optval = so->so_rcv.sb_hiwat;
1773 
1774 			so->so_rcv.sb_lowat = optval;
1775 			break;
1776 		}
1777 		break;
1778 
1779 #ifdef COMPAT_50
1780 	case SO_OSNDTIMEO:
1781 	case SO_ORCVTIMEO: {
1782 		struct timeval50 otv;
1783 		error = sockopt_get(sopt, &otv, sizeof(otv));
1784 		if (error) {
1785 			solock(so);
1786 			break;
1787 		}
1788 		timeval50_to_timeval(&otv, &tv);
1789 		opt = opt == SO_OSNDTIMEO ? SO_SNDTIMEO : SO_RCVTIMEO;
1790 		error = 0;
1791 		/*FALLTHROUGH*/
1792 	}
1793 #endif /* COMPAT_50 */
1794 
1795 	case SO_SNDTIMEO:
1796 	case SO_RCVTIMEO:
1797 		if (error)
1798 			error = sockopt_get(sopt, &tv, sizeof(tv));
1799 		solock(so);
1800 		if (error)
1801 			break;
1802 
1803 		if (tv.tv_sec > (INT_MAX - tv.tv_usec / tick) / hz) {
1804 			error = EDOM;
1805 			break;
1806 		}
1807 
1808 		optval = tv.tv_sec * hz + tv.tv_usec / tick;
1809 		if (optval == 0 && tv.tv_usec != 0)
1810 			optval = 1;
1811 
1812 		switch (opt) {
1813 		case SO_SNDTIMEO:
1814 			so->so_snd.sb_timeo = optval;
1815 			break;
1816 		case SO_RCVTIMEO:
1817 			so->so_rcv.sb_timeo = optval;
1818 			break;
1819 		}
1820 		break;
1821 
1822 	default:
1823 		solock(so);
1824 		error = ENOPROTOOPT;
1825 		break;
1826 	}
1827 	KASSERT(solocked(so));
1828 	return error;
1829 }
1830 
1831 int
1832 sosetopt(struct socket *so, struct sockopt *sopt)
1833 {
1834 	int error, prerr;
1835 
1836 	if (sopt->sopt_level == SOL_SOCKET) {
1837 		error = sosetopt1(so, sopt);
1838 		KASSERT(solocked(so));
1839 	} else {
1840 		error = ENOPROTOOPT;
1841 		solock(so);
1842 	}
1843 
1844 	if ((error == 0 || error == ENOPROTOOPT) &&
1845 	    so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) {
1846 		/* give the protocol stack a shot */
1847 		prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, sopt);
1848 		if (prerr == 0)
1849 			error = 0;
1850 		else if (prerr != ENOPROTOOPT)
1851 			error = prerr;
1852 	}
1853 	sounlock(so);
1854 	return error;
1855 }
1856 
1857 /*
1858  * so_setsockopt() is a wrapper providing a sockopt structure for sosetopt()
1859  */
1860 int
1861 so_setsockopt(struct lwp *l, struct socket *so, int level, int name,
1862     const void *val, size_t valsize)
1863 {
1864 	struct sockopt sopt;
1865 	int error;
1866 
1867 	KASSERT(valsize == 0 || val != NULL);
1868 
1869 	sockopt_init(&sopt, level, name, valsize);
1870 	sockopt_set(&sopt, val, valsize);
1871 
1872 	error = sosetopt(so, &sopt);
1873 
1874 	sockopt_destroy(&sopt);
1875 
1876 	return error;
1877 }
1878 
1879 /*
1880  * internal get SOL_SOCKET options
1881  */
1882 static int
1883 sogetopt1(struct socket *so, struct sockopt *sopt)
1884 {
1885 	int error, optval, opt;
1886 	struct linger l;
1887 	struct timeval tv;
1888 
1889 	switch ((opt = sopt->sopt_name)) {
1890 
1891 	case SO_ACCEPTFILTER:
1892 		error = accept_filt_getopt(so, sopt);
1893 		break;
1894 
1895 	case SO_LINGER:
1896 		l.l_onoff = (so->so_options & SO_LINGER) ? 1 : 0;
1897 		l.l_linger = so->so_linger;
1898 
1899 		error = sockopt_set(sopt, &l, sizeof(l));
1900 		break;
1901 
1902 	case SO_USELOOPBACK:
1903 	case SO_DONTROUTE:
1904 	case SO_DEBUG:
1905 	case SO_KEEPALIVE:
1906 	case SO_REUSEADDR:
1907 	case SO_REUSEPORT:
1908 	case SO_BROADCAST:
1909 	case SO_OOBINLINE:
1910 	case SO_TIMESTAMP:
1911 	case SO_NOSIGPIPE:
1912 #ifdef SO_OTIMESTAMP
1913 	case SO_OTIMESTAMP:
1914 #endif
1915 	case SO_ACCEPTCONN:
1916 		error = sockopt_setint(sopt, (so->so_options & opt) ? 1 : 0);
1917 		break;
1918 
1919 	case SO_TYPE:
1920 		error = sockopt_setint(sopt, so->so_type);
1921 		break;
1922 
1923 	case SO_ERROR:
1924 		error = sockopt_setint(sopt, so->so_error);
1925 		so->so_error = 0;
1926 		break;
1927 
1928 	case SO_SNDBUF:
1929 		error = sockopt_setint(sopt, so->so_snd.sb_hiwat);
1930 		break;
1931 
1932 	case SO_RCVBUF:
1933 		error = sockopt_setint(sopt, so->so_rcv.sb_hiwat);
1934 		break;
1935 
1936 	case SO_SNDLOWAT:
1937 		error = sockopt_setint(sopt, so->so_snd.sb_lowat);
1938 		break;
1939 
1940 	case SO_RCVLOWAT:
1941 		error = sockopt_setint(sopt, so->so_rcv.sb_lowat);
1942 		break;
1943 
1944 #ifdef COMPAT_50
1945 	case SO_OSNDTIMEO:
1946 	case SO_ORCVTIMEO: {
1947 		struct timeval50 otv;
1948 
1949 		optval = (opt == SO_OSNDTIMEO ?
1950 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1951 
1952 		otv.tv_sec = optval / hz;
1953 		otv.tv_usec = (optval % hz) * tick;
1954 
1955 		error = sockopt_set(sopt, &otv, sizeof(otv));
1956 		break;
1957 	}
1958 #endif /* COMPAT_50 */
1959 
1960 	case SO_SNDTIMEO:
1961 	case SO_RCVTIMEO:
1962 		optval = (opt == SO_SNDTIMEO ?
1963 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1964 
1965 		tv.tv_sec = optval / hz;
1966 		tv.tv_usec = (optval % hz) * tick;
1967 
1968 		error = sockopt_set(sopt, &tv, sizeof(tv));
1969 		break;
1970 
1971 	case SO_OVERFLOWED:
1972 		error = sockopt_setint(sopt, so->so_rcv.sb_overflowed);
1973 		break;
1974 
1975 	default:
1976 		error = ENOPROTOOPT;
1977 		break;
1978 	}
1979 
1980 	return (error);
1981 }
1982 
1983 int
1984 sogetopt(struct socket *so, struct sockopt *sopt)
1985 {
1986 	int		error;
1987 
1988 	solock(so);
1989 	if (sopt->sopt_level != SOL_SOCKET) {
1990 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1991 			error = ((*so->so_proto->pr_ctloutput)
1992 			    (PRCO_GETOPT, so, sopt));
1993 		} else
1994 			error = (ENOPROTOOPT);
1995 	} else {
1996 		error = sogetopt1(so, sopt);
1997 	}
1998 	sounlock(so);
1999 	return (error);
2000 }
2001 
2002 /*
2003  * alloc sockopt data buffer buffer
2004  *	- will be released at destroy
2005  */
2006 static int
2007 sockopt_alloc(struct sockopt *sopt, size_t len, km_flag_t kmflag)
2008 {
2009 
2010 	KASSERT(sopt->sopt_size == 0);
2011 
2012 	if (len > sizeof(sopt->sopt_buf)) {
2013 		sopt->sopt_data = kmem_zalloc(len, kmflag);
2014 		if (sopt->sopt_data == NULL)
2015 			return ENOMEM;
2016 	} else
2017 		sopt->sopt_data = sopt->sopt_buf;
2018 
2019 	sopt->sopt_size = len;
2020 	return 0;
2021 }
2022 
2023 /*
2024  * initialise sockopt storage
2025  *	- MAY sleep during allocation
2026  */
2027 void
2028 sockopt_init(struct sockopt *sopt, int level, int name, size_t size)
2029 {
2030 
2031 	memset(sopt, 0, sizeof(*sopt));
2032 
2033 	sopt->sopt_level = level;
2034 	sopt->sopt_name = name;
2035 	(void)sockopt_alloc(sopt, size, KM_SLEEP);
2036 }
2037 
2038 /*
2039  * destroy sockopt storage
2040  *	- will release any held memory references
2041  */
2042 void
2043 sockopt_destroy(struct sockopt *sopt)
2044 {
2045 
2046 	if (sopt->sopt_data != sopt->sopt_buf)
2047 		kmem_free(sopt->sopt_data, sopt->sopt_size);
2048 
2049 	memset(sopt, 0, sizeof(*sopt));
2050 }
2051 
2052 /*
2053  * set sockopt value
2054  *	- value is copied into sockopt
2055  * 	- memory is allocated when necessary, will not sleep
2056  */
2057 int
2058 sockopt_set(struct sockopt *sopt, const void *buf, size_t len)
2059 {
2060 	int error;
2061 
2062 	if (sopt->sopt_size == 0) {
2063 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2064 		if (error)
2065 			return error;
2066 	}
2067 
2068 	KASSERT(sopt->sopt_size == len);
2069 	memcpy(sopt->sopt_data, buf, len);
2070 	return 0;
2071 }
2072 
2073 /*
2074  * common case of set sockopt integer value
2075  */
2076 int
2077 sockopt_setint(struct sockopt *sopt, int val)
2078 {
2079 
2080 	return sockopt_set(sopt, &val, sizeof(int));
2081 }
2082 
2083 /*
2084  * get sockopt value
2085  *	- correct size must be given
2086  */
2087 int
2088 sockopt_get(const struct sockopt *sopt, void *buf, size_t len)
2089 {
2090 
2091 	if (sopt->sopt_size != len)
2092 		return EINVAL;
2093 
2094 	memcpy(buf, sopt->sopt_data, len);
2095 	return 0;
2096 }
2097 
2098 /*
2099  * common case of get sockopt integer value
2100  */
2101 int
2102 sockopt_getint(const struct sockopt *sopt, int *valp)
2103 {
2104 
2105 	return sockopt_get(sopt, valp, sizeof(int));
2106 }
2107 
2108 /*
2109  * set sockopt value from mbuf
2110  *	- ONLY for legacy code
2111  *	- mbuf is released by sockopt
2112  *	- will not sleep
2113  */
2114 int
2115 sockopt_setmbuf(struct sockopt *sopt, struct mbuf *m)
2116 {
2117 	size_t len;
2118 	int error;
2119 
2120 	len = m_length(m);
2121 
2122 	if (sopt->sopt_size == 0) {
2123 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2124 		if (error)
2125 			return error;
2126 	}
2127 
2128 	KASSERT(sopt->sopt_size == len);
2129 	m_copydata(m, 0, len, sopt->sopt_data);
2130 	m_freem(m);
2131 
2132 	return 0;
2133 }
2134 
2135 /*
2136  * get sockopt value into mbuf
2137  *	- ONLY for legacy code
2138  *	- mbuf to be released by the caller
2139  *	- will not sleep
2140  */
2141 struct mbuf *
2142 sockopt_getmbuf(const struct sockopt *sopt)
2143 {
2144 	struct mbuf *m;
2145 
2146 	if (sopt->sopt_size > MCLBYTES)
2147 		return NULL;
2148 
2149 	m = m_get(M_DONTWAIT, MT_SOOPTS);
2150 	if (m == NULL)
2151 		return NULL;
2152 
2153 	if (sopt->sopt_size > MLEN) {
2154 		MCLGET(m, M_DONTWAIT);
2155 		if ((m->m_flags & M_EXT) == 0) {
2156 			m_free(m);
2157 			return NULL;
2158 		}
2159 	}
2160 
2161 	memcpy(mtod(m, void *), sopt->sopt_data, sopt->sopt_size);
2162 	m->m_len = sopt->sopt_size;
2163 
2164 	return m;
2165 }
2166 
2167 void
2168 sohasoutofband(struct socket *so)
2169 {
2170 
2171 	fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
2172 	selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, NOTE_SUBMIT);
2173 }
2174 
2175 static void
2176 filt_sordetach(struct knote *kn)
2177 {
2178 	struct socket	*so;
2179 
2180 	so = ((file_t *)kn->kn_obj)->f_socket;
2181 	solock(so);
2182 	SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
2183 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
2184 		so->so_rcv.sb_flags &= ~SB_KNOTE;
2185 	sounlock(so);
2186 }
2187 
2188 /*ARGSUSED*/
2189 static int
2190 filt_soread(struct knote *kn, long hint)
2191 {
2192 	struct socket	*so;
2193 	int rv;
2194 
2195 	so = ((file_t *)kn->kn_obj)->f_socket;
2196 	if (hint != NOTE_SUBMIT)
2197 		solock(so);
2198 	kn->kn_data = so->so_rcv.sb_cc;
2199 	if (so->so_state & SS_CANTRCVMORE) {
2200 		kn->kn_flags |= EV_EOF;
2201 		kn->kn_fflags = so->so_error;
2202 		rv = 1;
2203 	} else if (so->so_error)	/* temporary udp error */
2204 		rv = 1;
2205 	else if (kn->kn_sfflags & NOTE_LOWAT)
2206 		rv = (kn->kn_data >= kn->kn_sdata);
2207 	else
2208 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
2209 	if (hint != NOTE_SUBMIT)
2210 		sounlock(so);
2211 	return rv;
2212 }
2213 
2214 static void
2215 filt_sowdetach(struct knote *kn)
2216 {
2217 	struct socket	*so;
2218 
2219 	so = ((file_t *)kn->kn_obj)->f_socket;
2220 	solock(so);
2221 	SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
2222 	if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
2223 		so->so_snd.sb_flags &= ~SB_KNOTE;
2224 	sounlock(so);
2225 }
2226 
2227 /*ARGSUSED*/
2228 static int
2229 filt_sowrite(struct knote *kn, long hint)
2230 {
2231 	struct socket	*so;
2232 	int rv;
2233 
2234 	so = ((file_t *)kn->kn_obj)->f_socket;
2235 	if (hint != NOTE_SUBMIT)
2236 		solock(so);
2237 	kn->kn_data = sbspace(&so->so_snd);
2238 	if (so->so_state & SS_CANTSENDMORE) {
2239 		kn->kn_flags |= EV_EOF;
2240 		kn->kn_fflags = so->so_error;
2241 		rv = 1;
2242 	} else if (so->so_error)	/* temporary udp error */
2243 		rv = 1;
2244 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2245 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
2246 		rv = 0;
2247 	else if (kn->kn_sfflags & NOTE_LOWAT)
2248 		rv = (kn->kn_data >= kn->kn_sdata);
2249 	else
2250 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
2251 	if (hint != NOTE_SUBMIT)
2252 		sounlock(so);
2253 	return rv;
2254 }
2255 
2256 /*ARGSUSED*/
2257 static int
2258 filt_solisten(struct knote *kn, long hint)
2259 {
2260 	struct socket	*so;
2261 	int rv;
2262 
2263 	so = ((file_t *)kn->kn_obj)->f_socket;
2264 
2265 	/*
2266 	 * Set kn_data to number of incoming connections, not
2267 	 * counting partial (incomplete) connections.
2268 	 */
2269 	if (hint != NOTE_SUBMIT)
2270 		solock(so);
2271 	kn->kn_data = so->so_qlen;
2272 	rv = (kn->kn_data > 0);
2273 	if (hint != NOTE_SUBMIT)
2274 		sounlock(so);
2275 	return rv;
2276 }
2277 
2278 static const struct filterops solisten_filtops =
2279 	{ 1, NULL, filt_sordetach, filt_solisten };
2280 static const struct filterops soread_filtops =
2281 	{ 1, NULL, filt_sordetach, filt_soread };
2282 static const struct filterops sowrite_filtops =
2283 	{ 1, NULL, filt_sowdetach, filt_sowrite };
2284 
2285 int
2286 soo_kqfilter(struct file *fp, struct knote *kn)
2287 {
2288 	struct socket	*so;
2289 	struct sockbuf	*sb;
2290 
2291 	so = ((file_t *)kn->kn_obj)->f_socket;
2292 	solock(so);
2293 	switch (kn->kn_filter) {
2294 	case EVFILT_READ:
2295 		if (so->so_options & SO_ACCEPTCONN)
2296 			kn->kn_fop = &solisten_filtops;
2297 		else
2298 			kn->kn_fop = &soread_filtops;
2299 		sb = &so->so_rcv;
2300 		break;
2301 	case EVFILT_WRITE:
2302 		kn->kn_fop = &sowrite_filtops;
2303 		sb = &so->so_snd;
2304 		break;
2305 	default:
2306 		sounlock(so);
2307 		return (EINVAL);
2308 	}
2309 	SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
2310 	sb->sb_flags |= SB_KNOTE;
2311 	sounlock(so);
2312 	return (0);
2313 }
2314 
2315 static int
2316 sodopoll(struct socket *so, int events)
2317 {
2318 	int revents;
2319 
2320 	revents = 0;
2321 
2322 	if (events & (POLLIN | POLLRDNORM))
2323 		if (soreadable(so))
2324 			revents |= events & (POLLIN | POLLRDNORM);
2325 
2326 	if (events & (POLLOUT | POLLWRNORM))
2327 		if (sowritable(so))
2328 			revents |= events & (POLLOUT | POLLWRNORM);
2329 
2330 	if (events & (POLLPRI | POLLRDBAND))
2331 		if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
2332 			revents |= events & (POLLPRI | POLLRDBAND);
2333 
2334 	return revents;
2335 }
2336 
2337 int
2338 sopoll(struct socket *so, int events)
2339 {
2340 	int revents = 0;
2341 
2342 #ifndef DIAGNOSTIC
2343 	/*
2344 	 * Do a quick, unlocked check in expectation that the socket
2345 	 * will be ready for I/O.  Don't do this check if DIAGNOSTIC,
2346 	 * as the solocked() assertions will fail.
2347 	 */
2348 	if ((revents = sodopoll(so, events)) != 0)
2349 		return revents;
2350 #endif
2351 
2352 	solock(so);
2353 	if ((revents = sodopoll(so, events)) == 0) {
2354 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2355 			selrecord(curlwp, &so->so_rcv.sb_sel);
2356 			so->so_rcv.sb_flags |= SB_NOTIFY;
2357 		}
2358 
2359 		if (events & (POLLOUT | POLLWRNORM)) {
2360 			selrecord(curlwp, &so->so_snd.sb_sel);
2361 			so->so_snd.sb_flags |= SB_NOTIFY;
2362 		}
2363 	}
2364 	sounlock(so);
2365 
2366 	return revents;
2367 }
2368 
2369 
2370 #include <sys/sysctl.h>
2371 
2372 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO);
2373 static int sysctl_kern_sbmax(SYSCTLFN_PROTO);
2374 
2375 /*
2376  * sysctl helper routine for kern.somaxkva.  ensures that the given
2377  * value is not too small.
2378  * (XXX should we maybe make sure it's not too large as well?)
2379  */
2380 static int
2381 sysctl_kern_somaxkva(SYSCTLFN_ARGS)
2382 {
2383 	int error, new_somaxkva;
2384 	struct sysctlnode node;
2385 
2386 	new_somaxkva = somaxkva;
2387 	node = *rnode;
2388 	node.sysctl_data = &new_somaxkva;
2389 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2390 	if (error || newp == NULL)
2391 		return (error);
2392 
2393 	if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */
2394 		return (EINVAL);
2395 
2396 	mutex_enter(&so_pendfree_lock);
2397 	somaxkva = new_somaxkva;
2398 	cv_broadcast(&socurkva_cv);
2399 	mutex_exit(&so_pendfree_lock);
2400 
2401 	return (error);
2402 }
2403 
2404 /*
2405  * sysctl helper routine for kern.sbmax. Basically just ensures that
2406  * any new value is not too small.
2407  */
2408 static int
2409 sysctl_kern_sbmax(SYSCTLFN_ARGS)
2410 {
2411 	int error, new_sbmax;
2412 	struct sysctlnode node;
2413 
2414 	new_sbmax = sb_max;
2415 	node = *rnode;
2416 	node.sysctl_data = &new_sbmax;
2417 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2418 	if (error || newp == NULL)
2419 		return (error);
2420 
2421 	KERNEL_LOCK(1, NULL);
2422 	error = sb_max_set(new_sbmax);
2423 	KERNEL_UNLOCK_ONE(NULL);
2424 
2425 	return (error);
2426 }
2427 
2428 static void
2429 sysctl_kern_socket_setup(void)
2430 {
2431 
2432 	KASSERT(socket_sysctllog == NULL);
2433 
2434 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2435 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2436 		       CTLTYPE_INT, "somaxkva",
2437 		       SYSCTL_DESCR("Maximum amount of kernel memory to be "
2438 				    "used for socket buffers"),
2439 		       sysctl_kern_somaxkva, 0, NULL, 0,
2440 		       CTL_KERN, KERN_SOMAXKVA, CTL_EOL);
2441 
2442 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2443 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2444 		       CTLTYPE_INT, "sbmax",
2445 		       SYSCTL_DESCR("Maximum socket buffer size"),
2446 		       sysctl_kern_sbmax, 0, NULL, 0,
2447 		       CTL_KERN, KERN_SBMAX, CTL_EOL);
2448 }
2449