xref: /netbsd-src/sys/kern/uipc_socket.c (revision a536ee5124e62c9a0051a252f7833dc8f50f44c9)
1 /*	$NetBSD: uipc_socket.c,v 1.212 2012/10/08 19:20:45 pooka 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 1.212 2012/10/08 19:20:45 pooka Exp $");
67 
68 #include "opt_compat_netbsd.h"
69 #include "opt_sock_counters.h"
70 #include "opt_sosend_loan.h"
71 #include "opt_mbuftrace.h"
72 #include "opt_somaxkva.h"
73 #include "opt_multiprocessor.h"	/* XXX */
74 
75 #include <sys/param.h>
76 #include <sys/systm.h>
77 #include <sys/proc.h>
78 #include <sys/file.h>
79 #include <sys/filedesc.h>
80 #include <sys/kmem.h>
81 #include <sys/mbuf.h>
82 #include <sys/domain.h>
83 #include <sys/kernel.h>
84 #include <sys/protosw.h>
85 #include <sys/socket.h>
86 #include <sys/socketvar.h>
87 #include <sys/signalvar.h>
88 #include <sys/resourcevar.h>
89 #include <sys/uidinfo.h>
90 #include <sys/event.h>
91 #include <sys/poll.h>
92 #include <sys/kauth.h>
93 #include <sys/mutex.h>
94 #include <sys/condvar.h>
95 #include <sys/kthread.h>
96 
97 #ifdef COMPAT_50
98 #include <compat/sys/time.h>
99 #include <compat/sys/socket.h>
100 #endif
101 
102 #include <uvm/uvm_extern.h>
103 #include <uvm/uvm_loan.h>
104 #include <uvm/uvm_page.h>
105 
106 MALLOC_DEFINE(M_SOOPTS, "soopts", "socket options");
107 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
108 
109 extern const struct fileops socketops;
110 
111 extern int	somaxconn;			/* patchable (XXX sysctl) */
112 int		somaxconn = SOMAXCONN;
113 kmutex_t	*softnet_lock;
114 
115 #ifdef SOSEND_COUNTERS
116 #include <sys/device.h>
117 
118 static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
119     NULL, "sosend", "loan big");
120 static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
121     NULL, "sosend", "copy big");
122 static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
123     NULL, "sosend", "copy small");
124 static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
125     NULL, "sosend", "kva limit");
126 
127 #define	SOSEND_COUNTER_INCR(ev)		(ev)->ev_count++
128 
129 EVCNT_ATTACH_STATIC(sosend_loan_big);
130 EVCNT_ATTACH_STATIC(sosend_copy_big);
131 EVCNT_ATTACH_STATIC(sosend_copy_small);
132 EVCNT_ATTACH_STATIC(sosend_kvalimit);
133 #else
134 
135 #define	SOSEND_COUNTER_INCR(ev)		/* nothing */
136 
137 #endif /* SOSEND_COUNTERS */
138 
139 #if defined(SOSEND_NO_LOAN) || defined(MULTIPROCESSOR)
140 int sock_loan_thresh = -1;
141 #else
142 int sock_loan_thresh = 4096;
143 #endif
144 
145 static kmutex_t so_pendfree_lock;
146 static struct mbuf *so_pendfree = NULL;
147 
148 #ifndef SOMAXKVA
149 #define	SOMAXKVA (16 * 1024 * 1024)
150 #endif
151 int somaxkva = SOMAXKVA;
152 static int socurkva;
153 static kcondvar_t socurkva_cv;
154 
155 static kauth_listener_t socket_listener;
156 
157 #define	SOCK_LOAN_CHUNK		65536
158 
159 static void sopendfree_thread(void *);
160 static kcondvar_t pendfree_thread_cv;
161 static lwp_t *sopendfree_lwp;
162 
163 static void sysctl_kern_socket_setup(void);
164 static struct sysctllog *socket_sysctllog;
165 
166 static vsize_t
167 sokvareserve(struct socket *so, vsize_t len)
168 {
169 	int error;
170 
171 	mutex_enter(&so_pendfree_lock);
172 	while (socurkva + len > somaxkva) {
173 		SOSEND_COUNTER_INCR(&sosend_kvalimit);
174 		error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock);
175 		if (error) {
176 			len = 0;
177 			break;
178 		}
179 	}
180 	socurkva += len;
181 	mutex_exit(&so_pendfree_lock);
182 	return len;
183 }
184 
185 static void
186 sokvaunreserve(vsize_t len)
187 {
188 
189 	mutex_enter(&so_pendfree_lock);
190 	socurkva -= len;
191 	cv_broadcast(&socurkva_cv);
192 	mutex_exit(&so_pendfree_lock);
193 }
194 
195 /*
196  * sokvaalloc: allocate kva for loan.
197  */
198 
199 vaddr_t
200 sokvaalloc(vaddr_t sva, vsize_t len, struct socket *so)
201 {
202 	vaddr_t lva;
203 
204 	/*
205 	 * reserve kva.
206 	 */
207 
208 	if (sokvareserve(so, len) == 0)
209 		return 0;
210 
211 	/*
212 	 * allocate kva.
213 	 */
214 
215 	lva = uvm_km_alloc(kernel_map, len, atop(sva) & uvmexp.colormask,
216 	    UVM_KMF_COLORMATCH | UVM_KMF_VAONLY | UVM_KMF_WAITVA);
217 	if (lva == 0) {
218 		sokvaunreserve(len);
219 		return (0);
220 	}
221 
222 	return lva;
223 }
224 
225 /*
226  * sokvafree: free kva for loan.
227  */
228 
229 void
230 sokvafree(vaddr_t sva, vsize_t len)
231 {
232 
233 	/*
234 	 * free kva.
235 	 */
236 
237 	uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY);
238 
239 	/*
240 	 * unreserve kva.
241 	 */
242 
243 	sokvaunreserve(len);
244 }
245 
246 static void
247 sodoloanfree(struct vm_page **pgs, void *buf, size_t size)
248 {
249 	vaddr_t sva, eva;
250 	vsize_t len;
251 	int npgs;
252 
253 	KASSERT(pgs != NULL);
254 
255 	eva = round_page((vaddr_t) buf + size);
256 	sva = trunc_page((vaddr_t) buf);
257 	len = eva - sva;
258 	npgs = len >> PAGE_SHIFT;
259 
260 	pmap_kremove(sva, len);
261 	pmap_update(pmap_kernel());
262 	uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE);
263 	sokvafree(sva, len);
264 }
265 
266 /*
267  * sopendfree_thread: free mbufs on "pendfree" list.
268  * unlock and relock so_pendfree_lock when freeing mbufs.
269  */
270 
271 static void
272 sopendfree_thread(void *v)
273 {
274 	struct mbuf *m, *next;
275 	size_t rv;
276 
277 	mutex_enter(&so_pendfree_lock);
278 
279 	for (;;) {
280 		rv = 0;
281 		while (so_pendfree != NULL) {
282 			m = so_pendfree;
283 			so_pendfree = NULL;
284 			mutex_exit(&so_pendfree_lock);
285 
286 			for (; m != NULL; m = next) {
287 				next = m->m_next;
288 				KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0);
289 				KASSERT(m->m_ext.ext_refcnt == 0);
290 
291 				rv += m->m_ext.ext_size;
292 				sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf,
293 				    m->m_ext.ext_size);
294 				pool_cache_put(mb_cache, m);
295 			}
296 
297 			mutex_enter(&so_pendfree_lock);
298 		}
299 		if (rv)
300 			cv_broadcast(&socurkva_cv);
301 		cv_wait(&pendfree_thread_cv, &so_pendfree_lock);
302 	}
303 	panic("sopendfree_thread");
304 	/* NOTREACHED */
305 }
306 
307 void
308 soloanfree(struct mbuf *m, void *buf, size_t size, void *arg)
309 {
310 
311 	KASSERT(m != NULL);
312 
313 	/*
314 	 * postpone freeing mbuf.
315 	 *
316 	 * we can't do it in interrupt context
317 	 * because we need to put kva back to kernel_map.
318 	 */
319 
320 	mutex_enter(&so_pendfree_lock);
321 	m->m_next = so_pendfree;
322 	so_pendfree = m;
323 	cv_signal(&pendfree_thread_cv);
324 	mutex_exit(&so_pendfree_lock);
325 }
326 
327 static long
328 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space)
329 {
330 	struct iovec *iov = uio->uio_iov;
331 	vaddr_t sva, eva;
332 	vsize_t len;
333 	vaddr_t lva;
334 	int npgs, error;
335 	vaddr_t va;
336 	int i;
337 
338 	if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace))
339 		return (0);
340 
341 	if (iov->iov_len < (size_t) space)
342 		space = iov->iov_len;
343 	if (space > SOCK_LOAN_CHUNK)
344 		space = SOCK_LOAN_CHUNK;
345 
346 	eva = round_page((vaddr_t) iov->iov_base + space);
347 	sva = trunc_page((vaddr_t) iov->iov_base);
348 	len = eva - sva;
349 	npgs = len >> PAGE_SHIFT;
350 
351 	KASSERT(npgs <= M_EXT_MAXPAGES);
352 
353 	lva = sokvaalloc(sva, len, so);
354 	if (lva == 0)
355 		return 0;
356 
357 	error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len,
358 	    m->m_ext.ext_pgs, UVM_LOAN_TOPAGE);
359 	if (error) {
360 		sokvafree(lva, len);
361 		return (0);
362 	}
363 
364 	for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE)
365 		pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
366 		    VM_PROT_READ, 0);
367 	pmap_update(pmap_kernel());
368 
369 	lva += (vaddr_t) iov->iov_base & PAGE_MASK;
370 
371 	MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so);
372 	m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
373 
374 	uio->uio_resid -= space;
375 	/* uio_offset not updated, not set/used for write(2) */
376 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space;
377 	uio->uio_iov->iov_len -= space;
378 	if (uio->uio_iov->iov_len == 0) {
379 		uio->uio_iov++;
380 		uio->uio_iovcnt--;
381 	}
382 
383 	return (space);
384 }
385 
386 struct mbuf *
387 getsombuf(struct socket *so, int type)
388 {
389 	struct mbuf *m;
390 
391 	m = m_get(M_WAIT, type);
392 	MCLAIM(m, so->so_mowner);
393 	return m;
394 }
395 
396 static int
397 socket_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
398     void *arg0, void *arg1, void *arg2, void *arg3)
399 {
400 	int result;
401 	enum kauth_network_req req;
402 
403 	result = KAUTH_RESULT_DEFER;
404 	req = (enum kauth_network_req)arg0;
405 
406 	if ((action != KAUTH_NETWORK_SOCKET) &&
407 	    (action != KAUTH_NETWORK_BIND))
408 		return result;
409 
410 	switch (req) {
411 	case KAUTH_REQ_NETWORK_BIND_PORT:
412 		result = KAUTH_RESULT_ALLOW;
413 		break;
414 
415 	case KAUTH_REQ_NETWORK_SOCKET_DROP: {
416 		/* Normal users can only drop their own connections. */
417 		struct socket *so = (struct socket *)arg1;
418 
419 		if (proc_uidmatch(cred, so->so_cred))
420 			result = KAUTH_RESULT_ALLOW;
421 
422 		break;
423 		}
424 
425 	case KAUTH_REQ_NETWORK_SOCKET_OPEN:
426 		/* We allow "raw" routing/bluetooth sockets to anyone. */
427 		if ((u_long)arg1 == PF_ROUTE || (u_long)arg1 == PF_OROUTE
428 		    || (u_long)arg1 == PF_BLUETOOTH) {
429 			result = KAUTH_RESULT_ALLOW;
430 		} else {
431 			/* Privileged, let secmodel handle this. */
432 			if ((u_long)arg2 == SOCK_RAW)
433 				break;
434 		}
435 
436 		result = KAUTH_RESULT_ALLOW;
437 
438 		break;
439 
440 	case KAUTH_REQ_NETWORK_SOCKET_CANSEE:
441 		result = KAUTH_RESULT_ALLOW;
442 
443 		break;
444 
445 	default:
446 		break;
447 	}
448 
449 	return result;
450 }
451 
452 void
453 soinit(void)
454 {
455 
456 	sysctl_kern_socket_setup();
457 
458 	mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM);
459 	softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
460 	cv_init(&socurkva_cv, "sokva");
461 	cv_init(&pendfree_thread_cv, "sopendfr");
462 	soinit2();
463 
464 	/* Set the initial adjusted socket buffer size. */
465 	if (sb_max_set(sb_max))
466 		panic("bad initial sb_max value: %lu", sb_max);
467 
468 	socket_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK,
469 	    socket_listener_cb, NULL);
470 }
471 
472 void
473 soinit1(void)
474 {
475 	int error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL,
476 	    sopendfree_thread, NULL, &sopendfree_lwp, "sopendfree");
477 	if (error)
478 		panic("soinit1 %d", error);
479 }
480 
481 /*
482  * Socket operation routines.
483  * These routines are called by the routines in
484  * sys_socket.c or from a system process, and
485  * implement the semantics of socket operations by
486  * switching out to the protocol specific routines.
487  */
488 /*ARGSUSED*/
489 int
490 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l,
491 	 struct socket *lockso)
492 {
493 	const struct protosw	*prp;
494 	struct socket	*so;
495 	uid_t		uid;
496 	int		error;
497 	kmutex_t	*lock;
498 
499 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET,
500 	    KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type),
501 	    KAUTH_ARG(proto));
502 	if (error != 0)
503 		return error;
504 
505 	if (proto)
506 		prp = pffindproto(dom, proto, type);
507 	else
508 		prp = pffindtype(dom, type);
509 	if (prp == NULL) {
510 		/* no support for domain */
511 		if (pffinddomain(dom) == 0)
512 			return EAFNOSUPPORT;
513 		/* no support for socket type */
514 		if (proto == 0 && type != 0)
515 			return EPROTOTYPE;
516 		return EPROTONOSUPPORT;
517 	}
518 	if (prp->pr_usrreq == NULL)
519 		return EPROTONOSUPPORT;
520 	if (prp->pr_type != type)
521 		return EPROTOTYPE;
522 
523 	so = soget(true);
524 	so->so_type = type;
525 	so->so_proto = prp;
526 	so->so_send = sosend;
527 	so->so_receive = soreceive;
528 #ifdef MBUFTRACE
529 	so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner;
530 	so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner;
531 	so->so_mowner = &prp->pr_domain->dom_mowner;
532 #endif
533 	uid = kauth_cred_geteuid(l->l_cred);
534 	so->so_uidinfo = uid_find(uid);
535 	so->so_cpid = l->l_proc->p_pid;
536 	if (lockso != NULL) {
537 		/* Caller wants us to share a lock. */
538 		lock = lockso->so_lock;
539 		so->so_lock = lock;
540 		mutex_obj_hold(lock);
541 		mutex_enter(lock);
542 	} else {
543 		/* Lock assigned and taken during PRU_ATTACH. */
544 	}
545 	error = (*prp->pr_usrreq)(so, PRU_ATTACH, NULL,
546 	    (struct mbuf *)(long)proto, NULL, l);
547 	KASSERT(solocked(so));
548 	if (error != 0) {
549 		so->so_state |= SS_NOFDREF;
550 		sofree(so);
551 		return error;
552 	}
553 	so->so_cred = kauth_cred_dup(l->l_cred);
554 	sounlock(so);
555 	*aso = so;
556 	return 0;
557 }
558 
559 /* On success, write file descriptor to fdout and return zero.  On
560  * failure, return non-zero; *fdout will be undefined.
561  */
562 int
563 fsocreate(int domain, struct socket **sop, int type, int protocol,
564     struct lwp *l, int *fdout)
565 {
566 	struct socket	*so;
567 	struct file	*fp;
568 	int		fd, error;
569 	int		flags = type & SOCK_FLAGS_MASK;
570 
571 	type &= ~SOCK_FLAGS_MASK;
572 	if ((error = fd_allocfile(&fp, &fd)) != 0)
573 		return error;
574 	fd_set_exclose(l, fd, (flags & SOCK_CLOEXEC) != 0);
575 	fp->f_flag = FREAD|FWRITE|((flags & SOCK_NONBLOCK) ? FNONBLOCK : 0)|
576 	    ((flags & SOCK_NOSIGPIPE) ? FNOSIGPIPE : 0);
577 	fp->f_type = DTYPE_SOCKET;
578 	fp->f_ops = &socketops;
579 	error = socreate(domain, &so, type, protocol, l, NULL);
580 	if (error != 0) {
581 		fd_abort(curproc, fp, fd);
582 	} else {
583 		if (sop != NULL)
584 			*sop = so;
585 		fp->f_data = so;
586 		fd_affix(curproc, fp, fd);
587 		*fdout = fd;
588 	}
589 	return error;
590 }
591 
592 int
593 sofamily(const struct socket *so)
594 {
595 	const struct protosw *pr;
596 	const struct domain *dom;
597 
598 	if ((pr = so->so_proto) == NULL)
599 		return AF_UNSPEC;
600 	if ((dom = pr->pr_domain) == NULL)
601 		return AF_UNSPEC;
602 	return dom->dom_family;
603 }
604 
605 int
606 sobind(struct socket *so, struct mbuf *nam, struct lwp *l)
607 {
608 	int	error;
609 
610 	solock(so);
611 	error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, NULL, nam, NULL, l);
612 	sounlock(so);
613 	return error;
614 }
615 
616 int
617 solisten(struct socket *so, int backlog, struct lwp *l)
618 {
619 	int	error;
620 
621 	solock(so);
622 	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
623 	    SS_ISDISCONNECTING)) != 0) {
624 	    	sounlock(so);
625 		return (EINVAL);
626 	}
627 	error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, NULL,
628 	    NULL, NULL, l);
629 	if (error != 0) {
630 		sounlock(so);
631 		return error;
632 	}
633 	if (TAILQ_EMPTY(&so->so_q))
634 		so->so_options |= SO_ACCEPTCONN;
635 	if (backlog < 0)
636 		backlog = 0;
637 	so->so_qlimit = min(backlog, somaxconn);
638 	sounlock(so);
639 	return 0;
640 }
641 
642 void
643 sofree(struct socket *so)
644 {
645 	u_int refs;
646 
647 	KASSERT(solocked(so));
648 
649 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
650 		sounlock(so);
651 		return;
652 	}
653 	if (so->so_head) {
654 		/*
655 		 * We must not decommission a socket that's on the accept(2)
656 		 * queue.  If we do, then accept(2) may hang after select(2)
657 		 * indicated that the listening socket was ready.
658 		 */
659 		if (!soqremque(so, 0)) {
660 			sounlock(so);
661 			return;
662 		}
663 	}
664 	if (so->so_rcv.sb_hiwat)
665 		(void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0,
666 		    RLIM_INFINITY);
667 	if (so->so_snd.sb_hiwat)
668 		(void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0,
669 		    RLIM_INFINITY);
670 	sbrelease(&so->so_snd, so);
671 	KASSERT(!cv_has_waiters(&so->so_cv));
672 	KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv));
673 	KASSERT(!cv_has_waiters(&so->so_snd.sb_cv));
674 	sorflush(so);
675 	refs = so->so_aborting;	/* XXX */
676 	/* Remove acccept filter if one is present. */
677 	if (so->so_accf != NULL)
678 		(void)accept_filt_clear(so);
679 	sounlock(so);
680 	if (refs == 0)		/* XXX */
681 		soput(so);
682 }
683 
684 /*
685  * Close a socket on last file table reference removal.
686  * Initiate disconnect if connected.
687  * Free socket when disconnect complete.
688  */
689 int
690 soclose(struct socket *so)
691 {
692 	struct socket	*so2;
693 	int		error;
694 	int		error2;
695 
696 	error = 0;
697 	solock(so);
698 	if (so->so_options & SO_ACCEPTCONN) {
699 		for (;;) {
700 			if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) {
701 				KASSERT(solocked2(so, so2));
702 				(void) soqremque(so2, 0);
703 				/* soabort drops the lock. */
704 				(void) soabort(so2);
705 				solock(so);
706 				continue;
707 			}
708 			if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) {
709 				KASSERT(solocked2(so, so2));
710 				(void) soqremque(so2, 1);
711 				/* soabort drops the lock. */
712 				(void) soabort(so2);
713 				solock(so);
714 				continue;
715 			}
716 			break;
717 		}
718 	}
719 	if (so->so_pcb == 0)
720 		goto discard;
721 	if (so->so_state & SS_ISCONNECTED) {
722 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
723 			error = sodisconnect(so);
724 			if (error)
725 				goto drop;
726 		}
727 		if (so->so_options & SO_LINGER) {
728 			if ((so->so_state & (SS_ISDISCONNECTING|SS_NBIO)) ==
729 			    (SS_ISDISCONNECTING|SS_NBIO))
730 				goto drop;
731 			while (so->so_state & SS_ISCONNECTED) {
732 				error = sowait(so, true, so->so_linger * hz);
733 				if (error)
734 					break;
735 			}
736 		}
737 	}
738  drop:
739 	if (so->so_pcb) {
740 		error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH,
741 		    NULL, NULL, NULL, NULL);
742 		if (error == 0)
743 			error = error2;
744 	}
745  discard:
746 	if (so->so_state & SS_NOFDREF)
747 		panic("soclose: NOFDREF");
748 	kauth_cred_free(so->so_cred);
749 	so->so_state |= SS_NOFDREF;
750 	sofree(so);
751 	return (error);
752 }
753 
754 /*
755  * Must be called with the socket locked..  Will return with it unlocked.
756  */
757 int
758 soabort(struct socket *so)
759 {
760 	u_int refs;
761 	int error;
762 
763 	KASSERT(solocked(so));
764 	KASSERT(so->so_head == NULL);
765 
766 	so->so_aborting++;		/* XXX */
767 	error = (*so->so_proto->pr_usrreq)(so, PRU_ABORT, NULL,
768 	    NULL, NULL, NULL);
769 	refs = --so->so_aborting;	/* XXX */
770 	if (error || (refs == 0)) {
771 		sofree(so);
772 	} else {
773 		sounlock(so);
774 	}
775 	return error;
776 }
777 
778 int
779 soaccept(struct socket *so, struct mbuf *nam)
780 {
781 	int	error;
782 
783 	KASSERT(solocked(so));
784 
785 	error = 0;
786 	if ((so->so_state & SS_NOFDREF) == 0)
787 		panic("soaccept: !NOFDREF");
788 	so->so_state &= ~SS_NOFDREF;
789 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
790 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
791 		error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT,
792 		    NULL, nam, NULL, NULL);
793 	else
794 		error = ECONNABORTED;
795 
796 	return (error);
797 }
798 
799 int
800 soconnect(struct socket *so, struct mbuf *nam, struct lwp *l)
801 {
802 	int		error;
803 
804 	KASSERT(solocked(so));
805 
806 	if (so->so_options & SO_ACCEPTCONN)
807 		return (EOPNOTSUPP);
808 	/*
809 	 * If protocol is connection-based, can only connect once.
810 	 * Otherwise, if connected, try to disconnect first.
811 	 * This allows user to disconnect by connecting to, e.g.,
812 	 * a null address.
813 	 */
814 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
815 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
816 	    (error = sodisconnect(so))))
817 		error = EISCONN;
818 	else
819 		error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT,
820 		    NULL, nam, NULL, l);
821 	return (error);
822 }
823 
824 int
825 soconnect2(struct socket *so1, struct socket *so2)
826 {
827 	int	error;
828 
829 	KASSERT(solocked2(so1, so2));
830 
831 	error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2,
832 	    NULL, (struct mbuf *)so2, NULL, NULL);
833 	return (error);
834 }
835 
836 int
837 sodisconnect(struct socket *so)
838 {
839 	int	error;
840 
841 	KASSERT(solocked(so));
842 
843 	if ((so->so_state & SS_ISCONNECTED) == 0) {
844 		error = ENOTCONN;
845 	} else if (so->so_state & SS_ISDISCONNECTING) {
846 		error = EALREADY;
847 	} else {
848 		error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT,
849 		    NULL, NULL, NULL, NULL);
850 	}
851 	return (error);
852 }
853 
854 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
855 /*
856  * Send on a socket.
857  * If send must go all at once and message is larger than
858  * send buffering, then hard error.
859  * Lock against other senders.
860  * If must go all at once and not enough room now, then
861  * inform user that this would block and do nothing.
862  * Otherwise, if nonblocking, send as much as possible.
863  * The data to be sent is described by "uio" if nonzero,
864  * otherwise by the mbuf chain "top" (which must be null
865  * if uio is not).  Data provided in mbuf chain must be small
866  * enough to send all at once.
867  *
868  * Returns nonzero on error, timeout or signal; callers
869  * must check for short counts if EINTR/ERESTART are returned.
870  * Data and control buffers are freed on return.
871  */
872 int
873 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
874 	struct mbuf *control, int flags, struct lwp *l)
875 {
876 	struct mbuf	**mp, *m;
877 	struct proc	*p;
878 	long		space, len, resid, clen, mlen;
879 	int		error, s, dontroute, atomic;
880 	short		wakeup_state = 0;
881 
882 	p = l->l_proc;
883 	clen = 0;
884 
885 	/*
886 	 * solock() provides atomicity of access.  splsoftnet() prevents
887 	 * protocol processing soft interrupts from interrupting us and
888 	 * blocking (expensive).
889 	 */
890 	s = splsoftnet();
891 	solock(so);
892 	atomic = sosendallatonce(so) || top;
893 	if (uio)
894 		resid = uio->uio_resid;
895 	else
896 		resid = top->m_pkthdr.len;
897 	/*
898 	 * In theory resid should be unsigned.
899 	 * However, space must be signed, as it might be less than 0
900 	 * if we over-committed, and we must use a signed comparison
901 	 * of space and resid.  On the other hand, a negative resid
902 	 * causes us to loop sending 0-length segments to the protocol.
903 	 */
904 	if (resid < 0) {
905 		error = EINVAL;
906 		goto out;
907 	}
908 	dontroute =
909 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
910 	    (so->so_proto->pr_flags & PR_ATOMIC);
911 	l->l_ru.ru_msgsnd++;
912 	if (control)
913 		clen = control->m_len;
914  restart:
915 	if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
916 		goto out;
917 	do {
918 		if (so->so_state & SS_CANTSENDMORE) {
919 			error = EPIPE;
920 			goto release;
921 		}
922 		if (so->so_error) {
923 			error = so->so_error;
924 			so->so_error = 0;
925 			goto release;
926 		}
927 		if ((so->so_state & SS_ISCONNECTED) == 0) {
928 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
929 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
930 				    !(resid == 0 && 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 	if (so->so_state & SS_ISCONFIRMING && uio->uio_resid)
1188 		(*pr->pr_usrreq)(so, PRU_RCVD, NULL, NULL, NULL, l);
1189 
1190  restart:
1191 	if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) {
1192 		sounlock(so);
1193 		splx(s);
1194 		return error;
1195 	}
1196 
1197 	m = so->so_rcv.sb_mb;
1198 	/*
1199 	 * If we have less data than requested, block awaiting more
1200 	 * (subject to any timeout) if:
1201 	 *   1. the current count is less than the low water mark,
1202 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1203 	 *	receive operation at once if we block (resid <= hiwat), or
1204 	 *   3. MSG_DONTWAIT is not set.
1205 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1206 	 * we have to do the receive in sections, and thus risk returning
1207 	 * a short count if a timeout or signal occurs after we start.
1208 	 */
1209 	if (m == NULL ||
1210 	    ((flags & MSG_DONTWAIT) == 0 &&
1211 	     so->so_rcv.sb_cc < uio->uio_resid &&
1212 	     (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1213 	      ((flags & MSG_WAITALL) &&
1214 	       uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1215 	     m->m_nextpkt == NULL && !atomic)) {
1216 #ifdef DIAGNOSTIC
1217 		if (m == NULL && so->so_rcv.sb_cc)
1218 			panic("receive 1");
1219 #endif
1220 		if (so->so_error) {
1221 			if (m != NULL)
1222 				goto dontblock;
1223 			error = so->so_error;
1224 			if ((flags & MSG_PEEK) == 0)
1225 				so->so_error = 0;
1226 			goto release;
1227 		}
1228 		if (so->so_state & SS_CANTRCVMORE) {
1229 			if (m != NULL)
1230 				goto dontblock;
1231 			else
1232 				goto release;
1233 		}
1234 		for (; m != NULL; m = m->m_next)
1235 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1236 				m = so->so_rcv.sb_mb;
1237 				goto dontblock;
1238 			}
1239 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1240 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1241 			error = ENOTCONN;
1242 			goto release;
1243 		}
1244 		if (uio->uio_resid == 0)
1245 			goto release;
1246 		if ((so->so_state & SS_NBIO) ||
1247 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1248 			error = EWOULDBLOCK;
1249 			goto release;
1250 		}
1251 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
1252 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
1253 		sbunlock(&so->so_rcv);
1254 		if (wakeup_state & SS_RESTARTSYS)
1255 			error = ERESTART;
1256 		else
1257 			error = sbwait(&so->so_rcv);
1258 		if (error != 0) {
1259 			sounlock(so);
1260 			splx(s);
1261 			return error;
1262 		}
1263 		wakeup_state = so->so_state;
1264 		goto restart;
1265 	}
1266  dontblock:
1267 	/*
1268 	 * On entry here, m points to the first record of the socket buffer.
1269 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1270 	 * pointer to the next record in the socket buffer.  We must keep the
1271 	 * various socket buffer pointers and local stack versions of the
1272 	 * pointers in sync, pushing out modifications before dropping the
1273 	 * socket lock, and re-reading them when picking it up.
1274 	 *
1275 	 * Otherwise, we will race with the network stack appending new data
1276 	 * or records onto the socket buffer by using inconsistent/stale
1277 	 * versions of the field, possibly resulting in socket buffer
1278 	 * corruption.
1279 	 *
1280 	 * By holding the high-level sblock(), we prevent simultaneous
1281 	 * readers from pulling off the front of the socket buffer.
1282 	 */
1283 	if (l != NULL)
1284 		l->l_ru.ru_msgrcv++;
1285 	KASSERT(m == so->so_rcv.sb_mb);
1286 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
1287 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
1288 	nextrecord = m->m_nextpkt;
1289 	if (pr->pr_flags & PR_ADDR) {
1290 #ifdef DIAGNOSTIC
1291 		if (m->m_type != MT_SONAME)
1292 			panic("receive 1a");
1293 #endif
1294 		orig_resid = 0;
1295 		if (flags & MSG_PEEK) {
1296 			if (paddr)
1297 				*paddr = m_copy(m, 0, m->m_len);
1298 			m = m->m_next;
1299 		} else {
1300 			sbfree(&so->so_rcv, m);
1301 			mbuf_removed = 1;
1302 			if (paddr != NULL) {
1303 				*paddr = m;
1304 				so->so_rcv.sb_mb = m->m_next;
1305 				m->m_next = NULL;
1306 				m = so->so_rcv.sb_mb;
1307 			} else {
1308 				MFREE(m, so->so_rcv.sb_mb);
1309 				m = so->so_rcv.sb_mb;
1310 			}
1311 			sbsync(&so->so_rcv, nextrecord);
1312 		}
1313 	}
1314 
1315 	/*
1316 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1317 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1318 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1319 	 * perform externalization (or freeing if controlp == NULL).
1320 	 */
1321 	if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) {
1322 		struct mbuf *cm = NULL, *cmn;
1323 		struct mbuf **cme = &cm;
1324 
1325 		do {
1326 			if (flags & MSG_PEEK) {
1327 				if (controlp != NULL) {
1328 					*controlp = m_copy(m, 0, m->m_len);
1329 					controlp = &(*controlp)->m_next;
1330 				}
1331 				m = m->m_next;
1332 			} else {
1333 				sbfree(&so->so_rcv, m);
1334 				so->so_rcv.sb_mb = m->m_next;
1335 				m->m_next = NULL;
1336 				*cme = m;
1337 				cme = &(*cme)->m_next;
1338 				m = so->so_rcv.sb_mb;
1339 			}
1340 		} while (m != NULL && m->m_type == MT_CONTROL);
1341 		if ((flags & MSG_PEEK) == 0)
1342 			sbsync(&so->so_rcv, nextrecord);
1343 		for (; cm != NULL; cm = cmn) {
1344 			cmn = cm->m_next;
1345 			cm->m_next = NULL;
1346 			type = mtod(cm, struct cmsghdr *)->cmsg_type;
1347 			if (controlp != NULL) {
1348 				if (dom->dom_externalize != NULL &&
1349 				    type == SCM_RIGHTS) {
1350 					sounlock(so);
1351 					splx(s);
1352 					error = (*dom->dom_externalize)(cm, l,
1353 					    (flags & MSG_CMSG_CLOEXEC) ?
1354 					    O_CLOEXEC : 0);
1355 					s = splsoftnet();
1356 					solock(so);
1357 				}
1358 				*controlp = cm;
1359 				while (*controlp != NULL)
1360 					controlp = &(*controlp)->m_next;
1361 			} else {
1362 				/*
1363 				 * Dispose of any SCM_RIGHTS message that went
1364 				 * through the read path rather than recv.
1365 				 */
1366 				if (dom->dom_dispose != NULL &&
1367 				    type == SCM_RIGHTS) {
1368 				    	sounlock(so);
1369 					(*dom->dom_dispose)(cm);
1370 					solock(so);
1371 				}
1372 				m_freem(cm);
1373 			}
1374 		}
1375 		if (m != NULL)
1376 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1377 		else
1378 			nextrecord = so->so_rcv.sb_mb;
1379 		orig_resid = 0;
1380 	}
1381 
1382 	/* If m is non-NULL, we have some data to read. */
1383 	if (__predict_true(m != NULL)) {
1384 		type = m->m_type;
1385 		if (type == MT_OOBDATA)
1386 			flags |= MSG_OOB;
1387 	}
1388 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1389 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1390 
1391 	moff = 0;
1392 	offset = 0;
1393 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1394 		if (m->m_type == MT_OOBDATA) {
1395 			if (type != MT_OOBDATA)
1396 				break;
1397 		} else if (type == MT_OOBDATA)
1398 			break;
1399 #ifdef DIAGNOSTIC
1400 		else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
1401 			panic("receive 3");
1402 #endif
1403 		so->so_state &= ~SS_RCVATMARK;
1404 		wakeup_state = 0;
1405 		len = uio->uio_resid;
1406 		if (so->so_oobmark && len > so->so_oobmark - offset)
1407 			len = so->so_oobmark - offset;
1408 		if (len > m->m_len - moff)
1409 			len = m->m_len - moff;
1410 		/*
1411 		 * If mp is set, just pass back the mbufs.
1412 		 * Otherwise copy them out via the uio, then free.
1413 		 * Sockbuf must be consistent here (points to current mbuf,
1414 		 * it points to next record) when we drop priority;
1415 		 * we must note any additions to the sockbuf when we
1416 		 * block interrupts again.
1417 		 */
1418 		if (mp == NULL) {
1419 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1420 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1421 			sounlock(so);
1422 			splx(s);
1423 			error = uiomove(mtod(m, char *) + moff, len, uio);
1424 			s = splsoftnet();
1425 			solock(so);
1426 			if (error != 0) {
1427 				/*
1428 				 * If any part of the record has been removed
1429 				 * (such as the MT_SONAME mbuf, which will
1430 				 * happen when PR_ADDR, and thus also
1431 				 * PR_ATOMIC, is set), then drop the entire
1432 				 * record to maintain the atomicity of the
1433 				 * receive operation.
1434 				 *
1435 				 * This avoids a later panic("receive 1a")
1436 				 * when compiled with DIAGNOSTIC.
1437 				 */
1438 				if (m && mbuf_removed && atomic)
1439 					(void) sbdroprecord(&so->so_rcv);
1440 
1441 				goto release;
1442 			}
1443 		} else
1444 			uio->uio_resid -= len;
1445 		if (len == m->m_len - moff) {
1446 			if (m->m_flags & M_EOR)
1447 				flags |= MSG_EOR;
1448 			if (flags & MSG_PEEK) {
1449 				m = m->m_next;
1450 				moff = 0;
1451 			} else {
1452 				nextrecord = m->m_nextpkt;
1453 				sbfree(&so->so_rcv, m);
1454 				if (mp) {
1455 					*mp = m;
1456 					mp = &m->m_next;
1457 					so->so_rcv.sb_mb = m = m->m_next;
1458 					*mp = NULL;
1459 				} else {
1460 					MFREE(m, so->so_rcv.sb_mb);
1461 					m = so->so_rcv.sb_mb;
1462 				}
1463 				/*
1464 				 * If m != NULL, we also know that
1465 				 * so->so_rcv.sb_mb != NULL.
1466 				 */
1467 				KASSERT(so->so_rcv.sb_mb == m);
1468 				if (m) {
1469 					m->m_nextpkt = nextrecord;
1470 					if (nextrecord == NULL)
1471 						so->so_rcv.sb_lastrecord = m;
1472 				} else {
1473 					so->so_rcv.sb_mb = nextrecord;
1474 					SB_EMPTY_FIXUP(&so->so_rcv);
1475 				}
1476 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1477 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1478 			}
1479 		} else if (flags & MSG_PEEK)
1480 			moff += len;
1481 		else {
1482 			if (mp != NULL) {
1483 				mt = m_copym(m, 0, len, M_NOWAIT);
1484 				if (__predict_false(mt == NULL)) {
1485 					sounlock(so);
1486 					mt = m_copym(m, 0, len, M_WAIT);
1487 					solock(so);
1488 				}
1489 				*mp = mt;
1490 			}
1491 			m->m_data += len;
1492 			m->m_len -= len;
1493 			so->so_rcv.sb_cc -= len;
1494 		}
1495 		if (so->so_oobmark) {
1496 			if ((flags & MSG_PEEK) == 0) {
1497 				so->so_oobmark -= len;
1498 				if (so->so_oobmark == 0) {
1499 					so->so_state |= SS_RCVATMARK;
1500 					break;
1501 				}
1502 			} else {
1503 				offset += len;
1504 				if (offset == so->so_oobmark)
1505 					break;
1506 			}
1507 		}
1508 		if (flags & MSG_EOR)
1509 			break;
1510 		/*
1511 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1512 		 * we must not quit until "uio->uio_resid == 0" or an error
1513 		 * termination.  If a signal/timeout occurs, return
1514 		 * with a short count but without error.
1515 		 * Keep sockbuf locked against other readers.
1516 		 */
1517 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1518 		    !sosendallatonce(so) && !nextrecord) {
1519 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1520 				break;
1521 			/*
1522 			 * If we are peeking and the socket receive buffer is
1523 			 * full, stop since we can't get more data to peek at.
1524 			 */
1525 			if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
1526 				break;
1527 			/*
1528 			 * If we've drained the socket buffer, tell the
1529 			 * protocol in case it needs to do something to
1530 			 * get it filled again.
1531 			 */
1532 			if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1533 				(*pr->pr_usrreq)(so, PRU_RCVD,
1534 				    NULL, (struct mbuf *)(long)flags, NULL, l);
1535 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1536 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1537 			if (wakeup_state & SS_RESTARTSYS)
1538 				error = ERESTART;
1539 			else
1540 				error = sbwait(&so->so_rcv);
1541 			if (error != 0) {
1542 				sbunlock(&so->so_rcv);
1543 				sounlock(so);
1544 				splx(s);
1545 				return 0;
1546 			}
1547 			if ((m = so->so_rcv.sb_mb) != NULL)
1548 				nextrecord = m->m_nextpkt;
1549 			wakeup_state = so->so_state;
1550 		}
1551 	}
1552 
1553 	if (m && atomic) {
1554 		flags |= MSG_TRUNC;
1555 		if ((flags & MSG_PEEK) == 0)
1556 			(void) sbdroprecord(&so->so_rcv);
1557 	}
1558 	if ((flags & MSG_PEEK) == 0) {
1559 		if (m == NULL) {
1560 			/*
1561 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1562 			 * part makes sure sb_lastrecord is up-to-date if
1563 			 * there is still data in the socket buffer.
1564 			 */
1565 			so->so_rcv.sb_mb = nextrecord;
1566 			if (so->so_rcv.sb_mb == NULL) {
1567 				so->so_rcv.sb_mbtail = NULL;
1568 				so->so_rcv.sb_lastrecord = NULL;
1569 			} else if (nextrecord->m_nextpkt == NULL)
1570 				so->so_rcv.sb_lastrecord = nextrecord;
1571 		}
1572 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1573 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1574 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1575 			(*pr->pr_usrreq)(so, PRU_RCVD, NULL,
1576 			    (struct mbuf *)(long)flags, NULL, l);
1577 	}
1578 	if (orig_resid == uio->uio_resid && orig_resid &&
1579 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1580 		sbunlock(&so->so_rcv);
1581 		goto restart;
1582 	}
1583 
1584 	if (flagsp != NULL)
1585 		*flagsp |= flags;
1586  release:
1587 	sbunlock(&so->so_rcv);
1588 	sounlock(so);
1589 	splx(s);
1590 	return error;
1591 }
1592 
1593 int
1594 soshutdown(struct socket *so, int how)
1595 {
1596 	const struct protosw	*pr;
1597 	int	error;
1598 
1599 	KASSERT(solocked(so));
1600 
1601 	pr = so->so_proto;
1602 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1603 		return (EINVAL);
1604 
1605 	if (how == SHUT_RD || how == SHUT_RDWR) {
1606 		sorflush(so);
1607 		error = 0;
1608 	}
1609 	if (how == SHUT_WR || how == SHUT_RDWR)
1610 		error = (*pr->pr_usrreq)(so, PRU_SHUTDOWN, NULL,
1611 		    NULL, NULL, NULL);
1612 
1613 	return error;
1614 }
1615 
1616 void
1617 sorestart(struct socket *so)
1618 {
1619 	/*
1620 	 * An application has called close() on an fd on which another
1621 	 * of its threads has called a socket system call.
1622 	 * Mark this and wake everyone up, and code that would block again
1623 	 * instead returns ERESTART.
1624 	 * On system call re-entry the fd is validated and EBADF returned.
1625 	 * Any other fd will block again on the 2nd syscall.
1626 	 */
1627 	solock(so);
1628 	so->so_state |= SS_RESTARTSYS;
1629 	cv_broadcast(&so->so_cv);
1630 	cv_broadcast(&so->so_snd.sb_cv);
1631 	cv_broadcast(&so->so_rcv.sb_cv);
1632 	sounlock(so);
1633 }
1634 
1635 void
1636 sorflush(struct socket *so)
1637 {
1638 	struct sockbuf	*sb, asb;
1639 	const struct protosw	*pr;
1640 
1641 	KASSERT(solocked(so));
1642 
1643 	sb = &so->so_rcv;
1644 	pr = so->so_proto;
1645 	socantrcvmore(so);
1646 	sb->sb_flags |= SB_NOINTR;
1647 	(void )sblock(sb, M_WAITOK);
1648 	sbunlock(sb);
1649 	asb = *sb;
1650 	/*
1651 	 * Clear most of the sockbuf structure, but leave some of the
1652 	 * fields valid.
1653 	 */
1654 	memset(&sb->sb_startzero, 0,
1655 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1656 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) {
1657 		sounlock(so);
1658 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1659 		solock(so);
1660 	}
1661 	sbrelease(&asb, so);
1662 }
1663 
1664 /*
1665  * internal set SOL_SOCKET options
1666  */
1667 static int
1668 sosetopt1(struct socket *so, const struct sockopt *sopt)
1669 {
1670 	int error = EINVAL, optval, opt;
1671 	struct linger l;
1672 	struct timeval tv;
1673 
1674 	switch ((opt = sopt->sopt_name)) {
1675 
1676 	case SO_ACCEPTFILTER:
1677 		error = accept_filt_setopt(so, sopt);
1678 		KASSERT(solocked(so));
1679 		break;
1680 
1681   	case SO_LINGER:
1682  		error = sockopt_get(sopt, &l, sizeof(l));
1683 		solock(so);
1684  		if (error)
1685  			break;
1686  		if (l.l_linger < 0 || l.l_linger > USHRT_MAX ||
1687  		    l.l_linger > (INT_MAX / hz)) {
1688 			error = EDOM;
1689 			break;
1690 		}
1691  		so->so_linger = l.l_linger;
1692  		if (l.l_onoff)
1693  			so->so_options |= SO_LINGER;
1694  		else
1695  			so->so_options &= ~SO_LINGER;
1696    		break;
1697 
1698 	case SO_DEBUG:
1699 	case SO_KEEPALIVE:
1700 	case SO_DONTROUTE:
1701 	case SO_USELOOPBACK:
1702 	case SO_BROADCAST:
1703 	case SO_REUSEADDR:
1704 	case SO_REUSEPORT:
1705 	case SO_OOBINLINE:
1706 	case SO_TIMESTAMP:
1707 	case SO_NOSIGPIPE:
1708 #ifdef SO_OTIMESTAMP
1709 	case SO_OTIMESTAMP:
1710 #endif
1711 		error = sockopt_getint(sopt, &optval);
1712 		solock(so);
1713 		if (error)
1714 			break;
1715 		if (optval)
1716 			so->so_options |= opt;
1717 		else
1718 			so->so_options &= ~opt;
1719 		break;
1720 
1721 	case SO_SNDBUF:
1722 	case SO_RCVBUF:
1723 	case SO_SNDLOWAT:
1724 	case SO_RCVLOWAT:
1725 		error = sockopt_getint(sopt, &optval);
1726 		solock(so);
1727 		if (error)
1728 			break;
1729 
1730 		/*
1731 		 * Values < 1 make no sense for any of these
1732 		 * options, so disallow them.
1733 		 */
1734 		if (optval < 1) {
1735 			error = EINVAL;
1736 			break;
1737 		}
1738 
1739 		switch (opt) {
1740 		case SO_SNDBUF:
1741 			if (sbreserve(&so->so_snd, (u_long)optval, so) == 0) {
1742 				error = ENOBUFS;
1743 				break;
1744 			}
1745 			so->so_snd.sb_flags &= ~SB_AUTOSIZE;
1746 			break;
1747 
1748 		case SO_RCVBUF:
1749 			if (sbreserve(&so->so_rcv, (u_long)optval, so) == 0) {
1750 				error = ENOBUFS;
1751 				break;
1752 			}
1753 			so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1754 			break;
1755 
1756 		/*
1757 		 * Make sure the low-water is never greater than
1758 		 * the high-water.
1759 		 */
1760 		case SO_SNDLOWAT:
1761 			if (optval > so->so_snd.sb_hiwat)
1762 				optval = so->so_snd.sb_hiwat;
1763 
1764 			so->so_snd.sb_lowat = optval;
1765 			break;
1766 
1767 		case SO_RCVLOWAT:
1768 			if (optval > so->so_rcv.sb_hiwat)
1769 				optval = so->so_rcv.sb_hiwat;
1770 
1771 			so->so_rcv.sb_lowat = optval;
1772 			break;
1773 		}
1774 		break;
1775 
1776 #ifdef COMPAT_50
1777 	case SO_OSNDTIMEO:
1778 	case SO_ORCVTIMEO: {
1779 		struct timeval50 otv;
1780 		error = sockopt_get(sopt, &otv, sizeof(otv));
1781 		if (error) {
1782 			solock(so);
1783 			break;
1784 		}
1785 		timeval50_to_timeval(&otv, &tv);
1786 		opt = opt == SO_OSNDTIMEO ? SO_SNDTIMEO : SO_RCVTIMEO;
1787 		error = 0;
1788 		/*FALLTHROUGH*/
1789 	}
1790 #endif /* COMPAT_50 */
1791 
1792 	case SO_SNDTIMEO:
1793 	case SO_RCVTIMEO:
1794 		if (error)
1795 			error = sockopt_get(sopt, &tv, sizeof(tv));
1796 		solock(so);
1797 		if (error)
1798 			break;
1799 
1800 		if (tv.tv_sec > (INT_MAX - tv.tv_usec / tick) / hz) {
1801 			error = EDOM;
1802 			break;
1803 		}
1804 
1805 		optval = tv.tv_sec * hz + tv.tv_usec / tick;
1806 		if (optval == 0 && tv.tv_usec != 0)
1807 			optval = 1;
1808 
1809 		switch (opt) {
1810 		case SO_SNDTIMEO:
1811 			so->so_snd.sb_timeo = optval;
1812 			break;
1813 		case SO_RCVTIMEO:
1814 			so->so_rcv.sb_timeo = optval;
1815 			break;
1816 		}
1817 		break;
1818 
1819 	default:
1820 		solock(so);
1821 		error = ENOPROTOOPT;
1822 		break;
1823 	}
1824 	KASSERT(solocked(so));
1825 	return error;
1826 }
1827 
1828 int
1829 sosetopt(struct socket *so, struct sockopt *sopt)
1830 {
1831 	int error, prerr;
1832 
1833 	if (sopt->sopt_level == SOL_SOCKET) {
1834 		error = sosetopt1(so, sopt);
1835 		KASSERT(solocked(so));
1836 	} else {
1837 		error = ENOPROTOOPT;
1838 		solock(so);
1839 	}
1840 
1841 	if ((error == 0 || error == ENOPROTOOPT) &&
1842 	    so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) {
1843 		/* give the protocol stack a shot */
1844 		prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, sopt);
1845 		if (prerr == 0)
1846 			error = 0;
1847 		else if (prerr != ENOPROTOOPT)
1848 			error = prerr;
1849 	}
1850 	sounlock(so);
1851 	return error;
1852 }
1853 
1854 /*
1855  * so_setsockopt() is a wrapper providing a sockopt structure for sosetopt()
1856  */
1857 int
1858 so_setsockopt(struct lwp *l, struct socket *so, int level, int name,
1859     const void *val, size_t valsize)
1860 {
1861 	struct sockopt sopt;
1862 	int error;
1863 
1864 	KASSERT(valsize == 0 || val != NULL);
1865 
1866 	sockopt_init(&sopt, level, name, valsize);
1867 	sockopt_set(&sopt, val, valsize);
1868 
1869 	error = sosetopt(so, &sopt);
1870 
1871 	sockopt_destroy(&sopt);
1872 
1873 	return error;
1874 }
1875 
1876 /*
1877  * internal get SOL_SOCKET options
1878  */
1879 static int
1880 sogetopt1(struct socket *so, struct sockopt *sopt)
1881 {
1882 	int error, optval, opt;
1883 	struct linger l;
1884 	struct timeval tv;
1885 
1886 	switch ((opt = sopt->sopt_name)) {
1887 
1888 	case SO_ACCEPTFILTER:
1889 		error = accept_filt_getopt(so, sopt);
1890 		break;
1891 
1892 	case SO_LINGER:
1893 		l.l_onoff = (so->so_options & SO_LINGER) ? 1 : 0;
1894 		l.l_linger = so->so_linger;
1895 
1896 		error = sockopt_set(sopt, &l, sizeof(l));
1897 		break;
1898 
1899 	case SO_USELOOPBACK:
1900 	case SO_DONTROUTE:
1901 	case SO_DEBUG:
1902 	case SO_KEEPALIVE:
1903 	case SO_REUSEADDR:
1904 	case SO_REUSEPORT:
1905 	case SO_BROADCAST:
1906 	case SO_OOBINLINE:
1907 	case SO_TIMESTAMP:
1908 	case SO_NOSIGPIPE:
1909 #ifdef SO_OTIMESTAMP
1910 	case SO_OTIMESTAMP:
1911 #endif
1912 		error = sockopt_setint(sopt, (so->so_options & opt) ? 1 : 0);
1913 		break;
1914 
1915 	case SO_TYPE:
1916 		error = sockopt_setint(sopt, so->so_type);
1917 		break;
1918 
1919 	case SO_ERROR:
1920 		error = sockopt_setint(sopt, so->so_error);
1921 		so->so_error = 0;
1922 		break;
1923 
1924 	case SO_SNDBUF:
1925 		error = sockopt_setint(sopt, so->so_snd.sb_hiwat);
1926 		break;
1927 
1928 	case SO_RCVBUF:
1929 		error = sockopt_setint(sopt, so->so_rcv.sb_hiwat);
1930 		break;
1931 
1932 	case SO_SNDLOWAT:
1933 		error = sockopt_setint(sopt, so->so_snd.sb_lowat);
1934 		break;
1935 
1936 	case SO_RCVLOWAT:
1937 		error = sockopt_setint(sopt, so->so_rcv.sb_lowat);
1938 		break;
1939 
1940 #ifdef COMPAT_50
1941 	case SO_OSNDTIMEO:
1942 	case SO_ORCVTIMEO: {
1943 		struct timeval50 otv;
1944 
1945 		optval = (opt == SO_OSNDTIMEO ?
1946 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1947 
1948 		otv.tv_sec = optval / hz;
1949 		otv.tv_usec = (optval % hz) * tick;
1950 
1951 		error = sockopt_set(sopt, &otv, sizeof(otv));
1952 		break;
1953 	}
1954 #endif /* COMPAT_50 */
1955 
1956 	case SO_SNDTIMEO:
1957 	case SO_RCVTIMEO:
1958 		optval = (opt == SO_SNDTIMEO ?
1959 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1960 
1961 		tv.tv_sec = optval / hz;
1962 		tv.tv_usec = (optval % hz) * tick;
1963 
1964 		error = sockopt_set(sopt, &tv, sizeof(tv));
1965 		break;
1966 
1967 	case SO_OVERFLOWED:
1968 		error = sockopt_setint(sopt, so->so_rcv.sb_overflowed);
1969 		break;
1970 
1971 	default:
1972 		error = ENOPROTOOPT;
1973 		break;
1974 	}
1975 
1976 	return (error);
1977 }
1978 
1979 int
1980 sogetopt(struct socket *so, struct sockopt *sopt)
1981 {
1982 	int		error;
1983 
1984 	solock(so);
1985 	if (sopt->sopt_level != SOL_SOCKET) {
1986 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1987 			error = ((*so->so_proto->pr_ctloutput)
1988 			    (PRCO_GETOPT, so, sopt));
1989 		} else
1990 			error = (ENOPROTOOPT);
1991 	} else {
1992 		error = sogetopt1(so, sopt);
1993 	}
1994 	sounlock(so);
1995 	return (error);
1996 }
1997 
1998 /*
1999  * alloc sockopt data buffer buffer
2000  *	- will be released at destroy
2001  */
2002 static int
2003 sockopt_alloc(struct sockopt *sopt, size_t len, km_flag_t kmflag)
2004 {
2005 
2006 	KASSERT(sopt->sopt_size == 0);
2007 
2008 	if (len > sizeof(sopt->sopt_buf)) {
2009 		sopt->sopt_data = kmem_zalloc(len, kmflag);
2010 		if (sopt->sopt_data == NULL)
2011 			return ENOMEM;
2012 	} else
2013 		sopt->sopt_data = sopt->sopt_buf;
2014 
2015 	sopt->sopt_size = len;
2016 	return 0;
2017 }
2018 
2019 /*
2020  * initialise sockopt storage
2021  *	- MAY sleep during allocation
2022  */
2023 void
2024 sockopt_init(struct sockopt *sopt, int level, int name, size_t size)
2025 {
2026 
2027 	memset(sopt, 0, sizeof(*sopt));
2028 
2029 	sopt->sopt_level = level;
2030 	sopt->sopt_name = name;
2031 	(void)sockopt_alloc(sopt, size, KM_SLEEP);
2032 }
2033 
2034 /*
2035  * destroy sockopt storage
2036  *	- will release any held memory references
2037  */
2038 void
2039 sockopt_destroy(struct sockopt *sopt)
2040 {
2041 
2042 	if (sopt->sopt_data != sopt->sopt_buf)
2043 		kmem_free(sopt->sopt_data, sopt->sopt_size);
2044 
2045 	memset(sopt, 0, sizeof(*sopt));
2046 }
2047 
2048 /*
2049  * set sockopt value
2050  *	- value is copied into sockopt
2051  * 	- memory is allocated when necessary, will not sleep
2052  */
2053 int
2054 sockopt_set(struct sockopt *sopt, const void *buf, size_t len)
2055 {
2056 	int error;
2057 
2058 	if (sopt->sopt_size == 0) {
2059 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2060 		if (error)
2061 			return error;
2062 	}
2063 
2064 	KASSERT(sopt->sopt_size == len);
2065 	memcpy(sopt->sopt_data, buf, len);
2066 	return 0;
2067 }
2068 
2069 /*
2070  * common case of set sockopt integer value
2071  */
2072 int
2073 sockopt_setint(struct sockopt *sopt, int val)
2074 {
2075 
2076 	return sockopt_set(sopt, &val, sizeof(int));
2077 }
2078 
2079 /*
2080  * get sockopt value
2081  *	- correct size must be given
2082  */
2083 int
2084 sockopt_get(const struct sockopt *sopt, void *buf, size_t len)
2085 {
2086 
2087 	if (sopt->sopt_size != len)
2088 		return EINVAL;
2089 
2090 	memcpy(buf, sopt->sopt_data, len);
2091 	return 0;
2092 }
2093 
2094 /*
2095  * common case of get sockopt integer value
2096  */
2097 int
2098 sockopt_getint(const struct sockopt *sopt, int *valp)
2099 {
2100 
2101 	return sockopt_get(sopt, valp, sizeof(int));
2102 }
2103 
2104 /*
2105  * set sockopt value from mbuf
2106  *	- ONLY for legacy code
2107  *	- mbuf is released by sockopt
2108  *	- will not sleep
2109  */
2110 int
2111 sockopt_setmbuf(struct sockopt *sopt, struct mbuf *m)
2112 {
2113 	size_t len;
2114 	int error;
2115 
2116 	len = m_length(m);
2117 
2118 	if (sopt->sopt_size == 0) {
2119 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2120 		if (error)
2121 			return error;
2122 	}
2123 
2124 	KASSERT(sopt->sopt_size == len);
2125 	m_copydata(m, 0, len, sopt->sopt_data);
2126 	m_freem(m);
2127 
2128 	return 0;
2129 }
2130 
2131 /*
2132  * get sockopt value into mbuf
2133  *	- ONLY for legacy code
2134  *	- mbuf to be released by the caller
2135  *	- will not sleep
2136  */
2137 struct mbuf *
2138 sockopt_getmbuf(const struct sockopt *sopt)
2139 {
2140 	struct mbuf *m;
2141 
2142 	if (sopt->sopt_size > MCLBYTES)
2143 		return NULL;
2144 
2145 	m = m_get(M_DONTWAIT, MT_SOOPTS);
2146 	if (m == NULL)
2147 		return NULL;
2148 
2149 	if (sopt->sopt_size > MLEN) {
2150 		MCLGET(m, M_DONTWAIT);
2151 		if ((m->m_flags & M_EXT) == 0) {
2152 			m_free(m);
2153 			return NULL;
2154 		}
2155 	}
2156 
2157 	memcpy(mtod(m, void *), sopt->sopt_data, sopt->sopt_size);
2158 	m->m_len = sopt->sopt_size;
2159 
2160 	return m;
2161 }
2162 
2163 void
2164 sohasoutofband(struct socket *so)
2165 {
2166 
2167 	fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
2168 	selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, NOTE_SUBMIT);
2169 }
2170 
2171 static void
2172 filt_sordetach(struct knote *kn)
2173 {
2174 	struct socket	*so;
2175 
2176 	so = ((file_t *)kn->kn_obj)->f_data;
2177 	solock(so);
2178 	SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
2179 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
2180 		so->so_rcv.sb_flags &= ~SB_KNOTE;
2181 	sounlock(so);
2182 }
2183 
2184 /*ARGSUSED*/
2185 static int
2186 filt_soread(struct knote *kn, long hint)
2187 {
2188 	struct socket	*so;
2189 	int rv;
2190 
2191 	so = ((file_t *)kn->kn_obj)->f_data;
2192 	if (hint != NOTE_SUBMIT)
2193 		solock(so);
2194 	kn->kn_data = so->so_rcv.sb_cc;
2195 	if (so->so_state & SS_CANTRCVMORE) {
2196 		kn->kn_flags |= EV_EOF;
2197 		kn->kn_fflags = so->so_error;
2198 		rv = 1;
2199 	} else if (so->so_error)	/* temporary udp error */
2200 		rv = 1;
2201 	else if (kn->kn_sfflags & NOTE_LOWAT)
2202 		rv = (kn->kn_data >= kn->kn_sdata);
2203 	else
2204 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
2205 	if (hint != NOTE_SUBMIT)
2206 		sounlock(so);
2207 	return rv;
2208 }
2209 
2210 static void
2211 filt_sowdetach(struct knote *kn)
2212 {
2213 	struct socket	*so;
2214 
2215 	so = ((file_t *)kn->kn_obj)->f_data;
2216 	solock(so);
2217 	SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
2218 	if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
2219 		so->so_snd.sb_flags &= ~SB_KNOTE;
2220 	sounlock(so);
2221 }
2222 
2223 /*ARGSUSED*/
2224 static int
2225 filt_sowrite(struct knote *kn, long hint)
2226 {
2227 	struct socket	*so;
2228 	int rv;
2229 
2230 	so = ((file_t *)kn->kn_obj)->f_data;
2231 	if (hint != NOTE_SUBMIT)
2232 		solock(so);
2233 	kn->kn_data = sbspace(&so->so_snd);
2234 	if (so->so_state & SS_CANTSENDMORE) {
2235 		kn->kn_flags |= EV_EOF;
2236 		kn->kn_fflags = so->so_error;
2237 		rv = 1;
2238 	} else if (so->so_error)	/* temporary udp error */
2239 		rv = 1;
2240 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2241 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
2242 		rv = 0;
2243 	else if (kn->kn_sfflags & NOTE_LOWAT)
2244 		rv = (kn->kn_data >= kn->kn_sdata);
2245 	else
2246 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
2247 	if (hint != NOTE_SUBMIT)
2248 		sounlock(so);
2249 	return rv;
2250 }
2251 
2252 /*ARGSUSED*/
2253 static int
2254 filt_solisten(struct knote *kn, long hint)
2255 {
2256 	struct socket	*so;
2257 	int rv;
2258 
2259 	so = ((file_t *)kn->kn_obj)->f_data;
2260 
2261 	/*
2262 	 * Set kn_data to number of incoming connections, not
2263 	 * counting partial (incomplete) connections.
2264 	 */
2265 	if (hint != NOTE_SUBMIT)
2266 		solock(so);
2267 	kn->kn_data = so->so_qlen;
2268 	rv = (kn->kn_data > 0);
2269 	if (hint != NOTE_SUBMIT)
2270 		sounlock(so);
2271 	return rv;
2272 }
2273 
2274 static const struct filterops solisten_filtops =
2275 	{ 1, NULL, filt_sordetach, filt_solisten };
2276 static const struct filterops soread_filtops =
2277 	{ 1, NULL, filt_sordetach, filt_soread };
2278 static const struct filterops sowrite_filtops =
2279 	{ 1, NULL, filt_sowdetach, filt_sowrite };
2280 
2281 int
2282 soo_kqfilter(struct file *fp, struct knote *kn)
2283 {
2284 	struct socket	*so;
2285 	struct sockbuf	*sb;
2286 
2287 	so = ((file_t *)kn->kn_obj)->f_data;
2288 	solock(so);
2289 	switch (kn->kn_filter) {
2290 	case EVFILT_READ:
2291 		if (so->so_options & SO_ACCEPTCONN)
2292 			kn->kn_fop = &solisten_filtops;
2293 		else
2294 			kn->kn_fop = &soread_filtops;
2295 		sb = &so->so_rcv;
2296 		break;
2297 	case EVFILT_WRITE:
2298 		kn->kn_fop = &sowrite_filtops;
2299 		sb = &so->so_snd;
2300 		break;
2301 	default:
2302 		sounlock(so);
2303 		return (EINVAL);
2304 	}
2305 	SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
2306 	sb->sb_flags |= SB_KNOTE;
2307 	sounlock(so);
2308 	return (0);
2309 }
2310 
2311 static int
2312 sodopoll(struct socket *so, int events)
2313 {
2314 	int revents;
2315 
2316 	revents = 0;
2317 
2318 	if (events & (POLLIN | POLLRDNORM))
2319 		if (soreadable(so))
2320 			revents |= events & (POLLIN | POLLRDNORM);
2321 
2322 	if (events & (POLLOUT | POLLWRNORM))
2323 		if (sowritable(so))
2324 			revents |= events & (POLLOUT | POLLWRNORM);
2325 
2326 	if (events & (POLLPRI | POLLRDBAND))
2327 		if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
2328 			revents |= events & (POLLPRI | POLLRDBAND);
2329 
2330 	return revents;
2331 }
2332 
2333 int
2334 sopoll(struct socket *so, int events)
2335 {
2336 	int revents = 0;
2337 
2338 #ifndef DIAGNOSTIC
2339 	/*
2340 	 * Do a quick, unlocked check in expectation that the socket
2341 	 * will be ready for I/O.  Don't do this check if DIAGNOSTIC,
2342 	 * as the solocked() assertions will fail.
2343 	 */
2344 	if ((revents = sodopoll(so, events)) != 0)
2345 		return revents;
2346 #endif
2347 
2348 	solock(so);
2349 	if ((revents = sodopoll(so, events)) == 0) {
2350 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2351 			selrecord(curlwp, &so->so_rcv.sb_sel);
2352 			so->so_rcv.sb_flags |= SB_NOTIFY;
2353 		}
2354 
2355 		if (events & (POLLOUT | POLLWRNORM)) {
2356 			selrecord(curlwp, &so->so_snd.sb_sel);
2357 			so->so_snd.sb_flags |= SB_NOTIFY;
2358 		}
2359 	}
2360 	sounlock(so);
2361 
2362 	return revents;
2363 }
2364 
2365 
2366 #include <sys/sysctl.h>
2367 
2368 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO);
2369 static int sysctl_kern_sbmax(SYSCTLFN_PROTO);
2370 
2371 /*
2372  * sysctl helper routine for kern.somaxkva.  ensures that the given
2373  * value is not too small.
2374  * (XXX should we maybe make sure it's not too large as well?)
2375  */
2376 static int
2377 sysctl_kern_somaxkva(SYSCTLFN_ARGS)
2378 {
2379 	int error, new_somaxkva;
2380 	struct sysctlnode node;
2381 
2382 	new_somaxkva = somaxkva;
2383 	node = *rnode;
2384 	node.sysctl_data = &new_somaxkva;
2385 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2386 	if (error || newp == NULL)
2387 		return (error);
2388 
2389 	if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */
2390 		return (EINVAL);
2391 
2392 	mutex_enter(&so_pendfree_lock);
2393 	somaxkva = new_somaxkva;
2394 	cv_broadcast(&socurkva_cv);
2395 	mutex_exit(&so_pendfree_lock);
2396 
2397 	return (error);
2398 }
2399 
2400 /*
2401  * sysctl helper routine for kern.sbmax. Basically just ensures that
2402  * any new value is not too small.
2403  */
2404 static int
2405 sysctl_kern_sbmax(SYSCTLFN_ARGS)
2406 {
2407 	int error, new_sbmax;
2408 	struct sysctlnode node;
2409 
2410 	new_sbmax = sb_max;
2411 	node = *rnode;
2412 	node.sysctl_data = &new_sbmax;
2413 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2414 	if (error || newp == NULL)
2415 		return (error);
2416 
2417 	KERNEL_LOCK(1, NULL);
2418 	error = sb_max_set(new_sbmax);
2419 	KERNEL_UNLOCK_ONE(NULL);
2420 
2421 	return (error);
2422 }
2423 
2424 static void
2425 sysctl_kern_socket_setup(void)
2426 {
2427 
2428 	KASSERT(socket_sysctllog == NULL);
2429 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2430 		       CTLFLAG_PERMANENT,
2431 		       CTLTYPE_NODE, "kern", NULL,
2432 		       NULL, 0, NULL, 0,
2433 		       CTL_KERN, CTL_EOL);
2434 
2435 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2436 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2437 		       CTLTYPE_INT, "somaxkva",
2438 		       SYSCTL_DESCR("Maximum amount of kernel memory to be "
2439 				    "used for socket buffers"),
2440 		       sysctl_kern_somaxkva, 0, NULL, 0,
2441 		       CTL_KERN, KERN_SOMAXKVA, CTL_EOL);
2442 
2443 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2444 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2445 		       CTLTYPE_INT, "sbmax",
2446 		       SYSCTL_DESCR("Maximum socket buffer size"),
2447 		       sysctl_kern_sbmax, 0, NULL, 0,
2448 		       CTL_KERN, KERN_SBMAX, CTL_EOL);
2449 }
2450