xref: /netbsd-src/sys/kern/uipc_socket.c (revision 466a16a118933bd295a8a104f095714fadf9cf68)
1 /*	$NetBSD: uipc_socket.c,v 1.178 2008/12/07 20:58:46 pooka Exp $	*/
2 
3 /*-
4  * Copyright (c) 2002, 2007, 2008 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.
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 1.178 2008/12/07 20:58:46 pooka Exp $");
67 
68 #include "opt_sock_counters.h"
69 #include "opt_sosend_loan.h"
70 #include "opt_mbuftrace.h"
71 #include "opt_somaxkva.h"
72 #include "opt_multiprocessor.h"	/* XXX */
73 
74 #include <sys/param.h>
75 #include <sys/systm.h>
76 #include <sys/proc.h>
77 #include <sys/file.h>
78 #include <sys/filedesc.h>
79 #include <sys/kmem.h>
80 #include <sys/mbuf.h>
81 #include <sys/domain.h>
82 #include <sys/kernel.h>
83 #include <sys/protosw.h>
84 #include <sys/socket.h>
85 #include <sys/socketvar.h>
86 #include <sys/signalvar.h>
87 #include <sys/resourcevar.h>
88 #include <sys/uidinfo.h>
89 #include <sys/event.h>
90 #include <sys/poll.h>
91 #include <sys/kauth.h>
92 #include <sys/mutex.h>
93 #include <sys/condvar.h>
94 
95 #include <uvm/uvm.h>
96 
97 MALLOC_DEFINE(M_SOOPTS, "soopts", "socket options");
98 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
99 
100 extern const struct fileops socketops;
101 
102 extern int	somaxconn;			/* patchable (XXX sysctl) */
103 int		somaxconn = SOMAXCONN;
104 kmutex_t	*softnet_lock;
105 
106 #ifdef SOSEND_COUNTERS
107 #include <sys/device.h>
108 
109 static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
110     NULL, "sosend", "loan big");
111 static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
112     NULL, "sosend", "copy big");
113 static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
114     NULL, "sosend", "copy small");
115 static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
116     NULL, "sosend", "kva limit");
117 
118 #define	SOSEND_COUNTER_INCR(ev)		(ev)->ev_count++
119 
120 EVCNT_ATTACH_STATIC(sosend_loan_big);
121 EVCNT_ATTACH_STATIC(sosend_copy_big);
122 EVCNT_ATTACH_STATIC(sosend_copy_small);
123 EVCNT_ATTACH_STATIC(sosend_kvalimit);
124 #else
125 
126 #define	SOSEND_COUNTER_INCR(ev)		/* nothing */
127 
128 #endif /* SOSEND_COUNTERS */
129 
130 static struct callback_entry sokva_reclaimerentry;
131 
132 #if defined(SOSEND_NO_LOAN) || defined(MULTIPROCESSOR)
133 int sock_loan_thresh = -1;
134 #else
135 int sock_loan_thresh = 4096;
136 #endif
137 
138 static kmutex_t so_pendfree_lock;
139 static struct mbuf *so_pendfree;
140 
141 #ifndef SOMAXKVA
142 #define	SOMAXKVA (16 * 1024 * 1024)
143 #endif
144 int somaxkva = SOMAXKVA;
145 static int socurkva;
146 static kcondvar_t socurkva_cv;
147 
148 #define	SOCK_LOAN_CHUNK		65536
149 
150 static size_t sodopendfree(void);
151 static size_t sodopendfreel(void);
152 
153 static void sysctl_kern_somaxkva_setup(void);
154 static struct sysctllog *socket_sysctllog;
155 
156 static vsize_t
157 sokvareserve(struct socket *so, vsize_t len)
158 {
159 	int error;
160 
161 	mutex_enter(&so_pendfree_lock);
162 	while (socurkva + len > somaxkva) {
163 		size_t freed;
164 
165 		/*
166 		 * try to do pendfree.
167 		 */
168 
169 		freed = sodopendfreel();
170 
171 		/*
172 		 * if some kva was freed, try again.
173 		 */
174 
175 		if (freed)
176 			continue;
177 
178 		SOSEND_COUNTER_INCR(&sosend_kvalimit);
179 		error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock);
180 		if (error) {
181 			len = 0;
182 			break;
183 		}
184 	}
185 	socurkva += len;
186 	mutex_exit(&so_pendfree_lock);
187 	return len;
188 }
189 
190 static void
191 sokvaunreserve(vsize_t len)
192 {
193 
194 	mutex_enter(&so_pendfree_lock);
195 	socurkva -= len;
196 	cv_broadcast(&socurkva_cv);
197 	mutex_exit(&so_pendfree_lock);
198 }
199 
200 /*
201  * sokvaalloc: allocate kva for loan.
202  */
203 
204 vaddr_t
205 sokvaalloc(vsize_t len, struct socket *so)
206 {
207 	vaddr_t lva;
208 
209 	/*
210 	 * reserve kva.
211 	 */
212 
213 	if (sokvareserve(so, len) == 0)
214 		return 0;
215 
216 	/*
217 	 * allocate kva.
218 	 */
219 
220 	lva = uvm_km_alloc(kernel_map, len, 0, UVM_KMF_VAONLY | UVM_KMF_WAITVA);
221 	if (lva == 0) {
222 		sokvaunreserve(len);
223 		return (0);
224 	}
225 
226 	return lva;
227 }
228 
229 /*
230  * sokvafree: free kva for loan.
231  */
232 
233 void
234 sokvafree(vaddr_t sva, vsize_t len)
235 {
236 
237 	/*
238 	 * free kva.
239 	 */
240 
241 	uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY);
242 
243 	/*
244 	 * unreserve kva.
245 	 */
246 
247 	sokvaunreserve(len);
248 }
249 
250 static void
251 sodoloanfree(struct vm_page **pgs, void *buf, size_t size)
252 {
253 	vaddr_t sva, eva;
254 	vsize_t len;
255 	int npgs;
256 
257 	KASSERT(pgs != NULL);
258 
259 	eva = round_page((vaddr_t) buf + size);
260 	sva = trunc_page((vaddr_t) buf);
261 	len = eva - sva;
262 	npgs = len >> PAGE_SHIFT;
263 
264 	pmap_kremove(sva, len);
265 	pmap_update(pmap_kernel());
266 	uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE);
267 	sokvafree(sva, len);
268 }
269 
270 static size_t
271 sodopendfree(void)
272 {
273 	size_t rv;
274 
275 	if (__predict_true(so_pendfree == NULL))
276 		return 0;
277 
278 	mutex_enter(&so_pendfree_lock);
279 	rv = sodopendfreel();
280 	mutex_exit(&so_pendfree_lock);
281 
282 	return rv;
283 }
284 
285 /*
286  * sodopendfreel: free mbufs on "pendfree" list.
287  * unlock and relock so_pendfree_lock when freeing mbufs.
288  *
289  * => called with so_pendfree_lock held.
290  */
291 
292 static size_t
293 sodopendfreel(void)
294 {
295 	struct mbuf *m, *next;
296 	size_t rv = 0;
297 
298 	KASSERT(mutex_owned(&so_pendfree_lock));
299 
300 	while (so_pendfree != NULL) {
301 		m = so_pendfree;
302 		so_pendfree = NULL;
303 		mutex_exit(&so_pendfree_lock);
304 
305 		for (; m != NULL; m = next) {
306 			next = m->m_next;
307 			KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0);
308 			KASSERT(m->m_ext.ext_refcnt == 0);
309 
310 			rv += m->m_ext.ext_size;
311 			sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf,
312 			    m->m_ext.ext_size);
313 			pool_cache_put(mb_cache, m);
314 		}
315 
316 		mutex_enter(&so_pendfree_lock);
317 	}
318 
319 	return (rv);
320 }
321 
322 void
323 soloanfree(struct mbuf *m, void *buf, size_t size, void *arg)
324 {
325 
326 	KASSERT(m != NULL);
327 
328 	/*
329 	 * postpone freeing mbuf.
330 	 *
331 	 * we can't do it in interrupt context
332 	 * because we need to put kva back to kernel_map.
333 	 */
334 
335 	mutex_enter(&so_pendfree_lock);
336 	m->m_next = so_pendfree;
337 	so_pendfree = m;
338 	cv_broadcast(&socurkva_cv);
339 	mutex_exit(&so_pendfree_lock);
340 }
341 
342 static long
343 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space)
344 {
345 	struct iovec *iov = uio->uio_iov;
346 	vaddr_t sva, eva;
347 	vsize_t len;
348 	vaddr_t lva;
349 	int npgs, error;
350 	vaddr_t va;
351 	int i;
352 
353 	if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace))
354 		return (0);
355 
356 	if (iov->iov_len < (size_t) space)
357 		space = iov->iov_len;
358 	if (space > SOCK_LOAN_CHUNK)
359 		space = SOCK_LOAN_CHUNK;
360 
361 	eva = round_page((vaddr_t) iov->iov_base + space);
362 	sva = trunc_page((vaddr_t) iov->iov_base);
363 	len = eva - sva;
364 	npgs = len >> PAGE_SHIFT;
365 
366 	KASSERT(npgs <= M_EXT_MAXPAGES);
367 
368 	lva = sokvaalloc(len, so);
369 	if (lva == 0)
370 		return 0;
371 
372 	error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len,
373 	    m->m_ext.ext_pgs, UVM_LOAN_TOPAGE);
374 	if (error) {
375 		sokvafree(lva, len);
376 		return (0);
377 	}
378 
379 	for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE)
380 		pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
381 		    VM_PROT_READ);
382 	pmap_update(pmap_kernel());
383 
384 	lva += (vaddr_t) iov->iov_base & PAGE_MASK;
385 
386 	MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so);
387 	m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
388 
389 	uio->uio_resid -= space;
390 	/* uio_offset not updated, not set/used for write(2) */
391 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space;
392 	uio->uio_iov->iov_len -= space;
393 	if (uio->uio_iov->iov_len == 0) {
394 		uio->uio_iov++;
395 		uio->uio_iovcnt--;
396 	}
397 
398 	return (space);
399 }
400 
401 static int
402 sokva_reclaim_callback(struct callback_entry *ce, void *obj, void *arg)
403 {
404 
405 	KASSERT(ce == &sokva_reclaimerentry);
406 	KASSERT(obj == NULL);
407 
408 	sodopendfree();
409 	if (!vm_map_starved_p(kernel_map)) {
410 		return CALLBACK_CHAIN_ABORT;
411 	}
412 	return CALLBACK_CHAIN_CONTINUE;
413 }
414 
415 struct mbuf *
416 getsombuf(struct socket *so, int type)
417 {
418 	struct mbuf *m;
419 
420 	m = m_get(M_WAIT, type);
421 	MCLAIM(m, so->so_mowner);
422 	return m;
423 }
424 
425 void
426 soinit(void)
427 {
428 
429 	sysctl_kern_somaxkva_setup();
430 
431 	mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM);
432 	softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
433 	cv_init(&socurkva_cv, "sokva");
434 	soinit2();
435 
436 	/* Set the initial adjusted socket buffer size. */
437 	if (sb_max_set(sb_max))
438 		panic("bad initial sb_max value: %lu", sb_max);
439 
440 	callback_register(&vm_map_to_kernel(kernel_map)->vmk_reclaim_callback,
441 	    &sokva_reclaimerentry, NULL, sokva_reclaim_callback);
442 }
443 
444 /*
445  * Socket operation routines.
446  * These routines are called by the routines in
447  * sys_socket.c or from a system process, and
448  * implement the semantics of socket operations by
449  * switching out to the protocol specific routines.
450  */
451 /*ARGSUSED*/
452 int
453 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l,
454 	 struct socket *lockso)
455 {
456 	const struct protosw	*prp;
457 	struct socket	*so;
458 	uid_t		uid;
459 	int		error;
460 	kmutex_t	*lock;
461 
462 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET,
463 	    KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type),
464 	    KAUTH_ARG(proto));
465 	if (error != 0)
466 		return error;
467 
468 	if (proto)
469 		prp = pffindproto(dom, proto, type);
470 	else
471 		prp = pffindtype(dom, type);
472 	if (prp == NULL) {
473 		/* no support for domain */
474 		if (pffinddomain(dom) == 0)
475 			return EAFNOSUPPORT;
476 		/* no support for socket type */
477 		if (proto == 0 && type != 0)
478 			return EPROTOTYPE;
479 		return EPROTONOSUPPORT;
480 	}
481 	if (prp->pr_usrreq == NULL)
482 		return EPROTONOSUPPORT;
483 	if (prp->pr_type != type)
484 		return EPROTOTYPE;
485 
486 	so = soget(true);
487 	so->so_type = type;
488 	so->so_proto = prp;
489 	so->so_send = sosend;
490 	so->so_receive = soreceive;
491 #ifdef MBUFTRACE
492 	so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner;
493 	so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner;
494 	so->so_mowner = &prp->pr_domain->dom_mowner;
495 #endif
496 	uid = kauth_cred_geteuid(l->l_cred);
497 	so->so_uidinfo = uid_find(uid);
498 	so->so_egid = kauth_cred_getegid(l->l_cred);
499 	so->so_cpid = l->l_proc->p_pid;
500 	if (lockso != NULL) {
501 		/* Caller wants us to share a lock. */
502 		lock = lockso->so_lock;
503 		so->so_lock = lock;
504 		mutex_obj_hold(lock);
505 		mutex_enter(lock);
506 	} else {
507 		/* Lock assigned and taken during PRU_ATTACH. */
508 	}
509 	error = (*prp->pr_usrreq)(so, PRU_ATTACH, NULL,
510 	    (struct mbuf *)(long)proto, NULL, l);
511 	KASSERT(solocked(so));
512 	if (error != 0) {
513 		so->so_state |= SS_NOFDREF;
514 		sofree(so);
515 		return error;
516 	}
517 	sounlock(so);
518 	*aso = so;
519 	return 0;
520 }
521 
522 /* On success, write file descriptor to fdout and return zero.  On
523  * failure, return non-zero; *fdout will be undefined.
524  */
525 int
526 fsocreate(int domain, struct socket **sop, int type, int protocol,
527     struct lwp *l, int *fdout)
528 {
529 	struct socket	*so;
530 	struct file	*fp;
531 	int		fd, error;
532 
533 	if ((error = fd_allocfile(&fp, &fd)) != 0)
534 		return (error);
535 	fp->f_flag = FREAD|FWRITE;
536 	fp->f_type = DTYPE_SOCKET;
537 	fp->f_ops = &socketops;
538 	error = socreate(domain, &so, type, protocol, l, NULL);
539 	if (error != 0) {
540 		fd_abort(curproc, fp, fd);
541 	} else {
542 		if (sop != NULL)
543 			*sop = so;
544 		fp->f_data = so;
545 		fd_affix(curproc, fp, fd);
546 		*fdout = fd;
547 	}
548 	return error;
549 }
550 
551 int
552 sobind(struct socket *so, struct mbuf *nam, struct lwp *l)
553 {
554 	int	error;
555 
556 	solock(so);
557 	error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, NULL, nam, NULL, l);
558 	sounlock(so);
559 	return error;
560 }
561 
562 int
563 solisten(struct socket *so, int backlog, struct lwp *l)
564 {
565 	int	error;
566 
567 	solock(so);
568 	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
569 	    SS_ISDISCONNECTING)) != 0) {
570 	    	sounlock(so);
571 		return (EOPNOTSUPP);
572 	}
573 	error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, NULL,
574 	    NULL, NULL, l);
575 	if (error != 0) {
576 		sounlock(so);
577 		return error;
578 	}
579 	if (TAILQ_EMPTY(&so->so_q))
580 		so->so_options |= SO_ACCEPTCONN;
581 	if (backlog < 0)
582 		backlog = 0;
583 	so->so_qlimit = min(backlog, somaxconn);
584 	sounlock(so);
585 	return 0;
586 }
587 
588 void
589 sofree(struct socket *so)
590 {
591 	u_int refs;
592 
593 	KASSERT(solocked(so));
594 
595 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
596 		sounlock(so);
597 		return;
598 	}
599 	if (so->so_head) {
600 		/*
601 		 * We must not decommission a socket that's on the accept(2)
602 		 * queue.  If we do, then accept(2) may hang after select(2)
603 		 * indicated that the listening socket was ready.
604 		 */
605 		if (!soqremque(so, 0)) {
606 			sounlock(so);
607 			return;
608 		}
609 	}
610 	if (so->so_rcv.sb_hiwat)
611 		(void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0,
612 		    RLIM_INFINITY);
613 	if (so->so_snd.sb_hiwat)
614 		(void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0,
615 		    RLIM_INFINITY);
616 	sbrelease(&so->so_snd, so);
617 	KASSERT(!cv_has_waiters(&so->so_cv));
618 	KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv));
619 	KASSERT(!cv_has_waiters(&so->so_snd.sb_cv));
620 	sorflush(so);
621 	refs = so->so_aborting;	/* XXX */
622 	/* Remove acccept filter if one is present. */
623 	if (so->so_accf != NULL)
624 		(void)accept_filt_clear(so);
625 	sounlock(so);
626 	if (refs == 0)		/* XXX */
627 		soput(so);
628 }
629 
630 /*
631  * Close a socket on last file table reference removal.
632  * Initiate disconnect if connected.
633  * Free socket when disconnect complete.
634  */
635 int
636 soclose(struct socket *so)
637 {
638 	struct socket	*so2;
639 	int		error;
640 	int		error2;
641 
642 	error = 0;
643 	solock(so);
644 	if (so->so_options & SO_ACCEPTCONN) {
645 		for (;;) {
646 			if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) {
647 				KASSERT(solocked2(so, so2));
648 				(void) soqremque(so2, 0);
649 				/* soabort drops the lock. */
650 				(void) soabort(so2);
651 				solock(so);
652 				continue;
653 			}
654 			if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) {
655 				KASSERT(solocked2(so, so2));
656 				(void) soqremque(so2, 1);
657 				/* soabort drops the lock. */
658 				(void) soabort(so2);
659 				solock(so);
660 				continue;
661 			}
662 			break;
663 		}
664 	}
665 	if (so->so_pcb == 0)
666 		goto discard;
667 	if (so->so_state & SS_ISCONNECTED) {
668 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
669 			error = sodisconnect(so);
670 			if (error)
671 				goto drop;
672 		}
673 		if (so->so_options & SO_LINGER) {
674 			if ((so->so_state & SS_ISDISCONNECTING) && so->so_nbio)
675 				goto drop;
676 			while (so->so_state & SS_ISCONNECTED) {
677 				error = sowait(so, so->so_linger * hz);
678 				if (error)
679 					break;
680 			}
681 		}
682 	}
683  drop:
684 	if (so->so_pcb) {
685 		error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH,
686 		    NULL, NULL, NULL, NULL);
687 		if (error == 0)
688 			error = error2;
689 	}
690  discard:
691 	if (so->so_state & SS_NOFDREF)
692 		panic("soclose: NOFDREF");
693 	so->so_state |= SS_NOFDREF;
694 	sofree(so);
695 	return (error);
696 }
697 
698 /*
699  * Must be called with the socket locked..  Will return with it unlocked.
700  */
701 int
702 soabort(struct socket *so)
703 {
704 	u_int refs;
705 	int error;
706 
707 	KASSERT(solocked(so));
708 	KASSERT(so->so_head == NULL);
709 
710 	so->so_aborting++;		/* XXX */
711 	error = (*so->so_proto->pr_usrreq)(so, PRU_ABORT, NULL,
712 	    NULL, NULL, NULL);
713 	refs = --so->so_aborting;	/* XXX */
714 	if (error || (refs == 0)) {
715 		sofree(so);
716 	} else {
717 		sounlock(so);
718 	}
719 	return error;
720 }
721 
722 int
723 soaccept(struct socket *so, struct mbuf *nam)
724 {
725 	int	error;
726 
727 	KASSERT(solocked(so));
728 
729 	error = 0;
730 	if ((so->so_state & SS_NOFDREF) == 0)
731 		panic("soaccept: !NOFDREF");
732 	so->so_state &= ~SS_NOFDREF;
733 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
734 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
735 		error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT,
736 		    NULL, nam, NULL, NULL);
737 	else
738 		error = ECONNABORTED;
739 
740 	return (error);
741 }
742 
743 int
744 soconnect(struct socket *so, struct mbuf *nam, struct lwp *l)
745 {
746 	int		error;
747 
748 	KASSERT(solocked(so));
749 
750 	if (so->so_options & SO_ACCEPTCONN)
751 		return (EOPNOTSUPP);
752 	/*
753 	 * If protocol is connection-based, can only connect once.
754 	 * Otherwise, if connected, try to disconnect first.
755 	 * This allows user to disconnect by connecting to, e.g.,
756 	 * a null address.
757 	 */
758 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
759 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
760 	    (error = sodisconnect(so))))
761 		error = EISCONN;
762 	else
763 		error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT,
764 		    NULL, nam, NULL, l);
765 	return (error);
766 }
767 
768 int
769 soconnect2(struct socket *so1, struct socket *so2)
770 {
771 	int	error;
772 
773 	KASSERT(solocked2(so1, so2));
774 
775 	error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2,
776 	    NULL, (struct mbuf *)so2, NULL, NULL);
777 	return (error);
778 }
779 
780 int
781 sodisconnect(struct socket *so)
782 {
783 	int	error;
784 
785 	KASSERT(solocked(so));
786 
787 	if ((so->so_state & SS_ISCONNECTED) == 0) {
788 		error = ENOTCONN;
789 	} else if (so->so_state & SS_ISDISCONNECTING) {
790 		error = EALREADY;
791 	} else {
792 		error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT,
793 		    NULL, NULL, NULL, NULL);
794 	}
795 	sodopendfree();
796 	return (error);
797 }
798 
799 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
800 /*
801  * Send on a socket.
802  * If send must go all at once and message is larger than
803  * send buffering, then hard error.
804  * Lock against other senders.
805  * If must go all at once and not enough room now, then
806  * inform user that this would block and do nothing.
807  * Otherwise, if nonblocking, send as much as possible.
808  * The data to be sent is described by "uio" if nonzero,
809  * otherwise by the mbuf chain "top" (which must be null
810  * if uio is not).  Data provided in mbuf chain must be small
811  * enough to send all at once.
812  *
813  * Returns nonzero on error, timeout or signal; callers
814  * must check for short counts if EINTR/ERESTART are returned.
815  * Data and control buffers are freed on return.
816  */
817 int
818 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
819 	struct mbuf *control, int flags, struct lwp *l)
820 {
821 	struct mbuf	**mp, *m;
822 	struct proc	*p;
823 	long		space, len, resid, clen, mlen;
824 	int		error, s, dontroute, atomic;
825 
826 	p = l->l_proc;
827 	sodopendfree();
828 	clen = 0;
829 
830 	/*
831 	 * solock() provides atomicity of access.  splsoftnet() prevents
832 	 * protocol processing soft interrupts from interrupting us and
833 	 * blocking (expensive).
834 	 */
835 	s = splsoftnet();
836 	solock(so);
837 	atomic = sosendallatonce(so) || top;
838 	if (uio)
839 		resid = uio->uio_resid;
840 	else
841 		resid = top->m_pkthdr.len;
842 	/*
843 	 * In theory resid should be unsigned.
844 	 * However, space must be signed, as it might be less than 0
845 	 * if we over-committed, and we must use a signed comparison
846 	 * of space and resid.  On the other hand, a negative resid
847 	 * causes us to loop sending 0-length segments to the protocol.
848 	 */
849 	if (resid < 0) {
850 		error = EINVAL;
851 		goto out;
852 	}
853 	dontroute =
854 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
855 	    (so->so_proto->pr_flags & PR_ATOMIC);
856 	l->l_ru.ru_msgsnd++;
857 	if (control)
858 		clen = control->m_len;
859  restart:
860 	if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
861 		goto out;
862 	do {
863 		if (so->so_state & SS_CANTSENDMORE) {
864 			error = EPIPE;
865 			goto release;
866 		}
867 		if (so->so_error) {
868 			error = so->so_error;
869 			so->so_error = 0;
870 			goto release;
871 		}
872 		if ((so->so_state & SS_ISCONNECTED) == 0) {
873 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
874 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
875 				    !(resid == 0 && clen != 0)) {
876 					error = ENOTCONN;
877 					goto release;
878 				}
879 			} else if (addr == 0) {
880 				error = EDESTADDRREQ;
881 				goto release;
882 			}
883 		}
884 		space = sbspace(&so->so_snd);
885 		if (flags & MSG_OOB)
886 			space += 1024;
887 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
888 		    clen > so->so_snd.sb_hiwat) {
889 			error = EMSGSIZE;
890 			goto release;
891 		}
892 		if (space < resid + clen &&
893 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
894 			if (so->so_nbio) {
895 				error = EWOULDBLOCK;
896 				goto release;
897 			}
898 			sbunlock(&so->so_snd);
899 			error = sbwait(&so->so_snd);
900 			if (error)
901 				goto out;
902 			goto restart;
903 		}
904 		mp = &top;
905 		space -= clen;
906 		do {
907 			if (uio == NULL) {
908 				/*
909 				 * Data is prepackaged in "top".
910 				 */
911 				resid = 0;
912 				if (flags & MSG_EOR)
913 					top->m_flags |= M_EOR;
914 			} else do {
915 				sounlock(so);
916 				splx(s);
917 				if (top == NULL) {
918 					m = m_gethdr(M_WAIT, MT_DATA);
919 					mlen = MHLEN;
920 					m->m_pkthdr.len = 0;
921 					m->m_pkthdr.rcvif = NULL;
922 				} else {
923 					m = m_get(M_WAIT, MT_DATA);
924 					mlen = MLEN;
925 				}
926 				MCLAIM(m, so->so_snd.sb_mowner);
927 				if (sock_loan_thresh >= 0 &&
928 				    uio->uio_iov->iov_len >= sock_loan_thresh &&
929 				    space >= sock_loan_thresh &&
930 				    (len = sosend_loan(so, uio, m,
931 						       space)) != 0) {
932 					SOSEND_COUNTER_INCR(&sosend_loan_big);
933 					space -= len;
934 					goto have_data;
935 				}
936 				if (resid >= MINCLSIZE && space >= MCLBYTES) {
937 					SOSEND_COUNTER_INCR(&sosend_copy_big);
938 					m_clget(m, M_WAIT);
939 					if ((m->m_flags & M_EXT) == 0)
940 						goto nopages;
941 					mlen = MCLBYTES;
942 					if (atomic && top == 0) {
943 						len = lmin(MCLBYTES - max_hdr,
944 						    resid);
945 						m->m_data += max_hdr;
946 					} else
947 						len = lmin(MCLBYTES, resid);
948 					space -= len;
949 				} else {
950  nopages:
951 					SOSEND_COUNTER_INCR(&sosend_copy_small);
952 					len = lmin(lmin(mlen, resid), space);
953 					space -= len;
954 					/*
955 					 * For datagram protocols, leave room
956 					 * for protocol headers in first mbuf.
957 					 */
958 					if (atomic && top == 0 && len < mlen)
959 						MH_ALIGN(m, len);
960 				}
961 				error = uiomove(mtod(m, void *), (int)len, uio);
962  have_data:
963 				resid = uio->uio_resid;
964 				m->m_len = len;
965 				*mp = m;
966 				top->m_pkthdr.len += len;
967 				s = splsoftnet();
968 				solock(so);
969 				if (error != 0)
970 					goto release;
971 				mp = &m->m_next;
972 				if (resid <= 0) {
973 					if (flags & MSG_EOR)
974 						top->m_flags |= M_EOR;
975 					break;
976 				}
977 			} while (space > 0 && atomic);
978 
979 			if (so->so_state & SS_CANTSENDMORE) {
980 				error = EPIPE;
981 				goto release;
982 			}
983 			if (dontroute)
984 				so->so_options |= SO_DONTROUTE;
985 			if (resid > 0)
986 				so->so_state |= SS_MORETOCOME;
987 			error = (*so->so_proto->pr_usrreq)(so,
988 			    (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND,
989 			    top, addr, control, curlwp);
990 			if (dontroute)
991 				so->so_options &= ~SO_DONTROUTE;
992 			if (resid > 0)
993 				so->so_state &= ~SS_MORETOCOME;
994 			clen = 0;
995 			control = NULL;
996 			top = NULL;
997 			mp = &top;
998 			if (error != 0)
999 				goto release;
1000 		} while (resid && space > 0);
1001 	} while (resid);
1002 
1003  release:
1004 	sbunlock(&so->so_snd);
1005  out:
1006 	sounlock(so);
1007 	splx(s);
1008 	if (top)
1009 		m_freem(top);
1010 	if (control)
1011 		m_freem(control);
1012 	return (error);
1013 }
1014 
1015 /*
1016  * Following replacement or removal of the first mbuf on the first
1017  * mbuf chain of a socket buffer, push necessary state changes back
1018  * into the socket buffer so that other consumers see the values
1019  * consistently.  'nextrecord' is the callers locally stored value of
1020  * the original value of sb->sb_mb->m_nextpkt which must be restored
1021  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
1022  */
1023 static void
1024 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
1025 {
1026 
1027 	KASSERT(solocked(sb->sb_so));
1028 
1029 	/*
1030 	 * First, update for the new value of nextrecord.  If necessary,
1031 	 * make it the first record.
1032 	 */
1033 	if (sb->sb_mb != NULL)
1034 		sb->sb_mb->m_nextpkt = nextrecord;
1035 	else
1036 		sb->sb_mb = nextrecord;
1037 
1038         /*
1039          * Now update any dependent socket buffer fields to reflect
1040          * the new state.  This is an inline of SB_EMPTY_FIXUP, with
1041          * the addition of a second clause that takes care of the
1042          * case where sb_mb has been updated, but remains the last
1043          * record.
1044          */
1045         if (sb->sb_mb == NULL) {
1046                 sb->sb_mbtail = NULL;
1047                 sb->sb_lastrecord = NULL;
1048         } else if (sb->sb_mb->m_nextpkt == NULL)
1049                 sb->sb_lastrecord = sb->sb_mb;
1050 }
1051 
1052 /*
1053  * Implement receive operations on a socket.
1054  * We depend on the way that records are added to the sockbuf
1055  * by sbappend*.  In particular, each record (mbufs linked through m_next)
1056  * must begin with an address if the protocol so specifies,
1057  * followed by an optional mbuf or mbufs containing ancillary data,
1058  * and then zero or more mbufs of data.
1059  * In order to avoid blocking network interrupts for the entire time here,
1060  * we splx() while doing the actual copy to user space.
1061  * Although the sockbuf is locked, new data may still be appended,
1062  * and thus we must maintain consistency of the sockbuf during that time.
1063  *
1064  * The caller may receive the data as a single mbuf chain by supplying
1065  * an mbuf **mp0 for use in returning the chain.  The uio is then used
1066  * only for the count in uio_resid.
1067  */
1068 int
1069 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
1070 	struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1071 {
1072 	struct lwp *l = curlwp;
1073 	struct mbuf	*m, **mp, *mt;
1074 	int atomic, flags, len, error, s, offset, moff, type, orig_resid;
1075 	const struct protosw	*pr;
1076 	struct mbuf	*nextrecord;
1077 	int		mbuf_removed = 0;
1078 	const struct domain *dom;
1079 
1080 	pr = so->so_proto;
1081 	atomic = pr->pr_flags & PR_ATOMIC;
1082 	dom = pr->pr_domain;
1083 	mp = mp0;
1084 	type = 0;
1085 	orig_resid = uio->uio_resid;
1086 
1087 	if (paddr != NULL)
1088 		*paddr = NULL;
1089 	if (controlp != NULL)
1090 		*controlp = NULL;
1091 	if (flagsp != NULL)
1092 		flags = *flagsp &~ MSG_EOR;
1093 	else
1094 		flags = 0;
1095 
1096 	if ((flags & MSG_DONTWAIT) == 0)
1097 		sodopendfree();
1098 
1099 	if (flags & MSG_OOB) {
1100 		m = m_get(M_WAIT, MT_DATA);
1101 		solock(so);
1102 		error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m,
1103 		    (struct mbuf *)(long)(flags & MSG_PEEK), NULL, l);
1104 		sounlock(so);
1105 		if (error)
1106 			goto bad;
1107 		do {
1108 			error = uiomove(mtod(m, void *),
1109 			    (int) min(uio->uio_resid, m->m_len), uio);
1110 			m = m_free(m);
1111 		} while (uio->uio_resid > 0 && error == 0 && m);
1112  bad:
1113 		if (m != NULL)
1114 			m_freem(m);
1115 		return error;
1116 	}
1117 	if (mp != NULL)
1118 		*mp = NULL;
1119 
1120 	/*
1121 	 * solock() provides atomicity of access.  splsoftnet() prevents
1122 	 * protocol processing soft interrupts from interrupting us and
1123 	 * blocking (expensive).
1124 	 */
1125 	s = splsoftnet();
1126 	solock(so);
1127 	if (so->so_state & SS_ISCONFIRMING && uio->uio_resid)
1128 		(*pr->pr_usrreq)(so, PRU_RCVD, NULL, NULL, NULL, l);
1129 
1130  restart:
1131 	if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) {
1132 		sounlock(so);
1133 		splx(s);
1134 		return error;
1135 	}
1136 
1137 	m = so->so_rcv.sb_mb;
1138 	/*
1139 	 * If we have less data than requested, block awaiting more
1140 	 * (subject to any timeout) if:
1141 	 *   1. the current count is less than the low water mark,
1142 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1143 	 *	receive operation at once if we block (resid <= hiwat), or
1144 	 *   3. MSG_DONTWAIT is not set.
1145 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1146 	 * we have to do the receive in sections, and thus risk returning
1147 	 * a short count if a timeout or signal occurs after we start.
1148 	 */
1149 	if (m == NULL ||
1150 	    ((flags & MSG_DONTWAIT) == 0 &&
1151 	     so->so_rcv.sb_cc < uio->uio_resid &&
1152 	     (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1153 	      ((flags & MSG_WAITALL) &&
1154 	       uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1155 	     m->m_nextpkt == NULL && !atomic)) {
1156 #ifdef DIAGNOSTIC
1157 		if (m == NULL && so->so_rcv.sb_cc)
1158 			panic("receive 1");
1159 #endif
1160 		if (so->so_error) {
1161 			if (m != NULL)
1162 				goto dontblock;
1163 			error = so->so_error;
1164 			if ((flags & MSG_PEEK) == 0)
1165 				so->so_error = 0;
1166 			goto release;
1167 		}
1168 		if (so->so_state & SS_CANTRCVMORE) {
1169 			if (m != NULL)
1170 				goto dontblock;
1171 			else
1172 				goto release;
1173 		}
1174 		for (; m != NULL; m = m->m_next)
1175 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1176 				m = so->so_rcv.sb_mb;
1177 				goto dontblock;
1178 			}
1179 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1180 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1181 			error = ENOTCONN;
1182 			goto release;
1183 		}
1184 		if (uio->uio_resid == 0)
1185 			goto release;
1186 		if (so->so_nbio || (flags & MSG_DONTWAIT)) {
1187 			error = EWOULDBLOCK;
1188 			goto release;
1189 		}
1190 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
1191 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
1192 		sbunlock(&so->so_rcv);
1193 		error = sbwait(&so->so_rcv);
1194 		if (error != 0) {
1195 			sounlock(so);
1196 			splx(s);
1197 			return error;
1198 		}
1199 		goto restart;
1200 	}
1201  dontblock:
1202 	/*
1203 	 * On entry here, m points to the first record of the socket buffer.
1204 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1205 	 * pointer to the next record in the socket buffer.  We must keep the
1206 	 * various socket buffer pointers and local stack versions of the
1207 	 * pointers in sync, pushing out modifications before dropping the
1208 	 * socket lock, and re-reading them when picking it up.
1209 	 *
1210 	 * Otherwise, we will race with the network stack appending new data
1211 	 * or records onto the socket buffer by using inconsistent/stale
1212 	 * versions of the field, possibly resulting in socket buffer
1213 	 * corruption.
1214 	 *
1215 	 * By holding the high-level sblock(), we prevent simultaneous
1216 	 * readers from pulling off the front of the socket buffer.
1217 	 */
1218 	if (l != NULL)
1219 		l->l_ru.ru_msgrcv++;
1220 	KASSERT(m == so->so_rcv.sb_mb);
1221 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
1222 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
1223 	nextrecord = m->m_nextpkt;
1224 	if (pr->pr_flags & PR_ADDR) {
1225 #ifdef DIAGNOSTIC
1226 		if (m->m_type != MT_SONAME)
1227 			panic("receive 1a");
1228 #endif
1229 		orig_resid = 0;
1230 		if (flags & MSG_PEEK) {
1231 			if (paddr)
1232 				*paddr = m_copy(m, 0, m->m_len);
1233 			m = m->m_next;
1234 		} else {
1235 			sbfree(&so->so_rcv, m);
1236 			mbuf_removed = 1;
1237 			if (paddr != NULL) {
1238 				*paddr = m;
1239 				so->so_rcv.sb_mb = m->m_next;
1240 				m->m_next = NULL;
1241 				m = so->so_rcv.sb_mb;
1242 			} else {
1243 				MFREE(m, so->so_rcv.sb_mb);
1244 				m = so->so_rcv.sb_mb;
1245 			}
1246 			sbsync(&so->so_rcv, nextrecord);
1247 		}
1248 	}
1249 
1250 	/*
1251 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1252 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1253 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1254 	 * perform externalization (or freeing if controlp == NULL).
1255 	 */
1256 	if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) {
1257 		struct mbuf *cm = NULL, *cmn;
1258 		struct mbuf **cme = &cm;
1259 
1260 		do {
1261 			if (flags & MSG_PEEK) {
1262 				if (controlp != NULL) {
1263 					*controlp = m_copy(m, 0, m->m_len);
1264 					controlp = &(*controlp)->m_next;
1265 				}
1266 				m = m->m_next;
1267 			} else {
1268 				sbfree(&so->so_rcv, m);
1269 				so->so_rcv.sb_mb = m->m_next;
1270 				m->m_next = NULL;
1271 				*cme = m;
1272 				cme = &(*cme)->m_next;
1273 				m = so->so_rcv.sb_mb;
1274 			}
1275 		} while (m != NULL && m->m_type == MT_CONTROL);
1276 		if ((flags & MSG_PEEK) == 0)
1277 			sbsync(&so->so_rcv, nextrecord);
1278 		for (; cm != NULL; cm = cmn) {
1279 			cmn = cm->m_next;
1280 			cm->m_next = NULL;
1281 			type = mtod(cm, struct cmsghdr *)->cmsg_type;
1282 			if (controlp != NULL) {
1283 				if (dom->dom_externalize != NULL &&
1284 				    type == SCM_RIGHTS) {
1285 					sounlock(so);
1286 					splx(s);
1287 					error = (*dom->dom_externalize)(cm, l);
1288 					s = splsoftnet();
1289 					solock(so);
1290 				}
1291 				*controlp = cm;
1292 				while (*controlp != NULL)
1293 					controlp = &(*controlp)->m_next;
1294 			} else {
1295 				/*
1296 				 * Dispose of any SCM_RIGHTS message that went
1297 				 * through the read path rather than recv.
1298 				 */
1299 				if (dom->dom_dispose != NULL &&
1300 				    type == SCM_RIGHTS) {
1301 				    	sounlock(so);
1302 					(*dom->dom_dispose)(cm);
1303 					solock(so);
1304 				}
1305 				m_freem(cm);
1306 			}
1307 		}
1308 		if (m != NULL)
1309 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1310 		else
1311 			nextrecord = so->so_rcv.sb_mb;
1312 		orig_resid = 0;
1313 	}
1314 
1315 	/* If m is non-NULL, we have some data to read. */
1316 	if (__predict_true(m != NULL)) {
1317 		type = m->m_type;
1318 		if (type == MT_OOBDATA)
1319 			flags |= MSG_OOB;
1320 	}
1321 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1322 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1323 
1324 	moff = 0;
1325 	offset = 0;
1326 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1327 		if (m->m_type == MT_OOBDATA) {
1328 			if (type != MT_OOBDATA)
1329 				break;
1330 		} else if (type == MT_OOBDATA)
1331 			break;
1332 #ifdef DIAGNOSTIC
1333 		else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
1334 			panic("receive 3");
1335 #endif
1336 		so->so_state &= ~SS_RCVATMARK;
1337 		len = uio->uio_resid;
1338 		if (so->so_oobmark && len > so->so_oobmark - offset)
1339 			len = so->so_oobmark - offset;
1340 		if (len > m->m_len - moff)
1341 			len = m->m_len - moff;
1342 		/*
1343 		 * If mp is set, just pass back the mbufs.
1344 		 * Otherwise copy them out via the uio, then free.
1345 		 * Sockbuf must be consistent here (points to current mbuf,
1346 		 * it points to next record) when we drop priority;
1347 		 * we must note any additions to the sockbuf when we
1348 		 * block interrupts again.
1349 		 */
1350 		if (mp == NULL) {
1351 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1352 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1353 			sounlock(so);
1354 			splx(s);
1355 			error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1356 			s = splsoftnet();
1357 			solock(so);
1358 			if (error != 0) {
1359 				/*
1360 				 * If any part of the record has been removed
1361 				 * (such as the MT_SONAME mbuf, which will
1362 				 * happen when PR_ADDR, and thus also
1363 				 * PR_ATOMIC, is set), then drop the entire
1364 				 * record to maintain the atomicity of the
1365 				 * receive operation.
1366 				 *
1367 				 * This avoids a later panic("receive 1a")
1368 				 * when compiled with DIAGNOSTIC.
1369 				 */
1370 				if (m && mbuf_removed && atomic)
1371 					(void) sbdroprecord(&so->so_rcv);
1372 
1373 				goto release;
1374 			}
1375 		} else
1376 			uio->uio_resid -= len;
1377 		if (len == m->m_len - moff) {
1378 			if (m->m_flags & M_EOR)
1379 				flags |= MSG_EOR;
1380 			if (flags & MSG_PEEK) {
1381 				m = m->m_next;
1382 				moff = 0;
1383 			} else {
1384 				nextrecord = m->m_nextpkt;
1385 				sbfree(&so->so_rcv, m);
1386 				if (mp) {
1387 					*mp = m;
1388 					mp = &m->m_next;
1389 					so->so_rcv.sb_mb = m = m->m_next;
1390 					*mp = NULL;
1391 				} else {
1392 					MFREE(m, so->so_rcv.sb_mb);
1393 					m = so->so_rcv.sb_mb;
1394 				}
1395 				/*
1396 				 * If m != NULL, we also know that
1397 				 * so->so_rcv.sb_mb != NULL.
1398 				 */
1399 				KASSERT(so->so_rcv.sb_mb == m);
1400 				if (m) {
1401 					m->m_nextpkt = nextrecord;
1402 					if (nextrecord == NULL)
1403 						so->so_rcv.sb_lastrecord = m;
1404 				} else {
1405 					so->so_rcv.sb_mb = nextrecord;
1406 					SB_EMPTY_FIXUP(&so->so_rcv);
1407 				}
1408 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1409 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1410 			}
1411 		} else if (flags & MSG_PEEK)
1412 			moff += len;
1413 		else {
1414 			if (mp != NULL) {
1415 				mt = m_copym(m, 0, len, M_NOWAIT);
1416 				if (__predict_false(mt == NULL)) {
1417 					sounlock(so);
1418 					mt = m_copym(m, 0, len, M_WAIT);
1419 					solock(so);
1420 				}
1421 				*mp = mt;
1422 			}
1423 			m->m_data += len;
1424 			m->m_len -= len;
1425 			so->so_rcv.sb_cc -= len;
1426 		}
1427 		if (so->so_oobmark) {
1428 			if ((flags & MSG_PEEK) == 0) {
1429 				so->so_oobmark -= len;
1430 				if (so->so_oobmark == 0) {
1431 					so->so_state |= SS_RCVATMARK;
1432 					break;
1433 				}
1434 			} else {
1435 				offset += len;
1436 				if (offset == so->so_oobmark)
1437 					break;
1438 			}
1439 		}
1440 		if (flags & MSG_EOR)
1441 			break;
1442 		/*
1443 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1444 		 * we must not quit until "uio->uio_resid == 0" or an error
1445 		 * termination.  If a signal/timeout occurs, return
1446 		 * with a short count but without error.
1447 		 * Keep sockbuf locked against other readers.
1448 		 */
1449 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1450 		    !sosendallatonce(so) && !nextrecord) {
1451 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1452 				break;
1453 			/*
1454 			 * If we are peeking and the socket receive buffer is
1455 			 * full, stop since we can't get more data to peek at.
1456 			 */
1457 			if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
1458 				break;
1459 			/*
1460 			 * If we've drained the socket buffer, tell the
1461 			 * protocol in case it needs to do something to
1462 			 * get it filled again.
1463 			 */
1464 			if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1465 				(*pr->pr_usrreq)(so, PRU_RCVD,
1466 				    NULL, (struct mbuf *)(long)flags, NULL, l);
1467 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1468 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1469 			error = sbwait(&so->so_rcv);
1470 			if (error != 0) {
1471 				sbunlock(&so->so_rcv);
1472 				sounlock(so);
1473 				splx(s);
1474 				return 0;
1475 			}
1476 			if ((m = so->so_rcv.sb_mb) != NULL)
1477 				nextrecord = m->m_nextpkt;
1478 		}
1479 	}
1480 
1481 	if (m && atomic) {
1482 		flags |= MSG_TRUNC;
1483 		if ((flags & MSG_PEEK) == 0)
1484 			(void) sbdroprecord(&so->so_rcv);
1485 	}
1486 	if ((flags & MSG_PEEK) == 0) {
1487 		if (m == NULL) {
1488 			/*
1489 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1490 			 * part makes sure sb_lastrecord is up-to-date if
1491 			 * there is still data in the socket buffer.
1492 			 */
1493 			so->so_rcv.sb_mb = nextrecord;
1494 			if (so->so_rcv.sb_mb == NULL) {
1495 				so->so_rcv.sb_mbtail = NULL;
1496 				so->so_rcv.sb_lastrecord = NULL;
1497 			} else if (nextrecord->m_nextpkt == NULL)
1498 				so->so_rcv.sb_lastrecord = nextrecord;
1499 		}
1500 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1501 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1502 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1503 			(*pr->pr_usrreq)(so, PRU_RCVD, NULL,
1504 			    (struct mbuf *)(long)flags, NULL, l);
1505 	}
1506 	if (orig_resid == uio->uio_resid && orig_resid &&
1507 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1508 		sbunlock(&so->so_rcv);
1509 		goto restart;
1510 	}
1511 
1512 	if (flagsp != NULL)
1513 		*flagsp |= flags;
1514  release:
1515 	sbunlock(&so->so_rcv);
1516 	sounlock(so);
1517 	splx(s);
1518 	return error;
1519 }
1520 
1521 int
1522 soshutdown(struct socket *so, int how)
1523 {
1524 	const struct protosw	*pr;
1525 	int	error;
1526 
1527 	KASSERT(solocked(so));
1528 
1529 	pr = so->so_proto;
1530 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1531 		return (EINVAL);
1532 
1533 	if (how == SHUT_RD || how == SHUT_RDWR) {
1534 		sorflush(so);
1535 		error = 0;
1536 	}
1537 	if (how == SHUT_WR || how == SHUT_RDWR)
1538 		error = (*pr->pr_usrreq)(so, PRU_SHUTDOWN, NULL,
1539 		    NULL, NULL, NULL);
1540 
1541 	return error;
1542 }
1543 
1544 void
1545 sorflush(struct socket *so)
1546 {
1547 	struct sockbuf	*sb, asb;
1548 	const struct protosw	*pr;
1549 
1550 	KASSERT(solocked(so));
1551 
1552 	sb = &so->so_rcv;
1553 	pr = so->so_proto;
1554 	socantrcvmore(so);
1555 	sb->sb_flags |= SB_NOINTR;
1556 	(void )sblock(sb, M_WAITOK);
1557 	sbunlock(sb);
1558 	asb = *sb;
1559 	/*
1560 	 * Clear most of the sockbuf structure, but leave some of the
1561 	 * fields valid.
1562 	 */
1563 	memset(&sb->sb_startzero, 0,
1564 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1565 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) {
1566 		sounlock(so);
1567 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1568 		solock(so);
1569 	}
1570 	sbrelease(&asb, so);
1571 }
1572 
1573 /*
1574  * internal set SOL_SOCKET options
1575  */
1576 static int
1577 sosetopt1(struct socket *so, const struct sockopt *sopt)
1578 {
1579 	int error, optval;
1580 	struct linger l;
1581 	struct timeval tv;
1582 
1583 	switch (sopt->sopt_name) {
1584 
1585 	case SO_ACCEPTFILTER:
1586 		error = accept_filt_setopt(so, sopt);
1587 		KASSERT(solocked(so));
1588 		break;
1589 
1590   	case SO_LINGER:
1591  		error = sockopt_get(sopt, &l, sizeof(l));
1592 		solock(so);
1593  		if (error)
1594  			break;
1595  		if (l.l_linger < 0 || l.l_linger > USHRT_MAX ||
1596  		    l.l_linger > (INT_MAX / hz)) {
1597 			error = EDOM;
1598 			break;
1599 		}
1600  		so->so_linger = l.l_linger;
1601  		if (l.l_onoff)
1602  			so->so_options |= SO_LINGER;
1603  		else
1604  			so->so_options &= ~SO_LINGER;
1605    		break;
1606 
1607 	case SO_DEBUG:
1608 	case SO_KEEPALIVE:
1609 	case SO_DONTROUTE:
1610 	case SO_USELOOPBACK:
1611 	case SO_BROADCAST:
1612 	case SO_REUSEADDR:
1613 	case SO_REUSEPORT:
1614 	case SO_OOBINLINE:
1615 	case SO_TIMESTAMP:
1616 		error = sockopt_getint(sopt, &optval);
1617 		solock(so);
1618 		if (error)
1619 			break;
1620 		if (optval)
1621 			so->so_options |= sopt->sopt_name;
1622 		else
1623 			so->so_options &= ~sopt->sopt_name;
1624 		break;
1625 
1626 	case SO_SNDBUF:
1627 	case SO_RCVBUF:
1628 	case SO_SNDLOWAT:
1629 	case SO_RCVLOWAT:
1630 		error = sockopt_getint(sopt, &optval);
1631 		solock(so);
1632 		if (error)
1633 			break;
1634 
1635 		/*
1636 		 * Values < 1 make no sense for any of these
1637 		 * options, so disallow them.
1638 		 */
1639 		if (optval < 1) {
1640 			error = EINVAL;
1641 			break;
1642 		}
1643 
1644 		switch (sopt->sopt_name) {
1645 		case SO_SNDBUF:
1646 			if (sbreserve(&so->so_snd, (u_long)optval, so) == 0) {
1647 				error = ENOBUFS;
1648 				break;
1649 			}
1650 			so->so_snd.sb_flags &= ~SB_AUTOSIZE;
1651 			break;
1652 
1653 		case SO_RCVBUF:
1654 			if (sbreserve(&so->so_rcv, (u_long)optval, so) == 0) {
1655 				error = ENOBUFS;
1656 				break;
1657 			}
1658 			so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1659 			break;
1660 
1661 		/*
1662 		 * Make sure the low-water is never greater than
1663 		 * the high-water.
1664 		 */
1665 		case SO_SNDLOWAT:
1666 			if (optval > so->so_snd.sb_hiwat)
1667 				optval = so->so_snd.sb_hiwat;
1668 
1669 			so->so_snd.sb_lowat = optval;
1670 			break;
1671 
1672 		case SO_RCVLOWAT:
1673 			if (optval > so->so_rcv.sb_hiwat)
1674 				optval = so->so_rcv.sb_hiwat;
1675 
1676 			so->so_rcv.sb_lowat = optval;
1677 			break;
1678 		}
1679 		break;
1680 
1681 	case SO_SNDTIMEO:
1682 	case SO_RCVTIMEO:
1683 		error = sockopt_get(sopt, &tv, sizeof(tv));
1684 		solock(so);
1685 		if (error)
1686 			break;
1687 
1688 		if (tv.tv_sec > (INT_MAX - tv.tv_usec / tick) / hz) {
1689 			error = EDOM;
1690 			break;
1691 		}
1692 
1693 		optval = tv.tv_sec * hz + tv.tv_usec / tick;
1694 		if (optval == 0 && tv.tv_usec != 0)
1695 			optval = 1;
1696 
1697 		switch (sopt->sopt_name) {
1698 		case SO_SNDTIMEO:
1699 			so->so_snd.sb_timeo = optval;
1700 			break;
1701 		case SO_RCVTIMEO:
1702 			so->so_rcv.sb_timeo = optval;
1703 			break;
1704 		}
1705 		break;
1706 
1707 	default:
1708 		solock(so);
1709 		error = ENOPROTOOPT;
1710 		break;
1711 	}
1712 	KASSERT(solocked(so));
1713 	return error;
1714 }
1715 
1716 int
1717 sosetopt(struct socket *so, struct sockopt *sopt)
1718 {
1719 	int error, prerr;
1720 
1721 	if (sopt->sopt_level == SOL_SOCKET) {
1722 		error = sosetopt1(so, sopt);
1723 		KASSERT(solocked(so));
1724 	} else {
1725 		error = ENOPROTOOPT;
1726 		solock(so);
1727 	}
1728 
1729 	if ((error == 0 || error == ENOPROTOOPT) &&
1730 	    so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) {
1731 		/* give the protocol stack a shot */
1732 		prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, sopt);
1733 		if (prerr == 0)
1734 			error = 0;
1735 		else if (prerr != ENOPROTOOPT)
1736 			error = prerr;
1737 	}
1738 	sounlock(so);
1739 	return error;
1740 }
1741 
1742 /*
1743  * so_setsockopt() is a wrapper providing a sockopt structure for sosetopt()
1744  */
1745 int
1746 so_setsockopt(struct lwp *l, struct socket *so, int level, int name,
1747     const void *val, size_t valsize)
1748 {
1749 	struct sockopt sopt;
1750 	int error;
1751 
1752 	KASSERT(valsize == 0 || val != NULL);
1753 
1754 	sockopt_init(&sopt, level, name, valsize);
1755 	sockopt_set(&sopt, val, valsize);
1756 
1757 	error = sosetopt(so, &sopt);
1758 
1759 	sockopt_destroy(&sopt);
1760 
1761 	return error;
1762 }
1763 
1764 /*
1765  * internal get SOL_SOCKET options
1766  */
1767 static int
1768 sogetopt1(struct socket *so, struct sockopt *sopt)
1769 {
1770 	int error, optval;
1771 	struct linger l;
1772 	struct timeval tv;
1773 
1774 	switch (sopt->sopt_name) {
1775 
1776 	case SO_ACCEPTFILTER:
1777 		error = accept_filt_getopt(so, sopt);
1778 		break;
1779 
1780 	case SO_LINGER:
1781 		l.l_onoff = (so->so_options & SO_LINGER) ? 1 : 0;
1782 		l.l_linger = so->so_linger;
1783 
1784 		error = sockopt_set(sopt, &l, sizeof(l));
1785 		break;
1786 
1787 	case SO_USELOOPBACK:
1788 	case SO_DONTROUTE:
1789 	case SO_DEBUG:
1790 	case SO_KEEPALIVE:
1791 	case SO_REUSEADDR:
1792 	case SO_REUSEPORT:
1793 	case SO_BROADCAST:
1794 	case SO_OOBINLINE:
1795 	case SO_TIMESTAMP:
1796 		error = sockopt_setint(sopt,
1797 		    (so->so_options & sopt->sopt_name) ? 1 : 0);
1798 		break;
1799 
1800 	case SO_TYPE:
1801 		error = sockopt_setint(sopt, so->so_type);
1802 		break;
1803 
1804 	case SO_ERROR:
1805 		error = sockopt_setint(sopt, so->so_error);
1806 		so->so_error = 0;
1807 		break;
1808 
1809 	case SO_SNDBUF:
1810 		error = sockopt_setint(sopt, so->so_snd.sb_hiwat);
1811 		break;
1812 
1813 	case SO_RCVBUF:
1814 		error = sockopt_setint(sopt, so->so_rcv.sb_hiwat);
1815 		break;
1816 
1817 	case SO_SNDLOWAT:
1818 		error = sockopt_setint(sopt, so->so_snd.sb_lowat);
1819 		break;
1820 
1821 	case SO_RCVLOWAT:
1822 		error = sockopt_setint(sopt, so->so_rcv.sb_lowat);
1823 		break;
1824 
1825 	case SO_SNDTIMEO:
1826 	case SO_RCVTIMEO:
1827 		optval = (sopt->sopt_name == SO_SNDTIMEO ?
1828 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1829 
1830 		tv.tv_sec = optval / hz;
1831 		tv.tv_usec = (optval % hz) * tick;
1832 
1833 		error = sockopt_set(sopt, &tv, sizeof(tv));
1834 		break;
1835 
1836 	case SO_OVERFLOWED:
1837 		error = sockopt_setint(sopt, so->so_rcv.sb_overflowed);
1838 		break;
1839 
1840 	default:
1841 		error = ENOPROTOOPT;
1842 		break;
1843 	}
1844 
1845 	return (error);
1846 }
1847 
1848 int
1849 sogetopt(struct socket *so, struct sockopt *sopt)
1850 {
1851 	int		error;
1852 
1853 	solock(so);
1854 	if (sopt->sopt_level != SOL_SOCKET) {
1855 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1856 			error = ((*so->so_proto->pr_ctloutput)
1857 			    (PRCO_GETOPT, so, sopt));
1858 		} else
1859 			error = (ENOPROTOOPT);
1860 	} else {
1861 		error = sogetopt1(so, sopt);
1862 	}
1863 	sounlock(so);
1864 	return (error);
1865 }
1866 
1867 /*
1868  * alloc sockopt data buffer buffer
1869  *	- will be released at destroy
1870  */
1871 static int
1872 sockopt_alloc(struct sockopt *sopt, size_t len, km_flag_t kmflag)
1873 {
1874 
1875 	KASSERT(sopt->sopt_size == 0);
1876 
1877 	if (len > sizeof(sopt->sopt_buf)) {
1878 		sopt->sopt_data = kmem_zalloc(len, kmflag);
1879 		if (sopt->sopt_data == NULL)
1880 			return ENOMEM;
1881 	} else
1882 		sopt->sopt_data = sopt->sopt_buf;
1883 
1884 	sopt->sopt_size = len;
1885 	return 0;
1886 }
1887 
1888 /*
1889  * initialise sockopt storage
1890  *	- MAY sleep during allocation
1891  */
1892 void
1893 sockopt_init(struct sockopt *sopt, int level, int name, size_t size)
1894 {
1895 
1896 	memset(sopt, 0, sizeof(*sopt));
1897 
1898 	sopt->sopt_level = level;
1899 	sopt->sopt_name = name;
1900 	(void)sockopt_alloc(sopt, size, KM_SLEEP);
1901 }
1902 
1903 /*
1904  * destroy sockopt storage
1905  *	- will release any held memory references
1906  */
1907 void
1908 sockopt_destroy(struct sockopt *sopt)
1909 {
1910 
1911 	if (sopt->sopt_data != sopt->sopt_buf)
1912 		kmem_free(sopt->sopt_data, sopt->sopt_size);
1913 
1914 	memset(sopt, 0, sizeof(*sopt));
1915 }
1916 
1917 /*
1918  * set sockopt value
1919  *	- value is copied into sockopt
1920  * 	- memory is allocated when necessary, will not sleep
1921  */
1922 int
1923 sockopt_set(struct sockopt *sopt, const void *buf, size_t len)
1924 {
1925 	int error;
1926 
1927 	if (sopt->sopt_size == 0) {
1928 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
1929 		if (error)
1930 			return error;
1931 	}
1932 
1933 	KASSERT(sopt->sopt_size == len);
1934 	memcpy(sopt->sopt_data, buf, len);
1935 	return 0;
1936 }
1937 
1938 /*
1939  * common case of set sockopt integer value
1940  */
1941 int
1942 sockopt_setint(struct sockopt *sopt, int val)
1943 {
1944 
1945 	return sockopt_set(sopt, &val, sizeof(int));
1946 }
1947 
1948 /*
1949  * get sockopt value
1950  *	- correct size must be given
1951  */
1952 int
1953 sockopt_get(const struct sockopt *sopt, void *buf, size_t len)
1954 {
1955 
1956 	if (sopt->sopt_size != len)
1957 		return EINVAL;
1958 
1959 	memcpy(buf, sopt->sopt_data, len);
1960 	return 0;
1961 }
1962 
1963 /*
1964  * common case of get sockopt integer value
1965  */
1966 int
1967 sockopt_getint(const struct sockopt *sopt, int *valp)
1968 {
1969 
1970 	return sockopt_get(sopt, valp, sizeof(int));
1971 }
1972 
1973 /*
1974  * set sockopt value from mbuf
1975  *	- ONLY for legacy code
1976  *	- mbuf is released by sockopt
1977  *	- will not sleep
1978  */
1979 int
1980 sockopt_setmbuf(struct sockopt *sopt, struct mbuf *m)
1981 {
1982 	size_t len;
1983 	int error;
1984 
1985 	len = m_length(m);
1986 
1987 	if (sopt->sopt_size == 0) {
1988 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
1989 		if (error)
1990 			return error;
1991 	}
1992 
1993 	KASSERT(sopt->sopt_size == len);
1994 	m_copydata(m, 0, len, sopt->sopt_data);
1995 	m_freem(m);
1996 
1997 	return 0;
1998 }
1999 
2000 /*
2001  * get sockopt value into mbuf
2002  *	- ONLY for legacy code
2003  *	- mbuf to be released by the caller
2004  *	- will not sleep
2005  */
2006 struct mbuf *
2007 sockopt_getmbuf(const struct sockopt *sopt)
2008 {
2009 	struct mbuf *m;
2010 
2011 	if (sopt->sopt_size > MCLBYTES)
2012 		return NULL;
2013 
2014 	m = m_get(M_DONTWAIT, MT_SOOPTS);
2015 	if (m == NULL)
2016 		return NULL;
2017 
2018 	if (sopt->sopt_size > MLEN) {
2019 		MCLGET(m, M_DONTWAIT);
2020 		if ((m->m_flags & M_EXT) == 0) {
2021 			m_free(m);
2022 			return NULL;
2023 		}
2024 	}
2025 
2026 	memcpy(mtod(m, void *), sopt->sopt_data, sopt->sopt_size);
2027 	m->m_len = sopt->sopt_size;
2028 
2029 	return m;
2030 }
2031 
2032 void
2033 sohasoutofband(struct socket *so)
2034 {
2035 
2036 	fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
2037 	selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, 0);
2038 }
2039 
2040 static void
2041 filt_sordetach(struct knote *kn)
2042 {
2043 	struct socket	*so;
2044 
2045 	so = ((file_t *)kn->kn_obj)->f_data;
2046 	solock(so);
2047 	SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
2048 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
2049 		so->so_rcv.sb_flags &= ~SB_KNOTE;
2050 	sounlock(so);
2051 }
2052 
2053 /*ARGSUSED*/
2054 static int
2055 filt_soread(struct knote *kn, long hint)
2056 {
2057 	struct socket	*so;
2058 	int rv;
2059 
2060 	so = ((file_t *)kn->kn_obj)->f_data;
2061 	if (hint != NOTE_SUBMIT)
2062 		solock(so);
2063 	kn->kn_data = so->so_rcv.sb_cc;
2064 	if (so->so_state & SS_CANTRCVMORE) {
2065 		kn->kn_flags |= EV_EOF;
2066 		kn->kn_fflags = so->so_error;
2067 		rv = 1;
2068 	} else if (so->so_error)	/* temporary udp error */
2069 		rv = 1;
2070 	else if (kn->kn_sfflags & NOTE_LOWAT)
2071 		rv = (kn->kn_data >= kn->kn_sdata);
2072 	else
2073 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
2074 	if (hint != NOTE_SUBMIT)
2075 		sounlock(so);
2076 	return rv;
2077 }
2078 
2079 static void
2080 filt_sowdetach(struct knote *kn)
2081 {
2082 	struct socket	*so;
2083 
2084 	so = ((file_t *)kn->kn_obj)->f_data;
2085 	solock(so);
2086 	SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
2087 	if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
2088 		so->so_snd.sb_flags &= ~SB_KNOTE;
2089 	sounlock(so);
2090 }
2091 
2092 /*ARGSUSED*/
2093 static int
2094 filt_sowrite(struct knote *kn, long hint)
2095 {
2096 	struct socket	*so;
2097 	int rv;
2098 
2099 	so = ((file_t *)kn->kn_obj)->f_data;
2100 	if (hint != NOTE_SUBMIT)
2101 		solock(so);
2102 	kn->kn_data = sbspace(&so->so_snd);
2103 	if (so->so_state & SS_CANTSENDMORE) {
2104 		kn->kn_flags |= EV_EOF;
2105 		kn->kn_fflags = so->so_error;
2106 		rv = 1;
2107 	} else if (so->so_error)	/* temporary udp error */
2108 		rv = 1;
2109 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2110 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
2111 		rv = 0;
2112 	else if (kn->kn_sfflags & NOTE_LOWAT)
2113 		rv = (kn->kn_data >= kn->kn_sdata);
2114 	else
2115 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
2116 	if (hint != NOTE_SUBMIT)
2117 		sounlock(so);
2118 	return rv;
2119 }
2120 
2121 /*ARGSUSED*/
2122 static int
2123 filt_solisten(struct knote *kn, long hint)
2124 {
2125 	struct socket	*so;
2126 	int rv;
2127 
2128 	so = ((file_t *)kn->kn_obj)->f_data;
2129 
2130 	/*
2131 	 * Set kn_data to number of incoming connections, not
2132 	 * counting partial (incomplete) connections.
2133 	 */
2134 	if (hint != NOTE_SUBMIT)
2135 		solock(so);
2136 	kn->kn_data = so->so_qlen;
2137 	rv = (kn->kn_data > 0);
2138 	if (hint != NOTE_SUBMIT)
2139 		sounlock(so);
2140 	return rv;
2141 }
2142 
2143 static const struct filterops solisten_filtops =
2144 	{ 1, NULL, filt_sordetach, filt_solisten };
2145 static const struct filterops soread_filtops =
2146 	{ 1, NULL, filt_sordetach, filt_soread };
2147 static const struct filterops sowrite_filtops =
2148 	{ 1, NULL, filt_sowdetach, filt_sowrite };
2149 
2150 int
2151 soo_kqfilter(struct file *fp, struct knote *kn)
2152 {
2153 	struct socket	*so;
2154 	struct sockbuf	*sb;
2155 
2156 	so = ((file_t *)kn->kn_obj)->f_data;
2157 	solock(so);
2158 	switch (kn->kn_filter) {
2159 	case EVFILT_READ:
2160 		if (so->so_options & SO_ACCEPTCONN)
2161 			kn->kn_fop = &solisten_filtops;
2162 		else
2163 			kn->kn_fop = &soread_filtops;
2164 		sb = &so->so_rcv;
2165 		break;
2166 	case EVFILT_WRITE:
2167 		kn->kn_fop = &sowrite_filtops;
2168 		sb = &so->so_snd;
2169 		break;
2170 	default:
2171 		sounlock(so);
2172 		return (EINVAL);
2173 	}
2174 	SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
2175 	sb->sb_flags |= SB_KNOTE;
2176 	sounlock(so);
2177 	return (0);
2178 }
2179 
2180 static int
2181 sodopoll(struct socket *so, int events)
2182 {
2183 	int revents;
2184 
2185 	revents = 0;
2186 
2187 	if (events & (POLLIN | POLLRDNORM))
2188 		if (soreadable(so))
2189 			revents |= events & (POLLIN | POLLRDNORM);
2190 
2191 	if (events & (POLLOUT | POLLWRNORM))
2192 		if (sowritable(so))
2193 			revents |= events & (POLLOUT | POLLWRNORM);
2194 
2195 	if (events & (POLLPRI | POLLRDBAND))
2196 		if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
2197 			revents |= events & (POLLPRI | POLLRDBAND);
2198 
2199 	return revents;
2200 }
2201 
2202 int
2203 sopoll(struct socket *so, int events)
2204 {
2205 	int revents = 0;
2206 
2207 #ifndef DIAGNOSTIC
2208 	/*
2209 	 * Do a quick, unlocked check in expectation that the socket
2210 	 * will be ready for I/O.  Don't do this check if DIAGNOSTIC,
2211 	 * as the solocked() assertions will fail.
2212 	 */
2213 	if ((revents = sodopoll(so, events)) != 0)
2214 		return revents;
2215 #endif
2216 
2217 	solock(so);
2218 	if ((revents = sodopoll(so, events)) == 0) {
2219 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2220 			selrecord(curlwp, &so->so_rcv.sb_sel);
2221 			so->so_rcv.sb_flags |= SB_NOTIFY;
2222 		}
2223 
2224 		if (events & (POLLOUT | POLLWRNORM)) {
2225 			selrecord(curlwp, &so->so_snd.sb_sel);
2226 			so->so_snd.sb_flags |= SB_NOTIFY;
2227 		}
2228 	}
2229 	sounlock(so);
2230 
2231 	return revents;
2232 }
2233 
2234 
2235 #include <sys/sysctl.h>
2236 
2237 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO);
2238 
2239 /*
2240  * sysctl helper routine for kern.somaxkva.  ensures that the given
2241  * value is not too small.
2242  * (XXX should we maybe make sure it's not too large as well?)
2243  */
2244 static int
2245 sysctl_kern_somaxkva(SYSCTLFN_ARGS)
2246 {
2247 	int error, new_somaxkva;
2248 	struct sysctlnode node;
2249 
2250 	new_somaxkva = somaxkva;
2251 	node = *rnode;
2252 	node.sysctl_data = &new_somaxkva;
2253 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2254 	if (error || newp == NULL)
2255 		return (error);
2256 
2257 	if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */
2258 		return (EINVAL);
2259 
2260 	mutex_enter(&so_pendfree_lock);
2261 	somaxkva = new_somaxkva;
2262 	cv_broadcast(&socurkva_cv);
2263 	mutex_exit(&so_pendfree_lock);
2264 
2265 	return (error);
2266 }
2267 
2268 static void
2269 sysctl_kern_somaxkva_setup()
2270 {
2271 
2272 	KASSERT(socket_sysctllog == NULL);
2273 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2274 		       CTLFLAG_PERMANENT,
2275 		       CTLTYPE_NODE, "kern", NULL,
2276 		       NULL, 0, NULL, 0,
2277 		       CTL_KERN, CTL_EOL);
2278 
2279 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2280 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2281 		       CTLTYPE_INT, "somaxkva",
2282 		       SYSCTL_DESCR("Maximum amount of kernel memory to be "
2283 				    "used for socket buffers"),
2284 		       sysctl_kern_somaxkva, 0, NULL, 0,
2285 		       CTL_KERN, KERN_SOMAXKVA, CTL_EOL);
2286 }
2287