xref: /netbsd-src/sys/kern/uipc_socket.c (revision e89934bbf778a6d6d6894877c4da59d0c7835b0f)
1 /*	$NetBSD: uipc_socket.c,v 1.252 2016/10/13 19:10:23 uwe Exp $	*/
2 
3 /*-
4  * Copyright (c) 2002, 2007, 2008, 2009 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe of Wasabi Systems, Inc, and by Andrew Doran.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Copyright (c) 2004 The FreeBSD Foundation
34  * Copyright (c) 2004 Robert Watson
35  * Copyright (c) 1982, 1986, 1988, 1990, 1993
36  *	The Regents of the University of California.  All rights reserved.
37  *
38  * Redistribution and use in source and binary forms, with or without
39  * modification, are permitted provided that the following conditions
40  * are met:
41  * 1. Redistributions of source code must retain the above copyright
42  *    notice, this list of conditions and the following disclaimer.
43  * 2. Redistributions in binary form must reproduce the above copyright
44  *    notice, this list of conditions and the following disclaimer in the
45  *    documentation and/or other materials provided with the distribution.
46  * 3. Neither the name of the University nor the names of its contributors
47  *    may be used to endorse or promote products derived from this software
48  *    without specific prior written permission.
49  *
50  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
52  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
53  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
54  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
55  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
56  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
57  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
58  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
59  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
60  * SUCH DAMAGE.
61  *
62  *	@(#)uipc_socket.c	8.6 (Berkeley) 5/2/95
63  */
64 
65 /*
66  * Socket operation routines.
67  *
68  * These routines are called by the routines in sys_socket.c or from a
69  * system process, and implement the semantics of socket operations by
70  * switching out to the protocol specific routines.
71  */
72 
73 #include <sys/cdefs.h>
74 __KERNEL_RCSID(0, "$NetBSD: uipc_socket.c,v 1.252 2016/10/13 19:10:23 uwe Exp $");
75 
76 #ifdef _KERNEL_OPT
77 #include "opt_compat_netbsd.h"
78 #include "opt_sock_counters.h"
79 #include "opt_sosend_loan.h"
80 #include "opt_mbuftrace.h"
81 #include "opt_somaxkva.h"
82 #include "opt_multiprocessor.h"	/* XXX */
83 #include "opt_sctp.h"
84 #endif
85 
86 #include <sys/param.h>
87 #include <sys/systm.h>
88 #include <sys/proc.h>
89 #include <sys/file.h>
90 #include <sys/filedesc.h>
91 #include <sys/kmem.h>
92 #include <sys/mbuf.h>
93 #include <sys/domain.h>
94 #include <sys/kernel.h>
95 #include <sys/protosw.h>
96 #include <sys/socket.h>
97 #include <sys/socketvar.h>
98 #include <sys/signalvar.h>
99 #include <sys/resourcevar.h>
100 #include <sys/uidinfo.h>
101 #include <sys/event.h>
102 #include <sys/poll.h>
103 #include <sys/kauth.h>
104 #include <sys/mutex.h>
105 #include <sys/condvar.h>
106 #include <sys/kthread.h>
107 
108 #ifdef COMPAT_50
109 #include <compat/sys/time.h>
110 #include <compat/sys/socket.h>
111 #endif
112 
113 #include <uvm/uvm_extern.h>
114 #include <uvm/uvm_loan.h>
115 #include <uvm/uvm_page.h>
116 
117 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
118 
119 extern const struct fileops socketops;
120 
121 extern int	somaxconn;			/* patchable (XXX sysctl) */
122 int		somaxconn = SOMAXCONN;
123 kmutex_t	*softnet_lock;
124 
125 #ifdef SOSEND_COUNTERS
126 #include <sys/device.h>
127 
128 static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
129     NULL, "sosend", "loan big");
130 static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
131     NULL, "sosend", "copy big");
132 static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
133     NULL, "sosend", "copy small");
134 static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
135     NULL, "sosend", "kva limit");
136 
137 #define	SOSEND_COUNTER_INCR(ev)		(ev)->ev_count++
138 
139 EVCNT_ATTACH_STATIC(sosend_loan_big);
140 EVCNT_ATTACH_STATIC(sosend_copy_big);
141 EVCNT_ATTACH_STATIC(sosend_copy_small);
142 EVCNT_ATTACH_STATIC(sosend_kvalimit);
143 #else
144 
145 #define	SOSEND_COUNTER_INCR(ev)		/* nothing */
146 
147 #endif /* SOSEND_COUNTERS */
148 
149 #if defined(SOSEND_NO_LOAN) || defined(MULTIPROCESSOR)
150 int sock_loan_thresh = -1;
151 #else
152 int sock_loan_thresh = 4096;
153 #endif
154 
155 static kmutex_t so_pendfree_lock;
156 static struct mbuf *so_pendfree = NULL;
157 
158 #ifndef SOMAXKVA
159 #define	SOMAXKVA (16 * 1024 * 1024)
160 #endif
161 int somaxkva = SOMAXKVA;
162 static int socurkva;
163 static kcondvar_t socurkva_cv;
164 
165 static kauth_listener_t socket_listener;
166 
167 #define	SOCK_LOAN_CHUNK		65536
168 
169 static void sopendfree_thread(void *);
170 static kcondvar_t pendfree_thread_cv;
171 static lwp_t *sopendfree_lwp;
172 
173 static void sysctl_kern_socket_setup(void);
174 static struct sysctllog *socket_sysctllog;
175 
176 static vsize_t
177 sokvareserve(struct socket *so, vsize_t len)
178 {
179 	int error;
180 
181 	mutex_enter(&so_pendfree_lock);
182 	while (socurkva + len > somaxkva) {
183 		SOSEND_COUNTER_INCR(&sosend_kvalimit);
184 		error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock);
185 		if (error) {
186 			len = 0;
187 			break;
188 		}
189 	}
190 	socurkva += len;
191 	mutex_exit(&so_pendfree_lock);
192 	return len;
193 }
194 
195 static void
196 sokvaunreserve(vsize_t len)
197 {
198 
199 	mutex_enter(&so_pendfree_lock);
200 	socurkva -= len;
201 	cv_broadcast(&socurkva_cv);
202 	mutex_exit(&so_pendfree_lock);
203 }
204 
205 /*
206  * sokvaalloc: allocate kva for loan.
207  */
208 
209 vaddr_t
210 sokvaalloc(vaddr_t sva, vsize_t len, struct socket *so)
211 {
212 	vaddr_t lva;
213 
214 	/*
215 	 * reserve kva.
216 	 */
217 
218 	if (sokvareserve(so, len) == 0)
219 		return 0;
220 
221 	/*
222 	 * allocate kva.
223 	 */
224 
225 	lva = uvm_km_alloc(kernel_map, len, atop(sva) & uvmexp.colormask,
226 	    UVM_KMF_COLORMATCH | UVM_KMF_VAONLY | UVM_KMF_WAITVA);
227 	if (lva == 0) {
228 		sokvaunreserve(len);
229 		return (0);
230 	}
231 
232 	return lva;
233 }
234 
235 /*
236  * sokvafree: free kva for loan.
237  */
238 
239 void
240 sokvafree(vaddr_t sva, vsize_t len)
241 {
242 
243 	/*
244 	 * free kva.
245 	 */
246 
247 	uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY);
248 
249 	/*
250 	 * unreserve kva.
251 	 */
252 
253 	sokvaunreserve(len);
254 }
255 
256 static void
257 sodoloanfree(struct vm_page **pgs, void *buf, size_t size)
258 {
259 	vaddr_t sva, eva;
260 	vsize_t len;
261 	int npgs;
262 
263 	KASSERT(pgs != NULL);
264 
265 	eva = round_page((vaddr_t) buf + size);
266 	sva = trunc_page((vaddr_t) buf);
267 	len = eva - sva;
268 	npgs = len >> PAGE_SHIFT;
269 
270 	pmap_kremove(sva, len);
271 	pmap_update(pmap_kernel());
272 	uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE);
273 	sokvafree(sva, len);
274 }
275 
276 /*
277  * sopendfree_thread: free mbufs on "pendfree" list.
278  * unlock and relock so_pendfree_lock when freeing mbufs.
279  */
280 
281 static void
282 sopendfree_thread(void *v)
283 {
284 	struct mbuf *m, *next;
285 	size_t rv;
286 
287 	mutex_enter(&so_pendfree_lock);
288 
289 	for (;;) {
290 		rv = 0;
291 		while (so_pendfree != NULL) {
292 			m = so_pendfree;
293 			so_pendfree = NULL;
294 			mutex_exit(&so_pendfree_lock);
295 
296 			for (; m != NULL; m = next) {
297 				next = m->m_next;
298 				KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0);
299 				KASSERT(m->m_ext.ext_refcnt == 0);
300 
301 				rv += m->m_ext.ext_size;
302 				sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf,
303 				    m->m_ext.ext_size);
304 				pool_cache_put(mb_cache, m);
305 			}
306 
307 			mutex_enter(&so_pendfree_lock);
308 		}
309 		if (rv)
310 			cv_broadcast(&socurkva_cv);
311 		cv_wait(&pendfree_thread_cv, &so_pendfree_lock);
312 	}
313 	panic("sopendfree_thread");
314 	/* NOTREACHED */
315 }
316 
317 void
318 soloanfree(struct mbuf *m, void *buf, size_t size, void *arg)
319 {
320 
321 	KASSERT(m != NULL);
322 
323 	/*
324 	 * postpone freeing mbuf.
325 	 *
326 	 * we can't do it in interrupt context
327 	 * because we need to put kva back to kernel_map.
328 	 */
329 
330 	mutex_enter(&so_pendfree_lock);
331 	m->m_next = so_pendfree;
332 	so_pendfree = m;
333 	cv_signal(&pendfree_thread_cv);
334 	mutex_exit(&so_pendfree_lock);
335 }
336 
337 static long
338 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space)
339 {
340 	struct iovec *iov = uio->uio_iov;
341 	vaddr_t sva, eva;
342 	vsize_t len;
343 	vaddr_t lva;
344 	int npgs, error;
345 	vaddr_t va;
346 	int i;
347 
348 	if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace))
349 		return (0);
350 
351 	if (iov->iov_len < (size_t) space)
352 		space = iov->iov_len;
353 	if (space > SOCK_LOAN_CHUNK)
354 		space = SOCK_LOAN_CHUNK;
355 
356 	eva = round_page((vaddr_t) iov->iov_base + space);
357 	sva = trunc_page((vaddr_t) iov->iov_base);
358 	len = eva - sva;
359 	npgs = len >> PAGE_SHIFT;
360 
361 	KASSERT(npgs <= M_EXT_MAXPAGES);
362 
363 	lva = sokvaalloc(sva, len, so);
364 	if (lva == 0)
365 		return 0;
366 
367 	error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len,
368 	    m->m_ext.ext_pgs, UVM_LOAN_TOPAGE);
369 	if (error) {
370 		sokvafree(lva, len);
371 		return (0);
372 	}
373 
374 	for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE)
375 		pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
376 		    VM_PROT_READ, 0);
377 	pmap_update(pmap_kernel());
378 
379 	lva += (vaddr_t) iov->iov_base & PAGE_MASK;
380 
381 	MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so);
382 	m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
383 
384 	uio->uio_resid -= space;
385 	/* uio_offset not updated, not set/used for write(2) */
386 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space;
387 	uio->uio_iov->iov_len -= space;
388 	if (uio->uio_iov->iov_len == 0) {
389 		uio->uio_iov++;
390 		uio->uio_iovcnt--;
391 	}
392 
393 	return (space);
394 }
395 
396 struct mbuf *
397 getsombuf(struct socket *so, int type)
398 {
399 	struct mbuf *m;
400 
401 	m = m_get(M_WAIT, type);
402 	MCLAIM(m, so->so_mowner);
403 	return m;
404 }
405 
406 static int
407 socket_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
408     void *arg0, void *arg1, void *arg2, void *arg3)
409 {
410 	int result;
411 	enum kauth_network_req req;
412 
413 	result = KAUTH_RESULT_DEFER;
414 	req = (enum kauth_network_req)arg0;
415 
416 	if ((action != KAUTH_NETWORK_SOCKET) &&
417 	    (action != KAUTH_NETWORK_BIND))
418 		return result;
419 
420 	switch (req) {
421 	case KAUTH_REQ_NETWORK_BIND_PORT:
422 		result = KAUTH_RESULT_ALLOW;
423 		break;
424 
425 	case KAUTH_REQ_NETWORK_SOCKET_DROP: {
426 		/* Normal users can only drop their own connections. */
427 		struct socket *so = (struct socket *)arg1;
428 
429 		if (so->so_cred && proc_uidmatch(cred, so->so_cred) == 0)
430 			result = KAUTH_RESULT_ALLOW;
431 
432 		break;
433 		}
434 
435 	case KAUTH_REQ_NETWORK_SOCKET_OPEN:
436 		/* We allow "raw" routing/bluetooth sockets to anyone. */
437 		if ((u_long)arg1 == PF_ROUTE || (u_long)arg1 == PF_OROUTE
438 		    || (u_long)arg1 == PF_BLUETOOTH) {
439 			result = KAUTH_RESULT_ALLOW;
440 		} else {
441 			/* Privileged, let secmodel handle this. */
442 			if ((u_long)arg2 == SOCK_RAW)
443 				break;
444 		}
445 
446 		result = KAUTH_RESULT_ALLOW;
447 
448 		break;
449 
450 	case KAUTH_REQ_NETWORK_SOCKET_CANSEE:
451 		result = KAUTH_RESULT_ALLOW;
452 
453 		break;
454 
455 	default:
456 		break;
457 	}
458 
459 	return result;
460 }
461 
462 void
463 soinit(void)
464 {
465 
466 	sysctl_kern_socket_setup();
467 
468 	mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM);
469 	softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
470 	cv_init(&socurkva_cv, "sokva");
471 	cv_init(&pendfree_thread_cv, "sopendfr");
472 	soinit2();
473 
474 	/* Set the initial adjusted socket buffer size. */
475 	if (sb_max_set(sb_max))
476 		panic("bad initial sb_max value: %lu", sb_max);
477 
478 	socket_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK,
479 	    socket_listener_cb, NULL);
480 }
481 
482 void
483 soinit1(void)
484 {
485 	int error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL,
486 	    sopendfree_thread, NULL, &sopendfree_lwp, "sopendfree");
487 	if (error)
488 		panic("soinit1 %d", error);
489 }
490 
491 /*
492  * socreate: create a new socket of the specified type and the protocol.
493  *
494  * => Caller may specify another socket for lock sharing (must not be held).
495  * => Returns the new socket without lock held.
496  */
497 int
498 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l,
499 	 struct socket *lockso)
500 {
501 	const struct protosw	*prp;
502 	struct socket	*so;
503 	uid_t		uid;
504 	int		error;
505 	kmutex_t	*lock;
506 
507 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET,
508 	    KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type),
509 	    KAUTH_ARG(proto));
510 	if (error != 0)
511 		return error;
512 
513 	if (proto)
514 		prp = pffindproto(dom, proto, type);
515 	else
516 		prp = pffindtype(dom, type);
517 	if (prp == NULL) {
518 		/* no support for domain */
519 		if (pffinddomain(dom) == 0)
520 			return EAFNOSUPPORT;
521 		/* no support for socket type */
522 		if (proto == 0 && type != 0)
523 			return EPROTOTYPE;
524 		return EPROTONOSUPPORT;
525 	}
526 	if (prp->pr_usrreqs == NULL)
527 		return EPROTONOSUPPORT;
528 	if (prp->pr_type != type)
529 		return EPROTOTYPE;
530 
531 	so = soget(true);
532 	so->so_type = type;
533 	so->so_proto = prp;
534 	so->so_send = sosend;
535 	so->so_receive = soreceive;
536 #ifdef MBUFTRACE
537 	so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner;
538 	so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner;
539 	so->so_mowner = &prp->pr_domain->dom_mowner;
540 #endif
541 	uid = kauth_cred_geteuid(l->l_cred);
542 	so->so_uidinfo = uid_find(uid);
543 	so->so_cpid = l->l_proc->p_pid;
544 
545 	/*
546 	 * Lock assigned and taken during PCB attach, unless we share
547 	 * the lock with another socket, e.g. socketpair(2) case.
548 	 */
549 	if (lockso) {
550 		lock = lockso->so_lock;
551 		so->so_lock = lock;
552 		mutex_obj_hold(lock);
553 		mutex_enter(lock);
554 	}
555 
556 	/* Attach the PCB (returns with the socket lock held). */
557 	error = (*prp->pr_usrreqs->pr_attach)(so, proto);
558 	KASSERT(solocked(so));
559 
560 	if (error) {
561 		KASSERT(so->so_pcb == NULL);
562 		so->so_state |= SS_NOFDREF;
563 		sofree(so);
564 		return error;
565 	}
566 	so->so_cred = kauth_cred_dup(l->l_cred);
567 	sounlock(so);
568 
569 	*aso = so;
570 	return 0;
571 }
572 
573 /*
574  * fsocreate: create a socket and a file descriptor associated with it.
575  *
576  * => On success, write file descriptor to fdout and return zero.
577  * => On failure, return non-zero; *fdout will be undefined.
578  */
579 int
580 fsocreate(int domain, struct socket **sop, int type, int proto, int *fdout)
581 {
582 	lwp_t *l = curlwp;
583 	int error, fd, flags;
584 	struct socket *so;
585 	struct file *fp;
586 
587 	if ((error = fd_allocfile(&fp, &fd)) != 0) {
588 		return error;
589 	}
590 	flags = type & SOCK_FLAGS_MASK;
591 	fd_set_exclose(l, fd, (flags & SOCK_CLOEXEC) != 0);
592 	fp->f_flag = FREAD|FWRITE|((flags & SOCK_NONBLOCK) ? FNONBLOCK : 0)|
593 	    ((flags & SOCK_NOSIGPIPE) ? FNOSIGPIPE : 0);
594 	fp->f_type = DTYPE_SOCKET;
595 	fp->f_ops = &socketops;
596 
597 	type &= ~SOCK_FLAGS_MASK;
598 	error = socreate(domain, &so, type, proto, l, NULL);
599 	if (error) {
600 		fd_abort(curproc, fp, fd);
601 		return error;
602 	}
603 	if (flags & SOCK_NONBLOCK) {
604 		so->so_state |= SS_NBIO;
605 	}
606 	fp->f_socket = so;
607 	fd_affix(curproc, fp, fd);
608 
609 	if (sop != NULL) {
610 		*sop = so;
611 	}
612 	*fdout = fd;
613 	return error;
614 }
615 
616 int
617 sofamily(const struct socket *so)
618 {
619 	const struct protosw *pr;
620 	const struct domain *dom;
621 
622 	if ((pr = so->so_proto) == NULL)
623 		return AF_UNSPEC;
624 	if ((dom = pr->pr_domain) == NULL)
625 		return AF_UNSPEC;
626 	return dom->dom_family;
627 }
628 
629 int
630 sobind(struct socket *so, struct sockaddr *nam, struct lwp *l)
631 {
632 	int	error;
633 
634 	solock(so);
635 	if (nam->sa_family != so->so_proto->pr_domain->dom_family) {
636 		sounlock(so);
637 		return EAFNOSUPPORT;
638 	}
639 	error = (*so->so_proto->pr_usrreqs->pr_bind)(so, nam, l);
640 	sounlock(so);
641 	return error;
642 }
643 
644 int
645 solisten(struct socket *so, int backlog, struct lwp *l)
646 {
647 	int	error;
648 	short	oldopt, oldqlimit;
649 
650 	solock(so);
651 	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
652 	    SS_ISDISCONNECTING)) != 0) {
653 		sounlock(so);
654 		return EINVAL;
655 	}
656 	oldopt = so->so_options;
657 	oldqlimit = so->so_qlimit;
658 	if (TAILQ_EMPTY(&so->so_q))
659 		so->so_options |= SO_ACCEPTCONN;
660 	if (backlog < 0)
661 		backlog = 0;
662 	so->so_qlimit = min(backlog, somaxconn);
663 
664 	error = (*so->so_proto->pr_usrreqs->pr_listen)(so, l);
665 	if (error != 0) {
666 		so->so_options = oldopt;
667 		so->so_qlimit = oldqlimit;
668 		sounlock(so);
669 		return error;
670 	}
671 	sounlock(so);
672 	return 0;
673 }
674 
675 void
676 sofree(struct socket *so)
677 {
678 	u_int refs;
679 
680 	KASSERT(solocked(so));
681 
682 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
683 		sounlock(so);
684 		return;
685 	}
686 	if (so->so_head) {
687 		/*
688 		 * We must not decommission a socket that's on the accept(2)
689 		 * queue.  If we do, then accept(2) may hang after select(2)
690 		 * indicated that the listening socket was ready.
691 		 */
692 		if (!soqremque(so, 0)) {
693 			sounlock(so);
694 			return;
695 		}
696 	}
697 	if (so->so_rcv.sb_hiwat)
698 		(void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0,
699 		    RLIM_INFINITY);
700 	if (so->so_snd.sb_hiwat)
701 		(void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0,
702 		    RLIM_INFINITY);
703 	sbrelease(&so->so_snd, so);
704 	KASSERT(!cv_has_waiters(&so->so_cv));
705 	KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv));
706 	KASSERT(!cv_has_waiters(&so->so_snd.sb_cv));
707 	sorflush(so);
708 	refs = so->so_aborting;	/* XXX */
709 	/* Remove acccept filter if one is present. */
710 	if (so->so_accf != NULL)
711 		(void)accept_filt_clear(so);
712 	sounlock(so);
713 	if (refs == 0)		/* XXX */
714 		soput(so);
715 }
716 
717 /*
718  * soclose: close a socket on last file table reference removal.
719  * Initiate disconnect if connected.  Free socket when disconnect complete.
720  */
721 int
722 soclose(struct socket *so)
723 {
724 	struct socket *so2;
725 	int error = 0;
726 
727 	solock(so);
728 	if (so->so_options & SO_ACCEPTCONN) {
729 		for (;;) {
730 			if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) {
731 				KASSERT(solocked2(so, so2));
732 				(void) soqremque(so2, 0);
733 				/* soabort drops the lock. */
734 				(void) soabort(so2);
735 				solock(so);
736 				continue;
737 			}
738 			if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) {
739 				KASSERT(solocked2(so, so2));
740 				(void) soqremque(so2, 1);
741 				/* soabort drops the lock. */
742 				(void) soabort(so2);
743 				solock(so);
744 				continue;
745 			}
746 			break;
747 		}
748 	}
749 	if (so->so_pcb == NULL)
750 		goto discard;
751 	if (so->so_state & SS_ISCONNECTED) {
752 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
753 			error = sodisconnect(so);
754 			if (error)
755 				goto drop;
756 		}
757 		if (so->so_options & SO_LINGER) {
758 			if ((so->so_state & (SS_ISDISCONNECTING|SS_NBIO)) ==
759 			    (SS_ISDISCONNECTING|SS_NBIO))
760 				goto drop;
761 			while (so->so_state & SS_ISCONNECTED) {
762 				error = sowait(so, true, so->so_linger * hz);
763 				if (error)
764 					break;
765 			}
766 		}
767 	}
768  drop:
769 	if (so->so_pcb) {
770 		KASSERT(solocked(so));
771 		(*so->so_proto->pr_usrreqs->pr_detach)(so);
772 	}
773  discard:
774 	KASSERT((so->so_state & SS_NOFDREF) == 0);
775 	kauth_cred_free(so->so_cred);
776 	so->so_state |= SS_NOFDREF;
777 	sofree(so);
778 	return error;
779 }
780 
781 /*
782  * Must be called with the socket locked..  Will return with it unlocked.
783  */
784 int
785 soabort(struct socket *so)
786 {
787 	u_int refs;
788 	int error;
789 
790 	KASSERT(solocked(so));
791 	KASSERT(so->so_head == NULL);
792 
793 	so->so_aborting++;		/* XXX */
794 	error = (*so->so_proto->pr_usrreqs->pr_abort)(so);
795 	refs = --so->so_aborting;	/* XXX */
796 	if (error || (refs == 0)) {
797 		sofree(so);
798 	} else {
799 		sounlock(so);
800 	}
801 	return error;
802 }
803 
804 int
805 soaccept(struct socket *so, struct sockaddr *nam)
806 {
807 	int error;
808 
809 	KASSERT(solocked(so));
810 	KASSERT((so->so_state & SS_NOFDREF) != 0);
811 
812 	so->so_state &= ~SS_NOFDREF;
813 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
814 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
815 		error = (*so->so_proto->pr_usrreqs->pr_accept)(so, nam);
816 	else
817 		error = ECONNABORTED;
818 
819 	return error;
820 }
821 
822 int
823 soconnect(struct socket *so, struct sockaddr *nam, struct lwp *l)
824 {
825 	int error;
826 
827 	KASSERT(solocked(so));
828 
829 	if (so->so_options & SO_ACCEPTCONN)
830 		return EOPNOTSUPP;
831 	/*
832 	 * If protocol is connection-based, can only connect once.
833 	 * Otherwise, if connected, try to disconnect first.
834 	 * This allows user to disconnect by connecting to, e.g.,
835 	 * a null address.
836 	 */
837 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
838 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
839 	    (error = sodisconnect(so)))) {
840 		error = EISCONN;
841 	} else {
842 		if (nam->sa_family != so->so_proto->pr_domain->dom_family) {
843 			return EAFNOSUPPORT;
844 		}
845 		error = (*so->so_proto->pr_usrreqs->pr_connect)(so, nam, l);
846 	}
847 
848 	return error;
849 }
850 
851 int
852 soconnect2(struct socket *so1, struct socket *so2)
853 {
854 	KASSERT(solocked2(so1, so2));
855 
856 	return (*so1->so_proto->pr_usrreqs->pr_connect2)(so1, so2);
857 }
858 
859 int
860 sodisconnect(struct socket *so)
861 {
862 	int	error;
863 
864 	KASSERT(solocked(so));
865 
866 	if ((so->so_state & SS_ISCONNECTED) == 0) {
867 		error = ENOTCONN;
868 	} else if (so->so_state & SS_ISDISCONNECTING) {
869 		error = EALREADY;
870 	} else {
871 		error = (*so->so_proto->pr_usrreqs->pr_disconnect)(so);
872 	}
873 	return (error);
874 }
875 
876 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
877 /*
878  * Send on a socket.
879  * If send must go all at once and message is larger than
880  * send buffering, then hard error.
881  * Lock against other senders.
882  * If must go all at once and not enough room now, then
883  * inform user that this would block and do nothing.
884  * Otherwise, if nonblocking, send as much as possible.
885  * The data to be sent is described by "uio" if nonzero,
886  * otherwise by the mbuf chain "top" (which must be null
887  * if uio is not).  Data provided in mbuf chain must be small
888  * enough to send all at once.
889  *
890  * Returns nonzero on error, timeout or signal; callers
891  * must check for short counts if EINTR/ERESTART are returned.
892  * Data and control buffers are freed on return.
893  */
894 int
895 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
896 	struct mbuf *top, struct mbuf *control, int flags, struct lwp *l)
897 {
898 	struct mbuf	**mp, *m;
899 	long		space, len, resid, clen, mlen;
900 	int		error, s, dontroute, atomic;
901 	short		wakeup_state = 0;
902 
903 	clen = 0;
904 
905 	/*
906 	 * solock() provides atomicity of access.  splsoftnet() prevents
907 	 * protocol processing soft interrupts from interrupting us and
908 	 * blocking (expensive).
909 	 */
910 	s = splsoftnet();
911 	solock(so);
912 	atomic = sosendallatonce(so) || top;
913 	if (uio)
914 		resid = uio->uio_resid;
915 	else
916 		resid = top->m_pkthdr.len;
917 	/*
918 	 * In theory resid should be unsigned.
919 	 * However, space must be signed, as it might be less than 0
920 	 * if we over-committed, and we must use a signed comparison
921 	 * of space and resid.  On the other hand, a negative resid
922 	 * causes us to loop sending 0-length segments to the protocol.
923 	 */
924 	if (resid < 0) {
925 		error = EINVAL;
926 		goto out;
927 	}
928 	dontroute =
929 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
930 	    (so->so_proto->pr_flags & PR_ATOMIC);
931 	l->l_ru.ru_msgsnd++;
932 	if (control)
933 		clen = control->m_len;
934  restart:
935 	if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
936 		goto out;
937 	do {
938 		if (so->so_state & SS_CANTSENDMORE) {
939 			error = EPIPE;
940 			goto release;
941 		}
942 		if (so->so_error) {
943 			error = so->so_error;
944 			so->so_error = 0;
945 			goto release;
946 		}
947 		if ((so->so_state & SS_ISCONNECTED) == 0) {
948 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
949 				if (resid || clen == 0) {
950 					error = ENOTCONN;
951 					goto release;
952 				}
953 			} else if (addr == NULL) {
954 				error = EDESTADDRREQ;
955 				goto release;
956 			}
957 		}
958 		space = sbspace(&so->so_snd);
959 		if (flags & MSG_OOB)
960 			space += 1024;
961 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
962 		    clen > so->so_snd.sb_hiwat) {
963 			error = EMSGSIZE;
964 			goto release;
965 		}
966 		if (space < resid + clen &&
967 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
968 			if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
969 				error = EWOULDBLOCK;
970 				goto release;
971 			}
972 			sbunlock(&so->so_snd);
973 			if (wakeup_state & SS_RESTARTSYS) {
974 				error = ERESTART;
975 				goto out;
976 			}
977 			error = sbwait(&so->so_snd);
978 			if (error)
979 				goto out;
980 			wakeup_state = so->so_state;
981 			goto restart;
982 		}
983 		wakeup_state = 0;
984 		mp = &top;
985 		space -= clen;
986 		do {
987 			if (uio == NULL) {
988 				/*
989 				 * Data is prepackaged in "top".
990 				 */
991 				resid = 0;
992 				if (flags & MSG_EOR)
993 					top->m_flags |= M_EOR;
994 			} else do {
995 				sounlock(so);
996 				splx(s);
997 				if (top == NULL) {
998 					m = m_gethdr(M_WAIT, MT_DATA);
999 					mlen = MHLEN;
1000 					m->m_pkthdr.len = 0;
1001 					m_reset_rcvif(m);
1002 				} else {
1003 					m = m_get(M_WAIT, MT_DATA);
1004 					mlen = MLEN;
1005 				}
1006 				MCLAIM(m, so->so_snd.sb_mowner);
1007 				if (sock_loan_thresh >= 0 &&
1008 				    uio->uio_iov->iov_len >= sock_loan_thresh &&
1009 				    space >= sock_loan_thresh &&
1010 				    (len = sosend_loan(so, uio, m,
1011 						       space)) != 0) {
1012 					SOSEND_COUNTER_INCR(&sosend_loan_big);
1013 					space -= len;
1014 					goto have_data;
1015 				}
1016 				if (resid >= MINCLSIZE && space >= MCLBYTES) {
1017 					SOSEND_COUNTER_INCR(&sosend_copy_big);
1018 					m_clget(m, M_DONTWAIT);
1019 					if ((m->m_flags & M_EXT) == 0)
1020 						goto nopages;
1021 					mlen = MCLBYTES;
1022 					if (atomic && top == 0) {
1023 						len = lmin(MCLBYTES - max_hdr,
1024 						    resid);
1025 						m->m_data += max_hdr;
1026 					} else
1027 						len = lmin(MCLBYTES, resid);
1028 					space -= len;
1029 				} else {
1030  nopages:
1031 					SOSEND_COUNTER_INCR(&sosend_copy_small);
1032 					len = lmin(lmin(mlen, resid), space);
1033 					space -= len;
1034 					/*
1035 					 * For datagram protocols, leave room
1036 					 * for protocol headers in first mbuf.
1037 					 */
1038 					if (atomic && top == 0 && len < mlen)
1039 						MH_ALIGN(m, len);
1040 				}
1041 				error = uiomove(mtod(m, void *), (int)len, uio);
1042  have_data:
1043 				resid = uio->uio_resid;
1044 				m->m_len = len;
1045 				*mp = m;
1046 				top->m_pkthdr.len += len;
1047 				s = splsoftnet();
1048 				solock(so);
1049 				if (error != 0)
1050 					goto release;
1051 				mp = &m->m_next;
1052 				if (resid <= 0) {
1053 					if (flags & MSG_EOR)
1054 						top->m_flags |= M_EOR;
1055 					break;
1056 				}
1057 			} while (space > 0 && atomic);
1058 
1059 			if (so->so_state & SS_CANTSENDMORE) {
1060 				error = EPIPE;
1061 				goto release;
1062 			}
1063 			if (dontroute)
1064 				so->so_options |= SO_DONTROUTE;
1065 			if (resid > 0)
1066 				so->so_state |= SS_MORETOCOME;
1067 			if (flags & MSG_OOB) {
1068 				error = (*so->so_proto->pr_usrreqs->pr_sendoob)(so,
1069 				    top, control);
1070 			} else {
1071 				error = (*so->so_proto->pr_usrreqs->pr_send)(so,
1072 				    top, addr, control, l);
1073 			}
1074 			if (dontroute)
1075 				so->so_options &= ~SO_DONTROUTE;
1076 			if (resid > 0)
1077 				so->so_state &= ~SS_MORETOCOME;
1078 			clen = 0;
1079 			control = NULL;
1080 			top = NULL;
1081 			mp = &top;
1082 			if (error != 0)
1083 				goto release;
1084 		} while (resid && space > 0);
1085 	} while (resid);
1086 
1087  release:
1088 	sbunlock(&so->so_snd);
1089  out:
1090 	sounlock(so);
1091 	splx(s);
1092 	if (top)
1093 		m_freem(top);
1094 	if (control)
1095 		m_freem(control);
1096 	return (error);
1097 }
1098 
1099 /*
1100  * Following replacement or removal of the first mbuf on the first
1101  * mbuf chain of a socket buffer, push necessary state changes back
1102  * into the socket buffer so that other consumers see the values
1103  * consistently.  'nextrecord' is the callers locally stored value of
1104  * the original value of sb->sb_mb->m_nextpkt which must be restored
1105  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
1106  */
1107 static void
1108 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
1109 {
1110 
1111 	KASSERT(solocked(sb->sb_so));
1112 
1113 	/*
1114 	 * First, update for the new value of nextrecord.  If necessary,
1115 	 * make it the first record.
1116 	 */
1117 	if (sb->sb_mb != NULL)
1118 		sb->sb_mb->m_nextpkt = nextrecord;
1119 	else
1120 		sb->sb_mb = nextrecord;
1121 
1122         /*
1123          * Now update any dependent socket buffer fields to reflect
1124          * the new state.  This is an inline of SB_EMPTY_FIXUP, with
1125          * the addition of a second clause that takes care of the
1126          * case where sb_mb has been updated, but remains the last
1127          * record.
1128          */
1129         if (sb->sb_mb == NULL) {
1130                 sb->sb_mbtail = NULL;
1131                 sb->sb_lastrecord = NULL;
1132         } else if (sb->sb_mb->m_nextpkt == NULL)
1133                 sb->sb_lastrecord = sb->sb_mb;
1134 }
1135 
1136 /*
1137  * Implement receive operations on a socket.
1138  * We depend on the way that records are added to the sockbuf
1139  * by sbappend*.  In particular, each record (mbufs linked through m_next)
1140  * must begin with an address if the protocol so specifies,
1141  * followed by an optional mbuf or mbufs containing ancillary data,
1142  * and then zero or more mbufs of data.
1143  * In order to avoid blocking network interrupts for the entire time here,
1144  * we splx() while doing the actual copy to user space.
1145  * Although the sockbuf is locked, new data may still be appended,
1146  * and thus we must maintain consistency of the sockbuf during that time.
1147  *
1148  * The caller may receive the data as a single mbuf chain by supplying
1149  * an mbuf **mp0 for use in returning the chain.  The uio is then used
1150  * only for the count in uio_resid.
1151  */
1152 int
1153 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
1154 	struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1155 {
1156 	struct lwp *l = curlwp;
1157 	struct mbuf	*m, **mp, *mt;
1158 	size_t len, offset, moff, orig_resid;
1159 	int atomic, flags, error, s, type;
1160 	const struct protosw	*pr;
1161 	struct mbuf	*nextrecord;
1162 	int		mbuf_removed = 0;
1163 	const struct domain *dom;
1164 	short		wakeup_state = 0;
1165 
1166 	pr = so->so_proto;
1167 	atomic = pr->pr_flags & PR_ATOMIC;
1168 	dom = pr->pr_domain;
1169 	mp = mp0;
1170 	type = 0;
1171 	orig_resid = uio->uio_resid;
1172 
1173 	if (paddr != NULL)
1174 		*paddr = NULL;
1175 	if (controlp != NULL)
1176 		*controlp = NULL;
1177 	if (flagsp != NULL)
1178 		flags = *flagsp &~ MSG_EOR;
1179 	else
1180 		flags = 0;
1181 
1182 	if (flags & MSG_OOB) {
1183 		m = m_get(M_WAIT, MT_DATA);
1184 		solock(so);
1185 		error = (*pr->pr_usrreqs->pr_recvoob)(so, m, flags & MSG_PEEK);
1186 		sounlock(so);
1187 		if (error)
1188 			goto bad;
1189 		do {
1190 			error = uiomove(mtod(m, void *),
1191 			    MIN(uio->uio_resid, m->m_len), uio);
1192 			m = m_free(m);
1193 		} while (uio->uio_resid > 0 && error == 0 && m);
1194  bad:
1195 		if (m != NULL)
1196 			m_freem(m);
1197 		return error;
1198 	}
1199 	if (mp != NULL)
1200 		*mp = NULL;
1201 
1202 	/*
1203 	 * solock() provides atomicity of access.  splsoftnet() prevents
1204 	 * protocol processing soft interrupts from interrupting us and
1205 	 * blocking (expensive).
1206 	 */
1207 	s = splsoftnet();
1208 	solock(so);
1209  restart:
1210 	if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) {
1211 		sounlock(so);
1212 		splx(s);
1213 		return error;
1214 	}
1215 
1216 	m = so->so_rcv.sb_mb;
1217 	/*
1218 	 * If we have less data than requested, block awaiting more
1219 	 * (subject to any timeout) if:
1220 	 *   1. the current count is less than the low water mark,
1221 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1222 	 *	receive operation at once if we block (resid <= hiwat), or
1223 	 *   3. MSG_DONTWAIT is not set.
1224 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1225 	 * we have to do the receive in sections, and thus risk returning
1226 	 * a short count if a timeout or signal occurs after we start.
1227 	 */
1228 	if (m == NULL ||
1229 	    ((flags & MSG_DONTWAIT) == 0 &&
1230 	     so->so_rcv.sb_cc < uio->uio_resid &&
1231 	     (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1232 	      ((flags & MSG_WAITALL) &&
1233 	       uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1234 	     m->m_nextpkt == NULL && !atomic)) {
1235 #ifdef DIAGNOSTIC
1236 		if (m == NULL && so->so_rcv.sb_cc)
1237 			panic("receive 1");
1238 #endif
1239 		if (so->so_error) {
1240 			if (m != NULL)
1241 				goto dontblock;
1242 			error = so->so_error;
1243 			if ((flags & MSG_PEEK) == 0)
1244 				so->so_error = 0;
1245 			goto release;
1246 		}
1247 		if (so->so_state & SS_CANTRCVMORE) {
1248 			if (m != NULL)
1249 				goto dontblock;
1250 			else
1251 				goto release;
1252 		}
1253 		for (; m != NULL; m = m->m_next)
1254 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1255 				m = so->so_rcv.sb_mb;
1256 				goto dontblock;
1257 			}
1258 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1259 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1260 			error = ENOTCONN;
1261 			goto release;
1262 		}
1263 		if (uio->uio_resid == 0)
1264 			goto release;
1265 		if ((so->so_state & SS_NBIO) ||
1266 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1267 			error = EWOULDBLOCK;
1268 			goto release;
1269 		}
1270 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
1271 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
1272 		sbunlock(&so->so_rcv);
1273 		if (wakeup_state & SS_RESTARTSYS)
1274 			error = ERESTART;
1275 		else
1276 			error = sbwait(&so->so_rcv);
1277 		if (error != 0) {
1278 			sounlock(so);
1279 			splx(s);
1280 			return error;
1281 		}
1282 		wakeup_state = so->so_state;
1283 		goto restart;
1284 	}
1285  dontblock:
1286 	/*
1287 	 * On entry here, m points to the first record of the socket buffer.
1288 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1289 	 * pointer to the next record in the socket buffer.  We must keep the
1290 	 * various socket buffer pointers and local stack versions of the
1291 	 * pointers in sync, pushing out modifications before dropping the
1292 	 * socket lock, and re-reading them when picking it up.
1293 	 *
1294 	 * Otherwise, we will race with the network stack appending new data
1295 	 * or records onto the socket buffer by using inconsistent/stale
1296 	 * versions of the field, possibly resulting in socket buffer
1297 	 * corruption.
1298 	 *
1299 	 * By holding the high-level sblock(), we prevent simultaneous
1300 	 * readers from pulling off the front of the socket buffer.
1301 	 */
1302 	if (l != NULL)
1303 		l->l_ru.ru_msgrcv++;
1304 	KASSERT(m == so->so_rcv.sb_mb);
1305 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
1306 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
1307 	nextrecord = m->m_nextpkt;
1308 	if (pr->pr_flags & PR_ADDR) {
1309 #ifdef DIAGNOSTIC
1310 		if (m->m_type != MT_SONAME)
1311 			panic("receive 1a");
1312 #endif
1313 		orig_resid = 0;
1314 		if (flags & MSG_PEEK) {
1315 			if (paddr)
1316 				*paddr = m_copy(m, 0, m->m_len);
1317 			m = m->m_next;
1318 		} else {
1319 			sbfree(&so->so_rcv, m);
1320 			mbuf_removed = 1;
1321 			if (paddr != NULL) {
1322 				*paddr = m;
1323 				so->so_rcv.sb_mb = m->m_next;
1324 				m->m_next = NULL;
1325 				m = so->so_rcv.sb_mb;
1326 			} else {
1327 				m = so->so_rcv.sb_mb = m_free(m);
1328 			}
1329 			sbsync(&so->so_rcv, nextrecord);
1330 		}
1331 	}
1332 	if (pr->pr_flags & PR_ADDR_OPT) {
1333 		/*
1334 		 * For SCTP we may be getting a
1335 		 * whole message OR a partial delivery.
1336 		 */
1337 		if (m->m_type == MT_SONAME) {
1338 			orig_resid = 0;
1339 			if (flags & MSG_PEEK) {
1340 				if (paddr)
1341 					*paddr = m_copy(m, 0, m->m_len);
1342 				m = m->m_next;
1343 			} else {
1344 				sbfree(&so->so_rcv, m);
1345 				if (paddr) {
1346 					*paddr = m;
1347 					so->so_rcv.sb_mb = m->m_next;
1348 					m->m_next = 0;
1349 					m = so->so_rcv.sb_mb;
1350 				} else {
1351 					m = so->so_rcv.sb_mb = m_free(m);
1352 				}
1353 			}
1354 		}
1355 	}
1356 
1357 	/*
1358 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1359 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1360 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1361 	 * perform externalization (or freeing if controlp == NULL).
1362 	 */
1363 	if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) {
1364 		struct mbuf *cm = NULL, *cmn;
1365 		struct mbuf **cme = &cm;
1366 
1367 		do {
1368 			if (flags & MSG_PEEK) {
1369 				if (controlp != NULL) {
1370 					*controlp = m_copy(m, 0, m->m_len);
1371 					controlp = &(*controlp)->m_next;
1372 				}
1373 				m = m->m_next;
1374 			} else {
1375 				sbfree(&so->so_rcv, m);
1376 				so->so_rcv.sb_mb = m->m_next;
1377 				m->m_next = NULL;
1378 				*cme = m;
1379 				cme = &(*cme)->m_next;
1380 				m = so->so_rcv.sb_mb;
1381 			}
1382 		} while (m != NULL && m->m_type == MT_CONTROL);
1383 		if ((flags & MSG_PEEK) == 0)
1384 			sbsync(&so->so_rcv, nextrecord);
1385 		for (; cm != NULL; cm = cmn) {
1386 			cmn = cm->m_next;
1387 			cm->m_next = NULL;
1388 			type = mtod(cm, struct cmsghdr *)->cmsg_type;
1389 			if (controlp != NULL) {
1390 				if (dom->dom_externalize != NULL &&
1391 				    type == SCM_RIGHTS) {
1392 					sounlock(so);
1393 					splx(s);
1394 					error = (*dom->dom_externalize)(cm, l,
1395 					    (flags & MSG_CMSG_CLOEXEC) ?
1396 					    O_CLOEXEC : 0);
1397 					s = splsoftnet();
1398 					solock(so);
1399 				}
1400 				*controlp = cm;
1401 				while (*controlp != NULL)
1402 					controlp = &(*controlp)->m_next;
1403 			} else {
1404 				/*
1405 				 * Dispose of any SCM_RIGHTS message that went
1406 				 * through the read path rather than recv.
1407 				 */
1408 				if (dom->dom_dispose != NULL &&
1409 				    type == SCM_RIGHTS) {
1410 				    	sounlock(so);
1411 					(*dom->dom_dispose)(cm);
1412 					solock(so);
1413 				}
1414 				m_freem(cm);
1415 			}
1416 		}
1417 		if (m != NULL)
1418 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1419 		else
1420 			nextrecord = so->so_rcv.sb_mb;
1421 		orig_resid = 0;
1422 	}
1423 
1424 	/* If m is non-NULL, we have some data to read. */
1425 	if (__predict_true(m != NULL)) {
1426 		type = m->m_type;
1427 		if (type == MT_OOBDATA)
1428 			flags |= MSG_OOB;
1429 	}
1430 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1431 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1432 
1433 	moff = 0;
1434 	offset = 0;
1435 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1436 		if (m->m_type == MT_OOBDATA) {
1437 			if (type != MT_OOBDATA)
1438 				break;
1439 		} else if (type == MT_OOBDATA)
1440 			break;
1441 #ifdef DIAGNOSTIC
1442 		else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
1443 			panic("receive 3");
1444 #endif
1445 		so->so_state &= ~SS_RCVATMARK;
1446 		wakeup_state = 0;
1447 		len = uio->uio_resid;
1448 		if (so->so_oobmark && len > so->so_oobmark - offset)
1449 			len = so->so_oobmark - offset;
1450 		if (len > m->m_len - moff)
1451 			len = m->m_len - moff;
1452 		/*
1453 		 * If mp is set, just pass back the mbufs.
1454 		 * Otherwise copy them out via the uio, then free.
1455 		 * Sockbuf must be consistent here (points to current mbuf,
1456 		 * it points to next record) when we drop priority;
1457 		 * we must note any additions to the sockbuf when we
1458 		 * block interrupts again.
1459 		 */
1460 		if (mp == NULL) {
1461 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1462 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1463 			sounlock(so);
1464 			splx(s);
1465 			error = uiomove(mtod(m, char *) + moff, len, uio);
1466 			s = splsoftnet();
1467 			solock(so);
1468 			if (error != 0) {
1469 				/*
1470 				 * If any part of the record has been removed
1471 				 * (such as the MT_SONAME mbuf, which will
1472 				 * happen when PR_ADDR, and thus also
1473 				 * PR_ATOMIC, is set), then drop the entire
1474 				 * record to maintain the atomicity of the
1475 				 * receive operation.
1476 				 *
1477 				 * This avoids a later panic("receive 1a")
1478 				 * when compiled with DIAGNOSTIC.
1479 				 */
1480 				if (m && mbuf_removed && atomic)
1481 					(void) sbdroprecord(&so->so_rcv);
1482 
1483 				goto release;
1484 			}
1485 		} else
1486 			uio->uio_resid -= len;
1487 		if (len == m->m_len - moff) {
1488 			if (m->m_flags & M_EOR)
1489 				flags |= MSG_EOR;
1490 #ifdef SCTP
1491 			if (m->m_flags & M_NOTIFICATION)
1492 				flags |= MSG_NOTIFICATION;
1493 #endif /* SCTP */
1494 			if (flags & MSG_PEEK) {
1495 				m = m->m_next;
1496 				moff = 0;
1497 			} else {
1498 				nextrecord = m->m_nextpkt;
1499 				sbfree(&so->so_rcv, m);
1500 				if (mp) {
1501 					*mp = m;
1502 					mp = &m->m_next;
1503 					so->so_rcv.sb_mb = m = m->m_next;
1504 					*mp = NULL;
1505 				} else {
1506 					m = so->so_rcv.sb_mb = m_free(m);
1507 				}
1508 				/*
1509 				 * If m != NULL, we also know that
1510 				 * so->so_rcv.sb_mb != NULL.
1511 				 */
1512 				KASSERT(so->so_rcv.sb_mb == m);
1513 				if (m) {
1514 					m->m_nextpkt = nextrecord;
1515 					if (nextrecord == NULL)
1516 						so->so_rcv.sb_lastrecord = m;
1517 				} else {
1518 					so->so_rcv.sb_mb = nextrecord;
1519 					SB_EMPTY_FIXUP(&so->so_rcv);
1520 				}
1521 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1522 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1523 			}
1524 		} else if (flags & MSG_PEEK)
1525 			moff += len;
1526 		else {
1527 			if (mp != NULL) {
1528 				mt = m_copym(m, 0, len, M_NOWAIT);
1529 				if (__predict_false(mt == NULL)) {
1530 					sounlock(so);
1531 					mt = m_copym(m, 0, len, M_WAIT);
1532 					solock(so);
1533 				}
1534 				*mp = mt;
1535 			}
1536 			m->m_data += len;
1537 			m->m_len -= len;
1538 			so->so_rcv.sb_cc -= len;
1539 		}
1540 		if (so->so_oobmark) {
1541 			if ((flags & MSG_PEEK) == 0) {
1542 				so->so_oobmark -= len;
1543 				if (so->so_oobmark == 0) {
1544 					so->so_state |= SS_RCVATMARK;
1545 					break;
1546 				}
1547 			} else {
1548 				offset += len;
1549 				if (offset == so->so_oobmark)
1550 					break;
1551 			}
1552 		}
1553 		if (flags & MSG_EOR)
1554 			break;
1555 		/*
1556 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1557 		 * we must not quit until "uio->uio_resid == 0" or an error
1558 		 * termination.  If a signal/timeout occurs, return
1559 		 * with a short count but without error.
1560 		 * Keep sockbuf locked against other readers.
1561 		 */
1562 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1563 		    !sosendallatonce(so) && !nextrecord) {
1564 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1565 				break;
1566 			/*
1567 			 * If we are peeking and the socket receive buffer is
1568 			 * full, stop since we can't get more data to peek at.
1569 			 */
1570 			if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
1571 				break;
1572 			/*
1573 			 * If we've drained the socket buffer, tell the
1574 			 * protocol in case it needs to do something to
1575 			 * get it filled again.
1576 			 */
1577 			if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1578 				(*pr->pr_usrreqs->pr_rcvd)(so, flags, l);
1579 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1580 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1581 			if (wakeup_state & SS_RESTARTSYS)
1582 				error = ERESTART;
1583 			else
1584 				error = sbwait(&so->so_rcv);
1585 			if (error != 0) {
1586 				sbunlock(&so->so_rcv);
1587 				sounlock(so);
1588 				splx(s);
1589 				return 0;
1590 			}
1591 			if ((m = so->so_rcv.sb_mb) != NULL)
1592 				nextrecord = m->m_nextpkt;
1593 			wakeup_state = so->so_state;
1594 		}
1595 	}
1596 
1597 	if (m && atomic) {
1598 		flags |= MSG_TRUNC;
1599 		if ((flags & MSG_PEEK) == 0)
1600 			(void) sbdroprecord(&so->so_rcv);
1601 	}
1602 	if ((flags & MSG_PEEK) == 0) {
1603 		if (m == NULL) {
1604 			/*
1605 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1606 			 * part makes sure sb_lastrecord is up-to-date if
1607 			 * there is still data in the socket buffer.
1608 			 */
1609 			so->so_rcv.sb_mb = nextrecord;
1610 			if (so->so_rcv.sb_mb == NULL) {
1611 				so->so_rcv.sb_mbtail = NULL;
1612 				so->so_rcv.sb_lastrecord = NULL;
1613 			} else if (nextrecord->m_nextpkt == NULL)
1614 				so->so_rcv.sb_lastrecord = nextrecord;
1615 		}
1616 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1617 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1618 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1619 			(*pr->pr_usrreqs->pr_rcvd)(so, flags, l);
1620 	}
1621 	if (orig_resid == uio->uio_resid && orig_resid &&
1622 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1623 		sbunlock(&so->so_rcv);
1624 		goto restart;
1625 	}
1626 
1627 	if (flagsp != NULL)
1628 		*flagsp |= flags;
1629  release:
1630 	sbunlock(&so->so_rcv);
1631 	sounlock(so);
1632 	splx(s);
1633 	return error;
1634 }
1635 
1636 int
1637 soshutdown(struct socket *so, int how)
1638 {
1639 	const struct protosw	*pr;
1640 	int	error;
1641 
1642 	KASSERT(solocked(so));
1643 
1644 	pr = so->so_proto;
1645 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1646 		return (EINVAL);
1647 
1648 	if (how == SHUT_RD || how == SHUT_RDWR) {
1649 		sorflush(so);
1650 		error = 0;
1651 	}
1652 	if (how == SHUT_WR || how == SHUT_RDWR)
1653 		error = (*pr->pr_usrreqs->pr_shutdown)(so);
1654 
1655 	return error;
1656 }
1657 
1658 void
1659 sorestart(struct socket *so)
1660 {
1661 	/*
1662 	 * An application has called close() on an fd on which another
1663 	 * of its threads has called a socket system call.
1664 	 * Mark this and wake everyone up, and code that would block again
1665 	 * instead returns ERESTART.
1666 	 * On system call re-entry the fd is validated and EBADF returned.
1667 	 * Any other fd will block again on the 2nd syscall.
1668 	 */
1669 	solock(so);
1670 	so->so_state |= SS_RESTARTSYS;
1671 	cv_broadcast(&so->so_cv);
1672 	cv_broadcast(&so->so_snd.sb_cv);
1673 	cv_broadcast(&so->so_rcv.sb_cv);
1674 	sounlock(so);
1675 }
1676 
1677 void
1678 sorflush(struct socket *so)
1679 {
1680 	struct sockbuf	*sb, asb;
1681 	const struct protosw	*pr;
1682 
1683 	KASSERT(solocked(so));
1684 
1685 	sb = &so->so_rcv;
1686 	pr = so->so_proto;
1687 	socantrcvmore(so);
1688 	sb->sb_flags |= SB_NOINTR;
1689 	(void )sblock(sb, M_WAITOK);
1690 	sbunlock(sb);
1691 	asb = *sb;
1692 	/*
1693 	 * Clear most of the sockbuf structure, but leave some of the
1694 	 * fields valid.
1695 	 */
1696 	memset(&sb->sb_startzero, 0,
1697 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1698 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) {
1699 		sounlock(so);
1700 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1701 		solock(so);
1702 	}
1703 	sbrelease(&asb, so);
1704 }
1705 
1706 /*
1707  * internal set SOL_SOCKET options
1708  */
1709 static int
1710 sosetopt1(struct socket *so, const struct sockopt *sopt)
1711 {
1712 	int error = EINVAL, opt;
1713 	int optval = 0; /* XXX: gcc */
1714 	struct linger l;
1715 	struct timeval tv;
1716 
1717 	switch ((opt = sopt->sopt_name)) {
1718 
1719 	case SO_ACCEPTFILTER:
1720 		error = accept_filt_setopt(so, sopt);
1721 		KASSERT(solocked(so));
1722 		break;
1723 
1724   	case SO_LINGER:
1725  		error = sockopt_get(sopt, &l, sizeof(l));
1726 		solock(so);
1727  		if (error)
1728  			break;
1729  		if (l.l_linger < 0 || l.l_linger > USHRT_MAX ||
1730  		    l.l_linger > (INT_MAX / hz)) {
1731 			error = EDOM;
1732 			break;
1733 		}
1734  		so->so_linger = l.l_linger;
1735  		if (l.l_onoff)
1736  			so->so_options |= SO_LINGER;
1737  		else
1738  			so->so_options &= ~SO_LINGER;
1739    		break;
1740 
1741 	case SO_DEBUG:
1742 	case SO_KEEPALIVE:
1743 	case SO_DONTROUTE:
1744 	case SO_USELOOPBACK:
1745 	case SO_BROADCAST:
1746 	case SO_REUSEADDR:
1747 	case SO_REUSEPORT:
1748 	case SO_OOBINLINE:
1749 	case SO_TIMESTAMP:
1750 	case SO_NOSIGPIPE:
1751 #ifdef SO_OTIMESTAMP
1752 	case SO_OTIMESTAMP:
1753 #endif
1754 		error = sockopt_getint(sopt, &optval);
1755 		solock(so);
1756 		if (error)
1757 			break;
1758 		if (optval)
1759 			so->so_options |= opt;
1760 		else
1761 			so->so_options &= ~opt;
1762 		break;
1763 
1764 	case SO_SNDBUF:
1765 	case SO_RCVBUF:
1766 	case SO_SNDLOWAT:
1767 	case SO_RCVLOWAT:
1768 		error = sockopt_getint(sopt, &optval);
1769 		solock(so);
1770 		if (error)
1771 			break;
1772 
1773 		/*
1774 		 * Values < 1 make no sense for any of these
1775 		 * options, so disallow them.
1776 		 */
1777 		if (optval < 1) {
1778 			error = EINVAL;
1779 			break;
1780 		}
1781 
1782 		switch (opt) {
1783 		case SO_SNDBUF:
1784 			if (sbreserve(&so->so_snd, (u_long)optval, so) == 0) {
1785 				error = ENOBUFS;
1786 				break;
1787 			}
1788 			so->so_snd.sb_flags &= ~SB_AUTOSIZE;
1789 			break;
1790 
1791 		case SO_RCVBUF:
1792 			if (sbreserve(&so->so_rcv, (u_long)optval, so) == 0) {
1793 				error = ENOBUFS;
1794 				break;
1795 			}
1796 			so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1797 			break;
1798 
1799 		/*
1800 		 * Make sure the low-water is never greater than
1801 		 * the high-water.
1802 		 */
1803 		case SO_SNDLOWAT:
1804 			if (optval > so->so_snd.sb_hiwat)
1805 				optval = so->so_snd.sb_hiwat;
1806 
1807 			so->so_snd.sb_lowat = optval;
1808 			break;
1809 
1810 		case SO_RCVLOWAT:
1811 			if (optval > so->so_rcv.sb_hiwat)
1812 				optval = so->so_rcv.sb_hiwat;
1813 
1814 			so->so_rcv.sb_lowat = optval;
1815 			break;
1816 		}
1817 		break;
1818 
1819 #ifdef COMPAT_50
1820 	case SO_OSNDTIMEO:
1821 	case SO_ORCVTIMEO: {
1822 		struct timeval50 otv;
1823 		error = sockopt_get(sopt, &otv, sizeof(otv));
1824 		if (error) {
1825 			solock(so);
1826 			break;
1827 		}
1828 		timeval50_to_timeval(&otv, &tv);
1829 		opt = opt == SO_OSNDTIMEO ? SO_SNDTIMEO : SO_RCVTIMEO;
1830 		error = 0;
1831 		/*FALLTHROUGH*/
1832 	}
1833 #endif /* COMPAT_50 */
1834 
1835 	case SO_SNDTIMEO:
1836 	case SO_RCVTIMEO:
1837 		if (error)
1838 			error = sockopt_get(sopt, &tv, sizeof(tv));
1839 		solock(so);
1840 		if (error)
1841 			break;
1842 
1843 		if (tv.tv_sec > (INT_MAX - tv.tv_usec / tick) / hz) {
1844 			error = EDOM;
1845 			break;
1846 		}
1847 
1848 		optval = tv.tv_sec * hz + tv.tv_usec / tick;
1849 		if (optval == 0 && tv.tv_usec != 0)
1850 			optval = 1;
1851 
1852 		switch (opt) {
1853 		case SO_SNDTIMEO:
1854 			so->so_snd.sb_timeo = optval;
1855 			break;
1856 		case SO_RCVTIMEO:
1857 			so->so_rcv.sb_timeo = optval;
1858 			break;
1859 		}
1860 		break;
1861 
1862 	default:
1863 		solock(so);
1864 		error = ENOPROTOOPT;
1865 		break;
1866 	}
1867 	KASSERT(solocked(so));
1868 	return error;
1869 }
1870 
1871 int
1872 sosetopt(struct socket *so, struct sockopt *sopt)
1873 {
1874 	int error, prerr;
1875 
1876 	if (sopt->sopt_level == SOL_SOCKET) {
1877 		error = sosetopt1(so, sopt);
1878 		KASSERT(solocked(so));
1879 	} else {
1880 		error = ENOPROTOOPT;
1881 		solock(so);
1882 	}
1883 
1884 	if ((error == 0 || error == ENOPROTOOPT) &&
1885 	    so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) {
1886 		/* give the protocol stack a shot */
1887 		prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, sopt);
1888 		if (prerr == 0)
1889 			error = 0;
1890 		else if (prerr != ENOPROTOOPT)
1891 			error = prerr;
1892 	}
1893 	sounlock(so);
1894 	return error;
1895 }
1896 
1897 /*
1898  * so_setsockopt() is a wrapper providing a sockopt structure for sosetopt()
1899  */
1900 int
1901 so_setsockopt(struct lwp *l, struct socket *so, int level, int name,
1902     const void *val, size_t valsize)
1903 {
1904 	struct sockopt sopt;
1905 	int error;
1906 
1907 	KASSERT(valsize == 0 || val != NULL);
1908 
1909 	sockopt_init(&sopt, level, name, valsize);
1910 	sockopt_set(&sopt, val, valsize);
1911 
1912 	error = sosetopt(so, &sopt);
1913 
1914 	sockopt_destroy(&sopt);
1915 
1916 	return error;
1917 }
1918 
1919 /*
1920  * internal get SOL_SOCKET options
1921  */
1922 static int
1923 sogetopt1(struct socket *so, struct sockopt *sopt)
1924 {
1925 	int error, optval, opt;
1926 	struct linger l;
1927 	struct timeval tv;
1928 
1929 	switch ((opt = sopt->sopt_name)) {
1930 
1931 	case SO_ACCEPTFILTER:
1932 		error = accept_filt_getopt(so, sopt);
1933 		break;
1934 
1935 	case SO_LINGER:
1936 		l.l_onoff = (so->so_options & SO_LINGER) ? 1 : 0;
1937 		l.l_linger = so->so_linger;
1938 
1939 		error = sockopt_set(sopt, &l, sizeof(l));
1940 		break;
1941 
1942 	case SO_USELOOPBACK:
1943 	case SO_DONTROUTE:
1944 	case SO_DEBUG:
1945 	case SO_KEEPALIVE:
1946 	case SO_REUSEADDR:
1947 	case SO_REUSEPORT:
1948 	case SO_BROADCAST:
1949 	case SO_OOBINLINE:
1950 	case SO_TIMESTAMP:
1951 	case SO_NOSIGPIPE:
1952 #ifdef SO_OTIMESTAMP
1953 	case SO_OTIMESTAMP:
1954 #endif
1955 	case SO_ACCEPTCONN:
1956 		error = sockopt_setint(sopt, (so->so_options & opt) ? 1 : 0);
1957 		break;
1958 
1959 	case SO_TYPE:
1960 		error = sockopt_setint(sopt, so->so_type);
1961 		break;
1962 
1963 	case SO_ERROR:
1964 		error = sockopt_setint(sopt, so->so_error);
1965 		so->so_error = 0;
1966 		break;
1967 
1968 	case SO_SNDBUF:
1969 		error = sockopt_setint(sopt, so->so_snd.sb_hiwat);
1970 		break;
1971 
1972 	case SO_RCVBUF:
1973 		error = sockopt_setint(sopt, so->so_rcv.sb_hiwat);
1974 		break;
1975 
1976 	case SO_SNDLOWAT:
1977 		error = sockopt_setint(sopt, so->so_snd.sb_lowat);
1978 		break;
1979 
1980 	case SO_RCVLOWAT:
1981 		error = sockopt_setint(sopt, so->so_rcv.sb_lowat);
1982 		break;
1983 
1984 #ifdef COMPAT_50
1985 	case SO_OSNDTIMEO:
1986 	case SO_ORCVTIMEO: {
1987 		struct timeval50 otv;
1988 
1989 		optval = (opt == SO_OSNDTIMEO ?
1990 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1991 
1992 		otv.tv_sec = optval / hz;
1993 		otv.tv_usec = (optval % hz) * tick;
1994 
1995 		error = sockopt_set(sopt, &otv, sizeof(otv));
1996 		break;
1997 	}
1998 #endif /* COMPAT_50 */
1999 
2000 	case SO_SNDTIMEO:
2001 	case SO_RCVTIMEO:
2002 		optval = (opt == SO_SNDTIMEO ?
2003 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2004 
2005 		tv.tv_sec = optval / hz;
2006 		tv.tv_usec = (optval % hz) * tick;
2007 
2008 		error = sockopt_set(sopt, &tv, sizeof(tv));
2009 		break;
2010 
2011 	case SO_OVERFLOWED:
2012 		error = sockopt_setint(sopt, so->so_rcv.sb_overflowed);
2013 		break;
2014 
2015 	default:
2016 		error = ENOPROTOOPT;
2017 		break;
2018 	}
2019 
2020 	return (error);
2021 }
2022 
2023 int
2024 sogetopt(struct socket *so, struct sockopt *sopt)
2025 {
2026 	int		error;
2027 
2028 	solock(so);
2029 	if (sopt->sopt_level != SOL_SOCKET) {
2030 		if (so->so_proto && so->so_proto->pr_ctloutput) {
2031 			error = ((*so->so_proto->pr_ctloutput)
2032 			    (PRCO_GETOPT, so, sopt));
2033 		} else
2034 			error = (ENOPROTOOPT);
2035 	} else {
2036 		error = sogetopt1(so, sopt);
2037 	}
2038 	sounlock(so);
2039 	return (error);
2040 }
2041 
2042 /*
2043  * alloc sockopt data buffer buffer
2044  *	- will be released at destroy
2045  */
2046 static int
2047 sockopt_alloc(struct sockopt *sopt, size_t len, km_flag_t kmflag)
2048 {
2049 
2050 	KASSERT(sopt->sopt_size == 0);
2051 
2052 	if (len > sizeof(sopt->sopt_buf)) {
2053 		sopt->sopt_data = kmem_zalloc(len, kmflag);
2054 		if (sopt->sopt_data == NULL)
2055 			return ENOMEM;
2056 	} else
2057 		sopt->sopt_data = sopt->sopt_buf;
2058 
2059 	sopt->sopt_size = len;
2060 	return 0;
2061 }
2062 
2063 /*
2064  * initialise sockopt storage
2065  *	- MAY sleep during allocation
2066  */
2067 void
2068 sockopt_init(struct sockopt *sopt, int level, int name, size_t size)
2069 {
2070 
2071 	memset(sopt, 0, sizeof(*sopt));
2072 
2073 	sopt->sopt_level = level;
2074 	sopt->sopt_name = name;
2075 	(void)sockopt_alloc(sopt, size, KM_SLEEP);
2076 }
2077 
2078 /*
2079  * destroy sockopt storage
2080  *	- will release any held memory references
2081  */
2082 void
2083 sockopt_destroy(struct sockopt *sopt)
2084 {
2085 
2086 	if (sopt->sopt_data != sopt->sopt_buf)
2087 		kmem_free(sopt->sopt_data, sopt->sopt_size);
2088 
2089 	memset(sopt, 0, sizeof(*sopt));
2090 }
2091 
2092 /*
2093  * set sockopt value
2094  *	- value is copied into sockopt
2095  * 	- memory is allocated when necessary, will not sleep
2096  */
2097 int
2098 sockopt_set(struct sockopt *sopt, const void *buf, size_t len)
2099 {
2100 	int error;
2101 
2102 	if (sopt->sopt_size == 0) {
2103 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2104 		if (error)
2105 			return error;
2106 	}
2107 
2108 	KASSERT(sopt->sopt_size == len);
2109 	memcpy(sopt->sopt_data, buf, len);
2110 	return 0;
2111 }
2112 
2113 /*
2114  * common case of set sockopt integer value
2115  */
2116 int
2117 sockopt_setint(struct sockopt *sopt, int val)
2118 {
2119 
2120 	return sockopt_set(sopt, &val, sizeof(int));
2121 }
2122 
2123 /*
2124  * get sockopt value
2125  *	- correct size must be given
2126  */
2127 int
2128 sockopt_get(const struct sockopt *sopt, void *buf, size_t len)
2129 {
2130 
2131 	if (sopt->sopt_size != len)
2132 		return EINVAL;
2133 
2134 	memcpy(buf, sopt->sopt_data, len);
2135 	return 0;
2136 }
2137 
2138 /*
2139  * common case of get sockopt integer value
2140  */
2141 int
2142 sockopt_getint(const struct sockopt *sopt, int *valp)
2143 {
2144 
2145 	return sockopt_get(sopt, valp, sizeof(int));
2146 }
2147 
2148 /*
2149  * set sockopt value from mbuf
2150  *	- ONLY for legacy code
2151  *	- mbuf is released by sockopt
2152  *	- will not sleep
2153  */
2154 int
2155 sockopt_setmbuf(struct sockopt *sopt, struct mbuf *m)
2156 {
2157 	size_t len;
2158 	int error;
2159 
2160 	len = m_length(m);
2161 
2162 	if (sopt->sopt_size == 0) {
2163 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2164 		if (error)
2165 			return error;
2166 	}
2167 
2168 	KASSERT(sopt->sopt_size == len);
2169 	m_copydata(m, 0, len, sopt->sopt_data);
2170 	m_freem(m);
2171 
2172 	return 0;
2173 }
2174 
2175 /*
2176  * get sockopt value into mbuf
2177  *	- ONLY for legacy code
2178  *	- mbuf to be released by the caller
2179  *	- will not sleep
2180  */
2181 struct mbuf *
2182 sockopt_getmbuf(const struct sockopt *sopt)
2183 {
2184 	struct mbuf *m;
2185 
2186 	if (sopt->sopt_size > MCLBYTES)
2187 		return NULL;
2188 
2189 	m = m_get(M_DONTWAIT, MT_SOOPTS);
2190 	if (m == NULL)
2191 		return NULL;
2192 
2193 	if (sopt->sopt_size > MLEN) {
2194 		MCLGET(m, M_DONTWAIT);
2195 		if ((m->m_flags & M_EXT) == 0) {
2196 			m_free(m);
2197 			return NULL;
2198 		}
2199 	}
2200 
2201 	memcpy(mtod(m, void *), sopt->sopt_data, sopt->sopt_size);
2202 	m->m_len = sopt->sopt_size;
2203 
2204 	return m;
2205 }
2206 
2207 void
2208 sohasoutofband(struct socket *so)
2209 {
2210 
2211 	fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
2212 	selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, NOTE_SUBMIT);
2213 }
2214 
2215 static void
2216 filt_sordetach(struct knote *kn)
2217 {
2218 	struct socket	*so;
2219 
2220 	so = ((file_t *)kn->kn_obj)->f_socket;
2221 	solock(so);
2222 	SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
2223 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
2224 		so->so_rcv.sb_flags &= ~SB_KNOTE;
2225 	sounlock(so);
2226 }
2227 
2228 /*ARGSUSED*/
2229 static int
2230 filt_soread(struct knote *kn, long hint)
2231 {
2232 	struct socket	*so;
2233 	int rv;
2234 
2235 	so = ((file_t *)kn->kn_obj)->f_socket;
2236 	if (hint != NOTE_SUBMIT)
2237 		solock(so);
2238 	kn->kn_data = so->so_rcv.sb_cc;
2239 	if (so->so_state & SS_CANTRCVMORE) {
2240 		kn->kn_flags |= EV_EOF;
2241 		kn->kn_fflags = so->so_error;
2242 		rv = 1;
2243 	} else if (so->so_error)	/* temporary udp error */
2244 		rv = 1;
2245 	else if (kn->kn_sfflags & NOTE_LOWAT)
2246 		rv = (kn->kn_data >= kn->kn_sdata);
2247 	else
2248 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
2249 	if (hint != NOTE_SUBMIT)
2250 		sounlock(so);
2251 	return rv;
2252 }
2253 
2254 static void
2255 filt_sowdetach(struct knote *kn)
2256 {
2257 	struct socket	*so;
2258 
2259 	so = ((file_t *)kn->kn_obj)->f_socket;
2260 	solock(so);
2261 	SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
2262 	if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
2263 		so->so_snd.sb_flags &= ~SB_KNOTE;
2264 	sounlock(so);
2265 }
2266 
2267 /*ARGSUSED*/
2268 static int
2269 filt_sowrite(struct knote *kn, long hint)
2270 {
2271 	struct socket	*so;
2272 	int rv;
2273 
2274 	so = ((file_t *)kn->kn_obj)->f_socket;
2275 	if (hint != NOTE_SUBMIT)
2276 		solock(so);
2277 	kn->kn_data = sbspace(&so->so_snd);
2278 	if (so->so_state & SS_CANTSENDMORE) {
2279 		kn->kn_flags |= EV_EOF;
2280 		kn->kn_fflags = so->so_error;
2281 		rv = 1;
2282 	} else if (so->so_error)	/* temporary udp error */
2283 		rv = 1;
2284 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2285 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
2286 		rv = 0;
2287 	else if (kn->kn_sfflags & NOTE_LOWAT)
2288 		rv = (kn->kn_data >= kn->kn_sdata);
2289 	else
2290 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
2291 	if (hint != NOTE_SUBMIT)
2292 		sounlock(so);
2293 	return rv;
2294 }
2295 
2296 /*ARGSUSED*/
2297 static int
2298 filt_solisten(struct knote *kn, long hint)
2299 {
2300 	struct socket	*so;
2301 	int rv;
2302 
2303 	so = ((file_t *)kn->kn_obj)->f_socket;
2304 
2305 	/*
2306 	 * Set kn_data to number of incoming connections, not
2307 	 * counting partial (incomplete) connections.
2308 	 */
2309 	if (hint != NOTE_SUBMIT)
2310 		solock(so);
2311 	kn->kn_data = so->so_qlen;
2312 	rv = (kn->kn_data > 0);
2313 	if (hint != NOTE_SUBMIT)
2314 		sounlock(so);
2315 	return rv;
2316 }
2317 
2318 static const struct filterops solisten_filtops =
2319 	{ 1, NULL, filt_sordetach, filt_solisten };
2320 static const struct filterops soread_filtops =
2321 	{ 1, NULL, filt_sordetach, filt_soread };
2322 static const struct filterops sowrite_filtops =
2323 	{ 1, NULL, filt_sowdetach, filt_sowrite };
2324 
2325 int
2326 soo_kqfilter(struct file *fp, struct knote *kn)
2327 {
2328 	struct socket	*so;
2329 	struct sockbuf	*sb;
2330 
2331 	so = ((file_t *)kn->kn_obj)->f_socket;
2332 	solock(so);
2333 	switch (kn->kn_filter) {
2334 	case EVFILT_READ:
2335 		if (so->so_options & SO_ACCEPTCONN)
2336 			kn->kn_fop = &solisten_filtops;
2337 		else
2338 			kn->kn_fop = &soread_filtops;
2339 		sb = &so->so_rcv;
2340 		break;
2341 	case EVFILT_WRITE:
2342 		kn->kn_fop = &sowrite_filtops;
2343 		sb = &so->so_snd;
2344 		break;
2345 	default:
2346 		sounlock(so);
2347 		return (EINVAL);
2348 	}
2349 	SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
2350 	sb->sb_flags |= SB_KNOTE;
2351 	sounlock(so);
2352 	return (0);
2353 }
2354 
2355 static int
2356 sodopoll(struct socket *so, int events)
2357 {
2358 	int revents;
2359 
2360 	revents = 0;
2361 
2362 	if (events & (POLLIN | POLLRDNORM))
2363 		if (soreadable(so))
2364 			revents |= events & (POLLIN | POLLRDNORM);
2365 
2366 	if (events & (POLLOUT | POLLWRNORM))
2367 		if (sowritable(so))
2368 			revents |= events & (POLLOUT | POLLWRNORM);
2369 
2370 	if (events & (POLLPRI | POLLRDBAND))
2371 		if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
2372 			revents |= events & (POLLPRI | POLLRDBAND);
2373 
2374 	return revents;
2375 }
2376 
2377 int
2378 sopoll(struct socket *so, int events)
2379 {
2380 	int revents = 0;
2381 
2382 #ifndef DIAGNOSTIC
2383 	/*
2384 	 * Do a quick, unlocked check in expectation that the socket
2385 	 * will be ready for I/O.  Don't do this check if DIAGNOSTIC,
2386 	 * as the solocked() assertions will fail.
2387 	 */
2388 	if ((revents = sodopoll(so, events)) != 0)
2389 		return revents;
2390 #endif
2391 
2392 	solock(so);
2393 	if ((revents = sodopoll(so, events)) == 0) {
2394 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2395 			selrecord(curlwp, &so->so_rcv.sb_sel);
2396 			so->so_rcv.sb_flags |= SB_NOTIFY;
2397 		}
2398 
2399 		if (events & (POLLOUT | POLLWRNORM)) {
2400 			selrecord(curlwp, &so->so_snd.sb_sel);
2401 			so->so_snd.sb_flags |= SB_NOTIFY;
2402 		}
2403 	}
2404 	sounlock(so);
2405 
2406 	return revents;
2407 }
2408 
2409 
2410 #include <sys/sysctl.h>
2411 
2412 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO);
2413 static int sysctl_kern_sbmax(SYSCTLFN_PROTO);
2414 
2415 /*
2416  * sysctl helper routine for kern.somaxkva.  ensures that the given
2417  * value is not too small.
2418  * (XXX should we maybe make sure it's not too large as well?)
2419  */
2420 static int
2421 sysctl_kern_somaxkva(SYSCTLFN_ARGS)
2422 {
2423 	int error, new_somaxkva;
2424 	struct sysctlnode node;
2425 
2426 	new_somaxkva = somaxkva;
2427 	node = *rnode;
2428 	node.sysctl_data = &new_somaxkva;
2429 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2430 	if (error || newp == NULL)
2431 		return (error);
2432 
2433 	if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */
2434 		return (EINVAL);
2435 
2436 	mutex_enter(&so_pendfree_lock);
2437 	somaxkva = new_somaxkva;
2438 	cv_broadcast(&socurkva_cv);
2439 	mutex_exit(&so_pendfree_lock);
2440 
2441 	return (error);
2442 }
2443 
2444 /*
2445  * sysctl helper routine for kern.sbmax. Basically just ensures that
2446  * any new value is not too small.
2447  */
2448 static int
2449 sysctl_kern_sbmax(SYSCTLFN_ARGS)
2450 {
2451 	int error, new_sbmax;
2452 	struct sysctlnode node;
2453 
2454 	new_sbmax = sb_max;
2455 	node = *rnode;
2456 	node.sysctl_data = &new_sbmax;
2457 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2458 	if (error || newp == NULL)
2459 		return (error);
2460 
2461 	KERNEL_LOCK(1, NULL);
2462 	error = sb_max_set(new_sbmax);
2463 	KERNEL_UNLOCK_ONE(NULL);
2464 
2465 	return (error);
2466 }
2467 
2468 static void
2469 sysctl_kern_socket_setup(void)
2470 {
2471 
2472 	KASSERT(socket_sysctllog == NULL);
2473 
2474 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2475 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2476 		       CTLTYPE_INT, "somaxkva",
2477 		       SYSCTL_DESCR("Maximum amount of kernel memory to be "
2478 				    "used for socket buffers"),
2479 		       sysctl_kern_somaxkva, 0, NULL, 0,
2480 		       CTL_KERN, KERN_SOMAXKVA, CTL_EOL);
2481 
2482 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2483 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2484 		       CTLTYPE_INT, "sbmax",
2485 		       SYSCTL_DESCR("Maximum socket buffer size"),
2486 		       sysctl_kern_sbmax, 0, NULL, 0,
2487 		       CTL_KERN, KERN_SBMAX, CTL_EOL);
2488 }
2489