xref: /netbsd-src/sys/kern/uipc_socket.c (revision f3cfa6f6ce31685c6c4a758bc430e69eb99f50a4)
1 /*	$NetBSD: uipc_socket.c,v 1.279 2019/05/08 14:05:18 christos 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 1.279 2019/05/08 14:05:18 christos 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 #include <sys/compat_stub.h>
108 
109 #include <compat/sys/time.h>
110 #include <compat/sys/socket.h>
111 
112 #include <uvm/uvm_extern.h>
113 #include <uvm/uvm_loan.h>
114 #include <uvm/uvm_page.h>
115 
116 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
117 
118 extern const struct fileops socketops;
119 
120 static int	sooptions;
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 vaddr_t
209 sokvaalloc(vaddr_t sva, vsize_t len, struct socket *so)
210 {
211 	vaddr_t lva;
212 
213 	if (sokvareserve(so, len) == 0)
214 		return 0;
215 
216 	lva = uvm_km_alloc(kernel_map, len, atop(sva) & uvmexp.colormask,
217 	    UVM_KMF_COLORMATCH | UVM_KMF_VAONLY | UVM_KMF_WAITVA);
218 	if (lva == 0) {
219 		sokvaunreserve(len);
220 		return 0;
221 	}
222 
223 	return lva;
224 }
225 
226 /*
227  * sokvafree: free kva for loan.
228  */
229 void
230 sokvafree(vaddr_t sva, vsize_t len)
231 {
232 
233 	uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY);
234 	sokvaunreserve(len);
235 }
236 
237 static void
238 sodoloanfree(struct vm_page **pgs, void *buf, size_t size)
239 {
240 	vaddr_t sva, eva;
241 	vsize_t len;
242 	int npgs;
243 
244 	KASSERT(pgs != NULL);
245 
246 	eva = round_page((vaddr_t) buf + size);
247 	sva = trunc_page((vaddr_t) buf);
248 	len = eva - sva;
249 	npgs = len >> PAGE_SHIFT;
250 
251 	pmap_kremove(sva, len);
252 	pmap_update(pmap_kernel());
253 	uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE);
254 	sokvafree(sva, len);
255 }
256 
257 /*
258  * sopendfree_thread: free mbufs on "pendfree" list. Unlock and relock
259  * so_pendfree_lock when freeing mbufs.
260  */
261 static void
262 sopendfree_thread(void *v)
263 {
264 	struct mbuf *m, *next;
265 	size_t rv;
266 
267 	mutex_enter(&so_pendfree_lock);
268 
269 	for (;;) {
270 		rv = 0;
271 		while (so_pendfree != NULL) {
272 			m = so_pendfree;
273 			so_pendfree = NULL;
274 			mutex_exit(&so_pendfree_lock);
275 
276 			for (; m != NULL; m = next) {
277 				next = m->m_next;
278 				KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) ==
279 				    0);
280 				KASSERT(m->m_ext.ext_refcnt == 0);
281 
282 				rv += m->m_ext.ext_size;
283 				sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf,
284 				    m->m_ext.ext_size);
285 				pool_cache_put(mb_cache, m);
286 			}
287 
288 			mutex_enter(&so_pendfree_lock);
289 		}
290 		if (rv)
291 			cv_broadcast(&socurkva_cv);
292 		cv_wait(&pendfree_thread_cv, &so_pendfree_lock);
293 	}
294 	panic("sopendfree_thread");
295 	/* NOTREACHED */
296 }
297 
298 void
299 soloanfree(struct mbuf *m, void *buf, size_t size, void *arg)
300 {
301 
302 	KASSERT(m != NULL);
303 
304 	/*
305 	 * postpone freeing mbuf.
306 	 *
307 	 * we can't do it in interrupt context
308 	 * because we need to put kva back to kernel_map.
309 	 */
310 
311 	mutex_enter(&so_pendfree_lock);
312 	m->m_next = so_pendfree;
313 	so_pendfree = m;
314 	cv_signal(&pendfree_thread_cv);
315 	mutex_exit(&so_pendfree_lock);
316 }
317 
318 static long
319 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space)
320 {
321 	struct iovec *iov = uio->uio_iov;
322 	vaddr_t sva, eva;
323 	vsize_t len;
324 	vaddr_t lva;
325 	int npgs, error;
326 	vaddr_t va;
327 	int i;
328 
329 	if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace))
330 		return 0;
331 
332 	if (iov->iov_len < (size_t) space)
333 		space = iov->iov_len;
334 	if (space > SOCK_LOAN_CHUNK)
335 		space = SOCK_LOAN_CHUNK;
336 
337 	eva = round_page((vaddr_t) iov->iov_base + space);
338 	sva = trunc_page((vaddr_t) iov->iov_base);
339 	len = eva - sva;
340 	npgs = len >> PAGE_SHIFT;
341 
342 	KASSERT(npgs <= M_EXT_MAXPAGES);
343 
344 	lva = sokvaalloc(sva, len, so);
345 	if (lva == 0)
346 		return 0;
347 
348 	error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len,
349 	    m->m_ext.ext_pgs, UVM_LOAN_TOPAGE);
350 	if (error) {
351 		sokvafree(lva, len);
352 		return 0;
353 	}
354 
355 	for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE)
356 		pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
357 		    VM_PROT_READ, 0);
358 	pmap_update(pmap_kernel());
359 
360 	lva += (vaddr_t) iov->iov_base & PAGE_MASK;
361 
362 	MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so);
363 	m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
364 
365 	uio->uio_resid -= space;
366 	/* uio_offset not updated, not set/used for write(2) */
367 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space;
368 	uio->uio_iov->iov_len -= space;
369 	if (uio->uio_iov->iov_len == 0) {
370 		uio->uio_iov++;
371 		uio->uio_iovcnt--;
372 	}
373 
374 	return space;
375 }
376 
377 static int
378 socket_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
379     void *arg0, void *arg1, void *arg2, void *arg3)
380 {
381 	int result;
382 	enum kauth_network_req req;
383 
384 	result = KAUTH_RESULT_DEFER;
385 	req = (enum kauth_network_req)arg0;
386 
387 	if ((action != KAUTH_NETWORK_SOCKET) &&
388 	    (action != KAUTH_NETWORK_BIND))
389 		return result;
390 
391 	switch (req) {
392 	case KAUTH_REQ_NETWORK_BIND_PORT:
393 		result = KAUTH_RESULT_ALLOW;
394 		break;
395 
396 	case KAUTH_REQ_NETWORK_SOCKET_DROP: {
397 		/* Normal users can only drop their own connections. */
398 		struct socket *so = (struct socket *)arg1;
399 
400 		if (so->so_cred && proc_uidmatch(cred, so->so_cred) == 0)
401 			result = KAUTH_RESULT_ALLOW;
402 
403 		break;
404 		}
405 
406 	case KAUTH_REQ_NETWORK_SOCKET_OPEN:
407 		/* We allow "raw" routing/bluetooth sockets to anyone. */
408 		switch ((u_long)arg1) {
409 		case PF_ROUTE:
410 		case PF_OROUTE:
411 		case PF_BLUETOOTH:
412 		case PF_CAN:
413 			result = KAUTH_RESULT_ALLOW;
414 			break;
415 		default:
416 			/* Privileged, let secmodel handle this. */
417 			if ((u_long)arg2 == SOCK_RAW)
418 				break;
419 			result = KAUTH_RESULT_ALLOW;
420 			break;
421 		}
422 		break;
423 
424 	case KAUTH_REQ_NETWORK_SOCKET_CANSEE:
425 		result = KAUTH_RESULT_ALLOW;
426 
427 		break;
428 
429 	default:
430 		break;
431 	}
432 
433 	return result;
434 }
435 
436 void
437 soinit(void)
438 {
439 
440 	sysctl_kern_socket_setup();
441 
442 	mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM);
443 	softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
444 	cv_init(&socurkva_cv, "sokva");
445 	cv_init(&pendfree_thread_cv, "sopendfr");
446 	soinit2();
447 
448 	/* Set the initial adjusted socket buffer size. */
449 	if (sb_max_set(sb_max))
450 		panic("bad initial sb_max value: %lu", sb_max);
451 
452 	socket_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK,
453 	    socket_listener_cb, NULL);
454 }
455 
456 void
457 soinit1(void)
458 {
459 	int error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL,
460 	    sopendfree_thread, NULL, &sopendfree_lwp, "sopendfree");
461 	if (error)
462 		panic("soinit1 %d", error);
463 }
464 
465 /*
466  * socreate: create a new socket of the specified type and the protocol.
467  *
468  * => Caller may specify another socket for lock sharing (must not be held).
469  * => Returns the new socket without lock held.
470  */
471 int
472 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l,
473     struct socket *lockso)
474 {
475 	const struct protosw *prp;
476 	struct socket *so;
477 	uid_t uid;
478 	int error;
479 	kmutex_t *lock;
480 
481 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET,
482 	    KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type),
483 	    KAUTH_ARG(proto));
484 	if (error != 0)
485 		return error;
486 
487 	if (proto)
488 		prp = pffindproto(dom, proto, type);
489 	else
490 		prp = pffindtype(dom, type);
491 	if (prp == NULL) {
492 		/* no support for domain */
493 		if (pffinddomain(dom) == 0)
494 			return EAFNOSUPPORT;
495 		/* no support for socket type */
496 		if (proto == 0 && type != 0)
497 			return EPROTOTYPE;
498 		return EPROTONOSUPPORT;
499 	}
500 	if (prp->pr_usrreqs == NULL)
501 		return EPROTONOSUPPORT;
502 	if (prp->pr_type != type)
503 		return EPROTOTYPE;
504 
505 	so = soget(true);
506 	so->so_type = type;
507 	so->so_proto = prp;
508 	so->so_send = sosend;
509 	so->so_receive = soreceive;
510 	so->so_options = sooptions;
511 #ifdef MBUFTRACE
512 	so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner;
513 	so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner;
514 	so->so_mowner = &prp->pr_domain->dom_mowner;
515 #endif
516 	uid = kauth_cred_geteuid(l->l_cred);
517 	so->so_uidinfo = uid_find(uid);
518 	so->so_cpid = l->l_proc->p_pid;
519 
520 	/*
521 	 * Lock assigned and taken during PCB attach, unless we share
522 	 * the lock with another socket, e.g. socketpair(2) case.
523 	 */
524 	if (lockso) {
525 		lock = lockso->so_lock;
526 		so->so_lock = lock;
527 		mutex_obj_hold(lock);
528 		mutex_enter(lock);
529 	}
530 
531 	/* Attach the PCB (returns with the socket lock held). */
532 	error = (*prp->pr_usrreqs->pr_attach)(so, proto);
533 	KASSERT(solocked(so));
534 
535 	if (error) {
536 		KASSERT(so->so_pcb == NULL);
537 		so->so_state |= SS_NOFDREF;
538 		sofree(so);
539 		return error;
540 	}
541 	so->so_cred = kauth_cred_dup(l->l_cred);
542 	sounlock(so);
543 
544 	*aso = so;
545 	return 0;
546 }
547 
548 /*
549  * fsocreate: create a socket and a file descriptor associated with it.
550  *
551  * => On success, write file descriptor to fdout and return zero.
552  * => On failure, return non-zero; *fdout will be undefined.
553  */
554 int
555 fsocreate(int domain, struct socket **sop, int type, int proto, int *fdout)
556 {
557 	lwp_t *l = curlwp;
558 	int error, fd, flags;
559 	struct socket *so;
560 	struct file *fp;
561 
562 	if ((error = fd_allocfile(&fp, &fd)) != 0) {
563 		return error;
564 	}
565 	flags = type & SOCK_FLAGS_MASK;
566 	fd_set_exclose(l, fd, (flags & SOCK_CLOEXEC) != 0);
567 	fp->f_flag = FREAD|FWRITE|((flags & SOCK_NONBLOCK) ? FNONBLOCK : 0)|
568 	    ((flags & SOCK_NOSIGPIPE) ? FNOSIGPIPE : 0);
569 	fp->f_type = DTYPE_SOCKET;
570 	fp->f_ops = &socketops;
571 
572 	type &= ~SOCK_FLAGS_MASK;
573 	error = socreate(domain, &so, type, proto, l, NULL);
574 	if (error) {
575 		fd_abort(curproc, fp, fd);
576 		return error;
577 	}
578 	if (flags & SOCK_NONBLOCK) {
579 		so->so_state |= SS_NBIO;
580 	}
581 	fp->f_socket = so;
582 	fd_affix(curproc, fp, fd);
583 
584 	if (sop != NULL) {
585 		*sop = so;
586 	}
587 	*fdout = fd;
588 	return error;
589 }
590 
591 int
592 sofamily(const struct socket *so)
593 {
594 	const struct protosw *pr;
595 	const struct domain *dom;
596 
597 	if ((pr = so->so_proto) == NULL)
598 		return AF_UNSPEC;
599 	if ((dom = pr->pr_domain) == NULL)
600 		return AF_UNSPEC;
601 	return dom->dom_family;
602 }
603 
604 int
605 sobind(struct socket *so, struct sockaddr *nam, struct lwp *l)
606 {
607 	int error;
608 
609 	solock(so);
610 	if (nam->sa_family != so->so_proto->pr_domain->dom_family) {
611 		sounlock(so);
612 		return EAFNOSUPPORT;
613 	}
614 	error = (*so->so_proto->pr_usrreqs->pr_bind)(so, nam, l);
615 	sounlock(so);
616 	return error;
617 }
618 
619 int
620 solisten(struct socket *so, int backlog, struct lwp *l)
621 {
622 	int error;
623 	short oldopt, oldqlimit;
624 
625 	solock(so);
626 	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
627 	    SS_ISDISCONNECTING)) != 0) {
628 		sounlock(so);
629 		return EINVAL;
630 	}
631 	oldopt = so->so_options;
632 	oldqlimit = so->so_qlimit;
633 	if (TAILQ_EMPTY(&so->so_q))
634 		so->so_options |= SO_ACCEPTCONN;
635 	if (backlog < 0)
636 		backlog = 0;
637 	so->so_qlimit = uimin(backlog, somaxconn);
638 
639 	error = (*so->so_proto->pr_usrreqs->pr_listen)(so, l);
640 	if (error != 0) {
641 		so->so_options = oldopt;
642 		so->so_qlimit = oldqlimit;
643 		sounlock(so);
644 		return error;
645 	}
646 	sounlock(so);
647 	return 0;
648 }
649 
650 void
651 sofree(struct socket *so)
652 {
653 	u_int refs;
654 
655 	KASSERT(solocked(so));
656 
657 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
658 		sounlock(so);
659 		return;
660 	}
661 	if (so->so_head) {
662 		/*
663 		 * We must not decommission a socket that's on the accept(2)
664 		 * queue.  If we do, then accept(2) may hang after select(2)
665 		 * indicated that the listening socket was ready.
666 		 */
667 		if (!soqremque(so, 0)) {
668 			sounlock(so);
669 			return;
670 		}
671 	}
672 	if (so->so_rcv.sb_hiwat)
673 		(void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0,
674 		    RLIM_INFINITY);
675 	if (so->so_snd.sb_hiwat)
676 		(void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0,
677 		    RLIM_INFINITY);
678 	sbrelease(&so->so_snd, so);
679 	KASSERT(!cv_has_waiters(&so->so_cv));
680 	KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv));
681 	KASSERT(!cv_has_waiters(&so->so_snd.sb_cv));
682 	sorflush(so);
683 	refs = so->so_aborting;	/* XXX */
684 	/* Remove acccept filter if one is present. */
685 	if (so->so_accf != NULL)
686 		(void)accept_filt_clear(so);
687 	sounlock(so);
688 	if (refs == 0)		/* XXX */
689 		soput(so);
690 }
691 
692 /*
693  * soclose: close a socket on last file table reference removal.
694  * Initiate disconnect if connected.  Free socket when disconnect complete.
695  */
696 int
697 soclose(struct socket *so)
698 {
699 	struct socket *so2;
700 	int error = 0;
701 
702 	solock(so);
703 	if (so->so_options & SO_ACCEPTCONN) {
704 		for (;;) {
705 			if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) {
706 				KASSERT(solocked2(so, so2));
707 				(void) soqremque(so2, 0);
708 				/* soabort drops the lock. */
709 				(void) soabort(so2);
710 				solock(so);
711 				continue;
712 			}
713 			if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) {
714 				KASSERT(solocked2(so, so2));
715 				(void) soqremque(so2, 1);
716 				/* soabort drops the lock. */
717 				(void) soabort(so2);
718 				solock(so);
719 				continue;
720 			}
721 			break;
722 		}
723 	}
724 	if (so->so_pcb == NULL)
725 		goto discard;
726 	if (so->so_state & SS_ISCONNECTED) {
727 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
728 			error = sodisconnect(so);
729 			if (error)
730 				goto drop;
731 		}
732 		if (so->so_options & SO_LINGER) {
733 			if ((so->so_state & (SS_ISDISCONNECTING|SS_NBIO)) ==
734 			    (SS_ISDISCONNECTING|SS_NBIO))
735 				goto drop;
736 			while (so->so_state & SS_ISCONNECTED) {
737 				error = sowait(so, true, so->so_linger * hz);
738 				if (error)
739 					break;
740 			}
741 		}
742 	}
743  drop:
744 	if (so->so_pcb) {
745 		KASSERT(solocked(so));
746 		(*so->so_proto->pr_usrreqs->pr_detach)(so);
747 	}
748  discard:
749 	KASSERT((so->so_state & SS_NOFDREF) == 0);
750 	kauth_cred_free(so->so_cred);
751 	so->so_cred = NULL;
752 	so->so_state |= SS_NOFDREF;
753 	sofree(so);
754 	return error;
755 }
756 
757 /*
758  * Must be called with the socket locked..  Will return with it unlocked.
759  */
760 int
761 soabort(struct socket *so)
762 {
763 	u_int refs;
764 	int error;
765 
766 	KASSERT(solocked(so));
767 	KASSERT(so->so_head == NULL);
768 
769 	so->so_aborting++;		/* XXX */
770 	error = (*so->so_proto->pr_usrreqs->pr_abort)(so);
771 	refs = --so->so_aborting;	/* XXX */
772 	if (error || (refs == 0)) {
773 		sofree(so);
774 	} else {
775 		sounlock(so);
776 	}
777 	return error;
778 }
779 
780 int
781 soaccept(struct socket *so, struct sockaddr *nam)
782 {
783 	int error;
784 
785 	KASSERT(solocked(so));
786 	KASSERT((so->so_state & SS_NOFDREF) != 0);
787 
788 	so->so_state &= ~SS_NOFDREF;
789 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
790 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
791 		error = (*so->so_proto->pr_usrreqs->pr_accept)(so, nam);
792 	else
793 		error = ECONNABORTED;
794 
795 	return error;
796 }
797 
798 int
799 soconnect(struct socket *so, struct sockaddr *nam, struct lwp *l)
800 {
801 	int error;
802 
803 	KASSERT(solocked(so));
804 
805 	if (so->so_options & SO_ACCEPTCONN)
806 		return EOPNOTSUPP;
807 	/*
808 	 * If protocol is connection-based, can only connect once.
809 	 * Otherwise, if connected, try to disconnect first.
810 	 * This allows user to disconnect by connecting to, e.g.,
811 	 * a null address.
812 	 */
813 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
814 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
815 	    (error = sodisconnect(so)))) {
816 		error = EISCONN;
817 	} else {
818 		if (nam->sa_family != so->so_proto->pr_domain->dom_family) {
819 			return EAFNOSUPPORT;
820 		}
821 		error = (*so->so_proto->pr_usrreqs->pr_connect)(so, nam, l);
822 	}
823 
824 	return error;
825 }
826 
827 int
828 soconnect2(struct socket *so1, struct socket *so2)
829 {
830 	KASSERT(solocked2(so1, so2));
831 
832 	return (*so1->so_proto->pr_usrreqs->pr_connect2)(so1, so2);
833 }
834 
835 int
836 sodisconnect(struct socket *so)
837 {
838 	int error;
839 
840 	KASSERT(solocked(so));
841 
842 	if ((so->so_state & SS_ISCONNECTED) == 0) {
843 		error = ENOTCONN;
844 	} else if (so->so_state & SS_ISDISCONNECTING) {
845 		error = EALREADY;
846 	} else {
847 		error = (*so->so_proto->pr_usrreqs->pr_disconnect)(so);
848 	}
849 	return error;
850 }
851 
852 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
853 /*
854  * Send on a socket.
855  * If send must go all at once and message is larger than
856  * send buffering, then hard error.
857  * Lock against other senders.
858  * If must go all at once and not enough room now, then
859  * inform user that this would block and do nothing.
860  * Otherwise, if nonblocking, send as much as possible.
861  * The data to be sent is described by "uio" if nonzero,
862  * otherwise by the mbuf chain "top" (which must be null
863  * if uio is not).  Data provided in mbuf chain must be small
864  * enough to send all at once.
865  *
866  * Returns nonzero on error, timeout or signal; callers
867  * must check for short counts if EINTR/ERESTART are returned.
868  * Data and control buffers are freed on return.
869  */
870 int
871 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
872 	struct mbuf *top, struct mbuf *control, int flags, struct lwp *l)
873 {
874 	struct mbuf **mp, *m;
875 	long space, len, resid, clen, mlen;
876 	int error, s, dontroute, atomic;
877 	short wakeup_state = 0;
878 
879 	clen = 0;
880 
881 	/*
882 	 * solock() provides atomicity of access.  splsoftnet() prevents
883 	 * protocol processing soft interrupts from interrupting us and
884 	 * blocking (expensive).
885 	 */
886 	s = splsoftnet();
887 	solock(so);
888 	atomic = sosendallatonce(so) || top;
889 	if (uio)
890 		resid = uio->uio_resid;
891 	else
892 		resid = top->m_pkthdr.len;
893 	/*
894 	 * In theory resid should be unsigned.
895 	 * However, space must be signed, as it might be less than 0
896 	 * if we over-committed, and we must use a signed comparison
897 	 * of space and resid.  On the other hand, a negative resid
898 	 * causes us to loop sending 0-length segments to the protocol.
899 	 */
900 	if (resid < 0) {
901 		error = EINVAL;
902 		goto out;
903 	}
904 	dontroute =
905 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
906 	    (so->so_proto->pr_flags & PR_ATOMIC);
907 	l->l_ru.ru_msgsnd++;
908 	if (control)
909 		clen = control->m_len;
910  restart:
911 	if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
912 		goto out;
913 	do {
914 		if (so->so_state & SS_CANTSENDMORE) {
915 			error = EPIPE;
916 			goto release;
917 		}
918 		if (so->so_error) {
919 			error = so->so_error;
920 			so->so_error = 0;
921 			goto release;
922 		}
923 		if ((so->so_state & SS_ISCONNECTED) == 0) {
924 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
925 				if (resid || clen == 0) {
926 					error = ENOTCONN;
927 					goto release;
928 				}
929 			} else if (addr == NULL) {
930 				error = EDESTADDRREQ;
931 				goto release;
932 			}
933 		}
934 		space = sbspace(&so->so_snd);
935 		if (flags & MSG_OOB)
936 			space += 1024;
937 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
938 		    clen > so->so_snd.sb_hiwat) {
939 			error = EMSGSIZE;
940 			goto release;
941 		}
942 		if (space < resid + clen &&
943 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
944 			if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
945 				error = EWOULDBLOCK;
946 				goto release;
947 			}
948 			sbunlock(&so->so_snd);
949 			if (wakeup_state & SS_RESTARTSYS) {
950 				error = ERESTART;
951 				goto out;
952 			}
953 			error = sbwait(&so->so_snd);
954 			if (error)
955 				goto out;
956 			wakeup_state = so->so_state;
957 			goto restart;
958 		}
959 		wakeup_state = 0;
960 		mp = &top;
961 		space -= clen;
962 		do {
963 			if (uio == NULL) {
964 				/*
965 				 * Data is prepackaged in "top".
966 				 */
967 				resid = 0;
968 				if (flags & MSG_EOR)
969 					top->m_flags |= M_EOR;
970 			} else do {
971 				sounlock(so);
972 				splx(s);
973 				if (top == NULL) {
974 					m = m_gethdr(M_WAIT, MT_DATA);
975 					mlen = MHLEN;
976 					m->m_pkthdr.len = 0;
977 					m_reset_rcvif(m);
978 				} else {
979 					m = m_get(M_WAIT, MT_DATA);
980 					mlen = MLEN;
981 				}
982 				MCLAIM(m, so->so_snd.sb_mowner);
983 				if (sock_loan_thresh >= 0 &&
984 				    uio->uio_iov->iov_len >= sock_loan_thresh &&
985 				    space >= sock_loan_thresh &&
986 				    (len = sosend_loan(so, uio, m,
987 						       space)) != 0) {
988 					SOSEND_COUNTER_INCR(&sosend_loan_big);
989 					space -= len;
990 					goto have_data;
991 				}
992 				if (resid >= MINCLSIZE && space >= MCLBYTES) {
993 					SOSEND_COUNTER_INCR(&sosend_copy_big);
994 					m_clget(m, M_DONTWAIT);
995 					if ((m->m_flags & M_EXT) == 0)
996 						goto nopages;
997 					mlen = MCLBYTES;
998 					if (atomic && top == 0) {
999 						len = lmin(MCLBYTES - max_hdr,
1000 						    resid);
1001 						m->m_data += max_hdr;
1002 					} else
1003 						len = lmin(MCLBYTES, resid);
1004 					space -= len;
1005 				} else {
1006  nopages:
1007 					SOSEND_COUNTER_INCR(&sosend_copy_small);
1008 					len = lmin(lmin(mlen, resid), space);
1009 					space -= len;
1010 					/*
1011 					 * For datagram protocols, leave room
1012 					 * for protocol headers in first mbuf.
1013 					 */
1014 					if (atomic && top == 0 && len < mlen)
1015 						m_align(m, len);
1016 				}
1017 				error = uiomove(mtod(m, void *), (int)len, uio);
1018  have_data:
1019 				resid = uio->uio_resid;
1020 				m->m_len = len;
1021 				*mp = m;
1022 				top->m_pkthdr.len += len;
1023 				s = splsoftnet();
1024 				solock(so);
1025 				if (error != 0)
1026 					goto release;
1027 				mp = &m->m_next;
1028 				if (resid <= 0) {
1029 					if (flags & MSG_EOR)
1030 						top->m_flags |= M_EOR;
1031 					break;
1032 				}
1033 			} while (space > 0 && atomic);
1034 
1035 			if (so->so_state & SS_CANTSENDMORE) {
1036 				error = EPIPE;
1037 				goto release;
1038 			}
1039 			if (dontroute)
1040 				so->so_options |= SO_DONTROUTE;
1041 			if (resid > 0)
1042 				so->so_state |= SS_MORETOCOME;
1043 			if (flags & MSG_OOB) {
1044 				error = (*so->so_proto->pr_usrreqs->pr_sendoob)(
1045 				    so, top, control);
1046 			} else {
1047 				error = (*so->so_proto->pr_usrreqs->pr_send)(so,
1048 				    top, addr, control, l);
1049 			}
1050 			if (dontroute)
1051 				so->so_options &= ~SO_DONTROUTE;
1052 			if (resid > 0)
1053 				so->so_state &= ~SS_MORETOCOME;
1054 			clen = 0;
1055 			control = NULL;
1056 			top = NULL;
1057 			mp = &top;
1058 			if (error != 0)
1059 				goto release;
1060 		} while (resid && space > 0);
1061 	} while (resid);
1062 
1063  release:
1064 	sbunlock(&so->so_snd);
1065  out:
1066 	sounlock(so);
1067 	splx(s);
1068 	if (top)
1069 		m_freem(top);
1070 	if (control)
1071 		m_freem(control);
1072 	return error;
1073 }
1074 
1075 /*
1076  * Following replacement or removal of the first mbuf on the first
1077  * mbuf chain of a socket buffer, push necessary state changes back
1078  * into the socket buffer so that other consumers see the values
1079  * consistently.  'nextrecord' is the caller's locally stored value of
1080  * the original value of sb->sb_mb->m_nextpkt which must be restored
1081  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
1082  */
1083 static void
1084 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
1085 {
1086 
1087 	KASSERT(solocked(sb->sb_so));
1088 
1089 	/*
1090 	 * First, update for the new value of nextrecord.  If necessary,
1091 	 * make it the first record.
1092 	 */
1093 	if (sb->sb_mb != NULL)
1094 		sb->sb_mb->m_nextpkt = nextrecord;
1095 	else
1096 		sb->sb_mb = nextrecord;
1097 
1098         /*
1099          * Now update any dependent socket buffer fields to reflect
1100          * the new state.  This is an inline of SB_EMPTY_FIXUP, with
1101          * the addition of a second clause that takes care of the
1102          * case where sb_mb has been updated, but remains the last
1103          * record.
1104          */
1105         if (sb->sb_mb == NULL) {
1106                 sb->sb_mbtail = NULL;
1107                 sb->sb_lastrecord = NULL;
1108         } else if (sb->sb_mb->m_nextpkt == NULL)
1109                 sb->sb_lastrecord = sb->sb_mb;
1110 }
1111 
1112 /*
1113  * Implement receive operations on a socket.
1114  *
1115  * We depend on the way that records are added to the sockbuf by sbappend*. In
1116  * particular, each record (mbufs linked through m_next) must begin with an
1117  * address if the protocol so specifies, followed by an optional mbuf or mbufs
1118  * containing ancillary data, and then zero or more mbufs of data.
1119  *
1120  * In order to avoid blocking network interrupts for the entire time here, we
1121  * splx() while doing the actual copy to user space. Although the sockbuf is
1122  * locked, new data may still be appended, and thus we must maintain
1123  * consistency of the sockbuf during that time.
1124  *
1125  * The caller may receive the data as a single mbuf chain by supplying an mbuf
1126  * **mp0 for use in returning the chain. The uio is then used only for the
1127  * count in uio_resid.
1128  */
1129 int
1130 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
1131     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1132 {
1133 	struct lwp *l = curlwp;
1134 	struct mbuf *m, **mp, *mt;
1135 	size_t len, offset, moff, orig_resid;
1136 	int atomic, flags, error, s, type;
1137 	const struct protosw *pr;
1138 	struct mbuf *nextrecord;
1139 	int mbuf_removed = 0;
1140 	const struct domain *dom;
1141 	short wakeup_state = 0;
1142 
1143 	pr = so->so_proto;
1144 	atomic = pr->pr_flags & PR_ATOMIC;
1145 	dom = pr->pr_domain;
1146 	mp = mp0;
1147 	type = 0;
1148 	orig_resid = uio->uio_resid;
1149 
1150 	if (paddr != NULL)
1151 		*paddr = NULL;
1152 	if (controlp != NULL)
1153 		*controlp = NULL;
1154 	if (flagsp != NULL)
1155 		flags = *flagsp &~ MSG_EOR;
1156 	else
1157 		flags = 0;
1158 
1159 	if (flags & MSG_OOB) {
1160 		m = m_get(M_WAIT, MT_DATA);
1161 		solock(so);
1162 		error = (*pr->pr_usrreqs->pr_recvoob)(so, m, flags & MSG_PEEK);
1163 		sounlock(so);
1164 		if (error)
1165 			goto bad;
1166 		do {
1167 			error = uiomove(mtod(m, void *),
1168 			    MIN(uio->uio_resid, m->m_len), uio);
1169 			m = m_free(m);
1170 		} while (uio->uio_resid > 0 && error == 0 && m);
1171 bad:
1172 		if (m != NULL)
1173 			m_freem(m);
1174 		return error;
1175 	}
1176 	if (mp != NULL)
1177 		*mp = NULL;
1178 
1179 	/*
1180 	 * solock() provides atomicity of access.  splsoftnet() prevents
1181 	 * protocol processing soft interrupts from interrupting us and
1182 	 * blocking (expensive).
1183 	 */
1184 	s = splsoftnet();
1185 	solock(so);
1186 restart:
1187 	if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) {
1188 		sounlock(so);
1189 		splx(s);
1190 		return error;
1191 	}
1192 	m = so->so_rcv.sb_mb;
1193 
1194 	/*
1195 	 * If we have less data than requested, block awaiting more
1196 	 * (subject to any timeout) if:
1197 	 *   1. the current count is less than the low water mark,
1198 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1199 	 *	receive operation at once if we block (resid <= hiwat), or
1200 	 *   3. MSG_DONTWAIT is not set.
1201 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1202 	 * we have to do the receive in sections, and thus risk returning
1203 	 * a short count if a timeout or signal occurs after we start.
1204 	 */
1205 	if (m == NULL ||
1206 	    ((flags & MSG_DONTWAIT) == 0 &&
1207 	     so->so_rcv.sb_cc < uio->uio_resid &&
1208 	     (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1209 	      ((flags & MSG_WAITALL) &&
1210 	       uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1211 	     m->m_nextpkt == NULL && !atomic)) {
1212 #ifdef DIAGNOSTIC
1213 		if (m == NULL && so->so_rcv.sb_cc)
1214 			panic("receive 1");
1215 #endif
1216 		if (so->so_error || so->so_rerror) {
1217 			if (m != NULL)
1218 				goto dontblock;
1219 			if (so->so_error) {
1220 				error = so->so_error;
1221 				so->so_error = 0;
1222 			} else {
1223 				error = so->so_rerror;
1224 				so->so_rerror = 0;
1225 			}
1226 			goto release;
1227 		}
1228 		if (so->so_state & SS_CANTRCVMORE) {
1229 			if (m != NULL)
1230 				goto dontblock;
1231 			else
1232 				goto release;
1233 		}
1234 		for (; m != NULL; m = m->m_next)
1235 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1236 				m = so->so_rcv.sb_mb;
1237 				goto dontblock;
1238 			}
1239 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1240 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1241 			error = ENOTCONN;
1242 			goto release;
1243 		}
1244 		if (uio->uio_resid == 0)
1245 			goto release;
1246 		if ((so->so_state & SS_NBIO) ||
1247 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1248 			error = EWOULDBLOCK;
1249 			goto release;
1250 		}
1251 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
1252 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
1253 		sbunlock(&so->so_rcv);
1254 		if (wakeup_state & SS_RESTARTSYS)
1255 			error = ERESTART;
1256 		else
1257 			error = sbwait(&so->so_rcv);
1258 		if (error != 0) {
1259 			sounlock(so);
1260 			splx(s);
1261 			return error;
1262 		}
1263 		wakeup_state = so->so_state;
1264 		goto restart;
1265 	}
1266 
1267 dontblock:
1268 	/*
1269 	 * On entry here, m points to the first record of the socket buffer.
1270 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1271 	 * pointer to the next record in the socket buffer.  We must keep the
1272 	 * various socket buffer pointers and local stack versions of the
1273 	 * pointers in sync, pushing out modifications before dropping the
1274 	 * socket lock, and re-reading them when picking it up.
1275 	 *
1276 	 * Otherwise, we will race with the network stack appending new data
1277 	 * or records onto the socket buffer by using inconsistent/stale
1278 	 * versions of the field, possibly resulting in socket buffer
1279 	 * corruption.
1280 	 *
1281 	 * By holding the high-level sblock(), we prevent simultaneous
1282 	 * readers from pulling off the front of the socket buffer.
1283 	 */
1284 	if (l != NULL)
1285 		l->l_ru.ru_msgrcv++;
1286 	KASSERT(m == so->so_rcv.sb_mb);
1287 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
1288 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
1289 	nextrecord = m->m_nextpkt;
1290 
1291 	if (pr->pr_flags & PR_ADDR) {
1292 		KASSERT(m->m_type == MT_SONAME);
1293 		orig_resid = 0;
1294 		if (flags & MSG_PEEK) {
1295 			if (paddr)
1296 				*paddr = m_copym(m, 0, m->m_len, M_DONTWAIT);
1297 			m = m->m_next;
1298 		} else {
1299 			sbfree(&so->so_rcv, m);
1300 			mbuf_removed = 1;
1301 			if (paddr != NULL) {
1302 				*paddr = m;
1303 				so->so_rcv.sb_mb = m->m_next;
1304 				m->m_next = NULL;
1305 				m = so->so_rcv.sb_mb;
1306 			} else {
1307 				m = so->so_rcv.sb_mb = m_free(m);
1308 			}
1309 			sbsync(&so->so_rcv, nextrecord);
1310 		}
1311 	}
1312 
1313 	if (pr->pr_flags & PR_ADDR_OPT) {
1314 		/*
1315 		 * For SCTP we may be getting a whole message OR a partial
1316 		 * delivery.
1317 		 */
1318 		if (m->m_type == MT_SONAME) {
1319 			orig_resid = 0;
1320 			if (flags & MSG_PEEK) {
1321 				if (paddr)
1322 					*paddr = m_copym(m, 0, m->m_len, M_DONTWAIT);
1323 				m = m->m_next;
1324 			} else {
1325 				sbfree(&so->so_rcv, m);
1326 				if (paddr) {
1327 					*paddr = m;
1328 					so->so_rcv.sb_mb = m->m_next;
1329 					m->m_next = 0;
1330 					m = so->so_rcv.sb_mb;
1331 				} else {
1332 					m = so->so_rcv.sb_mb = m_free(m);
1333 				}
1334 			}
1335 		}
1336 	}
1337 
1338 	/*
1339 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1340 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1341 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1342 	 * perform externalization (or freeing if controlp == NULL).
1343 	 */
1344 	if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) {
1345 		struct mbuf *cm = NULL, *cmn;
1346 		struct mbuf **cme = &cm;
1347 
1348 		do {
1349 			if (flags & MSG_PEEK) {
1350 				if (controlp != NULL) {
1351 					*controlp = m_copym(m, 0, m->m_len, M_DONTWAIT);
1352 					controlp = &(*controlp)->m_next;
1353 				}
1354 				m = m->m_next;
1355 			} else {
1356 				sbfree(&so->so_rcv, m);
1357 				so->so_rcv.sb_mb = m->m_next;
1358 				m->m_next = NULL;
1359 				*cme = m;
1360 				cme = &(*cme)->m_next;
1361 				m = so->so_rcv.sb_mb;
1362 			}
1363 		} while (m != NULL && m->m_type == MT_CONTROL);
1364 		if ((flags & MSG_PEEK) == 0)
1365 			sbsync(&so->so_rcv, nextrecord);
1366 
1367 		for (; cm != NULL; cm = cmn) {
1368 			cmn = cm->m_next;
1369 			cm->m_next = NULL;
1370 			type = mtod(cm, struct cmsghdr *)->cmsg_type;
1371 			if (controlp != NULL) {
1372 				if (dom->dom_externalize != NULL &&
1373 				    type == SCM_RIGHTS) {
1374 					sounlock(so);
1375 					splx(s);
1376 					error = (*dom->dom_externalize)(cm, l,
1377 					    (flags & MSG_CMSG_CLOEXEC) ?
1378 					    O_CLOEXEC : 0);
1379 					s = splsoftnet();
1380 					solock(so);
1381 				}
1382 				*controlp = cm;
1383 				while (*controlp != NULL)
1384 					controlp = &(*controlp)->m_next;
1385 			} else {
1386 				/*
1387 				 * Dispose of any SCM_RIGHTS message that went
1388 				 * through the read path rather than recv.
1389 				 */
1390 				if (dom->dom_dispose != NULL &&
1391 				    type == SCM_RIGHTS) {
1392 					sounlock(so);
1393 					(*dom->dom_dispose)(cm);
1394 					solock(so);
1395 				}
1396 				m_freem(cm);
1397 			}
1398 		}
1399 		if (m != NULL)
1400 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1401 		else
1402 			nextrecord = so->so_rcv.sb_mb;
1403 		orig_resid = 0;
1404 	}
1405 
1406 	/* If m is non-NULL, we have some data to read. */
1407 	if (__predict_true(m != NULL)) {
1408 		type = m->m_type;
1409 		if (type == MT_OOBDATA)
1410 			flags |= MSG_OOB;
1411 	}
1412 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1413 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1414 
1415 	moff = 0;
1416 	offset = 0;
1417 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1418 		/*
1419 		 * If the type of mbuf has changed, end the receive
1420 		 * operation and do a short read.
1421 		 */
1422 		if (m->m_type == MT_OOBDATA) {
1423 			if (type != MT_OOBDATA)
1424 				break;
1425 		} else if (type == MT_OOBDATA) {
1426 			break;
1427 		} else if (m->m_type == MT_CONTROL) {
1428 			break;
1429 		}
1430 #ifdef DIAGNOSTIC
1431 		else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) {
1432 			panic("%s: m_type=%d", __func__, m->m_type);
1433 		}
1434 #endif
1435 
1436 		so->so_state &= ~SS_RCVATMARK;
1437 		wakeup_state = 0;
1438 		len = uio->uio_resid;
1439 		if (so->so_oobmark && len > so->so_oobmark - offset)
1440 			len = so->so_oobmark - offset;
1441 		if (len > m->m_len - moff)
1442 			len = m->m_len - moff;
1443 
1444 		/*
1445 		 * If mp is set, just pass back the mbufs.
1446 		 * Otherwise copy them out via the uio, then free.
1447 		 * Sockbuf must be consistent here (points to current mbuf,
1448 		 * it points to next record) when we drop priority;
1449 		 * we must note any additions to the sockbuf when we
1450 		 * block interrupts again.
1451 		 */
1452 		if (mp == NULL) {
1453 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1454 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1455 			sounlock(so);
1456 			splx(s);
1457 			error = uiomove(mtod(m, char *) + moff, len, uio);
1458 			s = splsoftnet();
1459 			solock(so);
1460 			if (error != 0) {
1461 				/*
1462 				 * If any part of the record has been removed
1463 				 * (such as the MT_SONAME mbuf, which will
1464 				 * happen when PR_ADDR, and thus also
1465 				 * PR_ATOMIC, is set), then drop the entire
1466 				 * record to maintain the atomicity of the
1467 				 * receive operation.
1468 				 *
1469 				 * This avoids a later panic("receive 1a")
1470 				 * when compiled with DIAGNOSTIC.
1471 				 */
1472 				if (m && mbuf_removed && atomic)
1473 					(void) sbdroprecord(&so->so_rcv);
1474 
1475 				goto release;
1476 			}
1477 		} else {
1478 			uio->uio_resid -= len;
1479 		}
1480 
1481 		if (len == m->m_len - moff) {
1482 			if (m->m_flags & M_EOR)
1483 				flags |= MSG_EOR;
1484 #ifdef SCTP
1485 			if (m->m_flags & M_NOTIFICATION)
1486 				flags |= MSG_NOTIFICATION;
1487 #endif
1488 			if (flags & MSG_PEEK) {
1489 				m = m->m_next;
1490 				moff = 0;
1491 			} else {
1492 				nextrecord = m->m_nextpkt;
1493 				sbfree(&so->so_rcv, m);
1494 				if (mp) {
1495 					*mp = m;
1496 					mp = &m->m_next;
1497 					so->so_rcv.sb_mb = m = m->m_next;
1498 					*mp = NULL;
1499 				} else {
1500 					m = so->so_rcv.sb_mb = m_free(m);
1501 				}
1502 				/*
1503 				 * If m != NULL, we also know that
1504 				 * so->so_rcv.sb_mb != NULL.
1505 				 */
1506 				KASSERT(so->so_rcv.sb_mb == m);
1507 				if (m) {
1508 					m->m_nextpkt = nextrecord;
1509 					if (nextrecord == NULL)
1510 						so->so_rcv.sb_lastrecord = m;
1511 				} else {
1512 					so->so_rcv.sb_mb = nextrecord;
1513 					SB_EMPTY_FIXUP(&so->so_rcv);
1514 				}
1515 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1516 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1517 			}
1518 		} else if (flags & MSG_PEEK) {
1519 			moff += len;
1520 		} else {
1521 			if (mp != NULL) {
1522 				mt = m_copym(m, 0, len, M_NOWAIT);
1523 				if (__predict_false(mt == NULL)) {
1524 					sounlock(so);
1525 					mt = m_copym(m, 0, len, M_WAIT);
1526 					solock(so);
1527 				}
1528 				*mp = mt;
1529 			}
1530 			m->m_data += len;
1531 			m->m_len -= len;
1532 			so->so_rcv.sb_cc -= len;
1533 		}
1534 
1535 		if (so->so_oobmark) {
1536 			if ((flags & MSG_PEEK) == 0) {
1537 				so->so_oobmark -= len;
1538 				if (so->so_oobmark == 0) {
1539 					so->so_state |= SS_RCVATMARK;
1540 					break;
1541 				}
1542 			} else {
1543 				offset += len;
1544 				if (offset == so->so_oobmark)
1545 					break;
1546 			}
1547 		}
1548 		if (flags & MSG_EOR)
1549 			break;
1550 
1551 		/*
1552 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1553 		 * we must not quit until "uio->uio_resid == 0" or an error
1554 		 * termination.  If a signal/timeout occurs, return
1555 		 * with a short count but without error.
1556 		 * Keep sockbuf locked against other readers.
1557 		 */
1558 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1559 		    !sosendallatonce(so) && !nextrecord) {
1560 			if (so->so_error || so->so_rerror ||
1561 			    so->so_state & SS_CANTRCVMORE)
1562 				break;
1563 			/*
1564 			 * If we are peeking and the socket receive buffer is
1565 			 * full, stop since we can't get more data to peek at.
1566 			 */
1567 			if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
1568 				break;
1569 			/*
1570 			 * If we've drained the socket buffer, tell the
1571 			 * protocol in case it needs to do something to
1572 			 * get it filled again.
1573 			 */
1574 			if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1575 				(*pr->pr_usrreqs->pr_rcvd)(so, flags, l);
1576 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1577 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1578 			if (wakeup_state & SS_RESTARTSYS)
1579 				error = ERESTART;
1580 			else
1581 				error = sbwait(&so->so_rcv);
1582 			if (error != 0) {
1583 				sbunlock(&so->so_rcv);
1584 				sounlock(so);
1585 				splx(s);
1586 				return 0;
1587 			}
1588 			if ((m = so->so_rcv.sb_mb) != NULL)
1589 				nextrecord = m->m_nextpkt;
1590 			wakeup_state = so->so_state;
1591 		}
1592 	}
1593 
1594 	if (m && atomic) {
1595 		flags |= MSG_TRUNC;
1596 		if ((flags & MSG_PEEK) == 0)
1597 			(void) sbdroprecord(&so->so_rcv);
1598 	}
1599 	if ((flags & MSG_PEEK) == 0) {
1600 		if (m == NULL) {
1601 			/*
1602 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1603 			 * part makes sure sb_lastrecord is up-to-date if
1604 			 * there is still data in the socket buffer.
1605 			 */
1606 			so->so_rcv.sb_mb = nextrecord;
1607 			if (so->so_rcv.sb_mb == NULL) {
1608 				so->so_rcv.sb_mbtail = NULL;
1609 				so->so_rcv.sb_lastrecord = NULL;
1610 			} else if (nextrecord->m_nextpkt == NULL)
1611 				so->so_rcv.sb_lastrecord = nextrecord;
1612 		}
1613 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1614 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1615 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1616 			(*pr->pr_usrreqs->pr_rcvd)(so, flags, l);
1617 	}
1618 	if (orig_resid == uio->uio_resid && orig_resid &&
1619 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1620 		sbunlock(&so->so_rcv);
1621 		goto restart;
1622 	}
1623 
1624 	if (flagsp != NULL)
1625 		*flagsp |= flags;
1626 release:
1627 	sbunlock(&so->so_rcv);
1628 	sounlock(so);
1629 	splx(s);
1630 	return error;
1631 }
1632 
1633 int
1634 soshutdown(struct socket *so, int how)
1635 {
1636 	const struct protosw *pr;
1637 	int error;
1638 
1639 	KASSERT(solocked(so));
1640 
1641 	pr = so->so_proto;
1642 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1643 		return EINVAL;
1644 
1645 	if (how == SHUT_RD || how == SHUT_RDWR) {
1646 		sorflush(so);
1647 		error = 0;
1648 	}
1649 	if (how == SHUT_WR || how == SHUT_RDWR)
1650 		error = (*pr->pr_usrreqs->pr_shutdown)(so);
1651 
1652 	return error;
1653 }
1654 
1655 void
1656 sorestart(struct socket *so)
1657 {
1658 	/*
1659 	 * An application has called close() on an fd on which another
1660 	 * of its threads has called a socket system call.
1661 	 * Mark this and wake everyone up, and code that would block again
1662 	 * instead returns ERESTART.
1663 	 * On system call re-entry the fd is validated and EBADF returned.
1664 	 * Any other fd will block again on the 2nd syscall.
1665 	 */
1666 	solock(so);
1667 	so->so_state |= SS_RESTARTSYS;
1668 	cv_broadcast(&so->so_cv);
1669 	cv_broadcast(&so->so_snd.sb_cv);
1670 	cv_broadcast(&so->so_rcv.sb_cv);
1671 	sounlock(so);
1672 }
1673 
1674 void
1675 sorflush(struct socket *so)
1676 {
1677 	struct sockbuf *sb, asb;
1678 	const struct protosw *pr;
1679 
1680 	KASSERT(solocked(so));
1681 
1682 	sb = &so->so_rcv;
1683 	pr = so->so_proto;
1684 	socantrcvmore(so);
1685 	sb->sb_flags |= SB_NOINTR;
1686 	(void )sblock(sb, M_WAITOK);
1687 	sbunlock(sb);
1688 	asb = *sb;
1689 	/*
1690 	 * Clear most of the sockbuf structure, but leave some of the
1691 	 * fields valid.
1692 	 */
1693 	memset(&sb->sb_startzero, 0,
1694 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1695 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) {
1696 		sounlock(so);
1697 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1698 		solock(so);
1699 	}
1700 	sbrelease(&asb, so);
1701 }
1702 
1703 /*
1704  * internal set SOL_SOCKET options
1705  */
1706 static int
1707 sosetopt1(struct socket *so, const struct sockopt *sopt)
1708 {
1709 	int error, opt;
1710 	int optval = 0; /* XXX: gcc */
1711 	struct linger l;
1712 	struct timeval tv;
1713 
1714 	opt = sopt->sopt_name;
1715 
1716 	switch (opt) {
1717 
1718 	case SO_ACCEPTFILTER:
1719 		error = accept_filt_setopt(so, sopt);
1720 		KASSERT(solocked(so));
1721 		break;
1722 
1723 	case SO_LINGER:
1724 		error = sockopt_get(sopt, &l, sizeof(l));
1725 		solock(so);
1726 		if (error)
1727 			break;
1728 		if (l.l_linger < 0 || l.l_linger > USHRT_MAX ||
1729 		    l.l_linger > (INT_MAX / hz)) {
1730 			error = EDOM;
1731 			break;
1732 		}
1733 		so->so_linger = l.l_linger;
1734 		if (l.l_onoff)
1735 			so->so_options |= SO_LINGER;
1736 		else
1737 			so->so_options &= ~SO_LINGER;
1738 		break;
1739 
1740 	case SO_DEBUG:
1741 	case SO_KEEPALIVE:
1742 	case SO_DONTROUTE:
1743 	case SO_USELOOPBACK:
1744 	case SO_BROADCAST:
1745 	case SO_REUSEADDR:
1746 	case SO_REUSEPORT:
1747 	case SO_OOBINLINE:
1748 	case SO_TIMESTAMP:
1749 	case SO_NOSIGPIPE:
1750 	case SO_RERROR:
1751 		error = sockopt_getint(sopt, &optval);
1752 		solock(so);
1753 		if (error)
1754 			break;
1755 		if (optval)
1756 			so->so_options |= opt;
1757 		else
1758 			so->so_options &= ~opt;
1759 		break;
1760 
1761 	case SO_SNDBUF:
1762 	case SO_RCVBUF:
1763 	case SO_SNDLOWAT:
1764 	case SO_RCVLOWAT:
1765 		error = sockopt_getint(sopt, &optval);
1766 		solock(so);
1767 		if (error)
1768 			break;
1769 
1770 		/*
1771 		 * Values < 1 make no sense for any of these
1772 		 * options, so disallow them.
1773 		 */
1774 		if (optval < 1) {
1775 			error = EINVAL;
1776 			break;
1777 		}
1778 
1779 		switch (opt) {
1780 		case SO_SNDBUF:
1781 			if (sbreserve(&so->so_snd, (u_long)optval, so) == 0) {
1782 				error = ENOBUFS;
1783 				break;
1784 			}
1785 			so->so_snd.sb_flags &= ~SB_AUTOSIZE;
1786 			break;
1787 
1788 		case SO_RCVBUF:
1789 			if (sbreserve(&so->so_rcv, (u_long)optval, so) == 0) {
1790 				error = ENOBUFS;
1791 				break;
1792 			}
1793 			so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1794 			break;
1795 
1796 		/*
1797 		 * Make sure the low-water is never greater than
1798 		 * the high-water.
1799 		 */
1800 		case SO_SNDLOWAT:
1801 			if (optval > so->so_snd.sb_hiwat)
1802 				optval = so->so_snd.sb_hiwat;
1803 
1804 			so->so_snd.sb_lowat = optval;
1805 			break;
1806 
1807 		case SO_RCVLOWAT:
1808 			if (optval > so->so_rcv.sb_hiwat)
1809 				optval = so->so_rcv.sb_hiwat;
1810 
1811 			so->so_rcv.sb_lowat = optval;
1812 			break;
1813 		}
1814 		break;
1815 
1816 	case SO_SNDTIMEO:
1817 	case SO_RCVTIMEO:
1818 		solock(so);
1819 		error = sockopt_get(sopt, &tv, sizeof(tv));
1820 		if (error)
1821 			break;
1822 
1823 		if (tv.tv_sec < 0 || tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
1824 			error = EDOM;
1825 			break;
1826 		}
1827 		if (tv.tv_sec > (INT_MAX - tv.tv_usec / tick) / hz) {
1828 			error = EDOM;
1829 			break;
1830 		}
1831 
1832 		optval = tv.tv_sec * hz + tv.tv_usec / tick;
1833 		if (optval == 0 && tv.tv_usec != 0)
1834 			optval = 1;
1835 
1836 		switch (opt) {
1837 		case SO_SNDTIMEO:
1838 			so->so_snd.sb_timeo = optval;
1839 			break;
1840 		case SO_RCVTIMEO:
1841 			so->so_rcv.sb_timeo = optval;
1842 			break;
1843 		}
1844 		break;
1845 
1846 	default:
1847 		MODULE_HOOK_CALL(uipc_socket_50_setopt1_hook,
1848 		    (opt, so, sopt), enosys(), error);
1849 		if (error == ENOSYS || error == EPASSTHROUGH) {
1850 			solock(so);
1851 			error = ENOPROTOOPT;
1852 		}
1853 		break;
1854 	}
1855 	KASSERT(solocked(so));
1856 	return error;
1857 }
1858 
1859 int
1860 sosetopt(struct socket *so, struct sockopt *sopt)
1861 {
1862 	int error, prerr;
1863 
1864 	if (sopt->sopt_level == SOL_SOCKET) {
1865 		error = sosetopt1(so, sopt);
1866 		KASSERT(solocked(so));
1867 	} else {
1868 		error = ENOPROTOOPT;
1869 		solock(so);
1870 	}
1871 
1872 	if ((error == 0 || error == ENOPROTOOPT) &&
1873 	    so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) {
1874 		/* give the protocol stack a shot */
1875 		prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, sopt);
1876 		if (prerr == 0)
1877 			error = 0;
1878 		else if (prerr != ENOPROTOOPT)
1879 			error = prerr;
1880 	}
1881 	sounlock(so);
1882 	return error;
1883 }
1884 
1885 /*
1886  * so_setsockopt() is a wrapper providing a sockopt structure for sosetopt()
1887  */
1888 int
1889 so_setsockopt(struct lwp *l, struct socket *so, int level, int name,
1890     const void *val, size_t valsize)
1891 {
1892 	struct sockopt sopt;
1893 	int error;
1894 
1895 	KASSERT(valsize == 0 || val != NULL);
1896 
1897 	sockopt_init(&sopt, level, name, valsize);
1898 	sockopt_set(&sopt, val, valsize);
1899 
1900 	error = sosetopt(so, &sopt);
1901 
1902 	sockopt_destroy(&sopt);
1903 
1904 	return error;
1905 }
1906 
1907 /*
1908  * internal get SOL_SOCKET options
1909  */
1910 static int
1911 sogetopt1(struct socket *so, struct sockopt *sopt)
1912 {
1913 	int error, optval, opt;
1914 	struct linger l;
1915 	struct timeval tv;
1916 
1917 	switch ((opt = sopt->sopt_name)) {
1918 
1919 	case SO_ACCEPTFILTER:
1920 		error = accept_filt_getopt(so, sopt);
1921 		break;
1922 
1923 	case SO_LINGER:
1924 		l.l_onoff = (so->so_options & SO_LINGER) ? 1 : 0;
1925 		l.l_linger = so->so_linger;
1926 
1927 		error = sockopt_set(sopt, &l, sizeof(l));
1928 		break;
1929 
1930 	case SO_USELOOPBACK:
1931 	case SO_DONTROUTE:
1932 	case SO_DEBUG:
1933 	case SO_KEEPALIVE:
1934 	case SO_REUSEADDR:
1935 	case SO_REUSEPORT:
1936 	case SO_BROADCAST:
1937 	case SO_OOBINLINE:
1938 	case SO_TIMESTAMP:
1939 	case SO_NOSIGPIPE:
1940 	case SO_RERROR:
1941 	case SO_ACCEPTCONN:
1942 		error = sockopt_setint(sopt, (so->so_options & opt) ? 1 : 0);
1943 		break;
1944 
1945 	case SO_TYPE:
1946 		error = sockopt_setint(sopt, so->so_type);
1947 		break;
1948 
1949 	case SO_ERROR:
1950 		if (so->so_error == 0) {
1951 			so->so_error = so->so_rerror;
1952 			so->so_rerror = 0;
1953 		}
1954 		error = sockopt_setint(sopt, so->so_error);
1955 		so->so_error = 0;
1956 		break;
1957 
1958 	case SO_SNDBUF:
1959 		error = sockopt_setint(sopt, so->so_snd.sb_hiwat);
1960 		break;
1961 
1962 	case SO_RCVBUF:
1963 		error = sockopt_setint(sopt, so->so_rcv.sb_hiwat);
1964 		break;
1965 
1966 	case SO_SNDLOWAT:
1967 		error = sockopt_setint(sopt, so->so_snd.sb_lowat);
1968 		break;
1969 
1970 	case SO_RCVLOWAT:
1971 		error = sockopt_setint(sopt, so->so_rcv.sb_lowat);
1972 		break;
1973 
1974 	case SO_SNDTIMEO:
1975 	case SO_RCVTIMEO:
1976 		optval = (opt == SO_SNDTIMEO ?
1977 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1978 
1979 		tv.tv_sec = optval / hz;
1980 		tv.tv_usec = (optval % hz) * tick;
1981 
1982 		error = sockopt_set(sopt, &tv, sizeof(tv));
1983 		break;
1984 
1985 	case SO_OVERFLOWED:
1986 		error = sockopt_setint(sopt, so->so_rcv.sb_overflowed);
1987 		break;
1988 
1989 	default:
1990 		MODULE_HOOK_CALL(uipc_socket_50_getopt1_hook,
1991 		    (opt, so, sopt), enosys(), error);
1992 		if (error)
1993 			error = ENOPROTOOPT;
1994 		break;
1995 	}
1996 
1997 	return error;
1998 }
1999 
2000 int
2001 sogetopt(struct socket *so, struct sockopt *sopt)
2002 {
2003 	int error;
2004 
2005 	solock(so);
2006 	if (sopt->sopt_level != SOL_SOCKET) {
2007 		if (so->so_proto && so->so_proto->pr_ctloutput) {
2008 			error = ((*so->so_proto->pr_ctloutput)
2009 			    (PRCO_GETOPT, so, sopt));
2010 		} else
2011 			error = (ENOPROTOOPT);
2012 	} else {
2013 		error = sogetopt1(so, sopt);
2014 	}
2015 	sounlock(so);
2016 	return error;
2017 }
2018 
2019 /*
2020  * alloc sockopt data buffer buffer
2021  *	- will be released at destroy
2022  */
2023 static int
2024 sockopt_alloc(struct sockopt *sopt, size_t len, km_flag_t kmflag)
2025 {
2026 
2027 	KASSERT(sopt->sopt_size == 0);
2028 
2029 	if (len > sizeof(sopt->sopt_buf)) {
2030 		sopt->sopt_data = kmem_zalloc(len, kmflag);
2031 		if (sopt->sopt_data == NULL)
2032 			return ENOMEM;
2033 	} else
2034 		sopt->sopt_data = sopt->sopt_buf;
2035 
2036 	sopt->sopt_size = len;
2037 	return 0;
2038 }
2039 
2040 /*
2041  * initialise sockopt storage
2042  *	- MAY sleep during allocation
2043  */
2044 void
2045 sockopt_init(struct sockopt *sopt, int level, int name, size_t size)
2046 {
2047 
2048 	memset(sopt, 0, sizeof(*sopt));
2049 
2050 	sopt->sopt_level = level;
2051 	sopt->sopt_name = name;
2052 	(void)sockopt_alloc(sopt, size, KM_SLEEP);
2053 }
2054 
2055 /*
2056  * destroy sockopt storage
2057  *	- will release any held memory references
2058  */
2059 void
2060 sockopt_destroy(struct sockopt *sopt)
2061 {
2062 
2063 	if (sopt->sopt_data != sopt->sopt_buf)
2064 		kmem_free(sopt->sopt_data, sopt->sopt_size);
2065 
2066 	memset(sopt, 0, sizeof(*sopt));
2067 }
2068 
2069 /*
2070  * set sockopt value
2071  *	- value is copied into sockopt
2072  *	- memory is allocated when necessary, will not sleep
2073  */
2074 int
2075 sockopt_set(struct sockopt *sopt, const void *buf, size_t len)
2076 {
2077 	int error;
2078 
2079 	if (sopt->sopt_size == 0) {
2080 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2081 		if (error)
2082 			return error;
2083 	}
2084 
2085 	sopt->sopt_retsize = MIN(sopt->sopt_size, len);
2086 	memcpy(sopt->sopt_data, buf, sopt->sopt_retsize);
2087 
2088 	return 0;
2089 }
2090 
2091 /*
2092  * common case of set sockopt integer value
2093  */
2094 int
2095 sockopt_setint(struct sockopt *sopt, int val)
2096 {
2097 
2098 	return sockopt_set(sopt, &val, sizeof(int));
2099 }
2100 
2101 /*
2102  * get sockopt value
2103  *	- correct size must be given
2104  */
2105 int
2106 sockopt_get(const struct sockopt *sopt, void *buf, size_t len)
2107 {
2108 
2109 	if (sopt->sopt_size != len)
2110 		return EINVAL;
2111 
2112 	memcpy(buf, sopt->sopt_data, len);
2113 	return 0;
2114 }
2115 
2116 /*
2117  * common case of get sockopt integer value
2118  */
2119 int
2120 sockopt_getint(const struct sockopt *sopt, int *valp)
2121 {
2122 
2123 	return sockopt_get(sopt, valp, sizeof(int));
2124 }
2125 
2126 /*
2127  * set sockopt value from mbuf
2128  *	- ONLY for legacy code
2129  *	- mbuf is released by sockopt
2130  *	- will not sleep
2131  */
2132 int
2133 sockopt_setmbuf(struct sockopt *sopt, struct mbuf *m)
2134 {
2135 	size_t len;
2136 	int error;
2137 
2138 	len = m_length(m);
2139 
2140 	if (sopt->sopt_size == 0) {
2141 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2142 		if (error)
2143 			return error;
2144 	}
2145 
2146 	sopt->sopt_retsize = MIN(sopt->sopt_size, len);
2147 	m_copydata(m, 0, sopt->sopt_retsize, sopt->sopt_data);
2148 	m_freem(m);
2149 
2150 	return 0;
2151 }
2152 
2153 /*
2154  * get sockopt value into mbuf
2155  *	- ONLY for legacy code
2156  *	- mbuf to be released by the caller
2157  *	- will not sleep
2158  */
2159 struct mbuf *
2160 sockopt_getmbuf(const struct sockopt *sopt)
2161 {
2162 	struct mbuf *m;
2163 
2164 	if (sopt->sopt_size > MCLBYTES)
2165 		return NULL;
2166 
2167 	m = m_get(M_DONTWAIT, MT_SOOPTS);
2168 	if (m == NULL)
2169 		return NULL;
2170 
2171 	if (sopt->sopt_size > MLEN) {
2172 		MCLGET(m, M_DONTWAIT);
2173 		if ((m->m_flags & M_EXT) == 0) {
2174 			m_free(m);
2175 			return NULL;
2176 		}
2177 	}
2178 
2179 	memcpy(mtod(m, void *), sopt->sopt_data, sopt->sopt_size);
2180 	m->m_len = sopt->sopt_size;
2181 
2182 	return m;
2183 }
2184 
2185 void
2186 sohasoutofband(struct socket *so)
2187 {
2188 
2189 	fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
2190 	selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, NOTE_SUBMIT);
2191 }
2192 
2193 static void
2194 filt_sordetach(struct knote *kn)
2195 {
2196 	struct socket *so;
2197 
2198 	so = ((file_t *)kn->kn_obj)->f_socket;
2199 	solock(so);
2200 	SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
2201 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
2202 		so->so_rcv.sb_flags &= ~SB_KNOTE;
2203 	sounlock(so);
2204 }
2205 
2206 /*ARGSUSED*/
2207 static int
2208 filt_soread(struct knote *kn, long hint)
2209 {
2210 	struct socket *so;
2211 	int rv;
2212 
2213 	so = ((file_t *)kn->kn_obj)->f_socket;
2214 	if (hint != NOTE_SUBMIT)
2215 		solock(so);
2216 	kn->kn_data = so->so_rcv.sb_cc;
2217 	if (so->so_state & SS_CANTRCVMORE) {
2218 		kn->kn_flags |= EV_EOF;
2219 		kn->kn_fflags = so->so_error;
2220 		rv = 1;
2221 	} else if (so->so_error || so->so_rerror)
2222 		rv = 1;
2223 	else if (kn->kn_sfflags & NOTE_LOWAT)
2224 		rv = (kn->kn_data >= kn->kn_sdata);
2225 	else
2226 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
2227 	if (hint != NOTE_SUBMIT)
2228 		sounlock(so);
2229 	return rv;
2230 }
2231 
2232 static void
2233 filt_sowdetach(struct knote *kn)
2234 {
2235 	struct socket *so;
2236 
2237 	so = ((file_t *)kn->kn_obj)->f_socket;
2238 	solock(so);
2239 	SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
2240 	if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
2241 		so->so_snd.sb_flags &= ~SB_KNOTE;
2242 	sounlock(so);
2243 }
2244 
2245 /*ARGSUSED*/
2246 static int
2247 filt_sowrite(struct knote *kn, long hint)
2248 {
2249 	struct socket *so;
2250 	int rv;
2251 
2252 	so = ((file_t *)kn->kn_obj)->f_socket;
2253 	if (hint != NOTE_SUBMIT)
2254 		solock(so);
2255 	kn->kn_data = sbspace(&so->so_snd);
2256 	if (so->so_state & SS_CANTSENDMORE) {
2257 		kn->kn_flags |= EV_EOF;
2258 		kn->kn_fflags = so->so_error;
2259 		rv = 1;
2260 	} else if (so->so_error)
2261 		rv = 1;
2262 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2263 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
2264 		rv = 0;
2265 	else if (kn->kn_sfflags & NOTE_LOWAT)
2266 		rv = (kn->kn_data >= kn->kn_sdata);
2267 	else
2268 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
2269 	if (hint != NOTE_SUBMIT)
2270 		sounlock(so);
2271 	return rv;
2272 }
2273 
2274 /*ARGSUSED*/
2275 static int
2276 filt_solisten(struct knote *kn, long hint)
2277 {
2278 	struct socket *so;
2279 	int rv;
2280 
2281 	so = ((file_t *)kn->kn_obj)->f_socket;
2282 
2283 	/*
2284 	 * Set kn_data to number of incoming connections, not
2285 	 * counting partial (incomplete) connections.
2286 	 */
2287 	if (hint != NOTE_SUBMIT)
2288 		solock(so);
2289 	kn->kn_data = so->so_qlen;
2290 	rv = (kn->kn_data > 0);
2291 	if (hint != NOTE_SUBMIT)
2292 		sounlock(so);
2293 	return rv;
2294 }
2295 
2296 static const struct filterops solisten_filtops = {
2297 	.f_isfd = 1,
2298 	.f_attach = NULL,
2299 	.f_detach = filt_sordetach,
2300 	.f_event = filt_solisten,
2301 };
2302 
2303 static const struct filterops soread_filtops = {
2304 	.f_isfd = 1,
2305 	.f_attach = NULL,
2306 	.f_detach = filt_sordetach,
2307 	.f_event = filt_soread,
2308 };
2309 
2310 static const struct filterops sowrite_filtops = {
2311 	.f_isfd = 1,
2312 	.f_attach = NULL,
2313 	.f_detach = filt_sowdetach,
2314 	.f_event = filt_sowrite,
2315 };
2316 
2317 int
2318 soo_kqfilter(struct file *fp, struct knote *kn)
2319 {
2320 	struct socket *so;
2321 	struct sockbuf *sb;
2322 
2323 	so = ((file_t *)kn->kn_obj)->f_socket;
2324 	solock(so);
2325 	switch (kn->kn_filter) {
2326 	case EVFILT_READ:
2327 		if (so->so_options & SO_ACCEPTCONN)
2328 			kn->kn_fop = &solisten_filtops;
2329 		else
2330 			kn->kn_fop = &soread_filtops;
2331 		sb = &so->so_rcv;
2332 		break;
2333 	case EVFILT_WRITE:
2334 		kn->kn_fop = &sowrite_filtops;
2335 		sb = &so->so_snd;
2336 		break;
2337 	default:
2338 		sounlock(so);
2339 		return EINVAL;
2340 	}
2341 	SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
2342 	sb->sb_flags |= SB_KNOTE;
2343 	sounlock(so);
2344 	return 0;
2345 }
2346 
2347 static int
2348 sodopoll(struct socket *so, int events)
2349 {
2350 	int revents;
2351 
2352 	revents = 0;
2353 
2354 	if (events & (POLLIN | POLLRDNORM))
2355 		if (soreadable(so))
2356 			revents |= events & (POLLIN | POLLRDNORM);
2357 
2358 	if (events & (POLLOUT | POLLWRNORM))
2359 		if (sowritable(so))
2360 			revents |= events & (POLLOUT | POLLWRNORM);
2361 
2362 	if (events & (POLLPRI | POLLRDBAND))
2363 		if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
2364 			revents |= events & (POLLPRI | POLLRDBAND);
2365 
2366 	return revents;
2367 }
2368 
2369 int
2370 sopoll(struct socket *so, int events)
2371 {
2372 	int revents = 0;
2373 
2374 #ifndef DIAGNOSTIC
2375 	/*
2376 	 * Do a quick, unlocked check in expectation that the socket
2377 	 * will be ready for I/O.  Don't do this check if DIAGNOSTIC,
2378 	 * as the solocked() assertions will fail.
2379 	 */
2380 	if ((revents = sodopoll(so, events)) != 0)
2381 		return revents;
2382 #endif
2383 
2384 	solock(so);
2385 	if ((revents = sodopoll(so, events)) == 0) {
2386 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2387 			selrecord(curlwp, &so->so_rcv.sb_sel);
2388 			so->so_rcv.sb_flags |= SB_NOTIFY;
2389 		}
2390 
2391 		if (events & (POLLOUT | POLLWRNORM)) {
2392 			selrecord(curlwp, &so->so_snd.sb_sel);
2393 			so->so_snd.sb_flags |= SB_NOTIFY;
2394 		}
2395 	}
2396 	sounlock(so);
2397 
2398 	return revents;
2399 }
2400 
2401 struct mbuf **
2402 sbsavetimestamp(int opt, struct mbuf **mp)
2403 {
2404 	struct timeval tv;
2405 	int error;
2406 
2407 	microtime(&tv);
2408 
2409 	MODULE_HOOK_CALL(uipc_socket_50_sbts_hook, (opt, mp), enosys(), error);
2410 	if (error == 0)
2411 		return mp;
2412 
2413 	if (opt & SO_TIMESTAMP) {
2414 		*mp = sbcreatecontrol(&tv, sizeof(tv),
2415 		    SCM_TIMESTAMP, SOL_SOCKET);
2416 		if (*mp)
2417 			mp = &(*mp)->m_next;
2418 	}
2419 	return mp;
2420 }
2421 
2422 
2423 #include <sys/sysctl.h>
2424 
2425 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO);
2426 static int sysctl_kern_sbmax(SYSCTLFN_PROTO);
2427 
2428 /*
2429  * sysctl helper routine for kern.somaxkva.  ensures that the given
2430  * value is not too small.
2431  * (XXX should we maybe make sure it's not too large as well?)
2432  */
2433 static int
2434 sysctl_kern_somaxkva(SYSCTLFN_ARGS)
2435 {
2436 	int error, new_somaxkva;
2437 	struct sysctlnode node;
2438 
2439 	new_somaxkva = somaxkva;
2440 	node = *rnode;
2441 	node.sysctl_data = &new_somaxkva;
2442 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2443 	if (error || newp == NULL)
2444 		return error;
2445 
2446 	if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */
2447 		return EINVAL;
2448 
2449 	mutex_enter(&so_pendfree_lock);
2450 	somaxkva = new_somaxkva;
2451 	cv_broadcast(&socurkva_cv);
2452 	mutex_exit(&so_pendfree_lock);
2453 
2454 	return error;
2455 }
2456 
2457 /*
2458  * sysctl helper routine for kern.sbmax. Basically just ensures that
2459  * any new value is not too small.
2460  */
2461 static int
2462 sysctl_kern_sbmax(SYSCTLFN_ARGS)
2463 {
2464 	int error, new_sbmax;
2465 	struct sysctlnode node;
2466 
2467 	new_sbmax = sb_max;
2468 	node = *rnode;
2469 	node.sysctl_data = &new_sbmax;
2470 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2471 	if (error || newp == NULL)
2472 		return error;
2473 
2474 	KERNEL_LOCK(1, NULL);
2475 	error = sb_max_set(new_sbmax);
2476 	KERNEL_UNLOCK_ONE(NULL);
2477 
2478 	return error;
2479 }
2480 
2481 /*
2482  * sysctl helper routine for kern.sooptions. Ensures that only allowed
2483  * options can be set.
2484  */
2485 static int
2486 sysctl_kern_sooptions(SYSCTLFN_ARGS)
2487 {
2488 	int error, new_options;
2489 	struct sysctlnode node;
2490 
2491 	new_options = sooptions;
2492 	node = *rnode;
2493 	node.sysctl_data = &new_options;
2494 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2495 	if (error || newp == NULL)
2496 		return error;
2497 
2498 	if (new_options & ~SO_DEFOPTS)
2499 		return EINVAL;
2500 
2501 	sooptions = new_options;
2502 
2503 	return 0;
2504 }
2505 
2506 static void
2507 sysctl_kern_socket_setup(void)
2508 {
2509 
2510 	KASSERT(socket_sysctllog == NULL);
2511 
2512 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2513 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2514 		       CTLTYPE_INT, "somaxkva",
2515 		       SYSCTL_DESCR("Maximum amount of kernel memory to be "
2516 		                    "used for socket buffers"),
2517 		       sysctl_kern_somaxkva, 0, NULL, 0,
2518 		       CTL_KERN, KERN_SOMAXKVA, CTL_EOL);
2519 
2520 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2521 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2522 		       CTLTYPE_INT, "sbmax",
2523 		       SYSCTL_DESCR("Maximum socket buffer size"),
2524 		       sysctl_kern_sbmax, 0, NULL, 0,
2525 		       CTL_KERN, KERN_SBMAX, CTL_EOL);
2526 
2527 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2528 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2529 		       CTLTYPE_INT, "sooptions",
2530 		       SYSCTL_DESCR("Default socket options"),
2531 		       sysctl_kern_sooptions, 0, NULL, 0,
2532 		       CTL_KERN, CTL_CREATE, CTL_EOL);
2533 }
2534