xref: /netbsd-src/sys/kern/uipc_socket.c (revision b7b7574d3bf8eeb51a1fa3977b59142ec6434a55)
1 /*	$NetBSD: uipc_socket.c,v 1.224 2014/05/19 02:51:24 rmind 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 1.224 2014/05/19 02:51:24 rmind Exp $");
75 
76 #include "opt_compat_netbsd.h"
77 #include "opt_sock_counters.h"
78 #include "opt_sosend_loan.h"
79 #include "opt_mbuftrace.h"
80 #include "opt_somaxkva.h"
81 #include "opt_multiprocessor.h"	/* XXX */
82 
83 #include <sys/param.h>
84 #include <sys/systm.h>
85 #include <sys/proc.h>
86 #include <sys/file.h>
87 #include <sys/filedesc.h>
88 #include <sys/kmem.h>
89 #include <sys/mbuf.h>
90 #include <sys/domain.h>
91 #include <sys/kernel.h>
92 #include <sys/protosw.h>
93 #include <sys/socket.h>
94 #include <sys/socketvar.h>
95 #include <sys/signalvar.h>
96 #include <sys/resourcevar.h>
97 #include <sys/uidinfo.h>
98 #include <sys/event.h>
99 #include <sys/poll.h>
100 #include <sys/kauth.h>
101 #include <sys/mutex.h>
102 #include <sys/condvar.h>
103 #include <sys/kthread.h>
104 
105 #ifdef COMPAT_50
106 #include <compat/sys/time.h>
107 #include <compat/sys/socket.h>
108 #endif
109 
110 #include <uvm/uvm_extern.h>
111 #include <uvm/uvm_loan.h>
112 #include <uvm/uvm_page.h>
113 
114 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
115 
116 extern const struct fileops socketops;
117 
118 extern int	somaxconn;			/* patchable (XXX sysctl) */
119 int		somaxconn = SOMAXCONN;
120 kmutex_t	*softnet_lock;
121 
122 #ifdef SOSEND_COUNTERS
123 #include <sys/device.h>
124 
125 static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
126     NULL, "sosend", "loan big");
127 static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
128     NULL, "sosend", "copy big");
129 static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
130     NULL, "sosend", "copy small");
131 static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
132     NULL, "sosend", "kva limit");
133 
134 #define	SOSEND_COUNTER_INCR(ev)		(ev)->ev_count++
135 
136 EVCNT_ATTACH_STATIC(sosend_loan_big);
137 EVCNT_ATTACH_STATIC(sosend_copy_big);
138 EVCNT_ATTACH_STATIC(sosend_copy_small);
139 EVCNT_ATTACH_STATIC(sosend_kvalimit);
140 #else
141 
142 #define	SOSEND_COUNTER_INCR(ev)		/* nothing */
143 
144 #endif /* SOSEND_COUNTERS */
145 
146 #if defined(SOSEND_NO_LOAN) || defined(MULTIPROCESSOR)
147 int sock_loan_thresh = -1;
148 #else
149 int sock_loan_thresh = 4096;
150 #endif
151 
152 static kmutex_t so_pendfree_lock;
153 static struct mbuf *so_pendfree = NULL;
154 
155 #ifndef SOMAXKVA
156 #define	SOMAXKVA (16 * 1024 * 1024)
157 #endif
158 int somaxkva = SOMAXKVA;
159 static int socurkva;
160 static kcondvar_t socurkva_cv;
161 
162 static kauth_listener_t socket_listener;
163 
164 #define	SOCK_LOAN_CHUNK		65536
165 
166 static void sopendfree_thread(void *);
167 static kcondvar_t pendfree_thread_cv;
168 static lwp_t *sopendfree_lwp;
169 
170 static void sysctl_kern_socket_setup(void);
171 static struct sysctllog *socket_sysctllog;
172 
173 static vsize_t
174 sokvareserve(struct socket *so, vsize_t len)
175 {
176 	int error;
177 
178 	mutex_enter(&so_pendfree_lock);
179 	while (socurkva + len > somaxkva) {
180 		SOSEND_COUNTER_INCR(&sosend_kvalimit);
181 		error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock);
182 		if (error) {
183 			len = 0;
184 			break;
185 		}
186 	}
187 	socurkva += len;
188 	mutex_exit(&so_pendfree_lock);
189 	return len;
190 }
191 
192 static void
193 sokvaunreserve(vsize_t len)
194 {
195 
196 	mutex_enter(&so_pendfree_lock);
197 	socurkva -= len;
198 	cv_broadcast(&socurkva_cv);
199 	mutex_exit(&so_pendfree_lock);
200 }
201 
202 /*
203  * sokvaalloc: allocate kva for loan.
204  */
205 
206 vaddr_t
207 sokvaalloc(vaddr_t sva, vsize_t len, struct socket *so)
208 {
209 	vaddr_t lva;
210 
211 	/*
212 	 * reserve kva.
213 	 */
214 
215 	if (sokvareserve(so, len) == 0)
216 		return 0;
217 
218 	/*
219 	 * allocate kva.
220 	 */
221 
222 	lva = uvm_km_alloc(kernel_map, len, atop(sva) & uvmexp.colormask,
223 	    UVM_KMF_COLORMATCH | UVM_KMF_VAONLY | UVM_KMF_WAITVA);
224 	if (lva == 0) {
225 		sokvaunreserve(len);
226 		return (0);
227 	}
228 
229 	return lva;
230 }
231 
232 /*
233  * sokvafree: free kva for loan.
234  */
235 
236 void
237 sokvafree(vaddr_t sva, vsize_t len)
238 {
239 
240 	/*
241 	 * free kva.
242 	 */
243 
244 	uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY);
245 
246 	/*
247 	 * unreserve kva.
248 	 */
249 
250 	sokvaunreserve(len);
251 }
252 
253 static void
254 sodoloanfree(struct vm_page **pgs, void *buf, size_t size)
255 {
256 	vaddr_t sva, eva;
257 	vsize_t len;
258 	int npgs;
259 
260 	KASSERT(pgs != NULL);
261 
262 	eva = round_page((vaddr_t) buf + size);
263 	sva = trunc_page((vaddr_t) buf);
264 	len = eva - sva;
265 	npgs = len >> PAGE_SHIFT;
266 
267 	pmap_kremove(sva, len);
268 	pmap_update(pmap_kernel());
269 	uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE);
270 	sokvafree(sva, len);
271 }
272 
273 /*
274  * sopendfree_thread: free mbufs on "pendfree" list.
275  * unlock and relock so_pendfree_lock when freeing mbufs.
276  */
277 
278 static void
279 sopendfree_thread(void *v)
280 {
281 	struct mbuf *m, *next;
282 	size_t rv;
283 
284 	mutex_enter(&so_pendfree_lock);
285 
286 	for (;;) {
287 		rv = 0;
288 		while (so_pendfree != NULL) {
289 			m = so_pendfree;
290 			so_pendfree = NULL;
291 			mutex_exit(&so_pendfree_lock);
292 
293 			for (; m != NULL; m = next) {
294 				next = m->m_next;
295 				KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0);
296 				KASSERT(m->m_ext.ext_refcnt == 0);
297 
298 				rv += m->m_ext.ext_size;
299 				sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf,
300 				    m->m_ext.ext_size);
301 				pool_cache_put(mb_cache, m);
302 			}
303 
304 			mutex_enter(&so_pendfree_lock);
305 		}
306 		if (rv)
307 			cv_broadcast(&socurkva_cv);
308 		cv_wait(&pendfree_thread_cv, &so_pendfree_lock);
309 	}
310 	panic("sopendfree_thread");
311 	/* NOTREACHED */
312 }
313 
314 void
315 soloanfree(struct mbuf *m, void *buf, size_t size, void *arg)
316 {
317 
318 	KASSERT(m != NULL);
319 
320 	/*
321 	 * postpone freeing mbuf.
322 	 *
323 	 * we can't do it in interrupt context
324 	 * because we need to put kva back to kernel_map.
325 	 */
326 
327 	mutex_enter(&so_pendfree_lock);
328 	m->m_next = so_pendfree;
329 	so_pendfree = m;
330 	cv_signal(&pendfree_thread_cv);
331 	mutex_exit(&so_pendfree_lock);
332 }
333 
334 static long
335 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space)
336 {
337 	struct iovec *iov = uio->uio_iov;
338 	vaddr_t sva, eva;
339 	vsize_t len;
340 	vaddr_t lva;
341 	int npgs, error;
342 	vaddr_t va;
343 	int i;
344 
345 	if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace))
346 		return (0);
347 
348 	if (iov->iov_len < (size_t) space)
349 		space = iov->iov_len;
350 	if (space > SOCK_LOAN_CHUNK)
351 		space = SOCK_LOAN_CHUNK;
352 
353 	eva = round_page((vaddr_t) iov->iov_base + space);
354 	sva = trunc_page((vaddr_t) iov->iov_base);
355 	len = eva - sva;
356 	npgs = len >> PAGE_SHIFT;
357 
358 	KASSERT(npgs <= M_EXT_MAXPAGES);
359 
360 	lva = sokvaalloc(sva, len, so);
361 	if (lva == 0)
362 		return 0;
363 
364 	error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len,
365 	    m->m_ext.ext_pgs, UVM_LOAN_TOPAGE);
366 	if (error) {
367 		sokvafree(lva, len);
368 		return (0);
369 	}
370 
371 	for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE)
372 		pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
373 		    VM_PROT_READ, 0);
374 	pmap_update(pmap_kernel());
375 
376 	lva += (vaddr_t) iov->iov_base & PAGE_MASK;
377 
378 	MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so);
379 	m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
380 
381 	uio->uio_resid -= space;
382 	/* uio_offset not updated, not set/used for write(2) */
383 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space;
384 	uio->uio_iov->iov_len -= space;
385 	if (uio->uio_iov->iov_len == 0) {
386 		uio->uio_iov++;
387 		uio->uio_iovcnt--;
388 	}
389 
390 	return (space);
391 }
392 
393 struct mbuf *
394 getsombuf(struct socket *so, int type)
395 {
396 	struct mbuf *m;
397 
398 	m = m_get(M_WAIT, type);
399 	MCLAIM(m, so->so_mowner);
400 	return m;
401 }
402 
403 static int
404 socket_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
405     void *arg0, void *arg1, void *arg2, void *arg3)
406 {
407 	int result;
408 	enum kauth_network_req req;
409 
410 	result = KAUTH_RESULT_DEFER;
411 	req = (enum kauth_network_req)arg0;
412 
413 	if ((action != KAUTH_NETWORK_SOCKET) &&
414 	    (action != KAUTH_NETWORK_BIND))
415 		return result;
416 
417 	switch (req) {
418 	case KAUTH_REQ_NETWORK_BIND_PORT:
419 		result = KAUTH_RESULT_ALLOW;
420 		break;
421 
422 	case KAUTH_REQ_NETWORK_SOCKET_DROP: {
423 		/* Normal users can only drop their own connections. */
424 		struct socket *so = (struct socket *)arg1;
425 
426 		if (so->so_cred && proc_uidmatch(cred, so->so_cred) == 0)
427 			result = KAUTH_RESULT_ALLOW;
428 
429 		break;
430 		}
431 
432 	case KAUTH_REQ_NETWORK_SOCKET_OPEN:
433 		/* We allow "raw" routing/bluetooth sockets to anyone. */
434 		if ((u_long)arg1 == PF_ROUTE || (u_long)arg1 == PF_OROUTE
435 		    || (u_long)arg1 == PF_BLUETOOTH) {
436 			result = KAUTH_RESULT_ALLOW;
437 		} else {
438 			/* Privileged, let secmodel handle this. */
439 			if ((u_long)arg2 == SOCK_RAW)
440 				break;
441 		}
442 
443 		result = KAUTH_RESULT_ALLOW;
444 
445 		break;
446 
447 	case KAUTH_REQ_NETWORK_SOCKET_CANSEE:
448 		result = KAUTH_RESULT_ALLOW;
449 
450 		break;
451 
452 	default:
453 		break;
454 	}
455 
456 	return result;
457 }
458 
459 void
460 soinit(void)
461 {
462 
463 	sysctl_kern_socket_setup();
464 
465 	mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM);
466 	softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
467 	cv_init(&socurkva_cv, "sokva");
468 	cv_init(&pendfree_thread_cv, "sopendfr");
469 	soinit2();
470 
471 	/* Set the initial adjusted socket buffer size. */
472 	if (sb_max_set(sb_max))
473 		panic("bad initial sb_max value: %lu", sb_max);
474 
475 	socket_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK,
476 	    socket_listener_cb, NULL);
477 }
478 
479 void
480 soinit1(void)
481 {
482 	int error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL,
483 	    sopendfree_thread, NULL, &sopendfree_lwp, "sopendfree");
484 	if (error)
485 		panic("soinit1 %d", error);
486 }
487 
488 /*
489  * socreate: create a new socket of the specified type and the protocol.
490  *
491  * => Caller may specify another socket for lock sharing (must not be held).
492  * => Returns the new socket without lock held.
493  */
494 int
495 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l,
496 	 struct socket *lockso)
497 {
498 	const struct protosw	*prp;
499 	struct socket	*so;
500 	uid_t		uid;
501 	int		error;
502 	kmutex_t	*lock;
503 
504 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET,
505 	    KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type),
506 	    KAUTH_ARG(proto));
507 	if (error != 0)
508 		return error;
509 
510 	if (proto)
511 		prp = pffindproto(dom, proto, type);
512 	else
513 		prp = pffindtype(dom, type);
514 	if (prp == NULL) {
515 		/* no support for domain */
516 		if (pffinddomain(dom) == 0)
517 			return EAFNOSUPPORT;
518 		/* no support for socket type */
519 		if (proto == 0 && type != 0)
520 			return EPROTOTYPE;
521 		return EPROTONOSUPPORT;
522 	}
523 	if (prp->pr_usrreqs == NULL)
524 		return EPROTONOSUPPORT;
525 	if (prp->pr_type != type)
526 		return EPROTOTYPE;
527 
528 	so = soget(true);
529 	so->so_type = type;
530 	so->so_proto = prp;
531 	so->so_send = sosend;
532 	so->so_receive = soreceive;
533 #ifdef MBUFTRACE
534 	so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner;
535 	so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner;
536 	so->so_mowner = &prp->pr_domain->dom_mowner;
537 #endif
538 	uid = kauth_cred_geteuid(l->l_cred);
539 	so->so_uidinfo = uid_find(uid);
540 	so->so_cpid = l->l_proc->p_pid;
541 
542 	/*
543 	 * Lock assigned and taken during PCB attach, unless we share
544 	 * the lock with another socket, e.g. socketpair(2) case.
545 	 */
546 	if (lockso) {
547 		lock = lockso->so_lock;
548 		so->so_lock = lock;
549 		mutex_obj_hold(lock);
550 		mutex_enter(lock);
551 	}
552 
553 	/* Attach the PCB (returns with the socket lock held). */
554 	error = (*prp->pr_usrreqs->pr_attach)(so, proto);
555 	KASSERT(solocked(so));
556 
557 	if (error) {
558 		KASSERT(so->so_pcb == NULL);
559 		so->so_state |= SS_NOFDREF;
560 		sofree(so);
561 		return error;
562 	}
563 	so->so_cred = kauth_cred_dup(l->l_cred);
564 	sounlock(so);
565 
566 	*aso = so;
567 	return 0;
568 }
569 
570 /*
571  * fsocreate: create a socket and a file descriptor associated with it.
572  *
573  * => On success, write file descriptor to fdout and return zero.
574  * => On failure, return non-zero; *fdout will be undefined.
575  */
576 int
577 fsocreate(int domain, struct socket **sop, int type, int proto, int *fdout)
578 {
579 	lwp_t *l = curlwp;
580 	int error, fd, flags;
581 	struct socket *so;
582 	struct file *fp;
583 
584 	if ((error = fd_allocfile(&fp, &fd)) != 0) {
585 		return error;
586 	}
587 	flags = type & SOCK_FLAGS_MASK;
588 	fd_set_exclose(l, fd, (flags & SOCK_CLOEXEC) != 0);
589 	fp->f_flag = FREAD|FWRITE|((flags & SOCK_NONBLOCK) ? FNONBLOCK : 0)|
590 	    ((flags & SOCK_NOSIGPIPE) ? FNOSIGPIPE : 0);
591 	fp->f_type = DTYPE_SOCKET;
592 	fp->f_ops = &socketops;
593 
594 	type &= ~SOCK_FLAGS_MASK;
595 	error = socreate(domain, &so, type, proto, l, NULL);
596 	if (error) {
597 		fd_abort(curproc, fp, fd);
598 		return error;
599 	}
600 	if (flags & SOCK_NONBLOCK) {
601 		so->so_state |= SS_NBIO;
602 	}
603 	fp->f_data = so;
604 	fd_affix(curproc, fp, fd);
605 
606 	if (sop != NULL) {
607 		*sop = so;
608 	}
609 	*fdout = fd;
610 	return error;
611 }
612 
613 int
614 sofamily(const struct socket *so)
615 {
616 	const struct protosw *pr;
617 	const struct domain *dom;
618 
619 	if ((pr = so->so_proto) == NULL)
620 		return AF_UNSPEC;
621 	if ((dom = pr->pr_domain) == NULL)
622 		return AF_UNSPEC;
623 	return dom->dom_family;
624 }
625 
626 int
627 sobind(struct socket *so, struct mbuf *nam, struct lwp *l)
628 {
629 	int	error;
630 
631 	solock(so);
632 	error = (*so->so_proto->pr_usrreqs->pr_generic)(so,
633 	    PRU_BIND, NULL, nam, NULL, l);
634 	sounlock(so);
635 	return error;
636 }
637 
638 int
639 solisten(struct socket *so, int backlog, struct lwp *l)
640 {
641 	int	error;
642 
643 	solock(so);
644 	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
645 	    SS_ISDISCONNECTING)) != 0) {
646 		sounlock(so);
647 		return EINVAL;
648 	}
649 	error = (*so->so_proto->pr_usrreqs->pr_generic)(so,
650 	    PRU_LISTEN, NULL, NULL, NULL, l);
651 	if (error != 0) {
652 		sounlock(so);
653 		return error;
654 	}
655 	if (TAILQ_EMPTY(&so->so_q))
656 		so->so_options |= SO_ACCEPTCONN;
657 	if (backlog < 0)
658 		backlog = 0;
659 	so->so_qlimit = min(backlog, somaxconn);
660 	sounlock(so);
661 	return 0;
662 }
663 
664 void
665 sofree(struct socket *so)
666 {
667 	u_int refs;
668 
669 	KASSERT(solocked(so));
670 
671 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
672 		sounlock(so);
673 		return;
674 	}
675 	if (so->so_head) {
676 		/*
677 		 * We must not decommission a socket that's on the accept(2)
678 		 * queue.  If we do, then accept(2) may hang after select(2)
679 		 * indicated that the listening socket was ready.
680 		 */
681 		if (!soqremque(so, 0)) {
682 			sounlock(so);
683 			return;
684 		}
685 	}
686 	if (so->so_rcv.sb_hiwat)
687 		(void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0,
688 		    RLIM_INFINITY);
689 	if (so->so_snd.sb_hiwat)
690 		(void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0,
691 		    RLIM_INFINITY);
692 	sbrelease(&so->so_snd, so);
693 	KASSERT(!cv_has_waiters(&so->so_cv));
694 	KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv));
695 	KASSERT(!cv_has_waiters(&so->so_snd.sb_cv));
696 	sorflush(so);
697 	refs = so->so_aborting;	/* XXX */
698 	/* Remove acccept filter if one is present. */
699 	if (so->so_accf != NULL)
700 		(void)accept_filt_clear(so);
701 	sounlock(so);
702 	if (refs == 0)		/* XXX */
703 		soput(so);
704 }
705 
706 /*
707  * soclose: close a socket on last file table reference removal.
708  * Initiate disconnect if connected.  Free socket when disconnect complete.
709  */
710 int
711 soclose(struct socket *so)
712 {
713 	struct socket *so2;
714 	int error = 0;
715 
716 	solock(so);
717 	if (so->so_options & SO_ACCEPTCONN) {
718 		for (;;) {
719 			if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) {
720 				KASSERT(solocked2(so, so2));
721 				(void) soqremque(so2, 0);
722 				/* soabort drops the lock. */
723 				(void) soabort(so2);
724 				solock(so);
725 				continue;
726 			}
727 			if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) {
728 				KASSERT(solocked2(so, so2));
729 				(void) soqremque(so2, 1);
730 				/* soabort drops the lock. */
731 				(void) soabort(so2);
732 				solock(so);
733 				continue;
734 			}
735 			break;
736 		}
737 	}
738 	if (so->so_pcb == NULL)
739 		goto discard;
740 	if (so->so_state & SS_ISCONNECTED) {
741 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
742 			error = sodisconnect(so);
743 			if (error)
744 				goto drop;
745 		}
746 		if (so->so_options & SO_LINGER) {
747 			if ((so->so_state & (SS_ISDISCONNECTING|SS_NBIO)) ==
748 			    (SS_ISDISCONNECTING|SS_NBIO))
749 				goto drop;
750 			while (so->so_state & SS_ISCONNECTED) {
751 				error = sowait(so, true, so->so_linger * hz);
752 				if (error)
753 					break;
754 			}
755 		}
756 	}
757  drop:
758 	if (so->so_pcb) {
759 		KASSERT(solocked(so));
760 		(*so->so_proto->pr_usrreqs->pr_detach)(so);
761 	}
762  discard:
763 	KASSERT((so->so_state & SS_NOFDREF) == 0);
764 	kauth_cred_free(so->so_cred);
765 	so->so_state |= SS_NOFDREF;
766 	sofree(so);
767 	return error;
768 }
769 
770 /*
771  * Must be called with the socket locked..  Will return with it unlocked.
772  */
773 int
774 soabort(struct socket *so)
775 {
776 	u_int refs;
777 	int error;
778 
779 	KASSERT(solocked(so));
780 	KASSERT(so->so_head == NULL);
781 
782 	so->so_aborting++;		/* XXX */
783 	error = (*so->so_proto->pr_usrreqs->pr_generic)(so,
784 	    PRU_ABORT, NULL, NULL, NULL, NULL);
785 	refs = --so->so_aborting;	/* XXX */
786 	if (error || (refs == 0)) {
787 		sofree(so);
788 	} else {
789 		sounlock(so);
790 	}
791 	return error;
792 }
793 
794 int
795 soaccept(struct socket *so, struct mbuf *nam)
796 {
797 	int error;
798 
799 	KASSERT(solocked(so));
800 	KASSERT((so->so_state & SS_NOFDREF) != 0);
801 
802 	so->so_state &= ~SS_NOFDREF;
803 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
804 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
805 		error = (*so->so_proto->pr_usrreqs->pr_generic)(so,
806 		    PRU_ACCEPT, NULL, nam, NULL, NULL);
807 	else
808 		error = ECONNABORTED;
809 
810 	return error;
811 }
812 
813 int
814 soconnect(struct socket *so, struct mbuf *nam, struct lwp *l)
815 {
816 	int error;
817 
818 	KASSERT(solocked(so));
819 
820 	if (so->so_options & SO_ACCEPTCONN)
821 		return EOPNOTSUPP;
822 	/*
823 	 * If protocol is connection-based, can only connect once.
824 	 * Otherwise, if connected, try to disconnect first.
825 	 * This allows user to disconnect by connecting to, e.g.,
826 	 * a null address.
827 	 */
828 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
829 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
830 	    (error = sodisconnect(so))))
831 		error = EISCONN;
832 	else
833 		error = (*so->so_proto->pr_usrreqs->pr_generic)(so,
834 		    PRU_CONNECT, NULL, nam, NULL, l);
835 
836 	return error;
837 }
838 
839 int
840 soconnect2(struct socket *so1, struct socket *so2)
841 {
842 	KASSERT(solocked2(so1, so2));
843 
844 	return (*so1->so_proto->pr_usrreqs->pr_generic)(so1,
845 	    PRU_CONNECT2, NULL, (struct mbuf *)so2, NULL, NULL);
846 }
847 
848 int
849 sodisconnect(struct socket *so)
850 {
851 	int	error;
852 
853 	KASSERT(solocked(so));
854 
855 	if ((so->so_state & SS_ISCONNECTED) == 0) {
856 		error = ENOTCONN;
857 	} else if (so->so_state & SS_ISDISCONNECTING) {
858 		error = EALREADY;
859 	} else {
860 		error = (*so->so_proto->pr_usrreqs->pr_generic)(so,
861 		    PRU_DISCONNECT, NULL, NULL, NULL, NULL);
862 	}
863 	return (error);
864 }
865 
866 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
867 /*
868  * Send on a socket.
869  * If send must go all at once and message is larger than
870  * send buffering, then hard error.
871  * Lock against other senders.
872  * If must go all at once and not enough room now, then
873  * inform user that this would block and do nothing.
874  * Otherwise, if nonblocking, send as much as possible.
875  * The data to be sent is described by "uio" if nonzero,
876  * otherwise by the mbuf chain "top" (which must be null
877  * if uio is not).  Data provided in mbuf chain must be small
878  * enough to send all at once.
879  *
880  * Returns nonzero on error, timeout or signal; callers
881  * must check for short counts if EINTR/ERESTART are returned.
882  * Data and control buffers are freed on return.
883  */
884 int
885 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
886 	struct mbuf *control, int flags, struct lwp *l)
887 {
888 	struct mbuf	**mp, *m;
889 	long		space, len, resid, clen, mlen;
890 	int		error, s, dontroute, atomic;
891 	short		wakeup_state = 0;
892 
893 	clen = 0;
894 
895 	/*
896 	 * solock() provides atomicity of access.  splsoftnet() prevents
897 	 * protocol processing soft interrupts from interrupting us and
898 	 * blocking (expensive).
899 	 */
900 	s = splsoftnet();
901 	solock(so);
902 	atomic = sosendallatonce(so) || top;
903 	if (uio)
904 		resid = uio->uio_resid;
905 	else
906 		resid = top->m_pkthdr.len;
907 	/*
908 	 * In theory resid should be unsigned.
909 	 * However, space must be signed, as it might be less than 0
910 	 * if we over-committed, and we must use a signed comparison
911 	 * of space and resid.  On the other hand, a negative resid
912 	 * causes us to loop sending 0-length segments to the protocol.
913 	 */
914 	if (resid < 0) {
915 		error = EINVAL;
916 		goto out;
917 	}
918 	dontroute =
919 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
920 	    (so->so_proto->pr_flags & PR_ATOMIC);
921 	l->l_ru.ru_msgsnd++;
922 	if (control)
923 		clen = control->m_len;
924  restart:
925 	if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
926 		goto out;
927 	do {
928 		if (so->so_state & SS_CANTSENDMORE) {
929 			error = EPIPE;
930 			goto release;
931 		}
932 		if (so->so_error) {
933 			error = so->so_error;
934 			so->so_error = 0;
935 			goto release;
936 		}
937 		if ((so->so_state & SS_ISCONNECTED) == 0) {
938 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
939 				if (resid || clen == 0) {
940 					error = ENOTCONN;
941 					goto release;
942 				}
943 			} else if (addr == 0) {
944 				error = EDESTADDRREQ;
945 				goto release;
946 			}
947 		}
948 		space = sbspace(&so->so_snd);
949 		if (flags & MSG_OOB)
950 			space += 1024;
951 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
952 		    clen > so->so_snd.sb_hiwat) {
953 			error = EMSGSIZE;
954 			goto release;
955 		}
956 		if (space < resid + clen &&
957 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
958 			if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
959 				error = EWOULDBLOCK;
960 				goto release;
961 			}
962 			sbunlock(&so->so_snd);
963 			if (wakeup_state & SS_RESTARTSYS) {
964 				error = ERESTART;
965 				goto out;
966 			}
967 			error = sbwait(&so->so_snd);
968 			if (error)
969 				goto out;
970 			wakeup_state = so->so_state;
971 			goto restart;
972 		}
973 		wakeup_state = 0;
974 		mp = &top;
975 		space -= clen;
976 		do {
977 			if (uio == NULL) {
978 				/*
979 				 * Data is prepackaged in "top".
980 				 */
981 				resid = 0;
982 				if (flags & MSG_EOR)
983 					top->m_flags |= M_EOR;
984 			} else do {
985 				sounlock(so);
986 				splx(s);
987 				if (top == NULL) {
988 					m = m_gethdr(M_WAIT, MT_DATA);
989 					mlen = MHLEN;
990 					m->m_pkthdr.len = 0;
991 					m->m_pkthdr.rcvif = NULL;
992 				} else {
993 					m = m_get(M_WAIT, MT_DATA);
994 					mlen = MLEN;
995 				}
996 				MCLAIM(m, so->so_snd.sb_mowner);
997 				if (sock_loan_thresh >= 0 &&
998 				    uio->uio_iov->iov_len >= sock_loan_thresh &&
999 				    space >= sock_loan_thresh &&
1000 				    (len = sosend_loan(so, uio, m,
1001 						       space)) != 0) {
1002 					SOSEND_COUNTER_INCR(&sosend_loan_big);
1003 					space -= len;
1004 					goto have_data;
1005 				}
1006 				if (resid >= MINCLSIZE && space >= MCLBYTES) {
1007 					SOSEND_COUNTER_INCR(&sosend_copy_big);
1008 					m_clget(m, M_DONTWAIT);
1009 					if ((m->m_flags & M_EXT) == 0)
1010 						goto nopages;
1011 					mlen = MCLBYTES;
1012 					if (atomic && top == 0) {
1013 						len = lmin(MCLBYTES - max_hdr,
1014 						    resid);
1015 						m->m_data += max_hdr;
1016 					} else
1017 						len = lmin(MCLBYTES, resid);
1018 					space -= len;
1019 				} else {
1020  nopages:
1021 					SOSEND_COUNTER_INCR(&sosend_copy_small);
1022 					len = lmin(lmin(mlen, resid), space);
1023 					space -= len;
1024 					/*
1025 					 * For datagram protocols, leave room
1026 					 * for protocol headers in first mbuf.
1027 					 */
1028 					if (atomic && top == 0 && len < mlen)
1029 						MH_ALIGN(m, len);
1030 				}
1031 				error = uiomove(mtod(m, void *), (int)len, uio);
1032  have_data:
1033 				resid = uio->uio_resid;
1034 				m->m_len = len;
1035 				*mp = m;
1036 				top->m_pkthdr.len += len;
1037 				s = splsoftnet();
1038 				solock(so);
1039 				if (error != 0)
1040 					goto release;
1041 				mp = &m->m_next;
1042 				if (resid <= 0) {
1043 					if (flags & MSG_EOR)
1044 						top->m_flags |= M_EOR;
1045 					break;
1046 				}
1047 			} while (space > 0 && atomic);
1048 
1049 			if (so->so_state & SS_CANTSENDMORE) {
1050 				error = EPIPE;
1051 				goto release;
1052 			}
1053 			if (dontroute)
1054 				so->so_options |= SO_DONTROUTE;
1055 			if (resid > 0)
1056 				so->so_state |= SS_MORETOCOME;
1057 			error = (*so->so_proto->pr_usrreqs->pr_generic)(so,
1058 			    (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND,
1059 			    top, addr, control, curlwp);
1060 			if (dontroute)
1061 				so->so_options &= ~SO_DONTROUTE;
1062 			if (resid > 0)
1063 				so->so_state &= ~SS_MORETOCOME;
1064 			clen = 0;
1065 			control = NULL;
1066 			top = NULL;
1067 			mp = &top;
1068 			if (error != 0)
1069 				goto release;
1070 		} while (resid && space > 0);
1071 	} while (resid);
1072 
1073  release:
1074 	sbunlock(&so->so_snd);
1075  out:
1076 	sounlock(so);
1077 	splx(s);
1078 	if (top)
1079 		m_freem(top);
1080 	if (control)
1081 		m_freem(control);
1082 	return (error);
1083 }
1084 
1085 /*
1086  * Following replacement or removal of the first mbuf on the first
1087  * mbuf chain of a socket buffer, push necessary state changes back
1088  * into the socket buffer so that other consumers see the values
1089  * consistently.  'nextrecord' is the callers locally stored value of
1090  * the original value of sb->sb_mb->m_nextpkt which must be restored
1091  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
1092  */
1093 static void
1094 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
1095 {
1096 
1097 	KASSERT(solocked(sb->sb_so));
1098 
1099 	/*
1100 	 * First, update for the new value of nextrecord.  If necessary,
1101 	 * make it the first record.
1102 	 */
1103 	if (sb->sb_mb != NULL)
1104 		sb->sb_mb->m_nextpkt = nextrecord;
1105 	else
1106 		sb->sb_mb = nextrecord;
1107 
1108         /*
1109          * Now update any dependent socket buffer fields to reflect
1110          * the new state.  This is an inline of SB_EMPTY_FIXUP, with
1111          * the addition of a second clause that takes care of the
1112          * case where sb_mb has been updated, but remains the last
1113          * record.
1114          */
1115         if (sb->sb_mb == NULL) {
1116                 sb->sb_mbtail = NULL;
1117                 sb->sb_lastrecord = NULL;
1118         } else if (sb->sb_mb->m_nextpkt == NULL)
1119                 sb->sb_lastrecord = sb->sb_mb;
1120 }
1121 
1122 /*
1123  * Implement receive operations on a socket.
1124  * We depend on the way that records are added to the sockbuf
1125  * by sbappend*.  In particular, each record (mbufs linked through m_next)
1126  * must begin with an address if the protocol so specifies,
1127  * followed by an optional mbuf or mbufs containing ancillary data,
1128  * and then zero or more mbufs of data.
1129  * In order to avoid blocking network interrupts for the entire time here,
1130  * we splx() while doing the actual copy to user space.
1131  * Although the sockbuf is locked, new data may still be appended,
1132  * and thus we must maintain consistency of the sockbuf during that time.
1133  *
1134  * The caller may receive the data as a single mbuf chain by supplying
1135  * an mbuf **mp0 for use in returning the chain.  The uio is then used
1136  * only for the count in uio_resid.
1137  */
1138 int
1139 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
1140 	struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1141 {
1142 	struct lwp *l = curlwp;
1143 	struct mbuf	*m, **mp, *mt;
1144 	size_t len, offset, moff, orig_resid;
1145 	int atomic, flags, error, s, type;
1146 	const struct protosw	*pr;
1147 	struct mbuf	*nextrecord;
1148 	int		mbuf_removed = 0;
1149 	const struct domain *dom;
1150 	short		wakeup_state = 0;
1151 
1152 	pr = so->so_proto;
1153 	atomic = pr->pr_flags & PR_ATOMIC;
1154 	dom = pr->pr_domain;
1155 	mp = mp0;
1156 	type = 0;
1157 	orig_resid = uio->uio_resid;
1158 
1159 	if (paddr != NULL)
1160 		*paddr = NULL;
1161 	if (controlp != NULL)
1162 		*controlp = NULL;
1163 	if (flagsp != NULL)
1164 		flags = *flagsp &~ MSG_EOR;
1165 	else
1166 		flags = 0;
1167 
1168 	if (flags & MSG_OOB) {
1169 		m = m_get(M_WAIT, MT_DATA);
1170 		solock(so);
1171 		error = (*pr->pr_usrreqs->pr_generic)(so, PRU_RCVOOB, m,
1172 		    (struct mbuf *)(long)(flags & MSG_PEEK), NULL, l);
1173 		sounlock(so);
1174 		if (error)
1175 			goto bad;
1176 		do {
1177 			error = uiomove(mtod(m, void *),
1178 			    MIN(uio->uio_resid, m->m_len), uio);
1179 			m = m_free(m);
1180 		} while (uio->uio_resid > 0 && error == 0 && m);
1181  bad:
1182 		if (m != NULL)
1183 			m_freem(m);
1184 		return error;
1185 	}
1186 	if (mp != NULL)
1187 		*mp = NULL;
1188 
1189 	/*
1190 	 * solock() provides atomicity of access.  splsoftnet() prevents
1191 	 * protocol processing soft interrupts from interrupting us and
1192 	 * blocking (expensive).
1193 	 */
1194 	s = splsoftnet();
1195 	solock(so);
1196  restart:
1197 	if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) {
1198 		sounlock(so);
1199 		splx(s);
1200 		return error;
1201 	}
1202 
1203 	m = so->so_rcv.sb_mb;
1204 	/*
1205 	 * If we have less data than requested, block awaiting more
1206 	 * (subject to any timeout) if:
1207 	 *   1. the current count is less than the low water mark,
1208 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1209 	 *	receive operation at once if we block (resid <= hiwat), or
1210 	 *   3. MSG_DONTWAIT is not set.
1211 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1212 	 * we have to do the receive in sections, and thus risk returning
1213 	 * a short count if a timeout or signal occurs after we start.
1214 	 */
1215 	if (m == NULL ||
1216 	    ((flags & MSG_DONTWAIT) == 0 &&
1217 	     so->so_rcv.sb_cc < uio->uio_resid &&
1218 	     (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1219 	      ((flags & MSG_WAITALL) &&
1220 	       uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1221 	     m->m_nextpkt == NULL && !atomic)) {
1222 #ifdef DIAGNOSTIC
1223 		if (m == NULL && so->so_rcv.sb_cc)
1224 			panic("receive 1");
1225 #endif
1226 		if (so->so_error) {
1227 			if (m != NULL)
1228 				goto dontblock;
1229 			error = so->so_error;
1230 			if ((flags & MSG_PEEK) == 0)
1231 				so->so_error = 0;
1232 			goto release;
1233 		}
1234 		if (so->so_state & SS_CANTRCVMORE) {
1235 			if (m != NULL)
1236 				goto dontblock;
1237 			else
1238 				goto release;
1239 		}
1240 		for (; m != NULL; m = m->m_next)
1241 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1242 				m = so->so_rcv.sb_mb;
1243 				goto dontblock;
1244 			}
1245 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1246 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1247 			error = ENOTCONN;
1248 			goto release;
1249 		}
1250 		if (uio->uio_resid == 0)
1251 			goto release;
1252 		if ((so->so_state & SS_NBIO) ||
1253 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1254 			error = EWOULDBLOCK;
1255 			goto release;
1256 		}
1257 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
1258 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
1259 		sbunlock(&so->so_rcv);
1260 		if (wakeup_state & SS_RESTARTSYS)
1261 			error = ERESTART;
1262 		else
1263 			error = sbwait(&so->so_rcv);
1264 		if (error != 0) {
1265 			sounlock(so);
1266 			splx(s);
1267 			return error;
1268 		}
1269 		wakeup_state = so->so_state;
1270 		goto restart;
1271 	}
1272  dontblock:
1273 	/*
1274 	 * On entry here, m points to the first record of the socket buffer.
1275 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1276 	 * pointer to the next record in the socket buffer.  We must keep the
1277 	 * various socket buffer pointers and local stack versions of the
1278 	 * pointers in sync, pushing out modifications before dropping the
1279 	 * socket lock, and re-reading them when picking it up.
1280 	 *
1281 	 * Otherwise, we will race with the network stack appending new data
1282 	 * or records onto the socket buffer by using inconsistent/stale
1283 	 * versions of the field, possibly resulting in socket buffer
1284 	 * corruption.
1285 	 *
1286 	 * By holding the high-level sblock(), we prevent simultaneous
1287 	 * readers from pulling off the front of the socket buffer.
1288 	 */
1289 	if (l != NULL)
1290 		l->l_ru.ru_msgrcv++;
1291 	KASSERT(m == so->so_rcv.sb_mb);
1292 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
1293 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
1294 	nextrecord = m->m_nextpkt;
1295 	if (pr->pr_flags & PR_ADDR) {
1296 #ifdef DIAGNOSTIC
1297 		if (m->m_type != MT_SONAME)
1298 			panic("receive 1a");
1299 #endif
1300 		orig_resid = 0;
1301 		if (flags & MSG_PEEK) {
1302 			if (paddr)
1303 				*paddr = m_copy(m, 0, m->m_len);
1304 			m = m->m_next;
1305 		} else {
1306 			sbfree(&so->so_rcv, m);
1307 			mbuf_removed = 1;
1308 			if (paddr != NULL) {
1309 				*paddr = m;
1310 				so->so_rcv.sb_mb = m->m_next;
1311 				m->m_next = NULL;
1312 				m = so->so_rcv.sb_mb;
1313 			} else {
1314 				MFREE(m, so->so_rcv.sb_mb);
1315 				m = so->so_rcv.sb_mb;
1316 			}
1317 			sbsync(&so->so_rcv, nextrecord);
1318 		}
1319 	}
1320 
1321 	/*
1322 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1323 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1324 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1325 	 * perform externalization (or freeing if controlp == NULL).
1326 	 */
1327 	if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) {
1328 		struct mbuf *cm = NULL, *cmn;
1329 		struct mbuf **cme = &cm;
1330 
1331 		do {
1332 			if (flags & MSG_PEEK) {
1333 				if (controlp != NULL) {
1334 					*controlp = m_copy(m, 0, m->m_len);
1335 					controlp = &(*controlp)->m_next;
1336 				}
1337 				m = m->m_next;
1338 			} else {
1339 				sbfree(&so->so_rcv, m);
1340 				so->so_rcv.sb_mb = m->m_next;
1341 				m->m_next = NULL;
1342 				*cme = m;
1343 				cme = &(*cme)->m_next;
1344 				m = so->so_rcv.sb_mb;
1345 			}
1346 		} while (m != NULL && m->m_type == MT_CONTROL);
1347 		if ((flags & MSG_PEEK) == 0)
1348 			sbsync(&so->so_rcv, nextrecord);
1349 		for (; cm != NULL; cm = cmn) {
1350 			cmn = cm->m_next;
1351 			cm->m_next = NULL;
1352 			type = mtod(cm, struct cmsghdr *)->cmsg_type;
1353 			if (controlp != NULL) {
1354 				if (dom->dom_externalize != NULL &&
1355 				    type == SCM_RIGHTS) {
1356 					sounlock(so);
1357 					splx(s);
1358 					error = (*dom->dom_externalize)(cm, l,
1359 					    (flags & MSG_CMSG_CLOEXEC) ?
1360 					    O_CLOEXEC : 0);
1361 					s = splsoftnet();
1362 					solock(so);
1363 				}
1364 				*controlp = cm;
1365 				while (*controlp != NULL)
1366 					controlp = &(*controlp)->m_next;
1367 			} else {
1368 				/*
1369 				 * Dispose of any SCM_RIGHTS message that went
1370 				 * through the read path rather than recv.
1371 				 */
1372 				if (dom->dom_dispose != NULL &&
1373 				    type == SCM_RIGHTS) {
1374 				    	sounlock(so);
1375 					(*dom->dom_dispose)(cm);
1376 					solock(so);
1377 				}
1378 				m_freem(cm);
1379 			}
1380 		}
1381 		if (m != NULL)
1382 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1383 		else
1384 			nextrecord = so->so_rcv.sb_mb;
1385 		orig_resid = 0;
1386 	}
1387 
1388 	/* If m is non-NULL, we have some data to read. */
1389 	if (__predict_true(m != NULL)) {
1390 		type = m->m_type;
1391 		if (type == MT_OOBDATA)
1392 			flags |= MSG_OOB;
1393 	}
1394 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1395 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1396 
1397 	moff = 0;
1398 	offset = 0;
1399 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1400 		if (m->m_type == MT_OOBDATA) {
1401 			if (type != MT_OOBDATA)
1402 				break;
1403 		} else if (type == MT_OOBDATA)
1404 			break;
1405 #ifdef DIAGNOSTIC
1406 		else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
1407 			panic("receive 3");
1408 #endif
1409 		so->so_state &= ~SS_RCVATMARK;
1410 		wakeup_state = 0;
1411 		len = uio->uio_resid;
1412 		if (so->so_oobmark && len > so->so_oobmark - offset)
1413 			len = so->so_oobmark - offset;
1414 		if (len > m->m_len - moff)
1415 			len = m->m_len - moff;
1416 		/*
1417 		 * If mp is set, just pass back the mbufs.
1418 		 * Otherwise copy them out via the uio, then free.
1419 		 * Sockbuf must be consistent here (points to current mbuf,
1420 		 * it points to next record) when we drop priority;
1421 		 * we must note any additions to the sockbuf when we
1422 		 * block interrupts again.
1423 		 */
1424 		if (mp == NULL) {
1425 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1426 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1427 			sounlock(so);
1428 			splx(s);
1429 			error = uiomove(mtod(m, char *) + moff, len, uio);
1430 			s = splsoftnet();
1431 			solock(so);
1432 			if (error != 0) {
1433 				/*
1434 				 * If any part of the record has been removed
1435 				 * (such as the MT_SONAME mbuf, which will
1436 				 * happen when PR_ADDR, and thus also
1437 				 * PR_ATOMIC, is set), then drop the entire
1438 				 * record to maintain the atomicity of the
1439 				 * receive operation.
1440 				 *
1441 				 * This avoids a later panic("receive 1a")
1442 				 * when compiled with DIAGNOSTIC.
1443 				 */
1444 				if (m && mbuf_removed && atomic)
1445 					(void) sbdroprecord(&so->so_rcv);
1446 
1447 				goto release;
1448 			}
1449 		} else
1450 			uio->uio_resid -= len;
1451 		if (len == m->m_len - moff) {
1452 			if (m->m_flags & M_EOR)
1453 				flags |= MSG_EOR;
1454 			if (flags & MSG_PEEK) {
1455 				m = m->m_next;
1456 				moff = 0;
1457 			} else {
1458 				nextrecord = m->m_nextpkt;
1459 				sbfree(&so->so_rcv, m);
1460 				if (mp) {
1461 					*mp = m;
1462 					mp = &m->m_next;
1463 					so->so_rcv.sb_mb = m = m->m_next;
1464 					*mp = NULL;
1465 				} else {
1466 					MFREE(m, so->so_rcv.sb_mb);
1467 					m = so->so_rcv.sb_mb;
1468 				}
1469 				/*
1470 				 * If m != NULL, we also know that
1471 				 * so->so_rcv.sb_mb != NULL.
1472 				 */
1473 				KASSERT(so->so_rcv.sb_mb == m);
1474 				if (m) {
1475 					m->m_nextpkt = nextrecord;
1476 					if (nextrecord == NULL)
1477 						so->so_rcv.sb_lastrecord = m;
1478 				} else {
1479 					so->so_rcv.sb_mb = nextrecord;
1480 					SB_EMPTY_FIXUP(&so->so_rcv);
1481 				}
1482 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1483 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1484 			}
1485 		} else if (flags & MSG_PEEK)
1486 			moff += len;
1487 		else {
1488 			if (mp != NULL) {
1489 				mt = m_copym(m, 0, len, M_NOWAIT);
1490 				if (__predict_false(mt == NULL)) {
1491 					sounlock(so);
1492 					mt = m_copym(m, 0, len, M_WAIT);
1493 					solock(so);
1494 				}
1495 				*mp = mt;
1496 			}
1497 			m->m_data += len;
1498 			m->m_len -= len;
1499 			so->so_rcv.sb_cc -= len;
1500 		}
1501 		if (so->so_oobmark) {
1502 			if ((flags & MSG_PEEK) == 0) {
1503 				so->so_oobmark -= len;
1504 				if (so->so_oobmark == 0) {
1505 					so->so_state |= SS_RCVATMARK;
1506 					break;
1507 				}
1508 			} else {
1509 				offset += len;
1510 				if (offset == so->so_oobmark)
1511 					break;
1512 			}
1513 		}
1514 		if (flags & MSG_EOR)
1515 			break;
1516 		/*
1517 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1518 		 * we must not quit until "uio->uio_resid == 0" or an error
1519 		 * termination.  If a signal/timeout occurs, return
1520 		 * with a short count but without error.
1521 		 * Keep sockbuf locked against other readers.
1522 		 */
1523 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1524 		    !sosendallatonce(so) && !nextrecord) {
1525 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1526 				break;
1527 			/*
1528 			 * If we are peeking and the socket receive buffer is
1529 			 * full, stop since we can't get more data to peek at.
1530 			 */
1531 			if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
1532 				break;
1533 			/*
1534 			 * If we've drained the socket buffer, tell the
1535 			 * protocol in case it needs to do something to
1536 			 * get it filled again.
1537 			 */
1538 			if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1539 				(*pr->pr_usrreqs->pr_generic)(so, PRU_RCVD,
1540 				    NULL, (struct mbuf *)(long)flags, NULL, l);
1541 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1542 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1543 			if (wakeup_state & SS_RESTARTSYS)
1544 				error = ERESTART;
1545 			else
1546 				error = sbwait(&so->so_rcv);
1547 			if (error != 0) {
1548 				sbunlock(&so->so_rcv);
1549 				sounlock(so);
1550 				splx(s);
1551 				return 0;
1552 			}
1553 			if ((m = so->so_rcv.sb_mb) != NULL)
1554 				nextrecord = m->m_nextpkt;
1555 			wakeup_state = so->so_state;
1556 		}
1557 	}
1558 
1559 	if (m && atomic) {
1560 		flags |= MSG_TRUNC;
1561 		if ((flags & MSG_PEEK) == 0)
1562 			(void) sbdroprecord(&so->so_rcv);
1563 	}
1564 	if ((flags & MSG_PEEK) == 0) {
1565 		if (m == NULL) {
1566 			/*
1567 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1568 			 * part makes sure sb_lastrecord is up-to-date if
1569 			 * there is still data in the socket buffer.
1570 			 */
1571 			so->so_rcv.sb_mb = nextrecord;
1572 			if (so->so_rcv.sb_mb == NULL) {
1573 				so->so_rcv.sb_mbtail = NULL;
1574 				so->so_rcv.sb_lastrecord = NULL;
1575 			} else if (nextrecord->m_nextpkt == NULL)
1576 				so->so_rcv.sb_lastrecord = nextrecord;
1577 		}
1578 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1579 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1580 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1581 			(*pr->pr_usrreqs->pr_generic)(so, PRU_RCVD, NULL,
1582 			    (struct mbuf *)(long)flags, NULL, l);
1583 	}
1584 	if (orig_resid == uio->uio_resid && orig_resid &&
1585 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1586 		sbunlock(&so->so_rcv);
1587 		goto restart;
1588 	}
1589 
1590 	if (flagsp != NULL)
1591 		*flagsp |= flags;
1592  release:
1593 	sbunlock(&so->so_rcv);
1594 	sounlock(so);
1595 	splx(s);
1596 	return error;
1597 }
1598 
1599 int
1600 soshutdown(struct socket *so, int how)
1601 {
1602 	const struct protosw	*pr;
1603 	int	error;
1604 
1605 	KASSERT(solocked(so));
1606 
1607 	pr = so->so_proto;
1608 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1609 		return (EINVAL);
1610 
1611 	if (how == SHUT_RD || how == SHUT_RDWR) {
1612 		sorflush(so);
1613 		error = 0;
1614 	}
1615 	if (how == SHUT_WR || how == SHUT_RDWR)
1616 		error = (*pr->pr_usrreqs->pr_generic)(so,
1617 		    PRU_SHUTDOWN, NULL, NULL, NULL, NULL);
1618 
1619 	return error;
1620 }
1621 
1622 void
1623 sorestart(struct socket *so)
1624 {
1625 	/*
1626 	 * An application has called close() on an fd on which another
1627 	 * of its threads has called a socket system call.
1628 	 * Mark this and wake everyone up, and code that would block again
1629 	 * instead returns ERESTART.
1630 	 * On system call re-entry the fd is validated and EBADF returned.
1631 	 * Any other fd will block again on the 2nd syscall.
1632 	 */
1633 	solock(so);
1634 	so->so_state |= SS_RESTARTSYS;
1635 	cv_broadcast(&so->so_cv);
1636 	cv_broadcast(&so->so_snd.sb_cv);
1637 	cv_broadcast(&so->so_rcv.sb_cv);
1638 	sounlock(so);
1639 }
1640 
1641 void
1642 sorflush(struct socket *so)
1643 {
1644 	struct sockbuf	*sb, asb;
1645 	const struct protosw	*pr;
1646 
1647 	KASSERT(solocked(so));
1648 
1649 	sb = &so->so_rcv;
1650 	pr = so->so_proto;
1651 	socantrcvmore(so);
1652 	sb->sb_flags |= SB_NOINTR;
1653 	(void )sblock(sb, M_WAITOK);
1654 	sbunlock(sb);
1655 	asb = *sb;
1656 	/*
1657 	 * Clear most of the sockbuf structure, but leave some of the
1658 	 * fields valid.
1659 	 */
1660 	memset(&sb->sb_startzero, 0,
1661 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1662 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) {
1663 		sounlock(so);
1664 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1665 		solock(so);
1666 	}
1667 	sbrelease(&asb, so);
1668 }
1669 
1670 /*
1671  * internal set SOL_SOCKET options
1672  */
1673 static int
1674 sosetopt1(struct socket *so, const struct sockopt *sopt)
1675 {
1676 	int error = EINVAL, opt;
1677 	int optval = 0; /* XXX: gcc */
1678 	struct linger l;
1679 	struct timeval tv;
1680 
1681 	switch ((opt = sopt->sopt_name)) {
1682 
1683 	case SO_ACCEPTFILTER:
1684 		error = accept_filt_setopt(so, sopt);
1685 		KASSERT(solocked(so));
1686 		break;
1687 
1688   	case SO_LINGER:
1689  		error = sockopt_get(sopt, &l, sizeof(l));
1690 		solock(so);
1691  		if (error)
1692  			break;
1693  		if (l.l_linger < 0 || l.l_linger > USHRT_MAX ||
1694  		    l.l_linger > (INT_MAX / hz)) {
1695 			error = EDOM;
1696 			break;
1697 		}
1698  		so->so_linger = l.l_linger;
1699  		if (l.l_onoff)
1700  			so->so_options |= SO_LINGER;
1701  		else
1702  			so->so_options &= ~SO_LINGER;
1703    		break;
1704 
1705 	case SO_DEBUG:
1706 	case SO_KEEPALIVE:
1707 	case SO_DONTROUTE:
1708 	case SO_USELOOPBACK:
1709 	case SO_BROADCAST:
1710 	case SO_REUSEADDR:
1711 	case SO_REUSEPORT:
1712 	case SO_OOBINLINE:
1713 	case SO_TIMESTAMP:
1714 	case SO_NOSIGPIPE:
1715 #ifdef SO_OTIMESTAMP
1716 	case SO_OTIMESTAMP:
1717 #endif
1718 		error = sockopt_getint(sopt, &optval);
1719 		solock(so);
1720 		if (error)
1721 			break;
1722 		if (optval)
1723 			so->so_options |= opt;
1724 		else
1725 			so->so_options &= ~opt;
1726 		break;
1727 
1728 	case SO_SNDBUF:
1729 	case SO_RCVBUF:
1730 	case SO_SNDLOWAT:
1731 	case SO_RCVLOWAT:
1732 		error = sockopt_getint(sopt, &optval);
1733 		solock(so);
1734 		if (error)
1735 			break;
1736 
1737 		/*
1738 		 * Values < 1 make no sense for any of these
1739 		 * options, so disallow them.
1740 		 */
1741 		if (optval < 1) {
1742 			error = EINVAL;
1743 			break;
1744 		}
1745 
1746 		switch (opt) {
1747 		case SO_SNDBUF:
1748 			if (sbreserve(&so->so_snd, (u_long)optval, so) == 0) {
1749 				error = ENOBUFS;
1750 				break;
1751 			}
1752 			so->so_snd.sb_flags &= ~SB_AUTOSIZE;
1753 			break;
1754 
1755 		case SO_RCVBUF:
1756 			if (sbreserve(&so->so_rcv, (u_long)optval, so) == 0) {
1757 				error = ENOBUFS;
1758 				break;
1759 			}
1760 			so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1761 			break;
1762 
1763 		/*
1764 		 * Make sure the low-water is never greater than
1765 		 * the high-water.
1766 		 */
1767 		case SO_SNDLOWAT:
1768 			if (optval > so->so_snd.sb_hiwat)
1769 				optval = so->so_snd.sb_hiwat;
1770 
1771 			so->so_snd.sb_lowat = optval;
1772 			break;
1773 
1774 		case SO_RCVLOWAT:
1775 			if (optval > so->so_rcv.sb_hiwat)
1776 				optval = so->so_rcv.sb_hiwat;
1777 
1778 			so->so_rcv.sb_lowat = optval;
1779 			break;
1780 		}
1781 		break;
1782 
1783 #ifdef COMPAT_50
1784 	case SO_OSNDTIMEO:
1785 	case SO_ORCVTIMEO: {
1786 		struct timeval50 otv;
1787 		error = sockopt_get(sopt, &otv, sizeof(otv));
1788 		if (error) {
1789 			solock(so);
1790 			break;
1791 		}
1792 		timeval50_to_timeval(&otv, &tv);
1793 		opt = opt == SO_OSNDTIMEO ? SO_SNDTIMEO : SO_RCVTIMEO;
1794 		error = 0;
1795 		/*FALLTHROUGH*/
1796 	}
1797 #endif /* COMPAT_50 */
1798 
1799 	case SO_SNDTIMEO:
1800 	case SO_RCVTIMEO:
1801 		if (error)
1802 			error = sockopt_get(sopt, &tv, sizeof(tv));
1803 		solock(so);
1804 		if (error)
1805 			break;
1806 
1807 		if (tv.tv_sec > (INT_MAX - tv.tv_usec / tick) / hz) {
1808 			error = EDOM;
1809 			break;
1810 		}
1811 
1812 		optval = tv.tv_sec * hz + tv.tv_usec / tick;
1813 		if (optval == 0 && tv.tv_usec != 0)
1814 			optval = 1;
1815 
1816 		switch (opt) {
1817 		case SO_SNDTIMEO:
1818 			so->so_snd.sb_timeo = optval;
1819 			break;
1820 		case SO_RCVTIMEO:
1821 			so->so_rcv.sb_timeo = optval;
1822 			break;
1823 		}
1824 		break;
1825 
1826 	default:
1827 		solock(so);
1828 		error = ENOPROTOOPT;
1829 		break;
1830 	}
1831 	KASSERT(solocked(so));
1832 	return error;
1833 }
1834 
1835 int
1836 sosetopt(struct socket *so, struct sockopt *sopt)
1837 {
1838 	int error, prerr;
1839 
1840 	if (sopt->sopt_level == SOL_SOCKET) {
1841 		error = sosetopt1(so, sopt);
1842 		KASSERT(solocked(so));
1843 	} else {
1844 		error = ENOPROTOOPT;
1845 		solock(so);
1846 	}
1847 
1848 	if ((error == 0 || error == ENOPROTOOPT) &&
1849 	    so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) {
1850 		/* give the protocol stack a shot */
1851 		prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, sopt);
1852 		if (prerr == 0)
1853 			error = 0;
1854 		else if (prerr != ENOPROTOOPT)
1855 			error = prerr;
1856 	}
1857 	sounlock(so);
1858 	return error;
1859 }
1860 
1861 /*
1862  * so_setsockopt() is a wrapper providing a sockopt structure for sosetopt()
1863  */
1864 int
1865 so_setsockopt(struct lwp *l, struct socket *so, int level, int name,
1866     const void *val, size_t valsize)
1867 {
1868 	struct sockopt sopt;
1869 	int error;
1870 
1871 	KASSERT(valsize == 0 || val != NULL);
1872 
1873 	sockopt_init(&sopt, level, name, valsize);
1874 	sockopt_set(&sopt, val, valsize);
1875 
1876 	error = sosetopt(so, &sopt);
1877 
1878 	sockopt_destroy(&sopt);
1879 
1880 	return error;
1881 }
1882 
1883 /*
1884  * internal get SOL_SOCKET options
1885  */
1886 static int
1887 sogetopt1(struct socket *so, struct sockopt *sopt)
1888 {
1889 	int error, optval, opt;
1890 	struct linger l;
1891 	struct timeval tv;
1892 
1893 	switch ((opt = sopt->sopt_name)) {
1894 
1895 	case SO_ACCEPTFILTER:
1896 		error = accept_filt_getopt(so, sopt);
1897 		break;
1898 
1899 	case SO_LINGER:
1900 		l.l_onoff = (so->so_options & SO_LINGER) ? 1 : 0;
1901 		l.l_linger = so->so_linger;
1902 
1903 		error = sockopt_set(sopt, &l, sizeof(l));
1904 		break;
1905 
1906 	case SO_USELOOPBACK:
1907 	case SO_DONTROUTE:
1908 	case SO_DEBUG:
1909 	case SO_KEEPALIVE:
1910 	case SO_REUSEADDR:
1911 	case SO_REUSEPORT:
1912 	case SO_BROADCAST:
1913 	case SO_OOBINLINE:
1914 	case SO_TIMESTAMP:
1915 	case SO_NOSIGPIPE:
1916 #ifdef SO_OTIMESTAMP
1917 	case SO_OTIMESTAMP:
1918 #endif
1919 	case SO_ACCEPTCONN:
1920 		error = sockopt_setint(sopt, (so->so_options & opt) ? 1 : 0);
1921 		break;
1922 
1923 	case SO_TYPE:
1924 		error = sockopt_setint(sopt, so->so_type);
1925 		break;
1926 
1927 	case SO_ERROR:
1928 		error = sockopt_setint(sopt, so->so_error);
1929 		so->so_error = 0;
1930 		break;
1931 
1932 	case SO_SNDBUF:
1933 		error = sockopt_setint(sopt, so->so_snd.sb_hiwat);
1934 		break;
1935 
1936 	case SO_RCVBUF:
1937 		error = sockopt_setint(sopt, so->so_rcv.sb_hiwat);
1938 		break;
1939 
1940 	case SO_SNDLOWAT:
1941 		error = sockopt_setint(sopt, so->so_snd.sb_lowat);
1942 		break;
1943 
1944 	case SO_RCVLOWAT:
1945 		error = sockopt_setint(sopt, so->so_rcv.sb_lowat);
1946 		break;
1947 
1948 #ifdef COMPAT_50
1949 	case SO_OSNDTIMEO:
1950 	case SO_ORCVTIMEO: {
1951 		struct timeval50 otv;
1952 
1953 		optval = (opt == SO_OSNDTIMEO ?
1954 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1955 
1956 		otv.tv_sec = optval / hz;
1957 		otv.tv_usec = (optval % hz) * tick;
1958 
1959 		error = sockopt_set(sopt, &otv, sizeof(otv));
1960 		break;
1961 	}
1962 #endif /* COMPAT_50 */
1963 
1964 	case SO_SNDTIMEO:
1965 	case SO_RCVTIMEO:
1966 		optval = (opt == SO_SNDTIMEO ?
1967 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1968 
1969 		tv.tv_sec = optval / hz;
1970 		tv.tv_usec = (optval % hz) * tick;
1971 
1972 		error = sockopt_set(sopt, &tv, sizeof(tv));
1973 		break;
1974 
1975 	case SO_OVERFLOWED:
1976 		error = sockopt_setint(sopt, so->so_rcv.sb_overflowed);
1977 		break;
1978 
1979 	default:
1980 		error = ENOPROTOOPT;
1981 		break;
1982 	}
1983 
1984 	return (error);
1985 }
1986 
1987 int
1988 sogetopt(struct socket *so, struct sockopt *sopt)
1989 {
1990 	int		error;
1991 
1992 	solock(so);
1993 	if (sopt->sopt_level != SOL_SOCKET) {
1994 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1995 			error = ((*so->so_proto->pr_ctloutput)
1996 			    (PRCO_GETOPT, so, sopt));
1997 		} else
1998 			error = (ENOPROTOOPT);
1999 	} else {
2000 		error = sogetopt1(so, sopt);
2001 	}
2002 	sounlock(so);
2003 	return (error);
2004 }
2005 
2006 /*
2007  * alloc sockopt data buffer buffer
2008  *	- will be released at destroy
2009  */
2010 static int
2011 sockopt_alloc(struct sockopt *sopt, size_t len, km_flag_t kmflag)
2012 {
2013 
2014 	KASSERT(sopt->sopt_size == 0);
2015 
2016 	if (len > sizeof(sopt->sopt_buf)) {
2017 		sopt->sopt_data = kmem_zalloc(len, kmflag);
2018 		if (sopt->sopt_data == NULL)
2019 			return ENOMEM;
2020 	} else
2021 		sopt->sopt_data = sopt->sopt_buf;
2022 
2023 	sopt->sopt_size = len;
2024 	return 0;
2025 }
2026 
2027 /*
2028  * initialise sockopt storage
2029  *	- MAY sleep during allocation
2030  */
2031 void
2032 sockopt_init(struct sockopt *sopt, int level, int name, size_t size)
2033 {
2034 
2035 	memset(sopt, 0, sizeof(*sopt));
2036 
2037 	sopt->sopt_level = level;
2038 	sopt->sopt_name = name;
2039 	(void)sockopt_alloc(sopt, size, KM_SLEEP);
2040 }
2041 
2042 /*
2043  * destroy sockopt storage
2044  *	- will release any held memory references
2045  */
2046 void
2047 sockopt_destroy(struct sockopt *sopt)
2048 {
2049 
2050 	if (sopt->sopt_data != sopt->sopt_buf)
2051 		kmem_free(sopt->sopt_data, sopt->sopt_size);
2052 
2053 	memset(sopt, 0, sizeof(*sopt));
2054 }
2055 
2056 /*
2057  * set sockopt value
2058  *	- value is copied into sockopt
2059  * 	- memory is allocated when necessary, will not sleep
2060  */
2061 int
2062 sockopt_set(struct sockopt *sopt, const void *buf, size_t len)
2063 {
2064 	int error;
2065 
2066 	if (sopt->sopt_size == 0) {
2067 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2068 		if (error)
2069 			return error;
2070 	}
2071 
2072 	KASSERT(sopt->sopt_size == len);
2073 	memcpy(sopt->sopt_data, buf, len);
2074 	return 0;
2075 }
2076 
2077 /*
2078  * common case of set sockopt integer value
2079  */
2080 int
2081 sockopt_setint(struct sockopt *sopt, int val)
2082 {
2083 
2084 	return sockopt_set(sopt, &val, sizeof(int));
2085 }
2086 
2087 /*
2088  * get sockopt value
2089  *	- correct size must be given
2090  */
2091 int
2092 sockopt_get(const struct sockopt *sopt, void *buf, size_t len)
2093 {
2094 
2095 	if (sopt->sopt_size != len)
2096 		return EINVAL;
2097 
2098 	memcpy(buf, sopt->sopt_data, len);
2099 	return 0;
2100 }
2101 
2102 /*
2103  * common case of get sockopt integer value
2104  */
2105 int
2106 sockopt_getint(const struct sockopt *sopt, int *valp)
2107 {
2108 
2109 	return sockopt_get(sopt, valp, sizeof(int));
2110 }
2111 
2112 /*
2113  * set sockopt value from mbuf
2114  *	- ONLY for legacy code
2115  *	- mbuf is released by sockopt
2116  *	- will not sleep
2117  */
2118 int
2119 sockopt_setmbuf(struct sockopt *sopt, struct mbuf *m)
2120 {
2121 	size_t len;
2122 	int error;
2123 
2124 	len = m_length(m);
2125 
2126 	if (sopt->sopt_size == 0) {
2127 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2128 		if (error)
2129 			return error;
2130 	}
2131 
2132 	KASSERT(sopt->sopt_size == len);
2133 	m_copydata(m, 0, len, sopt->sopt_data);
2134 	m_freem(m);
2135 
2136 	return 0;
2137 }
2138 
2139 /*
2140  * get sockopt value into mbuf
2141  *	- ONLY for legacy code
2142  *	- mbuf to be released by the caller
2143  *	- will not sleep
2144  */
2145 struct mbuf *
2146 sockopt_getmbuf(const struct sockopt *sopt)
2147 {
2148 	struct mbuf *m;
2149 
2150 	if (sopt->sopt_size > MCLBYTES)
2151 		return NULL;
2152 
2153 	m = m_get(M_DONTWAIT, MT_SOOPTS);
2154 	if (m == NULL)
2155 		return NULL;
2156 
2157 	if (sopt->sopt_size > MLEN) {
2158 		MCLGET(m, M_DONTWAIT);
2159 		if ((m->m_flags & M_EXT) == 0) {
2160 			m_free(m);
2161 			return NULL;
2162 		}
2163 	}
2164 
2165 	memcpy(mtod(m, void *), sopt->sopt_data, sopt->sopt_size);
2166 	m->m_len = sopt->sopt_size;
2167 
2168 	return m;
2169 }
2170 
2171 void
2172 sohasoutofband(struct socket *so)
2173 {
2174 
2175 	fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
2176 	selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, NOTE_SUBMIT);
2177 }
2178 
2179 static void
2180 filt_sordetach(struct knote *kn)
2181 {
2182 	struct socket	*so;
2183 
2184 	so = ((file_t *)kn->kn_obj)->f_data;
2185 	solock(so);
2186 	SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
2187 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
2188 		so->so_rcv.sb_flags &= ~SB_KNOTE;
2189 	sounlock(so);
2190 }
2191 
2192 /*ARGSUSED*/
2193 static int
2194 filt_soread(struct knote *kn, long hint)
2195 {
2196 	struct socket	*so;
2197 	int rv;
2198 
2199 	so = ((file_t *)kn->kn_obj)->f_data;
2200 	if (hint != NOTE_SUBMIT)
2201 		solock(so);
2202 	kn->kn_data = so->so_rcv.sb_cc;
2203 	if (so->so_state & SS_CANTRCVMORE) {
2204 		kn->kn_flags |= EV_EOF;
2205 		kn->kn_fflags = so->so_error;
2206 		rv = 1;
2207 	} else if (so->so_error)	/* temporary udp error */
2208 		rv = 1;
2209 	else if (kn->kn_sfflags & NOTE_LOWAT)
2210 		rv = (kn->kn_data >= kn->kn_sdata);
2211 	else
2212 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
2213 	if (hint != NOTE_SUBMIT)
2214 		sounlock(so);
2215 	return rv;
2216 }
2217 
2218 static void
2219 filt_sowdetach(struct knote *kn)
2220 {
2221 	struct socket	*so;
2222 
2223 	so = ((file_t *)kn->kn_obj)->f_data;
2224 	solock(so);
2225 	SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
2226 	if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
2227 		so->so_snd.sb_flags &= ~SB_KNOTE;
2228 	sounlock(so);
2229 }
2230 
2231 /*ARGSUSED*/
2232 static int
2233 filt_sowrite(struct knote *kn, long hint)
2234 {
2235 	struct socket	*so;
2236 	int rv;
2237 
2238 	so = ((file_t *)kn->kn_obj)->f_data;
2239 	if (hint != NOTE_SUBMIT)
2240 		solock(so);
2241 	kn->kn_data = sbspace(&so->so_snd);
2242 	if (so->so_state & SS_CANTSENDMORE) {
2243 		kn->kn_flags |= EV_EOF;
2244 		kn->kn_fflags = so->so_error;
2245 		rv = 1;
2246 	} else if (so->so_error)	/* temporary udp error */
2247 		rv = 1;
2248 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2249 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
2250 		rv = 0;
2251 	else if (kn->kn_sfflags & NOTE_LOWAT)
2252 		rv = (kn->kn_data >= kn->kn_sdata);
2253 	else
2254 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
2255 	if (hint != NOTE_SUBMIT)
2256 		sounlock(so);
2257 	return rv;
2258 }
2259 
2260 /*ARGSUSED*/
2261 static int
2262 filt_solisten(struct knote *kn, long hint)
2263 {
2264 	struct socket	*so;
2265 	int rv;
2266 
2267 	so = ((file_t *)kn->kn_obj)->f_data;
2268 
2269 	/*
2270 	 * Set kn_data to number of incoming connections, not
2271 	 * counting partial (incomplete) connections.
2272 	 */
2273 	if (hint != NOTE_SUBMIT)
2274 		solock(so);
2275 	kn->kn_data = so->so_qlen;
2276 	rv = (kn->kn_data > 0);
2277 	if (hint != NOTE_SUBMIT)
2278 		sounlock(so);
2279 	return rv;
2280 }
2281 
2282 static const struct filterops solisten_filtops =
2283 	{ 1, NULL, filt_sordetach, filt_solisten };
2284 static const struct filterops soread_filtops =
2285 	{ 1, NULL, filt_sordetach, filt_soread };
2286 static const struct filterops sowrite_filtops =
2287 	{ 1, NULL, filt_sowdetach, filt_sowrite };
2288 
2289 int
2290 soo_kqfilter(struct file *fp, struct knote *kn)
2291 {
2292 	struct socket	*so;
2293 	struct sockbuf	*sb;
2294 
2295 	so = ((file_t *)kn->kn_obj)->f_data;
2296 	solock(so);
2297 	switch (kn->kn_filter) {
2298 	case EVFILT_READ:
2299 		if (so->so_options & SO_ACCEPTCONN)
2300 			kn->kn_fop = &solisten_filtops;
2301 		else
2302 			kn->kn_fop = &soread_filtops;
2303 		sb = &so->so_rcv;
2304 		break;
2305 	case EVFILT_WRITE:
2306 		kn->kn_fop = &sowrite_filtops;
2307 		sb = &so->so_snd;
2308 		break;
2309 	default:
2310 		sounlock(so);
2311 		return (EINVAL);
2312 	}
2313 	SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
2314 	sb->sb_flags |= SB_KNOTE;
2315 	sounlock(so);
2316 	return (0);
2317 }
2318 
2319 static int
2320 sodopoll(struct socket *so, int events)
2321 {
2322 	int revents;
2323 
2324 	revents = 0;
2325 
2326 	if (events & (POLLIN | POLLRDNORM))
2327 		if (soreadable(so))
2328 			revents |= events & (POLLIN | POLLRDNORM);
2329 
2330 	if (events & (POLLOUT | POLLWRNORM))
2331 		if (sowritable(so))
2332 			revents |= events & (POLLOUT | POLLWRNORM);
2333 
2334 	if (events & (POLLPRI | POLLRDBAND))
2335 		if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
2336 			revents |= events & (POLLPRI | POLLRDBAND);
2337 
2338 	return revents;
2339 }
2340 
2341 int
2342 sopoll(struct socket *so, int events)
2343 {
2344 	int revents = 0;
2345 
2346 #ifndef DIAGNOSTIC
2347 	/*
2348 	 * Do a quick, unlocked check in expectation that the socket
2349 	 * will be ready for I/O.  Don't do this check if DIAGNOSTIC,
2350 	 * as the solocked() assertions will fail.
2351 	 */
2352 	if ((revents = sodopoll(so, events)) != 0)
2353 		return revents;
2354 #endif
2355 
2356 	solock(so);
2357 	if ((revents = sodopoll(so, events)) == 0) {
2358 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2359 			selrecord(curlwp, &so->so_rcv.sb_sel);
2360 			so->so_rcv.sb_flags |= SB_NOTIFY;
2361 		}
2362 
2363 		if (events & (POLLOUT | POLLWRNORM)) {
2364 			selrecord(curlwp, &so->so_snd.sb_sel);
2365 			so->so_snd.sb_flags |= SB_NOTIFY;
2366 		}
2367 	}
2368 	sounlock(so);
2369 
2370 	return revents;
2371 }
2372 
2373 
2374 #include <sys/sysctl.h>
2375 
2376 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO);
2377 static int sysctl_kern_sbmax(SYSCTLFN_PROTO);
2378 
2379 /*
2380  * sysctl helper routine for kern.somaxkva.  ensures that the given
2381  * value is not too small.
2382  * (XXX should we maybe make sure it's not too large as well?)
2383  */
2384 static int
2385 sysctl_kern_somaxkva(SYSCTLFN_ARGS)
2386 {
2387 	int error, new_somaxkva;
2388 	struct sysctlnode node;
2389 
2390 	new_somaxkva = somaxkva;
2391 	node = *rnode;
2392 	node.sysctl_data = &new_somaxkva;
2393 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2394 	if (error || newp == NULL)
2395 		return (error);
2396 
2397 	if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */
2398 		return (EINVAL);
2399 
2400 	mutex_enter(&so_pendfree_lock);
2401 	somaxkva = new_somaxkva;
2402 	cv_broadcast(&socurkva_cv);
2403 	mutex_exit(&so_pendfree_lock);
2404 
2405 	return (error);
2406 }
2407 
2408 /*
2409  * sysctl helper routine for kern.sbmax. Basically just ensures that
2410  * any new value is not too small.
2411  */
2412 static int
2413 sysctl_kern_sbmax(SYSCTLFN_ARGS)
2414 {
2415 	int error, new_sbmax;
2416 	struct sysctlnode node;
2417 
2418 	new_sbmax = sb_max;
2419 	node = *rnode;
2420 	node.sysctl_data = &new_sbmax;
2421 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2422 	if (error || newp == NULL)
2423 		return (error);
2424 
2425 	KERNEL_LOCK(1, NULL);
2426 	error = sb_max_set(new_sbmax);
2427 	KERNEL_UNLOCK_ONE(NULL);
2428 
2429 	return (error);
2430 }
2431 
2432 static void
2433 sysctl_kern_socket_setup(void)
2434 {
2435 
2436 	KASSERT(socket_sysctllog == NULL);
2437 
2438 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2439 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2440 		       CTLTYPE_INT, "somaxkva",
2441 		       SYSCTL_DESCR("Maximum amount of kernel memory to be "
2442 				    "used for socket buffers"),
2443 		       sysctl_kern_somaxkva, 0, NULL, 0,
2444 		       CTL_KERN, KERN_SOMAXKVA, CTL_EOL);
2445 
2446 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2447 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2448 		       CTLTYPE_INT, "sbmax",
2449 		       SYSCTL_DESCR("Maximum socket buffer size"),
2450 		       sysctl_kern_sbmax, 0, NULL, 0,
2451 		       CTL_KERN, KERN_SBMAX, CTL_EOL);
2452 }
2453