xref: /netbsd-src/sys/kern/uipc_socket.c (revision 37afb7eb6895c833050f8bfb1d1bb2f99f332539)
1 /*	$NetBSD: uipc_socket.c,v 1.245 2015/05/09 15:22:47 rtr 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 1.245 2015/05/09 15:22:47 rtr 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_socket = 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 sockaddr *nam, struct lwp *l)
628 {
629 	int	error;
630 
631 	solock(so);
632 	if (nam->sa_family != so->so_proto->pr_domain->dom_family) {
633 		sounlock(so);
634 		return EAFNOSUPPORT;
635 	}
636 	error = (*so->so_proto->pr_usrreqs->pr_bind)(so, nam, l);
637 	sounlock(so);
638 	return error;
639 }
640 
641 int
642 solisten(struct socket *so, int backlog, struct lwp *l)
643 {
644 	int	error;
645 
646 	solock(so);
647 	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
648 	    SS_ISDISCONNECTING)) != 0) {
649 		sounlock(so);
650 		return EINVAL;
651 	}
652 	error = (*so->so_proto->pr_usrreqs->pr_listen)(so, l);
653 	if (error != 0) {
654 		sounlock(so);
655 		return error;
656 	}
657 	if (TAILQ_EMPTY(&so->so_q))
658 		so->so_options |= SO_ACCEPTCONN;
659 	if (backlog < 0)
660 		backlog = 0;
661 	so->so_qlimit = min(backlog, somaxconn);
662 	sounlock(so);
663 	return 0;
664 }
665 
666 void
667 sofree(struct socket *so)
668 {
669 	u_int refs;
670 
671 	KASSERT(solocked(so));
672 
673 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
674 		sounlock(so);
675 		return;
676 	}
677 	if (so->so_head) {
678 		/*
679 		 * We must not decommission a socket that's on the accept(2)
680 		 * queue.  If we do, then accept(2) may hang after select(2)
681 		 * indicated that the listening socket was ready.
682 		 */
683 		if (!soqremque(so, 0)) {
684 			sounlock(so);
685 			return;
686 		}
687 	}
688 	if (so->so_rcv.sb_hiwat)
689 		(void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0,
690 		    RLIM_INFINITY);
691 	if (so->so_snd.sb_hiwat)
692 		(void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0,
693 		    RLIM_INFINITY);
694 	sbrelease(&so->so_snd, so);
695 	KASSERT(!cv_has_waiters(&so->so_cv));
696 	KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv));
697 	KASSERT(!cv_has_waiters(&so->so_snd.sb_cv));
698 	sorflush(so);
699 	refs = so->so_aborting;	/* XXX */
700 	/* Remove acccept filter if one is present. */
701 	if (so->so_accf != NULL)
702 		(void)accept_filt_clear(so);
703 	sounlock(so);
704 	if (refs == 0)		/* XXX */
705 		soput(so);
706 }
707 
708 /*
709  * soclose: close a socket on last file table reference removal.
710  * Initiate disconnect if connected.  Free socket when disconnect complete.
711  */
712 int
713 soclose(struct socket *so)
714 {
715 	struct socket *so2;
716 	int error = 0;
717 
718 	solock(so);
719 	if (so->so_options & SO_ACCEPTCONN) {
720 		for (;;) {
721 			if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) {
722 				KASSERT(solocked2(so, so2));
723 				(void) soqremque(so2, 0);
724 				/* soabort drops the lock. */
725 				(void) soabort(so2);
726 				solock(so);
727 				continue;
728 			}
729 			if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) {
730 				KASSERT(solocked2(so, so2));
731 				(void) soqremque(so2, 1);
732 				/* soabort drops the lock. */
733 				(void) soabort(so2);
734 				solock(so);
735 				continue;
736 			}
737 			break;
738 		}
739 	}
740 	if (so->so_pcb == NULL)
741 		goto discard;
742 	if (so->so_state & SS_ISCONNECTED) {
743 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
744 			error = sodisconnect(so);
745 			if (error)
746 				goto drop;
747 		}
748 		if (so->so_options & SO_LINGER) {
749 			if ((so->so_state & (SS_ISDISCONNECTING|SS_NBIO)) ==
750 			    (SS_ISDISCONNECTING|SS_NBIO))
751 				goto drop;
752 			while (so->so_state & SS_ISCONNECTED) {
753 				error = sowait(so, true, so->so_linger * hz);
754 				if (error)
755 					break;
756 			}
757 		}
758 	}
759  drop:
760 	if (so->so_pcb) {
761 		KASSERT(solocked(so));
762 		(*so->so_proto->pr_usrreqs->pr_detach)(so);
763 	}
764  discard:
765 	KASSERT((so->so_state & SS_NOFDREF) == 0);
766 	kauth_cred_free(so->so_cred);
767 	so->so_state |= SS_NOFDREF;
768 	sofree(so);
769 	return error;
770 }
771 
772 /*
773  * Must be called with the socket locked..  Will return with it unlocked.
774  */
775 int
776 soabort(struct socket *so)
777 {
778 	u_int refs;
779 	int error;
780 
781 	KASSERT(solocked(so));
782 	KASSERT(so->so_head == NULL);
783 
784 	so->so_aborting++;		/* XXX */
785 	error = (*so->so_proto->pr_usrreqs->pr_abort)(so);
786 	refs = --so->so_aborting;	/* XXX */
787 	if (error || (refs == 0)) {
788 		sofree(so);
789 	} else {
790 		sounlock(so);
791 	}
792 	return error;
793 }
794 
795 int
796 soaccept(struct socket *so, struct sockaddr *nam)
797 {
798 	int error;
799 
800 	KASSERT(solocked(so));
801 	KASSERT((so->so_state & SS_NOFDREF) != 0);
802 
803 	so->so_state &= ~SS_NOFDREF;
804 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
805 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
806 		error = (*so->so_proto->pr_usrreqs->pr_accept)(so, nam);
807 	else
808 		error = ECONNABORTED;
809 
810 	return error;
811 }
812 
813 int
814 soconnect(struct socket *so, struct sockaddr *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 		if (nam->sa_family != so->so_proto->pr_domain->dom_family) {
834 			return EAFNOSUPPORT;
835 		}
836 		error = (*so->so_proto->pr_usrreqs->pr_connect)(so, nam, l);
837 	}
838 
839 	return error;
840 }
841 
842 int
843 soconnect2(struct socket *so1, struct socket *so2)
844 {
845 	KASSERT(solocked2(so1, so2));
846 
847 	return (*so1->so_proto->pr_usrreqs->pr_connect2)(so1, so2);
848 }
849 
850 int
851 sodisconnect(struct socket *so)
852 {
853 	int	error;
854 
855 	KASSERT(solocked(so));
856 
857 	if ((so->so_state & SS_ISCONNECTED) == 0) {
858 		error = ENOTCONN;
859 	} else if (so->so_state & SS_ISDISCONNECTING) {
860 		error = EALREADY;
861 	} else {
862 		error = (*so->so_proto->pr_usrreqs->pr_disconnect)(so);
863 	}
864 	return (error);
865 }
866 
867 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
868 /*
869  * Send on a socket.
870  * If send must go all at once and message is larger than
871  * send buffering, then hard error.
872  * Lock against other senders.
873  * If must go all at once and not enough room now, then
874  * inform user that this would block and do nothing.
875  * Otherwise, if nonblocking, send as much as possible.
876  * The data to be sent is described by "uio" if nonzero,
877  * otherwise by the mbuf chain "top" (which must be null
878  * if uio is not).  Data provided in mbuf chain must be small
879  * enough to send all at once.
880  *
881  * Returns nonzero on error, timeout or signal; callers
882  * must check for short counts if EINTR/ERESTART are returned.
883  * Data and control buffers are freed on return.
884  */
885 int
886 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
887 	struct mbuf *top, struct mbuf *control, int flags, struct lwp *l)
888 {
889 	struct mbuf	**mp, *m;
890 	long		space, len, resid, clen, mlen;
891 	int		error, s, dontroute, atomic;
892 	short		wakeup_state = 0;
893 
894 	clen = 0;
895 
896 	/*
897 	 * solock() provides atomicity of access.  splsoftnet() prevents
898 	 * protocol processing soft interrupts from interrupting us and
899 	 * blocking (expensive).
900 	 */
901 	s = splsoftnet();
902 	solock(so);
903 	atomic = sosendallatonce(so) || top;
904 	if (uio)
905 		resid = uio->uio_resid;
906 	else
907 		resid = top->m_pkthdr.len;
908 	/*
909 	 * In theory resid should be unsigned.
910 	 * However, space must be signed, as it might be less than 0
911 	 * if we over-committed, and we must use a signed comparison
912 	 * of space and resid.  On the other hand, a negative resid
913 	 * causes us to loop sending 0-length segments to the protocol.
914 	 */
915 	if (resid < 0) {
916 		error = EINVAL;
917 		goto out;
918 	}
919 	dontroute =
920 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
921 	    (so->so_proto->pr_flags & PR_ATOMIC);
922 	l->l_ru.ru_msgsnd++;
923 	if (control)
924 		clen = control->m_len;
925  restart:
926 	if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
927 		goto out;
928 	do {
929 		if (so->so_state & SS_CANTSENDMORE) {
930 			error = EPIPE;
931 			goto release;
932 		}
933 		if (so->so_error) {
934 			error = so->so_error;
935 			so->so_error = 0;
936 			goto release;
937 		}
938 		if ((so->so_state & SS_ISCONNECTED) == 0) {
939 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
940 				if (resid || clen == 0) {
941 					error = ENOTCONN;
942 					goto release;
943 				}
944 			} else if (addr == NULL) {
945 				error = EDESTADDRREQ;
946 				goto release;
947 			}
948 		}
949 		space = sbspace(&so->so_snd);
950 		if (flags & MSG_OOB)
951 			space += 1024;
952 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
953 		    clen > so->so_snd.sb_hiwat) {
954 			error = EMSGSIZE;
955 			goto release;
956 		}
957 		if (space < resid + clen &&
958 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
959 			if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
960 				error = EWOULDBLOCK;
961 				goto release;
962 			}
963 			sbunlock(&so->so_snd);
964 			if (wakeup_state & SS_RESTARTSYS) {
965 				error = ERESTART;
966 				goto out;
967 			}
968 			error = sbwait(&so->so_snd);
969 			if (error)
970 				goto out;
971 			wakeup_state = so->so_state;
972 			goto restart;
973 		}
974 		wakeup_state = 0;
975 		mp = &top;
976 		space -= clen;
977 		do {
978 			if (uio == NULL) {
979 				/*
980 				 * Data is prepackaged in "top".
981 				 */
982 				resid = 0;
983 				if (flags & MSG_EOR)
984 					top->m_flags |= M_EOR;
985 			} else do {
986 				sounlock(so);
987 				splx(s);
988 				if (top == NULL) {
989 					m = m_gethdr(M_WAIT, MT_DATA);
990 					mlen = MHLEN;
991 					m->m_pkthdr.len = 0;
992 					m->m_pkthdr.rcvif = NULL;
993 				} else {
994 					m = m_get(M_WAIT, MT_DATA);
995 					mlen = MLEN;
996 				}
997 				MCLAIM(m, so->so_snd.sb_mowner);
998 				if (sock_loan_thresh >= 0 &&
999 				    uio->uio_iov->iov_len >= sock_loan_thresh &&
1000 				    space >= sock_loan_thresh &&
1001 				    (len = sosend_loan(so, uio, m,
1002 						       space)) != 0) {
1003 					SOSEND_COUNTER_INCR(&sosend_loan_big);
1004 					space -= len;
1005 					goto have_data;
1006 				}
1007 				if (resid >= MINCLSIZE && space >= MCLBYTES) {
1008 					SOSEND_COUNTER_INCR(&sosend_copy_big);
1009 					m_clget(m, M_DONTWAIT);
1010 					if ((m->m_flags & M_EXT) == 0)
1011 						goto nopages;
1012 					mlen = MCLBYTES;
1013 					if (atomic && top == 0) {
1014 						len = lmin(MCLBYTES - max_hdr,
1015 						    resid);
1016 						m->m_data += max_hdr;
1017 					} else
1018 						len = lmin(MCLBYTES, resid);
1019 					space -= len;
1020 				} else {
1021  nopages:
1022 					SOSEND_COUNTER_INCR(&sosend_copy_small);
1023 					len = lmin(lmin(mlen, resid), space);
1024 					space -= len;
1025 					/*
1026 					 * For datagram protocols, leave room
1027 					 * for protocol headers in first mbuf.
1028 					 */
1029 					if (atomic && top == 0 && len < mlen)
1030 						MH_ALIGN(m, len);
1031 				}
1032 				error = uiomove(mtod(m, void *), (int)len, uio);
1033  have_data:
1034 				resid = uio->uio_resid;
1035 				m->m_len = len;
1036 				*mp = m;
1037 				top->m_pkthdr.len += len;
1038 				s = splsoftnet();
1039 				solock(so);
1040 				if (error != 0)
1041 					goto release;
1042 				mp = &m->m_next;
1043 				if (resid <= 0) {
1044 					if (flags & MSG_EOR)
1045 						top->m_flags |= M_EOR;
1046 					break;
1047 				}
1048 			} while (space > 0 && atomic);
1049 
1050 			if (so->so_state & SS_CANTSENDMORE) {
1051 				error = EPIPE;
1052 				goto release;
1053 			}
1054 			if (dontroute)
1055 				so->so_options |= SO_DONTROUTE;
1056 			if (resid > 0)
1057 				so->so_state |= SS_MORETOCOME;
1058 			if (flags & MSG_OOB) {
1059 				error = (*so->so_proto->pr_usrreqs->pr_sendoob)(so,
1060 				    top, control);
1061 			} else {
1062 				error = (*so->so_proto->pr_usrreqs->pr_send)(so,
1063 				    top, addr, control, l);
1064 			}
1065 			if (dontroute)
1066 				so->so_options &= ~SO_DONTROUTE;
1067 			if (resid > 0)
1068 				so->so_state &= ~SS_MORETOCOME;
1069 			clen = 0;
1070 			control = NULL;
1071 			top = NULL;
1072 			mp = &top;
1073 			if (error != 0)
1074 				goto release;
1075 		} while (resid && space > 0);
1076 	} while (resid);
1077 
1078  release:
1079 	sbunlock(&so->so_snd);
1080  out:
1081 	sounlock(so);
1082 	splx(s);
1083 	if (top)
1084 		m_freem(top);
1085 	if (control)
1086 		m_freem(control);
1087 	return (error);
1088 }
1089 
1090 /*
1091  * Following replacement or removal of the first mbuf on the first
1092  * mbuf chain of a socket buffer, push necessary state changes back
1093  * into the socket buffer so that other consumers see the values
1094  * consistently.  'nextrecord' is the callers locally stored value of
1095  * the original value of sb->sb_mb->m_nextpkt which must be restored
1096  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
1097  */
1098 static void
1099 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
1100 {
1101 
1102 	KASSERT(solocked(sb->sb_so));
1103 
1104 	/*
1105 	 * First, update for the new value of nextrecord.  If necessary,
1106 	 * make it the first record.
1107 	 */
1108 	if (sb->sb_mb != NULL)
1109 		sb->sb_mb->m_nextpkt = nextrecord;
1110 	else
1111 		sb->sb_mb = nextrecord;
1112 
1113         /*
1114          * Now update any dependent socket buffer fields to reflect
1115          * the new state.  This is an inline of SB_EMPTY_FIXUP, with
1116          * the addition of a second clause that takes care of the
1117          * case where sb_mb has been updated, but remains the last
1118          * record.
1119          */
1120         if (sb->sb_mb == NULL) {
1121                 sb->sb_mbtail = NULL;
1122                 sb->sb_lastrecord = NULL;
1123         } else if (sb->sb_mb->m_nextpkt == NULL)
1124                 sb->sb_lastrecord = sb->sb_mb;
1125 }
1126 
1127 /*
1128  * Implement receive operations on a socket.
1129  * We depend on the way that records are added to the sockbuf
1130  * by sbappend*.  In particular, each record (mbufs linked through m_next)
1131  * must begin with an address if the protocol so specifies,
1132  * followed by an optional mbuf or mbufs containing ancillary data,
1133  * and then zero or more mbufs of data.
1134  * In order to avoid blocking network interrupts for the entire time here,
1135  * we splx() while doing the actual copy to user space.
1136  * Although the sockbuf is locked, new data may still be appended,
1137  * and thus we must maintain consistency of the sockbuf during that time.
1138  *
1139  * The caller may receive the data as a single mbuf chain by supplying
1140  * an mbuf **mp0 for use in returning the chain.  The uio is then used
1141  * only for the count in uio_resid.
1142  */
1143 int
1144 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
1145 	struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1146 {
1147 	struct lwp *l = curlwp;
1148 	struct mbuf	*m, **mp, *mt;
1149 	size_t len, offset, moff, orig_resid;
1150 	int atomic, flags, error, s, type;
1151 	const struct protosw	*pr;
1152 	struct mbuf	*nextrecord;
1153 	int		mbuf_removed = 0;
1154 	const struct domain *dom;
1155 	short		wakeup_state = 0;
1156 
1157 	pr = so->so_proto;
1158 	atomic = pr->pr_flags & PR_ATOMIC;
1159 	dom = pr->pr_domain;
1160 	mp = mp0;
1161 	type = 0;
1162 	orig_resid = uio->uio_resid;
1163 
1164 	if (paddr != NULL)
1165 		*paddr = NULL;
1166 	if (controlp != NULL)
1167 		*controlp = NULL;
1168 	if (flagsp != NULL)
1169 		flags = *flagsp &~ MSG_EOR;
1170 	else
1171 		flags = 0;
1172 
1173 	if (flags & MSG_OOB) {
1174 		m = m_get(M_WAIT, MT_DATA);
1175 		solock(so);
1176 		error = (*pr->pr_usrreqs->pr_recvoob)(so, m, flags & MSG_PEEK);
1177 		sounlock(so);
1178 		if (error)
1179 			goto bad;
1180 		do {
1181 			error = uiomove(mtod(m, void *),
1182 			    MIN(uio->uio_resid, m->m_len), uio);
1183 			m = m_free(m);
1184 		} while (uio->uio_resid > 0 && error == 0 && m);
1185  bad:
1186 		if (m != NULL)
1187 			m_freem(m);
1188 		return error;
1189 	}
1190 	if (mp != NULL)
1191 		*mp = NULL;
1192 
1193 	/*
1194 	 * solock() provides atomicity of access.  splsoftnet() prevents
1195 	 * protocol processing soft interrupts from interrupting us and
1196 	 * blocking (expensive).
1197 	 */
1198 	s = splsoftnet();
1199 	solock(so);
1200  restart:
1201 	if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) {
1202 		sounlock(so);
1203 		splx(s);
1204 		return error;
1205 	}
1206 
1207 	m = so->so_rcv.sb_mb;
1208 	/*
1209 	 * If we have less data than requested, block awaiting more
1210 	 * (subject to any timeout) if:
1211 	 *   1. the current count is less than the low water mark,
1212 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1213 	 *	receive operation at once if we block (resid <= hiwat), or
1214 	 *   3. MSG_DONTWAIT is not set.
1215 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1216 	 * we have to do the receive in sections, and thus risk returning
1217 	 * a short count if a timeout or signal occurs after we start.
1218 	 */
1219 	if (m == NULL ||
1220 	    ((flags & MSG_DONTWAIT) == 0 &&
1221 	     so->so_rcv.sb_cc < uio->uio_resid &&
1222 	     (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1223 	      ((flags & MSG_WAITALL) &&
1224 	       uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1225 	     m->m_nextpkt == NULL && !atomic)) {
1226 #ifdef DIAGNOSTIC
1227 		if (m == NULL && so->so_rcv.sb_cc)
1228 			panic("receive 1");
1229 #endif
1230 		if (so->so_error) {
1231 			if (m != NULL)
1232 				goto dontblock;
1233 			error = so->so_error;
1234 			if ((flags & MSG_PEEK) == 0)
1235 				so->so_error = 0;
1236 			goto release;
1237 		}
1238 		if (so->so_state & SS_CANTRCVMORE) {
1239 			if (m != NULL)
1240 				goto dontblock;
1241 			else
1242 				goto release;
1243 		}
1244 		for (; m != NULL; m = m->m_next)
1245 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1246 				m = so->so_rcv.sb_mb;
1247 				goto dontblock;
1248 			}
1249 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1250 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1251 			error = ENOTCONN;
1252 			goto release;
1253 		}
1254 		if (uio->uio_resid == 0)
1255 			goto release;
1256 		if ((so->so_state & SS_NBIO) ||
1257 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1258 			error = EWOULDBLOCK;
1259 			goto release;
1260 		}
1261 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
1262 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
1263 		sbunlock(&so->so_rcv);
1264 		if (wakeup_state & SS_RESTARTSYS)
1265 			error = ERESTART;
1266 		else
1267 			error = sbwait(&so->so_rcv);
1268 		if (error != 0) {
1269 			sounlock(so);
1270 			splx(s);
1271 			return error;
1272 		}
1273 		wakeup_state = so->so_state;
1274 		goto restart;
1275 	}
1276  dontblock:
1277 	/*
1278 	 * On entry here, m points to the first record of the socket buffer.
1279 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1280 	 * pointer to the next record in the socket buffer.  We must keep the
1281 	 * various socket buffer pointers and local stack versions of the
1282 	 * pointers in sync, pushing out modifications before dropping the
1283 	 * socket lock, and re-reading them when picking it up.
1284 	 *
1285 	 * Otherwise, we will race with the network stack appending new data
1286 	 * or records onto the socket buffer by using inconsistent/stale
1287 	 * versions of the field, possibly resulting in socket buffer
1288 	 * corruption.
1289 	 *
1290 	 * By holding the high-level sblock(), we prevent simultaneous
1291 	 * readers from pulling off the front of the socket buffer.
1292 	 */
1293 	if (l != NULL)
1294 		l->l_ru.ru_msgrcv++;
1295 	KASSERT(m == so->so_rcv.sb_mb);
1296 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
1297 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
1298 	nextrecord = m->m_nextpkt;
1299 	if (pr->pr_flags & PR_ADDR) {
1300 #ifdef DIAGNOSTIC
1301 		if (m->m_type != MT_SONAME)
1302 			panic("receive 1a");
1303 #endif
1304 		orig_resid = 0;
1305 		if (flags & MSG_PEEK) {
1306 			if (paddr)
1307 				*paddr = m_copy(m, 0, m->m_len);
1308 			m = m->m_next;
1309 		} else {
1310 			sbfree(&so->so_rcv, m);
1311 			mbuf_removed = 1;
1312 			if (paddr != NULL) {
1313 				*paddr = m;
1314 				so->so_rcv.sb_mb = m->m_next;
1315 				m->m_next = NULL;
1316 				m = so->so_rcv.sb_mb;
1317 			} else {
1318 				MFREE(m, so->so_rcv.sb_mb);
1319 				m = so->so_rcv.sb_mb;
1320 			}
1321 			sbsync(&so->so_rcv, nextrecord);
1322 		}
1323 	}
1324 
1325 	/*
1326 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1327 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1328 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1329 	 * perform externalization (or freeing if controlp == NULL).
1330 	 */
1331 	if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) {
1332 		struct mbuf *cm = NULL, *cmn;
1333 		struct mbuf **cme = &cm;
1334 
1335 		do {
1336 			if (flags & MSG_PEEK) {
1337 				if (controlp != NULL) {
1338 					*controlp = m_copy(m, 0, m->m_len);
1339 					controlp = &(*controlp)->m_next;
1340 				}
1341 				m = m->m_next;
1342 			} else {
1343 				sbfree(&so->so_rcv, m);
1344 				so->so_rcv.sb_mb = m->m_next;
1345 				m->m_next = NULL;
1346 				*cme = m;
1347 				cme = &(*cme)->m_next;
1348 				m = so->so_rcv.sb_mb;
1349 			}
1350 		} while (m != NULL && m->m_type == MT_CONTROL);
1351 		if ((flags & MSG_PEEK) == 0)
1352 			sbsync(&so->so_rcv, nextrecord);
1353 		for (; cm != NULL; cm = cmn) {
1354 			cmn = cm->m_next;
1355 			cm->m_next = NULL;
1356 			type = mtod(cm, struct cmsghdr *)->cmsg_type;
1357 			if (controlp != NULL) {
1358 				if (dom->dom_externalize != NULL &&
1359 				    type == SCM_RIGHTS) {
1360 					sounlock(so);
1361 					splx(s);
1362 					error = (*dom->dom_externalize)(cm, l,
1363 					    (flags & MSG_CMSG_CLOEXEC) ?
1364 					    O_CLOEXEC : 0);
1365 					s = splsoftnet();
1366 					solock(so);
1367 				}
1368 				*controlp = cm;
1369 				while (*controlp != NULL)
1370 					controlp = &(*controlp)->m_next;
1371 			} else {
1372 				/*
1373 				 * Dispose of any SCM_RIGHTS message that went
1374 				 * through the read path rather than recv.
1375 				 */
1376 				if (dom->dom_dispose != NULL &&
1377 				    type == SCM_RIGHTS) {
1378 				    	sounlock(so);
1379 					(*dom->dom_dispose)(cm);
1380 					solock(so);
1381 				}
1382 				m_freem(cm);
1383 			}
1384 		}
1385 		if (m != NULL)
1386 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1387 		else
1388 			nextrecord = so->so_rcv.sb_mb;
1389 		orig_resid = 0;
1390 	}
1391 
1392 	/* If m is non-NULL, we have some data to read. */
1393 	if (__predict_true(m != NULL)) {
1394 		type = m->m_type;
1395 		if (type == MT_OOBDATA)
1396 			flags |= MSG_OOB;
1397 	}
1398 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1399 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1400 
1401 	moff = 0;
1402 	offset = 0;
1403 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1404 		if (m->m_type == MT_OOBDATA) {
1405 			if (type != MT_OOBDATA)
1406 				break;
1407 		} else if (type == MT_OOBDATA)
1408 			break;
1409 #ifdef DIAGNOSTIC
1410 		else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
1411 			panic("receive 3");
1412 #endif
1413 		so->so_state &= ~SS_RCVATMARK;
1414 		wakeup_state = 0;
1415 		len = uio->uio_resid;
1416 		if (so->so_oobmark && len > so->so_oobmark - offset)
1417 			len = so->so_oobmark - offset;
1418 		if (len > m->m_len - moff)
1419 			len = m->m_len - moff;
1420 		/*
1421 		 * If mp is set, just pass back the mbufs.
1422 		 * Otherwise copy them out via the uio, then free.
1423 		 * Sockbuf must be consistent here (points to current mbuf,
1424 		 * it points to next record) when we drop priority;
1425 		 * we must note any additions to the sockbuf when we
1426 		 * block interrupts again.
1427 		 */
1428 		if (mp == NULL) {
1429 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1430 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1431 			sounlock(so);
1432 			splx(s);
1433 			error = uiomove(mtod(m, char *) + moff, len, uio);
1434 			s = splsoftnet();
1435 			solock(so);
1436 			if (error != 0) {
1437 				/*
1438 				 * If any part of the record has been removed
1439 				 * (such as the MT_SONAME mbuf, which will
1440 				 * happen when PR_ADDR, and thus also
1441 				 * PR_ATOMIC, is set), then drop the entire
1442 				 * record to maintain the atomicity of the
1443 				 * receive operation.
1444 				 *
1445 				 * This avoids a later panic("receive 1a")
1446 				 * when compiled with DIAGNOSTIC.
1447 				 */
1448 				if (m && mbuf_removed && atomic)
1449 					(void) sbdroprecord(&so->so_rcv);
1450 
1451 				goto release;
1452 			}
1453 		} else
1454 			uio->uio_resid -= len;
1455 		if (len == m->m_len - moff) {
1456 			if (m->m_flags & M_EOR)
1457 				flags |= MSG_EOR;
1458 			if (flags & MSG_PEEK) {
1459 				m = m->m_next;
1460 				moff = 0;
1461 			} else {
1462 				nextrecord = m->m_nextpkt;
1463 				sbfree(&so->so_rcv, m);
1464 				if (mp) {
1465 					*mp = m;
1466 					mp = &m->m_next;
1467 					so->so_rcv.sb_mb = m = m->m_next;
1468 					*mp = NULL;
1469 				} else {
1470 					MFREE(m, so->so_rcv.sb_mb);
1471 					m = so->so_rcv.sb_mb;
1472 				}
1473 				/*
1474 				 * If m != NULL, we also know that
1475 				 * so->so_rcv.sb_mb != NULL.
1476 				 */
1477 				KASSERT(so->so_rcv.sb_mb == m);
1478 				if (m) {
1479 					m->m_nextpkt = nextrecord;
1480 					if (nextrecord == NULL)
1481 						so->so_rcv.sb_lastrecord = m;
1482 				} else {
1483 					so->so_rcv.sb_mb = nextrecord;
1484 					SB_EMPTY_FIXUP(&so->so_rcv);
1485 				}
1486 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1487 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1488 			}
1489 		} else if (flags & MSG_PEEK)
1490 			moff += len;
1491 		else {
1492 			if (mp != NULL) {
1493 				mt = m_copym(m, 0, len, M_NOWAIT);
1494 				if (__predict_false(mt == NULL)) {
1495 					sounlock(so);
1496 					mt = m_copym(m, 0, len, M_WAIT);
1497 					solock(so);
1498 				}
1499 				*mp = mt;
1500 			}
1501 			m->m_data += len;
1502 			m->m_len -= len;
1503 			so->so_rcv.sb_cc -= len;
1504 		}
1505 		if (so->so_oobmark) {
1506 			if ((flags & MSG_PEEK) == 0) {
1507 				so->so_oobmark -= len;
1508 				if (so->so_oobmark == 0) {
1509 					so->so_state |= SS_RCVATMARK;
1510 					break;
1511 				}
1512 			} else {
1513 				offset += len;
1514 				if (offset == so->so_oobmark)
1515 					break;
1516 			}
1517 		}
1518 		if (flags & MSG_EOR)
1519 			break;
1520 		/*
1521 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1522 		 * we must not quit until "uio->uio_resid == 0" or an error
1523 		 * termination.  If a signal/timeout occurs, return
1524 		 * with a short count but without error.
1525 		 * Keep sockbuf locked against other readers.
1526 		 */
1527 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1528 		    !sosendallatonce(so) && !nextrecord) {
1529 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1530 				break;
1531 			/*
1532 			 * If we are peeking and the socket receive buffer is
1533 			 * full, stop since we can't get more data to peek at.
1534 			 */
1535 			if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
1536 				break;
1537 			/*
1538 			 * If we've drained the socket buffer, tell the
1539 			 * protocol in case it needs to do something to
1540 			 * get it filled again.
1541 			 */
1542 			if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1543 				(*pr->pr_usrreqs->pr_rcvd)(so, flags, l);
1544 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1545 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1546 			if (wakeup_state & SS_RESTARTSYS)
1547 				error = ERESTART;
1548 			else
1549 				error = sbwait(&so->so_rcv);
1550 			if (error != 0) {
1551 				sbunlock(&so->so_rcv);
1552 				sounlock(so);
1553 				splx(s);
1554 				return 0;
1555 			}
1556 			if ((m = so->so_rcv.sb_mb) != NULL)
1557 				nextrecord = m->m_nextpkt;
1558 			wakeup_state = so->so_state;
1559 		}
1560 	}
1561 
1562 	if (m && atomic) {
1563 		flags |= MSG_TRUNC;
1564 		if ((flags & MSG_PEEK) == 0)
1565 			(void) sbdroprecord(&so->so_rcv);
1566 	}
1567 	if ((flags & MSG_PEEK) == 0) {
1568 		if (m == NULL) {
1569 			/*
1570 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1571 			 * part makes sure sb_lastrecord is up-to-date if
1572 			 * there is still data in the socket buffer.
1573 			 */
1574 			so->so_rcv.sb_mb = nextrecord;
1575 			if (so->so_rcv.sb_mb == NULL) {
1576 				so->so_rcv.sb_mbtail = NULL;
1577 				so->so_rcv.sb_lastrecord = NULL;
1578 			} else if (nextrecord->m_nextpkt == NULL)
1579 				so->so_rcv.sb_lastrecord = nextrecord;
1580 		}
1581 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1582 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1583 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1584 			(*pr->pr_usrreqs->pr_rcvd)(so, flags, l);
1585 	}
1586 	if (orig_resid == uio->uio_resid && orig_resid &&
1587 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1588 		sbunlock(&so->so_rcv);
1589 		goto restart;
1590 	}
1591 
1592 	if (flagsp != NULL)
1593 		*flagsp |= flags;
1594  release:
1595 	sbunlock(&so->so_rcv);
1596 	sounlock(so);
1597 	splx(s);
1598 	return error;
1599 }
1600 
1601 int
1602 soshutdown(struct socket *so, int how)
1603 {
1604 	const struct protosw	*pr;
1605 	int	error;
1606 
1607 	KASSERT(solocked(so));
1608 
1609 	pr = so->so_proto;
1610 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1611 		return (EINVAL);
1612 
1613 	if (how == SHUT_RD || how == SHUT_RDWR) {
1614 		sorflush(so);
1615 		error = 0;
1616 	}
1617 	if (how == SHUT_WR || how == SHUT_RDWR)
1618 		error = (*pr->pr_usrreqs->pr_shutdown)(so);
1619 
1620 	return error;
1621 }
1622 
1623 void
1624 sorestart(struct socket *so)
1625 {
1626 	/*
1627 	 * An application has called close() on an fd on which another
1628 	 * of its threads has called a socket system call.
1629 	 * Mark this and wake everyone up, and code that would block again
1630 	 * instead returns ERESTART.
1631 	 * On system call re-entry the fd is validated and EBADF returned.
1632 	 * Any other fd will block again on the 2nd syscall.
1633 	 */
1634 	solock(so);
1635 	so->so_state |= SS_RESTARTSYS;
1636 	cv_broadcast(&so->so_cv);
1637 	cv_broadcast(&so->so_snd.sb_cv);
1638 	cv_broadcast(&so->so_rcv.sb_cv);
1639 	sounlock(so);
1640 }
1641 
1642 void
1643 sorflush(struct socket *so)
1644 {
1645 	struct sockbuf	*sb, asb;
1646 	const struct protosw	*pr;
1647 
1648 	KASSERT(solocked(so));
1649 
1650 	sb = &so->so_rcv;
1651 	pr = so->so_proto;
1652 	socantrcvmore(so);
1653 	sb->sb_flags |= SB_NOINTR;
1654 	(void )sblock(sb, M_WAITOK);
1655 	sbunlock(sb);
1656 	asb = *sb;
1657 	/*
1658 	 * Clear most of the sockbuf structure, but leave some of the
1659 	 * fields valid.
1660 	 */
1661 	memset(&sb->sb_startzero, 0,
1662 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1663 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) {
1664 		sounlock(so);
1665 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1666 		solock(so);
1667 	}
1668 	sbrelease(&asb, so);
1669 }
1670 
1671 /*
1672  * internal set SOL_SOCKET options
1673  */
1674 static int
1675 sosetopt1(struct socket *so, const struct sockopt *sopt)
1676 {
1677 	int error = EINVAL, opt;
1678 	int optval = 0; /* XXX: gcc */
1679 	struct linger l;
1680 	struct timeval tv;
1681 
1682 	switch ((opt = sopt->sopt_name)) {
1683 
1684 	case SO_ACCEPTFILTER:
1685 		error = accept_filt_setopt(so, sopt);
1686 		KASSERT(solocked(so));
1687 		break;
1688 
1689   	case SO_LINGER:
1690  		error = sockopt_get(sopt, &l, sizeof(l));
1691 		solock(so);
1692  		if (error)
1693  			break;
1694  		if (l.l_linger < 0 || l.l_linger > USHRT_MAX ||
1695  		    l.l_linger > (INT_MAX / hz)) {
1696 			error = EDOM;
1697 			break;
1698 		}
1699  		so->so_linger = l.l_linger;
1700  		if (l.l_onoff)
1701  			so->so_options |= SO_LINGER;
1702  		else
1703  			so->so_options &= ~SO_LINGER;
1704    		break;
1705 
1706 	case SO_DEBUG:
1707 	case SO_KEEPALIVE:
1708 	case SO_DONTROUTE:
1709 	case SO_USELOOPBACK:
1710 	case SO_BROADCAST:
1711 	case SO_REUSEADDR:
1712 	case SO_REUSEPORT:
1713 	case SO_OOBINLINE:
1714 	case SO_TIMESTAMP:
1715 	case SO_NOSIGPIPE:
1716 #ifdef SO_OTIMESTAMP
1717 	case SO_OTIMESTAMP:
1718 #endif
1719 		error = sockopt_getint(sopt, &optval);
1720 		solock(so);
1721 		if (error)
1722 			break;
1723 		if (optval)
1724 			so->so_options |= opt;
1725 		else
1726 			so->so_options &= ~opt;
1727 		break;
1728 
1729 	case SO_SNDBUF:
1730 	case SO_RCVBUF:
1731 	case SO_SNDLOWAT:
1732 	case SO_RCVLOWAT:
1733 		error = sockopt_getint(sopt, &optval);
1734 		solock(so);
1735 		if (error)
1736 			break;
1737 
1738 		/*
1739 		 * Values < 1 make no sense for any of these
1740 		 * options, so disallow them.
1741 		 */
1742 		if (optval < 1) {
1743 			error = EINVAL;
1744 			break;
1745 		}
1746 
1747 		switch (opt) {
1748 		case SO_SNDBUF:
1749 			if (sbreserve(&so->so_snd, (u_long)optval, so) == 0) {
1750 				error = ENOBUFS;
1751 				break;
1752 			}
1753 			so->so_snd.sb_flags &= ~SB_AUTOSIZE;
1754 			break;
1755 
1756 		case SO_RCVBUF:
1757 			if (sbreserve(&so->so_rcv, (u_long)optval, so) == 0) {
1758 				error = ENOBUFS;
1759 				break;
1760 			}
1761 			so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1762 			break;
1763 
1764 		/*
1765 		 * Make sure the low-water is never greater than
1766 		 * the high-water.
1767 		 */
1768 		case SO_SNDLOWAT:
1769 			if (optval > so->so_snd.sb_hiwat)
1770 				optval = so->so_snd.sb_hiwat;
1771 
1772 			so->so_snd.sb_lowat = optval;
1773 			break;
1774 
1775 		case SO_RCVLOWAT:
1776 			if (optval > so->so_rcv.sb_hiwat)
1777 				optval = so->so_rcv.sb_hiwat;
1778 
1779 			so->so_rcv.sb_lowat = optval;
1780 			break;
1781 		}
1782 		break;
1783 
1784 #ifdef COMPAT_50
1785 	case SO_OSNDTIMEO:
1786 	case SO_ORCVTIMEO: {
1787 		struct timeval50 otv;
1788 		error = sockopt_get(sopt, &otv, sizeof(otv));
1789 		if (error) {
1790 			solock(so);
1791 			break;
1792 		}
1793 		timeval50_to_timeval(&otv, &tv);
1794 		opt = opt == SO_OSNDTIMEO ? SO_SNDTIMEO : SO_RCVTIMEO;
1795 		error = 0;
1796 		/*FALLTHROUGH*/
1797 	}
1798 #endif /* COMPAT_50 */
1799 
1800 	case SO_SNDTIMEO:
1801 	case SO_RCVTIMEO:
1802 		if (error)
1803 			error = sockopt_get(sopt, &tv, sizeof(tv));
1804 		solock(so);
1805 		if (error)
1806 			break;
1807 
1808 		if (tv.tv_sec > (INT_MAX - tv.tv_usec / tick) / hz) {
1809 			error = EDOM;
1810 			break;
1811 		}
1812 
1813 		optval = tv.tv_sec * hz + tv.tv_usec / tick;
1814 		if (optval == 0 && tv.tv_usec != 0)
1815 			optval = 1;
1816 
1817 		switch (opt) {
1818 		case SO_SNDTIMEO:
1819 			so->so_snd.sb_timeo = optval;
1820 			break;
1821 		case SO_RCVTIMEO:
1822 			so->so_rcv.sb_timeo = optval;
1823 			break;
1824 		}
1825 		break;
1826 
1827 	default:
1828 		solock(so);
1829 		error = ENOPROTOOPT;
1830 		break;
1831 	}
1832 	KASSERT(solocked(so));
1833 	return error;
1834 }
1835 
1836 int
1837 sosetopt(struct socket *so, struct sockopt *sopt)
1838 {
1839 	int error, prerr;
1840 
1841 	if (sopt->sopt_level == SOL_SOCKET) {
1842 		error = sosetopt1(so, sopt);
1843 		KASSERT(solocked(so));
1844 	} else {
1845 		error = ENOPROTOOPT;
1846 		solock(so);
1847 	}
1848 
1849 	if ((error == 0 || error == ENOPROTOOPT) &&
1850 	    so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) {
1851 		/* give the protocol stack a shot */
1852 		prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, sopt);
1853 		if (prerr == 0)
1854 			error = 0;
1855 		else if (prerr != ENOPROTOOPT)
1856 			error = prerr;
1857 	}
1858 	sounlock(so);
1859 	return error;
1860 }
1861 
1862 /*
1863  * so_setsockopt() is a wrapper providing a sockopt structure for sosetopt()
1864  */
1865 int
1866 so_setsockopt(struct lwp *l, struct socket *so, int level, int name,
1867     const void *val, size_t valsize)
1868 {
1869 	struct sockopt sopt;
1870 	int error;
1871 
1872 	KASSERT(valsize == 0 || val != NULL);
1873 
1874 	sockopt_init(&sopt, level, name, valsize);
1875 	sockopt_set(&sopt, val, valsize);
1876 
1877 	error = sosetopt(so, &sopt);
1878 
1879 	sockopt_destroy(&sopt);
1880 
1881 	return error;
1882 }
1883 
1884 /*
1885  * internal get SOL_SOCKET options
1886  */
1887 static int
1888 sogetopt1(struct socket *so, struct sockopt *sopt)
1889 {
1890 	int error, optval, opt;
1891 	struct linger l;
1892 	struct timeval tv;
1893 
1894 	switch ((opt = sopt->sopt_name)) {
1895 
1896 	case SO_ACCEPTFILTER:
1897 		error = accept_filt_getopt(so, sopt);
1898 		break;
1899 
1900 	case SO_LINGER:
1901 		l.l_onoff = (so->so_options & SO_LINGER) ? 1 : 0;
1902 		l.l_linger = so->so_linger;
1903 
1904 		error = sockopt_set(sopt, &l, sizeof(l));
1905 		break;
1906 
1907 	case SO_USELOOPBACK:
1908 	case SO_DONTROUTE:
1909 	case SO_DEBUG:
1910 	case SO_KEEPALIVE:
1911 	case SO_REUSEADDR:
1912 	case SO_REUSEPORT:
1913 	case SO_BROADCAST:
1914 	case SO_OOBINLINE:
1915 	case SO_TIMESTAMP:
1916 	case SO_NOSIGPIPE:
1917 #ifdef SO_OTIMESTAMP
1918 	case SO_OTIMESTAMP:
1919 #endif
1920 	case SO_ACCEPTCONN:
1921 		error = sockopt_setint(sopt, (so->so_options & opt) ? 1 : 0);
1922 		break;
1923 
1924 	case SO_TYPE:
1925 		error = sockopt_setint(sopt, so->so_type);
1926 		break;
1927 
1928 	case SO_ERROR:
1929 		error = sockopt_setint(sopt, so->so_error);
1930 		so->so_error = 0;
1931 		break;
1932 
1933 	case SO_SNDBUF:
1934 		error = sockopt_setint(sopt, so->so_snd.sb_hiwat);
1935 		break;
1936 
1937 	case SO_RCVBUF:
1938 		error = sockopt_setint(sopt, so->so_rcv.sb_hiwat);
1939 		break;
1940 
1941 	case SO_SNDLOWAT:
1942 		error = sockopt_setint(sopt, so->so_snd.sb_lowat);
1943 		break;
1944 
1945 	case SO_RCVLOWAT:
1946 		error = sockopt_setint(sopt, so->so_rcv.sb_lowat);
1947 		break;
1948 
1949 #ifdef COMPAT_50
1950 	case SO_OSNDTIMEO:
1951 	case SO_ORCVTIMEO: {
1952 		struct timeval50 otv;
1953 
1954 		optval = (opt == SO_OSNDTIMEO ?
1955 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1956 
1957 		otv.tv_sec = optval / hz;
1958 		otv.tv_usec = (optval % hz) * tick;
1959 
1960 		error = sockopt_set(sopt, &otv, sizeof(otv));
1961 		break;
1962 	}
1963 #endif /* COMPAT_50 */
1964 
1965 	case SO_SNDTIMEO:
1966 	case SO_RCVTIMEO:
1967 		optval = (opt == SO_SNDTIMEO ?
1968 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1969 
1970 		tv.tv_sec = optval / hz;
1971 		tv.tv_usec = (optval % hz) * tick;
1972 
1973 		error = sockopt_set(sopt, &tv, sizeof(tv));
1974 		break;
1975 
1976 	case SO_OVERFLOWED:
1977 		error = sockopt_setint(sopt, so->so_rcv.sb_overflowed);
1978 		break;
1979 
1980 	default:
1981 		error = ENOPROTOOPT;
1982 		break;
1983 	}
1984 
1985 	return (error);
1986 }
1987 
1988 int
1989 sogetopt(struct socket *so, struct sockopt *sopt)
1990 {
1991 	int		error;
1992 
1993 	solock(so);
1994 	if (sopt->sopt_level != SOL_SOCKET) {
1995 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1996 			error = ((*so->so_proto->pr_ctloutput)
1997 			    (PRCO_GETOPT, so, sopt));
1998 		} else
1999 			error = (ENOPROTOOPT);
2000 	} else {
2001 		error = sogetopt1(so, sopt);
2002 	}
2003 	sounlock(so);
2004 	return (error);
2005 }
2006 
2007 /*
2008  * alloc sockopt data buffer buffer
2009  *	- will be released at destroy
2010  */
2011 static int
2012 sockopt_alloc(struct sockopt *sopt, size_t len, km_flag_t kmflag)
2013 {
2014 
2015 	KASSERT(sopt->sopt_size == 0);
2016 
2017 	if (len > sizeof(sopt->sopt_buf)) {
2018 		sopt->sopt_data = kmem_zalloc(len, kmflag);
2019 		if (sopt->sopt_data == NULL)
2020 			return ENOMEM;
2021 	} else
2022 		sopt->sopt_data = sopt->sopt_buf;
2023 
2024 	sopt->sopt_size = len;
2025 	return 0;
2026 }
2027 
2028 /*
2029  * initialise sockopt storage
2030  *	- MAY sleep during allocation
2031  */
2032 void
2033 sockopt_init(struct sockopt *sopt, int level, int name, size_t size)
2034 {
2035 
2036 	memset(sopt, 0, sizeof(*sopt));
2037 
2038 	sopt->sopt_level = level;
2039 	sopt->sopt_name = name;
2040 	(void)sockopt_alloc(sopt, size, KM_SLEEP);
2041 }
2042 
2043 /*
2044  * destroy sockopt storage
2045  *	- will release any held memory references
2046  */
2047 void
2048 sockopt_destroy(struct sockopt *sopt)
2049 {
2050 
2051 	if (sopt->sopt_data != sopt->sopt_buf)
2052 		kmem_free(sopt->sopt_data, sopt->sopt_size);
2053 
2054 	memset(sopt, 0, sizeof(*sopt));
2055 }
2056 
2057 /*
2058  * set sockopt value
2059  *	- value is copied into sockopt
2060  * 	- memory is allocated when necessary, will not sleep
2061  */
2062 int
2063 sockopt_set(struct sockopt *sopt, const void *buf, size_t len)
2064 {
2065 	int error;
2066 
2067 	if (sopt->sopt_size == 0) {
2068 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2069 		if (error)
2070 			return error;
2071 	}
2072 
2073 	KASSERT(sopt->sopt_size == len);
2074 	memcpy(sopt->sopt_data, buf, len);
2075 	return 0;
2076 }
2077 
2078 /*
2079  * common case of set sockopt integer value
2080  */
2081 int
2082 sockopt_setint(struct sockopt *sopt, int val)
2083 {
2084 
2085 	return sockopt_set(sopt, &val, sizeof(int));
2086 }
2087 
2088 /*
2089  * get sockopt value
2090  *	- correct size must be given
2091  */
2092 int
2093 sockopt_get(const struct sockopt *sopt, void *buf, size_t len)
2094 {
2095 
2096 	if (sopt->sopt_size != len)
2097 		return EINVAL;
2098 
2099 	memcpy(buf, sopt->sopt_data, len);
2100 	return 0;
2101 }
2102 
2103 /*
2104  * common case of get sockopt integer value
2105  */
2106 int
2107 sockopt_getint(const struct sockopt *sopt, int *valp)
2108 {
2109 
2110 	return sockopt_get(sopt, valp, sizeof(int));
2111 }
2112 
2113 /*
2114  * set sockopt value from mbuf
2115  *	- ONLY for legacy code
2116  *	- mbuf is released by sockopt
2117  *	- will not sleep
2118  */
2119 int
2120 sockopt_setmbuf(struct sockopt *sopt, struct mbuf *m)
2121 {
2122 	size_t len;
2123 	int error;
2124 
2125 	len = m_length(m);
2126 
2127 	if (sopt->sopt_size == 0) {
2128 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2129 		if (error)
2130 			return error;
2131 	}
2132 
2133 	KASSERT(sopt->sopt_size == len);
2134 	m_copydata(m, 0, len, sopt->sopt_data);
2135 	m_freem(m);
2136 
2137 	return 0;
2138 }
2139 
2140 /*
2141  * get sockopt value into mbuf
2142  *	- ONLY for legacy code
2143  *	- mbuf to be released by the caller
2144  *	- will not sleep
2145  */
2146 struct mbuf *
2147 sockopt_getmbuf(const struct sockopt *sopt)
2148 {
2149 	struct mbuf *m;
2150 
2151 	if (sopt->sopt_size > MCLBYTES)
2152 		return NULL;
2153 
2154 	m = m_get(M_DONTWAIT, MT_SOOPTS);
2155 	if (m == NULL)
2156 		return NULL;
2157 
2158 	if (sopt->sopt_size > MLEN) {
2159 		MCLGET(m, M_DONTWAIT);
2160 		if ((m->m_flags & M_EXT) == 0) {
2161 			m_free(m);
2162 			return NULL;
2163 		}
2164 	}
2165 
2166 	memcpy(mtod(m, void *), sopt->sopt_data, sopt->sopt_size);
2167 	m->m_len = sopt->sopt_size;
2168 
2169 	return m;
2170 }
2171 
2172 void
2173 sohasoutofband(struct socket *so)
2174 {
2175 
2176 	fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
2177 	selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, NOTE_SUBMIT);
2178 }
2179 
2180 static void
2181 filt_sordetach(struct knote *kn)
2182 {
2183 	struct socket	*so;
2184 
2185 	so = ((file_t *)kn->kn_obj)->f_socket;
2186 	solock(so);
2187 	SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
2188 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
2189 		so->so_rcv.sb_flags &= ~SB_KNOTE;
2190 	sounlock(so);
2191 }
2192 
2193 /*ARGSUSED*/
2194 static int
2195 filt_soread(struct knote *kn, long hint)
2196 {
2197 	struct socket	*so;
2198 	int rv;
2199 
2200 	so = ((file_t *)kn->kn_obj)->f_socket;
2201 	if (hint != NOTE_SUBMIT)
2202 		solock(so);
2203 	kn->kn_data = so->so_rcv.sb_cc;
2204 	if (so->so_state & SS_CANTRCVMORE) {
2205 		kn->kn_flags |= EV_EOF;
2206 		kn->kn_fflags = so->so_error;
2207 		rv = 1;
2208 	} else if (so->so_error)	/* temporary udp error */
2209 		rv = 1;
2210 	else if (kn->kn_sfflags & NOTE_LOWAT)
2211 		rv = (kn->kn_data >= kn->kn_sdata);
2212 	else
2213 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
2214 	if (hint != NOTE_SUBMIT)
2215 		sounlock(so);
2216 	return rv;
2217 }
2218 
2219 static void
2220 filt_sowdetach(struct knote *kn)
2221 {
2222 	struct socket	*so;
2223 
2224 	so = ((file_t *)kn->kn_obj)->f_socket;
2225 	solock(so);
2226 	SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
2227 	if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
2228 		so->so_snd.sb_flags &= ~SB_KNOTE;
2229 	sounlock(so);
2230 }
2231 
2232 /*ARGSUSED*/
2233 static int
2234 filt_sowrite(struct knote *kn, long hint)
2235 {
2236 	struct socket	*so;
2237 	int rv;
2238 
2239 	so = ((file_t *)kn->kn_obj)->f_socket;
2240 	if (hint != NOTE_SUBMIT)
2241 		solock(so);
2242 	kn->kn_data = sbspace(&so->so_snd);
2243 	if (so->so_state & SS_CANTSENDMORE) {
2244 		kn->kn_flags |= EV_EOF;
2245 		kn->kn_fflags = so->so_error;
2246 		rv = 1;
2247 	} else if (so->so_error)	/* temporary udp error */
2248 		rv = 1;
2249 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2250 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
2251 		rv = 0;
2252 	else if (kn->kn_sfflags & NOTE_LOWAT)
2253 		rv = (kn->kn_data >= kn->kn_sdata);
2254 	else
2255 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
2256 	if (hint != NOTE_SUBMIT)
2257 		sounlock(so);
2258 	return rv;
2259 }
2260 
2261 /*ARGSUSED*/
2262 static int
2263 filt_solisten(struct knote *kn, long hint)
2264 {
2265 	struct socket	*so;
2266 	int rv;
2267 
2268 	so = ((file_t *)kn->kn_obj)->f_socket;
2269 
2270 	/*
2271 	 * Set kn_data to number of incoming connections, not
2272 	 * counting partial (incomplete) connections.
2273 	 */
2274 	if (hint != NOTE_SUBMIT)
2275 		solock(so);
2276 	kn->kn_data = so->so_qlen;
2277 	rv = (kn->kn_data > 0);
2278 	if (hint != NOTE_SUBMIT)
2279 		sounlock(so);
2280 	return rv;
2281 }
2282 
2283 static const struct filterops solisten_filtops =
2284 	{ 1, NULL, filt_sordetach, filt_solisten };
2285 static const struct filterops soread_filtops =
2286 	{ 1, NULL, filt_sordetach, filt_soread };
2287 static const struct filterops sowrite_filtops =
2288 	{ 1, NULL, filt_sowdetach, filt_sowrite };
2289 
2290 int
2291 soo_kqfilter(struct file *fp, struct knote *kn)
2292 {
2293 	struct socket	*so;
2294 	struct sockbuf	*sb;
2295 
2296 	so = ((file_t *)kn->kn_obj)->f_socket;
2297 	solock(so);
2298 	switch (kn->kn_filter) {
2299 	case EVFILT_READ:
2300 		if (so->so_options & SO_ACCEPTCONN)
2301 			kn->kn_fop = &solisten_filtops;
2302 		else
2303 			kn->kn_fop = &soread_filtops;
2304 		sb = &so->so_rcv;
2305 		break;
2306 	case EVFILT_WRITE:
2307 		kn->kn_fop = &sowrite_filtops;
2308 		sb = &so->so_snd;
2309 		break;
2310 	default:
2311 		sounlock(so);
2312 		return (EINVAL);
2313 	}
2314 	SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
2315 	sb->sb_flags |= SB_KNOTE;
2316 	sounlock(so);
2317 	return (0);
2318 }
2319 
2320 static int
2321 sodopoll(struct socket *so, int events)
2322 {
2323 	int revents;
2324 
2325 	revents = 0;
2326 
2327 	if (events & (POLLIN | POLLRDNORM))
2328 		if (soreadable(so))
2329 			revents |= events & (POLLIN | POLLRDNORM);
2330 
2331 	if (events & (POLLOUT | POLLWRNORM))
2332 		if (sowritable(so))
2333 			revents |= events & (POLLOUT | POLLWRNORM);
2334 
2335 	if (events & (POLLPRI | POLLRDBAND))
2336 		if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
2337 			revents |= events & (POLLPRI | POLLRDBAND);
2338 
2339 	return revents;
2340 }
2341 
2342 int
2343 sopoll(struct socket *so, int events)
2344 {
2345 	int revents = 0;
2346 
2347 #ifndef DIAGNOSTIC
2348 	/*
2349 	 * Do a quick, unlocked check in expectation that the socket
2350 	 * will be ready for I/O.  Don't do this check if DIAGNOSTIC,
2351 	 * as the solocked() assertions will fail.
2352 	 */
2353 	if ((revents = sodopoll(so, events)) != 0)
2354 		return revents;
2355 #endif
2356 
2357 	solock(so);
2358 	if ((revents = sodopoll(so, events)) == 0) {
2359 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2360 			selrecord(curlwp, &so->so_rcv.sb_sel);
2361 			so->so_rcv.sb_flags |= SB_NOTIFY;
2362 		}
2363 
2364 		if (events & (POLLOUT | POLLWRNORM)) {
2365 			selrecord(curlwp, &so->so_snd.sb_sel);
2366 			so->so_snd.sb_flags |= SB_NOTIFY;
2367 		}
2368 	}
2369 	sounlock(so);
2370 
2371 	return revents;
2372 }
2373 
2374 
2375 #include <sys/sysctl.h>
2376 
2377 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO);
2378 static int sysctl_kern_sbmax(SYSCTLFN_PROTO);
2379 
2380 /*
2381  * sysctl helper routine for kern.somaxkva.  ensures that the given
2382  * value is not too small.
2383  * (XXX should we maybe make sure it's not too large as well?)
2384  */
2385 static int
2386 sysctl_kern_somaxkva(SYSCTLFN_ARGS)
2387 {
2388 	int error, new_somaxkva;
2389 	struct sysctlnode node;
2390 
2391 	new_somaxkva = somaxkva;
2392 	node = *rnode;
2393 	node.sysctl_data = &new_somaxkva;
2394 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2395 	if (error || newp == NULL)
2396 		return (error);
2397 
2398 	if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */
2399 		return (EINVAL);
2400 
2401 	mutex_enter(&so_pendfree_lock);
2402 	somaxkva = new_somaxkva;
2403 	cv_broadcast(&socurkva_cv);
2404 	mutex_exit(&so_pendfree_lock);
2405 
2406 	return (error);
2407 }
2408 
2409 /*
2410  * sysctl helper routine for kern.sbmax. Basically just ensures that
2411  * any new value is not too small.
2412  */
2413 static int
2414 sysctl_kern_sbmax(SYSCTLFN_ARGS)
2415 {
2416 	int error, new_sbmax;
2417 	struct sysctlnode node;
2418 
2419 	new_sbmax = sb_max;
2420 	node = *rnode;
2421 	node.sysctl_data = &new_sbmax;
2422 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2423 	if (error || newp == NULL)
2424 		return (error);
2425 
2426 	KERNEL_LOCK(1, NULL);
2427 	error = sb_max_set(new_sbmax);
2428 	KERNEL_UNLOCK_ONE(NULL);
2429 
2430 	return (error);
2431 }
2432 
2433 static void
2434 sysctl_kern_socket_setup(void)
2435 {
2436 
2437 	KASSERT(socket_sysctllog == NULL);
2438 
2439 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2440 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2441 		       CTLTYPE_INT, "somaxkva",
2442 		       SYSCTL_DESCR("Maximum amount of kernel memory to be "
2443 				    "used for socket buffers"),
2444 		       sysctl_kern_somaxkva, 0, NULL, 0,
2445 		       CTL_KERN, KERN_SOMAXKVA, CTL_EOL);
2446 
2447 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2448 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2449 		       CTLTYPE_INT, "sbmax",
2450 		       SYSCTL_DESCR("Maximum socket buffer size"),
2451 		       sysctl_kern_sbmax, 0, NULL, 0,
2452 		       CTL_KERN, KERN_SBMAX, CTL_EOL);
2453 }
2454