xref: /netbsd-src/sys/kern/uipc_socket.c (revision d48f14661dda8638fee055ba15d35bdfb29b9fa8)
1 /*	$NetBSD: uipc_socket.c,v 1.121 2006/06/21 12:55:12 yamt Exp $	*/
2 
3 /*-
4  * Copyright (c) 2002 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe of Wasabi Systems, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the NetBSD
21  *	Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 
39 /*
40  * Copyright (c) 1982, 1986, 1988, 1990, 1993
41  *	The Regents of the University of California.  All rights reserved.
42  *
43  * Redistribution and use in source and binary forms, with or without
44  * modification, are permitted provided that the following conditions
45  * are met:
46  * 1. Redistributions of source code must retain the above copyright
47  *    notice, this list of conditions and the following disclaimer.
48  * 2. Redistributions in binary form must reproduce the above copyright
49  *    notice, this list of conditions and the following disclaimer in the
50  *    documentation and/or other materials provided with the distribution.
51  * 3. Neither the name of the University nor the names of its contributors
52  *    may be used to endorse or promote products derived from this software
53  *    without specific prior written permission.
54  *
55  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
56  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
57  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
58  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
59  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
60  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
61  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
62  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
63  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
64  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
65  * SUCH DAMAGE.
66  *
67  *	@(#)uipc_socket.c	8.6 (Berkeley) 5/2/95
68  */
69 
70 #include <sys/cdefs.h>
71 __KERNEL_RCSID(0, "$NetBSD: uipc_socket.c,v 1.121 2006/06/21 12:55:12 yamt Exp $");
72 
73 #include "opt_sock_counters.h"
74 #include "opt_sosend_loan.h"
75 #include "opt_mbuftrace.h"
76 #include "opt_somaxkva.h"
77 
78 #include <sys/param.h>
79 #include <sys/systm.h>
80 #include <sys/proc.h>
81 #include <sys/file.h>
82 #include <sys/malloc.h>
83 #include <sys/mbuf.h>
84 #include <sys/domain.h>
85 #include <sys/kernel.h>
86 #include <sys/protosw.h>
87 #include <sys/socket.h>
88 #include <sys/socketvar.h>
89 #include <sys/signalvar.h>
90 #include <sys/resourcevar.h>
91 #include <sys/pool.h>
92 #include <sys/event.h>
93 #include <sys/poll.h>
94 #include <sys/kauth.h>
95 
96 #include <uvm/uvm.h>
97 
98 POOL_INIT(socket_pool, sizeof(struct socket), 0, 0, 0, "sockpl", NULL);
99 
100 MALLOC_DEFINE(M_SOOPTS, "soopts", "socket options");
101 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
102 
103 extern int	somaxconn;			/* patchable (XXX sysctl) */
104 int		somaxconn = SOMAXCONN;
105 
106 #ifdef SOSEND_COUNTERS
107 #include <sys/device.h>
108 
109 static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
110     NULL, "sosend", "loan big");
111 static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
112     NULL, "sosend", "copy big");
113 static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
114     NULL, "sosend", "copy small");
115 static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
116     NULL, "sosend", "kva limit");
117 
118 #define	SOSEND_COUNTER_INCR(ev)		(ev)->ev_count++
119 
120 EVCNT_ATTACH_STATIC(sosend_loan_big);
121 EVCNT_ATTACH_STATIC(sosend_copy_big);
122 EVCNT_ATTACH_STATIC(sosend_copy_small);
123 EVCNT_ATTACH_STATIC(sosend_kvalimit);
124 #else
125 
126 #define	SOSEND_COUNTER_INCR(ev)		/* nothing */
127 
128 #endif /* SOSEND_COUNTERS */
129 
130 static struct callback_entry sokva_reclaimerentry;
131 
132 #ifdef SOSEND_NO_LOAN
133 int sock_loan_thresh = -1;
134 #else
135 int sock_loan_thresh = 4096;
136 #endif
137 
138 static struct simplelock so_pendfree_slock = SIMPLELOCK_INITIALIZER;
139 static struct mbuf *so_pendfree;
140 
141 #ifndef SOMAXKVA
142 #define	SOMAXKVA (16 * 1024 * 1024)
143 #endif
144 int somaxkva = SOMAXKVA;
145 static int socurkva;
146 static int sokvawaiters;
147 
148 #define	SOCK_LOAN_CHUNK		65536
149 
150 static size_t sodopendfree(void);
151 static size_t sodopendfreel(void);
152 
153 static vsize_t
154 sokvareserve(struct socket *so, vsize_t len)
155 {
156 	int s;
157 	int error;
158 
159 	s = splvm();
160 	simple_lock(&so_pendfree_slock);
161 	while (socurkva + len > somaxkva) {
162 		size_t freed;
163 
164 		/*
165 		 * try to do pendfree.
166 		 */
167 
168 		freed = sodopendfreel();
169 
170 		/*
171 		 * if some kva was freed, try again.
172 		 */
173 
174 		if (freed)
175 			continue;
176 
177 		SOSEND_COUNTER_INCR(&sosend_kvalimit);
178 		sokvawaiters++;
179 		error = ltsleep(&socurkva, PVM | PCATCH, "sokva", 0,
180 		    &so_pendfree_slock);
181 		sokvawaiters--;
182 		if (error) {
183 			len = 0;
184 			break;
185 		}
186 	}
187 	socurkva += len;
188 	simple_unlock(&so_pendfree_slock);
189 	splx(s);
190 	return len;
191 }
192 
193 static void
194 sokvaunreserve(vsize_t len)
195 {
196 	int s;
197 
198 	s = splvm();
199 	simple_lock(&so_pendfree_slock);
200 	socurkva -= len;
201 	if (sokvawaiters)
202 		wakeup(&socurkva);
203 	simple_unlock(&so_pendfree_slock);
204 	splx(s);
205 }
206 
207 /*
208  * sokvaalloc: allocate kva for loan.
209  */
210 
211 vaddr_t
212 sokvaalloc(vsize_t len, struct socket *so)
213 {
214 	vaddr_t lva;
215 
216 	/*
217 	 * reserve kva.
218 	 */
219 
220 	if (sokvareserve(so, len) == 0)
221 		return 0;
222 
223 	/*
224 	 * allocate kva.
225 	 */
226 
227 	lva = uvm_km_alloc(kernel_map, len, 0, UVM_KMF_VAONLY | UVM_KMF_WAITVA);
228 	if (lva == 0) {
229 		sokvaunreserve(len);
230 		return (0);
231 	}
232 
233 	return lva;
234 }
235 
236 /*
237  * sokvafree: free kva for loan.
238  */
239 
240 void
241 sokvafree(vaddr_t sva, vsize_t len)
242 {
243 
244 	/*
245 	 * free kva.
246 	 */
247 
248 	uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY);
249 
250 	/*
251 	 * unreserve kva.
252 	 */
253 
254 	sokvaunreserve(len);
255 }
256 
257 static void
258 sodoloanfree(struct vm_page **pgs, caddr_t buf, size_t size)
259 {
260 	vaddr_t va, sva, eva;
261 	vsize_t len;
262 	paddr_t pa;
263 	int i, npgs;
264 
265 	eva = round_page((vaddr_t) buf + size);
266 	sva = trunc_page((vaddr_t) buf);
267 	len = eva - sva;
268 	npgs = len >> PAGE_SHIFT;
269 
270 	if (__predict_false(pgs == NULL)) {
271 		pgs = alloca(npgs * sizeof(*pgs));
272 
273 		for (i = 0, va = sva; va < eva; i++, va += PAGE_SIZE) {
274 			if (pmap_extract(pmap_kernel(), va, &pa) == FALSE)
275 				panic("sodoloanfree: va 0x%lx not mapped", va);
276 			pgs[i] = PHYS_TO_VM_PAGE(pa);
277 		}
278 	}
279 
280 	pmap_kremove(sva, len);
281 	pmap_update(pmap_kernel());
282 	uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE);
283 	sokvafree(sva, len);
284 }
285 
286 static size_t
287 sodopendfree()
288 {
289 	int s;
290 	size_t rv;
291 
292 	s = splvm();
293 	simple_lock(&so_pendfree_slock);
294 	rv = sodopendfreel();
295 	simple_unlock(&so_pendfree_slock);
296 	splx(s);
297 
298 	return rv;
299 }
300 
301 /*
302  * sodopendfreel: free mbufs on "pendfree" list.
303  * unlock and relock so_pendfree_slock when freeing mbufs.
304  *
305  * => called with so_pendfree_slock held.
306  * => called at splvm.
307  */
308 
309 static size_t
310 sodopendfreel()
311 {
312 	size_t rv = 0;
313 
314 	LOCK_ASSERT(simple_lock_held(&so_pendfree_slock));
315 
316 	for (;;) {
317 		struct mbuf *m;
318 		struct mbuf *next;
319 
320 		m = so_pendfree;
321 		if (m == NULL)
322 			break;
323 		so_pendfree = NULL;
324 		simple_unlock(&so_pendfree_slock);
325 		/* XXX splx */
326 
327 		for (; m != NULL; m = next) {
328 			next = m->m_next;
329 
330 			rv += m->m_ext.ext_size;
331 			sodoloanfree((m->m_flags & M_EXT_PAGES) ?
332 			    m->m_ext.ext_pgs : NULL, m->m_ext.ext_buf,
333 			    m->m_ext.ext_size);
334 			pool_cache_put(&mbpool_cache, m);
335 		}
336 
337 		/* XXX splvm */
338 		simple_lock(&so_pendfree_slock);
339 	}
340 
341 	return (rv);
342 }
343 
344 void
345 soloanfree(struct mbuf *m, caddr_t buf, size_t size, void *arg)
346 {
347 	int s;
348 
349 	if (m == NULL) {
350 
351 		/*
352 		 * called from MEXTREMOVE.
353 		 */
354 
355 		sodoloanfree(NULL, buf, size);
356 		return;
357 	}
358 
359 	/*
360 	 * postpone freeing mbuf.
361 	 *
362 	 * we can't do it in interrupt context
363 	 * because we need to put kva back to kernel_map.
364 	 */
365 
366 	s = splvm();
367 	simple_lock(&so_pendfree_slock);
368 	m->m_next = so_pendfree;
369 	so_pendfree = m;
370 	if (sokvawaiters)
371 		wakeup(&socurkva);
372 	simple_unlock(&so_pendfree_slock);
373 	splx(s);
374 }
375 
376 static long
377 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space)
378 {
379 	struct iovec *iov = uio->uio_iov;
380 	vaddr_t sva, eva;
381 	vsize_t len;
382 	vaddr_t lva, va;
383 	int npgs, i, error;
384 
385 	if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace))
386 		return (0);
387 
388 	if (iov->iov_len < (size_t) space)
389 		space = iov->iov_len;
390 	if (space > SOCK_LOAN_CHUNK)
391 		space = SOCK_LOAN_CHUNK;
392 
393 	eva = round_page((vaddr_t) iov->iov_base + space);
394 	sva = trunc_page((vaddr_t) iov->iov_base);
395 	len = eva - sva;
396 	npgs = len >> PAGE_SHIFT;
397 
398 	/* XXX KDASSERT */
399 	KASSERT(npgs <= M_EXT_MAXPAGES);
400 
401 	lva = sokvaalloc(len, so);
402 	if (lva == 0)
403 		return 0;
404 
405 	error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len,
406 	    m->m_ext.ext_pgs, UVM_LOAN_TOPAGE);
407 	if (error) {
408 		sokvafree(lva, len);
409 		return (0);
410 	}
411 
412 	for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE)
413 		pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
414 		    VM_PROT_READ);
415 	pmap_update(pmap_kernel());
416 
417 	lva += (vaddr_t) iov->iov_base & PAGE_MASK;
418 
419 	MEXTADD(m, (caddr_t) lva, space, M_MBUF, soloanfree, so);
420 	m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
421 
422 	uio->uio_resid -= space;
423 	/* uio_offset not updated, not set/used for write(2) */
424 	uio->uio_iov->iov_base = (caddr_t) uio->uio_iov->iov_base + space;
425 	uio->uio_iov->iov_len -= space;
426 	if (uio->uio_iov->iov_len == 0) {
427 		uio->uio_iov++;
428 		uio->uio_iovcnt--;
429 	}
430 
431 	return (space);
432 }
433 
434 static int
435 sokva_reclaim_callback(struct callback_entry *ce, void *obj, void *arg)
436 {
437 
438 	KASSERT(ce == &sokva_reclaimerentry);
439 	KASSERT(obj == NULL);
440 
441 	sodopendfree();
442 	if (!vm_map_starved_p(kernel_map)) {
443 		return CALLBACK_CHAIN_ABORT;
444 	}
445 	return CALLBACK_CHAIN_CONTINUE;
446 }
447 
448 void
449 soinit(void)
450 {
451 
452 	/* Set the initial adjusted socket buffer size. */
453 	if (sb_max_set(sb_max))
454 		panic("bad initial sb_max value: %lu", sb_max);
455 
456 	callback_register(&vm_map_to_kernel(kernel_map)->vmk_reclaim_callback,
457 	    &sokva_reclaimerentry, NULL, sokva_reclaim_callback);
458 }
459 
460 /*
461  * Socket operation routines.
462  * These routines are called by the routines in
463  * sys_socket.c or from a system process, and
464  * implement the semantics of socket operations by
465  * switching out to the protocol specific routines.
466  */
467 /*ARGSUSED*/
468 int
469 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l)
470 {
471 	const struct protosw	*prp;
472 	struct socket	*so;
473 	uid_t		uid;
474 	int		error, s;
475 
476 	if (proto)
477 		prp = pffindproto(dom, proto, type);
478 	else
479 		prp = pffindtype(dom, type);
480 	if (prp == 0) {
481 		/* no support for domain */
482 		if (pffinddomain(dom) == 0)
483 			return (EAFNOSUPPORT);
484 		/* no support for socket type */
485 		if (proto == 0 && type != 0)
486 			return (EPROTOTYPE);
487 		return (EPROTONOSUPPORT);
488 	}
489 	if (prp->pr_usrreq == 0)
490 		return (EPROTONOSUPPORT);
491 	if (prp->pr_type != type)
492 		return (EPROTOTYPE);
493 	s = splsoftnet();
494 	so = pool_get(&socket_pool, PR_WAITOK);
495 	memset((caddr_t)so, 0, sizeof(*so));
496 	TAILQ_INIT(&so->so_q0);
497 	TAILQ_INIT(&so->so_q);
498 	so->so_type = type;
499 	so->so_proto = prp;
500 	so->so_send = sosend;
501 	so->so_receive = soreceive;
502 #ifdef MBUFTRACE
503 	so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner;
504 	so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner;
505 	so->so_mowner = &prp->pr_domain->dom_mowner;
506 #endif
507 	if (l != NULL) {
508 		uid = kauth_cred_geteuid(l->l_proc->p_cred);
509 	} else {
510 		uid = 0;
511 	}
512 	so->so_uidinfo = uid_find(uid);
513 	error = (*prp->pr_usrreq)(so, PRU_ATTACH, (struct mbuf *)0,
514 	    (struct mbuf *)(long)proto, (struct mbuf *)0, l);
515 	if (error) {
516 		so->so_state |= SS_NOFDREF;
517 		sofree(so);
518 		splx(s);
519 		return (error);
520 	}
521 	splx(s);
522 	*aso = so;
523 	return (0);
524 }
525 
526 int
527 sobind(struct socket *so, struct mbuf *nam, struct lwp *l)
528 {
529 	int	s, error;
530 
531 	s = splsoftnet();
532 	error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, (struct mbuf *)0,
533 	    nam, (struct mbuf *)0, l);
534 	splx(s);
535 	return (error);
536 }
537 
538 int
539 solisten(struct socket *so, int backlog)
540 {
541 	int	s, error;
542 
543 	s = splsoftnet();
544 	error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, (struct mbuf *)0,
545 	    (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0);
546 	if (error) {
547 		splx(s);
548 		return (error);
549 	}
550 	if (TAILQ_EMPTY(&so->so_q))
551 		so->so_options |= SO_ACCEPTCONN;
552 	if (backlog < 0)
553 		backlog = 0;
554 	so->so_qlimit = min(backlog, somaxconn);
555 	splx(s);
556 	return (0);
557 }
558 
559 void
560 sofree(struct socket *so)
561 {
562 
563 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0)
564 		return;
565 	if (so->so_head) {
566 		/*
567 		 * We must not decommission a socket that's on the accept(2)
568 		 * queue.  If we do, then accept(2) may hang after select(2)
569 		 * indicated that the listening socket was ready.
570 		 */
571 		if (!soqremque(so, 0))
572 			return;
573 	}
574 	if (so->so_rcv.sb_hiwat)
575 		(void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0,
576 		    RLIM_INFINITY);
577 	if (so->so_snd.sb_hiwat)
578 		(void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0,
579 		    RLIM_INFINITY);
580 	sbrelease(&so->so_snd, so);
581 	sorflush(so);
582 	pool_put(&socket_pool, so);
583 }
584 
585 /*
586  * Close a socket on last file table reference removal.
587  * Initiate disconnect if connected.
588  * Free socket when disconnect complete.
589  */
590 int
591 soclose(struct socket *so)
592 {
593 	struct socket	*so2;
594 	int		s, error;
595 
596 	error = 0;
597 	s = splsoftnet();		/* conservative */
598 	if (so->so_options & SO_ACCEPTCONN) {
599 		while ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) {
600 			(void) soqremque(so2, 0);
601 			(void) soabort(so2);
602 		}
603 		while ((so2 = TAILQ_FIRST(&so->so_q)) != 0) {
604 			(void) soqremque(so2, 1);
605 			(void) soabort(so2);
606 		}
607 	}
608 	if (so->so_pcb == 0)
609 		goto discard;
610 	if (so->so_state & SS_ISCONNECTED) {
611 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
612 			error = sodisconnect(so);
613 			if (error)
614 				goto drop;
615 		}
616 		if (so->so_options & SO_LINGER) {
617 			if ((so->so_state & SS_ISDISCONNECTING) &&
618 			    (so->so_state & SS_NBIO))
619 				goto drop;
620 			while (so->so_state & SS_ISCONNECTED) {
621 				error = tsleep((caddr_t)&so->so_timeo,
622 					       PSOCK | PCATCH, netcls,
623 					       so->so_linger * hz);
624 				if (error)
625 					break;
626 			}
627 		}
628 	}
629  drop:
630 	if (so->so_pcb) {
631 		int error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH,
632 		    (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0,
633 		    (struct lwp *)0);
634 		if (error == 0)
635 			error = error2;
636 	}
637  discard:
638 	if (so->so_state & SS_NOFDREF)
639 		panic("soclose: NOFDREF");
640 	so->so_state |= SS_NOFDREF;
641 	sofree(so);
642 	splx(s);
643 	return (error);
644 }
645 
646 /*
647  * Must be called at splsoftnet...
648  */
649 int
650 soabort(struct socket *so)
651 {
652 
653 	return (*so->so_proto->pr_usrreq)(so, PRU_ABORT, (struct mbuf *)0,
654 	    (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0);
655 }
656 
657 int
658 soaccept(struct socket *so, struct mbuf *nam)
659 {
660 	int	s, error;
661 
662 	error = 0;
663 	s = splsoftnet();
664 	if ((so->so_state & SS_NOFDREF) == 0)
665 		panic("soaccept: !NOFDREF");
666 	so->so_state &= ~SS_NOFDREF;
667 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
668 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
669 		error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT,
670 		    (struct mbuf *)0, nam, (struct mbuf *)0, (struct lwp *)0);
671 	else
672 		error = ECONNABORTED;
673 
674 	splx(s);
675 	return (error);
676 }
677 
678 int
679 soconnect(struct socket *so, struct mbuf *nam, struct lwp *l)
680 {
681 	int		s, error;
682 
683 	if (so->so_options & SO_ACCEPTCONN)
684 		return (EOPNOTSUPP);
685 	s = splsoftnet();
686 	/*
687 	 * If protocol is connection-based, can only connect once.
688 	 * Otherwise, if connected, try to disconnect first.
689 	 * This allows user to disconnect by connecting to, e.g.,
690 	 * a null address.
691 	 */
692 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
693 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
694 	    (error = sodisconnect(so))))
695 		error = EISCONN;
696 	else
697 		error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT,
698 		    (struct mbuf *)0, nam, (struct mbuf *)0, l);
699 	splx(s);
700 	return (error);
701 }
702 
703 int
704 soconnect2(struct socket *so1, struct socket *so2)
705 {
706 	int	s, error;
707 
708 	s = splsoftnet();
709 	error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2,
710 	    (struct mbuf *)0, (struct mbuf *)so2, (struct mbuf *)0,
711 	    (struct lwp *)0);
712 	splx(s);
713 	return (error);
714 }
715 
716 int
717 sodisconnect(struct socket *so)
718 {
719 	int	s, error;
720 
721 	s = splsoftnet();
722 	if ((so->so_state & SS_ISCONNECTED) == 0) {
723 		error = ENOTCONN;
724 		goto bad;
725 	}
726 	if (so->so_state & SS_ISDISCONNECTING) {
727 		error = EALREADY;
728 		goto bad;
729 	}
730 	error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT,
731 	    (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0,
732 	    (struct lwp *)0);
733  bad:
734 	splx(s);
735 	sodopendfree();
736 	return (error);
737 }
738 
739 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
740 /*
741  * Send on a socket.
742  * If send must go all at once and message is larger than
743  * send buffering, then hard error.
744  * Lock against other senders.
745  * If must go all at once and not enough room now, then
746  * inform user that this would block and do nothing.
747  * Otherwise, if nonblocking, send as much as possible.
748  * The data to be sent is described by "uio" if nonzero,
749  * otherwise by the mbuf chain "top" (which must be null
750  * if uio is not).  Data provided in mbuf chain must be small
751  * enough to send all at once.
752  *
753  * Returns nonzero on error, timeout or signal; callers
754  * must check for short counts if EINTR/ERESTART are returned.
755  * Data and control buffers are freed on return.
756  */
757 int
758 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
759 	struct mbuf *control, int flags, struct lwp *l)
760 {
761 	struct mbuf	**mp, *m;
762 	struct proc	*p;
763 	long		space, len, resid, clen, mlen;
764 	int		error, s, dontroute, atomic;
765 
766 	p = l->l_proc;
767 	sodopendfree();
768 
769 	clen = 0;
770 	atomic = sosendallatonce(so) || top;
771 	if (uio)
772 		resid = uio->uio_resid;
773 	else
774 		resid = top->m_pkthdr.len;
775 	/*
776 	 * In theory resid should be unsigned.
777 	 * However, space must be signed, as it might be less than 0
778 	 * if we over-committed, and we must use a signed comparison
779 	 * of space and resid.  On the other hand, a negative resid
780 	 * causes us to loop sending 0-length segments to the protocol.
781 	 */
782 	if (resid < 0) {
783 		error = EINVAL;
784 		goto out;
785 	}
786 	dontroute =
787 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
788 	    (so->so_proto->pr_flags & PR_ATOMIC);
789 	if (p)
790 		p->p_stats->p_ru.ru_msgsnd++;
791 	if (control)
792 		clen = control->m_len;
793 #define	snderr(errno)	{ error = errno; splx(s); goto release; }
794 
795  restart:
796 	if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
797 		goto out;
798 	do {
799 		s = splsoftnet();
800 		if (so->so_state & SS_CANTSENDMORE)
801 			snderr(EPIPE);
802 		if (so->so_error) {
803 			error = so->so_error;
804 			so->so_error = 0;
805 			splx(s);
806 			goto release;
807 		}
808 		if ((so->so_state & SS_ISCONNECTED) == 0) {
809 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
810 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
811 				    !(resid == 0 && clen != 0))
812 					snderr(ENOTCONN);
813 			} else if (addr == 0)
814 				snderr(EDESTADDRREQ);
815 		}
816 		space = sbspace(&so->so_snd);
817 		if (flags & MSG_OOB)
818 			space += 1024;
819 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
820 		    clen > so->so_snd.sb_hiwat)
821 			snderr(EMSGSIZE);
822 		if (space < resid + clen &&
823 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
824 			if (so->so_state & SS_NBIO)
825 				snderr(EWOULDBLOCK);
826 			sbunlock(&so->so_snd);
827 			error = sbwait(&so->so_snd);
828 			splx(s);
829 			if (error)
830 				goto out;
831 			goto restart;
832 		}
833 		splx(s);
834 		mp = &top;
835 		space -= clen;
836 		do {
837 			if (uio == NULL) {
838 				/*
839 				 * Data is prepackaged in "top".
840 				 */
841 				resid = 0;
842 				if (flags & MSG_EOR)
843 					top->m_flags |= M_EOR;
844 			} else do {
845 				if (top == 0) {
846 					m = m_gethdr(M_WAIT, MT_DATA);
847 					mlen = MHLEN;
848 					m->m_pkthdr.len = 0;
849 					m->m_pkthdr.rcvif = (struct ifnet *)0;
850 				} else {
851 					m = m_get(M_WAIT, MT_DATA);
852 					mlen = MLEN;
853 				}
854 				MCLAIM(m, so->so_snd.sb_mowner);
855 				if (sock_loan_thresh >= 0 &&
856 				    uio->uio_iov->iov_len >= sock_loan_thresh &&
857 				    space >= sock_loan_thresh &&
858 				    (len = sosend_loan(so, uio, m,
859 						       space)) != 0) {
860 					SOSEND_COUNTER_INCR(&sosend_loan_big);
861 					space -= len;
862 					goto have_data;
863 				}
864 				if (resid >= MINCLSIZE && space >= MCLBYTES) {
865 					SOSEND_COUNTER_INCR(&sosend_copy_big);
866 					m_clget(m, M_WAIT);
867 					if ((m->m_flags & M_EXT) == 0)
868 						goto nopages;
869 					mlen = MCLBYTES;
870 					if (atomic && top == 0) {
871 						len = lmin(MCLBYTES - max_hdr,
872 						    resid);
873 						m->m_data += max_hdr;
874 					} else
875 						len = lmin(MCLBYTES, resid);
876 					space -= len;
877 				} else {
878  nopages:
879 					SOSEND_COUNTER_INCR(&sosend_copy_small);
880 					len = lmin(lmin(mlen, resid), space);
881 					space -= len;
882 					/*
883 					 * For datagram protocols, leave room
884 					 * for protocol headers in first mbuf.
885 					 */
886 					if (atomic && top == 0 && len < mlen)
887 						MH_ALIGN(m, len);
888 				}
889 				error = uiomove(mtod(m, caddr_t), (int)len,
890 				    uio);
891  have_data:
892 				resid = uio->uio_resid;
893 				m->m_len = len;
894 				*mp = m;
895 				top->m_pkthdr.len += len;
896 				if (error)
897 					goto release;
898 				mp = &m->m_next;
899 				if (resid <= 0) {
900 					if (flags & MSG_EOR)
901 						top->m_flags |= M_EOR;
902 					break;
903 				}
904 			} while (space > 0 && atomic);
905 
906 			s = splsoftnet();
907 
908 			if (so->so_state & SS_CANTSENDMORE)
909 				snderr(EPIPE);
910 
911 			if (dontroute)
912 				so->so_options |= SO_DONTROUTE;
913 			if (resid > 0)
914 				so->so_state |= SS_MORETOCOME;
915 			error = (*so->so_proto->pr_usrreq)(so,
916 			    (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND,
917 			    top, addr, control, curlwp);	/* XXX */
918 			if (dontroute)
919 				so->so_options &= ~SO_DONTROUTE;
920 			if (resid > 0)
921 				so->so_state &= ~SS_MORETOCOME;
922 			splx(s);
923 
924 			clen = 0;
925 			control = 0;
926 			top = 0;
927 			mp = &top;
928 			if (error)
929 				goto release;
930 		} while (resid && space > 0);
931 	} while (resid);
932 
933  release:
934 	sbunlock(&so->so_snd);
935  out:
936 	if (top)
937 		m_freem(top);
938 	if (control)
939 		m_freem(control);
940 	return (error);
941 }
942 
943 /*
944  * Implement receive operations on a socket.
945  * We depend on the way that records are added to the sockbuf
946  * by sbappend*.  In particular, each record (mbufs linked through m_next)
947  * must begin with an address if the protocol so specifies,
948  * followed by an optional mbuf or mbufs containing ancillary data,
949  * and then zero or more mbufs of data.
950  * In order to avoid blocking network interrupts for the entire time here,
951  * we splx() while doing the actual copy to user space.
952  * Although the sockbuf is locked, new data may still be appended,
953  * and thus we must maintain consistency of the sockbuf during that time.
954  *
955  * The caller may receive the data as a single mbuf chain by supplying
956  * an mbuf **mp0 for use in returning the chain.  The uio is then used
957  * only for the count in uio_resid.
958  */
959 int
960 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
961 	struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
962 {
963 	struct lwp *l = curlwp;
964 	struct mbuf	*m, **mp;
965 	int		flags, len, error, s, offset, moff, type, orig_resid;
966 	const struct protosw	*pr;
967 	struct mbuf	*nextrecord;
968 	int		mbuf_removed = 0;
969 
970 	pr = so->so_proto;
971 	mp = mp0;
972 	type = 0;
973 	orig_resid = uio->uio_resid;
974 
975 	if (paddr)
976 		*paddr = 0;
977 	if (controlp)
978 		*controlp = 0;
979 	if (flagsp)
980 		flags = *flagsp &~ MSG_EOR;
981 	else
982 		flags = 0;
983 
984 	if ((flags & MSG_DONTWAIT) == 0)
985 		sodopendfree();
986 
987 	if (flags & MSG_OOB) {
988 		m = m_get(M_WAIT, MT_DATA);
989 		error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m,
990 		    (struct mbuf *)(long)(flags & MSG_PEEK),
991 		    (struct mbuf *)0, l);
992 		if (error)
993 			goto bad;
994 		do {
995 			error = uiomove(mtod(m, caddr_t),
996 			    (int) min(uio->uio_resid, m->m_len), uio);
997 			m = m_free(m);
998 		} while (uio->uio_resid && error == 0 && m);
999  bad:
1000 		if (m)
1001 			m_freem(m);
1002 		return (error);
1003 	}
1004 	if (mp)
1005 		*mp = (struct mbuf *)0;
1006 	if (so->so_state & SS_ISCONFIRMING && uio->uio_resid)
1007 		(*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0,
1008 		    (struct mbuf *)0, (struct mbuf *)0, l);
1009 
1010  restart:
1011 	if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0)
1012 		return (error);
1013 	s = splsoftnet();
1014 
1015 	m = so->so_rcv.sb_mb;
1016 	/*
1017 	 * If we have less data than requested, block awaiting more
1018 	 * (subject to any timeout) if:
1019 	 *   1. the current count is less than the low water mark,
1020 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1021 	 *	receive operation at once if we block (resid <= hiwat), or
1022 	 *   3. MSG_DONTWAIT is not set.
1023 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1024 	 * we have to do the receive in sections, and thus risk returning
1025 	 * a short count if a timeout or signal occurs after we start.
1026 	 */
1027 	if (m == 0 || (((flags & MSG_DONTWAIT) == 0 &&
1028 	    so->so_rcv.sb_cc < uio->uio_resid) &&
1029 	    (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1030 	    ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1031 	    m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
1032 #ifdef DIAGNOSTIC
1033 		if (m == 0 && so->so_rcv.sb_cc)
1034 			panic("receive 1");
1035 #endif
1036 		if (so->so_error) {
1037 			if (m)
1038 				goto dontblock;
1039 			error = so->so_error;
1040 			if ((flags & MSG_PEEK) == 0)
1041 				so->so_error = 0;
1042 			goto release;
1043 		}
1044 		if (so->so_state & SS_CANTRCVMORE) {
1045 			if (m)
1046 				goto dontblock;
1047 			else
1048 				goto release;
1049 		}
1050 		for (; m; m = m->m_next)
1051 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1052 				m = so->so_rcv.sb_mb;
1053 				goto dontblock;
1054 			}
1055 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1056 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1057 			error = ENOTCONN;
1058 			goto release;
1059 		}
1060 		if (uio->uio_resid == 0)
1061 			goto release;
1062 		if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT)) {
1063 			error = EWOULDBLOCK;
1064 			goto release;
1065 		}
1066 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
1067 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
1068 		sbunlock(&so->so_rcv);
1069 		error = sbwait(&so->so_rcv);
1070 		splx(s);
1071 		if (error)
1072 			return (error);
1073 		goto restart;
1074 	}
1075  dontblock:
1076 	/*
1077 	 * On entry here, m points to the first record of the socket buffer.
1078 	 * While we process the initial mbufs containing address and control
1079 	 * info, we save a copy of m->m_nextpkt into nextrecord.
1080 	 */
1081 	if (l)
1082 		l->l_proc->p_stats->p_ru.ru_msgrcv++;
1083 	KASSERT(m == so->so_rcv.sb_mb);
1084 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
1085 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
1086 	nextrecord = m->m_nextpkt;
1087 	if (pr->pr_flags & PR_ADDR) {
1088 #ifdef DIAGNOSTIC
1089 		if (m->m_type != MT_SONAME)
1090 			panic("receive 1a");
1091 #endif
1092 		orig_resid = 0;
1093 		if (flags & MSG_PEEK) {
1094 			if (paddr)
1095 				*paddr = m_copy(m, 0, m->m_len);
1096 			m = m->m_next;
1097 		} else {
1098 			sbfree(&so->so_rcv, m);
1099 			mbuf_removed = 1;
1100 			if (paddr) {
1101 				*paddr = m;
1102 				so->so_rcv.sb_mb = m->m_next;
1103 				m->m_next = 0;
1104 				m = so->so_rcv.sb_mb;
1105 			} else {
1106 				MFREE(m, so->so_rcv.sb_mb);
1107 				m = so->so_rcv.sb_mb;
1108 			}
1109 		}
1110 	}
1111 	while (m && m->m_type == MT_CONTROL && error == 0) {
1112 		if (flags & MSG_PEEK) {
1113 			if (controlp)
1114 				*controlp = m_copy(m, 0, m->m_len);
1115 			m = m->m_next;
1116 		} else {
1117 			sbfree(&so->so_rcv, m);
1118 			mbuf_removed = 1;
1119 			if (controlp) {
1120 				struct domain *dom = pr->pr_domain;
1121 				if (dom->dom_externalize && l &&
1122 				    mtod(m, struct cmsghdr *)->cmsg_type ==
1123 				    SCM_RIGHTS)
1124 					error = (*dom->dom_externalize)(m, l);
1125 				*controlp = m;
1126 				so->so_rcv.sb_mb = m->m_next;
1127 				m->m_next = 0;
1128 				m = so->so_rcv.sb_mb;
1129 			} else {
1130 				/*
1131 				 * Dispose of any SCM_RIGHTS message that went
1132 				 * through the read path rather than recv.
1133 				 */
1134 				if (pr->pr_domain->dom_dispose &&
1135 				    mtod(m, struct cmsghdr *)->cmsg_type == SCM_RIGHTS)
1136 					(*pr->pr_domain->dom_dispose)(m);
1137 				MFREE(m, so->so_rcv.sb_mb);
1138 				m = so->so_rcv.sb_mb;
1139 			}
1140 		}
1141 		if (controlp) {
1142 			orig_resid = 0;
1143 			controlp = &(*controlp)->m_next;
1144 		}
1145 	}
1146 
1147 	/*
1148 	 * If m is non-NULL, we have some data to read.  From now on,
1149 	 * make sure to keep sb_lastrecord consistent when working on
1150 	 * the last packet on the chain (nextrecord == NULL) and we
1151 	 * change m->m_nextpkt.
1152 	 */
1153 	if (m) {
1154 		if ((flags & MSG_PEEK) == 0) {
1155 			m->m_nextpkt = nextrecord;
1156 			/*
1157 			 * If nextrecord == NULL (this is a single chain),
1158 			 * then sb_lastrecord may not be valid here if m
1159 			 * was changed earlier.
1160 			 */
1161 			if (nextrecord == NULL) {
1162 				KASSERT(so->so_rcv.sb_mb == m);
1163 				so->so_rcv.sb_lastrecord = m;
1164 			}
1165 		}
1166 		type = m->m_type;
1167 		if (type == MT_OOBDATA)
1168 			flags |= MSG_OOB;
1169 	} else {
1170 		if ((flags & MSG_PEEK) == 0) {
1171 			KASSERT(so->so_rcv.sb_mb == m);
1172 			so->so_rcv.sb_mb = nextrecord;
1173 			SB_EMPTY_FIXUP(&so->so_rcv);
1174 		}
1175 	}
1176 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1177 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1178 
1179 	moff = 0;
1180 	offset = 0;
1181 	while (m && uio->uio_resid > 0 && error == 0) {
1182 		if (m->m_type == MT_OOBDATA) {
1183 			if (type != MT_OOBDATA)
1184 				break;
1185 		} else if (type == MT_OOBDATA)
1186 			break;
1187 #ifdef DIAGNOSTIC
1188 		else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
1189 			panic("receive 3");
1190 #endif
1191 		so->so_state &= ~SS_RCVATMARK;
1192 		len = uio->uio_resid;
1193 		if (so->so_oobmark && len > so->so_oobmark - offset)
1194 			len = so->so_oobmark - offset;
1195 		if (len > m->m_len - moff)
1196 			len = m->m_len - moff;
1197 		/*
1198 		 * If mp is set, just pass back the mbufs.
1199 		 * Otherwise copy them out via the uio, then free.
1200 		 * Sockbuf must be consistent here (points to current mbuf,
1201 		 * it points to next record) when we drop priority;
1202 		 * we must note any additions to the sockbuf when we
1203 		 * block interrupts again.
1204 		 */
1205 		if (mp == 0) {
1206 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1207 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1208 			splx(s);
1209 			error = uiomove(mtod(m, caddr_t) + moff, (int)len, uio);
1210 			s = splsoftnet();
1211 			if (error) {
1212 				/*
1213 				 * If any part of the record has been removed
1214 				 * (such as the MT_SONAME mbuf, which will
1215 				 * happen when PR_ADDR, and thus also
1216 				 * PR_ATOMIC, is set), then drop the entire
1217 				 * record to maintain the atomicity of the
1218 				 * receive operation.
1219 				 *
1220 				 * This avoids a later panic("receive 1a")
1221 				 * when compiled with DIAGNOSTIC.
1222 				 */
1223 				if (m && mbuf_removed
1224 				    && (pr->pr_flags & PR_ATOMIC))
1225 					(void) sbdroprecord(&so->so_rcv);
1226 
1227 				goto release;
1228 			}
1229 		} else
1230 			uio->uio_resid -= len;
1231 		if (len == m->m_len - moff) {
1232 			if (m->m_flags & M_EOR)
1233 				flags |= MSG_EOR;
1234 			if (flags & MSG_PEEK) {
1235 				m = m->m_next;
1236 				moff = 0;
1237 			} else {
1238 				nextrecord = m->m_nextpkt;
1239 				sbfree(&so->so_rcv, m);
1240 				if (mp) {
1241 					*mp = m;
1242 					mp = &m->m_next;
1243 					so->so_rcv.sb_mb = m = m->m_next;
1244 					*mp = (struct mbuf *)0;
1245 				} else {
1246 					MFREE(m, so->so_rcv.sb_mb);
1247 					m = so->so_rcv.sb_mb;
1248 				}
1249 				/*
1250 				 * If m != NULL, we also know that
1251 				 * so->so_rcv.sb_mb != NULL.
1252 				 */
1253 				KASSERT(so->so_rcv.sb_mb == m);
1254 				if (m) {
1255 					m->m_nextpkt = nextrecord;
1256 					if (nextrecord == NULL)
1257 						so->so_rcv.sb_lastrecord = m;
1258 				} else {
1259 					so->so_rcv.sb_mb = nextrecord;
1260 					SB_EMPTY_FIXUP(&so->so_rcv);
1261 				}
1262 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1263 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1264 			}
1265 		} else {
1266 			if (flags & MSG_PEEK)
1267 				moff += len;
1268 			else {
1269 				if (mp)
1270 					*mp = m_copym(m, 0, len, M_WAIT);
1271 				m->m_data += len;
1272 				m->m_len -= len;
1273 				so->so_rcv.sb_cc -= len;
1274 			}
1275 		}
1276 		if (so->so_oobmark) {
1277 			if ((flags & MSG_PEEK) == 0) {
1278 				so->so_oobmark -= len;
1279 				if (so->so_oobmark == 0) {
1280 					so->so_state |= SS_RCVATMARK;
1281 					break;
1282 				}
1283 			} else {
1284 				offset += len;
1285 				if (offset == so->so_oobmark)
1286 					break;
1287 			}
1288 		}
1289 		if (flags & MSG_EOR)
1290 			break;
1291 		/*
1292 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1293 		 * we must not quit until "uio->uio_resid == 0" or an error
1294 		 * termination.  If a signal/timeout occurs, return
1295 		 * with a short count but without error.
1296 		 * Keep sockbuf locked against other readers.
1297 		 */
1298 		while (flags & MSG_WAITALL && m == 0 && uio->uio_resid > 0 &&
1299 		    !sosendallatonce(so) && !nextrecord) {
1300 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1301 				break;
1302 			/*
1303 			 * If we are peeking and the socket receive buffer is
1304 			 * full, stop since we can't get more data to peek at.
1305 			 */
1306 			if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
1307 				break;
1308 			/*
1309 			 * If we've drained the socket buffer, tell the
1310 			 * protocol in case it needs to do something to
1311 			 * get it filled again.
1312 			 */
1313 			if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1314 				(*pr->pr_usrreq)(so, PRU_RCVD,
1315 				    (struct mbuf *)0,
1316 				    (struct mbuf *)(long)flags,
1317 				    (struct mbuf *)0, l);
1318 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1319 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1320 			error = sbwait(&so->so_rcv);
1321 			if (error) {
1322 				sbunlock(&so->so_rcv);
1323 				splx(s);
1324 				return (0);
1325 			}
1326 			if ((m = so->so_rcv.sb_mb) != NULL)
1327 				nextrecord = m->m_nextpkt;
1328 		}
1329 	}
1330 
1331 	if (m && pr->pr_flags & PR_ATOMIC) {
1332 		flags |= MSG_TRUNC;
1333 		if ((flags & MSG_PEEK) == 0)
1334 			(void) sbdroprecord(&so->so_rcv);
1335 	}
1336 	if ((flags & MSG_PEEK) == 0) {
1337 		if (m == 0) {
1338 			/*
1339 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1340 			 * part makes sure sb_lastrecord is up-to-date if
1341 			 * there is still data in the socket buffer.
1342 			 */
1343 			so->so_rcv.sb_mb = nextrecord;
1344 			if (so->so_rcv.sb_mb == NULL) {
1345 				so->so_rcv.sb_mbtail = NULL;
1346 				so->so_rcv.sb_lastrecord = NULL;
1347 			} else if (nextrecord->m_nextpkt == NULL)
1348 				so->so_rcv.sb_lastrecord = nextrecord;
1349 		}
1350 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1351 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1352 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1353 			(*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0,
1354 			    (struct mbuf *)(long)flags, (struct mbuf *)0, l);
1355 	}
1356 	if (orig_resid == uio->uio_resid && orig_resid &&
1357 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1358 		sbunlock(&so->so_rcv);
1359 		splx(s);
1360 		goto restart;
1361 	}
1362 
1363 	if (flagsp)
1364 		*flagsp |= flags;
1365  release:
1366 	sbunlock(&so->so_rcv);
1367 	splx(s);
1368 	return (error);
1369 }
1370 
1371 int
1372 soshutdown(struct socket *so, int how)
1373 {
1374 	const struct protosw	*pr;
1375 
1376 	pr = so->so_proto;
1377 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1378 		return (EINVAL);
1379 
1380 	if (how == SHUT_RD || how == SHUT_RDWR)
1381 		sorflush(so);
1382 	if (how == SHUT_WR || how == SHUT_RDWR)
1383 		return (*pr->pr_usrreq)(so, PRU_SHUTDOWN, (struct mbuf *)0,
1384 		    (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0);
1385 	return (0);
1386 }
1387 
1388 void
1389 sorflush(struct socket *so)
1390 {
1391 	struct sockbuf	*sb, asb;
1392 	const struct protosw	*pr;
1393 	int		s;
1394 
1395 	sb = &so->so_rcv;
1396 	pr = so->so_proto;
1397 	sb->sb_flags |= SB_NOINTR;
1398 	(void) sblock(sb, M_WAITOK);
1399 	s = splnet();
1400 	socantrcvmore(so);
1401 	sbunlock(sb);
1402 	asb = *sb;
1403 	/*
1404 	 * Clear most of the sockbuf structure, but leave some of the
1405 	 * fields valid.
1406 	 */
1407 	memset(&sb->sb_startzero, 0,
1408 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1409 	splx(s);
1410 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose)
1411 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1412 	sbrelease(&asb, so);
1413 }
1414 
1415 int
1416 sosetopt(struct socket *so, int level, int optname, struct mbuf *m0)
1417 {
1418 	int		error;
1419 	struct mbuf	*m;
1420 
1421 	error = 0;
1422 	m = m0;
1423 	if (level != SOL_SOCKET) {
1424 		if (so->so_proto && so->so_proto->pr_ctloutput)
1425 			return ((*so->so_proto->pr_ctloutput)
1426 				  (PRCO_SETOPT, so, level, optname, &m0));
1427 		error = ENOPROTOOPT;
1428 	} else {
1429 		switch (optname) {
1430 
1431 		case SO_LINGER:
1432 			if (m == NULL || m->m_len != sizeof(struct linger)) {
1433 				error = EINVAL;
1434 				goto bad;
1435 			}
1436 			if (mtod(m, struct linger *)->l_linger < 0 ||
1437 			    mtod(m, struct linger *)->l_linger > (INT_MAX / hz)) {
1438 				error = EDOM;
1439 				goto bad;
1440 			}
1441 			so->so_linger = mtod(m, struct linger *)->l_linger;
1442 			/* fall thru... */
1443 
1444 		case SO_DEBUG:
1445 		case SO_KEEPALIVE:
1446 		case SO_DONTROUTE:
1447 		case SO_USELOOPBACK:
1448 		case SO_BROADCAST:
1449 		case SO_REUSEADDR:
1450 		case SO_REUSEPORT:
1451 		case SO_OOBINLINE:
1452 		case SO_TIMESTAMP:
1453 			if (m == NULL || m->m_len < sizeof(int)) {
1454 				error = EINVAL;
1455 				goto bad;
1456 			}
1457 			if (*mtod(m, int *))
1458 				so->so_options |= optname;
1459 			else
1460 				so->so_options &= ~optname;
1461 			break;
1462 
1463 		case SO_SNDBUF:
1464 		case SO_RCVBUF:
1465 		case SO_SNDLOWAT:
1466 		case SO_RCVLOWAT:
1467 		    {
1468 			int optval;
1469 
1470 			if (m == NULL || m->m_len < sizeof(int)) {
1471 				error = EINVAL;
1472 				goto bad;
1473 			}
1474 
1475 			/*
1476 			 * Values < 1 make no sense for any of these
1477 			 * options, so disallow them.
1478 			 */
1479 			optval = *mtod(m, int *);
1480 			if (optval < 1) {
1481 				error = EINVAL;
1482 				goto bad;
1483 			}
1484 
1485 			switch (optname) {
1486 
1487 			case SO_SNDBUF:
1488 			case SO_RCVBUF:
1489 				if (sbreserve(optname == SO_SNDBUF ?
1490 				    &so->so_snd : &so->so_rcv,
1491 				    (u_long) optval, so) == 0) {
1492 					error = ENOBUFS;
1493 					goto bad;
1494 				}
1495 				break;
1496 
1497 			/*
1498 			 * Make sure the low-water is never greater than
1499 			 * the high-water.
1500 			 */
1501 			case SO_SNDLOWAT:
1502 				so->so_snd.sb_lowat =
1503 				    (optval > so->so_snd.sb_hiwat) ?
1504 				    so->so_snd.sb_hiwat : optval;
1505 				break;
1506 			case SO_RCVLOWAT:
1507 				so->so_rcv.sb_lowat =
1508 				    (optval > so->so_rcv.sb_hiwat) ?
1509 				    so->so_rcv.sb_hiwat : optval;
1510 				break;
1511 			}
1512 			break;
1513 		    }
1514 
1515 		case SO_SNDTIMEO:
1516 		case SO_RCVTIMEO:
1517 		    {
1518 			struct timeval *tv;
1519 			int val;
1520 
1521 			if (m == NULL || m->m_len < sizeof(*tv)) {
1522 				error = EINVAL;
1523 				goto bad;
1524 			}
1525 			tv = mtod(m, struct timeval *);
1526 			if (tv->tv_sec > (INT_MAX - tv->tv_usec / tick) / hz) {
1527 				error = EDOM;
1528 				goto bad;
1529 			}
1530 			val = tv->tv_sec * hz + tv->tv_usec / tick;
1531 			if (val == 0 && tv->tv_usec != 0)
1532 				val = 1;
1533 
1534 			switch (optname) {
1535 
1536 			case SO_SNDTIMEO:
1537 				so->so_snd.sb_timeo = val;
1538 				break;
1539 			case SO_RCVTIMEO:
1540 				so->so_rcv.sb_timeo = val;
1541 				break;
1542 			}
1543 			break;
1544 		    }
1545 
1546 		default:
1547 			error = ENOPROTOOPT;
1548 			break;
1549 		}
1550 		if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
1551 			(void) ((*so->so_proto->pr_ctloutput)
1552 				  (PRCO_SETOPT, so, level, optname, &m0));
1553 			m = NULL;	/* freed by protocol */
1554 		}
1555 	}
1556  bad:
1557 	if (m)
1558 		(void) m_free(m);
1559 	return (error);
1560 }
1561 
1562 int
1563 sogetopt(struct socket *so, int level, int optname, struct mbuf **mp)
1564 {
1565 	struct mbuf	*m;
1566 
1567 	if (level != SOL_SOCKET) {
1568 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1569 			return ((*so->so_proto->pr_ctloutput)
1570 				  (PRCO_GETOPT, so, level, optname, mp));
1571 		} else
1572 			return (ENOPROTOOPT);
1573 	} else {
1574 		m = m_get(M_WAIT, MT_SOOPTS);
1575 		m->m_len = sizeof(int);
1576 
1577 		switch (optname) {
1578 
1579 		case SO_LINGER:
1580 			m->m_len = sizeof(struct linger);
1581 			mtod(m, struct linger *)->l_onoff =
1582 				so->so_options & SO_LINGER;
1583 			mtod(m, struct linger *)->l_linger = so->so_linger;
1584 			break;
1585 
1586 		case SO_USELOOPBACK:
1587 		case SO_DONTROUTE:
1588 		case SO_DEBUG:
1589 		case SO_KEEPALIVE:
1590 		case SO_REUSEADDR:
1591 		case SO_REUSEPORT:
1592 		case SO_BROADCAST:
1593 		case SO_OOBINLINE:
1594 		case SO_TIMESTAMP:
1595 			*mtod(m, int *) = so->so_options & optname;
1596 			break;
1597 
1598 		case SO_TYPE:
1599 			*mtod(m, int *) = so->so_type;
1600 			break;
1601 
1602 		case SO_ERROR:
1603 			*mtod(m, int *) = so->so_error;
1604 			so->so_error = 0;
1605 			break;
1606 
1607 		case SO_SNDBUF:
1608 			*mtod(m, int *) = so->so_snd.sb_hiwat;
1609 			break;
1610 
1611 		case SO_RCVBUF:
1612 			*mtod(m, int *) = so->so_rcv.sb_hiwat;
1613 			break;
1614 
1615 		case SO_SNDLOWAT:
1616 			*mtod(m, int *) = so->so_snd.sb_lowat;
1617 			break;
1618 
1619 		case SO_RCVLOWAT:
1620 			*mtod(m, int *) = so->so_rcv.sb_lowat;
1621 			break;
1622 
1623 		case SO_SNDTIMEO:
1624 		case SO_RCVTIMEO:
1625 		    {
1626 			int val = (optname == SO_SNDTIMEO ?
1627 			     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1628 
1629 			m->m_len = sizeof(struct timeval);
1630 			mtod(m, struct timeval *)->tv_sec = val / hz;
1631 			mtod(m, struct timeval *)->tv_usec =
1632 			    (val % hz) * tick;
1633 			break;
1634 		    }
1635 
1636 		case SO_OVERFLOWED:
1637 			*mtod(m, int *) = so->so_rcv.sb_overflowed;
1638 			break;
1639 
1640 		default:
1641 			(void)m_free(m);
1642 			return (ENOPROTOOPT);
1643 		}
1644 		*mp = m;
1645 		return (0);
1646 	}
1647 }
1648 
1649 void
1650 sohasoutofband(struct socket *so)
1651 {
1652 	fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
1653 	selwakeup(&so->so_rcv.sb_sel);
1654 }
1655 
1656 static void
1657 filt_sordetach(struct knote *kn)
1658 {
1659 	struct socket	*so;
1660 
1661 	so = (struct socket *)kn->kn_fp->f_data;
1662 	SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
1663 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
1664 		so->so_rcv.sb_flags &= ~SB_KNOTE;
1665 }
1666 
1667 /*ARGSUSED*/
1668 static int
1669 filt_soread(struct knote *kn, long hint)
1670 {
1671 	struct socket	*so;
1672 
1673 	so = (struct socket *)kn->kn_fp->f_data;
1674 	kn->kn_data = so->so_rcv.sb_cc;
1675 	if (so->so_state & SS_CANTRCVMORE) {
1676 		kn->kn_flags |= EV_EOF;
1677 		kn->kn_fflags = so->so_error;
1678 		return (1);
1679 	}
1680 	if (so->so_error)	/* temporary udp error */
1681 		return (1);
1682 	if (kn->kn_sfflags & NOTE_LOWAT)
1683 		return (kn->kn_data >= kn->kn_sdata);
1684 	return (kn->kn_data >= so->so_rcv.sb_lowat);
1685 }
1686 
1687 static void
1688 filt_sowdetach(struct knote *kn)
1689 {
1690 	struct socket	*so;
1691 
1692 	so = (struct socket *)kn->kn_fp->f_data;
1693 	SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
1694 	if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
1695 		so->so_snd.sb_flags &= ~SB_KNOTE;
1696 }
1697 
1698 /*ARGSUSED*/
1699 static int
1700 filt_sowrite(struct knote *kn, long hint)
1701 {
1702 	struct socket	*so;
1703 
1704 	so = (struct socket *)kn->kn_fp->f_data;
1705 	kn->kn_data = sbspace(&so->so_snd);
1706 	if (so->so_state & SS_CANTSENDMORE) {
1707 		kn->kn_flags |= EV_EOF;
1708 		kn->kn_fflags = so->so_error;
1709 		return (1);
1710 	}
1711 	if (so->so_error)	/* temporary udp error */
1712 		return (1);
1713 	if (((so->so_state & SS_ISCONNECTED) == 0) &&
1714 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
1715 		return (0);
1716 	if (kn->kn_sfflags & NOTE_LOWAT)
1717 		return (kn->kn_data >= kn->kn_sdata);
1718 	return (kn->kn_data >= so->so_snd.sb_lowat);
1719 }
1720 
1721 /*ARGSUSED*/
1722 static int
1723 filt_solisten(struct knote *kn, long hint)
1724 {
1725 	struct socket	*so;
1726 
1727 	so = (struct socket *)kn->kn_fp->f_data;
1728 
1729 	/*
1730 	 * Set kn_data to number of incoming connections, not
1731 	 * counting partial (incomplete) connections.
1732 	 */
1733 	kn->kn_data = so->so_qlen;
1734 	return (kn->kn_data > 0);
1735 }
1736 
1737 static const struct filterops solisten_filtops =
1738 	{ 1, NULL, filt_sordetach, filt_solisten };
1739 static const struct filterops soread_filtops =
1740 	{ 1, NULL, filt_sordetach, filt_soread };
1741 static const struct filterops sowrite_filtops =
1742 	{ 1, NULL, filt_sowdetach, filt_sowrite };
1743 
1744 int
1745 soo_kqfilter(struct file *fp, struct knote *kn)
1746 {
1747 	struct socket	*so;
1748 	struct sockbuf	*sb;
1749 
1750 	so = (struct socket *)kn->kn_fp->f_data;
1751 	switch (kn->kn_filter) {
1752 	case EVFILT_READ:
1753 		if (so->so_options & SO_ACCEPTCONN)
1754 			kn->kn_fop = &solisten_filtops;
1755 		else
1756 			kn->kn_fop = &soread_filtops;
1757 		sb = &so->so_rcv;
1758 		break;
1759 	case EVFILT_WRITE:
1760 		kn->kn_fop = &sowrite_filtops;
1761 		sb = &so->so_snd;
1762 		break;
1763 	default:
1764 		return (1);
1765 	}
1766 	SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
1767 	sb->sb_flags |= SB_KNOTE;
1768 	return (0);
1769 }
1770 
1771 #include <sys/sysctl.h>
1772 
1773 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO);
1774 
1775 /*
1776  * sysctl helper routine for kern.somaxkva.  ensures that the given
1777  * value is not too small.
1778  * (XXX should we maybe make sure it's not too large as well?)
1779  */
1780 static int
1781 sysctl_kern_somaxkva(SYSCTLFN_ARGS)
1782 {
1783 	int error, new_somaxkva;
1784 	struct sysctlnode node;
1785 	int s;
1786 
1787 	new_somaxkva = somaxkva;
1788 	node = *rnode;
1789 	node.sysctl_data = &new_somaxkva;
1790 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
1791 	if (error || newp == NULL)
1792 		return (error);
1793 
1794 	if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */
1795 		return (EINVAL);
1796 
1797 	s = splvm();
1798 	simple_lock(&so_pendfree_slock);
1799 	somaxkva = new_somaxkva;
1800 	wakeup(&socurkva);
1801 	simple_unlock(&so_pendfree_slock);
1802 	splx(s);
1803 
1804 	return (error);
1805 }
1806 
1807 SYSCTL_SETUP(sysctl_kern_somaxkva_setup, "sysctl kern.somaxkva setup")
1808 {
1809 
1810 	sysctl_createv(clog, 0, NULL, NULL,
1811 		       CTLFLAG_PERMANENT,
1812 		       CTLTYPE_NODE, "kern", NULL,
1813 		       NULL, 0, NULL, 0,
1814 		       CTL_KERN, CTL_EOL);
1815 
1816 	sysctl_createv(clog, 0, NULL, NULL,
1817 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1818 		       CTLTYPE_INT, "somaxkva",
1819 		       SYSCTL_DESCR("Maximum amount of kernel memory to be "
1820 				    "used for socket buffers"),
1821 		       sysctl_kern_somaxkva, 0, NULL, 0,
1822 		       CTL_KERN, KERN_SOMAXKVA, CTL_EOL);
1823 }
1824