xref: /netbsd-src/sys/kern/uipc_socket.c (revision b78992537496bc71ee3d761f9fe0be0fc0a9a001)
1 /*	$NetBSD: uipc_socket.c,v 1.171 2008/08/06 15:01:23 plunky Exp $	*/
2 
3 /*-
4  * Copyright (c) 2002, 2007, 2008 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  *
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 #include <sys/cdefs.h>
66 __KERNEL_RCSID(0, "$NetBSD: uipc_socket.c,v 1.171 2008/08/06 15:01:23 plunky Exp $");
67 
68 #include "opt_inet.h"
69 #include "opt_sock_counters.h"
70 #include "opt_sosend_loan.h"
71 #include "opt_mbuftrace.h"
72 #include "opt_somaxkva.h"
73 #include "opt_multiprocessor.h"	/* XXX */
74 
75 #include <sys/param.h>
76 #include <sys/systm.h>
77 #include <sys/proc.h>
78 #include <sys/file.h>
79 #include <sys/filedesc.h>
80 #include <sys/malloc.h>
81 #include <sys/mbuf.h>
82 #include <sys/domain.h>
83 #include <sys/kernel.h>
84 #include <sys/protosw.h>
85 #include <sys/socket.h>
86 #include <sys/socketvar.h>
87 #include <sys/signalvar.h>
88 #include <sys/resourcevar.h>
89 #include <sys/event.h>
90 #include <sys/poll.h>
91 #include <sys/kauth.h>
92 #include <sys/mutex.h>
93 #include <sys/condvar.h>
94 
95 #include <uvm/uvm.h>
96 
97 MALLOC_DEFINE(M_SOOPTS, "soopts", "socket options");
98 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
99 
100 extern const struct fileops socketops;
101 
102 extern int	somaxconn;			/* patchable (XXX sysctl) */
103 int		somaxconn = SOMAXCONN;
104 kmutex_t	*softnet_lock;
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 #if defined(SOSEND_NO_LOAN) || defined(MULTIPROCESSOR)
133 int sock_loan_thresh = -1;
134 #else
135 int sock_loan_thresh = 4096;
136 #endif
137 
138 static kmutex_t so_pendfree_lock;
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 kcondvar_t socurkva_cv;
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 error;
157 
158 	mutex_enter(&so_pendfree_lock);
159 	while (socurkva + len > somaxkva) {
160 		size_t freed;
161 
162 		/*
163 		 * try to do pendfree.
164 		 */
165 
166 		freed = sodopendfreel();
167 
168 		/*
169 		 * if some kva was freed, try again.
170 		 */
171 
172 		if (freed)
173 			continue;
174 
175 		SOSEND_COUNTER_INCR(&sosend_kvalimit);
176 		error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock);
177 		if (error) {
178 			len = 0;
179 			break;
180 		}
181 	}
182 	socurkva += len;
183 	mutex_exit(&so_pendfree_lock);
184 	return len;
185 }
186 
187 static void
188 sokvaunreserve(vsize_t len)
189 {
190 
191 	mutex_enter(&so_pendfree_lock);
192 	socurkva -= len;
193 	cv_broadcast(&socurkva_cv);
194 	mutex_exit(&so_pendfree_lock);
195 }
196 
197 /*
198  * sokvaalloc: allocate kva for loan.
199  */
200 
201 vaddr_t
202 sokvaalloc(vsize_t len, struct socket *so)
203 {
204 	vaddr_t lva;
205 
206 	/*
207 	 * reserve kva.
208 	 */
209 
210 	if (sokvareserve(so, len) == 0)
211 		return 0;
212 
213 	/*
214 	 * allocate kva.
215 	 */
216 
217 	lva = uvm_km_alloc(kernel_map, len, 0, UVM_KMF_VAONLY | UVM_KMF_WAITVA);
218 	if (lva == 0) {
219 		sokvaunreserve(len);
220 		return (0);
221 	}
222 
223 	return lva;
224 }
225 
226 /*
227  * sokvafree: free kva for loan.
228  */
229 
230 void
231 sokvafree(vaddr_t sva, vsize_t len)
232 {
233 
234 	/*
235 	 * free kva.
236 	 */
237 
238 	uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY);
239 
240 	/*
241 	 * unreserve kva.
242 	 */
243 
244 	sokvaunreserve(len);
245 }
246 
247 static void
248 sodoloanfree(struct vm_page **pgs, void *buf, size_t size)
249 {
250 	vaddr_t sva, eva;
251 	vsize_t len;
252 	int npgs;
253 
254 	KASSERT(pgs != NULL);
255 
256 	eva = round_page((vaddr_t) buf + size);
257 	sva = trunc_page((vaddr_t) buf);
258 	len = eva - sva;
259 	npgs = len >> PAGE_SHIFT;
260 
261 	pmap_kremove(sva, len);
262 	pmap_update(pmap_kernel());
263 	uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE);
264 	sokvafree(sva, len);
265 }
266 
267 static size_t
268 sodopendfree(void)
269 {
270 	size_t rv;
271 
272 	if (__predict_true(so_pendfree == NULL))
273 		return 0;
274 
275 	mutex_enter(&so_pendfree_lock);
276 	rv = sodopendfreel();
277 	mutex_exit(&so_pendfree_lock);
278 
279 	return rv;
280 }
281 
282 /*
283  * sodopendfreel: free mbufs on "pendfree" list.
284  * unlock and relock so_pendfree_lock when freeing mbufs.
285  *
286  * => called with so_pendfree_lock held.
287  */
288 
289 static size_t
290 sodopendfreel(void)
291 {
292 	struct mbuf *m, *next;
293 	size_t rv = 0;
294 
295 	KASSERT(mutex_owned(&so_pendfree_lock));
296 
297 	while (so_pendfree != NULL) {
298 		m = so_pendfree;
299 		so_pendfree = NULL;
300 		mutex_exit(&so_pendfree_lock);
301 
302 		for (; m != NULL; m = next) {
303 			next = m->m_next;
304 			KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0);
305 			KASSERT(m->m_ext.ext_refcnt == 0);
306 
307 			rv += m->m_ext.ext_size;
308 			sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf,
309 			    m->m_ext.ext_size);
310 			pool_cache_put(mb_cache, m);
311 		}
312 
313 		mutex_enter(&so_pendfree_lock);
314 	}
315 
316 	return (rv);
317 }
318 
319 void
320 soloanfree(struct mbuf *m, void *buf, size_t size, void *arg)
321 {
322 
323 	KASSERT(m != NULL);
324 
325 	/*
326 	 * postpone freeing mbuf.
327 	 *
328 	 * we can't do it in interrupt context
329 	 * because we need to put kva back to kernel_map.
330 	 */
331 
332 	mutex_enter(&so_pendfree_lock);
333 	m->m_next = so_pendfree;
334 	so_pendfree = m;
335 	cv_broadcast(&socurkva_cv);
336 	mutex_exit(&so_pendfree_lock);
337 }
338 
339 static long
340 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space)
341 {
342 	struct iovec *iov = uio->uio_iov;
343 	vaddr_t sva, eva;
344 	vsize_t len;
345 	vaddr_t lva;
346 	int npgs, error;
347 	vaddr_t va;
348 	int i;
349 
350 	if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace))
351 		return (0);
352 
353 	if (iov->iov_len < (size_t) space)
354 		space = iov->iov_len;
355 	if (space > SOCK_LOAN_CHUNK)
356 		space = SOCK_LOAN_CHUNK;
357 
358 	eva = round_page((vaddr_t) iov->iov_base + space);
359 	sva = trunc_page((vaddr_t) iov->iov_base);
360 	len = eva - sva;
361 	npgs = len >> PAGE_SHIFT;
362 
363 	KASSERT(npgs <= M_EXT_MAXPAGES);
364 
365 	lva = sokvaalloc(len, so);
366 	if (lva == 0)
367 		return 0;
368 
369 	error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len,
370 	    m->m_ext.ext_pgs, UVM_LOAN_TOPAGE);
371 	if (error) {
372 		sokvafree(lva, len);
373 		return (0);
374 	}
375 
376 	for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE)
377 		pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
378 		    VM_PROT_READ);
379 	pmap_update(pmap_kernel());
380 
381 	lva += (vaddr_t) iov->iov_base & PAGE_MASK;
382 
383 	MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so);
384 	m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
385 
386 	uio->uio_resid -= space;
387 	/* uio_offset not updated, not set/used for write(2) */
388 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space;
389 	uio->uio_iov->iov_len -= space;
390 	if (uio->uio_iov->iov_len == 0) {
391 		uio->uio_iov++;
392 		uio->uio_iovcnt--;
393 	}
394 
395 	return (space);
396 }
397 
398 static int
399 sokva_reclaim_callback(struct callback_entry *ce, void *obj, void *arg)
400 {
401 
402 	KASSERT(ce == &sokva_reclaimerentry);
403 	KASSERT(obj == NULL);
404 
405 	sodopendfree();
406 	if (!vm_map_starved_p(kernel_map)) {
407 		return CALLBACK_CHAIN_ABORT;
408 	}
409 	return CALLBACK_CHAIN_CONTINUE;
410 }
411 
412 struct mbuf *
413 getsombuf(struct socket *so, int type)
414 {
415 	struct mbuf *m;
416 
417 	m = m_get(M_WAIT, type);
418 	MCLAIM(m, so->so_mowner);
419 	return m;
420 }
421 
422 void
423 soinit(void)
424 {
425 
426 	mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM);
427 	softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
428 	cv_init(&socurkva_cv, "sokva");
429 	soinit2();
430 
431 	/* Set the initial adjusted socket buffer size. */
432 	if (sb_max_set(sb_max))
433 		panic("bad initial sb_max value: %lu", sb_max);
434 
435 	callback_register(&vm_map_to_kernel(kernel_map)->vmk_reclaim_callback,
436 	    &sokva_reclaimerentry, NULL, sokva_reclaim_callback);
437 }
438 
439 /*
440  * Socket operation routines.
441  * These routines are called by the routines in
442  * sys_socket.c or from a system process, and
443  * implement the semantics of socket operations by
444  * switching out to the protocol specific routines.
445  */
446 /*ARGSUSED*/
447 int
448 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l,
449 	 struct socket *lockso)
450 {
451 	const struct protosw	*prp;
452 	struct socket	*so;
453 	uid_t		uid;
454 	int		error;
455 	kmutex_t	*lock;
456 
457 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET,
458 	    KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type),
459 	    KAUTH_ARG(proto));
460 	if (error != 0)
461 		return error;
462 
463 	if (proto)
464 		prp = pffindproto(dom, proto, type);
465 	else
466 		prp = pffindtype(dom, type);
467 	if (prp == NULL) {
468 		/* no support for domain */
469 		if (pffinddomain(dom) == 0)
470 			return EAFNOSUPPORT;
471 		/* no support for socket type */
472 		if (proto == 0 && type != 0)
473 			return EPROTOTYPE;
474 		return EPROTONOSUPPORT;
475 	}
476 	if (prp->pr_usrreq == NULL)
477 		return EPROTONOSUPPORT;
478 	if (prp->pr_type != type)
479 		return EPROTOTYPE;
480 
481 	so = soget(true);
482 	so->so_type = type;
483 	so->so_proto = prp;
484 	so->so_send = sosend;
485 	so->so_receive = soreceive;
486 #ifdef MBUFTRACE
487 	so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner;
488 	so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner;
489 	so->so_mowner = &prp->pr_domain->dom_mowner;
490 #endif
491 	uid = kauth_cred_geteuid(l->l_cred);
492 	so->so_uidinfo = uid_find(uid);
493 	so->so_egid = kauth_cred_getegid(l->l_cred);
494 	so->so_cpid = l->l_proc->p_pid;
495 	if (lockso != NULL) {
496 		/* Caller wants us to share a lock. */
497 		lock = lockso->so_lock;
498 		so->so_lock = lock;
499 		mutex_obj_hold(lock);
500 		mutex_enter(lock);
501 	} else {
502 		/* Lock assigned and taken during PRU_ATTACH. */
503 	}
504 	error = (*prp->pr_usrreq)(so, PRU_ATTACH, NULL,
505 	    (struct mbuf *)(long)proto, NULL, l);
506 	KASSERT(solocked(so));
507 	if (error != 0) {
508 		so->so_state |= SS_NOFDREF;
509 		sofree(so);
510 		return error;
511 	}
512 	sounlock(so);
513 	*aso = so;
514 	return 0;
515 }
516 
517 /* On success, write file descriptor to fdout and return zero.  On
518  * failure, return non-zero; *fdout will be undefined.
519  */
520 int
521 fsocreate(int domain, struct socket **sop, int type, int protocol,
522     struct lwp *l, int *fdout)
523 {
524 	struct socket	*so;
525 	struct file	*fp;
526 	int		fd, error;
527 
528 	if ((error = fd_allocfile(&fp, &fd)) != 0)
529 		return (error);
530 	fp->f_flag = FREAD|FWRITE;
531 	fp->f_type = DTYPE_SOCKET;
532 	fp->f_ops = &socketops;
533 	error = socreate(domain, &so, type, protocol, l, NULL);
534 	if (error != 0) {
535 		fd_abort(curproc, fp, fd);
536 	} else {
537 		if (sop != NULL)
538 			*sop = so;
539 		fp->f_data = so;
540 		fd_affix(curproc, fp, fd);
541 		*fdout = fd;
542 	}
543 	return error;
544 }
545 
546 int
547 sobind(struct socket *so, struct mbuf *nam, struct lwp *l)
548 {
549 	int	error;
550 
551 	solock(so);
552 	error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, NULL, nam, NULL, l);
553 	sounlock(so);
554 	return error;
555 }
556 
557 int
558 solisten(struct socket *so, int backlog, struct lwp *l)
559 {
560 	int	error;
561 
562 	solock(so);
563 	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
564 	    SS_ISDISCONNECTING)) != 0) {
565 	    	sounlock(so);
566 		return (EOPNOTSUPP);
567 	}
568 	error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, NULL,
569 	    NULL, NULL, l);
570 	if (error != 0) {
571 		sounlock(so);
572 		return error;
573 	}
574 	if (TAILQ_EMPTY(&so->so_q))
575 		so->so_options |= SO_ACCEPTCONN;
576 	if (backlog < 0)
577 		backlog = 0;
578 	so->so_qlimit = min(backlog, somaxconn);
579 	sounlock(so);
580 	return 0;
581 }
582 
583 void
584 sofree(struct socket *so)
585 {
586 	u_int refs;
587 
588 	KASSERT(solocked(so));
589 
590 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
591 		sounlock(so);
592 		return;
593 	}
594 	if (so->so_head) {
595 		/*
596 		 * We must not decommission a socket that's on the accept(2)
597 		 * queue.  If we do, then accept(2) may hang after select(2)
598 		 * indicated that the listening socket was ready.
599 		 */
600 		if (!soqremque(so, 0)) {
601 			sounlock(so);
602 			return;
603 		}
604 	}
605 	if (so->so_rcv.sb_hiwat)
606 		(void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0,
607 		    RLIM_INFINITY);
608 	if (so->so_snd.sb_hiwat)
609 		(void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0,
610 		    RLIM_INFINITY);
611 	sbrelease(&so->so_snd, so);
612 	KASSERT(!cv_has_waiters(&so->so_cv));
613 	KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv));
614 	KASSERT(!cv_has_waiters(&so->so_snd.sb_cv));
615 	sorflush(so);
616 	refs = so->so_aborting;	/* XXX */
617 #ifdef INET
618 	/* remove acccept filter if one is present. */
619 	if (so->so_accf != NULL)
620 		do_setopt_accept_filter(so, NULL);
621 #endif
622 	sounlock(so);
623 	if (refs == 0)		/* XXX */
624 		soput(so);
625 }
626 
627 /*
628  * Close a socket on last file table reference removal.
629  * Initiate disconnect if connected.
630  * Free socket when disconnect complete.
631  */
632 int
633 soclose(struct socket *so)
634 {
635 	struct socket	*so2;
636 	int		error;
637 	int		error2;
638 
639 	error = 0;
640 	solock(so);
641 	if (so->so_options & SO_ACCEPTCONN) {
642 		do {
643 			while ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) {
644 				KASSERT(solocked2(so, so2));
645 				(void) soqremque(so2, 0);
646 				/* soabort drops the lock. */
647 				(void) soabort(so2);
648 				solock(so);
649 			}
650 			while ((so2 = TAILQ_FIRST(&so->so_q)) != 0) {
651 				KASSERT(solocked2(so, so2));
652 				(void) soqremque(so2, 1);
653 				/* soabort drops the lock. */
654 				(void) soabort(so2);
655 				solock(so);
656 			}
657 		} while (!TAILQ_EMPTY(&so->so_q0));
658 	}
659 	if (so->so_pcb == 0)
660 		goto discard;
661 	if (so->so_state & SS_ISCONNECTED) {
662 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
663 			error = sodisconnect(so);
664 			if (error)
665 				goto drop;
666 		}
667 		if (so->so_options & SO_LINGER) {
668 			if ((so->so_state & SS_ISDISCONNECTING) && so->so_nbio)
669 				goto drop;
670 			while (so->so_state & SS_ISCONNECTED) {
671 				error = sowait(so, so->so_linger * hz);
672 				if (error)
673 					break;
674 			}
675 		}
676 	}
677  drop:
678 	if (so->so_pcb) {
679 		error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH,
680 		    NULL, NULL, NULL, NULL);
681 		if (error == 0)
682 			error = error2;
683 	}
684  discard:
685 	if (so->so_state & SS_NOFDREF)
686 		panic("soclose: NOFDREF");
687 	so->so_state |= SS_NOFDREF;
688 	sofree(so);
689 	return (error);
690 }
691 
692 /*
693  * Must be called with the socket locked..  Will return with it unlocked.
694  */
695 int
696 soabort(struct socket *so)
697 {
698 	u_int refs;
699 	int error;
700 
701 	KASSERT(solocked(so));
702 	KASSERT(so->so_head == NULL);
703 
704 	so->so_aborting++;		/* XXX */
705 	error = (*so->so_proto->pr_usrreq)(so, PRU_ABORT, NULL,
706 	    NULL, NULL, NULL);
707 	refs = --so->so_aborting;	/* XXX */
708 	if (error || (refs == 0)) {
709 		sofree(so);
710 	} else {
711 		sounlock(so);
712 	}
713 	return error;
714 }
715 
716 int
717 soaccept(struct socket *so, struct mbuf *nam)
718 {
719 	int	error;
720 
721 	KASSERT(solocked(so));
722 
723 	error = 0;
724 	if ((so->so_state & SS_NOFDREF) == 0)
725 		panic("soaccept: !NOFDREF");
726 	so->so_state &= ~SS_NOFDREF;
727 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
728 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
729 		error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT,
730 		    NULL, nam, NULL, NULL);
731 	else
732 		error = ECONNABORTED;
733 
734 	return (error);
735 }
736 
737 int
738 soconnect(struct socket *so, struct mbuf *nam, struct lwp *l)
739 {
740 	int		error;
741 
742 	KASSERT(solocked(so));
743 
744 	if (so->so_options & SO_ACCEPTCONN)
745 		return (EOPNOTSUPP);
746 	/*
747 	 * If protocol is connection-based, can only connect once.
748 	 * Otherwise, if connected, try to disconnect first.
749 	 * This allows user to disconnect by connecting to, e.g.,
750 	 * a null address.
751 	 */
752 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
753 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
754 	    (error = sodisconnect(so))))
755 		error = EISCONN;
756 	else
757 		error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT,
758 		    NULL, nam, NULL, l);
759 	return (error);
760 }
761 
762 int
763 soconnect2(struct socket *so1, struct socket *so2)
764 {
765 	int	error;
766 
767 	KASSERT(solocked2(so1, so2));
768 
769 	error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2,
770 	    NULL, (struct mbuf *)so2, NULL, NULL);
771 	return (error);
772 }
773 
774 int
775 sodisconnect(struct socket *so)
776 {
777 	int	error;
778 
779 	KASSERT(solocked(so));
780 
781 	if ((so->so_state & SS_ISCONNECTED) == 0) {
782 		error = ENOTCONN;
783 	} else if (so->so_state & SS_ISDISCONNECTING) {
784 		error = EALREADY;
785 	} else {
786 		error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT,
787 		    NULL, NULL, NULL, NULL);
788 	}
789 	sodopendfree();
790 	return (error);
791 }
792 
793 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
794 /*
795  * Send on a socket.
796  * If send must go all at once and message is larger than
797  * send buffering, then hard error.
798  * Lock against other senders.
799  * If must go all at once and not enough room now, then
800  * inform user that this would block and do nothing.
801  * Otherwise, if nonblocking, send as much as possible.
802  * The data to be sent is described by "uio" if nonzero,
803  * otherwise by the mbuf chain "top" (which must be null
804  * if uio is not).  Data provided in mbuf chain must be small
805  * enough to send all at once.
806  *
807  * Returns nonzero on error, timeout or signal; callers
808  * must check for short counts if EINTR/ERESTART are returned.
809  * Data and control buffers are freed on return.
810  */
811 int
812 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
813 	struct mbuf *control, int flags, struct lwp *l)
814 {
815 	struct mbuf	**mp, *m;
816 	struct proc	*p;
817 	long		space, len, resid, clen, mlen;
818 	int		error, s, dontroute, atomic;
819 
820 	p = l->l_proc;
821 	sodopendfree();
822 	clen = 0;
823 
824 	/*
825 	 * solock() provides atomicity of access.  splsoftnet() prevents
826 	 * protocol processing soft interrupts from interrupting us and
827 	 * blocking (expensive).
828 	 */
829 	s = splsoftnet();
830 	solock(so);
831 	atomic = sosendallatonce(so) || top;
832 	if (uio)
833 		resid = uio->uio_resid;
834 	else
835 		resid = top->m_pkthdr.len;
836 	/*
837 	 * In theory resid should be unsigned.
838 	 * However, space must be signed, as it might be less than 0
839 	 * if we over-committed, and we must use a signed comparison
840 	 * of space and resid.  On the other hand, a negative resid
841 	 * causes us to loop sending 0-length segments to the protocol.
842 	 */
843 	if (resid < 0) {
844 		error = EINVAL;
845 		goto out;
846 	}
847 	dontroute =
848 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
849 	    (so->so_proto->pr_flags & PR_ATOMIC);
850 	l->l_ru.ru_msgsnd++;
851 	if (control)
852 		clen = control->m_len;
853  restart:
854 	if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
855 		goto out;
856 	do {
857 		if (so->so_state & SS_CANTSENDMORE) {
858 			error = EPIPE;
859 			goto release;
860 		}
861 		if (so->so_error) {
862 			error = so->so_error;
863 			so->so_error = 0;
864 			goto release;
865 		}
866 		if ((so->so_state & SS_ISCONNECTED) == 0) {
867 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
868 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
869 				    !(resid == 0 && clen != 0)) {
870 					error = ENOTCONN;
871 					goto release;
872 				}
873 			} else if (addr == 0) {
874 				error = EDESTADDRREQ;
875 				goto release;
876 			}
877 		}
878 		space = sbspace(&so->so_snd);
879 		if (flags & MSG_OOB)
880 			space += 1024;
881 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
882 		    clen > so->so_snd.sb_hiwat) {
883 			error = EMSGSIZE;
884 			goto release;
885 		}
886 		if (space < resid + clen &&
887 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
888 			if (so->so_nbio) {
889 				error = EWOULDBLOCK;
890 				goto release;
891 			}
892 			sbunlock(&so->so_snd);
893 			error = sbwait(&so->so_snd);
894 			if (error)
895 				goto out;
896 			goto restart;
897 		}
898 		mp = &top;
899 		space -= clen;
900 		do {
901 			if (uio == NULL) {
902 				/*
903 				 * Data is prepackaged in "top".
904 				 */
905 				resid = 0;
906 				if (flags & MSG_EOR)
907 					top->m_flags |= M_EOR;
908 			} else do {
909 				sounlock(so);
910 				splx(s);
911 				if (top == NULL) {
912 					m = m_gethdr(M_WAIT, MT_DATA);
913 					mlen = MHLEN;
914 					m->m_pkthdr.len = 0;
915 					m->m_pkthdr.rcvif = NULL;
916 				} else {
917 					m = m_get(M_WAIT, MT_DATA);
918 					mlen = MLEN;
919 				}
920 				MCLAIM(m, so->so_snd.sb_mowner);
921 				if (sock_loan_thresh >= 0 &&
922 				    uio->uio_iov->iov_len >= sock_loan_thresh &&
923 				    space >= sock_loan_thresh &&
924 				    (len = sosend_loan(so, uio, m,
925 						       space)) != 0) {
926 					SOSEND_COUNTER_INCR(&sosend_loan_big);
927 					space -= len;
928 					goto have_data;
929 				}
930 				if (resid >= MINCLSIZE && space >= MCLBYTES) {
931 					SOSEND_COUNTER_INCR(&sosend_copy_big);
932 					m_clget(m, M_WAIT);
933 					if ((m->m_flags & M_EXT) == 0)
934 						goto nopages;
935 					mlen = MCLBYTES;
936 					if (atomic && top == 0) {
937 						len = lmin(MCLBYTES - max_hdr,
938 						    resid);
939 						m->m_data += max_hdr;
940 					} else
941 						len = lmin(MCLBYTES, resid);
942 					space -= len;
943 				} else {
944  nopages:
945 					SOSEND_COUNTER_INCR(&sosend_copy_small);
946 					len = lmin(lmin(mlen, resid), space);
947 					space -= len;
948 					/*
949 					 * For datagram protocols, leave room
950 					 * for protocol headers in first mbuf.
951 					 */
952 					if (atomic && top == 0 && len < mlen)
953 						MH_ALIGN(m, len);
954 				}
955 				error = uiomove(mtod(m, void *), (int)len, uio);
956  have_data:
957 				resid = uio->uio_resid;
958 				m->m_len = len;
959 				*mp = m;
960 				top->m_pkthdr.len += len;
961 				s = splsoftnet();
962 				solock(so);
963 				if (error != 0)
964 					goto release;
965 				mp = &m->m_next;
966 				if (resid <= 0) {
967 					if (flags & MSG_EOR)
968 						top->m_flags |= M_EOR;
969 					break;
970 				}
971 			} while (space > 0 && atomic);
972 
973 			if (so->so_state & SS_CANTSENDMORE) {
974 				error = EPIPE;
975 				goto release;
976 			}
977 			if (dontroute)
978 				so->so_options |= SO_DONTROUTE;
979 			if (resid > 0)
980 				so->so_state |= SS_MORETOCOME;
981 			error = (*so->so_proto->pr_usrreq)(so,
982 			    (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND,
983 			    top, addr, control, curlwp);
984 			if (dontroute)
985 				so->so_options &= ~SO_DONTROUTE;
986 			if (resid > 0)
987 				so->so_state &= ~SS_MORETOCOME;
988 			clen = 0;
989 			control = NULL;
990 			top = NULL;
991 			mp = &top;
992 			if (error != 0)
993 				goto release;
994 		} while (resid && space > 0);
995 	} while (resid);
996 
997  release:
998 	sbunlock(&so->so_snd);
999  out:
1000 	sounlock(so);
1001 	splx(s);
1002 	if (top)
1003 		m_freem(top);
1004 	if (control)
1005 		m_freem(control);
1006 	return (error);
1007 }
1008 
1009 /*
1010  * Following replacement or removal of the first mbuf on the first
1011  * mbuf chain of a socket buffer, push necessary state changes back
1012  * into the socket buffer so that other consumers see the values
1013  * consistently.  'nextrecord' is the callers locally stored value of
1014  * the original value of sb->sb_mb->m_nextpkt which must be restored
1015  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
1016  */
1017 static void
1018 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
1019 {
1020 
1021 	KASSERT(solocked(sb->sb_so));
1022 
1023 	/*
1024 	 * First, update for the new value of nextrecord.  If necessary,
1025 	 * make it the first record.
1026 	 */
1027 	if (sb->sb_mb != NULL)
1028 		sb->sb_mb->m_nextpkt = nextrecord;
1029 	else
1030 		sb->sb_mb = nextrecord;
1031 
1032         /*
1033          * Now update any dependent socket buffer fields to reflect
1034          * the new state.  This is an inline of SB_EMPTY_FIXUP, with
1035          * the addition of a second clause that takes care of the
1036          * case where sb_mb has been updated, but remains the last
1037          * record.
1038          */
1039         if (sb->sb_mb == NULL) {
1040                 sb->sb_mbtail = NULL;
1041                 sb->sb_lastrecord = NULL;
1042         } else if (sb->sb_mb->m_nextpkt == NULL)
1043                 sb->sb_lastrecord = sb->sb_mb;
1044 }
1045 
1046 /*
1047  * Implement receive operations on a socket.
1048  * We depend on the way that records are added to the sockbuf
1049  * by sbappend*.  In particular, each record (mbufs linked through m_next)
1050  * must begin with an address if the protocol so specifies,
1051  * followed by an optional mbuf or mbufs containing ancillary data,
1052  * and then zero or more mbufs of data.
1053  * In order to avoid blocking network interrupts for the entire time here,
1054  * we splx() while doing the actual copy to user space.
1055  * Although the sockbuf is locked, new data may still be appended,
1056  * and thus we must maintain consistency of the sockbuf during that time.
1057  *
1058  * The caller may receive the data as a single mbuf chain by supplying
1059  * an mbuf **mp0 for use in returning the chain.  The uio is then used
1060  * only for the count in uio_resid.
1061  */
1062 int
1063 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
1064 	struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1065 {
1066 	struct lwp *l = curlwp;
1067 	struct mbuf	*m, **mp, *mt;
1068 	int atomic, flags, len, error, s, offset, moff, type, orig_resid;
1069 	const struct protosw	*pr;
1070 	struct mbuf	*nextrecord;
1071 	int		mbuf_removed = 0;
1072 	const struct domain *dom;
1073 
1074 	pr = so->so_proto;
1075 	atomic = pr->pr_flags & PR_ATOMIC;
1076 	dom = pr->pr_domain;
1077 	mp = mp0;
1078 	type = 0;
1079 	orig_resid = uio->uio_resid;
1080 
1081 	if (paddr != NULL)
1082 		*paddr = NULL;
1083 	if (controlp != NULL)
1084 		*controlp = NULL;
1085 	if (flagsp != NULL)
1086 		flags = *flagsp &~ MSG_EOR;
1087 	else
1088 		flags = 0;
1089 
1090 	if ((flags & MSG_DONTWAIT) == 0)
1091 		sodopendfree();
1092 
1093 	if (flags & MSG_OOB) {
1094 		m = m_get(M_WAIT, MT_DATA);
1095 		solock(so);
1096 		error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m,
1097 		    (struct mbuf *)(long)(flags & MSG_PEEK), NULL, l);
1098 		sounlock(so);
1099 		if (error)
1100 			goto bad;
1101 		do {
1102 			error = uiomove(mtod(m, void *),
1103 			    (int) min(uio->uio_resid, m->m_len), uio);
1104 			m = m_free(m);
1105 		} while (uio->uio_resid > 0 && error == 0 && m);
1106  bad:
1107 		if (m != NULL)
1108 			m_freem(m);
1109 		return error;
1110 	}
1111 	if (mp != NULL)
1112 		*mp = NULL;
1113 
1114 	/*
1115 	 * solock() provides atomicity of access.  splsoftnet() prevents
1116 	 * protocol processing soft interrupts from interrupting us and
1117 	 * blocking (expensive).
1118 	 */
1119 	s = splsoftnet();
1120 	solock(so);
1121 	if (so->so_state & SS_ISCONFIRMING && uio->uio_resid)
1122 		(*pr->pr_usrreq)(so, PRU_RCVD, NULL, NULL, NULL, l);
1123 
1124  restart:
1125 	if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) {
1126 		sounlock(so);
1127 		splx(s);
1128 		return error;
1129 	}
1130 
1131 	m = so->so_rcv.sb_mb;
1132 	/*
1133 	 * If we have less data than requested, block awaiting more
1134 	 * (subject to any timeout) if:
1135 	 *   1. the current count is less than the low water mark,
1136 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1137 	 *	receive operation at once if we block (resid <= hiwat), or
1138 	 *   3. MSG_DONTWAIT is not set.
1139 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1140 	 * we have to do the receive in sections, and thus risk returning
1141 	 * a short count if a timeout or signal occurs after we start.
1142 	 */
1143 	if (m == NULL ||
1144 	    ((flags & MSG_DONTWAIT) == 0 &&
1145 	     so->so_rcv.sb_cc < uio->uio_resid &&
1146 	     (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1147 	      ((flags & MSG_WAITALL) &&
1148 	       uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1149 	     m->m_nextpkt == NULL && !atomic)) {
1150 #ifdef DIAGNOSTIC
1151 		if (m == NULL && so->so_rcv.sb_cc)
1152 			panic("receive 1");
1153 #endif
1154 		if (so->so_error) {
1155 			if (m != NULL)
1156 				goto dontblock;
1157 			error = so->so_error;
1158 			if ((flags & MSG_PEEK) == 0)
1159 				so->so_error = 0;
1160 			goto release;
1161 		}
1162 		if (so->so_state & SS_CANTRCVMORE) {
1163 			if (m != NULL)
1164 				goto dontblock;
1165 			else
1166 				goto release;
1167 		}
1168 		for (; m != NULL; m = m->m_next)
1169 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1170 				m = so->so_rcv.sb_mb;
1171 				goto dontblock;
1172 			}
1173 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1174 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1175 			error = ENOTCONN;
1176 			goto release;
1177 		}
1178 		if (uio->uio_resid == 0)
1179 			goto release;
1180 		if (so->so_nbio || (flags & MSG_DONTWAIT)) {
1181 			error = EWOULDBLOCK;
1182 			goto release;
1183 		}
1184 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
1185 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
1186 		sbunlock(&so->so_rcv);
1187 		error = sbwait(&so->so_rcv);
1188 		if (error != 0) {
1189 			sounlock(so);
1190 			splx(s);
1191 			return error;
1192 		}
1193 		goto restart;
1194 	}
1195  dontblock:
1196 	/*
1197 	 * On entry here, m points to the first record of the socket buffer.
1198 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1199 	 * pointer to the next record in the socket buffer.  We must keep the
1200 	 * various socket buffer pointers and local stack versions of the
1201 	 * pointers in sync, pushing out modifications before dropping the
1202 	 * socket lock, and re-reading them when picking it up.
1203 	 *
1204 	 * Otherwise, we will race with the network stack appending new data
1205 	 * or records onto the socket buffer by using inconsistent/stale
1206 	 * versions of the field, possibly resulting in socket buffer
1207 	 * corruption.
1208 	 *
1209 	 * By holding the high-level sblock(), we prevent simultaneous
1210 	 * readers from pulling off the front of the socket buffer.
1211 	 */
1212 	if (l != NULL)
1213 		l->l_ru.ru_msgrcv++;
1214 	KASSERT(m == so->so_rcv.sb_mb);
1215 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
1216 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
1217 	nextrecord = m->m_nextpkt;
1218 	if (pr->pr_flags & PR_ADDR) {
1219 #ifdef DIAGNOSTIC
1220 		if (m->m_type != MT_SONAME)
1221 			panic("receive 1a");
1222 #endif
1223 		orig_resid = 0;
1224 		if (flags & MSG_PEEK) {
1225 			if (paddr)
1226 				*paddr = m_copy(m, 0, m->m_len);
1227 			m = m->m_next;
1228 		} else {
1229 			sbfree(&so->so_rcv, m);
1230 			mbuf_removed = 1;
1231 			if (paddr != NULL) {
1232 				*paddr = m;
1233 				so->so_rcv.sb_mb = m->m_next;
1234 				m->m_next = NULL;
1235 				m = so->so_rcv.sb_mb;
1236 			} else {
1237 				MFREE(m, so->so_rcv.sb_mb);
1238 				m = so->so_rcv.sb_mb;
1239 			}
1240 			sbsync(&so->so_rcv, nextrecord);
1241 		}
1242 	}
1243 
1244 	/*
1245 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1246 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1247 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1248 	 * perform externalization (or freeing if controlp == NULL).
1249 	 */
1250 	if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) {
1251 		struct mbuf *cm = NULL, *cmn;
1252 		struct mbuf **cme = &cm;
1253 
1254 		do {
1255 			if (flags & MSG_PEEK) {
1256 				if (controlp != NULL) {
1257 					*controlp = m_copy(m, 0, m->m_len);
1258 					controlp = &(*controlp)->m_next;
1259 				}
1260 				m = m->m_next;
1261 			} else {
1262 				sbfree(&so->so_rcv, m);
1263 				so->so_rcv.sb_mb = m->m_next;
1264 				m->m_next = NULL;
1265 				*cme = m;
1266 				cme = &(*cme)->m_next;
1267 				m = so->so_rcv.sb_mb;
1268 			}
1269 		} while (m != NULL && m->m_type == MT_CONTROL);
1270 		if ((flags & MSG_PEEK) == 0)
1271 			sbsync(&so->so_rcv, nextrecord);
1272 		for (; cm != NULL; cm = cmn) {
1273 			cmn = cm->m_next;
1274 			cm->m_next = NULL;
1275 			type = mtod(cm, struct cmsghdr *)->cmsg_type;
1276 			if (controlp != NULL) {
1277 				if (dom->dom_externalize != NULL &&
1278 				    type == SCM_RIGHTS) {
1279 					sounlock(so);
1280 					splx(s);
1281 					error = (*dom->dom_externalize)(cm, l);
1282 					s = splsoftnet();
1283 					solock(so);
1284 				}
1285 				*controlp = cm;
1286 				while (*controlp != NULL)
1287 					controlp = &(*controlp)->m_next;
1288 			} else {
1289 				/*
1290 				 * Dispose of any SCM_RIGHTS message that went
1291 				 * through the read path rather than recv.
1292 				 */
1293 				if (dom->dom_dispose != NULL &&
1294 				    type == SCM_RIGHTS) {
1295 				    	sounlock(so);
1296 					(*dom->dom_dispose)(cm);
1297 					solock(so);
1298 				}
1299 				m_freem(cm);
1300 			}
1301 		}
1302 		if (m != NULL)
1303 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1304 		else
1305 			nextrecord = so->so_rcv.sb_mb;
1306 		orig_resid = 0;
1307 	}
1308 
1309 	/* If m is non-NULL, we have some data to read. */
1310 	if (__predict_true(m != NULL)) {
1311 		type = m->m_type;
1312 		if (type == MT_OOBDATA)
1313 			flags |= MSG_OOB;
1314 	}
1315 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1316 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1317 
1318 	moff = 0;
1319 	offset = 0;
1320 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1321 		if (m->m_type == MT_OOBDATA) {
1322 			if (type != MT_OOBDATA)
1323 				break;
1324 		} else if (type == MT_OOBDATA)
1325 			break;
1326 #ifdef DIAGNOSTIC
1327 		else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
1328 			panic("receive 3");
1329 #endif
1330 		so->so_state &= ~SS_RCVATMARK;
1331 		len = uio->uio_resid;
1332 		if (so->so_oobmark && len > so->so_oobmark - offset)
1333 			len = so->so_oobmark - offset;
1334 		if (len > m->m_len - moff)
1335 			len = m->m_len - moff;
1336 		/*
1337 		 * If mp is set, just pass back the mbufs.
1338 		 * Otherwise copy them out via the uio, then free.
1339 		 * Sockbuf must be consistent here (points to current mbuf,
1340 		 * it points to next record) when we drop priority;
1341 		 * we must note any additions to the sockbuf when we
1342 		 * block interrupts again.
1343 		 */
1344 		if (mp == NULL) {
1345 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1346 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1347 			sounlock(so);
1348 			splx(s);
1349 			error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1350 			s = splsoftnet();
1351 			solock(so);
1352 			if (error != 0) {
1353 				/*
1354 				 * If any part of the record has been removed
1355 				 * (such as the MT_SONAME mbuf, which will
1356 				 * happen when PR_ADDR, and thus also
1357 				 * PR_ATOMIC, is set), then drop the entire
1358 				 * record to maintain the atomicity of the
1359 				 * receive operation.
1360 				 *
1361 				 * This avoids a later panic("receive 1a")
1362 				 * when compiled with DIAGNOSTIC.
1363 				 */
1364 				if (m && mbuf_removed && atomic)
1365 					(void) sbdroprecord(&so->so_rcv);
1366 
1367 				goto release;
1368 			}
1369 		} else
1370 			uio->uio_resid -= len;
1371 		if (len == m->m_len - moff) {
1372 			if (m->m_flags & M_EOR)
1373 				flags |= MSG_EOR;
1374 			if (flags & MSG_PEEK) {
1375 				m = m->m_next;
1376 				moff = 0;
1377 			} else {
1378 				nextrecord = m->m_nextpkt;
1379 				sbfree(&so->so_rcv, m);
1380 				if (mp) {
1381 					*mp = m;
1382 					mp = &m->m_next;
1383 					so->so_rcv.sb_mb = m = m->m_next;
1384 					*mp = NULL;
1385 				} else {
1386 					MFREE(m, so->so_rcv.sb_mb);
1387 					m = so->so_rcv.sb_mb;
1388 				}
1389 				/*
1390 				 * If m != NULL, we also know that
1391 				 * so->so_rcv.sb_mb != NULL.
1392 				 */
1393 				KASSERT(so->so_rcv.sb_mb == m);
1394 				if (m) {
1395 					m->m_nextpkt = nextrecord;
1396 					if (nextrecord == NULL)
1397 						so->so_rcv.sb_lastrecord = m;
1398 				} else {
1399 					so->so_rcv.sb_mb = nextrecord;
1400 					SB_EMPTY_FIXUP(&so->so_rcv);
1401 				}
1402 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1403 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1404 			}
1405 		} else if (flags & MSG_PEEK)
1406 			moff += len;
1407 		else {
1408 			if (mp != NULL) {
1409 				mt = m_copym(m, 0, len, M_NOWAIT);
1410 				if (__predict_false(mt == NULL)) {
1411 					sounlock(so);
1412 					mt = m_copym(m, 0, len, M_WAIT);
1413 					solock(so);
1414 				}
1415 				*mp = mt;
1416 			}
1417 			m->m_data += len;
1418 			m->m_len -= len;
1419 			so->so_rcv.sb_cc -= len;
1420 		}
1421 		if (so->so_oobmark) {
1422 			if ((flags & MSG_PEEK) == 0) {
1423 				so->so_oobmark -= len;
1424 				if (so->so_oobmark == 0) {
1425 					so->so_state |= SS_RCVATMARK;
1426 					break;
1427 				}
1428 			} else {
1429 				offset += len;
1430 				if (offset == so->so_oobmark)
1431 					break;
1432 			}
1433 		}
1434 		if (flags & MSG_EOR)
1435 			break;
1436 		/*
1437 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1438 		 * we must not quit until "uio->uio_resid == 0" or an error
1439 		 * termination.  If a signal/timeout occurs, return
1440 		 * with a short count but without error.
1441 		 * Keep sockbuf locked against other readers.
1442 		 */
1443 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1444 		    !sosendallatonce(so) && !nextrecord) {
1445 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1446 				break;
1447 			/*
1448 			 * If we are peeking and the socket receive buffer is
1449 			 * full, stop since we can't get more data to peek at.
1450 			 */
1451 			if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
1452 				break;
1453 			/*
1454 			 * If we've drained the socket buffer, tell the
1455 			 * protocol in case it needs to do something to
1456 			 * get it filled again.
1457 			 */
1458 			if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1459 				(*pr->pr_usrreq)(so, PRU_RCVD,
1460 				    NULL, (struct mbuf *)(long)flags, NULL, l);
1461 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1462 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1463 			error = sbwait(&so->so_rcv);
1464 			if (error != 0) {
1465 				sbunlock(&so->so_rcv);
1466 				sounlock(so);
1467 				splx(s);
1468 				return 0;
1469 			}
1470 			if ((m = so->so_rcv.sb_mb) != NULL)
1471 				nextrecord = m->m_nextpkt;
1472 		}
1473 	}
1474 
1475 	if (m && atomic) {
1476 		flags |= MSG_TRUNC;
1477 		if ((flags & MSG_PEEK) == 0)
1478 			(void) sbdroprecord(&so->so_rcv);
1479 	}
1480 	if ((flags & MSG_PEEK) == 0) {
1481 		if (m == NULL) {
1482 			/*
1483 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1484 			 * part makes sure sb_lastrecord is up-to-date if
1485 			 * there is still data in the socket buffer.
1486 			 */
1487 			so->so_rcv.sb_mb = nextrecord;
1488 			if (so->so_rcv.sb_mb == NULL) {
1489 				so->so_rcv.sb_mbtail = NULL;
1490 				so->so_rcv.sb_lastrecord = NULL;
1491 			} else if (nextrecord->m_nextpkt == NULL)
1492 				so->so_rcv.sb_lastrecord = nextrecord;
1493 		}
1494 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1495 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1496 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1497 			(*pr->pr_usrreq)(so, PRU_RCVD, NULL,
1498 			    (struct mbuf *)(long)flags, NULL, l);
1499 	}
1500 	if (orig_resid == uio->uio_resid && orig_resid &&
1501 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1502 		sbunlock(&so->so_rcv);
1503 		goto restart;
1504 	}
1505 
1506 	if (flagsp != NULL)
1507 		*flagsp |= flags;
1508  release:
1509 	sbunlock(&so->so_rcv);
1510 	sounlock(so);
1511 	splx(s);
1512 	return error;
1513 }
1514 
1515 int
1516 soshutdown(struct socket *so, int how)
1517 {
1518 	const struct protosw	*pr;
1519 	int	error;
1520 
1521 	KASSERT(solocked(so));
1522 
1523 	pr = so->so_proto;
1524 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1525 		return (EINVAL);
1526 
1527 	if (how == SHUT_RD || how == SHUT_RDWR) {
1528 		sorflush(so);
1529 		error = 0;
1530 	}
1531 	if (how == SHUT_WR || how == SHUT_RDWR)
1532 		error = (*pr->pr_usrreq)(so, PRU_SHUTDOWN, NULL,
1533 		    NULL, NULL, NULL);
1534 
1535 	return error;
1536 }
1537 
1538 void
1539 sorflush(struct socket *so)
1540 {
1541 	struct sockbuf	*sb, asb;
1542 	const struct protosw	*pr;
1543 
1544 	KASSERT(solocked(so));
1545 
1546 	sb = &so->so_rcv;
1547 	pr = so->so_proto;
1548 	socantrcvmore(so);
1549 	sb->sb_flags |= SB_NOINTR;
1550 	(void )sblock(sb, M_WAITOK);
1551 	sbunlock(sb);
1552 	asb = *sb;
1553 	/*
1554 	 * Clear most of the sockbuf structure, but leave some of the
1555 	 * fields valid.
1556 	 */
1557 	memset(&sb->sb_startzero, 0,
1558 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1559 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) {
1560 		sounlock(so);
1561 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1562 		solock(so);
1563 	}
1564 	sbrelease(&asb, so);
1565 }
1566 
1567 /*
1568  * internal set SOL_SOCKET options
1569  */
1570 static int
1571 sosetopt1(struct socket *so, const struct sockopt *sopt)
1572 {
1573 	int error, optval;
1574 	struct linger l;
1575 	struct timeval tv;
1576 
1577 	switch (sopt->sopt_name) {
1578 
1579 #ifdef INET
1580 	case SO_ACCEPTFILTER:
1581 		error = do_setopt_accept_filter(so, sopt);
1582 		if (error)
1583 			return error;
1584 		break;
1585 #endif
1586 
1587   	case SO_LINGER:
1588  		error = sockopt_get(sopt, &l, sizeof(l));
1589  		if (error)
1590  			return (error);
1591 
1592  		if (l.l_linger < 0 || l.l_linger > USHRT_MAX ||
1593  		    l.l_linger > (INT_MAX / hz))
1594 			return EDOM;
1595  		so->so_linger = l.l_linger;
1596  		if (l.l_onoff)
1597  			so->so_options |= SO_LINGER;
1598  		else
1599  			so->so_options &= ~SO_LINGER;
1600 
1601   		break;
1602 
1603 	case SO_DEBUG:
1604 	case SO_KEEPALIVE:
1605 	case SO_DONTROUTE:
1606 	case SO_USELOOPBACK:
1607 	case SO_BROADCAST:
1608 	case SO_REUSEADDR:
1609 	case SO_REUSEPORT:
1610 	case SO_OOBINLINE:
1611 	case SO_TIMESTAMP:
1612 		error = sockopt_getint(sopt, &optval);
1613 		if (error)
1614 			return (error);
1615 
1616 		if (optval)
1617 			so->so_options |= sopt->sopt_name;
1618 		else
1619 			so->so_options &= ~sopt->sopt_name;
1620 		break;
1621 
1622 	case SO_SNDBUF:
1623 	case SO_RCVBUF:
1624 	case SO_SNDLOWAT:
1625 	case SO_RCVLOWAT:
1626 		error = sockopt_getint(sopt, &optval);
1627 		if (error)
1628 			return (error);
1629 
1630 		/*
1631 		 * Values < 1 make no sense for any of these
1632 		 * options, so disallow them.
1633 		 */
1634 		if (optval < 1)
1635 			return EINVAL;
1636 
1637 		switch (sopt->sopt_name) {
1638 		case SO_SNDBUF:
1639 			if (sbreserve(&so->so_snd, (u_long)optval, so) == 0)
1640 				return ENOBUFS;
1641 
1642 			so->so_snd.sb_flags &= ~SB_AUTOSIZE;
1643 			break;
1644 
1645 		case SO_RCVBUF:
1646 			if (sbreserve(&so->so_rcv, (u_long)optval, so) == 0)
1647 				return ENOBUFS;
1648 
1649 			so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1650 			break;
1651 
1652 		/*
1653 		 * Make sure the low-water is never greater than
1654 		 * the high-water.
1655 		 */
1656 		case SO_SNDLOWAT:
1657 			if (optval > so->so_snd.sb_hiwat)
1658 				optval = so->so_snd.sb_hiwat;
1659 
1660 			so->so_snd.sb_lowat = optval;
1661 			break;
1662 
1663 		case SO_RCVLOWAT:
1664 			if (optval > so->so_rcv.sb_hiwat)
1665 				optval = so->so_rcv.sb_hiwat;
1666 
1667 			so->so_rcv.sb_lowat = optval;
1668 			break;
1669 		}
1670 		break;
1671 
1672 	case SO_SNDTIMEO:
1673 	case SO_RCVTIMEO:
1674 		error = sockopt_get(sopt, &tv, sizeof(tv));
1675 		if (error)
1676 			return (error);
1677 
1678 		if (tv.tv_sec > (INT_MAX - tv.tv_usec / tick) / hz)
1679 			return EDOM;
1680 
1681 		optval = tv.tv_sec * hz + tv.tv_usec / tick;
1682 		if (optval == 0 && tv.tv_usec != 0)
1683 			optval = 1;
1684 
1685 		switch (sopt->sopt_name) {
1686 		case SO_SNDTIMEO:
1687 			so->so_snd.sb_timeo = optval;
1688 			break;
1689 		case SO_RCVTIMEO:
1690 			so->so_rcv.sb_timeo = optval;
1691 			break;
1692 		}
1693 		break;
1694 
1695 	default:
1696 		return ENOPROTOOPT;
1697 	}
1698 	return 0;
1699 }
1700 
1701 int
1702 sosetopt(struct socket *so, struct sockopt *sopt)
1703 {
1704 	int error, prerr;
1705 
1706 	solock(so);
1707 	if (sopt->sopt_level == SOL_SOCKET)
1708 		error = sosetopt1(so, sopt);
1709 	else
1710 		error = ENOPROTOOPT;
1711 
1712 	if ((error == 0 || error == ENOPROTOOPT) &&
1713 	    so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) {
1714 		/* give the protocol stack a shot */
1715 		prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, sopt);
1716 		if (prerr == 0)
1717 			error = 0;
1718 		else if (prerr != ENOPROTOOPT)
1719 			error = prerr;
1720 	}
1721 	sounlock(so);
1722 	return error;
1723 }
1724 
1725 /*
1726  * so_setsockopt() is a wrapper providing a sockopt structure for sosetopt()
1727  */
1728 int
1729 so_setsockopt(struct lwp *l, struct socket *so, int level, int name,
1730     const void *val, size_t valsize)
1731 {
1732 	struct sockopt sopt;
1733 	int error;
1734 
1735 	KASSERT(valsize == 0 || val != NULL);
1736 
1737 	sockopt_init(&sopt, level, name, valsize);
1738 	sockopt_set(&sopt, val, valsize);
1739 
1740 	error = sosetopt(so, &sopt);
1741 
1742 	sockopt_destroy(&sopt);
1743 
1744 	return error;
1745 }
1746 
1747 /*
1748  * internal get SOL_SOCKET options
1749  */
1750 static int
1751 sogetopt1(struct socket *so, struct sockopt *sopt)
1752 {
1753 	int error, optval;
1754 	struct linger l;
1755 	struct timeval tv;
1756 
1757 	switch (sopt->sopt_name) {
1758 
1759 #ifdef INET
1760 	case SO_ACCEPTFILTER:
1761 		error = do_getopt_accept_filter(so, sopt);
1762 		break;
1763 #endif
1764 
1765 	case SO_LINGER:
1766 		l.l_onoff = (so->so_options & SO_LINGER) ? 1 : 0;
1767 		l.l_linger = so->so_linger;
1768 
1769 		error = sockopt_set(sopt, &l, sizeof(l));
1770 		break;
1771 
1772 	case SO_USELOOPBACK:
1773 	case SO_DONTROUTE:
1774 	case SO_DEBUG:
1775 	case SO_KEEPALIVE:
1776 	case SO_REUSEADDR:
1777 	case SO_REUSEPORT:
1778 	case SO_BROADCAST:
1779 	case SO_OOBINLINE:
1780 	case SO_TIMESTAMP:
1781 		error = sockopt_setint(sopt,
1782 		    (so->so_options & sopt->sopt_name) ? 1 : 0);
1783 		break;
1784 
1785 	case SO_TYPE:
1786 		error = sockopt_setint(sopt, so->so_type);
1787 		break;
1788 
1789 	case SO_ERROR:
1790 		error = sockopt_setint(sopt, so->so_error);
1791 		so->so_error = 0;
1792 		break;
1793 
1794 	case SO_SNDBUF:
1795 		error = sockopt_setint(sopt, so->so_snd.sb_hiwat);
1796 		break;
1797 
1798 	case SO_RCVBUF:
1799 		error = sockopt_setint(sopt, so->so_rcv.sb_hiwat);
1800 		break;
1801 
1802 	case SO_SNDLOWAT:
1803 		error = sockopt_setint(sopt, so->so_snd.sb_lowat);
1804 		break;
1805 
1806 	case SO_RCVLOWAT:
1807 		error = sockopt_setint(sopt, so->so_rcv.sb_lowat);
1808 		break;
1809 
1810 	case SO_SNDTIMEO:
1811 	case SO_RCVTIMEO:
1812 		optval = (sopt->sopt_name == SO_SNDTIMEO ?
1813 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1814 
1815 		tv.tv_sec = optval / hz;
1816 		tv.tv_usec = (optval % hz) * tick;
1817 
1818 		error = sockopt_set(sopt, &tv, sizeof(tv));
1819 		break;
1820 
1821 	case SO_OVERFLOWED:
1822 		error = sockopt_setint(sopt, so->so_rcv.sb_overflowed);
1823 		break;
1824 
1825 	default:
1826 		error = ENOPROTOOPT;
1827 		break;
1828 	}
1829 
1830 	return (error);
1831 }
1832 
1833 int
1834 sogetopt(struct socket *so, struct sockopt *sopt)
1835 {
1836 	int		error;
1837 
1838 	solock(so);
1839 	if (sopt->sopt_level != SOL_SOCKET) {
1840 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1841 			error = ((*so->so_proto->pr_ctloutput)
1842 			    (PRCO_GETOPT, so, sopt));
1843 		} else
1844 			error = (ENOPROTOOPT);
1845 	} else {
1846 		error = sogetopt1(so, sopt);
1847 	}
1848 	sounlock(so);
1849 	return (error);
1850 }
1851 
1852 /*
1853  * alloc sockopt data buffer buffer
1854  *	- will be released at destroy
1855  */
1856 static void
1857 sockopt_alloc(struct sockopt *sopt, size_t len)
1858 {
1859 
1860 	KASSERT(sopt->sopt_size == 0);
1861 
1862 	if (len > sizeof(sopt->sopt_buf))
1863 		sopt->sopt_data = malloc(len, M_SOOPTS, M_WAITOK | M_ZERO);
1864 	else
1865 		sopt->sopt_data = sopt->sopt_buf;
1866 
1867 	sopt->sopt_size = len;
1868 }
1869 
1870 /*
1871  * initialise sockopt storage
1872  */
1873 void
1874 sockopt_init(struct sockopt *sopt, int level, int name, size_t size)
1875 {
1876 
1877 	memset(sopt, 0, sizeof(*sopt));
1878 
1879 	sopt->sopt_level = level;
1880 	sopt->sopt_name = name;
1881 	sockopt_alloc(sopt, size);
1882 }
1883 
1884 /*
1885  * destroy sockopt storage
1886  *	- will release any held memory references
1887  */
1888 void
1889 sockopt_destroy(struct sockopt *sopt)
1890 {
1891 
1892 	if (sopt->sopt_data != sopt->sopt_buf)
1893 		free(sopt->sopt_data, M_SOOPTS);
1894 
1895 	memset(sopt, 0, sizeof(*sopt));
1896 }
1897 
1898 /*
1899  * set sockopt value
1900  *	- value is copied into sockopt
1901  * 	- memory is allocated when necessary
1902  */
1903 int
1904 sockopt_set(struct sockopt *sopt, const void *buf, size_t len)
1905 {
1906 
1907 	if (sopt->sopt_size == 0)
1908 		sockopt_alloc(sopt, len);
1909 
1910 	KASSERT(sopt->sopt_size == len);
1911 	memcpy(sopt->sopt_data, buf, len);
1912 	return 0;
1913 }
1914 
1915 /*
1916  * common case of set sockopt integer value
1917  */
1918 int
1919 sockopt_setint(struct sockopt *sopt, int val)
1920 {
1921 
1922 	return sockopt_set(sopt, &val, sizeof(int));
1923 }
1924 
1925 /*
1926  * get sockopt value
1927  *	- correct size must be given
1928  */
1929 int
1930 sockopt_get(const struct sockopt *sopt, void *buf, size_t len)
1931 {
1932 
1933 	if (sopt->sopt_size != len)
1934 		return EINVAL;
1935 
1936 	memcpy(buf, sopt->sopt_data, len);
1937 	return 0;
1938 }
1939 
1940 /*
1941  * common case of get sockopt integer value
1942  */
1943 int
1944 sockopt_getint(const struct sockopt *sopt, int *valp)
1945 {
1946 
1947 	return sockopt_get(sopt, valp, sizeof(int));
1948 }
1949 
1950 /*
1951  * set sockopt value from mbuf
1952  *	- ONLY for legacy code
1953  *	- mbuf is released by sockopt
1954  */
1955 int
1956 sockopt_setmbuf(struct sockopt *sopt, struct mbuf *m)
1957 {
1958 	size_t len;
1959 
1960 	len = m_length(m);
1961 
1962 	if (sopt->sopt_size == 0)
1963 		sockopt_alloc(sopt, len);
1964 
1965 	KASSERT(sopt->sopt_size == len);
1966 	m_copydata(m, 0, len, sopt->sopt_data);
1967 	m_freem(m);
1968 
1969 	return 0;
1970 }
1971 
1972 /*
1973  * get sockopt value into mbuf
1974  *	- ONLY for legacy code
1975  *	- mbuf to be released by the caller
1976  */
1977 struct mbuf *
1978 sockopt_getmbuf(const struct sockopt *sopt)
1979 {
1980 	struct mbuf *m;
1981 
1982 	m = m_get(M_WAIT, MT_SOOPTS);
1983 	if (m == NULL)
1984 		return NULL;
1985 
1986 	m->m_len = MLEN;
1987 	m_copyback(m, 0, sopt->sopt_size, sopt->sopt_data);
1988 	if (m_length(m) != max(sopt->sopt_size, MLEN)) {
1989 		m_freem(m);
1990 		return NULL;
1991 	}
1992 	m->m_len = min(sopt->sopt_size, MLEN);
1993 
1994 	return m;
1995 }
1996 
1997 void
1998 sohasoutofband(struct socket *so)
1999 {
2000 
2001 	fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
2002 	selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, 0);
2003 }
2004 
2005 static void
2006 filt_sordetach(struct knote *kn)
2007 {
2008 	struct socket	*so;
2009 
2010 	so = ((file_t *)kn->kn_obj)->f_data;
2011 	solock(so);
2012 	SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
2013 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
2014 		so->so_rcv.sb_flags &= ~SB_KNOTE;
2015 	sounlock(so);
2016 }
2017 
2018 /*ARGSUSED*/
2019 static int
2020 filt_soread(struct knote *kn, long hint)
2021 {
2022 	struct socket	*so;
2023 	int rv;
2024 
2025 	so = ((file_t *)kn->kn_obj)->f_data;
2026 	if (hint != NOTE_SUBMIT)
2027 		solock(so);
2028 	kn->kn_data = so->so_rcv.sb_cc;
2029 	if (so->so_state & SS_CANTRCVMORE) {
2030 		kn->kn_flags |= EV_EOF;
2031 		kn->kn_fflags = so->so_error;
2032 		rv = 1;
2033 	} else if (so->so_error)	/* temporary udp error */
2034 		rv = 1;
2035 	else if (kn->kn_sfflags & NOTE_LOWAT)
2036 		rv = (kn->kn_data >= kn->kn_sdata);
2037 	else
2038 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
2039 	if (hint != NOTE_SUBMIT)
2040 		sounlock(so);
2041 	return rv;
2042 }
2043 
2044 static void
2045 filt_sowdetach(struct knote *kn)
2046 {
2047 	struct socket	*so;
2048 
2049 	so = ((file_t *)kn->kn_obj)->f_data;
2050 	solock(so);
2051 	SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
2052 	if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
2053 		so->so_snd.sb_flags &= ~SB_KNOTE;
2054 	sounlock(so);
2055 }
2056 
2057 /*ARGSUSED*/
2058 static int
2059 filt_sowrite(struct knote *kn, long hint)
2060 {
2061 	struct socket	*so;
2062 	int rv;
2063 
2064 	so = ((file_t *)kn->kn_obj)->f_data;
2065 	if (hint != NOTE_SUBMIT)
2066 		solock(so);
2067 	kn->kn_data = sbspace(&so->so_snd);
2068 	if (so->so_state & SS_CANTSENDMORE) {
2069 		kn->kn_flags |= EV_EOF;
2070 		kn->kn_fflags = so->so_error;
2071 		rv = 1;
2072 	} else if (so->so_error)	/* temporary udp error */
2073 		rv = 1;
2074 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2075 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
2076 		rv = 0;
2077 	else if (kn->kn_sfflags & NOTE_LOWAT)
2078 		rv = (kn->kn_data >= kn->kn_sdata);
2079 	else
2080 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
2081 	if (hint != NOTE_SUBMIT)
2082 		sounlock(so);
2083 	return rv;
2084 }
2085 
2086 /*ARGSUSED*/
2087 static int
2088 filt_solisten(struct knote *kn, long hint)
2089 {
2090 	struct socket	*so;
2091 	int rv;
2092 
2093 	so = ((file_t *)kn->kn_obj)->f_data;
2094 
2095 	/*
2096 	 * Set kn_data to number of incoming connections, not
2097 	 * counting partial (incomplete) connections.
2098 	 */
2099 	if (hint != NOTE_SUBMIT)
2100 		solock(so);
2101 	kn->kn_data = so->so_qlen;
2102 	rv = (kn->kn_data > 0);
2103 	if (hint != NOTE_SUBMIT)
2104 		sounlock(so);
2105 	return rv;
2106 }
2107 
2108 static const struct filterops solisten_filtops =
2109 	{ 1, NULL, filt_sordetach, filt_solisten };
2110 static const struct filterops soread_filtops =
2111 	{ 1, NULL, filt_sordetach, filt_soread };
2112 static const struct filterops sowrite_filtops =
2113 	{ 1, NULL, filt_sowdetach, filt_sowrite };
2114 
2115 int
2116 soo_kqfilter(struct file *fp, struct knote *kn)
2117 {
2118 	struct socket	*so;
2119 	struct sockbuf	*sb;
2120 
2121 	so = ((file_t *)kn->kn_obj)->f_data;
2122 	solock(so);
2123 	switch (kn->kn_filter) {
2124 	case EVFILT_READ:
2125 		if (so->so_options & SO_ACCEPTCONN)
2126 			kn->kn_fop = &solisten_filtops;
2127 		else
2128 			kn->kn_fop = &soread_filtops;
2129 		sb = &so->so_rcv;
2130 		break;
2131 	case EVFILT_WRITE:
2132 		kn->kn_fop = &sowrite_filtops;
2133 		sb = &so->so_snd;
2134 		break;
2135 	default:
2136 		sounlock(so);
2137 		return (EINVAL);
2138 	}
2139 	SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
2140 	sb->sb_flags |= SB_KNOTE;
2141 	sounlock(so);
2142 	return (0);
2143 }
2144 
2145 static int
2146 sodopoll(struct socket *so, int events)
2147 {
2148 	int revents;
2149 
2150 	revents = 0;
2151 
2152 	if (events & (POLLIN | POLLRDNORM))
2153 		if (soreadable(so))
2154 			revents |= events & (POLLIN | POLLRDNORM);
2155 
2156 	if (events & (POLLOUT | POLLWRNORM))
2157 		if (sowritable(so))
2158 			revents |= events & (POLLOUT | POLLWRNORM);
2159 
2160 	if (events & (POLLPRI | POLLRDBAND))
2161 		if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
2162 			revents |= events & (POLLPRI | POLLRDBAND);
2163 
2164 	return revents;
2165 }
2166 
2167 int
2168 sopoll(struct socket *so, int events)
2169 {
2170 	int revents = 0;
2171 
2172 #ifndef DIAGNOSTIC
2173 	/*
2174 	 * Do a quick, unlocked check in expectation that the socket
2175 	 * will be ready for I/O.  Don't do this check if DIAGNOSTIC,
2176 	 * as the solocked() assertions will fail.
2177 	 */
2178 	if ((revents = sodopoll(so, events)) != 0)
2179 		return revents;
2180 #endif
2181 
2182 	solock(so);
2183 	if ((revents = sodopoll(so, events)) == 0) {
2184 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2185 			selrecord(curlwp, &so->so_rcv.sb_sel);
2186 			so->so_rcv.sb_flags |= SB_NOTIFY;
2187 		}
2188 
2189 		if (events & (POLLOUT | POLLWRNORM)) {
2190 			selrecord(curlwp, &so->so_snd.sb_sel);
2191 			so->so_snd.sb_flags |= SB_NOTIFY;
2192 		}
2193 	}
2194 	sounlock(so);
2195 
2196 	return revents;
2197 }
2198 
2199 
2200 #include <sys/sysctl.h>
2201 
2202 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO);
2203 
2204 /*
2205  * sysctl helper routine for kern.somaxkva.  ensures that the given
2206  * value is not too small.
2207  * (XXX should we maybe make sure it's not too large as well?)
2208  */
2209 static int
2210 sysctl_kern_somaxkva(SYSCTLFN_ARGS)
2211 {
2212 	int error, new_somaxkva;
2213 	struct sysctlnode node;
2214 
2215 	new_somaxkva = somaxkva;
2216 	node = *rnode;
2217 	node.sysctl_data = &new_somaxkva;
2218 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2219 	if (error || newp == NULL)
2220 		return (error);
2221 
2222 	if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */
2223 		return (EINVAL);
2224 
2225 	mutex_enter(&so_pendfree_lock);
2226 	somaxkva = new_somaxkva;
2227 	cv_broadcast(&socurkva_cv);
2228 	mutex_exit(&so_pendfree_lock);
2229 
2230 	return (error);
2231 }
2232 
2233 SYSCTL_SETUP(sysctl_kern_somaxkva_setup, "sysctl kern.somaxkva setup")
2234 {
2235 
2236 	sysctl_createv(clog, 0, NULL, NULL,
2237 		       CTLFLAG_PERMANENT,
2238 		       CTLTYPE_NODE, "kern", NULL,
2239 		       NULL, 0, NULL, 0,
2240 		       CTL_KERN, CTL_EOL);
2241 
2242 	sysctl_createv(clog, 0, NULL, NULL,
2243 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2244 		       CTLTYPE_INT, "somaxkva",
2245 		       SYSCTL_DESCR("Maximum amount of kernel memory to be "
2246 				    "used for socket buffers"),
2247 		       sysctl_kern_somaxkva, 0, NULL, 0,
2248 		       CTL_KERN, KERN_SOMAXKVA, CTL_EOL);
2249 }
2250