xref: /netbsd-src/sys/kern/uipc_socket2.c (revision b1c86f5f087524e68db12794ee9c3e3da1ab17a0)
1 /*	$NetBSD: uipc_socket2.c,v 1.106 2009/12/30 22:12:12 elad Exp $	*/
2 
3 /*-
4  * Copyright (c) 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
17  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
18  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
19  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
20  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
21  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
22  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
23  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
24  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
25  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
26  * POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 /*
30  * Copyright (c) 1982, 1986, 1988, 1990, 1993
31  *	The Regents of the University of California.  All rights reserved.
32  *
33  * Redistribution and use in source and binary forms, with or without
34  * modification, are permitted provided that the following conditions
35  * are met:
36  * 1. Redistributions of source code must retain the above copyright
37  *    notice, this list of conditions and the following disclaimer.
38  * 2. Redistributions in binary form must reproduce the above copyright
39  *    notice, this list of conditions and the following disclaimer in the
40  *    documentation and/or other materials provided with the distribution.
41  * 3. Neither the name of the University nor the names of its contributors
42  *    may be used to endorse or promote products derived from this software
43  *    without specific prior written permission.
44  *
45  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
46  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
47  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
48  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
49  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
50  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
51  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
52  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
53  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
54  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
55  * SUCH DAMAGE.
56  *
57  *	@(#)uipc_socket2.c	8.2 (Berkeley) 2/14/95
58  */
59 
60 #include <sys/cdefs.h>
61 __KERNEL_RCSID(0, "$NetBSD: uipc_socket2.c,v 1.106 2009/12/30 22:12:12 elad Exp $");
62 
63 #include "opt_mbuftrace.h"
64 #include "opt_sb_max.h"
65 
66 #include <sys/param.h>
67 #include <sys/systm.h>
68 #include <sys/proc.h>
69 #include <sys/file.h>
70 #include <sys/buf.h>
71 #include <sys/malloc.h>
72 #include <sys/mbuf.h>
73 #include <sys/protosw.h>
74 #include <sys/domain.h>
75 #include <sys/poll.h>
76 #include <sys/socket.h>
77 #include <sys/socketvar.h>
78 #include <sys/signalvar.h>
79 #include <sys/kauth.h>
80 #include <sys/pool.h>
81 #include <sys/uidinfo.h>
82 
83 /*
84  * Primitive routines for operating on sockets and socket buffers.
85  *
86  * Locking rules and assumptions:
87  *
88  * o socket::so_lock can change on the fly.  The low level routines used
89  *   to lock sockets are aware of this.  When so_lock is acquired, the
90  *   routine locking must check to see if so_lock still points to the
91  *   lock that was acquired.  If so_lock has changed in the meantime, the
92  *   now irellevant lock that was acquired must be dropped and the lock
93  *   operation retried.  Although not proven here, this is completely safe
94  *   on a multiprocessor system, even with relaxed memory ordering, given
95  *   the next two rules:
96  *
97  * o In order to mutate so_lock, the lock pointed to by the current value
98  *   of so_lock must be held: i.e., the socket must be held locked by the
99  *   changing thread.  The thread must issue membar_exit() to prevent
100  *   memory accesses being reordered, and can set so_lock to the desired
101  *   value.  If the lock pointed to by the new value of so_lock is not
102  *   held by the changing thread, the socket must then be considered
103  *   unlocked.
104  *
105  * o If so_lock is mutated, and the previous lock referred to by so_lock
106  *   could still be visible to other threads in the system (e.g. via file
107  *   descriptor or protocol-internal reference), then the old lock must
108  *   remain valid until the socket and/or protocol control block has been
109  *   torn down.
110  *
111  * o If a socket has a non-NULL so_head value (i.e. is in the process of
112  *   connecting), then locking the socket must also lock the socket pointed
113  *   to by so_head: their lock pointers must match.
114  *
115  * o If a socket has connections in progress (so_q, so_q0 not empty) then
116  *   locking the socket must also lock the sockets attached to both queues.
117  *   Again, their lock pointers must match.
118  *
119  * o Beyond the initial lock assigment in socreate(), assigning locks to
120  *   sockets is the responsibility of the individual protocols / protocol
121  *   domains.
122  */
123 
124 static pool_cache_t socket_cache;
125 
126 u_long	sb_max = SB_MAX;	/* maximum socket buffer size */
127 static u_long sb_max_adj;	/* adjusted sb_max */
128 
129 /*
130  * Procedures to manipulate state flags of socket
131  * and do appropriate wakeups.  Normal sequence from the
132  * active (originating) side is that soisconnecting() is
133  * called during processing of connect() call,
134  * resulting in an eventual call to soisconnected() if/when the
135  * connection is established.  When the connection is torn down
136  * soisdisconnecting() is called during processing of disconnect() call,
137  * and soisdisconnected() is called when the connection to the peer
138  * is totally severed.  The semantics of these routines are such that
139  * connectionless protocols can call soisconnected() and soisdisconnected()
140  * only, bypassing the in-progress calls when setting up a ``connection''
141  * takes no time.
142  *
143  * From the passive side, a socket is created with
144  * two queues of sockets: so_q0 for connections in progress
145  * and so_q for connections already made and awaiting user acceptance.
146  * As a protocol is preparing incoming connections, it creates a socket
147  * structure queued on so_q0 by calling sonewconn().  When the connection
148  * is established, soisconnected() is called, and transfers the
149  * socket structure to so_q, making it available to accept().
150  *
151  * If a socket is closed with sockets on either
152  * so_q0 or so_q, these sockets are dropped.
153  *
154  * If higher level protocols are implemented in
155  * the kernel, the wakeups done here will sometimes
156  * cause software-interrupt process scheduling.
157  */
158 
159 void
160 soisconnecting(struct socket *so)
161 {
162 
163 	KASSERT(solocked(so));
164 
165 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
166 	so->so_state |= SS_ISCONNECTING;
167 }
168 
169 void
170 soisconnected(struct socket *so)
171 {
172 	struct socket	*head;
173 
174 	head = so->so_head;
175 
176 	KASSERT(solocked(so));
177 	KASSERT(head == NULL || solocked2(so, head));
178 
179 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
180 	so->so_state |= SS_ISCONNECTED;
181 	if (head && so->so_onq == &head->so_q0) {
182 		if ((so->so_options & SO_ACCEPTFILTER) == 0) {
183 			soqremque(so, 0);
184 			soqinsque(head, so, 1);
185 			sorwakeup(head);
186 			cv_broadcast(&head->so_cv);
187 		} else {
188 			so->so_upcall =
189 			    head->so_accf->so_accept_filter->accf_callback;
190 			so->so_upcallarg = head->so_accf->so_accept_filter_arg;
191 			so->so_rcv.sb_flags |= SB_UPCALL;
192 			so->so_options &= ~SO_ACCEPTFILTER;
193 			(*so->so_upcall)(so, so->so_upcallarg,
194 					 POLLIN|POLLRDNORM, M_DONTWAIT);
195 		}
196 	} else {
197 		cv_broadcast(&so->so_cv);
198 		sorwakeup(so);
199 		sowwakeup(so);
200 	}
201 }
202 
203 void
204 soisdisconnecting(struct socket *so)
205 {
206 
207 	KASSERT(solocked(so));
208 
209 	so->so_state &= ~SS_ISCONNECTING;
210 	so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE);
211 	cv_broadcast(&so->so_cv);
212 	sowwakeup(so);
213 	sorwakeup(so);
214 }
215 
216 void
217 soisdisconnected(struct socket *so)
218 {
219 
220 	KASSERT(solocked(so));
221 
222 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
223 	so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED);
224 	cv_broadcast(&so->so_cv);
225 	sowwakeup(so);
226 	sorwakeup(so);
227 }
228 
229 void
230 soinit2(void)
231 {
232 
233 	socket_cache = pool_cache_init(sizeof(struct socket), 0, 0, 0,
234 	    "socket", NULL, IPL_SOFTNET, NULL, NULL, NULL);
235 }
236 
237 /*
238  * When an attempt at a new connection is noted on a socket
239  * which accepts connections, sonewconn is called.  If the
240  * connection is possible (subject to space constraints, etc.)
241  * then we allocate a new structure, propoerly linked into the
242  * data structure of the original socket, and return this.
243  * Connstatus may be 0, SS_ISCONFIRMING, or SS_ISCONNECTED.
244  */
245 struct socket *
246 sonewconn(struct socket *head, int connstatus)
247 {
248 	struct socket	*so;
249 	int		soqueue, error;
250 
251 	KASSERT(connstatus == 0 || connstatus == SS_ISCONFIRMING ||
252 	    connstatus == SS_ISCONNECTED);
253 	KASSERT(solocked(head));
254 
255 	if ((head->so_options & SO_ACCEPTFILTER) != 0)
256 		connstatus = 0;
257 	soqueue = connstatus ? 1 : 0;
258 	if (head->so_qlen + head->so_q0len > 3 * head->so_qlimit / 2)
259 		return NULL;
260 	so = soget(false);
261 	if (so == NULL)
262 		return NULL;
263 	mutex_obj_hold(head->so_lock);
264 	so->so_lock = head->so_lock;
265 	so->so_type = head->so_type;
266 	so->so_options = head->so_options &~ SO_ACCEPTCONN;
267 	so->so_linger = head->so_linger;
268 	so->so_state = head->so_state | SS_NOFDREF;
269 	so->so_nbio = head->so_nbio;
270 	so->so_proto = head->so_proto;
271 	so->so_timeo = head->so_timeo;
272 	so->so_pgid = head->so_pgid;
273 	so->so_send = head->so_send;
274 	so->so_receive = head->so_receive;
275 	so->so_uidinfo = head->so_uidinfo;
276 	so->so_cpid = head->so_cpid;
277 #ifdef MBUFTRACE
278 	so->so_mowner = head->so_mowner;
279 	so->so_rcv.sb_mowner = head->so_rcv.sb_mowner;
280 	so->so_snd.sb_mowner = head->so_snd.sb_mowner;
281 #endif
282 	if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) != 0)
283 		goto out;
284 	so->so_snd.sb_lowat = head->so_snd.sb_lowat;
285 	so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
286 	so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
287 	so->so_snd.sb_timeo = head->so_snd.sb_timeo;
288 	so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
289 	so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
290 	soqinsque(head, so, soqueue);
291 	error = (*so->so_proto->pr_usrreq)(so, PRU_ATTACH, NULL, NULL,
292 	    NULL, NULL);
293 	KASSERT(solocked(so));
294 	if (error != 0) {
295 		(void) soqremque(so, soqueue);
296 out:
297 		/*
298 		 * Remove acccept filter if one is present.
299 		 * XXX Is this really needed?
300 		 */
301 		if (so->so_accf != NULL)
302 			(void)accept_filt_clear(so);
303 		soput(so);
304 		return NULL;
305 	}
306 	if (connstatus) {
307 		sorwakeup(head);
308 		cv_broadcast(&head->so_cv);
309 		so->so_state |= connstatus;
310 	}
311 	return so;
312 }
313 
314 struct socket *
315 soget(bool waitok)
316 {
317 	struct socket *so;
318 
319 	so = pool_cache_get(socket_cache, (waitok ? PR_WAITOK : PR_NOWAIT));
320 	if (__predict_false(so == NULL))
321 		return (NULL);
322 	memset(so, 0, sizeof(*so));
323 	TAILQ_INIT(&so->so_q0);
324 	TAILQ_INIT(&so->so_q);
325 	cv_init(&so->so_cv, "socket");
326 	cv_init(&so->so_rcv.sb_cv, "netio");
327 	cv_init(&so->so_snd.sb_cv, "netio");
328 	selinit(&so->so_rcv.sb_sel);
329 	selinit(&so->so_snd.sb_sel);
330 	so->so_rcv.sb_so = so;
331 	so->so_snd.sb_so = so;
332 	return so;
333 }
334 
335 void
336 soput(struct socket *so)
337 {
338 
339 	KASSERT(!cv_has_waiters(&so->so_cv));
340 	KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv));
341 	KASSERT(!cv_has_waiters(&so->so_snd.sb_cv));
342 	seldestroy(&so->so_rcv.sb_sel);
343 	seldestroy(&so->so_snd.sb_sel);
344 	mutex_obj_free(so->so_lock);
345 	cv_destroy(&so->so_cv);
346 	cv_destroy(&so->so_rcv.sb_cv);
347 	cv_destroy(&so->so_snd.sb_cv);
348 	pool_cache_put(socket_cache, so);
349 }
350 
351 void
352 soqinsque(struct socket *head, struct socket *so, int q)
353 {
354 
355 	KASSERT(solocked2(head, so));
356 
357 #ifdef DIAGNOSTIC
358 	if (so->so_onq != NULL)
359 		panic("soqinsque");
360 #endif
361 
362 	so->so_head = head;
363 	if (q == 0) {
364 		head->so_q0len++;
365 		so->so_onq = &head->so_q0;
366 	} else {
367 		head->so_qlen++;
368 		so->so_onq = &head->so_q;
369 	}
370 	TAILQ_INSERT_TAIL(so->so_onq, so, so_qe);
371 }
372 
373 int
374 soqremque(struct socket *so, int q)
375 {
376 	struct socket	*head;
377 
378 	head = so->so_head;
379 
380 	KASSERT(solocked(so));
381 	if (q == 0) {
382 		if (so->so_onq != &head->so_q0)
383 			return (0);
384 		head->so_q0len--;
385 	} else {
386 		if (so->so_onq != &head->so_q)
387 			return (0);
388 		head->so_qlen--;
389 	}
390 	KASSERT(solocked2(so, head));
391 	TAILQ_REMOVE(so->so_onq, so, so_qe);
392 	so->so_onq = NULL;
393 	so->so_head = NULL;
394 	return (1);
395 }
396 
397 /*
398  * Socantsendmore indicates that no more data will be sent on the
399  * socket; it would normally be applied to a socket when the user
400  * informs the system that no more data is to be sent, by the protocol
401  * code (in case PRU_SHUTDOWN).  Socantrcvmore indicates that no more data
402  * will be received, and will normally be applied to the socket by a
403  * protocol when it detects that the peer will send no more data.
404  * Data queued for reading in the socket may yet be read.
405  */
406 
407 void
408 socantsendmore(struct socket *so)
409 {
410 
411 	KASSERT(solocked(so));
412 
413 	so->so_state |= SS_CANTSENDMORE;
414 	sowwakeup(so);
415 }
416 
417 void
418 socantrcvmore(struct socket *so)
419 {
420 
421 	KASSERT(solocked(so));
422 
423 	so->so_state |= SS_CANTRCVMORE;
424 	sorwakeup(so);
425 }
426 
427 /*
428  * Wait for data to arrive at/drain from a socket buffer.
429  */
430 int
431 sbwait(struct sockbuf *sb)
432 {
433 	struct socket *so;
434 	kmutex_t *lock;
435 	int error;
436 
437 	so = sb->sb_so;
438 
439 	KASSERT(solocked(so));
440 
441 	sb->sb_flags |= SB_NOTIFY;
442 	lock = so->so_lock;
443 	if ((sb->sb_flags & SB_NOINTR) != 0)
444 		error = cv_timedwait(&sb->sb_cv, lock, sb->sb_timeo);
445 	else
446 		error = cv_timedwait_sig(&sb->sb_cv, lock, sb->sb_timeo);
447 	if (__predict_false(lock != so->so_lock))
448 		solockretry(so, lock);
449 	return error;
450 }
451 
452 /*
453  * Wakeup processes waiting on a socket buffer.
454  * Do asynchronous notification via SIGIO
455  * if the socket buffer has the SB_ASYNC flag set.
456  */
457 void
458 sowakeup(struct socket *so, struct sockbuf *sb, int code)
459 {
460 	int band;
461 
462 	KASSERT(solocked(so));
463 	KASSERT(sb->sb_so == so);
464 
465 	if (code == POLL_IN)
466 		band = POLLIN|POLLRDNORM;
467 	else
468 		band = POLLOUT|POLLWRNORM;
469 	sb->sb_flags &= ~SB_NOTIFY;
470 	selnotify(&sb->sb_sel, band, NOTE_SUBMIT);
471 	cv_broadcast(&sb->sb_cv);
472 	if (sb->sb_flags & SB_ASYNC)
473 		fownsignal(so->so_pgid, SIGIO, code, band, so);
474 	if (sb->sb_flags & SB_UPCALL)
475 		(*so->so_upcall)(so, so->so_upcallarg, band, M_DONTWAIT);
476 }
477 
478 /*
479  * Reset a socket's lock pointer.  Wake all threads waiting on the
480  * socket's condition variables so that they can restart their waits
481  * using the new lock.  The existing lock must be held.
482  */
483 void
484 solockreset(struct socket *so, kmutex_t *lock)
485 {
486 
487 	KASSERT(solocked(so));
488 
489 	so->so_lock = lock;
490 	cv_broadcast(&so->so_snd.sb_cv);
491 	cv_broadcast(&so->so_rcv.sb_cv);
492 	cv_broadcast(&so->so_cv);
493 }
494 
495 /*
496  * Socket buffer (struct sockbuf) utility routines.
497  *
498  * Each socket contains two socket buffers: one for sending data and
499  * one for receiving data.  Each buffer contains a queue of mbufs,
500  * information about the number of mbufs and amount of data in the
501  * queue, and other fields allowing poll() statements and notification
502  * on data availability to be implemented.
503  *
504  * Data stored in a socket buffer is maintained as a list of records.
505  * Each record is a list of mbufs chained together with the m_next
506  * field.  Records are chained together with the m_nextpkt field. The upper
507  * level routine soreceive() expects the following conventions to be
508  * observed when placing information in the receive buffer:
509  *
510  * 1. If the protocol requires each message be preceded by the sender's
511  *    name, then a record containing that name must be present before
512  *    any associated data (mbuf's must be of type MT_SONAME).
513  * 2. If the protocol supports the exchange of ``access rights'' (really
514  *    just additional data associated with the message), and there are
515  *    ``rights'' to be received, then a record containing this data
516  *    should be present (mbuf's must be of type MT_CONTROL).
517  * 3. If a name or rights record exists, then it must be followed by
518  *    a data record, perhaps of zero length.
519  *
520  * Before using a new socket structure it is first necessary to reserve
521  * buffer space to the socket, by calling sbreserve().  This should commit
522  * some of the available buffer space in the system buffer pool for the
523  * socket (currently, it does nothing but enforce limits).  The space
524  * should be released by calling sbrelease() when the socket is destroyed.
525  */
526 
527 int
528 sb_max_set(u_long new_sbmax)
529 {
530 	int s;
531 
532 	if (new_sbmax < (16 * 1024))
533 		return (EINVAL);
534 
535 	s = splsoftnet();
536 	sb_max = new_sbmax;
537 	sb_max_adj = (u_quad_t)new_sbmax * MCLBYTES / (MSIZE + MCLBYTES);
538 	splx(s);
539 
540 	return (0);
541 }
542 
543 int
544 soreserve(struct socket *so, u_long sndcc, u_long rcvcc)
545 {
546 
547 	KASSERT(so->so_lock == NULL || solocked(so));
548 
549 	/*
550 	 * there's at least one application (a configure script of screen)
551 	 * which expects a fifo is writable even if it has "some" bytes
552 	 * in its buffer.
553 	 * so we want to make sure (hiwat - lowat) >= (some bytes).
554 	 *
555 	 * PIPE_BUF here is an arbitrary value chosen as (some bytes) above.
556 	 * we expect it's large enough for such applications.
557 	 */
558 	u_long  lowat = MAX(sock_loan_thresh, MCLBYTES);
559 	u_long  hiwat = lowat + PIPE_BUF;
560 
561 	if (sndcc < hiwat)
562 		sndcc = hiwat;
563 	if (sbreserve(&so->so_snd, sndcc, so) == 0)
564 		goto bad;
565 	if (sbreserve(&so->so_rcv, rcvcc, so) == 0)
566 		goto bad2;
567 	if (so->so_rcv.sb_lowat == 0)
568 		so->so_rcv.sb_lowat = 1;
569 	if (so->so_snd.sb_lowat == 0)
570 		so->so_snd.sb_lowat = lowat;
571 	if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
572 		so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
573 	return (0);
574  bad2:
575 	sbrelease(&so->so_snd, so);
576  bad:
577 	return (ENOBUFS);
578 }
579 
580 /*
581  * Allot mbufs to a sockbuf.
582  * Attempt to scale mbmax so that mbcnt doesn't become limiting
583  * if buffering efficiency is near the normal case.
584  */
585 int
586 sbreserve(struct sockbuf *sb, u_long cc, struct socket *so)
587 {
588 	struct lwp *l = curlwp; /* XXX */
589 	rlim_t maxcc;
590 	struct uidinfo *uidinfo;
591 
592 	KASSERT(so->so_lock == NULL || solocked(so));
593 	KASSERT(sb->sb_so == so);
594 	KASSERT(sb_max_adj != 0);
595 
596 	if (cc == 0 || cc > sb_max_adj)
597 		return (0);
598 
599 	maxcc = l->l_proc->p_rlimit[RLIMIT_SBSIZE].rlim_cur;
600 
601 	uidinfo = so->so_uidinfo;
602 	if (!chgsbsize(uidinfo, &sb->sb_hiwat, cc, maxcc))
603 		return 0;
604 	sb->sb_mbmax = min(cc * 2, sb_max);
605 	if (sb->sb_lowat > sb->sb_hiwat)
606 		sb->sb_lowat = sb->sb_hiwat;
607 	return (1);
608 }
609 
610 /*
611  * Free mbufs held by a socket, and reserved mbuf space.  We do not assert
612  * that the socket is held locked here: see sorflush().
613  */
614 void
615 sbrelease(struct sockbuf *sb, struct socket *so)
616 {
617 
618 	KASSERT(sb->sb_so == so);
619 
620 	sbflush(sb);
621 	(void)chgsbsize(so->so_uidinfo, &sb->sb_hiwat, 0, RLIM_INFINITY);
622 	sb->sb_mbmax = 0;
623 }
624 
625 /*
626  * Routines to add and remove
627  * data from an mbuf queue.
628  *
629  * The routines sbappend() or sbappendrecord() are normally called to
630  * append new mbufs to a socket buffer, after checking that adequate
631  * space is available, comparing the function sbspace() with the amount
632  * of data to be added.  sbappendrecord() differs from sbappend() in
633  * that data supplied is treated as the beginning of a new record.
634  * To place a sender's address, optional access rights, and data in a
635  * socket receive buffer, sbappendaddr() should be used.  To place
636  * access rights and data in a socket receive buffer, sbappendrights()
637  * should be used.  In either case, the new data begins a new record.
638  * Note that unlike sbappend() and sbappendrecord(), these routines check
639  * for the caller that there will be enough space to store the data.
640  * Each fails if there is not enough space, or if it cannot find mbufs
641  * to store additional information in.
642  *
643  * Reliable protocols may use the socket send buffer to hold data
644  * awaiting acknowledgement.  Data is normally copied from a socket
645  * send buffer in a protocol with m_copy for output to a peer,
646  * and then removing the data from the socket buffer with sbdrop()
647  * or sbdroprecord() when the data is acknowledged by the peer.
648  */
649 
650 #ifdef SOCKBUF_DEBUG
651 void
652 sblastrecordchk(struct sockbuf *sb, const char *where)
653 {
654 	struct mbuf *m = sb->sb_mb;
655 
656 	KASSERT(solocked(sb->sb_so));
657 
658 	while (m && m->m_nextpkt)
659 		m = m->m_nextpkt;
660 
661 	if (m != sb->sb_lastrecord) {
662 		printf("sblastrecordchk: sb_mb %p sb_lastrecord %p last %p\n",
663 		    sb->sb_mb, sb->sb_lastrecord, m);
664 		printf("packet chain:\n");
665 		for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt)
666 			printf("\t%p\n", m);
667 		panic("sblastrecordchk from %s", where);
668 	}
669 }
670 
671 void
672 sblastmbufchk(struct sockbuf *sb, const char *where)
673 {
674 	struct mbuf *m = sb->sb_mb;
675 	struct mbuf *n;
676 
677 	KASSERT(solocked(sb->sb_so));
678 
679 	while (m && m->m_nextpkt)
680 		m = m->m_nextpkt;
681 
682 	while (m && m->m_next)
683 		m = m->m_next;
684 
685 	if (m != sb->sb_mbtail) {
686 		printf("sblastmbufchk: sb_mb %p sb_mbtail %p last %p\n",
687 		    sb->sb_mb, sb->sb_mbtail, m);
688 		printf("packet tree:\n");
689 		for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) {
690 			printf("\t");
691 			for (n = m; n != NULL; n = n->m_next)
692 				printf("%p ", n);
693 			printf("\n");
694 		}
695 		panic("sblastmbufchk from %s", where);
696 	}
697 }
698 #endif /* SOCKBUF_DEBUG */
699 
700 /*
701  * Link a chain of records onto a socket buffer
702  */
703 #define	SBLINKRECORDCHAIN(sb, m0, mlast)				\
704 do {									\
705 	if ((sb)->sb_lastrecord != NULL)				\
706 		(sb)->sb_lastrecord->m_nextpkt = (m0);			\
707 	else								\
708 		(sb)->sb_mb = (m0);					\
709 	(sb)->sb_lastrecord = (mlast);					\
710 } while (/*CONSTCOND*/0)
711 
712 
713 #define	SBLINKRECORD(sb, m0)						\
714     SBLINKRECORDCHAIN(sb, m0, m0)
715 
716 /*
717  * Append mbuf chain m to the last record in the
718  * socket buffer sb.  The additional space associated
719  * the mbuf chain is recorded in sb.  Empty mbufs are
720  * discarded and mbufs are compacted where possible.
721  */
722 void
723 sbappend(struct sockbuf *sb, struct mbuf *m)
724 {
725 	struct mbuf	*n;
726 
727 	KASSERT(solocked(sb->sb_so));
728 
729 	if (m == 0)
730 		return;
731 
732 #ifdef MBUFTRACE
733 	m_claimm(m, sb->sb_mowner);
734 #endif
735 
736 	SBLASTRECORDCHK(sb, "sbappend 1");
737 
738 	if ((n = sb->sb_lastrecord) != NULL) {
739 		/*
740 		 * XXX Would like to simply use sb_mbtail here, but
741 		 * XXX I need to verify that I won't miss an EOR that
742 		 * XXX way.
743 		 */
744 		do {
745 			if (n->m_flags & M_EOR) {
746 				sbappendrecord(sb, m); /* XXXXXX!!!! */
747 				return;
748 			}
749 		} while (n->m_next && (n = n->m_next));
750 	} else {
751 		/*
752 		 * If this is the first record in the socket buffer, it's
753 		 * also the last record.
754 		 */
755 		sb->sb_lastrecord = m;
756 	}
757 	sbcompress(sb, m, n);
758 	SBLASTRECORDCHK(sb, "sbappend 2");
759 }
760 
761 /*
762  * This version of sbappend() should only be used when the caller
763  * absolutely knows that there will never be more than one record
764  * in the socket buffer, that is, a stream protocol (such as TCP).
765  */
766 void
767 sbappendstream(struct sockbuf *sb, struct mbuf *m)
768 {
769 
770 	KASSERT(solocked(sb->sb_so));
771 	KDASSERT(m->m_nextpkt == NULL);
772 	KASSERT(sb->sb_mb == sb->sb_lastrecord);
773 
774 	SBLASTMBUFCHK(sb, __func__);
775 
776 #ifdef MBUFTRACE
777 	m_claimm(m, sb->sb_mowner);
778 #endif
779 
780 	sbcompress(sb, m, sb->sb_mbtail);
781 
782 	sb->sb_lastrecord = sb->sb_mb;
783 	SBLASTRECORDCHK(sb, __func__);
784 }
785 
786 #ifdef SOCKBUF_DEBUG
787 void
788 sbcheck(struct sockbuf *sb)
789 {
790 	struct mbuf	*m, *m2;
791 	u_long		len, mbcnt;
792 
793 	KASSERT(solocked(sb->sb_so));
794 
795 	len = 0;
796 	mbcnt = 0;
797 	for (m = sb->sb_mb; m; m = m->m_nextpkt) {
798 		for (m2 = m; m2 != NULL; m2 = m2->m_next) {
799 			len += m2->m_len;
800 			mbcnt += MSIZE;
801 			if (m2->m_flags & M_EXT)
802 				mbcnt += m2->m_ext.ext_size;
803 			if (m2->m_nextpkt != NULL)
804 				panic("sbcheck nextpkt");
805 		}
806 	}
807 	if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) {
808 		printf("cc %lu != %lu || mbcnt %lu != %lu\n", len, sb->sb_cc,
809 		    mbcnt, sb->sb_mbcnt);
810 		panic("sbcheck");
811 	}
812 }
813 #endif
814 
815 /*
816  * As above, except the mbuf chain
817  * begins a new record.
818  */
819 void
820 sbappendrecord(struct sockbuf *sb, struct mbuf *m0)
821 {
822 	struct mbuf	*m;
823 
824 	KASSERT(solocked(sb->sb_so));
825 
826 	if (m0 == 0)
827 		return;
828 
829 #ifdef MBUFTRACE
830 	m_claimm(m0, sb->sb_mowner);
831 #endif
832 	/*
833 	 * Put the first mbuf on the queue.
834 	 * Note this permits zero length records.
835 	 */
836 	sballoc(sb, m0);
837 	SBLASTRECORDCHK(sb, "sbappendrecord 1");
838 	SBLINKRECORD(sb, m0);
839 	m = m0->m_next;
840 	m0->m_next = 0;
841 	if (m && (m0->m_flags & M_EOR)) {
842 		m0->m_flags &= ~M_EOR;
843 		m->m_flags |= M_EOR;
844 	}
845 	sbcompress(sb, m, m0);
846 	SBLASTRECORDCHK(sb, "sbappendrecord 2");
847 }
848 
849 /*
850  * As above except that OOB data
851  * is inserted at the beginning of the sockbuf,
852  * but after any other OOB data.
853  */
854 void
855 sbinsertoob(struct sockbuf *sb, struct mbuf *m0)
856 {
857 	struct mbuf	*m, **mp;
858 
859 	KASSERT(solocked(sb->sb_so));
860 
861 	if (m0 == 0)
862 		return;
863 
864 	SBLASTRECORDCHK(sb, "sbinsertoob 1");
865 
866 	for (mp = &sb->sb_mb; (m = *mp) != NULL; mp = &((*mp)->m_nextpkt)) {
867 	    again:
868 		switch (m->m_type) {
869 
870 		case MT_OOBDATA:
871 			continue;		/* WANT next train */
872 
873 		case MT_CONTROL:
874 			if ((m = m->m_next) != NULL)
875 				goto again;	/* inspect THIS train further */
876 		}
877 		break;
878 	}
879 	/*
880 	 * Put the first mbuf on the queue.
881 	 * Note this permits zero length records.
882 	 */
883 	sballoc(sb, m0);
884 	m0->m_nextpkt = *mp;
885 	if (*mp == NULL) {
886 		/* m0 is actually the new tail */
887 		sb->sb_lastrecord = m0;
888 	}
889 	*mp = m0;
890 	m = m0->m_next;
891 	m0->m_next = 0;
892 	if (m && (m0->m_flags & M_EOR)) {
893 		m0->m_flags &= ~M_EOR;
894 		m->m_flags |= M_EOR;
895 	}
896 	sbcompress(sb, m, m0);
897 	SBLASTRECORDCHK(sb, "sbinsertoob 2");
898 }
899 
900 /*
901  * Append address and data, and optionally, control (ancillary) data
902  * to the receive queue of a socket.  If present,
903  * m0 must include a packet header with total length.
904  * Returns 0 if no space in sockbuf or insufficient mbufs.
905  */
906 int
907 sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa, struct mbuf *m0,
908 	struct mbuf *control)
909 {
910 	struct mbuf	*m, *n, *nlast;
911 	int		space, len;
912 
913 	KASSERT(solocked(sb->sb_so));
914 
915 	space = asa->sa_len;
916 
917 	if (m0 != NULL) {
918 		if ((m0->m_flags & M_PKTHDR) == 0)
919 			panic("sbappendaddr");
920 		space += m0->m_pkthdr.len;
921 #ifdef MBUFTRACE
922 		m_claimm(m0, sb->sb_mowner);
923 #endif
924 	}
925 	for (n = control; n; n = n->m_next) {
926 		space += n->m_len;
927 		MCLAIM(n, sb->sb_mowner);
928 		if (n->m_next == 0)	/* keep pointer to last control buf */
929 			break;
930 	}
931 	if (space > sbspace(sb))
932 		return (0);
933 	MGET(m, M_DONTWAIT, MT_SONAME);
934 	if (m == 0)
935 		return (0);
936 	MCLAIM(m, sb->sb_mowner);
937 	/*
938 	 * XXX avoid 'comparison always true' warning which isn't easily
939 	 * avoided.
940 	 */
941 	len = asa->sa_len;
942 	if (len > MLEN) {
943 		MEXTMALLOC(m, asa->sa_len, M_NOWAIT);
944 		if ((m->m_flags & M_EXT) == 0) {
945 			m_free(m);
946 			return (0);
947 		}
948 	}
949 	m->m_len = asa->sa_len;
950 	memcpy(mtod(m, void *), asa, asa->sa_len);
951 	if (n)
952 		n->m_next = m0;		/* concatenate data to control */
953 	else
954 		control = m0;
955 	m->m_next = control;
956 
957 	SBLASTRECORDCHK(sb, "sbappendaddr 1");
958 
959 	for (n = m; n->m_next != NULL; n = n->m_next)
960 		sballoc(sb, n);
961 	sballoc(sb, n);
962 	nlast = n;
963 	SBLINKRECORD(sb, m);
964 
965 	sb->sb_mbtail = nlast;
966 	SBLASTMBUFCHK(sb, "sbappendaddr");
967 	SBLASTRECORDCHK(sb, "sbappendaddr 2");
968 
969 	return (1);
970 }
971 
972 /*
973  * Helper for sbappendchainaddr: prepend a struct sockaddr* to
974  * an mbuf chain.
975  */
976 static inline struct mbuf *
977 m_prepend_sockaddr(struct sockbuf *sb, struct mbuf *m0,
978 		   const struct sockaddr *asa)
979 {
980 	struct mbuf *m;
981 	const int salen = asa->sa_len;
982 
983 	KASSERT(solocked(sb->sb_so));
984 
985 	/* only the first in each chain need be a pkthdr */
986 	MGETHDR(m, M_DONTWAIT, MT_SONAME);
987 	if (m == 0)
988 		return (0);
989 	MCLAIM(m, sb->sb_mowner);
990 #ifdef notyet
991 	if (salen > MHLEN) {
992 		MEXTMALLOC(m, salen, M_NOWAIT);
993 		if ((m->m_flags & M_EXT) == 0) {
994 			m_free(m);
995 			return (0);
996 		}
997 	}
998 #else
999 	KASSERT(salen <= MHLEN);
1000 #endif
1001 	m->m_len = salen;
1002 	memcpy(mtod(m, void *), asa, salen);
1003 	m->m_next = m0;
1004 	m->m_pkthdr.len = salen + m0->m_pkthdr.len;
1005 
1006 	return m;
1007 }
1008 
1009 int
1010 sbappendaddrchain(struct sockbuf *sb, const struct sockaddr *asa,
1011 		  struct mbuf *m0, int sbprio)
1012 {
1013 	int space;
1014 	struct mbuf *m, *n, *n0, *nlast;
1015 	int error;
1016 
1017 	KASSERT(solocked(sb->sb_so));
1018 
1019 	/*
1020 	 * XXX sbprio reserved for encoding priority of this* request:
1021 	 *  SB_PRIO_NONE --> honour normal sb limits
1022 	 *  SB_PRIO_ONESHOT_OVERFLOW --> if socket has any space,
1023 	 *	take whole chain. Intended for large requests
1024 	 *      that should be delivered atomically (all, or none).
1025 	 * SB_PRIO_OVERDRAFT -- allow a small (2*MLEN) overflow
1026 	 *       over normal socket limits, for messages indicating
1027 	 *       buffer overflow in earlier normal/lower-priority messages
1028 	 * SB_PRIO_BESTEFFORT -->  ignore limits entirely.
1029 	 *       Intended for  kernel-generated messages only.
1030 	 *        Up to generator to avoid total mbuf resource exhaustion.
1031 	 */
1032 	(void)sbprio;
1033 
1034 	if (m0 && (m0->m_flags & M_PKTHDR) == 0)
1035 		panic("sbappendaddrchain");
1036 
1037 	space = sbspace(sb);
1038 
1039 #ifdef notyet
1040 	/*
1041 	 * Enforce SB_PRIO_* limits as described above.
1042 	 */
1043 #endif
1044 
1045 	n0 = NULL;
1046 	nlast = NULL;
1047 	for (m = m0; m; m = m->m_nextpkt) {
1048 		struct mbuf *np;
1049 
1050 #ifdef MBUFTRACE
1051 		m_claimm(m, sb->sb_mowner);
1052 #endif
1053 
1054 		/* Prepend sockaddr to this record (m) of input chain m0 */
1055 	  	n = m_prepend_sockaddr(sb, m, asa);
1056 		if (n == NULL) {
1057 			error = ENOBUFS;
1058 			goto bad;
1059 		}
1060 
1061 		/* Append record (asa+m) to end of new chain n0 */
1062 		if (n0 == NULL) {
1063 			n0 = n;
1064 		} else {
1065 			nlast->m_nextpkt = n;
1066 		}
1067 		/* Keep track of last record on new chain */
1068 		nlast = n;
1069 
1070 		for (np = n; np; np = np->m_next)
1071 			sballoc(sb, np);
1072 	}
1073 
1074 	SBLASTRECORDCHK(sb, "sbappendaddrchain 1");
1075 
1076 	/* Drop the entire chain of (asa+m) records onto the socket */
1077 	SBLINKRECORDCHAIN(sb, n0, nlast);
1078 
1079 	SBLASTRECORDCHK(sb, "sbappendaddrchain 2");
1080 
1081 	for (m = nlast; m->m_next; m = m->m_next)
1082 		;
1083 	sb->sb_mbtail = m;
1084 	SBLASTMBUFCHK(sb, "sbappendaddrchain");
1085 
1086 	return (1);
1087 
1088 bad:
1089 	/*
1090 	 * On error, free the prepended addreseses. For consistency
1091 	 * with sbappendaddr(), leave it to our caller to free
1092 	 * the input record chain passed to us as m0.
1093 	 */
1094 	while ((n = n0) != NULL) {
1095 	  	struct mbuf *np;
1096 
1097 		/* Undo the sballoc() of this record */
1098 		for (np = n; np; np = np->m_next)
1099 			sbfree(sb, np);
1100 
1101 		n0 = n->m_nextpkt;	/* iterate at next prepended address */
1102 		MFREE(n, np);		/* free prepended address (not data) */
1103 	}
1104 	return 0;
1105 }
1106 
1107 
1108 int
1109 sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control)
1110 {
1111 	struct mbuf	*m, *mlast, *n;
1112 	int		space;
1113 
1114 	KASSERT(solocked(sb->sb_so));
1115 
1116 	space = 0;
1117 	if (control == 0)
1118 		panic("sbappendcontrol");
1119 	for (m = control; ; m = m->m_next) {
1120 		space += m->m_len;
1121 		MCLAIM(m, sb->sb_mowner);
1122 		if (m->m_next == 0)
1123 			break;
1124 	}
1125 	n = m;			/* save pointer to last control buffer */
1126 	for (m = m0; m; m = m->m_next) {
1127 		MCLAIM(m, sb->sb_mowner);
1128 		space += m->m_len;
1129 	}
1130 	if (space > sbspace(sb))
1131 		return (0);
1132 	n->m_next = m0;			/* concatenate data to control */
1133 
1134 	SBLASTRECORDCHK(sb, "sbappendcontrol 1");
1135 
1136 	for (m = control; m->m_next != NULL; m = m->m_next)
1137 		sballoc(sb, m);
1138 	sballoc(sb, m);
1139 	mlast = m;
1140 	SBLINKRECORD(sb, control);
1141 
1142 	sb->sb_mbtail = mlast;
1143 	SBLASTMBUFCHK(sb, "sbappendcontrol");
1144 	SBLASTRECORDCHK(sb, "sbappendcontrol 2");
1145 
1146 	return (1);
1147 }
1148 
1149 /*
1150  * Compress mbuf chain m into the socket
1151  * buffer sb following mbuf n.  If n
1152  * is null, the buffer is presumed empty.
1153  */
1154 void
1155 sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n)
1156 {
1157 	int		eor;
1158 	struct mbuf	*o;
1159 
1160 	KASSERT(solocked(sb->sb_so));
1161 
1162 	eor = 0;
1163 	while (m) {
1164 		eor |= m->m_flags & M_EOR;
1165 		if (m->m_len == 0 &&
1166 		    (eor == 0 ||
1167 		     (((o = m->m_next) || (o = n)) &&
1168 		      o->m_type == m->m_type))) {
1169 			if (sb->sb_lastrecord == m)
1170 				sb->sb_lastrecord = m->m_next;
1171 			m = m_free(m);
1172 			continue;
1173 		}
1174 		if (n && (n->m_flags & M_EOR) == 0 &&
1175 		    /* M_TRAILINGSPACE() checks buffer writeability */
1176 		    m->m_len <= MCLBYTES / 4 && /* XXX Don't copy too much */
1177 		    m->m_len <= M_TRAILINGSPACE(n) &&
1178 		    n->m_type == m->m_type) {
1179 			memcpy(mtod(n, char *) + n->m_len, mtod(m, void *),
1180 			    (unsigned)m->m_len);
1181 			n->m_len += m->m_len;
1182 			sb->sb_cc += m->m_len;
1183 			m = m_free(m);
1184 			continue;
1185 		}
1186 		if (n)
1187 			n->m_next = m;
1188 		else
1189 			sb->sb_mb = m;
1190 		sb->sb_mbtail = m;
1191 		sballoc(sb, m);
1192 		n = m;
1193 		m->m_flags &= ~M_EOR;
1194 		m = m->m_next;
1195 		n->m_next = 0;
1196 	}
1197 	if (eor) {
1198 		if (n)
1199 			n->m_flags |= eor;
1200 		else
1201 			printf("semi-panic: sbcompress\n");
1202 	}
1203 	SBLASTMBUFCHK(sb, __func__);
1204 }
1205 
1206 /*
1207  * Free all mbufs in a sockbuf.
1208  * Check that all resources are reclaimed.
1209  */
1210 void
1211 sbflush(struct sockbuf *sb)
1212 {
1213 
1214 	KASSERT(solocked(sb->sb_so));
1215 	KASSERT((sb->sb_flags & SB_LOCK) == 0);
1216 
1217 	while (sb->sb_mbcnt)
1218 		sbdrop(sb, (int)sb->sb_cc);
1219 
1220 	KASSERT(sb->sb_cc == 0);
1221 	KASSERT(sb->sb_mb == NULL);
1222 	KASSERT(sb->sb_mbtail == NULL);
1223 	KASSERT(sb->sb_lastrecord == NULL);
1224 }
1225 
1226 /*
1227  * Drop data from (the front of) a sockbuf.
1228  */
1229 void
1230 sbdrop(struct sockbuf *sb, int len)
1231 {
1232 	struct mbuf	*m, *mn, *next;
1233 
1234 	KASSERT(solocked(sb->sb_so));
1235 
1236 	next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
1237 	while (len > 0) {
1238 		if (m == 0) {
1239 			if (next == 0)
1240 				panic("sbdrop");
1241 			m = next;
1242 			next = m->m_nextpkt;
1243 			continue;
1244 		}
1245 		if (m->m_len > len) {
1246 			m->m_len -= len;
1247 			m->m_data += len;
1248 			sb->sb_cc -= len;
1249 			break;
1250 		}
1251 		len -= m->m_len;
1252 		sbfree(sb, m);
1253 		MFREE(m, mn);
1254 		m = mn;
1255 	}
1256 	while (m && m->m_len == 0) {
1257 		sbfree(sb, m);
1258 		MFREE(m, mn);
1259 		m = mn;
1260 	}
1261 	if (m) {
1262 		sb->sb_mb = m;
1263 		m->m_nextpkt = next;
1264 	} else
1265 		sb->sb_mb = next;
1266 	/*
1267 	 * First part is an inline SB_EMPTY_FIXUP().  Second part
1268 	 * makes sure sb_lastrecord is up-to-date if we dropped
1269 	 * part of the last record.
1270 	 */
1271 	m = sb->sb_mb;
1272 	if (m == NULL) {
1273 		sb->sb_mbtail = NULL;
1274 		sb->sb_lastrecord = NULL;
1275 	} else if (m->m_nextpkt == NULL)
1276 		sb->sb_lastrecord = m;
1277 }
1278 
1279 /*
1280  * Drop a record off the front of a sockbuf
1281  * and move the next record to the front.
1282  */
1283 void
1284 sbdroprecord(struct sockbuf *sb)
1285 {
1286 	struct mbuf	*m, *mn;
1287 
1288 	KASSERT(solocked(sb->sb_so));
1289 
1290 	m = sb->sb_mb;
1291 	if (m) {
1292 		sb->sb_mb = m->m_nextpkt;
1293 		do {
1294 			sbfree(sb, m);
1295 			MFREE(m, mn);
1296 		} while ((m = mn) != NULL);
1297 	}
1298 	SB_EMPTY_FIXUP(sb);
1299 }
1300 
1301 /*
1302  * Create a "control" mbuf containing the specified data
1303  * with the specified type for presentation on a socket buffer.
1304  */
1305 struct mbuf *
1306 sbcreatecontrol(void *p, int size, int type, int level)
1307 {
1308 	struct cmsghdr	*cp;
1309 	struct mbuf	*m;
1310 
1311 	if (CMSG_SPACE(size) > MCLBYTES) {
1312 		printf("sbcreatecontrol: message too large %d\n", size);
1313 		return NULL;
1314 	}
1315 
1316 	if ((m = m_get(M_DONTWAIT, MT_CONTROL)) == NULL)
1317 		return ((struct mbuf *) NULL);
1318 	if (CMSG_SPACE(size) > MLEN) {
1319 		MCLGET(m, M_DONTWAIT);
1320 		if ((m->m_flags & M_EXT) == 0) {
1321 			m_free(m);
1322 			return NULL;
1323 		}
1324 	}
1325 	cp = mtod(m, struct cmsghdr *);
1326 	memcpy(CMSG_DATA(cp), p, size);
1327 	m->m_len = CMSG_SPACE(size);
1328 	cp->cmsg_len = CMSG_LEN(size);
1329 	cp->cmsg_level = level;
1330 	cp->cmsg_type = type;
1331 	return (m);
1332 }
1333 
1334 void
1335 solockretry(struct socket *so, kmutex_t *lock)
1336 {
1337 
1338 	while (lock != so->so_lock) {
1339 		mutex_exit(lock);
1340 		lock = so->so_lock;
1341 		mutex_enter(lock);
1342 	}
1343 }
1344 
1345 bool
1346 solocked(struct socket *so)
1347 {
1348 
1349 	return mutex_owned(so->so_lock);
1350 }
1351 
1352 bool
1353 solocked2(struct socket *so1, struct socket *so2)
1354 {
1355 	kmutex_t *lock;
1356 
1357 	lock = so1->so_lock;
1358 	if (lock != so2->so_lock)
1359 		return false;
1360 	return mutex_owned(lock);
1361 }
1362 
1363 /*
1364  * Assign a default lock to a new socket.  For PRU_ATTACH, and done by
1365  * protocols that do not have special locking requirements.
1366  */
1367 void
1368 sosetlock(struct socket *so)
1369 {
1370 	kmutex_t *lock;
1371 
1372 	if (so->so_lock == NULL) {
1373 		lock = softnet_lock;
1374 		so->so_lock = lock;
1375 		mutex_obj_hold(lock);
1376 		mutex_enter(lock);
1377 	}
1378 
1379 	/* In all cases, lock must be held on return from PRU_ATTACH. */
1380 	KASSERT(solocked(so));
1381 }
1382 
1383 /*
1384  * Set lock on sockbuf sb; sleep if lock is already held.
1385  * Unless SB_NOINTR is set on sockbuf, sleep is interruptible.
1386  * Returns error without lock if sleep is interrupted.
1387  */
1388 int
1389 sblock(struct sockbuf *sb, int wf)
1390 {
1391 	struct socket *so;
1392 	kmutex_t *lock;
1393 	int error;
1394 
1395 	KASSERT(solocked(sb->sb_so));
1396 
1397 	for (;;) {
1398 		if (__predict_true((sb->sb_flags & SB_LOCK) == 0)) {
1399 			sb->sb_flags |= SB_LOCK;
1400 			return 0;
1401 		}
1402 		if (wf != M_WAITOK)
1403 			return EWOULDBLOCK;
1404 		so = sb->sb_so;
1405 		lock = so->so_lock;
1406 		if ((sb->sb_flags & SB_NOINTR) != 0) {
1407 			cv_wait(&so->so_cv, lock);
1408 			error = 0;
1409 		} else
1410 			error = cv_wait_sig(&so->so_cv, lock);
1411 		if (__predict_false(lock != so->so_lock))
1412 			solockretry(so, lock);
1413 		if (error != 0)
1414 			return error;
1415 	}
1416 }
1417 
1418 void
1419 sbunlock(struct sockbuf *sb)
1420 {
1421 	struct socket *so;
1422 
1423 	so = sb->sb_so;
1424 
1425 	KASSERT(solocked(so));
1426 	KASSERT((sb->sb_flags & SB_LOCK) != 0);
1427 
1428 	sb->sb_flags &= ~SB_LOCK;
1429 	cv_broadcast(&so->so_cv);
1430 }
1431 
1432 int
1433 sowait(struct socket *so, bool catch, int timo)
1434 {
1435 	kmutex_t *lock;
1436 	int error;
1437 
1438 	KASSERT(solocked(so));
1439 	KASSERT(catch || timo != 0);
1440 
1441 	lock = so->so_lock;
1442 	if (catch)
1443 		error = cv_timedwait_sig(&so->so_cv, lock, timo);
1444 	else
1445 		error = cv_timedwait(&so->so_cv, lock, timo);
1446 	if (__predict_false(lock != so->so_lock))
1447 		solockretry(so, lock);
1448 	return error;
1449 }
1450