xref: /dflybsd-src/sys/kern/uipc_socket2.c (revision 93bffecadc0caefc46f12b736eab0e62c2b6f42e)
1 /*
2  * Copyright (c) 2005 Jeffrey M. Hsu.  All rights reserved.
3  * Copyright (c) 1982, 1986, 1988, 1990, 1993
4  *	The Regents of the University of California.  All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 3. All advertising materials mentioning features or use of this software
15  *    must display the following acknowledgement:
16  *	This product includes software developed by the University of
17  *	California, Berkeley and its contributors.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)uipc_socket2.c	8.1 (Berkeley) 6/10/93
35  * $FreeBSD: src/sys/kern/uipc_socket2.c,v 1.55.2.17 2002/08/31 19:04:55 dwmalone Exp $
36  * $DragonFly: src/sys/kern/uipc_socket2.c,v 1.33 2008/09/02 16:17:52 dillon Exp $
37  */
38 
39 #include "opt_param.h"
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/domain.h>
43 #include <sys/file.h>	/* for maxfiles */
44 #include <sys/kernel.h>
45 #include <sys/proc.h>
46 #include <sys/malloc.h>
47 #include <sys/mbuf.h>
48 #include <sys/protosw.h>
49 #include <sys/resourcevar.h>
50 #include <sys/stat.h>
51 #include <sys/socket.h>
52 #include <sys/socketvar.h>
53 #include <sys/signalvar.h>
54 #include <sys/sysctl.h>
55 #include <sys/aio.h> /* for aio_swake proto */
56 #include <sys/event.h>
57 
58 #include <sys/thread2.h>
59 #include <sys/msgport2.h>
60 #include <sys/socketvar2.h>
61 
62 int	maxsockets;
63 
64 /*
65  * Primitive routines for operating on sockets and socket buffers
66  */
67 
68 u_long	sb_max = SB_MAX;
69 u_long	sb_max_adj =
70     SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */
71 
72 static	u_long sb_efficiency = 8;	/* parameter for sbreserve() */
73 
74 /************************************************************************
75  * signalsockbuf procedures						*
76  ************************************************************************/
77 
78 /*
79  * Wait for data to arrive at/drain from a socket buffer.
80  *
81  * NOTE: Caller must generally hold the ssb_lock (client side lock) since
82  *	 WAIT/WAKEUP only works for one client at a time.
83  *
84  * NOTE: Caller always retries whatever operation it was waiting on.
85  */
86 int
87 ssb_wait(struct signalsockbuf *ssb)
88 {
89 	uint32_t flags;
90 	int pflags;
91 	int error;
92 
93 	pflags = (ssb->ssb_flags & SSB_NOINTR) ? 0 : PCATCH;
94 
95 	for (;;) {
96 		flags = ssb->ssb_flags;
97 		cpu_ccfence();
98 
99 		/*
100 		 * WAKEUP and WAIT interlock eachother.  We can catch the
101 		 * race by checking to see if WAKEUP has already been set,
102 		 * and only setting WAIT if WAKEUP is clear.
103 		 */
104 		if (flags & SSB_WAKEUP) {
105 			if (atomic_cmpset_int(&ssb->ssb_flags, flags,
106 					      flags & ~SSB_WAKEUP)) {
107 				error = 0;
108 				break;
109 			}
110 			continue;
111 		}
112 
113 		/*
114 		 * Only set WAIT if WAKEUP is clear.
115 		 */
116 		tsleep_interlock(&ssb->ssb_cc, pflags);
117 		if (atomic_cmpset_int(&ssb->ssb_flags, flags,
118 				      flags | SSB_WAIT)) {
119 			error = tsleep(&ssb->ssb_cc, pflags | PINTERLOCKED,
120 				       "sbwait", ssb->ssb_timeo);
121 			break;
122 		}
123 	}
124 	return (error);
125 }
126 
127 /*
128  * Lock a sockbuf already known to be locked;
129  * return any error returned from sleep (EINTR).
130  */
131 int
132 _ssb_lock(struct signalsockbuf *ssb)
133 {
134 	uint32_t flags;
135 	int pflags;
136 	int error;
137 
138 	pflags = (ssb->ssb_flags & SSB_NOINTR) ? 0 : PCATCH;
139 
140 	for (;;) {
141 		flags = ssb->ssb_flags;
142 		cpu_ccfence();
143 		if (flags & SSB_LOCK) {
144 			tsleep_interlock(&ssb->ssb_flags, pflags);
145 			if (atomic_cmpset_int(&ssb->ssb_flags, flags,
146 					      flags | SSB_WANT)) {
147 				error = tsleep(&ssb->ssb_flags,
148 					       pflags | PINTERLOCKED,
149 					       "sblock", 0);
150 				if (error)
151 					break;
152 			}
153 		} else {
154 			if (atomic_cmpset_int(&ssb->ssb_flags, flags,
155 					      flags | SSB_LOCK)) {
156 				lwkt_gettoken(&ssb->ssb_token);
157 				error = 0;
158 				break;
159 			}
160 		}
161 	}
162 	return (error);
163 }
164 
165 /*
166  * This does the same for sockbufs.  Note that the xsockbuf structure,
167  * since it is always embedded in a socket, does not include a self
168  * pointer nor a length.  We make this entry point public in case
169  * some other mechanism needs it.
170  */
171 void
172 ssbtoxsockbuf(struct signalsockbuf *ssb, struct xsockbuf *xsb)
173 {
174 	xsb->sb_cc = ssb->ssb_cc;
175 	xsb->sb_hiwat = ssb->ssb_hiwat;
176 	xsb->sb_mbcnt = ssb->ssb_mbcnt;
177 	xsb->sb_mbmax = ssb->ssb_mbmax;
178 	xsb->sb_lowat = ssb->ssb_lowat;
179 	xsb->sb_flags = ssb->ssb_flags;
180 	xsb->sb_timeo = ssb->ssb_timeo;
181 }
182 
183 
184 /************************************************************************
185  * Procedures which manipulate socket state flags, wakeups, etc.	*
186  ************************************************************************
187  *
188  * Normal sequence from the active (originating) side is that
189  * soisconnecting() is called during processing of connect() call, resulting
190  * in an eventual call to soisconnected() if/when the connection is
191  * established.  When the connection is torn down soisdisconnecting() is
192  * called during processing of disconnect() call, and soisdisconnected() is
193  * called when the connection to the peer is totally severed.
194  *
195  * The semantics of these routines are such that connectionless protocols
196  * can call soisconnected() and soisdisconnected() only, bypassing the
197  * in-progress calls when setting up a ``connection'' takes no time.
198  *
199  * From the passive side, a socket is created with two queues of sockets:
200  * so_incomp for connections in progress and so_comp for connections
201  * already made and awaiting user acceptance.  As a protocol is preparing
202  * incoming connections, it creates a socket structure queued on so_incomp
203  * by calling sonewconn().  When the connection is established,
204  * soisconnected() is called, and transfers the socket structure to so_comp,
205  * making it available to accept().
206  *
207  * If a socket is closed with sockets on either so_incomp or so_comp, these
208  * sockets are dropped.
209  *
210  * If higher level protocols are implemented in the kernel, the wakeups
211  * done here will sometimes cause software-interrupt process scheduling.
212  */
213 
214 void
215 soisconnecting(struct socket *so)
216 {
217 	soclrstate(so, SS_ISCONNECTED | SS_ISDISCONNECTING);
218 	sosetstate(so, SS_ISCONNECTING);
219 }
220 
221 void
222 soisconnected(struct socket *so)
223 {
224 	struct socket *head = so->so_head;
225 
226 	soclrstate(so, SS_ISCONNECTING | SS_ISDISCONNECTING | SS_ISCONFIRMING);
227 	sosetstate(so, SS_ISCONNECTED);
228 	if (head && (so->so_state & SS_INCOMP)) {
229 		if ((so->so_options & SO_ACCEPTFILTER) != 0) {
230 			so->so_upcall = head->so_accf->so_accept_filter->accf_callback;
231 			so->so_upcallarg = head->so_accf->so_accept_filter_arg;
232 			atomic_set_int(&so->so_rcv.ssb_flags, SSB_UPCALL);
233 			so->so_options &= ~SO_ACCEPTFILTER;
234 			so->so_upcall(so, so->so_upcallarg, 0);
235 			return;
236 		}
237 
238 		/*
239 		 * Listen socket are not per-cpu.
240 		 */
241 		lwkt_gettoken(&head->so_rcv.ssb_token);
242 		TAILQ_REMOVE(&head->so_incomp, so, so_list);
243 		head->so_incqlen--;
244 		soclrstate(so, SS_INCOMP);
245 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
246 		head->so_qlen++;
247 		sosetstate(so, SS_COMP);
248 		lwkt_reltoken(&head->so_rcv.ssb_token);
249 
250 		sorwakeup(head);
251 		wakeup_one(&head->so_timeo);
252 	} else {
253 		wakeup(&so->so_timeo);
254 		sorwakeup(so);
255 		sowwakeup(so);
256 	}
257 }
258 
259 void
260 soisdisconnecting(struct socket *so)
261 {
262 	soclrstate(so, SS_ISCONNECTING);
263 	sosetstate(so, SS_ISDISCONNECTING | SS_CANTRCVMORE | SS_CANTSENDMORE);
264 	wakeup((caddr_t)&so->so_timeo);
265 	sowwakeup(so);
266 	sorwakeup(so);
267 }
268 
269 void
270 soisdisconnected(struct socket *so)
271 {
272 	soclrstate(so, SS_ISCONNECTING | SS_ISCONNECTED | SS_ISDISCONNECTING);
273 	sosetstate(so, SS_CANTRCVMORE | SS_CANTSENDMORE | SS_ISDISCONNECTED);
274 	wakeup((caddr_t)&so->so_timeo);
275 	sbdrop(&so->so_snd.sb, so->so_snd.ssb_cc);
276 	sowwakeup(so);
277 	sorwakeup(so);
278 }
279 
280 void
281 soisreconnecting(struct socket *so)
282 {
283         soclrstate(so, SS_ISDISCONNECTING | SS_ISDISCONNECTED |
284 		       SS_CANTRCVMORE | SS_CANTSENDMORE);
285 	sosetstate(so, SS_ISCONNECTING);
286 }
287 
288 void
289 soisreconnected(struct socket *so)
290 {
291 	soclrstate(so, SS_ISDISCONNECTED | SS_CANTRCVMORE | SS_CANTSENDMORE);
292 	soisconnected(so);
293 }
294 
295 /*
296  * Set or change the message port a socket receives commands on.
297  *
298  * XXX
299  */
300 void
301 sosetport(struct socket *so, lwkt_port_t port)
302 {
303 	so->so_port = port;
304 }
305 
306 /*
307  * When an attempt at a new connection is noted on a socket
308  * which accepts connections, sonewconn is called.  If the
309  * connection is possible (subject to space constraints, etc.)
310  * then we allocate a new structure, propoerly linked into the
311  * data structure of the original socket, and return this.
312  * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
313  *
314  * The new socket is returned with one ref and so_pcb assigned.
315  * The reference is implied by so_pcb.
316  */
317 struct socket *
318 sonewconn(struct socket *head, int connstatus)
319 {
320 	struct socket *so;
321 	struct socket *sp;
322 	struct pru_attach_info ai;
323 
324 	if (head->so_qlen > 3 * head->so_qlimit / 2)
325 		return (NULL);
326 	so = soalloc(1);
327 	if (so == NULL)
328 		return (NULL);
329 	if ((head->so_options & SO_ACCEPTFILTER) != 0)
330 		connstatus = 0;
331 	so->so_head = head;
332 	so->so_type = head->so_type;
333 	so->so_options = head->so_options &~ SO_ACCEPTCONN;
334 	so->so_linger = head->so_linger;
335 
336 	/*
337 	 * NOTE: Clearing NOFDREF implies referencing the so with
338 	 *	 soreference().
339 	 */
340 	so->so_state = head->so_state | SS_NOFDREF;
341 	so->so_proto = head->so_proto;
342 	so->so_cred = crhold(head->so_cred);
343 	ai.sb_rlimit = NULL;
344 	ai.p_ucred = NULL;
345 	ai.fd_rdir = NULL;		/* jail code cruft XXX JH */
346 
347 	/*
348 	 * Reserve space and call pru_attach.  We can directl call the
349 	 * function since we're already in the protocol thread.
350 	 */
351 	if (soreserve(so, head->so_snd.ssb_hiwat,
352 		      head->so_rcv.ssb_hiwat, NULL) ||
353 	    (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, &ai)) {
354 		so->so_head = NULL;
355 		sofree(so);		/* remove implied pcb ref */
356 		return (NULL);
357 	}
358 	KKASSERT(so->so_refs == 2);	/* attach + our base ref */
359 	sofree(so);
360 	KKASSERT(so->so_port != NULL);
361 	so->so_rcv.ssb_lowat = head->so_rcv.ssb_lowat;
362 	so->so_snd.ssb_lowat = head->so_snd.ssb_lowat;
363 	so->so_rcv.ssb_timeo = head->so_rcv.ssb_timeo;
364 	so->so_snd.ssb_timeo = head->so_snd.ssb_timeo;
365 	so->so_rcv.ssb_flags |= head->so_rcv.ssb_flags &
366 				(SSB_AUTOSIZE | SSB_AUTOLOWAT);
367 	so->so_snd.ssb_flags |= head->so_snd.ssb_flags &
368 				(SSB_AUTOSIZE | SSB_AUTOLOWAT);
369 	lwkt_gettoken(&head->so_rcv.ssb_token);
370 	if (connstatus) {
371 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
372 		sosetstate(so, SS_COMP);
373 		head->so_qlen++;
374 	} else {
375 		if (head->so_incqlen > head->so_qlimit) {
376 			sp = TAILQ_FIRST(&head->so_incomp);
377 			TAILQ_REMOVE(&head->so_incomp, sp, so_list);
378 			head->so_incqlen--;
379 			soclrstate(sp, SS_INCOMP);
380 			sp->so_head = NULL;
381 			soaborta(sp);
382 		}
383 		TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
384 		sosetstate(so, SS_INCOMP);
385 		head->so_incqlen++;
386 	}
387 	lwkt_reltoken(&head->so_rcv.ssb_token);
388 	if (connstatus) {
389 		sorwakeup(head);
390 		wakeup((caddr_t)&head->so_timeo);
391 		sosetstate(so, connstatus);
392 	}
393 	return (so);
394 }
395 
396 /*
397  * Socantsendmore indicates that no more data will be sent on the
398  * socket; it would normally be applied to a socket when the user
399  * informs the system that no more data is to be sent, by the protocol
400  * code (in case PRU_SHUTDOWN).  Socantrcvmore indicates that no more data
401  * will be received, and will normally be applied to the socket by a
402  * protocol when it detects that the peer will send no more data.
403  * Data queued for reading in the socket may yet be read.
404  */
405 void
406 socantsendmore(struct socket *so)
407 {
408 	sosetstate(so, SS_CANTSENDMORE);
409 	sowwakeup(so);
410 }
411 
412 void
413 socantrcvmore(struct socket *so)
414 {
415 	sosetstate(so, SS_CANTRCVMORE);
416 	sorwakeup(so);
417 }
418 
419 /*
420  * Wakeup processes waiting on a socket buffer.  Do asynchronous notification
421  * via SIGIO if the socket has the SS_ASYNC flag set.
422  *
423  * For users waiting on send/recv try to avoid unnecessary context switch
424  * thrashing.  Particularly for senders of large buffers (needs to be
425  * extended to sel and aio? XXX)
426  */
427 void
428 sowakeup(struct socket *so, struct signalsockbuf *ssb)
429 {
430 	struct kqinfo *kqinfo = &ssb->ssb_kq;
431 	uint32_t flags;
432 
433 	/*
434 	 * Check conditions, set the WAKEUP flag, and clear and signal if
435 	 * the WAIT flag is found to be set.  This interlocks against the
436 	 * client side.
437 	 */
438 	for (;;) {
439 		flags = ssb->ssb_flags;
440 		cpu_ccfence();
441 
442 		if ((ssb == &so->so_snd && ssb_space(ssb) >= ssb->ssb_lowat) ||
443 		    (ssb == &so->so_rcv && ssb->ssb_cc >= ssb->ssb_lowat) ||
444 		    (ssb == &so->so_snd && (so->so_state & SS_CANTSENDMORE)) ||
445 		    (ssb == &so->so_rcv && (so->so_state & SS_CANTRCVMORE))
446 		) {
447 			if (atomic_cmpset_int(&ssb->ssb_flags, flags,
448 					  (flags | SSB_WAKEUP) & ~SSB_WAIT)) {
449 				if (flags & SSB_WAIT)
450 					wakeup(&ssb->ssb_cc);
451 				break;
452 			}
453 		} else {
454 			break;
455 		}
456 	}
457 
458 	/*
459 	 * Misc other events
460 	 */
461 	if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL)
462 		pgsigio(so->so_sigio, SIGIO, 0);
463 	if (ssb->ssb_flags & SSB_UPCALL)
464 		(*so->so_upcall)(so, so->so_upcallarg, MB_DONTWAIT);
465 	if (ssb->ssb_flags & SSB_AIO)
466 		aio_swake(so, ssb);
467 	KNOTE(&kqinfo->ki_note, 0);
468 	if (ssb->ssb_flags & SSB_MEVENT) {
469 		struct netmsg_so_notify *msg, *nmsg;
470 
471 		lwkt_gettoken(&kq_token);
472 		TAILQ_FOREACH_MUTABLE(msg, &kqinfo->ki_mlist, nm_list, nmsg) {
473 			if (msg->nm_predicate(&msg->nm_netmsg)) {
474 				TAILQ_REMOVE(&kqinfo->ki_mlist, msg, nm_list);
475 				lwkt_replymsg(&msg->nm_netmsg.nm_lmsg,
476 					      msg->nm_netmsg.nm_lmsg.ms_error);
477 			}
478 		}
479 		if (TAILQ_EMPTY(&ssb->ssb_kq.ki_mlist))
480 			atomic_clear_int(&ssb->ssb_flags, SSB_MEVENT);
481 		lwkt_reltoken(&kq_token);
482 	}
483 }
484 
485 /*
486  * Socket buffer (struct signalsockbuf) utility routines.
487  *
488  * Each socket contains two socket buffers: one for sending data and
489  * one for receiving data.  Each buffer contains a queue of mbufs,
490  * information about the number of mbufs and amount of data in the
491  * queue, and other fields allowing kevent()/select()/poll() statements
492  * and notification on data availability to be implemented.
493  *
494  * Data stored in a socket buffer is maintained as a list of records.
495  * Each record is a list of mbufs chained together with the m_next
496  * field.  Records are chained together with the m_nextpkt field. The upper
497  * level routine soreceive() expects the following conventions to be
498  * observed when placing information in the receive buffer:
499  *
500  * 1. If the protocol requires each message be preceded by the sender's
501  *    name, then a record containing that name must be present before
502  *    any associated data (mbuf's must be of type MT_SONAME).
503  * 2. If the protocol supports the exchange of ``access rights'' (really
504  *    just additional data associated with the message), and there are
505  *    ``rights'' to be received, then a record containing this data
506  *    should be present (mbuf's must be of type MT_RIGHTS).
507  * 3. If a name or rights record exists, then it must be followed by
508  *    a data record, perhaps of zero length.
509  *
510  * Before using a new socket structure it is first necessary to reserve
511  * buffer space to the socket, by calling sbreserve().  This should commit
512  * some of the available buffer space in the system buffer pool for the
513  * socket (currently, it does nothing but enforce limits).  The space
514  * should be released by calling ssb_release() when the socket is destroyed.
515  */
516 int
517 soreserve(struct socket *so, u_long sndcc, u_long rcvcc, struct rlimit *rl)
518 {
519 	if (so->so_snd.ssb_lowat == 0)
520 		atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOLOWAT);
521 	if (ssb_reserve(&so->so_snd, sndcc, so, rl) == 0)
522 		goto bad;
523 	if (ssb_reserve(&so->so_rcv, rcvcc, so, rl) == 0)
524 		goto bad2;
525 	if (so->so_rcv.ssb_lowat == 0)
526 		so->so_rcv.ssb_lowat = 1;
527 	if (so->so_snd.ssb_lowat == 0)
528 		so->so_snd.ssb_lowat = MCLBYTES;
529 	if (so->so_snd.ssb_lowat > so->so_snd.ssb_hiwat)
530 		so->so_snd.ssb_lowat = so->so_snd.ssb_hiwat;
531 	return (0);
532 bad2:
533 	ssb_release(&so->so_snd, so);
534 bad:
535 	return (ENOBUFS);
536 }
537 
538 static int
539 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS)
540 {
541 	int error = 0;
542 	u_long old_sb_max = sb_max;
543 
544 	error = SYSCTL_OUT(req, arg1, sizeof(int));
545 	if (error || !req->newptr)
546 		return (error);
547 	error = SYSCTL_IN(req, arg1, sizeof(int));
548 	if (error)
549 		return (error);
550 	if (sb_max < MSIZE + MCLBYTES) {
551 		sb_max = old_sb_max;
552 		return (EINVAL);
553 	}
554 	sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES);
555 	return (0);
556 }
557 
558 /*
559  * Allot mbufs to a signalsockbuf.
560  *
561  * Attempt to scale mbmax so that mbcnt doesn't become limiting
562  * if buffering efficiency is near the normal case.
563  *
564  * sb_max only applies to user-sockets (where rl != NULL).  It does
565  * not apply to kernel sockets or kernel-controlled sockets.  Note
566  * that NFS overrides the sockbuf limits created when nfsd creates
567  * a socket.
568  */
569 int
570 ssb_reserve(struct signalsockbuf *ssb, u_long cc, struct socket *so,
571 	    struct rlimit *rl)
572 {
573 	/*
574 	 * rl will only be NULL when we're in an interrupt (eg, in tcp_input)
575 	 * or when called from netgraph (ie, ngd_attach)
576 	 */
577 	if (rl && cc > sb_max_adj)
578 		cc = sb_max_adj;
579 	if (!chgsbsize(so->so_cred->cr_uidinfo, &ssb->ssb_hiwat, cc,
580 		       rl ? rl->rlim_cur : RLIM_INFINITY)) {
581 		return (0);
582 	}
583 	if (rl)
584 		ssb->ssb_mbmax = min(cc * sb_efficiency, sb_max);
585 	else
586 		ssb->ssb_mbmax = cc * sb_efficiency;
587 
588 	/*
589 	 * AUTOLOWAT is set on send buffers and prevents large writes
590 	 * from generating a huge number of context switches.
591 	 */
592 	if (ssb->ssb_flags & SSB_AUTOLOWAT) {
593 		ssb->ssb_lowat = ssb->ssb_hiwat / 2;
594 		if (ssb->ssb_lowat < MCLBYTES)
595 			ssb->ssb_lowat = MCLBYTES;
596 	}
597 	if (ssb->ssb_lowat > ssb->ssb_hiwat)
598 		ssb->ssb_lowat = ssb->ssb_hiwat;
599 	return (1);
600 }
601 
602 /*
603  * Free mbufs held by a socket, and reserved mbuf space.
604  */
605 void
606 ssb_release(struct signalsockbuf *ssb, struct socket *so)
607 {
608 	sbflush(&ssb->sb);
609 	(void)chgsbsize(so->so_cred->cr_uidinfo, &ssb->ssb_hiwat, 0,
610 	    RLIM_INFINITY);
611 	ssb->ssb_mbmax = 0;
612 }
613 
614 /*
615  * Some routines that return EOPNOTSUPP for entry points that are not
616  * supported by a protocol.  Fill in as needed.
617  */
618 int
619 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
620 {
621 	return EOPNOTSUPP;
622 }
623 
624 int
625 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
626 {
627 	return EOPNOTSUPP;
628 }
629 
630 int
631 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
632 {
633 	return EOPNOTSUPP;
634 }
635 
636 int
637 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
638 {
639 	return EOPNOTSUPP;
640 }
641 
642 int
643 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
644 		    struct ifnet *ifp, struct thread *td)
645 {
646 	return EOPNOTSUPP;
647 }
648 
649 int
650 pru_disconnect_notsupp(struct socket *so)
651 {
652 	return EOPNOTSUPP;
653 }
654 
655 int
656 pru_listen_notsupp(struct socket *so, struct thread *td)
657 {
658 	return EOPNOTSUPP;
659 }
660 
661 int
662 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
663 {
664 	return EOPNOTSUPP;
665 }
666 
667 int
668 pru_rcvd_notsupp(struct socket *so, int flags)
669 {
670 	return EOPNOTSUPP;
671 }
672 
673 int
674 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
675 {
676 	return EOPNOTSUPP;
677 }
678 
679 int
680 pru_shutdown_notsupp(struct socket *so)
681 {
682 	return EOPNOTSUPP;
683 }
684 
685 int
686 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
687 {
688 	return EOPNOTSUPP;
689 }
690 
691 int
692 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
693 	   struct mbuf *top, struct mbuf *control, int flags,
694 	   struct thread *td)
695 {
696 	if (top)
697 		m_freem(top);
698 	if (control)
699 		m_freem(control);
700 	return (EOPNOTSUPP);
701 }
702 
703 int
704 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
705 		      struct uio *uio, struct sockbuf *sio,
706 		      struct mbuf **controlp, int *flagsp)
707 {
708 	return (EOPNOTSUPP);
709 }
710 
711 int
712 pru_ctloutput_notsupp(struct socket *so, struct sockopt *sopt)
713 {
714 	return (EOPNOTSUPP);
715 }
716 
717 /*
718  * This isn't really a ``null'' operation, but it's the default one
719  * and doesn't do anything destructive.
720  */
721 int
722 pru_sense_null(struct socket *so, struct stat *sb)
723 {
724 	sb->st_blksize = so->so_snd.ssb_hiwat;
725 	return 0;
726 }
727 
728 /*
729  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.  Callers
730  * of this routine assume that it always succeeds, so we have to use a
731  * blockable allocation even though we might be called from a critical thread.
732  */
733 struct sockaddr *
734 dup_sockaddr(const struct sockaddr *sa)
735 {
736 	struct sockaddr *sa2;
737 
738 	sa2 = kmalloc(sa->sa_len, M_SONAME, M_INTWAIT);
739 	bcopy(sa, sa2, sa->sa_len);
740 	return (sa2);
741 }
742 
743 /*
744  * Create an external-format (``xsocket'') structure using the information
745  * in the kernel-format socket structure pointed to by so.  This is done
746  * to reduce the spew of irrelevant information over this interface,
747  * to isolate user code from changes in the kernel structure, and
748  * potentially to provide information-hiding if we decide that
749  * some of this information should be hidden from users.
750  */
751 void
752 sotoxsocket(struct socket *so, struct xsocket *xso)
753 {
754 	xso->xso_len = sizeof *xso;
755 	xso->xso_so = so;
756 	xso->so_type = so->so_type;
757 	xso->so_options = so->so_options;
758 	xso->so_linger = so->so_linger;
759 	xso->so_state = so->so_state;
760 	xso->so_pcb = so->so_pcb;
761 	xso->xso_protocol = so->so_proto->pr_protocol;
762 	xso->xso_family = so->so_proto->pr_domain->dom_family;
763 	xso->so_qlen = so->so_qlen;
764 	xso->so_incqlen = so->so_incqlen;
765 	xso->so_qlimit = so->so_qlimit;
766 	xso->so_timeo = so->so_timeo;
767 	xso->so_error = so->so_error;
768 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
769 	xso->so_oobmark = so->so_oobmark;
770 	ssbtoxsockbuf(&so->so_snd, &xso->so_snd);
771 	ssbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
772 	xso->so_uid = so->so_cred->cr_uid;
773 }
774 
775 /*
776  * Here is the definition of some of the basic objects in the kern.ipc
777  * branch of the MIB.
778  */
779 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
780 
781 /*
782  * This takes the place of kern.maxsockbuf, which moved to kern.ipc.
783  *
784  * NOTE! sb_max only applies to user-created socket buffers.
785  */
786 static int dummy;
787 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, "");
788 SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_INT|CTLFLAG_RW,
789     &sb_max, 0, sysctl_handle_sb_max, "I", "Maximum socket buffer size");
790 SYSCTL_INT(_kern_ipc, OID_AUTO, maxsockets, CTLFLAG_RD,
791     &maxsockets, 0, "Maximum number of sockets available");
792 SYSCTL_INT(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW,
793     &sb_efficiency, 0, "");
794 
795 /*
796  * Initialize maxsockets
797  */
798 static void
799 init_maxsockets(void *ignored)
800 {
801     TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
802     maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
803 }
804 SYSINIT(param, SI_BOOT1_TUNABLES, SI_ORDER_ANY,
805 	init_maxsockets, NULL);
806 
807