xref: /dflybsd-src/sys/kern/uipc_socket2.c (revision 30e3ae034c9501c319c415ada6d5e23372649c88)
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 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
245 		head->so_qlen++;
246 		sosetstate(so, SS_COMP);
247 		soclrstate(so, SS_INCOMP);
248 		lwkt_reltoken(&head->so_rcv.ssb_token);
249 
250 		/*
251 		 * XXX head may be on a different protocol thread.
252 		 *     sorwakeup()->sowakeup() is hacked atm.
253 		 */
254 		sorwakeup(head);
255 		wakeup_one(&head->so_timeo);
256 	} else {
257 		wakeup(&so->so_timeo);
258 		sorwakeup(so);
259 		sowwakeup(so);
260 	}
261 }
262 
263 void
264 soisdisconnecting(struct socket *so)
265 {
266 	soclrstate(so, SS_ISCONNECTING);
267 	sosetstate(so, SS_ISDISCONNECTING | SS_CANTRCVMORE | SS_CANTSENDMORE);
268 	wakeup((caddr_t)&so->so_timeo);
269 	sowwakeup(so);
270 	sorwakeup(so);
271 }
272 
273 void
274 soisdisconnected(struct socket *so)
275 {
276 	soclrstate(so, SS_ISCONNECTING | SS_ISCONNECTED | SS_ISDISCONNECTING);
277 	sosetstate(so, SS_CANTRCVMORE | SS_CANTSENDMORE | SS_ISDISCONNECTED);
278 	wakeup((caddr_t)&so->so_timeo);
279 	sbdrop(&so->so_snd.sb, so->so_snd.ssb_cc);
280 	sowwakeup(so);
281 	sorwakeup(so);
282 }
283 
284 void
285 soisreconnecting(struct socket *so)
286 {
287         soclrstate(so, SS_ISDISCONNECTING | SS_ISDISCONNECTED |
288 		       SS_CANTRCVMORE | SS_CANTSENDMORE);
289 	sosetstate(so, SS_ISCONNECTING);
290 }
291 
292 void
293 soisreconnected(struct socket *so)
294 {
295 	soclrstate(so, SS_ISDISCONNECTED | SS_CANTRCVMORE | SS_CANTSENDMORE);
296 	soisconnected(so);
297 }
298 
299 /*
300  * Set or change the message port a socket receives commands on.
301  *
302  * XXX
303  */
304 void
305 sosetport(struct socket *so, lwkt_port_t port)
306 {
307 	so->so_port = port;
308 }
309 
310 /*
311  * When an attempt at a new connection is noted on a socket
312  * which accepts connections, sonewconn is called.  If the
313  * connection is possible (subject to space constraints, etc.)
314  * then we allocate a new structure, propoerly linked into the
315  * data structure of the original socket, and return this.
316  * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
317  *
318  * The new socket is returned with one ref and so_pcb assigned.
319  * The reference is implied by so_pcb.
320  */
321 struct socket *
322 sonewconn(struct socket *head, int connstatus)
323 {
324 	struct socket *so;
325 	struct socket *sp;
326 	struct pru_attach_info ai;
327 
328 	if (head->so_qlen > 3 * head->so_qlimit / 2)
329 		return (NULL);
330 	so = soalloc(1);
331 	if (so == NULL)
332 		return (NULL);
333 
334 	/*
335 	 * Set the port prior to attaching the inpcb to the current
336 	 * cpu's protocol thread (which should be the current thread
337 	 * but might not be in all cases).  This serializes any pcb ops
338 	 * which occur to our cpu allowing us to complete the attachment
339 	 * without racing anything.
340 	 */
341 	sosetport(so, cpu_portfn(mycpu->gd_cpuid));
342 	if ((head->so_options & SO_ACCEPTFILTER) != 0)
343 		connstatus = 0;
344 	so->so_head = head;
345 	so->so_type = head->so_type;
346 	so->so_options = head->so_options &~ SO_ACCEPTCONN;
347 	so->so_linger = head->so_linger;
348 
349 	/*
350 	 * NOTE: Clearing NOFDREF implies referencing the so with
351 	 *	 soreference().
352 	 */
353 	so->so_state = head->so_state | SS_NOFDREF | SS_ASSERTINPROG;
354 	so->so_proto = head->so_proto;
355 	so->so_cred = crhold(head->so_cred);
356 	ai.sb_rlimit = NULL;
357 	ai.p_ucred = NULL;
358 	ai.fd_rdir = NULL;		/* jail code cruft XXX JH */
359 
360 	/*
361 	 * Reserve space and call pru_attach.  We can directl call the
362 	 * function since we're already in the protocol thread.
363 	 */
364 	if (soreserve(so, head->so_snd.ssb_hiwat,
365 		      head->so_rcv.ssb_hiwat, NULL) ||
366 	    (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, &ai)) {
367 		so->so_head = NULL;
368 		soclrstate(so, SS_ASSERTINPROG);
369 		sofree(so);		/* remove implied pcb ref */
370 		return (NULL);
371 	}
372 	KKASSERT(so->so_refs == 2);	/* attach + our base ref */
373 	sofree(so);
374 	KKASSERT(so->so_port != NULL);
375 	so->so_rcv.ssb_lowat = head->so_rcv.ssb_lowat;
376 	so->so_snd.ssb_lowat = head->so_snd.ssb_lowat;
377 	so->so_rcv.ssb_timeo = head->so_rcv.ssb_timeo;
378 	so->so_snd.ssb_timeo = head->so_snd.ssb_timeo;
379 	so->so_rcv.ssb_flags |= head->so_rcv.ssb_flags &
380 				(SSB_AUTOSIZE | SSB_AUTOLOWAT);
381 	so->so_snd.ssb_flags |= head->so_snd.ssb_flags &
382 				(SSB_AUTOSIZE | SSB_AUTOLOWAT);
383 	lwkt_gettoken(&head->so_rcv.ssb_token);
384 	if (connstatus) {
385 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
386 		sosetstate(so, SS_COMP);
387 		head->so_qlen++;
388 	} else {
389 		if (head->so_incqlen > head->so_qlimit) {
390 			sp = TAILQ_FIRST(&head->so_incomp);
391 			TAILQ_REMOVE(&head->so_incomp, sp, so_list);
392 			head->so_incqlen--;
393 			soclrstate(sp, SS_INCOMP);
394 			sp->so_head = NULL;
395 			soaborta(sp);
396 		}
397 		TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
398 		sosetstate(so, SS_INCOMP);
399 		head->so_incqlen++;
400 	}
401 	lwkt_reltoken(&head->so_rcv.ssb_token);
402 	if (connstatus) {
403 		/*
404 		 * XXX head may be on a different protocol thread.
405 		 *     sorwakeup()->sowakeup() is hacked atm.
406 		 */
407 		sorwakeup(head);
408 		wakeup((caddr_t)&head->so_timeo);
409 		sosetstate(so, connstatus);
410 	}
411 	soclrstate(so, SS_ASSERTINPROG);
412 	return (so);
413 }
414 
415 /*
416  * Socantsendmore indicates that no more data will be sent on the
417  * socket; it would normally be applied to a socket when the user
418  * informs the system that no more data is to be sent, by the protocol
419  * code (in case PRU_SHUTDOWN).  Socantrcvmore indicates that no more data
420  * will be received, and will normally be applied to the socket by a
421  * protocol when it detects that the peer will send no more data.
422  * Data queued for reading in the socket may yet be read.
423  */
424 void
425 socantsendmore(struct socket *so)
426 {
427 	sosetstate(so, SS_CANTSENDMORE);
428 	sowwakeup(so);
429 }
430 
431 void
432 socantrcvmore(struct socket *so)
433 {
434 	sosetstate(so, SS_CANTRCVMORE);
435 	sorwakeup(so);
436 }
437 
438 /*
439  * Wakeup processes waiting on a socket buffer.  Do asynchronous notification
440  * via SIGIO if the socket has the SS_ASYNC flag set.
441  *
442  * For users waiting on send/recv try to avoid unnecessary context switch
443  * thrashing.  Particularly for senders of large buffers (needs to be
444  * extended to sel and aio? XXX)
445  *
446  * WARNING!  Can be called on a foreign socket from the wrong protocol
447  *	     thread.  aka is called on the 'head' listen socket when
448  *	     a new connection comes in.
449  */
450 void
451 sowakeup(struct socket *so, struct signalsockbuf *ssb)
452 {
453 	struct kqinfo *kqinfo = &ssb->ssb_kq;
454 	uint32_t flags;
455 
456 	/*
457 	 * Check conditions, set the WAKEUP flag, and clear and signal if
458 	 * the WAIT flag is found to be set.  This interlocks against the
459 	 * client side.
460 	 */
461 	for (;;) {
462 		flags = ssb->ssb_flags;
463 		cpu_ccfence();
464 
465 		if ((ssb == &so->so_snd && ssb_space(ssb) >= ssb->ssb_lowat) ||
466 		    (ssb == &so->so_rcv && ssb->ssb_cc >= ssb->ssb_lowat) ||
467 		    (ssb == &so->so_snd && (so->so_state & SS_CANTSENDMORE)) ||
468 		    (ssb == &so->so_rcv && (so->so_state & SS_CANTRCVMORE))
469 		) {
470 			if (atomic_cmpset_int(&ssb->ssb_flags, flags,
471 					  (flags | SSB_WAKEUP) & ~SSB_WAIT)) {
472 				if (flags & SSB_WAIT)
473 					wakeup(&ssb->ssb_cc);
474 				break;
475 			}
476 		} else {
477 			break;
478 		}
479 	}
480 
481 	/*
482 	 * Misc other events
483 	 */
484 	if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL)
485 		pgsigio(so->so_sigio, SIGIO, 0);
486 	if (ssb->ssb_flags & SSB_UPCALL)
487 		(*so->so_upcall)(so, so->so_upcallarg, MB_DONTWAIT);
488 	if (ssb->ssb_flags & SSB_AIO)
489 		aio_swake(so, ssb);
490 	KNOTE(&kqinfo->ki_note, 0);
491 
492 	/*
493 	 * This is a bit of a hack.  Multiple threads can wind up scanning
494 	 * ki_mlist concurrently due to the fact that this function can be
495 	 * called on a foreign socket, so we can't afford to block here.
496 	 */
497 	if (ssb->ssb_flags & SSB_MEVENT) {
498 		struct netmsg_so_notify *msg, *nmsg;
499 
500 		lwkt_gettoken(&kq_token);
501 		lwkt_gettoken_hard(&ssb->ssb_token);
502 		TAILQ_FOREACH_MUTABLE(msg, &kqinfo->ki_mlist, nm_list, nmsg) {
503 			if (msg->nm_predicate(&msg->nm_netmsg)) {
504 				TAILQ_REMOVE(&kqinfo->ki_mlist, msg, nm_list);
505 				lwkt_replymsg(&msg->nm_netmsg.nm_lmsg,
506 					      msg->nm_netmsg.nm_lmsg.ms_error);
507 			}
508 		}
509 		if (TAILQ_EMPTY(&ssb->ssb_kq.ki_mlist))
510 			atomic_clear_int(&ssb->ssb_flags, SSB_MEVENT);
511 		lwkt_reltoken_hard(&ssb->ssb_token);
512 		lwkt_reltoken(&kq_token);
513 	}
514 }
515 
516 /*
517  * Socket buffer (struct signalsockbuf) utility routines.
518  *
519  * Each socket contains two socket buffers: one for sending data and
520  * one for receiving data.  Each buffer contains a queue of mbufs,
521  * information about the number of mbufs and amount of data in the
522  * queue, and other fields allowing kevent()/select()/poll() statements
523  * and notification on data availability to be implemented.
524  *
525  * Data stored in a socket buffer is maintained as a list of records.
526  * Each record is a list of mbufs chained together with the m_next
527  * field.  Records are chained together with the m_nextpkt field. The upper
528  * level routine soreceive() expects the following conventions to be
529  * observed when placing information in the receive buffer:
530  *
531  * 1. If the protocol requires each message be preceded by the sender's
532  *    name, then a record containing that name must be present before
533  *    any associated data (mbuf's must be of type MT_SONAME).
534  * 2. If the protocol supports the exchange of ``access rights'' (really
535  *    just additional data associated with the message), and there are
536  *    ``rights'' to be received, then a record containing this data
537  *    should be present (mbuf's must be of type MT_RIGHTS).
538  * 3. If a name or rights record exists, then it must be followed by
539  *    a data record, perhaps of zero length.
540  *
541  * Before using a new socket structure it is first necessary to reserve
542  * buffer space to the socket, by calling sbreserve().  This should commit
543  * some of the available buffer space in the system buffer pool for the
544  * socket (currently, it does nothing but enforce limits).  The space
545  * should be released by calling ssb_release() when the socket is destroyed.
546  */
547 int
548 soreserve(struct socket *so, u_long sndcc, u_long rcvcc, struct rlimit *rl)
549 {
550 	if (so->so_snd.ssb_lowat == 0)
551 		atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOLOWAT);
552 	if (ssb_reserve(&so->so_snd, sndcc, so, rl) == 0)
553 		goto bad;
554 	if (ssb_reserve(&so->so_rcv, rcvcc, so, rl) == 0)
555 		goto bad2;
556 	if (so->so_rcv.ssb_lowat == 0)
557 		so->so_rcv.ssb_lowat = 1;
558 	if (so->so_snd.ssb_lowat == 0)
559 		so->so_snd.ssb_lowat = MCLBYTES;
560 	if (so->so_snd.ssb_lowat > so->so_snd.ssb_hiwat)
561 		so->so_snd.ssb_lowat = so->so_snd.ssb_hiwat;
562 	return (0);
563 bad2:
564 	ssb_release(&so->so_snd, so);
565 bad:
566 	return (ENOBUFS);
567 }
568 
569 static int
570 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS)
571 {
572 	int error = 0;
573 	u_long old_sb_max = sb_max;
574 
575 	error = SYSCTL_OUT(req, arg1, sizeof(int));
576 	if (error || !req->newptr)
577 		return (error);
578 	error = SYSCTL_IN(req, arg1, sizeof(int));
579 	if (error)
580 		return (error);
581 	if (sb_max < MSIZE + MCLBYTES) {
582 		sb_max = old_sb_max;
583 		return (EINVAL);
584 	}
585 	sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES);
586 	return (0);
587 }
588 
589 /*
590  * Allot mbufs to a signalsockbuf.
591  *
592  * Attempt to scale mbmax so that mbcnt doesn't become limiting
593  * if buffering efficiency is near the normal case.
594  *
595  * sb_max only applies to user-sockets (where rl != NULL).  It does
596  * not apply to kernel sockets or kernel-controlled sockets.  Note
597  * that NFS overrides the sockbuf limits created when nfsd creates
598  * a socket.
599  */
600 int
601 ssb_reserve(struct signalsockbuf *ssb, u_long cc, struct socket *so,
602 	    struct rlimit *rl)
603 {
604 	/*
605 	 * rl will only be NULL when we're in an interrupt (eg, in tcp_input)
606 	 * or when called from netgraph (ie, ngd_attach)
607 	 */
608 	if (rl && cc > sb_max_adj)
609 		cc = sb_max_adj;
610 	if (!chgsbsize(so->so_cred->cr_uidinfo, &ssb->ssb_hiwat, cc,
611 		       rl ? rl->rlim_cur : RLIM_INFINITY)) {
612 		return (0);
613 	}
614 	if (rl)
615 		ssb->ssb_mbmax = min(cc * sb_efficiency, sb_max);
616 	else
617 		ssb->ssb_mbmax = cc * sb_efficiency;
618 
619 	/*
620 	 * AUTOLOWAT is set on send buffers and prevents large writes
621 	 * from generating a huge number of context switches.
622 	 */
623 	if (ssb->ssb_flags & SSB_AUTOLOWAT) {
624 		ssb->ssb_lowat = ssb->ssb_hiwat / 2;
625 		if (ssb->ssb_lowat < MCLBYTES)
626 			ssb->ssb_lowat = MCLBYTES;
627 	}
628 	if (ssb->ssb_lowat > ssb->ssb_hiwat)
629 		ssb->ssb_lowat = ssb->ssb_hiwat;
630 	return (1);
631 }
632 
633 /*
634  * Free mbufs held by a socket, and reserved mbuf space.
635  */
636 void
637 ssb_release(struct signalsockbuf *ssb, struct socket *so)
638 {
639 	sbflush(&ssb->sb);
640 	(void)chgsbsize(so->so_cred->cr_uidinfo, &ssb->ssb_hiwat, 0,
641 	    RLIM_INFINITY);
642 	ssb->ssb_mbmax = 0;
643 }
644 
645 /*
646  * Some routines that return EOPNOTSUPP for entry points that are not
647  * supported by a protocol.  Fill in as needed.
648  */
649 int
650 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
651 {
652 	return EOPNOTSUPP;
653 }
654 
655 int
656 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
657 {
658 	return EOPNOTSUPP;
659 }
660 
661 int
662 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
663 {
664 	return EOPNOTSUPP;
665 }
666 
667 int
668 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
669 {
670 	return EOPNOTSUPP;
671 }
672 
673 int
674 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
675 		    struct ifnet *ifp, struct thread *td)
676 {
677 	return EOPNOTSUPP;
678 }
679 
680 int
681 pru_disconnect_notsupp(struct socket *so)
682 {
683 	return EOPNOTSUPP;
684 }
685 
686 int
687 pru_listen_notsupp(struct socket *so, struct thread *td)
688 {
689 	return EOPNOTSUPP;
690 }
691 
692 int
693 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
694 {
695 	return EOPNOTSUPP;
696 }
697 
698 int
699 pru_rcvd_notsupp(struct socket *so, int flags)
700 {
701 	return EOPNOTSUPP;
702 }
703 
704 int
705 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
706 {
707 	return EOPNOTSUPP;
708 }
709 
710 int
711 pru_shutdown_notsupp(struct socket *so)
712 {
713 	return EOPNOTSUPP;
714 }
715 
716 int
717 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
718 {
719 	return EOPNOTSUPP;
720 }
721 
722 int
723 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
724 	   struct mbuf *top, struct mbuf *control, int flags,
725 	   struct thread *td)
726 {
727 	if (top)
728 		m_freem(top);
729 	if (control)
730 		m_freem(control);
731 	return (EOPNOTSUPP);
732 }
733 
734 int
735 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
736 		      struct uio *uio, struct sockbuf *sio,
737 		      struct mbuf **controlp, int *flagsp)
738 {
739 	return (EOPNOTSUPP);
740 }
741 
742 int
743 pru_ctloutput_notsupp(struct socket *so, struct sockopt *sopt)
744 {
745 	return (EOPNOTSUPP);
746 }
747 
748 /*
749  * This isn't really a ``null'' operation, but it's the default one
750  * and doesn't do anything destructive.
751  */
752 int
753 pru_sense_null(struct socket *so, struct stat *sb)
754 {
755 	sb->st_blksize = so->so_snd.ssb_hiwat;
756 	return 0;
757 }
758 
759 /*
760  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.  Callers
761  * of this routine assume that it always succeeds, so we have to use a
762  * blockable allocation even though we might be called from a critical thread.
763  */
764 struct sockaddr *
765 dup_sockaddr(const struct sockaddr *sa)
766 {
767 	struct sockaddr *sa2;
768 
769 	sa2 = kmalloc(sa->sa_len, M_SONAME, M_INTWAIT);
770 	bcopy(sa, sa2, sa->sa_len);
771 	return (sa2);
772 }
773 
774 /*
775  * Create an external-format (``xsocket'') structure using the information
776  * in the kernel-format socket structure pointed to by so.  This is done
777  * to reduce the spew of irrelevant information over this interface,
778  * to isolate user code from changes in the kernel structure, and
779  * potentially to provide information-hiding if we decide that
780  * some of this information should be hidden from users.
781  */
782 void
783 sotoxsocket(struct socket *so, struct xsocket *xso)
784 {
785 	xso->xso_len = sizeof *xso;
786 	xso->xso_so = so;
787 	xso->so_type = so->so_type;
788 	xso->so_options = so->so_options;
789 	xso->so_linger = so->so_linger;
790 	xso->so_state = so->so_state;
791 	xso->so_pcb = so->so_pcb;
792 	xso->xso_protocol = so->so_proto->pr_protocol;
793 	xso->xso_family = so->so_proto->pr_domain->dom_family;
794 	xso->so_qlen = so->so_qlen;
795 	xso->so_incqlen = so->so_incqlen;
796 	xso->so_qlimit = so->so_qlimit;
797 	xso->so_timeo = so->so_timeo;
798 	xso->so_error = so->so_error;
799 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
800 	xso->so_oobmark = so->so_oobmark;
801 	ssbtoxsockbuf(&so->so_snd, &xso->so_snd);
802 	ssbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
803 	xso->so_uid = so->so_cred->cr_uid;
804 }
805 
806 /*
807  * Here is the definition of some of the basic objects in the kern.ipc
808  * branch of the MIB.
809  */
810 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
811 
812 /*
813  * This takes the place of kern.maxsockbuf, which moved to kern.ipc.
814  *
815  * NOTE! sb_max only applies to user-created socket buffers.
816  */
817 static int dummy;
818 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, "");
819 SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_INT|CTLFLAG_RW,
820     &sb_max, 0, sysctl_handle_sb_max, "I", "Maximum socket buffer size");
821 SYSCTL_INT(_kern_ipc, OID_AUTO, maxsockets, CTLFLAG_RD,
822     &maxsockets, 0, "Maximum number of sockets available");
823 SYSCTL_INT(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW,
824     &sb_efficiency, 0, "");
825 
826 /*
827  * Initialize maxsockets
828  */
829 static void
830 init_maxsockets(void *ignored)
831 {
832     TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
833     maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
834 }
835 SYSINIT(param, SI_BOOT1_TUNABLES, SI_ORDER_ANY,
836 	init_maxsockets, NULL);
837 
838