xref: /dflybsd-src/sys/kern/uipc_socket2.c (revision 67bf99c4e3c62e257027c8f0d3b312f44cfe622f)
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/socketops.h>
54 #include <sys/signalvar.h>
55 #include <sys/sysctl.h>
56 #include <sys/aio.h> /* for aio_swake proto */
57 #include <sys/event.h>
58 
59 #include <sys/thread2.h>
60 #include <sys/msgport2.h>
61 #include <sys/socketvar2.h>
62 
63 int	maxsockets;
64 
65 /*
66  * Primitive routines for operating on sockets and socket buffers
67  */
68 
69 u_long	sb_max = SB_MAX;
70 u_long	sb_max_adj =
71     SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */
72 
73 static	u_long sb_efficiency = 8;	/* parameter for sbreserve() */
74 
75 /************************************************************************
76  * signalsockbuf procedures						*
77  ************************************************************************/
78 
79 /*
80  * Wait for data to arrive at/drain from a socket buffer.
81  *
82  * NOTE: Caller must generally hold the ssb_lock (client side lock) since
83  *	 WAIT/WAKEUP only works for one client at a time.
84  *
85  * NOTE: Caller always retries whatever operation it was waiting on.
86  */
87 int
88 ssb_wait(struct signalsockbuf *ssb)
89 {
90 	uint32_t flags;
91 	int pflags;
92 	int error;
93 
94 	pflags = (ssb->ssb_flags & SSB_NOINTR) ? 0 : PCATCH;
95 
96 	for (;;) {
97 		flags = ssb->ssb_flags;
98 		cpu_ccfence();
99 
100 		/*
101 		 * WAKEUP and WAIT interlock eachother.  We can catch the
102 		 * race by checking to see if WAKEUP has already been set,
103 		 * and only setting WAIT if WAKEUP is clear.
104 		 */
105 		if (flags & SSB_WAKEUP) {
106 			if (atomic_cmpset_int(&ssb->ssb_flags, flags,
107 					      flags & ~SSB_WAKEUP)) {
108 				error = 0;
109 				break;
110 			}
111 			continue;
112 		}
113 
114 		/*
115 		 * Only set WAIT if WAKEUP is clear.
116 		 */
117 		tsleep_interlock(&ssb->ssb_cc, pflags);
118 		if (atomic_cmpset_int(&ssb->ssb_flags, flags,
119 				      flags | SSB_WAIT)) {
120 			error = tsleep(&ssb->ssb_cc, pflags | PINTERLOCKED,
121 				       "sbwait", ssb->ssb_timeo);
122 			break;
123 		}
124 	}
125 	return (error);
126 }
127 
128 /*
129  * Lock a sockbuf already known to be locked;
130  * return any error returned from sleep (EINTR).
131  */
132 int
133 _ssb_lock(struct signalsockbuf *ssb)
134 {
135 	uint32_t flags;
136 	int pflags;
137 	int error;
138 
139 	pflags = (ssb->ssb_flags & SSB_NOINTR) ? 0 : PCATCH;
140 
141 	for (;;) {
142 		flags = ssb->ssb_flags;
143 		cpu_ccfence();
144 		if (flags & SSB_LOCK) {
145 			tsleep_interlock(&ssb->ssb_flags, pflags);
146 			if (atomic_cmpset_int(&ssb->ssb_flags, flags,
147 					      flags | SSB_WANT)) {
148 				error = tsleep(&ssb->ssb_flags,
149 					       pflags | PINTERLOCKED,
150 					       "sblock", 0);
151 				if (error)
152 					break;
153 			}
154 		} else {
155 			if (atomic_cmpset_int(&ssb->ssb_flags, flags,
156 					      flags | SSB_LOCK)) {
157 				lwkt_gettoken(&ssb->ssb_token);
158 				error = 0;
159 				break;
160 			}
161 		}
162 	}
163 	return (error);
164 }
165 
166 /*
167  * This does the same for sockbufs.  Note that the xsockbuf structure,
168  * since it is always embedded in a socket, does not include a self
169  * pointer nor a length.  We make this entry point public in case
170  * some other mechanism needs it.
171  */
172 void
173 ssbtoxsockbuf(struct signalsockbuf *ssb, struct xsockbuf *xsb)
174 {
175 	xsb->sb_cc = ssb->ssb_cc;
176 	xsb->sb_hiwat = ssb->ssb_hiwat;
177 	xsb->sb_mbcnt = ssb->ssb_mbcnt;
178 	xsb->sb_mbmax = ssb->ssb_mbmax;
179 	xsb->sb_lowat = ssb->ssb_lowat;
180 	xsb->sb_flags = ssb->ssb_flags;
181 	xsb->sb_timeo = ssb->ssb_timeo;
182 }
183 
184 
185 /************************************************************************
186  * Procedures which manipulate socket state flags, wakeups, etc.	*
187  ************************************************************************
188  *
189  * Normal sequence from the active (originating) side is that
190  * soisconnecting() is called during processing of connect() call, resulting
191  * in an eventual call to soisconnected() if/when the connection is
192  * established.  When the connection is torn down soisdisconnecting() is
193  * called during processing of disconnect() call, and soisdisconnected() is
194  * called when the connection to the peer is totally severed.
195  *
196  * The semantics of these routines are such that connectionless protocols
197  * can call soisconnected() and soisdisconnected() only, bypassing the
198  * in-progress calls when setting up a ``connection'' takes no time.
199  *
200  * From the passive side, a socket is created with two queues of sockets:
201  * so_incomp for connections in progress and so_comp for connections
202  * already made and awaiting user acceptance.  As a protocol is preparing
203  * incoming connections, it creates a socket structure queued on so_incomp
204  * by calling sonewconn().  When the connection is established,
205  * soisconnected() is called, and transfers the socket structure to so_comp,
206  * making it available to accept().
207  *
208  * If a socket is closed with sockets on either so_incomp or so_comp, these
209  * sockets are dropped.
210  *
211  * If higher level protocols are implemented in the kernel, the wakeups
212  * done here will sometimes cause software-interrupt process scheduling.
213  */
214 
215 void
216 soisconnecting(struct socket *so)
217 {
218 	soclrstate(so, SS_ISCONNECTED | SS_ISDISCONNECTING);
219 	sosetstate(so, SS_ISCONNECTING);
220 }
221 
222 void
223 soisconnected(struct socket *so)
224 {
225 	struct socket *head = so->so_head;
226 
227 	soclrstate(so, SS_ISCONNECTING | SS_ISDISCONNECTING | SS_ISCONFIRMING);
228 	sosetstate(so, SS_ISCONNECTED);
229 	if (head && (so->so_state & SS_INCOMP)) {
230 		if ((so->so_options & SO_ACCEPTFILTER) != 0) {
231 			so->so_upcall = head->so_accf->so_accept_filter->accf_callback;
232 			so->so_upcallarg = head->so_accf->so_accept_filter_arg;
233 			atomic_set_int(&so->so_rcv.ssb_flags, SSB_UPCALL);
234 			so->so_options &= ~SO_ACCEPTFILTER;
235 			so->so_upcall(so, so->so_upcallarg, 0);
236 			return;
237 		}
238 
239 		/*
240 		 * Listen socket are not per-cpu.
241 		 */
242 		lwkt_gettoken(&head->so_rcv.ssb_token);
243 		TAILQ_REMOVE(&head->so_incomp, so, so_list);
244 		head->so_incqlen--;
245 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
246 		head->so_qlen++;
247 		sosetstate(so, SS_COMP);
248 		soclrstate(so, SS_INCOMP);
249 		lwkt_reltoken(&head->so_rcv.ssb_token);
250 
251 		/*
252 		 * XXX head may be on a different protocol thread.
253 		 *     sorwakeup()->sowakeup() is hacked atm.
254 		 */
255 		sorwakeup(head);
256 		wakeup_one(&head->so_timeo);
257 	} else {
258 		wakeup(&so->so_timeo);
259 		sorwakeup(so);
260 		sowwakeup(so);
261 	}
262 }
263 
264 void
265 soisdisconnecting(struct socket *so)
266 {
267 	soclrstate(so, SS_ISCONNECTING);
268 	sosetstate(so, SS_ISDISCONNECTING | SS_CANTRCVMORE | SS_CANTSENDMORE);
269 	wakeup((caddr_t)&so->so_timeo);
270 	sowwakeup(so);
271 	sorwakeup(so);
272 }
273 
274 void
275 soisdisconnected(struct socket *so)
276 {
277 	soclrstate(so, SS_ISCONNECTING | SS_ISCONNECTED | SS_ISDISCONNECTING);
278 	sosetstate(so, SS_CANTRCVMORE | SS_CANTSENDMORE | SS_ISDISCONNECTED);
279 	wakeup((caddr_t)&so->so_timeo);
280 	sbdrop(&so->so_snd.sb, so->so_snd.ssb_cc);
281 	sowwakeup(so);
282 	sorwakeup(so);
283 }
284 
285 void
286 soisreconnecting(struct socket *so)
287 {
288         soclrstate(so, SS_ISDISCONNECTING | SS_ISDISCONNECTED |
289 		       SS_CANTRCVMORE | SS_CANTSENDMORE);
290 	sosetstate(so, SS_ISCONNECTING);
291 }
292 
293 void
294 soisreconnected(struct socket *so)
295 {
296 	soclrstate(so, SS_ISDISCONNECTED | SS_CANTRCVMORE | SS_CANTSENDMORE);
297 	soisconnected(so);
298 }
299 
300 /*
301  * Set or change the message port a socket receives commands on.
302  *
303  * XXX
304  */
305 void
306 sosetport(struct socket *so, lwkt_port_t port)
307 {
308 	so->so_port = port;
309 }
310 
311 /*
312  * When an attempt at a new connection is noted on a socket
313  * which accepts connections, sonewconn is called.  If the
314  * connection is possible (subject to space constraints, etc.)
315  * then we allocate a new structure, propoerly linked into the
316  * data structure of the original socket, and return this.
317  * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
318  *
319  * The new socket is returned with one ref and so_pcb assigned.
320  * The reference is implied by so_pcb.
321  */
322 struct socket *
323 sonewconn(struct socket *head, int connstatus)
324 {
325 	struct socket *so;
326 	struct socket *sp;
327 	struct pru_attach_info ai;
328 
329 	if (head->so_qlen > 3 * head->so_qlimit / 2)
330 		return (NULL);
331 	so = soalloc(1);
332 	if (so == NULL)
333 		return (NULL);
334 
335 	/*
336 	 * Set the port prior to attaching the inpcb to the current
337 	 * cpu's protocol thread (which should be the current thread
338 	 * but might not be in all cases).  This serializes any pcb ops
339 	 * which occur to our cpu allowing us to complete the attachment
340 	 * without racing anything.
341 	 */
342 	sosetport(so, cpu_portfn(mycpu->gd_cpuid));
343 	if ((head->so_options & SO_ACCEPTFILTER) != 0)
344 		connstatus = 0;
345 	so->so_head = head;
346 	so->so_type = head->so_type;
347 	so->so_options = head->so_options &~ SO_ACCEPTCONN;
348 	so->so_linger = head->so_linger;
349 
350 	/*
351 	 * NOTE: Clearing NOFDREF implies referencing the so with
352 	 *	 soreference().
353 	 */
354 	so->so_state = head->so_state | SS_NOFDREF | SS_ASSERTINPROG;
355 	so->so_proto = head->so_proto;
356 	so->so_cred = crhold(head->so_cred);
357 	ai.sb_rlimit = NULL;
358 	ai.p_ucred = NULL;
359 	ai.fd_rdir = NULL;		/* jail code cruft XXX JH */
360 
361 	/*
362 	 * Reserve space and call pru_attach.  We can direct-call the
363 	 * function since we're already in the protocol thread.
364 	 */
365 	if (soreserve(so, head->so_snd.ssb_hiwat,
366 		      head->so_rcv.ssb_hiwat, NULL) ||
367 	    so_pru_attach_direct(so, 0, &ai)) {
368 		so->so_head = NULL;
369 		soclrstate(so, SS_ASSERTINPROG);
370 		sofree(so);		/* remove implied pcb ref */
371 		return (NULL);
372 	}
373 	KKASSERT(so->so_refs == 2);	/* attach + our base ref */
374 	sofree(so);
375 	KKASSERT(so->so_port != NULL);
376 	so->so_rcv.ssb_lowat = head->so_rcv.ssb_lowat;
377 	so->so_snd.ssb_lowat = head->so_snd.ssb_lowat;
378 	so->so_rcv.ssb_timeo = head->so_rcv.ssb_timeo;
379 	so->so_snd.ssb_timeo = head->so_snd.ssb_timeo;
380 	so->so_rcv.ssb_flags |= head->so_rcv.ssb_flags &
381 				(SSB_AUTOSIZE | SSB_AUTOLOWAT);
382 	so->so_snd.ssb_flags |= head->so_snd.ssb_flags &
383 				(SSB_AUTOSIZE | SSB_AUTOLOWAT);
384 	lwkt_gettoken(&head->so_rcv.ssb_token);
385 	if (connstatus) {
386 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
387 		sosetstate(so, SS_COMP);
388 		head->so_qlen++;
389 	} else {
390 		if (head->so_incqlen > head->so_qlimit) {
391 			sp = TAILQ_FIRST(&head->so_incomp);
392 			TAILQ_REMOVE(&head->so_incomp, sp, so_list);
393 			head->so_incqlen--;
394 			soclrstate(sp, SS_INCOMP);
395 			sp->so_head = NULL;
396 			soaborta(sp);
397 		}
398 		TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
399 		sosetstate(so, SS_INCOMP);
400 		head->so_incqlen++;
401 	}
402 	lwkt_reltoken(&head->so_rcv.ssb_token);
403 	if (connstatus) {
404 		/*
405 		 * XXX head may be on a different protocol thread.
406 		 *     sorwakeup()->sowakeup() is hacked atm.
407 		 */
408 		sorwakeup(head);
409 		wakeup((caddr_t)&head->so_timeo);
410 		sosetstate(so, connstatus);
411 	}
412 	soclrstate(so, SS_ASSERTINPROG);
413 	return (so);
414 }
415 
416 /*
417  * Socantsendmore indicates that no more data will be sent on the
418  * socket; it would normally be applied to a socket when the user
419  * informs the system that no more data is to be sent, by the protocol
420  * code (in case PRU_SHUTDOWN).  Socantrcvmore indicates that no more data
421  * will be received, and will normally be applied to the socket by a
422  * protocol when it detects that the peer will send no more data.
423  * Data queued for reading in the socket may yet be read.
424  */
425 void
426 socantsendmore(struct socket *so)
427 {
428 	sosetstate(so, SS_CANTSENDMORE);
429 	sowwakeup(so);
430 }
431 
432 void
433 socantrcvmore(struct socket *so)
434 {
435 	sosetstate(so, SS_CANTRCVMORE);
436 	sorwakeup(so);
437 }
438 
439 /*
440  * Wakeup processes waiting on a socket buffer.  Do asynchronous notification
441  * via SIGIO if the socket has the SS_ASYNC flag set.
442  *
443  * For users waiting on send/recv try to avoid unnecessary context switch
444  * thrashing.  Particularly for senders of large buffers (needs to be
445  * extended to sel and aio? XXX)
446  *
447  * WARNING!  Can be called on a foreign socket from the wrong protocol
448  *	     thread.  aka is called on the 'head' listen socket when
449  *	     a new connection comes in.
450  */
451 void
452 sowakeup(struct socket *so, struct signalsockbuf *ssb)
453 {
454 	struct kqinfo *kqinfo = &ssb->ssb_kq;
455 	uint32_t flags;
456 
457 	/*
458 	 * Check conditions, set the WAKEUP flag, and clear and signal if
459 	 * the WAIT flag is found to be set.  This interlocks against the
460 	 * client side.
461 	 */
462 	for (;;) {
463 		flags = ssb->ssb_flags;
464 		cpu_ccfence();
465 
466 		if ((ssb == &so->so_snd && ssb_space(ssb) >= ssb->ssb_lowat) ||
467 		    (ssb == &so->so_rcv && ssb->ssb_cc >= ssb->ssb_lowat) ||
468 		    (ssb == &so->so_snd && (so->so_state & SS_CANTSENDMORE)) ||
469 		    (ssb == &so->so_rcv && (so->so_state & SS_CANTRCVMORE))
470 		) {
471 			if (atomic_cmpset_int(&ssb->ssb_flags, flags,
472 					  (flags | SSB_WAKEUP) & ~SSB_WAIT)) {
473 				if (flags & SSB_WAIT)
474 					wakeup(&ssb->ssb_cc);
475 				break;
476 			}
477 		} else {
478 			break;
479 		}
480 	}
481 
482 	/*
483 	 * Misc other events
484 	 */
485 	if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL)
486 		pgsigio(so->so_sigio, SIGIO, 0);
487 	if (ssb->ssb_flags & SSB_UPCALL)
488 		(*so->so_upcall)(so, so->so_upcallarg, MB_DONTWAIT);
489 	if (ssb->ssb_flags & SSB_AIO)
490 		aio_swake(so, ssb);
491 	KNOTE(&kqinfo->ki_note, 0);
492 
493 	/*
494 	 * This is a bit of a hack.  Multiple threads can wind up scanning
495 	 * ki_mlist concurrently due to the fact that this function can be
496 	 * called on a foreign socket, so we can't afford to block here.
497 	 */
498 	if (ssb->ssb_flags & SSB_MEVENT) {
499 		struct netmsg_so_notify *msg, *nmsg;
500 
501 		lwkt_gettoken(&kq_token);
502 		lwkt_gettoken_hard(&ssb->ssb_token);
503 		TAILQ_FOREACH_MUTABLE(msg, &kqinfo->ki_mlist, nm_list, nmsg) {
504 			if (msg->nm_predicate(msg)) {
505 				TAILQ_REMOVE(&kqinfo->ki_mlist, msg, nm_list);
506 				lwkt_replymsg(&msg->base.lmsg,
507 					      msg->base.lmsg.ms_error);
508 			}
509 		}
510 		if (TAILQ_EMPTY(&ssb->ssb_kq.ki_mlist))
511 			atomic_clear_int(&ssb->ssb_flags, SSB_MEVENT);
512 		lwkt_reltoken_hard(&ssb->ssb_token);
513 		lwkt_reltoken(&kq_token);
514 	}
515 }
516 
517 /*
518  * Socket buffer (struct signalsockbuf) utility routines.
519  *
520  * Each socket contains two socket buffers: one for sending data and
521  * one for receiving data.  Each buffer contains a queue of mbufs,
522  * information about the number of mbufs and amount of data in the
523  * queue, and other fields allowing kevent()/select()/poll() statements
524  * and notification on data availability to be implemented.
525  *
526  * Data stored in a socket buffer is maintained as a list of records.
527  * Each record is a list of mbufs chained together with the m_next
528  * field.  Records are chained together with the m_nextpkt field. The upper
529  * level routine soreceive() expects the following conventions to be
530  * observed when placing information in the receive buffer:
531  *
532  * 1. If the protocol requires each message be preceded by the sender's
533  *    name, then a record containing that name must be present before
534  *    any associated data (mbuf's must be of type MT_SONAME).
535  * 2. If the protocol supports the exchange of ``access rights'' (really
536  *    just additional data associated with the message), and there are
537  *    ``rights'' to be received, then a record containing this data
538  *    should be present (mbuf's must be of type MT_RIGHTS).
539  * 3. If a name or rights record exists, then it must be followed by
540  *    a data record, perhaps of zero length.
541  *
542  * Before using a new socket structure it is first necessary to reserve
543  * buffer space to the socket, by calling sbreserve().  This should commit
544  * some of the available buffer space in the system buffer pool for the
545  * socket (currently, it does nothing but enforce limits).  The space
546  * should be released by calling ssb_release() when the socket is destroyed.
547  */
548 int
549 soreserve(struct socket *so, u_long sndcc, u_long rcvcc, struct rlimit *rl)
550 {
551 	if (so->so_snd.ssb_lowat == 0)
552 		atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOLOWAT);
553 	if (ssb_reserve(&so->so_snd, sndcc, so, rl) == 0)
554 		goto bad;
555 	if (ssb_reserve(&so->so_rcv, rcvcc, so, rl) == 0)
556 		goto bad2;
557 	if (so->so_rcv.ssb_lowat == 0)
558 		so->so_rcv.ssb_lowat = 1;
559 	if (so->so_snd.ssb_lowat == 0)
560 		so->so_snd.ssb_lowat = MCLBYTES;
561 	if (so->so_snd.ssb_lowat > so->so_snd.ssb_hiwat)
562 		so->so_snd.ssb_lowat = so->so_snd.ssb_hiwat;
563 	return (0);
564 bad2:
565 	ssb_release(&so->so_snd, so);
566 bad:
567 	return (ENOBUFS);
568 }
569 
570 static int
571 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS)
572 {
573 	int error = 0;
574 	u_long old_sb_max = sb_max;
575 
576 	error = SYSCTL_OUT(req, arg1, sizeof(int));
577 	if (error || !req->newptr)
578 		return (error);
579 	error = SYSCTL_IN(req, arg1, sizeof(int));
580 	if (error)
581 		return (error);
582 	if (sb_max < MSIZE + MCLBYTES) {
583 		sb_max = old_sb_max;
584 		return (EINVAL);
585 	}
586 	sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES);
587 	return (0);
588 }
589 
590 /*
591  * Allot mbufs to a signalsockbuf.
592  *
593  * Attempt to scale mbmax so that mbcnt doesn't become limiting
594  * if buffering efficiency is near the normal case.
595  *
596  * sb_max only applies to user-sockets (where rl != NULL).  It does
597  * not apply to kernel sockets or kernel-controlled sockets.  Note
598  * that NFS overrides the sockbuf limits created when nfsd creates
599  * a socket.
600  */
601 int
602 ssb_reserve(struct signalsockbuf *ssb, u_long cc, struct socket *so,
603 	    struct rlimit *rl)
604 {
605 	/*
606 	 * rl will only be NULL when we're in an interrupt (eg, in tcp_input)
607 	 * or when called from netgraph (ie, ngd_attach)
608 	 */
609 	if (rl && cc > sb_max_adj)
610 		cc = sb_max_adj;
611 	if (!chgsbsize(so->so_cred->cr_uidinfo, &ssb->ssb_hiwat, cc,
612 		       rl ? rl->rlim_cur : RLIM_INFINITY)) {
613 		return (0);
614 	}
615 	if (rl)
616 		ssb->ssb_mbmax = min(cc * sb_efficiency, sb_max);
617 	else
618 		ssb->ssb_mbmax = cc * sb_efficiency;
619 
620 	/*
621 	 * AUTOLOWAT is set on send buffers and prevents large writes
622 	 * from generating a huge number of context switches.
623 	 */
624 	if (ssb->ssb_flags & SSB_AUTOLOWAT) {
625 		ssb->ssb_lowat = ssb->ssb_hiwat / 2;
626 		if (ssb->ssb_lowat < MCLBYTES)
627 			ssb->ssb_lowat = MCLBYTES;
628 	}
629 	if (ssb->ssb_lowat > ssb->ssb_hiwat)
630 		ssb->ssb_lowat = ssb->ssb_hiwat;
631 	return (1);
632 }
633 
634 /*
635  * Free mbufs held by a socket, and reserved mbuf space.
636  */
637 void
638 ssb_release(struct signalsockbuf *ssb, struct socket *so)
639 {
640 	sbflush(&ssb->sb);
641 	(void)chgsbsize(so->so_cred->cr_uidinfo, &ssb->ssb_hiwat, 0,
642 	    RLIM_INFINITY);
643 	ssb->ssb_mbmax = 0;
644 }
645 
646 /*
647  * Some routines that return EOPNOTSUPP for entry points that are not
648  * supported by a protocol.  Fill in as needed.
649  */
650 void
651 pr_generic_notsupp(netmsg_t msg)
652 {
653 	lwkt_replymsg(&msg->lmsg, EOPNOTSUPP);
654 }
655 
656 int
657 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
658 	   struct mbuf *top, struct mbuf *control, int flags,
659 	   struct thread *td)
660 {
661 	if (top)
662 		m_freem(top);
663 	if (control)
664 		m_freem(control);
665 	return (EOPNOTSUPP);
666 }
667 
668 int
669 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
670 		      struct uio *uio, struct sockbuf *sio,
671 		      struct mbuf **controlp, int *flagsp)
672 {
673 	return (EOPNOTSUPP);
674 }
675 
676 /*
677  * This isn't really a ``null'' operation, but it's the default one
678  * and doesn't do anything destructive.
679  */
680 void
681 pru_sense_null(netmsg_t msg)
682 {
683 	msg->sense.nm_stat->st_blksize = msg->base.nm_so->so_snd.ssb_hiwat;
684 	lwkt_replymsg(&msg->lmsg, 0);
685 }
686 
687 /*
688  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.  Callers
689  * of this routine assume that it always succeeds, so we have to use a
690  * blockable allocation even though we might be called from a critical thread.
691  */
692 struct sockaddr *
693 dup_sockaddr(const struct sockaddr *sa)
694 {
695 	struct sockaddr *sa2;
696 
697 	sa2 = kmalloc(sa->sa_len, M_SONAME, M_INTWAIT);
698 	bcopy(sa, sa2, sa->sa_len);
699 	return (sa2);
700 }
701 
702 /*
703  * Create an external-format (``xsocket'') structure using the information
704  * in the kernel-format socket structure pointed to by so.  This is done
705  * to reduce the spew of irrelevant information over this interface,
706  * to isolate user code from changes in the kernel structure, and
707  * potentially to provide information-hiding if we decide that
708  * some of this information should be hidden from users.
709  */
710 void
711 sotoxsocket(struct socket *so, struct xsocket *xso)
712 {
713 	xso->xso_len = sizeof *xso;
714 	xso->xso_so = so;
715 	xso->so_type = so->so_type;
716 	xso->so_options = so->so_options;
717 	xso->so_linger = so->so_linger;
718 	xso->so_state = so->so_state;
719 	xso->so_pcb = so->so_pcb;
720 	xso->xso_protocol = so->so_proto->pr_protocol;
721 	xso->xso_family = so->so_proto->pr_domain->dom_family;
722 	xso->so_qlen = so->so_qlen;
723 	xso->so_incqlen = so->so_incqlen;
724 	xso->so_qlimit = so->so_qlimit;
725 	xso->so_timeo = so->so_timeo;
726 	xso->so_error = so->so_error;
727 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
728 	xso->so_oobmark = so->so_oobmark;
729 	ssbtoxsockbuf(&so->so_snd, &xso->so_snd);
730 	ssbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
731 	xso->so_uid = so->so_cred->cr_uid;
732 }
733 
734 /*
735  * Here is the definition of some of the basic objects in the kern.ipc
736  * branch of the MIB.
737  */
738 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
739 
740 /*
741  * This takes the place of kern.maxsockbuf, which moved to kern.ipc.
742  *
743  * NOTE! sb_max only applies to user-created socket buffers.
744  */
745 static int dummy;
746 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, "");
747 SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_INT|CTLFLAG_RW,
748     &sb_max, 0, sysctl_handle_sb_max, "I", "Maximum socket buffer size");
749 SYSCTL_INT(_kern_ipc, OID_AUTO, maxsockets, CTLFLAG_RD,
750     &maxsockets, 0, "Maximum number of sockets available");
751 SYSCTL_INT(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW,
752     &sb_efficiency, 0, "");
753 
754 /*
755  * Initialize maxsockets
756  */
757 static void
758 init_maxsockets(void *ignored)
759 {
760     TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
761     maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
762 }
763 SYSINIT(param, SI_BOOT1_TUNABLES, SI_ORDER_ANY,
764 	init_maxsockets, NULL);
765 
766