xref: /dflybsd-src/sys/kern/uipc_socket2.c (revision 3cc0afc616515a4d253a4a2bf5437858ae934c06)
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. Neither the name of the University nor the names of its contributors
15  *    may be used to endorse or promote products derived from this software
16  *    without specific prior written permission.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  *
30  *	@(#)uipc_socket2.c	8.1 (Berkeley) 6/10/93
31  * $FreeBSD: src/sys/kern/uipc_socket2.c,v 1.55.2.17 2002/08/31 19:04:55 dwmalone Exp $
32  */
33 
34 #include "opt_param.h"
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/domain.h>
38 #include <sys/file.h>	/* for maxfiles */
39 #include <sys/kernel.h>
40 #include <sys/ktr.h>
41 #include <sys/proc.h>
42 #include <sys/malloc.h>
43 #include <sys/mbuf.h>
44 #include <sys/protosw.h>
45 #include <sys/resourcevar.h>
46 #include <sys/stat.h>
47 #include <sys/socket.h>
48 #include <sys/socketvar.h>
49 #include <sys/socketops.h>
50 #include <sys/signalvar.h>
51 #include <sys/sysctl.h>
52 #include <sys/event.h>
53 
54 #include <sys/thread2.h>
55 #include <sys/msgport2.h>
56 #include <sys/socketvar2.h>
57 
58 #include <net/netisr2.h>
59 
60 #ifndef KTR_SOWAKEUP
61 #define KTR_SOWAKEUP	KTR_ALL
62 #endif
63 KTR_INFO_MASTER(sowakeup);
64 KTR_INFO(KTR_SOWAKEUP, sowakeup, nconn_start, 0, "newconn sorwakeup start");
65 KTR_INFO(KTR_SOWAKEUP, sowakeup, nconn_end, 1, "newconn sorwakeup end");
66 KTR_INFO(KTR_SOWAKEUP, sowakeup, nconn_wakeupstart, 2, "newconn wakeup start");
67 KTR_INFO(KTR_SOWAKEUP, sowakeup, nconn_wakeupend, 3, "newconn wakeup end");
68 #define logsowakeup(name)	KTR_LOG(sowakeup_ ## name)
69 
70 int	maxsockets;
71 
72 /*
73  * Primitive routines for operating on sockets and socket buffers
74  */
75 
76 u_long	sb_max = SB_MAX;
77 u_long	sb_max_adj =
78     SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */
79 
80 static	u_long sb_efficiency = 8;	/* parameter for sbreserve() */
81 
82 /************************************************************************
83  * signalsockbuf procedures						*
84  ************************************************************************/
85 
86 /*
87  * Wait for data to arrive at/drain from a socket buffer.
88  *
89  * NOTE: Caller must generally hold the ssb_lock (client side lock) since
90  *	 WAIT/WAKEUP only works for one client at a time.
91  *
92  * NOTE: Caller always retries whatever operation it was waiting on.
93  */
94 int
95 ssb_wait(struct signalsockbuf *ssb)
96 {
97 	uint32_t flags;
98 	int pflags;
99 	int error;
100 
101 	pflags = (ssb->ssb_flags & SSB_NOINTR) ? 0 : PCATCH;
102 
103 	for (;;) {
104 		flags = ssb->ssb_flags;
105 		cpu_ccfence();
106 
107 		/*
108 		 * WAKEUP and WAIT interlock each other.  We can catch the
109 		 * race by checking to see if WAKEUP has already been set,
110 		 * and only setting WAIT if WAKEUP is clear.
111 		 */
112 		if (flags & SSB_WAKEUP) {
113 			if (atomic_cmpset_int(&ssb->ssb_flags, flags,
114 					      flags & ~SSB_WAKEUP)) {
115 				error = 0;
116 				break;
117 			}
118 			continue;
119 		}
120 
121 		/*
122 		 * Only set WAIT if WAKEUP is clear.
123 		 */
124 		tsleep_interlock(&ssb->ssb_cc, pflags);
125 		if (atomic_cmpset_int(&ssb->ssb_flags, flags,
126 				      flags | SSB_WAIT)) {
127 			error = tsleep(&ssb->ssb_cc, pflags | PINTERLOCKED,
128 				       "sbwait", ssb->ssb_timeo);
129 			break;
130 		}
131 	}
132 	return (error);
133 }
134 
135 /*
136  * Lock a sockbuf already known to be locked;
137  * return any error returned from sleep (EINTR).
138  */
139 int
140 _ssb_lock(struct signalsockbuf *ssb)
141 {
142 	uint32_t flags;
143 	int pflags;
144 	int error;
145 
146 	pflags = (ssb->ssb_flags & SSB_NOINTR) ? 0 : PCATCH;
147 
148 	for (;;) {
149 		flags = ssb->ssb_flags;
150 		cpu_ccfence();
151 		if (flags & SSB_LOCK) {
152 			tsleep_interlock(&ssb->ssb_flags, pflags);
153 			if (atomic_cmpset_int(&ssb->ssb_flags, flags,
154 					      flags | SSB_WANT)) {
155 				error = tsleep(&ssb->ssb_flags,
156 					       pflags | PINTERLOCKED,
157 					       "sblock", 0);
158 				if (error)
159 					break;
160 			}
161 		} else {
162 			if (atomic_cmpset_int(&ssb->ssb_flags, flags,
163 					      flags | SSB_LOCK)) {
164 				lwkt_gettoken(&ssb->ssb_token);
165 				error = 0;
166 				break;
167 			}
168 		}
169 	}
170 	return (error);
171 }
172 
173 /*
174  * This does the same for sockbufs.  Note that the xsockbuf structure,
175  * since it is always embedded in a socket, does not include a self
176  * pointer nor a length.  We make this entry point public in case
177  * some other mechanism needs it.
178  */
179 void
180 ssbtoxsockbuf(struct signalsockbuf *ssb, struct xsockbuf *xsb)
181 {
182 	xsb->sb_cc = ssb->ssb_cc;
183 	xsb->sb_hiwat = ssb->ssb_hiwat;
184 	xsb->sb_mbcnt = ssb->ssb_mbcnt;
185 	xsb->sb_mbmax = ssb->ssb_mbmax;
186 	xsb->sb_lowat = ssb->ssb_lowat;
187 	xsb->sb_flags = ssb->ssb_flags;
188 	xsb->sb_timeo = ssb->ssb_timeo;
189 }
190 
191 
192 /************************************************************************
193  * Procedures which manipulate socket state flags, wakeups, etc.	*
194  ************************************************************************
195  *
196  * Normal sequence from the active (originating) side is that
197  * soisconnecting() is called during processing of connect() call, resulting
198  * in an eventual call to soisconnected() if/when the connection is
199  * established.  When the connection is torn down soisdisconnecting() is
200  * called during processing of disconnect() call, and soisdisconnected() is
201  * called when the connection to the peer is totally severed.
202  *
203  * The semantics of these routines are such that connectionless protocols
204  * can call soisconnected() and soisdisconnected() only, bypassing the
205  * in-progress calls when setting up a ``connection'' takes no time.
206  *
207  * From the passive side, a socket is created with two queues of sockets:
208  * so_incomp for connections in progress and so_comp for connections
209  * already made and awaiting user acceptance.  As a protocol is preparing
210  * incoming connections, it creates a socket structure queued on so_incomp
211  * by calling sonewconn().  When the connection is established,
212  * soisconnected() is called, and transfers the socket structure to so_comp,
213  * making it available to accept().
214  *
215  * If a socket is closed with sockets on either so_incomp or so_comp, these
216  * sockets are dropped.
217  *
218  * If higher level protocols are implemented in the kernel, the wakeups
219  * done here will sometimes cause software-interrupt process scheduling.
220  */
221 
222 void
223 soisconnecting(struct socket *so)
224 {
225 	soclrstate(so, SS_ISCONNECTED | SS_ISDISCONNECTING);
226 	sosetstate(so, SS_ISCONNECTING);
227 }
228 
229 void
230 soisconnected(struct socket *so)
231 {
232 	struct socket *head;
233 
234 	while ((head = so->so_head) != NULL) {
235 		lwkt_getpooltoken(head);
236 		if (so->so_head == head)
237 			break;
238 		lwkt_relpooltoken(head);
239 	}
240 
241 	soclrstate(so, SS_ISCONNECTING | SS_ISDISCONNECTING | SS_ISCONFIRMING);
242 	sosetstate(so, SS_ISCONNECTED);
243 	if (head && (so->so_state & SS_INCOMP)) {
244 		if ((so->so_options & SO_ACCEPTFILTER) != 0) {
245 			so->so_upcall = head->so_accf->so_accept_filter->accf_callback;
246 			so->so_upcallarg = head->so_accf->so_accept_filter_arg;
247 			atomic_set_int(&so->so_rcv.ssb_flags, SSB_UPCALL);
248 			so->so_options &= ~SO_ACCEPTFILTER;
249 			so->so_upcall(so, so->so_upcallarg, 0);
250 			lwkt_relpooltoken(head);
251 			return;
252 		}
253 
254 		/*
255 		 * Listen socket are not per-cpu.
256 		 */
257 		KKASSERT((so->so_state & (SS_COMP | SS_INCOMP)) == SS_INCOMP);
258 		TAILQ_REMOVE(&head->so_incomp, so, so_list);
259 		head->so_incqlen--;
260 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
261 		head->so_qlen++;
262 		sosetstate(so, SS_COMP);
263 		soclrstate(so, SS_INCOMP);
264 
265 		/*
266 		 * XXX head may be on a different protocol thread.
267 		 *     sorwakeup()->sowakeup() is hacked atm.
268 		 */
269 		sorwakeup(head);
270 		wakeup_one(&head->so_timeo);
271 	} else {
272 		wakeup(&so->so_timeo);
273 		sorwakeup(so);
274 		sowwakeup(so);
275 	}
276 	if (head)
277 		lwkt_relpooltoken(head);
278 }
279 
280 void
281 soisdisconnecting(struct socket *so)
282 {
283 	soclrstate(so, SS_ISCONNECTING);
284 	sosetstate(so, SS_ISDISCONNECTING | SS_CANTRCVMORE | SS_CANTSENDMORE);
285 	wakeup((caddr_t)&so->so_timeo);
286 	sowwakeup(so);
287 	sorwakeup(so);
288 }
289 
290 void
291 soisdisconnected(struct socket *so)
292 {
293 	soclrstate(so, SS_ISCONNECTING | SS_ISCONNECTED | SS_ISDISCONNECTING);
294 	sosetstate(so, SS_CANTRCVMORE | SS_CANTSENDMORE | SS_ISDISCONNECTED);
295 	wakeup((caddr_t)&so->so_timeo);
296 	sbdrop(&so->so_snd.sb, so->so_snd.ssb_cc);
297 	sowwakeup(so);
298 	sorwakeup(so);
299 }
300 
301 void
302 soisreconnecting(struct socket *so)
303 {
304         soclrstate(so, SS_ISDISCONNECTING | SS_ISDISCONNECTED |
305 		       SS_CANTRCVMORE | SS_CANTSENDMORE);
306 	sosetstate(so, SS_ISCONNECTING);
307 }
308 
309 void
310 soisreconnected(struct socket *so)
311 {
312 	soclrstate(so, SS_ISDISCONNECTED | SS_CANTRCVMORE | SS_CANTSENDMORE);
313 	soisconnected(so);
314 }
315 
316 /*
317  * Set or change the message port a socket receives commands on.
318  *
319  * XXX
320  */
321 void
322 sosetport(struct socket *so, lwkt_port_t port)
323 {
324 	so->so_port = port;
325 }
326 
327 /*
328  * When an attempt at a new connection is noted on a socket
329  * which accepts connections, sonewconn is called.  If the
330  * connection is possible (subject to space constraints, etc.)
331  * then we allocate a new structure, propoerly linked into the
332  * data structure of the original socket, and return this.
333  * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
334  *
335  * The new socket is returned with one ref and so_pcb assigned.
336  * The reference is implied by so_pcb.
337  */
338 struct socket *
339 sonewconn_faddr(struct socket *head, int connstatus,
340     const struct sockaddr *faddr)
341 {
342 	struct socket *so;
343 	struct socket *sp;
344 	struct pru_attach_info ai;
345 
346 	if (head->so_qlen > 3 * head->so_qlimit / 2)
347 		return (NULL);
348 	so = soalloc(1, head->so_proto);
349 	if (so == NULL)
350 		return (NULL);
351 
352 	/*
353 	 * Set the port prior to attaching the inpcb to the current
354 	 * cpu's protocol thread (which should be the current thread
355 	 * but might not be in all cases).  This serializes any pcb ops
356 	 * which occur to our cpu allowing us to complete the attachment
357 	 * without racing anything.
358 	 */
359 	if (head->so_proto->pr_flags & PR_SYNC_PORT)
360 		sosetport(so, &netisr_sync_port);
361 	else
362 		sosetport(so, netisr_cpuport(mycpuid));
363 	if ((head->so_options & SO_ACCEPTFILTER) != 0)
364 		connstatus = 0;
365 	so->so_head = head;
366 	so->so_type = head->so_type;
367 	so->so_options = head->so_options &~ SO_ACCEPTCONN;
368 	so->so_linger = head->so_linger;
369 
370 	/*
371 	 * NOTE: Clearing NOFDREF implies referencing the so with
372 	 *	 soreference().
373 	 */
374 	so->so_state = head->so_state | SS_NOFDREF | SS_ASSERTINPROG;
375 	so->so_cred = crhold(head->so_cred);
376 	ai.sb_rlimit = NULL;
377 	ai.p_ucred = NULL;
378 	ai.fd_rdir = NULL;		/* jail code cruft XXX JH */
379 
380 	/*
381 	 * Reserve space and call pru_attach.  We can direct-call the
382 	 * function since we're already in the protocol thread.
383 	 */
384 	if (soreserve(so, head->so_snd.ssb_hiwat,
385 		      head->so_rcv.ssb_hiwat, NULL) ||
386 	    so_pru_attach_direct(so, 0, &ai)) {
387 		so->so_head = NULL;
388 		soclrstate(so, SS_ASSERTINPROG);
389 		sofree(so);		/* remove implied pcb ref */
390 		return (NULL);
391 	}
392 	KKASSERT(((so->so_proto->pr_flags & PR_ASYNC_RCVD) == 0 &&
393 	    so->so_refs == 2) ||	/* attach + our base ref */
394 	   ((so->so_proto->pr_flags & PR_ASYNC_RCVD) &&
395 	    so->so_refs == 3));		/* + async rcvd ref */
396 	sofree(so);
397 	KKASSERT(so->so_port != NULL);
398 	so->so_rcv.ssb_lowat = head->so_rcv.ssb_lowat;
399 	so->so_snd.ssb_lowat = head->so_snd.ssb_lowat;
400 	so->so_rcv.ssb_timeo = head->so_rcv.ssb_timeo;
401 	so->so_snd.ssb_timeo = head->so_snd.ssb_timeo;
402 
403 	if (head->so_rcv.ssb_flags & SSB_AUTOLOWAT)
404 		so->so_rcv.ssb_flags |= SSB_AUTOLOWAT;
405 	else
406 		so->so_rcv.ssb_flags &= ~SSB_AUTOLOWAT;
407 
408 	if (head->so_snd.ssb_flags & SSB_AUTOLOWAT)
409 		so->so_snd.ssb_flags |= SSB_AUTOLOWAT;
410 	else
411 		so->so_snd.ssb_flags &= ~SSB_AUTOLOWAT;
412 
413 	if (head->so_rcv.ssb_flags & SSB_AUTOSIZE)
414 		so->so_rcv.ssb_flags |= SSB_AUTOSIZE;
415 	else
416 		so->so_rcv.ssb_flags &= ~SSB_AUTOSIZE;
417 
418 	if (head->so_snd.ssb_flags & SSB_AUTOSIZE)
419 		so->so_snd.ssb_flags |= SSB_AUTOSIZE;
420 	else
421 		so->so_snd.ssb_flags &= ~SSB_AUTOSIZE;
422 
423 	/*
424 	 * Save the faddr, if the information is provided and
425 	 * the protocol can perform the saving opertation.
426 	 */
427 	if (faddr != NULL && so->so_proto->pr_usrreqs->pru_savefaddr != NULL)
428 		so->so_proto->pr_usrreqs->pru_savefaddr(so, faddr);
429 
430 	lwkt_getpooltoken(head);
431 	if (connstatus) {
432 		KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) == 0);
433 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
434 		sosetstate(so, SS_COMP);
435 		head->so_qlen++;
436 	} else {
437 		if (head->so_incqlen > head->so_qlimit) {
438 			sp = TAILQ_FIRST(&head->so_incomp);
439 			KKASSERT((sp->so_state & (SS_INCOMP | SS_COMP)) ==
440 			    SS_INCOMP);
441 			TAILQ_REMOVE(&head->so_incomp, sp, so_list);
442 			head->so_incqlen--;
443 			soclrstate(sp, SS_INCOMP);
444 			soabort_async(sp, TRUE);
445 		}
446 		KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) == 0);
447 		TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
448 		sosetstate(so, SS_INCOMP);
449 		head->so_incqlen++;
450 	}
451 	lwkt_relpooltoken(head);
452 	if (connstatus) {
453 		/*
454 		 * XXX head may be on a different protocol thread.
455 		 *     sorwakeup()->sowakeup() is hacked atm.
456 		 */
457 		logsowakeup(nconn_start);
458 		sorwakeup(head);
459 		logsowakeup(nconn_end);
460 
461 		logsowakeup(nconn_wakeupstart);
462 		wakeup((caddr_t)&head->so_timeo);
463 		logsowakeup(nconn_wakeupend);
464 
465 		sosetstate(so, connstatus);
466 	}
467 	soclrstate(so, SS_ASSERTINPROG);
468 	return (so);
469 }
470 
471 struct socket *
472 sonewconn(struct socket *head, int connstatus)
473 {
474 	return sonewconn_faddr(head, connstatus, NULL);
475 }
476 
477 /*
478  * Socantsendmore indicates that no more data will be sent on the
479  * socket; it would normally be applied to a socket when the user
480  * informs the system that no more data is to be sent, by the protocol
481  * code (in case PRU_SHUTDOWN).  Socantrcvmore indicates that no more data
482  * will be received, and will normally be applied to the socket by a
483  * protocol when it detects that the peer will send no more data.
484  * Data queued for reading in the socket may yet be read.
485  */
486 void
487 socantsendmore(struct socket *so)
488 {
489 	sosetstate(so, SS_CANTSENDMORE);
490 	sowwakeup(so);
491 }
492 
493 void
494 socantrcvmore(struct socket *so)
495 {
496 	sosetstate(so, SS_CANTRCVMORE);
497 	sorwakeup(so);
498 }
499 
500 /*
501  * Wakeup processes waiting on a socket buffer.  Do asynchronous notification
502  * via SIGIO if the socket has the SS_ASYNC flag set.
503  *
504  * For users waiting on send/recv try to avoid unnecessary context switch
505  * thrashing.  Particularly for senders of large buffers (needs to be
506  * extended to sel and aio? XXX)
507  *
508  * WARNING!  Can be called on a foreign socket from the wrong protocol
509  *	     thread.  aka is called on the 'head' listen socket when
510  *	     a new connection comes in.
511  */
512 
513 void
514 sowakeup(struct socket *so, struct signalsockbuf *ssb)
515 {
516 	struct kqinfo *kqinfo = &ssb->ssb_kq;
517 	uint32_t flags;
518 
519 	/*
520 	 * Atomically check the flags.  When no special features are being
521 	 * used, WAIT is clear, and WAKEUP is already set, we can simply
522 	 * return.  The upcoming synchronous waiter will not block.
523 	 */
524 	flags = atomic_fetchadd_int(&ssb->ssb_flags, 0);
525 	if ((flags & SSB_NOTIFY_MASK) == 0) {
526 		if (flags & SSB_WAKEUP)
527 			return;
528 	}
529 
530 	/*
531 	 * Check conditions, set the WAKEUP flag, and clear and signal if
532 	 * the WAIT flag is found to be set.  This interlocks against the
533 	 * client side.
534 	 */
535 	for (;;) {
536 		long space;
537 
538 		flags = ssb->ssb_flags;
539 		cpu_ccfence();
540 		if (ssb->ssb_flags & SSB_PREALLOC)
541 			space = ssb_space_prealloc(ssb);
542 		else
543 			space = ssb_space(ssb);
544 
545 		if ((ssb == &so->so_snd && space >= ssb->ssb_lowat) ||
546 		    (ssb == &so->so_rcv && ssb->ssb_cc >= ssb->ssb_lowat) ||
547 		    (ssb == &so->so_snd && (so->so_state & SS_CANTSENDMORE)) ||
548 		    (ssb == &so->so_rcv && (so->so_state & SS_CANTRCVMORE))
549 		) {
550 			if (atomic_cmpset_int(&ssb->ssb_flags, flags,
551 					  (flags | SSB_WAKEUP) & ~SSB_WAIT)) {
552 				if (flags & SSB_WAIT)
553 					wakeup(&ssb->ssb_cc);
554 				break;
555 			}
556 		} else {
557 			break;
558 		}
559 	}
560 
561 	/*
562 	 * Misc other events
563 	 */
564 	if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL)
565 		pgsigio(so->so_sigio, SIGIO, 0);
566 	if (ssb->ssb_flags & SSB_UPCALL)
567 		(*so->so_upcall)(so, so->so_upcallarg, M_NOWAIT);
568 	KNOTE(&kqinfo->ki_note, 0);
569 
570 	/*
571 	 * This is a bit of a hack.  Multiple threads can wind up scanning
572 	 * ki_mlist concurrently due to the fact that this function can be
573 	 * called on a foreign socket, so we can't afford to block here.
574 	 *
575 	 * We need the pool token for (so) (likely the listne socket if
576 	 * SSB_MEVENT is set) because the predicate function may have
577 	 * to access the accept queue.
578 	 */
579 	if (ssb->ssb_flags & SSB_MEVENT) {
580 		struct netmsg_so_notify *msg, *nmsg;
581 
582 		lwkt_getpooltoken(so);
583 		TAILQ_FOREACH_MUTABLE(msg, &kqinfo->ki_mlist, nm_list, nmsg) {
584 			if (msg->nm_predicate(msg)) {
585 				TAILQ_REMOVE(&kqinfo->ki_mlist, msg, nm_list);
586 				lwkt_replymsg(&msg->base.lmsg,
587 					      msg->base.lmsg.ms_error);
588 			}
589 		}
590 		if (TAILQ_EMPTY(&ssb->ssb_kq.ki_mlist))
591 			atomic_clear_int(&ssb->ssb_flags, SSB_MEVENT);
592 		lwkt_relpooltoken(so);
593 	}
594 }
595 
596 /*
597  * Socket buffer (struct signalsockbuf) utility routines.
598  *
599  * Each socket contains two socket buffers: one for sending data and
600  * one for receiving data.  Each buffer contains a queue of mbufs,
601  * information about the number of mbufs and amount of data in the
602  * queue, and other fields allowing kevent()/select()/poll() statements
603  * and notification on data availability to be implemented.
604  *
605  * Data stored in a socket buffer is maintained as a list of records.
606  * Each record is a list of mbufs chained together with the m_next
607  * field.  Records are chained together with the m_nextpkt field. The upper
608  * level routine soreceive() expects the following conventions to be
609  * observed when placing information in the receive buffer:
610  *
611  * 1. If the protocol requires each message be preceded by the sender's
612  *    name, then a record containing that name must be present before
613  *    any associated data (mbuf's must be of type MT_SONAME).
614  * 2. If the protocol supports the exchange of ``access rights'' (really
615  *    just additional data associated with the message), and there are
616  *    ``rights'' to be received, then a record containing this data
617  *    should be present (mbuf's must be of type MT_RIGHTS).
618  * 3. If a name or rights record exists, then it must be followed by
619  *    a data record, perhaps of zero length.
620  *
621  * Before using a new socket structure it is first necessary to reserve
622  * buffer space to the socket, by calling sbreserve().  This should commit
623  * some of the available buffer space in the system buffer pool for the
624  * socket (currently, it does nothing but enforce limits).  The space
625  * should be released by calling ssb_release() when the socket is destroyed.
626  */
627 int
628 soreserve(struct socket *so, u_long sndcc, u_long rcvcc, struct rlimit *rl)
629 {
630 	if (so->so_snd.ssb_lowat == 0)
631 		atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOLOWAT);
632 	if (ssb_reserve(&so->so_snd, sndcc, so, rl) == 0)
633 		goto bad;
634 	if (ssb_reserve(&so->so_rcv, rcvcc, so, rl) == 0)
635 		goto bad2;
636 	if (so->so_rcv.ssb_lowat == 0)
637 		so->so_rcv.ssb_lowat = 1;
638 	if (so->so_snd.ssb_lowat == 0)
639 		so->so_snd.ssb_lowat = MCLBYTES;
640 	if (so->so_snd.ssb_lowat > so->so_snd.ssb_hiwat)
641 		so->so_snd.ssb_lowat = so->so_snd.ssb_hiwat;
642 	return (0);
643 bad2:
644 	ssb_release(&so->so_snd, so);
645 bad:
646 	return (ENOBUFS);
647 }
648 
649 static int
650 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS)
651 {
652 	int error = 0;
653 	u_long old_sb_max = sb_max;
654 
655 	error = SYSCTL_OUT(req, arg1, sizeof(int));
656 	if (error || !req->newptr)
657 		return (error);
658 	error = SYSCTL_IN(req, arg1, sizeof(int));
659 	if (error)
660 		return (error);
661 	if (sb_max < MSIZE + MCLBYTES) {
662 		sb_max = old_sb_max;
663 		return (EINVAL);
664 	}
665 	sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES);
666 	return (0);
667 }
668 
669 /*
670  * Allot mbufs to a signalsockbuf.
671  *
672  * Attempt to scale mbmax so that mbcnt doesn't become limiting
673  * if buffering efficiency is near the normal case.
674  *
675  * sb_max only applies to user-sockets (where rl != NULL).  It does
676  * not apply to kernel sockets or kernel-controlled sockets.  Note
677  * that NFS overrides the sockbuf limits created when nfsd creates
678  * a socket.
679  */
680 int
681 ssb_reserve(struct signalsockbuf *ssb, u_long cc, struct socket *so,
682 	    struct rlimit *rl)
683 {
684 	/*
685 	 * rl will only be NULL when we're in an interrupt (eg, in tcp_input)
686 	 * or when called from netgraph (ie, ngd_attach)
687 	 */
688 	if (rl && cc > sb_max_adj)
689 		cc = sb_max_adj;
690 	if (!chgsbsize(so->so_cred->cr_uidinfo, &ssb->ssb_hiwat, cc,
691 		       rl ? rl->rlim_cur : RLIM_INFINITY)) {
692 		return (0);
693 	}
694 	if (rl)
695 		ssb->ssb_mbmax = min(cc * sb_efficiency, sb_max);
696 	else
697 		ssb->ssb_mbmax = cc * sb_efficiency;
698 
699 	/*
700 	 * AUTOLOWAT is set on send buffers and prevents large writes
701 	 * from generating a huge number of context switches.
702 	 */
703 	if (ssb->ssb_flags & SSB_AUTOLOWAT) {
704 		ssb->ssb_lowat = ssb->ssb_hiwat / 4;
705 		if (ssb->ssb_lowat < MCLBYTES)
706 			ssb->ssb_lowat = MCLBYTES;
707 	}
708 	if (ssb->ssb_lowat > ssb->ssb_hiwat)
709 		ssb->ssb_lowat = ssb->ssb_hiwat;
710 	return (1);
711 }
712 
713 /*
714  * Free mbufs held by a socket, and reserved mbuf space.
715  */
716 void
717 ssb_release(struct signalsockbuf *ssb, struct socket *so)
718 {
719 	sbflush(&ssb->sb);
720 	(void)chgsbsize(so->so_cred->cr_uidinfo, &ssb->ssb_hiwat, 0,
721 	    RLIM_INFINITY);
722 	ssb->ssb_mbmax = 0;
723 }
724 
725 /*
726  * Some routines that return EOPNOTSUPP for entry points that are not
727  * supported by a protocol.  Fill in as needed.
728  */
729 void
730 pr_generic_notsupp(netmsg_t msg)
731 {
732 	lwkt_replymsg(&msg->lmsg, EOPNOTSUPP);
733 }
734 
735 int
736 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
737 	   struct mbuf *top, struct mbuf *control, int flags,
738 	   struct thread *td)
739 {
740 	if (top)
741 		m_freem(top);
742 	if (control)
743 		m_freem(control);
744 	return (EOPNOTSUPP);
745 }
746 
747 int
748 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
749 		      struct uio *uio, struct sockbuf *sio,
750 		      struct mbuf **controlp, int *flagsp)
751 {
752 	return (EOPNOTSUPP);
753 }
754 
755 /*
756  * This isn't really a ``null'' operation, but it's the default one
757  * and doesn't do anything destructive.
758  */
759 void
760 pru_sense_null(netmsg_t msg)
761 {
762 	msg->sense.nm_stat->st_blksize = msg->base.nm_so->so_snd.ssb_hiwat;
763 	lwkt_replymsg(&msg->lmsg, 0);
764 }
765 
766 /*
767  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.  Callers
768  * of this routine assume that it always succeeds, so we have to use a
769  * blockable allocation even though we might be called from a critical thread.
770  */
771 struct sockaddr *
772 dup_sockaddr(const struct sockaddr *sa)
773 {
774 	struct sockaddr *sa2;
775 
776 	sa2 = kmalloc(sa->sa_len, M_SONAME, M_INTWAIT);
777 	bcopy(sa, sa2, sa->sa_len);
778 	return (sa2);
779 }
780 
781 /*
782  * Create an external-format (``xsocket'') structure using the information
783  * in the kernel-format socket structure pointed to by so.  This is done
784  * to reduce the spew of irrelevant information over this interface,
785  * to isolate user code from changes in the kernel structure, and
786  * potentially to provide information-hiding if we decide that
787  * some of this information should be hidden from users.
788  */
789 void
790 sotoxsocket(struct socket *so, struct xsocket *xso)
791 {
792 	xso->xso_len = sizeof *xso;
793 	xso->xso_so = so;
794 	xso->so_type = so->so_type;
795 	xso->so_options = so->so_options;
796 	xso->so_linger = so->so_linger;
797 	xso->so_state = so->so_state;
798 	xso->so_pcb = so->so_pcb;
799 	xso->xso_protocol = so->so_proto->pr_protocol;
800 	xso->xso_family = so->so_proto->pr_domain->dom_family;
801 	xso->so_qlen = so->so_qlen;
802 	xso->so_incqlen = so->so_incqlen;
803 	xso->so_qlimit = so->so_qlimit;
804 	xso->so_timeo = so->so_timeo;
805 	xso->so_error = so->so_error;
806 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
807 	xso->so_oobmark = so->so_oobmark;
808 	ssbtoxsockbuf(&so->so_snd, &xso->so_snd);
809 	ssbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
810 	xso->so_uid = so->so_cred->cr_uid;
811 }
812 
813 /*
814  * Here is the definition of some of the basic objects in the kern.ipc
815  * branch of the MIB.
816  */
817 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
818 
819 /*
820  * This takes the place of kern.maxsockbuf, which moved to kern.ipc.
821  *
822  * NOTE! sb_max only applies to user-created socket buffers.
823  */
824 static int dummy;
825 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, "");
826 SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_INT|CTLFLAG_RW,
827     &sb_max, 0, sysctl_handle_sb_max, "I", "Maximum socket buffer size");
828 SYSCTL_INT(_kern_ipc, OID_AUTO, maxsockets, CTLFLAG_RD,
829     &maxsockets, 0, "Maximum number of sockets available");
830 SYSCTL_INT(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW,
831     &sb_efficiency, 0,
832     "Socket buffer limit scaler");
833 
834 /*
835  * Initialize maxsockets
836  */
837 static void
838 init_maxsockets(void *ignored)
839 {
840     TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
841     maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
842 }
843 SYSINIT(param, SI_BOOT1_TUNABLES, SI_ORDER_ANY,
844 	init_maxsockets, NULL);
845 
846