1/*
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 *
33 * @(#)uipc_socket2.c 8.1 (Berkeley) 6/10/93
34 * $FreeBSD: head/sys/kern/uipc_sockbuf.c 77598 2001-06-01 21:47:34Z jesper $
35 */
36
37#include "opt_param.h"
38#include <sys/param.h>
39#include <sys/systm.h>
40#include <sys/domain.h>
41#include <sys/file.h> /* for maxfiles */
42#include <sys/kernel.h>
43#include <sys/lock.h>
44#include <sys/mutex.h>
45#include <sys/malloc.h>
46#include <sys/mbuf.h>
47#include <sys/proc.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
58int maxsockets;
59
60/*
61 * Primitive routines for operating on sockets and socket buffers
62 */
63
64u_long sb_max = SB_MAX; /* XXX should be static */
65
66static u_long sb_efficiency = 8; /* parameter for sbreserve() */
67
68/*
69 * Procedures to manipulate state flags of socket
70 * and do appropriate wakeups. Normal sequence from the
71 * active (originating) side is that soisconnecting() is
72 * called during processing of connect() call,
73 * resulting in an eventual call to soisconnected() if/when the
74 * connection is established. When the connection is torn down
75 * soisdisconnecting() is called during processing of disconnect() call,
76 * and soisdisconnected() is called when the connection to the peer
77 * is totally severed. The semantics of these routines are such that
78 * connectionless protocols can call soisconnected() and soisdisconnected()
79 * only, bypassing the in-progress calls when setting up a ``connection''
80 * takes no time.
81 *
82 * From the passive side, a socket is created with
83 * two queues of sockets: so_incomp for connections in progress
84 * and so_comp for connections already made and awaiting user acceptance.
85 * As a protocol is preparing incoming connections, it creates a socket
86 * structure queued on so_incomp by calling sonewconn(). When the connection
87 * is established, soisconnected() is called, and transfers the
88 * socket structure to so_comp, making it available to accept().
89 *
90 * If a socket is closed with sockets on either
91 * so_incomp or so_comp, these sockets are dropped.
92 *
93 * If higher level protocols are implemented in
94 * the kernel, the wakeups done here will sometimes
95 * cause software-interrupt process scheduling.
96 */
97
98void
99soisconnecting(so)
100 register struct socket *so;
101{
102
103 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
104 so->so_state |= SS_ISCONNECTING;
105}
106
107void
108soisconnected(so)
109 struct socket *so;
110{
111 struct socket *head = so->so_head;
112
113 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
114 so->so_state |= SS_ISCONNECTED;
115 if (head && (so->so_state & SS_INCOMP)) {
116 if ((so->so_options & SO_ACCEPTFILTER) != 0) {
117 so->so_upcall = head->so_accf->so_accept_filter->accf_callback;
118 so->so_upcallarg = head->so_accf->so_accept_filter_arg;
119 so->so_rcv.sb_flags |= SB_UPCALL;
120 so->so_options &= ~SO_ACCEPTFILTER;
121 so->so_upcall(so, so->so_upcallarg, 0);
122 return;
123 }
124 TAILQ_REMOVE(&head->so_incomp, so, so_list);
125 head->so_incqlen--;
126 so->so_state &= ~SS_INCOMP;
127 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
128 so->so_state |= SS_COMP;
129 sorwakeup(head);
130 wakeup_one(&head->so_timeo);
131 } else {
132 wakeup(&so->so_timeo);
133 sorwakeup(so);
134 sowwakeup(so);
135 }
136}
137
138void
139soisdisconnecting(so)
140 register struct socket *so;
141{
142
143 so->so_state &= ~SS_ISCONNECTING;
144 so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE);
145 wakeup((caddr_t)&so->so_timeo);
146 sowwakeup(so);
147 sorwakeup(so);
148}
149
150void
151soisdisconnected(so)
152 register struct socket *so;
153{
154
155 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
156 so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED);
157 wakeup((caddr_t)&so->so_timeo);
158 sowwakeup(so);
159 sorwakeup(so);
160}
161
162/*
163 * Return a random connection that hasn't been serviced yet and
164 * is eligible for discard. There is a one in qlen chance that
165 * we will return a null, saying that there are no dropable
166 * requests. In this case, the protocol specific code should drop
167 * the new request. This insures fairness.
168 *
169 * This may be used in conjunction with protocol specific queue
170 * congestion routines.
171 */
172struct socket *
173sodropablereq(head)
174 register struct socket *head;
175{
176 register struct socket *so;
177 unsigned int i, j, qlen;
178 static int rnd;
179 static struct timeval old_runtime;
180 static unsigned int cur_cnt, old_cnt;
181 struct timeval tv;
182
183 getmicrouptime(&tv);
184 if ((i = (tv.tv_sec - old_runtime.tv_sec)) != 0) {
185 old_runtime = tv;
186 old_cnt = cur_cnt / i;
187 cur_cnt = 0;
188 }
189
190 so = TAILQ_FIRST(&head->so_incomp);
191 if (!so)
192 return (so);
193
194 qlen = head->so_incqlen;
195 if (++cur_cnt > qlen || old_cnt > qlen) {
196 rnd = (314159 * rnd + 66329) & 0xffff;
197 j = ((qlen + 1) * rnd) >> 16;
198
199 while (j-- && so)
200 so = TAILQ_NEXT(so, so_list);
201 }
202
203 return (so);
204}
205
206/*
207 * When an attempt at a new connection is noted on a socket
208 * which accepts connections, sonewconn is called. If the
209 * connection is possible (subject to space constraints, etc.)
210 * then we allocate a new structure, propoerly linked into the
211 * data structure of the original socket, and return this.
212 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
213 */
214struct socket *
215sonewconn(head, connstatus)
216 register struct socket *head;
217 int connstatus;
218{
219
220 return (sonewconn3(head, connstatus, NULL));
221}
222
223struct socket *
224sonewconn3(head, connstatus, p)
225 register struct socket *head;
226 int connstatus;
227 struct proc *p;
228{
229 register struct socket *so;
230
231 if (head->so_qlen > 3 * head->so_qlimit / 2)
232 return ((struct socket *)0);
233 so = soalloc(0);
234 if (so == NULL)
235 return ((struct socket *)0);
236 so->so_head = head;
237 so->so_type = head->so_type;
238 so->so_options = head->so_options &~ SO_ACCEPTCONN;
239 so->so_linger = head->so_linger;
240 so->so_state = head->so_state | SS_NOFDREF;
241 so->so_proto = head->so_proto;
242 so->so_timeo = head->so_timeo;
243 so->so_cred = p ? p->p_ucred : head->so_cred;
244 crhold(so->so_cred);
245 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
246 (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
247 sodealloc(so);
248 return ((struct socket *)0);
249 }
250
251 if (connstatus) {
252 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
253 so->so_state |= SS_COMP;
254 } else {
255 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
256 so->so_state |= SS_INCOMP;
257 head->so_incqlen++;
258 }
259 head->so_qlen++;
260 if (connstatus) {
261 sorwakeup(head);
262 wakeup((caddr_t)&head->so_timeo);
263 so->so_state |= connstatus;
264 }
265 return (so);
266}
267
268/*
269 * Socantsendmore indicates that no more data will be sent on the
270 * socket; it would normally be applied to a socket when the user
271 * informs the system that no more data is to be sent, by the protocol
272 * code (in case PRU_SHUTDOWN). Socantrcvmore indicates that no more data
273 * will be received, and will normally be applied to the socket by a
274 * protocol when it detects that the peer will send no more data.
275 * Data queued for reading in the socket may yet be read.
276 */
277
278void
279socantsendmore(so)
280 struct socket *so;
281{
282
283 so->so_state |= SS_CANTSENDMORE;
284 sowwakeup(so);
285}
286
287void
288socantrcvmore(so)
289 struct socket *so;
290{
291
292 so->so_state |= SS_CANTRCVMORE;
293 sorwakeup(so);
294}
295
296/*
297 * Wait for data to arrive at/drain from a socket buffer.
298 */
299int
300sbwait(sb)
301 struct sockbuf *sb;
302{
303
304 sb->sb_flags |= SB_WAIT;
305 return (tsleep((caddr_t)&sb->sb_cc,
306 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait",
307 sb->sb_timeo));
308}
309
310/*
311 * Lock a sockbuf already known to be locked;
312 * return any error returned from sleep (EINTR).
313 */
314int
315sb_lock(sb)
316 register struct sockbuf *sb;
317{
318 int error;
319
320 while (sb->sb_flags & SB_LOCK) {
321 sb->sb_flags |= SB_WANT;
322 error = tsleep((caddr_t)&sb->sb_flags,
323 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK|PCATCH,
324 "sblock", 0);
325 if (error)
326 return (error);
327 }
328 sb->sb_flags |= SB_LOCK;
329 return (0);
330}
331
332/*
333 * Wakeup processes waiting on a socket buffer.
334 * Do asynchronous notification via SIGIO
335 * if the socket has the SS_ASYNC flag set.
336 */
337void
338sowakeup(so, sb)
339 register struct socket *so;
340 register struct sockbuf *sb;
341{
342 selwakeup(&sb->sb_sel);
343 sb->sb_flags &= ~SB_SEL;
344 if (sb->sb_flags & SB_WAIT) {
345 sb->sb_flags &= ~SB_WAIT;
346 wakeup((caddr_t)&sb->sb_cc);
347 }
348 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL)
349 pgsigio(so->so_sigio, SIGIO, 0);
350 if (sb->sb_flags & SB_UPCALL)
351 (*so->so_upcall)(so, so->so_upcallarg, M_DONTWAIT);
352 if (sb->sb_flags & SB_AIO)
353 aio_swake(so, sb);
354 KNOTE(&sb->sb_sel.si_note, 0);
355}
356
357/*
358 * Socket buffer (struct sockbuf) utility routines.
359 *
360 * Each socket contains two socket buffers: one for sending data and
361 * one for receiving data. Each buffer contains a queue of mbufs,
362 * information about the number of mbufs and amount of data in the
363 * queue, and other fields allowing select() statements and notification
364 * on data availability to be implemented.
365 *
366 * Data stored in a socket buffer is maintained as a list of records.
367 * Each record is a list of mbufs chained together with the m_next
368 * field. Records are chained together with the m_nextpkt field. The upper
369 * level routine soreceive() expects the following conventions to be
370 * observed when placing information in the receive buffer:
371 *
372 * 1. If the protocol requires each message be preceded by the sender's
373 * name, then a record containing that name must be present before
374 * any associated data (mbuf's must be of type MT_SONAME).
375 * 2. If the protocol supports the exchange of ``access rights'' (really
376 * just additional data associated with the message), and there are
377 * ``rights'' to be received, then a record containing this data
378 * should be present (mbuf's must be of type MT_RIGHTS).
379 * 3. If a name or rights record exists, then it must be followed by
380 * a data record, perhaps of zero length.
381 *
382 * Before using a new socket structure it is first necessary to reserve
383 * buffer space to the socket, by calling sbreserve(). This should commit
384 * some of the available buffer space in the system buffer pool for the
385 * socket (currently, it does nothing but enforce limits). The space
386 * should be released by calling sbrelease() when the socket is destroyed.
387 */
388
389int
390soreserve(so, sndcc, rcvcc)
391 register struct socket *so;
392 u_long sndcc, rcvcc;
393{
394 struct proc *p = curproc;
395
396 if (sbreserve(&so->so_snd, sndcc, so, p) == 0)
397 goto bad;
398 if (sbreserve(&so->so_rcv, rcvcc, so, p) == 0)
399 goto bad2;
400 if (so->so_rcv.sb_lowat == 0)
401 so->so_rcv.sb_lowat = 1;
402 if (so->so_snd.sb_lowat == 0)
403 so->so_snd.sb_lowat = MCLBYTES;
404 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
405 so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
406 return (0);
407bad2:
408 sbrelease(&so->so_snd, so);
409bad:
410 return (ENOBUFS);
411}
412
413/*
414 * Allot mbufs to a sockbuf.
415 * Attempt to scale mbmax so that mbcnt doesn't become limiting
416 * if buffering efficiency is near the normal case.
417 */
418int
419sbreserve(sb, cc, so, p)
420 struct sockbuf *sb;
421 u_long cc;
422 struct socket *so;
423 struct proc *p;
424{
425
426 /*
427 * p will only be NULL when we're in an interrupt
428 * (e.g. in tcp_input())
429 */
430 if ((u_quad_t)cc > (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES))
431 return (0);
432 if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc,
433 p ? p->p_rlimit[RLIMIT_SBSIZE].rlim_cur : RLIM_INFINITY)) {
434 return (0);
435 }
436 sb->sb_mbmax = min(cc * sb_efficiency, sb_max);
437 if (sb->sb_lowat > sb->sb_hiwat)
438 sb->sb_lowat = sb->sb_hiwat;
439 return (1);
440}
441
442/*
443 * Free mbufs held by a socket, and reserved mbuf space.
444 */
445void
446sbrelease(sb, so)
447 struct sockbuf *sb;
448 struct socket *so;
449{
450
451 sbflush(sb);
452 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0,
453 RLIM_INFINITY);
454 sb->sb_mbmax = 0;
455}
456
457/*
458 * Routines to add and remove
459 * data from an mbuf queue.
460 *
461 * The routines sbappend() or sbappendrecord() are normally called to
462 * append new mbufs to a socket buffer, after checking that adequate
463 * space is available, comparing the function sbspace() with the amount
464 * of data to be added. sbappendrecord() differs from sbappend() in
465 * that data supplied is treated as the beginning of a new record.
466 * To place a sender's address, optional access rights, and data in a
467 * socket receive buffer, sbappendaddr() should be used. To place
468 * access rights and data in a socket receive buffer, sbappendrights()
469 * should be used. In either case, the new data begins a new record.
470 * Note that unlike sbappend() and sbappendrecord(), these routines check
471 * for the caller that there will be enough space to store the data.
472 * Each fails if there is not enough space, or if it cannot find mbufs
473 * to store additional information in.
474 *
475 * Reliable protocols may use the socket send buffer to hold data
476 * awaiting acknowledgement. Data is normally copied from a socket
477 * send buffer in a protocol with m_copy for output to a peer,
478 * and then removing the data from the socket buffer with sbdrop()
479 * or sbdroprecord() when the data is acknowledged by the peer.
480 */
481
482/*
483 * Append mbuf chain m to the last record in the
484 * socket buffer sb. The additional space associated
485 * the mbuf chain is recorded in sb. Empty mbufs are
486 * discarded and mbufs are compacted where possible.
487 */
488void
489sbappend(sb, m)
490 struct sockbuf *sb;
491 struct mbuf *m;
492{
493 register struct mbuf *n;
494
495 if (m == 0)
496 return;
497 n = sb->sb_mb;
498 if (n) {
499 while (n->m_nextpkt)
500 n = n->m_nextpkt;
501 do {
502 if (n->m_flags & M_EOR) {
503 sbappendrecord(sb, m); /* XXXXXX!!!! */
504 return;
505 }
506 } while (n->m_next && (n = n->m_next));
507 }
508 sbcompress(sb, m, n);
509}
510
511#ifdef SOCKBUF_DEBUG
512void
513sbcheck(sb)
514 register struct sockbuf *sb;
515{
516 register struct mbuf *m;
517 register struct mbuf *n = 0;
518 register u_long len = 0, mbcnt = 0;
519
520 for (m = sb->sb_mb; m; m = n) {
521 n = m->m_nextpkt;
522 for (; m; m = m->m_next) {
523 len += m->m_len;
524 mbcnt += MSIZE;
525 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */
526 mbcnt += m->m_ext.ext_size;
527 }
528 }
529 if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) {
530 printf("cc %ld != %ld || mbcnt %ld != %ld\n", len, sb->sb_cc,
531 mbcnt, sb->sb_mbcnt);
532 panic("sbcheck");
533 }
534}
535#endif
536
537/*
538 * As above, except the mbuf chain
539 * begins a new record.
540 */
541void
542sbappendrecord(sb, m0)
543 register struct sockbuf *sb;
544 register struct mbuf *m0;
545{
546 register struct mbuf *m;
547
548 if (m0 == 0)
549 return;
550 m = sb->sb_mb;
551 if (m)
552 while (m->m_nextpkt)
553 m = m->m_nextpkt;
554 /*
555 * Put the first mbuf on the queue.
556 * Note this permits zero length records.
557 */
558 sballoc(sb, m0);
559 if (m)
560 m->m_nextpkt = m0;
561 else
562 sb->sb_mb = m0;
563 m = m0->m_next;
564 m0->m_next = 0;
565 if (m && (m0->m_flags & M_EOR)) {
566 m0->m_flags &= ~M_EOR;
567 m->m_flags |= M_EOR;
568 }
569 sbcompress(sb, m, m0);
570}
571
572/*
573 * As above except that OOB data
574 * is inserted at the beginning of the sockbuf,
575 * but after any other OOB data.
576 */
577void
578sbinsertoob(sb, m0)
579 register struct sockbuf *sb;
580 register struct mbuf *m0;
581{
582 register struct mbuf *m;
583 register struct mbuf **mp;
584
585 if (m0 == 0)
586 return;
587 for (mp = &sb->sb_mb; *mp ; mp = &((*mp)->m_nextpkt)) {
588 m = *mp;
589 again:
590 switch (m->m_type) {
591
592 case MT_OOBDATA:
593 continue; /* WANT next train */
594
595 case MT_CONTROL:
596 m = m->m_next;
597 if (m)
598 goto again; /* inspect THIS train further */
599 }
600 break;
601 }
602 /*
603 * Put the first mbuf on the queue.
604 * Note this permits zero length records.
605 */
606 sballoc(sb, m0);
607 m0->m_nextpkt = *mp;
608 *mp = m0;
609 m = m0->m_next;
610 m0->m_next = 0;
611 if (m && (m0->m_flags & M_EOR)) {
612 m0->m_flags &= ~M_EOR;
613 m->m_flags |= M_EOR;
614 }
615 sbcompress(sb, m, m0);
616}
617
618/*
619 * Append address and data, and optionally, control (ancillary) data
620 * to the receive queue of a socket. If present,
621 * m0 must include a packet header with total length.
622 * Returns 0 if no space in sockbuf or insufficient mbufs.
623 */
624int
625sbappendaddr(sb, asa, m0, control)
626 register struct sockbuf *sb;
627 struct sockaddr *asa;
628 struct mbuf *m0, *control;
629{
630 register struct mbuf *m, *n;
631 int space = asa->sa_len;
632
633if (m0 && (m0->m_flags & M_PKTHDR) == 0)
634panic("sbappendaddr");
635 if (m0)
636 space += m0->m_pkthdr.len;
637 for (n = control; n; n = n->m_next) {
638 space += n->m_len;
639 if (n->m_next == 0) /* keep pointer to last control buf */
640 break;
641 }
642 if (space > sbspace(sb))
643 return (0);
644 if (asa->sa_len > MLEN)
645 return (0);
646 MGET(m, M_DONTWAIT, MT_SONAME);
647 if (m == 0)
648 return (0);
649 m->m_len = asa->sa_len;
650 bcopy((caddr_t)asa, mtod(m, caddr_t), asa->sa_len);
651 if (n)
652 n->m_next = m0; /* concatenate data to control */
653 else
654 control = m0;
655 m->m_next = control;
656 for (n = m; n; n = n->m_next)
657 sballoc(sb, n);
658 n = sb->sb_mb;
659 if (n) {
660 while (n->m_nextpkt)
661 n = n->m_nextpkt;
662 n->m_nextpkt = m;
663 } else
664 sb->sb_mb = m;
665 return (1);
666}
667
668int
669sbappendcontrol(sb, m0, control)
670 struct sockbuf *sb;
671 struct mbuf *control, *m0;
672{
673 register struct mbuf *m, *n;
674 int space = 0;
675
676 if (control == 0)
677 panic("sbappendcontrol");
678 for (m = control; ; m = m->m_next) {
679 space += m->m_len;
680 if (m->m_next == 0)
681 break;
682 }
683 n = m; /* save pointer to last control buffer */
684 for (m = m0; m; m = m->m_next)
685 space += m->m_len;
686 if (space > sbspace(sb))
687 return (0);
688 n->m_next = m0; /* concatenate data to control */
689 for (m = control; m; m = m->m_next)
690 sballoc(sb, m);
691 n = sb->sb_mb;
692 if (n) {
693 while (n->m_nextpkt)
694 n = n->m_nextpkt;
695 n->m_nextpkt = control;
696 } else
697 sb->sb_mb = control;
698 return (1);
699}
700
701/*
702 * Compress mbuf chain m into the socket
703 * buffer sb following mbuf n. If n
704 * is null, the buffer is presumed empty.
705 */
706void
707sbcompress(sb, m, n)
708 register struct sockbuf *sb;
709 register struct mbuf *m, *n;
710{
711 register int eor = 0;
712 register struct mbuf *o;
713
714 while (m) {
715 eor |= m->m_flags & M_EOR;
716 if (m->m_len == 0 &&
717 (eor == 0 ||
718 (((o = m->m_next) || (o = n)) &&
719 o->m_type == m->m_type))) {
720 m = m_free(m);
721 continue;
722 }
723 if (n && (n->m_flags & M_EOR) == 0 &&
724 M_WRITABLE(n) &&
725 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */
726 m->m_len <= M_TRAILINGSPACE(n) &&
727 n->m_type == m->m_type) {
728 bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len,
729 (unsigned)m->m_len);
730 n->m_len += m->m_len;
731 sb->sb_cc += m->m_len;
732 m = m_free(m);
733 continue;
734 }
735 if (n)
736 n->m_next = m;
737 else
738 sb->sb_mb = m;
739 sballoc(sb, m);
740 n = m;
741 m->m_flags &= ~M_EOR;
742 m = m->m_next;
743 n->m_next = 0;
744 }
745 if (eor) {
746 if (n)
747 n->m_flags |= eor;
748 else
749 printf("semi-panic: sbcompress\n");
750 }
751}
752
753/*
754 * Free all mbufs in a sockbuf.
755 * Check that all resources are reclaimed.
756 */
757void
758sbflush(sb)
759 register struct sockbuf *sb;
760{
761
762 if (sb->sb_flags & SB_LOCK)
763 panic("sbflush: locked");
764 while (sb->sb_mbcnt) {
765 /*
766 * Don't call sbdrop(sb, 0) if the leading mbuf is non-empty:
767 * we would loop forever. Panic instead.
768 */
769 if (!sb->sb_cc && (sb->sb_mb == NULL || sb->sb_mb->m_len))
770 break;
771 sbdrop(sb, (int)sb->sb_cc);
772 }
773 if (sb->sb_cc || sb->sb_mb || sb->sb_mbcnt)
774 panic("sbflush: cc %ld || mb %p || mbcnt %ld", sb->sb_cc, (void *)sb->sb_mb, sb->sb_mbcnt);
775}
776
777/*
778 * Drop data from (the front of) a sockbuf.
779 */
780void
781sbdrop(sb, len)
782 register struct sockbuf *sb;
783 register int len;
784{
785 register struct mbuf *m, *mn;
786 struct mbuf *next;
787
788 next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
789 while (len > 0) {
790 if (m == 0) {
791 if (next == 0)
792 panic("sbdrop");
793 m = next;
794 next = m->m_nextpkt;
795 continue;
796 }
797 if (m->m_len > len) {
798 m->m_len -= len;
799 m->m_data += len;
800 sb->sb_cc -= len;
801 break;
802 }
803 len -= m->m_len;
804 sbfree(sb, m);
805 MFREE(m, mn);
806 m = mn;
807 }
808 while (m && m->m_len == 0) {
809 sbfree(sb, m);
810 MFREE(m, mn);
811 m = mn;
812 }
813 if (m) {
814 sb->sb_mb = m;
815 m->m_nextpkt = next;
816 } else
817 sb->sb_mb = next;
818}
819
820/*
821 * Drop a record off the front of a sockbuf
822 * and move the next record to the front.
823 */
824void
825sbdroprecord(sb)
826 register struct sockbuf *sb;
827{
828 register struct mbuf *m, *mn;
829
830 m = sb->sb_mb;
831 if (m) {
832 sb->sb_mb = m->m_nextpkt;
833 do {
834 sbfree(sb, m);
835 MFREE(m, mn);
836 m = mn;
837 } while (m);
838 }
839}
840
841/*
842 * Create a "control" mbuf containing the specified data
843 * with the specified type for presentation on a socket buffer.
844 */
845struct mbuf *
846sbcreatecontrol(p, size, type, level)
847 caddr_t p;
848 register int size;
849 int type, level;
850{
851 register struct cmsghdr *cp;
852 struct mbuf *m;
853
854 if (CMSG_SPACE((u_int)size) > MLEN)
855 return ((struct mbuf *) NULL);
856 if ((m = m_get(M_DONTWAIT, MT_CONTROL)) == NULL)
857 return ((struct mbuf *) NULL);
858 cp = mtod(m, struct cmsghdr *);
859 /* XXX check size? */
860 (void)memcpy(CMSG_DATA(cp), p, size);
861 m->m_len = CMSG_SPACE(size);
862 cp->cmsg_len = CMSG_LEN(size);
863 cp->cmsg_level = level;
864 cp->cmsg_type = type;
865 return (m);
866}
867
868/*
869 * Some routines that return EOPNOTSUPP for entry points that are not
870 * supported by a protocol. Fill in as needed.
871 */
872int
873pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
874{
875 return EOPNOTSUPP;
876}
877
878int
879pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct proc *p)
880{
881 return EOPNOTSUPP;
882}
883
884int
885pru_connect2_notsupp(struct socket *so1, struct socket *so2)
886{
887 return EOPNOTSUPP;
888}
889
890int
891pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
892 struct ifnet *ifp, struct proc *p)
893{
894 return EOPNOTSUPP;
895}
896
897int
898pru_listen_notsupp(struct socket *so, struct proc *p)
899{
900 return EOPNOTSUPP;
901}
902
903int
904pru_rcvd_notsupp(struct socket *so, int flags)
905{
906 return EOPNOTSUPP;
907}
908
909int
910pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
911{
912 return EOPNOTSUPP;
913}
914
915/*
916 * This isn't really a ``null'' operation, but it's the default one
917 * and doesn't do anything destructive.
918 */
919int
920pru_sense_null(struct socket *so, struct stat *sb)
921{
922 sb->st_blksize = so->so_snd.sb_hiwat;
923 return 0;
924}
925
926/*
927 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
928 */
929struct sockaddr *
930dup_sockaddr(sa, canwait)
931 struct sockaddr *sa;
932 int canwait;
933{
934 struct sockaddr *sa2;
935
936 MALLOC(sa2, struct sockaddr *, sa->sa_len, M_SONAME,
937 canwait ? M_WAITOK : M_NOWAIT);
938 if (sa2)
939 bcopy(sa, sa2, sa->sa_len);
940 return sa2;
941}
942
943/*
944 * Create an external-format (``xsocket'') structure using the information
945 * in the kernel-format socket structure pointed to by so. This is done
946 * to reduce the spew of irrelevant information over this interface,
947 * to isolate user code from changes in the kernel structure, and
948 * potentially to provide information-hiding if we decide that
949 * some of this information should be hidden from users.
950 */
951void
952sotoxsocket(struct socket *so, struct xsocket *xso)
953{
954 xso->xso_len = sizeof *xso;
955 xso->xso_so = so;
956 xso->so_type = so->so_type;
957 xso->so_options = so->so_options;
958 xso->so_linger = so->so_linger;
959 xso->so_state = so->so_state;
960 xso->so_pcb = so->so_pcb;
961 xso->xso_protocol = so->so_proto->pr_protocol;
962 xso->xso_family = so->so_proto->pr_domain->dom_family;
963 xso->so_qlen = so->so_qlen;
964 xso->so_incqlen = so->so_incqlen;
965 xso->so_qlimit = so->so_qlimit;
966 xso->so_timeo = so->so_timeo;
967 xso->so_error = so->so_error;
968 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
969 xso->so_oobmark = so->so_oobmark;
970 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
971 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
972 xso->so_uid = so->so_cred->cr_uid;
973}
974
975/*
976 * This does the same for sockbufs. Note that the xsockbuf structure,
977 * since it is always embedded in a socket, does not include a self
978 * pointer nor a length. We make this entry point public in case
979 * some other mechanism needs it.
980 */
981void
982sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb)
983{
984 xsb->sb_cc = sb->sb_cc;
985 xsb->sb_hiwat = sb->sb_hiwat;
986 xsb->sb_mbcnt = sb->sb_mbcnt;
987 xsb->sb_mbmax = sb->sb_mbmax;
988 xsb->sb_lowat = sb->sb_lowat;
989 xsb->sb_flags = sb->sb_flags;
990 xsb->sb_timeo = sb->sb_timeo;
991}
992
993/*
994 * Here is the definition of some of the basic objects in the kern.ipc
995 * branch of the MIB.
996 */
997SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
998
999/* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */
1000static int dummy;
1001SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, "");
1002
1003SYSCTL_INT(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLFLAG_RW,
1004 &sb_max, 0, "Maximum socket buffer size");
1005SYSCTL_INT(_kern_ipc, OID_AUTO, maxsockets, CTLFLAG_RD,
1006 &maxsockets, 0, "Maximum number of sockets avaliable");
1007SYSCTL_INT(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW,
1008 &sb_efficiency, 0, "");
1009
1010/*
1011 * Initialise maxsockets
1012 */
1013static void init_maxsockets(void *ignored)
1014{
1015 TUNABLE_INT_FETCH("kern.ipc.maxsockets", 0, maxsockets);
1016 maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
1017}
1018SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 *
33 * @(#)uipc_socket2.c 8.1 (Berkeley) 6/10/93
34 * $FreeBSD: head/sys/kern/uipc_sockbuf.c 77598 2001-06-01 21:47:34Z jesper $
35 */
36
37#include "opt_param.h"
38#include <sys/param.h>
39#include <sys/systm.h>
40#include <sys/domain.h>
41#include <sys/file.h> /* for maxfiles */
42#include <sys/kernel.h>
43#include <sys/lock.h>
44#include <sys/mutex.h>
45#include <sys/malloc.h>
46#include <sys/mbuf.h>
47#include <sys/proc.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
58int maxsockets;
59
60/*
61 * Primitive routines for operating on sockets and socket buffers
62 */
63
64u_long sb_max = SB_MAX; /* XXX should be static */
65
66static u_long sb_efficiency = 8; /* parameter for sbreserve() */
67
68/*
69 * Procedures to manipulate state flags of socket
70 * and do appropriate wakeups. Normal sequence from the
71 * active (originating) side is that soisconnecting() is
72 * called during processing of connect() call,
73 * resulting in an eventual call to soisconnected() if/when the
74 * connection is established. When the connection is torn down
75 * soisdisconnecting() is called during processing of disconnect() call,
76 * and soisdisconnected() is called when the connection to the peer
77 * is totally severed. The semantics of these routines are such that
78 * connectionless protocols can call soisconnected() and soisdisconnected()
79 * only, bypassing the in-progress calls when setting up a ``connection''
80 * takes no time.
81 *
82 * From the passive side, a socket is created with
83 * two queues of sockets: so_incomp for connections in progress
84 * and so_comp for connections already made and awaiting user acceptance.
85 * As a protocol is preparing incoming connections, it creates a socket
86 * structure queued on so_incomp by calling sonewconn(). When the connection
87 * is established, soisconnected() is called, and transfers the
88 * socket structure to so_comp, making it available to accept().
89 *
90 * If a socket is closed with sockets on either
91 * so_incomp or so_comp, these sockets are dropped.
92 *
93 * If higher level protocols are implemented in
94 * the kernel, the wakeups done here will sometimes
95 * cause software-interrupt process scheduling.
96 */
97
98void
99soisconnecting(so)
100 register struct socket *so;
101{
102
103 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
104 so->so_state |= SS_ISCONNECTING;
105}
106
107void
108soisconnected(so)
109 struct socket *so;
110{
111 struct socket *head = so->so_head;
112
113 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
114 so->so_state |= SS_ISCONNECTED;
115 if (head && (so->so_state & SS_INCOMP)) {
116 if ((so->so_options & SO_ACCEPTFILTER) != 0) {
117 so->so_upcall = head->so_accf->so_accept_filter->accf_callback;
118 so->so_upcallarg = head->so_accf->so_accept_filter_arg;
119 so->so_rcv.sb_flags |= SB_UPCALL;
120 so->so_options &= ~SO_ACCEPTFILTER;
121 so->so_upcall(so, so->so_upcallarg, 0);
122 return;
123 }
124 TAILQ_REMOVE(&head->so_incomp, so, so_list);
125 head->so_incqlen--;
126 so->so_state &= ~SS_INCOMP;
127 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
128 so->so_state |= SS_COMP;
129 sorwakeup(head);
130 wakeup_one(&head->so_timeo);
131 } else {
132 wakeup(&so->so_timeo);
133 sorwakeup(so);
134 sowwakeup(so);
135 }
136}
137
138void
139soisdisconnecting(so)
140 register struct socket *so;
141{
142
143 so->so_state &= ~SS_ISCONNECTING;
144 so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE);
145 wakeup((caddr_t)&so->so_timeo);
146 sowwakeup(so);
147 sorwakeup(so);
148}
149
150void
151soisdisconnected(so)
152 register struct socket *so;
153{
154
155 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
156 so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED);
157 wakeup((caddr_t)&so->so_timeo);
158 sowwakeup(so);
159 sorwakeup(so);
160}
161
162/*
163 * Return a random connection that hasn't been serviced yet and
164 * is eligible for discard. There is a one in qlen chance that
165 * we will return a null, saying that there are no dropable
166 * requests. In this case, the protocol specific code should drop
167 * the new request. This insures fairness.
168 *
169 * This may be used in conjunction with protocol specific queue
170 * congestion routines.
171 */
172struct socket *
173sodropablereq(head)
174 register struct socket *head;
175{
176 register struct socket *so;
177 unsigned int i, j, qlen;
178 static int rnd;
179 static struct timeval old_runtime;
180 static unsigned int cur_cnt, old_cnt;
181 struct timeval tv;
182
183 getmicrouptime(&tv);
184 if ((i = (tv.tv_sec - old_runtime.tv_sec)) != 0) {
185 old_runtime = tv;
186 old_cnt = cur_cnt / i;
187 cur_cnt = 0;
188 }
189
190 so = TAILQ_FIRST(&head->so_incomp);
191 if (!so)
192 return (so);
193
194 qlen = head->so_incqlen;
195 if (++cur_cnt > qlen || old_cnt > qlen) {
196 rnd = (314159 * rnd + 66329) & 0xffff;
197 j = ((qlen + 1) * rnd) >> 16;
198
199 while (j-- && so)
200 so = TAILQ_NEXT(so, so_list);
201 }
202
203 return (so);
204}
205
206/*
207 * When an attempt at a new connection is noted on a socket
208 * which accepts connections, sonewconn is called. If the
209 * connection is possible (subject to space constraints, etc.)
210 * then we allocate a new structure, propoerly linked into the
211 * data structure of the original socket, and return this.
212 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
213 */
214struct socket *
215sonewconn(head, connstatus)
216 register struct socket *head;
217 int connstatus;
218{
219
220 return (sonewconn3(head, connstatus, NULL));
221}
222
223struct socket *
224sonewconn3(head, connstatus, p)
225 register struct socket *head;
226 int connstatus;
227 struct proc *p;
228{
229 register struct socket *so;
230
231 if (head->so_qlen > 3 * head->so_qlimit / 2)
232 return ((struct socket *)0);
233 so = soalloc(0);
234 if (so == NULL)
235 return ((struct socket *)0);
236 so->so_head = head;
237 so->so_type = head->so_type;
238 so->so_options = head->so_options &~ SO_ACCEPTCONN;
239 so->so_linger = head->so_linger;
240 so->so_state = head->so_state | SS_NOFDREF;
241 so->so_proto = head->so_proto;
242 so->so_timeo = head->so_timeo;
243 so->so_cred = p ? p->p_ucred : head->so_cred;
244 crhold(so->so_cred);
245 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
246 (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
247 sodealloc(so);
248 return ((struct socket *)0);
249 }
250
251 if (connstatus) {
252 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
253 so->so_state |= SS_COMP;
254 } else {
255 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
256 so->so_state |= SS_INCOMP;
257 head->so_incqlen++;
258 }
259 head->so_qlen++;
260 if (connstatus) {
261 sorwakeup(head);
262 wakeup((caddr_t)&head->so_timeo);
263 so->so_state |= connstatus;
264 }
265 return (so);
266}
267
268/*
269 * Socantsendmore indicates that no more data will be sent on the
270 * socket; it would normally be applied to a socket when the user
271 * informs the system that no more data is to be sent, by the protocol
272 * code (in case PRU_SHUTDOWN). Socantrcvmore indicates that no more data
273 * will be received, and will normally be applied to the socket by a
274 * protocol when it detects that the peer will send no more data.
275 * Data queued for reading in the socket may yet be read.
276 */
277
278void
279socantsendmore(so)
280 struct socket *so;
281{
282
283 so->so_state |= SS_CANTSENDMORE;
284 sowwakeup(so);
285}
286
287void
288socantrcvmore(so)
289 struct socket *so;
290{
291
292 so->so_state |= SS_CANTRCVMORE;
293 sorwakeup(so);
294}
295
296/*
297 * Wait for data to arrive at/drain from a socket buffer.
298 */
299int
300sbwait(sb)
301 struct sockbuf *sb;
302{
303
304 sb->sb_flags |= SB_WAIT;
305 return (tsleep((caddr_t)&sb->sb_cc,
306 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait",
307 sb->sb_timeo));
308}
309
310/*
311 * Lock a sockbuf already known to be locked;
312 * return any error returned from sleep (EINTR).
313 */
314int
315sb_lock(sb)
316 register struct sockbuf *sb;
317{
318 int error;
319
320 while (sb->sb_flags & SB_LOCK) {
321 sb->sb_flags |= SB_WANT;
322 error = tsleep((caddr_t)&sb->sb_flags,
323 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK|PCATCH,
324 "sblock", 0);
325 if (error)
326 return (error);
327 }
328 sb->sb_flags |= SB_LOCK;
329 return (0);
330}
331
332/*
333 * Wakeup processes waiting on a socket buffer.
334 * Do asynchronous notification via SIGIO
335 * if the socket has the SS_ASYNC flag set.
336 */
337void
338sowakeup(so, sb)
339 register struct socket *so;
340 register struct sockbuf *sb;
341{
342 selwakeup(&sb->sb_sel);
343 sb->sb_flags &= ~SB_SEL;
344 if (sb->sb_flags & SB_WAIT) {
345 sb->sb_flags &= ~SB_WAIT;
346 wakeup((caddr_t)&sb->sb_cc);
347 }
348 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL)
349 pgsigio(so->so_sigio, SIGIO, 0);
350 if (sb->sb_flags & SB_UPCALL)
351 (*so->so_upcall)(so, so->so_upcallarg, M_DONTWAIT);
352 if (sb->sb_flags & SB_AIO)
353 aio_swake(so, sb);
354 KNOTE(&sb->sb_sel.si_note, 0);
355}
356
357/*
358 * Socket buffer (struct sockbuf) utility routines.
359 *
360 * Each socket contains two socket buffers: one for sending data and
361 * one for receiving data. Each buffer contains a queue of mbufs,
362 * information about the number of mbufs and amount of data in the
363 * queue, and other fields allowing select() statements and notification
364 * on data availability to be implemented.
365 *
366 * Data stored in a socket buffer is maintained as a list of records.
367 * Each record is a list of mbufs chained together with the m_next
368 * field. Records are chained together with the m_nextpkt field. The upper
369 * level routine soreceive() expects the following conventions to be
370 * observed when placing information in the receive buffer:
371 *
372 * 1. If the protocol requires each message be preceded by the sender's
373 * name, then a record containing that name must be present before
374 * any associated data (mbuf's must be of type MT_SONAME).
375 * 2. If the protocol supports the exchange of ``access rights'' (really
376 * just additional data associated with the message), and there are
377 * ``rights'' to be received, then a record containing this data
378 * should be present (mbuf's must be of type MT_RIGHTS).
379 * 3. If a name or rights record exists, then it must be followed by
380 * a data record, perhaps of zero length.
381 *
382 * Before using a new socket structure it is first necessary to reserve
383 * buffer space to the socket, by calling sbreserve(). This should commit
384 * some of the available buffer space in the system buffer pool for the
385 * socket (currently, it does nothing but enforce limits). The space
386 * should be released by calling sbrelease() when the socket is destroyed.
387 */
388
389int
390soreserve(so, sndcc, rcvcc)
391 register struct socket *so;
392 u_long sndcc, rcvcc;
393{
394 struct proc *p = curproc;
395
396 if (sbreserve(&so->so_snd, sndcc, so, p) == 0)
397 goto bad;
398 if (sbreserve(&so->so_rcv, rcvcc, so, p) == 0)
399 goto bad2;
400 if (so->so_rcv.sb_lowat == 0)
401 so->so_rcv.sb_lowat = 1;
402 if (so->so_snd.sb_lowat == 0)
403 so->so_snd.sb_lowat = MCLBYTES;
404 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
405 so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
406 return (0);
407bad2:
408 sbrelease(&so->so_snd, so);
409bad:
410 return (ENOBUFS);
411}
412
413/*
414 * Allot mbufs to a sockbuf.
415 * Attempt to scale mbmax so that mbcnt doesn't become limiting
416 * if buffering efficiency is near the normal case.
417 */
418int
419sbreserve(sb, cc, so, p)
420 struct sockbuf *sb;
421 u_long cc;
422 struct socket *so;
423 struct proc *p;
424{
425
426 /*
427 * p will only be NULL when we're in an interrupt
428 * (e.g. in tcp_input())
429 */
430 if ((u_quad_t)cc > (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES))
431 return (0);
432 if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc,
433 p ? p->p_rlimit[RLIMIT_SBSIZE].rlim_cur : RLIM_INFINITY)) {
434 return (0);
435 }
436 sb->sb_mbmax = min(cc * sb_efficiency, sb_max);
437 if (sb->sb_lowat > sb->sb_hiwat)
438 sb->sb_lowat = sb->sb_hiwat;
439 return (1);
440}
441
442/*
443 * Free mbufs held by a socket, and reserved mbuf space.
444 */
445void
446sbrelease(sb, so)
447 struct sockbuf *sb;
448 struct socket *so;
449{
450
451 sbflush(sb);
452 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0,
453 RLIM_INFINITY);
454 sb->sb_mbmax = 0;
455}
456
457/*
458 * Routines to add and remove
459 * data from an mbuf queue.
460 *
461 * The routines sbappend() or sbappendrecord() are normally called to
462 * append new mbufs to a socket buffer, after checking that adequate
463 * space is available, comparing the function sbspace() with the amount
464 * of data to be added. sbappendrecord() differs from sbappend() in
465 * that data supplied is treated as the beginning of a new record.
466 * To place a sender's address, optional access rights, and data in a
467 * socket receive buffer, sbappendaddr() should be used. To place
468 * access rights and data in a socket receive buffer, sbappendrights()
469 * should be used. In either case, the new data begins a new record.
470 * Note that unlike sbappend() and sbappendrecord(), these routines check
471 * for the caller that there will be enough space to store the data.
472 * Each fails if there is not enough space, or if it cannot find mbufs
473 * to store additional information in.
474 *
475 * Reliable protocols may use the socket send buffer to hold data
476 * awaiting acknowledgement. Data is normally copied from a socket
477 * send buffer in a protocol with m_copy for output to a peer,
478 * and then removing the data from the socket buffer with sbdrop()
479 * or sbdroprecord() when the data is acknowledged by the peer.
480 */
481
482/*
483 * Append mbuf chain m to the last record in the
484 * socket buffer sb. The additional space associated
485 * the mbuf chain is recorded in sb. Empty mbufs are
486 * discarded and mbufs are compacted where possible.
487 */
488void
489sbappend(sb, m)
490 struct sockbuf *sb;
491 struct mbuf *m;
492{
493 register struct mbuf *n;
494
495 if (m == 0)
496 return;
497 n = sb->sb_mb;
498 if (n) {
499 while (n->m_nextpkt)
500 n = n->m_nextpkt;
501 do {
502 if (n->m_flags & M_EOR) {
503 sbappendrecord(sb, m); /* XXXXXX!!!! */
504 return;
505 }
506 } while (n->m_next && (n = n->m_next));
507 }
508 sbcompress(sb, m, n);
509}
510
511#ifdef SOCKBUF_DEBUG
512void
513sbcheck(sb)
514 register struct sockbuf *sb;
515{
516 register struct mbuf *m;
517 register struct mbuf *n = 0;
518 register u_long len = 0, mbcnt = 0;
519
520 for (m = sb->sb_mb; m; m = n) {
521 n = m->m_nextpkt;
522 for (; m; m = m->m_next) {
523 len += m->m_len;
524 mbcnt += MSIZE;
525 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */
526 mbcnt += m->m_ext.ext_size;
527 }
528 }
529 if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) {
530 printf("cc %ld != %ld || mbcnt %ld != %ld\n", len, sb->sb_cc,
531 mbcnt, sb->sb_mbcnt);
532 panic("sbcheck");
533 }
534}
535#endif
536
537/*
538 * As above, except the mbuf chain
539 * begins a new record.
540 */
541void
542sbappendrecord(sb, m0)
543 register struct sockbuf *sb;
544 register struct mbuf *m0;
545{
546 register struct mbuf *m;
547
548 if (m0 == 0)
549 return;
550 m = sb->sb_mb;
551 if (m)
552 while (m->m_nextpkt)
553 m = m->m_nextpkt;
554 /*
555 * Put the first mbuf on the queue.
556 * Note this permits zero length records.
557 */
558 sballoc(sb, m0);
559 if (m)
560 m->m_nextpkt = m0;
561 else
562 sb->sb_mb = m0;
563 m = m0->m_next;
564 m0->m_next = 0;
565 if (m && (m0->m_flags & M_EOR)) {
566 m0->m_flags &= ~M_EOR;
567 m->m_flags |= M_EOR;
568 }
569 sbcompress(sb, m, m0);
570}
571
572/*
573 * As above except that OOB data
574 * is inserted at the beginning of the sockbuf,
575 * but after any other OOB data.
576 */
577void
578sbinsertoob(sb, m0)
579 register struct sockbuf *sb;
580 register struct mbuf *m0;
581{
582 register struct mbuf *m;
583 register struct mbuf **mp;
584
585 if (m0 == 0)
586 return;
587 for (mp = &sb->sb_mb; *mp ; mp = &((*mp)->m_nextpkt)) {
588 m = *mp;
589 again:
590 switch (m->m_type) {
591
592 case MT_OOBDATA:
593 continue; /* WANT next train */
594
595 case MT_CONTROL:
596 m = m->m_next;
597 if (m)
598 goto again; /* inspect THIS train further */
599 }
600 break;
601 }
602 /*
603 * Put the first mbuf on the queue.
604 * Note this permits zero length records.
605 */
606 sballoc(sb, m0);
607 m0->m_nextpkt = *mp;
608 *mp = m0;
609 m = m0->m_next;
610 m0->m_next = 0;
611 if (m && (m0->m_flags & M_EOR)) {
612 m0->m_flags &= ~M_EOR;
613 m->m_flags |= M_EOR;
614 }
615 sbcompress(sb, m, m0);
616}
617
618/*
619 * Append address and data, and optionally, control (ancillary) data
620 * to the receive queue of a socket. If present,
621 * m0 must include a packet header with total length.
622 * Returns 0 if no space in sockbuf or insufficient mbufs.
623 */
624int
625sbappendaddr(sb, asa, m0, control)
626 register struct sockbuf *sb;
627 struct sockaddr *asa;
628 struct mbuf *m0, *control;
629{
630 register struct mbuf *m, *n;
631 int space = asa->sa_len;
632
633if (m0 && (m0->m_flags & M_PKTHDR) == 0)
634panic("sbappendaddr");
635 if (m0)
636 space += m0->m_pkthdr.len;
637 for (n = control; n; n = n->m_next) {
638 space += n->m_len;
639 if (n->m_next == 0) /* keep pointer to last control buf */
640 break;
641 }
642 if (space > sbspace(sb))
643 return (0);
644 if (asa->sa_len > MLEN)
645 return (0);
646 MGET(m, M_DONTWAIT, MT_SONAME);
647 if (m == 0)
648 return (0);
649 m->m_len = asa->sa_len;
650 bcopy((caddr_t)asa, mtod(m, caddr_t), asa->sa_len);
651 if (n)
652 n->m_next = m0; /* concatenate data to control */
653 else
654 control = m0;
655 m->m_next = control;
656 for (n = m; n; n = n->m_next)
657 sballoc(sb, n);
658 n = sb->sb_mb;
659 if (n) {
660 while (n->m_nextpkt)
661 n = n->m_nextpkt;
662 n->m_nextpkt = m;
663 } else
664 sb->sb_mb = m;
665 return (1);
666}
667
668int
669sbappendcontrol(sb, m0, control)
670 struct sockbuf *sb;
671 struct mbuf *control, *m0;
672{
673 register struct mbuf *m, *n;
674 int space = 0;
675
676 if (control == 0)
677 panic("sbappendcontrol");
678 for (m = control; ; m = m->m_next) {
679 space += m->m_len;
680 if (m->m_next == 0)
681 break;
682 }
683 n = m; /* save pointer to last control buffer */
684 for (m = m0; m; m = m->m_next)
685 space += m->m_len;
686 if (space > sbspace(sb))
687 return (0);
688 n->m_next = m0; /* concatenate data to control */
689 for (m = control; m; m = m->m_next)
690 sballoc(sb, m);
691 n = sb->sb_mb;
692 if (n) {
693 while (n->m_nextpkt)
694 n = n->m_nextpkt;
695 n->m_nextpkt = control;
696 } else
697 sb->sb_mb = control;
698 return (1);
699}
700
701/*
702 * Compress mbuf chain m into the socket
703 * buffer sb following mbuf n. If n
704 * is null, the buffer is presumed empty.
705 */
706void
707sbcompress(sb, m, n)
708 register struct sockbuf *sb;
709 register struct mbuf *m, *n;
710{
711 register int eor = 0;
712 register struct mbuf *o;
713
714 while (m) {
715 eor |= m->m_flags & M_EOR;
716 if (m->m_len == 0 &&
717 (eor == 0 ||
718 (((o = m->m_next) || (o = n)) &&
719 o->m_type == m->m_type))) {
720 m = m_free(m);
721 continue;
722 }
723 if (n && (n->m_flags & M_EOR) == 0 &&
724 M_WRITABLE(n) &&
725 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */
726 m->m_len <= M_TRAILINGSPACE(n) &&
727 n->m_type == m->m_type) {
728 bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len,
729 (unsigned)m->m_len);
730 n->m_len += m->m_len;
731 sb->sb_cc += m->m_len;
732 m = m_free(m);
733 continue;
734 }
735 if (n)
736 n->m_next = m;
737 else
738 sb->sb_mb = m;
739 sballoc(sb, m);
740 n = m;
741 m->m_flags &= ~M_EOR;
742 m = m->m_next;
743 n->m_next = 0;
744 }
745 if (eor) {
746 if (n)
747 n->m_flags |= eor;
748 else
749 printf("semi-panic: sbcompress\n");
750 }
751}
752
753/*
754 * Free all mbufs in a sockbuf.
755 * Check that all resources are reclaimed.
756 */
757void
758sbflush(sb)
759 register struct sockbuf *sb;
760{
761
762 if (sb->sb_flags & SB_LOCK)
763 panic("sbflush: locked");
764 while (sb->sb_mbcnt) {
765 /*
766 * Don't call sbdrop(sb, 0) if the leading mbuf is non-empty:
767 * we would loop forever. Panic instead.
768 */
769 if (!sb->sb_cc && (sb->sb_mb == NULL || sb->sb_mb->m_len))
770 break;
771 sbdrop(sb, (int)sb->sb_cc);
772 }
773 if (sb->sb_cc || sb->sb_mb || sb->sb_mbcnt)
774 panic("sbflush: cc %ld || mb %p || mbcnt %ld", sb->sb_cc, (void *)sb->sb_mb, sb->sb_mbcnt);
775}
776
777/*
778 * Drop data from (the front of) a sockbuf.
779 */
780void
781sbdrop(sb, len)
782 register struct sockbuf *sb;
783 register int len;
784{
785 register struct mbuf *m, *mn;
786 struct mbuf *next;
787
788 next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
789 while (len > 0) {
790 if (m == 0) {
791 if (next == 0)
792 panic("sbdrop");
793 m = next;
794 next = m->m_nextpkt;
795 continue;
796 }
797 if (m->m_len > len) {
798 m->m_len -= len;
799 m->m_data += len;
800 sb->sb_cc -= len;
801 break;
802 }
803 len -= m->m_len;
804 sbfree(sb, m);
805 MFREE(m, mn);
806 m = mn;
807 }
808 while (m && m->m_len == 0) {
809 sbfree(sb, m);
810 MFREE(m, mn);
811 m = mn;
812 }
813 if (m) {
814 sb->sb_mb = m;
815 m->m_nextpkt = next;
816 } else
817 sb->sb_mb = next;
818}
819
820/*
821 * Drop a record off the front of a sockbuf
822 * and move the next record to the front.
823 */
824void
825sbdroprecord(sb)
826 register struct sockbuf *sb;
827{
828 register struct mbuf *m, *mn;
829
830 m = sb->sb_mb;
831 if (m) {
832 sb->sb_mb = m->m_nextpkt;
833 do {
834 sbfree(sb, m);
835 MFREE(m, mn);
836 m = mn;
837 } while (m);
838 }
839}
840
841/*
842 * Create a "control" mbuf containing the specified data
843 * with the specified type for presentation on a socket buffer.
844 */
845struct mbuf *
846sbcreatecontrol(p, size, type, level)
847 caddr_t p;
848 register int size;
849 int type, level;
850{
851 register struct cmsghdr *cp;
852 struct mbuf *m;
853
854 if (CMSG_SPACE((u_int)size) > MLEN)
855 return ((struct mbuf *) NULL);
856 if ((m = m_get(M_DONTWAIT, MT_CONTROL)) == NULL)
857 return ((struct mbuf *) NULL);
858 cp = mtod(m, struct cmsghdr *);
859 /* XXX check size? */
860 (void)memcpy(CMSG_DATA(cp), p, size);
861 m->m_len = CMSG_SPACE(size);
862 cp->cmsg_len = CMSG_LEN(size);
863 cp->cmsg_level = level;
864 cp->cmsg_type = type;
865 return (m);
866}
867
868/*
869 * Some routines that return EOPNOTSUPP for entry points that are not
870 * supported by a protocol. Fill in as needed.
871 */
872int
873pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
874{
875 return EOPNOTSUPP;
876}
877
878int
879pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct proc *p)
880{
881 return EOPNOTSUPP;
882}
883
884int
885pru_connect2_notsupp(struct socket *so1, struct socket *so2)
886{
887 return EOPNOTSUPP;
888}
889
890int
891pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
892 struct ifnet *ifp, struct proc *p)
893{
894 return EOPNOTSUPP;
895}
896
897int
898pru_listen_notsupp(struct socket *so, struct proc *p)
899{
900 return EOPNOTSUPP;
901}
902
903int
904pru_rcvd_notsupp(struct socket *so, int flags)
905{
906 return EOPNOTSUPP;
907}
908
909int
910pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
911{
912 return EOPNOTSUPP;
913}
914
915/*
916 * This isn't really a ``null'' operation, but it's the default one
917 * and doesn't do anything destructive.
918 */
919int
920pru_sense_null(struct socket *so, struct stat *sb)
921{
922 sb->st_blksize = so->so_snd.sb_hiwat;
923 return 0;
924}
925
926/*
927 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
928 */
929struct sockaddr *
930dup_sockaddr(sa, canwait)
931 struct sockaddr *sa;
932 int canwait;
933{
934 struct sockaddr *sa2;
935
936 MALLOC(sa2, struct sockaddr *, sa->sa_len, M_SONAME,
937 canwait ? M_WAITOK : M_NOWAIT);
938 if (sa2)
939 bcopy(sa, sa2, sa->sa_len);
940 return sa2;
941}
942
943/*
944 * Create an external-format (``xsocket'') structure using the information
945 * in the kernel-format socket structure pointed to by so. This is done
946 * to reduce the spew of irrelevant information over this interface,
947 * to isolate user code from changes in the kernel structure, and
948 * potentially to provide information-hiding if we decide that
949 * some of this information should be hidden from users.
950 */
951void
952sotoxsocket(struct socket *so, struct xsocket *xso)
953{
954 xso->xso_len = sizeof *xso;
955 xso->xso_so = so;
956 xso->so_type = so->so_type;
957 xso->so_options = so->so_options;
958 xso->so_linger = so->so_linger;
959 xso->so_state = so->so_state;
960 xso->so_pcb = so->so_pcb;
961 xso->xso_protocol = so->so_proto->pr_protocol;
962 xso->xso_family = so->so_proto->pr_domain->dom_family;
963 xso->so_qlen = so->so_qlen;
964 xso->so_incqlen = so->so_incqlen;
965 xso->so_qlimit = so->so_qlimit;
966 xso->so_timeo = so->so_timeo;
967 xso->so_error = so->so_error;
968 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
969 xso->so_oobmark = so->so_oobmark;
970 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
971 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
972 xso->so_uid = so->so_cred->cr_uid;
973}
974
975/*
976 * This does the same for sockbufs. Note that the xsockbuf structure,
977 * since it is always embedded in a socket, does not include a self
978 * pointer nor a length. We make this entry point public in case
979 * some other mechanism needs it.
980 */
981void
982sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb)
983{
984 xsb->sb_cc = sb->sb_cc;
985 xsb->sb_hiwat = sb->sb_hiwat;
986 xsb->sb_mbcnt = sb->sb_mbcnt;
987 xsb->sb_mbmax = sb->sb_mbmax;
988 xsb->sb_lowat = sb->sb_lowat;
989 xsb->sb_flags = sb->sb_flags;
990 xsb->sb_timeo = sb->sb_timeo;
991}
992
993/*
994 * Here is the definition of some of the basic objects in the kern.ipc
995 * branch of the MIB.
996 */
997SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
998
999/* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */
1000static int dummy;
1001SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, "");
1002
1003SYSCTL_INT(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLFLAG_RW,
1004 &sb_max, 0, "Maximum socket buffer size");
1005SYSCTL_INT(_kern_ipc, OID_AUTO, maxsockets, CTLFLAG_RD,
1006 &maxsockets, 0, "Maximum number of sockets avaliable");
1007SYSCTL_INT(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW,
1008 &sb_efficiency, 0, "");
1009
1010/*
1011 * Initialise maxsockets
1012 */
1013static void init_maxsockets(void *ignored)
1014{
1015 TUNABLE_INT_FETCH("kern.ipc.maxsockets", 0, maxsockets);
1016 maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
1017}
1018SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);