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 * $FreeBSD$
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/proc.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/protosw.h>
47 #include <sys/resourcevar.h>
48 #include <sys/stat.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/signalvar.h>
52 #include <sys/sysctl.h>
53 #include <sys/aio.h> /* for aio_swake proto */
54 #include <sys/event.h>
55
56 int maxsockets;
57
58 /*
59 * Primitive routines for operating on sockets and socket buffers
60 */
61
62 u_long sb_max = SB_MAX;
63 u_long sb_max_adj =
64 SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */
65
66 static 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
98 void
99 soisconnecting(so)
100 register struct socket *so;
101 {
102
103 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
104 so->so_state |= SS_ISCONNECTING;
105 }
106
107 void
108 soisconnected(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 head->so_qlen++;
129 so->so_state |= SS_COMP;
130 sorwakeup(head);
131 wakeup_one(&head->so_timeo);
132 } else {
133 wakeup(&so->so_timeo);
134 sorwakeup(so);
135 sowwakeup(so);
136 }
137 }
138
139 void
140 soisdisconnecting(so)
141 register struct socket *so;
142 {
143
144 so->so_state &= ~SS_ISCONNECTING;
145 so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE);
146 wakeup((caddr_t)&so->so_timeo);
147 sowwakeup(so);
148 sorwakeup(so);
149 }
150
151 void
152 soisdisconnected(so)
153 register struct socket *so;
154 {
155
156 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
157 so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED);
158 wakeup((caddr_t)&so->so_timeo);
159 sbdrop(&so->so_snd, so->so_snd.sb_cc);
160 sowwakeup(so);
161 sorwakeup(so);
162 }
163
164 /*
165 * When an attempt at a new connection is noted on a socket
166 * which accepts connections, sonewconn is called. If the
167 * connection is possible (subject to space constraints, etc.)
168 * then we allocate a new structure, propoerly linked into the
169 * data structure of the original socket, and return this.
170 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
171 */
172 struct socket *
173 sonewconn(head, connstatus)
174 register struct socket *head;
175 int connstatus;
176 {
177
178 return (sonewconn3(head, connstatus, NULL));
179 }
180
181 struct socket *
182 sonewconn3(head, connstatus, p)
183 register struct socket *head;
184 int connstatus;
185 struct proc *p;
186 {
187 register struct socket *so;
188
189 if (head->so_qlen > 3 * head->so_qlimit / 2)
190 return ((struct socket *)0);
191 so = soalloc(0);
192 if (so == NULL)
193 return ((struct socket *)0);
194 if ((head->so_options & SO_ACCEPTFILTER) != 0)
195 connstatus = 0;
196 so->so_head = head;
197 so->so_type = head->so_type;
198 so->so_options = head->so_options &~ SO_ACCEPTCONN;
199 so->so_linger = head->so_linger;
200 so->so_state = head->so_state | SS_NOFDREF;
201 so->so_proto = head->so_proto;
202 so->so_timeo = head->so_timeo;
203 so->so_cred = p ? p->p_ucred : head->so_cred;
204 crhold(so->so_cred);
205 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
206 (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
207 sodealloc(so);
208 return ((struct socket *)0);
209 }
210
211 if (connstatus) {
212 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
213 so->so_state |= SS_COMP;
214 head->so_qlen++;
215 } else {
216 if (head->so_incqlen > head->so_qlimit) {
217 struct socket *sp;
218 sp = TAILQ_FIRST(&head->so_incomp);
219 (void) soabort(sp);
220 }
221 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
222 so->so_state |= SS_INCOMP;
223 head->so_incqlen++;
224 }
225 if (connstatus) {
226 sorwakeup(head);
227 wakeup_one(&head->so_timeo);
228 so->so_state |= connstatus;
229 }
230 return (so);
231 }
232
233 /*
234 * Socantsendmore indicates that no more data will be sent on the
235 * socket; it would normally be applied to a socket when the user
236 * informs the system that no more data is to be sent, by the protocol
237 * code (in case PRU_SHUTDOWN). Socantrcvmore indicates that no more data
238 * will be received, and will normally be applied to the socket by a
239 * protocol when it detects that the peer will send no more data.
240 * Data queued for reading in the socket may yet be read.
241 */
242
243 void
244 socantsendmore(so)
245 struct socket *so;
246 {
247
248 so->so_state |= SS_CANTSENDMORE;
249 sowwakeup(so);
250 }
251
252 void
253 socantrcvmore(so)
254 struct socket *so;
255 {
256
257 so->so_state |= SS_CANTRCVMORE;
258 sorwakeup(so);
259 }
260
261 /*
262 * Wait for data to arrive at/drain from a socket buffer.
263 */
264 int
265 sbwait(sb)
266 struct sockbuf *sb;
267 {
268
269 sb->sb_flags |= SB_WAIT;
270 return (tsleep((caddr_t)&sb->sb_cc,
271 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait",
272 sb->sb_timeo));
273 }
274
275 /*
276 * Lock a sockbuf already known to be locked;
277 * return any error returned from sleep (EINTR).
278 */
279 int
280 sb_lock(sb)
281 register struct sockbuf *sb;
282 {
283 int error;
284
285 while (sb->sb_flags & SB_LOCK) {
286 sb->sb_flags |= SB_WANT;
287 error = tsleep((caddr_t)&sb->sb_flags,
288 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK|PCATCH,
289 "sblock", 0);
290 if (error)
291 return (error);
292 }
293 sb->sb_flags |= SB_LOCK;
294 return (0);
295 }
296
297 /*
298 * Wakeup processes waiting on a socket buffer.
299 * Do asynchronous notification via SIGIO
300 * if the socket has the SS_ASYNC flag set.
301 */
302 void
303 sowakeup(so, sb)
304 register struct socket *so;
305 register struct sockbuf *sb;
306 {
307 selwakeup(&sb->sb_sel);
308 sb->sb_flags &= ~SB_SEL;
309 if (sb->sb_flags & SB_WAIT) {
310 sb->sb_flags &= ~SB_WAIT;
311 wakeup((caddr_t)&sb->sb_cc);
312 }
313 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL)
314 pgsigio(so->so_sigio, SIGIO, 0);
315 if (sb->sb_flags & SB_UPCALL)
316 (*so->so_upcall)(so, so->so_upcallarg, M_DONTWAIT);
317 if (sb->sb_flags & SB_AIO)
318 aio_swake(so, sb);
319 KNOTE(&sb->sb_sel.si_note, 0);
320 }
321
322 /*
323 * Socket buffer (struct sockbuf) utility routines.
324 *
325 * Each socket contains two socket buffers: one for sending data and
326 * one for receiving data. Each buffer contains a queue of mbufs,
327 * information about the number of mbufs and amount of data in the
328 * queue, and other fields allowing select() statements and notification
329 * on data availability to be implemented.
330 *
331 * Data stored in a socket buffer is maintained as a list of records.
332 * Each record is a list of mbufs chained together with the m_next
333 * field. Records are chained together with the m_nextpkt field. The upper
334 * level routine soreceive() expects the following conventions to be
335 * observed when placing information in the receive buffer:
336 *
337 * 1. If the protocol requires each message be preceded by the sender's
338 * name, then a record containing that name must be present before
339 * any associated data (mbuf's must be of type MT_SONAME).
340 * 2. If the protocol supports the exchange of ``access rights'' (really
341 * just additional data associated with the message), and there are
342 * ``rights'' to be received, then a record containing this data
343 * should be present (mbuf's must be of type MT_RIGHTS).
344 * 3. If a name or rights record exists, then it must be followed by
345 * a data record, perhaps of zero length.
346 *
347 * Before using a new socket structure it is first necessary to reserve
348 * buffer space to the socket, by calling sbreserve(). This should commit
349 * some of the available buffer space in the system buffer pool for the
350 * socket (currently, it does nothing but enforce limits). The space
351 * should be released by calling sbrelease() when the socket is destroyed.
352 */
353
354 int
355 soreserve(so, sndcc, rcvcc)
356 register struct socket *so;
357 u_long sndcc, rcvcc;
358 {
359 struct proc *p = curproc;
360
361 if (sbreserve(&so->so_snd, sndcc, so, p) == 0)
362 goto bad;
363 if (sbreserve(&so->so_rcv, rcvcc, so, p) == 0)
364 goto bad2;
365 if (so->so_rcv.sb_lowat == 0)
366 so->so_rcv.sb_lowat = 1;
367 if (so->so_snd.sb_lowat == 0)
368 so->so_snd.sb_lowat = MCLBYTES;
369 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
370 so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
371 return (0);
372 bad2:
373 sbrelease(&so->so_snd, so);
374 bad:
375 return (ENOBUFS);
376 }
377
378 static int
379 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS)
380 {
381 int error = 0;
382 u_long old_sb_max = sb_max;
383
384 error = SYSCTL_OUT(req, arg1, sizeof(int));
385 if (error || !req->newptr)
386 return (error);
387 error = SYSCTL_IN(req, arg1, sizeof(int));
388 if (error)
389 return (error);
390 if (sb_max < MSIZE + MCLBYTES) {
391 sb_max = old_sb_max;
392 return (EINVAL);
393 }
394 sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES);
395 return (0);
396 }
397
398 /*
399 * Allot mbufs to a sockbuf.
400 * Attempt to scale mbmax so that mbcnt doesn't become limiting
401 * if buffering efficiency is near the normal case.
402 */
403 int
404 sbreserve(sb, cc, so, p)
405 struct sockbuf *sb;
406 u_long cc;
407 struct socket *so;
408 struct proc *p;
409 {
410
411 /*
412 * p will only be NULL when we're in an interrupt
413 * (e.g. in tcp_input())
414 */
415 if (cc > sb_max_adj)
416 return (0);
417 if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc,
418 p ? p->p_rlimit[RLIMIT_SBSIZE].rlim_cur : RLIM_INFINITY)) {
419 return (0);
420 }
421 sb->sb_mbmax = min(cc * sb_efficiency, sb_max);
422 if (sb->sb_lowat > sb->sb_hiwat)
423 sb->sb_lowat = sb->sb_hiwat;
424 return (1);
425 }
426
427 /*
428 * Free mbufs held by a socket, and reserved mbuf space.
429 */
430 void
431 sbrelease(sb, so)
432 struct sockbuf *sb;
433 struct socket *so;
434 {
435
436 sbflush(sb);
437 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0,
438 RLIM_INFINITY);
439 sb->sb_mbmax = 0;
440 }
441
442 /*
443 * Routines to add and remove
444 * data from an mbuf queue.
445 *
446 * The routines sbappend() or sbappendrecord() are normally called to
447 * append new mbufs to a socket buffer, after checking that adequate
448 * space is available, comparing the function sbspace() with the amount
449 * of data to be added. sbappendrecord() differs from sbappend() in
450 * that data supplied is treated as the beginning of a new record.
451 * To place a sender's address, optional access rights, and data in a
452 * socket receive buffer, sbappendaddr() should be used. To place
453 * access rights and data in a socket receive buffer, sbappendrights()
454 * should be used. In either case, the new data begins a new record.
455 * Note that unlike sbappend() and sbappendrecord(), these routines check
456 * for the caller that there will be enough space to store the data.
457 * Each fails if there is not enough space, or if it cannot find mbufs
458 * to store additional information in.
459 *
460 * Reliable protocols may use the socket send buffer to hold data
461 * awaiting acknowledgement. Data is normally copied from a socket
462 * send buffer in a protocol with m_copy for output to a peer,
463 * and then removing the data from the socket buffer with sbdrop()
464 * or sbdroprecord() when the data is acknowledged by the peer.
465 */
466
467 #ifdef SOCKBUF_DEBUG
468 void
469 sblastrecordchk(struct sockbuf *sb, const char *file, int line)
470 {
471 struct mbuf *m = sb->sb_mb;
472
473 while (m && m->m_nextpkt)
474 m = m->m_nextpkt;
475
476 if (m != sb->sb_lastrecord) {
477 printf("%s: sb_mb %p sb_lastrecord %p last %p\n",
478 __func__, sb->sb_mb, sb->sb_lastrecord, m);
479 printf("packet chain:\n");
480 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt)
481 printf("\t%p\n", m);
482 panic("%s from %s:%u", __func__, file, line);
483 }
484 }
485
486 void
487 sblastmbufchk(struct sockbuf *sb, const char *file, int line)
488 {
489 struct mbuf *m = sb->sb_mb;
490 struct mbuf *n;
491
492 while (m && m->m_nextpkt)
493 m = m->m_nextpkt;
494
495 while (m && m->m_next)
496 m = m->m_next;
497
498 if (m != sb->sb_mbtail) {
499 printf("%s: sb_mb %p sb_mbtail %p last %p\n",
500 __func__, sb->sb_mb, sb->sb_mbtail, m);
501 printf("packet tree:\n");
502 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) {
503 printf("\t");
504 for (n = m; n != NULL; n = n->m_next)
505 printf("%p ", n);
506 printf("\n");
507 }
508 panic("%s from %s:%u", __func__, file, line);
509 }
510 }
511 #endif /* SOCKBUF_DEBUG */
512
513 #define SBLINKRECORD(sb, m0) do { \
514 if ((sb)->sb_lastrecord != NULL) \
515 (sb)->sb_lastrecord->m_nextpkt = (m0); \
516 else \
517 (sb)->sb_mb = (m0); \
518 (sb)->sb_lastrecord = (m0); \
519 } while (/*CONSTCOND*/0)
520
521 /*
522 * Append mbuf chain m to the last record in the
523 * socket buffer sb. The additional space associated
524 * the mbuf chain is recorded in sb. Empty mbufs are
525 * discarded and mbufs are compacted where possible.
526 */
527 void
528 sbappend(sb, m)
529 struct sockbuf *sb;
530 struct mbuf *m;
531 {
532 register struct mbuf *n;
533
534 if (m == 0)
535 return;
536 SBLASTRECORDCHK(sb);
537 n = sb->sb_mb;
538 if (n) {
539 while (n->m_nextpkt)
540 n = n->m_nextpkt;
541 do {
542 if (n->m_flags & M_EOR) {
543 sbappendrecord(sb, m); /* XXXXXX!!!! */
544 return;
545 }
546 } while (n->m_next && (n = n->m_next));
547 } else {
548 /*
549 * XXX Would like to simply use sb_mbtail here, but
550 * XXX I need to verify that I won't miss an EOR that
551 * XXX way.
552 */
553 if ((n = sb->sb_lastrecord) != NULL) {
554 do {
555 if (n->m_flags & M_EOR) {
556 sbappendrecord(sb, m); /* XXXXXX!!!! */
557 return;
558 }
559 } while (n->m_next && (n = n->m_next));
560 } else {
561 /*
562 * If this is the first record in the socket buffer,
563 * it's also the last record.
564 */
565 sb->sb_lastrecord = m;
566 }
567 }
568 sbcompress(sb, m, n);
569 SBLASTRECORDCHK(sb);
570 }
571
572 /*
573 * This version of sbappend() should only be used when the caller
574 * absolutely knows that there will never be more than one record
575 * in the socket buffer, that is, a stream protocol (such as TCP).
576 */
577 void
578 sbappendstream(struct sockbuf *sb, struct mbuf *m)
579 {
580
581 KASSERT(m->m_nextpkt == NULL,("sbappendstream 0"));
582 KASSERT(sb->sb_mb == sb->sb_lastrecord,("sbappendstream 1"));
583
584 SBLASTMBUFCHK(sb);
585
586 sbcompress(sb, m, sb->sb_mbtail);
587
588 sb->sb_lastrecord = sb->sb_mb;
589 SBLASTRECORDCHK(sb);
590 }
591
592 #ifdef SOCKBUF_DEBUG
593 void
594 sbcheck(sb)
595 register struct sockbuf *sb;
596 {
597 register struct mbuf *m;
598 register struct mbuf *n = 0;
599 register u_long len = 0, mbcnt = 0;
600
601 for (m = sb->sb_mb; m; m = n) {
602 n = m->m_nextpkt;
603 for (; m; m = m->m_next) {
604 len += m->m_len;
605 mbcnt += MSIZE;
606 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */
607 mbcnt += m->m_ext.ext_size;
608 }
609 }
610 if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) {
611 printf("cc %ld != %u || mbcnt %ld != %u\n", len, sb->sb_cc,
612 mbcnt, sb->sb_mbcnt);
613 panic("sbcheck");
614 }
615 }
616 #endif
617
618 /*
619 * As above, except the mbuf chain
620 * begins a new record.
621 */
622 void
623 sbappendrecord(sb, m0)
624 register struct sockbuf *sb;
625 register struct mbuf *m0;
626 {
627 register struct mbuf *m;
628
629 if (m0 == 0)
630 return;
631 m = sb->sb_mb;
632 if (m)
633 while (m->m_nextpkt)
634 m = m->m_nextpkt;
635 /*
636 * Put the first mbuf on the queue.
637 * Note this permits zero length records.
638 */
639 sballoc(sb, m0);
640 SBLASTRECORDCHK(sb);
641 SBLINKRECORD(sb, m0);
642 if (m)
643 m->m_nextpkt = m0;
644 else
645 sb->sb_mb = m0;
646 m = m0->m_next;
647 m0->m_next = 0;
648 if (m && (m0->m_flags & M_EOR)) {
649 m0->m_flags &= ~M_EOR;
650 m->m_flags |= M_EOR;
651 }
652 sbcompress(sb, m, m0);
653 }
654
655 /*
656 * As above except that OOB data
657 * is inserted at the beginning of the sockbuf,
658 * but after any other OOB data.
659 */
660 void
661 sbinsertoob(sb, m0)
662 register struct sockbuf *sb;
663 register struct mbuf *m0;
664 {
665 register struct mbuf *m;
666 register struct mbuf **mp;
667
668 if (m0 == 0)
669 return;
670 for (mp = &sb->sb_mb; *mp ; mp = &((*mp)->m_nextpkt)) {
671 m = *mp;
672 again:
673 switch (m->m_type) {
674
675 case MT_OOBDATA:
676 continue; /* WANT next train */
677
678 case MT_CONTROL:
679 m = m->m_next;
680 if (m)
681 goto again; /* inspect THIS train further */
682 }
683 break;
684 }
685 /*
686 * Put the first mbuf on the queue.
687 * Note this permits zero length records.
688 */
689 sballoc(sb, m0);
690 m0->m_nextpkt = *mp;
691 *mp = m0;
692 m = m0->m_next;
693 m0->m_next = 0;
694 if (m && (m0->m_flags & M_EOR)) {
695 m0->m_flags &= ~M_EOR;
696 m->m_flags |= M_EOR;
697 }
698 sbcompress(sb, m, m0);
699 }
700
701 /*
702 * Append address and data, and optionally, control (ancillary) data
703 * to the receive queue of a socket. If present,
704 * m0 must include a packet header with total length.
705 * Returns 0 if no space in sockbuf or insufficient mbufs.
706 */
707 int
708 sbappendaddr(sb, asa, m0, control)
709 register struct sockbuf *sb;
710 struct sockaddr *asa;
711 struct mbuf *m0, *control;
712 {
713 register struct mbuf *m, *n, *nlast;
714 int space = asa->sa_len;
715
716 if (m0 && (m0->m_flags & M_PKTHDR) == 0)
717 panic("sbappendaddr");
718 if (m0)
719 space += m0->m_pkthdr.len;
720 for (n = control; n; n = n->m_next) {
721 space += n->m_len;
722 if (n->m_next == 0) /* keep pointer to last control buf */
723 break;
724 }
725 if (space > sbspace(sb))
726 return (0);
727 if (asa->sa_len > MLEN)
728 return (0);
729 MGET(m, M_DONTWAIT, MT_SONAME);
730 if (m == 0)
731 return (0);
732 m->m_len = asa->sa_len;
733 bcopy((caddr_t)asa, mtod(m, caddr_t), asa->sa_len);
734 if (n)
735 n->m_next = m0; /* concatenate data to control */
736 else
737 control = m0;
738 m->m_next = control;
739 for (n = m; n->m_next != NULL; n = n->m_next)
740 sballoc(sb, n);
741 sballoc(sb, n);
742 nlast = n;
743 SBLINKRECORD(sb, m);
744
745 sb->sb_mbtail = nlast;
746 SBLASTMBUFCHK(sb);
747
748 SBLASTRECORDCHK(sb);
749 return (1);
750 }
751
752 int
753 sbappendcontrol(sb, m0, control)
754 struct sockbuf *sb;
755 struct mbuf *control, *m0;
756 {
757 register struct mbuf *m, *n, *mlast;
758 int space = 0;
759
760 if (control == 0)
761 panic("sbappendcontrol");
762 for (m = control; ; m = m->m_next) {
763 space += m->m_len;
764 if (m->m_next == 0)
765 break;
766 }
767 n = m; /* save pointer to last control buffer */
768 for (m = m0; m; m = m->m_next)
769 space += m->m_len;
770 if (space > sbspace(sb))
771 return (0);
772 n->m_next = m0; /* concatenate data to control */
773
774 SBLASTRECORDCHK(sb);
775
776 for (m = control; m->m_next; m = m->m_next)
777 sballoc(sb, m);
778 sballoc(sb, m);
779 mlast = m;
780 SBLINKRECORD(sb, control);
781
782 sb->sb_mbtail = mlast;
783 SBLASTMBUFCHK(sb);
784
785 SBLASTRECORDCHK(sb);
786 return (1);
787 }
788
789 /*
790 * Compress mbuf chain m into the socket
791 * buffer sb following mbuf n. If n
792 * is null, the buffer is presumed empty.
793 */
794 void
795 sbcompress(sb, m, n)
796 register struct sockbuf *sb;
797 register struct mbuf *m, *n;
798 {
799 register int eor = 0;
800 register struct mbuf *o;
801
802 while (m) {
803 eor |= m->m_flags & M_EOR;
804 if (m->m_len == 0 &&
805 (eor == 0 ||
806 (((o = m->m_next) || (o = n)) &&
807 o->m_type == m->m_type))) {
808 if (sb->sb_lastrecord == m)
809 sb->sb_lastrecord = m->m_next;
810 m = m_free(m);
811 continue;
812 }
813 if (n && (n->m_flags & M_EOR) == 0 &&
814 M_WRITABLE(n) &&
815 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */
816 m->m_len <= M_TRAILINGSPACE(n) &&
817 n->m_type == m->m_type) {
818 bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len,
819 (unsigned)m->m_len);
820 n->m_len += m->m_len;
821 sb->sb_cc += m->m_len;
822 m = m_free(m);
823 continue;
824 }
825 if (n)
826 n->m_next = m;
827 else
828 sb->sb_mb = m;
829 sb->sb_mbtail = m;
830 sballoc(sb, m);
831 n = m;
832 m->m_flags &= ~M_EOR;
833 m = m->m_next;
834 n->m_next = 0;
835 }
836 if (eor) {
837 if (n)
838 n->m_flags |= eor;
839 else
840 printf("semi-panic: sbcompress\n");
841 }
842 SBLASTMBUFCHK(sb);
843 }
844
845 /*
846 * Free all mbufs in a sockbuf.
847 * Check that all resources are reclaimed.
848 */
849 void
850 sbflush(sb)
851 register struct sockbuf *sb;
852 {
853
854 if (sb->sb_flags & SB_LOCK)
855 panic("sbflush: locked");
856 while (sb->sb_mbcnt) {
857 /*
858 * Don't call sbdrop(sb, 0) if the leading mbuf is non-empty:
859 * we would loop forever. Panic instead.
860 */
861 if (!sb->sb_cc && (sb->sb_mb == NULL || sb->sb_mb->m_len))
862 break;
863 sbdrop(sb, (int)sb->sb_cc);
864 }
865 if (sb->sb_cc || sb->sb_mb || sb->sb_mbcnt)
866 panic("sbflush: cc %ld || mb %p || mbcnt %ld", sb->sb_cc, (void *)sb->sb_mb, sb->sb_mbcnt);
867 }
868
869 /*
870 * Drop data from (the front of) a sockbuf.
871 */
872 void
873 sbdrop(sb, len)
874 register struct sockbuf *sb;
875 register int len;
876 {
877 register struct mbuf *m;
878 struct mbuf *next;
879
880 next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
881 while (len > 0) {
882 if (m == 0) {
883 if (next == 0)
884 panic("sbdrop");
885 m = next;
886 next = m->m_nextpkt;
887 continue;
888 }
889 if (m->m_len > len) {
890 m->m_len -= len;
891 m->m_data += len;
892 sb->sb_cc -= len;
893 break;
894 }
895 len -= m->m_len;
896 sbfree(sb, m);
897 m = m_free(m);
898 }
899 while (m && m->m_len == 0) {
900 sbfree(sb, m);
901 m = m_free(m);
902 }
903 if (m) {
904 sb->sb_mb = m;
905 m->m_nextpkt = next;
906 } else
907 sb->sb_mb = next;
908 /*
909 * First part is an inline SB_EMPTY_FIXUP(). Second part
910 * makes sure sb_lastrecord is up-to-date if we dropped
911 * part of the last record.
912 */
913 m = sb->sb_mb;
914 if (m == NULL) {
915 sb->sb_mbtail = NULL;
916 sb->sb_lastrecord = NULL;
917 } else if (m->m_nextpkt == NULL) {
918 sb->sb_lastrecord = m;
919 }
920 }
921
922 /*
923 * Drop a record off the front of a sockbuf
924 * and move the next record to the front.
925 */
926 void
927 sbdroprecord(sb)
928 register struct sockbuf *sb;
929 {
930 register struct mbuf *m;
931
932 m = sb->sb_mb;
933 if (m) {
934 sb->sb_mb = m->m_nextpkt;
935 do {
936 sbfree(sb, m);
937 m = m_free(m);
938 } while (m);
939 }
940 SB_EMPTY_FIXUP(sb);
941 }
942
943 /*
944 * Create a "control" mbuf containing the specified data
945 * with the specified type for presentation on a socket buffer.
946 */
947 struct mbuf *
948 sbcreatecontrol(p, size, type, level)
949 caddr_t p;
950 register int size;
951 int type, level;
952 {
953 register struct cmsghdr *cp;
954 struct mbuf *m;
955
956 if (CMSG_SPACE((u_int)size) > MCLBYTES)
957 return ((struct mbuf *) NULL);
958 if ((m = m_get(M_DONTWAIT, MT_CONTROL)) == NULL)
959 return ((struct mbuf *) NULL);
960 if (CMSG_SPACE((u_int)size) > MLEN) {
961 MCLGET(m, M_DONTWAIT);
962 if ((m->m_flags & M_EXT) == 0) {
963 m_free(m);
964 return ((struct mbuf *) NULL);
965 }
966 }
967 cp = mtod(m, struct cmsghdr *);
968 m->m_len = 0;
969 KASSERT(CMSG_SPACE((u_int)size) <= M_TRAILINGSPACE(m),
970 ("sbcreatecontrol: short mbuf"));
971 if (p != NULL)
972 (void)memcpy(CMSG_DATA(cp), p, size);
973 m->m_len = CMSG_SPACE(size);
974 cp->cmsg_len = CMSG_LEN(size);
975 cp->cmsg_level = level;
976 cp->cmsg_type = type;
977 return (m);
978 }
979
980 /*
981 * Some routines that return EOPNOTSUPP for entry points that are not
982 * supported by a protocol. Fill in as needed.
983 */
984 int
985 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
986 {
987 return EOPNOTSUPP;
988 }
989
990 int
991 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct proc *p)
992 {
993 return EOPNOTSUPP;
994 }
995
996 int
997 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
998 {
999 return EOPNOTSUPP;
1000 }
1001
1002 int
1003 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
1004 struct ifnet *ifp, struct proc *p)
1005 {
1006 return EOPNOTSUPP;
1007 }
1008
1009 int
1010 pru_listen_notsupp(struct socket *so, struct proc *p)
1011 {
1012 return EOPNOTSUPP;
1013 }
1014
1015 int
1016 pru_rcvd_notsupp(struct socket *so, int flags)
1017 {
1018 return EOPNOTSUPP;
1019 }
1020
1021 int
1022 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
1023 {
1024 return EOPNOTSUPP;
1025 }
1026
1027 /*
1028 * This isn't really a ``null'' operation, but it's the default one
1029 * and doesn't do anything destructive.
1030 */
1031 int
1032 pru_sense_null(struct socket *so, struct stat *sb)
1033 {
1034 sb->st_blksize = so->so_snd.sb_hiwat;
1035 return 0;
1036 }
1037
1038 /*
1039 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
1040 */
1041 struct sockaddr *
1042 dup_sockaddr(sa, canwait)
1043 struct sockaddr *sa;
1044 int canwait;
1045 {
1046 struct sockaddr *sa2;
1047
1048 MALLOC(sa2, struct sockaddr *, sa->sa_len, M_SONAME,
1049 canwait ? M_WAITOK : M_NOWAIT);
1050 if (sa2)
1051 bcopy(sa, sa2, sa->sa_len);
1052 return sa2;
1053 }
1054
1055 /*
1056 * Create an external-format (``xsocket'') structure using the information
1057 * in the kernel-format socket structure pointed to by so. This is done
1058 * to reduce the spew of irrelevant information over this interface,
1059 * to isolate user code from changes in the kernel structure, and
1060 * potentially to provide information-hiding if we decide that
1061 * some of this information should be hidden from users.
1062 */
1063 void
1064 sotoxsocket(struct socket *so, struct xsocket *xso)
1065 {
1066 xso->xso_len = sizeof *xso;
1067 xso->xso_so = so;
1068 xso->so_type = so->so_type;
1069 xso->so_options = so->so_options;
1070 xso->so_linger = so->so_linger;
1071 xso->so_state = so->so_state;
1072 xso->so_pcb = so->so_pcb;
1073 xso->xso_protocol = so->so_proto->pr_protocol;
1074 xso->xso_family = so->so_proto->pr_domain->dom_family;
1075 xso->so_qlen = so->so_qlen;
1076 xso->so_incqlen = so->so_incqlen;
1077 xso->so_qlimit = so->so_qlimit;
1078 xso->so_timeo = so->so_timeo;
1079 xso->so_error = so->so_error;
1080 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
1081 xso->so_oobmark = so->so_oobmark;
1082 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
1083 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
1084 xso->so_uid = so->so_cred->cr_uid;
1085 }
1086
1087 /*
1088 * This does the same for sockbufs. Note that the xsockbuf structure,
1089 * since it is always embedded in a socket, does not include a self
1090 * pointer nor a length. We make this entry point public in case
1091 * some other mechanism needs it.
1092 */
1093 void
1094 sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb)
1095 {
1096 xsb->sb_cc = sb->sb_cc;
1097 xsb->sb_hiwat = sb->sb_hiwat;
1098 xsb->sb_mbcnt = sb->sb_mbcnt;
1099 xsb->sb_mbmax = sb->sb_mbmax;
1100 xsb->sb_lowat = sb->sb_lowat;
1101 xsb->sb_flags = sb->sb_flags;
1102 xsb->sb_timeo = sb->sb_timeo;
1103 }
1104
1105 /*
1106 * Here is the definition of some of the basic objects in the kern.ipc
1107 * branch of the MIB.
1108 */
1109 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
1110
1111 /* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */
1112 static int dummy;
1113 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, "");
1114 SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_INT|CTLFLAG_RW,
1115 &sb_max, 0, sysctl_handle_sb_max, "I", "Maximum socket buffer size");
1116 SYSCTL_INT(_kern_ipc, OID_AUTO, maxsockets, CTLFLAG_RD,
1117 &maxsockets, 0, "Maximum number of sockets avaliable");
1118 SYSCTL_INT(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW,
1119 &sb_efficiency, 0, "");
1120
1121 /*
1122 * Initialise maxsockets
1123 */
1124 static void init_maxsockets(void *ignored)
1125 {
1126 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
1127 maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
1128 }
1129 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
Cache object: 8279e7522451eb1e6e6357880c928bc6
|