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 */
35
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD: releng/5.2/sys/kern/uipc_socket2.c 122875 2003-11-18 00:39:07Z rwatson $");
38
39 #include "opt_mac.h"
40 #include "opt_param.h"
41
42 #include <sys/param.h>
43 #include <sys/aio.h> /* for aio_swake proto */
44 #include <sys/domain.h>
45 #include <sys/event.h>
46 #include <sys/file.h> /* for maxfiles */
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/mac.h>
50 #include <sys/malloc.h>
51 #include <sys/mbuf.h>
52 #include <sys/mutex.h>
53 #include <sys/proc.h>
54 #include <sys/protosw.h>
55 #include <sys/resourcevar.h>
56 #include <sys/signalvar.h>
57 #include <sys/socket.h>
58 #include <sys/socketvar.h>
59 #include <sys/stat.h>
60 #include <sys/sysctl.h>
61 #include <sys/systm.h>
62
63 int maxsockets;
64
65 void (*aio_swake)(struct socket *, struct sockbuf *);
66
67 /*
68 * Primitive routines for operating on sockets and socket buffers
69 */
70
71 u_long sb_max = SB_MAX;
72 static u_long sb_max_adj =
73 SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */
74
75 static u_long sb_efficiency = 8; /* parameter for sbreserve() */
76
77 /*
78 * Procedures to manipulate state flags of socket
79 * and do appropriate wakeups. Normal sequence from the
80 * active (originating) side is that soisconnecting() is
81 * called during processing of connect() call,
82 * resulting in an eventual call to soisconnected() if/when the
83 * connection is established. When the connection is torn down
84 * soisdisconnecting() is called during processing of disconnect() call,
85 * and soisdisconnected() is called when the connection to the peer
86 * is totally severed. The semantics of these routines are such that
87 * connectionless protocols can call soisconnected() and soisdisconnected()
88 * only, bypassing the in-progress calls when setting up a ``connection''
89 * takes no time.
90 *
91 * From the passive side, a socket is created with
92 * two queues of sockets: so_incomp for connections in progress
93 * and so_comp for connections already made and awaiting user acceptance.
94 * As a protocol is preparing incoming connections, it creates a socket
95 * structure queued on so_incomp by calling sonewconn(). When the connection
96 * is established, soisconnected() is called, and transfers the
97 * socket structure to so_comp, making it available to accept().
98 *
99 * If a socket is closed with sockets on either
100 * so_incomp or so_comp, these sockets are dropped.
101 *
102 * If higher level protocols are implemented in
103 * the kernel, the wakeups done here will sometimes
104 * cause software-interrupt process scheduling.
105 */
106
107 void
108 soisconnecting(so)
109 register struct socket *so;
110 {
111
112 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
113 so->so_state |= SS_ISCONNECTING;
114 }
115
116 void
117 soisconnected(so)
118 struct socket *so;
119 {
120 struct socket *head = so->so_head;
121
122 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
123 so->so_state |= SS_ISCONNECTED;
124 if (head && (so->so_state & SS_INCOMP)) {
125 if ((so->so_options & SO_ACCEPTFILTER) != 0) {
126 so->so_upcall = head->so_accf->so_accept_filter->accf_callback;
127 so->so_upcallarg = head->so_accf->so_accept_filter_arg;
128 so->so_rcv.sb_flags |= SB_UPCALL;
129 so->so_options &= ~SO_ACCEPTFILTER;
130 so->so_upcall(so, so->so_upcallarg, M_TRYWAIT);
131 return;
132 }
133 TAILQ_REMOVE(&head->so_incomp, so, so_list);
134 head->so_incqlen--;
135 so->so_state &= ~SS_INCOMP;
136 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
137 head->so_qlen++;
138 so->so_state |= SS_COMP;
139 sorwakeup(head);
140 wakeup_one(&head->so_timeo);
141 } else {
142 wakeup(&so->so_timeo);
143 sorwakeup(so);
144 sowwakeup(so);
145 }
146 }
147
148 void
149 soisdisconnecting(so)
150 register struct socket *so;
151 {
152
153 so->so_state &= ~SS_ISCONNECTING;
154 so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE);
155 wakeup(&so->so_timeo);
156 sowwakeup(so);
157 sorwakeup(so);
158 }
159
160 void
161 soisdisconnected(so)
162 register struct socket *so;
163 {
164
165 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
166 so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED);
167 wakeup(&so->so_timeo);
168 sbdrop(&so->so_snd, so->so_snd.sb_cc);
169 sowwakeup(so);
170 sorwakeup(so);
171 }
172
173 /*
174 * When an attempt at a new connection is noted on a socket
175 * which accepts connections, sonewconn is called. If the
176 * connection is possible (subject to space constraints, etc.)
177 * then we allocate a new structure, propoerly linked into the
178 * data structure of the original socket, and return this.
179 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
180 *
181 * note: the ref count on the socket is 0 on return
182 */
183 struct socket *
184 sonewconn(head, connstatus)
185 register struct socket *head;
186 int connstatus;
187 {
188 register struct socket *so;
189
190 if (head->so_qlen > 3 * head->so_qlimit / 2)
191 return ((struct socket *)0);
192 so = soalloc(0);
193 if (so == NULL)
194 return ((struct socket *)0);
195 if ((head->so_options & SO_ACCEPTFILTER) != 0)
196 connstatus = 0;
197 so->so_head = head;
198 so->so_type = head->so_type;
199 so->so_options = head->so_options &~ SO_ACCEPTCONN;
200 so->so_linger = head->so_linger;
201 so->so_state = head->so_state | SS_NOFDREF;
202 so->so_proto = head->so_proto;
203 so->so_timeo = head->so_timeo;
204 so->so_cred = crhold(head->so_cred);
205 #ifdef MAC
206 mac_create_socket_from_socket(head, so);
207 #endif
208 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
209 (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
210 sodealloc(so);
211 return ((struct socket *)0);
212 }
213
214 if (connstatus) {
215 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
216 so->so_state |= SS_COMP;
217 head->so_qlen++;
218 } else {
219 if (head->so_incqlen > head->so_qlimit) {
220 struct socket *sp;
221 sp = TAILQ_FIRST(&head->so_incomp);
222 (void) soabort(sp);
223 }
224 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
225 so->so_state |= SS_INCOMP;
226 head->so_incqlen++;
227 }
228 if (connstatus) {
229 sorwakeup(head);
230 wakeup(&head->so_timeo);
231 so->so_state |= connstatus;
232 }
233 return (so);
234 }
235
236 /*
237 * Socantsendmore indicates that no more data will be sent on the
238 * socket; it would normally be applied to a socket when the user
239 * informs the system that no more data is to be sent, by the protocol
240 * code (in case PRU_SHUTDOWN). Socantrcvmore indicates that no more data
241 * will be received, and will normally be applied to the socket by a
242 * protocol when it detects that the peer will send no more data.
243 * Data queued for reading in the socket may yet be read.
244 */
245
246 void
247 socantsendmore(so)
248 struct socket *so;
249 {
250
251 so->so_state |= SS_CANTSENDMORE;
252 sowwakeup(so);
253 }
254
255 void
256 socantrcvmore(so)
257 struct socket *so;
258 {
259
260 so->so_state |= SS_CANTRCVMORE;
261 sorwakeup(so);
262 }
263
264 /*
265 * Wait for data to arrive at/drain from a socket buffer.
266 */
267 int
268 sbwait(sb)
269 struct sockbuf *sb;
270 {
271
272 sb->sb_flags |= SB_WAIT;
273 return (tsleep(&sb->sb_cc,
274 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait",
275 sb->sb_timeo));
276 }
277
278 /*
279 * Lock a sockbuf already known to be locked;
280 * return any error returned from sleep (EINTR).
281 */
282 int
283 sb_lock(sb)
284 register struct sockbuf *sb;
285 {
286 int error;
287
288 while (sb->sb_flags & SB_LOCK) {
289 sb->sb_flags |= SB_WANT;
290 error = tsleep(&sb->sb_flags,
291 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK|PCATCH,
292 "sblock", 0);
293 if (error)
294 return (error);
295 }
296 sb->sb_flags |= SB_LOCK;
297 return (0);
298 }
299
300 /*
301 * Wakeup processes waiting on a socket buffer.
302 * Do asynchronous notification via SIGIO
303 * if the socket has the SS_ASYNC flag set.
304 */
305 void
306 sowakeup(so, sb)
307 register struct socket *so;
308 register struct sockbuf *sb;
309 {
310
311 selwakeuppri(&sb->sb_sel, PSOCK);
312 sb->sb_flags &= ~SB_SEL;
313 if (sb->sb_flags & SB_WAIT) {
314 sb->sb_flags &= ~SB_WAIT;
315 wakeup(&sb->sb_cc);
316 }
317 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL)
318 pgsigio(&so->so_sigio, SIGIO, 0);
319 if (sb->sb_flags & SB_UPCALL)
320 (*so->so_upcall)(so, so->so_upcallarg, M_DONTWAIT);
321 if (sb->sb_flags & SB_AIO)
322 aio_swake(so, sb);
323 KNOTE(&sb->sb_sel.si_note, 0);
324 }
325
326 /*
327 * Socket buffer (struct sockbuf) utility routines.
328 *
329 * Each socket contains two socket buffers: one for sending data and
330 * one for receiving data. Each buffer contains a queue of mbufs,
331 * information about the number of mbufs and amount of data in the
332 * queue, and other fields allowing select() statements and notification
333 * on data availability to be implemented.
334 *
335 * Data stored in a socket buffer is maintained as a list of records.
336 * Each record is a list of mbufs chained together with the m_next
337 * field. Records are chained together with the m_nextpkt field. The upper
338 * level routine soreceive() expects the following conventions to be
339 * observed when placing information in the receive buffer:
340 *
341 * 1. If the protocol requires each message be preceded by the sender's
342 * name, then a record containing that name must be present before
343 * any associated data (mbuf's must be of type MT_SONAME).
344 * 2. If the protocol supports the exchange of ``access rights'' (really
345 * just additional data associated with the message), and there are
346 * ``rights'' to be received, then a record containing this data
347 * should be present (mbuf's must be of type MT_RIGHTS).
348 * 3. If a name or rights record exists, then it must be followed by
349 * a data record, perhaps of zero length.
350 *
351 * Before using a new socket structure it is first necessary to reserve
352 * buffer space to the socket, by calling sbreserve(). This should commit
353 * some of the available buffer space in the system buffer pool for the
354 * socket (currently, it does nothing but enforce limits). The space
355 * should be released by calling sbrelease() when the socket is destroyed.
356 */
357
358 int
359 soreserve(so, sndcc, rcvcc)
360 register struct socket *so;
361 u_long sndcc, rcvcc;
362 {
363 struct thread *td = curthread;
364
365 if (sbreserve(&so->so_snd, sndcc, so, td) == 0)
366 goto bad;
367 if (sbreserve(&so->so_rcv, rcvcc, so, td) == 0)
368 goto bad2;
369 if (so->so_rcv.sb_lowat == 0)
370 so->so_rcv.sb_lowat = 1;
371 if (so->so_snd.sb_lowat == 0)
372 so->so_snd.sb_lowat = MCLBYTES;
373 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
374 so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
375 return (0);
376 bad2:
377 sbrelease(&so->so_snd, so);
378 bad:
379 return (ENOBUFS);
380 }
381
382 static int
383 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS)
384 {
385 int error = 0;
386 u_long old_sb_max = sb_max;
387
388 error = SYSCTL_OUT(req, arg1, sizeof(u_long));
389 if (error || !req->newptr)
390 return (error);
391 error = SYSCTL_IN(req, arg1, sizeof(u_long));
392 if (error)
393 return (error);
394 if (sb_max < MSIZE + MCLBYTES) {
395 sb_max = old_sb_max;
396 return (EINVAL);
397 }
398 sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES);
399 return (0);
400 }
401
402 /*
403 * Allot mbufs to a sockbuf.
404 * Attempt to scale mbmax so that mbcnt doesn't become limiting
405 * if buffering efficiency is near the normal case.
406 */
407 int
408 sbreserve(sb, cc, so, td)
409 struct sockbuf *sb;
410 u_long cc;
411 struct socket *so;
412 struct thread *td;
413 {
414
415 /*
416 * td will only be NULL when we're in an interrupt
417 * (e.g. in tcp_input())
418 */
419 if (cc > sb_max_adj)
420 return (0);
421 if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc,
422 td ? td->td_proc->p_rlimit[RLIMIT_SBSIZE].rlim_cur : RLIM_INFINITY)) {
423 return (0);
424 }
425 sb->sb_mbmax = min(cc * sb_efficiency, sb_max);
426 if (sb->sb_lowat > sb->sb_hiwat)
427 sb->sb_lowat = sb->sb_hiwat;
428 return (1);
429 }
430
431 /*
432 * Free mbufs held by a socket, and reserved mbuf space.
433 */
434 void
435 sbrelease(sb, so)
436 struct sockbuf *sb;
437 struct socket *so;
438 {
439
440 sbflush(sb);
441 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0,
442 RLIM_INFINITY);
443 sb->sb_mbmax = 0;
444 }
445
446 /*
447 * Routines to add and remove
448 * data from an mbuf queue.
449 *
450 * The routines sbappend() or sbappendrecord() are normally called to
451 * append new mbufs to a socket buffer, after checking that adequate
452 * space is available, comparing the function sbspace() with the amount
453 * of data to be added. sbappendrecord() differs from sbappend() in
454 * that data supplied is treated as the beginning of a new record.
455 * To place a sender's address, optional access rights, and data in a
456 * socket receive buffer, sbappendaddr() should be used. To place
457 * access rights and data in a socket receive buffer, sbappendrights()
458 * should be used. In either case, the new data begins a new record.
459 * Note that unlike sbappend() and sbappendrecord(), these routines check
460 * for the caller that there will be enough space to store the data.
461 * Each fails if there is not enough space, or if it cannot find mbufs
462 * to store additional information in.
463 *
464 * Reliable protocols may use the socket send buffer to hold data
465 * awaiting acknowledgement. Data is normally copied from a socket
466 * send buffer in a protocol with m_copy for output to a peer,
467 * and then removing the data from the socket buffer with sbdrop()
468 * or sbdroprecord() when the data is acknowledged by the peer.
469 */
470
471 #ifdef SOCKBUF_DEBUG
472 void
473 sblastrecordchk(struct sockbuf *sb, const char *file, int line)
474 {
475 struct mbuf *m = sb->sb_mb;
476
477 while (m && m->m_nextpkt)
478 m = m->m_nextpkt;
479
480 if (m != sb->sb_lastrecord) {
481 printf("%s: sb_mb %p sb_lastrecord %p last %p\n",
482 __func__, sb->sb_mb, sb->sb_lastrecord, m);
483 printf("packet chain:\n");
484 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt)
485 printf("\t%p\n", m);
486 panic("%s from %s:%u", __func__, file, line);
487 }
488 }
489
490 void
491 sblastmbufchk(struct sockbuf *sb, const char *file, int line)
492 {
493 struct mbuf *m = sb->sb_mb;
494 struct mbuf *n;
495
496 while (m && m->m_nextpkt)
497 m = m->m_nextpkt;
498
499 while (m && m->m_next)
500 m = m->m_next;
501
502 if (m != sb->sb_mbtail) {
503 printf("%s: sb_mb %p sb_mbtail %p last %p\n",
504 __func__, sb->sb_mb, sb->sb_mbtail, m);
505 printf("packet tree:\n");
506 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) {
507 printf("\t");
508 for (n = m; n != NULL; n = n->m_next)
509 printf("%p ", n);
510 printf("\n");
511 }
512 panic("%s from %s:%u", __func__, file, line);
513 }
514 }
515 #endif /* SOCKBUF_DEBUG */
516
517 #define SBLINKRECORD(sb, m0) do { \
518 if ((sb)->sb_lastrecord != NULL) \
519 (sb)->sb_lastrecord->m_nextpkt = (m0); \
520 else \
521 (sb)->sb_mb = (m0); \
522 (sb)->sb_lastrecord = (m0); \
523 } while (/*CONSTCOND*/0)
524
525 /*
526 * Append mbuf chain m to the last record in the
527 * socket buffer sb. The additional space associated
528 * the mbuf chain is recorded in sb. Empty mbufs are
529 * discarded and mbufs are compacted where possible.
530 */
531 void
532 sbappend(sb, m)
533 struct sockbuf *sb;
534 struct mbuf *m;
535 {
536 register struct mbuf *n;
537
538 if (m == 0)
539 return;
540 SBLASTRECORDCHK(sb);
541 n = sb->sb_mb;
542 if (n) {
543 while (n->m_nextpkt)
544 n = n->m_nextpkt;
545 do {
546 if (n->m_flags & M_EOR) {
547 sbappendrecord(sb, m); /* XXXXXX!!!! */
548 return;
549 }
550 } while (n->m_next && (n = n->m_next));
551 } else {
552 /*
553 * XXX Would like to simply use sb_mbtail here, but
554 * XXX I need to verify that I won't miss an EOR that
555 * XXX way.
556 */
557 if ((n = sb->sb_lastrecord) != NULL) {
558 do {
559 if (n->m_flags & M_EOR) {
560 sbappendrecord(sb, m); /* XXXXXX!!!! */
561 return;
562 }
563 } while (n->m_next && (n = n->m_next));
564 } else {
565 /*
566 * If this is the first record in the socket buffer,
567 * it's also the last record.
568 */
569 sb->sb_lastrecord = m;
570 }
571 }
572 sbcompress(sb, m, n);
573 SBLASTRECORDCHK(sb);
574 }
575
576 /*
577 * This version of sbappend() should only be used when the caller
578 * absolutely knows that there will never be more than one record
579 * in the socket buffer, that is, a stream protocol (such as TCP).
580 */
581 void
582 sbappendstream(struct sockbuf *sb, struct mbuf *m)
583 {
584
585 KASSERT(m->m_nextpkt == NULL,("sbappendstream 0"));
586 KASSERT(sb->sb_mb == sb->sb_lastrecord,("sbappendstream 1"));
587
588 SBLASTMBUFCHK(sb);
589
590 #ifdef MBUFTRACE
591 m_claim(m, sb->sb_mowner);
592 #endif
593
594 sbcompress(sb, m, sb->sb_mbtail);
595
596 sb->sb_lastrecord = sb->sb_mb;
597 SBLASTRECORDCHK(sb);
598 }
599
600 #ifdef SOCKBUF_DEBUG
601 void
602 sbcheck(sb)
603 struct sockbuf *sb;
604 {
605 struct mbuf *m;
606 struct mbuf *n = 0;
607 u_long len = 0, mbcnt = 0;
608
609 for (m = sb->sb_mb; m; m = n) {
610 n = m->m_nextpkt;
611 for (; m; m = m->m_next) {
612 len += m->m_len;
613 mbcnt += MSIZE;
614 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */
615 mbcnt += m->m_ext.ext_size;
616 }
617 }
618 if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) {
619 printf("cc %ld != %u || mbcnt %ld != %u\n", len, sb->sb_cc,
620 mbcnt, sb->sb_mbcnt);
621 panic("sbcheck");
622 }
623 }
624 #endif
625
626 /*
627 * As above, except the mbuf chain
628 * begins a new record.
629 */
630 void
631 sbappendrecord(sb, m0)
632 register struct sockbuf *sb;
633 register struct mbuf *m0;
634 {
635 register struct mbuf *m;
636
637 if (m0 == 0)
638 return;
639 m = sb->sb_mb;
640 if (m)
641 while (m->m_nextpkt)
642 m = m->m_nextpkt;
643 /*
644 * Put the first mbuf on the queue.
645 * Note this permits zero length records.
646 */
647 sballoc(sb, m0);
648 SBLASTRECORDCHK(sb);
649 SBLINKRECORD(sb, m0);
650 if (m)
651 m->m_nextpkt = m0;
652 else
653 sb->sb_mb = m0;
654 m = m0->m_next;
655 m0->m_next = 0;
656 if (m && (m0->m_flags & M_EOR)) {
657 m0->m_flags &= ~M_EOR;
658 m->m_flags |= M_EOR;
659 }
660 sbcompress(sb, m, m0);
661 }
662
663 /*
664 * As above except that OOB data
665 * is inserted at the beginning of the sockbuf,
666 * but after any other OOB data.
667 */
668 void
669 sbinsertoob(sb, m0)
670 register struct sockbuf *sb;
671 register struct mbuf *m0;
672 {
673 register struct mbuf *m;
674 register struct mbuf **mp;
675
676 if (m0 == 0)
677 return;
678 for (mp = &sb->sb_mb; *mp ; mp = &((*mp)->m_nextpkt)) {
679 m = *mp;
680 again:
681 switch (m->m_type) {
682
683 case MT_OOBDATA:
684 continue; /* WANT next train */
685
686 case MT_CONTROL:
687 m = m->m_next;
688 if (m)
689 goto again; /* inspect THIS train further */
690 }
691 break;
692 }
693 /*
694 * Put the first mbuf on the queue.
695 * Note this permits zero length records.
696 */
697 sballoc(sb, m0);
698 m0->m_nextpkt = *mp;
699 *mp = m0;
700 m = m0->m_next;
701 m0->m_next = 0;
702 if (m && (m0->m_flags & M_EOR)) {
703 m0->m_flags &= ~M_EOR;
704 m->m_flags |= M_EOR;
705 }
706 sbcompress(sb, m, m0);
707 }
708
709 /*
710 * Append address and data, and optionally, control (ancillary) data
711 * to the receive queue of a socket. If present,
712 * m0 must include a packet header with total length.
713 * Returns 0 if no space in sockbuf or insufficient mbufs.
714 */
715 int
716 sbappendaddr(sb, asa, m0, control)
717 struct sockbuf *sb;
718 struct sockaddr *asa;
719 struct mbuf *m0, *control;
720 {
721 struct mbuf *m, *n, *nlast;
722 int space = asa->sa_len;
723
724 if (m0 && (m0->m_flags & M_PKTHDR) == 0)
725 panic("sbappendaddr");
726 if (m0)
727 space += m0->m_pkthdr.len;
728 space += m_length(control, &n);
729 if (space > sbspace(sb))
730 return (0);
731 #if MSIZE <= 256
732 if (asa->sa_len > MLEN)
733 return (0);
734 #endif
735 MGET(m, M_DONTWAIT, MT_SONAME);
736 if (m == 0)
737 return (0);
738 m->m_len = asa->sa_len;
739 bcopy(asa, mtod(m, caddr_t), asa->sa_len);
740 if (n)
741 n->m_next = m0; /* concatenate data to control */
742 else
743 control = m0;
744 m->m_next = control;
745 for (n = m; n->m_next != NULL; n = n->m_next)
746 sballoc(sb, n);
747 sballoc(sb, n);
748 nlast = n;
749 SBLINKRECORD(sb, m);
750
751 sb->sb_mbtail = nlast;
752 SBLASTMBUFCHK(sb);
753
754 SBLASTRECORDCHK(sb);
755 return (1);
756 }
757
758 int
759 sbappendcontrol(sb, m0, control)
760 struct sockbuf *sb;
761 struct mbuf *control, *m0;
762 {
763 struct mbuf *m, *n, *mlast;
764 int space;
765
766 if (control == 0)
767 panic("sbappendcontrol");
768 space = m_length(control, &n) + m_length(m0, NULL);
769 if (space > sbspace(sb))
770 return (0);
771 n->m_next = m0; /* concatenate data to control */
772
773 SBLASTRECORDCHK(sb);
774
775 for (m = control; m->m_next; m = m->m_next)
776 sballoc(sb, m);
777 sballoc(sb, m);
778 mlast = m;
779 SBLINKRECORD(sb, control);
780
781 sb->sb_mbtail = mlast;
782 SBLASTMBUFCHK(sb);
783
784 SBLASTRECORDCHK(sb);
785 return (1);
786 }
787
788 /*
789 * Compress mbuf chain m into the socket
790 * buffer sb following mbuf n. If n
791 * is null, the buffer is presumed empty.
792 */
793 void
794 sbcompress(sb, m, n)
795 register struct sockbuf *sb;
796 register struct mbuf *m, *n;
797 {
798 register int eor = 0;
799 register struct mbuf *o;
800
801 while (m) {
802 eor |= m->m_flags & M_EOR;
803 if (m->m_len == 0 &&
804 (eor == 0 ||
805 (((o = m->m_next) || (o = n)) &&
806 o->m_type == m->m_type))) {
807 if (sb->sb_lastrecord == m)
808 sb->sb_lastrecord = m->m_next;
809 m = m_free(m);
810 continue;
811 }
812 if (n && (n->m_flags & M_EOR) == 0 &&
813 M_WRITABLE(n) &&
814 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */
815 m->m_len <= M_TRAILINGSPACE(n) &&
816 n->m_type == m->m_type) {
817 bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len,
818 (unsigned)m->m_len);
819 n->m_len += m->m_len;
820 sb->sb_cc += m->m_len;
821 if (m->m_type != MT_DATA && m->m_type != MT_HEADER &&
822 m->m_type != MT_OOBDATA)
823 /* XXX: Probably don't need.*/
824 sb->sb_ctl += m->m_len;
825 m = m_free(m);
826 continue;
827 }
828 if (n)
829 n->m_next = m;
830 else
831 sb->sb_mb = m;
832 sb->sb_mbtail = m;
833 sballoc(sb, m);
834 n = m;
835 m->m_flags &= ~M_EOR;
836 m = m->m_next;
837 n->m_next = 0;
838 }
839 if (eor) {
840 if (n)
841 n->m_flags |= eor;
842 else
843 printf("semi-panic: sbcompress\n");
844 }
845 SBLASTMBUFCHK(sb);
846 }
847
848 /*
849 * Free all mbufs in a sockbuf.
850 * Check that all resources are reclaimed.
851 */
852 void
853 sbflush(sb)
854 register struct sockbuf *sb;
855 {
856
857 if (sb->sb_flags & SB_LOCK)
858 panic("sbflush: locked");
859 while (sb->sb_mbcnt) {
860 /*
861 * Don't call sbdrop(sb, 0) if the leading mbuf is non-empty:
862 * we would loop forever. Panic instead.
863 */
864 if (!sb->sb_cc && (sb->sb_mb == NULL || sb->sb_mb->m_len))
865 break;
866 sbdrop(sb, (int)sb->sb_cc);
867 }
868 if (sb->sb_cc || sb->sb_mb || sb->sb_mbcnt)
869 panic("sbflush: cc %u || mb %p || mbcnt %u", sb->sb_cc, (void *)sb->sb_mb, sb->sb_mbcnt);
870 }
871
872 /*
873 * Drop data from (the front of) a sockbuf.
874 */
875 void
876 sbdrop(sb, len)
877 register struct sockbuf *sb;
878 register int len;
879 {
880 register struct mbuf *m;
881 struct mbuf *next;
882
883 next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
884 while (len > 0) {
885 if (m == 0) {
886 if (next == 0)
887 panic("sbdrop");
888 m = next;
889 next = m->m_nextpkt;
890 continue;
891 }
892 if (m->m_len > len) {
893 m->m_len -= len;
894 m->m_data += len;
895 sb->sb_cc -= len;
896 if (m->m_type != MT_DATA && m->m_type != MT_HEADER &&
897 m->m_type != MT_OOBDATA)
898 sb->sb_ctl -= len;
899 break;
900 }
901 len -= m->m_len;
902 sbfree(sb, m);
903 m = m_free(m);
904 }
905 while (m && m->m_len == 0) {
906 sbfree(sb, m);
907 m = m_free(m);
908 }
909 if (m) {
910 sb->sb_mb = m;
911 m->m_nextpkt = next;
912 } else
913 sb->sb_mb = next;
914 /*
915 * First part is an inline SB_EMPTY_FIXUP(). Second part
916 * makes sure sb_lastrecord is up-to-date if we dropped
917 * part of the last record.
918 */
919 m = sb->sb_mb;
920 if (m == NULL) {
921 sb->sb_mbtail = NULL;
922 sb->sb_lastrecord = NULL;
923 } else if (m->m_nextpkt == NULL) {
924 sb->sb_lastrecord = m;
925 }
926 }
927
928 /*
929 * Drop a record off the front of a sockbuf
930 * and move the next record to the front.
931 */
932 void
933 sbdroprecord(sb)
934 register struct sockbuf *sb;
935 {
936 register struct mbuf *m;
937
938 m = sb->sb_mb;
939 if (m) {
940 sb->sb_mb = m->m_nextpkt;
941 do {
942 sbfree(sb, m);
943 m = m_free(m);
944 } while (m);
945 }
946 SB_EMPTY_FIXUP(sb);
947 }
948
949 /*
950 * Create a "control" mbuf containing the specified data
951 * with the specified type for presentation on a socket buffer.
952 */
953 struct mbuf *
954 sbcreatecontrol(p, size, type, level)
955 caddr_t p;
956 register int size;
957 int type, level;
958 {
959 register struct cmsghdr *cp;
960 struct mbuf *m;
961
962 if (CMSG_SPACE((u_int)size) > MCLBYTES)
963 return ((struct mbuf *) NULL);
964 if ((m = m_get(M_DONTWAIT, MT_CONTROL)) == NULL)
965 return ((struct mbuf *) NULL);
966 if (CMSG_SPACE((u_int)size) > MLEN) {
967 MCLGET(m, M_DONTWAIT);
968 if ((m->m_flags & M_EXT) == 0) {
969 m_free(m);
970 return ((struct mbuf *) NULL);
971 }
972 }
973 cp = mtod(m, struct cmsghdr *);
974 m->m_len = 0;
975 KASSERT(CMSG_SPACE((u_int)size) <= M_TRAILINGSPACE(m),
976 ("sbcreatecontrol: short mbuf"));
977 if (p != NULL)
978 (void)memcpy(CMSG_DATA(cp), p, size);
979 m->m_len = CMSG_SPACE(size);
980 cp->cmsg_len = CMSG_LEN(size);
981 cp->cmsg_level = level;
982 cp->cmsg_type = type;
983 return (m);
984 }
985
986 /*
987 * Some routines that return EOPNOTSUPP for entry points that are not
988 * supported by a protocol. Fill in as needed.
989 */
990 int
991 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
992 {
993 return EOPNOTSUPP;
994 }
995
996 int
997 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
998 {
999 return EOPNOTSUPP;
1000 }
1001
1002 int
1003 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
1004 {
1005 return EOPNOTSUPP;
1006 }
1007
1008 int
1009 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
1010 struct ifnet *ifp, struct thread *td)
1011 {
1012 return EOPNOTSUPP;
1013 }
1014
1015 int
1016 pru_listen_notsupp(struct socket *so, struct thread *td)
1017 {
1018 return EOPNOTSUPP;
1019 }
1020
1021 int
1022 pru_rcvd_notsupp(struct socket *so, int flags)
1023 {
1024 return EOPNOTSUPP;
1025 }
1026
1027 int
1028 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
1029 {
1030 return EOPNOTSUPP;
1031 }
1032
1033 /*
1034 * This isn't really a ``null'' operation, but it's the default one
1035 * and doesn't do anything destructive.
1036 */
1037 int
1038 pru_sense_null(struct socket *so, struct stat *sb)
1039 {
1040 sb->st_blksize = so->so_snd.sb_hiwat;
1041 return 0;
1042 }
1043
1044 /*
1045 * For protocol types that don't keep cached copies of labels in their
1046 * pcbs, provide a null sosetlabel that does a NOOP.
1047 */
1048 void
1049 pru_sosetlabel_null(struct socket *so)
1050 {
1051
1052 }
1053
1054 /*
1055 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
1056 */
1057 struct sockaddr *
1058 dup_sockaddr(sa, canwait)
1059 struct sockaddr *sa;
1060 int canwait;
1061 {
1062 struct sockaddr *sa2;
1063
1064 MALLOC(sa2, struct sockaddr *, sa->sa_len, M_SONAME,
1065 canwait ? M_WAITOK : M_NOWAIT);
1066 if (sa2)
1067 bcopy(sa, sa2, sa->sa_len);
1068 return sa2;
1069 }
1070
1071 /*
1072 * Create an external-format (``xsocket'') structure using the information
1073 * in the kernel-format socket structure pointed to by so. This is done
1074 * to reduce the spew of irrelevant information over this interface,
1075 * to isolate user code from changes in the kernel structure, and
1076 * potentially to provide information-hiding if we decide that
1077 * some of this information should be hidden from users.
1078 */
1079 void
1080 sotoxsocket(struct socket *so, struct xsocket *xso)
1081 {
1082 xso->xso_len = sizeof *xso;
1083 xso->xso_so = so;
1084 xso->so_type = so->so_type;
1085 xso->so_options = so->so_options;
1086 xso->so_linger = so->so_linger;
1087 xso->so_state = so->so_state;
1088 xso->so_pcb = so->so_pcb;
1089 xso->xso_protocol = so->so_proto->pr_protocol;
1090 xso->xso_family = so->so_proto->pr_domain->dom_family;
1091 xso->so_qlen = so->so_qlen;
1092 xso->so_incqlen = so->so_incqlen;
1093 xso->so_qlimit = so->so_qlimit;
1094 xso->so_timeo = so->so_timeo;
1095 xso->so_error = so->so_error;
1096 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
1097 xso->so_oobmark = so->so_oobmark;
1098 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
1099 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
1100 xso->so_uid = so->so_cred->cr_uid;
1101 }
1102
1103 /*
1104 * This does the same for sockbufs. Note that the xsockbuf structure,
1105 * since it is always embedded in a socket, does not include a self
1106 * pointer nor a length. We make this entry point public in case
1107 * some other mechanism needs it.
1108 */
1109 void
1110 sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb)
1111 {
1112 xsb->sb_cc = sb->sb_cc;
1113 xsb->sb_hiwat = sb->sb_hiwat;
1114 xsb->sb_mbcnt = sb->sb_mbcnt;
1115 xsb->sb_mbmax = sb->sb_mbmax;
1116 xsb->sb_lowat = sb->sb_lowat;
1117 xsb->sb_flags = sb->sb_flags;
1118 xsb->sb_timeo = sb->sb_timeo;
1119 }
1120
1121 /*
1122 * Here is the definition of some of the basic objects in the kern.ipc
1123 * branch of the MIB.
1124 */
1125 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
1126
1127 /* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */
1128 static int dummy;
1129 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, "");
1130 SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_ULONG|CTLFLAG_RW,
1131 &sb_max, 0, sysctl_handle_sb_max, "LU", "Maximum socket buffer size");
1132 SYSCTL_INT(_kern_ipc, OID_AUTO, maxsockets, CTLFLAG_RDTUN,
1133 &maxsockets, 0, "Maximum number of sockets avaliable");
1134 SYSCTL_ULONG(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW,
1135 &sb_efficiency, 0, "");
1136
1137 /*
1138 * Initialise maxsockets
1139 */
1140 static void init_maxsockets(void *ignored)
1141 {
1142 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
1143 maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
1144 }
1145 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
Cache object: 48d7b9ca580f1796c9af986b4d6c8f24
|