uipc_socket2.c revision 1.89 1 /* $NetBSD: uipc_socket2.c,v 1.89 2008/02/07 12:14:43 ad Exp $ */
2
3 /*
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993
5 * The Regents of the University of California. All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 *
31 * @(#)uipc_socket2.c 8.2 (Berkeley) 2/14/95
32 */
33
34 #include <sys/cdefs.h>
35 __KERNEL_RCSID(0, "$NetBSD: uipc_socket2.c,v 1.89 2008/02/07 12:14:43 ad Exp $");
36
37 #include "opt_mbuftrace.h"
38 #include "opt_sb_max.h"
39
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/proc.h>
43 #include <sys/file.h>
44 #include <sys/buf.h>
45 #include <sys/malloc.h>
46 #include <sys/mbuf.h>
47 #include <sys/protosw.h>
48 #include <sys/poll.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/signalvar.h>
52 #include <sys/kauth.h>
53
54 /*
55 * Primitive routines for operating on sockets and socket buffers
56 */
57
58 /* strings for sleep message: */
59 const char netcon[] = "netcon";
60 const char netcls[] = "netcls";
61 const char netio[] = "netio";
62 const char netlck[] = "netlck";
63
64 u_long sb_max = SB_MAX; /* maximum socket buffer size */
65 static u_long sb_max_adj; /* adjusted sb_max */
66
67 /*
68 * Procedures to manipulate state flags of socket
69 * and do appropriate wakeups. Normal sequence from the
70 * active (originating) side is that soisconnecting() is
71 * called during processing of connect() call,
72 * resulting in an eventual call to soisconnected() if/when the
73 * connection is established. When the connection is torn down
74 * soisdisconnecting() is called during processing of disconnect() call,
75 * and soisdisconnected() is called when the connection to the peer
76 * is totally severed. The semantics of these routines are such that
77 * connectionless protocols can call soisconnected() and soisdisconnected()
78 * only, bypassing the in-progress calls when setting up a ``connection''
79 * takes no time.
80 *
81 * From the passive side, a socket is created with
82 * two queues of sockets: so_q0 for connections in progress
83 * and so_q for connections already made and awaiting user acceptance.
84 * As a protocol is preparing incoming connections, it creates a socket
85 * structure queued on so_q0 by calling sonewconn(). When the connection
86 * is established, soisconnected() is called, and transfers the
87 * socket structure to so_q, making it available to accept().
88 *
89 * If a socket is closed with sockets on either
90 * so_q0 or so_q, these sockets are dropped.
91 *
92 * If higher level protocols are implemented in
93 * the kernel, the wakeups done here will sometimes
94 * cause software-interrupt process scheduling.
95 */
96
97 void
98 soisconnecting(struct socket *so)
99 {
100
101 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
102 so->so_state |= SS_ISCONNECTING;
103 }
104
105 void
106 soisconnected(struct socket *so)
107 {
108 struct socket *head;
109
110 head = so->so_head;
111 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
112 so->so_state |= SS_ISCONNECTED;
113 if (head && soqremque(so, 0)) {
114 soqinsque(head, so, 1);
115 sorwakeup(head);
116 wakeup((void *)&head->so_timeo);
117 } else {
118 wakeup((void *)&so->so_timeo);
119 sorwakeup(so);
120 sowwakeup(so);
121 }
122 }
123
124 void
125 soisdisconnecting(struct socket *so)
126 {
127
128 so->so_state &= ~SS_ISCONNECTING;
129 so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE);
130 wakeup((void *)&so->so_timeo);
131 sowwakeup(so);
132 sorwakeup(so);
133 }
134
135 void
136 soisdisconnected(struct socket *so)
137 {
138
139 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
140 so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED);
141 wakeup((void *)&so->so_timeo);
142 sowwakeup(so);
143 sorwakeup(so);
144 }
145
146 /*
147 * When an attempt at a new connection is noted on a socket
148 * which accepts connections, sonewconn is called. If the
149 * connection is possible (subject to space constraints, etc.)
150 * then we allocate a new structure, propoerly linked into the
151 * data structure of the original socket, and return this.
152 * Connstatus may be 0, SS_ISCONFIRMING, or SS_ISCONNECTED.
153 */
154 struct socket *
155 sonewconn(struct socket *head, int connstatus)
156 {
157 struct socket *so;
158 int soqueue;
159
160 soqueue = connstatus ? 1 : 0;
161 if (head->so_qlen + head->so_q0len > 3 * head->so_qlimit / 2)
162 return ((struct socket *)0);
163 so = pool_get(&socket_pool, PR_NOWAIT);
164 if (so == NULL)
165 return (NULL);
166 memset((void *)so, 0, sizeof(*so));
167 so->so_type = head->so_type;
168 so->so_options = head->so_options &~ SO_ACCEPTCONN;
169 so->so_linger = head->so_linger;
170 so->so_state = head->so_state | SS_NOFDREF;
171 so->so_nbio = head->so_nbio;
172 so->so_proto = head->so_proto;
173 so->so_timeo = head->so_timeo;
174 so->so_pgid = head->so_pgid;
175 so->so_send = head->so_send;
176 so->so_receive = head->so_receive;
177 so->so_uidinfo = head->so_uidinfo;
178 #ifdef MBUFTRACE
179 so->so_mowner = head->so_mowner;
180 so->so_rcv.sb_mowner = head->so_rcv.sb_mowner;
181 so->so_snd.sb_mowner = head->so_snd.sb_mowner;
182 #endif
183 selinit(&so->so_rcv.sb_sel);
184 selinit(&so->so_snd.sb_sel);
185 (void) soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat);
186 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
187 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
188 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
189 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
190 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
191 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
192 soqinsque(head, so, soqueue);
193 if ((*so->so_proto->pr_usrreq)(so, PRU_ATTACH,
194 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0,
195 (struct lwp *)0)) {
196 (void) soqremque(so, soqueue);
197 seldestroy(&so->so_rcv.sb_sel);
198 seldestroy(&so->so_snd.sb_sel);
199 pool_put(&socket_pool, so);
200 return (NULL);
201 }
202 if (connstatus) {
203 sorwakeup(head);
204 wakeup((void *)&head->so_timeo);
205 so->so_state |= connstatus;
206 }
207 return (so);
208 }
209
210 void
211 soqinsque(struct socket *head, struct socket *so, int q)
212 {
213
214 #ifdef DIAGNOSTIC
215 if (so->so_onq != NULL)
216 panic("soqinsque");
217 #endif
218
219 so->so_head = head;
220 if (q == 0) {
221 head->so_q0len++;
222 so->so_onq = &head->so_q0;
223 } else {
224 head->so_qlen++;
225 so->so_onq = &head->so_q;
226 }
227 TAILQ_INSERT_TAIL(so->so_onq, so, so_qe);
228 }
229
230 int
231 soqremque(struct socket *so, int q)
232 {
233 struct socket *head;
234
235 head = so->so_head;
236 if (q == 0) {
237 if (so->so_onq != &head->so_q0)
238 return (0);
239 head->so_q0len--;
240 } else {
241 if (so->so_onq != &head->so_q)
242 return (0);
243 head->so_qlen--;
244 }
245 TAILQ_REMOVE(so->so_onq, so, so_qe);
246 so->so_onq = NULL;
247 so->so_head = NULL;
248 return (1);
249 }
250
251 /*
252 * Socantsendmore indicates that no more data will be sent on the
253 * socket; it would normally be applied to a socket when the user
254 * informs the system that no more data is to be sent, by the protocol
255 * code (in case PRU_SHUTDOWN). Socantrcvmore indicates that no more data
256 * will be received, and will normally be applied to the socket by a
257 * protocol when it detects that the peer will send no more data.
258 * Data queued for reading in the socket may yet be read.
259 */
260
261 void
262 socantsendmore(struct socket *so)
263 {
264
265 so->so_state |= SS_CANTSENDMORE;
266 sowwakeup(so);
267 }
268
269 void
270 socantrcvmore(struct socket *so)
271 {
272
273 so->so_state |= SS_CANTRCVMORE;
274 sorwakeup(so);
275 }
276
277 /*
278 * Wait for data to arrive at/drain from a socket buffer.
279 */
280 int
281 sbwait(struct sockbuf *sb)
282 {
283
284 sb->sb_flags |= SB_WAIT;
285 return (tsleep((void *)&sb->sb_cc,
286 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, netio,
287 sb->sb_timeo));
288 }
289
290 /*
291 * Lock a sockbuf already known to be locked;
292 * return any error returned from sleep (EINTR).
293 */
294 int
295 sb_lock(struct sockbuf *sb)
296 {
297 int error;
298
299 while (sb->sb_flags & SB_LOCK) {
300 sb->sb_flags |= SB_WANT;
301 error = tsleep((void *)&sb->sb_flags,
302 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK|PCATCH,
303 netlck, 0);
304 if (error)
305 return (error);
306 }
307 sb->sb_flags |= SB_LOCK;
308 return (0);
309 }
310
311 /*
312 * Wakeup processes waiting on a socket buffer.
313 * Do asynchronous notification via SIGIO
314 * if the socket buffer has the SB_ASYNC flag set.
315 */
316 void
317 sowakeup(struct socket *so, struct sockbuf *sb, int code)
318 {
319 selnotify(&sb->sb_sel, 0);
320 sb->sb_flags &= ~SB_SEL;
321 if (sb->sb_flags & SB_WAIT) {
322 sb->sb_flags &= ~SB_WAIT;
323 wakeup((void *)&sb->sb_cc);
324 }
325 if (sb->sb_flags & SB_ASYNC) {
326 int band;
327 if (code == POLL_IN)
328 band = POLLIN|POLLRDNORM;
329 else
330 band = POLLOUT|POLLWRNORM;
331 fownsignal(so->so_pgid, SIGIO, code, band, so);
332 }
333 if (sb->sb_flags & SB_UPCALL)
334 (*so->so_upcall)(so, so->so_upcallarg, M_DONTWAIT);
335 }
336
337 /*
338 * Socket buffer (struct sockbuf) utility routines.
339 *
340 * Each socket contains two socket buffers: one for sending data and
341 * one for receiving data. Each buffer contains a queue of mbufs,
342 * information about the number of mbufs and amount of data in the
343 * queue, and other fields allowing poll() statements and notification
344 * on data availability to be implemented.
345 *
346 * Data stored in a socket buffer is maintained as a list of records.
347 * Each record is a list of mbufs chained together with the m_next
348 * field. Records are chained together with the m_nextpkt field. The upper
349 * level routine soreceive() expects the following conventions to be
350 * observed when placing information in the receive buffer:
351 *
352 * 1. If the protocol requires each message be preceded by the sender's
353 * name, then a record containing that name must be present before
354 * any associated data (mbuf's must be of type MT_SONAME).
355 * 2. If the protocol supports the exchange of ``access rights'' (really
356 * just additional data associated with the message), and there are
357 * ``rights'' to be received, then a record containing this data
358 * should be present (mbuf's must be of type MT_CONTROL).
359 * 3. If a name or rights record exists, then it must be followed by
360 * a data record, perhaps of zero length.
361 *
362 * Before using a new socket structure it is first necessary to reserve
363 * buffer space to the socket, by calling sbreserve(). This should commit
364 * some of the available buffer space in the system buffer pool for the
365 * socket (currently, it does nothing but enforce limits). The space
366 * should be released by calling sbrelease() when the socket is destroyed.
367 */
368
369 int
370 sb_max_set(u_long new_sbmax)
371 {
372 int s;
373
374 if (new_sbmax < (16 * 1024))
375 return (EINVAL);
376
377 s = splsoftnet();
378 sb_max = new_sbmax;
379 sb_max_adj = (u_quad_t)new_sbmax * MCLBYTES / (MSIZE + MCLBYTES);
380 splx(s);
381
382 return (0);
383 }
384
385 int
386 soreserve(struct socket *so, u_long sndcc, u_long rcvcc)
387 {
388 /*
389 * there's at least one application (a configure script of screen)
390 * which expects a fifo is writable even if it has "some" bytes
391 * in its buffer.
392 * so we want to make sure (hiwat - lowat) >= (some bytes).
393 *
394 * PIPE_BUF here is an arbitrary value chosen as (some bytes) above.
395 * we expect it's large enough for such applications.
396 */
397 u_long lowat = MAX(sock_loan_thresh, MCLBYTES);
398 u_long hiwat = lowat + PIPE_BUF;
399
400 if (sndcc < hiwat)
401 sndcc = hiwat;
402 if (sbreserve(&so->so_snd, sndcc, so) == 0)
403 goto bad;
404 if (sbreserve(&so->so_rcv, rcvcc, so) == 0)
405 goto bad2;
406 if (so->so_rcv.sb_lowat == 0)
407 so->so_rcv.sb_lowat = 1;
408 if (so->so_snd.sb_lowat == 0)
409 so->so_snd.sb_lowat = lowat;
410 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
411 so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
412 return (0);
413 bad2:
414 sbrelease(&so->so_snd, so);
415 bad:
416 return (ENOBUFS);
417 }
418
419 /*
420 * Allot mbufs to a sockbuf.
421 * Attempt to scale mbmax so that mbcnt doesn't become limiting
422 * if buffering efficiency is near the normal case.
423 */
424 int
425 sbreserve(struct sockbuf *sb, u_long cc, struct socket *so)
426 {
427 struct lwp *l = curlwp; /* XXX */
428 rlim_t maxcc;
429 struct uidinfo *uidinfo;
430
431 KDASSERT(sb_max_adj != 0);
432 if (cc == 0 || cc > sb_max_adj)
433 return (0);
434 if (so) {
435 if (kauth_cred_geteuid(l->l_cred) == so->so_uidinfo->ui_uid)
436 maxcc = l->l_proc->p_rlimit[RLIMIT_SBSIZE].rlim_cur;
437 else
438 maxcc = RLIM_INFINITY;
439 uidinfo = so->so_uidinfo;
440 } else {
441 uidinfo = uid_find(0); /* XXX: nothing better */
442 maxcc = RLIM_INFINITY;
443 }
444 if (!chgsbsize(uidinfo, &sb->sb_hiwat, cc, maxcc))
445 return 0;
446 sb->sb_mbmax = min(cc * 2, sb_max);
447 if (sb->sb_lowat > sb->sb_hiwat)
448 sb->sb_lowat = sb->sb_hiwat;
449 return (1);
450 }
451
452 /*
453 * Free mbufs held by a socket, and reserved mbuf space.
454 */
455 void
456 sbrelease(struct sockbuf *sb, struct socket *so)
457 {
458
459 sbflush(sb);
460 (void)chgsbsize(so->so_uidinfo, &sb->sb_hiwat, 0, RLIM_INFINITY);
461 sb->sb_mbmax = 0;
462 }
463
464 /*
465 * Routines to add and remove
466 * data from an mbuf queue.
467 *
468 * The routines sbappend() or sbappendrecord() are normally called to
469 * append new mbufs to a socket buffer, after checking that adequate
470 * space is available, comparing the function sbspace() with the amount
471 * of data to be added. sbappendrecord() differs from sbappend() in
472 * that data supplied is treated as the beginning of a new record.
473 * To place a sender's address, optional access rights, and data in a
474 * socket receive buffer, sbappendaddr() should be used. To place
475 * access rights and data in a socket receive buffer, sbappendrights()
476 * should be used. In either case, the new data begins a new record.
477 * Note that unlike sbappend() and sbappendrecord(), these routines check
478 * for the caller that there will be enough space to store the data.
479 * Each fails if there is not enough space, or if it cannot find mbufs
480 * to store additional information in.
481 *
482 * Reliable protocols may use the socket send buffer to hold data
483 * awaiting acknowledgement. Data is normally copied from a socket
484 * send buffer in a protocol with m_copy for output to a peer,
485 * and then removing the data from the socket buffer with sbdrop()
486 * or sbdroprecord() when the data is acknowledged by the peer.
487 */
488
489 #ifdef SOCKBUF_DEBUG
490 void
491 sblastrecordchk(struct sockbuf *sb, const char *where)
492 {
493 struct mbuf *m = sb->sb_mb;
494
495 while (m && m->m_nextpkt)
496 m = m->m_nextpkt;
497
498 if (m != sb->sb_lastrecord) {
499 printf("sblastrecordchk: sb_mb %p sb_lastrecord %p last %p\n",
500 sb->sb_mb, sb->sb_lastrecord, m);
501 printf("packet chain:\n");
502 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt)
503 printf("\t%p\n", m);
504 panic("sblastrecordchk from %s", where);
505 }
506 }
507
508 void
509 sblastmbufchk(struct sockbuf *sb, const char *where)
510 {
511 struct mbuf *m = sb->sb_mb;
512 struct mbuf *n;
513
514 while (m && m->m_nextpkt)
515 m = m->m_nextpkt;
516
517 while (m && m->m_next)
518 m = m->m_next;
519
520 if (m != sb->sb_mbtail) {
521 printf("sblastmbufchk: sb_mb %p sb_mbtail %p last %p\n",
522 sb->sb_mb, sb->sb_mbtail, m);
523 printf("packet tree:\n");
524 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) {
525 printf("\t");
526 for (n = m; n != NULL; n = n->m_next)
527 printf("%p ", n);
528 printf("\n");
529 }
530 panic("sblastmbufchk from %s", where);
531 }
532 }
533 #endif /* SOCKBUF_DEBUG */
534
535 /*
536 * Link a chain of records onto a socket buffer
537 */
538 #define SBLINKRECORDCHAIN(sb, m0, mlast) \
539 do { \
540 if ((sb)->sb_lastrecord != NULL) \
541 (sb)->sb_lastrecord->m_nextpkt = (m0); \
542 else \
543 (sb)->sb_mb = (m0); \
544 (sb)->sb_lastrecord = (mlast); \
545 } while (/*CONSTCOND*/0)
546
547
548 #define SBLINKRECORD(sb, m0) \
549 SBLINKRECORDCHAIN(sb, m0, m0)
550
551 /*
552 * Append mbuf chain m to the last record in the
553 * socket buffer sb. The additional space associated
554 * the mbuf chain is recorded in sb. Empty mbufs are
555 * discarded and mbufs are compacted where possible.
556 */
557 void
558 sbappend(struct sockbuf *sb, struct mbuf *m)
559 {
560 struct mbuf *n;
561
562 if (m == 0)
563 return;
564
565 #ifdef MBUFTRACE
566 m_claimm(m, sb->sb_mowner);
567 #endif
568
569 SBLASTRECORDCHK(sb, "sbappend 1");
570
571 if ((n = sb->sb_lastrecord) != NULL) {
572 /*
573 * XXX Would like to simply use sb_mbtail here, but
574 * XXX I need to verify that I won't miss an EOR that
575 * XXX way.
576 */
577 do {
578 if (n->m_flags & M_EOR) {
579 sbappendrecord(sb, m); /* XXXXXX!!!! */
580 return;
581 }
582 } while (n->m_next && (n = n->m_next));
583 } else {
584 /*
585 * If this is the first record in the socket buffer, it's
586 * also the last record.
587 */
588 sb->sb_lastrecord = m;
589 }
590 sbcompress(sb, m, n);
591 SBLASTRECORDCHK(sb, "sbappend 2");
592 }
593
594 /*
595 * This version of sbappend() should only be used when the caller
596 * absolutely knows that there will never be more than one record
597 * in the socket buffer, that is, a stream protocol (such as TCP).
598 */
599 void
600 sbappendstream(struct sockbuf *sb, struct mbuf *m)
601 {
602
603 KDASSERT(m->m_nextpkt == NULL);
604 KASSERT(sb->sb_mb == sb->sb_lastrecord);
605
606 SBLASTMBUFCHK(sb, __func__);
607
608 #ifdef MBUFTRACE
609 m_claimm(m, sb->sb_mowner);
610 #endif
611
612 sbcompress(sb, m, sb->sb_mbtail);
613
614 sb->sb_lastrecord = sb->sb_mb;
615 SBLASTRECORDCHK(sb, __func__);
616 }
617
618 #ifdef SOCKBUF_DEBUG
619 void
620 sbcheck(struct sockbuf *sb)
621 {
622 struct mbuf *m;
623 u_long len, mbcnt;
624
625 len = 0;
626 mbcnt = 0;
627 for (m = sb->sb_mb; m; m = m->m_next) {
628 len += m->m_len;
629 mbcnt += MSIZE;
630 if (m->m_flags & M_EXT)
631 mbcnt += m->m_ext.ext_size;
632 if (m->m_nextpkt)
633 panic("sbcheck nextpkt");
634 }
635 if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) {
636 printf("cc %lu != %lu || mbcnt %lu != %lu\n", len, sb->sb_cc,
637 mbcnt, sb->sb_mbcnt);
638 panic("sbcheck");
639 }
640 }
641 #endif
642
643 /*
644 * As above, except the mbuf chain
645 * begins a new record.
646 */
647 void
648 sbappendrecord(struct sockbuf *sb, struct mbuf *m0)
649 {
650 struct mbuf *m;
651
652 if (m0 == 0)
653 return;
654
655 #ifdef MBUFTRACE
656 m_claimm(m0, sb->sb_mowner);
657 #endif
658 /*
659 * Put the first mbuf on the queue.
660 * Note this permits zero length records.
661 */
662 sballoc(sb, m0);
663 SBLASTRECORDCHK(sb, "sbappendrecord 1");
664 SBLINKRECORD(sb, m0);
665 m = m0->m_next;
666 m0->m_next = 0;
667 if (m && (m0->m_flags & M_EOR)) {
668 m0->m_flags &= ~M_EOR;
669 m->m_flags |= M_EOR;
670 }
671 sbcompress(sb, m, m0);
672 SBLASTRECORDCHK(sb, "sbappendrecord 2");
673 }
674
675 /*
676 * As above except that OOB data
677 * is inserted at the beginning of the sockbuf,
678 * but after any other OOB data.
679 */
680 void
681 sbinsertoob(struct sockbuf *sb, struct mbuf *m0)
682 {
683 struct mbuf *m, **mp;
684
685 if (m0 == 0)
686 return;
687
688 SBLASTRECORDCHK(sb, "sbinsertoob 1");
689
690 for (mp = &sb->sb_mb; (m = *mp) != NULL; mp = &((*mp)->m_nextpkt)) {
691 again:
692 switch (m->m_type) {
693
694 case MT_OOBDATA:
695 continue; /* WANT next train */
696
697 case MT_CONTROL:
698 if ((m = m->m_next) != NULL)
699 goto again; /* inspect THIS train further */
700 }
701 break;
702 }
703 /*
704 * Put the first mbuf on the queue.
705 * Note this permits zero length records.
706 */
707 sballoc(sb, m0);
708 m0->m_nextpkt = *mp;
709 if (*mp == NULL) {
710 /* m0 is actually the new tail */
711 sb->sb_lastrecord = m0;
712 }
713 *mp = m0;
714 m = m0->m_next;
715 m0->m_next = 0;
716 if (m && (m0->m_flags & M_EOR)) {
717 m0->m_flags &= ~M_EOR;
718 m->m_flags |= M_EOR;
719 }
720 sbcompress(sb, m, m0);
721 SBLASTRECORDCHK(sb, "sbinsertoob 2");
722 }
723
724 /*
725 * Append address and data, and optionally, control (ancillary) data
726 * to the receive queue of a socket. If present,
727 * m0 must include a packet header with total length.
728 * Returns 0 if no space in sockbuf or insufficient mbufs.
729 */
730 int
731 sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa, struct mbuf *m0,
732 struct mbuf *control)
733 {
734 struct mbuf *m, *n, *nlast;
735 int space, len;
736
737 space = asa->sa_len;
738
739 if (m0 != NULL) {
740 if ((m0->m_flags & M_PKTHDR) == 0)
741 panic("sbappendaddr");
742 space += m0->m_pkthdr.len;
743 #ifdef MBUFTRACE
744 m_claimm(m0, sb->sb_mowner);
745 #endif
746 }
747 for (n = control; n; n = n->m_next) {
748 space += n->m_len;
749 MCLAIM(n, sb->sb_mowner);
750 if (n->m_next == 0) /* keep pointer to last control buf */
751 break;
752 }
753 if (space > sbspace(sb))
754 return (0);
755 MGET(m, M_DONTWAIT, MT_SONAME);
756 if (m == 0)
757 return (0);
758 MCLAIM(m, sb->sb_mowner);
759 /*
760 * XXX avoid 'comparison always true' warning which isn't easily
761 * avoided.
762 */
763 len = asa->sa_len;
764 if (len > MLEN) {
765 MEXTMALLOC(m, asa->sa_len, M_NOWAIT);
766 if ((m->m_flags & M_EXT) == 0) {
767 m_free(m);
768 return (0);
769 }
770 }
771 m->m_len = asa->sa_len;
772 memcpy(mtod(m, void *), asa, asa->sa_len);
773 if (n)
774 n->m_next = m0; /* concatenate data to control */
775 else
776 control = m0;
777 m->m_next = control;
778
779 SBLASTRECORDCHK(sb, "sbappendaddr 1");
780
781 for (n = m; n->m_next != NULL; n = n->m_next)
782 sballoc(sb, n);
783 sballoc(sb, n);
784 nlast = n;
785 SBLINKRECORD(sb, m);
786
787 sb->sb_mbtail = nlast;
788 SBLASTMBUFCHK(sb, "sbappendaddr");
789
790 SBLASTRECORDCHK(sb, "sbappendaddr 2");
791
792 return (1);
793 }
794
795 /*
796 * Helper for sbappendchainaddr: prepend a struct sockaddr* to
797 * an mbuf chain.
798 */
799 static inline struct mbuf *
800 m_prepend_sockaddr(struct sockbuf *sb, struct mbuf *m0,
801 const struct sockaddr *asa)
802 {
803 struct mbuf *m;
804 const int salen = asa->sa_len;
805
806 /* only the first in each chain need be a pkthdr */
807 MGETHDR(m, M_DONTWAIT, MT_SONAME);
808 if (m == 0)
809 return (0);
810 MCLAIM(m, sb->sb_mowner);
811 #ifdef notyet
812 if (salen > MHLEN) {
813 MEXTMALLOC(m, salen, M_NOWAIT);
814 if ((m->m_flags & M_EXT) == 0) {
815 m_free(m);
816 return (0);
817 }
818 }
819 #else
820 KASSERT(salen <= MHLEN);
821 #endif
822 m->m_len = salen;
823 memcpy(mtod(m, void *), asa, salen);
824 m->m_next = m0;
825 m->m_pkthdr.len = salen + m0->m_pkthdr.len;
826
827 return m;
828 }
829
830 int
831 sbappendaddrchain(struct sockbuf *sb, const struct sockaddr *asa,
832 struct mbuf *m0, int sbprio)
833 {
834 int space;
835 struct mbuf *m, *n, *n0, *nlast;
836 int error;
837
838 /*
839 * XXX sbprio reserved for encoding priority of this* request:
840 * SB_PRIO_NONE --> honour normal sb limits
841 * SB_PRIO_ONESHOT_OVERFLOW --> if socket has any space,
842 * take whole chain. Intended for large requests
843 * that should be delivered atomically (all, or none).
844 * SB_PRIO_OVERDRAFT -- allow a small (2*MLEN) overflow
845 * over normal socket limits, for messages indicating
846 * buffer overflow in earlier normal/lower-priority messages
847 * SB_PRIO_BESTEFFORT --> ignore limits entirely.
848 * Intended for kernel-generated messages only.
849 * Up to generator to avoid total mbuf resource exhaustion.
850 */
851 (void)sbprio;
852
853 if (m0 && (m0->m_flags & M_PKTHDR) == 0)
854 panic("sbappendaddrchain");
855
856 space = sbspace(sb);
857
858 #ifdef notyet
859 /*
860 * Enforce SB_PRIO_* limits as described above.
861 */
862 #endif
863
864 n0 = NULL;
865 nlast = NULL;
866 for (m = m0; m; m = m->m_nextpkt) {
867 struct mbuf *np;
868
869 #ifdef MBUFTRACE
870 m_claimm(m, sb->sb_mowner);
871 #endif
872
873 /* Prepend sockaddr to this record (m) of input chain m0 */
874 n = m_prepend_sockaddr(sb, m, asa);
875 if (n == NULL) {
876 error = ENOBUFS;
877 goto bad;
878 }
879
880 /* Append record (asa+m) to end of new chain n0 */
881 if (n0 == NULL) {
882 n0 = n;
883 } else {
884 nlast->m_nextpkt = n;
885 }
886 /* Keep track of last record on new chain */
887 nlast = n;
888
889 for (np = n; np; np = np->m_next)
890 sballoc(sb, np);
891 }
892
893 SBLASTRECORDCHK(sb, "sbappendaddrchain 1");
894
895 /* Drop the entire chain of (asa+m) records onto the socket */
896 SBLINKRECORDCHAIN(sb, n0, nlast);
897
898 SBLASTRECORDCHK(sb, "sbappendaddrchain 2");
899
900 for (m = nlast; m->m_next; m = m->m_next)
901 ;
902 sb->sb_mbtail = m;
903 SBLASTMBUFCHK(sb, "sbappendaddrchain");
904
905 return (1);
906
907 bad:
908 /*
909 * On error, free the prepended addreseses. For consistency
910 * with sbappendaddr(), leave it to our caller to free
911 * the input record chain passed to us as m0.
912 */
913 while ((n = n0) != NULL) {
914 struct mbuf *np;
915
916 /* Undo the sballoc() of this record */
917 for (np = n; np; np = np->m_next)
918 sbfree(sb, np);
919
920 n0 = n->m_nextpkt; /* iterate at next prepended address */
921 MFREE(n, np); /* free prepended address (not data) */
922 }
923 return 0;
924 }
925
926
927 int
928 sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control)
929 {
930 struct mbuf *m, *mlast, *n;
931 int space;
932
933 space = 0;
934 if (control == 0)
935 panic("sbappendcontrol");
936 for (m = control; ; m = m->m_next) {
937 space += m->m_len;
938 MCLAIM(m, sb->sb_mowner);
939 if (m->m_next == 0)
940 break;
941 }
942 n = m; /* save pointer to last control buffer */
943 for (m = m0; m; m = m->m_next) {
944 MCLAIM(m, sb->sb_mowner);
945 space += m->m_len;
946 }
947 if (space > sbspace(sb))
948 return (0);
949 n->m_next = m0; /* concatenate data to control */
950
951 SBLASTRECORDCHK(sb, "sbappendcontrol 1");
952
953 for (m = control; m->m_next != NULL; m = m->m_next)
954 sballoc(sb, m);
955 sballoc(sb, m);
956 mlast = m;
957 SBLINKRECORD(sb, control);
958
959 sb->sb_mbtail = mlast;
960 SBLASTMBUFCHK(sb, "sbappendcontrol");
961
962 SBLASTRECORDCHK(sb, "sbappendcontrol 2");
963
964 return (1);
965 }
966
967 /*
968 * Compress mbuf chain m into the socket
969 * buffer sb following mbuf n. If n
970 * is null, the buffer is presumed empty.
971 */
972 void
973 sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n)
974 {
975 int eor;
976 struct mbuf *o;
977
978 eor = 0;
979 while (m) {
980 eor |= m->m_flags & M_EOR;
981 if (m->m_len == 0 &&
982 (eor == 0 ||
983 (((o = m->m_next) || (o = n)) &&
984 o->m_type == m->m_type))) {
985 if (sb->sb_lastrecord == m)
986 sb->sb_lastrecord = m->m_next;
987 m = m_free(m);
988 continue;
989 }
990 if (n && (n->m_flags & M_EOR) == 0 &&
991 /* M_TRAILINGSPACE() checks buffer writeability */
992 m->m_len <= MCLBYTES / 4 && /* XXX Don't copy too much */
993 m->m_len <= M_TRAILINGSPACE(n) &&
994 n->m_type == m->m_type) {
995 memcpy(mtod(n, char *) + n->m_len, mtod(m, void *),
996 (unsigned)m->m_len);
997 n->m_len += m->m_len;
998 sb->sb_cc += m->m_len;
999 m = m_free(m);
1000 continue;
1001 }
1002 if (n)
1003 n->m_next = m;
1004 else
1005 sb->sb_mb = m;
1006 sb->sb_mbtail = m;
1007 sballoc(sb, m);
1008 n = m;
1009 m->m_flags &= ~M_EOR;
1010 m = m->m_next;
1011 n->m_next = 0;
1012 }
1013 if (eor) {
1014 if (n)
1015 n->m_flags |= eor;
1016 else
1017 printf("semi-panic: sbcompress\n");
1018 }
1019 SBLASTMBUFCHK(sb, __func__);
1020 }
1021
1022 /*
1023 * Free all mbufs in a sockbuf.
1024 * Check that all resources are reclaimed.
1025 */
1026 void
1027 sbflush(struct sockbuf *sb)
1028 {
1029
1030 KASSERT((sb->sb_flags & SB_LOCK) == 0);
1031
1032 while (sb->sb_mbcnt)
1033 sbdrop(sb, (int)sb->sb_cc);
1034
1035 KASSERT(sb->sb_cc == 0);
1036 KASSERT(sb->sb_mb == NULL);
1037 KASSERT(sb->sb_mbtail == NULL);
1038 KASSERT(sb->sb_lastrecord == NULL);
1039 }
1040
1041 /*
1042 * Drop data from (the front of) a sockbuf.
1043 */
1044 void
1045 sbdrop(struct sockbuf *sb, int len)
1046 {
1047 struct mbuf *m, *mn, *next;
1048
1049 next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
1050 while (len > 0) {
1051 if (m == 0) {
1052 if (next == 0)
1053 panic("sbdrop");
1054 m = next;
1055 next = m->m_nextpkt;
1056 continue;
1057 }
1058 if (m->m_len > len) {
1059 m->m_len -= len;
1060 m->m_data += len;
1061 sb->sb_cc -= len;
1062 break;
1063 }
1064 len -= m->m_len;
1065 sbfree(sb, m);
1066 MFREE(m, mn);
1067 m = mn;
1068 }
1069 while (m && m->m_len == 0) {
1070 sbfree(sb, m);
1071 MFREE(m, mn);
1072 m = mn;
1073 }
1074 if (m) {
1075 sb->sb_mb = m;
1076 m->m_nextpkt = next;
1077 } else
1078 sb->sb_mb = next;
1079 /*
1080 * First part is an inline SB_EMPTY_FIXUP(). Second part
1081 * makes sure sb_lastrecord is up-to-date if we dropped
1082 * part of the last record.
1083 */
1084 m = sb->sb_mb;
1085 if (m == NULL) {
1086 sb->sb_mbtail = NULL;
1087 sb->sb_lastrecord = NULL;
1088 } else if (m->m_nextpkt == NULL)
1089 sb->sb_lastrecord = m;
1090 }
1091
1092 /*
1093 * Drop a record off the front of a sockbuf
1094 * and move the next record to the front.
1095 */
1096 void
1097 sbdroprecord(struct sockbuf *sb)
1098 {
1099 struct mbuf *m, *mn;
1100
1101 m = sb->sb_mb;
1102 if (m) {
1103 sb->sb_mb = m->m_nextpkt;
1104 do {
1105 sbfree(sb, m);
1106 MFREE(m, mn);
1107 } while ((m = mn) != NULL);
1108 }
1109 SB_EMPTY_FIXUP(sb);
1110 }
1111
1112 /*
1113 * Create a "control" mbuf containing the specified data
1114 * with the specified type for presentation on a socket buffer.
1115 */
1116 struct mbuf *
1117 sbcreatecontrol(void *p, int size, int type, int level)
1118 {
1119 struct cmsghdr *cp;
1120 struct mbuf *m;
1121
1122 if (CMSG_SPACE(size) > MCLBYTES) {
1123 printf("sbcreatecontrol: message too large %d\n", size);
1124 return NULL;
1125 }
1126
1127 if ((m = m_get(M_DONTWAIT, MT_CONTROL)) == NULL)
1128 return ((struct mbuf *) NULL);
1129 if (CMSG_SPACE(size) > MLEN) {
1130 MCLGET(m, M_DONTWAIT);
1131 if ((m->m_flags & M_EXT) == 0) {
1132 m_free(m);
1133 return NULL;
1134 }
1135 }
1136 cp = mtod(m, struct cmsghdr *);
1137 memcpy(CMSG_DATA(cp), p, size);
1138 m->m_len = CMSG_SPACE(size);
1139 cp->cmsg_len = CMSG_LEN(size);
1140 cp->cmsg_level = level;
1141 cp->cmsg_type = type;
1142 return (m);
1143 }
1144