FreeBSD/Linux Kernel Cross Reference
sys/kern/uipc_mbuf.c
1 /* $NetBSD: uipc_mbuf.c,v 1.92.6.3 2006/09/08 12:06:51 ghen Exp $ */
2
3 /*-
4 * Copyright (c) 1999, 2001 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9 * NASA Ames Research Center.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. All advertising materials mentioning features or use of this software
20 * must display the following acknowledgement:
21 * This product includes software developed by the NetBSD
22 * Foundation, Inc. and its contributors.
23 * 4. Neither the name of The NetBSD Foundation nor the names of its
24 * contributors may be used to endorse or promote products derived
25 * from this software without specific prior written permission.
26 *
27 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
28 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
29 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
30 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
31 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37 * POSSIBILITY OF SUCH DAMAGE.
38 */
39
40 /*
41 * Copyright (c) 1982, 1986, 1988, 1991, 1993
42 * The Regents of the University of California. All rights reserved.
43 *
44 * Redistribution and use in source and binary forms, with or without
45 * modification, are permitted provided that the following conditions
46 * are met:
47 * 1. Redistributions of source code must retain the above copyright
48 * notice, this list of conditions and the following disclaimer.
49 * 2. Redistributions in binary form must reproduce the above copyright
50 * notice, this list of conditions and the following disclaimer in the
51 * documentation and/or other materials provided with the distribution.
52 * 3. Neither the name of the University nor the names of its contributors
53 * may be used to endorse or promote products derived from this software
54 * without specific prior written permission.
55 *
56 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
57 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
58 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
59 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
60 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
61 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
62 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
63 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
64 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
65 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
66 * SUCH DAMAGE.
67 *
68 * @(#)uipc_mbuf.c 8.4 (Berkeley) 2/14/95
69 */
70
71 #include <sys/cdefs.h>
72 __KERNEL_RCSID(0, "$NetBSD: uipc_mbuf.c,v 1.92.6.3 2006/09/08 12:06:51 ghen Exp $");
73
74 #include "opt_mbuftrace.h"
75
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/proc.h>
79 #include <sys/malloc.h>
80 #define MBTYPES
81 #include <sys/mbuf.h>
82 #include <sys/kernel.h>
83 #include <sys/syslog.h>
84 #include <sys/domain.h>
85 #include <sys/protosw.h>
86 #include <sys/pool.h>
87 #include <sys/socket.h>
88 #include <sys/sysctl.h>
89
90 #include <net/if.h>
91
92 #include <uvm/uvm.h>
93
94
95 struct pool mbpool; /* mbuf pool */
96 struct pool mclpool; /* mbuf cluster pool */
97
98 struct pool_cache mbpool_cache;
99 struct pool_cache mclpool_cache;
100
101 struct mbstat mbstat;
102 int max_linkhdr;
103 int max_protohdr;
104 int max_hdr;
105 int max_datalen;
106
107 static int mb_ctor(void *, void *, int);
108
109 void *mclpool_alloc(struct pool *, int);
110 void mclpool_release(struct pool *, void *);
111
112 struct pool_allocator mclpool_allocator = {
113 mclpool_alloc, mclpool_release, 0,
114 };
115
116 static struct mbuf *m_copym0(struct mbuf *, int, int, int, int);
117 static struct mbuf *m_split0(struct mbuf *, int, int, int);
118 static int m_copyback0(struct mbuf **, int, int, const void *, int, int);
119
120 /* flags for m_copyback0 */
121 #define M_COPYBACK0_COPYBACK 0x0001 /* copyback from cp */
122 #define M_COPYBACK0_PRESERVE 0x0002 /* preserve original data */
123 #define M_COPYBACK0_COW 0x0004 /* do copy-on-write */
124 #define M_COPYBACK0_EXTEND 0x0008 /* extend chain */
125
126 const char mclpool_warnmsg[] =
127 "WARNING: mclpool limit reached; increase NMBCLUSTERS";
128
129 MALLOC_DEFINE(M_MBUF, "mbuf", "mbuf");
130
131 #ifdef MBUFTRACE
132 struct mownerhead mowners = LIST_HEAD_INITIALIZER(mowners);
133 struct mowner unknown_mowners[] = {
134 { "unknown", "free" },
135 { "unknown", "data" },
136 { "unknown", "header" },
137 { "unknown", "soname" },
138 { "unknown", "soopts" },
139 { "unknown", "ftable" },
140 { "unknown", "control" },
141 { "unknown", "oobdata" },
142 };
143 struct mowner revoked_mowner = { "revoked", "" };
144 #endif
145
146 /*
147 * Initialize the mbuf allocator.
148 */
149 void
150 mbinit(void)
151 {
152
153 KASSERT(sizeof(struct _m_ext) <= MHLEN);
154 KASSERT(sizeof(struct mbuf) == MSIZE);
155
156 pool_init(&mbpool, msize, 0, 0, 0, "mbpl", NULL);
157 pool_init(&mclpool, mclbytes, 0, 0, 0, "mclpl", &mclpool_allocator);
158
159 pool_set_drain_hook(&mbpool, m_reclaim, NULL);
160 pool_set_drain_hook(&mclpool, m_reclaim, NULL);
161
162 pool_cache_init(&mbpool_cache, &mbpool, mb_ctor, NULL, NULL);
163 pool_cache_init(&mclpool_cache, &mclpool, NULL, NULL, NULL);
164
165 /*
166 * Set the hard limit on the mclpool to the number of
167 * mbuf clusters the kernel is to support. Log the limit
168 * reached message max once a minute.
169 */
170 pool_sethardlimit(&mclpool, nmbclusters, mclpool_warnmsg, 60);
171
172 /*
173 * Set a low water mark for both mbufs and clusters. This should
174 * help ensure that they can be allocated in a memory starvation
175 * situation. This is important for e.g. diskless systems which
176 * must allocate mbufs in order for the pagedaemon to clean pages.
177 */
178 pool_setlowat(&mbpool, mblowat);
179 pool_setlowat(&mclpool, mcllowat);
180
181 #ifdef MBUFTRACE
182 {
183 /*
184 * Attach the unknown mowners.
185 */
186 int i;
187 MOWNER_ATTACH(&revoked_mowner);
188 for (i = sizeof(unknown_mowners)/sizeof(unknown_mowners[0]);
189 i-- > 0; )
190 MOWNER_ATTACH(&unknown_mowners[i]);
191 }
192 #endif
193 }
194
195 /*
196 * sysctl helper routine for the kern.mbuf subtree. nmbclusters may
197 * or may not be writable, and mblowat and mcllowat need range
198 * checking and pool tweaking after being reset.
199 */
200 static int
201 sysctl_kern_mbuf(SYSCTLFN_ARGS)
202 {
203 int error, newval;
204 struct sysctlnode node;
205
206 node = *rnode;
207 node.sysctl_data = &newval;
208 switch (rnode->sysctl_num) {
209 case MBUF_NMBCLUSTERS:
210 if (mb_map != NULL) {
211 node.sysctl_flags &= ~CTLFLAG_READWRITE;
212 node.sysctl_flags |= CTLFLAG_READONLY;
213 }
214 /* FALLTHROUGH */
215 case MBUF_MBLOWAT:
216 case MBUF_MCLLOWAT:
217 newval = *(int*)rnode->sysctl_data;
218 break;
219 default:
220 return (EOPNOTSUPP);
221 }
222
223 error = sysctl_lookup(SYSCTLFN_CALL(&node));
224 if (error || newp == NULL)
225 return (error);
226 if (newval < 0)
227 return (EINVAL);
228
229 switch (node.sysctl_num) {
230 case MBUF_NMBCLUSTERS:
231 if (newval < nmbclusters)
232 return (EINVAL);
233 nmbclusters = newval;
234 pool_sethardlimit(&mclpool, nmbclusters, mclpool_warnmsg, 60);
235 break;
236 case MBUF_MBLOWAT:
237 mblowat = newval;
238 pool_setlowat(&mbpool, mblowat);
239 break;
240 case MBUF_MCLLOWAT:
241 mcllowat = newval;
242 pool_setlowat(&mclpool, mcllowat);
243 break;
244 }
245
246 return (0);
247 }
248
249 #ifdef MBUFTRACE
250 static int
251 sysctl_kern_mbuf_mowners(SYSCTLFN_ARGS)
252 {
253 struct mowner *mo;
254 size_t len = 0;
255 int error = 0;
256
257 if (namelen != 0)
258 return (EINVAL);
259 if (newp != NULL)
260 return (EPERM);
261
262 LIST_FOREACH(mo, &mowners, mo_link) {
263 if (oldp != NULL) {
264 if (*oldlenp - len < sizeof(*mo)) {
265 error = ENOMEM;
266 break;
267 }
268 error = copyout(mo, (caddr_t) oldp + len,
269 sizeof(*mo));
270 if (error)
271 break;
272 }
273 len += sizeof(*mo);
274 }
275
276 if (error == 0)
277 *oldlenp = len;
278
279 return (error);
280 }
281 #endif /* MBUFTRACE */
282
283 SYSCTL_SETUP(sysctl_kern_mbuf_setup, "sysctl kern.mbuf subtree setup")
284 {
285
286 sysctl_createv(clog, 0, NULL, NULL,
287 CTLFLAG_PERMANENT,
288 CTLTYPE_NODE, "kern", NULL,
289 NULL, 0, NULL, 0,
290 CTL_KERN, CTL_EOL);
291 sysctl_createv(clog, 0, NULL, NULL,
292 CTLFLAG_PERMANENT,
293 CTLTYPE_NODE, "mbuf",
294 SYSCTL_DESCR("mbuf control variables"),
295 NULL, 0, NULL, 0,
296 CTL_KERN, KERN_MBUF, CTL_EOL);
297
298 sysctl_createv(clog, 0, NULL, NULL,
299 CTLFLAG_PERMANENT|CTLFLAG_IMMEDIATE,
300 CTLTYPE_INT, "msize",
301 SYSCTL_DESCR("mbuf base size"),
302 NULL, msize, NULL, 0,
303 CTL_KERN, KERN_MBUF, MBUF_MSIZE, CTL_EOL);
304 sysctl_createv(clog, 0, NULL, NULL,
305 CTLFLAG_PERMANENT|CTLFLAG_IMMEDIATE,
306 CTLTYPE_INT, "mclbytes",
307 SYSCTL_DESCR("mbuf cluster size"),
308 NULL, mclbytes, NULL, 0,
309 CTL_KERN, KERN_MBUF, MBUF_MCLBYTES, CTL_EOL);
310 sysctl_createv(clog, 0, NULL, NULL,
311 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
312 CTLTYPE_INT, "nmbclusters",
313 SYSCTL_DESCR("Limit on the number of mbuf clusters"),
314 sysctl_kern_mbuf, 0, &nmbclusters, 0,
315 CTL_KERN, KERN_MBUF, MBUF_NMBCLUSTERS, CTL_EOL);
316 sysctl_createv(clog, 0, NULL, NULL,
317 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
318 CTLTYPE_INT, "mblowat",
319 SYSCTL_DESCR("mbuf low water mark"),
320 sysctl_kern_mbuf, 0, &mblowat, 0,
321 CTL_KERN, KERN_MBUF, MBUF_MBLOWAT, CTL_EOL);
322 sysctl_createv(clog, 0, NULL, NULL,
323 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
324 CTLTYPE_INT, "mcllowat",
325 SYSCTL_DESCR("mbuf cluster low water mark"),
326 sysctl_kern_mbuf, 0, &mcllowat, 0,
327 CTL_KERN, KERN_MBUF, MBUF_MCLLOWAT, CTL_EOL);
328 sysctl_createv(clog, 0, NULL, NULL,
329 CTLFLAG_PERMANENT,
330 CTLTYPE_STRUCT, "stats",
331 SYSCTL_DESCR("mbuf allocation statistics"),
332 NULL, 0, &mbstat, sizeof(mbstat),
333 CTL_KERN, KERN_MBUF, MBUF_STATS, CTL_EOL);
334 #ifdef MBUFTRACE
335 sysctl_createv(clog, 0, NULL, NULL,
336 CTLFLAG_PERMANENT,
337 CTLTYPE_STRUCT, "mowners",
338 SYSCTL_DESCR("Information about mbuf owners"),
339 sysctl_kern_mbuf_mowners, 0, NULL, 0,
340 CTL_KERN, KERN_MBUF, MBUF_MOWNERS, CTL_EOL);
341 #endif /* MBUFTRACE */
342 }
343
344 void *
345 mclpool_alloc(struct pool *pp, int flags)
346 {
347 boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
348
349 return ((void *)uvm_km_alloc_poolpage1(mb_map, NULL, waitok));
350 }
351
352 void
353 mclpool_release(struct pool *pp, void *v)
354 {
355
356 uvm_km_free_poolpage1(mb_map, (vaddr_t)v);
357 }
358
359 /*ARGSUSED*/
360 static int
361 mb_ctor(void *arg, void *object, int flags)
362 {
363 struct mbuf *m = object;
364
365 #ifdef POOL_VTOPHYS
366 m->m_paddr = POOL_VTOPHYS(m);
367 #else
368 m->m_paddr = M_PADDR_INVALID;
369 #endif
370 return (0);
371 }
372
373 void
374 m_reclaim(void *arg, int flags)
375 {
376 struct domain *dp;
377 const struct protosw *pr;
378 struct ifnet *ifp;
379 int s = splvm();
380
381 DOMAIN_FOREACH(dp) {
382 for (pr = dp->dom_protosw;
383 pr < dp->dom_protoswNPROTOSW; pr++)
384 if (pr->pr_drain)
385 (*pr->pr_drain)();
386 }
387 IFNET_FOREACH(ifp) {
388 if (ifp->if_drain)
389 (*ifp->if_drain)(ifp);
390 }
391 splx(s);
392 mbstat.m_drain++;
393 }
394
395 /*
396 * Space allocation routines.
397 * These are also available as macros
398 * for critical paths.
399 */
400 struct mbuf *
401 m_get(int nowait, int type)
402 {
403 struct mbuf *m;
404
405 MGET(m, nowait, type);
406 return (m);
407 }
408
409 struct mbuf *
410 m_gethdr(int nowait, int type)
411 {
412 struct mbuf *m;
413
414 MGETHDR(m, nowait, type);
415 return (m);
416 }
417
418 struct mbuf *
419 m_getclr(int nowait, int type)
420 {
421 struct mbuf *m;
422
423 MGET(m, nowait, type);
424 if (m == 0)
425 return (NULL);
426 memset(mtod(m, caddr_t), 0, MLEN);
427 return (m);
428 }
429
430 void
431 m_clget(struct mbuf *m, int nowait)
432 {
433
434 MCLGET(m, nowait);
435 }
436
437 struct mbuf *
438 m_free(struct mbuf *m)
439 {
440 struct mbuf *n;
441
442 MFREE(m, n);
443 return (n);
444 }
445
446 void
447 m_freem(struct mbuf *m)
448 {
449 struct mbuf *n;
450
451 if (m == NULL)
452 return;
453 do {
454 MFREE(m, n);
455 m = n;
456 } while (m);
457 }
458
459 #ifdef MBUFTRACE
460 /*
461 * Walk a chain of mbufs, claiming ownership of each mbuf in the chain.
462 */
463 void
464 m_claimm(struct mbuf *m, struct mowner *mo)
465 {
466
467 for (; m != NULL; m = m->m_next)
468 MCLAIM(m, mo);
469 }
470 #endif
471
472 /*
473 * Mbuffer utility routines.
474 */
475
476 /*
477 * Lesser-used path for M_PREPEND:
478 * allocate new mbuf to prepend to chain,
479 * copy junk along.
480 */
481 struct mbuf *
482 m_prepend(struct mbuf *m, int len, int how)
483 {
484 struct mbuf *mn;
485
486 MGET(mn, how, m->m_type);
487 if (mn == (struct mbuf *)NULL) {
488 m_freem(m);
489 return ((struct mbuf *)NULL);
490 }
491 if (m->m_flags & M_PKTHDR) {
492 M_COPY_PKTHDR(mn, m);
493 m_tag_delete_chain(m, NULL);
494 m->m_flags &= ~M_PKTHDR;
495 } else {
496 MCLAIM(mn, m->m_owner);
497 }
498 mn->m_next = m;
499 m = mn;
500 if (len < MHLEN)
501 MH_ALIGN(m, len);
502 m->m_len = len;
503 return (m);
504 }
505
506 /*
507 * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
508 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf.
509 * The wait parameter is a choice of M_WAIT/M_DONTWAIT from caller.
510 */
511 int MCFail;
512
513 struct mbuf *
514 m_copym(struct mbuf *m, int off0, int len, int wait)
515 {
516
517 return m_copym0(m, off0, len, wait, 0); /* shallow copy on M_EXT */
518 }
519
520 struct mbuf *
521 m_dup(struct mbuf *m, int off0, int len, int wait)
522 {
523
524 return m_copym0(m, off0, len, wait, 1); /* deep copy */
525 }
526
527 static struct mbuf *
528 m_copym0(struct mbuf *m, int off0, int len, int wait, int deep)
529 {
530 struct mbuf *n, **np;
531 int off = off0;
532 struct mbuf *top;
533 int copyhdr = 0;
534
535 if (off < 0 || len < 0)
536 panic("m_copym: off %d, len %d", off, len);
537 if (off == 0 && m->m_flags & M_PKTHDR)
538 copyhdr = 1;
539 while (off > 0) {
540 if (m == 0)
541 panic("m_copym: m == 0, off %d", off);
542 if (off < m->m_len)
543 break;
544 off -= m->m_len;
545 m = m->m_next;
546 }
547 np = ⊤
548 top = 0;
549 while (len > 0) {
550 if (m == 0) {
551 if (len != M_COPYALL)
552 panic("m_copym: m == 0, len %d [!COPYALL]",
553 len);
554 break;
555 }
556 MGET(n, wait, m->m_type);
557 *np = n;
558 if (n == 0)
559 goto nospace;
560 MCLAIM(n, m->m_owner);
561 if (copyhdr) {
562 M_COPY_PKTHDR(n, m);
563 if (len == M_COPYALL)
564 n->m_pkthdr.len -= off0;
565 else
566 n->m_pkthdr.len = len;
567 copyhdr = 0;
568 }
569 n->m_len = min(len, m->m_len - off);
570 if (m->m_flags & M_EXT) {
571 if (!deep) {
572 n->m_data = m->m_data + off;
573 n->m_ext = m->m_ext;
574 MCLADDREFERENCE(m, n);
575 } else {
576 /*
577 * we are unsure about the way m was allocated.
578 * copy into multiple MCLBYTES cluster mbufs.
579 */
580 MCLGET(n, wait);
581 n->m_len = 0;
582 n->m_len = M_TRAILINGSPACE(n);
583 n->m_len = min(n->m_len, len);
584 n->m_len = min(n->m_len, m->m_len - off);
585 memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + off,
586 (unsigned)n->m_len);
587 }
588 } else
589 memcpy(mtod(n, caddr_t), mtod(m, caddr_t)+off,
590 (unsigned)n->m_len);
591 if (len != M_COPYALL)
592 len -= n->m_len;
593 off += n->m_len;
594 #ifdef DIAGNOSTIC
595 if (off > m->m_len)
596 panic("m_copym0 overrun");
597 #endif
598 if (off == m->m_len) {
599 m = m->m_next;
600 off = 0;
601 }
602 np = &n->m_next;
603 }
604 if (top == 0)
605 MCFail++;
606 return (top);
607 nospace:
608 m_freem(top);
609 MCFail++;
610 return (NULL);
611 }
612
613 /*
614 * Copy an entire packet, including header (which must be present).
615 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
616 */
617 struct mbuf *
618 m_copypacket(struct mbuf *m, int how)
619 {
620 struct mbuf *top, *n, *o;
621
622 MGET(n, how, m->m_type);
623 top = n;
624 if (!n)
625 goto nospace;
626
627 MCLAIM(n, m->m_owner);
628 M_COPY_PKTHDR(n, m);
629 n->m_len = m->m_len;
630 if (m->m_flags & M_EXT) {
631 n->m_data = m->m_data;
632 n->m_ext = m->m_ext;
633 MCLADDREFERENCE(m, n);
634 } else {
635 memcpy(mtod(n, char *), mtod(m, char *), n->m_len);
636 }
637
638 m = m->m_next;
639 while (m) {
640 MGET(o, how, m->m_type);
641 if (!o)
642 goto nospace;
643
644 MCLAIM(o, m->m_owner);
645 n->m_next = o;
646 n = n->m_next;
647
648 n->m_len = m->m_len;
649 if (m->m_flags & M_EXT) {
650 n->m_data = m->m_data;
651 n->m_ext = m->m_ext;
652 MCLADDREFERENCE(m, n);
653 } else {
654 memcpy(mtod(n, char *), mtod(m, char *), n->m_len);
655 }
656
657 m = m->m_next;
658 }
659 return top;
660 nospace:
661 m_freem(top);
662 MCFail++;
663 return NULL;
664 }
665
666 /*
667 * Copy data from an mbuf chain starting "off" bytes from the beginning,
668 * continuing for "len" bytes, into the indicated buffer.
669 */
670 void
671 m_copydata(struct mbuf *m, int off, int len, void *vp)
672 {
673 unsigned count;
674 char *cp = vp;
675
676 if (off < 0 || len < 0)
677 panic("m_copydata: off %d, len %d", off, len);
678 while (off > 0) {
679 if (m == 0)
680 panic("m_copydata: m == 0, off %d", off);
681 if (off < m->m_len)
682 break;
683 off -= m->m_len;
684 m = m->m_next;
685 }
686 while (len > 0) {
687 if (m == 0)
688 panic("m_copydata: m == 0, len %d", len);
689 count = min(m->m_len - off, len);
690 memcpy(cp, mtod(m, caddr_t) + off, count);
691 len -= count;
692 cp += count;
693 off = 0;
694 m = m->m_next;
695 }
696 }
697
698 /*
699 * Concatenate mbuf chain n to m.
700 * n might be copied into m (when n->m_len is small), therefore data portion of
701 * n could be copied into an mbuf of different mbuf type.
702 * Any m_pkthdr is not updated.
703 */
704 void
705 m_cat(struct mbuf *m, struct mbuf *n)
706 {
707
708 while (m->m_next)
709 m = m->m_next;
710 while (n) {
711 if (M_READONLY(m) || n->m_len > M_TRAILINGSPACE(m)) {
712 /* just join the two chains */
713 m->m_next = n;
714 return;
715 }
716 /* splat the data from one into the other */
717 memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
718 (u_int)n->m_len);
719 m->m_len += n->m_len;
720 n = m_free(n);
721 }
722 }
723
724 void
725 m_adj(struct mbuf *mp, int req_len)
726 {
727 int len = req_len;
728 struct mbuf *m;
729 int count;
730
731 if ((m = mp) == NULL)
732 return;
733 if (len >= 0) {
734 /*
735 * Trim from head.
736 */
737 while (m != NULL && len > 0) {
738 if (m->m_len <= len) {
739 len -= m->m_len;
740 m->m_len = 0;
741 m = m->m_next;
742 } else {
743 m->m_len -= len;
744 m->m_data += len;
745 len = 0;
746 }
747 }
748 m = mp;
749 if (mp->m_flags & M_PKTHDR)
750 m->m_pkthdr.len -= (req_len - len);
751 } else {
752 /*
753 * Trim from tail. Scan the mbuf chain,
754 * calculating its length and finding the last mbuf.
755 * If the adjustment only affects this mbuf, then just
756 * adjust and return. Otherwise, rescan and truncate
757 * after the remaining size.
758 */
759 len = -len;
760 count = 0;
761 for (;;) {
762 count += m->m_len;
763 if (m->m_next == (struct mbuf *)0)
764 break;
765 m = m->m_next;
766 }
767 if (m->m_len >= len) {
768 m->m_len -= len;
769 if (mp->m_flags & M_PKTHDR)
770 mp->m_pkthdr.len -= len;
771 return;
772 }
773 count -= len;
774 if (count < 0)
775 count = 0;
776 /*
777 * Correct length for chain is "count".
778 * Find the mbuf with last data, adjust its length,
779 * and toss data from remaining mbufs on chain.
780 */
781 m = mp;
782 if (m->m_flags & M_PKTHDR)
783 m->m_pkthdr.len = count;
784 for (; m; m = m->m_next) {
785 if (m->m_len >= count) {
786 m->m_len = count;
787 break;
788 }
789 count -= m->m_len;
790 }
791 while (m->m_next)
792 (m = m->m_next) ->m_len = 0;
793 }
794 }
795
796 /*
797 * Rearange an mbuf chain so that len bytes are contiguous
798 * and in the data area of an mbuf (so that mtod and dtom
799 * will work for a structure of size len). Returns the resulting
800 * mbuf chain on success, frees it and returns null on failure.
801 * If there is room, it will add up to max_protohdr-len extra bytes to the
802 * contiguous region in an attempt to avoid being called next time.
803 */
804 int MPFail;
805
806 struct mbuf *
807 m_pullup(struct mbuf *n, int len)
808 {
809 struct mbuf *m;
810 int count;
811 int space;
812
813 /*
814 * If first mbuf has no cluster, and has room for len bytes
815 * without shifting current data, pullup into it,
816 * otherwise allocate a new mbuf to prepend to the chain.
817 */
818 if ((n->m_flags & M_EXT) == 0 &&
819 n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
820 if (n->m_len >= len)
821 return (n);
822 m = n;
823 n = n->m_next;
824 len -= m->m_len;
825 } else {
826 if (len > MHLEN)
827 goto bad;
828 MGET(m, M_DONTWAIT, n->m_type);
829 if (m == 0)
830 goto bad;
831 MCLAIM(m, n->m_owner);
832 m->m_len = 0;
833 if (n->m_flags & M_PKTHDR) {
834 M_COPY_PKTHDR(m, n);
835 m_tag_delete_chain(n, NULL);
836 n->m_flags &= ~M_PKTHDR;
837 }
838 }
839 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
840 do {
841 count = min(min(max(len, max_protohdr), space), n->m_len);
842 memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
843 (unsigned)count);
844 len -= count;
845 m->m_len += count;
846 n->m_len -= count;
847 space -= count;
848 if (n->m_len)
849 n->m_data += count;
850 else
851 n = m_free(n);
852 } while (len > 0 && n);
853 if (len > 0) {
854 (void) m_free(m);
855 goto bad;
856 }
857 m->m_next = n;
858 return (m);
859 bad:
860 m_freem(n);
861 MPFail++;
862 return (NULL);
863 }
864
865 /*
866 * Like m_pullup(), except a new mbuf is always allocated, and we allow
867 * the amount of empty space before the data in the new mbuf to be specified
868 * (in the event that the caller expects to prepend later).
869 */
870 int MSFail;
871
872 struct mbuf *
873 m_copyup(struct mbuf *n, int len, int dstoff)
874 {
875 struct mbuf *m;
876 int count, space;
877
878 if (len > (MHLEN - dstoff))
879 goto bad;
880 MGET(m, M_DONTWAIT, n->m_type);
881 if (m == NULL)
882 goto bad;
883 MCLAIM(m, n->m_owner);
884 m->m_len = 0;
885 if (n->m_flags & M_PKTHDR) {
886 M_COPY_PKTHDR(m, n);
887 m_tag_delete_chain(n, NULL);
888 n->m_flags &= ~M_PKTHDR;
889 }
890 m->m_data += dstoff;
891 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
892 do {
893 count = min(min(max(len, max_protohdr), space), n->m_len);
894 memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
895 (unsigned)count);
896 len -= count;
897 m->m_len += count;
898 n->m_len -= count;
899 space -= count;
900 if (n->m_len)
901 n->m_data += count;
902 else
903 n = m_free(n);
904 } while (len > 0 && n);
905 if (len > 0) {
906 (void) m_free(m);
907 goto bad;
908 }
909 m->m_next = n;
910 return (m);
911 bad:
912 m_freem(n);
913 MSFail++;
914 return (NULL);
915 }
916
917 /*
918 * Partition an mbuf chain in two pieces, returning the tail --
919 * all but the first len0 bytes. In case of failure, it returns NULL and
920 * attempts to restore the chain to its original state.
921 */
922 struct mbuf *
923 m_split(struct mbuf *m0, int len0, int wait)
924 {
925
926 return m_split0(m0, len0, wait, 1);
927 }
928
929 static struct mbuf *
930 m_split0(struct mbuf *m0, int len0, int wait, int copyhdr)
931 {
932 struct mbuf *m, *n;
933 unsigned len = len0, remain, len_save;
934
935 for (m = m0; m && len > m->m_len; m = m->m_next)
936 len -= m->m_len;
937 if (m == 0)
938 return (NULL);
939 remain = m->m_len - len;
940 if (copyhdr && (m0->m_flags & M_PKTHDR)) {
941 MGETHDR(n, wait, m0->m_type);
942 if (n == 0)
943 return (NULL);
944 MCLAIM(n, m0->m_owner);
945 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
946 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
947 len_save = m0->m_pkthdr.len;
948 m0->m_pkthdr.len = len0;
949 if (m->m_flags & M_EXT)
950 goto extpacket;
951 if (remain > MHLEN) {
952 /* m can't be the lead packet */
953 MH_ALIGN(n, 0);
954 n->m_next = m_split(m, len, wait);
955 if (n->m_next == 0) {
956 (void) m_free(n);
957 m0->m_pkthdr.len = len_save;
958 return (NULL);
959 } else
960 return (n);
961 } else
962 MH_ALIGN(n, remain);
963 } else if (remain == 0) {
964 n = m->m_next;
965 m->m_next = 0;
966 return (n);
967 } else {
968 MGET(n, wait, m->m_type);
969 if (n == 0)
970 return (NULL);
971 MCLAIM(n, m->m_owner);
972 M_ALIGN(n, remain);
973 }
974 extpacket:
975 if (m->m_flags & M_EXT) {
976 n->m_ext = m->m_ext;
977 MCLADDREFERENCE(m, n);
978 n->m_data = m->m_data + len;
979 } else {
980 memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + len, remain);
981 }
982 n->m_len = remain;
983 m->m_len = len;
984 n->m_next = m->m_next;
985 m->m_next = 0;
986 return (n);
987 }
988 /*
989 * Routine to copy from device local memory into mbufs.
990 */
991 struct mbuf *
992 m_devget(char *buf, int totlen, int off0, struct ifnet *ifp,
993 void (*copy)(const void *from, void *to, size_t len))
994 {
995 struct mbuf *m;
996 struct mbuf *top = 0, **mp = ⊤
997 int off = off0, len;
998 char *cp;
999 char *epkt;
1000
1001 cp = buf;
1002 epkt = cp + totlen;
1003 if (off) {
1004 /*
1005 * If 'off' is non-zero, packet is trailer-encapsulated,
1006 * so we have to skip the type and length fields.
1007 */
1008 cp += off + 2 * sizeof(u_int16_t);
1009 totlen -= 2 * sizeof(u_int16_t);
1010 }
1011 MGETHDR(m, M_DONTWAIT, MT_DATA);
1012 if (m == 0)
1013 return (NULL);
1014 m->m_pkthdr.rcvif = ifp;
1015 m->m_pkthdr.len = totlen;
1016 m->m_len = MHLEN;
1017
1018 while (totlen > 0) {
1019 if (top) {
1020 MGET(m, M_DONTWAIT, MT_DATA);
1021 if (m == 0) {
1022 m_freem(top);
1023 return (NULL);
1024 }
1025 m->m_len = MLEN;
1026 }
1027 len = min(totlen, epkt - cp);
1028 if (len >= MINCLSIZE) {
1029 MCLGET(m, M_DONTWAIT);
1030 if ((m->m_flags & M_EXT) == 0) {
1031 m_free(m);
1032 m_freem(top);
1033 return (NULL);
1034 }
1035 m->m_len = len = min(len, MCLBYTES);
1036 } else {
1037 /*
1038 * Place initial small packet/header at end of mbuf.
1039 */
1040 if (len < m->m_len) {
1041 if (top == 0 && len + max_linkhdr <= m->m_len)
1042 m->m_data += max_linkhdr;
1043 m->m_len = len;
1044 } else
1045 len = m->m_len;
1046 }
1047 if (copy)
1048 copy(cp, mtod(m, caddr_t), (size_t)len);
1049 else
1050 memcpy(mtod(m, caddr_t), cp, (size_t)len);
1051 cp += len;
1052 *mp = m;
1053 mp = &m->m_next;
1054 totlen -= len;
1055 if (cp == epkt)
1056 cp = buf;
1057 }
1058 return (top);
1059 }
1060
1061 /*
1062 * Copy data from a buffer back into the indicated mbuf chain,
1063 * starting "off" bytes from the beginning, extending the mbuf
1064 * chain if necessary.
1065 */
1066 void
1067 m_copyback(struct mbuf *m0, int off, int len, const void *cp)
1068 {
1069 #if defined(DEBUG)
1070 struct mbuf *origm = m0;
1071 int error;
1072 #endif /* defined(DEBUG) */
1073
1074 if (m0 == NULL)
1075 return;
1076
1077 #if defined(DEBUG)
1078 error =
1079 #endif /* defined(DEBUG) */
1080 m_copyback0(&m0, off, len, cp,
1081 M_COPYBACK0_COPYBACK|M_COPYBACK0_EXTEND, M_DONTWAIT);
1082
1083 #if defined(DEBUG)
1084 if (error != 0 || (m0 != NULL && origm != m0))
1085 panic("m_copyback");
1086 #endif /* defined(DEBUG) */
1087 }
1088
1089 struct mbuf *
1090 m_copyback_cow(struct mbuf *m0, int off, int len, const void *cp, int how)
1091 {
1092 int error;
1093
1094 /* don't support chain expansion */
1095 KDASSERT(off + len <= m_length(m0));
1096
1097 error = m_copyback0(&m0, off, len, cp,
1098 M_COPYBACK0_COPYBACK|M_COPYBACK0_COW, how);
1099 if (error) {
1100 /*
1101 * no way to recover from partial success.
1102 * just free the chain.
1103 */
1104 m_freem(m0);
1105 return NULL;
1106 }
1107 return m0;
1108 }
1109
1110 /*
1111 * m_makewritable: ensure the specified range writable.
1112 */
1113 int
1114 m_makewritable(struct mbuf **mp, int off, int len, int how)
1115 {
1116 int error;
1117 #if defined(DEBUG)
1118 struct mbuf *n;
1119 int origlen, reslen;
1120
1121 origlen = m_length(*mp);
1122 #endif /* defined(DEBUG) */
1123
1124 #if 0 /* M_COPYALL is large enough */
1125 if (len == M_COPYALL)
1126 len = m_length(*mp) - off; /* XXX */
1127 #endif
1128
1129 error = m_copyback0(mp, off, len, NULL,
1130 M_COPYBACK0_PRESERVE|M_COPYBACK0_COW, how);
1131
1132 #if defined(DEBUG)
1133 reslen = 0;
1134 for (n = *mp; n; n = n->m_next)
1135 reslen += n->m_len;
1136 if (origlen != reslen)
1137 panic("m_makewritable: length changed");
1138 if (((*mp)->m_flags & M_PKTHDR) != 0 && reslen != (*mp)->m_pkthdr.len)
1139 panic("m_makewritable: inconsist");
1140 #endif /* defined(DEBUG) */
1141
1142 return error;
1143 }
1144
1145 int
1146 m_copyback0(struct mbuf **mp0, int off, int len, const void *vp, int flags,
1147 int how)
1148 {
1149 int mlen;
1150 struct mbuf *m, *n;
1151 struct mbuf **mp;
1152 int totlen = 0;
1153 const char *cp = vp;
1154
1155 KASSERT(mp0 != NULL);
1156 KASSERT(*mp0 != NULL);
1157 KASSERT((flags & M_COPYBACK0_PRESERVE) == 0 || cp == NULL);
1158 KASSERT((flags & M_COPYBACK0_COPYBACK) == 0 || cp != NULL);
1159
1160 mp = mp0;
1161 m = *mp;
1162 while (off > (mlen = m->m_len)) {
1163 off -= mlen;
1164 totlen += mlen;
1165 if (m->m_next == 0) {
1166 if ((flags & M_COPYBACK0_EXTEND) == 0)
1167 goto out;
1168 n = m_getclr(how, m->m_type);
1169 if (n == 0)
1170 goto out;
1171 n->m_len = min(MLEN, len + off);
1172 m->m_next = n;
1173 }
1174 mp = &m->m_next;
1175 m = m->m_next;
1176 }
1177 while (len > 0) {
1178 mlen = m->m_len - off;
1179 if (mlen != 0 && M_READONLY(m)) {
1180 char *datap;
1181 int eatlen;
1182
1183 /*
1184 * this mbuf is read-only.
1185 * allocate a new writable mbuf and try again.
1186 */
1187
1188 #if defined(DIAGNOSTIC)
1189 if ((flags & M_COPYBACK0_COW) == 0)
1190 panic("m_copyback0: read-only");
1191 #endif /* defined(DIAGNOSTIC) */
1192
1193 /*
1194 * if we're going to write into the middle of
1195 * a mbuf, split it first.
1196 */
1197 if (off > 0 && len < mlen) {
1198 n = m_split0(m, off, how, 0);
1199 if (n == NULL)
1200 goto enobufs;
1201 m->m_next = n;
1202 mp = &m->m_next;
1203 m = n;
1204 off = 0;
1205 continue;
1206 }
1207
1208 /*
1209 * XXX TODO coalesce into the trailingspace of
1210 * the previous mbuf when possible.
1211 */
1212
1213 /*
1214 * allocate a new mbuf. copy packet header if needed.
1215 */
1216 MGET(n, how, m->m_type);
1217 if (n == NULL)
1218 goto enobufs;
1219 MCLAIM(n, m->m_owner);
1220 if (off == 0 && (m->m_flags & M_PKTHDR) != 0) {
1221 /* XXX M_MOVE_PKTHDR */
1222 M_COPY_PKTHDR(n, m);
1223 n->m_len = MHLEN;
1224 } else {
1225 if (len >= MINCLSIZE)
1226 MCLGET(n, M_DONTWAIT);
1227 n->m_len =
1228 (n->m_flags & M_EXT) ? MCLBYTES : MLEN;
1229 }
1230 if (n->m_len > len)
1231 n->m_len = len;
1232
1233 /*
1234 * free the region which has been overwritten.
1235 * copying data from old mbufs if requested.
1236 */
1237 if (flags & M_COPYBACK0_PRESERVE)
1238 datap = mtod(n, char *);
1239 else
1240 datap = NULL;
1241 eatlen = n->m_len;
1242 KDASSERT(off == 0 || eatlen >= mlen);
1243 if (off > 0) {
1244 KDASSERT(len >= mlen);
1245 m->m_len = off;
1246 m->m_next = n;
1247 if (datap) {
1248 m_copydata(m, off, mlen, datap);
1249 datap += mlen;
1250 }
1251 eatlen -= mlen;
1252 mp = &m->m_next;
1253 m = m->m_next;
1254 }
1255 while (m != NULL && M_READONLY(m) &&
1256 n->m_type == m->m_type && eatlen > 0) {
1257 mlen = min(eatlen, m->m_len);
1258 if (datap) {
1259 m_copydata(m, 0, mlen, datap);
1260 datap += mlen;
1261 }
1262 m->m_data += mlen;
1263 m->m_len -= mlen;
1264 eatlen -= mlen;
1265 if (m->m_len == 0)
1266 *mp = m = m_free(m);
1267 }
1268 if (eatlen > 0)
1269 n->m_len -= eatlen;
1270 n->m_next = m;
1271 *mp = m = n;
1272 continue;
1273 }
1274 mlen = min(mlen, len);
1275 if (flags & M_COPYBACK0_COPYBACK) {
1276 memcpy(mtod(m, caddr_t) + off, cp, (unsigned)mlen);
1277 cp += mlen;
1278 }
1279 len -= mlen;
1280 mlen += off;
1281 off = 0;
1282 totlen += mlen;
1283 if (len == 0)
1284 break;
1285 if (m->m_next == 0) {
1286 if ((flags & M_COPYBACK0_EXTEND) == 0)
1287 goto out;
1288 n = m_get(how, m->m_type);
1289 if (n == 0)
1290 break;
1291 n->m_len = min(MLEN, len);
1292 m->m_next = n;
1293 }
1294 mp = &m->m_next;
1295 m = m->m_next;
1296 }
1297 out: if (((m = *mp0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
1298 m->m_pkthdr.len = totlen;
1299
1300 return 0;
1301
1302 enobufs:
1303 return ENOBUFS;
1304 }
1305
1306 /*
1307 * Apply function f to the data in an mbuf chain starting "off" bytes from the
1308 * beginning, continuing for "len" bytes.
1309 */
1310 int
1311 m_apply(struct mbuf *m, int off, int len,
1312 int (*f)(void *, caddr_t, unsigned int), void *arg)
1313 {
1314 unsigned int count;
1315 int rval;
1316
1317 KASSERT(len >= 0);
1318 KASSERT(off >= 0);
1319
1320 while (off > 0) {
1321 KASSERT(m != NULL);
1322 if (off < m->m_len)
1323 break;
1324 off -= m->m_len;
1325 m = m->m_next;
1326 }
1327 while (len > 0) {
1328 KASSERT(m != NULL);
1329 count = min(m->m_len - off, len);
1330
1331 rval = (*f)(arg, mtod(m, caddr_t) + off, count);
1332 if (rval)
1333 return (rval);
1334
1335 len -= count;
1336 off = 0;
1337 m = m->m_next;
1338 }
1339
1340 return (0);
1341 }
1342
1343 /*
1344 * Return a pointer to mbuf/offset of location in mbuf chain.
1345 */
1346 struct mbuf *
1347 m_getptr(struct mbuf *m, int loc, int *off)
1348 {
1349
1350 while (loc >= 0) {
1351 /* Normal end of search */
1352 if (m->m_len > loc) {
1353 *off = loc;
1354 return (m);
1355 } else {
1356 loc -= m->m_len;
1357
1358 if (m->m_next == NULL) {
1359 if (loc == 0) {
1360 /* Point at the end of valid data */
1361 *off = m->m_len;
1362 return (m);
1363 } else
1364 return (NULL);
1365 } else
1366 m = m->m_next;
1367 }
1368 }
1369
1370 return (NULL);
1371 }
Cache object: 3231edc31fdce8c185070b41bcad2153
|