1/*-
2 * Copyright (c) 1982, 1986, 1988, 1991, 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 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * @(#)uipc_mbuf.c 8.2 (Berkeley) 1/4/94
30 */
31
32#include <sys/cdefs.h>
33__FBSDID("$FreeBSD: head/sys/kern/uipc_mbuf.c 155780 2006-02-17 14:14:15Z andre $");
34
35#include "opt_mac.h"
36#include "opt_param.h"
37#include "opt_mbuf_stress_test.h"
38
39#include <sys/param.h>
40#include <sys/systm.h>
41#include <sys/kernel.h>
42#include <sys/limits.h>
43#include <sys/lock.h>
44#include <sys/mac.h>
45#include <sys/malloc.h>
46#include <sys/mbuf.h>
47#include <sys/sysctl.h>
48#include <sys/domain.h>
49#include <sys/protosw.h>
50#include <sys/uio.h>
51
52int max_linkhdr;
53int max_protohdr;
54int max_hdr;
55int max_datalen;
56#ifdef MBUF_STRESS_TEST
57int m_defragpackets;
58int m_defragbytes;
59int m_defraguseless;
60int m_defragfailure;
61int m_defragrandomfailures;
62#endif
63
64/*
65 * sysctl(8) exported objects
66 */
67SYSCTL_DECL(_kern_ipc);
68SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RW,
69 &max_linkhdr, 0, "");
70SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RW,
71 &max_protohdr, 0, "");
72SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RW, &max_hdr, 0, "");
73SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RW,
74 &max_datalen, 0, "");
75#ifdef MBUF_STRESS_TEST
76SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD,
77 &m_defragpackets, 0, "");
78SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD,
79 &m_defragbytes, 0, "");
80SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD,
81 &m_defraguseless, 0, "");
82SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD,
83 &m_defragfailure, 0, "");
84SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW,
85 &m_defragrandomfailures, 0, "");
86#endif
87
88/*
89 * Allocate a given length worth of mbufs and/or clusters (whatever fits
90 * best) and return a pointer to the top of the allocated chain. If an
91 * existing mbuf chain is provided, then we will append the new chain
92 * to the existing one but still return the top of the newly allocated
93 * chain.
94 */
95struct mbuf *
96m_getm(struct mbuf *m, int len, int how, short type)
97{
98 struct mbuf *mb, *top, *cur, *mtail;
99 int num, rem;
100 int i;
101
102 KASSERT(len >= 0, ("m_getm(): len is < 0"));
103
104 /* If m != NULL, we will append to the end of that chain. */
105 if (m != NULL)
106 for (mtail = m; mtail->m_next != NULL; mtail = mtail->m_next);
107 else
108 mtail = NULL;
109
110 /*
111 * Calculate how many mbufs+clusters ("packets") we need and how much
112 * leftover there is after that and allocate the first mbuf+cluster
113 * if required.
114 */
115 num = len / MCLBYTES;
116 rem = len % MCLBYTES;
117 top = cur = NULL;
118 if (num > 0) {
119 if ((top = cur = m_getcl(how, type, 0)) == NULL)
120 goto failed;
121 top->m_len = 0;
122 }
123 num--;
124
125 for (i = 0; i < num; i++) {
126 mb = m_getcl(how, type, 0);
127 if (mb == NULL)
128 goto failed;
129 mb->m_len = 0;
130 cur = (cur->m_next = mb);
131 }
132 if (rem > 0) {
133 mb = (rem > MINCLSIZE) ?
134 m_getcl(how, type, 0) : m_get(how, type);
135 if (mb == NULL)
136 goto failed;
137 mb->m_len = 0;
138 if (cur == NULL)
139 top = mb;
140 else
141 cur->m_next = mb;
142 }
143
144 if (mtail != NULL)
145 mtail->m_next = top;
146 return top;
147failed:
148 if (top != NULL)
149 m_freem(top);
150 return NULL;
151}
152
153/*
154 * Free an entire chain of mbufs and associated external buffers, if
155 * applicable.
156 */
157void
158m_freem(struct mbuf *mb)
159{
160
161 while (mb != NULL)
162 mb = m_free(mb);
163}
164
165/*-
166 * Configure a provided mbuf to refer to the provided external storage
167 * buffer and setup a reference count for said buffer. If the setting
168 * up of the reference count fails, the M_EXT bit will not be set. If
169 * successfull, the M_EXT bit is set in the mbuf's flags.
170 *
171 * Arguments:
172 * mb The existing mbuf to which to attach the provided buffer.
173 * buf The address of the provided external storage buffer.
174 * size The size of the provided buffer.
175 * freef A pointer to a routine that is responsible for freeing the
176 * provided external storage buffer.
177 * args A pointer to an argument structure (of any type) to be passed
178 * to the provided freef routine (may be NULL).
179 * flags Any other flags to be passed to the provided mbuf.
180 * type The type that the external storage buffer should be
181 * labeled with.
182 *
183 * Returns:
184 * Nothing.
185 */
186void
187m_extadd(struct mbuf *mb, caddr_t buf, u_int size,
188 void (*freef)(void *, void *), void *args, int flags, int type)
189{
190 KASSERT(type != EXT_CLUSTER, ("%s: EXT_CLUSTER not allowed", __func__));
191
192 if (type != EXT_EXTREF)
193 mb->m_ext.ref_cnt = (u_int *)uma_zalloc(zone_ext_refcnt, M_NOWAIT);
194 if (mb->m_ext.ref_cnt != NULL) {
195 *(mb->m_ext.ref_cnt) = 1;
196 mb->m_flags |= (M_EXT | flags);
197 mb->m_ext.ext_buf = buf;
198 mb->m_data = mb->m_ext.ext_buf;
199 mb->m_ext.ext_size = size;
200 mb->m_ext.ext_free = freef;
201 mb->m_ext.ext_args = args;
202 mb->m_ext.ext_type = type;
203 }
204}
205
206/*
207 * Non-directly-exported function to clean up after mbufs with M_EXT
208 * storage attached to them if the reference count hits 1.
209 */
210void
211mb_free_ext(struct mbuf *m)
212{
213 KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__));
214 KASSERT(m->m_ext.ref_cnt != NULL, ("%s: ref_cnt not set", __func__));
215
216 /* Free attached storage if this mbuf is the only reference to it. */
217 if (*(m->m_ext.ref_cnt) == 1 ||
218 atomic_fetchadd_int(m->m_ext.ref_cnt, -1) == 0) {
219 switch (m->m_ext.ext_type) {
220 case EXT_PACKET: /* The packet zone is special. */
221 if (*(m->m_ext.ref_cnt) == 0)
222 *(m->m_ext.ref_cnt) = 1;
223 uma_zfree(zone_pack, m);
224 return; /* Job done. */
225 case EXT_CLUSTER:
226 uma_zfree(zone_clust, m->m_ext.ext_buf);
227 break;
228 case EXT_JUMBOP:
229 uma_zfree(zone_jumbop, m->m_ext.ext_buf);
230 break;
231 case EXT_JUMBO9:
232 uma_zfree(zone_jumbo9, m->m_ext.ext_buf);
233 break;
234 case EXT_JUMBO16:
235 uma_zfree(zone_jumbo16, m->m_ext.ext_buf);
236 break;
237 case EXT_SFBUF:
238 case EXT_NET_DRV:
239 case EXT_MOD_TYPE:
240 case EXT_DISPOSABLE:
241 *(m->m_ext.ref_cnt) = 0;
242 uma_zfree(zone_ext_refcnt, __DEVOLATILE(u_int *,
243 m->m_ext.ref_cnt));
244 /* FALLTHROUGH */
245 case EXT_EXTREF:
246 KASSERT(m->m_ext.ext_free != NULL,
247 ("%s: ext_free not set", __func__));
248 (*(m->m_ext.ext_free))(m->m_ext.ext_buf,
249 m->m_ext.ext_args);
250 break;
251 default:
252 KASSERT(m->m_ext.ext_type == 0,
253 ("%s: unknown ext_type", __func__));
254 }
255 }
256 /*
257 * Free this mbuf back to the mbuf zone with all m_ext
258 * information purged.
259 */
260 m->m_ext.ext_buf = NULL;
261 m->m_ext.ext_free = NULL;
262 m->m_ext.ext_args = NULL;
263 m->m_ext.ref_cnt = NULL;
264 m->m_ext.ext_size = 0;
265 m->m_ext.ext_type = 0;
266 m->m_flags &= ~M_EXT;
267 uma_zfree(zone_mbuf, m);
268}
269
270/*
271 * Attach the the cluster from *m to *n, set up m_ext in *n
272 * and bump the refcount of the cluster.
273 */
274static void
275mb_dupcl(struct mbuf *n, struct mbuf *m)
276{
277 KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__));
278 KASSERT(m->m_ext.ref_cnt != NULL, ("%s: ref_cnt not set", __func__));
279 KASSERT((n->m_flags & M_EXT) == 0, ("%s: M_EXT set", __func__));
280
281 if (*(m->m_ext.ref_cnt) == 1)
282 *(m->m_ext.ref_cnt) += 1;
283 else
284 atomic_add_int(m->m_ext.ref_cnt, 1);
285 n->m_ext.ext_buf = m->m_ext.ext_buf;
286 n->m_ext.ext_free = m->m_ext.ext_free;
287 n->m_ext.ext_args = m->m_ext.ext_args;
288 n->m_ext.ext_size = m->m_ext.ext_size;
289 n->m_ext.ref_cnt = m->m_ext.ref_cnt;
290 n->m_ext.ext_type = m->m_ext.ext_type;
291 n->m_flags |= M_EXT;
292}
293
294/*
295 * Clean up mbuf (chain) from any tags and packet headers.
296 * If "all" is set then the first mbuf in the chain will be
297 * cleaned too.
298 */
299void
300m_demote(struct mbuf *m0, int all)
301{
302 struct mbuf *m;
303
304 for (m = all ? m0 : m0->m_next; m != NULL; m = m->m_next) {
305 if (m->m_flags & M_PKTHDR) {
306 m_tag_delete_chain(m, NULL);
307 m->m_flags &= ~M_PKTHDR;
308 bzero(&m->m_pkthdr, sizeof(struct pkthdr));
309 }
310 if (m->m_type == MT_HEADER)
311 m->m_type = MT_DATA;
312 if (m != m0 && m->m_nextpkt != NULL)
313 m->m_nextpkt = NULL;
314 m->m_flags = m->m_flags & (M_EXT|M_EOR|M_RDONLY|M_FREELIST);
315 }
316}
317
318/*
319 * Sanity checks on mbuf (chain) for use in KASSERT() and general
320 * debugging.
321 * Returns 0 or panics when bad and 1 on all tests passed.
322 * Sanitize, 0 to run M_SANITY_ACTION, 1 to garble things so they
323 * blow up later.
324 */
325int
326m_sanity(struct mbuf *m0, int sanitize)
327{
328 struct mbuf *m;
329 caddr_t a, b;
330 int pktlen = 0;
331
332#define M_SANITY_ACTION(s) return (0)
333/* #define M_SANITY_ACTION(s) panic("mbuf %p: " s, m) */
334
335 for (m = m0; m != NULL; m = m->m_next) {
336 /*
337 * Basic pointer checks. If any of these fails then some
338 * unrelated kernel memory before or after us is trashed.
339 * No way to recover from that.
340 */
341 a = ((m->m_flags & M_EXT) ? m->m_ext.ext_buf :
342 ((m->m_flags & M_PKTHDR) ? (caddr_t)(&m->m_pktdat) :
343 (caddr_t)(&m->m_dat)) );
344 b = (caddr_t)(a + (m->m_flags & M_EXT ? m->m_ext.ext_size :
345 ((m->m_flags & M_PKTHDR) ? MHLEN : MLEN)));
346 if ((caddr_t)m->m_data < a)
347 M_SANITY_ACTION("m_data outside mbuf data range left");
348 if ((caddr_t)m->m_data > b)
349 M_SANITY_ACTION("m_data outside mbuf data range right");
350 if ((caddr_t)m->m_data + m->m_len > b)
351 M_SANITY_ACTION("m_data + m_len exeeds mbuf space");
352 if ((m->m_flags & M_PKTHDR) && m->m_pkthdr.header) {
353 if ((caddr_t)m->m_pkthdr.header < a ||
354 (caddr_t)m->m_pkthdr.header > b)
355 M_SANITY_ACTION("m_pkthdr.header outside mbuf data range");
356 }
357
358 /* m->m_nextpkt may only be set on first mbuf in chain. */
359 if (m != m0 && m->m_nextpkt != NULL) {
360 if (sanitize) {
361 m_freem(m->m_nextpkt);
362 m->m_nextpkt = (struct mbuf *)0xDEADC0DE;
363 } else
364 M_SANITY_ACTION("m->m_nextpkt on in-chain mbuf");
365 }
366
367 /* correct type correlations. */
368 if (m->m_type == MT_HEADER && !(m->m_flags & M_PKTHDR)) {
369 if (sanitize)
370 m->m_type = MT_DATA;
371 else
372 M_SANITY_ACTION("MT_HEADER set but not M_PKTHDR");
373 }
374
375 /* packet length (not mbuf length!) calculation */
376 if (m0->m_flags & M_PKTHDR)
377 pktlen += m->m_len;
378
379 /* m_tags may only be attached to first mbuf in chain. */
380 if (m != m0 && m->m_flags & M_PKTHDR &&
381 !SLIST_EMPTY(&m->m_pkthdr.tags)) {
382 if (sanitize) {
383 m_tag_delete_chain(m, NULL);
384 /* put in 0xDEADC0DE perhaps? */
385 } else
386 M_SANITY_ACTION("m_tags on in-chain mbuf");
387 }
388
389 /* M_PKTHDR may only be set on first mbuf in chain */
390 if (m != m0 && m->m_flags & M_PKTHDR) {
391 if (sanitize) {
392 bzero(&m->m_pkthdr, sizeof(m->m_pkthdr));
393 m->m_flags &= ~M_PKTHDR;
394 /* put in 0xDEADCODE and leave hdr flag in */
395 } else
396 M_SANITY_ACTION("M_PKTHDR on in-chain mbuf");
397 }
398 }
399 m = m0;
400 if (pktlen && pktlen != m->m_pkthdr.len) {
401 if (sanitize)
402 m->m_pkthdr.len = 0;
403 else
404 M_SANITY_ACTION("m_pkthdr.len != mbuf chain length");
405 }
406 return 1;
407
408#undef M_SANITY_ACTION
409}
410
411
412/*
413 * "Move" mbuf pkthdr from "from" to "to".
414 * "from" must have M_PKTHDR set, and "to" must be empty.
415 */
416void
417m_move_pkthdr(struct mbuf *to, struct mbuf *from)
418{
419
420#if 0
421 /* see below for why these are not enabled */
422 M_ASSERTPKTHDR(to);
423 /* Note: with MAC, this may not be a good assertion. */
424 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags),
425 ("m_move_pkthdr: to has tags"));
426#endif
427#ifdef MAC
428 /*
429 * XXXMAC: It could be this should also occur for non-MAC?
430 */
431 if (to->m_flags & M_PKTHDR)
432 m_tag_delete_chain(to, NULL);
433#endif
434 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT);
435 if ((to->m_flags & M_EXT) == 0)
436 to->m_data = to->m_pktdat;
437 to->m_pkthdr = from->m_pkthdr; /* especially tags */
438 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */
439 from->m_flags &= ~M_PKTHDR;
440}
441
442/*
443 * Duplicate "from"'s mbuf pkthdr in "to".
444 * "from" must have M_PKTHDR set, and "to" must be empty.
445 * In particular, this does a deep copy of the packet tags.
446 */
447int
448m_dup_pkthdr(struct mbuf *to, struct mbuf *from, int how)
449{
450
451#if 0
452 /*
453 * The mbuf allocator only initializes the pkthdr
454 * when the mbuf is allocated with MGETHDR. Many users
455 * (e.g. m_copy*, m_prepend) use MGET and then
456 * smash the pkthdr as needed causing these
457 * assertions to trip. For now just disable them.
458 */
459 M_ASSERTPKTHDR(to);
460 /* Note: with MAC, this may not be a good assertion. */
461 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), ("m_dup_pkthdr: to has tags"));
462#endif
463 MBUF_CHECKSLEEP(how);
464#ifdef MAC
465 if (to->m_flags & M_PKTHDR)
466 m_tag_delete_chain(to, NULL);
467#endif
468 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT);
469 if ((to->m_flags & M_EXT) == 0)
470 to->m_data = to->m_pktdat;
471 to->m_pkthdr = from->m_pkthdr;
472 SLIST_INIT(&to->m_pkthdr.tags);
473 return (m_tag_copy_chain(to, from, MBTOM(how)));
474}
475
476/*
477 * Lesser-used path for M_PREPEND:
478 * allocate new mbuf to prepend to chain,
479 * copy junk along.
480 */
481struct mbuf *
482m_prepend(struct mbuf *m, int len, int how)
483{
484 struct mbuf *mn;
485
486 if (m->m_flags & M_PKTHDR)
487 MGETHDR(mn, how, m->m_type);
488 else
489 MGET(mn, how, m->m_type);
490 if (mn == NULL) {
491 m_freem(m);
492 return (NULL);
493 }
494 if (m->m_flags & M_PKTHDR)
495 M_MOVE_PKTHDR(mn, m);
496 mn->m_next = m;
497 m = mn;
498 if (len < MHLEN)
499 MH_ALIGN(m, len);
500 m->m_len = len;
501 return (m);
502}
503
504/*
505 * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
506 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf.
507 * The wait parameter is a choice of M_TRYWAIT/M_DONTWAIT from caller.
508 * Note that the copy is read-only, because clusters are not copied,
509 * only their reference counts are incremented.
510 */
511struct mbuf *
512m_copym(struct mbuf *m, int off0, int len, int wait)
513{
514 struct mbuf *n, **np;
515 int off = off0;
516 struct mbuf *top;
517 int copyhdr = 0;
518
519 KASSERT(off >= 0, ("m_copym, negative off %d", off));
520 KASSERT(len >= 0, ("m_copym, negative len %d", len));
521 MBUF_CHECKSLEEP(wait);
522 if (off == 0 && m->m_flags & M_PKTHDR)
523 copyhdr = 1;
524 while (off > 0) {
525 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain"));
526 if (off < m->m_len)
527 break;
528 off -= m->m_len;
529 m = m->m_next;
530 }
531 np = ⊤
532 top = 0;
533 while (len > 0) {
534 if (m == NULL) {
535 KASSERT(len == M_COPYALL,
536 ("m_copym, length > size of mbuf chain"));
537 break;
538 }
539 if (copyhdr)
540 MGETHDR(n, wait, m->m_type);
541 else
542 MGET(n, wait, m->m_type);
543 *np = n;
544 if (n == NULL)
545 goto nospace;
546 if (copyhdr) {
547 if (!m_dup_pkthdr(n, m, wait))
548 goto nospace;
549 if (len == M_COPYALL)
550 n->m_pkthdr.len -= off0;
551 else
552 n->m_pkthdr.len = len;
553 copyhdr = 0;
554 }
555 n->m_len = min(len, m->m_len - off);
556 if (m->m_flags & M_EXT) {
557 n->m_data = m->m_data + off;
558 mb_dupcl(n, m);
559 } else
560 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
561 (u_int)n->m_len);
562 if (len != M_COPYALL)
563 len -= n->m_len;
564 off = 0;
565 m = m->m_next;
566 np = &n->m_next;
567 }
568 if (top == NULL)
569 mbstat.m_mcfail++; /* XXX: No consistency. */
570
571 return (top);
572nospace:
573 m_freem(top);
574 mbstat.m_mcfail++; /* XXX: No consistency. */
575 return (NULL);
576}
577
578/*
579 * Returns mbuf chain with new head for the prepending case.
580 * Copies from mbuf (chain) n from off for len to mbuf (chain) m
581 * either prepending or appending the data.
582 * The resulting mbuf (chain) m is fully writeable.
583 * m is destination (is made writeable)
584 * n is source, off is offset in source, len is len from offset
585 * dir, 0 append, 1 prepend
586 * how, wait or nowait
587 */
588
589static int
590m_bcopyxxx(void *s, void *t, u_int len)
591{
592 bcopy(s, t, (size_t)len);
593 return 0;
594}
595
596struct mbuf *
597m_copymdata(struct mbuf *m, struct mbuf *n, int off, int len,
598 int prep, int how)
599{
600 struct mbuf *mm, *x, *z, *prev = NULL;
601 caddr_t p;
602 int i, nlen = 0;
603 caddr_t buf[MLEN];
604
605 KASSERT(m != NULL && n != NULL, ("m_copymdata, no target or source"));
606 KASSERT(off >= 0, ("m_copymdata, negative off %d", off));
607 KASSERT(len >= 0, ("m_copymdata, negative len %d", len));
608 KASSERT(prep == 0 || prep == 1, ("m_copymdata, unknown direction %d", prep));
609
610 mm = m;
611 if (!prep) {
612 while(mm->m_next) {
613 prev = mm;
614 mm = mm->m_next;
615 }
616 }
617 for (z = n; z != NULL; z = z->m_next)
618 nlen += z->m_len;
619 if (len == M_COPYALL)
620 len = nlen - off;
621 if (off + len > nlen || len < 1)
622 return NULL;
623
624 if (!M_WRITABLE(mm)) {
625 /* XXX: Use proper m_xxx function instead. */
626 x = m_getcl(how, MT_DATA, mm->m_flags);
627 if (x == NULL)
628 return NULL;
629 bcopy(mm->m_ext.ext_buf, x->m_ext.ext_buf, x->m_ext.ext_size);
630 p = x->m_ext.ext_buf + (mm->m_data - mm->m_ext.ext_buf);
631 x->m_data = p;
632 mm->m_next = NULL;
633 if (mm != m)
634 prev->m_next = x;
635 m_free(mm);
636 mm = x;
637 }
638
639 /*
640 * Append/prepend the data. Allocating mbufs as necessary.
641 */
642 /* Shortcut if enough free space in first/last mbuf. */
643 if (!prep && M_TRAILINGSPACE(mm) >= len) {
644 m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t) +
645 mm->m_len);
646 mm->m_len += len;
647 mm->m_pkthdr.len += len;
648 return m;
649 }
650 if (prep && M_LEADINGSPACE(mm) >= len) {
651 mm->m_data = mtod(mm, caddr_t) - len;
652 m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t));
653 mm->m_len += len;
654 mm->m_pkthdr.len += len;
655 return mm;
656 }
657
658 /* Expand first/last mbuf to cluster if possible. */
659 if (!prep && !(mm->m_flags & M_EXT) && len > M_TRAILINGSPACE(mm)) {
660 bcopy(mm->m_data, &buf, mm->m_len);
661 m_clget(mm, how);
662 if (!(mm->m_flags & M_EXT))
663 return NULL;
664 bcopy(&buf, mm->m_ext.ext_buf, mm->m_len);
665 mm->m_data = mm->m_ext.ext_buf;
666 mm->m_pkthdr.header = NULL;
667 }
668 if (prep && !(mm->m_flags & M_EXT) && len > M_LEADINGSPACE(mm)) {
669 bcopy(mm->m_data, &buf, mm->m_len);
670 m_clget(mm, how);
671 if (!(mm->m_flags & M_EXT))
672 return NULL;
673 bcopy(&buf, (caddr_t *)mm->m_ext.ext_buf +
674 mm->m_ext.ext_size - mm->m_len, mm->m_len);
675 mm->m_data = (caddr_t)mm->m_ext.ext_buf +
676 mm->m_ext.ext_size - mm->m_len;
677 mm->m_pkthdr.header = NULL;
678 }
679
680 /* Append/prepend as many mbuf (clusters) as necessary to fit len. */
681 if (!prep && len > M_TRAILINGSPACE(mm)) {
682 if (!m_getm(mm, len - M_TRAILINGSPACE(mm), how, MT_DATA))
683 return NULL;
684 }
685 if (prep && len > M_LEADINGSPACE(mm)) {
686 if (!(z = m_getm(NULL, len - M_LEADINGSPACE(mm), how, MT_DATA)))
687 return NULL;
688 i = 0;
689 for (x = z; x != NULL; x = x->m_next) {
690 i += x->m_flags & M_EXT ? x->m_ext.ext_size :
691 (x->m_flags & M_PKTHDR ? MHLEN : MLEN);
692 if (!x->m_next)
693 break;
694 }
695 z->m_data += i - len;
696 m_move_pkthdr(mm, z);
697 x->m_next = mm;
698 mm = z;
699 }
700
701 /* Seek to start position in source mbuf. Optimization for long chains. */
702 while (off > 0) {
703 if (off < n->m_len)
704 break;
705 off -= n->m_len;
706 n = n->m_next;
707 }
708
709 /* Copy data into target mbuf. */
710 z = mm;
711 while (len > 0) {
712 KASSERT(z != NULL, ("m_copymdata, falling off target edge"));
713 i = M_TRAILINGSPACE(z);
714 m_apply(n, off, i, m_bcopyxxx, mtod(z, caddr_t) + z->m_len);
715 z->m_len += i;
716 /* fixup pkthdr.len if necessary */
717 if ((prep ? mm : m)->m_flags & M_PKTHDR)
718 (prep ? mm : m)->m_pkthdr.len += i;
719 off += i;
720 len -= i;
721 z = z->m_next;
722 }
723 return (prep ? mm : m);
724}
725
726/*
727 * Copy an entire packet, including header (which must be present).
728 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
729 * Note that the copy is read-only, because clusters are not copied,
730 * only their reference counts are incremented.
731 * Preserve alignment of the first mbuf so if the creator has left
732 * some room at the beginning (e.g. for inserting protocol headers)
733 * the copies still have the room available.
734 */
735struct mbuf *
736m_copypacket(struct mbuf *m, int how)
737{
738 struct mbuf *top, *n, *o;
739
740 MBUF_CHECKSLEEP(how);
741 MGET(n, how, m->m_type);
742 top = n;
743 if (n == NULL)
744 goto nospace;
745
746 if (!m_dup_pkthdr(n, m, how))
747 goto nospace;
748 n->m_len = m->m_len;
749 if (m->m_flags & M_EXT) {
750 n->m_data = m->m_data;
751 mb_dupcl(n, m);
752 } else {
753 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat );
754 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
755 }
756
757 m = m->m_next;
758 while (m) {
759 MGET(o, how, m->m_type);
760 if (o == NULL)
761 goto nospace;
762
763 n->m_next = o;
764 n = n->m_next;
765
766 n->m_len = m->m_len;
767 if (m->m_flags & M_EXT) {
768 n->m_data = m->m_data;
769 mb_dupcl(n, m);
770 } else {
771 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
772 }
773
774 m = m->m_next;
775 }
776 return top;
777nospace:
778 m_freem(top);
779 mbstat.m_mcfail++; /* XXX: No consistency. */
780 return (NULL);
781}
782
783/*
784 * Copy data from an mbuf chain starting "off" bytes from the beginning,
785 * continuing for "len" bytes, into the indicated buffer.
786 */
787void
788m_copydata(const struct mbuf *m, int off, int len, caddr_t cp)
789{
790 u_int count;
791
792 KASSERT(off >= 0, ("m_copydata, negative off %d", off));
793 KASSERT(len >= 0, ("m_copydata, negative len %d", len));
794 while (off > 0) {
795 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain"));
796 if (off < m->m_len)
797 break;
798 off -= m->m_len;
799 m = m->m_next;
800 }
801 while (len > 0) {
802 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain"));
803 count = min(m->m_len - off, len);
804 bcopy(mtod(m, caddr_t) + off, cp, count);
805 len -= count;
806 cp += count;
807 off = 0;
808 m = m->m_next;
809 }
810}
811
812/*
813 * Copy a packet header mbuf chain into a completely new chain, including
814 * copying any mbuf clusters. Use this instead of m_copypacket() when
815 * you need a writable copy of an mbuf chain.
816 */
817struct mbuf *
818m_dup(struct mbuf *m, int how)
819{
820 struct mbuf **p, *top = NULL;
821 int remain, moff, nsize;
822
823 MBUF_CHECKSLEEP(how);
824 /* Sanity check */
825 if (m == NULL)
826 return (NULL);
827 M_ASSERTPKTHDR(m);
828
829 /* While there's more data, get a new mbuf, tack it on, and fill it */
830 remain = m->m_pkthdr.len;
831 moff = 0;
832 p = ⊤
833 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */
834 struct mbuf *n;
835
836 /* Get the next new mbuf */
837 if (remain >= MINCLSIZE) {
838 n = m_getcl(how, m->m_type, 0);
839 nsize = MCLBYTES;
840 } else {
841 n = m_get(how, m->m_type);
842 nsize = MLEN;
843 }
844 if (n == NULL)
845 goto nospace;
846
847 if (top == NULL) { /* First one, must be PKTHDR */
848 if (!m_dup_pkthdr(n, m, how)) {
849 m_free(n);
850 goto nospace;
851 }
852 if ((n->m_flags & M_EXT) == 0)
853 nsize = MHLEN;
854 }
855 n->m_len = 0;
856
857 /* Link it into the new chain */
858 *p = n;
859 p = &n->m_next;
860
861 /* Copy data from original mbuf(s) into new mbuf */
862 while (n->m_len < nsize && m != NULL) {
863 int chunk = min(nsize - n->m_len, m->m_len - moff);
864
865 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
866 moff += chunk;
867 n->m_len += chunk;
868 remain -= chunk;
869 if (moff == m->m_len) {
870 m = m->m_next;
871 moff = 0;
872 }
873 }
874
875 /* Check correct total mbuf length */
876 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL),
877 ("%s: bogus m_pkthdr.len", __func__));
878 }
879 return (top);
880
881nospace:
882 m_freem(top);
883 mbstat.m_mcfail++; /* XXX: No consistency. */
884 return (NULL);
885}
886
887/*
888 * Concatenate mbuf chain n to m.
889 * Both chains must be of the same type (e.g. MT_DATA).
890 * Any m_pkthdr is not updated.
891 */
892void
893m_cat(struct mbuf *m, struct mbuf *n)
894{
895 while (m->m_next)
896 m = m->m_next;
897 while (n) {
898 if (m->m_flags & M_EXT ||
899 m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) {
900 /* just join the two chains */
901 m->m_next = n;
902 return;
903 }
904 /* splat the data from one into the other */
905 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
906 (u_int)n->m_len);
907 m->m_len += n->m_len;
908 n = m_free(n);
909 }
910}
911
912void
913m_adj(struct mbuf *mp, int req_len)
914{
915 int len = req_len;
916 struct mbuf *m;
917 int count;
918
919 if ((m = mp) == NULL)
920 return;
921 if (len >= 0) {
922 /*
923 * Trim from head.
924 */
925 while (m != NULL && len > 0) {
926 if (m->m_len <= len) {
927 len -= m->m_len;
928 m->m_len = 0;
929 m = m->m_next;
930 } else {
931 m->m_len -= len;
932 m->m_data += len;
933 len = 0;
934 }
935 }
936 m = mp;
937 if (mp->m_flags & M_PKTHDR)
938 m->m_pkthdr.len -= (req_len - len);
939 } else {
940 /*
941 * Trim from tail. Scan the mbuf chain,
942 * calculating its length and finding the last mbuf.
943 * If the adjustment only affects this mbuf, then just
944 * adjust and return. Otherwise, rescan and truncate
945 * after the remaining size.
946 */
947 len = -len;
948 count = 0;
949 for (;;) {
950 count += m->m_len;
951 if (m->m_next == (struct mbuf *)0)
952 break;
953 m = m->m_next;
954 }
955 if (m->m_len >= len) {
956 m->m_len -= len;
957 if (mp->m_flags & M_PKTHDR)
958 mp->m_pkthdr.len -= len;
959 return;
960 }
961 count -= len;
962 if (count < 0)
963 count = 0;
964 /*
965 * Correct length for chain is "count".
966 * Find the mbuf with last data, adjust its length,
967 * and toss data from remaining mbufs on chain.
968 */
969 m = mp;
970 if (m->m_flags & M_PKTHDR)
971 m->m_pkthdr.len = count;
972 for (; m; m = m->m_next) {
973 if (m->m_len >= count) {
974 m->m_len = count;
975 if (m->m_next != NULL) {
976 m_freem(m->m_next);
977 m->m_next = NULL;
978 }
979 break;
980 }
981 count -= m->m_len;
982 }
983 }
984}
985
986/*
987 * Rearange an mbuf chain so that len bytes are contiguous
988 * and in the data area of an mbuf (so that mtod and dtom
989 * will work for a structure of size len). Returns the resulting
990 * mbuf chain on success, frees it and returns null on failure.
991 * If there is room, it will add up to max_protohdr-len extra bytes to the
992 * contiguous region in an attempt to avoid being called next time.
993 */
994struct mbuf *
995m_pullup(struct mbuf *n, int len)
996{
997 struct mbuf *m;
998 int count;
999 int space;
1000
1001 /*
1002 * If first mbuf has no cluster, and has room for len bytes
1003 * without shifting current data, pullup into it,
1004 * otherwise allocate a new mbuf to prepend to the chain.
1005 */
1006 if ((n->m_flags & M_EXT) == 0 &&
1007 n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
1008 if (n->m_len >= len)
1009 return (n);
1010 m = n;
1011 n = n->m_next;
1012 len -= m->m_len;
1013 } else {
1014 if (len > MHLEN)
1015 goto bad;
1016 MGET(m, M_DONTWAIT, n->m_type);
1017 if (m == NULL)
1018 goto bad;
1019 m->m_len = 0;
1020 if (n->m_flags & M_PKTHDR)
1021 M_MOVE_PKTHDR(m, n);
1022 }
1023 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1024 do {
1025 count = min(min(max(len, max_protohdr), space), n->m_len);
1026 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
1027 (u_int)count);
1028 len -= count;
1029 m->m_len += count;
1030 n->m_len -= count;
1031 space -= count;
1032 if (n->m_len)
1033 n->m_data += count;
1034 else
1035 n = m_free(n);
1036 } while (len > 0 && n);
1037 if (len > 0) {
1038 (void) m_free(m);
1039 goto bad;
1040 }
1041 m->m_next = n;
1042 return (m);
1043bad:
1044 m_freem(n);
1045 mbstat.m_mpfail++; /* XXX: No consistency. */
1046 return (NULL);
1047}
1048
1049/*
1050 * Like m_pullup(), except a new mbuf is always allocated, and we allow
1051 * the amount of empty space before the data in the new mbuf to be specified
1052 * (in the event that the caller expects to prepend later).
1053 */
1054int MSFail;
1055
1056struct mbuf *
1057m_copyup(struct mbuf *n, int len, int dstoff)
1058{
1059 struct mbuf *m;
1060 int count, space;
1061
1062 if (len > (MHLEN - dstoff))
1063 goto bad;
1064 MGET(m, M_DONTWAIT, n->m_type);
1065 if (m == NULL)
1066 goto bad;
1067 m->m_len = 0;
1068 if (n->m_flags & M_PKTHDR)
1069 M_MOVE_PKTHDR(m, n);
1070 m->m_data += dstoff;
1071 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1072 do {
1073 count = min(min(max(len, max_protohdr), space), n->m_len);
1074 memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
1075 (unsigned)count);
1076 len -= count;
1077 m->m_len += count;
1078 n->m_len -= count;
1079 space -= count;
1080 if (n->m_len)
1081 n->m_data += count;
1082 else
1083 n = m_free(n);
1084 } while (len > 0 && n);
1085 if (len > 0) {
1086 (void) m_free(m);
1087 goto bad;
1088 }
1089 m->m_next = n;
1090 return (m);
1091 bad:
1092 m_freem(n);
1093 MSFail++;
1094 return (NULL);
1095}
1096
1097/*
1098 * Partition an mbuf chain in two pieces, returning the tail --
1099 * all but the first len0 bytes. In case of failure, it returns NULL and
1100 * attempts to restore the chain to its original state.
1101 *
1102 * Note that the resulting mbufs might be read-only, because the new
1103 * mbuf can end up sharing an mbuf cluster with the original mbuf if
1104 * the "breaking point" happens to lie within a cluster mbuf. Use the
1105 * M_WRITABLE() macro to check for this case.
1106 */
1107struct mbuf *
1108m_split(struct mbuf *m0, int len0, int wait)
1109{
1110 struct mbuf *m, *n;
1111 u_int len = len0, remain;
1112
1113 MBUF_CHECKSLEEP(wait);
1114 for (m = m0; m && len > m->m_len; m = m->m_next)
1115 len -= m->m_len;
1116 if (m == NULL)
1117 return (NULL);
1118 remain = m->m_len - len;
1119 if (m0->m_flags & M_PKTHDR) {
1120 MGETHDR(n, wait, m0->m_type);
1121 if (n == NULL)
1122 return (NULL);
1123 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
1124 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
1125 m0->m_pkthdr.len = len0;
1126 if (m->m_flags & M_EXT)
1127 goto extpacket;
1128 if (remain > MHLEN) {
1129 /* m can't be the lead packet */
1130 MH_ALIGN(n, 0);
1131 n->m_next = m_split(m, len, wait);
1132 if (n->m_next == NULL) {
1133 (void) m_free(n);
1134 return (NULL);
1135 } else {
1136 n->m_len = 0;
1137 return (n);
1138 }
1139 } else
1140 MH_ALIGN(n, remain);
1141 } else if (remain == 0) {
1142 n = m->m_next;
1143 m->m_next = NULL;
1144 return (n);
1145 } else {
1146 MGET(n, wait, m->m_type);
1147 if (n == NULL)
1148 return (NULL);
1149 M_ALIGN(n, remain);
1150 }
1151extpacket:
1152 if (m->m_flags & M_EXT) {
1153 n->m_data = m->m_data + len;
1154 mb_dupcl(n, m);
1155 } else {
1156 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
1157 }
1158 n->m_len = remain;
1159 m->m_len = len;
1160 n->m_next = m->m_next;
1161 m->m_next = NULL;
1162 return (n);
1163}
1164/*
1165 * Routine to copy from device local memory into mbufs.
1166 * Note that `off' argument is offset into first mbuf of target chain from
1167 * which to begin copying the data to.
1168 */
1169struct mbuf *
1170m_devget(char *buf, int totlen, int off, struct ifnet *ifp,
1171 void (*copy)(char *from, caddr_t to, u_int len))
1172{
1173 struct mbuf *m;
1174 struct mbuf *top = NULL, **mp = ⊤
1175 int len;
1176
1177 if (off < 0 || off > MHLEN)
1178 return (NULL);
1179
1180 while (totlen > 0) {
1181 if (top == NULL) { /* First one, must be PKTHDR */
1182 if (totlen + off >= MINCLSIZE) {
1183 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1184 len = MCLBYTES;
1185 } else {
1186 m = m_gethdr(M_DONTWAIT, MT_DATA);
1187 len = MHLEN;
1188
1189 /* Place initial small packet/header at end of mbuf */
1190 if (m && totlen + off + max_linkhdr <= MLEN) {
1191 m->m_data += max_linkhdr;
1192 len -= max_linkhdr;
1193 }
1194 }
1195 if (m == NULL)
1196 return NULL;
1197 m->m_pkthdr.rcvif = ifp;
1198 m->m_pkthdr.len = totlen;
1199 } else {
1200 if (totlen + off >= MINCLSIZE) {
1201 m = m_getcl(M_DONTWAIT, MT_DATA, 0);
1202 len = MCLBYTES;
1203 } else {
1204 m = m_get(M_DONTWAIT, MT_DATA);
1205 len = MLEN;
1206 }
1207 if (m == NULL) {
1208 m_freem(top);
1209 return NULL;
1210 }
1211 }
1212 if (off) {
1213 m->m_data += off;
1214 len -= off;
1215 off = 0;
1216 }
1217 m->m_len = len = min(totlen, len);
1218 if (copy)
1219 copy(buf, mtod(m, caddr_t), (u_int)len);
1220 else
1221 bcopy(buf, mtod(m, caddr_t), (u_int)len);
1222 buf += len;
1223 *mp = m;
1224 mp = &m->m_next;
1225 totlen -= len;
1226 }
1227 return (top);
1228}
1229
1230/*
1231 * Copy data from a buffer back into the indicated mbuf chain,
1232 * starting "off" bytes from the beginning, extending the mbuf
1233 * chain if necessary.
1234 */
1235void
1236m_copyback(struct mbuf *m0, int off, int len, c_caddr_t cp)
1237{
1238 int mlen;
1239 struct mbuf *m = m0, *n;
1240 int totlen = 0;
1241
1242 if (m0 == NULL)
1243 return;
1244 while (off > (mlen = m->m_len)) {
1245 off -= mlen;
1246 totlen += mlen;
1247 if (m->m_next == NULL) {
1248 n = m_get(M_DONTWAIT, m->m_type);
1249 if (n == NULL)
1250 goto out;
1251 bzero(mtod(n, caddr_t), MLEN);
1252 n->m_len = min(MLEN, len + off);
1253 m->m_next = n;
1254 }
1255 m = m->m_next;
1256 }
1257 while (len > 0) {
1258 mlen = min (m->m_len - off, len);
1259 bcopy(cp, off + mtod(m, caddr_t), (u_int)mlen);
1260 cp += mlen;
1261 len -= mlen;
1262 mlen += off;
1263 off = 0;
1264 totlen += mlen;
1265 if (len == 0)
1266 break;
1267 if (m->m_next == NULL) {
1268 n = m_get(M_DONTWAIT, m->m_type);
1269 if (n == NULL)
1270 break;
1271 n->m_len = min(MLEN, len);
1272 m->m_next = n;
1273 }
1274 m = m->m_next;
1275 }
1276out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
1277 m->m_pkthdr.len = totlen;
1278}
1279
1280/*
1281 * Append the specified data to the indicated mbuf chain,
1282 * Extend the mbuf chain if the new data does not fit in
1283 * existing space.
1284 *
1285 * Return 1 if able to complete the job; otherwise 0.
1286 */
1287int
1288m_append(struct mbuf *m0, int len, c_caddr_t cp)
1289{
1290 struct mbuf *m, *n;
1291 int remainder, space;
1292
1293 for (m = m0; m->m_next != NULL; m = m->m_next)
1294 ;
1295 remainder = len;
1296 space = M_TRAILINGSPACE(m);
1297 if (space > 0) {
1298 /*
1299 * Copy into available space.
1300 */
1301 if (space > remainder)
1302 space = remainder;
1303 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1304 m->m_len += space;
1305 cp += space, remainder -= space;
1306 }
1307 while (remainder > 0) {
1308 /*
1309 * Allocate a new mbuf; could check space
1310 * and allocate a cluster instead.
1311 */
1312 n = m_get(M_DONTWAIT, m->m_type);
1313 if (n == NULL)
1314 break;
1315 n->m_len = min(MLEN, remainder);
1316 bcopy(cp, mtod(n, caddr_t), n->m_len);
1317 cp += n->m_len, remainder -= n->m_len;
1318 m->m_next = n;
1319 m = n;
1320 }
1321 if (m0->m_flags & M_PKTHDR)
1322 m0->m_pkthdr.len += len - remainder;
1323 return (remainder == 0);
1324}
1325
1326/*
1327 * Apply function f to the data in an mbuf chain starting "off" bytes from
1328 * the beginning, continuing for "len" bytes.
1329 */
1330int
1331m_apply(struct mbuf *m, int off, int len,
1332 int (*f)(void *, void *, u_int), void *arg)
1333{
1334 u_int count;
1335 int rval;
1336
1337 KASSERT(off >= 0, ("m_apply, negative off %d", off));
1338 KASSERT(len >= 0, ("m_apply, negative len %d", len));
1339 while (off > 0) {
1340 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1341 if (off < m->m_len)
1342 break;
1343 off -= m->m_len;
1344 m = m->m_next;
1345 }
1346 while (len > 0) {
1347 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1348 count = min(m->m_len - off, len);
1349 rval = (*f)(arg, mtod(m, caddr_t) + off, count);
1350 if (rval)
1351 return (rval);
1352 len -= count;
1353 off = 0;
1354 m = m->m_next;
1355 }
1356 return (0);
1357}
1358
1359/*
1360 * Return a pointer to mbuf/offset of location in mbuf chain.
1361 */
1362struct mbuf *
1363m_getptr(struct mbuf *m, int loc, int *off)
1364{
1365
1366 while (loc >= 0) {
1367 /* Normal end of search. */
1368 if (m->m_len > loc) {
1369 *off = loc;
1370 return (m);
1371 } else {
1372 loc -= m->m_len;
1373 if (m->m_next == NULL) {
1374 if (loc == 0) {
1375 /* Point at the end of valid data. */
1376 *off = m->m_len;
1377 return (m);
1378 }
1379 return (NULL);
1380 }
1381 m = m->m_next;
1382 }
1383 }
1384 return (NULL);
1385}
1386
1387void
1388m_print(const struct mbuf *m, int maxlen)
1389{
1390 int len;
1391 int pdata;
1392 const struct mbuf *m2;
1393
1394 if (m->m_flags & M_PKTHDR)
1395 len = m->m_pkthdr.len;
1396 else
1397 len = -1;
1398 m2 = m;
1399 while (m2 != NULL && (len == -1 || len)) {
1400 pdata = m2->m_len;
1401 if (maxlen != -1 && pdata > maxlen)
1402 pdata = maxlen;
1403 printf("mbuf: %p len: %d, next: %p, %b%s", m2, m2->m_len,
1404 m2->m_next, m2->m_flags, "\20\20freelist\17skipfw"
1405 "\11proto5\10proto4\7proto3\6proto2\5proto1\4rdonly"
1406 "\3eor\2pkthdr\1ext", pdata ? "" : "\n");
1407 if (pdata)
1408 printf(", %*D\n", m2->m_len, (u_char *)m2->m_data, "-");
1409 if (len != -1)
1410 len -= m2->m_len;
1411 m2 = m2->m_next;
1412 }
1413 if (len > 0)
1414 printf("%d bytes unaccounted for.\n", len);
1415 return;
1416}
1417
1418u_int
1419m_fixhdr(struct mbuf *m0)
1420{
1421 u_int len;
1422
1423 len = m_length(m0, NULL);
1424 m0->m_pkthdr.len = len;
1425 return (len);
1426}
1427
1428u_int
1429m_length(struct mbuf *m0, struct mbuf **last)
1430{
1431 struct mbuf *m;
1432 u_int len;
1433
1434 len = 0;
1435 for (m = m0; m != NULL; m = m->m_next) {
1436 len += m->m_len;
1437 if (m->m_next == NULL)
1438 break;
1439 }
1440 if (last != NULL)
1441 *last = m;
1442 return (len);
1443}
1444
1445/*
1446 * Defragment a mbuf chain, returning the shortest possible
1447 * chain of mbufs and clusters. If allocation fails and
1448 * this cannot be completed, NULL will be returned, but
1449 * the passed in chain will be unchanged. Upon success,
1450 * the original chain will be freed, and the new chain
1451 * will be returned.
1452 *
1453 * If a non-packet header is passed in, the original
1454 * mbuf (chain?) will be returned unharmed.
1455 */
1456struct mbuf *
1457m_defrag(struct mbuf *m0, int how)
1458{
1459 struct mbuf *m_new = NULL, *m_final = NULL;
1460 int progress = 0, length;
1461
1462 MBUF_CHECKSLEEP(how);
1463 if (!(m0->m_flags & M_PKTHDR))
1464 return (m0);
1465
1466 m_fixhdr(m0); /* Needed sanity check */
1467
1468#ifdef MBUF_STRESS_TEST
1469 if (m_defragrandomfailures) {
1470 int temp = arc4random() & 0xff;
1471 if (temp == 0xba)
1472 goto nospace;
1473 }
1474#endif
1475
1476 if (m0->m_pkthdr.len > MHLEN)
1477 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1478 else
1479 m_final = m_gethdr(how, MT_DATA);
1480
1481 if (m_final == NULL)
1482 goto nospace;
1483
1484 if (m_dup_pkthdr(m_final, m0, how) == 0)
1485 goto nospace;
1486
1487 m_new = m_final;
1488
1489 while (progress < m0->m_pkthdr.len) {
1490 length = m0->m_pkthdr.len - progress;
1491 if (length > MCLBYTES)
1492 length = MCLBYTES;
1493
1494 if (m_new == NULL) {
1495 if (length > MLEN)
1496 m_new = m_getcl(how, MT_DATA, 0);
1497 else
1498 m_new = m_get(how, MT_DATA);
1499 if (m_new == NULL)
1500 goto nospace;
1501 }
1502
1503 m_copydata(m0, progress, length, mtod(m_new, caddr_t));
1504 progress += length;
1505 m_new->m_len = length;
1506 if (m_new != m_final)
1507 m_cat(m_final, m_new);
1508 m_new = NULL;
1509 }
1510#ifdef MBUF_STRESS_TEST
1511 if (m0->m_next == NULL)
1512 m_defraguseless++;
1513#endif
1514 m_freem(m0);
1515 m0 = m_final;
1516#ifdef MBUF_STRESS_TEST
1517 m_defragpackets++;
1518 m_defragbytes += m0->m_pkthdr.len;
1519#endif
1520 return (m0);
1521nospace:
1522#ifdef MBUF_STRESS_TEST
1523 m_defragfailure++;
1524#endif
1525 if (m_final)
1526 m_freem(m_final);
1527 return (NULL);
1528}
1529
1530#ifdef MBUF_STRESS_TEST
1531
1532/*
1533 * Fragment an mbuf chain. There's no reason you'd ever want to do
1534 * this in normal usage, but it's great for stress testing various
1535 * mbuf consumers.
1536 *
1537 * If fragmentation is not possible, the original chain will be
1538 * returned.
1539 *
1540 * Possible length values:
1541 * 0 no fragmentation will occur
1542 * > 0 each fragment will be of the specified length
1543 * -1 each fragment will be the same random value in length
1544 * -2 each fragment's length will be entirely random
1545 * (Random values range from 1 to 256)
1546 */
1547struct mbuf *
1548m_fragment(struct mbuf *m0, int how, int length)
1549{
1550 struct mbuf *m_new = NULL, *m_final = NULL;
1551 int progress = 0;
1552
1553 if (!(m0->m_flags & M_PKTHDR))
1554 return (m0);
1555
1556 if ((length == 0) || (length < -2))
1557 return (m0);
1558
1559 m_fixhdr(m0); /* Needed sanity check */
1560
1561 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1562
1563 if (m_final == NULL)
1564 goto nospace;
1565
1566 if (m_dup_pkthdr(m_final, m0, how) == 0)
1567 goto nospace;
1568
1569 m_new = m_final;
1570
1571 if (length == -1)
1572 length = 1 + (arc4random() & 255);
1573
1574 while (progress < m0->m_pkthdr.len) {
1575 int fraglen;
1576
1577 if (length > 0)
1578 fraglen = length;
1579 else
1580 fraglen = 1 + (arc4random() & 255);
1581 if (fraglen > m0->m_pkthdr.len - progress)
1582 fraglen = m0->m_pkthdr.len - progress;
1583
1584 if (fraglen > MCLBYTES)
1585 fraglen = MCLBYTES;
1586
1587 if (m_new == NULL) {
1588 m_new = m_getcl(how, MT_DATA, 0);
1589 if (m_new == NULL)
1590 goto nospace;
1591 }
1592
1593 m_copydata(m0, progress, fraglen, mtod(m_new, caddr_t));
1594 progress += fraglen;
1595 m_new->m_len = fraglen;
1596 if (m_new != m_final)
1597 m_cat(m_final, m_new);
1598 m_new = NULL;
1599 }
1600 m_freem(m0);
1601 m0 = m_final;
1602 return (m0);
1603nospace:
1604 if (m_final)
1605 m_freem(m_final);
1606 /* Return the original chain on failure */
1607 return (m0);
1608}
1609
1610#endif
1611
1612struct mbuf *
1613m_uiotombuf(struct uio *uio, int how, int len, int align)
1614{
1615 struct mbuf *m_new = NULL, *m_final = NULL;
1616 int progress = 0, error = 0, length, total;
1617
1618 if (len > 0)
1619 total = min(uio->uio_resid, len);
1620 else
1621 total = uio->uio_resid;
1622 if (align >= MHLEN)
1623 goto nospace;
1624 if (total + align > MHLEN)
1625 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1626 else
1627 m_final = m_gethdr(how, MT_DATA);
1628 if (m_final == NULL)
1629 goto nospace;
1630 m_final->m_data += align;
1631 m_new = m_final;
1632 while (progress < total) {
1633 length = total - progress;
1634 if (length > MCLBYTES)
1635 length = MCLBYTES;
1636 if (m_new == NULL) {
1637 if (length > MLEN)
1638 m_new = m_getcl(how, MT_DATA, 0);
1639 else
1640 m_new = m_get(how, MT_DATA);
1641 if (m_new == NULL)
1642 goto nospace;
1643 }
1644 error = uiomove(mtod(m_new, void *), length, uio);
1645 if (error)
1646 goto nospace;
1647 progress += length;
1648 m_new->m_len = length;
1649 if (m_new != m_final)
1650 m_cat(m_final, m_new);
1651 m_new = NULL;
1652 }
1653 m_fixhdr(m_final);
1654 return (m_final);
1655nospace:
1656 if (m_new)
1657 m_free(m_new);
1658 if (m_final)
1659 m_freem(m_final);
1660 return (NULL);
1661}
1662
1663/*
1664 * Set the m_data pointer of a newly-allocated mbuf
1665 * to place an object of the specified size at the
1666 * end of the mbuf, longword aligned.
1667 */
1668void
1669m_align(struct mbuf *m, int len)
1670{
1671 int adjust;
1672
1673 if (m->m_flags & M_EXT)
1674 adjust = m->m_ext.ext_size - len;
1675 else if (m->m_flags & M_PKTHDR)
1676 adjust = MHLEN - len;
1677 else
1678 adjust = MLEN - len;
1679 m->m_data += adjust &~ (sizeof(long)-1);
1680}
2 * Copyright (c) 1982, 1986, 1988, 1991, 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 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * @(#)uipc_mbuf.c 8.2 (Berkeley) 1/4/94
30 */
31
32#include <sys/cdefs.h>
33__FBSDID("$FreeBSD: head/sys/kern/uipc_mbuf.c 155780 2006-02-17 14:14:15Z andre $");
34
35#include "opt_mac.h"
36#include "opt_param.h"
37#include "opt_mbuf_stress_test.h"
38
39#include <sys/param.h>
40#include <sys/systm.h>
41#include <sys/kernel.h>
42#include <sys/limits.h>
43#include <sys/lock.h>
44#include <sys/mac.h>
45#include <sys/malloc.h>
46#include <sys/mbuf.h>
47#include <sys/sysctl.h>
48#include <sys/domain.h>
49#include <sys/protosw.h>
50#include <sys/uio.h>
51
52int max_linkhdr;
53int max_protohdr;
54int max_hdr;
55int max_datalen;
56#ifdef MBUF_STRESS_TEST
57int m_defragpackets;
58int m_defragbytes;
59int m_defraguseless;
60int m_defragfailure;
61int m_defragrandomfailures;
62#endif
63
64/*
65 * sysctl(8) exported objects
66 */
67SYSCTL_DECL(_kern_ipc);
68SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RW,
69 &max_linkhdr, 0, "");
70SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RW,
71 &max_protohdr, 0, "");
72SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RW, &max_hdr, 0, "");
73SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RW,
74 &max_datalen, 0, "");
75#ifdef MBUF_STRESS_TEST
76SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD,
77 &m_defragpackets, 0, "");
78SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD,
79 &m_defragbytes, 0, "");
80SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD,
81 &m_defraguseless, 0, "");
82SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD,
83 &m_defragfailure, 0, "");
84SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW,
85 &m_defragrandomfailures, 0, "");
86#endif
87
88/*
89 * Allocate a given length worth of mbufs and/or clusters (whatever fits
90 * best) and return a pointer to the top of the allocated chain. If an
91 * existing mbuf chain is provided, then we will append the new chain
92 * to the existing one but still return the top of the newly allocated
93 * chain.
94 */
95struct mbuf *
96m_getm(struct mbuf *m, int len, int how, short type)
97{
98 struct mbuf *mb, *top, *cur, *mtail;
99 int num, rem;
100 int i;
101
102 KASSERT(len >= 0, ("m_getm(): len is < 0"));
103
104 /* If m != NULL, we will append to the end of that chain. */
105 if (m != NULL)
106 for (mtail = m; mtail->m_next != NULL; mtail = mtail->m_next);
107 else
108 mtail = NULL;
109
110 /*
111 * Calculate how many mbufs+clusters ("packets") we need and how much
112 * leftover there is after that and allocate the first mbuf+cluster
113 * if required.
114 */
115 num = len / MCLBYTES;
116 rem = len % MCLBYTES;
117 top = cur = NULL;
118 if (num > 0) {
119 if ((top = cur = m_getcl(how, type, 0)) == NULL)
120 goto failed;
121 top->m_len = 0;
122 }
123 num--;
124
125 for (i = 0; i < num; i++) {
126 mb = m_getcl(how, type, 0);
127 if (mb == NULL)
128 goto failed;
129 mb->m_len = 0;
130 cur = (cur->m_next = mb);
131 }
132 if (rem > 0) {
133 mb = (rem > MINCLSIZE) ?
134 m_getcl(how, type, 0) : m_get(how, type);
135 if (mb == NULL)
136 goto failed;
137 mb->m_len = 0;
138 if (cur == NULL)
139 top = mb;
140 else
141 cur->m_next = mb;
142 }
143
144 if (mtail != NULL)
145 mtail->m_next = top;
146 return top;
147failed:
148 if (top != NULL)
149 m_freem(top);
150 return NULL;
151}
152
153/*
154 * Free an entire chain of mbufs and associated external buffers, if
155 * applicable.
156 */
157void
158m_freem(struct mbuf *mb)
159{
160
161 while (mb != NULL)
162 mb = m_free(mb);
163}
164
165/*-
166 * Configure a provided mbuf to refer to the provided external storage
167 * buffer and setup a reference count for said buffer. If the setting
168 * up of the reference count fails, the M_EXT bit will not be set. If
169 * successfull, the M_EXT bit is set in the mbuf's flags.
170 *
171 * Arguments:
172 * mb The existing mbuf to which to attach the provided buffer.
173 * buf The address of the provided external storage buffer.
174 * size The size of the provided buffer.
175 * freef A pointer to a routine that is responsible for freeing the
176 * provided external storage buffer.
177 * args A pointer to an argument structure (of any type) to be passed
178 * to the provided freef routine (may be NULL).
179 * flags Any other flags to be passed to the provided mbuf.
180 * type The type that the external storage buffer should be
181 * labeled with.
182 *
183 * Returns:
184 * Nothing.
185 */
186void
187m_extadd(struct mbuf *mb, caddr_t buf, u_int size,
188 void (*freef)(void *, void *), void *args, int flags, int type)
189{
190 KASSERT(type != EXT_CLUSTER, ("%s: EXT_CLUSTER not allowed", __func__));
191
192 if (type != EXT_EXTREF)
193 mb->m_ext.ref_cnt = (u_int *)uma_zalloc(zone_ext_refcnt, M_NOWAIT);
194 if (mb->m_ext.ref_cnt != NULL) {
195 *(mb->m_ext.ref_cnt) = 1;
196 mb->m_flags |= (M_EXT | flags);
197 mb->m_ext.ext_buf = buf;
198 mb->m_data = mb->m_ext.ext_buf;
199 mb->m_ext.ext_size = size;
200 mb->m_ext.ext_free = freef;
201 mb->m_ext.ext_args = args;
202 mb->m_ext.ext_type = type;
203 }
204}
205
206/*
207 * Non-directly-exported function to clean up after mbufs with M_EXT
208 * storage attached to them if the reference count hits 1.
209 */
210void
211mb_free_ext(struct mbuf *m)
212{
213 KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__));
214 KASSERT(m->m_ext.ref_cnt != NULL, ("%s: ref_cnt not set", __func__));
215
216 /* Free attached storage if this mbuf is the only reference to it. */
217 if (*(m->m_ext.ref_cnt) == 1 ||
218 atomic_fetchadd_int(m->m_ext.ref_cnt, -1) == 0) {
219 switch (m->m_ext.ext_type) {
220 case EXT_PACKET: /* The packet zone is special. */
221 if (*(m->m_ext.ref_cnt) == 0)
222 *(m->m_ext.ref_cnt) = 1;
223 uma_zfree(zone_pack, m);
224 return; /* Job done. */
225 case EXT_CLUSTER:
226 uma_zfree(zone_clust, m->m_ext.ext_buf);
227 break;
228 case EXT_JUMBOP:
229 uma_zfree(zone_jumbop, m->m_ext.ext_buf);
230 break;
231 case EXT_JUMBO9:
232 uma_zfree(zone_jumbo9, m->m_ext.ext_buf);
233 break;
234 case EXT_JUMBO16:
235 uma_zfree(zone_jumbo16, m->m_ext.ext_buf);
236 break;
237 case EXT_SFBUF:
238 case EXT_NET_DRV:
239 case EXT_MOD_TYPE:
240 case EXT_DISPOSABLE:
241 *(m->m_ext.ref_cnt) = 0;
242 uma_zfree(zone_ext_refcnt, __DEVOLATILE(u_int *,
243 m->m_ext.ref_cnt));
244 /* FALLTHROUGH */
245 case EXT_EXTREF:
246 KASSERT(m->m_ext.ext_free != NULL,
247 ("%s: ext_free not set", __func__));
248 (*(m->m_ext.ext_free))(m->m_ext.ext_buf,
249 m->m_ext.ext_args);
250 break;
251 default:
252 KASSERT(m->m_ext.ext_type == 0,
253 ("%s: unknown ext_type", __func__));
254 }
255 }
256 /*
257 * Free this mbuf back to the mbuf zone with all m_ext
258 * information purged.
259 */
260 m->m_ext.ext_buf = NULL;
261 m->m_ext.ext_free = NULL;
262 m->m_ext.ext_args = NULL;
263 m->m_ext.ref_cnt = NULL;
264 m->m_ext.ext_size = 0;
265 m->m_ext.ext_type = 0;
266 m->m_flags &= ~M_EXT;
267 uma_zfree(zone_mbuf, m);
268}
269
270/*
271 * Attach the the cluster from *m to *n, set up m_ext in *n
272 * and bump the refcount of the cluster.
273 */
274static void
275mb_dupcl(struct mbuf *n, struct mbuf *m)
276{
277 KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__));
278 KASSERT(m->m_ext.ref_cnt != NULL, ("%s: ref_cnt not set", __func__));
279 KASSERT((n->m_flags & M_EXT) == 0, ("%s: M_EXT set", __func__));
280
281 if (*(m->m_ext.ref_cnt) == 1)
282 *(m->m_ext.ref_cnt) += 1;
283 else
284 atomic_add_int(m->m_ext.ref_cnt, 1);
285 n->m_ext.ext_buf = m->m_ext.ext_buf;
286 n->m_ext.ext_free = m->m_ext.ext_free;
287 n->m_ext.ext_args = m->m_ext.ext_args;
288 n->m_ext.ext_size = m->m_ext.ext_size;
289 n->m_ext.ref_cnt = m->m_ext.ref_cnt;
290 n->m_ext.ext_type = m->m_ext.ext_type;
291 n->m_flags |= M_EXT;
292}
293
294/*
295 * Clean up mbuf (chain) from any tags and packet headers.
296 * If "all" is set then the first mbuf in the chain will be
297 * cleaned too.
298 */
299void
300m_demote(struct mbuf *m0, int all)
301{
302 struct mbuf *m;
303
304 for (m = all ? m0 : m0->m_next; m != NULL; m = m->m_next) {
305 if (m->m_flags & M_PKTHDR) {
306 m_tag_delete_chain(m, NULL);
307 m->m_flags &= ~M_PKTHDR;
308 bzero(&m->m_pkthdr, sizeof(struct pkthdr));
309 }
310 if (m->m_type == MT_HEADER)
311 m->m_type = MT_DATA;
312 if (m != m0 && m->m_nextpkt != NULL)
313 m->m_nextpkt = NULL;
314 m->m_flags = m->m_flags & (M_EXT|M_EOR|M_RDONLY|M_FREELIST);
315 }
316}
317
318/*
319 * Sanity checks on mbuf (chain) for use in KASSERT() and general
320 * debugging.
321 * Returns 0 or panics when bad and 1 on all tests passed.
322 * Sanitize, 0 to run M_SANITY_ACTION, 1 to garble things so they
323 * blow up later.
324 */
325int
326m_sanity(struct mbuf *m0, int sanitize)
327{
328 struct mbuf *m;
329 caddr_t a, b;
330 int pktlen = 0;
331
332#define M_SANITY_ACTION(s) return (0)
333/* #define M_SANITY_ACTION(s) panic("mbuf %p: " s, m) */
334
335 for (m = m0; m != NULL; m = m->m_next) {
336 /*
337 * Basic pointer checks. If any of these fails then some
338 * unrelated kernel memory before or after us is trashed.
339 * No way to recover from that.
340 */
341 a = ((m->m_flags & M_EXT) ? m->m_ext.ext_buf :
342 ((m->m_flags & M_PKTHDR) ? (caddr_t)(&m->m_pktdat) :
343 (caddr_t)(&m->m_dat)) );
344 b = (caddr_t)(a + (m->m_flags & M_EXT ? m->m_ext.ext_size :
345 ((m->m_flags & M_PKTHDR) ? MHLEN : MLEN)));
346 if ((caddr_t)m->m_data < a)
347 M_SANITY_ACTION("m_data outside mbuf data range left");
348 if ((caddr_t)m->m_data > b)
349 M_SANITY_ACTION("m_data outside mbuf data range right");
350 if ((caddr_t)m->m_data + m->m_len > b)
351 M_SANITY_ACTION("m_data + m_len exeeds mbuf space");
352 if ((m->m_flags & M_PKTHDR) && m->m_pkthdr.header) {
353 if ((caddr_t)m->m_pkthdr.header < a ||
354 (caddr_t)m->m_pkthdr.header > b)
355 M_SANITY_ACTION("m_pkthdr.header outside mbuf data range");
356 }
357
358 /* m->m_nextpkt may only be set on first mbuf in chain. */
359 if (m != m0 && m->m_nextpkt != NULL) {
360 if (sanitize) {
361 m_freem(m->m_nextpkt);
362 m->m_nextpkt = (struct mbuf *)0xDEADC0DE;
363 } else
364 M_SANITY_ACTION("m->m_nextpkt on in-chain mbuf");
365 }
366
367 /* correct type correlations. */
368 if (m->m_type == MT_HEADER && !(m->m_flags & M_PKTHDR)) {
369 if (sanitize)
370 m->m_type = MT_DATA;
371 else
372 M_SANITY_ACTION("MT_HEADER set but not M_PKTHDR");
373 }
374
375 /* packet length (not mbuf length!) calculation */
376 if (m0->m_flags & M_PKTHDR)
377 pktlen += m->m_len;
378
379 /* m_tags may only be attached to first mbuf in chain. */
380 if (m != m0 && m->m_flags & M_PKTHDR &&
381 !SLIST_EMPTY(&m->m_pkthdr.tags)) {
382 if (sanitize) {
383 m_tag_delete_chain(m, NULL);
384 /* put in 0xDEADC0DE perhaps? */
385 } else
386 M_SANITY_ACTION("m_tags on in-chain mbuf");
387 }
388
389 /* M_PKTHDR may only be set on first mbuf in chain */
390 if (m != m0 && m->m_flags & M_PKTHDR) {
391 if (sanitize) {
392 bzero(&m->m_pkthdr, sizeof(m->m_pkthdr));
393 m->m_flags &= ~M_PKTHDR;
394 /* put in 0xDEADCODE and leave hdr flag in */
395 } else
396 M_SANITY_ACTION("M_PKTHDR on in-chain mbuf");
397 }
398 }
399 m = m0;
400 if (pktlen && pktlen != m->m_pkthdr.len) {
401 if (sanitize)
402 m->m_pkthdr.len = 0;
403 else
404 M_SANITY_ACTION("m_pkthdr.len != mbuf chain length");
405 }
406 return 1;
407
408#undef M_SANITY_ACTION
409}
410
411
412/*
413 * "Move" mbuf pkthdr from "from" to "to".
414 * "from" must have M_PKTHDR set, and "to" must be empty.
415 */
416void
417m_move_pkthdr(struct mbuf *to, struct mbuf *from)
418{
419
420#if 0
421 /* see below for why these are not enabled */
422 M_ASSERTPKTHDR(to);
423 /* Note: with MAC, this may not be a good assertion. */
424 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags),
425 ("m_move_pkthdr: to has tags"));
426#endif
427#ifdef MAC
428 /*
429 * XXXMAC: It could be this should also occur for non-MAC?
430 */
431 if (to->m_flags & M_PKTHDR)
432 m_tag_delete_chain(to, NULL);
433#endif
434 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT);
435 if ((to->m_flags & M_EXT) == 0)
436 to->m_data = to->m_pktdat;
437 to->m_pkthdr = from->m_pkthdr; /* especially tags */
438 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */
439 from->m_flags &= ~M_PKTHDR;
440}
441
442/*
443 * Duplicate "from"'s mbuf pkthdr in "to".
444 * "from" must have M_PKTHDR set, and "to" must be empty.
445 * In particular, this does a deep copy of the packet tags.
446 */
447int
448m_dup_pkthdr(struct mbuf *to, struct mbuf *from, int how)
449{
450
451#if 0
452 /*
453 * The mbuf allocator only initializes the pkthdr
454 * when the mbuf is allocated with MGETHDR. Many users
455 * (e.g. m_copy*, m_prepend) use MGET and then
456 * smash the pkthdr as needed causing these
457 * assertions to trip. For now just disable them.
458 */
459 M_ASSERTPKTHDR(to);
460 /* Note: with MAC, this may not be a good assertion. */
461 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), ("m_dup_pkthdr: to has tags"));
462#endif
463 MBUF_CHECKSLEEP(how);
464#ifdef MAC
465 if (to->m_flags & M_PKTHDR)
466 m_tag_delete_chain(to, NULL);
467#endif
468 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT);
469 if ((to->m_flags & M_EXT) == 0)
470 to->m_data = to->m_pktdat;
471 to->m_pkthdr = from->m_pkthdr;
472 SLIST_INIT(&to->m_pkthdr.tags);
473 return (m_tag_copy_chain(to, from, MBTOM(how)));
474}
475
476/*
477 * Lesser-used path for M_PREPEND:
478 * allocate new mbuf to prepend to chain,
479 * copy junk along.
480 */
481struct mbuf *
482m_prepend(struct mbuf *m, int len, int how)
483{
484 struct mbuf *mn;
485
486 if (m->m_flags & M_PKTHDR)
487 MGETHDR(mn, how, m->m_type);
488 else
489 MGET(mn, how, m->m_type);
490 if (mn == NULL) {
491 m_freem(m);
492 return (NULL);
493 }
494 if (m->m_flags & M_PKTHDR)
495 M_MOVE_PKTHDR(mn, m);
496 mn->m_next = m;
497 m = mn;
498 if (len < MHLEN)
499 MH_ALIGN(m, len);
500 m->m_len = len;
501 return (m);
502}
503
504/*
505 * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
506 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf.
507 * The wait parameter is a choice of M_TRYWAIT/M_DONTWAIT from caller.
508 * Note that the copy is read-only, because clusters are not copied,
509 * only their reference counts are incremented.
510 */
511struct mbuf *
512m_copym(struct mbuf *m, int off0, int len, int wait)
513{
514 struct mbuf *n, **np;
515 int off = off0;
516 struct mbuf *top;
517 int copyhdr = 0;
518
519 KASSERT(off >= 0, ("m_copym, negative off %d", off));
520 KASSERT(len >= 0, ("m_copym, negative len %d", len));
521 MBUF_CHECKSLEEP(wait);
522 if (off == 0 && m->m_flags & M_PKTHDR)
523 copyhdr = 1;
524 while (off > 0) {
525 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain"));
526 if (off < m->m_len)
527 break;
528 off -= m->m_len;
529 m = m->m_next;
530 }
531 np = ⊤
532 top = 0;
533 while (len > 0) {
534 if (m == NULL) {
535 KASSERT(len == M_COPYALL,
536 ("m_copym, length > size of mbuf chain"));
537 break;
538 }
539 if (copyhdr)
540 MGETHDR(n, wait, m->m_type);
541 else
542 MGET(n, wait, m->m_type);
543 *np = n;
544 if (n == NULL)
545 goto nospace;
546 if (copyhdr) {
547 if (!m_dup_pkthdr(n, m, wait))
548 goto nospace;
549 if (len == M_COPYALL)
550 n->m_pkthdr.len -= off0;
551 else
552 n->m_pkthdr.len = len;
553 copyhdr = 0;
554 }
555 n->m_len = min(len, m->m_len - off);
556 if (m->m_flags & M_EXT) {
557 n->m_data = m->m_data + off;
558 mb_dupcl(n, m);
559 } else
560 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
561 (u_int)n->m_len);
562 if (len != M_COPYALL)
563 len -= n->m_len;
564 off = 0;
565 m = m->m_next;
566 np = &n->m_next;
567 }
568 if (top == NULL)
569 mbstat.m_mcfail++; /* XXX: No consistency. */
570
571 return (top);
572nospace:
573 m_freem(top);
574 mbstat.m_mcfail++; /* XXX: No consistency. */
575 return (NULL);
576}
577
578/*
579 * Returns mbuf chain with new head for the prepending case.
580 * Copies from mbuf (chain) n from off for len to mbuf (chain) m
581 * either prepending or appending the data.
582 * The resulting mbuf (chain) m is fully writeable.
583 * m is destination (is made writeable)
584 * n is source, off is offset in source, len is len from offset
585 * dir, 0 append, 1 prepend
586 * how, wait or nowait
587 */
588
589static int
590m_bcopyxxx(void *s, void *t, u_int len)
591{
592 bcopy(s, t, (size_t)len);
593 return 0;
594}
595
596struct mbuf *
597m_copymdata(struct mbuf *m, struct mbuf *n, int off, int len,
598 int prep, int how)
599{
600 struct mbuf *mm, *x, *z, *prev = NULL;
601 caddr_t p;
602 int i, nlen = 0;
603 caddr_t buf[MLEN];
604
605 KASSERT(m != NULL && n != NULL, ("m_copymdata, no target or source"));
606 KASSERT(off >= 0, ("m_copymdata, negative off %d", off));
607 KASSERT(len >= 0, ("m_copymdata, negative len %d", len));
608 KASSERT(prep == 0 || prep == 1, ("m_copymdata, unknown direction %d", prep));
609
610 mm = m;
611 if (!prep) {
612 while(mm->m_next) {
613 prev = mm;
614 mm = mm->m_next;
615 }
616 }
617 for (z = n; z != NULL; z = z->m_next)
618 nlen += z->m_len;
619 if (len == M_COPYALL)
620 len = nlen - off;
621 if (off + len > nlen || len < 1)
622 return NULL;
623
624 if (!M_WRITABLE(mm)) {
625 /* XXX: Use proper m_xxx function instead. */
626 x = m_getcl(how, MT_DATA, mm->m_flags);
627 if (x == NULL)
628 return NULL;
629 bcopy(mm->m_ext.ext_buf, x->m_ext.ext_buf, x->m_ext.ext_size);
630 p = x->m_ext.ext_buf + (mm->m_data - mm->m_ext.ext_buf);
631 x->m_data = p;
632 mm->m_next = NULL;
633 if (mm != m)
634 prev->m_next = x;
635 m_free(mm);
636 mm = x;
637 }
638
639 /*
640 * Append/prepend the data. Allocating mbufs as necessary.
641 */
642 /* Shortcut if enough free space in first/last mbuf. */
643 if (!prep && M_TRAILINGSPACE(mm) >= len) {
644 m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t) +
645 mm->m_len);
646 mm->m_len += len;
647 mm->m_pkthdr.len += len;
648 return m;
649 }
650 if (prep && M_LEADINGSPACE(mm) >= len) {
651 mm->m_data = mtod(mm, caddr_t) - len;
652 m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t));
653 mm->m_len += len;
654 mm->m_pkthdr.len += len;
655 return mm;
656 }
657
658 /* Expand first/last mbuf to cluster if possible. */
659 if (!prep && !(mm->m_flags & M_EXT) && len > M_TRAILINGSPACE(mm)) {
660 bcopy(mm->m_data, &buf, mm->m_len);
661 m_clget(mm, how);
662 if (!(mm->m_flags & M_EXT))
663 return NULL;
664 bcopy(&buf, mm->m_ext.ext_buf, mm->m_len);
665 mm->m_data = mm->m_ext.ext_buf;
666 mm->m_pkthdr.header = NULL;
667 }
668 if (prep && !(mm->m_flags & M_EXT) && len > M_LEADINGSPACE(mm)) {
669 bcopy(mm->m_data, &buf, mm->m_len);
670 m_clget(mm, how);
671 if (!(mm->m_flags & M_EXT))
672 return NULL;
673 bcopy(&buf, (caddr_t *)mm->m_ext.ext_buf +
674 mm->m_ext.ext_size - mm->m_len, mm->m_len);
675 mm->m_data = (caddr_t)mm->m_ext.ext_buf +
676 mm->m_ext.ext_size - mm->m_len;
677 mm->m_pkthdr.header = NULL;
678 }
679
680 /* Append/prepend as many mbuf (clusters) as necessary to fit len. */
681 if (!prep && len > M_TRAILINGSPACE(mm)) {
682 if (!m_getm(mm, len - M_TRAILINGSPACE(mm), how, MT_DATA))
683 return NULL;
684 }
685 if (prep && len > M_LEADINGSPACE(mm)) {
686 if (!(z = m_getm(NULL, len - M_LEADINGSPACE(mm), how, MT_DATA)))
687 return NULL;
688 i = 0;
689 for (x = z; x != NULL; x = x->m_next) {
690 i += x->m_flags & M_EXT ? x->m_ext.ext_size :
691 (x->m_flags & M_PKTHDR ? MHLEN : MLEN);
692 if (!x->m_next)
693 break;
694 }
695 z->m_data += i - len;
696 m_move_pkthdr(mm, z);
697 x->m_next = mm;
698 mm = z;
699 }
700
701 /* Seek to start position in source mbuf. Optimization for long chains. */
702 while (off > 0) {
703 if (off < n->m_len)
704 break;
705 off -= n->m_len;
706 n = n->m_next;
707 }
708
709 /* Copy data into target mbuf. */
710 z = mm;
711 while (len > 0) {
712 KASSERT(z != NULL, ("m_copymdata, falling off target edge"));
713 i = M_TRAILINGSPACE(z);
714 m_apply(n, off, i, m_bcopyxxx, mtod(z, caddr_t) + z->m_len);
715 z->m_len += i;
716 /* fixup pkthdr.len if necessary */
717 if ((prep ? mm : m)->m_flags & M_PKTHDR)
718 (prep ? mm : m)->m_pkthdr.len += i;
719 off += i;
720 len -= i;
721 z = z->m_next;
722 }
723 return (prep ? mm : m);
724}
725
726/*
727 * Copy an entire packet, including header (which must be present).
728 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
729 * Note that the copy is read-only, because clusters are not copied,
730 * only their reference counts are incremented.
731 * Preserve alignment of the first mbuf so if the creator has left
732 * some room at the beginning (e.g. for inserting protocol headers)
733 * the copies still have the room available.
734 */
735struct mbuf *
736m_copypacket(struct mbuf *m, int how)
737{
738 struct mbuf *top, *n, *o;
739
740 MBUF_CHECKSLEEP(how);
741 MGET(n, how, m->m_type);
742 top = n;
743 if (n == NULL)
744 goto nospace;
745
746 if (!m_dup_pkthdr(n, m, how))
747 goto nospace;
748 n->m_len = m->m_len;
749 if (m->m_flags & M_EXT) {
750 n->m_data = m->m_data;
751 mb_dupcl(n, m);
752 } else {
753 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat );
754 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
755 }
756
757 m = m->m_next;
758 while (m) {
759 MGET(o, how, m->m_type);
760 if (o == NULL)
761 goto nospace;
762
763 n->m_next = o;
764 n = n->m_next;
765
766 n->m_len = m->m_len;
767 if (m->m_flags & M_EXT) {
768 n->m_data = m->m_data;
769 mb_dupcl(n, m);
770 } else {
771 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
772 }
773
774 m = m->m_next;
775 }
776 return top;
777nospace:
778 m_freem(top);
779 mbstat.m_mcfail++; /* XXX: No consistency. */
780 return (NULL);
781}
782
783/*
784 * Copy data from an mbuf chain starting "off" bytes from the beginning,
785 * continuing for "len" bytes, into the indicated buffer.
786 */
787void
788m_copydata(const struct mbuf *m, int off, int len, caddr_t cp)
789{
790 u_int count;
791
792 KASSERT(off >= 0, ("m_copydata, negative off %d", off));
793 KASSERT(len >= 0, ("m_copydata, negative len %d", len));
794 while (off > 0) {
795 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain"));
796 if (off < m->m_len)
797 break;
798 off -= m->m_len;
799 m = m->m_next;
800 }
801 while (len > 0) {
802 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain"));
803 count = min(m->m_len - off, len);
804 bcopy(mtod(m, caddr_t) + off, cp, count);
805 len -= count;
806 cp += count;
807 off = 0;
808 m = m->m_next;
809 }
810}
811
812/*
813 * Copy a packet header mbuf chain into a completely new chain, including
814 * copying any mbuf clusters. Use this instead of m_copypacket() when
815 * you need a writable copy of an mbuf chain.
816 */
817struct mbuf *
818m_dup(struct mbuf *m, int how)
819{
820 struct mbuf **p, *top = NULL;
821 int remain, moff, nsize;
822
823 MBUF_CHECKSLEEP(how);
824 /* Sanity check */
825 if (m == NULL)
826 return (NULL);
827 M_ASSERTPKTHDR(m);
828
829 /* While there's more data, get a new mbuf, tack it on, and fill it */
830 remain = m->m_pkthdr.len;
831 moff = 0;
832 p = ⊤
833 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */
834 struct mbuf *n;
835
836 /* Get the next new mbuf */
837 if (remain >= MINCLSIZE) {
838 n = m_getcl(how, m->m_type, 0);
839 nsize = MCLBYTES;
840 } else {
841 n = m_get(how, m->m_type);
842 nsize = MLEN;
843 }
844 if (n == NULL)
845 goto nospace;
846
847 if (top == NULL) { /* First one, must be PKTHDR */
848 if (!m_dup_pkthdr(n, m, how)) {
849 m_free(n);
850 goto nospace;
851 }
852 if ((n->m_flags & M_EXT) == 0)
853 nsize = MHLEN;
854 }
855 n->m_len = 0;
856
857 /* Link it into the new chain */
858 *p = n;
859 p = &n->m_next;
860
861 /* Copy data from original mbuf(s) into new mbuf */
862 while (n->m_len < nsize && m != NULL) {
863 int chunk = min(nsize - n->m_len, m->m_len - moff);
864
865 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
866 moff += chunk;
867 n->m_len += chunk;
868 remain -= chunk;
869 if (moff == m->m_len) {
870 m = m->m_next;
871 moff = 0;
872 }
873 }
874
875 /* Check correct total mbuf length */
876 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL),
877 ("%s: bogus m_pkthdr.len", __func__));
878 }
879 return (top);
880
881nospace:
882 m_freem(top);
883 mbstat.m_mcfail++; /* XXX: No consistency. */
884 return (NULL);
885}
886
887/*
888 * Concatenate mbuf chain n to m.
889 * Both chains must be of the same type (e.g. MT_DATA).
890 * Any m_pkthdr is not updated.
891 */
892void
893m_cat(struct mbuf *m, struct mbuf *n)
894{
895 while (m->m_next)
896 m = m->m_next;
897 while (n) {
898 if (m->m_flags & M_EXT ||
899 m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) {
900 /* just join the two chains */
901 m->m_next = n;
902 return;
903 }
904 /* splat the data from one into the other */
905 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
906 (u_int)n->m_len);
907 m->m_len += n->m_len;
908 n = m_free(n);
909 }
910}
911
912void
913m_adj(struct mbuf *mp, int req_len)
914{
915 int len = req_len;
916 struct mbuf *m;
917 int count;
918
919 if ((m = mp) == NULL)
920 return;
921 if (len >= 0) {
922 /*
923 * Trim from head.
924 */
925 while (m != NULL && len > 0) {
926 if (m->m_len <= len) {
927 len -= m->m_len;
928 m->m_len = 0;
929 m = m->m_next;
930 } else {
931 m->m_len -= len;
932 m->m_data += len;
933 len = 0;
934 }
935 }
936 m = mp;
937 if (mp->m_flags & M_PKTHDR)
938 m->m_pkthdr.len -= (req_len - len);
939 } else {
940 /*
941 * Trim from tail. Scan the mbuf chain,
942 * calculating its length and finding the last mbuf.
943 * If the adjustment only affects this mbuf, then just
944 * adjust and return. Otherwise, rescan and truncate
945 * after the remaining size.
946 */
947 len = -len;
948 count = 0;
949 for (;;) {
950 count += m->m_len;
951 if (m->m_next == (struct mbuf *)0)
952 break;
953 m = m->m_next;
954 }
955 if (m->m_len >= len) {
956 m->m_len -= len;
957 if (mp->m_flags & M_PKTHDR)
958 mp->m_pkthdr.len -= len;
959 return;
960 }
961 count -= len;
962 if (count < 0)
963 count = 0;
964 /*
965 * Correct length for chain is "count".
966 * Find the mbuf with last data, adjust its length,
967 * and toss data from remaining mbufs on chain.
968 */
969 m = mp;
970 if (m->m_flags & M_PKTHDR)
971 m->m_pkthdr.len = count;
972 for (; m; m = m->m_next) {
973 if (m->m_len >= count) {
974 m->m_len = count;
975 if (m->m_next != NULL) {
976 m_freem(m->m_next);
977 m->m_next = NULL;
978 }
979 break;
980 }
981 count -= m->m_len;
982 }
983 }
984}
985
986/*
987 * Rearange an mbuf chain so that len bytes are contiguous
988 * and in the data area of an mbuf (so that mtod and dtom
989 * will work for a structure of size len). Returns the resulting
990 * mbuf chain on success, frees it and returns null on failure.
991 * If there is room, it will add up to max_protohdr-len extra bytes to the
992 * contiguous region in an attempt to avoid being called next time.
993 */
994struct mbuf *
995m_pullup(struct mbuf *n, int len)
996{
997 struct mbuf *m;
998 int count;
999 int space;
1000
1001 /*
1002 * If first mbuf has no cluster, and has room for len bytes
1003 * without shifting current data, pullup into it,
1004 * otherwise allocate a new mbuf to prepend to the chain.
1005 */
1006 if ((n->m_flags & M_EXT) == 0 &&
1007 n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
1008 if (n->m_len >= len)
1009 return (n);
1010 m = n;
1011 n = n->m_next;
1012 len -= m->m_len;
1013 } else {
1014 if (len > MHLEN)
1015 goto bad;
1016 MGET(m, M_DONTWAIT, n->m_type);
1017 if (m == NULL)
1018 goto bad;
1019 m->m_len = 0;
1020 if (n->m_flags & M_PKTHDR)
1021 M_MOVE_PKTHDR(m, n);
1022 }
1023 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1024 do {
1025 count = min(min(max(len, max_protohdr), space), n->m_len);
1026 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
1027 (u_int)count);
1028 len -= count;
1029 m->m_len += count;
1030 n->m_len -= count;
1031 space -= count;
1032 if (n->m_len)
1033 n->m_data += count;
1034 else
1035 n = m_free(n);
1036 } while (len > 0 && n);
1037 if (len > 0) {
1038 (void) m_free(m);
1039 goto bad;
1040 }
1041 m->m_next = n;
1042 return (m);
1043bad:
1044 m_freem(n);
1045 mbstat.m_mpfail++; /* XXX: No consistency. */
1046 return (NULL);
1047}
1048
1049/*
1050 * Like m_pullup(), except a new mbuf is always allocated, and we allow
1051 * the amount of empty space before the data in the new mbuf to be specified
1052 * (in the event that the caller expects to prepend later).
1053 */
1054int MSFail;
1055
1056struct mbuf *
1057m_copyup(struct mbuf *n, int len, int dstoff)
1058{
1059 struct mbuf *m;
1060 int count, space;
1061
1062 if (len > (MHLEN - dstoff))
1063 goto bad;
1064 MGET(m, M_DONTWAIT, n->m_type);
1065 if (m == NULL)
1066 goto bad;
1067 m->m_len = 0;
1068 if (n->m_flags & M_PKTHDR)
1069 M_MOVE_PKTHDR(m, n);
1070 m->m_data += dstoff;
1071 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1072 do {
1073 count = min(min(max(len, max_protohdr), space), n->m_len);
1074 memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
1075 (unsigned)count);
1076 len -= count;
1077 m->m_len += count;
1078 n->m_len -= count;
1079 space -= count;
1080 if (n->m_len)
1081 n->m_data += count;
1082 else
1083 n = m_free(n);
1084 } while (len > 0 && n);
1085 if (len > 0) {
1086 (void) m_free(m);
1087 goto bad;
1088 }
1089 m->m_next = n;
1090 return (m);
1091 bad:
1092 m_freem(n);
1093 MSFail++;
1094 return (NULL);
1095}
1096
1097/*
1098 * Partition an mbuf chain in two pieces, returning the tail --
1099 * all but the first len0 bytes. In case of failure, it returns NULL and
1100 * attempts to restore the chain to its original state.
1101 *
1102 * Note that the resulting mbufs might be read-only, because the new
1103 * mbuf can end up sharing an mbuf cluster with the original mbuf if
1104 * the "breaking point" happens to lie within a cluster mbuf. Use the
1105 * M_WRITABLE() macro to check for this case.
1106 */
1107struct mbuf *
1108m_split(struct mbuf *m0, int len0, int wait)
1109{
1110 struct mbuf *m, *n;
1111 u_int len = len0, remain;
1112
1113 MBUF_CHECKSLEEP(wait);
1114 for (m = m0; m && len > m->m_len; m = m->m_next)
1115 len -= m->m_len;
1116 if (m == NULL)
1117 return (NULL);
1118 remain = m->m_len - len;
1119 if (m0->m_flags & M_PKTHDR) {
1120 MGETHDR(n, wait, m0->m_type);
1121 if (n == NULL)
1122 return (NULL);
1123 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
1124 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
1125 m0->m_pkthdr.len = len0;
1126 if (m->m_flags & M_EXT)
1127 goto extpacket;
1128 if (remain > MHLEN) {
1129 /* m can't be the lead packet */
1130 MH_ALIGN(n, 0);
1131 n->m_next = m_split(m, len, wait);
1132 if (n->m_next == NULL) {
1133 (void) m_free(n);
1134 return (NULL);
1135 } else {
1136 n->m_len = 0;
1137 return (n);
1138 }
1139 } else
1140 MH_ALIGN(n, remain);
1141 } else if (remain == 0) {
1142 n = m->m_next;
1143 m->m_next = NULL;
1144 return (n);
1145 } else {
1146 MGET(n, wait, m->m_type);
1147 if (n == NULL)
1148 return (NULL);
1149 M_ALIGN(n, remain);
1150 }
1151extpacket:
1152 if (m->m_flags & M_EXT) {
1153 n->m_data = m->m_data + len;
1154 mb_dupcl(n, m);
1155 } else {
1156 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
1157 }
1158 n->m_len = remain;
1159 m->m_len = len;
1160 n->m_next = m->m_next;
1161 m->m_next = NULL;
1162 return (n);
1163}
1164/*
1165 * Routine to copy from device local memory into mbufs.
1166 * Note that `off' argument is offset into first mbuf of target chain from
1167 * which to begin copying the data to.
1168 */
1169struct mbuf *
1170m_devget(char *buf, int totlen, int off, struct ifnet *ifp,
1171 void (*copy)(char *from, caddr_t to, u_int len))
1172{
1173 struct mbuf *m;
1174 struct mbuf *top = NULL, **mp = ⊤
1175 int len;
1176
1177 if (off < 0 || off > MHLEN)
1178 return (NULL);
1179
1180 while (totlen > 0) {
1181 if (top == NULL) { /* First one, must be PKTHDR */
1182 if (totlen + off >= MINCLSIZE) {
1183 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1184 len = MCLBYTES;
1185 } else {
1186 m = m_gethdr(M_DONTWAIT, MT_DATA);
1187 len = MHLEN;
1188
1189 /* Place initial small packet/header at end of mbuf */
1190 if (m && totlen + off + max_linkhdr <= MLEN) {
1191 m->m_data += max_linkhdr;
1192 len -= max_linkhdr;
1193 }
1194 }
1195 if (m == NULL)
1196 return NULL;
1197 m->m_pkthdr.rcvif = ifp;
1198 m->m_pkthdr.len = totlen;
1199 } else {
1200 if (totlen + off >= MINCLSIZE) {
1201 m = m_getcl(M_DONTWAIT, MT_DATA, 0);
1202 len = MCLBYTES;
1203 } else {
1204 m = m_get(M_DONTWAIT, MT_DATA);
1205 len = MLEN;
1206 }
1207 if (m == NULL) {
1208 m_freem(top);
1209 return NULL;
1210 }
1211 }
1212 if (off) {
1213 m->m_data += off;
1214 len -= off;
1215 off = 0;
1216 }
1217 m->m_len = len = min(totlen, len);
1218 if (copy)
1219 copy(buf, mtod(m, caddr_t), (u_int)len);
1220 else
1221 bcopy(buf, mtod(m, caddr_t), (u_int)len);
1222 buf += len;
1223 *mp = m;
1224 mp = &m->m_next;
1225 totlen -= len;
1226 }
1227 return (top);
1228}
1229
1230/*
1231 * Copy data from a buffer back into the indicated mbuf chain,
1232 * starting "off" bytes from the beginning, extending the mbuf
1233 * chain if necessary.
1234 */
1235void
1236m_copyback(struct mbuf *m0, int off, int len, c_caddr_t cp)
1237{
1238 int mlen;
1239 struct mbuf *m = m0, *n;
1240 int totlen = 0;
1241
1242 if (m0 == NULL)
1243 return;
1244 while (off > (mlen = m->m_len)) {
1245 off -= mlen;
1246 totlen += mlen;
1247 if (m->m_next == NULL) {
1248 n = m_get(M_DONTWAIT, m->m_type);
1249 if (n == NULL)
1250 goto out;
1251 bzero(mtod(n, caddr_t), MLEN);
1252 n->m_len = min(MLEN, len + off);
1253 m->m_next = n;
1254 }
1255 m = m->m_next;
1256 }
1257 while (len > 0) {
1258 mlen = min (m->m_len - off, len);
1259 bcopy(cp, off + mtod(m, caddr_t), (u_int)mlen);
1260 cp += mlen;
1261 len -= mlen;
1262 mlen += off;
1263 off = 0;
1264 totlen += mlen;
1265 if (len == 0)
1266 break;
1267 if (m->m_next == NULL) {
1268 n = m_get(M_DONTWAIT, m->m_type);
1269 if (n == NULL)
1270 break;
1271 n->m_len = min(MLEN, len);
1272 m->m_next = n;
1273 }
1274 m = m->m_next;
1275 }
1276out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
1277 m->m_pkthdr.len = totlen;
1278}
1279
1280/*
1281 * Append the specified data to the indicated mbuf chain,
1282 * Extend the mbuf chain if the new data does not fit in
1283 * existing space.
1284 *
1285 * Return 1 if able to complete the job; otherwise 0.
1286 */
1287int
1288m_append(struct mbuf *m0, int len, c_caddr_t cp)
1289{
1290 struct mbuf *m, *n;
1291 int remainder, space;
1292
1293 for (m = m0; m->m_next != NULL; m = m->m_next)
1294 ;
1295 remainder = len;
1296 space = M_TRAILINGSPACE(m);
1297 if (space > 0) {
1298 /*
1299 * Copy into available space.
1300 */
1301 if (space > remainder)
1302 space = remainder;
1303 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1304 m->m_len += space;
1305 cp += space, remainder -= space;
1306 }
1307 while (remainder > 0) {
1308 /*
1309 * Allocate a new mbuf; could check space
1310 * and allocate a cluster instead.
1311 */
1312 n = m_get(M_DONTWAIT, m->m_type);
1313 if (n == NULL)
1314 break;
1315 n->m_len = min(MLEN, remainder);
1316 bcopy(cp, mtod(n, caddr_t), n->m_len);
1317 cp += n->m_len, remainder -= n->m_len;
1318 m->m_next = n;
1319 m = n;
1320 }
1321 if (m0->m_flags & M_PKTHDR)
1322 m0->m_pkthdr.len += len - remainder;
1323 return (remainder == 0);
1324}
1325
1326/*
1327 * Apply function f to the data in an mbuf chain starting "off" bytes from
1328 * the beginning, continuing for "len" bytes.
1329 */
1330int
1331m_apply(struct mbuf *m, int off, int len,
1332 int (*f)(void *, void *, u_int), void *arg)
1333{
1334 u_int count;
1335 int rval;
1336
1337 KASSERT(off >= 0, ("m_apply, negative off %d", off));
1338 KASSERT(len >= 0, ("m_apply, negative len %d", len));
1339 while (off > 0) {
1340 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1341 if (off < m->m_len)
1342 break;
1343 off -= m->m_len;
1344 m = m->m_next;
1345 }
1346 while (len > 0) {
1347 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1348 count = min(m->m_len - off, len);
1349 rval = (*f)(arg, mtod(m, caddr_t) + off, count);
1350 if (rval)
1351 return (rval);
1352 len -= count;
1353 off = 0;
1354 m = m->m_next;
1355 }
1356 return (0);
1357}
1358
1359/*
1360 * Return a pointer to mbuf/offset of location in mbuf chain.
1361 */
1362struct mbuf *
1363m_getptr(struct mbuf *m, int loc, int *off)
1364{
1365
1366 while (loc >= 0) {
1367 /* Normal end of search. */
1368 if (m->m_len > loc) {
1369 *off = loc;
1370 return (m);
1371 } else {
1372 loc -= m->m_len;
1373 if (m->m_next == NULL) {
1374 if (loc == 0) {
1375 /* Point at the end of valid data. */
1376 *off = m->m_len;
1377 return (m);
1378 }
1379 return (NULL);
1380 }
1381 m = m->m_next;
1382 }
1383 }
1384 return (NULL);
1385}
1386
1387void
1388m_print(const struct mbuf *m, int maxlen)
1389{
1390 int len;
1391 int pdata;
1392 const struct mbuf *m2;
1393
1394 if (m->m_flags & M_PKTHDR)
1395 len = m->m_pkthdr.len;
1396 else
1397 len = -1;
1398 m2 = m;
1399 while (m2 != NULL && (len == -1 || len)) {
1400 pdata = m2->m_len;
1401 if (maxlen != -1 && pdata > maxlen)
1402 pdata = maxlen;
1403 printf("mbuf: %p len: %d, next: %p, %b%s", m2, m2->m_len,
1404 m2->m_next, m2->m_flags, "\20\20freelist\17skipfw"
1405 "\11proto5\10proto4\7proto3\6proto2\5proto1\4rdonly"
1406 "\3eor\2pkthdr\1ext", pdata ? "" : "\n");
1407 if (pdata)
1408 printf(", %*D\n", m2->m_len, (u_char *)m2->m_data, "-");
1409 if (len != -1)
1410 len -= m2->m_len;
1411 m2 = m2->m_next;
1412 }
1413 if (len > 0)
1414 printf("%d bytes unaccounted for.\n", len);
1415 return;
1416}
1417
1418u_int
1419m_fixhdr(struct mbuf *m0)
1420{
1421 u_int len;
1422
1423 len = m_length(m0, NULL);
1424 m0->m_pkthdr.len = len;
1425 return (len);
1426}
1427
1428u_int
1429m_length(struct mbuf *m0, struct mbuf **last)
1430{
1431 struct mbuf *m;
1432 u_int len;
1433
1434 len = 0;
1435 for (m = m0; m != NULL; m = m->m_next) {
1436 len += m->m_len;
1437 if (m->m_next == NULL)
1438 break;
1439 }
1440 if (last != NULL)
1441 *last = m;
1442 return (len);
1443}
1444
1445/*
1446 * Defragment a mbuf chain, returning the shortest possible
1447 * chain of mbufs and clusters. If allocation fails and
1448 * this cannot be completed, NULL will be returned, but
1449 * the passed in chain will be unchanged. Upon success,
1450 * the original chain will be freed, and the new chain
1451 * will be returned.
1452 *
1453 * If a non-packet header is passed in, the original
1454 * mbuf (chain?) will be returned unharmed.
1455 */
1456struct mbuf *
1457m_defrag(struct mbuf *m0, int how)
1458{
1459 struct mbuf *m_new = NULL, *m_final = NULL;
1460 int progress = 0, length;
1461
1462 MBUF_CHECKSLEEP(how);
1463 if (!(m0->m_flags & M_PKTHDR))
1464 return (m0);
1465
1466 m_fixhdr(m0); /* Needed sanity check */
1467
1468#ifdef MBUF_STRESS_TEST
1469 if (m_defragrandomfailures) {
1470 int temp = arc4random() & 0xff;
1471 if (temp == 0xba)
1472 goto nospace;
1473 }
1474#endif
1475
1476 if (m0->m_pkthdr.len > MHLEN)
1477 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1478 else
1479 m_final = m_gethdr(how, MT_DATA);
1480
1481 if (m_final == NULL)
1482 goto nospace;
1483
1484 if (m_dup_pkthdr(m_final, m0, how) == 0)
1485 goto nospace;
1486
1487 m_new = m_final;
1488
1489 while (progress < m0->m_pkthdr.len) {
1490 length = m0->m_pkthdr.len - progress;
1491 if (length > MCLBYTES)
1492 length = MCLBYTES;
1493
1494 if (m_new == NULL) {
1495 if (length > MLEN)
1496 m_new = m_getcl(how, MT_DATA, 0);
1497 else
1498 m_new = m_get(how, MT_DATA);
1499 if (m_new == NULL)
1500 goto nospace;
1501 }
1502
1503 m_copydata(m0, progress, length, mtod(m_new, caddr_t));
1504 progress += length;
1505 m_new->m_len = length;
1506 if (m_new != m_final)
1507 m_cat(m_final, m_new);
1508 m_new = NULL;
1509 }
1510#ifdef MBUF_STRESS_TEST
1511 if (m0->m_next == NULL)
1512 m_defraguseless++;
1513#endif
1514 m_freem(m0);
1515 m0 = m_final;
1516#ifdef MBUF_STRESS_TEST
1517 m_defragpackets++;
1518 m_defragbytes += m0->m_pkthdr.len;
1519#endif
1520 return (m0);
1521nospace:
1522#ifdef MBUF_STRESS_TEST
1523 m_defragfailure++;
1524#endif
1525 if (m_final)
1526 m_freem(m_final);
1527 return (NULL);
1528}
1529
1530#ifdef MBUF_STRESS_TEST
1531
1532/*
1533 * Fragment an mbuf chain. There's no reason you'd ever want to do
1534 * this in normal usage, but it's great for stress testing various
1535 * mbuf consumers.
1536 *
1537 * If fragmentation is not possible, the original chain will be
1538 * returned.
1539 *
1540 * Possible length values:
1541 * 0 no fragmentation will occur
1542 * > 0 each fragment will be of the specified length
1543 * -1 each fragment will be the same random value in length
1544 * -2 each fragment's length will be entirely random
1545 * (Random values range from 1 to 256)
1546 */
1547struct mbuf *
1548m_fragment(struct mbuf *m0, int how, int length)
1549{
1550 struct mbuf *m_new = NULL, *m_final = NULL;
1551 int progress = 0;
1552
1553 if (!(m0->m_flags & M_PKTHDR))
1554 return (m0);
1555
1556 if ((length == 0) || (length < -2))
1557 return (m0);
1558
1559 m_fixhdr(m0); /* Needed sanity check */
1560
1561 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1562
1563 if (m_final == NULL)
1564 goto nospace;
1565
1566 if (m_dup_pkthdr(m_final, m0, how) == 0)
1567 goto nospace;
1568
1569 m_new = m_final;
1570
1571 if (length == -1)
1572 length = 1 + (arc4random() & 255);
1573
1574 while (progress < m0->m_pkthdr.len) {
1575 int fraglen;
1576
1577 if (length > 0)
1578 fraglen = length;
1579 else
1580 fraglen = 1 + (arc4random() & 255);
1581 if (fraglen > m0->m_pkthdr.len - progress)
1582 fraglen = m0->m_pkthdr.len - progress;
1583
1584 if (fraglen > MCLBYTES)
1585 fraglen = MCLBYTES;
1586
1587 if (m_new == NULL) {
1588 m_new = m_getcl(how, MT_DATA, 0);
1589 if (m_new == NULL)
1590 goto nospace;
1591 }
1592
1593 m_copydata(m0, progress, fraglen, mtod(m_new, caddr_t));
1594 progress += fraglen;
1595 m_new->m_len = fraglen;
1596 if (m_new != m_final)
1597 m_cat(m_final, m_new);
1598 m_new = NULL;
1599 }
1600 m_freem(m0);
1601 m0 = m_final;
1602 return (m0);
1603nospace:
1604 if (m_final)
1605 m_freem(m_final);
1606 /* Return the original chain on failure */
1607 return (m0);
1608}
1609
1610#endif
1611
1612struct mbuf *
1613m_uiotombuf(struct uio *uio, int how, int len, int align)
1614{
1615 struct mbuf *m_new = NULL, *m_final = NULL;
1616 int progress = 0, error = 0, length, total;
1617
1618 if (len > 0)
1619 total = min(uio->uio_resid, len);
1620 else
1621 total = uio->uio_resid;
1622 if (align >= MHLEN)
1623 goto nospace;
1624 if (total + align > MHLEN)
1625 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1626 else
1627 m_final = m_gethdr(how, MT_DATA);
1628 if (m_final == NULL)
1629 goto nospace;
1630 m_final->m_data += align;
1631 m_new = m_final;
1632 while (progress < total) {
1633 length = total - progress;
1634 if (length > MCLBYTES)
1635 length = MCLBYTES;
1636 if (m_new == NULL) {
1637 if (length > MLEN)
1638 m_new = m_getcl(how, MT_DATA, 0);
1639 else
1640 m_new = m_get(how, MT_DATA);
1641 if (m_new == NULL)
1642 goto nospace;
1643 }
1644 error = uiomove(mtod(m_new, void *), length, uio);
1645 if (error)
1646 goto nospace;
1647 progress += length;
1648 m_new->m_len = length;
1649 if (m_new != m_final)
1650 m_cat(m_final, m_new);
1651 m_new = NULL;
1652 }
1653 m_fixhdr(m_final);
1654 return (m_final);
1655nospace:
1656 if (m_new)
1657 m_free(m_new);
1658 if (m_final)
1659 m_freem(m_final);
1660 return (NULL);
1661}
1662
1663/*
1664 * Set the m_data pointer of a newly-allocated mbuf
1665 * to place an object of the specified size at the
1666 * end of the mbuf, longword aligned.
1667 */
1668void
1669m_align(struct mbuf *m, int len)
1670{
1671 int adjust;
1672
1673 if (m->m_flags & M_EXT)
1674 adjust = m->m_ext.ext_size - len;
1675 else if (m->m_flags & M_PKTHDR)
1676 adjust = MHLEN - len;
1677 else
1678 adjust = MLEN - len;
1679 m->m_data += adjust &~ (sizeof(long)-1);
1680}