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 152035 2005-11-04 17:20:53Z 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_CLUSTER: /* 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 break;
226 case EXT_JUMBO9:
227 uma_zfree(zone_jumbo9, m->m_ext.ext_buf);
228 break;
229 case EXT_JUMBO16:
230 uma_zfree(zone_jumbo16, m->m_ext.ext_buf);
231 break;
232 case EXT_SFBUF:
233 case EXT_NET_DRV:
234 case EXT_MOD_TYPE:
235 case EXT_DISPOSABLE:
236 *(m->m_ext.ref_cnt) = 0;
237 uma_zfree(zone_ext_refcnt, __DEVOLATILE(u_int *,
238 m->m_ext.ref_cnt));
239 /* FALLTHROUGH */
240 case EXT_EXTREF:
241 KASSERT(m->m_ext.ext_free != NULL,
242 ("%s: ext_free not set", __func__));
243 (*(m->m_ext.ext_free))(m->m_ext.ext_buf,
244 m->m_ext.ext_args);
245 break;
246 default:
247 KASSERT(m->m_ext.ext_type == 0,
248 ("%s: unknown ext_type", __func__));
249 }
250 }
251 /*
252 * Free this mbuf back to the mbuf zone with all m_ext
253 * information purged.
254 */
255 m->m_ext.ext_buf = NULL;
256 m->m_ext.ext_free = NULL;
257 m->m_ext.ext_args = NULL;
258 m->m_ext.ref_cnt = NULL;
259 m->m_ext.ext_size = 0;
260 m->m_ext.ext_type = 0;
261 m->m_flags &= ~M_EXT;
262 uma_zfree(zone_mbuf, m);
263}
264
265/*
266 * Attach the the cluster from *m to *n, set up m_ext in *n
267 * and bump the refcount of the cluster.
268 */
269static void
270mb_dupcl(struct mbuf *n, struct mbuf *m)
271{
272 KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__));
273 KASSERT(m->m_ext.ref_cnt != NULL, ("%s: ref_cnt not set", __func__));
274 KASSERT((n->m_flags & M_EXT) == 0, ("%s: M_EXT set", __func__));
275
276 if (*(m->m_ext.ref_cnt) == 1)
277 *(m->m_ext.ref_cnt) += 1;
278 else
279 atomic_add_int(m->m_ext.ref_cnt, 1);
280 n->m_ext.ext_buf = m->m_ext.ext_buf;
281 n->m_ext.ext_free = m->m_ext.ext_free;
282 n->m_ext.ext_args = m->m_ext.ext_args;
283 n->m_ext.ext_size = m->m_ext.ext_size;
284 n->m_ext.ref_cnt = m->m_ext.ref_cnt;
285 n->m_ext.ext_type = m->m_ext.ext_type;
286 n->m_flags |= M_EXT;
287}
288
289/*
290 * Clean up mbuf (chain) from any tags and packet headers.
291 * If "all" is set then the first mbuf in the chain will be
292 * cleaned too.
293 */
294void
295m_demote(struct mbuf *m0, int all)
296{
297 struct mbuf *m;
298
299 for (m = all ? m0 : m0->m_next; m != NULL; m = m->m_next) {
300 if (m->m_flags & M_PKTHDR) {
301 m_tag_delete_chain(m, NULL);
302 m->m_flags &= ~M_PKTHDR;
303 bzero(&m->m_pkthdr, sizeof(struct pkthdr));
304 }
305 if (m->m_type == MT_HEADER)
306 m->m_type = MT_DATA;
307 if (m != m0 && m->m_nextpkt != NULL)
308 m->m_nextpkt = NULL;
309 m->m_flags = m->m_flags & (M_EXT|M_EOR|M_RDONLY|M_FREELIST);
310 }
311}
312
313/*
314 * Sanity checks on mbuf (chain) for use in KASSERT() and general
315 * debugging.
316 * Returns 0 or panics when bad and 1 on all tests passed.
317 * Sanitize, 0 to run M_SANITY_ACTION, 1 to garble things so they
318 * blow up later.
319 */
320int
321m_sanity(struct mbuf *m0, int sanitize)
322{
323 struct mbuf *m;
324 caddr_t a, b;
325 int pktlen = 0;
326
327#define M_SANITY_ACTION(s) return (0)
328/* #define M_SANITY_ACTION(s) panic("mbuf %p: " s, m) */
329
330 for (m = m0; m != NULL; m = m->m_next) {
331 /*
332 * Basic pointer checks. If any of these fails then some
333 * unrelated kernel memory before or after us is trashed.
334 * No way to recover from that.
335 */
336 a = ((m->m_flags & M_EXT) ? m->m_ext.ext_buf :
337 ((m->m_flags & M_PKTHDR) ? (caddr_t)(&m->m_pktdat) :
338 (caddr_t)(&m->m_dat)) );
339 b = (caddr_t)(a + (m->m_flags & M_EXT ? m->m_ext.ext_size :
340 ((m->m_flags & M_PKTHDR) ? MHLEN : MLEN)));
341 if ((caddr_t)m->m_data < a)
342 M_SANITY_ACTION("m_data outside mbuf data range left");
343 if ((caddr_t)m->m_data > b)
344 M_SANITY_ACTION("m_data outside mbuf data range right");
345 if ((caddr_t)m->m_data + m->m_len > b)
346 M_SANITY_ACTION("m_data + m_len exeeds mbuf space");
347 if ((m->m_flags & M_PKTHDR) && m->m_pkthdr.header) {
348 if ((caddr_t)m->m_pkthdr.header < a ||
349 (caddr_t)m->m_pkthdr.header > b)
350 M_SANITY_ACTION("m_pkthdr.header outside mbuf data range");
351 }
352
353 /* m->m_nextpkt may only be set on first mbuf in chain. */
354 if (m != m0 && m->m_nextpkt != NULL) {
355 if (sanitize) {
356 m_freem(m->m_nextpkt);
357 m->m_nextpkt = (struct mbuf *)0xDEADC0DE;
358 } else
359 M_SANITY_ACTION("m->m_nextpkt on in-chain mbuf");
360 }
361
362 /* correct type correlations. */
363 if (m->m_type == MT_HEADER && !(m->m_flags & M_PKTHDR)) {
364 if (sanitize)
365 m->m_type = MT_DATA;
366 else
367 M_SANITY_ACTION("MT_HEADER set but not M_PKTHDR");
368 }
369
370 /* packet length (not mbuf length!) calculation */
371 if (m0->m_flags & M_PKTHDR)
372 pktlen += m->m_len;
373
374 /* m_tags may only be attached to first mbuf in chain. */
375 if (m != m0 && m->m_flags & M_PKTHDR &&
376 !SLIST_EMPTY(&m->m_pkthdr.tags)) {
377 if (sanitize) {
378 m_tag_delete_chain(m, NULL);
379 /* put in 0xDEADC0DE perhaps? */
380 } else
381 M_SANITY_ACTION("m_tags on in-chain mbuf");
382 }
383
384 /* M_PKTHDR may only be set on first mbuf in chain */
385 if (m != m0 && m->m_flags & M_PKTHDR) {
386 if (sanitize) {
387 bzero(&m->m_pkthdr, sizeof(m->m_pkthdr));
388 m->m_flags &= ~M_PKTHDR;
389 /* put in 0xDEADCODE and leave hdr flag in */
390 } else
391 M_SANITY_ACTION("M_PKTHDR on in-chain mbuf");
392 }
393 }
394 m = m0;
395 if (pktlen && pktlen != m->m_pkthdr.len) {
396 if (sanitize)
397 m->m_pkthdr.len = 0;
398 else
399 M_SANITY_ACTION("m_pkthdr.len != mbuf chain length");
400 }
401 return 1;
402
403#undef M_SANITY_ACTION
404}
405
406
407/*
408 * "Move" mbuf pkthdr from "from" to "to".
409 * "from" must have M_PKTHDR set, and "to" must be empty.
410 */
411void
412m_move_pkthdr(struct mbuf *to, struct mbuf *from)
413{
414
415#if 0
416 /* see below for why these are not enabled */
417 M_ASSERTPKTHDR(to);
418 /* Note: with MAC, this may not be a good assertion. */
419 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags),
420 ("m_move_pkthdr: to has tags"));
421#endif
422#ifdef MAC
423 /*
424 * XXXMAC: It could be this should also occur for non-MAC?
425 */
426 if (to->m_flags & M_PKTHDR)
427 m_tag_delete_chain(to, NULL);
428#endif
429 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT);
430 if ((to->m_flags & M_EXT) == 0)
431 to->m_data = to->m_pktdat;
432 to->m_pkthdr = from->m_pkthdr; /* especially tags */
433 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */
434 from->m_flags &= ~M_PKTHDR;
435}
436
437/*
438 * Duplicate "from"'s mbuf pkthdr in "to".
439 * "from" must have M_PKTHDR set, and "to" must be empty.
440 * In particular, this does a deep copy of the packet tags.
441 */
442int
443m_dup_pkthdr(struct mbuf *to, struct mbuf *from, int how)
444{
445
446#if 0
447 /*
448 * The mbuf allocator only initializes the pkthdr
449 * when the mbuf is allocated with MGETHDR. Many users
450 * (e.g. m_copy*, m_prepend) use MGET and then
451 * smash the pkthdr as needed causing these
452 * assertions to trip. For now just disable them.
453 */
454 M_ASSERTPKTHDR(to);
455 /* Note: with MAC, this may not be a good assertion. */
456 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), ("m_dup_pkthdr: to has tags"));
457#endif
458 MBUF_CHECKSLEEP(how);
459#ifdef MAC
460 if (to->m_flags & M_PKTHDR)
461 m_tag_delete_chain(to, NULL);
462#endif
463 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT);
464 if ((to->m_flags & M_EXT) == 0)
465 to->m_data = to->m_pktdat;
466 to->m_pkthdr = from->m_pkthdr;
467 SLIST_INIT(&to->m_pkthdr.tags);
468 return (m_tag_copy_chain(to, from, MBTOM(how)));
469}
470
471/*
472 * Lesser-used path for M_PREPEND:
473 * allocate new mbuf to prepend to chain,
474 * copy junk along.
475 */
476struct mbuf *
477m_prepend(struct mbuf *m, int len, int how)
478{
479 struct mbuf *mn;
480
481 if (m->m_flags & M_PKTHDR)
482 MGETHDR(mn, how, m->m_type);
483 else
484 MGET(mn, how, m->m_type);
485 if (mn == NULL) {
486 m_freem(m);
487 return (NULL);
488 }
489 if (m->m_flags & M_PKTHDR)
490 M_MOVE_PKTHDR(mn, m);
491 mn->m_next = m;
492 m = mn;
493 if (len < MHLEN)
494 MH_ALIGN(m, len);
495 m->m_len = len;
496 return (m);
497}
498
499/*
500 * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
501 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf.
502 * The wait parameter is a choice of M_TRYWAIT/M_DONTWAIT from caller.
503 * Note that the copy is read-only, because clusters are not copied,
504 * only their reference counts are incremented.
505 */
506struct mbuf *
507m_copym(struct mbuf *m, int off0, int len, int wait)
508{
509 struct mbuf *n, **np;
510 int off = off0;
511 struct mbuf *top;
512 int copyhdr = 0;
513
514 KASSERT(off >= 0, ("m_copym, negative off %d", off));
515 KASSERT(len >= 0, ("m_copym, negative len %d", len));
516 MBUF_CHECKSLEEP(wait);
517 if (off == 0 && m->m_flags & M_PKTHDR)
518 copyhdr = 1;
519 while (off > 0) {
520 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain"));
521 if (off < m->m_len)
522 break;
523 off -= m->m_len;
524 m = m->m_next;
525 }
526 np = ⊤
527 top = 0;
528 while (len > 0) {
529 if (m == NULL) {
530 KASSERT(len == M_COPYALL,
531 ("m_copym, length > size of mbuf chain"));
532 break;
533 }
534 if (copyhdr)
535 MGETHDR(n, wait, m->m_type);
536 else
537 MGET(n, wait, m->m_type);
538 *np = n;
539 if (n == NULL)
540 goto nospace;
541 if (copyhdr) {
542 if (!m_dup_pkthdr(n, m, wait))
543 goto nospace;
544 if (len == M_COPYALL)
545 n->m_pkthdr.len -= off0;
546 else
547 n->m_pkthdr.len = len;
548 copyhdr = 0;
549 }
550 n->m_len = min(len, m->m_len - off);
551 if (m->m_flags & M_EXT) {
552 n->m_data = m->m_data + off;
553 mb_dupcl(n, m);
554 } else
555 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
556 (u_int)n->m_len);
557 if (len != M_COPYALL)
558 len -= n->m_len;
559 off = 0;
560 m = m->m_next;
561 np = &n->m_next;
562 }
563 if (top == NULL)
564 mbstat.m_mcfail++; /* XXX: No consistency. */
565
566 return (top);
567nospace:
568 m_freem(top);
569 mbstat.m_mcfail++; /* XXX: No consistency. */
570 return (NULL);
571}
572
573/*
574 * Returns mbuf chain with new head for the prepending case.
575 * Copies from mbuf (chain) n from off for len to mbuf (chain) m
576 * either prepending or appending the data.
577 * The resulting mbuf (chain) m is fully writeable.
578 * m is destination (is made writeable)
579 * n is source, off is offset in source, len is len from offset
580 * dir, 0 append, 1 prepend
581 * how, wait or nowait
582 */
583
584static int
585m_bcopyxxx(void *s, void *t, u_int len)
586{
587 bcopy(s, t, (size_t)len);
588 return 0;
589}
590
591struct mbuf *
592m_copymdata(struct mbuf *m, struct mbuf *n, int off, int len,
593 int prep, int how)
594{
595 struct mbuf *mm, *x, *z, *prev = NULL;
596 caddr_t p;
597 int i, nlen = 0;
598 caddr_t buf[MLEN];
599
600 KASSERT(m != NULL && n != NULL, ("m_copymdata, no target or source"));
601 KASSERT(off >= 0, ("m_copymdata, negative off %d", off));
602 KASSERT(len >= 0, ("m_copymdata, negative len %d", len));
603 KASSERT(prep == 0 || prep == 1, ("m_copymdata, unknown direction %d", prep));
604
605 mm = m;
606 if (!prep) {
607 while(mm->m_next) {
608 prev = mm;
609 mm = mm->m_next;
610 }
611 }
612 for (z = n; z != NULL; z = z->m_next)
613 nlen += z->m_len;
614 if (len == M_COPYALL)
615 len = nlen - off;
616 if (off + len > nlen || len < 1)
617 return NULL;
618
619 if (!M_WRITABLE(mm)) {
620 /* XXX: Use proper m_xxx function instead. */
621 x = m_getcl(how, MT_DATA, mm->m_flags);
622 if (x == NULL)
623 return NULL;
624 bcopy(mm->m_ext.ext_buf, x->m_ext.ext_buf, x->m_ext.ext_size);
625 p = x->m_ext.ext_buf + (mm->m_data - mm->m_ext.ext_buf);
626 x->m_data = p;
627 mm->m_next = NULL;
628 if (mm != m)
629 prev->m_next = x;
630 m_free(mm);
631 mm = x;
632 }
633
634 /*
635 * Append/prepend the data. Allocating mbufs as necessary.
636 */
637 /* Shortcut if enough free space in first/last mbuf. */
638 if (!prep && M_TRAILINGSPACE(mm) >= len) {
639 m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t) +
640 mm->m_len);
641 mm->m_len += len;
642 mm->m_pkthdr.len += len;
643 return m;
644 }
645 if (prep && M_LEADINGSPACE(mm) >= len) {
646 mm->m_data = mtod(mm, caddr_t) - len;
647 m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t));
648 mm->m_len += len;
649 mm->m_pkthdr.len += len;
650 return mm;
651 }
652
653 /* Expand first/last mbuf to cluster if possible. */
654 if (!prep && !(mm->m_flags & M_EXT) && len > M_TRAILINGSPACE(mm)) {
655 bcopy(mm->m_data, &buf, mm->m_len);
656 m_clget(mm, how);
657 if (!(mm->m_flags & M_EXT))
658 return NULL;
659 bcopy(&buf, mm->m_ext.ext_buf, mm->m_len);
660 mm->m_data = mm->m_ext.ext_buf;
661 mm->m_pkthdr.header = NULL;
662 }
663 if (prep && !(mm->m_flags & M_EXT) && len > M_LEADINGSPACE(mm)) {
664 bcopy(mm->m_data, &buf, mm->m_len);
665 m_clget(mm, how);
666 if (!(mm->m_flags & M_EXT))
667 return NULL;
668 bcopy(&buf, (caddr_t *)mm->m_ext.ext_buf +
669 mm->m_ext.ext_size - mm->m_len, mm->m_len);
670 mm->m_data = (caddr_t)mm->m_ext.ext_buf +
671 mm->m_ext.ext_size - mm->m_len;
672 mm->m_pkthdr.header = NULL;
673 }
674
675 /* Append/prepend as many mbuf (clusters) as necessary to fit len. */
676 if (!prep && len > M_TRAILINGSPACE(mm)) {
677 if (!m_getm(mm, len - M_TRAILINGSPACE(mm), how, MT_DATA))
678 return NULL;
679 }
680 if (prep && len > M_LEADINGSPACE(mm)) {
681 if (!(z = m_getm(NULL, len - M_LEADINGSPACE(mm), how, MT_DATA)))
682 return NULL;
683 i = 0;
684 for (x = z; x != NULL; x = x->m_next) {
685 i += x->m_flags & M_EXT ? x->m_ext.ext_size :
686 (x->m_flags & M_PKTHDR ? MHLEN : MLEN);
687 if (!x->m_next)
688 break;
689 }
690 z->m_data += i - len;
691 m_move_pkthdr(mm, z);
692 x->m_next = mm;
693 mm = z;
694 }
695
696 /* Seek to start position in source mbuf. Optimization for long chains. */
697 while (off > 0) {
698 if (off < n->m_len)
699 break;
700 off -= n->m_len;
701 n = n->m_next;
702 }
703
704 /* Copy data into target mbuf. */
705 z = mm;
706 while (len > 0) {
707 KASSERT(z != NULL, ("m_copymdata, falling off target edge"));
708 i = M_TRAILINGSPACE(z);
709 m_apply(n, off, i, m_bcopyxxx, mtod(z, caddr_t) + z->m_len);
710 z->m_len += i;
711 /* fixup pkthdr.len if necessary */
712 if ((prep ? mm : m)->m_flags & M_PKTHDR)
713 (prep ? mm : m)->m_pkthdr.len += i;
714 off += i;
715 len -= i;
716 z = z->m_next;
717 }
718 return (prep ? mm : m);
719}
720
721/*
722 * Copy an entire packet, including header (which must be present).
723 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
724 * Note that the copy is read-only, because clusters are not copied,
725 * only their reference counts are incremented.
726 * Preserve alignment of the first mbuf so if the creator has left
727 * some room at the beginning (e.g. for inserting protocol headers)
728 * the copies still have the room available.
729 */
730struct mbuf *
731m_copypacket(struct mbuf *m, int how)
732{
733 struct mbuf *top, *n, *o;
734
735 MBUF_CHECKSLEEP(how);
736 MGET(n, how, m->m_type);
737 top = n;
738 if (n == NULL)
739 goto nospace;
740
741 if (!m_dup_pkthdr(n, m, how))
742 goto nospace;
743 n->m_len = m->m_len;
744 if (m->m_flags & M_EXT) {
745 n->m_data = m->m_data;
746 mb_dupcl(n, m);
747 } else {
748 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat );
749 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
750 }
751
752 m = m->m_next;
753 while (m) {
754 MGET(o, how, m->m_type);
755 if (o == NULL)
756 goto nospace;
757
758 n->m_next = o;
759 n = n->m_next;
760
761 n->m_len = m->m_len;
762 if (m->m_flags & M_EXT) {
763 n->m_data = m->m_data;
764 mb_dupcl(n, m);
765 } else {
766 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
767 }
768
769 m = m->m_next;
770 }
771 return top;
772nospace:
773 m_freem(top);
774 mbstat.m_mcfail++; /* XXX: No consistency. */
775 return (NULL);
776}
777
778/*
779 * Copy data from an mbuf chain starting "off" bytes from the beginning,
780 * continuing for "len" bytes, into the indicated buffer.
781 */
782void
783m_copydata(const struct mbuf *m, int off, int len, caddr_t cp)
784{
785 u_int count;
786
787 KASSERT(off >= 0, ("m_copydata, negative off %d", off));
788 KASSERT(len >= 0, ("m_copydata, negative len %d", len));
789 while (off > 0) {
790 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain"));
791 if (off < m->m_len)
792 break;
793 off -= m->m_len;
794 m = m->m_next;
795 }
796 while (len > 0) {
797 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain"));
798 count = min(m->m_len - off, len);
799 bcopy(mtod(m, caddr_t) + off, cp, count);
800 len -= count;
801 cp += count;
802 off = 0;
803 m = m->m_next;
804 }
805}
806
807/*
808 * Copy a packet header mbuf chain into a completely new chain, including
809 * copying any mbuf clusters. Use this instead of m_copypacket() when
810 * you need a writable copy of an mbuf chain.
811 */
812struct mbuf *
813m_dup(struct mbuf *m, int how)
814{
815 struct mbuf **p, *top = NULL;
816 int remain, moff, nsize;
817
818 MBUF_CHECKSLEEP(how);
819 /* Sanity check */
820 if (m == NULL)
821 return (NULL);
822 M_ASSERTPKTHDR(m);
823
824 /* While there's more data, get a new mbuf, tack it on, and fill it */
825 remain = m->m_pkthdr.len;
826 moff = 0;
827 p = ⊤
828 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */
829 struct mbuf *n;
830
831 /* Get the next new mbuf */
832 if (remain >= MINCLSIZE) {
833 n = m_getcl(how, m->m_type, 0);
834 nsize = MCLBYTES;
835 } else {
836 n = m_get(how, m->m_type);
837 nsize = MLEN;
838 }
839 if (n == NULL)
840 goto nospace;
841
842 if (top == NULL) { /* First one, must be PKTHDR */
843 if (!m_dup_pkthdr(n, m, how)) {
844 m_free(n);
845 goto nospace;
846 }
847 nsize = MHLEN;
848 }
849 n->m_len = 0;
850
851 /* Link it into the new chain */
852 *p = n;
853 p = &n->m_next;
854
855 /* Copy data from original mbuf(s) into new mbuf */
856 while (n->m_len < nsize && m != NULL) {
857 int chunk = min(nsize - n->m_len, m->m_len - moff);
858
859 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
860 moff += chunk;
861 n->m_len += chunk;
862 remain -= chunk;
863 if (moff == m->m_len) {
864 m = m->m_next;
865 moff = 0;
866 }
867 }
868
869 /* Check correct total mbuf length */
870 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL),
871 ("%s: bogus m_pkthdr.len", __func__));
872 }
873 return (top);
874
875nospace:
876 m_freem(top);
877 mbstat.m_mcfail++; /* XXX: No consistency. */
878 return (NULL);
879}
880
881/*
882 * Concatenate mbuf chain n to m.
883 * Both chains must be of the same type (e.g. MT_DATA).
884 * Any m_pkthdr is not updated.
885 */
886void
887m_cat(struct mbuf *m, struct mbuf *n)
888{
889 while (m->m_next)
890 m = m->m_next;
891 while (n) {
892 if (m->m_flags & M_EXT ||
893 m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) {
894 /* just join the two chains */
895 m->m_next = n;
896 return;
897 }
898 /* splat the data from one into the other */
899 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
900 (u_int)n->m_len);
901 m->m_len += n->m_len;
902 n = m_free(n);
903 }
904}
905
906void
907m_adj(struct mbuf *mp, int req_len)
908{
909 int len = req_len;
910 struct mbuf *m;
911 int count;
912
913 if ((m = mp) == NULL)
914 return;
915 if (len >= 0) {
916 /*
917 * Trim from head.
918 */
919 while (m != NULL && len > 0) {
920 if (m->m_len <= len) {
921 len -= m->m_len;
922 m->m_len = 0;
923 m = m->m_next;
924 } else {
925 m->m_len -= len;
926 m->m_data += len;
927 len = 0;
928 }
929 }
930 m = mp;
931 if (mp->m_flags & M_PKTHDR)
932 m->m_pkthdr.len -= (req_len - len);
933 } else {
934 /*
935 * Trim from tail. Scan the mbuf chain,
936 * calculating its length and finding the last mbuf.
937 * If the adjustment only affects this mbuf, then just
938 * adjust and return. Otherwise, rescan and truncate
939 * after the remaining size.
940 */
941 len = -len;
942 count = 0;
943 for (;;) {
944 count += m->m_len;
945 if (m->m_next == (struct mbuf *)0)
946 break;
947 m = m->m_next;
948 }
949 if (m->m_len >= len) {
950 m->m_len -= len;
951 if (mp->m_flags & M_PKTHDR)
952 mp->m_pkthdr.len -= len;
953 return;
954 }
955 count -= len;
956 if (count < 0)
957 count = 0;
958 /*
959 * Correct length for chain is "count".
960 * Find the mbuf with last data, adjust its length,
961 * and toss data from remaining mbufs on chain.
962 */
963 m = mp;
964 if (m->m_flags & M_PKTHDR)
965 m->m_pkthdr.len = count;
966 for (; m; m = m->m_next) {
967 if (m->m_len >= count) {
968 m->m_len = count;
969 if (m->m_next != NULL) {
970 m_freem(m->m_next);
971 m->m_next = NULL;
972 }
973 break;
974 }
975 count -= m->m_len;
976 }
977 }
978}
979
980/*
981 * Rearange an mbuf chain so that len bytes are contiguous
982 * and in the data area of an mbuf (so that mtod and dtom
983 * will work for a structure of size len). Returns the resulting
984 * mbuf chain on success, frees it and returns null on failure.
985 * If there is room, it will add up to max_protohdr-len extra bytes to the
986 * contiguous region in an attempt to avoid being called next time.
987 */
988struct mbuf *
989m_pullup(struct mbuf *n, int len)
990{
991 struct mbuf *m;
992 int count;
993 int space;
994
995 /*
996 * If first mbuf has no cluster, and has room for len bytes
997 * without shifting current data, pullup into it,
998 * otherwise allocate a new mbuf to prepend to the chain.
999 */
1000 if ((n->m_flags & M_EXT) == 0 &&
1001 n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
1002 if (n->m_len >= len)
1003 return (n);
1004 m = n;
1005 n = n->m_next;
1006 len -= m->m_len;
1007 } else {
1008 if (len > MHLEN)
1009 goto bad;
1010 MGET(m, M_DONTWAIT, n->m_type);
1011 if (m == NULL)
1012 goto bad;
1013 m->m_len = 0;
1014 if (n->m_flags & M_PKTHDR)
1015 M_MOVE_PKTHDR(m, n);
1016 }
1017 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1018 do {
1019 count = min(min(max(len, max_protohdr), space), n->m_len);
1020 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
1021 (u_int)count);
1022 len -= count;
1023 m->m_len += count;
1024 n->m_len -= count;
1025 space -= count;
1026 if (n->m_len)
1027 n->m_data += count;
1028 else
1029 n = m_free(n);
1030 } while (len > 0 && n);
1031 if (len > 0) {
1032 (void) m_free(m);
1033 goto bad;
1034 }
1035 m->m_next = n;
1036 return (m);
1037bad:
1038 m_freem(n);
1039 mbstat.m_mpfail++; /* XXX: No consistency. */
1040 return (NULL);
1041}
1042
1043/*
1044 * Like m_pullup(), except a new mbuf is always allocated, and we allow
1045 * the amount of empty space before the data in the new mbuf to be specified
1046 * (in the event that the caller expects to prepend later).
1047 */
1048int MSFail;
1049
1050struct mbuf *
1051m_copyup(struct mbuf *n, int len, int dstoff)
1052{
1053 struct mbuf *m;
1054 int count, space;
1055
1056 if (len > (MHLEN - dstoff))
1057 goto bad;
1058 MGET(m, M_DONTWAIT, n->m_type);
1059 if (m == NULL)
1060 goto bad;
1061 m->m_len = 0;
1062 if (n->m_flags & M_PKTHDR)
1063 M_MOVE_PKTHDR(m, n);
1064 m->m_data += dstoff;
1065 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1066 do {
1067 count = min(min(max(len, max_protohdr), space), n->m_len);
1068 memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
1069 (unsigned)count);
1070 len -= count;
1071 m->m_len += count;
1072 n->m_len -= count;
1073 space -= count;
1074 if (n->m_len)
1075 n->m_data += count;
1076 else
1077 n = m_free(n);
1078 } while (len > 0 && n);
1079 if (len > 0) {
1080 (void) m_free(m);
1081 goto bad;
1082 }
1083 m->m_next = n;
1084 return (m);
1085 bad:
1086 m_freem(n);
1087 MSFail++;
1088 return (NULL);
1089}
1090
1091/*
1092 * Partition an mbuf chain in two pieces, returning the tail --
1093 * all but the first len0 bytes. In case of failure, it returns NULL and
1094 * attempts to restore the chain to its original state.
1095 *
1096 * Note that the resulting mbufs might be read-only, because the new
1097 * mbuf can end up sharing an mbuf cluster with the original mbuf if
1098 * the "breaking point" happens to lie within a cluster mbuf. Use the
1099 * M_WRITABLE() macro to check for this case.
1100 */
1101struct mbuf *
1102m_split(struct mbuf *m0, int len0, int wait)
1103{
1104 struct mbuf *m, *n;
1105 u_int len = len0, remain;
1106
1107 MBUF_CHECKSLEEP(wait);
1108 for (m = m0; m && len > m->m_len; m = m->m_next)
1109 len -= m->m_len;
1110 if (m == NULL)
1111 return (NULL);
1112 remain = m->m_len - len;
1113 if (m0->m_flags & M_PKTHDR) {
1114 MGETHDR(n, wait, m0->m_type);
1115 if (n == NULL)
1116 return (NULL);
1117 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
1118 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
1119 m0->m_pkthdr.len = len0;
1120 if (m->m_flags & M_EXT)
1121 goto extpacket;
1122 if (remain > MHLEN) {
1123 /* m can't be the lead packet */
1124 MH_ALIGN(n, 0);
1125 n->m_next = m_split(m, len, wait);
1126 if (n->m_next == NULL) {
1127 (void) m_free(n);
1128 return (NULL);
1129 } else {
1130 n->m_len = 0;
1131 return (n);
1132 }
1133 } else
1134 MH_ALIGN(n, remain);
1135 } else if (remain == 0) {
1136 n = m->m_next;
1137 m->m_next = NULL;
1138 return (n);
1139 } else {
1140 MGET(n, wait, m->m_type);
1141 if (n == NULL)
1142 return (NULL);
1143 M_ALIGN(n, remain);
1144 }
1145extpacket:
1146 if (m->m_flags & M_EXT) {
1147 n->m_data = m->m_data + len;
1148 mb_dupcl(n, m);
1149 } else {
1150 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
1151 }
1152 n->m_len = remain;
1153 m->m_len = len;
1154 n->m_next = m->m_next;
1155 m->m_next = NULL;
1156 return (n);
1157}
1158/*
1159 * Routine to copy from device local memory into mbufs.
1160 * Note that `off' argument is offset into first mbuf of target chain from
1161 * which to begin copying the data to.
1162 */
1163struct mbuf *
1164m_devget(char *buf, int totlen, int off, struct ifnet *ifp,
1165 void (*copy)(char *from, caddr_t to, u_int len))
1166{
1167 struct mbuf *m;
1168 struct mbuf *top = NULL, **mp = ⊤
1169 int len;
1170
1171 if (off < 0 || off > MHLEN)
1172 return (NULL);
1173
1174 while (totlen > 0) {
1175 if (top == NULL) { /* First one, must be PKTHDR */
1176 if (totlen + off >= MINCLSIZE) {
1177 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1178 len = MCLBYTES;
1179 } else {
1180 m = m_gethdr(M_DONTWAIT, MT_DATA);
1181 len = MHLEN;
1182
1183 /* Place initial small packet/header at end of mbuf */
1184 if (m && totlen + off + max_linkhdr <= MLEN) {
1185 m->m_data += max_linkhdr;
1186 len -= max_linkhdr;
1187 }
1188 }
1189 if (m == NULL)
1190 return NULL;
1191 m->m_pkthdr.rcvif = ifp;
1192 m->m_pkthdr.len = totlen;
1193 } else {
1194 if (totlen + off >= MINCLSIZE) {
1195 m = m_getcl(M_DONTWAIT, MT_DATA, 0);
1196 len = MCLBYTES;
1197 } else {
1198 m = m_get(M_DONTWAIT, MT_DATA);
1199 len = MLEN;
1200 }
1201 if (m == NULL) {
1202 m_freem(top);
1203 return NULL;
1204 }
1205 }
1206 if (off) {
1207 m->m_data += off;
1208 len -= off;
1209 off = 0;
1210 }
1211 m->m_len = len = min(totlen, len);
1212 if (copy)
1213 copy(buf, mtod(m, caddr_t), (u_int)len);
1214 else
1215 bcopy(buf, mtod(m, caddr_t), (u_int)len);
1216 buf += len;
1217 *mp = m;
1218 mp = &m->m_next;
1219 totlen -= len;
1220 }
1221 return (top);
1222}
1223
1224/*
1225 * Copy data from a buffer back into the indicated mbuf chain,
1226 * starting "off" bytes from the beginning, extending the mbuf
1227 * chain if necessary.
1228 */
1229void
1230m_copyback(struct mbuf *m0, int off, int len, c_caddr_t cp)
1231{
1232 int mlen;
1233 struct mbuf *m = m0, *n;
1234 int totlen = 0;
1235
1236 if (m0 == NULL)
1237 return;
1238 while (off > (mlen = m->m_len)) {
1239 off -= mlen;
1240 totlen += mlen;
1241 if (m->m_next == NULL) {
1242 n = m_get(M_DONTWAIT, m->m_type);
1243 if (n == NULL)
1244 goto out;
1245 bzero(mtod(n, caddr_t), MLEN);
1246 n->m_len = min(MLEN, len + off);
1247 m->m_next = n;
1248 }
1249 m = m->m_next;
1250 }
1251 while (len > 0) {
1252 mlen = min (m->m_len - off, len);
1253 bcopy(cp, off + mtod(m, caddr_t), (u_int)mlen);
1254 cp += mlen;
1255 len -= mlen;
1256 mlen += off;
1257 off = 0;
1258 totlen += mlen;
1259 if (len == 0)
1260 break;
1261 if (m->m_next == NULL) {
1262 n = m_get(M_DONTWAIT, m->m_type);
1263 if (n == NULL)
1264 break;
1265 n->m_len = min(MLEN, len);
1266 m->m_next = n;
1267 }
1268 m = m->m_next;
1269 }
1270out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
1271 m->m_pkthdr.len = totlen;
1272}
1273
1274/*
1275 * Append the specified data to the indicated mbuf chain,
1276 * Extend the mbuf chain if the new data does not fit in
1277 * existing space.
1278 *
1279 * Return 1 if able to complete the job; otherwise 0.
1280 */
1281int
1282m_append(struct mbuf *m0, int len, c_caddr_t cp)
1283{
1284 struct mbuf *m, *n;
1285 int remainder, space;
1286
1287 for (m = m0; m->m_next != NULL; m = m->m_next)
1288 ;
1289 remainder = len;
1290 space = M_TRAILINGSPACE(m);
1291 if (space > 0) {
1292 /*
1293 * Copy into available space.
1294 */
1295 if (space > remainder)
1296 space = remainder;
1297 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1298 m->m_len += space;
1299 cp += space, remainder -= space;
1300 }
1301 while (remainder > 0) {
1302 /*
1303 * Allocate a new mbuf; could check space
1304 * and allocate a cluster instead.
1305 */
1306 n = m_get(M_DONTWAIT, m->m_type);
1307 if (n == NULL)
1308 break;
1309 n->m_len = min(MLEN, remainder);
1310 bcopy(cp, mtod(n, caddr_t), n->m_len);
1311 cp += n->m_len, remainder -= n->m_len;
1312 m->m_next = n;
1313 m = n;
1314 }
1315 if (m0->m_flags & M_PKTHDR)
1316 m0->m_pkthdr.len += len - remainder;
1317 return (remainder == 0);
1318}
1319
1320/*
1321 * Apply function f to the data in an mbuf chain starting "off" bytes from
1322 * the beginning, continuing for "len" bytes.
1323 */
1324int
1325m_apply(struct mbuf *m, int off, int len,
1326 int (*f)(void *, void *, u_int), void *arg)
1327{
1328 u_int count;
1329 int rval;
1330
1331 KASSERT(off >= 0, ("m_apply, negative off %d", off));
1332 KASSERT(len >= 0, ("m_apply, negative len %d", len));
1333 while (off > 0) {
1334 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1335 if (off < m->m_len)
1336 break;
1337 off -= m->m_len;
1338 m = m->m_next;
1339 }
1340 while (len > 0) {
1341 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1342 count = min(m->m_len - off, len);
1343 rval = (*f)(arg, mtod(m, caddr_t) + off, count);
1344 if (rval)
1345 return (rval);
1346 len -= count;
1347 off = 0;
1348 m = m->m_next;
1349 }
1350 return (0);
1351}
1352
1353/*
1354 * Return a pointer to mbuf/offset of location in mbuf chain.
1355 */
1356struct mbuf *
1357m_getptr(struct mbuf *m, int loc, int *off)
1358{
1359
1360 while (loc >= 0) {
1361 /* Normal end of search. */
1362 if (m->m_len > loc) {
1363 *off = loc;
1364 return (m);
1365 } else {
1366 loc -= m->m_len;
1367 if (m->m_next == NULL) {
1368 if (loc == 0) {
1369 /* Point at the end of valid data. */
1370 *off = m->m_len;
1371 return (m);
1372 }
1373 return (NULL);
1374 }
1375 m = m->m_next;
1376 }
1377 }
1378 return (NULL);
1379}
1380
1381void
1382m_print(const struct mbuf *m, int maxlen)
1383{
1384 int len;
1385 int pdata;
1386 const struct mbuf *m2;
1387
1388 if (m->m_flags & M_PKTHDR)
1389 len = m->m_pkthdr.len;
1390 else
1391 len = -1;
1392 m2 = m;
1393 while (m2 != NULL && (len == -1 || len)) {
1394 pdata = m2->m_len;
1395 if (maxlen != -1 && pdata > maxlen)
1396 pdata = maxlen;
1397 printf("mbuf: %p len: %d, next: %p, %b%s", m2, m2->m_len,
1398 m2->m_next, m2->m_flags, "\20\20freelist\17skipfw"
1399 "\11proto5\10proto4\7proto3\6proto2\5proto1\4rdonly"
1400 "\3eor\2pkthdr\1ext", pdata ? "" : "\n");
1401 if (pdata)
1402 printf(", %*D\n", m2->m_len, (u_char *)m2->m_data, "-");
1403 if (len != -1)
1404 len -= m2->m_len;
1405 m2 = m2->m_next;
1406 }
1407 if (len > 0)
1408 printf("%d bytes unaccounted for.\n", len);
1409 return;
1410}
1411
1412u_int
1413m_fixhdr(struct mbuf *m0)
1414{
1415 u_int len;
1416
1417 len = m_length(m0, NULL);
1418 m0->m_pkthdr.len = len;
1419 return (len);
1420}
1421
1422u_int
1423m_length(struct mbuf *m0, struct mbuf **last)
1424{
1425 struct mbuf *m;
1426 u_int len;
1427
1428 len = 0;
1429 for (m = m0; m != NULL; m = m->m_next) {
1430 len += m->m_len;
1431 if (m->m_next == NULL)
1432 break;
1433 }
1434 if (last != NULL)
1435 *last = m;
1436 return (len);
1437}
1438
1439/*
1440 * Defragment a mbuf chain, returning the shortest possible
1441 * chain of mbufs and clusters. If allocation fails and
1442 * this cannot be completed, NULL will be returned, but
1443 * the passed in chain will be unchanged. Upon success,
1444 * the original chain will be freed, and the new chain
1445 * will be returned.
1446 *
1447 * If a non-packet header is passed in, the original
1448 * mbuf (chain?) will be returned unharmed.
1449 */
1450struct mbuf *
1451m_defrag(struct mbuf *m0, int how)
1452{
1453 struct mbuf *m_new = NULL, *m_final = NULL;
1454 int progress = 0, length;
1455
1456 MBUF_CHECKSLEEP(how);
1457 if (!(m0->m_flags & M_PKTHDR))
1458 return (m0);
1459
1460 m_fixhdr(m0); /* Needed sanity check */
1461
1462#ifdef MBUF_STRESS_TEST
1463 if (m_defragrandomfailures) {
1464 int temp = arc4random() & 0xff;
1465 if (temp == 0xba)
1466 goto nospace;
1467 }
1468#endif
1469
1470 if (m0->m_pkthdr.len > MHLEN)
1471 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1472 else
1473 m_final = m_gethdr(how, MT_DATA);
1474
1475 if (m_final == NULL)
1476 goto nospace;
1477
1478 if (m_dup_pkthdr(m_final, m0, how) == 0)
1479 goto nospace;
1480
1481 m_new = m_final;
1482
1483 while (progress < m0->m_pkthdr.len) {
1484 length = m0->m_pkthdr.len - progress;
1485 if (length > MCLBYTES)
1486 length = MCLBYTES;
1487
1488 if (m_new == NULL) {
1489 if (length > MLEN)
1490 m_new = m_getcl(how, MT_DATA, 0);
1491 else
1492 m_new = m_get(how, MT_DATA);
1493 if (m_new == NULL)
1494 goto nospace;
1495 }
1496
1497 m_copydata(m0, progress, length, mtod(m_new, caddr_t));
1498 progress += length;
1499 m_new->m_len = length;
1500 if (m_new != m_final)
1501 m_cat(m_final, m_new);
1502 m_new = NULL;
1503 }
1504#ifdef MBUF_STRESS_TEST
1505 if (m0->m_next == NULL)
1506 m_defraguseless++;
1507#endif
1508 m_freem(m0);
1509 m0 = m_final;
1510#ifdef MBUF_STRESS_TEST
1511 m_defragpackets++;
1512 m_defragbytes += m0->m_pkthdr.len;
1513#endif
1514 return (m0);
1515nospace:
1516#ifdef MBUF_STRESS_TEST
1517 m_defragfailure++;
1518#endif
1519 if (m_final)
1520 m_freem(m_final);
1521 return (NULL);
1522}
1523
1524#ifdef MBUF_STRESS_TEST
1525
1526/*
1527 * Fragment an mbuf chain. There's no reason you'd ever want to do
1528 * this in normal usage, but it's great for stress testing various
1529 * mbuf consumers.
1530 *
1531 * If fragmentation is not possible, the original chain will be
1532 * returned.
1533 *
1534 * Possible length values:
1535 * 0 no fragmentation will occur
1536 * > 0 each fragment will be of the specified length
1537 * -1 each fragment will be the same random value in length
1538 * -2 each fragment's length will be entirely random
1539 * (Random values range from 1 to 256)
1540 */
1541struct mbuf *
1542m_fragment(struct mbuf *m0, int how, int length)
1543{
1544 struct mbuf *m_new = NULL, *m_final = NULL;
1545 int progress = 0;
1546
1547 if (!(m0->m_flags & M_PKTHDR))
1548 return (m0);
1549
1550 if ((length == 0) || (length < -2))
1551 return (m0);
1552
1553 m_fixhdr(m0); /* Needed sanity check */
1554
1555 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1556
1557 if (m_final == NULL)
1558 goto nospace;
1559
1560 if (m_dup_pkthdr(m_final, m0, how) == 0)
1561 goto nospace;
1562
1563 m_new = m_final;
1564
1565 if (length == -1)
1566 length = 1 + (arc4random() & 255);
1567
1568 while (progress < m0->m_pkthdr.len) {
1569 int fraglen;
1570
1571 if (length > 0)
1572 fraglen = length;
1573 else
1574 fraglen = 1 + (arc4random() & 255);
1575 if (fraglen > m0->m_pkthdr.len - progress)
1576 fraglen = m0->m_pkthdr.len - progress;
1577
1578 if (fraglen > MCLBYTES)
1579 fraglen = MCLBYTES;
1580
1581 if (m_new == NULL) {
1582 m_new = m_getcl(how, MT_DATA, 0);
1583 if (m_new == NULL)
1584 goto nospace;
1585 }
1586
1587 m_copydata(m0, progress, fraglen, mtod(m_new, caddr_t));
1588 progress += fraglen;
1589 m_new->m_len = fraglen;
1590 if (m_new != m_final)
1591 m_cat(m_final, m_new);
1592 m_new = NULL;
1593 }
1594 m_freem(m0);
1595 m0 = m_final;
1596 return (m0);
1597nospace:
1598 if (m_final)
1599 m_freem(m_final);
1600 /* Return the original chain on failure */
1601 return (m0);
1602}
1603
1604#endif
1605
1606struct mbuf *
1607m_uiotombuf(struct uio *uio, int how, int len, int align)
1608{
1609 struct mbuf *m_new = NULL, *m_final = NULL;
1610 int progress = 0, error = 0, length, total;
1611
1612 if (len > 0)
1613 total = min(uio->uio_resid, len);
1614 else
1615 total = uio->uio_resid;
1616 if (align >= MHLEN)
1617 goto nospace;
1618 if (total + align > MHLEN)
1619 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1620 else
1621 m_final = m_gethdr(how, MT_DATA);
1622 if (m_final == NULL)
1623 goto nospace;
1624 m_final->m_data += align;
1625 m_new = m_final;
1626 while (progress < total) {
1627 length = total - progress;
1628 if (length > MCLBYTES)
1629 length = MCLBYTES;
1630 if (m_new == NULL) {
1631 if (length > MLEN)
1632 m_new = m_getcl(how, MT_DATA, 0);
1633 else
1634 m_new = m_get(how, MT_DATA);
1635 if (m_new == NULL)
1636 goto nospace;
1637 }
1638 error = uiomove(mtod(m_new, void *), length, uio);
1639 if (error)
1640 goto nospace;
1641 progress += length;
1642 m_new->m_len = length;
1643 if (m_new != m_final)
1644 m_cat(m_final, m_new);
1645 m_new = NULL;
1646 }
1647 m_fixhdr(m_final);
1648 return (m_final);
1649nospace:
1650 if (m_new)
1651 m_free(m_new);
1652 if (m_final)
1653 m_freem(m_final);
1654 return (NULL);
1655}
1656
1657/*
1658 * Set the m_data pointer of a newly-allocated mbuf
1659 * to place an object of the specified size at the
1660 * end of the mbuf, longword aligned.
1661 */
1662void
1663m_align(struct mbuf *m, int len)
1664{
1665 int adjust;
1666
1667 if (m->m_flags & M_EXT)
1668 adjust = m->m_ext.ext_size - len;
1669 else if (m->m_flags & M_PKTHDR)
1670 adjust = MHLEN - len;
1671 else
1672 adjust = MLEN - len;
1673 m->m_data += adjust &~ (sizeof(long)-1);
1674}
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 152035 2005-11-04 17:20:53Z 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_CLUSTER: /* 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 break;
226 case EXT_JUMBO9:
227 uma_zfree(zone_jumbo9, m->m_ext.ext_buf);
228 break;
229 case EXT_JUMBO16:
230 uma_zfree(zone_jumbo16, m->m_ext.ext_buf);
231 break;
232 case EXT_SFBUF:
233 case EXT_NET_DRV:
234 case EXT_MOD_TYPE:
235 case EXT_DISPOSABLE:
236 *(m->m_ext.ref_cnt) = 0;
237 uma_zfree(zone_ext_refcnt, __DEVOLATILE(u_int *,
238 m->m_ext.ref_cnt));
239 /* FALLTHROUGH */
240 case EXT_EXTREF:
241 KASSERT(m->m_ext.ext_free != NULL,
242 ("%s: ext_free not set", __func__));
243 (*(m->m_ext.ext_free))(m->m_ext.ext_buf,
244 m->m_ext.ext_args);
245 break;
246 default:
247 KASSERT(m->m_ext.ext_type == 0,
248 ("%s: unknown ext_type", __func__));
249 }
250 }
251 /*
252 * Free this mbuf back to the mbuf zone with all m_ext
253 * information purged.
254 */
255 m->m_ext.ext_buf = NULL;
256 m->m_ext.ext_free = NULL;
257 m->m_ext.ext_args = NULL;
258 m->m_ext.ref_cnt = NULL;
259 m->m_ext.ext_size = 0;
260 m->m_ext.ext_type = 0;
261 m->m_flags &= ~M_EXT;
262 uma_zfree(zone_mbuf, m);
263}
264
265/*
266 * Attach the the cluster from *m to *n, set up m_ext in *n
267 * and bump the refcount of the cluster.
268 */
269static void
270mb_dupcl(struct mbuf *n, struct mbuf *m)
271{
272 KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__));
273 KASSERT(m->m_ext.ref_cnt != NULL, ("%s: ref_cnt not set", __func__));
274 KASSERT((n->m_flags & M_EXT) == 0, ("%s: M_EXT set", __func__));
275
276 if (*(m->m_ext.ref_cnt) == 1)
277 *(m->m_ext.ref_cnt) += 1;
278 else
279 atomic_add_int(m->m_ext.ref_cnt, 1);
280 n->m_ext.ext_buf = m->m_ext.ext_buf;
281 n->m_ext.ext_free = m->m_ext.ext_free;
282 n->m_ext.ext_args = m->m_ext.ext_args;
283 n->m_ext.ext_size = m->m_ext.ext_size;
284 n->m_ext.ref_cnt = m->m_ext.ref_cnt;
285 n->m_ext.ext_type = m->m_ext.ext_type;
286 n->m_flags |= M_EXT;
287}
288
289/*
290 * Clean up mbuf (chain) from any tags and packet headers.
291 * If "all" is set then the first mbuf in the chain will be
292 * cleaned too.
293 */
294void
295m_demote(struct mbuf *m0, int all)
296{
297 struct mbuf *m;
298
299 for (m = all ? m0 : m0->m_next; m != NULL; m = m->m_next) {
300 if (m->m_flags & M_PKTHDR) {
301 m_tag_delete_chain(m, NULL);
302 m->m_flags &= ~M_PKTHDR;
303 bzero(&m->m_pkthdr, sizeof(struct pkthdr));
304 }
305 if (m->m_type == MT_HEADER)
306 m->m_type = MT_DATA;
307 if (m != m0 && m->m_nextpkt != NULL)
308 m->m_nextpkt = NULL;
309 m->m_flags = m->m_flags & (M_EXT|M_EOR|M_RDONLY|M_FREELIST);
310 }
311}
312
313/*
314 * Sanity checks on mbuf (chain) for use in KASSERT() and general
315 * debugging.
316 * Returns 0 or panics when bad and 1 on all tests passed.
317 * Sanitize, 0 to run M_SANITY_ACTION, 1 to garble things so they
318 * blow up later.
319 */
320int
321m_sanity(struct mbuf *m0, int sanitize)
322{
323 struct mbuf *m;
324 caddr_t a, b;
325 int pktlen = 0;
326
327#define M_SANITY_ACTION(s) return (0)
328/* #define M_SANITY_ACTION(s) panic("mbuf %p: " s, m) */
329
330 for (m = m0; m != NULL; m = m->m_next) {
331 /*
332 * Basic pointer checks. If any of these fails then some
333 * unrelated kernel memory before or after us is trashed.
334 * No way to recover from that.
335 */
336 a = ((m->m_flags & M_EXT) ? m->m_ext.ext_buf :
337 ((m->m_flags & M_PKTHDR) ? (caddr_t)(&m->m_pktdat) :
338 (caddr_t)(&m->m_dat)) );
339 b = (caddr_t)(a + (m->m_flags & M_EXT ? m->m_ext.ext_size :
340 ((m->m_flags & M_PKTHDR) ? MHLEN : MLEN)));
341 if ((caddr_t)m->m_data < a)
342 M_SANITY_ACTION("m_data outside mbuf data range left");
343 if ((caddr_t)m->m_data > b)
344 M_SANITY_ACTION("m_data outside mbuf data range right");
345 if ((caddr_t)m->m_data + m->m_len > b)
346 M_SANITY_ACTION("m_data + m_len exeeds mbuf space");
347 if ((m->m_flags & M_PKTHDR) && m->m_pkthdr.header) {
348 if ((caddr_t)m->m_pkthdr.header < a ||
349 (caddr_t)m->m_pkthdr.header > b)
350 M_SANITY_ACTION("m_pkthdr.header outside mbuf data range");
351 }
352
353 /* m->m_nextpkt may only be set on first mbuf in chain. */
354 if (m != m0 && m->m_nextpkt != NULL) {
355 if (sanitize) {
356 m_freem(m->m_nextpkt);
357 m->m_nextpkt = (struct mbuf *)0xDEADC0DE;
358 } else
359 M_SANITY_ACTION("m->m_nextpkt on in-chain mbuf");
360 }
361
362 /* correct type correlations. */
363 if (m->m_type == MT_HEADER && !(m->m_flags & M_PKTHDR)) {
364 if (sanitize)
365 m->m_type = MT_DATA;
366 else
367 M_SANITY_ACTION("MT_HEADER set but not M_PKTHDR");
368 }
369
370 /* packet length (not mbuf length!) calculation */
371 if (m0->m_flags & M_PKTHDR)
372 pktlen += m->m_len;
373
374 /* m_tags may only be attached to first mbuf in chain. */
375 if (m != m0 && m->m_flags & M_PKTHDR &&
376 !SLIST_EMPTY(&m->m_pkthdr.tags)) {
377 if (sanitize) {
378 m_tag_delete_chain(m, NULL);
379 /* put in 0xDEADC0DE perhaps? */
380 } else
381 M_SANITY_ACTION("m_tags on in-chain mbuf");
382 }
383
384 /* M_PKTHDR may only be set on first mbuf in chain */
385 if (m != m0 && m->m_flags & M_PKTHDR) {
386 if (sanitize) {
387 bzero(&m->m_pkthdr, sizeof(m->m_pkthdr));
388 m->m_flags &= ~M_PKTHDR;
389 /* put in 0xDEADCODE and leave hdr flag in */
390 } else
391 M_SANITY_ACTION("M_PKTHDR on in-chain mbuf");
392 }
393 }
394 m = m0;
395 if (pktlen && pktlen != m->m_pkthdr.len) {
396 if (sanitize)
397 m->m_pkthdr.len = 0;
398 else
399 M_SANITY_ACTION("m_pkthdr.len != mbuf chain length");
400 }
401 return 1;
402
403#undef M_SANITY_ACTION
404}
405
406
407/*
408 * "Move" mbuf pkthdr from "from" to "to".
409 * "from" must have M_PKTHDR set, and "to" must be empty.
410 */
411void
412m_move_pkthdr(struct mbuf *to, struct mbuf *from)
413{
414
415#if 0
416 /* see below for why these are not enabled */
417 M_ASSERTPKTHDR(to);
418 /* Note: with MAC, this may not be a good assertion. */
419 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags),
420 ("m_move_pkthdr: to has tags"));
421#endif
422#ifdef MAC
423 /*
424 * XXXMAC: It could be this should also occur for non-MAC?
425 */
426 if (to->m_flags & M_PKTHDR)
427 m_tag_delete_chain(to, NULL);
428#endif
429 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT);
430 if ((to->m_flags & M_EXT) == 0)
431 to->m_data = to->m_pktdat;
432 to->m_pkthdr = from->m_pkthdr; /* especially tags */
433 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */
434 from->m_flags &= ~M_PKTHDR;
435}
436
437/*
438 * Duplicate "from"'s mbuf pkthdr in "to".
439 * "from" must have M_PKTHDR set, and "to" must be empty.
440 * In particular, this does a deep copy of the packet tags.
441 */
442int
443m_dup_pkthdr(struct mbuf *to, struct mbuf *from, int how)
444{
445
446#if 0
447 /*
448 * The mbuf allocator only initializes the pkthdr
449 * when the mbuf is allocated with MGETHDR. Many users
450 * (e.g. m_copy*, m_prepend) use MGET and then
451 * smash the pkthdr as needed causing these
452 * assertions to trip. For now just disable them.
453 */
454 M_ASSERTPKTHDR(to);
455 /* Note: with MAC, this may not be a good assertion. */
456 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), ("m_dup_pkthdr: to has tags"));
457#endif
458 MBUF_CHECKSLEEP(how);
459#ifdef MAC
460 if (to->m_flags & M_PKTHDR)
461 m_tag_delete_chain(to, NULL);
462#endif
463 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT);
464 if ((to->m_flags & M_EXT) == 0)
465 to->m_data = to->m_pktdat;
466 to->m_pkthdr = from->m_pkthdr;
467 SLIST_INIT(&to->m_pkthdr.tags);
468 return (m_tag_copy_chain(to, from, MBTOM(how)));
469}
470
471/*
472 * Lesser-used path for M_PREPEND:
473 * allocate new mbuf to prepend to chain,
474 * copy junk along.
475 */
476struct mbuf *
477m_prepend(struct mbuf *m, int len, int how)
478{
479 struct mbuf *mn;
480
481 if (m->m_flags & M_PKTHDR)
482 MGETHDR(mn, how, m->m_type);
483 else
484 MGET(mn, how, m->m_type);
485 if (mn == NULL) {
486 m_freem(m);
487 return (NULL);
488 }
489 if (m->m_flags & M_PKTHDR)
490 M_MOVE_PKTHDR(mn, m);
491 mn->m_next = m;
492 m = mn;
493 if (len < MHLEN)
494 MH_ALIGN(m, len);
495 m->m_len = len;
496 return (m);
497}
498
499/*
500 * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
501 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf.
502 * The wait parameter is a choice of M_TRYWAIT/M_DONTWAIT from caller.
503 * Note that the copy is read-only, because clusters are not copied,
504 * only their reference counts are incremented.
505 */
506struct mbuf *
507m_copym(struct mbuf *m, int off0, int len, int wait)
508{
509 struct mbuf *n, **np;
510 int off = off0;
511 struct mbuf *top;
512 int copyhdr = 0;
513
514 KASSERT(off >= 0, ("m_copym, negative off %d", off));
515 KASSERT(len >= 0, ("m_copym, negative len %d", len));
516 MBUF_CHECKSLEEP(wait);
517 if (off == 0 && m->m_flags & M_PKTHDR)
518 copyhdr = 1;
519 while (off > 0) {
520 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain"));
521 if (off < m->m_len)
522 break;
523 off -= m->m_len;
524 m = m->m_next;
525 }
526 np = ⊤
527 top = 0;
528 while (len > 0) {
529 if (m == NULL) {
530 KASSERT(len == M_COPYALL,
531 ("m_copym, length > size of mbuf chain"));
532 break;
533 }
534 if (copyhdr)
535 MGETHDR(n, wait, m->m_type);
536 else
537 MGET(n, wait, m->m_type);
538 *np = n;
539 if (n == NULL)
540 goto nospace;
541 if (copyhdr) {
542 if (!m_dup_pkthdr(n, m, wait))
543 goto nospace;
544 if (len == M_COPYALL)
545 n->m_pkthdr.len -= off0;
546 else
547 n->m_pkthdr.len = len;
548 copyhdr = 0;
549 }
550 n->m_len = min(len, m->m_len - off);
551 if (m->m_flags & M_EXT) {
552 n->m_data = m->m_data + off;
553 mb_dupcl(n, m);
554 } else
555 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
556 (u_int)n->m_len);
557 if (len != M_COPYALL)
558 len -= n->m_len;
559 off = 0;
560 m = m->m_next;
561 np = &n->m_next;
562 }
563 if (top == NULL)
564 mbstat.m_mcfail++; /* XXX: No consistency. */
565
566 return (top);
567nospace:
568 m_freem(top);
569 mbstat.m_mcfail++; /* XXX: No consistency. */
570 return (NULL);
571}
572
573/*
574 * Returns mbuf chain with new head for the prepending case.
575 * Copies from mbuf (chain) n from off for len to mbuf (chain) m
576 * either prepending or appending the data.
577 * The resulting mbuf (chain) m is fully writeable.
578 * m is destination (is made writeable)
579 * n is source, off is offset in source, len is len from offset
580 * dir, 0 append, 1 prepend
581 * how, wait or nowait
582 */
583
584static int
585m_bcopyxxx(void *s, void *t, u_int len)
586{
587 bcopy(s, t, (size_t)len);
588 return 0;
589}
590
591struct mbuf *
592m_copymdata(struct mbuf *m, struct mbuf *n, int off, int len,
593 int prep, int how)
594{
595 struct mbuf *mm, *x, *z, *prev = NULL;
596 caddr_t p;
597 int i, nlen = 0;
598 caddr_t buf[MLEN];
599
600 KASSERT(m != NULL && n != NULL, ("m_copymdata, no target or source"));
601 KASSERT(off >= 0, ("m_copymdata, negative off %d", off));
602 KASSERT(len >= 0, ("m_copymdata, negative len %d", len));
603 KASSERT(prep == 0 || prep == 1, ("m_copymdata, unknown direction %d", prep));
604
605 mm = m;
606 if (!prep) {
607 while(mm->m_next) {
608 prev = mm;
609 mm = mm->m_next;
610 }
611 }
612 for (z = n; z != NULL; z = z->m_next)
613 nlen += z->m_len;
614 if (len == M_COPYALL)
615 len = nlen - off;
616 if (off + len > nlen || len < 1)
617 return NULL;
618
619 if (!M_WRITABLE(mm)) {
620 /* XXX: Use proper m_xxx function instead. */
621 x = m_getcl(how, MT_DATA, mm->m_flags);
622 if (x == NULL)
623 return NULL;
624 bcopy(mm->m_ext.ext_buf, x->m_ext.ext_buf, x->m_ext.ext_size);
625 p = x->m_ext.ext_buf + (mm->m_data - mm->m_ext.ext_buf);
626 x->m_data = p;
627 mm->m_next = NULL;
628 if (mm != m)
629 prev->m_next = x;
630 m_free(mm);
631 mm = x;
632 }
633
634 /*
635 * Append/prepend the data. Allocating mbufs as necessary.
636 */
637 /* Shortcut if enough free space in first/last mbuf. */
638 if (!prep && M_TRAILINGSPACE(mm) >= len) {
639 m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t) +
640 mm->m_len);
641 mm->m_len += len;
642 mm->m_pkthdr.len += len;
643 return m;
644 }
645 if (prep && M_LEADINGSPACE(mm) >= len) {
646 mm->m_data = mtod(mm, caddr_t) - len;
647 m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t));
648 mm->m_len += len;
649 mm->m_pkthdr.len += len;
650 return mm;
651 }
652
653 /* Expand first/last mbuf to cluster if possible. */
654 if (!prep && !(mm->m_flags & M_EXT) && len > M_TRAILINGSPACE(mm)) {
655 bcopy(mm->m_data, &buf, mm->m_len);
656 m_clget(mm, how);
657 if (!(mm->m_flags & M_EXT))
658 return NULL;
659 bcopy(&buf, mm->m_ext.ext_buf, mm->m_len);
660 mm->m_data = mm->m_ext.ext_buf;
661 mm->m_pkthdr.header = NULL;
662 }
663 if (prep && !(mm->m_flags & M_EXT) && len > M_LEADINGSPACE(mm)) {
664 bcopy(mm->m_data, &buf, mm->m_len);
665 m_clget(mm, how);
666 if (!(mm->m_flags & M_EXT))
667 return NULL;
668 bcopy(&buf, (caddr_t *)mm->m_ext.ext_buf +
669 mm->m_ext.ext_size - mm->m_len, mm->m_len);
670 mm->m_data = (caddr_t)mm->m_ext.ext_buf +
671 mm->m_ext.ext_size - mm->m_len;
672 mm->m_pkthdr.header = NULL;
673 }
674
675 /* Append/prepend as many mbuf (clusters) as necessary to fit len. */
676 if (!prep && len > M_TRAILINGSPACE(mm)) {
677 if (!m_getm(mm, len - M_TRAILINGSPACE(mm), how, MT_DATA))
678 return NULL;
679 }
680 if (prep && len > M_LEADINGSPACE(mm)) {
681 if (!(z = m_getm(NULL, len - M_LEADINGSPACE(mm), how, MT_DATA)))
682 return NULL;
683 i = 0;
684 for (x = z; x != NULL; x = x->m_next) {
685 i += x->m_flags & M_EXT ? x->m_ext.ext_size :
686 (x->m_flags & M_PKTHDR ? MHLEN : MLEN);
687 if (!x->m_next)
688 break;
689 }
690 z->m_data += i - len;
691 m_move_pkthdr(mm, z);
692 x->m_next = mm;
693 mm = z;
694 }
695
696 /* Seek to start position in source mbuf. Optimization for long chains. */
697 while (off > 0) {
698 if (off < n->m_len)
699 break;
700 off -= n->m_len;
701 n = n->m_next;
702 }
703
704 /* Copy data into target mbuf. */
705 z = mm;
706 while (len > 0) {
707 KASSERT(z != NULL, ("m_copymdata, falling off target edge"));
708 i = M_TRAILINGSPACE(z);
709 m_apply(n, off, i, m_bcopyxxx, mtod(z, caddr_t) + z->m_len);
710 z->m_len += i;
711 /* fixup pkthdr.len if necessary */
712 if ((prep ? mm : m)->m_flags & M_PKTHDR)
713 (prep ? mm : m)->m_pkthdr.len += i;
714 off += i;
715 len -= i;
716 z = z->m_next;
717 }
718 return (prep ? mm : m);
719}
720
721/*
722 * Copy an entire packet, including header (which must be present).
723 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
724 * Note that the copy is read-only, because clusters are not copied,
725 * only their reference counts are incremented.
726 * Preserve alignment of the first mbuf so if the creator has left
727 * some room at the beginning (e.g. for inserting protocol headers)
728 * the copies still have the room available.
729 */
730struct mbuf *
731m_copypacket(struct mbuf *m, int how)
732{
733 struct mbuf *top, *n, *o;
734
735 MBUF_CHECKSLEEP(how);
736 MGET(n, how, m->m_type);
737 top = n;
738 if (n == NULL)
739 goto nospace;
740
741 if (!m_dup_pkthdr(n, m, how))
742 goto nospace;
743 n->m_len = m->m_len;
744 if (m->m_flags & M_EXT) {
745 n->m_data = m->m_data;
746 mb_dupcl(n, m);
747 } else {
748 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat );
749 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
750 }
751
752 m = m->m_next;
753 while (m) {
754 MGET(o, how, m->m_type);
755 if (o == NULL)
756 goto nospace;
757
758 n->m_next = o;
759 n = n->m_next;
760
761 n->m_len = m->m_len;
762 if (m->m_flags & M_EXT) {
763 n->m_data = m->m_data;
764 mb_dupcl(n, m);
765 } else {
766 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
767 }
768
769 m = m->m_next;
770 }
771 return top;
772nospace:
773 m_freem(top);
774 mbstat.m_mcfail++; /* XXX: No consistency. */
775 return (NULL);
776}
777
778/*
779 * Copy data from an mbuf chain starting "off" bytes from the beginning,
780 * continuing for "len" bytes, into the indicated buffer.
781 */
782void
783m_copydata(const struct mbuf *m, int off, int len, caddr_t cp)
784{
785 u_int count;
786
787 KASSERT(off >= 0, ("m_copydata, negative off %d", off));
788 KASSERT(len >= 0, ("m_copydata, negative len %d", len));
789 while (off > 0) {
790 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain"));
791 if (off < m->m_len)
792 break;
793 off -= m->m_len;
794 m = m->m_next;
795 }
796 while (len > 0) {
797 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain"));
798 count = min(m->m_len - off, len);
799 bcopy(mtod(m, caddr_t) + off, cp, count);
800 len -= count;
801 cp += count;
802 off = 0;
803 m = m->m_next;
804 }
805}
806
807/*
808 * Copy a packet header mbuf chain into a completely new chain, including
809 * copying any mbuf clusters. Use this instead of m_copypacket() when
810 * you need a writable copy of an mbuf chain.
811 */
812struct mbuf *
813m_dup(struct mbuf *m, int how)
814{
815 struct mbuf **p, *top = NULL;
816 int remain, moff, nsize;
817
818 MBUF_CHECKSLEEP(how);
819 /* Sanity check */
820 if (m == NULL)
821 return (NULL);
822 M_ASSERTPKTHDR(m);
823
824 /* While there's more data, get a new mbuf, tack it on, and fill it */
825 remain = m->m_pkthdr.len;
826 moff = 0;
827 p = ⊤
828 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */
829 struct mbuf *n;
830
831 /* Get the next new mbuf */
832 if (remain >= MINCLSIZE) {
833 n = m_getcl(how, m->m_type, 0);
834 nsize = MCLBYTES;
835 } else {
836 n = m_get(how, m->m_type);
837 nsize = MLEN;
838 }
839 if (n == NULL)
840 goto nospace;
841
842 if (top == NULL) { /* First one, must be PKTHDR */
843 if (!m_dup_pkthdr(n, m, how)) {
844 m_free(n);
845 goto nospace;
846 }
847 nsize = MHLEN;
848 }
849 n->m_len = 0;
850
851 /* Link it into the new chain */
852 *p = n;
853 p = &n->m_next;
854
855 /* Copy data from original mbuf(s) into new mbuf */
856 while (n->m_len < nsize && m != NULL) {
857 int chunk = min(nsize - n->m_len, m->m_len - moff);
858
859 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
860 moff += chunk;
861 n->m_len += chunk;
862 remain -= chunk;
863 if (moff == m->m_len) {
864 m = m->m_next;
865 moff = 0;
866 }
867 }
868
869 /* Check correct total mbuf length */
870 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL),
871 ("%s: bogus m_pkthdr.len", __func__));
872 }
873 return (top);
874
875nospace:
876 m_freem(top);
877 mbstat.m_mcfail++; /* XXX: No consistency. */
878 return (NULL);
879}
880
881/*
882 * Concatenate mbuf chain n to m.
883 * Both chains must be of the same type (e.g. MT_DATA).
884 * Any m_pkthdr is not updated.
885 */
886void
887m_cat(struct mbuf *m, struct mbuf *n)
888{
889 while (m->m_next)
890 m = m->m_next;
891 while (n) {
892 if (m->m_flags & M_EXT ||
893 m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) {
894 /* just join the two chains */
895 m->m_next = n;
896 return;
897 }
898 /* splat the data from one into the other */
899 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
900 (u_int)n->m_len);
901 m->m_len += n->m_len;
902 n = m_free(n);
903 }
904}
905
906void
907m_adj(struct mbuf *mp, int req_len)
908{
909 int len = req_len;
910 struct mbuf *m;
911 int count;
912
913 if ((m = mp) == NULL)
914 return;
915 if (len >= 0) {
916 /*
917 * Trim from head.
918 */
919 while (m != NULL && len > 0) {
920 if (m->m_len <= len) {
921 len -= m->m_len;
922 m->m_len = 0;
923 m = m->m_next;
924 } else {
925 m->m_len -= len;
926 m->m_data += len;
927 len = 0;
928 }
929 }
930 m = mp;
931 if (mp->m_flags & M_PKTHDR)
932 m->m_pkthdr.len -= (req_len - len);
933 } else {
934 /*
935 * Trim from tail. Scan the mbuf chain,
936 * calculating its length and finding the last mbuf.
937 * If the adjustment only affects this mbuf, then just
938 * adjust and return. Otherwise, rescan and truncate
939 * after the remaining size.
940 */
941 len = -len;
942 count = 0;
943 for (;;) {
944 count += m->m_len;
945 if (m->m_next == (struct mbuf *)0)
946 break;
947 m = m->m_next;
948 }
949 if (m->m_len >= len) {
950 m->m_len -= len;
951 if (mp->m_flags & M_PKTHDR)
952 mp->m_pkthdr.len -= len;
953 return;
954 }
955 count -= len;
956 if (count < 0)
957 count = 0;
958 /*
959 * Correct length for chain is "count".
960 * Find the mbuf with last data, adjust its length,
961 * and toss data from remaining mbufs on chain.
962 */
963 m = mp;
964 if (m->m_flags & M_PKTHDR)
965 m->m_pkthdr.len = count;
966 for (; m; m = m->m_next) {
967 if (m->m_len >= count) {
968 m->m_len = count;
969 if (m->m_next != NULL) {
970 m_freem(m->m_next);
971 m->m_next = NULL;
972 }
973 break;
974 }
975 count -= m->m_len;
976 }
977 }
978}
979
980/*
981 * Rearange an mbuf chain so that len bytes are contiguous
982 * and in the data area of an mbuf (so that mtod and dtom
983 * will work for a structure of size len). Returns the resulting
984 * mbuf chain on success, frees it and returns null on failure.
985 * If there is room, it will add up to max_protohdr-len extra bytes to the
986 * contiguous region in an attempt to avoid being called next time.
987 */
988struct mbuf *
989m_pullup(struct mbuf *n, int len)
990{
991 struct mbuf *m;
992 int count;
993 int space;
994
995 /*
996 * If first mbuf has no cluster, and has room for len bytes
997 * without shifting current data, pullup into it,
998 * otherwise allocate a new mbuf to prepend to the chain.
999 */
1000 if ((n->m_flags & M_EXT) == 0 &&
1001 n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
1002 if (n->m_len >= len)
1003 return (n);
1004 m = n;
1005 n = n->m_next;
1006 len -= m->m_len;
1007 } else {
1008 if (len > MHLEN)
1009 goto bad;
1010 MGET(m, M_DONTWAIT, n->m_type);
1011 if (m == NULL)
1012 goto bad;
1013 m->m_len = 0;
1014 if (n->m_flags & M_PKTHDR)
1015 M_MOVE_PKTHDR(m, n);
1016 }
1017 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1018 do {
1019 count = min(min(max(len, max_protohdr), space), n->m_len);
1020 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
1021 (u_int)count);
1022 len -= count;
1023 m->m_len += count;
1024 n->m_len -= count;
1025 space -= count;
1026 if (n->m_len)
1027 n->m_data += count;
1028 else
1029 n = m_free(n);
1030 } while (len > 0 && n);
1031 if (len > 0) {
1032 (void) m_free(m);
1033 goto bad;
1034 }
1035 m->m_next = n;
1036 return (m);
1037bad:
1038 m_freem(n);
1039 mbstat.m_mpfail++; /* XXX: No consistency. */
1040 return (NULL);
1041}
1042
1043/*
1044 * Like m_pullup(), except a new mbuf is always allocated, and we allow
1045 * the amount of empty space before the data in the new mbuf to be specified
1046 * (in the event that the caller expects to prepend later).
1047 */
1048int MSFail;
1049
1050struct mbuf *
1051m_copyup(struct mbuf *n, int len, int dstoff)
1052{
1053 struct mbuf *m;
1054 int count, space;
1055
1056 if (len > (MHLEN - dstoff))
1057 goto bad;
1058 MGET(m, M_DONTWAIT, n->m_type);
1059 if (m == NULL)
1060 goto bad;
1061 m->m_len = 0;
1062 if (n->m_flags & M_PKTHDR)
1063 M_MOVE_PKTHDR(m, n);
1064 m->m_data += dstoff;
1065 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1066 do {
1067 count = min(min(max(len, max_protohdr), space), n->m_len);
1068 memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
1069 (unsigned)count);
1070 len -= count;
1071 m->m_len += count;
1072 n->m_len -= count;
1073 space -= count;
1074 if (n->m_len)
1075 n->m_data += count;
1076 else
1077 n = m_free(n);
1078 } while (len > 0 && n);
1079 if (len > 0) {
1080 (void) m_free(m);
1081 goto bad;
1082 }
1083 m->m_next = n;
1084 return (m);
1085 bad:
1086 m_freem(n);
1087 MSFail++;
1088 return (NULL);
1089}
1090
1091/*
1092 * Partition an mbuf chain in two pieces, returning the tail --
1093 * all but the first len0 bytes. In case of failure, it returns NULL and
1094 * attempts to restore the chain to its original state.
1095 *
1096 * Note that the resulting mbufs might be read-only, because the new
1097 * mbuf can end up sharing an mbuf cluster with the original mbuf if
1098 * the "breaking point" happens to lie within a cluster mbuf. Use the
1099 * M_WRITABLE() macro to check for this case.
1100 */
1101struct mbuf *
1102m_split(struct mbuf *m0, int len0, int wait)
1103{
1104 struct mbuf *m, *n;
1105 u_int len = len0, remain;
1106
1107 MBUF_CHECKSLEEP(wait);
1108 for (m = m0; m && len > m->m_len; m = m->m_next)
1109 len -= m->m_len;
1110 if (m == NULL)
1111 return (NULL);
1112 remain = m->m_len - len;
1113 if (m0->m_flags & M_PKTHDR) {
1114 MGETHDR(n, wait, m0->m_type);
1115 if (n == NULL)
1116 return (NULL);
1117 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
1118 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
1119 m0->m_pkthdr.len = len0;
1120 if (m->m_flags & M_EXT)
1121 goto extpacket;
1122 if (remain > MHLEN) {
1123 /* m can't be the lead packet */
1124 MH_ALIGN(n, 0);
1125 n->m_next = m_split(m, len, wait);
1126 if (n->m_next == NULL) {
1127 (void) m_free(n);
1128 return (NULL);
1129 } else {
1130 n->m_len = 0;
1131 return (n);
1132 }
1133 } else
1134 MH_ALIGN(n, remain);
1135 } else if (remain == 0) {
1136 n = m->m_next;
1137 m->m_next = NULL;
1138 return (n);
1139 } else {
1140 MGET(n, wait, m->m_type);
1141 if (n == NULL)
1142 return (NULL);
1143 M_ALIGN(n, remain);
1144 }
1145extpacket:
1146 if (m->m_flags & M_EXT) {
1147 n->m_data = m->m_data + len;
1148 mb_dupcl(n, m);
1149 } else {
1150 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
1151 }
1152 n->m_len = remain;
1153 m->m_len = len;
1154 n->m_next = m->m_next;
1155 m->m_next = NULL;
1156 return (n);
1157}
1158/*
1159 * Routine to copy from device local memory into mbufs.
1160 * Note that `off' argument is offset into first mbuf of target chain from
1161 * which to begin copying the data to.
1162 */
1163struct mbuf *
1164m_devget(char *buf, int totlen, int off, struct ifnet *ifp,
1165 void (*copy)(char *from, caddr_t to, u_int len))
1166{
1167 struct mbuf *m;
1168 struct mbuf *top = NULL, **mp = ⊤
1169 int len;
1170
1171 if (off < 0 || off > MHLEN)
1172 return (NULL);
1173
1174 while (totlen > 0) {
1175 if (top == NULL) { /* First one, must be PKTHDR */
1176 if (totlen + off >= MINCLSIZE) {
1177 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1178 len = MCLBYTES;
1179 } else {
1180 m = m_gethdr(M_DONTWAIT, MT_DATA);
1181 len = MHLEN;
1182
1183 /* Place initial small packet/header at end of mbuf */
1184 if (m && totlen + off + max_linkhdr <= MLEN) {
1185 m->m_data += max_linkhdr;
1186 len -= max_linkhdr;
1187 }
1188 }
1189 if (m == NULL)
1190 return NULL;
1191 m->m_pkthdr.rcvif = ifp;
1192 m->m_pkthdr.len = totlen;
1193 } else {
1194 if (totlen + off >= MINCLSIZE) {
1195 m = m_getcl(M_DONTWAIT, MT_DATA, 0);
1196 len = MCLBYTES;
1197 } else {
1198 m = m_get(M_DONTWAIT, MT_DATA);
1199 len = MLEN;
1200 }
1201 if (m == NULL) {
1202 m_freem(top);
1203 return NULL;
1204 }
1205 }
1206 if (off) {
1207 m->m_data += off;
1208 len -= off;
1209 off = 0;
1210 }
1211 m->m_len = len = min(totlen, len);
1212 if (copy)
1213 copy(buf, mtod(m, caddr_t), (u_int)len);
1214 else
1215 bcopy(buf, mtod(m, caddr_t), (u_int)len);
1216 buf += len;
1217 *mp = m;
1218 mp = &m->m_next;
1219 totlen -= len;
1220 }
1221 return (top);
1222}
1223
1224/*
1225 * Copy data from a buffer back into the indicated mbuf chain,
1226 * starting "off" bytes from the beginning, extending the mbuf
1227 * chain if necessary.
1228 */
1229void
1230m_copyback(struct mbuf *m0, int off, int len, c_caddr_t cp)
1231{
1232 int mlen;
1233 struct mbuf *m = m0, *n;
1234 int totlen = 0;
1235
1236 if (m0 == NULL)
1237 return;
1238 while (off > (mlen = m->m_len)) {
1239 off -= mlen;
1240 totlen += mlen;
1241 if (m->m_next == NULL) {
1242 n = m_get(M_DONTWAIT, m->m_type);
1243 if (n == NULL)
1244 goto out;
1245 bzero(mtod(n, caddr_t), MLEN);
1246 n->m_len = min(MLEN, len + off);
1247 m->m_next = n;
1248 }
1249 m = m->m_next;
1250 }
1251 while (len > 0) {
1252 mlen = min (m->m_len - off, len);
1253 bcopy(cp, off + mtod(m, caddr_t), (u_int)mlen);
1254 cp += mlen;
1255 len -= mlen;
1256 mlen += off;
1257 off = 0;
1258 totlen += mlen;
1259 if (len == 0)
1260 break;
1261 if (m->m_next == NULL) {
1262 n = m_get(M_DONTWAIT, m->m_type);
1263 if (n == NULL)
1264 break;
1265 n->m_len = min(MLEN, len);
1266 m->m_next = n;
1267 }
1268 m = m->m_next;
1269 }
1270out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
1271 m->m_pkthdr.len = totlen;
1272}
1273
1274/*
1275 * Append the specified data to the indicated mbuf chain,
1276 * Extend the mbuf chain if the new data does not fit in
1277 * existing space.
1278 *
1279 * Return 1 if able to complete the job; otherwise 0.
1280 */
1281int
1282m_append(struct mbuf *m0, int len, c_caddr_t cp)
1283{
1284 struct mbuf *m, *n;
1285 int remainder, space;
1286
1287 for (m = m0; m->m_next != NULL; m = m->m_next)
1288 ;
1289 remainder = len;
1290 space = M_TRAILINGSPACE(m);
1291 if (space > 0) {
1292 /*
1293 * Copy into available space.
1294 */
1295 if (space > remainder)
1296 space = remainder;
1297 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1298 m->m_len += space;
1299 cp += space, remainder -= space;
1300 }
1301 while (remainder > 0) {
1302 /*
1303 * Allocate a new mbuf; could check space
1304 * and allocate a cluster instead.
1305 */
1306 n = m_get(M_DONTWAIT, m->m_type);
1307 if (n == NULL)
1308 break;
1309 n->m_len = min(MLEN, remainder);
1310 bcopy(cp, mtod(n, caddr_t), n->m_len);
1311 cp += n->m_len, remainder -= n->m_len;
1312 m->m_next = n;
1313 m = n;
1314 }
1315 if (m0->m_flags & M_PKTHDR)
1316 m0->m_pkthdr.len += len - remainder;
1317 return (remainder == 0);
1318}
1319
1320/*
1321 * Apply function f to the data in an mbuf chain starting "off" bytes from
1322 * the beginning, continuing for "len" bytes.
1323 */
1324int
1325m_apply(struct mbuf *m, int off, int len,
1326 int (*f)(void *, void *, u_int), void *arg)
1327{
1328 u_int count;
1329 int rval;
1330
1331 KASSERT(off >= 0, ("m_apply, negative off %d", off));
1332 KASSERT(len >= 0, ("m_apply, negative len %d", len));
1333 while (off > 0) {
1334 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1335 if (off < m->m_len)
1336 break;
1337 off -= m->m_len;
1338 m = m->m_next;
1339 }
1340 while (len > 0) {
1341 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1342 count = min(m->m_len - off, len);
1343 rval = (*f)(arg, mtod(m, caddr_t) + off, count);
1344 if (rval)
1345 return (rval);
1346 len -= count;
1347 off = 0;
1348 m = m->m_next;
1349 }
1350 return (0);
1351}
1352
1353/*
1354 * Return a pointer to mbuf/offset of location in mbuf chain.
1355 */
1356struct mbuf *
1357m_getptr(struct mbuf *m, int loc, int *off)
1358{
1359
1360 while (loc >= 0) {
1361 /* Normal end of search. */
1362 if (m->m_len > loc) {
1363 *off = loc;
1364 return (m);
1365 } else {
1366 loc -= m->m_len;
1367 if (m->m_next == NULL) {
1368 if (loc == 0) {
1369 /* Point at the end of valid data. */
1370 *off = m->m_len;
1371 return (m);
1372 }
1373 return (NULL);
1374 }
1375 m = m->m_next;
1376 }
1377 }
1378 return (NULL);
1379}
1380
1381void
1382m_print(const struct mbuf *m, int maxlen)
1383{
1384 int len;
1385 int pdata;
1386 const struct mbuf *m2;
1387
1388 if (m->m_flags & M_PKTHDR)
1389 len = m->m_pkthdr.len;
1390 else
1391 len = -1;
1392 m2 = m;
1393 while (m2 != NULL && (len == -1 || len)) {
1394 pdata = m2->m_len;
1395 if (maxlen != -1 && pdata > maxlen)
1396 pdata = maxlen;
1397 printf("mbuf: %p len: %d, next: %p, %b%s", m2, m2->m_len,
1398 m2->m_next, m2->m_flags, "\20\20freelist\17skipfw"
1399 "\11proto5\10proto4\7proto3\6proto2\5proto1\4rdonly"
1400 "\3eor\2pkthdr\1ext", pdata ? "" : "\n");
1401 if (pdata)
1402 printf(", %*D\n", m2->m_len, (u_char *)m2->m_data, "-");
1403 if (len != -1)
1404 len -= m2->m_len;
1405 m2 = m2->m_next;
1406 }
1407 if (len > 0)
1408 printf("%d bytes unaccounted for.\n", len);
1409 return;
1410}
1411
1412u_int
1413m_fixhdr(struct mbuf *m0)
1414{
1415 u_int len;
1416
1417 len = m_length(m0, NULL);
1418 m0->m_pkthdr.len = len;
1419 return (len);
1420}
1421
1422u_int
1423m_length(struct mbuf *m0, struct mbuf **last)
1424{
1425 struct mbuf *m;
1426 u_int len;
1427
1428 len = 0;
1429 for (m = m0; m != NULL; m = m->m_next) {
1430 len += m->m_len;
1431 if (m->m_next == NULL)
1432 break;
1433 }
1434 if (last != NULL)
1435 *last = m;
1436 return (len);
1437}
1438
1439/*
1440 * Defragment a mbuf chain, returning the shortest possible
1441 * chain of mbufs and clusters. If allocation fails and
1442 * this cannot be completed, NULL will be returned, but
1443 * the passed in chain will be unchanged. Upon success,
1444 * the original chain will be freed, and the new chain
1445 * will be returned.
1446 *
1447 * If a non-packet header is passed in, the original
1448 * mbuf (chain?) will be returned unharmed.
1449 */
1450struct mbuf *
1451m_defrag(struct mbuf *m0, int how)
1452{
1453 struct mbuf *m_new = NULL, *m_final = NULL;
1454 int progress = 0, length;
1455
1456 MBUF_CHECKSLEEP(how);
1457 if (!(m0->m_flags & M_PKTHDR))
1458 return (m0);
1459
1460 m_fixhdr(m0); /* Needed sanity check */
1461
1462#ifdef MBUF_STRESS_TEST
1463 if (m_defragrandomfailures) {
1464 int temp = arc4random() & 0xff;
1465 if (temp == 0xba)
1466 goto nospace;
1467 }
1468#endif
1469
1470 if (m0->m_pkthdr.len > MHLEN)
1471 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1472 else
1473 m_final = m_gethdr(how, MT_DATA);
1474
1475 if (m_final == NULL)
1476 goto nospace;
1477
1478 if (m_dup_pkthdr(m_final, m0, how) == 0)
1479 goto nospace;
1480
1481 m_new = m_final;
1482
1483 while (progress < m0->m_pkthdr.len) {
1484 length = m0->m_pkthdr.len - progress;
1485 if (length > MCLBYTES)
1486 length = MCLBYTES;
1487
1488 if (m_new == NULL) {
1489 if (length > MLEN)
1490 m_new = m_getcl(how, MT_DATA, 0);
1491 else
1492 m_new = m_get(how, MT_DATA);
1493 if (m_new == NULL)
1494 goto nospace;
1495 }
1496
1497 m_copydata(m0, progress, length, mtod(m_new, caddr_t));
1498 progress += length;
1499 m_new->m_len = length;
1500 if (m_new != m_final)
1501 m_cat(m_final, m_new);
1502 m_new = NULL;
1503 }
1504#ifdef MBUF_STRESS_TEST
1505 if (m0->m_next == NULL)
1506 m_defraguseless++;
1507#endif
1508 m_freem(m0);
1509 m0 = m_final;
1510#ifdef MBUF_STRESS_TEST
1511 m_defragpackets++;
1512 m_defragbytes += m0->m_pkthdr.len;
1513#endif
1514 return (m0);
1515nospace:
1516#ifdef MBUF_STRESS_TEST
1517 m_defragfailure++;
1518#endif
1519 if (m_final)
1520 m_freem(m_final);
1521 return (NULL);
1522}
1523
1524#ifdef MBUF_STRESS_TEST
1525
1526/*
1527 * Fragment an mbuf chain. There's no reason you'd ever want to do
1528 * this in normal usage, but it's great for stress testing various
1529 * mbuf consumers.
1530 *
1531 * If fragmentation is not possible, the original chain will be
1532 * returned.
1533 *
1534 * Possible length values:
1535 * 0 no fragmentation will occur
1536 * > 0 each fragment will be of the specified length
1537 * -1 each fragment will be the same random value in length
1538 * -2 each fragment's length will be entirely random
1539 * (Random values range from 1 to 256)
1540 */
1541struct mbuf *
1542m_fragment(struct mbuf *m0, int how, int length)
1543{
1544 struct mbuf *m_new = NULL, *m_final = NULL;
1545 int progress = 0;
1546
1547 if (!(m0->m_flags & M_PKTHDR))
1548 return (m0);
1549
1550 if ((length == 0) || (length < -2))
1551 return (m0);
1552
1553 m_fixhdr(m0); /* Needed sanity check */
1554
1555 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1556
1557 if (m_final == NULL)
1558 goto nospace;
1559
1560 if (m_dup_pkthdr(m_final, m0, how) == 0)
1561 goto nospace;
1562
1563 m_new = m_final;
1564
1565 if (length == -1)
1566 length = 1 + (arc4random() & 255);
1567
1568 while (progress < m0->m_pkthdr.len) {
1569 int fraglen;
1570
1571 if (length > 0)
1572 fraglen = length;
1573 else
1574 fraglen = 1 + (arc4random() & 255);
1575 if (fraglen > m0->m_pkthdr.len - progress)
1576 fraglen = m0->m_pkthdr.len - progress;
1577
1578 if (fraglen > MCLBYTES)
1579 fraglen = MCLBYTES;
1580
1581 if (m_new == NULL) {
1582 m_new = m_getcl(how, MT_DATA, 0);
1583 if (m_new == NULL)
1584 goto nospace;
1585 }
1586
1587 m_copydata(m0, progress, fraglen, mtod(m_new, caddr_t));
1588 progress += fraglen;
1589 m_new->m_len = fraglen;
1590 if (m_new != m_final)
1591 m_cat(m_final, m_new);
1592 m_new = NULL;
1593 }
1594 m_freem(m0);
1595 m0 = m_final;
1596 return (m0);
1597nospace:
1598 if (m_final)
1599 m_freem(m_final);
1600 /* Return the original chain on failure */
1601 return (m0);
1602}
1603
1604#endif
1605
1606struct mbuf *
1607m_uiotombuf(struct uio *uio, int how, int len, int align)
1608{
1609 struct mbuf *m_new = NULL, *m_final = NULL;
1610 int progress = 0, error = 0, length, total;
1611
1612 if (len > 0)
1613 total = min(uio->uio_resid, len);
1614 else
1615 total = uio->uio_resid;
1616 if (align >= MHLEN)
1617 goto nospace;
1618 if (total + align > MHLEN)
1619 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1620 else
1621 m_final = m_gethdr(how, MT_DATA);
1622 if (m_final == NULL)
1623 goto nospace;
1624 m_final->m_data += align;
1625 m_new = m_final;
1626 while (progress < total) {
1627 length = total - progress;
1628 if (length > MCLBYTES)
1629 length = MCLBYTES;
1630 if (m_new == NULL) {
1631 if (length > MLEN)
1632 m_new = m_getcl(how, MT_DATA, 0);
1633 else
1634 m_new = m_get(how, MT_DATA);
1635 if (m_new == NULL)
1636 goto nospace;
1637 }
1638 error = uiomove(mtod(m_new, void *), length, uio);
1639 if (error)
1640 goto nospace;
1641 progress += length;
1642 m_new->m_len = length;
1643 if (m_new != m_final)
1644 m_cat(m_final, m_new);
1645 m_new = NULL;
1646 }
1647 m_fixhdr(m_final);
1648 return (m_final);
1649nospace:
1650 if (m_new)
1651 m_free(m_new);
1652 if (m_final)
1653 m_freem(m_final);
1654 return (NULL);
1655}
1656
1657/*
1658 * Set the m_data pointer of a newly-allocated mbuf
1659 * to place an object of the specified size at the
1660 * end of the mbuf, longword aligned.
1661 */
1662void
1663m_align(struct mbuf *m, int len)
1664{
1665 int adjust;
1666
1667 if (m->m_flags & M_EXT)
1668 adjust = m->m_ext.ext_size - len;
1669 else if (m->m_flags & M_PKTHDR)
1670 adjust = MHLEN - len;
1671 else
1672 adjust = MLEN - len;
1673 m->m_data += adjust &~ (sizeof(long)-1);
1674}