FreeBSD/Linux Kernel Cross Reference
sys/kern/uipc_mbuf.c
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: src/sys/kern/uipc_mbuf.c,v 1.135.2.7 2005/12/22 04:29:03 emaste Exp $");
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
52 int max_linkhdr;
53 int max_protohdr;
54 int max_hdr;
55 int max_datalen;
56 #ifdef MBUF_STRESS_TEST
57 int m_defragpackets;
58 int m_defragbytes;
59 int m_defraguseless;
60 int m_defragfailure;
61 int m_defragrandomfailures;
62 #endif
63
64 /*
65 * sysctl(8) exported objects
66 */
67 SYSCTL_DECL(_kern_ipc);
68 SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RW,
69 &max_linkhdr, 0, "");
70 SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RW,
71 &max_protohdr, 0, "");
72 SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RW, &max_hdr, 0, "");
73 SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RW,
74 &max_datalen, 0, "");
75 #ifdef MBUF_STRESS_TEST
76 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD,
77 &m_defragpackets, 0, "");
78 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD,
79 &m_defragbytes, 0, "");
80 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD,
81 &m_defraguseless, 0, "");
82 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD,
83 &m_defragfailure, 0, "");
84 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW,
85 &m_defragrandomfailures, 0, "");
86 #endif
87
88 /*
89 * Malloc-type for external ext_buf ref counts.
90 */
91 MALLOC_DEFINE(M_MBUF, "mbextcnt", "mbuf external ref counts");
92
93 /*
94 * Allocate a given length worth of mbufs and/or clusters (whatever fits
95 * best) and return a pointer to the top of the allocated chain. If an
96 * existing mbuf chain is provided, then we will append the new chain
97 * to the existing one but still return the top of the newly allocated
98 * chain.
99 */
100 struct mbuf *
101 m_getm(struct mbuf *m, int len, int how, short type)
102 {
103 struct mbuf *mb, *top, *cur, *mtail;
104 int num, rem;
105 int i;
106
107 KASSERT(len >= 0, ("m_getm(): len is < 0"));
108
109 /* If m != NULL, we will append to the end of that chain. */
110 if (m != NULL)
111 for (mtail = m; mtail->m_next != NULL; mtail = mtail->m_next);
112 else
113 mtail = NULL;
114
115 /*
116 * Calculate how many mbufs+clusters ("packets") we need and how much
117 * leftover there is after that and allocate the first mbuf+cluster
118 * if required.
119 */
120 num = len / MCLBYTES;
121 rem = len % MCLBYTES;
122 top = cur = NULL;
123 if (num > 0) {
124 if ((top = cur = m_getcl(how, type, 0)) == NULL)
125 goto failed;
126 top->m_len = 0;
127 }
128 num--;
129
130 for (i = 0; i < num; i++) {
131 mb = m_getcl(how, type, 0);
132 if (mb == NULL)
133 goto failed;
134 mb->m_len = 0;
135 cur = (cur->m_next = mb);
136 }
137 if (rem > 0) {
138 mb = (rem > MINCLSIZE) ?
139 m_getcl(how, type, 0) : m_get(how, type);
140 if (mb == NULL)
141 goto failed;
142 mb->m_len = 0;
143 if (cur == NULL)
144 top = mb;
145 else
146 cur->m_next = mb;
147 }
148
149 if (mtail != NULL)
150 mtail->m_next = top;
151 return top;
152 failed:
153 if (top != NULL)
154 m_freem(top);
155 return NULL;
156 }
157
158 /*
159 * Free an entire chain of mbufs and associated external buffers, if
160 * applicable.
161 */
162 void
163 m_freem(struct mbuf *mb)
164 {
165
166 while (mb != NULL)
167 mb = m_free(mb);
168 }
169
170 /*-
171 * Configure a provided mbuf to refer to the provided external storage
172 * buffer and setup a reference count for said buffer. If the setting
173 * up of the reference count fails, the M_EXT bit will not be set. If
174 * successfull, the M_EXT bit is set in the mbuf's flags.
175 *
176 * Arguments:
177 * mb The existing mbuf to which to attach the provided buffer.
178 * buf The address of the provided external storage buffer.
179 * size The size of the provided buffer.
180 * freef A pointer to a routine that is responsible for freeing the
181 * provided external storage buffer.
182 * args A pointer to an argument structure (of any type) to be passed
183 * to the provided freef routine (may be NULL).
184 * flags Any other flags to be passed to the provided mbuf.
185 * type The type that the external storage buffer should be
186 * labeled with.
187 *
188 * Returns:
189 * Nothing.
190 */
191 void
192 m_extadd(struct mbuf *mb, caddr_t buf, u_int size,
193 void (*freef)(void *, void *), void *args, int flags, int type)
194 {
195 u_int *ref_cnt = NULL;
196
197 /* XXX Shouldn't be adding EXT_CLUSTER with this API */
198 if (type == EXT_CLUSTER)
199 ref_cnt = (u_int *)uma_find_refcnt(zone_clust,
200 mb->m_ext.ext_buf);
201 else if (type == EXT_EXTREF)
202 ref_cnt = __DEVOLATILE(u_int *, mb->m_ext.ref_cnt);
203 mb->m_ext.ref_cnt = (ref_cnt == NULL) ?
204 malloc(sizeof(u_int), M_MBUF, M_NOWAIT) : (u_int *)ref_cnt;
205 if (mb->m_ext.ref_cnt != NULL) {
206 *(mb->m_ext.ref_cnt) = 1;
207 mb->m_flags |= (M_EXT | flags);
208 mb->m_ext.ext_buf = buf;
209 mb->m_data = mb->m_ext.ext_buf;
210 mb->m_ext.ext_size = size;
211 mb->m_ext.ext_free = freef;
212 mb->m_ext.ext_args = args;
213 mb->m_ext.ext_type = type;
214 }
215 }
216
217 /*
218 * Non-directly-exported function to clean up after mbufs with M_EXT
219 * storage attached to them if the reference count hits 0.
220 */
221 void
222 mb_free_ext(struct mbuf *m)
223 {
224 u_int cnt;
225 int dofree;
226
227 /* Account for lazy ref count assign. */
228 if (m->m_ext.ref_cnt == NULL)
229 dofree = 1;
230 else
231 dofree = 0;
232
233 /*
234 * This is tricky. We need to make sure to decrement the
235 * refcount in a safe way but to also clean up if we're the
236 * last reference. This method seems to do it without race.
237 */
238 while (dofree == 0) {
239 cnt = *(m->m_ext.ref_cnt);
240 if (atomic_cmpset_int(m->m_ext.ref_cnt, cnt, cnt - 1)) {
241 if (cnt == 1)
242 dofree = 1;
243 break;
244 }
245 }
246
247 if (dofree) {
248 /*
249 * Do the free, should be safe.
250 */
251 if (m->m_ext.ext_type == EXT_PACKET) {
252 uma_zfree(zone_pack, m);
253 return;
254 } else if (m->m_ext.ext_type == EXT_CLUSTER) {
255 uma_zfree(zone_clust, m->m_ext.ext_buf);
256 m->m_ext.ext_buf = NULL;
257 } else {
258 (*(m->m_ext.ext_free))(m->m_ext.ext_buf,
259 m->m_ext.ext_args);
260 if (m->m_ext.ext_type != EXT_EXTREF) {
261 if (m->m_ext.ref_cnt != NULL)
262 free(__DEVOLATILE(u_int *,
263 m->m_ext.ref_cnt), M_MBUF);
264 m->m_ext.ref_cnt = NULL;
265 }
266 m->m_ext.ext_buf = NULL;
267 }
268 }
269 uma_zfree(zone_mbuf, m);
270 }
271
272 /*
273 * "Move" mbuf pkthdr from "from" to "to".
274 * "from" must have M_PKTHDR set, and "to" must be empty.
275 */
276 void
277 m_move_pkthdr(struct mbuf *to, struct mbuf *from)
278 {
279
280 #if 0
281 /* see below for why these are not enabled */
282 M_ASSERTPKTHDR(to);
283 /* Note: with MAC, this may not be a good assertion. */
284 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags),
285 ("m_move_pkthdr: to has tags"));
286 #endif
287 KASSERT((to->m_flags & M_EXT) == 0, ("m_move_pkthdr: to has cluster"));
288 #ifdef MAC
289 /*
290 * XXXMAC: It could be this should also occur for non-MAC?
291 */
292 if (to->m_flags & M_PKTHDR)
293 m_tag_delete_chain(to, NULL);
294 #endif
295 to->m_flags = from->m_flags & M_COPYFLAGS;
296 to->m_data = to->m_pktdat;
297 to->m_pkthdr = from->m_pkthdr; /* especially tags */
298 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */
299 from->m_flags &= ~M_PKTHDR;
300 }
301
302 /*
303 * Duplicate "from"'s mbuf pkthdr in "to".
304 * "from" must have M_PKTHDR set, and "to" must be empty.
305 * In particular, this does a deep copy of the packet tags.
306 */
307 int
308 m_dup_pkthdr(struct mbuf *to, struct mbuf *from, int how)
309 {
310
311 #if 0
312 /*
313 * The mbuf allocator only initializes the pkthdr
314 * when the mbuf is allocated with MGETHDR. Many users
315 * (e.g. m_copy*, m_prepend) use MGET and then
316 * smash the pkthdr as needed causing these
317 * assertions to trip. For now just disable them.
318 */
319 M_ASSERTPKTHDR(to);
320 /* Note: with MAC, this may not be a good assertion. */
321 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), ("m_dup_pkthdr: to has tags"));
322 #endif
323 MBUF_CHECKSLEEP(how);
324 #ifdef MAC
325 if (to->m_flags & M_PKTHDR)
326 m_tag_delete_chain(to, NULL);
327 #endif
328 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT);
329 if ((to->m_flags & M_EXT) == 0)
330 to->m_data = to->m_pktdat;
331 to->m_pkthdr = from->m_pkthdr;
332 SLIST_INIT(&to->m_pkthdr.tags);
333 return (m_tag_copy_chain(to, from, MBTOM(how)));
334 }
335
336 /*
337 * Lesser-used path for M_PREPEND:
338 * allocate new mbuf to prepend to chain,
339 * copy junk along.
340 */
341 struct mbuf *
342 m_prepend(struct mbuf *m, int len, int how)
343 {
344 struct mbuf *mn;
345
346 if (m->m_flags & M_PKTHDR)
347 MGETHDR(mn, how, m->m_type);
348 else
349 MGET(mn, how, m->m_type);
350 if (mn == NULL) {
351 m_freem(m);
352 return (NULL);
353 }
354 if (m->m_flags & M_PKTHDR)
355 M_MOVE_PKTHDR(mn, m);
356 mn->m_next = m;
357 m = mn;
358 if (len < MHLEN)
359 MH_ALIGN(m, len);
360 m->m_len = len;
361 return (m);
362 }
363
364 /*
365 * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
366 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf.
367 * The wait parameter is a choice of M_TRYWAIT/M_DONTWAIT from caller.
368 * Note that the copy is read-only, because clusters are not copied,
369 * only their reference counts are incremented.
370 */
371 struct mbuf *
372 m_copym(struct mbuf *m, int off0, int len, int wait)
373 {
374 struct mbuf *n, **np;
375 int off = off0;
376 struct mbuf *top;
377 int copyhdr = 0;
378
379 KASSERT(off >= 0, ("m_copym, negative off %d", off));
380 KASSERT(len >= 0, ("m_copym, negative len %d", len));
381 MBUF_CHECKSLEEP(wait);
382 if (off == 0 && m->m_flags & M_PKTHDR)
383 copyhdr = 1;
384 while (off > 0) {
385 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain"));
386 if (off < m->m_len)
387 break;
388 off -= m->m_len;
389 m = m->m_next;
390 }
391 np = ⊤
392 top = 0;
393 while (len > 0) {
394 if (m == NULL) {
395 KASSERT(len == M_COPYALL,
396 ("m_copym, length > size of mbuf chain"));
397 break;
398 }
399 if (copyhdr)
400 MGETHDR(n, wait, m->m_type);
401 else
402 MGET(n, wait, m->m_type);
403 *np = n;
404 if (n == NULL)
405 goto nospace;
406 if (copyhdr) {
407 if (!m_dup_pkthdr(n, m, wait))
408 goto nospace;
409 if (len == M_COPYALL)
410 n->m_pkthdr.len -= off0;
411 else
412 n->m_pkthdr.len = len;
413 copyhdr = 0;
414 }
415 n->m_len = min(len, m->m_len - off);
416 if (m->m_flags & M_EXT) {
417 n->m_data = m->m_data + off;
418 n->m_ext = m->m_ext;
419 n->m_flags |= M_EXT;
420 MEXT_ADD_REF(m);
421 n->m_ext.ref_cnt = m->m_ext.ref_cnt;
422 } else
423 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
424 (u_int)n->m_len);
425 if (len != M_COPYALL)
426 len -= n->m_len;
427 off = 0;
428 m = m->m_next;
429 np = &n->m_next;
430 }
431 if (top == NULL)
432 mbstat.m_mcfail++; /* XXX: No consistency. */
433
434 return (top);
435 nospace:
436 m_freem(top);
437 mbstat.m_mcfail++; /* XXX: No consistency. */
438 return (NULL);
439 }
440
441 /*
442 * Copy an entire packet, including header (which must be present).
443 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
444 * Note that the copy is read-only, because clusters are not copied,
445 * only their reference counts are incremented.
446 * Preserve alignment of the first mbuf so if the creator has left
447 * some room at the beginning (e.g. for inserting protocol headers)
448 * the copies still have the room available.
449 */
450 struct mbuf *
451 m_copypacket(struct mbuf *m, int how)
452 {
453 struct mbuf *top, *n, *o;
454
455 MBUF_CHECKSLEEP(how);
456 MGET(n, how, m->m_type);
457 top = n;
458 if (n == NULL)
459 goto nospace;
460
461 if (!m_dup_pkthdr(n, m, how))
462 goto nospace;
463 n->m_len = m->m_len;
464 if (m->m_flags & M_EXT) {
465 n->m_data = m->m_data;
466 n->m_ext = m->m_ext;
467 n->m_flags |= M_EXT;
468 MEXT_ADD_REF(m);
469 n->m_ext.ref_cnt = m->m_ext.ref_cnt;
470 } else {
471 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat );
472 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
473 }
474
475 m = m->m_next;
476 while (m) {
477 MGET(o, how, m->m_type);
478 if (o == NULL)
479 goto nospace;
480
481 n->m_next = o;
482 n = n->m_next;
483
484 n->m_len = m->m_len;
485 if (m->m_flags & M_EXT) {
486 n->m_data = m->m_data;
487 n->m_ext = m->m_ext;
488 n->m_flags |= M_EXT;
489 MEXT_ADD_REF(m);
490 n->m_ext.ref_cnt = m->m_ext.ref_cnt;
491 } else {
492 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
493 }
494
495 m = m->m_next;
496 }
497 return top;
498 nospace:
499 m_freem(top);
500 mbstat.m_mcfail++; /* XXX: No consistency. */
501 return (NULL);
502 }
503
504 /*
505 * Copy data from an mbuf chain starting "off" bytes from the beginning,
506 * continuing for "len" bytes, into the indicated buffer.
507 */
508 void
509 m_copydata(const struct mbuf *m, int off, int len, caddr_t cp)
510 {
511 u_int count;
512
513 KASSERT(off >= 0, ("m_copydata, negative off %d", off));
514 KASSERT(len >= 0, ("m_copydata, negative len %d", len));
515 while (off > 0) {
516 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain"));
517 if (off < m->m_len)
518 break;
519 off -= m->m_len;
520 m = m->m_next;
521 }
522 while (len > 0) {
523 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain"));
524 count = min(m->m_len - off, len);
525 bcopy(mtod(m, caddr_t) + off, cp, count);
526 len -= count;
527 cp += count;
528 off = 0;
529 m = m->m_next;
530 }
531 }
532
533 /*
534 * Copy a packet header mbuf chain into a completely new chain, including
535 * copying any mbuf clusters. Use this instead of m_copypacket() when
536 * you need a writable copy of an mbuf chain.
537 */
538 struct mbuf *
539 m_dup(struct mbuf *m, int how)
540 {
541 struct mbuf **p, *top = NULL;
542 int remain, moff, nsize;
543
544 MBUF_CHECKSLEEP(how);
545 /* Sanity check */
546 if (m == NULL)
547 return (NULL);
548 M_ASSERTPKTHDR(m);
549
550 /* While there's more data, get a new mbuf, tack it on, and fill it */
551 remain = m->m_pkthdr.len;
552 moff = 0;
553 p = ⊤
554 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */
555 struct mbuf *n;
556
557 /* Get the next new mbuf */
558 if (remain >= MINCLSIZE) {
559 n = m_getcl(how, m->m_type, 0);
560 nsize = MCLBYTES;
561 } else {
562 n = m_get(how, m->m_type);
563 nsize = MLEN;
564 }
565 if (n == NULL)
566 goto nospace;
567
568 if (top == NULL) { /* First one, must be PKTHDR */
569 if (!m_dup_pkthdr(n, m, how)) {
570 m_free(n);
571 goto nospace;
572 }
573 if ((n->m_flags & M_EXT) == 0)
574 nsize = MHLEN;
575 }
576 n->m_len = 0;
577
578 /* Link it into the new chain */
579 *p = n;
580 p = &n->m_next;
581
582 /* Copy data from original mbuf(s) into new mbuf */
583 while (n->m_len < nsize && m != NULL) {
584 int chunk = min(nsize - n->m_len, m->m_len - moff);
585
586 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
587 moff += chunk;
588 n->m_len += chunk;
589 remain -= chunk;
590 if (moff == m->m_len) {
591 m = m->m_next;
592 moff = 0;
593 }
594 }
595
596 /* Check correct total mbuf length */
597 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL),
598 ("%s: bogus m_pkthdr.len", __func__));
599 }
600 return (top);
601
602 nospace:
603 m_freem(top);
604 mbstat.m_mcfail++; /* XXX: No consistency. */
605 return (NULL);
606 }
607
608 /*
609 * Concatenate mbuf chain n to m.
610 * Both chains must be of the same type (e.g. MT_DATA).
611 * Any m_pkthdr is not updated.
612 */
613 void
614 m_cat(struct mbuf *m, struct mbuf *n)
615 {
616 while (m->m_next)
617 m = m->m_next;
618 while (n) {
619 if (m->m_flags & M_EXT ||
620 m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) {
621 /* just join the two chains */
622 m->m_next = n;
623 return;
624 }
625 /* splat the data from one into the other */
626 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
627 (u_int)n->m_len);
628 m->m_len += n->m_len;
629 n = m_free(n);
630 }
631 }
632
633 void
634 m_adj(struct mbuf *mp, int req_len)
635 {
636 int len = req_len;
637 struct mbuf *m;
638 int count;
639
640 if ((m = mp) == NULL)
641 return;
642 if (len >= 0) {
643 /*
644 * Trim from head.
645 */
646 while (m != NULL && len > 0) {
647 if (m->m_len <= len) {
648 len -= m->m_len;
649 m->m_len = 0;
650 m = m->m_next;
651 } else {
652 m->m_len -= len;
653 m->m_data += len;
654 len = 0;
655 }
656 }
657 m = mp;
658 if (mp->m_flags & M_PKTHDR)
659 m->m_pkthdr.len -= (req_len - len);
660 } else {
661 /*
662 * Trim from tail. Scan the mbuf chain,
663 * calculating its length and finding the last mbuf.
664 * If the adjustment only affects this mbuf, then just
665 * adjust and return. Otherwise, rescan and truncate
666 * after the remaining size.
667 */
668 len = -len;
669 count = 0;
670 for (;;) {
671 count += m->m_len;
672 if (m->m_next == (struct mbuf *)0)
673 break;
674 m = m->m_next;
675 }
676 if (m->m_len >= len) {
677 m->m_len -= len;
678 if (mp->m_flags & M_PKTHDR)
679 mp->m_pkthdr.len -= len;
680 return;
681 }
682 count -= len;
683 if (count < 0)
684 count = 0;
685 /*
686 * Correct length for chain is "count".
687 * Find the mbuf with last data, adjust its length,
688 * and toss data from remaining mbufs on chain.
689 */
690 m = mp;
691 if (m->m_flags & M_PKTHDR)
692 m->m_pkthdr.len = count;
693 for (; m; m = m->m_next) {
694 if (m->m_len >= count) {
695 m->m_len = count;
696 break;
697 }
698 count -= m->m_len;
699 }
700 while (m->m_next)
701 (m = m->m_next) ->m_len = 0;
702 }
703 }
704
705 /*
706 * Rearange an mbuf chain so that len bytes are contiguous
707 * and in the data area of an mbuf (so that mtod and dtom
708 * will work for a structure of size len). Returns the resulting
709 * mbuf chain on success, frees it and returns null on failure.
710 * If there is room, it will add up to max_protohdr-len extra bytes to the
711 * contiguous region in an attempt to avoid being called next time.
712 */
713 struct mbuf *
714 m_pullup(struct mbuf *n, int len)
715 {
716 struct mbuf *m;
717 int count;
718 int space;
719
720 /*
721 * If first mbuf has no cluster, and has room for len bytes
722 * without shifting current data, pullup into it,
723 * otherwise allocate a new mbuf to prepend to the chain.
724 */
725 if ((n->m_flags & M_EXT) == 0 &&
726 n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
727 if (n->m_len >= len)
728 return (n);
729 m = n;
730 n = n->m_next;
731 len -= m->m_len;
732 } else {
733 if (len > MHLEN)
734 goto bad;
735 MGET(m, M_DONTWAIT, n->m_type);
736 if (m == NULL)
737 goto bad;
738 m->m_len = 0;
739 if (n->m_flags & M_PKTHDR)
740 M_MOVE_PKTHDR(m, n);
741 }
742 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
743 do {
744 count = min(min(max(len, max_protohdr), space), n->m_len);
745 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
746 (u_int)count);
747 len -= count;
748 m->m_len += count;
749 n->m_len -= count;
750 space -= count;
751 if (n->m_len)
752 n->m_data += count;
753 else
754 n = m_free(n);
755 } while (len > 0 && n);
756 if (len > 0) {
757 (void) m_free(m);
758 goto bad;
759 }
760 m->m_next = n;
761 return (m);
762 bad:
763 m_freem(n);
764 mbstat.m_mpfail++; /* XXX: No consistency. */
765 return (NULL);
766 }
767
768 /*
769 * Like m_pullup(), except a new mbuf is always allocated, and we allow
770 * the amount of empty space before the data in the new mbuf to be specified
771 * (in the event that the caller expects to prepend later).
772 */
773 int MSFail;
774
775 struct mbuf *
776 m_copyup(struct mbuf *n, int len, int dstoff)
777 {
778 struct mbuf *m;
779 int count, space;
780
781 if (len > (MHLEN - dstoff))
782 goto bad;
783 MGET(m, M_DONTWAIT, n->m_type);
784 if (m == NULL)
785 goto bad;
786 m->m_len = 0;
787 if (n->m_flags & M_PKTHDR)
788 M_MOVE_PKTHDR(m, n);
789 m->m_data += dstoff;
790 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
791 do {
792 count = min(min(max(len, max_protohdr), space), n->m_len);
793 memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
794 (unsigned)count);
795 len -= count;
796 m->m_len += count;
797 n->m_len -= count;
798 space -= count;
799 if (n->m_len)
800 n->m_data += count;
801 else
802 n = m_free(n);
803 } while (len > 0 && n);
804 if (len > 0) {
805 (void) m_free(m);
806 goto bad;
807 }
808 m->m_next = n;
809 return (m);
810 bad:
811 m_freem(n);
812 MSFail++;
813 return (NULL);
814 }
815
816 /*
817 * Partition an mbuf chain in two pieces, returning the tail --
818 * all but the first len0 bytes. In case of failure, it returns NULL and
819 * attempts to restore the chain to its original state.
820 *
821 * Note that the resulting mbufs might be read-only, because the new
822 * mbuf can end up sharing an mbuf cluster with the original mbuf if
823 * the "breaking point" happens to lie within a cluster mbuf. Use the
824 * M_WRITABLE() macro to check for this case.
825 */
826 struct mbuf *
827 m_split(struct mbuf *m0, int len0, int wait)
828 {
829 struct mbuf *m, *n;
830 u_int len = len0, remain;
831
832 MBUF_CHECKSLEEP(wait);
833 for (m = m0; m && len > m->m_len; m = m->m_next)
834 len -= m->m_len;
835 if (m == NULL)
836 return (NULL);
837 remain = m->m_len - len;
838 if (m0->m_flags & M_PKTHDR) {
839 MGETHDR(n, wait, m0->m_type);
840 if (n == NULL)
841 return (NULL);
842 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
843 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
844 m0->m_pkthdr.len = len0;
845 if (m->m_flags & M_EXT)
846 goto extpacket;
847 if (remain > MHLEN) {
848 /* m can't be the lead packet */
849 MH_ALIGN(n, 0);
850 n->m_next = m_split(m, len, wait);
851 if (n->m_next == NULL) {
852 (void) m_free(n);
853 return (NULL);
854 } else {
855 n->m_len = 0;
856 return (n);
857 }
858 } else
859 MH_ALIGN(n, remain);
860 } else if (remain == 0) {
861 n = m->m_next;
862 m->m_next = NULL;
863 return (n);
864 } else {
865 MGET(n, wait, m->m_type);
866 if (n == NULL)
867 return (NULL);
868 M_ALIGN(n, remain);
869 }
870 extpacket:
871 if (m->m_flags & M_EXT) {
872 n->m_flags |= M_EXT;
873 n->m_ext = m->m_ext;
874 MEXT_ADD_REF(m);
875 n->m_ext.ref_cnt = m->m_ext.ref_cnt;
876 n->m_data = m->m_data + len;
877 } else {
878 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
879 }
880 n->m_len = remain;
881 m->m_len = len;
882 n->m_next = m->m_next;
883 m->m_next = NULL;
884 return (n);
885 }
886 /*
887 * Routine to copy from device local memory into mbufs.
888 * Note that `off' argument is offset into first mbuf of target chain from
889 * which to begin copying the data to.
890 */
891 struct mbuf *
892 m_devget(char *buf, int totlen, int off, struct ifnet *ifp,
893 void (*copy)(char *from, caddr_t to, u_int len))
894 {
895 struct mbuf *m;
896 struct mbuf *top = NULL, **mp = ⊤
897 int len;
898
899 if (off < 0 || off > MHLEN)
900 return (NULL);
901
902 while (totlen > 0) {
903 if (top == NULL) { /* First one, must be PKTHDR */
904 if (totlen + off >= MINCLSIZE) {
905 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
906 len = MCLBYTES;
907 } else {
908 m = m_gethdr(M_DONTWAIT, MT_DATA);
909 len = MHLEN;
910
911 /* Place initial small packet/header at end of mbuf */
912 if (m && totlen + off + max_linkhdr <= MLEN) {
913 m->m_data += max_linkhdr;
914 len -= max_linkhdr;
915 }
916 }
917 if (m == NULL)
918 return NULL;
919 m->m_pkthdr.rcvif = ifp;
920 m->m_pkthdr.len = totlen;
921 } else {
922 if (totlen + off >= MINCLSIZE) {
923 m = m_getcl(M_DONTWAIT, MT_DATA, 0);
924 len = MCLBYTES;
925 } else {
926 m = m_get(M_DONTWAIT, MT_DATA);
927 len = MLEN;
928 }
929 if (m == NULL) {
930 m_freem(top);
931 return NULL;
932 }
933 }
934 if (off) {
935 m->m_data += off;
936 len -= off;
937 off = 0;
938 }
939 m->m_len = len = min(totlen, len);
940 if (copy)
941 copy(buf, mtod(m, caddr_t), (u_int)len);
942 else
943 bcopy(buf, mtod(m, caddr_t), (u_int)len);
944 buf += len;
945 *mp = m;
946 mp = &m->m_next;
947 totlen -= len;
948 }
949 return (top);
950 }
951
952 /*
953 * Copy data from a buffer back into the indicated mbuf chain,
954 * starting "off" bytes from the beginning, extending the mbuf
955 * chain if necessary.
956 */
957 void
958 m_copyback(struct mbuf *m0, int off, int len, c_caddr_t cp)
959 {
960 int mlen;
961 struct mbuf *m = m0, *n;
962 int totlen = 0;
963
964 if (m0 == NULL)
965 return;
966 while (off > (mlen = m->m_len)) {
967 off -= mlen;
968 totlen += mlen;
969 if (m->m_next == NULL) {
970 n = m_get(M_DONTWAIT, m->m_type);
971 if (n == NULL)
972 goto out;
973 bzero(mtod(n, caddr_t), MLEN);
974 n->m_len = min(MLEN, len + off);
975 m->m_next = n;
976 }
977 m = m->m_next;
978 }
979 while (len > 0) {
980 mlen = min (m->m_len - off, len);
981 bcopy(cp, off + mtod(m, caddr_t), (u_int)mlen);
982 cp += mlen;
983 len -= mlen;
984 mlen += off;
985 off = 0;
986 totlen += mlen;
987 if (len == 0)
988 break;
989 if (m->m_next == NULL) {
990 n = m_get(M_DONTWAIT, m->m_type);
991 if (n == NULL)
992 break;
993 n->m_len = min(MLEN, len);
994 m->m_next = n;
995 }
996 m = m->m_next;
997 }
998 out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
999 m->m_pkthdr.len = totlen;
1000 }
1001
1002 /*
1003 * Apply function f to the data in an mbuf chain starting "off" bytes from
1004 * the beginning, continuing for "len" bytes.
1005 */
1006 int
1007 m_apply(struct mbuf *m, int off, int len,
1008 int (*f)(void *, void *, u_int), void *arg)
1009 {
1010 u_int count;
1011 int rval;
1012
1013 KASSERT(off >= 0, ("m_apply, negative off %d", off));
1014 KASSERT(len >= 0, ("m_apply, negative len %d", len));
1015 while (off > 0) {
1016 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1017 if (off < m->m_len)
1018 break;
1019 off -= m->m_len;
1020 m = m->m_next;
1021 }
1022 while (len > 0) {
1023 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1024 count = min(m->m_len - off, len);
1025 rval = (*f)(arg, mtod(m, caddr_t) + off, count);
1026 if (rval)
1027 return (rval);
1028 len -= count;
1029 off = 0;
1030 m = m->m_next;
1031 }
1032 return (0);
1033 }
1034
1035 /*
1036 * Return a pointer to mbuf/offset of location in mbuf chain.
1037 */
1038 struct mbuf *
1039 m_getptr(struct mbuf *m, int loc, int *off)
1040 {
1041
1042 while (loc >= 0) {
1043 /* Normal end of search. */
1044 if (m->m_len > loc) {
1045 *off = loc;
1046 return (m);
1047 } else {
1048 loc -= m->m_len;
1049 if (m->m_next == NULL) {
1050 if (loc == 0) {
1051 /* Point at the end of valid data. */
1052 *off = m->m_len;
1053 return (m);
1054 }
1055 return (NULL);
1056 }
1057 m = m->m_next;
1058 }
1059 }
1060 return (NULL);
1061 }
1062
1063 void
1064 m_print(const struct mbuf *m, int maxlen)
1065 {
1066 int len;
1067 int pdata;
1068 const struct mbuf *m2;
1069
1070 if (m->m_flags & M_PKTHDR)
1071 len = m->m_pkthdr.len;
1072 else
1073 len = -1;
1074 m2 = m;
1075 while (m2 != NULL && (len == -1 || len)) {
1076 pdata = m2->m_len;
1077 if (maxlen != -1 && pdata > maxlen)
1078 pdata = maxlen;
1079 printf("mbuf: %p len: %d, next: %p, %b%s", m2, m2->m_len,
1080 m2->m_next, m2->m_flags, "\2\20freelist\17skipfw"
1081 "\11proto5\10proto4\7proto3\6proto2\5proto1\4rdonly"
1082 "\3eor\2pkthdr\1ext", pdata ? "" : "\n");
1083 if (pdata)
1084 printf(", %*D\n", m2->m_len, (u_char *)m2->m_data, "-");
1085 if (len != -1)
1086 len -= m2->m_len;
1087 m2 = m2->m_next;
1088 }
1089 if (len > 0)
1090 printf("%d bytes unaccounted for.\n", len);
1091 return;
1092 }
1093
1094 u_int
1095 m_fixhdr(struct mbuf *m0)
1096 {
1097 u_int len;
1098
1099 len = m_length(m0, NULL);
1100 m0->m_pkthdr.len = len;
1101 return (len);
1102 }
1103
1104 u_int
1105 m_length(struct mbuf *m0, struct mbuf **last)
1106 {
1107 struct mbuf *m;
1108 u_int len;
1109
1110 len = 0;
1111 for (m = m0; m != NULL; m = m->m_next) {
1112 len += m->m_len;
1113 if (m->m_next == NULL)
1114 break;
1115 }
1116 if (last != NULL)
1117 *last = m;
1118 return (len);
1119 }
1120
1121 /*
1122 * Defragment a mbuf chain, returning the shortest possible
1123 * chain of mbufs and clusters. If allocation fails and
1124 * this cannot be completed, NULL will be returned, but
1125 * the passed in chain will be unchanged. Upon success,
1126 * the original chain will be freed, and the new chain
1127 * will be returned.
1128 *
1129 * If a non-packet header is passed in, the original
1130 * mbuf (chain?) will be returned unharmed.
1131 */
1132 struct mbuf *
1133 m_defrag(struct mbuf *m0, int how)
1134 {
1135 struct mbuf *m_new = NULL, *m_final = NULL;
1136 int progress = 0, length;
1137
1138 MBUF_CHECKSLEEP(how);
1139 if (!(m0->m_flags & M_PKTHDR))
1140 return (m0);
1141
1142 m_fixhdr(m0); /* Needed sanity check */
1143
1144 #ifdef MBUF_STRESS_TEST
1145 if (m_defragrandomfailures) {
1146 int temp = arc4random() & 0xff;
1147 if (temp == 0xba)
1148 goto nospace;
1149 }
1150 #endif
1151
1152 if (m0->m_pkthdr.len > MHLEN)
1153 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1154 else
1155 m_final = m_gethdr(how, MT_DATA);
1156
1157 if (m_final == NULL)
1158 goto nospace;
1159
1160 if (m_dup_pkthdr(m_final, m0, how) == 0)
1161 goto nospace;
1162
1163 m_new = m_final;
1164
1165 while (progress < m0->m_pkthdr.len) {
1166 length = m0->m_pkthdr.len - progress;
1167 if (length > MCLBYTES)
1168 length = MCLBYTES;
1169
1170 if (m_new == NULL) {
1171 if (length > MLEN)
1172 m_new = m_getcl(how, MT_DATA, 0);
1173 else
1174 m_new = m_get(how, MT_DATA);
1175 if (m_new == NULL)
1176 goto nospace;
1177 }
1178
1179 m_copydata(m0, progress, length, mtod(m_new, caddr_t));
1180 progress += length;
1181 m_new->m_len = length;
1182 if (m_new != m_final)
1183 m_cat(m_final, m_new);
1184 m_new = NULL;
1185 }
1186 #ifdef MBUF_STRESS_TEST
1187 if (m0->m_next == NULL)
1188 m_defraguseless++;
1189 #endif
1190 m_freem(m0);
1191 m0 = m_final;
1192 #ifdef MBUF_STRESS_TEST
1193 m_defragpackets++;
1194 m_defragbytes += m0->m_pkthdr.len;
1195 #endif
1196 return (m0);
1197 nospace:
1198 #ifdef MBUF_STRESS_TEST
1199 m_defragfailure++;
1200 #endif
1201 if (m_new)
1202 m_free(m_new);
1203 if (m_final)
1204 m_freem(m_final);
1205 return (NULL);
1206 }
1207
1208 #ifdef MBUF_STRESS_TEST
1209
1210 /*
1211 * Fragment an mbuf chain. There's no reason you'd ever want to do
1212 * this in normal usage, but it's great for stress testing various
1213 * mbuf consumers.
1214 *
1215 * If fragmentation is not possible, the original chain will be
1216 * returned.
1217 *
1218 * Possible length values:
1219 * 0 no fragmentation will occur
1220 * > 0 each fragment will be of the specified length
1221 * -1 each fragment will be the same random value in length
1222 * -2 each fragment's length will be entirely random
1223 * (Random values range from 1 to 256)
1224 */
1225 struct mbuf *
1226 m_fragment(struct mbuf *m0, int how, int length)
1227 {
1228 struct mbuf *m_new = NULL, *m_final = NULL;
1229 int progress = 0;
1230
1231 if (!(m0->m_flags & M_PKTHDR))
1232 return (m0);
1233
1234 if ((length == 0) || (length < -2))
1235 return (m0);
1236
1237 m_fixhdr(m0); /* Needed sanity check */
1238
1239 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1240
1241 if (m_final == NULL)
1242 goto nospace;
1243
1244 if (m_dup_pkthdr(m_final, m0, how) == 0)
1245 goto nospace;
1246
1247 m_new = m_final;
1248
1249 if (length == -1)
1250 length = 1 + (arc4random() & 255);
1251
1252 while (progress < m0->m_pkthdr.len) {
1253 int fraglen;
1254
1255 if (length > 0)
1256 fraglen = length;
1257 else
1258 fraglen = 1 + (arc4random() & 255);
1259 if (fraglen > m0->m_pkthdr.len - progress)
1260 fraglen = m0->m_pkthdr.len - progress;
1261
1262 if (fraglen > MCLBYTES)
1263 fraglen = MCLBYTES;
1264
1265 if (m_new == NULL) {
1266 m_new = m_getcl(how, MT_DATA, 0);
1267 if (m_new == NULL)
1268 goto nospace;
1269 }
1270
1271 m_copydata(m0, progress, fraglen, mtod(m_new, caddr_t));
1272 progress += fraglen;
1273 m_new->m_len = fraglen;
1274 if (m_new != m_final)
1275 m_cat(m_final, m_new);
1276 m_new = NULL;
1277 }
1278 m_freem(m0);
1279 m0 = m_final;
1280 return (m0);
1281 nospace:
1282 if (m_new)
1283 m_free(m_new);
1284 if (m_final)
1285 m_freem(m_final);
1286 /* Return the original chain on failure */
1287 return (m0);
1288 }
1289
1290 #endif
1291
1292 struct mbuf *
1293 m_uiotombuf(struct uio *uio, int how, int len, int align)
1294 {
1295 struct mbuf *m_new = NULL, *m_final = NULL;
1296 int progress = 0, error = 0, length, total;
1297
1298 if (len > 0)
1299 total = min(uio->uio_resid, len);
1300 else
1301 total = uio->uio_resid;
1302 if (align >= MHLEN)
1303 goto nospace;
1304 if (total + align > MHLEN)
1305 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1306 else
1307 m_final = m_gethdr(how, MT_DATA);
1308 if (m_final == NULL)
1309 goto nospace;
1310 m_final->m_data += align;
1311 m_new = m_final;
1312 while (progress < total) {
1313 length = total - progress;
1314 if (length > MCLBYTES)
1315 length = MCLBYTES;
1316 if (m_new == NULL) {
1317 if (length > MLEN)
1318 m_new = m_getcl(how, MT_DATA, 0);
1319 else
1320 m_new = m_get(how, MT_DATA);
1321 if (m_new == NULL)
1322 goto nospace;
1323 }
1324 error = uiomove(mtod(m_new, void *), length, uio);
1325 if (error)
1326 goto nospace;
1327 progress += length;
1328 m_new->m_len = length;
1329 if (m_new != m_final)
1330 m_cat(m_final, m_new);
1331 m_new = NULL;
1332 }
1333 m_fixhdr(m_final);
1334 return (m_final);
1335 nospace:
1336 if (m_new)
1337 m_free(m_new);
1338 if (m_final)
1339 m_freem(m_final);
1340 return (NULL);
1341 }
Cache object: 6d443537b417cb94dbb9611d64d47583
|