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: releng/5.4/sys/kern/uipc_mbuf.c 145335 2005-04-20 19:11:07Z cvs2svn $");
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 nsize = MHLEN;
574 }
575 n->m_len = 0;
576
577 /* Link it into the new chain */
578 *p = n;
579 p = &n->m_next;
580
581 /* Copy data from original mbuf(s) into new mbuf */
582 while (n->m_len < nsize && m != NULL) {
583 int chunk = min(nsize - n->m_len, m->m_len - moff);
584
585 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
586 moff += chunk;
587 n->m_len += chunk;
588 remain -= chunk;
589 if (moff == m->m_len) {
590 m = m->m_next;
591 moff = 0;
592 }
593 }
594
595 /* Check correct total mbuf length */
596 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL),
597 ("%s: bogus m_pkthdr.len", __func__));
598 }
599 return (top);
600
601 nospace:
602 m_freem(top);
603 mbstat.m_mcfail++; /* XXX: No consistency. */
604 return (NULL);
605 }
606
607 /*
608 * Concatenate mbuf chain n to m.
609 * Both chains must be of the same type (e.g. MT_DATA).
610 * Any m_pkthdr is not updated.
611 */
612 void
613 m_cat(struct mbuf *m, struct mbuf *n)
614 {
615 while (m->m_next)
616 m = m->m_next;
617 while (n) {
618 if (m->m_flags & M_EXT ||
619 m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) {
620 /* just join the two chains */
621 m->m_next = n;
622 return;
623 }
624 /* splat the data from one into the other */
625 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
626 (u_int)n->m_len);
627 m->m_len += n->m_len;
628 n = m_free(n);
629 }
630 }
631
632 void
633 m_adj(struct mbuf *mp, int req_len)
634 {
635 int len = req_len;
636 struct mbuf *m;
637 int count;
638
639 if ((m = mp) == NULL)
640 return;
641 if (len >= 0) {
642 /*
643 * Trim from head.
644 */
645 while (m != NULL && len > 0) {
646 if (m->m_len <= len) {
647 len -= m->m_len;
648 m->m_len = 0;
649 m = m->m_next;
650 } else {
651 m->m_len -= len;
652 m->m_data += len;
653 len = 0;
654 }
655 }
656 m = mp;
657 if (mp->m_flags & M_PKTHDR)
658 m->m_pkthdr.len -= (req_len - len);
659 } else {
660 /*
661 * Trim from tail. Scan the mbuf chain,
662 * calculating its length and finding the last mbuf.
663 * If the adjustment only affects this mbuf, then just
664 * adjust and return. Otherwise, rescan and truncate
665 * after the remaining size.
666 */
667 len = -len;
668 count = 0;
669 for (;;) {
670 count += m->m_len;
671 if (m->m_next == (struct mbuf *)0)
672 break;
673 m = m->m_next;
674 }
675 if (m->m_len >= len) {
676 m->m_len -= len;
677 if (mp->m_flags & M_PKTHDR)
678 mp->m_pkthdr.len -= len;
679 return;
680 }
681 count -= len;
682 if (count < 0)
683 count = 0;
684 /*
685 * Correct length for chain is "count".
686 * Find the mbuf with last data, adjust its length,
687 * and toss data from remaining mbufs on chain.
688 */
689 m = mp;
690 if (m->m_flags & M_PKTHDR)
691 m->m_pkthdr.len = count;
692 for (; m; m = m->m_next) {
693 if (m->m_len >= count) {
694 m->m_len = count;
695 break;
696 }
697 count -= m->m_len;
698 }
699 while (m->m_next)
700 (m = m->m_next) ->m_len = 0;
701 }
702 }
703
704 /*
705 * Rearange an mbuf chain so that len bytes are contiguous
706 * and in the data area of an mbuf (so that mtod and dtom
707 * will work for a structure of size len). Returns the resulting
708 * mbuf chain on success, frees it and returns null on failure.
709 * If there is room, it will add up to max_protohdr-len extra bytes to the
710 * contiguous region in an attempt to avoid being called next time.
711 */
712 struct mbuf *
713 m_pullup(struct mbuf *n, int len)
714 {
715 struct mbuf *m;
716 int count;
717 int space;
718
719 /*
720 * If first mbuf has no cluster, and has room for len bytes
721 * without shifting current data, pullup into it,
722 * otherwise allocate a new mbuf to prepend to the chain.
723 */
724 if ((n->m_flags & M_EXT) == 0 &&
725 n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
726 if (n->m_len >= len)
727 return (n);
728 m = n;
729 n = n->m_next;
730 len -= m->m_len;
731 } else {
732 if (len > MHLEN)
733 goto bad;
734 MGET(m, M_DONTWAIT, n->m_type);
735 if (m == NULL)
736 goto bad;
737 m->m_len = 0;
738 if (n->m_flags & M_PKTHDR)
739 M_MOVE_PKTHDR(m, n);
740 }
741 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
742 do {
743 count = min(min(max(len, max_protohdr), space), n->m_len);
744 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
745 (u_int)count);
746 len -= count;
747 m->m_len += count;
748 n->m_len -= count;
749 space -= count;
750 if (n->m_len)
751 n->m_data += count;
752 else
753 n = m_free(n);
754 } while (len > 0 && n);
755 if (len > 0) {
756 (void) m_free(m);
757 goto bad;
758 }
759 m->m_next = n;
760 return (m);
761 bad:
762 m_freem(n);
763 mbstat.m_mpfail++; /* XXX: No consistency. */
764 return (NULL);
765 }
766
767 /*
768 * Partition an mbuf chain in two pieces, returning the tail --
769 * all but the first len0 bytes. In case of failure, it returns NULL and
770 * attempts to restore the chain to its original state.
771 *
772 * Note that the resulting mbufs might be read-only, because the new
773 * mbuf can end up sharing an mbuf cluster with the original mbuf if
774 * the "breaking point" happens to lie within a cluster mbuf. Use the
775 * M_WRITABLE() macro to check for this case.
776 */
777 struct mbuf *
778 m_split(struct mbuf *m0, int len0, int wait)
779 {
780 struct mbuf *m, *n;
781 u_int len = len0, remain;
782
783 MBUF_CHECKSLEEP(wait);
784 for (m = m0; m && len > m->m_len; m = m->m_next)
785 len -= m->m_len;
786 if (m == NULL)
787 return (NULL);
788 remain = m->m_len - len;
789 if (m0->m_flags & M_PKTHDR) {
790 MGETHDR(n, wait, m0->m_type);
791 if (n == NULL)
792 return (NULL);
793 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
794 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
795 m0->m_pkthdr.len = len0;
796 if (m->m_flags & M_EXT)
797 goto extpacket;
798 if (remain > MHLEN) {
799 /* m can't be the lead packet */
800 MH_ALIGN(n, 0);
801 n->m_next = m_split(m, len, wait);
802 if (n->m_next == NULL) {
803 (void) m_free(n);
804 return (NULL);
805 } else {
806 n->m_len = 0;
807 return (n);
808 }
809 } else
810 MH_ALIGN(n, remain);
811 } else if (remain == 0) {
812 n = m->m_next;
813 m->m_next = NULL;
814 return (n);
815 } else {
816 MGET(n, wait, m->m_type);
817 if (n == NULL)
818 return (NULL);
819 M_ALIGN(n, remain);
820 }
821 extpacket:
822 if (m->m_flags & M_EXT) {
823 n->m_flags |= M_EXT;
824 n->m_ext = m->m_ext;
825 MEXT_ADD_REF(m);
826 n->m_ext.ref_cnt = m->m_ext.ref_cnt;
827 n->m_data = m->m_data + len;
828 } else {
829 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
830 }
831 n->m_len = remain;
832 m->m_len = len;
833 n->m_next = m->m_next;
834 m->m_next = NULL;
835 return (n);
836 }
837 /*
838 * Routine to copy from device local memory into mbufs.
839 * Note that `off' argument is offset into first mbuf of target chain from
840 * which to begin copying the data to.
841 */
842 struct mbuf *
843 m_devget(char *buf, int totlen, int off, struct ifnet *ifp,
844 void (*copy)(char *from, caddr_t to, u_int len))
845 {
846 struct mbuf *m;
847 struct mbuf *top = NULL, **mp = ⊤
848 int len;
849
850 if (off < 0 || off > MHLEN)
851 return (NULL);
852
853 while (totlen > 0) {
854 if (top == NULL) { /* First one, must be PKTHDR */
855 if (totlen + off >= MINCLSIZE) {
856 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
857 len = MCLBYTES;
858 } else {
859 m = m_gethdr(M_DONTWAIT, MT_DATA);
860 len = MHLEN;
861
862 /* Place initial small packet/header at end of mbuf */
863 if (m && totlen + off + max_linkhdr <= MLEN) {
864 m->m_data += max_linkhdr;
865 len -= max_linkhdr;
866 }
867 }
868 if (m == NULL)
869 return NULL;
870 m->m_pkthdr.rcvif = ifp;
871 m->m_pkthdr.len = totlen;
872 } else {
873 if (totlen + off >= MINCLSIZE) {
874 m = m_getcl(M_DONTWAIT, MT_DATA, 0);
875 len = MCLBYTES;
876 } else {
877 m = m_get(M_DONTWAIT, MT_DATA);
878 len = MLEN;
879 }
880 if (m == NULL) {
881 m_freem(top);
882 return NULL;
883 }
884 }
885 if (off) {
886 m->m_data += off;
887 len -= off;
888 off = 0;
889 }
890 m->m_len = len = min(totlen, len);
891 if (copy)
892 copy(buf, mtod(m, caddr_t), (u_int)len);
893 else
894 bcopy(buf, mtod(m, caddr_t), (u_int)len);
895 buf += len;
896 *mp = m;
897 mp = &m->m_next;
898 totlen -= len;
899 }
900 return (top);
901 }
902
903 /*
904 * Copy data from a buffer back into the indicated mbuf chain,
905 * starting "off" bytes from the beginning, extending the mbuf
906 * chain if necessary.
907 */
908 void
909 m_copyback(struct mbuf *m0, int off, int len, c_caddr_t cp)
910 {
911 int mlen;
912 struct mbuf *m = m0, *n;
913 int totlen = 0;
914
915 if (m0 == NULL)
916 return;
917 while (off > (mlen = m->m_len)) {
918 off -= mlen;
919 totlen += mlen;
920 if (m->m_next == NULL) {
921 n = m_get(M_DONTWAIT, m->m_type);
922 if (n == NULL)
923 goto out;
924 bzero(mtod(n, caddr_t), MLEN);
925 n->m_len = min(MLEN, len + off);
926 m->m_next = n;
927 }
928 m = m->m_next;
929 }
930 while (len > 0) {
931 mlen = min (m->m_len - off, len);
932 bcopy(cp, off + mtod(m, caddr_t), (u_int)mlen);
933 cp += mlen;
934 len -= mlen;
935 mlen += off;
936 off = 0;
937 totlen += mlen;
938 if (len == 0)
939 break;
940 if (m->m_next == NULL) {
941 n = m_get(M_DONTWAIT, m->m_type);
942 if (n == NULL)
943 break;
944 n->m_len = min(MLEN, len);
945 m->m_next = n;
946 }
947 m = m->m_next;
948 }
949 out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
950 m->m_pkthdr.len = totlen;
951 }
952
953 /*
954 * Apply function f to the data in an mbuf chain starting "off" bytes from
955 * the beginning, continuing for "len" bytes.
956 */
957 int
958 m_apply(struct mbuf *m, int off, int len,
959 int (*f)(void *, void *, u_int), void *arg)
960 {
961 u_int count;
962 int rval;
963
964 KASSERT(off >= 0, ("m_apply, negative off %d", off));
965 KASSERT(len >= 0, ("m_apply, negative len %d", len));
966 while (off > 0) {
967 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
968 if (off < m->m_len)
969 break;
970 off -= m->m_len;
971 m = m->m_next;
972 }
973 while (len > 0) {
974 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
975 count = min(m->m_len - off, len);
976 rval = (*f)(arg, mtod(m, caddr_t) + off, count);
977 if (rval)
978 return (rval);
979 len -= count;
980 off = 0;
981 m = m->m_next;
982 }
983 return (0);
984 }
985
986 /*
987 * Return a pointer to mbuf/offset of location in mbuf chain.
988 */
989 struct mbuf *
990 m_getptr(struct mbuf *m, int loc, int *off)
991 {
992
993 while (loc >= 0) {
994 /* Normal end of search. */
995 if (m->m_len > loc) {
996 *off = loc;
997 return (m);
998 } else {
999 loc -= m->m_len;
1000 if (m->m_next == NULL) {
1001 if (loc == 0) {
1002 /* Point at the end of valid data. */
1003 *off = m->m_len;
1004 return (m);
1005 }
1006 return (NULL);
1007 }
1008 m = m->m_next;
1009 }
1010 }
1011 return (NULL);
1012 }
1013
1014 void
1015 m_print(const struct mbuf *m, int maxlen)
1016 {
1017 int len;
1018 int pdata;
1019 const struct mbuf *m2;
1020
1021 if (m->m_flags & M_PKTHDR)
1022 len = m->m_pkthdr.len;
1023 else
1024 len = -1;
1025 m2 = m;
1026 while (m2 != NULL && (len == -1 || len)) {
1027 pdata = m2->m_len;
1028 if (maxlen != -1 && pdata > maxlen)
1029 pdata = maxlen;
1030 printf("mbuf: %p len: %d, next: %p, %b%s", m2, m2->m_len,
1031 m2->m_next, m2->m_flags, "\2\20freelist\17skipfw"
1032 "\11proto5\10proto4\7proto3\6proto2\5proto1\4rdonly"
1033 "\3eor\2pkthdr\1ext", pdata ? "" : "\n");
1034 if (pdata)
1035 printf(", %*D\n", m2->m_len, (u_char *)m2->m_data, "-");
1036 if (len != -1)
1037 len -= m2->m_len;
1038 m2 = m2->m_next;
1039 }
1040 if (len > 0)
1041 printf("%d bytes unaccounted for.\n", len);
1042 return;
1043 }
1044
1045 u_int
1046 m_fixhdr(struct mbuf *m0)
1047 {
1048 u_int len;
1049
1050 len = m_length(m0, NULL);
1051 m0->m_pkthdr.len = len;
1052 return (len);
1053 }
1054
1055 u_int
1056 m_length(struct mbuf *m0, struct mbuf **last)
1057 {
1058 struct mbuf *m;
1059 u_int len;
1060
1061 len = 0;
1062 for (m = m0; m != NULL; m = m->m_next) {
1063 len += m->m_len;
1064 if (m->m_next == NULL)
1065 break;
1066 }
1067 if (last != NULL)
1068 *last = m;
1069 return (len);
1070 }
1071
1072 /*
1073 * Defragment a mbuf chain, returning the shortest possible
1074 * chain of mbufs and clusters. If allocation fails and
1075 * this cannot be completed, NULL will be returned, but
1076 * the passed in chain will be unchanged. Upon success,
1077 * the original chain will be freed, and the new chain
1078 * will be returned.
1079 *
1080 * If a non-packet header is passed in, the original
1081 * mbuf (chain?) will be returned unharmed.
1082 */
1083 struct mbuf *
1084 m_defrag(struct mbuf *m0, int how)
1085 {
1086 struct mbuf *m_new = NULL, *m_final = NULL;
1087 int progress = 0, length;
1088
1089 MBUF_CHECKSLEEP(how);
1090 if (!(m0->m_flags & M_PKTHDR))
1091 return (m0);
1092
1093 m_fixhdr(m0); /* Needed sanity check */
1094
1095 #ifdef MBUF_STRESS_TEST
1096 if (m_defragrandomfailures) {
1097 int temp = arc4random() & 0xff;
1098 if (temp == 0xba)
1099 goto nospace;
1100 }
1101 #endif
1102
1103 if (m0->m_pkthdr.len > MHLEN)
1104 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1105 else
1106 m_final = m_gethdr(how, MT_DATA);
1107
1108 if (m_final == NULL)
1109 goto nospace;
1110
1111 if (m_dup_pkthdr(m_final, m0, how) == 0)
1112 goto nospace;
1113
1114 m_new = m_final;
1115
1116 while (progress < m0->m_pkthdr.len) {
1117 length = m0->m_pkthdr.len - progress;
1118 if (length > MCLBYTES)
1119 length = MCLBYTES;
1120
1121 if (m_new == NULL) {
1122 if (length > MLEN)
1123 m_new = m_getcl(how, MT_DATA, 0);
1124 else
1125 m_new = m_get(how, MT_DATA);
1126 if (m_new == NULL)
1127 goto nospace;
1128 }
1129
1130 m_copydata(m0, progress, length, mtod(m_new, caddr_t));
1131 progress += length;
1132 m_new->m_len = length;
1133 if (m_new != m_final)
1134 m_cat(m_final, m_new);
1135 m_new = NULL;
1136 }
1137 #ifdef MBUF_STRESS_TEST
1138 if (m0->m_next == NULL)
1139 m_defraguseless++;
1140 #endif
1141 m_freem(m0);
1142 m0 = m_final;
1143 #ifdef MBUF_STRESS_TEST
1144 m_defragpackets++;
1145 m_defragbytes += m0->m_pkthdr.len;
1146 #endif
1147 return (m0);
1148 nospace:
1149 #ifdef MBUF_STRESS_TEST
1150 m_defragfailure++;
1151 #endif
1152 if (m_new)
1153 m_free(m_new);
1154 if (m_final)
1155 m_freem(m_final);
1156 return (NULL);
1157 }
1158
1159 #ifdef MBUF_STRESS_TEST
1160
1161 /*
1162 * Fragment an mbuf chain. There's no reason you'd ever want to do
1163 * this in normal usage, but it's great for stress testing various
1164 * mbuf consumers.
1165 *
1166 * If fragmentation is not possible, the original chain will be
1167 * returned.
1168 *
1169 * Possible length values:
1170 * 0 no fragmentation will occur
1171 * > 0 each fragment will be of the specified length
1172 * -1 each fragment will be the same random value in length
1173 * -2 each fragment's length will be entirely random
1174 * (Random values range from 1 to 256)
1175 */
1176 struct mbuf *
1177 m_fragment(struct mbuf *m0, int how, int length)
1178 {
1179 struct mbuf *m_new = NULL, *m_final = NULL;
1180 int progress = 0;
1181
1182 if (!(m0->m_flags & M_PKTHDR))
1183 return (m0);
1184
1185 if ((length == 0) || (length < -2))
1186 return (m0);
1187
1188 m_fixhdr(m0); /* Needed sanity check */
1189
1190 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1191
1192 if (m_final == NULL)
1193 goto nospace;
1194
1195 if (m_dup_pkthdr(m_final, m0, how) == 0)
1196 goto nospace;
1197
1198 m_new = m_final;
1199
1200 if (length == -1)
1201 length = 1 + (arc4random() & 255);
1202
1203 while (progress < m0->m_pkthdr.len) {
1204 int fraglen;
1205
1206 if (length > 0)
1207 fraglen = length;
1208 else
1209 fraglen = 1 + (arc4random() & 255);
1210 if (fraglen > m0->m_pkthdr.len - progress)
1211 fraglen = m0->m_pkthdr.len - progress;
1212
1213 if (fraglen > MCLBYTES)
1214 fraglen = MCLBYTES;
1215
1216 if (m_new == NULL) {
1217 m_new = m_getcl(how, MT_DATA, 0);
1218 if (m_new == NULL)
1219 goto nospace;
1220 }
1221
1222 m_copydata(m0, progress, fraglen, mtod(m_new, caddr_t));
1223 progress += fraglen;
1224 m_new->m_len = fraglen;
1225 if (m_new != m_final)
1226 m_cat(m_final, m_new);
1227 m_new = NULL;
1228 }
1229 m_freem(m0);
1230 m0 = m_final;
1231 return (m0);
1232 nospace:
1233 if (m_new)
1234 m_free(m_new);
1235 if (m_final)
1236 m_freem(m_final);
1237 /* Return the original chain on failure */
1238 return (m0);
1239 }
1240
1241 #endif
1242
1243 struct mbuf *
1244 m_uiotombuf(struct uio *uio, int how, int len)
1245 {
1246 struct mbuf *m_new = NULL, *m_final = NULL;
1247 int progress = 0, error = 0, length, total;
1248
1249 if (len > 0)
1250 total = min(uio->uio_resid, len);
1251 else
1252 total = uio->uio_resid;
1253 if (total > MHLEN)
1254 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1255 else
1256 m_final = m_gethdr(how, MT_DATA);
1257 if (m_final == NULL)
1258 goto nospace;
1259 m_new = m_final;
1260 while (progress < total) {
1261 length = total - progress;
1262 if (length > MCLBYTES)
1263 length = MCLBYTES;
1264 if (m_new == NULL) {
1265 if (length > MLEN)
1266 m_new = m_getcl(how, MT_DATA, 0);
1267 else
1268 m_new = m_get(how, MT_DATA);
1269 if (m_new == NULL)
1270 goto nospace;
1271 }
1272 error = uiomove(mtod(m_new, void *), length, uio);
1273 if (error)
1274 goto nospace;
1275 progress += length;
1276 m_new->m_len = length;
1277 if (m_new != m_final)
1278 m_cat(m_final, m_new);
1279 m_new = NULL;
1280 }
1281 m_fixhdr(m_final);
1282 return (m_final);
1283 nospace:
1284 if (m_new)
1285 m_free(m_new);
1286 if (m_final)
1287 m_freem(m_final);
1288 return (NULL);
1289 }
Cache object: fa0a25fcc17a1a40626a11f3cc0a34de
|