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/11.2/sys/kern/uipc_mbuf.c 331847 2018-03-31 17:28:30Z avos $");
34
35 #include "opt_param.h"
36 #include "opt_mbuf_stress_test.h"
37 #include "opt_mbuf_profiling.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/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/sysctl.h>
47 #include <sys/domain.h>
48 #include <sys/protosw.h>
49 #include <sys/uio.h>
50 #include <sys/sdt.h>
51
52 SDT_PROBE_DEFINE5_XLATE(sdt, , , m__init,
53 "struct mbuf *", "mbufinfo_t *",
54 "uint32_t", "uint32_t",
55 "uint16_t", "uint16_t",
56 "uint32_t", "uint32_t",
57 "uint32_t", "uint32_t");
58
59 SDT_PROBE_DEFINE3_XLATE(sdt, , , m__gethdr,
60 "uint32_t", "uint32_t",
61 "uint16_t", "uint16_t",
62 "struct mbuf *", "mbufinfo_t *");
63
64 SDT_PROBE_DEFINE3_XLATE(sdt, , , m__get,
65 "uint32_t", "uint32_t",
66 "uint16_t", "uint16_t",
67 "struct mbuf *", "mbufinfo_t *");
68
69 SDT_PROBE_DEFINE4_XLATE(sdt, , , m__getcl,
70 "uint32_t", "uint32_t",
71 "uint16_t", "uint16_t",
72 "uint32_t", "uint32_t",
73 "struct mbuf *", "mbufinfo_t *");
74
75 SDT_PROBE_DEFINE3_XLATE(sdt, , , m__clget,
76 "struct mbuf *", "mbufinfo_t *",
77 "uint32_t", "uint32_t",
78 "uint32_t", "uint32_t");
79
80 SDT_PROBE_DEFINE4_XLATE(sdt, , , m__cljget,
81 "struct mbuf *", "mbufinfo_t *",
82 "uint32_t", "uint32_t",
83 "uint32_t", "uint32_t",
84 "void*", "void*");
85
86 SDT_PROBE_DEFINE(sdt, , , m__cljset);
87
88 SDT_PROBE_DEFINE1_XLATE(sdt, , , m__free,
89 "struct mbuf *", "mbufinfo_t *");
90
91 SDT_PROBE_DEFINE1_XLATE(sdt, , , m__freem,
92 "struct mbuf *", "mbufinfo_t *");
93
94 #include <security/mac/mac_framework.h>
95
96 int max_linkhdr;
97 int max_protohdr;
98 int max_hdr;
99 int max_datalen;
100 #ifdef MBUF_STRESS_TEST
101 int m_defragpackets;
102 int m_defragbytes;
103 int m_defraguseless;
104 int m_defragfailure;
105 int m_defragrandomfailures;
106 #endif
107
108 /*
109 * sysctl(8) exported objects
110 */
111 SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RD,
112 &max_linkhdr, 0, "Size of largest link layer header");
113 SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RD,
114 &max_protohdr, 0, "Size of largest protocol layer header");
115 SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RD,
116 &max_hdr, 0, "Size of largest link plus protocol header");
117 SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RD,
118 &max_datalen, 0, "Minimum space left in mbuf after max_hdr");
119 #ifdef MBUF_STRESS_TEST
120 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD,
121 &m_defragpackets, 0, "");
122 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD,
123 &m_defragbytes, 0, "");
124 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD,
125 &m_defraguseless, 0, "");
126 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD,
127 &m_defragfailure, 0, "");
128 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW,
129 &m_defragrandomfailures, 0, "");
130 #endif
131
132 /*
133 * Ensure the correct size of various mbuf parameters. It could be off due
134 * to compiler-induced padding and alignment artifacts.
135 */
136 CTASSERT(MSIZE - offsetof(struct mbuf, m_dat) == MLEN);
137 CTASSERT(MSIZE - offsetof(struct mbuf, m_pktdat) == MHLEN);
138
139 /*
140 * mbuf data storage should be 64-bit aligned regardless of architectural
141 * pointer size; check this is the case with and without a packet header.
142 */
143 CTASSERT(offsetof(struct mbuf, m_dat) % 8 == 0);
144 CTASSERT(offsetof(struct mbuf, m_pktdat) % 8 == 0);
145
146 /*
147 * While the specific values here don't matter too much (i.e., +/- a few
148 * words), we do want to ensure that changes to these values are carefully
149 * reasoned about and properly documented. This is especially the case as
150 * network-protocol and device-driver modules encode these layouts, and must
151 * be recompiled if the structures change. Check these values at compile time
152 * against the ones documented in comments in mbuf.h.
153 *
154 * NB: Possibly they should be documented there via #define's and not just
155 * comments.
156 */
157 #if defined(__LP64__)
158 CTASSERT(offsetof(struct mbuf, m_dat) == 32);
159 CTASSERT(sizeof(struct pkthdr) == 56);
160 CTASSERT(sizeof(struct m_ext) == 48);
161 #else
162 CTASSERT(offsetof(struct mbuf, m_dat) == 24);
163 CTASSERT(sizeof(struct pkthdr) == 48);
164 CTASSERT(sizeof(struct m_ext) == 28);
165 #endif
166
167 /*
168 * Assert that the queue(3) macros produce code of the same size as an old
169 * plain pointer does.
170 */
171 #ifdef INVARIANTS
172 static struct mbuf __used m_assertbuf;
173 CTASSERT(sizeof(m_assertbuf.m_slist) == sizeof(m_assertbuf.m_next));
174 CTASSERT(sizeof(m_assertbuf.m_stailq) == sizeof(m_assertbuf.m_next));
175 CTASSERT(sizeof(m_assertbuf.m_slistpkt) == sizeof(m_assertbuf.m_nextpkt));
176 CTASSERT(sizeof(m_assertbuf.m_stailqpkt) == sizeof(m_assertbuf.m_nextpkt));
177 #endif
178
179 /*
180 * Attach the cluster from *m to *n, set up m_ext in *n
181 * and bump the refcount of the cluster.
182 */
183 void
184 mb_dupcl(struct mbuf *n, struct mbuf *m)
185 {
186 volatile u_int *refcnt;
187
188 KASSERT(m->m_flags & M_EXT, ("%s: M_EXT not set on %p", __func__, m));
189 KASSERT(!(n->m_flags & M_EXT), ("%s: M_EXT set on %p", __func__, n));
190
191 n->m_ext = m->m_ext;
192 n->m_flags |= M_EXT;
193 n->m_flags |= m->m_flags & M_RDONLY;
194
195 /* See if this is the mbuf that holds the embedded refcount. */
196 if (m->m_ext.ext_flags & EXT_FLAG_EMBREF) {
197 refcnt = n->m_ext.ext_cnt = &m->m_ext.ext_count;
198 n->m_ext.ext_flags &= ~EXT_FLAG_EMBREF;
199 } else {
200 KASSERT(m->m_ext.ext_cnt != NULL,
201 ("%s: no refcounting pointer on %p", __func__, m));
202 refcnt = m->m_ext.ext_cnt;
203 }
204
205 if (*refcnt == 1)
206 *refcnt += 1;
207 else
208 atomic_add_int(refcnt, 1);
209 }
210
211 void
212 m_demote_pkthdr(struct mbuf *m)
213 {
214
215 M_ASSERTPKTHDR(m);
216
217 m_tag_delete_chain(m, NULL);
218 m->m_flags &= ~M_PKTHDR;
219 bzero(&m->m_pkthdr, sizeof(struct pkthdr));
220 }
221
222 /*
223 * Clean up mbuf (chain) from any tags and packet headers.
224 * If "all" is set then the first mbuf in the chain will be
225 * cleaned too.
226 */
227 void
228 m_demote(struct mbuf *m0, int all, int flags)
229 {
230 struct mbuf *m;
231
232 for (m = all ? m0 : m0->m_next; m != NULL; m = m->m_next) {
233 KASSERT(m->m_nextpkt == NULL, ("%s: m_nextpkt in m %p, m0 %p",
234 __func__, m, m0));
235 if (m->m_flags & M_PKTHDR)
236 m_demote_pkthdr(m);
237 m->m_flags = m->m_flags & (M_EXT | M_RDONLY | M_NOFREE | flags);
238 }
239 }
240
241 /*
242 * Sanity checks on mbuf (chain) for use in KASSERT() and general
243 * debugging.
244 * Returns 0 or panics when bad and 1 on all tests passed.
245 * Sanitize, 0 to run M_SANITY_ACTION, 1 to garble things so they
246 * blow up later.
247 */
248 int
249 m_sanity(struct mbuf *m0, int sanitize)
250 {
251 struct mbuf *m;
252 caddr_t a, b;
253 int pktlen = 0;
254
255 #ifdef INVARIANTS
256 #define M_SANITY_ACTION(s) panic("mbuf %p: " s, m)
257 #else
258 #define M_SANITY_ACTION(s) printf("mbuf %p: " s, m)
259 #endif
260
261 for (m = m0; m != NULL; m = m->m_next) {
262 /*
263 * Basic pointer checks. If any of these fails then some
264 * unrelated kernel memory before or after us is trashed.
265 * No way to recover from that.
266 */
267 a = M_START(m);
268 b = a + M_SIZE(m);
269 if ((caddr_t)m->m_data < a)
270 M_SANITY_ACTION("m_data outside mbuf data range left");
271 if ((caddr_t)m->m_data > b)
272 M_SANITY_ACTION("m_data outside mbuf data range right");
273 if ((caddr_t)m->m_data + m->m_len > b)
274 M_SANITY_ACTION("m_data + m_len exeeds mbuf space");
275
276 /* m->m_nextpkt may only be set on first mbuf in chain. */
277 if (m != m0 && m->m_nextpkt != NULL) {
278 if (sanitize) {
279 m_freem(m->m_nextpkt);
280 m->m_nextpkt = (struct mbuf *)0xDEADC0DE;
281 } else
282 M_SANITY_ACTION("m->m_nextpkt on in-chain mbuf");
283 }
284
285 /* packet length (not mbuf length!) calculation */
286 if (m0->m_flags & M_PKTHDR)
287 pktlen += m->m_len;
288
289 /* m_tags may only be attached to first mbuf in chain. */
290 if (m != m0 && m->m_flags & M_PKTHDR &&
291 !SLIST_EMPTY(&m->m_pkthdr.tags)) {
292 if (sanitize) {
293 m_tag_delete_chain(m, NULL);
294 /* put in 0xDEADC0DE perhaps? */
295 } else
296 M_SANITY_ACTION("m_tags on in-chain mbuf");
297 }
298
299 /* M_PKTHDR may only be set on first mbuf in chain */
300 if (m != m0 && m->m_flags & M_PKTHDR) {
301 if (sanitize) {
302 bzero(&m->m_pkthdr, sizeof(m->m_pkthdr));
303 m->m_flags &= ~M_PKTHDR;
304 /* put in 0xDEADCODE and leave hdr flag in */
305 } else
306 M_SANITY_ACTION("M_PKTHDR on in-chain mbuf");
307 }
308 }
309 m = m0;
310 if (pktlen && pktlen != m->m_pkthdr.len) {
311 if (sanitize)
312 m->m_pkthdr.len = 0;
313 else
314 M_SANITY_ACTION("m_pkthdr.len != mbuf chain length");
315 }
316 return 1;
317
318 #undef M_SANITY_ACTION
319 }
320
321 /*
322 * Non-inlined part of m_init().
323 */
324 int
325 m_pkthdr_init(struct mbuf *m, int how)
326 {
327 #ifdef MAC
328 int error;
329 #endif
330 m->m_data = m->m_pktdat;
331 bzero(&m->m_pkthdr, sizeof(m->m_pkthdr));
332 #ifdef MAC
333 /* If the label init fails, fail the alloc */
334 error = mac_mbuf_init(m, how);
335 if (error)
336 return (error);
337 #endif
338
339 return (0);
340 }
341
342 /*
343 * "Move" mbuf pkthdr from "from" to "to".
344 * "from" must have M_PKTHDR set, and "to" must be empty.
345 */
346 void
347 m_move_pkthdr(struct mbuf *to, struct mbuf *from)
348 {
349
350 #if 0
351 /* see below for why these are not enabled */
352 M_ASSERTPKTHDR(to);
353 /* Note: with MAC, this may not be a good assertion. */
354 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags),
355 ("m_move_pkthdr: to has tags"));
356 #endif
357 #ifdef MAC
358 /*
359 * XXXMAC: It could be this should also occur for non-MAC?
360 */
361 if (to->m_flags & M_PKTHDR)
362 m_tag_delete_chain(to, NULL);
363 #endif
364 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT);
365 if ((to->m_flags & M_EXT) == 0)
366 to->m_data = to->m_pktdat;
367 to->m_pkthdr = from->m_pkthdr; /* especially tags */
368 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */
369 from->m_flags &= ~M_PKTHDR;
370 }
371
372 /*
373 * Duplicate "from"'s mbuf pkthdr in "to".
374 * "from" must have M_PKTHDR set, and "to" must be empty.
375 * In particular, this does a deep copy of the packet tags.
376 */
377 int
378 m_dup_pkthdr(struct mbuf *to, const struct mbuf *from, int how)
379 {
380
381 #if 0
382 /*
383 * The mbuf allocator only initializes the pkthdr
384 * when the mbuf is allocated with m_gethdr(). Many users
385 * (e.g. m_copy*, m_prepend) use m_get() and then
386 * smash the pkthdr as needed causing these
387 * assertions to trip. For now just disable them.
388 */
389 M_ASSERTPKTHDR(to);
390 /* Note: with MAC, this may not be a good assertion. */
391 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), ("m_dup_pkthdr: to has tags"));
392 #endif
393 MBUF_CHECKSLEEP(how);
394 #ifdef MAC
395 if (to->m_flags & M_PKTHDR)
396 m_tag_delete_chain(to, NULL);
397 #endif
398 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT);
399 if ((to->m_flags & M_EXT) == 0)
400 to->m_data = to->m_pktdat;
401 to->m_pkthdr = from->m_pkthdr;
402 SLIST_INIT(&to->m_pkthdr.tags);
403 return (m_tag_copy_chain(to, from, how));
404 }
405
406 /*
407 * Lesser-used path for M_PREPEND:
408 * allocate new mbuf to prepend to chain,
409 * copy junk along.
410 */
411 struct mbuf *
412 m_prepend(struct mbuf *m, int len, int how)
413 {
414 struct mbuf *mn;
415
416 if (m->m_flags & M_PKTHDR)
417 mn = m_gethdr(how, m->m_type);
418 else
419 mn = m_get(how, m->m_type);
420 if (mn == NULL) {
421 m_freem(m);
422 return (NULL);
423 }
424 if (m->m_flags & M_PKTHDR)
425 m_move_pkthdr(mn, m);
426 mn->m_next = m;
427 m = mn;
428 if (len < M_SIZE(m))
429 M_ALIGN(m, len);
430 m->m_len = len;
431 return (m);
432 }
433
434 /*
435 * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
436 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf.
437 * The wait parameter is a choice of M_WAITOK/M_NOWAIT from caller.
438 * Note that the copy is read-only, because clusters are not copied,
439 * only their reference counts are incremented.
440 */
441 struct mbuf *
442 m_copym(struct mbuf *m, int off0, int len, int wait)
443 {
444 struct mbuf *n, **np;
445 int off = off0;
446 struct mbuf *top;
447 int copyhdr = 0;
448
449 KASSERT(off >= 0, ("m_copym, negative off %d", off));
450 KASSERT(len >= 0, ("m_copym, negative len %d", len));
451 MBUF_CHECKSLEEP(wait);
452 if (off == 0 && m->m_flags & M_PKTHDR)
453 copyhdr = 1;
454 while (off > 0) {
455 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain"));
456 if (off < m->m_len)
457 break;
458 off -= m->m_len;
459 m = m->m_next;
460 }
461 np = ⊤
462 top = NULL;
463 while (len > 0) {
464 if (m == NULL) {
465 KASSERT(len == M_COPYALL,
466 ("m_copym, length > size of mbuf chain"));
467 break;
468 }
469 if (copyhdr)
470 n = m_gethdr(wait, m->m_type);
471 else
472 n = m_get(wait, m->m_type);
473 *np = n;
474 if (n == NULL)
475 goto nospace;
476 if (copyhdr) {
477 if (!m_dup_pkthdr(n, m, wait))
478 goto nospace;
479 if (len == M_COPYALL)
480 n->m_pkthdr.len -= off0;
481 else
482 n->m_pkthdr.len = len;
483 copyhdr = 0;
484 }
485 n->m_len = min(len, m->m_len - off);
486 if (m->m_flags & M_EXT) {
487 n->m_data = m->m_data + off;
488 mb_dupcl(n, m);
489 } else
490 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
491 (u_int)n->m_len);
492 if (len != M_COPYALL)
493 len -= n->m_len;
494 off = 0;
495 m = m->m_next;
496 np = &n->m_next;
497 }
498
499 return (top);
500 nospace:
501 m_freem(top);
502 return (NULL);
503 }
504
505 /*
506 * Copy an entire packet, including header (which must be present).
507 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
508 * Note that the copy is read-only, because clusters are not copied,
509 * only their reference counts are incremented.
510 * Preserve alignment of the first mbuf so if the creator has left
511 * some room at the beginning (e.g. for inserting protocol headers)
512 * the copies still have the room available.
513 */
514 struct mbuf *
515 m_copypacket(struct mbuf *m, int how)
516 {
517 struct mbuf *top, *n, *o;
518
519 MBUF_CHECKSLEEP(how);
520 n = m_get(how, m->m_type);
521 top = n;
522 if (n == NULL)
523 goto nospace;
524
525 if (!m_dup_pkthdr(n, m, how))
526 goto nospace;
527 n->m_len = m->m_len;
528 if (m->m_flags & M_EXT) {
529 n->m_data = m->m_data;
530 mb_dupcl(n, m);
531 } else {
532 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat );
533 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
534 }
535
536 m = m->m_next;
537 while (m) {
538 o = m_get(how, m->m_type);
539 if (o == NULL)
540 goto nospace;
541
542 n->m_next = o;
543 n = n->m_next;
544
545 n->m_len = m->m_len;
546 if (m->m_flags & M_EXT) {
547 n->m_data = m->m_data;
548 mb_dupcl(n, m);
549 } else {
550 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
551 }
552
553 m = m->m_next;
554 }
555 return top;
556 nospace:
557 m_freem(top);
558 return (NULL);
559 }
560
561 /*
562 * Copy data from an mbuf chain starting "off" bytes from the beginning,
563 * continuing for "len" bytes, into the indicated buffer.
564 */
565 void
566 m_copydata(const struct mbuf *m, int off, int len, caddr_t cp)
567 {
568 u_int count;
569
570 KASSERT(off >= 0, ("m_copydata, negative off %d", off));
571 KASSERT(len >= 0, ("m_copydata, negative len %d", len));
572 while (off > 0) {
573 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain"));
574 if (off < m->m_len)
575 break;
576 off -= m->m_len;
577 m = m->m_next;
578 }
579 while (len > 0) {
580 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain"));
581 count = min(m->m_len - off, len);
582 bcopy(mtod(m, caddr_t) + off, cp, count);
583 len -= count;
584 cp += count;
585 off = 0;
586 m = m->m_next;
587 }
588 }
589
590 /*
591 * Copy a packet header mbuf chain into a completely new chain, including
592 * copying any mbuf clusters. Use this instead of m_copypacket() when
593 * you need a writable copy of an mbuf chain.
594 */
595 struct mbuf *
596 m_dup(const struct mbuf *m, int how)
597 {
598 struct mbuf **p, *top = NULL;
599 int remain, moff, nsize;
600
601 MBUF_CHECKSLEEP(how);
602 /* Sanity check */
603 if (m == NULL)
604 return (NULL);
605 M_ASSERTPKTHDR(m);
606
607 /* While there's more data, get a new mbuf, tack it on, and fill it */
608 remain = m->m_pkthdr.len;
609 moff = 0;
610 p = ⊤
611 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */
612 struct mbuf *n;
613
614 /* Get the next new mbuf */
615 if (remain >= MINCLSIZE) {
616 n = m_getcl(how, m->m_type, 0);
617 nsize = MCLBYTES;
618 } else {
619 n = m_get(how, m->m_type);
620 nsize = MLEN;
621 }
622 if (n == NULL)
623 goto nospace;
624
625 if (top == NULL) { /* First one, must be PKTHDR */
626 if (!m_dup_pkthdr(n, m, how)) {
627 m_free(n);
628 goto nospace;
629 }
630 if ((n->m_flags & M_EXT) == 0)
631 nsize = MHLEN;
632 n->m_flags &= ~M_RDONLY;
633 }
634 n->m_len = 0;
635
636 /* Link it into the new chain */
637 *p = n;
638 p = &n->m_next;
639
640 /* Copy data from original mbuf(s) into new mbuf */
641 while (n->m_len < nsize && m != NULL) {
642 int chunk = min(nsize - n->m_len, m->m_len - moff);
643
644 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
645 moff += chunk;
646 n->m_len += chunk;
647 remain -= chunk;
648 if (moff == m->m_len) {
649 m = m->m_next;
650 moff = 0;
651 }
652 }
653
654 /* Check correct total mbuf length */
655 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL),
656 ("%s: bogus m_pkthdr.len", __func__));
657 }
658 return (top);
659
660 nospace:
661 m_freem(top);
662 return (NULL);
663 }
664
665 /*
666 * Concatenate mbuf chain n to m.
667 * Both chains must be of the same type (e.g. MT_DATA).
668 * Any m_pkthdr is not updated.
669 */
670 void
671 m_cat(struct mbuf *m, struct mbuf *n)
672 {
673 while (m->m_next)
674 m = m->m_next;
675 while (n) {
676 if (!M_WRITABLE(m) ||
677 M_TRAILINGSPACE(m) < n->m_len) {
678 /* just join the two chains */
679 m->m_next = n;
680 return;
681 }
682 /* splat the data from one into the other */
683 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
684 (u_int)n->m_len);
685 m->m_len += n->m_len;
686 n = m_free(n);
687 }
688 }
689
690 /*
691 * Concatenate two pkthdr mbuf chains.
692 */
693 void
694 m_catpkt(struct mbuf *m, struct mbuf *n)
695 {
696
697 M_ASSERTPKTHDR(m);
698 M_ASSERTPKTHDR(n);
699
700 m->m_pkthdr.len += n->m_pkthdr.len;
701 m_demote(n, 1, 0);
702
703 m_cat(m, n);
704 }
705
706 void
707 m_adj(struct mbuf *mp, int req_len)
708 {
709 int len = req_len;
710 struct mbuf *m;
711 int count;
712
713 if ((m = mp) == NULL)
714 return;
715 if (len >= 0) {
716 /*
717 * Trim from head.
718 */
719 while (m != NULL && len > 0) {
720 if (m->m_len <= len) {
721 len -= m->m_len;
722 m->m_len = 0;
723 m = m->m_next;
724 } else {
725 m->m_len -= len;
726 m->m_data += len;
727 len = 0;
728 }
729 }
730 if (mp->m_flags & M_PKTHDR)
731 mp->m_pkthdr.len -= (req_len - len);
732 } else {
733 /*
734 * Trim from tail. Scan the mbuf chain,
735 * calculating its length and finding the last mbuf.
736 * If the adjustment only affects this mbuf, then just
737 * adjust and return. Otherwise, rescan and truncate
738 * after the remaining size.
739 */
740 len = -len;
741 count = 0;
742 for (;;) {
743 count += m->m_len;
744 if (m->m_next == (struct mbuf *)0)
745 break;
746 m = m->m_next;
747 }
748 if (m->m_len >= len) {
749 m->m_len -= len;
750 if (mp->m_flags & M_PKTHDR)
751 mp->m_pkthdr.len -= len;
752 return;
753 }
754 count -= len;
755 if (count < 0)
756 count = 0;
757 /*
758 * Correct length for chain is "count".
759 * Find the mbuf with last data, adjust its length,
760 * and toss data from remaining mbufs on chain.
761 */
762 m = mp;
763 if (m->m_flags & M_PKTHDR)
764 m->m_pkthdr.len = count;
765 for (; m; m = m->m_next) {
766 if (m->m_len >= count) {
767 m->m_len = count;
768 if (m->m_next != NULL) {
769 m_freem(m->m_next);
770 m->m_next = NULL;
771 }
772 break;
773 }
774 count -= m->m_len;
775 }
776 }
777 }
778
779 /*
780 * Rearange an mbuf chain so that len bytes are contiguous
781 * and in the data area of an mbuf (so that mtod will work
782 * for a structure of size len). Returns the resulting
783 * mbuf chain on success, frees it and returns null on failure.
784 * If there is room, it will add up to max_protohdr-len extra bytes to the
785 * contiguous region in an attempt to avoid being called next time.
786 */
787 struct mbuf *
788 m_pullup(struct mbuf *n, int len)
789 {
790 struct mbuf *m;
791 int count;
792 int space;
793
794 /*
795 * If first mbuf has no cluster, and has room for len bytes
796 * without shifting current data, pullup into it,
797 * otherwise allocate a new mbuf to prepend to the chain.
798 */
799 if ((n->m_flags & M_EXT) == 0 &&
800 n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
801 if (n->m_len >= len)
802 return (n);
803 m = n;
804 n = n->m_next;
805 len -= m->m_len;
806 } else {
807 if (len > MHLEN)
808 goto bad;
809 m = m_get(M_NOWAIT, n->m_type);
810 if (m == NULL)
811 goto bad;
812 if (n->m_flags & M_PKTHDR)
813 m_move_pkthdr(m, n);
814 }
815 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
816 do {
817 count = min(min(max(len, max_protohdr), space), n->m_len);
818 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
819 (u_int)count);
820 len -= count;
821 m->m_len += count;
822 n->m_len -= count;
823 space -= count;
824 if (n->m_len)
825 n->m_data += count;
826 else
827 n = m_free(n);
828 } while (len > 0 && n);
829 if (len > 0) {
830 (void) m_free(m);
831 goto bad;
832 }
833 m->m_next = n;
834 return (m);
835 bad:
836 m_freem(n);
837 return (NULL);
838 }
839
840 /*
841 * Like m_pullup(), except a new mbuf is always allocated, and we allow
842 * the amount of empty space before the data in the new mbuf to be specified
843 * (in the event that the caller expects to prepend later).
844 */
845 struct mbuf *
846 m_copyup(struct mbuf *n, int len, int dstoff)
847 {
848 struct mbuf *m;
849 int count, space;
850
851 if (len > (MHLEN - dstoff))
852 goto bad;
853 m = m_get(M_NOWAIT, n->m_type);
854 if (m == NULL)
855 goto bad;
856 if (n->m_flags & M_PKTHDR)
857 m_move_pkthdr(m, n);
858 m->m_data += dstoff;
859 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
860 do {
861 count = min(min(max(len, max_protohdr), space), n->m_len);
862 memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
863 (unsigned)count);
864 len -= count;
865 m->m_len += count;
866 n->m_len -= count;
867 space -= count;
868 if (n->m_len)
869 n->m_data += count;
870 else
871 n = m_free(n);
872 } while (len > 0 && n);
873 if (len > 0) {
874 (void) m_free(m);
875 goto bad;
876 }
877 m->m_next = n;
878 return (m);
879 bad:
880 m_freem(n);
881 return (NULL);
882 }
883
884 /*
885 * Partition an mbuf chain in two pieces, returning the tail --
886 * all but the first len0 bytes. In case of failure, it returns NULL and
887 * attempts to restore the chain to its original state.
888 *
889 * Note that the resulting mbufs might be read-only, because the new
890 * mbuf can end up sharing an mbuf cluster with the original mbuf if
891 * the "breaking point" happens to lie within a cluster mbuf. Use the
892 * M_WRITABLE() macro to check for this case.
893 */
894 struct mbuf *
895 m_split(struct mbuf *m0, int len0, int wait)
896 {
897 struct mbuf *m, *n;
898 u_int len = len0, remain;
899
900 MBUF_CHECKSLEEP(wait);
901 for (m = m0; m && len > m->m_len; m = m->m_next)
902 len -= m->m_len;
903 if (m == NULL)
904 return (NULL);
905 remain = m->m_len - len;
906 if (m0->m_flags & M_PKTHDR && remain == 0) {
907 n = m_gethdr(wait, m0->m_type);
908 if (n == NULL)
909 return (NULL);
910 n->m_next = m->m_next;
911 m->m_next = NULL;
912 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
913 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
914 m0->m_pkthdr.len = len0;
915 return (n);
916 } else if (m0->m_flags & M_PKTHDR) {
917 n = m_gethdr(wait, m0->m_type);
918 if (n == NULL)
919 return (NULL);
920 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
921 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
922 m0->m_pkthdr.len = len0;
923 if (m->m_flags & M_EXT)
924 goto extpacket;
925 if (remain > MHLEN) {
926 /* m can't be the lead packet */
927 M_ALIGN(n, 0);
928 n->m_next = m_split(m, len, wait);
929 if (n->m_next == NULL) {
930 (void) m_free(n);
931 return (NULL);
932 } else {
933 n->m_len = 0;
934 return (n);
935 }
936 } else
937 M_ALIGN(n, remain);
938 } else if (remain == 0) {
939 n = m->m_next;
940 m->m_next = NULL;
941 return (n);
942 } else {
943 n = m_get(wait, m->m_type);
944 if (n == NULL)
945 return (NULL);
946 M_ALIGN(n, remain);
947 }
948 extpacket:
949 if (m->m_flags & M_EXT) {
950 n->m_data = m->m_data + len;
951 mb_dupcl(n, m);
952 } else {
953 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
954 }
955 n->m_len = remain;
956 m->m_len = len;
957 n->m_next = m->m_next;
958 m->m_next = NULL;
959 return (n);
960 }
961 /*
962 * Routine to copy from device local memory into mbufs.
963 * Note that `off' argument is offset into first mbuf of target chain from
964 * which to begin copying the data to.
965 */
966 struct mbuf *
967 m_devget(char *buf, int totlen, int off, struct ifnet *ifp,
968 void (*copy)(char *from, caddr_t to, u_int len))
969 {
970 struct mbuf *m;
971 struct mbuf *top = NULL, **mp = ⊤
972 int len;
973
974 if (off < 0 || off > MHLEN)
975 return (NULL);
976
977 while (totlen > 0) {
978 if (top == NULL) { /* First one, must be PKTHDR */
979 if (totlen + off >= MINCLSIZE) {
980 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
981 len = MCLBYTES;
982 } else {
983 m = m_gethdr(M_NOWAIT, MT_DATA);
984 len = MHLEN;
985
986 /* Place initial small packet/header at end of mbuf */
987 if (m && totlen + off + max_linkhdr <= MHLEN) {
988 m->m_data += max_linkhdr;
989 len -= max_linkhdr;
990 }
991 }
992 if (m == NULL)
993 return NULL;
994 m->m_pkthdr.rcvif = ifp;
995 m->m_pkthdr.len = totlen;
996 } else {
997 if (totlen + off >= MINCLSIZE) {
998 m = m_getcl(M_NOWAIT, MT_DATA, 0);
999 len = MCLBYTES;
1000 } else {
1001 m = m_get(M_NOWAIT, MT_DATA);
1002 len = MLEN;
1003 }
1004 if (m == NULL) {
1005 m_freem(top);
1006 return NULL;
1007 }
1008 }
1009 if (off) {
1010 m->m_data += off;
1011 len -= off;
1012 off = 0;
1013 }
1014 m->m_len = len = min(totlen, len);
1015 if (copy)
1016 copy(buf, mtod(m, caddr_t), (u_int)len);
1017 else
1018 bcopy(buf, mtod(m, caddr_t), (u_int)len);
1019 buf += len;
1020 *mp = m;
1021 mp = &m->m_next;
1022 totlen -= len;
1023 }
1024 return (top);
1025 }
1026
1027 /*
1028 * Copy data from a buffer back into the indicated mbuf chain,
1029 * starting "off" bytes from the beginning, extending the mbuf
1030 * chain if necessary.
1031 */
1032 void
1033 m_copyback(struct mbuf *m0, int off, int len, c_caddr_t cp)
1034 {
1035 int mlen;
1036 struct mbuf *m = m0, *n;
1037 int totlen = 0;
1038
1039 if (m0 == NULL)
1040 return;
1041 while (off > (mlen = m->m_len)) {
1042 off -= mlen;
1043 totlen += mlen;
1044 if (m->m_next == NULL) {
1045 n = m_get(M_NOWAIT, m->m_type);
1046 if (n == NULL)
1047 goto out;
1048 bzero(mtod(n, caddr_t), MLEN);
1049 n->m_len = min(MLEN, len + off);
1050 m->m_next = n;
1051 }
1052 m = m->m_next;
1053 }
1054 while (len > 0) {
1055 if (m->m_next == NULL && (len > m->m_len - off)) {
1056 m->m_len += min(len - (m->m_len - off),
1057 M_TRAILINGSPACE(m));
1058 }
1059 mlen = min (m->m_len - off, len);
1060 bcopy(cp, off + mtod(m, caddr_t), (u_int)mlen);
1061 cp += mlen;
1062 len -= mlen;
1063 mlen += off;
1064 off = 0;
1065 totlen += mlen;
1066 if (len == 0)
1067 break;
1068 if (m->m_next == NULL) {
1069 n = m_get(M_NOWAIT, m->m_type);
1070 if (n == NULL)
1071 break;
1072 n->m_len = min(MLEN, len);
1073 m->m_next = n;
1074 }
1075 m = m->m_next;
1076 }
1077 out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
1078 m->m_pkthdr.len = totlen;
1079 }
1080
1081 /*
1082 * Append the specified data to the indicated mbuf chain,
1083 * Extend the mbuf chain if the new data does not fit in
1084 * existing space.
1085 *
1086 * Return 1 if able to complete the job; otherwise 0.
1087 */
1088 int
1089 m_append(struct mbuf *m0, int len, c_caddr_t cp)
1090 {
1091 struct mbuf *m, *n;
1092 int remainder, space;
1093
1094 for (m = m0; m->m_next != NULL; m = m->m_next)
1095 ;
1096 remainder = len;
1097 space = M_TRAILINGSPACE(m);
1098 if (space > 0) {
1099 /*
1100 * Copy into available space.
1101 */
1102 if (space > remainder)
1103 space = remainder;
1104 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1105 m->m_len += space;
1106 cp += space, remainder -= space;
1107 }
1108 while (remainder > 0) {
1109 /*
1110 * Allocate a new mbuf; could check space
1111 * and allocate a cluster instead.
1112 */
1113 n = m_get(M_NOWAIT, m->m_type);
1114 if (n == NULL)
1115 break;
1116 n->m_len = min(MLEN, remainder);
1117 bcopy(cp, mtod(n, caddr_t), n->m_len);
1118 cp += n->m_len, remainder -= n->m_len;
1119 m->m_next = n;
1120 m = n;
1121 }
1122 if (m0->m_flags & M_PKTHDR)
1123 m0->m_pkthdr.len += len - remainder;
1124 return (remainder == 0);
1125 }
1126
1127 /*
1128 * Apply function f to the data in an mbuf chain starting "off" bytes from
1129 * the beginning, continuing for "len" bytes.
1130 */
1131 int
1132 m_apply(struct mbuf *m, int off, int len,
1133 int (*f)(void *, void *, u_int), void *arg)
1134 {
1135 u_int count;
1136 int rval;
1137
1138 KASSERT(off >= 0, ("m_apply, negative off %d", off));
1139 KASSERT(len >= 0, ("m_apply, negative len %d", len));
1140 while (off > 0) {
1141 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1142 if (off < m->m_len)
1143 break;
1144 off -= m->m_len;
1145 m = m->m_next;
1146 }
1147 while (len > 0) {
1148 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1149 count = min(m->m_len - off, len);
1150 rval = (*f)(arg, mtod(m, caddr_t) + off, count);
1151 if (rval)
1152 return (rval);
1153 len -= count;
1154 off = 0;
1155 m = m->m_next;
1156 }
1157 return (0);
1158 }
1159
1160 /*
1161 * Return a pointer to mbuf/offset of location in mbuf chain.
1162 */
1163 struct mbuf *
1164 m_getptr(struct mbuf *m, int loc, int *off)
1165 {
1166
1167 while (loc >= 0) {
1168 /* Normal end of search. */
1169 if (m->m_len > loc) {
1170 *off = loc;
1171 return (m);
1172 } else {
1173 loc -= m->m_len;
1174 if (m->m_next == NULL) {
1175 if (loc == 0) {
1176 /* Point at the end of valid data. */
1177 *off = m->m_len;
1178 return (m);
1179 }
1180 return (NULL);
1181 }
1182 m = m->m_next;
1183 }
1184 }
1185 return (NULL);
1186 }
1187
1188 void
1189 m_print(const struct mbuf *m, int maxlen)
1190 {
1191 int len;
1192 int pdata;
1193 const struct mbuf *m2;
1194
1195 if (m == NULL) {
1196 printf("mbuf: %p\n", m);
1197 return;
1198 }
1199
1200 if (m->m_flags & M_PKTHDR)
1201 len = m->m_pkthdr.len;
1202 else
1203 len = -1;
1204 m2 = m;
1205 while (m2 != NULL && (len == -1 || len)) {
1206 pdata = m2->m_len;
1207 if (maxlen != -1 && pdata > maxlen)
1208 pdata = maxlen;
1209 printf("mbuf: %p len: %d, next: %p, %b%s", m2, m2->m_len,
1210 m2->m_next, m2->m_flags, "\2\20freelist\17skipfw"
1211 "\11proto5\10proto4\7proto3\6proto2\5proto1\4rdonly"
1212 "\3eor\2pkthdr\1ext", pdata ? "" : "\n");
1213 if (pdata)
1214 printf(", %*D\n", pdata, (u_char *)m2->m_data, "-");
1215 if (len != -1)
1216 len -= m2->m_len;
1217 m2 = m2->m_next;
1218 }
1219 if (len > 0)
1220 printf("%d bytes unaccounted for.\n", len);
1221 return;
1222 }
1223
1224 u_int
1225 m_fixhdr(struct mbuf *m0)
1226 {
1227 u_int len;
1228
1229 len = m_length(m0, NULL);
1230 m0->m_pkthdr.len = len;
1231 return (len);
1232 }
1233
1234 u_int
1235 m_length(struct mbuf *m0, struct mbuf **last)
1236 {
1237 struct mbuf *m;
1238 u_int len;
1239
1240 len = 0;
1241 for (m = m0; m != NULL; m = m->m_next) {
1242 len += m->m_len;
1243 if (m->m_next == NULL)
1244 break;
1245 }
1246 if (last != NULL)
1247 *last = m;
1248 return (len);
1249 }
1250
1251 /*
1252 * Defragment a mbuf chain, returning the shortest possible
1253 * chain of mbufs and clusters. If allocation fails and
1254 * this cannot be completed, NULL will be returned, but
1255 * the passed in chain will be unchanged. Upon success,
1256 * the original chain will be freed, and the new chain
1257 * will be returned.
1258 *
1259 * If a non-packet header is passed in, the original
1260 * mbuf (chain?) will be returned unharmed.
1261 */
1262 struct mbuf *
1263 m_defrag(struct mbuf *m0, int how)
1264 {
1265 struct mbuf *m_new = NULL, *m_final = NULL;
1266 int progress = 0, length;
1267
1268 MBUF_CHECKSLEEP(how);
1269 if (!(m0->m_flags & M_PKTHDR))
1270 return (m0);
1271
1272 m_fixhdr(m0); /* Needed sanity check */
1273
1274 #ifdef MBUF_STRESS_TEST
1275 if (m_defragrandomfailures) {
1276 int temp = arc4random() & 0xff;
1277 if (temp == 0xba)
1278 goto nospace;
1279 }
1280 #endif
1281
1282 if (m0->m_pkthdr.len > MHLEN)
1283 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1284 else
1285 m_final = m_gethdr(how, MT_DATA);
1286
1287 if (m_final == NULL)
1288 goto nospace;
1289
1290 if (m_dup_pkthdr(m_final, m0, how) == 0)
1291 goto nospace;
1292
1293 m_new = m_final;
1294
1295 while (progress < m0->m_pkthdr.len) {
1296 length = m0->m_pkthdr.len - progress;
1297 if (length > MCLBYTES)
1298 length = MCLBYTES;
1299
1300 if (m_new == NULL) {
1301 if (length > MLEN)
1302 m_new = m_getcl(how, MT_DATA, 0);
1303 else
1304 m_new = m_get(how, MT_DATA);
1305 if (m_new == NULL)
1306 goto nospace;
1307 }
1308
1309 m_copydata(m0, progress, length, mtod(m_new, caddr_t));
1310 progress += length;
1311 m_new->m_len = length;
1312 if (m_new != m_final)
1313 m_cat(m_final, m_new);
1314 m_new = NULL;
1315 }
1316 #ifdef MBUF_STRESS_TEST
1317 if (m0->m_next == NULL)
1318 m_defraguseless++;
1319 #endif
1320 m_freem(m0);
1321 m0 = m_final;
1322 #ifdef MBUF_STRESS_TEST
1323 m_defragpackets++;
1324 m_defragbytes += m0->m_pkthdr.len;
1325 #endif
1326 return (m0);
1327 nospace:
1328 #ifdef MBUF_STRESS_TEST
1329 m_defragfailure++;
1330 #endif
1331 if (m_final)
1332 m_freem(m_final);
1333 return (NULL);
1334 }
1335
1336 /*
1337 * Defragment an mbuf chain, returning at most maxfrags separate
1338 * mbufs+clusters. If this is not possible NULL is returned and
1339 * the original mbuf chain is left in it's present (potentially
1340 * modified) state. We use two techniques: collapsing consecutive
1341 * mbufs and replacing consecutive mbufs by a cluster.
1342 *
1343 * NB: this should really be named m_defrag but that name is taken
1344 */
1345 struct mbuf *
1346 m_collapse(struct mbuf *m0, int how, int maxfrags)
1347 {
1348 struct mbuf *m, *n, *n2, **prev;
1349 u_int curfrags;
1350
1351 /*
1352 * Calculate the current number of frags.
1353 */
1354 curfrags = 0;
1355 for (m = m0; m != NULL; m = m->m_next)
1356 curfrags++;
1357 /*
1358 * First, try to collapse mbufs. Note that we always collapse
1359 * towards the front so we don't need to deal with moving the
1360 * pkthdr. This may be suboptimal if the first mbuf has much
1361 * less data than the following.
1362 */
1363 m = m0;
1364 again:
1365 for (;;) {
1366 n = m->m_next;
1367 if (n == NULL)
1368 break;
1369 if (M_WRITABLE(m) &&
1370 n->m_len < M_TRAILINGSPACE(m)) {
1371 bcopy(mtod(n, void *), mtod(m, char *) + m->m_len,
1372 n->m_len);
1373 m->m_len += n->m_len;
1374 m->m_next = n->m_next;
1375 m_free(n);
1376 if (--curfrags <= maxfrags)
1377 return m0;
1378 } else
1379 m = n;
1380 }
1381 KASSERT(maxfrags > 1,
1382 ("maxfrags %u, but normal collapse failed", maxfrags));
1383 /*
1384 * Collapse consecutive mbufs to a cluster.
1385 */
1386 prev = &m0->m_next; /* NB: not the first mbuf */
1387 while ((n = *prev) != NULL) {
1388 if ((n2 = n->m_next) != NULL &&
1389 n->m_len + n2->m_len < MCLBYTES) {
1390 m = m_getcl(how, MT_DATA, 0);
1391 if (m == NULL)
1392 goto bad;
1393 bcopy(mtod(n, void *), mtod(m, void *), n->m_len);
1394 bcopy(mtod(n2, void *), mtod(m, char *) + n->m_len,
1395 n2->m_len);
1396 m->m_len = n->m_len + n2->m_len;
1397 m->m_next = n2->m_next;
1398 *prev = m;
1399 m_free(n);
1400 m_free(n2);
1401 if (--curfrags <= maxfrags) /* +1 cl -2 mbufs */
1402 return m0;
1403 /*
1404 * Still not there, try the normal collapse
1405 * again before we allocate another cluster.
1406 */
1407 goto again;
1408 }
1409 prev = &n->m_next;
1410 }
1411 /*
1412 * No place where we can collapse to a cluster; punt.
1413 * This can occur if, for example, you request 2 frags
1414 * but the packet requires that both be clusters (we
1415 * never reallocate the first mbuf to avoid moving the
1416 * packet header).
1417 */
1418 bad:
1419 return NULL;
1420 }
1421
1422 #ifdef MBUF_STRESS_TEST
1423
1424 /*
1425 * Fragment an mbuf chain. There's no reason you'd ever want to do
1426 * this in normal usage, but it's great for stress testing various
1427 * mbuf consumers.
1428 *
1429 * If fragmentation is not possible, the original chain will be
1430 * returned.
1431 *
1432 * Possible length values:
1433 * 0 no fragmentation will occur
1434 * > 0 each fragment will be of the specified length
1435 * -1 each fragment will be the same random value in length
1436 * -2 each fragment's length will be entirely random
1437 * (Random values range from 1 to 256)
1438 */
1439 struct mbuf *
1440 m_fragment(struct mbuf *m0, int how, int length)
1441 {
1442 struct mbuf *m_first, *m_last;
1443 int divisor = 255, progress = 0, fraglen;
1444
1445 if (!(m0->m_flags & M_PKTHDR))
1446 return (m0);
1447
1448 if (length == 0 || length < -2)
1449 return (m0);
1450 if (length > MCLBYTES)
1451 length = MCLBYTES;
1452 if (length < 0 && divisor > MCLBYTES)
1453 divisor = MCLBYTES;
1454 if (length == -1)
1455 length = 1 + (arc4random() % divisor);
1456 if (length > 0)
1457 fraglen = length;
1458
1459 m_fixhdr(m0); /* Needed sanity check */
1460
1461 m_first = m_getcl(how, MT_DATA, M_PKTHDR);
1462 if (m_first == NULL)
1463 goto nospace;
1464
1465 if (m_dup_pkthdr(m_first, m0, how) == 0)
1466 goto nospace;
1467
1468 m_last = m_first;
1469
1470 while (progress < m0->m_pkthdr.len) {
1471 if (length == -2)
1472 fraglen = 1 + (arc4random() % divisor);
1473 if (fraglen > m0->m_pkthdr.len - progress)
1474 fraglen = m0->m_pkthdr.len - progress;
1475
1476 if (progress != 0) {
1477 struct mbuf *m_new = m_getcl(how, MT_DATA, 0);
1478 if (m_new == NULL)
1479 goto nospace;
1480
1481 m_last->m_next = m_new;
1482 m_last = m_new;
1483 }
1484
1485 m_copydata(m0, progress, fraglen, mtod(m_last, caddr_t));
1486 progress += fraglen;
1487 m_last->m_len = fraglen;
1488 }
1489 m_freem(m0);
1490 m0 = m_first;
1491 return (m0);
1492 nospace:
1493 if (m_first)
1494 m_freem(m_first);
1495 /* Return the original chain on failure */
1496 return (m0);
1497 }
1498
1499 #endif
1500
1501 /*
1502 * Copy the contents of uio into a properly sized mbuf chain.
1503 */
1504 struct mbuf *
1505 m_uiotombuf(struct uio *uio, int how, int len, int align, int flags)
1506 {
1507 struct mbuf *m, *mb;
1508 int error, length;
1509 ssize_t total;
1510 int progress = 0;
1511
1512 /*
1513 * len can be zero or an arbitrary large value bound by
1514 * the total data supplied by the uio.
1515 */
1516 if (len > 0)
1517 total = min(uio->uio_resid, len);
1518 else
1519 total = uio->uio_resid;
1520
1521 /*
1522 * The smallest unit returned by m_getm2() is a single mbuf
1523 * with pkthdr. We can't align past it.
1524 */
1525 if (align >= MHLEN)
1526 return (NULL);
1527
1528 /*
1529 * Give us the full allocation or nothing.
1530 * If len is zero return the smallest empty mbuf.
1531 */
1532 m = m_getm2(NULL, max(total + align, 1), how, MT_DATA, flags);
1533 if (m == NULL)
1534 return (NULL);
1535 m->m_data += align;
1536
1537 /* Fill all mbufs with uio data and update header information. */
1538 for (mb = m; mb != NULL; mb = mb->m_next) {
1539 length = min(M_TRAILINGSPACE(mb), total - progress);
1540
1541 error = uiomove(mtod(mb, void *), length, uio);
1542 if (error) {
1543 m_freem(m);
1544 return (NULL);
1545 }
1546
1547 mb->m_len = length;
1548 progress += length;
1549 if (flags & M_PKTHDR)
1550 m->m_pkthdr.len += length;
1551 }
1552 KASSERT(progress == total, ("%s: progress != total", __func__));
1553
1554 return (m);
1555 }
1556
1557 /*
1558 * Copy an mbuf chain into a uio limited by len if set.
1559 */
1560 int
1561 m_mbuftouio(struct uio *uio, struct mbuf *m, int len)
1562 {
1563 int error, length, total;
1564 int progress = 0;
1565
1566 if (len > 0)
1567 total = min(uio->uio_resid, len);
1568 else
1569 total = uio->uio_resid;
1570
1571 /* Fill the uio with data from the mbufs. */
1572 for (; m != NULL; m = m->m_next) {
1573 length = min(m->m_len, total - progress);
1574
1575 error = uiomove(mtod(m, void *), length, uio);
1576 if (error)
1577 return (error);
1578
1579 progress += length;
1580 }
1581
1582 return (0);
1583 }
1584
1585 /*
1586 * Create a writable copy of the mbuf chain. While doing this
1587 * we compact the chain with a goal of producing a chain with
1588 * at most two mbufs. The second mbuf in this chain is likely
1589 * to be a cluster. The primary purpose of this work is to create
1590 * a writable packet for encryption, compression, etc. The
1591 * secondary goal is to linearize the data so the data can be
1592 * passed to crypto hardware in the most efficient manner possible.
1593 */
1594 struct mbuf *
1595 m_unshare(struct mbuf *m0, int how)
1596 {
1597 struct mbuf *m, *mprev;
1598 struct mbuf *n, *mfirst, *mlast;
1599 int len, off;
1600
1601 mprev = NULL;
1602 for (m = m0; m != NULL; m = mprev->m_next) {
1603 /*
1604 * Regular mbufs are ignored unless there's a cluster
1605 * in front of it that we can use to coalesce. We do
1606 * the latter mainly so later clusters can be coalesced
1607 * also w/o having to handle them specially (i.e. convert
1608 * mbuf+cluster -> cluster). This optimization is heavily
1609 * influenced by the assumption that we're running over
1610 * Ethernet where MCLBYTES is large enough that the max
1611 * packet size will permit lots of coalescing into a
1612 * single cluster. This in turn permits efficient
1613 * crypto operations, especially when using hardware.
1614 */
1615 if ((m->m_flags & M_EXT) == 0) {
1616 if (mprev && (mprev->m_flags & M_EXT) &&
1617 m->m_len <= M_TRAILINGSPACE(mprev)) {
1618 /* XXX: this ignores mbuf types */
1619 memcpy(mtod(mprev, caddr_t) + mprev->m_len,
1620 mtod(m, caddr_t), m->m_len);
1621 mprev->m_len += m->m_len;
1622 mprev->m_next = m->m_next; /* unlink from chain */
1623 m_free(m); /* reclaim mbuf */
1624 #if 0
1625 newipsecstat.ips_mbcoalesced++;
1626 #endif
1627 } else {
1628 mprev = m;
1629 }
1630 continue;
1631 }
1632 /*
1633 * Writable mbufs are left alone (for now).
1634 */
1635 if (M_WRITABLE(m)) {
1636 mprev = m;
1637 continue;
1638 }
1639
1640 /*
1641 * Not writable, replace with a copy or coalesce with
1642 * the previous mbuf if possible (since we have to copy
1643 * it anyway, we try to reduce the number of mbufs and
1644 * clusters so that future work is easier).
1645 */
1646 KASSERT(m->m_flags & M_EXT, ("m_flags 0x%x", m->m_flags));
1647 /* NB: we only coalesce into a cluster or larger */
1648 if (mprev != NULL && (mprev->m_flags & M_EXT) &&
1649 m->m_len <= M_TRAILINGSPACE(mprev)) {
1650 /* XXX: this ignores mbuf types */
1651 memcpy(mtod(mprev, caddr_t) + mprev->m_len,
1652 mtod(m, caddr_t), m->m_len);
1653 mprev->m_len += m->m_len;
1654 mprev->m_next = m->m_next; /* unlink from chain */
1655 m_free(m); /* reclaim mbuf */
1656 #if 0
1657 newipsecstat.ips_clcoalesced++;
1658 #endif
1659 continue;
1660 }
1661
1662 /*
1663 * Allocate new space to hold the copy and copy the data.
1664 * We deal with jumbo mbufs (i.e. m_len > MCLBYTES) by
1665 * splitting them into clusters. We could just malloc a
1666 * buffer and make it external but too many device drivers
1667 * don't know how to break up the non-contiguous memory when
1668 * doing DMA.
1669 */
1670 n = m_getcl(how, m->m_type, m->m_flags & M_COPYFLAGS);
1671 if (n == NULL) {
1672 m_freem(m0);
1673 return (NULL);
1674 }
1675 if (m->m_flags & M_PKTHDR) {
1676 KASSERT(mprev == NULL, ("%s: m0 %p, m %p has M_PKTHDR",
1677 __func__, m0, m));
1678 m_move_pkthdr(n, m);
1679 }
1680 len = m->m_len;
1681 off = 0;
1682 mfirst = n;
1683 mlast = NULL;
1684 for (;;) {
1685 int cc = min(len, MCLBYTES);
1686 memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + off, cc);
1687 n->m_len = cc;
1688 if (mlast != NULL)
1689 mlast->m_next = n;
1690 mlast = n;
1691 #if 0
1692 newipsecstat.ips_clcopied++;
1693 #endif
1694
1695 len -= cc;
1696 if (len <= 0)
1697 break;
1698 off += cc;
1699
1700 n = m_getcl(how, m->m_type, m->m_flags & M_COPYFLAGS);
1701 if (n == NULL) {
1702 m_freem(mfirst);
1703 m_freem(m0);
1704 return (NULL);
1705 }
1706 }
1707 n->m_next = m->m_next;
1708 if (mprev == NULL)
1709 m0 = mfirst; /* new head of chain */
1710 else
1711 mprev->m_next = mfirst; /* replace old mbuf */
1712 m_free(m); /* release old mbuf */
1713 mprev = mfirst;
1714 }
1715 return (m0);
1716 }
1717
1718 #ifdef MBUF_PROFILING
1719
1720 #define MP_BUCKETS 32 /* don't just change this as things may overflow.*/
1721 struct mbufprofile {
1722 uintmax_t wasted[MP_BUCKETS];
1723 uintmax_t used[MP_BUCKETS];
1724 uintmax_t segments[MP_BUCKETS];
1725 } mbprof;
1726
1727 #define MP_MAXDIGITS 21 /* strlen("16,000,000,000,000,000,000") == 21 */
1728 #define MP_NUMLINES 6
1729 #define MP_NUMSPERLINE 16
1730 #define MP_EXTRABYTES 64 /* > strlen("used:\nwasted:\nsegments:\n") */
1731 /* work out max space needed and add a bit of spare space too */
1732 #define MP_MAXLINE ((MP_MAXDIGITS+1) * MP_NUMSPERLINE)
1733 #define MP_BUFSIZE ((MP_MAXLINE * MP_NUMLINES) + 1 + MP_EXTRABYTES)
1734
1735 char mbprofbuf[MP_BUFSIZE];
1736
1737 void
1738 m_profile(struct mbuf *m)
1739 {
1740 int segments = 0;
1741 int used = 0;
1742 int wasted = 0;
1743
1744 while (m) {
1745 segments++;
1746 used += m->m_len;
1747 if (m->m_flags & M_EXT) {
1748 wasted += MHLEN - sizeof(m->m_ext) +
1749 m->m_ext.ext_size - m->m_len;
1750 } else {
1751 if (m->m_flags & M_PKTHDR)
1752 wasted += MHLEN - m->m_len;
1753 else
1754 wasted += MLEN - m->m_len;
1755 }
1756 m = m->m_next;
1757 }
1758 /* be paranoid.. it helps */
1759 if (segments > MP_BUCKETS - 1)
1760 segments = MP_BUCKETS - 1;
1761 if (used > 100000)
1762 used = 100000;
1763 if (wasted > 100000)
1764 wasted = 100000;
1765 /* store in the appropriate bucket */
1766 /* don't bother locking. if it's slightly off, so what? */
1767 mbprof.segments[segments]++;
1768 mbprof.used[fls(used)]++;
1769 mbprof.wasted[fls(wasted)]++;
1770 }
1771
1772 static void
1773 mbprof_textify(void)
1774 {
1775 int offset;
1776 char *c;
1777 uint64_t *p;
1778
1779 p = &mbprof.wasted[0];
1780 c = mbprofbuf;
1781 offset = snprintf(c, MP_MAXLINE + 10,
1782 "wasted:\n"
1783 "%ju %ju %ju %ju %ju %ju %ju %ju "
1784 "%ju %ju %ju %ju %ju %ju %ju %ju\n",
1785 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
1786 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
1787 #ifdef BIG_ARRAY
1788 p = &mbprof.wasted[16];
1789 c += offset;
1790 offset = snprintf(c, MP_MAXLINE,
1791 "%ju %ju %ju %ju %ju %ju %ju %ju "
1792 "%ju %ju %ju %ju %ju %ju %ju %ju\n",
1793 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
1794 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
1795 #endif
1796 p = &mbprof.used[0];
1797 c += offset;
1798 offset = snprintf(c, MP_MAXLINE + 10,
1799 "used:\n"
1800 "%ju %ju %ju %ju %ju %ju %ju %ju "
1801 "%ju %ju %ju %ju %ju %ju %ju %ju\n",
1802 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
1803 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
1804 #ifdef BIG_ARRAY
1805 p = &mbprof.used[16];
1806 c += offset;
1807 offset = snprintf(c, MP_MAXLINE,
1808 "%ju %ju %ju %ju %ju %ju %ju %ju "
1809 "%ju %ju %ju %ju %ju %ju %ju %ju\n",
1810 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
1811 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
1812 #endif
1813 p = &mbprof.segments[0];
1814 c += offset;
1815 offset = snprintf(c, MP_MAXLINE + 10,
1816 "segments:\n"
1817 "%ju %ju %ju %ju %ju %ju %ju %ju "
1818 "%ju %ju %ju %ju %ju %ju %ju %ju\n",
1819 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
1820 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
1821 #ifdef BIG_ARRAY
1822 p = &mbprof.segments[16];
1823 c += offset;
1824 offset = snprintf(c, MP_MAXLINE,
1825 "%ju %ju %ju %ju %ju %ju %ju %ju "
1826 "%ju %ju %ju %ju %ju %ju %ju %jju",
1827 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
1828 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
1829 #endif
1830 }
1831
1832 static int
1833 mbprof_handler(SYSCTL_HANDLER_ARGS)
1834 {
1835 int error;
1836
1837 mbprof_textify();
1838 error = SYSCTL_OUT(req, mbprofbuf, strlen(mbprofbuf) + 1);
1839 return (error);
1840 }
1841
1842 static int
1843 mbprof_clr_handler(SYSCTL_HANDLER_ARGS)
1844 {
1845 int clear, error;
1846
1847 clear = 0;
1848 error = sysctl_handle_int(oidp, &clear, 0, req);
1849 if (error || !req->newptr)
1850 return (error);
1851
1852 if (clear) {
1853 bzero(&mbprof, sizeof(mbprof));
1854 }
1855
1856 return (error);
1857 }
1858
1859
1860 SYSCTL_PROC(_kern_ipc, OID_AUTO, mbufprofile, CTLTYPE_STRING|CTLFLAG_RD,
1861 NULL, 0, mbprof_handler, "A", "mbuf profiling statistics");
1862
1863 SYSCTL_PROC(_kern_ipc, OID_AUTO, mbufprofileclr, CTLTYPE_INT|CTLFLAG_RW,
1864 NULL, 0, mbprof_clr_handler, "I", "clear mbuf profiling statistics");
1865 #endif
1866
Cache object: bf7a912ccdccc11f455909994f44a3d1
|