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/10.3/sys/kern/uipc_mbuf.c 283198 2015-05-21 08:28:35Z ae $");
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
51 int max_linkhdr;
52 int max_protohdr;
53 int max_hdr;
54 int max_datalen;
55 #ifdef MBUF_STRESS_TEST
56 int m_defragpackets;
57 int m_defragbytes;
58 int m_defraguseless;
59 int m_defragfailure;
60 int m_defragrandomfailures;
61 #endif
62
63 /*
64 * sysctl(8) exported objects
65 */
66 SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RD,
67 &max_linkhdr, 0, "Size of largest link layer header");
68 SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RD,
69 &max_protohdr, 0, "Size of largest protocol layer header");
70 SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RD,
71 &max_hdr, 0, "Size of largest link plus protocol header");
72 SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RD,
73 &max_datalen, 0, "Minimum space left in mbuf after max_hdr");
74 #ifdef MBUF_STRESS_TEST
75 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD,
76 &m_defragpackets, 0, "");
77 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD,
78 &m_defragbytes, 0, "");
79 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD,
80 &m_defraguseless, 0, "");
81 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD,
82 &m_defragfailure, 0, "");
83 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW,
84 &m_defragrandomfailures, 0, "");
85 #endif
86
87 /*
88 * Ensure the correct size of various mbuf parameters. It could be off due
89 * to compiler-induced padding and alignment artifacts.
90 */
91 CTASSERT(sizeof(struct mbuf) == MSIZE);
92 CTASSERT(MSIZE - offsetof(struct mbuf, m_dat) == MLEN);
93 CTASSERT(MSIZE - offsetof(struct mbuf, m_pktdat) == MHLEN);
94
95 /*
96 * m_get2() allocates minimum mbuf that would fit "size" argument.
97 */
98 struct mbuf *
99 m_get2(int size, int how, short type, int flags)
100 {
101 struct mb_args args;
102 struct mbuf *m, *n;
103
104 args.flags = flags;
105 args.type = type;
106
107 if (size <= MHLEN || (size <= MLEN && (flags & M_PKTHDR) == 0))
108 return (uma_zalloc_arg(zone_mbuf, &args, how));
109 if (size <= MCLBYTES)
110 return (uma_zalloc_arg(zone_pack, &args, how));
111
112 if (size > MJUMPAGESIZE)
113 return (NULL);
114
115 m = uma_zalloc_arg(zone_mbuf, &args, how);
116 if (m == NULL)
117 return (NULL);
118
119 n = uma_zalloc_arg(zone_jumbop, m, how);
120 if (n == NULL) {
121 uma_zfree(zone_mbuf, m);
122 return (NULL);
123 }
124
125 return (m);
126 }
127
128 /*
129 * m_getjcl() returns an mbuf with a cluster of the specified size attached.
130 * For size it takes MCLBYTES, MJUMPAGESIZE, MJUM9BYTES, MJUM16BYTES.
131 */
132 struct mbuf *
133 m_getjcl(int how, short type, int flags, int size)
134 {
135 struct mb_args args;
136 struct mbuf *m, *n;
137 uma_zone_t zone;
138
139 if (size == MCLBYTES)
140 return m_getcl(how, type, flags);
141
142 args.flags = flags;
143 args.type = type;
144
145 m = uma_zalloc_arg(zone_mbuf, &args, how);
146 if (m == NULL)
147 return (NULL);
148
149 zone = m_getzone(size);
150 n = uma_zalloc_arg(zone, m, how);
151 if (n == NULL) {
152 uma_zfree(zone_mbuf, m);
153 return (NULL);
154 }
155 return (m);
156 }
157
158 /*
159 * Allocate a given length worth of mbufs and/or clusters (whatever fits
160 * best) and return a pointer to the top of the allocated chain. If an
161 * existing mbuf chain is provided, then we will append the new chain
162 * to the existing one but still return the top of the newly allocated
163 * chain.
164 */
165 struct mbuf *
166 m_getm2(struct mbuf *m, int len, int how, short type, int flags)
167 {
168 struct mbuf *mb, *nm = NULL, *mtail = NULL;
169
170 KASSERT(len >= 0, ("%s: len is < 0", __func__));
171
172 /* Validate flags. */
173 flags &= (M_PKTHDR | M_EOR);
174
175 /* Packet header mbuf must be first in chain. */
176 if ((flags & M_PKTHDR) && m != NULL)
177 flags &= ~M_PKTHDR;
178
179 /* Loop and append maximum sized mbufs to the chain tail. */
180 while (len > 0) {
181 if (len > MCLBYTES)
182 mb = m_getjcl(how, type, (flags & M_PKTHDR),
183 MJUMPAGESIZE);
184 else if (len >= MINCLSIZE)
185 mb = m_getcl(how, type, (flags & M_PKTHDR));
186 else if (flags & M_PKTHDR)
187 mb = m_gethdr(how, type);
188 else
189 mb = m_get(how, type);
190
191 /* Fail the whole operation if one mbuf can't be allocated. */
192 if (mb == NULL) {
193 if (nm != NULL)
194 m_freem(nm);
195 return (NULL);
196 }
197
198 /* Book keeping. */
199 len -= (mb->m_flags & M_EXT) ? mb->m_ext.ext_size :
200 ((mb->m_flags & M_PKTHDR) ? MHLEN : MLEN);
201 if (mtail != NULL)
202 mtail->m_next = mb;
203 else
204 nm = mb;
205 mtail = mb;
206 flags &= ~M_PKTHDR; /* Only valid on the first mbuf. */
207 }
208 if (flags & M_EOR)
209 mtail->m_flags |= M_EOR; /* Only valid on the last mbuf. */
210
211 /* If mbuf was supplied, append new chain to the end of it. */
212 if (m != NULL) {
213 for (mtail = m; mtail->m_next != NULL; mtail = mtail->m_next)
214 ;
215 mtail->m_next = nm;
216 mtail->m_flags &= ~M_EOR;
217 } else
218 m = nm;
219
220 return (m);
221 }
222
223 /*
224 * Free an entire chain of mbufs and associated external buffers, if
225 * applicable.
226 */
227 void
228 m_freem(struct mbuf *mb)
229 {
230
231 while (mb != NULL)
232 mb = m_free(mb);
233 }
234
235 /*-
236 * Configure a provided mbuf to refer to the provided external storage
237 * buffer and setup a reference count for said buffer. If the setting
238 * up of the reference count fails, the M_EXT bit will not be set. If
239 * successfull, the M_EXT bit is set in the mbuf's flags.
240 *
241 * Arguments:
242 * mb The existing mbuf to which to attach the provided buffer.
243 * buf The address of the provided external storage buffer.
244 * size The size of the provided buffer.
245 * freef A pointer to a routine that is responsible for freeing the
246 * provided external storage buffer.
247 * args A pointer to an argument structure (of any type) to be passed
248 * to the provided freef routine (may be NULL).
249 * flags Any other flags to be passed to the provided mbuf.
250 * type The type that the external storage buffer should be
251 * labeled with.
252 *
253 * Returns:
254 * Nothing.
255 */
256 int
257 m_extadd(struct mbuf *mb, caddr_t buf, u_int size,
258 int (*freef)(struct mbuf *, void *, void *), void *arg1, void *arg2,
259 int flags, int type, int wait)
260 {
261 KASSERT(type != EXT_CLUSTER, ("%s: EXT_CLUSTER not allowed", __func__));
262
263 if (type != EXT_EXTREF)
264 mb->m_ext.ref_cnt = uma_zalloc(zone_ext_refcnt, wait);
265
266 if (mb->m_ext.ref_cnt == NULL)
267 return (ENOMEM);
268
269 *(mb->m_ext.ref_cnt) = 1;
270 mb->m_flags |= (M_EXT | flags);
271 mb->m_ext.ext_buf = buf;
272 mb->m_data = mb->m_ext.ext_buf;
273 mb->m_ext.ext_size = size;
274 mb->m_ext.ext_free = freef;
275 mb->m_ext.ext_arg1 = arg1;
276 mb->m_ext.ext_arg2 = arg2;
277 mb->m_ext.ext_type = type;
278 mb->m_ext.ext_flags = 0;
279
280 return (0);
281 }
282
283 /*
284 * Non-directly-exported function to clean up after mbufs with M_EXT
285 * storage attached to them if the reference count hits 1.
286 */
287 void
288 mb_free_ext(struct mbuf *m)
289 {
290 int skipmbuf;
291
292 KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__));
293 KASSERT(m->m_ext.ref_cnt != NULL, ("%s: ref_cnt not set", __func__));
294
295 /*
296 * check if the header is embedded in the cluster
297 */
298 skipmbuf = (m->m_flags & M_NOFREE);
299
300 /* Free attached storage if this mbuf is the only reference to it. */
301 if (*(m->m_ext.ref_cnt) == 1 ||
302 atomic_fetchadd_int(m->m_ext.ref_cnt, -1) == 1) {
303 switch (m->m_ext.ext_type) {
304 case EXT_PACKET: /* The packet zone is special. */
305 if (*(m->m_ext.ref_cnt) == 0)
306 *(m->m_ext.ref_cnt) = 1;
307 uma_zfree(zone_pack, m);
308 return; /* Job done. */
309 case EXT_CLUSTER:
310 uma_zfree(zone_clust, m->m_ext.ext_buf);
311 break;
312 case EXT_JUMBOP:
313 uma_zfree(zone_jumbop, m->m_ext.ext_buf);
314 break;
315 case EXT_JUMBO9:
316 uma_zfree(zone_jumbo9, m->m_ext.ext_buf);
317 break;
318 case EXT_JUMBO16:
319 uma_zfree(zone_jumbo16, m->m_ext.ext_buf);
320 break;
321 case EXT_SFBUF:
322 case EXT_NET_DRV:
323 case EXT_MOD_TYPE:
324 case EXT_DISPOSABLE:
325 *(m->m_ext.ref_cnt) = 0;
326 uma_zfree(zone_ext_refcnt, __DEVOLATILE(u_int *,
327 m->m_ext.ref_cnt));
328 /* FALLTHROUGH */
329 case EXT_EXTREF:
330 KASSERT(m->m_ext.ext_free != NULL,
331 ("%s: ext_free not set", __func__));
332 (void)(*(m->m_ext.ext_free))(m, m->m_ext.ext_arg1,
333 m->m_ext.ext_arg2);
334 break;
335 default:
336 KASSERT(m->m_ext.ext_type == 0,
337 ("%s: unknown ext_type", __func__));
338 }
339 }
340 if (skipmbuf)
341 return;
342
343 /*
344 * Free this mbuf back to the mbuf zone with all m_ext
345 * information purged.
346 */
347 m->m_ext.ext_buf = NULL;
348 m->m_ext.ext_free = NULL;
349 m->m_ext.ext_arg1 = NULL;
350 m->m_ext.ext_arg2 = NULL;
351 m->m_ext.ref_cnt = NULL;
352 m->m_ext.ext_size = 0;
353 m->m_ext.ext_type = 0;
354 m->m_ext.ext_flags = 0;
355 m->m_flags &= ~M_EXT;
356 uma_zfree(zone_mbuf, m);
357 }
358
359 /*
360 * Attach the cluster from *m to *n, set up m_ext in *n
361 * and bump the refcount of the cluster.
362 */
363 static void
364 mb_dupcl(struct mbuf *n, struct mbuf *m)
365 {
366 KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__));
367 KASSERT(m->m_ext.ref_cnt != NULL, ("%s: ref_cnt not set", __func__));
368 KASSERT((n->m_flags & M_EXT) == 0, ("%s: M_EXT set", __func__));
369
370 if (*(m->m_ext.ref_cnt) == 1)
371 *(m->m_ext.ref_cnt) += 1;
372 else
373 atomic_add_int(m->m_ext.ref_cnt, 1);
374 n->m_ext.ext_buf = m->m_ext.ext_buf;
375 n->m_ext.ext_free = m->m_ext.ext_free;
376 n->m_ext.ext_arg1 = m->m_ext.ext_arg1;
377 n->m_ext.ext_arg2 = m->m_ext.ext_arg2;
378 n->m_ext.ext_size = m->m_ext.ext_size;
379 n->m_ext.ref_cnt = m->m_ext.ref_cnt;
380 n->m_ext.ext_type = m->m_ext.ext_type;
381 n->m_ext.ext_flags = m->m_ext.ext_flags;
382 n->m_flags |= M_EXT;
383 n->m_flags |= m->m_flags & M_RDONLY;
384 }
385
386 /*
387 * Clean up mbuf (chain) from any tags and packet headers.
388 * If "all" is set then the first mbuf in the chain will be
389 * cleaned too.
390 */
391 void
392 m_demote(struct mbuf *m0, int all)
393 {
394 struct mbuf *m;
395
396 for (m = all ? m0 : m0->m_next; m != NULL; m = m->m_next) {
397 if (m->m_flags & M_PKTHDR) {
398 m_tag_delete_chain(m, NULL);
399 m->m_flags &= ~M_PKTHDR;
400 bzero(&m->m_pkthdr, sizeof(struct pkthdr));
401 }
402 if (m != m0 && m->m_nextpkt != NULL) {
403 KASSERT(m->m_nextpkt == NULL,
404 ("%s: m_nextpkt not NULL", __func__));
405 m_freem(m->m_nextpkt);
406 m->m_nextpkt = NULL;
407 }
408 m->m_flags = m->m_flags & (M_EXT|M_RDONLY|M_NOFREE);
409 }
410 }
411
412 /*
413 * Sanity checks on mbuf (chain) for use in KASSERT() and general
414 * debugging.
415 * Returns 0 or panics when bad and 1 on all tests passed.
416 * Sanitize, 0 to run M_SANITY_ACTION, 1 to garble things so they
417 * blow up later.
418 */
419 int
420 m_sanity(struct mbuf *m0, int sanitize)
421 {
422 struct mbuf *m;
423 caddr_t a, b;
424 int pktlen = 0;
425
426 #ifdef INVARIANTS
427 #define M_SANITY_ACTION(s) panic("mbuf %p: " s, m)
428 #else
429 #define M_SANITY_ACTION(s) printf("mbuf %p: " s, m)
430 #endif
431
432 for (m = m0; m != NULL; m = m->m_next) {
433 /*
434 * Basic pointer checks. If any of these fails then some
435 * unrelated kernel memory before or after us is trashed.
436 * No way to recover from that.
437 */
438 a = ((m->m_flags & M_EXT) ? m->m_ext.ext_buf :
439 ((m->m_flags & M_PKTHDR) ? (caddr_t)(&m->m_pktdat) :
440 (caddr_t)(&m->m_dat)) );
441 b = (caddr_t)(a + (m->m_flags & M_EXT ? m->m_ext.ext_size :
442 ((m->m_flags & M_PKTHDR) ? MHLEN : MLEN)));
443 if ((caddr_t)m->m_data < a)
444 M_SANITY_ACTION("m_data outside mbuf data range left");
445 if ((caddr_t)m->m_data > b)
446 M_SANITY_ACTION("m_data outside mbuf data range right");
447 if ((caddr_t)m->m_data + m->m_len > b)
448 M_SANITY_ACTION("m_data + m_len exeeds mbuf space");
449
450 /* m->m_nextpkt may only be set on first mbuf in chain. */
451 if (m != m0 && m->m_nextpkt != NULL) {
452 if (sanitize) {
453 m_freem(m->m_nextpkt);
454 m->m_nextpkt = (struct mbuf *)0xDEADC0DE;
455 } else
456 M_SANITY_ACTION("m->m_nextpkt on in-chain mbuf");
457 }
458
459 /* packet length (not mbuf length!) calculation */
460 if (m0->m_flags & M_PKTHDR)
461 pktlen += m->m_len;
462
463 /* m_tags may only be attached to first mbuf in chain. */
464 if (m != m0 && m->m_flags & M_PKTHDR &&
465 !SLIST_EMPTY(&m->m_pkthdr.tags)) {
466 if (sanitize) {
467 m_tag_delete_chain(m, NULL);
468 /* put in 0xDEADC0DE perhaps? */
469 } else
470 M_SANITY_ACTION("m_tags on in-chain mbuf");
471 }
472
473 /* M_PKTHDR may only be set on first mbuf in chain */
474 if (m != m0 && m->m_flags & M_PKTHDR) {
475 if (sanitize) {
476 bzero(&m->m_pkthdr, sizeof(m->m_pkthdr));
477 m->m_flags &= ~M_PKTHDR;
478 /* put in 0xDEADCODE and leave hdr flag in */
479 } else
480 M_SANITY_ACTION("M_PKTHDR on in-chain mbuf");
481 }
482 }
483 m = m0;
484 if (pktlen && pktlen != m->m_pkthdr.len) {
485 if (sanitize)
486 m->m_pkthdr.len = 0;
487 else
488 M_SANITY_ACTION("m_pkthdr.len != mbuf chain length");
489 }
490 return 1;
491
492 #undef M_SANITY_ACTION
493 }
494
495
496 /*
497 * "Move" mbuf pkthdr from "from" to "to".
498 * "from" must have M_PKTHDR set, and "to" must be empty.
499 */
500 void
501 m_move_pkthdr(struct mbuf *to, struct mbuf *from)
502 {
503
504 #if 0
505 /* see below for why these are not enabled */
506 M_ASSERTPKTHDR(to);
507 /* Note: with MAC, this may not be a good assertion. */
508 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags),
509 ("m_move_pkthdr: to has tags"));
510 #endif
511 #ifdef MAC
512 /*
513 * XXXMAC: It could be this should also occur for non-MAC?
514 */
515 if (to->m_flags & M_PKTHDR)
516 m_tag_delete_chain(to, NULL);
517 #endif
518 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT);
519 if ((to->m_flags & M_EXT) == 0)
520 to->m_data = to->m_pktdat;
521 to->m_pkthdr = from->m_pkthdr; /* especially tags */
522 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */
523 from->m_flags &= ~M_PKTHDR;
524 }
525
526 /*
527 * Duplicate "from"'s mbuf pkthdr in "to".
528 * "from" must have M_PKTHDR set, and "to" must be empty.
529 * In particular, this does a deep copy of the packet tags.
530 */
531 int
532 m_dup_pkthdr(struct mbuf *to, struct mbuf *from, int how)
533 {
534
535 #if 0
536 /*
537 * The mbuf allocator only initializes the pkthdr
538 * when the mbuf is allocated with m_gethdr(). Many users
539 * (e.g. m_copy*, m_prepend) use m_get() and then
540 * smash the pkthdr as needed causing these
541 * assertions to trip. For now just disable them.
542 */
543 M_ASSERTPKTHDR(to);
544 /* Note: with MAC, this may not be a good assertion. */
545 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), ("m_dup_pkthdr: to has tags"));
546 #endif
547 MBUF_CHECKSLEEP(how);
548 #ifdef MAC
549 if (to->m_flags & M_PKTHDR)
550 m_tag_delete_chain(to, NULL);
551 #endif
552 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT);
553 if ((to->m_flags & M_EXT) == 0)
554 to->m_data = to->m_pktdat;
555 to->m_pkthdr = from->m_pkthdr;
556 SLIST_INIT(&to->m_pkthdr.tags);
557 return (m_tag_copy_chain(to, from, MBTOM(how)));
558 }
559
560 /*
561 * Lesser-used path for M_PREPEND:
562 * allocate new mbuf to prepend to chain,
563 * copy junk along.
564 */
565 struct mbuf *
566 m_prepend(struct mbuf *m, int len, int how)
567 {
568 struct mbuf *mn;
569
570 if (m->m_flags & M_PKTHDR)
571 mn = m_gethdr(how, m->m_type);
572 else
573 mn = m_get(how, m->m_type);
574 if (mn == NULL) {
575 m_freem(m);
576 return (NULL);
577 }
578 if (m->m_flags & M_PKTHDR)
579 m_move_pkthdr(mn, m);
580 mn->m_next = m;
581 m = mn;
582 if(m->m_flags & M_PKTHDR) {
583 if (len < MHLEN)
584 MH_ALIGN(m, len);
585 } else {
586 if (len < MLEN)
587 M_ALIGN(m, len);
588 }
589 m->m_len = len;
590 return (m);
591 }
592
593 /*
594 * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
595 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf.
596 * The wait parameter is a choice of M_WAITOK/M_NOWAIT from caller.
597 * Note that the copy is read-only, because clusters are not copied,
598 * only their reference counts are incremented.
599 */
600 struct mbuf *
601 m_copym(struct mbuf *m, int off0, int len, int wait)
602 {
603 struct mbuf *n, **np;
604 int off = off0;
605 struct mbuf *top;
606 int copyhdr = 0;
607
608 KASSERT(off >= 0, ("m_copym, negative off %d", off));
609 KASSERT(len >= 0, ("m_copym, negative len %d", len));
610 MBUF_CHECKSLEEP(wait);
611 if (off == 0 && m->m_flags & M_PKTHDR)
612 copyhdr = 1;
613 while (off > 0) {
614 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain"));
615 if (off < m->m_len)
616 break;
617 off -= m->m_len;
618 m = m->m_next;
619 }
620 np = ⊤
621 top = 0;
622 while (len > 0) {
623 if (m == NULL) {
624 KASSERT(len == M_COPYALL,
625 ("m_copym, length > size of mbuf chain"));
626 break;
627 }
628 if (copyhdr)
629 n = m_gethdr(wait, m->m_type);
630 else
631 n = m_get(wait, m->m_type);
632 *np = n;
633 if (n == NULL)
634 goto nospace;
635 if (copyhdr) {
636 if (!m_dup_pkthdr(n, m, wait))
637 goto nospace;
638 if (len == M_COPYALL)
639 n->m_pkthdr.len -= off0;
640 else
641 n->m_pkthdr.len = len;
642 copyhdr = 0;
643 }
644 n->m_len = min(len, m->m_len - off);
645 if (m->m_flags & M_EXT) {
646 n->m_data = m->m_data + off;
647 mb_dupcl(n, m);
648 } else
649 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
650 (u_int)n->m_len);
651 if (len != M_COPYALL)
652 len -= n->m_len;
653 off = 0;
654 m = m->m_next;
655 np = &n->m_next;
656 }
657
658 return (top);
659 nospace:
660 m_freem(top);
661 return (NULL);
662 }
663
664 /*
665 * Returns mbuf chain with new head for the prepending case.
666 * Copies from mbuf (chain) n from off for len to mbuf (chain) m
667 * either prepending or appending the data.
668 * The resulting mbuf (chain) m is fully writeable.
669 * m is destination (is made writeable)
670 * n is source, off is offset in source, len is len from offset
671 * dir, 0 append, 1 prepend
672 * how, wait or nowait
673 */
674
675 static int
676 m_bcopyxxx(void *s, void *t, u_int len)
677 {
678 bcopy(s, t, (size_t)len);
679 return 0;
680 }
681
682 struct mbuf *
683 m_copymdata(struct mbuf *m, struct mbuf *n, int off, int len,
684 int prep, int how)
685 {
686 struct mbuf *mm, *x, *z, *prev = NULL;
687 caddr_t p;
688 int i, nlen = 0;
689 caddr_t buf[MLEN];
690
691 KASSERT(m != NULL && n != NULL, ("m_copymdata, no target or source"));
692 KASSERT(off >= 0, ("m_copymdata, negative off %d", off));
693 KASSERT(len >= 0, ("m_copymdata, negative len %d", len));
694 KASSERT(prep == 0 || prep == 1, ("m_copymdata, unknown direction %d", prep));
695
696 mm = m;
697 if (!prep) {
698 while(mm->m_next) {
699 prev = mm;
700 mm = mm->m_next;
701 }
702 }
703 for (z = n; z != NULL; z = z->m_next)
704 nlen += z->m_len;
705 if (len == M_COPYALL)
706 len = nlen - off;
707 if (off + len > nlen || len < 1)
708 return NULL;
709
710 if (!M_WRITABLE(mm)) {
711 /* XXX: Use proper m_xxx function instead. */
712 x = m_getcl(how, MT_DATA, mm->m_flags);
713 if (x == NULL)
714 return NULL;
715 bcopy(mm->m_ext.ext_buf, x->m_ext.ext_buf, x->m_ext.ext_size);
716 p = x->m_ext.ext_buf + (mm->m_data - mm->m_ext.ext_buf);
717 x->m_data = p;
718 mm->m_next = NULL;
719 if (mm != m)
720 prev->m_next = x;
721 m_free(mm);
722 mm = x;
723 }
724
725 /*
726 * Append/prepend the data. Allocating mbufs as necessary.
727 */
728 /* Shortcut if enough free space in first/last mbuf. */
729 if (!prep && M_TRAILINGSPACE(mm) >= len) {
730 m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t) +
731 mm->m_len);
732 mm->m_len += len;
733 mm->m_pkthdr.len += len;
734 return m;
735 }
736 if (prep && M_LEADINGSPACE(mm) >= len) {
737 mm->m_data = mtod(mm, caddr_t) - len;
738 m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t));
739 mm->m_len += len;
740 mm->m_pkthdr.len += len;
741 return mm;
742 }
743
744 /* Expand first/last mbuf to cluster if possible. */
745 if (!prep && !(mm->m_flags & M_EXT) && len > M_TRAILINGSPACE(mm)) {
746 bcopy(mm->m_data, &buf, mm->m_len);
747 m_clget(mm, how);
748 if (!(mm->m_flags & M_EXT))
749 return NULL;
750 bcopy(&buf, mm->m_ext.ext_buf, mm->m_len);
751 mm->m_data = mm->m_ext.ext_buf;
752 }
753 if (prep && !(mm->m_flags & M_EXT) && len > M_LEADINGSPACE(mm)) {
754 bcopy(mm->m_data, &buf, mm->m_len);
755 m_clget(mm, how);
756 if (!(mm->m_flags & M_EXT))
757 return NULL;
758 bcopy(&buf, (caddr_t *)mm->m_ext.ext_buf +
759 mm->m_ext.ext_size - mm->m_len, mm->m_len);
760 mm->m_data = (caddr_t)mm->m_ext.ext_buf +
761 mm->m_ext.ext_size - mm->m_len;
762 }
763
764 /* Append/prepend as many mbuf (clusters) as necessary to fit len. */
765 if (!prep && len > M_TRAILINGSPACE(mm)) {
766 if (!m_getm(mm, len - M_TRAILINGSPACE(mm), how, MT_DATA))
767 return NULL;
768 }
769 if (prep && len > M_LEADINGSPACE(mm)) {
770 if (!(z = m_getm(NULL, len - M_LEADINGSPACE(mm), how, MT_DATA)))
771 return NULL;
772 i = 0;
773 for (x = z; x != NULL; x = x->m_next) {
774 i += x->m_flags & M_EXT ? x->m_ext.ext_size :
775 (x->m_flags & M_PKTHDR ? MHLEN : MLEN);
776 if (!x->m_next)
777 break;
778 }
779 z->m_data += i - len;
780 m_move_pkthdr(mm, z);
781 x->m_next = mm;
782 mm = z;
783 }
784
785 /* Seek to start position in source mbuf. Optimization for long chains. */
786 while (off > 0) {
787 if (off < n->m_len)
788 break;
789 off -= n->m_len;
790 n = n->m_next;
791 }
792
793 /* Copy data into target mbuf. */
794 z = mm;
795 while (len > 0) {
796 KASSERT(z != NULL, ("m_copymdata, falling off target edge"));
797 i = M_TRAILINGSPACE(z);
798 m_apply(n, off, i, m_bcopyxxx, mtod(z, caddr_t) + z->m_len);
799 z->m_len += i;
800 /* fixup pkthdr.len if necessary */
801 if ((prep ? mm : m)->m_flags & M_PKTHDR)
802 (prep ? mm : m)->m_pkthdr.len += i;
803 off += i;
804 len -= i;
805 z = z->m_next;
806 }
807 return (prep ? mm : m);
808 }
809
810 /*
811 * Copy an entire packet, including header (which must be present).
812 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
813 * Note that the copy is read-only, because clusters are not copied,
814 * only their reference counts are incremented.
815 * Preserve alignment of the first mbuf so if the creator has left
816 * some room at the beginning (e.g. for inserting protocol headers)
817 * the copies still have the room available.
818 */
819 struct mbuf *
820 m_copypacket(struct mbuf *m, int how)
821 {
822 struct mbuf *top, *n, *o;
823
824 MBUF_CHECKSLEEP(how);
825 n = m_get(how, m->m_type);
826 top = n;
827 if (n == NULL)
828 goto nospace;
829
830 if (!m_dup_pkthdr(n, m, how))
831 goto nospace;
832 n->m_len = m->m_len;
833 if (m->m_flags & M_EXT) {
834 n->m_data = m->m_data;
835 mb_dupcl(n, m);
836 } else {
837 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat );
838 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
839 }
840
841 m = m->m_next;
842 while (m) {
843 o = m_get(how, m->m_type);
844 if (o == NULL)
845 goto nospace;
846
847 n->m_next = o;
848 n = n->m_next;
849
850 n->m_len = m->m_len;
851 if (m->m_flags & M_EXT) {
852 n->m_data = m->m_data;
853 mb_dupcl(n, m);
854 } else {
855 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
856 }
857
858 m = m->m_next;
859 }
860 return top;
861 nospace:
862 m_freem(top);
863 return (NULL);
864 }
865
866 /*
867 * Copy data from an mbuf chain starting "off" bytes from the beginning,
868 * continuing for "len" bytes, into the indicated buffer.
869 */
870 void
871 m_copydata(const struct mbuf *m, int off, int len, caddr_t cp)
872 {
873 u_int count;
874
875 KASSERT(off >= 0, ("m_copydata, negative off %d", off));
876 KASSERT(len >= 0, ("m_copydata, negative len %d", len));
877 while (off > 0) {
878 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain"));
879 if (off < m->m_len)
880 break;
881 off -= m->m_len;
882 m = m->m_next;
883 }
884 while (len > 0) {
885 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain"));
886 count = min(m->m_len - off, len);
887 bcopy(mtod(m, caddr_t) + off, cp, count);
888 len -= count;
889 cp += count;
890 off = 0;
891 m = m->m_next;
892 }
893 }
894
895 /*
896 * Copy a packet header mbuf chain into a completely new chain, including
897 * copying any mbuf clusters. Use this instead of m_copypacket() when
898 * you need a writable copy of an mbuf chain.
899 */
900 struct mbuf *
901 m_dup(struct mbuf *m, int how)
902 {
903 struct mbuf **p, *top = NULL;
904 int remain, moff, nsize;
905
906 MBUF_CHECKSLEEP(how);
907 /* Sanity check */
908 if (m == NULL)
909 return (NULL);
910 M_ASSERTPKTHDR(m);
911
912 /* While there's more data, get a new mbuf, tack it on, and fill it */
913 remain = m->m_pkthdr.len;
914 moff = 0;
915 p = ⊤
916 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */
917 struct mbuf *n;
918
919 /* Get the next new mbuf */
920 if (remain >= MINCLSIZE) {
921 n = m_getcl(how, m->m_type, 0);
922 nsize = MCLBYTES;
923 } else {
924 n = m_get(how, m->m_type);
925 nsize = MLEN;
926 }
927 if (n == NULL)
928 goto nospace;
929
930 if (top == NULL) { /* First one, must be PKTHDR */
931 if (!m_dup_pkthdr(n, m, how)) {
932 m_free(n);
933 goto nospace;
934 }
935 if ((n->m_flags & M_EXT) == 0)
936 nsize = MHLEN;
937 n->m_flags &= ~M_RDONLY;
938 }
939 n->m_len = 0;
940
941 /* Link it into the new chain */
942 *p = n;
943 p = &n->m_next;
944
945 /* Copy data from original mbuf(s) into new mbuf */
946 while (n->m_len < nsize && m != NULL) {
947 int chunk = min(nsize - n->m_len, m->m_len - moff);
948
949 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
950 moff += chunk;
951 n->m_len += chunk;
952 remain -= chunk;
953 if (moff == m->m_len) {
954 m = m->m_next;
955 moff = 0;
956 }
957 }
958
959 /* Check correct total mbuf length */
960 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL),
961 ("%s: bogus m_pkthdr.len", __func__));
962 }
963 return (top);
964
965 nospace:
966 m_freem(top);
967 return (NULL);
968 }
969
970 /*
971 * Concatenate mbuf chain n to m.
972 * Both chains must be of the same type (e.g. MT_DATA).
973 * Any m_pkthdr is not updated.
974 */
975 void
976 m_cat(struct mbuf *m, struct mbuf *n)
977 {
978 while (m->m_next)
979 m = m->m_next;
980 while (n) {
981 if (!M_WRITABLE(m) ||
982 M_TRAILINGSPACE(m) < n->m_len) {
983 /* just join the two chains */
984 m->m_next = n;
985 return;
986 }
987 /* splat the data from one into the other */
988 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
989 (u_int)n->m_len);
990 m->m_len += n->m_len;
991 n = m_free(n);
992 }
993 }
994
995 void
996 m_adj(struct mbuf *mp, int req_len)
997 {
998 int len = req_len;
999 struct mbuf *m;
1000 int count;
1001
1002 if ((m = mp) == NULL)
1003 return;
1004 if (len >= 0) {
1005 /*
1006 * Trim from head.
1007 */
1008 while (m != NULL && len > 0) {
1009 if (m->m_len <= len) {
1010 len -= m->m_len;
1011 m->m_len = 0;
1012 m = m->m_next;
1013 } else {
1014 m->m_len -= len;
1015 m->m_data += len;
1016 len = 0;
1017 }
1018 }
1019 if (mp->m_flags & M_PKTHDR)
1020 mp->m_pkthdr.len -= (req_len - len);
1021 } else {
1022 /*
1023 * Trim from tail. Scan the mbuf chain,
1024 * calculating its length and finding the last mbuf.
1025 * If the adjustment only affects this mbuf, then just
1026 * adjust and return. Otherwise, rescan and truncate
1027 * after the remaining size.
1028 */
1029 len = -len;
1030 count = 0;
1031 for (;;) {
1032 count += m->m_len;
1033 if (m->m_next == (struct mbuf *)0)
1034 break;
1035 m = m->m_next;
1036 }
1037 if (m->m_len >= len) {
1038 m->m_len -= len;
1039 if (mp->m_flags & M_PKTHDR)
1040 mp->m_pkthdr.len -= len;
1041 return;
1042 }
1043 count -= len;
1044 if (count < 0)
1045 count = 0;
1046 /*
1047 * Correct length for chain is "count".
1048 * Find the mbuf with last data, adjust its length,
1049 * and toss data from remaining mbufs on chain.
1050 */
1051 m = mp;
1052 if (m->m_flags & M_PKTHDR)
1053 m->m_pkthdr.len = count;
1054 for (; m; m = m->m_next) {
1055 if (m->m_len >= count) {
1056 m->m_len = count;
1057 if (m->m_next != NULL) {
1058 m_freem(m->m_next);
1059 m->m_next = NULL;
1060 }
1061 break;
1062 }
1063 count -= m->m_len;
1064 }
1065 }
1066 }
1067
1068 /*
1069 * Rearange an mbuf chain so that len bytes are contiguous
1070 * and in the data area of an mbuf (so that mtod will work
1071 * for a structure of size len). Returns the resulting
1072 * mbuf chain on success, frees it and returns null on failure.
1073 * If there is room, it will add up to max_protohdr-len extra bytes to the
1074 * contiguous region in an attempt to avoid being called next time.
1075 */
1076 struct mbuf *
1077 m_pullup(struct mbuf *n, int len)
1078 {
1079 struct mbuf *m;
1080 int count;
1081 int space;
1082
1083 /*
1084 * If first mbuf has no cluster, and has room for len bytes
1085 * without shifting current data, pullup into it,
1086 * otherwise allocate a new mbuf to prepend to the chain.
1087 */
1088 if ((n->m_flags & M_EXT) == 0 &&
1089 n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
1090 if (n->m_len >= len)
1091 return (n);
1092 m = n;
1093 n = n->m_next;
1094 len -= m->m_len;
1095 } else {
1096 if (len > MHLEN)
1097 goto bad;
1098 m = m_get(M_NOWAIT, n->m_type);
1099 if (m == NULL)
1100 goto bad;
1101 if (n->m_flags & M_PKTHDR)
1102 m_move_pkthdr(m, n);
1103 }
1104 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1105 do {
1106 count = min(min(max(len, max_protohdr), space), n->m_len);
1107 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
1108 (u_int)count);
1109 len -= count;
1110 m->m_len += count;
1111 n->m_len -= count;
1112 space -= count;
1113 if (n->m_len)
1114 n->m_data += count;
1115 else
1116 n = m_free(n);
1117 } while (len > 0 && n);
1118 if (len > 0) {
1119 (void) m_free(m);
1120 goto bad;
1121 }
1122 m->m_next = n;
1123 return (m);
1124 bad:
1125 m_freem(n);
1126 return (NULL);
1127 }
1128
1129 /*
1130 * Like m_pullup(), except a new mbuf is always allocated, and we allow
1131 * the amount of empty space before the data in the new mbuf to be specified
1132 * (in the event that the caller expects to prepend later).
1133 */
1134 int MSFail;
1135
1136 struct mbuf *
1137 m_copyup(struct mbuf *n, int len, int dstoff)
1138 {
1139 struct mbuf *m;
1140 int count, space;
1141
1142 if (len > (MHLEN - dstoff))
1143 goto bad;
1144 m = m_get(M_NOWAIT, n->m_type);
1145 if (m == NULL)
1146 goto bad;
1147 if (n->m_flags & M_PKTHDR)
1148 m_move_pkthdr(m, n);
1149 m->m_data += dstoff;
1150 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1151 do {
1152 count = min(min(max(len, max_protohdr), space), n->m_len);
1153 memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
1154 (unsigned)count);
1155 len -= count;
1156 m->m_len += count;
1157 n->m_len -= count;
1158 space -= count;
1159 if (n->m_len)
1160 n->m_data += count;
1161 else
1162 n = m_free(n);
1163 } while (len > 0 && n);
1164 if (len > 0) {
1165 (void) m_free(m);
1166 goto bad;
1167 }
1168 m->m_next = n;
1169 return (m);
1170 bad:
1171 m_freem(n);
1172 MSFail++;
1173 return (NULL);
1174 }
1175
1176 /*
1177 * Partition an mbuf chain in two pieces, returning the tail --
1178 * all but the first len0 bytes. In case of failure, it returns NULL and
1179 * attempts to restore the chain to its original state.
1180 *
1181 * Note that the resulting mbufs might be read-only, because the new
1182 * mbuf can end up sharing an mbuf cluster with the original mbuf if
1183 * the "breaking point" happens to lie within a cluster mbuf. Use the
1184 * M_WRITABLE() macro to check for this case.
1185 */
1186 struct mbuf *
1187 m_split(struct mbuf *m0, int len0, int wait)
1188 {
1189 struct mbuf *m, *n;
1190 u_int len = len0, remain;
1191
1192 MBUF_CHECKSLEEP(wait);
1193 for (m = m0; m && len > m->m_len; m = m->m_next)
1194 len -= m->m_len;
1195 if (m == NULL)
1196 return (NULL);
1197 remain = m->m_len - len;
1198 if (m0->m_flags & M_PKTHDR && remain == 0) {
1199 n = m_gethdr(wait, m0->m_type);
1200 if (n == NULL)
1201 return (NULL);
1202 n->m_next = m->m_next;
1203 m->m_next = NULL;
1204 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
1205 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
1206 m0->m_pkthdr.len = len0;
1207 return (n);
1208 } else if (m0->m_flags & M_PKTHDR) {
1209 n = m_gethdr(wait, m0->m_type);
1210 if (n == NULL)
1211 return (NULL);
1212 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
1213 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
1214 m0->m_pkthdr.len = len0;
1215 if (m->m_flags & M_EXT)
1216 goto extpacket;
1217 if (remain > MHLEN) {
1218 /* m can't be the lead packet */
1219 MH_ALIGN(n, 0);
1220 n->m_next = m_split(m, len, wait);
1221 if (n->m_next == NULL) {
1222 (void) m_free(n);
1223 return (NULL);
1224 } else {
1225 n->m_len = 0;
1226 return (n);
1227 }
1228 } else
1229 MH_ALIGN(n, remain);
1230 } else if (remain == 0) {
1231 n = m->m_next;
1232 m->m_next = NULL;
1233 return (n);
1234 } else {
1235 n = m_get(wait, m->m_type);
1236 if (n == NULL)
1237 return (NULL);
1238 M_ALIGN(n, remain);
1239 }
1240 extpacket:
1241 if (m->m_flags & M_EXT) {
1242 n->m_data = m->m_data + len;
1243 mb_dupcl(n, m);
1244 } else {
1245 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
1246 }
1247 n->m_len = remain;
1248 m->m_len = len;
1249 n->m_next = m->m_next;
1250 m->m_next = NULL;
1251 return (n);
1252 }
1253 /*
1254 * Routine to copy from device local memory into mbufs.
1255 * Note that `off' argument is offset into first mbuf of target chain from
1256 * which to begin copying the data to.
1257 */
1258 struct mbuf *
1259 m_devget(char *buf, int totlen, int off, struct ifnet *ifp,
1260 void (*copy)(char *from, caddr_t to, u_int len))
1261 {
1262 struct mbuf *m;
1263 struct mbuf *top = NULL, **mp = ⊤
1264 int len;
1265
1266 if (off < 0 || off > MHLEN)
1267 return (NULL);
1268
1269 while (totlen > 0) {
1270 if (top == NULL) { /* First one, must be PKTHDR */
1271 if (totlen + off >= MINCLSIZE) {
1272 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
1273 len = MCLBYTES;
1274 } else {
1275 m = m_gethdr(M_NOWAIT, MT_DATA);
1276 len = MHLEN;
1277
1278 /* Place initial small packet/header at end of mbuf */
1279 if (m && totlen + off + max_linkhdr <= MLEN) {
1280 m->m_data += max_linkhdr;
1281 len -= max_linkhdr;
1282 }
1283 }
1284 if (m == NULL)
1285 return NULL;
1286 m->m_pkthdr.rcvif = ifp;
1287 m->m_pkthdr.len = totlen;
1288 } else {
1289 if (totlen + off >= MINCLSIZE) {
1290 m = m_getcl(M_NOWAIT, MT_DATA, 0);
1291 len = MCLBYTES;
1292 } else {
1293 m = m_get(M_NOWAIT, MT_DATA);
1294 len = MLEN;
1295 }
1296 if (m == NULL) {
1297 m_freem(top);
1298 return NULL;
1299 }
1300 }
1301 if (off) {
1302 m->m_data += off;
1303 len -= off;
1304 off = 0;
1305 }
1306 m->m_len = len = min(totlen, len);
1307 if (copy)
1308 copy(buf, mtod(m, caddr_t), (u_int)len);
1309 else
1310 bcopy(buf, mtod(m, caddr_t), (u_int)len);
1311 buf += len;
1312 *mp = m;
1313 mp = &m->m_next;
1314 totlen -= len;
1315 }
1316 return (top);
1317 }
1318
1319 /*
1320 * Copy data from a buffer back into the indicated mbuf chain,
1321 * starting "off" bytes from the beginning, extending the mbuf
1322 * chain if necessary.
1323 */
1324 void
1325 m_copyback(struct mbuf *m0, int off, int len, c_caddr_t cp)
1326 {
1327 int mlen;
1328 struct mbuf *m = m0, *n;
1329 int totlen = 0;
1330
1331 if (m0 == NULL)
1332 return;
1333 while (off > (mlen = m->m_len)) {
1334 off -= mlen;
1335 totlen += mlen;
1336 if (m->m_next == NULL) {
1337 n = m_get(M_NOWAIT, m->m_type);
1338 if (n == NULL)
1339 goto out;
1340 bzero(mtod(n, caddr_t), MLEN);
1341 n->m_len = min(MLEN, len + off);
1342 m->m_next = n;
1343 }
1344 m = m->m_next;
1345 }
1346 while (len > 0) {
1347 if (m->m_next == NULL && (len > m->m_len - off)) {
1348 m->m_len += min(len - (m->m_len - off),
1349 M_TRAILINGSPACE(m));
1350 }
1351 mlen = min (m->m_len - off, len);
1352 bcopy(cp, off + mtod(m, caddr_t), (u_int)mlen);
1353 cp += mlen;
1354 len -= mlen;
1355 mlen += off;
1356 off = 0;
1357 totlen += mlen;
1358 if (len == 0)
1359 break;
1360 if (m->m_next == NULL) {
1361 n = m_get(M_NOWAIT, m->m_type);
1362 if (n == NULL)
1363 break;
1364 n->m_len = min(MLEN, len);
1365 m->m_next = n;
1366 }
1367 m = m->m_next;
1368 }
1369 out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
1370 m->m_pkthdr.len = totlen;
1371 }
1372
1373 /*
1374 * Append the specified data to the indicated mbuf chain,
1375 * Extend the mbuf chain if the new data does not fit in
1376 * existing space.
1377 *
1378 * Return 1 if able to complete the job; otherwise 0.
1379 */
1380 int
1381 m_append(struct mbuf *m0, int len, c_caddr_t cp)
1382 {
1383 struct mbuf *m, *n;
1384 int remainder, space;
1385
1386 for (m = m0; m->m_next != NULL; m = m->m_next)
1387 ;
1388 remainder = len;
1389 space = M_TRAILINGSPACE(m);
1390 if (space > 0) {
1391 /*
1392 * Copy into available space.
1393 */
1394 if (space > remainder)
1395 space = remainder;
1396 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1397 m->m_len += space;
1398 cp += space, remainder -= space;
1399 }
1400 while (remainder > 0) {
1401 /*
1402 * Allocate a new mbuf; could check space
1403 * and allocate a cluster instead.
1404 */
1405 n = m_get(M_NOWAIT, m->m_type);
1406 if (n == NULL)
1407 break;
1408 n->m_len = min(MLEN, remainder);
1409 bcopy(cp, mtod(n, caddr_t), n->m_len);
1410 cp += n->m_len, remainder -= n->m_len;
1411 m->m_next = n;
1412 m = n;
1413 }
1414 if (m0->m_flags & M_PKTHDR)
1415 m0->m_pkthdr.len += len - remainder;
1416 return (remainder == 0);
1417 }
1418
1419 /*
1420 * Apply function f to the data in an mbuf chain starting "off" bytes from
1421 * the beginning, continuing for "len" bytes.
1422 */
1423 int
1424 m_apply(struct mbuf *m, int off, int len,
1425 int (*f)(void *, void *, u_int), void *arg)
1426 {
1427 u_int count;
1428 int rval;
1429
1430 KASSERT(off >= 0, ("m_apply, negative off %d", off));
1431 KASSERT(len >= 0, ("m_apply, negative len %d", len));
1432 while (off > 0) {
1433 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1434 if (off < m->m_len)
1435 break;
1436 off -= m->m_len;
1437 m = m->m_next;
1438 }
1439 while (len > 0) {
1440 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1441 count = min(m->m_len - off, len);
1442 rval = (*f)(arg, mtod(m, caddr_t) + off, count);
1443 if (rval)
1444 return (rval);
1445 len -= count;
1446 off = 0;
1447 m = m->m_next;
1448 }
1449 return (0);
1450 }
1451
1452 /*
1453 * Return a pointer to mbuf/offset of location in mbuf chain.
1454 */
1455 struct mbuf *
1456 m_getptr(struct mbuf *m, int loc, int *off)
1457 {
1458
1459 while (loc >= 0) {
1460 /* Normal end of search. */
1461 if (m->m_len > loc) {
1462 *off = loc;
1463 return (m);
1464 } else {
1465 loc -= m->m_len;
1466 if (m->m_next == NULL) {
1467 if (loc == 0) {
1468 /* Point at the end of valid data. */
1469 *off = m->m_len;
1470 return (m);
1471 }
1472 return (NULL);
1473 }
1474 m = m->m_next;
1475 }
1476 }
1477 return (NULL);
1478 }
1479
1480 void
1481 m_print(const struct mbuf *m, int maxlen)
1482 {
1483 int len;
1484 int pdata;
1485 const struct mbuf *m2;
1486
1487 if (m == NULL) {
1488 printf("mbuf: %p\n", m);
1489 return;
1490 }
1491
1492 if (m->m_flags & M_PKTHDR)
1493 len = m->m_pkthdr.len;
1494 else
1495 len = -1;
1496 m2 = m;
1497 while (m2 != NULL && (len == -1 || len)) {
1498 pdata = m2->m_len;
1499 if (maxlen != -1 && pdata > maxlen)
1500 pdata = maxlen;
1501 printf("mbuf: %p len: %d, next: %p, %b%s", m2, m2->m_len,
1502 m2->m_next, m2->m_flags, "\2\20freelist\17skipfw"
1503 "\11proto5\10proto4\7proto3\6proto2\5proto1\4rdonly"
1504 "\3eor\2pkthdr\1ext", pdata ? "" : "\n");
1505 if (pdata)
1506 printf(", %*D\n", pdata, (u_char *)m2->m_data, "-");
1507 if (len != -1)
1508 len -= m2->m_len;
1509 m2 = m2->m_next;
1510 }
1511 if (len > 0)
1512 printf("%d bytes unaccounted for.\n", len);
1513 return;
1514 }
1515
1516 u_int
1517 m_fixhdr(struct mbuf *m0)
1518 {
1519 u_int len;
1520
1521 len = m_length(m0, NULL);
1522 m0->m_pkthdr.len = len;
1523 return (len);
1524 }
1525
1526 u_int
1527 m_length(struct mbuf *m0, struct mbuf **last)
1528 {
1529 struct mbuf *m;
1530 u_int len;
1531
1532 len = 0;
1533 for (m = m0; m != NULL; m = m->m_next) {
1534 len += m->m_len;
1535 if (m->m_next == NULL)
1536 break;
1537 }
1538 if (last != NULL)
1539 *last = m;
1540 return (len);
1541 }
1542
1543 /*
1544 * Defragment a mbuf chain, returning the shortest possible
1545 * chain of mbufs and clusters. If allocation fails and
1546 * this cannot be completed, NULL will be returned, but
1547 * the passed in chain will be unchanged. Upon success,
1548 * the original chain will be freed, and the new chain
1549 * will be returned.
1550 *
1551 * If a non-packet header is passed in, the original
1552 * mbuf (chain?) will be returned unharmed.
1553 */
1554 struct mbuf *
1555 m_defrag(struct mbuf *m0, int how)
1556 {
1557 struct mbuf *m_new = NULL, *m_final = NULL;
1558 int progress = 0, length;
1559
1560 MBUF_CHECKSLEEP(how);
1561 if (!(m0->m_flags & M_PKTHDR))
1562 return (m0);
1563
1564 m_fixhdr(m0); /* Needed sanity check */
1565
1566 #ifdef MBUF_STRESS_TEST
1567 if (m_defragrandomfailures) {
1568 int temp = arc4random() & 0xff;
1569 if (temp == 0xba)
1570 goto nospace;
1571 }
1572 #endif
1573
1574 if (m0->m_pkthdr.len > MHLEN)
1575 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1576 else
1577 m_final = m_gethdr(how, MT_DATA);
1578
1579 if (m_final == NULL)
1580 goto nospace;
1581
1582 if (m_dup_pkthdr(m_final, m0, how) == 0)
1583 goto nospace;
1584
1585 m_new = m_final;
1586
1587 while (progress < m0->m_pkthdr.len) {
1588 length = m0->m_pkthdr.len - progress;
1589 if (length > MCLBYTES)
1590 length = MCLBYTES;
1591
1592 if (m_new == NULL) {
1593 if (length > MLEN)
1594 m_new = m_getcl(how, MT_DATA, 0);
1595 else
1596 m_new = m_get(how, MT_DATA);
1597 if (m_new == NULL)
1598 goto nospace;
1599 }
1600
1601 m_copydata(m0, progress, length, mtod(m_new, caddr_t));
1602 progress += length;
1603 m_new->m_len = length;
1604 if (m_new != m_final)
1605 m_cat(m_final, m_new);
1606 m_new = NULL;
1607 }
1608 #ifdef MBUF_STRESS_TEST
1609 if (m0->m_next == NULL)
1610 m_defraguseless++;
1611 #endif
1612 m_freem(m0);
1613 m0 = m_final;
1614 #ifdef MBUF_STRESS_TEST
1615 m_defragpackets++;
1616 m_defragbytes += m0->m_pkthdr.len;
1617 #endif
1618 return (m0);
1619 nospace:
1620 #ifdef MBUF_STRESS_TEST
1621 m_defragfailure++;
1622 #endif
1623 if (m_final)
1624 m_freem(m_final);
1625 return (NULL);
1626 }
1627
1628 /*
1629 * Defragment an mbuf chain, returning at most maxfrags separate
1630 * mbufs+clusters. If this is not possible NULL is returned and
1631 * the original mbuf chain is left in it's present (potentially
1632 * modified) state. We use two techniques: collapsing consecutive
1633 * mbufs and replacing consecutive mbufs by a cluster.
1634 *
1635 * NB: this should really be named m_defrag but that name is taken
1636 */
1637 struct mbuf *
1638 m_collapse(struct mbuf *m0, int how, int maxfrags)
1639 {
1640 struct mbuf *m, *n, *n2, **prev;
1641 u_int curfrags;
1642
1643 /*
1644 * Calculate the current number of frags.
1645 */
1646 curfrags = 0;
1647 for (m = m0; m != NULL; m = m->m_next)
1648 curfrags++;
1649 /*
1650 * First, try to collapse mbufs. Note that we always collapse
1651 * towards the front so we don't need to deal with moving the
1652 * pkthdr. This may be suboptimal if the first mbuf has much
1653 * less data than the following.
1654 */
1655 m = m0;
1656 again:
1657 for (;;) {
1658 n = m->m_next;
1659 if (n == NULL)
1660 break;
1661 if (M_WRITABLE(m) &&
1662 n->m_len < M_TRAILINGSPACE(m)) {
1663 bcopy(mtod(n, void *), mtod(m, char *) + m->m_len,
1664 n->m_len);
1665 m->m_len += n->m_len;
1666 m->m_next = n->m_next;
1667 m_free(n);
1668 if (--curfrags <= maxfrags)
1669 return m0;
1670 } else
1671 m = n;
1672 }
1673 KASSERT(maxfrags > 1,
1674 ("maxfrags %u, but normal collapse failed", maxfrags));
1675 /*
1676 * Collapse consecutive mbufs to a cluster.
1677 */
1678 prev = &m0->m_next; /* NB: not the first mbuf */
1679 while ((n = *prev) != NULL) {
1680 if ((n2 = n->m_next) != NULL &&
1681 n->m_len + n2->m_len < MCLBYTES) {
1682 m = m_getcl(how, MT_DATA, 0);
1683 if (m == NULL)
1684 goto bad;
1685 bcopy(mtod(n, void *), mtod(m, void *), n->m_len);
1686 bcopy(mtod(n2, void *), mtod(m, char *) + n->m_len,
1687 n2->m_len);
1688 m->m_len = n->m_len + n2->m_len;
1689 m->m_next = n2->m_next;
1690 *prev = m;
1691 m_free(n);
1692 m_free(n2);
1693 if (--curfrags <= maxfrags) /* +1 cl -2 mbufs */
1694 return m0;
1695 /*
1696 * Still not there, try the normal collapse
1697 * again before we allocate another cluster.
1698 */
1699 goto again;
1700 }
1701 prev = &n->m_next;
1702 }
1703 /*
1704 * No place where we can collapse to a cluster; punt.
1705 * This can occur if, for example, you request 2 frags
1706 * but the packet requires that both be clusters (we
1707 * never reallocate the first mbuf to avoid moving the
1708 * packet header).
1709 */
1710 bad:
1711 return NULL;
1712 }
1713
1714 #ifdef MBUF_STRESS_TEST
1715
1716 /*
1717 * Fragment an mbuf chain. There's no reason you'd ever want to do
1718 * this in normal usage, but it's great for stress testing various
1719 * mbuf consumers.
1720 *
1721 * If fragmentation is not possible, the original chain will be
1722 * returned.
1723 *
1724 * Possible length values:
1725 * 0 no fragmentation will occur
1726 * > 0 each fragment will be of the specified length
1727 * -1 each fragment will be the same random value in length
1728 * -2 each fragment's length will be entirely random
1729 * (Random values range from 1 to 256)
1730 */
1731 struct mbuf *
1732 m_fragment(struct mbuf *m0, int how, int length)
1733 {
1734 struct mbuf *m_new = NULL, *m_final = NULL;
1735 int progress = 0;
1736
1737 if (!(m0->m_flags & M_PKTHDR))
1738 return (m0);
1739
1740 if ((length == 0) || (length < -2))
1741 return (m0);
1742
1743 m_fixhdr(m0); /* Needed sanity check */
1744
1745 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1746
1747 if (m_final == NULL)
1748 goto nospace;
1749
1750 if (m_dup_pkthdr(m_final, m0, how) == 0)
1751 goto nospace;
1752
1753 m_new = m_final;
1754
1755 if (length == -1)
1756 length = 1 + (arc4random() & 255);
1757
1758 while (progress < m0->m_pkthdr.len) {
1759 int fraglen;
1760
1761 if (length > 0)
1762 fraglen = length;
1763 else
1764 fraglen = 1 + (arc4random() & 255);
1765 if (fraglen > m0->m_pkthdr.len - progress)
1766 fraglen = m0->m_pkthdr.len - progress;
1767
1768 if (fraglen > MCLBYTES)
1769 fraglen = MCLBYTES;
1770
1771 if (m_new == NULL) {
1772 m_new = m_getcl(how, MT_DATA, 0);
1773 if (m_new == NULL)
1774 goto nospace;
1775 }
1776
1777 m_copydata(m0, progress, fraglen, mtod(m_new, caddr_t));
1778 progress += fraglen;
1779 m_new->m_len = fraglen;
1780 if (m_new != m_final)
1781 m_cat(m_final, m_new);
1782 m_new = NULL;
1783 }
1784 m_freem(m0);
1785 m0 = m_final;
1786 return (m0);
1787 nospace:
1788 if (m_final)
1789 m_freem(m_final);
1790 /* Return the original chain on failure */
1791 return (m0);
1792 }
1793
1794 #endif
1795
1796 /*
1797 * Copy the contents of uio into a properly sized mbuf chain.
1798 */
1799 struct mbuf *
1800 m_uiotombuf(struct uio *uio, int how, int len, int align, int flags)
1801 {
1802 struct mbuf *m, *mb;
1803 int error, length;
1804 ssize_t total;
1805 int progress = 0;
1806
1807 /*
1808 * len can be zero or an arbitrary large value bound by
1809 * the total data supplied by the uio.
1810 */
1811 if (len > 0)
1812 total = min(uio->uio_resid, len);
1813 else
1814 total = uio->uio_resid;
1815
1816 /*
1817 * The smallest unit returned by m_getm2() is a single mbuf
1818 * with pkthdr. We can't align past it.
1819 */
1820 if (align >= MHLEN)
1821 return (NULL);
1822
1823 /*
1824 * Give us the full allocation or nothing.
1825 * If len is zero return the smallest empty mbuf.
1826 */
1827 m = m_getm2(NULL, max(total + align, 1), how, MT_DATA, flags);
1828 if (m == NULL)
1829 return (NULL);
1830 m->m_data += align;
1831
1832 /* Fill all mbufs with uio data and update header information. */
1833 for (mb = m; mb != NULL; mb = mb->m_next) {
1834 length = min(M_TRAILINGSPACE(mb), total - progress);
1835
1836 error = uiomove(mtod(mb, void *), length, uio);
1837 if (error) {
1838 m_freem(m);
1839 return (NULL);
1840 }
1841
1842 mb->m_len = length;
1843 progress += length;
1844 if (flags & M_PKTHDR)
1845 m->m_pkthdr.len += length;
1846 }
1847 KASSERT(progress == total, ("%s: progress != total", __func__));
1848
1849 return (m);
1850 }
1851
1852 /*
1853 * Copy an mbuf chain into a uio limited by len if set.
1854 */
1855 int
1856 m_mbuftouio(struct uio *uio, struct mbuf *m, int len)
1857 {
1858 int error, length, total;
1859 int progress = 0;
1860
1861 if (len > 0)
1862 total = min(uio->uio_resid, len);
1863 else
1864 total = uio->uio_resid;
1865
1866 /* Fill the uio with data from the mbufs. */
1867 for (; m != NULL; m = m->m_next) {
1868 length = min(m->m_len, total - progress);
1869
1870 error = uiomove(mtod(m, void *), length, uio);
1871 if (error)
1872 return (error);
1873
1874 progress += length;
1875 }
1876
1877 return (0);
1878 }
1879
1880 /*
1881 * Set the m_data pointer of a newly-allocated mbuf
1882 * to place an object of the specified size at the
1883 * end of the mbuf, longword aligned.
1884 */
1885 void
1886 m_align(struct mbuf *m, int len)
1887 {
1888 #ifdef INVARIANTS
1889 const char *msg = "%s: not a virgin mbuf";
1890 #endif
1891 int adjust;
1892
1893 if (m->m_flags & M_EXT) {
1894 KASSERT(m->m_data == m->m_ext.ext_buf, (msg, __func__));
1895 adjust = m->m_ext.ext_size - len;
1896 } else if (m->m_flags & M_PKTHDR) {
1897 KASSERT(m->m_data == m->m_pktdat, (msg, __func__));
1898 adjust = MHLEN - len;
1899 } else {
1900 KASSERT(m->m_data == m->m_dat, (msg, __func__));
1901 adjust = MLEN - len;
1902 }
1903
1904 m->m_data += adjust &~ (sizeof(long)-1);
1905 }
1906
1907 /*
1908 * Create a writable copy of the mbuf chain. While doing this
1909 * we compact the chain with a goal of producing a chain with
1910 * at most two mbufs. The second mbuf in this chain is likely
1911 * to be a cluster. The primary purpose of this work is to create
1912 * a writable packet for encryption, compression, etc. The
1913 * secondary goal is to linearize the data so the data can be
1914 * passed to crypto hardware in the most efficient manner possible.
1915 */
1916 struct mbuf *
1917 m_unshare(struct mbuf *m0, int how)
1918 {
1919 struct mbuf *m, *mprev;
1920 struct mbuf *n, *mfirst, *mlast;
1921 int len, off;
1922
1923 mprev = NULL;
1924 for (m = m0; m != NULL; m = mprev->m_next) {
1925 /*
1926 * Regular mbufs are ignored unless there's a cluster
1927 * in front of it that we can use to coalesce. We do
1928 * the latter mainly so later clusters can be coalesced
1929 * also w/o having to handle them specially (i.e. convert
1930 * mbuf+cluster -> cluster). This optimization is heavily
1931 * influenced by the assumption that we're running over
1932 * Ethernet where MCLBYTES is large enough that the max
1933 * packet size will permit lots of coalescing into a
1934 * single cluster. This in turn permits efficient
1935 * crypto operations, especially when using hardware.
1936 */
1937 if ((m->m_flags & M_EXT) == 0) {
1938 if (mprev && (mprev->m_flags & M_EXT) &&
1939 m->m_len <= M_TRAILINGSPACE(mprev)) {
1940 /* XXX: this ignores mbuf types */
1941 memcpy(mtod(mprev, caddr_t) + mprev->m_len,
1942 mtod(m, caddr_t), m->m_len);
1943 mprev->m_len += m->m_len;
1944 mprev->m_next = m->m_next; /* unlink from chain */
1945 m_free(m); /* reclaim mbuf */
1946 #if 0
1947 newipsecstat.ips_mbcoalesced++;
1948 #endif
1949 } else {
1950 mprev = m;
1951 }
1952 continue;
1953 }
1954 /*
1955 * Writable mbufs are left alone (for now).
1956 */
1957 if (M_WRITABLE(m)) {
1958 mprev = m;
1959 continue;
1960 }
1961
1962 /*
1963 * Not writable, replace with a copy or coalesce with
1964 * the previous mbuf if possible (since we have to copy
1965 * it anyway, we try to reduce the number of mbufs and
1966 * clusters so that future work is easier).
1967 */
1968 KASSERT(m->m_flags & M_EXT, ("m_flags 0x%x", m->m_flags));
1969 /* NB: we only coalesce into a cluster or larger */
1970 if (mprev != NULL && (mprev->m_flags & M_EXT) &&
1971 m->m_len <= M_TRAILINGSPACE(mprev)) {
1972 /* XXX: this ignores mbuf types */
1973 memcpy(mtod(mprev, caddr_t) + mprev->m_len,
1974 mtod(m, caddr_t), m->m_len);
1975 mprev->m_len += m->m_len;
1976 mprev->m_next = m->m_next; /* unlink from chain */
1977 m_free(m); /* reclaim mbuf */
1978 #if 0
1979 newipsecstat.ips_clcoalesced++;
1980 #endif
1981 continue;
1982 }
1983
1984 /*
1985 * Allocate new space to hold the copy and copy the data.
1986 * We deal with jumbo mbufs (i.e. m_len > MCLBYTES) by
1987 * splitting them into clusters. We could just malloc a
1988 * buffer and make it external but too many device drivers
1989 * don't know how to break up the non-contiguous memory when
1990 * doing DMA.
1991 */
1992 n = m_getcl(how, m->m_type, m->m_flags);
1993 if (n == NULL) {
1994 m_freem(m0);
1995 return (NULL);
1996 }
1997 len = m->m_len;
1998 off = 0;
1999 mfirst = n;
2000 mlast = NULL;
2001 for (;;) {
2002 int cc = min(len, MCLBYTES);
2003 memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + off, cc);
2004 n->m_len = cc;
2005 if (mlast != NULL)
2006 mlast->m_next = n;
2007 mlast = n;
2008 #if 0
2009 newipsecstat.ips_clcopied++;
2010 #endif
2011
2012 len -= cc;
2013 if (len <= 0)
2014 break;
2015 off += cc;
2016
2017 n = m_getcl(how, m->m_type, m->m_flags);
2018 if (n == NULL) {
2019 m_freem(mfirst);
2020 m_freem(m0);
2021 return (NULL);
2022 }
2023 }
2024 n->m_next = m->m_next;
2025 if (mprev == NULL)
2026 m0 = mfirst; /* new head of chain */
2027 else
2028 mprev->m_next = mfirst; /* replace old mbuf */
2029 m_free(m); /* release old mbuf */
2030 mprev = mfirst;
2031 }
2032 return (m0);
2033 }
2034
2035 #ifdef MBUF_PROFILING
2036
2037 #define MP_BUCKETS 32 /* don't just change this as things may overflow.*/
2038 struct mbufprofile {
2039 uintmax_t wasted[MP_BUCKETS];
2040 uintmax_t used[MP_BUCKETS];
2041 uintmax_t segments[MP_BUCKETS];
2042 } mbprof;
2043
2044 #define MP_MAXDIGITS 21 /* strlen("16,000,000,000,000,000,000") == 21 */
2045 #define MP_NUMLINES 6
2046 #define MP_NUMSPERLINE 16
2047 #define MP_EXTRABYTES 64 /* > strlen("used:\nwasted:\nsegments:\n") */
2048 /* work out max space needed and add a bit of spare space too */
2049 #define MP_MAXLINE ((MP_MAXDIGITS+1) * MP_NUMSPERLINE)
2050 #define MP_BUFSIZE ((MP_MAXLINE * MP_NUMLINES) + 1 + MP_EXTRABYTES)
2051
2052 char mbprofbuf[MP_BUFSIZE];
2053
2054 void
2055 m_profile(struct mbuf *m)
2056 {
2057 int segments = 0;
2058 int used = 0;
2059 int wasted = 0;
2060
2061 while (m) {
2062 segments++;
2063 used += m->m_len;
2064 if (m->m_flags & M_EXT) {
2065 wasted += MHLEN - sizeof(m->m_ext) +
2066 m->m_ext.ext_size - m->m_len;
2067 } else {
2068 if (m->m_flags & M_PKTHDR)
2069 wasted += MHLEN - m->m_len;
2070 else
2071 wasted += MLEN - m->m_len;
2072 }
2073 m = m->m_next;
2074 }
2075 /* be paranoid.. it helps */
2076 if (segments > MP_BUCKETS - 1)
2077 segments = MP_BUCKETS - 1;
2078 if (used > 100000)
2079 used = 100000;
2080 if (wasted > 100000)
2081 wasted = 100000;
2082 /* store in the appropriate bucket */
2083 /* don't bother locking. if it's slightly off, so what? */
2084 mbprof.segments[segments]++;
2085 mbprof.used[fls(used)]++;
2086 mbprof.wasted[fls(wasted)]++;
2087 }
2088
2089 static void
2090 mbprof_textify(void)
2091 {
2092 int offset;
2093 char *c;
2094 uint64_t *p;
2095
2096
2097 p = &mbprof.wasted[0];
2098 c = mbprofbuf;
2099 offset = snprintf(c, MP_MAXLINE + 10,
2100 "wasted:\n"
2101 "%ju %ju %ju %ju %ju %ju %ju %ju "
2102 "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2103 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2104 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2105 #ifdef BIG_ARRAY
2106 p = &mbprof.wasted[16];
2107 c += offset;
2108 offset = snprintf(c, MP_MAXLINE,
2109 "%ju %ju %ju %ju %ju %ju %ju %ju "
2110 "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2111 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2112 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2113 #endif
2114 p = &mbprof.used[0];
2115 c += offset;
2116 offset = snprintf(c, MP_MAXLINE + 10,
2117 "used:\n"
2118 "%ju %ju %ju %ju %ju %ju %ju %ju "
2119 "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2120 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2121 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2122 #ifdef BIG_ARRAY
2123 p = &mbprof.used[16];
2124 c += offset;
2125 offset = snprintf(c, MP_MAXLINE,
2126 "%ju %ju %ju %ju %ju %ju %ju %ju "
2127 "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2128 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2129 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2130 #endif
2131 p = &mbprof.segments[0];
2132 c += offset;
2133 offset = snprintf(c, MP_MAXLINE + 10,
2134 "segments:\n"
2135 "%ju %ju %ju %ju %ju %ju %ju %ju "
2136 "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2137 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2138 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2139 #ifdef BIG_ARRAY
2140 p = &mbprof.segments[16];
2141 c += offset;
2142 offset = snprintf(c, MP_MAXLINE,
2143 "%ju %ju %ju %ju %ju %ju %ju %ju "
2144 "%ju %ju %ju %ju %ju %ju %ju %jju",
2145 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2146 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2147 #endif
2148 }
2149
2150 static int
2151 mbprof_handler(SYSCTL_HANDLER_ARGS)
2152 {
2153 int error;
2154
2155 mbprof_textify();
2156 error = SYSCTL_OUT(req, mbprofbuf, strlen(mbprofbuf) + 1);
2157 return (error);
2158 }
2159
2160 static int
2161 mbprof_clr_handler(SYSCTL_HANDLER_ARGS)
2162 {
2163 int clear, error;
2164
2165 clear = 0;
2166 error = sysctl_handle_int(oidp, &clear, 0, req);
2167 if (error || !req->newptr)
2168 return (error);
2169
2170 if (clear) {
2171 bzero(&mbprof, sizeof(mbprof));
2172 }
2173
2174 return (error);
2175 }
2176
2177
2178 SYSCTL_PROC(_kern_ipc, OID_AUTO, mbufprofile, CTLTYPE_STRING|CTLFLAG_RD,
2179 NULL, 0, mbprof_handler, "A", "mbuf profiling statistics");
2180
2181 SYSCTL_PROC(_kern_ipc, OID_AUTO, mbufprofileclr, CTLTYPE_INT|CTLFLAG_RW,
2182 NULL, 0, mbprof_clr_handler, "I", "clear mbuf profiling statistics");
2183 #endif
2184
Cache object: b1d0c913bd04c7fc03e4b91fced61890
|