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 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 *
33 * @(#)uipc_mbuf.c 8.2 (Berkeley) 1/4/94
34 * $FreeBSD$
35 */
36
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/malloc.h>
40 #include <sys/mbuf.h>
41 #include <sys/kernel.h>
42 #include <sys/sysctl.h>
43 #include <sys/domain.h>
44 #include <sys/protosw.h>
45
46 #include <vm/vm.h>
47 #include <vm/vm_kern.h>
48 #include <vm/vm_extern.h>
49
50 #ifdef INVARIANTS
51 #include <machine/cpu.h>
52 #endif
53
54 static void mbinit __P((void *));
55 SYSINIT(mbuf, SI_SUB_MBUF, SI_ORDER_FIRST, mbinit, NULL)
56
57 struct mbuf *mbutl;
58 char *mclrefcnt;
59 struct mbstat mbstat;
60 struct mbuf *mmbfree;
61 union mcluster *mclfree;
62 int max_linkhdr;
63 int max_protohdr;
64 int max_hdr;
65 int max_datalen;
66 u_int m_mballoc_wid = 0;
67 u_int m_clalloc_wid = 0;
68
69 SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RW,
70 &max_linkhdr, 0, "");
71 SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RW,
72 &max_protohdr, 0, "");
73 SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RW, &max_hdr, 0, "");
74 SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RW,
75 &max_datalen, 0, "");
76 SYSCTL_INT(_kern_ipc, OID_AUTO, mbuf_wait, CTLFLAG_RW,
77 &mbuf_wait, 0, "");
78 SYSCTL_STRUCT(_kern_ipc, KIPC_MBSTAT, mbstat, CTLFLAG_RW, &mbstat, mbstat, "");
79
80 static void m_reclaim __P((void));
81
82 /* "number of clusters of pages" */
83 #define NCL_INIT 1
84
85 #define NMB_INIT 16
86
87 /* ARGSUSED*/
88 static void
89 mbinit(dummy)
90 void *dummy;
91 {
92 int s;
93
94 mmbfree = NULL; mclfree = NULL;
95 mbstat.m_msize = MSIZE;
96 mbstat.m_mclbytes = MCLBYTES;
97 mbstat.m_minclsize = MINCLSIZE;
98 mbstat.m_mlen = MLEN;
99 mbstat.m_mhlen = MHLEN;
100
101 s = splimp();
102 if (m_mballoc(NMB_INIT, M_DONTWAIT) == 0)
103 goto bad;
104 #if MCLBYTES <= PAGE_SIZE
105 if (m_clalloc(NCL_INIT, M_DONTWAIT) == 0)
106 goto bad;
107 #else
108 /* It's OK to call contigmalloc in this context. */
109 if (m_clalloc(16, M_WAIT) == 0)
110 goto bad;
111 #endif
112 splx(s);
113 return;
114 bad:
115 panic("mbinit");
116 }
117
118 /*
119 * Allocate at least nmb mbufs and place on mbuf free list.
120 * Must be called at splimp.
121 */
122 /* ARGSUSED */
123 int
124 m_mballoc(nmb, how)
125 register int nmb;
126 int how;
127 {
128 register caddr_t p;
129 register int i;
130 int nbytes;
131
132 /*
133 * If we've hit the mbuf limit, stop allocating from mb_map,
134 * (or trying to) in order to avoid dipping into the section of
135 * mb_map which we've "reserved" for clusters.
136 */
137 if ((nmb + mbstat.m_mbufs) > nmbufs)
138 return (0);
139
140 /*
141 * Once we run out of map space, it will be impossible to get
142 * any more (nothing is ever freed back to the map)
143 * -- however you are not dead as m_reclaim might
144 * still be able to free a substantial amount of space.
145 *
146 * XXX Furthermore, we can also work with "recycled" mbufs (when
147 * we're calling with M_WAIT the sleep procedure will be woken
148 * up when an mbuf is freed. See m_mballoc_wait()).
149 */
150 if (mb_map_full)
151 return (0);
152
153 nbytes = round_page(nmb * MSIZE);
154 p = (caddr_t)kmem_malloc(mb_map, nbytes, M_NOWAIT);
155 if (p == 0 && how == M_WAIT) {
156 mbstat.m_wait++;
157 p = (caddr_t)kmem_malloc(mb_map, nbytes, M_WAITOK);
158 }
159
160 /*
161 * Either the map is now full, or `how' is M_NOWAIT and there
162 * are no pages left.
163 */
164 if (p == NULL)
165 return (0);
166
167 nmb = nbytes / MSIZE;
168 for (i = 0; i < nmb; i++) {
169 ((struct mbuf *)p)->m_next = mmbfree;
170 mmbfree = (struct mbuf *)p;
171 p += MSIZE;
172 }
173 mbstat.m_mbufs += nmb;
174 return (1);
175 }
176
177 /*
178 * Once the mb_map has been exhausted and if the call to the allocation macros
179 * (or, in some cases, functions) is with M_WAIT, then it is necessary to rely
180 * solely on reclaimed mbufs. Here we wait for an mbuf to be freed for a
181 * designated (mbuf_wait) time.
182 */
183 struct mbuf *
184 m_mballoc_wait(int caller, int type)
185 {
186 struct mbuf *p;
187 int s;
188
189 m_mballoc_wid++;
190 if ((tsleep(&m_mballoc_wid, PVM, "mballc", mbuf_wait)) == EWOULDBLOCK)
191 m_mballoc_wid--;
192
193 /*
194 * Now that we (think) that we've got something, we will redo an
195 * MGET, but avoid getting into another instance of m_mballoc_wait()
196 * XXX: We retry to fetch _even_ if the sleep timed out. This is left
197 * this way, purposely, in the [unlikely] case that an mbuf was
198 * freed but the sleep was not awakened in time.
199 */
200 p = NULL;
201 switch (caller) {
202 case MGET_C:
203 MGET(p, M_DONTWAIT, type);
204 break;
205 case MGETHDR_C:
206 MGETHDR(p, M_DONTWAIT, type);
207 break;
208 default:
209 panic("m_mballoc_wait: invalid caller (%d)", caller);
210 }
211
212 s = splimp();
213 if (p != NULL) { /* We waited and got something... */
214 mbstat.m_wait++;
215 /* Wake up another if we have more free. */
216 if (mmbfree != NULL)
217 MMBWAKEUP();
218 }
219 splx(s);
220 return (p);
221 }
222
223 #if MCLBYTES > PAGE_SIZE
224 static int i_want_my_mcl;
225
226 static void
227 kproc_mclalloc(void)
228 {
229 int status;
230
231 while (1) {
232 tsleep(&i_want_my_mcl, PVM, "mclalloc", 0);
233
234 for (; i_want_my_mcl; i_want_my_mcl--) {
235 if (m_clalloc(1, M_WAIT) == 0)
236 printf("m_clalloc failed even in process context!\n");
237 }
238 }
239 }
240
241 static struct proc *mclallocproc;
242 static struct kproc_desc mclalloc_kp = {
243 "mclalloc",
244 kproc_mclalloc,
245 &mclallocproc
246 };
247 SYSINIT_KT(mclallocproc, SI_SUB_KTHREAD_UPDATE, SI_ORDER_ANY, kproc_start,
248 &mclalloc_kp);
249 #endif
250
251 /*
252 * Allocate some number of mbuf clusters
253 * and place on cluster free list.
254 * Must be called at splimp.
255 */
256 /* ARGSUSED */
257 int
258 m_clalloc(ncl, how)
259 register int ncl;
260 int how;
261 {
262 register caddr_t p;
263 register int i;
264 int npg;
265
266 /*
267 * If we've hit the mcluster number limit, stop allocating from
268 * mb_map, (or trying to) in order to avoid dipping into the section
269 * of mb_map which we've "reserved" for mbufs.
270 */
271 if ((ncl + mbstat.m_clusters) > nmbclusters) {
272 mbstat.m_drops++;
273 return (0);
274 }
275
276 /*
277 * Once we run out of map space, it will be impossible
278 * to get any more (nothing is ever freed back to the
279 * map). From this point on, we solely rely on freed
280 * mclusters.
281 */
282 if (mb_map_full) {
283 mbstat.m_drops++;
284 return (0);
285 }
286
287 #if MCLBYTES > PAGE_SIZE
288 if (how != M_WAIT) {
289 i_want_my_mcl += ncl;
290 wakeup(&i_want_my_mcl);
291 mbstat.m_wait++;
292 p = 0;
293 } else {
294 p = contigmalloc1(MCLBYTES * ncl, M_DEVBUF, M_WAITOK, 0ul,
295 ~0ul, PAGE_SIZE, 0, mb_map);
296 }
297 #else
298 npg = ncl;
299 p = (caddr_t)kmem_malloc(mb_map, ctob(npg),
300 how != M_WAIT ? M_NOWAIT : M_WAITOK);
301 ncl = ncl * PAGE_SIZE / MCLBYTES;
302 #endif
303 /*
304 * Either the map is now full, or `how' is M_NOWAIT and there
305 * are no pages left.
306 */
307 if (p == NULL) {
308 mbstat.m_drops++;
309 return (0);
310 }
311
312 for (i = 0; i < ncl; i++) {
313 ((union mcluster *)p)->mcl_next = mclfree;
314 mclfree = (union mcluster *)p;
315 p += MCLBYTES;
316 mbstat.m_clfree++;
317 }
318 mbstat.m_clusters += ncl;
319 return (1);
320 }
321
322 /*
323 * Once the mb_map submap has been exhausted and the allocation is called with
324 * M_WAIT, we rely on the mclfree union pointers. If nothing is free, we will
325 * sleep for a designated amount of time (mbuf_wait) or until we're woken up
326 * due to sudden mcluster availability.
327 */
328 caddr_t
329 m_clalloc_wait(void)
330 {
331 caddr_t p;
332 int s;
333
334 #ifdef __i386__
335 /* If in interrupt context, and INVARIANTS, maintain sanity and die. */
336 KASSERT(intr_nesting_level == 0, ("CLALLOC: CANNOT WAIT IN INTERRUPT"));
337 #endif
338
339 /* Sleep until something's available or until we expire. */
340 m_clalloc_wid++;
341 if ((tsleep(&m_clalloc_wid, PVM, "mclalc", mbuf_wait)) == EWOULDBLOCK)
342 m_clalloc_wid--;
343
344 /*
345 * Now that we (think) that we've got something, we will redo and
346 * MGET, but avoid getting into another instance of m_clalloc_wait()
347 */
348 p = NULL;
349 MCLALLOC(p, M_DONTWAIT);
350
351 s = splimp();
352 if (p != NULL) { /* We waited and got something... */
353 mbstat.m_wait++;
354 /* Wake up another if we have more free. */
355 if (mclfree != NULL)
356 MCLWAKEUP();
357 }
358
359 splx(s);
360 return (p);
361 }
362
363 /*
364 * When MGET fails, ask protocols to free space when short of memory,
365 * then re-attempt to allocate an mbuf.
366 */
367 struct mbuf *
368 m_retry(i, t)
369 int i, t;
370 {
371 register struct mbuf *m;
372
373 /*
374 * Must only do the reclaim if not in an interrupt context.
375 */
376 if (i == M_WAIT) {
377 #ifdef __i386__
378 KASSERT(intr_nesting_level == 0,
379 ("MBALLOC: CANNOT WAIT IN INTERRUPT"));
380 #endif
381 m_reclaim();
382 }
383
384 /*
385 * Both m_mballoc_wait and m_retry must be nulled because
386 * when the MGET macro is run from here, we deffinately do _not_
387 * want to enter an instance of m_mballoc_wait() or m_retry() (again!)
388 */
389 #define m_mballoc_wait(caller,type) (struct mbuf *)0
390 #define m_retry(i, t) (struct mbuf *)0
391 MGET(m, i, t);
392 #undef m_retry
393 #undef m_mballoc_wait
394
395 if (m != NULL)
396 mbstat.m_wait++;
397 else
398 mbstat.m_drops++;
399
400 return (m);
401 }
402
403 /*
404 * As above; retry an MGETHDR.
405 */
406 struct mbuf *
407 m_retryhdr(i, t)
408 int i, t;
409 {
410 register struct mbuf *m;
411
412 /*
413 * Must only do the reclaim if not in an interrupt context.
414 */
415 if (i == M_WAIT) {
416 #ifdef __i386__
417 KASSERT(intr_nesting_level == 0,
418 ("MBALLOC: CANNOT WAIT IN INTERRUPT"));
419 #endif
420 m_reclaim();
421 }
422
423 #define m_mballoc_wait(caller,type) (struct mbuf *)0
424 #define m_retryhdr(i, t) (struct mbuf *)0
425 MGETHDR(m, i, t);
426 #undef m_retryhdr
427 #undef m_mballoc_wait
428
429 if (m != NULL)
430 mbstat.m_wait++;
431 else
432 mbstat.m_drops++;
433
434 return (m);
435 }
436
437 static void
438 m_reclaim()
439 {
440 register struct domain *dp;
441 register struct protosw *pr;
442 int s = splimp();
443
444 for (dp = domains; dp; dp = dp->dom_next)
445 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++)
446 if (pr->pr_drain)
447 (*pr->pr_drain)();
448 splx(s);
449 mbstat.m_drain++;
450 }
451
452 /*
453 * Space allocation routines.
454 * These are also available as macros
455 * for critical paths.
456 */
457 struct mbuf *
458 m_get(how, type)
459 int how, type;
460 {
461 register struct mbuf *m;
462
463 MGET(m, how, type);
464 return (m);
465 }
466
467 struct mbuf *
468 m_gethdr(how, type)
469 int how, type;
470 {
471 register struct mbuf *m;
472
473 MGETHDR(m, how, type);
474 return (m);
475 }
476
477 struct mbuf *
478 m_getclr(how, type)
479 int how, type;
480 {
481 register struct mbuf *m;
482
483 MGET(m, how, type);
484 if (m == 0)
485 return (0);
486 bzero(mtod(m, caddr_t), MLEN);
487 return (m);
488 }
489
490 struct mbuf *
491 m_free(m)
492 struct mbuf *m;
493 {
494 register struct mbuf *n;
495
496 MFREE(m, n);
497 return (n);
498 }
499
500 void
501 m_freem(m)
502 register struct mbuf *m;
503 {
504 register struct mbuf *n;
505
506 if (m == NULL)
507 return;
508 do {
509 MFREE(m, n);
510 m = n;
511 } while (m);
512 }
513
514 /*
515 * Mbuffer utility routines.
516 */
517
518 /*
519 * Lesser-used path for M_PREPEND:
520 * allocate new mbuf to prepend to chain,
521 * copy junk along.
522 */
523 struct mbuf *
524 m_prepend(m, len, how)
525 register struct mbuf *m;
526 int len, how;
527 {
528 struct mbuf *mn;
529
530 MGET(mn, how, m->m_type);
531 if (mn == (struct mbuf *)NULL) {
532 m_freem(m);
533 return ((struct mbuf *)NULL);
534 }
535 if (m->m_flags & M_PKTHDR) {
536 M_COPY_PKTHDR(mn, m);
537 m->m_flags &= ~M_PKTHDR;
538 }
539 mn->m_next = m;
540 m = mn;
541 if (len < MHLEN)
542 MH_ALIGN(m, len);
543 m->m_len = len;
544 return (m);
545 }
546
547 /*
548 * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
549 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf.
550 * The wait parameter is a choice of M_WAIT/M_DONTWAIT from caller.
551 * Note that the copy is read-only, because clusters are not copied,
552 * only their reference counts are incremented.
553 */
554 #define MCFail (mbstat.m_mcfail)
555
556 struct mbuf *
557 m_copym(m, off0, len, wait)
558 register struct mbuf *m;
559 int off0, wait;
560 register int len;
561 {
562 register struct mbuf *n, **np;
563 register int off = off0;
564 struct mbuf *top;
565 int copyhdr = 0;
566
567 if (off < 0 || len < 0)
568 panic("m_copym");
569 if (off == 0 && m->m_flags & M_PKTHDR)
570 copyhdr = 1;
571 while (off > 0) {
572 if (m == 0)
573 panic("m_copym");
574 if (off < m->m_len)
575 break;
576 off -= m->m_len;
577 m = m->m_next;
578 }
579 np = ⊤
580 top = 0;
581 while (len > 0) {
582 if (m == 0) {
583 if (len != M_COPYALL)
584 panic("m_copym");
585 break;
586 }
587 MGET(n, wait, m->m_type);
588 *np = n;
589 if (n == 0)
590 goto nospace;
591 if (copyhdr) {
592 M_COPY_PKTHDR(n, m);
593 if (len == M_COPYALL)
594 n->m_pkthdr.len -= off0;
595 else
596 n->m_pkthdr.len = len;
597 copyhdr = 0;
598 }
599 n->m_len = min(len, m->m_len - off);
600 if (m->m_flags & M_EXT) {
601 n->m_data = m->m_data + off;
602 if(!m->m_ext.ext_ref)
603 mclrefcnt[mtocl(m->m_ext.ext_buf)]++;
604 else
605 (*(m->m_ext.ext_ref))(m->m_ext.ext_buf,
606 m->m_ext.ext_size);
607 n->m_ext = m->m_ext;
608 n->m_flags |= M_EXT;
609 } else
610 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
611 (unsigned)n->m_len);
612 if (len != M_COPYALL)
613 len -= n->m_len;
614 off = 0;
615 m = m->m_next;
616 np = &n->m_next;
617 }
618 if (top == 0)
619 MCFail++;
620 return (top);
621 nospace:
622 m_freem(top);
623 MCFail++;
624 return (0);
625 }
626
627 /*
628 * Copy an entire packet, including header (which must be present).
629 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
630 * Note that the copy is read-only, because clusters are not copied,
631 * only their reference counts are incremented.
632 */
633 struct mbuf *
634 m_copypacket(m, how)
635 struct mbuf *m;
636 int how;
637 {
638 struct mbuf *top, *n, *o;
639
640 MGET(n, how, m->m_type);
641 top = n;
642 if (!n)
643 goto nospace;
644
645 M_COPY_PKTHDR(n, m);
646 n->m_len = m->m_len;
647 if (m->m_flags & M_EXT) {
648 n->m_data = m->m_data;
649 if(!m->m_ext.ext_ref)
650 mclrefcnt[mtocl(m->m_ext.ext_buf)]++;
651 else
652 (*(m->m_ext.ext_ref))(m->m_ext.ext_buf,
653 m->m_ext.ext_size);
654 n->m_ext = m->m_ext;
655 n->m_flags |= M_EXT;
656 } else {
657 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
658 }
659
660 m = m->m_next;
661 while (m) {
662 MGET(o, how, m->m_type);
663 if (!o)
664 goto nospace;
665
666 n->m_next = o;
667 n = n->m_next;
668
669 n->m_len = m->m_len;
670 if (m->m_flags & M_EXT) {
671 n->m_data = m->m_data;
672 if(!m->m_ext.ext_ref)
673 mclrefcnt[mtocl(m->m_ext.ext_buf)]++;
674 else
675 (*(m->m_ext.ext_ref))(m->m_ext.ext_buf,
676 m->m_ext.ext_size);
677 n->m_ext = m->m_ext;
678 n->m_flags |= M_EXT;
679 } else {
680 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
681 }
682
683 m = m->m_next;
684 }
685 return top;
686 nospace:
687 m_freem(top);
688 MCFail++;
689 return 0;
690 }
691
692 /*
693 * Copy data from an mbuf chain starting "off" bytes from the beginning,
694 * continuing for "len" bytes, into the indicated buffer.
695 */
696 void
697 m_copydata(m, off, len, cp)
698 register struct mbuf *m;
699 register int off;
700 register int len;
701 caddr_t cp;
702 {
703 register unsigned count;
704
705 if (off < 0 || len < 0)
706 panic("m_copydata");
707 while (off > 0) {
708 if (m == 0)
709 panic("m_copydata");
710 if (off < m->m_len)
711 break;
712 off -= m->m_len;
713 m = m->m_next;
714 }
715 while (len > 0) {
716 if (m == 0)
717 panic("m_copydata");
718 count = min(m->m_len - off, len);
719 bcopy(mtod(m, caddr_t) + off, cp, count);
720 len -= count;
721 cp += count;
722 off = 0;
723 m = m->m_next;
724 }
725 }
726
727 /*
728 * Copy a packet header mbuf chain into a completely new chain, including
729 * copying any mbuf clusters. Use this instead of m_copypacket() when
730 * you need a writable copy of an mbuf chain.
731 */
732 struct mbuf *
733 m_dup(m, how)
734 struct mbuf *m;
735 int how;
736 {
737 struct mbuf **p, *top = NULL;
738 int remain, moff, nsize;
739
740 /* Sanity check */
741 if (m == NULL)
742 return (0);
743 KASSERT((m->m_flags & M_PKTHDR) != 0, ("%s: !PKTHDR", __FUNCTION__));
744
745 /* While there's more data, get a new mbuf, tack it on, and fill it */
746 remain = m->m_pkthdr.len;
747 moff = 0;
748 p = ⊤
749 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */
750 struct mbuf *n;
751
752 /* Get the next new mbuf */
753 MGET(n, how, m->m_type);
754 if (n == NULL)
755 goto nospace;
756 if (top == NULL) { /* first one, must be PKTHDR */
757 M_COPY_PKTHDR(n, m);
758 nsize = MHLEN;
759 } else /* not the first one */
760 nsize = MLEN;
761 if (remain >= MINCLSIZE) {
762 MCLGET(n, how);
763 if ((n->m_flags & M_EXT) == 0) {
764 (void)m_free(n);
765 goto nospace;
766 }
767 nsize = MCLBYTES;
768 }
769 n->m_len = 0;
770
771 /* Link it into the new chain */
772 *p = n;
773 p = &n->m_next;
774
775 /* Copy data from original mbuf(s) into new mbuf */
776 while (n->m_len < nsize && m != NULL) {
777 int chunk = min(nsize - n->m_len, m->m_len - moff);
778
779 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
780 moff += chunk;
781 n->m_len += chunk;
782 remain -= chunk;
783 if (moff == m->m_len) {
784 m = m->m_next;
785 moff = 0;
786 }
787 }
788
789 /* Check correct total mbuf length */
790 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL),
791 ("%s: bogus m_pkthdr.len", __FUNCTION__));
792 }
793 return (top);
794
795 nospace:
796 m_freem(top);
797 MCFail++;
798 return (0);
799 }
800
801 /*
802 * Concatenate mbuf chain n to m.
803 * Both chains must be of the same type (e.g. MT_DATA).
804 * Any m_pkthdr is not updated.
805 */
806 void
807 m_cat(m, n)
808 register struct mbuf *m, *n;
809 {
810 while (m->m_next)
811 m = m->m_next;
812 while (n) {
813 if (m->m_flags & M_EXT ||
814 m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) {
815 /* just join the two chains */
816 m->m_next = n;
817 return;
818 }
819 /* splat the data from one into the other */
820 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
821 (u_int)n->m_len);
822 m->m_len += n->m_len;
823 n = m_free(n);
824 }
825 }
826
827 void
828 m_adj(mp, req_len)
829 struct mbuf *mp;
830 int req_len;
831 {
832 register int len = req_len;
833 register struct mbuf *m;
834 register int count;
835
836 if ((m = mp) == NULL)
837 return;
838 if (len >= 0) {
839 /*
840 * Trim from head.
841 */
842 while (m != NULL && len > 0) {
843 if (m->m_len <= len) {
844 len -= m->m_len;
845 m->m_len = 0;
846 m = m->m_next;
847 } else {
848 m->m_len -= len;
849 m->m_data += len;
850 len = 0;
851 }
852 }
853 m = mp;
854 if (mp->m_flags & M_PKTHDR)
855 m->m_pkthdr.len -= (req_len - len);
856 } else {
857 /*
858 * Trim from tail. Scan the mbuf chain,
859 * calculating its length and finding the last mbuf.
860 * If the adjustment only affects this mbuf, then just
861 * adjust and return. Otherwise, rescan and truncate
862 * after the remaining size.
863 */
864 len = -len;
865 count = 0;
866 for (;;) {
867 count += m->m_len;
868 if (m->m_next == (struct mbuf *)0)
869 break;
870 m = m->m_next;
871 }
872 if (m->m_len >= len) {
873 m->m_len -= len;
874 if (mp->m_flags & M_PKTHDR)
875 mp->m_pkthdr.len -= len;
876 return;
877 }
878 count -= len;
879 if (count < 0)
880 count = 0;
881 /*
882 * Correct length for chain is "count".
883 * Find the mbuf with last data, adjust its length,
884 * and toss data from remaining mbufs on chain.
885 */
886 m = mp;
887 if (m->m_flags & M_PKTHDR)
888 m->m_pkthdr.len = count;
889 for (; m; m = m->m_next) {
890 if (m->m_len >= count) {
891 m->m_len = count;
892 break;
893 }
894 count -= m->m_len;
895 }
896 while (m->m_next)
897 (m = m->m_next) ->m_len = 0;
898 }
899 }
900
901 /*
902 * Rearange an mbuf chain so that len bytes are contiguous
903 * and in the data area of an mbuf (so that mtod and dtom
904 * will work for a structure of size len). Returns the resulting
905 * mbuf chain on success, frees it and returns null on failure.
906 * If there is room, it will add up to max_protohdr-len extra bytes to the
907 * contiguous region in an attempt to avoid being called next time.
908 */
909 #define MPFail (mbstat.m_mpfail)
910
911 struct mbuf *
912 m_pullup(n, len)
913 register struct mbuf *n;
914 int len;
915 {
916 register struct mbuf *m;
917 register int count;
918 int space;
919
920 /*
921 * If first mbuf has no cluster, and has room for len bytes
922 * without shifting current data, pullup into it,
923 * otherwise allocate a new mbuf to prepend to the chain.
924 */
925 if ((n->m_flags & M_EXT) == 0 &&
926 n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
927 if (n->m_len >= len)
928 return (n);
929 m = n;
930 n = n->m_next;
931 len -= m->m_len;
932 } else {
933 if (len > MHLEN)
934 goto bad;
935 MGET(m, M_DONTWAIT, n->m_type);
936 if (m == 0)
937 goto bad;
938 m->m_len = 0;
939 if (n->m_flags & M_PKTHDR) {
940 M_COPY_PKTHDR(m, n);
941 n->m_flags &= ~M_PKTHDR;
942 }
943 }
944 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
945 do {
946 count = min(min(max(len, max_protohdr), space), n->m_len);
947 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
948 (unsigned)count);
949 len -= count;
950 m->m_len += count;
951 n->m_len -= count;
952 space -= count;
953 if (n->m_len)
954 n->m_data += count;
955 else
956 n = m_free(n);
957 } while (len > 0 && n);
958 if (len > 0) {
959 (void) m_free(m);
960 goto bad;
961 }
962 m->m_next = n;
963 return (m);
964 bad:
965 m_freem(n);
966 MPFail++;
967 return (0);
968 }
969
970 /*
971 * Partition an mbuf chain in two pieces, returning the tail --
972 * all but the first len0 bytes. In case of failure, it returns NULL and
973 * attempts to restore the chain to its original state.
974 */
975 struct mbuf *
976 m_split(m0, len0, wait)
977 register struct mbuf *m0;
978 int len0, wait;
979 {
980 register struct mbuf *m, *n;
981 unsigned len = len0, remain;
982
983 for (m = m0; m && len > m->m_len; m = m->m_next)
984 len -= m->m_len;
985 if (m == 0)
986 return (0);
987 remain = m->m_len - len;
988 if (m0->m_flags & M_PKTHDR) {
989 MGETHDR(n, wait, m0->m_type);
990 if (n == 0)
991 return (0);
992 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
993 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
994 m0->m_pkthdr.len = len0;
995 if (m->m_flags & M_EXT)
996 goto extpacket;
997 if (remain > MHLEN) {
998 /* m can't be the lead packet */
999 MH_ALIGN(n, 0);
1000 n->m_next = m_split(m, len, wait);
1001 if (n->m_next == 0) {
1002 (void) m_free(n);
1003 return (0);
1004 } else
1005 return (n);
1006 } else
1007 MH_ALIGN(n, remain);
1008 } else if (remain == 0) {
1009 n = m->m_next;
1010 m->m_next = 0;
1011 return (n);
1012 } else {
1013 MGET(n, wait, m->m_type);
1014 if (n == 0)
1015 return (0);
1016 M_ALIGN(n, remain);
1017 }
1018 extpacket:
1019 if (m->m_flags & M_EXT) {
1020 n->m_flags |= M_EXT;
1021 n->m_ext = m->m_ext;
1022 if(!m->m_ext.ext_ref)
1023 mclrefcnt[mtocl(m->m_ext.ext_buf)]++;
1024 else
1025 (*(m->m_ext.ext_ref))(m->m_ext.ext_buf,
1026 m->m_ext.ext_size);
1027 m->m_ext.ext_size = 0; /* For Accounting XXXXXX danger */
1028 n->m_data = m->m_data + len;
1029 } else {
1030 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
1031 }
1032 n->m_len = remain;
1033 m->m_len = len;
1034 n->m_next = m->m_next;
1035 m->m_next = 0;
1036 return (n);
1037 }
1038 /*
1039 * Routine to copy from device local memory into mbufs.
1040 */
1041 struct mbuf *
1042 m_devget(buf, totlen, off0, ifp, copy)
1043 char *buf;
1044 int totlen, off0;
1045 struct ifnet *ifp;
1046 void (*copy) __P((char *from, caddr_t to, u_int len));
1047 {
1048 register struct mbuf *m;
1049 struct mbuf *top = 0, **mp = ⊤
1050 register int off = off0, len;
1051 register char *cp;
1052 char *epkt;
1053
1054 cp = buf;
1055 epkt = cp + totlen;
1056 if (off) {
1057 cp += off + 2 * sizeof(u_short);
1058 totlen -= 2 * sizeof(u_short);
1059 }
1060 MGETHDR(m, M_DONTWAIT, MT_DATA);
1061 if (m == 0)
1062 return (0);
1063 m->m_pkthdr.rcvif = ifp;
1064 m->m_pkthdr.len = totlen;
1065 m->m_len = MHLEN;
1066
1067 while (totlen > 0) {
1068 if (top) {
1069 MGET(m, M_DONTWAIT, MT_DATA);
1070 if (m == 0) {
1071 m_freem(top);
1072 return (0);
1073 }
1074 m->m_len = MLEN;
1075 }
1076 len = min(totlen, epkt - cp);
1077 if (len >= MINCLSIZE) {
1078 MCLGET(m, M_DONTWAIT);
1079 if (m->m_flags & M_EXT)
1080 m->m_len = len = min(len, MCLBYTES);
1081 else
1082 len = m->m_len;
1083 } else {
1084 /*
1085 * Place initial small packet/header at end of mbuf.
1086 */
1087 if (len < m->m_len) {
1088 if (top == 0 && len + max_linkhdr <= m->m_len)
1089 m->m_data += max_linkhdr;
1090 m->m_len = len;
1091 } else
1092 len = m->m_len;
1093 }
1094 if (copy)
1095 copy(cp, mtod(m, caddr_t), (unsigned)len);
1096 else
1097 bcopy(cp, mtod(m, caddr_t), (unsigned)len);
1098 cp += len;
1099 *mp = m;
1100 mp = &m->m_next;
1101 totlen -= len;
1102 if (cp == epkt)
1103 cp = buf;
1104 }
1105 return (top);
1106 }
1107
1108 /*
1109 * Copy data from a buffer back into the indicated mbuf chain,
1110 * starting "off" bytes from the beginning, extending the mbuf
1111 * chain if necessary.
1112 */
1113 void
1114 m_copyback(m0, off, len, cp)
1115 struct mbuf *m0;
1116 register int off;
1117 register int len;
1118 caddr_t cp;
1119 {
1120 register int mlen;
1121 register struct mbuf *m = m0, *n;
1122 int totlen = 0;
1123
1124 if (m0 == 0)
1125 return;
1126 while (off > (mlen = m->m_len)) {
1127 off -= mlen;
1128 totlen += mlen;
1129 if (m->m_next == 0) {
1130 n = m_getclr(M_DONTWAIT, m->m_type);
1131 if (n == 0)
1132 goto out;
1133 n->m_len = min(MLEN, len + off);
1134 m->m_next = n;
1135 }
1136 m = m->m_next;
1137 }
1138 while (len > 0) {
1139 mlen = min (m->m_len - off, len);
1140 bcopy(cp, off + mtod(m, caddr_t), (unsigned)mlen);
1141 cp += mlen;
1142 len -= mlen;
1143 mlen += off;
1144 off = 0;
1145 totlen += mlen;
1146 if (len == 0)
1147 break;
1148 if (m->m_next == 0) {
1149 n = m_get(M_DONTWAIT, m->m_type);
1150 if (n == 0)
1151 break;
1152 n->m_len = min(MLEN, len);
1153 m->m_next = n;
1154 }
1155 m = m->m_next;
1156 }
1157 out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
1158 m->m_pkthdr.len = totlen;
1159 }
Cache object: 89605f036febb9043a040f522775ea90
|