blob: d285f2f7e812e6666df3a2da213c9f790a4f4abc [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Routines having to do with the 'struct sk_buff' memory handlers.
3 *
4 * Authors: Alan Cox <iiitac@pyr.swan.ac.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
6 *
7 * Version: $Id: skbuff.c,v 1.90 2001/11/07 05:56:19 davem Exp $
8 *
9 * Fixes:
10 * Alan Cox : Fixed the worst of the load
11 * balancer bugs.
12 * Dave Platt : Interrupt stacking fix.
13 * Richard Kooijman : Timestamp fixes.
14 * Alan Cox : Changed buffer format.
15 * Alan Cox : destructor hook for AF_UNIX etc.
16 * Linus Torvalds : Better skb_clone.
17 * Alan Cox : Added skb_copy.
18 * Alan Cox : Added all the changed routines Linus
19 * only put in the headers
20 * Ray VanTassle : Fixed --skb->lock in free
21 * Alan Cox : skb_copy copy arp field
22 * Andi Kleen : slabified it.
23 * Robert Olsson : Removed skb_head_pool
24 *
25 * NOTE:
26 * The __skb_ routines should be called with interrupts
27 * disabled, or you better be *real* sure that the operation is atomic
28 * with respect to whatever list is being frobbed (e.g. via lock_sock()
29 * or via disabling bottom half handlers, etc).
30 *
31 * This program is free software; you can redistribute it and/or
32 * modify it under the terms of the GNU General Public License
33 * as published by the Free Software Foundation; either version
34 * 2 of the License, or (at your option) any later version.
35 */
36
37/*
38 * The functions in this file will not compile correctly with gcc 2.4.x
39 */
40
41#include <linux/config.h>
42#include <linux/module.h>
43#include <linux/types.h>
44#include <linux/kernel.h>
45#include <linux/sched.h>
46#include <linux/mm.h>
47#include <linux/interrupt.h>
48#include <linux/in.h>
49#include <linux/inet.h>
50#include <linux/slab.h>
51#include <linux/netdevice.h>
52#ifdef CONFIG_NET_CLS_ACT
53#include <net/pkt_sched.h>
54#endif
55#include <linux/string.h>
56#include <linux/skbuff.h>
57#include <linux/cache.h>
58#include <linux/rtnetlink.h>
59#include <linux/init.h>
60#include <linux/highmem.h>
61
62#include <net/protocol.h>
63#include <net/dst.h>
64#include <net/sock.h>
65#include <net/checksum.h>
66#include <net/xfrm.h>
67
68#include <asm/uaccess.h>
69#include <asm/system.h>
70
71static kmem_cache_t *skbuff_head_cache;
72
73/*
74 * Keep out-of-line to prevent kernel bloat.
75 * __builtin_return_address is not used because it is not always
76 * reliable.
77 */
78
79/**
80 * skb_over_panic - private function
81 * @skb: buffer
82 * @sz: size
83 * @here: address
84 *
85 * Out of line support code for skb_put(). Not user callable.
86 */
87void skb_over_panic(struct sk_buff *skb, int sz, void *here)
88{
Patrick McHardy26095452005-04-21 16:43:02 -070089 printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
90 "data:%p tail:%p end:%p dev:%s\n",
91 here, skb->len, sz, skb->head, skb->data, skb->tail, skb->end,
92 skb->dev ? skb->dev->name : "<NULL>");
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 BUG();
94}
95
96/**
97 * skb_under_panic - private function
98 * @skb: buffer
99 * @sz: size
100 * @here: address
101 *
102 * Out of line support code for skb_push(). Not user callable.
103 */
104
105void skb_under_panic(struct sk_buff *skb, int sz, void *here)
106{
Patrick McHardy26095452005-04-21 16:43:02 -0700107 printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
108 "data:%p tail:%p end:%p dev:%s\n",
109 here, skb->len, sz, skb->head, skb->data, skb->tail, skb->end,
110 skb->dev ? skb->dev->name : "<NULL>");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 BUG();
112}
113
114/* Allocate a new skbuff. We do this ourselves so we can fill in a few
115 * 'private' fields and also do memory statistics to find all the
116 * [BEEP] leaks.
117 *
118 */
119
120/**
121 * alloc_skb - allocate a network buffer
122 * @size: size to allocate
123 * @gfp_mask: allocation mask
124 *
125 * Allocate a new &sk_buff. The returned buffer has no headroom and a
126 * tail room of size bytes. The object has a reference count of one.
127 * The return is the buffer. On a failure the return is %NULL.
128 *
129 * Buffers may only be allocated from interrupts using a @gfp_mask of
130 * %GFP_ATOMIC.
131 */
132struct sk_buff *alloc_skb(unsigned int size, int gfp_mask)
133{
134 struct sk_buff *skb;
135 u8 *data;
136
137 /* Get the HEAD */
138 skb = kmem_cache_alloc(skbuff_head_cache,
139 gfp_mask & ~__GFP_DMA);
140 if (!skb)
141 goto out;
142
143 /* Get the DATA. Size must match skb_add_mtu(). */
144 size = SKB_DATA_ALIGN(size);
145 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
146 if (!data)
147 goto nodata;
148
149 memset(skb, 0, offsetof(struct sk_buff, truesize));
150 skb->truesize = size + sizeof(struct sk_buff);
151 atomic_set(&skb->users, 1);
152 skb->head = data;
153 skb->data = data;
154 skb->tail = data;
155 skb->end = data + size;
156
157 atomic_set(&(skb_shinfo(skb)->dataref), 1);
158 skb_shinfo(skb)->nr_frags = 0;
159 skb_shinfo(skb)->tso_size = 0;
160 skb_shinfo(skb)->tso_segs = 0;
161 skb_shinfo(skb)->frag_list = NULL;
162out:
163 return skb;
164nodata:
165 kmem_cache_free(skbuff_head_cache, skb);
166 skb = NULL;
167 goto out;
168}
169
170/**
171 * alloc_skb_from_cache - allocate a network buffer
172 * @cp: kmem_cache from which to allocate the data area
173 * (object size must be big enough for @size bytes + skb overheads)
174 * @size: size to allocate
175 * @gfp_mask: allocation mask
176 *
177 * Allocate a new &sk_buff. The returned buffer has no headroom and
178 * tail room of size bytes. The object has a reference count of one.
179 * The return is the buffer. On a failure the return is %NULL.
180 *
181 * Buffers may only be allocated from interrupts using a @gfp_mask of
182 * %GFP_ATOMIC.
183 */
184struct sk_buff *alloc_skb_from_cache(kmem_cache_t *cp,
185 unsigned int size, int gfp_mask)
186{
187 struct sk_buff *skb;
188 u8 *data;
189
190 /* Get the HEAD */
191 skb = kmem_cache_alloc(skbuff_head_cache,
192 gfp_mask & ~__GFP_DMA);
193 if (!skb)
194 goto out;
195
196 /* Get the DATA. */
197 size = SKB_DATA_ALIGN(size);
198 data = kmem_cache_alloc(cp, gfp_mask);
199 if (!data)
200 goto nodata;
201
202 memset(skb, 0, offsetof(struct sk_buff, truesize));
203 skb->truesize = size + sizeof(struct sk_buff);
204 atomic_set(&skb->users, 1);
205 skb->head = data;
206 skb->data = data;
207 skb->tail = data;
208 skb->end = data + size;
209
210 atomic_set(&(skb_shinfo(skb)->dataref), 1);
211 skb_shinfo(skb)->nr_frags = 0;
212 skb_shinfo(skb)->tso_size = 0;
213 skb_shinfo(skb)->tso_segs = 0;
214 skb_shinfo(skb)->frag_list = NULL;
215out:
216 return skb;
217nodata:
218 kmem_cache_free(skbuff_head_cache, skb);
219 skb = NULL;
220 goto out;
221}
222
223
224static void skb_drop_fraglist(struct sk_buff *skb)
225{
226 struct sk_buff *list = skb_shinfo(skb)->frag_list;
227
228 skb_shinfo(skb)->frag_list = NULL;
229
230 do {
231 struct sk_buff *this = list;
232 list = list->next;
233 kfree_skb(this);
234 } while (list);
235}
236
237static void skb_clone_fraglist(struct sk_buff *skb)
238{
239 struct sk_buff *list;
240
241 for (list = skb_shinfo(skb)->frag_list; list; list = list->next)
242 skb_get(list);
243}
244
245void skb_release_data(struct sk_buff *skb)
246{
247 if (!skb->cloned ||
248 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
249 &skb_shinfo(skb)->dataref)) {
250 if (skb_shinfo(skb)->nr_frags) {
251 int i;
252 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
253 put_page(skb_shinfo(skb)->frags[i].page);
254 }
255
256 if (skb_shinfo(skb)->frag_list)
257 skb_drop_fraglist(skb);
258
259 kfree(skb->head);
260 }
261}
262
263/*
264 * Free an skbuff by memory without cleaning the state.
265 */
266void kfree_skbmem(struct sk_buff *skb)
267{
268 skb_release_data(skb);
269 kmem_cache_free(skbuff_head_cache, skb);
270}
271
272/**
273 * __kfree_skb - private function
274 * @skb: buffer
275 *
276 * Free an sk_buff. Release anything attached to the buffer.
277 * Clean the state. This is an internal helper function. Users should
278 * always call kfree_skb
279 */
280
281void __kfree_skb(struct sk_buff *skb)
282{
Stephen Hemminger9c2b3322005-04-19 22:39:42 -0700283 BUG_ON(skb->list != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284
285 dst_release(skb->dst);
286#ifdef CONFIG_XFRM
287 secpath_put(skb->sp);
288#endif
Stephen Hemminger9c2b3322005-04-19 22:39:42 -0700289 if (skb->destructor) {
290 WARN_ON(in_irq());
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 skb->destructor(skb);
292 }
293#ifdef CONFIG_NETFILTER
294 nf_conntrack_put(skb->nfct);
295#ifdef CONFIG_BRIDGE_NETFILTER
296 nf_bridge_put(skb->nf_bridge);
297#endif
298#endif
299/* XXX: IS this still necessary? - JHS */
300#ifdef CONFIG_NET_SCHED
301 skb->tc_index = 0;
302#ifdef CONFIG_NET_CLS_ACT
303 skb->tc_verd = 0;
304 skb->tc_classid = 0;
305#endif
306#endif
307
308 kfree_skbmem(skb);
309}
310
311/**
312 * skb_clone - duplicate an sk_buff
313 * @skb: buffer to clone
314 * @gfp_mask: allocation priority
315 *
316 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
317 * copies share the same packet data but not structure. The new
318 * buffer has a reference count of 1. If the allocation fails the
319 * function returns %NULL otherwise the new buffer is returned.
320 *
321 * If this function is called from an interrupt gfp_mask() must be
322 * %GFP_ATOMIC.
323 */
324
325struct sk_buff *skb_clone(struct sk_buff *skb, int gfp_mask)
326{
327 struct sk_buff *n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
328
329 if (!n)
330 return NULL;
331
332#define C(x) n->x = skb->x
333
334 n->next = n->prev = NULL;
335 n->list = NULL;
336 n->sk = NULL;
337 C(stamp);
338 C(dev);
339 C(real_dev);
340 C(h);
341 C(nh);
342 C(mac);
343 C(dst);
344 dst_clone(skb->dst);
345 C(sp);
346#ifdef CONFIG_INET
347 secpath_get(skb->sp);
348#endif
349 memcpy(n->cb, skb->cb, sizeof(skb->cb));
350 C(len);
351 C(data_len);
352 C(csum);
353 C(local_df);
354 n->cloned = 1;
355 n->nohdr = 0;
356 C(pkt_type);
357 C(ip_summed);
358 C(priority);
359 C(protocol);
360 C(security);
361 n->destructor = NULL;
362#ifdef CONFIG_NETFILTER
363 C(nfmark);
364 C(nfcache);
365 C(nfct);
366 nf_conntrack_get(skb->nfct);
367 C(nfctinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368#ifdef CONFIG_BRIDGE_NETFILTER
369 C(nf_bridge);
370 nf_bridge_get(skb->nf_bridge);
371#endif
372#endif /*CONFIG_NETFILTER*/
373#if defined(CONFIG_HIPPI)
374 C(private);
375#endif
376#ifdef CONFIG_NET_SCHED
377 C(tc_index);
378#ifdef CONFIG_NET_CLS_ACT
379 n->tc_verd = SET_TC_VERD(skb->tc_verd,0);
380 n->tc_verd = CLR_TC_OK2MUNGE(skb->tc_verd);
381 n->tc_verd = CLR_TC_MUNGED(skb->tc_verd);
382 C(input_dev);
383 C(tc_classid);
384#endif
385
386#endif
387 C(truesize);
388 atomic_set(&n->users, 1);
389 C(head);
390 C(data);
391 C(tail);
392 C(end);
393
394 atomic_inc(&(skb_shinfo(skb)->dataref));
395 skb->cloned = 1;
396
397 return n;
398}
399
400static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
401{
402 /*
403 * Shift between the two data areas in bytes
404 */
405 unsigned long offset = new->data - old->data;
406
407 new->list = NULL;
408 new->sk = NULL;
409 new->dev = old->dev;
410 new->real_dev = old->real_dev;
411 new->priority = old->priority;
412 new->protocol = old->protocol;
413 new->dst = dst_clone(old->dst);
414#ifdef CONFIG_INET
415 new->sp = secpath_get(old->sp);
416#endif
417 new->h.raw = old->h.raw + offset;
418 new->nh.raw = old->nh.raw + offset;
419 new->mac.raw = old->mac.raw + offset;
420 memcpy(new->cb, old->cb, sizeof(old->cb));
421 new->local_df = old->local_df;
422 new->pkt_type = old->pkt_type;
423 new->stamp = old->stamp;
424 new->destructor = NULL;
425 new->security = old->security;
426#ifdef CONFIG_NETFILTER
427 new->nfmark = old->nfmark;
428 new->nfcache = old->nfcache;
429 new->nfct = old->nfct;
430 nf_conntrack_get(old->nfct);
431 new->nfctinfo = old->nfctinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432#ifdef CONFIG_BRIDGE_NETFILTER
433 new->nf_bridge = old->nf_bridge;
434 nf_bridge_get(old->nf_bridge);
435#endif
436#endif
437#ifdef CONFIG_NET_SCHED
438#ifdef CONFIG_NET_CLS_ACT
439 new->tc_verd = old->tc_verd;
440#endif
441 new->tc_index = old->tc_index;
442#endif
443 atomic_set(&new->users, 1);
444 skb_shinfo(new)->tso_size = skb_shinfo(old)->tso_size;
445 skb_shinfo(new)->tso_segs = skb_shinfo(old)->tso_segs;
446}
447
448/**
449 * skb_copy - create private copy of an sk_buff
450 * @skb: buffer to copy
451 * @gfp_mask: allocation priority
452 *
453 * Make a copy of both an &sk_buff and its data. This is used when the
454 * caller wishes to modify the data and needs a private copy of the
455 * data to alter. Returns %NULL on failure or the pointer to the buffer
456 * on success. The returned buffer has a reference count of 1.
457 *
458 * As by-product this function converts non-linear &sk_buff to linear
459 * one, so that &sk_buff becomes completely private and caller is allowed
460 * to modify all the data of returned buffer. This means that this
461 * function is not recommended for use in circumstances when only
462 * header is going to be modified. Use pskb_copy() instead.
463 */
464
465struct sk_buff *skb_copy(const struct sk_buff *skb, int gfp_mask)
466{
467 int headerlen = skb->data - skb->head;
468 /*
469 * Allocate the copy buffer
470 */
471 struct sk_buff *n = alloc_skb(skb->end - skb->head + skb->data_len,
472 gfp_mask);
473 if (!n)
474 return NULL;
475
476 /* Set the data pointer */
477 skb_reserve(n, headerlen);
478 /* Set the tail pointer and length */
479 skb_put(n, skb->len);
480 n->csum = skb->csum;
481 n->ip_summed = skb->ip_summed;
482
483 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
484 BUG();
485
486 copy_skb_header(n, skb);
487 return n;
488}
489
490
491/**
492 * pskb_copy - create copy of an sk_buff with private head.
493 * @skb: buffer to copy
494 * @gfp_mask: allocation priority
495 *
496 * Make a copy of both an &sk_buff and part of its data, located
497 * in header. Fragmented data remain shared. This is used when
498 * the caller wishes to modify only header of &sk_buff and needs
499 * private copy of the header to alter. Returns %NULL on failure
500 * or the pointer to the buffer on success.
501 * The returned buffer has a reference count of 1.
502 */
503
504struct sk_buff *pskb_copy(struct sk_buff *skb, int gfp_mask)
505{
506 /*
507 * Allocate the copy buffer
508 */
509 struct sk_buff *n = alloc_skb(skb->end - skb->head, gfp_mask);
510
511 if (!n)
512 goto out;
513
514 /* Set the data pointer */
515 skb_reserve(n, skb->data - skb->head);
516 /* Set the tail pointer and length */
517 skb_put(n, skb_headlen(skb));
518 /* Copy the bytes */
519 memcpy(n->data, skb->data, n->len);
520 n->csum = skb->csum;
521 n->ip_summed = skb->ip_summed;
522
523 n->data_len = skb->data_len;
524 n->len = skb->len;
525
526 if (skb_shinfo(skb)->nr_frags) {
527 int i;
528
529 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
530 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
531 get_page(skb_shinfo(n)->frags[i].page);
532 }
533 skb_shinfo(n)->nr_frags = i;
534 }
535
536 if (skb_shinfo(skb)->frag_list) {
537 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
538 skb_clone_fraglist(n);
539 }
540
541 copy_skb_header(n, skb);
542out:
543 return n;
544}
545
546/**
547 * pskb_expand_head - reallocate header of &sk_buff
548 * @skb: buffer to reallocate
549 * @nhead: room to add at head
550 * @ntail: room to add at tail
551 * @gfp_mask: allocation priority
552 *
553 * Expands (or creates identical copy, if &nhead and &ntail are zero)
554 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
555 * reference count of 1. Returns zero in the case of success or error,
556 * if expansion failed. In the last case, &sk_buff is not changed.
557 *
558 * All the pointers pointing into skb header may change and must be
559 * reloaded after call to this function.
560 */
561
562int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, int gfp_mask)
563{
564 int i;
565 u8 *data;
566 int size = nhead + (skb->end - skb->head) + ntail;
567 long off;
568
569 if (skb_shared(skb))
570 BUG();
571
572 size = SKB_DATA_ALIGN(size);
573
574 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
575 if (!data)
576 goto nodata;
577
578 /* Copy only real data... and, alas, header. This should be
579 * optimized for the cases when header is void. */
580 memcpy(data + nhead, skb->head, skb->tail - skb->head);
581 memcpy(data + size, skb->end, sizeof(struct skb_shared_info));
582
583 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
584 get_page(skb_shinfo(skb)->frags[i].page);
585
586 if (skb_shinfo(skb)->frag_list)
587 skb_clone_fraglist(skb);
588
589 skb_release_data(skb);
590
591 off = (data + nhead) - skb->head;
592
593 skb->head = data;
594 skb->end = data + size;
595 skb->data += off;
596 skb->tail += off;
597 skb->mac.raw += off;
598 skb->h.raw += off;
599 skb->nh.raw += off;
600 skb->cloned = 0;
601 skb->nohdr = 0;
602 atomic_set(&skb_shinfo(skb)->dataref, 1);
603 return 0;
604
605nodata:
606 return -ENOMEM;
607}
608
609/* Make private copy of skb with writable head and some headroom */
610
611struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
612{
613 struct sk_buff *skb2;
614 int delta = headroom - skb_headroom(skb);
615
616 if (delta <= 0)
617 skb2 = pskb_copy(skb, GFP_ATOMIC);
618 else {
619 skb2 = skb_clone(skb, GFP_ATOMIC);
620 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
621 GFP_ATOMIC)) {
622 kfree_skb(skb2);
623 skb2 = NULL;
624 }
625 }
626 return skb2;
627}
628
629
630/**
631 * skb_copy_expand - copy and expand sk_buff
632 * @skb: buffer to copy
633 * @newheadroom: new free bytes at head
634 * @newtailroom: new free bytes at tail
635 * @gfp_mask: allocation priority
636 *
637 * Make a copy of both an &sk_buff and its data and while doing so
638 * allocate additional space.
639 *
640 * This is used when the caller wishes to modify the data and needs a
641 * private copy of the data to alter as well as more space for new fields.
642 * Returns %NULL on failure or the pointer to the buffer
643 * on success. The returned buffer has a reference count of 1.
644 *
645 * You must pass %GFP_ATOMIC as the allocation priority if this function
646 * is called from an interrupt.
647 *
648 * BUG ALERT: ip_summed is not copied. Why does this work? Is it used
649 * only by netfilter in the cases when checksum is recalculated? --ANK
650 */
651struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
652 int newheadroom, int newtailroom, int gfp_mask)
653{
654 /*
655 * Allocate the copy buffer
656 */
657 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
658 gfp_mask);
659 int head_copy_len, head_copy_off;
660
661 if (!n)
662 return NULL;
663
664 skb_reserve(n, newheadroom);
665
666 /* Set the tail pointer and length */
667 skb_put(n, skb->len);
668
669 head_copy_len = skb_headroom(skb);
670 head_copy_off = 0;
671 if (newheadroom <= head_copy_len)
672 head_copy_len = newheadroom;
673 else
674 head_copy_off = newheadroom - head_copy_len;
675
676 /* Copy the linear header and data. */
677 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
678 skb->len + head_copy_len))
679 BUG();
680
681 copy_skb_header(n, skb);
682
683 return n;
684}
685
686/**
687 * skb_pad - zero pad the tail of an skb
688 * @skb: buffer to pad
689 * @pad: space to pad
690 *
691 * Ensure that a buffer is followed by a padding area that is zero
692 * filled. Used by network drivers which may DMA or transfer data
693 * beyond the buffer end onto the wire.
694 *
695 * May return NULL in out of memory cases.
696 */
697
698struct sk_buff *skb_pad(struct sk_buff *skb, int pad)
699{
700 struct sk_buff *nskb;
701
702 /* If the skbuff is non linear tailroom is always zero.. */
703 if (skb_tailroom(skb) >= pad) {
704 memset(skb->data+skb->len, 0, pad);
705 return skb;
706 }
707
708 nskb = skb_copy_expand(skb, skb_headroom(skb), skb_tailroom(skb) + pad, GFP_ATOMIC);
709 kfree_skb(skb);
710 if (nskb)
711 memset(nskb->data+nskb->len, 0, pad);
712 return nskb;
713}
714
715/* Trims skb to length len. It can change skb pointers, if "realloc" is 1.
716 * If realloc==0 and trimming is impossible without change of data,
717 * it is BUG().
718 */
719
720int ___pskb_trim(struct sk_buff *skb, unsigned int len, int realloc)
721{
722 int offset = skb_headlen(skb);
723 int nfrags = skb_shinfo(skb)->nr_frags;
724 int i;
725
726 for (i = 0; i < nfrags; i++) {
727 int end = offset + skb_shinfo(skb)->frags[i].size;
728 if (end > len) {
729 if (skb_cloned(skb)) {
730 if (!realloc)
731 BUG();
732 if (pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
733 return -ENOMEM;
734 }
735 if (len <= offset) {
736 put_page(skb_shinfo(skb)->frags[i].page);
737 skb_shinfo(skb)->nr_frags--;
738 } else {
739 skb_shinfo(skb)->frags[i].size = len - offset;
740 }
741 }
742 offset = end;
743 }
744
745 if (offset < len) {
746 skb->data_len -= skb->len - len;
747 skb->len = len;
748 } else {
749 if (len <= skb_headlen(skb)) {
750 skb->len = len;
751 skb->data_len = 0;
752 skb->tail = skb->data + len;
753 if (skb_shinfo(skb)->frag_list && !skb_cloned(skb))
754 skb_drop_fraglist(skb);
755 } else {
756 skb->data_len -= skb->len - len;
757 skb->len = len;
758 }
759 }
760
761 return 0;
762}
763
764/**
765 * __pskb_pull_tail - advance tail of skb header
766 * @skb: buffer to reallocate
767 * @delta: number of bytes to advance tail
768 *
769 * The function makes a sense only on a fragmented &sk_buff,
770 * it expands header moving its tail forward and copying necessary
771 * data from fragmented part.
772 *
773 * &sk_buff MUST have reference count of 1.
774 *
775 * Returns %NULL (and &sk_buff does not change) if pull failed
776 * or value of new tail of skb in the case of success.
777 *
778 * All the pointers pointing into skb header may change and must be
779 * reloaded after call to this function.
780 */
781
782/* Moves tail of skb head forward, copying data from fragmented part,
783 * when it is necessary.
784 * 1. It may fail due to malloc failure.
785 * 2. It may change skb pointers.
786 *
787 * It is pretty complicated. Luckily, it is called only in exceptional cases.
788 */
789unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
790{
791 /* If skb has not enough free space at tail, get new one
792 * plus 128 bytes for future expansions. If we have enough
793 * room at tail, reallocate without expansion only if skb is cloned.
794 */
795 int i, k, eat = (skb->tail + delta) - skb->end;
796
797 if (eat > 0 || skb_cloned(skb)) {
798 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
799 GFP_ATOMIC))
800 return NULL;
801 }
802
803 if (skb_copy_bits(skb, skb_headlen(skb), skb->tail, delta))
804 BUG();
805
806 /* Optimization: no fragments, no reasons to preestimate
807 * size of pulled pages. Superb.
808 */
809 if (!skb_shinfo(skb)->frag_list)
810 goto pull_pages;
811
812 /* Estimate size of pulled pages. */
813 eat = delta;
814 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
815 if (skb_shinfo(skb)->frags[i].size >= eat)
816 goto pull_pages;
817 eat -= skb_shinfo(skb)->frags[i].size;
818 }
819
820 /* If we need update frag list, we are in troubles.
821 * Certainly, it possible to add an offset to skb data,
822 * but taking into account that pulling is expected to
823 * be very rare operation, it is worth to fight against
824 * further bloating skb head and crucify ourselves here instead.
825 * Pure masohism, indeed. 8)8)
826 */
827 if (eat) {
828 struct sk_buff *list = skb_shinfo(skb)->frag_list;
829 struct sk_buff *clone = NULL;
830 struct sk_buff *insp = NULL;
831
832 do {
833 if (!list)
834 BUG();
835
836 if (list->len <= eat) {
837 /* Eaten as whole. */
838 eat -= list->len;
839 list = list->next;
840 insp = list;
841 } else {
842 /* Eaten partially. */
843
844 if (skb_shared(list)) {
845 /* Sucks! We need to fork list. :-( */
846 clone = skb_clone(list, GFP_ATOMIC);
847 if (!clone)
848 return NULL;
849 insp = list->next;
850 list = clone;
851 } else {
852 /* This may be pulled without
853 * problems. */
854 insp = list;
855 }
856 if (!pskb_pull(list, eat)) {
857 if (clone)
858 kfree_skb(clone);
859 return NULL;
860 }
861 break;
862 }
863 } while (eat);
864
865 /* Free pulled out fragments. */
866 while ((list = skb_shinfo(skb)->frag_list) != insp) {
867 skb_shinfo(skb)->frag_list = list->next;
868 kfree_skb(list);
869 }
870 /* And insert new clone at head. */
871 if (clone) {
872 clone->next = list;
873 skb_shinfo(skb)->frag_list = clone;
874 }
875 }
876 /* Success! Now we may commit changes to skb data. */
877
878pull_pages:
879 eat = delta;
880 k = 0;
881 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
882 if (skb_shinfo(skb)->frags[i].size <= eat) {
883 put_page(skb_shinfo(skb)->frags[i].page);
884 eat -= skb_shinfo(skb)->frags[i].size;
885 } else {
886 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
887 if (eat) {
888 skb_shinfo(skb)->frags[k].page_offset += eat;
889 skb_shinfo(skb)->frags[k].size -= eat;
890 eat = 0;
891 }
892 k++;
893 }
894 }
895 skb_shinfo(skb)->nr_frags = k;
896
897 skb->tail += delta;
898 skb->data_len -= delta;
899
900 return skb->tail;
901}
902
903/* Copy some data bits from skb to kernel buffer. */
904
905int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
906{
907 int i, copy;
908 int start = skb_headlen(skb);
909
910 if (offset > (int)skb->len - len)
911 goto fault;
912
913 /* Copy header. */
914 if ((copy = start - offset) > 0) {
915 if (copy > len)
916 copy = len;
917 memcpy(to, skb->data + offset, copy);
918 if ((len -= copy) == 0)
919 return 0;
920 offset += copy;
921 to += copy;
922 }
923
924 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
925 int end;
926
927 BUG_TRAP(start <= offset + len);
928
929 end = start + skb_shinfo(skb)->frags[i].size;
930 if ((copy = end - offset) > 0) {
931 u8 *vaddr;
932
933 if (copy > len)
934 copy = len;
935
936 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
937 memcpy(to,
938 vaddr + skb_shinfo(skb)->frags[i].page_offset+
939 offset - start, copy);
940 kunmap_skb_frag(vaddr);
941
942 if ((len -= copy) == 0)
943 return 0;
944 offset += copy;
945 to += copy;
946 }
947 start = end;
948 }
949
950 if (skb_shinfo(skb)->frag_list) {
951 struct sk_buff *list = skb_shinfo(skb)->frag_list;
952
953 for (; list; list = list->next) {
954 int end;
955
956 BUG_TRAP(start <= offset + len);
957
958 end = start + list->len;
959 if ((copy = end - offset) > 0) {
960 if (copy > len)
961 copy = len;
962 if (skb_copy_bits(list, offset - start,
963 to, copy))
964 goto fault;
965 if ((len -= copy) == 0)
966 return 0;
967 offset += copy;
968 to += copy;
969 }
970 start = end;
971 }
972 }
973 if (!len)
974 return 0;
975
976fault:
977 return -EFAULT;
978}
979
Herbert Xu357b40a2005-04-19 22:30:14 -0700980/**
981 * skb_store_bits - store bits from kernel buffer to skb
982 * @skb: destination buffer
983 * @offset: offset in destination
984 * @from: source buffer
985 * @len: number of bytes to copy
986 *
987 * Copy the specified number of bytes from the source buffer to the
988 * destination skb. This function handles all the messy bits of
989 * traversing fragment lists and such.
990 */
991
992int skb_store_bits(const struct sk_buff *skb, int offset, void *from, int len)
993{
994 int i, copy;
995 int start = skb_headlen(skb);
996
997 if (offset > (int)skb->len - len)
998 goto fault;
999
1000 if ((copy = start - offset) > 0) {
1001 if (copy > len)
1002 copy = len;
1003 memcpy(skb->data + offset, from, copy);
1004 if ((len -= copy) == 0)
1005 return 0;
1006 offset += copy;
1007 from += copy;
1008 }
1009
1010 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1011 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1012 int end;
1013
1014 BUG_TRAP(start <= offset + len);
1015
1016 end = start + frag->size;
1017 if ((copy = end - offset) > 0) {
1018 u8 *vaddr;
1019
1020 if (copy > len)
1021 copy = len;
1022
1023 vaddr = kmap_skb_frag(frag);
1024 memcpy(vaddr + frag->page_offset + offset - start,
1025 from, copy);
1026 kunmap_skb_frag(vaddr);
1027
1028 if ((len -= copy) == 0)
1029 return 0;
1030 offset += copy;
1031 from += copy;
1032 }
1033 start = end;
1034 }
1035
1036 if (skb_shinfo(skb)->frag_list) {
1037 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1038
1039 for (; list; list = list->next) {
1040 int end;
1041
1042 BUG_TRAP(start <= offset + len);
1043
1044 end = start + list->len;
1045 if ((copy = end - offset) > 0) {
1046 if (copy > len)
1047 copy = len;
1048 if (skb_store_bits(list, offset - start,
1049 from, copy))
1050 goto fault;
1051 if ((len -= copy) == 0)
1052 return 0;
1053 offset += copy;
1054 from += copy;
1055 }
1056 start = end;
1057 }
1058 }
1059 if (!len)
1060 return 0;
1061
1062fault:
1063 return -EFAULT;
1064}
1065
1066EXPORT_SYMBOL(skb_store_bits);
1067
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068/* Checksum skb data. */
1069
1070unsigned int skb_checksum(const struct sk_buff *skb, int offset,
1071 int len, unsigned int csum)
1072{
1073 int start = skb_headlen(skb);
1074 int i, copy = start - offset;
1075 int pos = 0;
1076
1077 /* Checksum header. */
1078 if (copy > 0) {
1079 if (copy > len)
1080 copy = len;
1081 csum = csum_partial(skb->data + offset, copy, csum);
1082 if ((len -= copy) == 0)
1083 return csum;
1084 offset += copy;
1085 pos = copy;
1086 }
1087
1088 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1089 int end;
1090
1091 BUG_TRAP(start <= offset + len);
1092
1093 end = start + skb_shinfo(skb)->frags[i].size;
1094 if ((copy = end - offset) > 0) {
1095 unsigned int csum2;
1096 u8 *vaddr;
1097 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1098
1099 if (copy > len)
1100 copy = len;
1101 vaddr = kmap_skb_frag(frag);
1102 csum2 = csum_partial(vaddr + frag->page_offset +
1103 offset - start, copy, 0);
1104 kunmap_skb_frag(vaddr);
1105 csum = csum_block_add(csum, csum2, pos);
1106 if (!(len -= copy))
1107 return csum;
1108 offset += copy;
1109 pos += copy;
1110 }
1111 start = end;
1112 }
1113
1114 if (skb_shinfo(skb)->frag_list) {
1115 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1116
1117 for (; list; list = list->next) {
1118 int end;
1119
1120 BUG_TRAP(start <= offset + len);
1121
1122 end = start + list->len;
1123 if ((copy = end - offset) > 0) {
1124 unsigned int csum2;
1125 if (copy > len)
1126 copy = len;
1127 csum2 = skb_checksum(list, offset - start,
1128 copy, 0);
1129 csum = csum_block_add(csum, csum2, pos);
1130 if ((len -= copy) == 0)
1131 return csum;
1132 offset += copy;
1133 pos += copy;
1134 }
1135 start = end;
1136 }
1137 }
1138 if (len)
1139 BUG();
1140
1141 return csum;
1142}
1143
1144/* Both of above in one bottle. */
1145
1146unsigned int skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1147 u8 *to, int len, unsigned int csum)
1148{
1149 int start = skb_headlen(skb);
1150 int i, copy = start - offset;
1151 int pos = 0;
1152
1153 /* Copy header. */
1154 if (copy > 0) {
1155 if (copy > len)
1156 copy = len;
1157 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1158 copy, csum);
1159 if ((len -= copy) == 0)
1160 return csum;
1161 offset += copy;
1162 to += copy;
1163 pos = copy;
1164 }
1165
1166 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1167 int end;
1168
1169 BUG_TRAP(start <= offset + len);
1170
1171 end = start + skb_shinfo(skb)->frags[i].size;
1172 if ((copy = end - offset) > 0) {
1173 unsigned int csum2;
1174 u8 *vaddr;
1175 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1176
1177 if (copy > len)
1178 copy = len;
1179 vaddr = kmap_skb_frag(frag);
1180 csum2 = csum_partial_copy_nocheck(vaddr +
1181 frag->page_offset +
1182 offset - start, to,
1183 copy, 0);
1184 kunmap_skb_frag(vaddr);
1185 csum = csum_block_add(csum, csum2, pos);
1186 if (!(len -= copy))
1187 return csum;
1188 offset += copy;
1189 to += copy;
1190 pos += copy;
1191 }
1192 start = end;
1193 }
1194
1195 if (skb_shinfo(skb)->frag_list) {
1196 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1197
1198 for (; list; list = list->next) {
1199 unsigned int csum2;
1200 int end;
1201
1202 BUG_TRAP(start <= offset + len);
1203
1204 end = start + list->len;
1205 if ((copy = end - offset) > 0) {
1206 if (copy > len)
1207 copy = len;
1208 csum2 = skb_copy_and_csum_bits(list,
1209 offset - start,
1210 to, copy, 0);
1211 csum = csum_block_add(csum, csum2, pos);
1212 if ((len -= copy) == 0)
1213 return csum;
1214 offset += copy;
1215 to += copy;
1216 pos += copy;
1217 }
1218 start = end;
1219 }
1220 }
1221 if (len)
1222 BUG();
1223 return csum;
1224}
1225
1226void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1227{
1228 unsigned int csum;
1229 long csstart;
1230
1231 if (skb->ip_summed == CHECKSUM_HW)
1232 csstart = skb->h.raw - skb->data;
1233 else
1234 csstart = skb_headlen(skb);
1235
1236 if (csstart > skb_headlen(skb))
1237 BUG();
1238
1239 memcpy(to, skb->data, csstart);
1240
1241 csum = 0;
1242 if (csstart != skb->len)
1243 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1244 skb->len - csstart, 0);
1245
1246 if (skb->ip_summed == CHECKSUM_HW) {
1247 long csstuff = csstart + skb->csum;
1248
1249 *((unsigned short *)(to + csstuff)) = csum_fold(csum);
1250 }
1251}
1252
1253/**
1254 * skb_dequeue - remove from the head of the queue
1255 * @list: list to dequeue from
1256 *
1257 * Remove the head of the list. The list lock is taken so the function
1258 * may be used safely with other locking list functions. The head item is
1259 * returned or %NULL if the list is empty.
1260 */
1261
1262struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1263{
1264 unsigned long flags;
1265 struct sk_buff *result;
1266
1267 spin_lock_irqsave(&list->lock, flags);
1268 result = __skb_dequeue(list);
1269 spin_unlock_irqrestore(&list->lock, flags);
1270 return result;
1271}
1272
1273/**
1274 * skb_dequeue_tail - remove from the tail of the queue
1275 * @list: list to dequeue from
1276 *
1277 * Remove the tail of the list. The list lock is taken so the function
1278 * may be used safely with other locking list functions. The tail item is
1279 * returned or %NULL if the list is empty.
1280 */
1281struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1282{
1283 unsigned long flags;
1284 struct sk_buff *result;
1285
1286 spin_lock_irqsave(&list->lock, flags);
1287 result = __skb_dequeue_tail(list);
1288 spin_unlock_irqrestore(&list->lock, flags);
1289 return result;
1290}
1291
1292/**
1293 * skb_queue_purge - empty a list
1294 * @list: list to empty
1295 *
1296 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1297 * the list and one reference dropped. This function takes the list
1298 * lock and is atomic with respect to other list locking functions.
1299 */
1300void skb_queue_purge(struct sk_buff_head *list)
1301{
1302 struct sk_buff *skb;
1303 while ((skb = skb_dequeue(list)) != NULL)
1304 kfree_skb(skb);
1305}
1306
1307/**
1308 * skb_queue_head - queue a buffer at the list head
1309 * @list: list to use
1310 * @newsk: buffer to queue
1311 *
1312 * Queue a buffer at the start of the list. This function takes the
1313 * list lock and can be used safely with other locking &sk_buff functions
1314 * safely.
1315 *
1316 * A buffer cannot be placed on two lists at the same time.
1317 */
1318void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1319{
1320 unsigned long flags;
1321
1322 spin_lock_irqsave(&list->lock, flags);
1323 __skb_queue_head(list, newsk);
1324 spin_unlock_irqrestore(&list->lock, flags);
1325}
1326
1327/**
1328 * skb_queue_tail - queue a buffer at the list tail
1329 * @list: list to use
1330 * @newsk: buffer to queue
1331 *
1332 * Queue a buffer at the tail of the list. This function takes the
1333 * list lock and can be used safely with other locking &sk_buff functions
1334 * safely.
1335 *
1336 * A buffer cannot be placed on two lists at the same time.
1337 */
1338void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1339{
1340 unsigned long flags;
1341
1342 spin_lock_irqsave(&list->lock, flags);
1343 __skb_queue_tail(list, newsk);
1344 spin_unlock_irqrestore(&list->lock, flags);
1345}
1346/**
1347 * skb_unlink - remove a buffer from a list
1348 * @skb: buffer to remove
1349 *
1350 * Place a packet after a given packet in a list. The list locks are taken
1351 * and this function is atomic with respect to other list locked calls
1352 *
1353 * Works even without knowing the list it is sitting on, which can be
1354 * handy at times. It also means that THE LIST MUST EXIST when you
1355 * unlink. Thus a list must have its contents unlinked before it is
1356 * destroyed.
1357 */
1358void skb_unlink(struct sk_buff *skb)
1359{
1360 struct sk_buff_head *list = skb->list;
1361
1362 if (list) {
1363 unsigned long flags;
1364
1365 spin_lock_irqsave(&list->lock, flags);
1366 if (skb->list == list)
1367 __skb_unlink(skb, skb->list);
1368 spin_unlock_irqrestore(&list->lock, flags);
1369 }
1370}
1371
1372
1373/**
1374 * skb_append - append a buffer
1375 * @old: buffer to insert after
1376 * @newsk: buffer to insert
1377 *
1378 * Place a packet after a given packet in a list. The list locks are taken
1379 * and this function is atomic with respect to other list locked calls.
1380 * A buffer cannot be placed on two lists at the same time.
1381 */
1382
1383void skb_append(struct sk_buff *old, struct sk_buff *newsk)
1384{
1385 unsigned long flags;
1386
1387 spin_lock_irqsave(&old->list->lock, flags);
1388 __skb_append(old, newsk);
1389 spin_unlock_irqrestore(&old->list->lock, flags);
1390}
1391
1392
1393/**
1394 * skb_insert - insert a buffer
1395 * @old: buffer to insert before
1396 * @newsk: buffer to insert
1397 *
1398 * Place a packet before a given packet in a list. The list locks are taken
1399 * and this function is atomic with respect to other list locked calls
1400 * A buffer cannot be placed on two lists at the same time.
1401 */
1402
1403void skb_insert(struct sk_buff *old, struct sk_buff *newsk)
1404{
1405 unsigned long flags;
1406
1407 spin_lock_irqsave(&old->list->lock, flags);
1408 __skb_insert(newsk, old->prev, old, old->list);
1409 spin_unlock_irqrestore(&old->list->lock, flags);
1410}
1411
1412#if 0
1413/*
1414 * Tune the memory allocator for a new MTU size.
1415 */
1416void skb_add_mtu(int mtu)
1417{
1418 /* Must match allocation in alloc_skb */
1419 mtu = SKB_DATA_ALIGN(mtu) + sizeof(struct skb_shared_info);
1420
1421 kmem_add_cache_size(mtu);
1422}
1423#endif
1424
1425static inline void skb_split_inside_header(struct sk_buff *skb,
1426 struct sk_buff* skb1,
1427 const u32 len, const int pos)
1428{
1429 int i;
1430
1431 memcpy(skb_put(skb1, pos - len), skb->data + len, pos - len);
1432
1433 /* And move data appendix as is. */
1434 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1435 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
1436
1437 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
1438 skb_shinfo(skb)->nr_frags = 0;
1439 skb1->data_len = skb->data_len;
1440 skb1->len += skb1->data_len;
1441 skb->data_len = 0;
1442 skb->len = len;
1443 skb->tail = skb->data + len;
1444}
1445
1446static inline void skb_split_no_header(struct sk_buff *skb,
1447 struct sk_buff* skb1,
1448 const u32 len, int pos)
1449{
1450 int i, k = 0;
1451 const int nfrags = skb_shinfo(skb)->nr_frags;
1452
1453 skb_shinfo(skb)->nr_frags = 0;
1454 skb1->len = skb1->data_len = skb->len - len;
1455 skb->len = len;
1456 skb->data_len = len - pos;
1457
1458 for (i = 0; i < nfrags; i++) {
1459 int size = skb_shinfo(skb)->frags[i].size;
1460
1461 if (pos + size > len) {
1462 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
1463
1464 if (pos < len) {
1465 /* Split frag.
1466 * We have two variants in this case:
1467 * 1. Move all the frag to the second
1468 * part, if it is possible. F.e.
1469 * this approach is mandatory for TUX,
1470 * where splitting is expensive.
1471 * 2. Split is accurately. We make this.
1472 */
1473 get_page(skb_shinfo(skb)->frags[i].page);
1474 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
1475 skb_shinfo(skb1)->frags[0].size -= len - pos;
1476 skb_shinfo(skb)->frags[i].size = len - pos;
1477 skb_shinfo(skb)->nr_frags++;
1478 }
1479 k++;
1480 } else
1481 skb_shinfo(skb)->nr_frags++;
1482 pos += size;
1483 }
1484 skb_shinfo(skb1)->nr_frags = k;
1485}
1486
1487/**
1488 * skb_split - Split fragmented skb to two parts at length len.
1489 * @skb: the buffer to split
1490 * @skb1: the buffer to receive the second part
1491 * @len: new length for skb
1492 */
1493void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
1494{
1495 int pos = skb_headlen(skb);
1496
1497 if (len < pos) /* Split line is inside header. */
1498 skb_split_inside_header(skb, skb1, len, pos);
1499 else /* Second chunk has no header, nothing to copy. */
1500 skb_split_no_header(skb, skb1, len, pos);
1501}
1502
Thomas Graf677e90e2005-06-23 20:59:51 -07001503/**
1504 * skb_prepare_seq_read - Prepare a sequential read of skb data
1505 * @skb: the buffer to read
1506 * @from: lower offset of data to be read
1507 * @to: upper offset of data to be read
1508 * @st: state variable
1509 *
1510 * Initializes the specified state variable. Must be called before
1511 * invoking skb_seq_read() for the first time.
1512 */
1513void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1514 unsigned int to, struct skb_seq_state *st)
1515{
1516 st->lower_offset = from;
1517 st->upper_offset = to;
1518 st->root_skb = st->cur_skb = skb;
1519 st->frag_idx = st->stepped_offset = 0;
1520 st->frag_data = NULL;
1521}
1522
1523/**
1524 * skb_seq_read - Sequentially read skb data
1525 * @consumed: number of bytes consumed by the caller so far
1526 * @data: destination pointer for data to be returned
1527 * @st: state variable
1528 *
1529 * Reads a block of skb data at &consumed relative to the
1530 * lower offset specified to skb_prepare_seq_read(). Assigns
1531 * the head of the data block to &data and returns the length
1532 * of the block or 0 if the end of the skb data or the upper
1533 * offset has been reached.
1534 *
1535 * The caller is not required to consume all of the data
1536 * returned, i.e. &consumed is typically set to the number
1537 * of bytes already consumed and the next call to
1538 * skb_seq_read() will return the remaining part of the block.
1539 *
1540 * Note: The size of each block of data returned can be arbitary,
1541 * this limitation is the cost for zerocopy seqeuental
1542 * reads of potentially non linear data.
1543 *
1544 * Note: Fragment lists within fragments are not implemented
1545 * at the moment, state->root_skb could be replaced with
1546 * a stack for this purpose.
1547 */
1548unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1549 struct skb_seq_state *st)
1550{
1551 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
1552 skb_frag_t *frag;
1553
1554 if (unlikely(abs_offset >= st->upper_offset))
1555 return 0;
1556
1557next_skb:
1558 block_limit = skb_headlen(st->cur_skb);
1559
1560 if (abs_offset < block_limit) {
1561 *data = st->cur_skb->data + abs_offset;
1562 return block_limit - abs_offset;
1563 }
1564
1565 if (st->frag_idx == 0 && !st->frag_data)
1566 st->stepped_offset += skb_headlen(st->cur_skb);
1567
1568 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
1569 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
1570 block_limit = frag->size + st->stepped_offset;
1571
1572 if (abs_offset < block_limit) {
1573 if (!st->frag_data)
1574 st->frag_data = kmap_skb_frag(frag);
1575
1576 *data = (u8 *) st->frag_data + frag->page_offset +
1577 (abs_offset - st->stepped_offset);
1578
1579 return block_limit - abs_offset;
1580 }
1581
1582 if (st->frag_data) {
1583 kunmap_skb_frag(st->frag_data);
1584 st->frag_data = NULL;
1585 }
1586
1587 st->frag_idx++;
1588 st->stepped_offset += frag->size;
1589 }
1590
1591 if (st->cur_skb->next) {
1592 st->cur_skb = st->cur_skb->next;
1593 st->frag_idx = 0;
1594 goto next_skb;
1595 } else if (st->root_skb == st->cur_skb &&
1596 skb_shinfo(st->root_skb)->frag_list) {
1597 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
1598 goto next_skb;
1599 }
1600
1601 return 0;
1602}
1603
1604/**
1605 * skb_abort_seq_read - Abort a sequential read of skb data
1606 * @st: state variable
1607 *
1608 * Must be called if skb_seq_read() was not called until it
1609 * returned 0.
1610 */
1611void skb_abort_seq_read(struct skb_seq_state *st)
1612{
1613 if (st->frag_data)
1614 kunmap_skb_frag(st->frag_data);
1615}
1616
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617void __init skb_init(void)
1618{
1619 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
1620 sizeof(struct sk_buff),
1621 0,
1622 SLAB_HWCACHE_ALIGN,
1623 NULL, NULL);
1624 if (!skbuff_head_cache)
1625 panic("cannot create skbuff cache");
1626}
1627
1628EXPORT_SYMBOL(___pskb_trim);
1629EXPORT_SYMBOL(__kfree_skb);
1630EXPORT_SYMBOL(__pskb_pull_tail);
1631EXPORT_SYMBOL(alloc_skb);
1632EXPORT_SYMBOL(pskb_copy);
1633EXPORT_SYMBOL(pskb_expand_head);
1634EXPORT_SYMBOL(skb_checksum);
1635EXPORT_SYMBOL(skb_clone);
1636EXPORT_SYMBOL(skb_clone_fraglist);
1637EXPORT_SYMBOL(skb_copy);
1638EXPORT_SYMBOL(skb_copy_and_csum_bits);
1639EXPORT_SYMBOL(skb_copy_and_csum_dev);
1640EXPORT_SYMBOL(skb_copy_bits);
1641EXPORT_SYMBOL(skb_copy_expand);
1642EXPORT_SYMBOL(skb_over_panic);
1643EXPORT_SYMBOL(skb_pad);
1644EXPORT_SYMBOL(skb_realloc_headroom);
1645EXPORT_SYMBOL(skb_under_panic);
1646EXPORT_SYMBOL(skb_dequeue);
1647EXPORT_SYMBOL(skb_dequeue_tail);
1648EXPORT_SYMBOL(skb_insert);
1649EXPORT_SYMBOL(skb_queue_purge);
1650EXPORT_SYMBOL(skb_queue_head);
1651EXPORT_SYMBOL(skb_queue_tail);
1652EXPORT_SYMBOL(skb_unlink);
1653EXPORT_SYMBOL(skb_append);
1654EXPORT_SYMBOL(skb_split);
Thomas Graf677e90e2005-06-23 20:59:51 -07001655EXPORT_SYMBOL(skb_prepare_seq_read);
1656EXPORT_SYMBOL(skb_seq_read);
1657EXPORT_SYMBOL(skb_abort_seq_read);