blob: 7eab867ede5938a4fb698df683876937ce106364 [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 */
Victor Fusco86a76ca2005-07-08 14:57:47 -0700132struct sk_buff *alloc_skb(unsigned int size, unsigned int __nocast gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133{
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,
Victor Fusco86a76ca2005-07-08 14:57:47 -0700185 unsigned int size,
186 unsigned int __nocast gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187{
188 struct sk_buff *skb;
189 u8 *data;
190
191 /* Get the HEAD */
192 skb = kmem_cache_alloc(skbuff_head_cache,
193 gfp_mask & ~__GFP_DMA);
194 if (!skb)
195 goto out;
196
197 /* Get the DATA. */
198 size = SKB_DATA_ALIGN(size);
199 data = kmem_cache_alloc(cp, gfp_mask);
200 if (!data)
201 goto nodata;
202
203 memset(skb, 0, offsetof(struct sk_buff, truesize));
204 skb->truesize = size + sizeof(struct sk_buff);
205 atomic_set(&skb->users, 1);
206 skb->head = data;
207 skb->data = data;
208 skb->tail = data;
209 skb->end = data + size;
210
211 atomic_set(&(skb_shinfo(skb)->dataref), 1);
212 skb_shinfo(skb)->nr_frags = 0;
213 skb_shinfo(skb)->tso_size = 0;
214 skb_shinfo(skb)->tso_segs = 0;
215 skb_shinfo(skb)->frag_list = NULL;
216out:
217 return skb;
218nodata:
219 kmem_cache_free(skbuff_head_cache, skb);
220 skb = NULL;
221 goto out;
222}
223
224
225static void skb_drop_fraglist(struct sk_buff *skb)
226{
227 struct sk_buff *list = skb_shinfo(skb)->frag_list;
228
229 skb_shinfo(skb)->frag_list = NULL;
230
231 do {
232 struct sk_buff *this = list;
233 list = list->next;
234 kfree_skb(this);
235 } while (list);
236}
237
238static void skb_clone_fraglist(struct sk_buff *skb)
239{
240 struct sk_buff *list;
241
242 for (list = skb_shinfo(skb)->frag_list; list; list = list->next)
243 skb_get(list);
244}
245
246void skb_release_data(struct sk_buff *skb)
247{
248 if (!skb->cloned ||
249 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
250 &skb_shinfo(skb)->dataref)) {
251 if (skb_shinfo(skb)->nr_frags) {
252 int i;
253 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
254 put_page(skb_shinfo(skb)->frags[i].page);
255 }
256
257 if (skb_shinfo(skb)->frag_list)
258 skb_drop_fraglist(skb);
259
260 kfree(skb->head);
261 }
262}
263
264/*
265 * Free an skbuff by memory without cleaning the state.
266 */
267void kfree_skbmem(struct sk_buff *skb)
268{
269 skb_release_data(skb);
270 kmem_cache_free(skbuff_head_cache, skb);
271}
272
273/**
274 * __kfree_skb - private function
275 * @skb: buffer
276 *
277 * Free an sk_buff. Release anything attached to the buffer.
278 * Clean the state. This is an internal helper function. Users should
279 * always call kfree_skb
280 */
281
282void __kfree_skb(struct sk_buff *skb)
283{
Stephen Hemminger9c2b3322005-04-19 22:39:42 -0700284 BUG_ON(skb->list != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285
286 dst_release(skb->dst);
287#ifdef CONFIG_XFRM
288 secpath_put(skb->sp);
289#endif
Stephen Hemminger9c2b3322005-04-19 22:39:42 -0700290 if (skb->destructor) {
291 WARN_ON(in_irq());
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 skb->destructor(skb);
293 }
294#ifdef CONFIG_NETFILTER
295 nf_conntrack_put(skb->nfct);
296#ifdef CONFIG_BRIDGE_NETFILTER
297 nf_bridge_put(skb->nf_bridge);
298#endif
299#endif
300/* XXX: IS this still necessary? - JHS */
301#ifdef CONFIG_NET_SCHED
302 skb->tc_index = 0;
303#ifdef CONFIG_NET_CLS_ACT
304 skb->tc_verd = 0;
305 skb->tc_classid = 0;
306#endif
307#endif
308
309 kfree_skbmem(skb);
310}
311
312/**
313 * skb_clone - duplicate an sk_buff
314 * @skb: buffer to clone
315 * @gfp_mask: allocation priority
316 *
317 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
318 * copies share the same packet data but not structure. The new
319 * buffer has a reference count of 1. If the allocation fails the
320 * function returns %NULL otherwise the new buffer is returned.
321 *
322 * If this function is called from an interrupt gfp_mask() must be
323 * %GFP_ATOMIC.
324 */
325
Victor Fusco86a76ca2005-07-08 14:57:47 -0700326struct sk_buff *skb_clone(struct sk_buff *skb, unsigned int __nocast gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327{
328 struct sk_buff *n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
329
330 if (!n)
331 return NULL;
332
333#define C(x) n->x = skb->x
334
335 n->next = n->prev = NULL;
336 n->list = NULL;
337 n->sk = NULL;
338 C(stamp);
339 C(dev);
340 C(real_dev);
341 C(h);
342 C(nh);
343 C(mac);
344 C(dst);
345 dst_clone(skb->dst);
346 C(sp);
347#ifdef CONFIG_INET
348 secpath_get(skb->sp);
349#endif
350 memcpy(n->cb, skb->cb, sizeof(skb->cb));
351 C(len);
352 C(data_len);
353 C(csum);
354 C(local_df);
355 n->cloned = 1;
356 n->nohdr = 0;
357 C(pkt_type);
358 C(ip_summed);
359 C(priority);
360 C(protocol);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 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);
David S. Millerb72f6ec2005-07-19 14:13:54 -0700380 n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd);
381 n->tc_verd = CLR_TC_MUNGED(n->tc_verd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425#ifdef CONFIG_NETFILTER
426 new->nfmark = old->nfmark;
427 new->nfcache = old->nfcache;
428 new->nfct = old->nfct;
429 nf_conntrack_get(old->nfct);
430 new->nfctinfo = old->nfctinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431#ifdef CONFIG_BRIDGE_NETFILTER
432 new->nf_bridge = old->nf_bridge;
433 nf_bridge_get(old->nf_bridge);
434#endif
435#endif
436#ifdef CONFIG_NET_SCHED
437#ifdef CONFIG_NET_CLS_ACT
438 new->tc_verd = old->tc_verd;
439#endif
440 new->tc_index = old->tc_index;
441#endif
442 atomic_set(&new->users, 1);
443 skb_shinfo(new)->tso_size = skb_shinfo(old)->tso_size;
444 skb_shinfo(new)->tso_segs = skb_shinfo(old)->tso_segs;
445}
446
447/**
448 * skb_copy - create private copy of an sk_buff
449 * @skb: buffer to copy
450 * @gfp_mask: allocation priority
451 *
452 * Make a copy of both an &sk_buff and its data. This is used when the
453 * caller wishes to modify the data and needs a private copy of the
454 * data to alter. Returns %NULL on failure or the pointer to the buffer
455 * on success. The returned buffer has a reference count of 1.
456 *
457 * As by-product this function converts non-linear &sk_buff to linear
458 * one, so that &sk_buff becomes completely private and caller is allowed
459 * to modify all the data of returned buffer. This means that this
460 * function is not recommended for use in circumstances when only
461 * header is going to be modified. Use pskb_copy() instead.
462 */
463
Victor Fusco86a76ca2005-07-08 14:57:47 -0700464struct sk_buff *skb_copy(const struct sk_buff *skb, unsigned int __nocast gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465{
466 int headerlen = skb->data - skb->head;
467 /*
468 * Allocate the copy buffer
469 */
470 struct sk_buff *n = alloc_skb(skb->end - skb->head + skb->data_len,
471 gfp_mask);
472 if (!n)
473 return NULL;
474
475 /* Set the data pointer */
476 skb_reserve(n, headerlen);
477 /* Set the tail pointer and length */
478 skb_put(n, skb->len);
479 n->csum = skb->csum;
480 n->ip_summed = skb->ip_summed;
481
482 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
483 BUG();
484
485 copy_skb_header(n, skb);
486 return n;
487}
488
489
490/**
491 * pskb_copy - create copy of an sk_buff with private head.
492 * @skb: buffer to copy
493 * @gfp_mask: allocation priority
494 *
495 * Make a copy of both an &sk_buff and part of its data, located
496 * in header. Fragmented data remain shared. This is used when
497 * the caller wishes to modify only header of &sk_buff and needs
498 * private copy of the header to alter. Returns %NULL on failure
499 * or the pointer to the buffer on success.
500 * The returned buffer has a reference count of 1.
501 */
502
Victor Fusco86a76ca2005-07-08 14:57:47 -0700503struct sk_buff *pskb_copy(struct sk_buff *skb, unsigned int __nocast gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504{
505 /*
506 * Allocate the copy buffer
507 */
508 struct sk_buff *n = alloc_skb(skb->end - skb->head, gfp_mask);
509
510 if (!n)
511 goto out;
512
513 /* Set the data pointer */
514 skb_reserve(n, skb->data - skb->head);
515 /* Set the tail pointer and length */
516 skb_put(n, skb_headlen(skb));
517 /* Copy the bytes */
518 memcpy(n->data, skb->data, n->len);
519 n->csum = skb->csum;
520 n->ip_summed = skb->ip_summed;
521
522 n->data_len = skb->data_len;
523 n->len = skb->len;
524
525 if (skb_shinfo(skb)->nr_frags) {
526 int i;
527
528 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
529 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
530 get_page(skb_shinfo(n)->frags[i].page);
531 }
532 skb_shinfo(n)->nr_frags = i;
533 }
534
535 if (skb_shinfo(skb)->frag_list) {
536 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
537 skb_clone_fraglist(n);
538 }
539
540 copy_skb_header(n, skb);
541out:
542 return n;
543}
544
545/**
546 * pskb_expand_head - reallocate header of &sk_buff
547 * @skb: buffer to reallocate
548 * @nhead: room to add at head
549 * @ntail: room to add at tail
550 * @gfp_mask: allocation priority
551 *
552 * Expands (or creates identical copy, if &nhead and &ntail are zero)
553 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
554 * reference count of 1. Returns zero in the case of success or error,
555 * if expansion failed. In the last case, &sk_buff is not changed.
556 *
557 * All the pointers pointing into skb header may change and must be
558 * reloaded after call to this function.
559 */
560
Victor Fusco86a76ca2005-07-08 14:57:47 -0700561int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
562 unsigned int __nocast gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563{
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,
Victor Fusco86a76ca2005-07-08 14:57:47 -0700652 int newheadroom, int newtailroom,
653 unsigned int __nocast gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654{
655 /*
656 * Allocate the copy buffer
657 */
658 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
659 gfp_mask);
660 int head_copy_len, head_copy_off;
661
662 if (!n)
663 return NULL;
664
665 skb_reserve(n, newheadroom);
666
667 /* Set the tail pointer and length */
668 skb_put(n, skb->len);
669
670 head_copy_len = skb_headroom(skb);
671 head_copy_off = 0;
672 if (newheadroom <= head_copy_len)
673 head_copy_len = newheadroom;
674 else
675 head_copy_off = newheadroom - head_copy_len;
676
677 /* Copy the linear header and data. */
678 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
679 skb->len + head_copy_len))
680 BUG();
681
682 copy_skb_header(n, skb);
683
684 return n;
685}
686
687/**
688 * skb_pad - zero pad the tail of an skb
689 * @skb: buffer to pad
690 * @pad: space to pad
691 *
692 * Ensure that a buffer is followed by a padding area that is zero
693 * filled. Used by network drivers which may DMA or transfer data
694 * beyond the buffer end onto the wire.
695 *
696 * May return NULL in out of memory cases.
697 */
698
699struct sk_buff *skb_pad(struct sk_buff *skb, int pad)
700{
701 struct sk_buff *nskb;
702
703 /* If the skbuff is non linear tailroom is always zero.. */
704 if (skb_tailroom(skb) >= pad) {
705 memset(skb->data+skb->len, 0, pad);
706 return skb;
707 }
708
709 nskb = skb_copy_expand(skb, skb_headroom(skb), skb_tailroom(skb) + pad, GFP_ATOMIC);
710 kfree_skb(skb);
711 if (nskb)
712 memset(nskb->data+nskb->len, 0, pad);
713 return nskb;
714}
715
716/* Trims skb to length len. It can change skb pointers, if "realloc" is 1.
717 * If realloc==0 and trimming is impossible without change of data,
718 * it is BUG().
719 */
720
721int ___pskb_trim(struct sk_buff *skb, unsigned int len, int realloc)
722{
723 int offset = skb_headlen(skb);
724 int nfrags = skb_shinfo(skb)->nr_frags;
725 int i;
726
727 for (i = 0; i < nfrags; i++) {
728 int end = offset + skb_shinfo(skb)->frags[i].size;
729 if (end > len) {
730 if (skb_cloned(skb)) {
731 if (!realloc)
732 BUG();
733 if (pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
734 return -ENOMEM;
735 }
736 if (len <= offset) {
737 put_page(skb_shinfo(skb)->frags[i].page);
738 skb_shinfo(skb)->nr_frags--;
739 } else {
740 skb_shinfo(skb)->frags[i].size = len - offset;
741 }
742 }
743 offset = end;
744 }
745
746 if (offset < len) {
747 skb->data_len -= skb->len - len;
748 skb->len = len;
749 } else {
750 if (len <= skb_headlen(skb)) {
751 skb->len = len;
752 skb->data_len = 0;
753 skb->tail = skb->data + len;
754 if (skb_shinfo(skb)->frag_list && !skb_cloned(skb))
755 skb_drop_fraglist(skb);
756 } else {
757 skb->data_len -= skb->len - len;
758 skb->len = len;
759 }
760 }
761
762 return 0;
763}
764
765/**
766 * __pskb_pull_tail - advance tail of skb header
767 * @skb: buffer to reallocate
768 * @delta: number of bytes to advance tail
769 *
770 * The function makes a sense only on a fragmented &sk_buff,
771 * it expands header moving its tail forward and copying necessary
772 * data from fragmented part.
773 *
774 * &sk_buff MUST have reference count of 1.
775 *
776 * Returns %NULL (and &sk_buff does not change) if pull failed
777 * or value of new tail of skb in the case of success.
778 *
779 * All the pointers pointing into skb header may change and must be
780 * reloaded after call to this function.
781 */
782
783/* Moves tail of skb head forward, copying data from fragmented part,
784 * when it is necessary.
785 * 1. It may fail due to malloc failure.
786 * 2. It may change skb pointers.
787 *
788 * It is pretty complicated. Luckily, it is called only in exceptional cases.
789 */
790unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
791{
792 /* If skb has not enough free space at tail, get new one
793 * plus 128 bytes for future expansions. If we have enough
794 * room at tail, reallocate without expansion only if skb is cloned.
795 */
796 int i, k, eat = (skb->tail + delta) - skb->end;
797
798 if (eat > 0 || skb_cloned(skb)) {
799 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
800 GFP_ATOMIC))
801 return NULL;
802 }
803
804 if (skb_copy_bits(skb, skb_headlen(skb), skb->tail, delta))
805 BUG();
806
807 /* Optimization: no fragments, no reasons to preestimate
808 * size of pulled pages. Superb.
809 */
810 if (!skb_shinfo(skb)->frag_list)
811 goto pull_pages;
812
813 /* Estimate size of pulled pages. */
814 eat = delta;
815 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
816 if (skb_shinfo(skb)->frags[i].size >= eat)
817 goto pull_pages;
818 eat -= skb_shinfo(skb)->frags[i].size;
819 }
820
821 /* If we need update frag list, we are in troubles.
822 * Certainly, it possible to add an offset to skb data,
823 * but taking into account that pulling is expected to
824 * be very rare operation, it is worth to fight against
825 * further bloating skb head and crucify ourselves here instead.
826 * Pure masohism, indeed. 8)8)
827 */
828 if (eat) {
829 struct sk_buff *list = skb_shinfo(skb)->frag_list;
830 struct sk_buff *clone = NULL;
831 struct sk_buff *insp = NULL;
832
833 do {
834 if (!list)
835 BUG();
836
837 if (list->len <= eat) {
838 /* Eaten as whole. */
839 eat -= list->len;
840 list = list->next;
841 insp = list;
842 } else {
843 /* Eaten partially. */
844
845 if (skb_shared(list)) {
846 /* Sucks! We need to fork list. :-( */
847 clone = skb_clone(list, GFP_ATOMIC);
848 if (!clone)
849 return NULL;
850 insp = list->next;
851 list = clone;
852 } else {
853 /* This may be pulled without
854 * problems. */
855 insp = list;
856 }
857 if (!pskb_pull(list, eat)) {
858 if (clone)
859 kfree_skb(clone);
860 return NULL;
861 }
862 break;
863 }
864 } while (eat);
865
866 /* Free pulled out fragments. */
867 while ((list = skb_shinfo(skb)->frag_list) != insp) {
868 skb_shinfo(skb)->frag_list = list->next;
869 kfree_skb(list);
870 }
871 /* And insert new clone at head. */
872 if (clone) {
873 clone->next = list;
874 skb_shinfo(skb)->frag_list = clone;
875 }
876 }
877 /* Success! Now we may commit changes to skb data. */
878
879pull_pages:
880 eat = delta;
881 k = 0;
882 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
883 if (skb_shinfo(skb)->frags[i].size <= eat) {
884 put_page(skb_shinfo(skb)->frags[i].page);
885 eat -= skb_shinfo(skb)->frags[i].size;
886 } else {
887 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
888 if (eat) {
889 skb_shinfo(skb)->frags[k].page_offset += eat;
890 skb_shinfo(skb)->frags[k].size -= eat;
891 eat = 0;
892 }
893 k++;
894 }
895 }
896 skb_shinfo(skb)->nr_frags = k;
897
898 skb->tail += delta;
899 skb->data_len -= delta;
900
901 return skb->tail;
902}
903
904/* Copy some data bits from skb to kernel buffer. */
905
906int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
907{
908 int i, copy;
909 int start = skb_headlen(skb);
910
911 if (offset > (int)skb->len - len)
912 goto fault;
913
914 /* Copy header. */
915 if ((copy = start - offset) > 0) {
916 if (copy > len)
917 copy = len;
918 memcpy(to, skb->data + offset, copy);
919 if ((len -= copy) == 0)
920 return 0;
921 offset += copy;
922 to += copy;
923 }
924
925 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
926 int end;
927
928 BUG_TRAP(start <= offset + len);
929
930 end = start + skb_shinfo(skb)->frags[i].size;
931 if ((copy = end - offset) > 0) {
932 u8 *vaddr;
933
934 if (copy > len)
935 copy = len;
936
937 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
938 memcpy(to,
939 vaddr + skb_shinfo(skb)->frags[i].page_offset+
940 offset - start, copy);
941 kunmap_skb_frag(vaddr);
942
943 if ((len -= copy) == 0)
944 return 0;
945 offset += copy;
946 to += copy;
947 }
948 start = end;
949 }
950
951 if (skb_shinfo(skb)->frag_list) {
952 struct sk_buff *list = skb_shinfo(skb)->frag_list;
953
954 for (; list; list = list->next) {
955 int end;
956
957 BUG_TRAP(start <= offset + len);
958
959 end = start + list->len;
960 if ((copy = end - offset) > 0) {
961 if (copy > len)
962 copy = len;
963 if (skb_copy_bits(list, offset - start,
964 to, copy))
965 goto fault;
966 if ((len -= copy) == 0)
967 return 0;
968 offset += copy;
969 to += copy;
970 }
971 start = end;
972 }
973 }
974 if (!len)
975 return 0;
976
977fault:
978 return -EFAULT;
979}
980
Herbert Xu357b40a2005-04-19 22:30:14 -0700981/**
982 * skb_store_bits - store bits from kernel buffer to skb
983 * @skb: destination buffer
984 * @offset: offset in destination
985 * @from: source buffer
986 * @len: number of bytes to copy
987 *
988 * Copy the specified number of bytes from the source buffer to the
989 * destination skb. This function handles all the messy bits of
990 * traversing fragment lists and such.
991 */
992
993int skb_store_bits(const struct sk_buff *skb, int offset, void *from, int len)
994{
995 int i, copy;
996 int start = skb_headlen(skb);
997
998 if (offset > (int)skb->len - len)
999 goto fault;
1000
1001 if ((copy = start - offset) > 0) {
1002 if (copy > len)
1003 copy = len;
1004 memcpy(skb->data + offset, from, copy);
1005 if ((len -= copy) == 0)
1006 return 0;
1007 offset += copy;
1008 from += copy;
1009 }
1010
1011 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1012 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1013 int end;
1014
1015 BUG_TRAP(start <= offset + len);
1016
1017 end = start + frag->size;
1018 if ((copy = end - offset) > 0) {
1019 u8 *vaddr;
1020
1021 if (copy > len)
1022 copy = len;
1023
1024 vaddr = kmap_skb_frag(frag);
1025 memcpy(vaddr + frag->page_offset + offset - start,
1026 from, copy);
1027 kunmap_skb_frag(vaddr);
1028
1029 if ((len -= copy) == 0)
1030 return 0;
1031 offset += copy;
1032 from += copy;
1033 }
1034 start = end;
1035 }
1036
1037 if (skb_shinfo(skb)->frag_list) {
1038 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1039
1040 for (; list; list = list->next) {
1041 int end;
1042
1043 BUG_TRAP(start <= offset + len);
1044
1045 end = start + list->len;
1046 if ((copy = end - offset) > 0) {
1047 if (copy > len)
1048 copy = len;
1049 if (skb_store_bits(list, offset - start,
1050 from, copy))
1051 goto fault;
1052 if ((len -= copy) == 0)
1053 return 0;
1054 offset += copy;
1055 from += copy;
1056 }
1057 start = end;
1058 }
1059 }
1060 if (!len)
1061 return 0;
1062
1063fault:
1064 return -EFAULT;
1065}
1066
1067EXPORT_SYMBOL(skb_store_bits);
1068
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069/* Checksum skb data. */
1070
1071unsigned int skb_checksum(const struct sk_buff *skb, int offset,
1072 int len, unsigned int csum)
1073{
1074 int start = skb_headlen(skb);
1075 int i, copy = start - offset;
1076 int pos = 0;
1077
1078 /* Checksum header. */
1079 if (copy > 0) {
1080 if (copy > len)
1081 copy = len;
1082 csum = csum_partial(skb->data + offset, copy, csum);
1083 if ((len -= copy) == 0)
1084 return csum;
1085 offset += copy;
1086 pos = copy;
1087 }
1088
1089 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1090 int end;
1091
1092 BUG_TRAP(start <= offset + len);
1093
1094 end = start + skb_shinfo(skb)->frags[i].size;
1095 if ((copy = end - offset) > 0) {
1096 unsigned int csum2;
1097 u8 *vaddr;
1098 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1099
1100 if (copy > len)
1101 copy = len;
1102 vaddr = kmap_skb_frag(frag);
1103 csum2 = csum_partial(vaddr + frag->page_offset +
1104 offset - start, copy, 0);
1105 kunmap_skb_frag(vaddr);
1106 csum = csum_block_add(csum, csum2, pos);
1107 if (!(len -= copy))
1108 return csum;
1109 offset += copy;
1110 pos += copy;
1111 }
1112 start = end;
1113 }
1114
1115 if (skb_shinfo(skb)->frag_list) {
1116 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1117
1118 for (; list; list = list->next) {
1119 int end;
1120
1121 BUG_TRAP(start <= offset + len);
1122
1123 end = start + list->len;
1124 if ((copy = end - offset) > 0) {
1125 unsigned int csum2;
1126 if (copy > len)
1127 copy = len;
1128 csum2 = skb_checksum(list, offset - start,
1129 copy, 0);
1130 csum = csum_block_add(csum, csum2, pos);
1131 if ((len -= copy) == 0)
1132 return csum;
1133 offset += copy;
1134 pos += copy;
1135 }
1136 start = end;
1137 }
1138 }
1139 if (len)
1140 BUG();
1141
1142 return csum;
1143}
1144
1145/* Both of above in one bottle. */
1146
1147unsigned int skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1148 u8 *to, int len, unsigned int csum)
1149{
1150 int start = skb_headlen(skb);
1151 int i, copy = start - offset;
1152 int pos = 0;
1153
1154 /* Copy header. */
1155 if (copy > 0) {
1156 if (copy > len)
1157 copy = len;
1158 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1159 copy, csum);
1160 if ((len -= copy) == 0)
1161 return csum;
1162 offset += copy;
1163 to += copy;
1164 pos = copy;
1165 }
1166
1167 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1168 int end;
1169
1170 BUG_TRAP(start <= offset + len);
1171
1172 end = start + skb_shinfo(skb)->frags[i].size;
1173 if ((copy = end - offset) > 0) {
1174 unsigned int csum2;
1175 u8 *vaddr;
1176 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1177
1178 if (copy > len)
1179 copy = len;
1180 vaddr = kmap_skb_frag(frag);
1181 csum2 = csum_partial_copy_nocheck(vaddr +
1182 frag->page_offset +
1183 offset - start, to,
1184 copy, 0);
1185 kunmap_skb_frag(vaddr);
1186 csum = csum_block_add(csum, csum2, pos);
1187 if (!(len -= copy))
1188 return csum;
1189 offset += copy;
1190 to += copy;
1191 pos += copy;
1192 }
1193 start = end;
1194 }
1195
1196 if (skb_shinfo(skb)->frag_list) {
1197 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1198
1199 for (; list; list = list->next) {
1200 unsigned int csum2;
1201 int end;
1202
1203 BUG_TRAP(start <= offset + len);
1204
1205 end = start + list->len;
1206 if ((copy = end - offset) > 0) {
1207 if (copy > len)
1208 copy = len;
1209 csum2 = skb_copy_and_csum_bits(list,
1210 offset - start,
1211 to, copy, 0);
1212 csum = csum_block_add(csum, csum2, pos);
1213 if ((len -= copy) == 0)
1214 return csum;
1215 offset += copy;
1216 to += copy;
1217 pos += copy;
1218 }
1219 start = end;
1220 }
1221 }
1222 if (len)
1223 BUG();
1224 return csum;
1225}
1226
1227void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1228{
1229 unsigned int csum;
1230 long csstart;
1231
1232 if (skb->ip_summed == CHECKSUM_HW)
1233 csstart = skb->h.raw - skb->data;
1234 else
1235 csstart = skb_headlen(skb);
1236
1237 if (csstart > skb_headlen(skb))
1238 BUG();
1239
1240 memcpy(to, skb->data, csstart);
1241
1242 csum = 0;
1243 if (csstart != skb->len)
1244 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1245 skb->len - csstart, 0);
1246
1247 if (skb->ip_summed == CHECKSUM_HW) {
1248 long csstuff = csstart + skb->csum;
1249
1250 *((unsigned short *)(to + csstuff)) = csum_fold(csum);
1251 }
1252}
1253
1254/**
1255 * skb_dequeue - remove from the head of the queue
1256 * @list: list to dequeue from
1257 *
1258 * Remove the head of the list. The list lock is taken so the function
1259 * may be used safely with other locking list functions. The head item is
1260 * returned or %NULL if the list is empty.
1261 */
1262
1263struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1264{
1265 unsigned long flags;
1266 struct sk_buff *result;
1267
1268 spin_lock_irqsave(&list->lock, flags);
1269 result = __skb_dequeue(list);
1270 spin_unlock_irqrestore(&list->lock, flags);
1271 return result;
1272}
1273
1274/**
1275 * skb_dequeue_tail - remove from the tail of the queue
1276 * @list: list to dequeue from
1277 *
1278 * Remove the tail of the list. The list lock is taken so the function
1279 * may be used safely with other locking list functions. The tail item is
1280 * returned or %NULL if the list is empty.
1281 */
1282struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1283{
1284 unsigned long flags;
1285 struct sk_buff *result;
1286
1287 spin_lock_irqsave(&list->lock, flags);
1288 result = __skb_dequeue_tail(list);
1289 spin_unlock_irqrestore(&list->lock, flags);
1290 return result;
1291}
1292
1293/**
1294 * skb_queue_purge - empty a list
1295 * @list: list to empty
1296 *
1297 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1298 * the list and one reference dropped. This function takes the list
1299 * lock and is atomic with respect to other list locking functions.
1300 */
1301void skb_queue_purge(struct sk_buff_head *list)
1302{
1303 struct sk_buff *skb;
1304 while ((skb = skb_dequeue(list)) != NULL)
1305 kfree_skb(skb);
1306}
1307
1308/**
1309 * skb_queue_head - queue a buffer at the list head
1310 * @list: list to use
1311 * @newsk: buffer to queue
1312 *
1313 * Queue a buffer at the start of the list. This function takes the
1314 * list lock and can be used safely with other locking &sk_buff functions
1315 * safely.
1316 *
1317 * A buffer cannot be placed on two lists at the same time.
1318 */
1319void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1320{
1321 unsigned long flags;
1322
1323 spin_lock_irqsave(&list->lock, flags);
1324 __skb_queue_head(list, newsk);
1325 spin_unlock_irqrestore(&list->lock, flags);
1326}
1327
1328/**
1329 * skb_queue_tail - queue a buffer at the list tail
1330 * @list: list to use
1331 * @newsk: buffer to queue
1332 *
1333 * Queue a buffer at the tail of the list. This function takes the
1334 * list lock and can be used safely with other locking &sk_buff functions
1335 * safely.
1336 *
1337 * A buffer cannot be placed on two lists at the same time.
1338 */
1339void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1340{
1341 unsigned long flags;
1342
1343 spin_lock_irqsave(&list->lock, flags);
1344 __skb_queue_tail(list, newsk);
1345 spin_unlock_irqrestore(&list->lock, flags);
1346}
1347/**
1348 * skb_unlink - remove a buffer from a list
1349 * @skb: buffer to remove
1350 *
1351 * Place a packet after a given packet in a list. The list locks are taken
1352 * and this function is atomic with respect to other list locked calls
1353 *
1354 * Works even without knowing the list it is sitting on, which can be
1355 * handy at times. It also means that THE LIST MUST EXIST when you
1356 * unlink. Thus a list must have its contents unlinked before it is
1357 * destroyed.
1358 */
1359void skb_unlink(struct sk_buff *skb)
1360{
1361 struct sk_buff_head *list = skb->list;
1362
1363 if (list) {
1364 unsigned long flags;
1365
1366 spin_lock_irqsave(&list->lock, flags);
1367 if (skb->list == list)
1368 __skb_unlink(skb, skb->list);
1369 spin_unlock_irqrestore(&list->lock, flags);
1370 }
1371}
1372
1373
1374/**
1375 * skb_append - append a buffer
1376 * @old: buffer to insert after
1377 * @newsk: buffer to insert
1378 *
1379 * Place a packet after a given packet in a list. The list locks are taken
1380 * and this function is atomic with respect to other list locked calls.
1381 * A buffer cannot be placed on two lists at the same time.
1382 */
1383
1384void skb_append(struct sk_buff *old, struct sk_buff *newsk)
1385{
1386 unsigned long flags;
1387
1388 spin_lock_irqsave(&old->list->lock, flags);
1389 __skb_append(old, newsk);
1390 spin_unlock_irqrestore(&old->list->lock, flags);
1391}
1392
1393
1394/**
1395 * skb_insert - insert a buffer
1396 * @old: buffer to insert before
1397 * @newsk: buffer to insert
1398 *
1399 * Place a packet before a given packet in a list. The list locks are taken
1400 * and this function is atomic with respect to other list locked calls
1401 * A buffer cannot be placed on two lists at the same time.
1402 */
1403
1404void skb_insert(struct sk_buff *old, struct sk_buff *newsk)
1405{
1406 unsigned long flags;
1407
1408 spin_lock_irqsave(&old->list->lock, flags);
1409 __skb_insert(newsk, old->prev, old, old->list);
1410 spin_unlock_irqrestore(&old->list->lock, flags);
1411}
1412
1413#if 0
1414/*
1415 * Tune the memory allocator for a new MTU size.
1416 */
1417void skb_add_mtu(int mtu)
1418{
1419 /* Must match allocation in alloc_skb */
1420 mtu = SKB_DATA_ALIGN(mtu) + sizeof(struct skb_shared_info);
1421
1422 kmem_add_cache_size(mtu);
1423}
1424#endif
1425
1426static inline void skb_split_inside_header(struct sk_buff *skb,
1427 struct sk_buff* skb1,
1428 const u32 len, const int pos)
1429{
1430 int i;
1431
1432 memcpy(skb_put(skb1, pos - len), skb->data + len, pos - len);
1433
1434 /* And move data appendix as is. */
1435 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1436 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
1437
1438 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
1439 skb_shinfo(skb)->nr_frags = 0;
1440 skb1->data_len = skb->data_len;
1441 skb1->len += skb1->data_len;
1442 skb->data_len = 0;
1443 skb->len = len;
1444 skb->tail = skb->data + len;
1445}
1446
1447static inline void skb_split_no_header(struct sk_buff *skb,
1448 struct sk_buff* skb1,
1449 const u32 len, int pos)
1450{
1451 int i, k = 0;
1452 const int nfrags = skb_shinfo(skb)->nr_frags;
1453
1454 skb_shinfo(skb)->nr_frags = 0;
1455 skb1->len = skb1->data_len = skb->len - len;
1456 skb->len = len;
1457 skb->data_len = len - pos;
1458
1459 for (i = 0; i < nfrags; i++) {
1460 int size = skb_shinfo(skb)->frags[i].size;
1461
1462 if (pos + size > len) {
1463 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
1464
1465 if (pos < len) {
1466 /* Split frag.
1467 * We have two variants in this case:
1468 * 1. Move all the frag to the second
1469 * part, if it is possible. F.e.
1470 * this approach is mandatory for TUX,
1471 * where splitting is expensive.
1472 * 2. Split is accurately. We make this.
1473 */
1474 get_page(skb_shinfo(skb)->frags[i].page);
1475 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
1476 skb_shinfo(skb1)->frags[0].size -= len - pos;
1477 skb_shinfo(skb)->frags[i].size = len - pos;
1478 skb_shinfo(skb)->nr_frags++;
1479 }
1480 k++;
1481 } else
1482 skb_shinfo(skb)->nr_frags++;
1483 pos += size;
1484 }
1485 skb_shinfo(skb1)->nr_frags = k;
1486}
1487
1488/**
1489 * skb_split - Split fragmented skb to two parts at length len.
1490 * @skb: the buffer to split
1491 * @skb1: the buffer to receive the second part
1492 * @len: new length for skb
1493 */
1494void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
1495{
1496 int pos = skb_headlen(skb);
1497
1498 if (len < pos) /* Split line is inside header. */
1499 skb_split_inside_header(skb, skb1, len, pos);
1500 else /* Second chunk has no header, nothing to copy. */
1501 skb_split_no_header(skb, skb1, len, pos);
1502}
1503
Thomas Graf677e90e2005-06-23 20:59:51 -07001504/**
1505 * skb_prepare_seq_read - Prepare a sequential read of skb data
1506 * @skb: the buffer to read
1507 * @from: lower offset of data to be read
1508 * @to: upper offset of data to be read
1509 * @st: state variable
1510 *
1511 * Initializes the specified state variable. Must be called before
1512 * invoking skb_seq_read() for the first time.
1513 */
1514void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1515 unsigned int to, struct skb_seq_state *st)
1516{
1517 st->lower_offset = from;
1518 st->upper_offset = to;
1519 st->root_skb = st->cur_skb = skb;
1520 st->frag_idx = st->stepped_offset = 0;
1521 st->frag_data = NULL;
1522}
1523
1524/**
1525 * skb_seq_read - Sequentially read skb data
1526 * @consumed: number of bytes consumed by the caller so far
1527 * @data: destination pointer for data to be returned
1528 * @st: state variable
1529 *
1530 * Reads a block of skb data at &consumed relative to the
1531 * lower offset specified to skb_prepare_seq_read(). Assigns
1532 * the head of the data block to &data and returns the length
1533 * of the block or 0 if the end of the skb data or the upper
1534 * offset has been reached.
1535 *
1536 * The caller is not required to consume all of the data
1537 * returned, i.e. &consumed is typically set to the number
1538 * of bytes already consumed and the next call to
1539 * skb_seq_read() will return the remaining part of the block.
1540 *
1541 * Note: The size of each block of data returned can be arbitary,
1542 * this limitation is the cost for zerocopy seqeuental
1543 * reads of potentially non linear data.
1544 *
1545 * Note: Fragment lists within fragments are not implemented
1546 * at the moment, state->root_skb could be replaced with
1547 * a stack for this purpose.
1548 */
1549unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1550 struct skb_seq_state *st)
1551{
1552 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
1553 skb_frag_t *frag;
1554
1555 if (unlikely(abs_offset >= st->upper_offset))
1556 return 0;
1557
1558next_skb:
1559 block_limit = skb_headlen(st->cur_skb);
1560
1561 if (abs_offset < block_limit) {
1562 *data = st->cur_skb->data + abs_offset;
1563 return block_limit - abs_offset;
1564 }
1565
1566 if (st->frag_idx == 0 && !st->frag_data)
1567 st->stepped_offset += skb_headlen(st->cur_skb);
1568
1569 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
1570 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
1571 block_limit = frag->size + st->stepped_offset;
1572
1573 if (abs_offset < block_limit) {
1574 if (!st->frag_data)
1575 st->frag_data = kmap_skb_frag(frag);
1576
1577 *data = (u8 *) st->frag_data + frag->page_offset +
1578 (abs_offset - st->stepped_offset);
1579
1580 return block_limit - abs_offset;
1581 }
1582
1583 if (st->frag_data) {
1584 kunmap_skb_frag(st->frag_data);
1585 st->frag_data = NULL;
1586 }
1587
1588 st->frag_idx++;
1589 st->stepped_offset += frag->size;
1590 }
1591
1592 if (st->cur_skb->next) {
1593 st->cur_skb = st->cur_skb->next;
1594 st->frag_idx = 0;
1595 goto next_skb;
1596 } else if (st->root_skb == st->cur_skb &&
1597 skb_shinfo(st->root_skb)->frag_list) {
1598 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
1599 goto next_skb;
1600 }
1601
1602 return 0;
1603}
1604
1605/**
1606 * skb_abort_seq_read - Abort a sequential read of skb data
1607 * @st: state variable
1608 *
1609 * Must be called if skb_seq_read() was not called until it
1610 * returned 0.
1611 */
1612void skb_abort_seq_read(struct skb_seq_state *st)
1613{
1614 if (st->frag_data)
1615 kunmap_skb_frag(st->frag_data);
1616}
1617
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07001618#define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
1619
1620static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
1621 struct ts_config *conf,
1622 struct ts_state *state)
1623{
1624 return skb_seq_read(offset, text, TS_SKB_CB(state));
1625}
1626
1627static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
1628{
1629 skb_abort_seq_read(TS_SKB_CB(state));
1630}
1631
1632/**
1633 * skb_find_text - Find a text pattern in skb data
1634 * @skb: the buffer to look in
1635 * @from: search offset
1636 * @to: search limit
1637 * @config: textsearch configuration
1638 * @state: uninitialized textsearch state variable
1639 *
1640 * Finds a pattern in the skb data according to the specified
1641 * textsearch configuration. Use textsearch_next() to retrieve
1642 * subsequent occurrences of the pattern. Returns the offset
1643 * to the first occurrence or UINT_MAX if no match was found.
1644 */
1645unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1646 unsigned int to, struct ts_config *config,
1647 struct ts_state *state)
1648{
1649 config->get_next_block = skb_ts_get_next_block;
1650 config->finish = skb_ts_finish;
1651
1652 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
1653
1654 return textsearch_find(config, state);
1655}
1656
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657void __init skb_init(void)
1658{
1659 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
1660 sizeof(struct sk_buff),
1661 0,
1662 SLAB_HWCACHE_ALIGN,
1663 NULL, NULL);
1664 if (!skbuff_head_cache)
1665 panic("cannot create skbuff cache");
1666}
1667
1668EXPORT_SYMBOL(___pskb_trim);
1669EXPORT_SYMBOL(__kfree_skb);
1670EXPORT_SYMBOL(__pskb_pull_tail);
1671EXPORT_SYMBOL(alloc_skb);
1672EXPORT_SYMBOL(pskb_copy);
1673EXPORT_SYMBOL(pskb_expand_head);
1674EXPORT_SYMBOL(skb_checksum);
1675EXPORT_SYMBOL(skb_clone);
1676EXPORT_SYMBOL(skb_clone_fraglist);
1677EXPORT_SYMBOL(skb_copy);
1678EXPORT_SYMBOL(skb_copy_and_csum_bits);
1679EXPORT_SYMBOL(skb_copy_and_csum_dev);
1680EXPORT_SYMBOL(skb_copy_bits);
1681EXPORT_SYMBOL(skb_copy_expand);
1682EXPORT_SYMBOL(skb_over_panic);
1683EXPORT_SYMBOL(skb_pad);
1684EXPORT_SYMBOL(skb_realloc_headroom);
1685EXPORT_SYMBOL(skb_under_panic);
1686EXPORT_SYMBOL(skb_dequeue);
1687EXPORT_SYMBOL(skb_dequeue_tail);
1688EXPORT_SYMBOL(skb_insert);
1689EXPORT_SYMBOL(skb_queue_purge);
1690EXPORT_SYMBOL(skb_queue_head);
1691EXPORT_SYMBOL(skb_queue_tail);
1692EXPORT_SYMBOL(skb_unlink);
1693EXPORT_SYMBOL(skb_append);
1694EXPORT_SYMBOL(skb_split);
Thomas Graf677e90e2005-06-23 20:59:51 -07001695EXPORT_SYMBOL(skb_prepare_seq_read);
1696EXPORT_SYMBOL(skb_seq_read);
1697EXPORT_SYMBOL(skb_abort_seq_read);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07001698EXPORT_SYMBOL(skb_find_text);