blob: be9b541f536f8a9de759b6fe266cd5ad8b207499 [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
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include <linux/module.h>
42#include <linux/types.h>
43#include <linux/kernel.h>
44#include <linux/sched.h>
45#include <linux/mm.h>
46#include <linux/interrupt.h>
47#include <linux/in.h>
48#include <linux/inet.h>
49#include <linux/slab.h>
50#include <linux/netdevice.h>
51#ifdef CONFIG_NET_CLS_ACT
52#include <net/pkt_sched.h>
53#endif
54#include <linux/string.h>
55#include <linux/skbuff.h>
56#include <linux/cache.h>
57#include <linux/rtnetlink.h>
58#include <linux/init.h>
59#include <linux/highmem.h>
60
61#include <net/protocol.h>
62#include <net/dst.h>
63#include <net/sock.h>
64#include <net/checksum.h>
65#include <net/xfrm.h>
66
67#include <asm/uaccess.h>
68#include <asm/system.h>
69
Eric Dumazetba899662005-08-26 12:05:31 -070070static kmem_cache_t *skbuff_head_cache __read_mostly;
71static kmem_cache_t *skbuff_fclone_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
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
David S. Millerdc6de332006-04-20 00:10:50 -0700114void skb_truesize_bug(struct sk_buff *skb)
115{
116 printk(KERN_ERR "SKB BUG: Invalid truesize (%u) "
117 "len=%u, sizeof(sk_buff)=%Zd\n",
118 skb->truesize, skb->len, sizeof(struct sk_buff));
119}
120EXPORT_SYMBOL(skb_truesize_bug);
121
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122/* Allocate a new skbuff. We do this ourselves so we can fill in a few
123 * 'private' fields and also do memory statistics to find all the
124 * [BEEP] leaks.
125 *
126 */
127
128/**
David S. Millerd179cd12005-08-17 14:57:30 -0700129 * __alloc_skb - allocate a network buffer
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 * @size: size to allocate
131 * @gfp_mask: allocation mask
Randy Dunlapc83c2482005-10-18 22:07:41 -0700132 * @fclone: allocate from fclone cache instead of head cache
133 * and allocate a cloned (child) skb
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 *
135 * Allocate a new &sk_buff. The returned buffer has no headroom and a
136 * tail room of size bytes. The object has a reference count of one.
137 * The return is the buffer. On a failure the return is %NULL.
138 *
139 * Buffers may only be allocated from interrupts using a @gfp_mask of
140 * %GFP_ATOMIC.
141 */
Al Virodd0fc662005-10-07 07:46:04 +0100142struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
David S. Millerd179cd12005-08-17 14:57:30 -0700143 int fclone)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144{
Herbert Xu8798b3f2006-01-23 16:32:45 -0800145 kmem_cache_t *cache;
Benjamin LaHaise4947d3e2006-01-03 14:06:50 -0800146 struct skb_shared_info *shinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 struct sk_buff *skb;
148 u8 *data;
149
Herbert Xu8798b3f2006-01-23 16:32:45 -0800150 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
151
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 /* Get the HEAD */
Herbert Xu8798b3f2006-01-23 16:32:45 -0800153 skb = kmem_cache_alloc(cache, gfp_mask & ~__GFP_DMA);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 if (!skb)
155 goto out;
156
157 /* Get the DATA. Size must match skb_add_mtu(). */
158 size = SKB_DATA_ALIGN(size);
Christoph Hellwig1d2c8ee2006-10-04 02:15:25 -0700159 data = kmalloc_track_caller(size + sizeof(struct skb_shared_info),
160 gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 if (!data)
162 goto nodata;
163
164 memset(skb, 0, offsetof(struct sk_buff, truesize));
165 skb->truesize = size + sizeof(struct sk_buff);
166 atomic_set(&skb->users, 1);
167 skb->head = data;
168 skb->data = data;
169 skb->tail = data;
170 skb->end = data + size;
Benjamin LaHaise4947d3e2006-01-03 14:06:50 -0800171 /* make sure we initialize shinfo sequentially */
172 shinfo = skb_shinfo(skb);
173 atomic_set(&shinfo->dataref, 1);
174 shinfo->nr_frags = 0;
Herbert Xu79671682006-06-22 02:40:14 -0700175 shinfo->gso_size = 0;
176 shinfo->gso_segs = 0;
177 shinfo->gso_type = 0;
Benjamin LaHaise4947d3e2006-01-03 14:06:50 -0800178 shinfo->ip6_frag_id = 0;
179 shinfo->frag_list = NULL;
180
David S. Millerd179cd12005-08-17 14:57:30 -0700181 if (fclone) {
182 struct sk_buff *child = skb + 1;
183 atomic_t *fclone_ref = (atomic_t *) (child + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184
David S. Millerd179cd12005-08-17 14:57:30 -0700185 skb->fclone = SKB_FCLONE_ORIG;
186 atomic_set(fclone_ref, 1);
187
188 child->fclone = SKB_FCLONE_UNAVAILABLE;
189 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190out:
191 return skb;
192nodata:
Herbert Xu8798b3f2006-01-23 16:32:45 -0800193 kmem_cache_free(cache, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 skb = NULL;
195 goto out;
196}
197
198/**
199 * alloc_skb_from_cache - allocate a network buffer
200 * @cp: kmem_cache from which to allocate the data area
201 * (object size must be big enough for @size bytes + skb overheads)
202 * @size: size to allocate
203 * @gfp_mask: allocation mask
204 *
205 * Allocate a new &sk_buff. The returned buffer has no headroom and
206 * tail room of size bytes. The object has a reference count of one.
207 * The return is the buffer. On a failure the return is %NULL.
208 *
209 * Buffers may only be allocated from interrupts using a @gfp_mask of
210 * %GFP_ATOMIC.
211 */
212struct sk_buff *alloc_skb_from_cache(kmem_cache_t *cp,
Victor Fusco86a76ca2005-07-08 14:57:47 -0700213 unsigned int size,
Al Virodd0fc662005-10-07 07:46:04 +0100214 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215{
216 struct sk_buff *skb;
217 u8 *data;
218
219 /* Get the HEAD */
220 skb = kmem_cache_alloc(skbuff_head_cache,
221 gfp_mask & ~__GFP_DMA);
222 if (!skb)
223 goto out;
224
225 /* Get the DATA. */
226 size = SKB_DATA_ALIGN(size);
227 data = kmem_cache_alloc(cp, gfp_mask);
228 if (!data)
229 goto nodata;
230
231 memset(skb, 0, offsetof(struct sk_buff, truesize));
232 skb->truesize = size + sizeof(struct sk_buff);
233 atomic_set(&skb->users, 1);
234 skb->head = data;
235 skb->data = data;
236 skb->tail = data;
237 skb->end = data + size;
238
239 atomic_set(&(skb_shinfo(skb)->dataref), 1);
240 skb_shinfo(skb)->nr_frags = 0;
Herbert Xu79671682006-06-22 02:40:14 -0700241 skb_shinfo(skb)->gso_size = 0;
242 skb_shinfo(skb)->gso_segs = 0;
243 skb_shinfo(skb)->gso_type = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 skb_shinfo(skb)->frag_list = NULL;
245out:
246 return skb;
247nodata:
248 kmem_cache_free(skbuff_head_cache, skb);
249 skb = NULL;
250 goto out;
251}
252
Christoph Hellwig8af27452006-07-31 22:35:23 -0700253/**
254 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
255 * @dev: network device to receive on
256 * @length: length to allocate
257 * @gfp_mask: get_free_pages mask, passed to alloc_skb
258 *
259 * Allocate a new &sk_buff and assign it a usage count of one. The
260 * buffer has unspecified headroom built in. Users should allocate
261 * the headroom they think they need without accounting for the
262 * built in space. The built in space is used for optimisations.
263 *
264 * %NULL is returned if there is no free memory.
265 */
266struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
267 unsigned int length, gfp_t gfp_mask)
268{
269 struct sk_buff *skb;
270
271 skb = alloc_skb(length + NET_SKB_PAD, gfp_mask);
Christoph Hellwig7b2e4972006-08-07 16:09:04 -0700272 if (likely(skb)) {
Christoph Hellwig8af27452006-07-31 22:35:23 -0700273 skb_reserve(skb, NET_SKB_PAD);
Christoph Hellwig7b2e4972006-08-07 16:09:04 -0700274 skb->dev = dev;
275 }
Christoph Hellwig8af27452006-07-31 22:35:23 -0700276 return skb;
277}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278
Herbert Xu27b437c2006-07-13 19:26:39 -0700279static void skb_drop_list(struct sk_buff **listp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280{
Herbert Xu27b437c2006-07-13 19:26:39 -0700281 struct sk_buff *list = *listp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282
Herbert Xu27b437c2006-07-13 19:26:39 -0700283 *listp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284
285 do {
286 struct sk_buff *this = list;
287 list = list->next;
288 kfree_skb(this);
289 } while (list);
290}
291
Herbert Xu27b437c2006-07-13 19:26:39 -0700292static inline void skb_drop_fraglist(struct sk_buff *skb)
293{
294 skb_drop_list(&skb_shinfo(skb)->frag_list);
295}
296
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297static void skb_clone_fraglist(struct sk_buff *skb)
298{
299 struct sk_buff *list;
300
301 for (list = skb_shinfo(skb)->frag_list; list; list = list->next)
302 skb_get(list);
303}
304
Adrian Bunk5bba1712006-06-29 13:02:35 -0700305static void skb_release_data(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306{
307 if (!skb->cloned ||
308 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
309 &skb_shinfo(skb)->dataref)) {
310 if (skb_shinfo(skb)->nr_frags) {
311 int i;
312 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
313 put_page(skb_shinfo(skb)->frags[i].page);
314 }
315
316 if (skb_shinfo(skb)->frag_list)
317 skb_drop_fraglist(skb);
318
319 kfree(skb->head);
320 }
321}
322
323/*
324 * Free an skbuff by memory without cleaning the state.
325 */
326void kfree_skbmem(struct sk_buff *skb)
327{
David S. Millerd179cd12005-08-17 14:57:30 -0700328 struct sk_buff *other;
329 atomic_t *fclone_ref;
330
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 skb_release_data(skb);
David S. Millerd179cd12005-08-17 14:57:30 -0700332 switch (skb->fclone) {
333 case SKB_FCLONE_UNAVAILABLE:
334 kmem_cache_free(skbuff_head_cache, skb);
335 break;
336
337 case SKB_FCLONE_ORIG:
338 fclone_ref = (atomic_t *) (skb + 2);
339 if (atomic_dec_and_test(fclone_ref))
340 kmem_cache_free(skbuff_fclone_cache, skb);
341 break;
342
343 case SKB_FCLONE_CLONE:
344 fclone_ref = (atomic_t *) (skb + 1);
345 other = skb - 1;
346
347 /* The clone portion is available for
348 * fast-cloning again.
349 */
350 skb->fclone = SKB_FCLONE_UNAVAILABLE;
351
352 if (atomic_dec_and_test(fclone_ref))
353 kmem_cache_free(skbuff_fclone_cache, other);
354 break;
355 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356}
357
358/**
359 * __kfree_skb - private function
360 * @skb: buffer
361 *
362 * Free an sk_buff. Release anything attached to the buffer.
363 * Clean the state. This is an internal helper function. Users should
364 * always call kfree_skb
365 */
366
367void __kfree_skb(struct sk_buff *skb)
368{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 dst_release(skb->dst);
370#ifdef CONFIG_XFRM
371 secpath_put(skb->sp);
372#endif
Stephen Hemminger9c2b3322005-04-19 22:39:42 -0700373 if (skb->destructor) {
374 WARN_ON(in_irq());
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 skb->destructor(skb);
376 }
377#ifdef CONFIG_NETFILTER
378 nf_conntrack_put(skb->nfct);
Yasuyuki Kozakai9fb9cbb2005-11-09 16:38:16 -0800379#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
380 nf_conntrack_put_reasm(skb->nfct_reasm);
381#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382#ifdef CONFIG_BRIDGE_NETFILTER
383 nf_bridge_put(skb->nf_bridge);
384#endif
385#endif
386/* XXX: IS this still necessary? - JHS */
387#ifdef CONFIG_NET_SCHED
388 skb->tc_index = 0;
389#ifdef CONFIG_NET_CLS_ACT
390 skb->tc_verd = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391#endif
392#endif
393
394 kfree_skbmem(skb);
395}
396
397/**
Jörn Engel231d06a2006-03-20 21:28:35 -0800398 * kfree_skb - free an sk_buff
399 * @skb: buffer to free
400 *
401 * Drop a reference to the buffer and free it if the usage count has
402 * hit zero.
403 */
404void kfree_skb(struct sk_buff *skb)
405{
406 if (unlikely(!skb))
407 return;
408 if (likely(atomic_read(&skb->users) == 1))
409 smp_rmb();
410 else if (likely(!atomic_dec_and_test(&skb->users)))
411 return;
412 __kfree_skb(skb);
413}
414
415/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 * skb_clone - duplicate an sk_buff
417 * @skb: buffer to clone
418 * @gfp_mask: allocation priority
419 *
420 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
421 * copies share the same packet data but not structure. The new
422 * buffer has a reference count of 1. If the allocation fails the
423 * function returns %NULL otherwise the new buffer is returned.
424 *
425 * If this function is called from an interrupt gfp_mask() must be
426 * %GFP_ATOMIC.
427 */
428
Al Virodd0fc662005-10-07 07:46:04 +0100429struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430{
David S. Millerd179cd12005-08-17 14:57:30 -0700431 struct sk_buff *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432
David S. Millerd179cd12005-08-17 14:57:30 -0700433 n = skb + 1;
434 if (skb->fclone == SKB_FCLONE_ORIG &&
435 n->fclone == SKB_FCLONE_UNAVAILABLE) {
436 atomic_t *fclone_ref = (atomic_t *) (n + 1);
437 n->fclone = SKB_FCLONE_CLONE;
438 atomic_inc(fclone_ref);
439 } else {
440 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
441 if (!n)
442 return NULL;
443 n->fclone = SKB_FCLONE_UNAVAILABLE;
444 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445
446#define C(x) n->x = skb->x
447
448 n->next = n->prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 n->sk = NULL;
Patrick McHardya61bbcf2005-08-14 17:24:31 -0700450 C(tstamp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 C(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 C(h);
453 C(nh);
454 C(mac);
455 C(dst);
456 dst_clone(skb->dst);
457 C(sp);
458#ifdef CONFIG_INET
459 secpath_get(skb->sp);
460#endif
461 memcpy(n->cb, skb->cb, sizeof(skb->cb));
462 C(len);
463 C(data_len);
464 C(csum);
465 C(local_df);
466 n->cloned = 1;
467 n->nohdr = 0;
468 C(pkt_type);
469 C(ip_summed);
470 C(priority);
YOSHIFUJI Hideakia8372f02006-02-19 22:32:06 -0800471#if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
472 C(ipvs_property);
473#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 C(protocol);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 n->destructor = NULL;
Thomas Graf82e91ff2006-11-09 15:19:14 -0800476 C(mark);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477#ifdef CONFIG_NETFILTER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 C(nfct);
479 nf_conntrack_get(skb->nfct);
480 C(nfctinfo);
Yasuyuki Kozakai9fb9cbb2005-11-09 16:38:16 -0800481#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
482 C(nfct_reasm);
483 nf_conntrack_get_reasm(skb->nfct_reasm);
484#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485#ifdef CONFIG_BRIDGE_NETFILTER
486 C(nf_bridge);
487 nf_bridge_get(skb->nf_bridge);
488#endif
489#endif /*CONFIG_NETFILTER*/
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490#ifdef CONFIG_NET_SCHED
491 C(tc_index);
492#ifdef CONFIG_NET_CLS_ACT
493 n->tc_verd = SET_TC_VERD(skb->tc_verd,0);
David S. Millerb72f6ec2005-07-19 14:13:54 -0700494 n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd);
495 n->tc_verd = CLR_TC_MUNGED(n->tc_verd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 C(input_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497#endif
James Morris984bc162006-06-09 00:29:17 -0700498 skb_copy_secmark(n, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499#endif
500 C(truesize);
501 atomic_set(&n->users, 1);
502 C(head);
503 C(data);
504 C(tail);
505 C(end);
506
507 atomic_inc(&(skb_shinfo(skb)->dataref));
508 skb->cloned = 1;
509
510 return n;
511}
512
513static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
514{
515 /*
516 * Shift between the two data areas in bytes
517 */
518 unsigned long offset = new->data - old->data;
519
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 new->sk = NULL;
521 new->dev = old->dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 new->priority = old->priority;
523 new->protocol = old->protocol;
524 new->dst = dst_clone(old->dst);
525#ifdef CONFIG_INET
526 new->sp = secpath_get(old->sp);
527#endif
528 new->h.raw = old->h.raw + offset;
529 new->nh.raw = old->nh.raw + offset;
530 new->mac.raw = old->mac.raw + offset;
531 memcpy(new->cb, old->cb, sizeof(old->cb));
532 new->local_df = old->local_df;
David S. Millerd179cd12005-08-17 14:57:30 -0700533 new->fclone = SKB_FCLONE_UNAVAILABLE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 new->pkt_type = old->pkt_type;
Patrick McHardya61bbcf2005-08-14 17:24:31 -0700535 new->tstamp = old->tstamp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 new->destructor = NULL;
Thomas Graf82e91ff2006-11-09 15:19:14 -0800537 new->mark = old->mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538#ifdef CONFIG_NETFILTER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 new->nfct = old->nfct;
540 nf_conntrack_get(old->nfct);
541 new->nfctinfo = old->nfctinfo;
Yasuyuki Kozakai9fb9cbb2005-11-09 16:38:16 -0800542#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
543 new->nfct_reasm = old->nfct_reasm;
544 nf_conntrack_get_reasm(old->nfct_reasm);
545#endif
Julian Anastasovc98d80e2005-10-22 13:39:21 +0300546#if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
547 new->ipvs_property = old->ipvs_property;
548#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549#ifdef CONFIG_BRIDGE_NETFILTER
550 new->nf_bridge = old->nf_bridge;
551 nf_bridge_get(old->nf_bridge);
552#endif
553#endif
554#ifdef CONFIG_NET_SCHED
555#ifdef CONFIG_NET_CLS_ACT
556 new->tc_verd = old->tc_verd;
557#endif
558 new->tc_index = old->tc_index;
559#endif
James Morris984bc162006-06-09 00:29:17 -0700560 skb_copy_secmark(new, old);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 atomic_set(&new->users, 1);
Herbert Xu79671682006-06-22 02:40:14 -0700562 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
563 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
564 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565}
566
567/**
568 * skb_copy - create private copy of an sk_buff
569 * @skb: buffer to copy
570 * @gfp_mask: allocation priority
571 *
572 * Make a copy of both an &sk_buff and its data. This is used when the
573 * caller wishes to modify the data and needs a private copy of the
574 * data to alter. Returns %NULL on failure or the pointer to the buffer
575 * on success. The returned buffer has a reference count of 1.
576 *
577 * As by-product this function converts non-linear &sk_buff to linear
578 * one, so that &sk_buff becomes completely private and caller is allowed
579 * to modify all the data of returned buffer. This means that this
580 * function is not recommended for use in circumstances when only
581 * header is going to be modified. Use pskb_copy() instead.
582 */
583
Al Virodd0fc662005-10-07 07:46:04 +0100584struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585{
586 int headerlen = skb->data - skb->head;
587 /*
588 * Allocate the copy buffer
589 */
590 struct sk_buff *n = alloc_skb(skb->end - skb->head + skb->data_len,
591 gfp_mask);
592 if (!n)
593 return NULL;
594
595 /* Set the data pointer */
596 skb_reserve(n, headerlen);
597 /* Set the tail pointer and length */
598 skb_put(n, skb->len);
599 n->csum = skb->csum;
600 n->ip_summed = skb->ip_summed;
601
602 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
603 BUG();
604
605 copy_skb_header(n, skb);
606 return n;
607}
608
609
610/**
611 * pskb_copy - create copy of an sk_buff with private head.
612 * @skb: buffer to copy
613 * @gfp_mask: allocation priority
614 *
615 * Make a copy of both an &sk_buff and part of its data, located
616 * in header. Fragmented data remain shared. This is used when
617 * the caller wishes to modify only header of &sk_buff and needs
618 * private copy of the header to alter. Returns %NULL on failure
619 * or the pointer to the buffer on success.
620 * The returned buffer has a reference count of 1.
621 */
622
Al Virodd0fc662005-10-07 07:46:04 +0100623struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624{
625 /*
626 * Allocate the copy buffer
627 */
628 struct sk_buff *n = alloc_skb(skb->end - skb->head, gfp_mask);
629
630 if (!n)
631 goto out;
632
633 /* Set the data pointer */
634 skb_reserve(n, skb->data - skb->head);
635 /* Set the tail pointer and length */
636 skb_put(n, skb_headlen(skb));
637 /* Copy the bytes */
638 memcpy(n->data, skb->data, n->len);
639 n->csum = skb->csum;
640 n->ip_summed = skb->ip_summed;
641
Herbert Xu25f484a2006-11-07 14:57:15 -0800642 n->truesize += skb->data_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 n->data_len = skb->data_len;
644 n->len = skb->len;
645
646 if (skb_shinfo(skb)->nr_frags) {
647 int i;
648
649 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
650 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
651 get_page(skb_shinfo(n)->frags[i].page);
652 }
653 skb_shinfo(n)->nr_frags = i;
654 }
655
656 if (skb_shinfo(skb)->frag_list) {
657 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
658 skb_clone_fraglist(n);
659 }
660
661 copy_skb_header(n, skb);
662out:
663 return n;
664}
665
666/**
667 * pskb_expand_head - reallocate header of &sk_buff
668 * @skb: buffer to reallocate
669 * @nhead: room to add at head
670 * @ntail: room to add at tail
671 * @gfp_mask: allocation priority
672 *
673 * Expands (or creates identical copy, if &nhead and &ntail are zero)
674 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
675 * reference count of 1. Returns zero in the case of success or error,
676 * if expansion failed. In the last case, &sk_buff is not changed.
677 *
678 * All the pointers pointing into skb header may change and must be
679 * reloaded after call to this function.
680 */
681
Victor Fusco86a76ca2005-07-08 14:57:47 -0700682int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
Al Virodd0fc662005-10-07 07:46:04 +0100683 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684{
685 int i;
686 u8 *data;
687 int size = nhead + (skb->end - skb->head) + ntail;
688 long off;
689
690 if (skb_shared(skb))
691 BUG();
692
693 size = SKB_DATA_ALIGN(size);
694
695 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
696 if (!data)
697 goto nodata;
698
699 /* Copy only real data... and, alas, header. This should be
700 * optimized for the cases when header is void. */
701 memcpy(data + nhead, skb->head, skb->tail - skb->head);
702 memcpy(data + size, skb->end, sizeof(struct skb_shared_info));
703
704 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
705 get_page(skb_shinfo(skb)->frags[i].page);
706
707 if (skb_shinfo(skb)->frag_list)
708 skb_clone_fraglist(skb);
709
710 skb_release_data(skb);
711
712 off = (data + nhead) - skb->head;
713
714 skb->head = data;
715 skb->end = data + size;
716 skb->data += off;
717 skb->tail += off;
718 skb->mac.raw += off;
719 skb->h.raw += off;
720 skb->nh.raw += off;
721 skb->cloned = 0;
722 skb->nohdr = 0;
723 atomic_set(&skb_shinfo(skb)->dataref, 1);
724 return 0;
725
726nodata:
727 return -ENOMEM;
728}
729
730/* Make private copy of skb with writable head and some headroom */
731
732struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
733{
734 struct sk_buff *skb2;
735 int delta = headroom - skb_headroom(skb);
736
737 if (delta <= 0)
738 skb2 = pskb_copy(skb, GFP_ATOMIC);
739 else {
740 skb2 = skb_clone(skb, GFP_ATOMIC);
741 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
742 GFP_ATOMIC)) {
743 kfree_skb(skb2);
744 skb2 = NULL;
745 }
746 }
747 return skb2;
748}
749
750
751/**
752 * skb_copy_expand - copy and expand sk_buff
753 * @skb: buffer to copy
754 * @newheadroom: new free bytes at head
755 * @newtailroom: new free bytes at tail
756 * @gfp_mask: allocation priority
757 *
758 * Make a copy of both an &sk_buff and its data and while doing so
759 * allocate additional space.
760 *
761 * This is used when the caller wishes to modify the data and needs a
762 * private copy of the data to alter as well as more space for new fields.
763 * Returns %NULL on failure or the pointer to the buffer
764 * on success. The returned buffer has a reference count of 1.
765 *
766 * You must pass %GFP_ATOMIC as the allocation priority if this function
767 * is called from an interrupt.
768 *
769 * BUG ALERT: ip_summed is not copied. Why does this work? Is it used
770 * only by netfilter in the cases when checksum is recalculated? --ANK
771 */
772struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
Victor Fusco86a76ca2005-07-08 14:57:47 -0700773 int newheadroom, int newtailroom,
Al Virodd0fc662005-10-07 07:46:04 +0100774 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775{
776 /*
777 * Allocate the copy buffer
778 */
779 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
780 gfp_mask);
781 int head_copy_len, head_copy_off;
782
783 if (!n)
784 return NULL;
785
786 skb_reserve(n, newheadroom);
787
788 /* Set the tail pointer and length */
789 skb_put(n, skb->len);
790
791 head_copy_len = skb_headroom(skb);
792 head_copy_off = 0;
793 if (newheadroom <= head_copy_len)
794 head_copy_len = newheadroom;
795 else
796 head_copy_off = newheadroom - head_copy_len;
797
798 /* Copy the linear header and data. */
799 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
800 skb->len + head_copy_len))
801 BUG();
802
803 copy_skb_header(n, skb);
804
805 return n;
806}
807
808/**
809 * skb_pad - zero pad the tail of an skb
810 * @skb: buffer to pad
811 * @pad: space to pad
812 *
813 * Ensure that a buffer is followed by a padding area that is zero
814 * filled. Used by network drivers which may DMA or transfer data
815 * beyond the buffer end onto the wire.
816 *
Herbert Xu5b057c62006-06-23 02:06:41 -0700817 * May return error in out of memory cases. The skb is freed on error.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 */
819
Herbert Xu5b057c62006-06-23 02:06:41 -0700820int skb_pad(struct sk_buff *skb, int pad)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821{
Herbert Xu5b057c62006-06-23 02:06:41 -0700822 int err;
823 int ntail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824
825 /* If the skbuff is non linear tailroom is always zero.. */
Herbert Xu5b057c62006-06-23 02:06:41 -0700826 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 memset(skb->data+skb->len, 0, pad);
Herbert Xu5b057c62006-06-23 02:06:41 -0700828 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 }
Herbert Xu5b057c62006-06-23 02:06:41 -0700830
831 ntail = skb->data_len + pad - (skb->end - skb->tail);
832 if (likely(skb_cloned(skb) || ntail > 0)) {
833 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
834 if (unlikely(err))
835 goto free_skb;
836 }
837
838 /* FIXME: The use of this function with non-linear skb's really needs
839 * to be audited.
840 */
841 err = skb_linearize(skb);
842 if (unlikely(err))
843 goto free_skb;
844
845 memset(skb->data + skb->len, 0, pad);
846 return 0;
847
848free_skb:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 kfree_skb(skb);
Herbert Xu5b057c62006-06-23 02:06:41 -0700850 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851}
852
Herbert Xu3cc0e872006-06-09 16:13:38 -0700853/* Trims skb to length len. It can change skb pointers.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 */
855
Herbert Xu3cc0e872006-06-09 16:13:38 -0700856int ___pskb_trim(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857{
Herbert Xu27b437c2006-07-13 19:26:39 -0700858 struct sk_buff **fragp;
859 struct sk_buff *frag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 int offset = skb_headlen(skb);
861 int nfrags = skb_shinfo(skb)->nr_frags;
862 int i;
Herbert Xu27b437c2006-07-13 19:26:39 -0700863 int err;
864
865 if (skb_cloned(skb) &&
866 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
867 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868
Herbert Xuf4d26fb2006-07-30 20:20:28 -0700869 i = 0;
870 if (offset >= len)
871 goto drop_pages;
872
873 for (; i < nfrags; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 int end = offset + skb_shinfo(skb)->frags[i].size;
Herbert Xu27b437c2006-07-13 19:26:39 -0700875
876 if (end < len) {
877 offset = end;
878 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 }
Herbert Xu27b437c2006-07-13 19:26:39 -0700880
Herbert Xuf4d26fb2006-07-30 20:20:28 -0700881 skb_shinfo(skb)->frags[i++].size = len - offset;
Herbert Xu27b437c2006-07-13 19:26:39 -0700882
Herbert Xuf4d26fb2006-07-30 20:20:28 -0700883drop_pages:
Herbert Xu27b437c2006-07-13 19:26:39 -0700884 skb_shinfo(skb)->nr_frags = i;
885
886 for (; i < nfrags; i++)
887 put_page(skb_shinfo(skb)->frags[i].page);
888
889 if (skb_shinfo(skb)->frag_list)
890 skb_drop_fraglist(skb);
Herbert Xuf4d26fb2006-07-30 20:20:28 -0700891 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 }
893
Herbert Xu27b437c2006-07-13 19:26:39 -0700894 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
895 fragp = &frag->next) {
896 int end = offset + frag->len;
897
898 if (skb_shared(frag)) {
899 struct sk_buff *nfrag;
900
901 nfrag = skb_clone(frag, GFP_ATOMIC);
902 if (unlikely(!nfrag))
903 return -ENOMEM;
904
905 nfrag->next = frag->next;
Herbert Xuf4d26fb2006-07-30 20:20:28 -0700906 kfree_skb(frag);
Herbert Xu27b437c2006-07-13 19:26:39 -0700907 frag = nfrag;
908 *fragp = frag;
909 }
910
911 if (end < len) {
912 offset = end;
913 continue;
914 }
915
916 if (end > len &&
917 unlikely((err = pskb_trim(frag, len - offset))))
918 return err;
919
920 if (frag->next)
921 skb_drop_list(&frag->next);
922 break;
923 }
924
Herbert Xuf4d26fb2006-07-30 20:20:28 -0700925done:
Herbert Xu27b437c2006-07-13 19:26:39 -0700926 if (len > skb_headlen(skb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 skb->data_len -= skb->len - len;
928 skb->len = len;
929 } else {
Herbert Xu27b437c2006-07-13 19:26:39 -0700930 skb->len = len;
931 skb->data_len = 0;
932 skb->tail = skb->data + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 }
934
935 return 0;
936}
937
938/**
939 * __pskb_pull_tail - advance tail of skb header
940 * @skb: buffer to reallocate
941 * @delta: number of bytes to advance tail
942 *
943 * The function makes a sense only on a fragmented &sk_buff,
944 * it expands header moving its tail forward and copying necessary
945 * data from fragmented part.
946 *
947 * &sk_buff MUST have reference count of 1.
948 *
949 * Returns %NULL (and &sk_buff does not change) if pull failed
950 * or value of new tail of skb in the case of success.
951 *
952 * All the pointers pointing into skb header may change and must be
953 * reloaded after call to this function.
954 */
955
956/* Moves tail of skb head forward, copying data from fragmented part,
957 * when it is necessary.
958 * 1. It may fail due to malloc failure.
959 * 2. It may change skb pointers.
960 *
961 * It is pretty complicated. Luckily, it is called only in exceptional cases.
962 */
963unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
964{
965 /* If skb has not enough free space at tail, get new one
966 * plus 128 bytes for future expansions. If we have enough
967 * room at tail, reallocate without expansion only if skb is cloned.
968 */
969 int i, k, eat = (skb->tail + delta) - skb->end;
970
971 if (eat > 0 || skb_cloned(skb)) {
972 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
973 GFP_ATOMIC))
974 return NULL;
975 }
976
977 if (skb_copy_bits(skb, skb_headlen(skb), skb->tail, delta))
978 BUG();
979
980 /* Optimization: no fragments, no reasons to preestimate
981 * size of pulled pages. Superb.
982 */
983 if (!skb_shinfo(skb)->frag_list)
984 goto pull_pages;
985
986 /* Estimate size of pulled pages. */
987 eat = delta;
988 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
989 if (skb_shinfo(skb)->frags[i].size >= eat)
990 goto pull_pages;
991 eat -= skb_shinfo(skb)->frags[i].size;
992 }
993
994 /* If we need update frag list, we are in troubles.
995 * Certainly, it possible to add an offset to skb data,
996 * but taking into account that pulling is expected to
997 * be very rare operation, it is worth to fight against
998 * further bloating skb head and crucify ourselves here instead.
999 * Pure masohism, indeed. 8)8)
1000 */
1001 if (eat) {
1002 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1003 struct sk_buff *clone = NULL;
1004 struct sk_buff *insp = NULL;
1005
1006 do {
Kris Katterjohn09a62662006-01-08 22:24:28 -08001007 BUG_ON(!list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008
1009 if (list->len <= eat) {
1010 /* Eaten as whole. */
1011 eat -= list->len;
1012 list = list->next;
1013 insp = list;
1014 } else {
1015 /* Eaten partially. */
1016
1017 if (skb_shared(list)) {
1018 /* Sucks! We need to fork list. :-( */
1019 clone = skb_clone(list, GFP_ATOMIC);
1020 if (!clone)
1021 return NULL;
1022 insp = list->next;
1023 list = clone;
1024 } else {
1025 /* This may be pulled without
1026 * problems. */
1027 insp = list;
1028 }
1029 if (!pskb_pull(list, eat)) {
1030 if (clone)
1031 kfree_skb(clone);
1032 return NULL;
1033 }
1034 break;
1035 }
1036 } while (eat);
1037
1038 /* Free pulled out fragments. */
1039 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1040 skb_shinfo(skb)->frag_list = list->next;
1041 kfree_skb(list);
1042 }
1043 /* And insert new clone at head. */
1044 if (clone) {
1045 clone->next = list;
1046 skb_shinfo(skb)->frag_list = clone;
1047 }
1048 }
1049 /* Success! Now we may commit changes to skb data. */
1050
1051pull_pages:
1052 eat = delta;
1053 k = 0;
1054 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1055 if (skb_shinfo(skb)->frags[i].size <= eat) {
1056 put_page(skb_shinfo(skb)->frags[i].page);
1057 eat -= skb_shinfo(skb)->frags[i].size;
1058 } else {
1059 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1060 if (eat) {
1061 skb_shinfo(skb)->frags[k].page_offset += eat;
1062 skb_shinfo(skb)->frags[k].size -= eat;
1063 eat = 0;
1064 }
1065 k++;
1066 }
1067 }
1068 skb_shinfo(skb)->nr_frags = k;
1069
1070 skb->tail += delta;
1071 skb->data_len -= delta;
1072
1073 return skb->tail;
1074}
1075
1076/* Copy some data bits from skb to kernel buffer. */
1077
1078int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1079{
1080 int i, copy;
1081 int start = skb_headlen(skb);
1082
1083 if (offset > (int)skb->len - len)
1084 goto fault;
1085
1086 /* Copy header. */
1087 if ((copy = start - offset) > 0) {
1088 if (copy > len)
1089 copy = len;
1090 memcpy(to, skb->data + offset, copy);
1091 if ((len -= copy) == 0)
1092 return 0;
1093 offset += copy;
1094 to += copy;
1095 }
1096
1097 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1098 int end;
1099
1100 BUG_TRAP(start <= offset + len);
1101
1102 end = start + skb_shinfo(skb)->frags[i].size;
1103 if ((copy = end - offset) > 0) {
1104 u8 *vaddr;
1105
1106 if (copy > len)
1107 copy = len;
1108
1109 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1110 memcpy(to,
1111 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1112 offset - start, copy);
1113 kunmap_skb_frag(vaddr);
1114
1115 if ((len -= copy) == 0)
1116 return 0;
1117 offset += copy;
1118 to += copy;
1119 }
1120 start = end;
1121 }
1122
1123 if (skb_shinfo(skb)->frag_list) {
1124 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1125
1126 for (; list; list = list->next) {
1127 int end;
1128
1129 BUG_TRAP(start <= offset + len);
1130
1131 end = start + list->len;
1132 if ((copy = end - offset) > 0) {
1133 if (copy > len)
1134 copy = len;
1135 if (skb_copy_bits(list, offset - start,
1136 to, copy))
1137 goto fault;
1138 if ((len -= copy) == 0)
1139 return 0;
1140 offset += copy;
1141 to += copy;
1142 }
1143 start = end;
1144 }
1145 }
1146 if (!len)
1147 return 0;
1148
1149fault:
1150 return -EFAULT;
1151}
1152
Herbert Xu357b40a2005-04-19 22:30:14 -07001153/**
1154 * skb_store_bits - store bits from kernel buffer to skb
1155 * @skb: destination buffer
1156 * @offset: offset in destination
1157 * @from: source buffer
1158 * @len: number of bytes to copy
1159 *
1160 * Copy the specified number of bytes from the source buffer to the
1161 * destination skb. This function handles all the messy bits of
1162 * traversing fragment lists and such.
1163 */
1164
1165int skb_store_bits(const struct sk_buff *skb, int offset, void *from, int len)
1166{
1167 int i, copy;
1168 int start = skb_headlen(skb);
1169
1170 if (offset > (int)skb->len - len)
1171 goto fault;
1172
1173 if ((copy = start - offset) > 0) {
1174 if (copy > len)
1175 copy = len;
1176 memcpy(skb->data + offset, from, copy);
1177 if ((len -= copy) == 0)
1178 return 0;
1179 offset += copy;
1180 from += copy;
1181 }
1182
1183 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1184 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1185 int end;
1186
1187 BUG_TRAP(start <= offset + len);
1188
1189 end = start + frag->size;
1190 if ((copy = end - offset) > 0) {
1191 u8 *vaddr;
1192
1193 if (copy > len)
1194 copy = len;
1195
1196 vaddr = kmap_skb_frag(frag);
1197 memcpy(vaddr + frag->page_offset + offset - start,
1198 from, copy);
1199 kunmap_skb_frag(vaddr);
1200
1201 if ((len -= copy) == 0)
1202 return 0;
1203 offset += copy;
1204 from += copy;
1205 }
1206 start = end;
1207 }
1208
1209 if (skb_shinfo(skb)->frag_list) {
1210 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1211
1212 for (; list; list = list->next) {
1213 int end;
1214
1215 BUG_TRAP(start <= offset + len);
1216
1217 end = start + list->len;
1218 if ((copy = end - offset) > 0) {
1219 if (copy > len)
1220 copy = len;
1221 if (skb_store_bits(list, offset - start,
1222 from, copy))
1223 goto fault;
1224 if ((len -= copy) == 0)
1225 return 0;
1226 offset += copy;
1227 from += copy;
1228 }
1229 start = end;
1230 }
1231 }
1232 if (!len)
1233 return 0;
1234
1235fault:
1236 return -EFAULT;
1237}
1238
1239EXPORT_SYMBOL(skb_store_bits);
1240
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241/* Checksum skb data. */
1242
Al Viro2bbbc862006-11-14 21:37:14 -08001243__wsum skb_checksum(const struct sk_buff *skb, int offset,
1244 int len, __wsum csum)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245{
1246 int start = skb_headlen(skb);
1247 int i, copy = start - offset;
1248 int pos = 0;
1249
1250 /* Checksum header. */
1251 if (copy > 0) {
1252 if (copy > len)
1253 copy = len;
1254 csum = csum_partial(skb->data + offset, copy, csum);
1255 if ((len -= copy) == 0)
1256 return csum;
1257 offset += copy;
1258 pos = copy;
1259 }
1260
1261 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1262 int end;
1263
1264 BUG_TRAP(start <= offset + len);
1265
1266 end = start + skb_shinfo(skb)->frags[i].size;
1267 if ((copy = end - offset) > 0) {
Al Viro44bb9362006-11-14 21:36:14 -08001268 __wsum csum2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 u8 *vaddr;
1270 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1271
1272 if (copy > len)
1273 copy = len;
1274 vaddr = kmap_skb_frag(frag);
1275 csum2 = csum_partial(vaddr + frag->page_offset +
1276 offset - start, copy, 0);
1277 kunmap_skb_frag(vaddr);
1278 csum = csum_block_add(csum, csum2, pos);
1279 if (!(len -= copy))
1280 return csum;
1281 offset += copy;
1282 pos += copy;
1283 }
1284 start = end;
1285 }
1286
1287 if (skb_shinfo(skb)->frag_list) {
1288 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1289
1290 for (; list; list = list->next) {
1291 int end;
1292
1293 BUG_TRAP(start <= offset + len);
1294
1295 end = start + list->len;
1296 if ((copy = end - offset) > 0) {
Al Viro5f92a732006-11-14 21:36:54 -08001297 __wsum csum2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 if (copy > len)
1299 copy = len;
1300 csum2 = skb_checksum(list, offset - start,
1301 copy, 0);
1302 csum = csum_block_add(csum, csum2, pos);
1303 if ((len -= copy) == 0)
1304 return csum;
1305 offset += copy;
1306 pos += copy;
1307 }
1308 start = end;
1309 }
1310 }
Kris Katterjohn09a62662006-01-08 22:24:28 -08001311 BUG_ON(len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312
1313 return csum;
1314}
1315
1316/* Both of above in one bottle. */
1317
1318unsigned int skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1319 u8 *to, int len, unsigned int csum)
1320{
1321 int start = skb_headlen(skb);
1322 int i, copy = start - offset;
1323 int pos = 0;
1324
1325 /* Copy header. */
1326 if (copy > 0) {
1327 if (copy > len)
1328 copy = len;
1329 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1330 copy, csum);
1331 if ((len -= copy) == 0)
1332 return csum;
1333 offset += copy;
1334 to += copy;
1335 pos = copy;
1336 }
1337
1338 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1339 int end;
1340
1341 BUG_TRAP(start <= offset + len);
1342
1343 end = start + skb_shinfo(skb)->frags[i].size;
1344 if ((copy = end - offset) > 0) {
Al Viro50842052006-11-14 21:36:34 -08001345 __wsum csum2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 u8 *vaddr;
1347 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1348
1349 if (copy > len)
1350 copy = len;
1351 vaddr = kmap_skb_frag(frag);
1352 csum2 = csum_partial_copy_nocheck(vaddr +
1353 frag->page_offset +
1354 offset - start, to,
1355 copy, 0);
1356 kunmap_skb_frag(vaddr);
1357 csum = csum_block_add(csum, csum2, pos);
1358 if (!(len -= copy))
1359 return csum;
1360 offset += copy;
1361 to += copy;
1362 pos += copy;
1363 }
1364 start = end;
1365 }
1366
1367 if (skb_shinfo(skb)->frag_list) {
1368 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1369
1370 for (; list; list = list->next) {
1371 unsigned int csum2;
1372 int end;
1373
1374 BUG_TRAP(start <= offset + len);
1375
1376 end = start + list->len;
1377 if ((copy = end - offset) > 0) {
1378 if (copy > len)
1379 copy = len;
1380 csum2 = skb_copy_and_csum_bits(list,
1381 offset - start,
1382 to, copy, 0);
1383 csum = csum_block_add(csum, csum2, pos);
1384 if ((len -= copy) == 0)
1385 return csum;
1386 offset += copy;
1387 to += copy;
1388 pos += copy;
1389 }
1390 start = end;
1391 }
1392 }
Kris Katterjohn09a62662006-01-08 22:24:28 -08001393 BUG_ON(len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 return csum;
1395}
1396
1397void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1398{
Al Virod3bc23e2006-11-14 21:24:49 -08001399 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 long csstart;
1401
Patrick McHardy84fa7932006-08-29 16:44:56 -07001402 if (skb->ip_summed == CHECKSUM_PARTIAL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 csstart = skb->h.raw - skb->data;
1404 else
1405 csstart = skb_headlen(skb);
1406
Kris Katterjohn09a62662006-01-08 22:24:28 -08001407 BUG_ON(csstart > skb_headlen(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408
1409 memcpy(to, skb->data, csstart);
1410
1411 csum = 0;
1412 if (csstart != skb->len)
1413 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1414 skb->len - csstart, 0);
1415
Patrick McHardy84fa7932006-08-29 16:44:56 -07001416 if (skb->ip_summed == CHECKSUM_PARTIAL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 long csstuff = csstart + skb->csum;
1418
Al Virod3bc23e2006-11-14 21:24:49 -08001419 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 }
1421}
1422
1423/**
1424 * skb_dequeue - remove from the head of the queue
1425 * @list: list to dequeue from
1426 *
1427 * Remove the head of the list. The list lock is taken so the function
1428 * may be used safely with other locking list functions. The head item is
1429 * returned or %NULL if the list is empty.
1430 */
1431
1432struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1433{
1434 unsigned long flags;
1435 struct sk_buff *result;
1436
1437 spin_lock_irqsave(&list->lock, flags);
1438 result = __skb_dequeue(list);
1439 spin_unlock_irqrestore(&list->lock, flags);
1440 return result;
1441}
1442
1443/**
1444 * skb_dequeue_tail - remove from the tail of the queue
1445 * @list: list to dequeue from
1446 *
1447 * Remove the tail of the list. The list lock is taken so the function
1448 * may be used safely with other locking list functions. The tail item is
1449 * returned or %NULL if the list is empty.
1450 */
1451struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1452{
1453 unsigned long flags;
1454 struct sk_buff *result;
1455
1456 spin_lock_irqsave(&list->lock, flags);
1457 result = __skb_dequeue_tail(list);
1458 spin_unlock_irqrestore(&list->lock, flags);
1459 return result;
1460}
1461
1462/**
1463 * skb_queue_purge - empty a list
1464 * @list: list to empty
1465 *
1466 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1467 * the list and one reference dropped. This function takes the list
1468 * lock and is atomic with respect to other list locking functions.
1469 */
1470void skb_queue_purge(struct sk_buff_head *list)
1471{
1472 struct sk_buff *skb;
1473 while ((skb = skb_dequeue(list)) != NULL)
1474 kfree_skb(skb);
1475}
1476
1477/**
1478 * skb_queue_head - queue a buffer at the list head
1479 * @list: list to use
1480 * @newsk: buffer to queue
1481 *
1482 * Queue a buffer at the start of the list. This function takes the
1483 * list lock and can be used safely with other locking &sk_buff functions
1484 * safely.
1485 *
1486 * A buffer cannot be placed on two lists at the same time.
1487 */
1488void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1489{
1490 unsigned long flags;
1491
1492 spin_lock_irqsave(&list->lock, flags);
1493 __skb_queue_head(list, newsk);
1494 spin_unlock_irqrestore(&list->lock, flags);
1495}
1496
1497/**
1498 * skb_queue_tail - queue a buffer at the list tail
1499 * @list: list to use
1500 * @newsk: buffer to queue
1501 *
1502 * Queue a buffer at the tail of the list. This function takes the
1503 * list lock and can be used safely with other locking &sk_buff functions
1504 * safely.
1505 *
1506 * A buffer cannot be placed on two lists at the same time.
1507 */
1508void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1509{
1510 unsigned long flags;
1511
1512 spin_lock_irqsave(&list->lock, flags);
1513 __skb_queue_tail(list, newsk);
1514 spin_unlock_irqrestore(&list->lock, flags);
1515}
David S. Miller8728b832005-08-09 19:25:21 -07001516
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517/**
1518 * skb_unlink - remove a buffer from a list
1519 * @skb: buffer to remove
David S. Miller8728b832005-08-09 19:25:21 -07001520 * @list: list to use
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 *
David S. Miller8728b832005-08-09 19:25:21 -07001522 * Remove a packet from a list. The list locks are taken and this
1523 * function is atomic with respect to other list locked calls
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 *
David S. Miller8728b832005-08-09 19:25:21 -07001525 * You must know what list the SKB is on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 */
David S. Miller8728b832005-08-09 19:25:21 -07001527void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528{
David S. Miller8728b832005-08-09 19:25:21 -07001529 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530
David S. Miller8728b832005-08-09 19:25:21 -07001531 spin_lock_irqsave(&list->lock, flags);
1532 __skb_unlink(skb, list);
1533 spin_unlock_irqrestore(&list->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534}
1535
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536/**
1537 * skb_append - append a buffer
1538 * @old: buffer to insert after
1539 * @newsk: buffer to insert
David S. Miller8728b832005-08-09 19:25:21 -07001540 * @list: list to use
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 *
1542 * Place a packet after a given packet in a list. The list locks are taken
1543 * and this function is atomic with respect to other list locked calls.
1544 * A buffer cannot be placed on two lists at the same time.
1545 */
David S. Miller8728b832005-08-09 19:25:21 -07001546void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547{
1548 unsigned long flags;
1549
David S. Miller8728b832005-08-09 19:25:21 -07001550 spin_lock_irqsave(&list->lock, flags);
1551 __skb_append(old, newsk, list);
1552 spin_unlock_irqrestore(&list->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553}
1554
1555
1556/**
1557 * skb_insert - insert a buffer
1558 * @old: buffer to insert before
1559 * @newsk: buffer to insert
David S. Miller8728b832005-08-09 19:25:21 -07001560 * @list: list to use
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 *
David S. Miller8728b832005-08-09 19:25:21 -07001562 * Place a packet before a given packet in a list. The list locks are
1563 * taken and this function is atomic with respect to other list locked
1564 * calls.
1565 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 * A buffer cannot be placed on two lists at the same time.
1567 */
David S. Miller8728b832005-08-09 19:25:21 -07001568void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569{
1570 unsigned long flags;
1571
David S. Miller8728b832005-08-09 19:25:21 -07001572 spin_lock_irqsave(&list->lock, flags);
1573 __skb_insert(newsk, old->prev, old, list);
1574 spin_unlock_irqrestore(&list->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575}
1576
1577#if 0
1578/*
1579 * Tune the memory allocator for a new MTU size.
1580 */
1581void skb_add_mtu(int mtu)
1582{
1583 /* Must match allocation in alloc_skb */
1584 mtu = SKB_DATA_ALIGN(mtu) + sizeof(struct skb_shared_info);
1585
1586 kmem_add_cache_size(mtu);
1587}
1588#endif
1589
1590static inline void skb_split_inside_header(struct sk_buff *skb,
1591 struct sk_buff* skb1,
1592 const u32 len, const int pos)
1593{
1594 int i;
1595
1596 memcpy(skb_put(skb1, pos - len), skb->data + len, pos - len);
1597
1598 /* And move data appendix as is. */
1599 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1600 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
1601
1602 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
1603 skb_shinfo(skb)->nr_frags = 0;
1604 skb1->data_len = skb->data_len;
1605 skb1->len += skb1->data_len;
1606 skb->data_len = 0;
1607 skb->len = len;
1608 skb->tail = skb->data + len;
1609}
1610
1611static inline void skb_split_no_header(struct sk_buff *skb,
1612 struct sk_buff* skb1,
1613 const u32 len, int pos)
1614{
1615 int i, k = 0;
1616 const int nfrags = skb_shinfo(skb)->nr_frags;
1617
1618 skb_shinfo(skb)->nr_frags = 0;
1619 skb1->len = skb1->data_len = skb->len - len;
1620 skb->len = len;
1621 skb->data_len = len - pos;
1622
1623 for (i = 0; i < nfrags; i++) {
1624 int size = skb_shinfo(skb)->frags[i].size;
1625
1626 if (pos + size > len) {
1627 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
1628
1629 if (pos < len) {
1630 /* Split frag.
1631 * We have two variants in this case:
1632 * 1. Move all the frag to the second
1633 * part, if it is possible. F.e.
1634 * this approach is mandatory for TUX,
1635 * where splitting is expensive.
1636 * 2. Split is accurately. We make this.
1637 */
1638 get_page(skb_shinfo(skb)->frags[i].page);
1639 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
1640 skb_shinfo(skb1)->frags[0].size -= len - pos;
1641 skb_shinfo(skb)->frags[i].size = len - pos;
1642 skb_shinfo(skb)->nr_frags++;
1643 }
1644 k++;
1645 } else
1646 skb_shinfo(skb)->nr_frags++;
1647 pos += size;
1648 }
1649 skb_shinfo(skb1)->nr_frags = k;
1650}
1651
1652/**
1653 * skb_split - Split fragmented skb to two parts at length len.
1654 * @skb: the buffer to split
1655 * @skb1: the buffer to receive the second part
1656 * @len: new length for skb
1657 */
1658void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
1659{
1660 int pos = skb_headlen(skb);
1661
1662 if (len < pos) /* Split line is inside header. */
1663 skb_split_inside_header(skb, skb1, len, pos);
1664 else /* Second chunk has no header, nothing to copy. */
1665 skb_split_no_header(skb, skb1, len, pos);
1666}
1667
Thomas Graf677e90e2005-06-23 20:59:51 -07001668/**
1669 * skb_prepare_seq_read - Prepare a sequential read of skb data
1670 * @skb: the buffer to read
1671 * @from: lower offset of data to be read
1672 * @to: upper offset of data to be read
1673 * @st: state variable
1674 *
1675 * Initializes the specified state variable. Must be called before
1676 * invoking skb_seq_read() for the first time.
1677 */
1678void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1679 unsigned int to, struct skb_seq_state *st)
1680{
1681 st->lower_offset = from;
1682 st->upper_offset = to;
1683 st->root_skb = st->cur_skb = skb;
1684 st->frag_idx = st->stepped_offset = 0;
1685 st->frag_data = NULL;
1686}
1687
1688/**
1689 * skb_seq_read - Sequentially read skb data
1690 * @consumed: number of bytes consumed by the caller so far
1691 * @data: destination pointer for data to be returned
1692 * @st: state variable
1693 *
1694 * Reads a block of skb data at &consumed relative to the
1695 * lower offset specified to skb_prepare_seq_read(). Assigns
1696 * the head of the data block to &data and returns the length
1697 * of the block or 0 if the end of the skb data or the upper
1698 * offset has been reached.
1699 *
1700 * The caller is not required to consume all of the data
1701 * returned, i.e. &consumed is typically set to the number
1702 * of bytes already consumed and the next call to
1703 * skb_seq_read() will return the remaining part of the block.
1704 *
1705 * Note: The size of each block of data returned can be arbitary,
1706 * this limitation is the cost for zerocopy seqeuental
1707 * reads of potentially non linear data.
1708 *
1709 * Note: Fragment lists within fragments are not implemented
1710 * at the moment, state->root_skb could be replaced with
1711 * a stack for this purpose.
1712 */
1713unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1714 struct skb_seq_state *st)
1715{
1716 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
1717 skb_frag_t *frag;
1718
1719 if (unlikely(abs_offset >= st->upper_offset))
1720 return 0;
1721
1722next_skb:
1723 block_limit = skb_headlen(st->cur_skb);
1724
1725 if (abs_offset < block_limit) {
1726 *data = st->cur_skb->data + abs_offset;
1727 return block_limit - abs_offset;
1728 }
1729
1730 if (st->frag_idx == 0 && !st->frag_data)
1731 st->stepped_offset += skb_headlen(st->cur_skb);
1732
1733 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
1734 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
1735 block_limit = frag->size + st->stepped_offset;
1736
1737 if (abs_offset < block_limit) {
1738 if (!st->frag_data)
1739 st->frag_data = kmap_skb_frag(frag);
1740
1741 *data = (u8 *) st->frag_data + frag->page_offset +
1742 (abs_offset - st->stepped_offset);
1743
1744 return block_limit - abs_offset;
1745 }
1746
1747 if (st->frag_data) {
1748 kunmap_skb_frag(st->frag_data);
1749 st->frag_data = NULL;
1750 }
1751
1752 st->frag_idx++;
1753 st->stepped_offset += frag->size;
1754 }
1755
1756 if (st->cur_skb->next) {
1757 st->cur_skb = st->cur_skb->next;
1758 st->frag_idx = 0;
1759 goto next_skb;
1760 } else if (st->root_skb == st->cur_skb &&
1761 skb_shinfo(st->root_skb)->frag_list) {
1762 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
1763 goto next_skb;
1764 }
1765
1766 return 0;
1767}
1768
1769/**
1770 * skb_abort_seq_read - Abort a sequential read of skb data
1771 * @st: state variable
1772 *
1773 * Must be called if skb_seq_read() was not called until it
1774 * returned 0.
1775 */
1776void skb_abort_seq_read(struct skb_seq_state *st)
1777{
1778 if (st->frag_data)
1779 kunmap_skb_frag(st->frag_data);
1780}
1781
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07001782#define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
1783
1784static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
1785 struct ts_config *conf,
1786 struct ts_state *state)
1787{
1788 return skb_seq_read(offset, text, TS_SKB_CB(state));
1789}
1790
1791static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
1792{
1793 skb_abort_seq_read(TS_SKB_CB(state));
1794}
1795
1796/**
1797 * skb_find_text - Find a text pattern in skb data
1798 * @skb: the buffer to look in
1799 * @from: search offset
1800 * @to: search limit
1801 * @config: textsearch configuration
1802 * @state: uninitialized textsearch state variable
1803 *
1804 * Finds a pattern in the skb data according to the specified
1805 * textsearch configuration. Use textsearch_next() to retrieve
1806 * subsequent occurrences of the pattern. Returns the offset
1807 * to the first occurrence or UINT_MAX if no match was found.
1808 */
1809unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1810 unsigned int to, struct ts_config *config,
1811 struct ts_state *state)
1812{
Phil Oesterf72b9482006-06-26 00:00:57 -07001813 unsigned int ret;
1814
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07001815 config->get_next_block = skb_ts_get_next_block;
1816 config->finish = skb_ts_finish;
1817
1818 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
1819
Phil Oesterf72b9482006-06-26 00:00:57 -07001820 ret = textsearch_find(config, state);
1821 return (ret <= to - from ? ret : UINT_MAX);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07001822}
1823
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001824/**
1825 * skb_append_datato_frags: - append the user data to a skb
1826 * @sk: sock structure
1827 * @skb: skb structure to be appened with user data.
1828 * @getfrag: call back function to be used for getting the user data
1829 * @from: pointer to user message iov
1830 * @length: length of the iov message
1831 *
1832 * Description: This procedure append the user data in the fragment part
1833 * of the skb if any page alloc fails user this procedure returns -ENOMEM
1834 */
1835int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
Martin Waitzdab96302005-12-05 13:40:12 -08001836 int (*getfrag)(void *from, char *to, int offset,
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001837 int len, int odd, struct sk_buff *skb),
1838 void *from, int length)
1839{
1840 int frg_cnt = 0;
1841 skb_frag_t *frag = NULL;
1842 struct page *page = NULL;
1843 int copy, left;
1844 int offset = 0;
1845 int ret;
1846
1847 do {
1848 /* Return error if we don't have space for new frag */
1849 frg_cnt = skb_shinfo(skb)->nr_frags;
1850 if (frg_cnt >= MAX_SKB_FRAGS)
1851 return -EFAULT;
1852
1853 /* allocate a new page for next frag */
1854 page = alloc_pages(sk->sk_allocation, 0);
1855
1856 /* If alloc_page fails just return failure and caller will
1857 * free previous allocated pages by doing kfree_skb()
1858 */
1859 if (page == NULL)
1860 return -ENOMEM;
1861
1862 /* initialize the next frag */
1863 sk->sk_sndmsg_page = page;
1864 sk->sk_sndmsg_off = 0;
1865 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
1866 skb->truesize += PAGE_SIZE;
1867 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1868
1869 /* get the new initialized frag */
1870 frg_cnt = skb_shinfo(skb)->nr_frags;
1871 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
1872
1873 /* copy the user data to page */
1874 left = PAGE_SIZE - frag->page_offset;
1875 copy = (length > left)? left : length;
1876
1877 ret = getfrag(from, (page_address(frag->page) +
1878 frag->page_offset + frag->size),
1879 offset, copy, 0, skb);
1880 if (ret < 0)
1881 return -EFAULT;
1882
1883 /* copy was successful so update the size parameters */
1884 sk->sk_sndmsg_off += copy;
1885 frag->size += copy;
1886 skb->len += copy;
1887 skb->data_len += copy;
1888 offset += copy;
1889 length -= copy;
1890
1891 } while (length > 0);
1892
1893 return 0;
1894}
1895
Herbert Xucbb042f2006-03-20 22:43:56 -08001896/**
1897 * skb_pull_rcsum - pull skb and update receive checksum
1898 * @skb: buffer to update
1899 * @start: start of data before pull
1900 * @len: length of data pulled
1901 *
1902 * This function performs an skb_pull on the packet and updates
Patrick McHardy84fa7932006-08-29 16:44:56 -07001903 * update the CHECKSUM_COMPLETE checksum. It should be used on
1904 * receive path processing instead of skb_pull unless you know
1905 * that the checksum difference is zero (e.g., a valid IP header)
1906 * or you are setting ip_summed to CHECKSUM_NONE.
Herbert Xucbb042f2006-03-20 22:43:56 -08001907 */
1908unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
1909{
1910 BUG_ON(len > skb->len);
1911 skb->len -= len;
1912 BUG_ON(skb->len < skb->data_len);
1913 skb_postpull_rcsum(skb, skb->data, len);
1914 return skb->data += len;
1915}
1916
Arnaldo Carvalho de Melof94691a2006-03-20 22:47:55 -08001917EXPORT_SYMBOL_GPL(skb_pull_rcsum);
1918
Herbert Xuf4c50d92006-06-22 03:02:40 -07001919/**
1920 * skb_segment - Perform protocol segmentation on skb.
1921 * @skb: buffer to segment
Herbert Xu576a30e2006-06-27 13:22:38 -07001922 * @features: features for the output path (see dev->features)
Herbert Xuf4c50d92006-06-22 03:02:40 -07001923 *
1924 * This function performs segmentation on the given skb. It returns
1925 * the segment at the given position. It returns NULL if there are
1926 * no more segments to generate, or when an error is encountered.
1927 */
Herbert Xu576a30e2006-06-27 13:22:38 -07001928struct sk_buff *skb_segment(struct sk_buff *skb, int features)
Herbert Xuf4c50d92006-06-22 03:02:40 -07001929{
1930 struct sk_buff *segs = NULL;
1931 struct sk_buff *tail = NULL;
1932 unsigned int mss = skb_shinfo(skb)->gso_size;
1933 unsigned int doffset = skb->data - skb->mac.raw;
1934 unsigned int offset = doffset;
1935 unsigned int headroom;
1936 unsigned int len;
Herbert Xu576a30e2006-06-27 13:22:38 -07001937 int sg = features & NETIF_F_SG;
Herbert Xuf4c50d92006-06-22 03:02:40 -07001938 int nfrags = skb_shinfo(skb)->nr_frags;
1939 int err = -ENOMEM;
1940 int i = 0;
1941 int pos;
1942
1943 __skb_push(skb, doffset);
1944 headroom = skb_headroom(skb);
1945 pos = skb_headlen(skb);
1946
1947 do {
1948 struct sk_buff *nskb;
1949 skb_frag_t *frag;
Herbert Xuc8884ed2006-10-29 15:59:41 -08001950 int hsize;
Herbert Xuf4c50d92006-06-22 03:02:40 -07001951 int k;
1952 int size;
1953
1954 len = skb->len - offset;
1955 if (len > mss)
1956 len = mss;
1957
1958 hsize = skb_headlen(skb) - offset;
1959 if (hsize < 0)
1960 hsize = 0;
Herbert Xuc8884ed2006-10-29 15:59:41 -08001961 if (hsize > len || !sg)
1962 hsize = len;
Herbert Xuf4c50d92006-06-22 03:02:40 -07001963
Herbert Xuc8884ed2006-10-29 15:59:41 -08001964 nskb = alloc_skb(hsize + doffset + headroom, GFP_ATOMIC);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001965 if (unlikely(!nskb))
1966 goto err;
1967
1968 if (segs)
1969 tail->next = nskb;
1970 else
1971 segs = nskb;
1972 tail = nskb;
1973
1974 nskb->dev = skb->dev;
1975 nskb->priority = skb->priority;
1976 nskb->protocol = skb->protocol;
1977 nskb->dst = dst_clone(skb->dst);
1978 memcpy(nskb->cb, skb->cb, sizeof(skb->cb));
1979 nskb->pkt_type = skb->pkt_type;
1980 nskb->mac_len = skb->mac_len;
1981
1982 skb_reserve(nskb, headroom);
1983 nskb->mac.raw = nskb->data;
1984 nskb->nh.raw = nskb->data + skb->mac_len;
1985 nskb->h.raw = nskb->nh.raw + (skb->h.raw - skb->nh.raw);
1986 memcpy(skb_put(nskb, doffset), skb->data, doffset);
1987
1988 if (!sg) {
1989 nskb->csum = skb_copy_and_csum_bits(skb, offset,
1990 skb_put(nskb, len),
1991 len, 0);
1992 continue;
1993 }
1994
1995 frag = skb_shinfo(nskb)->frags;
1996 k = 0;
1997
Patrick McHardy84fa7932006-08-29 16:44:56 -07001998 nskb->ip_summed = CHECKSUM_PARTIAL;
Herbert Xuf4c50d92006-06-22 03:02:40 -07001999 nskb->csum = skb->csum;
2000 memcpy(skb_put(nskb, hsize), skb->data + offset, hsize);
2001
2002 while (pos < offset + len) {
2003 BUG_ON(i >= nfrags);
2004
2005 *frag = skb_shinfo(skb)->frags[i];
2006 get_page(frag->page);
2007 size = frag->size;
2008
2009 if (pos < offset) {
2010 frag->page_offset += offset - pos;
2011 frag->size -= offset - pos;
2012 }
2013
2014 k++;
2015
2016 if (pos + size <= offset + len) {
2017 i++;
2018 pos += size;
2019 } else {
2020 frag->size -= pos + size - (offset + len);
2021 break;
2022 }
2023
2024 frag++;
2025 }
2026
2027 skb_shinfo(nskb)->nr_frags = k;
2028 nskb->data_len = len - hsize;
2029 nskb->len += nskb->data_len;
2030 nskb->truesize += nskb->data_len;
2031 } while ((offset += len) < skb->len);
2032
2033 return segs;
2034
2035err:
2036 while ((skb = segs)) {
2037 segs = skb->next;
2038 kfree(skb);
2039 }
2040 return ERR_PTR(err);
2041}
2042
2043EXPORT_SYMBOL_GPL(skb_segment);
2044
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045void __init skb_init(void)
2046{
2047 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2048 sizeof(struct sk_buff),
2049 0,
Alexey Dobriyane5d679f332006-08-26 19:25:52 -07002050 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 NULL, NULL);
David S. Millerd179cd12005-08-17 14:57:30 -07002052 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2053 (2*sizeof(struct sk_buff)) +
2054 sizeof(atomic_t),
2055 0,
Alexey Dobriyane5d679f332006-08-26 19:25:52 -07002056 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
David S. Millerd179cd12005-08-17 14:57:30 -07002057 NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058}
2059
2060EXPORT_SYMBOL(___pskb_trim);
2061EXPORT_SYMBOL(__kfree_skb);
Jörn Engel231d06a2006-03-20 21:28:35 -08002062EXPORT_SYMBOL(kfree_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063EXPORT_SYMBOL(__pskb_pull_tail);
David S. Millerd179cd12005-08-17 14:57:30 -07002064EXPORT_SYMBOL(__alloc_skb);
Christoph Hellwig8af27452006-07-31 22:35:23 -07002065EXPORT_SYMBOL(__netdev_alloc_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066EXPORT_SYMBOL(pskb_copy);
2067EXPORT_SYMBOL(pskb_expand_head);
2068EXPORT_SYMBOL(skb_checksum);
2069EXPORT_SYMBOL(skb_clone);
2070EXPORT_SYMBOL(skb_clone_fraglist);
2071EXPORT_SYMBOL(skb_copy);
2072EXPORT_SYMBOL(skb_copy_and_csum_bits);
2073EXPORT_SYMBOL(skb_copy_and_csum_dev);
2074EXPORT_SYMBOL(skb_copy_bits);
2075EXPORT_SYMBOL(skb_copy_expand);
2076EXPORT_SYMBOL(skb_over_panic);
2077EXPORT_SYMBOL(skb_pad);
2078EXPORT_SYMBOL(skb_realloc_headroom);
2079EXPORT_SYMBOL(skb_under_panic);
2080EXPORT_SYMBOL(skb_dequeue);
2081EXPORT_SYMBOL(skb_dequeue_tail);
2082EXPORT_SYMBOL(skb_insert);
2083EXPORT_SYMBOL(skb_queue_purge);
2084EXPORT_SYMBOL(skb_queue_head);
2085EXPORT_SYMBOL(skb_queue_tail);
2086EXPORT_SYMBOL(skb_unlink);
2087EXPORT_SYMBOL(skb_append);
2088EXPORT_SYMBOL(skb_split);
Thomas Graf677e90e2005-06-23 20:59:51 -07002089EXPORT_SYMBOL(skb_prepare_seq_read);
2090EXPORT_SYMBOL(skb_seq_read);
2091EXPORT_SYMBOL(skb_abort_seq_read);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002092EXPORT_SYMBOL(skb_find_text);
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002093EXPORT_SYMBOL(skb_append_datato_frags);