blob: 0d0fd28a9041c1d84c1545bb15209d45bff52338 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include <linux/mm.h>
45#include <linux/interrupt.h>
46#include <linux/in.h>
47#include <linux/inet.h>
48#include <linux/slab.h>
49#include <linux/netdevice.h>
50#ifdef CONFIG_NET_CLS_ACT
51#include <net/pkt_sched.h>
52#endif
53#include <linux/string.h>
54#include <linux/skbuff.h>
Jens Axboe9c55e012007-11-06 23:30:13 -080055#include <linux/splice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070056#include <linux/cache.h>
57#include <linux/rtnetlink.h>
58#include <linux/init.h>
David Howells716ea3a2007-04-02 20:19:53 -070059#include <linux/scatterlist.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
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
Al Viroa1f8e7f72006-10-19 16:08:53 -040070#include "kmap_skb.h"
71
Christoph Lametere18b8902006-12-06 20:33:20 -080072static struct kmem_cache *skbuff_head_cache __read_mostly;
73static struct kmem_cache *skbuff_fclone_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070074
Jens Axboe9c55e012007-11-06 23:30:13 -080075static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
76 struct pipe_buffer *buf)
77{
78 struct sk_buff *skb = (struct sk_buff *) buf->private;
79
80 kfree_skb(skb);
81}
82
83static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
84 struct pipe_buffer *buf)
85{
86 struct sk_buff *skb = (struct sk_buff *) buf->private;
87
88 skb_get(skb);
89}
90
91static int sock_pipe_buf_steal(struct pipe_inode_info *pipe,
92 struct pipe_buffer *buf)
93{
94 return 1;
95}
96
97
98/* Pipe buffer operations for a socket. */
99static struct pipe_buf_operations sock_pipe_buf_ops = {
100 .can_merge = 0,
101 .map = generic_pipe_buf_map,
102 .unmap = generic_pipe_buf_unmap,
103 .confirm = generic_pipe_buf_confirm,
104 .release = sock_pipe_buf_release,
105 .steal = sock_pipe_buf_steal,
106 .get = sock_pipe_buf_get,
107};
108
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109/*
110 * Keep out-of-line to prevent kernel bloat.
111 * __builtin_return_address is not used because it is not always
112 * reliable.
113 */
114
115/**
116 * skb_over_panic - private function
117 * @skb: buffer
118 * @sz: size
119 * @here: address
120 *
121 * Out of line support code for skb_put(). Not user callable.
122 */
123void skb_over_panic(struct sk_buff *skb, int sz, void *here)
124{
Patrick McHardy26095452005-04-21 16:43:02 -0700125 printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700126 "data:%p tail:%#lx end:%#lx dev:%s\n",
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700127 here, skb->len, sz, skb->head, skb->data,
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700128 (unsigned long)skb->tail, (unsigned long)skb->end,
Patrick McHardy26095452005-04-21 16:43:02 -0700129 skb->dev ? skb->dev->name : "<NULL>");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 BUG();
131}
132
133/**
134 * skb_under_panic - private function
135 * @skb: buffer
136 * @sz: size
137 * @here: address
138 *
139 * Out of line support code for skb_push(). Not user callable.
140 */
141
142void skb_under_panic(struct sk_buff *skb, int sz, void *here)
143{
Patrick McHardy26095452005-04-21 16:43:02 -0700144 printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700145 "data:%p tail:%#lx end:%#lx dev:%s\n",
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700146 here, skb->len, sz, skb->head, skb->data,
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700147 (unsigned long)skb->tail, (unsigned long)skb->end,
Patrick McHardy26095452005-04-21 16:43:02 -0700148 skb->dev ? skb->dev->name : "<NULL>");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 BUG();
150}
151
David S. Millerdc6de332006-04-20 00:10:50 -0700152void skb_truesize_bug(struct sk_buff *skb)
153{
154 printk(KERN_ERR "SKB BUG: Invalid truesize (%u) "
155 "len=%u, sizeof(sk_buff)=%Zd\n",
156 skb->truesize, skb->len, sizeof(struct sk_buff));
157}
158EXPORT_SYMBOL(skb_truesize_bug);
159
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160/* Allocate a new skbuff. We do this ourselves so we can fill in a few
161 * 'private' fields and also do memory statistics to find all the
162 * [BEEP] leaks.
163 *
164 */
165
166/**
David S. Millerd179cd12005-08-17 14:57:30 -0700167 * __alloc_skb - allocate a network buffer
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168 * @size: size to allocate
169 * @gfp_mask: allocation mask
Randy Dunlapc83c2482005-10-18 22:07:41 -0700170 * @fclone: allocate from fclone cache instead of head cache
171 * and allocate a cloned (child) skb
Christoph Hellwigb30973f2006-12-06 20:32:36 -0800172 * @node: numa node to allocate memory on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 *
174 * Allocate a new &sk_buff. The returned buffer has no headroom and a
175 * tail room of size bytes. The object has a reference count of one.
176 * The return is the buffer. On a failure the return is %NULL.
177 *
178 * Buffers may only be allocated from interrupts using a @gfp_mask of
179 * %GFP_ATOMIC.
180 */
Al Virodd0fc662005-10-07 07:46:04 +0100181struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
Christoph Hellwigb30973f2006-12-06 20:32:36 -0800182 int fclone, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183{
Christoph Lametere18b8902006-12-06 20:33:20 -0800184 struct kmem_cache *cache;
Benjamin LaHaise4947d3e2006-01-03 14:06:50 -0800185 struct skb_shared_info *shinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 struct sk_buff *skb;
187 u8 *data;
188
Herbert Xu8798b3f2006-01-23 16:32:45 -0800189 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
190
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 /* Get the HEAD */
Christoph Hellwigb30973f2006-12-06 20:32:36 -0800192 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 if (!skb)
194 goto out;
195
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 size = SKB_DATA_ALIGN(size);
Christoph Hellwigb30973f2006-12-06 20:32:36 -0800197 data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info),
198 gfp_mask, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 if (!data)
200 goto nodata;
201
Arnaldo Carvalho de Meloca0605a2007-03-19 10:48:59 -0300202 /*
203 * See comment in sk_buff definition, just before the 'tail' member
204 */
205 memset(skb, 0, offsetof(struct sk_buff, tail));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 skb->truesize = size + sizeof(struct sk_buff);
207 atomic_set(&skb->users, 1);
208 skb->head = data;
209 skb->data = data;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700210 skb_reset_tail_pointer(skb);
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700211 skb->end = skb->tail + size;
Benjamin LaHaise4947d3e2006-01-03 14:06:50 -0800212 /* make sure we initialize shinfo sequentially */
213 shinfo = skb_shinfo(skb);
214 atomic_set(&shinfo->dataref, 1);
215 shinfo->nr_frags = 0;
Herbert Xu79671682006-06-22 02:40:14 -0700216 shinfo->gso_size = 0;
217 shinfo->gso_segs = 0;
218 shinfo->gso_type = 0;
Benjamin LaHaise4947d3e2006-01-03 14:06:50 -0800219 shinfo->ip6_frag_id = 0;
220 shinfo->frag_list = NULL;
221
David S. Millerd179cd12005-08-17 14:57:30 -0700222 if (fclone) {
223 struct sk_buff *child = skb + 1;
224 atomic_t *fclone_ref = (atomic_t *) (child + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225
David S. Millerd179cd12005-08-17 14:57:30 -0700226 skb->fclone = SKB_FCLONE_ORIG;
227 atomic_set(fclone_ref, 1);
228
229 child->fclone = SKB_FCLONE_UNAVAILABLE;
230 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231out:
232 return skb;
233nodata:
Herbert Xu8798b3f2006-01-23 16:32:45 -0800234 kmem_cache_free(cache, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 skb = NULL;
236 goto out;
237}
238
239/**
Christoph Hellwig8af27452006-07-31 22:35:23 -0700240 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
241 * @dev: network device to receive on
242 * @length: length to allocate
243 * @gfp_mask: get_free_pages mask, passed to alloc_skb
244 *
245 * Allocate a new &sk_buff and assign it a usage count of one. The
246 * buffer has unspecified headroom built in. Users should allocate
247 * the headroom they think they need without accounting for the
248 * built in space. The built in space is used for optimisations.
249 *
250 * %NULL is returned if there is no free memory.
251 */
252struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
253 unsigned int length, gfp_t gfp_mask)
254{
Greg Kroah-Hartman43cb76d2002-04-09 12:14:34 -0700255 int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1;
Christoph Hellwig8af27452006-07-31 22:35:23 -0700256 struct sk_buff *skb;
257
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900258 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, node);
Christoph Hellwig7b2e4972006-08-07 16:09:04 -0700259 if (likely(skb)) {
Christoph Hellwig8af27452006-07-31 22:35:23 -0700260 skb_reserve(skb, NET_SKB_PAD);
Christoph Hellwig7b2e4972006-08-07 16:09:04 -0700261 skb->dev = dev;
262 }
Christoph Hellwig8af27452006-07-31 22:35:23 -0700263 return skb;
264}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265
Herbert Xu27b437c2006-07-13 19:26:39 -0700266static void skb_drop_list(struct sk_buff **listp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267{
Herbert Xu27b437c2006-07-13 19:26:39 -0700268 struct sk_buff *list = *listp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269
Herbert Xu27b437c2006-07-13 19:26:39 -0700270 *listp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271
272 do {
273 struct sk_buff *this = list;
274 list = list->next;
275 kfree_skb(this);
276 } while (list);
277}
278
Herbert Xu27b437c2006-07-13 19:26:39 -0700279static inline void skb_drop_fraglist(struct sk_buff *skb)
280{
281 skb_drop_list(&skb_shinfo(skb)->frag_list);
282}
283
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284static void skb_clone_fraglist(struct sk_buff *skb)
285{
286 struct sk_buff *list;
287
288 for (list = skb_shinfo(skb)->frag_list; list; list = list->next)
289 skb_get(list);
290}
291
Adrian Bunk5bba1712006-06-29 13:02:35 -0700292static void skb_release_data(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293{
294 if (!skb->cloned ||
295 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
296 &skb_shinfo(skb)->dataref)) {
297 if (skb_shinfo(skb)->nr_frags) {
298 int i;
299 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
300 put_page(skb_shinfo(skb)->frags[i].page);
301 }
302
303 if (skb_shinfo(skb)->frag_list)
304 skb_drop_fraglist(skb);
305
306 kfree(skb->head);
307 }
308}
309
310/*
311 * Free an skbuff by memory without cleaning the state.
312 */
Herbert Xu2d4baff2007-11-26 23:11:19 +0800313static void kfree_skbmem(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314{
David S. Millerd179cd12005-08-17 14:57:30 -0700315 struct sk_buff *other;
316 atomic_t *fclone_ref;
317
David S. Millerd179cd12005-08-17 14:57:30 -0700318 switch (skb->fclone) {
319 case SKB_FCLONE_UNAVAILABLE:
320 kmem_cache_free(skbuff_head_cache, skb);
321 break;
322
323 case SKB_FCLONE_ORIG:
324 fclone_ref = (atomic_t *) (skb + 2);
325 if (atomic_dec_and_test(fclone_ref))
326 kmem_cache_free(skbuff_fclone_cache, skb);
327 break;
328
329 case SKB_FCLONE_CLONE:
330 fclone_ref = (atomic_t *) (skb + 1);
331 other = skb - 1;
332
333 /* The clone portion is available for
334 * fast-cloning again.
335 */
336 skb->fclone = SKB_FCLONE_UNAVAILABLE;
337
338 if (atomic_dec_and_test(fclone_ref))
339 kmem_cache_free(skbuff_fclone_cache, other);
340 break;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -0700341 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342}
343
Herbert Xu2d4baff2007-11-26 23:11:19 +0800344/* Free everything but the sk_buff shell. */
345static void skb_release_all(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 dst_release(skb->dst);
348#ifdef CONFIG_XFRM
349 secpath_put(skb->sp);
350#endif
Stephen Hemminger9c2b3322005-04-19 22:39:42 -0700351 if (skb->destructor) {
352 WARN_ON(in_irq());
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 skb->destructor(skb);
354 }
Yasuyuki Kozakai9fb9cbb2005-11-09 16:38:16 -0800355#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -0700356 nf_conntrack_put(skb->nfct);
Yasuyuki Kozakai9fb9cbb2005-11-09 16:38:16 -0800357 nf_conntrack_put_reasm(skb->nfct_reasm);
358#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359#ifdef CONFIG_BRIDGE_NETFILTER
360 nf_bridge_put(skb->nf_bridge);
361#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362/* XXX: IS this still necessary? - JHS */
363#ifdef CONFIG_NET_SCHED
364 skb->tc_index = 0;
365#ifdef CONFIG_NET_CLS_ACT
366 skb->tc_verd = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367#endif
368#endif
Herbert Xu2d4baff2007-11-26 23:11:19 +0800369 skb_release_data(skb);
370}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371
Herbert Xu2d4baff2007-11-26 23:11:19 +0800372/**
373 * __kfree_skb - private function
374 * @skb: buffer
375 *
376 * Free an sk_buff. Release anything attached to the buffer.
377 * Clean the state. This is an internal helper function. Users should
378 * always call kfree_skb
379 */
380
381void __kfree_skb(struct sk_buff *skb)
382{
383 skb_release_all(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 kfree_skbmem(skb);
385}
386
387/**
Jörn Engel231d06a2006-03-20 21:28:35 -0800388 * kfree_skb - free an sk_buff
389 * @skb: buffer to free
390 *
391 * Drop a reference to the buffer and free it if the usage count has
392 * hit zero.
393 */
394void kfree_skb(struct sk_buff *skb)
395{
396 if (unlikely(!skb))
397 return;
398 if (likely(atomic_read(&skb->users) == 1))
399 smp_rmb();
400 else if (likely(!atomic_dec_and_test(&skb->users)))
401 return;
402 __kfree_skb(skb);
403}
404
Herbert Xudec18812007-10-14 00:37:30 -0700405static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
406{
407 new->tstamp = old->tstamp;
408 new->dev = old->dev;
409 new->transport_header = old->transport_header;
410 new->network_header = old->network_header;
411 new->mac_header = old->mac_header;
412 new->dst = dst_clone(old->dst);
413#ifdef CONFIG_INET
414 new->sp = secpath_get(old->sp);
415#endif
416 memcpy(new->cb, old->cb, sizeof(old->cb));
417 new->csum_start = old->csum_start;
418 new->csum_offset = old->csum_offset;
419 new->local_df = old->local_df;
420 new->pkt_type = old->pkt_type;
421 new->ip_summed = old->ip_summed;
422 skb_copy_queue_mapping(new, old);
423 new->priority = old->priority;
424#if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
425 new->ipvs_property = old->ipvs_property;
426#endif
427 new->protocol = old->protocol;
428 new->mark = old->mark;
429 __nf_copy(new, old);
430#if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
431 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
432 new->nf_trace = old->nf_trace;
433#endif
434#ifdef CONFIG_NET_SCHED
435 new->tc_index = old->tc_index;
436#ifdef CONFIG_NET_CLS_ACT
437 new->tc_verd = old->tc_verd;
438#endif
439#endif
440 skb_copy_secmark(new, old);
441}
442
Herbert Xue0053ec2007-10-14 00:37:52 -0700443static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445#define C(x) n->x = skb->x
446
447 n->next = n->prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 n->sk = NULL;
Herbert Xudec18812007-10-14 00:37:30 -0700449 __copy_skb_header(n, skb);
450
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 C(len);
452 C(data_len);
Alexey Dobriyan3e6b3b22007-03-16 15:00:46 -0700453 C(mac_len);
Patrick McHardy334a8132007-06-25 04:35:20 -0700454 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
Paul Moore02f1c892008-01-07 21:56:41 -0800455 n->cloned = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 n->nohdr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 n->destructor = NULL;
Paul Moore02f1c892008-01-07 21:56:41 -0800458 C(iif);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 C(tail);
460 C(end);
Paul Moore02f1c892008-01-07 21:56:41 -0800461 C(head);
462 C(data);
463 C(truesize);
464 atomic_set(&n->users, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465
466 atomic_inc(&(skb_shinfo(skb)->dataref));
467 skb->cloned = 1;
468
469 return n;
Herbert Xue0053ec2007-10-14 00:37:52 -0700470#undef C
471}
472
473/**
474 * skb_morph - morph one skb into another
475 * @dst: the skb to receive the contents
476 * @src: the skb to supply the contents
477 *
478 * This is identical to skb_clone except that the target skb is
479 * supplied by the user.
480 *
481 * The target skb is returned upon exit.
482 */
483struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
484{
Herbert Xu2d4baff2007-11-26 23:11:19 +0800485 skb_release_all(dst);
Herbert Xue0053ec2007-10-14 00:37:52 -0700486 return __skb_clone(dst, src);
487}
488EXPORT_SYMBOL_GPL(skb_morph);
489
490/**
491 * skb_clone - duplicate an sk_buff
492 * @skb: buffer to clone
493 * @gfp_mask: allocation priority
494 *
495 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
496 * copies share the same packet data but not structure. The new
497 * buffer has a reference count of 1. If the allocation fails the
498 * function returns %NULL otherwise the new buffer is returned.
499 *
500 * If this function is called from an interrupt gfp_mask() must be
501 * %GFP_ATOMIC.
502 */
503
504struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
505{
506 struct sk_buff *n;
507
508 n = skb + 1;
509 if (skb->fclone == SKB_FCLONE_ORIG &&
510 n->fclone == SKB_FCLONE_UNAVAILABLE) {
511 atomic_t *fclone_ref = (atomic_t *) (n + 1);
512 n->fclone = SKB_FCLONE_CLONE;
513 atomic_inc(fclone_ref);
514 } else {
515 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
516 if (!n)
517 return NULL;
518 n->fclone = SKB_FCLONE_UNAVAILABLE;
519 }
520
521 return __skb_clone(n, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522}
523
524static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
525{
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700526#ifndef NET_SKBUFF_DATA_USES_OFFSET
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 /*
528 * Shift between the two data areas in bytes
529 */
530 unsigned long offset = new->data - old->data;
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700531#endif
Herbert Xudec18812007-10-14 00:37:30 -0700532
533 __copy_skb_header(new, old);
534
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700535#ifndef NET_SKBUFF_DATA_USES_OFFSET
536 /* {transport,network,mac}_header are relative to skb->head */
537 new->transport_header += offset;
538 new->network_header += offset;
539 new->mac_header += offset;
540#endif
Herbert Xu79671682006-06-22 02:40:14 -0700541 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
542 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
543 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544}
545
546/**
547 * skb_copy - create private copy of an sk_buff
548 * @skb: buffer to copy
549 * @gfp_mask: allocation priority
550 *
551 * Make a copy of both an &sk_buff and its data. This is used when the
552 * caller wishes to modify the data and needs a private copy of the
553 * data to alter. Returns %NULL on failure or the pointer to the buffer
554 * on success. The returned buffer has a reference count of 1.
555 *
556 * As by-product this function converts non-linear &sk_buff to linear
557 * one, so that &sk_buff becomes completely private and caller is allowed
558 * to modify all the data of returned buffer. This means that this
559 * function is not recommended for use in circumstances when only
560 * header is going to be modified. Use pskb_copy() instead.
561 */
562
Al Virodd0fc662005-10-07 07:46:04 +0100563struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564{
565 int headerlen = skb->data - skb->head;
566 /*
567 * Allocate the copy buffer
568 */
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700569 struct sk_buff *n;
570#ifdef NET_SKBUFF_DATA_USES_OFFSET
571 n = alloc_skb(skb->end + skb->data_len, gfp_mask);
572#else
573 n = alloc_skb(skb->end - skb->head + skb->data_len, gfp_mask);
574#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 if (!n)
576 return NULL;
577
578 /* Set the data pointer */
579 skb_reserve(n, headerlen);
580 /* Set the tail pointer and length */
581 skb_put(n, skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582
583 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
584 BUG();
585
586 copy_skb_header(n, skb);
587 return n;
588}
589
590
591/**
592 * pskb_copy - create copy of an sk_buff with private head.
593 * @skb: buffer to copy
594 * @gfp_mask: allocation priority
595 *
596 * Make a copy of both an &sk_buff and part of its data, located
597 * in header. Fragmented data remain shared. This is used when
598 * the caller wishes to modify only header of &sk_buff and needs
599 * private copy of the header to alter. Returns %NULL on failure
600 * or the pointer to the buffer on success.
601 * The returned buffer has a reference count of 1.
602 */
603
Al Virodd0fc662005-10-07 07:46:04 +0100604struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605{
606 /*
607 * Allocate the copy buffer
608 */
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700609 struct sk_buff *n;
610#ifdef NET_SKBUFF_DATA_USES_OFFSET
611 n = alloc_skb(skb->end, gfp_mask);
612#else
613 n = alloc_skb(skb->end - skb->head, gfp_mask);
614#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 if (!n)
616 goto out;
617
618 /* Set the data pointer */
619 skb_reserve(n, skb->data - skb->head);
620 /* Set the tail pointer and length */
621 skb_put(n, skb_headlen(skb));
622 /* Copy the bytes */
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -0300623 skb_copy_from_linear_data(skb, n->data, n->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624
Herbert Xu25f484a2006-11-07 14:57:15 -0800625 n->truesize += skb->data_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 n->data_len = skb->data_len;
627 n->len = skb->len;
628
629 if (skb_shinfo(skb)->nr_frags) {
630 int i;
631
632 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
633 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
634 get_page(skb_shinfo(n)->frags[i].page);
635 }
636 skb_shinfo(n)->nr_frags = i;
637 }
638
639 if (skb_shinfo(skb)->frag_list) {
640 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
641 skb_clone_fraglist(n);
642 }
643
644 copy_skb_header(n, skb);
645out:
646 return n;
647}
648
649/**
650 * pskb_expand_head - reallocate header of &sk_buff
651 * @skb: buffer to reallocate
652 * @nhead: room to add at head
653 * @ntail: room to add at tail
654 * @gfp_mask: allocation priority
655 *
656 * Expands (or creates identical copy, if &nhead and &ntail are zero)
657 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
658 * reference count of 1. Returns zero in the case of success or error,
659 * if expansion failed. In the last case, &sk_buff is not changed.
660 *
661 * All the pointers pointing into skb header may change and must be
662 * reloaded after call to this function.
663 */
664
Victor Fusco86a76ca2005-07-08 14:57:47 -0700665int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
Al Virodd0fc662005-10-07 07:46:04 +0100666 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667{
668 int i;
669 u8 *data;
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700670#ifdef NET_SKBUFF_DATA_USES_OFFSET
671 int size = nhead + skb->end + ntail;
672#else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 int size = nhead + (skb->end - skb->head) + ntail;
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700674#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 long off;
676
677 if (skb_shared(skb))
678 BUG();
679
680 size = SKB_DATA_ALIGN(size);
681
682 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
683 if (!data)
684 goto nodata;
685
686 /* Copy only real data... and, alas, header. This should be
687 * optimized for the cases when header is void. */
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700688#ifdef NET_SKBUFF_DATA_USES_OFFSET
Mikael Petterssonb6ccc672007-05-19 13:55:25 -0700689 memcpy(data + nhead, skb->head, skb->tail);
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700690#else
Mikael Petterssonb6ccc672007-05-19 13:55:25 -0700691 memcpy(data + nhead, skb->head, skb->tail - skb->head);
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700692#endif
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700693 memcpy(data + size, skb_end_pointer(skb),
694 sizeof(struct skb_shared_info));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695
696 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
697 get_page(skb_shinfo(skb)->frags[i].page);
698
699 if (skb_shinfo(skb)->frag_list)
700 skb_clone_fraglist(skb);
701
702 skb_release_data(skb);
703
704 off = (data + nhead) - skb->head;
705
706 skb->head = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 skb->data += off;
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700708#ifdef NET_SKBUFF_DATA_USES_OFFSET
709 skb->end = size;
Patrick McHardy56eb8882007-04-09 11:45:04 -0700710 off = nhead;
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700711#else
712 skb->end = skb->head + size;
Patrick McHardy56eb8882007-04-09 11:45:04 -0700713#endif
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700714 /* {transport,network,mac}_header and tail are relative to skb->head */
715 skb->tail += off;
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700716 skb->transport_header += off;
717 skb->network_header += off;
718 skb->mac_header += off;
Herbert Xu172a8632007-10-15 01:46:08 -0700719 skb->csum_start += nhead;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 skb->cloned = 0;
Patrick McHardy334a8132007-06-25 04:35:20 -0700721 skb->hdr_len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 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.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 */
769struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
Victor Fusco86a76ca2005-07-08 14:57:47 -0700770 int newheadroom, int newtailroom,
Al Virodd0fc662005-10-07 07:46:04 +0100771 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772{
773 /*
774 * Allocate the copy buffer
775 */
776 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
777 gfp_mask);
Patrick McHardyefd1e8d2007-04-10 18:30:09 -0700778 int oldheadroom = skb_headroom(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 int head_copy_len, head_copy_off;
Herbert Xu52886052007-09-16 16:32:11 -0700780 int off;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781
782 if (!n)
783 return NULL;
784
785 skb_reserve(n, newheadroom);
786
787 /* Set the tail pointer and length */
788 skb_put(n, skb->len);
789
Patrick McHardyefd1e8d2007-04-10 18:30:09 -0700790 head_copy_len = oldheadroom;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 head_copy_off = 0;
792 if (newheadroom <= head_copy_len)
793 head_copy_len = newheadroom;
794 else
795 head_copy_off = newheadroom - head_copy_len;
796
797 /* Copy the linear header and data. */
798 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
799 skb->len + head_copy_len))
800 BUG();
801
802 copy_skb_header(n, skb);
803
Patrick McHardyefd1e8d2007-04-10 18:30:09 -0700804 off = newheadroom - oldheadroom;
Herbert Xu52886052007-09-16 16:32:11 -0700805 n->csum_start += off;
806#ifdef NET_SKBUFF_DATA_USES_OFFSET
Patrick McHardyefd1e8d2007-04-10 18:30:09 -0700807 n->transport_header += off;
808 n->network_header += off;
809 n->mac_header += off;
Herbert Xu52886052007-09-16 16:32:11 -0700810#endif
Patrick McHardyefd1e8d2007-04-10 18:30:09 -0700811
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 return n;
813}
814
815/**
816 * skb_pad - zero pad the tail of an skb
817 * @skb: buffer to pad
818 * @pad: space to pad
819 *
820 * Ensure that a buffer is followed by a padding area that is zero
821 * filled. Used by network drivers which may DMA or transfer data
822 * beyond the buffer end onto the wire.
823 *
Herbert Xu5b057c62006-06-23 02:06:41 -0700824 * May return error in out of memory cases. The skb is freed on error.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 */
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900826
Herbert Xu5b057c62006-06-23 02:06:41 -0700827int skb_pad(struct sk_buff *skb, int pad)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828{
Herbert Xu5b057c62006-06-23 02:06:41 -0700829 int err;
830 int ntail;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900831
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 /* If the skbuff is non linear tailroom is always zero.. */
Herbert Xu5b057c62006-06-23 02:06:41 -0700833 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 memset(skb->data+skb->len, 0, pad);
Herbert Xu5b057c62006-06-23 02:06:41 -0700835 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 }
Herbert Xu5b057c62006-06-23 02:06:41 -0700837
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700838 ntail = skb->data_len + pad - (skb->end - skb->tail);
Herbert Xu5b057c62006-06-23 02:06:41 -0700839 if (likely(skb_cloned(skb) || ntail > 0)) {
840 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
841 if (unlikely(err))
842 goto free_skb;
843 }
844
845 /* FIXME: The use of this function with non-linear skb's really needs
846 * to be audited.
847 */
848 err = skb_linearize(skb);
849 if (unlikely(err))
850 goto free_skb;
851
852 memset(skb->data + skb->len, 0, pad);
853 return 0;
854
855free_skb:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 kfree_skb(skb);
Herbert Xu5b057c62006-06-23 02:06:41 -0700857 return err;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900858}
859
Herbert Xu3cc0e872006-06-09 16:13:38 -0700860/* Trims skb to length len. It can change skb pointers.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 */
862
Herbert Xu3cc0e872006-06-09 16:13:38 -0700863int ___pskb_trim(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864{
Herbert Xu27b437c2006-07-13 19:26:39 -0700865 struct sk_buff **fragp;
866 struct sk_buff *frag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 int offset = skb_headlen(skb);
868 int nfrags = skb_shinfo(skb)->nr_frags;
869 int i;
Herbert Xu27b437c2006-07-13 19:26:39 -0700870 int err;
871
872 if (skb_cloned(skb) &&
873 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
874 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875
Herbert Xuf4d26fb2006-07-30 20:20:28 -0700876 i = 0;
877 if (offset >= len)
878 goto drop_pages;
879
880 for (; i < nfrags; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 int end = offset + skb_shinfo(skb)->frags[i].size;
Herbert Xu27b437c2006-07-13 19:26:39 -0700882
883 if (end < len) {
884 offset = end;
885 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 }
Herbert Xu27b437c2006-07-13 19:26:39 -0700887
Herbert Xuf4d26fb2006-07-30 20:20:28 -0700888 skb_shinfo(skb)->frags[i++].size = len - offset;
Herbert Xu27b437c2006-07-13 19:26:39 -0700889
Herbert Xuf4d26fb2006-07-30 20:20:28 -0700890drop_pages:
Herbert Xu27b437c2006-07-13 19:26:39 -0700891 skb_shinfo(skb)->nr_frags = i;
892
893 for (; i < nfrags; i++)
894 put_page(skb_shinfo(skb)->frags[i].page);
895
896 if (skb_shinfo(skb)->frag_list)
897 skb_drop_fraglist(skb);
Herbert Xuf4d26fb2006-07-30 20:20:28 -0700898 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 }
900
Herbert Xu27b437c2006-07-13 19:26:39 -0700901 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
902 fragp = &frag->next) {
903 int end = offset + frag->len;
904
905 if (skb_shared(frag)) {
906 struct sk_buff *nfrag;
907
908 nfrag = skb_clone(frag, GFP_ATOMIC);
909 if (unlikely(!nfrag))
910 return -ENOMEM;
911
912 nfrag->next = frag->next;
Herbert Xuf4d26fb2006-07-30 20:20:28 -0700913 kfree_skb(frag);
Herbert Xu27b437c2006-07-13 19:26:39 -0700914 frag = nfrag;
915 *fragp = frag;
916 }
917
918 if (end < len) {
919 offset = end;
920 continue;
921 }
922
923 if (end > len &&
924 unlikely((err = pskb_trim(frag, len - offset))))
925 return err;
926
927 if (frag->next)
928 skb_drop_list(&frag->next);
929 break;
930 }
931
Herbert Xuf4d26fb2006-07-30 20:20:28 -0700932done:
Herbert Xu27b437c2006-07-13 19:26:39 -0700933 if (len > skb_headlen(skb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 skb->data_len -= skb->len - len;
935 skb->len = len;
936 } else {
Herbert Xu27b437c2006-07-13 19:26:39 -0700937 skb->len = len;
938 skb->data_len = 0;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700939 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 }
941
942 return 0;
943}
944
945/**
946 * __pskb_pull_tail - advance tail of skb header
947 * @skb: buffer to reallocate
948 * @delta: number of bytes to advance tail
949 *
950 * The function makes a sense only on a fragmented &sk_buff,
951 * it expands header moving its tail forward and copying necessary
952 * data from fragmented part.
953 *
954 * &sk_buff MUST have reference count of 1.
955 *
956 * Returns %NULL (and &sk_buff does not change) if pull failed
957 * or value of new tail of skb in the case of success.
958 *
959 * All the pointers pointing into skb header may change and must be
960 * reloaded after call to this function.
961 */
962
963/* Moves tail of skb head forward, copying data from fragmented part,
964 * when it is necessary.
965 * 1. It may fail due to malloc failure.
966 * 2. It may change skb pointers.
967 *
968 * It is pretty complicated. Luckily, it is called only in exceptional cases.
969 */
970unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
971{
972 /* If skb has not enough free space at tail, get new one
973 * plus 128 bytes for future expansions. If we have enough
974 * room at tail, reallocate without expansion only if skb is cloned.
975 */
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700976 int i, k, eat = (skb->tail + delta) - skb->end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977
978 if (eat > 0 || skb_cloned(skb)) {
979 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
980 GFP_ATOMIC))
981 return NULL;
982 }
983
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700984 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 BUG();
986
987 /* Optimization: no fragments, no reasons to preestimate
988 * size of pulled pages. Superb.
989 */
990 if (!skb_shinfo(skb)->frag_list)
991 goto pull_pages;
992
993 /* Estimate size of pulled pages. */
994 eat = delta;
995 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
996 if (skb_shinfo(skb)->frags[i].size >= eat)
997 goto pull_pages;
998 eat -= skb_shinfo(skb)->frags[i].size;
999 }
1000
1001 /* If we need update frag list, we are in troubles.
1002 * Certainly, it possible to add an offset to skb data,
1003 * but taking into account that pulling is expected to
1004 * be very rare operation, it is worth to fight against
1005 * further bloating skb head and crucify ourselves here instead.
1006 * Pure masohism, indeed. 8)8)
1007 */
1008 if (eat) {
1009 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1010 struct sk_buff *clone = NULL;
1011 struct sk_buff *insp = NULL;
1012
1013 do {
Kris Katterjohn09a62662006-01-08 22:24:28 -08001014 BUG_ON(!list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015
1016 if (list->len <= eat) {
1017 /* Eaten as whole. */
1018 eat -= list->len;
1019 list = list->next;
1020 insp = list;
1021 } else {
1022 /* Eaten partially. */
1023
1024 if (skb_shared(list)) {
1025 /* Sucks! We need to fork list. :-( */
1026 clone = skb_clone(list, GFP_ATOMIC);
1027 if (!clone)
1028 return NULL;
1029 insp = list->next;
1030 list = clone;
1031 } else {
1032 /* This may be pulled without
1033 * problems. */
1034 insp = list;
1035 }
1036 if (!pskb_pull(list, eat)) {
1037 if (clone)
1038 kfree_skb(clone);
1039 return NULL;
1040 }
1041 break;
1042 }
1043 } while (eat);
1044
1045 /* Free pulled out fragments. */
1046 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1047 skb_shinfo(skb)->frag_list = list->next;
1048 kfree_skb(list);
1049 }
1050 /* And insert new clone at head. */
1051 if (clone) {
1052 clone->next = list;
1053 skb_shinfo(skb)->frag_list = clone;
1054 }
1055 }
1056 /* Success! Now we may commit changes to skb data. */
1057
1058pull_pages:
1059 eat = delta;
1060 k = 0;
1061 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1062 if (skb_shinfo(skb)->frags[i].size <= eat) {
1063 put_page(skb_shinfo(skb)->frags[i].page);
1064 eat -= skb_shinfo(skb)->frags[i].size;
1065 } else {
1066 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1067 if (eat) {
1068 skb_shinfo(skb)->frags[k].page_offset += eat;
1069 skb_shinfo(skb)->frags[k].size -= eat;
1070 eat = 0;
1071 }
1072 k++;
1073 }
1074 }
1075 skb_shinfo(skb)->nr_frags = k;
1076
1077 skb->tail += delta;
1078 skb->data_len -= delta;
1079
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001080 return skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081}
1082
1083/* Copy some data bits from skb to kernel buffer. */
1084
1085int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1086{
1087 int i, copy;
David S. Miller1a028e52007-04-27 15:21:23 -07001088 int start = skb_headlen(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089
1090 if (offset > (int)skb->len - len)
1091 goto fault;
1092
1093 /* Copy header. */
David S. Miller1a028e52007-04-27 15:21:23 -07001094 if ((copy = start - offset) > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 if (copy > len)
1096 copy = len;
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03001097 skb_copy_from_linear_data_offset(skb, offset, to, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 if ((len -= copy) == 0)
1099 return 0;
1100 offset += copy;
1101 to += copy;
1102 }
1103
1104 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07001105 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106
David S. Miller1a028e52007-04-27 15:21:23 -07001107 BUG_TRAP(start <= offset + len);
1108
1109 end = start + skb_shinfo(skb)->frags[i].size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 if ((copy = end - offset) > 0) {
1111 u8 *vaddr;
1112
1113 if (copy > len)
1114 copy = len;
1115
1116 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1117 memcpy(to,
David S. Miller1a028e52007-04-27 15:21:23 -07001118 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1119 offset - start, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 kunmap_skb_frag(vaddr);
1121
1122 if ((len -= copy) == 0)
1123 return 0;
1124 offset += copy;
1125 to += copy;
1126 }
David S. Miller1a028e52007-04-27 15:21:23 -07001127 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 }
1129
1130 if (skb_shinfo(skb)->frag_list) {
1131 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1132
1133 for (; list; list = list->next) {
David S. Miller1a028e52007-04-27 15:21:23 -07001134 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135
David S. Miller1a028e52007-04-27 15:21:23 -07001136 BUG_TRAP(start <= offset + len);
1137
1138 end = start + list->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 if ((copy = end - offset) > 0) {
1140 if (copy > len)
1141 copy = len;
David S. Miller1a028e52007-04-27 15:21:23 -07001142 if (skb_copy_bits(list, offset - start,
1143 to, copy))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 goto fault;
1145 if ((len -= copy) == 0)
1146 return 0;
1147 offset += copy;
1148 to += copy;
1149 }
David S. Miller1a028e52007-04-27 15:21:23 -07001150 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 }
1152 }
1153 if (!len)
1154 return 0;
1155
1156fault:
1157 return -EFAULT;
1158}
1159
Jens Axboe9c55e012007-11-06 23:30:13 -08001160/*
1161 * Callback from splice_to_pipe(), if we need to release some pages
1162 * at the end of the spd in case we error'ed out in filling the pipe.
1163 */
1164static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1165{
1166 struct sk_buff *skb = (struct sk_buff *) spd->partial[i].private;
1167
1168 kfree_skb(skb);
1169}
1170
1171/*
1172 * Fill page/offset/length into spd, if it can hold more pages.
1173 */
1174static inline int spd_fill_page(struct splice_pipe_desc *spd, struct page *page,
1175 unsigned int len, unsigned int offset,
1176 struct sk_buff *skb)
1177{
1178 if (unlikely(spd->nr_pages == PIPE_BUFFERS))
1179 return 1;
1180
1181 spd->pages[spd->nr_pages] = page;
1182 spd->partial[spd->nr_pages].len = len;
1183 spd->partial[spd->nr_pages].offset = offset;
1184 spd->partial[spd->nr_pages].private = (unsigned long) skb_get(skb);
1185 spd->nr_pages++;
1186 return 0;
1187}
1188
1189/*
1190 * Map linear and fragment data from the skb to spd. Returns number of
1191 * pages mapped.
1192 */
1193static int __skb_splice_bits(struct sk_buff *skb, unsigned int *offset,
1194 unsigned int *total_len,
1195 struct splice_pipe_desc *spd)
1196{
1197 unsigned int nr_pages = spd->nr_pages;
1198 unsigned int poff, plen, len, toff, tlen;
1199 int headlen, seg;
1200
1201 toff = *offset;
1202 tlen = *total_len;
1203 if (!tlen)
1204 goto err;
1205
1206 /*
1207 * if the offset is greater than the linear part, go directly to
1208 * the fragments.
1209 */
1210 headlen = skb_headlen(skb);
1211 if (toff >= headlen) {
1212 toff -= headlen;
1213 goto map_frag;
1214 }
1215
1216 /*
1217 * first map the linear region into the pages/partial map, skipping
1218 * any potential initial offset.
1219 */
1220 len = 0;
1221 while (len < headlen) {
1222 void *p = skb->data + len;
1223
1224 poff = (unsigned long) p & (PAGE_SIZE - 1);
1225 plen = min_t(unsigned int, headlen - len, PAGE_SIZE - poff);
1226 len += plen;
1227
1228 if (toff) {
1229 if (plen <= toff) {
1230 toff -= plen;
1231 continue;
1232 }
1233 plen -= toff;
1234 poff += toff;
1235 toff = 0;
1236 }
1237
1238 plen = min(plen, tlen);
1239 if (!plen)
1240 break;
1241
1242 /*
1243 * just jump directly to update and return, no point
1244 * in going over fragments when the output is full.
1245 */
1246 if (spd_fill_page(spd, virt_to_page(p), plen, poff, skb))
1247 goto done;
1248
1249 tlen -= plen;
1250 }
1251
1252 /*
1253 * then map the fragments
1254 */
1255map_frag:
1256 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1257 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1258
1259 plen = f->size;
1260 poff = f->page_offset;
1261
1262 if (toff) {
1263 if (plen <= toff) {
1264 toff -= plen;
1265 continue;
1266 }
1267 plen -= toff;
1268 poff += toff;
1269 toff = 0;
1270 }
1271
1272 plen = min(plen, tlen);
1273 if (!plen)
1274 break;
1275
1276 if (spd_fill_page(spd, f->page, plen, poff, skb))
1277 break;
1278
1279 tlen -= plen;
1280 }
1281
1282done:
1283 if (spd->nr_pages - nr_pages) {
1284 *offset = 0;
1285 *total_len = tlen;
1286 return 0;
1287 }
1288err:
1289 return 1;
1290}
1291
1292/*
1293 * Map data from the skb to a pipe. Should handle both the linear part,
1294 * the fragments, and the frag list. It does NOT handle frag lists within
1295 * the frag list, if such a thing exists. We'd probably need to recurse to
1296 * handle that cleanly.
1297 */
1298int skb_splice_bits(struct sk_buff *__skb, unsigned int offset,
1299 struct pipe_inode_info *pipe, unsigned int tlen,
1300 unsigned int flags)
1301{
1302 struct partial_page partial[PIPE_BUFFERS];
1303 struct page *pages[PIPE_BUFFERS];
1304 struct splice_pipe_desc spd = {
1305 .pages = pages,
1306 .partial = partial,
1307 .flags = flags,
1308 .ops = &sock_pipe_buf_ops,
1309 .spd_release = sock_spd_release,
1310 };
1311 struct sk_buff *skb;
1312
1313 /*
1314 * I'd love to avoid the clone here, but tcp_read_sock()
1315 * ignores reference counts and unconditonally kills the sk_buff
1316 * on return from the actor.
1317 */
1318 skb = skb_clone(__skb, GFP_KERNEL);
1319 if (unlikely(!skb))
1320 return -ENOMEM;
1321
1322 /*
1323 * __skb_splice_bits() only fails if the output has no room left,
1324 * so no point in going over the frag_list for the error case.
1325 */
1326 if (__skb_splice_bits(skb, &offset, &tlen, &spd))
1327 goto done;
1328 else if (!tlen)
1329 goto done;
1330
1331 /*
1332 * now see if we have a frag_list to map
1333 */
1334 if (skb_shinfo(skb)->frag_list) {
1335 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1336
1337 for (; list && tlen; list = list->next) {
1338 if (__skb_splice_bits(list, &offset, &tlen, &spd))
1339 break;
1340 }
1341 }
1342
1343done:
1344 /*
1345 * drop our reference to the clone, the pipe consumption will
1346 * drop the rest.
1347 */
1348 kfree_skb(skb);
1349
1350 if (spd.nr_pages) {
1351 int ret;
1352
1353 /*
1354 * Drop the socket lock, otherwise we have reverse
1355 * locking dependencies between sk_lock and i_mutex
1356 * here as compared to sendfile(). We enter here
1357 * with the socket lock held, and splice_to_pipe() will
1358 * grab the pipe inode lock. For sendfile() emulation,
1359 * we call into ->sendpage() with the i_mutex lock held
1360 * and networking will grab the socket lock.
1361 */
1362 release_sock(__skb->sk);
1363 ret = splice_to_pipe(pipe, &spd);
1364 lock_sock(__skb->sk);
1365 return ret;
1366 }
1367
1368 return 0;
1369}
1370
Herbert Xu357b40a2005-04-19 22:30:14 -07001371/**
1372 * skb_store_bits - store bits from kernel buffer to skb
1373 * @skb: destination buffer
1374 * @offset: offset in destination
1375 * @from: source buffer
1376 * @len: number of bytes to copy
1377 *
1378 * Copy the specified number of bytes from the source buffer to the
1379 * destination skb. This function handles all the messy bits of
1380 * traversing fragment lists and such.
1381 */
1382
Stephen Hemminger0c6fcc82007-04-20 16:40:01 -07001383int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
Herbert Xu357b40a2005-04-19 22:30:14 -07001384{
1385 int i, copy;
David S. Miller1a028e52007-04-27 15:21:23 -07001386 int start = skb_headlen(skb);
Herbert Xu357b40a2005-04-19 22:30:14 -07001387
1388 if (offset > (int)skb->len - len)
1389 goto fault;
1390
David S. Miller1a028e52007-04-27 15:21:23 -07001391 if ((copy = start - offset) > 0) {
Herbert Xu357b40a2005-04-19 22:30:14 -07001392 if (copy > len)
1393 copy = len;
Arnaldo Carvalho de Melo27d7ff42007-03-31 11:55:19 -03001394 skb_copy_to_linear_data_offset(skb, offset, from, copy);
Herbert Xu357b40a2005-04-19 22:30:14 -07001395 if ((len -= copy) == 0)
1396 return 0;
1397 offset += copy;
1398 from += copy;
1399 }
1400
1401 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1402 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
David S. Miller1a028e52007-04-27 15:21:23 -07001403 int end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001404
David S. Miller1a028e52007-04-27 15:21:23 -07001405 BUG_TRAP(start <= offset + len);
1406
1407 end = start + frag->size;
Herbert Xu357b40a2005-04-19 22:30:14 -07001408 if ((copy = end - offset) > 0) {
1409 u8 *vaddr;
1410
1411 if (copy > len)
1412 copy = len;
1413
1414 vaddr = kmap_skb_frag(frag);
David S. Miller1a028e52007-04-27 15:21:23 -07001415 memcpy(vaddr + frag->page_offset + offset - start,
1416 from, copy);
Herbert Xu357b40a2005-04-19 22:30:14 -07001417 kunmap_skb_frag(vaddr);
1418
1419 if ((len -= copy) == 0)
1420 return 0;
1421 offset += copy;
1422 from += copy;
1423 }
David S. Miller1a028e52007-04-27 15:21:23 -07001424 start = end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001425 }
1426
1427 if (skb_shinfo(skb)->frag_list) {
1428 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1429
1430 for (; list; list = list->next) {
David S. Miller1a028e52007-04-27 15:21:23 -07001431 int end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001432
David S. Miller1a028e52007-04-27 15:21:23 -07001433 BUG_TRAP(start <= offset + len);
1434
1435 end = start + list->len;
Herbert Xu357b40a2005-04-19 22:30:14 -07001436 if ((copy = end - offset) > 0) {
1437 if (copy > len)
1438 copy = len;
David S. Miller1a028e52007-04-27 15:21:23 -07001439 if (skb_store_bits(list, offset - start,
1440 from, copy))
Herbert Xu357b40a2005-04-19 22:30:14 -07001441 goto fault;
1442 if ((len -= copy) == 0)
1443 return 0;
1444 offset += copy;
1445 from += copy;
1446 }
David S. Miller1a028e52007-04-27 15:21:23 -07001447 start = end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001448 }
1449 }
1450 if (!len)
1451 return 0;
1452
1453fault:
1454 return -EFAULT;
1455}
1456
1457EXPORT_SYMBOL(skb_store_bits);
1458
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459/* Checksum skb data. */
1460
Al Viro2bbbc862006-11-14 21:37:14 -08001461__wsum skb_checksum(const struct sk_buff *skb, int offset,
1462 int len, __wsum csum)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463{
David S. Miller1a028e52007-04-27 15:21:23 -07001464 int start = skb_headlen(skb);
1465 int i, copy = start - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 int pos = 0;
1467
1468 /* Checksum header. */
1469 if (copy > 0) {
1470 if (copy > len)
1471 copy = len;
1472 csum = csum_partial(skb->data + offset, copy, csum);
1473 if ((len -= copy) == 0)
1474 return csum;
1475 offset += copy;
1476 pos = copy;
1477 }
1478
1479 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07001480 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481
David S. Miller1a028e52007-04-27 15:21:23 -07001482 BUG_TRAP(start <= offset + len);
1483
1484 end = start + skb_shinfo(skb)->frags[i].size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 if ((copy = end - offset) > 0) {
Al Viro44bb9362006-11-14 21:36:14 -08001486 __wsum csum2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 u8 *vaddr;
1488 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1489
1490 if (copy > len)
1491 copy = len;
1492 vaddr = kmap_skb_frag(frag);
David S. Miller1a028e52007-04-27 15:21:23 -07001493 csum2 = csum_partial(vaddr + frag->page_offset +
1494 offset - start, copy, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 kunmap_skb_frag(vaddr);
1496 csum = csum_block_add(csum, csum2, pos);
1497 if (!(len -= copy))
1498 return csum;
1499 offset += copy;
1500 pos += copy;
1501 }
David S. Miller1a028e52007-04-27 15:21:23 -07001502 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 }
1504
1505 if (skb_shinfo(skb)->frag_list) {
1506 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1507
1508 for (; list; list = list->next) {
David S. Miller1a028e52007-04-27 15:21:23 -07001509 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510
David S. Miller1a028e52007-04-27 15:21:23 -07001511 BUG_TRAP(start <= offset + len);
1512
1513 end = start + list->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 if ((copy = end - offset) > 0) {
Al Viro5f92a732006-11-14 21:36:54 -08001515 __wsum csum2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 if (copy > len)
1517 copy = len;
David S. Miller1a028e52007-04-27 15:21:23 -07001518 csum2 = skb_checksum(list, offset - start,
1519 copy, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 csum = csum_block_add(csum, csum2, pos);
1521 if ((len -= copy) == 0)
1522 return csum;
1523 offset += copy;
1524 pos += copy;
1525 }
David S. Miller1a028e52007-04-27 15:21:23 -07001526 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 }
1528 }
Kris Katterjohn09a62662006-01-08 22:24:28 -08001529 BUG_ON(len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530
1531 return csum;
1532}
1533
1534/* Both of above in one bottle. */
1535
Al Viro81d77662006-11-14 21:37:33 -08001536__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1537 u8 *to, int len, __wsum csum)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538{
David S. Miller1a028e52007-04-27 15:21:23 -07001539 int start = skb_headlen(skb);
1540 int i, copy = start - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 int pos = 0;
1542
1543 /* Copy header. */
1544 if (copy > 0) {
1545 if (copy > len)
1546 copy = len;
1547 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1548 copy, csum);
1549 if ((len -= copy) == 0)
1550 return csum;
1551 offset += copy;
1552 to += copy;
1553 pos = copy;
1554 }
1555
1556 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07001557 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558
David S. Miller1a028e52007-04-27 15:21:23 -07001559 BUG_TRAP(start <= offset + len);
1560
1561 end = start + skb_shinfo(skb)->frags[i].size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 if ((copy = end - offset) > 0) {
Al Viro50842052006-11-14 21:36:34 -08001563 __wsum csum2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 u8 *vaddr;
1565 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1566
1567 if (copy > len)
1568 copy = len;
1569 vaddr = kmap_skb_frag(frag);
1570 csum2 = csum_partial_copy_nocheck(vaddr +
David S. Miller1a028e52007-04-27 15:21:23 -07001571 frag->page_offset +
1572 offset - start, to,
1573 copy, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 kunmap_skb_frag(vaddr);
1575 csum = csum_block_add(csum, csum2, pos);
1576 if (!(len -= copy))
1577 return csum;
1578 offset += copy;
1579 to += copy;
1580 pos += copy;
1581 }
David S. Miller1a028e52007-04-27 15:21:23 -07001582 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 }
1584
1585 if (skb_shinfo(skb)->frag_list) {
1586 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1587
1588 for (; list; list = list->next) {
Al Viro81d77662006-11-14 21:37:33 -08001589 __wsum csum2;
David S. Miller1a028e52007-04-27 15:21:23 -07001590 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591
David S. Miller1a028e52007-04-27 15:21:23 -07001592 BUG_TRAP(start <= offset + len);
1593
1594 end = start + list->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 if ((copy = end - offset) > 0) {
1596 if (copy > len)
1597 copy = len;
David S. Miller1a028e52007-04-27 15:21:23 -07001598 csum2 = skb_copy_and_csum_bits(list,
1599 offset - start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 to, copy, 0);
1601 csum = csum_block_add(csum, csum2, pos);
1602 if ((len -= copy) == 0)
1603 return csum;
1604 offset += copy;
1605 to += copy;
1606 pos += copy;
1607 }
David S. Miller1a028e52007-04-27 15:21:23 -07001608 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 }
1610 }
Kris Katterjohn09a62662006-01-08 22:24:28 -08001611 BUG_ON(len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 return csum;
1613}
1614
1615void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1616{
Al Virod3bc23e2006-11-14 21:24:49 -08001617 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 long csstart;
1619
Patrick McHardy84fa7932006-08-29 16:44:56 -07001620 if (skb->ip_summed == CHECKSUM_PARTIAL)
Herbert Xu663ead32007-04-09 11:59:07 -07001621 csstart = skb->csum_start - skb_headroom(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 else
1623 csstart = skb_headlen(skb);
1624
Kris Katterjohn09a62662006-01-08 22:24:28 -08001625 BUG_ON(csstart > skb_headlen(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03001627 skb_copy_from_linear_data(skb, to, csstart);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628
1629 csum = 0;
1630 if (csstart != skb->len)
1631 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1632 skb->len - csstart, 0);
1633
Patrick McHardy84fa7932006-08-29 16:44:56 -07001634 if (skb->ip_summed == CHECKSUM_PARTIAL) {
Al Viroff1dcad2006-11-20 18:07:29 -08001635 long csstuff = csstart + skb->csum_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636
Al Virod3bc23e2006-11-14 21:24:49 -08001637 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 }
1639}
1640
1641/**
1642 * skb_dequeue - remove from the head of the queue
1643 * @list: list to dequeue from
1644 *
1645 * Remove the head of the list. The list lock is taken so the function
1646 * may be used safely with other locking list functions. The head item is
1647 * returned or %NULL if the list is empty.
1648 */
1649
1650struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1651{
1652 unsigned long flags;
1653 struct sk_buff *result;
1654
1655 spin_lock_irqsave(&list->lock, flags);
1656 result = __skb_dequeue(list);
1657 spin_unlock_irqrestore(&list->lock, flags);
1658 return result;
1659}
1660
1661/**
1662 * skb_dequeue_tail - remove from the tail of the queue
1663 * @list: list to dequeue from
1664 *
1665 * Remove the tail of the list. The list lock is taken so the function
1666 * may be used safely with other locking list functions. The tail item is
1667 * returned or %NULL if the list is empty.
1668 */
1669struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1670{
1671 unsigned long flags;
1672 struct sk_buff *result;
1673
1674 spin_lock_irqsave(&list->lock, flags);
1675 result = __skb_dequeue_tail(list);
1676 spin_unlock_irqrestore(&list->lock, flags);
1677 return result;
1678}
1679
1680/**
1681 * skb_queue_purge - empty a list
1682 * @list: list to empty
1683 *
1684 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1685 * the list and one reference dropped. This function takes the list
1686 * lock and is atomic with respect to other list locking functions.
1687 */
1688void skb_queue_purge(struct sk_buff_head *list)
1689{
1690 struct sk_buff *skb;
1691 while ((skb = skb_dequeue(list)) != NULL)
1692 kfree_skb(skb);
1693}
1694
1695/**
1696 * skb_queue_head - queue a buffer at the list head
1697 * @list: list to use
1698 * @newsk: buffer to queue
1699 *
1700 * Queue a buffer at the start of the list. This function takes the
1701 * list lock and can be used safely with other locking &sk_buff functions
1702 * safely.
1703 *
1704 * A buffer cannot be placed on two lists at the same time.
1705 */
1706void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1707{
1708 unsigned long flags;
1709
1710 spin_lock_irqsave(&list->lock, flags);
1711 __skb_queue_head(list, newsk);
1712 spin_unlock_irqrestore(&list->lock, flags);
1713}
1714
1715/**
1716 * skb_queue_tail - queue a buffer at the list tail
1717 * @list: list to use
1718 * @newsk: buffer to queue
1719 *
1720 * Queue a buffer at the tail of the list. This function takes the
1721 * list lock and can be used safely with other locking &sk_buff functions
1722 * safely.
1723 *
1724 * A buffer cannot be placed on two lists at the same time.
1725 */
1726void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1727{
1728 unsigned long flags;
1729
1730 spin_lock_irqsave(&list->lock, flags);
1731 __skb_queue_tail(list, newsk);
1732 spin_unlock_irqrestore(&list->lock, flags);
1733}
David S. Miller8728b832005-08-09 19:25:21 -07001734
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735/**
1736 * skb_unlink - remove a buffer from a list
1737 * @skb: buffer to remove
David S. Miller8728b832005-08-09 19:25:21 -07001738 * @list: list to use
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 *
David S. Miller8728b832005-08-09 19:25:21 -07001740 * Remove a packet from a list. The list locks are taken and this
1741 * function is atomic with respect to other list locked calls
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 *
David S. Miller8728b832005-08-09 19:25:21 -07001743 * You must know what list the SKB is on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 */
David S. Miller8728b832005-08-09 19:25:21 -07001745void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746{
David S. Miller8728b832005-08-09 19:25:21 -07001747 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748
David S. Miller8728b832005-08-09 19:25:21 -07001749 spin_lock_irqsave(&list->lock, flags);
1750 __skb_unlink(skb, list);
1751 spin_unlock_irqrestore(&list->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752}
1753
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754/**
1755 * skb_append - append a buffer
1756 * @old: buffer to insert after
1757 * @newsk: buffer to insert
David S. Miller8728b832005-08-09 19:25:21 -07001758 * @list: list to use
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 *
1760 * Place a packet after a given packet in a list. The list locks are taken
1761 * and this function is atomic with respect to other list locked calls.
1762 * A buffer cannot be placed on two lists at the same time.
1763 */
David S. Miller8728b832005-08-09 19:25:21 -07001764void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765{
1766 unsigned long flags;
1767
David S. Miller8728b832005-08-09 19:25:21 -07001768 spin_lock_irqsave(&list->lock, flags);
1769 __skb_append(old, newsk, list);
1770 spin_unlock_irqrestore(&list->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771}
1772
1773
1774/**
1775 * skb_insert - insert a buffer
1776 * @old: buffer to insert before
1777 * @newsk: buffer to insert
David S. Miller8728b832005-08-09 19:25:21 -07001778 * @list: list to use
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 *
David S. Miller8728b832005-08-09 19:25:21 -07001780 * Place a packet before a given packet in a list. The list locks are
1781 * taken and this function is atomic with respect to other list locked
1782 * calls.
1783 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 * A buffer cannot be placed on two lists at the same time.
1785 */
David S. Miller8728b832005-08-09 19:25:21 -07001786void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787{
1788 unsigned long flags;
1789
David S. Miller8728b832005-08-09 19:25:21 -07001790 spin_lock_irqsave(&list->lock, flags);
1791 __skb_insert(newsk, old->prev, old, list);
1792 spin_unlock_irqrestore(&list->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793}
1794
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795static inline void skb_split_inside_header(struct sk_buff *skb,
1796 struct sk_buff* skb1,
1797 const u32 len, const int pos)
1798{
1799 int i;
1800
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03001801 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
1802 pos - len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 /* And move data appendix as is. */
1804 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1805 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
1806
1807 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
1808 skb_shinfo(skb)->nr_frags = 0;
1809 skb1->data_len = skb->data_len;
1810 skb1->len += skb1->data_len;
1811 skb->data_len = 0;
1812 skb->len = len;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001813 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814}
1815
1816static inline void skb_split_no_header(struct sk_buff *skb,
1817 struct sk_buff* skb1,
1818 const u32 len, int pos)
1819{
1820 int i, k = 0;
1821 const int nfrags = skb_shinfo(skb)->nr_frags;
1822
1823 skb_shinfo(skb)->nr_frags = 0;
1824 skb1->len = skb1->data_len = skb->len - len;
1825 skb->len = len;
1826 skb->data_len = len - pos;
1827
1828 for (i = 0; i < nfrags; i++) {
1829 int size = skb_shinfo(skb)->frags[i].size;
1830
1831 if (pos + size > len) {
1832 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
1833
1834 if (pos < len) {
1835 /* Split frag.
1836 * We have two variants in this case:
1837 * 1. Move all the frag to the second
1838 * part, if it is possible. F.e.
1839 * this approach is mandatory for TUX,
1840 * where splitting is expensive.
1841 * 2. Split is accurately. We make this.
1842 */
1843 get_page(skb_shinfo(skb)->frags[i].page);
1844 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
1845 skb_shinfo(skb1)->frags[0].size -= len - pos;
1846 skb_shinfo(skb)->frags[i].size = len - pos;
1847 skb_shinfo(skb)->nr_frags++;
1848 }
1849 k++;
1850 } else
1851 skb_shinfo(skb)->nr_frags++;
1852 pos += size;
1853 }
1854 skb_shinfo(skb1)->nr_frags = k;
1855}
1856
1857/**
1858 * skb_split - Split fragmented skb to two parts at length len.
1859 * @skb: the buffer to split
1860 * @skb1: the buffer to receive the second part
1861 * @len: new length for skb
1862 */
1863void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
1864{
1865 int pos = skb_headlen(skb);
1866
1867 if (len < pos) /* Split line is inside header. */
1868 skb_split_inside_header(skb, skb1, len, pos);
1869 else /* Second chunk has no header, nothing to copy. */
1870 skb_split_no_header(skb, skb1, len, pos);
1871}
1872
Thomas Graf677e90e2005-06-23 20:59:51 -07001873/**
1874 * skb_prepare_seq_read - Prepare a sequential read of skb data
1875 * @skb: the buffer to read
1876 * @from: lower offset of data to be read
1877 * @to: upper offset of data to be read
1878 * @st: state variable
1879 *
1880 * Initializes the specified state variable. Must be called before
1881 * invoking skb_seq_read() for the first time.
1882 */
1883void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1884 unsigned int to, struct skb_seq_state *st)
1885{
1886 st->lower_offset = from;
1887 st->upper_offset = to;
1888 st->root_skb = st->cur_skb = skb;
1889 st->frag_idx = st->stepped_offset = 0;
1890 st->frag_data = NULL;
1891}
1892
1893/**
1894 * skb_seq_read - Sequentially read skb data
1895 * @consumed: number of bytes consumed by the caller so far
1896 * @data: destination pointer for data to be returned
1897 * @st: state variable
1898 *
1899 * Reads a block of skb data at &consumed relative to the
1900 * lower offset specified to skb_prepare_seq_read(). Assigns
1901 * the head of the data block to &data and returns the length
1902 * of the block or 0 if the end of the skb data or the upper
1903 * offset has been reached.
1904 *
1905 * The caller is not required to consume all of the data
1906 * returned, i.e. &consumed is typically set to the number
1907 * of bytes already consumed and the next call to
1908 * skb_seq_read() will return the remaining part of the block.
1909 *
Randy Dunlapbc2cda12008-02-13 15:03:25 -08001910 * Note 1: The size of each block of data returned can be arbitary,
Thomas Graf677e90e2005-06-23 20:59:51 -07001911 * this limitation is the cost for zerocopy seqeuental
1912 * reads of potentially non linear data.
1913 *
Randy Dunlapbc2cda12008-02-13 15:03:25 -08001914 * Note 2: Fragment lists within fragments are not implemented
Thomas Graf677e90e2005-06-23 20:59:51 -07001915 * at the moment, state->root_skb could be replaced with
1916 * a stack for this purpose.
1917 */
1918unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1919 struct skb_seq_state *st)
1920{
1921 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
1922 skb_frag_t *frag;
1923
1924 if (unlikely(abs_offset >= st->upper_offset))
1925 return 0;
1926
1927next_skb:
1928 block_limit = skb_headlen(st->cur_skb);
1929
1930 if (abs_offset < block_limit) {
1931 *data = st->cur_skb->data + abs_offset;
1932 return block_limit - abs_offset;
1933 }
1934
1935 if (st->frag_idx == 0 && !st->frag_data)
1936 st->stepped_offset += skb_headlen(st->cur_skb);
1937
1938 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
1939 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
1940 block_limit = frag->size + st->stepped_offset;
1941
1942 if (abs_offset < block_limit) {
1943 if (!st->frag_data)
1944 st->frag_data = kmap_skb_frag(frag);
1945
1946 *data = (u8 *) st->frag_data + frag->page_offset +
1947 (abs_offset - st->stepped_offset);
1948
1949 return block_limit - abs_offset;
1950 }
1951
1952 if (st->frag_data) {
1953 kunmap_skb_frag(st->frag_data);
1954 st->frag_data = NULL;
1955 }
1956
1957 st->frag_idx++;
1958 st->stepped_offset += frag->size;
1959 }
1960
Olaf Kirch5b5a60d2007-06-23 23:11:52 -07001961 if (st->frag_data) {
1962 kunmap_skb_frag(st->frag_data);
1963 st->frag_data = NULL;
1964 }
1965
Thomas Graf677e90e2005-06-23 20:59:51 -07001966 if (st->cur_skb->next) {
1967 st->cur_skb = st->cur_skb->next;
1968 st->frag_idx = 0;
1969 goto next_skb;
1970 } else if (st->root_skb == st->cur_skb &&
1971 skb_shinfo(st->root_skb)->frag_list) {
1972 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
1973 goto next_skb;
1974 }
1975
1976 return 0;
1977}
1978
1979/**
1980 * skb_abort_seq_read - Abort a sequential read of skb data
1981 * @st: state variable
1982 *
1983 * Must be called if skb_seq_read() was not called until it
1984 * returned 0.
1985 */
1986void skb_abort_seq_read(struct skb_seq_state *st)
1987{
1988 if (st->frag_data)
1989 kunmap_skb_frag(st->frag_data);
1990}
1991
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07001992#define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
1993
1994static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
1995 struct ts_config *conf,
1996 struct ts_state *state)
1997{
1998 return skb_seq_read(offset, text, TS_SKB_CB(state));
1999}
2000
2001static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2002{
2003 skb_abort_seq_read(TS_SKB_CB(state));
2004}
2005
2006/**
2007 * skb_find_text - Find a text pattern in skb data
2008 * @skb: the buffer to look in
2009 * @from: search offset
2010 * @to: search limit
2011 * @config: textsearch configuration
2012 * @state: uninitialized textsearch state variable
2013 *
2014 * Finds a pattern in the skb data according to the specified
2015 * textsearch configuration. Use textsearch_next() to retrieve
2016 * subsequent occurrences of the pattern. Returns the offset
2017 * to the first occurrence or UINT_MAX if no match was found.
2018 */
2019unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2020 unsigned int to, struct ts_config *config,
2021 struct ts_state *state)
2022{
Phil Oesterf72b9482006-06-26 00:00:57 -07002023 unsigned int ret;
2024
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002025 config->get_next_block = skb_ts_get_next_block;
2026 config->finish = skb_ts_finish;
2027
2028 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2029
Phil Oesterf72b9482006-06-26 00:00:57 -07002030 ret = textsearch_find(config, state);
2031 return (ret <= to - from ? ret : UINT_MAX);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002032}
2033
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002034/**
2035 * skb_append_datato_frags: - append the user data to a skb
2036 * @sk: sock structure
2037 * @skb: skb structure to be appened with user data.
2038 * @getfrag: call back function to be used for getting the user data
2039 * @from: pointer to user message iov
2040 * @length: length of the iov message
2041 *
2042 * Description: This procedure append the user data in the fragment part
2043 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2044 */
2045int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
Martin Waitzdab96302005-12-05 13:40:12 -08002046 int (*getfrag)(void *from, char *to, int offset,
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002047 int len, int odd, struct sk_buff *skb),
2048 void *from, int length)
2049{
2050 int frg_cnt = 0;
2051 skb_frag_t *frag = NULL;
2052 struct page *page = NULL;
2053 int copy, left;
2054 int offset = 0;
2055 int ret;
2056
2057 do {
2058 /* Return error if we don't have space for new frag */
2059 frg_cnt = skb_shinfo(skb)->nr_frags;
2060 if (frg_cnt >= MAX_SKB_FRAGS)
2061 return -EFAULT;
2062
2063 /* allocate a new page for next frag */
2064 page = alloc_pages(sk->sk_allocation, 0);
2065
2066 /* If alloc_page fails just return failure and caller will
2067 * free previous allocated pages by doing kfree_skb()
2068 */
2069 if (page == NULL)
2070 return -ENOMEM;
2071
2072 /* initialize the next frag */
2073 sk->sk_sndmsg_page = page;
2074 sk->sk_sndmsg_off = 0;
2075 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2076 skb->truesize += PAGE_SIZE;
2077 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2078
2079 /* get the new initialized frag */
2080 frg_cnt = skb_shinfo(skb)->nr_frags;
2081 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2082
2083 /* copy the user data to page */
2084 left = PAGE_SIZE - frag->page_offset;
2085 copy = (length > left)? left : length;
2086
2087 ret = getfrag(from, (page_address(frag->page) +
2088 frag->page_offset + frag->size),
2089 offset, copy, 0, skb);
2090 if (ret < 0)
2091 return -EFAULT;
2092
2093 /* copy was successful so update the size parameters */
2094 sk->sk_sndmsg_off += copy;
2095 frag->size += copy;
2096 skb->len += copy;
2097 skb->data_len += copy;
2098 offset += copy;
2099 length -= copy;
2100
2101 } while (length > 0);
2102
2103 return 0;
2104}
2105
Herbert Xucbb042f2006-03-20 22:43:56 -08002106/**
2107 * skb_pull_rcsum - pull skb and update receive checksum
2108 * @skb: buffer to update
Herbert Xucbb042f2006-03-20 22:43:56 -08002109 * @len: length of data pulled
2110 *
2111 * This function performs an skb_pull on the packet and updates
Urs Thuermannfee54fa2008-02-12 22:03:25 -08002112 * the CHECKSUM_COMPLETE checksum. It should be used on
Patrick McHardy84fa7932006-08-29 16:44:56 -07002113 * receive path processing instead of skb_pull unless you know
2114 * that the checksum difference is zero (e.g., a valid IP header)
2115 * or you are setting ip_summed to CHECKSUM_NONE.
Herbert Xucbb042f2006-03-20 22:43:56 -08002116 */
2117unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2118{
2119 BUG_ON(len > skb->len);
2120 skb->len -= len;
2121 BUG_ON(skb->len < skb->data_len);
2122 skb_postpull_rcsum(skb, skb->data, len);
2123 return skb->data += len;
2124}
2125
Arnaldo Carvalho de Melof94691a2006-03-20 22:47:55 -08002126EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2127
Herbert Xuf4c50d92006-06-22 03:02:40 -07002128/**
2129 * skb_segment - Perform protocol segmentation on skb.
2130 * @skb: buffer to segment
Herbert Xu576a30e2006-06-27 13:22:38 -07002131 * @features: features for the output path (see dev->features)
Herbert Xuf4c50d92006-06-22 03:02:40 -07002132 *
2133 * This function performs segmentation on the given skb. It returns
2134 * the segment at the given position. It returns NULL if there are
2135 * no more segments to generate, or when an error is encountered.
2136 */
Herbert Xu576a30e2006-06-27 13:22:38 -07002137struct sk_buff *skb_segment(struct sk_buff *skb, int features)
Herbert Xuf4c50d92006-06-22 03:02:40 -07002138{
2139 struct sk_buff *segs = NULL;
2140 struct sk_buff *tail = NULL;
2141 unsigned int mss = skb_shinfo(skb)->gso_size;
Arnaldo Carvalho de Melo98e399f2007-03-19 15:33:04 -07002142 unsigned int doffset = skb->data - skb_mac_header(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002143 unsigned int offset = doffset;
2144 unsigned int headroom;
2145 unsigned int len;
Herbert Xu576a30e2006-06-27 13:22:38 -07002146 int sg = features & NETIF_F_SG;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002147 int nfrags = skb_shinfo(skb)->nr_frags;
2148 int err = -ENOMEM;
2149 int i = 0;
2150 int pos;
2151
2152 __skb_push(skb, doffset);
2153 headroom = skb_headroom(skb);
2154 pos = skb_headlen(skb);
2155
2156 do {
2157 struct sk_buff *nskb;
2158 skb_frag_t *frag;
Herbert Xuc8884ed2006-10-29 15:59:41 -08002159 int hsize;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002160 int k;
2161 int size;
2162
2163 len = skb->len - offset;
2164 if (len > mss)
2165 len = mss;
2166
2167 hsize = skb_headlen(skb) - offset;
2168 if (hsize < 0)
2169 hsize = 0;
Herbert Xuc8884ed2006-10-29 15:59:41 -08002170 if (hsize > len || !sg)
2171 hsize = len;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002172
Herbert Xuc8884ed2006-10-29 15:59:41 -08002173 nskb = alloc_skb(hsize + doffset + headroom, GFP_ATOMIC);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002174 if (unlikely(!nskb))
2175 goto err;
2176
2177 if (segs)
2178 tail->next = nskb;
2179 else
2180 segs = nskb;
2181 tail = nskb;
2182
2183 nskb->dev = skb->dev;
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07002184 skb_copy_queue_mapping(nskb, skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002185 nskb->priority = skb->priority;
2186 nskb->protocol = skb->protocol;
2187 nskb->dst = dst_clone(skb->dst);
2188 memcpy(nskb->cb, skb->cb, sizeof(skb->cb));
2189 nskb->pkt_type = skb->pkt_type;
2190 nskb->mac_len = skb->mac_len;
2191
2192 skb_reserve(nskb, headroom);
Arnaldo Carvalho de Melo459a98e2007-03-19 15:30:44 -07002193 skb_reset_mac_header(nskb);
Arnaldo Carvalho de Meloddc7b8e2007-03-15 21:42:27 -03002194 skb_set_network_header(nskb, skb->mac_len);
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -07002195 nskb->transport_header = (nskb->network_header +
2196 skb_network_header_len(skb));
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03002197 skb_copy_from_linear_data(skb, skb_put(nskb, doffset),
2198 doffset);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002199 if (!sg) {
2200 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2201 skb_put(nskb, len),
2202 len, 0);
2203 continue;
2204 }
2205
2206 frag = skb_shinfo(nskb)->frags;
2207 k = 0;
2208
Patrick McHardy84fa7932006-08-29 16:44:56 -07002209 nskb->ip_summed = CHECKSUM_PARTIAL;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002210 nskb->csum = skb->csum;
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03002211 skb_copy_from_linear_data_offset(skb, offset,
2212 skb_put(nskb, hsize), hsize);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002213
2214 while (pos < offset + len) {
2215 BUG_ON(i >= nfrags);
2216
2217 *frag = skb_shinfo(skb)->frags[i];
2218 get_page(frag->page);
2219 size = frag->size;
2220
2221 if (pos < offset) {
2222 frag->page_offset += offset - pos;
2223 frag->size -= offset - pos;
2224 }
2225
2226 k++;
2227
2228 if (pos + size <= offset + len) {
2229 i++;
2230 pos += size;
2231 } else {
2232 frag->size -= pos + size - (offset + len);
2233 break;
2234 }
2235
2236 frag++;
2237 }
2238
2239 skb_shinfo(nskb)->nr_frags = k;
2240 nskb->data_len = len - hsize;
2241 nskb->len += nskb->data_len;
2242 nskb->truesize += nskb->data_len;
2243 } while ((offset += len) < skb->len);
2244
2245 return segs;
2246
2247err:
2248 while ((skb = segs)) {
2249 segs = skb->next;
Patrick McHardyb08d5842007-02-27 09:57:37 -08002250 kfree_skb(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002251 }
2252 return ERR_PTR(err);
2253}
2254
2255EXPORT_SYMBOL_GPL(skb_segment);
2256
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257void __init skb_init(void)
2258{
2259 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2260 sizeof(struct sk_buff),
2261 0,
Alexey Dobriyane5d679f332006-08-26 19:25:52 -07002262 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
Paul Mundt20c2df82007-07-20 10:11:58 +09002263 NULL);
David S. Millerd179cd12005-08-17 14:57:30 -07002264 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2265 (2*sizeof(struct sk_buff)) +
2266 sizeof(atomic_t),
2267 0,
Alexey Dobriyane5d679f332006-08-26 19:25:52 -07002268 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
Paul Mundt20c2df82007-07-20 10:11:58 +09002269 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270}
2271
David Howells716ea3a2007-04-02 20:19:53 -07002272/**
2273 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2274 * @skb: Socket buffer containing the buffers to be mapped
2275 * @sg: The scatter-gather list to map into
2276 * @offset: The offset into the buffer's contents to start mapping
2277 * @len: Length of buffer space to be mapped
2278 *
2279 * Fill the specified scatter-gather list with mappings/pointers into a
2280 * region of the buffer space attached to a socket buffer.
2281 */
David S. Miller51c739d2007-10-30 21:29:29 -07002282static int
2283__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
David Howells716ea3a2007-04-02 20:19:53 -07002284{
David S. Miller1a028e52007-04-27 15:21:23 -07002285 int start = skb_headlen(skb);
2286 int i, copy = start - offset;
David Howells716ea3a2007-04-02 20:19:53 -07002287 int elt = 0;
2288
2289 if (copy > 0) {
2290 if (copy > len)
2291 copy = len;
Jens Axboe642f1492007-10-24 11:20:47 +02002292 sg_set_buf(sg, skb->data + offset, copy);
David Howells716ea3a2007-04-02 20:19:53 -07002293 elt++;
2294 if ((len -= copy) == 0)
2295 return elt;
2296 offset += copy;
2297 }
2298
2299 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07002300 int end;
David Howells716ea3a2007-04-02 20:19:53 -07002301
David S. Miller1a028e52007-04-27 15:21:23 -07002302 BUG_TRAP(start <= offset + len);
2303
2304 end = start + skb_shinfo(skb)->frags[i].size;
David Howells716ea3a2007-04-02 20:19:53 -07002305 if ((copy = end - offset) > 0) {
2306 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2307
2308 if (copy > len)
2309 copy = len;
Jens Axboe642f1492007-10-24 11:20:47 +02002310 sg_set_page(&sg[elt], frag->page, copy,
2311 frag->page_offset+offset-start);
David Howells716ea3a2007-04-02 20:19:53 -07002312 elt++;
2313 if (!(len -= copy))
2314 return elt;
2315 offset += copy;
2316 }
David S. Miller1a028e52007-04-27 15:21:23 -07002317 start = end;
David Howells716ea3a2007-04-02 20:19:53 -07002318 }
2319
2320 if (skb_shinfo(skb)->frag_list) {
2321 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2322
2323 for (; list; list = list->next) {
David S. Miller1a028e52007-04-27 15:21:23 -07002324 int end;
David Howells716ea3a2007-04-02 20:19:53 -07002325
David S. Miller1a028e52007-04-27 15:21:23 -07002326 BUG_TRAP(start <= offset + len);
2327
2328 end = start + list->len;
David Howells716ea3a2007-04-02 20:19:53 -07002329 if ((copy = end - offset) > 0) {
2330 if (copy > len)
2331 copy = len;
David S. Miller51c739d2007-10-30 21:29:29 -07002332 elt += __skb_to_sgvec(list, sg+elt, offset - start,
2333 copy);
David Howells716ea3a2007-04-02 20:19:53 -07002334 if ((len -= copy) == 0)
2335 return elt;
2336 offset += copy;
2337 }
David S. Miller1a028e52007-04-27 15:21:23 -07002338 start = end;
David Howells716ea3a2007-04-02 20:19:53 -07002339 }
2340 }
2341 BUG_ON(len);
2342 return elt;
2343}
2344
David S. Miller51c739d2007-10-30 21:29:29 -07002345int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2346{
2347 int nsg = __skb_to_sgvec(skb, sg, offset, len);
2348
Jens Axboec46f2332007-10-31 12:06:37 +01002349 sg_mark_end(&sg[nsg - 1]);
David S. Miller51c739d2007-10-30 21:29:29 -07002350
2351 return nsg;
2352}
2353
David Howells716ea3a2007-04-02 20:19:53 -07002354/**
2355 * skb_cow_data - Check that a socket buffer's data buffers are writable
2356 * @skb: The socket buffer to check.
2357 * @tailbits: Amount of trailing space to be added
2358 * @trailer: Returned pointer to the skb where the @tailbits space begins
2359 *
2360 * Make sure that the data buffers attached to a socket buffer are
2361 * writable. If they are not, private copies are made of the data buffers
2362 * and the socket buffer is set to use these instead.
2363 *
2364 * If @tailbits is given, make sure that there is space to write @tailbits
2365 * bytes of data beyond current end of socket buffer. @trailer will be
2366 * set to point to the skb in which this space begins.
2367 *
2368 * The number of scatterlist elements required to completely map the
2369 * COW'd and extended socket buffer will be returned.
2370 */
2371int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
2372{
2373 int copyflag;
2374 int elt;
2375 struct sk_buff *skb1, **skb_p;
2376
2377 /* If skb is cloned or its head is paged, reallocate
2378 * head pulling out all the pages (pages are considered not writable
2379 * at the moment even if they are anonymous).
2380 */
2381 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
2382 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
2383 return -ENOMEM;
2384
2385 /* Easy case. Most of packets will go this way. */
2386 if (!skb_shinfo(skb)->frag_list) {
2387 /* A little of trouble, not enough of space for trailer.
2388 * This should not happen, when stack is tuned to generate
2389 * good frames. OK, on miss we reallocate and reserve even more
2390 * space, 128 bytes is fair. */
2391
2392 if (skb_tailroom(skb) < tailbits &&
2393 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
2394 return -ENOMEM;
2395
2396 /* Voila! */
2397 *trailer = skb;
2398 return 1;
2399 }
2400
2401 /* Misery. We are in troubles, going to mincer fragments... */
2402
2403 elt = 1;
2404 skb_p = &skb_shinfo(skb)->frag_list;
2405 copyflag = 0;
2406
2407 while ((skb1 = *skb_p) != NULL) {
2408 int ntail = 0;
2409
2410 /* The fragment is partially pulled by someone,
2411 * this can happen on input. Copy it and everything
2412 * after it. */
2413
2414 if (skb_shared(skb1))
2415 copyflag = 1;
2416
2417 /* If the skb is the last, worry about trailer. */
2418
2419 if (skb1->next == NULL && tailbits) {
2420 if (skb_shinfo(skb1)->nr_frags ||
2421 skb_shinfo(skb1)->frag_list ||
2422 skb_tailroom(skb1) < tailbits)
2423 ntail = tailbits + 128;
2424 }
2425
2426 if (copyflag ||
2427 skb_cloned(skb1) ||
2428 ntail ||
2429 skb_shinfo(skb1)->nr_frags ||
2430 skb_shinfo(skb1)->frag_list) {
2431 struct sk_buff *skb2;
2432
2433 /* Fuck, we are miserable poor guys... */
2434 if (ntail == 0)
2435 skb2 = skb_copy(skb1, GFP_ATOMIC);
2436 else
2437 skb2 = skb_copy_expand(skb1,
2438 skb_headroom(skb1),
2439 ntail,
2440 GFP_ATOMIC);
2441 if (unlikely(skb2 == NULL))
2442 return -ENOMEM;
2443
2444 if (skb1->sk)
2445 skb_set_owner_w(skb2, skb1->sk);
2446
2447 /* Looking around. Are we still alive?
2448 * OK, link new skb, drop old one */
2449
2450 skb2->next = skb1->next;
2451 *skb_p = skb2;
2452 kfree_skb(skb1);
2453 skb1 = skb2;
2454 }
2455 elt++;
2456 *trailer = skb1;
2457 skb_p = &skb1->next;
2458 }
2459
2460 return elt;
2461}
2462
Rusty Russellf35d9d82008-02-04 23:49:54 -05002463/**
2464 * skb_partial_csum_set - set up and verify partial csum values for packet
2465 * @skb: the skb to set
2466 * @start: the number of bytes after skb->data to start checksumming.
2467 * @off: the offset from start to place the checksum.
2468 *
2469 * For untrusted partially-checksummed packets, we need to make sure the values
2470 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
2471 *
2472 * This function checks and sets those values and skb->ip_summed: if this
2473 * returns false you should drop the packet.
2474 */
2475bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
2476{
2477 if (unlikely(start > skb->len - 2) ||
2478 unlikely((int)start + off > skb->len - 2)) {
2479 if (net_ratelimit())
2480 printk(KERN_WARNING
2481 "bad partial csum: csum=%u/%u len=%u\n",
2482 start, off, skb->len);
2483 return false;
2484 }
2485 skb->ip_summed = CHECKSUM_PARTIAL;
2486 skb->csum_start = skb_headroom(skb) + start;
2487 skb->csum_offset = off;
2488 return true;
2489}
2490
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491EXPORT_SYMBOL(___pskb_trim);
2492EXPORT_SYMBOL(__kfree_skb);
Jörn Engel231d06a2006-03-20 21:28:35 -08002493EXPORT_SYMBOL(kfree_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494EXPORT_SYMBOL(__pskb_pull_tail);
David S. Millerd179cd12005-08-17 14:57:30 -07002495EXPORT_SYMBOL(__alloc_skb);
Christoph Hellwig8af27452006-07-31 22:35:23 -07002496EXPORT_SYMBOL(__netdev_alloc_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497EXPORT_SYMBOL(pskb_copy);
2498EXPORT_SYMBOL(pskb_expand_head);
2499EXPORT_SYMBOL(skb_checksum);
2500EXPORT_SYMBOL(skb_clone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501EXPORT_SYMBOL(skb_copy);
2502EXPORT_SYMBOL(skb_copy_and_csum_bits);
2503EXPORT_SYMBOL(skb_copy_and_csum_dev);
2504EXPORT_SYMBOL(skb_copy_bits);
2505EXPORT_SYMBOL(skb_copy_expand);
2506EXPORT_SYMBOL(skb_over_panic);
2507EXPORT_SYMBOL(skb_pad);
2508EXPORT_SYMBOL(skb_realloc_headroom);
2509EXPORT_SYMBOL(skb_under_panic);
2510EXPORT_SYMBOL(skb_dequeue);
2511EXPORT_SYMBOL(skb_dequeue_tail);
2512EXPORT_SYMBOL(skb_insert);
2513EXPORT_SYMBOL(skb_queue_purge);
2514EXPORT_SYMBOL(skb_queue_head);
2515EXPORT_SYMBOL(skb_queue_tail);
2516EXPORT_SYMBOL(skb_unlink);
2517EXPORT_SYMBOL(skb_append);
2518EXPORT_SYMBOL(skb_split);
Thomas Graf677e90e2005-06-23 20:59:51 -07002519EXPORT_SYMBOL(skb_prepare_seq_read);
2520EXPORT_SYMBOL(skb_seq_read);
2521EXPORT_SYMBOL(skb_abort_seq_read);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002522EXPORT_SYMBOL(skb_find_text);
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002523EXPORT_SYMBOL(skb_append_datato_frags);
David Howells716ea3a2007-04-02 20:19:53 -07002524
2525EXPORT_SYMBOL_GPL(skb_to_sgvec);
2526EXPORT_SYMBOL_GPL(skb_cow_data);
Rusty Russellf35d9d82008-02-04 23:49:54 -05002527EXPORT_SYMBOL_GPL(skb_partial_csum_set);