blob: cf489b6329e853dae5f07abb41eb68ac288b6e95 [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
Ilpo Järvinen0dde3e12008-03-27 17:43:41 -0700860/**
861 * skb_put - add data to a buffer
862 * @skb: buffer to use
863 * @len: amount of data to add
864 *
865 * This function extends the used data area of the buffer. If this would
866 * exceed the total buffer size the kernel will panic. A pointer to the
867 * first byte of the extra data is returned.
868 */
869unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
870{
871 unsigned char *tmp = skb_tail_pointer(skb);
872 SKB_LINEAR_ASSERT(skb);
873 skb->tail += len;
874 skb->len += len;
875 if (unlikely(skb->tail > skb->end))
876 skb_over_panic(skb, len, __builtin_return_address(0));
877 return tmp;
878}
879EXPORT_SYMBOL(skb_put);
880
Ilpo Järvinen6be8ac22008-03-27 17:47:24 -0700881/**
882 * skb_pull - remove data from the start of a buffer
883 * @skb: buffer to use
884 * @len: amount of data to remove
885 *
886 * This function removes data from the start of a buffer, returning
887 * the memory to the headroom. A pointer to the next data in the buffer
888 * is returned. Once the data has been pulled future pushes will overwrite
889 * the old data.
890 */
891unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
892{
893 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
894}
895EXPORT_SYMBOL(skb_pull);
896
Herbert Xu3cc0e872006-06-09 16:13:38 -0700897/* Trims skb to length len. It can change skb pointers.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 */
899
Herbert Xu3cc0e872006-06-09 16:13:38 -0700900int ___pskb_trim(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901{
Herbert Xu27b437c2006-07-13 19:26:39 -0700902 struct sk_buff **fragp;
903 struct sk_buff *frag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 int offset = skb_headlen(skb);
905 int nfrags = skb_shinfo(skb)->nr_frags;
906 int i;
Herbert Xu27b437c2006-07-13 19:26:39 -0700907 int err;
908
909 if (skb_cloned(skb) &&
910 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
911 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912
Herbert Xuf4d26fb2006-07-30 20:20:28 -0700913 i = 0;
914 if (offset >= len)
915 goto drop_pages;
916
917 for (; i < nfrags; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 int end = offset + skb_shinfo(skb)->frags[i].size;
Herbert Xu27b437c2006-07-13 19:26:39 -0700919
920 if (end < len) {
921 offset = end;
922 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 }
Herbert Xu27b437c2006-07-13 19:26:39 -0700924
Herbert Xuf4d26fb2006-07-30 20:20:28 -0700925 skb_shinfo(skb)->frags[i++].size = len - offset;
Herbert Xu27b437c2006-07-13 19:26:39 -0700926
Herbert Xuf4d26fb2006-07-30 20:20:28 -0700927drop_pages:
Herbert Xu27b437c2006-07-13 19:26:39 -0700928 skb_shinfo(skb)->nr_frags = i;
929
930 for (; i < nfrags; i++)
931 put_page(skb_shinfo(skb)->frags[i].page);
932
933 if (skb_shinfo(skb)->frag_list)
934 skb_drop_fraglist(skb);
Herbert Xuf4d26fb2006-07-30 20:20:28 -0700935 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 }
937
Herbert Xu27b437c2006-07-13 19:26:39 -0700938 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
939 fragp = &frag->next) {
940 int end = offset + frag->len;
941
942 if (skb_shared(frag)) {
943 struct sk_buff *nfrag;
944
945 nfrag = skb_clone(frag, GFP_ATOMIC);
946 if (unlikely(!nfrag))
947 return -ENOMEM;
948
949 nfrag->next = frag->next;
Herbert Xuf4d26fb2006-07-30 20:20:28 -0700950 kfree_skb(frag);
Herbert Xu27b437c2006-07-13 19:26:39 -0700951 frag = nfrag;
952 *fragp = frag;
953 }
954
955 if (end < len) {
956 offset = end;
957 continue;
958 }
959
960 if (end > len &&
961 unlikely((err = pskb_trim(frag, len - offset))))
962 return err;
963
964 if (frag->next)
965 skb_drop_list(&frag->next);
966 break;
967 }
968
Herbert Xuf4d26fb2006-07-30 20:20:28 -0700969done:
Herbert Xu27b437c2006-07-13 19:26:39 -0700970 if (len > skb_headlen(skb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 skb->data_len -= skb->len - len;
972 skb->len = len;
973 } else {
Herbert Xu27b437c2006-07-13 19:26:39 -0700974 skb->len = len;
975 skb->data_len = 0;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700976 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 }
978
979 return 0;
980}
981
982/**
983 * __pskb_pull_tail - advance tail of skb header
984 * @skb: buffer to reallocate
985 * @delta: number of bytes to advance tail
986 *
987 * The function makes a sense only on a fragmented &sk_buff,
988 * it expands header moving its tail forward and copying necessary
989 * data from fragmented part.
990 *
991 * &sk_buff MUST have reference count of 1.
992 *
993 * Returns %NULL (and &sk_buff does not change) if pull failed
994 * or value of new tail of skb in the case of success.
995 *
996 * All the pointers pointing into skb header may change and must be
997 * reloaded after call to this function.
998 */
999
1000/* Moves tail of skb head forward, copying data from fragmented part,
1001 * when it is necessary.
1002 * 1. It may fail due to malloc failure.
1003 * 2. It may change skb pointers.
1004 *
1005 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1006 */
1007unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1008{
1009 /* If skb has not enough free space at tail, get new one
1010 * plus 128 bytes for future expansions. If we have enough
1011 * room at tail, reallocate without expansion only if skb is cloned.
1012 */
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001013 int i, k, eat = (skb->tail + delta) - skb->end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014
1015 if (eat > 0 || skb_cloned(skb)) {
1016 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1017 GFP_ATOMIC))
1018 return NULL;
1019 }
1020
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001021 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 BUG();
1023
1024 /* Optimization: no fragments, no reasons to preestimate
1025 * size of pulled pages. Superb.
1026 */
1027 if (!skb_shinfo(skb)->frag_list)
1028 goto pull_pages;
1029
1030 /* Estimate size of pulled pages. */
1031 eat = delta;
1032 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1033 if (skb_shinfo(skb)->frags[i].size >= eat)
1034 goto pull_pages;
1035 eat -= skb_shinfo(skb)->frags[i].size;
1036 }
1037
1038 /* If we need update frag list, we are in troubles.
1039 * Certainly, it possible to add an offset to skb data,
1040 * but taking into account that pulling is expected to
1041 * be very rare operation, it is worth to fight against
1042 * further bloating skb head and crucify ourselves here instead.
1043 * Pure masohism, indeed. 8)8)
1044 */
1045 if (eat) {
1046 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1047 struct sk_buff *clone = NULL;
1048 struct sk_buff *insp = NULL;
1049
1050 do {
Kris Katterjohn09a62662006-01-08 22:24:28 -08001051 BUG_ON(!list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052
1053 if (list->len <= eat) {
1054 /* Eaten as whole. */
1055 eat -= list->len;
1056 list = list->next;
1057 insp = list;
1058 } else {
1059 /* Eaten partially. */
1060
1061 if (skb_shared(list)) {
1062 /* Sucks! We need to fork list. :-( */
1063 clone = skb_clone(list, GFP_ATOMIC);
1064 if (!clone)
1065 return NULL;
1066 insp = list->next;
1067 list = clone;
1068 } else {
1069 /* This may be pulled without
1070 * problems. */
1071 insp = list;
1072 }
1073 if (!pskb_pull(list, eat)) {
1074 if (clone)
1075 kfree_skb(clone);
1076 return NULL;
1077 }
1078 break;
1079 }
1080 } while (eat);
1081
1082 /* Free pulled out fragments. */
1083 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1084 skb_shinfo(skb)->frag_list = list->next;
1085 kfree_skb(list);
1086 }
1087 /* And insert new clone at head. */
1088 if (clone) {
1089 clone->next = list;
1090 skb_shinfo(skb)->frag_list = clone;
1091 }
1092 }
1093 /* Success! Now we may commit changes to skb data. */
1094
1095pull_pages:
1096 eat = delta;
1097 k = 0;
1098 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1099 if (skb_shinfo(skb)->frags[i].size <= eat) {
1100 put_page(skb_shinfo(skb)->frags[i].page);
1101 eat -= skb_shinfo(skb)->frags[i].size;
1102 } else {
1103 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1104 if (eat) {
1105 skb_shinfo(skb)->frags[k].page_offset += eat;
1106 skb_shinfo(skb)->frags[k].size -= eat;
1107 eat = 0;
1108 }
1109 k++;
1110 }
1111 }
1112 skb_shinfo(skb)->nr_frags = k;
1113
1114 skb->tail += delta;
1115 skb->data_len -= delta;
1116
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001117 return skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118}
1119
1120/* Copy some data bits from skb to kernel buffer. */
1121
1122int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1123{
1124 int i, copy;
David S. Miller1a028e52007-04-27 15:21:23 -07001125 int start = skb_headlen(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126
1127 if (offset > (int)skb->len - len)
1128 goto fault;
1129
1130 /* Copy header. */
David S. Miller1a028e52007-04-27 15:21:23 -07001131 if ((copy = start - offset) > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 if (copy > len)
1133 copy = len;
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03001134 skb_copy_from_linear_data_offset(skb, offset, to, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 if ((len -= copy) == 0)
1136 return 0;
1137 offset += copy;
1138 to += copy;
1139 }
1140
1141 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07001142 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143
David S. Miller1a028e52007-04-27 15:21:23 -07001144 BUG_TRAP(start <= offset + len);
1145
1146 end = start + skb_shinfo(skb)->frags[i].size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 if ((copy = end - offset) > 0) {
1148 u8 *vaddr;
1149
1150 if (copy > len)
1151 copy = len;
1152
1153 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1154 memcpy(to,
David S. Miller1a028e52007-04-27 15:21:23 -07001155 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1156 offset - start, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 kunmap_skb_frag(vaddr);
1158
1159 if ((len -= copy) == 0)
1160 return 0;
1161 offset += copy;
1162 to += copy;
1163 }
David S. Miller1a028e52007-04-27 15:21:23 -07001164 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 }
1166
1167 if (skb_shinfo(skb)->frag_list) {
1168 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1169
1170 for (; list; list = list->next) {
David S. Miller1a028e52007-04-27 15:21:23 -07001171 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172
David S. Miller1a028e52007-04-27 15:21:23 -07001173 BUG_TRAP(start <= offset + len);
1174
1175 end = start + list->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 if ((copy = end - offset) > 0) {
1177 if (copy > len)
1178 copy = len;
David S. Miller1a028e52007-04-27 15:21:23 -07001179 if (skb_copy_bits(list, offset - start,
1180 to, copy))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 goto fault;
1182 if ((len -= copy) == 0)
1183 return 0;
1184 offset += copy;
1185 to += copy;
1186 }
David S. Miller1a028e52007-04-27 15:21:23 -07001187 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 }
1189 }
1190 if (!len)
1191 return 0;
1192
1193fault:
1194 return -EFAULT;
1195}
1196
Jens Axboe9c55e012007-11-06 23:30:13 -08001197/*
1198 * Callback from splice_to_pipe(), if we need to release some pages
1199 * at the end of the spd in case we error'ed out in filling the pipe.
1200 */
1201static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1202{
1203 struct sk_buff *skb = (struct sk_buff *) spd->partial[i].private;
1204
1205 kfree_skb(skb);
1206}
1207
1208/*
1209 * Fill page/offset/length into spd, if it can hold more pages.
1210 */
1211static inline int spd_fill_page(struct splice_pipe_desc *spd, struct page *page,
1212 unsigned int len, unsigned int offset,
1213 struct sk_buff *skb)
1214{
1215 if (unlikely(spd->nr_pages == PIPE_BUFFERS))
1216 return 1;
1217
1218 spd->pages[spd->nr_pages] = page;
1219 spd->partial[spd->nr_pages].len = len;
1220 spd->partial[spd->nr_pages].offset = offset;
1221 spd->partial[spd->nr_pages].private = (unsigned long) skb_get(skb);
1222 spd->nr_pages++;
1223 return 0;
1224}
1225
1226/*
1227 * Map linear and fragment data from the skb to spd. Returns number of
1228 * pages mapped.
1229 */
1230static int __skb_splice_bits(struct sk_buff *skb, unsigned int *offset,
1231 unsigned int *total_len,
1232 struct splice_pipe_desc *spd)
1233{
1234 unsigned int nr_pages = spd->nr_pages;
1235 unsigned int poff, plen, len, toff, tlen;
1236 int headlen, seg;
1237
1238 toff = *offset;
1239 tlen = *total_len;
1240 if (!tlen)
1241 goto err;
1242
1243 /*
1244 * if the offset is greater than the linear part, go directly to
1245 * the fragments.
1246 */
1247 headlen = skb_headlen(skb);
1248 if (toff >= headlen) {
1249 toff -= headlen;
1250 goto map_frag;
1251 }
1252
1253 /*
1254 * first map the linear region into the pages/partial map, skipping
1255 * any potential initial offset.
1256 */
1257 len = 0;
1258 while (len < headlen) {
1259 void *p = skb->data + len;
1260
1261 poff = (unsigned long) p & (PAGE_SIZE - 1);
1262 plen = min_t(unsigned int, headlen - len, PAGE_SIZE - poff);
1263 len += plen;
1264
1265 if (toff) {
1266 if (plen <= toff) {
1267 toff -= plen;
1268 continue;
1269 }
1270 plen -= toff;
1271 poff += toff;
1272 toff = 0;
1273 }
1274
1275 plen = min(plen, tlen);
1276 if (!plen)
1277 break;
1278
1279 /*
1280 * just jump directly to update and return, no point
1281 * in going over fragments when the output is full.
1282 */
1283 if (spd_fill_page(spd, virt_to_page(p), plen, poff, skb))
1284 goto done;
1285
1286 tlen -= plen;
1287 }
1288
1289 /*
1290 * then map the fragments
1291 */
1292map_frag:
1293 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1294 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1295
1296 plen = f->size;
1297 poff = f->page_offset;
1298
1299 if (toff) {
1300 if (plen <= toff) {
1301 toff -= plen;
1302 continue;
1303 }
1304 plen -= toff;
1305 poff += toff;
1306 toff = 0;
1307 }
1308
1309 plen = min(plen, tlen);
1310 if (!plen)
1311 break;
1312
1313 if (spd_fill_page(spd, f->page, plen, poff, skb))
1314 break;
1315
1316 tlen -= plen;
1317 }
1318
1319done:
1320 if (spd->nr_pages - nr_pages) {
1321 *offset = 0;
1322 *total_len = tlen;
1323 return 0;
1324 }
1325err:
1326 return 1;
1327}
1328
1329/*
1330 * Map data from the skb to a pipe. Should handle both the linear part,
1331 * the fragments, and the frag list. It does NOT handle frag lists within
1332 * the frag list, if such a thing exists. We'd probably need to recurse to
1333 * handle that cleanly.
1334 */
1335int skb_splice_bits(struct sk_buff *__skb, unsigned int offset,
1336 struct pipe_inode_info *pipe, unsigned int tlen,
1337 unsigned int flags)
1338{
1339 struct partial_page partial[PIPE_BUFFERS];
1340 struct page *pages[PIPE_BUFFERS];
1341 struct splice_pipe_desc spd = {
1342 .pages = pages,
1343 .partial = partial,
1344 .flags = flags,
1345 .ops = &sock_pipe_buf_ops,
1346 .spd_release = sock_spd_release,
1347 };
1348 struct sk_buff *skb;
1349
1350 /*
1351 * I'd love to avoid the clone here, but tcp_read_sock()
1352 * ignores reference counts and unconditonally kills the sk_buff
1353 * on return from the actor.
1354 */
1355 skb = skb_clone(__skb, GFP_KERNEL);
1356 if (unlikely(!skb))
1357 return -ENOMEM;
1358
1359 /*
1360 * __skb_splice_bits() only fails if the output has no room left,
1361 * so no point in going over the frag_list for the error case.
1362 */
1363 if (__skb_splice_bits(skb, &offset, &tlen, &spd))
1364 goto done;
1365 else if (!tlen)
1366 goto done;
1367
1368 /*
1369 * now see if we have a frag_list to map
1370 */
1371 if (skb_shinfo(skb)->frag_list) {
1372 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1373
1374 for (; list && tlen; list = list->next) {
1375 if (__skb_splice_bits(list, &offset, &tlen, &spd))
1376 break;
1377 }
1378 }
1379
1380done:
1381 /*
1382 * drop our reference to the clone, the pipe consumption will
1383 * drop the rest.
1384 */
1385 kfree_skb(skb);
1386
1387 if (spd.nr_pages) {
1388 int ret;
1389
1390 /*
1391 * Drop the socket lock, otherwise we have reverse
1392 * locking dependencies between sk_lock and i_mutex
1393 * here as compared to sendfile(). We enter here
1394 * with the socket lock held, and splice_to_pipe() will
1395 * grab the pipe inode lock. For sendfile() emulation,
1396 * we call into ->sendpage() with the i_mutex lock held
1397 * and networking will grab the socket lock.
1398 */
1399 release_sock(__skb->sk);
1400 ret = splice_to_pipe(pipe, &spd);
1401 lock_sock(__skb->sk);
1402 return ret;
1403 }
1404
1405 return 0;
1406}
1407
Herbert Xu357b40a2005-04-19 22:30:14 -07001408/**
1409 * skb_store_bits - store bits from kernel buffer to skb
1410 * @skb: destination buffer
1411 * @offset: offset in destination
1412 * @from: source buffer
1413 * @len: number of bytes to copy
1414 *
1415 * Copy the specified number of bytes from the source buffer to the
1416 * destination skb. This function handles all the messy bits of
1417 * traversing fragment lists and such.
1418 */
1419
Stephen Hemminger0c6fcc82007-04-20 16:40:01 -07001420int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
Herbert Xu357b40a2005-04-19 22:30:14 -07001421{
1422 int i, copy;
David S. Miller1a028e52007-04-27 15:21:23 -07001423 int start = skb_headlen(skb);
Herbert Xu357b40a2005-04-19 22:30:14 -07001424
1425 if (offset > (int)skb->len - len)
1426 goto fault;
1427
David S. Miller1a028e52007-04-27 15:21:23 -07001428 if ((copy = start - offset) > 0) {
Herbert Xu357b40a2005-04-19 22:30:14 -07001429 if (copy > len)
1430 copy = len;
Arnaldo Carvalho de Melo27d7ff42007-03-31 11:55:19 -03001431 skb_copy_to_linear_data_offset(skb, offset, from, copy);
Herbert Xu357b40a2005-04-19 22:30:14 -07001432 if ((len -= copy) == 0)
1433 return 0;
1434 offset += copy;
1435 from += copy;
1436 }
1437
1438 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1439 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
David S. Miller1a028e52007-04-27 15:21:23 -07001440 int end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001441
David S. Miller1a028e52007-04-27 15:21:23 -07001442 BUG_TRAP(start <= offset + len);
1443
1444 end = start + frag->size;
Herbert Xu357b40a2005-04-19 22:30:14 -07001445 if ((copy = end - offset) > 0) {
1446 u8 *vaddr;
1447
1448 if (copy > len)
1449 copy = len;
1450
1451 vaddr = kmap_skb_frag(frag);
David S. Miller1a028e52007-04-27 15:21:23 -07001452 memcpy(vaddr + frag->page_offset + offset - start,
1453 from, copy);
Herbert Xu357b40a2005-04-19 22:30:14 -07001454 kunmap_skb_frag(vaddr);
1455
1456 if ((len -= copy) == 0)
1457 return 0;
1458 offset += copy;
1459 from += copy;
1460 }
David S. Miller1a028e52007-04-27 15:21:23 -07001461 start = end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001462 }
1463
1464 if (skb_shinfo(skb)->frag_list) {
1465 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1466
1467 for (; list; list = list->next) {
David S. Miller1a028e52007-04-27 15:21:23 -07001468 int end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001469
David S. Miller1a028e52007-04-27 15:21:23 -07001470 BUG_TRAP(start <= offset + len);
1471
1472 end = start + list->len;
Herbert Xu357b40a2005-04-19 22:30:14 -07001473 if ((copy = end - offset) > 0) {
1474 if (copy > len)
1475 copy = len;
David S. Miller1a028e52007-04-27 15:21:23 -07001476 if (skb_store_bits(list, offset - start,
1477 from, copy))
Herbert Xu357b40a2005-04-19 22:30:14 -07001478 goto fault;
1479 if ((len -= copy) == 0)
1480 return 0;
1481 offset += copy;
1482 from += copy;
1483 }
David S. Miller1a028e52007-04-27 15:21:23 -07001484 start = end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001485 }
1486 }
1487 if (!len)
1488 return 0;
1489
1490fault:
1491 return -EFAULT;
1492}
1493
1494EXPORT_SYMBOL(skb_store_bits);
1495
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496/* Checksum skb data. */
1497
Al Viro2bbbc862006-11-14 21:37:14 -08001498__wsum skb_checksum(const struct sk_buff *skb, int offset,
1499 int len, __wsum csum)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500{
David S. Miller1a028e52007-04-27 15:21:23 -07001501 int start = skb_headlen(skb);
1502 int i, copy = start - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 int pos = 0;
1504
1505 /* Checksum header. */
1506 if (copy > 0) {
1507 if (copy > len)
1508 copy = len;
1509 csum = csum_partial(skb->data + offset, copy, csum);
1510 if ((len -= copy) == 0)
1511 return csum;
1512 offset += copy;
1513 pos = copy;
1514 }
1515
1516 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07001517 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518
David S. Miller1a028e52007-04-27 15:21:23 -07001519 BUG_TRAP(start <= offset + len);
1520
1521 end = start + skb_shinfo(skb)->frags[i].size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 if ((copy = end - offset) > 0) {
Al Viro44bb9362006-11-14 21:36:14 -08001523 __wsum csum2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 u8 *vaddr;
1525 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1526
1527 if (copy > len)
1528 copy = len;
1529 vaddr = kmap_skb_frag(frag);
David S. Miller1a028e52007-04-27 15:21:23 -07001530 csum2 = csum_partial(vaddr + frag->page_offset +
1531 offset - start, copy, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 kunmap_skb_frag(vaddr);
1533 csum = csum_block_add(csum, csum2, pos);
1534 if (!(len -= copy))
1535 return csum;
1536 offset += copy;
1537 pos += copy;
1538 }
David S. Miller1a028e52007-04-27 15:21:23 -07001539 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 }
1541
1542 if (skb_shinfo(skb)->frag_list) {
1543 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1544
1545 for (; list; list = list->next) {
David S. Miller1a028e52007-04-27 15:21:23 -07001546 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547
David S. Miller1a028e52007-04-27 15:21:23 -07001548 BUG_TRAP(start <= offset + len);
1549
1550 end = start + list->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 if ((copy = end - offset) > 0) {
Al Viro5f92a732006-11-14 21:36:54 -08001552 __wsum csum2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 if (copy > len)
1554 copy = len;
David S. Miller1a028e52007-04-27 15:21:23 -07001555 csum2 = skb_checksum(list, offset - start,
1556 copy, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 csum = csum_block_add(csum, csum2, pos);
1558 if ((len -= copy) == 0)
1559 return csum;
1560 offset += copy;
1561 pos += copy;
1562 }
David S. Miller1a028e52007-04-27 15:21:23 -07001563 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 }
1565 }
Kris Katterjohn09a62662006-01-08 22:24:28 -08001566 BUG_ON(len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567
1568 return csum;
1569}
1570
1571/* Both of above in one bottle. */
1572
Al Viro81d77662006-11-14 21:37:33 -08001573__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1574 u8 *to, int len, __wsum csum)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575{
David S. Miller1a028e52007-04-27 15:21:23 -07001576 int start = skb_headlen(skb);
1577 int i, copy = start - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 int pos = 0;
1579
1580 /* Copy header. */
1581 if (copy > 0) {
1582 if (copy > len)
1583 copy = len;
1584 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1585 copy, csum);
1586 if ((len -= copy) == 0)
1587 return csum;
1588 offset += copy;
1589 to += copy;
1590 pos = copy;
1591 }
1592
1593 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07001594 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595
David S. Miller1a028e52007-04-27 15:21:23 -07001596 BUG_TRAP(start <= offset + len);
1597
1598 end = start + skb_shinfo(skb)->frags[i].size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 if ((copy = end - offset) > 0) {
Al Viro50842052006-11-14 21:36:34 -08001600 __wsum csum2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 u8 *vaddr;
1602 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1603
1604 if (copy > len)
1605 copy = len;
1606 vaddr = kmap_skb_frag(frag);
1607 csum2 = csum_partial_copy_nocheck(vaddr +
David S. Miller1a028e52007-04-27 15:21:23 -07001608 frag->page_offset +
1609 offset - start, to,
1610 copy, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 kunmap_skb_frag(vaddr);
1612 csum = csum_block_add(csum, csum2, pos);
1613 if (!(len -= copy))
1614 return csum;
1615 offset += copy;
1616 to += copy;
1617 pos += copy;
1618 }
David S. Miller1a028e52007-04-27 15:21:23 -07001619 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 }
1621
1622 if (skb_shinfo(skb)->frag_list) {
1623 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1624
1625 for (; list; list = list->next) {
Al Viro81d77662006-11-14 21:37:33 -08001626 __wsum csum2;
David S. Miller1a028e52007-04-27 15:21:23 -07001627 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628
David S. Miller1a028e52007-04-27 15:21:23 -07001629 BUG_TRAP(start <= offset + len);
1630
1631 end = start + list->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 if ((copy = end - offset) > 0) {
1633 if (copy > len)
1634 copy = len;
David S. Miller1a028e52007-04-27 15:21:23 -07001635 csum2 = skb_copy_and_csum_bits(list,
1636 offset - start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 to, copy, 0);
1638 csum = csum_block_add(csum, csum2, pos);
1639 if ((len -= copy) == 0)
1640 return csum;
1641 offset += copy;
1642 to += copy;
1643 pos += copy;
1644 }
David S. Miller1a028e52007-04-27 15:21:23 -07001645 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 }
1647 }
Kris Katterjohn09a62662006-01-08 22:24:28 -08001648 BUG_ON(len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 return csum;
1650}
1651
1652void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1653{
Al Virod3bc23e2006-11-14 21:24:49 -08001654 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 long csstart;
1656
Patrick McHardy84fa7932006-08-29 16:44:56 -07001657 if (skb->ip_summed == CHECKSUM_PARTIAL)
Herbert Xu663ead32007-04-09 11:59:07 -07001658 csstart = skb->csum_start - skb_headroom(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 else
1660 csstart = skb_headlen(skb);
1661
Kris Katterjohn09a62662006-01-08 22:24:28 -08001662 BUG_ON(csstart > skb_headlen(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03001664 skb_copy_from_linear_data(skb, to, csstart);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665
1666 csum = 0;
1667 if (csstart != skb->len)
1668 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1669 skb->len - csstart, 0);
1670
Patrick McHardy84fa7932006-08-29 16:44:56 -07001671 if (skb->ip_summed == CHECKSUM_PARTIAL) {
Al Viroff1dcad2006-11-20 18:07:29 -08001672 long csstuff = csstart + skb->csum_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673
Al Virod3bc23e2006-11-14 21:24:49 -08001674 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 }
1676}
1677
1678/**
1679 * skb_dequeue - remove from the head of the queue
1680 * @list: list to dequeue from
1681 *
1682 * Remove the head of the list. The list lock is taken so the function
1683 * may be used safely with other locking list functions. The head item is
1684 * returned or %NULL if the list is empty.
1685 */
1686
1687struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1688{
1689 unsigned long flags;
1690 struct sk_buff *result;
1691
1692 spin_lock_irqsave(&list->lock, flags);
1693 result = __skb_dequeue(list);
1694 spin_unlock_irqrestore(&list->lock, flags);
1695 return result;
1696}
1697
1698/**
1699 * skb_dequeue_tail - remove from the tail of the queue
1700 * @list: list to dequeue from
1701 *
1702 * Remove the tail of the list. The list lock is taken so the function
1703 * may be used safely with other locking list functions. The tail item is
1704 * returned or %NULL if the list is empty.
1705 */
1706struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1707{
1708 unsigned long flags;
1709 struct sk_buff *result;
1710
1711 spin_lock_irqsave(&list->lock, flags);
1712 result = __skb_dequeue_tail(list);
1713 spin_unlock_irqrestore(&list->lock, flags);
1714 return result;
1715}
1716
1717/**
1718 * skb_queue_purge - empty a list
1719 * @list: list to empty
1720 *
1721 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1722 * the list and one reference dropped. This function takes the list
1723 * lock and is atomic with respect to other list locking functions.
1724 */
1725void skb_queue_purge(struct sk_buff_head *list)
1726{
1727 struct sk_buff *skb;
1728 while ((skb = skb_dequeue(list)) != NULL)
1729 kfree_skb(skb);
1730}
1731
1732/**
1733 * skb_queue_head - queue a buffer at the list head
1734 * @list: list to use
1735 * @newsk: buffer to queue
1736 *
1737 * Queue a buffer at the start of the list. This function takes the
1738 * list lock and can be used safely with other locking &sk_buff functions
1739 * safely.
1740 *
1741 * A buffer cannot be placed on two lists at the same time.
1742 */
1743void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1744{
1745 unsigned long flags;
1746
1747 spin_lock_irqsave(&list->lock, flags);
1748 __skb_queue_head(list, newsk);
1749 spin_unlock_irqrestore(&list->lock, flags);
1750}
1751
1752/**
1753 * skb_queue_tail - queue a buffer at the list tail
1754 * @list: list to use
1755 * @newsk: buffer to queue
1756 *
1757 * Queue a buffer at the tail of the list. This function takes the
1758 * list lock and can be used safely with other locking &sk_buff functions
1759 * safely.
1760 *
1761 * A buffer cannot be placed on two lists at the same time.
1762 */
1763void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1764{
1765 unsigned long flags;
1766
1767 spin_lock_irqsave(&list->lock, flags);
1768 __skb_queue_tail(list, newsk);
1769 spin_unlock_irqrestore(&list->lock, flags);
1770}
David S. Miller8728b832005-08-09 19:25:21 -07001771
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772/**
1773 * skb_unlink - remove a buffer from a list
1774 * @skb: buffer to remove
David S. Miller8728b832005-08-09 19:25:21 -07001775 * @list: list to use
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 *
David S. Miller8728b832005-08-09 19:25:21 -07001777 * Remove a packet from a list. The list locks are taken and this
1778 * function is atomic with respect to other list locked calls
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 *
David S. Miller8728b832005-08-09 19:25:21 -07001780 * You must know what list the SKB is on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 */
David S. Miller8728b832005-08-09 19:25:21 -07001782void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783{
David S. Miller8728b832005-08-09 19:25:21 -07001784 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785
David S. Miller8728b832005-08-09 19:25:21 -07001786 spin_lock_irqsave(&list->lock, flags);
1787 __skb_unlink(skb, list);
1788 spin_unlock_irqrestore(&list->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789}
1790
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791/**
1792 * skb_append - append a buffer
1793 * @old: buffer to insert after
1794 * @newsk: buffer to insert
David S. Miller8728b832005-08-09 19:25:21 -07001795 * @list: list to use
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 *
1797 * Place a packet after a given packet in a list. The list locks are taken
1798 * and this function is atomic with respect to other list locked calls.
1799 * A buffer cannot be placed on two lists at the same time.
1800 */
David S. Miller8728b832005-08-09 19:25:21 -07001801void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802{
1803 unsigned long flags;
1804
David S. Miller8728b832005-08-09 19:25:21 -07001805 spin_lock_irqsave(&list->lock, flags);
1806 __skb_append(old, newsk, list);
1807 spin_unlock_irqrestore(&list->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808}
1809
1810
1811/**
1812 * skb_insert - insert a buffer
1813 * @old: buffer to insert before
1814 * @newsk: buffer to insert
David S. Miller8728b832005-08-09 19:25:21 -07001815 * @list: list to use
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 *
David S. Miller8728b832005-08-09 19:25:21 -07001817 * Place a packet before a given packet in a list. The list locks are
1818 * taken and this function is atomic with respect to other list locked
1819 * calls.
1820 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 * A buffer cannot be placed on two lists at the same time.
1822 */
David S. Miller8728b832005-08-09 19:25:21 -07001823void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824{
1825 unsigned long flags;
1826
David S. Miller8728b832005-08-09 19:25:21 -07001827 spin_lock_irqsave(&list->lock, flags);
1828 __skb_insert(newsk, old->prev, old, list);
1829 spin_unlock_irqrestore(&list->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830}
1831
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832static inline void skb_split_inside_header(struct sk_buff *skb,
1833 struct sk_buff* skb1,
1834 const u32 len, const int pos)
1835{
1836 int i;
1837
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03001838 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
1839 pos - len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 /* And move data appendix as is. */
1841 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1842 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
1843
1844 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
1845 skb_shinfo(skb)->nr_frags = 0;
1846 skb1->data_len = skb->data_len;
1847 skb1->len += skb1->data_len;
1848 skb->data_len = 0;
1849 skb->len = len;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001850 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851}
1852
1853static inline void skb_split_no_header(struct sk_buff *skb,
1854 struct sk_buff* skb1,
1855 const u32 len, int pos)
1856{
1857 int i, k = 0;
1858 const int nfrags = skb_shinfo(skb)->nr_frags;
1859
1860 skb_shinfo(skb)->nr_frags = 0;
1861 skb1->len = skb1->data_len = skb->len - len;
1862 skb->len = len;
1863 skb->data_len = len - pos;
1864
1865 for (i = 0; i < nfrags; i++) {
1866 int size = skb_shinfo(skb)->frags[i].size;
1867
1868 if (pos + size > len) {
1869 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
1870
1871 if (pos < len) {
1872 /* Split frag.
1873 * We have two variants in this case:
1874 * 1. Move all the frag to the second
1875 * part, if it is possible. F.e.
1876 * this approach is mandatory for TUX,
1877 * where splitting is expensive.
1878 * 2. Split is accurately. We make this.
1879 */
1880 get_page(skb_shinfo(skb)->frags[i].page);
1881 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
1882 skb_shinfo(skb1)->frags[0].size -= len - pos;
1883 skb_shinfo(skb)->frags[i].size = len - pos;
1884 skb_shinfo(skb)->nr_frags++;
1885 }
1886 k++;
1887 } else
1888 skb_shinfo(skb)->nr_frags++;
1889 pos += size;
1890 }
1891 skb_shinfo(skb1)->nr_frags = k;
1892}
1893
1894/**
1895 * skb_split - Split fragmented skb to two parts at length len.
1896 * @skb: the buffer to split
1897 * @skb1: the buffer to receive the second part
1898 * @len: new length for skb
1899 */
1900void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
1901{
1902 int pos = skb_headlen(skb);
1903
1904 if (len < pos) /* Split line is inside header. */
1905 skb_split_inside_header(skb, skb1, len, pos);
1906 else /* Second chunk has no header, nothing to copy. */
1907 skb_split_no_header(skb, skb1, len, pos);
1908}
1909
Thomas Graf677e90e2005-06-23 20:59:51 -07001910/**
1911 * skb_prepare_seq_read - Prepare a sequential read of skb data
1912 * @skb: the buffer to read
1913 * @from: lower offset of data to be read
1914 * @to: upper offset of data to be read
1915 * @st: state variable
1916 *
1917 * Initializes the specified state variable. Must be called before
1918 * invoking skb_seq_read() for the first time.
1919 */
1920void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1921 unsigned int to, struct skb_seq_state *st)
1922{
1923 st->lower_offset = from;
1924 st->upper_offset = to;
1925 st->root_skb = st->cur_skb = skb;
1926 st->frag_idx = st->stepped_offset = 0;
1927 st->frag_data = NULL;
1928}
1929
1930/**
1931 * skb_seq_read - Sequentially read skb data
1932 * @consumed: number of bytes consumed by the caller so far
1933 * @data: destination pointer for data to be returned
1934 * @st: state variable
1935 *
1936 * Reads a block of skb data at &consumed relative to the
1937 * lower offset specified to skb_prepare_seq_read(). Assigns
1938 * the head of the data block to &data and returns the length
1939 * of the block or 0 if the end of the skb data or the upper
1940 * offset has been reached.
1941 *
1942 * The caller is not required to consume all of the data
1943 * returned, i.e. &consumed is typically set to the number
1944 * of bytes already consumed and the next call to
1945 * skb_seq_read() will return the remaining part of the block.
1946 *
Randy Dunlapbc2cda12008-02-13 15:03:25 -08001947 * Note 1: The size of each block of data returned can be arbitary,
Thomas Graf677e90e2005-06-23 20:59:51 -07001948 * this limitation is the cost for zerocopy seqeuental
1949 * reads of potentially non linear data.
1950 *
Randy Dunlapbc2cda12008-02-13 15:03:25 -08001951 * Note 2: Fragment lists within fragments are not implemented
Thomas Graf677e90e2005-06-23 20:59:51 -07001952 * at the moment, state->root_skb could be replaced with
1953 * a stack for this purpose.
1954 */
1955unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1956 struct skb_seq_state *st)
1957{
1958 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
1959 skb_frag_t *frag;
1960
1961 if (unlikely(abs_offset >= st->upper_offset))
1962 return 0;
1963
1964next_skb:
1965 block_limit = skb_headlen(st->cur_skb);
1966
1967 if (abs_offset < block_limit) {
1968 *data = st->cur_skb->data + abs_offset;
1969 return block_limit - abs_offset;
1970 }
1971
1972 if (st->frag_idx == 0 && !st->frag_data)
1973 st->stepped_offset += skb_headlen(st->cur_skb);
1974
1975 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
1976 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
1977 block_limit = frag->size + st->stepped_offset;
1978
1979 if (abs_offset < block_limit) {
1980 if (!st->frag_data)
1981 st->frag_data = kmap_skb_frag(frag);
1982
1983 *data = (u8 *) st->frag_data + frag->page_offset +
1984 (abs_offset - st->stepped_offset);
1985
1986 return block_limit - abs_offset;
1987 }
1988
1989 if (st->frag_data) {
1990 kunmap_skb_frag(st->frag_data);
1991 st->frag_data = NULL;
1992 }
1993
1994 st->frag_idx++;
1995 st->stepped_offset += frag->size;
1996 }
1997
Olaf Kirch5b5a60d2007-06-23 23:11:52 -07001998 if (st->frag_data) {
1999 kunmap_skb_frag(st->frag_data);
2000 st->frag_data = NULL;
2001 }
2002
Thomas Graf677e90e2005-06-23 20:59:51 -07002003 if (st->cur_skb->next) {
2004 st->cur_skb = st->cur_skb->next;
2005 st->frag_idx = 0;
2006 goto next_skb;
2007 } else if (st->root_skb == st->cur_skb &&
2008 skb_shinfo(st->root_skb)->frag_list) {
2009 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2010 goto next_skb;
2011 }
2012
2013 return 0;
2014}
2015
2016/**
2017 * skb_abort_seq_read - Abort a sequential read of skb data
2018 * @st: state variable
2019 *
2020 * Must be called if skb_seq_read() was not called until it
2021 * returned 0.
2022 */
2023void skb_abort_seq_read(struct skb_seq_state *st)
2024{
2025 if (st->frag_data)
2026 kunmap_skb_frag(st->frag_data);
2027}
2028
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002029#define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2030
2031static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2032 struct ts_config *conf,
2033 struct ts_state *state)
2034{
2035 return skb_seq_read(offset, text, TS_SKB_CB(state));
2036}
2037
2038static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2039{
2040 skb_abort_seq_read(TS_SKB_CB(state));
2041}
2042
2043/**
2044 * skb_find_text - Find a text pattern in skb data
2045 * @skb: the buffer to look in
2046 * @from: search offset
2047 * @to: search limit
2048 * @config: textsearch configuration
2049 * @state: uninitialized textsearch state variable
2050 *
2051 * Finds a pattern in the skb data according to the specified
2052 * textsearch configuration. Use textsearch_next() to retrieve
2053 * subsequent occurrences of the pattern. Returns the offset
2054 * to the first occurrence or UINT_MAX if no match was found.
2055 */
2056unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2057 unsigned int to, struct ts_config *config,
2058 struct ts_state *state)
2059{
Phil Oesterf72b9482006-06-26 00:00:57 -07002060 unsigned int ret;
2061
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002062 config->get_next_block = skb_ts_get_next_block;
2063 config->finish = skb_ts_finish;
2064
2065 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2066
Phil Oesterf72b9482006-06-26 00:00:57 -07002067 ret = textsearch_find(config, state);
2068 return (ret <= to - from ? ret : UINT_MAX);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002069}
2070
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002071/**
2072 * skb_append_datato_frags: - append the user data to a skb
2073 * @sk: sock structure
2074 * @skb: skb structure to be appened with user data.
2075 * @getfrag: call back function to be used for getting the user data
2076 * @from: pointer to user message iov
2077 * @length: length of the iov message
2078 *
2079 * Description: This procedure append the user data in the fragment part
2080 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2081 */
2082int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
Martin Waitzdab96302005-12-05 13:40:12 -08002083 int (*getfrag)(void *from, char *to, int offset,
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002084 int len, int odd, struct sk_buff *skb),
2085 void *from, int length)
2086{
2087 int frg_cnt = 0;
2088 skb_frag_t *frag = NULL;
2089 struct page *page = NULL;
2090 int copy, left;
2091 int offset = 0;
2092 int ret;
2093
2094 do {
2095 /* Return error if we don't have space for new frag */
2096 frg_cnt = skb_shinfo(skb)->nr_frags;
2097 if (frg_cnt >= MAX_SKB_FRAGS)
2098 return -EFAULT;
2099
2100 /* allocate a new page for next frag */
2101 page = alloc_pages(sk->sk_allocation, 0);
2102
2103 /* If alloc_page fails just return failure and caller will
2104 * free previous allocated pages by doing kfree_skb()
2105 */
2106 if (page == NULL)
2107 return -ENOMEM;
2108
2109 /* initialize the next frag */
2110 sk->sk_sndmsg_page = page;
2111 sk->sk_sndmsg_off = 0;
2112 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2113 skb->truesize += PAGE_SIZE;
2114 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2115
2116 /* get the new initialized frag */
2117 frg_cnt = skb_shinfo(skb)->nr_frags;
2118 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2119
2120 /* copy the user data to page */
2121 left = PAGE_SIZE - frag->page_offset;
2122 copy = (length > left)? left : length;
2123
2124 ret = getfrag(from, (page_address(frag->page) +
2125 frag->page_offset + frag->size),
2126 offset, copy, 0, skb);
2127 if (ret < 0)
2128 return -EFAULT;
2129
2130 /* copy was successful so update the size parameters */
2131 sk->sk_sndmsg_off += copy;
2132 frag->size += copy;
2133 skb->len += copy;
2134 skb->data_len += copy;
2135 offset += copy;
2136 length -= copy;
2137
2138 } while (length > 0);
2139
2140 return 0;
2141}
2142
Herbert Xucbb042f2006-03-20 22:43:56 -08002143/**
2144 * skb_pull_rcsum - pull skb and update receive checksum
2145 * @skb: buffer to update
Herbert Xucbb042f2006-03-20 22:43:56 -08002146 * @len: length of data pulled
2147 *
2148 * This function performs an skb_pull on the packet and updates
Urs Thuermannfee54fa2008-02-12 22:03:25 -08002149 * the CHECKSUM_COMPLETE checksum. It should be used on
Patrick McHardy84fa7932006-08-29 16:44:56 -07002150 * receive path processing instead of skb_pull unless you know
2151 * that the checksum difference is zero (e.g., a valid IP header)
2152 * or you are setting ip_summed to CHECKSUM_NONE.
Herbert Xucbb042f2006-03-20 22:43:56 -08002153 */
2154unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2155{
2156 BUG_ON(len > skb->len);
2157 skb->len -= len;
2158 BUG_ON(skb->len < skb->data_len);
2159 skb_postpull_rcsum(skb, skb->data, len);
2160 return skb->data += len;
2161}
2162
Arnaldo Carvalho de Melof94691a2006-03-20 22:47:55 -08002163EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2164
Herbert Xuf4c50d92006-06-22 03:02:40 -07002165/**
2166 * skb_segment - Perform protocol segmentation on skb.
2167 * @skb: buffer to segment
Herbert Xu576a30e2006-06-27 13:22:38 -07002168 * @features: features for the output path (see dev->features)
Herbert Xuf4c50d92006-06-22 03:02:40 -07002169 *
2170 * This function performs segmentation on the given skb. It returns
2171 * the segment at the given position. It returns NULL if there are
2172 * no more segments to generate, or when an error is encountered.
2173 */
Herbert Xu576a30e2006-06-27 13:22:38 -07002174struct sk_buff *skb_segment(struct sk_buff *skb, int features)
Herbert Xuf4c50d92006-06-22 03:02:40 -07002175{
2176 struct sk_buff *segs = NULL;
2177 struct sk_buff *tail = NULL;
2178 unsigned int mss = skb_shinfo(skb)->gso_size;
Arnaldo Carvalho de Melo98e399f2007-03-19 15:33:04 -07002179 unsigned int doffset = skb->data - skb_mac_header(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002180 unsigned int offset = doffset;
2181 unsigned int headroom;
2182 unsigned int len;
Herbert Xu576a30e2006-06-27 13:22:38 -07002183 int sg = features & NETIF_F_SG;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002184 int nfrags = skb_shinfo(skb)->nr_frags;
2185 int err = -ENOMEM;
2186 int i = 0;
2187 int pos;
2188
2189 __skb_push(skb, doffset);
2190 headroom = skb_headroom(skb);
2191 pos = skb_headlen(skb);
2192
2193 do {
2194 struct sk_buff *nskb;
2195 skb_frag_t *frag;
Herbert Xuc8884ed2006-10-29 15:59:41 -08002196 int hsize;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002197 int k;
2198 int size;
2199
2200 len = skb->len - offset;
2201 if (len > mss)
2202 len = mss;
2203
2204 hsize = skb_headlen(skb) - offset;
2205 if (hsize < 0)
2206 hsize = 0;
Herbert Xuc8884ed2006-10-29 15:59:41 -08002207 if (hsize > len || !sg)
2208 hsize = len;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002209
Herbert Xuc8884ed2006-10-29 15:59:41 -08002210 nskb = alloc_skb(hsize + doffset + headroom, GFP_ATOMIC);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002211 if (unlikely(!nskb))
2212 goto err;
2213
2214 if (segs)
2215 tail->next = nskb;
2216 else
2217 segs = nskb;
2218 tail = nskb;
2219
2220 nskb->dev = skb->dev;
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07002221 skb_copy_queue_mapping(nskb, skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002222 nskb->priority = skb->priority;
2223 nskb->protocol = skb->protocol;
2224 nskb->dst = dst_clone(skb->dst);
2225 memcpy(nskb->cb, skb->cb, sizeof(skb->cb));
2226 nskb->pkt_type = skb->pkt_type;
2227 nskb->mac_len = skb->mac_len;
2228
2229 skb_reserve(nskb, headroom);
Arnaldo Carvalho de Melo459a98e2007-03-19 15:30:44 -07002230 skb_reset_mac_header(nskb);
Arnaldo Carvalho de Meloddc7b8e2007-03-15 21:42:27 -03002231 skb_set_network_header(nskb, skb->mac_len);
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -07002232 nskb->transport_header = (nskb->network_header +
2233 skb_network_header_len(skb));
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03002234 skb_copy_from_linear_data(skb, skb_put(nskb, doffset),
2235 doffset);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002236 if (!sg) {
2237 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2238 skb_put(nskb, len),
2239 len, 0);
2240 continue;
2241 }
2242
2243 frag = skb_shinfo(nskb)->frags;
2244 k = 0;
2245
Patrick McHardy84fa7932006-08-29 16:44:56 -07002246 nskb->ip_summed = CHECKSUM_PARTIAL;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002247 nskb->csum = skb->csum;
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03002248 skb_copy_from_linear_data_offset(skb, offset,
2249 skb_put(nskb, hsize), hsize);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002250
2251 while (pos < offset + len) {
2252 BUG_ON(i >= nfrags);
2253
2254 *frag = skb_shinfo(skb)->frags[i];
2255 get_page(frag->page);
2256 size = frag->size;
2257
2258 if (pos < offset) {
2259 frag->page_offset += offset - pos;
2260 frag->size -= offset - pos;
2261 }
2262
2263 k++;
2264
2265 if (pos + size <= offset + len) {
2266 i++;
2267 pos += size;
2268 } else {
2269 frag->size -= pos + size - (offset + len);
2270 break;
2271 }
2272
2273 frag++;
2274 }
2275
2276 skb_shinfo(nskb)->nr_frags = k;
2277 nskb->data_len = len - hsize;
2278 nskb->len += nskb->data_len;
2279 nskb->truesize += nskb->data_len;
2280 } while ((offset += len) < skb->len);
2281
2282 return segs;
2283
2284err:
2285 while ((skb = segs)) {
2286 segs = skb->next;
Patrick McHardyb08d5842007-02-27 09:57:37 -08002287 kfree_skb(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002288 }
2289 return ERR_PTR(err);
2290}
2291
2292EXPORT_SYMBOL_GPL(skb_segment);
2293
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294void __init skb_init(void)
2295{
2296 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2297 sizeof(struct sk_buff),
2298 0,
Alexey Dobriyane5d679f332006-08-26 19:25:52 -07002299 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
Paul Mundt20c2df82007-07-20 10:11:58 +09002300 NULL);
David S. Millerd179cd12005-08-17 14:57:30 -07002301 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2302 (2*sizeof(struct sk_buff)) +
2303 sizeof(atomic_t),
2304 0,
Alexey Dobriyane5d679f332006-08-26 19:25:52 -07002305 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
Paul Mundt20c2df82007-07-20 10:11:58 +09002306 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307}
2308
David Howells716ea3a2007-04-02 20:19:53 -07002309/**
2310 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2311 * @skb: Socket buffer containing the buffers to be mapped
2312 * @sg: The scatter-gather list to map into
2313 * @offset: The offset into the buffer's contents to start mapping
2314 * @len: Length of buffer space to be mapped
2315 *
2316 * Fill the specified scatter-gather list with mappings/pointers into a
2317 * region of the buffer space attached to a socket buffer.
2318 */
David S. Miller51c739d2007-10-30 21:29:29 -07002319static int
2320__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
David Howells716ea3a2007-04-02 20:19:53 -07002321{
David S. Miller1a028e52007-04-27 15:21:23 -07002322 int start = skb_headlen(skb);
2323 int i, copy = start - offset;
David Howells716ea3a2007-04-02 20:19:53 -07002324 int elt = 0;
2325
2326 if (copy > 0) {
2327 if (copy > len)
2328 copy = len;
Jens Axboe642f1492007-10-24 11:20:47 +02002329 sg_set_buf(sg, skb->data + offset, copy);
David Howells716ea3a2007-04-02 20:19:53 -07002330 elt++;
2331 if ((len -= copy) == 0)
2332 return elt;
2333 offset += copy;
2334 }
2335
2336 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07002337 int end;
David Howells716ea3a2007-04-02 20:19:53 -07002338
David S. Miller1a028e52007-04-27 15:21:23 -07002339 BUG_TRAP(start <= offset + len);
2340
2341 end = start + skb_shinfo(skb)->frags[i].size;
David Howells716ea3a2007-04-02 20:19:53 -07002342 if ((copy = end - offset) > 0) {
2343 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2344
2345 if (copy > len)
2346 copy = len;
Jens Axboe642f1492007-10-24 11:20:47 +02002347 sg_set_page(&sg[elt], frag->page, copy,
2348 frag->page_offset+offset-start);
David Howells716ea3a2007-04-02 20:19:53 -07002349 elt++;
2350 if (!(len -= copy))
2351 return elt;
2352 offset += copy;
2353 }
David S. Miller1a028e52007-04-27 15:21:23 -07002354 start = end;
David Howells716ea3a2007-04-02 20:19:53 -07002355 }
2356
2357 if (skb_shinfo(skb)->frag_list) {
2358 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2359
2360 for (; list; list = list->next) {
David S. Miller1a028e52007-04-27 15:21:23 -07002361 int end;
David Howells716ea3a2007-04-02 20:19:53 -07002362
David S. Miller1a028e52007-04-27 15:21:23 -07002363 BUG_TRAP(start <= offset + len);
2364
2365 end = start + list->len;
David Howells716ea3a2007-04-02 20:19:53 -07002366 if ((copy = end - offset) > 0) {
2367 if (copy > len)
2368 copy = len;
David S. Miller51c739d2007-10-30 21:29:29 -07002369 elt += __skb_to_sgvec(list, sg+elt, offset - start,
2370 copy);
David Howells716ea3a2007-04-02 20:19:53 -07002371 if ((len -= copy) == 0)
2372 return elt;
2373 offset += copy;
2374 }
David S. Miller1a028e52007-04-27 15:21:23 -07002375 start = end;
David Howells716ea3a2007-04-02 20:19:53 -07002376 }
2377 }
2378 BUG_ON(len);
2379 return elt;
2380}
2381
David S. Miller51c739d2007-10-30 21:29:29 -07002382int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2383{
2384 int nsg = __skb_to_sgvec(skb, sg, offset, len);
2385
Jens Axboec46f2332007-10-31 12:06:37 +01002386 sg_mark_end(&sg[nsg - 1]);
David S. Miller51c739d2007-10-30 21:29:29 -07002387
2388 return nsg;
2389}
2390
David Howells716ea3a2007-04-02 20:19:53 -07002391/**
2392 * skb_cow_data - Check that a socket buffer's data buffers are writable
2393 * @skb: The socket buffer to check.
2394 * @tailbits: Amount of trailing space to be added
2395 * @trailer: Returned pointer to the skb where the @tailbits space begins
2396 *
2397 * Make sure that the data buffers attached to a socket buffer are
2398 * writable. If they are not, private copies are made of the data buffers
2399 * and the socket buffer is set to use these instead.
2400 *
2401 * If @tailbits is given, make sure that there is space to write @tailbits
2402 * bytes of data beyond current end of socket buffer. @trailer will be
2403 * set to point to the skb in which this space begins.
2404 *
2405 * The number of scatterlist elements required to completely map the
2406 * COW'd and extended socket buffer will be returned.
2407 */
2408int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
2409{
2410 int copyflag;
2411 int elt;
2412 struct sk_buff *skb1, **skb_p;
2413
2414 /* If skb is cloned or its head is paged, reallocate
2415 * head pulling out all the pages (pages are considered not writable
2416 * at the moment even if they are anonymous).
2417 */
2418 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
2419 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
2420 return -ENOMEM;
2421
2422 /* Easy case. Most of packets will go this way. */
2423 if (!skb_shinfo(skb)->frag_list) {
2424 /* A little of trouble, not enough of space for trailer.
2425 * This should not happen, when stack is tuned to generate
2426 * good frames. OK, on miss we reallocate and reserve even more
2427 * space, 128 bytes is fair. */
2428
2429 if (skb_tailroom(skb) < tailbits &&
2430 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
2431 return -ENOMEM;
2432
2433 /* Voila! */
2434 *trailer = skb;
2435 return 1;
2436 }
2437
2438 /* Misery. We are in troubles, going to mincer fragments... */
2439
2440 elt = 1;
2441 skb_p = &skb_shinfo(skb)->frag_list;
2442 copyflag = 0;
2443
2444 while ((skb1 = *skb_p) != NULL) {
2445 int ntail = 0;
2446
2447 /* The fragment is partially pulled by someone,
2448 * this can happen on input. Copy it and everything
2449 * after it. */
2450
2451 if (skb_shared(skb1))
2452 copyflag = 1;
2453
2454 /* If the skb is the last, worry about trailer. */
2455
2456 if (skb1->next == NULL && tailbits) {
2457 if (skb_shinfo(skb1)->nr_frags ||
2458 skb_shinfo(skb1)->frag_list ||
2459 skb_tailroom(skb1) < tailbits)
2460 ntail = tailbits + 128;
2461 }
2462
2463 if (copyflag ||
2464 skb_cloned(skb1) ||
2465 ntail ||
2466 skb_shinfo(skb1)->nr_frags ||
2467 skb_shinfo(skb1)->frag_list) {
2468 struct sk_buff *skb2;
2469
2470 /* Fuck, we are miserable poor guys... */
2471 if (ntail == 0)
2472 skb2 = skb_copy(skb1, GFP_ATOMIC);
2473 else
2474 skb2 = skb_copy_expand(skb1,
2475 skb_headroom(skb1),
2476 ntail,
2477 GFP_ATOMIC);
2478 if (unlikely(skb2 == NULL))
2479 return -ENOMEM;
2480
2481 if (skb1->sk)
2482 skb_set_owner_w(skb2, skb1->sk);
2483
2484 /* Looking around. Are we still alive?
2485 * OK, link new skb, drop old one */
2486
2487 skb2->next = skb1->next;
2488 *skb_p = skb2;
2489 kfree_skb(skb1);
2490 skb1 = skb2;
2491 }
2492 elt++;
2493 *trailer = skb1;
2494 skb_p = &skb1->next;
2495 }
2496
2497 return elt;
2498}
2499
Rusty Russellf35d9d82008-02-04 23:49:54 -05002500/**
2501 * skb_partial_csum_set - set up and verify partial csum values for packet
2502 * @skb: the skb to set
2503 * @start: the number of bytes after skb->data to start checksumming.
2504 * @off: the offset from start to place the checksum.
2505 *
2506 * For untrusted partially-checksummed packets, we need to make sure the values
2507 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
2508 *
2509 * This function checks and sets those values and skb->ip_summed: if this
2510 * returns false you should drop the packet.
2511 */
2512bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
2513{
2514 if (unlikely(start > skb->len - 2) ||
2515 unlikely((int)start + off > skb->len - 2)) {
2516 if (net_ratelimit())
2517 printk(KERN_WARNING
2518 "bad partial csum: csum=%u/%u len=%u\n",
2519 start, off, skb->len);
2520 return false;
2521 }
2522 skb->ip_summed = CHECKSUM_PARTIAL;
2523 skb->csum_start = skb_headroom(skb) + start;
2524 skb->csum_offset = off;
2525 return true;
2526}
2527
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528EXPORT_SYMBOL(___pskb_trim);
2529EXPORT_SYMBOL(__kfree_skb);
Jörn Engel231d06a2006-03-20 21:28:35 -08002530EXPORT_SYMBOL(kfree_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531EXPORT_SYMBOL(__pskb_pull_tail);
David S. Millerd179cd12005-08-17 14:57:30 -07002532EXPORT_SYMBOL(__alloc_skb);
Christoph Hellwig8af27452006-07-31 22:35:23 -07002533EXPORT_SYMBOL(__netdev_alloc_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534EXPORT_SYMBOL(pskb_copy);
2535EXPORT_SYMBOL(pskb_expand_head);
2536EXPORT_SYMBOL(skb_checksum);
2537EXPORT_SYMBOL(skb_clone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538EXPORT_SYMBOL(skb_copy);
2539EXPORT_SYMBOL(skb_copy_and_csum_bits);
2540EXPORT_SYMBOL(skb_copy_and_csum_dev);
2541EXPORT_SYMBOL(skb_copy_bits);
2542EXPORT_SYMBOL(skb_copy_expand);
2543EXPORT_SYMBOL(skb_over_panic);
2544EXPORT_SYMBOL(skb_pad);
2545EXPORT_SYMBOL(skb_realloc_headroom);
2546EXPORT_SYMBOL(skb_under_panic);
2547EXPORT_SYMBOL(skb_dequeue);
2548EXPORT_SYMBOL(skb_dequeue_tail);
2549EXPORT_SYMBOL(skb_insert);
2550EXPORT_SYMBOL(skb_queue_purge);
2551EXPORT_SYMBOL(skb_queue_head);
2552EXPORT_SYMBOL(skb_queue_tail);
2553EXPORT_SYMBOL(skb_unlink);
2554EXPORT_SYMBOL(skb_append);
2555EXPORT_SYMBOL(skb_split);
Thomas Graf677e90e2005-06-23 20:59:51 -07002556EXPORT_SYMBOL(skb_prepare_seq_read);
2557EXPORT_SYMBOL(skb_seq_read);
2558EXPORT_SYMBOL(skb_abort_seq_read);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002559EXPORT_SYMBOL(skb_find_text);
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002560EXPORT_SYMBOL(skb_append_datato_frags);
David Howells716ea3a2007-04-02 20:19:53 -07002561
2562EXPORT_SYMBOL_GPL(skb_to_sgvec);
2563EXPORT_SYMBOL_GPL(skb_cow_data);
Rusty Russellf35d9d82008-02-04 23:49:54 -05002564EXPORT_SYMBOL_GPL(skb_partial_csum_set);