blob: 74ebf4b5258a3f552c38836dd67469b86d0e64f2 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Routines having to do with the 'struct sk_buff' memory handlers.
3 *
Alan Cox113aa832008-10-13 19:01:08 -07004 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
Linus Torvalds1da177e2005-04-16 15:20:36 -07005 * Florian La Roche <rzsfl@rz.uni-sb.de>
6 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * Fixes:
8 * Alan Cox : Fixed the worst of the load
9 * balancer bugs.
10 * Dave Platt : Interrupt stacking fix.
11 * Richard Kooijman : Timestamp fixes.
12 * Alan Cox : Changed buffer format.
13 * Alan Cox : destructor hook for AF_UNIX etc.
14 * Linus Torvalds : Better skb_clone.
15 * Alan Cox : Added skb_copy.
16 * Alan Cox : Added all the changed routines Linus
17 * only put in the headers
18 * Ray VanTassle : Fixed --skb->lock in free
19 * Alan Cox : skb_copy copy arp field
20 * Andi Kleen : slabified it.
21 * Robert Olsson : Removed skb_head_pool
22 *
23 * NOTE:
24 * The __skb_ routines should be called with interrupts
25 * disabled, or you better be *real* sure that the operation is atomic
26 * with respect to whatever list is being frobbed (e.g. via lock_sock()
27 * or via disabling bottom half handlers, etc).
28 *
29 * This program is free software; you can redistribute it and/or
30 * modify it under the terms of the GNU General Public License
31 * as published by the Free Software Foundation; either version
32 * 2 of the License, or (at your option) any later version.
33 */
34
35/*
36 * The functions in this file will not compile correctly with gcc 2.4.x
37 */
38
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include <linux/module.h>
40#include <linux/types.h>
41#include <linux/kernel.h>
Vegard Nossumfe55f6d2008-08-30 12:16:35 +020042#include <linux/kmemcheck.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include <linux/mm.h>
44#include <linux/interrupt.h>
45#include <linux/in.h>
46#include <linux/inet.h>
47#include <linux/slab.h>
48#include <linux/netdevice.h>
49#ifdef CONFIG_NET_CLS_ACT
50#include <net/pkt_sched.h>
51#endif
52#include <linux/string.h>
53#include <linux/skbuff.h>
Jens Axboe9c55e012007-11-06 23:30:13 -080054#include <linux/splice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070055#include <linux/cache.h>
56#include <linux/rtnetlink.h>
57#include <linux/init.h>
David Howells716ea3a2007-04-02 20:19:53 -070058#include <linux/scatterlist.h>
Patrick Ohlyac45f602009-02-12 05:03:37 +000059#include <linux/errqueue.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>
Steven Rostedtad8d75f2009-04-14 19:39:12 -040069#include <trace/events/skb.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070070
Al Viroa1f8e7f72006-10-19 16:08:53 -040071#include "kmap_skb.h"
72
Christoph Lametere18b8902006-12-06 20:33:20 -080073static struct kmem_cache *skbuff_head_cache __read_mostly;
74static struct kmem_cache *skbuff_fclone_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070075
Jens Axboe9c55e012007-11-06 23:30:13 -080076static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
77 struct pipe_buffer *buf)
78{
Jarek Poplawski8b9d3722009-01-19 17:03:56 -080079 put_page(buf->page);
Jens Axboe9c55e012007-11-06 23:30:13 -080080}
81
82static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
83 struct pipe_buffer *buf)
84{
Jarek Poplawski8b9d3722009-01-19 17:03:56 -080085 get_page(buf->page);
Jens Axboe9c55e012007-11-06 23:30:13 -080086}
87
88static int sock_pipe_buf_steal(struct pipe_inode_info *pipe,
89 struct pipe_buffer *buf)
90{
91 return 1;
92}
93
94
95/* Pipe buffer operations for a socket. */
Alexey Dobriyan28dfef82009-12-15 16:46:48 -080096static const struct pipe_buf_operations sock_pipe_buf_ops = {
Jens Axboe9c55e012007-11-06 23:30:13 -080097 .can_merge = 0,
98 .map = generic_pipe_buf_map,
99 .unmap = generic_pipe_buf_unmap,
100 .confirm = generic_pipe_buf_confirm,
101 .release = sock_pipe_buf_release,
102 .steal = sock_pipe_buf_steal,
103 .get = sock_pipe_buf_get,
104};
105
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106/*
107 * Keep out-of-line to prevent kernel bloat.
108 * __builtin_return_address is not used because it is not always
109 * reliable.
110 */
111
112/**
113 * skb_over_panic - private function
114 * @skb: buffer
115 * @sz: size
116 * @here: address
117 *
118 * Out of line support code for skb_put(). Not user callable.
119 */
Rami Rosenccb7c772010-04-20 22:39:53 -0700120static void skb_over_panic(struct sk_buff *skb, int sz, void *here)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121{
Patrick McHardy26095452005-04-21 16:43:02 -0700122 printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700123 "data:%p tail:%#lx end:%#lx dev:%s\n",
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700124 here, skb->len, sz, skb->head, skb->data,
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700125 (unsigned long)skb->tail, (unsigned long)skb->end,
Patrick McHardy26095452005-04-21 16:43:02 -0700126 skb->dev ? skb->dev->name : "<NULL>");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 BUG();
128}
129
130/**
131 * skb_under_panic - private function
132 * @skb: buffer
133 * @sz: size
134 * @here: address
135 *
136 * Out of line support code for skb_push(). Not user callable.
137 */
138
Rami Rosenccb7c772010-04-20 22:39:53 -0700139static void skb_under_panic(struct sk_buff *skb, int sz, void *here)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140{
Patrick McHardy26095452005-04-21 16:43:02 -0700141 printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700142 "data:%p tail:%#lx end:%#lx dev:%s\n",
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700143 here, skb->len, sz, skb->head, skb->data,
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700144 (unsigned long)skb->tail, (unsigned long)skb->end,
Patrick McHardy26095452005-04-21 16:43:02 -0700145 skb->dev ? skb->dev->name : "<NULL>");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 BUG();
147}
148
149/* Allocate a new skbuff. We do this ourselves so we can fill in a few
150 * 'private' fields and also do memory statistics to find all the
151 * [BEEP] leaks.
152 *
153 */
154
155/**
David S. Millerd179cd12005-08-17 14:57:30 -0700156 * __alloc_skb - allocate a network buffer
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 * @size: size to allocate
158 * @gfp_mask: allocation mask
Randy Dunlapc83c2482005-10-18 22:07:41 -0700159 * @fclone: allocate from fclone cache instead of head cache
160 * and allocate a cloned (child) skb
Christoph Hellwigb30973f2006-12-06 20:32:36 -0800161 * @node: numa node to allocate memory on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 *
163 * Allocate a new &sk_buff. The returned buffer has no headroom and a
164 * tail room of size bytes. The object has a reference count of one.
165 * The return is the buffer. On a failure the return is %NULL.
166 *
167 * Buffers may only be allocated from interrupts using a @gfp_mask of
168 * %GFP_ATOMIC.
169 */
Al Virodd0fc662005-10-07 07:46:04 +0100170struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
Christoph Hellwigb30973f2006-12-06 20:32:36 -0800171 int fclone, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172{
Christoph Lametere18b8902006-12-06 20:33:20 -0800173 struct kmem_cache *cache;
Benjamin LaHaise4947d3e2006-01-03 14:06:50 -0800174 struct skb_shared_info *shinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 struct sk_buff *skb;
176 u8 *data;
177
Herbert Xu8798b3f2006-01-23 16:32:45 -0800178 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
179
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 /* Get the HEAD */
Christoph Hellwigb30973f2006-12-06 20:32:36 -0800181 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 if (!skb)
183 goto out;
Eric Dumazetec7d2f22010-05-05 01:07:37 -0700184 prefetchw(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 size = SKB_DATA_ALIGN(size);
Christoph Hellwigb30973f2006-12-06 20:32:36 -0800187 data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info),
188 gfp_mask, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 if (!data)
190 goto nodata;
Eric Dumazetec7d2f22010-05-05 01:07:37 -0700191 prefetchw(data + size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192
Arnaldo Carvalho de Meloca0605a2007-03-19 10:48:59 -0300193 /*
Johannes Bergc8005782008-05-03 20:56:42 -0700194 * Only clear those fields we need to clear, not those that we will
195 * actually initialise below. Hence, don't put any more fields after
196 * the tail pointer in struct sk_buff!
Arnaldo Carvalho de Meloca0605a2007-03-19 10:48:59 -0300197 */
198 memset(skb, 0, offsetof(struct sk_buff, tail));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 skb->truesize = size + sizeof(struct sk_buff);
200 atomic_set(&skb->users, 1);
201 skb->head = data;
202 skb->data = data;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700203 skb_reset_tail_pointer(skb);
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700204 skb->end = skb->tail + size;
Stephen Hemminger19633e12009-06-17 05:23:27 +0000205#ifdef NET_SKBUFF_DATA_USES_OFFSET
206 skb->mac_header = ~0U;
207#endif
208
Benjamin LaHaise4947d3e2006-01-03 14:06:50 -0800209 /* make sure we initialize shinfo sequentially */
210 shinfo = skb_shinfo(skb);
Eric Dumazetec7d2f22010-05-05 01:07:37 -0700211 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
Benjamin LaHaise4947d3e2006-01-03 14:06:50 -0800212 atomic_set(&shinfo->dataref, 1);
Benjamin LaHaise4947d3e2006-01-03 14:06:50 -0800213
David S. Millerd179cd12005-08-17 14:57:30 -0700214 if (fclone) {
215 struct sk_buff *child = skb + 1;
216 atomic_t *fclone_ref = (atomic_t *) (child + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200218 kmemcheck_annotate_bitfield(child, flags1);
219 kmemcheck_annotate_bitfield(child, flags2);
David S. Millerd179cd12005-08-17 14:57:30 -0700220 skb->fclone = SKB_FCLONE_ORIG;
221 atomic_set(fclone_ref, 1);
222
223 child->fclone = SKB_FCLONE_UNAVAILABLE;
224 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225out:
226 return skb;
227nodata:
Herbert Xu8798b3f2006-01-23 16:32:45 -0800228 kmem_cache_free(cache, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 skb = NULL;
230 goto out;
231}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800232EXPORT_SYMBOL(__alloc_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233
234/**
Christoph Hellwig8af27452006-07-31 22:35:23 -0700235 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
236 * @dev: network device to receive on
237 * @length: length to allocate
238 * @gfp_mask: get_free_pages mask, passed to alloc_skb
239 *
240 * Allocate a new &sk_buff and assign it a usage count of one. The
241 * buffer has unspecified headroom built in. Users should allocate
242 * the headroom they think they need without accounting for the
243 * built in space. The built in space is used for optimisations.
244 *
245 * %NULL is returned if there is no free memory.
246 */
247struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
248 unsigned int length, gfp_t gfp_mask)
249{
250 struct sk_buff *skb;
251
Eric Dumazet564824b2010-10-11 19:05:25 +0000252 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, NUMA_NO_NODE);
Christoph Hellwig7b2e4972006-08-07 16:09:04 -0700253 if (likely(skb)) {
Christoph Hellwig8af27452006-07-31 22:35:23 -0700254 skb_reserve(skb, NET_SKB_PAD);
Christoph Hellwig7b2e4972006-08-07 16:09:04 -0700255 skb->dev = dev;
256 }
Christoph Hellwig8af27452006-07-31 22:35:23 -0700257 return skb;
258}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800259EXPORT_SYMBOL(__netdev_alloc_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
Peter Zijlstra654bed12008-10-07 14:22:33 -0700261void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
262 int size)
263{
264 skb_fill_page_desc(skb, i, page, off, size);
265 skb->len += size;
266 skb->data_len += size;
267 skb->truesize += size;
268}
269EXPORT_SYMBOL(skb_add_rx_frag);
270
Ilpo Järvinenf58518e2008-03-27 17:51:31 -0700271/**
272 * dev_alloc_skb - allocate an skbuff for receiving
273 * @length: length to allocate
274 *
275 * Allocate a new &sk_buff and assign it a usage count of one. The
276 * buffer has unspecified headroom built in. Users should allocate
277 * the headroom they think they need without accounting for the
278 * built in space. The built in space is used for optimisations.
279 *
280 * %NULL is returned if there is no free memory. Although this function
281 * allocates memory it can be called from an interrupt.
282 */
283struct sk_buff *dev_alloc_skb(unsigned int length)
284{
Denys Vlasenko1483b872008-03-28 15:57:39 -0700285 /*
286 * There is more code here than it seems:
David S. Millera0f55e02008-03-28 18:22:32 -0700287 * __dev_alloc_skb is an inline
Denys Vlasenko1483b872008-03-28 15:57:39 -0700288 */
Ilpo Järvinenf58518e2008-03-27 17:51:31 -0700289 return __dev_alloc_skb(length, GFP_ATOMIC);
290}
291EXPORT_SYMBOL(dev_alloc_skb);
292
Herbert Xu27b437c2006-07-13 19:26:39 -0700293static void skb_drop_list(struct sk_buff **listp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294{
Herbert Xu27b437c2006-07-13 19:26:39 -0700295 struct sk_buff *list = *listp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296
Herbert Xu27b437c2006-07-13 19:26:39 -0700297 *listp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298
299 do {
300 struct sk_buff *this = list;
301 list = list->next;
302 kfree_skb(this);
303 } while (list);
304}
305
Herbert Xu27b437c2006-07-13 19:26:39 -0700306static inline void skb_drop_fraglist(struct sk_buff *skb)
307{
308 skb_drop_list(&skb_shinfo(skb)->frag_list);
309}
310
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311static void skb_clone_fraglist(struct sk_buff *skb)
312{
313 struct sk_buff *list;
314
David S. Millerfbb398a2009-06-09 00:18:59 -0700315 skb_walk_frags(skb, list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 skb_get(list);
317}
318
Adrian Bunk5bba1712006-06-29 13:02:35 -0700319static void skb_release_data(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320{
321 if (!skb->cloned ||
322 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
323 &skb_shinfo(skb)->dataref)) {
324 if (skb_shinfo(skb)->nr_frags) {
325 int i;
326 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
327 put_page(skb_shinfo(skb)->frags[i].page);
328 }
329
David S. Miller21dc3302010-08-23 00:13:46 -0700330 if (skb_has_frag_list(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 skb_drop_fraglist(skb);
332
333 kfree(skb->head);
334 }
335}
336
337/*
338 * Free an skbuff by memory without cleaning the state.
339 */
Herbert Xu2d4baff2007-11-26 23:11:19 +0800340static void kfree_skbmem(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341{
David S. Millerd179cd12005-08-17 14:57:30 -0700342 struct sk_buff *other;
343 atomic_t *fclone_ref;
344
David S. Millerd179cd12005-08-17 14:57:30 -0700345 switch (skb->fclone) {
346 case SKB_FCLONE_UNAVAILABLE:
347 kmem_cache_free(skbuff_head_cache, skb);
348 break;
349
350 case SKB_FCLONE_ORIG:
351 fclone_ref = (atomic_t *) (skb + 2);
352 if (atomic_dec_and_test(fclone_ref))
353 kmem_cache_free(skbuff_fclone_cache, skb);
354 break;
355
356 case SKB_FCLONE_CLONE:
357 fclone_ref = (atomic_t *) (skb + 1);
358 other = skb - 1;
359
360 /* The clone portion is available for
361 * fast-cloning again.
362 */
363 skb->fclone = SKB_FCLONE_UNAVAILABLE;
364
365 if (atomic_dec_and_test(fclone_ref))
366 kmem_cache_free(skbuff_fclone_cache, other);
367 break;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -0700368 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369}
370
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700371static void skb_release_head_state(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372{
Eric Dumazetadf30902009-06-02 05:19:30 +0000373 skb_dst_drop(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374#ifdef CONFIG_XFRM
375 secpath_put(skb->sp);
376#endif
Stephen Hemminger9c2b3322005-04-19 22:39:42 -0700377 if (skb->destructor) {
378 WARN_ON(in_irq());
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 skb->destructor(skb);
380 }
Yasuyuki Kozakai9fb9cbb2005-11-09 16:38:16 -0800381#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -0700382 nf_conntrack_put(skb->nfct);
KOVACS Krisztianae90bde2010-12-15 23:53:41 +0100383#endif
384#ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
Yasuyuki Kozakai9fb9cbb2005-11-09 16:38:16 -0800385 nf_conntrack_put_reasm(skb->nfct_reasm);
386#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387#ifdef CONFIG_BRIDGE_NETFILTER
388 nf_bridge_put(skb->nf_bridge);
389#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390/* XXX: IS this still necessary? - JHS */
391#ifdef CONFIG_NET_SCHED
392 skb->tc_index = 0;
393#ifdef CONFIG_NET_CLS_ACT
394 skb->tc_verd = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395#endif
396#endif
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700397}
398
399/* Free everything but the sk_buff shell. */
400static void skb_release_all(struct sk_buff *skb)
401{
402 skb_release_head_state(skb);
Herbert Xu2d4baff2007-11-26 23:11:19 +0800403 skb_release_data(skb);
404}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405
Herbert Xu2d4baff2007-11-26 23:11:19 +0800406/**
407 * __kfree_skb - private function
408 * @skb: buffer
409 *
410 * Free an sk_buff. Release anything attached to the buffer.
411 * Clean the state. This is an internal helper function. Users should
412 * always call kfree_skb
413 */
414
415void __kfree_skb(struct sk_buff *skb)
416{
417 skb_release_all(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 kfree_skbmem(skb);
419}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800420EXPORT_SYMBOL(__kfree_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421
422/**
Jörn Engel231d06a2006-03-20 21:28:35 -0800423 * kfree_skb - free an sk_buff
424 * @skb: buffer to free
425 *
426 * Drop a reference to the buffer and free it if the usage count has
427 * hit zero.
428 */
429void kfree_skb(struct sk_buff *skb)
430{
431 if (unlikely(!skb))
432 return;
433 if (likely(atomic_read(&skb->users) == 1))
434 smp_rmb();
435 else if (likely(!atomic_dec_and_test(&skb->users)))
436 return;
Neil Hormanead2ceb2009-03-11 09:49:55 +0000437 trace_kfree_skb(skb, __builtin_return_address(0));
Jörn Engel231d06a2006-03-20 21:28:35 -0800438 __kfree_skb(skb);
439}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800440EXPORT_SYMBOL(kfree_skb);
Jörn Engel231d06a2006-03-20 21:28:35 -0800441
Stephen Hemmingerd1a203e2008-11-01 21:01:09 -0700442/**
Neil Hormanead2ceb2009-03-11 09:49:55 +0000443 * consume_skb - free an skbuff
444 * @skb: buffer to free
445 *
446 * Drop a ref to the buffer and free it if the usage count has hit zero
447 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
448 * is being dropped after a failure and notes that
449 */
450void consume_skb(struct sk_buff *skb)
451{
452 if (unlikely(!skb))
453 return;
454 if (likely(atomic_read(&skb->users) == 1))
455 smp_rmb();
456 else if (likely(!atomic_dec_and_test(&skb->users)))
457 return;
Koki Sanagi07dc22e2010-08-23 18:46:12 +0900458 trace_consume_skb(skb);
Neil Hormanead2ceb2009-03-11 09:49:55 +0000459 __kfree_skb(skb);
460}
461EXPORT_SYMBOL(consume_skb);
462
463/**
Stephen Hemmingerd1a203e2008-11-01 21:01:09 -0700464 * skb_recycle_check - check if skb can be reused for receive
465 * @skb: buffer
466 * @skb_size: minimum receive buffer size
467 *
468 * Checks that the skb passed in is not shared or cloned, and
469 * that it is linear and its head portion at least as large as
470 * skb_size so that it can be recycled as a receive buffer.
471 * If these conditions are met, this function does any necessary
472 * reference count dropping and cleans up the skbuff as if it
473 * just came from __alloc_skb().
474 */
Changli Gao5b0daa32010-05-29 00:12:13 -0700475bool skb_recycle_check(struct sk_buff *skb, int skb_size)
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700476{
477 struct skb_shared_info *shinfo;
478
Anton Vorontsove84af6d2009-11-10 14:11:01 +0000479 if (irqs_disabled())
Changli Gao5b0daa32010-05-29 00:12:13 -0700480 return false;
Anton Vorontsove84af6d2009-11-10 14:11:01 +0000481
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700482 if (skb_is_nonlinear(skb) || skb->fclone != SKB_FCLONE_UNAVAILABLE)
Changli Gao5b0daa32010-05-29 00:12:13 -0700483 return false;
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700484
485 skb_size = SKB_DATA_ALIGN(skb_size + NET_SKB_PAD);
486 if (skb_end_pointer(skb) - skb->head < skb_size)
Changli Gao5b0daa32010-05-29 00:12:13 -0700487 return false;
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700488
489 if (skb_shared(skb) || skb_cloned(skb))
Changli Gao5b0daa32010-05-29 00:12:13 -0700490 return false;
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700491
492 skb_release_head_state(skb);
Eric Dumazetec7d2f22010-05-05 01:07:37 -0700493
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700494 shinfo = skb_shinfo(skb);
Eric Dumazetec7d2f22010-05-05 01:07:37 -0700495 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700496 atomic_set(&shinfo->dataref, 1);
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700497
498 memset(skb, 0, offsetof(struct sk_buff, tail));
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700499 skb->data = skb->head + NET_SKB_PAD;
Lennert Buytenhek5cd33db2008-11-10 21:45:05 -0800500 skb_reset_tail_pointer(skb);
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700501
Changli Gao5b0daa32010-05-29 00:12:13 -0700502 return true;
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700503}
504EXPORT_SYMBOL(skb_recycle_check);
505
Herbert Xudec18812007-10-14 00:37:30 -0700506static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
507{
508 new->tstamp = old->tstamp;
509 new->dev = old->dev;
510 new->transport_header = old->transport_header;
511 new->network_header = old->network_header;
512 new->mac_header = old->mac_header;
Eric Dumazet7fee2262010-05-11 23:19:48 +0000513 skb_dst_copy(new, old);
Tom Herbert0a9627f2010-03-16 08:03:29 +0000514 new->rxhash = old->rxhash;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -0700515#ifdef CONFIG_XFRM
Herbert Xudec18812007-10-14 00:37:30 -0700516 new->sp = secpath_get(old->sp);
517#endif
518 memcpy(new->cb, old->cb, sizeof(old->cb));
Herbert Xu9bcb97c2009-05-22 22:20:02 +0000519 new->csum = old->csum;
Herbert Xudec18812007-10-14 00:37:30 -0700520 new->local_df = old->local_df;
521 new->pkt_type = old->pkt_type;
522 new->ip_summed = old->ip_summed;
523 skb_copy_queue_mapping(new, old);
524 new->priority = old->priority;
John Fastabende8970822010-06-13 10:36:30 +0000525 new->deliver_no_wcard = old->deliver_no_wcard;
Herbert Xudec18812007-10-14 00:37:30 -0700526#if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
527 new->ipvs_property = old->ipvs_property;
528#endif
529 new->protocol = old->protocol;
530 new->mark = old->mark;
Eric Dumazet8964be42009-11-20 15:35:04 -0800531 new->skb_iif = old->skb_iif;
Herbert Xudec18812007-10-14 00:37:30 -0700532 __nf_copy(new, old);
533#if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
534 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
535 new->nf_trace = old->nf_trace;
536#endif
537#ifdef CONFIG_NET_SCHED
538 new->tc_index = old->tc_index;
539#ifdef CONFIG_NET_CLS_ACT
540 new->tc_verd = old->tc_verd;
541#endif
542#endif
Patrick McHardy6aa895b2008-07-14 22:49:06 -0700543 new->vlan_tci = old->vlan_tci;
544
Herbert Xudec18812007-10-14 00:37:30 -0700545 skb_copy_secmark(new, old);
546}
547
Herbert Xu82c49a32009-05-22 22:11:37 +0000548/*
549 * You should not add any new code to this function. Add it to
550 * __copy_skb_header above instead.
551 */
Herbert Xue0053ec2007-10-14 00:37:52 -0700552static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554#define C(x) n->x = skb->x
555
556 n->next = n->prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 n->sk = NULL;
Herbert Xudec18812007-10-14 00:37:30 -0700558 __copy_skb_header(n, skb);
559
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 C(len);
561 C(data_len);
Alexey Dobriyan3e6b3b22007-03-16 15:00:46 -0700562 C(mac_len);
Patrick McHardy334a8132007-06-25 04:35:20 -0700563 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
Paul Moore02f1c892008-01-07 21:56:41 -0800564 n->cloned = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 n->nohdr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 n->destructor = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 C(tail);
568 C(end);
Paul Moore02f1c892008-01-07 21:56:41 -0800569 C(head);
570 C(data);
571 C(truesize);
572 atomic_set(&n->users, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573
574 atomic_inc(&(skb_shinfo(skb)->dataref));
575 skb->cloned = 1;
576
577 return n;
Herbert Xue0053ec2007-10-14 00:37:52 -0700578#undef C
579}
580
581/**
582 * skb_morph - morph one skb into another
583 * @dst: the skb to receive the contents
584 * @src: the skb to supply the contents
585 *
586 * This is identical to skb_clone except that the target skb is
587 * supplied by the user.
588 *
589 * The target skb is returned upon exit.
590 */
591struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
592{
Herbert Xu2d4baff2007-11-26 23:11:19 +0800593 skb_release_all(dst);
Herbert Xue0053ec2007-10-14 00:37:52 -0700594 return __skb_clone(dst, src);
595}
596EXPORT_SYMBOL_GPL(skb_morph);
597
598/**
599 * skb_clone - duplicate an sk_buff
600 * @skb: buffer to clone
601 * @gfp_mask: allocation priority
602 *
603 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
604 * copies share the same packet data but not structure. The new
605 * buffer has a reference count of 1. If the allocation fails the
606 * function returns %NULL otherwise the new buffer is returned.
607 *
608 * If this function is called from an interrupt gfp_mask() must be
609 * %GFP_ATOMIC.
610 */
611
612struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
613{
614 struct sk_buff *n;
615
616 n = skb + 1;
617 if (skb->fclone == SKB_FCLONE_ORIG &&
618 n->fclone == SKB_FCLONE_UNAVAILABLE) {
619 atomic_t *fclone_ref = (atomic_t *) (n + 1);
620 n->fclone = SKB_FCLONE_CLONE;
621 atomic_inc(fclone_ref);
622 } else {
623 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
624 if (!n)
625 return NULL;
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200626
627 kmemcheck_annotate_bitfield(n, flags1);
628 kmemcheck_annotate_bitfield(n, flags2);
Herbert Xue0053ec2007-10-14 00:37:52 -0700629 n->fclone = SKB_FCLONE_UNAVAILABLE;
630 }
631
632 return __skb_clone(n, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800634EXPORT_SYMBOL(skb_clone);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635
636static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
637{
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700638#ifndef NET_SKBUFF_DATA_USES_OFFSET
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 /*
640 * Shift between the two data areas in bytes
641 */
642 unsigned long offset = new->data - old->data;
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700643#endif
Herbert Xudec18812007-10-14 00:37:30 -0700644
645 __copy_skb_header(new, old);
646
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700647#ifndef NET_SKBUFF_DATA_USES_OFFSET
648 /* {transport,network,mac}_header are relative to skb->head */
649 new->transport_header += offset;
650 new->network_header += offset;
Stephen Hemminger603a8bb2009-06-17 12:17:34 +0000651 if (skb_mac_header_was_set(new))
652 new->mac_header += offset;
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700653#endif
Herbert Xu79671682006-06-22 02:40:14 -0700654 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
655 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
656 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657}
658
659/**
660 * skb_copy - create private copy of an sk_buff
661 * @skb: buffer to copy
662 * @gfp_mask: allocation priority
663 *
664 * Make a copy of both an &sk_buff and its data. This is used when the
665 * caller wishes to modify the data and needs a private copy of the
666 * data to alter. Returns %NULL on failure or the pointer to the buffer
667 * on success. The returned buffer has a reference count of 1.
668 *
669 * As by-product this function converts non-linear &sk_buff to linear
670 * one, so that &sk_buff becomes completely private and caller is allowed
671 * to modify all the data of returned buffer. This means that this
672 * function is not recommended for use in circumstances when only
673 * header is going to be modified. Use pskb_copy() instead.
674 */
675
Al Virodd0fc662005-10-07 07:46:04 +0100676struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677{
Eric Dumazet6602ceb2010-09-01 05:25:10 +0000678 int headerlen = skb_headroom(skb);
679 unsigned int size = (skb_end_pointer(skb) - skb->head) + skb->data_len;
680 struct sk_buff *n = alloc_skb(size, gfp_mask);
681
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 if (!n)
683 return NULL;
684
685 /* Set the data pointer */
686 skb_reserve(n, headerlen);
687 /* Set the tail pointer and length */
688 skb_put(n, skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689
690 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
691 BUG();
692
693 copy_skb_header(n, skb);
694 return n;
695}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800696EXPORT_SYMBOL(skb_copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697
698/**
699 * pskb_copy - create copy of an sk_buff with private head.
700 * @skb: buffer to copy
701 * @gfp_mask: allocation priority
702 *
703 * Make a copy of both an &sk_buff and part of its data, located
704 * in header. Fragmented data remain shared. This is used when
705 * the caller wishes to modify only header of &sk_buff and needs
706 * private copy of the header to alter. Returns %NULL on failure
707 * or the pointer to the buffer on success.
708 * The returned buffer has a reference count of 1.
709 */
710
Al Virodd0fc662005-10-07 07:46:04 +0100711struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712{
Eric Dumazet6602ceb2010-09-01 05:25:10 +0000713 unsigned int size = skb_end_pointer(skb) - skb->head;
714 struct sk_buff *n = alloc_skb(size, gfp_mask);
715
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 if (!n)
717 goto out;
718
719 /* Set the data pointer */
Eric Dumazet6602ceb2010-09-01 05:25:10 +0000720 skb_reserve(n, skb_headroom(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 /* Set the tail pointer and length */
722 skb_put(n, skb_headlen(skb));
723 /* Copy the bytes */
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -0300724 skb_copy_from_linear_data(skb, n->data, n->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725
Herbert Xu25f484a2006-11-07 14:57:15 -0800726 n->truesize += skb->data_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 n->data_len = skb->data_len;
728 n->len = skb->len;
729
730 if (skb_shinfo(skb)->nr_frags) {
731 int i;
732
733 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
734 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
735 get_page(skb_shinfo(n)->frags[i].page);
736 }
737 skb_shinfo(n)->nr_frags = i;
738 }
739
David S. Miller21dc3302010-08-23 00:13:46 -0700740 if (skb_has_frag_list(skb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
742 skb_clone_fraglist(n);
743 }
744
745 copy_skb_header(n, skb);
746out:
747 return n;
748}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800749EXPORT_SYMBOL(pskb_copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750
751/**
752 * pskb_expand_head - reallocate header of &sk_buff
753 * @skb: buffer to reallocate
754 * @nhead: room to add at head
755 * @ntail: room to add at tail
756 * @gfp_mask: allocation priority
757 *
758 * Expands (or creates identical copy, if &nhead and &ntail are zero)
759 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
760 * reference count of 1. Returns zero in the case of success or error,
761 * if expansion failed. In the last case, &sk_buff is not changed.
762 *
763 * All the pointers pointing into skb header may change and must be
764 * reloaded after call to this function.
765 */
766
Victor Fusco86a76ca2005-07-08 14:57:47 -0700767int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
Al Virodd0fc662005-10-07 07:46:04 +0100768 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769{
770 int i;
771 u8 *data;
Eric Dumazet6602ceb2010-09-01 05:25:10 +0000772 int size = nhead + (skb_end_pointer(skb) - skb->head) + ntail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 long off;
Eric Dumazet1fd63042010-09-02 23:09:32 +0000774 bool fastpath;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775
Herbert Xu4edd87a2008-10-01 07:09:38 -0700776 BUG_ON(nhead < 0);
777
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 if (skb_shared(skb))
779 BUG();
780
781 size = SKB_DATA_ALIGN(size);
782
783 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
784 if (!data)
785 goto nodata;
786
787 /* Copy only real data... and, alas, header. This should be
Eric Dumazet6602ceb2010-09-01 05:25:10 +0000788 * optimized for the cases when header is void.
789 */
790 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
791
792 memcpy((struct skb_shared_info *)(data + size),
793 skb_shinfo(skb),
Eric Dumazetfed66382010-07-22 19:09:08 +0000794 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795
Eric Dumazet1fd63042010-09-02 23:09:32 +0000796 /* Check if we can avoid taking references on fragments if we own
797 * the last reference on skb->head. (see skb_release_data())
798 */
799 if (!skb->cloned)
800 fastpath = true;
801 else {
802 int delta = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803
Eric Dumazet1fd63042010-09-02 23:09:32 +0000804 fastpath = atomic_read(&skb_shinfo(skb)->dataref) == delta;
805 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806
Eric Dumazet1fd63042010-09-02 23:09:32 +0000807 if (fastpath) {
808 kfree(skb->head);
809 } else {
810 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
811 get_page(skb_shinfo(skb)->frags[i].page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812
Eric Dumazet1fd63042010-09-02 23:09:32 +0000813 if (skb_has_frag_list(skb))
814 skb_clone_fraglist(skb);
815
816 skb_release_data(skb);
817 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 off = (data + nhead) - skb->head;
819
820 skb->head = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 skb->data += off;
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700822#ifdef NET_SKBUFF_DATA_USES_OFFSET
823 skb->end = size;
Patrick McHardy56eb8882007-04-09 11:45:04 -0700824 off = nhead;
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700825#else
826 skb->end = skb->head + size;
Patrick McHardy56eb8882007-04-09 11:45:04 -0700827#endif
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700828 /* {transport,network,mac}_header and tail are relative to skb->head */
829 skb->tail += off;
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700830 skb->transport_header += off;
831 skb->network_header += off;
Stephen Hemminger603a8bb2009-06-17 12:17:34 +0000832 if (skb_mac_header_was_set(skb))
833 skb->mac_header += off;
Andrea Shepard00c5a982010-07-22 09:12:35 +0000834 /* Only adjust this if it actually is csum_start rather than csum */
835 if (skb->ip_summed == CHECKSUM_PARTIAL)
836 skb->csum_start += nhead;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 skb->cloned = 0;
Patrick McHardy334a8132007-06-25 04:35:20 -0700838 skb->hdr_len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 skb->nohdr = 0;
840 atomic_set(&skb_shinfo(skb)->dataref, 1);
841 return 0;
842
843nodata:
844 return -ENOMEM;
845}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800846EXPORT_SYMBOL(pskb_expand_head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847
848/* Make private copy of skb with writable head and some headroom */
849
850struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
851{
852 struct sk_buff *skb2;
853 int delta = headroom - skb_headroom(skb);
854
855 if (delta <= 0)
856 skb2 = pskb_copy(skb, GFP_ATOMIC);
857 else {
858 skb2 = skb_clone(skb, GFP_ATOMIC);
859 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
860 GFP_ATOMIC)) {
861 kfree_skb(skb2);
862 skb2 = NULL;
863 }
864 }
865 return skb2;
866}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800867EXPORT_SYMBOL(skb_realloc_headroom);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868
869/**
870 * skb_copy_expand - copy and expand sk_buff
871 * @skb: buffer to copy
872 * @newheadroom: new free bytes at head
873 * @newtailroom: new free bytes at tail
874 * @gfp_mask: allocation priority
875 *
876 * Make a copy of both an &sk_buff and its data and while doing so
877 * allocate additional space.
878 *
879 * This is used when the caller wishes to modify the data and needs a
880 * private copy of the data to alter as well as more space for new fields.
881 * Returns %NULL on failure or the pointer to the buffer
882 * on success. The returned buffer has a reference count of 1.
883 *
884 * You must pass %GFP_ATOMIC as the allocation priority if this function
885 * is called from an interrupt.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 */
887struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
Victor Fusco86a76ca2005-07-08 14:57:47 -0700888 int newheadroom, int newtailroom,
Al Virodd0fc662005-10-07 07:46:04 +0100889 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890{
891 /*
892 * Allocate the copy buffer
893 */
894 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
895 gfp_mask);
Patrick McHardyefd1e8d2007-04-10 18:30:09 -0700896 int oldheadroom = skb_headroom(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 int head_copy_len, head_copy_off;
Herbert Xu52886052007-09-16 16:32:11 -0700898 int off;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899
900 if (!n)
901 return NULL;
902
903 skb_reserve(n, newheadroom);
904
905 /* Set the tail pointer and length */
906 skb_put(n, skb->len);
907
Patrick McHardyefd1e8d2007-04-10 18:30:09 -0700908 head_copy_len = oldheadroom;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 head_copy_off = 0;
910 if (newheadroom <= head_copy_len)
911 head_copy_len = newheadroom;
912 else
913 head_copy_off = newheadroom - head_copy_len;
914
915 /* Copy the linear header and data. */
916 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
917 skb->len + head_copy_len))
918 BUG();
919
920 copy_skb_header(n, skb);
921
Patrick McHardyefd1e8d2007-04-10 18:30:09 -0700922 off = newheadroom - oldheadroom;
David S. Millerbe2b6e62010-07-22 13:27:09 -0700923 if (n->ip_summed == CHECKSUM_PARTIAL)
924 n->csum_start += off;
Herbert Xu52886052007-09-16 16:32:11 -0700925#ifdef NET_SKBUFF_DATA_USES_OFFSET
Patrick McHardyefd1e8d2007-04-10 18:30:09 -0700926 n->transport_header += off;
927 n->network_header += off;
Stephen Hemminger603a8bb2009-06-17 12:17:34 +0000928 if (skb_mac_header_was_set(skb))
929 n->mac_header += off;
Herbert Xu52886052007-09-16 16:32:11 -0700930#endif
Patrick McHardyefd1e8d2007-04-10 18:30:09 -0700931
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 return n;
933}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800934EXPORT_SYMBOL(skb_copy_expand);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935
936/**
937 * skb_pad - zero pad the tail of an skb
938 * @skb: buffer to pad
939 * @pad: space to pad
940 *
941 * Ensure that a buffer is followed by a padding area that is zero
942 * filled. Used by network drivers which may DMA or transfer data
943 * beyond the buffer end onto the wire.
944 *
Herbert Xu5b057c62006-06-23 02:06:41 -0700945 * May return error in out of memory cases. The skb is freed on error.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 */
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900947
Herbert Xu5b057c62006-06-23 02:06:41 -0700948int skb_pad(struct sk_buff *skb, int pad)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949{
Herbert Xu5b057c62006-06-23 02:06:41 -0700950 int err;
951 int ntail;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900952
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 /* If the skbuff is non linear tailroom is always zero.. */
Herbert Xu5b057c62006-06-23 02:06:41 -0700954 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 memset(skb->data+skb->len, 0, pad);
Herbert Xu5b057c62006-06-23 02:06:41 -0700956 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 }
Herbert Xu5b057c62006-06-23 02:06:41 -0700958
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700959 ntail = skb->data_len + pad - (skb->end - skb->tail);
Herbert Xu5b057c62006-06-23 02:06:41 -0700960 if (likely(skb_cloned(skb) || ntail > 0)) {
961 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
962 if (unlikely(err))
963 goto free_skb;
964 }
965
966 /* FIXME: The use of this function with non-linear skb's really needs
967 * to be audited.
968 */
969 err = skb_linearize(skb);
970 if (unlikely(err))
971 goto free_skb;
972
973 memset(skb->data + skb->len, 0, pad);
974 return 0;
975
976free_skb:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 kfree_skb(skb);
Herbert Xu5b057c62006-06-23 02:06:41 -0700978 return err;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900979}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800980EXPORT_SYMBOL(skb_pad);
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900981
Ilpo Järvinen0dde3e12008-03-27 17:43:41 -0700982/**
983 * skb_put - add data to a buffer
984 * @skb: buffer to use
985 * @len: amount of data to add
986 *
987 * This function extends the used data area of the buffer. If this would
988 * exceed the total buffer size the kernel will panic. A pointer to the
989 * first byte of the extra data is returned.
990 */
991unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
992{
993 unsigned char *tmp = skb_tail_pointer(skb);
994 SKB_LINEAR_ASSERT(skb);
995 skb->tail += len;
996 skb->len += len;
997 if (unlikely(skb->tail > skb->end))
998 skb_over_panic(skb, len, __builtin_return_address(0));
999 return tmp;
1000}
1001EXPORT_SYMBOL(skb_put);
1002
Ilpo Järvinen6be8ac22008-03-27 17:47:24 -07001003/**
Ilpo Järvinenc2aa2702008-03-27 17:52:40 -07001004 * skb_push - add data to the start of a buffer
1005 * @skb: buffer to use
1006 * @len: amount of data to add
1007 *
1008 * This function extends the used data area of the buffer at the buffer
1009 * start. If this would exceed the total buffer headroom the kernel will
1010 * panic. A pointer to the first byte of the extra data is returned.
1011 */
1012unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1013{
1014 skb->data -= len;
1015 skb->len += len;
1016 if (unlikely(skb->data<skb->head))
1017 skb_under_panic(skb, len, __builtin_return_address(0));
1018 return skb->data;
1019}
1020EXPORT_SYMBOL(skb_push);
1021
1022/**
Ilpo Järvinen6be8ac22008-03-27 17:47:24 -07001023 * skb_pull - remove data from the start of a buffer
1024 * @skb: buffer to use
1025 * @len: amount of data to remove
1026 *
1027 * This function removes data from the start of a buffer, returning
1028 * the memory to the headroom. A pointer to the next data in the buffer
1029 * is returned. Once the data has been pulled future pushes will overwrite
1030 * the old data.
1031 */
1032unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1033{
David S. Miller47d29642010-05-02 02:21:44 -07001034 return skb_pull_inline(skb, len);
Ilpo Järvinen6be8ac22008-03-27 17:47:24 -07001035}
1036EXPORT_SYMBOL(skb_pull);
1037
Ilpo Järvinen419ae742008-03-27 17:54:01 -07001038/**
1039 * skb_trim - remove end from a buffer
1040 * @skb: buffer to alter
1041 * @len: new length
1042 *
1043 * Cut the length of a buffer down by removing data from the tail. If
1044 * the buffer is already under the length specified it is not modified.
1045 * The skb must be linear.
1046 */
1047void skb_trim(struct sk_buff *skb, unsigned int len)
1048{
1049 if (skb->len > len)
1050 __skb_trim(skb, len);
1051}
1052EXPORT_SYMBOL(skb_trim);
1053
Herbert Xu3cc0e872006-06-09 16:13:38 -07001054/* Trims skb to length len. It can change skb pointers.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 */
1056
Herbert Xu3cc0e872006-06-09 16:13:38 -07001057int ___pskb_trim(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058{
Herbert Xu27b437c2006-07-13 19:26:39 -07001059 struct sk_buff **fragp;
1060 struct sk_buff *frag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 int offset = skb_headlen(skb);
1062 int nfrags = skb_shinfo(skb)->nr_frags;
1063 int i;
Herbert Xu27b437c2006-07-13 19:26:39 -07001064 int err;
1065
1066 if (skb_cloned(skb) &&
1067 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1068 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001070 i = 0;
1071 if (offset >= len)
1072 goto drop_pages;
1073
1074 for (; i < nfrags; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 int end = offset + skb_shinfo(skb)->frags[i].size;
Herbert Xu27b437c2006-07-13 19:26:39 -07001076
1077 if (end < len) {
1078 offset = end;
1079 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 }
Herbert Xu27b437c2006-07-13 19:26:39 -07001081
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001082 skb_shinfo(skb)->frags[i++].size = len - offset;
Herbert Xu27b437c2006-07-13 19:26:39 -07001083
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001084drop_pages:
Herbert Xu27b437c2006-07-13 19:26:39 -07001085 skb_shinfo(skb)->nr_frags = i;
1086
1087 for (; i < nfrags; i++)
1088 put_page(skb_shinfo(skb)->frags[i].page);
1089
David S. Miller21dc3302010-08-23 00:13:46 -07001090 if (skb_has_frag_list(skb))
Herbert Xu27b437c2006-07-13 19:26:39 -07001091 skb_drop_fraglist(skb);
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001092 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 }
1094
Herbert Xu27b437c2006-07-13 19:26:39 -07001095 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1096 fragp = &frag->next) {
1097 int end = offset + frag->len;
1098
1099 if (skb_shared(frag)) {
1100 struct sk_buff *nfrag;
1101
1102 nfrag = skb_clone(frag, GFP_ATOMIC);
1103 if (unlikely(!nfrag))
1104 return -ENOMEM;
1105
1106 nfrag->next = frag->next;
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001107 kfree_skb(frag);
Herbert Xu27b437c2006-07-13 19:26:39 -07001108 frag = nfrag;
1109 *fragp = frag;
1110 }
1111
1112 if (end < len) {
1113 offset = end;
1114 continue;
1115 }
1116
1117 if (end > len &&
1118 unlikely((err = pskb_trim(frag, len - offset))))
1119 return err;
1120
1121 if (frag->next)
1122 skb_drop_list(&frag->next);
1123 break;
1124 }
1125
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001126done:
Herbert Xu27b437c2006-07-13 19:26:39 -07001127 if (len > skb_headlen(skb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 skb->data_len -= skb->len - len;
1129 skb->len = len;
1130 } else {
Herbert Xu27b437c2006-07-13 19:26:39 -07001131 skb->len = len;
1132 skb->data_len = 0;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001133 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 }
1135
1136 return 0;
1137}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001138EXPORT_SYMBOL(___pskb_trim);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139
1140/**
1141 * __pskb_pull_tail - advance tail of skb header
1142 * @skb: buffer to reallocate
1143 * @delta: number of bytes to advance tail
1144 *
1145 * The function makes a sense only on a fragmented &sk_buff,
1146 * it expands header moving its tail forward and copying necessary
1147 * data from fragmented part.
1148 *
1149 * &sk_buff MUST have reference count of 1.
1150 *
1151 * Returns %NULL (and &sk_buff does not change) if pull failed
1152 * or value of new tail of skb in the case of success.
1153 *
1154 * All the pointers pointing into skb header may change and must be
1155 * reloaded after call to this function.
1156 */
1157
1158/* Moves tail of skb head forward, copying data from fragmented part,
1159 * when it is necessary.
1160 * 1. It may fail due to malloc failure.
1161 * 2. It may change skb pointers.
1162 *
1163 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1164 */
1165unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1166{
1167 /* If skb has not enough free space at tail, get new one
1168 * plus 128 bytes for future expansions. If we have enough
1169 * room at tail, reallocate without expansion only if skb is cloned.
1170 */
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001171 int i, k, eat = (skb->tail + delta) - skb->end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172
1173 if (eat > 0 || skb_cloned(skb)) {
1174 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1175 GFP_ATOMIC))
1176 return NULL;
1177 }
1178
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001179 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 BUG();
1181
1182 /* Optimization: no fragments, no reasons to preestimate
1183 * size of pulled pages. Superb.
1184 */
David S. Miller21dc3302010-08-23 00:13:46 -07001185 if (!skb_has_frag_list(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 goto pull_pages;
1187
1188 /* Estimate size of pulled pages. */
1189 eat = delta;
1190 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1191 if (skb_shinfo(skb)->frags[i].size >= eat)
1192 goto pull_pages;
1193 eat -= skb_shinfo(skb)->frags[i].size;
1194 }
1195
1196 /* If we need update frag list, we are in troubles.
1197 * Certainly, it possible to add an offset to skb data,
1198 * but taking into account that pulling is expected to
1199 * be very rare operation, it is worth to fight against
1200 * further bloating skb head and crucify ourselves here instead.
1201 * Pure masohism, indeed. 8)8)
1202 */
1203 if (eat) {
1204 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1205 struct sk_buff *clone = NULL;
1206 struct sk_buff *insp = NULL;
1207
1208 do {
Kris Katterjohn09a62662006-01-08 22:24:28 -08001209 BUG_ON(!list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210
1211 if (list->len <= eat) {
1212 /* Eaten as whole. */
1213 eat -= list->len;
1214 list = list->next;
1215 insp = list;
1216 } else {
1217 /* Eaten partially. */
1218
1219 if (skb_shared(list)) {
1220 /* Sucks! We need to fork list. :-( */
1221 clone = skb_clone(list, GFP_ATOMIC);
1222 if (!clone)
1223 return NULL;
1224 insp = list->next;
1225 list = clone;
1226 } else {
1227 /* This may be pulled without
1228 * problems. */
1229 insp = list;
1230 }
1231 if (!pskb_pull(list, eat)) {
Wei Yongjunf3fbbe02009-02-25 00:37:32 +00001232 kfree_skb(clone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 return NULL;
1234 }
1235 break;
1236 }
1237 } while (eat);
1238
1239 /* Free pulled out fragments. */
1240 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1241 skb_shinfo(skb)->frag_list = list->next;
1242 kfree_skb(list);
1243 }
1244 /* And insert new clone at head. */
1245 if (clone) {
1246 clone->next = list;
1247 skb_shinfo(skb)->frag_list = clone;
1248 }
1249 }
1250 /* Success! Now we may commit changes to skb data. */
1251
1252pull_pages:
1253 eat = delta;
1254 k = 0;
1255 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1256 if (skb_shinfo(skb)->frags[i].size <= eat) {
1257 put_page(skb_shinfo(skb)->frags[i].page);
1258 eat -= skb_shinfo(skb)->frags[i].size;
1259 } else {
1260 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1261 if (eat) {
1262 skb_shinfo(skb)->frags[k].page_offset += eat;
1263 skb_shinfo(skb)->frags[k].size -= eat;
1264 eat = 0;
1265 }
1266 k++;
1267 }
1268 }
1269 skb_shinfo(skb)->nr_frags = k;
1270
1271 skb->tail += delta;
1272 skb->data_len -= delta;
1273
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001274 return skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001276EXPORT_SYMBOL(__pskb_pull_tail);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277
1278/* Copy some data bits from skb to kernel buffer. */
1279
1280int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1281{
David S. Miller1a028e52007-04-27 15:21:23 -07001282 int start = skb_headlen(skb);
David S. Millerfbb398a2009-06-09 00:18:59 -07001283 struct sk_buff *frag_iter;
1284 int i, copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285
1286 if (offset > (int)skb->len - len)
1287 goto fault;
1288
1289 /* Copy header. */
David S. Miller1a028e52007-04-27 15:21:23 -07001290 if ((copy = start - offset) > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 if (copy > len)
1292 copy = len;
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03001293 skb_copy_from_linear_data_offset(skb, offset, to, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 if ((len -= copy) == 0)
1295 return 0;
1296 offset += copy;
1297 to += copy;
1298 }
1299
1300 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07001301 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001303 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07001304
1305 end = start + skb_shinfo(skb)->frags[i].size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 if ((copy = end - offset) > 0) {
1307 u8 *vaddr;
1308
1309 if (copy > len)
1310 copy = len;
1311
1312 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1313 memcpy(to,
David S. Miller1a028e52007-04-27 15:21:23 -07001314 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1315 offset - start, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 kunmap_skb_frag(vaddr);
1317
1318 if ((len -= copy) == 0)
1319 return 0;
1320 offset += copy;
1321 to += copy;
1322 }
David S. Miller1a028e52007-04-27 15:21:23 -07001323 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 }
1325
David S. Millerfbb398a2009-06-09 00:18:59 -07001326 skb_walk_frags(skb, frag_iter) {
1327 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328
David S. Millerfbb398a2009-06-09 00:18:59 -07001329 WARN_ON(start > offset + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330
David S. Millerfbb398a2009-06-09 00:18:59 -07001331 end = start + frag_iter->len;
1332 if ((copy = end - offset) > 0) {
1333 if (copy > len)
1334 copy = len;
1335 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1336 goto fault;
1337 if ((len -= copy) == 0)
1338 return 0;
1339 offset += copy;
1340 to += copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 }
David S. Millerfbb398a2009-06-09 00:18:59 -07001342 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 }
1344 if (!len)
1345 return 0;
1346
1347fault:
1348 return -EFAULT;
1349}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001350EXPORT_SYMBOL(skb_copy_bits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351
Jens Axboe9c55e012007-11-06 23:30:13 -08001352/*
1353 * Callback from splice_to_pipe(), if we need to release some pages
1354 * at the end of the spd in case we error'ed out in filling the pipe.
1355 */
1356static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1357{
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001358 put_page(spd->pages[i]);
1359}
Jens Axboe9c55e012007-11-06 23:30:13 -08001360
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001361static inline struct page *linear_to_page(struct page *page, unsigned int *len,
1362 unsigned int *offset,
Jarek Poplawski7a67e562009-04-30 05:41:19 -07001363 struct sk_buff *skb, struct sock *sk)
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001364{
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001365 struct page *p = sk->sk_sndmsg_page;
1366 unsigned int off;
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001367
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001368 if (!p) {
1369new_page:
1370 p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1371 if (!p)
1372 return NULL;
1373
1374 off = sk->sk_sndmsg_off = 0;
1375 /* hold one ref to this page until it's full */
1376 } else {
1377 unsigned int mlen;
1378
1379 off = sk->sk_sndmsg_off;
1380 mlen = PAGE_SIZE - off;
1381 if (mlen < 64 && mlen < *len) {
1382 put_page(p);
1383 goto new_page;
1384 }
1385
1386 *len = min_t(unsigned int, *len, mlen);
1387 }
1388
1389 memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1390 sk->sk_sndmsg_off += *len;
1391 *offset = off;
1392 get_page(p);
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001393
1394 return p;
Jens Axboe9c55e012007-11-06 23:30:13 -08001395}
1396
1397/*
1398 * Fill page/offset/length into spd, if it can hold more pages.
1399 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001400static inline int spd_fill_page(struct splice_pipe_desc *spd,
1401 struct pipe_inode_info *pipe, struct page *page,
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001402 unsigned int *len, unsigned int offset,
Jarek Poplawski7a67e562009-04-30 05:41:19 -07001403 struct sk_buff *skb, int linear,
1404 struct sock *sk)
Jens Axboe9c55e012007-11-06 23:30:13 -08001405{
Jens Axboe35f3d142010-05-20 10:43:18 +02001406 if (unlikely(spd->nr_pages == pipe->buffers))
Jens Axboe9c55e012007-11-06 23:30:13 -08001407 return 1;
1408
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001409 if (linear) {
Jarek Poplawski7a67e562009-04-30 05:41:19 -07001410 page = linear_to_page(page, len, &offset, skb, sk);
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001411 if (!page)
1412 return 1;
1413 } else
1414 get_page(page);
1415
Jens Axboe9c55e012007-11-06 23:30:13 -08001416 spd->pages[spd->nr_pages] = page;
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001417 spd->partial[spd->nr_pages].len = *len;
Jens Axboe9c55e012007-11-06 23:30:13 -08001418 spd->partial[spd->nr_pages].offset = offset;
Jens Axboe9c55e012007-11-06 23:30:13 -08001419 spd->nr_pages++;
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001420
Jens Axboe9c55e012007-11-06 23:30:13 -08001421 return 0;
1422}
1423
Octavian Purdila2870c432008-07-15 00:49:11 -07001424static inline void __segment_seek(struct page **page, unsigned int *poff,
1425 unsigned int *plen, unsigned int off)
Jens Axboe9c55e012007-11-06 23:30:13 -08001426{
Jarek Poplawskice3dd392009-02-12 16:51:43 -08001427 unsigned long n;
1428
Octavian Purdila2870c432008-07-15 00:49:11 -07001429 *poff += off;
Jarek Poplawskice3dd392009-02-12 16:51:43 -08001430 n = *poff / PAGE_SIZE;
1431 if (n)
1432 *page = nth_page(*page, n);
1433
Octavian Purdila2870c432008-07-15 00:49:11 -07001434 *poff = *poff % PAGE_SIZE;
1435 *plen -= off;
1436}
Jens Axboe9c55e012007-11-06 23:30:13 -08001437
Octavian Purdila2870c432008-07-15 00:49:11 -07001438static inline int __splice_segment(struct page *page, unsigned int poff,
1439 unsigned int plen, unsigned int *off,
1440 unsigned int *len, struct sk_buff *skb,
Jarek Poplawski7a67e562009-04-30 05:41:19 -07001441 struct splice_pipe_desc *spd, int linear,
Jens Axboe35f3d142010-05-20 10:43:18 +02001442 struct sock *sk,
1443 struct pipe_inode_info *pipe)
Octavian Purdila2870c432008-07-15 00:49:11 -07001444{
1445 if (!*len)
1446 return 1;
1447
1448 /* skip this segment if already processed */
1449 if (*off >= plen) {
1450 *off -= plen;
1451 return 0;
Octavian Purdiladb43a282008-06-27 17:27:21 -07001452 }
Jens Axboe9c55e012007-11-06 23:30:13 -08001453
Octavian Purdila2870c432008-07-15 00:49:11 -07001454 /* ignore any bits we already processed */
1455 if (*off) {
1456 __segment_seek(&page, &poff, &plen, *off);
1457 *off = 0;
1458 }
1459
1460 do {
1461 unsigned int flen = min(*len, plen);
1462
1463 /* the linear region may spread across several pages */
1464 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1465
Jens Axboe35f3d142010-05-20 10:43:18 +02001466 if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk))
Octavian Purdila2870c432008-07-15 00:49:11 -07001467 return 1;
1468
1469 __segment_seek(&page, &poff, &plen, flen);
1470 *len -= flen;
1471
1472 } while (*len && plen);
1473
1474 return 0;
1475}
1476
1477/*
1478 * Map linear and fragment data from the skb to spd. It reports failure if the
1479 * pipe is full or if we already spliced the requested length.
1480 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001481static int __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1482 unsigned int *offset, unsigned int *len,
1483 struct splice_pipe_desc *spd, struct sock *sk)
Octavian Purdila2870c432008-07-15 00:49:11 -07001484{
1485 int seg;
1486
Jens Axboe9c55e012007-11-06 23:30:13 -08001487 /*
Octavian Purdila2870c432008-07-15 00:49:11 -07001488 * map the linear part
Jens Axboe9c55e012007-11-06 23:30:13 -08001489 */
Octavian Purdila2870c432008-07-15 00:49:11 -07001490 if (__splice_segment(virt_to_page(skb->data),
1491 (unsigned long) skb->data & (PAGE_SIZE - 1),
1492 skb_headlen(skb),
Jens Axboe35f3d142010-05-20 10:43:18 +02001493 offset, len, skb, spd, 1, sk, pipe))
Octavian Purdila2870c432008-07-15 00:49:11 -07001494 return 1;
Jens Axboe9c55e012007-11-06 23:30:13 -08001495
1496 /*
1497 * then map the fragments
1498 */
Jens Axboe9c55e012007-11-06 23:30:13 -08001499 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1500 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1501
Octavian Purdila2870c432008-07-15 00:49:11 -07001502 if (__splice_segment(f->page, f->page_offset, f->size,
Jens Axboe35f3d142010-05-20 10:43:18 +02001503 offset, len, skb, spd, 0, sk, pipe))
Octavian Purdila2870c432008-07-15 00:49:11 -07001504 return 1;
Jens Axboe9c55e012007-11-06 23:30:13 -08001505 }
1506
Octavian Purdila2870c432008-07-15 00:49:11 -07001507 return 0;
Jens Axboe9c55e012007-11-06 23:30:13 -08001508}
1509
1510/*
1511 * Map data from the skb to a pipe. Should handle both the linear part,
1512 * the fragments, and the frag list. It does NOT handle frag lists within
1513 * the frag list, if such a thing exists. We'd probably need to recurse to
1514 * handle that cleanly.
1515 */
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001516int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
Jens Axboe9c55e012007-11-06 23:30:13 -08001517 struct pipe_inode_info *pipe, unsigned int tlen,
1518 unsigned int flags)
1519{
Jens Axboe35f3d142010-05-20 10:43:18 +02001520 struct partial_page partial[PIPE_DEF_BUFFERS];
1521 struct page *pages[PIPE_DEF_BUFFERS];
Jens Axboe9c55e012007-11-06 23:30:13 -08001522 struct splice_pipe_desc spd = {
1523 .pages = pages,
1524 .partial = partial,
1525 .flags = flags,
1526 .ops = &sock_pipe_buf_ops,
1527 .spd_release = sock_spd_release,
1528 };
David S. Millerfbb398a2009-06-09 00:18:59 -07001529 struct sk_buff *frag_iter;
Jarek Poplawski7a67e562009-04-30 05:41:19 -07001530 struct sock *sk = skb->sk;
Jens Axboe35f3d142010-05-20 10:43:18 +02001531 int ret = 0;
1532
1533 if (splice_grow_spd(pipe, &spd))
1534 return -ENOMEM;
Jens Axboe9c55e012007-11-06 23:30:13 -08001535
1536 /*
1537 * __skb_splice_bits() only fails if the output has no room left,
1538 * so no point in going over the frag_list for the error case.
1539 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001540 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
Jens Axboe9c55e012007-11-06 23:30:13 -08001541 goto done;
1542 else if (!tlen)
1543 goto done;
1544
1545 /*
1546 * now see if we have a frag_list to map
1547 */
David S. Millerfbb398a2009-06-09 00:18:59 -07001548 skb_walk_frags(skb, frag_iter) {
1549 if (!tlen)
1550 break;
Jens Axboe35f3d142010-05-20 10:43:18 +02001551 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
David S. Millerfbb398a2009-06-09 00:18:59 -07001552 break;
Jens Axboe9c55e012007-11-06 23:30:13 -08001553 }
1554
1555done:
Jens Axboe9c55e012007-11-06 23:30:13 -08001556 if (spd.nr_pages) {
Jens Axboe9c55e012007-11-06 23:30:13 -08001557 /*
1558 * Drop the socket lock, otherwise we have reverse
1559 * locking dependencies between sk_lock and i_mutex
1560 * here as compared to sendfile(). We enter here
1561 * with the socket lock held, and splice_to_pipe() will
1562 * grab the pipe inode lock. For sendfile() emulation,
1563 * we call into ->sendpage() with the i_mutex lock held
1564 * and networking will grab the socket lock.
1565 */
Octavian Purdila293ad602008-06-04 15:45:58 -07001566 release_sock(sk);
Jens Axboe9c55e012007-11-06 23:30:13 -08001567 ret = splice_to_pipe(pipe, &spd);
Octavian Purdila293ad602008-06-04 15:45:58 -07001568 lock_sock(sk);
Jens Axboe9c55e012007-11-06 23:30:13 -08001569 }
1570
Jens Axboe35f3d142010-05-20 10:43:18 +02001571 splice_shrink_spd(pipe, &spd);
1572 return ret;
Jens Axboe9c55e012007-11-06 23:30:13 -08001573}
1574
Herbert Xu357b40a2005-04-19 22:30:14 -07001575/**
1576 * skb_store_bits - store bits from kernel buffer to skb
1577 * @skb: destination buffer
1578 * @offset: offset in destination
1579 * @from: source buffer
1580 * @len: number of bytes to copy
1581 *
1582 * Copy the specified number of bytes from the source buffer to the
1583 * destination skb. This function handles all the messy bits of
1584 * traversing fragment lists and such.
1585 */
1586
Stephen Hemminger0c6fcc82007-04-20 16:40:01 -07001587int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
Herbert Xu357b40a2005-04-19 22:30:14 -07001588{
David S. Miller1a028e52007-04-27 15:21:23 -07001589 int start = skb_headlen(skb);
David S. Millerfbb398a2009-06-09 00:18:59 -07001590 struct sk_buff *frag_iter;
1591 int i, copy;
Herbert Xu357b40a2005-04-19 22:30:14 -07001592
1593 if (offset > (int)skb->len - len)
1594 goto fault;
1595
David S. Miller1a028e52007-04-27 15:21:23 -07001596 if ((copy = start - offset) > 0) {
Herbert Xu357b40a2005-04-19 22:30:14 -07001597 if (copy > len)
1598 copy = len;
Arnaldo Carvalho de Melo27d7ff42007-03-31 11:55:19 -03001599 skb_copy_to_linear_data_offset(skb, offset, from, copy);
Herbert Xu357b40a2005-04-19 22:30:14 -07001600 if ((len -= copy) == 0)
1601 return 0;
1602 offset += copy;
1603 from += copy;
1604 }
1605
1606 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1607 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
David S. Miller1a028e52007-04-27 15:21:23 -07001608 int end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001609
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001610 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07001611
1612 end = start + frag->size;
Herbert Xu357b40a2005-04-19 22:30:14 -07001613 if ((copy = end - offset) > 0) {
1614 u8 *vaddr;
1615
1616 if (copy > len)
1617 copy = len;
1618
1619 vaddr = kmap_skb_frag(frag);
David S. Miller1a028e52007-04-27 15:21:23 -07001620 memcpy(vaddr + frag->page_offset + offset - start,
1621 from, copy);
Herbert Xu357b40a2005-04-19 22:30:14 -07001622 kunmap_skb_frag(vaddr);
1623
1624 if ((len -= copy) == 0)
1625 return 0;
1626 offset += copy;
1627 from += copy;
1628 }
David S. Miller1a028e52007-04-27 15:21:23 -07001629 start = end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001630 }
1631
David S. Millerfbb398a2009-06-09 00:18:59 -07001632 skb_walk_frags(skb, frag_iter) {
1633 int end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001634
David S. Millerfbb398a2009-06-09 00:18:59 -07001635 WARN_ON(start > offset + len);
Herbert Xu357b40a2005-04-19 22:30:14 -07001636
David S. Millerfbb398a2009-06-09 00:18:59 -07001637 end = start + frag_iter->len;
1638 if ((copy = end - offset) > 0) {
1639 if (copy > len)
1640 copy = len;
1641 if (skb_store_bits(frag_iter, offset - start,
1642 from, copy))
1643 goto fault;
1644 if ((len -= copy) == 0)
1645 return 0;
1646 offset += copy;
1647 from += copy;
Herbert Xu357b40a2005-04-19 22:30:14 -07001648 }
David S. Millerfbb398a2009-06-09 00:18:59 -07001649 start = end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001650 }
1651 if (!len)
1652 return 0;
1653
1654fault:
1655 return -EFAULT;
1656}
Herbert Xu357b40a2005-04-19 22:30:14 -07001657EXPORT_SYMBOL(skb_store_bits);
1658
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659/* Checksum skb data. */
1660
Al Viro2bbbc862006-11-14 21:37:14 -08001661__wsum skb_checksum(const struct sk_buff *skb, int offset,
1662 int len, __wsum csum)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663{
David S. Miller1a028e52007-04-27 15:21:23 -07001664 int start = skb_headlen(skb);
1665 int i, copy = start - offset;
David S. Millerfbb398a2009-06-09 00:18:59 -07001666 struct sk_buff *frag_iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 int pos = 0;
1668
1669 /* Checksum header. */
1670 if (copy > 0) {
1671 if (copy > len)
1672 copy = len;
1673 csum = csum_partial(skb->data + offset, copy, csum);
1674 if ((len -= copy) == 0)
1675 return csum;
1676 offset += copy;
1677 pos = copy;
1678 }
1679
1680 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07001681 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001683 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07001684
1685 end = start + skb_shinfo(skb)->frags[i].size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 if ((copy = end - offset) > 0) {
Al Viro44bb9362006-11-14 21:36:14 -08001687 __wsum csum2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688 u8 *vaddr;
1689 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1690
1691 if (copy > len)
1692 copy = len;
1693 vaddr = kmap_skb_frag(frag);
David S. Miller1a028e52007-04-27 15:21:23 -07001694 csum2 = csum_partial(vaddr + frag->page_offset +
1695 offset - start, copy, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 kunmap_skb_frag(vaddr);
1697 csum = csum_block_add(csum, csum2, pos);
1698 if (!(len -= copy))
1699 return csum;
1700 offset += copy;
1701 pos += copy;
1702 }
David S. Miller1a028e52007-04-27 15:21:23 -07001703 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 }
1705
David S. Millerfbb398a2009-06-09 00:18:59 -07001706 skb_walk_frags(skb, frag_iter) {
1707 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708
David S. Millerfbb398a2009-06-09 00:18:59 -07001709 WARN_ON(start > offset + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710
David S. Millerfbb398a2009-06-09 00:18:59 -07001711 end = start + frag_iter->len;
1712 if ((copy = end - offset) > 0) {
1713 __wsum csum2;
1714 if (copy > len)
1715 copy = len;
1716 csum2 = skb_checksum(frag_iter, offset - start,
1717 copy, 0);
1718 csum = csum_block_add(csum, csum2, pos);
1719 if ((len -= copy) == 0)
1720 return csum;
1721 offset += copy;
1722 pos += copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 }
David S. Millerfbb398a2009-06-09 00:18:59 -07001724 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 }
Kris Katterjohn09a62662006-01-08 22:24:28 -08001726 BUG_ON(len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727
1728 return csum;
1729}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001730EXPORT_SYMBOL(skb_checksum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731
1732/* Both of above in one bottle. */
1733
Al Viro81d77662006-11-14 21:37:33 -08001734__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1735 u8 *to, int len, __wsum csum)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736{
David S. Miller1a028e52007-04-27 15:21:23 -07001737 int start = skb_headlen(skb);
1738 int i, copy = start - offset;
David S. Millerfbb398a2009-06-09 00:18:59 -07001739 struct sk_buff *frag_iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 int pos = 0;
1741
1742 /* Copy header. */
1743 if (copy > 0) {
1744 if (copy > len)
1745 copy = len;
1746 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1747 copy, csum);
1748 if ((len -= copy) == 0)
1749 return csum;
1750 offset += copy;
1751 to += copy;
1752 pos = copy;
1753 }
1754
1755 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07001756 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001758 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07001759
1760 end = start + skb_shinfo(skb)->frags[i].size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 if ((copy = end - offset) > 0) {
Al Viro50842052006-11-14 21:36:34 -08001762 __wsum csum2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 u8 *vaddr;
1764 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1765
1766 if (copy > len)
1767 copy = len;
1768 vaddr = kmap_skb_frag(frag);
1769 csum2 = csum_partial_copy_nocheck(vaddr +
David S. Miller1a028e52007-04-27 15:21:23 -07001770 frag->page_offset +
1771 offset - start, to,
1772 copy, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 kunmap_skb_frag(vaddr);
1774 csum = csum_block_add(csum, csum2, pos);
1775 if (!(len -= copy))
1776 return csum;
1777 offset += copy;
1778 to += copy;
1779 pos += copy;
1780 }
David S. Miller1a028e52007-04-27 15:21:23 -07001781 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 }
1783
David S. Millerfbb398a2009-06-09 00:18:59 -07001784 skb_walk_frags(skb, frag_iter) {
1785 __wsum csum2;
1786 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787
David S. Millerfbb398a2009-06-09 00:18:59 -07001788 WARN_ON(start > offset + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789
David S. Millerfbb398a2009-06-09 00:18:59 -07001790 end = start + frag_iter->len;
1791 if ((copy = end - offset) > 0) {
1792 if (copy > len)
1793 copy = len;
1794 csum2 = skb_copy_and_csum_bits(frag_iter,
1795 offset - start,
1796 to, copy, 0);
1797 csum = csum_block_add(csum, csum2, pos);
1798 if ((len -= copy) == 0)
1799 return csum;
1800 offset += copy;
1801 to += copy;
1802 pos += copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 }
David S. Millerfbb398a2009-06-09 00:18:59 -07001804 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 }
Kris Katterjohn09a62662006-01-08 22:24:28 -08001806 BUG_ON(len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 return csum;
1808}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001809EXPORT_SYMBOL(skb_copy_and_csum_bits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810
1811void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1812{
Al Virod3bc23e2006-11-14 21:24:49 -08001813 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 long csstart;
1815
Patrick McHardy84fa7932006-08-29 16:44:56 -07001816 if (skb->ip_summed == CHECKSUM_PARTIAL)
Herbert Xu663ead32007-04-09 11:59:07 -07001817 csstart = skb->csum_start - skb_headroom(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 else
1819 csstart = skb_headlen(skb);
1820
Kris Katterjohn09a62662006-01-08 22:24:28 -08001821 BUG_ON(csstart > skb_headlen(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03001823 skb_copy_from_linear_data(skb, to, csstart);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824
1825 csum = 0;
1826 if (csstart != skb->len)
1827 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1828 skb->len - csstart, 0);
1829
Patrick McHardy84fa7932006-08-29 16:44:56 -07001830 if (skb->ip_summed == CHECKSUM_PARTIAL) {
Al Viroff1dcad2006-11-20 18:07:29 -08001831 long csstuff = csstart + skb->csum_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832
Al Virod3bc23e2006-11-14 21:24:49 -08001833 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 }
1835}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001836EXPORT_SYMBOL(skb_copy_and_csum_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837
1838/**
1839 * skb_dequeue - remove from the head of the queue
1840 * @list: list to dequeue from
1841 *
1842 * Remove the head of the list. The list lock is taken so the function
1843 * may be used safely with other locking list functions. The head item is
1844 * returned or %NULL if the list is empty.
1845 */
1846
1847struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1848{
1849 unsigned long flags;
1850 struct sk_buff *result;
1851
1852 spin_lock_irqsave(&list->lock, flags);
1853 result = __skb_dequeue(list);
1854 spin_unlock_irqrestore(&list->lock, flags);
1855 return result;
1856}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001857EXPORT_SYMBOL(skb_dequeue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858
1859/**
1860 * skb_dequeue_tail - remove from the tail of the queue
1861 * @list: list to dequeue from
1862 *
1863 * Remove the tail of the list. The list lock is taken so the function
1864 * may be used safely with other locking list functions. The tail item is
1865 * returned or %NULL if the list is empty.
1866 */
1867struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1868{
1869 unsigned long flags;
1870 struct sk_buff *result;
1871
1872 spin_lock_irqsave(&list->lock, flags);
1873 result = __skb_dequeue_tail(list);
1874 spin_unlock_irqrestore(&list->lock, flags);
1875 return result;
1876}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001877EXPORT_SYMBOL(skb_dequeue_tail);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878
1879/**
1880 * skb_queue_purge - empty a list
1881 * @list: list to empty
1882 *
1883 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1884 * the list and one reference dropped. This function takes the list
1885 * lock and is atomic with respect to other list locking functions.
1886 */
1887void skb_queue_purge(struct sk_buff_head *list)
1888{
1889 struct sk_buff *skb;
1890 while ((skb = skb_dequeue(list)) != NULL)
1891 kfree_skb(skb);
1892}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001893EXPORT_SYMBOL(skb_queue_purge);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894
1895/**
1896 * skb_queue_head - queue a buffer at the list head
1897 * @list: list to use
1898 * @newsk: buffer to queue
1899 *
1900 * Queue a buffer at the start of the list. This function takes the
1901 * list lock and can be used safely with other locking &sk_buff functions
1902 * safely.
1903 *
1904 * A buffer cannot be placed on two lists at the same time.
1905 */
1906void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1907{
1908 unsigned long flags;
1909
1910 spin_lock_irqsave(&list->lock, flags);
1911 __skb_queue_head(list, newsk);
1912 spin_unlock_irqrestore(&list->lock, flags);
1913}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001914EXPORT_SYMBOL(skb_queue_head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915
1916/**
1917 * skb_queue_tail - queue a buffer at the list tail
1918 * @list: list to use
1919 * @newsk: buffer to queue
1920 *
1921 * Queue a buffer at the tail of the list. This function takes the
1922 * list lock and can be used safely with other locking &sk_buff functions
1923 * safely.
1924 *
1925 * A buffer cannot be placed on two lists at the same time.
1926 */
1927void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1928{
1929 unsigned long flags;
1930
1931 spin_lock_irqsave(&list->lock, flags);
1932 __skb_queue_tail(list, newsk);
1933 spin_unlock_irqrestore(&list->lock, flags);
1934}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001935EXPORT_SYMBOL(skb_queue_tail);
David S. Miller8728b832005-08-09 19:25:21 -07001936
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937/**
1938 * skb_unlink - remove a buffer from a list
1939 * @skb: buffer to remove
David S. Miller8728b832005-08-09 19:25:21 -07001940 * @list: list to use
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 *
David S. Miller8728b832005-08-09 19:25:21 -07001942 * Remove a packet from a list. The list locks are taken and this
1943 * function is atomic with respect to other list locked calls
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 *
David S. Miller8728b832005-08-09 19:25:21 -07001945 * You must know what list the SKB is on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 */
David S. Miller8728b832005-08-09 19:25:21 -07001947void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948{
David S. Miller8728b832005-08-09 19:25:21 -07001949 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950
David S. Miller8728b832005-08-09 19:25:21 -07001951 spin_lock_irqsave(&list->lock, flags);
1952 __skb_unlink(skb, list);
1953 spin_unlock_irqrestore(&list->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001955EXPORT_SYMBOL(skb_unlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957/**
1958 * skb_append - append a buffer
1959 * @old: buffer to insert after
1960 * @newsk: buffer to insert
David S. Miller8728b832005-08-09 19:25:21 -07001961 * @list: list to use
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 *
1963 * Place a packet after a given packet in a list. The list locks are taken
1964 * and this function is atomic with respect to other list locked calls.
1965 * A buffer cannot be placed on two lists at the same time.
1966 */
David S. Miller8728b832005-08-09 19:25:21 -07001967void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968{
1969 unsigned long flags;
1970
David S. Miller8728b832005-08-09 19:25:21 -07001971 spin_lock_irqsave(&list->lock, flags);
Gerrit Renker7de6c032008-04-14 00:05:09 -07001972 __skb_queue_after(list, old, newsk);
David S. Miller8728b832005-08-09 19:25:21 -07001973 spin_unlock_irqrestore(&list->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001975EXPORT_SYMBOL(skb_append);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976
1977/**
1978 * skb_insert - insert a buffer
1979 * @old: buffer to insert before
1980 * @newsk: buffer to insert
David S. Miller8728b832005-08-09 19:25:21 -07001981 * @list: list to use
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 *
David S. Miller8728b832005-08-09 19:25:21 -07001983 * Place a packet before a given packet in a list. The list locks are
1984 * taken and this function is atomic with respect to other list locked
1985 * calls.
1986 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 * A buffer cannot be placed on two lists at the same time.
1988 */
David S. Miller8728b832005-08-09 19:25:21 -07001989void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990{
1991 unsigned long flags;
1992
David S. Miller8728b832005-08-09 19:25:21 -07001993 spin_lock_irqsave(&list->lock, flags);
1994 __skb_insert(newsk, old->prev, old, list);
1995 spin_unlock_irqrestore(&list->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001997EXPORT_SYMBOL(skb_insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999static inline void skb_split_inside_header(struct sk_buff *skb,
2000 struct sk_buff* skb1,
2001 const u32 len, const int pos)
2002{
2003 int i;
2004
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03002005 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2006 pos - len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 /* And move data appendix as is. */
2008 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2009 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2010
2011 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2012 skb_shinfo(skb)->nr_frags = 0;
2013 skb1->data_len = skb->data_len;
2014 skb1->len += skb1->data_len;
2015 skb->data_len = 0;
2016 skb->len = len;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07002017 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018}
2019
2020static inline void skb_split_no_header(struct sk_buff *skb,
2021 struct sk_buff* skb1,
2022 const u32 len, int pos)
2023{
2024 int i, k = 0;
2025 const int nfrags = skb_shinfo(skb)->nr_frags;
2026
2027 skb_shinfo(skb)->nr_frags = 0;
2028 skb1->len = skb1->data_len = skb->len - len;
2029 skb->len = len;
2030 skb->data_len = len - pos;
2031
2032 for (i = 0; i < nfrags; i++) {
2033 int size = skb_shinfo(skb)->frags[i].size;
2034
2035 if (pos + size > len) {
2036 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2037
2038 if (pos < len) {
2039 /* Split frag.
2040 * We have two variants in this case:
2041 * 1. Move all the frag to the second
2042 * part, if it is possible. F.e.
2043 * this approach is mandatory for TUX,
2044 * where splitting is expensive.
2045 * 2. Split is accurately. We make this.
2046 */
2047 get_page(skb_shinfo(skb)->frags[i].page);
2048 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2049 skb_shinfo(skb1)->frags[0].size -= len - pos;
2050 skb_shinfo(skb)->frags[i].size = len - pos;
2051 skb_shinfo(skb)->nr_frags++;
2052 }
2053 k++;
2054 } else
2055 skb_shinfo(skb)->nr_frags++;
2056 pos += size;
2057 }
2058 skb_shinfo(skb1)->nr_frags = k;
2059}
2060
2061/**
2062 * skb_split - Split fragmented skb to two parts at length len.
2063 * @skb: the buffer to split
2064 * @skb1: the buffer to receive the second part
2065 * @len: new length for skb
2066 */
2067void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2068{
2069 int pos = skb_headlen(skb);
2070
2071 if (len < pos) /* Split line is inside header. */
2072 skb_split_inside_header(skb, skb1, len, pos);
2073 else /* Second chunk has no header, nothing to copy. */
2074 skb_split_no_header(skb, skb1, len, pos);
2075}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002076EXPORT_SYMBOL(skb_split);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077
Ilpo Järvinen9f782db2008-11-25 13:57:01 -08002078/* Shifting from/to a cloned skb is a no-go.
2079 *
2080 * Caller cannot keep skb_shinfo related pointers past calling here!
2081 */
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002082static int skb_prepare_for_shift(struct sk_buff *skb)
2083{
Ilpo Järvinen0ace2852008-11-24 21:30:21 -08002084 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002085}
2086
2087/**
2088 * skb_shift - Shifts paged data partially from skb to another
2089 * @tgt: buffer into which tail data gets added
2090 * @skb: buffer from which the paged data comes from
2091 * @shiftlen: shift up to this many bytes
2092 *
2093 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2094 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2095 * It's up to caller to free skb if everything was shifted.
2096 *
2097 * If @tgt runs out of frags, the whole operation is aborted.
2098 *
2099 * Skb cannot include anything else but paged data while tgt is allowed
2100 * to have non-paged data as well.
2101 *
2102 * TODO: full sized shift could be optimized but that would need
2103 * specialized skb free'er to handle frags without up-to-date nr_frags.
2104 */
2105int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2106{
2107 int from, to, merge, todo;
2108 struct skb_frag_struct *fragfrom, *fragto;
2109
2110 BUG_ON(shiftlen > skb->len);
2111 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2112
2113 todo = shiftlen;
2114 from = 0;
2115 to = skb_shinfo(tgt)->nr_frags;
2116 fragfrom = &skb_shinfo(skb)->frags[from];
2117
2118 /* Actual merge is delayed until the point when we know we can
2119 * commit all, so that we don't have to undo partial changes
2120 */
2121 if (!to ||
2122 !skb_can_coalesce(tgt, to, fragfrom->page, fragfrom->page_offset)) {
2123 merge = -1;
2124 } else {
2125 merge = to - 1;
2126
2127 todo -= fragfrom->size;
2128 if (todo < 0) {
2129 if (skb_prepare_for_shift(skb) ||
2130 skb_prepare_for_shift(tgt))
2131 return 0;
2132
Ilpo Järvinen9f782db2008-11-25 13:57:01 -08002133 /* All previous frag pointers might be stale! */
2134 fragfrom = &skb_shinfo(skb)->frags[from];
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002135 fragto = &skb_shinfo(tgt)->frags[merge];
2136
2137 fragto->size += shiftlen;
2138 fragfrom->size -= shiftlen;
2139 fragfrom->page_offset += shiftlen;
2140
2141 goto onlymerged;
2142 }
2143
2144 from++;
2145 }
2146
2147 /* Skip full, not-fitting skb to avoid expensive operations */
2148 if ((shiftlen == skb->len) &&
2149 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2150 return 0;
2151
2152 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2153 return 0;
2154
2155 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2156 if (to == MAX_SKB_FRAGS)
2157 return 0;
2158
2159 fragfrom = &skb_shinfo(skb)->frags[from];
2160 fragto = &skb_shinfo(tgt)->frags[to];
2161
2162 if (todo >= fragfrom->size) {
2163 *fragto = *fragfrom;
2164 todo -= fragfrom->size;
2165 from++;
2166 to++;
2167
2168 } else {
2169 get_page(fragfrom->page);
2170 fragto->page = fragfrom->page;
2171 fragto->page_offset = fragfrom->page_offset;
2172 fragto->size = todo;
2173
2174 fragfrom->page_offset += todo;
2175 fragfrom->size -= todo;
2176 todo = 0;
2177
2178 to++;
2179 break;
2180 }
2181 }
2182
2183 /* Ready to "commit" this state change to tgt */
2184 skb_shinfo(tgt)->nr_frags = to;
2185
2186 if (merge >= 0) {
2187 fragfrom = &skb_shinfo(skb)->frags[0];
2188 fragto = &skb_shinfo(tgt)->frags[merge];
2189
2190 fragto->size += fragfrom->size;
2191 put_page(fragfrom->page);
2192 }
2193
2194 /* Reposition in the original skb */
2195 to = 0;
2196 while (from < skb_shinfo(skb)->nr_frags)
2197 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2198 skb_shinfo(skb)->nr_frags = to;
2199
2200 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2201
2202onlymerged:
2203 /* Most likely the tgt won't ever need its checksum anymore, skb on
2204 * the other hand might need it if it needs to be resent
2205 */
2206 tgt->ip_summed = CHECKSUM_PARTIAL;
2207 skb->ip_summed = CHECKSUM_PARTIAL;
2208
2209 /* Yak, is it really working this way? Some helper please? */
2210 skb->len -= shiftlen;
2211 skb->data_len -= shiftlen;
2212 skb->truesize -= shiftlen;
2213 tgt->len += shiftlen;
2214 tgt->data_len += shiftlen;
2215 tgt->truesize += shiftlen;
2216
2217 return shiftlen;
2218}
2219
Thomas Graf677e90e2005-06-23 20:59:51 -07002220/**
2221 * skb_prepare_seq_read - Prepare a sequential read of skb data
2222 * @skb: the buffer to read
2223 * @from: lower offset of data to be read
2224 * @to: upper offset of data to be read
2225 * @st: state variable
2226 *
2227 * Initializes the specified state variable. Must be called before
2228 * invoking skb_seq_read() for the first time.
2229 */
2230void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2231 unsigned int to, struct skb_seq_state *st)
2232{
2233 st->lower_offset = from;
2234 st->upper_offset = to;
2235 st->root_skb = st->cur_skb = skb;
2236 st->frag_idx = st->stepped_offset = 0;
2237 st->frag_data = NULL;
2238}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002239EXPORT_SYMBOL(skb_prepare_seq_read);
Thomas Graf677e90e2005-06-23 20:59:51 -07002240
2241/**
2242 * skb_seq_read - Sequentially read skb data
2243 * @consumed: number of bytes consumed by the caller so far
2244 * @data: destination pointer for data to be returned
2245 * @st: state variable
2246 *
2247 * Reads a block of skb data at &consumed relative to the
2248 * lower offset specified to skb_prepare_seq_read(). Assigns
2249 * the head of the data block to &data and returns the length
2250 * of the block or 0 if the end of the skb data or the upper
2251 * offset has been reached.
2252 *
2253 * The caller is not required to consume all of the data
2254 * returned, i.e. &consumed is typically set to the number
2255 * of bytes already consumed and the next call to
2256 * skb_seq_read() will return the remaining part of the block.
2257 *
Randy Dunlapbc2cda12008-02-13 15:03:25 -08002258 * Note 1: The size of each block of data returned can be arbitary,
Thomas Graf677e90e2005-06-23 20:59:51 -07002259 * this limitation is the cost for zerocopy seqeuental
2260 * reads of potentially non linear data.
2261 *
Randy Dunlapbc2cda12008-02-13 15:03:25 -08002262 * Note 2: Fragment lists within fragments are not implemented
Thomas Graf677e90e2005-06-23 20:59:51 -07002263 * at the moment, state->root_skb could be replaced with
2264 * a stack for this purpose.
2265 */
2266unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2267 struct skb_seq_state *st)
2268{
2269 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2270 skb_frag_t *frag;
2271
2272 if (unlikely(abs_offset >= st->upper_offset))
2273 return 0;
2274
2275next_skb:
Herbert Xu95e3b242009-01-29 16:07:52 -08002276 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
Thomas Graf677e90e2005-06-23 20:59:51 -07002277
Thomas Chenault995b3372009-05-18 21:43:27 -07002278 if (abs_offset < block_limit && !st->frag_data) {
Herbert Xu95e3b242009-01-29 16:07:52 -08002279 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
Thomas Graf677e90e2005-06-23 20:59:51 -07002280 return block_limit - abs_offset;
2281 }
2282
2283 if (st->frag_idx == 0 && !st->frag_data)
2284 st->stepped_offset += skb_headlen(st->cur_skb);
2285
2286 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2287 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2288 block_limit = frag->size + st->stepped_offset;
2289
2290 if (abs_offset < block_limit) {
2291 if (!st->frag_data)
2292 st->frag_data = kmap_skb_frag(frag);
2293
2294 *data = (u8 *) st->frag_data + frag->page_offset +
2295 (abs_offset - st->stepped_offset);
2296
2297 return block_limit - abs_offset;
2298 }
2299
2300 if (st->frag_data) {
2301 kunmap_skb_frag(st->frag_data);
2302 st->frag_data = NULL;
2303 }
2304
2305 st->frag_idx++;
2306 st->stepped_offset += frag->size;
2307 }
2308
Olaf Kirch5b5a60d2007-06-23 23:11:52 -07002309 if (st->frag_data) {
2310 kunmap_skb_frag(st->frag_data);
2311 st->frag_data = NULL;
2312 }
2313
David S. Miller21dc3302010-08-23 00:13:46 -07002314 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
Shyam Iyer71b33462009-01-29 16:12:42 -08002315 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
Thomas Graf677e90e2005-06-23 20:59:51 -07002316 st->frag_idx = 0;
2317 goto next_skb;
Shyam Iyer71b33462009-01-29 16:12:42 -08002318 } else if (st->cur_skb->next) {
2319 st->cur_skb = st->cur_skb->next;
Herbert Xu95e3b242009-01-29 16:07:52 -08002320 st->frag_idx = 0;
Thomas Graf677e90e2005-06-23 20:59:51 -07002321 goto next_skb;
2322 }
2323
2324 return 0;
2325}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002326EXPORT_SYMBOL(skb_seq_read);
Thomas Graf677e90e2005-06-23 20:59:51 -07002327
2328/**
2329 * skb_abort_seq_read - Abort a sequential read of skb data
2330 * @st: state variable
2331 *
2332 * Must be called if skb_seq_read() was not called until it
2333 * returned 0.
2334 */
2335void skb_abort_seq_read(struct skb_seq_state *st)
2336{
2337 if (st->frag_data)
2338 kunmap_skb_frag(st->frag_data);
2339}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002340EXPORT_SYMBOL(skb_abort_seq_read);
Thomas Graf677e90e2005-06-23 20:59:51 -07002341
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002342#define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2343
2344static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2345 struct ts_config *conf,
2346 struct ts_state *state)
2347{
2348 return skb_seq_read(offset, text, TS_SKB_CB(state));
2349}
2350
2351static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2352{
2353 skb_abort_seq_read(TS_SKB_CB(state));
2354}
2355
2356/**
2357 * skb_find_text - Find a text pattern in skb data
2358 * @skb: the buffer to look in
2359 * @from: search offset
2360 * @to: search limit
2361 * @config: textsearch configuration
2362 * @state: uninitialized textsearch state variable
2363 *
2364 * Finds a pattern in the skb data according to the specified
2365 * textsearch configuration. Use textsearch_next() to retrieve
2366 * subsequent occurrences of the pattern. Returns the offset
2367 * to the first occurrence or UINT_MAX if no match was found.
2368 */
2369unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2370 unsigned int to, struct ts_config *config,
2371 struct ts_state *state)
2372{
Phil Oesterf72b9482006-06-26 00:00:57 -07002373 unsigned int ret;
2374
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002375 config->get_next_block = skb_ts_get_next_block;
2376 config->finish = skb_ts_finish;
2377
2378 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2379
Phil Oesterf72b9482006-06-26 00:00:57 -07002380 ret = textsearch_find(config, state);
2381 return (ret <= to - from ? ret : UINT_MAX);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002382}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002383EXPORT_SYMBOL(skb_find_text);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002384
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002385/**
2386 * skb_append_datato_frags: - append the user data to a skb
2387 * @sk: sock structure
2388 * @skb: skb structure to be appened with user data.
2389 * @getfrag: call back function to be used for getting the user data
2390 * @from: pointer to user message iov
2391 * @length: length of the iov message
2392 *
2393 * Description: This procedure append the user data in the fragment part
2394 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2395 */
2396int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
Martin Waitzdab96302005-12-05 13:40:12 -08002397 int (*getfrag)(void *from, char *to, int offset,
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002398 int len, int odd, struct sk_buff *skb),
2399 void *from, int length)
2400{
2401 int frg_cnt = 0;
2402 skb_frag_t *frag = NULL;
2403 struct page *page = NULL;
2404 int copy, left;
2405 int offset = 0;
2406 int ret;
2407
2408 do {
2409 /* Return error if we don't have space for new frag */
2410 frg_cnt = skb_shinfo(skb)->nr_frags;
2411 if (frg_cnt >= MAX_SKB_FRAGS)
2412 return -EFAULT;
2413
2414 /* allocate a new page for next frag */
2415 page = alloc_pages(sk->sk_allocation, 0);
2416
2417 /* If alloc_page fails just return failure and caller will
2418 * free previous allocated pages by doing kfree_skb()
2419 */
2420 if (page == NULL)
2421 return -ENOMEM;
2422
2423 /* initialize the next frag */
2424 sk->sk_sndmsg_page = page;
2425 sk->sk_sndmsg_off = 0;
2426 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2427 skb->truesize += PAGE_SIZE;
2428 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2429
2430 /* get the new initialized frag */
2431 frg_cnt = skb_shinfo(skb)->nr_frags;
2432 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2433
2434 /* copy the user data to page */
2435 left = PAGE_SIZE - frag->page_offset;
2436 copy = (length > left)? left : length;
2437
2438 ret = getfrag(from, (page_address(frag->page) +
2439 frag->page_offset + frag->size),
2440 offset, copy, 0, skb);
2441 if (ret < 0)
2442 return -EFAULT;
2443
2444 /* copy was successful so update the size parameters */
2445 sk->sk_sndmsg_off += copy;
2446 frag->size += copy;
2447 skb->len += copy;
2448 skb->data_len += copy;
2449 offset += copy;
2450 length -= copy;
2451
2452 } while (length > 0);
2453
2454 return 0;
2455}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002456EXPORT_SYMBOL(skb_append_datato_frags);
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002457
Herbert Xucbb042f2006-03-20 22:43:56 -08002458/**
2459 * skb_pull_rcsum - pull skb and update receive checksum
2460 * @skb: buffer to update
Herbert Xucbb042f2006-03-20 22:43:56 -08002461 * @len: length of data pulled
2462 *
2463 * This function performs an skb_pull on the packet and updates
Urs Thuermannfee54fa2008-02-12 22:03:25 -08002464 * the CHECKSUM_COMPLETE checksum. It should be used on
Patrick McHardy84fa7932006-08-29 16:44:56 -07002465 * receive path processing instead of skb_pull unless you know
2466 * that the checksum difference is zero (e.g., a valid IP header)
2467 * or you are setting ip_summed to CHECKSUM_NONE.
Herbert Xucbb042f2006-03-20 22:43:56 -08002468 */
2469unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2470{
2471 BUG_ON(len > skb->len);
2472 skb->len -= len;
2473 BUG_ON(skb->len < skb->data_len);
2474 skb_postpull_rcsum(skb, skb->data, len);
2475 return skb->data += len;
2476}
Arnaldo Carvalho de Melof94691a2006-03-20 22:47:55 -08002477EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2478
Herbert Xuf4c50d92006-06-22 03:02:40 -07002479/**
2480 * skb_segment - Perform protocol segmentation on skb.
2481 * @skb: buffer to segment
Herbert Xu576a30e2006-06-27 13:22:38 -07002482 * @features: features for the output path (see dev->features)
Herbert Xuf4c50d92006-06-22 03:02:40 -07002483 *
2484 * This function performs segmentation on the given skb. It returns
Ben Hutchings4c821d72008-04-13 21:52:48 -07002485 * a pointer to the first in a list of new skbs for the segments.
2486 * In case of error it returns ERR_PTR(err).
Herbert Xuf4c50d92006-06-22 03:02:40 -07002487 */
Herbert Xu576a30e2006-06-27 13:22:38 -07002488struct sk_buff *skb_segment(struct sk_buff *skb, int features)
Herbert Xuf4c50d92006-06-22 03:02:40 -07002489{
2490 struct sk_buff *segs = NULL;
2491 struct sk_buff *tail = NULL;
Herbert Xu89319d382008-12-15 23:26:06 -08002492 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002493 unsigned int mss = skb_shinfo(skb)->gso_size;
Arnaldo Carvalho de Melo98e399f2007-03-19 15:33:04 -07002494 unsigned int doffset = skb->data - skb_mac_header(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002495 unsigned int offset = doffset;
2496 unsigned int headroom;
2497 unsigned int len;
Herbert Xu576a30e2006-06-27 13:22:38 -07002498 int sg = features & NETIF_F_SG;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002499 int nfrags = skb_shinfo(skb)->nr_frags;
2500 int err = -ENOMEM;
2501 int i = 0;
2502 int pos;
2503
2504 __skb_push(skb, doffset);
2505 headroom = skb_headroom(skb);
2506 pos = skb_headlen(skb);
2507
2508 do {
2509 struct sk_buff *nskb;
2510 skb_frag_t *frag;
Herbert Xuc8884ed2006-10-29 15:59:41 -08002511 int hsize;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002512 int size;
2513
2514 len = skb->len - offset;
2515 if (len > mss)
2516 len = mss;
2517
2518 hsize = skb_headlen(skb) - offset;
2519 if (hsize < 0)
2520 hsize = 0;
Herbert Xuc8884ed2006-10-29 15:59:41 -08002521 if (hsize > len || !sg)
2522 hsize = len;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002523
Herbert Xu89319d382008-12-15 23:26:06 -08002524 if (!hsize && i >= nfrags) {
2525 BUG_ON(fskb->len != len);
2526
2527 pos += len;
2528 nskb = skb_clone(fskb, GFP_ATOMIC);
2529 fskb = fskb->next;
2530
2531 if (unlikely(!nskb))
2532 goto err;
2533
2534 hsize = skb_end_pointer(nskb) - nskb->head;
2535 if (skb_cow_head(nskb, doffset + headroom)) {
2536 kfree_skb(nskb);
2537 goto err;
2538 }
2539
2540 nskb->truesize += skb_end_pointer(nskb) - nskb->head -
2541 hsize;
2542 skb_release_head_state(nskb);
2543 __skb_push(nskb, doffset);
2544 } else {
2545 nskb = alloc_skb(hsize + doffset + headroom,
2546 GFP_ATOMIC);
2547
2548 if (unlikely(!nskb))
2549 goto err;
2550
2551 skb_reserve(nskb, headroom);
2552 __skb_put(nskb, doffset);
2553 }
Herbert Xuf4c50d92006-06-22 03:02:40 -07002554
2555 if (segs)
2556 tail->next = nskb;
2557 else
2558 segs = nskb;
2559 tail = nskb;
2560
Herbert Xu6f85a122008-08-15 14:55:02 -07002561 __copy_skb_header(nskb, skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002562 nskb->mac_len = skb->mac_len;
2563
Eric Dumazet3d3be432010-09-01 00:50:51 +00002564 /* nskb and skb might have different headroom */
2565 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2566 nskb->csum_start += skb_headroom(nskb) - headroom;
2567
Arnaldo Carvalho de Melo459a98e2007-03-19 15:30:44 -07002568 skb_reset_mac_header(nskb);
Arnaldo Carvalho de Meloddc7b8e2007-03-15 21:42:27 -03002569 skb_set_network_header(nskb, skb->mac_len);
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -07002570 nskb->transport_header = (nskb->network_header +
2571 skb_network_header_len(skb));
Herbert Xu89319d382008-12-15 23:26:06 -08002572 skb_copy_from_linear_data(skb, nskb->data, doffset);
2573
Herbert Xu2f181852009-03-28 23:39:18 -07002574 if (fskb != skb_shinfo(skb)->frag_list)
Herbert Xu89319d382008-12-15 23:26:06 -08002575 continue;
2576
Herbert Xuf4c50d92006-06-22 03:02:40 -07002577 if (!sg) {
Herbert Xu6f85a122008-08-15 14:55:02 -07002578 nskb->ip_summed = CHECKSUM_NONE;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002579 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2580 skb_put(nskb, len),
2581 len, 0);
2582 continue;
2583 }
2584
2585 frag = skb_shinfo(nskb)->frags;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002586
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03002587 skb_copy_from_linear_data_offset(skb, offset,
2588 skb_put(nskb, hsize), hsize);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002589
Herbert Xu89319d382008-12-15 23:26:06 -08002590 while (pos < offset + len && i < nfrags) {
Herbert Xuf4c50d92006-06-22 03:02:40 -07002591 *frag = skb_shinfo(skb)->frags[i];
2592 get_page(frag->page);
2593 size = frag->size;
2594
2595 if (pos < offset) {
2596 frag->page_offset += offset - pos;
2597 frag->size -= offset - pos;
2598 }
2599
Herbert Xu89319d382008-12-15 23:26:06 -08002600 skb_shinfo(nskb)->nr_frags++;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002601
2602 if (pos + size <= offset + len) {
2603 i++;
2604 pos += size;
2605 } else {
2606 frag->size -= pos + size - (offset + len);
Herbert Xu89319d382008-12-15 23:26:06 -08002607 goto skip_fraglist;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002608 }
2609
2610 frag++;
2611 }
2612
Herbert Xu89319d382008-12-15 23:26:06 -08002613 if (pos < offset + len) {
2614 struct sk_buff *fskb2 = fskb;
2615
2616 BUG_ON(pos + fskb->len != offset + len);
2617
2618 pos += fskb->len;
2619 fskb = fskb->next;
2620
2621 if (fskb2->next) {
2622 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2623 if (!fskb2)
2624 goto err;
2625 } else
2626 skb_get(fskb2);
2627
David S. Millerfbb398a2009-06-09 00:18:59 -07002628 SKB_FRAG_ASSERT(nskb);
Herbert Xu89319d382008-12-15 23:26:06 -08002629 skb_shinfo(nskb)->frag_list = fskb2;
2630 }
2631
2632skip_fraglist:
Herbert Xuf4c50d92006-06-22 03:02:40 -07002633 nskb->data_len = len - hsize;
2634 nskb->len += nskb->data_len;
2635 nskb->truesize += nskb->data_len;
2636 } while ((offset += len) < skb->len);
2637
2638 return segs;
2639
2640err:
2641 while ((skb = segs)) {
2642 segs = skb->next;
Patrick McHardyb08d5842007-02-27 09:57:37 -08002643 kfree_skb(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002644 }
2645 return ERR_PTR(err);
2646}
Herbert Xuf4c50d92006-06-22 03:02:40 -07002647EXPORT_SYMBOL_GPL(skb_segment);
2648
Herbert Xu71d93b32008-12-15 23:42:33 -08002649int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2650{
2651 struct sk_buff *p = *head;
2652 struct sk_buff *nskb;
Herbert Xu9aaa1562009-05-26 18:50:33 +00002653 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2654 struct skb_shared_info *pinfo = skb_shinfo(p);
Herbert Xu71d93b32008-12-15 23:42:33 -08002655 unsigned int headroom;
Herbert Xu86911732009-01-29 14:19:50 +00002656 unsigned int len = skb_gro_len(skb);
Herbert Xu67147ba2009-05-26 18:50:22 +00002657 unsigned int offset = skb_gro_offset(skb);
2658 unsigned int headlen = skb_headlen(skb);
Herbert Xu71d93b32008-12-15 23:42:33 -08002659
Herbert Xu86911732009-01-29 14:19:50 +00002660 if (p->len + len >= 65536)
Herbert Xu71d93b32008-12-15 23:42:33 -08002661 return -E2BIG;
2662
Herbert Xu9aaa1562009-05-26 18:50:33 +00002663 if (pinfo->frag_list)
Herbert Xu71d93b32008-12-15 23:42:33 -08002664 goto merge;
Herbert Xu67147ba2009-05-26 18:50:22 +00002665 else if (headlen <= offset) {
Herbert Xu42da6992009-05-26 18:50:19 +00002666 skb_frag_t *frag;
Herbert Xu66e92fc2009-05-26 18:50:32 +00002667 skb_frag_t *frag2;
Herbert Xu9aaa1562009-05-26 18:50:33 +00002668 int i = skbinfo->nr_frags;
2669 int nr_frags = pinfo->nr_frags + i;
Herbert Xu42da6992009-05-26 18:50:19 +00002670
Herbert Xu66e92fc2009-05-26 18:50:32 +00002671 offset -= headlen;
2672
2673 if (nr_frags > MAX_SKB_FRAGS)
Herbert Xu81705ad2009-01-29 14:19:51 +00002674 return -E2BIG;
2675
Herbert Xu9aaa1562009-05-26 18:50:33 +00002676 pinfo->nr_frags = nr_frags;
2677 skbinfo->nr_frags = 0;
Herbert Xuf5572062009-01-14 20:40:03 -08002678
Herbert Xu9aaa1562009-05-26 18:50:33 +00002679 frag = pinfo->frags + nr_frags;
2680 frag2 = skbinfo->frags + i;
Herbert Xu66e92fc2009-05-26 18:50:32 +00002681 do {
2682 *--frag = *--frag2;
2683 } while (--i);
2684
2685 frag->page_offset += offset;
2686 frag->size -= offset;
2687
Herbert Xuf5572062009-01-14 20:40:03 -08002688 skb->truesize -= skb->data_len;
2689 skb->len -= skb->data_len;
2690 skb->data_len = 0;
2691
Herbert Xu5d38a072009-01-04 16:13:40 -08002692 NAPI_GRO_CB(skb)->free = 1;
2693 goto done;
Herbert Xu69c0cab2009-11-17 05:18:18 -08002694 } else if (skb_gro_len(p) != pinfo->gso_size)
2695 return -E2BIG;
Herbert Xu71d93b32008-12-15 23:42:33 -08002696
2697 headroom = skb_headroom(p);
Eric Dumazet3d3be432010-09-01 00:50:51 +00002698 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
Herbert Xu71d93b32008-12-15 23:42:33 -08002699 if (unlikely(!nskb))
2700 return -ENOMEM;
2701
2702 __copy_skb_header(nskb, p);
2703 nskb->mac_len = p->mac_len;
2704
2705 skb_reserve(nskb, headroom);
Herbert Xu86911732009-01-29 14:19:50 +00002706 __skb_put(nskb, skb_gro_offset(p));
Herbert Xu71d93b32008-12-15 23:42:33 -08002707
Herbert Xu86911732009-01-29 14:19:50 +00002708 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
Herbert Xu71d93b32008-12-15 23:42:33 -08002709 skb_set_network_header(nskb, skb_network_offset(p));
2710 skb_set_transport_header(nskb, skb_transport_offset(p));
2711
Herbert Xu86911732009-01-29 14:19:50 +00002712 __skb_pull(p, skb_gro_offset(p));
2713 memcpy(skb_mac_header(nskb), skb_mac_header(p),
2714 p->data - skb_mac_header(p));
Herbert Xu71d93b32008-12-15 23:42:33 -08002715
2716 *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2717 skb_shinfo(nskb)->frag_list = p;
Herbert Xu9aaa1562009-05-26 18:50:33 +00002718 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
Herbert Xu622e0ca2010-05-20 23:07:56 -07002719 pinfo->gso_size = 0;
Herbert Xu71d93b32008-12-15 23:42:33 -08002720 skb_header_release(p);
2721 nskb->prev = p;
2722
2723 nskb->data_len += p->len;
2724 nskb->truesize += p->len;
2725 nskb->len += p->len;
2726
2727 *head = nskb;
2728 nskb->next = p->next;
2729 p->next = NULL;
2730
2731 p = nskb;
2732
2733merge:
Herbert Xu67147ba2009-05-26 18:50:22 +00002734 if (offset > headlen) {
Herbert Xu9aaa1562009-05-26 18:50:33 +00002735 skbinfo->frags[0].page_offset += offset - headlen;
2736 skbinfo->frags[0].size -= offset - headlen;
Herbert Xu67147ba2009-05-26 18:50:22 +00002737 offset = headlen;
Herbert Xu56035022009-02-05 21:26:52 -08002738 }
2739
Herbert Xu67147ba2009-05-26 18:50:22 +00002740 __skb_pull(skb, offset);
Herbert Xu56035022009-02-05 21:26:52 -08002741
Herbert Xu71d93b32008-12-15 23:42:33 -08002742 p->prev->next = skb;
2743 p->prev = skb;
2744 skb_header_release(skb);
2745
Herbert Xu5d38a072009-01-04 16:13:40 -08002746done:
2747 NAPI_GRO_CB(p)->count++;
Herbert Xu37fe4732009-01-17 19:48:13 +00002748 p->data_len += len;
2749 p->truesize += len;
2750 p->len += len;
Herbert Xu71d93b32008-12-15 23:42:33 -08002751
2752 NAPI_GRO_CB(skb)->same_flow = 1;
2753 return 0;
2754}
2755EXPORT_SYMBOL_GPL(skb_gro_receive);
2756
Linus Torvalds1da177e2005-04-16 15:20:36 -07002757void __init skb_init(void)
2758{
2759 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2760 sizeof(struct sk_buff),
2761 0,
Alexey Dobriyane5d679f332006-08-26 19:25:52 -07002762 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
Paul Mundt20c2df82007-07-20 10:11:58 +09002763 NULL);
David S. Millerd179cd12005-08-17 14:57:30 -07002764 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2765 (2*sizeof(struct sk_buff)) +
2766 sizeof(atomic_t),
2767 0,
Alexey Dobriyane5d679f332006-08-26 19:25:52 -07002768 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
Paul Mundt20c2df82007-07-20 10:11:58 +09002769 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770}
2771
David Howells716ea3a2007-04-02 20:19:53 -07002772/**
2773 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2774 * @skb: Socket buffer containing the buffers to be mapped
2775 * @sg: The scatter-gather list to map into
2776 * @offset: The offset into the buffer's contents to start mapping
2777 * @len: Length of buffer space to be mapped
2778 *
2779 * Fill the specified scatter-gather list with mappings/pointers into a
2780 * region of the buffer space attached to a socket buffer.
2781 */
David S. Miller51c739d2007-10-30 21:29:29 -07002782static int
2783__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
David Howells716ea3a2007-04-02 20:19:53 -07002784{
David S. Miller1a028e52007-04-27 15:21:23 -07002785 int start = skb_headlen(skb);
2786 int i, copy = start - offset;
David S. Millerfbb398a2009-06-09 00:18:59 -07002787 struct sk_buff *frag_iter;
David Howells716ea3a2007-04-02 20:19:53 -07002788 int elt = 0;
2789
2790 if (copy > 0) {
2791 if (copy > len)
2792 copy = len;
Jens Axboe642f1492007-10-24 11:20:47 +02002793 sg_set_buf(sg, skb->data + offset, copy);
David Howells716ea3a2007-04-02 20:19:53 -07002794 elt++;
2795 if ((len -= copy) == 0)
2796 return elt;
2797 offset += copy;
2798 }
2799
2800 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07002801 int end;
David Howells716ea3a2007-04-02 20:19:53 -07002802
Ilpo Järvinen547b7922008-07-25 21:43:18 -07002803 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07002804
2805 end = start + skb_shinfo(skb)->frags[i].size;
David Howells716ea3a2007-04-02 20:19:53 -07002806 if ((copy = end - offset) > 0) {
2807 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2808
2809 if (copy > len)
2810 copy = len;
Jens Axboe642f1492007-10-24 11:20:47 +02002811 sg_set_page(&sg[elt], frag->page, copy,
2812 frag->page_offset+offset-start);
David Howells716ea3a2007-04-02 20:19:53 -07002813 elt++;
2814 if (!(len -= copy))
2815 return elt;
2816 offset += copy;
2817 }
David S. Miller1a028e52007-04-27 15:21:23 -07002818 start = end;
David Howells716ea3a2007-04-02 20:19:53 -07002819 }
2820
David S. Millerfbb398a2009-06-09 00:18:59 -07002821 skb_walk_frags(skb, frag_iter) {
2822 int end;
David Howells716ea3a2007-04-02 20:19:53 -07002823
David S. Millerfbb398a2009-06-09 00:18:59 -07002824 WARN_ON(start > offset + len);
David Howells716ea3a2007-04-02 20:19:53 -07002825
David S. Millerfbb398a2009-06-09 00:18:59 -07002826 end = start + frag_iter->len;
2827 if ((copy = end - offset) > 0) {
2828 if (copy > len)
2829 copy = len;
2830 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
2831 copy);
2832 if ((len -= copy) == 0)
2833 return elt;
2834 offset += copy;
David Howells716ea3a2007-04-02 20:19:53 -07002835 }
David S. Millerfbb398a2009-06-09 00:18:59 -07002836 start = end;
David Howells716ea3a2007-04-02 20:19:53 -07002837 }
2838 BUG_ON(len);
2839 return elt;
2840}
2841
David S. Miller51c739d2007-10-30 21:29:29 -07002842int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2843{
2844 int nsg = __skb_to_sgvec(skb, sg, offset, len);
2845
Jens Axboec46f2332007-10-31 12:06:37 +01002846 sg_mark_end(&sg[nsg - 1]);
David S. Miller51c739d2007-10-30 21:29:29 -07002847
2848 return nsg;
2849}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002850EXPORT_SYMBOL_GPL(skb_to_sgvec);
David S. Miller51c739d2007-10-30 21:29:29 -07002851
David Howells716ea3a2007-04-02 20:19:53 -07002852/**
2853 * skb_cow_data - Check that a socket buffer's data buffers are writable
2854 * @skb: The socket buffer to check.
2855 * @tailbits: Amount of trailing space to be added
2856 * @trailer: Returned pointer to the skb where the @tailbits space begins
2857 *
2858 * Make sure that the data buffers attached to a socket buffer are
2859 * writable. If they are not, private copies are made of the data buffers
2860 * and the socket buffer is set to use these instead.
2861 *
2862 * If @tailbits is given, make sure that there is space to write @tailbits
2863 * bytes of data beyond current end of socket buffer. @trailer will be
2864 * set to point to the skb in which this space begins.
2865 *
2866 * The number of scatterlist elements required to completely map the
2867 * COW'd and extended socket buffer will be returned.
2868 */
2869int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
2870{
2871 int copyflag;
2872 int elt;
2873 struct sk_buff *skb1, **skb_p;
2874
2875 /* If skb is cloned or its head is paged, reallocate
2876 * head pulling out all the pages (pages are considered not writable
2877 * at the moment even if they are anonymous).
2878 */
2879 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
2880 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
2881 return -ENOMEM;
2882
2883 /* Easy case. Most of packets will go this way. */
David S. Miller21dc3302010-08-23 00:13:46 -07002884 if (!skb_has_frag_list(skb)) {
David Howells716ea3a2007-04-02 20:19:53 -07002885 /* A little of trouble, not enough of space for trailer.
2886 * This should not happen, when stack is tuned to generate
2887 * good frames. OK, on miss we reallocate and reserve even more
2888 * space, 128 bytes is fair. */
2889
2890 if (skb_tailroom(skb) < tailbits &&
2891 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
2892 return -ENOMEM;
2893
2894 /* Voila! */
2895 *trailer = skb;
2896 return 1;
2897 }
2898
2899 /* Misery. We are in troubles, going to mincer fragments... */
2900
2901 elt = 1;
2902 skb_p = &skb_shinfo(skb)->frag_list;
2903 copyflag = 0;
2904
2905 while ((skb1 = *skb_p) != NULL) {
2906 int ntail = 0;
2907
2908 /* The fragment is partially pulled by someone,
2909 * this can happen on input. Copy it and everything
2910 * after it. */
2911
2912 if (skb_shared(skb1))
2913 copyflag = 1;
2914
2915 /* If the skb is the last, worry about trailer. */
2916
2917 if (skb1->next == NULL && tailbits) {
2918 if (skb_shinfo(skb1)->nr_frags ||
David S. Miller21dc3302010-08-23 00:13:46 -07002919 skb_has_frag_list(skb1) ||
David Howells716ea3a2007-04-02 20:19:53 -07002920 skb_tailroom(skb1) < tailbits)
2921 ntail = tailbits + 128;
2922 }
2923
2924 if (copyflag ||
2925 skb_cloned(skb1) ||
2926 ntail ||
2927 skb_shinfo(skb1)->nr_frags ||
David S. Miller21dc3302010-08-23 00:13:46 -07002928 skb_has_frag_list(skb1)) {
David Howells716ea3a2007-04-02 20:19:53 -07002929 struct sk_buff *skb2;
2930
2931 /* Fuck, we are miserable poor guys... */
2932 if (ntail == 0)
2933 skb2 = skb_copy(skb1, GFP_ATOMIC);
2934 else
2935 skb2 = skb_copy_expand(skb1,
2936 skb_headroom(skb1),
2937 ntail,
2938 GFP_ATOMIC);
2939 if (unlikely(skb2 == NULL))
2940 return -ENOMEM;
2941
2942 if (skb1->sk)
2943 skb_set_owner_w(skb2, skb1->sk);
2944
2945 /* Looking around. Are we still alive?
2946 * OK, link new skb, drop old one */
2947
2948 skb2->next = skb1->next;
2949 *skb_p = skb2;
2950 kfree_skb(skb1);
2951 skb1 = skb2;
2952 }
2953 elt++;
2954 *trailer = skb1;
2955 skb_p = &skb1->next;
2956 }
2957
2958 return elt;
2959}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002960EXPORT_SYMBOL_GPL(skb_cow_data);
David Howells716ea3a2007-04-02 20:19:53 -07002961
Eric Dumazetb1faf562010-05-31 23:44:05 -07002962static void sock_rmem_free(struct sk_buff *skb)
2963{
2964 struct sock *sk = skb->sk;
2965
2966 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
2967}
2968
2969/*
2970 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
2971 */
2972int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
2973{
2974 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
2975 (unsigned)sk->sk_rcvbuf)
2976 return -ENOMEM;
2977
2978 skb_orphan(skb);
2979 skb->sk = sk;
2980 skb->destructor = sock_rmem_free;
2981 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2982
2983 skb_queue_tail(&sk->sk_error_queue, skb);
2984 if (!sock_flag(sk, SOCK_DEAD))
2985 sk->sk_data_ready(sk, skb->len);
2986 return 0;
2987}
2988EXPORT_SYMBOL(sock_queue_err_skb);
2989
Patrick Ohlyac45f602009-02-12 05:03:37 +00002990void skb_tstamp_tx(struct sk_buff *orig_skb,
2991 struct skb_shared_hwtstamps *hwtstamps)
2992{
2993 struct sock *sk = orig_skb->sk;
2994 struct sock_exterr_skb *serr;
2995 struct sk_buff *skb;
2996 int err;
2997
2998 if (!sk)
2999 return;
3000
3001 skb = skb_clone(orig_skb, GFP_ATOMIC);
3002 if (!skb)
3003 return;
3004
3005 if (hwtstamps) {
3006 *skb_hwtstamps(skb) =
3007 *hwtstamps;
3008 } else {
3009 /*
3010 * no hardware time stamps available,
Oliver Hartkopp2244d072010-08-17 08:59:14 +00003011 * so keep the shared tx_flags and only
Patrick Ohlyac45f602009-02-12 05:03:37 +00003012 * store software time stamp
3013 */
3014 skb->tstamp = ktime_get_real();
3015 }
3016
3017 serr = SKB_EXT_ERR(skb);
3018 memset(serr, 0, sizeof(*serr));
3019 serr->ee.ee_errno = ENOMSG;
3020 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
Eric Dumazet29030372010-05-29 00:20:48 -07003021
Patrick Ohlyac45f602009-02-12 05:03:37 +00003022 err = sock_queue_err_skb(sk, skb);
Eric Dumazet29030372010-05-29 00:20:48 -07003023
Patrick Ohlyac45f602009-02-12 05:03:37 +00003024 if (err)
3025 kfree_skb(skb);
3026}
3027EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3028
3029
Rusty Russellf35d9d82008-02-04 23:49:54 -05003030/**
3031 * skb_partial_csum_set - set up and verify partial csum values for packet
3032 * @skb: the skb to set
3033 * @start: the number of bytes after skb->data to start checksumming.
3034 * @off: the offset from start to place the checksum.
3035 *
3036 * For untrusted partially-checksummed packets, we need to make sure the values
3037 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3038 *
3039 * This function checks and sets those values and skb->ip_summed: if this
3040 * returns false you should drop the packet.
3041 */
3042bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3043{
Herbert Xu5ff8dda2009-06-04 01:22:01 +00003044 if (unlikely(start > skb_headlen(skb)) ||
3045 unlikely((int)start + off > skb_headlen(skb) - 2)) {
Rusty Russellf35d9d82008-02-04 23:49:54 -05003046 if (net_ratelimit())
3047 printk(KERN_WARNING
3048 "bad partial csum: csum=%u/%u len=%u\n",
Herbert Xu5ff8dda2009-06-04 01:22:01 +00003049 start, off, skb_headlen(skb));
Rusty Russellf35d9d82008-02-04 23:49:54 -05003050 return false;
3051 }
3052 skb->ip_summed = CHECKSUM_PARTIAL;
3053 skb->csum_start = skb_headroom(skb) + start;
3054 skb->csum_offset = off;
3055 return true;
3056}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08003057EXPORT_SYMBOL_GPL(skb_partial_csum_set);
Rusty Russellf35d9d82008-02-04 23:49:54 -05003058
Ben Hutchings4497b072008-06-19 16:22:28 -07003059void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3060{
3061 if (net_ratelimit())
3062 pr_warning("%s: received packets cannot be forwarded"
3063 " while LRO is enabled\n", skb->dev->name);
3064}
Ben Hutchings4497b072008-06-19 16:22:28 -07003065EXPORT_SYMBOL(__skb_warn_lro_forwarding);