blob: a690cae91cdd14641b61844f6ac1f903fbf9666e [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 Torvalds268bb0c2011-05-20 12:50:29 -070060#include <linux/prefetch.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
62#include <net/protocol.h>
63#include <net/dst.h>
64#include <net/sock.h>
65#include <net/checksum.h>
66#include <net/xfrm.h>
67
68#include <asm/uaccess.h>
69#include <asm/system.h>
Steven Rostedtad8d75f2009-04-14 19:39:12 -040070#include <trace/events/skb.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070071
Al Viroa1f8e7f72006-10-19 16:08:53 -040072#include "kmap_skb.h"
73
Christoph Lametere18b8902006-12-06 20:33:20 -080074static struct kmem_cache *skbuff_head_cache __read_mostly;
75static struct kmem_cache *skbuff_fclone_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070076
Jens Axboe9c55e012007-11-06 23:30:13 -080077static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
78 struct pipe_buffer *buf)
79{
Jarek Poplawski8b9d3722009-01-19 17:03:56 -080080 put_page(buf->page);
Jens Axboe9c55e012007-11-06 23:30:13 -080081}
82
83static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
84 struct pipe_buffer *buf)
85{
Jarek Poplawski8b9d3722009-01-19 17:03:56 -080086 get_page(buf->page);
Jens Axboe9c55e012007-11-06 23:30:13 -080087}
88
89static int sock_pipe_buf_steal(struct pipe_inode_info *pipe,
90 struct pipe_buffer *buf)
91{
92 return 1;
93}
94
95
96/* Pipe buffer operations for a socket. */
Alexey Dobriyan28dfef82009-12-15 16:46:48 -080097static const struct pipe_buf_operations sock_pipe_buf_ops = {
Jens Axboe9c55e012007-11-06 23:30:13 -080098 .can_merge = 0,
99 .map = generic_pipe_buf_map,
100 .unmap = generic_pipe_buf_unmap,
101 .confirm = generic_pipe_buf_confirm,
102 .release = sock_pipe_buf_release,
103 .steal = sock_pipe_buf_steal,
104 .get = sock_pipe_buf_get,
105};
106
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107/*
108 * Keep out-of-line to prevent kernel bloat.
109 * __builtin_return_address is not used because it is not always
110 * reliable.
111 */
112
113/**
114 * skb_over_panic - private function
115 * @skb: buffer
116 * @sz: size
117 * @here: address
118 *
119 * Out of line support code for skb_put(). Not user callable.
120 */
Rami Rosenccb7c772010-04-20 22:39:53 -0700121static void skb_over_panic(struct sk_buff *skb, int sz, void *here)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122{
Patrick McHardy26095452005-04-21 16:43:02 -0700123 printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700124 "data:%p tail:%#lx end:%#lx dev:%s\n",
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700125 here, skb->len, sz, skb->head, skb->data,
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700126 (unsigned long)skb->tail, (unsigned long)skb->end,
Patrick McHardy26095452005-04-21 16:43:02 -0700127 skb->dev ? skb->dev->name : "<NULL>");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 BUG();
129}
130
131/**
132 * skb_under_panic - private function
133 * @skb: buffer
134 * @sz: size
135 * @here: address
136 *
137 * Out of line support code for skb_push(). Not user callable.
138 */
139
Rami Rosenccb7c772010-04-20 22:39:53 -0700140static void skb_under_panic(struct sk_buff *skb, int sz, void *here)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141{
Patrick McHardy26095452005-04-21 16:43:02 -0700142 printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700143 "data:%p tail:%#lx end:%#lx dev:%s\n",
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700144 here, skb->len, sz, skb->head, skb->data,
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700145 (unsigned long)skb->tail, (unsigned long)skb->end,
Patrick McHardy26095452005-04-21 16:43:02 -0700146 skb->dev ? skb->dev->name : "<NULL>");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 BUG();
148}
149
150/* Allocate a new skbuff. We do this ourselves so we can fill in a few
151 * 'private' fields and also do memory statistics to find all the
152 * [BEEP] leaks.
153 *
154 */
155
156/**
David S. Millerd179cd12005-08-17 14:57:30 -0700157 * __alloc_skb - allocate a network buffer
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 * @size: size to allocate
159 * @gfp_mask: allocation mask
Randy Dunlapc83c2482005-10-18 22:07:41 -0700160 * @fclone: allocate from fclone cache instead of head cache
161 * and allocate a cloned (child) skb
Christoph Hellwigb30973f2006-12-06 20:32:36 -0800162 * @node: numa node to allocate memory on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 *
164 * Allocate a new &sk_buff. The returned buffer has no headroom and a
165 * tail room of size bytes. The object has a reference count of one.
166 * The return is the buffer. On a failure the return is %NULL.
167 *
168 * Buffers may only be allocated from interrupts using a @gfp_mask of
169 * %GFP_ATOMIC.
170 */
Al Virodd0fc662005-10-07 07:46:04 +0100171struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
Christoph Hellwigb30973f2006-12-06 20:32:36 -0800172 int fclone, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173{
Christoph Lametere18b8902006-12-06 20:33:20 -0800174 struct kmem_cache *cache;
Benjamin LaHaise4947d3e2006-01-03 14:06:50 -0800175 struct skb_shared_info *shinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 struct sk_buff *skb;
177 u8 *data;
178
Herbert Xu8798b3f2006-01-23 16:32:45 -0800179 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
180
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 /* Get the HEAD */
Christoph Hellwigb30973f2006-12-06 20:32:36 -0800182 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 if (!skb)
184 goto out;
Eric Dumazetec7d2f22010-05-05 01:07:37 -0700185 prefetchw(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186
Eric Dumazet87fb4b72011-10-13 07:28:54 +0000187 /* We do our best to align skb_shared_info on a separate cache
188 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
189 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
190 * Both skb->head and skb_shared_info are cache line aligned.
191 */
Tony Lindgrenbc417e32011-11-02 13:40:28 +0000192 size = SKB_DATA_ALIGN(size);
Eric Dumazet87fb4b72011-10-13 07:28:54 +0000193 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
194 data = kmalloc_node_track_caller(size, gfp_mask, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 if (!data)
196 goto nodata;
Eric Dumazet87fb4b72011-10-13 07:28:54 +0000197 /* kmalloc(size) might give us more room than requested.
198 * Put skb_shared_info exactly at the end of allocated zone,
199 * to allow max possible filling before reallocation.
200 */
201 size = SKB_WITH_OVERHEAD(ksize(data));
Eric Dumazetec7d2f22010-05-05 01:07:37 -0700202 prefetchw(data + size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203
Arnaldo Carvalho de Meloca0605a2007-03-19 10:48:59 -0300204 /*
Johannes Bergc8005782008-05-03 20:56:42 -0700205 * Only clear those fields we need to clear, not those that we will
206 * actually initialise below. Hence, don't put any more fields after
207 * the tail pointer in struct sk_buff!
Arnaldo Carvalho de Meloca0605a2007-03-19 10:48:59 -0300208 */
209 memset(skb, 0, offsetof(struct sk_buff, tail));
Eric Dumazet87fb4b72011-10-13 07:28:54 +0000210 /* Account for allocated memory : skb + skb->head */
211 skb->truesize = SKB_TRUESIZE(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 atomic_set(&skb->users, 1);
213 skb->head = data;
214 skb->data = data;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700215 skb_reset_tail_pointer(skb);
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700216 skb->end = skb->tail + size;
Stephen Hemminger19633e12009-06-17 05:23:27 +0000217#ifdef NET_SKBUFF_DATA_USES_OFFSET
218 skb->mac_header = ~0U;
219#endif
220
Benjamin LaHaise4947d3e2006-01-03 14:06:50 -0800221 /* make sure we initialize shinfo sequentially */
222 shinfo = skb_shinfo(skb);
Eric Dumazetec7d2f22010-05-05 01:07:37 -0700223 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
Benjamin LaHaise4947d3e2006-01-03 14:06:50 -0800224 atomic_set(&shinfo->dataref, 1);
Eric Dumazetc2aa3662011-01-25 23:18:38 +0000225 kmemcheck_annotate_variable(shinfo->destructor_arg);
Benjamin LaHaise4947d3e2006-01-03 14:06:50 -0800226
David S. Millerd179cd12005-08-17 14:57:30 -0700227 if (fclone) {
228 struct sk_buff *child = skb + 1;
229 atomic_t *fclone_ref = (atomic_t *) (child + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200231 kmemcheck_annotate_bitfield(child, flags1);
232 kmemcheck_annotate_bitfield(child, flags2);
David S. Millerd179cd12005-08-17 14:57:30 -0700233 skb->fclone = SKB_FCLONE_ORIG;
234 atomic_set(fclone_ref, 1);
235
236 child->fclone = SKB_FCLONE_UNAVAILABLE;
237 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238out:
239 return skb;
240nodata:
Herbert Xu8798b3f2006-01-23 16:32:45 -0800241 kmem_cache_free(cache, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 skb = NULL;
243 goto out;
244}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800245EXPORT_SYMBOL(__alloc_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246
247/**
Eric Dumazetb2b5ce92011-11-14 06:03:34 +0000248 * build_skb - build a network buffer
249 * @data: data buffer provided by caller
250 *
251 * Allocate a new &sk_buff. Caller provides space holding head and
252 * skb_shared_info. @data must have been allocated by kmalloc()
253 * The return is the new skb buffer.
254 * On a failure the return is %NULL, and @data is not freed.
255 * Notes :
256 * Before IO, driver allocates only data buffer where NIC put incoming frame
257 * Driver should add room at head (NET_SKB_PAD) and
258 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
259 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
260 * before giving packet to stack.
261 * RX rings only contains data buffers, not full skbs.
262 */
263struct sk_buff *build_skb(void *data)
264{
265 struct skb_shared_info *shinfo;
266 struct sk_buff *skb;
267 unsigned int size;
268
269 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
270 if (!skb)
271 return NULL;
272
273 size = ksize(data) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
274
275 memset(skb, 0, offsetof(struct sk_buff, tail));
276 skb->truesize = SKB_TRUESIZE(size);
277 atomic_set(&skb->users, 1);
278 skb->head = data;
279 skb->data = data;
280 skb_reset_tail_pointer(skb);
281 skb->end = skb->tail + size;
282#ifdef NET_SKBUFF_DATA_USES_OFFSET
283 skb->mac_header = ~0U;
284#endif
285
286 /* make sure we initialize shinfo sequentially */
287 shinfo = skb_shinfo(skb);
288 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
289 atomic_set(&shinfo->dataref, 1);
290 kmemcheck_annotate_variable(shinfo->destructor_arg);
291
292 return skb;
293}
294EXPORT_SYMBOL(build_skb);
295
296/**
Christoph Hellwig8af27452006-07-31 22:35:23 -0700297 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
298 * @dev: network device to receive on
299 * @length: length to allocate
300 * @gfp_mask: get_free_pages mask, passed to alloc_skb
301 *
302 * Allocate a new &sk_buff and assign it a usage count of one. The
303 * buffer has unspecified headroom built in. Users should allocate
304 * the headroom they think they need without accounting for the
305 * built in space. The built in space is used for optimisations.
306 *
307 * %NULL is returned if there is no free memory.
308 */
309struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
310 unsigned int length, gfp_t gfp_mask)
311{
312 struct sk_buff *skb;
313
Eric Dumazet564824b2010-10-11 19:05:25 +0000314 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, NUMA_NO_NODE);
Christoph Hellwig7b2e4972006-08-07 16:09:04 -0700315 if (likely(skb)) {
Christoph Hellwig8af27452006-07-31 22:35:23 -0700316 skb_reserve(skb, NET_SKB_PAD);
Christoph Hellwig7b2e4972006-08-07 16:09:04 -0700317 skb->dev = dev;
318 }
Christoph Hellwig8af27452006-07-31 22:35:23 -0700319 return skb;
320}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800321EXPORT_SYMBOL(__netdev_alloc_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322
Peter Zijlstra654bed12008-10-07 14:22:33 -0700323void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
Eric Dumazet50269e12012-03-23 23:59:33 +0000324 int size, unsigned int truesize)
Peter Zijlstra654bed12008-10-07 14:22:33 -0700325{
326 skb_fill_page_desc(skb, i, page, off, size);
327 skb->len += size;
328 skb->data_len += size;
Eric Dumazet50269e12012-03-23 23:59:33 +0000329 skb->truesize += truesize;
Peter Zijlstra654bed12008-10-07 14:22:33 -0700330}
331EXPORT_SYMBOL(skb_add_rx_frag);
332
Ilpo Järvinenf58518e2008-03-27 17:51:31 -0700333/**
334 * dev_alloc_skb - allocate an skbuff for receiving
335 * @length: length to allocate
336 *
337 * Allocate a new &sk_buff and assign it a usage count of one. The
338 * buffer has unspecified headroom built in. Users should allocate
339 * the headroom they think they need without accounting for the
340 * built in space. The built in space is used for optimisations.
341 *
342 * %NULL is returned if there is no free memory. Although this function
343 * allocates memory it can be called from an interrupt.
344 */
345struct sk_buff *dev_alloc_skb(unsigned int length)
346{
Denys Vlasenko1483b872008-03-28 15:57:39 -0700347 /*
348 * There is more code here than it seems:
David S. Millera0f55e02008-03-28 18:22:32 -0700349 * __dev_alloc_skb is an inline
Denys Vlasenko1483b872008-03-28 15:57:39 -0700350 */
Ilpo Järvinenf58518e2008-03-27 17:51:31 -0700351 return __dev_alloc_skb(length, GFP_ATOMIC);
352}
353EXPORT_SYMBOL(dev_alloc_skb);
354
Herbert Xu27b437c2006-07-13 19:26:39 -0700355static void skb_drop_list(struct sk_buff **listp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356{
Herbert Xu27b437c2006-07-13 19:26:39 -0700357 struct sk_buff *list = *listp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358
Herbert Xu27b437c2006-07-13 19:26:39 -0700359 *listp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360
361 do {
362 struct sk_buff *this = list;
363 list = list->next;
364 kfree_skb(this);
365 } while (list);
366}
367
Herbert Xu27b437c2006-07-13 19:26:39 -0700368static inline void skb_drop_fraglist(struct sk_buff *skb)
369{
370 skb_drop_list(&skb_shinfo(skb)->frag_list);
371}
372
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373static void skb_clone_fraglist(struct sk_buff *skb)
374{
375 struct sk_buff *list;
376
David S. Millerfbb398a2009-06-09 00:18:59 -0700377 skb_walk_frags(skb, list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 skb_get(list);
379}
380
Adrian Bunk5bba1712006-06-29 13:02:35 -0700381static void skb_release_data(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382{
383 if (!skb->cloned ||
384 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
385 &skb_shinfo(skb)->dataref)) {
386 if (skb_shinfo(skb)->nr_frags) {
387 int i;
388 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
Ian Campbellea2ab692011-08-22 23:44:58 +0000389 skb_frag_unref(skb, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 }
391
Shirley Maa6686f22011-07-06 12:22:12 +0000392 /*
393 * If skb buf is from userspace, we need to notify the caller
394 * the lower device DMA has done;
395 */
396 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
397 struct ubuf_info *uarg;
398
399 uarg = skb_shinfo(skb)->destructor_arg;
400 if (uarg->callback)
401 uarg->callback(uarg);
402 }
403
David S. Miller21dc3302010-08-23 00:13:46 -0700404 if (skb_has_frag_list(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 skb_drop_fraglist(skb);
406
407 kfree(skb->head);
408 }
409}
410
411/*
412 * Free an skbuff by memory without cleaning the state.
413 */
Herbert Xu2d4baff2007-11-26 23:11:19 +0800414static void kfree_skbmem(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415{
David S. Millerd179cd12005-08-17 14:57:30 -0700416 struct sk_buff *other;
417 atomic_t *fclone_ref;
418
David S. Millerd179cd12005-08-17 14:57:30 -0700419 switch (skb->fclone) {
420 case SKB_FCLONE_UNAVAILABLE:
421 kmem_cache_free(skbuff_head_cache, skb);
422 break;
423
424 case SKB_FCLONE_ORIG:
425 fclone_ref = (atomic_t *) (skb + 2);
426 if (atomic_dec_and_test(fclone_ref))
427 kmem_cache_free(skbuff_fclone_cache, skb);
428 break;
429
430 case SKB_FCLONE_CLONE:
431 fclone_ref = (atomic_t *) (skb + 1);
432 other = skb - 1;
433
434 /* The clone portion is available for
435 * fast-cloning again.
436 */
437 skb->fclone = SKB_FCLONE_UNAVAILABLE;
438
439 if (atomic_dec_and_test(fclone_ref))
440 kmem_cache_free(skbuff_fclone_cache, other);
441 break;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -0700442 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443}
444
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700445static void skb_release_head_state(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446{
Eric Dumazetadf30902009-06-02 05:19:30 +0000447 skb_dst_drop(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448#ifdef CONFIG_XFRM
449 secpath_put(skb->sp);
450#endif
Stephen Hemminger9c2b3322005-04-19 22:39:42 -0700451 if (skb->destructor) {
452 WARN_ON(in_irq());
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 skb->destructor(skb);
454 }
Igor Maravića3bf7ae2011-12-12 02:58:22 +0000455#if IS_ENABLED(CONFIG_NF_CONNTRACK)
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -0700456 nf_conntrack_put(skb->nfct);
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +0100457#endif
458#ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
Yasuyuki Kozakai9fb9cbb2005-11-09 16:38:16 -0800459 nf_conntrack_put_reasm(skb->nfct_reasm);
460#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461#ifdef CONFIG_BRIDGE_NETFILTER
462 nf_bridge_put(skb->nf_bridge);
463#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464/* XXX: IS this still necessary? - JHS */
465#ifdef CONFIG_NET_SCHED
466 skb->tc_index = 0;
467#ifdef CONFIG_NET_CLS_ACT
468 skb->tc_verd = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469#endif
470#endif
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700471}
472
473/* Free everything but the sk_buff shell. */
474static void skb_release_all(struct sk_buff *skb)
475{
476 skb_release_head_state(skb);
Herbert Xu2d4baff2007-11-26 23:11:19 +0800477 skb_release_data(skb);
478}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479
Herbert Xu2d4baff2007-11-26 23:11:19 +0800480/**
481 * __kfree_skb - private function
482 * @skb: buffer
483 *
484 * Free an sk_buff. Release anything attached to the buffer.
485 * Clean the state. This is an internal helper function. Users should
486 * always call kfree_skb
487 */
488
489void __kfree_skb(struct sk_buff *skb)
490{
491 skb_release_all(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 kfree_skbmem(skb);
493}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800494EXPORT_SYMBOL(__kfree_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495
496/**
Jörn Engel231d06a2006-03-20 21:28:35 -0800497 * kfree_skb - free an sk_buff
498 * @skb: buffer to free
499 *
500 * Drop a reference to the buffer and free it if the usage count has
501 * hit zero.
502 */
503void kfree_skb(struct sk_buff *skb)
504{
505 if (unlikely(!skb))
506 return;
507 if (likely(atomic_read(&skb->users) == 1))
508 smp_rmb();
509 else if (likely(!atomic_dec_and_test(&skb->users)))
510 return;
Neil Hormanead2ceb2009-03-11 09:49:55 +0000511 trace_kfree_skb(skb, __builtin_return_address(0));
Jörn Engel231d06a2006-03-20 21:28:35 -0800512 __kfree_skb(skb);
513}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800514EXPORT_SYMBOL(kfree_skb);
Jörn Engel231d06a2006-03-20 21:28:35 -0800515
Stephen Hemmingerd1a203e2008-11-01 21:01:09 -0700516/**
Neil Hormanead2ceb2009-03-11 09:49:55 +0000517 * consume_skb - free an skbuff
518 * @skb: buffer to free
519 *
520 * Drop a ref to the buffer and free it if the usage count has hit zero
521 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
522 * is being dropped after a failure and notes that
523 */
524void consume_skb(struct sk_buff *skb)
525{
526 if (unlikely(!skb))
527 return;
528 if (likely(atomic_read(&skb->users) == 1))
529 smp_rmb();
530 else if (likely(!atomic_dec_and_test(&skb->users)))
531 return;
Koki Sanagi07dc22e2010-08-23 18:46:12 +0900532 trace_consume_skb(skb);
Neil Hormanead2ceb2009-03-11 09:49:55 +0000533 __kfree_skb(skb);
534}
535EXPORT_SYMBOL(consume_skb);
536
537/**
Andy Fleming3d153a72011-10-13 04:33:54 +0000538 * skb_recycle - clean up an skb for reuse
539 * @skb: buffer
540 *
541 * Recycles the skb to be reused as a receive buffer. This
542 * function does any necessary reference count dropping, and
543 * cleans up the skbuff as if it just came from __alloc_skb().
544 */
545void skb_recycle(struct sk_buff *skb)
546{
547 struct skb_shared_info *shinfo;
548
549 skb_release_head_state(skb);
550
551 shinfo = skb_shinfo(skb);
552 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
553 atomic_set(&shinfo->dataref, 1);
554
555 memset(skb, 0, offsetof(struct sk_buff, tail));
556 skb->data = skb->head + NET_SKB_PAD;
557 skb_reset_tail_pointer(skb);
558}
559EXPORT_SYMBOL(skb_recycle);
560
561/**
Stephen Hemmingerd1a203e2008-11-01 21:01:09 -0700562 * skb_recycle_check - check if skb can be reused for receive
563 * @skb: buffer
564 * @skb_size: minimum receive buffer size
565 *
566 * Checks that the skb passed in is not shared or cloned, and
567 * that it is linear and its head portion at least as large as
568 * skb_size so that it can be recycled as a receive buffer.
569 * If these conditions are met, this function does any necessary
570 * reference count dropping and cleans up the skbuff as if it
571 * just came from __alloc_skb().
572 */
Changli Gao5b0daa32010-05-29 00:12:13 -0700573bool skb_recycle_check(struct sk_buff *skb, int skb_size)
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700574{
Andy Fleming3d153a72011-10-13 04:33:54 +0000575 if (!skb_is_recycleable(skb, skb_size))
Changli Gao5b0daa32010-05-29 00:12:13 -0700576 return false;
Anton Vorontsove84af6d2009-11-10 14:11:01 +0000577
Andy Fleming3d153a72011-10-13 04:33:54 +0000578 skb_recycle(skb);
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700579
Changli Gao5b0daa32010-05-29 00:12:13 -0700580 return true;
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700581}
582EXPORT_SYMBOL(skb_recycle_check);
583
Herbert Xudec18812007-10-14 00:37:30 -0700584static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
585{
586 new->tstamp = old->tstamp;
587 new->dev = old->dev;
588 new->transport_header = old->transport_header;
589 new->network_header = old->network_header;
590 new->mac_header = old->mac_header;
Eric Dumazet7fee2262010-05-11 23:19:48 +0000591 skb_dst_copy(new, old);
Tom Herbert0a9627f2010-03-16 08:03:29 +0000592 new->rxhash = old->rxhash;
Changli Gao6461be32011-08-19 04:44:18 +0000593 new->ooo_okay = old->ooo_okay;
Tom Herbertbdeab992011-08-14 19:45:55 +0000594 new->l4_rxhash = old->l4_rxhash;
Ben Greear3bdc0eb2012-02-11 15:39:30 +0000595 new->no_fcs = old->no_fcs;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -0700596#ifdef CONFIG_XFRM
Herbert Xudec18812007-10-14 00:37:30 -0700597 new->sp = secpath_get(old->sp);
598#endif
599 memcpy(new->cb, old->cb, sizeof(old->cb));
Herbert Xu9bcb97c2009-05-22 22:20:02 +0000600 new->csum = old->csum;
Herbert Xudec18812007-10-14 00:37:30 -0700601 new->local_df = old->local_df;
602 new->pkt_type = old->pkt_type;
603 new->ip_summed = old->ip_summed;
604 skb_copy_queue_mapping(new, old);
605 new->priority = old->priority;
Igor Maravića3bf7ae2011-12-12 02:58:22 +0000606#if IS_ENABLED(CONFIG_IP_VS)
Herbert Xudec18812007-10-14 00:37:30 -0700607 new->ipvs_property = old->ipvs_property;
608#endif
609 new->protocol = old->protocol;
610 new->mark = old->mark;
Eric Dumazet8964be42009-11-20 15:35:04 -0800611 new->skb_iif = old->skb_iif;
Herbert Xudec18812007-10-14 00:37:30 -0700612 __nf_copy(new, old);
Igor Maravića3bf7ae2011-12-12 02:58:22 +0000613#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
Herbert Xudec18812007-10-14 00:37:30 -0700614 new->nf_trace = old->nf_trace;
615#endif
616#ifdef CONFIG_NET_SCHED
617 new->tc_index = old->tc_index;
618#ifdef CONFIG_NET_CLS_ACT
619 new->tc_verd = old->tc_verd;
620#endif
621#endif
Patrick McHardy6aa895b02008-07-14 22:49:06 -0700622 new->vlan_tci = old->vlan_tci;
623
Herbert Xudec18812007-10-14 00:37:30 -0700624 skb_copy_secmark(new, old);
625}
626
Herbert Xu82c49a32009-05-22 22:11:37 +0000627/*
628 * You should not add any new code to this function. Add it to
629 * __copy_skb_header above instead.
630 */
Herbert Xue0053ec2007-10-14 00:37:52 -0700631static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633#define C(x) n->x = skb->x
634
635 n->next = n->prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 n->sk = NULL;
Herbert Xudec18812007-10-14 00:37:30 -0700637 __copy_skb_header(n, skb);
638
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 C(len);
640 C(data_len);
Alexey Dobriyan3e6b3b22007-03-16 15:00:46 -0700641 C(mac_len);
Patrick McHardy334a8132007-06-25 04:35:20 -0700642 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
Paul Moore02f1c892008-01-07 21:56:41 -0800643 n->cloned = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 n->nohdr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 n->destructor = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 C(tail);
647 C(end);
Paul Moore02f1c892008-01-07 21:56:41 -0800648 C(head);
649 C(data);
650 C(truesize);
651 atomic_set(&n->users, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
653 atomic_inc(&(skb_shinfo(skb)->dataref));
654 skb->cloned = 1;
655
656 return n;
Herbert Xue0053ec2007-10-14 00:37:52 -0700657#undef C
658}
659
660/**
661 * skb_morph - morph one skb into another
662 * @dst: the skb to receive the contents
663 * @src: the skb to supply the contents
664 *
665 * This is identical to skb_clone except that the target skb is
666 * supplied by the user.
667 *
668 * The target skb is returned upon exit.
669 */
670struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
671{
Herbert Xu2d4baff2007-11-26 23:11:19 +0800672 skb_release_all(dst);
Herbert Xue0053ec2007-10-14 00:37:52 -0700673 return __skb_clone(dst, src);
674}
675EXPORT_SYMBOL_GPL(skb_morph);
676
Michael S. Tsirkin48c83012011-08-31 08:03:29 +0000677/* skb_copy_ubufs - copy userspace skb frags buffers to kernel
678 * @skb: the skb to modify
679 * @gfp_mask: allocation priority
680 *
681 * This must be called on SKBTX_DEV_ZEROCOPY skb.
682 * It will copy all frags into kernel and drop the reference
683 * to userspace pages.
684 *
685 * If this function is called from an interrupt gfp_mask() must be
686 * %GFP_ATOMIC.
687 *
688 * Returns 0 on success or a negative error code on failure
689 * to allocate kernel memory to copy to.
690 */
691int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
Shirley Maa6686f22011-07-06 12:22:12 +0000692{
693 int i;
694 int num_frags = skb_shinfo(skb)->nr_frags;
695 struct page *page, *head = NULL;
696 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
697
698 for (i = 0; i < num_frags; i++) {
699 u8 *vaddr;
700 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
701
702 page = alloc_page(GFP_ATOMIC);
703 if (!page) {
704 while (head) {
705 struct page *next = (struct page *)head->private;
706 put_page(head);
707 head = next;
708 }
709 return -ENOMEM;
710 }
711 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
712 memcpy(page_address(page),
Eric Dumazet9e903e02011-10-18 21:00:24 +0000713 vaddr + f->page_offset, skb_frag_size(f));
Shirley Maa6686f22011-07-06 12:22:12 +0000714 kunmap_skb_frag(vaddr);
715 page->private = (unsigned long)head;
716 head = page;
717 }
718
719 /* skb frags release userspace buffers */
720 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
Ian Campbella8605c62011-10-19 23:01:49 +0000721 skb_frag_unref(skb, i);
Shirley Maa6686f22011-07-06 12:22:12 +0000722
723 uarg->callback(uarg);
724
725 /* skb frags point to kernel buffers */
726 for (i = skb_shinfo(skb)->nr_frags; i > 0; i--) {
Ian Campbella8605c62011-10-19 23:01:49 +0000727 __skb_fill_page_desc(skb, i-1, head, 0,
728 skb_shinfo(skb)->frags[i - 1].size);
Shirley Maa6686f22011-07-06 12:22:12 +0000729 head = (struct page *)head->private;
730 }
Michael S. Tsirkin48c83012011-08-31 08:03:29 +0000731
732 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
Shirley Maa6686f22011-07-06 12:22:12 +0000733 return 0;
734}
735
736
Herbert Xue0053ec2007-10-14 00:37:52 -0700737/**
738 * skb_clone - duplicate an sk_buff
739 * @skb: buffer to clone
740 * @gfp_mask: allocation priority
741 *
742 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
743 * copies share the same packet data but not structure. The new
744 * buffer has a reference count of 1. If the allocation fails the
745 * function returns %NULL otherwise the new buffer is returned.
746 *
747 * If this function is called from an interrupt gfp_mask() must be
748 * %GFP_ATOMIC.
749 */
750
751struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
752{
753 struct sk_buff *n;
754
Shirley Maa6686f22011-07-06 12:22:12 +0000755 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
756 if (skb_copy_ubufs(skb, gfp_mask))
757 return NULL;
758 }
759
Herbert Xue0053ec2007-10-14 00:37:52 -0700760 n = skb + 1;
761 if (skb->fclone == SKB_FCLONE_ORIG &&
762 n->fclone == SKB_FCLONE_UNAVAILABLE) {
763 atomic_t *fclone_ref = (atomic_t *) (n + 1);
764 n->fclone = SKB_FCLONE_CLONE;
765 atomic_inc(fclone_ref);
766 } else {
767 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
768 if (!n)
769 return NULL;
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200770
771 kmemcheck_annotate_bitfield(n, flags1);
772 kmemcheck_annotate_bitfield(n, flags2);
Herbert Xue0053ec2007-10-14 00:37:52 -0700773 n->fclone = SKB_FCLONE_UNAVAILABLE;
774 }
775
776 return __skb_clone(n, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800778EXPORT_SYMBOL(skb_clone);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779
780static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
781{
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700782#ifndef NET_SKBUFF_DATA_USES_OFFSET
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 /*
784 * Shift between the two data areas in bytes
785 */
786 unsigned long offset = new->data - old->data;
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700787#endif
Herbert Xudec18812007-10-14 00:37:30 -0700788
789 __copy_skb_header(new, old);
790
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700791#ifndef NET_SKBUFF_DATA_USES_OFFSET
792 /* {transport,network,mac}_header are relative to skb->head */
793 new->transport_header += offset;
794 new->network_header += offset;
Stephen Hemminger603a8bb2009-06-17 12:17:34 +0000795 if (skb_mac_header_was_set(new))
796 new->mac_header += offset;
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700797#endif
Herbert Xu79671682006-06-22 02:40:14 -0700798 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
799 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
800 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801}
802
803/**
804 * skb_copy - create private copy of an sk_buff
805 * @skb: buffer to copy
806 * @gfp_mask: allocation priority
807 *
808 * Make a copy of both an &sk_buff and its data. This is used when the
809 * caller wishes to modify the data and needs a private copy of the
810 * data to alter. Returns %NULL on failure or the pointer to the buffer
811 * on success. The returned buffer has a reference count of 1.
812 *
813 * As by-product this function converts non-linear &sk_buff to linear
814 * one, so that &sk_buff becomes completely private and caller is allowed
815 * to modify all the data of returned buffer. This means that this
816 * function is not recommended for use in circumstances when only
817 * header is going to be modified. Use pskb_copy() instead.
818 */
819
Al Virodd0fc662005-10-07 07:46:04 +0100820struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821{
Eric Dumazet6602ceb2010-09-01 05:25:10 +0000822 int headerlen = skb_headroom(skb);
823 unsigned int size = (skb_end_pointer(skb) - skb->head) + skb->data_len;
824 struct sk_buff *n = alloc_skb(size, gfp_mask);
825
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 if (!n)
827 return NULL;
828
829 /* Set the data pointer */
830 skb_reserve(n, headerlen);
831 /* Set the tail pointer and length */
832 skb_put(n, skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833
834 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
835 BUG();
836
837 copy_skb_header(n, skb);
838 return n;
839}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800840EXPORT_SYMBOL(skb_copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841
842/**
Eric Dumazet117632e2011-12-03 21:39:53 +0000843 * __pskb_copy - create copy of an sk_buff with private head.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 * @skb: buffer to copy
Eric Dumazet117632e2011-12-03 21:39:53 +0000845 * @headroom: headroom of new skb
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 * @gfp_mask: allocation priority
847 *
848 * Make a copy of both an &sk_buff and part of its data, located
849 * in header. Fragmented data remain shared. This is used when
850 * the caller wishes to modify only header of &sk_buff and needs
851 * private copy of the header to alter. Returns %NULL on failure
852 * or the pointer to the buffer on success.
853 * The returned buffer has a reference count of 1.
854 */
855
Eric Dumazet117632e2011-12-03 21:39:53 +0000856struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857{
Eric Dumazet117632e2011-12-03 21:39:53 +0000858 unsigned int size = skb_headlen(skb) + headroom;
Eric Dumazet6602ceb2010-09-01 05:25:10 +0000859 struct sk_buff *n = alloc_skb(size, gfp_mask);
860
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 if (!n)
862 goto out;
863
864 /* Set the data pointer */
Eric Dumazet117632e2011-12-03 21:39:53 +0000865 skb_reserve(n, headroom);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 /* Set the tail pointer and length */
867 skb_put(n, skb_headlen(skb));
868 /* Copy the bytes */
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -0300869 skb_copy_from_linear_data(skb, n->data, n->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870
Herbert Xu25f484a2006-11-07 14:57:15 -0800871 n->truesize += skb->data_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 n->data_len = skb->data_len;
873 n->len = skb->len;
874
875 if (skb_shinfo(skb)->nr_frags) {
876 int i;
877
Shirley Maa6686f22011-07-06 12:22:12 +0000878 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
879 if (skb_copy_ubufs(skb, gfp_mask)) {
Dan Carpenter15110222011-07-19 22:51:49 +0000880 kfree_skb(n);
881 n = NULL;
Shirley Maa6686f22011-07-06 12:22:12 +0000882 goto out;
883 }
884 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
886 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
Ian Campbellea2ab692011-08-22 23:44:58 +0000887 skb_frag_ref(skb, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 }
889 skb_shinfo(n)->nr_frags = i;
890 }
891
David S. Miller21dc3302010-08-23 00:13:46 -0700892 if (skb_has_frag_list(skb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
894 skb_clone_fraglist(n);
895 }
896
897 copy_skb_header(n, skb);
898out:
899 return n;
900}
Eric Dumazet117632e2011-12-03 21:39:53 +0000901EXPORT_SYMBOL(__pskb_copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902
903/**
904 * pskb_expand_head - reallocate header of &sk_buff
905 * @skb: buffer to reallocate
906 * @nhead: room to add at head
907 * @ntail: room to add at tail
908 * @gfp_mask: allocation priority
909 *
910 * Expands (or creates identical copy, if &nhead and &ntail are zero)
911 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
912 * reference count of 1. Returns zero in the case of success or error,
913 * if expansion failed. In the last case, &sk_buff is not changed.
914 *
915 * All the pointers pointing into skb header may change and must be
916 * reloaded after call to this function.
917 */
918
Victor Fusco86a76ca2005-07-08 14:57:47 -0700919int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
Al Virodd0fc662005-10-07 07:46:04 +0100920 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921{
922 int i;
923 u8 *data;
Eric Dumazet6602ceb2010-09-01 05:25:10 +0000924 int size = nhead + (skb_end_pointer(skb) - skb->head) + ntail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 long off;
Eric Dumazet1fd63042010-09-02 23:09:32 +0000926 bool fastpath;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927
Herbert Xu4edd87a2008-10-01 07:09:38 -0700928 BUG_ON(nhead < 0);
929
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 if (skb_shared(skb))
931 BUG();
932
933 size = SKB_DATA_ALIGN(size);
934
Changli Gaoca44ac32010-11-29 22:48:46 +0000935 /* Check if we can avoid taking references on fragments if we own
936 * the last reference on skb->head. (see skb_release_data())
937 */
938 if (!skb->cloned)
939 fastpath = true;
940 else {
941 int delta = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1;
Changli Gaoca44ac32010-11-29 22:48:46 +0000942 fastpath = atomic_read(&skb_shinfo(skb)->dataref) == delta;
943 }
944
945 if (fastpath &&
946 size + sizeof(struct skb_shared_info) <= ksize(skb->head)) {
947 memmove(skb->head + size, skb_shinfo(skb),
948 offsetof(struct skb_shared_info,
949 frags[skb_shinfo(skb)->nr_frags]));
950 memmove(skb->head + nhead, skb->head,
951 skb_tail_pointer(skb) - skb->head);
952 off = nhead;
953 goto adjust_others;
954 }
955
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
957 if (!data)
958 goto nodata;
959
960 /* Copy only real data... and, alas, header. This should be
Eric Dumazet6602ceb2010-09-01 05:25:10 +0000961 * optimized for the cases when header is void.
962 */
963 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
964
965 memcpy((struct skb_shared_info *)(data + size),
966 skb_shinfo(skb),
Eric Dumazetfed66382010-07-22 19:09:08 +0000967 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968
Eric Dumazet1fd63042010-09-02 23:09:32 +0000969 if (fastpath) {
970 kfree(skb->head);
971 } else {
Shirley Maa6686f22011-07-06 12:22:12 +0000972 /* copy this zero copy skb frags */
973 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
974 if (skb_copy_ubufs(skb, gfp_mask))
975 goto nofrags;
976 }
Eric Dumazet1fd63042010-09-02 23:09:32 +0000977 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
Ian Campbellea2ab692011-08-22 23:44:58 +0000978 skb_frag_ref(skb, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979
Eric Dumazet1fd63042010-09-02 23:09:32 +0000980 if (skb_has_frag_list(skb))
981 skb_clone_fraglist(skb);
982
983 skb_release_data(skb);
984 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 off = (data + nhead) - skb->head;
986
987 skb->head = data;
Changli Gaoca44ac32010-11-29 22:48:46 +0000988adjust_others:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 skb->data += off;
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700990#ifdef NET_SKBUFF_DATA_USES_OFFSET
991 skb->end = size;
Patrick McHardy56eb8882007-04-09 11:45:04 -0700992 off = nhead;
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700993#else
994 skb->end = skb->head + size;
Patrick McHardy56eb8882007-04-09 11:45:04 -0700995#endif
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700996 /* {transport,network,mac}_header and tail are relative to skb->head */
997 skb->tail += off;
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700998 skb->transport_header += off;
999 skb->network_header += off;
Stephen Hemminger603a8bb2009-06-17 12:17:34 +00001000 if (skb_mac_header_was_set(skb))
1001 skb->mac_header += off;
Andrea Shepard00c5a982010-07-22 09:12:35 +00001002 /* Only adjust this if it actually is csum_start rather than csum */
1003 if (skb->ip_summed == CHECKSUM_PARTIAL)
1004 skb->csum_start += nhead;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 skb->cloned = 0;
Patrick McHardy334a8132007-06-25 04:35:20 -07001006 skb->hdr_len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 skb->nohdr = 0;
1008 atomic_set(&skb_shinfo(skb)->dataref, 1);
1009 return 0;
1010
Shirley Maa6686f22011-07-06 12:22:12 +00001011nofrags:
1012 kfree(data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013nodata:
1014 return -ENOMEM;
1015}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001016EXPORT_SYMBOL(pskb_expand_head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017
1018/* Make private copy of skb with writable head and some headroom */
1019
1020struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1021{
1022 struct sk_buff *skb2;
1023 int delta = headroom - skb_headroom(skb);
1024
1025 if (delta <= 0)
1026 skb2 = pskb_copy(skb, GFP_ATOMIC);
1027 else {
1028 skb2 = skb_clone(skb, GFP_ATOMIC);
1029 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1030 GFP_ATOMIC)) {
1031 kfree_skb(skb2);
1032 skb2 = NULL;
1033 }
1034 }
1035 return skb2;
1036}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001037EXPORT_SYMBOL(skb_realloc_headroom);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038
1039/**
1040 * skb_copy_expand - copy and expand sk_buff
1041 * @skb: buffer to copy
1042 * @newheadroom: new free bytes at head
1043 * @newtailroom: new free bytes at tail
1044 * @gfp_mask: allocation priority
1045 *
1046 * Make a copy of both an &sk_buff and its data and while doing so
1047 * allocate additional space.
1048 *
1049 * This is used when the caller wishes to modify the data and needs a
1050 * private copy of the data to alter as well as more space for new fields.
1051 * Returns %NULL on failure or the pointer to the buffer
1052 * on success. The returned buffer has a reference count of 1.
1053 *
1054 * You must pass %GFP_ATOMIC as the allocation priority if this function
1055 * is called from an interrupt.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 */
1057struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
Victor Fusco86a76ca2005-07-08 14:57:47 -07001058 int newheadroom, int newtailroom,
Al Virodd0fc662005-10-07 07:46:04 +01001059 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060{
1061 /*
1062 * Allocate the copy buffer
1063 */
1064 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
1065 gfp_mask);
Patrick McHardyefd1e8d2007-04-10 18:30:09 -07001066 int oldheadroom = skb_headroom(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 int head_copy_len, head_copy_off;
Herbert Xu52886052007-09-16 16:32:11 -07001068 int off;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069
1070 if (!n)
1071 return NULL;
1072
1073 skb_reserve(n, newheadroom);
1074
1075 /* Set the tail pointer and length */
1076 skb_put(n, skb->len);
1077
Patrick McHardyefd1e8d2007-04-10 18:30:09 -07001078 head_copy_len = oldheadroom;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 head_copy_off = 0;
1080 if (newheadroom <= head_copy_len)
1081 head_copy_len = newheadroom;
1082 else
1083 head_copy_off = newheadroom - head_copy_len;
1084
1085 /* Copy the linear header and data. */
1086 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1087 skb->len + head_copy_len))
1088 BUG();
1089
1090 copy_skb_header(n, skb);
1091
Patrick McHardyefd1e8d2007-04-10 18:30:09 -07001092 off = newheadroom - oldheadroom;
David S. Millerbe2b6e62010-07-22 13:27:09 -07001093 if (n->ip_summed == CHECKSUM_PARTIAL)
1094 n->csum_start += off;
Herbert Xu52886052007-09-16 16:32:11 -07001095#ifdef NET_SKBUFF_DATA_USES_OFFSET
Patrick McHardyefd1e8d2007-04-10 18:30:09 -07001096 n->transport_header += off;
1097 n->network_header += off;
Stephen Hemminger603a8bb2009-06-17 12:17:34 +00001098 if (skb_mac_header_was_set(skb))
1099 n->mac_header += off;
Herbert Xu52886052007-09-16 16:32:11 -07001100#endif
Patrick McHardyefd1e8d2007-04-10 18:30:09 -07001101
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 return n;
1103}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001104EXPORT_SYMBOL(skb_copy_expand);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105
1106/**
1107 * skb_pad - zero pad the tail of an skb
1108 * @skb: buffer to pad
1109 * @pad: space to pad
1110 *
1111 * Ensure that a buffer is followed by a padding area that is zero
1112 * filled. Used by network drivers which may DMA or transfer data
1113 * beyond the buffer end onto the wire.
1114 *
Herbert Xu5b057c62006-06-23 02:06:41 -07001115 * May return error in out of memory cases. The skb is freed on error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 */
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001117
Herbert Xu5b057c62006-06-23 02:06:41 -07001118int skb_pad(struct sk_buff *skb, int pad)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119{
Herbert Xu5b057c62006-06-23 02:06:41 -07001120 int err;
1121 int ntail;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001122
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 /* If the skbuff is non linear tailroom is always zero.. */
Herbert Xu5b057c62006-06-23 02:06:41 -07001124 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 memset(skb->data+skb->len, 0, pad);
Herbert Xu5b057c62006-06-23 02:06:41 -07001126 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 }
Herbert Xu5b057c62006-06-23 02:06:41 -07001128
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001129 ntail = skb->data_len + pad - (skb->end - skb->tail);
Herbert Xu5b057c62006-06-23 02:06:41 -07001130 if (likely(skb_cloned(skb) || ntail > 0)) {
1131 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1132 if (unlikely(err))
1133 goto free_skb;
1134 }
1135
1136 /* FIXME: The use of this function with non-linear skb's really needs
1137 * to be audited.
1138 */
1139 err = skb_linearize(skb);
1140 if (unlikely(err))
1141 goto free_skb;
1142
1143 memset(skb->data + skb->len, 0, pad);
1144 return 0;
1145
1146free_skb:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 kfree_skb(skb);
Herbert Xu5b057c62006-06-23 02:06:41 -07001148 return err;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001149}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001150EXPORT_SYMBOL(skb_pad);
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001151
Ilpo Järvinen0dde3e12008-03-27 17:43:41 -07001152/**
1153 * skb_put - add data to a buffer
1154 * @skb: buffer to use
1155 * @len: amount of data to add
1156 *
1157 * This function extends the used data area of the buffer. If this would
1158 * exceed the total buffer size the kernel will panic. A pointer to the
1159 * first byte of the extra data is returned.
1160 */
1161unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1162{
1163 unsigned char *tmp = skb_tail_pointer(skb);
1164 SKB_LINEAR_ASSERT(skb);
1165 skb->tail += len;
1166 skb->len += len;
1167 if (unlikely(skb->tail > skb->end))
1168 skb_over_panic(skb, len, __builtin_return_address(0));
1169 return tmp;
1170}
1171EXPORT_SYMBOL(skb_put);
1172
Ilpo Järvinen6be8ac22008-03-27 17:47:24 -07001173/**
Ilpo Järvinenc2aa2702008-03-27 17:52:40 -07001174 * skb_push - add data to the start of a buffer
1175 * @skb: buffer to use
1176 * @len: amount of data to add
1177 *
1178 * This function extends the used data area of the buffer at the buffer
1179 * start. If this would exceed the total buffer headroom the kernel will
1180 * panic. A pointer to the first byte of the extra data is returned.
1181 */
1182unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1183{
1184 skb->data -= len;
1185 skb->len += len;
1186 if (unlikely(skb->data<skb->head))
1187 skb_under_panic(skb, len, __builtin_return_address(0));
1188 return skb->data;
1189}
1190EXPORT_SYMBOL(skb_push);
1191
1192/**
Ilpo Järvinen6be8ac22008-03-27 17:47:24 -07001193 * skb_pull - remove data from the start of a buffer
1194 * @skb: buffer to use
1195 * @len: amount of data to remove
1196 *
1197 * This function removes data from the start of a buffer, returning
1198 * the memory to the headroom. A pointer to the next data in the buffer
1199 * is returned. Once the data has been pulled future pushes will overwrite
1200 * the old data.
1201 */
1202unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1203{
David S. Miller47d29642010-05-02 02:21:44 -07001204 return skb_pull_inline(skb, len);
Ilpo Järvinen6be8ac22008-03-27 17:47:24 -07001205}
1206EXPORT_SYMBOL(skb_pull);
1207
Ilpo Järvinen419ae742008-03-27 17:54:01 -07001208/**
1209 * skb_trim - remove end from a buffer
1210 * @skb: buffer to alter
1211 * @len: new length
1212 *
1213 * Cut the length of a buffer down by removing data from the tail. If
1214 * the buffer is already under the length specified it is not modified.
1215 * The skb must be linear.
1216 */
1217void skb_trim(struct sk_buff *skb, unsigned int len)
1218{
1219 if (skb->len > len)
1220 __skb_trim(skb, len);
1221}
1222EXPORT_SYMBOL(skb_trim);
1223
Herbert Xu3cc0e872006-06-09 16:13:38 -07001224/* Trims skb to length len. It can change skb pointers.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 */
1226
Herbert Xu3cc0e872006-06-09 16:13:38 -07001227int ___pskb_trim(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228{
Herbert Xu27b437c2006-07-13 19:26:39 -07001229 struct sk_buff **fragp;
1230 struct sk_buff *frag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 int offset = skb_headlen(skb);
1232 int nfrags = skb_shinfo(skb)->nr_frags;
1233 int i;
Herbert Xu27b437c2006-07-13 19:26:39 -07001234 int err;
1235
1236 if (skb_cloned(skb) &&
1237 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1238 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001240 i = 0;
1241 if (offset >= len)
1242 goto drop_pages;
1243
1244 for (; i < nfrags; i++) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00001245 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
Herbert Xu27b437c2006-07-13 19:26:39 -07001246
1247 if (end < len) {
1248 offset = end;
1249 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 }
Herbert Xu27b437c2006-07-13 19:26:39 -07001251
Eric Dumazet9e903e02011-10-18 21:00:24 +00001252 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
Herbert Xu27b437c2006-07-13 19:26:39 -07001253
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001254drop_pages:
Herbert Xu27b437c2006-07-13 19:26:39 -07001255 skb_shinfo(skb)->nr_frags = i;
1256
1257 for (; i < nfrags; i++)
Ian Campbellea2ab692011-08-22 23:44:58 +00001258 skb_frag_unref(skb, i);
Herbert Xu27b437c2006-07-13 19:26:39 -07001259
David S. Miller21dc3302010-08-23 00:13:46 -07001260 if (skb_has_frag_list(skb))
Herbert Xu27b437c2006-07-13 19:26:39 -07001261 skb_drop_fraglist(skb);
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001262 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 }
1264
Herbert Xu27b437c2006-07-13 19:26:39 -07001265 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1266 fragp = &frag->next) {
1267 int end = offset + frag->len;
1268
1269 if (skb_shared(frag)) {
1270 struct sk_buff *nfrag;
1271
1272 nfrag = skb_clone(frag, GFP_ATOMIC);
1273 if (unlikely(!nfrag))
1274 return -ENOMEM;
1275
1276 nfrag->next = frag->next;
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001277 kfree_skb(frag);
Herbert Xu27b437c2006-07-13 19:26:39 -07001278 frag = nfrag;
1279 *fragp = frag;
1280 }
1281
1282 if (end < len) {
1283 offset = end;
1284 continue;
1285 }
1286
1287 if (end > len &&
1288 unlikely((err = pskb_trim(frag, len - offset))))
1289 return err;
1290
1291 if (frag->next)
1292 skb_drop_list(&frag->next);
1293 break;
1294 }
1295
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001296done:
Herbert Xu27b437c2006-07-13 19:26:39 -07001297 if (len > skb_headlen(skb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 skb->data_len -= skb->len - len;
1299 skb->len = len;
1300 } else {
Herbert Xu27b437c2006-07-13 19:26:39 -07001301 skb->len = len;
1302 skb->data_len = 0;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001303 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 }
1305
1306 return 0;
1307}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001308EXPORT_SYMBOL(___pskb_trim);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309
1310/**
1311 * __pskb_pull_tail - advance tail of skb header
1312 * @skb: buffer to reallocate
1313 * @delta: number of bytes to advance tail
1314 *
1315 * The function makes a sense only on a fragmented &sk_buff,
1316 * it expands header moving its tail forward and copying necessary
1317 * data from fragmented part.
1318 *
1319 * &sk_buff MUST have reference count of 1.
1320 *
1321 * Returns %NULL (and &sk_buff does not change) if pull failed
1322 * or value of new tail of skb in the case of success.
1323 *
1324 * All the pointers pointing into skb header may change and must be
1325 * reloaded after call to this function.
1326 */
1327
1328/* Moves tail of skb head forward, copying data from fragmented part,
1329 * when it is necessary.
1330 * 1. It may fail due to malloc failure.
1331 * 2. It may change skb pointers.
1332 *
1333 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1334 */
1335unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1336{
1337 /* If skb has not enough free space at tail, get new one
1338 * plus 128 bytes for future expansions. If we have enough
1339 * room at tail, reallocate without expansion only if skb is cloned.
1340 */
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001341 int i, k, eat = (skb->tail + delta) - skb->end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342
1343 if (eat > 0 || skb_cloned(skb)) {
1344 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1345 GFP_ATOMIC))
1346 return NULL;
1347 }
1348
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001349 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 BUG();
1351
1352 /* Optimization: no fragments, no reasons to preestimate
1353 * size of pulled pages. Superb.
1354 */
David S. Miller21dc3302010-08-23 00:13:46 -07001355 if (!skb_has_frag_list(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 goto pull_pages;
1357
1358 /* Estimate size of pulled pages. */
1359 eat = delta;
1360 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00001361 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1362
1363 if (size >= eat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 goto pull_pages;
Eric Dumazet9e903e02011-10-18 21:00:24 +00001365 eat -= size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 }
1367
1368 /* If we need update frag list, we are in troubles.
1369 * Certainly, it possible to add an offset to skb data,
1370 * but taking into account that pulling is expected to
1371 * be very rare operation, it is worth to fight against
1372 * further bloating skb head and crucify ourselves here instead.
1373 * Pure masohism, indeed. 8)8)
1374 */
1375 if (eat) {
1376 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1377 struct sk_buff *clone = NULL;
1378 struct sk_buff *insp = NULL;
1379
1380 do {
Kris Katterjohn09a62662006-01-08 22:24:28 -08001381 BUG_ON(!list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382
1383 if (list->len <= eat) {
1384 /* Eaten as whole. */
1385 eat -= list->len;
1386 list = list->next;
1387 insp = list;
1388 } else {
1389 /* Eaten partially. */
1390
1391 if (skb_shared(list)) {
1392 /* Sucks! We need to fork list. :-( */
1393 clone = skb_clone(list, GFP_ATOMIC);
1394 if (!clone)
1395 return NULL;
1396 insp = list->next;
1397 list = clone;
1398 } else {
1399 /* This may be pulled without
1400 * problems. */
1401 insp = list;
1402 }
1403 if (!pskb_pull(list, eat)) {
Wei Yongjunf3fbbe02009-02-25 00:37:32 +00001404 kfree_skb(clone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 return NULL;
1406 }
1407 break;
1408 }
1409 } while (eat);
1410
1411 /* Free pulled out fragments. */
1412 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1413 skb_shinfo(skb)->frag_list = list->next;
1414 kfree_skb(list);
1415 }
1416 /* And insert new clone at head. */
1417 if (clone) {
1418 clone->next = list;
1419 skb_shinfo(skb)->frag_list = clone;
1420 }
1421 }
1422 /* Success! Now we may commit changes to skb data. */
1423
1424pull_pages:
1425 eat = delta;
1426 k = 0;
1427 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00001428 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1429
1430 if (size <= eat) {
Ian Campbellea2ab692011-08-22 23:44:58 +00001431 skb_frag_unref(skb, i);
Eric Dumazet9e903e02011-10-18 21:00:24 +00001432 eat -= size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 } else {
1434 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1435 if (eat) {
1436 skb_shinfo(skb)->frags[k].page_offset += eat;
Eric Dumazet9e903e02011-10-18 21:00:24 +00001437 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 eat = 0;
1439 }
1440 k++;
1441 }
1442 }
1443 skb_shinfo(skb)->nr_frags = k;
1444
1445 skb->tail += delta;
1446 skb->data_len -= delta;
1447
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001448 return skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001450EXPORT_SYMBOL(__pskb_pull_tail);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451
Eric Dumazet22019b12011-07-29 18:37:31 +00001452/**
1453 * skb_copy_bits - copy bits from skb to kernel buffer
1454 * @skb: source skb
1455 * @offset: offset in source
1456 * @to: destination buffer
1457 * @len: number of bytes to copy
1458 *
1459 * Copy the specified number of bytes from the source skb to the
1460 * destination buffer.
1461 *
1462 * CAUTION ! :
1463 * If its prototype is ever changed,
1464 * check arch/{*}/net/{*}.S files,
1465 * since it is called from BPF assembly code.
1466 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1468{
David S. Miller1a028e52007-04-27 15:21:23 -07001469 int start = skb_headlen(skb);
David S. Millerfbb398a2009-06-09 00:18:59 -07001470 struct sk_buff *frag_iter;
1471 int i, copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472
1473 if (offset > (int)skb->len - len)
1474 goto fault;
1475
1476 /* Copy header. */
David S. Miller1a028e52007-04-27 15:21:23 -07001477 if ((copy = start - offset) > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 if (copy > len)
1479 copy = len;
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03001480 skb_copy_from_linear_data_offset(skb, offset, to, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 if ((len -= copy) == 0)
1482 return 0;
1483 offset += copy;
1484 to += copy;
1485 }
1486
1487 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07001488 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001490 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07001491
Eric Dumazet9e903e02011-10-18 21:00:24 +00001492 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 if ((copy = end - offset) > 0) {
1494 u8 *vaddr;
1495
1496 if (copy > len)
1497 copy = len;
1498
1499 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1500 memcpy(to,
David S. Miller1a028e52007-04-27 15:21:23 -07001501 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1502 offset - start, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 kunmap_skb_frag(vaddr);
1504
1505 if ((len -= copy) == 0)
1506 return 0;
1507 offset += copy;
1508 to += copy;
1509 }
David S. Miller1a028e52007-04-27 15:21:23 -07001510 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 }
1512
David S. Millerfbb398a2009-06-09 00:18:59 -07001513 skb_walk_frags(skb, frag_iter) {
1514 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515
David S. Millerfbb398a2009-06-09 00:18:59 -07001516 WARN_ON(start > offset + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517
David S. Millerfbb398a2009-06-09 00:18:59 -07001518 end = start + frag_iter->len;
1519 if ((copy = end - offset) > 0) {
1520 if (copy > len)
1521 copy = len;
1522 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1523 goto fault;
1524 if ((len -= copy) == 0)
1525 return 0;
1526 offset += copy;
1527 to += copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 }
David S. Millerfbb398a2009-06-09 00:18:59 -07001529 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 }
Shirley Maa6686f22011-07-06 12:22:12 +00001531
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 if (!len)
1533 return 0;
1534
1535fault:
1536 return -EFAULT;
1537}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001538EXPORT_SYMBOL(skb_copy_bits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539
Jens Axboe9c55e012007-11-06 23:30:13 -08001540/*
1541 * Callback from splice_to_pipe(), if we need to release some pages
1542 * at the end of the spd in case we error'ed out in filling the pipe.
1543 */
1544static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1545{
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001546 put_page(spd->pages[i]);
1547}
Jens Axboe9c55e012007-11-06 23:30:13 -08001548
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001549static inline struct page *linear_to_page(struct page *page, unsigned int *len,
1550 unsigned int *offset,
Jarek Poplawski7a67e562009-04-30 05:41:19 -07001551 struct sk_buff *skb, struct sock *sk)
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001552{
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001553 struct page *p = sk->sk_sndmsg_page;
1554 unsigned int off;
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001555
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001556 if (!p) {
1557new_page:
1558 p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1559 if (!p)
1560 return NULL;
1561
1562 off = sk->sk_sndmsg_off = 0;
1563 /* hold one ref to this page until it's full */
1564 } else {
1565 unsigned int mlen;
1566
1567 off = sk->sk_sndmsg_off;
1568 mlen = PAGE_SIZE - off;
1569 if (mlen < 64 && mlen < *len) {
1570 put_page(p);
1571 goto new_page;
1572 }
1573
1574 *len = min_t(unsigned int, *len, mlen);
1575 }
1576
1577 memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1578 sk->sk_sndmsg_off += *len;
1579 *offset = off;
1580 get_page(p);
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001581
1582 return p;
Jens Axboe9c55e012007-11-06 23:30:13 -08001583}
1584
1585/*
1586 * Fill page/offset/length into spd, if it can hold more pages.
1587 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001588static inline int spd_fill_page(struct splice_pipe_desc *spd,
1589 struct pipe_inode_info *pipe, struct page *page,
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001590 unsigned int *len, unsigned int offset,
Jarek Poplawski7a67e562009-04-30 05:41:19 -07001591 struct sk_buff *skb, int linear,
1592 struct sock *sk)
Jens Axboe9c55e012007-11-06 23:30:13 -08001593{
Jens Axboe35f3d142010-05-20 10:43:18 +02001594 if (unlikely(spd->nr_pages == pipe->buffers))
Jens Axboe9c55e012007-11-06 23:30:13 -08001595 return 1;
1596
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001597 if (linear) {
Jarek Poplawski7a67e562009-04-30 05:41:19 -07001598 page = linear_to_page(page, len, &offset, skb, sk);
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001599 if (!page)
1600 return 1;
1601 } else
1602 get_page(page);
1603
Jens Axboe9c55e012007-11-06 23:30:13 -08001604 spd->pages[spd->nr_pages] = page;
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001605 spd->partial[spd->nr_pages].len = *len;
Jens Axboe9c55e012007-11-06 23:30:13 -08001606 spd->partial[spd->nr_pages].offset = offset;
Jens Axboe9c55e012007-11-06 23:30:13 -08001607 spd->nr_pages++;
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001608
Jens Axboe9c55e012007-11-06 23:30:13 -08001609 return 0;
1610}
1611
Octavian Purdila2870c432008-07-15 00:49:11 -07001612static inline void __segment_seek(struct page **page, unsigned int *poff,
1613 unsigned int *plen, unsigned int off)
Jens Axboe9c55e012007-11-06 23:30:13 -08001614{
Jarek Poplawskice3dd392009-02-12 16:51:43 -08001615 unsigned long n;
1616
Octavian Purdila2870c432008-07-15 00:49:11 -07001617 *poff += off;
Jarek Poplawskice3dd392009-02-12 16:51:43 -08001618 n = *poff / PAGE_SIZE;
1619 if (n)
1620 *page = nth_page(*page, n);
1621
Octavian Purdila2870c432008-07-15 00:49:11 -07001622 *poff = *poff % PAGE_SIZE;
1623 *plen -= off;
1624}
Jens Axboe9c55e012007-11-06 23:30:13 -08001625
Octavian Purdila2870c432008-07-15 00:49:11 -07001626static inline int __splice_segment(struct page *page, unsigned int poff,
1627 unsigned int plen, unsigned int *off,
1628 unsigned int *len, struct sk_buff *skb,
Jarek Poplawski7a67e562009-04-30 05:41:19 -07001629 struct splice_pipe_desc *spd, int linear,
Jens Axboe35f3d142010-05-20 10:43:18 +02001630 struct sock *sk,
1631 struct pipe_inode_info *pipe)
Octavian Purdila2870c432008-07-15 00:49:11 -07001632{
1633 if (!*len)
1634 return 1;
1635
1636 /* skip this segment if already processed */
1637 if (*off >= plen) {
1638 *off -= plen;
1639 return 0;
Octavian Purdiladb43a282008-06-27 17:27:21 -07001640 }
Jens Axboe9c55e012007-11-06 23:30:13 -08001641
Octavian Purdila2870c432008-07-15 00:49:11 -07001642 /* ignore any bits we already processed */
1643 if (*off) {
1644 __segment_seek(&page, &poff, &plen, *off);
1645 *off = 0;
1646 }
1647
1648 do {
1649 unsigned int flen = min(*len, plen);
1650
1651 /* the linear region may spread across several pages */
1652 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1653
Jens Axboe35f3d142010-05-20 10:43:18 +02001654 if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk))
Octavian Purdila2870c432008-07-15 00:49:11 -07001655 return 1;
1656
1657 __segment_seek(&page, &poff, &plen, flen);
1658 *len -= flen;
1659
1660 } while (*len && plen);
1661
1662 return 0;
1663}
1664
1665/*
1666 * Map linear and fragment data from the skb to spd. It reports failure if the
1667 * pipe is full or if we already spliced the requested length.
1668 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001669static int __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1670 unsigned int *offset, unsigned int *len,
1671 struct splice_pipe_desc *spd, struct sock *sk)
Octavian Purdila2870c432008-07-15 00:49:11 -07001672{
1673 int seg;
1674
Jens Axboe9c55e012007-11-06 23:30:13 -08001675 /*
Octavian Purdila2870c432008-07-15 00:49:11 -07001676 * map the linear part
Jens Axboe9c55e012007-11-06 23:30:13 -08001677 */
Octavian Purdila2870c432008-07-15 00:49:11 -07001678 if (__splice_segment(virt_to_page(skb->data),
1679 (unsigned long) skb->data & (PAGE_SIZE - 1),
1680 skb_headlen(skb),
Jens Axboe35f3d142010-05-20 10:43:18 +02001681 offset, len, skb, spd, 1, sk, pipe))
Octavian Purdila2870c432008-07-15 00:49:11 -07001682 return 1;
Jens Axboe9c55e012007-11-06 23:30:13 -08001683
1684 /*
1685 * then map the fragments
1686 */
Jens Axboe9c55e012007-11-06 23:30:13 -08001687 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1688 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1689
Ian Campbellea2ab692011-08-22 23:44:58 +00001690 if (__splice_segment(skb_frag_page(f),
Eric Dumazet9e903e02011-10-18 21:00:24 +00001691 f->page_offset, skb_frag_size(f),
Jens Axboe35f3d142010-05-20 10:43:18 +02001692 offset, len, skb, spd, 0, sk, pipe))
Octavian Purdila2870c432008-07-15 00:49:11 -07001693 return 1;
Jens Axboe9c55e012007-11-06 23:30:13 -08001694 }
1695
Octavian Purdila2870c432008-07-15 00:49:11 -07001696 return 0;
Jens Axboe9c55e012007-11-06 23:30:13 -08001697}
1698
1699/*
1700 * Map data from the skb to a pipe. Should handle both the linear part,
1701 * the fragments, and the frag list. It does NOT handle frag lists within
1702 * the frag list, if such a thing exists. We'd probably need to recurse to
1703 * handle that cleanly.
1704 */
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001705int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
Jens Axboe9c55e012007-11-06 23:30:13 -08001706 struct pipe_inode_info *pipe, unsigned int tlen,
1707 unsigned int flags)
1708{
Jens Axboe35f3d142010-05-20 10:43:18 +02001709 struct partial_page partial[PIPE_DEF_BUFFERS];
1710 struct page *pages[PIPE_DEF_BUFFERS];
Jens Axboe9c55e012007-11-06 23:30:13 -08001711 struct splice_pipe_desc spd = {
1712 .pages = pages,
1713 .partial = partial,
1714 .flags = flags,
1715 .ops = &sock_pipe_buf_ops,
1716 .spd_release = sock_spd_release,
1717 };
David S. Millerfbb398a2009-06-09 00:18:59 -07001718 struct sk_buff *frag_iter;
Jarek Poplawski7a67e562009-04-30 05:41:19 -07001719 struct sock *sk = skb->sk;
Jens Axboe35f3d142010-05-20 10:43:18 +02001720 int ret = 0;
1721
1722 if (splice_grow_spd(pipe, &spd))
1723 return -ENOMEM;
Jens Axboe9c55e012007-11-06 23:30:13 -08001724
1725 /*
1726 * __skb_splice_bits() only fails if the output has no room left,
1727 * so no point in going over the frag_list for the error case.
1728 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001729 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
Jens Axboe9c55e012007-11-06 23:30:13 -08001730 goto done;
1731 else if (!tlen)
1732 goto done;
1733
1734 /*
1735 * now see if we have a frag_list to map
1736 */
David S. Millerfbb398a2009-06-09 00:18:59 -07001737 skb_walk_frags(skb, frag_iter) {
1738 if (!tlen)
1739 break;
Jens Axboe35f3d142010-05-20 10:43:18 +02001740 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
David S. Millerfbb398a2009-06-09 00:18:59 -07001741 break;
Jens Axboe9c55e012007-11-06 23:30:13 -08001742 }
1743
1744done:
Jens Axboe9c55e012007-11-06 23:30:13 -08001745 if (spd.nr_pages) {
Jens Axboe9c55e012007-11-06 23:30:13 -08001746 /*
1747 * Drop the socket lock, otherwise we have reverse
1748 * locking dependencies between sk_lock and i_mutex
1749 * here as compared to sendfile(). We enter here
1750 * with the socket lock held, and splice_to_pipe() will
1751 * grab the pipe inode lock. For sendfile() emulation,
1752 * we call into ->sendpage() with the i_mutex lock held
1753 * and networking will grab the socket lock.
1754 */
Octavian Purdila293ad602008-06-04 15:45:58 -07001755 release_sock(sk);
Jens Axboe9c55e012007-11-06 23:30:13 -08001756 ret = splice_to_pipe(pipe, &spd);
Octavian Purdila293ad602008-06-04 15:45:58 -07001757 lock_sock(sk);
Jens Axboe9c55e012007-11-06 23:30:13 -08001758 }
1759
Jens Axboe35f3d142010-05-20 10:43:18 +02001760 splice_shrink_spd(pipe, &spd);
1761 return ret;
Jens Axboe9c55e012007-11-06 23:30:13 -08001762}
1763
Herbert Xu357b40a2005-04-19 22:30:14 -07001764/**
1765 * skb_store_bits - store bits from kernel buffer to skb
1766 * @skb: destination buffer
1767 * @offset: offset in destination
1768 * @from: source buffer
1769 * @len: number of bytes to copy
1770 *
1771 * Copy the specified number of bytes from the source buffer to the
1772 * destination skb. This function handles all the messy bits of
1773 * traversing fragment lists and such.
1774 */
1775
Stephen Hemminger0c6fcc82007-04-20 16:40:01 -07001776int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
Herbert Xu357b40a2005-04-19 22:30:14 -07001777{
David S. Miller1a028e52007-04-27 15:21:23 -07001778 int start = skb_headlen(skb);
David S. Millerfbb398a2009-06-09 00:18:59 -07001779 struct sk_buff *frag_iter;
1780 int i, copy;
Herbert Xu357b40a2005-04-19 22:30:14 -07001781
1782 if (offset > (int)skb->len - len)
1783 goto fault;
1784
David S. Miller1a028e52007-04-27 15:21:23 -07001785 if ((copy = start - offset) > 0) {
Herbert Xu357b40a2005-04-19 22:30:14 -07001786 if (copy > len)
1787 copy = len;
Arnaldo Carvalho de Melo27d7ff42007-03-31 11:55:19 -03001788 skb_copy_to_linear_data_offset(skb, offset, from, copy);
Herbert Xu357b40a2005-04-19 22:30:14 -07001789 if ((len -= copy) == 0)
1790 return 0;
1791 offset += copy;
1792 from += copy;
1793 }
1794
1795 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1796 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
David S. Miller1a028e52007-04-27 15:21:23 -07001797 int end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001798
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001799 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07001800
Eric Dumazet9e903e02011-10-18 21:00:24 +00001801 end = start + skb_frag_size(frag);
Herbert Xu357b40a2005-04-19 22:30:14 -07001802 if ((copy = end - offset) > 0) {
1803 u8 *vaddr;
1804
1805 if (copy > len)
1806 copy = len;
1807
1808 vaddr = kmap_skb_frag(frag);
David S. Miller1a028e52007-04-27 15:21:23 -07001809 memcpy(vaddr + frag->page_offset + offset - start,
1810 from, copy);
Herbert Xu357b40a2005-04-19 22:30:14 -07001811 kunmap_skb_frag(vaddr);
1812
1813 if ((len -= copy) == 0)
1814 return 0;
1815 offset += copy;
1816 from += copy;
1817 }
David S. Miller1a028e52007-04-27 15:21:23 -07001818 start = end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001819 }
1820
David S. Millerfbb398a2009-06-09 00:18:59 -07001821 skb_walk_frags(skb, frag_iter) {
1822 int end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001823
David S. Millerfbb398a2009-06-09 00:18:59 -07001824 WARN_ON(start > offset + len);
Herbert Xu357b40a2005-04-19 22:30:14 -07001825
David S. Millerfbb398a2009-06-09 00:18:59 -07001826 end = start + frag_iter->len;
1827 if ((copy = end - offset) > 0) {
1828 if (copy > len)
1829 copy = len;
1830 if (skb_store_bits(frag_iter, offset - start,
1831 from, copy))
1832 goto fault;
1833 if ((len -= copy) == 0)
1834 return 0;
1835 offset += copy;
1836 from += copy;
Herbert Xu357b40a2005-04-19 22:30:14 -07001837 }
David S. Millerfbb398a2009-06-09 00:18:59 -07001838 start = end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001839 }
1840 if (!len)
1841 return 0;
1842
1843fault:
1844 return -EFAULT;
1845}
Herbert Xu357b40a2005-04-19 22:30:14 -07001846EXPORT_SYMBOL(skb_store_bits);
1847
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848/* Checksum skb data. */
1849
Al Viro2bbbc862006-11-14 21:37:14 -08001850__wsum skb_checksum(const struct sk_buff *skb, int offset,
1851 int len, __wsum csum)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852{
David S. Miller1a028e52007-04-27 15:21:23 -07001853 int start = skb_headlen(skb);
1854 int i, copy = start - offset;
David S. Millerfbb398a2009-06-09 00:18:59 -07001855 struct sk_buff *frag_iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 int pos = 0;
1857
1858 /* Checksum header. */
1859 if (copy > 0) {
1860 if (copy > len)
1861 copy = len;
1862 csum = csum_partial(skb->data + offset, copy, csum);
1863 if ((len -= copy) == 0)
1864 return csum;
1865 offset += copy;
1866 pos = copy;
1867 }
1868
1869 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07001870 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001872 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07001873
Eric Dumazet9e903e02011-10-18 21:00:24 +00001874 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 if ((copy = end - offset) > 0) {
Al Viro44bb9362006-11-14 21:36:14 -08001876 __wsum csum2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 u8 *vaddr;
1878 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1879
1880 if (copy > len)
1881 copy = len;
1882 vaddr = kmap_skb_frag(frag);
David S. Miller1a028e52007-04-27 15:21:23 -07001883 csum2 = csum_partial(vaddr + frag->page_offset +
1884 offset - start, copy, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 kunmap_skb_frag(vaddr);
1886 csum = csum_block_add(csum, csum2, pos);
1887 if (!(len -= copy))
1888 return csum;
1889 offset += copy;
1890 pos += copy;
1891 }
David S. Miller1a028e52007-04-27 15:21:23 -07001892 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 }
1894
David S. Millerfbb398a2009-06-09 00:18:59 -07001895 skb_walk_frags(skb, frag_iter) {
1896 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897
David S. Millerfbb398a2009-06-09 00:18:59 -07001898 WARN_ON(start > offset + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899
David S. Millerfbb398a2009-06-09 00:18:59 -07001900 end = start + frag_iter->len;
1901 if ((copy = end - offset) > 0) {
1902 __wsum csum2;
1903 if (copy > len)
1904 copy = len;
1905 csum2 = skb_checksum(frag_iter, offset - start,
1906 copy, 0);
1907 csum = csum_block_add(csum, csum2, pos);
1908 if ((len -= copy) == 0)
1909 return csum;
1910 offset += copy;
1911 pos += copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 }
David S. Millerfbb398a2009-06-09 00:18:59 -07001913 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 }
Kris Katterjohn09a62662006-01-08 22:24:28 -08001915 BUG_ON(len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916
1917 return csum;
1918}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001919EXPORT_SYMBOL(skb_checksum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920
1921/* Both of above in one bottle. */
1922
Al Viro81d77662006-11-14 21:37:33 -08001923__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1924 u8 *to, int len, __wsum csum)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925{
David S. Miller1a028e52007-04-27 15:21:23 -07001926 int start = skb_headlen(skb);
1927 int i, copy = start - offset;
David S. Millerfbb398a2009-06-09 00:18:59 -07001928 struct sk_buff *frag_iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 int pos = 0;
1930
1931 /* Copy header. */
1932 if (copy > 0) {
1933 if (copy > len)
1934 copy = len;
1935 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1936 copy, csum);
1937 if ((len -= copy) == 0)
1938 return csum;
1939 offset += copy;
1940 to += copy;
1941 pos = copy;
1942 }
1943
1944 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07001945 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001947 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07001948
Eric Dumazet9e903e02011-10-18 21:00:24 +00001949 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 if ((copy = end - offset) > 0) {
Al Viro50842052006-11-14 21:36:34 -08001951 __wsum csum2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 u8 *vaddr;
1953 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1954
1955 if (copy > len)
1956 copy = len;
1957 vaddr = kmap_skb_frag(frag);
1958 csum2 = csum_partial_copy_nocheck(vaddr +
David S. Miller1a028e52007-04-27 15:21:23 -07001959 frag->page_offset +
1960 offset - start, to,
1961 copy, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 kunmap_skb_frag(vaddr);
1963 csum = csum_block_add(csum, csum2, pos);
1964 if (!(len -= copy))
1965 return csum;
1966 offset += copy;
1967 to += copy;
1968 pos += copy;
1969 }
David S. Miller1a028e52007-04-27 15:21:23 -07001970 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 }
1972
David S. Millerfbb398a2009-06-09 00:18:59 -07001973 skb_walk_frags(skb, frag_iter) {
1974 __wsum csum2;
1975 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976
David S. Millerfbb398a2009-06-09 00:18:59 -07001977 WARN_ON(start > offset + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978
David S. Millerfbb398a2009-06-09 00:18:59 -07001979 end = start + frag_iter->len;
1980 if ((copy = end - offset) > 0) {
1981 if (copy > len)
1982 copy = len;
1983 csum2 = skb_copy_and_csum_bits(frag_iter,
1984 offset - start,
1985 to, copy, 0);
1986 csum = csum_block_add(csum, csum2, pos);
1987 if ((len -= copy) == 0)
1988 return csum;
1989 offset += copy;
1990 to += copy;
1991 pos += copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 }
David S. Millerfbb398a2009-06-09 00:18:59 -07001993 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 }
Kris Katterjohn09a62662006-01-08 22:24:28 -08001995 BUG_ON(len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 return csum;
1997}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001998EXPORT_SYMBOL(skb_copy_and_csum_bits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999
2000void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2001{
Al Virod3bc23e2006-11-14 21:24:49 -08002002 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003 long csstart;
2004
Patrick McHardy84fa7932006-08-29 16:44:56 -07002005 if (skb->ip_summed == CHECKSUM_PARTIAL)
Michał Mirosław55508d62010-12-14 15:24:08 +00002006 csstart = skb_checksum_start_offset(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 else
2008 csstart = skb_headlen(skb);
2009
Kris Katterjohn09a62662006-01-08 22:24:28 -08002010 BUG_ON(csstart > skb_headlen(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03002012 skb_copy_from_linear_data(skb, to, csstart);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013
2014 csum = 0;
2015 if (csstart != skb->len)
2016 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2017 skb->len - csstart, 0);
2018
Patrick McHardy84fa7932006-08-29 16:44:56 -07002019 if (skb->ip_summed == CHECKSUM_PARTIAL) {
Al Viroff1dcad2006-11-20 18:07:29 -08002020 long csstuff = csstart + skb->csum_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021
Al Virod3bc23e2006-11-14 21:24:49 -08002022 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 }
2024}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002025EXPORT_SYMBOL(skb_copy_and_csum_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026
2027/**
2028 * skb_dequeue - remove from the head of the queue
2029 * @list: list to dequeue from
2030 *
2031 * Remove the head of the list. The list lock is taken so the function
2032 * may be used safely with other locking list functions. The head item is
2033 * returned or %NULL if the list is empty.
2034 */
2035
2036struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2037{
2038 unsigned long flags;
2039 struct sk_buff *result;
2040
2041 spin_lock_irqsave(&list->lock, flags);
2042 result = __skb_dequeue(list);
2043 spin_unlock_irqrestore(&list->lock, flags);
2044 return result;
2045}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002046EXPORT_SYMBOL(skb_dequeue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047
2048/**
2049 * skb_dequeue_tail - remove from the tail of the queue
2050 * @list: list to dequeue from
2051 *
2052 * Remove the tail of the list. The list lock is taken so the function
2053 * may be used safely with other locking list functions. The tail item is
2054 * returned or %NULL if the list is empty.
2055 */
2056struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2057{
2058 unsigned long flags;
2059 struct sk_buff *result;
2060
2061 spin_lock_irqsave(&list->lock, flags);
2062 result = __skb_dequeue_tail(list);
2063 spin_unlock_irqrestore(&list->lock, flags);
2064 return result;
2065}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002066EXPORT_SYMBOL(skb_dequeue_tail);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067
2068/**
2069 * skb_queue_purge - empty a list
2070 * @list: list to empty
2071 *
2072 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2073 * the list and one reference dropped. This function takes the list
2074 * lock and is atomic with respect to other list locking functions.
2075 */
2076void skb_queue_purge(struct sk_buff_head *list)
2077{
2078 struct sk_buff *skb;
2079 while ((skb = skb_dequeue(list)) != NULL)
2080 kfree_skb(skb);
2081}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002082EXPORT_SYMBOL(skb_queue_purge);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083
2084/**
2085 * skb_queue_head - queue a buffer at the list head
2086 * @list: list to use
2087 * @newsk: buffer to queue
2088 *
2089 * Queue a buffer at the start of the list. This function takes the
2090 * list lock and can be used safely with other locking &sk_buff functions
2091 * safely.
2092 *
2093 * A buffer cannot be placed on two lists at the same time.
2094 */
2095void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2096{
2097 unsigned long flags;
2098
2099 spin_lock_irqsave(&list->lock, flags);
2100 __skb_queue_head(list, newsk);
2101 spin_unlock_irqrestore(&list->lock, flags);
2102}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002103EXPORT_SYMBOL(skb_queue_head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104
2105/**
2106 * skb_queue_tail - queue a buffer at the list tail
2107 * @list: list to use
2108 * @newsk: buffer to queue
2109 *
2110 * Queue a buffer at the tail of the list. This function takes the
2111 * list lock and can be used safely with other locking &sk_buff functions
2112 * safely.
2113 *
2114 * A buffer cannot be placed on two lists at the same time.
2115 */
2116void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2117{
2118 unsigned long flags;
2119
2120 spin_lock_irqsave(&list->lock, flags);
2121 __skb_queue_tail(list, newsk);
2122 spin_unlock_irqrestore(&list->lock, flags);
2123}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002124EXPORT_SYMBOL(skb_queue_tail);
David S. Miller8728b832005-08-09 19:25:21 -07002125
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126/**
2127 * skb_unlink - remove a buffer from a list
2128 * @skb: buffer to remove
David S. Miller8728b832005-08-09 19:25:21 -07002129 * @list: list to use
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 *
David S. Miller8728b832005-08-09 19:25:21 -07002131 * Remove a packet from a list. The list locks are taken and this
2132 * function is atomic with respect to other list locked calls
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 *
David S. Miller8728b832005-08-09 19:25:21 -07002134 * You must know what list the SKB is on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 */
David S. Miller8728b832005-08-09 19:25:21 -07002136void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137{
David S. Miller8728b832005-08-09 19:25:21 -07002138 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139
David S. Miller8728b832005-08-09 19:25:21 -07002140 spin_lock_irqsave(&list->lock, flags);
2141 __skb_unlink(skb, list);
2142 spin_unlock_irqrestore(&list->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002144EXPORT_SYMBOL(skb_unlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146/**
2147 * skb_append - append a buffer
2148 * @old: buffer to insert after
2149 * @newsk: buffer to insert
David S. Miller8728b832005-08-09 19:25:21 -07002150 * @list: list to use
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 *
2152 * Place a packet after a given packet in a list. The list locks are taken
2153 * and this function is atomic with respect to other list locked calls.
2154 * A buffer cannot be placed on two lists at the same time.
2155 */
David S. Miller8728b832005-08-09 19:25:21 -07002156void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157{
2158 unsigned long flags;
2159
David S. Miller8728b832005-08-09 19:25:21 -07002160 spin_lock_irqsave(&list->lock, flags);
Gerrit Renker7de6c032008-04-14 00:05:09 -07002161 __skb_queue_after(list, old, newsk);
David S. Miller8728b832005-08-09 19:25:21 -07002162 spin_unlock_irqrestore(&list->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002164EXPORT_SYMBOL(skb_append);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165
2166/**
2167 * skb_insert - insert a buffer
2168 * @old: buffer to insert before
2169 * @newsk: buffer to insert
David S. Miller8728b832005-08-09 19:25:21 -07002170 * @list: list to use
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 *
David S. Miller8728b832005-08-09 19:25:21 -07002172 * Place a packet before a given packet in a list. The list locks are
2173 * taken and this function is atomic with respect to other list locked
2174 * calls.
2175 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 * A buffer cannot be placed on two lists at the same time.
2177 */
David S. Miller8728b832005-08-09 19:25:21 -07002178void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179{
2180 unsigned long flags;
2181
David S. Miller8728b832005-08-09 19:25:21 -07002182 spin_lock_irqsave(&list->lock, flags);
2183 __skb_insert(newsk, old->prev, old, list);
2184 spin_unlock_irqrestore(&list->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002186EXPORT_SYMBOL(skb_insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188static inline void skb_split_inside_header(struct sk_buff *skb,
2189 struct sk_buff* skb1,
2190 const u32 len, const int pos)
2191{
2192 int i;
2193
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03002194 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2195 pos - len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 /* And move data appendix as is. */
2197 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2198 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2199
2200 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2201 skb_shinfo(skb)->nr_frags = 0;
2202 skb1->data_len = skb->data_len;
2203 skb1->len += skb1->data_len;
2204 skb->data_len = 0;
2205 skb->len = len;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07002206 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207}
2208
2209static inline void skb_split_no_header(struct sk_buff *skb,
2210 struct sk_buff* skb1,
2211 const u32 len, int pos)
2212{
2213 int i, k = 0;
2214 const int nfrags = skb_shinfo(skb)->nr_frags;
2215
2216 skb_shinfo(skb)->nr_frags = 0;
2217 skb1->len = skb1->data_len = skb->len - len;
2218 skb->len = len;
2219 skb->data_len = len - pos;
2220
2221 for (i = 0; i < nfrags; i++) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00002222 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223
2224 if (pos + size > len) {
2225 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2226
2227 if (pos < len) {
2228 /* Split frag.
2229 * We have two variants in this case:
2230 * 1. Move all the frag to the second
2231 * part, if it is possible. F.e.
2232 * this approach is mandatory for TUX,
2233 * where splitting is expensive.
2234 * 2. Split is accurately. We make this.
2235 */
Ian Campbellea2ab692011-08-22 23:44:58 +00002236 skb_frag_ref(skb, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
Eric Dumazet9e903e02011-10-18 21:00:24 +00002238 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2239 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 skb_shinfo(skb)->nr_frags++;
2241 }
2242 k++;
2243 } else
2244 skb_shinfo(skb)->nr_frags++;
2245 pos += size;
2246 }
2247 skb_shinfo(skb1)->nr_frags = k;
2248}
2249
2250/**
2251 * skb_split - Split fragmented skb to two parts at length len.
2252 * @skb: the buffer to split
2253 * @skb1: the buffer to receive the second part
2254 * @len: new length for skb
2255 */
2256void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2257{
2258 int pos = skb_headlen(skb);
2259
2260 if (len < pos) /* Split line is inside header. */
2261 skb_split_inside_header(skb, skb1, len, pos);
2262 else /* Second chunk has no header, nothing to copy. */
2263 skb_split_no_header(skb, skb1, len, pos);
2264}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002265EXPORT_SYMBOL(skb_split);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266
Ilpo Järvinen9f782db2008-11-25 13:57:01 -08002267/* Shifting from/to a cloned skb is a no-go.
2268 *
2269 * Caller cannot keep skb_shinfo related pointers past calling here!
2270 */
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002271static int skb_prepare_for_shift(struct sk_buff *skb)
2272{
Ilpo Järvinen0ace2852008-11-24 21:30:21 -08002273 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002274}
2275
2276/**
2277 * skb_shift - Shifts paged data partially from skb to another
2278 * @tgt: buffer into which tail data gets added
2279 * @skb: buffer from which the paged data comes from
2280 * @shiftlen: shift up to this many bytes
2281 *
2282 * Attempts to shift up to shiftlen worth of bytes, which may be less than
Feng King20e994a2011-11-21 01:47:11 +00002283 * the length of the skb, from skb to tgt. Returns number bytes shifted.
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002284 * It's up to caller to free skb if everything was shifted.
2285 *
2286 * If @tgt runs out of frags, the whole operation is aborted.
2287 *
2288 * Skb cannot include anything else but paged data while tgt is allowed
2289 * to have non-paged data as well.
2290 *
2291 * TODO: full sized shift could be optimized but that would need
2292 * specialized skb free'er to handle frags without up-to-date nr_frags.
2293 */
2294int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2295{
2296 int from, to, merge, todo;
2297 struct skb_frag_struct *fragfrom, *fragto;
2298
2299 BUG_ON(shiftlen > skb->len);
2300 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2301
2302 todo = shiftlen;
2303 from = 0;
2304 to = skb_shinfo(tgt)->nr_frags;
2305 fragfrom = &skb_shinfo(skb)->frags[from];
2306
2307 /* Actual merge is delayed until the point when we know we can
2308 * commit all, so that we don't have to undo partial changes
2309 */
2310 if (!to ||
Ian Campbellea2ab692011-08-22 23:44:58 +00002311 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2312 fragfrom->page_offset)) {
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002313 merge = -1;
2314 } else {
2315 merge = to - 1;
2316
Eric Dumazet9e903e02011-10-18 21:00:24 +00002317 todo -= skb_frag_size(fragfrom);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002318 if (todo < 0) {
2319 if (skb_prepare_for_shift(skb) ||
2320 skb_prepare_for_shift(tgt))
2321 return 0;
2322
Ilpo Järvinen9f782db2008-11-25 13:57:01 -08002323 /* All previous frag pointers might be stale! */
2324 fragfrom = &skb_shinfo(skb)->frags[from];
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002325 fragto = &skb_shinfo(tgt)->frags[merge];
2326
Eric Dumazet9e903e02011-10-18 21:00:24 +00002327 skb_frag_size_add(fragto, shiftlen);
2328 skb_frag_size_sub(fragfrom, shiftlen);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002329 fragfrom->page_offset += shiftlen;
2330
2331 goto onlymerged;
2332 }
2333
2334 from++;
2335 }
2336
2337 /* Skip full, not-fitting skb to avoid expensive operations */
2338 if ((shiftlen == skb->len) &&
2339 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2340 return 0;
2341
2342 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2343 return 0;
2344
2345 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2346 if (to == MAX_SKB_FRAGS)
2347 return 0;
2348
2349 fragfrom = &skb_shinfo(skb)->frags[from];
2350 fragto = &skb_shinfo(tgt)->frags[to];
2351
Eric Dumazet9e903e02011-10-18 21:00:24 +00002352 if (todo >= skb_frag_size(fragfrom)) {
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002353 *fragto = *fragfrom;
Eric Dumazet9e903e02011-10-18 21:00:24 +00002354 todo -= skb_frag_size(fragfrom);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002355 from++;
2356 to++;
2357
2358 } else {
Ian Campbellea2ab692011-08-22 23:44:58 +00002359 __skb_frag_ref(fragfrom);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002360 fragto->page = fragfrom->page;
2361 fragto->page_offset = fragfrom->page_offset;
Eric Dumazet9e903e02011-10-18 21:00:24 +00002362 skb_frag_size_set(fragto, todo);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002363
2364 fragfrom->page_offset += todo;
Eric Dumazet9e903e02011-10-18 21:00:24 +00002365 skb_frag_size_sub(fragfrom, todo);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002366 todo = 0;
2367
2368 to++;
2369 break;
2370 }
2371 }
2372
2373 /* Ready to "commit" this state change to tgt */
2374 skb_shinfo(tgt)->nr_frags = to;
2375
2376 if (merge >= 0) {
2377 fragfrom = &skb_shinfo(skb)->frags[0];
2378 fragto = &skb_shinfo(tgt)->frags[merge];
2379
Eric Dumazet9e903e02011-10-18 21:00:24 +00002380 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
Ian Campbellea2ab692011-08-22 23:44:58 +00002381 __skb_frag_unref(fragfrom);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002382 }
2383
2384 /* Reposition in the original skb */
2385 to = 0;
2386 while (from < skb_shinfo(skb)->nr_frags)
2387 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2388 skb_shinfo(skb)->nr_frags = to;
2389
2390 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2391
2392onlymerged:
2393 /* Most likely the tgt won't ever need its checksum anymore, skb on
2394 * the other hand might need it if it needs to be resent
2395 */
2396 tgt->ip_summed = CHECKSUM_PARTIAL;
2397 skb->ip_summed = CHECKSUM_PARTIAL;
2398
2399 /* Yak, is it really working this way? Some helper please? */
2400 skb->len -= shiftlen;
2401 skb->data_len -= shiftlen;
2402 skb->truesize -= shiftlen;
2403 tgt->len += shiftlen;
2404 tgt->data_len += shiftlen;
2405 tgt->truesize += shiftlen;
2406
2407 return shiftlen;
2408}
2409
Thomas Graf677e90e2005-06-23 20:59:51 -07002410/**
2411 * skb_prepare_seq_read - Prepare a sequential read of skb data
2412 * @skb: the buffer to read
2413 * @from: lower offset of data to be read
2414 * @to: upper offset of data to be read
2415 * @st: state variable
2416 *
2417 * Initializes the specified state variable. Must be called before
2418 * invoking skb_seq_read() for the first time.
2419 */
2420void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2421 unsigned int to, struct skb_seq_state *st)
2422{
2423 st->lower_offset = from;
2424 st->upper_offset = to;
2425 st->root_skb = st->cur_skb = skb;
2426 st->frag_idx = st->stepped_offset = 0;
2427 st->frag_data = NULL;
2428}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002429EXPORT_SYMBOL(skb_prepare_seq_read);
Thomas Graf677e90e2005-06-23 20:59:51 -07002430
2431/**
2432 * skb_seq_read - Sequentially read skb data
2433 * @consumed: number of bytes consumed by the caller so far
2434 * @data: destination pointer for data to be returned
2435 * @st: state variable
2436 *
2437 * Reads a block of skb data at &consumed relative to the
2438 * lower offset specified to skb_prepare_seq_read(). Assigns
2439 * the head of the data block to &data and returns the length
2440 * of the block or 0 if the end of the skb data or the upper
2441 * offset has been reached.
2442 *
2443 * The caller is not required to consume all of the data
2444 * returned, i.e. &consumed is typically set to the number
2445 * of bytes already consumed and the next call to
2446 * skb_seq_read() will return the remaining part of the block.
2447 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002448 * Note 1: The size of each block of data returned can be arbitrary,
Thomas Graf677e90e2005-06-23 20:59:51 -07002449 * this limitation is the cost for zerocopy seqeuental
2450 * reads of potentially non linear data.
2451 *
Randy Dunlapbc2cda12008-02-13 15:03:25 -08002452 * Note 2: Fragment lists within fragments are not implemented
Thomas Graf677e90e2005-06-23 20:59:51 -07002453 * at the moment, state->root_skb could be replaced with
2454 * a stack for this purpose.
2455 */
2456unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2457 struct skb_seq_state *st)
2458{
2459 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2460 skb_frag_t *frag;
2461
2462 if (unlikely(abs_offset >= st->upper_offset))
2463 return 0;
2464
2465next_skb:
Herbert Xu95e3b242009-01-29 16:07:52 -08002466 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
Thomas Graf677e90e2005-06-23 20:59:51 -07002467
Thomas Chenault995b3372009-05-18 21:43:27 -07002468 if (abs_offset < block_limit && !st->frag_data) {
Herbert Xu95e3b242009-01-29 16:07:52 -08002469 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
Thomas Graf677e90e2005-06-23 20:59:51 -07002470 return block_limit - abs_offset;
2471 }
2472
2473 if (st->frag_idx == 0 && !st->frag_data)
2474 st->stepped_offset += skb_headlen(st->cur_skb);
2475
2476 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2477 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
Eric Dumazet9e903e02011-10-18 21:00:24 +00002478 block_limit = skb_frag_size(frag) + st->stepped_offset;
Thomas Graf677e90e2005-06-23 20:59:51 -07002479
2480 if (abs_offset < block_limit) {
2481 if (!st->frag_data)
2482 st->frag_data = kmap_skb_frag(frag);
2483
2484 *data = (u8 *) st->frag_data + frag->page_offset +
2485 (abs_offset - st->stepped_offset);
2486
2487 return block_limit - abs_offset;
2488 }
2489
2490 if (st->frag_data) {
2491 kunmap_skb_frag(st->frag_data);
2492 st->frag_data = NULL;
2493 }
2494
2495 st->frag_idx++;
Eric Dumazet9e903e02011-10-18 21:00:24 +00002496 st->stepped_offset += skb_frag_size(frag);
Thomas Graf677e90e2005-06-23 20:59:51 -07002497 }
2498
Olaf Kirch5b5a60d2007-06-23 23:11:52 -07002499 if (st->frag_data) {
2500 kunmap_skb_frag(st->frag_data);
2501 st->frag_data = NULL;
2502 }
2503
David S. Miller21dc3302010-08-23 00:13:46 -07002504 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
Shyam Iyer71b33462009-01-29 16:12:42 -08002505 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
Thomas Graf677e90e2005-06-23 20:59:51 -07002506 st->frag_idx = 0;
2507 goto next_skb;
Shyam Iyer71b33462009-01-29 16:12:42 -08002508 } else if (st->cur_skb->next) {
2509 st->cur_skb = st->cur_skb->next;
Herbert Xu95e3b242009-01-29 16:07:52 -08002510 st->frag_idx = 0;
Thomas Graf677e90e2005-06-23 20:59:51 -07002511 goto next_skb;
2512 }
2513
2514 return 0;
2515}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002516EXPORT_SYMBOL(skb_seq_read);
Thomas Graf677e90e2005-06-23 20:59:51 -07002517
2518/**
2519 * skb_abort_seq_read - Abort a sequential read of skb data
2520 * @st: state variable
2521 *
2522 * Must be called if skb_seq_read() was not called until it
2523 * returned 0.
2524 */
2525void skb_abort_seq_read(struct skb_seq_state *st)
2526{
2527 if (st->frag_data)
2528 kunmap_skb_frag(st->frag_data);
2529}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002530EXPORT_SYMBOL(skb_abort_seq_read);
Thomas Graf677e90e2005-06-23 20:59:51 -07002531
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002532#define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2533
2534static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2535 struct ts_config *conf,
2536 struct ts_state *state)
2537{
2538 return skb_seq_read(offset, text, TS_SKB_CB(state));
2539}
2540
2541static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2542{
2543 skb_abort_seq_read(TS_SKB_CB(state));
2544}
2545
2546/**
2547 * skb_find_text - Find a text pattern in skb data
2548 * @skb: the buffer to look in
2549 * @from: search offset
2550 * @to: search limit
2551 * @config: textsearch configuration
2552 * @state: uninitialized textsearch state variable
2553 *
2554 * Finds a pattern in the skb data according to the specified
2555 * textsearch configuration. Use textsearch_next() to retrieve
2556 * subsequent occurrences of the pattern. Returns the offset
2557 * to the first occurrence or UINT_MAX if no match was found.
2558 */
2559unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2560 unsigned int to, struct ts_config *config,
2561 struct ts_state *state)
2562{
Phil Oesterf72b9482006-06-26 00:00:57 -07002563 unsigned int ret;
2564
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002565 config->get_next_block = skb_ts_get_next_block;
2566 config->finish = skb_ts_finish;
2567
2568 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2569
Phil Oesterf72b9482006-06-26 00:00:57 -07002570 ret = textsearch_find(config, state);
2571 return (ret <= to - from ? ret : UINT_MAX);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002572}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002573EXPORT_SYMBOL(skb_find_text);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002574
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002575/**
2576 * skb_append_datato_frags: - append the user data to a skb
2577 * @sk: sock structure
2578 * @skb: skb structure to be appened with user data.
2579 * @getfrag: call back function to be used for getting the user data
2580 * @from: pointer to user message iov
2581 * @length: length of the iov message
2582 *
2583 * Description: This procedure append the user data in the fragment part
2584 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2585 */
2586int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
Martin Waitzdab96302005-12-05 13:40:12 -08002587 int (*getfrag)(void *from, char *to, int offset,
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002588 int len, int odd, struct sk_buff *skb),
2589 void *from, int length)
2590{
2591 int frg_cnt = 0;
2592 skb_frag_t *frag = NULL;
2593 struct page *page = NULL;
2594 int copy, left;
2595 int offset = 0;
2596 int ret;
2597
2598 do {
2599 /* Return error if we don't have space for new frag */
2600 frg_cnt = skb_shinfo(skb)->nr_frags;
2601 if (frg_cnt >= MAX_SKB_FRAGS)
2602 return -EFAULT;
2603
2604 /* allocate a new page for next frag */
2605 page = alloc_pages(sk->sk_allocation, 0);
2606
2607 /* If alloc_page fails just return failure and caller will
2608 * free previous allocated pages by doing kfree_skb()
2609 */
2610 if (page == NULL)
2611 return -ENOMEM;
2612
2613 /* initialize the next frag */
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002614 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2615 skb->truesize += PAGE_SIZE;
2616 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2617
2618 /* get the new initialized frag */
2619 frg_cnt = skb_shinfo(skb)->nr_frags;
2620 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2621
2622 /* copy the user data to page */
2623 left = PAGE_SIZE - frag->page_offset;
2624 copy = (length > left)? left : length;
2625
Eric Dumazet9e903e02011-10-18 21:00:24 +00002626 ret = getfrag(from, skb_frag_address(frag) + skb_frag_size(frag),
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002627 offset, copy, 0, skb);
2628 if (ret < 0)
2629 return -EFAULT;
2630
2631 /* copy was successful so update the size parameters */
Eric Dumazet9e903e02011-10-18 21:00:24 +00002632 skb_frag_size_add(frag, copy);
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002633 skb->len += copy;
2634 skb->data_len += copy;
2635 offset += copy;
2636 length -= copy;
2637
2638 } while (length > 0);
2639
2640 return 0;
2641}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002642EXPORT_SYMBOL(skb_append_datato_frags);
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002643
Herbert Xucbb042f2006-03-20 22:43:56 -08002644/**
2645 * skb_pull_rcsum - pull skb and update receive checksum
2646 * @skb: buffer to update
Herbert Xucbb042f2006-03-20 22:43:56 -08002647 * @len: length of data pulled
2648 *
2649 * This function performs an skb_pull on the packet and updates
Urs Thuermannfee54fa2008-02-12 22:03:25 -08002650 * the CHECKSUM_COMPLETE checksum. It should be used on
Patrick McHardy84fa7932006-08-29 16:44:56 -07002651 * receive path processing instead of skb_pull unless you know
2652 * that the checksum difference is zero (e.g., a valid IP header)
2653 * or you are setting ip_summed to CHECKSUM_NONE.
Herbert Xucbb042f2006-03-20 22:43:56 -08002654 */
2655unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2656{
2657 BUG_ON(len > skb->len);
2658 skb->len -= len;
2659 BUG_ON(skb->len < skb->data_len);
2660 skb_postpull_rcsum(skb, skb->data, len);
2661 return skb->data += len;
2662}
Arnaldo Carvalho de Melof94691a2006-03-20 22:47:55 -08002663EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2664
Herbert Xuf4c50d92006-06-22 03:02:40 -07002665/**
2666 * skb_segment - Perform protocol segmentation on skb.
2667 * @skb: buffer to segment
Herbert Xu576a30e2006-06-27 13:22:38 -07002668 * @features: features for the output path (see dev->features)
Herbert Xuf4c50d92006-06-22 03:02:40 -07002669 *
2670 * This function performs segmentation on the given skb. It returns
Ben Hutchings4c821d72008-04-13 21:52:48 -07002671 * a pointer to the first in a list of new skbs for the segments.
2672 * In case of error it returns ERR_PTR(err).
Herbert Xuf4c50d92006-06-22 03:02:40 -07002673 */
Michał Mirosławc8f44af2011-11-15 15:29:55 +00002674struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features)
Herbert Xuf4c50d92006-06-22 03:02:40 -07002675{
2676 struct sk_buff *segs = NULL;
2677 struct sk_buff *tail = NULL;
Herbert Xu89319d382008-12-15 23:26:06 -08002678 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002679 unsigned int mss = skb_shinfo(skb)->gso_size;
Arnaldo Carvalho de Melo98e399f2007-03-19 15:33:04 -07002680 unsigned int doffset = skb->data - skb_mac_header(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002681 unsigned int offset = doffset;
2682 unsigned int headroom;
2683 unsigned int len;
Michał Mirosław04ed3e72011-01-24 15:32:47 -08002684 int sg = !!(features & NETIF_F_SG);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002685 int nfrags = skb_shinfo(skb)->nr_frags;
2686 int err = -ENOMEM;
2687 int i = 0;
2688 int pos;
2689
2690 __skb_push(skb, doffset);
2691 headroom = skb_headroom(skb);
2692 pos = skb_headlen(skb);
2693
2694 do {
2695 struct sk_buff *nskb;
2696 skb_frag_t *frag;
Herbert Xuc8884ed2006-10-29 15:59:41 -08002697 int hsize;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002698 int size;
2699
2700 len = skb->len - offset;
2701 if (len > mss)
2702 len = mss;
2703
2704 hsize = skb_headlen(skb) - offset;
2705 if (hsize < 0)
2706 hsize = 0;
Herbert Xuc8884ed2006-10-29 15:59:41 -08002707 if (hsize > len || !sg)
2708 hsize = len;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002709
Herbert Xu89319d382008-12-15 23:26:06 -08002710 if (!hsize && i >= nfrags) {
2711 BUG_ON(fskb->len != len);
2712
2713 pos += len;
2714 nskb = skb_clone(fskb, GFP_ATOMIC);
2715 fskb = fskb->next;
2716
2717 if (unlikely(!nskb))
2718 goto err;
2719
2720 hsize = skb_end_pointer(nskb) - nskb->head;
2721 if (skb_cow_head(nskb, doffset + headroom)) {
2722 kfree_skb(nskb);
2723 goto err;
2724 }
2725
2726 nskb->truesize += skb_end_pointer(nskb) - nskb->head -
2727 hsize;
2728 skb_release_head_state(nskb);
2729 __skb_push(nskb, doffset);
2730 } else {
2731 nskb = alloc_skb(hsize + doffset + headroom,
2732 GFP_ATOMIC);
2733
2734 if (unlikely(!nskb))
2735 goto err;
2736
2737 skb_reserve(nskb, headroom);
2738 __skb_put(nskb, doffset);
2739 }
Herbert Xuf4c50d92006-06-22 03:02:40 -07002740
2741 if (segs)
2742 tail->next = nskb;
2743 else
2744 segs = nskb;
2745 tail = nskb;
2746
Herbert Xu6f85a122008-08-15 14:55:02 -07002747 __copy_skb_header(nskb, skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002748 nskb->mac_len = skb->mac_len;
2749
Eric Dumazet3d3be432010-09-01 00:50:51 +00002750 /* nskb and skb might have different headroom */
2751 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2752 nskb->csum_start += skb_headroom(nskb) - headroom;
2753
Arnaldo Carvalho de Melo459a98e2007-03-19 15:30:44 -07002754 skb_reset_mac_header(nskb);
Arnaldo Carvalho de Meloddc7b8e2007-03-15 21:42:27 -03002755 skb_set_network_header(nskb, skb->mac_len);
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -07002756 nskb->transport_header = (nskb->network_header +
2757 skb_network_header_len(skb));
Herbert Xu89319d382008-12-15 23:26:06 -08002758 skb_copy_from_linear_data(skb, nskb->data, doffset);
2759
Herbert Xu2f181852009-03-28 23:39:18 -07002760 if (fskb != skb_shinfo(skb)->frag_list)
Herbert Xu89319d382008-12-15 23:26:06 -08002761 continue;
2762
Herbert Xuf4c50d92006-06-22 03:02:40 -07002763 if (!sg) {
Herbert Xu6f85a122008-08-15 14:55:02 -07002764 nskb->ip_summed = CHECKSUM_NONE;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002765 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2766 skb_put(nskb, len),
2767 len, 0);
2768 continue;
2769 }
2770
2771 frag = skb_shinfo(nskb)->frags;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002772
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03002773 skb_copy_from_linear_data_offset(skb, offset,
2774 skb_put(nskb, hsize), hsize);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002775
Herbert Xu89319d382008-12-15 23:26:06 -08002776 while (pos < offset + len && i < nfrags) {
Herbert Xuf4c50d92006-06-22 03:02:40 -07002777 *frag = skb_shinfo(skb)->frags[i];
Ian Campbellea2ab692011-08-22 23:44:58 +00002778 __skb_frag_ref(frag);
Eric Dumazet9e903e02011-10-18 21:00:24 +00002779 size = skb_frag_size(frag);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002780
2781 if (pos < offset) {
2782 frag->page_offset += offset - pos;
Eric Dumazet9e903e02011-10-18 21:00:24 +00002783 skb_frag_size_sub(frag, offset - pos);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002784 }
2785
Herbert Xu89319d382008-12-15 23:26:06 -08002786 skb_shinfo(nskb)->nr_frags++;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002787
2788 if (pos + size <= offset + len) {
2789 i++;
2790 pos += size;
2791 } else {
Eric Dumazet9e903e02011-10-18 21:00:24 +00002792 skb_frag_size_sub(frag, pos + size - (offset + len));
Herbert Xu89319d382008-12-15 23:26:06 -08002793 goto skip_fraglist;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002794 }
2795
2796 frag++;
2797 }
2798
Herbert Xu89319d382008-12-15 23:26:06 -08002799 if (pos < offset + len) {
2800 struct sk_buff *fskb2 = fskb;
2801
2802 BUG_ON(pos + fskb->len != offset + len);
2803
2804 pos += fskb->len;
2805 fskb = fskb->next;
2806
2807 if (fskb2->next) {
2808 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2809 if (!fskb2)
2810 goto err;
2811 } else
2812 skb_get(fskb2);
2813
David S. Millerfbb398a2009-06-09 00:18:59 -07002814 SKB_FRAG_ASSERT(nskb);
Herbert Xu89319d382008-12-15 23:26:06 -08002815 skb_shinfo(nskb)->frag_list = fskb2;
2816 }
2817
2818skip_fraglist:
Herbert Xuf4c50d92006-06-22 03:02:40 -07002819 nskb->data_len = len - hsize;
2820 nskb->len += nskb->data_len;
2821 nskb->truesize += nskb->data_len;
2822 } while ((offset += len) < skb->len);
2823
2824 return segs;
2825
2826err:
2827 while ((skb = segs)) {
2828 segs = skb->next;
Patrick McHardyb08d5842007-02-27 09:57:37 -08002829 kfree_skb(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002830 }
2831 return ERR_PTR(err);
2832}
Herbert Xuf4c50d92006-06-22 03:02:40 -07002833EXPORT_SYMBOL_GPL(skb_segment);
2834
Herbert Xu71d93b32008-12-15 23:42:33 -08002835int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2836{
2837 struct sk_buff *p = *head;
2838 struct sk_buff *nskb;
Herbert Xu9aaa1562009-05-26 18:50:33 +00002839 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2840 struct skb_shared_info *pinfo = skb_shinfo(p);
Herbert Xu71d93b32008-12-15 23:42:33 -08002841 unsigned int headroom;
Herbert Xu86911732009-01-29 14:19:50 +00002842 unsigned int len = skb_gro_len(skb);
Herbert Xu67147ba2009-05-26 18:50:22 +00002843 unsigned int offset = skb_gro_offset(skb);
2844 unsigned int headlen = skb_headlen(skb);
Herbert Xu71d93b32008-12-15 23:42:33 -08002845
Herbert Xu86911732009-01-29 14:19:50 +00002846 if (p->len + len >= 65536)
Herbert Xu71d93b32008-12-15 23:42:33 -08002847 return -E2BIG;
2848
Herbert Xu9aaa1562009-05-26 18:50:33 +00002849 if (pinfo->frag_list)
Herbert Xu71d93b32008-12-15 23:42:33 -08002850 goto merge;
Herbert Xu67147ba2009-05-26 18:50:22 +00002851 else if (headlen <= offset) {
Herbert Xu42da6992009-05-26 18:50:19 +00002852 skb_frag_t *frag;
Herbert Xu66e92fc2009-05-26 18:50:32 +00002853 skb_frag_t *frag2;
Herbert Xu9aaa1562009-05-26 18:50:33 +00002854 int i = skbinfo->nr_frags;
2855 int nr_frags = pinfo->nr_frags + i;
Herbert Xu42da6992009-05-26 18:50:19 +00002856
Herbert Xu66e92fc2009-05-26 18:50:32 +00002857 offset -= headlen;
2858
2859 if (nr_frags > MAX_SKB_FRAGS)
Herbert Xu81705ad2009-01-29 14:19:51 +00002860 return -E2BIG;
2861
Herbert Xu9aaa1562009-05-26 18:50:33 +00002862 pinfo->nr_frags = nr_frags;
2863 skbinfo->nr_frags = 0;
Herbert Xuf5572062009-01-14 20:40:03 -08002864
Herbert Xu9aaa1562009-05-26 18:50:33 +00002865 frag = pinfo->frags + nr_frags;
2866 frag2 = skbinfo->frags + i;
Herbert Xu66e92fc2009-05-26 18:50:32 +00002867 do {
2868 *--frag = *--frag2;
2869 } while (--i);
2870
2871 frag->page_offset += offset;
Eric Dumazet9e903e02011-10-18 21:00:24 +00002872 skb_frag_size_sub(frag, offset);
Herbert Xu66e92fc2009-05-26 18:50:32 +00002873
Herbert Xuf5572062009-01-14 20:40:03 -08002874 skb->truesize -= skb->data_len;
2875 skb->len -= skb->data_len;
2876 skb->data_len = 0;
2877
Herbert Xu5d38a072009-01-04 16:13:40 -08002878 NAPI_GRO_CB(skb)->free = 1;
2879 goto done;
Herbert Xu69c0cab2009-11-17 05:18:18 -08002880 } else if (skb_gro_len(p) != pinfo->gso_size)
2881 return -E2BIG;
Herbert Xu71d93b32008-12-15 23:42:33 -08002882
2883 headroom = skb_headroom(p);
Eric Dumazet3d3be432010-09-01 00:50:51 +00002884 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
Herbert Xu71d93b32008-12-15 23:42:33 -08002885 if (unlikely(!nskb))
2886 return -ENOMEM;
2887
2888 __copy_skb_header(nskb, p);
2889 nskb->mac_len = p->mac_len;
2890
2891 skb_reserve(nskb, headroom);
Herbert Xu86911732009-01-29 14:19:50 +00002892 __skb_put(nskb, skb_gro_offset(p));
Herbert Xu71d93b32008-12-15 23:42:33 -08002893
Herbert Xu86911732009-01-29 14:19:50 +00002894 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
Herbert Xu71d93b32008-12-15 23:42:33 -08002895 skb_set_network_header(nskb, skb_network_offset(p));
2896 skb_set_transport_header(nskb, skb_transport_offset(p));
2897
Herbert Xu86911732009-01-29 14:19:50 +00002898 __skb_pull(p, skb_gro_offset(p));
2899 memcpy(skb_mac_header(nskb), skb_mac_header(p),
2900 p->data - skb_mac_header(p));
Herbert Xu71d93b32008-12-15 23:42:33 -08002901
2902 *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2903 skb_shinfo(nskb)->frag_list = p;
Herbert Xu9aaa1562009-05-26 18:50:33 +00002904 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
Herbert Xu622e0ca2010-05-20 23:07:56 -07002905 pinfo->gso_size = 0;
Herbert Xu71d93b32008-12-15 23:42:33 -08002906 skb_header_release(p);
2907 nskb->prev = p;
2908
2909 nskb->data_len += p->len;
Eric Dumazetde8261c2012-02-13 04:09:20 +00002910 nskb->truesize += p->truesize;
Herbert Xu71d93b32008-12-15 23:42:33 -08002911 nskb->len += p->len;
2912
2913 *head = nskb;
2914 nskb->next = p->next;
2915 p->next = NULL;
2916
2917 p = nskb;
2918
2919merge:
Eric Dumazetde8261c2012-02-13 04:09:20 +00002920 p->truesize += skb->truesize - len;
Herbert Xu67147ba2009-05-26 18:50:22 +00002921 if (offset > headlen) {
Michal Schmidtd1dc7ab2011-01-24 12:08:48 +00002922 unsigned int eat = offset - headlen;
2923
2924 skbinfo->frags[0].page_offset += eat;
Eric Dumazet9e903e02011-10-18 21:00:24 +00002925 skb_frag_size_sub(&skbinfo->frags[0], eat);
Michal Schmidtd1dc7ab2011-01-24 12:08:48 +00002926 skb->data_len -= eat;
2927 skb->len -= eat;
Herbert Xu67147ba2009-05-26 18:50:22 +00002928 offset = headlen;
Herbert Xu56035022009-02-05 21:26:52 -08002929 }
2930
Herbert Xu67147ba2009-05-26 18:50:22 +00002931 __skb_pull(skb, offset);
Herbert Xu56035022009-02-05 21:26:52 -08002932
Herbert Xu71d93b32008-12-15 23:42:33 -08002933 p->prev->next = skb;
2934 p->prev = skb;
2935 skb_header_release(skb);
2936
Herbert Xu5d38a072009-01-04 16:13:40 -08002937done:
2938 NAPI_GRO_CB(p)->count++;
Herbert Xu37fe4732009-01-17 19:48:13 +00002939 p->data_len += len;
2940 p->truesize += len;
2941 p->len += len;
Herbert Xu71d93b32008-12-15 23:42:33 -08002942
2943 NAPI_GRO_CB(skb)->same_flow = 1;
2944 return 0;
2945}
2946EXPORT_SYMBOL_GPL(skb_gro_receive);
2947
Linus Torvalds1da177e2005-04-16 15:20:36 -07002948void __init skb_init(void)
2949{
2950 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2951 sizeof(struct sk_buff),
2952 0,
Alexey Dobriyane5d679f332006-08-26 19:25:52 -07002953 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
Paul Mundt20c2df82007-07-20 10:11:58 +09002954 NULL);
David S. Millerd179cd12005-08-17 14:57:30 -07002955 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2956 (2*sizeof(struct sk_buff)) +
2957 sizeof(atomic_t),
2958 0,
Alexey Dobriyane5d679f332006-08-26 19:25:52 -07002959 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
Paul Mundt20c2df82007-07-20 10:11:58 +09002960 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002961}
2962
David Howells716ea3a2007-04-02 20:19:53 -07002963/**
2964 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2965 * @skb: Socket buffer containing the buffers to be mapped
2966 * @sg: The scatter-gather list to map into
2967 * @offset: The offset into the buffer's contents to start mapping
2968 * @len: Length of buffer space to be mapped
2969 *
2970 * Fill the specified scatter-gather list with mappings/pointers into a
2971 * region of the buffer space attached to a socket buffer.
2972 */
David S. Miller51c739d2007-10-30 21:29:29 -07002973static int
2974__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
David Howells716ea3a2007-04-02 20:19:53 -07002975{
David S. Miller1a028e52007-04-27 15:21:23 -07002976 int start = skb_headlen(skb);
2977 int i, copy = start - offset;
David S. Millerfbb398a2009-06-09 00:18:59 -07002978 struct sk_buff *frag_iter;
David Howells716ea3a2007-04-02 20:19:53 -07002979 int elt = 0;
2980
2981 if (copy > 0) {
2982 if (copy > len)
2983 copy = len;
Jens Axboe642f1492007-10-24 11:20:47 +02002984 sg_set_buf(sg, skb->data + offset, copy);
David Howells716ea3a2007-04-02 20:19:53 -07002985 elt++;
2986 if ((len -= copy) == 0)
2987 return elt;
2988 offset += copy;
2989 }
2990
2991 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07002992 int end;
David Howells716ea3a2007-04-02 20:19:53 -07002993
Ilpo Järvinen547b7922008-07-25 21:43:18 -07002994 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07002995
Eric Dumazet9e903e02011-10-18 21:00:24 +00002996 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
David Howells716ea3a2007-04-02 20:19:53 -07002997 if ((copy = end - offset) > 0) {
2998 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2999
3000 if (copy > len)
3001 copy = len;
Ian Campbellea2ab692011-08-22 23:44:58 +00003002 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
Jens Axboe642f1492007-10-24 11:20:47 +02003003 frag->page_offset+offset-start);
David Howells716ea3a2007-04-02 20:19:53 -07003004 elt++;
3005 if (!(len -= copy))
3006 return elt;
3007 offset += copy;
3008 }
David S. Miller1a028e52007-04-27 15:21:23 -07003009 start = end;
David Howells716ea3a2007-04-02 20:19:53 -07003010 }
3011
David S. Millerfbb398a2009-06-09 00:18:59 -07003012 skb_walk_frags(skb, frag_iter) {
3013 int end;
David Howells716ea3a2007-04-02 20:19:53 -07003014
David S. Millerfbb398a2009-06-09 00:18:59 -07003015 WARN_ON(start > offset + len);
David Howells716ea3a2007-04-02 20:19:53 -07003016
David S. Millerfbb398a2009-06-09 00:18:59 -07003017 end = start + frag_iter->len;
3018 if ((copy = end - offset) > 0) {
3019 if (copy > len)
3020 copy = len;
3021 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3022 copy);
3023 if ((len -= copy) == 0)
3024 return elt;
3025 offset += copy;
David Howells716ea3a2007-04-02 20:19:53 -07003026 }
David S. Millerfbb398a2009-06-09 00:18:59 -07003027 start = end;
David Howells716ea3a2007-04-02 20:19:53 -07003028 }
3029 BUG_ON(len);
3030 return elt;
3031}
3032
David S. Miller51c739d2007-10-30 21:29:29 -07003033int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3034{
3035 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3036
Jens Axboec46f2332007-10-31 12:06:37 +01003037 sg_mark_end(&sg[nsg - 1]);
David S. Miller51c739d2007-10-30 21:29:29 -07003038
3039 return nsg;
3040}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08003041EXPORT_SYMBOL_GPL(skb_to_sgvec);
David S. Miller51c739d2007-10-30 21:29:29 -07003042
David Howells716ea3a2007-04-02 20:19:53 -07003043/**
3044 * skb_cow_data - Check that a socket buffer's data buffers are writable
3045 * @skb: The socket buffer to check.
3046 * @tailbits: Amount of trailing space to be added
3047 * @trailer: Returned pointer to the skb where the @tailbits space begins
3048 *
3049 * Make sure that the data buffers attached to a socket buffer are
3050 * writable. If they are not, private copies are made of the data buffers
3051 * and the socket buffer is set to use these instead.
3052 *
3053 * If @tailbits is given, make sure that there is space to write @tailbits
3054 * bytes of data beyond current end of socket buffer. @trailer will be
3055 * set to point to the skb in which this space begins.
3056 *
3057 * The number of scatterlist elements required to completely map the
3058 * COW'd and extended socket buffer will be returned.
3059 */
3060int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3061{
3062 int copyflag;
3063 int elt;
3064 struct sk_buff *skb1, **skb_p;
3065
3066 /* If skb is cloned or its head is paged, reallocate
3067 * head pulling out all the pages (pages are considered not writable
3068 * at the moment even if they are anonymous).
3069 */
3070 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3071 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3072 return -ENOMEM;
3073
3074 /* Easy case. Most of packets will go this way. */
David S. Miller21dc3302010-08-23 00:13:46 -07003075 if (!skb_has_frag_list(skb)) {
David Howells716ea3a2007-04-02 20:19:53 -07003076 /* A little of trouble, not enough of space for trailer.
3077 * This should not happen, when stack is tuned to generate
3078 * good frames. OK, on miss we reallocate and reserve even more
3079 * space, 128 bytes is fair. */
3080
3081 if (skb_tailroom(skb) < tailbits &&
3082 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3083 return -ENOMEM;
3084
3085 /* Voila! */
3086 *trailer = skb;
3087 return 1;
3088 }
3089
3090 /* Misery. We are in troubles, going to mincer fragments... */
3091
3092 elt = 1;
3093 skb_p = &skb_shinfo(skb)->frag_list;
3094 copyflag = 0;
3095
3096 while ((skb1 = *skb_p) != NULL) {
3097 int ntail = 0;
3098
3099 /* The fragment is partially pulled by someone,
3100 * this can happen on input. Copy it and everything
3101 * after it. */
3102
3103 if (skb_shared(skb1))
3104 copyflag = 1;
3105
3106 /* If the skb is the last, worry about trailer. */
3107
3108 if (skb1->next == NULL && tailbits) {
3109 if (skb_shinfo(skb1)->nr_frags ||
David S. Miller21dc3302010-08-23 00:13:46 -07003110 skb_has_frag_list(skb1) ||
David Howells716ea3a2007-04-02 20:19:53 -07003111 skb_tailroom(skb1) < tailbits)
3112 ntail = tailbits + 128;
3113 }
3114
3115 if (copyflag ||
3116 skb_cloned(skb1) ||
3117 ntail ||
3118 skb_shinfo(skb1)->nr_frags ||
David S. Miller21dc3302010-08-23 00:13:46 -07003119 skb_has_frag_list(skb1)) {
David Howells716ea3a2007-04-02 20:19:53 -07003120 struct sk_buff *skb2;
3121
3122 /* Fuck, we are miserable poor guys... */
3123 if (ntail == 0)
3124 skb2 = skb_copy(skb1, GFP_ATOMIC);
3125 else
3126 skb2 = skb_copy_expand(skb1,
3127 skb_headroom(skb1),
3128 ntail,
3129 GFP_ATOMIC);
3130 if (unlikely(skb2 == NULL))
3131 return -ENOMEM;
3132
3133 if (skb1->sk)
3134 skb_set_owner_w(skb2, skb1->sk);
3135
3136 /* Looking around. Are we still alive?
3137 * OK, link new skb, drop old one */
3138
3139 skb2->next = skb1->next;
3140 *skb_p = skb2;
3141 kfree_skb(skb1);
3142 skb1 = skb2;
3143 }
3144 elt++;
3145 *trailer = skb1;
3146 skb_p = &skb1->next;
3147 }
3148
3149 return elt;
3150}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08003151EXPORT_SYMBOL_GPL(skb_cow_data);
David Howells716ea3a2007-04-02 20:19:53 -07003152
Eric Dumazetb1faf562010-05-31 23:44:05 -07003153static void sock_rmem_free(struct sk_buff *skb)
3154{
3155 struct sock *sk = skb->sk;
3156
3157 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3158}
3159
3160/*
3161 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3162 */
3163int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3164{
3165 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3166 (unsigned)sk->sk_rcvbuf)
3167 return -ENOMEM;
3168
3169 skb_orphan(skb);
3170 skb->sk = sk;
3171 skb->destructor = sock_rmem_free;
3172 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3173
Eric Dumazetabb57ea2011-05-18 02:21:31 -04003174 /* before exiting rcu section, make sure dst is refcounted */
3175 skb_dst_force(skb);
3176
Eric Dumazetb1faf562010-05-31 23:44:05 -07003177 skb_queue_tail(&sk->sk_error_queue, skb);
3178 if (!sock_flag(sk, SOCK_DEAD))
3179 sk->sk_data_ready(sk, skb->len);
3180 return 0;
3181}
3182EXPORT_SYMBOL(sock_queue_err_skb);
3183
Patrick Ohlyac45f602009-02-12 05:03:37 +00003184void skb_tstamp_tx(struct sk_buff *orig_skb,
3185 struct skb_shared_hwtstamps *hwtstamps)
3186{
3187 struct sock *sk = orig_skb->sk;
3188 struct sock_exterr_skb *serr;
3189 struct sk_buff *skb;
3190 int err;
3191
3192 if (!sk)
3193 return;
3194
3195 skb = skb_clone(orig_skb, GFP_ATOMIC);
3196 if (!skb)
3197 return;
3198
3199 if (hwtstamps) {
3200 *skb_hwtstamps(skb) =
3201 *hwtstamps;
3202 } else {
3203 /*
3204 * no hardware time stamps available,
Oliver Hartkopp2244d072010-08-17 08:59:14 +00003205 * so keep the shared tx_flags and only
Patrick Ohlyac45f602009-02-12 05:03:37 +00003206 * store software time stamp
3207 */
3208 skb->tstamp = ktime_get_real();
3209 }
3210
3211 serr = SKB_EXT_ERR(skb);
3212 memset(serr, 0, sizeof(*serr));
3213 serr->ee.ee_errno = ENOMSG;
3214 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
Eric Dumazet29030372010-05-29 00:20:48 -07003215
Patrick Ohlyac45f602009-02-12 05:03:37 +00003216 err = sock_queue_err_skb(sk, skb);
Eric Dumazet29030372010-05-29 00:20:48 -07003217
Patrick Ohlyac45f602009-02-12 05:03:37 +00003218 if (err)
3219 kfree_skb(skb);
3220}
3221EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3222
Johannes Berg6e3e9392011-11-09 10:15:42 +01003223void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3224{
3225 struct sock *sk = skb->sk;
3226 struct sock_exterr_skb *serr;
3227 int err;
3228
3229 skb->wifi_acked_valid = 1;
3230 skb->wifi_acked = acked;
3231
3232 serr = SKB_EXT_ERR(skb);
3233 memset(serr, 0, sizeof(*serr));
3234 serr->ee.ee_errno = ENOMSG;
3235 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3236
3237 err = sock_queue_err_skb(sk, skb);
3238 if (err)
3239 kfree_skb(skb);
3240}
3241EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3242
Patrick Ohlyac45f602009-02-12 05:03:37 +00003243
Rusty Russellf35d9d82008-02-04 23:49:54 -05003244/**
3245 * skb_partial_csum_set - set up and verify partial csum values for packet
3246 * @skb: the skb to set
3247 * @start: the number of bytes after skb->data to start checksumming.
3248 * @off: the offset from start to place the checksum.
3249 *
3250 * For untrusted partially-checksummed packets, we need to make sure the values
3251 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3252 *
3253 * This function checks and sets those values and skb->ip_summed: if this
3254 * returns false you should drop the packet.
3255 */
3256bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3257{
Herbert Xu5ff8dda2009-06-04 01:22:01 +00003258 if (unlikely(start > skb_headlen(skb)) ||
3259 unlikely((int)start + off > skb_headlen(skb) - 2)) {
Rusty Russellf35d9d82008-02-04 23:49:54 -05003260 if (net_ratelimit())
3261 printk(KERN_WARNING
3262 "bad partial csum: csum=%u/%u len=%u\n",
Herbert Xu5ff8dda2009-06-04 01:22:01 +00003263 start, off, skb_headlen(skb));
Rusty Russellf35d9d82008-02-04 23:49:54 -05003264 return false;
3265 }
3266 skb->ip_summed = CHECKSUM_PARTIAL;
3267 skb->csum_start = skb_headroom(skb) + start;
3268 skb->csum_offset = off;
3269 return true;
3270}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08003271EXPORT_SYMBOL_GPL(skb_partial_csum_set);
Rusty Russellf35d9d82008-02-04 23:49:54 -05003272
Ben Hutchings4497b072008-06-19 16:22:28 -07003273void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3274{
3275 if (net_ratelimit())
3276 pr_warning("%s: received packets cannot be forwarded"
3277 " while LRO is enabled\n", skb->dev->name);
3278}
Ben Hutchings4497b072008-06-19 16:22:28 -07003279EXPORT_SYMBOL(__skb_warn_lro_forwarding);