blob: 46cbd28f40f9698bfbfad886047223c4bba16df2 [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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 size = SKB_DATA_ALIGN(size);
Christoph Hellwigb30973f2006-12-06 20:32:36 -0800188 data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info),
189 gfp_mask, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 if (!data)
191 goto nodata;
Eric Dumazetec7d2f22010-05-05 01:07:37 -0700192 prefetchw(data + size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
Arnaldo Carvalho de Meloca0605a2007-03-19 10:48:59 -0300194 /*
Johannes Bergc8005782008-05-03 20:56:42 -0700195 * Only clear those fields we need to clear, not those that we will
196 * actually initialise below. Hence, don't put any more fields after
197 * the tail pointer in struct sk_buff!
Arnaldo Carvalho de Meloca0605a2007-03-19 10:48:59 -0300198 */
199 memset(skb, 0, offsetof(struct sk_buff, tail));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 skb->truesize = size + sizeof(struct sk_buff);
201 atomic_set(&skb->users, 1);
202 skb->head = data;
203 skb->data = data;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700204 skb_reset_tail_pointer(skb);
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700205 skb->end = skb->tail + size;
Stephen Hemminger19633e12009-06-17 05:23:27 +0000206#ifdef NET_SKBUFF_DATA_USES_OFFSET
207 skb->mac_header = ~0U;
208#endif
209
Benjamin LaHaise4947d3e2006-01-03 14:06:50 -0800210 /* make sure we initialize shinfo sequentially */
211 shinfo = skb_shinfo(skb);
Eric Dumazetec7d2f22010-05-05 01:07:37 -0700212 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
Benjamin LaHaise4947d3e2006-01-03 14:06:50 -0800213 atomic_set(&shinfo->dataref, 1);
Eric Dumazetc2aa3662011-01-25 23:18:38 +0000214 kmemcheck_annotate_variable(shinfo->destructor_arg);
Benjamin LaHaise4947d3e2006-01-03 14:06:50 -0800215
David S. Millerd179cd12005-08-17 14:57:30 -0700216 if (fclone) {
217 struct sk_buff *child = skb + 1;
218 atomic_t *fclone_ref = (atomic_t *) (child + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200220 kmemcheck_annotate_bitfield(child, flags1);
221 kmemcheck_annotate_bitfield(child, flags2);
David S. Millerd179cd12005-08-17 14:57:30 -0700222 skb->fclone = SKB_FCLONE_ORIG;
223 atomic_set(fclone_ref, 1);
224
225 child->fclone = SKB_FCLONE_UNAVAILABLE;
226 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227out:
228 return skb;
229nodata:
Herbert Xu8798b3f2006-01-23 16:32:45 -0800230 kmem_cache_free(cache, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 skb = NULL;
232 goto out;
233}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800234EXPORT_SYMBOL(__alloc_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235
236/**
Christoph Hellwig8af27452006-07-31 22:35:23 -0700237 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
238 * @dev: network device to receive on
239 * @length: length to allocate
240 * @gfp_mask: get_free_pages mask, passed to alloc_skb
241 *
242 * Allocate a new &sk_buff and assign it a usage count of one. The
243 * buffer has unspecified headroom built in. Users should allocate
244 * the headroom they think they need without accounting for the
245 * built in space. The built in space is used for optimisations.
246 *
247 * %NULL is returned if there is no free memory.
248 */
249struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
250 unsigned int length, gfp_t gfp_mask)
251{
252 struct sk_buff *skb;
253
Eric Dumazet564824b2010-10-11 19:05:25 +0000254 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, NUMA_NO_NODE);
Christoph Hellwig7b2e4972006-08-07 16:09:04 -0700255 if (likely(skb)) {
Christoph Hellwig8af27452006-07-31 22:35:23 -0700256 skb_reserve(skb, NET_SKB_PAD);
Christoph Hellwig7b2e4972006-08-07 16:09:04 -0700257 skb->dev = dev;
258 }
Christoph Hellwig8af27452006-07-31 22:35:23 -0700259 return skb;
260}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800261EXPORT_SYMBOL(__netdev_alloc_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262
Peter Zijlstra654bed12008-10-07 14:22:33 -0700263void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
264 int size)
265{
266 skb_fill_page_desc(skb, i, page, off, size);
267 skb->len += size;
268 skb->data_len += size;
269 skb->truesize += size;
270}
271EXPORT_SYMBOL(skb_add_rx_frag);
272
Ilpo Järvinenf58518e2008-03-27 17:51:31 -0700273/**
274 * dev_alloc_skb - allocate an skbuff for receiving
275 * @length: length to allocate
276 *
277 * Allocate a new &sk_buff and assign it a usage count of one. The
278 * buffer has unspecified headroom built in. Users should allocate
279 * the headroom they think they need without accounting for the
280 * built in space. The built in space is used for optimisations.
281 *
282 * %NULL is returned if there is no free memory. Although this function
283 * allocates memory it can be called from an interrupt.
284 */
285struct sk_buff *dev_alloc_skb(unsigned int length)
286{
Denys Vlasenko1483b872008-03-28 15:57:39 -0700287 /*
288 * There is more code here than it seems:
David S. Millera0f55e02008-03-28 18:22:32 -0700289 * __dev_alloc_skb is an inline
Denys Vlasenko1483b872008-03-28 15:57:39 -0700290 */
Ilpo Järvinenf58518e2008-03-27 17:51:31 -0700291 return __dev_alloc_skb(length, GFP_ATOMIC);
292}
293EXPORT_SYMBOL(dev_alloc_skb);
294
Herbert Xu27b437c2006-07-13 19:26:39 -0700295static void skb_drop_list(struct sk_buff **listp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296{
Herbert Xu27b437c2006-07-13 19:26:39 -0700297 struct sk_buff *list = *listp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298
Herbert Xu27b437c2006-07-13 19:26:39 -0700299 *listp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300
301 do {
302 struct sk_buff *this = list;
303 list = list->next;
304 kfree_skb(this);
305 } while (list);
306}
307
Herbert Xu27b437c2006-07-13 19:26:39 -0700308static inline void skb_drop_fraglist(struct sk_buff *skb)
309{
310 skb_drop_list(&skb_shinfo(skb)->frag_list);
311}
312
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313static void skb_clone_fraglist(struct sk_buff *skb)
314{
315 struct sk_buff *list;
316
David S. Millerfbb398a2009-06-09 00:18:59 -0700317 skb_walk_frags(skb, list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 skb_get(list);
319}
320
Adrian Bunk5bba1712006-06-29 13:02:35 -0700321static void skb_release_data(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322{
323 if (!skb->cloned ||
324 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
325 &skb_shinfo(skb)->dataref)) {
326 if (skb_shinfo(skb)->nr_frags) {
327 int i;
328 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
329 put_page(skb_shinfo(skb)->frags[i].page);
330 }
331
David S. Miller21dc3302010-08-23 00:13:46 -0700332 if (skb_has_frag_list(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 skb_drop_fraglist(skb);
334
335 kfree(skb->head);
336 }
337}
338
339/*
340 * Free an skbuff by memory without cleaning the state.
341 */
Herbert Xu2d4baff2007-11-26 23:11:19 +0800342static void kfree_skbmem(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343{
David S. Millerd179cd12005-08-17 14:57:30 -0700344 struct sk_buff *other;
345 atomic_t *fclone_ref;
346
David S. Millerd179cd12005-08-17 14:57:30 -0700347 switch (skb->fclone) {
348 case SKB_FCLONE_UNAVAILABLE:
349 kmem_cache_free(skbuff_head_cache, skb);
350 break;
351
352 case SKB_FCLONE_ORIG:
353 fclone_ref = (atomic_t *) (skb + 2);
354 if (atomic_dec_and_test(fclone_ref))
355 kmem_cache_free(skbuff_fclone_cache, skb);
356 break;
357
358 case SKB_FCLONE_CLONE:
359 fclone_ref = (atomic_t *) (skb + 1);
360 other = skb - 1;
361
362 /* The clone portion is available for
363 * fast-cloning again.
364 */
365 skb->fclone = SKB_FCLONE_UNAVAILABLE;
366
367 if (atomic_dec_and_test(fclone_ref))
368 kmem_cache_free(skbuff_fclone_cache, other);
369 break;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -0700370 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371}
372
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700373static void skb_release_head_state(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374{
Eric Dumazetadf30902009-06-02 05:19:30 +0000375 skb_dst_drop(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376#ifdef CONFIG_XFRM
377 secpath_put(skb->sp);
378#endif
Stephen Hemminger9c2b3322005-04-19 22:39:42 -0700379 if (skb->destructor) {
380 WARN_ON(in_irq());
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 skb->destructor(skb);
382 }
Yasuyuki Kozakai9fb9cbb2005-11-09 16:38:16 -0800383#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -0700384 nf_conntrack_put(skb->nfct);
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +0100385#endif
386#ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
Yasuyuki Kozakai9fb9cbb2005-11-09 16:38:16 -0800387 nf_conntrack_put_reasm(skb->nfct_reasm);
388#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389#ifdef CONFIG_BRIDGE_NETFILTER
390 nf_bridge_put(skb->nf_bridge);
391#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392/* XXX: IS this still necessary? - JHS */
393#ifdef CONFIG_NET_SCHED
394 skb->tc_index = 0;
395#ifdef CONFIG_NET_CLS_ACT
396 skb->tc_verd = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397#endif
398#endif
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700399}
400
401/* Free everything but the sk_buff shell. */
402static void skb_release_all(struct sk_buff *skb)
403{
404 skb_release_head_state(skb);
Herbert Xu2d4baff2007-11-26 23:11:19 +0800405 skb_release_data(skb);
406}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407
Herbert Xu2d4baff2007-11-26 23:11:19 +0800408/**
409 * __kfree_skb - private function
410 * @skb: buffer
411 *
412 * Free an sk_buff. Release anything attached to the buffer.
413 * Clean the state. This is an internal helper function. Users should
414 * always call kfree_skb
415 */
416
417void __kfree_skb(struct sk_buff *skb)
418{
419 skb_release_all(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 kfree_skbmem(skb);
421}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800422EXPORT_SYMBOL(__kfree_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423
424/**
Jörn Engel231d06a2006-03-20 21:28:35 -0800425 * kfree_skb - free an sk_buff
426 * @skb: buffer to free
427 *
428 * Drop a reference to the buffer and free it if the usage count has
429 * hit zero.
430 */
431void kfree_skb(struct sk_buff *skb)
432{
433 if (unlikely(!skb))
434 return;
435 if (likely(atomic_read(&skb->users) == 1))
436 smp_rmb();
437 else if (likely(!atomic_dec_and_test(&skb->users)))
438 return;
Neil Hormanead2ceb2009-03-11 09:49:55 +0000439 trace_kfree_skb(skb, __builtin_return_address(0));
Jörn Engel231d06a2006-03-20 21:28:35 -0800440 __kfree_skb(skb);
441}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800442EXPORT_SYMBOL(kfree_skb);
Jörn Engel231d06a2006-03-20 21:28:35 -0800443
Stephen Hemmingerd1a203e2008-11-01 21:01:09 -0700444/**
Neil Hormanead2ceb2009-03-11 09:49:55 +0000445 * consume_skb - free an skbuff
446 * @skb: buffer to free
447 *
448 * Drop a ref to the buffer and free it if the usage count has hit zero
449 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
450 * is being dropped after a failure and notes that
451 */
452void consume_skb(struct sk_buff *skb)
453{
454 if (unlikely(!skb))
455 return;
456 if (likely(atomic_read(&skb->users) == 1))
457 smp_rmb();
458 else if (likely(!atomic_dec_and_test(&skb->users)))
459 return;
Koki Sanagi07dc22e2010-08-23 18:46:12 +0900460 trace_consume_skb(skb);
Neil Hormanead2ceb2009-03-11 09:49:55 +0000461 __kfree_skb(skb);
462}
463EXPORT_SYMBOL(consume_skb);
464
465/**
Stephen Hemmingerd1a203e2008-11-01 21:01:09 -0700466 * skb_recycle_check - check if skb can be reused for receive
467 * @skb: buffer
468 * @skb_size: minimum receive buffer size
469 *
470 * Checks that the skb passed in is not shared or cloned, and
471 * that it is linear and its head portion at least as large as
472 * skb_size so that it can be recycled as a receive buffer.
473 * If these conditions are met, this function does any necessary
474 * reference count dropping and cleans up the skbuff as if it
475 * just came from __alloc_skb().
476 */
Changli Gao5b0daa32010-05-29 00:12:13 -0700477bool skb_recycle_check(struct sk_buff *skb, int skb_size)
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700478{
479 struct skb_shared_info *shinfo;
480
Anton Vorontsove84af6d2009-11-10 14:11:01 +0000481 if (irqs_disabled())
Changli Gao5b0daa32010-05-29 00:12:13 -0700482 return false;
Anton Vorontsove84af6d2009-11-10 14:11:01 +0000483
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700484 if (skb_is_nonlinear(skb) || skb->fclone != SKB_FCLONE_UNAVAILABLE)
Changli Gao5b0daa32010-05-29 00:12:13 -0700485 return false;
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700486
487 skb_size = SKB_DATA_ALIGN(skb_size + NET_SKB_PAD);
488 if (skb_end_pointer(skb) - skb->head < skb_size)
Changli Gao5b0daa32010-05-29 00:12:13 -0700489 return false;
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700490
491 if (skb_shared(skb) || skb_cloned(skb))
Changli Gao5b0daa32010-05-29 00:12:13 -0700492 return false;
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700493
494 skb_release_head_state(skb);
Eric Dumazetec7d2f22010-05-05 01:07:37 -0700495
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700496 shinfo = skb_shinfo(skb);
Eric Dumazetec7d2f22010-05-05 01:07:37 -0700497 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700498 atomic_set(&shinfo->dataref, 1);
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700499
500 memset(skb, 0, offsetof(struct sk_buff, tail));
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700501 skb->data = skb->head + NET_SKB_PAD;
Lennert Buytenhek5cd33db2008-11-10 21:45:05 -0800502 skb_reset_tail_pointer(skb);
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700503
Changli Gao5b0daa32010-05-29 00:12:13 -0700504 return true;
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700505}
506EXPORT_SYMBOL(skb_recycle_check);
507
Herbert Xudec18812007-10-14 00:37:30 -0700508static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
509{
510 new->tstamp = old->tstamp;
511 new->dev = old->dev;
512 new->transport_header = old->transport_header;
513 new->network_header = old->network_header;
514 new->mac_header = old->mac_header;
Eric Dumazet7fee2262010-05-11 23:19:48 +0000515 skb_dst_copy(new, old);
Tom Herbert0a9627f2010-03-16 08:03:29 +0000516 new->rxhash = old->rxhash;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -0700517#ifdef CONFIG_XFRM
Herbert Xudec18812007-10-14 00:37:30 -0700518 new->sp = secpath_get(old->sp);
519#endif
520 memcpy(new->cb, old->cb, sizeof(old->cb));
Herbert Xu9bcb97c2009-05-22 22:20:02 +0000521 new->csum = old->csum;
Herbert Xudec18812007-10-14 00:37:30 -0700522 new->local_df = old->local_df;
523 new->pkt_type = old->pkt_type;
524 new->ip_summed = old->ip_summed;
525 skb_copy_queue_mapping(new, old);
526 new->priority = old->priority;
527#if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
528 new->ipvs_property = old->ipvs_property;
529#endif
530 new->protocol = old->protocol;
531 new->mark = old->mark;
Eric Dumazet8964be42009-11-20 15:35:04 -0800532 new->skb_iif = old->skb_iif;
Herbert Xudec18812007-10-14 00:37:30 -0700533 __nf_copy(new, old);
534#if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
535 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
536 new->nf_trace = old->nf_trace;
537#endif
538#ifdef CONFIG_NET_SCHED
539 new->tc_index = old->tc_index;
540#ifdef CONFIG_NET_CLS_ACT
541 new->tc_verd = old->tc_verd;
542#endif
543#endif
Patrick McHardy6aa895b02008-07-14 22:49:06 -0700544 new->vlan_tci = old->vlan_tci;
545
Herbert Xudec18812007-10-14 00:37:30 -0700546 skb_copy_secmark(new, old);
547}
548
Herbert Xu82c49a32009-05-22 22:11:37 +0000549/*
550 * You should not add any new code to this function. Add it to
551 * __copy_skb_header above instead.
552 */
Herbert Xue0053ec2007-10-14 00:37:52 -0700553static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555#define C(x) n->x = skb->x
556
557 n->next = n->prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 n->sk = NULL;
Herbert Xudec18812007-10-14 00:37:30 -0700559 __copy_skb_header(n, skb);
560
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 C(len);
562 C(data_len);
Alexey Dobriyan3e6b3b22007-03-16 15:00:46 -0700563 C(mac_len);
Patrick McHardy334a8132007-06-25 04:35:20 -0700564 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
Paul Moore02f1c892008-01-07 21:56:41 -0800565 n->cloned = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 n->nohdr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 n->destructor = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 C(tail);
569 C(end);
Paul Moore02f1c892008-01-07 21:56:41 -0800570 C(head);
571 C(data);
572 C(truesize);
573 atomic_set(&n->users, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574
575 atomic_inc(&(skb_shinfo(skb)->dataref));
576 skb->cloned = 1;
577
578 return n;
Herbert Xue0053ec2007-10-14 00:37:52 -0700579#undef C
580}
581
582/**
583 * skb_morph - morph one skb into another
584 * @dst: the skb to receive the contents
585 * @src: the skb to supply the contents
586 *
587 * This is identical to skb_clone except that the target skb is
588 * supplied by the user.
589 *
590 * The target skb is returned upon exit.
591 */
592struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
593{
Herbert Xu2d4baff2007-11-26 23:11:19 +0800594 skb_release_all(dst);
Herbert Xue0053ec2007-10-14 00:37:52 -0700595 return __skb_clone(dst, src);
596}
597EXPORT_SYMBOL_GPL(skb_morph);
598
599/**
600 * skb_clone - duplicate an sk_buff
601 * @skb: buffer to clone
602 * @gfp_mask: allocation priority
603 *
604 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
605 * copies share the same packet data but not structure. The new
606 * buffer has a reference count of 1. If the allocation fails the
607 * function returns %NULL otherwise the new buffer is returned.
608 *
609 * If this function is called from an interrupt gfp_mask() must be
610 * %GFP_ATOMIC.
611 */
612
613struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
614{
615 struct sk_buff *n;
616
617 n = skb + 1;
618 if (skb->fclone == SKB_FCLONE_ORIG &&
619 n->fclone == SKB_FCLONE_UNAVAILABLE) {
620 atomic_t *fclone_ref = (atomic_t *) (n + 1);
621 n->fclone = SKB_FCLONE_CLONE;
622 atomic_inc(fclone_ref);
623 } else {
624 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
625 if (!n)
626 return NULL;
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200627
628 kmemcheck_annotate_bitfield(n, flags1);
629 kmemcheck_annotate_bitfield(n, flags2);
Herbert Xue0053ec2007-10-14 00:37:52 -0700630 n->fclone = SKB_FCLONE_UNAVAILABLE;
631 }
632
633 return __skb_clone(n, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800635EXPORT_SYMBOL(skb_clone);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636
637static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
638{
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700639#ifndef NET_SKBUFF_DATA_USES_OFFSET
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 /*
641 * Shift between the two data areas in bytes
642 */
643 unsigned long offset = new->data - old->data;
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700644#endif
Herbert Xudec18812007-10-14 00:37:30 -0700645
646 __copy_skb_header(new, old);
647
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700648#ifndef NET_SKBUFF_DATA_USES_OFFSET
649 /* {transport,network,mac}_header are relative to skb->head */
650 new->transport_header += offset;
651 new->network_header += offset;
Stephen Hemminger603a8bb2009-06-17 12:17:34 +0000652 if (skb_mac_header_was_set(new))
653 new->mac_header += offset;
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700654#endif
Herbert Xu79671682006-06-22 02:40:14 -0700655 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
656 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
657 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658}
659
660/**
661 * skb_copy - create private copy of an sk_buff
662 * @skb: buffer to copy
663 * @gfp_mask: allocation priority
664 *
665 * Make a copy of both an &sk_buff and its data. This is used when the
666 * caller wishes to modify the data and needs a private copy of the
667 * data to alter. Returns %NULL on failure or the pointer to the buffer
668 * on success. The returned buffer has a reference count of 1.
669 *
670 * As by-product this function converts non-linear &sk_buff to linear
671 * one, so that &sk_buff becomes completely private and caller is allowed
672 * to modify all the data of returned buffer. This means that this
673 * function is not recommended for use in circumstances when only
674 * header is going to be modified. Use pskb_copy() instead.
675 */
676
Al Virodd0fc662005-10-07 07:46:04 +0100677struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678{
Eric Dumazet6602ceb2010-09-01 05:25:10 +0000679 int headerlen = skb_headroom(skb);
680 unsigned int size = (skb_end_pointer(skb) - skb->head) + skb->data_len;
681 struct sk_buff *n = alloc_skb(size, gfp_mask);
682
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 if (!n)
684 return NULL;
685
686 /* Set the data pointer */
687 skb_reserve(n, headerlen);
688 /* Set the tail pointer and length */
689 skb_put(n, skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690
691 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
692 BUG();
693
694 copy_skb_header(n, skb);
695 return n;
696}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800697EXPORT_SYMBOL(skb_copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698
699/**
700 * pskb_copy - create copy of an sk_buff with private head.
701 * @skb: buffer to copy
702 * @gfp_mask: allocation priority
703 *
704 * Make a copy of both an &sk_buff and part of its data, located
705 * in header. Fragmented data remain shared. This is used when
706 * the caller wishes to modify only header of &sk_buff and needs
707 * private copy of the header to alter. Returns %NULL on failure
708 * or the pointer to the buffer on success.
709 * The returned buffer has a reference count of 1.
710 */
711
Al Virodd0fc662005-10-07 07:46:04 +0100712struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713{
Eric Dumazet6602ceb2010-09-01 05:25:10 +0000714 unsigned int size = skb_end_pointer(skb) - skb->head;
715 struct sk_buff *n = alloc_skb(size, gfp_mask);
716
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 if (!n)
718 goto out;
719
720 /* Set the data pointer */
Eric Dumazet6602ceb2010-09-01 05:25:10 +0000721 skb_reserve(n, skb_headroom(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 /* Set the tail pointer and length */
723 skb_put(n, skb_headlen(skb));
724 /* Copy the bytes */
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -0300725 skb_copy_from_linear_data(skb, n->data, n->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726
Herbert Xu25f484a2006-11-07 14:57:15 -0800727 n->truesize += skb->data_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 n->data_len = skb->data_len;
729 n->len = skb->len;
730
731 if (skb_shinfo(skb)->nr_frags) {
732 int i;
733
734 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
735 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
736 get_page(skb_shinfo(n)->frags[i].page);
737 }
738 skb_shinfo(n)->nr_frags = i;
739 }
740
David S. Miller21dc3302010-08-23 00:13:46 -0700741 if (skb_has_frag_list(skb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
743 skb_clone_fraglist(n);
744 }
745
746 copy_skb_header(n, skb);
747out:
748 return n;
749}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800750EXPORT_SYMBOL(pskb_copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751
752/**
753 * pskb_expand_head - reallocate header of &sk_buff
754 * @skb: buffer to reallocate
755 * @nhead: room to add at head
756 * @ntail: room to add at tail
757 * @gfp_mask: allocation priority
758 *
759 * Expands (or creates identical copy, if &nhead and &ntail are zero)
760 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
761 * reference count of 1. Returns zero in the case of success or error,
762 * if expansion failed. In the last case, &sk_buff is not changed.
763 *
764 * All the pointers pointing into skb header may change and must be
765 * reloaded after call to this function.
766 */
767
Victor Fusco86a76ca2005-07-08 14:57:47 -0700768int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
Al Virodd0fc662005-10-07 07:46:04 +0100769 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770{
771 int i;
772 u8 *data;
Eric Dumazet6602ceb2010-09-01 05:25:10 +0000773 int size = nhead + (skb_end_pointer(skb) - skb->head) + ntail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 long off;
Eric Dumazet1fd63042010-09-02 23:09:32 +0000775 bool fastpath;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776
Herbert Xu4edd87a2008-10-01 07:09:38 -0700777 BUG_ON(nhead < 0);
778
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 if (skb_shared(skb))
780 BUG();
781
782 size = SKB_DATA_ALIGN(size);
783
Changli Gaoca44ac32010-11-29 22:48:46 +0000784 /* Check if we can avoid taking references on fragments if we own
785 * the last reference on skb->head. (see skb_release_data())
786 */
787 if (!skb->cloned)
788 fastpath = true;
789 else {
790 int delta = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1;
791
792 fastpath = atomic_read(&skb_shinfo(skb)->dataref) == delta;
793 }
794
795 if (fastpath &&
796 size + sizeof(struct skb_shared_info) <= ksize(skb->head)) {
797 memmove(skb->head + size, skb_shinfo(skb),
798 offsetof(struct skb_shared_info,
799 frags[skb_shinfo(skb)->nr_frags]));
800 memmove(skb->head + nhead, skb->head,
801 skb_tail_pointer(skb) - skb->head);
802 off = nhead;
803 goto adjust_others;
804 }
805
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
807 if (!data)
808 goto nodata;
809
810 /* Copy only real data... and, alas, header. This should be
Eric Dumazet6602ceb2010-09-01 05:25:10 +0000811 * optimized for the cases when header is void.
812 */
813 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
814
815 memcpy((struct skb_shared_info *)(data + size),
816 skb_shinfo(skb),
Eric Dumazetfed66382010-07-22 19:09:08 +0000817 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818
Eric Dumazet1fd63042010-09-02 23:09:32 +0000819 if (fastpath) {
820 kfree(skb->head);
821 } else {
822 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
823 get_page(skb_shinfo(skb)->frags[i].page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824
Eric Dumazet1fd63042010-09-02 23:09:32 +0000825 if (skb_has_frag_list(skb))
826 skb_clone_fraglist(skb);
827
828 skb_release_data(skb);
829 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 off = (data + nhead) - skb->head;
831
832 skb->head = data;
Changli Gaoca44ac32010-11-29 22:48:46 +0000833adjust_others:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 skb->data += off;
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700835#ifdef NET_SKBUFF_DATA_USES_OFFSET
836 skb->end = size;
Patrick McHardy56eb8882007-04-09 11:45:04 -0700837 off = nhead;
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700838#else
839 skb->end = skb->head + size;
Patrick McHardy56eb8882007-04-09 11:45:04 -0700840#endif
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700841 /* {transport,network,mac}_header and tail are relative to skb->head */
842 skb->tail += off;
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700843 skb->transport_header += off;
844 skb->network_header += off;
Stephen Hemminger603a8bb2009-06-17 12:17:34 +0000845 if (skb_mac_header_was_set(skb))
846 skb->mac_header += off;
Andrea Shepard00c5a982010-07-22 09:12:35 +0000847 /* Only adjust this if it actually is csum_start rather than csum */
848 if (skb->ip_summed == CHECKSUM_PARTIAL)
849 skb->csum_start += nhead;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 skb->cloned = 0;
Patrick McHardy334a8132007-06-25 04:35:20 -0700851 skb->hdr_len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 skb->nohdr = 0;
853 atomic_set(&skb_shinfo(skb)->dataref, 1);
854 return 0;
855
856nodata:
857 return -ENOMEM;
858}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800859EXPORT_SYMBOL(pskb_expand_head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860
861/* Make private copy of skb with writable head and some headroom */
862
863struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
864{
865 struct sk_buff *skb2;
866 int delta = headroom - skb_headroom(skb);
867
868 if (delta <= 0)
869 skb2 = pskb_copy(skb, GFP_ATOMIC);
870 else {
871 skb2 = skb_clone(skb, GFP_ATOMIC);
872 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
873 GFP_ATOMIC)) {
874 kfree_skb(skb2);
875 skb2 = NULL;
876 }
877 }
878 return skb2;
879}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800880EXPORT_SYMBOL(skb_realloc_headroom);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881
882/**
883 * skb_copy_expand - copy and expand sk_buff
884 * @skb: buffer to copy
885 * @newheadroom: new free bytes at head
886 * @newtailroom: new free bytes at tail
887 * @gfp_mask: allocation priority
888 *
889 * Make a copy of both an &sk_buff and its data and while doing so
890 * allocate additional space.
891 *
892 * This is used when the caller wishes to modify the data and needs a
893 * private copy of the data to alter as well as more space for new fields.
894 * Returns %NULL on failure or the pointer to the buffer
895 * on success. The returned buffer has a reference count of 1.
896 *
897 * You must pass %GFP_ATOMIC as the allocation priority if this function
898 * is called from an interrupt.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 */
900struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
Victor Fusco86a76ca2005-07-08 14:57:47 -0700901 int newheadroom, int newtailroom,
Al Virodd0fc662005-10-07 07:46:04 +0100902 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903{
904 /*
905 * Allocate the copy buffer
906 */
907 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
908 gfp_mask);
Patrick McHardyefd1e8d2007-04-10 18:30:09 -0700909 int oldheadroom = skb_headroom(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 int head_copy_len, head_copy_off;
Herbert Xu52886052007-09-16 16:32:11 -0700911 int off;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912
913 if (!n)
914 return NULL;
915
916 skb_reserve(n, newheadroom);
917
918 /* Set the tail pointer and length */
919 skb_put(n, skb->len);
920
Patrick McHardyefd1e8d2007-04-10 18:30:09 -0700921 head_copy_len = oldheadroom;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 head_copy_off = 0;
923 if (newheadroom <= head_copy_len)
924 head_copy_len = newheadroom;
925 else
926 head_copy_off = newheadroom - head_copy_len;
927
928 /* Copy the linear header and data. */
929 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
930 skb->len + head_copy_len))
931 BUG();
932
933 copy_skb_header(n, skb);
934
Patrick McHardyefd1e8d2007-04-10 18:30:09 -0700935 off = newheadroom - oldheadroom;
David S. Millerbe2b6e62010-07-22 13:27:09 -0700936 if (n->ip_summed == CHECKSUM_PARTIAL)
937 n->csum_start += off;
Herbert Xu52886052007-09-16 16:32:11 -0700938#ifdef NET_SKBUFF_DATA_USES_OFFSET
Patrick McHardyefd1e8d2007-04-10 18:30:09 -0700939 n->transport_header += off;
940 n->network_header += off;
Stephen Hemminger603a8bb2009-06-17 12:17:34 +0000941 if (skb_mac_header_was_set(skb))
942 n->mac_header += off;
Herbert Xu52886052007-09-16 16:32:11 -0700943#endif
Patrick McHardyefd1e8d2007-04-10 18:30:09 -0700944
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 return n;
946}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800947EXPORT_SYMBOL(skb_copy_expand);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948
949/**
950 * skb_pad - zero pad the tail of an skb
951 * @skb: buffer to pad
952 * @pad: space to pad
953 *
954 * Ensure that a buffer is followed by a padding area that is zero
955 * filled. Used by network drivers which may DMA or transfer data
956 * beyond the buffer end onto the wire.
957 *
Herbert Xu5b057c62006-06-23 02:06:41 -0700958 * May return error in out of memory cases. The skb is freed on error.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 */
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900960
Herbert Xu5b057c62006-06-23 02:06:41 -0700961int skb_pad(struct sk_buff *skb, int pad)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962{
Herbert Xu5b057c62006-06-23 02:06:41 -0700963 int err;
964 int ntail;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900965
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 /* If the skbuff is non linear tailroom is always zero.. */
Herbert Xu5b057c62006-06-23 02:06:41 -0700967 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 memset(skb->data+skb->len, 0, pad);
Herbert Xu5b057c62006-06-23 02:06:41 -0700969 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 }
Herbert Xu5b057c62006-06-23 02:06:41 -0700971
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700972 ntail = skb->data_len + pad - (skb->end - skb->tail);
Herbert Xu5b057c62006-06-23 02:06:41 -0700973 if (likely(skb_cloned(skb) || ntail > 0)) {
974 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
975 if (unlikely(err))
976 goto free_skb;
977 }
978
979 /* FIXME: The use of this function with non-linear skb's really needs
980 * to be audited.
981 */
982 err = skb_linearize(skb);
983 if (unlikely(err))
984 goto free_skb;
985
986 memset(skb->data + skb->len, 0, pad);
987 return 0;
988
989free_skb:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 kfree_skb(skb);
Herbert Xu5b057c62006-06-23 02:06:41 -0700991 return err;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900992}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800993EXPORT_SYMBOL(skb_pad);
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900994
Ilpo Järvinen0dde3e12008-03-27 17:43:41 -0700995/**
996 * skb_put - add data to a buffer
997 * @skb: buffer to use
998 * @len: amount of data to add
999 *
1000 * This function extends the used data area of the buffer. If this would
1001 * exceed the total buffer size the kernel will panic. A pointer to the
1002 * first byte of the extra data is returned.
1003 */
1004unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1005{
1006 unsigned char *tmp = skb_tail_pointer(skb);
1007 SKB_LINEAR_ASSERT(skb);
1008 skb->tail += len;
1009 skb->len += len;
1010 if (unlikely(skb->tail > skb->end))
1011 skb_over_panic(skb, len, __builtin_return_address(0));
1012 return tmp;
1013}
1014EXPORT_SYMBOL(skb_put);
1015
Ilpo Järvinen6be8ac22008-03-27 17:47:24 -07001016/**
Ilpo Järvinenc2aa2702008-03-27 17:52:40 -07001017 * skb_push - add data to the start of a buffer
1018 * @skb: buffer to use
1019 * @len: amount of data to add
1020 *
1021 * This function extends the used data area of the buffer at the buffer
1022 * start. If this would exceed the total buffer headroom the kernel will
1023 * panic. A pointer to the first byte of the extra data is returned.
1024 */
1025unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1026{
1027 skb->data -= len;
1028 skb->len += len;
1029 if (unlikely(skb->data<skb->head))
1030 skb_under_panic(skb, len, __builtin_return_address(0));
1031 return skb->data;
1032}
1033EXPORT_SYMBOL(skb_push);
1034
1035/**
Ilpo Järvinen6be8ac22008-03-27 17:47:24 -07001036 * skb_pull - remove data from the start of a buffer
1037 * @skb: buffer to use
1038 * @len: amount of data to remove
1039 *
1040 * This function removes data from the start of a buffer, returning
1041 * the memory to the headroom. A pointer to the next data in the buffer
1042 * is returned. Once the data has been pulled future pushes will overwrite
1043 * the old data.
1044 */
1045unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1046{
David S. Miller47d29642010-05-02 02:21:44 -07001047 return skb_pull_inline(skb, len);
Ilpo Järvinen6be8ac22008-03-27 17:47:24 -07001048}
1049EXPORT_SYMBOL(skb_pull);
1050
Ilpo Järvinen419ae742008-03-27 17:54:01 -07001051/**
1052 * skb_trim - remove end from a buffer
1053 * @skb: buffer to alter
1054 * @len: new length
1055 *
1056 * Cut the length of a buffer down by removing data from the tail. If
1057 * the buffer is already under the length specified it is not modified.
1058 * The skb must be linear.
1059 */
1060void skb_trim(struct sk_buff *skb, unsigned int len)
1061{
1062 if (skb->len > len)
1063 __skb_trim(skb, len);
1064}
1065EXPORT_SYMBOL(skb_trim);
1066
Herbert Xu3cc0e872006-06-09 16:13:38 -07001067/* Trims skb to length len. It can change skb pointers.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 */
1069
Herbert Xu3cc0e872006-06-09 16:13:38 -07001070int ___pskb_trim(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071{
Herbert Xu27b437c2006-07-13 19:26:39 -07001072 struct sk_buff **fragp;
1073 struct sk_buff *frag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 int offset = skb_headlen(skb);
1075 int nfrags = skb_shinfo(skb)->nr_frags;
1076 int i;
Herbert Xu27b437c2006-07-13 19:26:39 -07001077 int err;
1078
1079 if (skb_cloned(skb) &&
1080 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1081 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001083 i = 0;
1084 if (offset >= len)
1085 goto drop_pages;
1086
1087 for (; i < nfrags; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 int end = offset + skb_shinfo(skb)->frags[i].size;
Herbert Xu27b437c2006-07-13 19:26:39 -07001089
1090 if (end < len) {
1091 offset = end;
1092 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 }
Herbert Xu27b437c2006-07-13 19:26:39 -07001094
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001095 skb_shinfo(skb)->frags[i++].size = len - offset;
Herbert Xu27b437c2006-07-13 19:26:39 -07001096
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001097drop_pages:
Herbert Xu27b437c2006-07-13 19:26:39 -07001098 skb_shinfo(skb)->nr_frags = i;
1099
1100 for (; i < nfrags; i++)
1101 put_page(skb_shinfo(skb)->frags[i].page);
1102
David S. Miller21dc3302010-08-23 00:13:46 -07001103 if (skb_has_frag_list(skb))
Herbert Xu27b437c2006-07-13 19:26:39 -07001104 skb_drop_fraglist(skb);
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001105 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 }
1107
Herbert Xu27b437c2006-07-13 19:26:39 -07001108 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1109 fragp = &frag->next) {
1110 int end = offset + frag->len;
1111
1112 if (skb_shared(frag)) {
1113 struct sk_buff *nfrag;
1114
1115 nfrag = skb_clone(frag, GFP_ATOMIC);
1116 if (unlikely(!nfrag))
1117 return -ENOMEM;
1118
1119 nfrag->next = frag->next;
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001120 kfree_skb(frag);
Herbert Xu27b437c2006-07-13 19:26:39 -07001121 frag = nfrag;
1122 *fragp = frag;
1123 }
1124
1125 if (end < len) {
1126 offset = end;
1127 continue;
1128 }
1129
1130 if (end > len &&
1131 unlikely((err = pskb_trim(frag, len - offset))))
1132 return err;
1133
1134 if (frag->next)
1135 skb_drop_list(&frag->next);
1136 break;
1137 }
1138
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001139done:
Herbert Xu27b437c2006-07-13 19:26:39 -07001140 if (len > skb_headlen(skb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 skb->data_len -= skb->len - len;
1142 skb->len = len;
1143 } else {
Herbert Xu27b437c2006-07-13 19:26:39 -07001144 skb->len = len;
1145 skb->data_len = 0;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001146 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 }
1148
1149 return 0;
1150}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001151EXPORT_SYMBOL(___pskb_trim);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152
1153/**
1154 * __pskb_pull_tail - advance tail of skb header
1155 * @skb: buffer to reallocate
1156 * @delta: number of bytes to advance tail
1157 *
1158 * The function makes a sense only on a fragmented &sk_buff,
1159 * it expands header moving its tail forward and copying necessary
1160 * data from fragmented part.
1161 *
1162 * &sk_buff MUST have reference count of 1.
1163 *
1164 * Returns %NULL (and &sk_buff does not change) if pull failed
1165 * or value of new tail of skb in the case of success.
1166 *
1167 * All the pointers pointing into skb header may change and must be
1168 * reloaded after call to this function.
1169 */
1170
1171/* Moves tail of skb head forward, copying data from fragmented part,
1172 * when it is necessary.
1173 * 1. It may fail due to malloc failure.
1174 * 2. It may change skb pointers.
1175 *
1176 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1177 */
1178unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1179{
1180 /* If skb has not enough free space at tail, get new one
1181 * plus 128 bytes for future expansions. If we have enough
1182 * room at tail, reallocate without expansion only if skb is cloned.
1183 */
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001184 int i, k, eat = (skb->tail + delta) - skb->end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185
1186 if (eat > 0 || skb_cloned(skb)) {
1187 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1188 GFP_ATOMIC))
1189 return NULL;
1190 }
1191
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001192 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 BUG();
1194
1195 /* Optimization: no fragments, no reasons to preestimate
1196 * size of pulled pages. Superb.
1197 */
David S. Miller21dc3302010-08-23 00:13:46 -07001198 if (!skb_has_frag_list(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 goto pull_pages;
1200
1201 /* Estimate size of pulled pages. */
1202 eat = delta;
1203 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1204 if (skb_shinfo(skb)->frags[i].size >= eat)
1205 goto pull_pages;
1206 eat -= skb_shinfo(skb)->frags[i].size;
1207 }
1208
1209 /* If we need update frag list, we are in troubles.
1210 * Certainly, it possible to add an offset to skb data,
1211 * but taking into account that pulling is expected to
1212 * be very rare operation, it is worth to fight against
1213 * further bloating skb head and crucify ourselves here instead.
1214 * Pure masohism, indeed. 8)8)
1215 */
1216 if (eat) {
1217 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1218 struct sk_buff *clone = NULL;
1219 struct sk_buff *insp = NULL;
1220
1221 do {
Kris Katterjohn09a62662006-01-08 22:24:28 -08001222 BUG_ON(!list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223
1224 if (list->len <= eat) {
1225 /* Eaten as whole. */
1226 eat -= list->len;
1227 list = list->next;
1228 insp = list;
1229 } else {
1230 /* Eaten partially. */
1231
1232 if (skb_shared(list)) {
1233 /* Sucks! We need to fork list. :-( */
1234 clone = skb_clone(list, GFP_ATOMIC);
1235 if (!clone)
1236 return NULL;
1237 insp = list->next;
1238 list = clone;
1239 } else {
1240 /* This may be pulled without
1241 * problems. */
1242 insp = list;
1243 }
1244 if (!pskb_pull(list, eat)) {
Wei Yongjunf3fbbe02009-02-25 00:37:32 +00001245 kfree_skb(clone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 return NULL;
1247 }
1248 break;
1249 }
1250 } while (eat);
1251
1252 /* Free pulled out fragments. */
1253 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1254 skb_shinfo(skb)->frag_list = list->next;
1255 kfree_skb(list);
1256 }
1257 /* And insert new clone at head. */
1258 if (clone) {
1259 clone->next = list;
1260 skb_shinfo(skb)->frag_list = clone;
1261 }
1262 }
1263 /* Success! Now we may commit changes to skb data. */
1264
1265pull_pages:
1266 eat = delta;
1267 k = 0;
1268 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1269 if (skb_shinfo(skb)->frags[i].size <= eat) {
1270 put_page(skb_shinfo(skb)->frags[i].page);
1271 eat -= skb_shinfo(skb)->frags[i].size;
1272 } else {
1273 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1274 if (eat) {
1275 skb_shinfo(skb)->frags[k].page_offset += eat;
1276 skb_shinfo(skb)->frags[k].size -= eat;
1277 eat = 0;
1278 }
1279 k++;
1280 }
1281 }
1282 skb_shinfo(skb)->nr_frags = k;
1283
1284 skb->tail += delta;
1285 skb->data_len -= delta;
1286
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001287 return skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001289EXPORT_SYMBOL(__pskb_pull_tail);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290
1291/* Copy some data bits from skb to kernel buffer. */
1292
1293int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1294{
David S. Miller1a028e52007-04-27 15:21:23 -07001295 int start = skb_headlen(skb);
David S. Millerfbb398a2009-06-09 00:18:59 -07001296 struct sk_buff *frag_iter;
1297 int i, copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298
1299 if (offset > (int)skb->len - len)
1300 goto fault;
1301
1302 /* Copy header. */
David S. Miller1a028e52007-04-27 15:21:23 -07001303 if ((copy = start - offset) > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 if (copy > len)
1305 copy = len;
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03001306 skb_copy_from_linear_data_offset(skb, offset, to, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 if ((len -= copy) == 0)
1308 return 0;
1309 offset += copy;
1310 to += copy;
1311 }
1312
1313 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07001314 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001316 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07001317
1318 end = start + skb_shinfo(skb)->frags[i].size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 if ((copy = end - offset) > 0) {
1320 u8 *vaddr;
1321
1322 if (copy > len)
1323 copy = len;
1324
1325 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1326 memcpy(to,
David S. Miller1a028e52007-04-27 15:21:23 -07001327 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1328 offset - start, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 kunmap_skb_frag(vaddr);
1330
1331 if ((len -= copy) == 0)
1332 return 0;
1333 offset += copy;
1334 to += copy;
1335 }
David S. Miller1a028e52007-04-27 15:21:23 -07001336 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 }
1338
David S. Millerfbb398a2009-06-09 00:18:59 -07001339 skb_walk_frags(skb, frag_iter) {
1340 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341
David S. Millerfbb398a2009-06-09 00:18:59 -07001342 WARN_ON(start > offset + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343
David S. Millerfbb398a2009-06-09 00:18:59 -07001344 end = start + frag_iter->len;
1345 if ((copy = end - offset) > 0) {
1346 if (copy > len)
1347 copy = len;
1348 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1349 goto fault;
1350 if ((len -= copy) == 0)
1351 return 0;
1352 offset += copy;
1353 to += copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 }
David S. Millerfbb398a2009-06-09 00:18:59 -07001355 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 }
1357 if (!len)
1358 return 0;
1359
1360fault:
1361 return -EFAULT;
1362}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001363EXPORT_SYMBOL(skb_copy_bits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
Jens Axboe9c55e012007-11-06 23:30:13 -08001365/*
1366 * Callback from splice_to_pipe(), if we need to release some pages
1367 * at the end of the spd in case we error'ed out in filling the pipe.
1368 */
1369static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1370{
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001371 put_page(spd->pages[i]);
1372}
Jens Axboe9c55e012007-11-06 23:30:13 -08001373
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001374static inline struct page *linear_to_page(struct page *page, unsigned int *len,
1375 unsigned int *offset,
Jarek Poplawski7a67e562009-04-30 05:41:19 -07001376 struct sk_buff *skb, struct sock *sk)
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001377{
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001378 struct page *p = sk->sk_sndmsg_page;
1379 unsigned int off;
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001380
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001381 if (!p) {
1382new_page:
1383 p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1384 if (!p)
1385 return NULL;
1386
1387 off = sk->sk_sndmsg_off = 0;
1388 /* hold one ref to this page until it's full */
1389 } else {
1390 unsigned int mlen;
1391
1392 off = sk->sk_sndmsg_off;
1393 mlen = PAGE_SIZE - off;
1394 if (mlen < 64 && mlen < *len) {
1395 put_page(p);
1396 goto new_page;
1397 }
1398
1399 *len = min_t(unsigned int, *len, mlen);
1400 }
1401
1402 memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1403 sk->sk_sndmsg_off += *len;
1404 *offset = off;
1405 get_page(p);
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001406
1407 return p;
Jens Axboe9c55e012007-11-06 23:30:13 -08001408}
1409
1410/*
1411 * Fill page/offset/length into spd, if it can hold more pages.
1412 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001413static inline int spd_fill_page(struct splice_pipe_desc *spd,
1414 struct pipe_inode_info *pipe, struct page *page,
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001415 unsigned int *len, unsigned int offset,
Jarek Poplawski7a67e562009-04-30 05:41:19 -07001416 struct sk_buff *skb, int linear,
1417 struct sock *sk)
Jens Axboe9c55e012007-11-06 23:30:13 -08001418{
Jens Axboe35f3d142010-05-20 10:43:18 +02001419 if (unlikely(spd->nr_pages == pipe->buffers))
Jens Axboe9c55e012007-11-06 23:30:13 -08001420 return 1;
1421
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001422 if (linear) {
Jarek Poplawski7a67e562009-04-30 05:41:19 -07001423 page = linear_to_page(page, len, &offset, skb, sk);
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001424 if (!page)
1425 return 1;
1426 } else
1427 get_page(page);
1428
Jens Axboe9c55e012007-11-06 23:30:13 -08001429 spd->pages[spd->nr_pages] = page;
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001430 spd->partial[spd->nr_pages].len = *len;
Jens Axboe9c55e012007-11-06 23:30:13 -08001431 spd->partial[spd->nr_pages].offset = offset;
Jens Axboe9c55e012007-11-06 23:30:13 -08001432 spd->nr_pages++;
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001433
Jens Axboe9c55e012007-11-06 23:30:13 -08001434 return 0;
1435}
1436
Octavian Purdila2870c432008-07-15 00:49:11 -07001437static inline void __segment_seek(struct page **page, unsigned int *poff,
1438 unsigned int *plen, unsigned int off)
Jens Axboe9c55e012007-11-06 23:30:13 -08001439{
Jarek Poplawskice3dd392009-02-12 16:51:43 -08001440 unsigned long n;
1441
Octavian Purdila2870c432008-07-15 00:49:11 -07001442 *poff += off;
Jarek Poplawskice3dd392009-02-12 16:51:43 -08001443 n = *poff / PAGE_SIZE;
1444 if (n)
1445 *page = nth_page(*page, n);
1446
Octavian Purdila2870c432008-07-15 00:49:11 -07001447 *poff = *poff % PAGE_SIZE;
1448 *plen -= off;
1449}
Jens Axboe9c55e012007-11-06 23:30:13 -08001450
Octavian Purdila2870c432008-07-15 00:49:11 -07001451static inline int __splice_segment(struct page *page, unsigned int poff,
1452 unsigned int plen, unsigned int *off,
1453 unsigned int *len, struct sk_buff *skb,
Jarek Poplawski7a67e562009-04-30 05:41:19 -07001454 struct splice_pipe_desc *spd, int linear,
Jens Axboe35f3d142010-05-20 10:43:18 +02001455 struct sock *sk,
1456 struct pipe_inode_info *pipe)
Octavian Purdila2870c432008-07-15 00:49:11 -07001457{
1458 if (!*len)
1459 return 1;
1460
1461 /* skip this segment if already processed */
1462 if (*off >= plen) {
1463 *off -= plen;
1464 return 0;
Octavian Purdiladb43a282008-06-27 17:27:21 -07001465 }
Jens Axboe9c55e012007-11-06 23:30:13 -08001466
Octavian Purdila2870c432008-07-15 00:49:11 -07001467 /* ignore any bits we already processed */
1468 if (*off) {
1469 __segment_seek(&page, &poff, &plen, *off);
1470 *off = 0;
1471 }
1472
1473 do {
1474 unsigned int flen = min(*len, plen);
1475
1476 /* the linear region may spread across several pages */
1477 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1478
Jens Axboe35f3d142010-05-20 10:43:18 +02001479 if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk))
Octavian Purdila2870c432008-07-15 00:49:11 -07001480 return 1;
1481
1482 __segment_seek(&page, &poff, &plen, flen);
1483 *len -= flen;
1484
1485 } while (*len && plen);
1486
1487 return 0;
1488}
1489
1490/*
1491 * Map linear and fragment data from the skb to spd. It reports failure if the
1492 * pipe is full or if we already spliced the requested length.
1493 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001494static int __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1495 unsigned int *offset, unsigned int *len,
1496 struct splice_pipe_desc *spd, struct sock *sk)
Octavian Purdila2870c432008-07-15 00:49:11 -07001497{
1498 int seg;
1499
Jens Axboe9c55e012007-11-06 23:30:13 -08001500 /*
Octavian Purdila2870c432008-07-15 00:49:11 -07001501 * map the linear part
Jens Axboe9c55e012007-11-06 23:30:13 -08001502 */
Octavian Purdila2870c432008-07-15 00:49:11 -07001503 if (__splice_segment(virt_to_page(skb->data),
1504 (unsigned long) skb->data & (PAGE_SIZE - 1),
1505 skb_headlen(skb),
Jens Axboe35f3d142010-05-20 10:43:18 +02001506 offset, len, skb, spd, 1, sk, pipe))
Octavian Purdila2870c432008-07-15 00:49:11 -07001507 return 1;
Jens Axboe9c55e012007-11-06 23:30:13 -08001508
1509 /*
1510 * then map the fragments
1511 */
Jens Axboe9c55e012007-11-06 23:30:13 -08001512 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1513 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1514
Octavian Purdila2870c432008-07-15 00:49:11 -07001515 if (__splice_segment(f->page, f->page_offset, f->size,
Jens Axboe35f3d142010-05-20 10:43:18 +02001516 offset, len, skb, spd, 0, sk, pipe))
Octavian Purdila2870c432008-07-15 00:49:11 -07001517 return 1;
Jens Axboe9c55e012007-11-06 23:30:13 -08001518 }
1519
Octavian Purdila2870c432008-07-15 00:49:11 -07001520 return 0;
Jens Axboe9c55e012007-11-06 23:30:13 -08001521}
1522
1523/*
1524 * Map data from the skb to a pipe. Should handle both the linear part,
1525 * the fragments, and the frag list. It does NOT handle frag lists within
1526 * the frag list, if such a thing exists. We'd probably need to recurse to
1527 * handle that cleanly.
1528 */
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001529int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
Jens Axboe9c55e012007-11-06 23:30:13 -08001530 struct pipe_inode_info *pipe, unsigned int tlen,
1531 unsigned int flags)
1532{
Jens Axboe35f3d142010-05-20 10:43:18 +02001533 struct partial_page partial[PIPE_DEF_BUFFERS];
1534 struct page *pages[PIPE_DEF_BUFFERS];
Jens Axboe9c55e012007-11-06 23:30:13 -08001535 struct splice_pipe_desc spd = {
1536 .pages = pages,
1537 .partial = partial,
1538 .flags = flags,
1539 .ops = &sock_pipe_buf_ops,
1540 .spd_release = sock_spd_release,
1541 };
David S. Millerfbb398a2009-06-09 00:18:59 -07001542 struct sk_buff *frag_iter;
Jarek Poplawski7a67e562009-04-30 05:41:19 -07001543 struct sock *sk = skb->sk;
Jens Axboe35f3d142010-05-20 10:43:18 +02001544 int ret = 0;
1545
1546 if (splice_grow_spd(pipe, &spd))
1547 return -ENOMEM;
Jens Axboe9c55e012007-11-06 23:30:13 -08001548
1549 /*
1550 * __skb_splice_bits() only fails if the output has no room left,
1551 * so no point in going over the frag_list for the error case.
1552 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001553 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
Jens Axboe9c55e012007-11-06 23:30:13 -08001554 goto done;
1555 else if (!tlen)
1556 goto done;
1557
1558 /*
1559 * now see if we have a frag_list to map
1560 */
David S. Millerfbb398a2009-06-09 00:18:59 -07001561 skb_walk_frags(skb, frag_iter) {
1562 if (!tlen)
1563 break;
Jens Axboe35f3d142010-05-20 10:43:18 +02001564 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
David S. Millerfbb398a2009-06-09 00:18:59 -07001565 break;
Jens Axboe9c55e012007-11-06 23:30:13 -08001566 }
1567
1568done:
Jens Axboe9c55e012007-11-06 23:30:13 -08001569 if (spd.nr_pages) {
Jens Axboe9c55e012007-11-06 23:30:13 -08001570 /*
1571 * Drop the socket lock, otherwise we have reverse
1572 * locking dependencies between sk_lock and i_mutex
1573 * here as compared to sendfile(). We enter here
1574 * with the socket lock held, and splice_to_pipe() will
1575 * grab the pipe inode lock. For sendfile() emulation,
1576 * we call into ->sendpage() with the i_mutex lock held
1577 * and networking will grab the socket lock.
1578 */
Octavian Purdila293ad602008-06-04 15:45:58 -07001579 release_sock(sk);
Jens Axboe9c55e012007-11-06 23:30:13 -08001580 ret = splice_to_pipe(pipe, &spd);
Octavian Purdila293ad602008-06-04 15:45:58 -07001581 lock_sock(sk);
Jens Axboe9c55e012007-11-06 23:30:13 -08001582 }
1583
Jens Axboe35f3d142010-05-20 10:43:18 +02001584 splice_shrink_spd(pipe, &spd);
1585 return ret;
Jens Axboe9c55e012007-11-06 23:30:13 -08001586}
1587
Herbert Xu357b40a2005-04-19 22:30:14 -07001588/**
1589 * skb_store_bits - store bits from kernel buffer to skb
1590 * @skb: destination buffer
1591 * @offset: offset in destination
1592 * @from: source buffer
1593 * @len: number of bytes to copy
1594 *
1595 * Copy the specified number of bytes from the source buffer to the
1596 * destination skb. This function handles all the messy bits of
1597 * traversing fragment lists and such.
1598 */
1599
Stephen Hemminger0c6fcc82007-04-20 16:40:01 -07001600int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
Herbert Xu357b40a2005-04-19 22:30:14 -07001601{
David S. Miller1a028e52007-04-27 15:21:23 -07001602 int start = skb_headlen(skb);
David S. Millerfbb398a2009-06-09 00:18:59 -07001603 struct sk_buff *frag_iter;
1604 int i, copy;
Herbert Xu357b40a2005-04-19 22:30:14 -07001605
1606 if (offset > (int)skb->len - len)
1607 goto fault;
1608
David S. Miller1a028e52007-04-27 15:21:23 -07001609 if ((copy = start - offset) > 0) {
Herbert Xu357b40a2005-04-19 22:30:14 -07001610 if (copy > len)
1611 copy = len;
Arnaldo Carvalho de Melo27d7ff42007-03-31 11:55:19 -03001612 skb_copy_to_linear_data_offset(skb, offset, from, copy);
Herbert Xu357b40a2005-04-19 22:30:14 -07001613 if ((len -= copy) == 0)
1614 return 0;
1615 offset += copy;
1616 from += copy;
1617 }
1618
1619 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1620 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
David S. Miller1a028e52007-04-27 15:21:23 -07001621 int end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001622
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001623 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07001624
1625 end = start + frag->size;
Herbert Xu357b40a2005-04-19 22:30:14 -07001626 if ((copy = end - offset) > 0) {
1627 u8 *vaddr;
1628
1629 if (copy > len)
1630 copy = len;
1631
1632 vaddr = kmap_skb_frag(frag);
David S. Miller1a028e52007-04-27 15:21:23 -07001633 memcpy(vaddr + frag->page_offset + offset - start,
1634 from, copy);
Herbert Xu357b40a2005-04-19 22:30:14 -07001635 kunmap_skb_frag(vaddr);
1636
1637 if ((len -= copy) == 0)
1638 return 0;
1639 offset += copy;
1640 from += copy;
1641 }
David S. Miller1a028e52007-04-27 15:21:23 -07001642 start = end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001643 }
1644
David S. Millerfbb398a2009-06-09 00:18:59 -07001645 skb_walk_frags(skb, frag_iter) {
1646 int end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001647
David S. Millerfbb398a2009-06-09 00:18:59 -07001648 WARN_ON(start > offset + len);
Herbert Xu357b40a2005-04-19 22:30:14 -07001649
David S. Millerfbb398a2009-06-09 00:18:59 -07001650 end = start + frag_iter->len;
1651 if ((copy = end - offset) > 0) {
1652 if (copy > len)
1653 copy = len;
1654 if (skb_store_bits(frag_iter, offset - start,
1655 from, copy))
1656 goto fault;
1657 if ((len -= copy) == 0)
1658 return 0;
1659 offset += copy;
1660 from += copy;
Herbert Xu357b40a2005-04-19 22:30:14 -07001661 }
David S. Millerfbb398a2009-06-09 00:18:59 -07001662 start = end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001663 }
1664 if (!len)
1665 return 0;
1666
1667fault:
1668 return -EFAULT;
1669}
Herbert Xu357b40a2005-04-19 22:30:14 -07001670EXPORT_SYMBOL(skb_store_bits);
1671
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672/* Checksum skb data. */
1673
Al Viro2bbbc862006-11-14 21:37:14 -08001674__wsum skb_checksum(const struct sk_buff *skb, int offset,
1675 int len, __wsum csum)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676{
David S. Miller1a028e52007-04-27 15:21:23 -07001677 int start = skb_headlen(skb);
1678 int i, copy = start - offset;
David S. Millerfbb398a2009-06-09 00:18:59 -07001679 struct sk_buff *frag_iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 int pos = 0;
1681
1682 /* Checksum header. */
1683 if (copy > 0) {
1684 if (copy > len)
1685 copy = len;
1686 csum = csum_partial(skb->data + offset, copy, csum);
1687 if ((len -= copy) == 0)
1688 return csum;
1689 offset += copy;
1690 pos = copy;
1691 }
1692
1693 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07001694 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001696 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07001697
1698 end = start + skb_shinfo(skb)->frags[i].size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 if ((copy = end - offset) > 0) {
Al Viro44bb9362006-11-14 21:36:14 -08001700 __wsum csum2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 u8 *vaddr;
1702 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1703
1704 if (copy > len)
1705 copy = len;
1706 vaddr = kmap_skb_frag(frag);
David S. Miller1a028e52007-04-27 15:21:23 -07001707 csum2 = csum_partial(vaddr + frag->page_offset +
1708 offset - start, copy, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 kunmap_skb_frag(vaddr);
1710 csum = csum_block_add(csum, csum2, pos);
1711 if (!(len -= copy))
1712 return csum;
1713 offset += copy;
1714 pos += copy;
1715 }
David S. Miller1a028e52007-04-27 15:21:23 -07001716 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 }
1718
David S. Millerfbb398a2009-06-09 00:18:59 -07001719 skb_walk_frags(skb, frag_iter) {
1720 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721
David S. Millerfbb398a2009-06-09 00:18:59 -07001722 WARN_ON(start > offset + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723
David S. Millerfbb398a2009-06-09 00:18:59 -07001724 end = start + frag_iter->len;
1725 if ((copy = end - offset) > 0) {
1726 __wsum csum2;
1727 if (copy > len)
1728 copy = len;
1729 csum2 = skb_checksum(frag_iter, offset - start,
1730 copy, 0);
1731 csum = csum_block_add(csum, csum2, pos);
1732 if ((len -= copy) == 0)
1733 return csum;
1734 offset += copy;
1735 pos += copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 }
David S. Millerfbb398a2009-06-09 00:18:59 -07001737 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 }
Kris Katterjohn09a62662006-01-08 22:24:28 -08001739 BUG_ON(len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740
1741 return csum;
1742}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001743EXPORT_SYMBOL(skb_checksum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744
1745/* Both of above in one bottle. */
1746
Al Viro81d77662006-11-14 21:37:33 -08001747__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1748 u8 *to, int len, __wsum csum)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749{
David S. Miller1a028e52007-04-27 15:21:23 -07001750 int start = skb_headlen(skb);
1751 int i, copy = start - offset;
David S. Millerfbb398a2009-06-09 00:18:59 -07001752 struct sk_buff *frag_iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 int pos = 0;
1754
1755 /* Copy header. */
1756 if (copy > 0) {
1757 if (copy > len)
1758 copy = len;
1759 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1760 copy, csum);
1761 if ((len -= copy) == 0)
1762 return csum;
1763 offset += copy;
1764 to += copy;
1765 pos = copy;
1766 }
1767
1768 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07001769 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001771 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07001772
1773 end = start + skb_shinfo(skb)->frags[i].size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 if ((copy = end - offset) > 0) {
Al Viro50842052006-11-14 21:36:34 -08001775 __wsum csum2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 u8 *vaddr;
1777 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1778
1779 if (copy > len)
1780 copy = len;
1781 vaddr = kmap_skb_frag(frag);
1782 csum2 = csum_partial_copy_nocheck(vaddr +
David S. Miller1a028e52007-04-27 15:21:23 -07001783 frag->page_offset +
1784 offset - start, to,
1785 copy, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 kunmap_skb_frag(vaddr);
1787 csum = csum_block_add(csum, csum2, pos);
1788 if (!(len -= copy))
1789 return csum;
1790 offset += copy;
1791 to += copy;
1792 pos += copy;
1793 }
David S. Miller1a028e52007-04-27 15:21:23 -07001794 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 }
1796
David S. Millerfbb398a2009-06-09 00:18:59 -07001797 skb_walk_frags(skb, frag_iter) {
1798 __wsum csum2;
1799 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800
David S. Millerfbb398a2009-06-09 00:18:59 -07001801 WARN_ON(start > offset + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802
David S. Millerfbb398a2009-06-09 00:18:59 -07001803 end = start + frag_iter->len;
1804 if ((copy = end - offset) > 0) {
1805 if (copy > len)
1806 copy = len;
1807 csum2 = skb_copy_and_csum_bits(frag_iter,
1808 offset - start,
1809 to, copy, 0);
1810 csum = csum_block_add(csum, csum2, pos);
1811 if ((len -= copy) == 0)
1812 return csum;
1813 offset += copy;
1814 to += copy;
1815 pos += copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 }
David S. Millerfbb398a2009-06-09 00:18:59 -07001817 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 }
Kris Katterjohn09a62662006-01-08 22:24:28 -08001819 BUG_ON(len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 return csum;
1821}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001822EXPORT_SYMBOL(skb_copy_and_csum_bits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823
1824void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1825{
Al Virod3bc23e2006-11-14 21:24:49 -08001826 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 long csstart;
1828
Patrick McHardy84fa7932006-08-29 16:44:56 -07001829 if (skb->ip_summed == CHECKSUM_PARTIAL)
Michał Mirosław55508d62010-12-14 15:24:08 +00001830 csstart = skb_checksum_start_offset(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 else
1832 csstart = skb_headlen(skb);
1833
Kris Katterjohn09a62662006-01-08 22:24:28 -08001834 BUG_ON(csstart > skb_headlen(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03001836 skb_copy_from_linear_data(skb, to, csstart);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837
1838 csum = 0;
1839 if (csstart != skb->len)
1840 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1841 skb->len - csstart, 0);
1842
Patrick McHardy84fa7932006-08-29 16:44:56 -07001843 if (skb->ip_summed == CHECKSUM_PARTIAL) {
Al Viroff1dcad2006-11-20 18:07:29 -08001844 long csstuff = csstart + skb->csum_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845
Al Virod3bc23e2006-11-14 21:24:49 -08001846 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 }
1848}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001849EXPORT_SYMBOL(skb_copy_and_csum_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850
1851/**
1852 * skb_dequeue - remove from the head of the queue
1853 * @list: list to dequeue from
1854 *
1855 * Remove the head of the list. The list lock is taken so the function
1856 * may be used safely with other locking list functions. The head item is
1857 * returned or %NULL if the list is empty.
1858 */
1859
1860struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1861{
1862 unsigned long flags;
1863 struct sk_buff *result;
1864
1865 spin_lock_irqsave(&list->lock, flags);
1866 result = __skb_dequeue(list);
1867 spin_unlock_irqrestore(&list->lock, flags);
1868 return result;
1869}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001870EXPORT_SYMBOL(skb_dequeue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871
1872/**
1873 * skb_dequeue_tail - remove from the tail of the queue
1874 * @list: list to dequeue from
1875 *
1876 * Remove the tail of the list. The list lock is taken so the function
1877 * may be used safely with other locking list functions. The tail item is
1878 * returned or %NULL if the list is empty.
1879 */
1880struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1881{
1882 unsigned long flags;
1883 struct sk_buff *result;
1884
1885 spin_lock_irqsave(&list->lock, flags);
1886 result = __skb_dequeue_tail(list);
1887 spin_unlock_irqrestore(&list->lock, flags);
1888 return result;
1889}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001890EXPORT_SYMBOL(skb_dequeue_tail);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891
1892/**
1893 * skb_queue_purge - empty a list
1894 * @list: list to empty
1895 *
1896 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1897 * the list and one reference dropped. This function takes the list
1898 * lock and is atomic with respect to other list locking functions.
1899 */
1900void skb_queue_purge(struct sk_buff_head *list)
1901{
1902 struct sk_buff *skb;
1903 while ((skb = skb_dequeue(list)) != NULL)
1904 kfree_skb(skb);
1905}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001906EXPORT_SYMBOL(skb_queue_purge);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907
1908/**
1909 * skb_queue_head - queue a buffer at the list head
1910 * @list: list to use
1911 * @newsk: buffer to queue
1912 *
1913 * Queue a buffer at the start of the list. This function takes the
1914 * list lock and can be used safely with other locking &sk_buff functions
1915 * safely.
1916 *
1917 * A buffer cannot be placed on two lists at the same time.
1918 */
1919void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1920{
1921 unsigned long flags;
1922
1923 spin_lock_irqsave(&list->lock, flags);
1924 __skb_queue_head(list, newsk);
1925 spin_unlock_irqrestore(&list->lock, flags);
1926}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001927EXPORT_SYMBOL(skb_queue_head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928
1929/**
1930 * skb_queue_tail - queue a buffer at the list tail
1931 * @list: list to use
1932 * @newsk: buffer to queue
1933 *
1934 * Queue a buffer at the tail of the list. This function takes the
1935 * list lock and can be used safely with other locking &sk_buff functions
1936 * safely.
1937 *
1938 * A buffer cannot be placed on two lists at the same time.
1939 */
1940void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1941{
1942 unsigned long flags;
1943
1944 spin_lock_irqsave(&list->lock, flags);
1945 __skb_queue_tail(list, newsk);
1946 spin_unlock_irqrestore(&list->lock, flags);
1947}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001948EXPORT_SYMBOL(skb_queue_tail);
David S. Miller8728b832005-08-09 19:25:21 -07001949
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950/**
1951 * skb_unlink - remove a buffer from a list
1952 * @skb: buffer to remove
David S. Miller8728b832005-08-09 19:25:21 -07001953 * @list: list to use
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 *
David S. Miller8728b832005-08-09 19:25:21 -07001955 * Remove a packet from a list. The list locks are taken and this
1956 * function is atomic with respect to other list locked calls
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 *
David S. Miller8728b832005-08-09 19:25:21 -07001958 * You must know what list the SKB is on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 */
David S. Miller8728b832005-08-09 19:25:21 -07001960void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961{
David S. Miller8728b832005-08-09 19:25:21 -07001962 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963
David S. Miller8728b832005-08-09 19:25:21 -07001964 spin_lock_irqsave(&list->lock, flags);
1965 __skb_unlink(skb, list);
1966 spin_unlock_irqrestore(&list->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001968EXPORT_SYMBOL(skb_unlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970/**
1971 * skb_append - append a buffer
1972 * @old: buffer to insert after
1973 * @newsk: buffer to insert
David S. Miller8728b832005-08-09 19:25:21 -07001974 * @list: list to use
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 *
1976 * Place a packet after a given packet in a list. The list locks are taken
1977 * and this function is atomic with respect to other list locked calls.
1978 * A buffer cannot be placed on two lists at the same time.
1979 */
David S. Miller8728b832005-08-09 19:25:21 -07001980void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981{
1982 unsigned long flags;
1983
David S. Miller8728b832005-08-09 19:25:21 -07001984 spin_lock_irqsave(&list->lock, flags);
Gerrit Renker7de6c032008-04-14 00:05:09 -07001985 __skb_queue_after(list, old, newsk);
David S. Miller8728b832005-08-09 19:25:21 -07001986 spin_unlock_irqrestore(&list->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001988EXPORT_SYMBOL(skb_append);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989
1990/**
1991 * skb_insert - insert a buffer
1992 * @old: buffer to insert before
1993 * @newsk: buffer to insert
David S. Miller8728b832005-08-09 19:25:21 -07001994 * @list: list to use
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 *
David S. Miller8728b832005-08-09 19:25:21 -07001996 * Place a packet before a given packet in a list. The list locks are
1997 * taken and this function is atomic with respect to other list locked
1998 * calls.
1999 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 * A buffer cannot be placed on two lists at the same time.
2001 */
David S. Miller8728b832005-08-09 19:25:21 -07002002void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003{
2004 unsigned long flags;
2005
David S. Miller8728b832005-08-09 19:25:21 -07002006 spin_lock_irqsave(&list->lock, flags);
2007 __skb_insert(newsk, old->prev, old, list);
2008 spin_unlock_irqrestore(&list->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002010EXPORT_SYMBOL(skb_insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012static inline void skb_split_inside_header(struct sk_buff *skb,
2013 struct sk_buff* skb1,
2014 const u32 len, const int pos)
2015{
2016 int i;
2017
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03002018 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2019 pos - len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 /* And move data appendix as is. */
2021 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2022 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2023
2024 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2025 skb_shinfo(skb)->nr_frags = 0;
2026 skb1->data_len = skb->data_len;
2027 skb1->len += skb1->data_len;
2028 skb->data_len = 0;
2029 skb->len = len;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07002030 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031}
2032
2033static inline void skb_split_no_header(struct sk_buff *skb,
2034 struct sk_buff* skb1,
2035 const u32 len, int pos)
2036{
2037 int i, k = 0;
2038 const int nfrags = skb_shinfo(skb)->nr_frags;
2039
2040 skb_shinfo(skb)->nr_frags = 0;
2041 skb1->len = skb1->data_len = skb->len - len;
2042 skb->len = len;
2043 skb->data_len = len - pos;
2044
2045 for (i = 0; i < nfrags; i++) {
2046 int size = skb_shinfo(skb)->frags[i].size;
2047
2048 if (pos + size > len) {
2049 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2050
2051 if (pos < len) {
2052 /* Split frag.
2053 * We have two variants in this case:
2054 * 1. Move all the frag to the second
2055 * part, if it is possible. F.e.
2056 * this approach is mandatory for TUX,
2057 * where splitting is expensive.
2058 * 2. Split is accurately. We make this.
2059 */
2060 get_page(skb_shinfo(skb)->frags[i].page);
2061 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2062 skb_shinfo(skb1)->frags[0].size -= len - pos;
2063 skb_shinfo(skb)->frags[i].size = len - pos;
2064 skb_shinfo(skb)->nr_frags++;
2065 }
2066 k++;
2067 } else
2068 skb_shinfo(skb)->nr_frags++;
2069 pos += size;
2070 }
2071 skb_shinfo(skb1)->nr_frags = k;
2072}
2073
2074/**
2075 * skb_split - Split fragmented skb to two parts at length len.
2076 * @skb: the buffer to split
2077 * @skb1: the buffer to receive the second part
2078 * @len: new length for skb
2079 */
2080void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2081{
2082 int pos = skb_headlen(skb);
2083
2084 if (len < pos) /* Split line is inside header. */
2085 skb_split_inside_header(skb, skb1, len, pos);
2086 else /* Second chunk has no header, nothing to copy. */
2087 skb_split_no_header(skb, skb1, len, pos);
2088}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002089EXPORT_SYMBOL(skb_split);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090
Ilpo Järvinen9f782db2008-11-25 13:57:01 -08002091/* Shifting from/to a cloned skb is a no-go.
2092 *
2093 * Caller cannot keep skb_shinfo related pointers past calling here!
2094 */
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002095static int skb_prepare_for_shift(struct sk_buff *skb)
2096{
Ilpo Järvinen0ace2852008-11-24 21:30:21 -08002097 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002098}
2099
2100/**
2101 * skb_shift - Shifts paged data partially from skb to another
2102 * @tgt: buffer into which tail data gets added
2103 * @skb: buffer from which the paged data comes from
2104 * @shiftlen: shift up to this many bytes
2105 *
2106 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2107 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2108 * It's up to caller to free skb if everything was shifted.
2109 *
2110 * If @tgt runs out of frags, the whole operation is aborted.
2111 *
2112 * Skb cannot include anything else but paged data while tgt is allowed
2113 * to have non-paged data as well.
2114 *
2115 * TODO: full sized shift could be optimized but that would need
2116 * specialized skb free'er to handle frags without up-to-date nr_frags.
2117 */
2118int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2119{
2120 int from, to, merge, todo;
2121 struct skb_frag_struct *fragfrom, *fragto;
2122
2123 BUG_ON(shiftlen > skb->len);
2124 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2125
2126 todo = shiftlen;
2127 from = 0;
2128 to = skb_shinfo(tgt)->nr_frags;
2129 fragfrom = &skb_shinfo(skb)->frags[from];
2130
2131 /* Actual merge is delayed until the point when we know we can
2132 * commit all, so that we don't have to undo partial changes
2133 */
2134 if (!to ||
2135 !skb_can_coalesce(tgt, to, fragfrom->page, fragfrom->page_offset)) {
2136 merge = -1;
2137 } else {
2138 merge = to - 1;
2139
2140 todo -= fragfrom->size;
2141 if (todo < 0) {
2142 if (skb_prepare_for_shift(skb) ||
2143 skb_prepare_for_shift(tgt))
2144 return 0;
2145
Ilpo Järvinen9f782db2008-11-25 13:57:01 -08002146 /* All previous frag pointers might be stale! */
2147 fragfrom = &skb_shinfo(skb)->frags[from];
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002148 fragto = &skb_shinfo(tgt)->frags[merge];
2149
2150 fragto->size += shiftlen;
2151 fragfrom->size -= shiftlen;
2152 fragfrom->page_offset += shiftlen;
2153
2154 goto onlymerged;
2155 }
2156
2157 from++;
2158 }
2159
2160 /* Skip full, not-fitting skb to avoid expensive operations */
2161 if ((shiftlen == skb->len) &&
2162 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2163 return 0;
2164
2165 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2166 return 0;
2167
2168 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2169 if (to == MAX_SKB_FRAGS)
2170 return 0;
2171
2172 fragfrom = &skb_shinfo(skb)->frags[from];
2173 fragto = &skb_shinfo(tgt)->frags[to];
2174
2175 if (todo >= fragfrom->size) {
2176 *fragto = *fragfrom;
2177 todo -= fragfrom->size;
2178 from++;
2179 to++;
2180
2181 } else {
2182 get_page(fragfrom->page);
2183 fragto->page = fragfrom->page;
2184 fragto->page_offset = fragfrom->page_offset;
2185 fragto->size = todo;
2186
2187 fragfrom->page_offset += todo;
2188 fragfrom->size -= todo;
2189 todo = 0;
2190
2191 to++;
2192 break;
2193 }
2194 }
2195
2196 /* Ready to "commit" this state change to tgt */
2197 skb_shinfo(tgt)->nr_frags = to;
2198
2199 if (merge >= 0) {
2200 fragfrom = &skb_shinfo(skb)->frags[0];
2201 fragto = &skb_shinfo(tgt)->frags[merge];
2202
2203 fragto->size += fragfrom->size;
2204 put_page(fragfrom->page);
2205 }
2206
2207 /* Reposition in the original skb */
2208 to = 0;
2209 while (from < skb_shinfo(skb)->nr_frags)
2210 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2211 skb_shinfo(skb)->nr_frags = to;
2212
2213 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2214
2215onlymerged:
2216 /* Most likely the tgt won't ever need its checksum anymore, skb on
2217 * the other hand might need it if it needs to be resent
2218 */
2219 tgt->ip_summed = CHECKSUM_PARTIAL;
2220 skb->ip_summed = CHECKSUM_PARTIAL;
2221
2222 /* Yak, is it really working this way? Some helper please? */
2223 skb->len -= shiftlen;
2224 skb->data_len -= shiftlen;
2225 skb->truesize -= shiftlen;
2226 tgt->len += shiftlen;
2227 tgt->data_len += shiftlen;
2228 tgt->truesize += shiftlen;
2229
2230 return shiftlen;
2231}
2232
Thomas Graf677e90e2005-06-23 20:59:51 -07002233/**
2234 * skb_prepare_seq_read - Prepare a sequential read of skb data
2235 * @skb: the buffer to read
2236 * @from: lower offset of data to be read
2237 * @to: upper offset of data to be read
2238 * @st: state variable
2239 *
2240 * Initializes the specified state variable. Must be called before
2241 * invoking skb_seq_read() for the first time.
2242 */
2243void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2244 unsigned int to, struct skb_seq_state *st)
2245{
2246 st->lower_offset = from;
2247 st->upper_offset = to;
2248 st->root_skb = st->cur_skb = skb;
2249 st->frag_idx = st->stepped_offset = 0;
2250 st->frag_data = NULL;
2251}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002252EXPORT_SYMBOL(skb_prepare_seq_read);
Thomas Graf677e90e2005-06-23 20:59:51 -07002253
2254/**
2255 * skb_seq_read - Sequentially read skb data
2256 * @consumed: number of bytes consumed by the caller so far
2257 * @data: destination pointer for data to be returned
2258 * @st: state variable
2259 *
2260 * Reads a block of skb data at &consumed relative to the
2261 * lower offset specified to skb_prepare_seq_read(). Assigns
2262 * the head of the data block to &data and returns the length
2263 * of the block or 0 if the end of the skb data or the upper
2264 * offset has been reached.
2265 *
2266 * The caller is not required to consume all of the data
2267 * returned, i.e. &consumed is typically set to the number
2268 * of bytes already consumed and the next call to
2269 * skb_seq_read() will return the remaining part of the block.
2270 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002271 * Note 1: The size of each block of data returned can be arbitrary,
Thomas Graf677e90e2005-06-23 20:59:51 -07002272 * this limitation is the cost for zerocopy seqeuental
2273 * reads of potentially non linear data.
2274 *
Randy Dunlapbc2cda12008-02-13 15:03:25 -08002275 * Note 2: Fragment lists within fragments are not implemented
Thomas Graf677e90e2005-06-23 20:59:51 -07002276 * at the moment, state->root_skb could be replaced with
2277 * a stack for this purpose.
2278 */
2279unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2280 struct skb_seq_state *st)
2281{
2282 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2283 skb_frag_t *frag;
2284
2285 if (unlikely(abs_offset >= st->upper_offset))
2286 return 0;
2287
2288next_skb:
Herbert Xu95e3b242009-01-29 16:07:52 -08002289 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
Thomas Graf677e90e2005-06-23 20:59:51 -07002290
Thomas Chenault995b3372009-05-18 21:43:27 -07002291 if (abs_offset < block_limit && !st->frag_data) {
Herbert Xu95e3b242009-01-29 16:07:52 -08002292 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
Thomas Graf677e90e2005-06-23 20:59:51 -07002293 return block_limit - abs_offset;
2294 }
2295
2296 if (st->frag_idx == 0 && !st->frag_data)
2297 st->stepped_offset += skb_headlen(st->cur_skb);
2298
2299 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2300 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2301 block_limit = frag->size + st->stepped_offset;
2302
2303 if (abs_offset < block_limit) {
2304 if (!st->frag_data)
2305 st->frag_data = kmap_skb_frag(frag);
2306
2307 *data = (u8 *) st->frag_data + frag->page_offset +
2308 (abs_offset - st->stepped_offset);
2309
2310 return block_limit - abs_offset;
2311 }
2312
2313 if (st->frag_data) {
2314 kunmap_skb_frag(st->frag_data);
2315 st->frag_data = NULL;
2316 }
2317
2318 st->frag_idx++;
2319 st->stepped_offset += frag->size;
2320 }
2321
Olaf Kirch5b5a60d2007-06-23 23:11:52 -07002322 if (st->frag_data) {
2323 kunmap_skb_frag(st->frag_data);
2324 st->frag_data = NULL;
2325 }
2326
David S. Miller21dc3302010-08-23 00:13:46 -07002327 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
Shyam Iyer71b33462009-01-29 16:12:42 -08002328 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
Thomas Graf677e90e2005-06-23 20:59:51 -07002329 st->frag_idx = 0;
2330 goto next_skb;
Shyam Iyer71b33462009-01-29 16:12:42 -08002331 } else if (st->cur_skb->next) {
2332 st->cur_skb = st->cur_skb->next;
Herbert Xu95e3b242009-01-29 16:07:52 -08002333 st->frag_idx = 0;
Thomas Graf677e90e2005-06-23 20:59:51 -07002334 goto next_skb;
2335 }
2336
2337 return 0;
2338}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002339EXPORT_SYMBOL(skb_seq_read);
Thomas Graf677e90e2005-06-23 20:59:51 -07002340
2341/**
2342 * skb_abort_seq_read - Abort a sequential read of skb data
2343 * @st: state variable
2344 *
2345 * Must be called if skb_seq_read() was not called until it
2346 * returned 0.
2347 */
2348void skb_abort_seq_read(struct skb_seq_state *st)
2349{
2350 if (st->frag_data)
2351 kunmap_skb_frag(st->frag_data);
2352}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002353EXPORT_SYMBOL(skb_abort_seq_read);
Thomas Graf677e90e2005-06-23 20:59:51 -07002354
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002355#define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2356
2357static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2358 struct ts_config *conf,
2359 struct ts_state *state)
2360{
2361 return skb_seq_read(offset, text, TS_SKB_CB(state));
2362}
2363
2364static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2365{
2366 skb_abort_seq_read(TS_SKB_CB(state));
2367}
2368
2369/**
2370 * skb_find_text - Find a text pattern in skb data
2371 * @skb: the buffer to look in
2372 * @from: search offset
2373 * @to: search limit
2374 * @config: textsearch configuration
2375 * @state: uninitialized textsearch state variable
2376 *
2377 * Finds a pattern in the skb data according to the specified
2378 * textsearch configuration. Use textsearch_next() to retrieve
2379 * subsequent occurrences of the pattern. Returns the offset
2380 * to the first occurrence or UINT_MAX if no match was found.
2381 */
2382unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2383 unsigned int to, struct ts_config *config,
2384 struct ts_state *state)
2385{
Phil Oesterf72b9482006-06-26 00:00:57 -07002386 unsigned int ret;
2387
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002388 config->get_next_block = skb_ts_get_next_block;
2389 config->finish = skb_ts_finish;
2390
2391 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2392
Phil Oesterf72b9482006-06-26 00:00:57 -07002393 ret = textsearch_find(config, state);
2394 return (ret <= to - from ? ret : UINT_MAX);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002395}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002396EXPORT_SYMBOL(skb_find_text);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002397
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002398/**
2399 * skb_append_datato_frags: - append the user data to a skb
2400 * @sk: sock structure
2401 * @skb: skb structure to be appened with user data.
2402 * @getfrag: call back function to be used for getting the user data
2403 * @from: pointer to user message iov
2404 * @length: length of the iov message
2405 *
2406 * Description: This procedure append the user data in the fragment part
2407 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2408 */
2409int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
Martin Waitzdab96302005-12-05 13:40:12 -08002410 int (*getfrag)(void *from, char *to, int offset,
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002411 int len, int odd, struct sk_buff *skb),
2412 void *from, int length)
2413{
2414 int frg_cnt = 0;
2415 skb_frag_t *frag = NULL;
2416 struct page *page = NULL;
2417 int copy, left;
2418 int offset = 0;
2419 int ret;
2420
2421 do {
2422 /* Return error if we don't have space for new frag */
2423 frg_cnt = skb_shinfo(skb)->nr_frags;
2424 if (frg_cnt >= MAX_SKB_FRAGS)
2425 return -EFAULT;
2426
2427 /* allocate a new page for next frag */
2428 page = alloc_pages(sk->sk_allocation, 0);
2429
2430 /* If alloc_page fails just return failure and caller will
2431 * free previous allocated pages by doing kfree_skb()
2432 */
2433 if (page == NULL)
2434 return -ENOMEM;
2435
2436 /* initialize the next frag */
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002437 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2438 skb->truesize += PAGE_SIZE;
2439 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2440
2441 /* get the new initialized frag */
2442 frg_cnt = skb_shinfo(skb)->nr_frags;
2443 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2444
2445 /* copy the user data to page */
2446 left = PAGE_SIZE - frag->page_offset;
2447 copy = (length > left)? left : length;
2448
2449 ret = getfrag(from, (page_address(frag->page) +
2450 frag->page_offset + frag->size),
2451 offset, copy, 0, skb);
2452 if (ret < 0)
2453 return -EFAULT;
2454
2455 /* copy was successful so update the size parameters */
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002456 frag->size += copy;
2457 skb->len += copy;
2458 skb->data_len += copy;
2459 offset += copy;
2460 length -= copy;
2461
2462 } while (length > 0);
2463
2464 return 0;
2465}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002466EXPORT_SYMBOL(skb_append_datato_frags);
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002467
Herbert Xucbb042f2006-03-20 22:43:56 -08002468/**
2469 * skb_pull_rcsum - pull skb and update receive checksum
2470 * @skb: buffer to update
Herbert Xucbb042f2006-03-20 22:43:56 -08002471 * @len: length of data pulled
2472 *
2473 * This function performs an skb_pull on the packet and updates
Urs Thuermannfee54fa2008-02-12 22:03:25 -08002474 * the CHECKSUM_COMPLETE checksum. It should be used on
Patrick McHardy84fa7932006-08-29 16:44:56 -07002475 * receive path processing instead of skb_pull unless you know
2476 * that the checksum difference is zero (e.g., a valid IP header)
2477 * or you are setting ip_summed to CHECKSUM_NONE.
Herbert Xucbb042f2006-03-20 22:43:56 -08002478 */
2479unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2480{
2481 BUG_ON(len > skb->len);
2482 skb->len -= len;
2483 BUG_ON(skb->len < skb->data_len);
2484 skb_postpull_rcsum(skb, skb->data, len);
2485 return skb->data += len;
2486}
Arnaldo Carvalho de Melof94691a2006-03-20 22:47:55 -08002487EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2488
Herbert Xuf4c50d92006-06-22 03:02:40 -07002489/**
2490 * skb_segment - Perform protocol segmentation on skb.
2491 * @skb: buffer to segment
Herbert Xu576a30e2006-06-27 13:22:38 -07002492 * @features: features for the output path (see dev->features)
Herbert Xuf4c50d92006-06-22 03:02:40 -07002493 *
2494 * This function performs segmentation on the given skb. It returns
Ben Hutchings4c821d72008-04-13 21:52:48 -07002495 * a pointer to the first in a list of new skbs for the segments.
2496 * In case of error it returns ERR_PTR(err).
Herbert Xuf4c50d92006-06-22 03:02:40 -07002497 */
Michał Mirosław04ed3e72011-01-24 15:32:47 -08002498struct sk_buff *skb_segment(struct sk_buff *skb, u32 features)
Herbert Xuf4c50d92006-06-22 03:02:40 -07002499{
2500 struct sk_buff *segs = NULL;
2501 struct sk_buff *tail = NULL;
Herbert Xu89319d382008-12-15 23:26:06 -08002502 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002503 unsigned int mss = skb_shinfo(skb)->gso_size;
Arnaldo Carvalho de Melo98e399f2007-03-19 15:33:04 -07002504 unsigned int doffset = skb->data - skb_mac_header(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002505 unsigned int offset = doffset;
2506 unsigned int headroom;
2507 unsigned int len;
Michał Mirosław04ed3e72011-01-24 15:32:47 -08002508 int sg = !!(features & NETIF_F_SG);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002509 int nfrags = skb_shinfo(skb)->nr_frags;
2510 int err = -ENOMEM;
2511 int i = 0;
2512 int pos;
2513
2514 __skb_push(skb, doffset);
2515 headroom = skb_headroom(skb);
2516 pos = skb_headlen(skb);
2517
2518 do {
2519 struct sk_buff *nskb;
2520 skb_frag_t *frag;
Herbert Xuc8884ed2006-10-29 15:59:41 -08002521 int hsize;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002522 int size;
2523
2524 len = skb->len - offset;
2525 if (len > mss)
2526 len = mss;
2527
2528 hsize = skb_headlen(skb) - offset;
2529 if (hsize < 0)
2530 hsize = 0;
Herbert Xuc8884ed2006-10-29 15:59:41 -08002531 if (hsize > len || !sg)
2532 hsize = len;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002533
Herbert Xu89319d382008-12-15 23:26:06 -08002534 if (!hsize && i >= nfrags) {
2535 BUG_ON(fskb->len != len);
2536
2537 pos += len;
2538 nskb = skb_clone(fskb, GFP_ATOMIC);
2539 fskb = fskb->next;
2540
2541 if (unlikely(!nskb))
2542 goto err;
2543
2544 hsize = skb_end_pointer(nskb) - nskb->head;
2545 if (skb_cow_head(nskb, doffset + headroom)) {
2546 kfree_skb(nskb);
2547 goto err;
2548 }
2549
2550 nskb->truesize += skb_end_pointer(nskb) - nskb->head -
2551 hsize;
2552 skb_release_head_state(nskb);
2553 __skb_push(nskb, doffset);
2554 } else {
2555 nskb = alloc_skb(hsize + doffset + headroom,
2556 GFP_ATOMIC);
2557
2558 if (unlikely(!nskb))
2559 goto err;
2560
2561 skb_reserve(nskb, headroom);
2562 __skb_put(nskb, doffset);
2563 }
Herbert Xuf4c50d92006-06-22 03:02:40 -07002564
2565 if (segs)
2566 tail->next = nskb;
2567 else
2568 segs = nskb;
2569 tail = nskb;
2570
Herbert Xu6f85a122008-08-15 14:55:02 -07002571 __copy_skb_header(nskb, skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002572 nskb->mac_len = skb->mac_len;
2573
Eric Dumazet3d3be432010-09-01 00:50:51 +00002574 /* nskb and skb might have different headroom */
2575 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2576 nskb->csum_start += skb_headroom(nskb) - headroom;
2577
Arnaldo Carvalho de Melo459a98e2007-03-19 15:30:44 -07002578 skb_reset_mac_header(nskb);
Arnaldo Carvalho de Meloddc7b8e2007-03-15 21:42:27 -03002579 skb_set_network_header(nskb, skb->mac_len);
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -07002580 nskb->transport_header = (nskb->network_header +
2581 skb_network_header_len(skb));
Herbert Xu89319d382008-12-15 23:26:06 -08002582 skb_copy_from_linear_data(skb, nskb->data, doffset);
2583
Herbert Xu2f181852009-03-28 23:39:18 -07002584 if (fskb != skb_shinfo(skb)->frag_list)
Herbert Xu89319d382008-12-15 23:26:06 -08002585 continue;
2586
Herbert Xuf4c50d92006-06-22 03:02:40 -07002587 if (!sg) {
Herbert Xu6f85a122008-08-15 14:55:02 -07002588 nskb->ip_summed = CHECKSUM_NONE;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002589 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2590 skb_put(nskb, len),
2591 len, 0);
2592 continue;
2593 }
2594
2595 frag = skb_shinfo(nskb)->frags;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002596
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03002597 skb_copy_from_linear_data_offset(skb, offset,
2598 skb_put(nskb, hsize), hsize);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002599
Herbert Xu89319d382008-12-15 23:26:06 -08002600 while (pos < offset + len && i < nfrags) {
Herbert Xuf4c50d92006-06-22 03:02:40 -07002601 *frag = skb_shinfo(skb)->frags[i];
2602 get_page(frag->page);
2603 size = frag->size;
2604
2605 if (pos < offset) {
2606 frag->page_offset += offset - pos;
2607 frag->size -= offset - pos;
2608 }
2609
Herbert Xu89319d382008-12-15 23:26:06 -08002610 skb_shinfo(nskb)->nr_frags++;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002611
2612 if (pos + size <= offset + len) {
2613 i++;
2614 pos += size;
2615 } else {
2616 frag->size -= pos + size - (offset + len);
Herbert Xu89319d382008-12-15 23:26:06 -08002617 goto skip_fraglist;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002618 }
2619
2620 frag++;
2621 }
2622
Herbert Xu89319d382008-12-15 23:26:06 -08002623 if (pos < offset + len) {
2624 struct sk_buff *fskb2 = fskb;
2625
2626 BUG_ON(pos + fskb->len != offset + len);
2627
2628 pos += fskb->len;
2629 fskb = fskb->next;
2630
2631 if (fskb2->next) {
2632 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2633 if (!fskb2)
2634 goto err;
2635 } else
2636 skb_get(fskb2);
2637
David S. Millerfbb398a2009-06-09 00:18:59 -07002638 SKB_FRAG_ASSERT(nskb);
Herbert Xu89319d382008-12-15 23:26:06 -08002639 skb_shinfo(nskb)->frag_list = fskb2;
2640 }
2641
2642skip_fraglist:
Herbert Xuf4c50d92006-06-22 03:02:40 -07002643 nskb->data_len = len - hsize;
2644 nskb->len += nskb->data_len;
2645 nskb->truesize += nskb->data_len;
2646 } while ((offset += len) < skb->len);
2647
2648 return segs;
2649
2650err:
2651 while ((skb = segs)) {
2652 segs = skb->next;
Patrick McHardyb08d5842007-02-27 09:57:37 -08002653 kfree_skb(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002654 }
2655 return ERR_PTR(err);
2656}
Herbert Xuf4c50d92006-06-22 03:02:40 -07002657EXPORT_SYMBOL_GPL(skb_segment);
2658
Herbert Xu71d93b32008-12-15 23:42:33 -08002659int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2660{
2661 struct sk_buff *p = *head;
2662 struct sk_buff *nskb;
Herbert Xu9aaa1562009-05-26 18:50:33 +00002663 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2664 struct skb_shared_info *pinfo = skb_shinfo(p);
Herbert Xu71d93b32008-12-15 23:42:33 -08002665 unsigned int headroom;
Herbert Xu86911732009-01-29 14:19:50 +00002666 unsigned int len = skb_gro_len(skb);
Herbert Xu67147ba2009-05-26 18:50:22 +00002667 unsigned int offset = skb_gro_offset(skb);
2668 unsigned int headlen = skb_headlen(skb);
Herbert Xu71d93b32008-12-15 23:42:33 -08002669
Herbert Xu86911732009-01-29 14:19:50 +00002670 if (p->len + len >= 65536)
Herbert Xu71d93b32008-12-15 23:42:33 -08002671 return -E2BIG;
2672
Herbert Xu9aaa1562009-05-26 18:50:33 +00002673 if (pinfo->frag_list)
Herbert Xu71d93b32008-12-15 23:42:33 -08002674 goto merge;
Herbert Xu67147ba2009-05-26 18:50:22 +00002675 else if (headlen <= offset) {
Herbert Xu42da6992009-05-26 18:50:19 +00002676 skb_frag_t *frag;
Herbert Xu66e92fc2009-05-26 18:50:32 +00002677 skb_frag_t *frag2;
Herbert Xu9aaa1562009-05-26 18:50:33 +00002678 int i = skbinfo->nr_frags;
2679 int nr_frags = pinfo->nr_frags + i;
Herbert Xu42da6992009-05-26 18:50:19 +00002680
Herbert Xu66e92fc2009-05-26 18:50:32 +00002681 offset -= headlen;
2682
2683 if (nr_frags > MAX_SKB_FRAGS)
Herbert Xu81705ad2009-01-29 14:19:51 +00002684 return -E2BIG;
2685
Herbert Xu9aaa1562009-05-26 18:50:33 +00002686 pinfo->nr_frags = nr_frags;
2687 skbinfo->nr_frags = 0;
Herbert Xuf5572062009-01-14 20:40:03 -08002688
Herbert Xu9aaa1562009-05-26 18:50:33 +00002689 frag = pinfo->frags + nr_frags;
2690 frag2 = skbinfo->frags + i;
Herbert Xu66e92fc2009-05-26 18:50:32 +00002691 do {
2692 *--frag = *--frag2;
2693 } while (--i);
2694
2695 frag->page_offset += offset;
2696 frag->size -= offset;
2697
Herbert Xuf5572062009-01-14 20:40:03 -08002698 skb->truesize -= skb->data_len;
2699 skb->len -= skb->data_len;
2700 skb->data_len = 0;
2701
Herbert Xu5d38a072009-01-04 16:13:40 -08002702 NAPI_GRO_CB(skb)->free = 1;
2703 goto done;
Herbert Xu69c0cab2009-11-17 05:18:18 -08002704 } else if (skb_gro_len(p) != pinfo->gso_size)
2705 return -E2BIG;
Herbert Xu71d93b32008-12-15 23:42:33 -08002706
2707 headroom = skb_headroom(p);
Eric Dumazet3d3be432010-09-01 00:50:51 +00002708 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
Herbert Xu71d93b32008-12-15 23:42:33 -08002709 if (unlikely(!nskb))
2710 return -ENOMEM;
2711
2712 __copy_skb_header(nskb, p);
2713 nskb->mac_len = p->mac_len;
2714
2715 skb_reserve(nskb, headroom);
Herbert Xu86911732009-01-29 14:19:50 +00002716 __skb_put(nskb, skb_gro_offset(p));
Herbert Xu71d93b32008-12-15 23:42:33 -08002717
Herbert Xu86911732009-01-29 14:19:50 +00002718 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
Herbert Xu71d93b32008-12-15 23:42:33 -08002719 skb_set_network_header(nskb, skb_network_offset(p));
2720 skb_set_transport_header(nskb, skb_transport_offset(p));
2721
Herbert Xu86911732009-01-29 14:19:50 +00002722 __skb_pull(p, skb_gro_offset(p));
2723 memcpy(skb_mac_header(nskb), skb_mac_header(p),
2724 p->data - skb_mac_header(p));
Herbert Xu71d93b32008-12-15 23:42:33 -08002725
2726 *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2727 skb_shinfo(nskb)->frag_list = p;
Herbert Xu9aaa1562009-05-26 18:50:33 +00002728 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
Herbert Xu622e0ca2010-05-20 23:07:56 -07002729 pinfo->gso_size = 0;
Herbert Xu71d93b32008-12-15 23:42:33 -08002730 skb_header_release(p);
2731 nskb->prev = p;
2732
2733 nskb->data_len += p->len;
2734 nskb->truesize += p->len;
2735 nskb->len += p->len;
2736
2737 *head = nskb;
2738 nskb->next = p->next;
2739 p->next = NULL;
2740
2741 p = nskb;
2742
2743merge:
Herbert Xu67147ba2009-05-26 18:50:22 +00002744 if (offset > headlen) {
Michal Schmidtd1dc7ab2011-01-24 12:08:48 +00002745 unsigned int eat = offset - headlen;
2746
2747 skbinfo->frags[0].page_offset += eat;
2748 skbinfo->frags[0].size -= eat;
2749 skb->data_len -= eat;
2750 skb->len -= eat;
Herbert Xu67147ba2009-05-26 18:50:22 +00002751 offset = headlen;
Herbert Xu56035022009-02-05 21:26:52 -08002752 }
2753
Herbert Xu67147ba2009-05-26 18:50:22 +00002754 __skb_pull(skb, offset);
Herbert Xu56035022009-02-05 21:26:52 -08002755
Herbert Xu71d93b32008-12-15 23:42:33 -08002756 p->prev->next = skb;
2757 p->prev = skb;
2758 skb_header_release(skb);
2759
Herbert Xu5d38a072009-01-04 16:13:40 -08002760done:
2761 NAPI_GRO_CB(p)->count++;
Herbert Xu37fe4732009-01-17 19:48:13 +00002762 p->data_len += len;
2763 p->truesize += len;
2764 p->len += len;
Herbert Xu71d93b32008-12-15 23:42:33 -08002765
2766 NAPI_GRO_CB(skb)->same_flow = 1;
2767 return 0;
2768}
2769EXPORT_SYMBOL_GPL(skb_gro_receive);
2770
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771void __init skb_init(void)
2772{
2773 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2774 sizeof(struct sk_buff),
2775 0,
Alexey Dobriyane5d679f332006-08-26 19:25:52 -07002776 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
Paul Mundt20c2df82007-07-20 10:11:58 +09002777 NULL);
David S. Millerd179cd12005-08-17 14:57:30 -07002778 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2779 (2*sizeof(struct sk_buff)) +
2780 sizeof(atomic_t),
2781 0,
Alexey Dobriyane5d679f332006-08-26 19:25:52 -07002782 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
Paul Mundt20c2df82007-07-20 10:11:58 +09002783 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784}
2785
David Howells716ea3a2007-04-02 20:19:53 -07002786/**
2787 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2788 * @skb: Socket buffer containing the buffers to be mapped
2789 * @sg: The scatter-gather list to map into
2790 * @offset: The offset into the buffer's contents to start mapping
2791 * @len: Length of buffer space to be mapped
2792 *
2793 * Fill the specified scatter-gather list with mappings/pointers into a
2794 * region of the buffer space attached to a socket buffer.
2795 */
David S. Miller51c739d2007-10-30 21:29:29 -07002796static int
2797__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
David Howells716ea3a2007-04-02 20:19:53 -07002798{
David S. Miller1a028e52007-04-27 15:21:23 -07002799 int start = skb_headlen(skb);
2800 int i, copy = start - offset;
David S. Millerfbb398a2009-06-09 00:18:59 -07002801 struct sk_buff *frag_iter;
David Howells716ea3a2007-04-02 20:19:53 -07002802 int elt = 0;
2803
2804 if (copy > 0) {
2805 if (copy > len)
2806 copy = len;
Jens Axboe642f1492007-10-24 11:20:47 +02002807 sg_set_buf(sg, skb->data + offset, copy);
David Howells716ea3a2007-04-02 20:19:53 -07002808 elt++;
2809 if ((len -= copy) == 0)
2810 return elt;
2811 offset += copy;
2812 }
2813
2814 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07002815 int end;
David Howells716ea3a2007-04-02 20:19:53 -07002816
Ilpo Järvinen547b7922008-07-25 21:43:18 -07002817 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07002818
2819 end = start + skb_shinfo(skb)->frags[i].size;
David Howells716ea3a2007-04-02 20:19:53 -07002820 if ((copy = end - offset) > 0) {
2821 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2822
2823 if (copy > len)
2824 copy = len;
Jens Axboe642f1492007-10-24 11:20:47 +02002825 sg_set_page(&sg[elt], frag->page, copy,
2826 frag->page_offset+offset-start);
David Howells716ea3a2007-04-02 20:19:53 -07002827 elt++;
2828 if (!(len -= copy))
2829 return elt;
2830 offset += copy;
2831 }
David S. Miller1a028e52007-04-27 15:21:23 -07002832 start = end;
David Howells716ea3a2007-04-02 20:19:53 -07002833 }
2834
David S. Millerfbb398a2009-06-09 00:18:59 -07002835 skb_walk_frags(skb, frag_iter) {
2836 int end;
David Howells716ea3a2007-04-02 20:19:53 -07002837
David S. Millerfbb398a2009-06-09 00:18:59 -07002838 WARN_ON(start > offset + len);
David Howells716ea3a2007-04-02 20:19:53 -07002839
David S. Millerfbb398a2009-06-09 00:18:59 -07002840 end = start + frag_iter->len;
2841 if ((copy = end - offset) > 0) {
2842 if (copy > len)
2843 copy = len;
2844 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
2845 copy);
2846 if ((len -= copy) == 0)
2847 return elt;
2848 offset += copy;
David Howells716ea3a2007-04-02 20:19:53 -07002849 }
David S. Millerfbb398a2009-06-09 00:18:59 -07002850 start = end;
David Howells716ea3a2007-04-02 20:19:53 -07002851 }
2852 BUG_ON(len);
2853 return elt;
2854}
2855
David S. Miller51c739d2007-10-30 21:29:29 -07002856int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2857{
2858 int nsg = __skb_to_sgvec(skb, sg, offset, len);
2859
Jens Axboec46f2332007-10-31 12:06:37 +01002860 sg_mark_end(&sg[nsg - 1]);
David S. Miller51c739d2007-10-30 21:29:29 -07002861
2862 return nsg;
2863}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002864EXPORT_SYMBOL_GPL(skb_to_sgvec);
David S. Miller51c739d2007-10-30 21:29:29 -07002865
David Howells716ea3a2007-04-02 20:19:53 -07002866/**
2867 * skb_cow_data - Check that a socket buffer's data buffers are writable
2868 * @skb: The socket buffer to check.
2869 * @tailbits: Amount of trailing space to be added
2870 * @trailer: Returned pointer to the skb where the @tailbits space begins
2871 *
2872 * Make sure that the data buffers attached to a socket buffer are
2873 * writable. If they are not, private copies are made of the data buffers
2874 * and the socket buffer is set to use these instead.
2875 *
2876 * If @tailbits is given, make sure that there is space to write @tailbits
2877 * bytes of data beyond current end of socket buffer. @trailer will be
2878 * set to point to the skb in which this space begins.
2879 *
2880 * The number of scatterlist elements required to completely map the
2881 * COW'd and extended socket buffer will be returned.
2882 */
2883int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
2884{
2885 int copyflag;
2886 int elt;
2887 struct sk_buff *skb1, **skb_p;
2888
2889 /* If skb is cloned or its head is paged, reallocate
2890 * head pulling out all the pages (pages are considered not writable
2891 * at the moment even if they are anonymous).
2892 */
2893 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
2894 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
2895 return -ENOMEM;
2896
2897 /* Easy case. Most of packets will go this way. */
David S. Miller21dc3302010-08-23 00:13:46 -07002898 if (!skb_has_frag_list(skb)) {
David Howells716ea3a2007-04-02 20:19:53 -07002899 /* A little of trouble, not enough of space for trailer.
2900 * This should not happen, when stack is tuned to generate
2901 * good frames. OK, on miss we reallocate and reserve even more
2902 * space, 128 bytes is fair. */
2903
2904 if (skb_tailroom(skb) < tailbits &&
2905 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
2906 return -ENOMEM;
2907
2908 /* Voila! */
2909 *trailer = skb;
2910 return 1;
2911 }
2912
2913 /* Misery. We are in troubles, going to mincer fragments... */
2914
2915 elt = 1;
2916 skb_p = &skb_shinfo(skb)->frag_list;
2917 copyflag = 0;
2918
2919 while ((skb1 = *skb_p) != NULL) {
2920 int ntail = 0;
2921
2922 /* The fragment is partially pulled by someone,
2923 * this can happen on input. Copy it and everything
2924 * after it. */
2925
2926 if (skb_shared(skb1))
2927 copyflag = 1;
2928
2929 /* If the skb is the last, worry about trailer. */
2930
2931 if (skb1->next == NULL && tailbits) {
2932 if (skb_shinfo(skb1)->nr_frags ||
David S. Miller21dc3302010-08-23 00:13:46 -07002933 skb_has_frag_list(skb1) ||
David Howells716ea3a2007-04-02 20:19:53 -07002934 skb_tailroom(skb1) < tailbits)
2935 ntail = tailbits + 128;
2936 }
2937
2938 if (copyflag ||
2939 skb_cloned(skb1) ||
2940 ntail ||
2941 skb_shinfo(skb1)->nr_frags ||
David S. Miller21dc3302010-08-23 00:13:46 -07002942 skb_has_frag_list(skb1)) {
David Howells716ea3a2007-04-02 20:19:53 -07002943 struct sk_buff *skb2;
2944
2945 /* Fuck, we are miserable poor guys... */
2946 if (ntail == 0)
2947 skb2 = skb_copy(skb1, GFP_ATOMIC);
2948 else
2949 skb2 = skb_copy_expand(skb1,
2950 skb_headroom(skb1),
2951 ntail,
2952 GFP_ATOMIC);
2953 if (unlikely(skb2 == NULL))
2954 return -ENOMEM;
2955
2956 if (skb1->sk)
2957 skb_set_owner_w(skb2, skb1->sk);
2958
2959 /* Looking around. Are we still alive?
2960 * OK, link new skb, drop old one */
2961
2962 skb2->next = skb1->next;
2963 *skb_p = skb2;
2964 kfree_skb(skb1);
2965 skb1 = skb2;
2966 }
2967 elt++;
2968 *trailer = skb1;
2969 skb_p = &skb1->next;
2970 }
2971
2972 return elt;
2973}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002974EXPORT_SYMBOL_GPL(skb_cow_data);
David Howells716ea3a2007-04-02 20:19:53 -07002975
Eric Dumazetb1faf562010-05-31 23:44:05 -07002976static void sock_rmem_free(struct sk_buff *skb)
2977{
2978 struct sock *sk = skb->sk;
2979
2980 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
2981}
2982
2983/*
2984 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
2985 */
2986int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
2987{
2988 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
2989 (unsigned)sk->sk_rcvbuf)
2990 return -ENOMEM;
2991
2992 skb_orphan(skb);
2993 skb->sk = sk;
2994 skb->destructor = sock_rmem_free;
2995 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2996
Eric Dumazetabb57ea2011-05-18 02:21:31 -04002997 /* before exiting rcu section, make sure dst is refcounted */
2998 skb_dst_force(skb);
2999
Eric Dumazetb1faf562010-05-31 23:44:05 -07003000 skb_queue_tail(&sk->sk_error_queue, skb);
3001 if (!sock_flag(sk, SOCK_DEAD))
3002 sk->sk_data_ready(sk, skb->len);
3003 return 0;
3004}
3005EXPORT_SYMBOL(sock_queue_err_skb);
3006
Patrick Ohlyac45f602009-02-12 05:03:37 +00003007void skb_tstamp_tx(struct sk_buff *orig_skb,
3008 struct skb_shared_hwtstamps *hwtstamps)
3009{
3010 struct sock *sk = orig_skb->sk;
3011 struct sock_exterr_skb *serr;
3012 struct sk_buff *skb;
3013 int err;
3014
3015 if (!sk)
3016 return;
3017
3018 skb = skb_clone(orig_skb, GFP_ATOMIC);
3019 if (!skb)
3020 return;
3021
3022 if (hwtstamps) {
3023 *skb_hwtstamps(skb) =
3024 *hwtstamps;
3025 } else {
3026 /*
3027 * no hardware time stamps available,
Oliver Hartkopp2244d072010-08-17 08:59:14 +00003028 * so keep the shared tx_flags and only
Patrick Ohlyac45f602009-02-12 05:03:37 +00003029 * store software time stamp
3030 */
3031 skb->tstamp = ktime_get_real();
3032 }
3033
3034 serr = SKB_EXT_ERR(skb);
3035 memset(serr, 0, sizeof(*serr));
3036 serr->ee.ee_errno = ENOMSG;
3037 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
Eric Dumazet29030372010-05-29 00:20:48 -07003038
Patrick Ohlyac45f602009-02-12 05:03:37 +00003039 err = sock_queue_err_skb(sk, skb);
Eric Dumazet29030372010-05-29 00:20:48 -07003040
Patrick Ohlyac45f602009-02-12 05:03:37 +00003041 if (err)
3042 kfree_skb(skb);
3043}
3044EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3045
3046
Rusty Russellf35d9d82008-02-04 23:49:54 -05003047/**
3048 * skb_partial_csum_set - set up and verify partial csum values for packet
3049 * @skb: the skb to set
3050 * @start: the number of bytes after skb->data to start checksumming.
3051 * @off: the offset from start to place the checksum.
3052 *
3053 * For untrusted partially-checksummed packets, we need to make sure the values
3054 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3055 *
3056 * This function checks and sets those values and skb->ip_summed: if this
3057 * returns false you should drop the packet.
3058 */
3059bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3060{
Herbert Xu5ff8dda2009-06-04 01:22:01 +00003061 if (unlikely(start > skb_headlen(skb)) ||
3062 unlikely((int)start + off > skb_headlen(skb) - 2)) {
Rusty Russellf35d9d82008-02-04 23:49:54 -05003063 if (net_ratelimit())
3064 printk(KERN_WARNING
3065 "bad partial csum: csum=%u/%u len=%u\n",
Herbert Xu5ff8dda2009-06-04 01:22:01 +00003066 start, off, skb_headlen(skb));
Rusty Russellf35d9d82008-02-04 23:49:54 -05003067 return false;
3068 }
3069 skb->ip_summed = CHECKSUM_PARTIAL;
3070 skb->csum_start = skb_headroom(skb) + start;
3071 skb->csum_offset = off;
3072 return true;
3073}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08003074EXPORT_SYMBOL_GPL(skb_partial_csum_set);
Rusty Russellf35d9d82008-02-04 23:49:54 -05003075
Ben Hutchings4497b072008-06-19 16:22:28 -07003076void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3077{
3078 if (net_ratelimit())
3079 pr_warning("%s: received packets cannot be forwarded"
3080 " while LRO is enabled\n", skb->dev->name);
3081}
Ben Hutchings4497b072008-06-19 16:22:28 -07003082EXPORT_SYMBOL(__skb_warn_lro_forwarding);