blob: 43c490d52df63f5f959c04cb210d164b0eb43959 [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>
Steven Rostedtad8d75f2009-04-14 19:39:12 -040069#include <trace/events/skb.h>
Eric Dumazet51c56b02012-04-05 11:35:15 +020070#include <linux/highmem.h>
Al Viroa1f8e7f72006-10-19 16:08:53 -040071
Christoph Lametere18b8902006-12-06 20:33:20 -080072static struct kmem_cache *skbuff_head_cache __read_mostly;
73static struct kmem_cache *skbuff_fclone_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070074
Jens Axboe9c55e012007-11-06 23:30:13 -080075static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
76 struct pipe_buffer *buf)
77{
Jarek Poplawski8b9d3722009-01-19 17:03:56 -080078 put_page(buf->page);
Jens Axboe9c55e012007-11-06 23:30:13 -080079}
80
81static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
82 struct pipe_buffer *buf)
83{
Jarek Poplawski8b9d3722009-01-19 17:03:56 -080084 get_page(buf->page);
Jens Axboe9c55e012007-11-06 23:30:13 -080085}
86
87static int sock_pipe_buf_steal(struct pipe_inode_info *pipe,
88 struct pipe_buffer *buf)
89{
90 return 1;
91}
92
93
94/* Pipe buffer operations for a socket. */
Alexey Dobriyan28dfef82009-12-15 16:46:48 -080095static const struct pipe_buf_operations sock_pipe_buf_ops = {
Jens Axboe9c55e012007-11-06 23:30:13 -080096 .can_merge = 0,
97 .map = generic_pipe_buf_map,
98 .unmap = generic_pipe_buf_unmap,
99 .confirm = generic_pipe_buf_confirm,
100 .release = sock_pipe_buf_release,
101 .steal = sock_pipe_buf_steal,
102 .get = sock_pipe_buf_get,
103};
104
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105/*
106 * Keep out-of-line to prevent kernel bloat.
107 * __builtin_return_address is not used because it is not always
108 * reliable.
109 */
110
111/**
112 * skb_over_panic - private function
113 * @skb: buffer
114 * @sz: size
115 * @here: address
116 *
117 * Out of line support code for skb_put(). Not user callable.
118 */
Rami Rosenccb7c772010-04-20 22:39:53 -0700119static void skb_over_panic(struct sk_buff *skb, int sz, void *here)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120{
Patrick McHardy26095452005-04-21 16:43:02 -0700121 printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700122 "data:%p tail:%#lx end:%#lx dev:%s\n",
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700123 here, skb->len, sz, skb->head, skb->data,
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700124 (unsigned long)skb->tail, (unsigned long)skb->end,
Patrick McHardy26095452005-04-21 16:43:02 -0700125 skb->dev ? skb->dev->name : "<NULL>");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 BUG();
127}
128
129/**
130 * skb_under_panic - private function
131 * @skb: buffer
132 * @sz: size
133 * @here: address
134 *
135 * Out of line support code for skb_push(). Not user callable.
136 */
137
Rami Rosenccb7c772010-04-20 22:39:53 -0700138static void skb_under_panic(struct sk_buff *skb, int sz, void *here)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139{
Patrick McHardy26095452005-04-21 16:43:02 -0700140 printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700141 "data:%p tail:%#lx end:%#lx dev:%s\n",
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700142 here, skb->len, sz, skb->head, skb->data,
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700143 (unsigned long)skb->tail, (unsigned long)skb->end,
Patrick McHardy26095452005-04-21 16:43:02 -0700144 skb->dev ? skb->dev->name : "<NULL>");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 BUG();
146}
147
148/* Allocate a new skbuff. We do this ourselves so we can fill in a few
149 * 'private' fields and also do memory statistics to find all the
150 * [BEEP] leaks.
151 *
152 */
153
154/**
David S. Millerd179cd12005-08-17 14:57:30 -0700155 * __alloc_skb - allocate a network buffer
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 * @size: size to allocate
157 * @gfp_mask: allocation mask
Randy Dunlapc83c2482005-10-18 22:07:41 -0700158 * @fclone: allocate from fclone cache instead of head cache
159 * and allocate a cloned (child) skb
Christoph Hellwigb30973f2006-12-06 20:32:36 -0800160 * @node: numa node to allocate memory on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 *
162 * Allocate a new &sk_buff. The returned buffer has no headroom and a
163 * tail room of size bytes. The object has a reference count of one.
164 * The return is the buffer. On a failure the return is %NULL.
165 *
166 * Buffers may only be allocated from interrupts using a @gfp_mask of
167 * %GFP_ATOMIC.
168 */
Al Virodd0fc662005-10-07 07:46:04 +0100169struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
Christoph Hellwigb30973f2006-12-06 20:32:36 -0800170 int fclone, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171{
Christoph Lametere18b8902006-12-06 20:33:20 -0800172 struct kmem_cache *cache;
Benjamin LaHaise4947d3e2006-01-03 14:06:50 -0800173 struct skb_shared_info *shinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 struct sk_buff *skb;
175 u8 *data;
176
Herbert Xu8798b3f2006-01-23 16:32:45 -0800177 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
178
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 /* Get the HEAD */
Christoph Hellwigb30973f2006-12-06 20:32:36 -0800180 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 if (!skb)
182 goto out;
Eric Dumazetec7d2f22010-05-05 01:07:37 -0700183 prefetchw(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184
Eric Dumazet87fb4b72011-10-13 07:28:54 +0000185 /* We do our best to align skb_shared_info on a separate cache
186 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
187 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
188 * Both skb->head and skb_shared_info are cache line aligned.
189 */
Tony Lindgrenbc417e32011-11-02 13:40:28 +0000190 size = SKB_DATA_ALIGN(size);
Eric Dumazet87fb4b72011-10-13 07:28:54 +0000191 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
192 data = kmalloc_node_track_caller(size, gfp_mask, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 if (!data)
194 goto nodata;
Eric Dumazet87fb4b72011-10-13 07:28:54 +0000195 /* kmalloc(size) might give us more room than requested.
196 * Put skb_shared_info exactly at the end of allocated zone,
197 * to allow max possible filling before reallocation.
198 */
199 size = SKB_WITH_OVERHEAD(ksize(data));
Eric Dumazetec7d2f22010-05-05 01:07:37 -0700200 prefetchw(data + size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201
Arnaldo Carvalho de Meloca0605a2007-03-19 10:48:59 -0300202 /*
Johannes Bergc8005782008-05-03 20:56:42 -0700203 * Only clear those fields we need to clear, not those that we will
204 * actually initialise below. Hence, don't put any more fields after
205 * the tail pointer in struct sk_buff!
Arnaldo Carvalho de Meloca0605a2007-03-19 10:48:59 -0300206 */
207 memset(skb, 0, offsetof(struct sk_buff, tail));
Eric Dumazet87fb4b72011-10-13 07:28:54 +0000208 /* Account for allocated memory : skb + skb->head */
209 skb->truesize = SKB_TRUESIZE(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 atomic_set(&skb->users, 1);
211 skb->head = data;
212 skb->data = data;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700213 skb_reset_tail_pointer(skb);
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700214 skb->end = skb->tail + size;
Stephen Hemminger19633e12009-06-17 05:23:27 +0000215#ifdef NET_SKBUFF_DATA_USES_OFFSET
216 skb->mac_header = ~0U;
217#endif
218
Benjamin LaHaise4947d3e2006-01-03 14:06:50 -0800219 /* make sure we initialize shinfo sequentially */
220 shinfo = skb_shinfo(skb);
Eric Dumazetec7d2f22010-05-05 01:07:37 -0700221 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
Benjamin LaHaise4947d3e2006-01-03 14:06:50 -0800222 atomic_set(&shinfo->dataref, 1);
Eric Dumazetc2aa3662011-01-25 23:18:38 +0000223 kmemcheck_annotate_variable(shinfo->destructor_arg);
Benjamin LaHaise4947d3e2006-01-03 14:06:50 -0800224
David S. Millerd179cd12005-08-17 14:57:30 -0700225 if (fclone) {
226 struct sk_buff *child = skb + 1;
227 atomic_t *fclone_ref = (atomic_t *) (child + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200229 kmemcheck_annotate_bitfield(child, flags1);
230 kmemcheck_annotate_bitfield(child, flags2);
David S. Millerd179cd12005-08-17 14:57:30 -0700231 skb->fclone = SKB_FCLONE_ORIG;
232 atomic_set(fclone_ref, 1);
233
234 child->fclone = SKB_FCLONE_UNAVAILABLE;
235 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236out:
237 return skb;
238nodata:
Herbert Xu8798b3f2006-01-23 16:32:45 -0800239 kmem_cache_free(cache, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 skb = NULL;
241 goto out;
242}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800243EXPORT_SYMBOL(__alloc_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244
245/**
Eric Dumazetb2b5ce92011-11-14 06:03:34 +0000246 * build_skb - build a network buffer
247 * @data: data buffer provided by caller
248 *
249 * Allocate a new &sk_buff. Caller provides space holding head and
250 * skb_shared_info. @data must have been allocated by kmalloc()
251 * The return is the new skb buffer.
252 * On a failure the return is %NULL, and @data is not freed.
253 * Notes :
254 * Before IO, driver allocates only data buffer where NIC put incoming frame
255 * Driver should add room at head (NET_SKB_PAD) and
256 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
257 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
258 * before giving packet to stack.
259 * RX rings only contains data buffers, not full skbs.
260 */
261struct sk_buff *build_skb(void *data)
262{
263 struct skb_shared_info *shinfo;
264 struct sk_buff *skb;
265 unsigned int size;
266
267 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
268 if (!skb)
269 return NULL;
270
271 size = ksize(data) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
272
273 memset(skb, 0, offsetof(struct sk_buff, tail));
274 skb->truesize = SKB_TRUESIZE(size);
275 atomic_set(&skb->users, 1);
276 skb->head = data;
277 skb->data = data;
278 skb_reset_tail_pointer(skb);
279 skb->end = skb->tail + size;
280#ifdef NET_SKBUFF_DATA_USES_OFFSET
281 skb->mac_header = ~0U;
282#endif
283
284 /* make sure we initialize shinfo sequentially */
285 shinfo = skb_shinfo(skb);
286 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
287 atomic_set(&shinfo->dataref, 1);
288 kmemcheck_annotate_variable(shinfo->destructor_arg);
289
290 return skb;
291}
292EXPORT_SYMBOL(build_skb);
293
294/**
Christoph Hellwig8af27452006-07-31 22:35:23 -0700295 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
296 * @dev: network device to receive on
297 * @length: length to allocate
298 * @gfp_mask: get_free_pages mask, passed to alloc_skb
299 *
300 * Allocate a new &sk_buff and assign it a usage count of one. The
301 * buffer has unspecified headroom built in. Users should allocate
302 * the headroom they think they need without accounting for the
303 * built in space. The built in space is used for optimisations.
304 *
305 * %NULL is returned if there is no free memory.
306 */
307struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
308 unsigned int length, gfp_t gfp_mask)
309{
310 struct sk_buff *skb;
311
Eric Dumazet564824b2010-10-11 19:05:25 +0000312 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, NUMA_NO_NODE);
Christoph Hellwig7b2e4972006-08-07 16:09:04 -0700313 if (likely(skb)) {
Christoph Hellwig8af27452006-07-31 22:35:23 -0700314 skb_reserve(skb, NET_SKB_PAD);
Christoph Hellwig7b2e4972006-08-07 16:09:04 -0700315 skb->dev = dev;
316 }
Christoph Hellwig8af27452006-07-31 22:35:23 -0700317 return skb;
318}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800319EXPORT_SYMBOL(__netdev_alloc_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320
Peter Zijlstra654bed12008-10-07 14:22:33 -0700321void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
Eric Dumazet50269e12012-03-23 23:59:33 +0000322 int size, unsigned int truesize)
Peter Zijlstra654bed12008-10-07 14:22:33 -0700323{
324 skb_fill_page_desc(skb, i, page, off, size);
325 skb->len += size;
326 skb->data_len += size;
Eric Dumazet50269e12012-03-23 23:59:33 +0000327 skb->truesize += truesize;
Peter Zijlstra654bed12008-10-07 14:22:33 -0700328}
329EXPORT_SYMBOL(skb_add_rx_frag);
330
Ilpo Järvinenf58518e2008-03-27 17:51:31 -0700331/**
332 * dev_alloc_skb - allocate an skbuff for receiving
333 * @length: length to allocate
334 *
335 * Allocate a new &sk_buff and assign it a usage count of one. The
336 * buffer has unspecified headroom built in. Users should allocate
337 * the headroom they think they need without accounting for the
338 * built in space. The built in space is used for optimisations.
339 *
340 * %NULL is returned if there is no free memory. Although this function
341 * allocates memory it can be called from an interrupt.
342 */
343struct sk_buff *dev_alloc_skb(unsigned int length)
344{
Denys Vlasenko1483b872008-03-28 15:57:39 -0700345 /*
346 * There is more code here than it seems:
David S. Millera0f55e02008-03-28 18:22:32 -0700347 * __dev_alloc_skb is an inline
Denys Vlasenko1483b872008-03-28 15:57:39 -0700348 */
Ilpo Järvinenf58518e2008-03-27 17:51:31 -0700349 return __dev_alloc_skb(length, GFP_ATOMIC);
350}
351EXPORT_SYMBOL(dev_alloc_skb);
352
Herbert Xu27b437c2006-07-13 19:26:39 -0700353static void skb_drop_list(struct sk_buff **listp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354{
Herbert Xu27b437c2006-07-13 19:26:39 -0700355 struct sk_buff *list = *listp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356
Herbert Xu27b437c2006-07-13 19:26:39 -0700357 *listp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358
359 do {
360 struct sk_buff *this = list;
361 list = list->next;
362 kfree_skb(this);
363 } while (list);
364}
365
Herbert Xu27b437c2006-07-13 19:26:39 -0700366static inline void skb_drop_fraglist(struct sk_buff *skb)
367{
368 skb_drop_list(&skb_shinfo(skb)->frag_list);
369}
370
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371static void skb_clone_fraglist(struct sk_buff *skb)
372{
373 struct sk_buff *list;
374
David S. Millerfbb398a2009-06-09 00:18:59 -0700375 skb_walk_frags(skb, list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 skb_get(list);
377}
378
Adrian Bunk5bba1712006-06-29 13:02:35 -0700379static void skb_release_data(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380{
381 if (!skb->cloned ||
382 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
383 &skb_shinfo(skb)->dataref)) {
384 if (skb_shinfo(skb)->nr_frags) {
385 int i;
386 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
Ian Campbellea2ab692011-08-22 23:44:58 +0000387 skb_frag_unref(skb, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 }
389
Shirley Maa6686f22011-07-06 12:22:12 +0000390 /*
391 * If skb buf is from userspace, we need to notify the caller
392 * the lower device DMA has done;
393 */
394 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
395 struct ubuf_info *uarg;
396
397 uarg = skb_shinfo(skb)->destructor_arg;
398 if (uarg->callback)
399 uarg->callback(uarg);
400 }
401
David S. Miller21dc3302010-08-23 00:13:46 -0700402 if (skb_has_frag_list(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 skb_drop_fraglist(skb);
404
405 kfree(skb->head);
406 }
407}
408
409/*
410 * Free an skbuff by memory without cleaning the state.
411 */
Herbert Xu2d4baff2007-11-26 23:11:19 +0800412static void kfree_skbmem(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413{
David S. Millerd179cd12005-08-17 14:57:30 -0700414 struct sk_buff *other;
415 atomic_t *fclone_ref;
416
David S. Millerd179cd12005-08-17 14:57:30 -0700417 switch (skb->fclone) {
418 case SKB_FCLONE_UNAVAILABLE:
419 kmem_cache_free(skbuff_head_cache, skb);
420 break;
421
422 case SKB_FCLONE_ORIG:
423 fclone_ref = (atomic_t *) (skb + 2);
424 if (atomic_dec_and_test(fclone_ref))
425 kmem_cache_free(skbuff_fclone_cache, skb);
426 break;
427
428 case SKB_FCLONE_CLONE:
429 fclone_ref = (atomic_t *) (skb + 1);
430 other = skb - 1;
431
432 /* The clone portion is available for
433 * fast-cloning again.
434 */
435 skb->fclone = SKB_FCLONE_UNAVAILABLE;
436
437 if (atomic_dec_and_test(fclone_ref))
438 kmem_cache_free(skbuff_fclone_cache, other);
439 break;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -0700440 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441}
442
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700443static void skb_release_head_state(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444{
Eric Dumazetadf30902009-06-02 05:19:30 +0000445 skb_dst_drop(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446#ifdef CONFIG_XFRM
447 secpath_put(skb->sp);
448#endif
Stephen Hemminger9c2b3322005-04-19 22:39:42 -0700449 if (skb->destructor) {
450 WARN_ON(in_irq());
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 skb->destructor(skb);
452 }
Igor Maravića3bf7ae2011-12-12 02:58:22 +0000453#if IS_ENABLED(CONFIG_NF_CONNTRACK)
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -0700454 nf_conntrack_put(skb->nfct);
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +0100455#endif
456#ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
Yasuyuki Kozakai9fb9cbb2005-11-09 16:38:16 -0800457 nf_conntrack_put_reasm(skb->nfct_reasm);
458#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459#ifdef CONFIG_BRIDGE_NETFILTER
460 nf_bridge_put(skb->nf_bridge);
461#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462/* XXX: IS this still necessary? - JHS */
463#ifdef CONFIG_NET_SCHED
464 skb->tc_index = 0;
465#ifdef CONFIG_NET_CLS_ACT
466 skb->tc_verd = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467#endif
468#endif
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700469}
470
471/* Free everything but the sk_buff shell. */
472static void skb_release_all(struct sk_buff *skb)
473{
474 skb_release_head_state(skb);
Herbert Xu2d4baff2007-11-26 23:11:19 +0800475 skb_release_data(skb);
476}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477
Herbert Xu2d4baff2007-11-26 23:11:19 +0800478/**
479 * __kfree_skb - private function
480 * @skb: buffer
481 *
482 * Free an sk_buff. Release anything attached to the buffer.
483 * Clean the state. This is an internal helper function. Users should
484 * always call kfree_skb
485 */
486
487void __kfree_skb(struct sk_buff *skb)
488{
489 skb_release_all(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 kfree_skbmem(skb);
491}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800492EXPORT_SYMBOL(__kfree_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493
494/**
Jörn Engel231d06a2006-03-20 21:28:35 -0800495 * kfree_skb - free an sk_buff
496 * @skb: buffer to free
497 *
498 * Drop a reference to the buffer and free it if the usage count has
499 * hit zero.
500 */
501void kfree_skb(struct sk_buff *skb)
502{
503 if (unlikely(!skb))
504 return;
505 if (likely(atomic_read(&skb->users) == 1))
506 smp_rmb();
507 else if (likely(!atomic_dec_and_test(&skb->users)))
508 return;
Neil Hormanead2ceb2009-03-11 09:49:55 +0000509 trace_kfree_skb(skb, __builtin_return_address(0));
Jörn Engel231d06a2006-03-20 21:28:35 -0800510 __kfree_skb(skb);
511}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800512EXPORT_SYMBOL(kfree_skb);
Jörn Engel231d06a2006-03-20 21:28:35 -0800513
Stephen Hemmingerd1a203e2008-11-01 21:01:09 -0700514/**
Neil Hormanead2ceb2009-03-11 09:49:55 +0000515 * consume_skb - free an skbuff
516 * @skb: buffer to free
517 *
518 * Drop a ref to the buffer and free it if the usage count has hit zero
519 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
520 * is being dropped after a failure and notes that
521 */
522void consume_skb(struct sk_buff *skb)
523{
524 if (unlikely(!skb))
525 return;
526 if (likely(atomic_read(&skb->users) == 1))
527 smp_rmb();
528 else if (likely(!atomic_dec_and_test(&skb->users)))
529 return;
Koki Sanagi07dc22e2010-08-23 18:46:12 +0900530 trace_consume_skb(skb);
Neil Hormanead2ceb2009-03-11 09:49:55 +0000531 __kfree_skb(skb);
532}
533EXPORT_SYMBOL(consume_skb);
534
535/**
Andy Fleming3d153a72011-10-13 04:33:54 +0000536 * skb_recycle - clean up an skb for reuse
537 * @skb: buffer
538 *
539 * Recycles the skb to be reused as a receive buffer. This
540 * function does any necessary reference count dropping, and
541 * cleans up the skbuff as if it just came from __alloc_skb().
542 */
543void skb_recycle(struct sk_buff *skb)
544{
545 struct skb_shared_info *shinfo;
546
547 skb_release_head_state(skb);
548
549 shinfo = skb_shinfo(skb);
550 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
551 atomic_set(&shinfo->dataref, 1);
552
553 memset(skb, 0, offsetof(struct sk_buff, tail));
554 skb->data = skb->head + NET_SKB_PAD;
555 skb_reset_tail_pointer(skb);
556}
557EXPORT_SYMBOL(skb_recycle);
558
559/**
Stephen Hemmingerd1a203e2008-11-01 21:01:09 -0700560 * skb_recycle_check - check if skb can be reused for receive
561 * @skb: buffer
562 * @skb_size: minimum receive buffer size
563 *
564 * Checks that the skb passed in is not shared or cloned, and
565 * that it is linear and its head portion at least as large as
566 * skb_size so that it can be recycled as a receive buffer.
567 * If these conditions are met, this function does any necessary
568 * reference count dropping and cleans up the skbuff as if it
569 * just came from __alloc_skb().
570 */
Changli Gao5b0daa32010-05-29 00:12:13 -0700571bool skb_recycle_check(struct sk_buff *skb, int skb_size)
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700572{
Andy Fleming3d153a72011-10-13 04:33:54 +0000573 if (!skb_is_recycleable(skb, skb_size))
Changli Gao5b0daa32010-05-29 00:12:13 -0700574 return false;
Anton Vorontsove84af6d2009-11-10 14:11:01 +0000575
Andy Fleming3d153a72011-10-13 04:33:54 +0000576 skb_recycle(skb);
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700577
Changli Gao5b0daa32010-05-29 00:12:13 -0700578 return true;
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700579}
580EXPORT_SYMBOL(skb_recycle_check);
581
Herbert Xudec18812007-10-14 00:37:30 -0700582static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
583{
584 new->tstamp = old->tstamp;
585 new->dev = old->dev;
586 new->transport_header = old->transport_header;
587 new->network_header = old->network_header;
588 new->mac_header = old->mac_header;
Eric Dumazet7fee2262010-05-11 23:19:48 +0000589 skb_dst_copy(new, old);
Tom Herbert0a9627f2010-03-16 08:03:29 +0000590 new->rxhash = old->rxhash;
Changli Gao6461be32011-08-19 04:44:18 +0000591 new->ooo_okay = old->ooo_okay;
Tom Herbertbdeab992011-08-14 19:45:55 +0000592 new->l4_rxhash = old->l4_rxhash;
Ben Greear3bdc0eb2012-02-11 15:39:30 +0000593 new->no_fcs = old->no_fcs;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -0700594#ifdef CONFIG_XFRM
Herbert Xudec18812007-10-14 00:37:30 -0700595 new->sp = secpath_get(old->sp);
596#endif
597 memcpy(new->cb, old->cb, sizeof(old->cb));
Herbert Xu9bcb97c2009-05-22 22:20:02 +0000598 new->csum = old->csum;
Herbert Xudec18812007-10-14 00:37:30 -0700599 new->local_df = old->local_df;
600 new->pkt_type = old->pkt_type;
601 new->ip_summed = old->ip_summed;
602 skb_copy_queue_mapping(new, old);
603 new->priority = old->priority;
Igor Maravića3bf7ae2011-12-12 02:58:22 +0000604#if IS_ENABLED(CONFIG_IP_VS)
Herbert Xudec18812007-10-14 00:37:30 -0700605 new->ipvs_property = old->ipvs_property;
606#endif
607 new->protocol = old->protocol;
608 new->mark = old->mark;
Eric Dumazet8964be42009-11-20 15:35:04 -0800609 new->skb_iif = old->skb_iif;
Herbert Xudec18812007-10-14 00:37:30 -0700610 __nf_copy(new, old);
Igor Maravića3bf7ae2011-12-12 02:58:22 +0000611#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
Herbert Xudec18812007-10-14 00:37:30 -0700612 new->nf_trace = old->nf_trace;
613#endif
614#ifdef CONFIG_NET_SCHED
615 new->tc_index = old->tc_index;
616#ifdef CONFIG_NET_CLS_ACT
617 new->tc_verd = old->tc_verd;
618#endif
619#endif
Patrick McHardy6aa895b2008-07-14 22:49:06 -0700620 new->vlan_tci = old->vlan_tci;
621
Herbert Xudec18812007-10-14 00:37:30 -0700622 skb_copy_secmark(new, old);
623}
624
Herbert Xu82c49a32009-05-22 22:11:37 +0000625/*
626 * You should not add any new code to this function. Add it to
627 * __copy_skb_header above instead.
628 */
Herbert Xue0053ec2007-10-14 00:37:52 -0700629static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631#define C(x) n->x = skb->x
632
633 n->next = n->prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 n->sk = NULL;
Herbert Xudec18812007-10-14 00:37:30 -0700635 __copy_skb_header(n, skb);
636
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 C(len);
638 C(data_len);
Alexey Dobriyan3e6b3b22007-03-16 15:00:46 -0700639 C(mac_len);
Patrick McHardy334a8132007-06-25 04:35:20 -0700640 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
Paul Moore02f1c892008-01-07 21:56:41 -0800641 n->cloned = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 n->nohdr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 n->destructor = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 C(tail);
645 C(end);
Paul Moore02f1c892008-01-07 21:56:41 -0800646 C(head);
647 C(data);
648 C(truesize);
649 atomic_set(&n->users, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650
651 atomic_inc(&(skb_shinfo(skb)->dataref));
652 skb->cloned = 1;
653
654 return n;
Herbert Xue0053ec2007-10-14 00:37:52 -0700655#undef C
656}
657
658/**
659 * skb_morph - morph one skb into another
660 * @dst: the skb to receive the contents
661 * @src: the skb to supply the contents
662 *
663 * This is identical to skb_clone except that the target skb is
664 * supplied by the user.
665 *
666 * The target skb is returned upon exit.
667 */
668struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
669{
Herbert Xu2d4baff2007-11-26 23:11:19 +0800670 skb_release_all(dst);
Herbert Xue0053ec2007-10-14 00:37:52 -0700671 return __skb_clone(dst, src);
672}
673EXPORT_SYMBOL_GPL(skb_morph);
674
Michael S. Tsirkin48c83012011-08-31 08:03:29 +0000675/* skb_copy_ubufs - copy userspace skb frags buffers to kernel
676 * @skb: the skb to modify
677 * @gfp_mask: allocation priority
678 *
679 * This must be called on SKBTX_DEV_ZEROCOPY skb.
680 * It will copy all frags into kernel and drop the reference
681 * to userspace pages.
682 *
683 * If this function is called from an interrupt gfp_mask() must be
684 * %GFP_ATOMIC.
685 *
686 * Returns 0 on success or a negative error code on failure
687 * to allocate kernel memory to copy to.
688 */
689int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
Shirley Maa6686f22011-07-06 12:22:12 +0000690{
691 int i;
692 int num_frags = skb_shinfo(skb)->nr_frags;
693 struct page *page, *head = NULL;
694 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
695
696 for (i = 0; i < num_frags; i++) {
697 u8 *vaddr;
698 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
699
700 page = alloc_page(GFP_ATOMIC);
701 if (!page) {
702 while (head) {
703 struct page *next = (struct page *)head->private;
704 put_page(head);
705 head = next;
706 }
707 return -ENOMEM;
708 }
Eric Dumazet51c56b02012-04-05 11:35:15 +0200709 vaddr = kmap_atomic(skb_frag_page(f));
Shirley Maa6686f22011-07-06 12:22:12 +0000710 memcpy(page_address(page),
Eric Dumazet9e903e02011-10-18 21:00:24 +0000711 vaddr + f->page_offset, skb_frag_size(f));
Eric Dumazet51c56b02012-04-05 11:35:15 +0200712 kunmap_atomic(vaddr);
Shirley Maa6686f22011-07-06 12:22:12 +0000713 page->private = (unsigned long)head;
714 head = page;
715 }
716
717 /* skb frags release userspace buffers */
718 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
Ian Campbella8605c62011-10-19 23:01:49 +0000719 skb_frag_unref(skb, i);
Shirley Maa6686f22011-07-06 12:22:12 +0000720
721 uarg->callback(uarg);
722
723 /* skb frags point to kernel buffers */
724 for (i = skb_shinfo(skb)->nr_frags; i > 0; i--) {
Ian Campbella8605c62011-10-19 23:01:49 +0000725 __skb_fill_page_desc(skb, i-1, head, 0,
726 skb_shinfo(skb)->frags[i - 1].size);
Shirley Maa6686f22011-07-06 12:22:12 +0000727 head = (struct page *)head->private;
728 }
Michael S. Tsirkin48c83012011-08-31 08:03:29 +0000729
730 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
Shirley Maa6686f22011-07-06 12:22:12 +0000731 return 0;
732}
733
734
Herbert Xue0053ec2007-10-14 00:37:52 -0700735/**
736 * skb_clone - duplicate an sk_buff
737 * @skb: buffer to clone
738 * @gfp_mask: allocation priority
739 *
740 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
741 * copies share the same packet data but not structure. The new
742 * buffer has a reference count of 1. If the allocation fails the
743 * function returns %NULL otherwise the new buffer is returned.
744 *
745 * If this function is called from an interrupt gfp_mask() must be
746 * %GFP_ATOMIC.
747 */
748
749struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
750{
751 struct sk_buff *n;
752
Shirley Maa6686f22011-07-06 12:22:12 +0000753 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
754 if (skb_copy_ubufs(skb, gfp_mask))
755 return NULL;
756 }
757
Herbert Xue0053ec2007-10-14 00:37:52 -0700758 n = skb + 1;
759 if (skb->fclone == SKB_FCLONE_ORIG &&
760 n->fclone == SKB_FCLONE_UNAVAILABLE) {
761 atomic_t *fclone_ref = (atomic_t *) (n + 1);
762 n->fclone = SKB_FCLONE_CLONE;
763 atomic_inc(fclone_ref);
764 } else {
765 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
766 if (!n)
767 return NULL;
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200768
769 kmemcheck_annotate_bitfield(n, flags1);
770 kmemcheck_annotate_bitfield(n, flags2);
Herbert Xue0053ec2007-10-14 00:37:52 -0700771 n->fclone = SKB_FCLONE_UNAVAILABLE;
772 }
773
774 return __skb_clone(n, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800776EXPORT_SYMBOL(skb_clone);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777
778static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
779{
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700780#ifndef NET_SKBUFF_DATA_USES_OFFSET
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 /*
782 * Shift between the two data areas in bytes
783 */
784 unsigned long offset = new->data - old->data;
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700785#endif
Herbert Xudec18812007-10-14 00:37:30 -0700786
787 __copy_skb_header(new, old);
788
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700789#ifndef NET_SKBUFF_DATA_USES_OFFSET
790 /* {transport,network,mac}_header are relative to skb->head */
791 new->transport_header += offset;
792 new->network_header += offset;
Stephen Hemminger603a8bb2009-06-17 12:17:34 +0000793 if (skb_mac_header_was_set(new))
794 new->mac_header += offset;
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700795#endif
Herbert Xu79671682006-06-22 02:40:14 -0700796 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
797 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
798 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799}
800
801/**
802 * skb_copy - create private copy of an sk_buff
803 * @skb: buffer to copy
804 * @gfp_mask: allocation priority
805 *
806 * Make a copy of both an &sk_buff and its data. This is used when the
807 * caller wishes to modify the data and needs a private copy of the
808 * data to alter. Returns %NULL on failure or the pointer to the buffer
809 * on success. The returned buffer has a reference count of 1.
810 *
811 * As by-product this function converts non-linear &sk_buff to linear
812 * one, so that &sk_buff becomes completely private and caller is allowed
813 * to modify all the data of returned buffer. This means that this
814 * function is not recommended for use in circumstances when only
815 * header is going to be modified. Use pskb_copy() instead.
816 */
817
Al Virodd0fc662005-10-07 07:46:04 +0100818struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819{
Eric Dumazet6602ceb2010-09-01 05:25:10 +0000820 int headerlen = skb_headroom(skb);
821 unsigned int size = (skb_end_pointer(skb) - skb->head) + skb->data_len;
822 struct sk_buff *n = alloc_skb(size, gfp_mask);
823
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 if (!n)
825 return NULL;
826
827 /* Set the data pointer */
828 skb_reserve(n, headerlen);
829 /* Set the tail pointer and length */
830 skb_put(n, skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831
832 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
833 BUG();
834
835 copy_skb_header(n, skb);
836 return n;
837}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800838EXPORT_SYMBOL(skb_copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839
840/**
Eric Dumazet117632e2011-12-03 21:39:53 +0000841 * __pskb_copy - create copy of an sk_buff with private head.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 * @skb: buffer to copy
Eric Dumazet117632e2011-12-03 21:39:53 +0000843 * @headroom: headroom of new skb
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 * @gfp_mask: allocation priority
845 *
846 * Make a copy of both an &sk_buff and part of its data, located
847 * in header. Fragmented data remain shared. This is used when
848 * the caller wishes to modify only header of &sk_buff and needs
849 * private copy of the header to alter. Returns %NULL on failure
850 * or the pointer to the buffer on success.
851 * The returned buffer has a reference count of 1.
852 */
853
Eric Dumazet117632e2011-12-03 21:39:53 +0000854struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855{
Eric Dumazet117632e2011-12-03 21:39:53 +0000856 unsigned int size = skb_headlen(skb) + headroom;
Eric Dumazet6602ceb2010-09-01 05:25:10 +0000857 struct sk_buff *n = alloc_skb(size, gfp_mask);
858
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 if (!n)
860 goto out;
861
862 /* Set the data pointer */
Eric Dumazet117632e2011-12-03 21:39:53 +0000863 skb_reserve(n, headroom);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 /* Set the tail pointer and length */
865 skb_put(n, skb_headlen(skb));
866 /* Copy the bytes */
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -0300867 skb_copy_from_linear_data(skb, n->data, n->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868
Herbert Xu25f484a2006-11-07 14:57:15 -0800869 n->truesize += skb->data_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 n->data_len = skb->data_len;
871 n->len = skb->len;
872
873 if (skb_shinfo(skb)->nr_frags) {
874 int i;
875
Shirley Maa6686f22011-07-06 12:22:12 +0000876 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
877 if (skb_copy_ubufs(skb, gfp_mask)) {
Dan Carpenter15110222011-07-19 22:51:49 +0000878 kfree_skb(n);
879 n = NULL;
Shirley Maa6686f22011-07-06 12:22:12 +0000880 goto out;
881 }
882 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
884 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
Ian Campbellea2ab692011-08-22 23:44:58 +0000885 skb_frag_ref(skb, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 }
887 skb_shinfo(n)->nr_frags = i;
888 }
889
David S. Miller21dc3302010-08-23 00:13:46 -0700890 if (skb_has_frag_list(skb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
892 skb_clone_fraglist(n);
893 }
894
895 copy_skb_header(n, skb);
896out:
897 return n;
898}
Eric Dumazet117632e2011-12-03 21:39:53 +0000899EXPORT_SYMBOL(__pskb_copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900
901/**
902 * pskb_expand_head - reallocate header of &sk_buff
903 * @skb: buffer to reallocate
904 * @nhead: room to add at head
905 * @ntail: room to add at tail
906 * @gfp_mask: allocation priority
907 *
908 * Expands (or creates identical copy, if &nhead and &ntail are zero)
909 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
910 * reference count of 1. Returns zero in the case of success or error,
911 * if expansion failed. In the last case, &sk_buff is not changed.
912 *
913 * All the pointers pointing into skb header may change and must be
914 * reloaded after call to this function.
915 */
916
Victor Fusco86a76ca2005-07-08 14:57:47 -0700917int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
Al Virodd0fc662005-10-07 07:46:04 +0100918 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919{
920 int i;
921 u8 *data;
Eric Dumazet6602ceb2010-09-01 05:25:10 +0000922 int size = nhead + (skb_end_pointer(skb) - skb->head) + ntail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 long off;
Eric Dumazet1fd63042010-09-02 23:09:32 +0000924 bool fastpath;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925
Herbert Xu4edd87a2008-10-01 07:09:38 -0700926 BUG_ON(nhead < 0);
927
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 if (skb_shared(skb))
929 BUG();
930
931 size = SKB_DATA_ALIGN(size);
932
Changli Gaoca44ac32010-11-29 22:48:46 +0000933 /* Check if we can avoid taking references on fragments if we own
934 * the last reference on skb->head. (see skb_release_data())
935 */
936 if (!skb->cloned)
937 fastpath = true;
938 else {
939 int delta = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1;
Changli Gaoca44ac32010-11-29 22:48:46 +0000940 fastpath = atomic_read(&skb_shinfo(skb)->dataref) == delta;
941 }
942
943 if (fastpath &&
944 size + sizeof(struct skb_shared_info) <= ksize(skb->head)) {
945 memmove(skb->head + size, skb_shinfo(skb),
946 offsetof(struct skb_shared_info,
947 frags[skb_shinfo(skb)->nr_frags]));
948 memmove(skb->head + nhead, skb->head,
949 skb_tail_pointer(skb) - skb->head);
950 off = nhead;
951 goto adjust_others;
952 }
953
Eric Dumazet87151b82012-04-10 20:08:39 +0000954 data = kmalloc(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
955 gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 if (!data)
957 goto nodata;
Eric Dumazet87151b82012-04-10 20:08:39 +0000958 size = SKB_WITH_OVERHEAD(ksize(data));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959
960 /* Copy only real data... and, alas, header. This should be
Eric Dumazet6602ceb2010-09-01 05:25:10 +0000961 * optimized for the cases when header is void.
962 */
963 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
964
965 memcpy((struct skb_shared_info *)(data + size),
966 skb_shinfo(skb),
Eric Dumazetfed66382010-07-22 19:09:08 +0000967 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968
Eric Dumazet1fd63042010-09-02 23:09:32 +0000969 if (fastpath) {
970 kfree(skb->head);
971 } else {
Shirley Maa6686f22011-07-06 12:22:12 +0000972 /* copy this zero copy skb frags */
973 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
974 if (skb_copy_ubufs(skb, gfp_mask))
975 goto nofrags;
976 }
Eric Dumazet1fd63042010-09-02 23:09:32 +0000977 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
Ian Campbellea2ab692011-08-22 23:44:58 +0000978 skb_frag_ref(skb, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979
Eric Dumazet1fd63042010-09-02 23:09:32 +0000980 if (skb_has_frag_list(skb))
981 skb_clone_fraglist(skb);
982
983 skb_release_data(skb);
984 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 off = (data + nhead) - skb->head;
986
987 skb->head = data;
Changli Gaoca44ac32010-11-29 22:48:46 +0000988adjust_others:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 skb->data += off;
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700990#ifdef NET_SKBUFF_DATA_USES_OFFSET
991 skb->end = size;
Patrick McHardy56eb8882007-04-09 11:45:04 -0700992 off = nhead;
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700993#else
994 skb->end = skb->head + size;
Patrick McHardy56eb8882007-04-09 11:45:04 -0700995#endif
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700996 /* {transport,network,mac}_header and tail are relative to skb->head */
997 skb->tail += off;
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700998 skb->transport_header += off;
999 skb->network_header += off;
Stephen Hemminger603a8bb2009-06-17 12:17:34 +00001000 if (skb_mac_header_was_set(skb))
1001 skb->mac_header += off;
Andrea Shepard00c5a982010-07-22 09:12:35 +00001002 /* Only adjust this if it actually is csum_start rather than csum */
1003 if (skb->ip_summed == CHECKSUM_PARTIAL)
1004 skb->csum_start += nhead;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 skb->cloned = 0;
Patrick McHardy334a8132007-06-25 04:35:20 -07001006 skb->hdr_len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 skb->nohdr = 0;
1008 atomic_set(&skb_shinfo(skb)->dataref, 1);
1009 return 0;
1010
Shirley Maa6686f22011-07-06 12:22:12 +00001011nofrags:
1012 kfree(data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013nodata:
1014 return -ENOMEM;
1015}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001016EXPORT_SYMBOL(pskb_expand_head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017
1018/* Make private copy of skb with writable head and some headroom */
1019
1020struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1021{
1022 struct sk_buff *skb2;
1023 int delta = headroom - skb_headroom(skb);
1024
1025 if (delta <= 0)
1026 skb2 = pskb_copy(skb, GFP_ATOMIC);
1027 else {
1028 skb2 = skb_clone(skb, GFP_ATOMIC);
1029 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1030 GFP_ATOMIC)) {
1031 kfree_skb(skb2);
1032 skb2 = NULL;
1033 }
1034 }
1035 return skb2;
1036}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001037EXPORT_SYMBOL(skb_realloc_headroom);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038
1039/**
1040 * skb_copy_expand - copy and expand sk_buff
1041 * @skb: buffer to copy
1042 * @newheadroom: new free bytes at head
1043 * @newtailroom: new free bytes at tail
1044 * @gfp_mask: allocation priority
1045 *
1046 * Make a copy of both an &sk_buff and its data and while doing so
1047 * allocate additional space.
1048 *
1049 * This is used when the caller wishes to modify the data and needs a
1050 * private copy of the data to alter as well as more space for new fields.
1051 * Returns %NULL on failure or the pointer to the buffer
1052 * on success. The returned buffer has a reference count of 1.
1053 *
1054 * You must pass %GFP_ATOMIC as the allocation priority if this function
1055 * is called from an interrupt.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 */
1057struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
Victor Fusco86a76ca2005-07-08 14:57:47 -07001058 int newheadroom, int newtailroom,
Al Virodd0fc662005-10-07 07:46:04 +01001059 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060{
1061 /*
1062 * Allocate the copy buffer
1063 */
1064 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
1065 gfp_mask);
Patrick McHardyefd1e8d2007-04-10 18:30:09 -07001066 int oldheadroom = skb_headroom(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 int head_copy_len, head_copy_off;
Herbert Xu52886052007-09-16 16:32:11 -07001068 int off;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069
1070 if (!n)
1071 return NULL;
1072
1073 skb_reserve(n, newheadroom);
1074
1075 /* Set the tail pointer and length */
1076 skb_put(n, skb->len);
1077
Patrick McHardyefd1e8d2007-04-10 18:30:09 -07001078 head_copy_len = oldheadroom;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 head_copy_off = 0;
1080 if (newheadroom <= head_copy_len)
1081 head_copy_len = newheadroom;
1082 else
1083 head_copy_off = newheadroom - head_copy_len;
1084
1085 /* Copy the linear header and data. */
1086 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1087 skb->len + head_copy_len))
1088 BUG();
1089
1090 copy_skb_header(n, skb);
1091
Patrick McHardyefd1e8d2007-04-10 18:30:09 -07001092 off = newheadroom - oldheadroom;
David S. Millerbe2b6e62010-07-22 13:27:09 -07001093 if (n->ip_summed == CHECKSUM_PARTIAL)
1094 n->csum_start += off;
Herbert Xu52886052007-09-16 16:32:11 -07001095#ifdef NET_SKBUFF_DATA_USES_OFFSET
Patrick McHardyefd1e8d2007-04-10 18:30:09 -07001096 n->transport_header += off;
1097 n->network_header += off;
Stephen Hemminger603a8bb2009-06-17 12:17:34 +00001098 if (skb_mac_header_was_set(skb))
1099 n->mac_header += off;
Herbert Xu52886052007-09-16 16:32:11 -07001100#endif
Patrick McHardyefd1e8d2007-04-10 18:30:09 -07001101
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 return n;
1103}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001104EXPORT_SYMBOL(skb_copy_expand);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105
1106/**
1107 * skb_pad - zero pad the tail of an skb
1108 * @skb: buffer to pad
1109 * @pad: space to pad
1110 *
1111 * Ensure that a buffer is followed by a padding area that is zero
1112 * filled. Used by network drivers which may DMA or transfer data
1113 * beyond the buffer end onto the wire.
1114 *
Herbert Xu5b057c62006-06-23 02:06:41 -07001115 * May return error in out of memory cases. The skb is freed on error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 */
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001117
Herbert Xu5b057c62006-06-23 02:06:41 -07001118int skb_pad(struct sk_buff *skb, int pad)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119{
Herbert Xu5b057c62006-06-23 02:06:41 -07001120 int err;
1121 int ntail;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001122
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 /* If the skbuff is non linear tailroom is always zero.. */
Herbert Xu5b057c62006-06-23 02:06:41 -07001124 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 memset(skb->data+skb->len, 0, pad);
Herbert Xu5b057c62006-06-23 02:06:41 -07001126 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 }
Herbert Xu5b057c62006-06-23 02:06:41 -07001128
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001129 ntail = skb->data_len + pad - (skb->end - skb->tail);
Herbert Xu5b057c62006-06-23 02:06:41 -07001130 if (likely(skb_cloned(skb) || ntail > 0)) {
1131 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1132 if (unlikely(err))
1133 goto free_skb;
1134 }
1135
1136 /* FIXME: The use of this function with non-linear skb's really needs
1137 * to be audited.
1138 */
1139 err = skb_linearize(skb);
1140 if (unlikely(err))
1141 goto free_skb;
1142
1143 memset(skb->data + skb->len, 0, pad);
1144 return 0;
1145
1146free_skb:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 kfree_skb(skb);
Herbert Xu5b057c62006-06-23 02:06:41 -07001148 return err;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001149}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001150EXPORT_SYMBOL(skb_pad);
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001151
Ilpo Järvinen0dde3e12008-03-27 17:43:41 -07001152/**
1153 * skb_put - add data to a buffer
1154 * @skb: buffer to use
1155 * @len: amount of data to add
1156 *
1157 * This function extends the used data area of the buffer. If this would
1158 * exceed the total buffer size the kernel will panic. A pointer to the
1159 * first byte of the extra data is returned.
1160 */
1161unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1162{
1163 unsigned char *tmp = skb_tail_pointer(skb);
1164 SKB_LINEAR_ASSERT(skb);
1165 skb->tail += len;
1166 skb->len += len;
1167 if (unlikely(skb->tail > skb->end))
1168 skb_over_panic(skb, len, __builtin_return_address(0));
1169 return tmp;
1170}
1171EXPORT_SYMBOL(skb_put);
1172
Ilpo Järvinen6be8ac22008-03-27 17:47:24 -07001173/**
Ilpo Järvinenc2aa2702008-03-27 17:52:40 -07001174 * skb_push - add data to the start of a buffer
1175 * @skb: buffer to use
1176 * @len: amount of data to add
1177 *
1178 * This function extends the used data area of the buffer at the buffer
1179 * start. If this would exceed the total buffer headroom the kernel will
1180 * panic. A pointer to the first byte of the extra data is returned.
1181 */
1182unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1183{
1184 skb->data -= len;
1185 skb->len += len;
1186 if (unlikely(skb->data<skb->head))
1187 skb_under_panic(skb, len, __builtin_return_address(0));
1188 return skb->data;
1189}
1190EXPORT_SYMBOL(skb_push);
1191
1192/**
Ilpo Järvinen6be8ac22008-03-27 17:47:24 -07001193 * skb_pull - remove data from the start of a buffer
1194 * @skb: buffer to use
1195 * @len: amount of data to remove
1196 *
1197 * This function removes data from the start of a buffer, returning
1198 * the memory to the headroom. A pointer to the next data in the buffer
1199 * is returned. Once the data has been pulled future pushes will overwrite
1200 * the old data.
1201 */
1202unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1203{
David S. Miller47d29642010-05-02 02:21:44 -07001204 return skb_pull_inline(skb, len);
Ilpo Järvinen6be8ac22008-03-27 17:47:24 -07001205}
1206EXPORT_SYMBOL(skb_pull);
1207
Ilpo Järvinen419ae742008-03-27 17:54:01 -07001208/**
1209 * skb_trim - remove end from a buffer
1210 * @skb: buffer to alter
1211 * @len: new length
1212 *
1213 * Cut the length of a buffer down by removing data from the tail. If
1214 * the buffer is already under the length specified it is not modified.
1215 * The skb must be linear.
1216 */
1217void skb_trim(struct sk_buff *skb, unsigned int len)
1218{
1219 if (skb->len > len)
1220 __skb_trim(skb, len);
1221}
1222EXPORT_SYMBOL(skb_trim);
1223
Herbert Xu3cc0e872006-06-09 16:13:38 -07001224/* Trims skb to length len. It can change skb pointers.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 */
1226
Herbert Xu3cc0e872006-06-09 16:13:38 -07001227int ___pskb_trim(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228{
Herbert Xu27b437c2006-07-13 19:26:39 -07001229 struct sk_buff **fragp;
1230 struct sk_buff *frag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 int offset = skb_headlen(skb);
1232 int nfrags = skb_shinfo(skb)->nr_frags;
1233 int i;
Herbert Xu27b437c2006-07-13 19:26:39 -07001234 int err;
1235
1236 if (skb_cloned(skb) &&
1237 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1238 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001240 i = 0;
1241 if (offset >= len)
1242 goto drop_pages;
1243
1244 for (; i < nfrags; i++) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00001245 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
Herbert Xu27b437c2006-07-13 19:26:39 -07001246
1247 if (end < len) {
1248 offset = end;
1249 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 }
Herbert Xu27b437c2006-07-13 19:26:39 -07001251
Eric Dumazet9e903e02011-10-18 21:00:24 +00001252 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
Herbert Xu27b437c2006-07-13 19:26:39 -07001253
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001254drop_pages:
Herbert Xu27b437c2006-07-13 19:26:39 -07001255 skb_shinfo(skb)->nr_frags = i;
1256
1257 for (; i < nfrags; i++)
Ian Campbellea2ab692011-08-22 23:44:58 +00001258 skb_frag_unref(skb, i);
Herbert Xu27b437c2006-07-13 19:26:39 -07001259
David S. Miller21dc3302010-08-23 00:13:46 -07001260 if (skb_has_frag_list(skb))
Herbert Xu27b437c2006-07-13 19:26:39 -07001261 skb_drop_fraglist(skb);
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001262 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 }
1264
Herbert Xu27b437c2006-07-13 19:26:39 -07001265 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1266 fragp = &frag->next) {
1267 int end = offset + frag->len;
1268
1269 if (skb_shared(frag)) {
1270 struct sk_buff *nfrag;
1271
1272 nfrag = skb_clone(frag, GFP_ATOMIC);
1273 if (unlikely(!nfrag))
1274 return -ENOMEM;
1275
1276 nfrag->next = frag->next;
Eric Dumazet85bb2a62012-04-19 02:24:53 +00001277 consume_skb(frag);
Herbert Xu27b437c2006-07-13 19:26:39 -07001278 frag = nfrag;
1279 *fragp = frag;
1280 }
1281
1282 if (end < len) {
1283 offset = end;
1284 continue;
1285 }
1286
1287 if (end > len &&
1288 unlikely((err = pskb_trim(frag, len - offset))))
1289 return err;
1290
1291 if (frag->next)
1292 skb_drop_list(&frag->next);
1293 break;
1294 }
1295
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001296done:
Herbert Xu27b437c2006-07-13 19:26:39 -07001297 if (len > skb_headlen(skb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 skb->data_len -= skb->len - len;
1299 skb->len = len;
1300 } else {
Herbert Xu27b437c2006-07-13 19:26:39 -07001301 skb->len = len;
1302 skb->data_len = 0;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001303 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 }
1305
1306 return 0;
1307}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001308EXPORT_SYMBOL(___pskb_trim);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309
1310/**
1311 * __pskb_pull_tail - advance tail of skb header
1312 * @skb: buffer to reallocate
1313 * @delta: number of bytes to advance tail
1314 *
1315 * The function makes a sense only on a fragmented &sk_buff,
1316 * it expands header moving its tail forward and copying necessary
1317 * data from fragmented part.
1318 *
1319 * &sk_buff MUST have reference count of 1.
1320 *
1321 * Returns %NULL (and &sk_buff does not change) if pull failed
1322 * or value of new tail of skb in the case of success.
1323 *
1324 * All the pointers pointing into skb header may change and must be
1325 * reloaded after call to this function.
1326 */
1327
1328/* Moves tail of skb head forward, copying data from fragmented part,
1329 * when it is necessary.
1330 * 1. It may fail due to malloc failure.
1331 * 2. It may change skb pointers.
1332 *
1333 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1334 */
1335unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1336{
1337 /* If skb has not enough free space at tail, get new one
1338 * plus 128 bytes for future expansions. If we have enough
1339 * room at tail, reallocate without expansion only if skb is cloned.
1340 */
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001341 int i, k, eat = (skb->tail + delta) - skb->end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342
1343 if (eat > 0 || skb_cloned(skb)) {
1344 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1345 GFP_ATOMIC))
1346 return NULL;
1347 }
1348
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001349 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 BUG();
1351
1352 /* Optimization: no fragments, no reasons to preestimate
1353 * size of pulled pages. Superb.
1354 */
David S. Miller21dc3302010-08-23 00:13:46 -07001355 if (!skb_has_frag_list(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 goto pull_pages;
1357
1358 /* Estimate size of pulled pages. */
1359 eat = delta;
1360 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00001361 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1362
1363 if (size >= eat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 goto pull_pages;
Eric Dumazet9e903e02011-10-18 21:00:24 +00001365 eat -= size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 }
1367
1368 /* If we need update frag list, we are in troubles.
1369 * Certainly, it possible to add an offset to skb data,
1370 * but taking into account that pulling is expected to
1371 * be very rare operation, it is worth to fight against
1372 * further bloating skb head and crucify ourselves here instead.
1373 * Pure masohism, indeed. 8)8)
1374 */
1375 if (eat) {
1376 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1377 struct sk_buff *clone = NULL;
1378 struct sk_buff *insp = NULL;
1379
1380 do {
Kris Katterjohn09a62662006-01-08 22:24:28 -08001381 BUG_ON(!list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382
1383 if (list->len <= eat) {
1384 /* Eaten as whole. */
1385 eat -= list->len;
1386 list = list->next;
1387 insp = list;
1388 } else {
1389 /* Eaten partially. */
1390
1391 if (skb_shared(list)) {
1392 /* Sucks! We need to fork list. :-( */
1393 clone = skb_clone(list, GFP_ATOMIC);
1394 if (!clone)
1395 return NULL;
1396 insp = list->next;
1397 list = clone;
1398 } else {
1399 /* This may be pulled without
1400 * problems. */
1401 insp = list;
1402 }
1403 if (!pskb_pull(list, eat)) {
Wei Yongjunf3fbbe02009-02-25 00:37:32 +00001404 kfree_skb(clone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 return NULL;
1406 }
1407 break;
1408 }
1409 } while (eat);
1410
1411 /* Free pulled out fragments. */
1412 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1413 skb_shinfo(skb)->frag_list = list->next;
1414 kfree_skb(list);
1415 }
1416 /* And insert new clone at head. */
1417 if (clone) {
1418 clone->next = list;
1419 skb_shinfo(skb)->frag_list = clone;
1420 }
1421 }
1422 /* Success! Now we may commit changes to skb data. */
1423
1424pull_pages:
1425 eat = delta;
1426 k = 0;
1427 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00001428 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1429
1430 if (size <= eat) {
Ian Campbellea2ab692011-08-22 23:44:58 +00001431 skb_frag_unref(skb, i);
Eric Dumazet9e903e02011-10-18 21:00:24 +00001432 eat -= size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 } else {
1434 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1435 if (eat) {
1436 skb_shinfo(skb)->frags[k].page_offset += eat;
Eric Dumazet9e903e02011-10-18 21:00:24 +00001437 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 eat = 0;
1439 }
1440 k++;
1441 }
1442 }
1443 skb_shinfo(skb)->nr_frags = k;
1444
1445 skb->tail += delta;
1446 skb->data_len -= delta;
1447
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001448 return skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001450EXPORT_SYMBOL(__pskb_pull_tail);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451
Eric Dumazet22019b12011-07-29 18:37:31 +00001452/**
1453 * skb_copy_bits - copy bits from skb to kernel buffer
1454 * @skb: source skb
1455 * @offset: offset in source
1456 * @to: destination buffer
1457 * @len: number of bytes to copy
1458 *
1459 * Copy the specified number of bytes from the source skb to the
1460 * destination buffer.
1461 *
1462 * CAUTION ! :
1463 * If its prototype is ever changed,
1464 * check arch/{*}/net/{*}.S files,
1465 * since it is called from BPF assembly code.
1466 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1468{
David S. Miller1a028e52007-04-27 15:21:23 -07001469 int start = skb_headlen(skb);
David S. Millerfbb398a2009-06-09 00:18:59 -07001470 struct sk_buff *frag_iter;
1471 int i, copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472
1473 if (offset > (int)skb->len - len)
1474 goto fault;
1475
1476 /* Copy header. */
David S. Miller1a028e52007-04-27 15:21:23 -07001477 if ((copy = start - offset) > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 if (copy > len)
1479 copy = len;
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03001480 skb_copy_from_linear_data_offset(skb, offset, to, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 if ((len -= copy) == 0)
1482 return 0;
1483 offset += copy;
1484 to += copy;
1485 }
1486
1487 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07001488 int end;
Eric Dumazet51c56b02012-04-05 11:35:15 +02001489 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001491 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07001492
Eric Dumazet51c56b02012-04-05 11:35:15 +02001493 end = start + skb_frag_size(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 if ((copy = end - offset) > 0) {
1495 u8 *vaddr;
1496
1497 if (copy > len)
1498 copy = len;
1499
Eric Dumazet51c56b02012-04-05 11:35:15 +02001500 vaddr = kmap_atomic(skb_frag_page(f));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 memcpy(to,
Eric Dumazet51c56b02012-04-05 11:35:15 +02001502 vaddr + f->page_offset + offset - start,
1503 copy);
1504 kunmap_atomic(vaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505
1506 if ((len -= copy) == 0)
1507 return 0;
1508 offset += copy;
1509 to += copy;
1510 }
David S. Miller1a028e52007-04-27 15:21:23 -07001511 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 }
1513
David S. Millerfbb398a2009-06-09 00:18:59 -07001514 skb_walk_frags(skb, frag_iter) {
1515 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516
David S. Millerfbb398a2009-06-09 00:18:59 -07001517 WARN_ON(start > offset + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518
David S. Millerfbb398a2009-06-09 00:18:59 -07001519 end = start + frag_iter->len;
1520 if ((copy = end - offset) > 0) {
1521 if (copy > len)
1522 copy = len;
1523 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1524 goto fault;
1525 if ((len -= copy) == 0)
1526 return 0;
1527 offset += copy;
1528 to += copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 }
David S. Millerfbb398a2009-06-09 00:18:59 -07001530 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 }
Shirley Maa6686f22011-07-06 12:22:12 +00001532
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 if (!len)
1534 return 0;
1535
1536fault:
1537 return -EFAULT;
1538}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001539EXPORT_SYMBOL(skb_copy_bits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540
Jens Axboe9c55e012007-11-06 23:30:13 -08001541/*
1542 * Callback from splice_to_pipe(), if we need to release some pages
1543 * at the end of the spd in case we error'ed out in filling the pipe.
1544 */
1545static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1546{
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001547 put_page(spd->pages[i]);
1548}
Jens Axboe9c55e012007-11-06 23:30:13 -08001549
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001550static inline struct page *linear_to_page(struct page *page, unsigned int *len,
1551 unsigned int *offset,
Jarek Poplawski7a67e562009-04-30 05:41:19 -07001552 struct sk_buff *skb, struct sock *sk)
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001553{
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001554 struct page *p = sk->sk_sndmsg_page;
1555 unsigned int off;
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001556
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001557 if (!p) {
1558new_page:
1559 p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1560 if (!p)
1561 return NULL;
1562
1563 off = sk->sk_sndmsg_off = 0;
1564 /* hold one ref to this page until it's full */
1565 } else {
1566 unsigned int mlen;
1567
1568 off = sk->sk_sndmsg_off;
1569 mlen = PAGE_SIZE - off;
1570 if (mlen < 64 && mlen < *len) {
1571 put_page(p);
1572 goto new_page;
1573 }
1574
1575 *len = min_t(unsigned int, *len, mlen);
1576 }
1577
1578 memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1579 sk->sk_sndmsg_off += *len;
1580 *offset = off;
1581 get_page(p);
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001582
1583 return p;
Jens Axboe9c55e012007-11-06 23:30:13 -08001584}
1585
1586/*
1587 * Fill page/offset/length into spd, if it can hold more pages.
1588 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001589static inline int spd_fill_page(struct splice_pipe_desc *spd,
1590 struct pipe_inode_info *pipe, struct page *page,
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001591 unsigned int *len, unsigned int offset,
Jarek Poplawski7a67e562009-04-30 05:41:19 -07001592 struct sk_buff *skb, int linear,
1593 struct sock *sk)
Jens Axboe9c55e012007-11-06 23:30:13 -08001594{
Jens Axboe35f3d142010-05-20 10:43:18 +02001595 if (unlikely(spd->nr_pages == pipe->buffers))
Jens Axboe9c55e012007-11-06 23:30:13 -08001596 return 1;
1597
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001598 if (linear) {
Jarek Poplawski7a67e562009-04-30 05:41:19 -07001599 page = linear_to_page(page, len, &offset, skb, sk);
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001600 if (!page)
1601 return 1;
1602 } else
1603 get_page(page);
1604
Jens Axboe9c55e012007-11-06 23:30:13 -08001605 spd->pages[spd->nr_pages] = page;
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001606 spd->partial[spd->nr_pages].len = *len;
Jens Axboe9c55e012007-11-06 23:30:13 -08001607 spd->partial[spd->nr_pages].offset = offset;
Jens Axboe9c55e012007-11-06 23:30:13 -08001608 spd->nr_pages++;
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001609
Jens Axboe9c55e012007-11-06 23:30:13 -08001610 return 0;
1611}
1612
Octavian Purdila2870c432008-07-15 00:49:11 -07001613static inline void __segment_seek(struct page **page, unsigned int *poff,
1614 unsigned int *plen, unsigned int off)
Jens Axboe9c55e012007-11-06 23:30:13 -08001615{
Jarek Poplawskice3dd392009-02-12 16:51:43 -08001616 unsigned long n;
1617
Octavian Purdila2870c432008-07-15 00:49:11 -07001618 *poff += off;
Jarek Poplawskice3dd392009-02-12 16:51:43 -08001619 n = *poff / PAGE_SIZE;
1620 if (n)
1621 *page = nth_page(*page, n);
1622
Octavian Purdila2870c432008-07-15 00:49:11 -07001623 *poff = *poff % PAGE_SIZE;
1624 *plen -= off;
1625}
Jens Axboe9c55e012007-11-06 23:30:13 -08001626
Octavian Purdila2870c432008-07-15 00:49:11 -07001627static inline int __splice_segment(struct page *page, unsigned int poff,
1628 unsigned int plen, unsigned int *off,
1629 unsigned int *len, struct sk_buff *skb,
Jarek Poplawski7a67e562009-04-30 05:41:19 -07001630 struct splice_pipe_desc *spd, int linear,
Jens Axboe35f3d142010-05-20 10:43:18 +02001631 struct sock *sk,
1632 struct pipe_inode_info *pipe)
Octavian Purdila2870c432008-07-15 00:49:11 -07001633{
1634 if (!*len)
1635 return 1;
1636
1637 /* skip this segment if already processed */
1638 if (*off >= plen) {
1639 *off -= plen;
1640 return 0;
Octavian Purdiladb43a282008-06-27 17:27:21 -07001641 }
Jens Axboe9c55e012007-11-06 23:30:13 -08001642
Octavian Purdila2870c432008-07-15 00:49:11 -07001643 /* ignore any bits we already processed */
1644 if (*off) {
1645 __segment_seek(&page, &poff, &plen, *off);
1646 *off = 0;
1647 }
1648
1649 do {
1650 unsigned int flen = min(*len, plen);
1651
1652 /* the linear region may spread across several pages */
1653 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1654
Jens Axboe35f3d142010-05-20 10:43:18 +02001655 if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk))
Octavian Purdila2870c432008-07-15 00:49:11 -07001656 return 1;
1657
1658 __segment_seek(&page, &poff, &plen, flen);
1659 *len -= flen;
1660
1661 } while (*len && plen);
1662
1663 return 0;
1664}
1665
1666/*
1667 * Map linear and fragment data from the skb to spd. It reports failure if the
1668 * pipe is full or if we already spliced the requested length.
1669 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001670static int __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1671 unsigned int *offset, unsigned int *len,
1672 struct splice_pipe_desc *spd, struct sock *sk)
Octavian Purdila2870c432008-07-15 00:49:11 -07001673{
1674 int seg;
1675
Jens Axboe9c55e012007-11-06 23:30:13 -08001676 /*
Octavian Purdila2870c432008-07-15 00:49:11 -07001677 * map the linear part
Jens Axboe9c55e012007-11-06 23:30:13 -08001678 */
Octavian Purdila2870c432008-07-15 00:49:11 -07001679 if (__splice_segment(virt_to_page(skb->data),
1680 (unsigned long) skb->data & (PAGE_SIZE - 1),
1681 skb_headlen(skb),
Jens Axboe35f3d142010-05-20 10:43:18 +02001682 offset, len, skb, spd, 1, sk, pipe))
Octavian Purdila2870c432008-07-15 00:49:11 -07001683 return 1;
Jens Axboe9c55e012007-11-06 23:30:13 -08001684
1685 /*
1686 * then map the fragments
1687 */
Jens Axboe9c55e012007-11-06 23:30:13 -08001688 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1689 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1690
Ian Campbellea2ab692011-08-22 23:44:58 +00001691 if (__splice_segment(skb_frag_page(f),
Eric Dumazet9e903e02011-10-18 21:00:24 +00001692 f->page_offset, skb_frag_size(f),
Jens Axboe35f3d142010-05-20 10:43:18 +02001693 offset, len, skb, spd, 0, sk, pipe))
Octavian Purdila2870c432008-07-15 00:49:11 -07001694 return 1;
Jens Axboe9c55e012007-11-06 23:30:13 -08001695 }
1696
Octavian Purdila2870c432008-07-15 00:49:11 -07001697 return 0;
Jens Axboe9c55e012007-11-06 23:30:13 -08001698}
1699
1700/*
1701 * Map data from the skb to a pipe. Should handle both the linear part,
1702 * the fragments, and the frag list. It does NOT handle frag lists within
1703 * the frag list, if such a thing exists. We'd probably need to recurse to
1704 * handle that cleanly.
1705 */
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001706int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
Jens Axboe9c55e012007-11-06 23:30:13 -08001707 struct pipe_inode_info *pipe, unsigned int tlen,
1708 unsigned int flags)
1709{
Jens Axboe35f3d142010-05-20 10:43:18 +02001710 struct partial_page partial[PIPE_DEF_BUFFERS];
1711 struct page *pages[PIPE_DEF_BUFFERS];
Jens Axboe9c55e012007-11-06 23:30:13 -08001712 struct splice_pipe_desc spd = {
1713 .pages = pages,
1714 .partial = partial,
1715 .flags = flags,
1716 .ops = &sock_pipe_buf_ops,
1717 .spd_release = sock_spd_release,
1718 };
David S. Millerfbb398a2009-06-09 00:18:59 -07001719 struct sk_buff *frag_iter;
Jarek Poplawski7a67e562009-04-30 05:41:19 -07001720 struct sock *sk = skb->sk;
Jens Axboe35f3d142010-05-20 10:43:18 +02001721 int ret = 0;
1722
1723 if (splice_grow_spd(pipe, &spd))
1724 return -ENOMEM;
Jens Axboe9c55e012007-11-06 23:30:13 -08001725
1726 /*
1727 * __skb_splice_bits() only fails if the output has no room left,
1728 * so no point in going over the frag_list for the error case.
1729 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001730 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
Jens Axboe9c55e012007-11-06 23:30:13 -08001731 goto done;
1732 else if (!tlen)
1733 goto done;
1734
1735 /*
1736 * now see if we have a frag_list to map
1737 */
David S. Millerfbb398a2009-06-09 00:18:59 -07001738 skb_walk_frags(skb, frag_iter) {
1739 if (!tlen)
1740 break;
Jens Axboe35f3d142010-05-20 10:43:18 +02001741 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
David S. Millerfbb398a2009-06-09 00:18:59 -07001742 break;
Jens Axboe9c55e012007-11-06 23:30:13 -08001743 }
1744
1745done:
Jens Axboe9c55e012007-11-06 23:30:13 -08001746 if (spd.nr_pages) {
Jens Axboe9c55e012007-11-06 23:30:13 -08001747 /*
1748 * Drop the socket lock, otherwise we have reverse
1749 * locking dependencies between sk_lock and i_mutex
1750 * here as compared to sendfile(). We enter here
1751 * with the socket lock held, and splice_to_pipe() will
1752 * grab the pipe inode lock. For sendfile() emulation,
1753 * we call into ->sendpage() with the i_mutex lock held
1754 * and networking will grab the socket lock.
1755 */
Octavian Purdila293ad602008-06-04 15:45:58 -07001756 release_sock(sk);
Jens Axboe9c55e012007-11-06 23:30:13 -08001757 ret = splice_to_pipe(pipe, &spd);
Octavian Purdila293ad602008-06-04 15:45:58 -07001758 lock_sock(sk);
Jens Axboe9c55e012007-11-06 23:30:13 -08001759 }
1760
Jens Axboe35f3d142010-05-20 10:43:18 +02001761 splice_shrink_spd(pipe, &spd);
1762 return ret;
Jens Axboe9c55e012007-11-06 23:30:13 -08001763}
1764
Herbert Xu357b40a2005-04-19 22:30:14 -07001765/**
1766 * skb_store_bits - store bits from kernel buffer to skb
1767 * @skb: destination buffer
1768 * @offset: offset in destination
1769 * @from: source buffer
1770 * @len: number of bytes to copy
1771 *
1772 * Copy the specified number of bytes from the source buffer to the
1773 * destination skb. This function handles all the messy bits of
1774 * traversing fragment lists and such.
1775 */
1776
Stephen Hemminger0c6fcc82007-04-20 16:40:01 -07001777int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
Herbert Xu357b40a2005-04-19 22:30:14 -07001778{
David S. Miller1a028e52007-04-27 15:21:23 -07001779 int start = skb_headlen(skb);
David S. Millerfbb398a2009-06-09 00:18:59 -07001780 struct sk_buff *frag_iter;
1781 int i, copy;
Herbert Xu357b40a2005-04-19 22:30:14 -07001782
1783 if (offset > (int)skb->len - len)
1784 goto fault;
1785
David S. Miller1a028e52007-04-27 15:21:23 -07001786 if ((copy = start - offset) > 0) {
Herbert Xu357b40a2005-04-19 22:30:14 -07001787 if (copy > len)
1788 copy = len;
Arnaldo Carvalho de Melo27d7ff42007-03-31 11:55:19 -03001789 skb_copy_to_linear_data_offset(skb, offset, from, copy);
Herbert Xu357b40a2005-04-19 22:30:14 -07001790 if ((len -= copy) == 0)
1791 return 0;
1792 offset += copy;
1793 from += copy;
1794 }
1795
1796 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1797 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
David S. Miller1a028e52007-04-27 15:21:23 -07001798 int end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001799
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001800 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07001801
Eric Dumazet9e903e02011-10-18 21:00:24 +00001802 end = start + skb_frag_size(frag);
Herbert Xu357b40a2005-04-19 22:30:14 -07001803 if ((copy = end - offset) > 0) {
1804 u8 *vaddr;
1805
1806 if (copy > len)
1807 copy = len;
1808
Eric Dumazet51c56b02012-04-05 11:35:15 +02001809 vaddr = kmap_atomic(skb_frag_page(frag));
David S. Miller1a028e52007-04-27 15:21:23 -07001810 memcpy(vaddr + frag->page_offset + offset - start,
1811 from, copy);
Eric Dumazet51c56b02012-04-05 11:35:15 +02001812 kunmap_atomic(vaddr);
Herbert Xu357b40a2005-04-19 22:30:14 -07001813
1814 if ((len -= copy) == 0)
1815 return 0;
1816 offset += copy;
1817 from += copy;
1818 }
David S. Miller1a028e52007-04-27 15:21:23 -07001819 start = end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001820 }
1821
David S. Millerfbb398a2009-06-09 00:18:59 -07001822 skb_walk_frags(skb, frag_iter) {
1823 int end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001824
David S. Millerfbb398a2009-06-09 00:18:59 -07001825 WARN_ON(start > offset + len);
Herbert Xu357b40a2005-04-19 22:30:14 -07001826
David S. Millerfbb398a2009-06-09 00:18:59 -07001827 end = start + frag_iter->len;
1828 if ((copy = end - offset) > 0) {
1829 if (copy > len)
1830 copy = len;
1831 if (skb_store_bits(frag_iter, offset - start,
1832 from, copy))
1833 goto fault;
1834 if ((len -= copy) == 0)
1835 return 0;
1836 offset += copy;
1837 from += copy;
Herbert Xu357b40a2005-04-19 22:30:14 -07001838 }
David S. Millerfbb398a2009-06-09 00:18:59 -07001839 start = end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001840 }
1841 if (!len)
1842 return 0;
1843
1844fault:
1845 return -EFAULT;
1846}
Herbert Xu357b40a2005-04-19 22:30:14 -07001847EXPORT_SYMBOL(skb_store_bits);
1848
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849/* Checksum skb data. */
1850
Al Viro2bbbc862006-11-14 21:37:14 -08001851__wsum skb_checksum(const struct sk_buff *skb, int offset,
1852 int len, __wsum csum)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853{
David S. Miller1a028e52007-04-27 15:21:23 -07001854 int start = skb_headlen(skb);
1855 int i, copy = start - offset;
David S. Millerfbb398a2009-06-09 00:18:59 -07001856 struct sk_buff *frag_iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 int pos = 0;
1858
1859 /* Checksum header. */
1860 if (copy > 0) {
1861 if (copy > len)
1862 copy = len;
1863 csum = csum_partial(skb->data + offset, copy, csum);
1864 if ((len -= copy) == 0)
1865 return csum;
1866 offset += copy;
1867 pos = copy;
1868 }
1869
1870 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07001871 int end;
Eric Dumazet51c56b02012-04-05 11:35:15 +02001872 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001874 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07001875
Eric Dumazet51c56b02012-04-05 11:35:15 +02001876 end = start + skb_frag_size(frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 if ((copy = end - offset) > 0) {
Al Viro44bb9362006-11-14 21:36:14 -08001878 __wsum csum2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 u8 *vaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880
1881 if (copy > len)
1882 copy = len;
Eric Dumazet51c56b02012-04-05 11:35:15 +02001883 vaddr = kmap_atomic(skb_frag_page(frag));
David S. Miller1a028e52007-04-27 15:21:23 -07001884 csum2 = csum_partial(vaddr + frag->page_offset +
1885 offset - start, copy, 0);
Eric Dumazet51c56b02012-04-05 11:35:15 +02001886 kunmap_atomic(vaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 csum = csum_block_add(csum, csum2, pos);
1888 if (!(len -= copy))
1889 return csum;
1890 offset += copy;
1891 pos += copy;
1892 }
David S. Miller1a028e52007-04-27 15:21:23 -07001893 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 }
1895
David S. Millerfbb398a2009-06-09 00:18:59 -07001896 skb_walk_frags(skb, frag_iter) {
1897 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898
David S. Millerfbb398a2009-06-09 00:18:59 -07001899 WARN_ON(start > offset + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900
David S. Millerfbb398a2009-06-09 00:18:59 -07001901 end = start + frag_iter->len;
1902 if ((copy = end - offset) > 0) {
1903 __wsum csum2;
1904 if (copy > len)
1905 copy = len;
1906 csum2 = skb_checksum(frag_iter, offset - start,
1907 copy, 0);
1908 csum = csum_block_add(csum, csum2, pos);
1909 if ((len -= copy) == 0)
1910 return csum;
1911 offset += copy;
1912 pos += copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 }
David S. Millerfbb398a2009-06-09 00:18:59 -07001914 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 }
Kris Katterjohn09a62662006-01-08 22:24:28 -08001916 BUG_ON(len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917
1918 return csum;
1919}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001920EXPORT_SYMBOL(skb_checksum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921
1922/* Both of above in one bottle. */
1923
Al Viro81d77662006-11-14 21:37:33 -08001924__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1925 u8 *to, int len, __wsum csum)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926{
David S. Miller1a028e52007-04-27 15:21:23 -07001927 int start = skb_headlen(skb);
1928 int i, copy = start - offset;
David S. Millerfbb398a2009-06-09 00:18:59 -07001929 struct sk_buff *frag_iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 int pos = 0;
1931
1932 /* Copy header. */
1933 if (copy > 0) {
1934 if (copy > len)
1935 copy = len;
1936 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1937 copy, csum);
1938 if ((len -= copy) == 0)
1939 return csum;
1940 offset += copy;
1941 to += copy;
1942 pos = copy;
1943 }
1944
1945 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07001946 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001948 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07001949
Eric Dumazet9e903e02011-10-18 21:00:24 +00001950 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 if ((copy = end - offset) > 0) {
Al Viro50842052006-11-14 21:36:34 -08001952 __wsum csum2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 u8 *vaddr;
1954 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1955
1956 if (copy > len)
1957 copy = len;
Eric Dumazet51c56b02012-04-05 11:35:15 +02001958 vaddr = kmap_atomic(skb_frag_page(frag));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 csum2 = csum_partial_copy_nocheck(vaddr +
David S. Miller1a028e52007-04-27 15:21:23 -07001960 frag->page_offset +
1961 offset - start, to,
1962 copy, 0);
Eric Dumazet51c56b02012-04-05 11:35:15 +02001963 kunmap_atomic(vaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 csum = csum_block_add(csum, csum2, pos);
1965 if (!(len -= copy))
1966 return csum;
1967 offset += copy;
1968 to += copy;
1969 pos += copy;
1970 }
David S. Miller1a028e52007-04-27 15:21:23 -07001971 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 }
1973
David S. Millerfbb398a2009-06-09 00:18:59 -07001974 skb_walk_frags(skb, frag_iter) {
1975 __wsum csum2;
1976 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977
David S. Millerfbb398a2009-06-09 00:18:59 -07001978 WARN_ON(start > offset + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979
David S. Millerfbb398a2009-06-09 00:18:59 -07001980 end = start + frag_iter->len;
1981 if ((copy = end - offset) > 0) {
1982 if (copy > len)
1983 copy = len;
1984 csum2 = skb_copy_and_csum_bits(frag_iter,
1985 offset - start,
1986 to, copy, 0);
1987 csum = csum_block_add(csum, csum2, pos);
1988 if ((len -= copy) == 0)
1989 return csum;
1990 offset += copy;
1991 to += copy;
1992 pos += copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 }
David S. Millerfbb398a2009-06-09 00:18:59 -07001994 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 }
Kris Katterjohn09a62662006-01-08 22:24:28 -08001996 BUG_ON(len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 return csum;
1998}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001999EXPORT_SYMBOL(skb_copy_and_csum_bits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000
2001void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2002{
Al Virod3bc23e2006-11-14 21:24:49 -08002003 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 long csstart;
2005
Patrick McHardy84fa7932006-08-29 16:44:56 -07002006 if (skb->ip_summed == CHECKSUM_PARTIAL)
Michał Mirosław55508d62010-12-14 15:24:08 +00002007 csstart = skb_checksum_start_offset(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 else
2009 csstart = skb_headlen(skb);
2010
Kris Katterjohn09a62662006-01-08 22:24:28 -08002011 BUG_ON(csstart > skb_headlen(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03002013 skb_copy_from_linear_data(skb, to, csstart);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014
2015 csum = 0;
2016 if (csstart != skb->len)
2017 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2018 skb->len - csstart, 0);
2019
Patrick McHardy84fa7932006-08-29 16:44:56 -07002020 if (skb->ip_summed == CHECKSUM_PARTIAL) {
Al Viroff1dcad2006-11-20 18:07:29 -08002021 long csstuff = csstart + skb->csum_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022
Al Virod3bc23e2006-11-14 21:24:49 -08002023 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024 }
2025}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002026EXPORT_SYMBOL(skb_copy_and_csum_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027
2028/**
2029 * skb_dequeue - remove from the head of the queue
2030 * @list: list to dequeue from
2031 *
2032 * Remove the head of the list. The list lock is taken so the function
2033 * may be used safely with other locking list functions. The head item is
2034 * returned or %NULL if the list is empty.
2035 */
2036
2037struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2038{
2039 unsigned long flags;
2040 struct sk_buff *result;
2041
2042 spin_lock_irqsave(&list->lock, flags);
2043 result = __skb_dequeue(list);
2044 spin_unlock_irqrestore(&list->lock, flags);
2045 return result;
2046}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002047EXPORT_SYMBOL(skb_dequeue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048
2049/**
2050 * skb_dequeue_tail - remove from the tail of the queue
2051 * @list: list to dequeue from
2052 *
2053 * Remove the tail of the list. The list lock is taken so the function
2054 * may be used safely with other locking list functions. The tail item is
2055 * returned or %NULL if the list is empty.
2056 */
2057struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2058{
2059 unsigned long flags;
2060 struct sk_buff *result;
2061
2062 spin_lock_irqsave(&list->lock, flags);
2063 result = __skb_dequeue_tail(list);
2064 spin_unlock_irqrestore(&list->lock, flags);
2065 return result;
2066}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002067EXPORT_SYMBOL(skb_dequeue_tail);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068
2069/**
2070 * skb_queue_purge - empty a list
2071 * @list: list to empty
2072 *
2073 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2074 * the list and one reference dropped. This function takes the list
2075 * lock and is atomic with respect to other list locking functions.
2076 */
2077void skb_queue_purge(struct sk_buff_head *list)
2078{
2079 struct sk_buff *skb;
2080 while ((skb = skb_dequeue(list)) != NULL)
2081 kfree_skb(skb);
2082}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002083EXPORT_SYMBOL(skb_queue_purge);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084
2085/**
2086 * skb_queue_head - queue a buffer at the list head
2087 * @list: list to use
2088 * @newsk: buffer to queue
2089 *
2090 * Queue a buffer at the start of the list. This function takes the
2091 * list lock and can be used safely with other locking &sk_buff functions
2092 * safely.
2093 *
2094 * A buffer cannot be placed on two lists at the same time.
2095 */
2096void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2097{
2098 unsigned long flags;
2099
2100 spin_lock_irqsave(&list->lock, flags);
2101 __skb_queue_head(list, newsk);
2102 spin_unlock_irqrestore(&list->lock, flags);
2103}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002104EXPORT_SYMBOL(skb_queue_head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105
2106/**
2107 * skb_queue_tail - queue a buffer at the list tail
2108 * @list: list to use
2109 * @newsk: buffer to queue
2110 *
2111 * Queue a buffer at the tail of the list. This function takes the
2112 * list lock and can be used safely with other locking &sk_buff functions
2113 * safely.
2114 *
2115 * A buffer cannot be placed on two lists at the same time.
2116 */
2117void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2118{
2119 unsigned long flags;
2120
2121 spin_lock_irqsave(&list->lock, flags);
2122 __skb_queue_tail(list, newsk);
2123 spin_unlock_irqrestore(&list->lock, flags);
2124}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002125EXPORT_SYMBOL(skb_queue_tail);
David S. Miller8728b832005-08-09 19:25:21 -07002126
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127/**
2128 * skb_unlink - remove a buffer from a list
2129 * @skb: buffer to remove
David S. Miller8728b832005-08-09 19:25:21 -07002130 * @list: list to use
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 *
David S. Miller8728b832005-08-09 19:25:21 -07002132 * Remove a packet from a list. The list locks are taken and this
2133 * function is atomic with respect to other list locked calls
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 *
David S. Miller8728b832005-08-09 19:25:21 -07002135 * You must know what list the SKB is on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 */
David S. Miller8728b832005-08-09 19:25:21 -07002137void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138{
David S. Miller8728b832005-08-09 19:25:21 -07002139 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140
David S. Miller8728b832005-08-09 19:25:21 -07002141 spin_lock_irqsave(&list->lock, flags);
2142 __skb_unlink(skb, list);
2143 spin_unlock_irqrestore(&list->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002145EXPORT_SYMBOL(skb_unlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147/**
2148 * skb_append - append a buffer
2149 * @old: buffer to insert after
2150 * @newsk: buffer to insert
David S. Miller8728b832005-08-09 19:25:21 -07002151 * @list: list to use
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 *
2153 * Place a packet after a given packet in a list. The list locks are taken
2154 * and this function is atomic with respect to other list locked calls.
2155 * A buffer cannot be placed on two lists at the same time.
2156 */
David S. Miller8728b832005-08-09 19:25:21 -07002157void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158{
2159 unsigned long flags;
2160
David S. Miller8728b832005-08-09 19:25:21 -07002161 spin_lock_irqsave(&list->lock, flags);
Gerrit Renker7de6c032008-04-14 00:05:09 -07002162 __skb_queue_after(list, old, newsk);
David S. Miller8728b832005-08-09 19:25:21 -07002163 spin_unlock_irqrestore(&list->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002165EXPORT_SYMBOL(skb_append);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166
2167/**
2168 * skb_insert - insert a buffer
2169 * @old: buffer to insert before
2170 * @newsk: buffer to insert
David S. Miller8728b832005-08-09 19:25:21 -07002171 * @list: list to use
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 *
David S. Miller8728b832005-08-09 19:25:21 -07002173 * Place a packet before a given packet in a list. The list locks are
2174 * taken and this function is atomic with respect to other list locked
2175 * calls.
2176 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 * A buffer cannot be placed on two lists at the same time.
2178 */
David S. Miller8728b832005-08-09 19:25:21 -07002179void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180{
2181 unsigned long flags;
2182
David S. Miller8728b832005-08-09 19:25:21 -07002183 spin_lock_irqsave(&list->lock, flags);
2184 __skb_insert(newsk, old->prev, old, list);
2185 spin_unlock_irqrestore(&list->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002187EXPORT_SYMBOL(skb_insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189static inline void skb_split_inside_header(struct sk_buff *skb,
2190 struct sk_buff* skb1,
2191 const u32 len, const int pos)
2192{
2193 int i;
2194
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03002195 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2196 pos - len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197 /* And move data appendix as is. */
2198 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2199 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2200
2201 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2202 skb_shinfo(skb)->nr_frags = 0;
2203 skb1->data_len = skb->data_len;
2204 skb1->len += skb1->data_len;
2205 skb->data_len = 0;
2206 skb->len = len;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07002207 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208}
2209
2210static inline void skb_split_no_header(struct sk_buff *skb,
2211 struct sk_buff* skb1,
2212 const u32 len, int pos)
2213{
2214 int i, k = 0;
2215 const int nfrags = skb_shinfo(skb)->nr_frags;
2216
2217 skb_shinfo(skb)->nr_frags = 0;
2218 skb1->len = skb1->data_len = skb->len - len;
2219 skb->len = len;
2220 skb->data_len = len - pos;
2221
2222 for (i = 0; i < nfrags; i++) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00002223 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224
2225 if (pos + size > len) {
2226 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2227
2228 if (pos < len) {
2229 /* Split frag.
2230 * We have two variants in this case:
2231 * 1. Move all the frag to the second
2232 * part, if it is possible. F.e.
2233 * this approach is mandatory for TUX,
2234 * where splitting is expensive.
2235 * 2. Split is accurately. We make this.
2236 */
Ian Campbellea2ab692011-08-22 23:44:58 +00002237 skb_frag_ref(skb, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
Eric Dumazet9e903e02011-10-18 21:00:24 +00002239 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2240 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241 skb_shinfo(skb)->nr_frags++;
2242 }
2243 k++;
2244 } else
2245 skb_shinfo(skb)->nr_frags++;
2246 pos += size;
2247 }
2248 skb_shinfo(skb1)->nr_frags = k;
2249}
2250
2251/**
2252 * skb_split - Split fragmented skb to two parts at length len.
2253 * @skb: the buffer to split
2254 * @skb1: the buffer to receive the second part
2255 * @len: new length for skb
2256 */
2257void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2258{
2259 int pos = skb_headlen(skb);
2260
2261 if (len < pos) /* Split line is inside header. */
2262 skb_split_inside_header(skb, skb1, len, pos);
2263 else /* Second chunk has no header, nothing to copy. */
2264 skb_split_no_header(skb, skb1, len, pos);
2265}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002266EXPORT_SYMBOL(skb_split);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267
Ilpo Järvinen9f782db2008-11-25 13:57:01 -08002268/* Shifting from/to a cloned skb is a no-go.
2269 *
2270 * Caller cannot keep skb_shinfo related pointers past calling here!
2271 */
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002272static int skb_prepare_for_shift(struct sk_buff *skb)
2273{
Ilpo Järvinen0ace2852008-11-24 21:30:21 -08002274 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002275}
2276
2277/**
2278 * skb_shift - Shifts paged data partially from skb to another
2279 * @tgt: buffer into which tail data gets added
2280 * @skb: buffer from which the paged data comes from
2281 * @shiftlen: shift up to this many bytes
2282 *
2283 * Attempts to shift up to shiftlen worth of bytes, which may be less than
Feng King20e994a2011-11-21 01:47:11 +00002284 * the length of the skb, from skb to tgt. Returns number bytes shifted.
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002285 * It's up to caller to free skb if everything was shifted.
2286 *
2287 * If @tgt runs out of frags, the whole operation is aborted.
2288 *
2289 * Skb cannot include anything else but paged data while tgt is allowed
2290 * to have non-paged data as well.
2291 *
2292 * TODO: full sized shift could be optimized but that would need
2293 * specialized skb free'er to handle frags without up-to-date nr_frags.
2294 */
2295int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2296{
2297 int from, to, merge, todo;
2298 struct skb_frag_struct *fragfrom, *fragto;
2299
2300 BUG_ON(shiftlen > skb->len);
2301 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2302
2303 todo = shiftlen;
2304 from = 0;
2305 to = skb_shinfo(tgt)->nr_frags;
2306 fragfrom = &skb_shinfo(skb)->frags[from];
2307
2308 /* Actual merge is delayed until the point when we know we can
2309 * commit all, so that we don't have to undo partial changes
2310 */
2311 if (!to ||
Ian Campbellea2ab692011-08-22 23:44:58 +00002312 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2313 fragfrom->page_offset)) {
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002314 merge = -1;
2315 } else {
2316 merge = to - 1;
2317
Eric Dumazet9e903e02011-10-18 21:00:24 +00002318 todo -= skb_frag_size(fragfrom);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002319 if (todo < 0) {
2320 if (skb_prepare_for_shift(skb) ||
2321 skb_prepare_for_shift(tgt))
2322 return 0;
2323
Ilpo Järvinen9f782db2008-11-25 13:57:01 -08002324 /* All previous frag pointers might be stale! */
2325 fragfrom = &skb_shinfo(skb)->frags[from];
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002326 fragto = &skb_shinfo(tgt)->frags[merge];
2327
Eric Dumazet9e903e02011-10-18 21:00:24 +00002328 skb_frag_size_add(fragto, shiftlen);
2329 skb_frag_size_sub(fragfrom, shiftlen);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002330 fragfrom->page_offset += shiftlen;
2331
2332 goto onlymerged;
2333 }
2334
2335 from++;
2336 }
2337
2338 /* Skip full, not-fitting skb to avoid expensive operations */
2339 if ((shiftlen == skb->len) &&
2340 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2341 return 0;
2342
2343 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2344 return 0;
2345
2346 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2347 if (to == MAX_SKB_FRAGS)
2348 return 0;
2349
2350 fragfrom = &skb_shinfo(skb)->frags[from];
2351 fragto = &skb_shinfo(tgt)->frags[to];
2352
Eric Dumazet9e903e02011-10-18 21:00:24 +00002353 if (todo >= skb_frag_size(fragfrom)) {
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002354 *fragto = *fragfrom;
Eric Dumazet9e903e02011-10-18 21:00:24 +00002355 todo -= skb_frag_size(fragfrom);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002356 from++;
2357 to++;
2358
2359 } else {
Ian Campbellea2ab692011-08-22 23:44:58 +00002360 __skb_frag_ref(fragfrom);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002361 fragto->page = fragfrom->page;
2362 fragto->page_offset = fragfrom->page_offset;
Eric Dumazet9e903e02011-10-18 21:00:24 +00002363 skb_frag_size_set(fragto, todo);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002364
2365 fragfrom->page_offset += todo;
Eric Dumazet9e903e02011-10-18 21:00:24 +00002366 skb_frag_size_sub(fragfrom, todo);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002367 todo = 0;
2368
2369 to++;
2370 break;
2371 }
2372 }
2373
2374 /* Ready to "commit" this state change to tgt */
2375 skb_shinfo(tgt)->nr_frags = to;
2376
2377 if (merge >= 0) {
2378 fragfrom = &skb_shinfo(skb)->frags[0];
2379 fragto = &skb_shinfo(tgt)->frags[merge];
2380
Eric Dumazet9e903e02011-10-18 21:00:24 +00002381 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
Ian Campbellea2ab692011-08-22 23:44:58 +00002382 __skb_frag_unref(fragfrom);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002383 }
2384
2385 /* Reposition in the original skb */
2386 to = 0;
2387 while (from < skb_shinfo(skb)->nr_frags)
2388 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2389 skb_shinfo(skb)->nr_frags = to;
2390
2391 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2392
2393onlymerged:
2394 /* Most likely the tgt won't ever need its checksum anymore, skb on
2395 * the other hand might need it if it needs to be resent
2396 */
2397 tgt->ip_summed = CHECKSUM_PARTIAL;
2398 skb->ip_summed = CHECKSUM_PARTIAL;
2399
2400 /* Yak, is it really working this way? Some helper please? */
2401 skb->len -= shiftlen;
2402 skb->data_len -= shiftlen;
2403 skb->truesize -= shiftlen;
2404 tgt->len += shiftlen;
2405 tgt->data_len += shiftlen;
2406 tgt->truesize += shiftlen;
2407
2408 return shiftlen;
2409}
2410
Thomas Graf677e90e2005-06-23 20:59:51 -07002411/**
2412 * skb_prepare_seq_read - Prepare a sequential read of skb data
2413 * @skb: the buffer to read
2414 * @from: lower offset of data to be read
2415 * @to: upper offset of data to be read
2416 * @st: state variable
2417 *
2418 * Initializes the specified state variable. Must be called before
2419 * invoking skb_seq_read() for the first time.
2420 */
2421void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2422 unsigned int to, struct skb_seq_state *st)
2423{
2424 st->lower_offset = from;
2425 st->upper_offset = to;
2426 st->root_skb = st->cur_skb = skb;
2427 st->frag_idx = st->stepped_offset = 0;
2428 st->frag_data = NULL;
2429}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002430EXPORT_SYMBOL(skb_prepare_seq_read);
Thomas Graf677e90e2005-06-23 20:59:51 -07002431
2432/**
2433 * skb_seq_read - Sequentially read skb data
2434 * @consumed: number of bytes consumed by the caller so far
2435 * @data: destination pointer for data to be returned
2436 * @st: state variable
2437 *
2438 * Reads a block of skb data at &consumed relative to the
2439 * lower offset specified to skb_prepare_seq_read(). Assigns
2440 * the head of the data block to &data and returns the length
2441 * of the block or 0 if the end of the skb data or the upper
2442 * offset has been reached.
2443 *
2444 * The caller is not required to consume all of the data
2445 * returned, i.e. &consumed is typically set to the number
2446 * of bytes already consumed and the next call to
2447 * skb_seq_read() will return the remaining part of the block.
2448 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002449 * Note 1: The size of each block of data returned can be arbitrary,
Thomas Graf677e90e2005-06-23 20:59:51 -07002450 * this limitation is the cost for zerocopy seqeuental
2451 * reads of potentially non linear data.
2452 *
Randy Dunlapbc2cda12008-02-13 15:03:25 -08002453 * Note 2: Fragment lists within fragments are not implemented
Thomas Graf677e90e2005-06-23 20:59:51 -07002454 * at the moment, state->root_skb could be replaced with
2455 * a stack for this purpose.
2456 */
2457unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2458 struct skb_seq_state *st)
2459{
2460 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2461 skb_frag_t *frag;
2462
2463 if (unlikely(abs_offset >= st->upper_offset))
2464 return 0;
2465
2466next_skb:
Herbert Xu95e3b242009-01-29 16:07:52 -08002467 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
Thomas Graf677e90e2005-06-23 20:59:51 -07002468
Thomas Chenault995b3372009-05-18 21:43:27 -07002469 if (abs_offset < block_limit && !st->frag_data) {
Herbert Xu95e3b242009-01-29 16:07:52 -08002470 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
Thomas Graf677e90e2005-06-23 20:59:51 -07002471 return block_limit - abs_offset;
2472 }
2473
2474 if (st->frag_idx == 0 && !st->frag_data)
2475 st->stepped_offset += skb_headlen(st->cur_skb);
2476
2477 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2478 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
Eric Dumazet9e903e02011-10-18 21:00:24 +00002479 block_limit = skb_frag_size(frag) + st->stepped_offset;
Thomas Graf677e90e2005-06-23 20:59:51 -07002480
2481 if (abs_offset < block_limit) {
2482 if (!st->frag_data)
Eric Dumazet51c56b02012-04-05 11:35:15 +02002483 st->frag_data = kmap_atomic(skb_frag_page(frag));
Thomas Graf677e90e2005-06-23 20:59:51 -07002484
2485 *data = (u8 *) st->frag_data + frag->page_offset +
2486 (abs_offset - st->stepped_offset);
2487
2488 return block_limit - abs_offset;
2489 }
2490
2491 if (st->frag_data) {
Eric Dumazet51c56b02012-04-05 11:35:15 +02002492 kunmap_atomic(st->frag_data);
Thomas Graf677e90e2005-06-23 20:59:51 -07002493 st->frag_data = NULL;
2494 }
2495
2496 st->frag_idx++;
Eric Dumazet9e903e02011-10-18 21:00:24 +00002497 st->stepped_offset += skb_frag_size(frag);
Thomas Graf677e90e2005-06-23 20:59:51 -07002498 }
2499
Olaf Kirch5b5a60d2007-06-23 23:11:52 -07002500 if (st->frag_data) {
Eric Dumazet51c56b02012-04-05 11:35:15 +02002501 kunmap_atomic(st->frag_data);
Olaf Kirch5b5a60d2007-06-23 23:11:52 -07002502 st->frag_data = NULL;
2503 }
2504
David S. Miller21dc3302010-08-23 00:13:46 -07002505 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
Shyam Iyer71b33462009-01-29 16:12:42 -08002506 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
Thomas Graf677e90e2005-06-23 20:59:51 -07002507 st->frag_idx = 0;
2508 goto next_skb;
Shyam Iyer71b33462009-01-29 16:12:42 -08002509 } else if (st->cur_skb->next) {
2510 st->cur_skb = st->cur_skb->next;
Herbert Xu95e3b242009-01-29 16:07:52 -08002511 st->frag_idx = 0;
Thomas Graf677e90e2005-06-23 20:59:51 -07002512 goto next_skb;
2513 }
2514
2515 return 0;
2516}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002517EXPORT_SYMBOL(skb_seq_read);
Thomas Graf677e90e2005-06-23 20:59:51 -07002518
2519/**
2520 * skb_abort_seq_read - Abort a sequential read of skb data
2521 * @st: state variable
2522 *
2523 * Must be called if skb_seq_read() was not called until it
2524 * returned 0.
2525 */
2526void skb_abort_seq_read(struct skb_seq_state *st)
2527{
2528 if (st->frag_data)
Eric Dumazet51c56b02012-04-05 11:35:15 +02002529 kunmap_atomic(st->frag_data);
Thomas Graf677e90e2005-06-23 20:59:51 -07002530}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002531EXPORT_SYMBOL(skb_abort_seq_read);
Thomas Graf677e90e2005-06-23 20:59:51 -07002532
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002533#define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2534
2535static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2536 struct ts_config *conf,
2537 struct ts_state *state)
2538{
2539 return skb_seq_read(offset, text, TS_SKB_CB(state));
2540}
2541
2542static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2543{
2544 skb_abort_seq_read(TS_SKB_CB(state));
2545}
2546
2547/**
2548 * skb_find_text - Find a text pattern in skb data
2549 * @skb: the buffer to look in
2550 * @from: search offset
2551 * @to: search limit
2552 * @config: textsearch configuration
2553 * @state: uninitialized textsearch state variable
2554 *
2555 * Finds a pattern in the skb data according to the specified
2556 * textsearch configuration. Use textsearch_next() to retrieve
2557 * subsequent occurrences of the pattern. Returns the offset
2558 * to the first occurrence or UINT_MAX if no match was found.
2559 */
2560unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2561 unsigned int to, struct ts_config *config,
2562 struct ts_state *state)
2563{
Phil Oesterf72b9482006-06-26 00:00:57 -07002564 unsigned int ret;
2565
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002566 config->get_next_block = skb_ts_get_next_block;
2567 config->finish = skb_ts_finish;
2568
2569 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2570
Phil Oesterf72b9482006-06-26 00:00:57 -07002571 ret = textsearch_find(config, state);
2572 return (ret <= to - from ? ret : UINT_MAX);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002573}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002574EXPORT_SYMBOL(skb_find_text);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002575
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002576/**
2577 * skb_append_datato_frags: - append the user data to a skb
2578 * @sk: sock structure
2579 * @skb: skb structure to be appened with user data.
2580 * @getfrag: call back function to be used for getting the user data
2581 * @from: pointer to user message iov
2582 * @length: length of the iov message
2583 *
2584 * Description: This procedure append the user data in the fragment part
2585 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2586 */
2587int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
Martin Waitzdab96302005-12-05 13:40:12 -08002588 int (*getfrag)(void *from, char *to, int offset,
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002589 int len, int odd, struct sk_buff *skb),
2590 void *from, int length)
2591{
2592 int frg_cnt = 0;
2593 skb_frag_t *frag = NULL;
2594 struct page *page = NULL;
2595 int copy, left;
2596 int offset = 0;
2597 int ret;
2598
2599 do {
2600 /* Return error if we don't have space for new frag */
2601 frg_cnt = skb_shinfo(skb)->nr_frags;
2602 if (frg_cnt >= MAX_SKB_FRAGS)
2603 return -EFAULT;
2604
2605 /* allocate a new page for next frag */
2606 page = alloc_pages(sk->sk_allocation, 0);
2607
2608 /* If alloc_page fails just return failure and caller will
2609 * free previous allocated pages by doing kfree_skb()
2610 */
2611 if (page == NULL)
2612 return -ENOMEM;
2613
2614 /* initialize the next frag */
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002615 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2616 skb->truesize += PAGE_SIZE;
2617 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2618
2619 /* get the new initialized frag */
2620 frg_cnt = skb_shinfo(skb)->nr_frags;
2621 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2622
2623 /* copy the user data to page */
2624 left = PAGE_SIZE - frag->page_offset;
2625 copy = (length > left)? left : length;
2626
Eric Dumazet9e903e02011-10-18 21:00:24 +00002627 ret = getfrag(from, skb_frag_address(frag) + skb_frag_size(frag),
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002628 offset, copy, 0, skb);
2629 if (ret < 0)
2630 return -EFAULT;
2631
2632 /* copy was successful so update the size parameters */
Eric Dumazet9e903e02011-10-18 21:00:24 +00002633 skb_frag_size_add(frag, copy);
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002634 skb->len += copy;
2635 skb->data_len += copy;
2636 offset += copy;
2637 length -= copy;
2638
2639 } while (length > 0);
2640
2641 return 0;
2642}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002643EXPORT_SYMBOL(skb_append_datato_frags);
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002644
Herbert Xucbb042f2006-03-20 22:43:56 -08002645/**
2646 * skb_pull_rcsum - pull skb and update receive checksum
2647 * @skb: buffer to update
Herbert Xucbb042f2006-03-20 22:43:56 -08002648 * @len: length of data pulled
2649 *
2650 * This function performs an skb_pull on the packet and updates
Urs Thuermannfee54fa2008-02-12 22:03:25 -08002651 * the CHECKSUM_COMPLETE checksum. It should be used on
Patrick McHardy84fa7932006-08-29 16:44:56 -07002652 * receive path processing instead of skb_pull unless you know
2653 * that the checksum difference is zero (e.g., a valid IP header)
2654 * or you are setting ip_summed to CHECKSUM_NONE.
Herbert Xucbb042f2006-03-20 22:43:56 -08002655 */
2656unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2657{
2658 BUG_ON(len > skb->len);
2659 skb->len -= len;
2660 BUG_ON(skb->len < skb->data_len);
2661 skb_postpull_rcsum(skb, skb->data, len);
2662 return skb->data += len;
2663}
Arnaldo Carvalho de Melof94691a2006-03-20 22:47:55 -08002664EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2665
Herbert Xuf4c50d92006-06-22 03:02:40 -07002666/**
2667 * skb_segment - Perform protocol segmentation on skb.
2668 * @skb: buffer to segment
Herbert Xu576a30e2006-06-27 13:22:38 -07002669 * @features: features for the output path (see dev->features)
Herbert Xuf4c50d92006-06-22 03:02:40 -07002670 *
2671 * This function performs segmentation on the given skb. It returns
Ben Hutchings4c821d72008-04-13 21:52:48 -07002672 * a pointer to the first in a list of new skbs for the segments.
2673 * In case of error it returns ERR_PTR(err).
Herbert Xuf4c50d92006-06-22 03:02:40 -07002674 */
Michał Mirosławc8f44af2011-11-15 15:29:55 +00002675struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features)
Herbert Xuf4c50d92006-06-22 03:02:40 -07002676{
2677 struct sk_buff *segs = NULL;
2678 struct sk_buff *tail = NULL;
Herbert Xu89319d382008-12-15 23:26:06 -08002679 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002680 unsigned int mss = skb_shinfo(skb)->gso_size;
Arnaldo Carvalho de Melo98e399f2007-03-19 15:33:04 -07002681 unsigned int doffset = skb->data - skb_mac_header(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002682 unsigned int offset = doffset;
2683 unsigned int headroom;
2684 unsigned int len;
Michał Mirosław04ed3e72011-01-24 15:32:47 -08002685 int sg = !!(features & NETIF_F_SG);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002686 int nfrags = skb_shinfo(skb)->nr_frags;
2687 int err = -ENOMEM;
2688 int i = 0;
2689 int pos;
2690
2691 __skb_push(skb, doffset);
2692 headroom = skb_headroom(skb);
2693 pos = skb_headlen(skb);
2694
2695 do {
2696 struct sk_buff *nskb;
2697 skb_frag_t *frag;
Herbert Xuc8884ed2006-10-29 15:59:41 -08002698 int hsize;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002699 int size;
2700
2701 len = skb->len - offset;
2702 if (len > mss)
2703 len = mss;
2704
2705 hsize = skb_headlen(skb) - offset;
2706 if (hsize < 0)
2707 hsize = 0;
Herbert Xuc8884ed2006-10-29 15:59:41 -08002708 if (hsize > len || !sg)
2709 hsize = len;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002710
Herbert Xu89319d382008-12-15 23:26:06 -08002711 if (!hsize && i >= nfrags) {
2712 BUG_ON(fskb->len != len);
2713
2714 pos += len;
2715 nskb = skb_clone(fskb, GFP_ATOMIC);
2716 fskb = fskb->next;
2717
2718 if (unlikely(!nskb))
2719 goto err;
2720
2721 hsize = skb_end_pointer(nskb) - nskb->head;
2722 if (skb_cow_head(nskb, doffset + headroom)) {
2723 kfree_skb(nskb);
2724 goto err;
2725 }
2726
2727 nskb->truesize += skb_end_pointer(nskb) - nskb->head -
2728 hsize;
2729 skb_release_head_state(nskb);
2730 __skb_push(nskb, doffset);
2731 } else {
2732 nskb = alloc_skb(hsize + doffset + headroom,
2733 GFP_ATOMIC);
2734
2735 if (unlikely(!nskb))
2736 goto err;
2737
2738 skb_reserve(nskb, headroom);
2739 __skb_put(nskb, doffset);
2740 }
Herbert Xuf4c50d92006-06-22 03:02:40 -07002741
2742 if (segs)
2743 tail->next = nskb;
2744 else
2745 segs = nskb;
2746 tail = nskb;
2747
Herbert Xu6f85a122008-08-15 14:55:02 -07002748 __copy_skb_header(nskb, skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002749 nskb->mac_len = skb->mac_len;
2750
Eric Dumazet3d3be432010-09-01 00:50:51 +00002751 /* nskb and skb might have different headroom */
2752 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2753 nskb->csum_start += skb_headroom(nskb) - headroom;
2754
Arnaldo Carvalho de Melo459a98e2007-03-19 15:30:44 -07002755 skb_reset_mac_header(nskb);
Arnaldo Carvalho de Meloddc7b8e2007-03-15 21:42:27 -03002756 skb_set_network_header(nskb, skb->mac_len);
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -07002757 nskb->transport_header = (nskb->network_header +
2758 skb_network_header_len(skb));
Herbert Xu89319d382008-12-15 23:26:06 -08002759 skb_copy_from_linear_data(skb, nskb->data, doffset);
2760
Herbert Xu2f181852009-03-28 23:39:18 -07002761 if (fskb != skb_shinfo(skb)->frag_list)
Herbert Xu89319d382008-12-15 23:26:06 -08002762 continue;
2763
Herbert Xuf4c50d92006-06-22 03:02:40 -07002764 if (!sg) {
Herbert Xu6f85a122008-08-15 14:55:02 -07002765 nskb->ip_summed = CHECKSUM_NONE;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002766 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2767 skb_put(nskb, len),
2768 len, 0);
2769 continue;
2770 }
2771
2772 frag = skb_shinfo(nskb)->frags;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002773
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03002774 skb_copy_from_linear_data_offset(skb, offset,
2775 skb_put(nskb, hsize), hsize);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002776
Herbert Xu89319d382008-12-15 23:26:06 -08002777 while (pos < offset + len && i < nfrags) {
Herbert Xuf4c50d92006-06-22 03:02:40 -07002778 *frag = skb_shinfo(skb)->frags[i];
Ian Campbellea2ab692011-08-22 23:44:58 +00002779 __skb_frag_ref(frag);
Eric Dumazet9e903e02011-10-18 21:00:24 +00002780 size = skb_frag_size(frag);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002781
2782 if (pos < offset) {
2783 frag->page_offset += offset - pos;
Eric Dumazet9e903e02011-10-18 21:00:24 +00002784 skb_frag_size_sub(frag, offset - pos);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002785 }
2786
Herbert Xu89319d382008-12-15 23:26:06 -08002787 skb_shinfo(nskb)->nr_frags++;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002788
2789 if (pos + size <= offset + len) {
2790 i++;
2791 pos += size;
2792 } else {
Eric Dumazet9e903e02011-10-18 21:00:24 +00002793 skb_frag_size_sub(frag, pos + size - (offset + len));
Herbert Xu89319d382008-12-15 23:26:06 -08002794 goto skip_fraglist;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002795 }
2796
2797 frag++;
2798 }
2799
Herbert Xu89319d382008-12-15 23:26:06 -08002800 if (pos < offset + len) {
2801 struct sk_buff *fskb2 = fskb;
2802
2803 BUG_ON(pos + fskb->len != offset + len);
2804
2805 pos += fskb->len;
2806 fskb = fskb->next;
2807
2808 if (fskb2->next) {
2809 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2810 if (!fskb2)
2811 goto err;
2812 } else
2813 skb_get(fskb2);
2814
David S. Millerfbb398a2009-06-09 00:18:59 -07002815 SKB_FRAG_ASSERT(nskb);
Herbert Xu89319d382008-12-15 23:26:06 -08002816 skb_shinfo(nskb)->frag_list = fskb2;
2817 }
2818
2819skip_fraglist:
Herbert Xuf4c50d92006-06-22 03:02:40 -07002820 nskb->data_len = len - hsize;
2821 nskb->len += nskb->data_len;
2822 nskb->truesize += nskb->data_len;
2823 } while ((offset += len) < skb->len);
2824
2825 return segs;
2826
2827err:
2828 while ((skb = segs)) {
2829 segs = skb->next;
Patrick McHardyb08d5842007-02-27 09:57:37 -08002830 kfree_skb(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002831 }
2832 return ERR_PTR(err);
2833}
Herbert Xuf4c50d92006-06-22 03:02:40 -07002834EXPORT_SYMBOL_GPL(skb_segment);
2835
Herbert Xu71d93b32008-12-15 23:42:33 -08002836int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2837{
2838 struct sk_buff *p = *head;
2839 struct sk_buff *nskb;
Herbert Xu9aaa1562009-05-26 18:50:33 +00002840 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2841 struct skb_shared_info *pinfo = skb_shinfo(p);
Herbert Xu71d93b32008-12-15 23:42:33 -08002842 unsigned int headroom;
Herbert Xu86911732009-01-29 14:19:50 +00002843 unsigned int len = skb_gro_len(skb);
Herbert Xu67147ba2009-05-26 18:50:22 +00002844 unsigned int offset = skb_gro_offset(skb);
2845 unsigned int headlen = skb_headlen(skb);
Herbert Xu71d93b32008-12-15 23:42:33 -08002846
Herbert Xu86911732009-01-29 14:19:50 +00002847 if (p->len + len >= 65536)
Herbert Xu71d93b32008-12-15 23:42:33 -08002848 return -E2BIG;
2849
Herbert Xu9aaa1562009-05-26 18:50:33 +00002850 if (pinfo->frag_list)
Herbert Xu71d93b32008-12-15 23:42:33 -08002851 goto merge;
Herbert Xu67147ba2009-05-26 18:50:22 +00002852 else if (headlen <= offset) {
Herbert Xu42da6992009-05-26 18:50:19 +00002853 skb_frag_t *frag;
Herbert Xu66e92fc2009-05-26 18:50:32 +00002854 skb_frag_t *frag2;
Herbert Xu9aaa1562009-05-26 18:50:33 +00002855 int i = skbinfo->nr_frags;
2856 int nr_frags = pinfo->nr_frags + i;
Herbert Xu42da6992009-05-26 18:50:19 +00002857
Herbert Xu66e92fc2009-05-26 18:50:32 +00002858 offset -= headlen;
2859
2860 if (nr_frags > MAX_SKB_FRAGS)
Herbert Xu81705ad2009-01-29 14:19:51 +00002861 return -E2BIG;
2862
Herbert Xu9aaa1562009-05-26 18:50:33 +00002863 pinfo->nr_frags = nr_frags;
2864 skbinfo->nr_frags = 0;
Herbert Xuf5572062009-01-14 20:40:03 -08002865
Herbert Xu9aaa1562009-05-26 18:50:33 +00002866 frag = pinfo->frags + nr_frags;
2867 frag2 = skbinfo->frags + i;
Herbert Xu66e92fc2009-05-26 18:50:32 +00002868 do {
2869 *--frag = *--frag2;
2870 } while (--i);
2871
2872 frag->page_offset += offset;
Eric Dumazet9e903e02011-10-18 21:00:24 +00002873 skb_frag_size_sub(frag, offset);
Herbert Xu66e92fc2009-05-26 18:50:32 +00002874
Herbert Xuf5572062009-01-14 20:40:03 -08002875 skb->truesize -= skb->data_len;
2876 skb->len -= skb->data_len;
2877 skb->data_len = 0;
2878
Herbert Xu5d38a072009-01-04 16:13:40 -08002879 NAPI_GRO_CB(skb)->free = 1;
2880 goto done;
Herbert Xu69c0cab2009-11-17 05:18:18 -08002881 } else if (skb_gro_len(p) != pinfo->gso_size)
2882 return -E2BIG;
Herbert Xu71d93b32008-12-15 23:42:33 -08002883
2884 headroom = skb_headroom(p);
Eric Dumazet3d3be432010-09-01 00:50:51 +00002885 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
Herbert Xu71d93b32008-12-15 23:42:33 -08002886 if (unlikely(!nskb))
2887 return -ENOMEM;
2888
2889 __copy_skb_header(nskb, p);
2890 nskb->mac_len = p->mac_len;
2891
2892 skb_reserve(nskb, headroom);
Herbert Xu86911732009-01-29 14:19:50 +00002893 __skb_put(nskb, skb_gro_offset(p));
Herbert Xu71d93b32008-12-15 23:42:33 -08002894
Herbert Xu86911732009-01-29 14:19:50 +00002895 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
Herbert Xu71d93b32008-12-15 23:42:33 -08002896 skb_set_network_header(nskb, skb_network_offset(p));
2897 skb_set_transport_header(nskb, skb_transport_offset(p));
2898
Herbert Xu86911732009-01-29 14:19:50 +00002899 __skb_pull(p, skb_gro_offset(p));
2900 memcpy(skb_mac_header(nskb), skb_mac_header(p),
2901 p->data - skb_mac_header(p));
Herbert Xu71d93b32008-12-15 23:42:33 -08002902
2903 *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2904 skb_shinfo(nskb)->frag_list = p;
Herbert Xu9aaa1562009-05-26 18:50:33 +00002905 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
Herbert Xu622e0ca2010-05-20 23:07:56 -07002906 pinfo->gso_size = 0;
Herbert Xu71d93b32008-12-15 23:42:33 -08002907 skb_header_release(p);
2908 nskb->prev = p;
2909
2910 nskb->data_len += p->len;
Eric Dumazetde8261c2012-02-13 04:09:20 +00002911 nskb->truesize += p->truesize;
Herbert Xu71d93b32008-12-15 23:42:33 -08002912 nskb->len += p->len;
2913
2914 *head = nskb;
2915 nskb->next = p->next;
2916 p->next = NULL;
2917
2918 p = nskb;
2919
2920merge:
Eric Dumazetde8261c2012-02-13 04:09:20 +00002921 p->truesize += skb->truesize - len;
Herbert Xu67147ba2009-05-26 18:50:22 +00002922 if (offset > headlen) {
Michal Schmidtd1dc7ab2011-01-24 12:08:48 +00002923 unsigned int eat = offset - headlen;
2924
2925 skbinfo->frags[0].page_offset += eat;
Eric Dumazet9e903e02011-10-18 21:00:24 +00002926 skb_frag_size_sub(&skbinfo->frags[0], eat);
Michal Schmidtd1dc7ab2011-01-24 12:08:48 +00002927 skb->data_len -= eat;
2928 skb->len -= eat;
Herbert Xu67147ba2009-05-26 18:50:22 +00002929 offset = headlen;
Herbert Xu56035022009-02-05 21:26:52 -08002930 }
2931
Herbert Xu67147ba2009-05-26 18:50:22 +00002932 __skb_pull(skb, offset);
Herbert Xu56035022009-02-05 21:26:52 -08002933
Herbert Xu71d93b32008-12-15 23:42:33 -08002934 p->prev->next = skb;
2935 p->prev = skb;
2936 skb_header_release(skb);
2937
Herbert Xu5d38a072009-01-04 16:13:40 -08002938done:
2939 NAPI_GRO_CB(p)->count++;
Herbert Xu37fe4732009-01-17 19:48:13 +00002940 p->data_len += len;
2941 p->truesize += len;
2942 p->len += len;
Herbert Xu71d93b32008-12-15 23:42:33 -08002943
2944 NAPI_GRO_CB(skb)->same_flow = 1;
2945 return 0;
2946}
2947EXPORT_SYMBOL_GPL(skb_gro_receive);
2948
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949void __init skb_init(void)
2950{
2951 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2952 sizeof(struct sk_buff),
2953 0,
Alexey Dobriyane5d679f332006-08-26 19:25:52 -07002954 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
Paul Mundt20c2df82007-07-20 10:11:58 +09002955 NULL);
David S. Millerd179cd12005-08-17 14:57:30 -07002956 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2957 (2*sizeof(struct sk_buff)) +
2958 sizeof(atomic_t),
2959 0,
Alexey Dobriyane5d679f332006-08-26 19:25:52 -07002960 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
Paul Mundt20c2df82007-07-20 10:11:58 +09002961 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002962}
2963
David Howells716ea3a2007-04-02 20:19:53 -07002964/**
2965 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2966 * @skb: Socket buffer containing the buffers to be mapped
2967 * @sg: The scatter-gather list to map into
2968 * @offset: The offset into the buffer's contents to start mapping
2969 * @len: Length of buffer space to be mapped
2970 *
2971 * Fill the specified scatter-gather list with mappings/pointers into a
2972 * region of the buffer space attached to a socket buffer.
2973 */
David S. Miller51c739d2007-10-30 21:29:29 -07002974static int
2975__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
David Howells716ea3a2007-04-02 20:19:53 -07002976{
David S. Miller1a028e52007-04-27 15:21:23 -07002977 int start = skb_headlen(skb);
2978 int i, copy = start - offset;
David S. Millerfbb398a2009-06-09 00:18:59 -07002979 struct sk_buff *frag_iter;
David Howells716ea3a2007-04-02 20:19:53 -07002980 int elt = 0;
2981
2982 if (copy > 0) {
2983 if (copy > len)
2984 copy = len;
Jens Axboe642f149032007-10-24 11:20:47 +02002985 sg_set_buf(sg, skb->data + offset, copy);
David Howells716ea3a2007-04-02 20:19:53 -07002986 elt++;
2987 if ((len -= copy) == 0)
2988 return elt;
2989 offset += copy;
2990 }
2991
2992 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07002993 int end;
David Howells716ea3a2007-04-02 20:19:53 -07002994
Ilpo Järvinen547b7922008-07-25 21:43:18 -07002995 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07002996
Eric Dumazet9e903e02011-10-18 21:00:24 +00002997 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
David Howells716ea3a2007-04-02 20:19:53 -07002998 if ((copy = end - offset) > 0) {
2999 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3000
3001 if (copy > len)
3002 copy = len;
Ian Campbellea2ab692011-08-22 23:44:58 +00003003 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
Jens Axboe642f149032007-10-24 11:20:47 +02003004 frag->page_offset+offset-start);
David Howells716ea3a2007-04-02 20:19:53 -07003005 elt++;
3006 if (!(len -= copy))
3007 return elt;
3008 offset += copy;
3009 }
David S. Miller1a028e52007-04-27 15:21:23 -07003010 start = end;
David Howells716ea3a2007-04-02 20:19:53 -07003011 }
3012
David S. Millerfbb398a2009-06-09 00:18:59 -07003013 skb_walk_frags(skb, frag_iter) {
3014 int end;
David Howells716ea3a2007-04-02 20:19:53 -07003015
David S. Millerfbb398a2009-06-09 00:18:59 -07003016 WARN_ON(start > offset + len);
David Howells716ea3a2007-04-02 20:19:53 -07003017
David S. Millerfbb398a2009-06-09 00:18:59 -07003018 end = start + frag_iter->len;
3019 if ((copy = end - offset) > 0) {
3020 if (copy > len)
3021 copy = len;
3022 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3023 copy);
3024 if ((len -= copy) == 0)
3025 return elt;
3026 offset += copy;
David Howells716ea3a2007-04-02 20:19:53 -07003027 }
David S. Millerfbb398a2009-06-09 00:18:59 -07003028 start = end;
David Howells716ea3a2007-04-02 20:19:53 -07003029 }
3030 BUG_ON(len);
3031 return elt;
3032}
3033
David S. Miller51c739d2007-10-30 21:29:29 -07003034int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3035{
3036 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3037
Jens Axboec46f2332007-10-31 12:06:37 +01003038 sg_mark_end(&sg[nsg - 1]);
David S. Miller51c739d2007-10-30 21:29:29 -07003039
3040 return nsg;
3041}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08003042EXPORT_SYMBOL_GPL(skb_to_sgvec);
David S. Miller51c739d2007-10-30 21:29:29 -07003043
David Howells716ea3a2007-04-02 20:19:53 -07003044/**
3045 * skb_cow_data - Check that a socket buffer's data buffers are writable
3046 * @skb: The socket buffer to check.
3047 * @tailbits: Amount of trailing space to be added
3048 * @trailer: Returned pointer to the skb where the @tailbits space begins
3049 *
3050 * Make sure that the data buffers attached to a socket buffer are
3051 * writable. If they are not, private copies are made of the data buffers
3052 * and the socket buffer is set to use these instead.
3053 *
3054 * If @tailbits is given, make sure that there is space to write @tailbits
3055 * bytes of data beyond current end of socket buffer. @trailer will be
3056 * set to point to the skb in which this space begins.
3057 *
3058 * The number of scatterlist elements required to completely map the
3059 * COW'd and extended socket buffer will be returned.
3060 */
3061int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3062{
3063 int copyflag;
3064 int elt;
3065 struct sk_buff *skb1, **skb_p;
3066
3067 /* If skb is cloned or its head is paged, reallocate
3068 * head pulling out all the pages (pages are considered not writable
3069 * at the moment even if they are anonymous).
3070 */
3071 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3072 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3073 return -ENOMEM;
3074
3075 /* Easy case. Most of packets will go this way. */
David S. Miller21dc3302010-08-23 00:13:46 -07003076 if (!skb_has_frag_list(skb)) {
David Howells716ea3a2007-04-02 20:19:53 -07003077 /* A little of trouble, not enough of space for trailer.
3078 * This should not happen, when stack is tuned to generate
3079 * good frames. OK, on miss we reallocate and reserve even more
3080 * space, 128 bytes is fair. */
3081
3082 if (skb_tailroom(skb) < tailbits &&
3083 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3084 return -ENOMEM;
3085
3086 /* Voila! */
3087 *trailer = skb;
3088 return 1;
3089 }
3090
3091 /* Misery. We are in troubles, going to mincer fragments... */
3092
3093 elt = 1;
3094 skb_p = &skb_shinfo(skb)->frag_list;
3095 copyflag = 0;
3096
3097 while ((skb1 = *skb_p) != NULL) {
3098 int ntail = 0;
3099
3100 /* The fragment is partially pulled by someone,
3101 * this can happen on input. Copy it and everything
3102 * after it. */
3103
3104 if (skb_shared(skb1))
3105 copyflag = 1;
3106
3107 /* If the skb is the last, worry about trailer. */
3108
3109 if (skb1->next == NULL && tailbits) {
3110 if (skb_shinfo(skb1)->nr_frags ||
David S. Miller21dc3302010-08-23 00:13:46 -07003111 skb_has_frag_list(skb1) ||
David Howells716ea3a2007-04-02 20:19:53 -07003112 skb_tailroom(skb1) < tailbits)
3113 ntail = tailbits + 128;
3114 }
3115
3116 if (copyflag ||
3117 skb_cloned(skb1) ||
3118 ntail ||
3119 skb_shinfo(skb1)->nr_frags ||
David S. Miller21dc3302010-08-23 00:13:46 -07003120 skb_has_frag_list(skb1)) {
David Howells716ea3a2007-04-02 20:19:53 -07003121 struct sk_buff *skb2;
3122
3123 /* Fuck, we are miserable poor guys... */
3124 if (ntail == 0)
3125 skb2 = skb_copy(skb1, GFP_ATOMIC);
3126 else
3127 skb2 = skb_copy_expand(skb1,
3128 skb_headroom(skb1),
3129 ntail,
3130 GFP_ATOMIC);
3131 if (unlikely(skb2 == NULL))
3132 return -ENOMEM;
3133
3134 if (skb1->sk)
3135 skb_set_owner_w(skb2, skb1->sk);
3136
3137 /* Looking around. Are we still alive?
3138 * OK, link new skb, drop old one */
3139
3140 skb2->next = skb1->next;
3141 *skb_p = skb2;
3142 kfree_skb(skb1);
3143 skb1 = skb2;
3144 }
3145 elt++;
3146 *trailer = skb1;
3147 skb_p = &skb1->next;
3148 }
3149
3150 return elt;
3151}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08003152EXPORT_SYMBOL_GPL(skb_cow_data);
David Howells716ea3a2007-04-02 20:19:53 -07003153
Eric Dumazetb1faf562010-05-31 23:44:05 -07003154static void sock_rmem_free(struct sk_buff *skb)
3155{
3156 struct sock *sk = skb->sk;
3157
3158 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3159}
3160
3161/*
3162 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3163 */
3164int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3165{
Eric Dumazet110c4332012-04-06 10:49:10 +02003166 int len = skb->len;
3167
Eric Dumazetb1faf562010-05-31 23:44:05 -07003168 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
Eric Dumazet95c96172012-04-15 05:58:06 +00003169 (unsigned int)sk->sk_rcvbuf)
Eric Dumazetb1faf562010-05-31 23:44:05 -07003170 return -ENOMEM;
3171
3172 skb_orphan(skb);
3173 skb->sk = sk;
3174 skb->destructor = sock_rmem_free;
3175 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3176
Eric Dumazetabb57ea2011-05-18 02:21:31 -04003177 /* before exiting rcu section, make sure dst is refcounted */
3178 skb_dst_force(skb);
3179
Eric Dumazetb1faf562010-05-31 23:44:05 -07003180 skb_queue_tail(&sk->sk_error_queue, skb);
3181 if (!sock_flag(sk, SOCK_DEAD))
Eric Dumazet110c4332012-04-06 10:49:10 +02003182 sk->sk_data_ready(sk, len);
Eric Dumazetb1faf562010-05-31 23:44:05 -07003183 return 0;
3184}
3185EXPORT_SYMBOL(sock_queue_err_skb);
3186
Patrick Ohlyac45f602009-02-12 05:03:37 +00003187void skb_tstamp_tx(struct sk_buff *orig_skb,
3188 struct skb_shared_hwtstamps *hwtstamps)
3189{
3190 struct sock *sk = orig_skb->sk;
3191 struct sock_exterr_skb *serr;
3192 struct sk_buff *skb;
3193 int err;
3194
3195 if (!sk)
3196 return;
3197
3198 skb = skb_clone(orig_skb, GFP_ATOMIC);
3199 if (!skb)
3200 return;
3201
3202 if (hwtstamps) {
3203 *skb_hwtstamps(skb) =
3204 *hwtstamps;
3205 } else {
3206 /*
3207 * no hardware time stamps available,
Oliver Hartkopp2244d072010-08-17 08:59:14 +00003208 * so keep the shared tx_flags and only
Patrick Ohlyac45f602009-02-12 05:03:37 +00003209 * store software time stamp
3210 */
3211 skb->tstamp = ktime_get_real();
3212 }
3213
3214 serr = SKB_EXT_ERR(skb);
3215 memset(serr, 0, sizeof(*serr));
3216 serr->ee.ee_errno = ENOMSG;
3217 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
Eric Dumazet29030372010-05-29 00:20:48 -07003218
Patrick Ohlyac45f602009-02-12 05:03:37 +00003219 err = sock_queue_err_skb(sk, skb);
Eric Dumazet29030372010-05-29 00:20:48 -07003220
Patrick Ohlyac45f602009-02-12 05:03:37 +00003221 if (err)
3222 kfree_skb(skb);
3223}
3224EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3225
Johannes Berg6e3e9392011-11-09 10:15:42 +01003226void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3227{
3228 struct sock *sk = skb->sk;
3229 struct sock_exterr_skb *serr;
3230 int err;
3231
3232 skb->wifi_acked_valid = 1;
3233 skb->wifi_acked = acked;
3234
3235 serr = SKB_EXT_ERR(skb);
3236 memset(serr, 0, sizeof(*serr));
3237 serr->ee.ee_errno = ENOMSG;
3238 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3239
3240 err = sock_queue_err_skb(sk, skb);
3241 if (err)
3242 kfree_skb(skb);
3243}
3244EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3245
Patrick Ohlyac45f602009-02-12 05:03:37 +00003246
Rusty Russellf35d9d82008-02-04 23:49:54 -05003247/**
3248 * skb_partial_csum_set - set up and verify partial csum values for packet
3249 * @skb: the skb to set
3250 * @start: the number of bytes after skb->data to start checksumming.
3251 * @off: the offset from start to place the checksum.
3252 *
3253 * For untrusted partially-checksummed packets, we need to make sure the values
3254 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3255 *
3256 * This function checks and sets those values and skb->ip_summed: if this
3257 * returns false you should drop the packet.
3258 */
3259bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3260{
Herbert Xu5ff8dda2009-06-04 01:22:01 +00003261 if (unlikely(start > skb_headlen(skb)) ||
3262 unlikely((int)start + off > skb_headlen(skb) - 2)) {
Rusty Russellf35d9d82008-02-04 23:49:54 -05003263 if (net_ratelimit())
3264 printk(KERN_WARNING
3265 "bad partial csum: csum=%u/%u len=%u\n",
Herbert Xu5ff8dda2009-06-04 01:22:01 +00003266 start, off, skb_headlen(skb));
Rusty Russellf35d9d82008-02-04 23:49:54 -05003267 return false;
3268 }
3269 skb->ip_summed = CHECKSUM_PARTIAL;
3270 skb->csum_start = skb_headroom(skb) + start;
3271 skb->csum_offset = off;
3272 return true;
3273}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08003274EXPORT_SYMBOL_GPL(skb_partial_csum_set);
Rusty Russellf35d9d82008-02-04 23:49:54 -05003275
Ben Hutchings4497b072008-06-19 16:22:28 -07003276void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3277{
3278 if (net_ratelimit())
3279 pr_warning("%s: received packets cannot be forwarded"
3280 " while LRO is enabled\n", skb->dev->name);
3281}
Ben Hutchings4497b072008-06-19 16:22:28 -07003282EXPORT_SYMBOL(__skb_warn_lro_forwarding);