blob: 1b62343f58378b3d8fc0e3ea048dbb45ce1e3a76 [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
Joe Perchese005d192012-05-16 19:58:40 +000039#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
40
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include <linux/module.h>
42#include <linux/types.h>
43#include <linux/kernel.h>
Vegard Nossumfe55f6d2008-08-30 12:16:35 +020044#include <linux/kmemcheck.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070045#include <linux/mm.h>
46#include <linux/interrupt.h>
47#include <linux/in.h>
48#include <linux/inet.h>
49#include <linux/slab.h>
Florian Westphalde960aa2014-01-26 10:58:16 +010050#include <linux/tcp.h>
51#include <linux/udp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include <linux/netdevice.h>
53#ifdef CONFIG_NET_CLS_ACT
54#include <net/pkt_sched.h>
55#endif
56#include <linux/string.h>
57#include <linux/skbuff.h>
Jens Axboe9c55e012007-11-06 23:30:13 -080058#include <linux/splice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070059#include <linux/cache.h>
60#include <linux/rtnetlink.h>
61#include <linux/init.h>
David Howells716ea3a2007-04-02 20:19:53 -070062#include <linux/scatterlist.h>
Patrick Ohlyac45f602009-02-12 05:03:37 +000063#include <linux/errqueue.h>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070064#include <linux/prefetch.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070065
66#include <net/protocol.h>
67#include <net/dst.h>
68#include <net/sock.h>
69#include <net/checksum.h>
Paul Durranted1f50c2014-01-09 10:02:46 +000070#include <net/ip6_checksum.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070071#include <net/xfrm.h>
72
73#include <asm/uaccess.h>
Steven Rostedtad8d75f2009-04-14 19:39:12 -040074#include <trace/events/skb.h>
Eric Dumazet51c56b02012-04-05 11:35:15 +020075#include <linux/highmem.h>
Al Viroa1f8e7f72006-10-19 16:08:53 -040076
Eric Dumazetd7e88832012-04-30 08:10:34 +000077struct kmem_cache *skbuff_head_cache __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -080078static struct kmem_cache *skbuff_fclone_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070079
Linus Torvalds1da177e2005-04-16 15:20:36 -070080/**
Jean Sacrenf05de732013-02-11 13:30:38 +000081 * skb_panic - private function for out-of-line support
82 * @skb: buffer
83 * @sz: size
84 * @addr: address
James Hogan99d58512013-02-13 11:20:27 +000085 * @msg: skb_over_panic or skb_under_panic
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 *
Jean Sacrenf05de732013-02-11 13:30:38 +000087 * Out-of-line support for skb_put() and skb_push().
88 * Called via the wrapper skb_over_panic() or skb_under_panic().
89 * Keep out of line to prevent kernel bloat.
90 * __builtin_return_address is not used because it is not always reliable.
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 */
Jean Sacrenf05de732013-02-11 13:30:38 +000092static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
James Hogan99d58512013-02-13 11:20:27 +000093 const char msg[])
Linus Torvalds1da177e2005-04-16 15:20:36 -070094{
Joe Perchese005d192012-05-16 19:58:40 +000095 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
James Hogan99d58512013-02-13 11:20:27 +000096 msg, addr, skb->len, sz, skb->head, skb->data,
Joe Perchese005d192012-05-16 19:58:40 +000097 (unsigned long)skb->tail, (unsigned long)skb->end,
98 skb->dev ? skb->dev->name : "<NULL>");
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 BUG();
100}
101
Jean Sacrenf05de732013-02-11 13:30:38 +0000102static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103{
Jean Sacrenf05de732013-02-11 13:30:38 +0000104 skb_panic(skb, sz, addr, __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105}
106
Jean Sacrenf05de732013-02-11 13:30:38 +0000107static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
108{
109 skb_panic(skb, sz, addr, __func__);
110}
Mel Gormanc93bdd02012-07-31 16:44:19 -0700111
112/*
113 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
114 * the caller if emergency pfmemalloc reserves are being used. If it is and
115 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
116 * may be used. Otherwise, the packet data may be discarded until enough
117 * memory is free
118 */
119#define kmalloc_reserve(size, gfp, node, pfmemalloc) \
120 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
stephen hemminger61c5e882012-12-28 18:24:28 +0000121
122static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
123 unsigned long ip, bool *pfmemalloc)
Mel Gormanc93bdd02012-07-31 16:44:19 -0700124{
125 void *obj;
126 bool ret_pfmemalloc = false;
127
128 /*
129 * Try a regular allocation, when that fails and we're not entitled
130 * to the reserves, fail.
131 */
132 obj = kmalloc_node_track_caller(size,
133 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
134 node);
135 if (obj || !(gfp_pfmemalloc_allowed(flags)))
136 goto out;
137
138 /* Try again but now we are using pfmemalloc reserves */
139 ret_pfmemalloc = true;
140 obj = kmalloc_node_track_caller(size, flags, node);
141
142out:
143 if (pfmemalloc)
144 *pfmemalloc = ret_pfmemalloc;
145
146 return obj;
147}
148
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149/* Allocate a new skbuff. We do this ourselves so we can fill in a few
150 * 'private' fields and also do memory statistics to find all the
151 * [BEEP] leaks.
152 *
153 */
154
Patrick McHardy0ebd0ac2013-04-17 06:46:58 +0000155struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node)
156{
157 struct sk_buff *skb;
158
159 /* Get the HEAD */
160 skb = kmem_cache_alloc_node(skbuff_head_cache,
161 gfp_mask & ~__GFP_DMA, node);
162 if (!skb)
163 goto out;
164
165 /*
166 * Only clear those fields we need to clear, not those that we will
167 * actually initialise below. Hence, don't put any more fields after
168 * the tail pointer in struct sk_buff!
169 */
170 memset(skb, 0, offsetof(struct sk_buff, tail));
Pablo Neira5e71d9d2013-06-03 09:28:43 +0000171 skb->head = NULL;
Patrick McHardy0ebd0ac2013-04-17 06:46:58 +0000172 skb->truesize = sizeof(struct sk_buff);
173 atomic_set(&skb->users, 1);
174
Cong Wang35d04612013-05-29 15:16:05 +0800175 skb->mac_header = (typeof(skb->mac_header))~0U;
Patrick McHardy0ebd0ac2013-04-17 06:46:58 +0000176out:
177 return skb;
178}
179
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180/**
David S. Millerd179cd12005-08-17 14:57:30 -0700181 * __alloc_skb - allocate a network buffer
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 * @size: size to allocate
183 * @gfp_mask: allocation mask
Mel Gormanc93bdd02012-07-31 16:44:19 -0700184 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
185 * instead of head cache and allocate a cloned (child) skb.
186 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
187 * allocations in case the data is required for writeback
Christoph Hellwigb30973f2006-12-06 20:32:36 -0800188 * @node: numa node to allocate memory on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 *
190 * Allocate a new &sk_buff. The returned buffer has no headroom and a
Ben Hutchings94b60422012-06-06 15:23:37 +0000191 * tail room of at least size bytes. The object has a reference count
192 * of one. The return is the buffer. On a failure the return is %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 *
194 * Buffers may only be allocated from interrupts using a @gfp_mask of
195 * %GFP_ATOMIC.
196 */
Al Virodd0fc662005-10-07 07:46:04 +0100197struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
Mel Gormanc93bdd02012-07-31 16:44:19 -0700198 int flags, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199{
Christoph Lametere18b8902006-12-06 20:33:20 -0800200 struct kmem_cache *cache;
Benjamin LaHaise4947d3e2006-01-03 14:06:50 -0800201 struct skb_shared_info *shinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 struct sk_buff *skb;
203 u8 *data;
Mel Gormanc93bdd02012-07-31 16:44:19 -0700204 bool pfmemalloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205
Mel Gormanc93bdd02012-07-31 16:44:19 -0700206 cache = (flags & SKB_ALLOC_FCLONE)
207 ? skbuff_fclone_cache : skbuff_head_cache;
208
209 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
210 gfp_mask |= __GFP_MEMALLOC;
Herbert Xu8798b3f2006-01-23 16:32:45 -0800211
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 /* Get the HEAD */
Christoph Hellwigb30973f2006-12-06 20:32:36 -0800213 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 if (!skb)
215 goto out;
Eric Dumazetec7d2f22010-05-05 01:07:37 -0700216 prefetchw(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217
Eric Dumazet87fb4b72011-10-13 07:28:54 +0000218 /* We do our best to align skb_shared_info on a separate cache
219 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
220 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
221 * Both skb->head and skb_shared_info are cache line aligned.
222 */
Tony Lindgrenbc417e32011-11-02 13:40:28 +0000223 size = SKB_DATA_ALIGN(size);
Eric Dumazet87fb4b72011-10-13 07:28:54 +0000224 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
Mel Gormanc93bdd02012-07-31 16:44:19 -0700225 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 if (!data)
227 goto nodata;
Eric Dumazet87fb4b72011-10-13 07:28:54 +0000228 /* kmalloc(size) might give us more room than requested.
229 * Put skb_shared_info exactly at the end of allocated zone,
230 * to allow max possible filling before reallocation.
231 */
232 size = SKB_WITH_OVERHEAD(ksize(data));
Eric Dumazetec7d2f22010-05-05 01:07:37 -0700233 prefetchw(data + size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234
Arnaldo Carvalho de Meloca0605a2007-03-19 10:48:59 -0300235 /*
Johannes Bergc8005782008-05-03 20:56:42 -0700236 * Only clear those fields we need to clear, not those that we will
237 * actually initialise below. Hence, don't put any more fields after
238 * the tail pointer in struct sk_buff!
Arnaldo Carvalho de Meloca0605a2007-03-19 10:48:59 -0300239 */
240 memset(skb, 0, offsetof(struct sk_buff, tail));
Eric Dumazet87fb4b72011-10-13 07:28:54 +0000241 /* Account for allocated memory : skb + skb->head */
242 skb->truesize = SKB_TRUESIZE(size);
Mel Gormanc93bdd02012-07-31 16:44:19 -0700243 skb->pfmemalloc = pfmemalloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 atomic_set(&skb->users, 1);
245 skb->head = data;
246 skb->data = data;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700247 skb_reset_tail_pointer(skb);
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700248 skb->end = skb->tail + size;
Cong Wang35d04612013-05-29 15:16:05 +0800249 skb->mac_header = (typeof(skb->mac_header))~0U;
250 skb->transport_header = (typeof(skb->transport_header))~0U;
Stephen Hemminger19633e12009-06-17 05:23:27 +0000251
Benjamin LaHaise4947d3e2006-01-03 14:06:50 -0800252 /* make sure we initialize shinfo sequentially */
253 shinfo = skb_shinfo(skb);
Eric Dumazetec7d2f22010-05-05 01:07:37 -0700254 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
Benjamin LaHaise4947d3e2006-01-03 14:06:50 -0800255 atomic_set(&shinfo->dataref, 1);
Eric Dumazetc2aa3662011-01-25 23:18:38 +0000256 kmemcheck_annotate_variable(shinfo->destructor_arg);
Benjamin LaHaise4947d3e2006-01-03 14:06:50 -0800257
Mel Gormanc93bdd02012-07-31 16:44:19 -0700258 if (flags & SKB_ALLOC_FCLONE) {
David S. Millerd179cd12005-08-17 14:57:30 -0700259 struct sk_buff *child = skb + 1;
260 atomic_t *fclone_ref = (atomic_t *) (child + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200262 kmemcheck_annotate_bitfield(child, flags1);
263 kmemcheck_annotate_bitfield(child, flags2);
David S. Millerd179cd12005-08-17 14:57:30 -0700264 skb->fclone = SKB_FCLONE_ORIG;
265 atomic_set(fclone_ref, 1);
266
267 child->fclone = SKB_FCLONE_UNAVAILABLE;
Mel Gormanc93bdd02012-07-31 16:44:19 -0700268 child->pfmemalloc = pfmemalloc;
David S. Millerd179cd12005-08-17 14:57:30 -0700269 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270out:
271 return skb;
272nodata:
Herbert Xu8798b3f2006-01-23 16:32:45 -0800273 kmem_cache_free(cache, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 skb = NULL;
275 goto out;
276}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800277EXPORT_SYMBOL(__alloc_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278
279/**
Eric Dumazetb2b5ce92011-11-14 06:03:34 +0000280 * build_skb - build a network buffer
281 * @data: data buffer provided by caller
Eric Dumazetd3836f22012-04-27 00:33:38 +0000282 * @frag_size: size of fragment, or 0 if head was kmalloced
Eric Dumazetb2b5ce92011-11-14 06:03:34 +0000283 *
284 * Allocate a new &sk_buff. Caller provides space holding head and
Florian Fainellideceb4c2013-07-23 20:22:39 +0100285 * skb_shared_info. @data must have been allocated by kmalloc() only if
286 * @frag_size is 0, otherwise data should come from the page allocator.
Eric Dumazetb2b5ce92011-11-14 06:03:34 +0000287 * The return is the new skb buffer.
288 * On a failure the return is %NULL, and @data is not freed.
289 * Notes :
290 * Before IO, driver allocates only data buffer where NIC put incoming frame
291 * Driver should add room at head (NET_SKB_PAD) and
292 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
293 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
294 * before giving packet to stack.
295 * RX rings only contains data buffers, not full skbs.
296 */
Eric Dumazetd3836f22012-04-27 00:33:38 +0000297struct sk_buff *build_skb(void *data, unsigned int frag_size)
Eric Dumazetb2b5ce92011-11-14 06:03:34 +0000298{
299 struct skb_shared_info *shinfo;
300 struct sk_buff *skb;
Eric Dumazetd3836f22012-04-27 00:33:38 +0000301 unsigned int size = frag_size ? : ksize(data);
Eric Dumazetb2b5ce92011-11-14 06:03:34 +0000302
303 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
304 if (!skb)
305 return NULL;
306
Eric Dumazetd3836f22012-04-27 00:33:38 +0000307 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
Eric Dumazetb2b5ce92011-11-14 06:03:34 +0000308
309 memset(skb, 0, offsetof(struct sk_buff, tail));
310 skb->truesize = SKB_TRUESIZE(size);
Eric Dumazetd3836f22012-04-27 00:33:38 +0000311 skb->head_frag = frag_size != 0;
Eric Dumazetb2b5ce92011-11-14 06:03:34 +0000312 atomic_set(&skb->users, 1);
313 skb->head = data;
314 skb->data = data;
315 skb_reset_tail_pointer(skb);
316 skb->end = skb->tail + size;
Cong Wang35d04612013-05-29 15:16:05 +0800317 skb->mac_header = (typeof(skb->mac_header))~0U;
318 skb->transport_header = (typeof(skb->transport_header))~0U;
Eric Dumazetb2b5ce92011-11-14 06:03:34 +0000319
320 /* make sure we initialize shinfo sequentially */
321 shinfo = skb_shinfo(skb);
322 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
323 atomic_set(&shinfo->dataref, 1);
324 kmemcheck_annotate_variable(shinfo->destructor_arg);
325
326 return skb;
327}
328EXPORT_SYMBOL(build_skb);
329
Eric Dumazeta1c7fff2012-05-17 07:34:16 +0000330struct netdev_alloc_cache {
Eric Dumazet69b08f62012-09-26 06:46:57 +0000331 struct page_frag frag;
332 /* we maintain a pagecount bias, so that we dont dirty cache line
333 * containing page->_count every time we allocate a fragment.
334 */
335 unsigned int pagecnt_bias;
Eric Dumazeta1c7fff2012-05-17 07:34:16 +0000336};
337static DEFINE_PER_CPU(struct netdev_alloc_cache, netdev_alloc_cache);
338
Mel Gormanc93bdd02012-07-31 16:44:19 -0700339static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
Eric Dumazet6f532612012-05-18 05:12:12 +0000340{
341 struct netdev_alloc_cache *nc;
342 void *data = NULL;
Eric Dumazet69b08f62012-09-26 06:46:57 +0000343 int order;
Eric Dumazet6f532612012-05-18 05:12:12 +0000344 unsigned long flags;
345
346 local_irq_save(flags);
347 nc = &__get_cpu_var(netdev_alloc_cache);
Eric Dumazet69b08f62012-09-26 06:46:57 +0000348 if (unlikely(!nc->frag.page)) {
Eric Dumazet6f532612012-05-18 05:12:12 +0000349refill:
Eric Dumazet69b08f62012-09-26 06:46:57 +0000350 for (order = NETDEV_FRAG_PAGE_MAX_ORDER; ;) {
351 gfp_t gfp = gfp_mask;
352
353 if (order)
354 gfp |= __GFP_COMP | __GFP_NOWARN;
355 nc->frag.page = alloc_pages(gfp, order);
356 if (likely(nc->frag.page))
357 break;
358 if (--order < 0)
359 goto end;
360 }
361 nc->frag.size = PAGE_SIZE << order;
Alexander Duyck540eb7b2012-07-12 14:23:50 +0000362recycle:
Eric Dumazet69b08f62012-09-26 06:46:57 +0000363 atomic_set(&nc->frag.page->_count, NETDEV_PAGECNT_MAX_BIAS);
364 nc->pagecnt_bias = NETDEV_PAGECNT_MAX_BIAS;
365 nc->frag.offset = 0;
Eric Dumazet6f532612012-05-18 05:12:12 +0000366 }
Alexander Duyck540eb7b2012-07-12 14:23:50 +0000367
Eric Dumazet69b08f62012-09-26 06:46:57 +0000368 if (nc->frag.offset + fragsz > nc->frag.size) {
Alexander Duyck540eb7b2012-07-12 14:23:50 +0000369 /* avoid unnecessary locked operations if possible */
Eric Dumazet69b08f62012-09-26 06:46:57 +0000370 if ((atomic_read(&nc->frag.page->_count) == nc->pagecnt_bias) ||
371 atomic_sub_and_test(nc->pagecnt_bias, &nc->frag.page->_count))
Alexander Duyck540eb7b2012-07-12 14:23:50 +0000372 goto recycle;
373 goto refill;
Eric Dumazet6f532612012-05-18 05:12:12 +0000374 }
Alexander Duyck540eb7b2012-07-12 14:23:50 +0000375
Eric Dumazet69b08f62012-09-26 06:46:57 +0000376 data = page_address(nc->frag.page) + nc->frag.offset;
377 nc->frag.offset += fragsz;
Alexander Duyck540eb7b2012-07-12 14:23:50 +0000378 nc->pagecnt_bias--;
379end:
Eric Dumazet6f532612012-05-18 05:12:12 +0000380 local_irq_restore(flags);
381 return data;
382}
Mel Gormanc93bdd02012-07-31 16:44:19 -0700383
384/**
385 * netdev_alloc_frag - allocate a page fragment
386 * @fragsz: fragment size
387 *
388 * Allocates a frag from a page for receive buffer.
389 * Uses GFP_ATOMIC allocations.
390 */
391void *netdev_alloc_frag(unsigned int fragsz)
392{
393 return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
394}
Eric Dumazet6f532612012-05-18 05:12:12 +0000395EXPORT_SYMBOL(netdev_alloc_frag);
396
397/**
Christoph Hellwig8af27452006-07-31 22:35:23 -0700398 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
399 * @dev: network device to receive on
400 * @length: length to allocate
401 * @gfp_mask: get_free_pages mask, passed to alloc_skb
402 *
403 * Allocate a new &sk_buff and assign it a usage count of one. The
404 * buffer has unspecified headroom built in. Users should allocate
405 * the headroom they think they need without accounting for the
406 * built in space. The built in space is used for optimisations.
407 *
408 * %NULL is returned if there is no free memory.
409 */
410struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
Eric Dumazet6f532612012-05-18 05:12:12 +0000411 unsigned int length, gfp_t gfp_mask)
Christoph Hellwig8af27452006-07-31 22:35:23 -0700412{
Eric Dumazet6f532612012-05-18 05:12:12 +0000413 struct sk_buff *skb = NULL;
Eric Dumazeta1c7fff2012-05-17 07:34:16 +0000414 unsigned int fragsz = SKB_DATA_ALIGN(length + NET_SKB_PAD) +
415 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
Christoph Hellwig8af27452006-07-31 22:35:23 -0700416
Eric Dumazet310e1582012-07-16 13:15:52 +0200417 if (fragsz <= PAGE_SIZE && !(gfp_mask & (__GFP_WAIT | GFP_DMA))) {
Mel Gormanc93bdd02012-07-31 16:44:19 -0700418 void *data;
419
420 if (sk_memalloc_socks())
421 gfp_mask |= __GFP_MEMALLOC;
422
423 data = __netdev_alloc_frag(fragsz, gfp_mask);
Eric Dumazeta1c7fff2012-05-17 07:34:16 +0000424
Eric Dumazet6f532612012-05-18 05:12:12 +0000425 if (likely(data)) {
426 skb = build_skb(data, fragsz);
427 if (unlikely(!skb))
428 put_page(virt_to_head_page(data));
Eric Dumazeta1c7fff2012-05-17 07:34:16 +0000429 }
Eric Dumazeta1c7fff2012-05-17 07:34:16 +0000430 } else {
Mel Gormanc93bdd02012-07-31 16:44:19 -0700431 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask,
432 SKB_ALLOC_RX, NUMA_NO_NODE);
Eric Dumazeta1c7fff2012-05-17 07:34:16 +0000433 }
Christoph Hellwig7b2e4972006-08-07 16:09:04 -0700434 if (likely(skb)) {
Christoph Hellwig8af27452006-07-31 22:35:23 -0700435 skb_reserve(skb, NET_SKB_PAD);
Christoph Hellwig7b2e4972006-08-07 16:09:04 -0700436 skb->dev = dev;
437 }
Christoph Hellwig8af27452006-07-31 22:35:23 -0700438 return skb;
439}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800440EXPORT_SYMBOL(__netdev_alloc_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441
Peter Zijlstra654bed12008-10-07 14:22:33 -0700442void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
Eric Dumazet50269e12012-03-23 23:59:33 +0000443 int size, unsigned int truesize)
Peter Zijlstra654bed12008-10-07 14:22:33 -0700444{
445 skb_fill_page_desc(skb, i, page, off, size);
446 skb->len += size;
447 skb->data_len += size;
Eric Dumazet50269e12012-03-23 23:59:33 +0000448 skb->truesize += truesize;
Peter Zijlstra654bed12008-10-07 14:22:33 -0700449}
450EXPORT_SYMBOL(skb_add_rx_frag);
451
Jason Wangf8e617e2013-11-01 14:07:47 +0800452void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
453 unsigned int truesize)
454{
455 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
456
457 skb_frag_size_add(frag, size);
458 skb->len += size;
459 skb->data_len += size;
460 skb->truesize += truesize;
461}
462EXPORT_SYMBOL(skb_coalesce_rx_frag);
463
Herbert Xu27b437c2006-07-13 19:26:39 -0700464static void skb_drop_list(struct sk_buff **listp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465{
Eric Dumazetbd8a7032013-06-24 06:26:00 -0700466 kfree_skb_list(*listp);
Herbert Xu27b437c2006-07-13 19:26:39 -0700467 *listp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468}
469
Herbert Xu27b437c2006-07-13 19:26:39 -0700470static inline void skb_drop_fraglist(struct sk_buff *skb)
471{
472 skb_drop_list(&skb_shinfo(skb)->frag_list);
473}
474
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475static void skb_clone_fraglist(struct sk_buff *skb)
476{
477 struct sk_buff *list;
478
David S. Millerfbb398a2009-06-09 00:18:59 -0700479 skb_walk_frags(skb, list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 skb_get(list);
481}
482
Eric Dumazetd3836f22012-04-27 00:33:38 +0000483static void skb_free_head(struct sk_buff *skb)
484{
485 if (skb->head_frag)
486 put_page(virt_to_head_page(skb->head));
487 else
488 kfree(skb->head);
489}
490
Adrian Bunk5bba1712006-06-29 13:02:35 -0700491static void skb_release_data(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492{
493 if (!skb->cloned ||
494 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
495 &skb_shinfo(skb)->dataref)) {
496 if (skb_shinfo(skb)->nr_frags) {
497 int i;
498 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
Ian Campbellea2ab692011-08-22 23:44:58 +0000499 skb_frag_unref(skb, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 }
501
Shirley Maa6686f22011-07-06 12:22:12 +0000502 /*
503 * If skb buf is from userspace, we need to notify the caller
504 * the lower device DMA has done;
505 */
506 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
507 struct ubuf_info *uarg;
508
509 uarg = skb_shinfo(skb)->destructor_arg;
510 if (uarg->callback)
Michael S. Tsirkine19d6762012-11-01 09:16:22 +0000511 uarg->callback(uarg, true);
Shirley Maa6686f22011-07-06 12:22:12 +0000512 }
513
David S. Miller21dc3302010-08-23 00:13:46 -0700514 if (skb_has_frag_list(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 skb_drop_fraglist(skb);
516
Eric Dumazetd3836f22012-04-27 00:33:38 +0000517 skb_free_head(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 }
519}
520
521/*
522 * Free an skbuff by memory without cleaning the state.
523 */
Herbert Xu2d4baff2007-11-26 23:11:19 +0800524static void kfree_skbmem(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525{
David S. Millerd179cd12005-08-17 14:57:30 -0700526 struct sk_buff *other;
527 atomic_t *fclone_ref;
528
David S. Millerd179cd12005-08-17 14:57:30 -0700529 switch (skb->fclone) {
530 case SKB_FCLONE_UNAVAILABLE:
531 kmem_cache_free(skbuff_head_cache, skb);
532 break;
533
534 case SKB_FCLONE_ORIG:
535 fclone_ref = (atomic_t *) (skb + 2);
536 if (atomic_dec_and_test(fclone_ref))
537 kmem_cache_free(skbuff_fclone_cache, skb);
538 break;
539
540 case SKB_FCLONE_CLONE:
541 fclone_ref = (atomic_t *) (skb + 1);
542 other = skb - 1;
543
544 /* The clone portion is available for
545 * fast-cloning again.
546 */
547 skb->fclone = SKB_FCLONE_UNAVAILABLE;
548
549 if (atomic_dec_and_test(fclone_ref))
550 kmem_cache_free(skbuff_fclone_cache, other);
551 break;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -0700552 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553}
554
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700555static void skb_release_head_state(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556{
Eric Dumazetadf30902009-06-02 05:19:30 +0000557 skb_dst_drop(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558#ifdef CONFIG_XFRM
559 secpath_put(skb->sp);
560#endif
Stephen Hemminger9c2b3322005-04-19 22:39:42 -0700561 if (skb->destructor) {
562 WARN_ON(in_irq());
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 skb->destructor(skb);
564 }
Igor Maravića3bf7ae2011-12-12 02:58:22 +0000565#if IS_ENABLED(CONFIG_NF_CONNTRACK)
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -0700566 nf_conntrack_put(skb->nfct);
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +0100567#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568#ifdef CONFIG_BRIDGE_NETFILTER
569 nf_bridge_put(skb->nf_bridge);
570#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571/* XXX: IS this still necessary? - JHS */
572#ifdef CONFIG_NET_SCHED
573 skb->tc_index = 0;
574#ifdef CONFIG_NET_CLS_ACT
575 skb->tc_verd = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576#endif
577#endif
Lennert Buytenhek04a4bb52008-10-01 02:33:12 -0700578}
579
580/* Free everything but the sk_buff shell. */
581static void skb_release_all(struct sk_buff *skb)
582{
583 skb_release_head_state(skb);
Pablo Neira5e71d9d2013-06-03 09:28:43 +0000584 if (likely(skb->head))
Patrick McHardy0ebd0ac2013-04-17 06:46:58 +0000585 skb_release_data(skb);
Herbert Xu2d4baff2007-11-26 23:11:19 +0800586}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
Herbert Xu2d4baff2007-11-26 23:11:19 +0800588/**
589 * __kfree_skb - private function
590 * @skb: buffer
591 *
592 * Free an sk_buff. Release anything attached to the buffer.
593 * Clean the state. This is an internal helper function. Users should
594 * always call kfree_skb
595 */
596
597void __kfree_skb(struct sk_buff *skb)
598{
599 skb_release_all(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 kfree_skbmem(skb);
601}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800602EXPORT_SYMBOL(__kfree_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603
604/**
Jörn Engel231d06a2006-03-20 21:28:35 -0800605 * kfree_skb - free an sk_buff
606 * @skb: buffer to free
607 *
608 * Drop a reference to the buffer and free it if the usage count has
609 * hit zero.
610 */
611void kfree_skb(struct sk_buff *skb)
612{
613 if (unlikely(!skb))
614 return;
615 if (likely(atomic_read(&skb->users) == 1))
616 smp_rmb();
617 else if (likely(!atomic_dec_and_test(&skb->users)))
618 return;
Neil Hormanead2ceb2009-03-11 09:49:55 +0000619 trace_kfree_skb(skb, __builtin_return_address(0));
Jörn Engel231d06a2006-03-20 21:28:35 -0800620 __kfree_skb(skb);
621}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800622EXPORT_SYMBOL(kfree_skb);
Jörn Engel231d06a2006-03-20 21:28:35 -0800623
Eric Dumazetbd8a7032013-06-24 06:26:00 -0700624void kfree_skb_list(struct sk_buff *segs)
625{
626 while (segs) {
627 struct sk_buff *next = segs->next;
628
629 kfree_skb(segs);
630 segs = next;
631 }
632}
633EXPORT_SYMBOL(kfree_skb_list);
634
Stephen Hemmingerd1a203e2008-11-01 21:01:09 -0700635/**
Michael S. Tsirkin25121172012-11-01 09:16:28 +0000636 * skb_tx_error - report an sk_buff xmit error
637 * @skb: buffer that triggered an error
638 *
639 * Report xmit error if a device callback is tracking this skb.
640 * skb must be freed afterwards.
641 */
642void skb_tx_error(struct sk_buff *skb)
643{
644 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
645 struct ubuf_info *uarg;
646
647 uarg = skb_shinfo(skb)->destructor_arg;
648 if (uarg->callback)
649 uarg->callback(uarg, false);
650 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
651 }
652}
653EXPORT_SYMBOL(skb_tx_error);
654
655/**
Neil Hormanead2ceb2009-03-11 09:49:55 +0000656 * consume_skb - free an skbuff
657 * @skb: buffer to free
658 *
659 * Drop a ref to the buffer and free it if the usage count has hit zero
660 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
661 * is being dropped after a failure and notes that
662 */
663void consume_skb(struct sk_buff *skb)
664{
665 if (unlikely(!skb))
666 return;
667 if (likely(atomic_read(&skb->users) == 1))
668 smp_rmb();
669 else if (likely(!atomic_dec_and_test(&skb->users)))
670 return;
Koki Sanagi07dc22e2010-08-23 18:46:12 +0900671 trace_consume_skb(skb);
Neil Hormanead2ceb2009-03-11 09:49:55 +0000672 __kfree_skb(skb);
673}
674EXPORT_SYMBOL(consume_skb);
675
Herbert Xudec18812007-10-14 00:37:30 -0700676static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
677{
678 new->tstamp = old->tstamp;
679 new->dev = old->dev;
680 new->transport_header = old->transport_header;
681 new->network_header = old->network_header;
682 new->mac_header = old->mac_header;
Joe Stringer4bc41b82013-07-03 16:04:25 +0900683 new->inner_protocol = old->inner_protocol;
Joseph Gasparakis6a674e92012-12-07 14:14:14 +0000684 new->inner_transport_header = old->inner_transport_header;
Pravin B Shelar92df9b22013-02-01 15:18:49 +0000685 new->inner_network_header = old->inner_network_header;
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +0000686 new->inner_mac_header = old->inner_mac_header;
Eric Dumazet7fee2262010-05-11 23:19:48 +0000687 skb_dst_copy(new, old);
Tom Herbert3df7a742013-12-15 22:16:29 -0800688 skb_copy_hash(new, old);
Changli Gao6461be32011-08-19 04:44:18 +0000689 new->ooo_okay = old->ooo_okay;
Ben Greear3bdc0eb2012-02-11 15:39:30 +0000690 new->no_fcs = old->no_fcs;
Joseph Gasparakis6a674e92012-12-07 14:14:14 +0000691 new->encapsulation = old->encapsulation;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -0700692#ifdef CONFIG_XFRM
Herbert Xudec18812007-10-14 00:37:30 -0700693 new->sp = secpath_get(old->sp);
694#endif
695 memcpy(new->cb, old->cb, sizeof(old->cb));
Herbert Xu9bcb97c2009-05-22 22:20:02 +0000696 new->csum = old->csum;
Herbert Xudec18812007-10-14 00:37:30 -0700697 new->local_df = old->local_df;
698 new->pkt_type = old->pkt_type;
699 new->ip_summed = old->ip_summed;
700 skb_copy_queue_mapping(new, old);
701 new->priority = old->priority;
Igor Maravića3bf7ae2011-12-12 02:58:22 +0000702#if IS_ENABLED(CONFIG_IP_VS)
Herbert Xudec18812007-10-14 00:37:30 -0700703 new->ipvs_property = old->ipvs_property;
704#endif
Mel Gormanc93bdd02012-07-31 16:44:19 -0700705 new->pfmemalloc = old->pfmemalloc;
Herbert Xudec18812007-10-14 00:37:30 -0700706 new->protocol = old->protocol;
707 new->mark = old->mark;
Eric Dumazet8964be42009-11-20 15:35:04 -0800708 new->skb_iif = old->skb_iif;
Herbert Xudec18812007-10-14 00:37:30 -0700709 __nf_copy(new, old);
Herbert Xudec18812007-10-14 00:37:30 -0700710#ifdef CONFIG_NET_SCHED
711 new->tc_index = old->tc_index;
712#ifdef CONFIG_NET_CLS_ACT
713 new->tc_verd = old->tc_verd;
714#endif
715#endif
Patrick McHardy86a9bad2013-04-19 02:04:30 +0000716 new->vlan_proto = old->vlan_proto;
Patrick McHardy6aa895b02008-07-14 22:49:06 -0700717 new->vlan_tci = old->vlan_tci;
718
Herbert Xudec18812007-10-14 00:37:30 -0700719 skb_copy_secmark(new, old);
Eliezer Tamir06021292013-06-10 11:39:50 +0300720
Cong Wange0d10952013-08-01 11:10:25 +0800721#ifdef CONFIG_NET_RX_BUSY_POLL
Eliezer Tamir06021292013-06-10 11:39:50 +0300722 new->napi_id = old->napi_id;
723#endif
Herbert Xudec18812007-10-14 00:37:30 -0700724}
725
Herbert Xu82c49a32009-05-22 22:11:37 +0000726/*
727 * You should not add any new code to this function. Add it to
728 * __copy_skb_header above instead.
729 */
Herbert Xue0053ec2007-10-14 00:37:52 -0700730static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732#define C(x) n->x = skb->x
733
734 n->next = n->prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 n->sk = NULL;
Herbert Xudec18812007-10-14 00:37:30 -0700736 __copy_skb_header(n, skb);
737
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 C(len);
739 C(data_len);
Alexey Dobriyan3e6b3b22007-03-16 15:00:46 -0700740 C(mac_len);
Patrick McHardy334a8132007-06-25 04:35:20 -0700741 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
Paul Moore02f1c892008-01-07 21:56:41 -0800742 n->cloned = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 n->nohdr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 n->destructor = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 C(tail);
746 C(end);
Paul Moore02f1c892008-01-07 21:56:41 -0800747 C(head);
Eric Dumazetd3836f22012-04-27 00:33:38 +0000748 C(head_frag);
Paul Moore02f1c892008-01-07 21:56:41 -0800749 C(data);
750 C(truesize);
751 atomic_set(&n->users, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752
753 atomic_inc(&(skb_shinfo(skb)->dataref));
754 skb->cloned = 1;
755
756 return n;
Herbert Xue0053ec2007-10-14 00:37:52 -0700757#undef C
758}
759
760/**
761 * skb_morph - morph one skb into another
762 * @dst: the skb to receive the contents
763 * @src: the skb to supply the contents
764 *
765 * This is identical to skb_clone except that the target skb is
766 * supplied by the user.
767 *
768 * The target skb is returned upon exit.
769 */
770struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
771{
Herbert Xu2d4baff2007-11-26 23:11:19 +0800772 skb_release_all(dst);
Herbert Xue0053ec2007-10-14 00:37:52 -0700773 return __skb_clone(dst, src);
774}
775EXPORT_SYMBOL_GPL(skb_morph);
776
Ben Hutchings2c530402012-07-10 10:55:09 +0000777/**
778 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
Michael S. Tsirkin48c83012011-08-31 08:03:29 +0000779 * @skb: the skb to modify
780 * @gfp_mask: allocation priority
781 *
782 * This must be called on SKBTX_DEV_ZEROCOPY skb.
783 * It will copy all frags into kernel and drop the reference
784 * to userspace pages.
785 *
786 * If this function is called from an interrupt gfp_mask() must be
787 * %GFP_ATOMIC.
788 *
789 * Returns 0 on success or a negative error code on failure
790 * to allocate kernel memory to copy to.
791 */
792int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
Shirley Maa6686f22011-07-06 12:22:12 +0000793{
794 int i;
795 int num_frags = skb_shinfo(skb)->nr_frags;
796 struct page *page, *head = NULL;
797 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
798
799 for (i = 0; i < num_frags; i++) {
800 u8 *vaddr;
801 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
802
Krishna Kumar02756ed2012-07-17 02:05:29 +0000803 page = alloc_page(gfp_mask);
Shirley Maa6686f22011-07-06 12:22:12 +0000804 if (!page) {
805 while (head) {
Sunghan Suh40dadff2013-07-12 16:17:23 +0900806 struct page *next = (struct page *)page_private(head);
Shirley Maa6686f22011-07-06 12:22:12 +0000807 put_page(head);
808 head = next;
809 }
810 return -ENOMEM;
811 }
Eric Dumazet51c56b02012-04-05 11:35:15 +0200812 vaddr = kmap_atomic(skb_frag_page(f));
Shirley Maa6686f22011-07-06 12:22:12 +0000813 memcpy(page_address(page),
Eric Dumazet9e903e02011-10-18 21:00:24 +0000814 vaddr + f->page_offset, skb_frag_size(f));
Eric Dumazet51c56b02012-04-05 11:35:15 +0200815 kunmap_atomic(vaddr);
Sunghan Suh40dadff2013-07-12 16:17:23 +0900816 set_page_private(page, (unsigned long)head);
Shirley Maa6686f22011-07-06 12:22:12 +0000817 head = page;
818 }
819
820 /* skb frags release userspace buffers */
Krishna Kumar02756ed2012-07-17 02:05:29 +0000821 for (i = 0; i < num_frags; i++)
Ian Campbella8605c62011-10-19 23:01:49 +0000822 skb_frag_unref(skb, i);
Shirley Maa6686f22011-07-06 12:22:12 +0000823
Michael S. Tsirkine19d6762012-11-01 09:16:22 +0000824 uarg->callback(uarg, false);
Shirley Maa6686f22011-07-06 12:22:12 +0000825
826 /* skb frags point to kernel buffers */
Krishna Kumar02756ed2012-07-17 02:05:29 +0000827 for (i = num_frags - 1; i >= 0; i--) {
828 __skb_fill_page_desc(skb, i, head, 0,
829 skb_shinfo(skb)->frags[i].size);
Sunghan Suh40dadff2013-07-12 16:17:23 +0900830 head = (struct page *)page_private(head);
Shirley Maa6686f22011-07-06 12:22:12 +0000831 }
Michael S. Tsirkin48c83012011-08-31 08:03:29 +0000832
833 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
Shirley Maa6686f22011-07-06 12:22:12 +0000834 return 0;
835}
Michael S. Tsirkindcc0fb72012-07-20 09:23:20 +0000836EXPORT_SYMBOL_GPL(skb_copy_ubufs);
Shirley Maa6686f22011-07-06 12:22:12 +0000837
Herbert Xue0053ec2007-10-14 00:37:52 -0700838/**
839 * skb_clone - duplicate an sk_buff
840 * @skb: buffer to clone
841 * @gfp_mask: allocation priority
842 *
843 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
844 * copies share the same packet data but not structure. The new
845 * buffer has a reference count of 1. If the allocation fails the
846 * function returns %NULL otherwise the new buffer is returned.
847 *
848 * If this function is called from an interrupt gfp_mask() must be
849 * %GFP_ATOMIC.
850 */
851
852struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
853{
854 struct sk_buff *n;
855
Michael S. Tsirkin70008aa2012-07-20 09:23:10 +0000856 if (skb_orphan_frags(skb, gfp_mask))
857 return NULL;
Shirley Maa6686f22011-07-06 12:22:12 +0000858
Herbert Xue0053ec2007-10-14 00:37:52 -0700859 n = skb + 1;
860 if (skb->fclone == SKB_FCLONE_ORIG &&
861 n->fclone == SKB_FCLONE_UNAVAILABLE) {
862 atomic_t *fclone_ref = (atomic_t *) (n + 1);
863 n->fclone = SKB_FCLONE_CLONE;
864 atomic_inc(fclone_ref);
865 } else {
Mel Gormanc93bdd02012-07-31 16:44:19 -0700866 if (skb_pfmemalloc(skb))
867 gfp_mask |= __GFP_MEMALLOC;
868
Herbert Xue0053ec2007-10-14 00:37:52 -0700869 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
870 if (!n)
871 return NULL;
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200872
873 kmemcheck_annotate_bitfield(n, flags1);
874 kmemcheck_annotate_bitfield(n, flags2);
Herbert Xue0053ec2007-10-14 00:37:52 -0700875 n->fclone = SKB_FCLONE_UNAVAILABLE;
876 }
877
878 return __skb_clone(n, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800880EXPORT_SYMBOL(skb_clone);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881
Pravin B Shelarf5b17292013-03-07 13:21:40 +0000882static void skb_headers_offset_update(struct sk_buff *skb, int off)
883{
Eric Dumazet030737b2013-10-19 11:42:54 -0700884 /* Only adjust this if it actually is csum_start rather than csum */
885 if (skb->ip_summed == CHECKSUM_PARTIAL)
886 skb->csum_start += off;
Pravin B Shelarf5b17292013-03-07 13:21:40 +0000887 /* {transport,network,mac}_header and tail are relative to skb->head */
888 skb->transport_header += off;
889 skb->network_header += off;
890 if (skb_mac_header_was_set(skb))
891 skb->mac_header += off;
892 skb->inner_transport_header += off;
893 skb->inner_network_header += off;
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +0000894 skb->inner_mac_header += off;
Pravin B Shelarf5b17292013-03-07 13:21:40 +0000895}
896
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
898{
Herbert Xudec18812007-10-14 00:37:30 -0700899 __copy_skb_header(new, old);
900
Herbert Xu79671682006-06-22 02:40:14 -0700901 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
902 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
903 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904}
905
Mel Gormanc93bdd02012-07-31 16:44:19 -0700906static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
907{
908 if (skb_pfmemalloc(skb))
909 return SKB_ALLOC_RX;
910 return 0;
911}
912
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913/**
914 * skb_copy - create private copy of an sk_buff
915 * @skb: buffer to copy
916 * @gfp_mask: allocation priority
917 *
918 * Make a copy of both an &sk_buff and its data. This is used when the
919 * caller wishes to modify the data and needs a private copy of the
920 * data to alter. Returns %NULL on failure or the pointer to the buffer
921 * on success. The returned buffer has a reference count of 1.
922 *
923 * As by-product this function converts non-linear &sk_buff to linear
924 * one, so that &sk_buff becomes completely private and caller is allowed
925 * to modify all the data of returned buffer. This means that this
926 * function is not recommended for use in circumstances when only
927 * header is going to be modified. Use pskb_copy() instead.
928 */
929
Al Virodd0fc662005-10-07 07:46:04 +0100930struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931{
Eric Dumazet6602ceb2010-09-01 05:25:10 +0000932 int headerlen = skb_headroom(skb);
Alexander Duyckec47ea82012-05-04 14:26:56 +0000933 unsigned int size = skb_end_offset(skb) + skb->data_len;
Mel Gormanc93bdd02012-07-31 16:44:19 -0700934 struct sk_buff *n = __alloc_skb(size, gfp_mask,
935 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
Eric Dumazet6602ceb2010-09-01 05:25:10 +0000936
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 if (!n)
938 return NULL;
939
940 /* Set the data pointer */
941 skb_reserve(n, headerlen);
942 /* Set the tail pointer and length */
943 skb_put(n, skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944
945 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
946 BUG();
947
948 copy_skb_header(n, skb);
949 return n;
950}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800951EXPORT_SYMBOL(skb_copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952
953/**
Eric Dumazet117632e2011-12-03 21:39:53 +0000954 * __pskb_copy - create copy of an sk_buff with private head.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 * @skb: buffer to copy
Eric Dumazet117632e2011-12-03 21:39:53 +0000956 * @headroom: headroom of new skb
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 * @gfp_mask: allocation priority
958 *
959 * Make a copy of both an &sk_buff and part of its data, located
960 * in header. Fragmented data remain shared. This is used when
961 * the caller wishes to modify only header of &sk_buff and needs
962 * private copy of the header to alter. Returns %NULL on failure
963 * or the pointer to the buffer on success.
964 * The returned buffer has a reference count of 1.
965 */
966
Eric Dumazet117632e2011-12-03 21:39:53 +0000967struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968{
Eric Dumazet117632e2011-12-03 21:39:53 +0000969 unsigned int size = skb_headlen(skb) + headroom;
Mel Gormanc93bdd02012-07-31 16:44:19 -0700970 struct sk_buff *n = __alloc_skb(size, gfp_mask,
971 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
Eric Dumazet6602ceb2010-09-01 05:25:10 +0000972
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 if (!n)
974 goto out;
975
976 /* Set the data pointer */
Eric Dumazet117632e2011-12-03 21:39:53 +0000977 skb_reserve(n, headroom);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 /* Set the tail pointer and length */
979 skb_put(n, skb_headlen(skb));
980 /* Copy the bytes */
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -0300981 skb_copy_from_linear_data(skb, n->data, n->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982
Herbert Xu25f484a2006-11-07 14:57:15 -0800983 n->truesize += skb->data_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 n->data_len = skb->data_len;
985 n->len = skb->len;
986
987 if (skb_shinfo(skb)->nr_frags) {
988 int i;
989
Michael S. Tsirkin70008aa2012-07-20 09:23:10 +0000990 if (skb_orphan_frags(skb, gfp_mask)) {
991 kfree_skb(n);
992 n = NULL;
993 goto out;
Shirley Maa6686f22011-07-06 12:22:12 +0000994 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
996 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
Ian Campbellea2ab692011-08-22 23:44:58 +0000997 skb_frag_ref(skb, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 }
999 skb_shinfo(n)->nr_frags = i;
1000 }
1001
David S. Miller21dc3302010-08-23 00:13:46 -07001002 if (skb_has_frag_list(skb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1004 skb_clone_fraglist(n);
1005 }
1006
1007 copy_skb_header(n, skb);
1008out:
1009 return n;
1010}
Eric Dumazet117632e2011-12-03 21:39:53 +00001011EXPORT_SYMBOL(__pskb_copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012
1013/**
1014 * pskb_expand_head - reallocate header of &sk_buff
1015 * @skb: buffer to reallocate
1016 * @nhead: room to add at head
1017 * @ntail: room to add at tail
1018 * @gfp_mask: allocation priority
1019 *
Mathias Krausebc323832013-11-07 14:18:26 +01001020 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1021 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 * reference count of 1. Returns zero in the case of success or error,
1023 * if expansion failed. In the last case, &sk_buff is not changed.
1024 *
1025 * All the pointers pointing into skb header may change and must be
1026 * reloaded after call to this function.
1027 */
1028
Victor Fusco86a76ca2005-07-08 14:57:47 -07001029int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
Al Virodd0fc662005-10-07 07:46:04 +01001030 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031{
1032 int i;
1033 u8 *data;
Alexander Duyckec47ea82012-05-04 14:26:56 +00001034 int size = nhead + skb_end_offset(skb) + ntail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 long off;
1036
Herbert Xu4edd87a2008-10-01 07:09:38 -07001037 BUG_ON(nhead < 0);
1038
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 if (skb_shared(skb))
1040 BUG();
1041
1042 size = SKB_DATA_ALIGN(size);
1043
Mel Gormanc93bdd02012-07-31 16:44:19 -07001044 if (skb_pfmemalloc(skb))
1045 gfp_mask |= __GFP_MEMALLOC;
1046 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1047 gfp_mask, NUMA_NO_NODE, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 if (!data)
1049 goto nodata;
Eric Dumazet87151b82012-04-10 20:08:39 +00001050 size = SKB_WITH_OVERHEAD(ksize(data));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051
1052 /* Copy only real data... and, alas, header. This should be
Eric Dumazet6602ceb2010-09-01 05:25:10 +00001053 * optimized for the cases when header is void.
1054 */
1055 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1056
1057 memcpy((struct skb_shared_info *)(data + size),
1058 skb_shinfo(skb),
Eric Dumazetfed66382010-07-22 19:09:08 +00001059 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060
Alexander Duyck3e245912012-05-04 14:26:51 +00001061 /*
1062 * if shinfo is shared we must drop the old head gracefully, but if it
1063 * is not we can just drop the old head and let the existing refcount
1064 * be since all we did is relocate the values
1065 */
1066 if (skb_cloned(skb)) {
Shirley Maa6686f22011-07-06 12:22:12 +00001067 /* copy this zero copy skb frags */
Michael S. Tsirkin70008aa2012-07-20 09:23:10 +00001068 if (skb_orphan_frags(skb, gfp_mask))
1069 goto nofrags;
Eric Dumazet1fd63042010-09-02 23:09:32 +00001070 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
Ian Campbellea2ab692011-08-22 23:44:58 +00001071 skb_frag_ref(skb, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072
Eric Dumazet1fd63042010-09-02 23:09:32 +00001073 if (skb_has_frag_list(skb))
1074 skb_clone_fraglist(skb);
1075
1076 skb_release_data(skb);
Alexander Duyck3e245912012-05-04 14:26:51 +00001077 } else {
1078 skb_free_head(skb);
Eric Dumazet1fd63042010-09-02 23:09:32 +00001079 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 off = (data + nhead) - skb->head;
1081
1082 skb->head = data;
Eric Dumazetd3836f22012-04-27 00:33:38 +00001083 skb->head_frag = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 skb->data += off;
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001085#ifdef NET_SKBUFF_DATA_USES_OFFSET
1086 skb->end = size;
Patrick McHardy56eb8882007-04-09 11:45:04 -07001087 off = nhead;
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001088#else
1089 skb->end = skb->head + size;
Patrick McHardy56eb8882007-04-09 11:45:04 -07001090#endif
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001091 skb->tail += off;
Peter Pan(潘卫平)b41abb42013-06-06 21:27:21 +08001092 skb_headers_offset_update(skb, nhead);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 skb->cloned = 0;
Patrick McHardy334a8132007-06-25 04:35:20 -07001094 skb->hdr_len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 skb->nohdr = 0;
1096 atomic_set(&skb_shinfo(skb)->dataref, 1);
1097 return 0;
1098
Shirley Maa6686f22011-07-06 12:22:12 +00001099nofrags:
1100 kfree(data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101nodata:
1102 return -ENOMEM;
1103}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001104EXPORT_SYMBOL(pskb_expand_head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105
1106/* Make private copy of skb with writable head and some headroom */
1107
1108struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1109{
1110 struct sk_buff *skb2;
1111 int delta = headroom - skb_headroom(skb);
1112
1113 if (delta <= 0)
1114 skb2 = pskb_copy(skb, GFP_ATOMIC);
1115 else {
1116 skb2 = skb_clone(skb, GFP_ATOMIC);
1117 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1118 GFP_ATOMIC)) {
1119 kfree_skb(skb2);
1120 skb2 = NULL;
1121 }
1122 }
1123 return skb2;
1124}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001125EXPORT_SYMBOL(skb_realloc_headroom);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126
1127/**
1128 * skb_copy_expand - copy and expand sk_buff
1129 * @skb: buffer to copy
1130 * @newheadroom: new free bytes at head
1131 * @newtailroom: new free bytes at tail
1132 * @gfp_mask: allocation priority
1133 *
1134 * Make a copy of both an &sk_buff and its data and while doing so
1135 * allocate additional space.
1136 *
1137 * This is used when the caller wishes to modify the data and needs a
1138 * private copy of the data to alter as well as more space for new fields.
1139 * Returns %NULL on failure or the pointer to the buffer
1140 * on success. The returned buffer has a reference count of 1.
1141 *
1142 * You must pass %GFP_ATOMIC as the allocation priority if this function
1143 * is called from an interrupt.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 */
1145struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
Victor Fusco86a76ca2005-07-08 14:57:47 -07001146 int newheadroom, int newtailroom,
Al Virodd0fc662005-10-07 07:46:04 +01001147 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148{
1149 /*
1150 * Allocate the copy buffer
1151 */
Mel Gormanc93bdd02012-07-31 16:44:19 -07001152 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1153 gfp_mask, skb_alloc_rx_flag(skb),
1154 NUMA_NO_NODE);
Patrick McHardyefd1e8d2007-04-10 18:30:09 -07001155 int oldheadroom = skb_headroom(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 int head_copy_len, head_copy_off;
1157
1158 if (!n)
1159 return NULL;
1160
1161 skb_reserve(n, newheadroom);
1162
1163 /* Set the tail pointer and length */
1164 skb_put(n, skb->len);
1165
Patrick McHardyefd1e8d2007-04-10 18:30:09 -07001166 head_copy_len = oldheadroom;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 head_copy_off = 0;
1168 if (newheadroom <= head_copy_len)
1169 head_copy_len = newheadroom;
1170 else
1171 head_copy_off = newheadroom - head_copy_len;
1172
1173 /* Copy the linear header and data. */
1174 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1175 skb->len + head_copy_len))
1176 BUG();
1177
1178 copy_skb_header(n, skb);
1179
Eric Dumazet030737b2013-10-19 11:42:54 -07001180 skb_headers_offset_update(n, newheadroom - oldheadroom);
Patrick McHardyefd1e8d2007-04-10 18:30:09 -07001181
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 return n;
1183}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001184EXPORT_SYMBOL(skb_copy_expand);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185
1186/**
1187 * skb_pad - zero pad the tail of an skb
1188 * @skb: buffer to pad
1189 * @pad: space to pad
1190 *
1191 * Ensure that a buffer is followed by a padding area that is zero
1192 * filled. Used by network drivers which may DMA or transfer data
1193 * beyond the buffer end onto the wire.
1194 *
Herbert Xu5b057c62006-06-23 02:06:41 -07001195 * May return error in out of memory cases. The skb is freed on error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 */
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001197
Herbert Xu5b057c62006-06-23 02:06:41 -07001198int skb_pad(struct sk_buff *skb, int pad)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199{
Herbert Xu5b057c62006-06-23 02:06:41 -07001200 int err;
1201 int ntail;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001202
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 /* If the skbuff is non linear tailroom is always zero.. */
Herbert Xu5b057c62006-06-23 02:06:41 -07001204 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 memset(skb->data+skb->len, 0, pad);
Herbert Xu5b057c62006-06-23 02:06:41 -07001206 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 }
Herbert Xu5b057c62006-06-23 02:06:41 -07001208
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001209 ntail = skb->data_len + pad - (skb->end - skb->tail);
Herbert Xu5b057c62006-06-23 02:06:41 -07001210 if (likely(skb_cloned(skb) || ntail > 0)) {
1211 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1212 if (unlikely(err))
1213 goto free_skb;
1214 }
1215
1216 /* FIXME: The use of this function with non-linear skb's really needs
1217 * to be audited.
1218 */
1219 err = skb_linearize(skb);
1220 if (unlikely(err))
1221 goto free_skb;
1222
1223 memset(skb->data + skb->len, 0, pad);
1224 return 0;
1225
1226free_skb:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 kfree_skb(skb);
Herbert Xu5b057c62006-06-23 02:06:41 -07001228 return err;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001229}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001230EXPORT_SYMBOL(skb_pad);
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001231
Ilpo Järvinen0dde3e12008-03-27 17:43:41 -07001232/**
Mathias Krause0c7ddf32013-11-07 14:18:24 +01001233 * pskb_put - add data to the tail of a potentially fragmented buffer
1234 * @skb: start of the buffer to use
1235 * @tail: tail fragment of the buffer to use
1236 * @len: amount of data to add
1237 *
1238 * This function extends the used data area of the potentially
1239 * fragmented buffer. @tail must be the last fragment of @skb -- or
1240 * @skb itself. If this would exceed the total buffer size the kernel
1241 * will panic. A pointer to the first byte of the extra data is
1242 * returned.
1243 */
1244
1245unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1246{
1247 if (tail != skb) {
1248 skb->data_len += len;
1249 skb->len += len;
1250 }
1251 return skb_put(tail, len);
1252}
1253EXPORT_SYMBOL_GPL(pskb_put);
1254
1255/**
Ilpo Järvinen0dde3e12008-03-27 17:43:41 -07001256 * skb_put - add data to a buffer
1257 * @skb: buffer to use
1258 * @len: amount of data to add
1259 *
1260 * This function extends the used data area of the buffer. If this would
1261 * exceed the total buffer size the kernel will panic. A pointer to the
1262 * first byte of the extra data is returned.
1263 */
1264unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1265{
1266 unsigned char *tmp = skb_tail_pointer(skb);
1267 SKB_LINEAR_ASSERT(skb);
1268 skb->tail += len;
1269 skb->len += len;
1270 if (unlikely(skb->tail > skb->end))
1271 skb_over_panic(skb, len, __builtin_return_address(0));
1272 return tmp;
1273}
1274EXPORT_SYMBOL(skb_put);
1275
Ilpo Järvinen6be8ac22008-03-27 17:47:24 -07001276/**
Ilpo Järvinenc2aa2702008-03-27 17:52:40 -07001277 * skb_push - add data to the start of a buffer
1278 * @skb: buffer to use
1279 * @len: amount of data to add
1280 *
1281 * This function extends the used data area of the buffer at the buffer
1282 * start. If this would exceed the total buffer headroom the kernel will
1283 * panic. A pointer to the first byte of the extra data is returned.
1284 */
1285unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1286{
1287 skb->data -= len;
1288 skb->len += len;
1289 if (unlikely(skb->data<skb->head))
1290 skb_under_panic(skb, len, __builtin_return_address(0));
1291 return skb->data;
1292}
1293EXPORT_SYMBOL(skb_push);
1294
1295/**
Ilpo Järvinen6be8ac22008-03-27 17:47:24 -07001296 * skb_pull - remove data from the start of a buffer
1297 * @skb: buffer to use
1298 * @len: amount of data to remove
1299 *
1300 * This function removes data from the start of a buffer, returning
1301 * the memory to the headroom. A pointer to the next data in the buffer
1302 * is returned. Once the data has been pulled future pushes will overwrite
1303 * the old data.
1304 */
1305unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1306{
David S. Miller47d29642010-05-02 02:21:44 -07001307 return skb_pull_inline(skb, len);
Ilpo Järvinen6be8ac22008-03-27 17:47:24 -07001308}
1309EXPORT_SYMBOL(skb_pull);
1310
Ilpo Järvinen419ae742008-03-27 17:54:01 -07001311/**
1312 * skb_trim - remove end from a buffer
1313 * @skb: buffer to alter
1314 * @len: new length
1315 *
1316 * Cut the length of a buffer down by removing data from the tail. If
1317 * the buffer is already under the length specified it is not modified.
1318 * The skb must be linear.
1319 */
1320void skb_trim(struct sk_buff *skb, unsigned int len)
1321{
1322 if (skb->len > len)
1323 __skb_trim(skb, len);
1324}
1325EXPORT_SYMBOL(skb_trim);
1326
Herbert Xu3cc0e872006-06-09 16:13:38 -07001327/* Trims skb to length len. It can change skb pointers.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 */
1329
Herbert Xu3cc0e872006-06-09 16:13:38 -07001330int ___pskb_trim(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331{
Herbert Xu27b437c2006-07-13 19:26:39 -07001332 struct sk_buff **fragp;
1333 struct sk_buff *frag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 int offset = skb_headlen(skb);
1335 int nfrags = skb_shinfo(skb)->nr_frags;
1336 int i;
Herbert Xu27b437c2006-07-13 19:26:39 -07001337 int err;
1338
1339 if (skb_cloned(skb) &&
1340 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1341 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001343 i = 0;
1344 if (offset >= len)
1345 goto drop_pages;
1346
1347 for (; i < nfrags; i++) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00001348 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
Herbert Xu27b437c2006-07-13 19:26:39 -07001349
1350 if (end < len) {
1351 offset = end;
1352 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 }
Herbert Xu27b437c2006-07-13 19:26:39 -07001354
Eric Dumazet9e903e02011-10-18 21:00:24 +00001355 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
Herbert Xu27b437c2006-07-13 19:26:39 -07001356
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001357drop_pages:
Herbert Xu27b437c2006-07-13 19:26:39 -07001358 skb_shinfo(skb)->nr_frags = i;
1359
1360 for (; i < nfrags; i++)
Ian Campbellea2ab692011-08-22 23:44:58 +00001361 skb_frag_unref(skb, i);
Herbert Xu27b437c2006-07-13 19:26:39 -07001362
David S. Miller21dc3302010-08-23 00:13:46 -07001363 if (skb_has_frag_list(skb))
Herbert Xu27b437c2006-07-13 19:26:39 -07001364 skb_drop_fraglist(skb);
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001365 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 }
1367
Herbert Xu27b437c2006-07-13 19:26:39 -07001368 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1369 fragp = &frag->next) {
1370 int end = offset + frag->len;
1371
1372 if (skb_shared(frag)) {
1373 struct sk_buff *nfrag;
1374
1375 nfrag = skb_clone(frag, GFP_ATOMIC);
1376 if (unlikely(!nfrag))
1377 return -ENOMEM;
1378
1379 nfrag->next = frag->next;
Eric Dumazet85bb2a62012-04-19 02:24:53 +00001380 consume_skb(frag);
Herbert Xu27b437c2006-07-13 19:26:39 -07001381 frag = nfrag;
1382 *fragp = frag;
1383 }
1384
1385 if (end < len) {
1386 offset = end;
1387 continue;
1388 }
1389
1390 if (end > len &&
1391 unlikely((err = pskb_trim(frag, len - offset))))
1392 return err;
1393
1394 if (frag->next)
1395 skb_drop_list(&frag->next);
1396 break;
1397 }
1398
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001399done:
Herbert Xu27b437c2006-07-13 19:26:39 -07001400 if (len > skb_headlen(skb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 skb->data_len -= skb->len - len;
1402 skb->len = len;
1403 } else {
Herbert Xu27b437c2006-07-13 19:26:39 -07001404 skb->len = len;
1405 skb->data_len = 0;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001406 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 }
1408
1409 return 0;
1410}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001411EXPORT_SYMBOL(___pskb_trim);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412
1413/**
1414 * __pskb_pull_tail - advance tail of skb header
1415 * @skb: buffer to reallocate
1416 * @delta: number of bytes to advance tail
1417 *
1418 * The function makes a sense only on a fragmented &sk_buff,
1419 * it expands header moving its tail forward and copying necessary
1420 * data from fragmented part.
1421 *
1422 * &sk_buff MUST have reference count of 1.
1423 *
1424 * Returns %NULL (and &sk_buff does not change) if pull failed
1425 * or value of new tail of skb in the case of success.
1426 *
1427 * All the pointers pointing into skb header may change and must be
1428 * reloaded after call to this function.
1429 */
1430
1431/* Moves tail of skb head forward, copying data from fragmented part,
1432 * when it is necessary.
1433 * 1. It may fail due to malloc failure.
1434 * 2. It may change skb pointers.
1435 *
1436 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1437 */
1438unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1439{
1440 /* If skb has not enough free space at tail, get new one
1441 * plus 128 bytes for future expansions. If we have enough
1442 * room at tail, reallocate without expansion only if skb is cloned.
1443 */
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001444 int i, k, eat = (skb->tail + delta) - skb->end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445
1446 if (eat > 0 || skb_cloned(skb)) {
1447 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1448 GFP_ATOMIC))
1449 return NULL;
1450 }
1451
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001452 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 BUG();
1454
1455 /* Optimization: no fragments, no reasons to preestimate
1456 * size of pulled pages. Superb.
1457 */
David S. Miller21dc3302010-08-23 00:13:46 -07001458 if (!skb_has_frag_list(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 goto pull_pages;
1460
1461 /* Estimate size of pulled pages. */
1462 eat = delta;
1463 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00001464 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1465
1466 if (size >= eat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 goto pull_pages;
Eric Dumazet9e903e02011-10-18 21:00:24 +00001468 eat -= size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 }
1470
1471 /* If we need update frag list, we are in troubles.
1472 * Certainly, it possible to add an offset to skb data,
1473 * but taking into account that pulling is expected to
1474 * be very rare operation, it is worth to fight against
1475 * further bloating skb head and crucify ourselves here instead.
1476 * Pure masohism, indeed. 8)8)
1477 */
1478 if (eat) {
1479 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1480 struct sk_buff *clone = NULL;
1481 struct sk_buff *insp = NULL;
1482
1483 do {
Kris Katterjohn09a62662006-01-08 22:24:28 -08001484 BUG_ON(!list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485
1486 if (list->len <= eat) {
1487 /* Eaten as whole. */
1488 eat -= list->len;
1489 list = list->next;
1490 insp = list;
1491 } else {
1492 /* Eaten partially. */
1493
1494 if (skb_shared(list)) {
1495 /* Sucks! We need to fork list. :-( */
1496 clone = skb_clone(list, GFP_ATOMIC);
1497 if (!clone)
1498 return NULL;
1499 insp = list->next;
1500 list = clone;
1501 } else {
1502 /* This may be pulled without
1503 * problems. */
1504 insp = list;
1505 }
1506 if (!pskb_pull(list, eat)) {
Wei Yongjunf3fbbe02009-02-25 00:37:32 +00001507 kfree_skb(clone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 return NULL;
1509 }
1510 break;
1511 }
1512 } while (eat);
1513
1514 /* Free pulled out fragments. */
1515 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1516 skb_shinfo(skb)->frag_list = list->next;
1517 kfree_skb(list);
1518 }
1519 /* And insert new clone at head. */
1520 if (clone) {
1521 clone->next = list;
1522 skb_shinfo(skb)->frag_list = clone;
1523 }
1524 }
1525 /* Success! Now we may commit changes to skb data. */
1526
1527pull_pages:
1528 eat = delta;
1529 k = 0;
1530 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00001531 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1532
1533 if (size <= eat) {
Ian Campbellea2ab692011-08-22 23:44:58 +00001534 skb_frag_unref(skb, i);
Eric Dumazet9e903e02011-10-18 21:00:24 +00001535 eat -= size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 } else {
1537 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1538 if (eat) {
1539 skb_shinfo(skb)->frags[k].page_offset += eat;
Eric Dumazet9e903e02011-10-18 21:00:24 +00001540 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 eat = 0;
1542 }
1543 k++;
1544 }
1545 }
1546 skb_shinfo(skb)->nr_frags = k;
1547
1548 skb->tail += delta;
1549 skb->data_len -= delta;
1550
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001551 return skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001553EXPORT_SYMBOL(__pskb_pull_tail);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554
Eric Dumazet22019b12011-07-29 18:37:31 +00001555/**
1556 * skb_copy_bits - copy bits from skb to kernel buffer
1557 * @skb: source skb
1558 * @offset: offset in source
1559 * @to: destination buffer
1560 * @len: number of bytes to copy
1561 *
1562 * Copy the specified number of bytes from the source skb to the
1563 * destination buffer.
1564 *
1565 * CAUTION ! :
1566 * If its prototype is ever changed,
1567 * check arch/{*}/net/{*}.S files,
1568 * since it is called from BPF assembly code.
1569 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1571{
David S. Miller1a028e52007-04-27 15:21:23 -07001572 int start = skb_headlen(skb);
David S. Millerfbb398a2009-06-09 00:18:59 -07001573 struct sk_buff *frag_iter;
1574 int i, copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575
1576 if (offset > (int)skb->len - len)
1577 goto fault;
1578
1579 /* Copy header. */
David S. Miller1a028e52007-04-27 15:21:23 -07001580 if ((copy = start - offset) > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 if (copy > len)
1582 copy = len;
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03001583 skb_copy_from_linear_data_offset(skb, offset, to, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 if ((len -= copy) == 0)
1585 return 0;
1586 offset += copy;
1587 to += copy;
1588 }
1589
1590 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07001591 int end;
Eric Dumazet51c56b02012-04-05 11:35:15 +02001592 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001594 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07001595
Eric Dumazet51c56b02012-04-05 11:35:15 +02001596 end = start + skb_frag_size(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 if ((copy = end - offset) > 0) {
1598 u8 *vaddr;
1599
1600 if (copy > len)
1601 copy = len;
1602
Eric Dumazet51c56b02012-04-05 11:35:15 +02001603 vaddr = kmap_atomic(skb_frag_page(f));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 memcpy(to,
Eric Dumazet51c56b02012-04-05 11:35:15 +02001605 vaddr + f->page_offset + offset - start,
1606 copy);
1607 kunmap_atomic(vaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608
1609 if ((len -= copy) == 0)
1610 return 0;
1611 offset += copy;
1612 to += copy;
1613 }
David S. Miller1a028e52007-04-27 15:21:23 -07001614 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 }
1616
David S. Millerfbb398a2009-06-09 00:18:59 -07001617 skb_walk_frags(skb, frag_iter) {
1618 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619
David S. Millerfbb398a2009-06-09 00:18:59 -07001620 WARN_ON(start > offset + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621
David S. Millerfbb398a2009-06-09 00:18:59 -07001622 end = start + frag_iter->len;
1623 if ((copy = end - offset) > 0) {
1624 if (copy > len)
1625 copy = len;
1626 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1627 goto fault;
1628 if ((len -= copy) == 0)
1629 return 0;
1630 offset += copy;
1631 to += copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 }
David S. Millerfbb398a2009-06-09 00:18:59 -07001633 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 }
Shirley Maa6686f22011-07-06 12:22:12 +00001635
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 if (!len)
1637 return 0;
1638
1639fault:
1640 return -EFAULT;
1641}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001642EXPORT_SYMBOL(skb_copy_bits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643
Jens Axboe9c55e012007-11-06 23:30:13 -08001644/*
1645 * Callback from splice_to_pipe(), if we need to release some pages
1646 * at the end of the spd in case we error'ed out in filling the pipe.
1647 */
1648static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1649{
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001650 put_page(spd->pages[i]);
1651}
Jens Axboe9c55e012007-11-06 23:30:13 -08001652
David S. Millera108d5f2012-04-23 23:06:11 -04001653static struct page *linear_to_page(struct page *page, unsigned int *len,
1654 unsigned int *offset,
Eric Dumazet18aafc62013-01-11 14:46:37 +00001655 struct sock *sk)
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001656{
Eric Dumazet5640f762012-09-23 23:04:42 +00001657 struct page_frag *pfrag = sk_page_frag(sk);
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001658
Eric Dumazet5640f762012-09-23 23:04:42 +00001659 if (!sk_page_frag_refill(sk, pfrag))
1660 return NULL;
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001661
Eric Dumazet5640f762012-09-23 23:04:42 +00001662 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001663
Eric Dumazet5640f762012-09-23 23:04:42 +00001664 memcpy(page_address(pfrag->page) + pfrag->offset,
1665 page_address(page) + *offset, *len);
1666 *offset = pfrag->offset;
1667 pfrag->offset += *len;
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001668
Eric Dumazet5640f762012-09-23 23:04:42 +00001669 return pfrag->page;
Jens Axboe9c55e012007-11-06 23:30:13 -08001670}
1671
Eric Dumazet41c73a02012-04-22 12:26:16 +00001672static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1673 struct page *page,
1674 unsigned int offset)
1675{
1676 return spd->nr_pages &&
1677 spd->pages[spd->nr_pages - 1] == page &&
1678 (spd->partial[spd->nr_pages - 1].offset +
1679 spd->partial[spd->nr_pages - 1].len == offset);
1680}
1681
Jens Axboe9c55e012007-11-06 23:30:13 -08001682/*
1683 * Fill page/offset/length into spd, if it can hold more pages.
1684 */
David S. Millera108d5f2012-04-23 23:06:11 -04001685static bool spd_fill_page(struct splice_pipe_desc *spd,
1686 struct pipe_inode_info *pipe, struct page *page,
1687 unsigned int *len, unsigned int offset,
Eric Dumazet18aafc62013-01-11 14:46:37 +00001688 bool linear,
David S. Millera108d5f2012-04-23 23:06:11 -04001689 struct sock *sk)
Jens Axboe9c55e012007-11-06 23:30:13 -08001690{
Eric Dumazet41c73a02012-04-22 12:26:16 +00001691 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
David S. Millera108d5f2012-04-23 23:06:11 -04001692 return true;
Jens Axboe9c55e012007-11-06 23:30:13 -08001693
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001694 if (linear) {
Eric Dumazet18aafc62013-01-11 14:46:37 +00001695 page = linear_to_page(page, len, &offset, sk);
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001696 if (!page)
David S. Millera108d5f2012-04-23 23:06:11 -04001697 return true;
Eric Dumazet41c73a02012-04-22 12:26:16 +00001698 }
1699 if (spd_can_coalesce(spd, page, offset)) {
1700 spd->partial[spd->nr_pages - 1].len += *len;
David S. Millera108d5f2012-04-23 23:06:11 -04001701 return false;
Eric Dumazet41c73a02012-04-22 12:26:16 +00001702 }
1703 get_page(page);
Jens Axboe9c55e012007-11-06 23:30:13 -08001704 spd->pages[spd->nr_pages] = page;
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001705 spd->partial[spd->nr_pages].len = *len;
Jens Axboe9c55e012007-11-06 23:30:13 -08001706 spd->partial[spd->nr_pages].offset = offset;
Jens Axboe9c55e012007-11-06 23:30:13 -08001707 spd->nr_pages++;
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001708
David S. Millera108d5f2012-04-23 23:06:11 -04001709 return false;
Jens Axboe9c55e012007-11-06 23:30:13 -08001710}
1711
David S. Millera108d5f2012-04-23 23:06:11 -04001712static bool __splice_segment(struct page *page, unsigned int poff,
1713 unsigned int plen, unsigned int *off,
Eric Dumazet18aafc62013-01-11 14:46:37 +00001714 unsigned int *len,
Eric Dumazetd7ccf7c2012-04-23 23:35:04 -04001715 struct splice_pipe_desc *spd, bool linear,
David S. Millera108d5f2012-04-23 23:06:11 -04001716 struct sock *sk,
1717 struct pipe_inode_info *pipe)
Octavian Purdila2870c432008-07-15 00:49:11 -07001718{
1719 if (!*len)
David S. Millera108d5f2012-04-23 23:06:11 -04001720 return true;
Octavian Purdila2870c432008-07-15 00:49:11 -07001721
1722 /* skip this segment if already processed */
1723 if (*off >= plen) {
1724 *off -= plen;
David S. Millera108d5f2012-04-23 23:06:11 -04001725 return false;
Octavian Purdiladb43a282008-06-27 17:27:21 -07001726 }
Jens Axboe9c55e012007-11-06 23:30:13 -08001727
Octavian Purdila2870c432008-07-15 00:49:11 -07001728 /* ignore any bits we already processed */
Eric Dumazet9ca1b222013-01-05 21:31:18 +00001729 poff += *off;
1730 plen -= *off;
1731 *off = 0;
Octavian Purdila2870c432008-07-15 00:49:11 -07001732
Eric Dumazet18aafc62013-01-11 14:46:37 +00001733 do {
1734 unsigned int flen = min(*len, plen);
Octavian Purdila2870c432008-07-15 00:49:11 -07001735
Eric Dumazet18aafc62013-01-11 14:46:37 +00001736 if (spd_fill_page(spd, pipe, page, &flen, poff,
1737 linear, sk))
1738 return true;
1739 poff += flen;
1740 plen -= flen;
1741 *len -= flen;
1742 } while (*len && plen);
Octavian Purdila2870c432008-07-15 00:49:11 -07001743
David S. Millera108d5f2012-04-23 23:06:11 -04001744 return false;
Octavian Purdila2870c432008-07-15 00:49:11 -07001745}
1746
1747/*
David S. Millera108d5f2012-04-23 23:06:11 -04001748 * Map linear and fragment data from the skb to spd. It reports true if the
Octavian Purdila2870c432008-07-15 00:49:11 -07001749 * pipe is full or if we already spliced the requested length.
1750 */
David S. Millera108d5f2012-04-23 23:06:11 -04001751static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1752 unsigned int *offset, unsigned int *len,
1753 struct splice_pipe_desc *spd, struct sock *sk)
Octavian Purdila2870c432008-07-15 00:49:11 -07001754{
1755 int seg;
1756
Eric Dumazet1d0c0b32012-04-27 02:10:03 +00001757 /* map the linear part :
Alexander Duyck2996d312012-05-02 18:18:42 +00001758 * If skb->head_frag is set, this 'linear' part is backed by a
1759 * fragment, and if the head is not shared with any clones then
1760 * we can avoid a copy since we own the head portion of this page.
Jens Axboe9c55e012007-11-06 23:30:13 -08001761 */
Octavian Purdila2870c432008-07-15 00:49:11 -07001762 if (__splice_segment(virt_to_page(skb->data),
1763 (unsigned long) skb->data & (PAGE_SIZE - 1),
1764 skb_headlen(skb),
Eric Dumazet18aafc62013-01-11 14:46:37 +00001765 offset, len, spd,
Alexander Duyck3a7c1ee42012-05-03 01:09:42 +00001766 skb_head_is_locked(skb),
Eric Dumazet1d0c0b32012-04-27 02:10:03 +00001767 sk, pipe))
David S. Millera108d5f2012-04-23 23:06:11 -04001768 return true;
Jens Axboe9c55e012007-11-06 23:30:13 -08001769
1770 /*
1771 * then map the fragments
1772 */
Jens Axboe9c55e012007-11-06 23:30:13 -08001773 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1774 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1775
Ian Campbellea2ab692011-08-22 23:44:58 +00001776 if (__splice_segment(skb_frag_page(f),
Eric Dumazet9e903e02011-10-18 21:00:24 +00001777 f->page_offset, skb_frag_size(f),
Eric Dumazet18aafc62013-01-11 14:46:37 +00001778 offset, len, spd, false, sk, pipe))
David S. Millera108d5f2012-04-23 23:06:11 -04001779 return true;
Jens Axboe9c55e012007-11-06 23:30:13 -08001780 }
1781
David S. Millera108d5f2012-04-23 23:06:11 -04001782 return false;
Jens Axboe9c55e012007-11-06 23:30:13 -08001783}
1784
1785/*
1786 * Map data from the skb to a pipe. Should handle both the linear part,
1787 * the fragments, and the frag list. It does NOT handle frag lists within
1788 * the frag list, if such a thing exists. We'd probably need to recurse to
1789 * handle that cleanly.
1790 */
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001791int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
Jens Axboe9c55e012007-11-06 23:30:13 -08001792 struct pipe_inode_info *pipe, unsigned int tlen,
1793 unsigned int flags)
1794{
Eric Dumazet41c73a02012-04-22 12:26:16 +00001795 struct partial_page partial[MAX_SKB_FRAGS];
1796 struct page *pages[MAX_SKB_FRAGS];
Jens Axboe9c55e012007-11-06 23:30:13 -08001797 struct splice_pipe_desc spd = {
1798 .pages = pages,
1799 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +02001800 .nr_pages_max = MAX_SKB_FRAGS,
Jens Axboe9c55e012007-11-06 23:30:13 -08001801 .flags = flags,
Miklos Szeredi28a625c2014-01-22 19:36:57 +01001802 .ops = &nosteal_pipe_buf_ops,
Jens Axboe9c55e012007-11-06 23:30:13 -08001803 .spd_release = sock_spd_release,
1804 };
David S. Millerfbb398a2009-06-09 00:18:59 -07001805 struct sk_buff *frag_iter;
Jarek Poplawski7a67e562009-04-30 05:41:19 -07001806 struct sock *sk = skb->sk;
Jens Axboe35f3d142010-05-20 10:43:18 +02001807 int ret = 0;
1808
Jens Axboe9c55e012007-11-06 23:30:13 -08001809 /*
1810 * __skb_splice_bits() only fails if the output has no room left,
1811 * so no point in going over the frag_list for the error case.
1812 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001813 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
Jens Axboe9c55e012007-11-06 23:30:13 -08001814 goto done;
1815 else if (!tlen)
1816 goto done;
1817
1818 /*
1819 * now see if we have a frag_list to map
1820 */
David S. Millerfbb398a2009-06-09 00:18:59 -07001821 skb_walk_frags(skb, frag_iter) {
1822 if (!tlen)
1823 break;
Jens Axboe35f3d142010-05-20 10:43:18 +02001824 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
David S. Millerfbb398a2009-06-09 00:18:59 -07001825 break;
Jens Axboe9c55e012007-11-06 23:30:13 -08001826 }
1827
1828done:
Jens Axboe9c55e012007-11-06 23:30:13 -08001829 if (spd.nr_pages) {
Jens Axboe9c55e012007-11-06 23:30:13 -08001830 /*
1831 * Drop the socket lock, otherwise we have reverse
1832 * locking dependencies between sk_lock and i_mutex
1833 * here as compared to sendfile(). We enter here
1834 * with the socket lock held, and splice_to_pipe() will
1835 * grab the pipe inode lock. For sendfile() emulation,
1836 * we call into ->sendpage() with the i_mutex lock held
1837 * and networking will grab the socket lock.
1838 */
Octavian Purdila293ad602008-06-04 15:45:58 -07001839 release_sock(sk);
Jens Axboe9c55e012007-11-06 23:30:13 -08001840 ret = splice_to_pipe(pipe, &spd);
Octavian Purdila293ad602008-06-04 15:45:58 -07001841 lock_sock(sk);
Jens Axboe9c55e012007-11-06 23:30:13 -08001842 }
1843
Jens Axboe35f3d142010-05-20 10:43:18 +02001844 return ret;
Jens Axboe9c55e012007-11-06 23:30:13 -08001845}
1846
Herbert Xu357b40a2005-04-19 22:30:14 -07001847/**
1848 * skb_store_bits - store bits from kernel buffer to skb
1849 * @skb: destination buffer
1850 * @offset: offset in destination
1851 * @from: source buffer
1852 * @len: number of bytes to copy
1853 *
1854 * Copy the specified number of bytes from the source buffer to the
1855 * destination skb. This function handles all the messy bits of
1856 * traversing fragment lists and such.
1857 */
1858
Stephen Hemminger0c6fcc82007-04-20 16:40:01 -07001859int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
Herbert Xu357b40a2005-04-19 22:30:14 -07001860{
David S. Miller1a028e52007-04-27 15:21:23 -07001861 int start = skb_headlen(skb);
David S. Millerfbb398a2009-06-09 00:18:59 -07001862 struct sk_buff *frag_iter;
1863 int i, copy;
Herbert Xu357b40a2005-04-19 22:30:14 -07001864
1865 if (offset > (int)skb->len - len)
1866 goto fault;
1867
David S. Miller1a028e52007-04-27 15:21:23 -07001868 if ((copy = start - offset) > 0) {
Herbert Xu357b40a2005-04-19 22:30:14 -07001869 if (copy > len)
1870 copy = len;
Arnaldo Carvalho de Melo27d7ff42007-03-31 11:55:19 -03001871 skb_copy_to_linear_data_offset(skb, offset, from, copy);
Herbert Xu357b40a2005-04-19 22:30:14 -07001872 if ((len -= copy) == 0)
1873 return 0;
1874 offset += copy;
1875 from += copy;
1876 }
1877
1878 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1879 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
David S. Miller1a028e52007-04-27 15:21:23 -07001880 int end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001881
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001882 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07001883
Eric Dumazet9e903e02011-10-18 21:00:24 +00001884 end = start + skb_frag_size(frag);
Herbert Xu357b40a2005-04-19 22:30:14 -07001885 if ((copy = end - offset) > 0) {
1886 u8 *vaddr;
1887
1888 if (copy > len)
1889 copy = len;
1890
Eric Dumazet51c56b02012-04-05 11:35:15 +02001891 vaddr = kmap_atomic(skb_frag_page(frag));
David S. Miller1a028e52007-04-27 15:21:23 -07001892 memcpy(vaddr + frag->page_offset + offset - start,
1893 from, copy);
Eric Dumazet51c56b02012-04-05 11:35:15 +02001894 kunmap_atomic(vaddr);
Herbert Xu357b40a2005-04-19 22:30:14 -07001895
1896 if ((len -= copy) == 0)
1897 return 0;
1898 offset += copy;
1899 from += copy;
1900 }
David S. Miller1a028e52007-04-27 15:21:23 -07001901 start = end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001902 }
1903
David S. Millerfbb398a2009-06-09 00:18:59 -07001904 skb_walk_frags(skb, frag_iter) {
1905 int end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001906
David S. Millerfbb398a2009-06-09 00:18:59 -07001907 WARN_ON(start > offset + len);
Herbert Xu357b40a2005-04-19 22:30:14 -07001908
David S. Millerfbb398a2009-06-09 00:18:59 -07001909 end = start + frag_iter->len;
1910 if ((copy = end - offset) > 0) {
1911 if (copy > len)
1912 copy = len;
1913 if (skb_store_bits(frag_iter, offset - start,
1914 from, copy))
1915 goto fault;
1916 if ((len -= copy) == 0)
1917 return 0;
1918 offset += copy;
1919 from += copy;
Herbert Xu357b40a2005-04-19 22:30:14 -07001920 }
David S. Millerfbb398a2009-06-09 00:18:59 -07001921 start = end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001922 }
1923 if (!len)
1924 return 0;
1925
1926fault:
1927 return -EFAULT;
1928}
Herbert Xu357b40a2005-04-19 22:30:14 -07001929EXPORT_SYMBOL(skb_store_bits);
1930
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931/* Checksum skb data. */
Daniel Borkmann2817a332013-10-30 11:50:51 +01001932__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
1933 __wsum csum, const struct skb_checksum_ops *ops)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934{
David S. Miller1a028e52007-04-27 15:21:23 -07001935 int start = skb_headlen(skb);
1936 int i, copy = start - offset;
David S. Millerfbb398a2009-06-09 00:18:59 -07001937 struct sk_buff *frag_iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 int pos = 0;
1939
1940 /* Checksum header. */
1941 if (copy > 0) {
1942 if (copy > len)
1943 copy = len;
Daniel Borkmann2817a332013-10-30 11:50:51 +01001944 csum = ops->update(skb->data + offset, copy, csum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 if ((len -= copy) == 0)
1946 return csum;
1947 offset += copy;
1948 pos = copy;
1949 }
1950
1951 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07001952 int end;
Eric Dumazet51c56b02012-04-05 11:35:15 +02001953 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001955 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07001956
Eric Dumazet51c56b02012-04-05 11:35:15 +02001957 end = start + skb_frag_size(frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 if ((copy = end - offset) > 0) {
Al Viro44bb9362006-11-14 21:36:14 -08001959 __wsum csum2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 u8 *vaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961
1962 if (copy > len)
1963 copy = len;
Eric Dumazet51c56b02012-04-05 11:35:15 +02001964 vaddr = kmap_atomic(skb_frag_page(frag));
Daniel Borkmann2817a332013-10-30 11:50:51 +01001965 csum2 = ops->update(vaddr + frag->page_offset +
1966 offset - start, copy, 0);
Eric Dumazet51c56b02012-04-05 11:35:15 +02001967 kunmap_atomic(vaddr);
Daniel Borkmann2817a332013-10-30 11:50:51 +01001968 csum = ops->combine(csum, csum2, pos, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 if (!(len -= copy))
1970 return csum;
1971 offset += copy;
1972 pos += copy;
1973 }
David S. Miller1a028e52007-04-27 15:21:23 -07001974 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 }
1976
David S. Millerfbb398a2009-06-09 00:18:59 -07001977 skb_walk_frags(skb, frag_iter) {
1978 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979
David S. Millerfbb398a2009-06-09 00:18:59 -07001980 WARN_ON(start > offset + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981
David S. Millerfbb398a2009-06-09 00:18:59 -07001982 end = start + frag_iter->len;
1983 if ((copy = end - offset) > 0) {
1984 __wsum csum2;
1985 if (copy > len)
1986 copy = len;
Daniel Borkmann2817a332013-10-30 11:50:51 +01001987 csum2 = __skb_checksum(frag_iter, offset - start,
1988 copy, 0, ops);
1989 csum = ops->combine(csum, csum2, pos, copy);
David S. Millerfbb398a2009-06-09 00:18:59 -07001990 if ((len -= copy) == 0)
1991 return csum;
1992 offset += copy;
1993 pos += copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 }
David S. Millerfbb398a2009-06-09 00:18:59 -07001995 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 }
Kris Katterjohn09a62662006-01-08 22:24:28 -08001997 BUG_ON(len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998
1999 return csum;
2000}
Daniel Borkmann2817a332013-10-30 11:50:51 +01002001EXPORT_SYMBOL(__skb_checksum);
2002
2003__wsum skb_checksum(const struct sk_buff *skb, int offset,
2004 int len, __wsum csum)
2005{
2006 const struct skb_checksum_ops ops = {
Daniel Borkmanncea80ea2013-11-04 17:10:25 +01002007 .update = csum_partial_ext,
Daniel Borkmann2817a332013-10-30 11:50:51 +01002008 .combine = csum_block_add_ext,
2009 };
2010
2011 return __skb_checksum(skb, offset, len, csum, &ops);
2012}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002013EXPORT_SYMBOL(skb_checksum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014
2015/* Both of above in one bottle. */
2016
Al Viro81d77662006-11-14 21:37:33 -08002017__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2018 u8 *to, int len, __wsum csum)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019{
David S. Miller1a028e52007-04-27 15:21:23 -07002020 int start = skb_headlen(skb);
2021 int i, copy = start - offset;
David S. Millerfbb398a2009-06-09 00:18:59 -07002022 struct sk_buff *frag_iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 int pos = 0;
2024
2025 /* Copy header. */
2026 if (copy > 0) {
2027 if (copy > len)
2028 copy = len;
2029 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2030 copy, csum);
2031 if ((len -= copy) == 0)
2032 return csum;
2033 offset += copy;
2034 to += copy;
2035 pos = copy;
2036 }
2037
2038 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07002039 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040
Ilpo Järvinen547b7922008-07-25 21:43:18 -07002041 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07002042
Eric Dumazet9e903e02011-10-18 21:00:24 +00002043 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 if ((copy = end - offset) > 0) {
Al Viro50842052006-11-14 21:36:34 -08002045 __wsum csum2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 u8 *vaddr;
2047 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2048
2049 if (copy > len)
2050 copy = len;
Eric Dumazet51c56b02012-04-05 11:35:15 +02002051 vaddr = kmap_atomic(skb_frag_page(frag));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 csum2 = csum_partial_copy_nocheck(vaddr +
David S. Miller1a028e52007-04-27 15:21:23 -07002053 frag->page_offset +
2054 offset - start, to,
2055 copy, 0);
Eric Dumazet51c56b02012-04-05 11:35:15 +02002056 kunmap_atomic(vaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 csum = csum_block_add(csum, csum2, pos);
2058 if (!(len -= copy))
2059 return csum;
2060 offset += copy;
2061 to += copy;
2062 pos += copy;
2063 }
David S. Miller1a028e52007-04-27 15:21:23 -07002064 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 }
2066
David S. Millerfbb398a2009-06-09 00:18:59 -07002067 skb_walk_frags(skb, frag_iter) {
2068 __wsum csum2;
2069 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070
David S. Millerfbb398a2009-06-09 00:18:59 -07002071 WARN_ON(start > offset + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072
David S. Millerfbb398a2009-06-09 00:18:59 -07002073 end = start + frag_iter->len;
2074 if ((copy = end - offset) > 0) {
2075 if (copy > len)
2076 copy = len;
2077 csum2 = skb_copy_and_csum_bits(frag_iter,
2078 offset - start,
2079 to, copy, 0);
2080 csum = csum_block_add(csum, csum2, pos);
2081 if ((len -= copy) == 0)
2082 return csum;
2083 offset += copy;
2084 to += copy;
2085 pos += copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 }
David S. Millerfbb398a2009-06-09 00:18:59 -07002087 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 }
Kris Katterjohn09a62662006-01-08 22:24:28 -08002089 BUG_ON(len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 return csum;
2091}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002092EXPORT_SYMBOL(skb_copy_and_csum_bits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093
Thomas Grafaf2806f2013-12-13 15:22:17 +01002094 /**
2095 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2096 * @from: source buffer
2097 *
2098 * Calculates the amount of linear headroom needed in the 'to' skb passed
2099 * into skb_zerocopy().
2100 */
2101unsigned int
2102skb_zerocopy_headlen(const struct sk_buff *from)
2103{
2104 unsigned int hlen = 0;
2105
2106 if (!from->head_frag ||
2107 skb_headlen(from) < L1_CACHE_BYTES ||
2108 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2109 hlen = skb_headlen(from);
2110
2111 if (skb_has_frag_list(from))
2112 hlen = from->len;
2113
2114 return hlen;
2115}
2116EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2117
2118/**
2119 * skb_zerocopy - Zero copy skb to skb
2120 * @to: destination buffer
Masanari Iida7fceb4d2014-01-29 01:05:28 +09002121 * @from: source buffer
Thomas Grafaf2806f2013-12-13 15:22:17 +01002122 * @len: number of bytes to copy from source buffer
2123 * @hlen: size of linear headroom in destination buffer
2124 *
2125 * Copies up to `len` bytes from `from` to `to` by creating references
2126 * to the frags in the source buffer.
2127 *
2128 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2129 * headroom in the `to` buffer.
Zoltan Kiss36d5fe62014-03-26 22:37:45 +00002130 *
2131 * Return value:
2132 * 0: everything is OK
2133 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2134 * -EFAULT: skb_copy_bits() found some problem with skb geometry
Thomas Grafaf2806f2013-12-13 15:22:17 +01002135 */
Zoltan Kiss36d5fe62014-03-26 22:37:45 +00002136int
2137skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
Thomas Grafaf2806f2013-12-13 15:22:17 +01002138{
2139 int i, j = 0;
2140 int plen = 0; /* length of skb->head fragment */
Zoltan Kiss36d5fe62014-03-26 22:37:45 +00002141 int ret;
Thomas Grafaf2806f2013-12-13 15:22:17 +01002142 struct page *page;
2143 unsigned int offset;
2144
2145 BUG_ON(!from->head_frag && !hlen);
2146
2147 /* dont bother with small payloads */
Zoltan Kiss36d5fe62014-03-26 22:37:45 +00002148 if (len <= skb_tailroom(to))
2149 return skb_copy_bits(from, 0, skb_put(to, len), len);
Thomas Grafaf2806f2013-12-13 15:22:17 +01002150
2151 if (hlen) {
Zoltan Kiss36d5fe62014-03-26 22:37:45 +00002152 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2153 if (unlikely(ret))
2154 return ret;
Thomas Grafaf2806f2013-12-13 15:22:17 +01002155 len -= hlen;
2156 } else {
2157 plen = min_t(int, skb_headlen(from), len);
2158 if (plen) {
2159 page = virt_to_head_page(from->head);
2160 offset = from->data - (unsigned char *)page_address(page);
2161 __skb_fill_page_desc(to, 0, page, offset, plen);
2162 get_page(page);
2163 j = 1;
2164 len -= plen;
2165 }
2166 }
2167
2168 to->truesize += len + plen;
2169 to->len += len + plen;
2170 to->data_len += len + plen;
2171
Zoltan Kiss36d5fe62014-03-26 22:37:45 +00002172 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2173 skb_tx_error(from);
2174 return -ENOMEM;
2175 }
2176
Thomas Grafaf2806f2013-12-13 15:22:17 +01002177 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2178 if (!len)
2179 break;
2180 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2181 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
2182 len -= skb_shinfo(to)->frags[j].size;
2183 skb_frag_ref(to, j);
2184 j++;
2185 }
2186 skb_shinfo(to)->nr_frags = j;
Zoltan Kiss36d5fe62014-03-26 22:37:45 +00002187
2188 return 0;
Thomas Grafaf2806f2013-12-13 15:22:17 +01002189}
2190EXPORT_SYMBOL_GPL(skb_zerocopy);
2191
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2193{
Al Virod3bc23e2006-11-14 21:24:49 -08002194 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 long csstart;
2196
Patrick McHardy84fa7932006-08-29 16:44:56 -07002197 if (skb->ip_summed == CHECKSUM_PARTIAL)
Michał Mirosław55508d62010-12-14 15:24:08 +00002198 csstart = skb_checksum_start_offset(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 else
2200 csstart = skb_headlen(skb);
2201
Kris Katterjohn09a62662006-01-08 22:24:28 -08002202 BUG_ON(csstart > skb_headlen(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03002204 skb_copy_from_linear_data(skb, to, csstart);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205
2206 csum = 0;
2207 if (csstart != skb->len)
2208 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2209 skb->len - csstart, 0);
2210
Patrick McHardy84fa7932006-08-29 16:44:56 -07002211 if (skb->ip_summed == CHECKSUM_PARTIAL) {
Al Viroff1dcad2006-11-20 18:07:29 -08002212 long csstuff = csstart + skb->csum_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213
Al Virod3bc23e2006-11-14 21:24:49 -08002214 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215 }
2216}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002217EXPORT_SYMBOL(skb_copy_and_csum_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218
2219/**
2220 * skb_dequeue - remove from the head of the queue
2221 * @list: list to dequeue from
2222 *
2223 * Remove the head of the list. The list lock is taken so the function
2224 * may be used safely with other locking list functions. The head item is
2225 * returned or %NULL if the list is empty.
2226 */
2227
2228struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2229{
2230 unsigned long flags;
2231 struct sk_buff *result;
2232
2233 spin_lock_irqsave(&list->lock, flags);
2234 result = __skb_dequeue(list);
2235 spin_unlock_irqrestore(&list->lock, flags);
2236 return result;
2237}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002238EXPORT_SYMBOL(skb_dequeue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239
2240/**
2241 * skb_dequeue_tail - remove from the tail of the queue
2242 * @list: list to dequeue from
2243 *
2244 * Remove the tail of the list. The list lock is taken so the function
2245 * may be used safely with other locking list functions. The tail item is
2246 * returned or %NULL if the list is empty.
2247 */
2248struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2249{
2250 unsigned long flags;
2251 struct sk_buff *result;
2252
2253 spin_lock_irqsave(&list->lock, flags);
2254 result = __skb_dequeue_tail(list);
2255 spin_unlock_irqrestore(&list->lock, flags);
2256 return result;
2257}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002258EXPORT_SYMBOL(skb_dequeue_tail);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259
2260/**
2261 * skb_queue_purge - empty a list
2262 * @list: list to empty
2263 *
2264 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2265 * the list and one reference dropped. This function takes the list
2266 * lock and is atomic with respect to other list locking functions.
2267 */
2268void skb_queue_purge(struct sk_buff_head *list)
2269{
2270 struct sk_buff *skb;
2271 while ((skb = skb_dequeue(list)) != NULL)
2272 kfree_skb(skb);
2273}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002274EXPORT_SYMBOL(skb_queue_purge);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275
2276/**
2277 * skb_queue_head - queue a buffer at the list head
2278 * @list: list to use
2279 * @newsk: buffer to queue
2280 *
2281 * Queue a buffer at the start of the list. This function takes the
2282 * list lock and can be used safely with other locking &sk_buff functions
2283 * safely.
2284 *
2285 * A buffer cannot be placed on two lists at the same time.
2286 */
2287void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2288{
2289 unsigned long flags;
2290
2291 spin_lock_irqsave(&list->lock, flags);
2292 __skb_queue_head(list, newsk);
2293 spin_unlock_irqrestore(&list->lock, flags);
2294}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002295EXPORT_SYMBOL(skb_queue_head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296
2297/**
2298 * skb_queue_tail - queue a buffer at the list tail
2299 * @list: list to use
2300 * @newsk: buffer to queue
2301 *
2302 * Queue a buffer at the tail of the list. This function takes the
2303 * list lock and can be used safely with other locking &sk_buff functions
2304 * safely.
2305 *
2306 * A buffer cannot be placed on two lists at the same time.
2307 */
2308void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2309{
2310 unsigned long flags;
2311
2312 spin_lock_irqsave(&list->lock, flags);
2313 __skb_queue_tail(list, newsk);
2314 spin_unlock_irqrestore(&list->lock, flags);
2315}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002316EXPORT_SYMBOL(skb_queue_tail);
David S. Miller8728b832005-08-09 19:25:21 -07002317
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318/**
2319 * skb_unlink - remove a buffer from a list
2320 * @skb: buffer to remove
David S. Miller8728b832005-08-09 19:25:21 -07002321 * @list: list to use
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 *
David S. Miller8728b832005-08-09 19:25:21 -07002323 * Remove a packet from a list. The list locks are taken and this
2324 * function is atomic with respect to other list locked calls
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 *
David S. Miller8728b832005-08-09 19:25:21 -07002326 * You must know what list the SKB is on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 */
David S. Miller8728b832005-08-09 19:25:21 -07002328void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329{
David S. Miller8728b832005-08-09 19:25:21 -07002330 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331
David S. Miller8728b832005-08-09 19:25:21 -07002332 spin_lock_irqsave(&list->lock, flags);
2333 __skb_unlink(skb, list);
2334 spin_unlock_irqrestore(&list->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002336EXPORT_SYMBOL(skb_unlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338/**
2339 * skb_append - append a buffer
2340 * @old: buffer to insert after
2341 * @newsk: buffer to insert
David S. Miller8728b832005-08-09 19:25:21 -07002342 * @list: list to use
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 *
2344 * Place a packet after a given packet in a list. The list locks are taken
2345 * and this function is atomic with respect to other list locked calls.
2346 * A buffer cannot be placed on two lists at the same time.
2347 */
David S. Miller8728b832005-08-09 19:25:21 -07002348void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349{
2350 unsigned long flags;
2351
David S. Miller8728b832005-08-09 19:25:21 -07002352 spin_lock_irqsave(&list->lock, flags);
Gerrit Renker7de6c032008-04-14 00:05:09 -07002353 __skb_queue_after(list, old, newsk);
David S. Miller8728b832005-08-09 19:25:21 -07002354 spin_unlock_irqrestore(&list->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002356EXPORT_SYMBOL(skb_append);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357
2358/**
2359 * skb_insert - insert a buffer
2360 * @old: buffer to insert before
2361 * @newsk: buffer to insert
David S. Miller8728b832005-08-09 19:25:21 -07002362 * @list: list to use
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363 *
David S. Miller8728b832005-08-09 19:25:21 -07002364 * Place a packet before a given packet in a list. The list locks are
2365 * taken and this function is atomic with respect to other list locked
2366 * calls.
2367 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368 * A buffer cannot be placed on two lists at the same time.
2369 */
David S. Miller8728b832005-08-09 19:25:21 -07002370void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371{
2372 unsigned long flags;
2373
David S. Miller8728b832005-08-09 19:25:21 -07002374 spin_lock_irqsave(&list->lock, flags);
2375 __skb_insert(newsk, old->prev, old, list);
2376 spin_unlock_irqrestore(&list->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002378EXPORT_SYMBOL(skb_insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380static inline void skb_split_inside_header(struct sk_buff *skb,
2381 struct sk_buff* skb1,
2382 const u32 len, const int pos)
2383{
2384 int i;
2385
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03002386 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2387 pos - len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 /* And move data appendix as is. */
2389 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2390 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2391
2392 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2393 skb_shinfo(skb)->nr_frags = 0;
2394 skb1->data_len = skb->data_len;
2395 skb1->len += skb1->data_len;
2396 skb->data_len = 0;
2397 skb->len = len;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07002398 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399}
2400
2401static inline void skb_split_no_header(struct sk_buff *skb,
2402 struct sk_buff* skb1,
2403 const u32 len, int pos)
2404{
2405 int i, k = 0;
2406 const int nfrags = skb_shinfo(skb)->nr_frags;
2407
2408 skb_shinfo(skb)->nr_frags = 0;
2409 skb1->len = skb1->data_len = skb->len - len;
2410 skb->len = len;
2411 skb->data_len = len - pos;
2412
2413 for (i = 0; i < nfrags; i++) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00002414 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415
2416 if (pos + size > len) {
2417 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2418
2419 if (pos < len) {
2420 /* Split frag.
2421 * We have two variants in this case:
2422 * 1. Move all the frag to the second
2423 * part, if it is possible. F.e.
2424 * this approach is mandatory for TUX,
2425 * where splitting is expensive.
2426 * 2. Split is accurately. We make this.
2427 */
Ian Campbellea2ab692011-08-22 23:44:58 +00002428 skb_frag_ref(skb, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
Eric Dumazet9e903e02011-10-18 21:00:24 +00002430 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2431 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432 skb_shinfo(skb)->nr_frags++;
2433 }
2434 k++;
2435 } else
2436 skb_shinfo(skb)->nr_frags++;
2437 pos += size;
2438 }
2439 skb_shinfo(skb1)->nr_frags = k;
2440}
2441
2442/**
2443 * skb_split - Split fragmented skb to two parts at length len.
2444 * @skb: the buffer to split
2445 * @skb1: the buffer to receive the second part
2446 * @len: new length for skb
2447 */
2448void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2449{
2450 int pos = skb_headlen(skb);
2451
Amerigo Wang68534c62013-02-19 22:51:30 +00002452 skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453 if (len < pos) /* Split line is inside header. */
2454 skb_split_inside_header(skb, skb1, len, pos);
2455 else /* Second chunk has no header, nothing to copy. */
2456 skb_split_no_header(skb, skb1, len, pos);
2457}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002458EXPORT_SYMBOL(skb_split);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459
Ilpo Järvinen9f782db2008-11-25 13:57:01 -08002460/* Shifting from/to a cloned skb is a no-go.
2461 *
2462 * Caller cannot keep skb_shinfo related pointers past calling here!
2463 */
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002464static int skb_prepare_for_shift(struct sk_buff *skb)
2465{
Ilpo Järvinen0ace2852008-11-24 21:30:21 -08002466 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002467}
2468
2469/**
2470 * skb_shift - Shifts paged data partially from skb to another
2471 * @tgt: buffer into which tail data gets added
2472 * @skb: buffer from which the paged data comes from
2473 * @shiftlen: shift up to this many bytes
2474 *
2475 * Attempts to shift up to shiftlen worth of bytes, which may be less than
Feng King20e994a2011-11-21 01:47:11 +00002476 * the length of the skb, from skb to tgt. Returns number bytes shifted.
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002477 * It's up to caller to free skb if everything was shifted.
2478 *
2479 * If @tgt runs out of frags, the whole operation is aborted.
2480 *
2481 * Skb cannot include anything else but paged data while tgt is allowed
2482 * to have non-paged data as well.
2483 *
2484 * TODO: full sized shift could be optimized but that would need
2485 * specialized skb free'er to handle frags without up-to-date nr_frags.
2486 */
2487int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2488{
2489 int from, to, merge, todo;
2490 struct skb_frag_struct *fragfrom, *fragto;
2491
2492 BUG_ON(shiftlen > skb->len);
2493 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2494
2495 todo = shiftlen;
2496 from = 0;
2497 to = skb_shinfo(tgt)->nr_frags;
2498 fragfrom = &skb_shinfo(skb)->frags[from];
2499
2500 /* Actual merge is delayed until the point when we know we can
2501 * commit all, so that we don't have to undo partial changes
2502 */
2503 if (!to ||
Ian Campbellea2ab692011-08-22 23:44:58 +00002504 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2505 fragfrom->page_offset)) {
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002506 merge = -1;
2507 } else {
2508 merge = to - 1;
2509
Eric Dumazet9e903e02011-10-18 21:00:24 +00002510 todo -= skb_frag_size(fragfrom);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002511 if (todo < 0) {
2512 if (skb_prepare_for_shift(skb) ||
2513 skb_prepare_for_shift(tgt))
2514 return 0;
2515
Ilpo Järvinen9f782db2008-11-25 13:57:01 -08002516 /* All previous frag pointers might be stale! */
2517 fragfrom = &skb_shinfo(skb)->frags[from];
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002518 fragto = &skb_shinfo(tgt)->frags[merge];
2519
Eric Dumazet9e903e02011-10-18 21:00:24 +00002520 skb_frag_size_add(fragto, shiftlen);
2521 skb_frag_size_sub(fragfrom, shiftlen);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002522 fragfrom->page_offset += shiftlen;
2523
2524 goto onlymerged;
2525 }
2526
2527 from++;
2528 }
2529
2530 /* Skip full, not-fitting skb to avoid expensive operations */
2531 if ((shiftlen == skb->len) &&
2532 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2533 return 0;
2534
2535 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2536 return 0;
2537
2538 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2539 if (to == MAX_SKB_FRAGS)
2540 return 0;
2541
2542 fragfrom = &skb_shinfo(skb)->frags[from];
2543 fragto = &skb_shinfo(tgt)->frags[to];
2544
Eric Dumazet9e903e02011-10-18 21:00:24 +00002545 if (todo >= skb_frag_size(fragfrom)) {
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002546 *fragto = *fragfrom;
Eric Dumazet9e903e02011-10-18 21:00:24 +00002547 todo -= skb_frag_size(fragfrom);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002548 from++;
2549 to++;
2550
2551 } else {
Ian Campbellea2ab692011-08-22 23:44:58 +00002552 __skb_frag_ref(fragfrom);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002553 fragto->page = fragfrom->page;
2554 fragto->page_offset = fragfrom->page_offset;
Eric Dumazet9e903e02011-10-18 21:00:24 +00002555 skb_frag_size_set(fragto, todo);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002556
2557 fragfrom->page_offset += todo;
Eric Dumazet9e903e02011-10-18 21:00:24 +00002558 skb_frag_size_sub(fragfrom, todo);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002559 todo = 0;
2560
2561 to++;
2562 break;
2563 }
2564 }
2565
2566 /* Ready to "commit" this state change to tgt */
2567 skb_shinfo(tgt)->nr_frags = to;
2568
2569 if (merge >= 0) {
2570 fragfrom = &skb_shinfo(skb)->frags[0];
2571 fragto = &skb_shinfo(tgt)->frags[merge];
2572
Eric Dumazet9e903e02011-10-18 21:00:24 +00002573 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
Ian Campbellea2ab692011-08-22 23:44:58 +00002574 __skb_frag_unref(fragfrom);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002575 }
2576
2577 /* Reposition in the original skb */
2578 to = 0;
2579 while (from < skb_shinfo(skb)->nr_frags)
2580 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2581 skb_shinfo(skb)->nr_frags = to;
2582
2583 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2584
2585onlymerged:
2586 /* Most likely the tgt won't ever need its checksum anymore, skb on
2587 * the other hand might need it if it needs to be resent
2588 */
2589 tgt->ip_summed = CHECKSUM_PARTIAL;
2590 skb->ip_summed = CHECKSUM_PARTIAL;
2591
2592 /* Yak, is it really working this way? Some helper please? */
2593 skb->len -= shiftlen;
2594 skb->data_len -= shiftlen;
2595 skb->truesize -= shiftlen;
2596 tgt->len += shiftlen;
2597 tgt->data_len += shiftlen;
2598 tgt->truesize += shiftlen;
2599
2600 return shiftlen;
2601}
2602
Thomas Graf677e90e2005-06-23 20:59:51 -07002603/**
2604 * skb_prepare_seq_read - Prepare a sequential read of skb data
2605 * @skb: the buffer to read
2606 * @from: lower offset of data to be read
2607 * @to: upper offset of data to be read
2608 * @st: state variable
2609 *
2610 * Initializes the specified state variable. Must be called before
2611 * invoking skb_seq_read() for the first time.
2612 */
2613void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2614 unsigned int to, struct skb_seq_state *st)
2615{
2616 st->lower_offset = from;
2617 st->upper_offset = to;
2618 st->root_skb = st->cur_skb = skb;
2619 st->frag_idx = st->stepped_offset = 0;
2620 st->frag_data = NULL;
2621}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002622EXPORT_SYMBOL(skb_prepare_seq_read);
Thomas Graf677e90e2005-06-23 20:59:51 -07002623
2624/**
2625 * skb_seq_read - Sequentially read skb data
2626 * @consumed: number of bytes consumed by the caller so far
2627 * @data: destination pointer for data to be returned
2628 * @st: state variable
2629 *
Mathias Krausebc323832013-11-07 14:18:26 +01002630 * Reads a block of skb data at @consumed relative to the
Thomas Graf677e90e2005-06-23 20:59:51 -07002631 * lower offset specified to skb_prepare_seq_read(). Assigns
Mathias Krausebc323832013-11-07 14:18:26 +01002632 * the head of the data block to @data and returns the length
Thomas Graf677e90e2005-06-23 20:59:51 -07002633 * of the block or 0 if the end of the skb data or the upper
2634 * offset has been reached.
2635 *
2636 * The caller is not required to consume all of the data
Mathias Krausebc323832013-11-07 14:18:26 +01002637 * returned, i.e. @consumed is typically set to the number
Thomas Graf677e90e2005-06-23 20:59:51 -07002638 * of bytes already consumed and the next call to
2639 * skb_seq_read() will return the remaining part of the block.
2640 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002641 * Note 1: The size of each block of data returned can be arbitrary,
Thomas Graf677e90e2005-06-23 20:59:51 -07002642 * this limitation is the cost for zerocopy seqeuental
2643 * reads of potentially non linear data.
2644 *
Randy Dunlapbc2cda12008-02-13 15:03:25 -08002645 * Note 2: Fragment lists within fragments are not implemented
Thomas Graf677e90e2005-06-23 20:59:51 -07002646 * at the moment, state->root_skb could be replaced with
2647 * a stack for this purpose.
2648 */
2649unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2650 struct skb_seq_state *st)
2651{
2652 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2653 skb_frag_t *frag;
2654
Wedson Almeida Filhoaeb193e2013-06-23 23:33:48 -07002655 if (unlikely(abs_offset >= st->upper_offset)) {
2656 if (st->frag_data) {
2657 kunmap_atomic(st->frag_data);
2658 st->frag_data = NULL;
2659 }
Thomas Graf677e90e2005-06-23 20:59:51 -07002660 return 0;
Wedson Almeida Filhoaeb193e2013-06-23 23:33:48 -07002661 }
Thomas Graf677e90e2005-06-23 20:59:51 -07002662
2663next_skb:
Herbert Xu95e3b242009-01-29 16:07:52 -08002664 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
Thomas Graf677e90e2005-06-23 20:59:51 -07002665
Thomas Chenault995b3372009-05-18 21:43:27 -07002666 if (abs_offset < block_limit && !st->frag_data) {
Herbert Xu95e3b242009-01-29 16:07:52 -08002667 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
Thomas Graf677e90e2005-06-23 20:59:51 -07002668 return block_limit - abs_offset;
2669 }
2670
2671 if (st->frag_idx == 0 && !st->frag_data)
2672 st->stepped_offset += skb_headlen(st->cur_skb);
2673
2674 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2675 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
Eric Dumazet9e903e02011-10-18 21:00:24 +00002676 block_limit = skb_frag_size(frag) + st->stepped_offset;
Thomas Graf677e90e2005-06-23 20:59:51 -07002677
2678 if (abs_offset < block_limit) {
2679 if (!st->frag_data)
Eric Dumazet51c56b02012-04-05 11:35:15 +02002680 st->frag_data = kmap_atomic(skb_frag_page(frag));
Thomas Graf677e90e2005-06-23 20:59:51 -07002681
2682 *data = (u8 *) st->frag_data + frag->page_offset +
2683 (abs_offset - st->stepped_offset);
2684
2685 return block_limit - abs_offset;
2686 }
2687
2688 if (st->frag_data) {
Eric Dumazet51c56b02012-04-05 11:35:15 +02002689 kunmap_atomic(st->frag_data);
Thomas Graf677e90e2005-06-23 20:59:51 -07002690 st->frag_data = NULL;
2691 }
2692
2693 st->frag_idx++;
Eric Dumazet9e903e02011-10-18 21:00:24 +00002694 st->stepped_offset += skb_frag_size(frag);
Thomas Graf677e90e2005-06-23 20:59:51 -07002695 }
2696
Olaf Kirch5b5a60d2007-06-23 23:11:52 -07002697 if (st->frag_data) {
Eric Dumazet51c56b02012-04-05 11:35:15 +02002698 kunmap_atomic(st->frag_data);
Olaf Kirch5b5a60d2007-06-23 23:11:52 -07002699 st->frag_data = NULL;
2700 }
2701
David S. Miller21dc3302010-08-23 00:13:46 -07002702 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
Shyam Iyer71b33462009-01-29 16:12:42 -08002703 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
Thomas Graf677e90e2005-06-23 20:59:51 -07002704 st->frag_idx = 0;
2705 goto next_skb;
Shyam Iyer71b33462009-01-29 16:12:42 -08002706 } else if (st->cur_skb->next) {
2707 st->cur_skb = st->cur_skb->next;
Herbert Xu95e3b242009-01-29 16:07:52 -08002708 st->frag_idx = 0;
Thomas Graf677e90e2005-06-23 20:59:51 -07002709 goto next_skb;
2710 }
2711
2712 return 0;
2713}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002714EXPORT_SYMBOL(skb_seq_read);
Thomas Graf677e90e2005-06-23 20:59:51 -07002715
2716/**
2717 * skb_abort_seq_read - Abort a sequential read of skb data
2718 * @st: state variable
2719 *
2720 * Must be called if skb_seq_read() was not called until it
2721 * returned 0.
2722 */
2723void skb_abort_seq_read(struct skb_seq_state *st)
2724{
2725 if (st->frag_data)
Eric Dumazet51c56b02012-04-05 11:35:15 +02002726 kunmap_atomic(st->frag_data);
Thomas Graf677e90e2005-06-23 20:59:51 -07002727}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002728EXPORT_SYMBOL(skb_abort_seq_read);
Thomas Graf677e90e2005-06-23 20:59:51 -07002729
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002730#define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2731
2732static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2733 struct ts_config *conf,
2734 struct ts_state *state)
2735{
2736 return skb_seq_read(offset, text, TS_SKB_CB(state));
2737}
2738
2739static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2740{
2741 skb_abort_seq_read(TS_SKB_CB(state));
2742}
2743
2744/**
2745 * skb_find_text - Find a text pattern in skb data
2746 * @skb: the buffer to look in
2747 * @from: search offset
2748 * @to: search limit
2749 * @config: textsearch configuration
2750 * @state: uninitialized textsearch state variable
2751 *
2752 * Finds a pattern in the skb data according to the specified
2753 * textsearch configuration. Use textsearch_next() to retrieve
2754 * subsequent occurrences of the pattern. Returns the offset
2755 * to the first occurrence or UINT_MAX if no match was found.
2756 */
2757unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2758 unsigned int to, struct ts_config *config,
2759 struct ts_state *state)
2760{
Phil Oesterf72b9482006-06-26 00:00:57 -07002761 unsigned int ret;
2762
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002763 config->get_next_block = skb_ts_get_next_block;
2764 config->finish = skb_ts_finish;
2765
2766 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2767
Phil Oesterf72b9482006-06-26 00:00:57 -07002768 ret = textsearch_find(config, state);
2769 return (ret <= to - from ? ret : UINT_MAX);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002770}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002771EXPORT_SYMBOL(skb_find_text);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002772
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002773/**
Ben Hutchings2c530402012-07-10 10:55:09 +00002774 * skb_append_datato_frags - append the user data to a skb
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002775 * @sk: sock structure
2776 * @skb: skb structure to be appened with user data.
2777 * @getfrag: call back function to be used for getting the user data
2778 * @from: pointer to user message iov
2779 * @length: length of the iov message
2780 *
2781 * Description: This procedure append the user data in the fragment part
2782 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2783 */
2784int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
Martin Waitzdab96302005-12-05 13:40:12 -08002785 int (*getfrag)(void *from, char *to, int offset,
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002786 int len, int odd, struct sk_buff *skb),
2787 void *from, int length)
2788{
Eric Dumazetb2111722012-12-28 06:06:37 +00002789 int frg_cnt = skb_shinfo(skb)->nr_frags;
2790 int copy;
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002791 int offset = 0;
2792 int ret;
Eric Dumazetb2111722012-12-28 06:06:37 +00002793 struct page_frag *pfrag = &current->task_frag;
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002794
2795 do {
2796 /* Return error if we don't have space for new frag */
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002797 if (frg_cnt >= MAX_SKB_FRAGS)
Eric Dumazetb2111722012-12-28 06:06:37 +00002798 return -EMSGSIZE;
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002799
Eric Dumazetb2111722012-12-28 06:06:37 +00002800 if (!sk_page_frag_refill(sk, pfrag))
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002801 return -ENOMEM;
2802
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002803 /* copy the user data to page */
Eric Dumazetb2111722012-12-28 06:06:37 +00002804 copy = min_t(int, length, pfrag->size - pfrag->offset);
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002805
Eric Dumazetb2111722012-12-28 06:06:37 +00002806 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
2807 offset, copy, 0, skb);
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002808 if (ret < 0)
2809 return -EFAULT;
2810
2811 /* copy was successful so update the size parameters */
Eric Dumazetb2111722012-12-28 06:06:37 +00002812 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
2813 copy);
2814 frg_cnt++;
2815 pfrag->offset += copy;
2816 get_page(pfrag->page);
2817
2818 skb->truesize += copy;
2819 atomic_add(copy, &sk->sk_wmem_alloc);
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002820 skb->len += copy;
2821 skb->data_len += copy;
2822 offset += copy;
2823 length -= copy;
2824
2825 } while (length > 0);
2826
2827 return 0;
2828}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002829EXPORT_SYMBOL(skb_append_datato_frags);
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002830
Herbert Xucbb042f2006-03-20 22:43:56 -08002831/**
2832 * skb_pull_rcsum - pull skb and update receive checksum
2833 * @skb: buffer to update
Herbert Xucbb042f2006-03-20 22:43:56 -08002834 * @len: length of data pulled
2835 *
2836 * This function performs an skb_pull on the packet and updates
Urs Thuermannfee54fa2008-02-12 22:03:25 -08002837 * the CHECKSUM_COMPLETE checksum. It should be used on
Patrick McHardy84fa7932006-08-29 16:44:56 -07002838 * receive path processing instead of skb_pull unless you know
2839 * that the checksum difference is zero (e.g., a valid IP header)
2840 * or you are setting ip_summed to CHECKSUM_NONE.
Herbert Xucbb042f2006-03-20 22:43:56 -08002841 */
2842unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2843{
2844 BUG_ON(len > skb->len);
2845 skb->len -= len;
2846 BUG_ON(skb->len < skb->data_len);
2847 skb_postpull_rcsum(skb, skb->data, len);
2848 return skb->data += len;
2849}
Arnaldo Carvalho de Melof94691a2006-03-20 22:47:55 -08002850EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2851
Herbert Xuf4c50d92006-06-22 03:02:40 -07002852/**
2853 * skb_segment - Perform protocol segmentation on skb.
Michael S. Tsirkindf5771f2014-03-10 18:29:19 +02002854 * @head_skb: buffer to segment
Herbert Xu576a30e2006-06-27 13:22:38 -07002855 * @features: features for the output path (see dev->features)
Herbert Xuf4c50d92006-06-22 03:02:40 -07002856 *
2857 * This function performs segmentation on the given skb. It returns
Ben Hutchings4c821d72008-04-13 21:52:48 -07002858 * a pointer to the first in a list of new skbs for the segments.
2859 * In case of error it returns ERR_PTR(err).
Herbert Xuf4c50d92006-06-22 03:02:40 -07002860 */
Michael S. Tsirkindf5771f2014-03-10 18:29:19 +02002861struct sk_buff *skb_segment(struct sk_buff *head_skb,
2862 netdev_features_t features)
Herbert Xuf4c50d92006-06-22 03:02:40 -07002863{
2864 struct sk_buff *segs = NULL;
2865 struct sk_buff *tail = NULL;
Michael S. Tsirkin1a4ceda2014-03-10 19:27:59 +02002866 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
Michael S. Tsirkindf5771f2014-03-10 18:29:19 +02002867 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
2868 unsigned int mss = skb_shinfo(head_skb)->gso_size;
2869 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
Michael S. Tsirkin1fd819e2014-03-10 19:28:08 +02002870 struct sk_buff *frag_skb = head_skb;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002871 unsigned int offset = doffset;
Michael S. Tsirkindf5771f2014-03-10 18:29:19 +02002872 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002873 unsigned int headroom;
2874 unsigned int len;
Pravin B Shelarec5f0612013-03-07 09:28:01 +00002875 __be16 proto;
2876 bool csum;
Michał Mirosław04ed3e72011-01-24 15:32:47 -08002877 int sg = !!(features & NETIF_F_SG);
Michael S. Tsirkindf5771f2014-03-10 18:29:19 +02002878 int nfrags = skb_shinfo(head_skb)->nr_frags;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002879 int err = -ENOMEM;
2880 int i = 0;
2881 int pos;
Vlad Yasevich53d64712014-03-27 17:26:18 -04002882 int dummy;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002883
Vlad Yasevich53d64712014-03-27 17:26:18 -04002884 proto = skb_network_protocol(head_skb, &dummy);
Pravin B Shelarec5f0612013-03-07 09:28:01 +00002885 if (unlikely(!proto))
2886 return ERR_PTR(-EINVAL);
2887
2888 csum = !!can_checksum_protocol(features, proto);
Michael S. Tsirkindf5771f2014-03-10 18:29:19 +02002889 __skb_push(head_skb, doffset);
2890 headroom = skb_headroom(head_skb);
2891 pos = skb_headlen(head_skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002892
2893 do {
2894 struct sk_buff *nskb;
Michael S. Tsirkin8cb19902014-03-10 18:29:04 +02002895 skb_frag_t *nskb_frag;
Herbert Xuc8884ed2006-10-29 15:59:41 -08002896 int hsize;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002897 int size;
2898
Michael S. Tsirkindf5771f2014-03-10 18:29:19 +02002899 len = head_skb->len - offset;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002900 if (len > mss)
2901 len = mss;
2902
Michael S. Tsirkindf5771f2014-03-10 18:29:19 +02002903 hsize = skb_headlen(head_skb) - offset;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002904 if (hsize < 0)
2905 hsize = 0;
Herbert Xuc8884ed2006-10-29 15:59:41 -08002906 if (hsize > len || !sg)
2907 hsize = len;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002908
Michael S. Tsirkin1a4ceda2014-03-10 19:27:59 +02002909 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
2910 (skb_headlen(list_skb) == len || sg)) {
2911 BUG_ON(skb_headlen(list_skb) > len);
Herbert Xu89319d382008-12-15 23:26:06 -08002912
Herbert Xu9d8506c2013-11-21 11:10:04 -08002913 i = 0;
Michael S. Tsirkin1a4ceda2014-03-10 19:27:59 +02002914 nfrags = skb_shinfo(list_skb)->nr_frags;
2915 frag = skb_shinfo(list_skb)->frags;
Michael S. Tsirkin1fd819e2014-03-10 19:28:08 +02002916 frag_skb = list_skb;
Michael S. Tsirkin1a4ceda2014-03-10 19:27:59 +02002917 pos += skb_headlen(list_skb);
Herbert Xu9d8506c2013-11-21 11:10:04 -08002918
2919 while (pos < offset + len) {
2920 BUG_ON(i >= nfrags);
2921
Michael S. Tsirkin4e1beba2014-03-10 18:29:14 +02002922 size = skb_frag_size(frag);
Herbert Xu9d8506c2013-11-21 11:10:04 -08002923 if (pos + size > offset + len)
2924 break;
2925
2926 i++;
2927 pos += size;
Michael S. Tsirkin4e1beba2014-03-10 18:29:14 +02002928 frag++;
Herbert Xu9d8506c2013-11-21 11:10:04 -08002929 }
2930
Michael S. Tsirkin1a4ceda2014-03-10 19:27:59 +02002931 nskb = skb_clone(list_skb, GFP_ATOMIC);
2932 list_skb = list_skb->next;
Herbert Xu89319d382008-12-15 23:26:06 -08002933
2934 if (unlikely(!nskb))
2935 goto err;
2936
Herbert Xu9d8506c2013-11-21 11:10:04 -08002937 if (unlikely(pskb_trim(nskb, len))) {
2938 kfree_skb(nskb);
2939 goto err;
2940 }
2941
Alexander Duyckec47ea82012-05-04 14:26:56 +00002942 hsize = skb_end_offset(nskb);
Herbert Xu89319d382008-12-15 23:26:06 -08002943 if (skb_cow_head(nskb, doffset + headroom)) {
2944 kfree_skb(nskb);
2945 goto err;
2946 }
2947
Alexander Duyckec47ea82012-05-04 14:26:56 +00002948 nskb->truesize += skb_end_offset(nskb) - hsize;
Herbert Xu89319d382008-12-15 23:26:06 -08002949 skb_release_head_state(nskb);
2950 __skb_push(nskb, doffset);
2951 } else {
Mel Gormanc93bdd02012-07-31 16:44:19 -07002952 nskb = __alloc_skb(hsize + doffset + headroom,
Michael S. Tsirkindf5771f2014-03-10 18:29:19 +02002953 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
Mel Gormanc93bdd02012-07-31 16:44:19 -07002954 NUMA_NO_NODE);
Herbert Xu89319d382008-12-15 23:26:06 -08002955
2956 if (unlikely(!nskb))
2957 goto err;
2958
2959 skb_reserve(nskb, headroom);
2960 __skb_put(nskb, doffset);
2961 }
Herbert Xuf4c50d92006-06-22 03:02:40 -07002962
2963 if (segs)
2964 tail->next = nskb;
2965 else
2966 segs = nskb;
2967 tail = nskb;
2968
Michael S. Tsirkindf5771f2014-03-10 18:29:19 +02002969 __copy_skb_header(nskb, head_skb);
2970 nskb->mac_len = head_skb->mac_len;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002971
Eric Dumazet030737b2013-10-19 11:42:54 -07002972 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
Pravin B Shelar68c33162013-02-14 14:02:41 +00002973
Michael S. Tsirkindf5771f2014-03-10 18:29:19 +02002974 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
Pravin B Shelar68c33162013-02-14 14:02:41 +00002975 nskb->data - tnl_hlen,
2976 doffset + tnl_hlen);
Herbert Xu89319d382008-12-15 23:26:06 -08002977
Herbert Xu9d8506c2013-11-21 11:10:04 -08002978 if (nskb->len == len + doffset)
Simon Horman1cdbcb72013-05-19 15:46:49 +00002979 goto perform_csum_check;
Herbert Xu89319d382008-12-15 23:26:06 -08002980
Herbert Xuf4c50d92006-06-22 03:02:40 -07002981 if (!sg) {
Herbert Xu6f85a122008-08-15 14:55:02 -07002982 nskb->ip_summed = CHECKSUM_NONE;
Michael S. Tsirkindf5771f2014-03-10 18:29:19 +02002983 nskb->csum = skb_copy_and_csum_bits(head_skb, offset,
Herbert Xuf4c50d92006-06-22 03:02:40 -07002984 skb_put(nskb, len),
2985 len, 0);
2986 continue;
2987 }
2988
Michael S. Tsirkin8cb19902014-03-10 18:29:04 +02002989 nskb_frag = skb_shinfo(nskb)->frags;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002990
Michael S. Tsirkindf5771f2014-03-10 18:29:19 +02002991 skb_copy_from_linear_data_offset(head_skb, offset,
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03002992 skb_put(nskb, hsize), hsize);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002993
Michael S. Tsirkindf5771f2014-03-10 18:29:19 +02002994 skb_shinfo(nskb)->tx_flags = skb_shinfo(head_skb)->tx_flags &
2995 SKBTX_SHARED_FRAG;
Eric Dumazetcef401d2013-01-25 20:34:37 +00002996
Herbert Xu9d8506c2013-11-21 11:10:04 -08002997 while (pos < offset + len) {
2998 if (i >= nfrags) {
Michael S. Tsirkin1a4ceda2014-03-10 19:27:59 +02002999 BUG_ON(skb_headlen(list_skb));
Herbert Xu9d8506c2013-11-21 11:10:04 -08003000
3001 i = 0;
Michael S. Tsirkin1a4ceda2014-03-10 19:27:59 +02003002 nfrags = skb_shinfo(list_skb)->nr_frags;
3003 frag = skb_shinfo(list_skb)->frags;
Michael S. Tsirkin1fd819e2014-03-10 19:28:08 +02003004 frag_skb = list_skb;
Herbert Xu9d8506c2013-11-21 11:10:04 -08003005
3006 BUG_ON(!nfrags);
3007
Michael S. Tsirkin1a4ceda2014-03-10 19:27:59 +02003008 list_skb = list_skb->next;
Herbert Xu9d8506c2013-11-21 11:10:04 -08003009 }
3010
3011 if (unlikely(skb_shinfo(nskb)->nr_frags >=
3012 MAX_SKB_FRAGS)) {
3013 net_warn_ratelimited(
3014 "skb_segment: too many frags: %u %u\n",
3015 pos, mss);
3016 goto err;
3017 }
3018
Michael S. Tsirkin1fd819e2014-03-10 19:28:08 +02003019 if (unlikely(skb_orphan_frags(frag_skb, GFP_ATOMIC)))
3020 goto err;
3021
Michael S. Tsirkin4e1beba2014-03-10 18:29:14 +02003022 *nskb_frag = *frag;
Michael S. Tsirkin8cb19902014-03-10 18:29:04 +02003023 __skb_frag_ref(nskb_frag);
3024 size = skb_frag_size(nskb_frag);
Herbert Xuf4c50d92006-06-22 03:02:40 -07003025
3026 if (pos < offset) {
Michael S. Tsirkin8cb19902014-03-10 18:29:04 +02003027 nskb_frag->page_offset += offset - pos;
3028 skb_frag_size_sub(nskb_frag, offset - pos);
Herbert Xuf4c50d92006-06-22 03:02:40 -07003029 }
3030
Herbert Xu89319d382008-12-15 23:26:06 -08003031 skb_shinfo(nskb)->nr_frags++;
Herbert Xuf4c50d92006-06-22 03:02:40 -07003032
3033 if (pos + size <= offset + len) {
3034 i++;
Michael S. Tsirkin4e1beba2014-03-10 18:29:14 +02003035 frag++;
Herbert Xuf4c50d92006-06-22 03:02:40 -07003036 pos += size;
3037 } else {
Michael S. Tsirkin8cb19902014-03-10 18:29:04 +02003038 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
Herbert Xu89319d382008-12-15 23:26:06 -08003039 goto skip_fraglist;
Herbert Xuf4c50d92006-06-22 03:02:40 -07003040 }
3041
Michael S. Tsirkin8cb19902014-03-10 18:29:04 +02003042 nskb_frag++;
Herbert Xuf4c50d92006-06-22 03:02:40 -07003043 }
3044
Herbert Xu89319d382008-12-15 23:26:06 -08003045skip_fraglist:
Herbert Xuf4c50d92006-06-22 03:02:40 -07003046 nskb->data_len = len - hsize;
3047 nskb->len += nskb->data_len;
3048 nskb->truesize += nskb->data_len;
Pravin B Shelarec5f0612013-03-07 09:28:01 +00003049
Simon Horman1cdbcb72013-05-19 15:46:49 +00003050perform_csum_check:
Pravin B Shelarec5f0612013-03-07 09:28:01 +00003051 if (!csum) {
3052 nskb->csum = skb_checksum(nskb, doffset,
3053 nskb->len - doffset, 0);
3054 nskb->ip_summed = CHECKSUM_NONE;
3055 }
Michael S. Tsirkindf5771f2014-03-10 18:29:19 +02003056 } while ((offset += len) < head_skb->len);
Herbert Xuf4c50d92006-06-22 03:02:40 -07003057
3058 return segs;
3059
3060err:
Eric Dumazet289dccb2013-12-20 14:29:08 -08003061 kfree_skb_list(segs);
Herbert Xuf4c50d92006-06-22 03:02:40 -07003062 return ERR_PTR(err);
3063}
Herbert Xuf4c50d92006-06-22 03:02:40 -07003064EXPORT_SYMBOL_GPL(skb_segment);
3065
Herbert Xu71d93b32008-12-15 23:42:33 -08003066int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
3067{
Eric Dumazet8a291112013-10-08 09:02:23 -07003068 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
Herbert Xu67147ba2009-05-26 18:50:22 +00003069 unsigned int offset = skb_gro_offset(skb);
3070 unsigned int headlen = skb_headlen(skb);
Eric Dumazet8a291112013-10-08 09:02:23 -07003071 struct sk_buff *nskb, *lp, *p = *head;
3072 unsigned int len = skb_gro_len(skb);
Eric Dumazet715dc1f2012-05-02 23:33:21 +00003073 unsigned int delta_truesize;
Eric Dumazet8a291112013-10-08 09:02:23 -07003074 unsigned int headroom;
Herbert Xu71d93b32008-12-15 23:42:33 -08003075
Eric Dumazet8a291112013-10-08 09:02:23 -07003076 if (unlikely(p->len + len >= 65536))
Herbert Xu71d93b32008-12-15 23:42:33 -08003077 return -E2BIG;
3078
Eric Dumazet8a291112013-10-08 09:02:23 -07003079 lp = NAPI_GRO_CB(p)->last ?: p;
3080 pinfo = skb_shinfo(lp);
3081
3082 if (headlen <= offset) {
Herbert Xu42da6992009-05-26 18:50:19 +00003083 skb_frag_t *frag;
Herbert Xu66e92fc2009-05-26 18:50:32 +00003084 skb_frag_t *frag2;
Herbert Xu9aaa1562009-05-26 18:50:33 +00003085 int i = skbinfo->nr_frags;
3086 int nr_frags = pinfo->nr_frags + i;
Herbert Xu42da6992009-05-26 18:50:19 +00003087
Herbert Xu66e92fc2009-05-26 18:50:32 +00003088 if (nr_frags > MAX_SKB_FRAGS)
Eric Dumazet8a291112013-10-08 09:02:23 -07003089 goto merge;
Herbert Xu81705ad2009-01-29 14:19:51 +00003090
Eric Dumazet8a291112013-10-08 09:02:23 -07003091 offset -= headlen;
Herbert Xu9aaa1562009-05-26 18:50:33 +00003092 pinfo->nr_frags = nr_frags;
3093 skbinfo->nr_frags = 0;
Herbert Xuf5572062009-01-14 20:40:03 -08003094
Herbert Xu9aaa1562009-05-26 18:50:33 +00003095 frag = pinfo->frags + nr_frags;
3096 frag2 = skbinfo->frags + i;
Herbert Xu66e92fc2009-05-26 18:50:32 +00003097 do {
3098 *--frag = *--frag2;
3099 } while (--i);
3100
3101 frag->page_offset += offset;
Eric Dumazet9e903e02011-10-18 21:00:24 +00003102 skb_frag_size_sub(frag, offset);
Herbert Xu66e92fc2009-05-26 18:50:32 +00003103
Eric Dumazet715dc1f2012-05-02 23:33:21 +00003104 /* all fragments truesize : remove (head size + sk_buff) */
Alexander Duyckec47ea82012-05-04 14:26:56 +00003105 delta_truesize = skb->truesize -
3106 SKB_TRUESIZE(skb_end_offset(skb));
Eric Dumazet715dc1f2012-05-02 23:33:21 +00003107
Herbert Xuf5572062009-01-14 20:40:03 -08003108 skb->truesize -= skb->data_len;
3109 skb->len -= skb->data_len;
3110 skb->data_len = 0;
3111
Eric Dumazet715dc1f2012-05-02 23:33:21 +00003112 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
Herbert Xu5d38a072009-01-04 16:13:40 -08003113 goto done;
Eric Dumazetd7e88832012-04-30 08:10:34 +00003114 } else if (skb->head_frag) {
3115 int nr_frags = pinfo->nr_frags;
3116 skb_frag_t *frag = pinfo->frags + nr_frags;
3117 struct page *page = virt_to_head_page(skb->head);
3118 unsigned int first_size = headlen - offset;
3119 unsigned int first_offset;
3120
3121 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
Eric Dumazet8a291112013-10-08 09:02:23 -07003122 goto merge;
Eric Dumazetd7e88832012-04-30 08:10:34 +00003123
3124 first_offset = skb->data -
3125 (unsigned char *)page_address(page) +
3126 offset;
3127
3128 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3129
3130 frag->page.p = page;
3131 frag->page_offset = first_offset;
3132 skb_frag_size_set(frag, first_size);
3133
3134 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3135 /* We dont need to clear skbinfo->nr_frags here */
3136
Eric Dumazet715dc1f2012-05-02 23:33:21 +00003137 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
Eric Dumazetd7e88832012-04-30 08:10:34 +00003138 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3139 goto done;
Eric Dumazet8a291112013-10-08 09:02:23 -07003140 }
3141 if (pinfo->frag_list)
3142 goto merge;
3143 if (skb_gro_len(p) != pinfo->gso_size)
Herbert Xu69c0cab2009-11-17 05:18:18 -08003144 return -E2BIG;
Herbert Xu71d93b32008-12-15 23:42:33 -08003145
3146 headroom = skb_headroom(p);
Eric Dumazet3d3be432010-09-01 00:50:51 +00003147 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
Herbert Xu71d93b32008-12-15 23:42:33 -08003148 if (unlikely(!nskb))
3149 return -ENOMEM;
3150
3151 __copy_skb_header(nskb, p);
3152 nskb->mac_len = p->mac_len;
3153
3154 skb_reserve(nskb, headroom);
Herbert Xu86911732009-01-29 14:19:50 +00003155 __skb_put(nskb, skb_gro_offset(p));
Herbert Xu71d93b32008-12-15 23:42:33 -08003156
Herbert Xu86911732009-01-29 14:19:50 +00003157 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
Herbert Xu71d93b32008-12-15 23:42:33 -08003158 skb_set_network_header(nskb, skb_network_offset(p));
3159 skb_set_transport_header(nskb, skb_transport_offset(p));
3160
Herbert Xu86911732009-01-29 14:19:50 +00003161 __skb_pull(p, skb_gro_offset(p));
3162 memcpy(skb_mac_header(nskb), skb_mac_header(p),
3163 p->data - skb_mac_header(p));
Herbert Xu71d93b32008-12-15 23:42:33 -08003164
Herbert Xu71d93b32008-12-15 23:42:33 -08003165 skb_shinfo(nskb)->frag_list = p;
Herbert Xu9aaa1562009-05-26 18:50:33 +00003166 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
Herbert Xu622e0ca2010-05-20 23:07:56 -07003167 pinfo->gso_size = 0;
Herbert Xu71d93b32008-12-15 23:42:33 -08003168 skb_header_release(p);
Eric Dumazetc3c7c252012-12-06 13:54:59 +00003169 NAPI_GRO_CB(nskb)->last = p;
Herbert Xu71d93b32008-12-15 23:42:33 -08003170
3171 nskb->data_len += p->len;
Eric Dumazetde8261c2012-02-13 04:09:20 +00003172 nskb->truesize += p->truesize;
Herbert Xu71d93b32008-12-15 23:42:33 -08003173 nskb->len += p->len;
3174
3175 *head = nskb;
3176 nskb->next = p->next;
3177 p->next = NULL;
3178
3179 p = nskb;
3180
3181merge:
Eric Dumazet715dc1f2012-05-02 23:33:21 +00003182 delta_truesize = skb->truesize;
Herbert Xu67147ba2009-05-26 18:50:22 +00003183 if (offset > headlen) {
Michal Schmidtd1dc7ab2011-01-24 12:08:48 +00003184 unsigned int eat = offset - headlen;
3185
3186 skbinfo->frags[0].page_offset += eat;
Eric Dumazet9e903e02011-10-18 21:00:24 +00003187 skb_frag_size_sub(&skbinfo->frags[0], eat);
Michal Schmidtd1dc7ab2011-01-24 12:08:48 +00003188 skb->data_len -= eat;
3189 skb->len -= eat;
Herbert Xu67147ba2009-05-26 18:50:22 +00003190 offset = headlen;
Herbert Xu56035022009-02-05 21:26:52 -08003191 }
3192
Herbert Xu67147ba2009-05-26 18:50:22 +00003193 __skb_pull(skb, offset);
Herbert Xu56035022009-02-05 21:26:52 -08003194
Eric Dumazet8a291112013-10-08 09:02:23 -07003195 if (!NAPI_GRO_CB(p)->last)
3196 skb_shinfo(p)->frag_list = skb;
3197 else
3198 NAPI_GRO_CB(p)->last->next = skb;
Eric Dumazetc3c7c252012-12-06 13:54:59 +00003199 NAPI_GRO_CB(p)->last = skb;
Herbert Xu71d93b32008-12-15 23:42:33 -08003200 skb_header_release(skb);
Eric Dumazet8a291112013-10-08 09:02:23 -07003201 lp = p;
Herbert Xu71d93b32008-12-15 23:42:33 -08003202
Herbert Xu5d38a072009-01-04 16:13:40 -08003203done:
3204 NAPI_GRO_CB(p)->count++;
Herbert Xu37fe4732009-01-17 19:48:13 +00003205 p->data_len += len;
Eric Dumazet715dc1f2012-05-02 23:33:21 +00003206 p->truesize += delta_truesize;
Herbert Xu37fe4732009-01-17 19:48:13 +00003207 p->len += len;
Eric Dumazet8a291112013-10-08 09:02:23 -07003208 if (lp != p) {
3209 lp->data_len += len;
3210 lp->truesize += delta_truesize;
3211 lp->len += len;
3212 }
Herbert Xu71d93b32008-12-15 23:42:33 -08003213 NAPI_GRO_CB(skb)->same_flow = 1;
3214 return 0;
3215}
3216EXPORT_SYMBOL_GPL(skb_gro_receive);
3217
Linus Torvalds1da177e2005-04-16 15:20:36 -07003218void __init skb_init(void)
3219{
3220 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3221 sizeof(struct sk_buff),
3222 0,
Alexey Dobriyane5d679f332006-08-26 19:25:52 -07003223 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
Paul Mundt20c2df82007-07-20 10:11:58 +09003224 NULL);
David S. Millerd179cd12005-08-17 14:57:30 -07003225 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3226 (2*sizeof(struct sk_buff)) +
3227 sizeof(atomic_t),
3228 0,
Alexey Dobriyane5d679f332006-08-26 19:25:52 -07003229 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
Paul Mundt20c2df82007-07-20 10:11:58 +09003230 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003231}
3232
David Howells716ea3a2007-04-02 20:19:53 -07003233/**
3234 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3235 * @skb: Socket buffer containing the buffers to be mapped
3236 * @sg: The scatter-gather list to map into
3237 * @offset: The offset into the buffer's contents to start mapping
3238 * @len: Length of buffer space to be mapped
3239 *
3240 * Fill the specified scatter-gather list with mappings/pointers into a
3241 * region of the buffer space attached to a socket buffer.
3242 */
David S. Miller51c739d2007-10-30 21:29:29 -07003243static int
3244__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
David Howells716ea3a2007-04-02 20:19:53 -07003245{
David S. Miller1a028e52007-04-27 15:21:23 -07003246 int start = skb_headlen(skb);
3247 int i, copy = start - offset;
David S. Millerfbb398a2009-06-09 00:18:59 -07003248 struct sk_buff *frag_iter;
David Howells716ea3a2007-04-02 20:19:53 -07003249 int elt = 0;
3250
3251 if (copy > 0) {
3252 if (copy > len)
3253 copy = len;
Jens Axboe642f1492007-10-24 11:20:47 +02003254 sg_set_buf(sg, skb->data + offset, copy);
David Howells716ea3a2007-04-02 20:19:53 -07003255 elt++;
3256 if ((len -= copy) == 0)
3257 return elt;
3258 offset += copy;
3259 }
3260
3261 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07003262 int end;
David Howells716ea3a2007-04-02 20:19:53 -07003263
Ilpo Järvinen547b7922008-07-25 21:43:18 -07003264 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07003265
Eric Dumazet9e903e02011-10-18 21:00:24 +00003266 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
David Howells716ea3a2007-04-02 20:19:53 -07003267 if ((copy = end - offset) > 0) {
3268 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3269
3270 if (copy > len)
3271 copy = len;
Ian Campbellea2ab692011-08-22 23:44:58 +00003272 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
Jens Axboe642f1492007-10-24 11:20:47 +02003273 frag->page_offset+offset-start);
David Howells716ea3a2007-04-02 20:19:53 -07003274 elt++;
3275 if (!(len -= copy))
3276 return elt;
3277 offset += copy;
3278 }
David S. Miller1a028e52007-04-27 15:21:23 -07003279 start = end;
David Howells716ea3a2007-04-02 20:19:53 -07003280 }
3281
David S. Millerfbb398a2009-06-09 00:18:59 -07003282 skb_walk_frags(skb, frag_iter) {
3283 int end;
David Howells716ea3a2007-04-02 20:19:53 -07003284
David S. Millerfbb398a2009-06-09 00:18:59 -07003285 WARN_ON(start > offset + len);
David Howells716ea3a2007-04-02 20:19:53 -07003286
David S. Millerfbb398a2009-06-09 00:18:59 -07003287 end = start + frag_iter->len;
3288 if ((copy = end - offset) > 0) {
3289 if (copy > len)
3290 copy = len;
3291 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3292 copy);
3293 if ((len -= copy) == 0)
3294 return elt;
3295 offset += copy;
David Howells716ea3a2007-04-02 20:19:53 -07003296 }
David S. Millerfbb398a2009-06-09 00:18:59 -07003297 start = end;
David Howells716ea3a2007-04-02 20:19:53 -07003298 }
3299 BUG_ON(len);
3300 return elt;
3301}
3302
Fan Du25a91d82014-01-18 09:54:23 +08003303/* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
3304 * sglist without mark the sg which contain last skb data as the end.
3305 * So the caller can mannipulate sg list as will when padding new data after
3306 * the first call without calling sg_unmark_end to expend sg list.
3307 *
3308 * Scenario to use skb_to_sgvec_nomark:
3309 * 1. sg_init_table
3310 * 2. skb_to_sgvec_nomark(payload1)
3311 * 3. skb_to_sgvec_nomark(payload2)
3312 *
3313 * This is equivalent to:
3314 * 1. sg_init_table
3315 * 2. skb_to_sgvec(payload1)
3316 * 3. sg_unmark_end
3317 * 4. skb_to_sgvec(payload2)
3318 *
3319 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
3320 * is more preferable.
3321 */
3322int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
3323 int offset, int len)
3324{
3325 return __skb_to_sgvec(skb, sg, offset, len);
3326}
3327EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
3328
David S. Miller51c739d2007-10-30 21:29:29 -07003329int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3330{
3331 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3332
Jens Axboec46f2332007-10-31 12:06:37 +01003333 sg_mark_end(&sg[nsg - 1]);
David S. Miller51c739d2007-10-30 21:29:29 -07003334
3335 return nsg;
3336}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08003337EXPORT_SYMBOL_GPL(skb_to_sgvec);
David S. Miller51c739d2007-10-30 21:29:29 -07003338
David Howells716ea3a2007-04-02 20:19:53 -07003339/**
3340 * skb_cow_data - Check that a socket buffer's data buffers are writable
3341 * @skb: The socket buffer to check.
3342 * @tailbits: Amount of trailing space to be added
3343 * @trailer: Returned pointer to the skb where the @tailbits space begins
3344 *
3345 * Make sure that the data buffers attached to a socket buffer are
3346 * writable. If they are not, private copies are made of the data buffers
3347 * and the socket buffer is set to use these instead.
3348 *
3349 * If @tailbits is given, make sure that there is space to write @tailbits
3350 * bytes of data beyond current end of socket buffer. @trailer will be
3351 * set to point to the skb in which this space begins.
3352 *
3353 * The number of scatterlist elements required to completely map the
3354 * COW'd and extended socket buffer will be returned.
3355 */
3356int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3357{
3358 int copyflag;
3359 int elt;
3360 struct sk_buff *skb1, **skb_p;
3361
3362 /* If skb is cloned or its head is paged, reallocate
3363 * head pulling out all the pages (pages are considered not writable
3364 * at the moment even if they are anonymous).
3365 */
3366 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3367 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3368 return -ENOMEM;
3369
3370 /* Easy case. Most of packets will go this way. */
David S. Miller21dc3302010-08-23 00:13:46 -07003371 if (!skb_has_frag_list(skb)) {
David Howells716ea3a2007-04-02 20:19:53 -07003372 /* A little of trouble, not enough of space for trailer.
3373 * This should not happen, when stack is tuned to generate
3374 * good frames. OK, on miss we reallocate and reserve even more
3375 * space, 128 bytes is fair. */
3376
3377 if (skb_tailroom(skb) < tailbits &&
3378 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3379 return -ENOMEM;
3380
3381 /* Voila! */
3382 *trailer = skb;
3383 return 1;
3384 }
3385
3386 /* Misery. We are in troubles, going to mincer fragments... */
3387
3388 elt = 1;
3389 skb_p = &skb_shinfo(skb)->frag_list;
3390 copyflag = 0;
3391
3392 while ((skb1 = *skb_p) != NULL) {
3393 int ntail = 0;
3394
3395 /* The fragment is partially pulled by someone,
3396 * this can happen on input. Copy it and everything
3397 * after it. */
3398
3399 if (skb_shared(skb1))
3400 copyflag = 1;
3401
3402 /* If the skb is the last, worry about trailer. */
3403
3404 if (skb1->next == NULL && tailbits) {
3405 if (skb_shinfo(skb1)->nr_frags ||
David S. Miller21dc3302010-08-23 00:13:46 -07003406 skb_has_frag_list(skb1) ||
David Howells716ea3a2007-04-02 20:19:53 -07003407 skb_tailroom(skb1) < tailbits)
3408 ntail = tailbits + 128;
3409 }
3410
3411 if (copyflag ||
3412 skb_cloned(skb1) ||
3413 ntail ||
3414 skb_shinfo(skb1)->nr_frags ||
David S. Miller21dc3302010-08-23 00:13:46 -07003415 skb_has_frag_list(skb1)) {
David Howells716ea3a2007-04-02 20:19:53 -07003416 struct sk_buff *skb2;
3417
3418 /* Fuck, we are miserable poor guys... */
3419 if (ntail == 0)
3420 skb2 = skb_copy(skb1, GFP_ATOMIC);
3421 else
3422 skb2 = skb_copy_expand(skb1,
3423 skb_headroom(skb1),
3424 ntail,
3425 GFP_ATOMIC);
3426 if (unlikely(skb2 == NULL))
3427 return -ENOMEM;
3428
3429 if (skb1->sk)
3430 skb_set_owner_w(skb2, skb1->sk);
3431
3432 /* Looking around. Are we still alive?
3433 * OK, link new skb, drop old one */
3434
3435 skb2->next = skb1->next;
3436 *skb_p = skb2;
3437 kfree_skb(skb1);
3438 skb1 = skb2;
3439 }
3440 elt++;
3441 *trailer = skb1;
3442 skb_p = &skb1->next;
3443 }
3444
3445 return elt;
3446}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08003447EXPORT_SYMBOL_GPL(skb_cow_data);
David Howells716ea3a2007-04-02 20:19:53 -07003448
Eric Dumazetb1faf562010-05-31 23:44:05 -07003449static void sock_rmem_free(struct sk_buff *skb)
3450{
3451 struct sock *sk = skb->sk;
3452
3453 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3454}
3455
3456/*
3457 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3458 */
3459int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3460{
3461 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
Eric Dumazet95c96172012-04-15 05:58:06 +00003462 (unsigned int)sk->sk_rcvbuf)
Eric Dumazetb1faf562010-05-31 23:44:05 -07003463 return -ENOMEM;
3464
3465 skb_orphan(skb);
3466 skb->sk = sk;
3467 skb->destructor = sock_rmem_free;
3468 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3469
Eric Dumazetabb57ea2011-05-18 02:21:31 -04003470 /* before exiting rcu section, make sure dst is refcounted */
3471 skb_dst_force(skb);
3472
Eric Dumazetb1faf562010-05-31 23:44:05 -07003473 skb_queue_tail(&sk->sk_error_queue, skb);
3474 if (!sock_flag(sk, SOCK_DEAD))
David S. Miller676d2362014-04-11 16:15:36 -04003475 sk->sk_data_ready(sk);
Eric Dumazetb1faf562010-05-31 23:44:05 -07003476 return 0;
3477}
3478EXPORT_SYMBOL(sock_queue_err_skb);
3479
Patrick Ohlyac45f602009-02-12 05:03:37 +00003480void skb_tstamp_tx(struct sk_buff *orig_skb,
3481 struct skb_shared_hwtstamps *hwtstamps)
3482{
3483 struct sock *sk = orig_skb->sk;
3484 struct sock_exterr_skb *serr;
3485 struct sk_buff *skb;
3486 int err;
3487
3488 if (!sk)
3489 return;
3490
Patrick Ohlyac45f602009-02-12 05:03:37 +00003491 if (hwtstamps) {
Willem de Bruijn2e313962013-04-23 00:39:28 +00003492 *skb_hwtstamps(orig_skb) =
Patrick Ohlyac45f602009-02-12 05:03:37 +00003493 *hwtstamps;
3494 } else {
3495 /*
3496 * no hardware time stamps available,
Oliver Hartkopp2244d072010-08-17 08:59:14 +00003497 * so keep the shared tx_flags and only
Patrick Ohlyac45f602009-02-12 05:03:37 +00003498 * store software time stamp
3499 */
Willem de Bruijn2e313962013-04-23 00:39:28 +00003500 orig_skb->tstamp = ktime_get_real();
Patrick Ohlyac45f602009-02-12 05:03:37 +00003501 }
3502
Willem de Bruijn2e313962013-04-23 00:39:28 +00003503 skb = skb_clone(orig_skb, GFP_ATOMIC);
3504 if (!skb)
3505 return;
3506
Patrick Ohlyac45f602009-02-12 05:03:37 +00003507 serr = SKB_EXT_ERR(skb);
3508 memset(serr, 0, sizeof(*serr));
3509 serr->ee.ee_errno = ENOMSG;
3510 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
Eric Dumazet29030372010-05-29 00:20:48 -07003511
Patrick Ohlyac45f602009-02-12 05:03:37 +00003512 err = sock_queue_err_skb(sk, skb);
Eric Dumazet29030372010-05-29 00:20:48 -07003513
Patrick Ohlyac45f602009-02-12 05:03:37 +00003514 if (err)
3515 kfree_skb(skb);
3516}
3517EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3518
Johannes Berg6e3e9392011-11-09 10:15:42 +01003519void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3520{
3521 struct sock *sk = skb->sk;
3522 struct sock_exterr_skb *serr;
3523 int err;
3524
3525 skb->wifi_acked_valid = 1;
3526 skb->wifi_acked = acked;
3527
3528 serr = SKB_EXT_ERR(skb);
3529 memset(serr, 0, sizeof(*serr));
3530 serr->ee.ee_errno = ENOMSG;
3531 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3532
3533 err = sock_queue_err_skb(sk, skb);
3534 if (err)
3535 kfree_skb(skb);
3536}
3537EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3538
Patrick Ohlyac45f602009-02-12 05:03:37 +00003539
Rusty Russellf35d9d82008-02-04 23:49:54 -05003540/**
3541 * skb_partial_csum_set - set up and verify partial csum values for packet
3542 * @skb: the skb to set
3543 * @start: the number of bytes after skb->data to start checksumming.
3544 * @off: the offset from start to place the checksum.
3545 *
3546 * For untrusted partially-checksummed packets, we need to make sure the values
3547 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3548 *
3549 * This function checks and sets those values and skb->ip_summed: if this
3550 * returns false you should drop the packet.
3551 */
3552bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3553{
Herbert Xu5ff8dda2009-06-04 01:22:01 +00003554 if (unlikely(start > skb_headlen(skb)) ||
3555 unlikely((int)start + off > skb_headlen(skb) - 2)) {
Joe Perchese87cc472012-05-13 21:56:26 +00003556 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3557 start, off, skb_headlen(skb));
Rusty Russellf35d9d82008-02-04 23:49:54 -05003558 return false;
3559 }
3560 skb->ip_summed = CHECKSUM_PARTIAL;
3561 skb->csum_start = skb_headroom(skb) + start;
3562 skb->csum_offset = off;
Jason Wange5d5dec2013-03-26 23:11:20 +00003563 skb_set_transport_header(skb, start);
Rusty Russellf35d9d82008-02-04 23:49:54 -05003564 return true;
3565}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08003566EXPORT_SYMBOL_GPL(skb_partial_csum_set);
Rusty Russellf35d9d82008-02-04 23:49:54 -05003567
Paul Durranted1f50c2014-01-09 10:02:46 +00003568static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
3569 unsigned int max)
3570{
3571 if (skb_headlen(skb) >= len)
3572 return 0;
3573
3574 /* If we need to pullup then pullup to the max, so we
3575 * won't need to do it again.
3576 */
3577 if (max > skb->len)
3578 max = skb->len;
3579
3580 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
3581 return -ENOMEM;
3582
3583 if (skb_headlen(skb) < len)
3584 return -EPROTO;
3585
3586 return 0;
3587}
3588
Jan Beulichf9708b42014-03-11 13:56:05 +00003589#define MAX_TCP_HDR_LEN (15 * 4)
3590
3591static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
3592 typeof(IPPROTO_IP) proto,
3593 unsigned int off)
3594{
3595 switch (proto) {
3596 int err;
3597
3598 case IPPROTO_TCP:
3599 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
3600 off + MAX_TCP_HDR_LEN);
3601 if (!err && !skb_partial_csum_set(skb, off,
3602 offsetof(struct tcphdr,
3603 check)))
3604 err = -EPROTO;
3605 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
3606
3607 case IPPROTO_UDP:
3608 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
3609 off + sizeof(struct udphdr));
3610 if (!err && !skb_partial_csum_set(skb, off,
3611 offsetof(struct udphdr,
3612 check)))
3613 err = -EPROTO;
3614 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
3615 }
3616
3617 return ERR_PTR(-EPROTO);
3618}
3619
Paul Durranted1f50c2014-01-09 10:02:46 +00003620/* This value should be large enough to cover a tagged ethernet header plus
3621 * maximally sized IP and TCP or UDP headers.
3622 */
3623#define MAX_IP_HDR_LEN 128
3624
Jan Beulichf9708b42014-03-11 13:56:05 +00003625static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
Paul Durranted1f50c2014-01-09 10:02:46 +00003626{
3627 unsigned int off;
3628 bool fragment;
Jan Beulichf9708b42014-03-11 13:56:05 +00003629 __sum16 *csum;
Paul Durranted1f50c2014-01-09 10:02:46 +00003630 int err;
3631
3632 fragment = false;
3633
3634 err = skb_maybe_pull_tail(skb,
3635 sizeof(struct iphdr),
3636 MAX_IP_HDR_LEN);
3637 if (err < 0)
3638 goto out;
3639
3640 if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
3641 fragment = true;
3642
3643 off = ip_hdrlen(skb);
3644
3645 err = -EPROTO;
3646
3647 if (fragment)
3648 goto out;
3649
Jan Beulichf9708b42014-03-11 13:56:05 +00003650 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
3651 if (IS_ERR(csum))
3652 return PTR_ERR(csum);
Paul Durranted1f50c2014-01-09 10:02:46 +00003653
Jan Beulichf9708b42014-03-11 13:56:05 +00003654 if (recalculate)
3655 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3656 ip_hdr(skb)->daddr,
3657 skb->len - off,
3658 ip_hdr(skb)->protocol, 0);
Paul Durranted1f50c2014-01-09 10:02:46 +00003659 err = 0;
3660
3661out:
3662 return err;
3663}
3664
3665/* This value should be large enough to cover a tagged ethernet header plus
3666 * an IPv6 header, all options, and a maximal TCP or UDP header.
3667 */
3668#define MAX_IPV6_HDR_LEN 256
3669
3670#define OPT_HDR(type, skb, off) \
3671 (type *)(skb_network_header(skb) + (off))
3672
3673static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
3674{
3675 int err;
3676 u8 nexthdr;
3677 unsigned int off;
3678 unsigned int len;
3679 bool fragment;
3680 bool done;
Jan Beulichf9708b42014-03-11 13:56:05 +00003681 __sum16 *csum;
Paul Durranted1f50c2014-01-09 10:02:46 +00003682
3683 fragment = false;
3684 done = false;
3685
3686 off = sizeof(struct ipv6hdr);
3687
3688 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
3689 if (err < 0)
3690 goto out;
3691
3692 nexthdr = ipv6_hdr(skb)->nexthdr;
3693
3694 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
3695 while (off <= len && !done) {
3696 switch (nexthdr) {
3697 case IPPROTO_DSTOPTS:
3698 case IPPROTO_HOPOPTS:
3699 case IPPROTO_ROUTING: {
3700 struct ipv6_opt_hdr *hp;
3701
3702 err = skb_maybe_pull_tail(skb,
3703 off +
3704 sizeof(struct ipv6_opt_hdr),
3705 MAX_IPV6_HDR_LEN);
3706 if (err < 0)
3707 goto out;
3708
3709 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
3710 nexthdr = hp->nexthdr;
3711 off += ipv6_optlen(hp);
3712 break;
3713 }
3714 case IPPROTO_AH: {
3715 struct ip_auth_hdr *hp;
3716
3717 err = skb_maybe_pull_tail(skb,
3718 off +
3719 sizeof(struct ip_auth_hdr),
3720 MAX_IPV6_HDR_LEN);
3721 if (err < 0)
3722 goto out;
3723
3724 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
3725 nexthdr = hp->nexthdr;
3726 off += ipv6_authlen(hp);
3727 break;
3728 }
3729 case IPPROTO_FRAGMENT: {
3730 struct frag_hdr *hp;
3731
3732 err = skb_maybe_pull_tail(skb,
3733 off +
3734 sizeof(struct frag_hdr),
3735 MAX_IPV6_HDR_LEN);
3736 if (err < 0)
3737 goto out;
3738
3739 hp = OPT_HDR(struct frag_hdr, skb, off);
3740
3741 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
3742 fragment = true;
3743
3744 nexthdr = hp->nexthdr;
3745 off += sizeof(struct frag_hdr);
3746 break;
3747 }
3748 default:
3749 done = true;
3750 break;
3751 }
3752 }
3753
3754 err = -EPROTO;
3755
3756 if (!done || fragment)
3757 goto out;
3758
Jan Beulichf9708b42014-03-11 13:56:05 +00003759 csum = skb_checksum_setup_ip(skb, nexthdr, off);
3760 if (IS_ERR(csum))
3761 return PTR_ERR(csum);
Paul Durranted1f50c2014-01-09 10:02:46 +00003762
Jan Beulichf9708b42014-03-11 13:56:05 +00003763 if (recalculate)
3764 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3765 &ipv6_hdr(skb)->daddr,
3766 skb->len - off, nexthdr, 0);
Paul Durranted1f50c2014-01-09 10:02:46 +00003767 err = 0;
3768
3769out:
3770 return err;
3771}
3772
3773/**
3774 * skb_checksum_setup - set up partial checksum offset
3775 * @skb: the skb to set up
3776 * @recalculate: if true the pseudo-header checksum will be recalculated
3777 */
3778int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
3779{
3780 int err;
3781
3782 switch (skb->protocol) {
3783 case htons(ETH_P_IP):
Jan Beulichf9708b42014-03-11 13:56:05 +00003784 err = skb_checksum_setup_ipv4(skb, recalculate);
Paul Durranted1f50c2014-01-09 10:02:46 +00003785 break;
3786
3787 case htons(ETH_P_IPV6):
3788 err = skb_checksum_setup_ipv6(skb, recalculate);
3789 break;
3790
3791 default:
3792 err = -EPROTO;
3793 break;
3794 }
3795
3796 return err;
3797}
3798EXPORT_SYMBOL(skb_checksum_setup);
3799
Ben Hutchings4497b072008-06-19 16:22:28 -07003800void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3801{
Joe Perchese87cc472012-05-13 21:56:26 +00003802 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
3803 skb->dev->name);
Ben Hutchings4497b072008-06-19 16:22:28 -07003804}
Ben Hutchings4497b072008-06-19 16:22:28 -07003805EXPORT_SYMBOL(__skb_warn_lro_forwarding);
Eric Dumazetbad43ca2012-05-19 03:02:02 +00003806
3807void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
3808{
Eric Dumazet3d861f62012-10-22 09:03:40 +00003809 if (head_stolen) {
3810 skb_release_head_state(skb);
Eric Dumazetbad43ca2012-05-19 03:02:02 +00003811 kmem_cache_free(skbuff_head_cache, skb);
Eric Dumazet3d861f62012-10-22 09:03:40 +00003812 } else {
Eric Dumazetbad43ca2012-05-19 03:02:02 +00003813 __kfree_skb(skb);
Eric Dumazet3d861f62012-10-22 09:03:40 +00003814 }
Eric Dumazetbad43ca2012-05-19 03:02:02 +00003815}
3816EXPORT_SYMBOL(kfree_skb_partial);
3817
3818/**
3819 * skb_try_coalesce - try to merge skb to prior one
3820 * @to: prior buffer
3821 * @from: buffer to add
3822 * @fragstolen: pointer to boolean
Randy Dunlapc6c4b972012-06-08 14:01:44 +00003823 * @delta_truesize: how much more was allocated than was requested
Eric Dumazetbad43ca2012-05-19 03:02:02 +00003824 */
3825bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
3826 bool *fragstolen, int *delta_truesize)
3827{
3828 int i, delta, len = from->len;
3829
3830 *fragstolen = false;
3831
3832 if (skb_cloned(to))
3833 return false;
3834
3835 if (len <= skb_tailroom(to)) {
3836 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
3837 *delta_truesize = 0;
3838 return true;
3839 }
3840
3841 if (skb_has_frag_list(to) || skb_has_frag_list(from))
3842 return false;
3843
3844 if (skb_headlen(from) != 0) {
3845 struct page *page;
3846 unsigned int offset;
3847
3848 if (skb_shinfo(to)->nr_frags +
3849 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
3850 return false;
3851
3852 if (skb_head_is_locked(from))
3853 return false;
3854
3855 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3856
3857 page = virt_to_head_page(from->head);
3858 offset = from->data - (unsigned char *)page_address(page);
3859
3860 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
3861 page, offset, skb_headlen(from));
3862 *fragstolen = true;
3863 } else {
3864 if (skb_shinfo(to)->nr_frags +
3865 skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
3866 return false;
3867
Weiping Panf4b549a2012-09-28 20:15:30 +00003868 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
Eric Dumazetbad43ca2012-05-19 03:02:02 +00003869 }
3870
3871 WARN_ON_ONCE(delta < len);
3872
3873 memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
3874 skb_shinfo(from)->frags,
3875 skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
3876 skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
3877
3878 if (!skb_cloned(from))
3879 skb_shinfo(from)->nr_frags = 0;
3880
Li RongQing8ea853f2012-09-18 16:53:21 +00003881 /* if the skb is not cloned this does nothing
3882 * since we set nr_frags to 0.
3883 */
Eric Dumazetbad43ca2012-05-19 03:02:02 +00003884 for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
3885 skb_frag_ref(from, i);
3886
3887 to->truesize += delta;
3888 to->len += len;
3889 to->data_len += len;
3890
3891 *delta_truesize = delta;
3892 return true;
3893}
3894EXPORT_SYMBOL(skb_try_coalesce);
Nicolas Dichtel621e84d2013-06-26 16:11:27 +02003895
3896/**
Nicolas Dichtel8b27f272013-09-02 15:34:56 +02003897 * skb_scrub_packet - scrub an skb
Nicolas Dichtel621e84d2013-06-26 16:11:27 +02003898 *
3899 * @skb: buffer to clean
Nicolas Dichtel8b27f272013-09-02 15:34:56 +02003900 * @xnet: packet is crossing netns
Nicolas Dichtel621e84d2013-06-26 16:11:27 +02003901 *
Nicolas Dichtel8b27f272013-09-02 15:34:56 +02003902 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
3903 * into/from a tunnel. Some information have to be cleared during these
3904 * operations.
3905 * skb_scrub_packet can also be used to clean a skb before injecting it in
3906 * another namespace (@xnet == true). We have to clear all information in the
3907 * skb that could impact namespace isolation.
Nicolas Dichtel621e84d2013-06-26 16:11:27 +02003908 */
Nicolas Dichtel8b27f272013-09-02 15:34:56 +02003909void skb_scrub_packet(struct sk_buff *skb, bool xnet)
Nicolas Dichtel621e84d2013-06-26 16:11:27 +02003910{
Nicolas Dichtel8b27f272013-09-02 15:34:56 +02003911 if (xnet)
3912 skb_orphan(skb);
Nicolas Dichtel621e84d2013-06-26 16:11:27 +02003913 skb->tstamp.tv64 = 0;
3914 skb->pkt_type = PACKET_HOST;
3915 skb->skb_iif = 0;
Hannes Frederic Sowa239c78d2013-12-05 23:29:19 +01003916 skb->local_df = 0;
Nicolas Dichtel621e84d2013-06-26 16:11:27 +02003917 skb_dst_drop(skb);
3918 skb->mark = 0;
3919 secpath_reset(skb);
3920 nf_reset(skb);
3921 nf_reset_trace(skb);
3922}
3923EXPORT_SYMBOL_GPL(skb_scrub_packet);
Florian Westphalde960aa2014-01-26 10:58:16 +01003924
3925/**
3926 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
3927 *
3928 * @skb: GSO skb
3929 *
3930 * skb_gso_transport_seglen is used to determine the real size of the
3931 * individual segments, including Layer4 headers (TCP/UDP).
3932 *
3933 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
3934 */
3935unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
3936{
3937 const struct skb_shared_info *shinfo = skb_shinfo(skb);
Florian Westphalde960aa2014-01-26 10:58:16 +01003938
3939 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
Florian Westphal6d39d582014-04-09 10:28:50 +02003940 return tcp_hdrlen(skb) + shinfo->gso_size;
3941
3942 /* UFO sets gso_size to the size of the fragmentation
3943 * payload, i.e. the size of the L4 (UDP) header is already
3944 * accounted for.
3945 */
3946 return shinfo->gso_size;
Florian Westphalde960aa2014-01-26 10:58:16 +01003947}
3948EXPORT_SYMBOL_GPL(skb_gso_transport_seglen);