blob: 5976ef0846bdda08db6289bb91f05a63b85e6e3a [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);
Igor Maravića3bf7ae2011-12-12 02:58:22 +0000710#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
Herbert Xudec18812007-10-14 00:37:30 -0700711 new->nf_trace = old->nf_trace;
712#endif
713#ifdef CONFIG_NET_SCHED
714 new->tc_index = old->tc_index;
715#ifdef CONFIG_NET_CLS_ACT
716 new->tc_verd = old->tc_verd;
717#endif
718#endif
Patrick McHardy86a9bad2013-04-19 02:04:30 +0000719 new->vlan_proto = old->vlan_proto;
Patrick McHardy6aa895b02008-07-14 22:49:06 -0700720 new->vlan_tci = old->vlan_tci;
721
Herbert Xudec18812007-10-14 00:37:30 -0700722 skb_copy_secmark(new, old);
Eliezer Tamir06021292013-06-10 11:39:50 +0300723
Cong Wange0d10952013-08-01 11:10:25 +0800724#ifdef CONFIG_NET_RX_BUSY_POLL
Eliezer Tamir06021292013-06-10 11:39:50 +0300725 new->napi_id = old->napi_id;
726#endif
Herbert Xudec18812007-10-14 00:37:30 -0700727}
728
Herbert Xu82c49a32009-05-22 22:11:37 +0000729/*
730 * You should not add any new code to this function. Add it to
731 * __copy_skb_header above instead.
732 */
Herbert Xue0053ec2007-10-14 00:37:52 -0700733static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735#define C(x) n->x = skb->x
736
737 n->next = n->prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 n->sk = NULL;
Herbert Xudec18812007-10-14 00:37:30 -0700739 __copy_skb_header(n, skb);
740
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 C(len);
742 C(data_len);
Alexey Dobriyan3e6b3b22007-03-16 15:00:46 -0700743 C(mac_len);
Patrick McHardy334a8132007-06-25 04:35:20 -0700744 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
Paul Moore02f1c892008-01-07 21:56:41 -0800745 n->cloned = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 n->nohdr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 n->destructor = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 C(tail);
749 C(end);
Paul Moore02f1c892008-01-07 21:56:41 -0800750 C(head);
Eric Dumazetd3836f22012-04-27 00:33:38 +0000751 C(head_frag);
Paul Moore02f1c892008-01-07 21:56:41 -0800752 C(data);
753 C(truesize);
754 atomic_set(&n->users, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755
756 atomic_inc(&(skb_shinfo(skb)->dataref));
757 skb->cloned = 1;
758
759 return n;
Herbert Xue0053ec2007-10-14 00:37:52 -0700760#undef C
761}
762
763/**
764 * skb_morph - morph one skb into another
765 * @dst: the skb to receive the contents
766 * @src: the skb to supply the contents
767 *
768 * This is identical to skb_clone except that the target skb is
769 * supplied by the user.
770 *
771 * The target skb is returned upon exit.
772 */
773struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
774{
Herbert Xu2d4baff2007-11-26 23:11:19 +0800775 skb_release_all(dst);
Herbert Xue0053ec2007-10-14 00:37:52 -0700776 return __skb_clone(dst, src);
777}
778EXPORT_SYMBOL_GPL(skb_morph);
779
Ben Hutchings2c530402012-07-10 10:55:09 +0000780/**
781 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
Michael S. Tsirkin48c83012011-08-31 08:03:29 +0000782 * @skb: the skb to modify
783 * @gfp_mask: allocation priority
784 *
785 * This must be called on SKBTX_DEV_ZEROCOPY skb.
786 * It will copy all frags into kernel and drop the reference
787 * to userspace pages.
788 *
789 * If this function is called from an interrupt gfp_mask() must be
790 * %GFP_ATOMIC.
791 *
792 * Returns 0 on success or a negative error code on failure
793 * to allocate kernel memory to copy to.
794 */
795int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
Shirley Maa6686f22011-07-06 12:22:12 +0000796{
797 int i;
798 int num_frags = skb_shinfo(skb)->nr_frags;
799 struct page *page, *head = NULL;
800 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
801
802 for (i = 0; i < num_frags; i++) {
803 u8 *vaddr;
804 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
805
Krishna Kumar02756ed2012-07-17 02:05:29 +0000806 page = alloc_page(gfp_mask);
Shirley Maa6686f22011-07-06 12:22:12 +0000807 if (!page) {
808 while (head) {
Sunghan Suh40dadff2013-07-12 16:17:23 +0900809 struct page *next = (struct page *)page_private(head);
Shirley Maa6686f22011-07-06 12:22:12 +0000810 put_page(head);
811 head = next;
812 }
813 return -ENOMEM;
814 }
Eric Dumazet51c56b02012-04-05 11:35:15 +0200815 vaddr = kmap_atomic(skb_frag_page(f));
Shirley Maa6686f22011-07-06 12:22:12 +0000816 memcpy(page_address(page),
Eric Dumazet9e903e02011-10-18 21:00:24 +0000817 vaddr + f->page_offset, skb_frag_size(f));
Eric Dumazet51c56b02012-04-05 11:35:15 +0200818 kunmap_atomic(vaddr);
Sunghan Suh40dadff2013-07-12 16:17:23 +0900819 set_page_private(page, (unsigned long)head);
Shirley Maa6686f22011-07-06 12:22:12 +0000820 head = page;
821 }
822
823 /* skb frags release userspace buffers */
Krishna Kumar02756ed2012-07-17 02:05:29 +0000824 for (i = 0; i < num_frags; i++)
Ian Campbella8605c62011-10-19 23:01:49 +0000825 skb_frag_unref(skb, i);
Shirley Maa6686f22011-07-06 12:22:12 +0000826
Michael S. Tsirkine19d6762012-11-01 09:16:22 +0000827 uarg->callback(uarg, false);
Shirley Maa6686f22011-07-06 12:22:12 +0000828
829 /* skb frags point to kernel buffers */
Krishna Kumar02756ed2012-07-17 02:05:29 +0000830 for (i = num_frags - 1; i >= 0; i--) {
831 __skb_fill_page_desc(skb, i, head, 0,
832 skb_shinfo(skb)->frags[i].size);
Sunghan Suh40dadff2013-07-12 16:17:23 +0900833 head = (struct page *)page_private(head);
Shirley Maa6686f22011-07-06 12:22:12 +0000834 }
Michael S. Tsirkin48c83012011-08-31 08:03:29 +0000835
836 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
Shirley Maa6686f22011-07-06 12:22:12 +0000837 return 0;
838}
Michael S. Tsirkindcc0fb72012-07-20 09:23:20 +0000839EXPORT_SYMBOL_GPL(skb_copy_ubufs);
Shirley Maa6686f22011-07-06 12:22:12 +0000840
Herbert Xue0053ec2007-10-14 00:37:52 -0700841/**
842 * skb_clone - duplicate an sk_buff
843 * @skb: buffer to clone
844 * @gfp_mask: allocation priority
845 *
846 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
847 * copies share the same packet data but not structure. The new
848 * buffer has a reference count of 1. If the allocation fails the
849 * function returns %NULL otherwise the new buffer is returned.
850 *
851 * If this function is called from an interrupt gfp_mask() must be
852 * %GFP_ATOMIC.
853 */
854
855struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
856{
857 struct sk_buff *n;
858
Michael S. Tsirkin70008aa2012-07-20 09:23:10 +0000859 if (skb_orphan_frags(skb, gfp_mask))
860 return NULL;
Shirley Maa6686f22011-07-06 12:22:12 +0000861
Herbert Xue0053ec2007-10-14 00:37:52 -0700862 n = skb + 1;
863 if (skb->fclone == SKB_FCLONE_ORIG &&
864 n->fclone == SKB_FCLONE_UNAVAILABLE) {
865 atomic_t *fclone_ref = (atomic_t *) (n + 1);
866 n->fclone = SKB_FCLONE_CLONE;
867 atomic_inc(fclone_ref);
868 } else {
Mel Gormanc93bdd02012-07-31 16:44:19 -0700869 if (skb_pfmemalloc(skb))
870 gfp_mask |= __GFP_MEMALLOC;
871
Herbert Xue0053ec2007-10-14 00:37:52 -0700872 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
873 if (!n)
874 return NULL;
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200875
876 kmemcheck_annotate_bitfield(n, flags1);
877 kmemcheck_annotate_bitfield(n, flags2);
Herbert Xue0053ec2007-10-14 00:37:52 -0700878 n->fclone = SKB_FCLONE_UNAVAILABLE;
879 }
880
881 return __skb_clone(n, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800883EXPORT_SYMBOL(skb_clone);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884
Pravin B Shelarf5b17292013-03-07 13:21:40 +0000885static void skb_headers_offset_update(struct sk_buff *skb, int off)
886{
Eric Dumazet030737b2013-10-19 11:42:54 -0700887 /* Only adjust this if it actually is csum_start rather than csum */
888 if (skb->ip_summed == CHECKSUM_PARTIAL)
889 skb->csum_start += off;
Pravin B Shelarf5b17292013-03-07 13:21:40 +0000890 /* {transport,network,mac}_header and tail are relative to skb->head */
891 skb->transport_header += off;
892 skb->network_header += off;
893 if (skb_mac_header_was_set(skb))
894 skb->mac_header += off;
895 skb->inner_transport_header += off;
896 skb->inner_network_header += off;
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +0000897 skb->inner_mac_header += off;
Pravin B Shelarf5b17292013-03-07 13:21:40 +0000898}
899
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
901{
Herbert Xudec18812007-10-14 00:37:30 -0700902 __copy_skb_header(new, old);
903
Herbert Xu79671682006-06-22 02:40:14 -0700904 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
905 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
906 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907}
908
Mel Gormanc93bdd02012-07-31 16:44:19 -0700909static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
910{
911 if (skb_pfmemalloc(skb))
912 return SKB_ALLOC_RX;
913 return 0;
914}
915
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916/**
917 * skb_copy - create private copy of an sk_buff
918 * @skb: buffer to copy
919 * @gfp_mask: allocation priority
920 *
921 * Make a copy of both an &sk_buff and its data. This is used when the
922 * caller wishes to modify the data and needs a private copy of the
923 * data to alter. Returns %NULL on failure or the pointer to the buffer
924 * on success. The returned buffer has a reference count of 1.
925 *
926 * As by-product this function converts non-linear &sk_buff to linear
927 * one, so that &sk_buff becomes completely private and caller is allowed
928 * to modify all the data of returned buffer. This means that this
929 * function is not recommended for use in circumstances when only
930 * header is going to be modified. Use pskb_copy() instead.
931 */
932
Al Virodd0fc662005-10-07 07:46:04 +0100933struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934{
Eric Dumazet6602ceb2010-09-01 05:25:10 +0000935 int headerlen = skb_headroom(skb);
Alexander Duyckec47ea82012-05-04 14:26:56 +0000936 unsigned int size = skb_end_offset(skb) + skb->data_len;
Mel Gormanc93bdd02012-07-31 16:44:19 -0700937 struct sk_buff *n = __alloc_skb(size, gfp_mask,
938 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
Eric Dumazet6602ceb2010-09-01 05:25:10 +0000939
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 if (!n)
941 return NULL;
942
943 /* Set the data pointer */
944 skb_reserve(n, headerlen);
945 /* Set the tail pointer and length */
946 skb_put(n, skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947
948 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
949 BUG();
950
951 copy_skb_header(n, skb);
952 return n;
953}
David S. Millerb4ac530fc2009-02-10 02:09:24 -0800954EXPORT_SYMBOL(skb_copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955
956/**
Eric Dumazet117632e2011-12-03 21:39:53 +0000957 * __pskb_copy - create copy of an sk_buff with private head.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 * @skb: buffer to copy
Eric Dumazet117632e2011-12-03 21:39:53 +0000959 * @headroom: headroom of new skb
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 * @gfp_mask: allocation priority
961 *
962 * Make a copy of both an &sk_buff and part of its data, located
963 * in header. Fragmented data remain shared. This is used when
964 * the caller wishes to modify only header of &sk_buff and needs
965 * private copy of the header to alter. Returns %NULL on failure
966 * or the pointer to the buffer on success.
967 * The returned buffer has a reference count of 1.
968 */
969
Eric Dumazet117632e2011-12-03 21:39:53 +0000970struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971{
Eric Dumazet117632e2011-12-03 21:39:53 +0000972 unsigned int size = skb_headlen(skb) + headroom;
Mel Gormanc93bdd02012-07-31 16:44:19 -0700973 struct sk_buff *n = __alloc_skb(size, gfp_mask,
974 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
Eric Dumazet6602ceb2010-09-01 05:25:10 +0000975
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 if (!n)
977 goto out;
978
979 /* Set the data pointer */
Eric Dumazet117632e2011-12-03 21:39:53 +0000980 skb_reserve(n, headroom);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 /* Set the tail pointer and length */
982 skb_put(n, skb_headlen(skb));
983 /* Copy the bytes */
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -0300984 skb_copy_from_linear_data(skb, n->data, n->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985
Herbert Xu25f484a2006-11-07 14:57:15 -0800986 n->truesize += skb->data_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 n->data_len = skb->data_len;
988 n->len = skb->len;
989
990 if (skb_shinfo(skb)->nr_frags) {
991 int i;
992
Michael S. Tsirkin70008aa2012-07-20 09:23:10 +0000993 if (skb_orphan_frags(skb, gfp_mask)) {
994 kfree_skb(n);
995 n = NULL;
996 goto out;
Shirley Maa6686f22011-07-06 12:22:12 +0000997 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
999 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
Ian Campbellea2ab692011-08-22 23:44:58 +00001000 skb_frag_ref(skb, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 }
1002 skb_shinfo(n)->nr_frags = i;
1003 }
1004
David S. Miller21dc3302010-08-23 00:13:46 -07001005 if (skb_has_frag_list(skb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1007 skb_clone_fraglist(n);
1008 }
1009
1010 copy_skb_header(n, skb);
1011out:
1012 return n;
1013}
Eric Dumazet117632e2011-12-03 21:39:53 +00001014EXPORT_SYMBOL(__pskb_copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015
1016/**
1017 * pskb_expand_head - reallocate header of &sk_buff
1018 * @skb: buffer to reallocate
1019 * @nhead: room to add at head
1020 * @ntail: room to add at tail
1021 * @gfp_mask: allocation priority
1022 *
Mathias Krausebc323832013-11-07 14:18:26 +01001023 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1024 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 * reference count of 1. Returns zero in the case of success or error,
1026 * if expansion failed. In the last case, &sk_buff is not changed.
1027 *
1028 * All the pointers pointing into skb header may change and must be
1029 * reloaded after call to this function.
1030 */
1031
Victor Fusco86a76ca2005-07-08 14:57:47 -07001032int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
Al Virodd0fc662005-10-07 07:46:04 +01001033 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034{
1035 int i;
1036 u8 *data;
Alexander Duyckec47ea82012-05-04 14:26:56 +00001037 int size = nhead + skb_end_offset(skb) + ntail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 long off;
1039
Herbert Xu4edd87a2008-10-01 07:09:38 -07001040 BUG_ON(nhead < 0);
1041
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 if (skb_shared(skb))
1043 BUG();
1044
1045 size = SKB_DATA_ALIGN(size);
1046
Mel Gormanc93bdd02012-07-31 16:44:19 -07001047 if (skb_pfmemalloc(skb))
1048 gfp_mask |= __GFP_MEMALLOC;
1049 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1050 gfp_mask, NUMA_NO_NODE, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 if (!data)
1052 goto nodata;
Eric Dumazet87151b82012-04-10 20:08:39 +00001053 size = SKB_WITH_OVERHEAD(ksize(data));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054
1055 /* Copy only real data... and, alas, header. This should be
Eric Dumazet6602ceb2010-09-01 05:25:10 +00001056 * optimized for the cases when header is void.
1057 */
1058 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1059
1060 memcpy((struct skb_shared_info *)(data + size),
1061 skb_shinfo(skb),
Eric Dumazetfed66382010-07-22 19:09:08 +00001062 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063
Alexander Duyck3e245912012-05-04 14:26:51 +00001064 /*
1065 * if shinfo is shared we must drop the old head gracefully, but if it
1066 * is not we can just drop the old head and let the existing refcount
1067 * be since all we did is relocate the values
1068 */
1069 if (skb_cloned(skb)) {
Shirley Maa6686f22011-07-06 12:22:12 +00001070 /* copy this zero copy skb frags */
Michael S. Tsirkin70008aa2012-07-20 09:23:10 +00001071 if (skb_orphan_frags(skb, gfp_mask))
1072 goto nofrags;
Eric Dumazet1fd63042010-09-02 23:09:32 +00001073 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
Ian Campbellea2ab692011-08-22 23:44:58 +00001074 skb_frag_ref(skb, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075
Eric Dumazet1fd63042010-09-02 23:09:32 +00001076 if (skb_has_frag_list(skb))
1077 skb_clone_fraglist(skb);
1078
1079 skb_release_data(skb);
Alexander Duyck3e245912012-05-04 14:26:51 +00001080 } else {
1081 skb_free_head(skb);
Eric Dumazet1fd63042010-09-02 23:09:32 +00001082 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 off = (data + nhead) - skb->head;
1084
1085 skb->head = data;
Eric Dumazetd3836f22012-04-27 00:33:38 +00001086 skb->head_frag = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 skb->data += off;
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001088#ifdef NET_SKBUFF_DATA_USES_OFFSET
1089 skb->end = size;
Patrick McHardy56eb8882007-04-09 11:45:04 -07001090 off = nhead;
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001091#else
1092 skb->end = skb->head + size;
Patrick McHardy56eb8882007-04-09 11:45:04 -07001093#endif
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001094 skb->tail += off;
Peter Pan(潘卫平)b41abb42013-06-06 21:27:21 +08001095 skb_headers_offset_update(skb, nhead);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 skb->cloned = 0;
Patrick McHardy334a8132007-06-25 04:35:20 -07001097 skb->hdr_len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 skb->nohdr = 0;
1099 atomic_set(&skb_shinfo(skb)->dataref, 1);
1100 return 0;
1101
Shirley Maa6686f22011-07-06 12:22:12 +00001102nofrags:
1103 kfree(data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104nodata:
1105 return -ENOMEM;
1106}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001107EXPORT_SYMBOL(pskb_expand_head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108
1109/* Make private copy of skb with writable head and some headroom */
1110
1111struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1112{
1113 struct sk_buff *skb2;
1114 int delta = headroom - skb_headroom(skb);
1115
1116 if (delta <= 0)
1117 skb2 = pskb_copy(skb, GFP_ATOMIC);
1118 else {
1119 skb2 = skb_clone(skb, GFP_ATOMIC);
1120 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1121 GFP_ATOMIC)) {
1122 kfree_skb(skb2);
1123 skb2 = NULL;
1124 }
1125 }
1126 return skb2;
1127}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001128EXPORT_SYMBOL(skb_realloc_headroom);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129
1130/**
1131 * skb_copy_expand - copy and expand sk_buff
1132 * @skb: buffer to copy
1133 * @newheadroom: new free bytes at head
1134 * @newtailroom: new free bytes at tail
1135 * @gfp_mask: allocation priority
1136 *
1137 * Make a copy of both an &sk_buff and its data and while doing so
1138 * allocate additional space.
1139 *
1140 * This is used when the caller wishes to modify the data and needs a
1141 * private copy of the data to alter as well as more space for new fields.
1142 * Returns %NULL on failure or the pointer to the buffer
1143 * on success. The returned buffer has a reference count of 1.
1144 *
1145 * You must pass %GFP_ATOMIC as the allocation priority if this function
1146 * is called from an interrupt.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 */
1148struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
Victor Fusco86a76ca2005-07-08 14:57:47 -07001149 int newheadroom, int newtailroom,
Al Virodd0fc662005-10-07 07:46:04 +01001150 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151{
1152 /*
1153 * Allocate the copy buffer
1154 */
Mel Gormanc93bdd02012-07-31 16:44:19 -07001155 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1156 gfp_mask, skb_alloc_rx_flag(skb),
1157 NUMA_NO_NODE);
Patrick McHardyefd1e8d2007-04-10 18:30:09 -07001158 int oldheadroom = skb_headroom(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 int head_copy_len, head_copy_off;
1160
1161 if (!n)
1162 return NULL;
1163
1164 skb_reserve(n, newheadroom);
1165
1166 /* Set the tail pointer and length */
1167 skb_put(n, skb->len);
1168
Patrick McHardyefd1e8d2007-04-10 18:30:09 -07001169 head_copy_len = oldheadroom;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 head_copy_off = 0;
1171 if (newheadroom <= head_copy_len)
1172 head_copy_len = newheadroom;
1173 else
1174 head_copy_off = newheadroom - head_copy_len;
1175
1176 /* Copy the linear header and data. */
1177 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1178 skb->len + head_copy_len))
1179 BUG();
1180
1181 copy_skb_header(n, skb);
1182
Eric Dumazet030737b2013-10-19 11:42:54 -07001183 skb_headers_offset_update(n, newheadroom - oldheadroom);
Patrick McHardyefd1e8d2007-04-10 18:30:09 -07001184
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 return n;
1186}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001187EXPORT_SYMBOL(skb_copy_expand);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188
1189/**
1190 * skb_pad - zero pad the tail of an skb
1191 * @skb: buffer to pad
1192 * @pad: space to pad
1193 *
1194 * Ensure that a buffer is followed by a padding area that is zero
1195 * filled. Used by network drivers which may DMA or transfer data
1196 * beyond the buffer end onto the wire.
1197 *
Herbert Xu5b057c62006-06-23 02:06:41 -07001198 * May return error in out of memory cases. The skb is freed on error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 */
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001200
Herbert Xu5b057c62006-06-23 02:06:41 -07001201int skb_pad(struct sk_buff *skb, int pad)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202{
Herbert Xu5b057c62006-06-23 02:06:41 -07001203 int err;
1204 int ntail;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001205
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 /* If the skbuff is non linear tailroom is always zero.. */
Herbert Xu5b057c62006-06-23 02:06:41 -07001207 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 memset(skb->data+skb->len, 0, pad);
Herbert Xu5b057c62006-06-23 02:06:41 -07001209 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 }
Herbert Xu5b057c62006-06-23 02:06:41 -07001211
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001212 ntail = skb->data_len + pad - (skb->end - skb->tail);
Herbert Xu5b057c62006-06-23 02:06:41 -07001213 if (likely(skb_cloned(skb) || ntail > 0)) {
1214 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1215 if (unlikely(err))
1216 goto free_skb;
1217 }
1218
1219 /* FIXME: The use of this function with non-linear skb's really needs
1220 * to be audited.
1221 */
1222 err = skb_linearize(skb);
1223 if (unlikely(err))
1224 goto free_skb;
1225
1226 memset(skb->data + skb->len, 0, pad);
1227 return 0;
1228
1229free_skb:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 kfree_skb(skb);
Herbert Xu5b057c62006-06-23 02:06:41 -07001231 return err;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001232}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001233EXPORT_SYMBOL(skb_pad);
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001234
Ilpo Järvinen0dde3e12008-03-27 17:43:41 -07001235/**
Mathias Krause0c7ddf32013-11-07 14:18:24 +01001236 * pskb_put - add data to the tail of a potentially fragmented buffer
1237 * @skb: start of the buffer to use
1238 * @tail: tail fragment of the buffer to use
1239 * @len: amount of data to add
1240 *
1241 * This function extends the used data area of the potentially
1242 * fragmented buffer. @tail must be the last fragment of @skb -- or
1243 * @skb itself. If this would exceed the total buffer size the kernel
1244 * will panic. A pointer to the first byte of the extra data is
1245 * returned.
1246 */
1247
1248unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1249{
1250 if (tail != skb) {
1251 skb->data_len += len;
1252 skb->len += len;
1253 }
1254 return skb_put(tail, len);
1255}
1256EXPORT_SYMBOL_GPL(pskb_put);
1257
1258/**
Ilpo Järvinen0dde3e12008-03-27 17:43:41 -07001259 * skb_put - add data to a buffer
1260 * @skb: buffer to use
1261 * @len: amount of data to add
1262 *
1263 * This function extends the used data area of the buffer. If this would
1264 * exceed the total buffer size the kernel will panic. A pointer to the
1265 * first byte of the extra data is returned.
1266 */
1267unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1268{
1269 unsigned char *tmp = skb_tail_pointer(skb);
1270 SKB_LINEAR_ASSERT(skb);
1271 skb->tail += len;
1272 skb->len += len;
1273 if (unlikely(skb->tail > skb->end))
1274 skb_over_panic(skb, len, __builtin_return_address(0));
1275 return tmp;
1276}
1277EXPORT_SYMBOL(skb_put);
1278
Ilpo Järvinen6be8ac22008-03-27 17:47:24 -07001279/**
Ilpo Järvinenc2aa2702008-03-27 17:52:40 -07001280 * skb_push - add data to the start of a buffer
1281 * @skb: buffer to use
1282 * @len: amount of data to add
1283 *
1284 * This function extends the used data area of the buffer at the buffer
1285 * start. If this would exceed the total buffer headroom the kernel will
1286 * panic. A pointer to the first byte of the extra data is returned.
1287 */
1288unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1289{
1290 skb->data -= len;
1291 skb->len += len;
1292 if (unlikely(skb->data<skb->head))
1293 skb_under_panic(skb, len, __builtin_return_address(0));
1294 return skb->data;
1295}
1296EXPORT_SYMBOL(skb_push);
1297
1298/**
Ilpo Järvinen6be8ac22008-03-27 17:47:24 -07001299 * skb_pull - remove data from the start of a buffer
1300 * @skb: buffer to use
1301 * @len: amount of data to remove
1302 *
1303 * This function removes data from the start of a buffer, returning
1304 * the memory to the headroom. A pointer to the next data in the buffer
1305 * is returned. Once the data has been pulled future pushes will overwrite
1306 * the old data.
1307 */
1308unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1309{
David S. Miller47d29642010-05-02 02:21:44 -07001310 return skb_pull_inline(skb, len);
Ilpo Järvinen6be8ac22008-03-27 17:47:24 -07001311}
1312EXPORT_SYMBOL(skb_pull);
1313
Ilpo Järvinen419ae742008-03-27 17:54:01 -07001314/**
1315 * skb_trim - remove end from a buffer
1316 * @skb: buffer to alter
1317 * @len: new length
1318 *
1319 * Cut the length of a buffer down by removing data from the tail. If
1320 * the buffer is already under the length specified it is not modified.
1321 * The skb must be linear.
1322 */
1323void skb_trim(struct sk_buff *skb, unsigned int len)
1324{
1325 if (skb->len > len)
1326 __skb_trim(skb, len);
1327}
1328EXPORT_SYMBOL(skb_trim);
1329
Herbert Xu3cc0e872006-06-09 16:13:38 -07001330/* Trims skb to length len. It can change skb pointers.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 */
1332
Herbert Xu3cc0e872006-06-09 16:13:38 -07001333int ___pskb_trim(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334{
Herbert Xu27b437c2006-07-13 19:26:39 -07001335 struct sk_buff **fragp;
1336 struct sk_buff *frag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 int offset = skb_headlen(skb);
1338 int nfrags = skb_shinfo(skb)->nr_frags;
1339 int i;
Herbert Xu27b437c2006-07-13 19:26:39 -07001340 int err;
1341
1342 if (skb_cloned(skb) &&
1343 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1344 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001346 i = 0;
1347 if (offset >= len)
1348 goto drop_pages;
1349
1350 for (; i < nfrags; i++) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00001351 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
Herbert Xu27b437c2006-07-13 19:26:39 -07001352
1353 if (end < len) {
1354 offset = end;
1355 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 }
Herbert Xu27b437c2006-07-13 19:26:39 -07001357
Eric Dumazet9e903e02011-10-18 21:00:24 +00001358 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
Herbert Xu27b437c2006-07-13 19:26:39 -07001359
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001360drop_pages:
Herbert Xu27b437c2006-07-13 19:26:39 -07001361 skb_shinfo(skb)->nr_frags = i;
1362
1363 for (; i < nfrags; i++)
Ian Campbellea2ab692011-08-22 23:44:58 +00001364 skb_frag_unref(skb, i);
Herbert Xu27b437c2006-07-13 19:26:39 -07001365
David S. Miller21dc3302010-08-23 00:13:46 -07001366 if (skb_has_frag_list(skb))
Herbert Xu27b437c2006-07-13 19:26:39 -07001367 skb_drop_fraglist(skb);
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001368 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 }
1370
Herbert Xu27b437c2006-07-13 19:26:39 -07001371 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1372 fragp = &frag->next) {
1373 int end = offset + frag->len;
1374
1375 if (skb_shared(frag)) {
1376 struct sk_buff *nfrag;
1377
1378 nfrag = skb_clone(frag, GFP_ATOMIC);
1379 if (unlikely(!nfrag))
1380 return -ENOMEM;
1381
1382 nfrag->next = frag->next;
Eric Dumazet85bb2a62012-04-19 02:24:53 +00001383 consume_skb(frag);
Herbert Xu27b437c2006-07-13 19:26:39 -07001384 frag = nfrag;
1385 *fragp = frag;
1386 }
1387
1388 if (end < len) {
1389 offset = end;
1390 continue;
1391 }
1392
1393 if (end > len &&
1394 unlikely((err = pskb_trim(frag, len - offset))))
1395 return err;
1396
1397 if (frag->next)
1398 skb_drop_list(&frag->next);
1399 break;
1400 }
1401
Herbert Xuf4d26fb2006-07-30 20:20:28 -07001402done:
Herbert Xu27b437c2006-07-13 19:26:39 -07001403 if (len > skb_headlen(skb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 skb->data_len -= skb->len - len;
1405 skb->len = len;
1406 } else {
Herbert Xu27b437c2006-07-13 19:26:39 -07001407 skb->len = len;
1408 skb->data_len = 0;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001409 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 }
1411
1412 return 0;
1413}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001414EXPORT_SYMBOL(___pskb_trim);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415
1416/**
1417 * __pskb_pull_tail - advance tail of skb header
1418 * @skb: buffer to reallocate
1419 * @delta: number of bytes to advance tail
1420 *
1421 * The function makes a sense only on a fragmented &sk_buff,
1422 * it expands header moving its tail forward and copying necessary
1423 * data from fragmented part.
1424 *
1425 * &sk_buff MUST have reference count of 1.
1426 *
1427 * Returns %NULL (and &sk_buff does not change) if pull failed
1428 * or value of new tail of skb in the case of success.
1429 *
1430 * All the pointers pointing into skb header may change and must be
1431 * reloaded after call to this function.
1432 */
1433
1434/* Moves tail of skb head forward, copying data from fragmented part,
1435 * when it is necessary.
1436 * 1. It may fail due to malloc failure.
1437 * 2. It may change skb pointers.
1438 *
1439 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1440 */
1441unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1442{
1443 /* If skb has not enough free space at tail, get new one
1444 * plus 128 bytes for future expansions. If we have enough
1445 * room at tail, reallocate without expansion only if skb is cloned.
1446 */
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001447 int i, k, eat = (skb->tail + delta) - skb->end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448
1449 if (eat > 0 || skb_cloned(skb)) {
1450 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1451 GFP_ATOMIC))
1452 return NULL;
1453 }
1454
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001455 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 BUG();
1457
1458 /* Optimization: no fragments, no reasons to preestimate
1459 * size of pulled pages. Superb.
1460 */
David S. Miller21dc3302010-08-23 00:13:46 -07001461 if (!skb_has_frag_list(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 goto pull_pages;
1463
1464 /* Estimate size of pulled pages. */
1465 eat = delta;
1466 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00001467 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1468
1469 if (size >= eat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 goto pull_pages;
Eric Dumazet9e903e02011-10-18 21:00:24 +00001471 eat -= size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 }
1473
1474 /* If we need update frag list, we are in troubles.
1475 * Certainly, it possible to add an offset to skb data,
1476 * but taking into account that pulling is expected to
1477 * be very rare operation, it is worth to fight against
1478 * further bloating skb head and crucify ourselves here instead.
1479 * Pure masohism, indeed. 8)8)
1480 */
1481 if (eat) {
1482 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1483 struct sk_buff *clone = NULL;
1484 struct sk_buff *insp = NULL;
1485
1486 do {
Kris Katterjohn09a62662006-01-08 22:24:28 -08001487 BUG_ON(!list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488
1489 if (list->len <= eat) {
1490 /* Eaten as whole. */
1491 eat -= list->len;
1492 list = list->next;
1493 insp = list;
1494 } else {
1495 /* Eaten partially. */
1496
1497 if (skb_shared(list)) {
1498 /* Sucks! We need to fork list. :-( */
1499 clone = skb_clone(list, GFP_ATOMIC);
1500 if (!clone)
1501 return NULL;
1502 insp = list->next;
1503 list = clone;
1504 } else {
1505 /* This may be pulled without
1506 * problems. */
1507 insp = list;
1508 }
1509 if (!pskb_pull(list, eat)) {
Wei Yongjunf3fbbe02009-02-25 00:37:32 +00001510 kfree_skb(clone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 return NULL;
1512 }
1513 break;
1514 }
1515 } while (eat);
1516
1517 /* Free pulled out fragments. */
1518 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1519 skb_shinfo(skb)->frag_list = list->next;
1520 kfree_skb(list);
1521 }
1522 /* And insert new clone at head. */
1523 if (clone) {
1524 clone->next = list;
1525 skb_shinfo(skb)->frag_list = clone;
1526 }
1527 }
1528 /* Success! Now we may commit changes to skb data. */
1529
1530pull_pages:
1531 eat = delta;
1532 k = 0;
1533 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00001534 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1535
1536 if (size <= eat) {
Ian Campbellea2ab692011-08-22 23:44:58 +00001537 skb_frag_unref(skb, i);
Eric Dumazet9e903e02011-10-18 21:00:24 +00001538 eat -= size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 } else {
1540 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1541 if (eat) {
1542 skb_shinfo(skb)->frags[k].page_offset += eat;
Eric Dumazet9e903e02011-10-18 21:00:24 +00001543 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 eat = 0;
1545 }
1546 k++;
1547 }
1548 }
1549 skb_shinfo(skb)->nr_frags = k;
1550
1551 skb->tail += delta;
1552 skb->data_len -= delta;
1553
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001554 return skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001556EXPORT_SYMBOL(__pskb_pull_tail);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557
Eric Dumazet22019b12011-07-29 18:37:31 +00001558/**
1559 * skb_copy_bits - copy bits from skb to kernel buffer
1560 * @skb: source skb
1561 * @offset: offset in source
1562 * @to: destination buffer
1563 * @len: number of bytes to copy
1564 *
1565 * Copy the specified number of bytes from the source skb to the
1566 * destination buffer.
1567 *
1568 * CAUTION ! :
1569 * If its prototype is ever changed,
1570 * check arch/{*}/net/{*}.S files,
1571 * since it is called from BPF assembly code.
1572 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1574{
David S. Miller1a028e52007-04-27 15:21:23 -07001575 int start = skb_headlen(skb);
David S. Millerfbb398a2009-06-09 00:18:59 -07001576 struct sk_buff *frag_iter;
1577 int i, copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578
1579 if (offset > (int)skb->len - len)
1580 goto fault;
1581
1582 /* Copy header. */
David S. Miller1a028e52007-04-27 15:21:23 -07001583 if ((copy = start - offset) > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 if (copy > len)
1585 copy = len;
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03001586 skb_copy_from_linear_data_offset(skb, offset, to, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 if ((len -= copy) == 0)
1588 return 0;
1589 offset += copy;
1590 to += copy;
1591 }
1592
1593 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07001594 int end;
Eric Dumazet51c56b02012-04-05 11:35:15 +02001595 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001597 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07001598
Eric Dumazet51c56b02012-04-05 11:35:15 +02001599 end = start + skb_frag_size(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 if ((copy = end - offset) > 0) {
1601 u8 *vaddr;
1602
1603 if (copy > len)
1604 copy = len;
1605
Eric Dumazet51c56b02012-04-05 11:35:15 +02001606 vaddr = kmap_atomic(skb_frag_page(f));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 memcpy(to,
Eric Dumazet51c56b02012-04-05 11:35:15 +02001608 vaddr + f->page_offset + offset - start,
1609 copy);
1610 kunmap_atomic(vaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611
1612 if ((len -= copy) == 0)
1613 return 0;
1614 offset += copy;
1615 to += copy;
1616 }
David S. Miller1a028e52007-04-27 15:21:23 -07001617 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 }
1619
David S. Millerfbb398a2009-06-09 00:18:59 -07001620 skb_walk_frags(skb, frag_iter) {
1621 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622
David S. Millerfbb398a2009-06-09 00:18:59 -07001623 WARN_ON(start > offset + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624
David S. Millerfbb398a2009-06-09 00:18:59 -07001625 end = start + frag_iter->len;
1626 if ((copy = end - offset) > 0) {
1627 if (copy > len)
1628 copy = len;
1629 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1630 goto fault;
1631 if ((len -= copy) == 0)
1632 return 0;
1633 offset += copy;
1634 to += copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 }
David S. Millerfbb398a2009-06-09 00:18:59 -07001636 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 }
Shirley Maa6686f22011-07-06 12:22:12 +00001638
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 if (!len)
1640 return 0;
1641
1642fault:
1643 return -EFAULT;
1644}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08001645EXPORT_SYMBOL(skb_copy_bits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646
Jens Axboe9c55e012007-11-06 23:30:13 -08001647/*
1648 * Callback from splice_to_pipe(), if we need to release some pages
1649 * at the end of the spd in case we error'ed out in filling the pipe.
1650 */
1651static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1652{
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001653 put_page(spd->pages[i]);
1654}
Jens Axboe9c55e012007-11-06 23:30:13 -08001655
David S. Millera108d5f2012-04-23 23:06:11 -04001656static struct page *linear_to_page(struct page *page, unsigned int *len,
1657 unsigned int *offset,
Eric Dumazet18aafc62013-01-11 14:46:37 +00001658 struct sock *sk)
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001659{
Eric Dumazet5640f762012-09-23 23:04:42 +00001660 struct page_frag *pfrag = sk_page_frag(sk);
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001661
Eric Dumazet5640f762012-09-23 23:04:42 +00001662 if (!sk_page_frag_refill(sk, pfrag))
1663 return NULL;
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001664
Eric Dumazet5640f762012-09-23 23:04:42 +00001665 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001666
Eric Dumazet5640f762012-09-23 23:04:42 +00001667 memcpy(page_address(pfrag->page) + pfrag->offset,
1668 page_address(page) + *offset, *len);
1669 *offset = pfrag->offset;
1670 pfrag->offset += *len;
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001671
Eric Dumazet5640f762012-09-23 23:04:42 +00001672 return pfrag->page;
Jens Axboe9c55e012007-11-06 23:30:13 -08001673}
1674
Eric Dumazet41c73a02012-04-22 12:26:16 +00001675static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1676 struct page *page,
1677 unsigned int offset)
1678{
1679 return spd->nr_pages &&
1680 spd->pages[spd->nr_pages - 1] == page &&
1681 (spd->partial[spd->nr_pages - 1].offset +
1682 spd->partial[spd->nr_pages - 1].len == offset);
1683}
1684
Jens Axboe9c55e012007-11-06 23:30:13 -08001685/*
1686 * Fill page/offset/length into spd, if it can hold more pages.
1687 */
David S. Millera108d5f2012-04-23 23:06:11 -04001688static bool spd_fill_page(struct splice_pipe_desc *spd,
1689 struct pipe_inode_info *pipe, struct page *page,
1690 unsigned int *len, unsigned int offset,
Eric Dumazet18aafc62013-01-11 14:46:37 +00001691 bool linear,
David S. Millera108d5f2012-04-23 23:06:11 -04001692 struct sock *sk)
Jens Axboe9c55e012007-11-06 23:30:13 -08001693{
Eric Dumazet41c73a02012-04-22 12:26:16 +00001694 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
David S. Millera108d5f2012-04-23 23:06:11 -04001695 return true;
Jens Axboe9c55e012007-11-06 23:30:13 -08001696
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001697 if (linear) {
Eric Dumazet18aafc62013-01-11 14:46:37 +00001698 page = linear_to_page(page, len, &offset, sk);
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001699 if (!page)
David S. Millera108d5f2012-04-23 23:06:11 -04001700 return true;
Eric Dumazet41c73a02012-04-22 12:26:16 +00001701 }
1702 if (spd_can_coalesce(spd, page, offset)) {
1703 spd->partial[spd->nr_pages - 1].len += *len;
David S. Millera108d5f2012-04-23 23:06:11 -04001704 return false;
Eric Dumazet41c73a02012-04-22 12:26:16 +00001705 }
1706 get_page(page);
Jens Axboe9c55e012007-11-06 23:30:13 -08001707 spd->pages[spd->nr_pages] = page;
Jarek Poplawski4fb66992009-02-01 00:41:42 -08001708 spd->partial[spd->nr_pages].len = *len;
Jens Axboe9c55e012007-11-06 23:30:13 -08001709 spd->partial[spd->nr_pages].offset = offset;
Jens Axboe9c55e012007-11-06 23:30:13 -08001710 spd->nr_pages++;
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001711
David S. Millera108d5f2012-04-23 23:06:11 -04001712 return false;
Jens Axboe9c55e012007-11-06 23:30:13 -08001713}
1714
David S. Millera108d5f2012-04-23 23:06:11 -04001715static bool __splice_segment(struct page *page, unsigned int poff,
1716 unsigned int plen, unsigned int *off,
Eric Dumazet18aafc62013-01-11 14:46:37 +00001717 unsigned int *len,
Eric Dumazetd7ccf7c2012-04-23 23:35:04 -04001718 struct splice_pipe_desc *spd, bool linear,
David S. Millera108d5f2012-04-23 23:06:11 -04001719 struct sock *sk,
1720 struct pipe_inode_info *pipe)
Octavian Purdila2870c432008-07-15 00:49:11 -07001721{
1722 if (!*len)
David S. Millera108d5f2012-04-23 23:06:11 -04001723 return true;
Octavian Purdila2870c432008-07-15 00:49:11 -07001724
1725 /* skip this segment if already processed */
1726 if (*off >= plen) {
1727 *off -= plen;
David S. Millera108d5f2012-04-23 23:06:11 -04001728 return false;
Octavian Purdiladb43a282008-06-27 17:27:21 -07001729 }
Jens Axboe9c55e012007-11-06 23:30:13 -08001730
Octavian Purdila2870c432008-07-15 00:49:11 -07001731 /* ignore any bits we already processed */
Eric Dumazet9ca1b222013-01-05 21:31:18 +00001732 poff += *off;
1733 plen -= *off;
1734 *off = 0;
Octavian Purdila2870c432008-07-15 00:49:11 -07001735
Eric Dumazet18aafc62013-01-11 14:46:37 +00001736 do {
1737 unsigned int flen = min(*len, plen);
Octavian Purdila2870c432008-07-15 00:49:11 -07001738
Eric Dumazet18aafc62013-01-11 14:46:37 +00001739 if (spd_fill_page(spd, pipe, page, &flen, poff,
1740 linear, sk))
1741 return true;
1742 poff += flen;
1743 plen -= flen;
1744 *len -= flen;
1745 } while (*len && plen);
Octavian Purdila2870c432008-07-15 00:49:11 -07001746
David S. Millera108d5f2012-04-23 23:06:11 -04001747 return false;
Octavian Purdila2870c432008-07-15 00:49:11 -07001748}
1749
1750/*
David S. Millera108d5f2012-04-23 23:06:11 -04001751 * Map linear and fragment data from the skb to spd. It reports true if the
Octavian Purdila2870c432008-07-15 00:49:11 -07001752 * pipe is full or if we already spliced the requested length.
1753 */
David S. Millera108d5f2012-04-23 23:06:11 -04001754static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1755 unsigned int *offset, unsigned int *len,
1756 struct splice_pipe_desc *spd, struct sock *sk)
Octavian Purdila2870c432008-07-15 00:49:11 -07001757{
1758 int seg;
1759
Eric Dumazet1d0c0b32012-04-27 02:10:03 +00001760 /* map the linear part :
Alexander Duyck2996d312012-05-02 18:18:42 +00001761 * If skb->head_frag is set, this 'linear' part is backed by a
1762 * fragment, and if the head is not shared with any clones then
1763 * we can avoid a copy since we own the head portion of this page.
Jens Axboe9c55e012007-11-06 23:30:13 -08001764 */
Octavian Purdila2870c432008-07-15 00:49:11 -07001765 if (__splice_segment(virt_to_page(skb->data),
1766 (unsigned long) skb->data & (PAGE_SIZE - 1),
1767 skb_headlen(skb),
Eric Dumazet18aafc62013-01-11 14:46:37 +00001768 offset, len, spd,
Alexander Duyck3a7c1ee42012-05-03 01:09:42 +00001769 skb_head_is_locked(skb),
Eric Dumazet1d0c0b32012-04-27 02:10:03 +00001770 sk, pipe))
David S. Millera108d5f2012-04-23 23:06:11 -04001771 return true;
Jens Axboe9c55e012007-11-06 23:30:13 -08001772
1773 /*
1774 * then map the fragments
1775 */
Jens Axboe9c55e012007-11-06 23:30:13 -08001776 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1777 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1778
Ian Campbellea2ab692011-08-22 23:44:58 +00001779 if (__splice_segment(skb_frag_page(f),
Eric Dumazet9e903e02011-10-18 21:00:24 +00001780 f->page_offset, skb_frag_size(f),
Eric Dumazet18aafc62013-01-11 14:46:37 +00001781 offset, len, spd, false, sk, pipe))
David S. Millera108d5f2012-04-23 23:06:11 -04001782 return true;
Jens Axboe9c55e012007-11-06 23:30:13 -08001783 }
1784
David S. Millera108d5f2012-04-23 23:06:11 -04001785 return false;
Jens Axboe9c55e012007-11-06 23:30:13 -08001786}
1787
1788/*
1789 * Map data from the skb to a pipe. Should handle both the linear part,
1790 * the fragments, and the frag list. It does NOT handle frag lists within
1791 * the frag list, if such a thing exists. We'd probably need to recurse to
1792 * handle that cleanly.
1793 */
Jarek Poplawski8b9d3722009-01-19 17:03:56 -08001794int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
Jens Axboe9c55e012007-11-06 23:30:13 -08001795 struct pipe_inode_info *pipe, unsigned int tlen,
1796 unsigned int flags)
1797{
Eric Dumazet41c73a02012-04-22 12:26:16 +00001798 struct partial_page partial[MAX_SKB_FRAGS];
1799 struct page *pages[MAX_SKB_FRAGS];
Jens Axboe9c55e012007-11-06 23:30:13 -08001800 struct splice_pipe_desc spd = {
1801 .pages = pages,
1802 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +02001803 .nr_pages_max = MAX_SKB_FRAGS,
Jens Axboe9c55e012007-11-06 23:30:13 -08001804 .flags = flags,
Miklos Szeredi28a625c2014-01-22 19:36:57 +01001805 .ops = &nosteal_pipe_buf_ops,
Jens Axboe9c55e012007-11-06 23:30:13 -08001806 .spd_release = sock_spd_release,
1807 };
David S. Millerfbb398a2009-06-09 00:18:59 -07001808 struct sk_buff *frag_iter;
Jarek Poplawski7a67e562009-04-30 05:41:19 -07001809 struct sock *sk = skb->sk;
Jens Axboe35f3d142010-05-20 10:43:18 +02001810 int ret = 0;
1811
Jens Axboe9c55e012007-11-06 23:30:13 -08001812 /*
1813 * __skb_splice_bits() only fails if the output has no room left,
1814 * so no point in going over the frag_list for the error case.
1815 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001816 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
Jens Axboe9c55e012007-11-06 23:30:13 -08001817 goto done;
1818 else if (!tlen)
1819 goto done;
1820
1821 /*
1822 * now see if we have a frag_list to map
1823 */
David S. Millerfbb398a2009-06-09 00:18:59 -07001824 skb_walk_frags(skb, frag_iter) {
1825 if (!tlen)
1826 break;
Jens Axboe35f3d142010-05-20 10:43:18 +02001827 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
David S. Millerfbb398a2009-06-09 00:18:59 -07001828 break;
Jens Axboe9c55e012007-11-06 23:30:13 -08001829 }
1830
1831done:
Jens Axboe9c55e012007-11-06 23:30:13 -08001832 if (spd.nr_pages) {
Jens Axboe9c55e012007-11-06 23:30:13 -08001833 /*
1834 * Drop the socket lock, otherwise we have reverse
1835 * locking dependencies between sk_lock and i_mutex
1836 * here as compared to sendfile(). We enter here
1837 * with the socket lock held, and splice_to_pipe() will
1838 * grab the pipe inode lock. For sendfile() emulation,
1839 * we call into ->sendpage() with the i_mutex lock held
1840 * and networking will grab the socket lock.
1841 */
Octavian Purdila293ad602008-06-04 15:45:58 -07001842 release_sock(sk);
Jens Axboe9c55e012007-11-06 23:30:13 -08001843 ret = splice_to_pipe(pipe, &spd);
Octavian Purdila293ad602008-06-04 15:45:58 -07001844 lock_sock(sk);
Jens Axboe9c55e012007-11-06 23:30:13 -08001845 }
1846
Jens Axboe35f3d142010-05-20 10:43:18 +02001847 return ret;
Jens Axboe9c55e012007-11-06 23:30:13 -08001848}
1849
Herbert Xu357b40a2005-04-19 22:30:14 -07001850/**
1851 * skb_store_bits - store bits from kernel buffer to skb
1852 * @skb: destination buffer
1853 * @offset: offset in destination
1854 * @from: source buffer
1855 * @len: number of bytes to copy
1856 *
1857 * Copy the specified number of bytes from the source buffer to the
1858 * destination skb. This function handles all the messy bits of
1859 * traversing fragment lists and such.
1860 */
1861
Stephen Hemminger0c6fcc82007-04-20 16:40:01 -07001862int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
Herbert Xu357b40a2005-04-19 22:30:14 -07001863{
David S. Miller1a028e52007-04-27 15:21:23 -07001864 int start = skb_headlen(skb);
David S. Millerfbb398a2009-06-09 00:18:59 -07001865 struct sk_buff *frag_iter;
1866 int i, copy;
Herbert Xu357b40a2005-04-19 22:30:14 -07001867
1868 if (offset > (int)skb->len - len)
1869 goto fault;
1870
David S. Miller1a028e52007-04-27 15:21:23 -07001871 if ((copy = start - offset) > 0) {
Herbert Xu357b40a2005-04-19 22:30:14 -07001872 if (copy > len)
1873 copy = len;
Arnaldo Carvalho de Melo27d7ff42007-03-31 11:55:19 -03001874 skb_copy_to_linear_data_offset(skb, offset, from, copy);
Herbert Xu357b40a2005-04-19 22:30:14 -07001875 if ((len -= copy) == 0)
1876 return 0;
1877 offset += copy;
1878 from += copy;
1879 }
1880
1881 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1882 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
David S. Miller1a028e52007-04-27 15:21:23 -07001883 int end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001884
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001885 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07001886
Eric Dumazet9e903e02011-10-18 21:00:24 +00001887 end = start + skb_frag_size(frag);
Herbert Xu357b40a2005-04-19 22:30:14 -07001888 if ((copy = end - offset) > 0) {
1889 u8 *vaddr;
1890
1891 if (copy > len)
1892 copy = len;
1893
Eric Dumazet51c56b02012-04-05 11:35:15 +02001894 vaddr = kmap_atomic(skb_frag_page(frag));
David S. Miller1a028e52007-04-27 15:21:23 -07001895 memcpy(vaddr + frag->page_offset + offset - start,
1896 from, copy);
Eric Dumazet51c56b02012-04-05 11:35:15 +02001897 kunmap_atomic(vaddr);
Herbert Xu357b40a2005-04-19 22:30:14 -07001898
1899 if ((len -= copy) == 0)
1900 return 0;
1901 offset += copy;
1902 from += copy;
1903 }
David S. Miller1a028e52007-04-27 15:21:23 -07001904 start = end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001905 }
1906
David S. Millerfbb398a2009-06-09 00:18:59 -07001907 skb_walk_frags(skb, frag_iter) {
1908 int end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001909
David S. Millerfbb398a2009-06-09 00:18:59 -07001910 WARN_ON(start > offset + len);
Herbert Xu357b40a2005-04-19 22:30:14 -07001911
David S. Millerfbb398a2009-06-09 00:18:59 -07001912 end = start + frag_iter->len;
1913 if ((copy = end - offset) > 0) {
1914 if (copy > len)
1915 copy = len;
1916 if (skb_store_bits(frag_iter, offset - start,
1917 from, copy))
1918 goto fault;
1919 if ((len -= copy) == 0)
1920 return 0;
1921 offset += copy;
1922 from += copy;
Herbert Xu357b40a2005-04-19 22:30:14 -07001923 }
David S. Millerfbb398a2009-06-09 00:18:59 -07001924 start = end;
Herbert Xu357b40a2005-04-19 22:30:14 -07001925 }
1926 if (!len)
1927 return 0;
1928
1929fault:
1930 return -EFAULT;
1931}
Herbert Xu357b40a2005-04-19 22:30:14 -07001932EXPORT_SYMBOL(skb_store_bits);
1933
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934/* Checksum skb data. */
Daniel Borkmann2817a332013-10-30 11:50:51 +01001935__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
1936 __wsum csum, const struct skb_checksum_ops *ops)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937{
David S. Miller1a028e52007-04-27 15:21:23 -07001938 int start = skb_headlen(skb);
1939 int i, copy = start - offset;
David S. Millerfbb398a2009-06-09 00:18:59 -07001940 struct sk_buff *frag_iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 int pos = 0;
1942
1943 /* Checksum header. */
1944 if (copy > 0) {
1945 if (copy > len)
1946 copy = len;
Daniel Borkmann2817a332013-10-30 11:50:51 +01001947 csum = ops->update(skb->data + offset, copy, csum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948 if ((len -= copy) == 0)
1949 return csum;
1950 offset += copy;
1951 pos = copy;
1952 }
1953
1954 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07001955 int end;
Eric Dumazet51c56b02012-04-05 11:35:15 +02001956 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001958 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07001959
Eric Dumazet51c56b02012-04-05 11:35:15 +02001960 end = start + skb_frag_size(frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 if ((copy = end - offset) > 0) {
Al Viro44bb9362006-11-14 21:36:14 -08001962 __wsum csum2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 u8 *vaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964
1965 if (copy > len)
1966 copy = len;
Eric Dumazet51c56b02012-04-05 11:35:15 +02001967 vaddr = kmap_atomic(skb_frag_page(frag));
Daniel Borkmann2817a332013-10-30 11:50:51 +01001968 csum2 = ops->update(vaddr + frag->page_offset +
1969 offset - start, copy, 0);
Eric Dumazet51c56b02012-04-05 11:35:15 +02001970 kunmap_atomic(vaddr);
Daniel Borkmann2817a332013-10-30 11:50:51 +01001971 csum = ops->combine(csum, csum2, pos, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 if (!(len -= copy))
1973 return csum;
1974 offset += copy;
1975 pos += copy;
1976 }
David S. Miller1a028e52007-04-27 15:21:23 -07001977 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978 }
1979
David S. Millerfbb398a2009-06-09 00:18:59 -07001980 skb_walk_frags(skb, frag_iter) {
1981 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982
David S. Millerfbb398a2009-06-09 00:18:59 -07001983 WARN_ON(start > offset + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984
David S. Millerfbb398a2009-06-09 00:18:59 -07001985 end = start + frag_iter->len;
1986 if ((copy = end - offset) > 0) {
1987 __wsum csum2;
1988 if (copy > len)
1989 copy = len;
Daniel Borkmann2817a332013-10-30 11:50:51 +01001990 csum2 = __skb_checksum(frag_iter, offset - start,
1991 copy, 0, ops);
1992 csum = ops->combine(csum, csum2, pos, copy);
David S. Millerfbb398a2009-06-09 00:18:59 -07001993 if ((len -= copy) == 0)
1994 return csum;
1995 offset += copy;
1996 pos += copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 }
David S. Millerfbb398a2009-06-09 00:18:59 -07001998 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 }
Kris Katterjohn09a62662006-01-08 22:24:28 -08002000 BUG_ON(len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001
2002 return csum;
2003}
Daniel Borkmann2817a332013-10-30 11:50:51 +01002004EXPORT_SYMBOL(__skb_checksum);
2005
2006__wsum skb_checksum(const struct sk_buff *skb, int offset,
2007 int len, __wsum csum)
2008{
2009 const struct skb_checksum_ops ops = {
Daniel Borkmanncea80ea2013-11-04 17:10:25 +01002010 .update = csum_partial_ext,
Daniel Borkmann2817a332013-10-30 11:50:51 +01002011 .combine = csum_block_add_ext,
2012 };
2013
2014 return __skb_checksum(skb, offset, len, csum, &ops);
2015}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002016EXPORT_SYMBOL(skb_checksum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017
2018/* Both of above in one bottle. */
2019
Al Viro81d77662006-11-14 21:37:33 -08002020__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2021 u8 *to, int len, __wsum csum)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022{
David S. Miller1a028e52007-04-27 15:21:23 -07002023 int start = skb_headlen(skb);
2024 int i, copy = start - offset;
David S. Millerfbb398a2009-06-09 00:18:59 -07002025 struct sk_buff *frag_iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 int pos = 0;
2027
2028 /* Copy header. */
2029 if (copy > 0) {
2030 if (copy > len)
2031 copy = len;
2032 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2033 copy, csum);
2034 if ((len -= copy) == 0)
2035 return csum;
2036 offset += copy;
2037 to += copy;
2038 pos = copy;
2039 }
2040
2041 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07002042 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043
Ilpo Järvinen547b7922008-07-25 21:43:18 -07002044 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07002045
Eric Dumazet9e903e02011-10-18 21:00:24 +00002046 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 if ((copy = end - offset) > 0) {
Al Viro50842052006-11-14 21:36:34 -08002048 __wsum csum2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 u8 *vaddr;
2050 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2051
2052 if (copy > len)
2053 copy = len;
Eric Dumazet51c56b02012-04-05 11:35:15 +02002054 vaddr = kmap_atomic(skb_frag_page(frag));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 csum2 = csum_partial_copy_nocheck(vaddr +
David S. Miller1a028e52007-04-27 15:21:23 -07002056 frag->page_offset +
2057 offset - start, to,
2058 copy, 0);
Eric Dumazet51c56b02012-04-05 11:35:15 +02002059 kunmap_atomic(vaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 csum = csum_block_add(csum, csum2, pos);
2061 if (!(len -= copy))
2062 return csum;
2063 offset += copy;
2064 to += copy;
2065 pos += copy;
2066 }
David S. Miller1a028e52007-04-27 15:21:23 -07002067 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 }
2069
David S. Millerfbb398a2009-06-09 00:18:59 -07002070 skb_walk_frags(skb, frag_iter) {
2071 __wsum csum2;
2072 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073
David S. Millerfbb398a2009-06-09 00:18:59 -07002074 WARN_ON(start > offset + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075
David S. Millerfbb398a2009-06-09 00:18:59 -07002076 end = start + frag_iter->len;
2077 if ((copy = end - offset) > 0) {
2078 if (copy > len)
2079 copy = len;
2080 csum2 = skb_copy_and_csum_bits(frag_iter,
2081 offset - start,
2082 to, copy, 0);
2083 csum = csum_block_add(csum, csum2, pos);
2084 if ((len -= copy) == 0)
2085 return csum;
2086 offset += copy;
2087 to += copy;
2088 pos += copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 }
David S. Millerfbb398a2009-06-09 00:18:59 -07002090 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091 }
Kris Katterjohn09a62662006-01-08 22:24:28 -08002092 BUG_ON(len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093 return csum;
2094}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002095EXPORT_SYMBOL(skb_copy_and_csum_bits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096
Thomas Grafaf2806f2013-12-13 15:22:17 +01002097 /**
2098 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2099 * @from: source buffer
2100 *
2101 * Calculates the amount of linear headroom needed in the 'to' skb passed
2102 * into skb_zerocopy().
2103 */
2104unsigned int
2105skb_zerocopy_headlen(const struct sk_buff *from)
2106{
2107 unsigned int hlen = 0;
2108
2109 if (!from->head_frag ||
2110 skb_headlen(from) < L1_CACHE_BYTES ||
2111 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2112 hlen = skb_headlen(from);
2113
2114 if (skb_has_frag_list(from))
2115 hlen = from->len;
2116
2117 return hlen;
2118}
2119EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2120
2121/**
2122 * skb_zerocopy - Zero copy skb to skb
2123 * @to: destination buffer
Masanari Iida7fceb4d2014-01-29 01:05:28 +09002124 * @from: source buffer
Thomas Grafaf2806f2013-12-13 15:22:17 +01002125 * @len: number of bytes to copy from source buffer
2126 * @hlen: size of linear headroom in destination buffer
2127 *
2128 * Copies up to `len` bytes from `from` to `to` by creating references
2129 * to the frags in the source buffer.
2130 *
2131 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2132 * headroom in the `to` buffer.
2133 */
2134void
2135skb_zerocopy(struct sk_buff *to, const struct sk_buff *from, int len, int hlen)
2136{
2137 int i, j = 0;
2138 int plen = 0; /* length of skb->head fragment */
2139 struct page *page;
2140 unsigned int offset;
2141
2142 BUG_ON(!from->head_frag && !hlen);
2143
2144 /* dont bother with small payloads */
2145 if (len <= skb_tailroom(to)) {
2146 skb_copy_bits(from, 0, skb_put(to, len), len);
2147 return;
2148 }
2149
2150 if (hlen) {
2151 skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2152 len -= hlen;
2153 } else {
2154 plen = min_t(int, skb_headlen(from), len);
2155 if (plen) {
2156 page = virt_to_head_page(from->head);
2157 offset = from->data - (unsigned char *)page_address(page);
2158 __skb_fill_page_desc(to, 0, page, offset, plen);
2159 get_page(page);
2160 j = 1;
2161 len -= plen;
2162 }
2163 }
2164
2165 to->truesize += len + plen;
2166 to->len += len + plen;
2167 to->data_len += len + plen;
2168
2169 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2170 if (!len)
2171 break;
2172 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2173 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
2174 len -= skb_shinfo(to)->frags[j].size;
2175 skb_frag_ref(to, j);
2176 j++;
2177 }
2178 skb_shinfo(to)->nr_frags = j;
2179}
2180EXPORT_SYMBOL_GPL(skb_zerocopy);
2181
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2183{
Al Virod3bc23e2006-11-14 21:24:49 -08002184 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 long csstart;
2186
Patrick McHardy84fa7932006-08-29 16:44:56 -07002187 if (skb->ip_summed == CHECKSUM_PARTIAL)
Michał Mirosław55508d62010-12-14 15:24:08 +00002188 csstart = skb_checksum_start_offset(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189 else
2190 csstart = skb_headlen(skb);
2191
Kris Katterjohn09a62662006-01-08 22:24:28 -08002192 BUG_ON(csstart > skb_headlen(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03002194 skb_copy_from_linear_data(skb, to, csstart);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195
2196 csum = 0;
2197 if (csstart != skb->len)
2198 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2199 skb->len - csstart, 0);
2200
Patrick McHardy84fa7932006-08-29 16:44:56 -07002201 if (skb->ip_summed == CHECKSUM_PARTIAL) {
Al Viroff1dcad2006-11-20 18:07:29 -08002202 long csstuff = csstart + skb->csum_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203
Al Virod3bc23e2006-11-14 21:24:49 -08002204 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 }
2206}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002207EXPORT_SYMBOL(skb_copy_and_csum_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208
2209/**
2210 * skb_dequeue - remove from the head of the queue
2211 * @list: list to dequeue from
2212 *
2213 * Remove the head of the list. The list lock is taken so the function
2214 * may be used safely with other locking list functions. The head item is
2215 * returned or %NULL if the list is empty.
2216 */
2217
2218struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2219{
2220 unsigned long flags;
2221 struct sk_buff *result;
2222
2223 spin_lock_irqsave(&list->lock, flags);
2224 result = __skb_dequeue(list);
2225 spin_unlock_irqrestore(&list->lock, flags);
2226 return result;
2227}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002228EXPORT_SYMBOL(skb_dequeue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229
2230/**
2231 * skb_dequeue_tail - remove from the tail of the queue
2232 * @list: list to dequeue from
2233 *
2234 * Remove the tail of the list. The list lock is taken so the function
2235 * may be used safely with other locking list functions. The tail item is
2236 * returned or %NULL if the list is empty.
2237 */
2238struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2239{
2240 unsigned long flags;
2241 struct sk_buff *result;
2242
2243 spin_lock_irqsave(&list->lock, flags);
2244 result = __skb_dequeue_tail(list);
2245 spin_unlock_irqrestore(&list->lock, flags);
2246 return result;
2247}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002248EXPORT_SYMBOL(skb_dequeue_tail);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249
2250/**
2251 * skb_queue_purge - empty a list
2252 * @list: list to empty
2253 *
2254 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2255 * the list and one reference dropped. This function takes the list
2256 * lock and is atomic with respect to other list locking functions.
2257 */
2258void skb_queue_purge(struct sk_buff_head *list)
2259{
2260 struct sk_buff *skb;
2261 while ((skb = skb_dequeue(list)) != NULL)
2262 kfree_skb(skb);
2263}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002264EXPORT_SYMBOL(skb_queue_purge);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265
2266/**
2267 * skb_queue_head - queue a buffer at the list head
2268 * @list: list to use
2269 * @newsk: buffer to queue
2270 *
2271 * Queue a buffer at the start of the list. This function takes the
2272 * list lock and can be used safely with other locking &sk_buff functions
2273 * safely.
2274 *
2275 * A buffer cannot be placed on two lists at the same time.
2276 */
2277void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2278{
2279 unsigned long flags;
2280
2281 spin_lock_irqsave(&list->lock, flags);
2282 __skb_queue_head(list, newsk);
2283 spin_unlock_irqrestore(&list->lock, flags);
2284}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002285EXPORT_SYMBOL(skb_queue_head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286
2287/**
2288 * skb_queue_tail - queue a buffer at the list tail
2289 * @list: list to use
2290 * @newsk: buffer to queue
2291 *
2292 * Queue a buffer at the tail of the list. This function takes the
2293 * list lock and can be used safely with other locking &sk_buff functions
2294 * safely.
2295 *
2296 * A buffer cannot be placed on two lists at the same time.
2297 */
2298void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2299{
2300 unsigned long flags;
2301
2302 spin_lock_irqsave(&list->lock, flags);
2303 __skb_queue_tail(list, newsk);
2304 spin_unlock_irqrestore(&list->lock, flags);
2305}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002306EXPORT_SYMBOL(skb_queue_tail);
David S. Miller8728b832005-08-09 19:25:21 -07002307
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308/**
2309 * skb_unlink - remove a buffer from a list
2310 * @skb: buffer to remove
David S. Miller8728b832005-08-09 19:25:21 -07002311 * @list: list to use
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312 *
David S. Miller8728b832005-08-09 19:25:21 -07002313 * Remove a packet from a list. The list locks are taken and this
2314 * function is atomic with respect to other list locked calls
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 *
David S. Miller8728b832005-08-09 19:25:21 -07002316 * You must know what list the SKB is on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317 */
David S. Miller8728b832005-08-09 19:25:21 -07002318void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319{
David S. Miller8728b832005-08-09 19:25:21 -07002320 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321
David S. Miller8728b832005-08-09 19:25:21 -07002322 spin_lock_irqsave(&list->lock, flags);
2323 __skb_unlink(skb, list);
2324 spin_unlock_irqrestore(&list->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002326EXPORT_SYMBOL(skb_unlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328/**
2329 * skb_append - append a buffer
2330 * @old: buffer to insert after
2331 * @newsk: buffer to insert
David S. Miller8728b832005-08-09 19:25:21 -07002332 * @list: list to use
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 *
2334 * Place a packet after a given packet in a list. The list locks are taken
2335 * and this function is atomic with respect to other list locked calls.
2336 * A buffer cannot be placed on two lists at the same time.
2337 */
David S. Miller8728b832005-08-09 19:25:21 -07002338void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339{
2340 unsigned long flags;
2341
David S. Miller8728b832005-08-09 19:25:21 -07002342 spin_lock_irqsave(&list->lock, flags);
Gerrit Renker7de6c032008-04-14 00:05:09 -07002343 __skb_queue_after(list, old, newsk);
David S. Miller8728b832005-08-09 19:25:21 -07002344 spin_unlock_irqrestore(&list->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002346EXPORT_SYMBOL(skb_append);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347
2348/**
2349 * skb_insert - insert a buffer
2350 * @old: buffer to insert before
2351 * @newsk: buffer to insert
David S. Miller8728b832005-08-09 19:25:21 -07002352 * @list: list to use
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353 *
David S. Miller8728b832005-08-09 19:25:21 -07002354 * Place a packet before a given packet in a list. The list locks are
2355 * taken and this function is atomic with respect to other list locked
2356 * calls.
2357 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358 * A buffer cannot be placed on two lists at the same time.
2359 */
David S. Miller8728b832005-08-09 19:25:21 -07002360void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361{
2362 unsigned long flags;
2363
David S. Miller8728b832005-08-09 19:25:21 -07002364 spin_lock_irqsave(&list->lock, flags);
2365 __skb_insert(newsk, old->prev, old, list);
2366 spin_unlock_irqrestore(&list->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002368EXPORT_SYMBOL(skb_insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370static inline void skb_split_inside_header(struct sk_buff *skb,
2371 struct sk_buff* skb1,
2372 const u32 len, const int pos)
2373{
2374 int i;
2375
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03002376 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2377 pos - len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378 /* And move data appendix as is. */
2379 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2380 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2381
2382 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2383 skb_shinfo(skb)->nr_frags = 0;
2384 skb1->data_len = skb->data_len;
2385 skb1->len += skb1->data_len;
2386 skb->data_len = 0;
2387 skb->len = len;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07002388 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389}
2390
2391static inline void skb_split_no_header(struct sk_buff *skb,
2392 struct sk_buff* skb1,
2393 const u32 len, int pos)
2394{
2395 int i, k = 0;
2396 const int nfrags = skb_shinfo(skb)->nr_frags;
2397
2398 skb_shinfo(skb)->nr_frags = 0;
2399 skb1->len = skb1->data_len = skb->len - len;
2400 skb->len = len;
2401 skb->data_len = len - pos;
2402
2403 for (i = 0; i < nfrags; i++) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00002404 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405
2406 if (pos + size > len) {
2407 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2408
2409 if (pos < len) {
2410 /* Split frag.
2411 * We have two variants in this case:
2412 * 1. Move all the frag to the second
2413 * part, if it is possible. F.e.
2414 * this approach is mandatory for TUX,
2415 * where splitting is expensive.
2416 * 2. Split is accurately. We make this.
2417 */
Ian Campbellea2ab692011-08-22 23:44:58 +00002418 skb_frag_ref(skb, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
Eric Dumazet9e903e02011-10-18 21:00:24 +00002420 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2421 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422 skb_shinfo(skb)->nr_frags++;
2423 }
2424 k++;
2425 } else
2426 skb_shinfo(skb)->nr_frags++;
2427 pos += size;
2428 }
2429 skb_shinfo(skb1)->nr_frags = k;
2430}
2431
2432/**
2433 * skb_split - Split fragmented skb to two parts at length len.
2434 * @skb: the buffer to split
2435 * @skb1: the buffer to receive the second part
2436 * @len: new length for skb
2437 */
2438void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2439{
2440 int pos = skb_headlen(skb);
2441
Amerigo Wang68534c62013-02-19 22:51:30 +00002442 skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 if (len < pos) /* Split line is inside header. */
2444 skb_split_inside_header(skb, skb1, len, pos);
2445 else /* Second chunk has no header, nothing to copy. */
2446 skb_split_no_header(skb, skb1, len, pos);
2447}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002448EXPORT_SYMBOL(skb_split);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449
Ilpo Järvinen9f782db2008-11-25 13:57:01 -08002450/* Shifting from/to a cloned skb is a no-go.
2451 *
2452 * Caller cannot keep skb_shinfo related pointers past calling here!
2453 */
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002454static int skb_prepare_for_shift(struct sk_buff *skb)
2455{
Ilpo Järvinen0ace2852008-11-24 21:30:21 -08002456 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002457}
2458
2459/**
2460 * skb_shift - Shifts paged data partially from skb to another
2461 * @tgt: buffer into which tail data gets added
2462 * @skb: buffer from which the paged data comes from
2463 * @shiftlen: shift up to this many bytes
2464 *
2465 * Attempts to shift up to shiftlen worth of bytes, which may be less than
Feng King20e994a2011-11-21 01:47:11 +00002466 * the length of the skb, from skb to tgt. Returns number bytes shifted.
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002467 * It's up to caller to free skb if everything was shifted.
2468 *
2469 * If @tgt runs out of frags, the whole operation is aborted.
2470 *
2471 * Skb cannot include anything else but paged data while tgt is allowed
2472 * to have non-paged data as well.
2473 *
2474 * TODO: full sized shift could be optimized but that would need
2475 * specialized skb free'er to handle frags without up-to-date nr_frags.
2476 */
2477int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2478{
2479 int from, to, merge, todo;
2480 struct skb_frag_struct *fragfrom, *fragto;
2481
2482 BUG_ON(shiftlen > skb->len);
2483 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2484
2485 todo = shiftlen;
2486 from = 0;
2487 to = skb_shinfo(tgt)->nr_frags;
2488 fragfrom = &skb_shinfo(skb)->frags[from];
2489
2490 /* Actual merge is delayed until the point when we know we can
2491 * commit all, so that we don't have to undo partial changes
2492 */
2493 if (!to ||
Ian Campbellea2ab692011-08-22 23:44:58 +00002494 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2495 fragfrom->page_offset)) {
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002496 merge = -1;
2497 } else {
2498 merge = to - 1;
2499
Eric Dumazet9e903e02011-10-18 21:00:24 +00002500 todo -= skb_frag_size(fragfrom);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002501 if (todo < 0) {
2502 if (skb_prepare_for_shift(skb) ||
2503 skb_prepare_for_shift(tgt))
2504 return 0;
2505
Ilpo Järvinen9f782db2008-11-25 13:57:01 -08002506 /* All previous frag pointers might be stale! */
2507 fragfrom = &skb_shinfo(skb)->frags[from];
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002508 fragto = &skb_shinfo(tgt)->frags[merge];
2509
Eric Dumazet9e903e02011-10-18 21:00:24 +00002510 skb_frag_size_add(fragto, shiftlen);
2511 skb_frag_size_sub(fragfrom, shiftlen);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002512 fragfrom->page_offset += shiftlen;
2513
2514 goto onlymerged;
2515 }
2516
2517 from++;
2518 }
2519
2520 /* Skip full, not-fitting skb to avoid expensive operations */
2521 if ((shiftlen == skb->len) &&
2522 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2523 return 0;
2524
2525 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2526 return 0;
2527
2528 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2529 if (to == MAX_SKB_FRAGS)
2530 return 0;
2531
2532 fragfrom = &skb_shinfo(skb)->frags[from];
2533 fragto = &skb_shinfo(tgt)->frags[to];
2534
Eric Dumazet9e903e02011-10-18 21:00:24 +00002535 if (todo >= skb_frag_size(fragfrom)) {
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002536 *fragto = *fragfrom;
Eric Dumazet9e903e02011-10-18 21:00:24 +00002537 todo -= skb_frag_size(fragfrom);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002538 from++;
2539 to++;
2540
2541 } else {
Ian Campbellea2ab692011-08-22 23:44:58 +00002542 __skb_frag_ref(fragfrom);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002543 fragto->page = fragfrom->page;
2544 fragto->page_offset = fragfrom->page_offset;
Eric Dumazet9e903e02011-10-18 21:00:24 +00002545 skb_frag_size_set(fragto, todo);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002546
2547 fragfrom->page_offset += todo;
Eric Dumazet9e903e02011-10-18 21:00:24 +00002548 skb_frag_size_sub(fragfrom, todo);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002549 todo = 0;
2550
2551 to++;
2552 break;
2553 }
2554 }
2555
2556 /* Ready to "commit" this state change to tgt */
2557 skb_shinfo(tgt)->nr_frags = to;
2558
2559 if (merge >= 0) {
2560 fragfrom = &skb_shinfo(skb)->frags[0];
2561 fragto = &skb_shinfo(tgt)->frags[merge];
2562
Eric Dumazet9e903e02011-10-18 21:00:24 +00002563 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
Ian Campbellea2ab692011-08-22 23:44:58 +00002564 __skb_frag_unref(fragfrom);
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08002565 }
2566
2567 /* Reposition in the original skb */
2568 to = 0;
2569 while (from < skb_shinfo(skb)->nr_frags)
2570 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2571 skb_shinfo(skb)->nr_frags = to;
2572
2573 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2574
2575onlymerged:
2576 /* Most likely the tgt won't ever need its checksum anymore, skb on
2577 * the other hand might need it if it needs to be resent
2578 */
2579 tgt->ip_summed = CHECKSUM_PARTIAL;
2580 skb->ip_summed = CHECKSUM_PARTIAL;
2581
2582 /* Yak, is it really working this way? Some helper please? */
2583 skb->len -= shiftlen;
2584 skb->data_len -= shiftlen;
2585 skb->truesize -= shiftlen;
2586 tgt->len += shiftlen;
2587 tgt->data_len += shiftlen;
2588 tgt->truesize += shiftlen;
2589
2590 return shiftlen;
2591}
2592
Thomas Graf677e90e2005-06-23 20:59:51 -07002593/**
2594 * skb_prepare_seq_read - Prepare a sequential read of skb data
2595 * @skb: the buffer to read
2596 * @from: lower offset of data to be read
2597 * @to: upper offset of data to be read
2598 * @st: state variable
2599 *
2600 * Initializes the specified state variable. Must be called before
2601 * invoking skb_seq_read() for the first time.
2602 */
2603void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2604 unsigned int to, struct skb_seq_state *st)
2605{
2606 st->lower_offset = from;
2607 st->upper_offset = to;
2608 st->root_skb = st->cur_skb = skb;
2609 st->frag_idx = st->stepped_offset = 0;
2610 st->frag_data = NULL;
2611}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002612EXPORT_SYMBOL(skb_prepare_seq_read);
Thomas Graf677e90e2005-06-23 20:59:51 -07002613
2614/**
2615 * skb_seq_read - Sequentially read skb data
2616 * @consumed: number of bytes consumed by the caller so far
2617 * @data: destination pointer for data to be returned
2618 * @st: state variable
2619 *
Mathias Krausebc323832013-11-07 14:18:26 +01002620 * Reads a block of skb data at @consumed relative to the
Thomas Graf677e90e2005-06-23 20:59:51 -07002621 * lower offset specified to skb_prepare_seq_read(). Assigns
Mathias Krausebc323832013-11-07 14:18:26 +01002622 * the head of the data block to @data and returns the length
Thomas Graf677e90e2005-06-23 20:59:51 -07002623 * of the block or 0 if the end of the skb data or the upper
2624 * offset has been reached.
2625 *
2626 * The caller is not required to consume all of the data
Mathias Krausebc323832013-11-07 14:18:26 +01002627 * returned, i.e. @consumed is typically set to the number
Thomas Graf677e90e2005-06-23 20:59:51 -07002628 * of bytes already consumed and the next call to
2629 * skb_seq_read() will return the remaining part of the block.
2630 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002631 * Note 1: The size of each block of data returned can be arbitrary,
Thomas Graf677e90e2005-06-23 20:59:51 -07002632 * this limitation is the cost for zerocopy seqeuental
2633 * reads of potentially non linear data.
2634 *
Randy Dunlapbc2cda12008-02-13 15:03:25 -08002635 * Note 2: Fragment lists within fragments are not implemented
Thomas Graf677e90e2005-06-23 20:59:51 -07002636 * at the moment, state->root_skb could be replaced with
2637 * a stack for this purpose.
2638 */
2639unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2640 struct skb_seq_state *st)
2641{
2642 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2643 skb_frag_t *frag;
2644
Wedson Almeida Filhoaeb193e2013-06-23 23:33:48 -07002645 if (unlikely(abs_offset >= st->upper_offset)) {
2646 if (st->frag_data) {
2647 kunmap_atomic(st->frag_data);
2648 st->frag_data = NULL;
2649 }
Thomas Graf677e90e2005-06-23 20:59:51 -07002650 return 0;
Wedson Almeida Filhoaeb193e2013-06-23 23:33:48 -07002651 }
Thomas Graf677e90e2005-06-23 20:59:51 -07002652
2653next_skb:
Herbert Xu95e3b242009-01-29 16:07:52 -08002654 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
Thomas Graf677e90e2005-06-23 20:59:51 -07002655
Thomas Chenault995b3372009-05-18 21:43:27 -07002656 if (abs_offset < block_limit && !st->frag_data) {
Herbert Xu95e3b242009-01-29 16:07:52 -08002657 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
Thomas Graf677e90e2005-06-23 20:59:51 -07002658 return block_limit - abs_offset;
2659 }
2660
2661 if (st->frag_idx == 0 && !st->frag_data)
2662 st->stepped_offset += skb_headlen(st->cur_skb);
2663
2664 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2665 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
Eric Dumazet9e903e02011-10-18 21:00:24 +00002666 block_limit = skb_frag_size(frag) + st->stepped_offset;
Thomas Graf677e90e2005-06-23 20:59:51 -07002667
2668 if (abs_offset < block_limit) {
2669 if (!st->frag_data)
Eric Dumazet51c56b02012-04-05 11:35:15 +02002670 st->frag_data = kmap_atomic(skb_frag_page(frag));
Thomas Graf677e90e2005-06-23 20:59:51 -07002671
2672 *data = (u8 *) st->frag_data + frag->page_offset +
2673 (abs_offset - st->stepped_offset);
2674
2675 return block_limit - abs_offset;
2676 }
2677
2678 if (st->frag_data) {
Eric Dumazet51c56b02012-04-05 11:35:15 +02002679 kunmap_atomic(st->frag_data);
Thomas Graf677e90e2005-06-23 20:59:51 -07002680 st->frag_data = NULL;
2681 }
2682
2683 st->frag_idx++;
Eric Dumazet9e903e02011-10-18 21:00:24 +00002684 st->stepped_offset += skb_frag_size(frag);
Thomas Graf677e90e2005-06-23 20:59:51 -07002685 }
2686
Olaf Kirch5b5a60d2007-06-23 23:11:52 -07002687 if (st->frag_data) {
Eric Dumazet51c56b02012-04-05 11:35:15 +02002688 kunmap_atomic(st->frag_data);
Olaf Kirch5b5a60d2007-06-23 23:11:52 -07002689 st->frag_data = NULL;
2690 }
2691
David S. Miller21dc3302010-08-23 00:13:46 -07002692 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
Shyam Iyer71b33462009-01-29 16:12:42 -08002693 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
Thomas Graf677e90e2005-06-23 20:59:51 -07002694 st->frag_idx = 0;
2695 goto next_skb;
Shyam Iyer71b33462009-01-29 16:12:42 -08002696 } else if (st->cur_skb->next) {
2697 st->cur_skb = st->cur_skb->next;
Herbert Xu95e3b242009-01-29 16:07:52 -08002698 st->frag_idx = 0;
Thomas Graf677e90e2005-06-23 20:59:51 -07002699 goto next_skb;
2700 }
2701
2702 return 0;
2703}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002704EXPORT_SYMBOL(skb_seq_read);
Thomas Graf677e90e2005-06-23 20:59:51 -07002705
2706/**
2707 * skb_abort_seq_read - Abort a sequential read of skb data
2708 * @st: state variable
2709 *
2710 * Must be called if skb_seq_read() was not called until it
2711 * returned 0.
2712 */
2713void skb_abort_seq_read(struct skb_seq_state *st)
2714{
2715 if (st->frag_data)
Eric Dumazet51c56b02012-04-05 11:35:15 +02002716 kunmap_atomic(st->frag_data);
Thomas Graf677e90e2005-06-23 20:59:51 -07002717}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002718EXPORT_SYMBOL(skb_abort_seq_read);
Thomas Graf677e90e2005-06-23 20:59:51 -07002719
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002720#define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2721
2722static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2723 struct ts_config *conf,
2724 struct ts_state *state)
2725{
2726 return skb_seq_read(offset, text, TS_SKB_CB(state));
2727}
2728
2729static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2730{
2731 skb_abort_seq_read(TS_SKB_CB(state));
2732}
2733
2734/**
2735 * skb_find_text - Find a text pattern in skb data
2736 * @skb: the buffer to look in
2737 * @from: search offset
2738 * @to: search limit
2739 * @config: textsearch configuration
2740 * @state: uninitialized textsearch state variable
2741 *
2742 * Finds a pattern in the skb data according to the specified
2743 * textsearch configuration. Use textsearch_next() to retrieve
2744 * subsequent occurrences of the pattern. Returns the offset
2745 * to the first occurrence or UINT_MAX if no match was found.
2746 */
2747unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2748 unsigned int to, struct ts_config *config,
2749 struct ts_state *state)
2750{
Phil Oesterf72b9482006-06-26 00:00:57 -07002751 unsigned int ret;
2752
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002753 config->get_next_block = skb_ts_get_next_block;
2754 config->finish = skb_ts_finish;
2755
2756 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2757
Phil Oesterf72b9482006-06-26 00:00:57 -07002758 ret = textsearch_find(config, state);
2759 return (ret <= to - from ? ret : UINT_MAX);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002760}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002761EXPORT_SYMBOL(skb_find_text);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07002762
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002763/**
Ben Hutchings2c530402012-07-10 10:55:09 +00002764 * skb_append_datato_frags - append the user data to a skb
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002765 * @sk: sock structure
2766 * @skb: skb structure to be appened with user data.
2767 * @getfrag: call back function to be used for getting the user data
2768 * @from: pointer to user message iov
2769 * @length: length of the iov message
2770 *
2771 * Description: This procedure append the user data in the fragment part
2772 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2773 */
2774int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
Martin Waitzdab96302005-12-05 13:40:12 -08002775 int (*getfrag)(void *from, char *to, int offset,
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002776 int len, int odd, struct sk_buff *skb),
2777 void *from, int length)
2778{
Eric Dumazetb2111722012-12-28 06:06:37 +00002779 int frg_cnt = skb_shinfo(skb)->nr_frags;
2780 int copy;
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002781 int offset = 0;
2782 int ret;
Eric Dumazetb2111722012-12-28 06:06:37 +00002783 struct page_frag *pfrag = &current->task_frag;
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002784
2785 do {
2786 /* Return error if we don't have space for new frag */
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002787 if (frg_cnt >= MAX_SKB_FRAGS)
Eric Dumazetb2111722012-12-28 06:06:37 +00002788 return -EMSGSIZE;
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002789
Eric Dumazetb2111722012-12-28 06:06:37 +00002790 if (!sk_page_frag_refill(sk, pfrag))
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002791 return -ENOMEM;
2792
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002793 /* copy the user data to page */
Eric Dumazetb2111722012-12-28 06:06:37 +00002794 copy = min_t(int, length, pfrag->size - pfrag->offset);
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002795
Eric Dumazetb2111722012-12-28 06:06:37 +00002796 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
2797 offset, copy, 0, skb);
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002798 if (ret < 0)
2799 return -EFAULT;
2800
2801 /* copy was successful so update the size parameters */
Eric Dumazetb2111722012-12-28 06:06:37 +00002802 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
2803 copy);
2804 frg_cnt++;
2805 pfrag->offset += copy;
2806 get_page(pfrag->page);
2807
2808 skb->truesize += copy;
2809 atomic_add(copy, &sk->sk_wmem_alloc);
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002810 skb->len += copy;
2811 skb->data_len += copy;
2812 offset += copy;
2813 length -= copy;
2814
2815 } while (length > 0);
2816
2817 return 0;
2818}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08002819EXPORT_SYMBOL(skb_append_datato_frags);
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002820
Herbert Xucbb042f2006-03-20 22:43:56 -08002821/**
2822 * skb_pull_rcsum - pull skb and update receive checksum
2823 * @skb: buffer to update
Herbert Xucbb042f2006-03-20 22:43:56 -08002824 * @len: length of data pulled
2825 *
2826 * This function performs an skb_pull on the packet and updates
Urs Thuermannfee54fa2008-02-12 22:03:25 -08002827 * the CHECKSUM_COMPLETE checksum. It should be used on
Patrick McHardy84fa7932006-08-29 16:44:56 -07002828 * receive path processing instead of skb_pull unless you know
2829 * that the checksum difference is zero (e.g., a valid IP header)
2830 * or you are setting ip_summed to CHECKSUM_NONE.
Herbert Xucbb042f2006-03-20 22:43:56 -08002831 */
2832unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2833{
2834 BUG_ON(len > skb->len);
2835 skb->len -= len;
2836 BUG_ON(skb->len < skb->data_len);
2837 skb_postpull_rcsum(skb, skb->data, len);
2838 return skb->data += len;
2839}
Arnaldo Carvalho de Melof94691a2006-03-20 22:47:55 -08002840EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2841
Herbert Xuf4c50d92006-06-22 03:02:40 -07002842/**
2843 * skb_segment - Perform protocol segmentation on skb.
2844 * @skb: buffer to segment
Herbert Xu576a30e2006-06-27 13:22:38 -07002845 * @features: features for the output path (see dev->features)
Herbert Xuf4c50d92006-06-22 03:02:40 -07002846 *
2847 * This function performs segmentation on the given skb. It returns
Ben Hutchings4c821d72008-04-13 21:52:48 -07002848 * a pointer to the first in a list of new skbs for the segments.
2849 * In case of error it returns ERR_PTR(err).
Herbert Xuf4c50d92006-06-22 03:02:40 -07002850 */
Michał Mirosławc8f44af2011-11-15 15:29:55 +00002851struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features)
Herbert Xuf4c50d92006-06-22 03:02:40 -07002852{
2853 struct sk_buff *segs = NULL;
2854 struct sk_buff *tail = NULL;
Herbert Xu89319d382008-12-15 23:26:06 -08002855 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
Herbert Xu9d8506c2013-11-21 11:10:04 -08002856 skb_frag_t *skb_frag = skb_shinfo(skb)->frags;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002857 unsigned int mss = skb_shinfo(skb)->gso_size;
Arnaldo Carvalho de Melo98e399f2007-03-19 15:33:04 -07002858 unsigned int doffset = skb->data - skb_mac_header(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002859 unsigned int offset = doffset;
Pravin B Shelar68c33162013-02-14 14:02:41 +00002860 unsigned int tnl_hlen = skb_tnl_header_len(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002861 unsigned int headroom;
2862 unsigned int len;
Pravin B Shelarec5f0612013-03-07 09:28:01 +00002863 __be16 proto;
2864 bool csum;
Michał Mirosław04ed3e72011-01-24 15:32:47 -08002865 int sg = !!(features & NETIF_F_SG);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002866 int nfrags = skb_shinfo(skb)->nr_frags;
2867 int err = -ENOMEM;
2868 int i = 0;
2869 int pos;
2870
Pravin B Shelarec5f0612013-03-07 09:28:01 +00002871 proto = skb_network_protocol(skb);
2872 if (unlikely(!proto))
2873 return ERR_PTR(-EINVAL);
2874
2875 csum = !!can_checksum_protocol(features, proto);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002876 __skb_push(skb, doffset);
2877 headroom = skb_headroom(skb);
2878 pos = skb_headlen(skb);
2879
2880 do {
2881 struct sk_buff *nskb;
2882 skb_frag_t *frag;
Herbert Xuc8884ed2006-10-29 15:59:41 -08002883 int hsize;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002884 int size;
2885
2886 len = skb->len - offset;
2887 if (len > mss)
2888 len = mss;
2889
2890 hsize = skb_headlen(skb) - offset;
2891 if (hsize < 0)
2892 hsize = 0;
Herbert Xuc8884ed2006-10-29 15:59:41 -08002893 if (hsize > len || !sg)
2894 hsize = len;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002895
Herbert Xu9d8506c2013-11-21 11:10:04 -08002896 if (!hsize && i >= nfrags && skb_headlen(fskb) &&
2897 (skb_headlen(fskb) == len || sg)) {
2898 BUG_ON(skb_headlen(fskb) > len);
Herbert Xu89319d382008-12-15 23:26:06 -08002899
Herbert Xu9d8506c2013-11-21 11:10:04 -08002900 i = 0;
2901 nfrags = skb_shinfo(fskb)->nr_frags;
2902 skb_frag = skb_shinfo(fskb)->frags;
2903 pos += skb_headlen(fskb);
2904
2905 while (pos < offset + len) {
2906 BUG_ON(i >= nfrags);
2907
2908 size = skb_frag_size(skb_frag);
2909 if (pos + size > offset + len)
2910 break;
2911
2912 i++;
2913 pos += size;
2914 skb_frag++;
2915 }
2916
Herbert Xu89319d382008-12-15 23:26:06 -08002917 nskb = skb_clone(fskb, GFP_ATOMIC);
2918 fskb = fskb->next;
2919
2920 if (unlikely(!nskb))
2921 goto err;
2922
Herbert Xu9d8506c2013-11-21 11:10:04 -08002923 if (unlikely(pskb_trim(nskb, len))) {
2924 kfree_skb(nskb);
2925 goto err;
2926 }
2927
Alexander Duyckec47ea82012-05-04 14:26:56 +00002928 hsize = skb_end_offset(nskb);
Herbert Xu89319d382008-12-15 23:26:06 -08002929 if (skb_cow_head(nskb, doffset + headroom)) {
2930 kfree_skb(nskb);
2931 goto err;
2932 }
2933
Alexander Duyckec47ea82012-05-04 14:26:56 +00002934 nskb->truesize += skb_end_offset(nskb) - hsize;
Herbert Xu89319d382008-12-15 23:26:06 -08002935 skb_release_head_state(nskb);
2936 __skb_push(nskb, doffset);
2937 } else {
Mel Gormanc93bdd02012-07-31 16:44:19 -07002938 nskb = __alloc_skb(hsize + doffset + headroom,
2939 GFP_ATOMIC, skb_alloc_rx_flag(skb),
2940 NUMA_NO_NODE);
Herbert Xu89319d382008-12-15 23:26:06 -08002941
2942 if (unlikely(!nskb))
2943 goto err;
2944
2945 skb_reserve(nskb, headroom);
2946 __skb_put(nskb, doffset);
2947 }
Herbert Xuf4c50d92006-06-22 03:02:40 -07002948
2949 if (segs)
2950 tail->next = nskb;
2951 else
2952 segs = nskb;
2953 tail = nskb;
2954
Herbert Xu6f85a122008-08-15 14:55:02 -07002955 __copy_skb_header(nskb, skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002956 nskb->mac_len = skb->mac_len;
2957
Eric Dumazet030737b2013-10-19 11:42:54 -07002958 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
Pravin B Shelar68c33162013-02-14 14:02:41 +00002959
2960 skb_copy_from_linear_data_offset(skb, -tnl_hlen,
2961 nskb->data - tnl_hlen,
2962 doffset + tnl_hlen);
Herbert Xu89319d382008-12-15 23:26:06 -08002963
Herbert Xu9d8506c2013-11-21 11:10:04 -08002964 if (nskb->len == len + doffset)
Simon Horman1cdbcb72013-05-19 15:46:49 +00002965 goto perform_csum_check;
Herbert Xu89319d382008-12-15 23:26:06 -08002966
Herbert Xuf4c50d92006-06-22 03:02:40 -07002967 if (!sg) {
Herbert Xu6f85a122008-08-15 14:55:02 -07002968 nskb->ip_summed = CHECKSUM_NONE;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002969 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2970 skb_put(nskb, len),
2971 len, 0);
2972 continue;
2973 }
2974
2975 frag = skb_shinfo(nskb)->frags;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002976
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03002977 skb_copy_from_linear_data_offset(skb, offset,
2978 skb_put(nskb, hsize), hsize);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002979
Pravin B Shelarc9af6db2013-02-11 09:27:41 +00002980 skb_shinfo(nskb)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
Eric Dumazetcef401d2013-01-25 20:34:37 +00002981
Herbert Xu9d8506c2013-11-21 11:10:04 -08002982 while (pos < offset + len) {
2983 if (i >= nfrags) {
2984 BUG_ON(skb_headlen(fskb));
2985
2986 i = 0;
2987 nfrags = skb_shinfo(fskb)->nr_frags;
2988 skb_frag = skb_shinfo(fskb)->frags;
2989
2990 BUG_ON(!nfrags);
2991
2992 fskb = fskb->next;
2993 }
2994
2995 if (unlikely(skb_shinfo(nskb)->nr_frags >=
2996 MAX_SKB_FRAGS)) {
2997 net_warn_ratelimited(
2998 "skb_segment: too many frags: %u %u\n",
2999 pos, mss);
3000 goto err;
3001 }
3002
3003 *frag = *skb_frag;
Ian Campbellea2ab692011-08-22 23:44:58 +00003004 __skb_frag_ref(frag);
Eric Dumazet9e903e02011-10-18 21:00:24 +00003005 size = skb_frag_size(frag);
Herbert Xuf4c50d92006-06-22 03:02:40 -07003006
3007 if (pos < offset) {
3008 frag->page_offset += offset - pos;
Eric Dumazet9e903e02011-10-18 21:00:24 +00003009 skb_frag_size_sub(frag, offset - pos);
Herbert Xuf4c50d92006-06-22 03:02:40 -07003010 }
3011
Herbert Xu89319d382008-12-15 23:26:06 -08003012 skb_shinfo(nskb)->nr_frags++;
Herbert Xuf4c50d92006-06-22 03:02:40 -07003013
3014 if (pos + size <= offset + len) {
3015 i++;
Herbert Xu9d8506c2013-11-21 11:10:04 -08003016 skb_frag++;
Herbert Xuf4c50d92006-06-22 03:02:40 -07003017 pos += size;
3018 } else {
Eric Dumazet9e903e02011-10-18 21:00:24 +00003019 skb_frag_size_sub(frag, pos + size - (offset + len));
Herbert Xu89319d382008-12-15 23:26:06 -08003020 goto skip_fraglist;
Herbert Xuf4c50d92006-06-22 03:02:40 -07003021 }
3022
3023 frag++;
3024 }
3025
Herbert Xu89319d382008-12-15 23:26:06 -08003026skip_fraglist:
Herbert Xuf4c50d92006-06-22 03:02:40 -07003027 nskb->data_len = len - hsize;
3028 nskb->len += nskb->data_len;
3029 nskb->truesize += nskb->data_len;
Pravin B Shelarec5f0612013-03-07 09:28:01 +00003030
Simon Horman1cdbcb72013-05-19 15:46:49 +00003031perform_csum_check:
Pravin B Shelarec5f0612013-03-07 09:28:01 +00003032 if (!csum) {
3033 nskb->csum = skb_checksum(nskb, doffset,
3034 nskb->len - doffset, 0);
3035 nskb->ip_summed = CHECKSUM_NONE;
3036 }
Herbert Xuf4c50d92006-06-22 03:02:40 -07003037 } while ((offset += len) < skb->len);
3038
3039 return segs;
3040
3041err:
Eric Dumazet289dccb2013-12-20 14:29:08 -08003042 kfree_skb_list(segs);
Herbert Xuf4c50d92006-06-22 03:02:40 -07003043 return ERR_PTR(err);
3044}
Herbert Xuf4c50d92006-06-22 03:02:40 -07003045EXPORT_SYMBOL_GPL(skb_segment);
3046
Herbert Xu71d93b32008-12-15 23:42:33 -08003047int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
3048{
Eric Dumazet8a291112013-10-08 09:02:23 -07003049 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
Herbert Xu67147ba2009-05-26 18:50:22 +00003050 unsigned int offset = skb_gro_offset(skb);
3051 unsigned int headlen = skb_headlen(skb);
Eric Dumazet8a291112013-10-08 09:02:23 -07003052 struct sk_buff *nskb, *lp, *p = *head;
3053 unsigned int len = skb_gro_len(skb);
Eric Dumazet715dc1f2012-05-02 23:33:21 +00003054 unsigned int delta_truesize;
Eric Dumazet8a291112013-10-08 09:02:23 -07003055 unsigned int headroom;
Herbert Xu71d93b32008-12-15 23:42:33 -08003056
Eric Dumazet8a291112013-10-08 09:02:23 -07003057 if (unlikely(p->len + len >= 65536))
Herbert Xu71d93b32008-12-15 23:42:33 -08003058 return -E2BIG;
3059
Eric Dumazet8a291112013-10-08 09:02:23 -07003060 lp = NAPI_GRO_CB(p)->last ?: p;
3061 pinfo = skb_shinfo(lp);
3062
3063 if (headlen <= offset) {
Herbert Xu42da6992009-05-26 18:50:19 +00003064 skb_frag_t *frag;
Herbert Xu66e92fc2009-05-26 18:50:32 +00003065 skb_frag_t *frag2;
Herbert Xu9aaa1562009-05-26 18:50:33 +00003066 int i = skbinfo->nr_frags;
3067 int nr_frags = pinfo->nr_frags + i;
Herbert Xu42da6992009-05-26 18:50:19 +00003068
Herbert Xu66e92fc2009-05-26 18:50:32 +00003069 if (nr_frags > MAX_SKB_FRAGS)
Eric Dumazet8a291112013-10-08 09:02:23 -07003070 goto merge;
Herbert Xu81705ad2009-01-29 14:19:51 +00003071
Eric Dumazet8a291112013-10-08 09:02:23 -07003072 offset -= headlen;
Herbert Xu9aaa1562009-05-26 18:50:33 +00003073 pinfo->nr_frags = nr_frags;
3074 skbinfo->nr_frags = 0;
Herbert Xuf5572062009-01-14 20:40:03 -08003075
Herbert Xu9aaa1562009-05-26 18:50:33 +00003076 frag = pinfo->frags + nr_frags;
3077 frag2 = skbinfo->frags + i;
Herbert Xu66e92fc2009-05-26 18:50:32 +00003078 do {
3079 *--frag = *--frag2;
3080 } while (--i);
3081
3082 frag->page_offset += offset;
Eric Dumazet9e903e02011-10-18 21:00:24 +00003083 skb_frag_size_sub(frag, offset);
Herbert Xu66e92fc2009-05-26 18:50:32 +00003084
Eric Dumazet715dc1f2012-05-02 23:33:21 +00003085 /* all fragments truesize : remove (head size + sk_buff) */
Alexander Duyckec47ea82012-05-04 14:26:56 +00003086 delta_truesize = skb->truesize -
3087 SKB_TRUESIZE(skb_end_offset(skb));
Eric Dumazet715dc1f2012-05-02 23:33:21 +00003088
Herbert Xuf5572062009-01-14 20:40:03 -08003089 skb->truesize -= skb->data_len;
3090 skb->len -= skb->data_len;
3091 skb->data_len = 0;
3092
Eric Dumazet715dc1f2012-05-02 23:33:21 +00003093 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
Herbert Xu5d38a072009-01-04 16:13:40 -08003094 goto done;
Eric Dumazetd7e88832012-04-30 08:10:34 +00003095 } else if (skb->head_frag) {
3096 int nr_frags = pinfo->nr_frags;
3097 skb_frag_t *frag = pinfo->frags + nr_frags;
3098 struct page *page = virt_to_head_page(skb->head);
3099 unsigned int first_size = headlen - offset;
3100 unsigned int first_offset;
3101
3102 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
Eric Dumazet8a291112013-10-08 09:02:23 -07003103 goto merge;
Eric Dumazetd7e88832012-04-30 08:10:34 +00003104
3105 first_offset = skb->data -
3106 (unsigned char *)page_address(page) +
3107 offset;
3108
3109 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3110
3111 frag->page.p = page;
3112 frag->page_offset = first_offset;
3113 skb_frag_size_set(frag, first_size);
3114
3115 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3116 /* We dont need to clear skbinfo->nr_frags here */
3117
Eric Dumazet715dc1f2012-05-02 23:33:21 +00003118 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
Eric Dumazetd7e88832012-04-30 08:10:34 +00003119 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3120 goto done;
Eric Dumazet8a291112013-10-08 09:02:23 -07003121 }
3122 if (pinfo->frag_list)
3123 goto merge;
3124 if (skb_gro_len(p) != pinfo->gso_size)
Herbert Xu69c0cab2009-11-17 05:18:18 -08003125 return -E2BIG;
Herbert Xu71d93b32008-12-15 23:42:33 -08003126
3127 headroom = skb_headroom(p);
Eric Dumazet3d3be432010-09-01 00:50:51 +00003128 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
Herbert Xu71d93b32008-12-15 23:42:33 -08003129 if (unlikely(!nskb))
3130 return -ENOMEM;
3131
3132 __copy_skb_header(nskb, p);
3133 nskb->mac_len = p->mac_len;
3134
3135 skb_reserve(nskb, headroom);
Herbert Xu86911732009-01-29 14:19:50 +00003136 __skb_put(nskb, skb_gro_offset(p));
Herbert Xu71d93b32008-12-15 23:42:33 -08003137
Herbert Xu86911732009-01-29 14:19:50 +00003138 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
Herbert Xu71d93b32008-12-15 23:42:33 -08003139 skb_set_network_header(nskb, skb_network_offset(p));
3140 skb_set_transport_header(nskb, skb_transport_offset(p));
3141
Herbert Xu86911732009-01-29 14:19:50 +00003142 __skb_pull(p, skb_gro_offset(p));
3143 memcpy(skb_mac_header(nskb), skb_mac_header(p),
3144 p->data - skb_mac_header(p));
Herbert Xu71d93b32008-12-15 23:42:33 -08003145
Herbert Xu71d93b32008-12-15 23:42:33 -08003146 skb_shinfo(nskb)->frag_list = p;
Herbert Xu9aaa1562009-05-26 18:50:33 +00003147 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
Herbert Xu622e0ca2010-05-20 23:07:56 -07003148 pinfo->gso_size = 0;
Herbert Xu71d93b32008-12-15 23:42:33 -08003149 skb_header_release(p);
Eric Dumazetc3c7c252012-12-06 13:54:59 +00003150 NAPI_GRO_CB(nskb)->last = p;
Herbert Xu71d93b32008-12-15 23:42:33 -08003151
3152 nskb->data_len += p->len;
Eric Dumazetde8261c2012-02-13 04:09:20 +00003153 nskb->truesize += p->truesize;
Herbert Xu71d93b32008-12-15 23:42:33 -08003154 nskb->len += p->len;
3155
3156 *head = nskb;
3157 nskb->next = p->next;
3158 p->next = NULL;
3159
3160 p = nskb;
3161
3162merge:
Eric Dumazet715dc1f2012-05-02 23:33:21 +00003163 delta_truesize = skb->truesize;
Herbert Xu67147ba2009-05-26 18:50:22 +00003164 if (offset > headlen) {
Michal Schmidtd1dc7ab2011-01-24 12:08:48 +00003165 unsigned int eat = offset - headlen;
3166
3167 skbinfo->frags[0].page_offset += eat;
Eric Dumazet9e903e02011-10-18 21:00:24 +00003168 skb_frag_size_sub(&skbinfo->frags[0], eat);
Michal Schmidtd1dc7ab2011-01-24 12:08:48 +00003169 skb->data_len -= eat;
3170 skb->len -= eat;
Herbert Xu67147ba2009-05-26 18:50:22 +00003171 offset = headlen;
Herbert Xu56035022009-02-05 21:26:52 -08003172 }
3173
Herbert Xu67147ba2009-05-26 18:50:22 +00003174 __skb_pull(skb, offset);
Herbert Xu56035022009-02-05 21:26:52 -08003175
Eric Dumazet8a291112013-10-08 09:02:23 -07003176 if (!NAPI_GRO_CB(p)->last)
3177 skb_shinfo(p)->frag_list = skb;
3178 else
3179 NAPI_GRO_CB(p)->last->next = skb;
Eric Dumazetc3c7c252012-12-06 13:54:59 +00003180 NAPI_GRO_CB(p)->last = skb;
Herbert Xu71d93b32008-12-15 23:42:33 -08003181 skb_header_release(skb);
Eric Dumazet8a291112013-10-08 09:02:23 -07003182 lp = p;
Herbert Xu71d93b32008-12-15 23:42:33 -08003183
Herbert Xu5d38a072009-01-04 16:13:40 -08003184done:
3185 NAPI_GRO_CB(p)->count++;
Herbert Xu37fe4732009-01-17 19:48:13 +00003186 p->data_len += len;
Eric Dumazet715dc1f2012-05-02 23:33:21 +00003187 p->truesize += delta_truesize;
Herbert Xu37fe4732009-01-17 19:48:13 +00003188 p->len += len;
Eric Dumazet8a291112013-10-08 09:02:23 -07003189 if (lp != p) {
3190 lp->data_len += len;
3191 lp->truesize += delta_truesize;
3192 lp->len += len;
3193 }
Herbert Xu71d93b32008-12-15 23:42:33 -08003194 NAPI_GRO_CB(skb)->same_flow = 1;
3195 return 0;
3196}
3197EXPORT_SYMBOL_GPL(skb_gro_receive);
3198
Linus Torvalds1da177e2005-04-16 15:20:36 -07003199void __init skb_init(void)
3200{
3201 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3202 sizeof(struct sk_buff),
3203 0,
Alexey Dobriyane5d679f332006-08-26 19:25:52 -07003204 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
Paul Mundt20c2df82007-07-20 10:11:58 +09003205 NULL);
David S. Millerd179cd12005-08-17 14:57:30 -07003206 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3207 (2*sizeof(struct sk_buff)) +
3208 sizeof(atomic_t),
3209 0,
Alexey Dobriyane5d679f332006-08-26 19:25:52 -07003210 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
Paul Mundt20c2df82007-07-20 10:11:58 +09003211 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003212}
3213
David Howells716ea3a2007-04-02 20:19:53 -07003214/**
3215 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3216 * @skb: Socket buffer containing the buffers to be mapped
3217 * @sg: The scatter-gather list to map into
3218 * @offset: The offset into the buffer's contents to start mapping
3219 * @len: Length of buffer space to be mapped
3220 *
3221 * Fill the specified scatter-gather list with mappings/pointers into a
3222 * region of the buffer space attached to a socket buffer.
3223 */
David S. Miller51c739d2007-10-30 21:29:29 -07003224static int
3225__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
David Howells716ea3a2007-04-02 20:19:53 -07003226{
David S. Miller1a028e52007-04-27 15:21:23 -07003227 int start = skb_headlen(skb);
3228 int i, copy = start - offset;
David S. Millerfbb398a2009-06-09 00:18:59 -07003229 struct sk_buff *frag_iter;
David Howells716ea3a2007-04-02 20:19:53 -07003230 int elt = 0;
3231
3232 if (copy > 0) {
3233 if (copy > len)
3234 copy = len;
Jens Axboe642f1492007-10-24 11:20:47 +02003235 sg_set_buf(sg, skb->data + offset, copy);
David Howells716ea3a2007-04-02 20:19:53 -07003236 elt++;
3237 if ((len -= copy) == 0)
3238 return elt;
3239 offset += copy;
3240 }
3241
3242 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -07003243 int end;
David Howells716ea3a2007-04-02 20:19:53 -07003244
Ilpo Järvinen547b7922008-07-25 21:43:18 -07003245 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -07003246
Eric Dumazet9e903e02011-10-18 21:00:24 +00003247 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
David Howells716ea3a2007-04-02 20:19:53 -07003248 if ((copy = end - offset) > 0) {
3249 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3250
3251 if (copy > len)
3252 copy = len;
Ian Campbellea2ab692011-08-22 23:44:58 +00003253 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
Jens Axboe642f1492007-10-24 11:20:47 +02003254 frag->page_offset+offset-start);
David Howells716ea3a2007-04-02 20:19:53 -07003255 elt++;
3256 if (!(len -= copy))
3257 return elt;
3258 offset += copy;
3259 }
David S. Miller1a028e52007-04-27 15:21:23 -07003260 start = end;
David Howells716ea3a2007-04-02 20:19:53 -07003261 }
3262
David S. Millerfbb398a2009-06-09 00:18:59 -07003263 skb_walk_frags(skb, frag_iter) {
3264 int end;
David Howells716ea3a2007-04-02 20:19:53 -07003265
David S. Millerfbb398a2009-06-09 00:18:59 -07003266 WARN_ON(start > offset + len);
David Howells716ea3a2007-04-02 20:19:53 -07003267
David S. Millerfbb398a2009-06-09 00:18:59 -07003268 end = start + frag_iter->len;
3269 if ((copy = end - offset) > 0) {
3270 if (copy > len)
3271 copy = len;
3272 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3273 copy);
3274 if ((len -= copy) == 0)
3275 return elt;
3276 offset += copy;
David Howells716ea3a2007-04-02 20:19:53 -07003277 }
David S. Millerfbb398a2009-06-09 00:18:59 -07003278 start = end;
David Howells716ea3a2007-04-02 20:19:53 -07003279 }
3280 BUG_ON(len);
3281 return elt;
3282}
3283
David S. Miller51c739d2007-10-30 21:29:29 -07003284int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3285{
3286 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3287
Jens Axboec46f2332007-10-31 12:06:37 +01003288 sg_mark_end(&sg[nsg - 1]);
David S. Miller51c739d2007-10-30 21:29:29 -07003289
3290 return nsg;
3291}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08003292EXPORT_SYMBOL_GPL(skb_to_sgvec);
David S. Miller51c739d2007-10-30 21:29:29 -07003293
David Howells716ea3a2007-04-02 20:19:53 -07003294/**
3295 * skb_cow_data - Check that a socket buffer's data buffers are writable
3296 * @skb: The socket buffer to check.
3297 * @tailbits: Amount of trailing space to be added
3298 * @trailer: Returned pointer to the skb where the @tailbits space begins
3299 *
3300 * Make sure that the data buffers attached to a socket buffer are
3301 * writable. If they are not, private copies are made of the data buffers
3302 * and the socket buffer is set to use these instead.
3303 *
3304 * If @tailbits is given, make sure that there is space to write @tailbits
3305 * bytes of data beyond current end of socket buffer. @trailer will be
3306 * set to point to the skb in which this space begins.
3307 *
3308 * The number of scatterlist elements required to completely map the
3309 * COW'd and extended socket buffer will be returned.
3310 */
3311int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3312{
3313 int copyflag;
3314 int elt;
3315 struct sk_buff *skb1, **skb_p;
3316
3317 /* If skb is cloned or its head is paged, reallocate
3318 * head pulling out all the pages (pages are considered not writable
3319 * at the moment even if they are anonymous).
3320 */
3321 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3322 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3323 return -ENOMEM;
3324
3325 /* Easy case. Most of packets will go this way. */
David S. Miller21dc3302010-08-23 00:13:46 -07003326 if (!skb_has_frag_list(skb)) {
David Howells716ea3a2007-04-02 20:19:53 -07003327 /* A little of trouble, not enough of space for trailer.
3328 * This should not happen, when stack is tuned to generate
3329 * good frames. OK, on miss we reallocate and reserve even more
3330 * space, 128 bytes is fair. */
3331
3332 if (skb_tailroom(skb) < tailbits &&
3333 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3334 return -ENOMEM;
3335
3336 /* Voila! */
3337 *trailer = skb;
3338 return 1;
3339 }
3340
3341 /* Misery. We are in troubles, going to mincer fragments... */
3342
3343 elt = 1;
3344 skb_p = &skb_shinfo(skb)->frag_list;
3345 copyflag = 0;
3346
3347 while ((skb1 = *skb_p) != NULL) {
3348 int ntail = 0;
3349
3350 /* The fragment is partially pulled by someone,
3351 * this can happen on input. Copy it and everything
3352 * after it. */
3353
3354 if (skb_shared(skb1))
3355 copyflag = 1;
3356
3357 /* If the skb is the last, worry about trailer. */
3358
3359 if (skb1->next == NULL && tailbits) {
3360 if (skb_shinfo(skb1)->nr_frags ||
David S. Miller21dc3302010-08-23 00:13:46 -07003361 skb_has_frag_list(skb1) ||
David Howells716ea3a2007-04-02 20:19:53 -07003362 skb_tailroom(skb1) < tailbits)
3363 ntail = tailbits + 128;
3364 }
3365
3366 if (copyflag ||
3367 skb_cloned(skb1) ||
3368 ntail ||
3369 skb_shinfo(skb1)->nr_frags ||
David S. Miller21dc3302010-08-23 00:13:46 -07003370 skb_has_frag_list(skb1)) {
David Howells716ea3a2007-04-02 20:19:53 -07003371 struct sk_buff *skb2;
3372
3373 /* Fuck, we are miserable poor guys... */
3374 if (ntail == 0)
3375 skb2 = skb_copy(skb1, GFP_ATOMIC);
3376 else
3377 skb2 = skb_copy_expand(skb1,
3378 skb_headroom(skb1),
3379 ntail,
3380 GFP_ATOMIC);
3381 if (unlikely(skb2 == NULL))
3382 return -ENOMEM;
3383
3384 if (skb1->sk)
3385 skb_set_owner_w(skb2, skb1->sk);
3386
3387 /* Looking around. Are we still alive?
3388 * OK, link new skb, drop old one */
3389
3390 skb2->next = skb1->next;
3391 *skb_p = skb2;
3392 kfree_skb(skb1);
3393 skb1 = skb2;
3394 }
3395 elt++;
3396 *trailer = skb1;
3397 skb_p = &skb1->next;
3398 }
3399
3400 return elt;
3401}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08003402EXPORT_SYMBOL_GPL(skb_cow_data);
David Howells716ea3a2007-04-02 20:19:53 -07003403
Eric Dumazetb1faf562010-05-31 23:44:05 -07003404static void sock_rmem_free(struct sk_buff *skb)
3405{
3406 struct sock *sk = skb->sk;
3407
3408 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3409}
3410
3411/*
3412 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3413 */
3414int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3415{
Eric Dumazet110c4332012-04-06 10:49:10 +02003416 int len = skb->len;
3417
Eric Dumazetb1faf562010-05-31 23:44:05 -07003418 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
Eric Dumazet95c96172012-04-15 05:58:06 +00003419 (unsigned int)sk->sk_rcvbuf)
Eric Dumazetb1faf562010-05-31 23:44:05 -07003420 return -ENOMEM;
3421
3422 skb_orphan(skb);
3423 skb->sk = sk;
3424 skb->destructor = sock_rmem_free;
3425 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3426
Eric Dumazetabb57ea2011-05-18 02:21:31 -04003427 /* before exiting rcu section, make sure dst is refcounted */
3428 skb_dst_force(skb);
3429
Eric Dumazetb1faf562010-05-31 23:44:05 -07003430 skb_queue_tail(&sk->sk_error_queue, skb);
3431 if (!sock_flag(sk, SOCK_DEAD))
Eric Dumazet110c4332012-04-06 10:49:10 +02003432 sk->sk_data_ready(sk, len);
Eric Dumazetb1faf562010-05-31 23:44:05 -07003433 return 0;
3434}
3435EXPORT_SYMBOL(sock_queue_err_skb);
3436
Patrick Ohlyac45f602009-02-12 05:03:37 +00003437void skb_tstamp_tx(struct sk_buff *orig_skb,
3438 struct skb_shared_hwtstamps *hwtstamps)
3439{
3440 struct sock *sk = orig_skb->sk;
3441 struct sock_exterr_skb *serr;
3442 struct sk_buff *skb;
3443 int err;
3444
3445 if (!sk)
3446 return;
3447
Patrick Ohlyac45f602009-02-12 05:03:37 +00003448 if (hwtstamps) {
Willem de Bruijn2e313962013-04-23 00:39:28 +00003449 *skb_hwtstamps(orig_skb) =
Patrick Ohlyac45f602009-02-12 05:03:37 +00003450 *hwtstamps;
3451 } else {
3452 /*
3453 * no hardware time stamps available,
Oliver Hartkopp2244d072010-08-17 08:59:14 +00003454 * so keep the shared tx_flags and only
Patrick Ohlyac45f602009-02-12 05:03:37 +00003455 * store software time stamp
3456 */
Willem de Bruijn2e313962013-04-23 00:39:28 +00003457 orig_skb->tstamp = ktime_get_real();
Patrick Ohlyac45f602009-02-12 05:03:37 +00003458 }
3459
Willem de Bruijn2e313962013-04-23 00:39:28 +00003460 skb = skb_clone(orig_skb, GFP_ATOMIC);
3461 if (!skb)
3462 return;
3463
Patrick Ohlyac45f602009-02-12 05:03:37 +00003464 serr = SKB_EXT_ERR(skb);
3465 memset(serr, 0, sizeof(*serr));
3466 serr->ee.ee_errno = ENOMSG;
3467 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
Eric Dumazet29030372010-05-29 00:20:48 -07003468
Patrick Ohlyac45f602009-02-12 05:03:37 +00003469 err = sock_queue_err_skb(sk, skb);
Eric Dumazet29030372010-05-29 00:20:48 -07003470
Patrick Ohlyac45f602009-02-12 05:03:37 +00003471 if (err)
3472 kfree_skb(skb);
3473}
3474EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3475
Johannes Berg6e3e9392011-11-09 10:15:42 +01003476void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3477{
3478 struct sock *sk = skb->sk;
3479 struct sock_exterr_skb *serr;
3480 int err;
3481
3482 skb->wifi_acked_valid = 1;
3483 skb->wifi_acked = acked;
3484
3485 serr = SKB_EXT_ERR(skb);
3486 memset(serr, 0, sizeof(*serr));
3487 serr->ee.ee_errno = ENOMSG;
3488 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3489
3490 err = sock_queue_err_skb(sk, skb);
3491 if (err)
3492 kfree_skb(skb);
3493}
3494EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3495
Patrick Ohlyac45f602009-02-12 05:03:37 +00003496
Rusty Russellf35d9d82008-02-04 23:49:54 -05003497/**
3498 * skb_partial_csum_set - set up and verify partial csum values for packet
3499 * @skb: the skb to set
3500 * @start: the number of bytes after skb->data to start checksumming.
3501 * @off: the offset from start to place the checksum.
3502 *
3503 * For untrusted partially-checksummed packets, we need to make sure the values
3504 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3505 *
3506 * This function checks and sets those values and skb->ip_summed: if this
3507 * returns false you should drop the packet.
3508 */
3509bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3510{
Herbert Xu5ff8dda2009-06-04 01:22:01 +00003511 if (unlikely(start > skb_headlen(skb)) ||
3512 unlikely((int)start + off > skb_headlen(skb) - 2)) {
Joe Perchese87cc472012-05-13 21:56:26 +00003513 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3514 start, off, skb_headlen(skb));
Rusty Russellf35d9d82008-02-04 23:49:54 -05003515 return false;
3516 }
3517 skb->ip_summed = CHECKSUM_PARTIAL;
3518 skb->csum_start = skb_headroom(skb) + start;
3519 skb->csum_offset = off;
Jason Wange5d5dec2013-03-26 23:11:20 +00003520 skb_set_transport_header(skb, start);
Rusty Russellf35d9d82008-02-04 23:49:54 -05003521 return true;
3522}
David S. Millerb4ac530fc2009-02-10 02:09:24 -08003523EXPORT_SYMBOL_GPL(skb_partial_csum_set);
Rusty Russellf35d9d82008-02-04 23:49:54 -05003524
Paul Durranted1f50c2014-01-09 10:02:46 +00003525static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
3526 unsigned int max)
3527{
3528 if (skb_headlen(skb) >= len)
3529 return 0;
3530
3531 /* If we need to pullup then pullup to the max, so we
3532 * won't need to do it again.
3533 */
3534 if (max > skb->len)
3535 max = skb->len;
3536
3537 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
3538 return -ENOMEM;
3539
3540 if (skb_headlen(skb) < len)
3541 return -EPROTO;
3542
3543 return 0;
3544}
3545
3546/* This value should be large enough to cover a tagged ethernet header plus
3547 * maximally sized IP and TCP or UDP headers.
3548 */
3549#define MAX_IP_HDR_LEN 128
3550
3551static int skb_checksum_setup_ip(struct sk_buff *skb, bool recalculate)
3552{
3553 unsigned int off;
3554 bool fragment;
3555 int err;
3556
3557 fragment = false;
3558
3559 err = skb_maybe_pull_tail(skb,
3560 sizeof(struct iphdr),
3561 MAX_IP_HDR_LEN);
3562 if (err < 0)
3563 goto out;
3564
3565 if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
3566 fragment = true;
3567
3568 off = ip_hdrlen(skb);
3569
3570 err = -EPROTO;
3571
3572 if (fragment)
3573 goto out;
3574
3575 switch (ip_hdr(skb)->protocol) {
3576 case IPPROTO_TCP:
3577 err = skb_maybe_pull_tail(skb,
3578 off + sizeof(struct tcphdr),
3579 MAX_IP_HDR_LEN);
3580 if (err < 0)
3581 goto out;
3582
3583 if (!skb_partial_csum_set(skb, off,
3584 offsetof(struct tcphdr, check))) {
3585 err = -EPROTO;
3586 goto out;
3587 }
3588
3589 if (recalculate)
3590 tcp_hdr(skb)->check =
3591 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3592 ip_hdr(skb)->daddr,
3593 skb->len - off,
3594 IPPROTO_TCP, 0);
3595 break;
3596 case IPPROTO_UDP:
3597 err = skb_maybe_pull_tail(skb,
3598 off + sizeof(struct udphdr),
3599 MAX_IP_HDR_LEN);
3600 if (err < 0)
3601 goto out;
3602
3603 if (!skb_partial_csum_set(skb, off,
3604 offsetof(struct udphdr, check))) {
3605 err = -EPROTO;
3606 goto out;
3607 }
3608
3609 if (recalculate)
3610 udp_hdr(skb)->check =
3611 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3612 ip_hdr(skb)->daddr,
3613 skb->len - off,
3614 IPPROTO_UDP, 0);
3615 break;
3616 default:
3617 goto out;
3618 }
3619
3620 err = 0;
3621
3622out:
3623 return err;
3624}
3625
3626/* This value should be large enough to cover a tagged ethernet header plus
3627 * an IPv6 header, all options, and a maximal TCP or UDP header.
3628 */
3629#define MAX_IPV6_HDR_LEN 256
3630
3631#define OPT_HDR(type, skb, off) \
3632 (type *)(skb_network_header(skb) + (off))
3633
3634static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
3635{
3636 int err;
3637 u8 nexthdr;
3638 unsigned int off;
3639 unsigned int len;
3640 bool fragment;
3641 bool done;
3642
3643 fragment = false;
3644 done = false;
3645
3646 off = sizeof(struct ipv6hdr);
3647
3648 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
3649 if (err < 0)
3650 goto out;
3651
3652 nexthdr = ipv6_hdr(skb)->nexthdr;
3653
3654 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
3655 while (off <= len && !done) {
3656 switch (nexthdr) {
3657 case IPPROTO_DSTOPTS:
3658 case IPPROTO_HOPOPTS:
3659 case IPPROTO_ROUTING: {
3660 struct ipv6_opt_hdr *hp;
3661
3662 err = skb_maybe_pull_tail(skb,
3663 off +
3664 sizeof(struct ipv6_opt_hdr),
3665 MAX_IPV6_HDR_LEN);
3666 if (err < 0)
3667 goto out;
3668
3669 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
3670 nexthdr = hp->nexthdr;
3671 off += ipv6_optlen(hp);
3672 break;
3673 }
3674 case IPPROTO_AH: {
3675 struct ip_auth_hdr *hp;
3676
3677 err = skb_maybe_pull_tail(skb,
3678 off +
3679 sizeof(struct ip_auth_hdr),
3680 MAX_IPV6_HDR_LEN);
3681 if (err < 0)
3682 goto out;
3683
3684 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
3685 nexthdr = hp->nexthdr;
3686 off += ipv6_authlen(hp);
3687 break;
3688 }
3689 case IPPROTO_FRAGMENT: {
3690 struct frag_hdr *hp;
3691
3692 err = skb_maybe_pull_tail(skb,
3693 off +
3694 sizeof(struct frag_hdr),
3695 MAX_IPV6_HDR_LEN);
3696 if (err < 0)
3697 goto out;
3698
3699 hp = OPT_HDR(struct frag_hdr, skb, off);
3700
3701 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
3702 fragment = true;
3703
3704 nexthdr = hp->nexthdr;
3705 off += sizeof(struct frag_hdr);
3706 break;
3707 }
3708 default:
3709 done = true;
3710 break;
3711 }
3712 }
3713
3714 err = -EPROTO;
3715
3716 if (!done || fragment)
3717 goto out;
3718
3719 switch (nexthdr) {
3720 case IPPROTO_TCP:
3721 err = skb_maybe_pull_tail(skb,
3722 off + sizeof(struct tcphdr),
3723 MAX_IPV6_HDR_LEN);
3724 if (err < 0)
3725 goto out;
3726
3727 if (!skb_partial_csum_set(skb, off,
3728 offsetof(struct tcphdr, check))) {
3729 err = -EPROTO;
3730 goto out;
3731 }
3732
3733 if (recalculate)
3734 tcp_hdr(skb)->check =
3735 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3736 &ipv6_hdr(skb)->daddr,
3737 skb->len - off,
3738 IPPROTO_TCP, 0);
3739 break;
3740 case IPPROTO_UDP:
3741 err = skb_maybe_pull_tail(skb,
3742 off + sizeof(struct udphdr),
3743 MAX_IPV6_HDR_LEN);
3744 if (err < 0)
3745 goto out;
3746
3747 if (!skb_partial_csum_set(skb, off,
3748 offsetof(struct udphdr, check))) {
3749 err = -EPROTO;
3750 goto out;
3751 }
3752
3753 if (recalculate)
3754 udp_hdr(skb)->check =
3755 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3756 &ipv6_hdr(skb)->daddr,
3757 skb->len - off,
3758 IPPROTO_UDP, 0);
3759 break;
3760 default:
3761 goto out;
3762 }
3763
3764 err = 0;
3765
3766out:
3767 return err;
3768}
3769
3770/**
3771 * skb_checksum_setup - set up partial checksum offset
3772 * @skb: the skb to set up
3773 * @recalculate: if true the pseudo-header checksum will be recalculated
3774 */
3775int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
3776{
3777 int err;
3778
3779 switch (skb->protocol) {
3780 case htons(ETH_P_IP):
3781 err = skb_checksum_setup_ip(skb, recalculate);
3782 break;
3783
3784 case htons(ETH_P_IPV6):
3785 err = skb_checksum_setup_ipv6(skb, recalculate);
3786 break;
3787
3788 default:
3789 err = -EPROTO;
3790 break;
3791 }
3792
3793 return err;
3794}
3795EXPORT_SYMBOL(skb_checksum_setup);
3796
Ben Hutchings4497b072008-06-19 16:22:28 -07003797void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3798{
Joe Perchese87cc472012-05-13 21:56:26 +00003799 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
3800 skb->dev->name);
Ben Hutchings4497b072008-06-19 16:22:28 -07003801}
Ben Hutchings4497b072008-06-19 16:22:28 -07003802EXPORT_SYMBOL(__skb_warn_lro_forwarding);
Eric Dumazetbad43ca2012-05-19 03:02:02 +00003803
3804void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
3805{
Eric Dumazet3d861f62012-10-22 09:03:40 +00003806 if (head_stolen) {
3807 skb_release_head_state(skb);
Eric Dumazetbad43ca2012-05-19 03:02:02 +00003808 kmem_cache_free(skbuff_head_cache, skb);
Eric Dumazet3d861f62012-10-22 09:03:40 +00003809 } else {
Eric Dumazetbad43ca2012-05-19 03:02:02 +00003810 __kfree_skb(skb);
Eric Dumazet3d861f62012-10-22 09:03:40 +00003811 }
Eric Dumazetbad43ca2012-05-19 03:02:02 +00003812}
3813EXPORT_SYMBOL(kfree_skb_partial);
3814
3815/**
3816 * skb_try_coalesce - try to merge skb to prior one
3817 * @to: prior buffer
3818 * @from: buffer to add
3819 * @fragstolen: pointer to boolean
Randy Dunlapc6c4b972012-06-08 14:01:44 +00003820 * @delta_truesize: how much more was allocated than was requested
Eric Dumazetbad43ca2012-05-19 03:02:02 +00003821 */
3822bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
3823 bool *fragstolen, int *delta_truesize)
3824{
3825 int i, delta, len = from->len;
3826
3827 *fragstolen = false;
3828
3829 if (skb_cloned(to))
3830 return false;
3831
3832 if (len <= skb_tailroom(to)) {
3833 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
3834 *delta_truesize = 0;
3835 return true;
3836 }
3837
3838 if (skb_has_frag_list(to) || skb_has_frag_list(from))
3839 return false;
3840
3841 if (skb_headlen(from) != 0) {
3842 struct page *page;
3843 unsigned int offset;
3844
3845 if (skb_shinfo(to)->nr_frags +
3846 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
3847 return false;
3848
3849 if (skb_head_is_locked(from))
3850 return false;
3851
3852 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3853
3854 page = virt_to_head_page(from->head);
3855 offset = from->data - (unsigned char *)page_address(page);
3856
3857 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
3858 page, offset, skb_headlen(from));
3859 *fragstolen = true;
3860 } else {
3861 if (skb_shinfo(to)->nr_frags +
3862 skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
3863 return false;
3864
Weiping Panf4b549a2012-09-28 20:15:30 +00003865 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
Eric Dumazetbad43ca2012-05-19 03:02:02 +00003866 }
3867
3868 WARN_ON_ONCE(delta < len);
3869
3870 memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
3871 skb_shinfo(from)->frags,
3872 skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
3873 skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
3874
3875 if (!skb_cloned(from))
3876 skb_shinfo(from)->nr_frags = 0;
3877
Li RongQing8ea853f2012-09-18 16:53:21 +00003878 /* if the skb is not cloned this does nothing
3879 * since we set nr_frags to 0.
3880 */
Eric Dumazetbad43ca2012-05-19 03:02:02 +00003881 for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
3882 skb_frag_ref(from, i);
3883
3884 to->truesize += delta;
3885 to->len += len;
3886 to->data_len += len;
3887
3888 *delta_truesize = delta;
3889 return true;
3890}
3891EXPORT_SYMBOL(skb_try_coalesce);
Nicolas Dichtel621e84d2013-06-26 16:11:27 +02003892
3893/**
Nicolas Dichtel8b27f272013-09-02 15:34:56 +02003894 * skb_scrub_packet - scrub an skb
Nicolas Dichtel621e84d2013-06-26 16:11:27 +02003895 *
3896 * @skb: buffer to clean
Nicolas Dichtel8b27f272013-09-02 15:34:56 +02003897 * @xnet: packet is crossing netns
Nicolas Dichtel621e84d2013-06-26 16:11:27 +02003898 *
Nicolas Dichtel8b27f272013-09-02 15:34:56 +02003899 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
3900 * into/from a tunnel. Some information have to be cleared during these
3901 * operations.
3902 * skb_scrub_packet can also be used to clean a skb before injecting it in
3903 * another namespace (@xnet == true). We have to clear all information in the
3904 * skb that could impact namespace isolation.
Nicolas Dichtel621e84d2013-06-26 16:11:27 +02003905 */
Nicolas Dichtel8b27f272013-09-02 15:34:56 +02003906void skb_scrub_packet(struct sk_buff *skb, bool xnet)
Nicolas Dichtel621e84d2013-06-26 16:11:27 +02003907{
Nicolas Dichtel8b27f272013-09-02 15:34:56 +02003908 if (xnet)
3909 skb_orphan(skb);
Nicolas Dichtel621e84d2013-06-26 16:11:27 +02003910 skb->tstamp.tv64 = 0;
3911 skb->pkt_type = PACKET_HOST;
3912 skb->skb_iif = 0;
Hannes Frederic Sowa239c78d2013-12-05 23:29:19 +01003913 skb->local_df = 0;
Nicolas Dichtel621e84d2013-06-26 16:11:27 +02003914 skb_dst_drop(skb);
3915 skb->mark = 0;
3916 secpath_reset(skb);
3917 nf_reset(skb);
3918 nf_reset_trace(skb);
3919}
3920EXPORT_SYMBOL_GPL(skb_scrub_packet);
Florian Westphalde960aa2014-01-26 10:58:16 +01003921
3922/**
3923 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
3924 *
3925 * @skb: GSO skb
3926 *
3927 * skb_gso_transport_seglen is used to determine the real size of the
3928 * individual segments, including Layer4 headers (TCP/UDP).
3929 *
3930 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
3931 */
3932unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
3933{
3934 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3935 unsigned int hdr_len;
3936
3937 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
3938 hdr_len = tcp_hdrlen(skb);
3939 else
3940 hdr_len = sizeof(struct udphdr);
3941 return hdr_len + shinfo->gso_size;
3942}
3943EXPORT_SYMBOL_GPL(skb_gso_transport_seglen);