Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Routines having to do with the 'struct sk_buff' memory handlers. |
| 3 | * |
Alan Cox | 113aa83 | 2008-10-13 19:01:08 -0700 | [diff] [blame] | 4 | * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5 | * Florian La Roche <rzsfl@rz.uni-sb.de> |
| 6 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7 | * Fixes: |
| 8 | * Alan Cox : Fixed the worst of the load |
| 9 | * balancer bugs. |
| 10 | * Dave Platt : Interrupt stacking fix. |
| 11 | * Richard Kooijman : Timestamp fixes. |
| 12 | * Alan Cox : Changed buffer format. |
| 13 | * Alan Cox : destructor hook for AF_UNIX etc. |
| 14 | * Linus Torvalds : Better skb_clone. |
| 15 | * Alan Cox : Added skb_copy. |
| 16 | * Alan Cox : Added all the changed routines Linus |
| 17 | * only put in the headers |
| 18 | * Ray VanTassle : Fixed --skb->lock in free |
| 19 | * Alan Cox : skb_copy copy arp field |
| 20 | * Andi Kleen : slabified it. |
| 21 | * Robert Olsson : Removed skb_head_pool |
| 22 | * |
| 23 | * NOTE: |
| 24 | * The __skb_ routines should be called with interrupts |
| 25 | * disabled, or you better be *real* sure that the operation is atomic |
| 26 | * with respect to whatever list is being frobbed (e.g. via lock_sock() |
| 27 | * or via disabling bottom half handlers, etc). |
| 28 | * |
| 29 | * This program is free software; you can redistribute it and/or |
| 30 | * modify it under the terms of the GNU General Public License |
| 31 | * as published by the Free Software Foundation; either version |
| 32 | * 2 of the License, or (at your option) any later version. |
| 33 | */ |
| 34 | |
| 35 | /* |
| 36 | * The functions in this file will not compile correctly with gcc 2.4.x |
| 37 | */ |
| 38 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 39 | #include <linux/module.h> |
| 40 | #include <linux/types.h> |
| 41 | #include <linux/kernel.h> |
Vegard Nossum | fe55f6d | 2008-08-30 12:16:35 +0200 | [diff] [blame] | 42 | #include <linux/kmemcheck.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 43 | #include <linux/mm.h> |
| 44 | #include <linux/interrupt.h> |
| 45 | #include <linux/in.h> |
| 46 | #include <linux/inet.h> |
| 47 | #include <linux/slab.h> |
| 48 | #include <linux/netdevice.h> |
| 49 | #ifdef CONFIG_NET_CLS_ACT |
| 50 | #include <net/pkt_sched.h> |
| 51 | #endif |
| 52 | #include <linux/string.h> |
| 53 | #include <linux/skbuff.h> |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 54 | #include <linux/splice.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 55 | #include <linux/cache.h> |
| 56 | #include <linux/rtnetlink.h> |
| 57 | #include <linux/init.h> |
David Howells | 716ea3a | 2007-04-02 20:19:53 -0700 | [diff] [blame] | 58 | #include <linux/scatterlist.h> |
Patrick Ohly | ac45f60 | 2009-02-12 05:03:37 +0000 | [diff] [blame] | 59 | #include <linux/errqueue.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 60 | |
| 61 | #include <net/protocol.h> |
| 62 | #include <net/dst.h> |
| 63 | #include <net/sock.h> |
| 64 | #include <net/checksum.h> |
| 65 | #include <net/xfrm.h> |
| 66 | |
| 67 | #include <asm/uaccess.h> |
| 68 | #include <asm/system.h> |
Steven Rostedt | ad8d75f | 2009-04-14 19:39:12 -0400 | [diff] [blame] | 69 | #include <trace/events/skb.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 70 | |
Al Viro | a1f8e7f7 | 2006-10-19 16:08:53 -0400 | [diff] [blame] | 71 | #include "kmap_skb.h" |
| 72 | |
Christoph Lameter | e18b890 | 2006-12-06 20:33:20 -0800 | [diff] [blame] | 73 | static struct kmem_cache *skbuff_head_cache __read_mostly; |
| 74 | static struct kmem_cache *skbuff_fclone_cache __read_mostly; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 75 | |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 76 | static void sock_pipe_buf_release(struct pipe_inode_info *pipe, |
| 77 | struct pipe_buffer *buf) |
| 78 | { |
Jarek Poplawski | 8b9d372 | 2009-01-19 17:03:56 -0800 | [diff] [blame] | 79 | put_page(buf->page); |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 80 | } |
| 81 | |
| 82 | static void sock_pipe_buf_get(struct pipe_inode_info *pipe, |
| 83 | struct pipe_buffer *buf) |
| 84 | { |
Jarek Poplawski | 8b9d372 | 2009-01-19 17:03:56 -0800 | [diff] [blame] | 85 | get_page(buf->page); |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 86 | } |
| 87 | |
| 88 | static int sock_pipe_buf_steal(struct pipe_inode_info *pipe, |
| 89 | struct pipe_buffer *buf) |
| 90 | { |
| 91 | return 1; |
| 92 | } |
| 93 | |
| 94 | |
| 95 | /* Pipe buffer operations for a socket. */ |
Alexey Dobriyan | 28dfef8 | 2009-12-15 16:46:48 -0800 | [diff] [blame] | 96 | static const struct pipe_buf_operations sock_pipe_buf_ops = { |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 97 | .can_merge = 0, |
| 98 | .map = generic_pipe_buf_map, |
| 99 | .unmap = generic_pipe_buf_unmap, |
| 100 | .confirm = generic_pipe_buf_confirm, |
| 101 | .release = sock_pipe_buf_release, |
| 102 | .steal = sock_pipe_buf_steal, |
| 103 | .get = sock_pipe_buf_get, |
| 104 | }; |
| 105 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 106 | /* |
| 107 | * Keep out-of-line to prevent kernel bloat. |
| 108 | * __builtin_return_address is not used because it is not always |
| 109 | * reliable. |
| 110 | */ |
| 111 | |
| 112 | /** |
| 113 | * skb_over_panic - private function |
| 114 | * @skb: buffer |
| 115 | * @sz: size |
| 116 | * @here: address |
| 117 | * |
| 118 | * Out of line support code for skb_put(). Not user callable. |
| 119 | */ |
Rami Rosen | ccb7c77 | 2010-04-20 22:39:53 -0700 | [diff] [blame] | 120 | static void skb_over_panic(struct sk_buff *skb, int sz, void *here) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 121 | { |
Patrick McHardy | 2609545 | 2005-04-21 16:43:02 -0700 | [diff] [blame] | 122 | printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p " |
Arnaldo Carvalho de Melo | 4305b54 | 2007-04-19 20:43:29 -0700 | [diff] [blame] | 123 | "data:%p tail:%#lx end:%#lx dev:%s\n", |
Arnaldo Carvalho de Melo | 27a884d | 2007-04-19 20:29:13 -0700 | [diff] [blame] | 124 | here, skb->len, sz, skb->head, skb->data, |
Arnaldo Carvalho de Melo | 4305b54 | 2007-04-19 20:43:29 -0700 | [diff] [blame] | 125 | (unsigned long)skb->tail, (unsigned long)skb->end, |
Patrick McHardy | 2609545 | 2005-04-21 16:43:02 -0700 | [diff] [blame] | 126 | skb->dev ? skb->dev->name : "<NULL>"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 127 | BUG(); |
| 128 | } |
| 129 | |
| 130 | /** |
| 131 | * skb_under_panic - private function |
| 132 | * @skb: buffer |
| 133 | * @sz: size |
| 134 | * @here: address |
| 135 | * |
| 136 | * Out of line support code for skb_push(). Not user callable. |
| 137 | */ |
| 138 | |
Rami Rosen | ccb7c77 | 2010-04-20 22:39:53 -0700 | [diff] [blame] | 139 | static void skb_under_panic(struct sk_buff *skb, int sz, void *here) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 140 | { |
Patrick McHardy | 2609545 | 2005-04-21 16:43:02 -0700 | [diff] [blame] | 141 | printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p " |
Arnaldo Carvalho de Melo | 4305b54 | 2007-04-19 20:43:29 -0700 | [diff] [blame] | 142 | "data:%p tail:%#lx end:%#lx dev:%s\n", |
Arnaldo Carvalho de Melo | 27a884d | 2007-04-19 20:29:13 -0700 | [diff] [blame] | 143 | here, skb->len, sz, skb->head, skb->data, |
Arnaldo Carvalho de Melo | 4305b54 | 2007-04-19 20:43:29 -0700 | [diff] [blame] | 144 | (unsigned long)skb->tail, (unsigned long)skb->end, |
Patrick McHardy | 2609545 | 2005-04-21 16:43:02 -0700 | [diff] [blame] | 145 | skb->dev ? skb->dev->name : "<NULL>"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 146 | BUG(); |
| 147 | } |
| 148 | |
| 149 | /* 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 | |
| 155 | /** |
David S. Miller | d179cd1 | 2005-08-17 14:57:30 -0700 | [diff] [blame] | 156 | * __alloc_skb - allocate a network buffer |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 157 | * @size: size to allocate |
| 158 | * @gfp_mask: allocation mask |
Randy Dunlap | c83c248 | 2005-10-18 22:07:41 -0700 | [diff] [blame] | 159 | * @fclone: allocate from fclone cache instead of head cache |
| 160 | * and allocate a cloned (child) skb |
Christoph Hellwig | b30973f | 2006-12-06 20:32:36 -0800 | [diff] [blame] | 161 | * @node: numa node to allocate memory on |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 162 | * |
| 163 | * Allocate a new &sk_buff. The returned buffer has no headroom and a |
| 164 | * tail room of size bytes. The object has a reference count of one. |
| 165 | * The return is the buffer. On a failure the return is %NULL. |
| 166 | * |
| 167 | * Buffers may only be allocated from interrupts using a @gfp_mask of |
| 168 | * %GFP_ATOMIC. |
| 169 | */ |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 170 | struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, |
Christoph Hellwig | b30973f | 2006-12-06 20:32:36 -0800 | [diff] [blame] | 171 | int fclone, int node) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 172 | { |
Christoph Lameter | e18b890 | 2006-12-06 20:33:20 -0800 | [diff] [blame] | 173 | struct kmem_cache *cache; |
Benjamin LaHaise | 4947d3e | 2006-01-03 14:06:50 -0800 | [diff] [blame] | 174 | struct skb_shared_info *shinfo; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 175 | struct sk_buff *skb; |
| 176 | u8 *data; |
| 177 | |
Herbert Xu | 8798b3f | 2006-01-23 16:32:45 -0800 | [diff] [blame] | 178 | cache = fclone ? skbuff_fclone_cache : skbuff_head_cache; |
| 179 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 180 | /* Get the HEAD */ |
Christoph Hellwig | b30973f | 2006-12-06 20:32:36 -0800 | [diff] [blame] | 181 | skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 182 | if (!skb) |
| 183 | goto out; |
Eric Dumazet | ec7d2f2 | 2010-05-05 01:07:37 -0700 | [diff] [blame] | 184 | prefetchw(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 185 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 186 | size = SKB_DATA_ALIGN(size); |
Christoph Hellwig | b30973f | 2006-12-06 20:32:36 -0800 | [diff] [blame] | 187 | data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info), |
| 188 | gfp_mask, node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 189 | if (!data) |
| 190 | goto nodata; |
Eric Dumazet | ec7d2f2 | 2010-05-05 01:07:37 -0700 | [diff] [blame] | 191 | prefetchw(data + size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 192 | |
Arnaldo Carvalho de Melo | ca0605a | 2007-03-19 10:48:59 -0300 | [diff] [blame] | 193 | /* |
Johannes Berg | c800578 | 2008-05-03 20:56:42 -0700 | [diff] [blame] | 194 | * Only clear those fields we need to clear, not those that we will |
| 195 | * actually initialise below. Hence, don't put any more fields after |
| 196 | * the tail pointer in struct sk_buff! |
Arnaldo Carvalho de Melo | ca0605a | 2007-03-19 10:48:59 -0300 | [diff] [blame] | 197 | */ |
| 198 | memset(skb, 0, offsetof(struct sk_buff, tail)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 199 | skb->truesize = size + sizeof(struct sk_buff); |
| 200 | atomic_set(&skb->users, 1); |
| 201 | skb->head = data; |
| 202 | skb->data = data; |
Arnaldo Carvalho de Melo | 27a884d | 2007-04-19 20:29:13 -0700 | [diff] [blame] | 203 | skb_reset_tail_pointer(skb); |
Arnaldo Carvalho de Melo | 4305b54 | 2007-04-19 20:43:29 -0700 | [diff] [blame] | 204 | skb->end = skb->tail + size; |
Stephen Hemminger | 19633e1 | 2009-06-17 05:23:27 +0000 | [diff] [blame] | 205 | #ifdef NET_SKBUFF_DATA_USES_OFFSET |
| 206 | skb->mac_header = ~0U; |
| 207 | #endif |
| 208 | |
Benjamin LaHaise | 4947d3e | 2006-01-03 14:06:50 -0800 | [diff] [blame] | 209 | /* make sure we initialize shinfo sequentially */ |
| 210 | shinfo = skb_shinfo(skb); |
Eric Dumazet | ec7d2f2 | 2010-05-05 01:07:37 -0700 | [diff] [blame] | 211 | memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); |
Benjamin LaHaise | 4947d3e | 2006-01-03 14:06:50 -0800 | [diff] [blame] | 212 | atomic_set(&shinfo->dataref, 1); |
Benjamin LaHaise | 4947d3e | 2006-01-03 14:06:50 -0800 | [diff] [blame] | 213 | |
David S. Miller | d179cd1 | 2005-08-17 14:57:30 -0700 | [diff] [blame] | 214 | if (fclone) { |
| 215 | struct sk_buff *child = skb + 1; |
| 216 | atomic_t *fclone_ref = (atomic_t *) (child + 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 217 | |
Vegard Nossum | fe55f6d | 2008-08-30 12:16:35 +0200 | [diff] [blame] | 218 | kmemcheck_annotate_bitfield(child, flags1); |
| 219 | kmemcheck_annotate_bitfield(child, flags2); |
David S. Miller | d179cd1 | 2005-08-17 14:57:30 -0700 | [diff] [blame] | 220 | skb->fclone = SKB_FCLONE_ORIG; |
| 221 | atomic_set(fclone_ref, 1); |
| 222 | |
| 223 | child->fclone = SKB_FCLONE_UNAVAILABLE; |
| 224 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 225 | out: |
| 226 | return skb; |
| 227 | nodata: |
Herbert Xu | 8798b3f | 2006-01-23 16:32:45 -0800 | [diff] [blame] | 228 | kmem_cache_free(cache, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 229 | skb = NULL; |
| 230 | goto out; |
| 231 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 232 | EXPORT_SYMBOL(__alloc_skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 233 | |
| 234 | /** |
Christoph Hellwig | 8af2745 | 2006-07-31 22:35:23 -0700 | [diff] [blame] | 235 | * __netdev_alloc_skb - allocate an skbuff for rx on a specific device |
| 236 | * @dev: network device to receive on |
| 237 | * @length: length to allocate |
| 238 | * @gfp_mask: get_free_pages mask, passed to alloc_skb |
| 239 | * |
| 240 | * Allocate a new &sk_buff and assign it a usage count of one. The |
| 241 | * buffer has unspecified headroom built in. Users should allocate |
| 242 | * the headroom they think they need without accounting for the |
| 243 | * built in space. The built in space is used for optimisations. |
| 244 | * |
| 245 | * %NULL is returned if there is no free memory. |
| 246 | */ |
| 247 | struct sk_buff *__netdev_alloc_skb(struct net_device *dev, |
| 248 | unsigned int length, gfp_t gfp_mask) |
| 249 | { |
| 250 | struct sk_buff *skb; |
| 251 | |
Eric Dumazet | 564824b | 2010-10-11 19:05:25 +0000 | [diff] [blame^] | 252 | skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, NUMA_NO_NODE); |
Christoph Hellwig | 7b2e497 | 2006-08-07 16:09:04 -0700 | [diff] [blame] | 253 | if (likely(skb)) { |
Christoph Hellwig | 8af2745 | 2006-07-31 22:35:23 -0700 | [diff] [blame] | 254 | skb_reserve(skb, NET_SKB_PAD); |
Christoph Hellwig | 7b2e497 | 2006-08-07 16:09:04 -0700 | [diff] [blame] | 255 | skb->dev = dev; |
| 256 | } |
Christoph Hellwig | 8af2745 | 2006-07-31 22:35:23 -0700 | [diff] [blame] | 257 | return skb; |
| 258 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 259 | EXPORT_SYMBOL(__netdev_alloc_skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 260 | |
Peter Zijlstra | 654bed1 | 2008-10-07 14:22:33 -0700 | [diff] [blame] | 261 | void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, |
| 262 | int size) |
| 263 | { |
| 264 | skb_fill_page_desc(skb, i, page, off, size); |
| 265 | skb->len += size; |
| 266 | skb->data_len += size; |
| 267 | skb->truesize += size; |
| 268 | } |
| 269 | EXPORT_SYMBOL(skb_add_rx_frag); |
| 270 | |
Ilpo Järvinen | f58518e | 2008-03-27 17:51:31 -0700 | [diff] [blame] | 271 | /** |
| 272 | * dev_alloc_skb - allocate an skbuff for receiving |
| 273 | * @length: length to allocate |
| 274 | * |
| 275 | * Allocate a new &sk_buff and assign it a usage count of one. The |
| 276 | * buffer has unspecified headroom built in. Users should allocate |
| 277 | * the headroom they think they need without accounting for the |
| 278 | * built in space. The built in space is used for optimisations. |
| 279 | * |
| 280 | * %NULL is returned if there is no free memory. Although this function |
| 281 | * allocates memory it can be called from an interrupt. |
| 282 | */ |
| 283 | struct sk_buff *dev_alloc_skb(unsigned int length) |
| 284 | { |
Denys Vlasenko | 1483b87 | 2008-03-28 15:57:39 -0700 | [diff] [blame] | 285 | /* |
| 286 | * There is more code here than it seems: |
David S. Miller | a0f55e0 | 2008-03-28 18:22:32 -0700 | [diff] [blame] | 287 | * __dev_alloc_skb is an inline |
Denys Vlasenko | 1483b87 | 2008-03-28 15:57:39 -0700 | [diff] [blame] | 288 | */ |
Ilpo Järvinen | f58518e | 2008-03-27 17:51:31 -0700 | [diff] [blame] | 289 | return __dev_alloc_skb(length, GFP_ATOMIC); |
| 290 | } |
| 291 | EXPORT_SYMBOL(dev_alloc_skb); |
| 292 | |
Herbert Xu | 27b437c | 2006-07-13 19:26:39 -0700 | [diff] [blame] | 293 | static void skb_drop_list(struct sk_buff **listp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 294 | { |
Herbert Xu | 27b437c | 2006-07-13 19:26:39 -0700 | [diff] [blame] | 295 | struct sk_buff *list = *listp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 296 | |
Herbert Xu | 27b437c | 2006-07-13 19:26:39 -0700 | [diff] [blame] | 297 | *listp = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 298 | |
| 299 | do { |
| 300 | struct sk_buff *this = list; |
| 301 | list = list->next; |
| 302 | kfree_skb(this); |
| 303 | } while (list); |
| 304 | } |
| 305 | |
Herbert Xu | 27b437c | 2006-07-13 19:26:39 -0700 | [diff] [blame] | 306 | static inline void skb_drop_fraglist(struct sk_buff *skb) |
| 307 | { |
| 308 | skb_drop_list(&skb_shinfo(skb)->frag_list); |
| 309 | } |
| 310 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 311 | static void skb_clone_fraglist(struct sk_buff *skb) |
| 312 | { |
| 313 | struct sk_buff *list; |
| 314 | |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 315 | skb_walk_frags(skb, list) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 316 | skb_get(list); |
| 317 | } |
| 318 | |
Adrian Bunk | 5bba171 | 2006-06-29 13:02:35 -0700 | [diff] [blame] | 319 | static void skb_release_data(struct sk_buff *skb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 320 | { |
| 321 | if (!skb->cloned || |
| 322 | !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, |
| 323 | &skb_shinfo(skb)->dataref)) { |
| 324 | if (skb_shinfo(skb)->nr_frags) { |
| 325 | int i; |
| 326 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) |
| 327 | put_page(skb_shinfo(skb)->frags[i].page); |
| 328 | } |
| 329 | |
David S. Miller | 21dc330 | 2010-08-23 00:13:46 -0700 | [diff] [blame] | 330 | if (skb_has_frag_list(skb)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 331 | skb_drop_fraglist(skb); |
| 332 | |
| 333 | kfree(skb->head); |
| 334 | } |
| 335 | } |
| 336 | |
| 337 | /* |
| 338 | * Free an skbuff by memory without cleaning the state. |
| 339 | */ |
Herbert Xu | 2d4baff | 2007-11-26 23:11:19 +0800 | [diff] [blame] | 340 | static void kfree_skbmem(struct sk_buff *skb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 341 | { |
David S. Miller | d179cd1 | 2005-08-17 14:57:30 -0700 | [diff] [blame] | 342 | struct sk_buff *other; |
| 343 | atomic_t *fclone_ref; |
| 344 | |
David S. Miller | d179cd1 | 2005-08-17 14:57:30 -0700 | [diff] [blame] | 345 | switch (skb->fclone) { |
| 346 | case SKB_FCLONE_UNAVAILABLE: |
| 347 | kmem_cache_free(skbuff_head_cache, skb); |
| 348 | break; |
| 349 | |
| 350 | case SKB_FCLONE_ORIG: |
| 351 | fclone_ref = (atomic_t *) (skb + 2); |
| 352 | if (atomic_dec_and_test(fclone_ref)) |
| 353 | kmem_cache_free(skbuff_fclone_cache, skb); |
| 354 | break; |
| 355 | |
| 356 | case SKB_FCLONE_CLONE: |
| 357 | fclone_ref = (atomic_t *) (skb + 1); |
| 358 | other = skb - 1; |
| 359 | |
| 360 | /* The clone portion is available for |
| 361 | * fast-cloning again. |
| 362 | */ |
| 363 | skb->fclone = SKB_FCLONE_UNAVAILABLE; |
| 364 | |
| 365 | if (atomic_dec_and_test(fclone_ref)) |
| 366 | kmem_cache_free(skbuff_fclone_cache, other); |
| 367 | break; |
Stephen Hemminger | 3ff50b7 | 2007-04-20 17:09:22 -0700 | [diff] [blame] | 368 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 369 | } |
| 370 | |
Lennert Buytenhek | 04a4bb5 | 2008-10-01 02:33:12 -0700 | [diff] [blame] | 371 | static void skb_release_head_state(struct sk_buff *skb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 372 | { |
Eric Dumazet | adf3090 | 2009-06-02 05:19:30 +0000 | [diff] [blame] | 373 | skb_dst_drop(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 374 | #ifdef CONFIG_XFRM |
| 375 | secpath_put(skb->sp); |
| 376 | #endif |
Stephen Hemminger | 9c2b332 | 2005-04-19 22:39:42 -0700 | [diff] [blame] | 377 | if (skb->destructor) { |
| 378 | WARN_ON(in_irq()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 379 | skb->destructor(skb); |
| 380 | } |
Yasuyuki Kozakai | 9fb9cbb | 2005-11-09 16:38:16 -0800 | [diff] [blame] | 381 | #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) |
Yasuyuki Kozakai | 5f79e0f | 2007-03-23 11:17:07 -0700 | [diff] [blame] | 382 | nf_conntrack_put(skb->nfct); |
Yasuyuki Kozakai | 9fb9cbb | 2005-11-09 16:38:16 -0800 | [diff] [blame] | 383 | nf_conntrack_put_reasm(skb->nfct_reasm); |
| 384 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 385 | #ifdef CONFIG_BRIDGE_NETFILTER |
| 386 | nf_bridge_put(skb->nf_bridge); |
| 387 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 388 | /* XXX: IS this still necessary? - JHS */ |
| 389 | #ifdef CONFIG_NET_SCHED |
| 390 | skb->tc_index = 0; |
| 391 | #ifdef CONFIG_NET_CLS_ACT |
| 392 | skb->tc_verd = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 393 | #endif |
| 394 | #endif |
Lennert Buytenhek | 04a4bb5 | 2008-10-01 02:33:12 -0700 | [diff] [blame] | 395 | } |
| 396 | |
| 397 | /* Free everything but the sk_buff shell. */ |
| 398 | static void skb_release_all(struct sk_buff *skb) |
| 399 | { |
| 400 | skb_release_head_state(skb); |
Herbert Xu | 2d4baff | 2007-11-26 23:11:19 +0800 | [diff] [blame] | 401 | skb_release_data(skb); |
| 402 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 403 | |
Herbert Xu | 2d4baff | 2007-11-26 23:11:19 +0800 | [diff] [blame] | 404 | /** |
| 405 | * __kfree_skb - private function |
| 406 | * @skb: buffer |
| 407 | * |
| 408 | * Free an sk_buff. Release anything attached to the buffer. |
| 409 | * Clean the state. This is an internal helper function. Users should |
| 410 | * always call kfree_skb |
| 411 | */ |
| 412 | |
| 413 | void __kfree_skb(struct sk_buff *skb) |
| 414 | { |
| 415 | skb_release_all(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 416 | kfree_skbmem(skb); |
| 417 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 418 | EXPORT_SYMBOL(__kfree_skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 419 | |
| 420 | /** |
Jörn Engel | 231d06a | 2006-03-20 21:28:35 -0800 | [diff] [blame] | 421 | * kfree_skb - free an sk_buff |
| 422 | * @skb: buffer to free |
| 423 | * |
| 424 | * Drop a reference to the buffer and free it if the usage count has |
| 425 | * hit zero. |
| 426 | */ |
| 427 | void kfree_skb(struct sk_buff *skb) |
| 428 | { |
| 429 | if (unlikely(!skb)) |
| 430 | return; |
| 431 | if (likely(atomic_read(&skb->users) == 1)) |
| 432 | smp_rmb(); |
| 433 | else if (likely(!atomic_dec_and_test(&skb->users))) |
| 434 | return; |
Neil Horman | ead2ceb | 2009-03-11 09:49:55 +0000 | [diff] [blame] | 435 | trace_kfree_skb(skb, __builtin_return_address(0)); |
Jörn Engel | 231d06a | 2006-03-20 21:28:35 -0800 | [diff] [blame] | 436 | __kfree_skb(skb); |
| 437 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 438 | EXPORT_SYMBOL(kfree_skb); |
Jörn Engel | 231d06a | 2006-03-20 21:28:35 -0800 | [diff] [blame] | 439 | |
Stephen Hemminger | d1a203e | 2008-11-01 21:01:09 -0700 | [diff] [blame] | 440 | /** |
Neil Horman | ead2ceb | 2009-03-11 09:49:55 +0000 | [diff] [blame] | 441 | * consume_skb - free an skbuff |
| 442 | * @skb: buffer to free |
| 443 | * |
| 444 | * Drop a ref to the buffer and free it if the usage count has hit zero |
| 445 | * Functions identically to kfree_skb, but kfree_skb assumes that the frame |
| 446 | * is being dropped after a failure and notes that |
| 447 | */ |
| 448 | void consume_skb(struct sk_buff *skb) |
| 449 | { |
| 450 | if (unlikely(!skb)) |
| 451 | return; |
| 452 | if (likely(atomic_read(&skb->users) == 1)) |
| 453 | smp_rmb(); |
| 454 | else if (likely(!atomic_dec_and_test(&skb->users))) |
| 455 | return; |
| 456 | __kfree_skb(skb); |
| 457 | } |
| 458 | EXPORT_SYMBOL(consume_skb); |
| 459 | |
| 460 | /** |
Stephen Hemminger | d1a203e | 2008-11-01 21:01:09 -0700 | [diff] [blame] | 461 | * skb_recycle_check - check if skb can be reused for receive |
| 462 | * @skb: buffer |
| 463 | * @skb_size: minimum receive buffer size |
| 464 | * |
| 465 | * Checks that the skb passed in is not shared or cloned, and |
| 466 | * that it is linear and its head portion at least as large as |
| 467 | * skb_size so that it can be recycled as a receive buffer. |
| 468 | * If these conditions are met, this function does any necessary |
| 469 | * reference count dropping and cleans up the skbuff as if it |
| 470 | * just came from __alloc_skb(). |
| 471 | */ |
Changli Gao | 5b0daa3 | 2010-05-29 00:12:13 -0700 | [diff] [blame] | 472 | bool skb_recycle_check(struct sk_buff *skb, int skb_size) |
Lennert Buytenhek | 04a4bb5 | 2008-10-01 02:33:12 -0700 | [diff] [blame] | 473 | { |
| 474 | struct skb_shared_info *shinfo; |
| 475 | |
Anton Vorontsov | e84af6d | 2009-11-10 14:11:01 +0000 | [diff] [blame] | 476 | if (irqs_disabled()) |
Changli Gao | 5b0daa3 | 2010-05-29 00:12:13 -0700 | [diff] [blame] | 477 | return false; |
Anton Vorontsov | e84af6d | 2009-11-10 14:11:01 +0000 | [diff] [blame] | 478 | |
Lennert Buytenhek | 04a4bb5 | 2008-10-01 02:33:12 -0700 | [diff] [blame] | 479 | if (skb_is_nonlinear(skb) || skb->fclone != SKB_FCLONE_UNAVAILABLE) |
Changli Gao | 5b0daa3 | 2010-05-29 00:12:13 -0700 | [diff] [blame] | 480 | return false; |
Lennert Buytenhek | 04a4bb5 | 2008-10-01 02:33:12 -0700 | [diff] [blame] | 481 | |
| 482 | skb_size = SKB_DATA_ALIGN(skb_size + NET_SKB_PAD); |
| 483 | if (skb_end_pointer(skb) - skb->head < skb_size) |
Changli Gao | 5b0daa3 | 2010-05-29 00:12:13 -0700 | [diff] [blame] | 484 | return false; |
Lennert Buytenhek | 04a4bb5 | 2008-10-01 02:33:12 -0700 | [diff] [blame] | 485 | |
| 486 | if (skb_shared(skb) || skb_cloned(skb)) |
Changli Gao | 5b0daa3 | 2010-05-29 00:12:13 -0700 | [diff] [blame] | 487 | return false; |
Lennert Buytenhek | 04a4bb5 | 2008-10-01 02:33:12 -0700 | [diff] [blame] | 488 | |
| 489 | skb_release_head_state(skb); |
Eric Dumazet | ec7d2f2 | 2010-05-05 01:07:37 -0700 | [diff] [blame] | 490 | |
Lennert Buytenhek | 04a4bb5 | 2008-10-01 02:33:12 -0700 | [diff] [blame] | 491 | shinfo = skb_shinfo(skb); |
Eric Dumazet | ec7d2f2 | 2010-05-05 01:07:37 -0700 | [diff] [blame] | 492 | memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); |
Lennert Buytenhek | 04a4bb5 | 2008-10-01 02:33:12 -0700 | [diff] [blame] | 493 | atomic_set(&shinfo->dataref, 1); |
Lennert Buytenhek | 04a4bb5 | 2008-10-01 02:33:12 -0700 | [diff] [blame] | 494 | |
| 495 | memset(skb, 0, offsetof(struct sk_buff, tail)); |
Lennert Buytenhek | 04a4bb5 | 2008-10-01 02:33:12 -0700 | [diff] [blame] | 496 | skb->data = skb->head + NET_SKB_PAD; |
Lennert Buytenhek | 5cd33db | 2008-11-10 21:45:05 -0800 | [diff] [blame] | 497 | skb_reset_tail_pointer(skb); |
Lennert Buytenhek | 04a4bb5 | 2008-10-01 02:33:12 -0700 | [diff] [blame] | 498 | |
Changli Gao | 5b0daa3 | 2010-05-29 00:12:13 -0700 | [diff] [blame] | 499 | return true; |
Lennert Buytenhek | 04a4bb5 | 2008-10-01 02:33:12 -0700 | [diff] [blame] | 500 | } |
| 501 | EXPORT_SYMBOL(skb_recycle_check); |
| 502 | |
Herbert Xu | dec1881 | 2007-10-14 00:37:30 -0700 | [diff] [blame] | 503 | static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old) |
| 504 | { |
| 505 | new->tstamp = old->tstamp; |
| 506 | new->dev = old->dev; |
| 507 | new->transport_header = old->transport_header; |
| 508 | new->network_header = old->network_header; |
| 509 | new->mac_header = old->mac_header; |
Eric Dumazet | 7fee226 | 2010-05-11 23:19:48 +0000 | [diff] [blame] | 510 | skb_dst_copy(new, old); |
Tom Herbert | 0a9627f | 2010-03-16 08:03:29 +0000 | [diff] [blame] | 511 | new->rxhash = old->rxhash; |
Alexey Dobriyan | def8b4f | 2008-10-28 13:24:06 -0700 | [diff] [blame] | 512 | #ifdef CONFIG_XFRM |
Herbert Xu | dec1881 | 2007-10-14 00:37:30 -0700 | [diff] [blame] | 513 | new->sp = secpath_get(old->sp); |
| 514 | #endif |
| 515 | memcpy(new->cb, old->cb, sizeof(old->cb)); |
Herbert Xu | 9bcb97c | 2009-05-22 22:20:02 +0000 | [diff] [blame] | 516 | new->csum = old->csum; |
Herbert Xu | dec1881 | 2007-10-14 00:37:30 -0700 | [diff] [blame] | 517 | new->local_df = old->local_df; |
| 518 | new->pkt_type = old->pkt_type; |
| 519 | new->ip_summed = old->ip_summed; |
| 520 | skb_copy_queue_mapping(new, old); |
| 521 | new->priority = old->priority; |
John Fastabend | e897082 | 2010-06-13 10:36:30 +0000 | [diff] [blame] | 522 | new->deliver_no_wcard = old->deliver_no_wcard; |
Herbert Xu | dec1881 | 2007-10-14 00:37:30 -0700 | [diff] [blame] | 523 | #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE) |
| 524 | new->ipvs_property = old->ipvs_property; |
| 525 | #endif |
| 526 | new->protocol = old->protocol; |
| 527 | new->mark = old->mark; |
Eric Dumazet | 8964be4 | 2009-11-20 15:35:04 -0800 | [diff] [blame] | 528 | new->skb_iif = old->skb_iif; |
Herbert Xu | dec1881 | 2007-10-14 00:37:30 -0700 | [diff] [blame] | 529 | __nf_copy(new, old); |
| 530 | #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \ |
| 531 | defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE) |
| 532 | new->nf_trace = old->nf_trace; |
| 533 | #endif |
| 534 | #ifdef CONFIG_NET_SCHED |
| 535 | new->tc_index = old->tc_index; |
| 536 | #ifdef CONFIG_NET_CLS_ACT |
| 537 | new->tc_verd = old->tc_verd; |
| 538 | #endif |
| 539 | #endif |
Patrick McHardy | 6aa895b0 | 2008-07-14 22:49:06 -0700 | [diff] [blame] | 540 | new->vlan_tci = old->vlan_tci; |
| 541 | |
Herbert Xu | dec1881 | 2007-10-14 00:37:30 -0700 | [diff] [blame] | 542 | skb_copy_secmark(new, old); |
| 543 | } |
| 544 | |
Herbert Xu | 82c49a3 | 2009-05-22 22:11:37 +0000 | [diff] [blame] | 545 | /* |
| 546 | * You should not add any new code to this function. Add it to |
| 547 | * __copy_skb_header above instead. |
| 548 | */ |
Herbert Xu | e0053ec | 2007-10-14 00:37:52 -0700 | [diff] [blame] | 549 | static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 550 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 551 | #define C(x) n->x = skb->x |
| 552 | |
| 553 | n->next = n->prev = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 554 | n->sk = NULL; |
Herbert Xu | dec1881 | 2007-10-14 00:37:30 -0700 | [diff] [blame] | 555 | __copy_skb_header(n, skb); |
| 556 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 557 | C(len); |
| 558 | C(data_len); |
Alexey Dobriyan | 3e6b3b2 | 2007-03-16 15:00:46 -0700 | [diff] [blame] | 559 | C(mac_len); |
Patrick McHardy | 334a813 | 2007-06-25 04:35:20 -0700 | [diff] [blame] | 560 | n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len; |
Paul Moore | 02f1c89 | 2008-01-07 21:56:41 -0800 | [diff] [blame] | 561 | n->cloned = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 562 | n->nohdr = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 563 | n->destructor = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 564 | C(tail); |
| 565 | C(end); |
Paul Moore | 02f1c89 | 2008-01-07 21:56:41 -0800 | [diff] [blame] | 566 | C(head); |
| 567 | C(data); |
| 568 | C(truesize); |
| 569 | atomic_set(&n->users, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 570 | |
| 571 | atomic_inc(&(skb_shinfo(skb)->dataref)); |
| 572 | skb->cloned = 1; |
| 573 | |
| 574 | return n; |
Herbert Xu | e0053ec | 2007-10-14 00:37:52 -0700 | [diff] [blame] | 575 | #undef C |
| 576 | } |
| 577 | |
| 578 | /** |
| 579 | * skb_morph - morph one skb into another |
| 580 | * @dst: the skb to receive the contents |
| 581 | * @src: the skb to supply the contents |
| 582 | * |
| 583 | * This is identical to skb_clone except that the target skb is |
| 584 | * supplied by the user. |
| 585 | * |
| 586 | * The target skb is returned upon exit. |
| 587 | */ |
| 588 | struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src) |
| 589 | { |
Herbert Xu | 2d4baff | 2007-11-26 23:11:19 +0800 | [diff] [blame] | 590 | skb_release_all(dst); |
Herbert Xu | e0053ec | 2007-10-14 00:37:52 -0700 | [diff] [blame] | 591 | return __skb_clone(dst, src); |
| 592 | } |
| 593 | EXPORT_SYMBOL_GPL(skb_morph); |
| 594 | |
| 595 | /** |
| 596 | * skb_clone - duplicate an sk_buff |
| 597 | * @skb: buffer to clone |
| 598 | * @gfp_mask: allocation priority |
| 599 | * |
| 600 | * Duplicate an &sk_buff. The new one is not owned by a socket. Both |
| 601 | * copies share the same packet data but not structure. The new |
| 602 | * buffer has a reference count of 1. If the allocation fails the |
| 603 | * function returns %NULL otherwise the new buffer is returned. |
| 604 | * |
| 605 | * If this function is called from an interrupt gfp_mask() must be |
| 606 | * %GFP_ATOMIC. |
| 607 | */ |
| 608 | |
| 609 | struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask) |
| 610 | { |
| 611 | struct sk_buff *n; |
| 612 | |
| 613 | n = skb + 1; |
| 614 | if (skb->fclone == SKB_FCLONE_ORIG && |
| 615 | n->fclone == SKB_FCLONE_UNAVAILABLE) { |
| 616 | atomic_t *fclone_ref = (atomic_t *) (n + 1); |
| 617 | n->fclone = SKB_FCLONE_CLONE; |
| 618 | atomic_inc(fclone_ref); |
| 619 | } else { |
| 620 | n = kmem_cache_alloc(skbuff_head_cache, gfp_mask); |
| 621 | if (!n) |
| 622 | return NULL; |
Vegard Nossum | fe55f6d | 2008-08-30 12:16:35 +0200 | [diff] [blame] | 623 | |
| 624 | kmemcheck_annotate_bitfield(n, flags1); |
| 625 | kmemcheck_annotate_bitfield(n, flags2); |
Herbert Xu | e0053ec | 2007-10-14 00:37:52 -0700 | [diff] [blame] | 626 | n->fclone = SKB_FCLONE_UNAVAILABLE; |
| 627 | } |
| 628 | |
| 629 | return __skb_clone(n, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 630 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 631 | EXPORT_SYMBOL(skb_clone); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 632 | |
| 633 | static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old) |
| 634 | { |
Arnaldo Carvalho de Melo | 2e07fa9 | 2007-04-10 21:22:35 -0700 | [diff] [blame] | 635 | #ifndef NET_SKBUFF_DATA_USES_OFFSET |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 636 | /* |
| 637 | * Shift between the two data areas in bytes |
| 638 | */ |
| 639 | unsigned long offset = new->data - old->data; |
Arnaldo Carvalho de Melo | 2e07fa9 | 2007-04-10 21:22:35 -0700 | [diff] [blame] | 640 | #endif |
Herbert Xu | dec1881 | 2007-10-14 00:37:30 -0700 | [diff] [blame] | 641 | |
| 642 | __copy_skb_header(new, old); |
| 643 | |
Arnaldo Carvalho de Melo | 2e07fa9 | 2007-04-10 21:22:35 -0700 | [diff] [blame] | 644 | #ifndef NET_SKBUFF_DATA_USES_OFFSET |
| 645 | /* {transport,network,mac}_header are relative to skb->head */ |
| 646 | new->transport_header += offset; |
| 647 | new->network_header += offset; |
Stephen Hemminger | 603a8bb | 2009-06-17 12:17:34 +0000 | [diff] [blame] | 648 | if (skb_mac_header_was_set(new)) |
| 649 | new->mac_header += offset; |
Arnaldo Carvalho de Melo | 2e07fa9 | 2007-04-10 21:22:35 -0700 | [diff] [blame] | 650 | #endif |
Herbert Xu | 7967168 | 2006-06-22 02:40:14 -0700 | [diff] [blame] | 651 | skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size; |
| 652 | skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs; |
| 653 | skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 654 | } |
| 655 | |
| 656 | /** |
| 657 | * skb_copy - create private copy of an sk_buff |
| 658 | * @skb: buffer to copy |
| 659 | * @gfp_mask: allocation priority |
| 660 | * |
| 661 | * Make a copy of both an &sk_buff and its data. This is used when the |
| 662 | * caller wishes to modify the data and needs a private copy of the |
| 663 | * data to alter. Returns %NULL on failure or the pointer to the buffer |
| 664 | * on success. The returned buffer has a reference count of 1. |
| 665 | * |
| 666 | * As by-product this function converts non-linear &sk_buff to linear |
| 667 | * one, so that &sk_buff becomes completely private and caller is allowed |
| 668 | * to modify all the data of returned buffer. This means that this |
| 669 | * function is not recommended for use in circumstances when only |
| 670 | * header is going to be modified. Use pskb_copy() instead. |
| 671 | */ |
| 672 | |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 673 | struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 674 | { |
Eric Dumazet | 6602ceb | 2010-09-01 05:25:10 +0000 | [diff] [blame] | 675 | int headerlen = skb_headroom(skb); |
| 676 | unsigned int size = (skb_end_pointer(skb) - skb->head) + skb->data_len; |
| 677 | struct sk_buff *n = alloc_skb(size, gfp_mask); |
| 678 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 679 | if (!n) |
| 680 | return NULL; |
| 681 | |
| 682 | /* Set the data pointer */ |
| 683 | skb_reserve(n, headerlen); |
| 684 | /* Set the tail pointer and length */ |
| 685 | skb_put(n, skb->len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 686 | |
| 687 | if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len)) |
| 688 | BUG(); |
| 689 | |
| 690 | copy_skb_header(n, skb); |
| 691 | return n; |
| 692 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 693 | EXPORT_SYMBOL(skb_copy); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 694 | |
| 695 | /** |
| 696 | * pskb_copy - create copy of an sk_buff with private head. |
| 697 | * @skb: buffer to copy |
| 698 | * @gfp_mask: allocation priority |
| 699 | * |
| 700 | * Make a copy of both an &sk_buff and part of its data, located |
| 701 | * in header. Fragmented data remain shared. This is used when |
| 702 | * the caller wishes to modify only header of &sk_buff and needs |
| 703 | * private copy of the header to alter. Returns %NULL on failure |
| 704 | * or the pointer to the buffer on success. |
| 705 | * The returned buffer has a reference count of 1. |
| 706 | */ |
| 707 | |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 708 | struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 709 | { |
Eric Dumazet | 6602ceb | 2010-09-01 05:25:10 +0000 | [diff] [blame] | 710 | unsigned int size = skb_end_pointer(skb) - skb->head; |
| 711 | struct sk_buff *n = alloc_skb(size, gfp_mask); |
| 712 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 713 | if (!n) |
| 714 | goto out; |
| 715 | |
| 716 | /* Set the data pointer */ |
Eric Dumazet | 6602ceb | 2010-09-01 05:25:10 +0000 | [diff] [blame] | 717 | skb_reserve(n, skb_headroom(skb)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 718 | /* Set the tail pointer and length */ |
| 719 | skb_put(n, skb_headlen(skb)); |
| 720 | /* Copy the bytes */ |
Arnaldo Carvalho de Melo | d626f62 | 2007-03-27 18:55:52 -0300 | [diff] [blame] | 721 | skb_copy_from_linear_data(skb, n->data, n->len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 722 | |
Herbert Xu | 25f484a | 2006-11-07 14:57:15 -0800 | [diff] [blame] | 723 | n->truesize += skb->data_len; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 724 | n->data_len = skb->data_len; |
| 725 | n->len = skb->len; |
| 726 | |
| 727 | if (skb_shinfo(skb)->nr_frags) { |
| 728 | int i; |
| 729 | |
| 730 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
| 731 | skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i]; |
| 732 | get_page(skb_shinfo(n)->frags[i].page); |
| 733 | } |
| 734 | skb_shinfo(n)->nr_frags = i; |
| 735 | } |
| 736 | |
David S. Miller | 21dc330 | 2010-08-23 00:13:46 -0700 | [diff] [blame] | 737 | if (skb_has_frag_list(skb)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 738 | skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list; |
| 739 | skb_clone_fraglist(n); |
| 740 | } |
| 741 | |
| 742 | copy_skb_header(n, skb); |
| 743 | out: |
| 744 | return n; |
| 745 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 746 | EXPORT_SYMBOL(pskb_copy); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 747 | |
| 748 | /** |
| 749 | * pskb_expand_head - reallocate header of &sk_buff |
| 750 | * @skb: buffer to reallocate |
| 751 | * @nhead: room to add at head |
| 752 | * @ntail: room to add at tail |
| 753 | * @gfp_mask: allocation priority |
| 754 | * |
| 755 | * Expands (or creates identical copy, if &nhead and &ntail are zero) |
| 756 | * header of skb. &sk_buff itself is not changed. &sk_buff MUST have |
| 757 | * reference count of 1. Returns zero in the case of success or error, |
| 758 | * if expansion failed. In the last case, &sk_buff is not changed. |
| 759 | * |
| 760 | * All the pointers pointing into skb header may change and must be |
| 761 | * reloaded after call to this function. |
| 762 | */ |
| 763 | |
Victor Fusco | 86a76ca | 2005-07-08 14:57:47 -0700 | [diff] [blame] | 764 | int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 765 | gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 766 | { |
| 767 | int i; |
| 768 | u8 *data; |
Eric Dumazet | 6602ceb | 2010-09-01 05:25:10 +0000 | [diff] [blame] | 769 | int size = nhead + (skb_end_pointer(skb) - skb->head) + ntail; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 770 | long off; |
Eric Dumazet | 1fd6304 | 2010-09-02 23:09:32 +0000 | [diff] [blame] | 771 | bool fastpath; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 772 | |
Herbert Xu | 4edd87a | 2008-10-01 07:09:38 -0700 | [diff] [blame] | 773 | BUG_ON(nhead < 0); |
| 774 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 775 | if (skb_shared(skb)) |
| 776 | BUG(); |
| 777 | |
| 778 | size = SKB_DATA_ALIGN(size); |
| 779 | |
| 780 | data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask); |
| 781 | if (!data) |
| 782 | goto nodata; |
| 783 | |
| 784 | /* Copy only real data... and, alas, header. This should be |
Eric Dumazet | 6602ceb | 2010-09-01 05:25:10 +0000 | [diff] [blame] | 785 | * optimized for the cases when header is void. |
| 786 | */ |
| 787 | memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head); |
| 788 | |
| 789 | memcpy((struct skb_shared_info *)(data + size), |
| 790 | skb_shinfo(skb), |
Eric Dumazet | fed6638 | 2010-07-22 19:09:08 +0000 | [diff] [blame] | 791 | offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags])); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 792 | |
Eric Dumazet | 1fd6304 | 2010-09-02 23:09:32 +0000 | [diff] [blame] | 793 | /* Check if we can avoid taking references on fragments if we own |
| 794 | * the last reference on skb->head. (see skb_release_data()) |
| 795 | */ |
| 796 | if (!skb->cloned) |
| 797 | fastpath = true; |
| 798 | else { |
| 799 | int delta = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 800 | |
Eric Dumazet | 1fd6304 | 2010-09-02 23:09:32 +0000 | [diff] [blame] | 801 | fastpath = atomic_read(&skb_shinfo(skb)->dataref) == delta; |
| 802 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 803 | |
Eric Dumazet | 1fd6304 | 2010-09-02 23:09:32 +0000 | [diff] [blame] | 804 | if (fastpath) { |
| 805 | kfree(skb->head); |
| 806 | } else { |
| 807 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) |
| 808 | get_page(skb_shinfo(skb)->frags[i].page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 809 | |
Eric Dumazet | 1fd6304 | 2010-09-02 23:09:32 +0000 | [diff] [blame] | 810 | if (skb_has_frag_list(skb)) |
| 811 | skb_clone_fraglist(skb); |
| 812 | |
| 813 | skb_release_data(skb); |
| 814 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 815 | off = (data + nhead) - skb->head; |
| 816 | |
| 817 | skb->head = data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 818 | skb->data += off; |
Arnaldo Carvalho de Melo | 4305b54 | 2007-04-19 20:43:29 -0700 | [diff] [blame] | 819 | #ifdef NET_SKBUFF_DATA_USES_OFFSET |
| 820 | skb->end = size; |
Patrick McHardy | 56eb888 | 2007-04-09 11:45:04 -0700 | [diff] [blame] | 821 | off = nhead; |
Arnaldo Carvalho de Melo | 4305b54 | 2007-04-19 20:43:29 -0700 | [diff] [blame] | 822 | #else |
| 823 | skb->end = skb->head + size; |
Patrick McHardy | 56eb888 | 2007-04-09 11:45:04 -0700 | [diff] [blame] | 824 | #endif |
Arnaldo Carvalho de Melo | 27a884d | 2007-04-19 20:29:13 -0700 | [diff] [blame] | 825 | /* {transport,network,mac}_header and tail are relative to skb->head */ |
| 826 | skb->tail += off; |
Arnaldo Carvalho de Melo | b0e380b | 2007-04-10 21:21:55 -0700 | [diff] [blame] | 827 | skb->transport_header += off; |
| 828 | skb->network_header += off; |
Stephen Hemminger | 603a8bb | 2009-06-17 12:17:34 +0000 | [diff] [blame] | 829 | if (skb_mac_header_was_set(skb)) |
| 830 | skb->mac_header += off; |
Andrea Shepard | 00c5a98 | 2010-07-22 09:12:35 +0000 | [diff] [blame] | 831 | /* Only adjust this if it actually is csum_start rather than csum */ |
| 832 | if (skb->ip_summed == CHECKSUM_PARTIAL) |
| 833 | skb->csum_start += nhead; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 834 | skb->cloned = 0; |
Patrick McHardy | 334a813 | 2007-06-25 04:35:20 -0700 | [diff] [blame] | 835 | skb->hdr_len = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 836 | skb->nohdr = 0; |
| 837 | atomic_set(&skb_shinfo(skb)->dataref, 1); |
| 838 | return 0; |
| 839 | |
| 840 | nodata: |
| 841 | return -ENOMEM; |
| 842 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 843 | EXPORT_SYMBOL(pskb_expand_head); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 844 | |
| 845 | /* Make private copy of skb with writable head and some headroom */ |
| 846 | |
| 847 | struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom) |
| 848 | { |
| 849 | struct sk_buff *skb2; |
| 850 | int delta = headroom - skb_headroom(skb); |
| 851 | |
| 852 | if (delta <= 0) |
| 853 | skb2 = pskb_copy(skb, GFP_ATOMIC); |
| 854 | else { |
| 855 | skb2 = skb_clone(skb, GFP_ATOMIC); |
| 856 | if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0, |
| 857 | GFP_ATOMIC)) { |
| 858 | kfree_skb(skb2); |
| 859 | skb2 = NULL; |
| 860 | } |
| 861 | } |
| 862 | return skb2; |
| 863 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 864 | EXPORT_SYMBOL(skb_realloc_headroom); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 865 | |
| 866 | /** |
| 867 | * skb_copy_expand - copy and expand sk_buff |
| 868 | * @skb: buffer to copy |
| 869 | * @newheadroom: new free bytes at head |
| 870 | * @newtailroom: new free bytes at tail |
| 871 | * @gfp_mask: allocation priority |
| 872 | * |
| 873 | * Make a copy of both an &sk_buff and its data and while doing so |
| 874 | * allocate additional space. |
| 875 | * |
| 876 | * This is used when the caller wishes to modify the data and needs a |
| 877 | * private copy of the data to alter as well as more space for new fields. |
| 878 | * Returns %NULL on failure or the pointer to the buffer |
| 879 | * on success. The returned buffer has a reference count of 1. |
| 880 | * |
| 881 | * You must pass %GFP_ATOMIC as the allocation priority if this function |
| 882 | * is called from an interrupt. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 883 | */ |
| 884 | struct sk_buff *skb_copy_expand(const struct sk_buff *skb, |
Victor Fusco | 86a76ca | 2005-07-08 14:57:47 -0700 | [diff] [blame] | 885 | int newheadroom, int newtailroom, |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 886 | gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 887 | { |
| 888 | /* |
| 889 | * Allocate the copy buffer |
| 890 | */ |
| 891 | struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom, |
| 892 | gfp_mask); |
Patrick McHardy | efd1e8d | 2007-04-10 18:30:09 -0700 | [diff] [blame] | 893 | int oldheadroom = skb_headroom(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 894 | int head_copy_len, head_copy_off; |
Herbert Xu | 5288605 | 2007-09-16 16:32:11 -0700 | [diff] [blame] | 895 | int off; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 896 | |
| 897 | if (!n) |
| 898 | return NULL; |
| 899 | |
| 900 | skb_reserve(n, newheadroom); |
| 901 | |
| 902 | /* Set the tail pointer and length */ |
| 903 | skb_put(n, skb->len); |
| 904 | |
Patrick McHardy | efd1e8d | 2007-04-10 18:30:09 -0700 | [diff] [blame] | 905 | head_copy_len = oldheadroom; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 906 | head_copy_off = 0; |
| 907 | if (newheadroom <= head_copy_len) |
| 908 | head_copy_len = newheadroom; |
| 909 | else |
| 910 | head_copy_off = newheadroom - head_copy_len; |
| 911 | |
| 912 | /* Copy the linear header and data. */ |
| 913 | if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off, |
| 914 | skb->len + head_copy_len)) |
| 915 | BUG(); |
| 916 | |
| 917 | copy_skb_header(n, skb); |
| 918 | |
Patrick McHardy | efd1e8d | 2007-04-10 18:30:09 -0700 | [diff] [blame] | 919 | off = newheadroom - oldheadroom; |
David S. Miller | be2b6e6 | 2010-07-22 13:27:09 -0700 | [diff] [blame] | 920 | if (n->ip_summed == CHECKSUM_PARTIAL) |
| 921 | n->csum_start += off; |
Herbert Xu | 5288605 | 2007-09-16 16:32:11 -0700 | [diff] [blame] | 922 | #ifdef NET_SKBUFF_DATA_USES_OFFSET |
Patrick McHardy | efd1e8d | 2007-04-10 18:30:09 -0700 | [diff] [blame] | 923 | n->transport_header += off; |
| 924 | n->network_header += off; |
Stephen Hemminger | 603a8bb | 2009-06-17 12:17:34 +0000 | [diff] [blame] | 925 | if (skb_mac_header_was_set(skb)) |
| 926 | n->mac_header += off; |
Herbert Xu | 5288605 | 2007-09-16 16:32:11 -0700 | [diff] [blame] | 927 | #endif |
Patrick McHardy | efd1e8d | 2007-04-10 18:30:09 -0700 | [diff] [blame] | 928 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 929 | return n; |
| 930 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 931 | EXPORT_SYMBOL(skb_copy_expand); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 932 | |
| 933 | /** |
| 934 | * skb_pad - zero pad the tail of an skb |
| 935 | * @skb: buffer to pad |
| 936 | * @pad: space to pad |
| 937 | * |
| 938 | * Ensure that a buffer is followed by a padding area that is zero |
| 939 | * filled. Used by network drivers which may DMA or transfer data |
| 940 | * beyond the buffer end onto the wire. |
| 941 | * |
Herbert Xu | 5b057c6 | 2006-06-23 02:06:41 -0700 | [diff] [blame] | 942 | * May return error in out of memory cases. The skb is freed on error. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 943 | */ |
YOSHIFUJI Hideaki | 4ec93ed | 2007-02-09 23:24:36 +0900 | [diff] [blame] | 944 | |
Herbert Xu | 5b057c6 | 2006-06-23 02:06:41 -0700 | [diff] [blame] | 945 | int skb_pad(struct sk_buff *skb, int pad) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 946 | { |
Herbert Xu | 5b057c6 | 2006-06-23 02:06:41 -0700 | [diff] [blame] | 947 | int err; |
| 948 | int ntail; |
YOSHIFUJI Hideaki | 4ec93ed | 2007-02-09 23:24:36 +0900 | [diff] [blame] | 949 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 950 | /* If the skbuff is non linear tailroom is always zero.. */ |
Herbert Xu | 5b057c6 | 2006-06-23 02:06:41 -0700 | [diff] [blame] | 951 | if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 952 | memset(skb->data+skb->len, 0, pad); |
Herbert Xu | 5b057c6 | 2006-06-23 02:06:41 -0700 | [diff] [blame] | 953 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 954 | } |
Herbert Xu | 5b057c6 | 2006-06-23 02:06:41 -0700 | [diff] [blame] | 955 | |
Arnaldo Carvalho de Melo | 4305b54 | 2007-04-19 20:43:29 -0700 | [diff] [blame] | 956 | ntail = skb->data_len + pad - (skb->end - skb->tail); |
Herbert Xu | 5b057c6 | 2006-06-23 02:06:41 -0700 | [diff] [blame] | 957 | if (likely(skb_cloned(skb) || ntail > 0)) { |
| 958 | err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC); |
| 959 | if (unlikely(err)) |
| 960 | goto free_skb; |
| 961 | } |
| 962 | |
| 963 | /* FIXME: The use of this function with non-linear skb's really needs |
| 964 | * to be audited. |
| 965 | */ |
| 966 | err = skb_linearize(skb); |
| 967 | if (unlikely(err)) |
| 968 | goto free_skb; |
| 969 | |
| 970 | memset(skb->data + skb->len, 0, pad); |
| 971 | return 0; |
| 972 | |
| 973 | free_skb: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 974 | kfree_skb(skb); |
Herbert Xu | 5b057c6 | 2006-06-23 02:06:41 -0700 | [diff] [blame] | 975 | return err; |
YOSHIFUJI Hideaki | 4ec93ed | 2007-02-09 23:24:36 +0900 | [diff] [blame] | 976 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 977 | EXPORT_SYMBOL(skb_pad); |
YOSHIFUJI Hideaki | 4ec93ed | 2007-02-09 23:24:36 +0900 | [diff] [blame] | 978 | |
Ilpo Järvinen | 0dde3e1 | 2008-03-27 17:43:41 -0700 | [diff] [blame] | 979 | /** |
| 980 | * skb_put - add data to a buffer |
| 981 | * @skb: buffer to use |
| 982 | * @len: amount of data to add |
| 983 | * |
| 984 | * This function extends the used data area of the buffer. If this would |
| 985 | * exceed the total buffer size the kernel will panic. A pointer to the |
| 986 | * first byte of the extra data is returned. |
| 987 | */ |
| 988 | unsigned char *skb_put(struct sk_buff *skb, unsigned int len) |
| 989 | { |
| 990 | unsigned char *tmp = skb_tail_pointer(skb); |
| 991 | SKB_LINEAR_ASSERT(skb); |
| 992 | skb->tail += len; |
| 993 | skb->len += len; |
| 994 | if (unlikely(skb->tail > skb->end)) |
| 995 | skb_over_panic(skb, len, __builtin_return_address(0)); |
| 996 | return tmp; |
| 997 | } |
| 998 | EXPORT_SYMBOL(skb_put); |
| 999 | |
Ilpo Järvinen | 6be8ac2 | 2008-03-27 17:47:24 -0700 | [diff] [blame] | 1000 | /** |
Ilpo Järvinen | c2aa270 | 2008-03-27 17:52:40 -0700 | [diff] [blame] | 1001 | * skb_push - add data to the start of a buffer |
| 1002 | * @skb: buffer to use |
| 1003 | * @len: amount of data to add |
| 1004 | * |
| 1005 | * This function extends the used data area of the buffer at the buffer |
| 1006 | * start. If this would exceed the total buffer headroom the kernel will |
| 1007 | * panic. A pointer to the first byte of the extra data is returned. |
| 1008 | */ |
| 1009 | unsigned char *skb_push(struct sk_buff *skb, unsigned int len) |
| 1010 | { |
| 1011 | skb->data -= len; |
| 1012 | skb->len += len; |
| 1013 | if (unlikely(skb->data<skb->head)) |
| 1014 | skb_under_panic(skb, len, __builtin_return_address(0)); |
| 1015 | return skb->data; |
| 1016 | } |
| 1017 | EXPORT_SYMBOL(skb_push); |
| 1018 | |
| 1019 | /** |
Ilpo Järvinen | 6be8ac2 | 2008-03-27 17:47:24 -0700 | [diff] [blame] | 1020 | * skb_pull - remove data from the start of a buffer |
| 1021 | * @skb: buffer to use |
| 1022 | * @len: amount of data to remove |
| 1023 | * |
| 1024 | * This function removes data from the start of a buffer, returning |
| 1025 | * the memory to the headroom. A pointer to the next data in the buffer |
| 1026 | * is returned. Once the data has been pulled future pushes will overwrite |
| 1027 | * the old data. |
| 1028 | */ |
| 1029 | unsigned char *skb_pull(struct sk_buff *skb, unsigned int len) |
| 1030 | { |
David S. Miller | 47d2964 | 2010-05-02 02:21:44 -0700 | [diff] [blame] | 1031 | return skb_pull_inline(skb, len); |
Ilpo Järvinen | 6be8ac2 | 2008-03-27 17:47:24 -0700 | [diff] [blame] | 1032 | } |
| 1033 | EXPORT_SYMBOL(skb_pull); |
| 1034 | |
Ilpo Järvinen | 419ae74 | 2008-03-27 17:54:01 -0700 | [diff] [blame] | 1035 | /** |
| 1036 | * skb_trim - remove end from a buffer |
| 1037 | * @skb: buffer to alter |
| 1038 | * @len: new length |
| 1039 | * |
| 1040 | * Cut the length of a buffer down by removing data from the tail. If |
| 1041 | * the buffer is already under the length specified it is not modified. |
| 1042 | * The skb must be linear. |
| 1043 | */ |
| 1044 | void skb_trim(struct sk_buff *skb, unsigned int len) |
| 1045 | { |
| 1046 | if (skb->len > len) |
| 1047 | __skb_trim(skb, len); |
| 1048 | } |
| 1049 | EXPORT_SYMBOL(skb_trim); |
| 1050 | |
Herbert Xu | 3cc0e87 | 2006-06-09 16:13:38 -0700 | [diff] [blame] | 1051 | /* Trims skb to length len. It can change skb pointers. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1052 | */ |
| 1053 | |
Herbert Xu | 3cc0e87 | 2006-06-09 16:13:38 -0700 | [diff] [blame] | 1054 | int ___pskb_trim(struct sk_buff *skb, unsigned int len) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1055 | { |
Herbert Xu | 27b437c | 2006-07-13 19:26:39 -0700 | [diff] [blame] | 1056 | struct sk_buff **fragp; |
| 1057 | struct sk_buff *frag; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1058 | int offset = skb_headlen(skb); |
| 1059 | int nfrags = skb_shinfo(skb)->nr_frags; |
| 1060 | int i; |
Herbert Xu | 27b437c | 2006-07-13 19:26:39 -0700 | [diff] [blame] | 1061 | int err; |
| 1062 | |
| 1063 | if (skb_cloned(skb) && |
| 1064 | unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))) |
| 1065 | return err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1066 | |
Herbert Xu | f4d26fb | 2006-07-30 20:20:28 -0700 | [diff] [blame] | 1067 | i = 0; |
| 1068 | if (offset >= len) |
| 1069 | goto drop_pages; |
| 1070 | |
| 1071 | for (; i < nfrags; i++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1072 | int end = offset + skb_shinfo(skb)->frags[i].size; |
Herbert Xu | 27b437c | 2006-07-13 19:26:39 -0700 | [diff] [blame] | 1073 | |
| 1074 | if (end < len) { |
| 1075 | offset = end; |
| 1076 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1077 | } |
Herbert Xu | 27b437c | 2006-07-13 19:26:39 -0700 | [diff] [blame] | 1078 | |
Herbert Xu | f4d26fb | 2006-07-30 20:20:28 -0700 | [diff] [blame] | 1079 | skb_shinfo(skb)->frags[i++].size = len - offset; |
Herbert Xu | 27b437c | 2006-07-13 19:26:39 -0700 | [diff] [blame] | 1080 | |
Herbert Xu | f4d26fb | 2006-07-30 20:20:28 -0700 | [diff] [blame] | 1081 | drop_pages: |
Herbert Xu | 27b437c | 2006-07-13 19:26:39 -0700 | [diff] [blame] | 1082 | skb_shinfo(skb)->nr_frags = i; |
| 1083 | |
| 1084 | for (; i < nfrags; i++) |
| 1085 | put_page(skb_shinfo(skb)->frags[i].page); |
| 1086 | |
David S. Miller | 21dc330 | 2010-08-23 00:13:46 -0700 | [diff] [blame] | 1087 | if (skb_has_frag_list(skb)) |
Herbert Xu | 27b437c | 2006-07-13 19:26:39 -0700 | [diff] [blame] | 1088 | skb_drop_fraglist(skb); |
Herbert Xu | f4d26fb | 2006-07-30 20:20:28 -0700 | [diff] [blame] | 1089 | goto done; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1090 | } |
| 1091 | |
Herbert Xu | 27b437c | 2006-07-13 19:26:39 -0700 | [diff] [blame] | 1092 | for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp); |
| 1093 | fragp = &frag->next) { |
| 1094 | int end = offset + frag->len; |
| 1095 | |
| 1096 | if (skb_shared(frag)) { |
| 1097 | struct sk_buff *nfrag; |
| 1098 | |
| 1099 | nfrag = skb_clone(frag, GFP_ATOMIC); |
| 1100 | if (unlikely(!nfrag)) |
| 1101 | return -ENOMEM; |
| 1102 | |
| 1103 | nfrag->next = frag->next; |
Herbert Xu | f4d26fb | 2006-07-30 20:20:28 -0700 | [diff] [blame] | 1104 | kfree_skb(frag); |
Herbert Xu | 27b437c | 2006-07-13 19:26:39 -0700 | [diff] [blame] | 1105 | frag = nfrag; |
| 1106 | *fragp = frag; |
| 1107 | } |
| 1108 | |
| 1109 | if (end < len) { |
| 1110 | offset = end; |
| 1111 | continue; |
| 1112 | } |
| 1113 | |
| 1114 | if (end > len && |
| 1115 | unlikely((err = pskb_trim(frag, len - offset)))) |
| 1116 | return err; |
| 1117 | |
| 1118 | if (frag->next) |
| 1119 | skb_drop_list(&frag->next); |
| 1120 | break; |
| 1121 | } |
| 1122 | |
Herbert Xu | f4d26fb | 2006-07-30 20:20:28 -0700 | [diff] [blame] | 1123 | done: |
Herbert Xu | 27b437c | 2006-07-13 19:26:39 -0700 | [diff] [blame] | 1124 | if (len > skb_headlen(skb)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1125 | skb->data_len -= skb->len - len; |
| 1126 | skb->len = len; |
| 1127 | } else { |
Herbert Xu | 27b437c | 2006-07-13 19:26:39 -0700 | [diff] [blame] | 1128 | skb->len = len; |
| 1129 | skb->data_len = 0; |
Arnaldo Carvalho de Melo | 27a884d | 2007-04-19 20:29:13 -0700 | [diff] [blame] | 1130 | skb_set_tail_pointer(skb, len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1131 | } |
| 1132 | |
| 1133 | return 0; |
| 1134 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 1135 | EXPORT_SYMBOL(___pskb_trim); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1136 | |
| 1137 | /** |
| 1138 | * __pskb_pull_tail - advance tail of skb header |
| 1139 | * @skb: buffer to reallocate |
| 1140 | * @delta: number of bytes to advance tail |
| 1141 | * |
| 1142 | * The function makes a sense only on a fragmented &sk_buff, |
| 1143 | * it expands header moving its tail forward and copying necessary |
| 1144 | * data from fragmented part. |
| 1145 | * |
| 1146 | * &sk_buff MUST have reference count of 1. |
| 1147 | * |
| 1148 | * Returns %NULL (and &sk_buff does not change) if pull failed |
| 1149 | * or value of new tail of skb in the case of success. |
| 1150 | * |
| 1151 | * All the pointers pointing into skb header may change and must be |
| 1152 | * reloaded after call to this function. |
| 1153 | */ |
| 1154 | |
| 1155 | /* Moves tail of skb head forward, copying data from fragmented part, |
| 1156 | * when it is necessary. |
| 1157 | * 1. It may fail due to malloc failure. |
| 1158 | * 2. It may change skb pointers. |
| 1159 | * |
| 1160 | * It is pretty complicated. Luckily, it is called only in exceptional cases. |
| 1161 | */ |
| 1162 | unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta) |
| 1163 | { |
| 1164 | /* If skb has not enough free space at tail, get new one |
| 1165 | * plus 128 bytes for future expansions. If we have enough |
| 1166 | * room at tail, reallocate without expansion only if skb is cloned. |
| 1167 | */ |
Arnaldo Carvalho de Melo | 4305b54 | 2007-04-19 20:43:29 -0700 | [diff] [blame] | 1168 | int i, k, eat = (skb->tail + delta) - skb->end; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1169 | |
| 1170 | if (eat > 0 || skb_cloned(skb)) { |
| 1171 | if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0, |
| 1172 | GFP_ATOMIC)) |
| 1173 | return NULL; |
| 1174 | } |
| 1175 | |
Arnaldo Carvalho de Melo | 27a884d | 2007-04-19 20:29:13 -0700 | [diff] [blame] | 1176 | if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1177 | BUG(); |
| 1178 | |
| 1179 | /* Optimization: no fragments, no reasons to preestimate |
| 1180 | * size of pulled pages. Superb. |
| 1181 | */ |
David S. Miller | 21dc330 | 2010-08-23 00:13:46 -0700 | [diff] [blame] | 1182 | if (!skb_has_frag_list(skb)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1183 | goto pull_pages; |
| 1184 | |
| 1185 | /* Estimate size of pulled pages. */ |
| 1186 | eat = delta; |
| 1187 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
| 1188 | if (skb_shinfo(skb)->frags[i].size >= eat) |
| 1189 | goto pull_pages; |
| 1190 | eat -= skb_shinfo(skb)->frags[i].size; |
| 1191 | } |
| 1192 | |
| 1193 | /* If we need update frag list, we are in troubles. |
| 1194 | * Certainly, it possible to add an offset to skb data, |
| 1195 | * but taking into account that pulling is expected to |
| 1196 | * be very rare operation, it is worth to fight against |
| 1197 | * further bloating skb head and crucify ourselves here instead. |
| 1198 | * Pure masohism, indeed. 8)8) |
| 1199 | */ |
| 1200 | if (eat) { |
| 1201 | struct sk_buff *list = skb_shinfo(skb)->frag_list; |
| 1202 | struct sk_buff *clone = NULL; |
| 1203 | struct sk_buff *insp = NULL; |
| 1204 | |
| 1205 | do { |
Kris Katterjohn | 09a6266 | 2006-01-08 22:24:28 -0800 | [diff] [blame] | 1206 | BUG_ON(!list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1207 | |
| 1208 | if (list->len <= eat) { |
| 1209 | /* Eaten as whole. */ |
| 1210 | eat -= list->len; |
| 1211 | list = list->next; |
| 1212 | insp = list; |
| 1213 | } else { |
| 1214 | /* Eaten partially. */ |
| 1215 | |
| 1216 | if (skb_shared(list)) { |
| 1217 | /* Sucks! We need to fork list. :-( */ |
| 1218 | clone = skb_clone(list, GFP_ATOMIC); |
| 1219 | if (!clone) |
| 1220 | return NULL; |
| 1221 | insp = list->next; |
| 1222 | list = clone; |
| 1223 | } else { |
| 1224 | /* This may be pulled without |
| 1225 | * problems. */ |
| 1226 | insp = list; |
| 1227 | } |
| 1228 | if (!pskb_pull(list, eat)) { |
Wei Yongjun | f3fbbe0 | 2009-02-25 00:37:32 +0000 | [diff] [blame] | 1229 | kfree_skb(clone); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1230 | return NULL; |
| 1231 | } |
| 1232 | break; |
| 1233 | } |
| 1234 | } while (eat); |
| 1235 | |
| 1236 | /* Free pulled out fragments. */ |
| 1237 | while ((list = skb_shinfo(skb)->frag_list) != insp) { |
| 1238 | skb_shinfo(skb)->frag_list = list->next; |
| 1239 | kfree_skb(list); |
| 1240 | } |
| 1241 | /* And insert new clone at head. */ |
| 1242 | if (clone) { |
| 1243 | clone->next = list; |
| 1244 | skb_shinfo(skb)->frag_list = clone; |
| 1245 | } |
| 1246 | } |
| 1247 | /* Success! Now we may commit changes to skb data. */ |
| 1248 | |
| 1249 | pull_pages: |
| 1250 | eat = delta; |
| 1251 | k = 0; |
| 1252 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
| 1253 | if (skb_shinfo(skb)->frags[i].size <= eat) { |
| 1254 | put_page(skb_shinfo(skb)->frags[i].page); |
| 1255 | eat -= skb_shinfo(skb)->frags[i].size; |
| 1256 | } else { |
| 1257 | skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i]; |
| 1258 | if (eat) { |
| 1259 | skb_shinfo(skb)->frags[k].page_offset += eat; |
| 1260 | skb_shinfo(skb)->frags[k].size -= eat; |
| 1261 | eat = 0; |
| 1262 | } |
| 1263 | k++; |
| 1264 | } |
| 1265 | } |
| 1266 | skb_shinfo(skb)->nr_frags = k; |
| 1267 | |
| 1268 | skb->tail += delta; |
| 1269 | skb->data_len -= delta; |
| 1270 | |
Arnaldo Carvalho de Melo | 27a884d | 2007-04-19 20:29:13 -0700 | [diff] [blame] | 1271 | return skb_tail_pointer(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1272 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 1273 | EXPORT_SYMBOL(__pskb_pull_tail); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1274 | |
| 1275 | /* Copy some data bits from skb to kernel buffer. */ |
| 1276 | |
| 1277 | int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len) |
| 1278 | { |
David S. Miller | 1a028e5 | 2007-04-27 15:21:23 -0700 | [diff] [blame] | 1279 | int start = skb_headlen(skb); |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 1280 | struct sk_buff *frag_iter; |
| 1281 | int i, copy; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1282 | |
| 1283 | if (offset > (int)skb->len - len) |
| 1284 | goto fault; |
| 1285 | |
| 1286 | /* Copy header. */ |
David S. Miller | 1a028e5 | 2007-04-27 15:21:23 -0700 | [diff] [blame] | 1287 | if ((copy = start - offset) > 0) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1288 | if (copy > len) |
| 1289 | copy = len; |
Arnaldo Carvalho de Melo | d626f62 | 2007-03-27 18:55:52 -0300 | [diff] [blame] | 1290 | skb_copy_from_linear_data_offset(skb, offset, to, copy); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1291 | if ((len -= copy) == 0) |
| 1292 | return 0; |
| 1293 | offset += copy; |
| 1294 | to += copy; |
| 1295 | } |
| 1296 | |
| 1297 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
David S. Miller | 1a028e5 | 2007-04-27 15:21:23 -0700 | [diff] [blame] | 1298 | int end; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1299 | |
Ilpo Järvinen | 547b792 | 2008-07-25 21:43:18 -0700 | [diff] [blame] | 1300 | WARN_ON(start > offset + len); |
David S. Miller | 1a028e5 | 2007-04-27 15:21:23 -0700 | [diff] [blame] | 1301 | |
| 1302 | end = start + skb_shinfo(skb)->frags[i].size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1303 | if ((copy = end - offset) > 0) { |
| 1304 | u8 *vaddr; |
| 1305 | |
| 1306 | if (copy > len) |
| 1307 | copy = len; |
| 1308 | |
| 1309 | vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]); |
| 1310 | memcpy(to, |
David S. Miller | 1a028e5 | 2007-04-27 15:21:23 -0700 | [diff] [blame] | 1311 | vaddr + skb_shinfo(skb)->frags[i].page_offset+ |
| 1312 | offset - start, copy); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1313 | kunmap_skb_frag(vaddr); |
| 1314 | |
| 1315 | if ((len -= copy) == 0) |
| 1316 | return 0; |
| 1317 | offset += copy; |
| 1318 | to += copy; |
| 1319 | } |
David S. Miller | 1a028e5 | 2007-04-27 15:21:23 -0700 | [diff] [blame] | 1320 | start = end; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1321 | } |
| 1322 | |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 1323 | skb_walk_frags(skb, frag_iter) { |
| 1324 | int end; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1325 | |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 1326 | WARN_ON(start > offset + len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1327 | |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 1328 | end = start + frag_iter->len; |
| 1329 | if ((copy = end - offset) > 0) { |
| 1330 | if (copy > len) |
| 1331 | copy = len; |
| 1332 | if (skb_copy_bits(frag_iter, offset - start, to, copy)) |
| 1333 | goto fault; |
| 1334 | if ((len -= copy) == 0) |
| 1335 | return 0; |
| 1336 | offset += copy; |
| 1337 | to += copy; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1338 | } |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 1339 | start = end; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1340 | } |
| 1341 | if (!len) |
| 1342 | return 0; |
| 1343 | |
| 1344 | fault: |
| 1345 | return -EFAULT; |
| 1346 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 1347 | EXPORT_SYMBOL(skb_copy_bits); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1348 | |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1349 | /* |
| 1350 | * Callback from splice_to_pipe(), if we need to release some pages |
| 1351 | * at the end of the spd in case we error'ed out in filling the pipe. |
| 1352 | */ |
| 1353 | static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i) |
| 1354 | { |
Jarek Poplawski | 8b9d372 | 2009-01-19 17:03:56 -0800 | [diff] [blame] | 1355 | put_page(spd->pages[i]); |
| 1356 | } |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1357 | |
Jarek Poplawski | 4fb6699 | 2009-02-01 00:41:42 -0800 | [diff] [blame] | 1358 | static inline struct page *linear_to_page(struct page *page, unsigned int *len, |
| 1359 | unsigned int *offset, |
Jarek Poplawski | 7a67e56 | 2009-04-30 05:41:19 -0700 | [diff] [blame] | 1360 | struct sk_buff *skb, struct sock *sk) |
Jarek Poplawski | 8b9d372 | 2009-01-19 17:03:56 -0800 | [diff] [blame] | 1361 | { |
Jarek Poplawski | 4fb6699 | 2009-02-01 00:41:42 -0800 | [diff] [blame] | 1362 | struct page *p = sk->sk_sndmsg_page; |
| 1363 | unsigned int off; |
Jarek Poplawski | 8b9d372 | 2009-01-19 17:03:56 -0800 | [diff] [blame] | 1364 | |
Jarek Poplawski | 4fb6699 | 2009-02-01 00:41:42 -0800 | [diff] [blame] | 1365 | if (!p) { |
| 1366 | new_page: |
| 1367 | p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0); |
| 1368 | if (!p) |
| 1369 | return NULL; |
| 1370 | |
| 1371 | off = sk->sk_sndmsg_off = 0; |
| 1372 | /* hold one ref to this page until it's full */ |
| 1373 | } else { |
| 1374 | unsigned int mlen; |
| 1375 | |
| 1376 | off = sk->sk_sndmsg_off; |
| 1377 | mlen = PAGE_SIZE - off; |
| 1378 | if (mlen < 64 && mlen < *len) { |
| 1379 | put_page(p); |
| 1380 | goto new_page; |
| 1381 | } |
| 1382 | |
| 1383 | *len = min_t(unsigned int, *len, mlen); |
| 1384 | } |
| 1385 | |
| 1386 | memcpy(page_address(p) + off, page_address(page) + *offset, *len); |
| 1387 | sk->sk_sndmsg_off += *len; |
| 1388 | *offset = off; |
| 1389 | get_page(p); |
Jarek Poplawski | 8b9d372 | 2009-01-19 17:03:56 -0800 | [diff] [blame] | 1390 | |
| 1391 | return p; |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1392 | } |
| 1393 | |
| 1394 | /* |
| 1395 | * Fill page/offset/length into spd, if it can hold more pages. |
| 1396 | */ |
Jens Axboe | 35f3d14 | 2010-05-20 10:43:18 +0200 | [diff] [blame] | 1397 | static inline int spd_fill_page(struct splice_pipe_desc *spd, |
| 1398 | struct pipe_inode_info *pipe, struct page *page, |
Jarek Poplawski | 4fb6699 | 2009-02-01 00:41:42 -0800 | [diff] [blame] | 1399 | unsigned int *len, unsigned int offset, |
Jarek Poplawski | 7a67e56 | 2009-04-30 05:41:19 -0700 | [diff] [blame] | 1400 | struct sk_buff *skb, int linear, |
| 1401 | struct sock *sk) |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1402 | { |
Jens Axboe | 35f3d14 | 2010-05-20 10:43:18 +0200 | [diff] [blame] | 1403 | if (unlikely(spd->nr_pages == pipe->buffers)) |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1404 | return 1; |
| 1405 | |
Jarek Poplawski | 8b9d372 | 2009-01-19 17:03:56 -0800 | [diff] [blame] | 1406 | if (linear) { |
Jarek Poplawski | 7a67e56 | 2009-04-30 05:41:19 -0700 | [diff] [blame] | 1407 | page = linear_to_page(page, len, &offset, skb, sk); |
Jarek Poplawski | 8b9d372 | 2009-01-19 17:03:56 -0800 | [diff] [blame] | 1408 | if (!page) |
| 1409 | return 1; |
| 1410 | } else |
| 1411 | get_page(page); |
| 1412 | |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1413 | spd->pages[spd->nr_pages] = page; |
Jarek Poplawski | 4fb6699 | 2009-02-01 00:41:42 -0800 | [diff] [blame] | 1414 | spd->partial[spd->nr_pages].len = *len; |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1415 | spd->partial[spd->nr_pages].offset = offset; |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1416 | spd->nr_pages++; |
Jarek Poplawski | 8b9d372 | 2009-01-19 17:03:56 -0800 | [diff] [blame] | 1417 | |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1418 | return 0; |
| 1419 | } |
| 1420 | |
Octavian Purdila | 2870c43 | 2008-07-15 00:49:11 -0700 | [diff] [blame] | 1421 | static inline void __segment_seek(struct page **page, unsigned int *poff, |
| 1422 | unsigned int *plen, unsigned int off) |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1423 | { |
Jarek Poplawski | ce3dd39 | 2009-02-12 16:51:43 -0800 | [diff] [blame] | 1424 | unsigned long n; |
| 1425 | |
Octavian Purdila | 2870c43 | 2008-07-15 00:49:11 -0700 | [diff] [blame] | 1426 | *poff += off; |
Jarek Poplawski | ce3dd39 | 2009-02-12 16:51:43 -0800 | [diff] [blame] | 1427 | n = *poff / PAGE_SIZE; |
| 1428 | if (n) |
| 1429 | *page = nth_page(*page, n); |
| 1430 | |
Octavian Purdila | 2870c43 | 2008-07-15 00:49:11 -0700 | [diff] [blame] | 1431 | *poff = *poff % PAGE_SIZE; |
| 1432 | *plen -= off; |
| 1433 | } |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1434 | |
Octavian Purdila | 2870c43 | 2008-07-15 00:49:11 -0700 | [diff] [blame] | 1435 | static inline int __splice_segment(struct page *page, unsigned int poff, |
| 1436 | unsigned int plen, unsigned int *off, |
| 1437 | unsigned int *len, struct sk_buff *skb, |
Jarek Poplawski | 7a67e56 | 2009-04-30 05:41:19 -0700 | [diff] [blame] | 1438 | struct splice_pipe_desc *spd, int linear, |
Jens Axboe | 35f3d14 | 2010-05-20 10:43:18 +0200 | [diff] [blame] | 1439 | struct sock *sk, |
| 1440 | struct pipe_inode_info *pipe) |
Octavian Purdila | 2870c43 | 2008-07-15 00:49:11 -0700 | [diff] [blame] | 1441 | { |
| 1442 | if (!*len) |
| 1443 | return 1; |
| 1444 | |
| 1445 | /* skip this segment if already processed */ |
| 1446 | if (*off >= plen) { |
| 1447 | *off -= plen; |
| 1448 | return 0; |
Octavian Purdila | db43a28 | 2008-06-27 17:27:21 -0700 | [diff] [blame] | 1449 | } |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1450 | |
Octavian Purdila | 2870c43 | 2008-07-15 00:49:11 -0700 | [diff] [blame] | 1451 | /* ignore any bits we already processed */ |
| 1452 | if (*off) { |
| 1453 | __segment_seek(&page, &poff, &plen, *off); |
| 1454 | *off = 0; |
| 1455 | } |
| 1456 | |
| 1457 | do { |
| 1458 | unsigned int flen = min(*len, plen); |
| 1459 | |
| 1460 | /* the linear region may spread across several pages */ |
| 1461 | flen = min_t(unsigned int, flen, PAGE_SIZE - poff); |
| 1462 | |
Jens Axboe | 35f3d14 | 2010-05-20 10:43:18 +0200 | [diff] [blame] | 1463 | if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk)) |
Octavian Purdila | 2870c43 | 2008-07-15 00:49:11 -0700 | [diff] [blame] | 1464 | return 1; |
| 1465 | |
| 1466 | __segment_seek(&page, &poff, &plen, flen); |
| 1467 | *len -= flen; |
| 1468 | |
| 1469 | } while (*len && plen); |
| 1470 | |
| 1471 | return 0; |
| 1472 | } |
| 1473 | |
| 1474 | /* |
| 1475 | * Map linear and fragment data from the skb to spd. It reports failure if the |
| 1476 | * pipe is full or if we already spliced the requested length. |
| 1477 | */ |
Jens Axboe | 35f3d14 | 2010-05-20 10:43:18 +0200 | [diff] [blame] | 1478 | static int __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe, |
| 1479 | unsigned int *offset, unsigned int *len, |
| 1480 | struct splice_pipe_desc *spd, struct sock *sk) |
Octavian Purdila | 2870c43 | 2008-07-15 00:49:11 -0700 | [diff] [blame] | 1481 | { |
| 1482 | int seg; |
| 1483 | |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1484 | /* |
Octavian Purdila | 2870c43 | 2008-07-15 00:49:11 -0700 | [diff] [blame] | 1485 | * map the linear part |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1486 | */ |
Octavian Purdila | 2870c43 | 2008-07-15 00:49:11 -0700 | [diff] [blame] | 1487 | if (__splice_segment(virt_to_page(skb->data), |
| 1488 | (unsigned long) skb->data & (PAGE_SIZE - 1), |
| 1489 | skb_headlen(skb), |
Jens Axboe | 35f3d14 | 2010-05-20 10:43:18 +0200 | [diff] [blame] | 1490 | offset, len, skb, spd, 1, sk, pipe)) |
Octavian Purdila | 2870c43 | 2008-07-15 00:49:11 -0700 | [diff] [blame] | 1491 | return 1; |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1492 | |
| 1493 | /* |
| 1494 | * then map the fragments |
| 1495 | */ |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1496 | for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) { |
| 1497 | const skb_frag_t *f = &skb_shinfo(skb)->frags[seg]; |
| 1498 | |
Octavian Purdila | 2870c43 | 2008-07-15 00:49:11 -0700 | [diff] [blame] | 1499 | if (__splice_segment(f->page, f->page_offset, f->size, |
Jens Axboe | 35f3d14 | 2010-05-20 10:43:18 +0200 | [diff] [blame] | 1500 | offset, len, skb, spd, 0, sk, pipe)) |
Octavian Purdila | 2870c43 | 2008-07-15 00:49:11 -0700 | [diff] [blame] | 1501 | return 1; |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1502 | } |
| 1503 | |
Octavian Purdila | 2870c43 | 2008-07-15 00:49:11 -0700 | [diff] [blame] | 1504 | return 0; |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1505 | } |
| 1506 | |
| 1507 | /* |
| 1508 | * Map data from the skb to a pipe. Should handle both the linear part, |
| 1509 | * the fragments, and the frag list. It does NOT handle frag lists within |
| 1510 | * the frag list, if such a thing exists. We'd probably need to recurse to |
| 1511 | * handle that cleanly. |
| 1512 | */ |
Jarek Poplawski | 8b9d372 | 2009-01-19 17:03:56 -0800 | [diff] [blame] | 1513 | int skb_splice_bits(struct sk_buff *skb, unsigned int offset, |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1514 | struct pipe_inode_info *pipe, unsigned int tlen, |
| 1515 | unsigned int flags) |
| 1516 | { |
Jens Axboe | 35f3d14 | 2010-05-20 10:43:18 +0200 | [diff] [blame] | 1517 | struct partial_page partial[PIPE_DEF_BUFFERS]; |
| 1518 | struct page *pages[PIPE_DEF_BUFFERS]; |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1519 | struct splice_pipe_desc spd = { |
| 1520 | .pages = pages, |
| 1521 | .partial = partial, |
| 1522 | .flags = flags, |
| 1523 | .ops = &sock_pipe_buf_ops, |
| 1524 | .spd_release = sock_spd_release, |
| 1525 | }; |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 1526 | struct sk_buff *frag_iter; |
Jarek Poplawski | 7a67e56 | 2009-04-30 05:41:19 -0700 | [diff] [blame] | 1527 | struct sock *sk = skb->sk; |
Jens Axboe | 35f3d14 | 2010-05-20 10:43:18 +0200 | [diff] [blame] | 1528 | int ret = 0; |
| 1529 | |
| 1530 | if (splice_grow_spd(pipe, &spd)) |
| 1531 | return -ENOMEM; |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1532 | |
| 1533 | /* |
| 1534 | * __skb_splice_bits() only fails if the output has no room left, |
| 1535 | * so no point in going over the frag_list for the error case. |
| 1536 | */ |
Jens Axboe | 35f3d14 | 2010-05-20 10:43:18 +0200 | [diff] [blame] | 1537 | if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk)) |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1538 | goto done; |
| 1539 | else if (!tlen) |
| 1540 | goto done; |
| 1541 | |
| 1542 | /* |
| 1543 | * now see if we have a frag_list to map |
| 1544 | */ |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 1545 | skb_walk_frags(skb, frag_iter) { |
| 1546 | if (!tlen) |
| 1547 | break; |
Jens Axboe | 35f3d14 | 2010-05-20 10:43:18 +0200 | [diff] [blame] | 1548 | if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk)) |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 1549 | break; |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1550 | } |
| 1551 | |
| 1552 | done: |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1553 | if (spd.nr_pages) { |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1554 | /* |
| 1555 | * Drop the socket lock, otherwise we have reverse |
| 1556 | * locking dependencies between sk_lock and i_mutex |
| 1557 | * here as compared to sendfile(). We enter here |
| 1558 | * with the socket lock held, and splice_to_pipe() will |
| 1559 | * grab the pipe inode lock. For sendfile() emulation, |
| 1560 | * we call into ->sendpage() with the i_mutex lock held |
| 1561 | * and networking will grab the socket lock. |
| 1562 | */ |
Octavian Purdila | 293ad60 | 2008-06-04 15:45:58 -0700 | [diff] [blame] | 1563 | release_sock(sk); |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1564 | ret = splice_to_pipe(pipe, &spd); |
Octavian Purdila | 293ad60 | 2008-06-04 15:45:58 -0700 | [diff] [blame] | 1565 | lock_sock(sk); |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1566 | } |
| 1567 | |
Jens Axboe | 35f3d14 | 2010-05-20 10:43:18 +0200 | [diff] [blame] | 1568 | splice_shrink_spd(pipe, &spd); |
| 1569 | return ret; |
Jens Axboe | 9c55e01 | 2007-11-06 23:30:13 -0800 | [diff] [blame] | 1570 | } |
| 1571 | |
Herbert Xu | 357b40a | 2005-04-19 22:30:14 -0700 | [diff] [blame] | 1572 | /** |
| 1573 | * skb_store_bits - store bits from kernel buffer to skb |
| 1574 | * @skb: destination buffer |
| 1575 | * @offset: offset in destination |
| 1576 | * @from: source buffer |
| 1577 | * @len: number of bytes to copy |
| 1578 | * |
| 1579 | * Copy the specified number of bytes from the source buffer to the |
| 1580 | * destination skb. This function handles all the messy bits of |
| 1581 | * traversing fragment lists and such. |
| 1582 | */ |
| 1583 | |
Stephen Hemminger | 0c6fcc8 | 2007-04-20 16:40:01 -0700 | [diff] [blame] | 1584 | int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len) |
Herbert Xu | 357b40a | 2005-04-19 22:30:14 -0700 | [diff] [blame] | 1585 | { |
David S. Miller | 1a028e5 | 2007-04-27 15:21:23 -0700 | [diff] [blame] | 1586 | int start = skb_headlen(skb); |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 1587 | struct sk_buff *frag_iter; |
| 1588 | int i, copy; |
Herbert Xu | 357b40a | 2005-04-19 22:30:14 -0700 | [diff] [blame] | 1589 | |
| 1590 | if (offset > (int)skb->len - len) |
| 1591 | goto fault; |
| 1592 | |
David S. Miller | 1a028e5 | 2007-04-27 15:21:23 -0700 | [diff] [blame] | 1593 | if ((copy = start - offset) > 0) { |
Herbert Xu | 357b40a | 2005-04-19 22:30:14 -0700 | [diff] [blame] | 1594 | if (copy > len) |
| 1595 | copy = len; |
Arnaldo Carvalho de Melo | 27d7ff4 | 2007-03-31 11:55:19 -0300 | [diff] [blame] | 1596 | skb_copy_to_linear_data_offset(skb, offset, from, copy); |
Herbert Xu | 357b40a | 2005-04-19 22:30:14 -0700 | [diff] [blame] | 1597 | if ((len -= copy) == 0) |
| 1598 | return 0; |
| 1599 | offset += copy; |
| 1600 | from += copy; |
| 1601 | } |
| 1602 | |
| 1603 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
| 1604 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
David S. Miller | 1a028e5 | 2007-04-27 15:21:23 -0700 | [diff] [blame] | 1605 | int end; |
Herbert Xu | 357b40a | 2005-04-19 22:30:14 -0700 | [diff] [blame] | 1606 | |
Ilpo Järvinen | 547b792 | 2008-07-25 21:43:18 -0700 | [diff] [blame] | 1607 | WARN_ON(start > offset + len); |
David S. Miller | 1a028e5 | 2007-04-27 15:21:23 -0700 | [diff] [blame] | 1608 | |
| 1609 | end = start + frag->size; |
Herbert Xu | 357b40a | 2005-04-19 22:30:14 -0700 | [diff] [blame] | 1610 | if ((copy = end - offset) > 0) { |
| 1611 | u8 *vaddr; |
| 1612 | |
| 1613 | if (copy > len) |
| 1614 | copy = len; |
| 1615 | |
| 1616 | vaddr = kmap_skb_frag(frag); |
David S. Miller | 1a028e5 | 2007-04-27 15:21:23 -0700 | [diff] [blame] | 1617 | memcpy(vaddr + frag->page_offset + offset - start, |
| 1618 | from, copy); |
Herbert Xu | 357b40a | 2005-04-19 22:30:14 -0700 | [diff] [blame] | 1619 | kunmap_skb_frag(vaddr); |
| 1620 | |
| 1621 | if ((len -= copy) == 0) |
| 1622 | return 0; |
| 1623 | offset += copy; |
| 1624 | from += copy; |
| 1625 | } |
David S. Miller | 1a028e5 | 2007-04-27 15:21:23 -0700 | [diff] [blame] | 1626 | start = end; |
Herbert Xu | 357b40a | 2005-04-19 22:30:14 -0700 | [diff] [blame] | 1627 | } |
| 1628 | |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 1629 | skb_walk_frags(skb, frag_iter) { |
| 1630 | int end; |
Herbert Xu | 357b40a | 2005-04-19 22:30:14 -0700 | [diff] [blame] | 1631 | |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 1632 | WARN_ON(start > offset + len); |
Herbert Xu | 357b40a | 2005-04-19 22:30:14 -0700 | [diff] [blame] | 1633 | |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 1634 | end = start + frag_iter->len; |
| 1635 | if ((copy = end - offset) > 0) { |
| 1636 | if (copy > len) |
| 1637 | copy = len; |
| 1638 | if (skb_store_bits(frag_iter, offset - start, |
| 1639 | from, copy)) |
| 1640 | goto fault; |
| 1641 | if ((len -= copy) == 0) |
| 1642 | return 0; |
| 1643 | offset += copy; |
| 1644 | from += copy; |
Herbert Xu | 357b40a | 2005-04-19 22:30:14 -0700 | [diff] [blame] | 1645 | } |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 1646 | start = end; |
Herbert Xu | 357b40a | 2005-04-19 22:30:14 -0700 | [diff] [blame] | 1647 | } |
| 1648 | if (!len) |
| 1649 | return 0; |
| 1650 | |
| 1651 | fault: |
| 1652 | return -EFAULT; |
| 1653 | } |
Herbert Xu | 357b40a | 2005-04-19 22:30:14 -0700 | [diff] [blame] | 1654 | EXPORT_SYMBOL(skb_store_bits); |
| 1655 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1656 | /* Checksum skb data. */ |
| 1657 | |
Al Viro | 2bbbc86 | 2006-11-14 21:37:14 -0800 | [diff] [blame] | 1658 | __wsum skb_checksum(const struct sk_buff *skb, int offset, |
| 1659 | int len, __wsum csum) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1660 | { |
David S. Miller | 1a028e5 | 2007-04-27 15:21:23 -0700 | [diff] [blame] | 1661 | int start = skb_headlen(skb); |
| 1662 | int i, copy = start - offset; |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 1663 | struct sk_buff *frag_iter; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1664 | int pos = 0; |
| 1665 | |
| 1666 | /* Checksum header. */ |
| 1667 | if (copy > 0) { |
| 1668 | if (copy > len) |
| 1669 | copy = len; |
| 1670 | csum = csum_partial(skb->data + offset, copy, csum); |
| 1671 | if ((len -= copy) == 0) |
| 1672 | return csum; |
| 1673 | offset += copy; |
| 1674 | pos = copy; |
| 1675 | } |
| 1676 | |
| 1677 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
David S. Miller | 1a028e5 | 2007-04-27 15:21:23 -0700 | [diff] [blame] | 1678 | int end; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1679 | |
Ilpo Järvinen | 547b792 | 2008-07-25 21:43:18 -0700 | [diff] [blame] | 1680 | WARN_ON(start > offset + len); |
David S. Miller | 1a028e5 | 2007-04-27 15:21:23 -0700 | [diff] [blame] | 1681 | |
| 1682 | end = start + skb_shinfo(skb)->frags[i].size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1683 | if ((copy = end - offset) > 0) { |
Al Viro | 44bb936 | 2006-11-14 21:36:14 -0800 | [diff] [blame] | 1684 | __wsum csum2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1685 | u8 *vaddr; |
| 1686 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
| 1687 | |
| 1688 | if (copy > len) |
| 1689 | copy = len; |
| 1690 | vaddr = kmap_skb_frag(frag); |
David S. Miller | 1a028e5 | 2007-04-27 15:21:23 -0700 | [diff] [blame] | 1691 | csum2 = csum_partial(vaddr + frag->page_offset + |
| 1692 | offset - start, copy, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1693 | kunmap_skb_frag(vaddr); |
| 1694 | csum = csum_block_add(csum, csum2, pos); |
| 1695 | if (!(len -= copy)) |
| 1696 | return csum; |
| 1697 | offset += copy; |
| 1698 | pos += copy; |
| 1699 | } |
David S. Miller | 1a028e5 | 2007-04-27 15:21:23 -0700 | [diff] [blame] | 1700 | start = end; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1701 | } |
| 1702 | |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 1703 | skb_walk_frags(skb, frag_iter) { |
| 1704 | int end; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1705 | |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 1706 | WARN_ON(start > offset + len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1707 | |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 1708 | end = start + frag_iter->len; |
| 1709 | if ((copy = end - offset) > 0) { |
| 1710 | __wsum csum2; |
| 1711 | if (copy > len) |
| 1712 | copy = len; |
| 1713 | csum2 = skb_checksum(frag_iter, offset - start, |
| 1714 | copy, 0); |
| 1715 | csum = csum_block_add(csum, csum2, pos); |
| 1716 | if ((len -= copy) == 0) |
| 1717 | return csum; |
| 1718 | offset += copy; |
| 1719 | pos += copy; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1720 | } |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 1721 | start = end; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1722 | } |
Kris Katterjohn | 09a6266 | 2006-01-08 22:24:28 -0800 | [diff] [blame] | 1723 | BUG_ON(len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1724 | |
| 1725 | return csum; |
| 1726 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 1727 | EXPORT_SYMBOL(skb_checksum); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1728 | |
| 1729 | /* Both of above in one bottle. */ |
| 1730 | |
Al Viro | 81d7766 | 2006-11-14 21:37:33 -0800 | [diff] [blame] | 1731 | __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, |
| 1732 | u8 *to, int len, __wsum csum) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1733 | { |
David S. Miller | 1a028e5 | 2007-04-27 15:21:23 -0700 | [diff] [blame] | 1734 | int start = skb_headlen(skb); |
| 1735 | int i, copy = start - offset; |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 1736 | struct sk_buff *frag_iter; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1737 | int pos = 0; |
| 1738 | |
| 1739 | /* Copy header. */ |
| 1740 | if (copy > 0) { |
| 1741 | if (copy > len) |
| 1742 | copy = len; |
| 1743 | csum = csum_partial_copy_nocheck(skb->data + offset, to, |
| 1744 | copy, csum); |
| 1745 | if ((len -= copy) == 0) |
| 1746 | return csum; |
| 1747 | offset += copy; |
| 1748 | to += copy; |
| 1749 | pos = copy; |
| 1750 | } |
| 1751 | |
| 1752 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
David S. Miller | 1a028e5 | 2007-04-27 15:21:23 -0700 | [diff] [blame] | 1753 | int end; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1754 | |
Ilpo Järvinen | 547b792 | 2008-07-25 21:43:18 -0700 | [diff] [blame] | 1755 | WARN_ON(start > offset + len); |
David S. Miller | 1a028e5 | 2007-04-27 15:21:23 -0700 | [diff] [blame] | 1756 | |
| 1757 | end = start + skb_shinfo(skb)->frags[i].size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1758 | if ((copy = end - offset) > 0) { |
Al Viro | 5084205 | 2006-11-14 21:36:34 -0800 | [diff] [blame] | 1759 | __wsum csum2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1760 | u8 *vaddr; |
| 1761 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
| 1762 | |
| 1763 | if (copy > len) |
| 1764 | copy = len; |
| 1765 | vaddr = kmap_skb_frag(frag); |
| 1766 | csum2 = csum_partial_copy_nocheck(vaddr + |
David S. Miller | 1a028e5 | 2007-04-27 15:21:23 -0700 | [diff] [blame] | 1767 | frag->page_offset + |
| 1768 | offset - start, to, |
| 1769 | copy, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1770 | kunmap_skb_frag(vaddr); |
| 1771 | csum = csum_block_add(csum, csum2, pos); |
| 1772 | if (!(len -= copy)) |
| 1773 | return csum; |
| 1774 | offset += copy; |
| 1775 | to += copy; |
| 1776 | pos += copy; |
| 1777 | } |
David S. Miller | 1a028e5 | 2007-04-27 15:21:23 -0700 | [diff] [blame] | 1778 | start = end; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1779 | } |
| 1780 | |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 1781 | skb_walk_frags(skb, frag_iter) { |
| 1782 | __wsum csum2; |
| 1783 | int end; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1784 | |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 1785 | WARN_ON(start > offset + len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1786 | |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 1787 | end = start + frag_iter->len; |
| 1788 | if ((copy = end - offset) > 0) { |
| 1789 | if (copy > len) |
| 1790 | copy = len; |
| 1791 | csum2 = skb_copy_and_csum_bits(frag_iter, |
| 1792 | offset - start, |
| 1793 | to, copy, 0); |
| 1794 | csum = csum_block_add(csum, csum2, pos); |
| 1795 | if ((len -= copy) == 0) |
| 1796 | return csum; |
| 1797 | offset += copy; |
| 1798 | to += copy; |
| 1799 | pos += copy; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1800 | } |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 1801 | start = end; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1802 | } |
Kris Katterjohn | 09a6266 | 2006-01-08 22:24:28 -0800 | [diff] [blame] | 1803 | BUG_ON(len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1804 | return csum; |
| 1805 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 1806 | EXPORT_SYMBOL(skb_copy_and_csum_bits); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1807 | |
| 1808 | void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to) |
| 1809 | { |
Al Viro | d3bc23e | 2006-11-14 21:24:49 -0800 | [diff] [blame] | 1810 | __wsum csum; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1811 | long csstart; |
| 1812 | |
Patrick McHardy | 84fa793 | 2006-08-29 16:44:56 -0700 | [diff] [blame] | 1813 | if (skb->ip_summed == CHECKSUM_PARTIAL) |
Herbert Xu | 663ead3 | 2007-04-09 11:59:07 -0700 | [diff] [blame] | 1814 | csstart = skb->csum_start - skb_headroom(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1815 | else |
| 1816 | csstart = skb_headlen(skb); |
| 1817 | |
Kris Katterjohn | 09a6266 | 2006-01-08 22:24:28 -0800 | [diff] [blame] | 1818 | BUG_ON(csstart > skb_headlen(skb)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1819 | |
Arnaldo Carvalho de Melo | d626f62 | 2007-03-27 18:55:52 -0300 | [diff] [blame] | 1820 | skb_copy_from_linear_data(skb, to, csstart); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1821 | |
| 1822 | csum = 0; |
| 1823 | if (csstart != skb->len) |
| 1824 | csum = skb_copy_and_csum_bits(skb, csstart, to + csstart, |
| 1825 | skb->len - csstart, 0); |
| 1826 | |
Patrick McHardy | 84fa793 | 2006-08-29 16:44:56 -0700 | [diff] [blame] | 1827 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
Al Viro | ff1dcad | 2006-11-20 18:07:29 -0800 | [diff] [blame] | 1828 | long csstuff = csstart + skb->csum_offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1829 | |
Al Viro | d3bc23e | 2006-11-14 21:24:49 -0800 | [diff] [blame] | 1830 | *((__sum16 *)(to + csstuff)) = csum_fold(csum); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1831 | } |
| 1832 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 1833 | EXPORT_SYMBOL(skb_copy_and_csum_dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1834 | |
| 1835 | /** |
| 1836 | * skb_dequeue - remove from the head of the queue |
| 1837 | * @list: list to dequeue from |
| 1838 | * |
| 1839 | * Remove the head of the list. The list lock is taken so the function |
| 1840 | * may be used safely with other locking list functions. The head item is |
| 1841 | * returned or %NULL if the list is empty. |
| 1842 | */ |
| 1843 | |
| 1844 | struct sk_buff *skb_dequeue(struct sk_buff_head *list) |
| 1845 | { |
| 1846 | unsigned long flags; |
| 1847 | struct sk_buff *result; |
| 1848 | |
| 1849 | spin_lock_irqsave(&list->lock, flags); |
| 1850 | result = __skb_dequeue(list); |
| 1851 | spin_unlock_irqrestore(&list->lock, flags); |
| 1852 | return result; |
| 1853 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 1854 | EXPORT_SYMBOL(skb_dequeue); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1855 | |
| 1856 | /** |
| 1857 | * skb_dequeue_tail - remove from the tail of the queue |
| 1858 | * @list: list to dequeue from |
| 1859 | * |
| 1860 | * Remove the tail of the list. The list lock is taken so the function |
| 1861 | * may be used safely with other locking list functions. The tail item is |
| 1862 | * returned or %NULL if the list is empty. |
| 1863 | */ |
| 1864 | struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list) |
| 1865 | { |
| 1866 | unsigned long flags; |
| 1867 | struct sk_buff *result; |
| 1868 | |
| 1869 | spin_lock_irqsave(&list->lock, flags); |
| 1870 | result = __skb_dequeue_tail(list); |
| 1871 | spin_unlock_irqrestore(&list->lock, flags); |
| 1872 | return result; |
| 1873 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 1874 | EXPORT_SYMBOL(skb_dequeue_tail); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1875 | |
| 1876 | /** |
| 1877 | * skb_queue_purge - empty a list |
| 1878 | * @list: list to empty |
| 1879 | * |
| 1880 | * Delete all buffers on an &sk_buff list. Each buffer is removed from |
| 1881 | * the list and one reference dropped. This function takes the list |
| 1882 | * lock and is atomic with respect to other list locking functions. |
| 1883 | */ |
| 1884 | void skb_queue_purge(struct sk_buff_head *list) |
| 1885 | { |
| 1886 | struct sk_buff *skb; |
| 1887 | while ((skb = skb_dequeue(list)) != NULL) |
| 1888 | kfree_skb(skb); |
| 1889 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 1890 | EXPORT_SYMBOL(skb_queue_purge); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1891 | |
| 1892 | /** |
| 1893 | * skb_queue_head - queue a buffer at the list head |
| 1894 | * @list: list to use |
| 1895 | * @newsk: buffer to queue |
| 1896 | * |
| 1897 | * Queue a buffer at the start of the list. This function takes the |
| 1898 | * list lock and can be used safely with other locking &sk_buff functions |
| 1899 | * safely. |
| 1900 | * |
| 1901 | * A buffer cannot be placed on two lists at the same time. |
| 1902 | */ |
| 1903 | void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk) |
| 1904 | { |
| 1905 | unsigned long flags; |
| 1906 | |
| 1907 | spin_lock_irqsave(&list->lock, flags); |
| 1908 | __skb_queue_head(list, newsk); |
| 1909 | spin_unlock_irqrestore(&list->lock, flags); |
| 1910 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 1911 | EXPORT_SYMBOL(skb_queue_head); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1912 | |
| 1913 | /** |
| 1914 | * skb_queue_tail - queue a buffer at the list tail |
| 1915 | * @list: list to use |
| 1916 | * @newsk: buffer to queue |
| 1917 | * |
| 1918 | * Queue a buffer at the tail of the list. This function takes the |
| 1919 | * list lock and can be used safely with other locking &sk_buff functions |
| 1920 | * safely. |
| 1921 | * |
| 1922 | * A buffer cannot be placed on two lists at the same time. |
| 1923 | */ |
| 1924 | void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk) |
| 1925 | { |
| 1926 | unsigned long flags; |
| 1927 | |
| 1928 | spin_lock_irqsave(&list->lock, flags); |
| 1929 | __skb_queue_tail(list, newsk); |
| 1930 | spin_unlock_irqrestore(&list->lock, flags); |
| 1931 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 1932 | EXPORT_SYMBOL(skb_queue_tail); |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 1933 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1934 | /** |
| 1935 | * skb_unlink - remove a buffer from a list |
| 1936 | * @skb: buffer to remove |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 1937 | * @list: list to use |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1938 | * |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 1939 | * Remove a packet from a list. The list locks are taken and this |
| 1940 | * function is atomic with respect to other list locked calls |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1941 | * |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 1942 | * You must know what list the SKB is on. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1943 | */ |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 1944 | void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1945 | { |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 1946 | unsigned long flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1947 | |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 1948 | spin_lock_irqsave(&list->lock, flags); |
| 1949 | __skb_unlink(skb, list); |
| 1950 | spin_unlock_irqrestore(&list->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1951 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 1952 | EXPORT_SYMBOL(skb_unlink); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1953 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1954 | /** |
| 1955 | * skb_append - append a buffer |
| 1956 | * @old: buffer to insert after |
| 1957 | * @newsk: buffer to insert |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 1958 | * @list: list to use |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1959 | * |
| 1960 | * Place a packet after a given packet in a list. The list locks are taken |
| 1961 | * and this function is atomic with respect to other list locked calls. |
| 1962 | * A buffer cannot be placed on two lists at the same time. |
| 1963 | */ |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 1964 | void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1965 | { |
| 1966 | unsigned long flags; |
| 1967 | |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 1968 | spin_lock_irqsave(&list->lock, flags); |
Gerrit Renker | 7de6c03 | 2008-04-14 00:05:09 -0700 | [diff] [blame] | 1969 | __skb_queue_after(list, old, newsk); |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 1970 | spin_unlock_irqrestore(&list->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1971 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 1972 | EXPORT_SYMBOL(skb_append); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1973 | |
| 1974 | /** |
| 1975 | * skb_insert - insert a buffer |
| 1976 | * @old: buffer to insert before |
| 1977 | * @newsk: buffer to insert |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 1978 | * @list: list to use |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1979 | * |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 1980 | * Place a packet before a given packet in a list. The list locks are |
| 1981 | * taken and this function is atomic with respect to other list locked |
| 1982 | * calls. |
| 1983 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1984 | * A buffer cannot be placed on two lists at the same time. |
| 1985 | */ |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 1986 | void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1987 | { |
| 1988 | unsigned long flags; |
| 1989 | |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 1990 | spin_lock_irqsave(&list->lock, flags); |
| 1991 | __skb_insert(newsk, old->prev, old, list); |
| 1992 | spin_unlock_irqrestore(&list->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1993 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 1994 | EXPORT_SYMBOL(skb_insert); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1995 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1996 | static inline void skb_split_inside_header(struct sk_buff *skb, |
| 1997 | struct sk_buff* skb1, |
| 1998 | const u32 len, const int pos) |
| 1999 | { |
| 2000 | int i; |
| 2001 | |
Arnaldo Carvalho de Melo | d626f62 | 2007-03-27 18:55:52 -0300 | [diff] [blame] | 2002 | skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len), |
| 2003 | pos - len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2004 | /* And move data appendix as is. */ |
| 2005 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) |
| 2006 | skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i]; |
| 2007 | |
| 2008 | skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags; |
| 2009 | skb_shinfo(skb)->nr_frags = 0; |
| 2010 | skb1->data_len = skb->data_len; |
| 2011 | skb1->len += skb1->data_len; |
| 2012 | skb->data_len = 0; |
| 2013 | skb->len = len; |
Arnaldo Carvalho de Melo | 27a884d | 2007-04-19 20:29:13 -0700 | [diff] [blame] | 2014 | skb_set_tail_pointer(skb, len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2015 | } |
| 2016 | |
| 2017 | static inline void skb_split_no_header(struct sk_buff *skb, |
| 2018 | struct sk_buff* skb1, |
| 2019 | const u32 len, int pos) |
| 2020 | { |
| 2021 | int i, k = 0; |
| 2022 | const int nfrags = skb_shinfo(skb)->nr_frags; |
| 2023 | |
| 2024 | skb_shinfo(skb)->nr_frags = 0; |
| 2025 | skb1->len = skb1->data_len = skb->len - len; |
| 2026 | skb->len = len; |
| 2027 | skb->data_len = len - pos; |
| 2028 | |
| 2029 | for (i = 0; i < nfrags; i++) { |
| 2030 | int size = skb_shinfo(skb)->frags[i].size; |
| 2031 | |
| 2032 | if (pos + size > len) { |
| 2033 | skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i]; |
| 2034 | |
| 2035 | if (pos < len) { |
| 2036 | /* Split frag. |
| 2037 | * We have two variants in this case: |
| 2038 | * 1. Move all the frag to the second |
| 2039 | * part, if it is possible. F.e. |
| 2040 | * this approach is mandatory for TUX, |
| 2041 | * where splitting is expensive. |
| 2042 | * 2. Split is accurately. We make this. |
| 2043 | */ |
| 2044 | get_page(skb_shinfo(skb)->frags[i].page); |
| 2045 | skb_shinfo(skb1)->frags[0].page_offset += len - pos; |
| 2046 | skb_shinfo(skb1)->frags[0].size -= len - pos; |
| 2047 | skb_shinfo(skb)->frags[i].size = len - pos; |
| 2048 | skb_shinfo(skb)->nr_frags++; |
| 2049 | } |
| 2050 | k++; |
| 2051 | } else |
| 2052 | skb_shinfo(skb)->nr_frags++; |
| 2053 | pos += size; |
| 2054 | } |
| 2055 | skb_shinfo(skb1)->nr_frags = k; |
| 2056 | } |
| 2057 | |
| 2058 | /** |
| 2059 | * skb_split - Split fragmented skb to two parts at length len. |
| 2060 | * @skb: the buffer to split |
| 2061 | * @skb1: the buffer to receive the second part |
| 2062 | * @len: new length for skb |
| 2063 | */ |
| 2064 | void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) |
| 2065 | { |
| 2066 | int pos = skb_headlen(skb); |
| 2067 | |
| 2068 | if (len < pos) /* Split line is inside header. */ |
| 2069 | skb_split_inside_header(skb, skb1, len, pos); |
| 2070 | else /* Second chunk has no header, nothing to copy. */ |
| 2071 | skb_split_no_header(skb, skb1, len, pos); |
| 2072 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 2073 | EXPORT_SYMBOL(skb_split); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2074 | |
Ilpo Järvinen | 9f782db | 2008-11-25 13:57:01 -0800 | [diff] [blame] | 2075 | /* Shifting from/to a cloned skb is a no-go. |
| 2076 | * |
| 2077 | * Caller cannot keep skb_shinfo related pointers past calling here! |
| 2078 | */ |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 2079 | static int skb_prepare_for_shift(struct sk_buff *skb) |
| 2080 | { |
Ilpo Järvinen | 0ace285 | 2008-11-24 21:30:21 -0800 | [diff] [blame] | 2081 | return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 2082 | } |
| 2083 | |
| 2084 | /** |
| 2085 | * skb_shift - Shifts paged data partially from skb to another |
| 2086 | * @tgt: buffer into which tail data gets added |
| 2087 | * @skb: buffer from which the paged data comes from |
| 2088 | * @shiftlen: shift up to this many bytes |
| 2089 | * |
| 2090 | * Attempts to shift up to shiftlen worth of bytes, which may be less than |
| 2091 | * the length of the skb, from tgt to skb. Returns number bytes shifted. |
| 2092 | * It's up to caller to free skb if everything was shifted. |
| 2093 | * |
| 2094 | * If @tgt runs out of frags, the whole operation is aborted. |
| 2095 | * |
| 2096 | * Skb cannot include anything else but paged data while tgt is allowed |
| 2097 | * to have non-paged data as well. |
| 2098 | * |
| 2099 | * TODO: full sized shift could be optimized but that would need |
| 2100 | * specialized skb free'er to handle frags without up-to-date nr_frags. |
| 2101 | */ |
| 2102 | int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen) |
| 2103 | { |
| 2104 | int from, to, merge, todo; |
| 2105 | struct skb_frag_struct *fragfrom, *fragto; |
| 2106 | |
| 2107 | BUG_ON(shiftlen > skb->len); |
| 2108 | BUG_ON(skb_headlen(skb)); /* Would corrupt stream */ |
| 2109 | |
| 2110 | todo = shiftlen; |
| 2111 | from = 0; |
| 2112 | to = skb_shinfo(tgt)->nr_frags; |
| 2113 | fragfrom = &skb_shinfo(skb)->frags[from]; |
| 2114 | |
| 2115 | /* Actual merge is delayed until the point when we know we can |
| 2116 | * commit all, so that we don't have to undo partial changes |
| 2117 | */ |
| 2118 | if (!to || |
| 2119 | !skb_can_coalesce(tgt, to, fragfrom->page, fragfrom->page_offset)) { |
| 2120 | merge = -1; |
| 2121 | } else { |
| 2122 | merge = to - 1; |
| 2123 | |
| 2124 | todo -= fragfrom->size; |
| 2125 | if (todo < 0) { |
| 2126 | if (skb_prepare_for_shift(skb) || |
| 2127 | skb_prepare_for_shift(tgt)) |
| 2128 | return 0; |
| 2129 | |
Ilpo Järvinen | 9f782db | 2008-11-25 13:57:01 -0800 | [diff] [blame] | 2130 | /* All previous frag pointers might be stale! */ |
| 2131 | fragfrom = &skb_shinfo(skb)->frags[from]; |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 2132 | fragto = &skb_shinfo(tgt)->frags[merge]; |
| 2133 | |
| 2134 | fragto->size += shiftlen; |
| 2135 | fragfrom->size -= shiftlen; |
| 2136 | fragfrom->page_offset += shiftlen; |
| 2137 | |
| 2138 | goto onlymerged; |
| 2139 | } |
| 2140 | |
| 2141 | from++; |
| 2142 | } |
| 2143 | |
| 2144 | /* Skip full, not-fitting skb to avoid expensive operations */ |
| 2145 | if ((shiftlen == skb->len) && |
| 2146 | (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to)) |
| 2147 | return 0; |
| 2148 | |
| 2149 | if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt)) |
| 2150 | return 0; |
| 2151 | |
| 2152 | while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) { |
| 2153 | if (to == MAX_SKB_FRAGS) |
| 2154 | return 0; |
| 2155 | |
| 2156 | fragfrom = &skb_shinfo(skb)->frags[from]; |
| 2157 | fragto = &skb_shinfo(tgt)->frags[to]; |
| 2158 | |
| 2159 | if (todo >= fragfrom->size) { |
| 2160 | *fragto = *fragfrom; |
| 2161 | todo -= fragfrom->size; |
| 2162 | from++; |
| 2163 | to++; |
| 2164 | |
| 2165 | } else { |
| 2166 | get_page(fragfrom->page); |
| 2167 | fragto->page = fragfrom->page; |
| 2168 | fragto->page_offset = fragfrom->page_offset; |
| 2169 | fragto->size = todo; |
| 2170 | |
| 2171 | fragfrom->page_offset += todo; |
| 2172 | fragfrom->size -= todo; |
| 2173 | todo = 0; |
| 2174 | |
| 2175 | to++; |
| 2176 | break; |
| 2177 | } |
| 2178 | } |
| 2179 | |
| 2180 | /* Ready to "commit" this state change to tgt */ |
| 2181 | skb_shinfo(tgt)->nr_frags = to; |
| 2182 | |
| 2183 | if (merge >= 0) { |
| 2184 | fragfrom = &skb_shinfo(skb)->frags[0]; |
| 2185 | fragto = &skb_shinfo(tgt)->frags[merge]; |
| 2186 | |
| 2187 | fragto->size += fragfrom->size; |
| 2188 | put_page(fragfrom->page); |
| 2189 | } |
| 2190 | |
| 2191 | /* Reposition in the original skb */ |
| 2192 | to = 0; |
| 2193 | while (from < skb_shinfo(skb)->nr_frags) |
| 2194 | skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++]; |
| 2195 | skb_shinfo(skb)->nr_frags = to; |
| 2196 | |
| 2197 | BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags); |
| 2198 | |
| 2199 | onlymerged: |
| 2200 | /* Most likely the tgt won't ever need its checksum anymore, skb on |
| 2201 | * the other hand might need it if it needs to be resent |
| 2202 | */ |
| 2203 | tgt->ip_summed = CHECKSUM_PARTIAL; |
| 2204 | skb->ip_summed = CHECKSUM_PARTIAL; |
| 2205 | |
| 2206 | /* Yak, is it really working this way? Some helper please? */ |
| 2207 | skb->len -= shiftlen; |
| 2208 | skb->data_len -= shiftlen; |
| 2209 | skb->truesize -= shiftlen; |
| 2210 | tgt->len += shiftlen; |
| 2211 | tgt->data_len += shiftlen; |
| 2212 | tgt->truesize += shiftlen; |
| 2213 | |
| 2214 | return shiftlen; |
| 2215 | } |
| 2216 | |
Thomas Graf | 677e90e | 2005-06-23 20:59:51 -0700 | [diff] [blame] | 2217 | /** |
| 2218 | * skb_prepare_seq_read - Prepare a sequential read of skb data |
| 2219 | * @skb: the buffer to read |
| 2220 | * @from: lower offset of data to be read |
| 2221 | * @to: upper offset of data to be read |
| 2222 | * @st: state variable |
| 2223 | * |
| 2224 | * Initializes the specified state variable. Must be called before |
| 2225 | * invoking skb_seq_read() for the first time. |
| 2226 | */ |
| 2227 | void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, |
| 2228 | unsigned int to, struct skb_seq_state *st) |
| 2229 | { |
| 2230 | st->lower_offset = from; |
| 2231 | st->upper_offset = to; |
| 2232 | st->root_skb = st->cur_skb = skb; |
| 2233 | st->frag_idx = st->stepped_offset = 0; |
| 2234 | st->frag_data = NULL; |
| 2235 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 2236 | EXPORT_SYMBOL(skb_prepare_seq_read); |
Thomas Graf | 677e90e | 2005-06-23 20:59:51 -0700 | [diff] [blame] | 2237 | |
| 2238 | /** |
| 2239 | * skb_seq_read - Sequentially read skb data |
| 2240 | * @consumed: number of bytes consumed by the caller so far |
| 2241 | * @data: destination pointer for data to be returned |
| 2242 | * @st: state variable |
| 2243 | * |
| 2244 | * Reads a block of skb data at &consumed relative to the |
| 2245 | * lower offset specified to skb_prepare_seq_read(). Assigns |
| 2246 | * the head of the data block to &data and returns the length |
| 2247 | * of the block or 0 if the end of the skb data or the upper |
| 2248 | * offset has been reached. |
| 2249 | * |
| 2250 | * The caller is not required to consume all of the data |
| 2251 | * returned, i.e. &consumed is typically set to the number |
| 2252 | * of bytes already consumed and the next call to |
| 2253 | * skb_seq_read() will return the remaining part of the block. |
| 2254 | * |
Randy Dunlap | bc2cda1 | 2008-02-13 15:03:25 -0800 | [diff] [blame] | 2255 | * Note 1: The size of each block of data returned can be arbitary, |
Thomas Graf | 677e90e | 2005-06-23 20:59:51 -0700 | [diff] [blame] | 2256 | * this limitation is the cost for zerocopy seqeuental |
| 2257 | * reads of potentially non linear data. |
| 2258 | * |
Randy Dunlap | bc2cda1 | 2008-02-13 15:03:25 -0800 | [diff] [blame] | 2259 | * Note 2: Fragment lists within fragments are not implemented |
Thomas Graf | 677e90e | 2005-06-23 20:59:51 -0700 | [diff] [blame] | 2260 | * at the moment, state->root_skb could be replaced with |
| 2261 | * a stack for this purpose. |
| 2262 | */ |
| 2263 | unsigned int skb_seq_read(unsigned int consumed, const u8 **data, |
| 2264 | struct skb_seq_state *st) |
| 2265 | { |
| 2266 | unsigned int block_limit, abs_offset = consumed + st->lower_offset; |
| 2267 | skb_frag_t *frag; |
| 2268 | |
| 2269 | if (unlikely(abs_offset >= st->upper_offset)) |
| 2270 | return 0; |
| 2271 | |
| 2272 | next_skb: |
Herbert Xu | 95e3b24 | 2009-01-29 16:07:52 -0800 | [diff] [blame] | 2273 | block_limit = skb_headlen(st->cur_skb) + st->stepped_offset; |
Thomas Graf | 677e90e | 2005-06-23 20:59:51 -0700 | [diff] [blame] | 2274 | |
Thomas Chenault | 995b337 | 2009-05-18 21:43:27 -0700 | [diff] [blame] | 2275 | if (abs_offset < block_limit && !st->frag_data) { |
Herbert Xu | 95e3b24 | 2009-01-29 16:07:52 -0800 | [diff] [blame] | 2276 | *data = st->cur_skb->data + (abs_offset - st->stepped_offset); |
Thomas Graf | 677e90e | 2005-06-23 20:59:51 -0700 | [diff] [blame] | 2277 | return block_limit - abs_offset; |
| 2278 | } |
| 2279 | |
| 2280 | if (st->frag_idx == 0 && !st->frag_data) |
| 2281 | st->stepped_offset += skb_headlen(st->cur_skb); |
| 2282 | |
| 2283 | while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) { |
| 2284 | frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx]; |
| 2285 | block_limit = frag->size + st->stepped_offset; |
| 2286 | |
| 2287 | if (abs_offset < block_limit) { |
| 2288 | if (!st->frag_data) |
| 2289 | st->frag_data = kmap_skb_frag(frag); |
| 2290 | |
| 2291 | *data = (u8 *) st->frag_data + frag->page_offset + |
| 2292 | (abs_offset - st->stepped_offset); |
| 2293 | |
| 2294 | return block_limit - abs_offset; |
| 2295 | } |
| 2296 | |
| 2297 | if (st->frag_data) { |
| 2298 | kunmap_skb_frag(st->frag_data); |
| 2299 | st->frag_data = NULL; |
| 2300 | } |
| 2301 | |
| 2302 | st->frag_idx++; |
| 2303 | st->stepped_offset += frag->size; |
| 2304 | } |
| 2305 | |
Olaf Kirch | 5b5a60d | 2007-06-23 23:11:52 -0700 | [diff] [blame] | 2306 | if (st->frag_data) { |
| 2307 | kunmap_skb_frag(st->frag_data); |
| 2308 | st->frag_data = NULL; |
| 2309 | } |
| 2310 | |
David S. Miller | 21dc330 | 2010-08-23 00:13:46 -0700 | [diff] [blame] | 2311 | if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) { |
Shyam Iyer | 71b3346 | 2009-01-29 16:12:42 -0800 | [diff] [blame] | 2312 | st->cur_skb = skb_shinfo(st->root_skb)->frag_list; |
Thomas Graf | 677e90e | 2005-06-23 20:59:51 -0700 | [diff] [blame] | 2313 | st->frag_idx = 0; |
| 2314 | goto next_skb; |
Shyam Iyer | 71b3346 | 2009-01-29 16:12:42 -0800 | [diff] [blame] | 2315 | } else if (st->cur_skb->next) { |
| 2316 | st->cur_skb = st->cur_skb->next; |
Herbert Xu | 95e3b24 | 2009-01-29 16:07:52 -0800 | [diff] [blame] | 2317 | st->frag_idx = 0; |
Thomas Graf | 677e90e | 2005-06-23 20:59:51 -0700 | [diff] [blame] | 2318 | goto next_skb; |
| 2319 | } |
| 2320 | |
| 2321 | return 0; |
| 2322 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 2323 | EXPORT_SYMBOL(skb_seq_read); |
Thomas Graf | 677e90e | 2005-06-23 20:59:51 -0700 | [diff] [blame] | 2324 | |
| 2325 | /** |
| 2326 | * skb_abort_seq_read - Abort a sequential read of skb data |
| 2327 | * @st: state variable |
| 2328 | * |
| 2329 | * Must be called if skb_seq_read() was not called until it |
| 2330 | * returned 0. |
| 2331 | */ |
| 2332 | void skb_abort_seq_read(struct skb_seq_state *st) |
| 2333 | { |
| 2334 | if (st->frag_data) |
| 2335 | kunmap_skb_frag(st->frag_data); |
| 2336 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 2337 | EXPORT_SYMBOL(skb_abort_seq_read); |
Thomas Graf | 677e90e | 2005-06-23 20:59:51 -0700 | [diff] [blame] | 2338 | |
Thomas Graf | 3fc7e8a | 2005-06-23 21:00:17 -0700 | [diff] [blame] | 2339 | #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb)) |
| 2340 | |
| 2341 | static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text, |
| 2342 | struct ts_config *conf, |
| 2343 | struct ts_state *state) |
| 2344 | { |
| 2345 | return skb_seq_read(offset, text, TS_SKB_CB(state)); |
| 2346 | } |
| 2347 | |
| 2348 | static void skb_ts_finish(struct ts_config *conf, struct ts_state *state) |
| 2349 | { |
| 2350 | skb_abort_seq_read(TS_SKB_CB(state)); |
| 2351 | } |
| 2352 | |
| 2353 | /** |
| 2354 | * skb_find_text - Find a text pattern in skb data |
| 2355 | * @skb: the buffer to look in |
| 2356 | * @from: search offset |
| 2357 | * @to: search limit |
| 2358 | * @config: textsearch configuration |
| 2359 | * @state: uninitialized textsearch state variable |
| 2360 | * |
| 2361 | * Finds a pattern in the skb data according to the specified |
| 2362 | * textsearch configuration. Use textsearch_next() to retrieve |
| 2363 | * subsequent occurrences of the pattern. Returns the offset |
| 2364 | * to the first occurrence or UINT_MAX if no match was found. |
| 2365 | */ |
| 2366 | unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, |
| 2367 | unsigned int to, struct ts_config *config, |
| 2368 | struct ts_state *state) |
| 2369 | { |
Phil Oester | f72b948 | 2006-06-26 00:00:57 -0700 | [diff] [blame] | 2370 | unsigned int ret; |
| 2371 | |
Thomas Graf | 3fc7e8a | 2005-06-23 21:00:17 -0700 | [diff] [blame] | 2372 | config->get_next_block = skb_ts_get_next_block; |
| 2373 | config->finish = skb_ts_finish; |
| 2374 | |
| 2375 | skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state)); |
| 2376 | |
Phil Oester | f72b948 | 2006-06-26 00:00:57 -0700 | [diff] [blame] | 2377 | ret = textsearch_find(config, state); |
| 2378 | return (ret <= to - from ? ret : UINT_MAX); |
Thomas Graf | 3fc7e8a | 2005-06-23 21:00:17 -0700 | [diff] [blame] | 2379 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 2380 | EXPORT_SYMBOL(skb_find_text); |
Thomas Graf | 3fc7e8a | 2005-06-23 21:00:17 -0700 | [diff] [blame] | 2381 | |
Ananda Raju | e89e9cf | 2005-10-18 15:46:41 -0700 | [diff] [blame] | 2382 | /** |
| 2383 | * skb_append_datato_frags: - append the user data to a skb |
| 2384 | * @sk: sock structure |
| 2385 | * @skb: skb structure to be appened with user data. |
| 2386 | * @getfrag: call back function to be used for getting the user data |
| 2387 | * @from: pointer to user message iov |
| 2388 | * @length: length of the iov message |
| 2389 | * |
| 2390 | * Description: This procedure append the user data in the fragment part |
| 2391 | * of the skb if any page alloc fails user this procedure returns -ENOMEM |
| 2392 | */ |
| 2393 | int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, |
Martin Waitz | dab9630 | 2005-12-05 13:40:12 -0800 | [diff] [blame] | 2394 | int (*getfrag)(void *from, char *to, int offset, |
Ananda Raju | e89e9cf | 2005-10-18 15:46:41 -0700 | [diff] [blame] | 2395 | int len, int odd, struct sk_buff *skb), |
| 2396 | void *from, int length) |
| 2397 | { |
| 2398 | int frg_cnt = 0; |
| 2399 | skb_frag_t *frag = NULL; |
| 2400 | struct page *page = NULL; |
| 2401 | int copy, left; |
| 2402 | int offset = 0; |
| 2403 | int ret; |
| 2404 | |
| 2405 | do { |
| 2406 | /* Return error if we don't have space for new frag */ |
| 2407 | frg_cnt = skb_shinfo(skb)->nr_frags; |
| 2408 | if (frg_cnt >= MAX_SKB_FRAGS) |
| 2409 | return -EFAULT; |
| 2410 | |
| 2411 | /* allocate a new page for next frag */ |
| 2412 | page = alloc_pages(sk->sk_allocation, 0); |
| 2413 | |
| 2414 | /* If alloc_page fails just return failure and caller will |
| 2415 | * free previous allocated pages by doing kfree_skb() |
| 2416 | */ |
| 2417 | if (page == NULL) |
| 2418 | return -ENOMEM; |
| 2419 | |
| 2420 | /* initialize the next frag */ |
| 2421 | sk->sk_sndmsg_page = page; |
| 2422 | sk->sk_sndmsg_off = 0; |
| 2423 | skb_fill_page_desc(skb, frg_cnt, page, 0, 0); |
| 2424 | skb->truesize += PAGE_SIZE; |
| 2425 | atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc); |
| 2426 | |
| 2427 | /* get the new initialized frag */ |
| 2428 | frg_cnt = skb_shinfo(skb)->nr_frags; |
| 2429 | frag = &skb_shinfo(skb)->frags[frg_cnt - 1]; |
| 2430 | |
| 2431 | /* copy the user data to page */ |
| 2432 | left = PAGE_SIZE - frag->page_offset; |
| 2433 | copy = (length > left)? left : length; |
| 2434 | |
| 2435 | ret = getfrag(from, (page_address(frag->page) + |
| 2436 | frag->page_offset + frag->size), |
| 2437 | offset, copy, 0, skb); |
| 2438 | if (ret < 0) |
| 2439 | return -EFAULT; |
| 2440 | |
| 2441 | /* copy was successful so update the size parameters */ |
| 2442 | sk->sk_sndmsg_off += copy; |
| 2443 | frag->size += copy; |
| 2444 | skb->len += copy; |
| 2445 | skb->data_len += copy; |
| 2446 | offset += copy; |
| 2447 | length -= copy; |
| 2448 | |
| 2449 | } while (length > 0); |
| 2450 | |
| 2451 | return 0; |
| 2452 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 2453 | EXPORT_SYMBOL(skb_append_datato_frags); |
Ananda Raju | e89e9cf | 2005-10-18 15:46:41 -0700 | [diff] [blame] | 2454 | |
Herbert Xu | cbb042f | 2006-03-20 22:43:56 -0800 | [diff] [blame] | 2455 | /** |
| 2456 | * skb_pull_rcsum - pull skb and update receive checksum |
| 2457 | * @skb: buffer to update |
Herbert Xu | cbb042f | 2006-03-20 22:43:56 -0800 | [diff] [blame] | 2458 | * @len: length of data pulled |
| 2459 | * |
| 2460 | * This function performs an skb_pull on the packet and updates |
Urs Thuermann | fee54fa | 2008-02-12 22:03:25 -0800 | [diff] [blame] | 2461 | * the CHECKSUM_COMPLETE checksum. It should be used on |
Patrick McHardy | 84fa793 | 2006-08-29 16:44:56 -0700 | [diff] [blame] | 2462 | * receive path processing instead of skb_pull unless you know |
| 2463 | * that the checksum difference is zero (e.g., a valid IP header) |
| 2464 | * or you are setting ip_summed to CHECKSUM_NONE. |
Herbert Xu | cbb042f | 2006-03-20 22:43:56 -0800 | [diff] [blame] | 2465 | */ |
| 2466 | unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len) |
| 2467 | { |
| 2468 | BUG_ON(len > skb->len); |
| 2469 | skb->len -= len; |
| 2470 | BUG_ON(skb->len < skb->data_len); |
| 2471 | skb_postpull_rcsum(skb, skb->data, len); |
| 2472 | return skb->data += len; |
| 2473 | } |
Arnaldo Carvalho de Melo | f94691a | 2006-03-20 22:47:55 -0800 | [diff] [blame] | 2474 | EXPORT_SYMBOL_GPL(skb_pull_rcsum); |
| 2475 | |
Herbert Xu | f4c50d9 | 2006-06-22 03:02:40 -0700 | [diff] [blame] | 2476 | /** |
| 2477 | * skb_segment - Perform protocol segmentation on skb. |
| 2478 | * @skb: buffer to segment |
Herbert Xu | 576a30e | 2006-06-27 13:22:38 -0700 | [diff] [blame] | 2479 | * @features: features for the output path (see dev->features) |
Herbert Xu | f4c50d9 | 2006-06-22 03:02:40 -0700 | [diff] [blame] | 2480 | * |
| 2481 | * This function performs segmentation on the given skb. It returns |
Ben Hutchings | 4c821d7 | 2008-04-13 21:52:48 -0700 | [diff] [blame] | 2482 | * a pointer to the first in a list of new skbs for the segments. |
| 2483 | * In case of error it returns ERR_PTR(err). |
Herbert Xu | f4c50d9 | 2006-06-22 03:02:40 -0700 | [diff] [blame] | 2484 | */ |
Herbert Xu | 576a30e | 2006-06-27 13:22:38 -0700 | [diff] [blame] | 2485 | struct sk_buff *skb_segment(struct sk_buff *skb, int features) |
Herbert Xu | f4c50d9 | 2006-06-22 03:02:40 -0700 | [diff] [blame] | 2486 | { |
| 2487 | struct sk_buff *segs = NULL; |
| 2488 | struct sk_buff *tail = NULL; |
Herbert Xu | 89319d38 | 2008-12-15 23:26:06 -0800 | [diff] [blame] | 2489 | struct sk_buff *fskb = skb_shinfo(skb)->frag_list; |
Herbert Xu | f4c50d9 | 2006-06-22 03:02:40 -0700 | [diff] [blame] | 2490 | unsigned int mss = skb_shinfo(skb)->gso_size; |
Arnaldo Carvalho de Melo | 98e399f | 2007-03-19 15:33:04 -0700 | [diff] [blame] | 2491 | unsigned int doffset = skb->data - skb_mac_header(skb); |
Herbert Xu | f4c50d9 | 2006-06-22 03:02:40 -0700 | [diff] [blame] | 2492 | unsigned int offset = doffset; |
| 2493 | unsigned int headroom; |
| 2494 | unsigned int len; |
Herbert Xu | 576a30e | 2006-06-27 13:22:38 -0700 | [diff] [blame] | 2495 | int sg = features & NETIF_F_SG; |
Herbert Xu | f4c50d9 | 2006-06-22 03:02:40 -0700 | [diff] [blame] | 2496 | int nfrags = skb_shinfo(skb)->nr_frags; |
| 2497 | int err = -ENOMEM; |
| 2498 | int i = 0; |
| 2499 | int pos; |
| 2500 | |
| 2501 | __skb_push(skb, doffset); |
| 2502 | headroom = skb_headroom(skb); |
| 2503 | pos = skb_headlen(skb); |
| 2504 | |
| 2505 | do { |
| 2506 | struct sk_buff *nskb; |
| 2507 | skb_frag_t *frag; |
Herbert Xu | c8884ed | 2006-10-29 15:59:41 -0800 | [diff] [blame] | 2508 | int hsize; |
Herbert Xu | f4c50d9 | 2006-06-22 03:02:40 -0700 | [diff] [blame] | 2509 | int size; |
| 2510 | |
| 2511 | len = skb->len - offset; |
| 2512 | if (len > mss) |
| 2513 | len = mss; |
| 2514 | |
| 2515 | hsize = skb_headlen(skb) - offset; |
| 2516 | if (hsize < 0) |
| 2517 | hsize = 0; |
Herbert Xu | c8884ed | 2006-10-29 15:59:41 -0800 | [diff] [blame] | 2518 | if (hsize > len || !sg) |
| 2519 | hsize = len; |
Herbert Xu | f4c50d9 | 2006-06-22 03:02:40 -0700 | [diff] [blame] | 2520 | |
Herbert Xu | 89319d38 | 2008-12-15 23:26:06 -0800 | [diff] [blame] | 2521 | if (!hsize && i >= nfrags) { |
| 2522 | BUG_ON(fskb->len != len); |
| 2523 | |
| 2524 | pos += len; |
| 2525 | nskb = skb_clone(fskb, GFP_ATOMIC); |
| 2526 | fskb = fskb->next; |
| 2527 | |
| 2528 | if (unlikely(!nskb)) |
| 2529 | goto err; |
| 2530 | |
| 2531 | hsize = skb_end_pointer(nskb) - nskb->head; |
| 2532 | if (skb_cow_head(nskb, doffset + headroom)) { |
| 2533 | kfree_skb(nskb); |
| 2534 | goto err; |
| 2535 | } |
| 2536 | |
| 2537 | nskb->truesize += skb_end_pointer(nskb) - nskb->head - |
| 2538 | hsize; |
| 2539 | skb_release_head_state(nskb); |
| 2540 | __skb_push(nskb, doffset); |
| 2541 | } else { |
| 2542 | nskb = alloc_skb(hsize + doffset + headroom, |
| 2543 | GFP_ATOMIC); |
| 2544 | |
| 2545 | if (unlikely(!nskb)) |
| 2546 | goto err; |
| 2547 | |
| 2548 | skb_reserve(nskb, headroom); |
| 2549 | __skb_put(nskb, doffset); |
| 2550 | } |
Herbert Xu | f4c50d9 | 2006-06-22 03:02:40 -0700 | [diff] [blame] | 2551 | |
| 2552 | if (segs) |
| 2553 | tail->next = nskb; |
| 2554 | else |
| 2555 | segs = nskb; |
| 2556 | tail = nskb; |
| 2557 | |
Herbert Xu | 6f85a12 | 2008-08-15 14:55:02 -0700 | [diff] [blame] | 2558 | __copy_skb_header(nskb, skb); |
Herbert Xu | f4c50d9 | 2006-06-22 03:02:40 -0700 | [diff] [blame] | 2559 | nskb->mac_len = skb->mac_len; |
| 2560 | |
Eric Dumazet | 3d3be43 | 2010-09-01 00:50:51 +0000 | [diff] [blame] | 2561 | /* nskb and skb might have different headroom */ |
| 2562 | if (nskb->ip_summed == CHECKSUM_PARTIAL) |
| 2563 | nskb->csum_start += skb_headroom(nskb) - headroom; |
| 2564 | |
Arnaldo Carvalho de Melo | 459a98e | 2007-03-19 15:30:44 -0700 | [diff] [blame] | 2565 | skb_reset_mac_header(nskb); |
Arnaldo Carvalho de Melo | ddc7b8e | 2007-03-15 21:42:27 -0300 | [diff] [blame] | 2566 | skb_set_network_header(nskb, skb->mac_len); |
Arnaldo Carvalho de Melo | b0e380b | 2007-04-10 21:21:55 -0700 | [diff] [blame] | 2567 | nskb->transport_header = (nskb->network_header + |
| 2568 | skb_network_header_len(skb)); |
Herbert Xu | 89319d38 | 2008-12-15 23:26:06 -0800 | [diff] [blame] | 2569 | skb_copy_from_linear_data(skb, nskb->data, doffset); |
| 2570 | |
Herbert Xu | 2f18185 | 2009-03-28 23:39:18 -0700 | [diff] [blame] | 2571 | if (fskb != skb_shinfo(skb)->frag_list) |
Herbert Xu | 89319d38 | 2008-12-15 23:26:06 -0800 | [diff] [blame] | 2572 | continue; |
| 2573 | |
Herbert Xu | f4c50d9 | 2006-06-22 03:02:40 -0700 | [diff] [blame] | 2574 | if (!sg) { |
Herbert Xu | 6f85a12 | 2008-08-15 14:55:02 -0700 | [diff] [blame] | 2575 | nskb->ip_summed = CHECKSUM_NONE; |
Herbert Xu | f4c50d9 | 2006-06-22 03:02:40 -0700 | [diff] [blame] | 2576 | nskb->csum = skb_copy_and_csum_bits(skb, offset, |
| 2577 | skb_put(nskb, len), |
| 2578 | len, 0); |
| 2579 | continue; |
| 2580 | } |
| 2581 | |
| 2582 | frag = skb_shinfo(nskb)->frags; |
Herbert Xu | f4c50d9 | 2006-06-22 03:02:40 -0700 | [diff] [blame] | 2583 | |
Arnaldo Carvalho de Melo | d626f62 | 2007-03-27 18:55:52 -0300 | [diff] [blame] | 2584 | skb_copy_from_linear_data_offset(skb, offset, |
| 2585 | skb_put(nskb, hsize), hsize); |
Herbert Xu | f4c50d9 | 2006-06-22 03:02:40 -0700 | [diff] [blame] | 2586 | |
Herbert Xu | 89319d38 | 2008-12-15 23:26:06 -0800 | [diff] [blame] | 2587 | while (pos < offset + len && i < nfrags) { |
Herbert Xu | f4c50d9 | 2006-06-22 03:02:40 -0700 | [diff] [blame] | 2588 | *frag = skb_shinfo(skb)->frags[i]; |
| 2589 | get_page(frag->page); |
| 2590 | size = frag->size; |
| 2591 | |
| 2592 | if (pos < offset) { |
| 2593 | frag->page_offset += offset - pos; |
| 2594 | frag->size -= offset - pos; |
| 2595 | } |
| 2596 | |
Herbert Xu | 89319d38 | 2008-12-15 23:26:06 -0800 | [diff] [blame] | 2597 | skb_shinfo(nskb)->nr_frags++; |
Herbert Xu | f4c50d9 | 2006-06-22 03:02:40 -0700 | [diff] [blame] | 2598 | |
| 2599 | if (pos + size <= offset + len) { |
| 2600 | i++; |
| 2601 | pos += size; |
| 2602 | } else { |
| 2603 | frag->size -= pos + size - (offset + len); |
Herbert Xu | 89319d38 | 2008-12-15 23:26:06 -0800 | [diff] [blame] | 2604 | goto skip_fraglist; |
Herbert Xu | f4c50d9 | 2006-06-22 03:02:40 -0700 | [diff] [blame] | 2605 | } |
| 2606 | |
| 2607 | frag++; |
| 2608 | } |
| 2609 | |
Herbert Xu | 89319d38 | 2008-12-15 23:26:06 -0800 | [diff] [blame] | 2610 | if (pos < offset + len) { |
| 2611 | struct sk_buff *fskb2 = fskb; |
| 2612 | |
| 2613 | BUG_ON(pos + fskb->len != offset + len); |
| 2614 | |
| 2615 | pos += fskb->len; |
| 2616 | fskb = fskb->next; |
| 2617 | |
| 2618 | if (fskb2->next) { |
| 2619 | fskb2 = skb_clone(fskb2, GFP_ATOMIC); |
| 2620 | if (!fskb2) |
| 2621 | goto err; |
| 2622 | } else |
| 2623 | skb_get(fskb2); |
| 2624 | |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 2625 | SKB_FRAG_ASSERT(nskb); |
Herbert Xu | 89319d38 | 2008-12-15 23:26:06 -0800 | [diff] [blame] | 2626 | skb_shinfo(nskb)->frag_list = fskb2; |
| 2627 | } |
| 2628 | |
| 2629 | skip_fraglist: |
Herbert Xu | f4c50d9 | 2006-06-22 03:02:40 -0700 | [diff] [blame] | 2630 | nskb->data_len = len - hsize; |
| 2631 | nskb->len += nskb->data_len; |
| 2632 | nskb->truesize += nskb->data_len; |
| 2633 | } while ((offset += len) < skb->len); |
| 2634 | |
| 2635 | return segs; |
| 2636 | |
| 2637 | err: |
| 2638 | while ((skb = segs)) { |
| 2639 | segs = skb->next; |
Patrick McHardy | b08d584 | 2007-02-27 09:57:37 -0800 | [diff] [blame] | 2640 | kfree_skb(skb); |
Herbert Xu | f4c50d9 | 2006-06-22 03:02:40 -0700 | [diff] [blame] | 2641 | } |
| 2642 | return ERR_PTR(err); |
| 2643 | } |
Herbert Xu | f4c50d9 | 2006-06-22 03:02:40 -0700 | [diff] [blame] | 2644 | EXPORT_SYMBOL_GPL(skb_segment); |
| 2645 | |
Herbert Xu | 71d93b3 | 2008-12-15 23:42:33 -0800 | [diff] [blame] | 2646 | int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb) |
| 2647 | { |
| 2648 | struct sk_buff *p = *head; |
| 2649 | struct sk_buff *nskb; |
Herbert Xu | 9aaa156 | 2009-05-26 18:50:33 +0000 | [diff] [blame] | 2650 | struct skb_shared_info *skbinfo = skb_shinfo(skb); |
| 2651 | struct skb_shared_info *pinfo = skb_shinfo(p); |
Herbert Xu | 71d93b3 | 2008-12-15 23:42:33 -0800 | [diff] [blame] | 2652 | unsigned int headroom; |
Herbert Xu | 8691173 | 2009-01-29 14:19:50 +0000 | [diff] [blame] | 2653 | unsigned int len = skb_gro_len(skb); |
Herbert Xu | 67147ba | 2009-05-26 18:50:22 +0000 | [diff] [blame] | 2654 | unsigned int offset = skb_gro_offset(skb); |
| 2655 | unsigned int headlen = skb_headlen(skb); |
Herbert Xu | 71d93b3 | 2008-12-15 23:42:33 -0800 | [diff] [blame] | 2656 | |
Herbert Xu | 8691173 | 2009-01-29 14:19:50 +0000 | [diff] [blame] | 2657 | if (p->len + len >= 65536) |
Herbert Xu | 71d93b3 | 2008-12-15 23:42:33 -0800 | [diff] [blame] | 2658 | return -E2BIG; |
| 2659 | |
Herbert Xu | 9aaa156 | 2009-05-26 18:50:33 +0000 | [diff] [blame] | 2660 | if (pinfo->frag_list) |
Herbert Xu | 71d93b3 | 2008-12-15 23:42:33 -0800 | [diff] [blame] | 2661 | goto merge; |
Herbert Xu | 67147ba | 2009-05-26 18:50:22 +0000 | [diff] [blame] | 2662 | else if (headlen <= offset) { |
Herbert Xu | 42da699 | 2009-05-26 18:50:19 +0000 | [diff] [blame] | 2663 | skb_frag_t *frag; |
Herbert Xu | 66e92fc | 2009-05-26 18:50:32 +0000 | [diff] [blame] | 2664 | skb_frag_t *frag2; |
Herbert Xu | 9aaa156 | 2009-05-26 18:50:33 +0000 | [diff] [blame] | 2665 | int i = skbinfo->nr_frags; |
| 2666 | int nr_frags = pinfo->nr_frags + i; |
Herbert Xu | 42da699 | 2009-05-26 18:50:19 +0000 | [diff] [blame] | 2667 | |
Herbert Xu | 66e92fc | 2009-05-26 18:50:32 +0000 | [diff] [blame] | 2668 | offset -= headlen; |
| 2669 | |
| 2670 | if (nr_frags > MAX_SKB_FRAGS) |
Herbert Xu | 81705ad | 2009-01-29 14:19:51 +0000 | [diff] [blame] | 2671 | return -E2BIG; |
| 2672 | |
Herbert Xu | 9aaa156 | 2009-05-26 18:50:33 +0000 | [diff] [blame] | 2673 | pinfo->nr_frags = nr_frags; |
| 2674 | skbinfo->nr_frags = 0; |
Herbert Xu | f557206 | 2009-01-14 20:40:03 -0800 | [diff] [blame] | 2675 | |
Herbert Xu | 9aaa156 | 2009-05-26 18:50:33 +0000 | [diff] [blame] | 2676 | frag = pinfo->frags + nr_frags; |
| 2677 | frag2 = skbinfo->frags + i; |
Herbert Xu | 66e92fc | 2009-05-26 18:50:32 +0000 | [diff] [blame] | 2678 | do { |
| 2679 | *--frag = *--frag2; |
| 2680 | } while (--i); |
| 2681 | |
| 2682 | frag->page_offset += offset; |
| 2683 | frag->size -= offset; |
| 2684 | |
Herbert Xu | f557206 | 2009-01-14 20:40:03 -0800 | [diff] [blame] | 2685 | skb->truesize -= skb->data_len; |
| 2686 | skb->len -= skb->data_len; |
| 2687 | skb->data_len = 0; |
| 2688 | |
Herbert Xu | 5d38a07 | 2009-01-04 16:13:40 -0800 | [diff] [blame] | 2689 | NAPI_GRO_CB(skb)->free = 1; |
| 2690 | goto done; |
Herbert Xu | 69c0cab | 2009-11-17 05:18:18 -0800 | [diff] [blame] | 2691 | } else if (skb_gro_len(p) != pinfo->gso_size) |
| 2692 | return -E2BIG; |
Herbert Xu | 71d93b3 | 2008-12-15 23:42:33 -0800 | [diff] [blame] | 2693 | |
| 2694 | headroom = skb_headroom(p); |
Eric Dumazet | 3d3be43 | 2010-09-01 00:50:51 +0000 | [diff] [blame] | 2695 | nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC); |
Herbert Xu | 71d93b3 | 2008-12-15 23:42:33 -0800 | [diff] [blame] | 2696 | if (unlikely(!nskb)) |
| 2697 | return -ENOMEM; |
| 2698 | |
| 2699 | __copy_skb_header(nskb, p); |
| 2700 | nskb->mac_len = p->mac_len; |
| 2701 | |
| 2702 | skb_reserve(nskb, headroom); |
Herbert Xu | 8691173 | 2009-01-29 14:19:50 +0000 | [diff] [blame] | 2703 | __skb_put(nskb, skb_gro_offset(p)); |
Herbert Xu | 71d93b3 | 2008-12-15 23:42:33 -0800 | [diff] [blame] | 2704 | |
Herbert Xu | 8691173 | 2009-01-29 14:19:50 +0000 | [diff] [blame] | 2705 | skb_set_mac_header(nskb, skb_mac_header(p) - p->data); |
Herbert Xu | 71d93b3 | 2008-12-15 23:42:33 -0800 | [diff] [blame] | 2706 | skb_set_network_header(nskb, skb_network_offset(p)); |
| 2707 | skb_set_transport_header(nskb, skb_transport_offset(p)); |
| 2708 | |
Herbert Xu | 8691173 | 2009-01-29 14:19:50 +0000 | [diff] [blame] | 2709 | __skb_pull(p, skb_gro_offset(p)); |
| 2710 | memcpy(skb_mac_header(nskb), skb_mac_header(p), |
| 2711 | p->data - skb_mac_header(p)); |
Herbert Xu | 71d93b3 | 2008-12-15 23:42:33 -0800 | [diff] [blame] | 2712 | |
| 2713 | *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p); |
| 2714 | skb_shinfo(nskb)->frag_list = p; |
Herbert Xu | 9aaa156 | 2009-05-26 18:50:33 +0000 | [diff] [blame] | 2715 | skb_shinfo(nskb)->gso_size = pinfo->gso_size; |
Herbert Xu | 622e0ca | 2010-05-20 23:07:56 -0700 | [diff] [blame] | 2716 | pinfo->gso_size = 0; |
Herbert Xu | 71d93b3 | 2008-12-15 23:42:33 -0800 | [diff] [blame] | 2717 | skb_header_release(p); |
| 2718 | nskb->prev = p; |
| 2719 | |
| 2720 | nskb->data_len += p->len; |
| 2721 | nskb->truesize += p->len; |
| 2722 | nskb->len += p->len; |
| 2723 | |
| 2724 | *head = nskb; |
| 2725 | nskb->next = p->next; |
| 2726 | p->next = NULL; |
| 2727 | |
| 2728 | p = nskb; |
| 2729 | |
| 2730 | merge: |
Herbert Xu | 67147ba | 2009-05-26 18:50:22 +0000 | [diff] [blame] | 2731 | if (offset > headlen) { |
Herbert Xu | 9aaa156 | 2009-05-26 18:50:33 +0000 | [diff] [blame] | 2732 | skbinfo->frags[0].page_offset += offset - headlen; |
| 2733 | skbinfo->frags[0].size -= offset - headlen; |
Herbert Xu | 67147ba | 2009-05-26 18:50:22 +0000 | [diff] [blame] | 2734 | offset = headlen; |
Herbert Xu | 5603502 | 2009-02-05 21:26:52 -0800 | [diff] [blame] | 2735 | } |
| 2736 | |
Herbert Xu | 67147ba | 2009-05-26 18:50:22 +0000 | [diff] [blame] | 2737 | __skb_pull(skb, offset); |
Herbert Xu | 5603502 | 2009-02-05 21:26:52 -0800 | [diff] [blame] | 2738 | |
Herbert Xu | 71d93b3 | 2008-12-15 23:42:33 -0800 | [diff] [blame] | 2739 | p->prev->next = skb; |
| 2740 | p->prev = skb; |
| 2741 | skb_header_release(skb); |
| 2742 | |
Herbert Xu | 5d38a07 | 2009-01-04 16:13:40 -0800 | [diff] [blame] | 2743 | done: |
| 2744 | NAPI_GRO_CB(p)->count++; |
Herbert Xu | 37fe473 | 2009-01-17 19:48:13 +0000 | [diff] [blame] | 2745 | p->data_len += len; |
| 2746 | p->truesize += len; |
| 2747 | p->len += len; |
Herbert Xu | 71d93b3 | 2008-12-15 23:42:33 -0800 | [diff] [blame] | 2748 | |
| 2749 | NAPI_GRO_CB(skb)->same_flow = 1; |
| 2750 | return 0; |
| 2751 | } |
| 2752 | EXPORT_SYMBOL_GPL(skb_gro_receive); |
| 2753 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2754 | void __init skb_init(void) |
| 2755 | { |
| 2756 | skbuff_head_cache = kmem_cache_create("skbuff_head_cache", |
| 2757 | sizeof(struct sk_buff), |
| 2758 | 0, |
Alexey Dobriyan | e5d679f33 | 2006-08-26 19:25:52 -0700 | [diff] [blame] | 2759 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, |
Paul Mundt | 20c2df8 | 2007-07-20 10:11:58 +0900 | [diff] [blame] | 2760 | NULL); |
David S. Miller | d179cd1 | 2005-08-17 14:57:30 -0700 | [diff] [blame] | 2761 | skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache", |
| 2762 | (2*sizeof(struct sk_buff)) + |
| 2763 | sizeof(atomic_t), |
| 2764 | 0, |
Alexey Dobriyan | e5d679f33 | 2006-08-26 19:25:52 -0700 | [diff] [blame] | 2765 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, |
Paul Mundt | 20c2df8 | 2007-07-20 10:11:58 +0900 | [diff] [blame] | 2766 | NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2767 | } |
| 2768 | |
David Howells | 716ea3a | 2007-04-02 20:19:53 -0700 | [diff] [blame] | 2769 | /** |
| 2770 | * skb_to_sgvec - Fill a scatter-gather list from a socket buffer |
| 2771 | * @skb: Socket buffer containing the buffers to be mapped |
| 2772 | * @sg: The scatter-gather list to map into |
| 2773 | * @offset: The offset into the buffer's contents to start mapping |
| 2774 | * @len: Length of buffer space to be mapped |
| 2775 | * |
| 2776 | * Fill the specified scatter-gather list with mappings/pointers into a |
| 2777 | * region of the buffer space attached to a socket buffer. |
| 2778 | */ |
David S. Miller | 51c739d | 2007-10-30 21:29:29 -0700 | [diff] [blame] | 2779 | static int |
| 2780 | __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) |
David Howells | 716ea3a | 2007-04-02 20:19:53 -0700 | [diff] [blame] | 2781 | { |
David S. Miller | 1a028e5 | 2007-04-27 15:21:23 -0700 | [diff] [blame] | 2782 | int start = skb_headlen(skb); |
| 2783 | int i, copy = start - offset; |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 2784 | struct sk_buff *frag_iter; |
David Howells | 716ea3a | 2007-04-02 20:19:53 -0700 | [diff] [blame] | 2785 | int elt = 0; |
| 2786 | |
| 2787 | if (copy > 0) { |
| 2788 | if (copy > len) |
| 2789 | copy = len; |
Jens Axboe | 642f149 | 2007-10-24 11:20:47 +0200 | [diff] [blame] | 2790 | sg_set_buf(sg, skb->data + offset, copy); |
David Howells | 716ea3a | 2007-04-02 20:19:53 -0700 | [diff] [blame] | 2791 | elt++; |
| 2792 | if ((len -= copy) == 0) |
| 2793 | return elt; |
| 2794 | offset += copy; |
| 2795 | } |
| 2796 | |
| 2797 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
David S. Miller | 1a028e5 | 2007-04-27 15:21:23 -0700 | [diff] [blame] | 2798 | int end; |
David Howells | 716ea3a | 2007-04-02 20:19:53 -0700 | [diff] [blame] | 2799 | |
Ilpo Järvinen | 547b792 | 2008-07-25 21:43:18 -0700 | [diff] [blame] | 2800 | WARN_ON(start > offset + len); |
David S. Miller | 1a028e5 | 2007-04-27 15:21:23 -0700 | [diff] [blame] | 2801 | |
| 2802 | end = start + skb_shinfo(skb)->frags[i].size; |
David Howells | 716ea3a | 2007-04-02 20:19:53 -0700 | [diff] [blame] | 2803 | if ((copy = end - offset) > 0) { |
| 2804 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
| 2805 | |
| 2806 | if (copy > len) |
| 2807 | copy = len; |
Jens Axboe | 642f149 | 2007-10-24 11:20:47 +0200 | [diff] [blame] | 2808 | sg_set_page(&sg[elt], frag->page, copy, |
| 2809 | frag->page_offset+offset-start); |
David Howells | 716ea3a | 2007-04-02 20:19:53 -0700 | [diff] [blame] | 2810 | elt++; |
| 2811 | if (!(len -= copy)) |
| 2812 | return elt; |
| 2813 | offset += copy; |
| 2814 | } |
David S. Miller | 1a028e5 | 2007-04-27 15:21:23 -0700 | [diff] [blame] | 2815 | start = end; |
David Howells | 716ea3a | 2007-04-02 20:19:53 -0700 | [diff] [blame] | 2816 | } |
| 2817 | |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 2818 | skb_walk_frags(skb, frag_iter) { |
| 2819 | int end; |
David Howells | 716ea3a | 2007-04-02 20:19:53 -0700 | [diff] [blame] | 2820 | |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 2821 | WARN_ON(start > offset + len); |
David Howells | 716ea3a | 2007-04-02 20:19:53 -0700 | [diff] [blame] | 2822 | |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 2823 | end = start + frag_iter->len; |
| 2824 | if ((copy = end - offset) > 0) { |
| 2825 | if (copy > len) |
| 2826 | copy = len; |
| 2827 | elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start, |
| 2828 | copy); |
| 2829 | if ((len -= copy) == 0) |
| 2830 | return elt; |
| 2831 | offset += copy; |
David Howells | 716ea3a | 2007-04-02 20:19:53 -0700 | [diff] [blame] | 2832 | } |
David S. Miller | fbb398a | 2009-06-09 00:18:59 -0700 | [diff] [blame] | 2833 | start = end; |
David Howells | 716ea3a | 2007-04-02 20:19:53 -0700 | [diff] [blame] | 2834 | } |
| 2835 | BUG_ON(len); |
| 2836 | return elt; |
| 2837 | } |
| 2838 | |
David S. Miller | 51c739d | 2007-10-30 21:29:29 -0700 | [diff] [blame] | 2839 | int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) |
| 2840 | { |
| 2841 | int nsg = __skb_to_sgvec(skb, sg, offset, len); |
| 2842 | |
Jens Axboe | c46f233 | 2007-10-31 12:06:37 +0100 | [diff] [blame] | 2843 | sg_mark_end(&sg[nsg - 1]); |
David S. Miller | 51c739d | 2007-10-30 21:29:29 -0700 | [diff] [blame] | 2844 | |
| 2845 | return nsg; |
| 2846 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 2847 | EXPORT_SYMBOL_GPL(skb_to_sgvec); |
David S. Miller | 51c739d | 2007-10-30 21:29:29 -0700 | [diff] [blame] | 2848 | |
David Howells | 716ea3a | 2007-04-02 20:19:53 -0700 | [diff] [blame] | 2849 | /** |
| 2850 | * skb_cow_data - Check that a socket buffer's data buffers are writable |
| 2851 | * @skb: The socket buffer to check. |
| 2852 | * @tailbits: Amount of trailing space to be added |
| 2853 | * @trailer: Returned pointer to the skb where the @tailbits space begins |
| 2854 | * |
| 2855 | * Make sure that the data buffers attached to a socket buffer are |
| 2856 | * writable. If they are not, private copies are made of the data buffers |
| 2857 | * and the socket buffer is set to use these instead. |
| 2858 | * |
| 2859 | * If @tailbits is given, make sure that there is space to write @tailbits |
| 2860 | * bytes of data beyond current end of socket buffer. @trailer will be |
| 2861 | * set to point to the skb in which this space begins. |
| 2862 | * |
| 2863 | * The number of scatterlist elements required to completely map the |
| 2864 | * COW'd and extended socket buffer will be returned. |
| 2865 | */ |
| 2866 | int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer) |
| 2867 | { |
| 2868 | int copyflag; |
| 2869 | int elt; |
| 2870 | struct sk_buff *skb1, **skb_p; |
| 2871 | |
| 2872 | /* If skb is cloned or its head is paged, reallocate |
| 2873 | * head pulling out all the pages (pages are considered not writable |
| 2874 | * at the moment even if they are anonymous). |
| 2875 | */ |
| 2876 | if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) && |
| 2877 | __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL) |
| 2878 | return -ENOMEM; |
| 2879 | |
| 2880 | /* Easy case. Most of packets will go this way. */ |
David S. Miller | 21dc330 | 2010-08-23 00:13:46 -0700 | [diff] [blame] | 2881 | if (!skb_has_frag_list(skb)) { |
David Howells | 716ea3a | 2007-04-02 20:19:53 -0700 | [diff] [blame] | 2882 | /* A little of trouble, not enough of space for trailer. |
| 2883 | * This should not happen, when stack is tuned to generate |
| 2884 | * good frames. OK, on miss we reallocate and reserve even more |
| 2885 | * space, 128 bytes is fair. */ |
| 2886 | |
| 2887 | if (skb_tailroom(skb) < tailbits && |
| 2888 | pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC)) |
| 2889 | return -ENOMEM; |
| 2890 | |
| 2891 | /* Voila! */ |
| 2892 | *trailer = skb; |
| 2893 | return 1; |
| 2894 | } |
| 2895 | |
| 2896 | /* Misery. We are in troubles, going to mincer fragments... */ |
| 2897 | |
| 2898 | elt = 1; |
| 2899 | skb_p = &skb_shinfo(skb)->frag_list; |
| 2900 | copyflag = 0; |
| 2901 | |
| 2902 | while ((skb1 = *skb_p) != NULL) { |
| 2903 | int ntail = 0; |
| 2904 | |
| 2905 | /* The fragment is partially pulled by someone, |
| 2906 | * this can happen on input. Copy it and everything |
| 2907 | * after it. */ |
| 2908 | |
| 2909 | if (skb_shared(skb1)) |
| 2910 | copyflag = 1; |
| 2911 | |
| 2912 | /* If the skb is the last, worry about trailer. */ |
| 2913 | |
| 2914 | if (skb1->next == NULL && tailbits) { |
| 2915 | if (skb_shinfo(skb1)->nr_frags || |
David S. Miller | 21dc330 | 2010-08-23 00:13:46 -0700 | [diff] [blame] | 2916 | skb_has_frag_list(skb1) || |
David Howells | 716ea3a | 2007-04-02 20:19:53 -0700 | [diff] [blame] | 2917 | skb_tailroom(skb1) < tailbits) |
| 2918 | ntail = tailbits + 128; |
| 2919 | } |
| 2920 | |
| 2921 | if (copyflag || |
| 2922 | skb_cloned(skb1) || |
| 2923 | ntail || |
| 2924 | skb_shinfo(skb1)->nr_frags || |
David S. Miller | 21dc330 | 2010-08-23 00:13:46 -0700 | [diff] [blame] | 2925 | skb_has_frag_list(skb1)) { |
David Howells | 716ea3a | 2007-04-02 20:19:53 -0700 | [diff] [blame] | 2926 | struct sk_buff *skb2; |
| 2927 | |
| 2928 | /* Fuck, we are miserable poor guys... */ |
| 2929 | if (ntail == 0) |
| 2930 | skb2 = skb_copy(skb1, GFP_ATOMIC); |
| 2931 | else |
| 2932 | skb2 = skb_copy_expand(skb1, |
| 2933 | skb_headroom(skb1), |
| 2934 | ntail, |
| 2935 | GFP_ATOMIC); |
| 2936 | if (unlikely(skb2 == NULL)) |
| 2937 | return -ENOMEM; |
| 2938 | |
| 2939 | if (skb1->sk) |
| 2940 | skb_set_owner_w(skb2, skb1->sk); |
| 2941 | |
| 2942 | /* Looking around. Are we still alive? |
| 2943 | * OK, link new skb, drop old one */ |
| 2944 | |
| 2945 | skb2->next = skb1->next; |
| 2946 | *skb_p = skb2; |
| 2947 | kfree_skb(skb1); |
| 2948 | skb1 = skb2; |
| 2949 | } |
| 2950 | elt++; |
| 2951 | *trailer = skb1; |
| 2952 | skb_p = &skb1->next; |
| 2953 | } |
| 2954 | |
| 2955 | return elt; |
| 2956 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 2957 | EXPORT_SYMBOL_GPL(skb_cow_data); |
David Howells | 716ea3a | 2007-04-02 20:19:53 -0700 | [diff] [blame] | 2958 | |
Eric Dumazet | b1faf56 | 2010-05-31 23:44:05 -0700 | [diff] [blame] | 2959 | static void sock_rmem_free(struct sk_buff *skb) |
| 2960 | { |
| 2961 | struct sock *sk = skb->sk; |
| 2962 | |
| 2963 | atomic_sub(skb->truesize, &sk->sk_rmem_alloc); |
| 2964 | } |
| 2965 | |
| 2966 | /* |
| 2967 | * Note: We dont mem charge error packets (no sk_forward_alloc changes) |
| 2968 | */ |
| 2969 | int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb) |
| 2970 | { |
| 2971 | if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= |
| 2972 | (unsigned)sk->sk_rcvbuf) |
| 2973 | return -ENOMEM; |
| 2974 | |
| 2975 | skb_orphan(skb); |
| 2976 | skb->sk = sk; |
| 2977 | skb->destructor = sock_rmem_free; |
| 2978 | atomic_add(skb->truesize, &sk->sk_rmem_alloc); |
| 2979 | |
| 2980 | skb_queue_tail(&sk->sk_error_queue, skb); |
| 2981 | if (!sock_flag(sk, SOCK_DEAD)) |
| 2982 | sk->sk_data_ready(sk, skb->len); |
| 2983 | return 0; |
| 2984 | } |
| 2985 | EXPORT_SYMBOL(sock_queue_err_skb); |
| 2986 | |
Patrick Ohly | ac45f60 | 2009-02-12 05:03:37 +0000 | [diff] [blame] | 2987 | void skb_tstamp_tx(struct sk_buff *orig_skb, |
| 2988 | struct skb_shared_hwtstamps *hwtstamps) |
| 2989 | { |
| 2990 | struct sock *sk = orig_skb->sk; |
| 2991 | struct sock_exterr_skb *serr; |
| 2992 | struct sk_buff *skb; |
| 2993 | int err; |
| 2994 | |
| 2995 | if (!sk) |
| 2996 | return; |
| 2997 | |
| 2998 | skb = skb_clone(orig_skb, GFP_ATOMIC); |
| 2999 | if (!skb) |
| 3000 | return; |
| 3001 | |
| 3002 | if (hwtstamps) { |
| 3003 | *skb_hwtstamps(skb) = |
| 3004 | *hwtstamps; |
| 3005 | } else { |
| 3006 | /* |
| 3007 | * no hardware time stamps available, |
Oliver Hartkopp | 2244d07 | 2010-08-17 08:59:14 +0000 | [diff] [blame] | 3008 | * so keep the shared tx_flags and only |
Patrick Ohly | ac45f60 | 2009-02-12 05:03:37 +0000 | [diff] [blame] | 3009 | * store software time stamp |
| 3010 | */ |
| 3011 | skb->tstamp = ktime_get_real(); |
| 3012 | } |
| 3013 | |
| 3014 | serr = SKB_EXT_ERR(skb); |
| 3015 | memset(serr, 0, sizeof(*serr)); |
| 3016 | serr->ee.ee_errno = ENOMSG; |
| 3017 | serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING; |
Eric Dumazet | 2903037 | 2010-05-29 00:20:48 -0700 | [diff] [blame] | 3018 | |
Patrick Ohly | ac45f60 | 2009-02-12 05:03:37 +0000 | [diff] [blame] | 3019 | err = sock_queue_err_skb(sk, skb); |
Eric Dumazet | 2903037 | 2010-05-29 00:20:48 -0700 | [diff] [blame] | 3020 | |
Patrick Ohly | ac45f60 | 2009-02-12 05:03:37 +0000 | [diff] [blame] | 3021 | if (err) |
| 3022 | kfree_skb(skb); |
| 3023 | } |
| 3024 | EXPORT_SYMBOL_GPL(skb_tstamp_tx); |
| 3025 | |
| 3026 | |
Rusty Russell | f35d9d8 | 2008-02-04 23:49:54 -0500 | [diff] [blame] | 3027 | /** |
| 3028 | * skb_partial_csum_set - set up and verify partial csum values for packet |
| 3029 | * @skb: the skb to set |
| 3030 | * @start: the number of bytes after skb->data to start checksumming. |
| 3031 | * @off: the offset from start to place the checksum. |
| 3032 | * |
| 3033 | * For untrusted partially-checksummed packets, we need to make sure the values |
| 3034 | * for skb->csum_start and skb->csum_offset are valid so we don't oops. |
| 3035 | * |
| 3036 | * This function checks and sets those values and skb->ip_summed: if this |
| 3037 | * returns false you should drop the packet. |
| 3038 | */ |
| 3039 | bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off) |
| 3040 | { |
Herbert Xu | 5ff8dda | 2009-06-04 01:22:01 +0000 | [diff] [blame] | 3041 | if (unlikely(start > skb_headlen(skb)) || |
| 3042 | unlikely((int)start + off > skb_headlen(skb) - 2)) { |
Rusty Russell | f35d9d8 | 2008-02-04 23:49:54 -0500 | [diff] [blame] | 3043 | if (net_ratelimit()) |
| 3044 | printk(KERN_WARNING |
| 3045 | "bad partial csum: csum=%u/%u len=%u\n", |
Herbert Xu | 5ff8dda | 2009-06-04 01:22:01 +0000 | [diff] [blame] | 3046 | start, off, skb_headlen(skb)); |
Rusty Russell | f35d9d8 | 2008-02-04 23:49:54 -0500 | [diff] [blame] | 3047 | return false; |
| 3048 | } |
| 3049 | skb->ip_summed = CHECKSUM_PARTIAL; |
| 3050 | skb->csum_start = skb_headroom(skb) + start; |
| 3051 | skb->csum_offset = off; |
| 3052 | return true; |
| 3053 | } |
David S. Miller | b4ac530fc | 2009-02-10 02:09:24 -0800 | [diff] [blame] | 3054 | EXPORT_SYMBOL_GPL(skb_partial_csum_set); |
Rusty Russell | f35d9d8 | 2008-02-04 23:49:54 -0500 | [diff] [blame] | 3055 | |
Ben Hutchings | 4497b07 | 2008-06-19 16:22:28 -0700 | [diff] [blame] | 3056 | void __skb_warn_lro_forwarding(const struct sk_buff *skb) |
| 3057 | { |
| 3058 | if (net_ratelimit()) |
| 3059 | pr_warning("%s: received packets cannot be forwarded" |
| 3060 | " while LRO is enabled\n", skb->dev->name); |
| 3061 | } |
Ben Hutchings | 4497b07 | 2008-06-19 16:22:28 -0700 | [diff] [blame] | 3062 | EXPORT_SYMBOL(__skb_warn_lro_forwarding); |