Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Routines having to do with the 'struct sk_buff' memory handlers. | |
3 | * | |
113aa838 | 4 | * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk> |
1da177e4 LT |
5 | * Florian La Roche <rzsfl@rz.uni-sb.de> |
6 | * | |
1da177e4 LT |
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 | ||
1da177e4 LT |
39 | #include <linux/module.h> |
40 | #include <linux/types.h> | |
41 | #include <linux/kernel.h> | |
fe55f6d5 | 42 | #include <linux/kmemcheck.h> |
1da177e4 LT |
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> | |
9c55e01c | 54 | #include <linux/splice.h> |
1da177e4 LT |
55 | #include <linux/cache.h> |
56 | #include <linux/rtnetlink.h> | |
57 | #include <linux/init.h> | |
716ea3a7 | 58 | #include <linux/scatterlist.h> |
ac45f602 | 59 | #include <linux/errqueue.h> |
1da177e4 LT |
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> | |
ad8d75ff | 69 | #include <trace/events/skb.h> |
1da177e4 | 70 | |
a1f8e7f7 AV |
71 | #include "kmap_skb.h" |
72 | ||
e18b890b CL |
73 | static struct kmem_cache *skbuff_head_cache __read_mostly; |
74 | static struct kmem_cache *skbuff_fclone_cache __read_mostly; | |
1da177e4 | 75 | |
9c55e01c JA |
76 | static void sock_pipe_buf_release(struct pipe_inode_info *pipe, |
77 | struct pipe_buffer *buf) | |
78 | { | |
8b9d3728 | 79 | put_page(buf->page); |
9c55e01c JA |
80 | } |
81 | ||
82 | static void sock_pipe_buf_get(struct pipe_inode_info *pipe, | |
83 | struct pipe_buffer *buf) | |
84 | { | |
8b9d3728 | 85 | get_page(buf->page); |
9c55e01c JA |
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. */ | |
28dfef8f | 96 | static const struct pipe_buf_operations sock_pipe_buf_ops = { |
9c55e01c JA |
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 | ||
1da177e4 LT |
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 | */ | |
ccb7c773 | 120 | static void skb_over_panic(struct sk_buff *skb, int sz, void *here) |
1da177e4 | 121 | { |
26095455 | 122 | printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p " |
4305b541 | 123 | "data:%p tail:%#lx end:%#lx dev:%s\n", |
27a884dc | 124 | here, skb->len, sz, skb->head, skb->data, |
4305b541 | 125 | (unsigned long)skb->tail, (unsigned long)skb->end, |
26095455 | 126 | skb->dev ? skb->dev->name : "<NULL>"); |
1da177e4 LT |
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 | ||
ccb7c773 | 139 | static void skb_under_panic(struct sk_buff *skb, int sz, void *here) |
1da177e4 | 140 | { |
26095455 | 141 | printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p " |
4305b541 | 142 | "data:%p tail:%#lx end:%#lx dev:%s\n", |
27a884dc | 143 | here, skb->len, sz, skb->head, skb->data, |
4305b541 | 144 | (unsigned long)skb->tail, (unsigned long)skb->end, |
26095455 | 145 | skb->dev ? skb->dev->name : "<NULL>"); |
1da177e4 LT |
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 | /** | |
d179cd12 | 156 | * __alloc_skb - allocate a network buffer |
1da177e4 LT |
157 | * @size: size to allocate |
158 | * @gfp_mask: allocation mask | |
c83c2486 RD |
159 | * @fclone: allocate from fclone cache instead of head cache |
160 | * and allocate a cloned (child) skb | |
b30973f8 | 161 | * @node: numa node to allocate memory on |
1da177e4 LT |
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 | */ | |
dd0fc66f | 170 | struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, |
b30973f8 | 171 | int fclone, int node) |
1da177e4 | 172 | { |
e18b890b | 173 | struct kmem_cache *cache; |
4947d3ef | 174 | struct skb_shared_info *shinfo; |
1da177e4 LT |
175 | struct sk_buff *skb; |
176 | u8 *data; | |
177 | ||
8798b3fb HX |
178 | cache = fclone ? skbuff_fclone_cache : skbuff_head_cache; |
179 | ||
1da177e4 | 180 | /* Get the HEAD */ |
b30973f8 | 181 | skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node); |
1da177e4 LT |
182 | if (!skb) |
183 | goto out; | |
ec7d2f2c | 184 | prefetchw(skb); |
1da177e4 | 185 | |
1da177e4 | 186 | size = SKB_DATA_ALIGN(size); |
b30973f8 CH |
187 | data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info), |
188 | gfp_mask, node); | |
1da177e4 LT |
189 | if (!data) |
190 | goto nodata; | |
ec7d2f2c | 191 | prefetchw(data + size); |
1da177e4 | 192 | |
ca0605a7 | 193 | /* |
c8005785 JB |
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! | |
ca0605a7 ACM |
197 | */ |
198 | memset(skb, 0, offsetof(struct sk_buff, tail)); | |
1da177e4 LT |
199 | skb->truesize = size + sizeof(struct sk_buff); |
200 | atomic_set(&skb->users, 1); | |
201 | skb->head = data; | |
202 | skb->data = data; | |
27a884dc | 203 | skb_reset_tail_pointer(skb); |
4305b541 | 204 | skb->end = skb->tail + size; |
19633e12 SH |
205 | #ifdef NET_SKBUFF_DATA_USES_OFFSET |
206 | skb->mac_header = ~0U; | |
207 | #endif | |
208 | ||
4947d3ef BL |
209 | /* make sure we initialize shinfo sequentially */ |
210 | shinfo = skb_shinfo(skb); | |
ec7d2f2c | 211 | memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); |
4947d3ef | 212 | atomic_set(&shinfo->dataref, 1); |
c2aa3665 | 213 | kmemcheck_annotate_variable(shinfo->destructor_arg); |
4947d3ef | 214 | |
d179cd12 DM |
215 | if (fclone) { |
216 | struct sk_buff *child = skb + 1; | |
217 | atomic_t *fclone_ref = (atomic_t *) (child + 1); | |
1da177e4 | 218 | |
fe55f6d5 VN |
219 | kmemcheck_annotate_bitfield(child, flags1); |
220 | kmemcheck_annotate_bitfield(child, flags2); | |
d179cd12 DM |
221 | skb->fclone = SKB_FCLONE_ORIG; |
222 | atomic_set(fclone_ref, 1); | |
223 | ||
224 | child->fclone = SKB_FCLONE_UNAVAILABLE; | |
225 | } | |
1da177e4 LT |
226 | out: |
227 | return skb; | |
228 | nodata: | |
8798b3fb | 229 | kmem_cache_free(cache, skb); |
1da177e4 LT |
230 | skb = NULL; |
231 | goto out; | |
1da177e4 | 232 | } |
b4ac530f | 233 | EXPORT_SYMBOL(__alloc_skb); |
1da177e4 | 234 | |
8af27456 CH |
235 | /** |
236 | * __netdev_alloc_skb - allocate an skbuff for rx on a specific device | |
237 | * @dev: network device to receive on | |
238 | * @length: length to allocate | |
239 | * @gfp_mask: get_free_pages mask, passed to alloc_skb | |
240 | * | |
241 | * Allocate a new &sk_buff and assign it a usage count of one. The | |
242 | * buffer has unspecified headroom built in. Users should allocate | |
243 | * the headroom they think they need without accounting for the | |
244 | * built in space. The built in space is used for optimisations. | |
245 | * | |
246 | * %NULL is returned if there is no free memory. | |
247 | */ | |
248 | struct sk_buff *__netdev_alloc_skb(struct net_device *dev, | |
249 | unsigned int length, gfp_t gfp_mask) | |
250 | { | |
251 | struct sk_buff *skb; | |
252 | ||
564824b0 | 253 | skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, NUMA_NO_NODE); |
7b2e497a | 254 | if (likely(skb)) { |
8af27456 | 255 | skb_reserve(skb, NET_SKB_PAD); |
7b2e497a CH |
256 | skb->dev = dev; |
257 | } | |
8af27456 CH |
258 | return skb; |
259 | } | |
b4ac530f | 260 | EXPORT_SYMBOL(__netdev_alloc_skb); |
1da177e4 | 261 | |
654bed16 PZ |
262 | void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, |
263 | int size) | |
264 | { | |
265 | skb_fill_page_desc(skb, i, page, off, size); | |
266 | skb->len += size; | |
267 | skb->data_len += size; | |
268 | skb->truesize += size; | |
269 | } | |
270 | EXPORT_SYMBOL(skb_add_rx_frag); | |
271 | ||
f58518e6 IJ |
272 | /** |
273 | * dev_alloc_skb - allocate an skbuff for receiving | |
274 | * @length: length to allocate | |
275 | * | |
276 | * Allocate a new &sk_buff and assign it a usage count of one. The | |
277 | * buffer has unspecified headroom built in. Users should allocate | |
278 | * the headroom they think they need without accounting for the | |
279 | * built in space. The built in space is used for optimisations. | |
280 | * | |
281 | * %NULL is returned if there is no free memory. Although this function | |
282 | * allocates memory it can be called from an interrupt. | |
283 | */ | |
284 | struct sk_buff *dev_alloc_skb(unsigned int length) | |
285 | { | |
1483b874 DV |
286 | /* |
287 | * There is more code here than it seems: | |
a0f55e0e | 288 | * __dev_alloc_skb is an inline |
1483b874 | 289 | */ |
f58518e6 IJ |
290 | return __dev_alloc_skb(length, GFP_ATOMIC); |
291 | } | |
292 | EXPORT_SYMBOL(dev_alloc_skb); | |
293 | ||
27b437c8 | 294 | static void skb_drop_list(struct sk_buff **listp) |
1da177e4 | 295 | { |
27b437c8 | 296 | struct sk_buff *list = *listp; |
1da177e4 | 297 | |
27b437c8 | 298 | *listp = NULL; |
1da177e4 LT |
299 | |
300 | do { | |
301 | struct sk_buff *this = list; | |
302 | list = list->next; | |
303 | kfree_skb(this); | |
304 | } while (list); | |
305 | } | |
306 | ||
27b437c8 HX |
307 | static inline void skb_drop_fraglist(struct sk_buff *skb) |
308 | { | |
309 | skb_drop_list(&skb_shinfo(skb)->frag_list); | |
310 | } | |
311 | ||
1da177e4 LT |
312 | static void skb_clone_fraglist(struct sk_buff *skb) |
313 | { | |
314 | struct sk_buff *list; | |
315 | ||
fbb398a8 | 316 | skb_walk_frags(skb, list) |
1da177e4 LT |
317 | skb_get(list); |
318 | } | |
319 | ||
5bba1712 | 320 | static void skb_release_data(struct sk_buff *skb) |
1da177e4 LT |
321 | { |
322 | if (!skb->cloned || | |
323 | !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, | |
324 | &skb_shinfo(skb)->dataref)) { | |
325 | if (skb_shinfo(skb)->nr_frags) { | |
326 | int i; | |
327 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) | |
328 | put_page(skb_shinfo(skb)->frags[i].page); | |
329 | } | |
330 | ||
21dc3301 | 331 | if (skb_has_frag_list(skb)) |
1da177e4 LT |
332 | skb_drop_fraglist(skb); |
333 | ||
334 | kfree(skb->head); | |
335 | } | |
336 | } | |
337 | ||
338 | /* | |
339 | * Free an skbuff by memory without cleaning the state. | |
340 | */ | |
2d4baff8 | 341 | static void kfree_skbmem(struct sk_buff *skb) |
1da177e4 | 342 | { |
d179cd12 DM |
343 | struct sk_buff *other; |
344 | atomic_t *fclone_ref; | |
345 | ||
d179cd12 DM |
346 | switch (skb->fclone) { |
347 | case SKB_FCLONE_UNAVAILABLE: | |
348 | kmem_cache_free(skbuff_head_cache, skb); | |
349 | break; | |
350 | ||
351 | case SKB_FCLONE_ORIG: | |
352 | fclone_ref = (atomic_t *) (skb + 2); | |
353 | if (atomic_dec_and_test(fclone_ref)) | |
354 | kmem_cache_free(skbuff_fclone_cache, skb); | |
355 | break; | |
356 | ||
357 | case SKB_FCLONE_CLONE: | |
358 | fclone_ref = (atomic_t *) (skb + 1); | |
359 | other = skb - 1; | |
360 | ||
361 | /* The clone portion is available for | |
362 | * fast-cloning again. | |
363 | */ | |
364 | skb->fclone = SKB_FCLONE_UNAVAILABLE; | |
365 | ||
366 | if (atomic_dec_and_test(fclone_ref)) | |
367 | kmem_cache_free(skbuff_fclone_cache, other); | |
368 | break; | |
3ff50b79 | 369 | } |
1da177e4 LT |
370 | } |
371 | ||
04a4bb55 | 372 | static void skb_release_head_state(struct sk_buff *skb) |
1da177e4 | 373 | { |
adf30907 | 374 | skb_dst_drop(skb); |
1da177e4 LT |
375 | #ifdef CONFIG_XFRM |
376 | secpath_put(skb->sp); | |
377 | #endif | |
9c2b3328 SH |
378 | if (skb->destructor) { |
379 | WARN_ON(in_irq()); | |
1da177e4 LT |
380 | skb->destructor(skb); |
381 | } | |
9fb9cbb1 | 382 | #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) |
5f79e0f9 | 383 | nf_conntrack_put(skb->nfct); |
2fc72c7b KK |
384 | #endif |
385 | #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED | |
9fb9cbb1 YK |
386 | nf_conntrack_put_reasm(skb->nfct_reasm); |
387 | #endif | |
1da177e4 LT |
388 | #ifdef CONFIG_BRIDGE_NETFILTER |
389 | nf_bridge_put(skb->nf_bridge); | |
390 | #endif | |
1da177e4 LT |
391 | /* XXX: IS this still necessary? - JHS */ |
392 | #ifdef CONFIG_NET_SCHED | |
393 | skb->tc_index = 0; | |
394 | #ifdef CONFIG_NET_CLS_ACT | |
395 | skb->tc_verd = 0; | |
1da177e4 LT |
396 | #endif |
397 | #endif | |
04a4bb55 LB |
398 | } |
399 | ||
400 | /* Free everything but the sk_buff shell. */ | |
401 | static void skb_release_all(struct sk_buff *skb) | |
402 | { | |
403 | skb_release_head_state(skb); | |
2d4baff8 HX |
404 | skb_release_data(skb); |
405 | } | |
406 | ||
407 | /** | |
408 | * __kfree_skb - private function | |
409 | * @skb: buffer | |
410 | * | |
411 | * Free an sk_buff. Release anything attached to the buffer. | |
412 | * Clean the state. This is an internal helper function. Users should | |
413 | * always call kfree_skb | |
414 | */ | |
1da177e4 | 415 | |
2d4baff8 HX |
416 | void __kfree_skb(struct sk_buff *skb) |
417 | { | |
418 | skb_release_all(skb); | |
1da177e4 LT |
419 | kfree_skbmem(skb); |
420 | } | |
b4ac530f | 421 | EXPORT_SYMBOL(__kfree_skb); |
1da177e4 | 422 |