Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Routines having to do with the 'struct sk_buff' memory handlers. | |
3 | * | |
4 | * Authors: Alan Cox <iiitac@pyr.swan.ac.uk> | |
5 | * Florian La Roche <rzsfl@rz.uni-sb.de> | |
6 | * | |
7 | * Version: $Id: skbuff.c,v 1.90 2001/11/07 05:56:19 davem Exp $ | |
8 | * | |
9 | * Fixes: | |
10 | * Alan Cox : Fixed the worst of the load | |
11 | * balancer bugs. | |
12 | * Dave Platt : Interrupt stacking fix. | |
13 | * Richard Kooijman : Timestamp fixes. | |
14 | * Alan Cox : Changed buffer format. | |
15 | * Alan Cox : destructor hook for AF_UNIX etc. | |
16 | * Linus Torvalds : Better skb_clone. | |
17 | * Alan Cox : Added skb_copy. | |
18 | * Alan Cox : Added all the changed routines Linus | |
19 | * only put in the headers | |
20 | * Ray VanTassle : Fixed --skb->lock in free | |
21 | * Alan Cox : skb_copy copy arp field | |
22 | * Andi Kleen : slabified it. | |
23 | * Robert Olsson : Removed skb_head_pool | |
24 | * | |
25 | * NOTE: | |
26 | * The __skb_ routines should be called with interrupts | |
27 | * disabled, or you better be *real* sure that the operation is atomic | |
28 | * with respect to whatever list is being frobbed (e.g. via lock_sock() | |
29 | * or via disabling bottom half handlers, etc). | |
30 | * | |
31 | * This program is free software; you can redistribute it and/or | |
32 | * modify it under the terms of the GNU General Public License | |
33 | * as published by the Free Software Foundation; either version | |
34 | * 2 of the License, or (at your option) any later version. | |
35 | */ | |
36 | ||
37 | /* | |
38 | * The functions in this file will not compile correctly with gcc 2.4.x | |
39 | */ | |
40 | ||
1da177e4 LT |
41 | #include <linux/module.h> |
42 | #include <linux/types.h> | |
43 | #include <linux/kernel.h> | |
1da177e4 LT |
44 | #include <linux/mm.h> |
45 | #include <linux/interrupt.h> | |
46 | #include <linux/in.h> | |
47 | #include <linux/inet.h> | |
48 | #include <linux/slab.h> | |
49 | #include <linux/netdevice.h> | |
50 | #ifdef CONFIG_NET_CLS_ACT | |
51 | #include <net/pkt_sched.h> | |
52 | #endif | |
53 | #include <linux/string.h> | |
54 | #include <linux/skbuff.h> | |
9c55e01c | 55 | #include <linux/splice.h> |
1da177e4 LT |
56 | #include <linux/cache.h> |
57 | #include <linux/rtnetlink.h> | |
58 | #include <linux/init.h> | |
716ea3a7 | 59 | #include <linux/scatterlist.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> | |
69 | ||
a1f8e7f7 AV |
70 | #include "kmap_skb.h" |
71 | ||
e18b890b CL |
72 | static struct kmem_cache *skbuff_head_cache __read_mostly; |
73 | static struct kmem_cache *skbuff_fclone_cache __read_mostly; | |
1da177e4 | 74 | |
9c55e01c JA |
75 | static void sock_pipe_buf_release(struct pipe_inode_info *pipe, |
76 | struct pipe_buffer *buf) | |
77 | { | |
78 | struct sk_buff *skb = (struct sk_buff *) buf->private; | |
79 | ||
80 | kfree_skb(skb); | |
81 | } | |
82 | ||
83 | static void sock_pipe_buf_get(struct pipe_inode_info *pipe, | |
84 | struct pipe_buffer *buf) | |
85 | { | |
86 | struct sk_buff *skb = (struct sk_buff *) buf->private; | |
87 | ||
88 | skb_get(skb); | |
89 | } | |
90 | ||
91 | static int sock_pipe_buf_steal(struct pipe_inode_info *pipe, | |
92 | struct pipe_buffer *buf) | |
93 | { | |
94 | return 1; | |
95 | } | |
96 | ||
97 | ||
98 | /* Pipe buffer operations for a socket. */ | |
99 | static struct pipe_buf_operations sock_pipe_buf_ops = { | |
100 | .can_merge = 0, | |
101 | .map = generic_pipe_buf_map, | |
102 | .unmap = generic_pipe_buf_unmap, | |
103 | .confirm = generic_pipe_buf_confirm, | |
104 | .release = sock_pipe_buf_release, | |
105 | .steal = sock_pipe_buf_steal, | |
106 | .get = sock_pipe_buf_get, | |
107 | }; | |
108 | ||
1da177e4 LT |
109 | /* |
110 | * Keep out-of-line to prevent kernel bloat. | |
111 | * __builtin_return_address is not used because it is not always | |
112 | * reliable. | |
113 | */ | |
114 | ||
115 | /** | |
116 | * skb_over_panic - private function | |
117 | * @skb: buffer | |
118 | * @sz: size | |
119 | * @here: address | |
120 | * | |
121 | * Out of line support code for skb_put(). Not user callable. | |
122 | */ | |
123 | void skb_over_panic(struct sk_buff *skb, int sz, void *here) | |
124 | { | |
26095455 | 125 | printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p " |
4305b541 | 126 | "data:%p tail:%#lx end:%#lx dev:%s\n", |
27a884dc | 127 | here, skb->len, sz, skb->head, skb->data, |
4305b541 | 128 | (unsigned long)skb->tail, (unsigned long)skb->end, |
26095455 | 129 | skb->dev ? skb->dev->name : "<NULL>"); |
1da177e4 LT |
130 | BUG(); |
131 | } | |
132 | ||
133 | /** | |
134 | * skb_under_panic - private function | |
135 | * @skb: buffer | |
136 | * @sz: size | |
137 | * @here: address | |
138 | * | |
139 | * Out of line support code for skb_push(). Not user callable. | |
140 | */ | |
141 | ||
142 | void skb_under_panic(struct sk_buff *skb, int sz, void *here) | |
143 | { | |
26095455 | 144 | printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p " |
4305b541 | 145 | "data:%p tail:%#lx end:%#lx dev:%s\n", |
27a884dc | 146 | here, skb->len, sz, skb->head, skb->data, |
4305b541 | 147 | (unsigned long)skb->tail, (unsigned long)skb->end, |
26095455 | 148 | skb->dev ? skb->dev->name : "<NULL>"); |
1da177e4 LT |
149 | BUG(); |
150 | } | |
151 | ||
dc6de336 DM |
152 | void skb_truesize_bug(struct sk_buff *skb) |
153 | { | |
154 | printk(KERN_ERR "SKB BUG: Invalid truesize (%u) " | |
155 | "len=%u, sizeof(sk_buff)=%Zd\n", | |
156 | skb->truesize, skb->len, sizeof(struct sk_buff)); | |
157 | } | |
158 | EXPORT_SYMBOL(skb_truesize_bug); | |
159 | ||
1da177e4 LT |
160 | /* Allocate a new skbuff. We do this ourselves so we can fill in a few |
161 | * 'private' fields and also do memory statistics to find all the | |
162 | * [BEEP] leaks. | |
163 | * | |
164 | */ | |
165 | ||
166 | /** | |
d179cd12 | 167 | * __alloc_skb - allocate a network buffer |
1da177e4 LT |
168 | * @size: size to allocate |
169 | * @gfp_mask: allocation mask | |
c83c2486 RD |
170 | * @fclone: allocate from fclone cache instead of head cache |
171 | * and allocate a cloned (child) skb | |
b30973f8 | 172 | * @node: numa node to allocate memory on |
1da177e4 LT |
173 | * |
174 | * Allocate a new &sk_buff. The returned buffer has no headroom and a | |
175 | * tail room of size bytes. The object has a reference count of one. | |
176 | * The return is the buffer. On a failure the return is %NULL. | |
177 | * | |
178 | * Buffers may only be allocated from interrupts using a @gfp_mask of | |
179 | * %GFP_ATOMIC. | |
180 | */ | |
dd0fc66f | 181 | struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, |
b30973f8 | 182 | int fclone, int node) |
1da177e4 | 183 | { |
e18b890b | 184 | struct kmem_cache *cache; |
4947d3ef | 185 | struct skb_shared_info *shinfo; |
1da177e4 LT |
186 | struct sk_buff *skb; |
187 | u8 *data; | |
188 | ||
8798b3fb HX |
189 | cache = fclone ? skbuff_fclone_cache : skbuff_head_cache; |
190 | ||
1da177e4 | 191 | /* Get the HEAD */ |
b30973f8 | 192 | skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node); |
1da177e4 LT |
193 | if (!skb) |
194 | goto out; | |
195 | ||
1da177e4 | 196 | size = SKB_DATA_ALIGN(size); |
b30973f8 CH |
197 | data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info), |
198 | gfp_mask, node); | |
1da177e4 LT |
199 | if (!data) |
200 | goto nodata; | |
201 | ||
ca0605a7 ACM |
202 | /* |
203 | * See comment in sk_buff definition, just before the 'tail' member | |
204 | */ | |
205 | memset(skb, 0, offsetof(struct sk_buff, tail)); | |
1da177e4 LT |
206 | skb->truesize = size + sizeof(struct sk_buff); |
207 | atomic_set(&skb->users, 1); | |
208 | skb->head = data; | |
209 | skb->data = data; | |
27a884dc | 210 | skb_reset_tail_pointer(skb); |
4305b541 | 211 | skb->end = skb->tail + size; |
4947d3ef BL |
212 | /* make sure we initialize shinfo sequentially */ |
213 | shinfo = skb_shinfo(skb); | |
214 | atomic_set(&shinfo->dataref, 1); | |
215 | shinfo->nr_frags = 0; | |
7967168c HX |
216 | shinfo->gso_size = 0; |
217 | shinfo->gso_segs = 0; | |
218 | shinfo->gso_type = 0; | |
4947d3ef BL |
219 | shinfo->ip6_frag_id = 0; |
220 | shinfo->frag_list = NULL; | |
221 | ||
d179cd12 DM |
222 | if (fclone) { |
223 | struct sk_buff *child = skb + 1; | |
224 | atomic_t *fclone_ref = (atomic_t *) (child + 1); | |
1da177e4 | 225 | |
d179cd12 DM |
226 | skb->fclone = SKB_FCLONE_ORIG; |
227 | atomic_set(fclone_ref, 1); | |
228 | ||
229 | child->fclone = SKB_FCLONE_UNAVAILABLE; | |
230 | } | |
1da177e4 LT |
231 | out: |
232 | return skb; | |
233 | nodata: | |
8798b3fb | 234 | kmem_cache_free(cache, skb); |
1da177e4 LT |
235 | skb = NULL; |
236 | goto out; | |
1da177e4 LT |
237 | } |
238 | ||
8af27456 CH |
239 | /** |
240 | * __netdev_alloc_skb - allocate an skbuff for rx on a specific device | |
241 | * @dev: network device to receive on | |
242 | * @length: length to allocate | |
243 | * @gfp_mask: get_free_pages mask, passed to alloc_skb | |
244 | * | |
245 | * Allocate a new &sk_buff and assign it a usage count of one. The | |
246 | * buffer has unspecified headroom built in. Users should allocate | |
247 | * the headroom they think they need without accounting for the | |
248 | * built in space. The built in space is used for optimisations. | |
249 | * | |
250 | * %NULL is returned if there is no free memory. | |
251 | */ | |
252 | struct sk_buff *__netdev_alloc_skb(struct net_device *dev, | |
253 | unsigned int length, gfp_t gfp_mask) | |
254 | { | |
43cb76d9 | 255 | int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1; |
8af27456 CH |
256 | struct sk_buff *skb; |
257 | ||
4ec93edb | 258 | skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, node); |
7b2e497a | 259 | if (likely(skb)) { |
8af27456 | 260 | skb_reserve(skb, NET_SKB_PAD); |
7b2e497a CH |
261 | skb->dev = dev; |
262 | } | |
8af27456 CH |
263 | return skb; |
264 | } | |
1da177e4 | 265 | |
f58518e6 IJ |
266 | /** |
267 | * dev_alloc_skb - allocate an skbuff for receiving | |
268 | * @length: length to allocate | |
269 | * | |
270 | * Allocate a new &sk_buff and assign it a usage count of one. The | |
271 | * buffer has unspecified headroom built in. Users should allocate | |
272 | * the headroom they think they need without accounting for the | |
273 | * built in space. The built in space is used for optimisations. | |
274 | * | |
275 | * %NULL is returned if there is no free memory. Although this function | |
276 | * allocates memory it can be called from an interrupt. | |
277 | */ | |
278 | struct sk_buff *dev_alloc_skb(unsigned int length) | |
279 | { | |
1483b874 DV |
280 | /* |
281 | * There is more code here than it seems: | |
a0f55e0e | 282 | * __dev_alloc_skb is an inline |
1483b874 | 283 | */ |
f58518e6 IJ |
284 | return __dev_alloc_skb(length, GFP_ATOMIC); |
285 | } | |
286 | EXPORT_SYMBOL(dev_alloc_skb); | |
287 | ||
27b437c8 | 288 | static void skb_drop_list(struct sk_buff **listp) |
1da177e4 | 289 | { |
27b437c8 | 290 | struct sk_buff *list = *listp; |
1da177e4 | 291 | |
27b437c8 | 292 | *listp = NULL; |
1da177e4 LT |
293 | |
294 | do { | |
295 | struct sk_buff *this = list; | |
296 | list = list->next; | |
297 | kfree_skb(this); | |
298 | } while (list); | |
299 | } | |
300 | ||
27b437c8 HX |
301 | static inline void skb_drop_fraglist(struct sk_buff *skb) |
302 | { | |
303 | skb_drop_list(&skb_shinfo(skb)->frag_list); | |
304 | } | |
305 | ||
1da177e4 LT |
306 | static void skb_clone_fraglist(struct sk_buff *skb) |
307 | { | |
308 | struct sk_buff *list; | |
309 | ||
310 | for (list = skb_shinfo(skb)->frag_list; list; list = list->next) | |
311 | skb_get(list); | |
312 | } | |
313 | ||
5bba1712 | 314 | static void skb_release_data(struct sk_buff *skb) |
1da177e4 LT |
315 | { |
316 | if (!skb->cloned || | |
317 | !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, | |
318 | &skb_shinfo(skb)->dataref)) { | |
319 | if (skb_shinfo(skb)->nr_frags) { | |
320 | int i; | |
321 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) | |
322 | put_page(skb_shinfo(skb)->frags[i].page); | |
323 | } | |
324 | ||
325 | if (skb_shinfo(skb)->frag_list) | |
326 | skb_drop_fraglist(skb); | |
327 | ||
328 | kfree(skb->head); | |
329 | } | |
330 | } | |
331 | ||
332 | /* | |
333 | * Free an skbuff by memory without cleaning the state. | |
334 | */ | |
2d4baff8 | 335 | static void kfree_skbmem(struct sk_buff *skb) |
1da177e4 | 336 | { |
d179cd12 DM |
337 | struct sk_buff *other; |
338 | atomic_t *fclone_ref; | |
339 | ||
d179cd12 DM |
340 | switch (skb->fclone) { |
341 | case SKB_FCLONE_UNAVAILABLE: | |
342 | kmem_cache_free(skbuff_head_cache, skb); | |
343 | break; | |
344 | ||
345 | case SKB_FCLONE_ORIG: | |
346 | fclone_ref = (atomic_t *) (skb + 2); | |
347 | if (atomic_dec_and_test(fclone_ref)) | |
348 | kmem_cache_free(skbuff_fclone_cache, skb); | |
349 | break; | |
350 | ||
351 | case SKB_FCLONE_CLONE: | |
352 | fclone_ref = (atomic_t *) (skb + 1); | |
353 | other = skb - 1; | |
354 | ||
355 | /* The clone portion is available for | |
356 | * fast-cloning again. | |
357 | */ | |
358 | skb->fclone = SKB_FCLONE_UNAVAILABLE; | |
359 | ||
360 | if (atomic_dec_and_test(fclone_ref)) | |
361 | kmem_cache_free(skbuff_fclone_cache, other); | |
362 | break; | |
3ff50b79 | 363 | } |
1da177e4 LT |
364 | } |
365 | ||
2d4baff8 HX |
366 | /* Free everything but the sk_buff shell. */ |
367 | static void skb_release_all(struct sk_buff *skb) | |
1da177e4 | 368 | { |
1da177e4 LT |
369 | dst_release(skb->dst); |
370 | #ifdef CONFIG_XFRM | |
371 | secpath_put(skb->sp); | |
372 | #endif | |
9c2b3328 SH |
373 | if (skb->destructor) { |
374 | WARN_ON(in_irq()); | |
1da177e4 LT |
375 | skb->destructor(skb); |
376 | } | |
9fb9cbb1 | 377 | #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) |
5f79e0f9 | 378 | nf_conntrack_put(skb->nfct); |
9fb9cbb1 YK |
379 | nf_conntrack_put_reasm(skb->nfct_reasm); |
380 | #endif | |
1da177e4 LT |
381 | #ifdef CONFIG_BRIDGE_NETFILTER |
382 | nf_bridge_put(skb->nf_bridge); | |
383 | #endif | |
1da177e4 LT |
384 | /* XXX: IS this still necessary? - JHS */ |
385 | #ifdef CONFIG_NET_SCHED | |
386 | skb->tc_index = 0; | |
387 | #ifdef CONFIG_NET_CLS_ACT | |
388 | skb->tc_verd = 0; | |
1da177e4 LT |
389 | #endif |
390 | #endif | |
2d4baff8 HX |
391 | skb_release_data(skb); |
392 | } | |
393 | ||
394 | /** | |
395 | * __kfree_skb - private function | |
396 | * @skb: buffer | |
397 | * | |
398 | * Free an sk_buff. Release anything attached to the buffer. | |
399 | * Clean the state. This is an internal helper function. Users should | |
400 | * always call kfree_skb | |
401 | */ | |
1da177e4 | 402 | |
2d4baff8 HX |
403 | void __kfree_skb(struct sk_buff *skb) |
404 | { | |
405 | skb_release_all(skb); | |
1da177e4 LT |
406 | kfree_skbmem(skb); |
407 | } | |
408 |