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> | |
1da177e4 LT |
42 | #include <linux/mm.h> |
43 | #include <linux/interrupt.h> | |
44 | #include <linux/in.h> | |
45 | #include <linux/inet.h> | |
46 | #include <linux/slab.h> | |
47 | #include <linux/netdevice.h> | |
48 | #ifdef CONFIG_NET_CLS_ACT | |
49 | #include <net/pkt_sched.h> | |
50 | #endif | |
51 | #include <linux/string.h> | |
52 | #include <linux/skbuff.h> | |
9c55e01c | 53 | #include <linux/splice.h> |
1da177e4 LT |
54 | #include <linux/cache.h> |
55 | #include <linux/rtnetlink.h> | |
56 | #include <linux/init.h> | |
716ea3a7 | 57 | #include <linux/scatterlist.h> |
1da177e4 LT |
58 | |
59 | #include <net/protocol.h> | |
60 | #include <net/dst.h> | |
61 | #include <net/sock.h> | |
62 | #include <net/checksum.h> | |
63 | #include <net/xfrm.h> | |
64 | ||
65 | #include <asm/uaccess.h> | |
66 | #include <asm/system.h> | |
67 | ||
a1f8e7f7 AV |
68 | #include "kmap_skb.h" |
69 | ||
e18b890b CL |
70 | static struct kmem_cache *skbuff_head_cache __read_mostly; |
71 | static struct kmem_cache *skbuff_fclone_cache __read_mostly; | |
1da177e4 | 72 | |
9c55e01c JA |
73 | static void sock_pipe_buf_release(struct pipe_inode_info *pipe, |
74 | struct pipe_buffer *buf) | |
75 | { | |
8b9d3728 | 76 | put_page(buf->page); |
9c55e01c JA |
77 | } |
78 | ||
79 | static void sock_pipe_buf_get(struct pipe_inode_info *pipe, | |
80 | struct pipe_buffer *buf) | |
81 | { | |
8b9d3728 | 82 | get_page(buf->page); |
9c55e01c JA |
83 | } |
84 | ||
85 | static int sock_pipe_buf_steal(struct pipe_inode_info *pipe, | |
86 | struct pipe_buffer *buf) | |
87 | { | |
88 | return 1; | |
89 | } | |
90 | ||
91 | ||
92 | /* Pipe buffer operations for a socket. */ | |
93 | static struct pipe_buf_operations sock_pipe_buf_ops = { | |
94 | .can_merge = 0, | |
95 | .map = generic_pipe_buf_map, | |
96 | .unmap = generic_pipe_buf_unmap, | |
97 | .confirm = generic_pipe_buf_confirm, | |
98 | .release = sock_pipe_buf_release, | |
99 | .steal = sock_pipe_buf_steal, | |
100 | .get = sock_pipe_buf_get, | |
101 | }; | |
102 | ||
1da177e4 LT |
103 | /* |
104 | * Keep out-of-line to prevent kernel bloat. | |
105 | * __builtin_return_address is not used because it is not always | |
106 | * reliable. | |
107 | */ | |
108 | ||
109 | /** | |
110 | * skb_over_panic - private function | |
111 | * @skb: buffer | |
112 | * @sz: size | |
113 | * @here: address | |
114 | * | |
115 | * Out of line support code for skb_put(). Not user callable. | |
116 | */ | |
117 | void skb_over_panic(struct sk_buff *skb, int sz, void *here) | |
118 | { | |
26095455 | 119 | printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p " |
4305b541 | 120 | "data:%p tail:%#lx end:%#lx dev:%s\n", |
27a884dc | 121 | here, skb->len, sz, skb->head, skb->data, |
4305b541 | 122 | (unsigned long)skb->tail, (unsigned long)skb->end, |
26095455 | 123 | skb->dev ? skb->dev->name : "<NULL>"); |
1da177e4 LT |
124 | BUG(); |
125 | } | |
126 | ||
127 | /** | |
128 | * skb_under_panic - private function | |
129 | * @skb: buffer | |
130 | * @sz: size | |
131 | * @here: address | |
132 | * | |
133 | * Out of line support code for skb_push(). Not user callable. | |
134 | */ | |
135 | ||
136 | void skb_under_panic(struct sk_buff *skb, int sz, void *here) | |
137 | { | |
26095455 | 138 | printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p " |
4305b541 | 139 | "data:%p tail:%#lx end:%#lx dev:%s\n", |
27a884dc | 140 | here, skb->len, sz, skb->head, skb->data, |
4305b541 | 141 | (unsigned long)skb->tail, (unsigned long)skb->end, |
26095455 | 142 | skb->dev ? skb->dev->name : "<NULL>"); |
1da177e4 LT |
143 | BUG(); |
144 | } | |
145 | ||
146 | /* Allocate a new skbuff. We do this ourselves so we can fill in a few | |
147 | * 'private' fields and also do memory statistics to find all the | |
148 | * [BEEP] leaks. | |
149 | * | |
150 | */ | |
151 | ||
152 | /** | |
d179cd12 | 153 | * __alloc_skb - allocate a network buffer |
1da177e4 LT |
154 | * @size: size to allocate |
155 | * @gfp_mask: allocation mask | |
c83c2486 RD |
156 | * @fclone: allocate from fclone cache instead of head cache |
157 | * and allocate a cloned (child) skb | |
b30973f8 | 158 | * @node: numa node to allocate memory on |
1da177e4 LT |
159 | * |
160 | * Allocate a new &sk_buff. The returned buffer has no headroom and a | |
161 | * tail room of size bytes. The object has a reference count of one. | |
162 | * The return is the buffer. On a failure the return is %NULL. | |
163 | * | |
164 | * Buffers may only be allocated from interrupts using a @gfp_mask of | |
165 | * %GFP_ATOMIC. | |
166 | */ | |
dd0fc66f | 167 | struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, |
b30973f8 | 168 | int fclone, int node) |
1da177e4 | 169 | { |
e18b890b | 170 | struct kmem_cache *cache; |
4947d3ef | 171 | struct skb_shared_info *shinfo; |
1da177e4 LT |
172 | struct sk_buff *skb; |
173 | u8 *data; | |
174 | ||
8798b3fb HX |
175 | cache = fclone ? skbuff_fclone_cache : skbuff_head_cache; |
176 | ||
1da177e4 | 177 | /* Get the HEAD */ |
b30973f8 | 178 | skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node); |
1da177e4 LT |
179 | if (!skb) |
180 | goto out; | |
181 | ||
1da177e4 | 182 | size = SKB_DATA_ALIGN(size); |
b30973f8 CH |
183 | data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info), |
184 | gfp_mask, node); | |
1da177e4 LT |
185 | if (!data) |
186 | goto nodata; | |
187 | ||
ca0605a7 | 188 | /* |
c8005785 JB |
189 | * Only clear those fields we need to clear, not those that we will |
190 | * actually initialise below. Hence, don't put any more fields after | |
191 | * the tail pointer in struct sk_buff! | |
ca0605a7 ACM |
192 | */ |
193 | memset(skb, 0, offsetof(struct sk_buff, tail)); | |
1da177e4 LT |
194 | skb->truesize = size + sizeof(struct sk_buff); |
195 | atomic_set(&skb->users, 1); | |
196 | skb->head = data; | |
197 | skb->data = data; | |
27a884dc | 198 | skb_reset_tail_pointer(skb); |
4305b541 | 199 | skb->end = skb->tail + size; |
4947d3ef BL |
200 | /* make sure we initialize shinfo sequentially */ |
201 | shinfo = skb_shinfo(skb); | |
202 | atomic_set(&shinfo->dataref, 1); | |
203 | shinfo->nr_frags = 0; | |
7967168c HX |
204 | shinfo->gso_size = 0; |
205 | shinfo->gso_segs = 0; | |
206 | shinfo->gso_type = 0; | |
4947d3ef BL |
207 | shinfo->ip6_frag_id = 0; |
208 | shinfo->frag_list = NULL; | |
209 | ||
d179cd12 DM |
210 | if (fclone) { |
211 | struct sk_buff *child = skb + 1; | |
212 | atomic_t *fclone_ref = (atomic_t *) (child + 1); | |
1da177e4 | 213 | |
d179cd12 DM |
214 | skb->fclone = SKB_FCLONE_ORIG; |
215 | atomic_set(fclone_ref, 1); | |
216 | ||
217 | child->fclone = SKB_FCLONE_UNAVAILABLE; | |
218 | } | |
1da177e4 LT |
219 | out: |
220 | return skb; | |
221 | nodata: | |
8798b3fb | 222 | kmem_cache_free(cache, skb); |
1da177e4 LT |
223 | skb = NULL; |
224 | goto out; | |
1da177e4 LT |
225 | } |
226 | ||
8af27456 CH |
227 | /** |
228 | * __netdev_alloc_skb - allocate an skbuff for rx on a specific device | |
229 | * @dev: network device to receive on | |
230 | * @length: length to allocate | |
231 | * @gfp_mask: get_free_pages mask, passed to alloc_skb | |
232 | * | |
233 | * Allocate a new &sk_buff and assign it a usage count of one. The | |
234 | * buffer has unspecified headroom built in. Users should allocate | |
235 | * the headroom they think they need without accounting for the | |
236 | * built in space. The built in space is used for optimisations. | |
237 | * | |
238 | * %NULL is returned if there is no free memory. | |
239 | */ | |
240 | struct sk_buff *__netdev_alloc_skb(struct net_device *dev, | |
241 | unsigned int length, gfp_t gfp_mask) | |
242 | { | |
43cb76d9 | 243 | int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1; |
8af27456 CH |
244 | struct sk_buff *skb; |
245 | ||
4ec93edb | 246 | skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, node); |
7b2e497a | 247 | if (likely(skb)) { |
8af27456 | 248 | skb_reserve(skb, NET_SKB_PAD); |
7b2e497a CH |
249 | skb->dev = dev; |
250 | } | |
8af27456 CH |
251 | return skb; |
252 | } | |
1da177e4 | 253 | |
654bed16 PZ |
254 | struct page *__netdev_alloc_page(struct net_device *dev, gfp_t gfp_mask) |
255 | { | |
256 | int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1; | |
257 | struct page *page; | |
258 | ||
259 | page = alloc_pages_node(node, gfp_mask, 0); | |
260 | return page; | |
261 | } | |
262 | EXPORT_SYMBOL(__netdev_alloc_page); | |
263 | ||
264 | void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, | |
265 | int size) | |
266 | { | |
267 | skb_fill_page_desc(skb, i, page, off, size); | |
268 | skb->len += size; | |
269 | skb->data_len += size; | |
270 | skb->truesize += size; | |
271 | } | |
272 | EXPORT_SYMBOL(skb_add_rx_frag); | |
273 | ||
f58518e6 IJ |
274 | /** |
275 | * dev_alloc_skb - allocate an skbuff for receiving | |
276 | * @length: length to allocate | |
277 | * | |
278 | * Allocate a new &sk_buff and assign it a usage count of one. The | |
279 | * buffer has unspecified headroom built in. Users should allocate | |
280 | * the headroom they think they need without accounting for the | |
281 | * built in space. The built in space is used for optimisations. | |
282 | * | |
283 | * %NULL is returned if there is no free memory. Although this function | |
284 | * allocates memory it can be called from an interrupt. | |
285 | */ | |
286 | struct sk_buff *dev_alloc_skb(unsigned int length) | |
287 | { | |
1483b874 DV |
288 | /* |
289 | * There is more code here than it seems: | |
a0f55e0e | 290 | * __dev_alloc_skb is an inline |
1483b874 | 291 | */ |
f58518e6 IJ |
292 | return __dev_alloc_skb(length, GFP_ATOMIC); |
293 | } | |
294 | EXPORT_SYMBOL(dev_alloc_skb); | |
295 | ||
27b437c8 | 296 | static void skb_drop_list(struct sk_buff **listp) |
1da177e4 | 297 | { |
27b437c8 | 298 | struct sk_buff *list = *listp; |
1da177e4 | 299 | |
27b437c8 | 300 | *listp = NULL; |
1da177e4 LT |
301 | |
302 | do { | |
303 | struct sk_buff *this = list; | |
304 | list = list->next; | |
305 | kfree_skb(this); | |
306 | } while (list); | |
307 | } | |
308 | ||
27b437c8 HX |
309 | static inline void skb_drop_fraglist(struct sk_buff *skb) |
310 | { | |
311 | skb_drop_list(&skb_shinfo(skb)->frag_list); | |
312 | } | |
313 | ||
1da177e4 LT |
314 | static void skb_clone_fraglist(struct sk_buff *skb) |
315 | { | |
316 | struct sk_buff *list; | |
317 | ||
318 | for (list = skb_shinfo(skb)->frag_list; list; list = list->next) | |
319 | skb_get(list); | |
320 | } | |
321 | ||
5bba1712 | 322 | static void skb_release_data(struct sk_buff *skb) |
1da177e4 LT |
323 | { |
324 | if (!skb->cloned || | |
325 | !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, | |
326 | &skb_shinfo(skb)->dataref)) { | |
327 | if (skb_shinfo(skb)->nr_frags) { | |
328 | int i; | |
329 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) | |
330 | put_page(skb_shinfo(skb)->frags[i].page); | |
331 | } | |
332 | ||
333 | if (skb_shinfo(skb)->frag_list) | |
334 | skb_drop_fraglist(skb); | |
335 | ||
336 | kfree(skb->head); | |
337 | } | |
338 | } | |
339 | ||
340 | /* | |
341 | * Free an skbuff by memory without cleaning the state. | |
342 | */ | |
2d4baff8 | 343 | static void kfree_skbmem(struct sk_buff *skb) |
1da177e4 | 344 | { |
d179cd12 DM |
345 | struct sk_buff *other; |
346 | atomic_t *fclone_ref; | |
347 | ||
d179cd12 DM |
348 | switch (skb->fclone) { |
349 | case SKB_FCLONE_UNAVAILABLE: | |
350 | kmem_cache_free(skbuff_head_cache, skb); | |
351 | break; | |
352 | ||
353 | case SKB_FCLONE_ORIG: | |
354 | fclone_ref = (atomic_t *) (skb + 2); | |
355 | if (atomic_dec_and_test(fclone_ref)) | |
356 | kmem_cache_free(skbuff_fclone_cache, skb); | |
357 | break; | |
358 | ||
359 | case SKB_FCLONE_CLONE: | |
360 | fclone_ref = (atomic_t *) (skb + 1); | |
361 | other = skb - 1; | |
362 | ||
363 | /* The clone portion is available for | |
364 | * fast-cloning again. | |
365 | */ | |
366 | skb->fclone = SKB_FCLONE_UNAVAILABLE; | |
367 | ||
368 | if (atomic_dec_and_test(fclone_ref)) | |
369 | kmem_cache_free(skbuff_fclone_cache, other); | |
370 | break; | |
3ff50b79 | 371 | } |
1da177e4 LT |
372 | } |
373 | ||
04a4bb55 | 374 | static void skb_release_head_state(struct sk_buff *skb) |
1da177e4 | 375 | { |
1da177e4 LT |
376 | dst_release(skb->dst); |
377 | #ifdef CONFIG_XFRM | |
378 | secpath_put(skb->sp); | |
379 | #endif | |
9c2b3328 SH |
380 | if (skb->destructor) { |
381 | WARN_ON(in_irq()); | |
1da177e4 LT |
382 | skb->destructor(skb); |
383 | } | |
9fb9cbb1 | 384 | #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) |
5f79e0f9 | 385 | nf_conntrack_put(skb->nfct); |
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 | } | |
421 |