]>
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> | |
44 | #include <linux/sched.h> | |
45 | #include <linux/mm.h> | |
46 | #include <linux/interrupt.h> | |
47 | #include <linux/in.h> | |
48 | #include <linux/inet.h> | |
49 | #include <linux/slab.h> | |
50 | #include <linux/netdevice.h> | |
51 | #ifdef CONFIG_NET_CLS_ACT | |
52 | #include <net/pkt_sched.h> | |
53 | #endif | |
54 | #include <linux/string.h> | |
55 | #include <linux/skbuff.h> | |
56 | #include <linux/cache.h> | |
57 | #include <linux/rtnetlink.h> | |
58 | #include <linux/init.h> | |
59 | #include <linux/highmem.h> | |
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 | ||
ba89966c ED |
70 | static kmem_cache_t *skbuff_head_cache __read_mostly; |
71 | static kmem_cache_t *skbuff_fclone_cache __read_mostly; | |
1da177e4 LT |
72 | |
73 | /* | |
74 | * Keep out-of-line to prevent kernel bloat. | |
75 | * __builtin_return_address is not used because it is not always | |
76 | * reliable. | |
77 | */ | |
78 | ||
79 | /** | |
80 | * skb_over_panic - private function | |
81 | * @skb: buffer | |
82 | * @sz: size | |
83 | * @here: address | |
84 | * | |
85 | * Out of line support code for skb_put(). Not user callable. | |
86 | */ | |
87 | void skb_over_panic(struct sk_buff *skb, int sz, void *here) | |
88 | { | |
26095455 PM |
89 | printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p " |
90 | "data:%p tail:%p end:%p dev:%s\n", | |
91 | here, skb->len, sz, skb->head, skb->data, skb->tail, skb->end, | |
92 | skb->dev ? skb->dev->name : "<NULL>"); | |
1da177e4 LT |
93 | BUG(); |
94 | } | |
95 | ||
96 | /** | |
97 | * skb_under_panic - private function | |
98 | * @skb: buffer | |
99 | * @sz: size | |
100 | * @here: address | |
101 | * | |
102 | * Out of line support code for skb_push(). Not user callable. | |
103 | */ | |
104 | ||
105 | void skb_under_panic(struct sk_buff *skb, int sz, void *here) | |
106 | { | |
26095455 PM |
107 | printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p " |
108 | "data:%p tail:%p end:%p dev:%s\n", | |
109 | here, skb->len, sz, skb->head, skb->data, skb->tail, skb->end, | |
110 | skb->dev ? skb->dev->name : "<NULL>"); | |
1da177e4 LT |
111 | BUG(); |
112 | } | |
113 | ||
dc6de336 DM |
114 | void skb_truesize_bug(struct sk_buff *skb) |
115 | { | |
116 | printk(KERN_ERR "SKB BUG: Invalid truesize (%u) " | |
117 | "len=%u, sizeof(sk_buff)=%Zd\n", | |
118 | skb->truesize, skb->len, sizeof(struct sk_buff)); | |
119 | } | |
120 | EXPORT_SYMBOL(skb_truesize_bug); | |
121 | ||
1da177e4 LT |
122 | /* Allocate a new skbuff. We do this ourselves so we can fill in a few |
123 | * 'private' fields and also do memory statistics to find all the | |
124 | * [BEEP] leaks. | |
125 | * | |
126 | */ | |
127 | ||
128 | /** | |
d179cd12 | 129 | * __alloc_skb - allocate a network buffer |
1da177e4 LT |
130 | * @size: size to allocate |
131 | * @gfp_mask: allocation mask | |
c83c2486 RD |
132 | * @fclone: allocate from fclone cache instead of head cache |
133 | * and allocate a cloned (child) skb | |
1da177e4 LT |
134 | * |
135 | * Allocate a new &sk_buff. The returned buffer has no headroom and a | |
136 | * tail room of size bytes. The object has a reference count of one. | |
137 | * The return is the buffer. On a failure the return is %NULL. | |
138 | * | |
139 | * Buffers may only be allocated from interrupts using a @gfp_mask of | |
140 | * %GFP_ATOMIC. | |
141 | */ | |
dd0fc66f | 142 | struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, |
d179cd12 | 143 | int fclone) |
1da177e4 | 144 | { |
8798b3fb | 145 | kmem_cache_t *cache; |
4947d3ef | 146 | struct skb_shared_info *shinfo; |
1da177e4 LT |
147 | struct sk_buff *skb; |
148 | u8 *data; | |
149 | ||
8798b3fb HX |
150 | cache = fclone ? skbuff_fclone_cache : skbuff_head_cache; |
151 | ||
1da177e4 | 152 | /* Get the HEAD */ |
8798b3fb | 153 | skb = kmem_cache_alloc(cache, gfp_mask & ~__GFP_DMA); |
1da177e4 LT |
154 | if (!skb) |
155 | goto out; | |
156 | ||
157 | /* Get the DATA. Size must match skb_add_mtu(). */ | |
158 | size = SKB_DATA_ALIGN(size); | |
871751e2 | 159 | data = ____kmalloc(size + sizeof(struct skb_shared_info), gfp_mask); |
1da177e4 LT |
160 | if (!data) |
161 | goto nodata; | |
162 | ||
163 | memset(skb, 0, offsetof(struct sk_buff, truesize)); | |
164 | skb->truesize = size + sizeof(struct sk_buff); | |
165 | atomic_set(&skb->users, 1); | |
166 | skb->head = data; | |
167 | skb->data = data; | |
168 | skb->tail = data; | |
169 | skb->end = data + size; | |
4947d3ef BL |
170 | /* make sure we initialize shinfo sequentially */ |
171 | shinfo = skb_shinfo(skb); | |
172 | atomic_set(&shinfo->dataref, 1); | |
173 | shinfo->nr_frags = 0; | |
7967168c HX |
174 | shinfo->gso_size = 0; |
175 | shinfo->gso_segs = 0; | |
176 | shinfo->gso_type = 0; | |
4947d3ef BL |
177 | shinfo->ip6_frag_id = 0; |
178 | shinfo->frag_list = NULL; | |
179 | ||
d179cd12 DM |
180 | if (fclone) { |
181 | struct sk_buff *child = skb + 1; | |
182 | atomic_t *fclone_ref = (atomic_t *) (child + 1); | |
1da177e4 | 183 | |
d179cd12 DM |
184 | skb->fclone = SKB_FCLONE_ORIG; |
185 | atomic_set(fclone_ref, 1); | |
186 | ||
187 | child->fclone = SKB_FCLONE_UNAVAILABLE; | |
188 | } | |
1da177e4 LT |
189 | out: |
190 | return skb; | |
191 | nodata: | |
8798b3fb | 192 | kmem_cache_free(cache, skb); |
1da177e4 LT |
193 | skb = NULL; |
194 | goto out; | |
195 | } | |
196 | ||
197 | /** | |
198 | * alloc_skb_from_cache - allocate a network buffer | |
199 | * @cp: kmem_cache from which to allocate the data area | |
200 | * (object size must be big enough for @size bytes + skb overheads) | |
201 | * @size: size to allocate | |
202 | * @gfp_mask: allocation mask | |
203 | * | |
204 | * Allocate a new &sk_buff. The returned buffer has no headroom and | |
205 | * tail room of size bytes. The object has a reference count of one. | |
206 | * The return is the buffer. On a failure the return is %NULL. | |
207 | * | |
208 | * Buffers may only be allocated from interrupts using a @gfp_mask of | |
209 | * %GFP_ATOMIC. | |
210 | */ | |
211 | struct sk_buff *alloc_skb_from_cache(kmem_cache_t *cp, | |
86a76caf | 212 | unsigned int size, |
dd0fc66f | 213 | gfp_t gfp_mask) |
1da177e4 LT |
214 | { |
215 | struct sk_buff *skb; | |
216 | u8 *data; | |
217 | ||
218 | /* Get the HEAD */ | |
219 | skb = kmem_cache_alloc(skbuff_head_cache, | |
220 | gfp_mask & ~__GFP_DMA); | |
221 | if (!skb) | |
222 | goto out; | |
223 | ||
224 | /* Get the DATA. */ | |
225 | size = SKB_DATA_ALIGN(size); | |
226 | data = kmem_cache_alloc(cp, gfp_mask); | |
227 | if (!data) | |
228 | goto nodata; | |
229 | ||
230 | memset(skb, 0, offsetof(struct sk_buff, truesize)); | |
231 | skb->truesize = size + sizeof(struct sk_buff); | |
232 | atomic_set(&skb->users, 1); | |
233 | skb->head = data; | |
234 | skb->data = data; | |
235 | skb->tail = data; | |
236 | skb->end = data + size; | |
237 | ||
238 | atomic_set(&(skb_shinfo(skb)->dataref), 1); | |
239 | skb_shinfo(skb)->nr_frags = 0; | |
7967168c HX |
240 | skb_shinfo(skb)->gso_size = 0; |
241 | skb_shinfo(skb)->gso_segs = 0; | |
242 | skb_shinfo(skb)->gso_type = 0; | |
1da177e4 LT |
243 | skb_shinfo(skb)->frag_list = NULL; |
244 | out: | |
245 | return skb; | |
246 | nodata: | |
247 | kmem_cache_free(skbuff_head_cache, skb); | |
248 | skb = NULL; | |
249 | goto out; | |
250 | } | |
251 | ||
8af27456 CH |
252 | /** |
253 | * __netdev_alloc_skb - allocate an skbuff for rx on a specific device | |
254 | * @dev: network device to receive on | |
255 | * @length: length to allocate | |
256 | * @gfp_mask: get_free_pages mask, passed to alloc_skb | |
257 | * | |
258 | * Allocate a new &sk_buff and assign it a usage count of one. The | |
259 | * buffer has unspecified headroom built in. Users should allocate | |
260 | * the headroom they think they need without accounting for the | |
261 | * built in space. The built in space is used for optimisations. | |
262 | * | |
263 | * %NULL is returned if there is no free memory. | |
264 | */ | |
265 | struct sk_buff *__netdev_alloc_skb(struct net_device *dev, | |
266 | unsigned int length, gfp_t gfp_mask) | |
267 | { | |
268 | struct sk_buff *skb; | |
269 | ||
270 | skb = alloc_skb(length + NET_SKB_PAD, gfp_mask); | |
271 | if (likely(skb)) | |
272 | skb_reserve(skb, NET_SKB_PAD); | |
273 | return skb; | |
274 | } | |
1da177e4 | 275 | |
27b437c8 | 276 | static void skb_drop_list(struct sk_buff **listp) |
1da177e4 | 277 | { |
27b437c8 | 278 | struct sk_buff *list = *listp; |
1da177e4 | 279 | |
27b437c8 | 280 | *listp = NULL; |
1da177e4 LT |
281 | |
282 | do { | |
283 | struct sk_buff *this = list; | |
284 | list = list->next; | |
285 | kfree_skb(this); | |
286 | } while (list); | |
287 | } | |
288 | ||
27b437c8 HX |
289 | static inline void skb_drop_fraglist(struct sk_buff *skb) |
290 | { | |
291 | skb_drop_list(&skb_shinfo(skb)->frag_list); | |
292 | } | |
293 | ||
1da177e4 LT |
294 | static void skb_clone_fraglist(struct sk_buff *skb) |
295 | { | |
296 | struct sk_buff *list; | |
297 | ||
298 | for (list = skb_shinfo(skb)->frag_list; list; list = list->next) | |
299 | skb_get(list); | |
300 | } | |
301 | ||
5bba1712 | 302 | static void skb_release_data(struct sk_buff *skb) |
1da177e4 LT |
303 | { |
304 | if (!skb->cloned || | |
305 | !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, | |
306 | &skb_shinfo(skb)->dataref)) { | |
307 | if (skb_shinfo(skb)->nr_frags) { | |
308 | int i; | |
309 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) | |
310 | put_page(skb_shinfo(skb)->frags[i].page); | |
311 | } | |
312 | ||
313 | if (skb_shinfo(skb)->frag_list) | |
314 | skb_drop_fraglist(skb); | |
315 | ||
316 | kfree(skb->head); | |
317 | } | |
318 | } | |
319 | ||
320 | /* | |
321 | * Free an skbuff by memory without cleaning the state. | |
322 | */ | |
323 | void kfree_skbmem(struct sk_buff *skb) | |
324 | { | |
d179cd12 DM |
325 | struct sk_buff *other; |
326 | atomic_t *fclone_ref; | |
327 | ||
1da177e4 | 328 | skb_release_data(skb); |
d179cd12 DM |
329 | switch (skb->fclone) { |
330 | case SKB_FCLONE_UNAVAILABLE: | |
331 | kmem_cache_free(skbuff_head_cache, skb); | |
332 | break; | |
333 | ||
334 | case SKB_FCLONE_ORIG: | |
335 | fclone_ref = (atomic_t *) (skb + 2); | |
336 | if (atomic_dec_and_test(fclone_ref)) | |
337 | kmem_cache_free(skbuff_fclone_cache, skb); | |
338 | break; | |
339 | ||
340 | case SKB_FCLONE_CLONE: | |
341 | fclone_ref = (atomic_t *) (skb + 1); | |
342 | other = skb - 1; | |
343 | ||
344 | /* The clone portion is available for | |
345 | * fast-cloning again. | |
346 | */ | |
347 | skb->fclone = SKB_FCLONE_UNAVAILABLE; | |
348 | ||
349 | if (atomic_dec_and_test(fclone_ref)) | |
350 | kmem_cache_free(skbuff_fclone_cache, other); | |
351 | break; | |
352 | }; | |
1da177e4 LT |
353 | } |
354 | ||
355 | /** | |
356 | * __kfree_skb - private function | |
357 | * @skb: buffer | |
358 | * | |
359 | * Free an sk_buff. Release anything attached to the buffer. | |
360 | * Clean the state. This is an internal helper function. Users should | |
361 | * always call kfree_skb | |
362 | */ | |
363 | ||
364 | void __kfree_skb(struct sk_buff *skb) | |
365 | { | |
1da177e4 LT |
366 | dst_release(skb->dst); |
367 | #ifdef CONFIG_XFRM | |
368 | secpath_put(skb->sp); | |
369 | #endif | |
9c2b3328 SH |
370 | if (skb->destructor) { |
371 | WARN_ON(in_irq()); | |
1da177e4 LT |
372 | skb->destructor(skb); |
373 | } | |
374 | #ifdef CONFIG_NETFILTER | |
375 | nf_conntrack_put(skb->nfct); | |
9fb9cbb1 YK |
376 | #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) |
377 | nf_conntrack_put_reasm(skb->nfct_reasm); | |
378 | #endif | |
1da177e4 LT |
379 | #ifdef CONFIG_BRIDGE_NETFILTER |
380 | nf_bridge_put(skb->nf_bridge); | |
381 | #endif | |
382 | #endif | |
383 | /* XXX: IS this still necessary? - JHS */ | |
384 | #ifdef CONFIG_NET_SCHED | |
385 | skb->tc_index = 0; | |
386 | #ifdef CONFIG_NET_CLS_ACT | |
387 | skb->tc_verd = 0; | |
1da177e4 LT |
388 | #endif |
389 | #endif | |
390 | ||
391 | kfree_skbmem(skb); | |
392 | } | |
393 | ||