]> bbs.cooldavid.org Git - net-next-2.6.git/blame - net/core/skbuff.c
[NET]: Move generic skbuff stuff from XFRM code to generic code
[net-next-2.6.git] / net / core / skbuff.c
CommitLineData
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>
55#include <linux/cache.h>
56#include <linux/rtnetlink.h>
57#include <linux/init.h>
716ea3a7 58#include <linux/scatterlist.h>
1da177e4
LT
59
60#include <net/protocol.h>
61#include <net/dst.h>
62#include <net/sock.h>
63#include <net/checksum.h>
64#include <net/xfrm.h>
65
66#include <asm/uaccess.h>
67#include <asm/system.h>
68
a1f8e7f7
AV
69#include "kmap_skb.h"
70
e18b890b
CL
71static struct kmem_cache *skbuff_head_cache __read_mostly;
72static struct kmem_cache *skbuff_fclone_cache __read_mostly;
1da177e4
LT
73
74/*
75 * Keep out-of-line to prevent kernel bloat.
76 * __builtin_return_address is not used because it is not always
77 * reliable.
78 */
79
80/**
81 * skb_over_panic - private function
82 * @skb: buffer
83 * @sz: size
84 * @here: address
85 *
86 * Out of line support code for skb_put(). Not user callable.
87 */
88void skb_over_panic(struct sk_buff *skb, int sz, void *here)
89{
26095455 90 printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
4305b541 91 "data:%p tail:%#lx end:%#lx dev:%s\n",
27a884dc 92 here, skb->len, sz, skb->head, skb->data,
4305b541 93 (unsigned long)skb->tail, (unsigned long)skb->end,
26095455 94 skb->dev ? skb->dev->name : "<NULL>");
1da177e4
LT
95 BUG();
96}
97
98/**
99 * skb_under_panic - private function
100 * @skb: buffer
101 * @sz: size
102 * @here: address
103 *
104 * Out of line support code for skb_push(). Not user callable.
105 */
106
107void skb_under_panic(struct sk_buff *skb, int sz, void *here)
108{
26095455 109 printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
4305b541 110 "data:%p tail:%#lx end:%#lx dev:%s\n",
27a884dc 111 here, skb->len, sz, skb->head, skb->data,
4305b541 112 (unsigned long)skb->tail, (unsigned long)skb->end,
26095455 113 skb->dev ? skb->dev->name : "<NULL>");
1da177e4
LT
114 BUG();
115}
116
dc6de336
DM
117void skb_truesize_bug(struct sk_buff *skb)
118{
119 printk(KERN_ERR "SKB BUG: Invalid truesize (%u) "
120 "len=%u, sizeof(sk_buff)=%Zd\n",
121 skb->truesize, skb->len, sizeof(struct sk_buff));
122}
123EXPORT_SYMBOL(skb_truesize_bug);
124
1da177e4
LT
125/* Allocate a new skbuff. We do this ourselves so we can fill in a few
126 * 'private' fields and also do memory statistics to find all the
127 * [BEEP] leaks.
128 *
129 */
130
131/**
d179cd12 132 * __alloc_skb - allocate a network buffer
1da177e4
LT
133 * @size: size to allocate
134 * @gfp_mask: allocation mask
c83c2486
RD
135 * @fclone: allocate from fclone cache instead of head cache
136 * and allocate a cloned (child) skb
b30973f8 137 * @node: numa node to allocate memory on
1da177e4
LT
138 *
139 * Allocate a new &sk_buff. The returned buffer has no headroom and a
140 * tail room of size bytes. The object has a reference count of one.
141 * The return is the buffer. On a failure the return is %NULL.
142 *
143 * Buffers may only be allocated from interrupts using a @gfp_mask of
144 * %GFP_ATOMIC.
145 */
dd0fc66f 146struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
b30973f8 147 int fclone, int node)
1da177e4 148{
e18b890b 149 struct kmem_cache *cache;
4947d3ef 150 struct skb_shared_info *shinfo;
1da177e4
LT
151 struct sk_buff *skb;
152 u8 *data;
153
8798b3fb
HX
154 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
155
1da177e4 156 /* Get the HEAD */
b30973f8 157 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
1da177e4
LT
158 if (!skb)
159 goto out;
160
1da177e4 161 size = SKB_DATA_ALIGN(size);
b30973f8
CH
162 data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info),
163 gfp_mask, node);
1da177e4
LT
164 if (!data)
165 goto nodata;
166
ca0605a7
ACM
167 /*
168 * See comment in sk_buff definition, just before the 'tail' member
169 */
170 memset(skb, 0, offsetof(struct sk_buff, tail));
1da177e4
LT
171 skb->truesize = size + sizeof(struct sk_buff);
172 atomic_set(&skb->users, 1);
173 skb->head = data;
174 skb->data = data;
27a884dc 175 skb_reset_tail_pointer(skb);
4305b541 176 skb->end = skb->tail + size;
4947d3ef
BL
177 /* make sure we initialize shinfo sequentially */
178 shinfo = skb_shinfo(skb);
179 atomic_set(&shinfo->dataref, 1);
180 shinfo->nr_frags = 0;
7967168c
HX
181 shinfo->gso_size = 0;
182 shinfo->gso_segs = 0;
183 shinfo->gso_type = 0;
4947d3ef
BL
184 shinfo->ip6_frag_id = 0;
185 shinfo->frag_list = NULL;
186
d179cd12
DM
187 if (fclone) {
188 struct sk_buff *child = skb + 1;
189 atomic_t *fclone_ref = (atomic_t *) (child + 1);
1da177e4 190
d179cd12
DM
191 skb->fclone = SKB_FCLONE_ORIG;
192 atomic_set(fclone_ref, 1);
193
194 child->fclone = SKB_FCLONE_UNAVAILABLE;
195 }
1da177e4
LT
196out:
197 return skb;
198nodata:
8798b3fb 199 kmem_cache_free(cache, skb);
1da177e4
LT
200 skb = NULL;
201 goto out;
1da177e4
LT
202}
203
8af27456
CH
204/**
205 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
206 * @dev: network device to receive on
207 * @length: length to allocate
208 * @gfp_mask: get_free_pages mask, passed to alloc_skb
209 *
210 * Allocate a new &sk_buff and assign it a usage count of one. The
211 * buffer has unspecified headroom built in. Users should allocate
212 * the headroom they think they need without accounting for the
213 * built in space. The built in space is used for optimisations.
214 *
215 * %NULL is returned if there is no free memory.
216 */
217struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
218 unsigned int length, gfp_t gfp_mask)
219{
43cb76d9 220 int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1;
8af27456
CH
221 struct sk_buff *skb;
222
4ec93edb 223 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, node);
7b2e497a 224 if (likely(skb)) {
8af27456 225 skb_reserve(skb, NET_SKB_PAD);
7b2e497a
CH
226 skb->dev = dev;
227 }
8af27456
CH
228 return skb;
229}
1da177e4 230
27b437c8 231static void skb_drop_list(struct sk_buff **listp)
1da177e4 232{
27b437c8 233 struct sk_buff *list = *listp;
1da177e4 234
27b437c8 235 *listp = NULL;
1da177e4
LT
236
237 do {
238 struct sk_buff *this = list;
239 list = list->next;
240 kfree_skb(this);
241 } while (list);
242}
243
27b437c8
HX
244static inline void skb_drop_fraglist(struct sk_buff *skb)
245{
246 skb_drop_list(&skb_shinfo(skb)->frag_list);
247}
248
1da177e4
LT
249static void skb_clone_fraglist(struct sk_buff *skb)
250{
251 struct sk_buff *list;
252
253 for (list = skb_shinfo(skb)->frag_list; list; list = list->next)
254 skb_get(list);
255}
256
5bba1712 257static void skb_release_data(struct sk_buff *skb)
1da177e4
LT
258{
259 if (!skb->cloned ||
260 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
261 &skb_shinfo(skb)->dataref)) {
262 if (skb_shinfo(skb)->nr_frags) {
263 int i;
264 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
265 put_page(skb_shinfo(skb)->frags[i].page);
266 }
267
268 if (skb_shinfo(skb)->frag_list)
269 skb_drop_fraglist(skb);
270
271 kfree(skb->head);
272 }
273}
274
275/*
276 * Free an skbuff by memory without cleaning the state.
277 */
278void kfree_skbmem(struct sk_buff *skb)
279{
d179cd12
DM
280 struct sk_buff *other;
281 atomic_t *fclone_ref;
282
1da177e4 283 skb_release_data(skb);
d179cd12
DM
284 switch (skb->fclone) {
285 case SKB_FCLONE_UNAVAILABLE:
286 kmem_cache_free(skbuff_head_cache, skb);
287 break;
288
289 case SKB_FCLONE_ORIG:
290 fclone_ref = (atomic_t *) (skb + 2);
291 if (atomic_dec_and_test(fclone_ref))
292 kmem_cache_free(skbuff_fclone_cache, skb);
293 break;
294
295 case SKB_FCLONE_CLONE:
296 fclone_ref = (atomic_t *) (skb + 1);
297 other = skb - 1;
298
299 /* The clone portion is available for
300 * fast-cloning again.
301 */
302 skb->fclone = SKB_FCLONE_UNAVAILABLE;
303
304 if (atomic_dec_and_test(fclone_ref))
305 kmem_cache_free(skbuff_fclone_cache, other);
306 break;
307 };
1da177e4
LT
308}
309
310/**
311 * __kfree_skb - private function
312 * @skb: buffer
313 *
314 * Free an sk_buff. Release anything attached to the buffer.
315 * Clean the state. This is an internal helper function. Users should
316 * always call kfree_skb
317 */
318
319void __kfree_skb(struct sk_buff *skb)
320{
1da177e4
LT
321 dst_release(skb->dst);
322#ifdef CONFIG_XFRM
323 secpath_put(skb->sp);
324#endif
9c2b3328
SH
325 if (skb->destructor) {
326 WARN_ON(in_irq());
1da177e4
LT
327 skb->destructor(skb);
328 }
9fb9cbb1 329#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
5f79e0f9 330 nf_conntrack_put(skb->nfct);
9fb9cbb1
YK
331 nf_conntrack_put_reasm(skb->nfct_reasm);
332#endif
1da177e4
LT
333#ifdef CONFIG_BRIDGE_NETFILTER
334 nf_bridge_put(skb->nf_bridge);
335#endif
1da177e4
LT
336/* XXX: IS this still necessary? - JHS */
337#ifdef CONFIG_NET_SCHED
338 skb->tc_index = 0;
339#ifdef CONFIG_NET_CLS_ACT
340 skb->tc_verd = 0;
1da177e4
LT
341#endif
342#endif
343
344 kfree_skbmem(skb);
345}
346