[TCP]: Process linger2 timeout consistently.
[linux-2.6-block.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>
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
70static kmem_cache_t *skbuff_head_cache __read_mostly;
71static kmem_cache_t *skbuff_fclone_cache __read_mostly;
1da177e4 72
06825ba3
IM
73/*
74 * lockdep: lock class key used by skb_queue_head_init():
75 */
76struct lock_class_key skb_queue_lock_key;
77
78EXPORT_SYMBOL(skb_queue_lock_key);
79
1da177e4
LT
80/*
81 * Keep out-of-line to prevent kernel bloat.
82 * __builtin_return_address is not used because it is not always
83 * reliable.
84 */
85
86/**
87 * skb_over_panic - private function
88 * @skb: buffer
89 * @sz: size
90 * @here: address
91 *
92 * Out of line support code for skb_put(). Not user callable.
93 */
94void skb_over_panic(struct sk_buff *skb, int sz, void *here)
95{
26095455
PM
96 printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
97 "data:%p tail:%p end:%p dev:%s\n",
98 here, skb->len, sz, skb->head, skb->data, skb->tail, skb->end,
99 skb->dev ? skb->dev->name : "<NULL>");
1da177e4
LT
100 BUG();
101}
102
103/**
104 * skb_under_panic - private function
105 * @skb: buffer
106 * @sz: size
107 * @here: address
108 *
109 * Out of line support code for skb_push(). Not user callable.
110 */
111
112void skb_under_panic(struct sk_buff *skb, int sz, void *here)
113{
26095455
PM
114 printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
115 "data:%p tail:%p end:%p dev:%s\n",
116 here, skb->len, sz, skb->head, skb->data, skb->tail, skb->end,
117 skb->dev ? skb->dev->name : "<NULL>");
1da177e4
LT
118 BUG();
119}
120
dc6de336
DM
121void skb_truesize_bug(struct sk_buff *skb)
122{
123 printk(KERN_ERR "SKB BUG: Invalid truesize (%u) "
124 "len=%u, sizeof(sk_buff)=%Zd\n",
125 skb->truesize, skb->len, sizeof(struct sk_buff));
126}
127EXPORT_SYMBOL(skb_truesize_bug);
128
1da177e4
LT
129/* Allocate a new skbuff. We do this ourselves so we can fill in a few
130 * 'private' fields and also do memory statistics to find all the
131 * [BEEP] leaks.
132 *
133 */
134
135/**
d179cd12 136 * __alloc_skb - allocate a network buffer
1da177e4
LT
137 * @size: size to allocate
138 * @gfp_mask: allocation mask
c83c2486
RD
139 * @fclone: allocate from fclone cache instead of head cache
140 * and allocate a cloned (child) skb
1da177e4
LT
141 *
142 * Allocate a new &sk_buff. The returned buffer has no headroom and a
143 * tail room of size bytes. The object has a reference count of one.
144 * The return is the buffer. On a failure the return is %NULL.
145 *
146 * Buffers may only be allocated from interrupts using a @gfp_mask of
147 * %GFP_ATOMIC.
148 */
dd0fc66f 149struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
d179cd12 150 int fclone)
1da177e4 151{
8798b3fb 152 kmem_cache_t *cache;
4947d3ef 153 struct skb_shared_info *shinfo;
1da177e4
LT
154 struct sk_buff *skb;
155 u8 *data;
156
8798b3fb
HX
157 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
158
1da177e4 159 /* Get the HEAD */
8798b3fb 160 skb = kmem_cache_alloc(cache, gfp_mask & ~__GFP_DMA);
1da177e4
LT
161 if (!skb)
162 goto out;
163
164 /* Get the DATA. Size must match skb_add_mtu(). */
165 size = SKB_DATA_ALIGN(size);
871751e2 166 data = ____kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1da177e4
LT
167 if (!data)
168 goto nodata;
169
170 memset(skb, 0, offsetof(struct sk_buff, truesize));
171 skb->truesize = size + sizeof(struct sk_buff);
172 atomic_set(&skb->users, 1);
173 skb->head = data;
174 skb->data = data;
175 skb->tail = data;
176 skb->end = data + 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;
202}
203
204/**
205 * alloc_skb_from_cache - allocate a network buffer
206 * @cp: kmem_cache from which to allocate the data area
207 * (object size must be big enough for @size bytes + skb overheads)
208 * @size: size to allocate
209 * @gfp_mask: allocation mask
210 *
211 * Allocate a new &sk_buff. The returned buffer has no headroom and
212 * tail room of size bytes. The object has a reference count of one.
213 * The return is the buffer. On a failure the return is %NULL.
214 *
215 * Buffers may only be allocated from interrupts using a @gfp_mask of
216 * %GFP_ATOMIC.
217 */
218struct sk_buff *alloc_skb_from_cache(kmem_cache_t *cp,
86a76caf 219 unsigned int size,
dd0fc66f 220 gfp_t gfp_mask)
1da177e4
LT
221{
222 struct sk_buff *skb;
223 u8 *data;
224
225 /* Get the HEAD */
226 skb = kmem_cache_alloc(skbuff_head_cache,
227 gfp_mask & ~__GFP_DMA);
228 if (!skb)
229 goto out;
230
231 /* Get the DATA. */
232 size = SKB_DATA_ALIGN(size);
233 data = kmem_cache_alloc(cp, gfp_mask);
234 if (!data)
235 goto nodata;
236
237 memset(skb, 0, offsetof(struct sk_buff, truesize));
238 skb->truesize = size + sizeof(struct sk_buff);
239 atomic_set(&skb->users, 1);
240 skb->head = data;
241 skb->data = data;
242 skb->tail = data;
243 skb->end = data + size;
244
245 atomic_set(&(skb_shinfo(skb)->dataref), 1);
246 skb_shinfo(skb)->nr_frags = 0;
7967168c
HX
247 skb_shinfo(skb)->gso_size = 0;
248 skb_shinfo(skb)->gso_segs = 0;
249 skb_shinfo(skb)->gso_type = 0;
1da177e4
LT
250 skb_shinfo(skb)->frag_list = NULL;
251out:
252 return skb;
253nodata:
254 kmem_cache_free(skbuff_head_cache, skb);
255 skb = NULL;
256 goto out;
257}
258
259
27b437c8 260static void skb_drop_list(struct sk_buff **listp)
1da177e4 261{
27b437c8 262 struct sk_buff *list = *listp;
1da177e4 263
27b437c8 264 *listp = NULL;
1da177e4
LT
265
266 do {
267 struct sk_buff *this = list;
268 list = list->next;
269 kfree_skb(this);
270 } while (list);
271}
272
27b437c8
HX
273static inline void skb_drop_fraglist(struct sk_buff *skb)
274{
275 skb_drop_list(&skb_shinfo(skb)->frag_list);
276}
277
1da177e4
LT
278static void skb_clone_fraglist(struct sk_buff *skb)
279{
280 struct sk_buff *list;
281
282 for (list = skb_shinfo(skb)->frag_list; list; list = list->next)
283 skb_get(list);
284}
285
5bba1712 286static void skb_release_data(struct sk_buff *skb)
1da177e4
LT
287{
288 if (!skb->cloned ||
289 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
290 &skb_shinfo(skb)->dataref)) {
291 if (skb_shinfo(skb)->nr_frags) {
292 int i;
293 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
294 put_page(skb_shinfo(skb)->frags[i].page);
295 }
296
297 if (skb_shinfo(skb)->frag_list)
298 skb_drop_fraglist(skb);
299
300 kfree(skb->head);
301 }
302}
303
304/*
305 * Free an skbuff by memory without cleaning the state.
306 */
307void kfree_skbmem(struct sk_buff *skb)
308{
d179cd12
DM
309 struct sk_buff *other;
310 atomic_t *fclone_ref;
311
1da177e4 312 skb_release_data(skb);
d179cd12
DM
313 switch (skb->fclone) {
314 case SKB_FCLONE_UNAVAILABLE:
315 kmem_cache_free(skbuff_head_cache, skb);
316 break;
317
318 case SKB_FCLONE_ORIG:
319 fclone_ref = (atomic_t *) (skb + 2);
320 if (atomic_dec_and_test(fclone_ref))
321 kmem_cache_free(skbuff_fclone_cache, skb);
322 break;
323
324 case SKB_FCLONE_CLONE:
325 fclone_ref = (atomic_t *) (skb + 1);
326 other = skb - 1;
327
328 /* The clone portion is available for
329 * fast-cloning again.
330 */
331 skb->fclone = SKB_FCLONE_UNAVAILABLE;
332
333 if (atomic_dec_and_test(fclone_ref))
334 kmem_cache_free(skbuff_fclone_cache, other);
335 break;
336 };
1da177e4
LT
337}
338
339/**
340 * __kfree_skb - private function
341 * @skb: buffer
342 *
343 * Free an sk_buff. Release anything attached to the buffer.
344 * Clean the state. This is an internal helper function. Users should
345 * always call kfree_skb
346 */
347
348void __kfree_skb(struct sk_buff *skb)
349{
1da177e4
LT
350 dst_release(skb->dst);
351#ifdef CONFIG_XFRM
352 secpath_put(skb->sp);
353#endif
9c2b3328
SH
354 if (skb->destructor) {
355 WARN_ON(in_irq());
1da177e4
LT
356 skb->destructor(skb);
357 }
358#ifdef CONFIG_NETFILTER
359 nf_conntrack_put(skb->nfct);
9fb9cbb1
YK
360#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
361 nf_conntrack_put_reasm(skb->nfct_reasm);
362#endif
1da177e4
LT
363#ifdef CONFIG_BRIDGE_NETFILTER
364 nf_bridge_put(skb->nf_bridge);
365#endif
366#endif
367/* XXX: IS this still necessary? - JHS */
368#ifdef CONFIG_NET_SCHED
369 skb->tc_index = 0;
370#ifdef CONFIG_NET_CLS_ACT
371 skb->tc_verd = 0;
1da177e4
LT
372#endif
373#endif
374
375 kfree_skbmem(skb);
376}
377