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 | ||
e005d193 JP |
39 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
40 | ||
1da177e4 LT |
41 | #include <linux/module.h> |
42 | #include <linux/types.h> | |
43 | #include <linux/kernel.h> | |
fe55f6d5 | 44 | #include <linux/kmemcheck.h> |
1da177e4 LT |
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> | |
3fb03b59 FW |
50 | #include <linux/tcp.h> |
51 | #include <linux/udp.h> | |
1da177e4 LT |
52 | #include <linux/netdevice.h> |
53 | #ifdef CONFIG_NET_CLS_ACT | |
54 | #include <net/pkt_sched.h> | |
55 | #endif | |
56 | #include <linux/string.h> | |
57 | #include <linux/skbuff.h> | |
9c55e01c | 58 | #include <linux/splice.h> |
1da177e4 LT |
59 | #include <linux/cache.h> |
60 | #include <linux/rtnetlink.h> | |
61 | #include <linux/init.h> | |
716ea3a7 | 62 | #include <linux/scatterlist.h> |
ac45f602 | 63 | #include <linux/errqueue.h> |
268bb0ce | 64 | #include <linux/prefetch.h> |
1da177e4 LT |
65 | |
66 | #include <net/protocol.h> | |
67 | #include <net/dst.h> | |
68 | #include <net/sock.h> | |
69 | #include <net/checksum.h> | |
70 | #include <net/xfrm.h> | |
71 | ||
72 | #include <asm/uaccess.h> | |
ad8d75ff | 73 | #include <trace/events/skb.h> |
51c56b00 | 74 | #include <linux/highmem.h> |
a1f8e7f7 | 75 | |
d7e8883c | 76 | struct kmem_cache *skbuff_head_cache __read_mostly; |
e18b890b | 77 | static struct kmem_cache *skbuff_fclone_cache __read_mostly; |
1da177e4 | 78 | |
1da177e4 | 79 | /** |
f05de73b JS |
80 | * skb_panic - private function for out-of-line support |
81 | * @skb: buffer | |
82 | * @sz: size | |
83 | * @addr: address | |
99d5851e | 84 | * @msg: skb_over_panic or skb_under_panic |
1da177e4 | 85 | * |
f05de73b JS |
86 | * Out-of-line support for skb_put() and skb_push(). |
87 | * Called via the wrapper skb_over_panic() or skb_under_panic(). | |
88 | * Keep out of line to prevent kernel bloat. | |
89 | * __builtin_return_address is not used because it is not always reliable. | |
1da177e4 | 90 | */ |
f05de73b | 91 | static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr, |
99d5851e | 92 | const char msg[]) |
1da177e4 | 93 | { |
e005d193 | 94 | pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n", |
99d5851e | 95 | msg, addr, skb->len, sz, skb->head, skb->data, |
e005d193 JP |
96 | (unsigned long)skb->tail, (unsigned long)skb->end, |
97 | skb->dev ? skb->dev->name : "<NULL>"); | |
1da177e4 LT |
98 | BUG(); |
99 | } | |
100 | ||
f05de73b | 101 | static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr) |
1da177e4 | 102 | { |
f05de73b | 103 | skb_panic(skb, sz, addr, __func__); |
1da177e4 LT |
104 | } |
105 | ||
f05de73b JS |
106 | static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr) |
107 | { | |
108 | skb_panic(skb, sz, addr, __func__); | |
109 | } | |
c93bdd0e MG |
110 | |
111 | /* | |
112 | * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells | |
113 | * the caller if emergency pfmemalloc reserves are being used. If it is and | |
114 | * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves | |
115 | * may be used. Otherwise, the packet data may be discarded until enough | |
116 | * memory is free | |
117 | */ | |
118 | #define kmalloc_reserve(size, gfp, node, pfmemalloc) \ | |
119 | __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc) | |
61c5e88a | 120 | |
121 | static void *__kmalloc_reserve(size_t size, gfp_t flags, int node, | |
122 | unsigned long ip, bool *pfmemalloc) | |
c93bdd0e MG |
123 | { |
124 | void *obj; | |
125 | bool ret_pfmemalloc = false; | |
126 | ||
127 | /* | |
128 | * Try a regular allocation, when that fails and we're not entitled | |
129 | * to the reserves, fail. | |
130 | */ | |
131 | obj = kmalloc_node_track_caller(size, | |
132 | flags | __GFP_NOMEMALLOC | __GFP_NOWARN, | |
133 | node); | |
134 | if (obj || !(gfp_pfmemalloc_allowed(flags))) | |
135 | goto out; | |
136 | ||
137 | /* Try again but now we are using pfmemalloc reserves */ | |
138 | ret_pfmemalloc = true; | |
139 | obj = kmalloc_node_track_caller(size, flags, node); | |
140 | ||
141 | out: | |
142 | if (pfmemalloc) | |
143 | *pfmemalloc = ret_pfmemalloc; | |
144 | ||
145 | return obj; | |
146 | } | |
147 | ||
1da177e4 LT |
148 | /* Allocate a new skbuff. We do this ourselves so we can fill in a few |
149 | * 'private' fields and also do memory statistics to find all the | |
150 | * [BEEP] leaks. | |
151 | * | |
152 | */ | |
153 | ||
0ebd0ac5 PM |
154 | struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node) |
155 | { | |
156 | struct sk_buff *skb; | |
157 | ||
158 | /* Get the HEAD */ | |
159 | skb = kmem_cache_alloc_node(skbuff_head_cache, | |
160 | gfp_mask & ~__GFP_DMA, node); | |
161 | if (!skb) | |
162 | goto out; | |
163 | ||
164 | /* | |
165 | * Only clear those fields we need to clear, not those that we will | |
166 | * actually initialise below. Hence, don't put any more fields after | |
167 | * the tail pointer in struct sk_buff! | |
168 | */ | |
169 | memset(skb, 0, offsetof(struct sk_buff, tail)); | |
5e71d9d7 | 170 | skb->head = NULL; |
0ebd0ac5 PM |
171 | skb->truesize = sizeof(struct sk_buff); |
172 | atomic_set(&skb->users, 1); | |
173 | ||
174 | #ifdef NET_SKBUFF_DATA_USES_OFFSET | |
175 | skb->mac_header = ~0U; | |
176 | #endif | |
177 | out: | |
178 | return skb; | |
179 | } | |
180 | ||
1da177e4 | 181 | /** |
d179cd12 | 182 | * __alloc_skb - allocate a network buffer |
1da177e4 LT |
183 | * @size: size to allocate |
184 | * @gfp_mask: allocation mask | |
c93bdd0e MG |
185 | * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache |
186 | * instead of head cache and allocate a cloned (child) skb. | |
187 | * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for | |
188 | * allocations in case the data is required for writeback | |
b30973f8 | 189 | * @node: numa node to allocate memory on |
1da177e4 LT |
190 | * |
191 | * Allocate a new &sk_buff. The returned buffer has no headroom and a | |
94b6042c BH |
192 | * tail room of at least size bytes. The object has a reference count |
193 | * of one. The return is the buffer. On a failure the return is %NULL. | |
1da177e4 LT |
194 | * |
195 | * Buffers may only be allocated from interrupts using a @gfp_mask of | |
196 | * %GFP_ATOMIC. | |
197 | */ | |
dd0fc66f | 198 | struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, |
c93bdd0e | 199 | int flags, int node) |
1da177e4 | 200 | { |
e18b890b | 201 | struct kmem_cache *cache; |
4947d3ef | 202 | struct skb_shared_info *shinfo; |
1da177e4 LT |
203 | struct sk_buff *skb; |
204 | u8 *data; | |
c93bdd0e | 205 | bool pfmemalloc; |
1da177e4 | 206 | |
c93bdd0e MG |
207 | cache = (flags & SKB_ALLOC_FCLONE) |
208 | ? skbuff_fclone_cache : skbuff_head_cache; | |
209 | ||
210 | if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX)) | |
211 | gfp_mask |= __GFP_MEMALLOC; | |
8798b3fb | 212 | |
1da177e4 | 213 | /* Get the HEAD */ |
b30973f8 | 214 | skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node); |
1da177e4 LT |
215 | if (!skb) |
216 | goto out; | |
ec7d2f2c | 217 | prefetchw(skb); |
1da177e4 | 218 | |
87fb4b7b ED |
219 | /* We do our best to align skb_shared_info on a separate cache |
220 | * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives | |
221 | * aligned memory blocks, unless SLUB/SLAB debug is enabled. | |
222 | * Both skb->head and skb_shared_info are cache line aligned. | |
223 | */ | |
bc417e30 | 224 | size = SKB_DATA_ALIGN(size); |
87fb4b7b | 225 | size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); |
c93bdd0e | 226 | data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc); |
1da177e4 LT |
227 | if (!data) |
228 | goto nodata; | |
87fb4b7b ED |
229 | /* kmalloc(size) might give us more room than requested. |
230 | * Put skb_shared_info exactly at the end of allocated zone, | |
231 | * to allow max possible filling before reallocation. | |
232 | */ | |
233 | size = SKB_WITH_OVERHEAD(ksize(data)); | |
ec7d2f2c | 234 | prefetchw(data + size); |
1da177e4 | 235 | |
ca0605a7 | 236 | /* |
c8005785 JB |
237 | * Only clear those fields we need to clear, not those that we will |
238 | * actually initialise below. Hence, don't put any more fields after | |
239 | * the tail pointer in struct sk_buff! | |
ca0605a7 ACM |
240 | */ |
241 | memset(skb, 0, offsetof(struct sk_buff, tail)); | |
87fb4b7b ED |
242 | /* Account for allocated memory : skb + skb->head */ |
243 | skb->truesize = SKB_TRUESIZE(size); | |
c93bdd0e | 244 | skb->pfmemalloc = pfmemalloc; |
1da177e4 LT |
245 | atomic_set(&skb->users, 1); |
246 | skb->head = data; | |
247 | skb->data = data; | |
27a884dc | 248 | skb_reset_tail_pointer(skb); |
4305b541 | 249 | skb->end = skb->tail + size; |
19633e12 SH |
250 | #ifdef NET_SKBUFF_DATA_USES_OFFSET |
251 | skb->mac_header = ~0U; | |
fda55eca | 252 | skb->transport_header = ~0U; |
19633e12 SH |
253 | #endif |
254 | ||
4947d3ef BL |
255 | /* make sure we initialize shinfo sequentially */ |
256 | shinfo = skb_shinfo(skb); | |
ec7d2f2c | 257 | memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); |
4947d3ef | 258 | atomic_set(&shinfo->dataref, 1); |
c2aa3665 | 259 | kmemcheck_annotate_variable(shinfo->destructor_arg); |
4947d3ef | 260 | |
c93bdd0e | 261 | if (flags & SKB_ALLOC_FCLONE) { |
d179cd12 DM |
262 | struct sk_buff *child = skb + 1; |
263 | atomic_t *fclone_ref = (atomic_t *) (child + 1); | |
1da177e4 | 264 | |
fe55f6d5 VN |
265 | kmemcheck_annotate_bitfield(child, flags1); |
266 | kmemcheck_annotate_bitfield(child, flags2); | |
d179cd12 DM |
267 | skb->fclone = SKB_FCLONE_ORIG; |
268 | atomic_set(fclone_ref, 1); | |
269 | ||
270 | child->fclone = SKB_FCLONE_UNAVAILABLE; | |
c93bdd0e | 271 | child->pfmemalloc = pfmemalloc; |
d179cd12 | 272 | } |
1da177e4 LT |
273 | out: |
274 | return skb; | |
275 | nodata: | |
8798b3fb | 276 | kmem_cache_free(cache, skb); |
1da177e4 LT |
277 | skb = NULL; |
278 | goto out; | |
1da177e4 | 279 | } |
b4ac530f | 280 | EXPORT_SYMBOL(__alloc_skb); |
1da177e4 | 281 | |
b2b5ce9d ED |
282 | /** |
283 | * build_skb - build a network buffer | |
284 | * @data: data buffer provided by caller | |
d3836f21 | 285 | * @frag_size: size of fragment, or 0 if head was kmalloced |
b2b5ce9d ED |
286 | * |
287 | * Allocate a new &sk_buff. Caller provides space holding head and | |
288 | * skb_shared_info. @data must have been allocated by kmalloc() | |
289 | * The return is the new skb buffer. | |
290 | * On a failure the return is %NULL, and @data is not freed. | |
291 | * Notes : | |
292 | * Before IO, driver allocates only data buffer where NIC put incoming frame | |
293 | * Driver should add room at head (NET_SKB_PAD) and | |
294 | * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info)) | |
295 | * After IO, driver calls build_skb(), to allocate sk_buff and populate it | |
296 | * before giving packet to stack. | |
297 | * RX rings only contains data buffers, not full skbs. | |
298 | */ | |
d3836f21 | 299 | struct sk_buff *build_skb(void *data, unsigned int frag_size) |
b2b5ce9d ED |
300 | { |
301 | struct skb_shared_info *shinfo; | |
302 | struct sk_buff *skb; | |
d3836f21 | 303 | unsigned int size = frag_size ? : ksize(data); |
b2b5ce9d ED |
304 | |
305 | skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC); | |
306 | if (!skb) | |
307 | return NULL; | |
308 | ||
d3836f21 | 309 | size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); |
b2b5ce9d ED |
310 | |
311 | memset(skb, 0, offsetof(struct sk_buff, tail)); | |
312 | skb->truesize = SKB_TRUESIZE(size); | |
d3836f21 | 313 | skb->head_frag = frag_size != 0; |
b2b5ce9d ED |
314 | atomic_set(&skb->users, 1); |
315 | skb->head = data; | |
316 | skb->data = data; | |
317 | skb_reset_tail_pointer(skb); | |
318 | skb->end = skb->tail + size; | |
319 | #ifdef NET_SKBUFF_DATA_USES_OFFSET | |
320 | skb->mac_header = ~0U; | |
fda55eca | 321 | skb->transport_header = ~0U; |
b2b5ce9d ED |
322 | #endif |
323 | ||
324 | /* make sure we initialize shinfo sequentially */ | |
325 | shinfo = skb_shinfo(skb); | |
326 | memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); | |
327 | atomic_set(&shinfo->dataref, 1); | |
328 | kmemcheck_annotate_variable(shinfo->destructor_arg); | |
329 | ||
330 | return skb; | |
331 | } | |
332 | EXPORT_SYMBOL(build_skb); | |
333 | ||
a1c7fff7 | 334 | struct netdev_alloc_cache { |
69b08f62 ED |
335 | struct page_frag frag; |
336 | /* we maintain a pagecount bias, so that we dont dirty cache line | |
337 | * containing page->_count every time we allocate a fragment. | |
338 | */ | |
339 | unsigned int pagecnt_bias; | |
a1c7fff7 ED |
340 | }; |
341 | static DEFINE_PER_CPU(struct netdev_alloc_cache, netdev_alloc_cache); | |
342 | ||
c93bdd0e | 343 | static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask) |
6f532612 ED |
344 | { |
345 | struct netdev_alloc_cache *nc; | |
346 | void *data = NULL; | |
69b08f62 | 347 | int order; |
6f532612 ED |
348 | unsigned long flags; |
349 | ||
350 | local_irq_save(flags); | |
351 | nc = &__get_cpu_var(netdev_alloc_cache); | |
69b08f62 | 352 | if (unlikely(!nc->frag.page)) { |
6f532612 | 353 | refill: |
69b08f62 ED |
354 | for (order = NETDEV_FRAG_PAGE_MAX_ORDER; ;) { |
355 | gfp_t gfp = gfp_mask; | |
356 | ||
357 | if (order) | |
358 | gfp |= __GFP_COMP | __GFP_NOWARN; | |
359 | nc->frag.page = alloc_pages(gfp, order); | |
360 | if (likely(nc->frag.page)) | |
361 | break; | |
362 | if (--order < 0) | |
363 | goto end; | |
364 | } | |
365 | nc->frag.size = PAGE_SIZE << order; | |
540eb7bf | 366 | recycle: |
69b08f62 ED |
367 | atomic_set(&nc->frag.page->_count, NETDEV_PAGECNT_MAX_BIAS); |
368 | nc->pagecnt_bias = NETDEV_PAGECNT_MAX_BIAS; | |
369 | nc->frag.offset = 0; | |
6f532612 | 370 | } |
540eb7bf | 371 | |
69b08f62 | 372 | if (nc->frag.offset + fragsz > nc->frag.size) { |
540eb7bf | 373 | /* avoid unnecessary locked operations if possible */ |
69b08f62 ED |
374 | if ((atomic_read(&nc->frag.page->_count) == nc->pagecnt_bias) || |
375 | atomic_sub_and_test(nc->pagecnt_bias, &nc->frag.page->_count)) | |
540eb7bf AD |
376 | goto recycle; |
377 | goto refill; | |
6f532612 | 378 | } |
540eb7bf | 379 | |
69b08f62 ED |
380 | data = page_address(nc->frag.page) + nc->frag.offset; |
381 | nc->frag.offset += fragsz; | |
540eb7bf AD |
382 | nc->pagecnt_bias--; |
383 | end: | |
6f532612 ED |
384 | local_irq_restore(flags); |
385 | return data; | |
386 | } | |
c93bdd0e MG |
387 | |
388 | /** | |
389 | * netdev_alloc_frag - allocate a page fragment | |
390 | * @fragsz: fragment size | |
391 | * | |
392 | * Allocates a frag from a page for receive buffer. | |
393 | * Uses GFP_ATOMIC allocations. | |
394 | */ | |
395 | void *netdev_alloc_frag(unsigned int fragsz) | |
396 | { | |
397 | return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD); | |
398 | } | |
6f532612 ED |
399 | EXPORT_SYMBOL(netdev_alloc_frag); |
400 | ||
8af27456 CH |
401 | /** |
402 | * __netdev_alloc_skb - allocate an skbuff for rx on a specific device | |
403 | * @dev: network device to receive on | |
404 | * @length: length to allocate | |
405 | * @gfp_mask: get_free_pages mask, passed to alloc_skb | |
406 | * | |
407 | * Allocate a new &sk_buff and assign it a usage count of one. The | |
408 | * buffer has unspecified headroom built in. Users should allocate | |
409 | * the headroom they think they need without accounting for the | |
410 | * built in space. The built in space is used for optimisations. | |
411 | * | |
412 | * %NULL is returned if there is no free memory. | |
413 | */ | |
414 | struct sk_buff *__netdev_alloc_skb(struct net_device *dev, | |
6f532612 | 415 | unsigned int length, gfp_t gfp_mask) |
8af27456 | 416 | { |
6f532612 | 417 | struct sk_buff *skb = NULL; |
a1c7fff7 ED |
418 | unsigned int fragsz = SKB_DATA_ALIGN(length + NET_SKB_PAD) + |
419 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
420 | ||
310e158c | 421 | if (fragsz <= PAGE_SIZE && !(gfp_mask & (__GFP_WAIT | GFP_DMA))) { |
c93bdd0e MG |
422 | void *data; |
423 | ||
424 | if (sk_memalloc_socks()) | |
425 | gfp_mask |= __GFP_MEMALLOC; | |
426 | ||
427 | data = __netdev_alloc_frag(fragsz, gfp_mask); | |
a1c7fff7 | 428 | |
6f532612 ED |
429 | if (likely(data)) { |
430 | skb = build_skb(data, fragsz); | |
431 | if (unlikely(!skb)) | |
432 | put_page(virt_to_head_page(data)); | |
a1c7fff7 | 433 | } |
a1c7fff7 | 434 | } else { |
c93bdd0e MG |
435 | skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, |
436 | SKB_ALLOC_RX, NUMA_NO_NODE); | |
a1c7fff7 | 437 | } |
7b2e497a | 438 | if (likely(skb)) { |
8af27456 | 439 | skb_reserve(skb, NET_SKB_PAD); |
7b2e497a CH |
440 | skb->dev = dev; |
441 | } | |
8af27456 CH |
442 | return skb; |
443 | } | |
b4ac530f | 444 | EXPORT_SYMBOL(__netdev_alloc_skb); |
1da177e4 | 445 | |
654bed16 | 446 | void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, |
50269e19 | 447 | int size, unsigned int truesize) |
654bed16 PZ |
448 | { |
449 | skb_fill_page_desc(skb, i, page, off, size); | |
450 | skb->len += size; | |
451 | skb->data_len += size; | |
50269e19 | 452 | skb->truesize += truesize; |
654bed16 PZ |
453 | } |
454 | EXPORT_SYMBOL(skb_add_rx_frag); | |
455 | ||
27b437c8 | 456 | static void skb_drop_list(struct sk_buff **listp) |
1da177e4 | 457 | { |
bd8a7036 | 458 | kfree_skb_list(*listp); |
27b437c8 | 459 | *listp = NULL; |
1da177e4 LT |
460 | } |
461 | ||
27b437c8 HX |
462 | static inline void skb_drop_fraglist(struct sk_buff *skb) |
463 | { | |
464 | skb_drop_list(&skb_shinfo(skb)->frag_list); | |
465 | } | |
466 | ||
1da177e4 LT |
467 | static void skb_clone_fraglist(struct sk_buff *skb) |
468 | { | |
469 | struct sk_buff *list; | |
470 | ||
fbb398a8 | 471 | skb_walk_frags(skb, list) |
1da177e4 LT |
472 | skb_get(list); |
473 | } | |
474 | ||
d3836f21 ED |
475 | static void skb_free_head(struct sk_buff *skb) |
476 | { | |
477 | if (skb->head_frag) | |
478 | put_page(virt_to_head_page(skb->head)); | |
479 | else | |
480 | kfree(skb->head); | |
481 | } | |
482 | ||
5bba1712 | 483 | static void skb_release_data(struct sk_buff *skb) |
1da177e4 LT |
484 | { |
485 | if (!skb->cloned || | |
486 | !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, | |
487 | &skb_shinfo(skb)->dataref)) { | |
488 | if (skb_shinfo(skb)->nr_frags) { | |
489 | int i; | |
490 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) | |
ea2ab693 | 491 | skb_frag_unref(skb, i); |
1da177e4 LT |
492 | } |
493 | ||
a6686f2f SM |
494 | /* |
495 | * If skb buf is from userspace, we need to notify the caller | |
496 | * the lower device DMA has done; | |
497 | */ | |
498 | if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { | |
499 | struct ubuf_info *uarg; | |
500 | ||
501 | uarg = skb_shinfo(skb)->destructor_arg; | |
502 | if (uarg->callback) | |
e19d6763 | 503 | uarg->callback(uarg, true); |
a6686f2f SM |
504 | } |
505 | ||
21dc3301 | 506 | if (skb_has_frag_list(skb)) |
1da177e4 LT |
507 | skb_drop_fraglist(skb); |
508 | ||
d3836f21 | 509 | skb_free_head(skb); |
1da177e4 LT |
510 | } |
511 | } | |
512 | ||
513 | /* | |
514 | * Free an skbuff by memory without cleaning the state. | |
515 | */ | |
2d4baff8 | 516 | static void kfree_skbmem(struct sk_buff *skb) |
1da177e4 | 517 | { |
d179cd12 DM |
518 | struct sk_buff *other; |
519 | atomic_t *fclone_ref; | |
520 | ||
d179cd12 DM |
521 | switch (skb->fclone) { |
522 | case SKB_FCLONE_UNAVAILABLE: | |
523 | kmem_cache_free(skbuff_head_cache, skb); | |
524 | break; | |
525 | ||
526 | case SKB_FCLONE_ORIG: | |
527 | fclone_ref = (atomic_t *) (skb + 2); | |
528 | if (atomic_dec_and_test(fclone_ref)) | |
529 | kmem_cache_free(skbuff_fclone_cache, skb); | |
530 | break; | |
531 | ||
532 | case SKB_FCLONE_CLONE: | |
533 | fclone_ref = (atomic_t *) (skb + 1); | |
534 | other = skb - 1; | |
535 | ||
536 | /* The clone portion is available for | |
537 | * fast-cloning again. | |
538 | */ | |
539 | skb->fclone = SKB_FCLONE_UNAVAILABLE; | |
540 | ||
541 | if (atomic_dec_and_test(fclone_ref)) | |
542 | kmem_cache_free(skbuff_fclone_cache, other); | |
543 | break; | |
3ff50b79 | 544 | } |
1da177e4 LT |
545 | } |
546 | ||
04a4bb55 | 547 | static void skb_release_head_state(struct sk_buff *skb) |
1da177e4 | 548 | { |
adf30907 | 549 | skb_dst_drop(skb); |
1da177e4 LT |
550 | #ifdef CONFIG_XFRM |
551 | secpath_put(skb->sp); | |
552 | #endif | |
9c2b3328 SH |
553 | if (skb->destructor) { |
554 | WARN_ON(in_irq()); | |
1da177e4 LT |
555 | skb->destructor(skb); |
556 | } | |
a3bf7ae9 | 557 | #if IS_ENABLED(CONFIG_NF_CONNTRACK) |
5f79e0f9 | 558 | nf_conntrack_put(skb->nfct); |
2fc72c7b | 559 | #endif |
1da177e4 LT |
560 | #ifdef CONFIG_BRIDGE_NETFILTER |
561 | nf_bridge_put(skb->nf_bridge); | |
562 | #endif | |
1da177e4 LT |
563 | /* XXX: IS this still necessary? - JHS */ |
564 | #ifdef CONFIG_NET_SCHED | |
565 | skb->tc_index = 0; | |
566 | #ifdef CONFIG_NET_CLS_ACT | |
567 | skb->tc_verd = 0; | |
1da177e4 LT |
568 | #endif |
569 | #endif | |
04a4bb55 LB |
570 | } |
571 | ||
572 | /* Free everything but the sk_buff shell. */ | |
573 | static void skb_release_all(struct sk_buff *skb) | |
574 | { | |
575 | skb_release_head_state(skb); | |
5e71d9d7 | 576 | if (likely(skb->head)) |
0ebd0ac5 | 577 | skb_release_data(skb); |
2d4baff8 HX |
578 | } |
579 | ||
580 | /** | |
581 | * __kfree_skb - private function | |
582 | * @skb: buffer | |
583 | * | |
584 | * Free an sk_buff. Release anything attached to the buffer. | |
585 | * Clean the state. This is an internal helper function. Users should | |
586 | * always call kfree_skb | |
587 | */ | |
1da177e4 | 588 | |
2d4baff8 HX |
589 | void __kfree_skb(struct sk_buff *skb) |
590 | { | |
591 | skb_release_all(skb); | |
1da177e4 LT |
592 | kfree_skbmem(skb); |
593 | } | |
b4ac530f | 594 | EXPORT_SYMBOL(__kfree_skb); |
1da177e4 | 595 | |