]>
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> | |
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> | |
9c55e01c | 56 | #include <linux/splice.h> |
1da177e4 LT |
57 | #include <linux/cache.h> |
58 | #include <linux/rtnetlink.h> | |
59 | #include <linux/init.h> | |
716ea3a7 | 60 | #include <linux/scatterlist.h> |
ac45f602 | 61 | #include <linux/errqueue.h> |
268bb0ce | 62 | #include <linux/prefetch.h> |
1da177e4 LT |
63 | |
64 | #include <net/protocol.h> | |
65 | #include <net/dst.h> | |
66 | #include <net/sock.h> | |
67 | #include <net/checksum.h> | |
68 | #include <net/xfrm.h> | |
69 | ||
70 | #include <asm/uaccess.h> | |
ad8d75ff | 71 | #include <trace/events/skb.h> |
51c56b00 | 72 | #include <linux/highmem.h> |
a1f8e7f7 | 73 | |
d7e8883c | 74 | struct kmem_cache *skbuff_head_cache __read_mostly; |
e18b890b | 75 | static struct kmem_cache *skbuff_fclone_cache __read_mostly; |
1da177e4 | 76 | |
9c55e01c JA |
77 | static void sock_pipe_buf_release(struct pipe_inode_info *pipe, |
78 | struct pipe_buffer *buf) | |
79 | { | |
8b9d3728 | 80 | put_page(buf->page); |
9c55e01c JA |
81 | } |
82 | ||
83 | static void sock_pipe_buf_get(struct pipe_inode_info *pipe, | |
84 | struct pipe_buffer *buf) | |
85 | { | |
8b9d3728 | 86 | get_page(buf->page); |
9c55e01c JA |
87 | } |
88 | ||
89 | static int sock_pipe_buf_steal(struct pipe_inode_info *pipe, | |
90 | struct pipe_buffer *buf) | |
91 | { | |
92 | return 1; | |
93 | } | |
94 | ||
95 | ||
96 | /* Pipe buffer operations for a socket. */ | |
28dfef8f | 97 | static const struct pipe_buf_operations sock_pipe_buf_ops = { |
9c55e01c JA |
98 | .can_merge = 0, |
99 | .map = generic_pipe_buf_map, | |
100 | .unmap = generic_pipe_buf_unmap, | |
101 | .confirm = generic_pipe_buf_confirm, | |
102 | .release = sock_pipe_buf_release, | |
103 | .steal = sock_pipe_buf_steal, | |
104 | .get = sock_pipe_buf_get, | |
105 | }; | |
106 | ||
1da177e4 | 107 | /** |
f05de73b JS |
108 | * skb_panic - private function for out-of-line support |
109 | * @skb: buffer | |
110 | * @sz: size | |
111 | * @addr: address | |
99d5851e | 112 | * @msg: skb_over_panic or skb_under_panic |
1da177e4 | 113 | * |
f05de73b JS |
114 | * Out-of-line support for skb_put() and skb_push(). |
115 | * Called via the wrapper skb_over_panic() or skb_under_panic(). | |
116 | * Keep out of line to prevent kernel bloat. | |
117 | * __builtin_return_address is not used because it is not always reliable. | |
1da177e4 | 118 | */ |
f05de73b | 119 | static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr, |
99d5851e | 120 | const char msg[]) |
1da177e4 | 121 | { |
e005d193 | 122 | pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n", |
99d5851e | 123 | msg, addr, skb->len, sz, skb->head, skb->data, |
e005d193 JP |
124 | (unsigned long)skb->tail, (unsigned long)skb->end, |
125 | skb->dev ? skb->dev->name : "<NULL>"); | |
1da177e4 LT |
126 | BUG(); |
127 | } | |
128 | ||
f05de73b | 129 | static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr) |
1da177e4 | 130 | { |
f05de73b | 131 | skb_panic(skb, sz, addr, __func__); |
1da177e4 LT |
132 | } |
133 | ||
f05de73b JS |
134 | static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr) |
135 | { | |
136 | skb_panic(skb, sz, addr, __func__); | |
137 | } | |
c93bdd0e MG |
138 | |
139 | /* | |
140 | * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells | |
141 | * the caller if emergency pfmemalloc reserves are being used. If it is and | |
142 | * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves | |
143 | * may be used. Otherwise, the packet data may be discarded until enough | |
144 | * memory is free | |
145 | */ | |
146 | #define kmalloc_reserve(size, gfp, node, pfmemalloc) \ | |
147 | __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc) | |
61c5e88a | 148 | |
149 | static void *__kmalloc_reserve(size_t size, gfp_t flags, int node, | |
150 | unsigned long ip, bool *pfmemalloc) | |
c93bdd0e MG |
151 | { |
152 | void *obj; | |
153 | bool ret_pfmemalloc = false; | |
154 | ||
155 | /* | |
156 | * Try a regular allocation, when that fails and we're not entitled | |
157 | * to the reserves, fail. | |
158 | */ | |
159 | obj = kmalloc_node_track_caller(size, | |
160 | flags | __GFP_NOMEMALLOC | __GFP_NOWARN, | |
161 | node); | |
162 | if (obj || !(gfp_pfmemalloc_allowed(flags))) | |
163 | goto out; | |
164 | ||
165 | /* Try again but now we are using pfmemalloc reserves */ | |
166 | ret_pfmemalloc = true; | |
167 | obj = kmalloc_node_track_caller(size, flags, node); | |
168 | ||
169 | out: | |
170 | if (pfmemalloc) | |
171 | *pfmemalloc = ret_pfmemalloc; | |
172 | ||
173 | return obj; | |
174 | } | |
175 | ||
1da177e4 LT |
176 | /* Allocate a new skbuff. We do this ourselves so we can fill in a few |
177 | * 'private' fields and also do memory statistics to find all the | |
178 | * [BEEP] leaks. | |
179 | * | |
180 | */ | |
181 | ||
0ebd0ac5 PM |
182 | struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node) |
183 | { | |
184 | struct sk_buff *skb; | |
185 | ||
186 | /* Get the HEAD */ | |
187 | skb = kmem_cache_alloc_node(skbuff_head_cache, | |
188 | gfp_mask & ~__GFP_DMA, node); | |
189 | if (!skb) | |
190 | goto out; | |
191 | ||
192 | /* | |
193 | * Only clear those fields we need to clear, not those that we will | |
194 | * actually initialise below. Hence, don't put any more fields after | |
195 | * the tail pointer in struct sk_buff! | |
196 | */ | |
197 | memset(skb, 0, offsetof(struct sk_buff, tail)); | |
5e71d9d7 | 198 | skb->head = NULL; |
0ebd0ac5 PM |
199 | skb->truesize = sizeof(struct sk_buff); |
200 | atomic_set(&skb->users, 1); | |
201 | ||
35d04610 | 202 | skb->mac_header = (typeof(skb->mac_header))~0U; |
0ebd0ac5 PM |
203 | out: |
204 | return skb; | |
205 | } | |
206 | ||
1da177e4 | 207 | /** |
d179cd12 | 208 | * __alloc_skb - allocate a network buffer |
1da177e4 LT |
209 | * @size: size to allocate |
210 | * @gfp_mask: allocation mask | |
c93bdd0e MG |
211 | * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache |
212 | * instead of head cache and allocate a cloned (child) skb. | |
213 | * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for | |
214 | * allocations in case the data is required for writeback | |
b30973f8 | 215 | * @node: numa node to allocate memory on |
1da177e4 LT |
216 | * |
217 | * Allocate a new &sk_buff. The returned buffer has no headroom and a | |
94b6042c BH |
218 | * tail room of at least size bytes. The object has a reference count |
219 | * of one. The return is the buffer. On a failure the return is %NULL. | |
1da177e4 LT |
220 | * |
221 | * Buffers may only be allocated from interrupts using a @gfp_mask of | |
222 | * %GFP_ATOMIC. | |
223 | */ | |
dd0fc66f | 224 | struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, |
c93bdd0e | 225 | int flags, int node) |
1da177e4 | 226 | { |
e18b890b | 227 | struct kmem_cache *cache; |
4947d3ef | 228 | struct skb_shared_info *shinfo; |
1da177e4 LT |
229 | struct sk_buff *skb; |
230 | u8 *data; | |
c93bdd0e | 231 | bool pfmemalloc; |
1da177e4 | 232 | |
c93bdd0e MG |
233 | cache = (flags & SKB_ALLOC_FCLONE) |
234 | ? skbuff_fclone_cache : skbuff_head_cache; | |
235 | ||
236 | if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX)) | |
237 | gfp_mask |= __GFP_MEMALLOC; | |
8798b3fb | 238 | |
1da177e4 | 239 | /* Get the HEAD */ |
b30973f8 | 240 | skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node); |
1da177e4 LT |
241 | if (!skb) |
242 | goto out; | |
ec7d2f2c | 243 | prefetchw(skb); |
1da177e4 | 244 | |
87fb4b7b ED |
245 | /* We do our best to align skb_shared_info on a separate cache |
246 | * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives | |
247 | * aligned memory blocks, unless SLUB/SLAB debug is enabled. | |
248 | * Both skb->head and skb_shared_info are cache line aligned. | |
249 | */ | |
bc417e30 | 250 | size = SKB_DATA_ALIGN(size); |
87fb4b7b | 251 | size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); |
c93bdd0e | 252 | data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc); |
1da177e4 LT |
253 | if (!data) |
254 | goto nodata; | |
87fb4b7b ED |
255 | /* kmalloc(size) might give us more room than requested. |
256 | * Put skb_shared_info exactly at the end of allocated zone, | |
257 | * to allow max possible filling before reallocation. | |
258 | */ | |
259 | size = SKB_WITH_OVERHEAD(ksize(data)); | |
ec7d2f2c | 260 | prefetchw(data + size); |
1da177e4 | 261 | |
ca0605a7 | 262 | /* |
c8005785 JB |
263 | * Only clear those fields we need to clear, not those that we will |
264 | * actually initialise below. Hence, don't put any more fields after | |
265 | * the tail pointer in struct sk_buff! | |
ca0605a7 ACM |
266 | */ |
267 | memset(skb, 0, offsetof(struct sk_buff, tail)); | |
87fb4b7b ED |
268 | /* Account for allocated memory : skb + skb->head */ |
269 | skb->truesize = SKB_TRUESIZE(size); | |
c93bdd0e | 270 | skb->pfmemalloc = pfmemalloc; |
1da177e4 LT |
271 | atomic_set(&skb->users, 1); |
272 | skb->head = data; | |
273 | skb->data = data; | |
27a884dc | 274 | skb_reset_tail_pointer(skb); |
4305b541 | 275 | skb->end = skb->tail + size; |
35d04610 CW |
276 | skb->mac_header = (typeof(skb->mac_header))~0U; |
277 | skb->transport_header = (typeof(skb->transport_header))~0U; | |
19633e12 | 278 | |
4947d3ef BL |
279 | /* make sure we initialize shinfo sequentially */ |
280 | shinfo = skb_shinfo(skb); | |
ec7d2f2c | 281 | memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); |
4947d3ef | 282 | atomic_set(&shinfo->dataref, 1); |
c2aa3665 | 283 | kmemcheck_annotate_variable(shinfo->destructor_arg); |
4947d3ef | 284 | |
c93bdd0e | 285 | if (flags & SKB_ALLOC_FCLONE) { |
d179cd12 DM |
286 | struct sk_buff *child = skb + 1; |
287 | atomic_t *fclone_ref = (atomic_t *) (child + 1); | |
1da177e4 | 288 | |
fe55f6d5 VN |
289 | kmemcheck_annotate_bitfield(child, flags1); |
290 | kmemcheck_annotate_bitfield(child, flags2); | |
d179cd12 DM |
291 | skb->fclone = SKB_FCLONE_ORIG; |
292 | atomic_set(fclone_ref, 1); | |
293 | ||
294 | child->fclone = SKB_FCLONE_UNAVAILABLE; | |
c93bdd0e | 295 | child->pfmemalloc = pfmemalloc; |
d179cd12 | 296 | } |
1da177e4 LT |
297 | out: |
298 | return skb; | |
299 | nodata: | |
8798b3fb | 300 | kmem_cache_free(cache, skb); |
1da177e4 LT |
301 | skb = NULL; |
302 | goto out; | |
1da177e4 | 303 | } |
b4ac530f | 304 | EXPORT_SYMBOL(__alloc_skb); |
1da177e4 | 305 | |
b2b5ce9d ED |
306 | /** |
307 | * build_skb - build a network buffer | |
308 | * @data: data buffer provided by caller | |
d3836f21 | 309 | * @frag_size: size of fragment, or 0 if head was kmalloced |
b2b5ce9d ED |
310 | * |
311 | * Allocate a new &sk_buff. Caller provides space holding head and | |
deceb4c0 FF |
312 | * skb_shared_info. @data must have been allocated by kmalloc() only if |
313 | * @frag_size is 0, otherwise data should come from the page allocator. | |
b2b5ce9d ED |
314 | * The return is the new skb buffer. |
315 | * On a failure the return is %NULL, and @data is not freed. | |
316 | * Notes : | |
317 | * Before IO, driver allocates only data buffer where NIC put incoming frame | |
318 | * Driver should add room at head (NET_SKB_PAD) and | |
319 | * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info)) | |
320 | * After IO, driver calls build_skb(), to allocate sk_buff and populate it | |
321 | * before giving packet to stack. | |
322 | * RX rings only contains data buffers, not full skbs. | |
323 | */ | |
d3836f21 | 324 | struct sk_buff *build_skb(void *data, unsigned int frag_size) |
b2b5ce9d ED |
325 | { |
326 | struct skb_shared_info *shinfo; | |
327 | struct sk_buff *skb; | |
d3836f21 | 328 | unsigned int size = frag_size ? : ksize(data); |
b2b5ce9d ED |
329 | |
330 | skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC); | |
331 | if (!skb) | |
332 | return NULL; | |
333 | ||
d3836f21 | 334 | size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); |
b2b5ce9d ED |
335 | |
336 | memset(skb, 0, offsetof(struct sk_buff, tail)); | |
337 | skb->truesize = SKB_TRUESIZE(size); | |
d3836f21 | 338 | skb->head_frag = frag_size != 0; |
b2b5ce9d ED |
339 | atomic_set(&skb->users, 1); |
340 | skb->head = data; | |
341 | skb->data = data; | |
342 | skb_reset_tail_pointer(skb); | |
343 | skb->end = skb->tail + size; | |
35d04610 CW |
344 | skb->mac_header = (typeof(skb->mac_header))~0U; |
345 | skb->transport_header = (typeof(skb->transport_header))~0U; | |
b2b5ce9d ED |
346 | |
347 | /* make sure we initialize shinfo sequentially */ | |
348 | shinfo = skb_shinfo(skb); | |
349 | memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); | |
350 | atomic_set(&shinfo->dataref, 1); | |
351 | kmemcheck_annotate_variable(shinfo->destructor_arg); | |
352 | ||
353 | return skb; | |
354 | } | |
355 | EXPORT_SYMBOL(build_skb); | |
356 | ||
a1c7fff7 | 357 | struct netdev_alloc_cache { |
69b08f62 ED |
358 | struct page_frag frag; |
359 | /* we maintain a pagecount bias, so that we dont dirty cache line | |
360 | * containing page->_count every time we allocate a fragment. | |
361 | */ | |
362 | unsigned int pagecnt_bias; | |
a1c7fff7 ED |
363 | }; |
364 | static DEFINE_PER_CPU(struct netdev_alloc_cache, netdev_alloc_cache); | |
365 | ||
c93bdd0e | 366 | static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask) |
6f532612 ED |
367 | { |
368 | struct netdev_alloc_cache *nc; | |
369 | void *data = NULL; | |
69b08f62 | 370 | int order; |
6f532612 ED |
371 | unsigned long flags; |
372 | ||
373 | local_irq_save(flags); | |
374 | nc = &__get_cpu_var(netdev_alloc_cache); | |
69b08f62 | 375 | if (unlikely(!nc->frag.page)) { |
6f532612 | 376 | refill: |
69b08f62 ED |
377 | for (order = NETDEV_FRAG_PAGE_MAX_ORDER; ;) { |
378 | gfp_t gfp = gfp_mask; | |
379 | ||
380 | if (order) | |
381 | gfp |= __GFP_COMP | __GFP_NOWARN; | |
382 | nc->frag.page = alloc_pages(gfp, order); | |
383 | if (likely(nc->frag.page)) | |
384 | break; | |
385 | if (--order < 0) | |
386 | goto end; | |
387 | } | |
388 | nc->frag.size = PAGE_SIZE << order; | |
540eb7bf | 389 | recycle: |
69b08f62 ED |
390 | atomic_set(&nc->frag.page->_count, NETDEV_PAGECNT_MAX_BIAS); |
391 | nc->pagecnt_bias = NETDEV_PAGECNT_MAX_BIAS; | |
392 | nc->frag.offset = 0; | |
6f532612 | 393 | } |
540eb7bf | 394 | |
69b08f62 | 395 | if (nc->frag.offset + fragsz > nc->frag.size) { |
540eb7bf | 396 | /* avoid unnecessary locked operations if possible */ |
69b08f62 ED |
397 | if ((atomic_read(&nc->frag.page->_count) == nc->pagecnt_bias) || |
398 | atomic_sub_and_test(nc->pagecnt_bias, &nc->frag.page->_count)) | |
540eb7bf AD |
399 | goto recycle; |
400 | goto refill; | |
6f532612 | 401 | } |
540eb7bf | 402 | |
69b08f62 ED |
403 | data = page_address(nc->frag.page) + nc->frag.offset; |
404 | nc->frag.offset += fragsz; | |
540eb7bf AD |
405 | nc->pagecnt_bias--; |
406 | end: | |
6f532612 ED |
407 | local_irq_restore(flags); |
408 | return data; | |
409 | } | |
c93bdd0e MG |
410 | |
411 | /** | |
412 | * netdev_alloc_frag - allocate a page fragment | |
413 | * @fragsz: fragment size | |
414 | * | |
415 | * Allocates a frag from a page for receive buffer. | |
416 | * Uses GFP_ATOMIC allocations. | |
417 | */ | |
418 | void *netdev_alloc_frag(unsigned int fragsz) | |
419 | { | |
420 | return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD); | |
421 | } | |
6f532612 ED |
422 | EXPORT_SYMBOL(netdev_alloc_frag); |
423 | ||
8af27456 CH |
424 | /** |
425 | * __netdev_alloc_skb - allocate an skbuff for rx on a specific device | |
426 | * @dev: network device to receive on | |
427 | * @length: length to allocate | |
428 | * @gfp_mask: get_free_pages mask, passed to alloc_skb | |
429 | * | |
430 | * Allocate a new &sk_buff and assign it a usage count of one. The | |
431 | * buffer has unspecified headroom built in. Users should allocate | |
432 | * the headroom they think they need without accounting for the | |
433 | * built in space. The built in space is used for optimisations. | |
434 | * | |
435 | * %NULL is returned if there is no free memory. | |
436 | */ | |
437 | struct sk_buff *__netdev_alloc_skb(struct net_device *dev, | |
6f532612 | 438 | unsigned int length, gfp_t gfp_mask) |
8af27456 | 439 | { |
6f532612 | 440 | struct sk_buff *skb = NULL; |
a1c7fff7 ED |
441 | unsigned int fragsz = SKB_DATA_ALIGN(length + NET_SKB_PAD) + |
442 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
443 | ||
310e158c | 444 | if (fragsz <= PAGE_SIZE && !(gfp_mask & (__GFP_WAIT | GFP_DMA))) { |
c93bdd0e MG |
445 | void *data; |
446 | ||
447 | if (sk_memalloc_socks()) | |
448 | gfp_mask |= __GFP_MEMALLOC; | |
449 | ||
450 | data = __netdev_alloc_frag(fragsz, gfp_mask); | |
a1c7fff7 | 451 | |
6f532612 ED |
452 | if (likely(data)) { |
453 | skb = build_skb(data, fragsz); | |
454 | if (unlikely(!skb)) | |
455 | put_page(virt_to_head_page(data)); | |
a1c7fff7 | 456 | } |
a1c7fff7 | 457 | } else { |
c93bdd0e MG |
458 | skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, |
459 | SKB_ALLOC_RX, NUMA_NO_NODE); | |
a1c7fff7 | 460 | } |
7b2e497a | 461 | if (likely(skb)) { |
8af27456 | 462 | skb_reserve(skb, NET_SKB_PAD); |
7b2e497a CH |
463 | skb->dev = dev; |
464 | } | |
8af27456 CH |
465 | return skb; |
466 | } | |
b4ac530f | 467 | EXPORT_SYMBOL(__netdev_alloc_skb); |
1da177e4 | 468 | |
654bed16 | 469 | void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, |
50269e19 | 470 | int size, unsigned int truesize) |
654bed16 PZ |
471 | { |
472 | skb_fill_page_desc(skb, i, page, off, size); | |
473 | skb->len += size; | |
474 | skb->data_len += size; | |
50269e19 | 475 | skb->truesize += truesize; |
654bed16 PZ |
476 | } |
477 | EXPORT_SYMBOL(skb_add_rx_frag); | |
478 | ||
27b437c8 | 479 | static void skb_drop_list(struct sk_buff **listp) |
1da177e4 | 480 | { |
bd8a7036 | 481 | kfree_skb_list(*listp); |
27b437c8 | 482 | *listp = NULL; |
1da177e4 LT |
483 | } |
484 | ||
27b437c8 HX |
485 | static inline void skb_drop_fraglist(struct sk_buff *skb) |
486 | { | |
487 | skb_drop_list(&skb_shinfo(skb)->frag_list); | |
488 | } | |
489 | ||
1da177e4 LT |
490 | static void skb_clone_fraglist(struct sk_buff *skb) |
491 | { | |
492 | struct sk_buff *list; | |
493 | ||
fbb398a8 | 494 | skb_walk_frags(skb, list) |
1da177e4 LT |
495 | skb_get(list); |
496 | } | |
497 | ||
d3836f21 ED |
498 | static void skb_free_head(struct sk_buff *skb) |
499 | { | |
500 | if (skb->head_frag) | |
501 | put_page(virt_to_head_page(skb->head)); | |
502 | else | |
503 | kfree(skb->head); | |
504 | } | |
505 | ||
5bba1712 | 506 | static void skb_release_data(struct sk_buff *skb) |
1da177e4 LT |
507 | { |
508 | if (!skb->cloned || | |
509 | !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, | |
510 | &skb_shinfo(skb)->dataref)) { | |
511 | if (skb_shinfo(skb)->nr_frags) { | |
512 | int i; | |
513 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) | |
ea2ab693 | 514 | skb_frag_unref(skb, i); |
1da177e4 LT |
515 | } |
516 | ||
a6686f2f SM |
517 | /* |
518 | * If skb buf is from userspace, we need to notify the caller | |
519 | * the lower device DMA has done; | |
520 | */ | |
521 | if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { | |
522 | struct ubuf_info *uarg; | |
523 | ||
524 | uarg = skb_shinfo(skb)->destructor_arg; | |
525 | if (uarg->callback) | |
e19d6763 | 526 | uarg->callback(uarg, true); |
a6686f2f SM |
527 | } |
528 | ||
21dc3301 | 529 | if (skb_has_frag_list(skb)) |
1da177e4 LT |
530 | skb_drop_fraglist(skb); |
531 | ||
d3836f21 | 532 | skb_free_head(skb); |
1da177e4 LT |
533 | } |
534 | } | |
535 | ||
536 | /* | |
537 | * Free an skbuff by memory without cleaning the state. | |
538 | */ | |
2d4baff8 | 539 | static void kfree_skbmem(struct sk_buff *skb) |
1da177e4 | 540 | { |
d179cd12 DM |
541 | struct sk_buff *other; |
542 | atomic_t *fclone_ref; | |
543 | ||
d179cd12 DM |
544 | switch (skb->fclone) { |
545 | case SKB_FCLONE_UNAVAILABLE: | |
546 | kmem_cache_free(skbuff_head_cache, skb); | |
547 | break; | |
548 | ||
549 | case SKB_FCLONE_ORIG: | |
550 | fclone_ref = (atomic_t *) (skb + 2); | |
551 | if (atomic_dec_and_test(fclone_ref)) | |
552 | kmem_cache_free(skbuff_fclone_cache, skb); | |
553 | break; | |
554 | ||
555 | case SKB_FCLONE_CLONE: | |
556 | fclone_ref = (atomic_t *) (skb + 1); | |
557 | other = skb - 1; | |
558 | ||
559 | /* The clone portion is available for | |
560 | * fast-cloning again. | |
561 | */ | |
562 | skb->fclone = SKB_FCLONE_UNAVAILABLE; | |
563 | ||
564 | if (atomic_dec_and_test(fclone_ref)) | |
565 | kmem_cache_free(skbuff_fclone_cache, other); | |
566 | break; | |
3ff50b79 | 567 | } |
1da177e4 LT |
568 | } |
569 | ||
04a4bb55 | 570 | static void skb_release_head_state(struct sk_buff *skb) |
1da177e4 | 571 | { |
adf30907 | 572 | skb_dst_drop(skb); |
1da177e4 LT |
573 | #ifdef CONFIG_XFRM |
574 | secpath_put(skb->sp); | |
575 | #endif | |
9c2b3328 SH |
576 | if (skb->destructor) { |
577 | WARN_ON(in_irq()); | |
1da177e4 LT |
578 | skb->destructor(skb); |
579 | } | |
a3bf7ae9 | 580 | #if IS_ENABLED(CONFIG_NF_CONNTRACK) |
5f79e0f9 | 581 | nf_conntrack_put(skb->nfct); |
2fc72c7b KK |
582 | #endif |
583 | #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED | |
9fb9cbb1 YK |
584 | nf_conntrack_put_reasm(skb->nfct_reasm); |
585 | #endif | |
1da177e4 LT |
586 | #ifdef CONFIG_BRIDGE_NETFILTER |
587 | nf_bridge_put(skb->nf_bridge); | |
588 | #endif | |
1da177e4 LT |
589 | /* XXX: IS this still necessary? - JHS */ |
590 | #ifdef CONFIG_NET_SCHED | |
591 | skb->tc_index = 0; | |
592 | #ifdef CONFIG_NET_CLS_ACT | |
593 | skb->tc_verd = 0; | |
1da177e4 LT |
594 | #endif |
595 | #endif | |
04a4bb55 LB |
596 | } |
597 | ||
598 | /* Free everything but the sk_buff shell. */ | |
599 | static void skb_release_all(struct sk_buff *skb) | |
600 | { | |
601 | skb_release_head_state(skb); | |
5e71d9d7 | 602 | if (likely(skb->head)) |
0ebd0ac5 | 603 | skb_release_data(skb); |
2d4baff8 HX |
604 | } |
605 | ||
606 | /** | |
607 | * __kfree_skb - private function | |
608 | * @skb: buffer | |
609 | * | |
610 | * Free an sk_buff. Release anything attached to the buffer. | |
611 | * Clean the state. This is an internal helper function. Users should | |
612 | * always call kfree_skb | |
613 | */ | |
1da177e4 | 614 | |
2d4baff8 HX |
615 | void __kfree_skb(struct sk_buff *skb) |
616 | { | |
617 | skb_release_all(skb); | |
1da177e4 LT |
618 | kfree_skbmem(skb); |
619 | } | |
b4ac530f | 620 | EXPORT_SYMBOL(__kfree_skb); |
1da177e4 | 621 | |