]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Routines having to do with the 'struct sk_buff' memory handlers. | |
3 | * | |
113aa838 | 4 | * Authors: Alan Cox <[email protected]> |
1da177e4 LT |
5 | * Florian La Roche <[email protected]> |
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> | |
de960aa9 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> |
0d5501c1 | 65 | #include <linux/if_vlan.h> |
1da177e4 LT |
66 | |
67 | #include <net/protocol.h> | |
68 | #include <net/dst.h> | |
69 | #include <net/sock.h> | |
70 | #include <net/checksum.h> | |
ed1f50c3 | 71 | #include <net/ip6_checksum.h> |
1da177e4 LT |
72 | #include <net/xfrm.h> |
73 | ||
74 | #include <asm/uaccess.h> | |
ad8d75ff | 75 | #include <trace/events/skb.h> |
51c56b00 | 76 | #include <linux/highmem.h> |
b245be1f WB |
77 | #include <linux/capability.h> |
78 | #include <linux/user_namespace.h> | |
a1f8e7f7 | 79 | |
d7e8883c | 80 | struct kmem_cache *skbuff_head_cache __read_mostly; |
e18b890b | 81 | static struct kmem_cache *skbuff_fclone_cache __read_mostly; |
1da177e4 | 82 | |
1da177e4 | 83 | /** |
f05de73b JS |
84 | * skb_panic - private function for out-of-line support |
85 | * @skb: buffer | |
86 | * @sz: size | |
87 | * @addr: address | |
99d5851e | 88 | * @msg: skb_over_panic or skb_under_panic |
1da177e4 | 89 | * |
f05de73b JS |
90 | * Out-of-line support for skb_put() and skb_push(). |
91 | * Called via the wrapper skb_over_panic() or skb_under_panic(). | |
92 | * Keep out of line to prevent kernel bloat. | |
93 | * __builtin_return_address is not used because it is not always reliable. | |
1da177e4 | 94 | */ |
f05de73b | 95 | static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr, |
99d5851e | 96 | const char msg[]) |
1da177e4 | 97 | { |
e005d193 | 98 | pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n", |
99d5851e | 99 | msg, addr, skb->len, sz, skb->head, skb->data, |
e005d193 JP |
100 | (unsigned long)skb->tail, (unsigned long)skb->end, |
101 | skb->dev ? skb->dev->name : "<NULL>"); | |
1da177e4 LT |
102 | BUG(); |
103 | } | |
104 | ||
f05de73b | 105 | static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr) |
1da177e4 | 106 | { |
f05de73b | 107 | skb_panic(skb, sz, addr, __func__); |
1da177e4 LT |
108 | } |
109 | ||
f05de73b JS |
110 | static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr) |
111 | { | |
112 | skb_panic(skb, sz, addr, __func__); | |
113 | } | |
c93bdd0e MG |
114 | |
115 | /* | |
116 | * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells | |
117 | * the caller if emergency pfmemalloc reserves are being used. If it is and | |
118 | * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves | |
119 | * may be used. Otherwise, the packet data may be discarded until enough | |
120 | * memory is free | |
121 | */ | |
122 | #define kmalloc_reserve(size, gfp, node, pfmemalloc) \ | |
123 | __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc) | |
61c5e88a | 124 | |
125 | static void *__kmalloc_reserve(size_t size, gfp_t flags, int node, | |
126 | unsigned long ip, bool *pfmemalloc) | |
c93bdd0e MG |
127 | { |
128 | void *obj; | |
129 | bool ret_pfmemalloc = false; | |
130 | ||
131 | /* | |
132 | * Try a regular allocation, when that fails and we're not entitled | |
133 | * to the reserves, fail. | |
134 | */ | |
135 | obj = kmalloc_node_track_caller(size, | |
136 | flags | __GFP_NOMEMALLOC | __GFP_NOWARN, | |
137 | node); | |
138 | if (obj || !(gfp_pfmemalloc_allowed(flags))) | |
139 | goto out; | |
140 | ||
141 | /* Try again but now we are using pfmemalloc reserves */ | |
142 | ret_pfmemalloc = true; | |
143 | obj = kmalloc_node_track_caller(size, flags, node); | |
144 | ||
145 | out: | |
146 | if (pfmemalloc) | |
147 | *pfmemalloc = ret_pfmemalloc; | |
148 | ||
149 | return obj; | |
150 | } | |
151 | ||
1da177e4 LT |
152 | /* Allocate a new skbuff. We do this ourselves so we can fill in a few |
153 | * 'private' fields and also do memory statistics to find all the | |
154 | * [BEEP] leaks. | |
155 | * | |
156 | */ | |
157 | ||
0ebd0ac5 PM |
158 | struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node) |
159 | { | |
160 | struct sk_buff *skb; | |
161 | ||
162 | /* Get the HEAD */ | |
163 | skb = kmem_cache_alloc_node(skbuff_head_cache, | |
164 | gfp_mask & ~__GFP_DMA, node); | |
165 | if (!skb) | |
166 | goto out; | |
167 | ||
168 | /* | |
169 | * Only clear those fields we need to clear, not those that we will | |
170 | * actually initialise below. Hence, don't put any more fields after | |
171 | * the tail pointer in struct sk_buff! | |
172 | */ | |
173 | memset(skb, 0, offsetof(struct sk_buff, tail)); | |
5e71d9d7 | 174 | skb->head = NULL; |
0ebd0ac5 PM |
175 | skb->truesize = sizeof(struct sk_buff); |
176 | atomic_set(&skb->users, 1); | |
177 | ||
35d04610 | 178 | skb->mac_header = (typeof(skb->mac_header))~0U; |
0ebd0ac5 PM |
179 | out: |
180 | return skb; | |
181 | } | |
182 | ||
1da177e4 | 183 | /** |
d179cd12 | 184 | * __alloc_skb - allocate a network buffer |
1da177e4 LT |
185 | * @size: size to allocate |
186 | * @gfp_mask: allocation mask | |
c93bdd0e MG |
187 | * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache |
188 | * instead of head cache and allocate a cloned (child) skb. | |
189 | * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for | |
190 | * allocations in case the data is required for writeback | |
b30973f8 | 191 | * @node: numa node to allocate memory on |
1da177e4 LT |
192 | * |
193 | * Allocate a new &sk_buff. The returned buffer has no headroom and a | |
94b6042c BH |
194 | * tail room of at least size bytes. The object has a reference count |
195 | * of one. The return is the buffer. On a failure the return is %NULL. | |
1da177e4 LT |
196 | * |
197 | * Buffers may only be allocated from interrupts using a @gfp_mask of | |
198 | * %GFP_ATOMIC. | |
199 | */ | |
dd0fc66f | 200 | struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, |
c93bdd0e | 201 | int flags, int node) |
1da177e4 | 202 | { |
e18b890b | 203 | struct kmem_cache *cache; |
4947d3ef | 204 | struct skb_shared_info *shinfo; |
1da177e4 LT |
205 | struct sk_buff *skb; |
206 | u8 *data; | |
c93bdd0e | 207 | bool pfmemalloc; |
1da177e4 | 208 | |
c93bdd0e MG |
209 | cache = (flags & SKB_ALLOC_FCLONE) |
210 | ? skbuff_fclone_cache : skbuff_head_cache; | |
211 | ||
212 | if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX)) | |
213 | gfp_mask |= __GFP_MEMALLOC; | |
8798b3fb | 214 | |
1da177e4 | 215 | /* Get the HEAD */ |
b30973f8 | 216 | skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node); |
1da177e4 LT |
217 | if (!skb) |
218 | goto out; | |
ec7d2f2c | 219 | prefetchw(skb); |
1da177e4 | 220 | |
87fb4b7b ED |
221 | /* We do our best to align skb_shared_info on a separate cache |
222 | * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives | |
223 | * aligned memory blocks, unless SLUB/SLAB debug is enabled. | |
224 | * Both skb->head and skb_shared_info are cache line aligned. | |
225 | */ | |
bc417e30 | 226 | size = SKB_DATA_ALIGN(size); |
87fb4b7b | 227 | size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); |
c93bdd0e | 228 | data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc); |
1da177e4 LT |
229 | if (!data) |
230 | goto nodata; | |
87fb4b7b ED |
231 | /* kmalloc(size) might give us more room than requested. |
232 | * Put skb_shared_info exactly at the end of allocated zone, | |
233 | * to allow max possible filling before reallocation. | |
234 | */ | |
235 | size = SKB_WITH_OVERHEAD(ksize(data)); | |
ec7d2f2c | 236 | prefetchw(data + size); |
1da177e4 | 237 | |
ca0605a7 | 238 | /* |
c8005785 JB |
239 | * Only clear those fields we need to clear, not those that we will |
240 | * actually initialise below. Hence, don't put any more fields after | |
241 | * the tail pointer in struct sk_buff! | |
ca0605a7 ACM |
242 | */ |
243 | memset(skb, 0, offsetof(struct sk_buff, tail)); | |
87fb4b7b ED |
244 | /* Account for allocated memory : skb + skb->head */ |
245 | skb->truesize = SKB_TRUESIZE(size); | |
c93bdd0e | 246 | skb->pfmemalloc = pfmemalloc; |
1da177e4 LT |
247 | atomic_set(&skb->users, 1); |
248 | skb->head = data; | |
249 | skb->data = data; | |
27a884dc | 250 | skb_reset_tail_pointer(skb); |
4305b541 | 251 | skb->end = skb->tail + size; |
35d04610 CW |
252 | skb->mac_header = (typeof(skb->mac_header))~0U; |
253 | skb->transport_header = (typeof(skb->transport_header))~0U; | |
19633e12 | 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) { |
d0bf4a9e | 262 | struct sk_buff_fclones *fclones; |
1da177e4 | 263 | |
d0bf4a9e ED |
264 | fclones = container_of(skb, struct sk_buff_fclones, skb1); |
265 | ||
266 | kmemcheck_annotate_bitfield(&fclones->skb2, flags1); | |
d179cd12 | 267 | skb->fclone = SKB_FCLONE_ORIG; |
d0bf4a9e | 268 | atomic_set(&fclones->fclone_ref, 1); |
d179cd12 | 269 | |
6ffe75eb | 270 | fclones->skb2.fclone = SKB_FCLONE_CLONE; |
d0bf4a9e | 271 | fclones->skb2.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 | 282 | /** |
2ea2f62c | 283 | * __build_skb - build a network buffer |
b2b5ce9d | 284 | * @data: data buffer provided by caller |
2ea2f62c | 285 | * @frag_size: size of data, or 0 if head was kmalloced |
b2b5ce9d ED |
286 | * |
287 | * Allocate a new &sk_buff. Caller provides space holding head and | |
deceb4c0 | 288 | * skb_shared_info. @data must have been allocated by kmalloc() only if |
2ea2f62c ED |
289 | * @frag_size is 0, otherwise data should come from the page allocator |
290 | * or vmalloc() | |
b2b5ce9d ED |
291 | * The return is the new skb buffer. |
292 | * On a failure the return is %NULL, and @data is not freed. | |
293 | * Notes : | |
294 | * Before IO, driver allocates only data buffer where NIC put incoming frame | |
295 | * Driver should add room at head (NET_SKB_PAD) and | |
296 | * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info)) | |
297 | * After IO, driver calls build_skb(), to allocate sk_buff and populate it | |
298 | * before giving packet to stack. | |
299 | * RX rings only contains data buffers, not full skbs. | |
300 | */ | |
2ea2f62c | 301 | struct sk_buff *__build_skb(void *data, unsigned int frag_size) |
b2b5ce9d ED |
302 | { |
303 | struct skb_shared_info *shinfo; | |
304 | struct sk_buff *skb; | |
d3836f21 | 305 | unsigned int size = frag_size ? : ksize(data); |
b2b5ce9d ED |
306 | |
307 | skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC); | |
308 | if (!skb) | |
309 | return NULL; | |
310 | ||
d3836f21 | 311 | size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); |
b2b5ce9d ED |
312 | |
313 | memset(skb, 0, offsetof(struct sk_buff, tail)); | |
314 | skb->truesize = SKB_TRUESIZE(size); | |
315 | atomic_set(&skb->users, 1); | |
316 | skb->head = data; | |
317 | skb->data = data; | |
318 | skb_reset_tail_pointer(skb); | |
319 | skb->end = skb->tail + size; | |
35d04610 CW |
320 | skb->mac_header = (typeof(skb->mac_header))~0U; |
321 | skb->transport_header = (typeof(skb->transport_header))~0U; | |
b2b5ce9d ED |
322 | |
323 | /* make sure we initialize shinfo sequentially */ | |
324 | shinfo = skb_shinfo(skb); | |
325 | memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); | |
326 | atomic_set(&shinfo->dataref, 1); | |
327 | kmemcheck_annotate_variable(shinfo->destructor_arg); | |
328 | ||
329 | return skb; | |
330 | } | |
2ea2f62c ED |
331 | |
332 | /* build_skb() is wrapper over __build_skb(), that specifically | |
333 | * takes care of skb->head and skb->pfmemalloc | |
334 | * This means that if @frag_size is not zero, then @data must be backed | |
335 | * by a page fragment, not kmalloc() or vmalloc() | |
336 | */ | |
337 | struct sk_buff *build_skb(void *data, unsigned int frag_size) | |
338 | { | |
339 | struct sk_buff *skb = __build_skb(data, frag_size); | |
340 | ||
341 | if (skb && frag_size) { | |
342 | skb->head_frag = 1; | |
343 | if (virt_to_head_page(data)->pfmemalloc) | |
344 | skb->pfmemalloc = 1; | |
345 | } | |
346 | return skb; | |
347 | } | |
b2b5ce9d ED |
348 | EXPORT_SYMBOL(build_skb); |
349 | ||
a1c7fff7 | 350 | struct netdev_alloc_cache { |
69b08f62 ED |
351 | struct page_frag frag; |
352 | /* we maintain a pagecount bias, so that we dont dirty cache line | |
353 | * containing page->_count every time we allocate a fragment. | |
354 | */ | |
355 | unsigned int pagecnt_bias; | |
a1c7fff7 ED |
356 | }; |
357 | static DEFINE_PER_CPU(struct netdev_alloc_cache, netdev_alloc_cache); | |
ffde7328 | 358 | static DEFINE_PER_CPU(struct netdev_alloc_cache, napi_alloc_cache); |
a1c7fff7 | 359 | |
ffde7328 AD |
360 | static struct page *__page_frag_refill(struct netdev_alloc_cache *nc, |
361 | gfp_t gfp_mask) | |
6f532612 | 362 | { |
ffde7328 AD |
363 | const unsigned int order = NETDEV_FRAG_PAGE_MAX_ORDER; |
364 | struct page *page = NULL; | |
365 | gfp_t gfp = gfp_mask; | |
366 | ||
367 | if (order) { | |
79930f58 ED |
368 | gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY | |
369 | __GFP_NOMEMALLOC; | |
ffde7328 AD |
370 | page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); |
371 | nc->frag.size = PAGE_SIZE << (page ? order : 0); | |
372 | } | |
6f532612 | 373 | |
ffde7328 AD |
374 | if (unlikely(!page)) |
375 | page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); | |
376 | ||
377 | nc->frag.page = page; | |
378 | ||
379 | return page; | |
380 | } | |
381 | ||
382 | static void *__alloc_page_frag(struct netdev_alloc_cache __percpu *cache, | |
383 | unsigned int fragsz, gfp_t gfp_mask) | |
384 | { | |
385 | struct netdev_alloc_cache *nc = this_cpu_ptr(cache); | |
386 | struct page *page = nc->frag.page; | |
387 | unsigned int size; | |
388 | int offset; | |
389 | ||
390 | if (unlikely(!page)) { | |
6f532612 | 391 | refill: |
ffde7328 AD |
392 | page = __page_frag_refill(nc, gfp_mask); |
393 | if (!page) | |
394 | return NULL; | |
395 | ||
396 | /* if size can vary use frag.size else just use PAGE_SIZE */ | |
397 | size = NETDEV_FRAG_PAGE_MAX_ORDER ? nc->frag.size : PAGE_SIZE; | |
398 | ||
4c450583 ED |
399 | /* Even if we own the page, we do not use atomic_set(). |
400 | * This would break get_page_unless_zero() users. | |
401 | */ | |
ffde7328 AD |
402 | atomic_add(size - 1, &page->_count); |
403 | ||
404 | /* reset page count bias and offset to start of new frag */ | |
405 | nc->pagecnt_bias = size; | |
406 | nc->frag.offset = size; | |
6f532612 | 407 | } |
540eb7bf | 408 | |
ffde7328 AD |
409 | offset = nc->frag.offset - fragsz; |
410 | if (unlikely(offset < 0)) { | |
411 | if (!atomic_sub_and_test(nc->pagecnt_bias, &page->_count)) | |
412 | goto refill; | |
413 | ||
414 | /* if size can vary use frag.size else just use PAGE_SIZE */ | |
415 | size = NETDEV_FRAG_PAGE_MAX_ORDER ? nc->frag.size : PAGE_SIZE; | |
416 | ||
417 | /* OK, page count is 0, we can safely set it */ | |
418 | atomic_set(&page->_count, size); | |
419 | ||
420 | /* reset page count bias and offset to start of new frag */ | |
421 | nc->pagecnt_bias = size; | |
422 | offset = size - fragsz; | |
6f532612 | 423 | } |
540eb7bf | 424 | |
540eb7bf | 425 | nc->pagecnt_bias--; |
ffde7328 AD |
426 | nc->frag.offset = offset; |
427 | ||
428 | return page_address(page) + offset; | |
429 | } | |
430 | ||
431 | static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask) | |
432 | { | |
433 | unsigned long flags; | |
434 | void *data; | |
435 | ||
436 | local_irq_save(flags); | |
437 | data = __alloc_page_frag(&netdev_alloc_cache, fragsz, gfp_mask); | |
6f532612 ED |
438 | local_irq_restore(flags); |
439 | return data; | |
440 | } | |
c93bdd0e MG |
441 | |
442 | /** | |
443 | * netdev_alloc_frag - allocate a page fragment | |
444 | * @fragsz: fragment size | |
445 | * | |
446 | * Allocates a frag from a page for receive buffer. | |
447 | * Uses GFP_ATOMIC allocations. | |
448 | */ | |
449 | void *netdev_alloc_frag(unsigned int fragsz) | |
450 | { | |
451 | return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD); | |
452 | } | |
6f532612 ED |
453 | EXPORT_SYMBOL(netdev_alloc_frag); |
454 | ||
ffde7328 AD |
455 | static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask) |
456 | { | |
457 | return __alloc_page_frag(&napi_alloc_cache, fragsz, gfp_mask); | |
458 | } | |
459 | ||
460 | void *napi_alloc_frag(unsigned int fragsz) | |
461 | { | |
462 | return __napi_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD); | |
463 | } | |
464 | EXPORT_SYMBOL(napi_alloc_frag); | |
465 | ||
8af27456 | 466 | /** |
fd11a83d | 467 | * __alloc_rx_skb - allocate an skbuff for rx |
8af27456 CH |
468 | * @length: length to allocate |
469 | * @gfp_mask: get_free_pages mask, passed to alloc_skb | |
fd11a83d AD |
470 | * @flags: If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for |
471 | * allocations in case we have to fallback to __alloc_skb() | |
472 | * If SKB_ALLOC_NAPI is set, page fragment will be allocated | |
473 | * from napi_cache instead of netdev_cache. | |
8af27456 CH |
474 | * |
475 | * Allocate a new &sk_buff and assign it a usage count of one. The | |
476 | * buffer has unspecified headroom built in. Users should allocate | |
477 | * the headroom they think they need without accounting for the | |
478 | * built in space. The built in space is used for optimisations. | |
479 | * | |
480 | * %NULL is returned if there is no free memory. | |
481 | */ | |
fd11a83d AD |
482 | static struct sk_buff *__alloc_rx_skb(unsigned int length, gfp_t gfp_mask, |
483 | int flags) | |
8af27456 | 484 | { |
6f532612 | 485 | struct sk_buff *skb = NULL; |
fd11a83d | 486 | unsigned int fragsz = SKB_DATA_ALIGN(length) + |
a1c7fff7 ED |
487 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); |
488 | ||
310e158c | 489 | if (fragsz <= PAGE_SIZE && !(gfp_mask & (__GFP_WAIT | GFP_DMA))) { |
c93bdd0e MG |
490 | void *data; |
491 | ||
492 | if (sk_memalloc_socks()) | |
493 | gfp_mask |= __GFP_MEMALLOC; | |
494 | ||
fd11a83d AD |
495 | data = (flags & SKB_ALLOC_NAPI) ? |
496 | __napi_alloc_frag(fragsz, gfp_mask) : | |
497 | __netdev_alloc_frag(fragsz, gfp_mask); | |
a1c7fff7 | 498 | |
6f532612 ED |
499 | if (likely(data)) { |
500 | skb = build_skb(data, fragsz); | |
501 | if (unlikely(!skb)) | |
502 | put_page(virt_to_head_page(data)); | |
a1c7fff7 | 503 | } |
a1c7fff7 | 504 | } else { |
fd11a83d | 505 | skb = __alloc_skb(length, gfp_mask, |
c93bdd0e | 506 | SKB_ALLOC_RX, NUMA_NO_NODE); |
a1c7fff7 | 507 | } |
fd11a83d AD |
508 | return skb; |
509 | } | |
510 | ||
511 | /** | |
512 | * __netdev_alloc_skb - allocate an skbuff for rx on a specific device | |
513 | * @dev: network device to receive on | |
514 | * @length: length to allocate | |
515 | * @gfp_mask: get_free_pages mask, passed to alloc_skb | |
516 | * | |
517 | * Allocate a new &sk_buff and assign it a usage count of one. The | |
518 | * buffer has NET_SKB_PAD headroom built in. Users should allocate | |
519 | * the headroom they think they need without accounting for the | |
520 | * built in space. The built in space is used for optimisations. | |
521 | * | |
522 | * %NULL is returned if there is no free memory. | |
523 | */ | |
524 | struct sk_buff *__netdev_alloc_skb(struct net_device *dev, | |
525 | unsigned int length, gfp_t gfp_mask) | |
526 | { | |
527 | struct sk_buff *skb; | |
528 | ||
529 | length += NET_SKB_PAD; | |
530 | skb = __alloc_rx_skb(length, gfp_mask, 0); | |
531 | ||
7b2e497a | 532 | if (likely(skb)) { |
8af27456 | 533 | skb_reserve(skb, NET_SKB_PAD); |
7b2e497a CH |
534 | skb->dev = dev; |
535 | } | |
fd11a83d | 536 | |
8af27456 CH |
537 | return skb; |
538 | } | |
b4ac530f | 539 | EXPORT_SYMBOL(__netdev_alloc_skb); |
1da177e4 | 540 | |
fd11a83d AD |
541 | /** |
542 | * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance | |
543 | * @napi: napi instance this buffer was allocated for | |
544 | * @length: length to allocate | |
545 | * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages | |
546 | * | |
547 | * Allocate a new sk_buff for use in NAPI receive. This buffer will | |
548 | * attempt to allocate the head from a special reserved region used | |
549 | * only for NAPI Rx allocation. By doing this we can save several | |
550 | * CPU cycles by avoiding having to disable and re-enable IRQs. | |
551 | * | |
552 | * %NULL is returned if there is no free memory. | |
553 | */ | |
554 | struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, | |
555 | unsigned int length, gfp_t gfp_mask) | |
556 | { | |
557 | struct sk_buff *skb; | |
558 | ||
559 | length += NET_SKB_PAD + NET_IP_ALIGN; | |
560 | skb = __alloc_rx_skb(length, gfp_mask, SKB_ALLOC_NAPI); | |
561 | ||
562 | if (likely(skb)) { | |
563 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); | |
564 | skb->dev = napi->dev; | |
565 | } | |
566 | ||
567 | return skb; | |
568 | } | |
569 | EXPORT_SYMBOL(__napi_alloc_skb); | |
570 | ||
654bed16 | 571 | void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, |
50269e19 | 572 | int size, unsigned int truesize) |
654bed16 PZ |
573 | { |
574 | skb_fill_page_desc(skb, i, page, off, size); | |
575 | skb->len += size; | |
576 | skb->data_len += size; | |
50269e19 | 577 | skb->truesize += truesize; |
654bed16 PZ |
578 | } |
579 | EXPORT_SYMBOL(skb_add_rx_frag); | |
580 | ||
f8e617e1 JW |
581 | void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, |
582 | unsigned int truesize) | |
583 | { | |
584 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
585 | ||
586 | skb_frag_size_add(frag, size); | |
587 | skb->len += size; | |
588 | skb->data_len += size; | |
589 | skb->truesize += truesize; | |
590 | } | |
591 | EXPORT_SYMBOL(skb_coalesce_rx_frag); | |
592 | ||
27b437c8 | 593 | static void skb_drop_list(struct sk_buff **listp) |
1da177e4 | 594 | { |
bd8a7036 | 595 | kfree_skb_list(*listp); |
27b437c8 | 596 | *listp = NULL; |
1da177e4 LT |
597 | } |
598 | ||
27b437c8 HX |
599 | static inline void skb_drop_fraglist(struct sk_buff *skb) |
600 | { | |
601 | skb_drop_list(&skb_shinfo(skb)->frag_list); | |
602 | } | |
603 | ||
1da177e4 LT |
604 | static void skb_clone_fraglist(struct sk_buff *skb) |
605 | { | |
606 | struct sk_buff *list; | |
607 | ||
fbb398a8 | 608 | skb_walk_frags(skb, list) |
1da177e4 LT |
609 | skb_get(list); |
610 | } | |
611 | ||
d3836f21 ED |
612 | static void skb_free_head(struct sk_buff *skb) |
613 | { | |
614 | if (skb->head_frag) | |
615 | put_page(virt_to_head_page(skb->head)); | |
616 | else | |
617 | kfree(skb->head); | |
618 | } | |
619 | ||
5bba1712 | 620 | static void skb_release_data(struct sk_buff *skb) |
1da177e4 | 621 | { |
ff04a771 ED |
622 | struct skb_shared_info *shinfo = skb_shinfo(skb); |
623 | int i; | |
1da177e4 | 624 | |
ff04a771 ED |
625 | if (skb->cloned && |
626 | atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, | |
627 | &shinfo->dataref)) | |
628 | return; | |
a6686f2f | 629 | |
ff04a771 ED |
630 | for (i = 0; i < shinfo->nr_frags; i++) |
631 | __skb_frag_unref(&shinfo->frags[i]); | |
a6686f2f | 632 | |
ff04a771 ED |
633 | /* |
634 | * If skb buf is from userspace, we need to notify the caller | |
635 | * the lower device DMA has done; | |
636 | */ | |
637 | if (shinfo->tx_flags & SKBTX_DEV_ZEROCOPY) { | |
638 | struct ubuf_info *uarg; | |
1da177e4 | 639 | |
ff04a771 ED |
640 | uarg = shinfo->destructor_arg; |
641 | if (uarg->callback) | |
642 | uarg->callback(uarg, true); | |
1da177e4 | 643 | } |
ff04a771 ED |
644 | |
645 | if (shinfo->frag_list) | |
646 | kfree_skb_list(shinfo->frag_list); | |
647 | ||
648 | skb_free_head(skb); | |
1da177e4 LT |
649 | } |
650 | ||
651 | /* | |
652 | * Free an skbuff by memory without cleaning the state. | |
653 | */ | |
2d4baff8 | 654 | static void kfree_skbmem(struct sk_buff *skb) |
1da177e4 | 655 | { |
d0bf4a9e | 656 | struct sk_buff_fclones *fclones; |
d179cd12 | 657 | |
d179cd12 DM |
658 | switch (skb->fclone) { |
659 | case SKB_FCLONE_UNAVAILABLE: | |
660 | kmem_cache_free(skbuff_head_cache, skb); | |
6ffe75eb | 661 | return; |
d179cd12 DM |
662 | |
663 | case SKB_FCLONE_ORIG: | |
d0bf4a9e | 664 | fclones = container_of(skb, struct sk_buff_fclones, skb1); |
d179cd12 | 665 | |
6ffe75eb ED |
666 | /* We usually free the clone (TX completion) before original skb |
667 | * This test would have no chance to be true for the clone, | |
668 | * while here, branch prediction will be good. | |
d179cd12 | 669 | */ |
6ffe75eb ED |
670 | if (atomic_read(&fclones->fclone_ref) == 1) |
671 | goto fastpath; | |
672 | break; | |
e7820e39 | 673 | |
6ffe75eb ED |
674 | default: /* SKB_FCLONE_CLONE */ |
675 | fclones = container_of(skb, struct sk_buff_fclones, skb2); | |
d179cd12 | 676 | break; |
3ff50b79 | 677 | } |
6ffe75eb ED |
678 | if (!atomic_dec_and_test(&fclones->fclone_ref)) |
679 | return; | |
680 | fastpath: | |
681 | kmem_cache_free(skbuff_fclone_cache, fclones); | |
1da177e4 LT |
682 | } |
683 | ||
04a4bb55 | 684 | static void skb_release_head_state(struct sk_buff *skb) |
1da177e4 | 685 | { |
adf30907 | 686 | skb_dst_drop(skb); |
1da177e4 LT |
687 | #ifdef CONFIG_XFRM |
688 | secpath_put(skb->sp); | |
689 | #endif | |
9c2b3328 SH |
690 | if (skb->destructor) { |
691 | WARN_ON(in_irq()); | |
1da177e4 LT |
692 | skb->destructor(skb); |
693 | } | |
a3bf7ae9 | 694 | #if IS_ENABLED(CONFIG_NF_CONNTRACK) |
5f79e0f9 | 695 | nf_conntrack_put(skb->nfct); |
2fc72c7b | 696 | #endif |
1109a90c | 697 | #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) |
1da177e4 LT |
698 | nf_bridge_put(skb->nf_bridge); |
699 | #endif | |
04a4bb55 LB |
700 | } |
701 | ||
702 | /* Free everything but the sk_buff shell. */ | |
703 | static void skb_release_all(struct sk_buff *skb) | |
704 | { | |
705 | skb_release_head_state(skb); | |
5e71d9d7 | 706 | if (likely(skb->head)) |
0ebd0ac5 | 707 | skb_release_data(skb); |
2d4baff8 HX |
708 | } |
709 | ||
710 | /** | |
711 | * __kfree_skb - private function | |
712 | * @skb: buffer | |
713 | * | |
714 | * Free an sk_buff. Release anything attached to the buffer. | |
715 | * Clean the state. This is an internal helper function. Users should | |
716 | * always call kfree_skb | |
717 | */ | |
1da177e4 | 718 | |
2d4baff8 HX |
719 | void __kfree_skb(struct sk_buff *skb) |
720 | { | |
721 | skb_release_all(skb); | |
1da177e4 LT |
722 | kfree_skbmem(skb); |
723 | } | |
b4ac530f | 724 | EXPORT_SYMBOL(__kfree_skb); |
1da177e4 | 725 | |