]>
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> | |
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> | |
de960aa9 FW |
49 | #include <linux/tcp.h> |
50 | #include <linux/udp.h> | |
90017acc | 51 | #include <linux/sctp.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 | ||
7c0f6ba6 | 74 | #include <linux/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 | |
08009a76 AD |
80 | struct kmem_cache *skbuff_head_cache __ro_after_init; |
81 | static struct kmem_cache *skbuff_fclone_cache __ro_after_init; | |
df5042f4 FW |
82 | #ifdef CONFIG_SKB_EXTENSIONS |
83 | static struct kmem_cache *skbuff_ext_cache __ro_after_init; | |
84 | #endif | |
5f74f82e HWR |
85 | int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS; |
86 | EXPORT_SYMBOL(sysctl_max_skb_frags); | |
1da177e4 | 87 | |
1da177e4 | 88 | /** |
f05de73b JS |
89 | * skb_panic - private function for out-of-line support |
90 | * @skb: buffer | |
91 | * @sz: size | |
92 | * @addr: address | |
99d5851e | 93 | * @msg: skb_over_panic or skb_under_panic |
1da177e4 | 94 | * |
f05de73b JS |
95 | * Out-of-line support for skb_put() and skb_push(). |
96 | * Called via the wrapper skb_over_panic() or skb_under_panic(). | |
97 | * Keep out of line to prevent kernel bloat. | |
98 | * __builtin_return_address is not used because it is not always reliable. | |
1da177e4 | 99 | */ |
f05de73b | 100 | static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr, |
99d5851e | 101 | const char msg[]) |
1da177e4 | 102 | { |
e005d193 | 103 | pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n", |
99d5851e | 104 | msg, addr, skb->len, sz, skb->head, skb->data, |
e005d193 JP |
105 | (unsigned long)skb->tail, (unsigned long)skb->end, |
106 | skb->dev ? skb->dev->name : "<NULL>"); | |
1da177e4 LT |
107 | BUG(); |
108 | } | |
109 | ||
f05de73b | 110 | static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr) |
1da177e4 | 111 | { |
f05de73b | 112 | skb_panic(skb, sz, addr, __func__); |
1da177e4 LT |
113 | } |
114 | ||
f05de73b JS |
115 | static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr) |
116 | { | |
117 | skb_panic(skb, sz, addr, __func__); | |
118 | } | |
c93bdd0e MG |
119 | |
120 | /* | |
121 | * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells | |
122 | * the caller if emergency pfmemalloc reserves are being used. If it is and | |
123 | * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves | |
124 | * may be used. Otherwise, the packet data may be discarded until enough | |
125 | * memory is free | |
126 | */ | |
127 | #define kmalloc_reserve(size, gfp, node, pfmemalloc) \ | |
128 | __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc) | |
61c5e88a | 129 | |
130 | static void *__kmalloc_reserve(size_t size, gfp_t flags, int node, | |
131 | unsigned long ip, bool *pfmemalloc) | |
c93bdd0e MG |
132 | { |
133 | void *obj; | |
134 | bool ret_pfmemalloc = false; | |
135 | ||
136 | /* | |
137 | * Try a regular allocation, when that fails and we're not entitled | |
138 | * to the reserves, fail. | |
139 | */ | |
140 | obj = kmalloc_node_track_caller(size, | |
141 | flags | __GFP_NOMEMALLOC | __GFP_NOWARN, | |
142 | node); | |
143 | if (obj || !(gfp_pfmemalloc_allowed(flags))) | |
144 | goto out; | |
145 | ||
146 | /* Try again but now we are using pfmemalloc reserves */ | |
147 | ret_pfmemalloc = true; | |
148 | obj = kmalloc_node_track_caller(size, flags, node); | |
149 | ||
150 | out: | |
151 | if (pfmemalloc) | |
152 | *pfmemalloc = ret_pfmemalloc; | |
153 | ||
154 | return obj; | |
155 | } | |
156 | ||
1da177e4 LT |
157 | /* Allocate a new skbuff. We do this ourselves so we can fill in a few |
158 | * 'private' fields and also do memory statistics to find all the | |
159 | * [BEEP] leaks. | |
160 | * | |
161 | */ | |
162 | ||
163 | /** | |
d179cd12 | 164 | * __alloc_skb - allocate a network buffer |
1da177e4 LT |
165 | * @size: size to allocate |
166 | * @gfp_mask: allocation mask | |
c93bdd0e MG |
167 | * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache |
168 | * instead of head cache and allocate a cloned (child) skb. | |
169 | * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for | |
170 | * allocations in case the data is required for writeback | |
b30973f8 | 171 | * @node: numa node to allocate memory on |
1da177e4 LT |
172 | * |
173 | * Allocate a new &sk_buff. The returned buffer has no headroom and a | |
94b6042c BH |
174 | * tail room of at least size bytes. The object has a reference count |
175 | * of one. The return is the buffer. On a failure the return is %NULL. | |
1da177e4 LT |
176 | * |
177 | * Buffers may only be allocated from interrupts using a @gfp_mask of | |
178 | * %GFP_ATOMIC. | |
179 | */ | |
dd0fc66f | 180 | struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, |
c93bdd0e | 181 | int flags, int node) |
1da177e4 | 182 | { |
e18b890b | 183 | struct kmem_cache *cache; |
4947d3ef | 184 | struct skb_shared_info *shinfo; |
1da177e4 LT |
185 | struct sk_buff *skb; |
186 | u8 *data; | |
c93bdd0e | 187 | bool pfmemalloc; |
1da177e4 | 188 | |
c93bdd0e MG |
189 | cache = (flags & SKB_ALLOC_FCLONE) |
190 | ? skbuff_fclone_cache : skbuff_head_cache; | |
191 | ||
192 | if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX)) | |
193 | gfp_mask |= __GFP_MEMALLOC; | |
8798b3fb | 194 | |
1da177e4 | 195 | /* Get the HEAD */ |
b30973f8 | 196 | skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node); |
1da177e4 LT |
197 | if (!skb) |
198 | goto out; | |
ec7d2f2c | 199 | prefetchw(skb); |
1da177e4 | 200 | |
87fb4b7b ED |
201 | /* We do our best to align skb_shared_info on a separate cache |
202 | * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives | |
203 | * aligned memory blocks, unless SLUB/SLAB debug is enabled. | |
204 | * Both skb->head and skb_shared_info are cache line aligned. | |
205 | */ | |
bc417e30 | 206 | size = SKB_DATA_ALIGN(size); |
87fb4b7b | 207 | size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); |
c93bdd0e | 208 | data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc); |
1da177e4 LT |
209 | if (!data) |
210 | goto nodata; | |
87fb4b7b ED |
211 | /* kmalloc(size) might give us more room than requested. |
212 | * Put skb_shared_info exactly at the end of allocated zone, | |
213 | * to allow max possible filling before reallocation. | |
214 | */ | |
215 | size = SKB_WITH_OVERHEAD(ksize(data)); | |
ec7d2f2c | 216 | prefetchw(data + size); |
1da177e4 | 217 | |
ca0605a7 | 218 | /* |
c8005785 JB |
219 | * Only clear those fields we need to clear, not those that we will |
220 | * actually initialise below. Hence, don't put any more fields after | |
221 | * the tail pointer in struct sk_buff! | |
ca0605a7 ACM |
222 | */ |
223 | memset(skb, 0, offsetof(struct sk_buff, tail)); | |
87fb4b7b ED |
224 | /* Account for allocated memory : skb + skb->head */ |
225 | skb->truesize = SKB_TRUESIZE(size); | |
c93bdd0e | 226 | skb->pfmemalloc = pfmemalloc; |
63354797 | 227 | refcount_set(&skb->users, 1); |
1da177e4 LT |
228 | skb->head = data; |
229 | skb->data = data; | |
27a884dc | 230 | skb_reset_tail_pointer(skb); |
4305b541 | 231 | skb->end = skb->tail + size; |
35d04610 CW |
232 | skb->mac_header = (typeof(skb->mac_header))~0U; |
233 | skb->transport_header = (typeof(skb->transport_header))~0U; | |
19633e12 | 234 | |
4947d3ef BL |
235 | /* make sure we initialize shinfo sequentially */ |
236 | shinfo = skb_shinfo(skb); | |
ec7d2f2c | 237 | memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); |
4947d3ef | 238 | atomic_set(&shinfo->dataref, 1); |
4947d3ef | 239 | |
c93bdd0e | 240 | if (flags & SKB_ALLOC_FCLONE) { |
d0bf4a9e | 241 | struct sk_buff_fclones *fclones; |
1da177e4 | 242 | |
d0bf4a9e ED |
243 | fclones = container_of(skb, struct sk_buff_fclones, skb1); |
244 | ||
d179cd12 | 245 | skb->fclone = SKB_FCLONE_ORIG; |
2638595a | 246 | refcount_set(&fclones->fclone_ref, 1); |
d179cd12 | 247 | |
6ffe75eb | 248 | fclones->skb2.fclone = SKB_FCLONE_CLONE; |
d179cd12 | 249 | } |
1da177e4 LT |
250 | out: |
251 | return skb; | |
252 | nodata: | |
8798b3fb | 253 | kmem_cache_free(cache, skb); |
1da177e4 LT |
254 | skb = NULL; |
255 | goto out; | |
1da177e4 | 256 | } |
b4ac530f | 257 | EXPORT_SYMBOL(__alloc_skb); |
1da177e4 | 258 | |
b2b5ce9d | 259 | /** |
2ea2f62c | 260 | * __build_skb - build a network buffer |
b2b5ce9d | 261 | * @data: data buffer provided by caller |
2ea2f62c | 262 | * @frag_size: size of data, or 0 if head was kmalloced |
b2b5ce9d ED |
263 | * |
264 | * Allocate a new &sk_buff. Caller provides space holding head and | |
deceb4c0 | 265 | * skb_shared_info. @data must have been allocated by kmalloc() only if |
2ea2f62c ED |
266 | * @frag_size is 0, otherwise data should come from the page allocator |
267 | * or vmalloc() | |
b2b5ce9d ED |
268 | * The return is the new skb buffer. |
269 | * On a failure the return is %NULL, and @data is not freed. | |
270 | * Notes : | |
271 | * Before IO, driver allocates only data buffer where NIC put incoming frame | |
272 | * Driver should add room at head (NET_SKB_PAD) and | |
273 | * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info)) | |
274 | * After IO, driver calls build_skb(), to allocate sk_buff and populate it | |
275 | * before giving packet to stack. | |
276 | * RX rings only contains data buffers, not full skbs. | |
277 | */ | |
2ea2f62c | 278 | struct sk_buff *__build_skb(void *data, unsigned int frag_size) |
b2b5ce9d ED |
279 | { |
280 | struct skb_shared_info *shinfo; | |
281 | struct sk_buff *skb; | |
d3836f21 | 282 | unsigned int size = frag_size ? : ksize(data); |
b2b5ce9d ED |
283 | |
284 | skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC); | |
285 | if (!skb) | |
286 | return NULL; | |
287 | ||
d3836f21 | 288 | size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); |
b2b5ce9d ED |
289 | |
290 | memset(skb, 0, offsetof(struct sk_buff, tail)); | |
291 | skb->truesize = SKB_TRUESIZE(size); | |
63354797 | 292 | refcount_set(&skb->users, 1); |
b2b5ce9d ED |
293 | skb->head = data; |
294 | skb->data = data; | |
295 | skb_reset_tail_pointer(skb); | |
296 | skb->end = skb->tail + size; | |
35d04610 CW |
297 | skb->mac_header = (typeof(skb->mac_header))~0U; |
298 | skb->transport_header = (typeof(skb->transport_header))~0U; | |
b2b5ce9d ED |
299 | |
300 | /* make sure we initialize shinfo sequentially */ | |
301 | shinfo = skb_shinfo(skb); | |
302 | memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); | |
303 | atomic_set(&shinfo->dataref, 1); | |
b2b5ce9d ED |
304 | |
305 | return skb; | |
306 | } | |
2ea2f62c ED |
307 | |
308 | /* build_skb() is wrapper over __build_skb(), that specifically | |
309 | * takes care of skb->head and skb->pfmemalloc | |
310 | * This means that if @frag_size is not zero, then @data must be backed | |
311 | * by a page fragment, not kmalloc() or vmalloc() | |
312 | */ | |
313 | struct sk_buff *build_skb(void *data, unsigned int frag_size) | |
314 | { | |
315 | struct sk_buff *skb = __build_skb(data, frag_size); | |
316 | ||
317 | if (skb && frag_size) { | |
318 | skb->head_frag = 1; | |
2f064f34 | 319 | if (page_is_pfmemalloc(virt_to_head_page(data))) |
2ea2f62c ED |
320 | skb->pfmemalloc = 1; |
321 | } | |
322 | return skb; | |
323 | } | |
b2b5ce9d ED |
324 | EXPORT_SYMBOL(build_skb); |
325 | ||
795bb1c0 JDB |
326 | #define NAPI_SKB_CACHE_SIZE 64 |
327 | ||
328 | struct napi_alloc_cache { | |
329 | struct page_frag_cache page; | |
e0d7924a | 330 | unsigned int skb_count; |
795bb1c0 JDB |
331 | void *skb_cache[NAPI_SKB_CACHE_SIZE]; |
332 | }; | |
333 | ||
b63ae8ca | 334 | static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache); |
795bb1c0 | 335 | static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache); |
ffde7328 AD |
336 | |
337 | static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask) | |
338 | { | |
b63ae8ca | 339 | struct page_frag_cache *nc; |
ffde7328 AD |
340 | unsigned long flags; |
341 | void *data; | |
342 | ||
343 | local_irq_save(flags); | |
9451980a | 344 | nc = this_cpu_ptr(&netdev_alloc_cache); |
8c2dd3e4 | 345 | data = page_frag_alloc(nc, fragsz, gfp_mask); |
6f532612 ED |
346 | local_irq_restore(flags); |
347 | return data; | |
348 | } | |
c93bdd0e MG |
349 | |
350 | /** | |
351 | * netdev_alloc_frag - allocate a page fragment | |
352 | * @fragsz: fragment size | |
353 | * | |
354 | * Allocates a frag from a page for receive buffer. | |
355 | * Uses GFP_ATOMIC allocations. | |
356 | */ | |
357 | void *netdev_alloc_frag(unsigned int fragsz) | |
358 | { | |
3bed3cc4 AD |
359 | fragsz = SKB_DATA_ALIGN(fragsz); |
360 | ||
453f85d4 | 361 | return __netdev_alloc_frag(fragsz, GFP_ATOMIC); |
c93bdd0e | 362 | } |
6f532612 ED |
363 | EXPORT_SYMBOL(netdev_alloc_frag); |
364 | ||
ffde7328 AD |
365 | static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask) |
366 | { | |
795bb1c0 | 367 | struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); |
9451980a | 368 | |
8c2dd3e4 | 369 | return page_frag_alloc(&nc->page, fragsz, gfp_mask); |
ffde7328 AD |
370 | } |
371 | ||
372 | void *napi_alloc_frag(unsigned int fragsz) | |
373 | { | |
3bed3cc4 AD |
374 | fragsz = SKB_DATA_ALIGN(fragsz); |
375 | ||
453f85d4 | 376 | return __napi_alloc_frag(fragsz, GFP_ATOMIC); |
ffde7328 AD |
377 | } |
378 | EXPORT_SYMBOL(napi_alloc_frag); | |
379 | ||
fd11a83d AD |
380 | /** |
381 | * __netdev_alloc_skb - allocate an skbuff for rx on a specific device | |
382 | * @dev: network device to receive on | |
d7499160 | 383 | * @len: length to allocate |
fd11a83d AD |
384 | * @gfp_mask: get_free_pages mask, passed to alloc_skb |
385 | * | |
386 | * Allocate a new &sk_buff and assign it a usage count of one. The | |
387 | * buffer has NET_SKB_PAD headroom built in. Users should allocate | |
388 | * the headroom they think they need without accounting for the | |
389 | * built in space. The built in space is used for optimisations. | |
390 | * | |
391 | * %NULL is returned if there is no free memory. | |
392 | */ | |
9451980a AD |
393 | struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len, |
394 | gfp_t gfp_mask) | |
fd11a83d | 395 | { |
b63ae8ca | 396 | struct page_frag_cache *nc; |
9451980a | 397 | unsigned long flags; |
fd11a83d | 398 | struct sk_buff *skb; |
9451980a AD |
399 | bool pfmemalloc; |
400 | void *data; | |
401 | ||
402 | len += NET_SKB_PAD; | |
fd11a83d | 403 | |
9451980a | 404 | if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) || |
d0164adc | 405 | (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { |
a080e7bd AD |
406 | skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); |
407 | if (!skb) | |
408 | goto skb_fail; | |
409 | goto skb_success; | |
410 | } | |
fd11a83d | 411 | |
9451980a AD |
412 | len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); |
413 | len = SKB_DATA_ALIGN(len); | |
414 | ||
415 | if (sk_memalloc_socks()) | |
416 | gfp_mask |= __GFP_MEMALLOC; | |
417 | ||
418 | local_irq_save(flags); | |
419 | ||
420 | nc = this_cpu_ptr(&netdev_alloc_cache); | |
8c2dd3e4 | 421 | data = page_frag_alloc(nc, len, gfp_mask); |
9451980a AD |
422 | pfmemalloc = nc->pfmemalloc; |
423 | ||
424 | local_irq_restore(flags); | |
425 | ||
426 | if (unlikely(!data)) | |
427 | return NULL; | |
428 | ||
429 | skb = __build_skb(data, len); | |
430 | if (unlikely(!skb)) { | |
181edb2b | 431 | skb_free_frag(data); |
9451980a | 432 | return NULL; |
7b2e497a | 433 | } |
fd11a83d | 434 | |
9451980a AD |
435 | /* use OR instead of assignment to avoid clearing of bits in mask */ |
436 | if (pfmemalloc) | |
437 | skb->pfmemalloc = 1; | |
438 | skb->head_frag = 1; | |
439 | ||
a080e7bd | 440 | skb_success: |
9451980a AD |
441 | skb_reserve(skb, NET_SKB_PAD); |
442 | skb->dev = dev; | |
443 | ||
a080e7bd | 444 | skb_fail: |
8af27456 CH |
445 | return skb; |
446 | } | |
b4ac530f | 447 | EXPORT_SYMBOL(__netdev_alloc_skb); |
1da177e4 | 448 | |
fd11a83d AD |
449 | /** |
450 | * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance | |
451 | * @napi: napi instance this buffer was allocated for | |
d7499160 | 452 | * @len: length to allocate |
fd11a83d AD |
453 | * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages |
454 | * | |
455 | * Allocate a new sk_buff for use in NAPI receive. This buffer will | |
456 | * attempt to allocate the head from a special reserved region used | |
457 | * only for NAPI Rx allocation. By doing this we can save several | |
458 | * CPU cycles by avoiding having to disable and re-enable IRQs. | |
459 | * | |
460 | * %NULL is returned if there is no free memory. | |
461 | */ | |
9451980a AD |
462 | struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len, |
463 | gfp_t gfp_mask) | |
fd11a83d | 464 | { |
795bb1c0 | 465 | struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); |
fd11a83d | 466 | struct sk_buff *skb; |
9451980a AD |
467 | void *data; |
468 | ||
469 | len += NET_SKB_PAD + NET_IP_ALIGN; | |
fd11a83d | 470 | |
9451980a | 471 | if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) || |
d0164adc | 472 | (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { |
a080e7bd AD |
473 | skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); |
474 | if (!skb) | |
475 | goto skb_fail; | |
476 | goto skb_success; | |
477 | } | |
9451980a AD |
478 | |
479 | len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
480 | len = SKB_DATA_ALIGN(len); | |
481 | ||
482 | if (sk_memalloc_socks()) | |
483 | gfp_mask |= __GFP_MEMALLOC; | |
fd11a83d | 484 | |
8c2dd3e4 | 485 | data = page_frag_alloc(&nc->page, len, gfp_mask); |
9451980a AD |
486 | if (unlikely(!data)) |
487 | return NULL; | |
488 | ||
489 | skb = __build_skb(data, len); | |
490 | if (unlikely(!skb)) { | |
181edb2b | 491 | skb_free_frag(data); |
9451980a | 492 | return NULL; |
fd11a83d AD |
493 | } |
494 | ||
9451980a | 495 | /* use OR instead of assignment to avoid clearing of bits in mask */ |
795bb1c0 | 496 | if (nc->page.pfmemalloc) |
9451980a AD |
497 | skb->pfmemalloc = 1; |
498 | skb->head_frag = 1; | |
499 | ||
a080e7bd | 500 | skb_success: |
9451980a AD |
501 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); |
502 | skb->dev = napi->dev; | |
503 | ||
a080e7bd | 504 | skb_fail: |
fd11a83d AD |
505 | return skb; |
506 | } | |
507 | EXPORT_SYMBOL(__napi_alloc_skb); | |
508 | ||
654bed16 | 509 | void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, |
50269e19 | 510 | int size, unsigned int truesize) |
654bed16 PZ |
511 | { |
512 | skb_fill_page_desc(skb, i, page, off, size); | |
513 | skb->len += size; | |
514 | skb->data_len += size; | |
50269e19 | 515 | skb->truesize += truesize; |
654bed16 PZ |
516 | } |
517 | EXPORT_SYMBOL(skb_add_rx_frag); | |
518 | ||
f8e617e1 JW |
519 | void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, |
520 | unsigned int truesize) | |
521 | { | |
522 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
523 | ||
524 | skb_frag_size_add(frag, size); | |
525 | skb->len += size; | |
526 | skb->data_len += size; | |
527 | skb->truesize += truesize; | |
528 | } | |
529 | EXPORT_SYMBOL(skb_coalesce_rx_frag); | |
530 | ||
27b437c8 | 531 | static void skb_drop_list(struct sk_buff **listp) |
1da177e4 | 532 | { |
bd8a7036 | 533 | kfree_skb_list(*listp); |
27b437c8 | 534 | *listp = NULL; |
1da177e4 LT |
535 | } |
536 | ||
27b437c8 HX |
537 | static inline void skb_drop_fraglist(struct sk_buff *skb) |
538 | { | |
539 | skb_drop_list(&skb_shinfo(skb)->frag_list); | |
540 | } | |
541 | ||
1da177e4 LT |
542 | static void skb_clone_fraglist(struct sk_buff *skb) |
543 | { | |
544 | struct sk_buff *list; | |
545 | ||
fbb398a8 | 546 | skb_walk_frags(skb, list) |
1da177e4 LT |
547 | skb_get(list); |
548 | } | |
549 | ||
d3836f21 ED |
550 | static void skb_free_head(struct sk_buff *skb) |
551 | { | |
181edb2b AD |
552 | unsigned char *head = skb->head; |
553 | ||
d3836f21 | 554 | if (skb->head_frag) |
181edb2b | 555 | skb_free_frag(head); |
d3836f21 | 556 | else |
181edb2b | 557 | kfree(head); |
d3836f21 ED |
558 | } |
559 | ||
5bba1712 | 560 | static void skb_release_data(struct sk_buff *skb) |
1da177e4 | 561 | { |
ff04a771 ED |
562 | struct skb_shared_info *shinfo = skb_shinfo(skb); |
563 | int i; | |
1da177e4 | 564 | |
ff04a771 ED |
565 | if (skb->cloned && |
566 | atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, | |
567 | &shinfo->dataref)) | |
568 | return; | |
a6686f2f | 569 | |
ff04a771 ED |
570 | for (i = 0; i < shinfo->nr_frags; i++) |
571 | __skb_frag_unref(&shinfo->frags[i]); | |
a6686f2f | 572 | |
ff04a771 ED |
573 | if (shinfo->frag_list) |
574 | kfree_skb_list(shinfo->frag_list); | |
575 | ||
1f8b977a | 576 | skb_zcopy_clear(skb, true); |
ff04a771 | 577 | skb_free_head(skb); |
1da177e4 LT |
578 | } |
579 | ||
580 | /* | |
581 | * Free an skbuff by memory without cleaning the state. | |
582 | */ | |
2d4baff8 | 583 | static void kfree_skbmem(struct sk_buff *skb) |
1da177e4 | 584 | { |
d0bf4a9e | 585 | struct sk_buff_fclones *fclones; |
d179cd12 | 586 | |
d179cd12 DM |
587 | switch (skb->fclone) { |
588 | case SKB_FCLONE_UNAVAILABLE: | |
589 | kmem_cache_free(skbuff_head_cache, skb); | |
6ffe75eb | 590 | return; |
d179cd12 DM |
591 | |
592 | case SKB_FCLONE_ORIG: | |
d0bf4a9e | 593 | fclones = container_of(skb, struct sk_buff_fclones, skb1); |
d179cd12 | 594 | |
6ffe75eb ED |
595 | /* We usually free the clone (TX completion) before original skb |
596 | * This test would have no chance to be true for the clone, | |
597 | * while here, branch prediction will be good. | |
d179cd12 | 598 | */ |
2638595a | 599 | if (refcount_read(&fclones->fclone_ref) == 1) |
6ffe75eb ED |
600 | goto fastpath; |
601 | break; | |
e7820e39 | 602 | |
6ffe75eb ED |
603 | default: /* SKB_FCLONE_CLONE */ |
604 | fclones = container_of(skb, struct sk_buff_fclones, skb2); | |
d179cd12 | 605 | break; |
3ff50b79 | 606 | } |
2638595a | 607 | if (!refcount_dec_and_test(&fclones->fclone_ref)) |
6ffe75eb ED |
608 | return; |
609 | fastpath: | |
610 | kmem_cache_free(skbuff_fclone_cache, fclones); | |
1da177e4 LT |
611 | } |
612 | ||
0a463c78 | 613 | void skb_release_head_state(struct sk_buff *skb) |
1da177e4 | 614 | { |
adf30907 | 615 | skb_dst_drop(skb); |
9c2b3328 SH |
616 | if (skb->destructor) { |
617 | WARN_ON(in_irq()); | |
1da177e4 LT |
618 | skb->destructor(skb); |
619 | } | |
a3bf7ae9 | 620 | #if IS_ENABLED(CONFIG_NF_CONNTRACK) |
cb9c6836 | 621 | nf_conntrack_put(skb_nfct(skb)); |
1da177e4 | 622 | #endif |
df5042f4 | 623 | skb_ext_put(skb); |
04a4bb55 LB |
624 | } |
625 | ||
626 | /* Free everything but the sk_buff shell. */ | |
627 | static void skb_release_all(struct sk_buff *skb) | |
628 | { | |
629 | skb_release_head_state(skb); | |
a28b1b90 FW |
630 | if (likely(skb->head)) |
631 | skb_release_data(skb); | |
2d4baff8 HX |
632 | } |
633 | ||
634 | /** | |
635 | * __kfree_skb - private function | |
636 | * @skb: buffer | |
637 | * | |
638 | * Free an sk_buff. Release anything attached to the buffer. | |
639 | * Clean the state. This is an internal helper function. Users should | |
640 | * always call kfree_skb | |
641 | */ | |
1da177e4 | 642 | |
2d4baff8 HX |
643 | void __kfree_skb(struct sk_buff *skb) |
644 | { | |
645 | skb_release_all(skb); | |
1da177e4 LT |
646 | kfree_skbmem(skb); |
647 | } | |
b4ac530f | 648 | EXPORT_SYMBOL(__kfree_skb); |
1da177e4 | 649 | |