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> | |
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> | |
1cac41cb | 79 | #include <soc/samsung/exynos-modem-ctrl.h> |
a1f8e7f7 | 80 | |
d7e8883c | 81 | struct kmem_cache *skbuff_head_cache __read_mostly; |
e18b890b | 82 | static struct kmem_cache *skbuff_fclone_cache __read_mostly; |
1bec5f40 HWR |
83 | int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS; |
84 | EXPORT_SYMBOL(sysctl_max_skb_frags); | |
1da177e4 | 85 | |
1da177e4 | 86 | /** |
f05de73b JS |
87 | * skb_panic - private function for out-of-line support |
88 | * @skb: buffer | |
89 | * @sz: size | |
90 | * @addr: address | |
99d5851e | 91 | * @msg: skb_over_panic or skb_under_panic |
1da177e4 | 92 | * |
f05de73b JS |
93 | * Out-of-line support for skb_put() and skb_push(). |
94 | * Called via the wrapper skb_over_panic() or skb_under_panic(). | |
95 | * Keep out of line to prevent kernel bloat. | |
96 | * __builtin_return_address is not used because it is not always reliable. | |
1da177e4 | 97 | */ |
f05de73b | 98 | static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr, |
99d5851e | 99 | const char msg[]) |
1da177e4 | 100 | { |
e005d193 | 101 | pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n", |
99d5851e | 102 | msg, addr, skb->len, sz, skb->head, skb->data, |
e005d193 JP |
103 | (unsigned long)skb->tail, (unsigned long)skb->end, |
104 | skb->dev ? skb->dev->name : "<NULL>"); | |
1da177e4 LT |
105 | BUG(); |
106 | } | |
107 | ||
f05de73b | 108 | static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr) |
1da177e4 | 109 | { |
f05de73b | 110 | skb_panic(skb, sz, addr, __func__); |
1da177e4 LT |
111 | } |
112 | ||
f05de73b JS |
113 | static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr) |
114 | { | |
115 | skb_panic(skb, sz, addr, __func__); | |
116 | } | |
c93bdd0e MG |
117 | |
118 | /* | |
119 | * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells | |
120 | * the caller if emergency pfmemalloc reserves are being used. If it is and | |
121 | * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves | |
122 | * may be used. Otherwise, the packet data may be discarded until enough | |
123 | * memory is free | |
124 | */ | |
125 | #define kmalloc_reserve(size, gfp, node, pfmemalloc) \ | |
126 | __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc) | |
61c5e88a | 127 | |
128 | static void *__kmalloc_reserve(size_t size, gfp_t flags, int node, | |
129 | unsigned long ip, bool *pfmemalloc) | |
c93bdd0e MG |
130 | { |
131 | void *obj; | |
132 | bool ret_pfmemalloc = false; | |
133 | ||
134 | /* | |
135 | * Try a regular allocation, when that fails and we're not entitled | |
136 | * to the reserves, fail. | |
137 | */ | |
138 | obj = kmalloc_node_track_caller(size, | |
139 | flags | __GFP_NOMEMALLOC | __GFP_NOWARN, | |
140 | node); | |
141 | if (obj || !(gfp_pfmemalloc_allowed(flags))) | |
142 | goto out; | |
143 | ||
144 | /* Try again but now we are using pfmemalloc reserves */ | |
145 | ret_pfmemalloc = true; | |
146 | obj = kmalloc_node_track_caller(size, flags, node); | |
147 | ||
148 | out: | |
149 | if (pfmemalloc) | |
150 | *pfmemalloc = ret_pfmemalloc; | |
151 | ||
152 | return obj; | |
153 | } | |
154 | ||
1da177e4 LT |
155 | /* Allocate a new skbuff. We do this ourselves so we can fill in a few |
156 | * 'private' fields and also do memory statistics to find all the | |
157 | * [BEEP] leaks. | |
158 | * | |
159 | */ | |
160 | ||
0ebd0ac5 PM |
161 | struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node) |
162 | { | |
163 | struct sk_buff *skb; | |
164 | ||
165 | /* Get the HEAD */ | |
166 | skb = kmem_cache_alloc_node(skbuff_head_cache, | |
167 | gfp_mask & ~__GFP_DMA, node); | |
168 | if (!skb) | |
169 | goto out; | |
170 | ||
171 | /* | |
172 | * Only clear those fields we need to clear, not those that we will | |
173 | * actually initialise below. Hence, don't put any more fields after | |
174 | * the tail pointer in struct sk_buff! | |
175 | */ | |
176 | memset(skb, 0, offsetof(struct sk_buff, tail)); | |
5e71d9d7 | 177 | skb->head = NULL; |
0ebd0ac5 PM |
178 | skb->truesize = sizeof(struct sk_buff); |
179 | atomic_set(&skb->users, 1); | |
180 | ||
35d04610 | 181 | skb->mac_header = (typeof(skb->mac_header))~0U; |
0ebd0ac5 PM |
182 | out: |
183 | return skb; | |
184 | } | |
185 | ||
1da177e4 | 186 | /** |
d179cd12 | 187 | * __alloc_skb - allocate a network buffer |
1da177e4 LT |
188 | * @size: size to allocate |
189 | * @gfp_mask: allocation mask | |
c93bdd0e MG |
190 | * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache |
191 | * instead of head cache and allocate a cloned (child) skb. | |
192 | * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for | |
193 | * allocations in case the data is required for writeback | |
b30973f8 | 194 | * @node: numa node to allocate memory on |
1da177e4 LT |
195 | * |
196 | * Allocate a new &sk_buff. The returned buffer has no headroom and a | |
94b6042c BH |
197 | * tail room of at least size bytes. The object has a reference count |
198 | * of one. The return is the buffer. On a failure the return is %NULL. | |
1da177e4 LT |
199 | * |
200 | * Buffers may only be allocated from interrupts using a @gfp_mask of | |
201 | * %GFP_ATOMIC. | |
202 | */ | |
dd0fc66f | 203 | struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, |
c93bdd0e | 204 | int flags, int node) |
1da177e4 | 205 | { |
e18b890b | 206 | struct kmem_cache *cache; |
4947d3ef | 207 | struct skb_shared_info *shinfo; |
1da177e4 LT |
208 | struct sk_buff *skb; |
209 | u8 *data; | |
c93bdd0e | 210 | bool pfmemalloc; |
1da177e4 | 211 | |
c93bdd0e MG |
212 | cache = (flags & SKB_ALLOC_FCLONE) |
213 | ? skbuff_fclone_cache : skbuff_head_cache; | |
214 | ||
215 | if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX)) | |
216 | gfp_mask |= __GFP_MEMALLOC; | |
8798b3fb | 217 | |
1da177e4 | 218 | /* Get the HEAD */ |
b30973f8 | 219 | skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node); |
1da177e4 LT |
220 | if (!skb) |
221 | goto out; | |
ec7d2f2c | 222 | prefetchw(skb); |
1da177e4 | 223 | |
87fb4b7b ED |
224 | /* We do our best to align skb_shared_info on a separate cache |
225 | * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives | |
226 | * aligned memory blocks, unless SLUB/SLAB debug is enabled. | |
227 | * Both skb->head and skb_shared_info are cache line aligned. | |
228 | */ | |
bc417e30 | 229 | size = SKB_DATA_ALIGN(size); |
87fb4b7b | 230 | size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); |
c93bdd0e | 231 | data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc); |
1da177e4 LT |
232 | if (!data) |
233 | goto nodata; | |
87fb4b7b ED |
234 | /* kmalloc(size) might give us more room than requested. |
235 | * Put skb_shared_info exactly at the end of allocated zone, | |
236 | * to allow max possible filling before reallocation. | |
237 | */ | |
238 | size = SKB_WITH_OVERHEAD(ksize(data)); | |
ec7d2f2c | 239 | prefetchw(data + size); |
1da177e4 | 240 | |
ca0605a7 | 241 | /* |
c8005785 JB |
242 | * Only clear those fields we need to clear, not those that we will |
243 | * actually initialise below. Hence, don't put any more fields after | |
244 | * the tail pointer in struct sk_buff! | |
ca0605a7 ACM |
245 | */ |
246 | memset(skb, 0, offsetof(struct sk_buff, tail)); | |
87fb4b7b ED |
247 | /* Account for allocated memory : skb + skb->head */ |
248 | skb->truesize = SKB_TRUESIZE(size); | |
c93bdd0e | 249 | skb->pfmemalloc = pfmemalloc; |
1da177e4 LT |
250 | atomic_set(&skb->users, 1); |
251 | skb->head = data; | |
252 | skb->data = data; | |
27a884dc | 253 | skb_reset_tail_pointer(skb); |
4305b541 | 254 | skb->end = skb->tail + size; |
35d04610 CW |
255 | skb->mac_header = (typeof(skb->mac_header))~0U; |
256 | skb->transport_header = (typeof(skb->transport_header))~0U; | |
19633e12 | 257 | |
4947d3ef BL |
258 | /* make sure we initialize shinfo sequentially */ |
259 | shinfo = skb_shinfo(skb); | |
ec7d2f2c | 260 | memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); |
4947d3ef | 261 | atomic_set(&shinfo->dataref, 1); |
c2aa3665 | 262 | kmemcheck_annotate_variable(shinfo->destructor_arg); |
4947d3ef | 263 | |
c93bdd0e | 264 | if (flags & SKB_ALLOC_FCLONE) { |
d0bf4a9e | 265 | struct sk_buff_fclones *fclones; |
1da177e4 | 266 | |
d0bf4a9e ED |
267 | fclones = container_of(skb, struct sk_buff_fclones, skb1); |
268 | ||
269 | kmemcheck_annotate_bitfield(&fclones->skb2, flags1); | |
d179cd12 | 270 | skb->fclone = SKB_FCLONE_ORIG; |
d0bf4a9e | 271 | atomic_set(&fclones->fclone_ref, 1); |
d179cd12 | 272 | |
6ffe75eb | 273 | fclones->skb2.fclone = SKB_FCLONE_CLONE; |
d0bf4a9e | 274 | fclones->skb2.pfmemalloc = pfmemalloc; |
d179cd12 | 275 | } |
1da177e4 LT |
276 | out: |
277 | return skb; | |
278 | nodata: | |
8798b3fb | 279 | kmem_cache_free(cache, skb); |
1da177e4 LT |
280 | skb = NULL; |
281 | goto out; | |
1da177e4 | 282 | } |
b4ac530f | 283 | EXPORT_SYMBOL(__alloc_skb); |
1da177e4 | 284 | |
b2b5ce9d | 285 | /** |
2ea2f62c | 286 | * __build_skb - build a network buffer |
b2b5ce9d | 287 | * @data: data buffer provided by caller |
2ea2f62c | 288 | * @frag_size: size of data, or 0 if head was kmalloced |
b2b5ce9d ED |
289 | * |
290 | * Allocate a new &sk_buff. Caller provides space holding head and | |
deceb4c0 | 291 | * skb_shared_info. @data must have been allocated by kmalloc() only if |
2ea2f62c ED |
292 | * @frag_size is 0, otherwise data should come from the page allocator |
293 | * or vmalloc() | |
b2b5ce9d ED |
294 | * The return is the new skb buffer. |
295 | * On a failure the return is %NULL, and @data is not freed. | |
296 | * Notes : | |
297 | * Before IO, driver allocates only data buffer where NIC put incoming frame | |
298 | * Driver should add room at head (NET_SKB_PAD) and | |
299 | * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info)) | |
300 | * After IO, driver calls build_skb(), to allocate sk_buff and populate it | |
301 | * before giving packet to stack. | |
302 | * RX rings only contains data buffers, not full skbs. | |
303 | */ | |
2ea2f62c | 304 | struct sk_buff *__build_skb(void *data, unsigned int frag_size) |
b2b5ce9d ED |
305 | { |
306 | struct skb_shared_info *shinfo; | |
307 | struct sk_buff *skb; | |
d3836f21 | 308 | unsigned int size = frag_size ? : ksize(data); |
b2b5ce9d ED |
309 | |
310 | skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC); | |
311 | if (!skb) | |
312 | return NULL; | |
313 | ||
d3836f21 | 314 | size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); |
b2b5ce9d ED |
315 | |
316 | memset(skb, 0, offsetof(struct sk_buff, tail)); | |
317 | skb->truesize = SKB_TRUESIZE(size); | |
318 | atomic_set(&skb->users, 1); | |
319 | skb->head = data; | |
320 | skb->data = data; | |
321 | skb_reset_tail_pointer(skb); | |
322 | skb->end = skb->tail + size; | |
35d04610 CW |
323 | skb->mac_header = (typeof(skb->mac_header))~0U; |
324 | skb->transport_header = (typeof(skb->transport_header))~0U; | |
b2b5ce9d ED |
325 | |
326 | /* make sure we initialize shinfo sequentially */ | |
327 | shinfo = skb_shinfo(skb); | |
328 | memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); | |
329 | atomic_set(&shinfo->dataref, 1); | |
330 | kmemcheck_annotate_variable(shinfo->destructor_arg); | |
331 | ||
332 | return skb; | |
333 | } | |
2ea2f62c ED |
334 | |
335 | /* build_skb() is wrapper over __build_skb(), that specifically | |
336 | * takes care of skb->head and skb->pfmemalloc | |
337 | * This means that if @frag_size is not zero, then @data must be backed | |
338 | * by a page fragment, not kmalloc() or vmalloc() | |
339 | */ | |
340 | struct sk_buff *build_skb(void *data, unsigned int frag_size) | |
341 | { | |
342 | struct sk_buff *skb = __build_skb(data, frag_size); | |
343 | ||
344 | if (skb && frag_size) { | |
345 | skb->head_frag = 1; | |
2f064f34 | 346 | if (page_is_pfmemalloc(virt_to_head_page(data))) |
2ea2f62c ED |
347 | skb->pfmemalloc = 1; |
348 | } | |
349 | return skb; | |
350 | } | |
b2b5ce9d ED |
351 | EXPORT_SYMBOL(build_skb); |
352 | ||
b63ae8ca AD |
353 | static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache); |
354 | static DEFINE_PER_CPU(struct page_frag_cache, napi_alloc_cache); | |
ffde7328 AD |
355 | |
356 | static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask) | |
357 | { | |
b63ae8ca | 358 | struct page_frag_cache *nc; |
ffde7328 AD |
359 | unsigned long flags; |
360 | void *data; | |
361 | ||
362 | local_irq_save(flags); | |
9451980a AD |
363 | nc = this_cpu_ptr(&netdev_alloc_cache); |
364 | data = __alloc_page_frag(nc, fragsz, gfp_mask); | |
6f532612 ED |
365 | local_irq_restore(flags); |
366 | return data; | |
367 | } | |
c93bdd0e MG |
368 | |
369 | /** | |
370 | * netdev_alloc_frag - allocate a page fragment | |
371 | * @fragsz: fragment size | |
372 | * | |
373 | * Allocates a frag from a page for receive buffer. | |
374 | * Uses GFP_ATOMIC allocations. | |
375 | */ | |
376 | void *netdev_alloc_frag(unsigned int fragsz) | |
377 | { | |
378 | return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD); | |
379 | } | |
6f532612 ED |
380 | EXPORT_SYMBOL(netdev_alloc_frag); |
381 | ||
ffde7328 AD |
382 | static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask) |
383 | { | |
b63ae8ca | 384 | struct page_frag_cache *nc = this_cpu_ptr(&napi_alloc_cache); |
9451980a AD |
385 | |
386 | return __alloc_page_frag(nc, fragsz, gfp_mask); | |
ffde7328 AD |
387 | } |
388 | ||
389 | void *napi_alloc_frag(unsigned int fragsz) | |
390 | { | |
391 | return __napi_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD); | |
392 | } | |
393 | EXPORT_SYMBOL(napi_alloc_frag); | |
394 | ||
fd11a83d AD |
395 | /** |
396 | * __netdev_alloc_skb - allocate an skbuff for rx on a specific device | |
397 | * @dev: network device to receive on | |
d7499160 | 398 | * @len: length to allocate |
fd11a83d AD |
399 | * @gfp_mask: get_free_pages mask, passed to alloc_skb |
400 | * | |
401 | * Allocate a new &sk_buff and assign it a usage count of one. The | |
402 | * buffer has NET_SKB_PAD headroom built in. Users should allocate | |
403 | * the headroom they think they need without accounting for the | |
404 | * built in space. The built in space is used for optimisations. | |
405 | * | |
406 | * %NULL is returned if there is no free memory. | |
407 | */ | |
9451980a AD |
408 | struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len, |
409 | gfp_t gfp_mask) | |
fd11a83d | 410 | { |
b63ae8ca | 411 | struct page_frag_cache *nc; |
9451980a | 412 | unsigned long flags; |
fd11a83d | 413 | struct sk_buff *skb; |
9451980a AD |
414 | bool pfmemalloc; |
415 | void *data; | |
416 | ||
417 | len += NET_SKB_PAD; | |
fd11a83d | 418 | |
9451980a | 419 | if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) || |
d0164adc | 420 | (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { |
a080e7bd AD |
421 | skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); |
422 | if (!skb) | |
423 | goto skb_fail; | |
424 | goto skb_success; | |
425 | } | |
fd11a83d | 426 | |
9451980a AD |
427 | len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); |
428 | len = SKB_DATA_ALIGN(len); | |
429 | ||
430 | if (sk_memalloc_socks()) | |
431 | gfp_mask |= __GFP_MEMALLOC; | |
432 | ||
433 | local_irq_save(flags); | |
434 | ||
435 | nc = this_cpu_ptr(&netdev_alloc_cache); | |
436 | data = __alloc_page_frag(nc, len, gfp_mask); | |
437 | pfmemalloc = nc->pfmemalloc; | |
438 | ||
439 | local_irq_restore(flags); | |
440 | ||
441 | if (unlikely(!data)) | |
442 | return NULL; | |
443 | ||
444 | skb = __build_skb(data, len); | |
445 | if (unlikely(!skb)) { | |
181edb2b | 446 | skb_free_frag(data); |
9451980a | 447 | return NULL; |
7b2e497a | 448 | } |
fd11a83d | 449 | |
9451980a AD |
450 | /* use OR instead of assignment to avoid clearing of bits in mask */ |
451 | if (pfmemalloc) | |
452 | skb->pfmemalloc = 1; | |
453 | skb->head_frag = 1; | |
454 | ||
a080e7bd | 455 | skb_success: |
9451980a AD |
456 | skb_reserve(skb, NET_SKB_PAD); |
457 | skb->dev = dev; | |
458 | ||
a080e7bd | 459 | skb_fail: |
8af27456 CH |
460 | return skb; |
461 | } | |
b4ac530f | 462 | EXPORT_SYMBOL(__netdev_alloc_skb); |
1da177e4 | 463 | |
fd11a83d AD |
464 | /** |
465 | * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance | |
466 | * @napi: napi instance this buffer was allocated for | |
d7499160 | 467 | * @len: length to allocate |
fd11a83d AD |
468 | * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages |
469 | * | |
470 | * Allocate a new sk_buff for use in NAPI receive. This buffer will | |
471 | * attempt to allocate the head from a special reserved region used | |
472 | * only for NAPI Rx allocation. By doing this we can save several | |
473 | * CPU cycles by avoiding having to disable and re-enable IRQs. | |
474 | * | |
475 | * %NULL is returned if there is no free memory. | |
476 | */ | |
9451980a AD |
477 | struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len, |
478 | gfp_t gfp_mask) | |
fd11a83d | 479 | { |
b63ae8ca | 480 | struct page_frag_cache *nc = this_cpu_ptr(&napi_alloc_cache); |
fd11a83d | 481 | struct sk_buff *skb; |
9451980a AD |
482 | void *data; |
483 | ||
484 | len += NET_SKB_PAD + NET_IP_ALIGN; | |
fd11a83d | 485 | |
9451980a | 486 | if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) || |
d0164adc | 487 | (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { |
a080e7bd AD |
488 | skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); |
489 | if (!skb) | |
490 | goto skb_fail; | |
491 | goto skb_success; | |
492 | } | |
9451980a AD |
493 | |
494 | len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
495 | len = SKB_DATA_ALIGN(len); | |
496 | ||
497 | if (sk_memalloc_socks()) | |
498 | gfp_mask |= __GFP_MEMALLOC; | |
fd11a83d | 499 | |
9451980a AD |
500 | data = __alloc_page_frag(nc, len, gfp_mask); |
501 | if (unlikely(!data)) | |
502 | return NULL; | |
503 | ||
504 | skb = __build_skb(data, len); | |
505 | if (unlikely(!skb)) { | |
181edb2b | 506 | skb_free_frag(data); |
9451980a | 507 | return NULL; |
fd11a83d AD |
508 | } |
509 | ||
9451980a AD |
510 | /* use OR instead of assignment to avoid clearing of bits in mask */ |
511 | if (nc->pfmemalloc) | |
512 | skb->pfmemalloc = 1; | |
513 | skb->head_frag = 1; | |
514 | ||
a080e7bd | 515 | skb_success: |
9451980a AD |
516 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); |
517 | skb->dev = napi->dev; | |
518 | ||
a080e7bd | 519 | skb_fail: |
fd11a83d AD |
520 | return skb; |
521 | } | |
522 | EXPORT_SYMBOL(__napi_alloc_skb); | |
523 | ||
654bed16 | 524 | void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, |
50269e19 | 525 | int size, unsigned int truesize) |
654bed16 PZ |
526 | { |
527 | skb_fill_page_desc(skb, i, page, off, size); | |
528 | skb->len += size; | |
529 | skb->data_len += size; | |
50269e19 | 530 | skb->truesize += truesize; |
654bed16 PZ |
531 | } |
532 | EXPORT_SYMBOL(skb_add_rx_frag); | |
533 | ||
f8e617e1 JW |
534 | void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, |
535 | unsigned int truesize) | |
536 | { | |
537 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
538 | ||
539 | skb_frag_size_add(frag, size); | |
540 | skb->len += size; | |
541 | skb->data_len += size; | |
542 | skb->truesize += truesize; | |
543 | } | |
544 | EXPORT_SYMBOL(skb_coalesce_rx_frag); | |
545 | ||
27b437c8 | 546 | static void skb_drop_list(struct sk_buff **listp) |
1da177e4 | 547 | { |
bd8a7036 | 548 | kfree_skb_list(*listp); |
27b437c8 | 549 | *listp = NULL; |
1da177e4 LT |
550 | } |
551 | ||
27b437c8 HX |
552 | static inline void skb_drop_fraglist(struct sk_buff *skb) |
553 | { | |
554 | skb_drop_list(&skb_shinfo(skb)->frag_list); | |
555 | } | |
556 | ||
1cac41cb MB |
557 | #ifndef CONFIG_MPTCP |
558 | static | |
559 | #endif | |
560 | void skb_clone_fraglist(struct sk_buff *skb) | |
1da177e4 LT |
561 | { |
562 | struct sk_buff *list; | |
563 | ||
fbb398a8 | 564 | skb_walk_frags(skb, list) |
1da177e4 LT |
565 | skb_get(list); |
566 | } | |
567 | ||
d3836f21 ED |
568 | static void skb_free_head(struct sk_buff *skb) |
569 | { | |
181edb2b AD |
570 | unsigned char *head = skb->head; |
571 | ||
1cac41cb MB |
572 | if (skb_free_head_cp_zerocopy(skb)) |
573 | return; | |
574 | ||
d3836f21 | 575 | if (skb->head_frag) |
181edb2b | 576 | skb_free_frag(head); |
d3836f21 | 577 | else |
181edb2b | 578 | kfree(head); |
d3836f21 ED |
579 | } |
580 | ||
5bba1712 | 581 | static void skb_release_data(struct sk_buff *skb) |
1da177e4 | 582 | { |
ff04a771 ED |
583 | struct skb_shared_info *shinfo = skb_shinfo(skb); |
584 | int i; | |
1da177e4 | 585 | |
ff04a771 ED |
586 | if (skb->cloned && |
587 | atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, | |
588 | &shinfo->dataref)) | |
589 | return; | |
a6686f2f | 590 | |
ff04a771 ED |
591 | for (i = 0; i < shinfo->nr_frags; i++) |
592 | __skb_frag_unref(&shinfo->frags[i]); | |
a6686f2f | 593 | |
ff04a771 ED |
594 | /* |
595 | * If skb buf is from userspace, we need to notify the caller | |
596 | * the lower device DMA has done; | |
597 | */ | |
598 | if (shinfo->tx_flags & SKBTX_DEV_ZEROCOPY) { | |
599 | struct ubuf_info *uarg; | |
1da177e4 | 600 | |
ff04a771 ED |
601 | uarg = shinfo->destructor_arg; |
602 | if (uarg->callback) | |
603 | uarg->callback(uarg, true); | |
1da177e4 | 604 | } |
ff04a771 ED |
605 | |
606 | if (shinfo->frag_list) | |
607 | kfree_skb_list(shinfo->frag_list); | |
608 | ||
609 | skb_free_head(skb); | |
1da177e4 LT |
610 | } |
611 | ||
612 | /* | |
613 | * Free an skbuff by memory without cleaning the state. | |
614 | */ | |
2d4baff8 | 615 | static void kfree_skbmem(struct sk_buff *skb) |
1da177e4 | 616 | { |
d0bf4a9e | 617 | struct sk_buff_fclones *fclones; |
d179cd12 | 618 | |
d179cd12 DM |
619 | switch (skb->fclone) { |
620 | case SKB_FCLONE_UNAVAILABLE: | |
621 | kmem_cache_free(skbuff_head_cache, skb); | |
6ffe75eb | 622 | return; |
d179cd12 DM |
623 | |
624 | case SKB_FCLONE_ORIG: | |
d0bf4a9e | 625 | fclones = container_of(skb, struct sk_buff_fclones, skb1); |
d179cd12 | 626 | |
6ffe75eb ED |
627 | /* We usually free the clone (TX completion) before original skb |
628 | * This test would have no chance to be true for the clone, | |
629 | * while here, branch prediction will be good. | |
d179cd12 | 630 | */ |
6ffe75eb ED |
631 | if (atomic_read(&fclones->fclone_ref) == 1) |
632 | goto fastpath; | |
633 | break; | |
e7820e39 | 634 | |
6ffe75eb ED |
635 | default: /* SKB_FCLONE_CLONE */ |
636 | fclones = container_of(skb, struct sk_buff_fclones, skb2); | |
d179cd12 | 637 | break; |
3ff50b79 | 638 | } |
6ffe75eb ED |
639 | if (!atomic_dec_and_test(&fclones->fclone_ref)) |
640 | return; | |
641 | fastpath: | |
642 | kmem_cache_free(skbuff_fclone_cache, fclones); | |
1da177e4 LT |
643 | } |
644 | ||
04a4bb55 | 645 | static void skb_release_head_state(struct sk_buff *skb) |
1da177e4 | 646 | { |
adf30907 | 647 | skb_dst_drop(skb); |
1da177e4 LT |
648 | #ifdef CONFIG_XFRM |
649 | secpath_put(skb->sp); | |
650 | #endif | |
9c2b3328 SH |
651 | if (skb->destructor) { |
652 | WARN_ON(in_irq()); | |
1da177e4 LT |
653 | skb->destructor(skb); |
654 | } | |
a3bf7ae9 | 655 | #if IS_ENABLED(CONFIG_NF_CONNTRACK) |
5f79e0f9 | 656 | nf_conntrack_put(skb->nfct); |
2fc72c7b | 657 | #endif |
1109a90c | 658 | #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) |
1da177e4 LT |
659 | nf_bridge_put(skb->nf_bridge); |
660 | #endif | |
04a4bb55 LB |
661 | } |
662 | ||
663 | /* Free everything but the sk_buff shell. */ | |
664 | static void skb_release_all(struct sk_buff *skb) | |
665 | { | |
666 | skb_release_head_state(skb); | |
5e71d9d7 | 667 | if (likely(skb->head)) |
0ebd0ac5 | 668 | skb_release_data(skb); |
2d4baff8 HX |
669 | } |
670 | ||
671 | /** | |
672 | * __kfree_skb - private function | |
673 | * @skb: buffer | |
674 | * | |
675 | * Free an sk_buff. Release anything attached to the buffer. | |
676 | * Clean the state. This is an internal helper function. Users should | |
677 | * always call kfree_skb | |
678 | */ | |
1da177e4 | 679 | |
2d4baff8 HX |
680 | void __kfree_skb(struct sk_buff *skb) |
681 | { | |
682 | skb_release_all(skb); | |
1da177e4 LT |
683 | kfree_skbmem(skb); |
684 | } | |
b4ac530f | 685 | EXPORT_SYMBOL(__kfree_skb); |
1da177e4 | 686 | |