Commit | Line | Data |
---|---|---|
2874c5fd | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 LT |
2 | /* |
3 | * Routines having to do with the 'struct sk_buff' memory handlers. | |
4 | * | |
113aa838 | 5 | * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk> |
1da177e4 LT |
6 | * Florian La Roche <rzsfl@rz.uni-sb.de> |
7 | * | |
1da177e4 LT |
8 | * Fixes: |
9 | * Alan Cox : Fixed the worst of the load | |
10 | * balancer bugs. | |
11 | * Dave Platt : Interrupt stacking fix. | |
12 | * Richard Kooijman : Timestamp fixes. | |
13 | * Alan Cox : Changed buffer format. | |
14 | * Alan Cox : destructor hook for AF_UNIX etc. | |
15 | * Linus Torvalds : Better skb_clone. | |
16 | * Alan Cox : Added skb_copy. | |
17 | * Alan Cox : Added all the changed routines Linus | |
18 | * only put in the headers | |
19 | * Ray VanTassle : Fixed --skb->lock in free | |
20 | * Alan Cox : skb_copy copy arp field | |
21 | * Andi Kleen : slabified it. | |
22 | * Robert Olsson : Removed skb_head_pool | |
23 | * | |
24 | * NOTE: | |
25 | * The __skb_ routines should be called with interrupts | |
26 | * disabled, or you better be *real* sure that the operation is atomic | |
27 | * with respect to whatever list is being frobbed (e.g. via lock_sock() | |
28 | * or via disabling bottom half handlers, etc). | |
1da177e4 LT |
29 | */ |
30 | ||
31 | /* | |
32 | * The functions in this file will not compile correctly with gcc 2.4.x | |
33 | */ | |
34 | ||
e005d193 JP |
35 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
36 | ||
1da177e4 LT |
37 | #include <linux/module.h> |
38 | #include <linux/types.h> | |
39 | #include <linux/kernel.h> | |
1da177e4 LT |
40 | #include <linux/mm.h> |
41 | #include <linux/interrupt.h> | |
42 | #include <linux/in.h> | |
43 | #include <linux/inet.h> | |
44 | #include <linux/slab.h> | |
de960aa9 FW |
45 | #include <linux/tcp.h> |
46 | #include <linux/udp.h> | |
90017acc | 47 | #include <linux/sctp.h> |
1da177e4 LT |
48 | #include <linux/netdevice.h> |
49 | #ifdef CONFIG_NET_CLS_ACT | |
50 | #include <net/pkt_sched.h> | |
51 | #endif | |
52 | #include <linux/string.h> | |
53 | #include <linux/skbuff.h> | |
9c55e01c | 54 | #include <linux/splice.h> |
1da177e4 LT |
55 | #include <linux/cache.h> |
56 | #include <linux/rtnetlink.h> | |
57 | #include <linux/init.h> | |
716ea3a7 | 58 | #include <linux/scatterlist.h> |
ac45f602 | 59 | #include <linux/errqueue.h> |
268bb0ce | 60 | #include <linux/prefetch.h> |
0d5501c1 | 61 | #include <linux/if_vlan.h> |
2a2ea508 | 62 | #include <linux/mpls.h> |
183f47fc | 63 | #include <linux/kcov.h> |
1da177e4 LT |
64 | |
65 | #include <net/protocol.h> | |
66 | #include <net/dst.h> | |
67 | #include <net/sock.h> | |
68 | #include <net/checksum.h> | |
ed1f50c3 | 69 | #include <net/ip6_checksum.h> |
1da177e4 | 70 | #include <net/xfrm.h> |
8822e270 | 71 | #include <net/mpls.h> |
3ee17bc7 | 72 | #include <net/mptcp.h> |
78476d31 | 73 | #include <net/mctp.h> |
6a5bcd84 | 74 | #include <net/page_pool.h> |
1da177e4 | 75 | |
7c0f6ba6 | 76 | #include <linux/uaccess.h> |
ad8d75ff | 77 | #include <trace/events/skb.h> |
51c56b00 | 78 | #include <linux/highmem.h> |
b245be1f WB |
79 | #include <linux/capability.h> |
80 | #include <linux/user_namespace.h> | |
2544af03 | 81 | #include <linux/indirect_call_wrapper.h> |
a1f8e7f7 | 82 | |
39564c3f | 83 | #include "dev.h" |
7f678def | 84 | #include "sock_destructor.h" |
7b7ed885 | 85 | |
08009a76 AD |
86 | struct kmem_cache *skbuff_head_cache __ro_after_init; |
87 | static struct kmem_cache *skbuff_fclone_cache __ro_after_init; | |
df5042f4 FW |
88 | #ifdef CONFIG_SKB_EXTENSIONS |
89 | static struct kmem_cache *skbuff_ext_cache __ro_after_init; | |
90 | #endif | |
5f74f82e HWR |
91 | int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS; |
92 | EXPORT_SYMBOL(sysctl_max_skb_frags); | |
1da177e4 | 93 | |
9cb252c4 MD |
94 | #undef FN |
95 | #define FN(reason) [SKB_DROP_REASON_##reason] = #reason, | |
96 | const char * const drop_reasons[] = { | |
97 | DEFINE_DROP_REASON(FN, FN) | |
98 | }; | |
ec43908d MD |
99 | EXPORT_SYMBOL(drop_reasons); |
100 | ||
1da177e4 | 101 | /** |
f05de73b JS |
102 | * skb_panic - private function for out-of-line support |
103 | * @skb: buffer | |
104 | * @sz: size | |
105 | * @addr: address | |
99d5851e | 106 | * @msg: skb_over_panic or skb_under_panic |
1da177e4 | 107 | * |
f05de73b JS |
108 | * Out-of-line support for skb_put() and skb_push(). |
109 | * Called via the wrapper skb_over_panic() or skb_under_panic(). | |
110 | * Keep out of line to prevent kernel bloat. | |
111 | * __builtin_return_address is not used because it is not always reliable. | |
1da177e4 | 112 | */ |
f05de73b | 113 | static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr, |
99d5851e | 114 | const char msg[]) |
1da177e4 | 115 | { |
41a46913 | 116 | pr_emerg("%s: text:%px len:%d put:%d head:%px data:%px tail:%#lx end:%#lx dev:%s\n", |
99d5851e | 117 | msg, addr, skb->len, sz, skb->head, skb->data, |
e005d193 JP |
118 | (unsigned long)skb->tail, (unsigned long)skb->end, |
119 | skb->dev ? skb->dev->name : "<NULL>"); | |
1da177e4 LT |
120 | BUG(); |
121 | } | |
122 | ||
f05de73b | 123 | static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr) |
1da177e4 | 124 | { |
f05de73b | 125 | skb_panic(skb, sz, addr, __func__); |
1da177e4 LT |
126 | } |
127 | ||
f05de73b JS |
128 | static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr) |
129 | { | |
130 | skb_panic(skb, sz, addr, __func__); | |
131 | } | |
c93bdd0e | 132 | |
50fad4b5 | 133 | #define NAPI_SKB_CACHE_SIZE 64 |
f450d539 AL |
134 | #define NAPI_SKB_CACHE_BULK 16 |
135 | #define NAPI_SKB_CACHE_HALF (NAPI_SKB_CACHE_SIZE / 2) | |
50fad4b5 AL |
136 | |
137 | struct napi_alloc_cache { | |
138 | struct page_frag_cache page; | |
139 | unsigned int skb_count; | |
140 | void *skb_cache[NAPI_SKB_CACHE_SIZE]; | |
141 | }; | |
142 | ||
143 | static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache); | |
144 | static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache); | |
145 | ||
32e3573f | 146 | void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask) |
50fad4b5 AL |
147 | { |
148 | struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); | |
149 | ||
50fad4b5 AL |
150 | fragsz = SKB_DATA_ALIGN(fragsz); |
151 | ||
32e3573f | 152 | return page_frag_alloc_align(&nc->page, fragsz, GFP_ATOMIC, align_mask); |
50fad4b5 AL |
153 | } |
154 | EXPORT_SYMBOL(__napi_alloc_frag_align); | |
155 | ||
156 | void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask) | |
157 | { | |
50fad4b5 AL |
158 | void *data; |
159 | ||
160 | fragsz = SKB_DATA_ALIGN(fragsz); | |
afa79d08 | 161 | if (in_hardirq() || irqs_disabled()) { |
32e3573f YD |
162 | struct page_frag_cache *nc = this_cpu_ptr(&netdev_alloc_cache); |
163 | ||
50fad4b5 AL |
164 | data = page_frag_alloc_align(nc, fragsz, GFP_ATOMIC, align_mask); |
165 | } else { | |
32e3573f YD |
166 | struct napi_alloc_cache *nc; |
167 | ||
50fad4b5 | 168 | local_bh_disable(); |
32e3573f YD |
169 | nc = this_cpu_ptr(&napi_alloc_cache); |
170 | data = page_frag_alloc_align(&nc->page, fragsz, GFP_ATOMIC, align_mask); | |
50fad4b5 AL |
171 | local_bh_enable(); |
172 | } | |
173 | return data; | |
174 | } | |
175 | EXPORT_SYMBOL(__netdev_alloc_frag_align); | |
176 | ||
f450d539 AL |
177 | static struct sk_buff *napi_skb_cache_get(void) |
178 | { | |
179 | struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); | |
180 | struct sk_buff *skb; | |
181 | ||
49ae83fc | 182 | if (unlikely(!nc->skb_count)) { |
f450d539 AL |
183 | nc->skb_count = kmem_cache_alloc_bulk(skbuff_head_cache, |
184 | GFP_ATOMIC, | |
185 | NAPI_SKB_CACHE_BULK, | |
186 | nc->skb_cache); | |
49ae83fc SPL |
187 | if (unlikely(!nc->skb_count)) |
188 | return NULL; | |
189 | } | |
f450d539 AL |
190 | |
191 | skb = nc->skb_cache[--nc->skb_count]; | |
192 | kasan_unpoison_object_data(skbuff_head_cache, skb); | |
193 | ||
194 | return skb; | |
195 | } | |
196 | ||
ba0509b6 | 197 | /* Caller must provide SKB that is memset cleared */ |
483126b3 AL |
198 | static void __build_skb_around(struct sk_buff *skb, void *data, |
199 | unsigned int frag_size) | |
ba0509b6 JDB |
200 | { |
201 | struct skb_shared_info *shinfo; | |
202 | unsigned int size = frag_size ? : ksize(data); | |
203 | ||
204 | size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
205 | ||
206 | /* Assumes caller memset cleared SKB */ | |
207 | skb->truesize = SKB_TRUESIZE(size); | |
208 | refcount_set(&skb->users, 1); | |
209 | skb->head = data; | |
210 | skb->data = data; | |
211 | skb_reset_tail_pointer(skb); | |
763087da | 212 | skb_set_end_offset(skb, size); |
ba0509b6 JDB |
213 | skb->mac_header = (typeof(skb->mac_header))~0U; |
214 | skb->transport_header = (typeof(skb->transport_header))~0U; | |
68822bdf | 215 | skb->alloc_cpu = raw_smp_processor_id(); |
ba0509b6 JDB |
216 | /* make sure we initialize shinfo sequentially */ |
217 | shinfo = skb_shinfo(skb); | |
218 | memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); | |
219 | atomic_set(&shinfo->dataref, 1); | |
220 | ||
6370cc3b | 221 | skb_set_kcov_handle(skb, kcov_common_handle()); |
ba0509b6 JDB |
222 | } |
223 | ||
b2b5ce9d | 224 | /** |
2ea2f62c | 225 | * __build_skb - build a network buffer |
b2b5ce9d | 226 | * @data: data buffer provided by caller |
2ea2f62c | 227 | * @frag_size: size of data, or 0 if head was kmalloced |
b2b5ce9d ED |
228 | * |
229 | * Allocate a new &sk_buff. Caller provides space holding head and | |
deceb4c0 | 230 | * skb_shared_info. @data must have been allocated by kmalloc() only if |
2ea2f62c ED |
231 | * @frag_size is 0, otherwise data should come from the page allocator |
232 | * or vmalloc() | |
b2b5ce9d ED |
233 | * The return is the new skb buffer. |
234 | * On a failure the return is %NULL, and @data is not freed. | |
235 | * Notes : | |
236 | * Before IO, driver allocates only data buffer where NIC put incoming frame | |
237 | * Driver should add room at head (NET_SKB_PAD) and | |
238 | * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info)) | |
239 | * After IO, driver calls build_skb(), to allocate sk_buff and populate it | |
240 | * before giving packet to stack. | |
241 | * RX rings only contains data buffers, not full skbs. | |
242 | */ | |
2ea2f62c | 243 | struct sk_buff *__build_skb(void *data, unsigned int frag_size) |
b2b5ce9d | 244 | { |
b2b5ce9d | 245 | struct sk_buff *skb; |
b2b5ce9d ED |
246 | |
247 | skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC); | |
ba0509b6 | 248 | if (unlikely(!skb)) |
b2b5ce9d ED |
249 | return NULL; |
250 | ||
b2b5ce9d | 251 | memset(skb, 0, offsetof(struct sk_buff, tail)); |
483126b3 | 252 | __build_skb_around(skb, data, frag_size); |
b2b5ce9d | 253 | |
483126b3 | 254 | return skb; |
b2b5ce9d | 255 | } |
2ea2f62c ED |
256 | |
257 | /* build_skb() is wrapper over __build_skb(), that specifically | |
258 | * takes care of skb->head and skb->pfmemalloc | |
259 | * This means that if @frag_size is not zero, then @data must be backed | |
260 | * by a page fragment, not kmalloc() or vmalloc() | |
261 | */ | |
262 | struct sk_buff *build_skb(void *data, unsigned int frag_size) | |
263 | { | |
264 | struct sk_buff *skb = __build_skb(data, frag_size); | |
265 | ||
266 | if (skb && frag_size) { | |
267 | skb->head_frag = 1; | |
2f064f34 | 268 | if (page_is_pfmemalloc(virt_to_head_page(data))) |
2ea2f62c ED |
269 | skb->pfmemalloc = 1; |
270 | } | |
271 | return skb; | |
272 | } | |
b2b5ce9d ED |
273 | EXPORT_SYMBOL(build_skb); |
274 | ||
ba0509b6 JDB |
275 | /** |
276 | * build_skb_around - build a network buffer around provided skb | |
277 | * @skb: sk_buff provide by caller, must be memset cleared | |
278 | * @data: data buffer provided by caller | |
279 | * @frag_size: size of data, or 0 if head was kmalloced | |
280 | */ | |
281 | struct sk_buff *build_skb_around(struct sk_buff *skb, | |
282 | void *data, unsigned int frag_size) | |
283 | { | |
284 | if (unlikely(!skb)) | |
285 | return NULL; | |
286 | ||
483126b3 | 287 | __build_skb_around(skb, data, frag_size); |
ba0509b6 | 288 | |
483126b3 | 289 | if (frag_size) { |
ba0509b6 JDB |
290 | skb->head_frag = 1; |
291 | if (page_is_pfmemalloc(virt_to_head_page(data))) | |
292 | skb->pfmemalloc = 1; | |
293 | } | |
294 | return skb; | |
295 | } | |
296 | EXPORT_SYMBOL(build_skb_around); | |
297 | ||
f450d539 AL |
298 | /** |
299 | * __napi_build_skb - build a network buffer | |
300 | * @data: data buffer provided by caller | |
301 | * @frag_size: size of data, or 0 if head was kmalloced | |
302 | * | |
303 | * Version of __build_skb() that uses NAPI percpu caches to obtain | |
304 | * skbuff_head instead of inplace allocation. | |
305 | * | |
306 | * Returns a new &sk_buff on success, %NULL on allocation failure. | |
307 | */ | |
308 | static struct sk_buff *__napi_build_skb(void *data, unsigned int frag_size) | |
309 | { | |
310 | struct sk_buff *skb; | |
311 | ||
312 | skb = napi_skb_cache_get(); | |
313 | if (unlikely(!skb)) | |
314 | return NULL; | |
315 | ||
316 | memset(skb, 0, offsetof(struct sk_buff, tail)); | |
317 | __build_skb_around(skb, data, frag_size); | |
318 | ||
319 | return skb; | |
320 | } | |
321 | ||
322 | /** | |
323 | * napi_build_skb - build a network buffer | |
324 | * @data: data buffer provided by caller | |
325 | * @frag_size: size of data, or 0 if head was kmalloced | |
326 | * | |
327 | * Version of __napi_build_skb() that takes care of skb->head_frag | |
328 | * and skb->pfmemalloc when the data is a page or page fragment. | |
329 | * | |
330 | * Returns a new &sk_buff on success, %NULL on allocation failure. | |
331 | */ | |
332 | struct sk_buff *napi_build_skb(void *data, unsigned int frag_size) | |
333 | { | |
334 | struct sk_buff *skb = __napi_build_skb(data, frag_size); | |
335 | ||
336 | if (likely(skb) && frag_size) { | |
337 | skb->head_frag = 1; | |
338 | skb_propagate_pfmemalloc(virt_to_head_page(data), skb); | |
339 | } | |
340 | ||
341 | return skb; | |
342 | } | |
343 | EXPORT_SYMBOL(napi_build_skb); | |
344 | ||
5381b23d AL |
345 | /* |
346 | * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells | |
347 | * the caller if emergency pfmemalloc reserves are being used. If it is and | |
348 | * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves | |
349 | * may be used. Otherwise, the packet data may be discarded until enough | |
350 | * memory is free | |
351 | */ | |
ef28095f AL |
352 | static void *kmalloc_reserve(size_t size, gfp_t flags, int node, |
353 | bool *pfmemalloc) | |
5381b23d AL |
354 | { |
355 | void *obj; | |
356 | bool ret_pfmemalloc = false; | |
357 | ||
358 | /* | |
359 | * Try a regular allocation, when that fails and we're not entitled | |
360 | * to the reserves, fail. | |
361 | */ | |
362 | obj = kmalloc_node_track_caller(size, | |
363 | flags | __GFP_NOMEMALLOC | __GFP_NOWARN, | |
364 | node); | |
365 | if (obj || !(gfp_pfmemalloc_allowed(flags))) | |
366 | goto out; | |
367 | ||
368 | /* Try again but now we are using pfmemalloc reserves */ | |
369 | ret_pfmemalloc = true; | |
370 | obj = kmalloc_node_track_caller(size, flags, node); | |
371 | ||
372 | out: | |
373 | if (pfmemalloc) | |
374 | *pfmemalloc = ret_pfmemalloc; | |
375 | ||
376 | return obj; | |
377 | } | |
378 | ||
379 | /* Allocate a new skbuff. We do this ourselves so we can fill in a few | |
380 | * 'private' fields and also do memory statistics to find all the | |
381 | * [BEEP] leaks. | |
382 | * | |
383 | */ | |
384 | ||
385 | /** | |
386 | * __alloc_skb - allocate a network buffer | |
387 | * @size: size to allocate | |
388 | * @gfp_mask: allocation mask | |
389 | * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache | |
390 | * instead of head cache and allocate a cloned (child) skb. | |
391 | * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for | |
392 | * allocations in case the data is required for writeback | |
393 | * @node: numa node to allocate memory on | |
394 | * | |
395 | * Allocate a new &sk_buff. The returned buffer has no headroom and a | |
396 | * tail room of at least size bytes. The object has a reference count | |
397 | * of one. The return is the buffer. On a failure the return is %NULL. | |
398 | * | |
399 | * Buffers may only be allocated from interrupts using a @gfp_mask of | |
400 | * %GFP_ATOMIC. | |
401 | */ | |
402 | struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, | |
403 | int flags, int node) | |
404 | { | |
405 | struct kmem_cache *cache; | |
5381b23d | 406 | struct sk_buff *skb; |
a5df6333 | 407 | unsigned int osize; |
5381b23d | 408 | bool pfmemalloc; |
a5df6333 | 409 | u8 *data; |
5381b23d AL |
410 | |
411 | cache = (flags & SKB_ALLOC_FCLONE) | |
412 | ? skbuff_fclone_cache : skbuff_head_cache; | |
413 | ||
414 | if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX)) | |
415 | gfp_mask |= __GFP_MEMALLOC; | |
416 | ||
417 | /* Get the HEAD */ | |
d13612b5 AL |
418 | if ((flags & (SKB_ALLOC_FCLONE | SKB_ALLOC_NAPI)) == SKB_ALLOC_NAPI && |
419 | likely(node == NUMA_NO_NODE || node == numa_mem_id())) | |
420 | skb = napi_skb_cache_get(); | |
421 | else | |
422 | skb = kmem_cache_alloc_node(cache, gfp_mask & ~GFP_DMA, node); | |
df1ae022 AL |
423 | if (unlikely(!skb)) |
424 | return NULL; | |
5381b23d AL |
425 | prefetchw(skb); |
426 | ||
427 | /* We do our best to align skb_shared_info on a separate cache | |
428 | * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives | |
429 | * aligned memory blocks, unless SLUB/SLAB debug is enabled. | |
430 | * Both skb->head and skb_shared_info are cache line aligned. | |
431 | */ | |
432 | size = SKB_DATA_ALIGN(size); | |
433 | size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
434 | data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc); | |
df1ae022 | 435 | if (unlikely(!data)) |
5381b23d AL |
436 | goto nodata; |
437 | /* kmalloc(size) might give us more room than requested. | |
438 | * Put skb_shared_info exactly at the end of allocated zone, | |
439 | * to allow max possible filling before reallocation. | |
440 | */ | |
a5df6333 LR |
441 | osize = ksize(data); |
442 | size = SKB_WITH_OVERHEAD(osize); | |
5381b23d AL |
443 | prefetchw(data + size); |
444 | ||
445 | /* | |
446 | * Only clear those fields we need to clear, not those that we will | |
447 | * actually initialise below. Hence, don't put any more fields after | |
448 | * the tail pointer in struct sk_buff! | |
449 | */ | |
450 | memset(skb, 0, offsetof(struct sk_buff, tail)); | |
a5df6333 | 451 | __build_skb_around(skb, data, osize); |
5381b23d | 452 | skb->pfmemalloc = pfmemalloc; |
5381b23d AL |
453 | |
454 | if (flags & SKB_ALLOC_FCLONE) { | |
455 | struct sk_buff_fclones *fclones; | |
456 | ||
457 | fclones = container_of(skb, struct sk_buff_fclones, skb1); | |
458 | ||
459 | skb->fclone = SKB_FCLONE_ORIG; | |
460 | refcount_set(&fclones->fclone_ref, 1); | |
5381b23d AL |
461 | } |
462 | ||
5381b23d | 463 | return skb; |
df1ae022 | 464 | |
5381b23d AL |
465 | nodata: |
466 | kmem_cache_free(cache, skb); | |
df1ae022 | 467 | return NULL; |
5381b23d AL |
468 | } |
469 | EXPORT_SYMBOL(__alloc_skb); | |
470 | ||
fd11a83d AD |
471 | /** |
472 | * __netdev_alloc_skb - allocate an skbuff for rx on a specific device | |
473 | * @dev: network device to receive on | |
d7499160 | 474 | * @len: length to allocate |
fd11a83d AD |
475 | * @gfp_mask: get_free_pages mask, passed to alloc_skb |
476 | * | |
477 | * Allocate a new &sk_buff and assign it a usage count of one. The | |
478 | * buffer has NET_SKB_PAD headroom built in. Users should allocate | |
479 | * the headroom they think they need without accounting for the | |
480 | * built in space. The built in space is used for optimisations. | |
481 | * | |
482 | * %NULL is returned if there is no free memory. | |
483 | */ | |
9451980a AD |
484 | struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len, |
485 | gfp_t gfp_mask) | |
fd11a83d | 486 | { |
b63ae8ca | 487 | struct page_frag_cache *nc; |
fd11a83d | 488 | struct sk_buff *skb; |
9451980a AD |
489 | bool pfmemalloc; |
490 | void *data; | |
491 | ||
492 | len += NET_SKB_PAD; | |
fd11a83d | 493 | |
66c55602 AL |
494 | /* If requested length is either too small or too big, |
495 | * we use kmalloc() for skb->head allocation. | |
496 | */ | |
497 | if (len <= SKB_WITH_OVERHEAD(1024) || | |
498 | len > SKB_WITH_OVERHEAD(PAGE_SIZE) || | |
d0164adc | 499 | (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { |
a080e7bd AD |
500 | skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); |
501 | if (!skb) | |
502 | goto skb_fail; | |
503 | goto skb_success; | |
504 | } | |
fd11a83d | 505 | |
9451980a AD |
506 | len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); |
507 | len = SKB_DATA_ALIGN(len); | |
508 | ||
509 | if (sk_memalloc_socks()) | |
510 | gfp_mask |= __GFP_MEMALLOC; | |
511 | ||
afa79d08 | 512 | if (in_hardirq() || irqs_disabled()) { |
92dcabd7 SAS |
513 | nc = this_cpu_ptr(&netdev_alloc_cache); |
514 | data = page_frag_alloc(nc, len, gfp_mask); | |
515 | pfmemalloc = nc->pfmemalloc; | |
516 | } else { | |
517 | local_bh_disable(); | |
518 | nc = this_cpu_ptr(&napi_alloc_cache.page); | |
519 | data = page_frag_alloc(nc, len, gfp_mask); | |
520 | pfmemalloc = nc->pfmemalloc; | |
521 | local_bh_enable(); | |
522 | } | |
9451980a AD |
523 | |
524 | if (unlikely(!data)) | |
525 | return NULL; | |
526 | ||
527 | skb = __build_skb(data, len); | |
528 | if (unlikely(!skb)) { | |
181edb2b | 529 | skb_free_frag(data); |
9451980a | 530 | return NULL; |
7b2e497a | 531 | } |
fd11a83d | 532 | |
9451980a AD |
533 | if (pfmemalloc) |
534 | skb->pfmemalloc = 1; | |
535 | skb->head_frag = 1; | |
536 | ||
a080e7bd | 537 | skb_success: |
9451980a AD |
538 | skb_reserve(skb, NET_SKB_PAD); |
539 | skb->dev = dev; | |
540 | ||
a080e7bd | 541 | skb_fail: |
8af27456 CH |
542 | return skb; |
543 | } | |
b4ac530f | 544 | EXPORT_SYMBOL(__netdev_alloc_skb); |
1da177e4 | 545 | |
fd11a83d AD |
546 | /** |
547 | * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance | |
548 | * @napi: napi instance this buffer was allocated for | |
d7499160 | 549 | * @len: length to allocate |
fd11a83d AD |
550 | * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages |
551 | * | |
552 | * Allocate a new sk_buff for use in NAPI receive. This buffer will | |
553 | * attempt to allocate the head from a special reserved region used | |
554 | * only for NAPI Rx allocation. By doing this we can save several | |
555 | * CPU cycles by avoiding having to disable and re-enable IRQs. | |
556 | * | |
557 | * %NULL is returned if there is no free memory. | |
558 | */ | |
9451980a AD |
559 | struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len, |
560 | gfp_t gfp_mask) | |
fd11a83d | 561 | { |
3226b158 | 562 | struct napi_alloc_cache *nc; |
fd11a83d | 563 | struct sk_buff *skb; |
9451980a AD |
564 | void *data; |
565 | ||
ee2640df | 566 | DEBUG_NET_WARN_ON_ONCE(!in_softirq()); |
9451980a | 567 | len += NET_SKB_PAD + NET_IP_ALIGN; |
fd11a83d | 568 | |
3226b158 ED |
569 | /* If requested length is either too small or too big, |
570 | * we use kmalloc() for skb->head allocation. | |
571 | */ | |
572 | if (len <= SKB_WITH_OVERHEAD(1024) || | |
573 | len > SKB_WITH_OVERHEAD(PAGE_SIZE) || | |
d0164adc | 574 | (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { |
cfb8ec65 AL |
575 | skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX | SKB_ALLOC_NAPI, |
576 | NUMA_NO_NODE); | |
a080e7bd AD |
577 | if (!skb) |
578 | goto skb_fail; | |
579 | goto skb_success; | |
580 | } | |
9451980a | 581 | |
3226b158 | 582 | nc = this_cpu_ptr(&napi_alloc_cache); |
9451980a AD |
583 | len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); |
584 | len = SKB_DATA_ALIGN(len); | |
585 | ||
586 | if (sk_memalloc_socks()) | |
587 | gfp_mask |= __GFP_MEMALLOC; | |
fd11a83d | 588 | |
8c2dd3e4 | 589 | data = page_frag_alloc(&nc->page, len, gfp_mask); |
9451980a AD |
590 | if (unlikely(!data)) |
591 | return NULL; | |
592 | ||
cfb8ec65 | 593 | skb = __napi_build_skb(data, len); |
9451980a | 594 | if (unlikely(!skb)) { |
181edb2b | 595 | skb_free_frag(data); |
9451980a | 596 | return NULL; |
fd11a83d AD |
597 | } |
598 | ||
795bb1c0 | 599 | if (nc->page.pfmemalloc) |
9451980a AD |
600 | skb->pfmemalloc = 1; |
601 | skb->head_frag = 1; | |
602 | ||
a080e7bd | 603 | skb_success: |
9451980a AD |
604 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); |
605 | skb->dev = napi->dev; | |
606 | ||
a080e7bd | 607 | skb_fail: |
fd11a83d AD |
608 | return skb; |
609 | } | |
610 | EXPORT_SYMBOL(__napi_alloc_skb); | |
611 | ||
654bed16 | 612 | void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, |
50269e19 | 613 | int size, unsigned int truesize) |
654bed16 PZ |
614 | { |
615 | skb_fill_page_desc(skb, i, page, off, size); | |
616 | skb->len += size; | |
617 | skb->data_len += size; | |
50269e19 | 618 | skb->truesize += truesize; |
654bed16 PZ |
619 | } |
620 | EXPORT_SYMBOL(skb_add_rx_frag); | |
621 | ||
f8e617e1 JW |
622 | void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, |
623 | unsigned int truesize) | |
624 | { | |
625 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
626 | ||
627 | skb_frag_size_add(frag, size); | |
628 | skb->len += size; | |
629 | skb->data_len += size; | |
630 | skb->truesize += truesize; | |
631 | } | |
632 | EXPORT_SYMBOL(skb_coalesce_rx_frag); | |
633 | ||
27b437c8 | 634 | static void skb_drop_list(struct sk_buff **listp) |
1da177e4 | 635 | { |
bd8a7036 | 636 | kfree_skb_list(*listp); |
27b437c8 | 637 | *listp = NULL; |
1da177e4 LT |
638 | } |
639 | ||
27b437c8 HX |
640 | static inline void skb_drop_fraglist(struct sk_buff *skb) |
641 | { | |
642 | skb_drop_list(&skb_shinfo(skb)->frag_list); | |
643 | } | |
644 | ||
1da177e4 LT |
645 | static void skb_clone_fraglist(struct sk_buff *skb) |
646 | { | |
647 | struct sk_buff *list; | |
648 | ||
fbb398a8 | 649 | skb_walk_frags(skb, list) |
1da177e4 LT |
650 | skb_get(list); |
651 | } | |
652 | ||
d3836f21 ED |
653 | static void skb_free_head(struct sk_buff *skb) |
654 | { | |
181edb2b AD |
655 | unsigned char *head = skb->head; |
656 | ||
6a5bcd84 IA |
657 | if (skb->head_frag) { |
658 | if (skb_pp_recycle(skb, head)) | |
659 | return; | |
181edb2b | 660 | skb_free_frag(head); |
6a5bcd84 | 661 | } else { |
181edb2b | 662 | kfree(head); |
6a5bcd84 | 663 | } |
d3836f21 ED |
664 | } |
665 | ||
5bba1712 | 666 | static void skb_release_data(struct sk_buff *skb) |
1da177e4 | 667 | { |
ff04a771 ED |
668 | struct skb_shared_info *shinfo = skb_shinfo(skb); |
669 | int i; | |
1da177e4 | 670 | |
ff04a771 ED |
671 | if (skb->cloned && |
672 | atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, | |
673 | &shinfo->dataref)) | |
2cc3aeb5 | 674 | goto exit; |
a6686f2f | 675 | |
753f1ca4 PB |
676 | if (skb_zcopy(skb)) { |
677 | bool skip_unref = shinfo->flags & SKBFL_MANAGED_FRAG_REFS; | |
678 | ||
679 | skb_zcopy_clear(skb, true); | |
680 | if (skip_unref) | |
681 | goto free_head; | |
682 | } | |
70c43167 | 683 | |
ff04a771 | 684 | for (i = 0; i < shinfo->nr_frags; i++) |
6a5bcd84 | 685 | __skb_frag_unref(&shinfo->frags[i], skb->pp_recycle); |
a6686f2f | 686 | |
753f1ca4 | 687 | free_head: |
ff04a771 ED |
688 | if (shinfo->frag_list) |
689 | kfree_skb_list(shinfo->frag_list); | |
690 | ||
691 | skb_free_head(skb); | |
2cc3aeb5 IA |
692 | exit: |
693 | /* When we clone an SKB we copy the reycling bit. The pp_recycle | |
694 | * bit is only set on the head though, so in order to avoid races | |
695 | * while trying to recycle fragments on __skb_frag_unref() we need | |
696 | * to make one SKB responsible for triggering the recycle path. | |
697 | * So disable the recycling bit if an SKB is cloned and we have | |
58e61e41 | 698 | * additional references to the fragmented part of the SKB. |
2cc3aeb5 IA |
699 | * Eventually the last SKB will have the recycling bit set and it's |
700 | * dataref set to 0, which will trigger the recycling | |
701 | */ | |
702 | skb->pp_recycle = 0; | |
1da177e4 LT |
703 | } |
704 | ||
705 | /* | |
706 | * Free an skbuff by memory without cleaning the state. | |
707 | */ | |
2d4baff8 | 708 | static void kfree_skbmem(struct sk_buff *skb) |
1da177e4 | 709 | { |
d0bf4a9e | 710 | struct sk_buff_fclones *fclones; |
d179cd12 | 711 | |
d179cd12 DM |
712 | switch (skb->fclone) { |
713 | case SKB_FCLONE_UNAVAILABLE: | |
714 | kmem_cache_free(skbuff_head_cache, skb); | |
6ffe75eb | 715 | return; |
d179cd12 DM |
716 | |
717 | case SKB_FCLONE_ORIG: | |
d0bf4a9e | 718 | fclones = container_of(skb, struct sk_buff_fclones, skb1); |
d179cd12 | 719 | |
6ffe75eb ED |
720 | /* We usually free the clone (TX completion) before original skb |
721 | * This test would have no chance to be true for the clone, | |
722 | * while here, branch prediction will be good. | |
d179cd12 | 723 | */ |
2638595a | 724 | if (refcount_read(&fclones->fclone_ref) == 1) |
6ffe75eb ED |
725 | goto fastpath; |
726 | break; | |
e7820e39 | 727 | |
6ffe75eb ED |
728 | default: /* SKB_FCLONE_CLONE */ |
729 | fclones = container_of(skb, struct sk_buff_fclones, skb2); | |
d179cd12 | 730 | break; |
3ff50b79 | 731 | } |
2638595a | 732 | if (!refcount_dec_and_test(&fclones->fclone_ref)) |
6ffe75eb ED |
733 | return; |
734 | fastpath: | |
735 | kmem_cache_free(skbuff_fclone_cache, fclones); | |
1da177e4 LT |
736 | } |
737 | ||
0a463c78 | 738 | void skb_release_head_state(struct sk_buff *skb) |
1da177e4 | 739 | { |
adf30907 | 740 | skb_dst_drop(skb); |
9c2b3328 | 741 | if (skb->destructor) { |
7890e2f0 | 742 | DEBUG_NET_WARN_ON_ONCE(in_hardirq()); |
1da177e4 LT |
743 | skb->destructor(skb); |
744 | } | |
a3bf7ae9 | 745 | #if IS_ENABLED(CONFIG_NF_CONNTRACK) |
cb9c6836 | 746 | nf_conntrack_put(skb_nfct(skb)); |
1da177e4 | 747 | #endif |
df5042f4 | 748 | skb_ext_put(skb); |
04a4bb55 LB |
749 | } |
750 | ||
751 | /* Free everything but the sk_buff shell. */ | |
752 | static void skb_release_all(struct sk_buff *skb) | |
753 | { | |
754 | skb_release_head_state(skb); | |
a28b1b90 FW |
755 | if (likely(skb->head)) |
756 | skb_release_data(skb); | |
2d4baff8 HX |
757 | } |
758 | ||
759 | /** | |
760 | * __kfree_skb - private function | |
761 | * @skb: buffer | |
762 | * | |
763 | * Free an sk_buff. Release anything attached to the buffer. | |
764 | * Clean the state. This is an internal helper function. Users should | |
765 | * always call kfree_skb | |
766 | */ | |
1da177e4 | 767 | |
2d4baff8 HX |
768 | void __kfree_skb(struct sk_buff *skb) |
769 | { | |
770 | skb_release_all(skb); | |
1da177e4 LT |
771 | kfree_skbmem(skb); |
772 | } | |
b4ac530f | 773 | EXPORT_SYMBOL(__kfree_skb); |
1da177e4 | 774 | |