1 /*
2 * Routines having to do with the 'struct sk_buff' memory handlers.
3 *
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
6 *
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
39 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
40
41 #include <linux/module.h>
42 #include <linux/types.h>
43 #include <linux/kernel.h>
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>
49 #include <linux/tcp.h>
50 #include <linux/udp.h>
51 #include <linux/sctp.h>
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>
58 #include <linux/splice.h>
59 #include <linux/cache.h>
60 #include <linux/rtnetlink.h>
61 #include <linux/init.h>
62 #include <linux/scatterlist.h>
63 #include <linux/errqueue.h>
64 #include <linux/prefetch.h>
65 #include <linux/if_vlan.h>
66
67 #include <net/protocol.h>
68 #include <net/dst.h>
69 #include <net/sock.h>
70 #include <net/checksum.h>
71 #include <net/ip6_checksum.h>
72 #include <net/xfrm.h>
73
74 #include <linux/uaccess.h>
75 #include <trace/events/skb.h>
76 #include <linux/highmem.h>
77 #include <linux/capability.h>
78 #include <linux/user_namespace.h>
79
80 struct kmem_cache *skbuff_head_cache __ro_after_init;
81 static struct kmem_cache *skbuff_fclone_cache __ro_after_init;
82 int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
83 EXPORT_SYMBOL(sysctl_max_skb_frags);
84
85 /**
86 * skb_panic - private function for out-of-line support
87 * @skb: buffer
88 * @sz: size
89 * @addr: address
90 * @msg: skb_over_panic or skb_under_panic
91 *
92 * Out-of-line support for skb_put() and skb_push().
93 * Called via the wrapper skb_over_panic() or skb_under_panic().
94 * Keep out of line to prevent kernel bloat.
95 * __builtin_return_address is not used because it is not always reliable.
96 */
skb_panic(struct sk_buff * skb,unsigned int sz,void * addr,const char msg[])97 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
98 const char msg[])
99 {
100 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
101 msg, addr, skb->len, sz, skb->head, skb->data,
102 (unsigned long)skb->tail, (unsigned long)skb->end,
103 skb->dev ? skb->dev->name : "<NULL>");
104 BUG();
105 }
106
skb_over_panic(struct sk_buff * skb,unsigned int sz,void * addr)107 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
108 {
109 skb_panic(skb, sz, addr, __func__);
110 }
111
skb_under_panic(struct sk_buff * skb,unsigned int sz,void * addr)112 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
113 {
114 skb_panic(skb, sz, addr, __func__);
115 }
116
117 /*
118 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
119 * the caller if emergency pfmemalloc reserves are being used. If it is and
120 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
121 * may be used. Otherwise, the packet data may be discarded until enough
122 * memory is free
123 */
124 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
125 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
126
__kmalloc_reserve(size_t size,gfp_t flags,int node,unsigned long ip,bool * pfmemalloc)127 static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
128 unsigned long ip, bool *pfmemalloc)
129 {
130 void *obj;
131 bool ret_pfmemalloc = false;
132
133 /*
134 * Try a regular allocation, when that fails and we're not entitled
135 * to the reserves, fail.
136 */
137 obj = kmalloc_node_track_caller(size,
138 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
139 node);
140 if (obj || !(gfp_pfmemalloc_allowed(flags)))
141 goto out;
142
143 /* Try again but now we are using pfmemalloc reserves */
144 ret_pfmemalloc = true;
145 obj = kmalloc_node_track_caller(size, flags, node);
146
147 out:
148 if (pfmemalloc)
149 *pfmemalloc = ret_pfmemalloc;
150
151 return obj;
152 }
153
154 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
155 * 'private' fields and also do memory statistics to find all the
156 * [BEEP] leaks.
157 *
158 */
159
160 /**
161 * __alloc_skb - allocate a network buffer
162 * @size: size to allocate
163 * @gfp_mask: allocation mask
164 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
165 * instead of head cache and allocate a cloned (child) skb.
166 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
167 * allocations in case the data is required for writeback
168 * @node: numa node to allocate memory on
169 *
170 * Allocate a new &sk_buff. The returned buffer has no headroom and a
171 * tail room of at least size bytes. The object has a reference count
172 * of one. The return is the buffer. On a failure the return is %NULL.
173 *
174 * Buffers may only be allocated from interrupts using a @gfp_mask of
175 * %GFP_ATOMIC.
176 */
__alloc_skb(unsigned int size,gfp_t gfp_mask,int flags,int node)177 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
178 int flags, int node)
179 {
180 struct kmem_cache *cache;
181 struct skb_shared_info *shinfo;
182 struct sk_buff *skb;
183 u8 *data;
184 bool pfmemalloc;
185
186 cache = (flags & SKB_ALLOC_FCLONE)
187 ? skbuff_fclone_cache : skbuff_head_cache;
188
189 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
190 gfp_mask |= __GFP_MEMALLOC;
191
192 /* Get the HEAD */
193 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
194 if (!skb)
195 goto out;
196 prefetchw(skb);
197
198 /* We do our best to align skb_shared_info on a separate cache
199 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
200 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
201 * Both skb->head and skb_shared_info are cache line aligned.
202 */
203 size = SKB_DATA_ALIGN(size);
204 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
205 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
206 if (!data)
207 goto nodata;
208 /* kmalloc(size) might give us more room than requested.
209 * Put skb_shared_info exactly at the end of allocated zone,
210 * to allow max possible filling before reallocation.
211 */
212 size = SKB_WITH_OVERHEAD(ksize(data));
213 prefetchw(data + size);
214
215 /*
216 * Only clear those fields we need to clear, not those that we will
217 * actually initialise below. Hence, don't put any more fields after
218 * the tail pointer in struct sk_buff!
219 */
220 memset(skb, 0, offsetof(struct sk_buff, tail));
221 /* Account for allocated memory : skb + skb->head */
222 skb->truesize = SKB_TRUESIZE(size);
223 skb->pfmemalloc = pfmemalloc;
224 refcount_set(&skb->users, 1);
225 skb->head = data;
226 skb->data = data;
227 skb_reset_tail_pointer(skb);
228 skb->end = skb->tail + size;
229 skb->mac_header = (typeof(skb->mac_header))~0U;
230 skb->transport_header = (typeof(skb->transport_header))~0U;
231
232 /* make sure we initialize shinfo sequentially */
233 shinfo = skb_shinfo(skb);
234 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
235 atomic_set(&shinfo->dataref, 1);
236
237 if (flags & SKB_ALLOC_FCLONE) {
238 struct sk_buff_fclones *fclones;
239
240 fclones = container_of(skb, struct sk_buff_fclones, skb1);
241
242 skb->fclone = SKB_FCLONE_ORIG;
243 refcount_set(&fclones->fclone_ref, 1);
244
245 fclones->skb2.fclone = SKB_FCLONE_CLONE;
246 }
247 out:
248 return skb;
249 nodata:
250 kmem_cache_free(cache, skb);
251 skb = NULL;
252 goto out;
253 }
254 EXPORT_SYMBOL(__alloc_skb);
255
256 /**
257 * __build_skb - build a network buffer
258 * @data: data buffer provided by caller
259 * @frag_size: size of data, or 0 if head was kmalloced
260 *
261 * Allocate a new &sk_buff. Caller provides space holding head and
262 * skb_shared_info. @data must have been allocated by kmalloc() only if
263 * @frag_size is 0, otherwise data should come from the page allocator
264 * or vmalloc()
265 * The return is the new skb buffer.
266 * On a failure the return is %NULL, and @data is not freed.
267 * Notes :
268 * Before IO, driver allocates only data buffer where NIC put incoming frame
269 * Driver should add room at head (NET_SKB_PAD) and
270 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
271 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
272 * before giving packet to stack.
273 * RX rings only contains data buffers, not full skbs.
274 */
__build_skb(void * data,unsigned int frag_size)275 struct sk_buff *__build_skb(void *data, unsigned int frag_size)
276 {
277 struct skb_shared_info *shinfo;
278 struct sk_buff *skb;
279 unsigned int size = frag_size ? : ksize(data);
280
281 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
282 if (!skb)
283 return NULL;
284
285 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
286
287 memset(skb, 0, offsetof(struct sk_buff, tail));
288 skb->truesize = SKB_TRUESIZE(size);
289 refcount_set(&skb->users, 1);
290 skb->head = data;
291 skb->data = data;
292 skb_reset_tail_pointer(skb);
293 skb->end = skb->tail + size;
294 skb->mac_header = (typeof(skb->mac_header))~0U;
295 skb->transport_header = (typeof(skb->transport_header))~0U;
296
297 /* make sure we initialize shinfo sequentially */
298 shinfo = skb_shinfo(skb);
299 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
300 atomic_set(&shinfo->dataref, 1);
301
302 return skb;
303 }
304
305 /* build_skb() is wrapper over __build_skb(), that specifically
306 * takes care of skb->head and skb->pfmemalloc
307 * This means that if @frag_size is not zero, then @data must be backed
308 * by a page fragment, not kmalloc() or vmalloc()
309 */
build_skb(void * data,unsigned int frag_size)310 struct sk_buff *build_skb(void *data, unsigned int frag_size)
311 {
312 struct sk_buff *skb = __build_skb(data, frag_size);
313
314 if (skb && frag_size) {
315 skb->head_frag = 1;
316 if (page_is_pfmemalloc(virt_to_head_page(data)))
317 skb->pfmemalloc = 1;
318 }
319 return skb;
320 }
321 EXPORT_SYMBOL(build_skb);
322
323 #define NAPI_SKB_CACHE_SIZE 64
324
325 struct napi_alloc_cache {
326 struct page_frag_cache page;
327 unsigned int skb_count;
328 void *skb_cache[NAPI_SKB_CACHE_SIZE];
329 };
330
331 static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
332 static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
333
__netdev_alloc_frag(unsigned int fragsz,gfp_t gfp_mask)334 static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
335 {
336 struct page_frag_cache *nc;
337 unsigned long flags;
338 void *data;
339
340 local_irq_save(flags);
341 nc = this_cpu_ptr(&netdev_alloc_cache);
342 data = page_frag_alloc(nc, fragsz, gfp_mask);
343 local_irq_restore(flags);
344 return data;
345 }
346
347 /**
348 * netdev_alloc_frag - allocate a page fragment
349 * @fragsz: fragment size
350 *
351 * Allocates a frag from a page for receive buffer.
352 * Uses GFP_ATOMIC allocations.
353 */
netdev_alloc_frag(unsigned int fragsz)354 void *netdev_alloc_frag(unsigned int fragsz)
355 {
356 return __netdev_alloc_frag(fragsz, GFP_ATOMIC);
357 }
358 EXPORT_SYMBOL(netdev_alloc_frag);
359
__napi_alloc_frag(unsigned int fragsz,gfp_t gfp_mask)360 static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
361 {
362 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
363
364 return page_frag_alloc(&nc->page, fragsz, gfp_mask);
365 }
366
napi_alloc_frag(unsigned int fragsz)367 void *napi_alloc_frag(unsigned int fragsz)
368 {
369 return __napi_alloc_frag(fragsz, GFP_ATOMIC);
370 }
371 EXPORT_SYMBOL(napi_alloc_frag);
372
373 /**
374 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
375 * @dev: network device to receive on
376 * @len: length to allocate
377 * @gfp_mask: get_free_pages mask, passed to alloc_skb
378 *
379 * Allocate a new &sk_buff and assign it a usage count of one. The
380 * buffer has NET_SKB_PAD headroom built in. Users should allocate
381 * the headroom they think they need without accounting for the
382 * built in space. The built in space is used for optimisations.
383 *
384 * %NULL is returned if there is no free memory.
385 */
__netdev_alloc_skb(struct net_device * dev,unsigned int len,gfp_t gfp_mask)386 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
387 gfp_t gfp_mask)
388 {
389 struct page_frag_cache *nc;
390 unsigned long flags;
391 struct sk_buff *skb;
392 bool pfmemalloc;
393 void *data;
394
395 len += NET_SKB_PAD;
396
397 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
398 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
399 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
400 if (!skb)
401 goto skb_fail;
402 goto skb_success;
403 }
404
405 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
406 len = SKB_DATA_ALIGN(len);
407
408 if (sk_memalloc_socks())
409 gfp_mask |= __GFP_MEMALLOC;
410
411 local_irq_save(flags);
412
413 nc = this_cpu_ptr(&netdev_alloc_cache);
414 data = page_frag_alloc(nc, len, gfp_mask);
415 pfmemalloc = nc->pfmemalloc;
416
417 local_irq_restore(flags);
418
419 if (unlikely(!data))
420 return NULL;
421
422 skb = __build_skb(data, len);
423 if (unlikely(!skb)) {
424 skb_free_frag(data);
425 return NULL;
426 }
427
428 /* use OR instead of assignment to avoid clearing of bits in mask */
429 if (pfmemalloc)
430 skb->pfmemalloc = 1;
431 skb->head_frag = 1;
432
433 skb_success:
434 skb_reserve(skb, NET_SKB_PAD);
435 skb->dev = dev;
436
437 skb_fail:
438 return skb;
439 }
440 EXPORT_SYMBOL(__netdev_alloc_skb);
441
442 /**
443 * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
444 * @napi: napi instance this buffer was allocated for
445 * @len: length to allocate
446 * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
447 *
448 * Allocate a new sk_buff for use in NAPI receive. This buffer will
449 * attempt to allocate the head from a special reserved region used
450 * only for NAPI Rx allocation. By doing this we can save several
451 * CPU cycles by avoiding having to disable and re-enable IRQs.
452 *
453 * %NULL is returned if there is no free memory.
454 */
__napi_alloc_skb(struct napi_struct * napi,unsigned int len,gfp_t gfp_mask)455 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
456 gfp_t gfp_mask)
457 {
458 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
459 struct sk_buff *skb;
460 void *data;
461
462 len += NET_SKB_PAD + NET_IP_ALIGN;
463
464 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
465 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
466 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
467 if (!skb)
468 goto skb_fail;
469 goto skb_success;
470 }
471
472 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
473 len = SKB_DATA_ALIGN(len);
474
475 if (sk_memalloc_socks())
476 gfp_mask |= __GFP_MEMALLOC;
477
478 data = page_frag_alloc(&nc->page, len, gfp_mask);
479 if (unlikely(!data))
480 return NULL;
481
482 skb = __build_skb(data, len);
483 if (unlikely(!skb)) {
484 skb_free_frag(data);
485 return NULL;
486 }
487
488 /* use OR instead of assignment to avoid clearing of bits in mask */
489 if (nc->page.pfmemalloc)
490 skb->pfmemalloc = 1;
491 skb->head_frag = 1;
492
493 skb_success:
494 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
495 skb->dev = napi->dev;
496
497 skb_fail:
498 return skb;
499 }
500 EXPORT_SYMBOL(__napi_alloc_skb);
501
skb_add_rx_frag(struct sk_buff * skb,int i,struct page * page,int off,int size,unsigned int truesize)502 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
503 int size, unsigned int truesize)
504 {
505 skb_fill_page_desc(skb, i, page, off, size);
506 skb->len += size;
507 skb->data_len += size;
508 skb->truesize += truesize;
509 }
510 EXPORT_SYMBOL(skb_add_rx_frag);
511
skb_coalesce_rx_frag(struct sk_buff * skb,int i,int size,unsigned int truesize)512 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
513 unsigned int truesize)
514 {
515 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
516
517 skb_frag_size_add(frag, size);
518 skb->len += size;
519 skb->data_len += size;
520 skb->truesize += truesize;
521 }
522 EXPORT_SYMBOL(skb_coalesce_rx_frag);
523
skb_drop_list(struct sk_buff ** listp)524 static void skb_drop_list(struct sk_buff **listp)
525 {
526 kfree_skb_list(*listp);
527 *listp = NULL;
528 }
529
skb_drop_fraglist(struct sk_buff * skb)530 static inline void skb_drop_fraglist(struct sk_buff *skb)
531 {
532 skb_drop_list(&skb_shinfo(skb)->frag_list);
533 }
534
skb_clone_fraglist(struct sk_buff * skb)535 static void skb_clone_fraglist(struct sk_buff *skb)
536 {
537 struct sk_buff *list;
538
539 skb_walk_frags(skb, list)
540 skb_get(list);
541 }
542
skb_free_head(struct sk_buff * skb)543 static void skb_free_head(struct sk_buff *skb)
544 {
545 unsigned char *head = skb->head;
546
547 if (skb->head_frag)
548 skb_free_frag(head);
549 else
550 kfree(head);
551 }
552
skb_release_data(struct sk_buff * skb)553 static void skb_release_data(struct sk_buff *skb)
554 {
555 struct skb_shared_info *shinfo = skb_shinfo(skb);
556 int i;
557
558 if (skb->cloned &&
559 atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
560 &shinfo->dataref))
561 return;
562
563 for (i = 0; i < shinfo->nr_frags; i++)
564 __skb_frag_unref(&shinfo->frags[i]);
565
566 if (shinfo->frag_list)
567 kfree_skb_list(shinfo->frag_list);
568
569 skb_zcopy_clear(skb, true);
570 skb_free_head(skb);
571 }
572
573 /*
574 * Free an skbuff by memory without cleaning the state.
575 */
kfree_skbmem(struct sk_buff * skb)576 static void kfree_skbmem(struct sk_buff *skb)
577 {
578 struct sk_buff_fclones *fclones;
579
580 switch (skb->fclone) {
581 case SKB_FCLONE_UNAVAILABLE:
582 kmem_cache_free(skbuff_head_cache, skb);
583 return;
584
585 case SKB_FCLONE_ORIG:
586 fclones = container_of(skb, struct sk_buff_fclones, skb1);
587
588 /* We usually free the clone (TX completion) before original skb
589 * This test would have no chance to be true for the clone,
590 * while here, branch prediction will be good.
591 */
592 if (refcount_read(&fclones->fclone_ref) == 1)
593 goto fastpath;
594 break;
595
596 default: /* SKB_FCLONE_CLONE */
597 fclones = container_of(skb, struct sk_buff_fclones, skb2);
598 break;
599 }
600 if (!refcount_dec_and_test(&fclones->fclone_ref))
601 return;
602 fastpath:
603 kmem_cache_free(skbuff_fclone_cache, fclones);
604 }
605
skb_release_head_state(struct sk_buff * skb)606 void skb_release_head_state(struct sk_buff *skb)
607 {
608 skb_dst_drop(skb);
609 secpath_reset(skb);
610 if (skb->destructor) {
611 WARN_ON(in_irq());
612 skb->destructor(skb);
613 }
614 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
615 nf_conntrack_put(skb_nfct(skb));
616 #endif
617 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
618 nf_bridge_put(skb->nf_bridge);
619 #endif
620 }
621
622 /* Free everything but the sk_buff shell. */
skb_release_all(struct sk_buff * skb)623 static void skb_release_all(struct sk_buff *skb)
624 {
625 skb_release_head_state(skb);
626 if (likely(skb->head))
627 skb_release_data(skb);
628 }
629
630 /**
631 * __kfree_skb - private function
632 * @skb: buffer
633 *
634 * Free an sk_buff. Release anything attached to the buffer.
635 * Clean the state. This is an internal helper function. Users should
636 * always call kfree_skb
637 */
638
__kfree_skb(struct sk_buff * skb)639 void __kfree_skb(struct sk_buff *skb)
640 {
641 skb_release_all(skb);
642 kfree_skbmem(skb);
643 }
644 EXPORT_SYMBOL(__kfree_skb);
645
646 /**
647 * kfree_skb - free an sk_buff
648 * @skb: buffer to free
649 *
650 * Drop a reference to the buffer and free it if the usage count has
651 * hit zero.
652 */
kfree_skb(struct sk_buff * skb)653 void kfree_skb(struct sk_buff *skb)
654 {
655 if (!skb_unref(skb))
656 return;
657
658 trace_kfree_skb(skb, __builtin_return_address(0));
659 __kfree_skb(skb);
660 }
661 EXPORT_SYMBOL(kfree_skb);
662
kfree_skb_list(struct sk_buff * segs)663 void kfree_skb_list(struct sk_buff *segs)
664 {
665 while (segs) {
666 struct sk_buff *next = segs->next;
667
668 kfree_skb(segs);
669 segs = next;
670 }
671 }
672 EXPORT_SYMBOL(kfree_skb_list);
673
674 /**
675 * skb_tx_error - report an sk_buff xmit error
676 * @skb: buffer that triggered an error
677 *
678 * Report xmit error if a device callback is tracking this skb.
679 * skb must be freed afterwards.
680 */
skb_tx_error(struct sk_buff * skb)681 void skb_tx_error(struct sk_buff *skb)
682 {
683 skb_zcopy_clear(skb, true);
684 }
685 EXPORT_SYMBOL(skb_tx_error);
686
687 /**
688 * consume_skb - free an skbuff
689 * @skb: buffer to free
690 *
691 * Drop a ref to the buffer and free it if the usage count has hit zero
692 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
693 * is being dropped after a failure and notes that
694 */
consume_skb(struct sk_buff * skb)695 void consume_skb(struct sk_buff *skb)
696 {
697 if (!skb_unref(skb))
698 return;
699
700 trace_consume_skb(skb);
701 __kfree_skb(skb);
702 }
703 EXPORT_SYMBOL(consume_skb);
704
705 /**
706 * consume_stateless_skb - free an skbuff, assuming it is stateless
707 * @skb: buffer to free
708 *
709 * Alike consume_skb(), but this variant assumes that this is the last
710 * skb reference and all the head states have been already dropped
711 */
__consume_stateless_skb(struct sk_buff * skb)712 void __consume_stateless_skb(struct sk_buff *skb)
713 {
714 trace_consume_skb(skb);
715 skb_release_data(skb);
716 kfree_skbmem(skb);
717 }
718
__kfree_skb_flush(void)719 void __kfree_skb_flush(void)
720 {
721 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
722
723 /* flush skb_cache if containing objects */
724 if (nc->skb_count) {
725 kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count,
726 nc->skb_cache);
727 nc->skb_count = 0;
728 }
729 }
730
_kfree_skb_defer(struct sk_buff * skb)731 static inline void _kfree_skb_defer(struct sk_buff *skb)
732 {
733 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
734
735 /* drop skb->head and call any destructors for packet */
736 skb_release_all(skb);
737
738 /* record skb to CPU local list */
739 nc->skb_cache[nc->skb_count++] = skb;
740
741 #ifdef CONFIG_SLUB
742 /* SLUB writes into objects when freeing */
743 prefetchw(skb);
744 #endif
745
746 /* flush skb_cache if it is filled */
747 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
748 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE,
749 nc->skb_cache);
750 nc->skb_count = 0;
751 }
752 }
__kfree_skb_defer(struct sk_buff * skb)753 void __kfree_skb_defer(struct sk_buff *skb)
754 {
755 _kfree_skb_defer(skb);
756 }
757
napi_consume_skb(struct sk_buff * skb,int budget)758 void napi_consume_skb(struct sk_buff *skb, int budget)
759 {
760 if (unlikely(!skb))
761 return;
762
763 /* Zero budget indicate non-NAPI context called us, like netpoll */
764 if (unlikely(!budget)) {
765 dev_consume_skb_any(skb);
766 return;
767 }
768
769 if (!skb_unref(skb))
770 return;
771
772 /* if reaching here SKB is ready to free */
773 trace_consume_skb(skb);
774
775 /* if SKB is a clone, don't handle this case */
776 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
777 __kfree_skb(skb);
778 return;
779 }
780
781 _kfree_skb_defer(skb);
782 }
783 EXPORT_SYMBOL(napi_consume_skb);
784
785 /* Make sure a field is enclosed inside headers_start/headers_end section */
786 #define CHECK_SKB_FIELD(field) \
787 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
788 offsetof(struct sk_buff, headers_start)); \
789 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
790 offsetof(struct sk_buff, headers_end)); \
791
__copy_skb_header(struct sk_buff * new,const struct sk_buff * old)792 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
793 {
794 new->tstamp = old->tstamp;
795 /* We do not copy old->sk */
796 new->dev = old->dev;
797 memcpy(new->cb, old->cb, sizeof(old->cb));
798 skb_dst_copy(new, old);
799 #ifdef CONFIG_XFRM
800 new->sp = secpath_get(old->sp);
801 #endif
802 __nf_copy(new, old, false);
803
804 /* Note : this field could be in headers_start/headers_end section
805 * It is not yet because we do not want to have a 16 bit hole
806 */
807 new->queue_mapping = old->queue_mapping;
808
809 memcpy(&new->headers_start, &old->headers_start,
810 offsetof(struct sk_buff, headers_end) -
811 offsetof(struct sk_buff, headers_start));
812 CHECK_SKB_FIELD(protocol);
813 CHECK_SKB_FIELD(csum);
814 CHECK_SKB_FIELD(hash);
815 CHECK_SKB_FIELD(priority);
816 CHECK_SKB_FIELD(skb_iif);
817 CHECK_SKB_FIELD(vlan_proto);
818 CHECK_SKB_FIELD(vlan_tci);
819 CHECK_SKB_FIELD(transport_header);
820 CHECK_SKB_FIELD(network_header);
821 CHECK_SKB_FIELD(mac_header);
822 CHECK_SKB_FIELD(inner_protocol);
823 CHECK_SKB_FIELD(inner_transport_header);
824 CHECK_SKB_FIELD(inner_network_header);
825 CHECK_SKB_FIELD(inner_mac_header);
826 CHECK_SKB_FIELD(mark);
827 #ifdef CONFIG_NETWORK_SECMARK
828 CHECK_SKB_FIELD(secmark);
829 #endif
830 #ifdef CONFIG_NET_RX_BUSY_POLL
831 CHECK_SKB_FIELD(napi_id);
832 #endif
833 #ifdef CONFIG_XPS
834 CHECK_SKB_FIELD(sender_cpu);
835 #endif
836 #ifdef CONFIG_NET_SCHED
837 CHECK_SKB_FIELD(tc_index);
838 #endif
839
840 }
841
842 /*
843 * You should not add any new code to this function. Add it to
844 * __copy_skb_header above instead.
845 */
__skb_clone(struct sk_buff * n,struct sk_buff * skb)846 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
847 {
848 #define C(x) n->x = skb->x
849
850 n->next = n->prev = NULL;
851 n->sk = NULL;
852 __copy_skb_header(n, skb);
853
854 C(len);
855 C(data_len);
856 C(mac_len);
857 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
858 n->cloned = 1;
859 n->nohdr = 0;
860 n->peeked = 0;
861 C(pfmemalloc);
862 n->destructor = NULL;
863 C(tail);
864 C(end);
865 C(head);
866 C(head_frag);
867 C(data);
868 C(truesize);
869 refcount_set(&n->users, 1);
870
871 atomic_inc(&(skb_shinfo(skb)->dataref));
872 skb->cloned = 1;
873
874 return n;
875 #undef C
876 }
877
878 /**
879 * skb_morph - morph one skb into another
880 * @dst: the skb to receive the contents
881 * @src: the skb to supply the contents
882 *
883 * This is identical to skb_clone except that the target skb is
884 * supplied by the user.
885 *
886 * The target skb is returned upon exit.
887 */
skb_morph(struct sk_buff * dst,struct sk_buff * src)888 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
889 {
890 skb_release_all(dst);
891 return __skb_clone(dst, src);
892 }
893 EXPORT_SYMBOL_GPL(skb_morph);
894
mm_account_pinned_pages(struct mmpin * mmp,size_t size)895 int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
896 {
897 unsigned long max_pg, num_pg, new_pg, old_pg;
898 struct user_struct *user;
899
900 if (capable(CAP_IPC_LOCK) || !size)
901 return 0;
902
903 num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */
904 max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
905 user = mmp->user ? : current_user();
906
907 do {
908 old_pg = atomic_long_read(&user->locked_vm);
909 new_pg = old_pg + num_pg;
910 if (new_pg > max_pg)
911 return -ENOBUFS;
912 } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) !=
913 old_pg);
914
915 if (!mmp->user) {
916 mmp->user = get_uid(user);
917 mmp->num_pg = num_pg;
918 } else {
919 mmp->num_pg += num_pg;
920 }
921
922 return 0;
923 }
924 EXPORT_SYMBOL_GPL(mm_account_pinned_pages);
925
mm_unaccount_pinned_pages(struct mmpin * mmp)926 void mm_unaccount_pinned_pages(struct mmpin *mmp)
927 {
928 if (mmp->user) {
929 atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
930 free_uid(mmp->user);
931 }
932 }
933 EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages);
934
sock_zerocopy_alloc(struct sock * sk,size_t size)935 struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size)
936 {
937 struct ubuf_info *uarg;
938 struct sk_buff *skb;
939
940 WARN_ON_ONCE(!in_task());
941
942 skb = sock_omalloc(sk, 0, GFP_KERNEL);
943 if (!skb)
944 return NULL;
945
946 BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
947 uarg = (void *)skb->cb;
948 uarg->mmp.user = NULL;
949
950 if (mm_account_pinned_pages(&uarg->mmp, size)) {
951 kfree_skb(skb);
952 return NULL;
953 }
954
955 uarg->callback = sock_zerocopy_callback;
956 uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
957 uarg->len = 1;
958 uarg->bytelen = size;
959 uarg->zerocopy = 1;
960 refcount_set(&uarg->refcnt, 1);
961 sock_hold(sk);
962
963 return uarg;
964 }
965 EXPORT_SYMBOL_GPL(sock_zerocopy_alloc);
966
skb_from_uarg(struct ubuf_info * uarg)967 static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg)
968 {
969 return container_of((void *)uarg, struct sk_buff, cb);
970 }
971
sock_zerocopy_realloc(struct sock * sk,size_t size,struct ubuf_info * uarg)972 struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
973 struct ubuf_info *uarg)
974 {
975 if (uarg) {
976 const u32 byte_limit = 1 << 19; /* limit to a few TSO */
977 u32 bytelen, next;
978
979 /* realloc only when socket is locked (TCP, UDP cork),
980 * so uarg->len and sk_zckey access is serialized
981 */
982 if (!sock_owned_by_user(sk)) {
983 WARN_ON_ONCE(1);
984 return NULL;
985 }
986
987 bytelen = uarg->bytelen + size;
988 if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) {
989 /* TCP can create new skb to attach new uarg */
990 if (sk->sk_type == SOCK_STREAM)
991 goto new_alloc;
992 return NULL;
993 }
994
995 next = (u32)atomic_read(&sk->sk_zckey);
996 if ((u32)(uarg->id + uarg->len) == next) {
997 if (mm_account_pinned_pages(&uarg->mmp, size))
998 return NULL;
999 uarg->len++;
1000 uarg->bytelen = bytelen;
1001 atomic_set(&sk->sk_zckey, ++next);
1002 sock_zerocopy_get(uarg);
1003 return uarg;
1004 }
1005 }
1006
1007 new_alloc:
1008 return sock_zerocopy_alloc(sk, size);
1009 }
1010 EXPORT_SYMBOL_GPL(sock_zerocopy_realloc);
1011
skb_zerocopy_notify_extend(struct sk_buff * skb,u32 lo,u16 len)1012 static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1013 {
1014 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1015 u32 old_lo, old_hi;
1016 u64 sum_len;
1017
1018 old_lo = serr->ee.ee_info;
1019 old_hi = serr->ee.ee_data;
1020 sum_len = old_hi - old_lo + 1ULL + len;
1021
1022 if (sum_len >= (1ULL << 32))
1023 return false;
1024
1025 if (lo != old_hi + 1)
1026 return false;
1027
1028 serr->ee.ee_data += len;
1029 return true;
1030 }
1031
sock_zerocopy_callback(struct ubuf_info * uarg,bool success)1032 void sock_zerocopy_callback(struct ubuf_info *uarg, bool success)
1033 {
1034 struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1035 struct sock_exterr_skb *serr;
1036 struct sock *sk = skb->sk;
1037 struct sk_buff_head *q;
1038 unsigned long flags;
1039 u32 lo, hi;
1040 u16 len;
1041
1042 mm_unaccount_pinned_pages(&uarg->mmp);
1043
1044 /* if !len, there was only 1 call, and it was aborted
1045 * so do not queue a completion notification
1046 */
1047 if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1048 goto release;
1049
1050 len = uarg->len;
1051 lo = uarg->id;
1052 hi = uarg->id + len - 1;
1053
1054 serr = SKB_EXT_ERR(skb);
1055 memset(serr, 0, sizeof(*serr));
1056 serr->ee.ee_errno = 0;
1057 serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1058 serr->ee.ee_data = hi;
1059 serr->ee.ee_info = lo;
1060 if (!success)
1061 serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1062
1063 q = &sk->sk_error_queue;
1064 spin_lock_irqsave(&q->lock, flags);
1065 tail = skb_peek_tail(q);
1066 if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1067 !skb_zerocopy_notify_extend(tail, lo, len)) {
1068 __skb_queue_tail(q, skb);
1069 skb = NULL;
1070 }
1071 spin_unlock_irqrestore(&q->lock, flags);
1072
1073 sk->sk_error_report(sk);
1074
1075 release:
1076 consume_skb(skb);
1077 sock_put(sk);
1078 }
1079 EXPORT_SYMBOL_GPL(sock_zerocopy_callback);
1080
sock_zerocopy_put(struct ubuf_info * uarg)1081 void sock_zerocopy_put(struct ubuf_info *uarg)
1082 {
1083 if (uarg && refcount_dec_and_test(&uarg->refcnt)) {
1084 if (uarg->callback)
1085 uarg->callback(uarg, uarg->zerocopy);
1086 else
1087 consume_skb(skb_from_uarg(uarg));
1088 }
1089 }
1090 EXPORT_SYMBOL_GPL(sock_zerocopy_put);
1091
sock_zerocopy_put_abort(struct ubuf_info * uarg)1092 void sock_zerocopy_put_abort(struct ubuf_info *uarg)
1093 {
1094 if (uarg) {
1095 struct sock *sk = skb_from_uarg(uarg)->sk;
1096
1097 atomic_dec(&sk->sk_zckey);
1098 uarg->len--;
1099
1100 sock_zerocopy_put(uarg);
1101 }
1102 }
1103 EXPORT_SYMBOL_GPL(sock_zerocopy_put_abort);
1104
1105 extern int __zerocopy_sg_from_iter(struct sock *sk, struct sk_buff *skb,
1106 struct iov_iter *from, size_t length);
1107
skb_zerocopy_iter_stream(struct sock * sk,struct sk_buff * skb,struct msghdr * msg,int len,struct ubuf_info * uarg)1108 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1109 struct msghdr *msg, int len,
1110 struct ubuf_info *uarg)
1111 {
1112 struct ubuf_info *orig_uarg = skb_zcopy(skb);
1113 struct iov_iter orig_iter = msg->msg_iter;
1114 int err, orig_len = skb->len;
1115
1116 /* An skb can only point to one uarg. This edge case happens when
1117 * TCP appends to an skb, but zerocopy_realloc triggered a new alloc.
1118 */
1119 if (orig_uarg && uarg != orig_uarg)
1120 return -EEXIST;
1121
1122 err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len);
1123 if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1124 struct sock *save_sk = skb->sk;
1125
1126 /* Streams do not free skb on error. Reset to prev state. */
1127 msg->msg_iter = orig_iter;
1128 skb->sk = sk;
1129 ___pskb_trim(skb, orig_len);
1130 skb->sk = save_sk;
1131 return err;
1132 }
1133
1134 skb_zcopy_set(skb, uarg);
1135 return skb->len - orig_len;
1136 }
1137 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1138
skb_zerocopy_clone(struct sk_buff * nskb,struct sk_buff * orig,gfp_t gfp_mask)1139 static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1140 gfp_t gfp_mask)
1141 {
1142 if (skb_zcopy(orig)) {
1143 if (skb_zcopy(nskb)) {
1144 /* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1145 if (!gfp_mask) {
1146 WARN_ON_ONCE(1);
1147 return -ENOMEM;
1148 }
1149 if (skb_uarg(nskb) == skb_uarg(orig))
1150 return 0;
1151 if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1152 return -EIO;
1153 }
1154 skb_zcopy_set(nskb, skb_uarg(orig));
1155 }
1156 return 0;
1157 }
1158
1159 /**
1160 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
1161 * @skb: the skb to modify
1162 * @gfp_mask: allocation priority
1163 *
1164 * This must be called on SKBTX_DEV_ZEROCOPY skb.
1165 * It will copy all frags into kernel and drop the reference
1166 * to userspace pages.
1167 *
1168 * If this function is called from an interrupt gfp_mask() must be
1169 * %GFP_ATOMIC.
1170 *
1171 * Returns 0 on success or a negative error code on failure
1172 * to allocate kernel memory to copy to.
1173 */
skb_copy_ubufs(struct sk_buff * skb,gfp_t gfp_mask)1174 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1175 {
1176 int num_frags = skb_shinfo(skb)->nr_frags;
1177 struct page *page, *head = NULL;
1178 int i, new_frags;
1179 u32 d_off;
1180
1181 if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1182 return -EINVAL;
1183
1184 if (!num_frags)
1185 goto release;
1186
1187 new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1188 for (i = 0; i < new_frags; i++) {
1189 page = alloc_page(gfp_mask);
1190 if (!page) {
1191 while (head) {
1192 struct page *next = (struct page *)page_private(head);
1193 put_page(head);
1194 head = next;
1195 }
1196 return -ENOMEM;
1197 }
1198 set_page_private(page, (unsigned long)head);
1199 head = page;
1200 }
1201
1202 page = head;
1203 d_off = 0;
1204 for (i = 0; i < num_frags; i++) {
1205 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1206 u32 p_off, p_len, copied;
1207 struct page *p;
1208 u8 *vaddr;
1209
1210 skb_frag_foreach_page(f, f->page_offset, skb_frag_size(f),
1211 p, p_off, p_len, copied) {
1212 u32 copy, done = 0;
1213 vaddr = kmap_atomic(p);
1214
1215 while (done < p_len) {
1216 if (d_off == PAGE_SIZE) {
1217 d_off = 0;
1218 page = (struct page *)page_private(page);
1219 }
1220 copy = min_t(u32, PAGE_SIZE - d_off, p_len - done);
1221 memcpy(page_address(page) + d_off,
1222 vaddr + p_off + done, copy);
1223 done += copy;
1224 d_off += copy;
1225 }
1226 kunmap_atomic(vaddr);
1227 }
1228 }
1229
1230 /* skb frags release userspace buffers */
1231 for (i = 0; i < num_frags; i++)
1232 skb_frag_unref(skb, i);
1233
1234 /* skb frags point to kernel buffers */
1235 for (i = 0; i < new_frags - 1; i++) {
1236 __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE);
1237 head = (struct page *)page_private(head);
1238 }
1239 __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off);
1240 skb_shinfo(skb)->nr_frags = new_frags;
1241
1242 release:
1243 skb_zcopy_clear(skb, false);
1244 return 0;
1245 }
1246 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
1247
1248 /**
1249 * skb_clone - duplicate an sk_buff
1250 * @skb: buffer to clone
1251 * @gfp_mask: allocation priority
1252 *
1253 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1254 * copies share the same packet data but not structure. The new
1255 * buffer has a reference count of 1. If the allocation fails the
1256 * function returns %NULL otherwise the new buffer is returned.
1257 *
1258 * If this function is called from an interrupt gfp_mask() must be
1259 * %GFP_ATOMIC.
1260 */
1261
skb_clone(struct sk_buff * skb,gfp_t gfp_mask)1262 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1263 {
1264 struct sk_buff_fclones *fclones = container_of(skb,
1265 struct sk_buff_fclones,
1266 skb1);
1267 struct sk_buff *n;
1268
1269 if (skb_orphan_frags(skb, gfp_mask))
1270 return NULL;
1271
1272 if (skb->fclone == SKB_FCLONE_ORIG &&
1273 refcount_read(&fclones->fclone_ref) == 1) {
1274 n = &fclones->skb2;
1275 refcount_set(&fclones->fclone_ref, 2);
1276 } else {
1277 if (skb_pfmemalloc(skb))
1278 gfp_mask |= __GFP_MEMALLOC;
1279
1280 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1281 if (!n)
1282 return NULL;
1283
1284 n->fclone = SKB_FCLONE_UNAVAILABLE;
1285 }
1286
1287 return __skb_clone(n, skb);
1288 }
1289 EXPORT_SYMBOL(skb_clone);
1290
skb_headers_offset_update(struct sk_buff * skb,int off)1291 void skb_headers_offset_update(struct sk_buff *skb, int off)
1292 {
1293 /* Only adjust this if it actually is csum_start rather than csum */
1294 if (skb->ip_summed == CHECKSUM_PARTIAL)
1295 skb->csum_start += off;
1296 /* {transport,network,mac}_header and tail are relative to skb->head */
1297 skb->transport_header += off;
1298 skb->network_header += off;
1299 if (skb_mac_header_was_set(skb))
1300 skb->mac_header += off;
1301 skb->inner_transport_header += off;
1302 skb->inner_network_header += off;
1303 skb->inner_mac_header += off;
1304 }
1305 EXPORT_SYMBOL(skb_headers_offset_update);
1306
skb_copy_header(struct sk_buff * new,const struct sk_buff * old)1307 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old)
1308 {
1309 __copy_skb_header(new, old);
1310
1311 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1312 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1313 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1314 }
1315 EXPORT_SYMBOL(skb_copy_header);
1316
skb_alloc_rx_flag(const struct sk_buff * skb)1317 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1318 {
1319 if (skb_pfmemalloc(skb))
1320 return SKB_ALLOC_RX;
1321 return 0;
1322 }
1323
1324 /**
1325 * skb_copy - create private copy of an sk_buff
1326 * @skb: buffer to copy
1327 * @gfp_mask: allocation priority
1328 *
1329 * Make a copy of both an &sk_buff and its data. This is used when the
1330 * caller wishes to modify the data and needs a private copy of the
1331 * data to alter. Returns %NULL on failure or the pointer to the buffer
1332 * on success. The returned buffer has a reference count of 1.
1333 *
1334 * As by-product this function converts non-linear &sk_buff to linear
1335 * one, so that &sk_buff becomes completely private and caller is allowed
1336 * to modify all the data of returned buffer. This means that this
1337 * function is not recommended for use in circumstances when only
1338 * header is going to be modified. Use pskb_copy() instead.
1339 */
1340
skb_copy(const struct sk_buff * skb,gfp_t gfp_mask)1341 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1342 {
1343 int headerlen = skb_headroom(skb);
1344 unsigned int size = skb_end_offset(skb) + skb->data_len;
1345 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1346 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1347
1348 if (!n)
1349 return NULL;
1350
1351 /* Set the data pointer */
1352 skb_reserve(n, headerlen);
1353 /* Set the tail pointer and length */
1354 skb_put(n, skb->len);
1355
1356 BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
1357
1358 skb_copy_header(n, skb);
1359 return n;
1360 }
1361 EXPORT_SYMBOL(skb_copy);
1362
1363 /**
1364 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1365 * @skb: buffer to copy
1366 * @headroom: headroom of new skb
1367 * @gfp_mask: allocation priority
1368 * @fclone: if true allocate the copy of the skb from the fclone
1369 * cache instead of the head cache; it is recommended to set this
1370 * to true for the cases where the copy will likely be cloned
1371 *
1372 * Make a copy of both an &sk_buff and part of its data, located
1373 * in header. Fragmented data remain shared. This is used when
1374 * the caller wishes to modify only header of &sk_buff and needs
1375 * private copy of the header to alter. Returns %NULL on failure
1376 * or the pointer to the buffer on success.
1377 * The returned buffer has a reference count of 1.
1378 */
1379
__pskb_copy_fclone(struct sk_buff * skb,int headroom,gfp_t gfp_mask,bool fclone)1380 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1381 gfp_t gfp_mask, bool fclone)
1382 {
1383 unsigned int size = skb_headlen(skb) + headroom;
1384 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1385 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1386
1387 if (!n)
1388 goto out;
1389
1390 /* Set the data pointer */
1391 skb_reserve(n, headroom);
1392 /* Set the tail pointer and length */
1393 skb_put(n, skb_headlen(skb));
1394 /* Copy the bytes */
1395 skb_copy_from_linear_data(skb, n->data, n->len);
1396
1397 n->truesize += skb->data_len;
1398 n->data_len = skb->data_len;
1399 n->len = skb->len;
1400
1401 if (skb_shinfo(skb)->nr_frags) {
1402 int i;
1403
1404 if (skb_orphan_frags(skb, gfp_mask) ||
1405 skb_zerocopy_clone(n, skb, gfp_mask)) {
1406 kfree_skb(n);
1407 n = NULL;
1408 goto out;
1409 }
1410 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1411 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1412 skb_frag_ref(skb, i);
1413 }
1414 skb_shinfo(n)->nr_frags = i;
1415 }
1416
1417 if (skb_has_frag_list(skb)) {
1418 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1419 skb_clone_fraglist(n);
1420 }
1421
1422 skb_copy_header(n, skb);
1423 out:
1424 return n;
1425 }
1426 EXPORT_SYMBOL(__pskb_copy_fclone);
1427
1428 /**
1429 * pskb_expand_head - reallocate header of &sk_buff
1430 * @skb: buffer to reallocate
1431 * @nhead: room to add at head
1432 * @ntail: room to add at tail
1433 * @gfp_mask: allocation priority
1434 *
1435 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1436 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1437 * reference count of 1. Returns zero in the case of success or error,
1438 * if expansion failed. In the last case, &sk_buff is not changed.
1439 *
1440 * All the pointers pointing into skb header may change and must be
1441 * reloaded after call to this function.
1442 */
1443
pskb_expand_head(struct sk_buff * skb,int nhead,int ntail,gfp_t gfp_mask)1444 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1445 gfp_t gfp_mask)
1446 {
1447 int i, osize = skb_end_offset(skb);
1448 int size = osize + nhead + ntail;
1449 long off;
1450 u8 *data;
1451
1452 BUG_ON(nhead < 0);
1453
1454 BUG_ON(skb_shared(skb));
1455
1456 size = SKB_DATA_ALIGN(size);
1457
1458 if (skb_pfmemalloc(skb))
1459 gfp_mask |= __GFP_MEMALLOC;
1460 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1461 gfp_mask, NUMA_NO_NODE, NULL);
1462 if (!data)
1463 goto nodata;
1464 size = SKB_WITH_OVERHEAD(ksize(data));
1465
1466 /* Copy only real data... and, alas, header. This should be
1467 * optimized for the cases when header is void.
1468 */
1469 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1470
1471 memcpy((struct skb_shared_info *)(data + size),
1472 skb_shinfo(skb),
1473 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1474
1475 /*
1476 * if shinfo is shared we must drop the old head gracefully, but if it
1477 * is not we can just drop the old head and let the existing refcount
1478 * be since all we did is relocate the values
1479 */
1480 if (skb_cloned(skb)) {
1481 if (skb_orphan_frags(skb, gfp_mask))
1482 goto nofrags;
1483 if (skb_zcopy(skb))
1484 refcount_inc(&skb_uarg(skb)->refcnt);
1485 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1486 skb_frag_ref(skb, i);
1487
1488 if (skb_has_frag_list(skb))
1489 skb_clone_fraglist(skb);
1490
1491 skb_release_data(skb);
1492 } else {
1493 skb_free_head(skb);
1494 }
1495 off = (data + nhead) - skb->head;
1496
1497 skb->head = data;
1498 skb->head_frag = 0;
1499 skb->data += off;
1500 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1501 skb->end = size;
1502 off = nhead;
1503 #else
1504 skb->end = skb->head + size;
1505 #endif
1506 skb->tail += off;
1507 skb_headers_offset_update(skb, nhead);
1508 skb->cloned = 0;
1509 skb->hdr_len = 0;
1510 skb->nohdr = 0;
1511 atomic_set(&skb_shinfo(skb)->dataref, 1);
1512
1513 skb_metadata_clear(skb);
1514
1515 /* It is not generally safe to change skb->truesize.
1516 * For the moment, we really care of rx path, or
1517 * when skb is orphaned (not attached to a socket).
1518 */
1519 if (!skb->sk || skb->destructor == sock_edemux)
1520 skb->truesize += size - osize;
1521
1522 return 0;
1523
1524 nofrags:
1525 kfree(data);
1526 nodata:
1527 return -ENOMEM;
1528 }
1529 EXPORT_SYMBOL(pskb_expand_head);
1530
1531 /* Make private copy of skb with writable head and some headroom */
1532
skb_realloc_headroom(struct sk_buff * skb,unsigned int headroom)1533 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1534 {
1535 struct sk_buff *skb2;
1536 int delta = headroom - skb_headroom(skb);
1537
1538 if (delta <= 0)
1539 skb2 = pskb_copy(skb, GFP_ATOMIC);
1540 else {
1541 skb2 = skb_clone(skb, GFP_ATOMIC);
1542 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1543 GFP_ATOMIC)) {
1544 kfree_skb(skb2);
1545 skb2 = NULL;
1546 }
1547 }
1548 return skb2;
1549 }
1550 EXPORT_SYMBOL(skb_realloc_headroom);
1551
1552 /**
1553 * skb_copy_expand - copy and expand sk_buff
1554 * @skb: buffer to copy
1555 * @newheadroom: new free bytes at head
1556 * @newtailroom: new free bytes at tail
1557 * @gfp_mask: allocation priority
1558 *
1559 * Make a copy of both an &sk_buff and its data and while doing so
1560 * allocate additional space.
1561 *
1562 * This is used when the caller wishes to modify the data and needs a
1563 * private copy of the data to alter as well as more space for new fields.
1564 * Returns %NULL on failure or the pointer to the buffer
1565 * on success. The returned buffer has a reference count of 1.
1566 *
1567 * You must pass %GFP_ATOMIC as the allocation priority if this function
1568 * is called from an interrupt.
1569 */
skb_copy_expand(const struct sk_buff * skb,int newheadroom,int newtailroom,gfp_t gfp_mask)1570 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1571 int newheadroom, int newtailroom,
1572 gfp_t gfp_mask)
1573 {
1574 /*
1575 * Allocate the copy buffer
1576 */
1577 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1578 gfp_mask, skb_alloc_rx_flag(skb),
1579 NUMA_NO_NODE);
1580 int oldheadroom = skb_headroom(skb);
1581 int head_copy_len, head_copy_off;
1582
1583 if (!n)
1584 return NULL;
1585
1586 skb_reserve(n, newheadroom);
1587
1588 /* Set the tail pointer and length */
1589 skb_put(n, skb->len);
1590
1591 head_copy_len = oldheadroom;
1592 head_copy_off = 0;
1593 if (newheadroom <= head_copy_len)
1594 head_copy_len = newheadroom;
1595 else
1596 head_copy_off = newheadroom - head_copy_len;
1597
1598 /* Copy the linear header and data. */
1599 BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1600 skb->len + head_copy_len));
1601
1602 skb_copy_header(n, skb);
1603
1604 skb_headers_offset_update(n, newheadroom - oldheadroom);
1605
1606 return n;
1607 }
1608 EXPORT_SYMBOL(skb_copy_expand);
1609
1610 /**
1611 * __skb_pad - zero pad the tail of an skb
1612 * @skb: buffer to pad
1613 * @pad: space to pad
1614 * @free_on_error: free buffer on error
1615 *
1616 * Ensure that a buffer is followed by a padding area that is zero
1617 * filled. Used by network drivers which may DMA or transfer data
1618 * beyond the buffer end onto the wire.
1619 *
1620 * May return error in out of memory cases. The skb is freed on error
1621 * if @free_on_error is true.
1622 */
1623
__skb_pad(struct sk_buff * skb,int pad,bool free_on_error)1624 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error)
1625 {
1626 int err;
1627 int ntail;
1628
1629 /* If the skbuff is non linear tailroom is always zero.. */
1630 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1631 memset(skb->data+skb->len, 0, pad);
1632 return 0;
1633 }
1634
1635 ntail = skb->data_len + pad - (skb->end - skb->tail);
1636 if (likely(skb_cloned(skb) || ntail > 0)) {
1637 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1638 if (unlikely(err))
1639 goto free_skb;
1640 }
1641
1642 /* FIXME: The use of this function with non-linear skb's really needs
1643 * to be audited.
1644 */
1645 err = skb_linearize(skb);
1646 if (unlikely(err))
1647 goto free_skb;
1648
1649 memset(skb->data + skb->len, 0, pad);
1650 return 0;
1651
1652 free_skb:
1653 if (free_on_error)
1654 kfree_skb(skb);
1655 return err;
1656 }
1657 EXPORT_SYMBOL(__skb_pad);
1658
1659 /**
1660 * pskb_put - add data to the tail of a potentially fragmented buffer
1661 * @skb: start of the buffer to use
1662 * @tail: tail fragment of the buffer to use
1663 * @len: amount of data to add
1664 *
1665 * This function extends the used data area of the potentially
1666 * fragmented buffer. @tail must be the last fragment of @skb -- or
1667 * @skb itself. If this would exceed the total buffer size the kernel
1668 * will panic. A pointer to the first byte of the extra data is
1669 * returned.
1670 */
1671
pskb_put(struct sk_buff * skb,struct sk_buff * tail,int len)1672 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1673 {
1674 if (tail != skb) {
1675 skb->data_len += len;
1676 skb->len += len;
1677 }
1678 return skb_put(tail, len);
1679 }
1680 EXPORT_SYMBOL_GPL(pskb_put);
1681
1682 /**
1683 * skb_put - add data to a buffer
1684 * @skb: buffer to use
1685 * @len: amount of data to add
1686 *
1687 * This function extends the used data area of the buffer. If this would
1688 * exceed the total buffer size the kernel will panic. A pointer to the
1689 * first byte of the extra data is returned.
1690 */
skb_put(struct sk_buff * skb,unsigned int len)1691 void *skb_put(struct sk_buff *skb, unsigned int len)
1692 {
1693 void *tmp = skb_tail_pointer(skb);
1694 SKB_LINEAR_ASSERT(skb);
1695 skb->tail += len;
1696 skb->len += len;
1697 if (unlikely(skb->tail > skb->end))
1698 skb_over_panic(skb, len, __builtin_return_address(0));
1699 return tmp;
1700 }
1701 EXPORT_SYMBOL(skb_put);
1702
1703 /**
1704 * skb_push - add data to the start of a buffer
1705 * @skb: buffer to use
1706 * @len: amount of data to add
1707 *
1708 * This function extends the used data area of the buffer at the buffer
1709 * start. If this would exceed the total buffer headroom the kernel will
1710 * panic. A pointer to the first byte of the extra data is returned.
1711 */
skb_push(struct sk_buff * skb,unsigned int len)1712 void *skb_push(struct sk_buff *skb, unsigned int len)
1713 {
1714 skb->data -= len;
1715 skb->len += len;
1716 if (unlikely(skb->data < skb->head))
1717 skb_under_panic(skb, len, __builtin_return_address(0));
1718 return skb->data;
1719 }
1720 EXPORT_SYMBOL(skb_push);
1721
1722 /**
1723 * skb_pull - remove data from the start of a buffer
1724 * @skb: buffer to use
1725 * @len: amount of data to remove
1726 *
1727 * This function removes data from the start of a buffer, returning
1728 * the memory to the headroom. A pointer to the next data in the buffer
1729 * is returned. Once the data has been pulled future pushes will overwrite
1730 * the old data.
1731 */
skb_pull(struct sk_buff * skb,unsigned int len)1732 void *skb_pull(struct sk_buff *skb, unsigned int len)
1733 {
1734 return skb_pull_inline(skb, len);
1735 }
1736 EXPORT_SYMBOL(skb_pull);
1737
1738 /**
1739 * skb_trim - remove end from a buffer
1740 * @skb: buffer to alter
1741 * @len: new length
1742 *
1743 * Cut the length of a buffer down by removing data from the tail. If
1744 * the buffer is already under the length specified it is not modified.
1745 * The skb must be linear.
1746 */
skb_trim(struct sk_buff * skb,unsigned int len)1747 void skb_trim(struct sk_buff *skb, unsigned int len)
1748 {
1749 if (skb->len > len)
1750 __skb_trim(skb, len);
1751 }
1752 EXPORT_SYMBOL(skb_trim);
1753
1754 /* Trims skb to length len. It can change skb pointers.
1755 */
1756
___pskb_trim(struct sk_buff * skb,unsigned int len)1757 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1758 {
1759 struct sk_buff **fragp;
1760 struct sk_buff *frag;
1761 int offset = skb_headlen(skb);
1762 int nfrags = skb_shinfo(skb)->nr_frags;
1763 int i;
1764 int err;
1765
1766 if (skb_cloned(skb) &&
1767 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1768 return err;
1769
1770 i = 0;
1771 if (offset >= len)
1772 goto drop_pages;
1773
1774 for (; i < nfrags; i++) {
1775 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1776
1777 if (end < len) {
1778 offset = end;
1779 continue;
1780 }
1781
1782 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1783
1784 drop_pages:
1785 skb_shinfo(skb)->nr_frags = i;
1786
1787 for (; i < nfrags; i++)
1788 skb_frag_unref(skb, i);
1789
1790 if (skb_has_frag_list(skb))
1791 skb_drop_fraglist(skb);
1792 goto done;
1793 }
1794
1795 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1796 fragp = &frag->next) {
1797 int end = offset + frag->len;
1798
1799 if (skb_shared(frag)) {
1800 struct sk_buff *nfrag;
1801
1802 nfrag = skb_clone(frag, GFP_ATOMIC);
1803 if (unlikely(!nfrag))
1804 return -ENOMEM;
1805
1806 nfrag->next = frag->next;
1807 consume_skb(frag);
1808 frag = nfrag;
1809 *fragp = frag;
1810 }
1811
1812 if (end < len) {
1813 offset = end;
1814 continue;
1815 }
1816
1817 if (end > len &&
1818 unlikely((err = pskb_trim(frag, len - offset))))
1819 return err;
1820
1821 if (frag->next)
1822 skb_drop_list(&frag->next);
1823 break;
1824 }
1825
1826 done:
1827 if (len > skb_headlen(skb)) {
1828 skb->data_len -= skb->len - len;
1829 skb->len = len;
1830 } else {
1831 skb->len = len;
1832 skb->data_len = 0;
1833 skb_set_tail_pointer(skb, len);
1834 }
1835
1836 if (!skb->sk || skb->destructor == sock_edemux)
1837 skb_condense(skb);
1838 return 0;
1839 }
1840 EXPORT_SYMBOL(___pskb_trim);
1841
1842 /* Note : use pskb_trim_rcsum() instead of calling this directly
1843 */
pskb_trim_rcsum_slow(struct sk_buff * skb,unsigned int len)1844 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len)
1845 {
1846 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1847 int delta = skb->len - len;
1848
1849 skb->csum = csum_block_sub(skb->csum,
1850 skb_checksum(skb, len, delta, 0),
1851 len);
1852 }
1853 return __pskb_trim(skb, len);
1854 }
1855 EXPORT_SYMBOL(pskb_trim_rcsum_slow);
1856
1857 /**
1858 * __pskb_pull_tail - advance tail of skb header
1859 * @skb: buffer to reallocate
1860 * @delta: number of bytes to advance tail
1861 *
1862 * The function makes a sense only on a fragmented &sk_buff,
1863 * it expands header moving its tail forward and copying necessary
1864 * data from fragmented part.
1865 *
1866 * &sk_buff MUST have reference count of 1.
1867 *
1868 * Returns %NULL (and &sk_buff does not change) if pull failed
1869 * or value of new tail of skb in the case of success.
1870 *
1871 * All the pointers pointing into skb header may change and must be
1872 * reloaded after call to this function.
1873 */
1874
1875 /* Moves tail of skb head forward, copying data from fragmented part,
1876 * when it is necessary.
1877 * 1. It may fail due to malloc failure.
1878 * 2. It may change skb pointers.
1879 *
1880 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1881 */
__pskb_pull_tail(struct sk_buff * skb,int delta)1882 void *__pskb_pull_tail(struct sk_buff *skb, int delta)
1883 {
1884 /* If skb has not enough free space at tail, get new one
1885 * plus 128 bytes for future expansions. If we have enough
1886 * room at tail, reallocate without expansion only if skb is cloned.
1887 */
1888 int i, k, eat = (skb->tail + delta) - skb->end;
1889
1890 if (eat > 0 || skb_cloned(skb)) {
1891 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1892 GFP_ATOMIC))
1893 return NULL;
1894 }
1895
1896 BUG_ON(skb_copy_bits(skb, skb_headlen(skb),
1897 skb_tail_pointer(skb), delta));
1898
1899 /* Optimization: no fragments, no reasons to preestimate
1900 * size of pulled pages. Superb.
1901 */
1902 if (!skb_has_frag_list(skb))
1903 goto pull_pages;
1904
1905 /* Estimate size of pulled pages. */
1906 eat = delta;
1907 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1908 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1909
1910 if (size >= eat)
1911 goto pull_pages;
1912 eat -= size;
1913 }
1914
1915 /* If we need update frag list, we are in troubles.
1916 * Certainly, it is possible to add an offset to skb data,
1917 * but taking into account that pulling is expected to
1918 * be very rare operation, it is worth to fight against
1919 * further bloating skb head and crucify ourselves here instead.
1920 * Pure masohism, indeed. 8)8)
1921 */
1922 if (eat) {
1923 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1924 struct sk_buff *clone = NULL;
1925 struct sk_buff *insp = NULL;
1926
1927 do {
1928 BUG_ON(!list);
1929
1930 if (list->len <= eat) {
1931 /* Eaten as whole. */
1932 eat -= list->len;
1933 list = list->next;
1934 insp = list;
1935 } else {
1936 /* Eaten partially. */
1937
1938 if (skb_shared(list)) {
1939 /* Sucks! We need to fork list. :-( */
1940 clone = skb_clone(list, GFP_ATOMIC);
1941 if (!clone)
1942 return NULL;
1943 insp = list->next;
1944 list = clone;
1945 } else {
1946 /* This may be pulled without
1947 * problems. */
1948 insp = list;
1949 }
1950 if (!pskb_pull(list, eat)) {
1951 kfree_skb(clone);
1952 return NULL;
1953 }
1954 break;
1955 }
1956 } while (eat);
1957
1958 /* Free pulled out fragments. */
1959 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1960 skb_shinfo(skb)->frag_list = list->next;
1961 kfree_skb(list);
1962 }
1963 /* And insert new clone at head. */
1964 if (clone) {
1965 clone->next = list;
1966 skb_shinfo(skb)->frag_list = clone;
1967 }
1968 }
1969 /* Success! Now we may commit changes to skb data. */
1970
1971 pull_pages:
1972 eat = delta;
1973 k = 0;
1974 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1975 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1976
1977 if (size <= eat) {
1978 skb_frag_unref(skb, i);
1979 eat -= size;
1980 } else {
1981 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1982 if (eat) {
1983 skb_shinfo(skb)->frags[k].page_offset += eat;
1984 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1985 if (!i)
1986 goto end;
1987 eat = 0;
1988 }
1989 k++;
1990 }
1991 }
1992 skb_shinfo(skb)->nr_frags = k;
1993
1994 end:
1995 skb->tail += delta;
1996 skb->data_len -= delta;
1997
1998 if (!skb->data_len)
1999 skb_zcopy_clear(skb, false);
2000
2001 return skb_tail_pointer(skb);
2002 }
2003 EXPORT_SYMBOL(__pskb_pull_tail);
2004
2005 /**
2006 * skb_copy_bits - copy bits from skb to kernel buffer
2007 * @skb: source skb
2008 * @offset: offset in source
2009 * @to: destination buffer
2010 * @len: number of bytes to copy
2011 *
2012 * Copy the specified number of bytes from the source skb to the
2013 * destination buffer.
2014 *
2015 * CAUTION ! :
2016 * If its prototype is ever changed,
2017 * check arch/{*}/net/{*}.S files,
2018 * since it is called from BPF assembly code.
2019 */
skb_copy_bits(const struct sk_buff * skb,int offset,void * to,int len)2020 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2021 {
2022 int start = skb_headlen(skb);
2023 struct sk_buff *frag_iter;
2024 int i, copy;
2025
2026 if (offset > (int)skb->len - len)
2027 goto fault;
2028
2029 /* Copy header. */
2030 if ((copy = start - offset) > 0) {
2031 if (copy > len)
2032 copy = len;
2033 skb_copy_from_linear_data_offset(skb, offset, to, copy);
2034 if ((len -= copy) == 0)
2035 return 0;
2036 offset += copy;
2037 to += copy;
2038 }
2039
2040 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2041 int end;
2042 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2043
2044 WARN_ON(start > offset + len);
2045
2046 end = start + skb_frag_size(f);
2047 if ((copy = end - offset) > 0) {
2048 u32 p_off, p_len, copied;
2049 struct page *p;
2050 u8 *vaddr;
2051
2052 if (copy > len)
2053 copy = len;
2054
2055 skb_frag_foreach_page(f,
2056 f->page_offset + offset - start,
2057 copy, p, p_off, p_len, copied) {
2058 vaddr = kmap_atomic(p);
2059 memcpy(to + copied, vaddr + p_off, p_len);
2060 kunmap_atomic(vaddr);
2061 }
2062
2063 if ((len -= copy) == 0)
2064 return 0;
2065 offset += copy;
2066 to += copy;
2067 }
2068 start = end;
2069 }
2070
2071 skb_walk_frags(skb, frag_iter) {
2072 int end;
2073
2074 WARN_ON(start > offset + len);
2075
2076 end = start + frag_iter->len;
2077 if ((copy = end - offset) > 0) {
2078 if (copy > len)
2079 copy = len;
2080 if (skb_copy_bits(frag_iter, offset - start, to, copy))
2081 goto fault;
2082 if ((len -= copy) == 0)
2083 return 0;
2084 offset += copy;
2085 to += copy;
2086 }
2087 start = end;
2088 }
2089
2090 if (!len)
2091 return 0;
2092
2093 fault:
2094 return -EFAULT;
2095 }
2096 EXPORT_SYMBOL(skb_copy_bits);
2097
2098 /*
2099 * Callback from splice_to_pipe(), if we need to release some pages
2100 * at the end of the spd in case we error'ed out in filling the pipe.
2101 */
sock_spd_release(struct splice_pipe_desc * spd,unsigned int i)2102 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
2103 {
2104 put_page(spd->pages[i]);
2105 }
2106
linear_to_page(struct page * page,unsigned int * len,unsigned int * offset,struct sock * sk)2107 static struct page *linear_to_page(struct page *page, unsigned int *len,
2108 unsigned int *offset,
2109 struct sock *sk)
2110 {
2111 struct page_frag *pfrag = sk_page_frag(sk);
2112
2113 if (!sk_page_frag_refill(sk, pfrag))
2114 return NULL;
2115
2116 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
2117
2118 memcpy(page_address(pfrag->page) + pfrag->offset,
2119 page_address(page) + *offset, *len);
2120 *offset = pfrag->offset;
2121 pfrag->offset += *len;
2122
2123 return pfrag->page;
2124 }
2125
spd_can_coalesce(const struct splice_pipe_desc * spd,struct page * page,unsigned int offset)2126 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
2127 struct page *page,
2128 unsigned int offset)
2129 {
2130 return spd->nr_pages &&
2131 spd->pages[spd->nr_pages - 1] == page &&
2132 (spd->partial[spd->nr_pages - 1].offset +
2133 spd->partial[spd->nr_pages - 1].len == offset);
2134 }
2135
2136 /*
2137 * Fill page/offset/length into spd, if it can hold more pages.
2138 */
spd_fill_page(struct splice_pipe_desc * spd,struct pipe_inode_info * pipe,struct page * page,unsigned int * len,unsigned int offset,bool linear,struct sock * sk)2139 static bool spd_fill_page(struct splice_pipe_desc *spd,
2140 struct pipe_inode_info *pipe, struct page *page,
2141 unsigned int *len, unsigned int offset,
2142 bool linear,
2143 struct sock *sk)
2144 {
2145 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
2146 return true;
2147
2148 if (linear) {
2149 page = linear_to_page(page, len, &offset, sk);
2150 if (!page)
2151 return true;
2152 }
2153 if (spd_can_coalesce(spd, page, offset)) {
2154 spd->partial[spd->nr_pages - 1].len += *len;
2155 return false;
2156 }
2157 get_page(page);
2158 spd->pages[spd->nr_pages] = page;
2159 spd->partial[spd->nr_pages].len = *len;
2160 spd->partial[spd->nr_pages].offset = offset;
2161 spd->nr_pages++;
2162
2163 return false;
2164 }
2165
__splice_segment(struct page * page,unsigned int poff,unsigned int plen,unsigned int * off,unsigned int * len,struct splice_pipe_desc * spd,bool linear,struct sock * sk,struct pipe_inode_info * pipe)2166 static bool __splice_segment(struct page *page, unsigned int poff,
2167 unsigned int plen, unsigned int *off,
2168 unsigned int *len,
2169 struct splice_pipe_desc *spd, bool linear,
2170 struct sock *sk,
2171 struct pipe_inode_info *pipe)
2172 {
2173 if (!*len)
2174 return true;
2175
2176 /* skip this segment if already processed */
2177 if (*off >= plen) {
2178 *off -= plen;
2179 return false;
2180 }
2181
2182 /* ignore any bits we already processed */
2183 poff += *off;
2184 plen -= *off;
2185 *off = 0;
2186
2187 do {
2188 unsigned int flen = min(*len, plen);
2189
2190 if (spd_fill_page(spd, pipe, page, &flen, poff,
2191 linear, sk))
2192 return true;
2193 poff += flen;
2194 plen -= flen;
2195 *len -= flen;
2196 } while (*len && plen);
2197
2198 return false;
2199 }
2200
2201 /*
2202 * Map linear and fragment data from the skb to spd. It reports true if the
2203 * pipe is full or if we already spliced the requested length.
2204 */
__skb_splice_bits(struct sk_buff * skb,struct pipe_inode_info * pipe,unsigned int * offset,unsigned int * len,struct splice_pipe_desc * spd,struct sock * sk)2205 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
2206 unsigned int *offset, unsigned int *len,
2207 struct splice_pipe_desc *spd, struct sock *sk)
2208 {
2209 int seg;
2210 struct sk_buff *iter;
2211
2212 /* map the linear part :
2213 * If skb->head_frag is set, this 'linear' part is backed by a
2214 * fragment, and if the head is not shared with any clones then
2215 * we can avoid a copy since we own the head portion of this page.
2216 */
2217 if (__splice_segment(virt_to_page(skb->data),
2218 (unsigned long) skb->data & (PAGE_SIZE - 1),
2219 skb_headlen(skb),
2220 offset, len, spd,
2221 skb_head_is_locked(skb),
2222 sk, pipe))
2223 return true;
2224
2225 /*
2226 * then map the fragments
2227 */
2228 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
2229 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
2230
2231 if (__splice_segment(skb_frag_page(f),
2232 f->page_offset, skb_frag_size(f),
2233 offset, len, spd, false, sk, pipe))
2234 return true;
2235 }
2236
2237 skb_walk_frags(skb, iter) {
2238 if (*offset >= iter->len) {
2239 *offset -= iter->len;
2240 continue;
2241 }
2242 /* __skb_splice_bits() only fails if the output has no room
2243 * left, so no point in going over the frag_list for the error
2244 * case.
2245 */
2246 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
2247 return true;
2248 }
2249
2250 return false;
2251 }
2252
2253 /*
2254 * Map data from the skb to a pipe. Should handle both the linear part,
2255 * the fragments, and the frag list.
2256 */
skb_splice_bits(struct sk_buff * skb,struct sock * sk,unsigned int offset,struct pipe_inode_info * pipe,unsigned int tlen,unsigned int flags)2257 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2258 struct pipe_inode_info *pipe, unsigned int tlen,
2259 unsigned int flags)
2260 {
2261 struct partial_page partial[MAX_SKB_FRAGS];
2262 struct page *pages[MAX_SKB_FRAGS];
2263 struct splice_pipe_desc spd = {
2264 .pages = pages,
2265 .partial = partial,
2266 .nr_pages_max = MAX_SKB_FRAGS,
2267 .ops = &nosteal_pipe_buf_ops,
2268 .spd_release = sock_spd_release,
2269 };
2270 int ret = 0;
2271
2272 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2273
2274 if (spd.nr_pages)
2275 ret = splice_to_pipe(pipe, &spd);
2276
2277 return ret;
2278 }
2279 EXPORT_SYMBOL_GPL(skb_splice_bits);
2280
2281 /* Send skb data on a socket. Socket must be locked. */
skb_send_sock_locked(struct sock * sk,struct sk_buff * skb,int offset,int len)2282 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
2283 int len)
2284 {
2285 unsigned int orig_len = len;
2286 struct sk_buff *head = skb;
2287 unsigned short fragidx;
2288 int slen, ret;
2289
2290 do_frag_list:
2291
2292 /* Deal with head data */
2293 while (offset < skb_headlen(skb) && len) {
2294 struct kvec kv;
2295 struct msghdr msg;
2296
2297 slen = min_t(int, len, skb_headlen(skb) - offset);
2298 kv.iov_base = skb->data + offset;
2299 kv.iov_len = slen;
2300 memset(&msg, 0, sizeof(msg));
2301
2302 ret = kernel_sendmsg_locked(sk, &msg, &kv, 1, slen);
2303 if (ret <= 0)
2304 goto error;
2305
2306 offset += ret;
2307 len -= ret;
2308 }
2309
2310 /* All the data was skb head? */
2311 if (!len)
2312 goto out;
2313
2314 /* Make offset relative to start of frags */
2315 offset -= skb_headlen(skb);
2316
2317 /* Find where we are in frag list */
2318 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2319 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2320
2321 if (offset < frag->size)
2322 break;
2323
2324 offset -= frag->size;
2325 }
2326
2327 for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2328 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2329
2330 slen = min_t(size_t, len, frag->size - offset);
2331
2332 while (slen) {
2333 ret = kernel_sendpage_locked(sk, frag->page.p,
2334 frag->page_offset + offset,
2335 slen, MSG_DONTWAIT);
2336 if (ret <= 0)
2337 goto error;
2338
2339 len -= ret;
2340 offset += ret;
2341 slen -= ret;
2342 }
2343
2344 offset = 0;
2345 }
2346
2347 if (len) {
2348 /* Process any frag lists */
2349
2350 if (skb == head) {
2351 if (skb_has_frag_list(skb)) {
2352 skb = skb_shinfo(skb)->frag_list;
2353 goto do_frag_list;
2354 }
2355 } else if (skb->next) {
2356 skb = skb->next;
2357 goto do_frag_list;
2358 }
2359 }
2360
2361 out:
2362 return orig_len - len;
2363
2364 error:
2365 return orig_len == len ? ret : orig_len - len;
2366 }
2367 EXPORT_SYMBOL_GPL(skb_send_sock_locked);
2368
2369 /* Send skb data on a socket. */
skb_send_sock(struct sock * sk,struct sk_buff * skb,int offset,int len)2370 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len)
2371 {
2372 int ret = 0;
2373
2374 lock_sock(sk);
2375 ret = skb_send_sock_locked(sk, skb, offset, len);
2376 release_sock(sk);
2377
2378 return ret;
2379 }
2380 EXPORT_SYMBOL_GPL(skb_send_sock);
2381
2382 /**
2383 * skb_store_bits - store bits from kernel buffer to skb
2384 * @skb: destination buffer
2385 * @offset: offset in destination
2386 * @from: source buffer
2387 * @len: number of bytes to copy
2388 *
2389 * Copy the specified number of bytes from the source buffer to the
2390 * destination skb. This function handles all the messy bits of
2391 * traversing fragment lists and such.
2392 */
2393
skb_store_bits(struct sk_buff * skb,int offset,const void * from,int len)2394 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2395 {
2396 int start = skb_headlen(skb);
2397 struct sk_buff *frag_iter;
2398 int i, copy;
2399
2400 if (offset > (int)skb->len - len)
2401 goto fault;
2402
2403 if ((copy = start - offset) > 0) {
2404 if (copy > len)
2405 copy = len;
2406 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2407 if ((len -= copy) == 0)
2408 return 0;
2409 offset += copy;
2410 from += copy;
2411 }
2412
2413 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2414 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2415 int end;
2416
2417 WARN_ON(start > offset + len);
2418
2419 end = start + skb_frag_size(frag);
2420 if ((copy = end - offset) > 0) {
2421 u32 p_off, p_len, copied;
2422 struct page *p;
2423 u8 *vaddr;
2424
2425 if (copy > len)
2426 copy = len;
2427
2428 skb_frag_foreach_page(frag,
2429 frag->page_offset + offset - start,
2430 copy, p, p_off, p_len, copied) {
2431 vaddr = kmap_atomic(p);
2432 memcpy(vaddr + p_off, from + copied, p_len);
2433 kunmap_atomic(vaddr);
2434 }
2435
2436 if ((len -= copy) == 0)
2437 return 0;
2438 offset += copy;
2439 from += copy;
2440 }
2441 start = end;
2442 }
2443
2444 skb_walk_frags(skb, frag_iter) {
2445 int end;
2446
2447 WARN_ON(start > offset + len);
2448
2449 end = start + frag_iter->len;
2450 if ((copy = end - offset) > 0) {
2451 if (copy > len)
2452 copy = len;
2453 if (skb_store_bits(frag_iter, offset - start,
2454 from, copy))
2455 goto fault;
2456 if ((len -= copy) == 0)
2457 return 0;
2458 offset += copy;
2459 from += copy;
2460 }
2461 start = end;
2462 }
2463 if (!len)
2464 return 0;
2465
2466 fault:
2467 return -EFAULT;
2468 }
2469 EXPORT_SYMBOL(skb_store_bits);
2470
2471 /* Checksum skb data. */
__skb_checksum(const struct sk_buff * skb,int offset,int len,__wsum csum,const struct skb_checksum_ops * ops)2472 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2473 __wsum csum, const struct skb_checksum_ops *ops)
2474 {
2475 int start = skb_headlen(skb);
2476 int i, copy = start - offset;
2477 struct sk_buff *frag_iter;
2478 int pos = 0;
2479
2480 /* Checksum header. */
2481 if (copy > 0) {
2482 if (copy > len)
2483 copy = len;
2484 csum = ops->update(skb->data + offset, copy, csum);
2485 if ((len -= copy) == 0)
2486 return csum;
2487 offset += copy;
2488 pos = copy;
2489 }
2490
2491 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2492 int end;
2493 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2494
2495 WARN_ON(start > offset + len);
2496
2497 end = start + skb_frag_size(frag);
2498 if ((copy = end - offset) > 0) {
2499 u32 p_off, p_len, copied;
2500 struct page *p;
2501 __wsum csum2;
2502 u8 *vaddr;
2503
2504 if (copy > len)
2505 copy = len;
2506
2507 skb_frag_foreach_page(frag,
2508 frag->page_offset + offset - start,
2509 copy, p, p_off, p_len, copied) {
2510 vaddr = kmap_atomic(p);
2511 csum2 = ops->update(vaddr + p_off, p_len, 0);
2512 kunmap_atomic(vaddr);
2513 csum = ops->combine(csum, csum2, pos, p_len);
2514 pos += p_len;
2515 }
2516
2517 if (!(len -= copy))
2518 return csum;
2519 offset += copy;
2520 }
2521 start = end;
2522 }
2523
2524 skb_walk_frags(skb, frag_iter) {
2525 int end;
2526
2527 WARN_ON(start > offset + len);
2528
2529 end = start + frag_iter->len;
2530 if ((copy = end - offset) > 0) {
2531 __wsum csum2;
2532 if (copy > len)
2533 copy = len;
2534 csum2 = __skb_checksum(frag_iter, offset - start,
2535 copy, 0, ops);
2536 csum = ops->combine(csum, csum2, pos, copy);
2537 if ((len -= copy) == 0)
2538 return csum;
2539 offset += copy;
2540 pos += copy;
2541 }
2542 start = end;
2543 }
2544 BUG_ON(len);
2545
2546 return csum;
2547 }
2548 EXPORT_SYMBOL(__skb_checksum);
2549
skb_checksum(const struct sk_buff * skb,int offset,int len,__wsum csum)2550 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2551 int len, __wsum csum)
2552 {
2553 const struct skb_checksum_ops ops = {
2554 .update = csum_partial_ext,
2555 .combine = csum_block_add_ext,
2556 };
2557
2558 return __skb_checksum(skb, offset, len, csum, &ops);
2559 }
2560 EXPORT_SYMBOL(skb_checksum);
2561
2562 /* Both of above in one bottle. */
2563
skb_copy_and_csum_bits(const struct sk_buff * skb,int offset,u8 * to,int len,__wsum csum)2564 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2565 u8 *to, int len, __wsum csum)
2566 {
2567 int start = skb_headlen(skb);
2568 int i, copy = start - offset;
2569 struct sk_buff *frag_iter;
2570 int pos = 0;
2571
2572 /* Copy header. */
2573 if (copy > 0) {
2574 if (copy > len)
2575 copy = len;
2576 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2577 copy, csum);
2578 if ((len -= copy) == 0)
2579 return csum;
2580 offset += copy;
2581 to += copy;
2582 pos = copy;
2583 }
2584
2585 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2586 int end;
2587
2588 WARN_ON(start > offset + len);
2589
2590 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2591 if ((copy = end - offset) > 0) {
2592 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2593 u32 p_off, p_len, copied;
2594 struct page *p;
2595 __wsum csum2;
2596 u8 *vaddr;
2597
2598 if (copy > len)
2599 copy = len;
2600
2601 skb_frag_foreach_page(frag,
2602 frag->page_offset + offset - start,
2603 copy, p, p_off, p_len, copied) {
2604 vaddr = kmap_atomic(p);
2605 csum2 = csum_partial_copy_nocheck(vaddr + p_off,
2606 to + copied,
2607 p_len, 0);
2608 kunmap_atomic(vaddr);
2609 csum = csum_block_add(csum, csum2, pos);
2610 pos += p_len;
2611 }
2612
2613 if (!(len -= copy))
2614 return csum;
2615 offset += copy;
2616 to += copy;
2617 }
2618 start = end;
2619 }
2620
2621 skb_walk_frags(skb, frag_iter) {
2622 __wsum csum2;
2623 int end;
2624
2625 WARN_ON(start > offset + len);
2626
2627 end = start + frag_iter->len;
2628 if ((copy = end - offset) > 0) {
2629 if (copy > len)
2630 copy = len;
2631 csum2 = skb_copy_and_csum_bits(frag_iter,
2632 offset - start,
2633 to, copy, 0);
2634 csum = csum_block_add(csum, csum2, pos);
2635 if ((len -= copy) == 0)
2636 return csum;
2637 offset += copy;
2638 to += copy;
2639 pos += copy;
2640 }
2641 start = end;
2642 }
2643 BUG_ON(len);
2644 return csum;
2645 }
2646 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2647
warn_crc32c_csum_update(const void * buff,int len,__wsum sum)2648 static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
2649 {
2650 net_warn_ratelimited(
2651 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2652 __func__);
2653 return 0;
2654 }
2655
warn_crc32c_csum_combine(__wsum csum,__wsum csum2,int offset,int len)2656 static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
2657 int offset, int len)
2658 {
2659 net_warn_ratelimited(
2660 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2661 __func__);
2662 return 0;
2663 }
2664
2665 static const struct skb_checksum_ops default_crc32c_ops = {
2666 .update = warn_crc32c_csum_update,
2667 .combine = warn_crc32c_csum_combine,
2668 };
2669
2670 const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
2671 &default_crc32c_ops;
2672 EXPORT_SYMBOL(crc32c_csum_stub);
2673
2674 /**
2675 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2676 * @from: source buffer
2677 *
2678 * Calculates the amount of linear headroom needed in the 'to' skb passed
2679 * into skb_zerocopy().
2680 */
2681 unsigned int
skb_zerocopy_headlen(const struct sk_buff * from)2682 skb_zerocopy_headlen(const struct sk_buff *from)
2683 {
2684 unsigned int hlen = 0;
2685
2686 if (!from->head_frag ||
2687 skb_headlen(from) < L1_CACHE_BYTES ||
2688 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2689 hlen = skb_headlen(from);
2690
2691 if (skb_has_frag_list(from))
2692 hlen = from->len;
2693
2694 return hlen;
2695 }
2696 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2697
2698 /**
2699 * skb_zerocopy - Zero copy skb to skb
2700 * @to: destination buffer
2701 * @from: source buffer
2702 * @len: number of bytes to copy from source buffer
2703 * @hlen: size of linear headroom in destination buffer
2704 *
2705 * Copies up to `len` bytes from `from` to `to` by creating references
2706 * to the frags in the source buffer.
2707 *
2708 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2709 * headroom in the `to` buffer.
2710 *
2711 * Return value:
2712 * 0: everything is OK
2713 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2714 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2715 */
2716 int
skb_zerocopy(struct sk_buff * to,struct sk_buff * from,int len,int hlen)2717 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2718 {
2719 int i, j = 0;
2720 int plen = 0; /* length of skb->head fragment */
2721 int ret;
2722 struct page *page;
2723 unsigned int offset;
2724
2725 BUG_ON(!from->head_frag && !hlen);
2726
2727 /* dont bother with small payloads */
2728 if (len <= skb_tailroom(to))
2729 return skb_copy_bits(from, 0, skb_put(to, len), len);
2730
2731 if (hlen) {
2732 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2733 if (unlikely(ret))
2734 return ret;
2735 len -= hlen;
2736 } else {
2737 plen = min_t(int, skb_headlen(from), len);
2738 if (plen) {
2739 page = virt_to_head_page(from->head);
2740 offset = from->data - (unsigned char *)page_address(page);
2741 __skb_fill_page_desc(to, 0, page, offset, plen);
2742 get_page(page);
2743 j = 1;
2744 len -= plen;
2745 }
2746 }
2747
2748 to->truesize += len + plen;
2749 to->len += len + plen;
2750 to->data_len += len + plen;
2751
2752 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2753 skb_tx_error(from);
2754 return -ENOMEM;
2755 }
2756 skb_zerocopy_clone(to, from, GFP_ATOMIC);
2757
2758 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2759 if (!len)
2760 break;
2761 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2762 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
2763 len -= skb_shinfo(to)->frags[j].size;
2764 skb_frag_ref(to, j);
2765 j++;
2766 }
2767 skb_shinfo(to)->nr_frags = j;
2768
2769 return 0;
2770 }
2771 EXPORT_SYMBOL_GPL(skb_zerocopy);
2772
skb_copy_and_csum_dev(const struct sk_buff * skb,u8 * to)2773 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2774 {
2775 __wsum csum;
2776 long csstart;
2777
2778 if (skb->ip_summed == CHECKSUM_PARTIAL)
2779 csstart = skb_checksum_start_offset(skb);
2780 else
2781 csstart = skb_headlen(skb);
2782
2783 BUG_ON(csstart > skb_headlen(skb));
2784
2785 skb_copy_from_linear_data(skb, to, csstart);
2786
2787 csum = 0;
2788 if (csstart != skb->len)
2789 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2790 skb->len - csstart, 0);
2791
2792 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2793 long csstuff = csstart + skb->csum_offset;
2794
2795 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2796 }
2797 }
2798 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2799
2800 /**
2801 * skb_dequeue - remove from the head of the queue
2802 * @list: list to dequeue from
2803 *
2804 * Remove the head of the list. The list lock is taken so the function
2805 * may be used safely with other locking list functions. The head item is
2806 * returned or %NULL if the list is empty.
2807 */
2808
skb_dequeue(struct sk_buff_head * list)2809 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2810 {
2811 unsigned long flags;
2812 struct sk_buff *result;
2813
2814 spin_lock_irqsave(&list->lock, flags);
2815 result = __skb_dequeue(list);
2816 spin_unlock_irqrestore(&list->lock, flags);
2817 return result;
2818 }
2819 EXPORT_SYMBOL(skb_dequeue);
2820
2821 /**
2822 * skb_dequeue_tail - remove from the tail of the queue
2823 * @list: list to dequeue from
2824 *
2825 * Remove the tail of the list. The list lock is taken so the function
2826 * may be used safely with other locking list functions. The tail item is
2827 * returned or %NULL if the list is empty.
2828 */
skb_dequeue_tail(struct sk_buff_head * list)2829 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2830 {
2831 unsigned long flags;
2832 struct sk_buff *result;
2833
2834 spin_lock_irqsave(&list->lock, flags);
2835 result = __skb_dequeue_tail(list);
2836 spin_unlock_irqrestore(&list->lock, flags);
2837 return result;
2838 }
2839 EXPORT_SYMBOL(skb_dequeue_tail);
2840
2841 /**
2842 * skb_queue_purge - empty a list
2843 * @list: list to empty
2844 *
2845 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2846 * the list and one reference dropped. This function takes the list
2847 * lock and is atomic with respect to other list locking functions.
2848 */
skb_queue_purge(struct sk_buff_head * list)2849 void skb_queue_purge(struct sk_buff_head *list)
2850 {
2851 struct sk_buff *skb;
2852 while ((skb = skb_dequeue(list)) != NULL)
2853 kfree_skb(skb);
2854 }
2855 EXPORT_SYMBOL(skb_queue_purge);
2856
2857 /**
2858 * skb_rbtree_purge - empty a skb rbtree
2859 * @root: root of the rbtree to empty
2860 * Return value: the sum of truesizes of all purged skbs.
2861 *
2862 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
2863 * the list and one reference dropped. This function does not take
2864 * any lock. Synchronization should be handled by the caller (e.g., TCP
2865 * out-of-order queue is protected by the socket lock).
2866 */
skb_rbtree_purge(struct rb_root * root)2867 unsigned int skb_rbtree_purge(struct rb_root *root)
2868 {
2869 struct rb_node *p = rb_first(root);
2870 unsigned int sum = 0;
2871
2872 while (p) {
2873 struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
2874
2875 p = rb_next(p);
2876 rb_erase(&skb->rbnode, root);
2877 sum += skb->truesize;
2878 kfree_skb(skb);
2879 }
2880 return sum;
2881 }
2882
2883 /**
2884 * skb_queue_head - queue a buffer at the list head
2885 * @list: list to use
2886 * @newsk: buffer to queue
2887 *
2888 * Queue a buffer at the start of the list. This function takes the
2889 * list lock and can be used safely with other locking &sk_buff functions
2890 * safely.
2891 *
2892 * A buffer cannot be placed on two lists at the same time.
2893 */
skb_queue_head(struct sk_buff_head * list,struct sk_buff * newsk)2894 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2895 {
2896 unsigned long flags;
2897
2898 spin_lock_irqsave(&list->lock, flags);
2899 __skb_queue_head(list, newsk);
2900 spin_unlock_irqrestore(&list->lock, flags);
2901 }
2902 EXPORT_SYMBOL(skb_queue_head);
2903
2904 /**
2905 * skb_queue_tail - queue a buffer at the list tail
2906 * @list: list to use
2907 * @newsk: buffer to queue
2908 *
2909 * Queue a buffer at the tail of the list. This function takes the
2910 * list lock and can be used safely with other locking &sk_buff functions
2911 * safely.
2912 *
2913 * A buffer cannot be placed on two lists at the same time.
2914 */
skb_queue_tail(struct sk_buff_head * list,struct sk_buff * newsk)2915 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2916 {
2917 unsigned long flags;
2918
2919 spin_lock_irqsave(&list->lock, flags);
2920 __skb_queue_tail(list, newsk);
2921 spin_unlock_irqrestore(&list->lock, flags);
2922 }
2923 EXPORT_SYMBOL(skb_queue_tail);
2924
2925 /**
2926 * skb_unlink - remove a buffer from a list
2927 * @skb: buffer to remove
2928 * @list: list to use
2929 *
2930 * Remove a packet from a list. The list locks are taken and this
2931 * function is atomic with respect to other list locked calls
2932 *
2933 * You must know what list the SKB is on.
2934 */
skb_unlink(struct sk_buff * skb,struct sk_buff_head * list)2935 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2936 {
2937 unsigned long flags;
2938
2939 spin_lock_irqsave(&list->lock, flags);
2940 __skb_unlink(skb, list);
2941 spin_unlock_irqrestore(&list->lock, flags);
2942 }
2943 EXPORT_SYMBOL(skb_unlink);
2944
2945 /**
2946 * skb_append - append a buffer
2947 * @old: buffer to insert after
2948 * @newsk: buffer to insert
2949 * @list: list to use
2950 *
2951 * Place a packet after a given packet in a list. The list locks are taken
2952 * and this function is atomic with respect to other list locked calls.
2953 * A buffer cannot be placed on two lists at the same time.
2954 */
skb_append(struct sk_buff * old,struct sk_buff * newsk,struct sk_buff_head * list)2955 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2956 {
2957 unsigned long flags;
2958
2959 spin_lock_irqsave(&list->lock, flags);
2960 __skb_queue_after(list, old, newsk);
2961 spin_unlock_irqrestore(&list->lock, flags);
2962 }
2963 EXPORT_SYMBOL(skb_append);
2964
2965 /**
2966 * skb_insert - insert a buffer
2967 * @old: buffer to insert before
2968 * @newsk: buffer to insert
2969 * @list: list to use
2970 *
2971 * Place a packet before a given packet in a list. The list locks are
2972 * taken and this function is atomic with respect to other list locked
2973 * calls.
2974 *
2975 * A buffer cannot be placed on two lists at the same time.
2976 */
skb_insert(struct sk_buff * old,struct sk_buff * newsk,struct sk_buff_head * list)2977 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2978 {
2979 unsigned long flags;
2980
2981 spin_lock_irqsave(&list->lock, flags);
2982 __skb_insert(newsk, old->prev, old, list);
2983 spin_unlock_irqrestore(&list->lock, flags);
2984 }
2985 EXPORT_SYMBOL(skb_insert);
2986
skb_split_inside_header(struct sk_buff * skb,struct sk_buff * skb1,const u32 len,const int pos)2987 static inline void skb_split_inside_header(struct sk_buff *skb,
2988 struct sk_buff* skb1,
2989 const u32 len, const int pos)
2990 {
2991 int i;
2992
2993 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2994 pos - len);
2995 /* And move data appendix as is. */
2996 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2997 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2998
2999 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
3000 skb_shinfo(skb)->nr_frags = 0;
3001 skb1->data_len = skb->data_len;
3002 skb1->len += skb1->data_len;
3003 skb->data_len = 0;
3004 skb->len = len;
3005 skb_set_tail_pointer(skb, len);
3006 }
3007
skb_split_no_header(struct sk_buff * skb,struct sk_buff * skb1,const u32 len,int pos)3008 static inline void skb_split_no_header(struct sk_buff *skb,
3009 struct sk_buff* skb1,
3010 const u32 len, int pos)
3011 {
3012 int i, k = 0;
3013 const int nfrags = skb_shinfo(skb)->nr_frags;
3014
3015 skb_shinfo(skb)->nr_frags = 0;
3016 skb1->len = skb1->data_len = skb->len - len;
3017 skb->len = len;
3018 skb->data_len = len - pos;
3019
3020 for (i = 0; i < nfrags; i++) {
3021 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
3022
3023 if (pos + size > len) {
3024 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
3025
3026 if (pos < len) {
3027 /* Split frag.
3028 * We have two variants in this case:
3029 * 1. Move all the frag to the second
3030 * part, if it is possible. F.e.
3031 * this approach is mandatory for TUX,
3032 * where splitting is expensive.
3033 * 2. Split is accurately. We make this.
3034 */
3035 skb_frag_ref(skb, i);
3036 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
3037 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
3038 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
3039 skb_shinfo(skb)->nr_frags++;
3040 }
3041 k++;
3042 } else
3043 skb_shinfo(skb)->nr_frags++;
3044 pos += size;
3045 }
3046 skb_shinfo(skb1)->nr_frags = k;
3047 }
3048
3049 /**
3050 * skb_split - Split fragmented skb to two parts at length len.
3051 * @skb: the buffer to split
3052 * @skb1: the buffer to receive the second part
3053 * @len: new length for skb
3054 */
skb_split(struct sk_buff * skb,struct sk_buff * skb1,const u32 len)3055 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
3056 {
3057 int pos = skb_headlen(skb);
3058
3059 skb_shinfo(skb1)->tx_flags |= skb_shinfo(skb)->tx_flags &
3060 SKBTX_SHARED_FRAG;
3061 skb_zerocopy_clone(skb1, skb, 0);
3062 if (len < pos) /* Split line is inside header. */
3063 skb_split_inside_header(skb, skb1, len, pos);
3064 else /* Second chunk has no header, nothing to copy. */
3065 skb_split_no_header(skb, skb1, len, pos);
3066 }
3067 EXPORT_SYMBOL(skb_split);
3068
3069 /* Shifting from/to a cloned skb is a no-go.
3070 *
3071 * Caller cannot keep skb_shinfo related pointers past calling here!
3072 */
skb_prepare_for_shift(struct sk_buff * skb)3073 static int skb_prepare_for_shift(struct sk_buff *skb)
3074 {
3075 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3076 }
3077
3078 /**
3079 * skb_shift - Shifts paged data partially from skb to another
3080 * @tgt: buffer into which tail data gets added
3081 * @skb: buffer from which the paged data comes from
3082 * @shiftlen: shift up to this many bytes
3083 *
3084 * Attempts to shift up to shiftlen worth of bytes, which may be less than
3085 * the length of the skb, from skb to tgt. Returns number bytes shifted.
3086 * It's up to caller to free skb if everything was shifted.
3087 *
3088 * If @tgt runs out of frags, the whole operation is aborted.
3089 *
3090 * Skb cannot include anything else but paged data while tgt is allowed
3091 * to have non-paged data as well.
3092 *
3093 * TODO: full sized shift could be optimized but that would need
3094 * specialized skb free'er to handle frags without up-to-date nr_frags.
3095 */
skb_shift(struct sk_buff * tgt,struct sk_buff * skb,int shiftlen)3096 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
3097 {
3098 int from, to, merge, todo;
3099 struct skb_frag_struct *fragfrom, *fragto;
3100
3101 BUG_ON(shiftlen > skb->len);
3102
3103 if (skb_headlen(skb))
3104 return 0;
3105 if (skb_zcopy(tgt) || skb_zcopy(skb))
3106 return 0;
3107
3108 todo = shiftlen;
3109 from = 0;
3110 to = skb_shinfo(tgt)->nr_frags;
3111 fragfrom = &skb_shinfo(skb)->frags[from];
3112
3113 /* Actual merge is delayed until the point when we know we can
3114 * commit all, so that we don't have to undo partial changes
3115 */
3116 if (!to ||
3117 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
3118 fragfrom->page_offset)) {
3119 merge = -1;
3120 } else {
3121 merge = to - 1;
3122
3123 todo -= skb_frag_size(fragfrom);
3124 if (todo < 0) {
3125 if (skb_prepare_for_shift(skb) ||
3126 skb_prepare_for_shift(tgt))
3127 return 0;
3128
3129 /* All previous frag pointers might be stale! */
3130 fragfrom = &skb_shinfo(skb)->frags[from];
3131 fragto = &skb_shinfo(tgt)->frags[merge];
3132
3133 skb_frag_size_add(fragto, shiftlen);
3134 skb_frag_size_sub(fragfrom, shiftlen);
3135 fragfrom->page_offset += shiftlen;
3136
3137 goto onlymerged;
3138 }
3139
3140 from++;
3141 }
3142
3143 /* Skip full, not-fitting skb to avoid expensive operations */
3144 if ((shiftlen == skb->len) &&
3145 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
3146 return 0;
3147
3148 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
3149 return 0;
3150
3151 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
3152 if (to == MAX_SKB_FRAGS)
3153 return 0;
3154
3155 fragfrom = &skb_shinfo(skb)->frags[from];
3156 fragto = &skb_shinfo(tgt)->frags[to];
3157
3158 if (todo >= skb_frag_size(fragfrom)) {
3159 *fragto = *fragfrom;
3160 todo -= skb_frag_size(fragfrom);
3161 from++;
3162 to++;
3163
3164 } else {
3165 __skb_frag_ref(fragfrom);
3166 fragto->page = fragfrom->page;
3167 fragto->page_offset = fragfrom->page_offset;
3168 skb_frag_size_set(fragto, todo);
3169
3170 fragfrom->page_offset += todo;
3171 skb_frag_size_sub(fragfrom, todo);
3172 todo = 0;
3173
3174 to++;
3175 break;
3176 }
3177 }
3178
3179 /* Ready to "commit" this state change to tgt */
3180 skb_shinfo(tgt)->nr_frags = to;
3181
3182 if (merge >= 0) {
3183 fragfrom = &skb_shinfo(skb)->frags[0];
3184 fragto = &skb_shinfo(tgt)->frags[merge];
3185
3186 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
3187 __skb_frag_unref(fragfrom);
3188 }
3189
3190 /* Reposition in the original skb */
3191 to = 0;
3192 while (from < skb_shinfo(skb)->nr_frags)
3193 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
3194 skb_shinfo(skb)->nr_frags = to;
3195
3196 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
3197
3198 onlymerged:
3199 /* Most likely the tgt won't ever need its checksum anymore, skb on
3200 * the other hand might need it if it needs to be resent
3201 */
3202 tgt->ip_summed = CHECKSUM_PARTIAL;
3203 skb->ip_summed = CHECKSUM_PARTIAL;
3204
3205 /* Yak, is it really working this way? Some helper please? */
3206 skb->len -= shiftlen;
3207 skb->data_len -= shiftlen;
3208 skb->truesize -= shiftlen;
3209 tgt->len += shiftlen;
3210 tgt->data_len += shiftlen;
3211 tgt->truesize += shiftlen;
3212
3213 return shiftlen;
3214 }
3215
3216 /**
3217 * skb_prepare_seq_read - Prepare a sequential read of skb data
3218 * @skb: the buffer to read
3219 * @from: lower offset of data to be read
3220 * @to: upper offset of data to be read
3221 * @st: state variable
3222 *
3223 * Initializes the specified state variable. Must be called before
3224 * invoking skb_seq_read() for the first time.
3225 */
skb_prepare_seq_read(struct sk_buff * skb,unsigned int from,unsigned int to,struct skb_seq_state * st)3226 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
3227 unsigned int to, struct skb_seq_state *st)
3228 {
3229 st->lower_offset = from;
3230 st->upper_offset = to;
3231 st->root_skb = st->cur_skb = skb;
3232 st->frag_idx = st->stepped_offset = 0;
3233 st->frag_data = NULL;
3234 }
3235 EXPORT_SYMBOL(skb_prepare_seq_read);
3236
3237 /**
3238 * skb_seq_read - Sequentially read skb data
3239 * @consumed: number of bytes consumed by the caller so far
3240 * @data: destination pointer for data to be returned
3241 * @st: state variable
3242 *
3243 * Reads a block of skb data at @consumed relative to the
3244 * lower offset specified to skb_prepare_seq_read(). Assigns
3245 * the head of the data block to @data and returns the length
3246 * of the block or 0 if the end of the skb data or the upper
3247 * offset has been reached.
3248 *
3249 * The caller is not required to consume all of the data
3250 * returned, i.e. @consumed is typically set to the number
3251 * of bytes already consumed and the next call to
3252 * skb_seq_read() will return the remaining part of the block.
3253 *
3254 * Note 1: The size of each block of data returned can be arbitrary,
3255 * this limitation is the cost for zerocopy sequential
3256 * reads of potentially non linear data.
3257 *
3258 * Note 2: Fragment lists within fragments are not implemented
3259 * at the moment, state->root_skb could be replaced with
3260 * a stack for this purpose.
3261 */
skb_seq_read(unsigned int consumed,const u8 ** data,struct skb_seq_state * st)3262 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
3263 struct skb_seq_state *st)
3264 {
3265 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
3266 skb_frag_t *frag;
3267
3268 if (unlikely(abs_offset >= st->upper_offset)) {
3269 if (st->frag_data) {
3270 kunmap_atomic(st->frag_data);
3271 st->frag_data = NULL;
3272 }
3273 return 0;
3274 }
3275
3276 next_skb:
3277 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
3278
3279 if (abs_offset < block_limit && !st->frag_data) {
3280 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
3281 return block_limit - abs_offset;
3282 }
3283
3284 if (st->frag_idx == 0 && !st->frag_data)
3285 st->stepped_offset += skb_headlen(st->cur_skb);
3286
3287 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
3288 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
3289 block_limit = skb_frag_size(frag) + st->stepped_offset;
3290
3291 if (abs_offset < block_limit) {
3292 if (!st->frag_data)
3293 st->frag_data = kmap_atomic(skb_frag_page(frag));
3294
3295 *data = (u8 *) st->frag_data + frag->page_offset +
3296 (abs_offset - st->stepped_offset);
3297
3298 return block_limit - abs_offset;
3299 }
3300
3301 if (st->frag_data) {
3302 kunmap_atomic(st->frag_data);
3303 st->frag_data = NULL;
3304 }
3305
3306 st->frag_idx++;
3307 st->stepped_offset += skb_frag_size(frag);
3308 }
3309
3310 if (st->frag_data) {
3311 kunmap_atomic(st->frag_data);
3312 st->frag_data = NULL;
3313 }
3314
3315 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
3316 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
3317 st->frag_idx = 0;
3318 goto next_skb;
3319 } else if (st->cur_skb->next) {
3320 st->cur_skb = st->cur_skb->next;
3321 st->frag_idx = 0;
3322 goto next_skb;
3323 }
3324
3325 return 0;
3326 }
3327 EXPORT_SYMBOL(skb_seq_read);
3328
3329 /**
3330 * skb_abort_seq_read - Abort a sequential read of skb data
3331 * @st: state variable
3332 *
3333 * Must be called if skb_seq_read() was not called until it
3334 * returned 0.
3335 */
skb_abort_seq_read(struct skb_seq_state * st)3336 void skb_abort_seq_read(struct skb_seq_state *st)
3337 {
3338 if (st->frag_data)
3339 kunmap_atomic(st->frag_data);
3340 }
3341 EXPORT_SYMBOL(skb_abort_seq_read);
3342
3343 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
3344
skb_ts_get_next_block(unsigned int offset,const u8 ** text,struct ts_config * conf,struct ts_state * state)3345 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
3346 struct ts_config *conf,
3347 struct ts_state *state)
3348 {
3349 return skb_seq_read(offset, text, TS_SKB_CB(state));
3350 }
3351
skb_ts_finish(struct ts_config * conf,struct ts_state * state)3352 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
3353 {
3354 skb_abort_seq_read(TS_SKB_CB(state));
3355 }
3356
3357 /**
3358 * skb_find_text - Find a text pattern in skb data
3359 * @skb: the buffer to look in
3360 * @from: search offset
3361 * @to: search limit
3362 * @config: textsearch configuration
3363 *
3364 * Finds a pattern in the skb data according to the specified
3365 * textsearch configuration. Use textsearch_next() to retrieve
3366 * subsequent occurrences of the pattern. Returns the offset
3367 * to the first occurrence or UINT_MAX if no match was found.
3368 */
skb_find_text(struct sk_buff * skb,unsigned int from,unsigned int to,struct ts_config * config)3369 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
3370 unsigned int to, struct ts_config *config)
3371 {
3372 struct ts_state state;
3373 unsigned int ret;
3374
3375 config->get_next_block = skb_ts_get_next_block;
3376 config->finish = skb_ts_finish;
3377
3378 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
3379
3380 ret = textsearch_find(config, &state);
3381 return (ret <= to - from ? ret : UINT_MAX);
3382 }
3383 EXPORT_SYMBOL(skb_find_text);
3384
3385 /**
3386 * skb_append_datato_frags - append the user data to a skb
3387 * @sk: sock structure
3388 * @skb: skb structure to be appended with user data.
3389 * @getfrag: call back function to be used for getting the user data
3390 * @from: pointer to user message iov
3391 * @length: length of the iov message
3392 *
3393 * Description: This procedure append the user data in the fragment part
3394 * of the skb if any page alloc fails user this procedure returns -ENOMEM
3395 */
skb_append_datato_frags(struct sock * sk,struct sk_buff * skb,int (* getfrag)(void * from,char * to,int offset,int len,int odd,struct sk_buff * skb),void * from,int length)3396 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
3397 int (*getfrag)(void *from, char *to, int offset,
3398 int len, int odd, struct sk_buff *skb),
3399 void *from, int length)
3400 {
3401 int frg_cnt = skb_shinfo(skb)->nr_frags;
3402 int copy;
3403 int offset = 0;
3404 int ret;
3405 struct page_frag *pfrag = ¤t->task_frag;
3406
3407 do {
3408 /* Return error if we don't have space for new frag */
3409 if (frg_cnt >= MAX_SKB_FRAGS)
3410 return -EMSGSIZE;
3411
3412 if (!sk_page_frag_refill(sk, pfrag))
3413 return -ENOMEM;
3414
3415 /* copy the user data to page */
3416 copy = min_t(int, length, pfrag->size - pfrag->offset);
3417
3418 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
3419 offset, copy, 0, skb);
3420 if (ret < 0)
3421 return -EFAULT;
3422
3423 /* copy was successful so update the size parameters */
3424 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
3425 copy);
3426 frg_cnt++;
3427 pfrag->offset += copy;
3428 get_page(pfrag->page);
3429
3430 skb->truesize += copy;
3431 refcount_add(copy, &sk->sk_wmem_alloc);
3432 skb->len += copy;
3433 skb->data_len += copy;
3434 offset += copy;
3435 length -= copy;
3436
3437 } while (length > 0);
3438
3439 return 0;
3440 }
3441 EXPORT_SYMBOL(skb_append_datato_frags);
3442
skb_append_pagefrags(struct sk_buff * skb,struct page * page,int offset,size_t size)3443 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3444 int offset, size_t size)
3445 {
3446 int i = skb_shinfo(skb)->nr_frags;
3447
3448 if (skb_can_coalesce(skb, i, page, offset)) {
3449 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3450 } else if (i < MAX_SKB_FRAGS) {
3451 get_page(page);
3452 skb_fill_page_desc(skb, i, page, offset, size);
3453 } else {
3454 return -EMSGSIZE;
3455 }
3456
3457 return 0;
3458 }
3459 EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3460
3461 /**
3462 * skb_pull_rcsum - pull skb and update receive checksum
3463 * @skb: buffer to update
3464 * @len: length of data pulled
3465 *
3466 * This function performs an skb_pull on the packet and updates
3467 * the CHECKSUM_COMPLETE checksum. It should be used on
3468 * receive path processing instead of skb_pull unless you know
3469 * that the checksum difference is zero (e.g., a valid IP header)
3470 * or you are setting ip_summed to CHECKSUM_NONE.
3471 */
skb_pull_rcsum(struct sk_buff * skb,unsigned int len)3472 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3473 {
3474 unsigned char *data = skb->data;
3475
3476 BUG_ON(len > skb->len);
3477 __skb_pull(skb, len);
3478 skb_postpull_rcsum(skb, data, len);
3479 return skb->data;
3480 }
3481 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3482
skb_head_frag_to_page_desc(struct sk_buff * frag_skb)3483 static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb)
3484 {
3485 skb_frag_t head_frag;
3486 struct page *page;
3487
3488 page = virt_to_head_page(frag_skb->head);
3489 head_frag.page.p = page;
3490 head_frag.page_offset = frag_skb->data -
3491 (unsigned char *)page_address(page);
3492 head_frag.size = skb_headlen(frag_skb);
3493 return head_frag;
3494 }
3495
3496 /**
3497 * skb_segment - Perform protocol segmentation on skb.
3498 * @head_skb: buffer to segment
3499 * @features: features for the output path (see dev->features)
3500 *
3501 * This function performs segmentation on the given skb. It returns
3502 * a pointer to the first in a list of new skbs for the segments.
3503 * In case of error it returns ERR_PTR(err).
3504 */
skb_segment(struct sk_buff * head_skb,netdev_features_t features)3505 struct sk_buff *skb_segment(struct sk_buff *head_skb,
3506 netdev_features_t features)
3507 {
3508 struct sk_buff *segs = NULL;
3509 struct sk_buff *tail = NULL;
3510 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3511 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3512 unsigned int mss = skb_shinfo(head_skb)->gso_size;
3513 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3514 struct sk_buff *frag_skb = head_skb;
3515 unsigned int offset = doffset;
3516 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3517 unsigned int partial_segs = 0;
3518 unsigned int headroom;
3519 unsigned int len = head_skb->len;
3520 __be16 proto;
3521 bool csum, sg;
3522 int nfrags = skb_shinfo(head_skb)->nr_frags;
3523 int err = -ENOMEM;
3524 int i = 0;
3525 int pos;
3526 int dummy;
3527
3528 __skb_push(head_skb, doffset);
3529 proto = skb_network_protocol(head_skb, &dummy);
3530 if (unlikely(!proto))
3531 return ERR_PTR(-EINVAL);
3532
3533 sg = !!(features & NETIF_F_SG);
3534 csum = !!can_checksum_protocol(features, proto);
3535
3536 if (sg && csum && (mss != GSO_BY_FRAGS)) {
3537 if (!(features & NETIF_F_GSO_PARTIAL)) {
3538 struct sk_buff *iter;
3539 unsigned int frag_len;
3540
3541 if (!list_skb ||
3542 !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
3543 goto normal;
3544
3545 /* If we get here then all the required
3546 * GSO features except frag_list are supported.
3547 * Try to split the SKB to multiple GSO SKBs
3548 * with no frag_list.
3549 * Currently we can do that only when the buffers don't
3550 * have a linear part and all the buffers except
3551 * the last are of the same length.
3552 */
3553 frag_len = list_skb->len;
3554 skb_walk_frags(head_skb, iter) {
3555 if (frag_len != iter->len && iter->next)
3556 goto normal;
3557 if (skb_headlen(iter) && !iter->head_frag)
3558 goto normal;
3559
3560 len -= iter->len;
3561 }
3562
3563 if (len != frag_len)
3564 goto normal;
3565 }
3566
3567 /* GSO partial only requires that we trim off any excess that
3568 * doesn't fit into an MSS sized block, so take care of that
3569 * now.
3570 */
3571 partial_segs = len / mss;
3572 if (partial_segs > 1)
3573 mss *= partial_segs;
3574 else
3575 partial_segs = 0;
3576 }
3577
3578 normal:
3579 headroom = skb_headroom(head_skb);
3580 pos = skb_headlen(head_skb);
3581
3582 do {
3583 struct sk_buff *nskb;
3584 skb_frag_t *nskb_frag;
3585 int hsize;
3586 int size;
3587
3588 if (unlikely(mss == GSO_BY_FRAGS)) {
3589 len = list_skb->len;
3590 } else {
3591 len = head_skb->len - offset;
3592 if (len > mss)
3593 len = mss;
3594 }
3595
3596 hsize = skb_headlen(head_skb) - offset;
3597 if (hsize < 0)
3598 hsize = 0;
3599 if (hsize > len || !sg)
3600 hsize = len;
3601
3602 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3603 (skb_headlen(list_skb) == len || sg)) {
3604 BUG_ON(skb_headlen(list_skb) > len);
3605
3606 i = 0;
3607 nfrags = skb_shinfo(list_skb)->nr_frags;
3608 frag = skb_shinfo(list_skb)->frags;
3609 frag_skb = list_skb;
3610 pos += skb_headlen(list_skb);
3611
3612 while (pos < offset + len) {
3613 BUG_ON(i >= nfrags);
3614
3615 size = skb_frag_size(frag);
3616 if (pos + size > offset + len)
3617 break;
3618
3619 i++;
3620 pos += size;
3621 frag++;
3622 }
3623
3624 nskb = skb_clone(list_skb, GFP_ATOMIC);
3625 list_skb = list_skb->next;
3626
3627 if (unlikely(!nskb))
3628 goto err;
3629
3630 if (unlikely(pskb_trim(nskb, len))) {
3631 kfree_skb(nskb);
3632 goto err;
3633 }
3634
3635 hsize = skb_end_offset(nskb);
3636 if (skb_cow_head(nskb, doffset + headroom)) {
3637 kfree_skb(nskb);
3638 goto err;
3639 }
3640
3641 nskb->truesize += skb_end_offset(nskb) - hsize;
3642 skb_release_head_state(nskb);
3643 __skb_push(nskb, doffset);
3644 } else {
3645 nskb = __alloc_skb(hsize + doffset + headroom,
3646 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3647 NUMA_NO_NODE);
3648
3649 if (unlikely(!nskb))
3650 goto err;
3651
3652 skb_reserve(nskb, headroom);
3653 __skb_put(nskb, doffset);
3654 }
3655
3656 if (segs)
3657 tail->next = nskb;
3658 else
3659 segs = nskb;
3660 tail = nskb;
3661
3662 __copy_skb_header(nskb, head_skb);
3663
3664 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3665 skb_reset_mac_len(nskb);
3666
3667 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3668 nskb->data - tnl_hlen,
3669 doffset + tnl_hlen);
3670
3671 if (nskb->len == len + doffset)
3672 goto perform_csum_check;
3673
3674 if (!sg) {
3675 if (!nskb->remcsum_offload)
3676 nskb->ip_summed = CHECKSUM_NONE;
3677 SKB_GSO_CB(nskb)->csum =
3678 skb_copy_and_csum_bits(head_skb, offset,
3679 skb_put(nskb, len),
3680 len, 0);
3681 SKB_GSO_CB(nskb)->csum_start =
3682 skb_headroom(nskb) + doffset;
3683 continue;
3684 }
3685
3686 nskb_frag = skb_shinfo(nskb)->frags;
3687
3688 skb_copy_from_linear_data_offset(head_skb, offset,
3689 skb_put(nskb, hsize), hsize);
3690
3691 skb_shinfo(nskb)->tx_flags |= skb_shinfo(head_skb)->tx_flags &
3692 SKBTX_SHARED_FRAG;
3693
3694 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
3695 skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
3696 goto err;
3697
3698 while (pos < offset + len) {
3699 if (i >= nfrags) {
3700 i = 0;
3701 nfrags = skb_shinfo(list_skb)->nr_frags;
3702 frag = skb_shinfo(list_skb)->frags;
3703 frag_skb = list_skb;
3704 if (!skb_headlen(list_skb)) {
3705 BUG_ON(!nfrags);
3706 } else {
3707 BUG_ON(!list_skb->head_frag);
3708
3709 /* to make room for head_frag. */
3710 i--;
3711 frag--;
3712 }
3713 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
3714 skb_zerocopy_clone(nskb, frag_skb,
3715 GFP_ATOMIC))
3716 goto err;
3717
3718 list_skb = list_skb->next;
3719 }
3720
3721 if (unlikely(skb_shinfo(nskb)->nr_frags >=
3722 MAX_SKB_FRAGS)) {
3723 net_warn_ratelimited(
3724 "skb_segment: too many frags: %u %u\n",
3725 pos, mss);
3726 err = -EINVAL;
3727 goto err;
3728 }
3729
3730 *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag;
3731 __skb_frag_ref(nskb_frag);
3732 size = skb_frag_size(nskb_frag);
3733
3734 if (pos < offset) {
3735 nskb_frag->page_offset += offset - pos;
3736 skb_frag_size_sub(nskb_frag, offset - pos);
3737 }
3738
3739 skb_shinfo(nskb)->nr_frags++;
3740
3741 if (pos + size <= offset + len) {
3742 i++;
3743 frag++;
3744 pos += size;
3745 } else {
3746 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
3747 goto skip_fraglist;
3748 }
3749
3750 nskb_frag++;
3751 }
3752
3753 skip_fraglist:
3754 nskb->data_len = len - hsize;
3755 nskb->len += nskb->data_len;
3756 nskb->truesize += nskb->data_len;
3757
3758 perform_csum_check:
3759 if (!csum) {
3760 if (skb_has_shared_frag(nskb) &&
3761 __skb_linearize(nskb))
3762 goto err;
3763
3764 if (!nskb->remcsum_offload)
3765 nskb->ip_summed = CHECKSUM_NONE;
3766 SKB_GSO_CB(nskb)->csum =
3767 skb_checksum(nskb, doffset,
3768 nskb->len - doffset, 0);
3769 SKB_GSO_CB(nskb)->csum_start =
3770 skb_headroom(nskb) + doffset;
3771 }
3772 } while ((offset += len) < head_skb->len);
3773
3774 /* Some callers want to get the end of the list.
3775 * Put it in segs->prev to avoid walking the list.
3776 * (see validate_xmit_skb_list() for example)
3777 */
3778 segs->prev = tail;
3779
3780 if (partial_segs) {
3781 struct sk_buff *iter;
3782 int type = skb_shinfo(head_skb)->gso_type;
3783 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
3784
3785 /* Update type to add partial and then remove dodgy if set */
3786 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
3787 type &= ~SKB_GSO_DODGY;
3788
3789 /* Update GSO info and prepare to start updating headers on
3790 * our way back down the stack of protocols.
3791 */
3792 for (iter = segs; iter; iter = iter->next) {
3793 skb_shinfo(iter)->gso_size = gso_size;
3794 skb_shinfo(iter)->gso_segs = partial_segs;
3795 skb_shinfo(iter)->gso_type = type;
3796 SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
3797 }
3798
3799 if (tail->len - doffset <= gso_size)
3800 skb_shinfo(tail)->gso_size = 0;
3801 else if (tail != segs)
3802 skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
3803 }
3804
3805 /* Following permits correct backpressure, for protocols
3806 * using skb_set_owner_w().
3807 * Idea is to tranfert ownership from head_skb to last segment.
3808 */
3809 if (head_skb->destructor == sock_wfree) {
3810 swap(tail->truesize, head_skb->truesize);
3811 swap(tail->destructor, head_skb->destructor);
3812 swap(tail->sk, head_skb->sk);
3813 }
3814 return segs;
3815
3816 err:
3817 kfree_skb_list(segs);
3818 return ERR_PTR(err);
3819 }
3820 EXPORT_SYMBOL_GPL(skb_segment);
3821
skb_gro_receive(struct sk_buff * p,struct sk_buff * skb)3822 int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb)
3823 {
3824 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
3825 unsigned int offset = skb_gro_offset(skb);
3826 unsigned int headlen = skb_headlen(skb);
3827 unsigned int len = skb_gro_len(skb);
3828 unsigned int delta_truesize;
3829 struct sk_buff *lp;
3830
3831 if (unlikely(p->len + len >= 65536))
3832 return -E2BIG;
3833
3834 lp = NAPI_GRO_CB(p)->last;
3835 pinfo = skb_shinfo(lp);
3836
3837 if (headlen <= offset) {
3838 skb_frag_t *frag;
3839 skb_frag_t *frag2;
3840 int i = skbinfo->nr_frags;
3841 int nr_frags = pinfo->nr_frags + i;
3842
3843 if (nr_frags > MAX_SKB_FRAGS)
3844 goto merge;
3845
3846 offset -= headlen;
3847 pinfo->nr_frags = nr_frags;
3848 skbinfo->nr_frags = 0;
3849
3850 frag = pinfo->frags + nr_frags;
3851 frag2 = skbinfo->frags + i;
3852 do {
3853 *--frag = *--frag2;
3854 } while (--i);
3855
3856 frag->page_offset += offset;
3857 skb_frag_size_sub(frag, offset);
3858
3859 /* all fragments truesize : remove (head size + sk_buff) */
3860 delta_truesize = skb->truesize -
3861 SKB_TRUESIZE(skb_end_offset(skb));
3862
3863 skb->truesize -= skb->data_len;
3864 skb->len -= skb->data_len;
3865 skb->data_len = 0;
3866
3867 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
3868 goto done;
3869 } else if (skb->head_frag) {
3870 int nr_frags = pinfo->nr_frags;
3871 skb_frag_t *frag = pinfo->frags + nr_frags;
3872 struct page *page = virt_to_head_page(skb->head);
3873 unsigned int first_size = headlen - offset;
3874 unsigned int first_offset;
3875
3876 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
3877 goto merge;
3878
3879 first_offset = skb->data -
3880 (unsigned char *)page_address(page) +
3881 offset;
3882
3883 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3884
3885 frag->page.p = page;
3886 frag->page_offset = first_offset;
3887 skb_frag_size_set(frag, first_size);
3888
3889 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3890 /* We dont need to clear skbinfo->nr_frags here */
3891
3892 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3893 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3894 goto done;
3895 }
3896
3897 merge:
3898 delta_truesize = skb->truesize;
3899 if (offset > headlen) {
3900 unsigned int eat = offset - headlen;
3901
3902 skbinfo->frags[0].page_offset += eat;
3903 skb_frag_size_sub(&skbinfo->frags[0], eat);
3904 skb->data_len -= eat;
3905 skb->len -= eat;
3906 offset = headlen;
3907 }
3908
3909 __skb_pull(skb, offset);
3910
3911 if (NAPI_GRO_CB(p)->last == p)
3912 skb_shinfo(p)->frag_list = skb;
3913 else
3914 NAPI_GRO_CB(p)->last->next = skb;
3915 NAPI_GRO_CB(p)->last = skb;
3916 __skb_header_release(skb);
3917 lp = p;
3918
3919 done:
3920 NAPI_GRO_CB(p)->count++;
3921 p->data_len += len;
3922 p->truesize += delta_truesize;
3923 p->len += len;
3924 if (lp != p) {
3925 lp->data_len += len;
3926 lp->truesize += delta_truesize;
3927 lp->len += len;
3928 }
3929 NAPI_GRO_CB(skb)->same_flow = 1;
3930 return 0;
3931 }
3932 EXPORT_SYMBOL_GPL(skb_gro_receive);
3933
skb_init(void)3934 void __init skb_init(void)
3935 {
3936 skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache",
3937 sizeof(struct sk_buff),
3938 0,
3939 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3940 offsetof(struct sk_buff, cb),
3941 sizeof_field(struct sk_buff, cb),
3942 NULL);
3943 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3944 sizeof(struct sk_buff_fclones),
3945 0,
3946 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3947 NULL);
3948 }
3949
3950 static int
__skb_to_sgvec(struct sk_buff * skb,struct scatterlist * sg,int offset,int len,unsigned int recursion_level)3951 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
3952 unsigned int recursion_level)
3953 {
3954 int start = skb_headlen(skb);
3955 int i, copy = start - offset;
3956 struct sk_buff *frag_iter;
3957 int elt = 0;
3958
3959 if (unlikely(recursion_level >= 24))
3960 return -EMSGSIZE;
3961
3962 if (copy > 0) {
3963 if (copy > len)
3964 copy = len;
3965 sg_set_buf(sg, skb->data + offset, copy);
3966 elt++;
3967 if ((len -= copy) == 0)
3968 return elt;
3969 offset += copy;
3970 }
3971
3972 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3973 int end;
3974
3975 WARN_ON(start > offset + len);
3976
3977 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3978 if ((copy = end - offset) > 0) {
3979 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3980 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
3981 return -EMSGSIZE;
3982
3983 if (copy > len)
3984 copy = len;
3985 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3986 frag->page_offset+offset-start);
3987 elt++;
3988 if (!(len -= copy))
3989 return elt;
3990 offset += copy;
3991 }
3992 start = end;
3993 }
3994
3995 skb_walk_frags(skb, frag_iter) {
3996 int end, ret;
3997
3998 WARN_ON(start > offset + len);
3999
4000 end = start + frag_iter->len;
4001 if ((copy = end - offset) > 0) {
4002 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4003 return -EMSGSIZE;
4004
4005 if (copy > len)
4006 copy = len;
4007 ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
4008 copy, recursion_level + 1);
4009 if (unlikely(ret < 0))
4010 return ret;
4011 elt += ret;
4012 if ((len -= copy) == 0)
4013 return elt;
4014 offset += copy;
4015 }
4016 start = end;
4017 }
4018 BUG_ON(len);
4019 return elt;
4020 }
4021
4022 /**
4023 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
4024 * @skb: Socket buffer containing the buffers to be mapped
4025 * @sg: The scatter-gather list to map into
4026 * @offset: The offset into the buffer's contents to start mapping
4027 * @len: Length of buffer space to be mapped
4028 *
4029 * Fill the specified scatter-gather list with mappings/pointers into a
4030 * region of the buffer space attached to a socket buffer. Returns either
4031 * the number of scatterlist items used, or -EMSGSIZE if the contents
4032 * could not fit.
4033 */
skb_to_sgvec(struct sk_buff * skb,struct scatterlist * sg,int offset,int len)4034 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
4035 {
4036 int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
4037
4038 if (nsg <= 0)
4039 return nsg;
4040
4041 sg_mark_end(&sg[nsg - 1]);
4042
4043 return nsg;
4044 }
4045 EXPORT_SYMBOL_GPL(skb_to_sgvec);
4046
4047 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
4048 * sglist without mark the sg which contain last skb data as the end.
4049 * So the caller can mannipulate sg list as will when padding new data after
4050 * the first call without calling sg_unmark_end to expend sg list.
4051 *
4052 * Scenario to use skb_to_sgvec_nomark:
4053 * 1. sg_init_table
4054 * 2. skb_to_sgvec_nomark(payload1)
4055 * 3. skb_to_sgvec_nomark(payload2)
4056 *
4057 * This is equivalent to:
4058 * 1. sg_init_table
4059 * 2. skb_to_sgvec(payload1)
4060 * 3. sg_unmark_end
4061 * 4. skb_to_sgvec(payload2)
4062 *
4063 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
4064 * is more preferable.
4065 */
skb_to_sgvec_nomark(struct sk_buff * skb,struct scatterlist * sg,int offset,int len)4066 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
4067 int offset, int len)
4068 {
4069 return __skb_to_sgvec(skb, sg, offset, len, 0);
4070 }
4071 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
4072
4073
4074
4075 /**
4076 * skb_cow_data - Check that a socket buffer's data buffers are writable
4077 * @skb: The socket buffer to check.
4078 * @tailbits: Amount of trailing space to be added
4079 * @trailer: Returned pointer to the skb where the @tailbits space begins
4080 *
4081 * Make sure that the data buffers attached to a socket buffer are
4082 * writable. If they are not, private copies are made of the data buffers
4083 * and the socket buffer is set to use these instead.
4084 *
4085 * If @tailbits is given, make sure that there is space to write @tailbits
4086 * bytes of data beyond current end of socket buffer. @trailer will be
4087 * set to point to the skb in which this space begins.
4088 *
4089 * The number of scatterlist elements required to completely map the
4090 * COW'd and extended socket buffer will be returned.
4091 */
skb_cow_data(struct sk_buff * skb,int tailbits,struct sk_buff ** trailer)4092 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
4093 {
4094 int copyflag;
4095 int elt;
4096 struct sk_buff *skb1, **skb_p;
4097
4098 /* If skb is cloned or its head is paged, reallocate
4099 * head pulling out all the pages (pages are considered not writable
4100 * at the moment even if they are anonymous).
4101 */
4102 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
4103 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
4104 return -ENOMEM;
4105
4106 /* Easy case. Most of packets will go this way. */
4107 if (!skb_has_frag_list(skb)) {
4108 /* A little of trouble, not enough of space for trailer.
4109 * This should not happen, when stack is tuned to generate
4110 * good frames. OK, on miss we reallocate and reserve even more
4111 * space, 128 bytes is fair. */
4112
4113 if (skb_tailroom(skb) < tailbits &&
4114 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
4115 return -ENOMEM;
4116
4117 /* Voila! */
4118 *trailer = skb;
4119 return 1;
4120 }
4121
4122 /* Misery. We are in troubles, going to mincer fragments... */
4123
4124 elt = 1;
4125 skb_p = &skb_shinfo(skb)->frag_list;
4126 copyflag = 0;
4127
4128 while ((skb1 = *skb_p) != NULL) {
4129 int ntail = 0;
4130
4131 /* The fragment is partially pulled by someone,
4132 * this can happen on input. Copy it and everything
4133 * after it. */
4134
4135 if (skb_shared(skb1))
4136 copyflag = 1;
4137
4138 /* If the skb is the last, worry about trailer. */
4139
4140 if (skb1->next == NULL && tailbits) {
4141 if (skb_shinfo(skb1)->nr_frags ||
4142 skb_has_frag_list(skb1) ||
4143 skb_tailroom(skb1) < tailbits)
4144 ntail = tailbits + 128;
4145 }
4146
4147 if (copyflag ||
4148 skb_cloned(skb1) ||
4149 ntail ||
4150 skb_shinfo(skb1)->nr_frags ||
4151 skb_has_frag_list(skb1)) {
4152 struct sk_buff *skb2;
4153
4154 /* Fuck, we are miserable poor guys... */
4155 if (ntail == 0)
4156 skb2 = skb_copy(skb1, GFP_ATOMIC);
4157 else
4158 skb2 = skb_copy_expand(skb1,
4159 skb_headroom(skb1),
4160 ntail,
4161 GFP_ATOMIC);
4162 if (unlikely(skb2 == NULL))
4163 return -ENOMEM;
4164
4165 if (skb1->sk)
4166 skb_set_owner_w(skb2, skb1->sk);
4167
4168 /* Looking around. Are we still alive?
4169 * OK, link new skb, drop old one */
4170
4171 skb2->next = skb1->next;
4172 *skb_p = skb2;
4173 kfree_skb(skb1);
4174 skb1 = skb2;
4175 }
4176 elt++;
4177 *trailer = skb1;
4178 skb_p = &skb1->next;
4179 }
4180
4181 return elt;
4182 }
4183 EXPORT_SYMBOL_GPL(skb_cow_data);
4184
sock_rmem_free(struct sk_buff * skb)4185 static void sock_rmem_free(struct sk_buff *skb)
4186 {
4187 struct sock *sk = skb->sk;
4188
4189 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
4190 }
4191
skb_set_err_queue(struct sk_buff * skb)4192 static void skb_set_err_queue(struct sk_buff *skb)
4193 {
4194 /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
4195 * So, it is safe to (mis)use it to mark skbs on the error queue.
4196 */
4197 skb->pkt_type = PACKET_OUTGOING;
4198 BUILD_BUG_ON(PACKET_OUTGOING == 0);
4199 }
4200
4201 /*
4202 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
4203 */
sock_queue_err_skb(struct sock * sk,struct sk_buff * skb)4204 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
4205 {
4206 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
4207 (unsigned int)sk->sk_rcvbuf)
4208 return -ENOMEM;
4209
4210 skb_orphan(skb);
4211 skb->sk = sk;
4212 skb->destructor = sock_rmem_free;
4213 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
4214 skb_set_err_queue(skb);
4215
4216 /* before exiting rcu section, make sure dst is refcounted */
4217 skb_dst_force(skb);
4218
4219 skb_queue_tail(&sk->sk_error_queue, skb);
4220 if (!sock_flag(sk, SOCK_DEAD))
4221 sk->sk_error_report(sk);
4222 return 0;
4223 }
4224 EXPORT_SYMBOL(sock_queue_err_skb);
4225
is_icmp_err_skb(const struct sk_buff * skb)4226 static bool is_icmp_err_skb(const struct sk_buff *skb)
4227 {
4228 return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
4229 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
4230 }
4231
sock_dequeue_err_skb(struct sock * sk)4232 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
4233 {
4234 struct sk_buff_head *q = &sk->sk_error_queue;
4235 struct sk_buff *skb, *skb_next = NULL;
4236 bool icmp_next = false;
4237 unsigned long flags;
4238
4239 spin_lock_irqsave(&q->lock, flags);
4240 skb = __skb_dequeue(q);
4241 if (skb && (skb_next = skb_peek(q))) {
4242 icmp_next = is_icmp_err_skb(skb_next);
4243 if (icmp_next)
4244 sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_origin;
4245 }
4246 spin_unlock_irqrestore(&q->lock, flags);
4247
4248 if (is_icmp_err_skb(skb) && !icmp_next)
4249 sk->sk_err = 0;
4250
4251 if (skb_next)
4252 sk->sk_error_report(sk);
4253
4254 return skb;
4255 }
4256 EXPORT_SYMBOL(sock_dequeue_err_skb);
4257
4258 /**
4259 * skb_clone_sk - create clone of skb, and take reference to socket
4260 * @skb: the skb to clone
4261 *
4262 * This function creates a clone of a buffer that holds a reference on
4263 * sk_refcnt. Buffers created via this function are meant to be
4264 * returned using sock_queue_err_skb, or free via kfree_skb.
4265 *
4266 * When passing buffers allocated with this function to sock_queue_err_skb
4267 * it is necessary to wrap the call with sock_hold/sock_put in order to
4268 * prevent the socket from being released prior to being enqueued on
4269 * the sk_error_queue.
4270 */
skb_clone_sk(struct sk_buff * skb)4271 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
4272 {
4273 struct sock *sk = skb->sk;
4274 struct sk_buff *clone;
4275
4276 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
4277 return NULL;
4278
4279 clone = skb_clone(skb, GFP_ATOMIC);
4280 if (!clone) {
4281 sock_put(sk);
4282 return NULL;
4283 }
4284
4285 clone->sk = sk;
4286 clone->destructor = sock_efree;
4287
4288 return clone;
4289 }
4290 EXPORT_SYMBOL(skb_clone_sk);
4291
__skb_complete_tx_timestamp(struct sk_buff * skb,struct sock * sk,int tstype,bool opt_stats)4292 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
4293 struct sock *sk,
4294 int tstype,
4295 bool opt_stats)
4296 {
4297 struct sock_exterr_skb *serr;
4298 int err;
4299
4300 BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
4301
4302 serr = SKB_EXT_ERR(skb);
4303 memset(serr, 0, sizeof(*serr));
4304 serr->ee.ee_errno = ENOMSG;
4305 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
4306 serr->ee.ee_info = tstype;
4307 serr->opt_stats = opt_stats;
4308 serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
4309 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
4310 serr->ee.ee_data = skb_shinfo(skb)->tskey;
4311 if (sk->sk_protocol == IPPROTO_TCP &&
4312 sk->sk_type == SOCK_STREAM)
4313 serr->ee.ee_data -= sk->sk_tskey;
4314 }
4315
4316 err = sock_queue_err_skb(sk, skb);
4317
4318 if (err)
4319 kfree_skb(skb);
4320 }
4321
skb_may_tx_timestamp(struct sock * sk,bool tsonly)4322 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
4323 {
4324 bool ret;
4325
4326 if (likely(sysctl_tstamp_allow_data || tsonly))
4327 return true;
4328
4329 read_lock_bh(&sk->sk_callback_lock);
4330 ret = sk->sk_socket && sk->sk_socket->file &&
4331 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
4332 read_unlock_bh(&sk->sk_callback_lock);
4333 return ret;
4334 }
4335
skb_complete_tx_timestamp(struct sk_buff * skb,struct skb_shared_hwtstamps * hwtstamps)4336 void skb_complete_tx_timestamp(struct sk_buff *skb,
4337 struct skb_shared_hwtstamps *hwtstamps)
4338 {
4339 struct sock *sk = skb->sk;
4340
4341 if (!skb_may_tx_timestamp(sk, false))
4342 goto err;
4343
4344 /* Take a reference to prevent skb_orphan() from freeing the socket,
4345 * but only if the socket refcount is not zero.
4346 */
4347 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4348 *skb_hwtstamps(skb) = *hwtstamps;
4349 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
4350 sock_put(sk);
4351 return;
4352 }
4353
4354 err:
4355 kfree_skb(skb);
4356 }
4357 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
4358
__skb_tstamp_tx(struct sk_buff * orig_skb,struct skb_shared_hwtstamps * hwtstamps,struct sock * sk,int tstype)4359 void __skb_tstamp_tx(struct sk_buff *orig_skb,
4360 struct skb_shared_hwtstamps *hwtstamps,
4361 struct sock *sk, int tstype)
4362 {
4363 struct sk_buff *skb;
4364 bool tsonly, opt_stats = false;
4365
4366 if (!sk)
4367 return;
4368
4369 if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
4370 skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
4371 return;
4372
4373 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
4374 if (!skb_may_tx_timestamp(sk, tsonly))
4375 return;
4376
4377 if (tsonly) {
4378 #ifdef CONFIG_INET
4379 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
4380 sk->sk_protocol == IPPROTO_TCP &&
4381 sk->sk_type == SOCK_STREAM) {
4382 skb = tcp_get_timestamping_opt_stats(sk);
4383 opt_stats = true;
4384 } else
4385 #endif
4386 skb = alloc_skb(0, GFP_ATOMIC);
4387 } else {
4388 skb = skb_clone(orig_skb, GFP_ATOMIC);
4389 }
4390 if (!skb)
4391 return;
4392
4393 if (tsonly) {
4394 skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
4395 SKBTX_ANY_TSTAMP;
4396 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
4397 }
4398
4399 if (hwtstamps)
4400 *skb_hwtstamps(skb) = *hwtstamps;
4401 else
4402 skb->tstamp = ktime_get_real();
4403
4404 __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
4405 }
4406 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
4407
skb_tstamp_tx(struct sk_buff * orig_skb,struct skb_shared_hwtstamps * hwtstamps)4408 void skb_tstamp_tx(struct sk_buff *orig_skb,
4409 struct skb_shared_hwtstamps *hwtstamps)
4410 {
4411 return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
4412 SCM_TSTAMP_SND);
4413 }
4414 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
4415
skb_complete_wifi_ack(struct sk_buff * skb,bool acked)4416 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
4417 {
4418 struct sock *sk = skb->sk;
4419 struct sock_exterr_skb *serr;
4420 int err = 1;
4421
4422 skb->wifi_acked_valid = 1;
4423 skb->wifi_acked = acked;
4424
4425 serr = SKB_EXT_ERR(skb);
4426 memset(serr, 0, sizeof(*serr));
4427 serr->ee.ee_errno = ENOMSG;
4428 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
4429
4430 /* Take a reference to prevent skb_orphan() from freeing the socket,
4431 * but only if the socket refcount is not zero.
4432 */
4433 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4434 err = sock_queue_err_skb(sk, skb);
4435 sock_put(sk);
4436 }
4437 if (err)
4438 kfree_skb(skb);
4439 }
4440 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
4441
4442 /**
4443 * skb_partial_csum_set - set up and verify partial csum values for packet
4444 * @skb: the skb to set
4445 * @start: the number of bytes after skb->data to start checksumming.
4446 * @off: the offset from start to place the checksum.
4447 *
4448 * For untrusted partially-checksummed packets, we need to make sure the values
4449 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
4450 *
4451 * This function checks and sets those values and skb->ip_summed: if this
4452 * returns false you should drop the packet.
4453 */
skb_partial_csum_set(struct sk_buff * skb,u16 start,u16 off)4454 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
4455 {
4456 u32 csum_end = (u32)start + (u32)off + sizeof(__sum16);
4457 u32 csum_start = skb_headroom(skb) + (u32)start;
4458
4459 if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) {
4460 net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n",
4461 start, off, skb_headroom(skb), skb_headlen(skb));
4462 return false;
4463 }
4464 skb->ip_summed = CHECKSUM_PARTIAL;
4465 skb->csum_start = csum_start;
4466 skb->csum_offset = off;
4467 skb_set_transport_header(skb, start);
4468 return true;
4469 }
4470 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
4471
skb_maybe_pull_tail(struct sk_buff * skb,unsigned int len,unsigned int max)4472 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
4473 unsigned int max)
4474 {
4475 if (skb_headlen(skb) >= len)
4476 return 0;
4477
4478 /* If we need to pullup then pullup to the max, so we
4479 * won't need to do it again.
4480 */
4481 if (max > skb->len)
4482 max = skb->len;
4483
4484 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
4485 return -ENOMEM;
4486
4487 if (skb_headlen(skb) < len)
4488 return -EPROTO;
4489
4490 return 0;
4491 }
4492
4493 #define MAX_TCP_HDR_LEN (15 * 4)
4494
skb_checksum_setup_ip(struct sk_buff * skb,typeof (IPPROTO_IP)proto,unsigned int off)4495 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
4496 typeof(IPPROTO_IP) proto,
4497 unsigned int off)
4498 {
4499 switch (proto) {
4500 int err;
4501
4502 case IPPROTO_TCP:
4503 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
4504 off + MAX_TCP_HDR_LEN);
4505 if (!err && !skb_partial_csum_set(skb, off,
4506 offsetof(struct tcphdr,
4507 check)))
4508 err = -EPROTO;
4509 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
4510
4511 case IPPROTO_UDP:
4512 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
4513 off + sizeof(struct udphdr));
4514 if (!err && !skb_partial_csum_set(skb, off,
4515 offsetof(struct udphdr,
4516 check)))
4517 err = -EPROTO;
4518 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
4519 }
4520
4521 return ERR_PTR(-EPROTO);
4522 }
4523
4524 /* This value should be large enough to cover a tagged ethernet header plus
4525 * maximally sized IP and TCP or UDP headers.
4526 */
4527 #define MAX_IP_HDR_LEN 128
4528
skb_checksum_setup_ipv4(struct sk_buff * skb,bool recalculate)4529 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
4530 {
4531 unsigned int off;
4532 bool fragment;
4533 __sum16 *csum;
4534 int err;
4535
4536 fragment = false;
4537
4538 err = skb_maybe_pull_tail(skb,
4539 sizeof(struct iphdr),
4540 MAX_IP_HDR_LEN);
4541 if (err < 0)
4542 goto out;
4543
4544 if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
4545 fragment = true;
4546
4547 off = ip_hdrlen(skb);
4548
4549 err = -EPROTO;
4550
4551 if (fragment)
4552 goto out;
4553
4554 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
4555 if (IS_ERR(csum))
4556 return PTR_ERR(csum);
4557
4558 if (recalculate)
4559 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
4560 ip_hdr(skb)->daddr,
4561 skb->len - off,
4562 ip_hdr(skb)->protocol, 0);
4563 err = 0;
4564
4565 out:
4566 return err;
4567 }
4568
4569 /* This value should be large enough to cover a tagged ethernet header plus
4570 * an IPv6 header, all options, and a maximal TCP or UDP header.
4571 */
4572 #define MAX_IPV6_HDR_LEN 256
4573
4574 #define OPT_HDR(type, skb, off) \
4575 (type *)(skb_network_header(skb) + (off))
4576
skb_checksum_setup_ipv6(struct sk_buff * skb,bool recalculate)4577 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
4578 {
4579 int err;
4580 u8 nexthdr;
4581 unsigned int off;
4582 unsigned int len;
4583 bool fragment;
4584 bool done;
4585 __sum16 *csum;
4586
4587 fragment = false;
4588 done = false;
4589
4590 off = sizeof(struct ipv6hdr);
4591
4592 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4593 if (err < 0)
4594 goto out;
4595
4596 nexthdr = ipv6_hdr(skb)->nexthdr;
4597
4598 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4599 while (off <= len && !done) {
4600 switch (nexthdr) {
4601 case IPPROTO_DSTOPTS:
4602 case IPPROTO_HOPOPTS:
4603 case IPPROTO_ROUTING: {
4604 struct ipv6_opt_hdr *hp;
4605
4606 err = skb_maybe_pull_tail(skb,
4607 off +
4608 sizeof(struct ipv6_opt_hdr),
4609 MAX_IPV6_HDR_LEN);
4610 if (err < 0)
4611 goto out;
4612
4613 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
4614 nexthdr = hp->nexthdr;
4615 off += ipv6_optlen(hp);
4616 break;
4617 }
4618 case IPPROTO_AH: {
4619 struct ip_auth_hdr *hp;
4620
4621 err = skb_maybe_pull_tail(skb,
4622 off +
4623 sizeof(struct ip_auth_hdr),
4624 MAX_IPV6_HDR_LEN);
4625 if (err < 0)
4626 goto out;
4627
4628 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
4629 nexthdr = hp->nexthdr;
4630 off += ipv6_authlen(hp);
4631 break;
4632 }
4633 case IPPROTO_FRAGMENT: {
4634 struct frag_hdr *hp;
4635
4636 err = skb_maybe_pull_tail(skb,
4637 off +
4638 sizeof(struct frag_hdr),
4639 MAX_IPV6_HDR_LEN);
4640 if (err < 0)
4641 goto out;
4642
4643 hp = OPT_HDR(struct frag_hdr, skb, off);
4644
4645 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
4646 fragment = true;
4647
4648 nexthdr = hp->nexthdr;
4649 off += sizeof(struct frag_hdr);
4650 break;
4651 }
4652 default:
4653 done = true;
4654 break;
4655 }
4656 }
4657
4658 err = -EPROTO;
4659
4660 if (!done || fragment)
4661 goto out;
4662
4663 csum = skb_checksum_setup_ip(skb, nexthdr, off);
4664 if (IS_ERR(csum))
4665 return PTR_ERR(csum);
4666
4667 if (recalculate)
4668 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4669 &ipv6_hdr(skb)->daddr,
4670 skb->len - off, nexthdr, 0);
4671 err = 0;
4672
4673 out:
4674 return err;
4675 }
4676
4677 /**
4678 * skb_checksum_setup - set up partial checksum offset
4679 * @skb: the skb to set up
4680 * @recalculate: if true the pseudo-header checksum will be recalculated
4681 */
skb_checksum_setup(struct sk_buff * skb,bool recalculate)4682 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
4683 {
4684 int err;
4685
4686 switch (skb->protocol) {
4687 case htons(ETH_P_IP):
4688 err = skb_checksum_setup_ipv4(skb, recalculate);
4689 break;
4690
4691 case htons(ETH_P_IPV6):
4692 err = skb_checksum_setup_ipv6(skb, recalculate);
4693 break;
4694
4695 default:
4696 err = -EPROTO;
4697 break;
4698 }
4699
4700 return err;
4701 }
4702 EXPORT_SYMBOL(skb_checksum_setup);
4703
4704 /**
4705 * skb_checksum_maybe_trim - maybe trims the given skb
4706 * @skb: the skb to check
4707 * @transport_len: the data length beyond the network header
4708 *
4709 * Checks whether the given skb has data beyond the given transport length.
4710 * If so, returns a cloned skb trimmed to this transport length.
4711 * Otherwise returns the provided skb. Returns NULL in error cases
4712 * (e.g. transport_len exceeds skb length or out-of-memory).
4713 *
4714 * Caller needs to set the skb transport header and free any returned skb if it
4715 * differs from the provided skb.
4716 */
skb_checksum_maybe_trim(struct sk_buff * skb,unsigned int transport_len)4717 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
4718 unsigned int transport_len)
4719 {
4720 struct sk_buff *skb_chk;
4721 unsigned int len = skb_transport_offset(skb) + transport_len;
4722 int ret;
4723
4724 if (skb->len < len)
4725 return NULL;
4726 else if (skb->len == len)
4727 return skb;
4728
4729 skb_chk = skb_clone(skb, GFP_ATOMIC);
4730 if (!skb_chk)
4731 return NULL;
4732
4733 ret = pskb_trim_rcsum(skb_chk, len);
4734 if (ret) {
4735 kfree_skb(skb_chk);
4736 return NULL;
4737 }
4738
4739 return skb_chk;
4740 }
4741
4742 /**
4743 * skb_checksum_trimmed - validate checksum of an skb
4744 * @skb: the skb to check
4745 * @transport_len: the data length beyond the network header
4746 * @skb_chkf: checksum function to use
4747 *
4748 * Applies the given checksum function skb_chkf to the provided skb.
4749 * Returns a checked and maybe trimmed skb. Returns NULL on error.
4750 *
4751 * If the skb has data beyond the given transport length, then a
4752 * trimmed & cloned skb is checked and returned.
4753 *
4754 * Caller needs to set the skb transport header and free any returned skb if it
4755 * differs from the provided skb.
4756 */
skb_checksum_trimmed(struct sk_buff * skb,unsigned int transport_len,__sum16 (* skb_chkf)(struct sk_buff * skb))4757 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4758 unsigned int transport_len,
4759 __sum16(*skb_chkf)(struct sk_buff *skb))
4760 {
4761 struct sk_buff *skb_chk;
4762 unsigned int offset = skb_transport_offset(skb);
4763 __sum16 ret;
4764
4765 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
4766 if (!skb_chk)
4767 goto err;
4768
4769 if (!pskb_may_pull(skb_chk, offset))
4770 goto err;
4771
4772 skb_pull_rcsum(skb_chk, offset);
4773 ret = skb_chkf(skb_chk);
4774 skb_push_rcsum(skb_chk, offset);
4775
4776 if (ret)
4777 goto err;
4778
4779 return skb_chk;
4780
4781 err:
4782 if (skb_chk && skb_chk != skb)
4783 kfree_skb(skb_chk);
4784
4785 return NULL;
4786
4787 }
4788 EXPORT_SYMBOL(skb_checksum_trimmed);
4789
__skb_warn_lro_forwarding(const struct sk_buff * skb)4790 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
4791 {
4792 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
4793 skb->dev->name);
4794 }
4795 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
4796
kfree_skb_partial(struct sk_buff * skb,bool head_stolen)4797 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
4798 {
4799 if (head_stolen) {
4800 skb_release_head_state(skb);
4801 kmem_cache_free(skbuff_head_cache, skb);
4802 } else {
4803 __kfree_skb(skb);
4804 }
4805 }
4806 EXPORT_SYMBOL(kfree_skb_partial);
4807
4808 /**
4809 * skb_try_coalesce - try to merge skb to prior one
4810 * @to: prior buffer
4811 * @from: buffer to add
4812 * @fragstolen: pointer to boolean
4813 * @delta_truesize: how much more was allocated than was requested
4814 */
skb_try_coalesce(struct sk_buff * to,struct sk_buff * from,bool * fragstolen,int * delta_truesize)4815 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
4816 bool *fragstolen, int *delta_truesize)
4817 {
4818 struct skb_shared_info *to_shinfo, *from_shinfo;
4819 int i, delta, len = from->len;
4820
4821 *fragstolen = false;
4822
4823 if (skb_cloned(to))
4824 return false;
4825
4826 if (len <= skb_tailroom(to)) {
4827 if (len)
4828 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
4829 *delta_truesize = 0;
4830 return true;
4831 }
4832
4833 to_shinfo = skb_shinfo(to);
4834 from_shinfo = skb_shinfo(from);
4835 if (to_shinfo->frag_list || from_shinfo->frag_list)
4836 return false;
4837 if (skb_zcopy(to) || skb_zcopy(from))
4838 return false;
4839
4840 if (skb_headlen(from) != 0) {
4841 struct page *page;
4842 unsigned int offset;
4843
4844 if (to_shinfo->nr_frags +
4845 from_shinfo->nr_frags >= MAX_SKB_FRAGS)
4846 return false;
4847
4848 if (skb_head_is_locked(from))
4849 return false;
4850
4851 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4852
4853 page = virt_to_head_page(from->head);
4854 offset = from->data - (unsigned char *)page_address(page);
4855
4856 skb_fill_page_desc(to, to_shinfo->nr_frags,
4857 page, offset, skb_headlen(from));
4858 *fragstolen = true;
4859 } else {
4860 if (to_shinfo->nr_frags +
4861 from_shinfo->nr_frags > MAX_SKB_FRAGS)
4862 return false;
4863
4864 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
4865 }
4866
4867 WARN_ON_ONCE(delta < len);
4868
4869 memcpy(to_shinfo->frags + to_shinfo->nr_frags,
4870 from_shinfo->frags,
4871 from_shinfo->nr_frags * sizeof(skb_frag_t));
4872 to_shinfo->nr_frags += from_shinfo->nr_frags;
4873
4874 if (!skb_cloned(from))
4875 from_shinfo->nr_frags = 0;
4876
4877 /* if the skb is not cloned this does nothing
4878 * since we set nr_frags to 0.
4879 */
4880 for (i = 0; i < from_shinfo->nr_frags; i++)
4881 __skb_frag_ref(&from_shinfo->frags[i]);
4882
4883 to->truesize += delta;
4884 to->len += len;
4885 to->data_len += len;
4886
4887 *delta_truesize = delta;
4888 return true;
4889 }
4890 EXPORT_SYMBOL(skb_try_coalesce);
4891
4892 /**
4893 * skb_scrub_packet - scrub an skb
4894 *
4895 * @skb: buffer to clean
4896 * @xnet: packet is crossing netns
4897 *
4898 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
4899 * into/from a tunnel. Some information have to be cleared during these
4900 * operations.
4901 * skb_scrub_packet can also be used to clean a skb before injecting it in
4902 * another namespace (@xnet == true). We have to clear all information in the
4903 * skb that could impact namespace isolation.
4904 */
skb_scrub_packet(struct sk_buff * skb,bool xnet)4905 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
4906 {
4907 skb->pkt_type = PACKET_HOST;
4908 skb->skb_iif = 0;
4909 skb->ignore_df = 0;
4910 skb_dst_drop(skb);
4911 secpath_reset(skb);
4912 nf_reset(skb);
4913 nf_reset_trace(skb);
4914
4915 if (!xnet)
4916 return;
4917
4918 ipvs_reset(skb);
4919 skb->mark = 0;
4920 skb->tstamp = 0;
4921 }
4922 EXPORT_SYMBOL_GPL(skb_scrub_packet);
4923
4924 /**
4925 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
4926 *
4927 * @skb: GSO skb
4928 *
4929 * skb_gso_transport_seglen is used to determine the real size of the
4930 * individual segments, including Layer4 headers (TCP/UDP).
4931 *
4932 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
4933 */
skb_gso_transport_seglen(const struct sk_buff * skb)4934 static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
4935 {
4936 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4937 unsigned int thlen = 0;
4938
4939 if (skb->encapsulation) {
4940 thlen = skb_inner_transport_header(skb) -
4941 skb_transport_header(skb);
4942
4943 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
4944 thlen += inner_tcp_hdrlen(skb);
4945 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
4946 thlen = tcp_hdrlen(skb);
4947 } else if (unlikely(skb_is_gso_sctp(skb))) {
4948 thlen = sizeof(struct sctphdr);
4949 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) {
4950 thlen = sizeof(struct udphdr);
4951 }
4952 /* UFO sets gso_size to the size of the fragmentation
4953 * payload, i.e. the size of the L4 (UDP) header is already
4954 * accounted for.
4955 */
4956 return thlen + shinfo->gso_size;
4957 }
4958
4959 /**
4960 * skb_gso_network_seglen - Return length of individual segments of a gso packet
4961 *
4962 * @skb: GSO skb
4963 *
4964 * skb_gso_network_seglen is used to determine the real size of the
4965 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
4966 *
4967 * The MAC/L2 header is not accounted for.
4968 */
skb_gso_network_seglen(const struct sk_buff * skb)4969 static unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
4970 {
4971 unsigned int hdr_len = skb_transport_header(skb) -
4972 skb_network_header(skb);
4973
4974 return hdr_len + skb_gso_transport_seglen(skb);
4975 }
4976
4977 /**
4978 * skb_gso_mac_seglen - Return length of individual segments of a gso packet
4979 *
4980 * @skb: GSO skb
4981 *
4982 * skb_gso_mac_seglen is used to determine the real size of the
4983 * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4
4984 * headers (TCP/UDP).
4985 */
skb_gso_mac_seglen(const struct sk_buff * skb)4986 static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb)
4987 {
4988 unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
4989
4990 return hdr_len + skb_gso_transport_seglen(skb);
4991 }
4992
4993 /**
4994 * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS
4995 *
4996 * There are a couple of instances where we have a GSO skb, and we
4997 * want to determine what size it would be after it is segmented.
4998 *
4999 * We might want to check:
5000 * - L3+L4+payload size (e.g. IP forwarding)
5001 * - L2+L3+L4+payload size (e.g. sanity check before passing to driver)
5002 *
5003 * This is a helper to do that correctly considering GSO_BY_FRAGS.
5004 *
5005 * @seg_len: The segmented length (from skb_gso_*_seglen). In the
5006 * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS].
5007 *
5008 * @max_len: The maximum permissible length.
5009 *
5010 * Returns true if the segmented length <= max length.
5011 */
skb_gso_size_check(const struct sk_buff * skb,unsigned int seg_len,unsigned int max_len)5012 static inline bool skb_gso_size_check(const struct sk_buff *skb,
5013 unsigned int seg_len,
5014 unsigned int max_len) {
5015 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5016 const struct sk_buff *iter;
5017
5018 if (shinfo->gso_size != GSO_BY_FRAGS)
5019 return seg_len <= max_len;
5020
5021 /* Undo this so we can re-use header sizes */
5022 seg_len -= GSO_BY_FRAGS;
5023
5024 skb_walk_frags(skb, iter) {
5025 if (seg_len + skb_headlen(iter) > max_len)
5026 return false;
5027 }
5028
5029 return true;
5030 }
5031
5032 /**
5033 * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU?
5034 *
5035 * @skb: GSO skb
5036 * @mtu: MTU to validate against
5037 *
5038 * skb_gso_validate_network_len validates if a given skb will fit a
5039 * wanted MTU once split. It considers L3 headers, L4 headers, and the
5040 * payload.
5041 */
skb_gso_validate_network_len(const struct sk_buff * skb,unsigned int mtu)5042 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu)
5043 {
5044 return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu);
5045 }
5046 EXPORT_SYMBOL_GPL(skb_gso_validate_network_len);
5047
5048 /**
5049 * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length?
5050 *
5051 * @skb: GSO skb
5052 * @len: length to validate against
5053 *
5054 * skb_gso_validate_mac_len validates if a given skb will fit a wanted
5055 * length once split, including L2, L3 and L4 headers and the payload.
5056 */
skb_gso_validate_mac_len(const struct sk_buff * skb,unsigned int len)5057 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len)
5058 {
5059 return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len);
5060 }
5061 EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len);
5062
skb_reorder_vlan_header(struct sk_buff * skb)5063 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
5064 {
5065 int mac_len;
5066
5067 if (skb_cow(skb, skb_headroom(skb)) < 0) {
5068 kfree_skb(skb);
5069 return NULL;
5070 }
5071
5072 mac_len = skb->data - skb_mac_header(skb);
5073 if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) {
5074 memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb),
5075 mac_len - VLAN_HLEN - ETH_TLEN);
5076 }
5077 skb->mac_header += VLAN_HLEN;
5078 return skb;
5079 }
5080
skb_vlan_untag(struct sk_buff * skb)5081 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
5082 {
5083 struct vlan_hdr *vhdr;
5084 u16 vlan_tci;
5085
5086 if (unlikely(skb_vlan_tag_present(skb))) {
5087 /* vlan_tci is already set-up so leave this for another time */
5088 return skb;
5089 }
5090
5091 skb = skb_share_check(skb, GFP_ATOMIC);
5092 if (unlikely(!skb))
5093 goto err_free;
5094
5095 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN)))
5096 goto err_free;
5097
5098 vhdr = (struct vlan_hdr *)skb->data;
5099 vlan_tci = ntohs(vhdr->h_vlan_TCI);
5100 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
5101
5102 skb_pull_rcsum(skb, VLAN_HLEN);
5103 vlan_set_encap_proto(skb, vhdr);
5104
5105 skb = skb_reorder_vlan_header(skb);
5106 if (unlikely(!skb))
5107 goto err_free;
5108
5109 skb_reset_network_header(skb);
5110 skb_reset_transport_header(skb);
5111 skb_reset_mac_len(skb);
5112
5113 return skb;
5114
5115 err_free:
5116 kfree_skb(skb);
5117 return NULL;
5118 }
5119 EXPORT_SYMBOL(skb_vlan_untag);
5120
skb_ensure_writable(struct sk_buff * skb,int write_len)5121 int skb_ensure_writable(struct sk_buff *skb, int write_len)
5122 {
5123 if (!pskb_may_pull(skb, write_len))
5124 return -ENOMEM;
5125
5126 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
5127 return 0;
5128
5129 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5130 }
5131 EXPORT_SYMBOL(skb_ensure_writable);
5132
5133 /* remove VLAN header from packet and update csum accordingly.
5134 * expects a non skb_vlan_tag_present skb with a vlan tag payload
5135 */
__skb_vlan_pop(struct sk_buff * skb,u16 * vlan_tci)5136 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
5137 {
5138 struct vlan_hdr *vhdr;
5139 int offset = skb->data - skb_mac_header(skb);
5140 int err;
5141
5142 if (WARN_ONCE(offset,
5143 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
5144 offset)) {
5145 return -EINVAL;
5146 }
5147
5148 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
5149 if (unlikely(err))
5150 return err;
5151
5152 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5153
5154 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
5155 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
5156
5157 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
5158 __skb_pull(skb, VLAN_HLEN);
5159
5160 vlan_set_encap_proto(skb, vhdr);
5161 skb->mac_header += VLAN_HLEN;
5162
5163 if (skb_network_offset(skb) < ETH_HLEN)
5164 skb_set_network_header(skb, ETH_HLEN);
5165
5166 skb_reset_mac_len(skb);
5167
5168 return err;
5169 }
5170 EXPORT_SYMBOL(__skb_vlan_pop);
5171
5172 /* Pop a vlan tag either from hwaccel or from payload.
5173 * Expects skb->data at mac header.
5174 */
skb_vlan_pop(struct sk_buff * skb)5175 int skb_vlan_pop(struct sk_buff *skb)
5176 {
5177 u16 vlan_tci;
5178 __be16 vlan_proto;
5179 int err;
5180
5181 if (likely(skb_vlan_tag_present(skb))) {
5182 skb->vlan_tci = 0;
5183 } else {
5184 if (unlikely(!eth_type_vlan(skb->protocol)))
5185 return 0;
5186
5187 err = __skb_vlan_pop(skb, &vlan_tci);
5188 if (err)
5189 return err;
5190 }
5191 /* move next vlan tag to hw accel tag */
5192 if (likely(!eth_type_vlan(skb->protocol)))
5193 return 0;
5194
5195 vlan_proto = skb->protocol;
5196 err = __skb_vlan_pop(skb, &vlan_tci);
5197 if (unlikely(err))
5198 return err;
5199
5200 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5201 return 0;
5202 }
5203 EXPORT_SYMBOL(skb_vlan_pop);
5204
5205 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
5206 * Expects skb->data at mac header.
5207 */
skb_vlan_push(struct sk_buff * skb,__be16 vlan_proto,u16 vlan_tci)5208 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
5209 {
5210 if (skb_vlan_tag_present(skb)) {
5211 int offset = skb->data - skb_mac_header(skb);
5212 int err;
5213
5214 if (WARN_ONCE(offset,
5215 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
5216 offset)) {
5217 return -EINVAL;
5218 }
5219
5220 err = __vlan_insert_tag(skb, skb->vlan_proto,
5221 skb_vlan_tag_get(skb));
5222 if (err)
5223 return err;
5224
5225 skb->protocol = skb->vlan_proto;
5226 skb->mac_len += VLAN_HLEN;
5227
5228 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5229 }
5230 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5231 return 0;
5232 }
5233 EXPORT_SYMBOL(skb_vlan_push);
5234
5235 /**
5236 * alloc_skb_with_frags - allocate skb with page frags
5237 *
5238 * @header_len: size of linear part
5239 * @data_len: needed length in frags
5240 * @max_page_order: max page order desired.
5241 * @errcode: pointer to error code if any
5242 * @gfp_mask: allocation mask
5243 *
5244 * This can be used to allocate a paged skb, given a maximal order for frags.
5245 */
alloc_skb_with_frags(unsigned long header_len,unsigned long data_len,int max_page_order,int * errcode,gfp_t gfp_mask)5246 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
5247 unsigned long data_len,
5248 int max_page_order,
5249 int *errcode,
5250 gfp_t gfp_mask)
5251 {
5252 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
5253 unsigned long chunk;
5254 struct sk_buff *skb;
5255 struct page *page;
5256 gfp_t gfp_head;
5257 int i;
5258
5259 *errcode = -EMSGSIZE;
5260 /* Note this test could be relaxed, if we succeed to allocate
5261 * high order pages...
5262 */
5263 if (npages > MAX_SKB_FRAGS)
5264 return NULL;
5265
5266 gfp_head = gfp_mask;
5267 if (gfp_head & __GFP_DIRECT_RECLAIM)
5268 gfp_head |= __GFP_RETRY_MAYFAIL;
5269
5270 *errcode = -ENOBUFS;
5271 skb = alloc_skb(header_len, gfp_head);
5272 if (!skb)
5273 return NULL;
5274
5275 skb->truesize += npages << PAGE_SHIFT;
5276
5277 for (i = 0; npages > 0; i++) {
5278 int order = max_page_order;
5279
5280 while (order) {
5281 if (npages >= 1 << order) {
5282 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
5283 __GFP_COMP |
5284 __GFP_NOWARN,
5285 order);
5286 if (page)
5287 goto fill_page;
5288 /* Do not retry other high order allocations */
5289 order = 1;
5290 max_page_order = 0;
5291 }
5292 order--;
5293 }
5294 page = alloc_page(gfp_mask);
5295 if (!page)
5296 goto failure;
5297 fill_page:
5298 chunk = min_t(unsigned long, data_len,
5299 PAGE_SIZE << order);
5300 skb_fill_page_desc(skb, i, page, 0, chunk);
5301 data_len -= chunk;
5302 npages -= 1 << order;
5303 }
5304 return skb;
5305
5306 failure:
5307 kfree_skb(skb);
5308 return NULL;
5309 }
5310 EXPORT_SYMBOL(alloc_skb_with_frags);
5311
5312 /* carve out the first off bytes from skb when off < headlen */
pskb_carve_inside_header(struct sk_buff * skb,const u32 off,const int headlen,gfp_t gfp_mask)5313 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
5314 const int headlen, gfp_t gfp_mask)
5315 {
5316 int i;
5317 int size = skb_end_offset(skb);
5318 int new_hlen = headlen - off;
5319 u8 *data;
5320
5321 size = SKB_DATA_ALIGN(size);
5322
5323 if (skb_pfmemalloc(skb))
5324 gfp_mask |= __GFP_MEMALLOC;
5325 data = kmalloc_reserve(size +
5326 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5327 gfp_mask, NUMA_NO_NODE, NULL);
5328 if (!data)
5329 return -ENOMEM;
5330
5331 size = SKB_WITH_OVERHEAD(ksize(data));
5332
5333 /* Copy real data, and all frags */
5334 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
5335 skb->len -= off;
5336
5337 memcpy((struct skb_shared_info *)(data + size),
5338 skb_shinfo(skb),
5339 offsetof(struct skb_shared_info,
5340 frags[skb_shinfo(skb)->nr_frags]));
5341 if (skb_cloned(skb)) {
5342 /* drop the old head gracefully */
5343 if (skb_orphan_frags(skb, gfp_mask)) {
5344 kfree(data);
5345 return -ENOMEM;
5346 }
5347 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
5348 skb_frag_ref(skb, i);
5349 if (skb_has_frag_list(skb))
5350 skb_clone_fraglist(skb);
5351 skb_release_data(skb);
5352 } else {
5353 /* we can reuse existing recount- all we did was
5354 * relocate values
5355 */
5356 skb_free_head(skb);
5357 }
5358
5359 skb->head = data;
5360 skb->data = data;
5361 skb->head_frag = 0;
5362 #ifdef NET_SKBUFF_DATA_USES_OFFSET
5363 skb->end = size;
5364 #else
5365 skb->end = skb->head + size;
5366 #endif
5367 skb_set_tail_pointer(skb, skb_headlen(skb));
5368 skb_headers_offset_update(skb, 0);
5369 skb->cloned = 0;
5370 skb->hdr_len = 0;
5371 skb->nohdr = 0;
5372 atomic_set(&skb_shinfo(skb)->dataref, 1);
5373
5374 return 0;
5375 }
5376
5377 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
5378
5379 /* carve out the first eat bytes from skb's frag_list. May recurse into
5380 * pskb_carve()
5381 */
pskb_carve_frag_list(struct sk_buff * skb,struct skb_shared_info * shinfo,int eat,gfp_t gfp_mask)5382 static int pskb_carve_frag_list(struct sk_buff *skb,
5383 struct skb_shared_info *shinfo, int eat,
5384 gfp_t gfp_mask)
5385 {
5386 struct sk_buff *list = shinfo->frag_list;
5387 struct sk_buff *clone = NULL;
5388 struct sk_buff *insp = NULL;
5389
5390 do {
5391 if (!list) {
5392 pr_err("Not enough bytes to eat. Want %d\n", eat);
5393 return -EFAULT;
5394 }
5395 if (list->len <= eat) {
5396 /* Eaten as whole. */
5397 eat -= list->len;
5398 list = list->next;
5399 insp = list;
5400 } else {
5401 /* Eaten partially. */
5402 if (skb_shared(list)) {
5403 clone = skb_clone(list, gfp_mask);
5404 if (!clone)
5405 return -ENOMEM;
5406 insp = list->next;
5407 list = clone;
5408 } else {
5409 /* This may be pulled without problems. */
5410 insp = list;
5411 }
5412 if (pskb_carve(list, eat, gfp_mask) < 0) {
5413 kfree_skb(clone);
5414 return -ENOMEM;
5415 }
5416 break;
5417 }
5418 } while (eat);
5419
5420 /* Free pulled out fragments. */
5421 while ((list = shinfo->frag_list) != insp) {
5422 shinfo->frag_list = list->next;
5423 kfree_skb(list);
5424 }
5425 /* And insert new clone at head. */
5426 if (clone) {
5427 clone->next = list;
5428 shinfo->frag_list = clone;
5429 }
5430 return 0;
5431 }
5432
5433 /* carve off first len bytes from skb. Split line (off) is in the
5434 * non-linear part of skb
5435 */
pskb_carve_inside_nonlinear(struct sk_buff * skb,const u32 off,int pos,gfp_t gfp_mask)5436 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
5437 int pos, gfp_t gfp_mask)
5438 {
5439 int i, k = 0;
5440 int size = skb_end_offset(skb);
5441 u8 *data;
5442 const int nfrags = skb_shinfo(skb)->nr_frags;
5443 struct skb_shared_info *shinfo;
5444
5445 size = SKB_DATA_ALIGN(size);
5446
5447 if (skb_pfmemalloc(skb))
5448 gfp_mask |= __GFP_MEMALLOC;
5449 data = kmalloc_reserve(size +
5450 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5451 gfp_mask, NUMA_NO_NODE, NULL);
5452 if (!data)
5453 return -ENOMEM;
5454
5455 size = SKB_WITH_OVERHEAD(ksize(data));
5456
5457 memcpy((struct skb_shared_info *)(data + size),
5458 skb_shinfo(skb), offsetof(struct skb_shared_info,
5459 frags[skb_shinfo(skb)->nr_frags]));
5460 if (skb_orphan_frags(skb, gfp_mask)) {
5461 kfree(data);
5462 return -ENOMEM;
5463 }
5464 shinfo = (struct skb_shared_info *)(data + size);
5465 for (i = 0; i < nfrags; i++) {
5466 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
5467
5468 if (pos + fsize > off) {
5469 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
5470
5471 if (pos < off) {
5472 /* Split frag.
5473 * We have two variants in this case:
5474 * 1. Move all the frag to the second
5475 * part, if it is possible. F.e.
5476 * this approach is mandatory for TUX,
5477 * where splitting is expensive.
5478 * 2. Split is accurately. We make this.
5479 */
5480 shinfo->frags[0].page_offset += off - pos;
5481 skb_frag_size_sub(&shinfo->frags[0], off - pos);
5482 }
5483 skb_frag_ref(skb, i);
5484 k++;
5485 }
5486 pos += fsize;
5487 }
5488 shinfo->nr_frags = k;
5489 if (skb_has_frag_list(skb))
5490 skb_clone_fraglist(skb);
5491
5492 if (k == 0) {
5493 /* split line is in frag list */
5494 pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask);
5495 }
5496 skb_release_data(skb);
5497
5498 skb->head = data;
5499 skb->head_frag = 0;
5500 skb->data = data;
5501 #ifdef NET_SKBUFF_DATA_USES_OFFSET
5502 skb->end = size;
5503 #else
5504 skb->end = skb->head + size;
5505 #endif
5506 skb_reset_tail_pointer(skb);
5507 skb_headers_offset_update(skb, 0);
5508 skb->cloned = 0;
5509 skb->hdr_len = 0;
5510 skb->nohdr = 0;
5511 skb->len -= off;
5512 skb->data_len = skb->len;
5513 atomic_set(&skb_shinfo(skb)->dataref, 1);
5514 return 0;
5515 }
5516
5517 /* remove len bytes from the beginning of the skb */
pskb_carve(struct sk_buff * skb,const u32 len,gfp_t gfp)5518 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
5519 {
5520 int headlen = skb_headlen(skb);
5521
5522 if (len < headlen)
5523 return pskb_carve_inside_header(skb, len, headlen, gfp);
5524 else
5525 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
5526 }
5527
5528 /* Extract to_copy bytes starting at off from skb, and return this in
5529 * a new skb
5530 */
pskb_extract(struct sk_buff * skb,int off,int to_copy,gfp_t gfp)5531 struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
5532 int to_copy, gfp_t gfp)
5533 {
5534 struct sk_buff *clone = skb_clone(skb, gfp);
5535
5536 if (!clone)
5537 return NULL;
5538
5539 if (pskb_carve(clone, off, gfp) < 0 ||
5540 pskb_trim(clone, to_copy)) {
5541 kfree_skb(clone);
5542 return NULL;
5543 }
5544 return clone;
5545 }
5546 EXPORT_SYMBOL(pskb_extract);
5547
5548 /**
5549 * skb_condense - try to get rid of fragments/frag_list if possible
5550 * @skb: buffer
5551 *
5552 * Can be used to save memory before skb is added to a busy queue.
5553 * If packet has bytes in frags and enough tail room in skb->head,
5554 * pull all of them, so that we can free the frags right now and adjust
5555 * truesize.
5556 * Notes:
5557 * We do not reallocate skb->head thus can not fail.
5558 * Caller must re-evaluate skb->truesize if needed.
5559 */
skb_condense(struct sk_buff * skb)5560 void skb_condense(struct sk_buff *skb)
5561 {
5562 if (skb->data_len) {
5563 if (skb->data_len > skb->end - skb->tail ||
5564 skb_cloned(skb))
5565 return;
5566
5567 /* Nice, we can free page frag(s) right now */
5568 __pskb_pull_tail(skb, skb->data_len);
5569 }
5570 /* At this point, skb->truesize might be over estimated,
5571 * because skb had a fragment, and fragments do not tell
5572 * their truesize.
5573 * When we pulled its content into skb->head, fragment
5574 * was freed, but __pskb_pull_tail() could not possibly
5575 * adjust skb->truesize, not knowing the frag truesize.
5576 */
5577 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
5578 }
5579