Version:  2.0.40 2.2.26 2.4.37 3.3 3.4 3.5 3.6 3.7 3.8 3.9 3.10 3.11 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19

Linux/net/core/skbuff.c

  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/kmemcheck.h>
 45 #include <linux/mm.h>
 46 #include <linux/interrupt.h>
 47 #include <linux/in.h>
 48 #include <linux/inet.h>
 49 #include <linux/slab.h>
 50 #include <linux/tcp.h>
 51 #include <linux/udp.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 <asm/uaccess.h>
 75 #include <trace/events/skb.h>
 76 #include <linux/highmem.h>
 77 
 78 struct kmem_cache *skbuff_head_cache __read_mostly;
 79 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
 80 
 81 /**
 82  *      skb_panic - private function for out-of-line support
 83  *      @skb:   buffer
 84  *      @sz:    size
 85  *      @addr:  address
 86  *      @msg:   skb_over_panic or skb_under_panic
 87  *
 88  *      Out-of-line support for skb_put() and skb_push().
 89  *      Called via the wrapper skb_over_panic() or skb_under_panic().
 90  *      Keep out of line to prevent kernel bloat.
 91  *      __builtin_return_address is not used because it is not always reliable.
 92  */
 93 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
 94                       const char msg[])
 95 {
 96         pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
 97                  msg, addr, skb->len, sz, skb->head, skb->data,
 98                  (unsigned long)skb->tail, (unsigned long)skb->end,
 99                  skb->dev ? skb->dev->name : "<NULL>");
100         BUG();
101 }
102 
103 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
104 {
105         skb_panic(skb, sz, addr, __func__);
106 }
107 
108 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
109 {
110         skb_panic(skb, sz, addr, __func__);
111 }
112 
113 /*
114  * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
115  * the caller if emergency pfmemalloc reserves are being used. If it is and
116  * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
117  * may be used. Otherwise, the packet data may be discarded until enough
118  * memory is free
119  */
120 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
121          __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
122 
123 static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
124                                unsigned long ip, bool *pfmemalloc)
125 {
126         void *obj;
127         bool ret_pfmemalloc = false;
128 
129         /*
130          * Try a regular allocation, when that fails and we're not entitled
131          * to the reserves, fail.
132          */
133         obj = kmalloc_node_track_caller(size,
134                                         flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
135                                         node);
136         if (obj || !(gfp_pfmemalloc_allowed(flags)))
137                 goto out;
138 
139         /* Try again but now we are using pfmemalloc reserves */
140         ret_pfmemalloc = true;
141         obj = kmalloc_node_track_caller(size, flags, node);
142 
143 out:
144         if (pfmemalloc)
145                 *pfmemalloc = ret_pfmemalloc;
146 
147         return obj;
148 }
149 
150 /*      Allocate a new skbuff. We do this ourselves so we can fill in a few
151  *      'private' fields and also do memory statistics to find all the
152  *      [BEEP] leaks.
153  *
154  */
155 
156 struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node)
157 {
158         struct sk_buff *skb;
159 
160         /* Get the HEAD */
161         skb = kmem_cache_alloc_node(skbuff_head_cache,
162                                     gfp_mask & ~__GFP_DMA, node);
163         if (!skb)
164                 goto out;
165 
166         /*
167          * Only clear those fields we need to clear, not those that we will
168          * actually initialise below. Hence, don't put any more fields after
169          * the tail pointer in struct sk_buff!
170          */
171         memset(skb, 0, offsetof(struct sk_buff, tail));
172         skb->head = NULL;
173         skb->truesize = sizeof(struct sk_buff);
174         atomic_set(&skb->users, 1);
175 
176         skb->mac_header = (typeof(skb->mac_header))~0U;
177 out:
178         return skb;
179 }
180 
181 /**
182  *      __alloc_skb     -       allocate a network buffer
183  *      @size: size to allocate
184  *      @gfp_mask: allocation mask
185  *      @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
186  *              instead of head cache and allocate a cloned (child) skb.
187  *              If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
188  *              allocations in case the data is required for writeback
189  *      @node: numa node to allocate memory on
190  *
191  *      Allocate a new &sk_buff. The returned buffer has no headroom and a
192  *      tail room of at least size bytes. The object has a reference count
193  *      of one. The return is the buffer. On a failure the return is %NULL.
194  *
195  *      Buffers may only be allocated from interrupts using a @gfp_mask of
196  *      %GFP_ATOMIC.
197  */
198 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
199                             int flags, int node)
200 {
201         struct kmem_cache *cache;
202         struct skb_shared_info *shinfo;
203         struct sk_buff *skb;
204         u8 *data;
205         bool pfmemalloc;
206 
207         cache = (flags & SKB_ALLOC_FCLONE)
208                 ? skbuff_fclone_cache : skbuff_head_cache;
209 
210         if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
211                 gfp_mask |= __GFP_MEMALLOC;
212 
213         /* Get the HEAD */
214         skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
215         if (!skb)
216                 goto out;
217         prefetchw(skb);
218 
219         /* We do our best to align skb_shared_info on a separate cache
220          * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
221          * aligned memory blocks, unless SLUB/SLAB debug is enabled.
222          * Both skb->head and skb_shared_info are cache line aligned.
223          */
224         size = SKB_DATA_ALIGN(size);
225         size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
226         data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
227         if (!data)
228                 goto nodata;
229         /* kmalloc(size) might give us more room than requested.
230          * Put skb_shared_info exactly at the end of allocated zone,
231          * to allow max possible filling before reallocation.
232          */
233         size = SKB_WITH_OVERHEAD(ksize(data));
234         prefetchw(data + size);
235 
236         /*
237          * Only clear those fields we need to clear, not those that we will
238          * actually initialise below. Hence, don't put any more fields after
239          * the tail pointer in struct sk_buff!
240          */
241         memset(skb, 0, offsetof(struct sk_buff, tail));
242         /* Account for allocated memory : skb + skb->head */
243         skb->truesize = SKB_TRUESIZE(size);
244         skb->pfmemalloc = pfmemalloc;
245         atomic_set(&skb->users, 1);
246         skb->head = data;
247         skb->data = data;
248         skb_reset_tail_pointer(skb);
249         skb->end = skb->tail + size;
250         skb->mac_header = (typeof(skb->mac_header))~0U;
251         skb->transport_header = (typeof(skb->transport_header))~0U;
252 
253         /* make sure we initialize shinfo sequentially */
254         shinfo = skb_shinfo(skb);
255         memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
256         atomic_set(&shinfo->dataref, 1);
257         kmemcheck_annotate_variable(shinfo->destructor_arg);
258 
259         if (flags & SKB_ALLOC_FCLONE) {
260                 struct sk_buff_fclones *fclones;
261 
262                 fclones = container_of(skb, struct sk_buff_fclones, skb1);
263 
264                 kmemcheck_annotate_bitfield(&fclones->skb2, flags1);
265                 skb->fclone = SKB_FCLONE_ORIG;
266                 atomic_set(&fclones->fclone_ref, 1);
267 
268                 fclones->skb2.fclone = SKB_FCLONE_CLONE;
269                 fclones->skb2.pfmemalloc = pfmemalloc;
270         }
271 out:
272         return skb;
273 nodata:
274         kmem_cache_free(cache, skb);
275         skb = NULL;
276         goto out;
277 }
278 EXPORT_SYMBOL(__alloc_skb);
279 
280 /**
281  * build_skb - build a network buffer
282  * @data: data buffer provided by caller
283  * @frag_size: size of fragment, or 0 if head was kmalloced
284  *
285  * Allocate a new &sk_buff. Caller provides space holding head and
286  * skb_shared_info. @data must have been allocated by kmalloc() only if
287  * @frag_size is 0, otherwise data should come from the page allocator.
288  * The return is the new skb buffer.
289  * On a failure the return is %NULL, and @data is not freed.
290  * Notes :
291  *  Before IO, driver allocates only data buffer where NIC put incoming frame
292  *  Driver should add room at head (NET_SKB_PAD) and
293  *  MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
294  *  After IO, driver calls build_skb(), to allocate sk_buff and populate it
295  *  before giving packet to stack.
296  *  RX rings only contains data buffers, not full skbs.
297  */
298 struct sk_buff *build_skb(void *data, unsigned int frag_size)
299 {
300         struct skb_shared_info *shinfo;
301         struct sk_buff *skb;
302         unsigned int size = frag_size ? : ksize(data);
303 
304         skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
305         if (!skb)
306                 return NULL;
307 
308         size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
309 
310         memset(skb, 0, offsetof(struct sk_buff, tail));
311         skb->truesize = SKB_TRUESIZE(size);
312         skb->head_frag = frag_size != 0;
313         atomic_set(&skb->users, 1);
314         skb->head = data;
315         skb->data = data;
316         skb_reset_tail_pointer(skb);
317         skb->end = skb->tail + size;
318         skb->mac_header = (typeof(skb->mac_header))~0U;
319         skb->transport_header = (typeof(skb->transport_header))~0U;
320 
321         /* make sure we initialize shinfo sequentially */
322         shinfo = skb_shinfo(skb);
323         memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
324         atomic_set(&shinfo->dataref, 1);
325         kmemcheck_annotate_variable(shinfo->destructor_arg);
326 
327         return skb;
328 }
329 EXPORT_SYMBOL(build_skb);
330 
331 struct netdev_alloc_cache {
332         struct page_frag        frag;
333         /* we maintain a pagecount bias, so that we dont dirty cache line
334          * containing page->_count every time we allocate a fragment.
335          */
336         unsigned int            pagecnt_bias;
337 };
338 static DEFINE_PER_CPU(struct netdev_alloc_cache, netdev_alloc_cache);
339 static DEFINE_PER_CPU(struct netdev_alloc_cache, napi_alloc_cache);
340 
341 static struct page *__page_frag_refill(struct netdev_alloc_cache *nc,
342                                        gfp_t gfp_mask)
343 {
344         const unsigned int order = NETDEV_FRAG_PAGE_MAX_ORDER;
345         struct page *page = NULL;
346         gfp_t gfp = gfp_mask;
347 
348         if (order) {
349                 gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY;
350                 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
351                 nc->frag.size = PAGE_SIZE << (page ? order : 0);
352         }
353 
354         if (unlikely(!page))
355                 page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
356 
357         nc->frag.page = page;
358 
359         return page;
360 }
361 
362 static void *__alloc_page_frag(struct netdev_alloc_cache __percpu *cache,
363                                unsigned int fragsz, gfp_t gfp_mask)
364 {
365         struct netdev_alloc_cache *nc = this_cpu_ptr(cache);
366         struct page *page = nc->frag.page;
367         unsigned int size;
368         int offset;
369 
370         if (unlikely(!page)) {
371 refill:
372                 page = __page_frag_refill(nc, gfp_mask);
373                 if (!page)
374                         return NULL;
375 
376                 /* if size can vary use frag.size else just use PAGE_SIZE */
377                 size = NETDEV_FRAG_PAGE_MAX_ORDER ? nc->frag.size : PAGE_SIZE;
378 
379                 /* Even if we own the page, we do not use atomic_set().
380                  * This would break get_page_unless_zero() users.
381                  */
382                 atomic_add(size - 1, &page->_count);
383 
384                 /* reset page count bias and offset to start of new frag */
385                 nc->pagecnt_bias = size;
386                 nc->frag.offset = size;
387         }
388 
389         offset = nc->frag.offset - fragsz;
390         if (unlikely(offset < 0)) {
391                 if (!atomic_sub_and_test(nc->pagecnt_bias, &page->_count))
392                         goto refill;
393 
394                 /* if size can vary use frag.size else just use PAGE_SIZE */
395                 size = NETDEV_FRAG_PAGE_MAX_ORDER ? nc->frag.size : PAGE_SIZE;
396 
397                 /* OK, page count is 0, we can safely set it */
398                 atomic_set(&page->_count, size);
399 
400                 /* reset page count bias and offset to start of new frag */
401                 nc->pagecnt_bias = size;
402                 offset = size - fragsz;
403         }
404 
405         nc->pagecnt_bias--;
406         nc->frag.offset = offset;
407 
408         return page_address(page) + offset;
409 }
410 
411 static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
412 {
413         unsigned long flags;
414         void *data;
415 
416         local_irq_save(flags);
417         data = __alloc_page_frag(&netdev_alloc_cache, fragsz, gfp_mask);
418         local_irq_restore(flags);
419         return data;
420 }
421 
422 /**
423  * netdev_alloc_frag - allocate a page fragment
424  * @fragsz: fragment size
425  *
426  * Allocates a frag from a page for receive buffer.
427  * Uses GFP_ATOMIC allocations.
428  */
429 void *netdev_alloc_frag(unsigned int fragsz)
430 {
431         return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
432 }
433 EXPORT_SYMBOL(netdev_alloc_frag);
434 
435 static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
436 {
437         return __alloc_page_frag(&napi_alloc_cache, fragsz, gfp_mask);
438 }
439 
440 void *napi_alloc_frag(unsigned int fragsz)
441 {
442         return __napi_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
443 }
444 EXPORT_SYMBOL(napi_alloc_frag);
445 
446 /**
447  *      __alloc_rx_skb - allocate an skbuff for rx
448  *      @length: length to allocate
449  *      @gfp_mask: get_free_pages mask, passed to alloc_skb
450  *      @flags: If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
451  *              allocations in case we have to fallback to __alloc_skb()
452  *              If SKB_ALLOC_NAPI is set, page fragment will be allocated
453  *              from napi_cache instead of netdev_cache.
454  *
455  *      Allocate a new &sk_buff and assign it a usage count of one. The
456  *      buffer has unspecified headroom built in. Users should allocate
457  *      the headroom they think they need without accounting for the
458  *      built in space. The built in space is used for optimisations.
459  *
460  *      %NULL is returned if there is no free memory.
461  */
462 static struct sk_buff *__alloc_rx_skb(unsigned int length, gfp_t gfp_mask,
463                                       int flags)
464 {
465         struct sk_buff *skb = NULL;
466         unsigned int fragsz = SKB_DATA_ALIGN(length) +
467                               SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
468 
469         if (fragsz <= PAGE_SIZE && !(gfp_mask & (__GFP_WAIT | GFP_DMA))) {
470                 void *data;
471 
472                 if (sk_memalloc_socks())
473                         gfp_mask |= __GFP_MEMALLOC;
474 
475                 data = (flags & SKB_ALLOC_NAPI) ?
476                         __napi_alloc_frag(fragsz, gfp_mask) :
477                         __netdev_alloc_frag(fragsz, gfp_mask);
478 
479                 if (likely(data)) {
480                         skb = build_skb(data, fragsz);
481                         if (unlikely(!skb))
482                                 put_page(virt_to_head_page(data));
483                 }
484         } else {
485                 skb = __alloc_skb(length, gfp_mask,
486                                   SKB_ALLOC_RX, NUMA_NO_NODE);
487         }
488         return skb;
489 }
490 
491 /**
492  *      __netdev_alloc_skb - allocate an skbuff for rx on a specific device
493  *      @dev: network device to receive on
494  *      @length: length to allocate
495  *      @gfp_mask: get_free_pages mask, passed to alloc_skb
496  *
497  *      Allocate a new &sk_buff and assign it a usage count of one. The
498  *      buffer has NET_SKB_PAD headroom built in. Users should allocate
499  *      the headroom they think they need without accounting for the
500  *      built in space. The built in space is used for optimisations.
501  *
502  *      %NULL is returned if there is no free memory.
503  */
504 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
505                                    unsigned int length, gfp_t gfp_mask)
506 {
507         struct sk_buff *skb;
508 
509         length += NET_SKB_PAD;
510         skb = __alloc_rx_skb(length, gfp_mask, 0);
511 
512         if (likely(skb)) {
513                 skb_reserve(skb, NET_SKB_PAD);
514                 skb->dev = dev;
515         }
516 
517         return skb;
518 }
519 EXPORT_SYMBOL(__netdev_alloc_skb);
520 
521 /**
522  *      __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
523  *      @napi: napi instance this buffer was allocated for
524  *      @length: length to allocate
525  *      @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
526  *
527  *      Allocate a new sk_buff for use in NAPI receive.  This buffer will
528  *      attempt to allocate the head from a special reserved region used
529  *      only for NAPI Rx allocation.  By doing this we can save several
530  *      CPU cycles by avoiding having to disable and re-enable IRQs.
531  *
532  *      %NULL is returned if there is no free memory.
533  */
534 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
535                                  unsigned int length, gfp_t gfp_mask)
536 {
537         struct sk_buff *skb;
538 
539         length += NET_SKB_PAD + NET_IP_ALIGN;
540         skb = __alloc_rx_skb(length, gfp_mask, SKB_ALLOC_NAPI);
541 
542         if (likely(skb)) {
543                 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
544                 skb->dev = napi->dev;
545         }
546 
547         return skb;
548 }
549 EXPORT_SYMBOL(__napi_alloc_skb);
550 
551 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
552                      int size, unsigned int truesize)
553 {
554         skb_fill_page_desc(skb, i, page, off, size);
555         skb->len += size;
556         skb->data_len += size;
557         skb->truesize += truesize;
558 }
559 EXPORT_SYMBOL(skb_add_rx_frag);
560 
561 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
562                           unsigned int truesize)
563 {
564         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
565 
566         skb_frag_size_add(frag, size);
567         skb->len += size;
568         skb->data_len += size;
569         skb->truesize += truesize;
570 }
571 EXPORT_SYMBOL(skb_coalesce_rx_frag);
572 
573 static void skb_drop_list(struct sk_buff **listp)
574 {
575         kfree_skb_list(*listp);
576         *listp = NULL;
577 }
578 
579 static inline void skb_drop_fraglist(struct sk_buff *skb)
580 {
581         skb_drop_list(&skb_shinfo(skb)->frag_list);
582 }
583 
584 static void skb_clone_fraglist(struct sk_buff *skb)
585 {
586         struct sk_buff *list;
587 
588         skb_walk_frags(skb, list)
589                 skb_get(list);
590 }
591 
592 static void skb_free_head(struct sk_buff *skb)
593 {
594         if (skb->head_frag)
595                 put_page(virt_to_head_page(skb->head));
596         else
597                 kfree(skb->head);
598 }
599 
600 static void skb_release_data(struct sk_buff *skb)
601 {
602         struct skb_shared_info *shinfo = skb_shinfo(skb);
603         int i;
604 
605         if (skb->cloned &&
606             atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
607                               &shinfo->dataref))
608                 return;
609 
610         for (i = 0; i < shinfo->nr_frags; i++)
611                 __skb_frag_unref(&shinfo->frags[i]);
612 
613         /*
614          * If skb buf is from userspace, we need to notify the caller
615          * the lower device DMA has done;
616          */
617         if (shinfo->tx_flags & SKBTX_DEV_ZEROCOPY) {
618                 struct ubuf_info *uarg;
619 
620                 uarg = shinfo->destructor_arg;
621                 if (uarg->callback)
622                         uarg->callback(uarg, true);
623         }
624 
625         if (shinfo->frag_list)
626                 kfree_skb_list(shinfo->frag_list);
627 
628         skb_free_head(skb);
629 }
630 
631 /*
632  *      Free an skbuff by memory without cleaning the state.
633  */
634 static void kfree_skbmem(struct sk_buff *skb)
635 {
636         struct sk_buff_fclones *fclones;
637 
638         switch (skb->fclone) {
639         case SKB_FCLONE_UNAVAILABLE:
640                 kmem_cache_free(skbuff_head_cache, skb);
641                 return;
642 
643         case SKB_FCLONE_ORIG:
644                 fclones = container_of(skb, struct sk_buff_fclones, skb1);
645 
646                 /* We usually free the clone (TX completion) before original skb
647                  * This test would have no chance to be true for the clone,
648                  * while here, branch prediction will be good.
649                  */
650                 if (atomic_read(&fclones->fclone_ref) == 1)
651                         goto fastpath;
652                 break;
653 
654         default: /* SKB_FCLONE_CLONE */
655                 fclones = container_of(skb, struct sk_buff_fclones, skb2);
656                 break;
657         }
658         if (!atomic_dec_and_test(&fclones->fclone_ref))
659                 return;
660 fastpath:
661         kmem_cache_free(skbuff_fclone_cache, fclones);
662 }
663 
664 static void skb_release_head_state(struct sk_buff *skb)
665 {
666         skb_dst_drop(skb);
667 #ifdef CONFIG_XFRM
668         secpath_put(skb->sp);
669 #endif
670         if (skb->destructor) {
671                 WARN_ON(in_irq());
672                 skb->destructor(skb);
673         }
674 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
675         nf_conntrack_put(skb->nfct);
676 #endif
677 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
678         nf_bridge_put(skb->nf_bridge);
679 #endif
680 /* XXX: IS this still necessary? - JHS */
681 #ifdef CONFIG_NET_SCHED
682         skb->tc_index = 0;
683 #ifdef CONFIG_NET_CLS_ACT
684         skb->tc_verd = 0;
685 #endif
686 #endif
687 }
688 
689 /* Free everything but the sk_buff shell. */
690 static void skb_release_all(struct sk_buff *skb)
691 {
692         skb_release_head_state(skb);
693         if (likely(skb->head))
694                 skb_release_data(skb);
695 }
696 
697 /**
698  *      __kfree_skb - private function
699  *      @skb: buffer
700  *
701  *      Free an sk_buff. Release anything attached to the buffer.
702  *      Clean the state. This is an internal helper function. Users should
703  *      always call kfree_skb
704  */
705 
706 void __kfree_skb(struct sk_buff *skb)
707 {
708         skb_release_all(skb);
709         kfree_skbmem(skb);
710 }
711 EXPORT_SYMBOL(__kfree_skb);
712 
713 /**
714  *      kfree_skb - free an sk_buff
715  *      @skb: buffer to free
716  *
717  *      Drop a reference to the buffer and free it if the usage count has
718  *      hit zero.
719  */
720 void kfree_skb(struct sk_buff *skb)
721 {
722         if (unlikely(!skb))
723                 return;
724         if (likely(atomic_read(&skb->users) == 1))
725                 smp_rmb();
726         else if (likely(!atomic_dec_and_test(&skb->users)))
727                 return;
728         trace_kfree_skb(skb, __builtin_return_address(0));
729         __kfree_skb(skb);
730 }
731 EXPORT_SYMBOL(kfree_skb);
732 
733 void kfree_skb_list(struct sk_buff *segs)
734 {
735         while (segs) {
736                 struct sk_buff *next = segs->next;
737 
738                 kfree_skb(segs);
739                 segs = next;
740         }
741 }
742 EXPORT_SYMBOL(kfree_skb_list);
743 
744 /**
745  *      skb_tx_error - report an sk_buff xmit error
746  *      @skb: buffer that triggered an error
747  *
748  *      Report xmit error if a device callback is tracking this skb.
749  *      skb must be freed afterwards.
750  */
751 void skb_tx_error(struct sk_buff *skb)
752 {
753         if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
754                 struct ubuf_info *uarg;
755 
756                 uarg = skb_shinfo(skb)->destructor_arg;
757                 if (uarg->callback)
758                         uarg->callback(uarg, false);
759                 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
760         }
761 }
762 EXPORT_SYMBOL(skb_tx_error);
763 
764 /**
765  *      consume_skb - free an skbuff
766  *      @skb: buffer to free
767  *
768  *      Drop a ref to the buffer and free it if the usage count has hit zero
769  *      Functions identically to kfree_skb, but kfree_skb assumes that the frame
770  *      is being dropped after a failure and notes that
771  */
772 void consume_skb(struct sk_buff *skb)
773 {
774         if (unlikely(!skb))
775                 return;
776         if (likely(atomic_read(&skb->users) == 1))
777                 smp_rmb();
778         else if (likely(!atomic_dec_and_test(&skb->users)))
779                 return;
780         trace_consume_skb(skb);
781         __kfree_skb(skb);
782 }
783 EXPORT_SYMBOL(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 
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_NET_SCHED
834         CHECK_SKB_FIELD(tc_index);
835 #ifdef CONFIG_NET_CLS_ACT
836         CHECK_SKB_FIELD(tc_verd);
837 #endif
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  */
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->destructor = NULL;
861         C(tail);
862         C(end);
863         C(head);
864         C(head_frag);
865         C(data);
866         C(truesize);
867         atomic_set(&n->users, 1);
868 
869         atomic_inc(&(skb_shinfo(skb)->dataref));
870         skb->cloned = 1;
871 
872         return n;
873 #undef C
874 }
875 
876 /**
877  *      skb_morph       -       morph one skb into another
878  *      @dst: the skb to receive the contents
879  *      @src: the skb to supply the contents
880  *
881  *      This is identical to skb_clone except that the target skb is
882  *      supplied by the user.
883  *
884  *      The target skb is returned upon exit.
885  */
886 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
887 {
888         skb_release_all(dst);
889         return __skb_clone(dst, src);
890 }
891 EXPORT_SYMBOL_GPL(skb_morph);
892 
893 /**
894  *      skb_copy_ubufs  -       copy userspace skb frags buffers to kernel
895  *      @skb: the skb to modify
896  *      @gfp_mask: allocation priority
897  *
898  *      This must be called on SKBTX_DEV_ZEROCOPY skb.
899  *      It will copy all frags into kernel and drop the reference
900  *      to userspace pages.
901  *
902  *      If this function is called from an interrupt gfp_mask() must be
903  *      %GFP_ATOMIC.
904  *
905  *      Returns 0 on success or a negative error code on failure
906  *      to allocate kernel memory to copy to.
907  */
908 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
909 {
910         int i;
911         int num_frags = skb_shinfo(skb)->nr_frags;
912         struct page *page, *head = NULL;
913         struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
914 
915         for (i = 0; i < num_frags; i++) {
916                 u8 *vaddr;
917                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
918 
919                 page = alloc_page(gfp_mask);
920                 if (!page) {
921                         while (head) {
922                                 struct page *next = (struct page *)page_private(head);
923                                 put_page(head);
924                                 head = next;
925                         }
926                         return -ENOMEM;
927                 }
928                 vaddr = kmap_atomic(skb_frag_page(f));
929                 memcpy(page_address(page),
930                        vaddr + f->page_offset, skb_frag_size(f));
931                 kunmap_atomic(vaddr);
932                 set_page_private(page, (unsigned long)head);
933                 head = page;
934         }
935 
936         /* skb frags release userspace buffers */
937         for (i = 0; i < num_frags; i++)
938                 skb_frag_unref(skb, i);
939 
940         uarg->callback(uarg, false);
941 
942         /* skb frags point to kernel buffers */
943         for (i = num_frags - 1; i >= 0; i--) {
944                 __skb_fill_page_desc(skb, i, head, 0,
945                                      skb_shinfo(skb)->frags[i].size);
946                 head = (struct page *)page_private(head);
947         }
948 
949         skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
950         return 0;
951 }
952 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
953 
954 /**
955  *      skb_clone       -       duplicate an sk_buff
956  *      @skb: buffer to clone
957  *      @gfp_mask: allocation priority
958  *
959  *      Duplicate an &sk_buff. The new one is not owned by a socket. Both
960  *      copies share the same packet data but not structure. The new
961  *      buffer has a reference count of 1. If the allocation fails the
962  *      function returns %NULL otherwise the new buffer is returned.
963  *
964  *      If this function is called from an interrupt gfp_mask() must be
965  *      %GFP_ATOMIC.
966  */
967 
968 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
969 {
970         struct sk_buff_fclones *fclones = container_of(skb,
971                                                        struct sk_buff_fclones,
972                                                        skb1);
973         struct sk_buff *n;
974 
975         if (skb_orphan_frags(skb, gfp_mask))
976                 return NULL;
977 
978         if (skb->fclone == SKB_FCLONE_ORIG &&
979             atomic_read(&fclones->fclone_ref) == 1) {
980                 n = &fclones->skb2;
981                 atomic_set(&fclones->fclone_ref, 2);
982         } else {
983                 if (skb_pfmemalloc(skb))
984                         gfp_mask |= __GFP_MEMALLOC;
985 
986                 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
987                 if (!n)
988                         return NULL;
989 
990                 kmemcheck_annotate_bitfield(n, flags1);
991                 n->fclone = SKB_FCLONE_UNAVAILABLE;
992         }
993 
994         return __skb_clone(n, skb);
995 }
996 EXPORT_SYMBOL(skb_clone);
997 
998 static void skb_headers_offset_update(struct sk_buff *skb, int off)
999 {
1000         /* Only adjust this if it actually is csum_start rather than csum */
1001         if (skb->ip_summed == CHECKSUM_PARTIAL)
1002                 skb->csum_start += off;
1003         /* {transport,network,mac}_header and tail are relative to skb->head */
1004         skb->transport_header += off;
1005         skb->network_header   += off;
1006         if (skb_mac_header_was_set(skb))
1007                 skb->mac_header += off;
1008         skb->inner_transport_header += off;
1009         skb->inner_network_header += off;
1010         skb->inner_mac_header += off;
1011 }
1012 
1013 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
1014 {
1015         __copy_skb_header(new, old);
1016 
1017         skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1018         skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1019         skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1020 }
1021 
1022 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1023 {
1024         if (skb_pfmemalloc(skb))
1025                 return SKB_ALLOC_RX;
1026         return 0;
1027 }
1028 
1029 /**
1030  *      skb_copy        -       create private copy of an sk_buff
1031  *      @skb: buffer to copy
1032  *      @gfp_mask: allocation priority
1033  *
1034  *      Make a copy of both an &sk_buff and its data. This is used when the
1035  *      caller wishes to modify the data and needs a private copy of the
1036  *      data to alter. Returns %NULL on failure or the pointer to the buffer
1037  *      on success. The returned buffer has a reference count of 1.
1038  *
1039  *      As by-product this function converts non-linear &sk_buff to linear
1040  *      one, so that &sk_buff becomes completely private and caller is allowed
1041  *      to modify all the data of returned buffer. This means that this
1042  *      function is not recommended for use in circumstances when only
1043  *      header is going to be modified. Use pskb_copy() instead.
1044  */
1045 
1046 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1047 {
1048         int headerlen = skb_headroom(skb);
1049         unsigned int size = skb_end_offset(skb) + skb->data_len;
1050         struct sk_buff *n = __alloc_skb(size, gfp_mask,
1051                                         skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1052 
1053         if (!n)
1054                 return NULL;
1055 
1056         /* Set the data pointer */
1057         skb_reserve(n, headerlen);
1058         /* Set the tail pointer and length */
1059         skb_put(n, skb->len);
1060 
1061         if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
1062                 BUG();
1063 
1064         copy_skb_header(n, skb);
1065         return n;
1066 }
1067 EXPORT_SYMBOL(skb_copy);
1068 
1069 /**
1070  *      __pskb_copy_fclone      -  create copy of an sk_buff with private head.
1071  *      @skb: buffer to copy
1072  *      @headroom: headroom of new skb
1073  *      @gfp_mask: allocation priority
1074  *      @fclone: if true allocate the copy of the skb from the fclone
1075  *      cache instead of the head cache; it is recommended to set this
1076  *      to true for the cases where the copy will likely be cloned
1077  *
1078  *      Make a copy of both an &sk_buff and part of its data, located
1079  *      in header. Fragmented data remain shared. This is used when
1080  *      the caller wishes to modify only header of &sk_buff and needs
1081  *      private copy of the header to alter. Returns %NULL on failure
1082  *      or the pointer to the buffer on success.
1083  *      The returned buffer has a reference count of 1.
1084  */
1085 
1086 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1087                                    gfp_t gfp_mask, bool fclone)
1088 {
1089         unsigned int size = skb_headlen(skb) + headroom;
1090         int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1091         struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1092 
1093         if (!n)
1094                 goto out;
1095 
1096         /* Set the data pointer */
1097         skb_reserve(n, headroom);
1098         /* Set the tail pointer and length */
1099         skb_put(n, skb_headlen(skb));
1100         /* Copy the bytes */
1101         skb_copy_from_linear_data(skb, n->data, n->len);
1102 
1103         n->truesize += skb->data_len;
1104         n->data_len  = skb->data_len;
1105         n->len       = skb->len;
1106 
1107         if (skb_shinfo(skb)->nr_frags) {
1108                 int i;
1109 
1110                 if (skb_orphan_frags(skb, gfp_mask)) {
1111                         kfree_skb(n);
1112                         n = NULL;
1113                         goto out;
1114                 }
1115                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1116                         skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1117                         skb_frag_ref(skb, i);
1118                 }
1119                 skb_shinfo(n)->nr_frags = i;
1120         }
1121 
1122         if (skb_has_frag_list(skb)) {
1123                 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1124                 skb_clone_fraglist(n);
1125         }
1126 
1127         copy_skb_header(n, skb);
1128 out:
1129         return n;
1130 }
1131 EXPORT_SYMBOL(__pskb_copy_fclone);
1132 
1133 /**
1134  *      pskb_expand_head - reallocate header of &sk_buff
1135  *      @skb: buffer to reallocate
1136  *      @nhead: room to add at head
1137  *      @ntail: room to add at tail
1138  *      @gfp_mask: allocation priority
1139  *
1140  *      Expands (or creates identical copy, if @nhead and @ntail are zero)
1141  *      header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1142  *      reference count of 1. Returns zero in the case of success or error,
1143  *      if expansion failed. In the last case, &sk_buff is not changed.
1144  *
1145  *      All the pointers pointing into skb header may change and must be
1146  *      reloaded after call to this function.
1147  */
1148 
1149 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1150                      gfp_t gfp_mask)
1151 {
1152         int i;
1153         u8 *data;
1154         int size = nhead + skb_end_offset(skb) + ntail;
1155         long off;
1156 
1157         BUG_ON(nhead < 0);
1158 
1159         if (skb_shared(skb))
1160                 BUG();
1161 
1162         size = SKB_DATA_ALIGN(size);
1163 
1164         if (skb_pfmemalloc(skb))
1165                 gfp_mask |= __GFP_MEMALLOC;
1166         data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1167                                gfp_mask, NUMA_NO_NODE, NULL);
1168         if (!data)
1169                 goto nodata;
1170         size = SKB_WITH_OVERHEAD(ksize(data));
1171 
1172         /* Copy only real data... and, alas, header. This should be
1173          * optimized for the cases when header is void.
1174          */
1175         memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1176 
1177         memcpy((struct skb_shared_info *)(data + size),
1178                skb_shinfo(skb),
1179                offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1180 
1181         /*
1182          * if shinfo is shared we must drop the old head gracefully, but if it
1183          * is not we can just drop the old head and let the existing refcount
1184          * be since all we did is relocate the values
1185          */
1186         if (skb_cloned(skb)) {
1187                 /* copy this zero copy skb frags */
1188                 if (skb_orphan_frags(skb, gfp_mask))
1189                         goto nofrags;
1190                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1191                         skb_frag_ref(skb, i);
1192 
1193                 if (skb_has_frag_list(skb))
1194                         skb_clone_fraglist(skb);
1195 
1196                 skb_release_data(skb);
1197         } else {
1198                 skb_free_head(skb);
1199         }
1200         off = (data + nhead) - skb->head;
1201 
1202         skb->head     = data;
1203         skb->head_frag = 0;
1204         skb->data    += off;
1205 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1206         skb->end      = size;
1207         off           = nhead;
1208 #else
1209         skb->end      = skb->head + size;
1210 #endif
1211         skb->tail             += off;
1212         skb_headers_offset_update(skb, nhead);
1213         skb->cloned   = 0;
1214         skb->hdr_len  = 0;
1215         skb->nohdr    = 0;
1216         atomic_set(&skb_shinfo(skb)->dataref, 1);
1217         return 0;
1218 
1219 nofrags:
1220         kfree(data);
1221 nodata:
1222         return -ENOMEM;
1223 }
1224 EXPORT_SYMBOL(pskb_expand_head);
1225 
1226 /* Make private copy of skb with writable head and some headroom */
1227 
1228 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1229 {
1230         struct sk_buff *skb2;
1231         int delta = headroom - skb_headroom(skb);
1232 
1233         if (delta <= 0)
1234                 skb2 = pskb_copy(skb, GFP_ATOMIC);
1235         else {
1236                 skb2 = skb_clone(skb, GFP_ATOMIC);
1237                 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1238                                              GFP_ATOMIC)) {
1239                         kfree_skb(skb2);
1240                         skb2 = NULL;
1241                 }
1242         }
1243         return skb2;
1244 }
1245 EXPORT_SYMBOL(skb_realloc_headroom);
1246 
1247 /**
1248  *      skb_copy_expand -       copy and expand sk_buff
1249  *      @skb: buffer to copy
1250  *      @newheadroom: new free bytes at head
1251  *      @newtailroom: new free bytes at tail
1252  *      @gfp_mask: allocation priority
1253  *
1254  *      Make a copy of both an &sk_buff and its data and while doing so
1255  *      allocate additional space.
1256  *
1257  *      This is used when the caller wishes to modify the data and needs a
1258  *      private copy of the data to alter as well as more space for new fields.
1259  *      Returns %NULL on failure or the pointer to the buffer
1260  *      on success. The returned buffer has a reference count of 1.
1261  *
1262  *      You must pass %GFP_ATOMIC as the allocation priority if this function
1263  *      is called from an interrupt.
1264  */
1265 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1266                                 int newheadroom, int newtailroom,
1267                                 gfp_t gfp_mask)
1268 {
1269         /*
1270          *      Allocate the copy buffer
1271          */
1272         struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1273                                         gfp_mask, skb_alloc_rx_flag(skb),
1274                                         NUMA_NO_NODE);
1275         int oldheadroom = skb_headroom(skb);
1276         int head_copy_len, head_copy_off;
1277 
1278         if (!n)
1279                 return NULL;
1280 
1281         skb_reserve(n, newheadroom);
1282 
1283         /* Set the tail pointer and length */
1284         skb_put(n, skb->len);
1285 
1286         head_copy_len = oldheadroom;
1287         head_copy_off = 0;
1288         if (newheadroom <= head_copy_len)
1289                 head_copy_len = newheadroom;
1290         else
1291                 head_copy_off = newheadroom - head_copy_len;
1292 
1293         /* Copy the linear header and data. */
1294         if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1295                           skb->len + head_copy_len))
1296                 BUG();
1297 
1298         copy_skb_header(n, skb);
1299 
1300         skb_headers_offset_update(n, newheadroom - oldheadroom);
1301 
1302         return n;
1303 }
1304 EXPORT_SYMBOL(skb_copy_expand);
1305 
1306 /**
1307  *      skb_pad                 -       zero pad the tail of an skb
1308  *      @skb: buffer to pad
1309  *      @pad: space to pad
1310  *
1311  *      Ensure that a buffer is followed by a padding area that is zero
1312  *      filled. Used by network drivers which may DMA or transfer data
1313  *      beyond the buffer end onto the wire.
1314  *
1315  *      May return error in out of memory cases. The skb is freed on error.
1316  */
1317 
1318 int skb_pad(struct sk_buff *skb, int pad)
1319 {
1320         int err;
1321         int ntail;
1322 
1323         /* If the skbuff is non linear tailroom is always zero.. */
1324         if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1325                 memset(skb->data+skb->len, 0, pad);
1326                 return 0;
1327         }
1328 
1329         ntail = skb->data_len + pad - (skb->end - skb->tail);
1330         if (likely(skb_cloned(skb) || ntail > 0)) {
1331                 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1332                 if (unlikely(err))
1333                         goto free_skb;
1334         }
1335 
1336         /* FIXME: The use of this function with non-linear skb's really needs
1337          * to be audited.
1338          */
1339         err = skb_linearize(skb);
1340         if (unlikely(err))
1341                 goto free_skb;
1342 
1343         memset(skb->data + skb->len, 0, pad);
1344         return 0;
1345 
1346 free_skb:
1347         kfree_skb(skb);
1348         return err;
1349 }
1350 EXPORT_SYMBOL(skb_pad);
1351 
1352 /**
1353  *      pskb_put - add data to the tail of a potentially fragmented buffer
1354  *      @skb: start of the buffer to use
1355  *      @tail: tail fragment of the buffer to use
1356  *      @len: amount of data to add
1357  *
1358  *      This function extends the used data area of the potentially
1359  *      fragmented buffer. @tail must be the last fragment of @skb -- or
1360  *      @skb itself. If this would exceed the total buffer size the kernel
1361  *      will panic. A pointer to the first byte of the extra data is
1362  *      returned.
1363  */
1364 
1365 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1366 {
1367         if (tail != skb) {
1368                 skb->data_len += len;
1369                 skb->len += len;
1370         }
1371         return skb_put(tail, len);
1372 }
1373 EXPORT_SYMBOL_GPL(pskb_put);
1374 
1375 /**
1376  *      skb_put - add data to a buffer
1377  *      @skb: buffer to use
1378  *      @len: amount of data to add
1379  *
1380  *      This function extends the used data area of the buffer. If this would
1381  *      exceed the total buffer size the kernel will panic. A pointer to the
1382  *      first byte of the extra data is returned.
1383  */
1384 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1385 {
1386         unsigned char *tmp = skb_tail_pointer(skb);
1387         SKB_LINEAR_ASSERT(skb);
1388         skb->tail += len;
1389         skb->len  += len;
1390         if (unlikely(skb->tail > skb->end))
1391                 skb_over_panic(skb, len, __builtin_return_address(0));
1392         return tmp;
1393 }
1394 EXPORT_SYMBOL(skb_put);
1395 
1396 /**
1397  *      skb_push - add data to the start of a buffer
1398  *      @skb: buffer to use
1399  *      @len: amount of data to add
1400  *
1401  *      This function extends the used data area of the buffer at the buffer
1402  *      start. If this would exceed the total buffer headroom the kernel will
1403  *      panic. A pointer to the first byte of the extra data is returned.
1404  */
1405 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1406 {
1407         skb->data -= len;
1408         skb->len  += len;
1409         if (unlikely(skb->data<skb->head))
1410                 skb_under_panic(skb, len, __builtin_return_address(0));
1411         return skb->data;
1412 }
1413 EXPORT_SYMBOL(skb_push);
1414 
1415 /**
1416  *      skb_pull - remove data from the start of a buffer
1417  *      @skb: buffer to use
1418  *      @len: amount of data to remove
1419  *
1420  *      This function removes data from the start of a buffer, returning
1421  *      the memory to the headroom. A pointer to the next data in the buffer
1422  *      is returned. Once the data has been pulled future pushes will overwrite
1423  *      the old data.
1424  */
1425 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1426 {
1427         return skb_pull_inline(skb, len);
1428 }
1429 EXPORT_SYMBOL(skb_pull);
1430 
1431 /**
1432  *      skb_trim - remove end from a buffer
1433  *      @skb: buffer to alter
1434  *      @len: new length
1435  *
1436  *      Cut the length of a buffer down by removing data from the tail. If
1437  *      the buffer is already under the length specified it is not modified.
1438  *      The skb must be linear.
1439  */
1440 void skb_trim(struct sk_buff *skb, unsigned int len)
1441 {
1442         if (skb->len > len)
1443                 __skb_trim(skb, len);
1444 }
1445 EXPORT_SYMBOL(skb_trim);
1446 
1447 /* Trims skb to length len. It can change skb pointers.
1448  */
1449 
1450 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1451 {
1452         struct sk_buff **fragp;
1453         struct sk_buff *frag;
1454         int offset = skb_headlen(skb);
1455         int nfrags = skb_shinfo(skb)->nr_frags;
1456         int i;
1457         int err;
1458 
1459         if (skb_cloned(skb) &&
1460             unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1461                 return err;
1462 
1463         i = 0;
1464         if (offset >= len)
1465                 goto drop_pages;
1466 
1467         for (; i < nfrags; i++) {
1468                 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1469 
1470                 if (end < len) {
1471                         offset = end;
1472                         continue;
1473                 }
1474 
1475                 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1476 
1477 drop_pages:
1478                 skb_shinfo(skb)->nr_frags = i;
1479 
1480                 for (; i < nfrags; i++)
1481                         skb_frag_unref(skb, i);
1482 
1483                 if (skb_has_frag_list(skb))
1484                         skb_drop_fraglist(skb);
1485                 goto done;
1486         }
1487 
1488         for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1489              fragp = &frag->next) {
1490                 int end = offset + frag->len;
1491 
1492                 if (skb_shared(frag)) {
1493                         struct sk_buff *nfrag;
1494 
1495                         nfrag = skb_clone(frag, GFP_ATOMIC);
1496                         if (unlikely(!nfrag))
1497                                 return -ENOMEM;
1498 
1499                         nfrag->next = frag->next;
1500                         consume_skb(frag);
1501                         frag = nfrag;
1502                         *fragp = frag;
1503                 }
1504 
1505                 if (end < len) {
1506                         offset = end;
1507                         continue;
1508                 }
1509 
1510                 if (end > len &&
1511                     unlikely((err = pskb_trim(frag, len - offset))))
1512                         return err;
1513 
1514                 if (frag->next)
1515                         skb_drop_list(&frag->next);
1516                 break;
1517         }
1518 
1519 done:
1520         if (len > skb_headlen(skb)) {
1521                 skb->data_len -= skb->len - len;
1522                 skb->len       = len;
1523         } else {
1524                 skb->len       = len;
1525                 skb->data_len  = 0;
1526                 skb_set_tail_pointer(skb, len);
1527         }
1528 
1529         return 0;
1530 }
1531 EXPORT_SYMBOL(___pskb_trim);
1532 
1533 /**
1534  *      __pskb_pull_tail - advance tail of skb header
1535  *      @skb: buffer to reallocate
1536  *      @delta: number of bytes to advance tail
1537  *
1538  *      The function makes a sense only on a fragmented &sk_buff,
1539  *      it expands header moving its tail forward and copying necessary
1540  *      data from fragmented part.
1541  *
1542  *      &sk_buff MUST have reference count of 1.
1543  *
1544  *      Returns %NULL (and &sk_buff does not change) if pull failed
1545  *      or value of new tail of skb in the case of success.
1546  *
1547  *      All the pointers pointing into skb header may change and must be
1548  *      reloaded after call to this function.
1549  */
1550 
1551 /* Moves tail of skb head forward, copying data from fragmented part,
1552  * when it is necessary.
1553  * 1. It may fail due to malloc failure.
1554  * 2. It may change skb pointers.
1555  *
1556  * It is pretty complicated. Luckily, it is called only in exceptional cases.
1557  */
1558 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1559 {
1560         /* If skb has not enough free space at tail, get new one
1561          * plus 128 bytes for future expansions. If we have enough
1562          * room at tail, reallocate without expansion only if skb is cloned.
1563          */
1564         int i, k, eat = (skb->tail + delta) - skb->end;
1565 
1566         if (eat > 0 || skb_cloned(skb)) {
1567                 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1568                                      GFP_ATOMIC))
1569                         return NULL;
1570         }
1571 
1572         if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1573                 BUG();
1574 
1575         /* Optimization: no fragments, no reasons to preestimate
1576          * size of pulled pages. Superb.
1577          */
1578         if (!skb_has_frag_list(skb))
1579                 goto pull_pages;
1580 
1581         /* Estimate size of pulled pages. */
1582         eat = delta;
1583         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1584                 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1585 
1586                 if (size >= eat)
1587                         goto pull_pages;
1588                 eat -= size;
1589         }
1590 
1591         /* If we need update frag list, we are in troubles.
1592          * Certainly, it possible to add an offset to skb data,
1593          * but taking into account that pulling is expected to
1594          * be very rare operation, it is worth to fight against
1595          * further bloating skb head and crucify ourselves here instead.
1596          * Pure masohism, indeed. 8)8)
1597          */
1598         if (eat) {
1599                 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1600                 struct sk_buff *clone = NULL;
1601                 struct sk_buff *insp = NULL;
1602 
1603                 do {
1604                         BUG_ON(!list);
1605 
1606                         if (list->len <= eat) {
1607                                 /* Eaten as whole. */
1608                                 eat -= list->len;
1609                                 list = list->next;
1610                                 insp = list;
1611                         } else {
1612                                 /* Eaten partially. */
1613 
1614                                 if (skb_shared(list)) {
1615                                         /* Sucks! We need to fork list. :-( */
1616                                         clone = skb_clone(list, GFP_ATOMIC);
1617                                         if (!clone)
1618                                                 return NULL;
1619                                         insp = list->next;
1620                                         list = clone;
1621                                 } else {
1622                                         /* This may be pulled without
1623                                          * problems. */
1624                                         insp = list;
1625                                 }
1626                                 if (!pskb_pull(list, eat)) {
1627                                         kfree_skb(clone);
1628                                         return NULL;
1629                                 }
1630                                 break;
1631                         }
1632                 } while (eat);
1633 
1634                 /* Free pulled out fragments. */
1635                 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1636                         skb_shinfo(skb)->frag_list = list->next;
1637                         kfree_skb(list);
1638                 }
1639                 /* And insert new clone at head. */
1640                 if (clone) {
1641                         clone->next = list;
1642                         skb_shinfo(skb)->frag_list = clone;
1643                 }
1644         }
1645         /* Success! Now we may commit changes to skb data. */
1646 
1647 pull_pages:
1648         eat = delta;
1649         k = 0;
1650         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1651                 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1652 
1653                 if (size <= eat) {
1654                         skb_frag_unref(skb, i);
1655                         eat -= size;
1656                 } else {
1657                         skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1658                         if (eat) {
1659                                 skb_shinfo(skb)->frags[k].page_offset += eat;
1660                                 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1661                                 eat = 0;
1662                         }
1663                         k++;
1664                 }
1665         }
1666         skb_shinfo(skb)->nr_frags = k;
1667 
1668         skb->tail     += delta;
1669         skb->data_len -= delta;
1670 
1671         return skb_tail_pointer(skb);
1672 }
1673 EXPORT_SYMBOL(__pskb_pull_tail);
1674 
1675 /**
1676  *      skb_copy_bits - copy bits from skb to kernel buffer
1677  *      @skb: source skb
1678  *      @offset: offset in source
1679  *      @to: destination buffer
1680  *      @len: number of bytes to copy
1681  *
1682  *      Copy the specified number of bytes from the source skb to the
1683  *      destination buffer.
1684  *
1685  *      CAUTION ! :
1686  *              If its prototype is ever changed,
1687  *              check arch/{*}/net/{*}.S files,
1688  *              since it is called from BPF assembly code.
1689  */
1690 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1691 {
1692         int start = skb_headlen(skb);
1693         struct sk_buff *frag_iter;
1694         int i, copy;
1695 
1696         if (offset > (int)skb->len - len)
1697                 goto fault;
1698 
1699         /* Copy header. */
1700         if ((copy = start - offset) > 0) {
1701                 if (copy > len)
1702                         copy = len;
1703                 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1704                 if ((len -= copy) == 0)
1705                         return 0;
1706                 offset += copy;
1707                 to     += copy;
1708         }
1709 
1710         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1711                 int end;
1712                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1713 
1714                 WARN_ON(start > offset + len);
1715 
1716                 end = start + skb_frag_size(f);
1717                 if ((copy = end - offset) > 0) {
1718                         u8 *vaddr;
1719 
1720                         if (copy > len)
1721                                 copy = len;
1722 
1723                         vaddr = kmap_atomic(skb_frag_page(f));
1724                         memcpy(to,
1725                                vaddr + f->page_offset + offset - start,
1726                                copy);
1727                         kunmap_atomic(vaddr);
1728 
1729                         if ((len -= copy) == 0)
1730                                 return 0;
1731                         offset += copy;
1732                         to     += copy;
1733                 }
1734                 start = end;
1735         }
1736 
1737         skb_walk_frags(skb, frag_iter) {
1738                 int end;
1739 
1740                 WARN_ON(start > offset + len);
1741 
1742                 end = start + frag_iter->len;
1743                 if ((copy = end - offset) > 0) {
1744                         if (copy > len)
1745                                 copy = len;
1746                         if (skb_copy_bits(frag_iter, offset - start, to, copy))
1747                                 goto fault;
1748                         if ((len -= copy) == 0)
1749                                 return 0;
1750                         offset += copy;
1751                         to     += copy;
1752                 }
1753                 start = end;
1754         }
1755 
1756         if (!len)
1757                 return 0;
1758 
1759 fault:
1760         return -EFAULT;
1761 }
1762 EXPORT_SYMBOL(skb_copy_bits);
1763 
1764 /*
1765  * Callback from splice_to_pipe(), if we need to release some pages
1766  * at the end of the spd in case we error'ed out in filling the pipe.
1767  */
1768 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1769 {
1770         put_page(spd->pages[i]);
1771 }
1772 
1773 static struct page *linear_to_page(struct page *page, unsigned int *len,
1774                                    unsigned int *offset,
1775                                    struct sock *sk)
1776 {
1777         struct page_frag *pfrag = sk_page_frag(sk);
1778 
1779         if (!sk_page_frag_refill(sk, pfrag))
1780                 return NULL;
1781 
1782         *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
1783 
1784         memcpy(page_address(pfrag->page) + pfrag->offset,
1785                page_address(page) + *offset, *len);
1786         *offset = pfrag->offset;
1787         pfrag->offset += *len;
1788 
1789         return pfrag->page;
1790 }
1791 
1792 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1793                              struct page *page,
1794                              unsigned int offset)
1795 {
1796         return  spd->nr_pages &&
1797                 spd->pages[spd->nr_pages - 1] == page &&
1798                 (spd->partial[spd->nr_pages - 1].offset +
1799                  spd->partial[spd->nr_pages - 1].len == offset);
1800 }
1801 
1802 /*
1803  * Fill page/offset/length into spd, if it can hold more pages.
1804  */
1805 static bool spd_fill_page(struct splice_pipe_desc *spd,
1806                           struct pipe_inode_info *pipe, struct page *page,
1807                           unsigned int *len, unsigned int offset,
1808                           bool linear,
1809                           struct sock *sk)
1810 {
1811         if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1812                 return true;
1813 
1814         if (linear) {
1815                 page = linear_to_page(page, len, &offset, sk);
1816                 if (!page)
1817                         return true;
1818         }
1819         if (spd_can_coalesce(spd, page, offset)) {
1820                 spd->partial[spd->nr_pages - 1].len += *len;
1821                 return false;
1822         }
1823         get_page(page);
1824         spd->pages[spd->nr_pages] = page;
1825         spd->partial[spd->nr_pages].len = *len;
1826         spd->partial[spd->nr_pages].offset = offset;
1827         spd->nr_pages++;
1828 
1829         return false;
1830 }
1831 
1832 static bool __splice_segment(struct page *page, unsigned int poff,
1833                              unsigned int plen, unsigned int *off,
1834                              unsigned int *len,
1835                              struct splice_pipe_desc *spd, bool linear,
1836                              struct sock *sk,
1837                              struct pipe_inode_info *pipe)
1838 {
1839         if (!*len)
1840                 return true;
1841 
1842         /* skip this segment if already processed */
1843         if (*off >= plen) {
1844                 *off -= plen;
1845                 return false;
1846         }
1847 
1848         /* ignore any bits we already processed */
1849         poff += *off;
1850         plen -= *off;
1851         *off = 0;
1852 
1853         do {
1854                 unsigned int flen = min(*len, plen);
1855 
1856                 if (spd_fill_page(spd, pipe, page, &flen, poff,
1857                                   linear, sk))
1858                         return true;
1859                 poff += flen;
1860                 plen -= flen;
1861                 *len -= flen;
1862         } while (*len && plen);
1863 
1864         return false;
1865 }
1866 
1867 /*
1868  * Map linear and fragment data from the skb to spd. It reports true if the
1869  * pipe is full or if we already spliced the requested length.
1870  */
1871 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1872                               unsigned int *offset, unsigned int *len,
1873                               struct splice_pipe_desc *spd, struct sock *sk)
1874 {
1875         int seg;
1876 
1877         /* map the linear part :
1878          * If skb->head_frag is set, this 'linear' part is backed by a
1879          * fragment, and if the head is not shared with any clones then
1880          * we can avoid a copy since we own the head portion of this page.
1881          */
1882         if (__splice_segment(virt_to_page(skb->data),
1883                              (unsigned long) skb->data & (PAGE_SIZE - 1),
1884                              skb_headlen(skb),
1885                              offset, len, spd,
1886                              skb_head_is_locked(skb),
1887                              sk, pipe))
1888                 return true;
1889 
1890         /*
1891          * then map the fragments
1892          */
1893         for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1894                 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1895 
1896                 if (__splice_segment(skb_frag_page(f),
1897                                      f->page_offset, skb_frag_size(f),
1898                                      offset, len, spd, false, sk, pipe))
1899                         return true;
1900         }
1901 
1902         return false;
1903 }
1904 
1905 /*
1906  * Map data from the skb to a pipe. Should handle both the linear part,
1907  * the fragments, and the frag list. It does NOT handle frag lists within
1908  * the frag list, if such a thing exists. We'd probably need to recurse to
1909  * handle that cleanly.
1910  */
1911 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1912                     struct pipe_inode_info *pipe, unsigned int tlen,
1913                     unsigned int flags)
1914 {
1915         struct partial_page partial[MAX_SKB_FRAGS];
1916         struct page *pages[MAX_SKB_FRAGS];
1917         struct splice_pipe_desc spd = {
1918                 .pages = pages,
1919                 .partial = partial,
1920                 .nr_pages_max = MAX_SKB_FRAGS,
1921                 .flags = flags,
1922                 .ops = &nosteal_pipe_buf_ops,
1923                 .spd_release = sock_spd_release,
1924         };
1925         struct sk_buff *frag_iter;
1926         struct sock *sk = skb->sk;
1927         int ret = 0;
1928 
1929         /*
1930          * __skb_splice_bits() only fails if the output has no room left,
1931          * so no point in going over the frag_list for the error case.
1932          */
1933         if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1934                 goto done;
1935         else if (!tlen)
1936                 goto done;
1937 
1938         /*
1939          * now see if we have a frag_list to map
1940          */
1941         skb_walk_frags(skb, frag_iter) {
1942                 if (!tlen)
1943                         break;
1944                 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1945                         break;
1946         }
1947 
1948 done:
1949         if (spd.nr_pages) {
1950                 /*
1951                  * Drop the socket lock, otherwise we have reverse
1952                  * locking dependencies between sk_lock and i_mutex
1953                  * here as compared to sendfile(). We enter here
1954                  * with the socket lock held, and splice_to_pipe() will
1955                  * grab the pipe inode lock. For sendfile() emulation,
1956                  * we call into ->sendpage() with the i_mutex lock held
1957                  * and networking will grab the socket lock.
1958                  */
1959                 release_sock(sk);
1960                 ret = splice_to_pipe(pipe, &spd);
1961                 lock_sock(sk);
1962         }
1963 
1964         return ret;
1965 }
1966 
1967 /**
1968  *      skb_store_bits - store bits from kernel buffer to skb
1969  *      @skb: destination buffer
1970  *      @offset: offset in destination
1971  *      @from: source buffer
1972  *      @len: number of bytes to copy
1973  *
1974  *      Copy the specified number of bytes from the source buffer to the
1975  *      destination skb.  This function handles all the messy bits of
1976  *      traversing fragment lists and such.
1977  */
1978 
1979 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1980 {
1981         int start = skb_headlen(skb);
1982         struct sk_buff *frag_iter;
1983         int i, copy;
1984 
1985         if (offset > (int)skb->len - len)
1986                 goto fault;
1987 
1988         if ((copy = start - offset) > 0) {
1989                 if (copy > len)
1990                         copy = len;
1991                 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1992                 if ((len -= copy) == 0)
1993                         return 0;
1994                 offset += copy;
1995                 from += copy;
1996         }
1997 
1998         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1999                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2000                 int end;
2001 
2002                 WARN_ON(start > offset + len);
2003 
2004                 end = start + skb_frag_size(frag);
2005                 if ((copy = end - offset) > 0) {
2006                         u8 *vaddr;
2007 
2008                         if (copy > len)
2009                                 copy = len;
2010 
2011                         vaddr = kmap_atomic(skb_frag_page(frag));
2012                         memcpy(vaddr + frag->page_offset + offset - start,
2013                                from, copy);
2014                         kunmap_atomic(vaddr);
2015 
2016                         if ((len -= copy) == 0)
2017                                 return 0;
2018                         offset += copy;
2019                         from += copy;
2020                 }
2021                 start = end;
2022         }
2023 
2024         skb_walk_frags(skb, frag_iter) {
2025                 int end;
2026 
2027                 WARN_ON(start > offset + len);
2028 
2029                 end = start + frag_iter->len;
2030                 if ((copy = end - offset) > 0) {
2031                         if (copy > len)
2032                                 copy = len;
2033                         if (skb_store_bits(frag_iter, offset - start,
2034                                            from, copy))
2035                                 goto fault;
2036                         if ((len -= copy) == 0)
2037                                 return 0;
2038                         offset += copy;
2039                         from += copy;
2040                 }
2041                 start = end;
2042         }
2043         if (!len)
2044                 return 0;
2045 
2046 fault:
2047         return -EFAULT;
2048 }
2049 EXPORT_SYMBOL(skb_store_bits);
2050 
2051 /* Checksum skb data. */
2052 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2053                       __wsum csum, const struct skb_checksum_ops *ops)
2054 {
2055         int start = skb_headlen(skb);
2056         int i, copy = start - offset;
2057         struct sk_buff *frag_iter;
2058         int pos = 0;
2059 
2060         /* Checksum header. */
2061         if (copy > 0) {
2062                 if (copy > len)
2063                         copy = len;
2064                 csum = ops->update(skb->data + offset, copy, csum);
2065                 if ((len -= copy) == 0)
2066                         return csum;
2067                 offset += copy;
2068                 pos     = copy;
2069         }
2070 
2071         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2072                 int end;
2073                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2074 
2075                 WARN_ON(start > offset + len);
2076 
2077                 end = start + skb_frag_size(frag);
2078                 if ((copy = end - offset) > 0) {
2079                         __wsum csum2;
2080                         u8 *vaddr;
2081 
2082                         if (copy > len)
2083                                 copy = len;
2084                         vaddr = kmap_atomic(skb_frag_page(frag));
2085                         csum2 = ops->update(vaddr + frag->page_offset +
2086                                             offset - start, copy, 0);
2087                         kunmap_atomic(vaddr);
2088                         csum = ops->combine(csum, csum2, pos, copy);
2089                         if (!(len -= copy))
2090                                 return csum;
2091                         offset += copy;
2092                         pos    += copy;
2093                 }
2094                 start = end;
2095         }
2096 
2097         skb_walk_frags(skb, frag_iter) {
2098                 int end;
2099 
2100                 WARN_ON(start > offset + len);
2101 
2102                 end = start + frag_iter->len;
2103                 if ((copy = end - offset) > 0) {
2104                         __wsum csum2;
2105                         if (copy > len)
2106                                 copy = len;
2107                         csum2 = __skb_checksum(frag_iter, offset - start,
2108                                                copy, 0, ops);
2109                         csum = ops->combine(csum, csum2, pos, copy);
2110                         if ((len -= copy) == 0)
2111                                 return csum;
2112                         offset += copy;
2113                         pos    += copy;
2114                 }
2115                 start = end;
2116         }
2117         BUG_ON(len);
2118 
2119         return csum;
2120 }
2121 EXPORT_SYMBOL(__skb_checksum);
2122 
2123 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2124                     int len, __wsum csum)
2125 {
2126         const struct skb_checksum_ops ops = {
2127                 .update  = csum_partial_ext,
2128                 .combine = csum_block_add_ext,
2129         };
2130 
2131         return __skb_checksum(skb, offset, len, csum, &ops);
2132 }
2133 EXPORT_SYMBOL(skb_checksum);
2134 
2135 /* Both of above in one bottle. */
2136 
2137 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2138                                     u8 *to, int len, __wsum csum)
2139 {
2140         int start = skb_headlen(skb);
2141         int i, copy = start - offset;
2142         struct sk_buff *frag_iter;
2143         int pos = 0;
2144 
2145         /* Copy header. */
2146         if (copy > 0) {
2147                 if (copy > len)
2148                         copy = len;
2149                 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2150                                                  copy, csum);
2151                 if ((len -= copy) == 0)
2152                         return csum;
2153                 offset += copy;
2154                 to     += copy;
2155                 pos     = copy;
2156         }
2157 
2158         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2159                 int end;
2160 
2161                 WARN_ON(start > offset + len);
2162 
2163                 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2164                 if ((copy = end - offset) > 0) {
2165                         __wsum csum2;
2166                         u8 *vaddr;
2167                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2168 
2169                         if (copy > len)
2170                                 copy = len;
2171                         vaddr = kmap_atomic(skb_frag_page(frag));
2172                         csum2 = csum_partial_copy_nocheck(vaddr +
2173                                                           frag->page_offset +
2174                                                           offset - start, to,
2175                                                           copy, 0);
2176                         kunmap_atomic(vaddr);
2177                         csum = csum_block_add(csum, csum2, pos);
2178                         if (!(len -= copy))
2179                                 return csum;
2180                         offset += copy;
2181                         to     += copy;
2182                         pos    += copy;
2183                 }
2184                 start = end;
2185         }
2186 
2187         skb_walk_frags(skb, frag_iter) {
2188                 __wsum csum2;
2189                 int end;
2190 
2191                 WARN_ON(start > offset + len);
2192 
2193                 end = start + frag_iter->len;
2194                 if ((copy = end - offset) > 0) {
2195                         if (copy > len)
2196                                 copy = len;
2197                         csum2 = skb_copy_and_csum_bits(frag_iter,
2198                                                        offset - start,
2199                                                        to, copy, 0);
2200                         csum = csum_block_add(csum, csum2, pos);
2201                         if ((len -= copy) == 0)
2202                                 return csum;
2203                         offset += copy;
2204                         to     += copy;
2205                         pos    += copy;
2206                 }
2207                 start = end;
2208         }
2209         BUG_ON(len);
2210         return csum;
2211 }
2212 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2213 
2214  /**
2215  *      skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2216  *      @from: source buffer
2217  *
2218  *      Calculates the amount of linear headroom needed in the 'to' skb passed
2219  *      into skb_zerocopy().
2220  */
2221 unsigned int
2222 skb_zerocopy_headlen(const struct sk_buff *from)
2223 {
2224         unsigned int hlen = 0;
2225 
2226         if (!from->head_frag ||
2227             skb_headlen(from) < L1_CACHE_BYTES ||
2228             skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2229                 hlen = skb_headlen(from);
2230 
2231         if (skb_has_frag_list(from))
2232                 hlen = from->len;
2233 
2234         return hlen;
2235 }
2236 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2237 
2238 /**
2239  *      skb_zerocopy - Zero copy skb to skb
2240  *      @to: destination buffer
2241  *      @from: source buffer
2242  *      @len: number of bytes to copy from source buffer
2243  *      @hlen: size of linear headroom in destination buffer
2244  *
2245  *      Copies up to `len` bytes from `from` to `to` by creating references
2246  *      to the frags in the source buffer.
2247  *
2248  *      The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2249  *      headroom in the `to` buffer.
2250  *
2251  *      Return value:
2252  *      0: everything is OK
2253  *      -ENOMEM: couldn't orphan frags of @from due to lack of memory
2254  *      -EFAULT: skb_copy_bits() found some problem with skb geometry
2255  */
2256 int
2257 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2258 {
2259         int i, j = 0;
2260         int plen = 0; /* length of skb->head fragment */
2261         int ret;
2262         struct page *page;
2263         unsigned int offset;
2264 
2265         BUG_ON(!from->head_frag && !hlen);
2266 
2267         /* dont bother with small payloads */
2268         if (len <= skb_tailroom(to))
2269                 return skb_copy_bits(from, 0, skb_put(to, len), len);
2270 
2271         if (hlen) {
2272                 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2273                 if (unlikely(ret))
2274                         return ret;
2275                 len -= hlen;
2276         } else {
2277                 plen = min_t(int, skb_headlen(from), len);
2278                 if (plen) {
2279                         page = virt_to_head_page(from->head);
2280                         offset = from->data - (unsigned char *)page_address(page);
2281                         __skb_fill_page_desc(to, 0, page, offset, plen);
2282                         get_page(page);
2283                         j = 1;
2284                         len -= plen;
2285                 }
2286         }
2287 
2288         to->truesize += len + plen;
2289         to->len += len + plen;
2290         to->data_len += len + plen;
2291 
2292         if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2293                 skb_tx_error(from);
2294                 return -ENOMEM;
2295         }
2296 
2297         for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2298                 if (!len)
2299                         break;
2300                 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2301                 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
2302                 len -= skb_shinfo(to)->frags[j].size;
2303                 skb_frag_ref(to, j);
2304                 j++;
2305         }
2306         skb_shinfo(to)->nr_frags = j;
2307 
2308         return 0;
2309 }
2310 EXPORT_SYMBOL_GPL(skb_zerocopy);
2311 
2312 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2313 {
2314         __wsum csum;
2315         long csstart;
2316 
2317         if (skb->ip_summed == CHECKSUM_PARTIAL)
2318                 csstart = skb_checksum_start_offset(skb);
2319         else
2320                 csstart = skb_headlen(skb);
2321 
2322         BUG_ON(csstart > skb_headlen(skb));
2323 
2324         skb_copy_from_linear_data(skb, to, csstart);
2325 
2326         csum = 0;
2327         if (csstart != skb->len)
2328                 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2329                                               skb->len - csstart, 0);
2330 
2331         if (skb->ip_summed == CHECKSUM_PARTIAL) {
2332                 long csstuff = csstart + skb->csum_offset;
2333 
2334                 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2335         }
2336 }
2337 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2338 
2339 /**
2340  *      skb_dequeue - remove from the head of the queue
2341  *      @list: list to dequeue from
2342  *
2343  *      Remove the head of the list. The list lock is taken so the function
2344  *      may be used safely with other locking list functions. The head item is
2345  *      returned or %NULL if the list is empty.
2346  */
2347 
2348 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2349 {
2350         unsigned long flags;
2351         struct sk_buff *result;
2352 
2353         spin_lock_irqsave(&list->lock, flags);
2354         result = __skb_dequeue(list);
2355         spin_unlock_irqrestore(&list->lock, flags);
2356         return result;
2357 }
2358 EXPORT_SYMBOL(skb_dequeue);
2359 
2360 /**
2361  *      skb_dequeue_tail - remove from the tail of the queue
2362  *      @list: list to dequeue from
2363  *
2364  *      Remove the tail of the list. The list lock is taken so the function
2365  *      may be used safely with other locking list functions. The tail item is
2366  *      returned or %NULL if the list is empty.
2367  */
2368 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2369 {
2370         unsigned long flags;
2371         struct sk_buff *result;
2372 
2373         spin_lock_irqsave(&list->lock, flags);
2374         result = __skb_dequeue_tail(list);
2375         spin_unlock_irqrestore(&list->lock, flags);
2376         return result;
2377 }
2378 EXPORT_SYMBOL(skb_dequeue_tail);
2379 
2380 /**
2381  *      skb_queue_purge - empty a list
2382  *      @list: list to empty
2383  *
2384  *      Delete all buffers on an &sk_buff list. Each buffer is removed from
2385  *      the list and one reference dropped. This function takes the list
2386  *      lock and is atomic with respect to other list locking functions.
2387  */
2388 void skb_queue_purge(struct sk_buff_head *list)
2389 {
2390         struct sk_buff *skb;
2391         while ((skb = skb_dequeue(list)) != NULL)
2392                 kfree_skb(skb);
2393 }
2394 EXPORT_SYMBOL(skb_queue_purge);
2395 
2396 /**
2397  *      skb_queue_head - queue a buffer at the list head
2398  *      @list: list to use
2399  *      @newsk: buffer to queue
2400  *
2401  *      Queue a buffer at the start of the list. This function takes the
2402  *      list lock and can be used safely with other locking &sk_buff functions
2403  *      safely.
2404  *
2405  *      A buffer cannot be placed on two lists at the same time.
2406  */
2407 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2408 {
2409         unsigned long flags;
2410 
2411         spin_lock_irqsave(&list->lock, flags);
2412         __skb_queue_head(list, newsk);
2413         spin_unlock_irqrestore(&list->lock, flags);
2414 }
2415 EXPORT_SYMBOL(skb_queue_head);
2416 
2417 /**
2418  *      skb_queue_tail - queue a buffer at the list tail
2419  *      @list: list to use
2420  *      @newsk: buffer to queue
2421  *
2422  *      Queue a buffer at the tail of the list. This function takes the
2423  *      list lock and can be used safely with other locking &sk_buff functions
2424  *      safely.
2425  *
2426  *      A buffer cannot be placed on two lists at the same time.
2427  */
2428 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2429 {
2430         unsigned long flags;
2431 
2432         spin_lock_irqsave(&list->lock, flags);
2433         __skb_queue_tail(list, newsk);
2434         spin_unlock_irqrestore(&list->lock, flags);
2435 }
2436 EXPORT_SYMBOL(skb_queue_tail);
2437 
2438 /**
2439  *      skb_unlink      -       remove a buffer from a list
2440  *      @skb: buffer to remove
2441  *      @list: list to use
2442  *
2443  *      Remove a packet from a list. The list locks are taken and this
2444  *      function is atomic with respect to other list locked calls
2445  *
2446  *      You must know what list the SKB is on.
2447  */
2448 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2449 {
2450         unsigned long flags;
2451 
2452         spin_lock_irqsave(&list->lock, flags);
2453         __skb_unlink(skb, list);
2454         spin_unlock_irqrestore(&list->lock, flags);
2455 }
2456 EXPORT_SYMBOL(skb_unlink);
2457 
2458 /**
2459  *      skb_append      -       append a buffer
2460  *      @old: buffer to insert after
2461  *      @newsk: buffer to insert
2462  *      @list: list to use
2463  *
2464  *      Place a packet after a given packet in a list. The list locks are taken
2465  *      and this function is atomic with respect to other list locked calls.
2466  *      A buffer cannot be placed on two lists at the same time.
2467  */
2468 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2469 {
2470         unsigned long flags;
2471 
2472         spin_lock_irqsave(&list->lock, flags);
2473         __skb_queue_after(list, old, newsk);
2474         spin_unlock_irqrestore(&list->lock, flags);
2475 }
2476 EXPORT_SYMBOL(skb_append);
2477 
2478 /**
2479  *      skb_insert      -       insert a buffer
2480  *      @old: buffer to insert before
2481  *      @newsk: buffer to insert
2482  *      @list: list to use
2483  *
2484  *      Place a packet before a given packet in a list. The list locks are
2485  *      taken and this function is atomic with respect to other list locked
2486  *      calls.
2487  *
2488  *      A buffer cannot be placed on two lists at the same time.
2489  */
2490 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2491 {
2492         unsigned long flags;
2493 
2494         spin_lock_irqsave(&list->lock, flags);
2495         __skb_insert(newsk, old->prev, old, list);
2496         spin_unlock_irqrestore(&list->lock, flags);
2497 }
2498 EXPORT_SYMBOL(skb_insert);
2499 
2500 static inline void skb_split_inside_header(struct sk_buff *skb,
2501                                            struct sk_buff* skb1,
2502                                            const u32 len, const int pos)
2503 {
2504         int i;
2505 
2506         skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2507                                          pos - len);
2508         /* And move data appendix as is. */
2509         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2510                 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2511 
2512         skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2513         skb_shinfo(skb)->nr_frags  = 0;
2514         skb1->data_len             = skb->data_len;
2515         skb1->len                  += skb1->data_len;
2516         skb->data_len              = 0;
2517         skb->len                   = len;
2518         skb_set_tail_pointer(skb, len);
2519 }
2520 
2521 static inline void skb_split_no_header(struct sk_buff *skb,
2522                                        struct sk_buff* skb1,
2523                                        const u32 len, int pos)
2524 {
2525         int i, k = 0;
2526         const int nfrags = skb_shinfo(skb)->nr_frags;
2527 
2528         skb_shinfo(skb)->nr_frags = 0;
2529         skb1->len                 = skb1->data_len = skb->len - len;
2530         skb->len                  = len;
2531         skb->data_len             = len - pos;
2532 
2533         for (i = 0; i < nfrags; i++) {
2534                 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2535 
2536                 if (pos + size > len) {
2537                         skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2538 
2539                         if (pos < len) {
2540                                 /* Split frag.
2541                                  * We have two variants in this case:
2542                                  * 1. Move all the frag to the second
2543                                  *    part, if it is possible. F.e.
2544                                  *    this approach is mandatory for TUX,
2545                                  *    where splitting is expensive.
2546                                  * 2. Split is accurately. We make this.
2547                                  */
2548                                 skb_frag_ref(skb, i);
2549                                 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2550                                 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2551                                 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2552                                 skb_shinfo(skb)->nr_frags++;
2553                         }
2554                         k++;
2555                 } else
2556                         skb_shinfo(skb)->nr_frags++;
2557                 pos += size;
2558         }
2559         skb_shinfo(skb1)->nr_frags = k;
2560 }
2561 
2562 /**
2563  * skb_split - Split fragmented skb to two parts at length len.
2564  * @skb: the buffer to split
2565  * @skb1: the buffer to receive the second part
2566  * @len: new length for skb
2567  */
2568 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2569 {
2570         int pos = skb_headlen(skb);
2571 
2572         skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2573         if (len < pos)  /* Split line is inside header. */
2574                 skb_split_inside_header(skb, skb1, len, pos);
2575         else            /* Second chunk has no header, nothing to copy. */
2576                 skb_split_no_header(skb, skb1, len, pos);
2577 }
2578 EXPORT_SYMBOL(skb_split);
2579 
2580 /* Shifting from/to a cloned skb is a no-go.
2581  *
2582  * Caller cannot keep skb_shinfo related pointers past calling here!
2583  */
2584 static int skb_prepare_for_shift(struct sk_buff *skb)
2585 {
2586         return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2587 }
2588 
2589 /**
2590  * skb_shift - Shifts paged data partially from skb to another
2591  * @tgt: buffer into which tail data gets added
2592  * @skb: buffer from which the paged data comes from
2593  * @shiftlen: shift up to this many bytes
2594  *
2595  * Attempts to shift up to shiftlen worth of bytes, which may be less than
2596  * the length of the skb, from skb to tgt. Returns number bytes shifted.
2597  * It's up to caller to free skb if everything was shifted.
2598  *
2599  * If @tgt runs out of frags, the whole operation is aborted.
2600  *
2601  * Skb cannot include anything else but paged data while tgt is allowed
2602  * to have non-paged data as well.
2603  *
2604  * TODO: full sized shift could be optimized but that would need
2605  * specialized skb free'er to handle frags without up-to-date nr_frags.
2606  */
2607 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2608 {
2609         int from, to, merge, todo;
2610         struct skb_frag_struct *fragfrom, *fragto;
2611 
2612         BUG_ON(shiftlen > skb->len);
2613         BUG_ON(skb_headlen(skb));       /* Would corrupt stream */
2614 
2615         todo = shiftlen;
2616         from = 0;
2617         to = skb_shinfo(tgt)->nr_frags;
2618         fragfrom = &skb_shinfo(skb)->frags[from];
2619 
2620         /* Actual merge is delayed until the point when we know we can
2621          * commit all, so that we don't have to undo partial changes
2622          */
2623         if (!to ||
2624             !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2625                               fragfrom->page_offset)) {
2626                 merge = -1;
2627         } else {
2628                 merge = to - 1;
2629 
2630                 todo -= skb_frag_size(fragfrom);
2631                 if (todo < 0) {
2632                         if (skb_prepare_for_shift(skb) ||
2633                             skb_prepare_for_shift(tgt))
2634                                 return 0;
2635 
2636                         /* All previous frag pointers might be stale! */
2637                         fragfrom = &skb_shinfo(skb)->frags[from];
2638                         fragto = &skb_shinfo(tgt)->frags[merge];
2639 
2640                         skb_frag_size_add(fragto, shiftlen);
2641                         skb_frag_size_sub(fragfrom, shiftlen);
2642                         fragfrom->page_offset += shiftlen;
2643 
2644                         goto onlymerged;
2645                 }
2646 
2647                 from++;
2648         }
2649 
2650         /* Skip full, not-fitting skb to avoid expensive operations */
2651         if ((shiftlen == skb->len) &&
2652             (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2653                 return 0;
2654 
2655         if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2656                 return 0;
2657 
2658         while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2659                 if (to == MAX_SKB_FRAGS)
2660                         return 0;
2661 
2662                 fragfrom = &skb_shinfo(skb)->frags[from];
2663                 fragto = &skb_shinfo(tgt)->frags[to];
2664 
2665                 if (todo >= skb_frag_size(fragfrom)) {
2666                         *fragto = *fragfrom;
2667                         todo -= skb_frag_size(fragfrom);
2668                         from++;
2669                         to++;
2670 
2671                 } else {
2672                         __skb_frag_ref(fragfrom);
2673                         fragto->page = fragfrom->page;
2674                         fragto->page_offset = fragfrom->page_offset;
2675                         skb_frag_size_set(fragto, todo);
2676 
2677                         fragfrom->page_offset += todo;
2678                         skb_frag_size_sub(fragfrom, todo);
2679                         todo = 0;
2680 
2681                         to++;
2682                         break;
2683                 }
2684         }
2685 
2686         /* Ready to "commit" this state change to tgt */
2687         skb_shinfo(tgt)->nr_frags = to;
2688 
2689         if (merge >= 0) {
2690                 fragfrom = &skb_shinfo(skb)->frags[0];
2691                 fragto = &skb_shinfo(tgt)->frags[merge];
2692 
2693                 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2694                 __skb_frag_unref(fragfrom);
2695         }
2696 
2697         /* Reposition in the original skb */
2698         to = 0;
2699         while (from < skb_shinfo(skb)->nr_frags)
2700                 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2701         skb_shinfo(skb)->nr_frags = to;
2702 
2703         BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2704 
2705 onlymerged:
2706         /* Most likely the tgt won't ever need its checksum anymore, skb on
2707          * the other hand might need it if it needs to be resent
2708          */
2709         tgt->ip_summed = CHECKSUM_PARTIAL;
2710         skb->ip_summed = CHECKSUM_PARTIAL;
2711 
2712         /* Yak, is it really working this way? Some helper please? */
2713         skb->len -= shiftlen;
2714         skb->data_len -= shiftlen;
2715         skb->truesize -= shiftlen;
2716         tgt->len += shiftlen;
2717         tgt->data_len += shiftlen;
2718         tgt->truesize += shiftlen;
2719 
2720         return shiftlen;
2721 }
2722 
2723 /**
2724  * skb_prepare_seq_read - Prepare a sequential read of skb data
2725  * @skb: the buffer to read
2726  * @from: lower offset of data to be read
2727  * @to: upper offset of data to be read
2728  * @st: state variable
2729  *
2730  * Initializes the specified state variable. Must be called before
2731  * invoking skb_seq_read() for the first time.
2732  */
2733 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2734                           unsigned int to, struct skb_seq_state *st)
2735 {
2736         st->lower_offset = from;
2737         st->upper_offset = to;
2738         st->root_skb = st->cur_skb = skb;
2739         st->frag_idx = st->stepped_offset = 0;
2740         st->frag_data = NULL;
2741 }
2742 EXPORT_SYMBOL(skb_prepare_seq_read);
2743 
2744 /**
2745  * skb_seq_read - Sequentially read skb data
2746  * @consumed: number of bytes consumed by the caller so far
2747  * @data: destination pointer for data to be returned
2748  * @st: state variable
2749  *
2750  * Reads a block of skb data at @consumed relative to the
2751  * lower offset specified to skb_prepare_seq_read(). Assigns
2752  * the head of the data block to @data and returns the length
2753  * of the block or 0 if the end of the skb data or the upper
2754  * offset has been reached.
2755  *
2756  * The caller is not required to consume all of the data
2757  * returned, i.e. @consumed is typically set to the number
2758  * of bytes already consumed and the next call to
2759  * skb_seq_read() will return the remaining part of the block.
2760  *
2761  * Note 1: The size of each block of data returned can be arbitrary,
2762  *       this limitation is the cost for zerocopy sequential
2763  *       reads of potentially non linear data.
2764  *
2765  * Note 2: Fragment lists within fragments are not implemented
2766  *       at the moment, state->root_skb could be replaced with
2767  *       a stack for this purpose.
2768  */
2769 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2770                           struct skb_seq_state *st)
2771 {
2772         unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2773         skb_frag_t *frag;
2774 
2775         if (unlikely(abs_offset >= st->upper_offset)) {
2776                 if (st->frag_data) {
2777                         kunmap_atomic(st->frag_data);
2778                         st->frag_data = NULL;
2779                 }
2780                 return 0;
2781         }
2782 
2783 next_skb:
2784         block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2785 
2786         if (abs_offset < block_limit && !st->frag_data) {
2787                 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2788                 return block_limit - abs_offset;
2789         }
2790 
2791         if (st->frag_idx == 0 && !st->frag_data)
2792                 st->stepped_offset += skb_headlen(st->cur_skb);
2793 
2794         while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2795                 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2796                 block_limit = skb_frag_size(frag) + st->stepped_offset;
2797 
2798                 if (abs_offset < block_limit) {
2799                         if (!st->frag_data)
2800                                 st->frag_data = kmap_atomic(skb_frag_page(frag));
2801 
2802                         *data = (u8 *) st->frag_data + frag->page_offset +
2803                                 (abs_offset - st->stepped_offset);
2804 
2805                         return block_limit - abs_offset;
2806                 }
2807 
2808                 if (st->frag_data) {
2809                         kunmap_atomic(st->frag_data);
2810                         st->frag_data = NULL;
2811                 }
2812 
2813                 st->frag_idx++;
2814                 st->stepped_offset += skb_frag_size(frag);
2815         }
2816 
2817         if (st->frag_data) {
2818                 kunmap_atomic(st->frag_data);
2819                 st->frag_data = NULL;
2820         }
2821 
2822         if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2823                 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2824                 st->frag_idx = 0;
2825                 goto next_skb;
2826         } else if (st->cur_skb->next) {
2827                 st->cur_skb = st->cur_skb->next;
2828                 st->frag_idx = 0;
2829                 goto next_skb;
2830         }
2831 
2832         return 0;
2833 }
2834 EXPORT_SYMBOL(skb_seq_read);
2835 
2836 /**
2837  * skb_abort_seq_read - Abort a sequential read of skb data
2838  * @st: state variable
2839  *
2840  * Must be called if skb_seq_read() was not called until it
2841  * returned 0.
2842  */
2843 void skb_abort_seq_read(struct skb_seq_state *st)
2844 {
2845         if (st->frag_data)
2846                 kunmap_atomic(st->frag_data);
2847 }
2848 EXPORT_SYMBOL(skb_abort_seq_read);
2849 
2850 #define TS_SKB_CB(state)        ((struct skb_seq_state *) &((state)->cb))
2851 
2852 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2853                                           struct ts_config *conf,
2854                                           struct ts_state *state)
2855 {
2856         return skb_seq_read(offset, text, TS_SKB_CB(state));
2857 }
2858 
2859 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2860 {
2861         skb_abort_seq_read(TS_SKB_CB(state));
2862 }
2863 
2864 /**
2865  * skb_find_text - Find a text pattern in skb data
2866  * @skb: the buffer to look in
2867  * @from: search offset
2868  * @to: search limit
2869  * @config: textsearch configuration
2870  * @state: uninitialized textsearch state variable
2871  *
2872  * Finds a pattern in the skb data according to the specified
2873  * textsearch configuration. Use textsearch_next() to retrieve
2874  * subsequent occurrences of the pattern. Returns the offset
2875  * to the first occurrence or UINT_MAX if no match was found.
2876  */
2877 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2878                            unsigned int to, struct ts_config *config,
2879                            struct ts_state *state)
2880 {
2881         unsigned int ret;
2882 
2883         config->get_next_block = skb_ts_get_next_block;
2884         config->finish = skb_ts_finish;
2885 
2886         skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2887 
2888         ret = textsearch_find(config, state);
2889         return (ret <= to - from ? ret : UINT_MAX);
2890 }
2891 EXPORT_SYMBOL(skb_find_text);
2892 
2893 /**
2894  * skb_append_datato_frags - append the user data to a skb
2895  * @sk: sock  structure
2896  * @skb: skb structure to be appended with user data.
2897  * @getfrag: call back function to be used for getting the user data
2898  * @from: pointer to user message iov
2899  * @length: length of the iov message
2900  *
2901  * Description: This procedure append the user data in the fragment part
2902  * of the skb if any page alloc fails user this procedure returns  -ENOMEM
2903  */
2904 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2905                         int (*getfrag)(void *from, char *to, int offset,
2906                                         int len, int odd, struct sk_buff *skb),
2907                         void *from, int length)
2908 {
2909         int frg_cnt = skb_shinfo(skb)->nr_frags;
2910         int copy;
2911         int offset = 0;
2912         int ret;
2913         struct page_frag *pfrag = &current->task_frag;
2914 
2915         do {
2916                 /* Return error if we don't have space for new frag */
2917                 if (frg_cnt >= MAX_SKB_FRAGS)
2918                         return -EMSGSIZE;
2919 
2920                 if (!sk_page_frag_refill(sk, pfrag))
2921                         return -ENOMEM;
2922 
2923                 /* copy the user data to page */
2924                 copy = min_t(int, length, pfrag->size - pfrag->offset);
2925 
2926                 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
2927                               offset, copy, 0, skb);
2928                 if (ret < 0)
2929                         return -EFAULT;
2930 
2931                 /* copy was successful so update the size parameters */
2932                 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
2933                                    copy);
2934                 frg_cnt++;
2935                 pfrag->offset += copy;
2936                 get_page(pfrag->page);
2937 
2938                 skb->truesize += copy;
2939                 atomic_add(copy, &sk->sk_wmem_alloc);
2940                 skb->len += copy;
2941                 skb->data_len += copy;
2942                 offset += copy;
2943                 length -= copy;
2944 
2945         } while (length > 0);
2946 
2947         return 0;
2948 }
2949 EXPORT_SYMBOL(skb_append_datato_frags);
2950 
2951 /**
2952  *      skb_pull_rcsum - pull skb and update receive checksum
2953  *      @skb: buffer to update
2954  *      @len: length of data pulled
2955  *
2956  *      This function performs an skb_pull on the packet and updates
2957  *      the CHECKSUM_COMPLETE checksum.  It should be used on
2958  *      receive path processing instead of skb_pull unless you know
2959  *      that the checksum difference is zero (e.g., a valid IP header)
2960  *      or you are setting ip_summed to CHECKSUM_NONE.
2961  */
2962 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2963 {
2964         BUG_ON(len > skb->len);
2965         skb->len -= len;
2966         BUG_ON(skb->len < skb->data_len);
2967         skb_postpull_rcsum(skb, skb->data, len);
2968         return skb->data += len;
2969 }
2970 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2971 
2972 /**
2973  *      skb_segment - Perform protocol segmentation on skb.
2974  *      @head_skb: buffer to segment
2975  *      @features: features for the output path (see dev->features)
2976  *
2977  *      This function performs segmentation on the given skb.  It returns
2978  *      a pointer to the first in a list of new skbs for the segments.
2979  *      In case of error it returns ERR_PTR(err).
2980  */
2981 struct sk_buff *skb_segment(struct sk_buff *head_skb,
2982                             netdev_features_t features)
2983 {
2984         struct sk_buff *segs = NULL;
2985         struct sk_buff *tail = NULL;
2986         struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
2987         skb_frag_t *frag = skb_shinfo(head_skb)->frags;
2988         unsigned int mss = skb_shinfo(head_skb)->gso_size;
2989         unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
2990         struct sk_buff *frag_skb = head_skb;
2991         unsigned int offset = doffset;
2992         unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
2993         unsigned int headroom;
2994         unsigned int len;
2995         __be16 proto;
2996         bool csum;
2997         int sg = !!(features & NETIF_F_SG);
2998         int nfrags = skb_shinfo(head_skb)->nr_frags;
2999         int err = -ENOMEM;
3000         int i = 0;
3001         int pos;
3002         int dummy;
3003 
3004         __skb_push(head_skb, doffset);
3005         proto = skb_network_protocol(head_skb, &dummy);
3006         if (unlikely(!proto))
3007                 return ERR_PTR(-EINVAL);
3008 
3009         csum = !head_skb->encap_hdr_csum &&
3010             !!can_checksum_protocol(features, proto);
3011 
3012         headroom = skb_headroom(head_skb);
3013         pos = skb_headlen(head_skb);
3014 
3015         do {
3016                 struct sk_buff *nskb;
3017                 skb_frag_t *nskb_frag;
3018                 int hsize;
3019                 int size;
3020 
3021                 len = head_skb->len - offset;
3022                 if (len > mss)
3023                         len = mss;
3024 
3025                 hsize = skb_headlen(head_skb) - offset;
3026                 if (hsize < 0)
3027                         hsize = 0;
3028                 if (hsize > len || !sg)
3029                         hsize = len;
3030 
3031                 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3032                     (skb_headlen(list_skb) == len || sg)) {
3033                         BUG_ON(skb_headlen(list_skb) > len);
3034 
3035                         i = 0;
3036                         nfrags = skb_shinfo(list_skb)->nr_frags;
3037                         frag = skb_shinfo(list_skb)->frags;
3038                         frag_skb = list_skb;
3039                         pos += skb_headlen(list_skb);
3040 
3041                         while (pos < offset + len) {
3042                                 BUG_ON(i >= nfrags);
3043 
3044                                 size = skb_frag_size(frag);
3045                                 if (pos + size > offset + len)
3046                                         break;
3047 
3048                                 i++;
3049                                 pos += size;
3050                                 frag++;
3051                         }
3052 
3053                         nskb = skb_clone(list_skb, GFP_ATOMIC);
3054                         list_skb = list_skb->next;
3055 
3056                         if (unlikely(!nskb))
3057                                 goto err;
3058 
3059                         if (unlikely(pskb_trim(nskb, len))) {
3060                                 kfree_skb(nskb);
3061                                 goto err;
3062                         }
3063 
3064                         hsize = skb_end_offset(nskb);
3065                         if (skb_cow_head(nskb, doffset + headroom)) {
3066                                 kfree_skb(nskb);
3067                                 goto err;
3068                         }
3069 
3070                         nskb->truesize += skb_end_offset(nskb) - hsize;
3071                         skb_release_head_state(nskb);
3072                         __skb_push(nskb, doffset);
3073                 } else {
3074                         nskb = __alloc_skb(hsize + doffset + headroom,
3075                                            GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3076                                            NUMA_NO_NODE);
3077 
3078                         if (unlikely(!nskb))
3079                                 goto err;
3080 
3081                         skb_reserve(nskb, headroom);
3082                         __skb_put(nskb, doffset);
3083                 }
3084 
3085                 if (segs)
3086                         tail->next = nskb;
3087                 else
3088                         segs = nskb;
3089                 tail = nskb;
3090 
3091                 __copy_skb_header(nskb, head_skb);
3092 
3093                 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3094                 skb_reset_mac_len(nskb);
3095 
3096                 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3097                                                  nskb->data - tnl_hlen,
3098                                                  doffset + tnl_hlen);
3099 
3100                 if (nskb->len == len + doffset)
3101                         goto perform_csum_check;
3102 
3103                 if (!sg && !nskb->remcsum_offload) {
3104                         nskb->ip_summed = CHECKSUM_NONE;
3105                         nskb->csum = skb_copy_and_csum_bits(head_skb, offset,
3106                                                             skb_put(nskb, len),
3107                                                             len, 0);
3108                         SKB_GSO_CB(nskb)->csum_start =
3109                             skb_headroom(nskb) + doffset;
3110                         continue;
3111                 }
3112 
3113                 nskb_frag = skb_shinfo(nskb)->frags;
3114 
3115                 skb_copy_from_linear_data_offset(head_skb, offset,
3116                                                  skb_put(nskb, hsize), hsize);
3117 
3118                 skb_shinfo(nskb)->tx_flags = skb_shinfo(head_skb)->tx_flags &
3119                         SKBTX_SHARED_FRAG;
3120 
3121                 while (pos < offset + len) {
3122                         if (i >= nfrags) {
3123                                 BUG_ON(skb_headlen(list_skb));
3124 
3125                                 i = 0;
3126                                 nfrags = skb_shinfo(list_skb)->nr_frags;
3127                                 frag = skb_shinfo(list_skb)->frags;
3128                                 frag_skb = list_skb;
3129 
3130                                 BUG_ON(!nfrags);
3131 
3132                                 list_skb = list_skb->next;
3133                         }
3134 
3135                         if (unlikely(skb_shinfo(nskb)->nr_frags >=
3136                                      MAX_SKB_FRAGS)) {
3137                                 net_warn_ratelimited(
3138                                         "skb_segment: too many frags: %u %u\n",
3139                                         pos, mss);
3140                                 goto err;
3141                         }
3142 
3143                         if (unlikely(skb_orphan_frags(frag_skb, GFP_ATOMIC)))
3144                                 goto err;
3145 
3146                         *nskb_frag = *frag;
3147                         __skb_frag_ref(nskb_frag);
3148                         size = skb_frag_size(nskb_frag);
3149 
3150                         if (pos < offset) {
3151                                 nskb_frag->page_offset += offset - pos;
3152                                 skb_frag_size_sub(nskb_frag, offset - pos);
3153                         }
3154 
3155                         skb_shinfo(nskb)->nr_frags++;
3156 
3157                         if (pos + size <= offset + len) {
3158                                 i++;
3159                                 frag++;
3160                                 pos += size;
3161                         } else {
3162                                 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
3163                                 goto skip_fraglist;
3164                         }
3165 
3166                         nskb_frag++;
3167                 }
3168 
3169 skip_fraglist:
3170                 nskb->data_len = len - hsize;
3171                 nskb->len += nskb->data_len;
3172                 nskb->truesize += nskb->data_len;
3173 
3174 perform_csum_check:
3175                 if (!csum && !nskb->remcsum_offload) {
3176                         nskb->csum = skb_checksum(nskb, doffset,
3177                                                   nskb->len - doffset, 0);
3178                         nskb->ip_summed = CHECKSUM_NONE;
3179                         SKB_GSO_CB(nskb)->csum_start =
3180                             skb_headroom(nskb) + doffset;
3181                 }
3182         } while ((offset += len) < head_skb->len);
3183 
3184         /* Some callers want to get the end of the list.
3185          * Put it in segs->prev to avoid walking the list.
3186          * (see validate_xmit_skb_list() for example)
3187          */
3188         segs->prev = tail;
3189 
3190         /* Following permits correct backpressure, for protocols
3191          * using skb_set_owner_w().
3192          * Idea is to tranfert ownership from head_skb to last segment.
3193          */
3194         if (head_skb->destructor == sock_wfree) {
3195                 swap(tail->truesize, head_skb->truesize);
3196                 swap(tail->destructor, head_skb->destructor);
3197                 swap(tail->sk, head_skb->sk);
3198         }
3199         return segs;
3200 
3201 err:
3202         kfree_skb_list(segs);
3203         return ERR_PTR(err);
3204 }
3205 EXPORT_SYMBOL_GPL(skb_segment);
3206 
3207 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
3208 {
3209         struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
3210         unsigned int offset = skb_gro_offset(skb);
3211         unsigned int headlen = skb_headlen(skb);
3212         struct sk_buff *nskb, *lp, *p = *head;
3213         unsigned int len = skb_gro_len(skb);
3214         unsigned int delta_truesize;
3215         unsigned int headroom;
3216 
3217         if (unlikely(p->len + len >= 65536))
3218                 return -E2BIG;
3219 
3220         lp = NAPI_GRO_CB(p)->last;
3221         pinfo = skb_shinfo(lp);
3222 
3223         if (headlen <= offset) {
3224                 skb_frag_t *frag;
3225                 skb_frag_t *frag2;
3226                 int i = skbinfo->nr_frags;
3227                 int nr_frags = pinfo->nr_frags + i;
3228 
3229                 if (nr_frags > MAX_SKB_FRAGS)
3230                         goto merge;
3231 
3232                 offset -= headlen;
3233                 pinfo->nr_frags = nr_frags;
3234                 skbinfo->nr_frags = 0;
3235 
3236                 frag = pinfo->frags + nr_frags;
3237                 frag2 = skbinfo->frags + i;
3238                 do {
3239                         *--frag = *--frag2;
3240                 } while (--i);
3241 
3242                 frag->page_offset += offset;
3243                 skb_frag_size_sub(frag, offset);
3244 
3245                 /* all fragments truesize : remove (head size + sk_buff) */
3246                 delta_truesize = skb->truesize -
3247                                  SKB_TRUESIZE(skb_end_offset(skb));
3248 
3249                 skb->truesize -= skb->data_len;
3250                 skb->len -= skb->data_len;
3251                 skb->data_len = 0;
3252 
3253                 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
3254                 goto done;
3255         } else if (skb->head_frag) {
3256                 int nr_frags = pinfo->nr_frags;
3257                 skb_frag_t *frag = pinfo->frags + nr_frags;
3258                 struct page *page = virt_to_head_page(skb->head);
3259                 unsigned int first_size = headlen - offset;
3260                 unsigned int first_offset;
3261 
3262                 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
3263                         goto merge;
3264 
3265                 first_offset = skb->data -
3266                                (unsigned char *)page_address(page) +
3267                                offset;
3268 
3269                 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3270 
3271                 frag->page.p      = page;
3272                 frag->page_offset = first_offset;
3273                 skb_frag_size_set(frag, first_size);
3274 
3275                 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3276                 /* We dont need to clear skbinfo->nr_frags here */
3277 
3278                 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3279                 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3280                 goto done;
3281         }
3282         /* switch back to head shinfo */
3283         pinfo = skb_shinfo(p);
3284 
3285         if (pinfo->frag_list)
3286                 goto merge;
3287         if (skb_gro_len(p) != pinfo->gso_size)
3288                 return -E2BIG;
3289 
3290         headroom = skb_headroom(p);
3291         nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
3292         if (unlikely(!nskb))
3293                 return -ENOMEM;
3294 
3295         __copy_skb_header(nskb, p);
3296         nskb->mac_len = p->mac_len;
3297 
3298         skb_reserve(nskb, headroom);
3299         __skb_put(nskb, skb_gro_offset(p));
3300 
3301         skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
3302         skb_set_network_header(nskb, skb_network_offset(p));
3303         skb_set_transport_header(nskb, skb_transport_offset(p));
3304 
3305         __skb_pull(p, skb_gro_offset(p));
3306         memcpy(skb_mac_header(nskb), skb_mac_header(p),
3307                p->data - skb_mac_header(p));
3308 
3309         skb_shinfo(nskb)->frag_list = p;
3310         skb_shinfo(nskb)->gso_size = pinfo->gso_size;
3311         pinfo->gso_size = 0;
3312         __skb_header_release(p);
3313         NAPI_GRO_CB(nskb)->last = p;
3314 
3315         nskb->data_len += p->len;
3316         nskb->truesize += p->truesize;
3317         nskb->len += p->len;
3318 
3319         *head = nskb;
3320         nskb->next = p->next;
3321         p->next = NULL;
3322 
3323         p = nskb;
3324 
3325 merge:
3326         delta_truesize = skb->truesize;
3327         if (offset > headlen) {
3328                 unsigned int eat = offset - headlen;
3329 
3330                 skbinfo->frags[0].page_offset += eat;
3331                 skb_frag_size_sub(&skbinfo->frags[0], eat);
3332                 skb->data_len -= eat;
3333                 skb->len -= eat;
3334                 offset = headlen;
3335         }
3336 
3337         __skb_pull(skb, offset);
3338 
3339         if (NAPI_GRO_CB(p)->last == p)
3340                 skb_shinfo(p)->frag_list = skb;
3341         else
3342                 NAPI_GRO_CB(p)->last->next = skb;
3343         NAPI_GRO_CB(p)->last = skb;
3344         __skb_header_release(skb);
3345         lp = p;
3346 
3347 done:
3348         NAPI_GRO_CB(p)->count++;
3349         p->data_len += len;
3350         p->truesize += delta_truesize;
3351         p->len += len;
3352         if (lp != p) {
3353                 lp->data_len += len;
3354                 lp->truesize += delta_truesize;
3355                 lp->len += len;
3356         }
3357         NAPI_GRO_CB(skb)->same_flow = 1;
3358         return 0;
3359 }
3360 
3361 void __init skb_init(void)
3362 {
3363         skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3364                                               sizeof(struct sk_buff),
3365                                               0,
3366                                               SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3367                                               NULL);
3368         skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3369                                                 sizeof(struct sk_buff_fclones),
3370                                                 0,
3371                                                 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3372                                                 NULL);
3373 }
3374 
3375 /**
3376  *      skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3377  *      @skb: Socket buffer containing the buffers to be mapped
3378  *      @sg: The scatter-gather list to map into
3379  *      @offset: The offset into the buffer's contents to start mapping
3380  *      @len: Length of buffer space to be mapped
3381  *
3382  *      Fill the specified scatter-gather list with mappings/pointers into a
3383  *      region of the buffer space attached to a socket buffer.
3384  */
3385 static int
3386 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3387 {
3388         int start = skb_headlen(skb);
3389         int i, copy = start - offset;
3390         struct sk_buff *frag_iter;
3391         int elt = 0;
3392 
3393         if (copy > 0) {
3394                 if (copy > len)
3395                         copy = len;
3396                 sg_set_buf(sg, skb->data + offset, copy);
3397                 elt++;
3398                 if ((len -= copy) == 0)
3399                         return elt;
3400                 offset += copy;
3401         }
3402 
3403         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3404                 int end;
3405 
3406                 WARN_ON(start > offset + len);
3407 
3408                 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3409                 if ((copy = end - offset) > 0) {
3410                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3411 
3412                         if (copy > len)
3413                                 copy = len;
3414                         sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3415                                         frag->page_offset+offset-start);
3416                         elt++;
3417                         if (!(len -= copy))
3418                                 return elt;
3419                         offset += copy;
3420                 }
3421                 start = end;
3422         }
3423 
3424         skb_walk_frags(skb, frag_iter) {
3425                 int end;
3426 
3427                 WARN_ON(start > offset + len);
3428 
3429                 end = start + frag_iter->len;
3430                 if ((copy = end - offset) > 0) {
3431                         if (copy > len)
3432                                 copy = len;
3433                         elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3434                                               copy);
3435                         if ((len -= copy) == 0)
3436                                 return elt;
3437                         offset += copy;
3438                 }
3439                 start = end;
3440         }
3441         BUG_ON(len);
3442         return elt;
3443 }
3444 
3445 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
3446  * sglist without mark the sg which contain last skb data as the end.
3447  * So the caller can mannipulate sg list as will when padding new data after
3448  * the first call without calling sg_unmark_end to expend sg list.
3449  *
3450  * Scenario to use skb_to_sgvec_nomark:
3451  * 1. sg_init_table
3452  * 2. skb_to_sgvec_nomark(payload1)
3453  * 3. skb_to_sgvec_nomark(payload2)
3454  *
3455  * This is equivalent to:
3456  * 1. sg_init_table
3457  * 2. skb_to_sgvec(payload1)
3458  * 3. sg_unmark_end
3459  * 4. skb_to_sgvec(payload2)
3460  *
3461  * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
3462  * is more preferable.
3463  */
3464 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
3465                         int offset, int len)
3466 {
3467         return __skb_to_sgvec(skb, sg, offset, len);
3468 }
3469 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
3470 
3471 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3472 {
3473         int nsg = __skb_to_sgvec(skb, sg, offset, len);
3474 
3475         sg_mark_end(&sg[nsg - 1]);
3476 
3477         return nsg;
3478 }
3479 EXPORT_SYMBOL_GPL(skb_to_sgvec);
3480 
3481 /**
3482  *      skb_cow_data - Check that a socket buffer's data buffers are writable
3483  *      @skb: The socket buffer to check.
3484  *      @tailbits: Amount of trailing space to be added
3485  *      @trailer: Returned pointer to the skb where the @tailbits space begins
3486  *
3487  *      Make sure that the data buffers attached to a socket buffer are
3488  *      writable. If they are not, private copies are made of the data buffers
3489  *      and the socket buffer is set to use these instead.
3490  *
3491  *      If @tailbits is given, make sure that there is space to write @tailbits
3492  *      bytes of data beyond current end of socket buffer.  @trailer will be
3493  *      set to point to the skb in which this space begins.
3494  *
3495  *      The number of scatterlist elements required to completely map the
3496  *      COW'd and extended socket buffer will be returned.
3497  */
3498 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3499 {
3500         int copyflag;
3501         int elt;
3502         struct sk_buff *skb1, **skb_p;
3503 
3504         /* If skb is cloned or its head is paged, reallocate
3505          * head pulling out all the pages (pages are considered not writable
3506          * at the moment even if they are anonymous).
3507          */
3508         if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3509             __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3510                 return -ENOMEM;
3511 
3512         /* Easy case. Most of packets will go this way. */
3513         if (!skb_has_frag_list(skb)) {
3514                 /* A little of trouble, not enough of space for trailer.
3515                  * This should not happen, when stack is tuned to generate
3516                  * good frames. OK, on miss we reallocate and reserve even more
3517                  * space, 128 bytes is fair. */
3518 
3519                 if (skb_tailroom(skb) < tailbits &&
3520                     pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3521                         return -ENOMEM;
3522 
3523                 /* Voila! */
3524                 *trailer = skb;
3525                 return 1;
3526         }
3527 
3528         /* Misery. We are in troubles, going to mincer fragments... */
3529 
3530         elt = 1;
3531         skb_p = &skb_shinfo(skb)->frag_list;
3532         copyflag = 0;
3533 
3534         while ((skb1 = *skb_p) != NULL) {
3535                 int ntail = 0;
3536 
3537                 /* The fragment is partially pulled by someone,
3538                  * this can happen on input. Copy it and everything
3539                  * after it. */
3540 
3541                 if (skb_shared(skb1))
3542                         copyflag = 1;
3543 
3544                 /* If the skb is the last, worry about trailer. */
3545 
3546                 if (skb1->next == NULL && tailbits) {
3547                         if (skb_shinfo(skb1)->nr_frags ||
3548                             skb_has_frag_list(skb1) ||
3549                             skb_tailroom(skb1) < tailbits)
3550                                 ntail = tailbits + 128;
3551                 }
3552 
3553                 if (copyflag ||
3554                     skb_cloned(skb1) ||
3555                     ntail ||
3556                     skb_shinfo(skb1)->nr_frags ||
3557                     skb_has_frag_list(skb1)) {
3558                         struct sk_buff *skb2;
3559 
3560                         /* Fuck, we are miserable poor guys... */
3561                         if (ntail == 0)
3562                                 skb2 = skb_copy(skb1, GFP_ATOMIC);
3563                         else
3564                                 skb2 = skb_copy_expand(skb1,
3565                                                        skb_headroom(skb1),
3566                                                        ntail,
3567                                                        GFP_ATOMIC);
3568                         if (unlikely(skb2 == NULL))
3569                                 return -ENOMEM;
3570 
3571                         if (skb1->sk)
3572                                 skb_set_owner_w(skb2, skb1->sk);
3573 
3574                         /* Looking around. Are we still alive?
3575                          * OK, link new skb, drop old one */
3576 
3577                         skb2->next = skb1->next;
3578                         *skb_p = skb2;
3579                         kfree_skb(skb1);
3580                         skb1 = skb2;
3581                 }
3582                 elt++;
3583                 *trailer = skb1;
3584                 skb_p = &skb1->next;
3585         }
3586 
3587         return elt;
3588 }
3589 EXPORT_SYMBOL_GPL(skb_cow_data);
3590 
3591 static void sock_rmem_free(struct sk_buff *skb)
3592 {
3593         struct sock *sk = skb->sk;
3594 
3595         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3596 }
3597 
3598 /*
3599  * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3600  */
3601 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3602 {
3603         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3604             (unsigned int)sk->sk_rcvbuf)
3605                 return -ENOMEM;
3606 
3607         skb_orphan(skb);
3608         skb->sk = sk;
3609         skb->destructor = sock_rmem_free;
3610         atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3611 
3612         /* before exiting rcu section, make sure dst is refcounted */
3613         skb_dst_force(skb);
3614 
3615         skb_queue_tail(&sk->sk_error_queue, skb);
3616         if (!sock_flag(sk, SOCK_DEAD))
3617                 sk->sk_data_ready(sk);
3618         return 0;
3619 }
3620 EXPORT_SYMBOL(sock_queue_err_skb);
3621 
3622 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
3623 {
3624         struct sk_buff_head *q = &sk->sk_error_queue;
3625         struct sk_buff *skb, *skb_next;
3626         int err = 0;
3627 
3628         spin_lock_bh(&q->lock);
3629         skb = __skb_dequeue(q);
3630         if (skb && (skb_next = skb_peek(q)))
3631                 err = SKB_EXT_ERR(skb_next)->ee.ee_errno;
3632         spin_unlock_bh(&q->lock);
3633 
3634         sk->sk_err = err;
3635         if (err)
3636                 sk->sk_error_report(sk);
3637 
3638         return skb;
3639 }
3640 EXPORT_SYMBOL(sock_dequeue_err_skb);
3641 
3642 /**
3643  * skb_clone_sk - create clone of skb, and take reference to socket
3644  * @skb: the skb to clone
3645  *
3646  * This function creates a clone of a buffer that holds a reference on
3647  * sk_refcnt.  Buffers created via this function are meant to be
3648  * returned using sock_queue_err_skb, or free via kfree_skb.
3649  *
3650  * When passing buffers allocated with this function to sock_queue_err_skb
3651  * it is necessary to wrap the call with sock_hold/sock_put in order to
3652  * prevent the socket from being released prior to being enqueued on
3653  * the sk_error_queue.
3654  */
3655 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
3656 {
3657         struct sock *sk = skb->sk;
3658         struct sk_buff *clone;
3659 
3660         if (!sk || !atomic_inc_not_zero(&sk->sk_refcnt))
3661                 return NULL;
3662 
3663         clone = skb_clone(skb, GFP_ATOMIC);
3664         if (!clone) {
3665                 sock_put(sk);
3666                 return NULL;
3667         }
3668 
3669         clone->sk = sk;
3670         clone->destructor = sock_efree;
3671 
3672         return clone;
3673 }
3674 EXPORT_SYMBOL(skb_clone_sk);
3675 
3676 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
3677                                         struct sock *sk,
3678                                         int tstype)
3679 {
3680         struct sock_exterr_skb *serr;
3681         int err;
3682 
3683         serr = SKB_EXT_ERR(skb);
3684         memset(serr, 0, sizeof(*serr));
3685         serr->ee.ee_errno = ENOMSG;
3686         serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3687         serr->ee.ee_info = tstype;
3688         if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
3689                 serr->ee.ee_data = skb_shinfo(skb)->tskey;
3690                 if (sk->sk_protocol == IPPROTO_TCP)
3691                         serr->ee.ee_data -= sk->sk_tskey;
3692         }
3693 
3694         err = sock_queue_err_skb(sk, skb);
3695 
3696         if (err)
3697                 kfree_skb(skb);
3698 }
3699 
3700 void skb_complete_tx_timestamp(struct sk_buff *skb,
3701                                struct skb_shared_hwtstamps *hwtstamps)
3702 {
3703         struct sock *sk = skb->sk;
3704 
3705         /* take a reference to prevent skb_orphan() from freeing the socket */
3706         sock_hold(sk);
3707 
3708         *skb_hwtstamps(skb) = *hwtstamps;
3709         __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND);
3710 
3711         sock_put(sk);
3712 }
3713 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
3714 
3715 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3716                      struct skb_shared_hwtstamps *hwtstamps,
3717                      struct sock *sk, int tstype)
3718 {
3719         struct sk_buff *skb;
3720 
3721         if (!sk)
3722                 return;
3723 
3724         if (hwtstamps)
3725                 *skb_hwtstamps(orig_skb) = *hwtstamps;
3726         else
3727                 orig_skb->tstamp = ktime_get_real();
3728 
3729         skb = skb_clone(orig_skb, GFP_ATOMIC);
3730         if (!skb)
3731                 return;
3732 
3733         __skb_complete_tx_timestamp(skb, sk, tstype);
3734 }
3735 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
3736 
3737 void skb_tstamp_tx(struct sk_buff *orig_skb,
3738                    struct skb_shared_hwtstamps *hwtstamps)
3739 {
3740         return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
3741                                SCM_TSTAMP_SND);
3742 }
3743 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3744 
3745 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3746 {
3747         struct sock *sk = skb->sk;
3748         struct sock_exterr_skb *serr;
3749         int err;
3750 
3751         skb->wifi_acked_valid = 1;
3752         skb->wifi_acked = acked;
3753 
3754         serr = SKB_EXT_ERR(skb);
3755         memset(serr, 0, sizeof(*serr));
3756         serr->ee.ee_errno = ENOMSG;
3757         serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3758 
3759         /* take a reference to prevent skb_orphan() from freeing the socket */
3760         sock_hold(sk);
3761 
3762         err = sock_queue_err_skb(sk, skb);
3763         if (err)
3764                 kfree_skb(skb);
3765 
3766         sock_put(sk);
3767 }
3768 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3769 
3770 
3771 /**
3772  * skb_partial_csum_set - set up and verify partial csum values for packet
3773  * @skb: the skb to set
3774  * @start: the number of bytes after skb->data to start checksumming.
3775  * @off: the offset from start to place the checksum.
3776  *
3777  * For untrusted partially-checksummed packets, we need to make sure the values
3778  * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3779  *
3780  * This function checks and sets those values and skb->ip_summed: if this
3781  * returns false you should drop the packet.
3782  */
3783 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3784 {
3785         if (unlikely(start > skb_headlen(skb)) ||
3786             unlikely((int)start + off > skb_headlen(skb) - 2)) {
3787                 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3788                                      start, off, skb_headlen(skb));
3789                 return false;
3790         }
3791         skb->ip_summed = CHECKSUM_PARTIAL;
3792         skb->csum_start = skb_headroom(skb) + start;
3793         skb->csum_offset = off;
3794         skb_set_transport_header(skb, start);
3795         return true;
3796 }
3797 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3798 
3799 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
3800                                unsigned int max)
3801 {
3802         if (skb_headlen(skb) >= len)
3803                 return 0;
3804 
3805         /* If we need to pullup then pullup to the max, so we
3806          * won't need to do it again.
3807          */
3808         if (max > skb->len)
3809                 max = skb->len;
3810 
3811         if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
3812                 return -ENOMEM;
3813 
3814         if (skb_headlen(skb) < len)
3815                 return -EPROTO;
3816 
3817         return 0;
3818 }
3819 
3820 #define MAX_TCP_HDR_LEN (15 * 4)
3821 
3822 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
3823                                       typeof(IPPROTO_IP) proto,
3824                                       unsigned int off)
3825 {
3826         switch (proto) {
3827                 int err;
3828 
3829         case IPPROTO_TCP:
3830                 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
3831                                           off + MAX_TCP_HDR_LEN);
3832                 if (!err && !skb_partial_csum_set(skb, off,
3833                                                   offsetof(struct tcphdr,
3834                                                            check)))
3835                         err = -EPROTO;
3836                 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
3837 
3838         case IPPROTO_UDP:
3839                 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
3840                                           off + sizeof(struct udphdr));
3841                 if (!err && !skb_partial_csum_set(skb, off,
3842                                                   offsetof(struct udphdr,
3843                                                            check)))
3844                         err = -EPROTO;
3845                 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
3846         }
3847 
3848         return ERR_PTR(-EPROTO);
3849 }
3850 
3851 /* This value should be large enough to cover a tagged ethernet header plus
3852  * maximally sized IP and TCP or UDP headers.
3853  */
3854 #define MAX_IP_HDR_LEN 128
3855 
3856 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
3857 {
3858         unsigned int off;
3859         bool fragment;
3860         __sum16 *csum;
3861         int err;
3862 
3863         fragment = false;
3864 
3865         err = skb_maybe_pull_tail(skb,
3866                                   sizeof(struct iphdr),
3867                                   MAX_IP_HDR_LEN);
3868         if (err < 0)
3869                 goto out;
3870 
3871         if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
3872                 fragment = true;
3873 
3874         off = ip_hdrlen(skb);
3875 
3876         err = -EPROTO;
3877 
3878         if (fragment)
3879                 goto out;
3880 
3881         csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
3882         if (IS_ERR(csum))
3883                 return PTR_ERR(csum);
3884 
3885         if (recalculate)
3886                 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3887                                            ip_hdr(skb)->daddr,
3888                                            skb->len - off,
3889                                            ip_hdr(skb)->protocol, 0);
3890         err = 0;
3891 
3892 out:
3893         return err;
3894 }
3895 
3896 /* This value should be large enough to cover a tagged ethernet header plus
3897  * an IPv6 header, all options, and a maximal TCP or UDP header.
3898  */
3899 #define MAX_IPV6_HDR_LEN 256
3900 
3901 #define OPT_HDR(type, skb, off) \
3902         (type *)(skb_network_header(skb) + (off))
3903 
3904 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
3905 {
3906         int err;
3907         u8 nexthdr;
3908         unsigned int off;
3909         unsigned int len;
3910         bool fragment;
3911         bool done;
3912         __sum16 *csum;
3913 
3914         fragment = false;
3915         done = false;
3916 
3917         off = sizeof(struct ipv6hdr);
3918 
3919         err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
3920         if (err < 0)
3921                 goto out;
3922 
3923         nexthdr = ipv6_hdr(skb)->nexthdr;
3924 
3925         len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
3926         while (off <= len && !done) {
3927                 switch (nexthdr) {
3928                 case IPPROTO_DSTOPTS:
3929                 case IPPROTO_HOPOPTS:
3930                 case IPPROTO_ROUTING: {
3931                         struct ipv6_opt_hdr *hp;
3932 
3933                         err = skb_maybe_pull_tail(skb,
3934                                                   off +
3935                                                   sizeof(struct ipv6_opt_hdr),
3936                                                   MAX_IPV6_HDR_LEN);
3937                         if (err < 0)
3938                                 goto out;
3939 
3940                         hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
3941                         nexthdr = hp->nexthdr;
3942                         off += ipv6_optlen(hp);
3943                         break;
3944                 }
3945                 case IPPROTO_AH: {
3946                         struct ip_auth_hdr *hp;
3947 
3948                         err = skb_maybe_pull_tail(skb,
3949                                                   off +
3950                                                   sizeof(struct ip_auth_hdr),
3951                                                   MAX_IPV6_HDR_LEN);
3952                         if (err < 0)
3953                                 goto out;
3954 
3955                         hp = OPT_HDR(struct ip_auth_hdr, skb, off);
3956                         nexthdr = hp->nexthdr;
3957                         off += ipv6_authlen(hp);
3958                         break;
3959                 }
3960                 case IPPROTO_FRAGMENT: {
3961                         struct frag_hdr *hp;
3962 
3963                         err = skb_maybe_pull_tail(skb,
3964                                                   off +
3965                                                   sizeof(struct frag_hdr),
3966                                                   MAX_IPV6_HDR_LEN);
3967                         if (err < 0)
3968                                 goto out;
3969 
3970                         hp = OPT_HDR(struct frag_hdr, skb, off);
3971 
3972                         if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
3973                                 fragment = true;
3974 
3975                         nexthdr = hp->nexthdr;
3976                         off += sizeof(struct frag_hdr);
3977                         break;
3978                 }
3979                 default:
3980                         done = true;
3981                         break;
3982                 }
3983         }
3984 
3985         err = -EPROTO;
3986 
3987         if (!done || fragment)
3988                 goto out;
3989 
3990         csum = skb_checksum_setup_ip(skb, nexthdr, off);
3991         if (IS_ERR(csum))
3992                 return PTR_ERR(csum);
3993 
3994         if (recalculate)
3995                 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3996                                          &ipv6_hdr(skb)->daddr,
3997                                          skb->len - off, nexthdr, 0);
3998         err = 0;
3999 
4000 out:
4001         return err;
4002 }
4003 
4004 /**
4005  * skb_checksum_setup - set up partial checksum offset
4006  * @skb: the skb to set up
4007  * @recalculate: if true the pseudo-header checksum will be recalculated
4008  */
4009 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
4010 {
4011         int err;
4012 
4013         switch (skb->protocol) {
4014         case htons(ETH_P_IP):
4015                 err = skb_checksum_setup_ipv4(skb, recalculate);
4016                 break;
4017 
4018         case htons(ETH_P_IPV6):
4019                 err = skb_checksum_setup_ipv6(skb, recalculate);
4020                 break;
4021 
4022         default:
4023                 err = -EPROTO;
4024                 break;
4025         }
4026 
4027         return err;
4028 }
4029 EXPORT_SYMBOL(skb_checksum_setup);
4030 
4031 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
4032 {
4033         net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
4034                              skb->dev->name);
4035 }
4036 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
4037 
4038 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
4039 {
4040         if (head_stolen) {
4041                 skb_release_head_state(skb);
4042                 kmem_cache_free(skbuff_head_cache, skb);
4043         } else {
4044                 __kfree_skb(skb);
4045         }
4046 }
4047 EXPORT_SYMBOL(kfree_skb_partial);
4048 
4049 /**
4050  * skb_try_coalesce - try to merge skb to prior one
4051  * @to: prior buffer
4052  * @from: buffer to add
4053  * @fragstolen: pointer to boolean
4054  * @delta_truesize: how much more was allocated than was requested
4055  */
4056 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
4057                       bool *fragstolen, int *delta_truesize)
4058 {
4059         int i, delta, len = from->len;
4060 
4061         *fragstolen = false;
4062 
4063         if (skb_cloned(to))
4064                 return false;
4065 
4066         if (len <= skb_tailroom(to)) {
4067                 if (len)
4068                         BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
4069                 *delta_truesize = 0;
4070                 return true;
4071         }
4072 
4073         if (skb_has_frag_list(to) || skb_has_frag_list(from))
4074                 return false;
4075 
4076         if (skb_headlen(from) != 0) {
4077                 struct page *page;
4078                 unsigned int offset;
4079 
4080                 if (skb_shinfo(to)->nr_frags +
4081                     skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
4082                         return false;
4083 
4084                 if (skb_head_is_locked(from))
4085                         return false;
4086 
4087                 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4088 
4089                 page = virt_to_head_page(from->head);
4090                 offset = from->data - (unsigned char *)page_address(page);
4091 
4092                 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
4093                                    page, offset, skb_headlen(from));
4094                 *fragstolen = true;
4095         } else {
4096                 if (skb_shinfo(to)->nr_frags +
4097                     skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
4098                         return false;
4099 
4100                 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
4101         }
4102 
4103         WARN_ON_ONCE(delta < len);
4104 
4105         memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
4106                skb_shinfo(from)->frags,
4107                skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
4108         skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
4109 
4110         if (!skb_cloned(from))
4111                 skb_shinfo(from)->nr_frags = 0;
4112 
4113         /* if the skb is not cloned this does nothing
4114          * since we set nr_frags to 0.
4115          */
4116         for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
4117                 skb_frag_ref(from, i);
4118 
4119         to->truesize += delta;
4120         to->len += len;
4121         to->data_len += len;
4122 
4123         *delta_truesize = delta;
4124         return true;
4125 }
4126 EXPORT_SYMBOL(skb_try_coalesce);
4127 
4128 /**
4129  * skb_scrub_packet - scrub an skb
4130  *
4131  * @skb: buffer to clean
4132  * @xnet: packet is crossing netns
4133  *
4134  * skb_scrub_packet can be used after encapsulating or decapsulting a packet
4135  * into/from a tunnel. Some information have to be cleared during these
4136  * operations.
4137  * skb_scrub_packet can also be used to clean a skb before injecting it in
4138  * another namespace (@xnet == true). We have to clear all information in the
4139  * skb that could impact namespace isolation.
4140  */
4141 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
4142 {
4143         if (xnet)
4144                 skb_orphan(skb);
4145         skb->tstamp.tv64 = 0;
4146         skb->pkt_type = PACKET_HOST;
4147         skb->skb_iif = 0;
4148         skb->ignore_df = 0;
4149         skb_dst_drop(skb);
4150         skb->mark = 0;
4151         skb_init_secmark(skb);
4152         secpath_reset(skb);
4153         nf_reset(skb);
4154         nf_reset_trace(skb);
4155 }
4156 EXPORT_SYMBOL_GPL(skb_scrub_packet);
4157 
4158 /**
4159  * skb_gso_transport_seglen - Return length of individual segments of a gso packet
4160  *
4161  * @skb: GSO skb
4162  *
4163  * skb_gso_transport_seglen is used to determine the real size of the
4164  * individual segments, including Layer4 headers (TCP/UDP).
4165  *
4166  * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
4167  */
4168 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
4169 {
4170         const struct skb_shared_info *shinfo = skb_shinfo(skb);
4171         unsigned int thlen = 0;
4172 
4173         if (skb->encapsulation) {
4174                 thlen = skb_inner_transport_header(skb) -
4175                         skb_transport_header(skb);
4176 
4177                 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
4178                         thlen += inner_tcp_hdrlen(skb);
4179         } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
4180                 thlen = tcp_hdrlen(skb);
4181         }
4182         /* UFO sets gso_size to the size of the fragmentation
4183          * payload, i.e. the size of the L4 (UDP) header is already
4184          * accounted for.
4185          */
4186         return thlen + shinfo->gso_size;
4187 }
4188 EXPORT_SYMBOL_GPL(skb_gso_transport_seglen);
4189 
4190 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
4191 {
4192         if (skb_cow(skb, skb_headroom(skb)) < 0) {
4193                 kfree_skb(skb);
4194                 return NULL;
4195         }
4196 
4197         memmove(skb->data - ETH_HLEN, skb->data - VLAN_ETH_HLEN, 2 * ETH_ALEN);
4198         skb->mac_header += VLAN_HLEN;
4199         return skb;
4200 }
4201 
4202 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
4203 {
4204         struct vlan_hdr *vhdr;
4205         u16 vlan_tci;
4206 
4207         if (unlikely(vlan_tx_tag_present(skb))) {
4208                 /* vlan_tci is already set-up so leave this for another time */
4209                 return skb;
4210         }
4211 
4212         skb = skb_share_check(skb, GFP_ATOMIC);
4213         if (unlikely(!skb))
4214                 goto err_free;
4215 
4216         if (unlikely(!pskb_may_pull(skb, VLAN_HLEN)))
4217                 goto err_free;
4218 
4219         vhdr = (struct vlan_hdr *)skb->data;
4220         vlan_tci = ntohs(vhdr->h_vlan_TCI);
4221         __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
4222 
4223         skb_pull_rcsum(skb, VLAN_HLEN);
4224         vlan_set_encap_proto(skb, vhdr);
4225 
4226         skb = skb_reorder_vlan_header(skb);
4227         if (unlikely(!skb))
4228                 goto err_free;
4229 
4230         skb_reset_network_header(skb);
4231         skb_reset_transport_header(skb);
4232         skb_reset_mac_len(skb);
4233 
4234         return skb;
4235 
4236 err_free:
4237         kfree_skb(skb);
4238         return NULL;
4239 }
4240 EXPORT_SYMBOL(skb_vlan_untag);
4241 
4242 int skb_ensure_writable(struct sk_buff *skb, int write_len)
4243 {
4244         if (!pskb_may_pull(skb, write_len))
4245                 return -ENOMEM;
4246 
4247         if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
4248                 return 0;
4249 
4250         return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4251 }
4252 EXPORT_SYMBOL(skb_ensure_writable);
4253 
4254 /* remove VLAN header from packet and update csum accordingly. */
4255 static int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
4256 {
4257         struct vlan_hdr *vhdr;
4258         unsigned int offset = skb->data - skb_mac_header(skb);
4259         int err;
4260 
4261         __skb_push(skb, offset);
4262         err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
4263         if (unlikely(err))
4264                 goto pull;
4265 
4266         skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
4267 
4268         vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
4269         *vlan_tci = ntohs(vhdr->h_vlan_TCI);
4270 
4271         memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
4272         __skb_pull(skb, VLAN_HLEN);
4273 
4274         vlan_set_encap_proto(skb, vhdr);
4275         skb->mac_header += VLAN_HLEN;
4276 
4277         if (skb_network_offset(skb) < ETH_HLEN)
4278                 skb_set_network_header(skb, ETH_HLEN);
4279 
4280         skb_reset_mac_len(skb);
4281 pull:
4282         __skb_pull(skb, offset);
4283 
4284         return err;
4285 }
4286 
4287 int skb_vlan_pop(struct sk_buff *skb)
4288 {
4289         u16 vlan_tci;
4290         __be16 vlan_proto;
4291         int err;
4292 
4293         if (likely(vlan_tx_tag_present(skb))) {
4294                 skb->vlan_tci = 0;
4295         } else {
4296                 if (unlikely((skb->protocol != htons(ETH_P_8021Q) &&
4297                               skb->protocol != htons(ETH_P_8021AD)) ||
4298                              skb->len < VLAN_ETH_HLEN))
4299                         return 0;
4300 
4301                 err = __skb_vlan_pop(skb, &vlan_tci);
4302                 if (err)
4303                         return err;
4304         }
4305         /* move next vlan tag to hw accel tag */
4306         if (likely((skb->protocol != htons(ETH_P_8021Q) &&
4307                     skb->protocol != htons(ETH_P_8021AD)) ||
4308                    skb->len < VLAN_ETH_HLEN))
4309                 return 0;
4310 
4311         vlan_proto = skb->protocol;
4312         err = __skb_vlan_pop(skb, &vlan_tci);
4313         if (unlikely(err))
4314                 return err;
4315 
4316         __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
4317         return 0;
4318 }
4319 EXPORT_SYMBOL(skb_vlan_pop);
4320 
4321 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
4322 {
4323         if (vlan_tx_tag_present(skb)) {
4324                 unsigned int offset = skb->data - skb_mac_header(skb);
4325                 int err;
4326 
4327                 /* __vlan_insert_tag expect skb->data pointing to mac header.
4328                  * So change skb->data before calling it and change back to
4329                  * original position later
4330                  */
4331                 __skb_push(skb, offset);
4332                 err = __vlan_insert_tag(skb, skb->vlan_proto,
4333                                         vlan_tx_tag_get(skb));
4334                 if (err)
4335                         return err;
4336                 skb->protocol = skb->vlan_proto;
4337                 skb->mac_len += VLAN_HLEN;
4338                 __skb_pull(skb, offset);
4339 
4340                 if (skb->ip_summed == CHECKSUM_COMPLETE)
4341                         skb->csum = csum_add(skb->csum, csum_partial(skb->data
4342                                         + (2 * ETH_ALEN), VLAN_HLEN, 0));
4343         }
4344         __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
4345         return 0;
4346 }
4347 EXPORT_SYMBOL(skb_vlan_push);
4348 
4349 /**
4350  * alloc_skb_with_frags - allocate skb with page frags
4351  *
4352  * @header_len: size of linear part
4353  * @data_len: needed length in frags
4354  * @max_page_order: max page order desired.
4355  * @errcode: pointer to error code if any
4356  * @gfp_mask: allocation mask
4357  *
4358  * This can be used to allocate a paged skb, given a maximal order for frags.
4359  */
4360 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
4361                                      unsigned long data_len,
4362                                      int max_page_order,
4363                                      int *errcode,
4364                                      gfp_t gfp_mask)
4365 {
4366         int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
4367         unsigned long chunk;
4368         struct sk_buff *skb;
4369         struct page *page;
4370         gfp_t gfp_head;
4371         int i;
4372 
4373         *errcode = -EMSGSIZE;
4374         /* Note this test could be relaxed, if we succeed to allocate
4375          * high order pages...
4376          */
4377         if (npages > MAX_SKB_FRAGS)
4378                 return NULL;
4379 
4380         gfp_head = gfp_mask;
4381         if (gfp_head & __GFP_WAIT)
4382                 gfp_head |= __GFP_REPEAT;
4383 
4384         *errcode = -ENOBUFS;
4385         skb = alloc_skb(header_len, gfp_head);
4386         if (!skb)
4387                 return NULL;
4388 
4389         skb->truesize += npages << PAGE_SHIFT;
4390 
4391         for (i = 0; npages > 0; i++) {
4392                 int order = max_page_order;
4393 
4394                 while (order) {
4395                         if (npages >= 1 << order) {
4396                                 page = alloc_pages(gfp_mask |
4397                                                    __GFP_COMP |
4398                                                    __GFP_NOWARN |
4399                                                    __GFP_NORETRY,
4400                                                    order);
4401                                 if (page)
4402                                         goto fill_page;
4403                                 /* Do not retry other high order allocations */
4404                                 order = 1;
4405                                 max_page_order = 0;
4406                         }
4407                         order--;
4408                 }
4409                 page = alloc_page(gfp_mask);
4410                 if (!page)
4411                         goto failure;
4412 fill_page:
4413                 chunk = min_t(unsigned long, data_len,
4414                               PAGE_SIZE << order);
4415                 skb_fill_page_desc(skb, i, page, 0, chunk);
4416                 data_len -= chunk;
4417                 npages -= 1 << order;
4418         }
4419         return skb;
4420 
4421 failure:
4422         kfree_skb(skb);
4423         return NULL;
4424 }
4425 EXPORT_SYMBOL(alloc_skb_with_frags);
4426 

This page was automatically generated by LXR 0.3.1 (source).  •  Linux is a registered trademark of Linus Torvalds  •  Contact us