Version:  2.0.40 2.2.26 2.4.37 3.1 3.2 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

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 *child = skb + 1;
261                 atomic_t *fclone_ref = (atomic_t *) (child + 1);
262 
263                 kmemcheck_annotate_bitfield(child, flags1);
264                 kmemcheck_annotate_bitfield(child, flags2);
265                 skb->fclone = SKB_FCLONE_ORIG;
266                 atomic_set(fclone_ref, 1);
267 
268                 child->fclone = SKB_FCLONE_UNAVAILABLE;
269                 child->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 
340 static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
341 {
342         struct netdev_alloc_cache *nc;
343         void *data = NULL;
344         int order;
345         unsigned long flags;
346 
347         local_irq_save(flags);
348         nc = &__get_cpu_var(netdev_alloc_cache);
349         if (unlikely(!nc->frag.page)) {
350 refill:
351                 for (order = NETDEV_FRAG_PAGE_MAX_ORDER; ;) {
352                         gfp_t gfp = gfp_mask;
353 
354                         if (order)
355                                 gfp |= __GFP_COMP | __GFP_NOWARN;
356                         nc->frag.page = alloc_pages(gfp, order);
357                         if (likely(nc->frag.page))
358                                 break;
359                         if (--order < 0)
360                                 goto end;
361                 }
362                 nc->frag.size = PAGE_SIZE << order;
363 recycle:
364                 atomic_set(&nc->frag.page->_count, NETDEV_PAGECNT_MAX_BIAS);
365                 nc->pagecnt_bias = NETDEV_PAGECNT_MAX_BIAS;
366                 nc->frag.offset = 0;
367         }
368 
369         if (nc->frag.offset + fragsz > nc->frag.size) {
370                 /* avoid unnecessary locked operations if possible */
371                 if ((atomic_read(&nc->frag.page->_count) == nc->pagecnt_bias) ||
372                     atomic_sub_and_test(nc->pagecnt_bias, &nc->frag.page->_count))
373                         goto recycle;
374                 goto refill;
375         }
376 
377         data = page_address(nc->frag.page) + nc->frag.offset;
378         nc->frag.offset += fragsz;
379         nc->pagecnt_bias--;
380 end:
381         local_irq_restore(flags);
382         return data;
383 }
384 
385 /**
386  * netdev_alloc_frag - allocate a page fragment
387  * @fragsz: fragment size
388  *
389  * Allocates a frag from a page for receive buffer.
390  * Uses GFP_ATOMIC allocations.
391  */
392 void *netdev_alloc_frag(unsigned int fragsz)
393 {
394         return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
395 }
396 EXPORT_SYMBOL(netdev_alloc_frag);
397 
398 /**
399  *      __netdev_alloc_skb - allocate an skbuff for rx on a specific device
400  *      @dev: network device to receive on
401  *      @length: length to allocate
402  *      @gfp_mask: get_free_pages mask, passed to alloc_skb
403  *
404  *      Allocate a new &sk_buff and assign it a usage count of one. The
405  *      buffer has unspecified headroom built in. Users should allocate
406  *      the headroom they think they need without accounting for the
407  *      built in space. The built in space is used for optimisations.
408  *
409  *      %NULL is returned if there is no free memory.
410  */
411 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
412                                    unsigned int length, gfp_t gfp_mask)
413 {
414         struct sk_buff *skb = NULL;
415         unsigned int fragsz = SKB_DATA_ALIGN(length + NET_SKB_PAD) +
416                               SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
417 
418         if (fragsz <= PAGE_SIZE && !(gfp_mask & (__GFP_WAIT | GFP_DMA))) {
419                 void *data;
420 
421                 if (sk_memalloc_socks())
422                         gfp_mask |= __GFP_MEMALLOC;
423 
424                 data = __netdev_alloc_frag(fragsz, gfp_mask);
425 
426                 if (likely(data)) {
427                         skb = build_skb(data, fragsz);
428                         if (unlikely(!skb))
429                                 put_page(virt_to_head_page(data));
430                 }
431         } else {
432                 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask,
433                                   SKB_ALLOC_RX, NUMA_NO_NODE);
434         }
435         if (likely(skb)) {
436                 skb_reserve(skb, NET_SKB_PAD);
437                 skb->dev = dev;
438         }
439         return skb;
440 }
441 EXPORT_SYMBOL(__netdev_alloc_skb);
442 
443 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
444                      int size, unsigned int truesize)
445 {
446         skb_fill_page_desc(skb, i, page, off, size);
447         skb->len += size;
448         skb->data_len += size;
449         skb->truesize += truesize;
450 }
451 EXPORT_SYMBOL(skb_add_rx_frag);
452 
453 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
454                           unsigned int truesize)
455 {
456         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
457 
458         skb_frag_size_add(frag, size);
459         skb->len += size;
460         skb->data_len += size;
461         skb->truesize += truesize;
462 }
463 EXPORT_SYMBOL(skb_coalesce_rx_frag);
464 
465 static void skb_drop_list(struct sk_buff **listp)
466 {
467         kfree_skb_list(*listp);
468         *listp = NULL;
469 }
470 
471 static inline void skb_drop_fraglist(struct sk_buff *skb)
472 {
473         skb_drop_list(&skb_shinfo(skb)->frag_list);
474 }
475 
476 static void skb_clone_fraglist(struct sk_buff *skb)
477 {
478         struct sk_buff *list;
479 
480         skb_walk_frags(skb, list)
481                 skb_get(list);
482 }
483 
484 static void skb_free_head(struct sk_buff *skb)
485 {
486         if (skb->head_frag)
487                 put_page(virt_to_head_page(skb->head));
488         else
489                 kfree(skb->head);
490 }
491 
492 static void skb_release_data(struct sk_buff *skb)
493 {
494         if (!skb->cloned ||
495             !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
496                                &skb_shinfo(skb)->dataref)) {
497                 if (skb_shinfo(skb)->nr_frags) {
498                         int i;
499                         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
500                                 skb_frag_unref(skb, i);
501                 }
502 
503                 /*
504                  * If skb buf is from userspace, we need to notify the caller
505                  * the lower device DMA has done;
506                  */
507                 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
508                         struct ubuf_info *uarg;
509 
510                         uarg = skb_shinfo(skb)->destructor_arg;
511                         if (uarg->callback)
512                                 uarg->callback(uarg, true);
513                 }
514 
515                 if (skb_has_frag_list(skb))
516                         skb_drop_fraglist(skb);
517 
518                 skb_free_head(skb);
519         }
520 }
521 
522 /*
523  *      Free an skbuff by memory without cleaning the state.
524  */
525 static void kfree_skbmem(struct sk_buff *skb)
526 {
527         struct sk_buff *other;
528         atomic_t *fclone_ref;
529 
530         switch (skb->fclone) {
531         case SKB_FCLONE_UNAVAILABLE:
532                 kmem_cache_free(skbuff_head_cache, skb);
533                 break;
534 
535         case SKB_FCLONE_ORIG:
536                 fclone_ref = (atomic_t *) (skb + 2);
537                 if (atomic_dec_and_test(fclone_ref))
538                         kmem_cache_free(skbuff_fclone_cache, skb);
539                 break;
540 
541         case SKB_FCLONE_CLONE:
542                 fclone_ref = (atomic_t *) (skb + 1);
543                 other = skb - 1;
544 
545                 /* The clone portion is available for
546                  * fast-cloning again.
547                  */
548                 skb->fclone = SKB_FCLONE_UNAVAILABLE;
549 
550                 if (atomic_dec_and_test(fclone_ref))
551                         kmem_cache_free(skbuff_fclone_cache, other);
552                 break;
553         }
554 }
555 
556 static void skb_release_head_state(struct sk_buff *skb)
557 {
558         skb_dst_drop(skb);
559 #ifdef CONFIG_XFRM
560         secpath_put(skb->sp);
561 #endif
562         if (skb->destructor) {
563                 WARN_ON(in_irq());
564                 skb->destructor(skb);
565         }
566 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
567         nf_conntrack_put(skb->nfct);
568 #endif
569 #ifdef CONFIG_BRIDGE_NETFILTER
570         nf_bridge_put(skb->nf_bridge);
571 #endif
572 /* XXX: IS this still necessary? - JHS */
573 #ifdef CONFIG_NET_SCHED
574         skb->tc_index = 0;
575 #ifdef CONFIG_NET_CLS_ACT
576         skb->tc_verd = 0;
577 #endif
578 #endif
579 }
580 
581 /* Free everything but the sk_buff shell. */
582 static void skb_release_all(struct sk_buff *skb)
583 {
584         skb_release_head_state(skb);
585         if (likely(skb->head))
586                 skb_release_data(skb);
587 }
588 
589 /**
590  *      __kfree_skb - private function
591  *      @skb: buffer
592  *
593  *      Free an sk_buff. Release anything attached to the buffer.
594  *      Clean the state. This is an internal helper function. Users should
595  *      always call kfree_skb
596  */
597 
598 void __kfree_skb(struct sk_buff *skb)
599 {
600         skb_release_all(skb);
601         kfree_skbmem(skb);
602 }
603 EXPORT_SYMBOL(__kfree_skb);
604 
605 /**
606  *      kfree_skb - free an sk_buff
607  *      @skb: buffer to free
608  *
609  *      Drop a reference to the buffer and free it if the usage count has
610  *      hit zero.
611  */
612 void kfree_skb(struct sk_buff *skb)
613 {
614         if (unlikely(!skb))
615                 return;
616         if (likely(atomic_read(&skb->users) == 1))
617                 smp_rmb();
618         else if (likely(!atomic_dec_and_test(&skb->users)))
619                 return;
620         trace_kfree_skb(skb, __builtin_return_address(0));
621         __kfree_skb(skb);
622 }
623 EXPORT_SYMBOL(kfree_skb);
624 
625 void kfree_skb_list(struct sk_buff *segs)
626 {
627         while (segs) {
628                 struct sk_buff *next = segs->next;
629 
630                 kfree_skb(segs);
631                 segs = next;
632         }
633 }
634 EXPORT_SYMBOL(kfree_skb_list);
635 
636 /**
637  *      skb_tx_error - report an sk_buff xmit error
638  *      @skb: buffer that triggered an error
639  *
640  *      Report xmit error if a device callback is tracking this skb.
641  *      skb must be freed afterwards.
642  */
643 void skb_tx_error(struct sk_buff *skb)
644 {
645         if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
646                 struct ubuf_info *uarg;
647 
648                 uarg = skb_shinfo(skb)->destructor_arg;
649                 if (uarg->callback)
650                         uarg->callback(uarg, false);
651                 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
652         }
653 }
654 EXPORT_SYMBOL(skb_tx_error);
655 
656 /**
657  *      consume_skb - free an skbuff
658  *      @skb: buffer to free
659  *
660  *      Drop a ref to the buffer and free it if the usage count has hit zero
661  *      Functions identically to kfree_skb, but kfree_skb assumes that the frame
662  *      is being dropped after a failure and notes that
663  */
664 void consume_skb(struct sk_buff *skb)
665 {
666         if (unlikely(!skb))
667                 return;
668         if (likely(atomic_read(&skb->users) == 1))
669                 smp_rmb();
670         else if (likely(!atomic_dec_and_test(&skb->users)))
671                 return;
672         trace_consume_skb(skb);
673         __kfree_skb(skb);
674 }
675 EXPORT_SYMBOL(consume_skb);
676 
677 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
678 {
679         new->tstamp             = old->tstamp;
680         new->dev                = old->dev;
681         new->transport_header   = old->transport_header;
682         new->network_header     = old->network_header;
683         new->mac_header         = old->mac_header;
684         new->inner_protocol     = old->inner_protocol;
685         new->inner_transport_header = old->inner_transport_header;
686         new->inner_network_header = old->inner_network_header;
687         new->inner_mac_header = old->inner_mac_header;
688         skb_dst_copy(new, old);
689         skb_copy_hash(new, old);
690         new->ooo_okay           = old->ooo_okay;
691         new->no_fcs             = old->no_fcs;
692         new->encapsulation      = old->encapsulation;
693         new->encap_hdr_csum     = old->encap_hdr_csum;
694         new->csum_valid         = old->csum_valid;
695         new->csum_complete_sw   = old->csum_complete_sw;
696 #ifdef CONFIG_XFRM
697         new->sp                 = secpath_get(old->sp);
698 #endif
699         memcpy(new->cb, old->cb, sizeof(old->cb));
700         new->csum               = old->csum;
701         new->ignore_df          = old->ignore_df;
702         new->pkt_type           = old->pkt_type;
703         new->ip_summed          = old->ip_summed;
704         skb_copy_queue_mapping(new, old);
705         new->priority           = old->priority;
706 #if IS_ENABLED(CONFIG_IP_VS)
707         new->ipvs_property      = old->ipvs_property;
708 #endif
709         new->pfmemalloc         = old->pfmemalloc;
710         new->protocol           = old->protocol;
711         new->mark               = old->mark;
712         new->skb_iif            = old->skb_iif;
713         __nf_copy(new, old);
714 #ifdef CONFIG_NET_SCHED
715         new->tc_index           = old->tc_index;
716 #ifdef CONFIG_NET_CLS_ACT
717         new->tc_verd            = old->tc_verd;
718 #endif
719 #endif
720         new->vlan_proto         = old->vlan_proto;
721         new->vlan_tci           = old->vlan_tci;
722 
723         skb_copy_secmark(new, old);
724 
725 #ifdef CONFIG_NET_RX_BUSY_POLL
726         new->napi_id    = old->napi_id;
727 #endif
728 }
729 
730 /*
731  * You should not add any new code to this function.  Add it to
732  * __copy_skb_header above instead.
733  */
734 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
735 {
736 #define C(x) n->x = skb->x
737 
738         n->next = n->prev = NULL;
739         n->sk = NULL;
740         __copy_skb_header(n, skb);
741 
742         C(len);
743         C(data_len);
744         C(mac_len);
745         n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
746         n->cloned = 1;
747         n->nohdr = 0;
748         n->destructor = NULL;
749         C(tail);
750         C(end);
751         C(head);
752         C(head_frag);
753         C(data);
754         C(truesize);
755         atomic_set(&n->users, 1);
756 
757         atomic_inc(&(skb_shinfo(skb)->dataref));
758         skb->cloned = 1;
759 
760         return n;
761 #undef C
762 }
763 
764 /**
765  *      skb_morph       -       morph one skb into another
766  *      @dst: the skb to receive the contents
767  *      @src: the skb to supply the contents
768  *
769  *      This is identical to skb_clone except that the target skb is
770  *      supplied by the user.
771  *
772  *      The target skb is returned upon exit.
773  */
774 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
775 {
776         skb_release_all(dst);
777         return __skb_clone(dst, src);
778 }
779 EXPORT_SYMBOL_GPL(skb_morph);
780 
781 /**
782  *      skb_copy_ubufs  -       copy userspace skb frags buffers to kernel
783  *      @skb: the skb to modify
784  *      @gfp_mask: allocation priority
785  *
786  *      This must be called on SKBTX_DEV_ZEROCOPY skb.
787  *      It will copy all frags into kernel and drop the reference
788  *      to userspace pages.
789  *
790  *      If this function is called from an interrupt gfp_mask() must be
791  *      %GFP_ATOMIC.
792  *
793  *      Returns 0 on success or a negative error code on failure
794  *      to allocate kernel memory to copy to.
795  */
796 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
797 {
798         int i;
799         int num_frags = skb_shinfo(skb)->nr_frags;
800         struct page *page, *head = NULL;
801         struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
802 
803         for (i = 0; i < num_frags; i++) {
804                 u8 *vaddr;
805                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
806 
807                 page = alloc_page(gfp_mask);
808                 if (!page) {
809                         while (head) {
810                                 struct page *next = (struct page *)page_private(head);
811                                 put_page(head);
812                                 head = next;
813                         }
814                         return -ENOMEM;
815                 }
816                 vaddr = kmap_atomic(skb_frag_page(f));
817                 memcpy(page_address(page),
818                        vaddr + f->page_offset, skb_frag_size(f));
819                 kunmap_atomic(vaddr);
820                 set_page_private(page, (unsigned long)head);
821                 head = page;
822         }
823 
824         /* skb frags release userspace buffers */
825         for (i = 0; i < num_frags; i++)
826                 skb_frag_unref(skb, i);
827 
828         uarg->callback(uarg, false);
829 
830         /* skb frags point to kernel buffers */
831         for (i = num_frags - 1; i >= 0; i--) {
832                 __skb_fill_page_desc(skb, i, head, 0,
833                                      skb_shinfo(skb)->frags[i].size);
834                 head = (struct page *)page_private(head);
835         }
836 
837         skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
838         return 0;
839 }
840 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
841 
842 /**
843  *      skb_clone       -       duplicate an sk_buff
844  *      @skb: buffer to clone
845  *      @gfp_mask: allocation priority
846  *
847  *      Duplicate an &sk_buff. The new one is not owned by a socket. Both
848  *      copies share the same packet data but not structure. The new
849  *      buffer has a reference count of 1. If the allocation fails the
850  *      function returns %NULL otherwise the new buffer is returned.
851  *
852  *      If this function is called from an interrupt gfp_mask() must be
853  *      %GFP_ATOMIC.
854  */
855 
856 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
857 {
858         struct sk_buff *n;
859 
860         if (skb_orphan_frags(skb, gfp_mask))
861                 return NULL;
862 
863         n = skb + 1;
864         if (skb->fclone == SKB_FCLONE_ORIG &&
865             n->fclone == SKB_FCLONE_UNAVAILABLE) {
866                 atomic_t *fclone_ref = (atomic_t *) (n + 1);
867                 n->fclone = SKB_FCLONE_CLONE;
868                 atomic_inc(fclone_ref);
869         } else {
870                 if (skb_pfmemalloc(skb))
871                         gfp_mask |= __GFP_MEMALLOC;
872 
873                 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
874                 if (!n)
875                         return NULL;
876 
877                 kmemcheck_annotate_bitfield(n, flags1);
878                 kmemcheck_annotate_bitfield(n, flags2);
879                 n->fclone = SKB_FCLONE_UNAVAILABLE;
880         }
881 
882         return __skb_clone(n, skb);
883 }
884 EXPORT_SYMBOL(skb_clone);
885 
886 static void skb_headers_offset_update(struct sk_buff *skb, int off)
887 {
888         /* Only adjust this if it actually is csum_start rather than csum */
889         if (skb->ip_summed == CHECKSUM_PARTIAL)
890                 skb->csum_start += off;
891         /* {transport,network,mac}_header and tail are relative to skb->head */
892         skb->transport_header += off;
893         skb->network_header   += off;
894         if (skb_mac_header_was_set(skb))
895                 skb->mac_header += off;
896         skb->inner_transport_header += off;
897         skb->inner_network_header += off;
898         skb->inner_mac_header += off;
899 }
900 
901 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
902 {
903         __copy_skb_header(new, old);
904 
905         skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
906         skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
907         skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
908 }
909 
910 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
911 {
912         if (skb_pfmemalloc(skb))
913                 return SKB_ALLOC_RX;
914         return 0;
915 }
916 
917 /**
918  *      skb_copy        -       create private copy of an sk_buff
919  *      @skb: buffer to copy
920  *      @gfp_mask: allocation priority
921  *
922  *      Make a copy of both an &sk_buff and its data. This is used when the
923  *      caller wishes to modify the data and needs a private copy of the
924  *      data to alter. Returns %NULL on failure or the pointer to the buffer
925  *      on success. The returned buffer has a reference count of 1.
926  *
927  *      As by-product this function converts non-linear &sk_buff to linear
928  *      one, so that &sk_buff becomes completely private and caller is allowed
929  *      to modify all the data of returned buffer. This means that this
930  *      function is not recommended for use in circumstances when only
931  *      header is going to be modified. Use pskb_copy() instead.
932  */
933 
934 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
935 {
936         int headerlen = skb_headroom(skb);
937         unsigned int size = skb_end_offset(skb) + skb->data_len;
938         struct sk_buff *n = __alloc_skb(size, gfp_mask,
939                                         skb_alloc_rx_flag(skb), NUMA_NO_NODE);
940 
941         if (!n)
942                 return NULL;
943 
944         /* Set the data pointer */
945         skb_reserve(n, headerlen);
946         /* Set the tail pointer and length */
947         skb_put(n, skb->len);
948 
949         if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
950                 BUG();
951 
952         copy_skb_header(n, skb);
953         return n;
954 }
955 EXPORT_SYMBOL(skb_copy);
956 
957 /**
958  *      __pskb_copy_fclone      -  create copy of an sk_buff with private head.
959  *      @skb: buffer to copy
960  *      @headroom: headroom of new skb
961  *      @gfp_mask: allocation priority
962  *      @fclone: if true allocate the copy of the skb from the fclone
963  *      cache instead of the head cache; it is recommended to set this
964  *      to true for the cases where the copy will likely be cloned
965  *
966  *      Make a copy of both an &sk_buff and part of its data, located
967  *      in header. Fragmented data remain shared. This is used when
968  *      the caller wishes to modify only header of &sk_buff and needs
969  *      private copy of the header to alter. Returns %NULL on failure
970  *      or the pointer to the buffer on success.
971  *      The returned buffer has a reference count of 1.
972  */
973 
974 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
975                                    gfp_t gfp_mask, bool fclone)
976 {
977         unsigned int size = skb_headlen(skb) + headroom;
978         int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
979         struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
980 
981         if (!n)
982                 goto out;
983 
984         /* Set the data pointer */
985         skb_reserve(n, headroom);
986         /* Set the tail pointer and length */
987         skb_put(n, skb_headlen(skb));
988         /* Copy the bytes */
989         skb_copy_from_linear_data(skb, n->data, n->len);
990 
991         n->truesize += skb->data_len;
992         n->data_len  = skb->data_len;
993         n->len       = skb->len;
994 
995         if (skb_shinfo(skb)->nr_frags) {
996                 int i;
997 
998                 if (skb_orphan_frags(skb, gfp_mask)) {
999                         kfree_skb(n);
1000                         n = NULL;
1001                         goto out;
1002                 }
1003                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1004                         skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1005                         skb_frag_ref(skb, i);
1006                 }
1007                 skb_shinfo(n)->nr_frags = i;
1008         }
1009 
1010         if (skb_has_frag_list(skb)) {
1011                 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1012                 skb_clone_fraglist(n);
1013         }
1014 
1015         copy_skb_header(n, skb);
1016 out:
1017         return n;
1018 }
1019 EXPORT_SYMBOL(__pskb_copy_fclone);
1020 
1021 /**
1022  *      pskb_expand_head - reallocate header of &sk_buff
1023  *      @skb: buffer to reallocate
1024  *      @nhead: room to add at head
1025  *      @ntail: room to add at tail
1026  *      @gfp_mask: allocation priority
1027  *
1028  *      Expands (or creates identical copy, if @nhead and @ntail are zero)
1029  *      header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1030  *      reference count of 1. Returns zero in the case of success or error,
1031  *      if expansion failed. In the last case, &sk_buff is not changed.
1032  *
1033  *      All the pointers pointing into skb header may change and must be
1034  *      reloaded after call to this function.
1035  */
1036 
1037 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1038                      gfp_t gfp_mask)
1039 {
1040         int i;
1041         u8 *data;
1042         int size = nhead + skb_end_offset(skb) + ntail;
1043         long off;
1044 
1045         BUG_ON(nhead < 0);
1046 
1047         if (skb_shared(skb))
1048                 BUG();
1049 
1050         size = SKB_DATA_ALIGN(size);
1051 
1052         if (skb_pfmemalloc(skb))
1053                 gfp_mask |= __GFP_MEMALLOC;
1054         data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1055                                gfp_mask, NUMA_NO_NODE, NULL);
1056         if (!data)
1057                 goto nodata;
1058         size = SKB_WITH_OVERHEAD(ksize(data));
1059 
1060         /* Copy only real data... and, alas, header. This should be
1061          * optimized for the cases when header is void.
1062          */
1063         memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1064 
1065         memcpy((struct skb_shared_info *)(data + size),
1066                skb_shinfo(skb),
1067                offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1068 
1069         /*
1070          * if shinfo is shared we must drop the old head gracefully, but if it
1071          * is not we can just drop the old head and let the existing refcount
1072          * be since all we did is relocate the values
1073          */
1074         if (skb_cloned(skb)) {
1075                 /* copy this zero copy skb frags */
1076                 if (skb_orphan_frags(skb, gfp_mask))
1077                         goto nofrags;
1078                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1079                         skb_frag_ref(skb, i);
1080 
1081                 if (skb_has_frag_list(skb))
1082                         skb_clone_fraglist(skb);
1083 
1084                 skb_release_data(skb);
1085         } else {
1086                 skb_free_head(skb);
1087         }
1088         off = (data + nhead) - skb->head;
1089 
1090         skb->head     = data;
1091         skb->head_frag = 0;
1092         skb->data    += off;
1093 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1094         skb->end      = size;
1095         off           = nhead;
1096 #else
1097         skb->end      = skb->head + size;
1098 #endif
1099         skb->tail             += off;
1100         skb_headers_offset_update(skb, nhead);
1101         skb->cloned   = 0;
1102         skb->hdr_len  = 0;
1103         skb->nohdr    = 0;
1104         atomic_set(&skb_shinfo(skb)->dataref, 1);
1105         return 0;
1106 
1107 nofrags:
1108         kfree(data);
1109 nodata:
1110         return -ENOMEM;
1111 }
1112 EXPORT_SYMBOL(pskb_expand_head);
1113 
1114 /* Make private copy of skb with writable head and some headroom */
1115 
1116 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1117 {
1118         struct sk_buff *skb2;
1119         int delta = headroom - skb_headroom(skb);
1120 
1121         if (delta <= 0)
1122                 skb2 = pskb_copy(skb, GFP_ATOMIC);
1123         else {
1124                 skb2 = skb_clone(skb, GFP_ATOMIC);
1125                 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1126                                              GFP_ATOMIC)) {
1127                         kfree_skb(skb2);
1128                         skb2 = NULL;
1129                 }
1130         }
1131         return skb2;
1132 }
1133 EXPORT_SYMBOL(skb_realloc_headroom);
1134 
1135 /**
1136  *      skb_copy_expand -       copy and expand sk_buff
1137  *      @skb: buffer to copy
1138  *      @newheadroom: new free bytes at head
1139  *      @newtailroom: new free bytes at tail
1140  *      @gfp_mask: allocation priority
1141  *
1142  *      Make a copy of both an &sk_buff and its data and while doing so
1143  *      allocate additional space.
1144  *
1145  *      This is used when the caller wishes to modify the data and needs a
1146  *      private copy of the data to alter as well as more space for new fields.
1147  *      Returns %NULL on failure or the pointer to the buffer
1148  *      on success. The returned buffer has a reference count of 1.
1149  *
1150  *      You must pass %GFP_ATOMIC as the allocation priority if this function
1151  *      is called from an interrupt.
1152  */
1153 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1154                                 int newheadroom, int newtailroom,
1155                                 gfp_t gfp_mask)
1156 {
1157         /*
1158          *      Allocate the copy buffer
1159          */
1160         struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1161                                         gfp_mask, skb_alloc_rx_flag(skb),
1162                                         NUMA_NO_NODE);
1163         int oldheadroom = skb_headroom(skb);
1164         int head_copy_len, head_copy_off;
1165 
1166         if (!n)
1167                 return NULL;
1168 
1169         skb_reserve(n, newheadroom);
1170 
1171         /* Set the tail pointer and length */
1172         skb_put(n, skb->len);
1173 
1174         head_copy_len = oldheadroom;
1175         head_copy_off = 0;
1176         if (newheadroom <= head_copy_len)
1177                 head_copy_len = newheadroom;
1178         else
1179                 head_copy_off = newheadroom - head_copy_len;
1180 
1181         /* Copy the linear header and data. */
1182         if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1183                           skb->len + head_copy_len))
1184                 BUG();
1185 
1186         copy_skb_header(n, skb);
1187 
1188         skb_headers_offset_update(n, newheadroom - oldheadroom);
1189 
1190         return n;
1191 }
1192 EXPORT_SYMBOL(skb_copy_expand);
1193 
1194 /**
1195  *      skb_pad                 -       zero pad the tail of an skb
1196  *      @skb: buffer to pad
1197  *      @pad: space to pad
1198  *
1199  *      Ensure that a buffer is followed by a padding area that is zero
1200  *      filled. Used by network drivers which may DMA or transfer data
1201  *      beyond the buffer end onto the wire.
1202  *
1203  *      May return error in out of memory cases. The skb is freed on error.
1204  */
1205 
1206 int skb_pad(struct sk_buff *skb, int pad)
1207 {
1208         int err;
1209         int ntail;
1210 
1211         /* If the skbuff is non linear tailroom is always zero.. */
1212         if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1213                 memset(skb->data+skb->len, 0, pad);
1214                 return 0;
1215         }
1216 
1217         ntail = skb->data_len + pad - (skb->end - skb->tail);
1218         if (likely(skb_cloned(skb) || ntail > 0)) {
1219                 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1220                 if (unlikely(err))
1221                         goto free_skb;
1222         }
1223 
1224         /* FIXME: The use of this function with non-linear skb's really needs
1225          * to be audited.
1226          */
1227         err = skb_linearize(skb);
1228         if (unlikely(err))
1229                 goto free_skb;
1230 
1231         memset(skb->data + skb->len, 0, pad);
1232         return 0;
1233 
1234 free_skb:
1235         kfree_skb(skb);
1236         return err;
1237 }
1238 EXPORT_SYMBOL(skb_pad);
1239 
1240 /**
1241  *      pskb_put - add data to the tail of a potentially fragmented buffer
1242  *      @skb: start of the buffer to use
1243  *      @tail: tail fragment of the buffer to use
1244  *      @len: amount of data to add
1245  *
1246  *      This function extends the used data area of the potentially
1247  *      fragmented buffer. @tail must be the last fragment of @skb -- or
1248  *      @skb itself. If this would exceed the total buffer size the kernel
1249  *      will panic. A pointer to the first byte of the extra data is
1250  *      returned.
1251  */
1252 
1253 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1254 {
1255         if (tail != skb) {
1256                 skb->data_len += len;
1257                 skb->len += len;
1258         }
1259         return skb_put(tail, len);
1260 }
1261 EXPORT_SYMBOL_GPL(pskb_put);
1262 
1263 /**
1264  *      skb_put - add data to a buffer
1265  *      @skb: buffer to use
1266  *      @len: amount of data to add
1267  *
1268  *      This function extends the used data area of the buffer. If this would
1269  *      exceed the total buffer size the kernel will panic. A pointer to the
1270  *      first byte of the extra data is returned.
1271  */
1272 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1273 {
1274         unsigned char *tmp = skb_tail_pointer(skb);
1275         SKB_LINEAR_ASSERT(skb);
1276         skb->tail += len;
1277         skb->len  += len;
1278         if (unlikely(skb->tail > skb->end))
1279                 skb_over_panic(skb, len, __builtin_return_address(0));
1280         return tmp;
1281 }
1282 EXPORT_SYMBOL(skb_put);
1283 
1284 /**
1285  *      skb_push - add data to the start of a buffer
1286  *      @skb: buffer to use
1287  *      @len: amount of data to add
1288  *
1289  *      This function extends the used data area of the buffer at the buffer
1290  *      start. If this would exceed the total buffer headroom the kernel will
1291  *      panic. A pointer to the first byte of the extra data is returned.
1292  */
1293 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1294 {
1295         skb->data -= len;
1296         skb->len  += len;
1297         if (unlikely(skb->data<skb->head))
1298                 skb_under_panic(skb, len, __builtin_return_address(0));
1299         return skb->data;
1300 }
1301 EXPORT_SYMBOL(skb_push);
1302 
1303 /**
1304  *      skb_pull - remove data from the start of a buffer
1305  *      @skb: buffer to use
1306  *      @len: amount of data to remove
1307  *
1308  *      This function removes data from the start of a buffer, returning
1309  *      the memory to the headroom. A pointer to the next data in the buffer
1310  *      is returned. Once the data has been pulled future pushes will overwrite
1311  *      the old data.
1312  */
1313 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1314 {
1315         return skb_pull_inline(skb, len);
1316 }
1317 EXPORT_SYMBOL(skb_pull);
1318 
1319 /**
1320  *      skb_trim - remove end from a buffer
1321  *      @skb: buffer to alter
1322  *      @len: new length
1323  *
1324  *      Cut the length of a buffer down by removing data from the tail. If
1325  *      the buffer is already under the length specified it is not modified.
1326  *      The skb must be linear.
1327  */
1328 void skb_trim(struct sk_buff *skb, unsigned int len)
1329 {
1330         if (skb->len > len)
1331                 __skb_trim(skb, len);
1332 }
1333 EXPORT_SYMBOL(skb_trim);
1334 
1335 /* Trims skb to length len. It can change skb pointers.
1336  */
1337 
1338 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1339 {
1340         struct sk_buff **fragp;
1341         struct sk_buff *frag;
1342         int offset = skb_headlen(skb);
1343         int nfrags = skb_shinfo(skb)->nr_frags;
1344         int i;
1345         int err;
1346 
1347         if (skb_cloned(skb) &&
1348             unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1349                 return err;
1350 
1351         i = 0;
1352         if (offset >= len)
1353                 goto drop_pages;
1354 
1355         for (; i < nfrags; i++) {
1356                 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1357 
1358                 if (end < len) {
1359                         offset = end;
1360                         continue;
1361                 }
1362 
1363                 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1364 
1365 drop_pages:
1366                 skb_shinfo(skb)->nr_frags = i;
1367 
1368                 for (; i < nfrags; i++)
1369                         skb_frag_unref(skb, i);
1370 
1371                 if (skb_has_frag_list(skb))
1372                         skb_drop_fraglist(skb);
1373                 goto done;
1374         }
1375 
1376         for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1377              fragp = &frag->next) {
1378                 int end = offset + frag->len;
1379 
1380                 if (skb_shared(frag)) {
1381                         struct sk_buff *nfrag;
1382 
1383                         nfrag = skb_clone(frag, GFP_ATOMIC);
1384                         if (unlikely(!nfrag))
1385                                 return -ENOMEM;
1386 
1387                         nfrag->next = frag->next;
1388                         consume_skb(frag);
1389                         frag = nfrag;
1390                         *fragp = frag;
1391                 }
1392 
1393                 if (end < len) {
1394                         offset = end;
1395                         continue;
1396                 }
1397 
1398                 if (end > len &&
1399                     unlikely((err = pskb_trim(frag, len - offset))))
1400                         return err;
1401 
1402                 if (frag->next)
1403                         skb_drop_list(&frag->next);
1404                 break;
1405         }
1406 
1407 done:
1408         if (len > skb_headlen(skb)) {
1409                 skb->data_len -= skb->len - len;
1410                 skb->len       = len;
1411         } else {
1412                 skb->len       = len;
1413                 skb->data_len  = 0;
1414                 skb_set_tail_pointer(skb, len);
1415         }
1416 
1417         return 0;
1418 }
1419 EXPORT_SYMBOL(___pskb_trim);
1420 
1421 /**
1422  *      __pskb_pull_tail - advance tail of skb header
1423  *      @skb: buffer to reallocate
1424  *      @delta: number of bytes to advance tail
1425  *
1426  *      The function makes a sense only on a fragmented &sk_buff,
1427  *      it expands header moving its tail forward and copying necessary
1428  *      data from fragmented part.
1429  *
1430  *      &sk_buff MUST have reference count of 1.
1431  *
1432  *      Returns %NULL (and &sk_buff does not change) if pull failed
1433  *      or value of new tail of skb in the case of success.
1434  *
1435  *      All the pointers pointing into skb header may change and must be
1436  *      reloaded after call to this function.
1437  */
1438 
1439 /* Moves tail of skb head forward, copying data from fragmented part,
1440  * when it is necessary.
1441  * 1. It may fail due to malloc failure.
1442  * 2. It may change skb pointers.
1443  *
1444  * It is pretty complicated. Luckily, it is called only in exceptional cases.
1445  */
1446 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1447 {
1448         /* If skb has not enough free space at tail, get new one
1449          * plus 128 bytes for future expansions. If we have enough
1450          * room at tail, reallocate without expansion only if skb is cloned.
1451          */
1452         int i, k, eat = (skb->tail + delta) - skb->end;
1453 
1454         if (eat > 0 || skb_cloned(skb)) {
1455                 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1456                                      GFP_ATOMIC))
1457                         return NULL;
1458         }
1459 
1460         if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1461                 BUG();
1462 
1463         /* Optimization: no fragments, no reasons to preestimate
1464          * size of pulled pages. Superb.
1465          */
1466         if (!skb_has_frag_list(skb))
1467                 goto pull_pages;
1468 
1469         /* Estimate size of pulled pages. */
1470         eat = delta;
1471         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1472                 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1473 
1474                 if (size >= eat)
1475                         goto pull_pages;
1476                 eat -= size;
1477         }
1478 
1479         /* If we need update frag list, we are in troubles.
1480          * Certainly, it possible to add an offset to skb data,
1481          * but taking into account that pulling is expected to
1482          * be very rare operation, it is worth to fight against
1483          * further bloating skb head and crucify ourselves here instead.
1484          * Pure masohism, indeed. 8)8)
1485          */
1486         if (eat) {
1487                 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1488                 struct sk_buff *clone = NULL;
1489                 struct sk_buff *insp = NULL;
1490 
1491                 do {
1492                         BUG_ON(!list);
1493 
1494                         if (list->len <= eat) {
1495                                 /* Eaten as whole. */
1496                                 eat -= list->len;
1497                                 list = list->next;
1498                                 insp = list;
1499                         } else {
1500                                 /* Eaten partially. */
1501 
1502                                 if (skb_shared(list)) {
1503                                         /* Sucks! We need to fork list. :-( */
1504                                         clone = skb_clone(list, GFP_ATOMIC);
1505                                         if (!clone)
1506                                                 return NULL;
1507                                         insp = list->next;
1508                                         list = clone;
1509                                 } else {
1510                                         /* This may be pulled without
1511                                          * problems. */
1512                                         insp = list;
1513                                 }
1514                                 if (!pskb_pull(list, eat)) {
1515                                         kfree_skb(clone);
1516                                         return NULL;
1517                                 }
1518                                 break;
1519                         }
1520                 } while (eat);
1521 
1522                 /* Free pulled out fragments. */
1523                 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1524                         skb_shinfo(skb)->frag_list = list->next;
1525                         kfree_skb(list);
1526                 }
1527                 /* And insert new clone at head. */
1528                 if (clone) {
1529                         clone->next = list;
1530                         skb_shinfo(skb)->frag_list = clone;
1531                 }
1532         }
1533         /* Success! Now we may commit changes to skb data. */
1534 
1535 pull_pages:
1536         eat = delta;
1537         k = 0;
1538         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1539                 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1540 
1541                 if (size <= eat) {
1542                         skb_frag_unref(skb, i);
1543                         eat -= size;
1544                 } else {
1545                         skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1546                         if (eat) {
1547                                 skb_shinfo(skb)->frags[k].page_offset += eat;
1548                                 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1549                                 eat = 0;
1550                         }
1551                         k++;
1552                 }
1553         }
1554         skb_shinfo(skb)->nr_frags = k;
1555 
1556         skb->tail     += delta;
1557         skb->data_len -= delta;
1558 
1559         return skb_tail_pointer(skb);
1560 }
1561 EXPORT_SYMBOL(__pskb_pull_tail);
1562 
1563 /**
1564  *      skb_copy_bits - copy bits from skb to kernel buffer
1565  *      @skb: source skb
1566  *      @offset: offset in source
1567  *      @to: destination buffer
1568  *      @len: number of bytes to copy
1569  *
1570  *      Copy the specified number of bytes from the source skb to the
1571  *      destination buffer.
1572  *
1573  *      CAUTION ! :
1574  *              If its prototype is ever changed,
1575  *              check arch/{*}/net/{*}.S files,
1576  *              since it is called from BPF assembly code.
1577  */
1578 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1579 {
1580         int start = skb_headlen(skb);
1581         struct sk_buff *frag_iter;
1582         int i, copy;
1583 
1584         if (offset > (int)skb->len - len)
1585                 goto fault;
1586 
1587         /* Copy header. */
1588         if ((copy = start - offset) > 0) {
1589                 if (copy > len)
1590                         copy = len;
1591                 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1592                 if ((len -= copy) == 0)
1593                         return 0;
1594                 offset += copy;
1595                 to     += copy;
1596         }
1597 
1598         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1599                 int end;
1600                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1601 
1602                 WARN_ON(start > offset + len);
1603 
1604                 end = start + skb_frag_size(f);
1605                 if ((copy = end - offset) > 0) {
1606                         u8 *vaddr;
1607 
1608                         if (copy > len)
1609                                 copy = len;
1610 
1611                         vaddr = kmap_atomic(skb_frag_page(f));
1612                         memcpy(to,
1613                                vaddr + f->page_offset + offset - start,
1614                                copy);
1615                         kunmap_atomic(vaddr);
1616 
1617                         if ((len -= copy) == 0)
1618                                 return 0;
1619                         offset += copy;
1620                         to     += copy;
1621                 }
1622                 start = end;
1623         }
1624 
1625         skb_walk_frags(skb, frag_iter) {
1626                 int end;
1627 
1628                 WARN_ON(start > offset + len);
1629 
1630                 end = start + frag_iter->len;
1631                 if ((copy = end - offset) > 0) {
1632                         if (copy > len)
1633                                 copy = len;
1634                         if (skb_copy_bits(frag_iter, offset - start, to, copy))
1635                                 goto fault;
1636                         if ((len -= copy) == 0)
1637                                 return 0;
1638                         offset += copy;
1639                         to     += copy;
1640                 }
1641                 start = end;
1642         }
1643 
1644         if (!len)
1645                 return 0;
1646 
1647 fault:
1648         return -EFAULT;
1649 }
1650 EXPORT_SYMBOL(skb_copy_bits);
1651 
1652 /*
1653  * Callback from splice_to_pipe(), if we need to release some pages
1654  * at the end of the spd in case we error'ed out in filling the pipe.
1655  */
1656 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1657 {
1658         put_page(spd->pages[i]);
1659 }
1660 
1661 static struct page *linear_to_page(struct page *page, unsigned int *len,
1662                                    unsigned int *offset,
1663                                    struct sock *sk)
1664 {
1665         struct page_frag *pfrag = sk_page_frag(sk);
1666 
1667         if (!sk_page_frag_refill(sk, pfrag))
1668                 return NULL;
1669 
1670         *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
1671 
1672         memcpy(page_address(pfrag->page) + pfrag->offset,
1673                page_address(page) + *offset, *len);
1674         *offset = pfrag->offset;
1675         pfrag->offset += *len;
1676 
1677         return pfrag->page;
1678 }
1679 
1680 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1681                              struct page *page,
1682                              unsigned int offset)
1683 {
1684         return  spd->nr_pages &&
1685                 spd->pages[spd->nr_pages - 1] == page &&
1686                 (spd->partial[spd->nr_pages - 1].offset +
1687                  spd->partial[spd->nr_pages - 1].len == offset);
1688 }
1689 
1690 /*
1691  * Fill page/offset/length into spd, if it can hold more pages.
1692  */
1693 static bool spd_fill_page(struct splice_pipe_desc *spd,
1694                           struct pipe_inode_info *pipe, struct page *page,
1695                           unsigned int *len, unsigned int offset,
1696                           bool linear,
1697                           struct sock *sk)
1698 {
1699         if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1700                 return true;
1701 
1702         if (linear) {
1703                 page = linear_to_page(page, len, &offset, sk);
1704                 if (!page)
1705                         return true;
1706         }
1707         if (spd_can_coalesce(spd, page, offset)) {
1708                 spd->partial[spd->nr_pages - 1].len += *len;
1709                 return false;
1710         }
1711         get_page(page);
1712         spd->pages[spd->nr_pages] = page;
1713         spd->partial[spd->nr_pages].len = *len;
1714         spd->partial[spd->nr_pages].offset = offset;
1715         spd->nr_pages++;
1716 
1717         return false;
1718 }
1719 
1720 static bool __splice_segment(struct page *page, unsigned int poff,
1721                              unsigned int plen, unsigned int *off,
1722                              unsigned int *len,
1723                              struct splice_pipe_desc *spd, bool linear,
1724                              struct sock *sk,
1725                              struct pipe_inode_info *pipe)
1726 {
1727         if (!*len)
1728                 return true;
1729 
1730         /* skip this segment if already processed */
1731         if (*off >= plen) {
1732                 *off -= plen;
1733                 return false;
1734         }
1735 
1736         /* ignore any bits we already processed */
1737         poff += *off;
1738         plen -= *off;
1739         *off = 0;
1740 
1741         do {
1742                 unsigned int flen = min(*len, plen);
1743 
1744                 if (spd_fill_page(spd, pipe, page, &flen, poff,
1745                                   linear, sk))
1746                         return true;
1747                 poff += flen;
1748                 plen -= flen;
1749                 *len -= flen;
1750         } while (*len && plen);
1751 
1752         return false;
1753 }
1754 
1755 /*
1756  * Map linear and fragment data from the skb to spd. It reports true if the
1757  * pipe is full or if we already spliced the requested length.
1758  */
1759 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1760                               unsigned int *offset, unsigned int *len,
1761                               struct splice_pipe_desc *spd, struct sock *sk)
1762 {
1763         int seg;
1764 
1765         /* map the linear part :
1766          * If skb->head_frag is set, this 'linear' part is backed by a
1767          * fragment, and if the head is not shared with any clones then
1768          * we can avoid a copy since we own the head portion of this page.
1769          */
1770         if (__splice_segment(virt_to_page(skb->data),
1771                              (unsigned long) skb->data & (PAGE_SIZE - 1),
1772                              skb_headlen(skb),
1773                              offset, len, spd,
1774                              skb_head_is_locked(skb),
1775                              sk, pipe))
1776                 return true;
1777 
1778         /*
1779          * then map the fragments
1780          */
1781         for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1782                 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1783 
1784                 if (__splice_segment(skb_frag_page(f),
1785                                      f->page_offset, skb_frag_size(f),
1786                                      offset, len, spd, false, sk, pipe))
1787                         return true;
1788         }
1789 
1790         return false;
1791 }
1792 
1793 /*
1794  * Map data from the skb to a pipe. Should handle both the linear part,
1795  * the fragments, and the frag list. It does NOT handle frag lists within
1796  * the frag list, if such a thing exists. We'd probably need to recurse to
1797  * handle that cleanly.
1798  */
1799 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1800                     struct pipe_inode_info *pipe, unsigned int tlen,
1801                     unsigned int flags)
1802 {
1803         struct partial_page partial[MAX_SKB_FRAGS];
1804         struct page *pages[MAX_SKB_FRAGS];
1805         struct splice_pipe_desc spd = {
1806                 .pages = pages,
1807                 .partial = partial,
1808                 .nr_pages_max = MAX_SKB_FRAGS,
1809                 .flags = flags,
1810                 .ops = &nosteal_pipe_buf_ops,
1811                 .spd_release = sock_spd_release,
1812         };
1813         struct sk_buff *frag_iter;
1814         struct sock *sk = skb->sk;
1815         int ret = 0;
1816 
1817         /*
1818          * __skb_splice_bits() only fails if the output has no room left,
1819          * so no point in going over the frag_list for the error case.
1820          */
1821         if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1822                 goto done;
1823         else if (!tlen)
1824                 goto done;
1825 
1826         /*
1827          * now see if we have a frag_list to map
1828          */
1829         skb_walk_frags(skb, frag_iter) {
1830                 if (!tlen)
1831                         break;
1832                 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1833                         break;
1834         }
1835 
1836 done:
1837         if (spd.nr_pages) {
1838                 /*
1839                  * Drop the socket lock, otherwise we have reverse
1840                  * locking dependencies between sk_lock and i_mutex
1841                  * here as compared to sendfile(). We enter here
1842                  * with the socket lock held, and splice_to_pipe() will
1843                  * grab the pipe inode lock. For sendfile() emulation,
1844                  * we call into ->sendpage() with the i_mutex lock held
1845                  * and networking will grab the socket lock.
1846                  */
1847                 release_sock(sk);
1848                 ret = splice_to_pipe(pipe, &spd);
1849                 lock_sock(sk);
1850         }
1851 
1852         return ret;
1853 }
1854 
1855 /**
1856  *      skb_store_bits - store bits from kernel buffer to skb
1857  *      @skb: destination buffer
1858  *      @offset: offset in destination
1859  *      @from: source buffer
1860  *      @len: number of bytes to copy
1861  *
1862  *      Copy the specified number of bytes from the source buffer to the
1863  *      destination skb.  This function handles all the messy bits of
1864  *      traversing fragment lists and such.
1865  */
1866 
1867 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1868 {
1869         int start = skb_headlen(skb);
1870         struct sk_buff *frag_iter;
1871         int i, copy;
1872 
1873         if (offset > (int)skb->len - len)
1874                 goto fault;
1875 
1876         if ((copy = start - offset) > 0) {
1877                 if (copy > len)
1878                         copy = len;
1879                 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1880                 if ((len -= copy) == 0)
1881                         return 0;
1882                 offset += copy;
1883                 from += copy;
1884         }
1885 
1886         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1887                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1888                 int end;
1889 
1890                 WARN_ON(start > offset + len);
1891 
1892                 end = start + skb_frag_size(frag);
1893                 if ((copy = end - offset) > 0) {
1894                         u8 *vaddr;
1895 
1896                         if (copy > len)
1897                                 copy = len;
1898 
1899                         vaddr = kmap_atomic(skb_frag_page(frag));
1900                         memcpy(vaddr + frag->page_offset + offset - start,
1901                                from, copy);
1902                         kunmap_atomic(vaddr);
1903 
1904                         if ((len -= copy) == 0)
1905                                 return 0;
1906                         offset += copy;
1907                         from += copy;
1908                 }
1909                 start = end;
1910         }
1911 
1912         skb_walk_frags(skb, frag_iter) {
1913                 int end;
1914 
1915                 WARN_ON(start > offset + len);
1916 
1917                 end = start + frag_iter->len;
1918                 if ((copy = end - offset) > 0) {
1919                         if (copy > len)
1920                                 copy = len;
1921                         if (skb_store_bits(frag_iter, offset - start,
1922                                            from, copy))
1923                                 goto fault;
1924                         if ((len -= copy) == 0)
1925                                 return 0;
1926                         offset += copy;
1927                         from += copy;
1928                 }
1929                 start = end;
1930         }
1931         if (!len)
1932                 return 0;
1933 
1934 fault:
1935         return -EFAULT;
1936 }
1937 EXPORT_SYMBOL(skb_store_bits);
1938 
1939 /* Checksum skb data. */
1940 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
1941                       __wsum csum, const struct skb_checksum_ops *ops)
1942 {
1943         int start = skb_headlen(skb);
1944         int i, copy = start - offset;
1945         struct sk_buff *frag_iter;
1946         int pos = 0;
1947 
1948         /* Checksum header. */
1949         if (copy > 0) {
1950                 if (copy > len)
1951                         copy = len;
1952                 csum = ops->update(skb->data + offset, copy, csum);
1953                 if ((len -= copy) == 0)
1954                         return csum;
1955                 offset += copy;
1956                 pos     = copy;
1957         }
1958 
1959         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1960                 int end;
1961                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1962 
1963                 WARN_ON(start > offset + len);
1964 
1965                 end = start + skb_frag_size(frag);
1966                 if ((copy = end - offset) > 0) {
1967                         __wsum csum2;
1968                         u8 *vaddr;
1969 
1970                         if (copy > len)
1971                                 copy = len;
1972                         vaddr = kmap_atomic(skb_frag_page(frag));
1973                         csum2 = ops->update(vaddr + frag->page_offset +
1974                                             offset - start, copy, 0);
1975                         kunmap_atomic(vaddr);
1976                         csum = ops->combine(csum, csum2, pos, copy);
1977                         if (!(len -= copy))
1978                                 return csum;
1979                         offset += copy;
1980                         pos    += copy;
1981                 }
1982                 start = end;
1983         }
1984 
1985         skb_walk_frags(skb, frag_iter) {
1986                 int end;
1987 
1988                 WARN_ON(start > offset + len);
1989 
1990                 end = start + frag_iter->len;
1991                 if ((copy = end - offset) > 0) {
1992                         __wsum csum2;
1993                         if (copy > len)
1994                                 copy = len;
1995                         csum2 = __skb_checksum(frag_iter, offset - start,
1996                                                copy, 0, ops);
1997                         csum = ops->combine(csum, csum2, pos, copy);
1998                         if ((len -= copy) == 0)
1999                                 return csum;
2000                         offset += copy;
2001                         pos    += copy;
2002                 }
2003                 start = end;
2004         }
2005         BUG_ON(len);
2006 
2007         return csum;
2008 }
2009 EXPORT_SYMBOL(__skb_checksum);
2010 
2011 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2012                     int len, __wsum csum)
2013 {
2014         const struct skb_checksum_ops ops = {
2015                 .update  = csum_partial_ext,
2016                 .combine = csum_block_add_ext,
2017         };
2018 
2019         return __skb_checksum(skb, offset, len, csum, &ops);
2020 }
2021 EXPORT_SYMBOL(skb_checksum);
2022 
2023 /* Both of above in one bottle. */
2024 
2025 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2026                                     u8 *to, int len, __wsum csum)
2027 {
2028         int start = skb_headlen(skb);
2029         int i, copy = start - offset;
2030         struct sk_buff *frag_iter;
2031         int pos = 0;
2032 
2033         /* Copy header. */
2034         if (copy > 0) {
2035                 if (copy > len)
2036                         copy = len;
2037                 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2038                                                  copy, csum);
2039                 if ((len -= copy) == 0)
2040                         return csum;
2041                 offset += copy;
2042                 to     += copy;
2043                 pos     = copy;
2044         }
2045 
2046         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2047                 int end;
2048 
2049                 WARN_ON(start > offset + len);
2050 
2051                 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2052                 if ((copy = end - offset) > 0) {
2053                         __wsum csum2;
2054                         u8 *vaddr;
2055                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2056 
2057                         if (copy > len)
2058                                 copy = len;
2059                         vaddr = kmap_atomic(skb_frag_page(frag));
2060                         csum2 = csum_partial_copy_nocheck(vaddr +
2061                                                           frag->page_offset +
2062                                                           offset - start, to,
2063                                                           copy, 0);
2064                         kunmap_atomic(vaddr);
2065                         csum = csum_block_add(csum, csum2, pos);
2066                         if (!(len -= copy))
2067                                 return csum;
2068                         offset += copy;
2069                         to     += copy;
2070                         pos    += copy;
2071                 }
2072                 start = end;
2073         }
2074 
2075         skb_walk_frags(skb, frag_iter) {
2076                 __wsum csum2;
2077                 int end;
2078 
2079                 WARN_ON(start > offset + len);
2080 
2081                 end = start + frag_iter->len;
2082                 if ((copy = end - offset) > 0) {
2083                         if (copy > len)
2084                                 copy = len;
2085                         csum2 = skb_copy_and_csum_bits(frag_iter,
2086                                                        offset - start,
2087                                                        to, copy, 0);
2088                         csum = csum_block_add(csum, csum2, pos);
2089                         if ((len -= copy) == 0)
2090                                 return csum;
2091                         offset += copy;
2092                         to     += copy;
2093                         pos    += copy;
2094                 }
2095                 start = end;
2096         }
2097         BUG_ON(len);
2098         return csum;
2099 }
2100 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2101 
2102  /**
2103  *      skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2104  *      @from: source buffer
2105  *
2106  *      Calculates the amount of linear headroom needed in the 'to' skb passed
2107  *      into skb_zerocopy().
2108  */
2109 unsigned int
2110 skb_zerocopy_headlen(const struct sk_buff *from)
2111 {
2112         unsigned int hlen = 0;
2113 
2114         if (!from->head_frag ||
2115             skb_headlen(from) < L1_CACHE_BYTES ||
2116             skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2117                 hlen = skb_headlen(from);
2118 
2119         if (skb_has_frag_list(from))
2120                 hlen = from->len;
2121 
2122         return hlen;
2123 }
2124 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2125 
2126 /**
2127  *      skb_zerocopy - Zero copy skb to skb
2128  *      @to: destination buffer
2129  *      @from: source buffer
2130  *      @len: number of bytes to copy from source buffer
2131  *      @hlen: size of linear headroom in destination buffer
2132  *
2133  *      Copies up to `len` bytes from `from` to `to` by creating references
2134  *      to the frags in the source buffer.
2135  *
2136  *      The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2137  *      headroom in the `to` buffer.
2138  *
2139  *      Return value:
2140  *      0: everything is OK
2141  *      -ENOMEM: couldn't orphan frags of @from due to lack of memory
2142  *      -EFAULT: skb_copy_bits() found some problem with skb geometry
2143  */
2144 int
2145 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2146 {
2147         int i, j = 0;
2148         int plen = 0; /* length of skb->head fragment */
2149         int ret;
2150         struct page *page;
2151         unsigned int offset;
2152 
2153         BUG_ON(!from->head_frag && !hlen);
2154 
2155         /* dont bother with small payloads */
2156         if (len <= skb_tailroom(to))
2157                 return skb_copy_bits(from, 0, skb_put(to, len), len);
2158 
2159         if (hlen) {
2160                 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2161                 if (unlikely(ret))
2162                         return ret;
2163                 len -= hlen;
2164         } else {
2165                 plen = min_t(int, skb_headlen(from), len);
2166                 if (plen) {
2167                         page = virt_to_head_page(from->head);
2168                         offset = from->data - (unsigned char *)page_address(page);
2169                         __skb_fill_page_desc(to, 0, page, offset, plen);
2170                         get_page(page);
2171                         j = 1;
2172                         len -= plen;
2173                 }
2174         }
2175 
2176         to->truesize += len + plen;
2177         to->len += len + plen;
2178         to->data_len += len + plen;
2179 
2180         if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2181                 skb_tx_error(from);
2182                 return -ENOMEM;
2183         }
2184 
2185         for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2186                 if (!len)
2187                         break;
2188                 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2189                 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
2190                 len -= skb_shinfo(to)->frags[j].size;
2191                 skb_frag_ref(to, j);
2192                 j++;
2193         }
2194         skb_shinfo(to)->nr_frags = j;
2195 
2196         return 0;
2197 }
2198 EXPORT_SYMBOL_GPL(skb_zerocopy);
2199 
2200 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2201 {
2202         __wsum csum;
2203         long csstart;
2204 
2205         if (skb->ip_summed == CHECKSUM_PARTIAL)
2206                 csstart = skb_checksum_start_offset(skb);
2207         else
2208                 csstart = skb_headlen(skb);
2209 
2210         BUG_ON(csstart > skb_headlen(skb));
2211 
2212         skb_copy_from_linear_data(skb, to, csstart);
2213 
2214         csum = 0;
2215         if (csstart != skb->len)
2216                 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2217                                               skb->len - csstart, 0);
2218 
2219         if (skb->ip_summed == CHECKSUM_PARTIAL) {
2220                 long csstuff = csstart + skb->csum_offset;
2221 
2222                 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2223         }
2224 }
2225 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2226 
2227 /**
2228  *      skb_dequeue - remove from the head of the queue
2229  *      @list: list to dequeue from
2230  *
2231  *      Remove the head of the list. The list lock is taken so the function
2232  *      may be used safely with other locking list functions. The head item is
2233  *      returned or %NULL if the list is empty.
2234  */
2235 
2236 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2237 {
2238         unsigned long flags;
2239         struct sk_buff *result;
2240 
2241         spin_lock_irqsave(&list->lock, flags);
2242         result = __skb_dequeue(list);
2243         spin_unlock_irqrestore(&list->lock, flags);
2244         return result;
2245 }
2246 EXPORT_SYMBOL(skb_dequeue);
2247 
2248 /**
2249  *      skb_dequeue_tail - remove from the tail of the queue
2250  *      @list: list to dequeue from
2251  *
2252  *      Remove the tail of the list. The list lock is taken so the function
2253  *      may be used safely with other locking list functions. The tail item is
2254  *      returned or %NULL if the list is empty.
2255  */
2256 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2257 {
2258         unsigned long flags;
2259         struct sk_buff *result;
2260 
2261         spin_lock_irqsave(&list->lock, flags);
2262         result = __skb_dequeue_tail(list);
2263         spin_unlock_irqrestore(&list->lock, flags);
2264         return result;
2265 }
2266 EXPORT_SYMBOL(skb_dequeue_tail);
2267 
2268 /**
2269  *      skb_queue_purge - empty a list
2270  *      @list: list to empty
2271  *
2272  *      Delete all buffers on an &sk_buff list. Each buffer is removed from
2273  *      the list and one reference dropped. This function takes the list
2274  *      lock and is atomic with respect to other list locking functions.
2275  */
2276 void skb_queue_purge(struct sk_buff_head *list)
2277 {
2278         struct sk_buff *skb;
2279         while ((skb = skb_dequeue(list)) != NULL)
2280                 kfree_skb(skb);
2281 }
2282 EXPORT_SYMBOL(skb_queue_purge);
2283 
2284 /**
2285  *      skb_queue_head - queue a buffer at the list head
2286  *      @list: list to use
2287  *      @newsk: buffer to queue
2288  *
2289  *      Queue a buffer at the start of the list. This function takes the
2290  *      list lock and can be used safely with other locking &sk_buff functions
2291  *      safely.
2292  *
2293  *      A buffer cannot be placed on two lists at the same time.
2294  */
2295 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2296 {
2297         unsigned long flags;
2298 
2299         spin_lock_irqsave(&list->lock, flags);
2300         __skb_queue_head(list, newsk);
2301         spin_unlock_irqrestore(&list->lock, flags);
2302 }
2303 EXPORT_SYMBOL(skb_queue_head);
2304 
2305 /**
2306  *      skb_queue_tail - queue a buffer at the list tail
2307  *      @list: list to use
2308  *      @newsk: buffer to queue
2309  *
2310  *      Queue a buffer at the tail of the list. This function takes the
2311  *      list lock and can be used safely with other locking &sk_buff functions
2312  *      safely.
2313  *
2314  *      A buffer cannot be placed on two lists at the same time.
2315  */
2316 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2317 {
2318         unsigned long flags;
2319 
2320         spin_lock_irqsave(&list->lock, flags);
2321         __skb_queue_tail(list, newsk);
2322         spin_unlock_irqrestore(&list->lock, flags);
2323 }
2324 EXPORT_SYMBOL(skb_queue_tail);
2325 
2326 /**
2327  *      skb_unlink      -       remove a buffer from a list
2328  *      @skb: buffer to remove
2329  *      @list: list to use
2330  *
2331  *      Remove a packet from a list. The list locks are taken and this
2332  *      function is atomic with respect to other list locked calls
2333  *
2334  *      You must know what list the SKB is on.
2335  */
2336 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2337 {
2338         unsigned long flags;
2339 
2340         spin_lock_irqsave(&list->lock, flags);
2341         __skb_unlink(skb, list);
2342         spin_unlock_irqrestore(&list->lock, flags);
2343 }
2344 EXPORT_SYMBOL(skb_unlink);
2345 
2346 /**
2347  *      skb_append      -       append a buffer
2348  *      @old: buffer to insert after
2349  *      @newsk: buffer to insert
2350  *      @list: list to use
2351  *
2352  *      Place a packet after a given packet in a list. The list locks are taken
2353  *      and this function is atomic with respect to other list locked calls.
2354  *      A buffer cannot be placed on two lists at the same time.
2355  */
2356 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2357 {
2358         unsigned long flags;
2359 
2360         spin_lock_irqsave(&list->lock, flags);
2361         __skb_queue_after(list, old, newsk);
2362         spin_unlock_irqrestore(&list->lock, flags);
2363 }
2364 EXPORT_SYMBOL(skb_append);
2365 
2366 /**
2367  *      skb_insert      -       insert a buffer
2368  *      @old: buffer to insert before
2369  *      @newsk: buffer to insert
2370  *      @list: list to use
2371  *
2372  *      Place a packet before a given packet in a list. The list locks are
2373  *      taken and this function is atomic with respect to other list locked
2374  *      calls.
2375  *
2376  *      A buffer cannot be placed on two lists at the same time.
2377  */
2378 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2379 {
2380         unsigned long flags;
2381 
2382         spin_lock_irqsave(&list->lock, flags);
2383         __skb_insert(newsk, old->prev, old, list);
2384         spin_unlock_irqrestore(&list->lock, flags);
2385 }
2386 EXPORT_SYMBOL(skb_insert);
2387 
2388 static inline void skb_split_inside_header(struct sk_buff *skb,
2389                                            struct sk_buff* skb1,
2390                                            const u32 len, const int pos)
2391 {
2392         int i;
2393 
2394         skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2395                                          pos - len);
2396         /* And move data appendix as is. */
2397         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2398                 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2399 
2400         skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2401         skb_shinfo(skb)->nr_frags  = 0;
2402         skb1->data_len             = skb->data_len;
2403         skb1->len                  += skb1->data_len;
2404         skb->data_len              = 0;
2405         skb->len                   = len;
2406         skb_set_tail_pointer(skb, len);
2407 }
2408 
2409 static inline void skb_split_no_header(struct sk_buff *skb,
2410                                        struct sk_buff* skb1,
2411                                        const u32 len, int pos)
2412 {
2413         int i, k = 0;
2414         const int nfrags = skb_shinfo(skb)->nr_frags;
2415 
2416         skb_shinfo(skb)->nr_frags = 0;
2417         skb1->len                 = skb1->data_len = skb->len - len;
2418         skb->len                  = len;
2419         skb->data_len             = len - pos;
2420 
2421         for (i = 0; i < nfrags; i++) {
2422                 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2423 
2424                 if (pos + size > len) {
2425                         skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2426 
2427                         if (pos < len) {
2428                                 /* Split frag.
2429                                  * We have two variants in this case:
2430                                  * 1. Move all the frag to the second
2431                                  *    part, if it is possible. F.e.
2432                                  *    this approach is mandatory for TUX,
2433                                  *    where splitting is expensive.
2434                                  * 2. Split is accurately. We make this.
2435                                  */
2436                                 skb_frag_ref(skb, i);
2437                                 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2438                                 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2439                                 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2440                                 skb_shinfo(skb)->nr_frags++;
2441                         }
2442                         k++;
2443                 } else
2444                         skb_shinfo(skb)->nr_frags++;
2445                 pos += size;
2446         }
2447         skb_shinfo(skb1)->nr_frags = k;
2448 }
2449 
2450 /**
2451  * skb_split - Split fragmented skb to two parts at length len.
2452  * @skb: the buffer to split
2453  * @skb1: the buffer to receive the second part
2454  * @len: new length for skb
2455  */
2456 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2457 {
2458         int pos = skb_headlen(skb);
2459 
2460         skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2461         if (len < pos)  /* Split line is inside header. */
2462                 skb_split_inside_header(skb, skb1, len, pos);
2463         else            /* Second chunk has no header, nothing to copy. */
2464                 skb_split_no_header(skb, skb1, len, pos);
2465 }
2466 EXPORT_SYMBOL(skb_split);
2467 
2468 /* Shifting from/to a cloned skb is a no-go.
2469  *
2470  * Caller cannot keep skb_shinfo related pointers past calling here!
2471  */
2472 static int skb_prepare_for_shift(struct sk_buff *skb)
2473 {
2474         return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2475 }
2476 
2477 /**
2478  * skb_shift - Shifts paged data partially from skb to another
2479  * @tgt: buffer into which tail data gets added
2480  * @skb: buffer from which the paged data comes from
2481  * @shiftlen: shift up to this many bytes
2482  *
2483  * Attempts to shift up to shiftlen worth of bytes, which may be less than
2484  * the length of the skb, from skb to tgt. Returns number bytes shifted.
2485  * It's up to caller to free skb if everything was shifted.
2486  *
2487  * If @tgt runs out of frags, the whole operation is aborted.
2488  *
2489  * Skb cannot include anything else but paged data while tgt is allowed
2490  * to have non-paged data as well.
2491  *
2492  * TODO: full sized shift could be optimized but that would need
2493  * specialized skb free'er to handle frags without up-to-date nr_frags.
2494  */
2495 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2496 {
2497         int from, to, merge, todo;
2498         struct skb_frag_struct *fragfrom, *fragto;
2499 
2500         BUG_ON(shiftlen > skb->len);
2501         BUG_ON(skb_headlen(skb));       /* Would corrupt stream */
2502 
2503         todo = shiftlen;
2504         from = 0;
2505         to = skb_shinfo(tgt)->nr_frags;
2506         fragfrom = &skb_shinfo(skb)->frags[from];
2507 
2508         /* Actual merge is delayed until the point when we know we can
2509          * commit all, so that we don't have to undo partial changes
2510          */
2511         if (!to ||
2512             !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2513                               fragfrom->page_offset)) {
2514                 merge = -1;
2515         } else {
2516                 merge = to - 1;
2517 
2518                 todo -= skb_frag_size(fragfrom);
2519                 if (todo < 0) {
2520                         if (skb_prepare_for_shift(skb) ||
2521                             skb_prepare_for_shift(tgt))
2522                                 return 0;
2523 
2524                         /* All previous frag pointers might be stale! */
2525                         fragfrom = &skb_shinfo(skb)->frags[from];
2526                         fragto = &skb_shinfo(tgt)->frags[merge];
2527 
2528                         skb_frag_size_add(fragto, shiftlen);
2529                         skb_frag_size_sub(fragfrom, shiftlen);
2530                         fragfrom->page_offset += shiftlen;
2531 
2532                         goto onlymerged;
2533                 }
2534 
2535                 from++;
2536         }
2537 
2538         /* Skip full, not-fitting skb to avoid expensive operations */
2539         if ((shiftlen == skb->len) &&
2540             (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2541                 return 0;
2542 
2543         if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2544                 return 0;
2545 
2546         while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2547                 if (to == MAX_SKB_FRAGS)
2548                         return 0;
2549 
2550                 fragfrom = &skb_shinfo(skb)->frags[from];
2551                 fragto = &skb_shinfo(tgt)->frags[to];
2552 
2553                 if (todo >= skb_frag_size(fragfrom)) {
2554                         *fragto = *fragfrom;
2555                         todo -= skb_frag_size(fragfrom);
2556                         from++;
2557                         to++;
2558 
2559                 } else {
2560                         __skb_frag_ref(fragfrom);
2561                         fragto->page = fragfrom->page;
2562                         fragto->page_offset = fragfrom->page_offset;
2563                         skb_frag_size_set(fragto, todo);
2564 
2565                         fragfrom->page_offset += todo;
2566                         skb_frag_size_sub(fragfrom, todo);
2567                         todo = 0;
2568 
2569                         to++;
2570                         break;
2571                 }
2572         }
2573 
2574         /* Ready to "commit" this state change to tgt */
2575         skb_shinfo(tgt)->nr_frags = to;
2576 
2577         if (merge >= 0) {
2578                 fragfrom = &skb_shinfo(skb)->frags[0];
2579                 fragto = &skb_shinfo(tgt)->frags[merge];
2580 
2581                 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2582                 __skb_frag_unref(fragfrom);
2583         }
2584 
2585         /* Reposition in the original skb */
2586         to = 0;
2587         while (from < skb_shinfo(skb)->nr_frags)
2588                 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2589         skb_shinfo(skb)->nr_frags = to;
2590 
2591         BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2592 
2593 onlymerged:
2594         /* Most likely the tgt won't ever need its checksum anymore, skb on
2595          * the other hand might need it if it needs to be resent
2596          */
2597         tgt->ip_summed = CHECKSUM_PARTIAL;
2598         skb->ip_summed = CHECKSUM_PARTIAL;
2599 
2600         /* Yak, is it really working this way? Some helper please? */
2601         skb->len -= shiftlen;
2602         skb->data_len -= shiftlen;
2603         skb->truesize -= shiftlen;
2604         tgt->len += shiftlen;
2605         tgt->data_len += shiftlen;
2606         tgt->truesize += shiftlen;
2607 
2608         return shiftlen;
2609 }
2610 
2611 /**
2612  * skb_prepare_seq_read - Prepare a sequential read of skb data
2613  * @skb: the buffer to read
2614  * @from: lower offset of data to be read
2615  * @to: upper offset of data to be read
2616  * @st: state variable
2617  *
2618  * Initializes the specified state variable. Must be called before
2619  * invoking skb_seq_read() for the first time.
2620  */
2621 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2622                           unsigned int to, struct skb_seq_state *st)
2623 {
2624         st->lower_offset = from;
2625         st->upper_offset = to;
2626         st->root_skb = st->cur_skb = skb;
2627         st->frag_idx = st->stepped_offset = 0;
2628         st->frag_data = NULL;
2629 }
2630 EXPORT_SYMBOL(skb_prepare_seq_read);
2631 
2632 /**
2633  * skb_seq_read - Sequentially read skb data
2634  * @consumed: number of bytes consumed by the caller so far
2635  * @data: destination pointer for data to be returned
2636  * @st: state variable
2637  *
2638  * Reads a block of skb data at @consumed relative to the
2639  * lower offset specified to skb_prepare_seq_read(). Assigns
2640  * the head of the data block to @data and returns the length
2641  * of the block or 0 if the end of the skb data or the upper
2642  * offset has been reached.
2643  *
2644  * The caller is not required to consume all of the data
2645  * returned, i.e. @consumed is typically set to the number
2646  * of bytes already consumed and the next call to
2647  * skb_seq_read() will return the remaining part of the block.
2648  *
2649  * Note 1: The size of each block of data returned can be arbitrary,
2650  *       this limitation is the cost for zerocopy sequential
2651  *       reads of potentially non linear data.
2652  *
2653  * Note 2: Fragment lists within fragments are not implemented
2654  *       at the moment, state->root_skb could be replaced with
2655  *       a stack for this purpose.
2656  */
2657 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2658                           struct skb_seq_state *st)
2659 {
2660         unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2661         skb_frag_t *frag;
2662 
2663         if (unlikely(abs_offset >= st->upper_offset)) {
2664                 if (st->frag_data) {
2665                         kunmap_atomic(st->frag_data);
2666                         st->frag_data = NULL;
2667                 }
2668                 return 0;
2669         }
2670 
2671 next_skb:
2672         block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2673 
2674         if (abs_offset < block_limit && !st->frag_data) {
2675                 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2676                 return block_limit - abs_offset;
2677         }
2678 
2679         if (st->frag_idx == 0 && !st->frag_data)
2680                 st->stepped_offset += skb_headlen(st->cur_skb);
2681 
2682         while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2683                 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2684                 block_limit = skb_frag_size(frag) + st->stepped_offset;
2685 
2686                 if (abs_offset < block_limit) {
2687                         if (!st->frag_data)
2688                                 st->frag_data = kmap_atomic(skb_frag_page(frag));
2689 
2690                         *data = (u8 *) st->frag_data + frag->page_offset +
2691                                 (abs_offset - st->stepped_offset);
2692 
2693                         return block_limit - abs_offset;
2694                 }
2695 
2696                 if (st->frag_data) {
2697                         kunmap_atomic(st->frag_data);
2698                         st->frag_data = NULL;
2699                 }
2700 
2701                 st->frag_idx++;
2702                 st->stepped_offset += skb_frag_size(frag);
2703         }
2704 
2705         if (st->frag_data) {
2706                 kunmap_atomic(st->frag_data);
2707                 st->frag_data = NULL;
2708         }
2709 
2710         if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2711                 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2712                 st->frag_idx = 0;
2713                 goto next_skb;
2714         } else if (st->cur_skb->next) {
2715                 st->cur_skb = st->cur_skb->next;
2716                 st->frag_idx = 0;
2717                 goto next_skb;
2718         }
2719 
2720         return 0;
2721 }
2722 EXPORT_SYMBOL(skb_seq_read);
2723 
2724 /**
2725  * skb_abort_seq_read - Abort a sequential read of skb data
2726  * @st: state variable
2727  *
2728  * Must be called if skb_seq_read() was not called until it
2729  * returned 0.
2730  */
2731 void skb_abort_seq_read(struct skb_seq_state *st)
2732 {
2733         if (st->frag_data)
2734                 kunmap_atomic(st->frag_data);
2735 }
2736 EXPORT_SYMBOL(skb_abort_seq_read);
2737 
2738 #define TS_SKB_CB(state)        ((struct skb_seq_state *) &((state)->cb))
2739 
2740 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2741                                           struct ts_config *conf,
2742                                           struct ts_state *state)
2743 {
2744         return skb_seq_read(offset, text, TS_SKB_CB(state));
2745 }
2746 
2747 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2748 {
2749         skb_abort_seq_read(TS_SKB_CB(state));
2750 }
2751 
2752 /**
2753  * skb_find_text - Find a text pattern in skb data
2754  * @skb: the buffer to look in
2755  * @from: search offset
2756  * @to: search limit
2757  * @config: textsearch configuration
2758  * @state: uninitialized textsearch state variable
2759  *
2760  * Finds a pattern in the skb data according to the specified
2761  * textsearch configuration. Use textsearch_next() to retrieve
2762  * subsequent occurrences of the pattern. Returns the offset
2763  * to the first occurrence or UINT_MAX if no match was found.
2764  */
2765 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2766                            unsigned int to, struct ts_config *config,
2767                            struct ts_state *state)
2768 {
2769         unsigned int ret;
2770 
2771         config->get_next_block = skb_ts_get_next_block;
2772         config->finish = skb_ts_finish;
2773 
2774         skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2775 
2776         ret = textsearch_find(config, state);
2777         return (ret <= to - from ? ret : UINT_MAX);
2778 }
2779 EXPORT_SYMBOL(skb_find_text);
2780 
2781 /**
2782  * skb_append_datato_frags - append the user data to a skb
2783  * @sk: sock  structure
2784  * @skb: skb structure to be appended with user data.
2785  * @getfrag: call back function to be used for getting the user data
2786  * @from: pointer to user message iov
2787  * @length: length of the iov message
2788  *
2789  * Description: This procedure append the user data in the fragment part
2790  * of the skb if any page alloc fails user this procedure returns  -ENOMEM
2791  */
2792 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2793                         int (*getfrag)(void *from, char *to, int offset,
2794                                         int len, int odd, struct sk_buff *skb),
2795                         void *from, int length)
2796 {
2797         int frg_cnt = skb_shinfo(skb)->nr_frags;
2798         int copy;
2799         int offset = 0;
2800         int ret;
2801         struct page_frag *pfrag = &current->task_frag;
2802 
2803         do {
2804                 /* Return error if we don't have space for new frag */
2805                 if (frg_cnt >= MAX_SKB_FRAGS)
2806                         return -EMSGSIZE;
2807 
2808                 if (!sk_page_frag_refill(sk, pfrag))
2809                         return -ENOMEM;
2810 
2811                 /* copy the user data to page */
2812                 copy = min_t(int, length, pfrag->size - pfrag->offset);
2813 
2814                 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
2815                               offset, copy, 0, skb);
2816                 if (ret < 0)
2817                         return -EFAULT;
2818 
2819                 /* copy was successful so update the size parameters */
2820                 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
2821                                    copy);
2822                 frg_cnt++;
2823                 pfrag->offset += copy;
2824                 get_page(pfrag->page);
2825 
2826                 skb->truesize += copy;
2827                 atomic_add(copy, &sk->sk_wmem_alloc);
2828                 skb->len += copy;
2829                 skb->data_len += copy;
2830                 offset += copy;
2831                 length -= copy;
2832 
2833         } while (length > 0);
2834 
2835         return 0;
2836 }
2837 EXPORT_SYMBOL(skb_append_datato_frags);
2838 
2839 /**
2840  *      skb_pull_rcsum - pull skb and update receive checksum
2841  *      @skb: buffer to update
2842  *      @len: length of data pulled
2843  *
2844  *      This function performs an skb_pull on the packet and updates
2845  *      the CHECKSUM_COMPLETE checksum.  It should be used on
2846  *      receive path processing instead of skb_pull unless you know
2847  *      that the checksum difference is zero (e.g., a valid IP header)
2848  *      or you are setting ip_summed to CHECKSUM_NONE.
2849  */
2850 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2851 {
2852         BUG_ON(len > skb->len);
2853         skb->len -= len;
2854         BUG_ON(skb->len < skb->data_len);
2855         skb_postpull_rcsum(skb, skb->data, len);
2856         return skb->data += len;
2857 }
2858 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2859 
2860 /**
2861  *      skb_segment - Perform protocol segmentation on skb.
2862  *      @head_skb: buffer to segment
2863  *      @features: features for the output path (see dev->features)
2864  *
2865  *      This function performs segmentation on the given skb.  It returns
2866  *      a pointer to the first in a list of new skbs for the segments.
2867  *      In case of error it returns ERR_PTR(err).
2868  */
2869 struct sk_buff *skb_segment(struct sk_buff *head_skb,
2870                             netdev_features_t features)
2871 {
2872         struct sk_buff *segs = NULL;
2873         struct sk_buff *tail = NULL;
2874         struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
2875         skb_frag_t *frag = skb_shinfo(head_skb)->frags;
2876         unsigned int mss = skb_shinfo(head_skb)->gso_size;
2877         unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
2878         struct sk_buff *frag_skb = head_skb;
2879         unsigned int offset = doffset;
2880         unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
2881         unsigned int headroom;
2882         unsigned int len;
2883         __be16 proto;
2884         bool csum;
2885         int sg = !!(features & NETIF_F_SG);
2886         int nfrags = skb_shinfo(head_skb)->nr_frags;
2887         int err = -ENOMEM;
2888         int i = 0;
2889         int pos;
2890         int dummy;
2891 
2892         __skb_push(head_skb, doffset);
2893         proto = skb_network_protocol(head_skb, &dummy);
2894         if (unlikely(!proto))
2895                 return ERR_PTR(-EINVAL);
2896 
2897         csum = !head_skb->encap_hdr_csum &&
2898             !!can_checksum_protocol(features, proto);
2899 
2900         headroom = skb_headroom(head_skb);
2901         pos = skb_headlen(head_skb);
2902 
2903         do {
2904                 struct sk_buff *nskb;
2905                 skb_frag_t *nskb_frag;
2906                 int hsize;
2907                 int size;
2908 
2909                 len = head_skb->len - offset;
2910                 if (len > mss)
2911                         len = mss;
2912 
2913                 hsize = skb_headlen(head_skb) - offset;
2914                 if (hsize < 0)
2915                         hsize = 0;
2916                 if (hsize > len || !sg)
2917                         hsize = len;
2918 
2919                 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
2920                     (skb_headlen(list_skb) == len || sg)) {
2921                         BUG_ON(skb_headlen(list_skb) > len);
2922 
2923                         i = 0;
2924                         nfrags = skb_shinfo(list_skb)->nr_frags;
2925                         frag = skb_shinfo(list_skb)->frags;
2926                         frag_skb = list_skb;
2927                         pos += skb_headlen(list_skb);
2928 
2929                         while (pos < offset + len) {
2930                                 BUG_ON(i >= nfrags);
2931 
2932                                 size = skb_frag_size(frag);
2933                                 if (pos + size > offset + len)
2934                                         break;
2935 
2936                                 i++;
2937                                 pos += size;
2938                                 frag++;
2939                         }
2940 
2941                         nskb = skb_clone(list_skb, GFP_ATOMIC);
2942                         list_skb = list_skb->next;
2943 
2944                         if (unlikely(!nskb))
2945                                 goto err;
2946 
2947                         if (unlikely(pskb_trim(nskb, len))) {
2948                                 kfree_skb(nskb);
2949                                 goto err;
2950                         }
2951 
2952                         hsize = skb_end_offset(nskb);
2953                         if (skb_cow_head(nskb, doffset + headroom)) {
2954                                 kfree_skb(nskb);
2955                                 goto err;
2956                         }
2957 
2958                         nskb->truesize += skb_end_offset(nskb) - hsize;
2959                         skb_release_head_state(nskb);
2960                         __skb_push(nskb, doffset);
2961                 } else {
2962                         nskb = __alloc_skb(hsize + doffset + headroom,
2963                                            GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
2964                                            NUMA_NO_NODE);
2965 
2966                         if (unlikely(!nskb))
2967                                 goto err;
2968 
2969                         skb_reserve(nskb, headroom);
2970                         __skb_put(nskb, doffset);
2971                 }
2972 
2973                 if (segs)
2974                         tail->next = nskb;
2975                 else
2976                         segs = nskb;
2977                 tail = nskb;
2978 
2979                 __copy_skb_header(nskb, head_skb);
2980 
2981                 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
2982                 skb_reset_mac_len(nskb);
2983 
2984                 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
2985                                                  nskb->data - tnl_hlen,
2986                                                  doffset + tnl_hlen);
2987 
2988                 if (nskb->len == len + doffset)
2989                         goto perform_csum_check;
2990 
2991                 if (!sg) {
2992                         nskb->ip_summed = CHECKSUM_NONE;
2993                         nskb->csum = skb_copy_and_csum_bits(head_skb, offset,
2994                                                             skb_put(nskb, len),
2995                                                             len, 0);
2996                         SKB_GSO_CB(nskb)->csum_start =
2997                             skb_headroom(nskb) + doffset;
2998                         continue;
2999                 }
3000 
3001                 nskb_frag = skb_shinfo(nskb)->frags;
3002 
3003                 skb_copy_from_linear_data_offset(head_skb, offset,
3004                                                  skb_put(nskb, hsize), hsize);
3005 
3006                 skb_shinfo(nskb)->tx_flags = skb_shinfo(head_skb)->tx_flags &
3007                         SKBTX_SHARED_FRAG;
3008 
3009                 while (pos < offset + len) {
3010                         if (i >= nfrags) {
3011                                 BUG_ON(skb_headlen(list_skb));
3012 
3013                                 i = 0;
3014                                 nfrags = skb_shinfo(list_skb)->nr_frags;
3015                                 frag = skb_shinfo(list_skb)->frags;
3016                                 frag_skb = list_skb;
3017 
3018                                 BUG_ON(!nfrags);
3019 
3020                                 list_skb = list_skb->next;
3021                         }
3022 
3023                         if (unlikely(skb_shinfo(nskb)->nr_frags >=
3024                                      MAX_SKB_FRAGS)) {
3025                                 net_warn_ratelimited(
3026                                         "skb_segment: too many frags: %u %u\n",
3027                                         pos, mss);
3028                                 goto err;
3029                         }
3030 
3031                         if (unlikely(skb_orphan_frags(frag_skb, GFP_ATOMIC)))
3032                                 goto err;
3033 
3034                         *nskb_frag = *frag;
3035                         __skb_frag_ref(nskb_frag);
3036                         size = skb_frag_size(nskb_frag);
3037 
3038                         if (pos < offset) {
3039                                 nskb_frag->page_offset += offset - pos;
3040                                 skb_frag_size_sub(nskb_frag, offset - pos);
3041                         }
3042 
3043                         skb_shinfo(nskb)->nr_frags++;
3044 
3045                         if (pos + size <= offset + len) {
3046                                 i++;
3047                                 frag++;
3048                                 pos += size;
3049                         } else {
3050                                 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
3051                                 goto skip_fraglist;
3052                         }
3053 
3054                         nskb_frag++;
3055                 }
3056 
3057 skip_fraglist:
3058                 nskb->data_len = len - hsize;
3059                 nskb->len += nskb->data_len;
3060                 nskb->truesize += nskb->data_len;
3061 
3062 perform_csum_check:
3063                 if (!csum) {
3064                         nskb->csum = skb_checksum(nskb, doffset,
3065                                                   nskb->len - doffset, 0);
3066                         nskb->ip_summed = CHECKSUM_NONE;
3067                         SKB_GSO_CB(nskb)->csum_start =
3068                             skb_headroom(nskb) + doffset;
3069                 }
3070         } while ((offset += len) < head_skb->len);
3071 
3072         return segs;
3073 
3074 err:
3075         kfree_skb_list(segs);
3076         return ERR_PTR(err);
3077 }
3078 EXPORT_SYMBOL_GPL(skb_segment);
3079 
3080 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
3081 {
3082         struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
3083         unsigned int offset = skb_gro_offset(skb);
3084         unsigned int headlen = skb_headlen(skb);
3085         struct sk_buff *nskb, *lp, *p = *head;
3086         unsigned int len = skb_gro_len(skb);
3087         unsigned int delta_truesize;
3088         unsigned int headroom;
3089 
3090         if (unlikely(p->len + len >= 65536))
3091                 return -E2BIG;
3092 
3093         lp = NAPI_GRO_CB(p)->last;
3094         pinfo = skb_shinfo(lp);
3095 
3096         if (headlen <= offset) {
3097                 skb_frag_t *frag;
3098                 skb_frag_t *frag2;
3099                 int i = skbinfo->nr_frags;
3100                 int nr_frags = pinfo->nr_frags + i;
3101 
3102                 if (nr_frags > MAX_SKB_FRAGS)
3103                         goto merge;
3104 
3105                 offset -= headlen;
3106                 pinfo->nr_frags = nr_frags;
3107                 skbinfo->nr_frags = 0;
3108 
3109                 frag = pinfo->frags + nr_frags;
3110                 frag2 = skbinfo->frags + i;
3111                 do {
3112                         *--frag = *--frag2;
3113                 } while (--i);
3114 
3115                 frag->page_offset += offset;
3116                 skb_frag_size_sub(frag, offset);
3117 
3118                 /* all fragments truesize : remove (head size + sk_buff) */
3119                 delta_truesize = skb->truesize -
3120                                  SKB_TRUESIZE(skb_end_offset(skb));
3121 
3122                 skb->truesize -= skb->data_len;
3123                 skb->len -= skb->data_len;
3124                 skb->data_len = 0;
3125 
3126                 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
3127                 goto done;
3128         } else if (skb->head_frag) {
3129                 int nr_frags = pinfo->nr_frags;
3130                 skb_frag_t *frag = pinfo->frags + nr_frags;
3131                 struct page *page = virt_to_head_page(skb->head);
3132                 unsigned int first_size = headlen - offset;
3133                 unsigned int first_offset;
3134 
3135                 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
3136                         goto merge;
3137 
3138                 first_offset = skb->data -
3139                                (unsigned char *)page_address(page) +
3140                                offset;
3141 
3142                 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3143 
3144                 frag->page.p      = page;
3145                 frag->page_offset = first_offset;
3146                 skb_frag_size_set(frag, first_size);
3147 
3148                 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3149                 /* We dont need to clear skbinfo->nr_frags here */
3150 
3151                 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3152                 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3153                 goto done;
3154         }
3155         /* switch back to head shinfo */
3156         pinfo = skb_shinfo(p);
3157 
3158         if (pinfo->frag_list)
3159                 goto merge;
3160         if (skb_gro_len(p) != pinfo->gso_size)
3161                 return -E2BIG;
3162 
3163         headroom = skb_headroom(p);
3164         nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
3165         if (unlikely(!nskb))
3166                 return -ENOMEM;
3167 
3168         __copy_skb_header(nskb, p);
3169         nskb->mac_len = p->mac_len;
3170 
3171         skb_reserve(nskb, headroom);
3172         __skb_put(nskb, skb_gro_offset(p));
3173 
3174         skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
3175         skb_set_network_header(nskb, skb_network_offset(p));
3176         skb_set_transport_header(nskb, skb_transport_offset(p));
3177 
3178         __skb_pull(p, skb_gro_offset(p));
3179         memcpy(skb_mac_header(nskb), skb_mac_header(p),
3180                p->data - skb_mac_header(p));
3181 
3182         skb_shinfo(nskb)->frag_list = p;
3183         skb_shinfo(nskb)->gso_size = pinfo->gso_size;
3184         pinfo->gso_size = 0;
3185         skb_header_release(p);
3186         NAPI_GRO_CB(nskb)->last = p;
3187 
3188         nskb->data_len += p->len;
3189         nskb->truesize += p->truesize;
3190         nskb->len += p->len;
3191 
3192         *head = nskb;
3193         nskb->next = p->next;
3194         p->next = NULL;
3195 
3196         p = nskb;
3197 
3198 merge:
3199         delta_truesize = skb->truesize;
3200         if (offset > headlen) {
3201                 unsigned int eat = offset - headlen;
3202 
3203                 skbinfo->frags[0].page_offset += eat;
3204                 skb_frag_size_sub(&skbinfo->frags[0], eat);
3205                 skb->data_len -= eat;
3206                 skb->len -= eat;
3207                 offset = headlen;
3208         }
3209 
3210         __skb_pull(skb, offset);
3211 
3212         if (NAPI_GRO_CB(p)->last == p)
3213                 skb_shinfo(p)->frag_list = skb;
3214         else
3215                 NAPI_GRO_CB(p)->last->next = skb;
3216         NAPI_GRO_CB(p)->last = skb;
3217         skb_header_release(skb);
3218         lp = p;
3219 
3220 done:
3221         NAPI_GRO_CB(p)->count++;
3222         p->data_len += len;
3223         p->truesize += delta_truesize;
3224         p->len += len;
3225         if (lp != p) {
3226                 lp->data_len += len;
3227                 lp->truesize += delta_truesize;
3228                 lp->len += len;
3229         }
3230         NAPI_GRO_CB(skb)->same_flow = 1;
3231         return 0;
3232 }
3233 EXPORT_SYMBOL_GPL(skb_gro_receive);
3234 
3235 void __init skb_init(void)
3236 {
3237         skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3238                                               sizeof(struct sk_buff),
3239                                               0,
3240                                               SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3241                                               NULL);
3242         skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3243                                                 (2*sizeof(struct sk_buff)) +
3244                                                 sizeof(atomic_t),
3245                                                 0,
3246                                                 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3247                                                 NULL);
3248 }
3249 
3250 /**
3251  *      skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3252  *      @skb: Socket buffer containing the buffers to be mapped
3253  *      @sg: The scatter-gather list to map into
3254  *      @offset: The offset into the buffer's contents to start mapping
3255  *      @len: Length of buffer space to be mapped
3256  *
3257  *      Fill the specified scatter-gather list with mappings/pointers into a
3258  *      region of the buffer space attached to a socket buffer.
3259  */
3260 static int
3261 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3262 {
3263         int start = skb_headlen(skb);
3264         int i, copy = start - offset;
3265         struct sk_buff *frag_iter;
3266         int elt = 0;
3267 
3268         if (copy > 0) {
3269                 if (copy > len)
3270                         copy = len;
3271                 sg_set_buf(sg, skb->data + offset, copy);
3272                 elt++;
3273                 if ((len -= copy) == 0)
3274                         return elt;
3275                 offset += copy;
3276         }
3277 
3278         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3279                 int end;
3280 
3281                 WARN_ON(start > offset + len);
3282 
3283                 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3284                 if ((copy = end - offset) > 0) {
3285                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3286 
3287                         if (copy > len)
3288                                 copy = len;
3289                         sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3290                                         frag->page_offset+offset-start);
3291                         elt++;
3292                         if (!(len -= copy))
3293                                 return elt;
3294                         offset += copy;
3295                 }
3296                 start = end;
3297         }
3298 
3299         skb_walk_frags(skb, frag_iter) {
3300                 int end;
3301 
3302                 WARN_ON(start > offset + len);
3303 
3304                 end = start + frag_iter->len;
3305                 if ((copy = end - offset) > 0) {
3306                         if (copy > len)
3307                                 copy = len;
3308                         elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3309                                               copy);
3310                         if ((len -= copy) == 0)
3311                                 return elt;
3312                         offset += copy;
3313                 }
3314                 start = end;
3315         }
3316         BUG_ON(len);
3317         return elt;
3318 }
3319 
3320 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
3321  * sglist without mark the sg which contain last skb data as the end.
3322  * So the caller can mannipulate sg list as will when padding new data after
3323  * the first call without calling sg_unmark_end to expend sg list.
3324  *
3325  * Scenario to use skb_to_sgvec_nomark:
3326  * 1. sg_init_table
3327  * 2. skb_to_sgvec_nomark(payload1)
3328  * 3. skb_to_sgvec_nomark(payload2)
3329  *
3330  * This is equivalent to:
3331  * 1. sg_init_table
3332  * 2. skb_to_sgvec(payload1)
3333  * 3. sg_unmark_end
3334  * 4. skb_to_sgvec(payload2)
3335  *
3336  * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
3337  * is more preferable.
3338  */
3339 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
3340                         int offset, int len)
3341 {
3342         return __skb_to_sgvec(skb, sg, offset, len);
3343 }
3344 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
3345 
3346 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3347 {
3348         int nsg = __skb_to_sgvec(skb, sg, offset, len);
3349 
3350         sg_mark_end(&sg[nsg - 1]);
3351 
3352         return nsg;
3353 }
3354 EXPORT_SYMBOL_GPL(skb_to_sgvec);
3355 
3356 /**
3357  *      skb_cow_data - Check that a socket buffer's data buffers are writable
3358  *      @skb: The socket buffer to check.
3359  *      @tailbits: Amount of trailing space to be added
3360  *      @trailer: Returned pointer to the skb where the @tailbits space begins
3361  *
3362  *      Make sure that the data buffers attached to a socket buffer are
3363  *      writable. If they are not, private copies are made of the data buffers
3364  *      and the socket buffer is set to use these instead.
3365  *
3366  *      If @tailbits is given, make sure that there is space to write @tailbits
3367  *      bytes of data beyond current end of socket buffer.  @trailer will be
3368  *      set to point to the skb in which this space begins.
3369  *
3370  *      The number of scatterlist elements required to completely map the
3371  *      COW'd and extended socket buffer will be returned.
3372  */
3373 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3374 {
3375         int copyflag;
3376         int elt;
3377         struct sk_buff *skb1, **skb_p;
3378 
3379         /* If skb is cloned or its head is paged, reallocate
3380          * head pulling out all the pages (pages are considered not writable
3381          * at the moment even if they are anonymous).
3382          */
3383         if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3384             __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3385                 return -ENOMEM;
3386 
3387         /* Easy case. Most of packets will go this way. */
3388         if (!skb_has_frag_list(skb)) {
3389                 /* A little of trouble, not enough of space for trailer.
3390                  * This should not happen, when stack is tuned to generate
3391                  * good frames. OK, on miss we reallocate and reserve even more
3392                  * space, 128 bytes is fair. */
3393 
3394                 if (skb_tailroom(skb) < tailbits &&
3395                     pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3396                         return -ENOMEM;
3397 
3398                 /* Voila! */
3399                 *trailer = skb;
3400                 return 1;
3401         }
3402 
3403         /* Misery. We are in troubles, going to mincer fragments... */
3404 
3405         elt = 1;
3406         skb_p = &skb_shinfo(skb)->frag_list;
3407         copyflag = 0;
3408 
3409         while ((skb1 = *skb_p) != NULL) {
3410                 int ntail = 0;
3411 
3412                 /* The fragment is partially pulled by someone,
3413                  * this can happen on input. Copy it and everything
3414                  * after it. */
3415 
3416                 if (skb_shared(skb1))
3417                         copyflag = 1;
3418 
3419                 /* If the skb is the last, worry about trailer. */
3420 
3421                 if (skb1->next == NULL && tailbits) {
3422                         if (skb_shinfo(skb1)->nr_frags ||
3423                             skb_has_frag_list(skb1) ||
3424                             skb_tailroom(skb1) < tailbits)
3425                                 ntail = tailbits + 128;
3426                 }
3427 
3428                 if (copyflag ||
3429                     skb_cloned(skb1) ||
3430                     ntail ||
3431                     skb_shinfo(skb1)->nr_frags ||
3432                     skb_has_frag_list(skb1)) {
3433                         struct sk_buff *skb2;
3434 
3435                         /* Fuck, we are miserable poor guys... */
3436                         if (ntail == 0)
3437                                 skb2 = skb_copy(skb1, GFP_ATOMIC);
3438                         else
3439                                 skb2 = skb_copy_expand(skb1,
3440                                                        skb_headroom(skb1),
3441                                                        ntail,
3442                                                        GFP_ATOMIC);
3443                         if (unlikely(skb2 == NULL))
3444                                 return -ENOMEM;
3445 
3446                         if (skb1->sk)
3447                                 skb_set_owner_w(skb2, skb1->sk);
3448 
3449                         /* Looking around. Are we still alive?
3450                          * OK, link new skb, drop old one */
3451 
3452                         skb2->next = skb1->next;
3453                         *skb_p = skb2;
3454                         kfree_skb(skb1);
3455                         skb1 = skb2;
3456                 }
3457                 elt++;
3458                 *trailer = skb1;
3459                 skb_p = &skb1->next;
3460         }
3461 
3462         return elt;
3463 }
3464 EXPORT_SYMBOL_GPL(skb_cow_data);
3465 
3466 static void sock_rmem_free(struct sk_buff *skb)
3467 {
3468         struct sock *sk = skb->sk;
3469 
3470         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3471 }
3472 
3473 /*
3474  * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3475  */
3476 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3477 {
3478         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3479             (unsigned int)sk->sk_rcvbuf)
3480                 return -ENOMEM;
3481 
3482         skb_orphan(skb);
3483         skb->sk = sk;
3484         skb->destructor = sock_rmem_free;
3485         atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3486 
3487         /* before exiting rcu section, make sure dst is refcounted */
3488         skb_dst_force(skb);
3489 
3490         skb_queue_tail(&sk->sk_error_queue, skb);
3491         if (!sock_flag(sk, SOCK_DEAD))
3492                 sk->sk_data_ready(sk);
3493         return 0;
3494 }
3495 EXPORT_SYMBOL(sock_queue_err_skb);
3496 
3497 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3498                      struct skb_shared_hwtstamps *hwtstamps,
3499                      struct sock *sk, int tstype)
3500 {
3501         struct sock_exterr_skb *serr;
3502         struct sk_buff *skb;
3503         int err;
3504 
3505         if (!sk)
3506                 return;
3507 
3508         if (hwtstamps) {
3509                 *skb_hwtstamps(orig_skb) =
3510                         *hwtstamps;
3511         } else {
3512                 /*
3513                  * no hardware time stamps available,
3514                  * so keep the shared tx_flags and only
3515                  * store software time stamp
3516                  */
3517                 orig_skb->tstamp = ktime_get_real();
3518         }
3519 
3520         skb = skb_clone(orig_skb, GFP_ATOMIC);
3521         if (!skb)
3522                 return;
3523 
3524         serr = SKB_EXT_ERR(skb);
3525         memset(serr, 0, sizeof(*serr));
3526         serr->ee.ee_errno = ENOMSG;
3527         serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3528         serr->ee.ee_info = tstype;
3529         if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
3530                 serr->ee.ee_data = skb_shinfo(skb)->tskey;
3531                 if (sk->sk_protocol == IPPROTO_TCP)
3532                         serr->ee.ee_data -= sk->sk_tskey;
3533         }
3534 
3535         err = sock_queue_err_skb(sk, skb);
3536 
3537         if (err)
3538                 kfree_skb(skb);
3539 }
3540 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
3541 
3542 void skb_tstamp_tx(struct sk_buff *orig_skb,
3543                    struct skb_shared_hwtstamps *hwtstamps)
3544 {
3545         return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
3546                                SCM_TSTAMP_SND);
3547 }
3548 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3549 
3550 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3551 {
3552         struct sock *sk = skb->sk;
3553         struct sock_exterr_skb *serr;
3554         int err;
3555 
3556         skb->wifi_acked_valid = 1;
3557         skb->wifi_acked = acked;
3558 
3559         serr = SKB_EXT_ERR(skb);
3560         memset(serr, 0, sizeof(*serr));
3561         serr->ee.ee_errno = ENOMSG;
3562         serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3563 
3564         err = sock_queue_err_skb(sk, skb);
3565         if (err)
3566                 kfree_skb(skb);
3567 }
3568 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3569 
3570 
3571 /**
3572  * skb_partial_csum_set - set up and verify partial csum values for packet
3573  * @skb: the skb to set
3574  * @start: the number of bytes after skb->data to start checksumming.
3575  * @off: the offset from start to place the checksum.
3576  *
3577  * For untrusted partially-checksummed packets, we need to make sure the values
3578  * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3579  *
3580  * This function checks and sets those values and skb->ip_summed: if this
3581  * returns false you should drop the packet.
3582  */
3583 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3584 {
3585         if (unlikely(start > skb_headlen(skb)) ||
3586             unlikely((int)start + off > skb_headlen(skb) - 2)) {
3587                 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3588                                      start, off, skb_headlen(skb));
3589                 return false;
3590         }
3591         skb->ip_summed = CHECKSUM_PARTIAL;
3592         skb->csum_start = skb_headroom(skb) + start;
3593         skb->csum_offset = off;
3594         skb_set_transport_header(skb, start);
3595         return true;
3596 }
3597 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3598 
3599 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
3600                                unsigned int max)
3601 {
3602         if (skb_headlen(skb) >= len)
3603                 return 0;
3604 
3605         /* If we need to pullup then pullup to the max, so we
3606          * won't need to do it again.
3607          */
3608         if (max > skb->len)
3609                 max = skb->len;
3610 
3611         if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
3612                 return -ENOMEM;
3613 
3614         if (skb_headlen(skb) < len)
3615                 return -EPROTO;
3616 
3617         return 0;
3618 }
3619 
3620 #define MAX_TCP_HDR_LEN (15 * 4)
3621 
3622 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
3623                                       typeof(IPPROTO_IP) proto,
3624                                       unsigned int off)
3625 {
3626         switch (proto) {
3627                 int err;
3628 
3629         case IPPROTO_TCP:
3630                 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
3631                                           off + MAX_TCP_HDR_LEN);
3632                 if (!err && !skb_partial_csum_set(skb, off,
3633                                                   offsetof(struct tcphdr,
3634                                                            check)))
3635                         err = -EPROTO;
3636                 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
3637 
3638         case IPPROTO_UDP:
3639                 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
3640                                           off + sizeof(struct udphdr));
3641                 if (!err && !skb_partial_csum_set(skb, off,
3642                                                   offsetof(struct udphdr,
3643                                                            check)))
3644                         err = -EPROTO;
3645                 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
3646         }
3647 
3648         return ERR_PTR(-EPROTO);
3649 }
3650 
3651 /* This value should be large enough to cover a tagged ethernet header plus
3652  * maximally sized IP and TCP or UDP headers.
3653  */
3654 #define MAX_IP_HDR_LEN 128
3655 
3656 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
3657 {
3658         unsigned int off;
3659         bool fragment;
3660         __sum16 *csum;
3661         int err;
3662 
3663         fragment = false;
3664 
3665         err = skb_maybe_pull_tail(skb,
3666                                   sizeof(struct iphdr),
3667                                   MAX_IP_HDR_LEN);
3668         if (err < 0)
3669                 goto out;
3670 
3671         if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
3672                 fragment = true;
3673 
3674         off = ip_hdrlen(skb);
3675 
3676         err = -EPROTO;
3677 
3678         if (fragment)
3679                 goto out;
3680 
3681         csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
3682         if (IS_ERR(csum))
3683                 return PTR_ERR(csum);
3684 
3685         if (recalculate)
3686                 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3687                                            ip_hdr(skb)->daddr,
3688                                            skb->len - off,
3689                                            ip_hdr(skb)->protocol, 0);
3690         err = 0;
3691 
3692 out:
3693         return err;
3694 }
3695 
3696 /* This value should be large enough to cover a tagged ethernet header plus
3697  * an IPv6 header, all options, and a maximal TCP or UDP header.
3698  */
3699 #define MAX_IPV6_HDR_LEN 256
3700 
3701 #define OPT_HDR(type, skb, off) \
3702         (type *)(skb_network_header(skb) + (off))
3703 
3704 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
3705 {
3706         int err;
3707         u8 nexthdr;
3708         unsigned int off;
3709         unsigned int len;
3710         bool fragment;
3711         bool done;
3712         __sum16 *csum;
3713 
3714         fragment = false;
3715         done = false;
3716 
3717         off = sizeof(struct ipv6hdr);
3718 
3719         err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
3720         if (err < 0)
3721                 goto out;
3722 
3723         nexthdr = ipv6_hdr(skb)->nexthdr;
3724 
3725         len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
3726         while (off <= len && !done) {
3727                 switch (nexthdr) {
3728                 case IPPROTO_DSTOPTS:
3729                 case IPPROTO_HOPOPTS:
3730                 case IPPROTO_ROUTING: {
3731                         struct ipv6_opt_hdr *hp;
3732 
3733                         err = skb_maybe_pull_tail(skb,
3734                                                   off +
3735                                                   sizeof(struct ipv6_opt_hdr),
3736                                                   MAX_IPV6_HDR_LEN);
3737                         if (err < 0)
3738                                 goto out;
3739 
3740                         hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
3741                         nexthdr = hp->nexthdr;
3742                         off += ipv6_optlen(hp);
3743                         break;
3744                 }
3745                 case IPPROTO_AH: {
3746                         struct ip_auth_hdr *hp;
3747 
3748                         err = skb_maybe_pull_tail(skb,
3749                                                   off +
3750                                                   sizeof(struct ip_auth_hdr),
3751                                                   MAX_IPV6_HDR_LEN);
3752                         if (err < 0)
3753                                 goto out;
3754 
3755                         hp = OPT_HDR(struct ip_auth_hdr, skb, off);
3756                         nexthdr = hp->nexthdr;
3757                         off += ipv6_authlen(hp);
3758                         break;
3759                 }
3760                 case IPPROTO_FRAGMENT: {
3761                         struct frag_hdr *hp;
3762 
3763                         err = skb_maybe_pull_tail(skb,
3764                                                   off +
3765                                                   sizeof(struct frag_hdr),
3766                                                   MAX_IPV6_HDR_LEN);
3767                         if (err < 0)
3768                                 goto out;
3769 
3770                         hp = OPT_HDR(struct frag_hdr, skb, off);
3771 
3772                         if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
3773                                 fragment = true;
3774 
3775                         nexthdr = hp->nexthdr;
3776                         off += sizeof(struct frag_hdr);
3777                         break;
3778                 }
3779                 default:
3780                         done = true;
3781                         break;
3782                 }
3783         }
3784 
3785         err = -EPROTO;
3786 
3787         if (!done || fragment)
3788                 goto out;
3789 
3790         csum = skb_checksum_setup_ip(skb, nexthdr, off);
3791         if (IS_ERR(csum))
3792                 return PTR_ERR(csum);
3793 
3794         if (recalculate)
3795                 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3796                                          &ipv6_hdr(skb)->daddr,
3797                                          skb->len - off, nexthdr, 0);
3798         err = 0;
3799 
3800 out:
3801         return err;
3802 }
3803 
3804 /**
3805  * skb_checksum_setup - set up partial checksum offset
3806  * @skb: the skb to set up
3807  * @recalculate: if true the pseudo-header checksum will be recalculated
3808  */
3809 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
3810 {
3811         int err;
3812 
3813         switch (skb->protocol) {
3814         case htons(ETH_P_IP):
3815                 err = skb_checksum_setup_ipv4(skb, recalculate);
3816                 break;
3817 
3818         case htons(ETH_P_IPV6):
3819                 err = skb_checksum_setup_ipv6(skb, recalculate);
3820                 break;
3821 
3822         default:
3823                 err = -EPROTO;
3824                 break;
3825         }
3826 
3827         return err;
3828 }
3829 EXPORT_SYMBOL(skb_checksum_setup);
3830 
3831 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3832 {
3833         net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
3834                              skb->dev->name);
3835 }
3836 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
3837 
3838 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
3839 {
3840         if (head_stolen) {
3841                 skb_release_head_state(skb);
3842                 kmem_cache_free(skbuff_head_cache, skb);
3843         } else {
3844                 __kfree_skb(skb);
3845         }
3846 }
3847 EXPORT_SYMBOL(kfree_skb_partial);
3848 
3849 /**
3850  * skb_try_coalesce - try to merge skb to prior one
3851  * @to: prior buffer
3852  * @from: buffer to add
3853  * @fragstolen: pointer to boolean
3854  * @delta_truesize: how much more was allocated than was requested
3855  */
3856 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
3857                       bool *fragstolen, int *delta_truesize)
3858 {
3859         int i, delta, len = from->len;
3860 
3861         *fragstolen = false;
3862 
3863         if (skb_cloned(to))
3864                 return false;
3865 
3866         if (len <= skb_tailroom(to)) {
3867                 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
3868                 *delta_truesize = 0;
3869                 return true;
3870         }
3871 
3872         if (skb_has_frag_list(to) || skb_has_frag_list(from))
3873                 return false;
3874 
3875         if (skb_headlen(from) != 0) {
3876                 struct page *page;
3877                 unsigned int offset;
3878 
3879                 if (skb_shinfo(to)->nr_frags +
3880                     skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
3881                         return false;
3882 
3883                 if (skb_head_is_locked(from))
3884                         return false;
3885 
3886                 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3887 
3888                 page = virt_to_head_page(from->head);
3889                 offset = from->data - (unsigned char *)page_address(page);
3890 
3891                 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
3892                                    page, offset, skb_headlen(from));
3893                 *fragstolen = true;
3894         } else {
3895                 if (skb_shinfo(to)->nr_frags +
3896                     skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
3897                         return false;
3898 
3899                 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
3900         }
3901 
3902         WARN_ON_ONCE(delta < len);
3903 
3904         memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
3905                skb_shinfo(from)->frags,
3906                skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
3907         skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
3908 
3909         if (!skb_cloned(from))
3910                 skb_shinfo(from)->nr_frags = 0;
3911 
3912         /* if the skb is not cloned this does nothing
3913          * since we set nr_frags to 0.
3914          */
3915         for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
3916                 skb_frag_ref(from, i);
3917 
3918         to->truesize += delta;
3919         to->len += len;
3920         to->data_len += len;
3921 
3922         *delta_truesize = delta;
3923         return true;
3924 }
3925 EXPORT_SYMBOL(skb_try_coalesce);
3926 
3927 /**
3928  * skb_scrub_packet - scrub an skb
3929  *
3930  * @skb: buffer to clean
3931  * @xnet: packet is crossing netns
3932  *
3933  * skb_scrub_packet can be used after encapsulating or decapsulting a packet
3934  * into/from a tunnel. Some information have to be cleared during these
3935  * operations.
3936  * skb_scrub_packet can also be used to clean a skb before injecting it in
3937  * another namespace (@xnet == true). We have to clear all information in the
3938  * skb that could impact namespace isolation.
3939  */
3940 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
3941 {
3942         if (xnet)
3943                 skb_orphan(skb);
3944         skb->tstamp.tv64 = 0;
3945         skb->pkt_type = PACKET_HOST;
3946         skb->skb_iif = 0;
3947         skb->ignore_df = 0;
3948         skb_dst_drop(skb);
3949         skb->mark = 0;
3950         secpath_reset(skb);
3951         nf_reset(skb);
3952         nf_reset_trace(skb);
3953 }
3954 EXPORT_SYMBOL_GPL(skb_scrub_packet);
3955 
3956 /**
3957  * skb_gso_transport_seglen - Return length of individual segments of a gso packet
3958  *
3959  * @skb: GSO skb
3960  *
3961  * skb_gso_transport_seglen is used to determine the real size of the
3962  * individual segments, including Layer4 headers (TCP/UDP).
3963  *
3964  * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
3965  */
3966 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
3967 {
3968         const struct skb_shared_info *shinfo = skb_shinfo(skb);
3969 
3970         if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
3971                 return tcp_hdrlen(skb) + shinfo->gso_size;
3972 
3973         /* UFO sets gso_size to the size of the fragmentation
3974          * payload, i.e. the size of the L4 (UDP) header is already
3975          * accounted for.
3976          */
3977         return shinfo->gso_size;
3978 }
3979 EXPORT_SYMBOL_GPL(skb_gso_transport_seglen);
3980 
3981 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
3982 {
3983         if (skb_cow(skb, skb_headroom(skb)) < 0) {
3984                 kfree_skb(skb);
3985                 return NULL;
3986         }
3987 
3988         memmove(skb->data - ETH_HLEN, skb->data - VLAN_ETH_HLEN, 2 * ETH_ALEN);
3989         skb->mac_header += VLAN_HLEN;
3990         return skb;
3991 }
3992 
3993 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
3994 {
3995         struct vlan_hdr *vhdr;
3996         u16 vlan_tci;
3997 
3998         if (unlikely(vlan_tx_tag_present(skb))) {
3999                 /* vlan_tci is already set-up so leave this for another time */
4000                 return skb;
4001         }
4002 
4003         skb = skb_share_check(skb, GFP_ATOMIC);
4004         if (unlikely(!skb))
4005                 goto err_free;
4006 
4007         if (unlikely(!pskb_may_pull(skb, VLAN_HLEN)))
4008                 goto err_free;
4009 
4010         vhdr = (struct vlan_hdr *)skb->data;
4011         vlan_tci = ntohs(vhdr->h_vlan_TCI);
4012         __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
4013 
4014         skb_pull_rcsum(skb, VLAN_HLEN);
4015         vlan_set_encap_proto(skb, vhdr);
4016 
4017         skb = skb_reorder_vlan_header(skb);
4018         if (unlikely(!skb))
4019                 goto err_free;
4020 
4021         skb_reset_network_header(skb);
4022         skb_reset_transport_header(skb);
4023         skb_reset_mac_len(skb);
4024 
4025         return skb;
4026 
4027 err_free:
4028         kfree_skb(skb);
4029         return NULL;
4030 }
4031 EXPORT_SYMBOL(skb_vlan_untag);
4032 

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