Version:  2.0.40 2.2.26 2.4.37 3.11 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8

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/sctp.h>
 53 #include <linux/netdevice.h>
 54 #ifdef CONFIG_NET_CLS_ACT
 55 #include <net/pkt_sched.h>
 56 #endif
 57 #include <linux/string.h>
 58 #include <linux/skbuff.h>
 59 #include <linux/splice.h>
 60 #include <linux/cache.h>
 61 #include <linux/rtnetlink.h>
 62 #include <linux/init.h>
 63 #include <linux/scatterlist.h>
 64 #include <linux/errqueue.h>
 65 #include <linux/prefetch.h>
 66 #include <linux/if_vlan.h>
 67 
 68 #include <net/protocol.h>
 69 #include <net/dst.h>
 70 #include <net/sock.h>
 71 #include <net/checksum.h>
 72 #include <net/ip6_checksum.h>
 73 #include <net/xfrm.h>
 74 
 75 #include <asm/uaccess.h>
 76 #include <trace/events/skb.h>
 77 #include <linux/highmem.h>
 78 #include <linux/capability.h>
 79 #include <linux/user_namespace.h>
 80 
 81 struct kmem_cache *skbuff_head_cache __read_mostly;
 82 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
 83 int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
 84 EXPORT_SYMBOL(sysctl_max_skb_frags);
 85 
 86 /**
 87  *      skb_panic - private function for out-of-line support
 88  *      @skb:   buffer
 89  *      @sz:    size
 90  *      @addr:  address
 91  *      @msg:   skb_over_panic or skb_under_panic
 92  *
 93  *      Out-of-line support for skb_put() and skb_push().
 94  *      Called via the wrapper skb_over_panic() or skb_under_panic().
 95  *      Keep out of line to prevent kernel bloat.
 96  *      __builtin_return_address is not used because it is not always reliable.
 97  */
 98 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
 99                       const char msg[])
100 {
101         pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
102                  msg, addr, skb->len, sz, skb->head, skb->data,
103                  (unsigned long)skb->tail, (unsigned long)skb->end,
104                  skb->dev ? skb->dev->name : "<NULL>");
105         BUG();
106 }
107 
108 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
109 {
110         skb_panic(skb, sz, addr, __func__);
111 }
112 
113 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
114 {
115         skb_panic(skb, sz, addr, __func__);
116 }
117 
118 /*
119  * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
120  * the caller if emergency pfmemalloc reserves are being used. If it is and
121  * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
122  * may be used. Otherwise, the packet data may be discarded until enough
123  * memory is free
124  */
125 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
126          __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
127 
128 static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
129                                unsigned long ip, bool *pfmemalloc)
130 {
131         void *obj;
132         bool ret_pfmemalloc = false;
133 
134         /*
135          * Try a regular allocation, when that fails and we're not entitled
136          * to the reserves, fail.
137          */
138         obj = kmalloc_node_track_caller(size,
139                                         flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
140                                         node);
141         if (obj || !(gfp_pfmemalloc_allowed(flags)))
142                 goto out;
143 
144         /* Try again but now we are using pfmemalloc reserves */
145         ret_pfmemalloc = true;
146         obj = kmalloc_node_track_caller(size, flags, node);
147 
148 out:
149         if (pfmemalloc)
150                 *pfmemalloc = ret_pfmemalloc;
151 
152         return obj;
153 }
154 
155 /*      Allocate a new skbuff. We do this ourselves so we can fill in a few
156  *      'private' fields and also do memory statistics to find all the
157  *      [BEEP] leaks.
158  *
159  */
160 
161 struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node)
162 {
163         struct sk_buff *skb;
164 
165         /* Get the HEAD */
166         skb = kmem_cache_alloc_node(skbuff_head_cache,
167                                     gfp_mask & ~__GFP_DMA, node);
168         if (!skb)
169                 goto out;
170 
171         /*
172          * Only clear those fields we need to clear, not those that we will
173          * actually initialise below. Hence, don't put any more fields after
174          * the tail pointer in struct sk_buff!
175          */
176         memset(skb, 0, offsetof(struct sk_buff, tail));
177         skb->head = NULL;
178         skb->truesize = sizeof(struct sk_buff);
179         atomic_set(&skb->users, 1);
180 
181         skb->mac_header = (typeof(skb->mac_header))~0U;
182 out:
183         return skb;
184 }
185 
186 /**
187  *      __alloc_skb     -       allocate a network buffer
188  *      @size: size to allocate
189  *      @gfp_mask: allocation mask
190  *      @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
191  *              instead of head cache and allocate a cloned (child) skb.
192  *              If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
193  *              allocations in case the data is required for writeback
194  *      @node: numa node to allocate memory on
195  *
196  *      Allocate a new &sk_buff. The returned buffer has no headroom and a
197  *      tail room of at least size bytes. The object has a reference count
198  *      of one. The return is the buffer. On a failure the return is %NULL.
199  *
200  *      Buffers may only be allocated from interrupts using a @gfp_mask of
201  *      %GFP_ATOMIC.
202  */
203 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
204                             int flags, int node)
205 {
206         struct kmem_cache *cache;
207         struct skb_shared_info *shinfo;
208         struct sk_buff *skb;
209         u8 *data;
210         bool pfmemalloc;
211 
212         cache = (flags & SKB_ALLOC_FCLONE)
213                 ? skbuff_fclone_cache : skbuff_head_cache;
214 
215         if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
216                 gfp_mask |= __GFP_MEMALLOC;
217 
218         /* Get the HEAD */
219         skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
220         if (!skb)
221                 goto out;
222         prefetchw(skb);
223 
224         /* We do our best to align skb_shared_info on a separate cache
225          * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
226          * aligned memory blocks, unless SLUB/SLAB debug is enabled.
227          * Both skb->head and skb_shared_info are cache line aligned.
228          */
229         size = SKB_DATA_ALIGN(size);
230         size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
231         data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
232         if (!data)
233                 goto nodata;
234         /* kmalloc(size) might give us more room than requested.
235          * Put skb_shared_info exactly at the end of allocated zone,
236          * to allow max possible filling before reallocation.
237          */
238         size = SKB_WITH_OVERHEAD(ksize(data));
239         prefetchw(data + size);
240 
241         /*
242          * Only clear those fields we need to clear, not those that we will
243          * actually initialise below. Hence, don't put any more fields after
244          * the tail pointer in struct sk_buff!
245          */
246         memset(skb, 0, offsetof(struct sk_buff, tail));
247         /* Account for allocated memory : skb + skb->head */
248         skb->truesize = SKB_TRUESIZE(size);
249         skb->pfmemalloc = pfmemalloc;
250         atomic_set(&skb->users, 1);
251         skb->head = data;
252         skb->data = data;
253         skb_reset_tail_pointer(skb);
254         skb->end = skb->tail + size;
255         skb->mac_header = (typeof(skb->mac_header))~0U;
256         skb->transport_header = (typeof(skb->transport_header))~0U;
257 
258         /* make sure we initialize shinfo sequentially */
259         shinfo = skb_shinfo(skb);
260         memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
261         atomic_set(&shinfo->dataref, 1);
262         kmemcheck_annotate_variable(shinfo->destructor_arg);
263 
264         if (flags & SKB_ALLOC_FCLONE) {
265                 struct sk_buff_fclones *fclones;
266 
267                 fclones = container_of(skb, struct sk_buff_fclones, skb1);
268 
269                 kmemcheck_annotate_bitfield(&fclones->skb2, flags1);
270                 skb->fclone = SKB_FCLONE_ORIG;
271                 atomic_set(&fclones->fclone_ref, 1);
272 
273                 fclones->skb2.fclone = SKB_FCLONE_CLONE;
274                 fclones->skb2.pfmemalloc = pfmemalloc;
275         }
276 out:
277         return skb;
278 nodata:
279         kmem_cache_free(cache, skb);
280         skb = NULL;
281         goto out;
282 }
283 EXPORT_SYMBOL(__alloc_skb);
284 
285 /**
286  * __build_skb - build a network buffer
287  * @data: data buffer provided by caller
288  * @frag_size: size of data, or 0 if head was kmalloced
289  *
290  * Allocate a new &sk_buff. Caller provides space holding head and
291  * skb_shared_info. @data must have been allocated by kmalloc() only if
292  * @frag_size is 0, otherwise data should come from the page allocator
293  *  or vmalloc()
294  * The return is the new skb buffer.
295  * On a failure the return is %NULL, and @data is not freed.
296  * Notes :
297  *  Before IO, driver allocates only data buffer where NIC put incoming frame
298  *  Driver should add room at head (NET_SKB_PAD) and
299  *  MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
300  *  After IO, driver calls build_skb(), to allocate sk_buff and populate it
301  *  before giving packet to stack.
302  *  RX rings only contains data buffers, not full skbs.
303  */
304 struct sk_buff *__build_skb(void *data, unsigned int frag_size)
305 {
306         struct skb_shared_info *shinfo;
307         struct sk_buff *skb;
308         unsigned int size = frag_size ? : ksize(data);
309 
310         skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
311         if (!skb)
312                 return NULL;
313 
314         size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
315 
316         memset(skb, 0, offsetof(struct sk_buff, tail));
317         skb->truesize = SKB_TRUESIZE(size);
318         atomic_set(&skb->users, 1);
319         skb->head = data;
320         skb->data = data;
321         skb_reset_tail_pointer(skb);
322         skb->end = skb->tail + size;
323         skb->mac_header = (typeof(skb->mac_header))~0U;
324         skb->transport_header = (typeof(skb->transport_header))~0U;
325 
326         /* make sure we initialize shinfo sequentially */
327         shinfo = skb_shinfo(skb);
328         memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
329         atomic_set(&shinfo->dataref, 1);
330         kmemcheck_annotate_variable(shinfo->destructor_arg);
331 
332         return skb;
333 }
334 
335 /* build_skb() is wrapper over __build_skb(), that specifically
336  * takes care of skb->head and skb->pfmemalloc
337  * This means that if @frag_size is not zero, then @data must be backed
338  * by a page fragment, not kmalloc() or vmalloc()
339  */
340 struct sk_buff *build_skb(void *data, unsigned int frag_size)
341 {
342         struct sk_buff *skb = __build_skb(data, frag_size);
343 
344         if (skb && frag_size) {
345                 skb->head_frag = 1;
346                 if (page_is_pfmemalloc(virt_to_head_page(data)))
347                         skb->pfmemalloc = 1;
348         }
349         return skb;
350 }
351 EXPORT_SYMBOL(build_skb);
352 
353 #define NAPI_SKB_CACHE_SIZE     64
354 
355 struct napi_alloc_cache {
356         struct page_frag_cache page;
357         size_t skb_count;
358         void *skb_cache[NAPI_SKB_CACHE_SIZE];
359 };
360 
361 static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
362 static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
363 
364 static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
365 {
366         struct page_frag_cache *nc;
367         unsigned long flags;
368         void *data;
369 
370         local_irq_save(flags);
371         nc = this_cpu_ptr(&netdev_alloc_cache);
372         data = __alloc_page_frag(nc, fragsz, gfp_mask);
373         local_irq_restore(flags);
374         return data;
375 }
376 
377 /**
378  * netdev_alloc_frag - allocate a page fragment
379  * @fragsz: fragment size
380  *
381  * Allocates a frag from a page for receive buffer.
382  * Uses GFP_ATOMIC allocations.
383  */
384 void *netdev_alloc_frag(unsigned int fragsz)
385 {
386         return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
387 }
388 EXPORT_SYMBOL(netdev_alloc_frag);
389 
390 static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
391 {
392         struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
393 
394         return __alloc_page_frag(&nc->page, fragsz, gfp_mask);
395 }
396 
397 void *napi_alloc_frag(unsigned int fragsz)
398 {
399         return __napi_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
400 }
401 EXPORT_SYMBOL(napi_alloc_frag);
402 
403 /**
404  *      __netdev_alloc_skb - allocate an skbuff for rx on a specific device
405  *      @dev: network device to receive on
406  *      @len: length to allocate
407  *      @gfp_mask: get_free_pages mask, passed to alloc_skb
408  *
409  *      Allocate a new &sk_buff and assign it a usage count of one. The
410  *      buffer has NET_SKB_PAD headroom built in. Users should allocate
411  *      the headroom they think they need without accounting for the
412  *      built in space. The built in space is used for optimisations.
413  *
414  *      %NULL is returned if there is no free memory.
415  */
416 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
417                                    gfp_t gfp_mask)
418 {
419         struct page_frag_cache *nc;
420         unsigned long flags;
421         struct sk_buff *skb;
422         bool pfmemalloc;
423         void *data;
424 
425         len += NET_SKB_PAD;
426 
427         if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
428             (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
429                 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
430                 if (!skb)
431                         goto skb_fail;
432                 goto skb_success;
433         }
434 
435         len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
436         len = SKB_DATA_ALIGN(len);
437 
438         if (sk_memalloc_socks())
439                 gfp_mask |= __GFP_MEMALLOC;
440 
441         local_irq_save(flags);
442 
443         nc = this_cpu_ptr(&netdev_alloc_cache);
444         data = __alloc_page_frag(nc, len, gfp_mask);
445         pfmemalloc = nc->pfmemalloc;
446 
447         local_irq_restore(flags);
448 
449         if (unlikely(!data))
450                 return NULL;
451 
452         skb = __build_skb(data, len);
453         if (unlikely(!skb)) {
454                 skb_free_frag(data);
455                 return NULL;
456         }
457 
458         /* use OR instead of assignment to avoid clearing of bits in mask */
459         if (pfmemalloc)
460                 skb->pfmemalloc = 1;
461         skb->head_frag = 1;
462 
463 skb_success:
464         skb_reserve(skb, NET_SKB_PAD);
465         skb->dev = dev;
466 
467 skb_fail:
468         return skb;
469 }
470 EXPORT_SYMBOL(__netdev_alloc_skb);
471 
472 /**
473  *      __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
474  *      @napi: napi instance this buffer was allocated for
475  *      @len: length to allocate
476  *      @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
477  *
478  *      Allocate a new sk_buff for use in NAPI receive.  This buffer will
479  *      attempt to allocate the head from a special reserved region used
480  *      only for NAPI Rx allocation.  By doing this we can save several
481  *      CPU cycles by avoiding having to disable and re-enable IRQs.
482  *
483  *      %NULL is returned if there is no free memory.
484  */
485 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
486                                  gfp_t gfp_mask)
487 {
488         struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
489         struct sk_buff *skb;
490         void *data;
491 
492         len += NET_SKB_PAD + NET_IP_ALIGN;
493 
494         if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
495             (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
496                 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
497                 if (!skb)
498                         goto skb_fail;
499                 goto skb_success;
500         }
501 
502         len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
503         len = SKB_DATA_ALIGN(len);
504 
505         if (sk_memalloc_socks())
506                 gfp_mask |= __GFP_MEMALLOC;
507 
508         data = __alloc_page_frag(&nc->page, len, gfp_mask);
509         if (unlikely(!data))
510                 return NULL;
511 
512         skb = __build_skb(data, len);
513         if (unlikely(!skb)) {
514                 skb_free_frag(data);
515                 return NULL;
516         }
517 
518         /* use OR instead of assignment to avoid clearing of bits in mask */
519         if (nc->page.pfmemalloc)
520                 skb->pfmemalloc = 1;
521         skb->head_frag = 1;
522 
523 skb_success:
524         skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
525         skb->dev = napi->dev;
526 
527 skb_fail:
528         return skb;
529 }
530 EXPORT_SYMBOL(__napi_alloc_skb);
531 
532 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
533                      int size, unsigned int truesize)
534 {
535         skb_fill_page_desc(skb, i, page, off, size);
536         skb->len += size;
537         skb->data_len += size;
538         skb->truesize += truesize;
539 }
540 EXPORT_SYMBOL(skb_add_rx_frag);
541 
542 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
543                           unsigned int truesize)
544 {
545         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
546 
547         skb_frag_size_add(frag, size);
548         skb->len += size;
549         skb->data_len += size;
550         skb->truesize += truesize;
551 }
552 EXPORT_SYMBOL(skb_coalesce_rx_frag);
553 
554 static void skb_drop_list(struct sk_buff **listp)
555 {
556         kfree_skb_list(*listp);
557         *listp = NULL;
558 }
559 
560 static inline void skb_drop_fraglist(struct sk_buff *skb)
561 {
562         skb_drop_list(&skb_shinfo(skb)->frag_list);
563 }
564 
565 static void skb_clone_fraglist(struct sk_buff *skb)
566 {
567         struct sk_buff *list;
568 
569         skb_walk_frags(skb, list)
570                 skb_get(list);
571 }
572 
573 static void skb_free_head(struct sk_buff *skb)
574 {
575         unsigned char *head = skb->head;
576 
577         if (skb->head_frag)
578                 skb_free_frag(head);
579         else
580                 kfree(head);
581 }
582 
583 static void skb_release_data(struct sk_buff *skb)
584 {
585         struct skb_shared_info *shinfo = skb_shinfo(skb);
586         int i;
587 
588         if (skb->cloned &&
589             atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
590                               &shinfo->dataref))
591                 return;
592 
593         for (i = 0; i < shinfo->nr_frags; i++)
594                 __skb_frag_unref(&shinfo->frags[i]);
595 
596         /*
597          * If skb buf is from userspace, we need to notify the caller
598          * the lower device DMA has done;
599          */
600         if (shinfo->tx_flags & SKBTX_DEV_ZEROCOPY) {
601                 struct ubuf_info *uarg;
602 
603                 uarg = shinfo->destructor_arg;
604                 if (uarg->callback)
605                         uarg->callback(uarg, true);
606         }
607 
608         if (shinfo->frag_list)
609                 kfree_skb_list(shinfo->frag_list);
610 
611         skb_free_head(skb);
612 }
613 
614 /*
615  *      Free an skbuff by memory without cleaning the state.
616  */
617 static void kfree_skbmem(struct sk_buff *skb)
618 {
619         struct sk_buff_fclones *fclones;
620 
621         switch (skb->fclone) {
622         case SKB_FCLONE_UNAVAILABLE:
623                 kmem_cache_free(skbuff_head_cache, skb);
624                 return;
625 
626         case SKB_FCLONE_ORIG:
627                 fclones = container_of(skb, struct sk_buff_fclones, skb1);
628 
629                 /* We usually free the clone (TX completion) before original skb
630                  * This test would have no chance to be true for the clone,
631                  * while here, branch prediction will be good.
632                  */
633                 if (atomic_read(&fclones->fclone_ref) == 1)
634                         goto fastpath;
635                 break;
636 
637         default: /* SKB_FCLONE_CLONE */
638                 fclones = container_of(skb, struct sk_buff_fclones, skb2);
639                 break;
640         }
641         if (!atomic_dec_and_test(&fclones->fclone_ref))
642                 return;
643 fastpath:
644         kmem_cache_free(skbuff_fclone_cache, fclones);
645 }
646 
647 static void skb_release_head_state(struct sk_buff *skb)
648 {
649         skb_dst_drop(skb);
650 #ifdef CONFIG_XFRM
651         secpath_put(skb->sp);
652 #endif
653         if (skb->destructor) {
654                 WARN_ON(in_irq());
655                 skb->destructor(skb);
656         }
657 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
658         nf_conntrack_put(skb->nfct);
659 #endif
660 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
661         nf_bridge_put(skb->nf_bridge);
662 #endif
663 }
664 
665 /* Free everything but the sk_buff shell. */
666 static void skb_release_all(struct sk_buff *skb)
667 {
668         skb_release_head_state(skb);
669         if (likely(skb->head))
670                 skb_release_data(skb);
671 }
672 
673 /**
674  *      __kfree_skb - private function
675  *      @skb: buffer
676  *
677  *      Free an sk_buff. Release anything attached to the buffer.
678  *      Clean the state. This is an internal helper function. Users should
679  *      always call kfree_skb
680  */
681 
682 void __kfree_skb(struct sk_buff *skb)
683 {
684         skb_release_all(skb);
685         kfree_skbmem(skb);
686 }
687 EXPORT_SYMBOL(__kfree_skb);
688 
689 /**
690  *      kfree_skb - free an sk_buff
691  *      @skb: buffer to free
692  *
693  *      Drop a reference to the buffer and free it if the usage count has
694  *      hit zero.
695  */
696 void kfree_skb(struct sk_buff *skb)
697 {
698         if (unlikely(!skb))
699                 return;
700         if (likely(atomic_read(&skb->users) == 1))
701                 smp_rmb();
702         else if (likely(!atomic_dec_and_test(&skb->users)))
703                 return;
704         trace_kfree_skb(skb, __builtin_return_address(0));
705         __kfree_skb(skb);
706 }
707 EXPORT_SYMBOL(kfree_skb);
708 
709 void kfree_skb_list(struct sk_buff *segs)
710 {
711         while (segs) {
712                 struct sk_buff *next = segs->next;
713 
714                 kfree_skb(segs);
715                 segs = next;
716         }
717 }
718 EXPORT_SYMBOL(kfree_skb_list);
719 
720 /**
721  *      skb_tx_error - report an sk_buff xmit error
722  *      @skb: buffer that triggered an error
723  *
724  *      Report xmit error if a device callback is tracking this skb.
725  *      skb must be freed afterwards.
726  */
727 void skb_tx_error(struct sk_buff *skb)
728 {
729         if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
730                 struct ubuf_info *uarg;
731 
732                 uarg = skb_shinfo(skb)->destructor_arg;
733                 if (uarg->callback)
734                         uarg->callback(uarg, false);
735                 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
736         }
737 }
738 EXPORT_SYMBOL(skb_tx_error);
739 
740 /**
741  *      consume_skb - free an skbuff
742  *      @skb: buffer to free
743  *
744  *      Drop a ref to the buffer and free it if the usage count has hit zero
745  *      Functions identically to kfree_skb, but kfree_skb assumes that the frame
746  *      is being dropped after a failure and notes that
747  */
748 void consume_skb(struct sk_buff *skb)
749 {
750         if (unlikely(!skb))
751                 return;
752         if (likely(atomic_read(&skb->users) == 1))
753                 smp_rmb();
754         else if (likely(!atomic_dec_and_test(&skb->users)))
755                 return;
756         trace_consume_skb(skb);
757         __kfree_skb(skb);
758 }
759 EXPORT_SYMBOL(consume_skb);
760 
761 void __kfree_skb_flush(void)
762 {
763         struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
764 
765         /* flush skb_cache if containing objects */
766         if (nc->skb_count) {
767                 kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count,
768                                      nc->skb_cache);
769                 nc->skb_count = 0;
770         }
771 }
772 
773 static inline void _kfree_skb_defer(struct sk_buff *skb)
774 {
775         struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
776 
777         /* drop skb->head and call any destructors for packet */
778         skb_release_all(skb);
779 
780         /* record skb to CPU local list */
781         nc->skb_cache[nc->skb_count++] = skb;
782 
783 #ifdef CONFIG_SLUB
784         /* SLUB writes into objects when freeing */
785         prefetchw(skb);
786 #endif
787 
788         /* flush skb_cache if it is filled */
789         if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
790                 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE,
791                                      nc->skb_cache);
792                 nc->skb_count = 0;
793         }
794 }
795 void __kfree_skb_defer(struct sk_buff *skb)
796 {
797         _kfree_skb_defer(skb);
798 }
799 
800 void napi_consume_skb(struct sk_buff *skb, int budget)
801 {
802         if (unlikely(!skb))
803                 return;
804 
805         /* Zero budget indicate non-NAPI context called us, like netpoll */
806         if (unlikely(!budget)) {
807                 dev_consume_skb_any(skb);
808                 return;
809         }
810 
811         if (likely(atomic_read(&skb->users) == 1))
812                 smp_rmb();
813         else if (likely(!atomic_dec_and_test(&skb->users)))
814                 return;
815         /* if reaching here SKB is ready to free */
816         trace_consume_skb(skb);
817 
818         /* if SKB is a clone, don't handle this case */
819         if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
820                 __kfree_skb(skb);
821                 return;
822         }
823 
824         _kfree_skb_defer(skb);
825 }
826 EXPORT_SYMBOL(napi_consume_skb);
827 
828 /* Make sure a field is enclosed inside headers_start/headers_end section */
829 #define CHECK_SKB_FIELD(field) \
830         BUILD_BUG_ON(offsetof(struct sk_buff, field) <          \
831                      offsetof(struct sk_buff, headers_start));  \
832         BUILD_BUG_ON(offsetof(struct sk_buff, field) >          \
833                      offsetof(struct sk_buff, headers_end));    \
834 
835 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
836 {
837         new->tstamp             = old->tstamp;
838         /* We do not copy old->sk */
839         new->dev                = old->dev;
840         memcpy(new->cb, old->cb, sizeof(old->cb));
841         skb_dst_copy(new, old);
842 #ifdef CONFIG_XFRM
843         new->sp                 = secpath_get(old->sp);
844 #endif
845         __nf_copy(new, old, false);
846 
847         /* Note : this field could be in headers_start/headers_end section
848          * It is not yet because we do not want to have a 16 bit hole
849          */
850         new->queue_mapping = old->queue_mapping;
851 
852         memcpy(&new->headers_start, &old->headers_start,
853                offsetof(struct sk_buff, headers_end) -
854                offsetof(struct sk_buff, headers_start));
855         CHECK_SKB_FIELD(protocol);
856         CHECK_SKB_FIELD(csum);
857         CHECK_SKB_FIELD(hash);
858         CHECK_SKB_FIELD(priority);
859         CHECK_SKB_FIELD(skb_iif);
860         CHECK_SKB_FIELD(vlan_proto);
861         CHECK_SKB_FIELD(vlan_tci);
862         CHECK_SKB_FIELD(transport_header);
863         CHECK_SKB_FIELD(network_header);
864         CHECK_SKB_FIELD(mac_header);
865         CHECK_SKB_FIELD(inner_protocol);
866         CHECK_SKB_FIELD(inner_transport_header);
867         CHECK_SKB_FIELD(inner_network_header);
868         CHECK_SKB_FIELD(inner_mac_header);
869         CHECK_SKB_FIELD(mark);
870 #ifdef CONFIG_NETWORK_SECMARK
871         CHECK_SKB_FIELD(secmark);
872 #endif
873 #ifdef CONFIG_NET_RX_BUSY_POLL
874         CHECK_SKB_FIELD(napi_id);
875 #endif
876 #ifdef CONFIG_XPS
877         CHECK_SKB_FIELD(sender_cpu);
878 #endif
879 #ifdef CONFIG_NET_SCHED
880         CHECK_SKB_FIELD(tc_index);
881 #ifdef CONFIG_NET_CLS_ACT
882         CHECK_SKB_FIELD(tc_verd);
883 #endif
884 #endif
885 
886 }
887 
888 /*
889  * You should not add any new code to this function.  Add it to
890  * __copy_skb_header above instead.
891  */
892 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
893 {
894 #define C(x) n->x = skb->x
895 
896         n->next = n->prev = NULL;
897         n->sk = NULL;
898         __copy_skb_header(n, skb);
899 
900         C(len);
901         C(data_len);
902         C(mac_len);
903         n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
904         n->cloned = 1;
905         n->nohdr = 0;
906         n->destructor = NULL;
907         C(tail);
908         C(end);
909         C(head);
910         C(head_frag);
911         C(data);
912         C(truesize);
913         atomic_set(&n->users, 1);
914 
915         atomic_inc(&(skb_shinfo(skb)->dataref));
916         skb->cloned = 1;
917 
918         return n;
919 #undef C
920 }
921 
922 /**
923  *      skb_morph       -       morph one skb into another
924  *      @dst: the skb to receive the contents
925  *      @src: the skb to supply the contents
926  *
927  *      This is identical to skb_clone except that the target skb is
928  *      supplied by the user.
929  *
930  *      The target skb is returned upon exit.
931  */
932 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
933 {
934         skb_release_all(dst);
935         return __skb_clone(dst, src);
936 }
937 EXPORT_SYMBOL_GPL(skb_morph);
938 
939 /**
940  *      skb_copy_ubufs  -       copy userspace skb frags buffers to kernel
941  *      @skb: the skb to modify
942  *      @gfp_mask: allocation priority
943  *
944  *      This must be called on SKBTX_DEV_ZEROCOPY skb.
945  *      It will copy all frags into kernel and drop the reference
946  *      to userspace pages.
947  *
948  *      If this function is called from an interrupt gfp_mask() must be
949  *      %GFP_ATOMIC.
950  *
951  *      Returns 0 on success or a negative error code on failure
952  *      to allocate kernel memory to copy to.
953  */
954 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
955 {
956         int i;
957         int num_frags = skb_shinfo(skb)->nr_frags;
958         struct page *page, *head = NULL;
959         struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
960 
961         for (i = 0; i < num_frags; i++) {
962                 u8 *vaddr;
963                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
964 
965                 page = alloc_page(gfp_mask);
966                 if (!page) {
967                         while (head) {
968                                 struct page *next = (struct page *)page_private(head);
969                                 put_page(head);
970                                 head = next;
971                         }
972                         return -ENOMEM;
973                 }
974                 vaddr = kmap_atomic(skb_frag_page(f));
975                 memcpy(page_address(page),
976                        vaddr + f->page_offset, skb_frag_size(f));
977                 kunmap_atomic(vaddr);
978                 set_page_private(page, (unsigned long)head);
979                 head = page;
980         }
981 
982         /* skb frags release userspace buffers */
983         for (i = 0; i < num_frags; i++)
984                 skb_frag_unref(skb, i);
985 
986         uarg->callback(uarg, false);
987 
988         /* skb frags point to kernel buffers */
989         for (i = num_frags - 1; i >= 0; i--) {
990                 __skb_fill_page_desc(skb, i, head, 0,
991                                      skb_shinfo(skb)->frags[i].size);
992                 head = (struct page *)page_private(head);
993         }
994 
995         skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
996         return 0;
997 }
998 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
999 
1000 /**
1001  *      skb_clone       -       duplicate an sk_buff
1002  *      @skb: buffer to clone
1003  *      @gfp_mask: allocation priority
1004  *
1005  *      Duplicate an &sk_buff. The new one is not owned by a socket. Both
1006  *      copies share the same packet data but not structure. The new
1007  *      buffer has a reference count of 1. If the allocation fails the
1008  *      function returns %NULL otherwise the new buffer is returned.
1009  *
1010  *      If this function is called from an interrupt gfp_mask() must be
1011  *      %GFP_ATOMIC.
1012  */
1013 
1014 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1015 {
1016         struct sk_buff_fclones *fclones = container_of(skb,
1017                                                        struct sk_buff_fclones,
1018                                                        skb1);
1019         struct sk_buff *n;
1020 
1021         if (skb_orphan_frags(skb, gfp_mask))
1022                 return NULL;
1023 
1024         if (skb->fclone == SKB_FCLONE_ORIG &&
1025             atomic_read(&fclones->fclone_ref) == 1) {
1026                 n = &fclones->skb2;
1027                 atomic_set(&fclones->fclone_ref, 2);
1028         } else {
1029                 if (skb_pfmemalloc(skb))
1030                         gfp_mask |= __GFP_MEMALLOC;
1031 
1032                 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1033                 if (!n)
1034                         return NULL;
1035 
1036                 kmemcheck_annotate_bitfield(n, flags1);
1037                 n->fclone = SKB_FCLONE_UNAVAILABLE;
1038         }
1039 
1040         return __skb_clone(n, skb);
1041 }
1042 EXPORT_SYMBOL(skb_clone);
1043 
1044 static void skb_headers_offset_update(struct sk_buff *skb, int off)
1045 {
1046         /* Only adjust this if it actually is csum_start rather than csum */
1047         if (skb->ip_summed == CHECKSUM_PARTIAL)
1048                 skb->csum_start += off;
1049         /* {transport,network,mac}_header and tail are relative to skb->head */
1050         skb->transport_header += off;
1051         skb->network_header   += off;
1052         if (skb_mac_header_was_set(skb))
1053                 skb->mac_header += off;
1054         skb->inner_transport_header += off;
1055         skb->inner_network_header += off;
1056         skb->inner_mac_header += off;
1057 }
1058 
1059 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
1060 {
1061         __copy_skb_header(new, old);
1062 
1063         skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1064         skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1065         skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1066 }
1067 
1068 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1069 {
1070         if (skb_pfmemalloc(skb))
1071                 return SKB_ALLOC_RX;
1072         return 0;
1073 }
1074 
1075 /**
1076  *      skb_copy        -       create private copy of an sk_buff
1077  *      @skb: buffer to copy
1078  *      @gfp_mask: allocation priority
1079  *
1080  *      Make a copy of both an &sk_buff and its data. This is used when the
1081  *      caller wishes to modify the data and needs a private copy of the
1082  *      data to alter. Returns %NULL on failure or the pointer to the buffer
1083  *      on success. The returned buffer has a reference count of 1.
1084  *
1085  *      As by-product this function converts non-linear &sk_buff to linear
1086  *      one, so that &sk_buff becomes completely private and caller is allowed
1087  *      to modify all the data of returned buffer. This means that this
1088  *      function is not recommended for use in circumstances when only
1089  *      header is going to be modified. Use pskb_copy() instead.
1090  */
1091 
1092 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1093 {
1094         int headerlen = skb_headroom(skb);
1095         unsigned int size = skb_end_offset(skb) + skb->data_len;
1096         struct sk_buff *n = __alloc_skb(size, gfp_mask,
1097                                         skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1098 
1099         if (!n)
1100                 return NULL;
1101 
1102         /* Set the data pointer */
1103         skb_reserve(n, headerlen);
1104         /* Set the tail pointer and length */
1105         skb_put(n, skb->len);
1106 
1107         if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
1108                 BUG();
1109 
1110         copy_skb_header(n, skb);
1111         return n;
1112 }
1113 EXPORT_SYMBOL(skb_copy);
1114 
1115 /**
1116  *      __pskb_copy_fclone      -  create copy of an sk_buff with private head.
1117  *      @skb: buffer to copy
1118  *      @headroom: headroom of new skb
1119  *      @gfp_mask: allocation priority
1120  *      @fclone: if true allocate the copy of the skb from the fclone
1121  *      cache instead of the head cache; it is recommended to set this
1122  *      to true for the cases where the copy will likely be cloned
1123  *
1124  *      Make a copy of both an &sk_buff and part of its data, located
1125  *      in header. Fragmented data remain shared. This is used when
1126  *      the caller wishes to modify only header of &sk_buff and needs
1127  *      private copy of the header to alter. Returns %NULL on failure
1128  *      or the pointer to the buffer on success.
1129  *      The returned buffer has a reference count of 1.
1130  */
1131 
1132 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1133                                    gfp_t gfp_mask, bool fclone)
1134 {
1135         unsigned int size = skb_headlen(skb) + headroom;
1136         int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1137         struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1138 
1139         if (!n)
1140                 goto out;
1141 
1142         /* Set the data pointer */
1143         skb_reserve(n, headroom);
1144         /* Set the tail pointer and length */
1145         skb_put(n, skb_headlen(skb));
1146         /* Copy the bytes */
1147         skb_copy_from_linear_data(skb, n->data, n->len);
1148 
1149         n->truesize += skb->data_len;
1150         n->data_len  = skb->data_len;
1151         n->len       = skb->len;
1152 
1153         if (skb_shinfo(skb)->nr_frags) {
1154                 int i;
1155 
1156                 if (skb_orphan_frags(skb, gfp_mask)) {
1157                         kfree_skb(n);
1158                         n = NULL;
1159                         goto out;
1160                 }
1161                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1162                         skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1163                         skb_frag_ref(skb, i);
1164                 }
1165                 skb_shinfo(n)->nr_frags = i;
1166         }
1167 
1168         if (skb_has_frag_list(skb)) {
1169                 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1170                 skb_clone_fraglist(n);
1171         }
1172 
1173         copy_skb_header(n, skb);
1174 out:
1175         return n;
1176 }
1177 EXPORT_SYMBOL(__pskb_copy_fclone);
1178 
1179 /**
1180  *      pskb_expand_head - reallocate header of &sk_buff
1181  *      @skb: buffer to reallocate
1182  *      @nhead: room to add at head
1183  *      @ntail: room to add at tail
1184  *      @gfp_mask: allocation priority
1185  *
1186  *      Expands (or creates identical copy, if @nhead and @ntail are zero)
1187  *      header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1188  *      reference count of 1. Returns zero in the case of success or error,
1189  *      if expansion failed. In the last case, &sk_buff is not changed.
1190  *
1191  *      All the pointers pointing into skb header may change and must be
1192  *      reloaded after call to this function.
1193  */
1194 
1195 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1196                      gfp_t gfp_mask)
1197 {
1198         int i;
1199         u8 *data;
1200         int size = nhead + skb_end_offset(skb) + ntail;
1201         long off;
1202 
1203         BUG_ON(nhead < 0);
1204 
1205         if (skb_shared(skb))
1206                 BUG();
1207 
1208         size = SKB_DATA_ALIGN(size);
1209 
1210         if (skb_pfmemalloc(skb))
1211                 gfp_mask |= __GFP_MEMALLOC;
1212         data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1213                                gfp_mask, NUMA_NO_NODE, NULL);
1214         if (!data)
1215                 goto nodata;
1216         size = SKB_WITH_OVERHEAD(ksize(data));
1217 
1218         /* Copy only real data... and, alas, header. This should be
1219          * optimized for the cases when header is void.
1220          */
1221         memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1222 
1223         memcpy((struct skb_shared_info *)(data + size),
1224                skb_shinfo(skb),
1225                offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1226 
1227         /*
1228          * if shinfo is shared we must drop the old head gracefully, but if it
1229          * is not we can just drop the old head and let the existing refcount
1230          * be since all we did is relocate the values
1231          */
1232         if (skb_cloned(skb)) {
1233                 /* copy this zero copy skb frags */
1234                 if (skb_orphan_frags(skb, gfp_mask))
1235                         goto nofrags;
1236                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1237                         skb_frag_ref(skb, i);
1238 
1239                 if (skb_has_frag_list(skb))
1240                         skb_clone_fraglist(skb);
1241 
1242                 skb_release_data(skb);
1243         } else {
1244                 skb_free_head(skb);
1245         }
1246         off = (data + nhead) - skb->head;
1247 
1248         skb->head     = data;
1249         skb->head_frag = 0;
1250         skb->data    += off;
1251 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1252         skb->end      = size;
1253         off           = nhead;
1254 #else
1255         skb->end      = skb->head + size;
1256 #endif
1257         skb->tail             += off;
1258         skb_headers_offset_update(skb, nhead);
1259         skb->cloned   = 0;
1260         skb->hdr_len  = 0;
1261         skb->nohdr    = 0;
1262         atomic_set(&skb_shinfo(skb)->dataref, 1);
1263         return 0;
1264 
1265 nofrags:
1266         kfree(data);
1267 nodata:
1268         return -ENOMEM;
1269 }
1270 EXPORT_SYMBOL(pskb_expand_head);
1271 
1272 /* Make private copy of skb with writable head and some headroom */
1273 
1274 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1275 {
1276         struct sk_buff *skb2;
1277         int delta = headroom - skb_headroom(skb);
1278 
1279         if (delta <= 0)
1280                 skb2 = pskb_copy(skb, GFP_ATOMIC);
1281         else {
1282                 skb2 = skb_clone(skb, GFP_ATOMIC);
1283                 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1284                                              GFP_ATOMIC)) {
1285                         kfree_skb(skb2);
1286                         skb2 = NULL;
1287                 }
1288         }
1289         return skb2;
1290 }
1291 EXPORT_SYMBOL(skb_realloc_headroom);
1292 
1293 /**
1294  *      skb_copy_expand -       copy and expand sk_buff
1295  *      @skb: buffer to copy
1296  *      @newheadroom: new free bytes at head
1297  *      @newtailroom: new free bytes at tail
1298  *      @gfp_mask: allocation priority
1299  *
1300  *      Make a copy of both an &sk_buff and its data and while doing so
1301  *      allocate additional space.
1302  *
1303  *      This is used when the caller wishes to modify the data and needs a
1304  *      private copy of the data to alter as well as more space for new fields.
1305  *      Returns %NULL on failure or the pointer to the buffer
1306  *      on success. The returned buffer has a reference count of 1.
1307  *
1308  *      You must pass %GFP_ATOMIC as the allocation priority if this function
1309  *      is called from an interrupt.
1310  */
1311 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1312                                 int newheadroom, int newtailroom,
1313                                 gfp_t gfp_mask)
1314 {
1315         /*
1316          *      Allocate the copy buffer
1317          */
1318         struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1319                                         gfp_mask, skb_alloc_rx_flag(skb),
1320                                         NUMA_NO_NODE);
1321         int oldheadroom = skb_headroom(skb);
1322         int head_copy_len, head_copy_off;
1323 
1324         if (!n)
1325                 return NULL;
1326 
1327         skb_reserve(n, newheadroom);
1328 
1329         /* Set the tail pointer and length */
1330         skb_put(n, skb->len);
1331 
1332         head_copy_len = oldheadroom;
1333         head_copy_off = 0;
1334         if (newheadroom <= head_copy_len)
1335                 head_copy_len = newheadroom;
1336         else
1337                 head_copy_off = newheadroom - head_copy_len;
1338 
1339         /* Copy the linear header and data. */
1340         if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1341                           skb->len + head_copy_len))
1342                 BUG();
1343 
1344         copy_skb_header(n, skb);
1345 
1346         skb_headers_offset_update(n, newheadroom - oldheadroom);
1347 
1348         return n;
1349 }
1350 EXPORT_SYMBOL(skb_copy_expand);
1351 
1352 /**
1353  *      skb_pad                 -       zero pad the tail of an skb
1354  *      @skb: buffer to pad
1355  *      @pad: space to pad
1356  *
1357  *      Ensure that a buffer is followed by a padding area that is zero
1358  *      filled. Used by network drivers which may DMA or transfer data
1359  *      beyond the buffer end onto the wire.
1360  *
1361  *      May return error in out of memory cases. The skb is freed on error.
1362  */
1363 
1364 int skb_pad(struct sk_buff *skb, int pad)
1365 {
1366         int err;
1367         int ntail;
1368 
1369         /* If the skbuff is non linear tailroom is always zero.. */
1370         if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1371                 memset(skb->data+skb->len, 0, pad);
1372                 return 0;
1373         }
1374 
1375         ntail = skb->data_len + pad - (skb->end - skb->tail);
1376         if (likely(skb_cloned(skb) || ntail > 0)) {
1377                 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1378                 if (unlikely(err))
1379                         goto free_skb;
1380         }
1381 
1382         /* FIXME: The use of this function with non-linear skb's really needs
1383          * to be audited.
1384          */
1385         err = skb_linearize(skb);
1386         if (unlikely(err))
1387                 goto free_skb;
1388 
1389         memset(skb->data + skb->len, 0, pad);
1390         return 0;
1391 
1392 free_skb:
1393         kfree_skb(skb);
1394         return err;
1395 }
1396 EXPORT_SYMBOL(skb_pad);
1397 
1398 /**
1399  *      pskb_put - add data to the tail of a potentially fragmented buffer
1400  *      @skb: start of the buffer to use
1401  *      @tail: tail fragment of the buffer to use
1402  *      @len: amount of data to add
1403  *
1404  *      This function extends the used data area of the potentially
1405  *      fragmented buffer. @tail must be the last fragment of @skb -- or
1406  *      @skb itself. If this would exceed the total buffer size the kernel
1407  *      will panic. A pointer to the first byte of the extra data is
1408  *      returned.
1409  */
1410 
1411 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1412 {
1413         if (tail != skb) {
1414                 skb->data_len += len;
1415                 skb->len += len;
1416         }
1417         return skb_put(tail, len);
1418 }
1419 EXPORT_SYMBOL_GPL(pskb_put);
1420 
1421 /**
1422  *      skb_put - add data to a buffer
1423  *      @skb: buffer to use
1424  *      @len: amount of data to add
1425  *
1426  *      This function extends the used data area of the buffer. If this would
1427  *      exceed the total buffer size the kernel will panic. A pointer to the
1428  *      first byte of the extra data is returned.
1429  */
1430 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1431 {
1432         unsigned char *tmp = skb_tail_pointer(skb);
1433         SKB_LINEAR_ASSERT(skb);
1434         skb->tail += len;
1435         skb->len  += len;
1436         if (unlikely(skb->tail > skb->end))
1437                 skb_over_panic(skb, len, __builtin_return_address(0));
1438         return tmp;
1439 }
1440 EXPORT_SYMBOL(skb_put);
1441 
1442 /**
1443  *      skb_push - add data to the start of a buffer
1444  *      @skb: buffer to use
1445  *      @len: amount of data to add
1446  *
1447  *      This function extends the used data area of the buffer at the buffer
1448  *      start. If this would exceed the total buffer headroom the kernel will
1449  *      panic. A pointer to the first byte of the extra data is returned.
1450  */
1451 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1452 {
1453         skb->data -= len;
1454         skb->len  += len;
1455         if (unlikely(skb->data<skb->head))
1456                 skb_under_panic(skb, len, __builtin_return_address(0));
1457         return skb->data;
1458 }
1459 EXPORT_SYMBOL(skb_push);
1460 
1461 /**
1462  *      skb_pull - remove data from the start of a buffer
1463  *      @skb: buffer to use
1464  *      @len: amount of data to remove
1465  *
1466  *      This function removes data from the start of a buffer, returning
1467  *      the memory to the headroom. A pointer to the next data in the buffer
1468  *      is returned. Once the data has been pulled future pushes will overwrite
1469  *      the old data.
1470  */
1471 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1472 {
1473         return skb_pull_inline(skb, len);
1474 }
1475 EXPORT_SYMBOL(skb_pull);
1476 
1477 /**
1478  *      skb_trim - remove end from a buffer
1479  *      @skb: buffer to alter
1480  *      @len: new length
1481  *
1482  *      Cut the length of a buffer down by removing data from the tail. If
1483  *      the buffer is already under the length specified it is not modified.
1484  *      The skb must be linear.
1485  */
1486 void skb_trim(struct sk_buff *skb, unsigned int len)
1487 {
1488         if (skb->len > len)
1489                 __skb_trim(skb, len);
1490 }
1491 EXPORT_SYMBOL(skb_trim);
1492 
1493 /* Trims skb to length len. It can change skb pointers.
1494  */
1495 
1496 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1497 {
1498         struct sk_buff **fragp;
1499         struct sk_buff *frag;
1500         int offset = skb_headlen(skb);
1501         int nfrags = skb_shinfo(skb)->nr_frags;
1502         int i;
1503         int err;
1504 
1505         if (skb_cloned(skb) &&
1506             unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1507                 return err;
1508 
1509         i = 0;
1510         if (offset >= len)
1511                 goto drop_pages;
1512 
1513         for (; i < nfrags; i++) {
1514                 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1515 
1516                 if (end < len) {
1517                         offset = end;
1518                         continue;
1519                 }
1520 
1521                 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1522 
1523 drop_pages:
1524                 skb_shinfo(skb)->nr_frags = i;
1525 
1526                 for (; i < nfrags; i++)
1527                         skb_frag_unref(skb, i);
1528 
1529                 if (skb_has_frag_list(skb))
1530                         skb_drop_fraglist(skb);
1531                 goto done;
1532         }
1533 
1534         for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1535              fragp = &frag->next) {
1536                 int end = offset + frag->len;
1537 
1538                 if (skb_shared(frag)) {
1539                         struct sk_buff *nfrag;
1540 
1541                         nfrag = skb_clone(frag, GFP_ATOMIC);
1542                         if (unlikely(!nfrag))
1543                                 return -ENOMEM;
1544 
1545                         nfrag->next = frag->next;
1546                         consume_skb(frag);
1547                         frag = nfrag;
1548                         *fragp = frag;
1549                 }
1550 
1551                 if (end < len) {
1552                         offset = end;
1553                         continue;
1554                 }
1555 
1556                 if (end > len &&
1557                     unlikely((err = pskb_trim(frag, len - offset))))
1558                         return err;
1559 
1560                 if (frag->next)
1561                         skb_drop_list(&frag->next);
1562                 break;
1563         }
1564 
1565 done:
1566         if (len > skb_headlen(skb)) {
1567                 skb->data_len -= skb->len - len;
1568                 skb->len       = len;
1569         } else {
1570                 skb->len       = len;
1571                 skb->data_len  = 0;
1572                 skb_set_tail_pointer(skb, len);
1573         }
1574 
1575         return 0;
1576 }
1577 EXPORT_SYMBOL(___pskb_trim);
1578 
1579 /**
1580  *      __pskb_pull_tail - advance tail of skb header
1581  *      @skb: buffer to reallocate
1582  *      @delta: number of bytes to advance tail
1583  *
1584  *      The function makes a sense only on a fragmented &sk_buff,
1585  *      it expands header moving its tail forward and copying necessary
1586  *      data from fragmented part.
1587  *
1588  *      &sk_buff MUST have reference count of 1.
1589  *
1590  *      Returns %NULL (and &sk_buff does not change) if pull failed
1591  *      or value of new tail of skb in the case of success.
1592  *
1593  *      All the pointers pointing into skb header may change and must be
1594  *      reloaded after call to this function.
1595  */
1596 
1597 /* Moves tail of skb head forward, copying data from fragmented part,
1598  * when it is necessary.
1599  * 1. It may fail due to malloc failure.
1600  * 2. It may change skb pointers.
1601  *
1602  * It is pretty complicated. Luckily, it is called only in exceptional cases.
1603  */
1604 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1605 {
1606         /* If skb has not enough free space at tail, get new one
1607          * plus 128 bytes for future expansions. If we have enough
1608          * room at tail, reallocate without expansion only if skb is cloned.
1609          */
1610         int i, k, eat = (skb->tail + delta) - skb->end;
1611 
1612         if (eat > 0 || skb_cloned(skb)) {
1613                 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1614                                      GFP_ATOMIC))
1615                         return NULL;
1616         }
1617 
1618         if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1619                 BUG();
1620 
1621         /* Optimization: no fragments, no reasons to preestimate
1622          * size of pulled pages. Superb.
1623          */
1624         if (!skb_has_frag_list(skb))
1625                 goto pull_pages;
1626 
1627         /* Estimate size of pulled pages. */
1628         eat = delta;
1629         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1630                 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1631 
1632                 if (size >= eat)
1633                         goto pull_pages;
1634                 eat -= size;
1635         }
1636 
1637         /* If we need update frag list, we are in troubles.
1638          * Certainly, it possible to add an offset to skb data,
1639          * but taking into account that pulling is expected to
1640          * be very rare operation, it is worth to fight against
1641          * further bloating skb head and crucify ourselves here instead.
1642          * Pure masohism, indeed. 8)8)
1643          */
1644         if (eat) {
1645                 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1646                 struct sk_buff *clone = NULL;
1647                 struct sk_buff *insp = NULL;
1648 
1649                 do {
1650                         BUG_ON(!list);
1651 
1652                         if (list->len <= eat) {
1653                                 /* Eaten as whole. */
1654                                 eat -= list->len;
1655                                 list = list->next;
1656                                 insp = list;
1657                         } else {
1658                                 /* Eaten partially. */
1659 
1660                                 if (skb_shared(list)) {
1661                                         /* Sucks! We need to fork list. :-( */
1662                                         clone = skb_clone(list, GFP_ATOMIC);
1663                                         if (!clone)
1664                                                 return NULL;
1665                                         insp = list->next;
1666                                         list = clone;
1667                                 } else {
1668                                         /* This may be pulled without
1669                                          * problems. */
1670                                         insp = list;
1671                                 }
1672                                 if (!pskb_pull(list, eat)) {
1673                                         kfree_skb(clone);
1674                                         return NULL;
1675                                 }
1676                                 break;
1677                         }
1678                 } while (eat);
1679 
1680                 /* Free pulled out fragments. */
1681                 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1682                         skb_shinfo(skb)->frag_list = list->next;
1683                         kfree_skb(list);
1684                 }
1685                 /* And insert new clone at head. */
1686                 if (clone) {
1687                         clone->next = list;
1688                         skb_shinfo(skb)->frag_list = clone;
1689                 }
1690         }
1691         /* Success! Now we may commit changes to skb data. */
1692 
1693 pull_pages:
1694         eat = delta;
1695         k = 0;
1696         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1697                 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1698 
1699                 if (size <= eat) {
1700                         skb_frag_unref(skb, i);
1701                         eat -= size;
1702                 } else {
1703                         skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1704                         if (eat) {
1705                                 skb_shinfo(skb)->frags[k].page_offset += eat;
1706                                 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1707                                 eat = 0;
1708                         }
1709                         k++;
1710                 }
1711         }
1712         skb_shinfo(skb)->nr_frags = k;
1713 
1714         skb->tail     += delta;
1715         skb->data_len -= delta;
1716 
1717         return skb_tail_pointer(skb);
1718 }
1719 EXPORT_SYMBOL(__pskb_pull_tail);
1720 
1721 /**
1722  *      skb_copy_bits - copy bits from skb to kernel buffer
1723  *      @skb: source skb
1724  *      @offset: offset in source
1725  *      @to: destination buffer
1726  *      @len: number of bytes to copy
1727  *
1728  *      Copy the specified number of bytes from the source skb to the
1729  *      destination buffer.
1730  *
1731  *      CAUTION ! :
1732  *              If its prototype is ever changed,
1733  *              check arch/{*}/net/{*}.S files,
1734  *              since it is called from BPF assembly code.
1735  */
1736 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1737 {
1738         int start = skb_headlen(skb);
1739         struct sk_buff *frag_iter;
1740         int i, copy;
1741 
1742         if (offset > (int)skb->len - len)
1743                 goto fault;
1744 
1745         /* Copy header. */
1746         if ((copy = start - offset) > 0) {
1747                 if (copy > len)
1748                         copy = len;
1749                 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1750                 if ((len -= copy) == 0)
1751                         return 0;
1752                 offset += copy;
1753                 to     += copy;
1754         }
1755 
1756         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1757                 int end;
1758                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1759 
1760                 WARN_ON(start > offset + len);
1761 
1762                 end = start + skb_frag_size(f);
1763                 if ((copy = end - offset) > 0) {
1764                         u8 *vaddr;
1765 
1766                         if (copy > len)
1767                                 copy = len;
1768 
1769                         vaddr = kmap_atomic(skb_frag_page(f));
1770                         memcpy(to,
1771                                vaddr + f->page_offset + offset - start,
1772                                copy);
1773                         kunmap_atomic(vaddr);
1774 
1775                         if ((len -= copy) == 0)
1776                                 return 0;
1777                         offset += copy;
1778                         to     += copy;
1779                 }
1780                 start = end;
1781         }
1782 
1783         skb_walk_frags(skb, frag_iter) {
1784                 int end;
1785 
1786                 WARN_ON(start > offset + len);
1787 
1788                 end = start + frag_iter->len;
1789                 if ((copy = end - offset) > 0) {
1790                         if (copy > len)
1791                                 copy = len;
1792                         if (skb_copy_bits(frag_iter, offset - start, to, copy))
1793                                 goto fault;
1794                         if ((len -= copy) == 0)
1795                                 return 0;
1796                         offset += copy;
1797                         to     += copy;
1798                 }
1799                 start = end;
1800         }
1801 
1802         if (!len)
1803                 return 0;
1804 
1805 fault:
1806         return -EFAULT;
1807 }
1808 EXPORT_SYMBOL(skb_copy_bits);
1809 
1810 /*
1811  * Callback from splice_to_pipe(), if we need to release some pages
1812  * at the end of the spd in case we error'ed out in filling the pipe.
1813  */
1814 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1815 {
1816         put_page(spd->pages[i]);
1817 }
1818 
1819 static struct page *linear_to_page(struct page *page, unsigned int *len,
1820                                    unsigned int *offset,
1821                                    struct sock *sk)
1822 {
1823         struct page_frag *pfrag = sk_page_frag(sk);
1824 
1825         if (!sk_page_frag_refill(sk, pfrag))
1826                 return NULL;
1827 
1828         *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
1829 
1830         memcpy(page_address(pfrag->page) + pfrag->offset,
1831                page_address(page) + *offset, *len);
1832         *offset = pfrag->offset;
1833         pfrag->offset += *len;
1834 
1835         return pfrag->page;
1836 }
1837 
1838 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1839                              struct page *page,
1840                              unsigned int offset)
1841 {
1842         return  spd->nr_pages &&
1843                 spd->pages[spd->nr_pages - 1] == page &&
1844                 (spd->partial[spd->nr_pages - 1].offset +
1845                  spd->partial[spd->nr_pages - 1].len == offset);
1846 }
1847 
1848 /*
1849  * Fill page/offset/length into spd, if it can hold more pages.
1850  */
1851 static bool spd_fill_page(struct splice_pipe_desc *spd,
1852                           struct pipe_inode_info *pipe, struct page *page,
1853                           unsigned int *len, unsigned int offset,
1854                           bool linear,
1855                           struct sock *sk)
1856 {
1857         if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1858                 return true;
1859 
1860         if (linear) {
1861                 page = linear_to_page(page, len, &offset, sk);
1862                 if (!page)
1863                         return true;
1864         }
1865         if (spd_can_coalesce(spd, page, offset)) {
1866                 spd->partial[spd->nr_pages - 1].len += *len;
1867                 return false;
1868         }
1869         get_page(page);
1870         spd->pages[spd->nr_pages] = page;
1871         spd->partial[spd->nr_pages].len = *len;
1872         spd->partial[spd->nr_pages].offset = offset;
1873         spd->nr_pages++;
1874 
1875         return false;
1876 }
1877 
1878 static bool __splice_segment(struct page *page, unsigned int poff,
1879                              unsigned int plen, unsigned int *off,
1880                              unsigned int *len,
1881                              struct splice_pipe_desc *spd, bool linear,
1882                              struct sock *sk,
1883                              struct pipe_inode_info *pipe)
1884 {
1885         if (!*len)
1886                 return true;
1887 
1888         /* skip this segment if already processed */
1889         if (*off >= plen) {
1890                 *off -= plen;
1891                 return false;
1892         }
1893 
1894         /* ignore any bits we already processed */
1895         poff += *off;
1896         plen -= *off;
1897         *off = 0;
1898 
1899         do {
1900                 unsigned int flen = min(*len, plen);
1901 
1902                 if (spd_fill_page(spd, pipe, page, &flen, poff,
1903                                   linear, sk))
1904                         return true;
1905                 poff += flen;
1906                 plen -= flen;
1907                 *len -= flen;
1908         } while (*len && plen);
1909 
1910         return false;
1911 }
1912 
1913 /*
1914  * Map linear and fragment data from the skb to spd. It reports true if the
1915  * pipe is full or if we already spliced the requested length.
1916  */
1917 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1918                               unsigned int *offset, unsigned int *len,
1919                               struct splice_pipe_desc *spd, struct sock *sk)
1920 {
1921         int seg;
1922         struct sk_buff *iter;
1923 
1924         /* map the linear part :
1925          * If skb->head_frag is set, this 'linear' part is backed by a
1926          * fragment, and if the head is not shared with any clones then
1927          * we can avoid a copy since we own the head portion of this page.
1928          */
1929         if (__splice_segment(virt_to_page(skb->data),
1930                              (unsigned long) skb->data & (PAGE_SIZE - 1),
1931                              skb_headlen(skb),
1932                              offset, len, spd,
1933                              skb_head_is_locked(skb),
1934                              sk, pipe))
1935                 return true;
1936 
1937         /*
1938          * then map the fragments
1939          */
1940         for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1941                 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1942 
1943                 if (__splice_segment(skb_frag_page(f),
1944                                      f->page_offset, skb_frag_size(f),
1945                                      offset, len, spd, false, sk, pipe))
1946                         return true;
1947         }
1948 
1949         skb_walk_frags(skb, iter) {
1950                 if (*offset >= iter->len) {
1951                         *offset -= iter->len;
1952                         continue;
1953                 }
1954                 /* __skb_splice_bits() only fails if the output has no room
1955                  * left, so no point in going over the frag_list for the error
1956                  * case.
1957                  */
1958                 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
1959                         return true;
1960         }
1961 
1962         return false;
1963 }
1964 
1965 ssize_t skb_socket_splice(struct sock *sk,
1966                           struct pipe_inode_info *pipe,
1967                           struct splice_pipe_desc *spd)
1968 {
1969         int ret;
1970 
1971         /* Drop the socket lock, otherwise we have reverse
1972          * locking dependencies between sk_lock and i_mutex
1973          * here as compared to sendfile(). We enter here
1974          * with the socket lock held, and splice_to_pipe() will
1975          * grab the pipe inode lock. For sendfile() emulation,
1976          * we call into ->sendpage() with the i_mutex lock held
1977          * and networking will grab the socket lock.
1978          */
1979         release_sock(sk);
1980         ret = splice_to_pipe(pipe, spd);
1981         lock_sock(sk);
1982 
1983         return ret;
1984 }
1985 
1986 /*
1987  * Map data from the skb to a pipe. Should handle both the linear part,
1988  * the fragments, and the frag list.
1989  */
1990 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
1991                     struct pipe_inode_info *pipe, unsigned int tlen,
1992                     unsigned int flags,
1993                     ssize_t (*splice_cb)(struct sock *,
1994                                          struct pipe_inode_info *,
1995                                          struct splice_pipe_desc *))
1996 {
1997         struct partial_page partial[MAX_SKB_FRAGS];
1998         struct page *pages[MAX_SKB_FRAGS];
1999         struct splice_pipe_desc spd = {
2000                 .pages = pages,
2001                 .partial = partial,
2002                 .nr_pages_max = MAX_SKB_FRAGS,
2003                 .flags = flags,
2004                 .ops = &nosteal_pipe_buf_ops,
2005                 .spd_release = sock_spd_release,
2006         };
2007         int ret = 0;
2008 
2009         __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2010 
2011         if (spd.nr_pages)
2012                 ret = splice_cb(sk, pipe, &spd);
2013 
2014         return ret;
2015 }
2016 EXPORT_SYMBOL_GPL(skb_splice_bits);
2017 
2018 /**
2019  *      skb_store_bits - store bits from kernel buffer to skb
2020  *      @skb: destination buffer
2021  *      @offset: offset in destination
2022  *      @from: source buffer
2023  *      @len: number of bytes to copy
2024  *
2025  *      Copy the specified number of bytes from the source buffer to the
2026  *      destination skb.  This function handles all the messy bits of
2027  *      traversing fragment lists and such.
2028  */
2029 
2030 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2031 {
2032         int start = skb_headlen(skb);
2033         struct sk_buff *frag_iter;
2034         int i, copy;
2035 
2036         if (offset > (int)skb->len - len)
2037                 goto fault;
2038 
2039         if ((copy = start - offset) > 0) {
2040                 if (copy > len)
2041                         copy = len;
2042                 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2043                 if ((len -= copy) == 0)
2044                         return 0;
2045                 offset += copy;
2046                 from += copy;
2047         }
2048 
2049         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2050                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2051                 int end;
2052 
2053                 WARN_ON(start > offset + len);
2054 
2055                 end = start + skb_frag_size(frag);
2056                 if ((copy = end - offset) > 0) {
2057                         u8 *vaddr;
2058 
2059                         if (copy > len)
2060                                 copy = len;
2061 
2062                         vaddr = kmap_atomic(skb_frag_page(frag));
2063                         memcpy(vaddr + frag->page_offset + offset - start,
2064                                from, copy);
2065                         kunmap_atomic(vaddr);
2066 
2067                         if ((len -= copy) == 0)
2068                                 return 0;
2069                         offset += copy;
2070                         from += copy;
2071                 }
2072                 start = end;
2073         }
2074 
2075         skb_walk_frags(skb, frag_iter) {
2076                 int end;
2077 
2078                 WARN_ON(start > offset + len);
2079 
2080                 end = start + frag_iter->len;
2081                 if ((copy = end - offset) > 0) {
2082                         if (copy > len)
2083                                 copy = len;
2084                         if (skb_store_bits(frag_iter, offset - start,
2085                                            from, copy))
2086                                 goto fault;
2087                         if ((len -= copy) == 0)
2088                                 return 0;
2089                         offset += copy;
2090                         from += copy;
2091                 }
2092                 start = end;
2093         }
2094         if (!len)
2095                 return 0;
2096 
2097 fault:
2098         return -EFAULT;
2099 }
2100 EXPORT_SYMBOL(skb_store_bits);
2101 
2102 /* Checksum skb data. */
2103 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2104                       __wsum csum, const struct skb_checksum_ops *ops)
2105 {
2106         int start = skb_headlen(skb);
2107         int i, copy = start - offset;
2108         struct sk_buff *frag_iter;
2109         int pos = 0;
2110 
2111         /* Checksum header. */
2112         if (copy > 0) {
2113                 if (copy > len)
2114                         copy = len;
2115                 csum = ops->update(skb->data + offset, copy, csum);
2116                 if ((len -= copy) == 0)
2117                         return csum;
2118                 offset += copy;
2119                 pos     = copy;
2120         }
2121 
2122         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2123                 int end;
2124                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2125 
2126                 WARN_ON(start > offset + len);
2127 
2128                 end = start + skb_frag_size(frag);
2129                 if ((copy = end - offset) > 0) {
2130                         __wsum csum2;
2131                         u8 *vaddr;
2132 
2133                         if (copy > len)
2134                                 copy = len;
2135                         vaddr = kmap_atomic(skb_frag_page(frag));
2136                         csum2 = ops->update(vaddr + frag->page_offset +
2137                                             offset - start, copy, 0);
2138                         kunmap_atomic(vaddr);
2139                         csum = ops->combine(csum, csum2, pos, copy);
2140                         if (!(len -= copy))
2141                                 return csum;
2142                         offset += copy;
2143                         pos    += copy;
2144                 }
2145                 start = end;
2146         }
2147 
2148         skb_walk_frags(skb, frag_iter) {
2149                 int end;
2150 
2151                 WARN_ON(start > offset + len);
2152 
2153                 end = start + frag_iter->len;
2154                 if ((copy = end - offset) > 0) {
2155                         __wsum csum2;
2156                         if (copy > len)
2157                                 copy = len;
2158                         csum2 = __skb_checksum(frag_iter, offset - start,
2159                                                copy, 0, ops);
2160                         csum = ops->combine(csum, csum2, pos, copy);
2161                         if ((len -= copy) == 0)
2162                                 return csum;
2163                         offset += copy;
2164                         pos    += copy;
2165                 }
2166                 start = end;
2167         }
2168         BUG_ON(len);
2169 
2170         return csum;
2171 }
2172 EXPORT_SYMBOL(__skb_checksum);
2173 
2174 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2175                     int len, __wsum csum)
2176 {
2177         const struct skb_checksum_ops ops = {
2178                 .update  = csum_partial_ext,
2179                 .combine = csum_block_add_ext,
2180         };
2181 
2182         return __skb_checksum(skb, offset, len, csum, &ops);
2183 }
2184 EXPORT_SYMBOL(skb_checksum);
2185 
2186 /* Both of above in one bottle. */
2187 
2188 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2189                                     u8 *to, int len, __wsum csum)
2190 {
2191         int start = skb_headlen(skb);
2192         int i, copy = start - offset;
2193         struct sk_buff *frag_iter;
2194         int pos = 0;
2195 
2196         /* Copy header. */
2197         if (copy > 0) {
2198                 if (copy > len)
2199                         copy = len;
2200                 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2201                                                  copy, csum);
2202                 if ((len -= copy) == 0)
2203                         return csum;
2204                 offset += copy;
2205                 to     += copy;
2206                 pos     = copy;
2207         }
2208 
2209         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2210                 int end;
2211 
2212                 WARN_ON(start > offset + len);
2213 
2214                 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2215                 if ((copy = end - offset) > 0) {
2216                         __wsum csum2;
2217                         u8 *vaddr;
2218                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2219 
2220                         if (copy > len)
2221                                 copy = len;
2222                         vaddr = kmap_atomic(skb_frag_page(frag));
2223                         csum2 = csum_partial_copy_nocheck(vaddr +
2224                                                           frag->page_offset +
2225                                                           offset - start, to,
2226                                                           copy, 0);
2227                         kunmap_atomic(vaddr);
2228                         csum = csum_block_add(csum, csum2, pos);
2229                         if (!(len -= copy))
2230                                 return csum;
2231                         offset += copy;
2232                         to     += copy;
2233                         pos    += copy;
2234                 }
2235                 start = end;
2236         }
2237 
2238         skb_walk_frags(skb, frag_iter) {
2239                 __wsum csum2;
2240                 int end;
2241 
2242                 WARN_ON(start > offset + len);
2243 
2244                 end = start + frag_iter->len;
2245                 if ((copy = end - offset) > 0) {
2246                         if (copy > len)
2247                                 copy = len;
2248                         csum2 = skb_copy_and_csum_bits(frag_iter,
2249                                                        offset - start,
2250                                                        to, copy, 0);
2251                         csum = csum_block_add(csum, csum2, pos);
2252                         if ((len -= copy) == 0)
2253                                 return csum;
2254                         offset += copy;
2255                         to     += copy;
2256                         pos    += copy;
2257                 }
2258                 start = end;
2259         }
2260         BUG_ON(len);
2261         return csum;
2262 }
2263 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2264 
2265  /**
2266  *      skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2267  *      @from: source buffer
2268  *
2269  *      Calculates the amount of linear headroom needed in the 'to' skb passed
2270  *      into skb_zerocopy().
2271  */
2272 unsigned int
2273 skb_zerocopy_headlen(const struct sk_buff *from)
2274 {
2275         unsigned int hlen = 0;
2276 
2277         if (!from->head_frag ||
2278             skb_headlen(from) < L1_CACHE_BYTES ||
2279             skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2280                 hlen = skb_headlen(from);
2281 
2282         if (skb_has_frag_list(from))
2283                 hlen = from->len;
2284 
2285         return hlen;
2286 }
2287 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2288 
2289 /**
2290  *      skb_zerocopy - Zero copy skb to skb
2291  *      @to: destination buffer
2292  *      @from: source buffer
2293  *      @len: number of bytes to copy from source buffer
2294  *      @hlen: size of linear headroom in destination buffer
2295  *
2296  *      Copies up to `len` bytes from `from` to `to` by creating references
2297  *      to the frags in the source buffer.
2298  *
2299  *      The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2300  *      headroom in the `to` buffer.
2301  *
2302  *      Return value:
2303  *      0: everything is OK
2304  *      -ENOMEM: couldn't orphan frags of @from due to lack of memory
2305  *      -EFAULT: skb_copy_bits() found some problem with skb geometry
2306  */
2307 int
2308 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2309 {
2310         int i, j = 0;
2311         int plen = 0; /* length of skb->head fragment */
2312         int ret;
2313         struct page *page;
2314         unsigned int offset;
2315 
2316         BUG_ON(!from->head_frag && !hlen);
2317 
2318         /* dont bother with small payloads */
2319         if (len <= skb_tailroom(to))
2320                 return skb_copy_bits(from, 0, skb_put(to, len), len);
2321 
2322         if (hlen) {
2323                 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2324                 if (unlikely(ret))
2325                         return ret;
2326                 len -= hlen;
2327         } else {
2328                 plen = min_t(int, skb_headlen(from), len);
2329                 if (plen) {
2330                         page = virt_to_head_page(from->head);
2331                         offset = from->data - (unsigned char *)page_address(page);
2332                         __skb_fill_page_desc(to, 0, page, offset, plen);
2333                         get_page(page);
2334                         j = 1;
2335                         len -= plen;
2336                 }
2337         }
2338 
2339         to->truesize += len + plen;
2340         to->len += len + plen;
2341         to->data_len += len + plen;
2342 
2343         if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2344                 skb_tx_error(from);
2345                 return -ENOMEM;
2346         }
2347 
2348         for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2349                 if (!len)
2350                         break;
2351                 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2352                 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
2353                 len -= skb_shinfo(to)->frags[j].size;
2354                 skb_frag_ref(to, j);
2355                 j++;
2356         }
2357         skb_shinfo(to)->nr_frags = j;
2358 
2359         return 0;
2360 }
2361 EXPORT_SYMBOL_GPL(skb_zerocopy);
2362 
2363 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2364 {
2365         __wsum csum;
2366         long csstart;
2367 
2368         if (skb->ip_summed == CHECKSUM_PARTIAL)
2369                 csstart = skb_checksum_start_offset(skb);
2370         else
2371                 csstart = skb_headlen(skb);
2372 
2373         BUG_ON(csstart > skb_headlen(skb));
2374 
2375         skb_copy_from_linear_data(skb, to, csstart);
2376 
2377         csum = 0;
2378         if (csstart != skb->len)
2379                 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2380                                               skb->len - csstart, 0);
2381 
2382         if (skb->ip_summed == CHECKSUM_PARTIAL) {
2383                 long csstuff = csstart + skb->csum_offset;
2384 
2385                 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2386         }
2387 }
2388 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2389 
2390 /**
2391  *      skb_dequeue - remove from the head of the queue
2392  *      @list: list to dequeue from
2393  *
2394  *      Remove the head of the list. The list lock is taken so the function
2395  *      may be used safely with other locking list functions. The head item is
2396  *      returned or %NULL if the list is empty.
2397  */
2398 
2399 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2400 {
2401         unsigned long flags;
2402         struct sk_buff *result;
2403 
2404         spin_lock_irqsave(&list->lock, flags);
2405         result = __skb_dequeue(list);
2406         spin_unlock_irqrestore(&list->lock, flags);
2407         return result;
2408 }
2409 EXPORT_SYMBOL(skb_dequeue);
2410 
2411 /**
2412  *      skb_dequeue_tail - remove from the tail of the queue
2413  *      @list: list to dequeue from
2414  *
2415  *      Remove the tail of the list. The list lock is taken so the function
2416  *      may be used safely with other locking list functions. The tail item is
2417  *      returned or %NULL if the list is empty.
2418  */
2419 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2420 {
2421         unsigned long flags;
2422         struct sk_buff *result;
2423 
2424         spin_lock_irqsave(&list->lock, flags);
2425         result = __skb_dequeue_tail(list);
2426         spin_unlock_irqrestore(&list->lock, flags);
2427         return result;
2428 }
2429 EXPORT_SYMBOL(skb_dequeue_tail);
2430 
2431 /**
2432  *      skb_queue_purge - empty a list
2433  *      @list: list to empty
2434  *
2435  *      Delete all buffers on an &sk_buff list. Each buffer is removed from
2436  *      the list and one reference dropped. This function takes the list
2437  *      lock and is atomic with respect to other list locking functions.
2438  */
2439 void skb_queue_purge(struct sk_buff_head *list)
2440 {
2441         struct sk_buff *skb;
2442         while ((skb = skb_dequeue(list)) != NULL)
2443                 kfree_skb(skb);
2444 }
2445 EXPORT_SYMBOL(skb_queue_purge);
2446 
2447 /**
2448  *      skb_queue_head - queue a buffer at the list head
2449  *      @list: list to use
2450  *      @newsk: buffer to queue
2451  *
2452  *      Queue a buffer at the start of the list. This function takes the
2453  *      list lock and can be used safely with other locking &sk_buff functions
2454  *      safely.
2455  *
2456  *      A buffer cannot be placed on two lists at the same time.
2457  */
2458 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2459 {
2460         unsigned long flags;
2461 
2462         spin_lock_irqsave(&list->lock, flags);
2463         __skb_queue_head(list, newsk);
2464         spin_unlock_irqrestore(&list->lock, flags);
2465 }
2466 EXPORT_SYMBOL(skb_queue_head);
2467 
2468 /**
2469  *      skb_queue_tail - queue a buffer at the list tail
2470  *      @list: list to use
2471  *      @newsk: buffer to queue
2472  *
2473  *      Queue a buffer at the tail of the list. This function takes the
2474  *      list lock and can be used safely with other locking &sk_buff functions
2475  *      safely.
2476  *
2477  *      A buffer cannot be placed on two lists at the same time.
2478  */
2479 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2480 {
2481         unsigned long flags;
2482 
2483         spin_lock_irqsave(&list->lock, flags);
2484         __skb_queue_tail(list, newsk);
2485         spin_unlock_irqrestore(&list->lock, flags);
2486 }
2487 EXPORT_SYMBOL(skb_queue_tail);
2488 
2489 /**
2490  *      skb_unlink      -       remove a buffer from a list
2491  *      @skb: buffer to remove
2492  *      @list: list to use
2493  *
2494  *      Remove a packet from a list. The list locks are taken and this
2495  *      function is atomic with respect to other list locked calls
2496  *
2497  *      You must know what list the SKB is on.
2498  */
2499 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2500 {
2501         unsigned long flags;
2502 
2503         spin_lock_irqsave(&list->lock, flags);
2504         __skb_unlink(skb, list);
2505         spin_unlock_irqrestore(&list->lock, flags);
2506 }
2507 EXPORT_SYMBOL(skb_unlink);
2508 
2509 /**
2510  *      skb_append      -       append a buffer
2511  *      @old: buffer to insert after
2512  *      @newsk: buffer to insert
2513  *      @list: list to use
2514  *
2515  *      Place a packet after a given packet in a list. The list locks are taken
2516  *      and this function is atomic with respect to other list locked calls.
2517  *      A buffer cannot be placed on two lists at the same time.
2518  */
2519 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2520 {
2521         unsigned long flags;
2522 
2523         spin_lock_irqsave(&list->lock, flags);
2524         __skb_queue_after(list, old, newsk);
2525         spin_unlock_irqrestore(&list->lock, flags);
2526 }
2527 EXPORT_SYMBOL(skb_append);
2528 
2529 /**
2530  *      skb_insert      -       insert a buffer
2531  *      @old: buffer to insert before
2532  *      @newsk: buffer to insert
2533  *      @list: list to use
2534  *
2535  *      Place a packet before a given packet in a list. The list locks are
2536  *      taken and this function is atomic with respect to other list locked
2537  *      calls.
2538  *
2539  *      A buffer cannot be placed on two lists at the same time.
2540  */
2541 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2542 {
2543         unsigned long flags;
2544 
2545         spin_lock_irqsave(&list->lock, flags);
2546         __skb_insert(newsk, old->prev, old, list);
2547         spin_unlock_irqrestore(&list->lock, flags);
2548 }
2549 EXPORT_SYMBOL(skb_insert);
2550 
2551 static inline void skb_split_inside_header(struct sk_buff *skb,
2552                                            struct sk_buff* skb1,
2553                                            const u32 len, const int pos)
2554 {
2555         int i;
2556 
2557         skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2558                                          pos - len);
2559         /* And move data appendix as is. */
2560         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2561                 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2562 
2563         skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2564         skb_shinfo(skb)->nr_frags  = 0;
2565         skb1->data_len             = skb->data_len;
2566         skb1->len                  += skb1->data_len;
2567         skb->data_len              = 0;
2568         skb->len                   = len;
2569         skb_set_tail_pointer(skb, len);
2570 }
2571 
2572 static inline void skb_split_no_header(struct sk_buff *skb,
2573                                        struct sk_buff* skb1,
2574                                        const u32 len, int pos)
2575 {
2576         int i, k = 0;
2577         const int nfrags = skb_shinfo(skb)->nr_frags;
2578 
2579         skb_shinfo(skb)->nr_frags = 0;
2580         skb1->len                 = skb1->data_len = skb->len - len;
2581         skb->len                  = len;
2582         skb->data_len             = len - pos;
2583 
2584         for (i = 0; i < nfrags; i++) {
2585                 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2586 
2587                 if (pos + size > len) {
2588                         skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2589 
2590                         if (pos < len) {
2591                                 /* Split frag.
2592                                  * We have two variants in this case:
2593                                  * 1. Move all the frag to the second
2594                                  *    part, if it is possible. F.e.
2595                                  *    this approach is mandatory for TUX,
2596                                  *    where splitting is expensive.
2597                                  * 2. Split is accurately. We make this.
2598                                  */
2599                                 skb_frag_ref(skb, i);
2600                                 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2601                                 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2602                                 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2603                                 skb_shinfo(skb)->nr_frags++;
2604                         }
2605                         k++;
2606                 } else
2607                         skb_shinfo(skb)->nr_frags++;
2608                 pos += size;
2609         }
2610         skb_shinfo(skb1)->nr_frags = k;
2611 }
2612 
2613 /**
2614  * skb_split - Split fragmented skb to two parts at length len.
2615  * @skb: the buffer to split
2616  * @skb1: the buffer to receive the second part
2617  * @len: new length for skb
2618  */
2619 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2620 {
2621         int pos = skb_headlen(skb);
2622 
2623         skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2624         if (len < pos)  /* Split line is inside header. */
2625                 skb_split_inside_header(skb, skb1, len, pos);
2626         else            /* Second chunk has no header, nothing to copy. */
2627                 skb_split_no_header(skb, skb1, len, pos);
2628 }
2629 EXPORT_SYMBOL(skb_split);
2630 
2631 /* Shifting from/to a cloned skb is a no-go.
2632  *
2633  * Caller cannot keep skb_shinfo related pointers past calling here!
2634  */
2635 static int skb_prepare_for_shift(struct sk_buff *skb)
2636 {
2637         return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2638 }
2639 
2640 /**
2641  * skb_shift - Shifts paged data partially from skb to another
2642  * @tgt: buffer into which tail data gets added
2643  * @skb: buffer from which the paged data comes from
2644  * @shiftlen: shift up to this many bytes
2645  *
2646  * Attempts to shift up to shiftlen worth of bytes, which may be less than
2647  * the length of the skb, from skb to tgt. Returns number bytes shifted.
2648  * It's up to caller to free skb if everything was shifted.
2649  *
2650  * If @tgt runs out of frags, the whole operation is aborted.
2651  *
2652  * Skb cannot include anything else but paged data while tgt is allowed
2653  * to have non-paged data as well.
2654  *
2655  * TODO: full sized shift could be optimized but that would need
2656  * specialized skb free'er to handle frags without up-to-date nr_frags.
2657  */
2658 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2659 {
2660         int from, to, merge, todo;
2661         struct skb_frag_struct *fragfrom, *fragto;
2662 
2663         BUG_ON(shiftlen > skb->len);
2664         BUG_ON(skb_headlen(skb));       /* Would corrupt stream */
2665 
2666         todo = shiftlen;
2667         from = 0;
2668         to = skb_shinfo(tgt)->nr_frags;
2669         fragfrom = &skb_shinfo(skb)->frags[from];
2670 
2671         /* Actual merge is delayed until the point when we know we can
2672          * commit all, so that we don't have to undo partial changes
2673          */
2674         if (!to ||
2675             !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2676                               fragfrom->page_offset)) {
2677                 merge = -1;
2678         } else {
2679                 merge = to - 1;
2680 
2681                 todo -= skb_frag_size(fragfrom);
2682                 if (todo < 0) {
2683                         if (skb_prepare_for_shift(skb) ||
2684                             skb_prepare_for_shift(tgt))
2685                                 return 0;
2686 
2687                         /* All previous frag pointers might be stale! */
2688                         fragfrom = &skb_shinfo(skb)->frags[from];
2689                         fragto = &skb_shinfo(tgt)->frags[merge];
2690 
2691                         skb_frag_size_add(fragto, shiftlen);
2692                         skb_frag_size_sub(fragfrom, shiftlen);
2693                         fragfrom->page_offset += shiftlen;
2694 
2695                         goto onlymerged;
2696                 }
2697 
2698                 from++;
2699         }
2700 
2701         /* Skip full, not-fitting skb to avoid expensive operations */
2702         if ((shiftlen == skb->len) &&
2703             (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2704                 return 0;
2705 
2706         if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2707                 return 0;
2708 
2709         while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2710                 if (to == MAX_SKB_FRAGS)
2711                         return 0;
2712 
2713                 fragfrom = &skb_shinfo(skb)->frags[from];
2714                 fragto = &skb_shinfo(tgt)->frags[to];
2715 
2716                 if (todo >= skb_frag_size(fragfrom)) {
2717                         *fragto = *fragfrom;
2718                         todo -= skb_frag_size(fragfrom);
2719                         from++;
2720                         to++;
2721 
2722                 } else {
2723                         __skb_frag_ref(fragfrom);
2724                         fragto->page = fragfrom->page;
2725                         fragto->page_offset = fragfrom->page_offset;
2726                         skb_frag_size_set(fragto, todo);
2727 
2728                         fragfrom->page_offset += todo;
2729                         skb_frag_size_sub(fragfrom, todo);
2730                         todo = 0;
2731 
2732                         to++;
2733                         break;
2734                 }
2735         }
2736 
2737         /* Ready to "commit" this state change to tgt */
2738         skb_shinfo(tgt)->nr_frags = to;
2739 
2740         if (merge >= 0) {
2741                 fragfrom = &skb_shinfo(skb)->frags[0];
2742                 fragto = &skb_shinfo(tgt)->frags[merge];
2743 
2744                 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2745                 __skb_frag_unref(fragfrom);
2746         }
2747 
2748         /* Reposition in the original skb */
2749         to = 0;
2750         while (from < skb_shinfo(skb)->nr_frags)
2751                 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2752         skb_shinfo(skb)->nr_frags = to;
2753 
2754         BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2755 
2756 onlymerged:
2757         /* Most likely the tgt won't ever need its checksum anymore, skb on
2758          * the other hand might need it if it needs to be resent
2759          */
2760         tgt->ip_summed = CHECKSUM_PARTIAL;
2761         skb->ip_summed = CHECKSUM_PARTIAL;
2762 
2763         /* Yak, is it really working this way? Some helper please? */
2764         skb->len -= shiftlen;
2765         skb->data_len -= shiftlen;
2766         skb->truesize -= shiftlen;
2767         tgt->len += shiftlen;
2768         tgt->data_len += shiftlen;
2769         tgt->truesize += shiftlen;
2770 
2771         return shiftlen;
2772 }
2773 
2774 /**
2775  * skb_prepare_seq_read - Prepare a sequential read of skb data
2776  * @skb: the buffer to read
2777  * @from: lower offset of data to be read
2778  * @to: upper offset of data to be read
2779  * @st: state variable
2780  *
2781  * Initializes the specified state variable. Must be called before
2782  * invoking skb_seq_read() for the first time.
2783  */
2784 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2785                           unsigned int to, struct skb_seq_state *st)
2786 {
2787         st->lower_offset = from;
2788         st->upper_offset = to;
2789         st->root_skb = st->cur_skb = skb;
2790         st->frag_idx = st->stepped_offset = 0;
2791         st->frag_data = NULL;
2792 }
2793 EXPORT_SYMBOL(skb_prepare_seq_read);
2794 
2795 /**
2796  * skb_seq_read - Sequentially read skb data
2797  * @consumed: number of bytes consumed by the caller so far
2798  * @data: destination pointer for data to be returned
2799  * @st: state variable
2800  *
2801  * Reads a block of skb data at @consumed relative to the
2802  * lower offset specified to skb_prepare_seq_read(). Assigns
2803  * the head of the data block to @data and returns the length
2804  * of the block or 0 if the end of the skb data or the upper
2805  * offset has been reached.
2806  *
2807  * The caller is not required to consume all of the data
2808  * returned, i.e. @consumed is typically set to the number
2809  * of bytes already consumed and the next call to
2810  * skb_seq_read() will return the remaining part of the block.
2811  *
2812  * Note 1: The size of each block of data returned can be arbitrary,
2813  *       this limitation is the cost for zerocopy sequential
2814  *       reads of potentially non linear data.
2815  *
2816  * Note 2: Fragment lists within fragments are not implemented
2817  *       at the moment, state->root_skb could be replaced with
2818  *       a stack for this purpose.
2819  */
2820 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2821                           struct skb_seq_state *st)
2822 {
2823         unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2824         skb_frag_t *frag;
2825 
2826         if (unlikely(abs_offset >= st->upper_offset)) {
2827                 if (st->frag_data) {
2828                         kunmap_atomic(st->frag_data);
2829                         st->frag_data = NULL;
2830                 }
2831                 return 0;
2832         }
2833 
2834 next_skb:
2835         block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2836 
2837         if (abs_offset < block_limit && !st->frag_data) {
2838                 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2839                 return block_limit - abs_offset;
2840         }
2841 
2842         if (st->frag_idx == 0 && !st->frag_data)
2843                 st->stepped_offset += skb_headlen(st->cur_skb);
2844 
2845         while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2846                 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2847                 block_limit = skb_frag_size(frag) + st->stepped_offset;
2848 
2849                 if (abs_offset < block_limit) {
2850                         if (!st->frag_data)
2851                                 st->frag_data = kmap_atomic(skb_frag_page(frag));
2852 
2853                         *data = (u8 *) st->frag_data + frag->page_offset +
2854                                 (abs_offset - st->stepped_offset);
2855 
2856                         return block_limit - abs_offset;
2857                 }
2858 
2859                 if (st->frag_data) {
2860                         kunmap_atomic(st->frag_data);
2861                         st->frag_data = NULL;
2862                 }
2863 
2864                 st->frag_idx++;
2865                 st->stepped_offset += skb_frag_size(frag);
2866         }
2867 
2868         if (st->frag_data) {
2869                 kunmap_atomic(st->frag_data);
2870                 st->frag_data = NULL;
2871         }
2872 
2873         if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2874                 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2875                 st->frag_idx = 0;
2876                 goto next_skb;
2877         } else if (st->cur_skb->next) {
2878                 st->cur_skb = st->cur_skb->next;
2879                 st->frag_idx = 0;
2880                 goto next_skb;
2881         }
2882 
2883         return 0;
2884 }
2885 EXPORT_SYMBOL(skb_seq_read);
2886 
2887 /**
2888  * skb_abort_seq_read - Abort a sequential read of skb data
2889  * @st: state variable
2890  *
2891  * Must be called if skb_seq_read() was not called until it
2892  * returned 0.
2893  */
2894 void skb_abort_seq_read(struct skb_seq_state *st)
2895 {
2896         if (st->frag_data)
2897                 kunmap_atomic(st->frag_data);
2898 }
2899 EXPORT_SYMBOL(skb_abort_seq_read);
2900 
2901 #define TS_SKB_CB(state)        ((struct skb_seq_state *) &((state)->cb))
2902 
2903 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2904                                           struct ts_config *conf,
2905                                           struct ts_state *state)
2906 {
2907         return skb_seq_read(offset, text, TS_SKB_CB(state));
2908 }
2909 
2910 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2911 {
2912         skb_abort_seq_read(TS_SKB_CB(state));
2913 }
2914 
2915 /**
2916  * skb_find_text - Find a text pattern in skb data
2917  * @skb: the buffer to look in
2918  * @from: search offset
2919  * @to: search limit
2920  * @config: textsearch configuration
2921  *
2922  * Finds a pattern in the skb data according to the specified
2923  * textsearch configuration. Use textsearch_next() to retrieve
2924  * subsequent occurrences of the pattern. Returns the offset
2925  * to the first occurrence or UINT_MAX if no match was found.
2926  */
2927 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2928                            unsigned int to, struct ts_config *config)
2929 {
2930         struct ts_state state;
2931         unsigned int ret;
2932 
2933         config->get_next_block = skb_ts_get_next_block;
2934         config->finish = skb_ts_finish;
2935 
2936         skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
2937 
2938         ret = textsearch_find(config, &state);
2939         return (ret <= to - from ? ret : UINT_MAX);
2940 }
2941 EXPORT_SYMBOL(skb_find_text);
2942 
2943 /**
2944  * skb_append_datato_frags - append the user data to a skb
2945  * @sk: sock  structure
2946  * @skb: skb structure to be appended with user data.
2947  * @getfrag: call back function to be used for getting the user data
2948  * @from: pointer to user message iov
2949  * @length: length of the iov message
2950  *
2951  * Description: This procedure append the user data in the fragment part
2952  * of the skb if any page alloc fails user this procedure returns  -ENOMEM
2953  */
2954 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2955                         int (*getfrag)(void *from, char *to, int offset,
2956                                         int len, int odd, struct sk_buff *skb),
2957                         void *from, int length)
2958 {
2959         int frg_cnt = skb_shinfo(skb)->nr_frags;
2960         int copy;
2961         int offset = 0;
2962         int ret;
2963         struct page_frag *pfrag = &current->task_frag;
2964 
2965         do {
2966                 /* Return error if we don't have space for new frag */
2967                 if (frg_cnt >= MAX_SKB_FRAGS)
2968                         return -EMSGSIZE;
2969 
2970                 if (!sk_page_frag_refill(sk, pfrag))
2971                         return -ENOMEM;
2972 
2973                 /* copy the user data to page */
2974                 copy = min_t(int, length, pfrag->size - pfrag->offset);
2975 
2976                 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
2977                               offset, copy, 0, skb);
2978                 if (ret < 0)
2979                         return -EFAULT;
2980 
2981                 /* copy was successful so update the size parameters */
2982                 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
2983                                    copy);
2984                 frg_cnt++;
2985                 pfrag->offset += copy;
2986                 get_page(pfrag->page);
2987 
2988                 skb->truesize += copy;
2989                 atomic_add(copy, &sk->sk_wmem_alloc);
2990                 skb->len += copy;
2991                 skb->data_len += copy;
2992                 offset += copy;
2993                 length -= copy;
2994 
2995         } while (length > 0);
2996 
2997         return 0;
2998 }
2999 EXPORT_SYMBOL(skb_append_datato_frags);
3000 
3001 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3002                          int offset, size_t size)
3003 {
3004         int i = skb_shinfo(skb)->nr_frags;
3005 
3006         if (skb_can_coalesce(skb, i, page, offset)) {
3007                 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3008         } else if (i < MAX_SKB_FRAGS) {
3009                 get_page(page);
3010                 skb_fill_page_desc(skb, i, page, offset, size);
3011         } else {
3012                 return -EMSGSIZE;
3013         }
3014 
3015         return 0;
3016 }
3017 EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3018 
3019 /**
3020  *      skb_pull_rcsum - pull skb and update receive checksum
3021  *      @skb: buffer to update
3022  *      @len: length of data pulled
3023  *
3024  *      This function performs an skb_pull on the packet and updates
3025  *      the CHECKSUM_COMPLETE checksum.  It should be used on
3026  *      receive path processing instead of skb_pull unless you know
3027  *      that the checksum difference is zero (e.g., a valid IP header)
3028  *      or you are setting ip_summed to CHECKSUM_NONE.
3029  */
3030 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3031 {
3032         unsigned char *data = skb->data;
3033 
3034         BUG_ON(len > skb->len);
3035         __skb_pull(skb, len);
3036         skb_postpull_rcsum(skb, data, len);
3037         return skb->data;
3038 }
3039 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3040 
3041 /**
3042  *      skb_segment - Perform protocol segmentation on skb.
3043  *      @head_skb: buffer to segment
3044  *      @features: features for the output path (see dev->features)
3045  *
3046  *      This function performs segmentation on the given skb.  It returns
3047  *      a pointer to the first in a list of new skbs for the segments.
3048  *      In case of error it returns ERR_PTR(err).
3049  */
3050 struct sk_buff *skb_segment(struct sk_buff *head_skb,
3051                             netdev_features_t features)
3052 {
3053         struct sk_buff *segs = NULL;
3054         struct sk_buff *tail = NULL;
3055         struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3056         skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3057         unsigned int mss = skb_shinfo(head_skb)->gso_size;
3058         unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3059         struct sk_buff *frag_skb = head_skb;
3060         unsigned int offset = doffset;
3061         unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3062         unsigned int partial_segs = 0;
3063         unsigned int headroom;
3064         unsigned int len = head_skb->len;
3065         __be16 proto;
3066         bool csum, sg;
3067         int nfrags = skb_shinfo(head_skb)->nr_frags;
3068         int err = -ENOMEM;
3069         int i = 0;
3070         int pos;
3071         int dummy;
3072 
3073         __skb_push(head_skb, doffset);
3074         proto = skb_network_protocol(head_skb, &dummy);
3075         if (unlikely(!proto))
3076                 return ERR_PTR(-EINVAL);
3077 
3078         sg = !!(features & NETIF_F_SG);
3079         csum = !!can_checksum_protocol(features, proto);
3080 
3081         /* GSO partial only requires that we trim off any excess that
3082          * doesn't fit into an MSS sized block, so take care of that
3083          * now.
3084          */
3085         if (sg && csum && (features & NETIF_F_GSO_PARTIAL)) {
3086                 partial_segs = len / mss;
3087                 if (partial_segs > 1)
3088                         mss *= partial_segs;
3089                 else
3090                         partial_segs = 0;
3091         }
3092 
3093         headroom = skb_headroom(head_skb);
3094         pos = skb_headlen(head_skb);
3095 
3096         do {
3097                 struct sk_buff *nskb;
3098                 skb_frag_t *nskb_frag;
3099                 int hsize;
3100                 int size;
3101 
3102                 if (unlikely(mss == GSO_BY_FRAGS)) {
3103                         len = list_skb->len;
3104                 } else {
3105                         len = head_skb->len - offset;
3106                         if (len > mss)
3107                                 len = mss;
3108                 }
3109 
3110                 hsize = skb_headlen(head_skb) - offset;
3111                 if (hsize < 0)
3112                         hsize = 0;
3113                 if (hsize > len || !sg)
3114                         hsize = len;
3115 
3116                 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3117                     (skb_headlen(list_skb) == len || sg)) {
3118                         BUG_ON(skb_headlen(list_skb) > len);
3119 
3120                         i = 0;
3121                         nfrags = skb_shinfo(list_skb)->nr_frags;
3122                         frag = skb_shinfo(list_skb)->frags;
3123                         frag_skb = list_skb;
3124                         pos += skb_headlen(list_skb);
3125 
3126                         while (pos < offset + len) {
3127                                 BUG_ON(i >= nfrags);
3128 
3129                                 size = skb_frag_size(frag);
3130                                 if (pos + size > offset + len)
3131                                         break;
3132 
3133                                 i++;
3134                                 pos += size;
3135                                 frag++;
3136                         }
3137 
3138                         nskb = skb_clone(list_skb, GFP_ATOMIC);
3139                         list_skb = list_skb->next;
3140 
3141                         if (unlikely(!nskb))
3142                                 goto err;
3143 
3144                         if (unlikely(pskb_trim(nskb, len))) {
3145                                 kfree_skb(nskb);
3146                                 goto err;
3147                         }
3148 
3149                         hsize = skb_end_offset(nskb);
3150                         if (skb_cow_head(nskb, doffset + headroom)) {
3151                                 kfree_skb(nskb);
3152                                 goto err;
3153                         }
3154 
3155                         nskb->truesize += skb_end_offset(nskb) - hsize;
3156                         skb_release_head_state(nskb);
3157                         __skb_push(nskb, doffset);
3158                 } else {
3159                         nskb = __alloc_skb(hsize + doffset + headroom,
3160                                            GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3161                                            NUMA_NO_NODE);
3162 
3163                         if (unlikely(!nskb))
3164                                 goto err;
3165 
3166                         skb_reserve(nskb, headroom);
3167                         __skb_put(nskb, doffset);
3168                 }
3169 
3170                 if (segs)
3171                         tail->next = nskb;
3172                 else
3173                         segs = nskb;
3174                 tail = nskb;
3175 
3176                 __copy_skb_header(nskb, head_skb);
3177 
3178                 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3179                 skb_reset_mac_len(nskb);
3180 
3181                 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3182                                                  nskb->data - tnl_hlen,
3183                                                  doffset + tnl_hlen);
3184 
3185                 if (nskb->len == len + doffset)
3186                         goto perform_csum_check;
3187 
3188                 if (!sg) {
3189                         if (!nskb->remcsum_offload)
3190                                 nskb->ip_summed = CHECKSUM_NONE;
3191                         SKB_GSO_CB(nskb)->csum =
3192                                 skb_copy_and_csum_bits(head_skb, offset,
3193                                                        skb_put(nskb, len),
3194                                                        len, 0);
3195                         SKB_GSO_CB(nskb)->csum_start =
3196                                 skb_headroom(nskb) + doffset;
3197                         continue;
3198                 }
3199 
3200                 nskb_frag = skb_shinfo(nskb)->frags;
3201 
3202                 skb_copy_from_linear_data_offset(head_skb, offset,
3203                                                  skb_put(nskb, hsize), hsize);
3204 
3205                 skb_shinfo(nskb)->tx_flags = skb_shinfo(head_skb)->tx_flags &
3206                         SKBTX_SHARED_FRAG;
3207 
3208                 while (pos < offset + len) {
3209                         if (i >= nfrags) {
3210                                 BUG_ON(skb_headlen(list_skb));
3211 
3212                                 i = 0;
3213                                 nfrags = skb_shinfo(list_skb)->nr_frags;
3214                                 frag = skb_shinfo(list_skb)->frags;
3215                                 frag_skb = list_skb;
3216 
3217                                 BUG_ON(!nfrags);
3218 
3219                                 list_skb = list_skb->next;
3220                         }
3221 
3222                         if (unlikely(skb_shinfo(nskb)->nr_frags >=
3223                                      MAX_SKB_FRAGS)) {
3224                                 net_warn_ratelimited(
3225                                         "skb_segment: too many frags: %u %u\n",
3226                                         pos, mss);
3227                                 goto err;
3228                         }
3229 
3230                         if (unlikely(skb_orphan_frags(frag_skb, GFP_ATOMIC)))
3231                                 goto err;
3232 
3233                         *nskb_frag = *frag;
3234                         __skb_frag_ref(nskb_frag);
3235                         size = skb_frag_size(nskb_frag);
3236 
3237                         if (pos < offset) {
3238                                 nskb_frag->page_offset += offset - pos;
3239                                 skb_frag_size_sub(nskb_frag, offset - pos);
3240                         }
3241 
3242                         skb_shinfo(nskb)->nr_frags++;
3243 
3244                         if (pos + size <= offset + len) {
3245                                 i++;
3246                                 frag++;
3247                                 pos += size;
3248                         } else {
3249                                 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
3250                                 goto skip_fraglist;
3251                         }
3252 
3253                         nskb_frag++;
3254                 }
3255 
3256 skip_fraglist:
3257                 nskb->data_len = len - hsize;
3258                 nskb->len += nskb->data_len;
3259                 nskb->truesize += nskb->data_len;
3260 
3261 perform_csum_check:
3262                 if (!csum) {
3263                         if (skb_has_shared_frag(nskb)) {
3264                                 err = __skb_linearize(nskb);
3265                                 if (err)
3266                                         goto err;
3267                         }
3268                         if (!nskb->remcsum_offload)
3269                                 nskb->ip_summed = CHECKSUM_NONE;
3270                         SKB_GSO_CB(nskb)->csum =
3271                                 skb_checksum(nskb, doffset,
3272                                              nskb->len - doffset, 0);
3273                         SKB_GSO_CB(nskb)->csum_start =
3274                                 skb_headroom(nskb) + doffset;
3275                 }
3276         } while ((offset += len) < head_skb->len);
3277 
3278         /* Some callers want to get the end of the list.
3279          * Put it in segs->prev to avoid walking the list.
3280          * (see validate_xmit_skb_list() for example)
3281          */
3282         segs->prev = tail;
3283 
3284         /* Update GSO info on first skb in partial sequence. */
3285         if (partial_segs) {
3286                 int type = skb_shinfo(head_skb)->gso_type;
3287 
3288                 /* Update type to add partial and then remove dodgy if set */
3289                 type |= SKB_GSO_PARTIAL;
3290                 type &= ~SKB_GSO_DODGY;
3291 
3292                 /* Update GSO info and prepare to start updating headers on
3293                  * our way back down the stack of protocols.
3294                  */
3295                 skb_shinfo(segs)->gso_size = skb_shinfo(head_skb)->gso_size;
3296                 skb_shinfo(segs)->gso_segs = partial_segs;
3297                 skb_shinfo(segs)->gso_type = type;
3298                 SKB_GSO_CB(segs)->data_offset = skb_headroom(segs) + doffset;
3299         }
3300 
3301         /* Following permits correct backpressure, for protocols
3302          * using skb_set_owner_w().
3303          * Idea is to tranfert ownership from head_skb to last segment.
3304          */
3305         if (head_skb->destructor == sock_wfree) {
3306                 swap(tail->truesize, head_skb->truesize);
3307                 swap(tail->destructor, head_skb->destructor);
3308                 swap(tail->sk, head_skb->sk);
3309         }
3310         return segs;
3311 
3312 err:
3313         kfree_skb_list(segs);
3314         return ERR_PTR(err);
3315 }
3316 EXPORT_SYMBOL_GPL(skb_segment);
3317 
3318 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
3319 {
3320         struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
3321         unsigned int offset = skb_gro_offset(skb);
3322         unsigned int headlen = skb_headlen(skb);
3323         unsigned int len = skb_gro_len(skb);
3324         struct sk_buff *lp, *p = *head;
3325         unsigned int delta_truesize;
3326 
3327         if (unlikely(p->len + len >= 65536))
3328                 return -E2BIG;
3329 
3330         lp = NAPI_GRO_CB(p)->last;
3331         pinfo = skb_shinfo(lp);
3332 
3333         if (headlen <= offset) {
3334                 skb_frag_t *frag;
3335                 skb_frag_t *frag2;
3336                 int i = skbinfo->nr_frags;
3337                 int nr_frags = pinfo->nr_frags + i;
3338 
3339                 if (nr_frags > MAX_SKB_FRAGS)
3340                         goto merge;
3341 
3342                 offset -= headlen;
3343                 pinfo->nr_frags = nr_frags;
3344                 skbinfo->nr_frags = 0;
3345 
3346                 frag = pinfo->frags + nr_frags;
3347                 frag2 = skbinfo->frags + i;
3348                 do {
3349                         *--frag = *--frag2;
3350                 } while (--i);
3351 
3352                 frag->page_offset += offset;
3353                 skb_frag_size_sub(frag, offset);
3354 
3355                 /* all fragments truesize : remove (head size + sk_buff) */
3356                 delta_truesize = skb->truesize -
3357                                  SKB_TRUESIZE(skb_end_offset(skb));
3358 
3359                 skb->truesize -= skb->data_len;
3360                 skb->len -= skb->data_len;
3361                 skb->data_len = 0;
3362 
3363                 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
3364                 goto done;
3365         } else if (skb->head_frag) {
3366                 int nr_frags = pinfo->nr_frags;
3367                 skb_frag_t *frag = pinfo->frags + nr_frags;
3368                 struct page *page = virt_to_head_page(skb->head);
3369                 unsigned int first_size = headlen - offset;
3370                 unsigned int first_offset;
3371 
3372                 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
3373                         goto merge;
3374 
3375                 first_offset = skb->data -
3376                                (unsigned char *)page_address(page) +
3377                                offset;
3378 
3379                 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3380 
3381                 frag->page.p      = page;
3382                 frag->page_offset = first_offset;
3383                 skb_frag_size_set(frag, first_size);
3384 
3385                 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3386                 /* We dont need to clear skbinfo->nr_frags here */
3387 
3388                 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3389                 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3390                 goto done;
3391         }
3392 
3393 merge:
3394         delta_truesize = skb->truesize;
3395         if (offset > headlen) {
3396                 unsigned int eat = offset - headlen;
3397 
3398                 skbinfo->frags[0].page_offset += eat;
3399                 skb_frag_size_sub(&skbinfo->frags[0], eat);
3400                 skb->data_len -= eat;
3401                 skb->len -= eat;
3402                 offset = headlen;
3403         }
3404 
3405         __skb_pull(skb, offset);
3406 
3407         if (NAPI_GRO_CB(p)->last == p)
3408                 skb_shinfo(p)->frag_list = skb;
3409         else
3410                 NAPI_GRO_CB(p)->last->next = skb;
3411         NAPI_GRO_CB(p)->last = skb;
3412         __skb_header_release(skb);
3413         lp = p;
3414 
3415 done:
3416         NAPI_GRO_CB(p)->count++;
3417         p->data_len += len;
3418         p->truesize += delta_truesize;
3419         p->len += len;
3420         if (lp != p) {
3421                 lp->data_len += len;
3422                 lp->truesize += delta_truesize;
3423                 lp->len += len;
3424         }
3425         NAPI_GRO_CB(skb)->same_flow = 1;
3426         return 0;
3427 }
3428 EXPORT_SYMBOL_GPL(skb_gro_receive);
3429 
3430 void __init skb_init(void)
3431 {
3432         skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3433                                               sizeof(struct sk_buff),
3434                                               0,
3435                                               SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3436                                               NULL);
3437         skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3438                                                 sizeof(struct sk_buff_fclones),
3439                                                 0,
3440                                                 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3441                                                 NULL);
3442 }
3443 
3444 /**
3445  *      skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3446  *      @skb: Socket buffer containing the buffers to be mapped
3447  *      @sg: The scatter-gather list to map into
3448  *      @offset: The offset into the buffer's contents to start mapping
3449  *      @len: Length of buffer space to be mapped
3450  *
3451  *      Fill the specified scatter-gather list with mappings/pointers into a
3452  *      region of the buffer space attached to a socket buffer.
3453  */
3454 static int
3455 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3456 {
3457         int start = skb_headlen(skb);
3458         int i, copy = start - offset;
3459         struct sk_buff *frag_iter;
3460         int elt = 0;
3461 
3462         if (copy > 0) {
3463                 if (copy > len)
3464                         copy = len;
3465                 sg_set_buf(sg, skb->data + offset, copy);
3466                 elt++;
3467                 if ((len -= copy) == 0)
3468                         return elt;
3469                 offset += copy;
3470         }
3471 
3472         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3473                 int end;
3474 
3475                 WARN_ON(start > offset + len);
3476 
3477                 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3478                 if ((copy = end - offset) > 0) {
3479                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3480 
3481                         if (copy > len)
3482                                 copy = len;
3483                         sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3484                                         frag->page_offset+offset-start);
3485                         elt++;
3486                         if (!(len -= copy))
3487                                 return elt;
3488                         offset += copy;
3489                 }
3490                 start = end;
3491         }
3492 
3493         skb_walk_frags(skb, frag_iter) {
3494                 int end;
3495 
3496                 WARN_ON(start > offset + len);
3497 
3498                 end = start + frag_iter->len;
3499                 if ((copy = end - offset) > 0) {
3500                         if (copy > len)
3501                                 copy = len;
3502                         elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3503                                               copy);
3504                         if ((len -= copy) == 0)
3505                                 return elt;
3506                         offset += copy;
3507                 }
3508                 start = end;
3509         }
3510         BUG_ON(len);
3511         return elt;
3512 }
3513 
3514 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
3515  * sglist without mark the sg which contain last skb data as the end.
3516  * So the caller can mannipulate sg list as will when padding new data after
3517  * the first call without calling sg_unmark_end to expend sg list.
3518  *
3519  * Scenario to use skb_to_sgvec_nomark:
3520  * 1. sg_init_table
3521  * 2. skb_to_sgvec_nomark(payload1)
3522  * 3. skb_to_sgvec_nomark(payload2)
3523  *
3524  * This is equivalent to:
3525  * 1. sg_init_table
3526  * 2. skb_to_sgvec(payload1)
3527  * 3. sg_unmark_end
3528  * 4. skb_to_sgvec(payload2)
3529  *
3530  * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
3531  * is more preferable.
3532  */
3533 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
3534                         int offset, int len)
3535 {
3536         return __skb_to_sgvec(skb, sg, offset, len);
3537 }
3538 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
3539 
3540 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3541 {
3542         int nsg = __skb_to_sgvec(skb, sg, offset, len);
3543 
3544         sg_mark_end(&sg[nsg - 1]);
3545 
3546         return nsg;
3547 }
3548 EXPORT_SYMBOL_GPL(skb_to_sgvec);
3549 
3550 /**
3551  *      skb_cow_data - Check that a socket buffer's data buffers are writable
3552  *      @skb: The socket buffer to check.
3553  *      @tailbits: Amount of trailing space to be added
3554  *      @trailer: Returned pointer to the skb where the @tailbits space begins
3555  *
3556  *      Make sure that the data buffers attached to a socket buffer are
3557  *      writable. If they are not, private copies are made of the data buffers
3558  *      and the socket buffer is set to use these instead.
3559  *
3560  *      If @tailbits is given, make sure that there is space to write @tailbits
3561  *      bytes of data beyond current end of socket buffer.  @trailer will be
3562  *      set to point to the skb in which this space begins.
3563  *
3564  *      The number of scatterlist elements required to completely map the
3565  *      COW'd and extended socket buffer will be returned.
3566  */
3567 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3568 {
3569         int copyflag;
3570         int elt;
3571         struct sk_buff *skb1, **skb_p;
3572 
3573         /* If skb is cloned or its head is paged, reallocate
3574          * head pulling out all the pages (pages are considered not writable
3575          * at the moment even if they are anonymous).
3576          */
3577         if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3578             __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3579                 return -ENOMEM;
3580 
3581         /* Easy case. Most of packets will go this way. */
3582         if (!skb_has_frag_list(skb)) {
3583                 /* A little of trouble, not enough of space for trailer.
3584                  * This should not happen, when stack is tuned to generate
3585                  * good frames. OK, on miss we reallocate and reserve even more
3586                  * space, 128 bytes is fair. */
3587 
3588                 if (skb_tailroom(skb) < tailbits &&
3589                     pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3590                         return -ENOMEM;
3591 
3592                 /* Voila! */
3593                 *trailer = skb;
3594                 return 1;
3595         }
3596 
3597         /* Misery. We are in troubles, going to mincer fragments... */
3598 
3599         elt = 1;
3600         skb_p = &skb_shinfo(skb)->frag_list;
3601         copyflag = 0;
3602 
3603         while ((skb1 = *skb_p) != NULL) {
3604                 int ntail = 0;
3605 
3606                 /* The fragment is partially pulled by someone,
3607                  * this can happen on input. Copy it and everything
3608                  * after it. */
3609 
3610                 if (skb_shared(skb1))
3611                         copyflag = 1;
3612 
3613                 /* If the skb is the last, worry about trailer. */
3614 
3615                 if (skb1->next == NULL && tailbits) {
3616                         if (skb_shinfo(skb1)->nr_frags ||
3617                             skb_has_frag_list(skb1) ||
3618                             skb_tailroom(skb1) < tailbits)
3619                                 ntail = tailbits + 128;
3620                 }
3621 
3622                 if (copyflag ||
3623                     skb_cloned(skb1) ||
3624                     ntail ||
3625                     skb_shinfo(skb1)->nr_frags ||
3626                     skb_has_frag_list(skb1)) {
3627                         struct sk_buff *skb2;
3628 
3629                         /* Fuck, we are miserable poor guys... */
3630                         if (ntail == 0)
3631                                 skb2 = skb_copy(skb1, GFP_ATOMIC);
3632                         else
3633                                 skb2 = skb_copy_expand(skb1,
3634                                                        skb_headroom(skb1),
3635                                                        ntail,
3636                                                        GFP_ATOMIC);
3637                         if (unlikely(skb2 == NULL))
3638                                 return -ENOMEM;
3639 
3640                         if (skb1->sk)
3641                                 skb_set_owner_w(skb2, skb1->sk);
3642 
3643                         /* Looking around. Are we still alive?
3644                          * OK, link new skb, drop old one */
3645 
3646                         skb2->next = skb1->next;
3647                         *skb_p = skb2;
3648                         kfree_skb(skb1);
3649                         skb1 = skb2;
3650                 }
3651                 elt++;
3652                 *trailer = skb1;
3653                 skb_p = &skb1->next;
3654         }
3655 
3656         return elt;
3657 }
3658 EXPORT_SYMBOL_GPL(skb_cow_data);
3659 
3660 static void sock_rmem_free(struct sk_buff *skb)
3661 {
3662         struct sock *sk = skb->sk;
3663 
3664         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3665 }
3666 
3667 /*
3668  * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3669  */
3670 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3671 {
3672         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3673             (unsigned int)sk->sk_rcvbuf)
3674                 return -ENOMEM;
3675 
3676         skb_orphan(skb);
3677         skb->sk = sk;
3678         skb->destructor = sock_rmem_free;
3679         atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3680 
3681         /* before exiting rcu section, make sure dst is refcounted */
3682         skb_dst_force(skb);
3683 
3684         skb_queue_tail(&sk->sk_error_queue, skb);
3685         if (!sock_flag(sk, SOCK_DEAD))
3686                 sk->sk_data_ready(sk);
3687         return 0;
3688 }
3689 EXPORT_SYMBOL(sock_queue_err_skb);
3690 
3691 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
3692 {
3693         struct sk_buff_head *q = &sk->sk_error_queue;
3694         struct sk_buff *skb, *skb_next;
3695         unsigned long flags;
3696         int err = 0;
3697 
3698         spin_lock_irqsave(&q->lock, flags);
3699         skb = __skb_dequeue(q);
3700         if (skb && (skb_next = skb_peek(q)))
3701                 err = SKB_EXT_ERR(skb_next)->ee.ee_errno;
3702         spin_unlock_irqrestore(&q->lock, flags);
3703 
3704         sk->sk_err = err;
3705         if (err)
3706                 sk->sk_error_report(sk);
3707 
3708         return skb;
3709 }
3710 EXPORT_SYMBOL(sock_dequeue_err_skb);
3711 
3712 /**
3713  * skb_clone_sk - create clone of skb, and take reference to socket
3714  * @skb: the skb to clone
3715  *
3716  * This function creates a clone of a buffer that holds a reference on
3717  * sk_refcnt.  Buffers created via this function are meant to be
3718  * returned using sock_queue_err_skb, or free via kfree_skb.
3719  *
3720  * When passing buffers allocated with this function to sock_queue_err_skb
3721  * it is necessary to wrap the call with sock_hold/sock_put in order to
3722  * prevent the socket from being released prior to being enqueued on
3723  * the sk_error_queue.
3724  */
3725 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
3726 {
3727         struct sock *sk = skb->sk;
3728         struct sk_buff *clone;
3729 
3730         if (!sk || !atomic_inc_not_zero(&sk->sk_refcnt))
3731                 return NULL;
3732 
3733         clone = skb_clone(skb, GFP_ATOMIC);
3734         if (!clone) {
3735                 sock_put(sk);
3736                 return NULL;
3737         }
3738 
3739         clone->sk = sk;
3740         clone->destructor = sock_efree;
3741 
3742         return clone;
3743 }
3744 EXPORT_SYMBOL(skb_clone_sk);
3745 
3746 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
3747                                         struct sock *sk,
3748                                         int tstype)
3749 {
3750         struct sock_exterr_skb *serr;
3751         int err;
3752 
3753         serr = SKB_EXT_ERR(skb);
3754         memset(serr, 0, sizeof(*serr));
3755         serr->ee.ee_errno = ENOMSG;
3756         serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3757         serr->ee.ee_info = tstype;
3758         if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
3759                 serr->ee.ee_data = skb_shinfo(skb)->tskey;
3760                 if (sk->sk_protocol == IPPROTO_TCP &&
3761                     sk->sk_type == SOCK_STREAM)
3762                         serr->ee.ee_data -= sk->sk_tskey;
3763         }
3764 
3765         err = sock_queue_err_skb(sk, skb);
3766 
3767         if (err)
3768                 kfree_skb(skb);
3769 }
3770 
3771 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
3772 {
3773         bool ret;
3774 
3775         if (likely(sysctl_tstamp_allow_data || tsonly))
3776                 return true;
3777 
3778         read_lock_bh(&sk->sk_callback_lock);
3779         ret = sk->sk_socket && sk->sk_socket->file &&
3780               file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
3781         read_unlock_bh(&sk->sk_callback_lock);
3782         return ret;
3783 }
3784 
3785 void skb_complete_tx_timestamp(struct sk_buff *skb,
3786                                struct skb_shared_hwtstamps *hwtstamps)
3787 {
3788         struct sock *sk = skb->sk;
3789 
3790         if (!skb_may_tx_timestamp(sk, false))
3791                 return;
3792 
3793         /* take a reference to prevent skb_orphan() from freeing the socket */
3794         sock_hold(sk);
3795 
3796         *skb_hwtstamps(skb) = *hwtstamps;
3797         __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND);
3798 
3799         sock_put(sk);
3800 }
3801 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
3802 
3803 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3804                      struct skb_shared_hwtstamps *hwtstamps,
3805                      struct sock *sk, int tstype)
3806 {
3807         struct sk_buff *skb;
3808         bool tsonly;
3809 
3810         if (!sk)
3811                 return;
3812 
3813         tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
3814         if (!skb_may_tx_timestamp(sk, tsonly))
3815                 return;
3816 
3817         if (tsonly)
3818                 skb = alloc_skb(0, GFP_ATOMIC);
3819         else
3820                 skb = skb_clone(orig_skb, GFP_ATOMIC);
3821         if (!skb)
3822                 return;
3823 
3824         if (tsonly) {
3825                 skb_shinfo(skb)->tx_flags = skb_shinfo(orig_skb)->tx_flags;
3826                 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
3827         }
3828 
3829         if (hwtstamps)
3830                 *skb_hwtstamps(skb) = *hwtstamps;
3831         else
3832                 skb->tstamp = ktime_get_real();
3833 
3834         __skb_complete_tx_timestamp(skb, sk, tstype);
3835 }
3836 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
3837 
3838 void skb_tstamp_tx(struct sk_buff *orig_skb,
3839                    struct skb_shared_hwtstamps *hwtstamps)
3840 {
3841         return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
3842                                SCM_TSTAMP_SND);
3843 }
3844 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3845 
3846 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3847 {
3848         struct sock *sk = skb->sk;
3849         struct sock_exterr_skb *serr;
3850         int err;
3851 
3852         skb->wifi_acked_valid = 1;
3853         skb->wifi_acked = acked;
3854 
3855         serr = SKB_EXT_ERR(skb);
3856         memset(serr, 0, sizeof(*serr));
3857         serr->ee.ee_errno = ENOMSG;
3858         serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3859 
3860         /* take a reference to prevent skb_orphan() from freeing the socket */
3861         sock_hold(sk);
3862 
3863         err = sock_queue_err_skb(sk, skb);
3864         if (err)
3865                 kfree_skb(skb);
3866 
3867         sock_put(sk);
3868 }
3869 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3870 
3871 /**
3872  * skb_partial_csum_set - set up and verify partial csum values for packet
3873  * @skb: the skb to set
3874  * @start: the number of bytes after skb->data to start checksumming.
3875  * @off: the offset from start to place the checksum.
3876  *
3877  * For untrusted partially-checksummed packets, we need to make sure the values
3878  * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3879  *
3880  * This function checks and sets those values and skb->ip_summed: if this
3881  * returns false you should drop the packet.
3882  */
3883 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3884 {
3885         if (unlikely(start > skb_headlen(skb)) ||
3886             unlikely((int)start + off > skb_headlen(skb) - 2)) {
3887                 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3888                                      start, off, skb_headlen(skb));
3889                 return false;
3890         }
3891         skb->ip_summed = CHECKSUM_PARTIAL;
3892         skb->csum_start = skb_headroom(skb) + start;
3893         skb->csum_offset = off;
3894         skb_set_transport_header(skb, start);
3895         return true;
3896 }
3897 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3898 
3899 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
3900                                unsigned int max)
3901 {
3902         if (skb_headlen(skb) >= len)
3903                 return 0;
3904 
3905         /* If we need to pullup then pullup to the max, so we
3906          * won't need to do it again.
3907          */
3908         if (max > skb->len)
3909                 max = skb->len;
3910 
3911         if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
3912                 return -ENOMEM;
3913 
3914         if (skb_headlen(skb) < len)
3915                 return -EPROTO;
3916 
3917         return 0;
3918 }
3919 
3920 #define MAX_TCP_HDR_LEN (15 * 4)
3921 
3922 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
3923                                       typeof(IPPROTO_IP) proto,
3924                                       unsigned int off)
3925 {
3926         switch (proto) {
3927                 int err;
3928 
3929         case IPPROTO_TCP:
3930                 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
3931                                           off + MAX_TCP_HDR_LEN);
3932                 if (!err && !skb_partial_csum_set(skb, off,
3933                                                   offsetof(struct tcphdr,
3934                                                            check)))
3935                         err = -EPROTO;
3936                 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
3937 
3938         case IPPROTO_UDP:
3939                 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
3940                                           off + sizeof(struct udphdr));
3941                 if (!err && !skb_partial_csum_set(skb, off,
3942                                                   offsetof(struct udphdr,
3943                                                            check)))
3944                         err = -EPROTO;
3945                 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
3946         }
3947 
3948         return ERR_PTR(-EPROTO);
3949 }
3950 
3951 /* This value should be large enough to cover a tagged ethernet header plus
3952  * maximally sized IP and TCP or UDP headers.
3953  */
3954 #define MAX_IP_HDR_LEN 128
3955 
3956 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
3957 {
3958         unsigned int off;
3959         bool fragment;
3960         __sum16 *csum;
3961         int err;
3962 
3963         fragment = false;
3964 
3965         err = skb_maybe_pull_tail(skb,
3966                                   sizeof(struct iphdr),
3967                                   MAX_IP_HDR_LEN);
3968         if (err < 0)
3969                 goto out;
3970 
3971         if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
3972                 fragment = true;
3973 
3974         off = ip_hdrlen(skb);
3975 
3976         err = -EPROTO;
3977 
3978         if (fragment)
3979                 goto out;
3980 
3981         csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
3982         if (IS_ERR(csum))
3983                 return PTR_ERR(csum);
3984 
3985         if (recalculate)
3986                 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3987                                            ip_hdr(skb)->daddr,
3988                                            skb->len - off,
3989                                            ip_hdr(skb)->protocol, 0);
3990         err = 0;
3991 
3992 out:
3993         return err;
3994 }
3995 
3996 /* This value should be large enough to cover a tagged ethernet header plus
3997  * an IPv6 header, all options, and a maximal TCP or UDP header.
3998  */
3999 #define MAX_IPV6_HDR_LEN 256
4000 
4001 #define OPT_HDR(type, skb, off) \
4002         (type *)(skb_network_header(skb) + (off))
4003 
4004 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
4005 {
4006         int err;
4007         u8 nexthdr;
4008         unsigned int off;
4009         unsigned int len;
4010         bool fragment;
4011         bool done;
4012         __sum16 *csum;
4013 
4014         fragment = false;
4015         done = false;
4016 
4017         off = sizeof(struct ipv6hdr);
4018 
4019         err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4020         if (err < 0)
4021                 goto out;
4022 
4023         nexthdr = ipv6_hdr(skb)->nexthdr;
4024 
4025         len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4026         while (off <= len && !done) {
4027                 switch (nexthdr) {
4028                 case IPPROTO_DSTOPTS:
4029                 case IPPROTO_HOPOPTS:
4030                 case IPPROTO_ROUTING: {
4031                         struct ipv6_opt_hdr *hp;
4032 
4033                         err = skb_maybe_pull_tail(skb,
4034                                                   off +
4035                                                   sizeof(struct ipv6_opt_hdr),
4036                                                   MAX_IPV6_HDR_LEN);
4037                         if (err < 0)
4038                                 goto out;
4039 
4040                         hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
4041                         nexthdr = hp->nexthdr;
4042                         off += ipv6_optlen(hp);
4043                         break;
4044                 }
4045                 case IPPROTO_AH: {
4046                         struct ip_auth_hdr *hp;
4047 
4048                         err = skb_maybe_pull_tail(skb,
4049                                                   off +
4050                                                   sizeof(struct ip_auth_hdr),
4051                                                   MAX_IPV6_HDR_LEN);
4052                         if (err < 0)
4053                                 goto out;
4054 
4055                         hp = OPT_HDR(struct ip_auth_hdr, skb, off);
4056                         nexthdr = hp->nexthdr;
4057                         off += ipv6_authlen(hp);
4058                         break;
4059                 }
4060                 case IPPROTO_FRAGMENT: {
4061                         struct frag_hdr *hp;
4062 
4063                         err = skb_maybe_pull_tail(skb,
4064                                                   off +
4065                                                   sizeof(struct frag_hdr),
4066                                                   MAX_IPV6_HDR_LEN);
4067                         if (err < 0)
4068                                 goto out;
4069 
4070                         hp = OPT_HDR(struct frag_hdr, skb, off);
4071 
4072                         if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
4073                                 fragment = true;
4074 
4075                         nexthdr = hp->nexthdr;
4076                         off += sizeof(struct frag_hdr);
4077                         break;
4078                 }
4079                 default:
4080                         done = true;
4081                         break;
4082                 }
4083         }
4084 
4085         err = -EPROTO;
4086 
4087         if (!done || fragment)
4088                 goto out;
4089 
4090         csum = skb_checksum_setup_ip(skb, nexthdr, off);
4091         if (IS_ERR(csum))
4092                 return PTR_ERR(csum);
4093 
4094         if (recalculate)
4095                 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4096                                          &ipv6_hdr(skb)->daddr,
4097                                          skb->len - off, nexthdr, 0);
4098         err = 0;
4099 
4100 out:
4101         return err;
4102 }
4103 
4104 /**
4105  * skb_checksum_setup - set up partial checksum offset
4106  * @skb: the skb to set up
4107  * @recalculate: if true the pseudo-header checksum will be recalculated
4108  */
4109 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
4110 {
4111         int err;
4112 
4113         switch (skb->protocol) {
4114         case htons(ETH_P_IP):
4115                 err = skb_checksum_setup_ipv4(skb, recalculate);
4116                 break;
4117 
4118         case htons(ETH_P_IPV6):
4119                 err = skb_checksum_setup_ipv6(skb, recalculate);
4120                 break;
4121 
4122         default:
4123                 err = -EPROTO;
4124                 break;
4125         }
4126 
4127         return err;
4128 }
4129 EXPORT_SYMBOL(skb_checksum_setup);
4130 
4131 /**
4132  * skb_checksum_maybe_trim - maybe trims the given skb
4133  * @skb: the skb to check
4134  * @transport_len: the data length beyond the network header
4135  *
4136  * Checks whether the given skb has data beyond the given transport length.
4137  * If so, returns a cloned skb trimmed to this transport length.
4138  * Otherwise returns the provided skb. Returns NULL in error cases
4139  * (e.g. transport_len exceeds skb length or out-of-memory).
4140  *
4141  * Caller needs to set the skb transport header and free any returned skb if it
4142  * differs from the provided skb.
4143  */
4144 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
4145                                                unsigned int transport_len)
4146 {
4147         struct sk_buff *skb_chk;
4148         unsigned int len = skb_transport_offset(skb) + transport_len;
4149         int ret;
4150 
4151         if (skb->len < len)
4152                 return NULL;
4153         else if (skb->len == len)
4154                 return skb;
4155 
4156         skb_chk = skb_clone(skb, GFP_ATOMIC);
4157         if (!skb_chk)
4158                 return NULL;
4159 
4160         ret = pskb_trim_rcsum(skb_chk, len);
4161         if (ret) {
4162                 kfree_skb(skb_chk);
4163                 return NULL;
4164         }
4165 
4166         return skb_chk;
4167 }
4168 
4169 /**
4170  * skb_checksum_trimmed - validate checksum of an skb
4171  * @skb: the skb to check
4172  * @transport_len: the data length beyond the network header
4173  * @skb_chkf: checksum function to use
4174  *
4175  * Applies the given checksum function skb_chkf to the provided skb.
4176  * Returns a checked and maybe trimmed skb. Returns NULL on error.
4177  *
4178  * If the skb has data beyond the given transport length, then a
4179  * trimmed & cloned skb is checked and returned.
4180  *
4181  * Caller needs to set the skb transport header and free any returned skb if it
4182  * differs from the provided skb.
4183  */
4184 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4185                                      unsigned int transport_len,
4186                                      __sum16(*skb_chkf)(struct sk_buff *skb))
4187 {
4188         struct sk_buff *skb_chk;
4189         unsigned int offset = skb_transport_offset(skb);
4190         __sum16 ret;
4191 
4192         skb_chk = skb_checksum_maybe_trim(skb, transport_len);
4193         if (!skb_chk)
4194                 goto err;
4195 
4196         if (!pskb_may_pull(skb_chk, offset))
4197                 goto err;
4198 
4199         skb_pull_rcsum(skb_chk, offset);
4200         ret = skb_chkf(skb_chk);
4201         skb_push_rcsum(skb_chk, offset);
4202 
4203         if (ret)
4204                 goto err;
4205 
4206         return skb_chk;
4207 
4208 err:
4209         if (skb_chk && skb_chk != skb)
4210                 kfree_skb(skb_chk);
4211 
4212         return NULL;
4213 
4214 }
4215 EXPORT_SYMBOL(skb_checksum_trimmed);
4216 
4217 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
4218 {
4219         net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
4220                              skb->dev->name);
4221 }
4222 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
4223 
4224 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
4225 {
4226         if (head_stolen) {
4227                 skb_release_head_state(skb);
4228                 kmem_cache_free(skbuff_head_cache, skb);
4229         } else {
4230                 __kfree_skb(skb);
4231         }
4232 }
4233 EXPORT_SYMBOL(kfree_skb_partial);
4234 
4235 /**
4236  * skb_try_coalesce - try to merge skb to prior one
4237  * @to: prior buffer
4238  * @from: buffer to add
4239  * @fragstolen: pointer to boolean
4240  * @delta_truesize: how much more was allocated than was requested
4241  */
4242 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
4243                       bool *fragstolen, int *delta_truesize)
4244 {
4245         int i, delta, len = from->len;
4246 
4247         *fragstolen = false;
4248 
4249         if (skb_cloned(to))
4250                 return false;
4251 
4252         if (len <= skb_tailroom(to)) {
4253                 if (len)
4254                         BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
4255                 *delta_truesize = 0;
4256                 return true;
4257         }
4258 
4259         if (skb_has_frag_list(to) || skb_has_frag_list(from))
4260                 return false;
4261 
4262         if (skb_headlen(from) != 0) {
4263                 struct page *page;
4264                 unsigned int offset;
4265 
4266                 if (skb_shinfo(to)->nr_frags +
4267                     skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
4268                         return false;
4269 
4270                 if (skb_head_is_locked(from))
4271                         return false;
4272 
4273                 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4274 
4275                 page = virt_to_head_page(from->head);
4276                 offset = from->data - (unsigned char *)page_address(page);
4277 
4278                 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
4279                                    page, offset, skb_headlen(from));
4280                 *fragstolen = true;
4281         } else {
4282                 if (skb_shinfo(to)->nr_frags +
4283                     skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
4284                         return false;
4285 
4286                 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
4287         }
4288 
4289         WARN_ON_ONCE(delta < len);
4290 
4291         memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
4292                skb_shinfo(from)->frags,
4293                skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
4294         skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
4295 
4296         if (!skb_cloned(from))
4297                 skb_shinfo(from)->nr_frags = 0;
4298 
4299         /* if the skb is not cloned this does nothing
4300          * since we set nr_frags to 0.
4301          */
4302         for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
4303                 skb_frag_ref(from, i);
4304 
4305         to->truesize += delta;
4306         to->len += len;
4307         to->data_len += len;
4308 
4309         *delta_truesize = delta;
4310         return true;
4311 }
4312 EXPORT_SYMBOL(skb_try_coalesce);
4313 
4314 /**
4315  * skb_scrub_packet - scrub an skb
4316  *
4317  * @skb: buffer to clean
4318  * @xnet: packet is crossing netns
4319  *
4320  * skb_scrub_packet can be used after encapsulating or decapsulting a packet
4321  * into/from a tunnel. Some information have to be cleared during these
4322  * operations.
4323  * skb_scrub_packet can also be used to clean a skb before injecting it in
4324  * another namespace (@xnet == true). We have to clear all information in the
4325  * skb that could impact namespace isolation.
4326  */
4327 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
4328 {
4329         skb->tstamp.tv64 = 0;
4330         skb->pkt_type = PACKET_HOST;
4331         skb->skb_iif = 0;
4332         skb->ignore_df = 0;
4333         skb_dst_drop(skb);
4334         secpath_reset(skb);
4335         nf_reset(skb);
4336         nf_reset_trace(skb);
4337 
4338         if (!xnet)
4339                 return;
4340 
4341         skb_orphan(skb);
4342         skb->mark = 0;
4343 }
4344 EXPORT_SYMBOL_GPL(skb_scrub_packet);
4345 
4346 /**
4347  * skb_gso_transport_seglen - Return length of individual segments of a gso packet
4348  *
4349  * @skb: GSO skb
4350  *
4351  * skb_gso_transport_seglen is used to determine the real size of the
4352  * individual segments, including Layer4 headers (TCP/UDP).
4353  *
4354  * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
4355  */
4356 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
4357 {
4358         const struct skb_shared_info *shinfo = skb_shinfo(skb);
4359         unsigned int thlen = 0;
4360 
4361         if (skb->encapsulation) {
4362                 thlen = skb_inner_transport_header(skb) -
4363                         skb_transport_header(skb);
4364 
4365                 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
4366                         thlen += inner_tcp_hdrlen(skb);
4367         } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
4368                 thlen = tcp_hdrlen(skb);
4369         } else if (unlikely(shinfo->gso_type & SKB_GSO_SCTP)) {
4370                 thlen = sizeof(struct sctphdr);
4371         }
4372         /* UFO sets gso_size to the size of the fragmentation
4373          * payload, i.e. the size of the L4 (UDP) header is already
4374          * accounted for.
4375          */
4376         return thlen + shinfo->gso_size;
4377 }
4378 EXPORT_SYMBOL_GPL(skb_gso_transport_seglen);
4379 
4380 /**
4381  * skb_gso_validate_mtu - Return in case such skb fits a given MTU
4382  *
4383  * @skb: GSO skb
4384  * @mtu: MTU to validate against
4385  *
4386  * skb_gso_validate_mtu validates if a given skb will fit a wanted MTU
4387  * once split.
4388  */
4389 bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu)
4390 {
4391         const struct skb_shared_info *shinfo = skb_shinfo(skb);
4392         const struct sk_buff *iter;
4393         unsigned int hlen;
4394 
4395         hlen = skb_gso_network_seglen(skb);
4396 
4397         if (shinfo->gso_size != GSO_BY_FRAGS)
4398                 return hlen <= mtu;
4399 
4400         /* Undo this so we can re-use header sizes */
4401         hlen -= GSO_BY_FRAGS;
4402 
4403         skb_walk_frags(skb, iter) {
4404                 if (hlen + skb_headlen(iter) > mtu)
4405                         return false;
4406         }
4407 
4408         return true;
4409 }
4410 EXPORT_SYMBOL_GPL(skb_gso_validate_mtu);
4411 
4412 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
4413 {
4414         if (skb_cow(skb, skb_headroom(skb)) < 0) {
4415                 kfree_skb(skb);
4416                 return NULL;
4417         }
4418 
4419         memmove(skb->data - ETH_HLEN, skb->data - skb->mac_len - VLAN_HLEN,
4420                 2 * ETH_ALEN);
4421         skb->mac_header += VLAN_HLEN;
4422         return skb;
4423 }
4424 
4425 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
4426 {
4427         struct vlan_hdr *vhdr;
4428         u16 vlan_tci;
4429 
4430         if (unlikely(skb_vlan_tag_present(skb))) {
4431                 /* vlan_tci is already set-up so leave this for another time */
4432                 return skb;
4433         }
4434 
4435         skb = skb_share_check(skb, GFP_ATOMIC);
4436         if (unlikely(!skb))
4437                 goto err_free;
4438 
4439         if (unlikely(!pskb_may_pull(skb, VLAN_HLEN)))
4440                 goto err_free;
4441 
4442         vhdr = (struct vlan_hdr *)skb->data;
4443         vlan_tci = ntohs(vhdr->h_vlan_TCI);
4444         __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
4445 
4446         skb_pull_rcsum(skb, VLAN_HLEN);
4447         vlan_set_encap_proto(skb, vhdr);
4448 
4449         skb = skb_reorder_vlan_header(skb);
4450         if (unlikely(!skb))
4451                 goto err_free;
4452 
4453         skb_reset_network_header(skb);
4454         skb_reset_transport_header(skb);
4455         skb_reset_mac_len(skb);
4456 
4457         return skb;
4458 
4459 err_free:
4460         kfree_skb(skb);
4461         return NULL;
4462 }
4463 EXPORT_SYMBOL(skb_vlan_untag);
4464 
4465 int skb_ensure_writable(struct sk_buff *skb, int write_len)
4466 {
4467         if (!pskb_may_pull(skb, write_len))
4468                 return -ENOMEM;
4469 
4470         if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
4471                 return 0;
4472 
4473         return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4474 }
4475 EXPORT_SYMBOL(skb_ensure_writable);
4476 
4477 /* remove VLAN header from packet and update csum accordingly. */
4478 static int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
4479 {
4480         struct vlan_hdr *vhdr;
4481         unsigned int offset = skb->data - skb_mac_header(skb);
4482         int err;
4483 
4484         __skb_push(skb, offset);
4485         err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
4486         if (unlikely(err))
4487                 goto pull;
4488 
4489         skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
4490 
4491         vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
4492         *vlan_tci = ntohs(vhdr->h_vlan_TCI);
4493 
4494         memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
4495         __skb_pull(skb, VLAN_HLEN);
4496 
4497         vlan_set_encap_proto(skb, vhdr);
4498         skb->mac_header += VLAN_HLEN;
4499 
4500         if (skb_network_offset(skb) < ETH_HLEN)
4501                 skb_set_network_header(skb, ETH_HLEN);
4502 
4503         skb_reset_mac_len(skb);
4504 pull:
4505         __skb_pull(skb, offset);
4506 
4507         return err;
4508 }
4509 
4510 int skb_vlan_pop(struct sk_buff *skb)
4511 {
4512         u16 vlan_tci;
4513         __be16 vlan_proto;
4514         int err;
4515 
4516         if (likely(skb_vlan_tag_present(skb))) {
4517                 skb->vlan_tci = 0;
4518         } else {
4519                 if (unlikely((skb->protocol != htons(ETH_P_8021Q) &&
4520                               skb->protocol != htons(ETH_P_8021AD)) ||
4521                              skb->len < VLAN_ETH_HLEN))
4522                         return 0;
4523 
4524                 err = __skb_vlan_pop(skb, &vlan_tci);
4525                 if (err)
4526                         return err;
4527         }
4528         /* move next vlan tag to hw accel tag */
4529         if (likely((skb->protocol != htons(ETH_P_8021Q) &&
4530                     skb->protocol != htons(ETH_P_8021AD)) ||
4531                    skb->len < VLAN_ETH_HLEN))
4532                 return 0;
4533 
4534         vlan_proto = skb->protocol;
4535         err = __skb_vlan_pop(skb, &vlan_tci);
4536         if (unlikely(err))
4537                 return err;
4538 
4539         __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
4540         return 0;
4541 }
4542 EXPORT_SYMBOL(skb_vlan_pop);
4543 
4544 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
4545 {
4546         if (skb_vlan_tag_present(skb)) {
4547                 unsigned int offset = skb->data - skb_mac_header(skb);
4548                 int err;
4549 
4550                 /* __vlan_insert_tag expect skb->data pointing to mac header.
4551                  * So change skb->data before calling it and change back to
4552                  * original position later
4553                  */
4554                 __skb_push(skb, offset);
4555                 err = __vlan_insert_tag(skb, skb->vlan_proto,
4556                                         skb_vlan_tag_get(skb));
4557                 if (err) {
4558                         __skb_pull(skb, offset);
4559                         return err;
4560                 }
4561 
4562                 skb->protocol = skb->vlan_proto;
4563                 skb->mac_len += VLAN_HLEN;
4564 
4565                 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
4566                 __skb_pull(skb, offset);
4567         }
4568         __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
4569         return 0;
4570 }
4571 EXPORT_SYMBOL(skb_vlan_push);
4572 
4573 /**
4574  * alloc_skb_with_frags - allocate skb with page frags
4575  *
4576  * @header_len: size of linear part
4577  * @data_len: needed length in frags
4578  * @max_page_order: max page order desired.
4579  * @errcode: pointer to error code if any
4580  * @gfp_mask: allocation mask
4581  *
4582  * This can be used to allocate a paged skb, given a maximal order for frags.
4583  */
4584 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
4585                                      unsigned long data_len,
4586                                      int max_page_order,
4587                                      int *errcode,
4588                                      gfp_t gfp_mask)
4589 {
4590         int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
4591         unsigned long chunk;
4592         struct sk_buff *skb;
4593         struct page *page;
4594         gfp_t gfp_head;
4595         int i;
4596 
4597         *errcode = -EMSGSIZE;
4598         /* Note this test could be relaxed, if we succeed to allocate
4599          * high order pages...
4600          */
4601         if (npages > MAX_SKB_FRAGS)
4602                 return NULL;
4603 
4604         gfp_head = gfp_mask;
4605         if (gfp_head & __GFP_DIRECT_RECLAIM)
4606                 gfp_head |= __GFP_REPEAT;
4607 
4608         *errcode = -ENOBUFS;
4609         skb = alloc_skb(header_len, gfp_head);
4610         if (!skb)
4611                 return NULL;
4612 
4613         skb->truesize += npages << PAGE_SHIFT;
4614 
4615         for (i = 0; npages > 0; i++) {
4616                 int order = max_page_order;
4617 
4618                 while (order) {
4619                         if (npages >= 1 << order) {
4620                                 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
4621                                                    __GFP_COMP |
4622                                                    __GFP_NOWARN |
4623                                                    __GFP_NORETRY,
4624                                                    order);
4625                                 if (page)
4626                                         goto fill_page;
4627                                 /* Do not retry other high order allocations */
4628                                 order = 1;
4629                                 max_page_order = 0;
4630                         }
4631                         order--;
4632                 }
4633                 page = alloc_page(gfp_mask);
4634                 if (!page)
4635                         goto failure;
4636 fill_page:
4637                 chunk = min_t(unsigned long, data_len,
4638                               PAGE_SIZE << order);
4639                 skb_fill_page_desc(skb, i, page, 0, chunk);
4640                 data_len -= chunk;
4641                 npages -= 1 << order;
4642         }
4643         return skb;
4644 
4645 failure:
4646         kfree_skb(skb);
4647         return NULL;
4648 }
4649 EXPORT_SYMBOL(alloc_skb_with_frags);
4650 
4651 /* carve out the first off bytes from skb when off < headlen */
4652 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
4653                                     const int headlen, gfp_t gfp_mask)
4654 {
4655         int i;
4656         int size = skb_end_offset(skb);
4657         int new_hlen = headlen - off;
4658         u8 *data;
4659 
4660         size = SKB_DATA_ALIGN(size);
4661 
4662         if (skb_pfmemalloc(skb))
4663                 gfp_mask |= __GFP_MEMALLOC;
4664         data = kmalloc_reserve(size +
4665                                SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
4666                                gfp_mask, NUMA_NO_NODE, NULL);
4667         if (!data)
4668                 return -ENOMEM;
4669 
4670         size = SKB_WITH_OVERHEAD(ksize(data));
4671 
4672         /* Copy real data, and all frags */
4673         skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
4674         skb->len -= off;
4675 
4676         memcpy((struct skb_shared_info *)(data + size),
4677                skb_shinfo(skb),
4678                offsetof(struct skb_shared_info,
4679                         frags[skb_shinfo(skb)->nr_frags]));
4680         if (skb_cloned(skb)) {
4681                 /* drop the old head gracefully */
4682                 if (skb_orphan_frags(skb, gfp_mask)) {
4683                         kfree(data);
4684                         return -ENOMEM;
4685                 }
4686                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
4687                         skb_frag_ref(skb, i);
4688                 if (skb_has_frag_list(skb))
4689                         skb_clone_fraglist(skb);
4690                 skb_release_data(skb);
4691         } else {
4692                 /* we can reuse existing recount- all we did was
4693                  * relocate values
4694                  */
4695                 skb_free_head(skb);
4696         }
4697 
4698         skb->head = data;
4699         skb->data = data;
4700         skb->head_frag = 0;
4701 #ifdef NET_SKBUFF_DATA_USES_OFFSET
4702         skb->end = size;
4703 #else
4704         skb->end = skb->head + size;
4705 #endif
4706         skb_set_tail_pointer(skb, skb_headlen(skb));
4707         skb_headers_offset_update(skb, 0);
4708         skb->cloned = 0;
4709         skb->hdr_len = 0;
4710         skb->nohdr = 0;
4711         atomic_set(&skb_shinfo(skb)->dataref, 1);
4712 
4713         return 0;
4714 }
4715 
4716 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
4717 
4718 /* carve out the first eat bytes from skb's frag_list. May recurse into
4719  * pskb_carve()
4720  */
4721 static int pskb_carve_frag_list(struct sk_buff *skb,
4722                                 struct skb_shared_info *shinfo, int eat,
4723                                 gfp_t gfp_mask)
4724 {
4725         struct sk_buff *list = shinfo->frag_list;
4726         struct sk_buff *clone = NULL;
4727         struct sk_buff *insp = NULL;
4728 
4729         do {
4730                 if (!list) {
4731                         pr_err("Not enough bytes to eat. Want %d\n", eat);
4732                         return -EFAULT;
4733                 }
4734                 if (list->len <= eat) {
4735                         /* Eaten as whole. */
4736                         eat -= list->len;
4737                         list = list->next;
4738                         insp = list;
4739                 } else {
4740                         /* Eaten partially. */
4741                         if (skb_shared(list)) {
4742                                 clone = skb_clone(list, gfp_mask);
4743                                 if (!clone)
4744                                         return -ENOMEM;
4745                                 insp = list->next;
4746                                 list = clone;
4747                         } else {
4748                                 /* This may be pulled without problems. */
4749                                 insp = list;
4750                         }
4751                         if (pskb_carve(list, eat, gfp_mask) < 0) {
4752                                 kfree_skb(clone);
4753                                 return -ENOMEM;
4754                         }
4755                         break;
4756                 }
4757         } while (eat);
4758 
4759         /* Free pulled out fragments. */
4760         while ((list = shinfo->frag_list) != insp) {
4761                 shinfo->frag_list = list->next;
4762                 kfree_skb(list);
4763         }
4764         /* And insert new clone at head. */
4765         if (clone) {
4766                 clone->next = list;
4767                 shinfo->frag_list = clone;
4768         }
4769         return 0;
4770 }
4771 
4772 /* carve off first len bytes from skb. Split line (off) is in the
4773  * non-linear part of skb
4774  */
4775 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
4776                                        int pos, gfp_t gfp_mask)
4777 {
4778         int i, k = 0;
4779         int size = skb_end_offset(skb);
4780         u8 *data;
4781         const int nfrags = skb_shinfo(skb)->nr_frags;
4782         struct skb_shared_info *shinfo;
4783 
4784         size = SKB_DATA_ALIGN(size);
4785 
4786         if (skb_pfmemalloc(skb))
4787                 gfp_mask |= __GFP_MEMALLOC;
4788         data = kmalloc_reserve(size +
4789                                SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
4790                                gfp_mask, NUMA_NO_NODE, NULL);
4791         if (!data)
4792                 return -ENOMEM;
4793 
4794         size = SKB_WITH_OVERHEAD(ksize(data));
4795 
4796         memcpy((struct skb_shared_info *)(data + size),
4797                skb_shinfo(skb), offsetof(struct skb_shared_info,
4798                                          frags[skb_shinfo(skb)->nr_frags]));
4799         if (skb_orphan_frags(skb, gfp_mask)) {
4800                 kfree(data);
4801                 return -ENOMEM;
4802         }
4803         shinfo = (struct skb_shared_info *)(data + size);
4804         for (i = 0; i < nfrags; i++) {
4805                 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
4806 
4807                 if (pos + fsize > off) {
4808                         shinfo->frags[k] = skb_shinfo(skb)->frags[i];
4809 
4810                         if (pos < off) {
4811                                 /* Split frag.
4812                                  * We have two variants in this case:
4813                                  * 1. Move all the frag to the second
4814                                  *    part, if it is possible. F.e.
4815                                  *    this approach is mandatory for TUX,
4816                                  *    where splitting is expensive.
4817                                  * 2. Split is accurately. We make this.
4818                                  */
4819                                 shinfo->frags[0].page_offset += off - pos;
4820                                 skb_frag_size_sub(&shinfo->frags[0], off - pos);
4821                         }
4822                         skb_frag_ref(skb, i);
4823                         k++;
4824                 }
4825                 pos += fsize;
4826         }
4827         shinfo->nr_frags = k;
4828         if (skb_has_frag_list(skb))
4829                 skb_clone_fraglist(skb);
4830 
4831         if (k == 0) {
4832                 /* split line is in frag list */
4833                 pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask);
4834         }
4835         skb_release_data(skb);
4836 
4837         skb->head = data;
4838         skb->head_frag = 0;
4839         skb->data = data;
4840 #ifdef NET_SKBUFF_DATA_USES_OFFSET
4841         skb->end = size;
4842 #else
4843         skb->end = skb->head + size;
4844 #endif
4845         skb_reset_tail_pointer(skb);
4846         skb_headers_offset_update(skb, 0);
4847         skb->cloned   = 0;
4848         skb->hdr_len  = 0;
4849         skb->nohdr    = 0;
4850         skb->len -= off;
4851         skb->data_len = skb->len;
4852         atomic_set(&skb_shinfo(skb)->dataref, 1);
4853         return 0;
4854 }
4855 
4856 /* remove len bytes from the beginning of the skb */
4857 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
4858 {
4859         int headlen = skb_headlen(skb);
4860 
4861         if (len < headlen)
4862                 return pskb_carve_inside_header(skb, len, headlen, gfp);
4863         else
4864                 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
4865 }
4866 
4867 /* Extract to_copy bytes starting at off from skb, and return this in
4868  * a new skb
4869  */
4870 struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
4871                              int to_copy, gfp_t gfp)
4872 {
4873         struct sk_buff  *clone = skb_clone(skb, gfp);
4874 
4875         if (!clone)
4876                 return NULL;
4877 
4878         if (pskb_carve(clone, off, gfp) < 0 ||
4879             pskb_trim(clone, to_copy)) {
4880                 kfree_skb(clone);
4881                 return NULL;
4882         }
4883         return clone;
4884 }
4885 EXPORT_SYMBOL(pskb_extract);
4886 

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