~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/net/core/skbuff.c

Version: ~ [ linux-5.2-rc5 ] ~ [ linux-5.1.12 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.53 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.128 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.182 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.182 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.68 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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 <linux/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         unsigned int 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 = page_frag_alloc(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 page_frag_alloc(&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 = page_frag_alloc(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 = page_frag_alloc(&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         if (!skb->sk || skb->destructor == sock_edemux)
1576                 skb_condense(skb);
1577         return 0;
1578 }
1579 EXPORT_SYMBOL(___pskb_trim);
1580 
1581 /**
1582  *      __pskb_pull_tail - advance tail of skb header
1583  *      @skb: buffer to reallocate
1584  *      @delta: number of bytes to advance tail
1585  *
1586  *      The function makes a sense only on a fragmented &sk_buff,
1587  *      it expands header moving its tail forward and copying necessary
1588  *      data from fragmented part.
1589  *
1590  *      &sk_buff MUST have reference count of 1.
1591  *
1592  *      Returns %NULL (and &sk_buff does not change) if pull failed
1593  *      or value of new tail of skb in the case of success.
1594  *
1595  *      All the pointers pointing into skb header may change and must be
1596  *      reloaded after call to this function.
1597  */
1598 
1599 /* Moves tail of skb head forward, copying data from fragmented part,
1600  * when it is necessary.
1601  * 1. It may fail due to malloc failure.
1602  * 2. It may change skb pointers.
1603  *
1604  * It is pretty complicated. Luckily, it is called only in exceptional cases.
1605  */
1606 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1607 {
1608         /* If skb has not enough free space at tail, get new one
1609          * plus 128 bytes for future expansions. If we have enough
1610          * room at tail, reallocate without expansion only if skb is cloned.
1611          */
1612         int i, k, eat = (skb->tail + delta) - skb->end;
1613 
1614         if (eat > 0 || skb_cloned(skb)) {
1615                 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1616                                      GFP_ATOMIC))
1617                         return NULL;
1618         }
1619 
1620         if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1621                 BUG();
1622 
1623         /* Optimization: no fragments, no reasons to preestimate
1624          * size of pulled pages. Superb.
1625          */
1626         if (!skb_has_frag_list(skb))
1627                 goto pull_pages;
1628 
1629         /* Estimate size of pulled pages. */
1630         eat = delta;
1631         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1632                 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1633 
1634                 if (size >= eat)
1635                         goto pull_pages;
1636                 eat -= size;
1637         }
1638 
1639         /* If we need update frag list, we are in troubles.
1640          * Certainly, it possible to add an offset to skb data,
1641          * but taking into account that pulling is expected to
1642          * be very rare operation, it is worth to fight against
1643          * further bloating skb head and crucify ourselves here instead.
1644          * Pure masohism, indeed. 8)8)
1645          */
1646         if (eat) {
1647                 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1648                 struct sk_buff *clone = NULL;
1649                 struct sk_buff *insp = NULL;
1650 
1651                 do {
1652                         BUG_ON(!list);
1653 
1654                         if (list->len <= eat) {
1655                                 /* Eaten as whole. */
1656                                 eat -= list->len;
1657                                 list = list->next;
1658                                 insp = list;
1659                         } else {
1660                                 /* Eaten partially. */
1661 
1662                                 if (skb_shared(list)) {
1663                                         /* Sucks! We need to fork list. :-( */
1664                                         clone = skb_clone(list, GFP_ATOMIC);
1665                                         if (!clone)
1666                                                 return NULL;
1667                                         insp = list->next;
1668                                         list = clone;
1669                                 } else {
1670                                         /* This may be pulled without
1671                                          * problems. */
1672                                         insp = list;
1673                                 }
1674                                 if (!pskb_pull(list, eat)) {
1675                                         kfree_skb(clone);
1676                                         return NULL;
1677                                 }
1678                                 break;
1679                         }
1680                 } while (eat);
1681 
1682                 /* Free pulled out fragments. */
1683                 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1684                         skb_shinfo(skb)->frag_list = list->next;
1685                         kfree_skb(list);
1686                 }
1687                 /* And insert new clone at head. */
1688                 if (clone) {
1689                         clone->next = list;
1690                         skb_shinfo(skb)->frag_list = clone;
1691                 }
1692         }
1693         /* Success! Now we may commit changes to skb data. */
1694 
1695 pull_pages:
1696         eat = delta;
1697         k = 0;
1698         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1699                 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1700 
1701                 if (size <= eat) {
1702                         skb_frag_unref(skb, i);
1703                         eat -= size;
1704                 } else {
1705                         skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1706                         if (eat) {
1707                                 skb_shinfo(skb)->frags[k].page_offset += eat;
1708                                 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1709                                 eat = 0;
1710                         }
1711                         k++;
1712                 }
1713         }
1714         skb_shinfo(skb)->nr_frags = k;
1715 
1716         skb->tail     += delta;
1717         skb->data_len -= delta;
1718 
1719         return skb_tail_pointer(skb);
1720 }
1721 EXPORT_SYMBOL(__pskb_pull_tail);
1722 
1723 /**
1724  *      skb_copy_bits - copy bits from skb to kernel buffer
1725  *      @skb: source skb
1726  *      @offset: offset in source
1727  *      @to: destination buffer
1728  *      @len: number of bytes to copy
1729  *
1730  *      Copy the specified number of bytes from the source skb to the
1731  *      destination buffer.
1732  *
1733  *      CAUTION ! :
1734  *              If its prototype is ever changed,
1735  *              check arch/{*}/net/{*}.S files,
1736  *              since it is called from BPF assembly code.
1737  */
1738 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1739 {
1740         int start = skb_headlen(skb);
1741         struct sk_buff *frag_iter;
1742         int i, copy;
1743 
1744         if (offset > (int)skb->len - len)
1745                 goto fault;
1746 
1747         /* Copy header. */
1748         if ((copy = start - offset) > 0) {
1749                 if (copy > len)
1750                         copy = len;
1751                 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1752                 if ((len -= copy) == 0)
1753                         return 0;
1754                 offset += copy;
1755                 to     += copy;
1756         }
1757 
1758         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1759                 int end;
1760                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1761 
1762                 WARN_ON(start > offset + len);
1763 
1764                 end = start + skb_frag_size(f);
1765                 if ((copy = end - offset) > 0) {
1766                         u8 *vaddr;
1767 
1768                         if (copy > len)
1769                                 copy = len;
1770 
1771                         vaddr = kmap_atomic(skb_frag_page(f));
1772                         memcpy(to,
1773                                vaddr + f->page_offset + offset - start,
1774                                copy);
1775                         kunmap_atomic(vaddr);
1776 
1777                         if ((len -= copy) == 0)
1778                                 return 0;
1779                         offset += copy;
1780                         to     += copy;
1781                 }
1782                 start = end;
1783         }
1784 
1785         skb_walk_frags(skb, frag_iter) {
1786                 int end;
1787 
1788                 WARN_ON(start > offset + len);
1789 
1790                 end = start + frag_iter->len;
1791                 if ((copy = end - offset) > 0) {
1792                         if (copy > len)
1793                                 copy = len;
1794                         if (skb_copy_bits(frag_iter, offset - start, to, copy))
1795                                 goto fault;
1796                         if ((len -= copy) == 0)
1797                                 return 0;
1798                         offset += copy;
1799                         to     += copy;
1800                 }
1801                 start = end;
1802         }
1803 
1804         if (!len)
1805                 return 0;
1806 
1807 fault:
1808         return -EFAULT;
1809 }
1810 EXPORT_SYMBOL(skb_copy_bits);
1811 
1812 /*
1813  * Callback from splice_to_pipe(), if we need to release some pages
1814  * at the end of the spd in case we error'ed out in filling the pipe.
1815  */
1816 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1817 {
1818         put_page(spd->pages[i]);
1819 }
1820 
1821 static struct page *linear_to_page(struct page *page, unsigned int *len,
1822                                    unsigned int *offset,
1823                                    struct sock *sk)
1824 {
1825         struct page_frag *pfrag = sk_page_frag(sk);
1826 
1827         if (!sk_page_frag_refill(sk, pfrag))
1828                 return NULL;
1829 
1830         *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
1831 
1832         memcpy(page_address(pfrag->page) + pfrag->offset,
1833                page_address(page) + *offset, *len);
1834         *offset = pfrag->offset;
1835         pfrag->offset += *len;
1836 
1837         return pfrag->page;
1838 }
1839 
1840 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1841                              struct page *page,
1842                              unsigned int offset)
1843 {
1844         return  spd->nr_pages &&
1845                 spd->pages[spd->nr_pages - 1] == page &&
1846                 (spd->partial[spd->nr_pages - 1].offset +
1847                  spd->partial[spd->nr_pages - 1].len == offset);
1848 }
1849 
1850 /*
1851  * Fill page/offset/length into spd, if it can hold more pages.
1852  */
1853 static bool spd_fill_page(struct splice_pipe_desc *spd,
1854                           struct pipe_inode_info *pipe, struct page *page,
1855                           unsigned int *len, unsigned int offset,
1856                           bool linear,
1857                           struct sock *sk)
1858 {
1859         if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1860                 return true;
1861 
1862         if (linear) {
1863                 page = linear_to_page(page, len, &offset, sk);
1864                 if (!page)
1865                         return true;
1866         }
1867         if (spd_can_coalesce(spd, page, offset)) {
1868                 spd->partial[spd->nr_pages - 1].len += *len;
1869                 return false;
1870         }
1871         get_page(page);
1872         spd->pages[spd->nr_pages] = page;
1873         spd->partial[spd->nr_pages].len = *len;
1874         spd->partial[spd->nr_pages].offset = offset;
1875         spd->nr_pages++;
1876 
1877         return false;
1878 }
1879 
1880 static bool __splice_segment(struct page *page, unsigned int poff,
1881                              unsigned int plen, unsigned int *off,
1882                              unsigned int *len,
1883                              struct splice_pipe_desc *spd, bool linear,
1884                              struct sock *sk,
1885                              struct pipe_inode_info *pipe)
1886 {
1887         if (!*len)
1888                 return true;
1889 
1890         /* skip this segment if already processed */
1891         if (*off >= plen) {
1892                 *off -= plen;
1893                 return false;
1894         }
1895 
1896         /* ignore any bits we already processed */
1897         poff += *off;
1898         plen -= *off;
1899         *off = 0;
1900 
1901         do {
1902                 unsigned int flen = min(*len, plen);
1903 
1904                 if (spd_fill_page(spd, pipe, page, &flen, poff,
1905                                   linear, sk))
1906                         return true;
1907                 poff += flen;
1908                 plen -= flen;
1909                 *len -= flen;
1910         } while (*len && plen);
1911 
1912         return false;
1913 }
1914 
1915 /*
1916  * Map linear and fragment data from the skb to spd. It reports true if the
1917  * pipe is full or if we already spliced the requested length.
1918  */
1919 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1920                               unsigned int *offset, unsigned int *len,
1921                               struct splice_pipe_desc *spd, struct sock *sk)
1922 {
1923         int seg;
1924         struct sk_buff *iter;
1925 
1926         /* map the linear part :
1927          * If skb->head_frag is set, this 'linear' part is backed by a
1928          * fragment, and if the head is not shared with any clones then
1929          * we can avoid a copy since we own the head portion of this page.
1930          */
1931         if (__splice_segment(virt_to_page(skb->data),
1932                              (unsigned long) skb->data & (PAGE_SIZE - 1),
1933                              skb_headlen(skb),
1934                              offset, len, spd,
1935                              skb_head_is_locked(skb),
1936                              sk, pipe))
1937                 return true;
1938 
1939         /*
1940          * then map the fragments
1941          */
1942         for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1943                 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1944 
1945                 if (__splice_segment(skb_frag_page(f),
1946                                      f->page_offset, skb_frag_size(f),
1947                                      offset, len, spd, false, sk, pipe))
1948                         return true;
1949         }
1950 
1951         skb_walk_frags(skb, iter) {
1952                 if (*offset >= iter->len) {
1953                         *offset -= iter->len;
1954                         continue;
1955                 }
1956                 /* __skb_splice_bits() only fails if the output has no room
1957                  * left, so no point in going over the frag_list for the error
1958                  * case.
1959                  */
1960                 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
1961                         return true;
1962         }
1963 
1964         return false;
1965 }
1966 
1967 /*
1968  * Map data from the skb to a pipe. Should handle both the linear part,
1969  * the fragments, and the frag list.
1970  */
1971 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
1972                     struct pipe_inode_info *pipe, unsigned int tlen,
1973                     unsigned int flags)
1974 {
1975         struct partial_page partial[MAX_SKB_FRAGS];
1976         struct page *pages[MAX_SKB_FRAGS];
1977         struct splice_pipe_desc spd = {
1978                 .pages = pages,
1979                 .partial = partial,
1980                 .nr_pages_max = MAX_SKB_FRAGS,
1981                 .flags = flags,
1982                 .ops = &nosteal_pipe_buf_ops,
1983                 .spd_release = sock_spd_release,
1984         };
1985         int ret = 0;
1986 
1987         __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
1988 
1989         if (spd.nr_pages)
1990                 ret = splice_to_pipe(pipe, &spd);
1991 
1992         return ret;
1993 }
1994 EXPORT_SYMBOL_GPL(skb_splice_bits);
1995 
1996 /**
1997  *      skb_store_bits - store bits from kernel buffer to skb
1998  *      @skb: destination buffer
1999  *      @offset: offset in destination
2000  *      @from: source buffer
2001  *      @len: number of bytes to copy
2002  *
2003  *      Copy the specified number of bytes from the source buffer to the
2004  *      destination skb.  This function handles all the messy bits of
2005  *      traversing fragment lists and such.
2006  */
2007 
2008 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2009 {
2010         int start = skb_headlen(skb);
2011         struct sk_buff *frag_iter;
2012         int i, copy;
2013 
2014         if (offset > (int)skb->len - len)
2015                 goto fault;
2016 
2017         if ((copy = start - offset) > 0) {
2018                 if (copy > len)
2019                         copy = len;
2020                 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2021                 if ((len -= copy) == 0)
2022                         return 0;
2023                 offset += copy;
2024                 from += copy;
2025         }
2026 
2027         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2028                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2029                 int end;
2030 
2031                 WARN_ON(start > offset + len);
2032 
2033                 end = start + skb_frag_size(frag);
2034                 if ((copy = end - offset) > 0) {
2035                         u8 *vaddr;
2036 
2037                         if (copy > len)
2038                                 copy = len;
2039 
2040                         vaddr = kmap_atomic(skb_frag_page(frag));
2041                         memcpy(vaddr + frag->page_offset + offset - start,
2042                                from, copy);
2043                         kunmap_atomic(vaddr);
2044 
2045                         if ((len -= copy) == 0)
2046                                 return 0;
2047                         offset += copy;
2048                         from += copy;
2049                 }
2050                 start = end;
2051         }
2052 
2053         skb_walk_frags(skb, frag_iter) {
2054                 int end;
2055 
2056                 WARN_ON(start > offset + len);
2057 
2058                 end = start + frag_iter->len;
2059                 if ((copy = end - offset) > 0) {
2060                         if (copy > len)
2061                                 copy = len;
2062                         if (skb_store_bits(frag_iter, offset - start,
2063                                            from, copy))
2064                                 goto fault;
2065                         if ((len -= copy) == 0)
2066                                 return 0;
2067                         offset += copy;
2068                         from += copy;
2069                 }
2070                 start = end;
2071         }
2072         if (!len)
2073                 return 0;
2074 
2075 fault:
2076         return -EFAULT;
2077 }
2078 EXPORT_SYMBOL(skb_store_bits);
2079 
2080 /* Checksum skb data. */
2081 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2082                       __wsum csum, const struct skb_checksum_ops *ops)
2083 {
2084         int start = skb_headlen(skb);
2085         int i, copy = start - offset;
2086         struct sk_buff *frag_iter;
2087         int pos = 0;
2088 
2089         /* Checksum header. */
2090         if (copy > 0) {
2091                 if (copy > len)
2092                         copy = len;
2093                 csum = ops->update(skb->data + offset, copy, csum);
2094                 if ((len -= copy) == 0)
2095                         return csum;
2096                 offset += copy;
2097                 pos     = copy;
2098         }
2099 
2100         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2101                 int end;
2102                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2103 
2104                 WARN_ON(start > offset + len);
2105 
2106                 end = start + skb_frag_size(frag);
2107                 if ((copy = end - offset) > 0) {
2108                         __wsum csum2;
2109                         u8 *vaddr;
2110 
2111                         if (copy > len)
2112                                 copy = len;
2113                         vaddr = kmap_atomic(skb_frag_page(frag));
2114                         csum2 = ops->update(vaddr + frag->page_offset +
2115                                             offset - start, copy, 0);
2116                         kunmap_atomic(vaddr);
2117                         csum = ops->combine(csum, csum2, pos, copy);
2118                         if (!(len -= copy))
2119                                 return csum;
2120                         offset += copy;
2121                         pos    += copy;
2122                 }
2123                 start = end;
2124         }
2125 
2126         skb_walk_frags(skb, frag_iter) {
2127                 int end;
2128 
2129                 WARN_ON(start > offset + len);
2130 
2131                 end = start + frag_iter->len;
2132                 if ((copy = end - offset) > 0) {
2133                         __wsum csum2;
2134                         if (copy > len)
2135                                 copy = len;
2136                         csum2 = __skb_checksum(frag_iter, offset - start,
2137                                                copy, 0, ops);
2138                         csum = ops->combine(csum, csum2, pos, copy);
2139                         if ((len -= copy) == 0)
2140                                 return csum;
2141                         offset += copy;
2142                         pos    += copy;
2143                 }
2144                 start = end;
2145         }
2146         BUG_ON(len);
2147 
2148         return csum;
2149 }
2150 EXPORT_SYMBOL(__skb_checksum);
2151 
2152 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2153                     int len, __wsum csum)
2154 {
2155         const struct skb_checksum_ops ops = {
2156                 .update  = csum_partial_ext,
2157                 .combine = csum_block_add_ext,
2158         };
2159 
2160         return __skb_checksum(skb, offset, len, csum, &ops);
2161 }
2162 EXPORT_SYMBOL(skb_checksum);
2163 
2164 /* Both of above in one bottle. */
2165 
2166 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2167                                     u8 *to, int len, __wsum csum)
2168 {
2169         int start = skb_headlen(skb);
2170         int i, copy = start - offset;
2171         struct sk_buff *frag_iter;
2172         int pos = 0;
2173 
2174         /* Copy header. */
2175         if (copy > 0) {
2176                 if (copy > len)
2177                         copy = len;
2178                 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2179                                                  copy, csum);
2180                 if ((len -= copy) == 0)
2181                         return csum;
2182                 offset += copy;
2183                 to     += copy;
2184                 pos     = copy;
2185         }
2186 
2187         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2188                 int end;
2189 
2190                 WARN_ON(start > offset + len);
2191 
2192                 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2193                 if ((copy = end - offset) > 0) {
2194                         __wsum csum2;
2195                         u8 *vaddr;
2196                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2197 
2198                         if (copy > len)
2199                                 copy = len;
2200                         vaddr = kmap_atomic(skb_frag_page(frag));
2201                         csum2 = csum_partial_copy_nocheck(vaddr +
2202                                                           frag->page_offset +
2203                                                           offset - start, to,
2204                                                           copy, 0);
2205                         kunmap_atomic(vaddr);
2206                         csum = csum_block_add(csum, csum2, pos);
2207                         if (!(len -= copy))
2208                                 return csum;
2209                         offset += copy;
2210                         to     += copy;
2211                         pos    += copy;
2212                 }
2213                 start = end;
2214         }
2215 
2216         skb_walk_frags(skb, frag_iter) {
2217                 __wsum csum2;
2218                 int end;
2219 
2220                 WARN_ON(start > offset + len);
2221 
2222                 end = start + frag_iter->len;
2223                 if ((copy = end - offset) > 0) {
2224                         if (copy > len)
2225                                 copy = len;
2226                         csum2 = skb_copy_and_csum_bits(frag_iter,
2227                                                        offset - start,
2228                                                        to, copy, 0);
2229                         csum = csum_block_add(csum, csum2, pos);
2230                         if ((len -= copy) == 0)
2231                                 return csum;
2232                         offset += copy;
2233                         to     += copy;
2234                         pos    += copy;
2235                 }
2236                 start = end;
2237         }
2238         BUG_ON(len);
2239         return csum;
2240 }
2241 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2242 
2243  /**
2244  *      skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2245  *      @from: source buffer
2246  *
2247  *      Calculates the amount of linear headroom needed in the 'to' skb passed
2248  *      into skb_zerocopy().
2249  */
2250 unsigned int
2251 skb_zerocopy_headlen(const struct sk_buff *from)
2252 {
2253         unsigned int hlen = 0;
2254 
2255         if (!from->head_frag ||
2256             skb_headlen(from) < L1_CACHE_BYTES ||
2257             skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2258                 hlen = skb_headlen(from);
2259 
2260         if (skb_has_frag_list(from))
2261                 hlen = from->len;
2262 
2263         return hlen;
2264 }
2265 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2266 
2267 /**
2268  *      skb_zerocopy - Zero copy skb to skb
2269  *      @to: destination buffer
2270  *      @from: source buffer
2271  *      @len: number of bytes to copy from source buffer
2272  *      @hlen: size of linear headroom in destination buffer
2273  *
2274  *      Copies up to `len` bytes from `from` to `to` by creating references
2275  *      to the frags in the source buffer.
2276  *
2277  *      The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2278  *      headroom in the `to` buffer.
2279  *
2280  *      Return value:
2281  *      0: everything is OK
2282  *      -ENOMEM: couldn't orphan frags of @from due to lack of memory
2283  *      -EFAULT: skb_copy_bits() found some problem with skb geometry
2284  */
2285 int
2286 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2287 {
2288         int i, j = 0;
2289         int plen = 0; /* length of skb->head fragment */
2290         int ret;
2291         struct page *page;
2292         unsigned int offset;
2293 
2294         BUG_ON(!from->head_frag && !hlen);
2295 
2296         /* dont bother with small payloads */
2297         if (len <= skb_tailroom(to))
2298                 return skb_copy_bits(from, 0, skb_put(to, len), len);
2299 
2300         if (hlen) {
2301                 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2302                 if (unlikely(ret))
2303                         return ret;
2304                 len -= hlen;
2305         } else {
2306                 plen = min_t(int, skb_headlen(from), len);
2307                 if (plen) {
2308                         page = virt_to_head_page(from->head);
2309                         offset = from->data - (unsigned char *)page_address(page);
2310                         __skb_fill_page_desc(to, 0, page, offset, plen);
2311                         get_page(page);
2312                         j = 1;
2313                         len -= plen;
2314                 }
2315         }
2316 
2317         to->truesize += len + plen;
2318         to->len += len + plen;
2319         to->data_len += len + plen;
2320 
2321         if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2322                 skb_tx_error(from);
2323                 return -ENOMEM;
2324         }
2325 
2326         for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2327                 if (!len)
2328                         break;
2329                 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2330                 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
2331                 len -= skb_shinfo(to)->frags[j].size;
2332                 skb_frag_ref(to, j);
2333                 j++;
2334         }
2335         skb_shinfo(to)->nr_frags = j;
2336 
2337         return 0;
2338 }
2339 EXPORT_SYMBOL_GPL(skb_zerocopy);
2340 
2341 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2342 {
2343         __wsum csum;
2344         long csstart;
2345 
2346         if (skb->ip_summed == CHECKSUM_PARTIAL)
2347                 csstart = skb_checksum_start_offset(skb);
2348         else
2349                 csstart = skb_headlen(skb);
2350 
2351         BUG_ON(csstart > skb_headlen(skb));
2352 
2353         skb_copy_from_linear_data(skb, to, csstart);
2354 
2355         csum = 0;
2356         if (csstart != skb->len)
2357                 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2358                                               skb->len - csstart, 0);
2359 
2360         if (skb->ip_summed == CHECKSUM_PARTIAL) {
2361                 long csstuff = csstart + skb->csum_offset;
2362 
2363                 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2364         }
2365 }
2366 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2367 
2368 /**
2369  *      skb_dequeue - remove from the head of the queue
2370  *      @list: list to dequeue from
2371  *
2372  *      Remove the head of the list. The list lock is taken so the function
2373  *      may be used safely with other locking list functions. The head item is
2374  *      returned or %NULL if the list is empty.
2375  */
2376 
2377 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2378 {
2379         unsigned long flags;
2380         struct sk_buff *result;
2381 
2382         spin_lock_irqsave(&list->lock, flags);
2383         result = __skb_dequeue(list);
2384         spin_unlock_irqrestore(&list->lock, flags);
2385         return result;
2386 }
2387 EXPORT_SYMBOL(skb_dequeue);
2388 
2389 /**
2390  *      skb_dequeue_tail - remove from the tail of the queue
2391  *      @list: list to dequeue from
2392  *
2393  *      Remove the tail of the list. The list lock is taken so the function
2394  *      may be used safely with other locking list functions. The tail item is
2395  *      returned or %NULL if the list is empty.
2396  */
2397 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2398 {
2399         unsigned long flags;
2400         struct sk_buff *result;
2401 
2402         spin_lock_irqsave(&list->lock, flags);
2403         result = __skb_dequeue_tail(list);
2404         spin_unlock_irqrestore(&list->lock, flags);
2405         return result;
2406 }
2407 EXPORT_SYMBOL(skb_dequeue_tail);
2408 
2409 /**
2410  *      skb_queue_purge - empty a list
2411  *      @list: list to empty
2412  *
2413  *      Delete all buffers on an &sk_buff list. Each buffer is removed from
2414  *      the list and one reference dropped. This function takes the list
2415  *      lock and is atomic with respect to other list locking functions.
2416  */
2417 void skb_queue_purge(struct sk_buff_head *list)
2418 {
2419         struct sk_buff *skb;
2420         while ((skb = skb_dequeue(list)) != NULL)
2421                 kfree_skb(skb);
2422 }
2423 EXPORT_SYMBOL(skb_queue_purge);
2424 
2425 /**
2426  *      skb_rbtree_purge - empty a skb rbtree
2427  *      @root: root of the rbtree to empty
2428  *
2429  *      Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
2430  *      the list and one reference dropped. This function does not take
2431  *      any lock. Synchronization should be handled by the caller (e.g., TCP
2432  *      out-of-order queue is protected by the socket lock).
2433  */
2434 void skb_rbtree_purge(struct rb_root *root)
2435 {
2436         struct sk_buff *skb, *next;
2437 
2438         rbtree_postorder_for_each_entry_safe(skb, next, root, rbnode)
2439                 kfree_skb(skb);
2440 
2441         *root = RB_ROOT;
2442 }
2443 
2444 /**
2445  *      skb_queue_head - queue a buffer at the list head
2446  *      @list: list to use
2447  *      @newsk: buffer to queue
2448  *
2449  *      Queue a buffer at the start of the list. This function takes the
2450  *      list lock and can be used safely with other locking &sk_buff functions
2451  *      safely.
2452  *
2453  *      A buffer cannot be placed on two lists at the same time.
2454  */
2455 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2456 {
2457         unsigned long flags;
2458 
2459         spin_lock_irqsave(&list->lock, flags);
2460         __skb_queue_head(list, newsk);
2461         spin_unlock_irqrestore(&list->lock, flags);
2462 }
2463 EXPORT_SYMBOL(skb_queue_head);
2464 
2465 /**
2466  *      skb_queue_tail - queue a buffer at the list tail
2467  *      @list: list to use
2468  *      @newsk: buffer to queue
2469  *
2470  *      Queue a buffer at the tail of the list. This function takes the
2471  *      list lock and can be used safely with other locking &sk_buff functions
2472  *      safely.
2473  *
2474  *      A buffer cannot be placed on two lists at the same time.
2475  */
2476 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2477 {
2478         unsigned long flags;
2479 
2480         spin_lock_irqsave(&list->lock, flags);
2481         __skb_queue_tail(list, newsk);
2482         spin_unlock_irqrestore(&list->lock, flags);
2483 }
2484 EXPORT_SYMBOL(skb_queue_tail);
2485 
2486 /**
2487  *      skb_unlink      -       remove a buffer from a list
2488  *      @skb: buffer to remove
2489  *      @list: list to use
2490  *
2491  *      Remove a packet from a list. The list locks are taken and this
2492  *      function is atomic with respect to other list locked calls
2493  *
2494  *      You must know what list the SKB is on.
2495  */
2496 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2497 {
2498         unsigned long flags;
2499 
2500         spin_lock_irqsave(&list->lock, flags);
2501         __skb_unlink(skb, list);
2502         spin_unlock_irqrestore(&list->lock, flags);
2503 }
2504 EXPORT_SYMBOL(skb_unlink);
2505 
2506 /**
2507  *      skb_append      -       append a buffer
2508  *      @old: buffer to insert after
2509  *      @newsk: buffer to insert
2510  *      @list: list to use
2511  *
2512  *      Place a packet after a given packet in a list. The list locks are taken
2513  *      and this function is atomic with respect to other list locked calls.
2514  *      A buffer cannot be placed on two lists at the same time.
2515  */
2516 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2517 {
2518         unsigned long flags;
2519 
2520         spin_lock_irqsave(&list->lock, flags);
2521         __skb_queue_after(list, old, newsk);
2522         spin_unlock_irqrestore(&list->lock, flags);
2523 }
2524 EXPORT_SYMBOL(skb_append);
2525 
2526 /**
2527  *      skb_insert      -       insert a buffer
2528  *      @old: buffer to insert before
2529  *      @newsk: buffer to insert
2530  *      @list: list to use
2531  *
2532  *      Place a packet before a given packet in a list. The list locks are
2533  *      taken and this function is atomic with respect to other list locked
2534  *      calls.
2535  *
2536  *      A buffer cannot be placed on two lists at the same time.
2537  */
2538 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2539 {
2540         unsigned long flags;
2541 
2542         spin_lock_irqsave(&list->lock, flags);
2543         __skb_insert(newsk, old->prev, old, list);
2544         spin_unlock_irqrestore(&list->lock, flags);
2545 }
2546 EXPORT_SYMBOL(skb_insert);
2547 
2548 static inline void skb_split_inside_header(struct sk_buff *skb,
2549                                            struct sk_buff* skb1,
2550                                            const u32 len, const int pos)
2551 {
2552         int i;
2553 
2554         skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2555                                          pos - len);
2556         /* And move data appendix as is. */
2557         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2558                 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2559 
2560         skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2561         skb_shinfo(skb)->nr_frags  = 0;
2562         skb1->data_len             = skb->data_len;
2563         skb1->len                  += skb1->data_len;
2564         skb->data_len              = 0;
2565         skb->len                   = len;
2566         skb_set_tail_pointer(skb, len);
2567 }
2568 
2569 static inline void skb_split_no_header(struct sk_buff *skb,
2570                                        struct sk_buff* skb1,
2571                                        const u32 len, int pos)
2572 {
2573         int i, k = 0;
2574         const int nfrags = skb_shinfo(skb)->nr_frags;
2575 
2576         skb_shinfo(skb)->nr_frags = 0;
2577         skb1->len                 = skb1->data_len = skb->len - len;
2578         skb->len                  = len;
2579         skb->data_len             = len - pos;
2580 
2581         for (i = 0; i < nfrags; i++) {
2582                 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2583 
2584                 if (pos + size > len) {
2585                         skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2586 
2587                         if (pos < len) {
2588                                 /* Split frag.
2589                                  * We have two variants in this case:
2590                                  * 1. Move all the frag to the second
2591                                  *    part, if it is possible. F.e.
2592                                  *    this approach is mandatory for TUX,
2593                                  *    where splitting is expensive.
2594                                  * 2. Split is accurately. We make this.
2595                                  */
2596                                 skb_frag_ref(skb, i);
2597                                 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2598                                 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2599                                 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2600                                 skb_shinfo(skb)->nr_frags++;
2601                         }
2602                         k++;
2603                 } else
2604                         skb_shinfo(skb)->nr_frags++;
2605                 pos += size;
2606         }
2607         skb_shinfo(skb1)->nr_frags = k;
2608 }
2609 
2610 /**
2611  * skb_split - Split fragmented skb to two parts at length len.
2612  * @skb: the buffer to split
2613  * @skb1: the buffer to receive the second part
2614  * @len: new length for skb
2615  */
2616 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2617 {
2618         int pos = skb_headlen(skb);
2619 
2620         skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2621         if (len < pos)  /* Split line is inside header. */
2622                 skb_split_inside_header(skb, skb1, len, pos);
2623         else            /* Second chunk has no header, nothing to copy. */
2624                 skb_split_no_header(skb, skb1, len, pos);
2625 }
2626 EXPORT_SYMBOL(skb_split);
2627 
2628 /* Shifting from/to a cloned skb is a no-go.
2629  *
2630  * Caller cannot keep skb_shinfo related pointers past calling here!
2631  */
2632 static int skb_prepare_for_shift(struct sk_buff *skb)
2633 {
2634         return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2635 }
2636 
2637 /**
2638  * skb_shift - Shifts paged data partially from skb to another
2639  * @tgt: buffer into which tail data gets added
2640  * @skb: buffer from which the paged data comes from
2641  * @shiftlen: shift up to this many bytes
2642  *
2643  * Attempts to shift up to shiftlen worth of bytes, which may be less than
2644  * the length of the skb, from skb to tgt. Returns number bytes shifted.
2645  * It's up to caller to free skb if everything was shifted.
2646  *
2647  * If @tgt runs out of frags, the whole operation is aborted.
2648  *
2649  * Skb cannot include anything else but paged data while tgt is allowed
2650  * to have non-paged data as well.
2651  *
2652  * TODO: full sized shift could be optimized but that would need
2653  * specialized skb free'er to handle frags without up-to-date nr_frags.
2654  */
2655 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2656 {
2657         int from, to, merge, todo;
2658         struct skb_frag_struct *fragfrom, *fragto;
2659 
2660         BUG_ON(shiftlen > skb->len);
2661 
2662         if (skb_headlen(skb))
2663                 return 0;
2664 
2665         todo = shiftlen;
2666         from = 0;
2667         to = skb_shinfo(tgt)->nr_frags;
2668         fragfrom = &skb_shinfo(skb)->frags[from];
2669 
2670         /* Actual merge is delayed until the point when we know we can
2671          * commit all, so that we don't have to undo partial changes
2672          */
2673         if (!to ||
2674             !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2675                               fragfrom->page_offset)) {
2676                 merge = -1;
2677         } else {
2678                 merge = to - 1;
2679 
2680                 todo -= skb_frag_size(fragfrom);
2681                 if (todo < 0) {
2682                         if (skb_prepare_for_shift(skb) ||
2683                             skb_prepare_for_shift(tgt))
2684                                 return 0;
2685 
2686                         /* All previous frag pointers might be stale! */
2687                         fragfrom = &skb_shinfo(skb)->frags[from];
2688                         fragto = &skb_shinfo(tgt)->frags[merge];
2689 
2690                         skb_frag_size_add(fragto, shiftlen);
2691                         skb_frag_size_sub(fragfrom, shiftlen);
2692                         fragfrom->page_offset += shiftlen;
2693 
2694                         goto onlymerged;
2695                 }
2696 
2697                 from++;
2698         }
2699 
2700         /* Skip full, not-fitting skb to avoid expensive operations */
2701         if ((shiftlen == skb->len) &&
2702             (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2703                 return 0;
2704 
2705         if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2706                 return 0;
2707 
2708         while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2709                 if (to == MAX_SKB_FRAGS)
2710                         return 0;
2711 
2712                 fragfrom = &skb_shinfo(skb)->frags[from];
2713                 fragto = &skb_shinfo(tgt)->frags[to];
2714 
2715                 if (todo >= skb_frag_size(fragfrom)) {
2716                         *fragto = *fragfrom;
2717                         todo -= skb_frag_size(fragfrom);
2718                         from++;
2719                         to++;
2720 
2721                 } else {
2722                         __skb_frag_ref(fragfrom);
2723                         fragto->page = fragfrom->page;
2724                         fragto->page_offset = fragfrom->page_offset;
2725                         skb_frag_size_set(fragto, todo);
2726 
2727                         fragfrom->page_offset += todo;
2728                         skb_frag_size_sub(fragfrom, todo);
2729                         todo = 0;
2730 
2731                         to++;
2732                         break;
2733                 }
2734         }
2735 
2736         /* Ready to "commit" this state change to tgt */
2737         skb_shinfo(tgt)->nr_frags = to;
2738 
2739         if (merge >= 0) {
2740                 fragfrom = &skb_shinfo(skb)->frags[0];
2741                 fragto = &skb_shinfo(tgt)->frags[merge];
2742 
2743                 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2744                 __skb_frag_unref(fragfrom);
2745         }
2746 
2747         /* Reposition in the original skb */
2748         to = 0;
2749         while (from < skb_shinfo(skb)->nr_frags)
2750                 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2751         skb_shinfo(skb)->nr_frags = to;
2752 
2753         BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2754 
2755 onlymerged:
2756         /* Most likely the tgt won't ever need its checksum anymore, skb on
2757          * the other hand might need it if it needs to be resent
2758          */
2759         tgt->ip_summed = CHECKSUM_PARTIAL;
2760         skb->ip_summed = CHECKSUM_PARTIAL;
2761 
2762         /* Yak, is it really working this way? Some helper please? */
2763         skb->len -= shiftlen;
2764         skb->data_len -= shiftlen;
2765         skb->truesize -= shiftlen;
2766         tgt->len += shiftlen;
2767         tgt->data_len += shiftlen;
2768         tgt->truesize += shiftlen;
2769 
2770         return shiftlen;
2771 }
2772 
2773 /**
2774  * skb_prepare_seq_read - Prepare a sequential read of skb data
2775  * @skb: the buffer to read
2776  * @from: lower offset of data to be read
2777  * @to: upper offset of data to be read
2778  * @st: state variable
2779  *
2780  * Initializes the specified state variable. Must be called before
2781  * invoking skb_seq_read() for the first time.
2782  */
2783 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2784                           unsigned int to, struct skb_seq_state *st)
2785 {
2786         st->lower_offset = from;
2787         st->upper_offset = to;
2788         st->root_skb = st->cur_skb = skb;
2789         st->frag_idx = st->stepped_offset = 0;
2790         st->frag_data = NULL;
2791 }
2792 EXPORT_SYMBOL(skb_prepare_seq_read);
2793 
2794 /**
2795  * skb_seq_read - Sequentially read skb data
2796  * @consumed: number of bytes consumed by the caller so far
2797  * @data: destination pointer for data to be returned
2798  * @st: state variable
2799  *
2800  * Reads a block of skb data at @consumed relative to the
2801  * lower offset specified to skb_prepare_seq_read(). Assigns
2802  * the head of the data block to @data and returns the length
2803  * of the block or 0 if the end of the skb data or the upper
2804  * offset has been reached.
2805  *
2806  * The caller is not required to consume all of the data
2807  * returned, i.e. @consumed is typically set to the number
2808  * of bytes already consumed and the next call to
2809  * skb_seq_read() will return the remaining part of the block.
2810  *
2811  * Note 1: The size of each block of data returned can be arbitrary,
2812  *       this limitation is the cost for zerocopy sequential
2813  *       reads of potentially non linear data.
2814  *
2815  * Note 2: Fragment lists within fragments are not implemented
2816  *       at the moment, state->root_skb could be replaced with
2817  *       a stack for this purpose.
2818  */
2819 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2820                           struct skb_seq_state *st)
2821 {
2822         unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2823         skb_frag_t *frag;
2824 
2825         if (unlikely(abs_offset >= st->upper_offset)) {
2826                 if (st->frag_data) {
2827                         kunmap_atomic(st->frag_data);
2828                         st->frag_data = NULL;
2829                 }
2830                 return 0;
2831         }
2832 
2833 next_skb:
2834         block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2835 
2836         if (abs_offset < block_limit && !st->frag_data) {
2837                 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2838                 return block_limit - abs_offset;
2839         }
2840 
2841         if (st->frag_idx == 0 && !st->frag_data)
2842                 st->stepped_offset += skb_headlen(st->cur_skb);
2843 
2844         while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2845                 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2846                 block_limit = skb_frag_size(frag) + st->stepped_offset;
2847 
2848                 if (abs_offset < block_limit) {
2849                         if (!st->frag_data)
2850                                 st->frag_data = kmap_atomic(skb_frag_page(frag));
2851 
2852                         *data = (u8 *) st->frag_data + frag->page_offset +
2853                                 (abs_offset - st->stepped_offset);
2854 
2855                         return block_limit - abs_offset;
2856                 }
2857 
2858                 if (st->frag_data) {
2859                         kunmap_atomic(st->frag_data);
2860                         st->frag_data = NULL;
2861                 }
2862 
2863                 st->frag_idx++;
2864                 st->stepped_offset += skb_frag_size(frag);
2865         }
2866 
2867         if (st->frag_data) {
2868                 kunmap_atomic(st->frag_data);
2869                 st->frag_data = NULL;
2870         }
2871 
2872         if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2873                 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2874                 st->frag_idx = 0;
2875                 goto next_skb;
2876         } else if (st->cur_skb->next) {
2877                 st->cur_skb = st->cur_skb->next;
2878                 st->frag_idx = 0;
2879                 goto next_skb;
2880         }
2881 
2882         return 0;
2883 }
2884 EXPORT_SYMBOL(skb_seq_read);
2885 
2886 /**
2887  * skb_abort_seq_read - Abort a sequential read of skb data
2888  * @st: state variable
2889  *
2890  * Must be called if skb_seq_read() was not called until it
2891  * returned 0.
2892  */
2893 void skb_abort_seq_read(struct skb_seq_state *st)
2894 {
2895         if (st->frag_data)
2896                 kunmap_atomic(st->frag_data);
2897 }
2898 EXPORT_SYMBOL(skb_abort_seq_read);
2899 
2900 #define TS_SKB_CB(state)        ((struct skb_seq_state *) &((state)->cb))
2901 
2902 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2903                                           struct ts_config *conf,
2904                                           struct ts_state *state)
2905 {
2906         return skb_seq_read(offset, text, TS_SKB_CB(state));
2907 }
2908 
2909 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2910 {
2911         skb_abort_seq_read(TS_SKB_CB(state));
2912 }
2913 
2914 /**
2915  * skb_find_text - Find a text pattern in skb data
2916  * @skb: the buffer to look in
2917  * @from: search offset
2918  * @to: search limit
2919  * @config: textsearch configuration
2920  *
2921  * Finds a pattern in the skb data according to the specified
2922  * textsearch configuration. Use textsearch_next() to retrieve
2923  * subsequent occurrences of the pattern. Returns the offset
2924  * to the first occurrence or UINT_MAX if no match was found.
2925  */
2926 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2927                            unsigned int to, struct ts_config *config)
2928 {
2929         struct ts_state state;
2930         unsigned int ret;
2931 
2932         config->get_next_block = skb_ts_get_next_block;
2933         config->finish = skb_ts_finish;
2934 
2935         skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
2936 
2937         ret = textsearch_find(config, &state);
2938         return (ret <= to - from ? ret : UINT_MAX);
2939 }
2940 EXPORT_SYMBOL(skb_find_text);
2941 
2942 /**
2943  * skb_append_datato_frags - append the user data to a skb
2944  * @sk: sock  structure
2945  * @skb: skb structure to be appended with user data.
2946  * @getfrag: call back function to be used for getting the user data
2947  * @from: pointer to user message iov
2948  * @length: length of the iov message
2949  *
2950  * Description: This procedure append the user data in the fragment part
2951  * of the skb if any page alloc fails user this procedure returns  -ENOMEM
2952  */
2953 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2954                         int (*getfrag)(void *from, char *to, int offset,
2955                                         int len, int odd, struct sk_buff *skb),
2956                         void *from, int length)
2957 {
2958         int frg_cnt = skb_shinfo(skb)->nr_frags;
2959         int copy;
2960         int offset = 0;
2961         int ret;
2962         struct page_frag *pfrag = &current->task_frag;
2963 
2964         do {
2965                 /* Return error if we don't have space for new frag */
2966                 if (frg_cnt >= MAX_SKB_FRAGS)
2967                         return -EMSGSIZE;
2968 
2969                 if (!sk_page_frag_refill(sk, pfrag))
2970                         return -ENOMEM;
2971 
2972                 /* copy the user data to page */
2973                 copy = min_t(int, length, pfrag->size - pfrag->offset);
2974 
2975                 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
2976                               offset, copy, 0, skb);
2977                 if (ret < 0)
2978                         return -EFAULT;
2979 
2980                 /* copy was successful so update the size parameters */
2981                 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
2982                                    copy);
2983                 frg_cnt++;
2984                 pfrag->offset += copy;
2985                 get_page(pfrag->page);
2986 
2987                 skb->truesize += copy;
2988                 atomic_add(copy, &sk->sk_wmem_alloc);
2989                 skb->len += copy;
2990                 skb->data_len += copy;
2991                 offset += copy;
2992                 length -= copy;
2993 
2994         } while (length > 0);
2995 
2996         return 0;
2997 }
2998 EXPORT_SYMBOL(skb_append_datato_frags);
2999 
3000 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3001                          int offset, size_t size)
3002 {
3003         int i = skb_shinfo(skb)->nr_frags;
3004 
3005         if (skb_can_coalesce(skb, i, page, offset)) {
3006                 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3007         } else if (i < MAX_SKB_FRAGS) {
3008                 get_page(page);
3009                 skb_fill_page_desc(skb, i, page, offset, size);
3010         } else {
3011                 return -EMSGSIZE;
3012         }
3013 
3014         return 0;
3015 }
3016 EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3017 
3018 /**
3019  *      skb_pull_rcsum - pull skb and update receive checksum
3020  *      @skb: buffer to update
3021  *      @len: length of data pulled
3022  *
3023  *      This function performs an skb_pull on the packet and updates
3024  *      the CHECKSUM_COMPLETE checksum.  It should be used on
3025  *      receive path processing instead of skb_pull unless you know
3026  *      that the checksum difference is zero (e.g., a valid IP header)
3027  *      or you are setting ip_summed to CHECKSUM_NONE.
3028  */
3029 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3030 {
3031         unsigned char *data = skb->data;
3032 
3033         BUG_ON(len > skb->len);
3034         __skb_pull(skb, len);
3035         skb_postpull_rcsum(skb, data, len);
3036         return skb->data;
3037 }
3038 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3039 
3040 /**
3041  *      skb_segment - Perform protocol segmentation on skb.
3042  *      @head_skb: buffer to segment
3043  *      @features: features for the output path (see dev->features)
3044  *
3045  *      This function performs segmentation on the given skb.  It returns
3046  *      a pointer to the first in a list of new skbs for the segments.
3047  *      In case of error it returns ERR_PTR(err).
3048  */
3049 struct sk_buff *skb_segment(struct sk_buff *head_skb,
3050                             netdev_features_t features)
3051 {
3052         struct sk_buff *segs = NULL;
3053         struct sk_buff *tail = NULL;
3054         struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3055         skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3056         unsigned int mss = skb_shinfo(head_skb)->gso_size;
3057         unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3058         struct sk_buff *frag_skb = head_skb;
3059         unsigned int offset = doffset;
3060         unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3061         unsigned int partial_segs = 0;
3062         unsigned int headroom;
3063         unsigned int len = head_skb->len;
3064         __be16 proto;
3065         bool csum, sg;
3066         int nfrags = skb_shinfo(head_skb)->nr_frags;
3067         int err = -ENOMEM;
3068         int i = 0;
3069         int pos;
3070         int dummy;
3071 
3072         __skb_push(head_skb, doffset);
3073         proto = skb_network_protocol(head_skb, &dummy);
3074         if (unlikely(!proto))
3075                 return ERR_PTR(-EINVAL);
3076 
3077         sg = !!(features & NETIF_F_SG);
3078         csum = !!can_checksum_protocol(features, proto);
3079 
3080         if (sg && csum && (mss != GSO_BY_FRAGS))  {
3081                 if (!(features & NETIF_F_GSO_PARTIAL)) {
3082                         struct sk_buff *iter;
3083                         unsigned int frag_len;
3084 
3085                         if (!list_skb ||
3086                             !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
3087                                 goto normal;
3088 
3089                         /* If we get here then all the required
3090                          * GSO features except frag_list are supported.
3091                          * Try to split the SKB to multiple GSO SKBs
3092                          * with no frag_list.
3093                          * Currently we can do that only when the buffers don't
3094                          * have a linear part and all the buffers except
3095                          * the last are of the same length.
3096                          */
3097                         frag_len = list_skb->len;
3098                         skb_walk_frags(head_skb, iter) {
3099                                 if (frag_len != iter->len && iter->next)
3100                                         goto normal;
3101                                 if (skb_headlen(iter))
3102                                         goto normal;
3103 
3104                                 len -= iter->len;
3105                         }
3106 
3107                         if (len != frag_len)
3108                                 goto normal;
3109                 }
3110 
3111                 /* GSO partial only requires that we trim off any excess that
3112                  * doesn't fit into an MSS sized block, so take care of that
3113                  * now.
3114                  */
3115                 partial_segs = len / mss;
3116                 if (partial_segs > 1)
3117                         mss *= partial_segs;
3118                 else
3119                         partial_segs = 0;
3120         }
3121 
3122 normal:
3123         headroom = skb_headroom(head_skb);
3124         pos = skb_headlen(head_skb);
3125 
3126         do {
3127                 struct sk_buff *nskb;
3128                 skb_frag_t *nskb_frag;
3129                 int hsize;
3130                 int size;
3131 
3132                 if (unlikely(mss == GSO_BY_FRAGS)) {
3133                         len = list_skb->len;
3134                 } else {
3135                         len = head_skb->len - offset;
3136                         if (len > mss)
3137                                 len = mss;
3138                 }
3139 
3140                 hsize = skb_headlen(head_skb) - offset;
3141                 if (hsize < 0)
3142                         hsize = 0;
3143                 if (hsize > len || !sg)
3144                         hsize = len;
3145 
3146                 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3147                     (skb_headlen(list_skb) == len || sg)) {
3148                         BUG_ON(skb_headlen(list_skb) > len);
3149 
3150                         i = 0;
3151                         nfrags = skb_shinfo(list_skb)->nr_frags;
3152                         frag = skb_shinfo(list_skb)->frags;
3153                         frag_skb = list_skb;
3154                         pos += skb_headlen(list_skb);
3155 
3156                         while (pos < offset + len) {
3157                                 BUG_ON(i >= nfrags);
3158 
3159                                 size = skb_frag_size(frag);
3160                                 if (pos + size > offset + len)
3161                                         break;
3162 
3163                                 i++;
3164                                 pos += size;
3165                                 frag++;
3166                         }
3167 
3168                         nskb = skb_clone(list_skb, GFP_ATOMIC);
3169                         list_skb = list_skb->next;
3170 
3171                         if (unlikely(!nskb))
3172                                 goto err;
3173 
3174                         if (unlikely(pskb_trim(nskb, len))) {
3175                                 kfree_skb(nskb);
3176                                 goto err;
3177                         }
3178 
3179                         hsize = skb_end_offset(nskb);
3180                         if (skb_cow_head(nskb, doffset + headroom)) {
3181                                 kfree_skb(nskb);
3182                                 goto err;
3183                         }
3184 
3185                         nskb->truesize += skb_end_offset(nskb) - hsize;
3186                         skb_release_head_state(nskb);
3187                         __skb_push(nskb, doffset);
3188                 } else {
3189                         nskb = __alloc_skb(hsize + doffset + headroom,
3190                                            GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3191                                            NUMA_NO_NODE);
3192 
3193                         if (unlikely(!nskb))
3194                                 goto err;
3195 
3196                         skb_reserve(nskb, headroom);
3197                         __skb_put(nskb, doffset);
3198                 }
3199 
3200                 if (segs)
3201                         tail->next = nskb;
3202                 else
3203                         segs = nskb;
3204                 tail = nskb;
3205 
3206                 __copy_skb_header(nskb, head_skb);
3207 
3208                 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3209                 skb_reset_mac_len(nskb);
3210 
3211                 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3212                                                  nskb->data - tnl_hlen,
3213                                                  doffset + tnl_hlen);
3214 
3215                 if (nskb->len == len + doffset)
3216                         goto perform_csum_check;
3217 
3218                 if (!sg) {
3219                         if (!nskb->remcsum_offload)
3220                                 nskb->ip_summed = CHECKSUM_NONE;
3221                         SKB_GSO_CB(nskb)->csum =
3222                                 skb_copy_and_csum_bits(head_skb, offset,
3223                                                        skb_put(nskb, len),
3224                                                        len, 0);
3225                         SKB_GSO_CB(nskb)->csum_start =
3226                                 skb_headroom(nskb) + doffset;
3227                         continue;
3228                 }
3229 
3230                 nskb_frag = skb_shinfo(nskb)->frags;
3231 
3232                 skb_copy_from_linear_data_offset(head_skb, offset,
3233                                                  skb_put(nskb, hsize), hsize);
3234 
3235                 skb_shinfo(nskb)->tx_flags = skb_shinfo(head_skb)->tx_flags &
3236                         SKBTX_SHARED_FRAG;
3237 
3238                 while (pos < offset + len) {
3239                         if (i >= nfrags) {
3240                                 BUG_ON(skb_headlen(list_skb));
3241 
3242                                 i = 0;
3243                                 nfrags = skb_shinfo(list_skb)->nr_frags;
3244                                 frag = skb_shinfo(list_skb)->frags;
3245                                 frag_skb = list_skb;
3246 
3247                                 BUG_ON(!nfrags);
3248 
3249                                 list_skb = list_skb->next;
3250                         }
3251 
3252                         if (unlikely(skb_shinfo(nskb)->nr_frags >=
3253                                      MAX_SKB_FRAGS)) {
3254                                 net_warn_ratelimited(
3255                                         "skb_segment: too many frags: %u %u\n",
3256                                         pos, mss);
3257                                 goto err;
3258                         }
3259 
3260                         if (unlikely(skb_orphan_frags(frag_skb, GFP_ATOMIC)))
3261                                 goto err;
3262 
3263                         *nskb_frag = *frag;
3264                         __skb_frag_ref(nskb_frag);
3265                         size = skb_frag_size(nskb_frag);
3266 
3267                         if (pos < offset) {
3268                                 nskb_frag->page_offset += offset - pos;
3269                                 skb_frag_size_sub(nskb_frag, offset - pos);
3270                         }
3271 
3272                         skb_shinfo(nskb)->nr_frags++;
3273 
3274                         if (pos + size <= offset + len) {
3275                                 i++;
3276                                 frag++;
3277                                 pos += size;
3278                         } else {
3279                                 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
3280                                 goto skip_fraglist;
3281                         }
3282 
3283                         nskb_frag++;
3284                 }
3285 
3286 skip_fraglist:
3287                 nskb->data_len = len - hsize;
3288                 nskb->len += nskb->data_len;
3289                 nskb->truesize += nskb->data_len;
3290 
3291 perform_csum_check:
3292                 if (!csum) {
3293                         if (skb_has_shared_frag(nskb)) {
3294                                 err = __skb_linearize(nskb);
3295                                 if (err)
3296                                         goto err;
3297                         }
3298                         if (!nskb->remcsum_offload)
3299                                 nskb->ip_summed = CHECKSUM_NONE;
3300                         SKB_GSO_CB(nskb)->csum =
3301                                 skb_checksum(nskb, doffset,
3302                                              nskb->len - doffset, 0);
3303                         SKB_GSO_CB(nskb)->csum_start =
3304                                 skb_headroom(nskb) + doffset;
3305                 }
3306         } while ((offset += len) < head_skb->len);
3307 
3308         /* Some callers want to get the end of the list.
3309          * Put it in segs->prev to avoid walking the list.
3310          * (see validate_xmit_skb_list() for example)
3311          */
3312         segs->prev = tail;
3313 
3314         if (partial_segs) {
3315                 struct sk_buff *iter;
3316                 int type = skb_shinfo(head_skb)->gso_type;
3317                 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
3318 
3319                 /* Update type to add partial and then remove dodgy if set */
3320                 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
3321                 type &= ~SKB_GSO_DODGY;
3322 
3323                 /* Update GSO info and prepare to start updating headers on
3324                  * our way back down the stack of protocols.
3325                  */
3326                 for (iter = segs; iter; iter = iter->next) {
3327                         skb_shinfo(iter)->gso_size = gso_size;
3328                         skb_shinfo(iter)->gso_segs = partial_segs;
3329                         skb_shinfo(iter)->gso_type = type;
3330                         SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
3331                 }
3332 
3333                 if (tail->len - doffset <= gso_size)
3334                         skb_shinfo(tail)->gso_size = 0;
3335                 else if (tail != segs)
3336                         skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
3337         }
3338 
3339         /* Following permits correct backpressure, for protocols
3340          * using skb_set_owner_w().
3341          * Idea is to tranfert ownership from head_skb to last segment.
3342          */
3343         if (head_skb->destructor == sock_wfree) {
3344                 swap(tail->truesize, head_skb->truesize);
3345                 swap(tail->destructor, head_skb->destructor);
3346                 swap(tail->sk, head_skb->sk);
3347         }
3348         return segs;
3349 
3350 err:
3351         kfree_skb_list(segs);
3352         return ERR_PTR(err);
3353 }
3354 EXPORT_SYMBOL_GPL(skb_segment);
3355 
3356 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
3357 {
3358         struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
3359         unsigned int offset = skb_gro_offset(skb);
3360         unsigned int headlen = skb_headlen(skb);
3361         unsigned int len = skb_gro_len(skb);
3362         struct sk_buff *lp, *p = *head;
3363         unsigned int delta_truesize;
3364 
3365         if (unlikely(p->len + len >= 65536))
3366                 return -E2BIG;
3367 
3368         lp = NAPI_GRO_CB(p)->last;
3369         pinfo = skb_shinfo(lp);
3370 
3371         if (headlen <= offset) {
3372                 skb_frag_t *frag;
3373                 skb_frag_t *frag2;
3374                 int i = skbinfo->nr_frags;
3375                 int nr_frags = pinfo->nr_frags + i;
3376 
3377                 if (nr_frags > MAX_SKB_FRAGS)
3378                         goto merge;
3379 
3380                 offset -= headlen;
3381                 pinfo->nr_frags = nr_frags;
3382                 skbinfo->nr_frags = 0;
3383 
3384                 frag = pinfo->frags + nr_frags;
3385                 frag2 = skbinfo->frags + i;
3386                 do {
3387                         *--frag = *--frag2;
3388                 } while (--i);
3389 
3390                 frag->page_offset += offset;
3391                 skb_frag_size_sub(frag, offset);
3392 
3393                 /* all fragments truesize : remove (head size + sk_buff) */
3394                 delta_truesize = skb->truesize -
3395                                  SKB_TRUESIZE(skb_end_offset(skb));
3396 
3397                 skb->truesize -= skb->data_len;
3398                 skb->len -= skb->data_len;
3399                 skb->data_len = 0;
3400 
3401                 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
3402                 goto done;
3403         } else if (skb->head_frag) {
3404                 int nr_frags = pinfo->nr_frags;
3405                 skb_frag_t *frag = pinfo->frags + nr_frags;
3406                 struct page *page = virt_to_head_page(skb->head);
3407                 unsigned int first_size = headlen - offset;
3408                 unsigned int first_offset;
3409 
3410                 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
3411                         goto merge;
3412 
3413                 first_offset = skb->data -
3414                                (unsigned char *)page_address(page) +
3415                                offset;
3416 
3417                 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3418 
3419                 frag->page.p      = page;
3420                 frag->page_offset = first_offset;
3421                 skb_frag_size_set(frag, first_size);
3422 
3423                 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3424                 /* We dont need to clear skbinfo->nr_frags here */
3425 
3426                 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3427                 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3428                 goto done;
3429         }
3430 
3431 merge:
3432         delta_truesize = skb->truesize;
3433         if (offset > headlen) {
3434                 unsigned int eat = offset - headlen;
3435 
3436                 skbinfo->frags[0].page_offset += eat;
3437                 skb_frag_size_sub(&skbinfo->frags[0], eat);
3438                 skb->data_len -= eat;
3439                 skb->len -= eat;
3440                 offset = headlen;
3441         }
3442 
3443         __skb_pull(skb, offset);
3444 
3445         if (NAPI_GRO_CB(p)->last == p)
3446                 skb_shinfo(p)->frag_list = skb;
3447         else
3448                 NAPI_GRO_CB(p)->last->next = skb;
3449         NAPI_GRO_CB(p)->last = skb;
3450         __skb_header_release(skb);
3451         lp = p;
3452 
3453 done:
3454         NAPI_GRO_CB(p)->count++;
3455         p->data_len += len;
3456         p->truesize += delta_truesize;
3457         p->len += len;
3458         if (lp != p) {
3459                 lp->data_len += len;
3460                 lp->truesize += delta_truesize;
3461                 lp->len += len;
3462         }
3463         NAPI_GRO_CB(skb)->same_flow = 1;
3464         return 0;
3465 }
3466 EXPORT_SYMBOL_GPL(skb_gro_receive);
3467 
3468 void __init skb_init(void)
3469 {
3470         skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3471                                               sizeof(struct sk_buff),
3472                                               0,
3473                                               SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3474                                               NULL);
3475         skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3476                                                 sizeof(struct sk_buff_fclones),
3477                                                 0,
3478                                                 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3479                                                 NULL);
3480 }
3481 
3482 /**
3483  *      skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3484  *      @skb: Socket buffer containing the buffers to be mapped
3485  *      @sg: The scatter-gather list to map into
3486  *      @offset: The offset into the buffer's contents to start mapping
3487  *      @len: Length of buffer space to be mapped
3488  *
3489  *      Fill the specified scatter-gather list with mappings/pointers into a
3490  *      region of the buffer space attached to a socket buffer.
3491  */
3492 static int
3493 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3494 {
3495         int start = skb_headlen(skb);
3496         int i, copy = start - offset;
3497         struct sk_buff *frag_iter;
3498         int elt = 0;
3499 
3500         if (copy > 0) {
3501                 if (copy > len)
3502                         copy = len;
3503                 sg_set_buf(sg, skb->data + offset, copy);
3504                 elt++;
3505                 if ((len -= copy) == 0)
3506                         return elt;
3507                 offset += copy;
3508         }
3509 
3510         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3511                 int end;
3512 
3513                 WARN_ON(start > offset + len);
3514 
3515                 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3516                 if ((copy = end - offset) > 0) {
3517                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3518 
3519                         if (copy > len)
3520                                 copy = len;
3521                         sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3522                                         frag->page_offset+offset-start);
3523                         elt++;
3524                         if (!(len -= copy))
3525                                 return elt;
3526                         offset += copy;
3527                 }
3528                 start = end;
3529         }
3530 
3531         skb_walk_frags(skb, frag_iter) {
3532                 int end;
3533 
3534                 WARN_ON(start > offset + len);
3535 
3536                 end = start + frag_iter->len;
3537                 if ((copy = end - offset) > 0) {
3538                         if (copy > len)
3539                                 copy = len;
3540                         elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3541                                               copy);
3542                         if ((len -= copy) == 0)
3543                                 return elt;
3544                         offset += copy;
3545                 }
3546                 start = end;
3547         }
3548         BUG_ON(len);
3549         return elt;
3550 }
3551 
3552 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
3553  * sglist without mark the sg which contain last skb data as the end.
3554  * So the caller can mannipulate sg list as will when padding new data after
3555  * the first call without calling sg_unmark_end to expend sg list.
3556  *
3557  * Scenario to use skb_to_sgvec_nomark:
3558  * 1. sg_init_table
3559  * 2. skb_to_sgvec_nomark(payload1)
3560  * 3. skb_to_sgvec_nomark(payload2)
3561  *
3562  * This is equivalent to:
3563  * 1. sg_init_table
3564  * 2. skb_to_sgvec(payload1)
3565  * 3. sg_unmark_end
3566  * 4. skb_to_sgvec(payload2)
3567  *
3568  * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
3569  * is more preferable.
3570  */
3571 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
3572                         int offset, int len)
3573 {
3574         return __skb_to_sgvec(skb, sg, offset, len);
3575 }
3576 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
3577 
3578 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3579 {
3580         int nsg = __skb_to_sgvec(skb, sg, offset, len);
3581 
3582         sg_mark_end(&sg[nsg - 1]);
3583 
3584         return nsg;
3585 }
3586 EXPORT_SYMBOL_GPL(skb_to_sgvec);
3587 
3588 /**
3589  *      skb_cow_data - Check that a socket buffer's data buffers are writable
3590  *      @skb: The socket buffer to check.
3591  *      @tailbits: Amount of trailing space to be added
3592  *      @trailer: Returned pointer to the skb where the @tailbits space begins
3593  *
3594  *      Make sure that the data buffers attached to a socket buffer are
3595  *      writable. If they are not, private copies are made of the data buffers
3596  *      and the socket buffer is set to use these instead.
3597  *
3598  *      If @tailbits is given, make sure that there is space to write @tailbits
3599  *      bytes of data beyond current end of socket buffer.  @trailer will be
3600  *      set to point to the skb in which this space begins.
3601  *
3602  *      The number of scatterlist elements required to completely map the
3603  *      COW'd and extended socket buffer will be returned.
3604  */
3605 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3606 {
3607         int copyflag;
3608         int elt;
3609         struct sk_buff *skb1, **skb_p;
3610 
3611         /* If skb is cloned or its head is paged, reallocate
3612          * head pulling out all the pages (pages are considered not writable
3613          * at the moment even if they are anonymous).
3614          */
3615         if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3616             __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3617                 return -ENOMEM;
3618 
3619         /* Easy case. Most of packets will go this way. */
3620         if (!skb_has_frag_list(skb)) {
3621                 /* A little of trouble, not enough of space for trailer.
3622                  * This should not happen, when stack is tuned to generate
3623                  * good frames. OK, on miss we reallocate and reserve even more
3624                  * space, 128 bytes is fair. */
3625 
3626                 if (skb_tailroom(skb) < tailbits &&
3627                     pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3628                         return -ENOMEM;
3629 
3630                 /* Voila! */
3631                 *trailer = skb;
3632                 return 1;
3633         }
3634 
3635         /* Misery. We are in troubles, going to mincer fragments... */
3636 
3637         elt = 1;
3638         skb_p = &skb_shinfo(skb)->frag_list;
3639         copyflag = 0;
3640 
3641         while ((skb1 = *skb_p) != NULL) {
3642                 int ntail = 0;
3643 
3644                 /* The fragment is partially pulled by someone,
3645                  * this can happen on input. Copy it and everything
3646                  * after it. */
3647 
3648                 if (skb_shared(skb1))
3649                         copyflag = 1;
3650 
3651                 /* If the skb is the last, worry about trailer. */
3652 
3653                 if (skb1->next == NULL && tailbits) {
3654                         if (skb_shinfo(skb1)->nr_frags ||
3655                             skb_has_frag_list(skb1) ||
3656                             skb_tailroom(skb1) < tailbits)
3657                                 ntail = tailbits + 128;
3658                 }
3659 
3660                 if (copyflag ||
3661                     skb_cloned(skb1) ||
3662                     ntail ||
3663                     skb_shinfo(skb1)->nr_frags ||
3664                     skb_has_frag_list(skb1)) {
3665                         struct sk_buff *skb2;
3666 
3667                         /* Fuck, we are miserable poor guys... */
3668                         if (ntail == 0)
3669                                 skb2 = skb_copy(skb1, GFP_ATOMIC);
3670                         else
3671                                 skb2 = skb_copy_expand(skb1,
3672                                                        skb_headroom(skb1),
3673                                                        ntail,
3674                                                        GFP_ATOMIC);
3675                         if (unlikely(skb2 == NULL))
3676                                 return -ENOMEM;
3677 
3678                         if (skb1->sk)
3679                                 skb_set_owner_w(skb2, skb1->sk);
3680 
3681                         /* Looking around. Are we still alive?
3682                          * OK, link new skb, drop old one */
3683 
3684                         skb2->next = skb1->next;
3685                         *skb_p = skb2;
3686                         kfree_skb(skb1);
3687                         skb1 = skb2;
3688                 }
3689                 elt++;
3690                 *trailer = skb1;
3691                 skb_p = &skb1->next;
3692         }
3693 
3694         return elt;
3695 }
3696 EXPORT_SYMBOL_GPL(skb_cow_data);
3697 
3698 static void sock_rmem_free(struct sk_buff *skb)
3699 {
3700         struct sock *sk = skb->sk;
3701 
3702         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3703 }
3704 
3705 static void skb_set_err_queue(struct sk_buff *skb)
3706 {
3707         /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
3708          * So, it is safe to (mis)use it to mark skbs on the error queue.
3709          */
3710         skb->pkt_type = PACKET_OUTGOING;
3711         BUILD_BUG_ON(PACKET_OUTGOING == 0);
3712 }
3713 
3714 /*
3715  * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3716  */
3717 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3718 {
3719         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3720             (unsigned int)sk->sk_rcvbuf)
3721                 return -ENOMEM;
3722 
3723         skb_orphan(skb);
3724         skb->sk = sk;
3725         skb->destructor = sock_rmem_free;
3726         atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3727         skb_set_err_queue(skb);
3728 
3729         /* before exiting rcu section, make sure dst is refcounted */
3730         skb_dst_force(skb);
3731 
3732         skb_queue_tail(&sk->sk_error_queue, skb);
3733         if (!sock_flag(sk, SOCK_DEAD))
3734                 sk->sk_data_ready(sk);
3735         return 0;
3736 }
3737 EXPORT_SYMBOL(sock_queue_err_skb);
3738 
3739 static bool is_icmp_err_skb(const struct sk_buff *skb)
3740 {
3741         return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
3742                        SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
3743 }
3744 
3745 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
3746 {
3747         struct sk_buff_head *q = &sk->sk_error_queue;
3748         struct sk_buff *skb, *skb_next = NULL;
3749         bool icmp_next = false;
3750         unsigned long flags;
3751 
3752         spin_lock_irqsave(&q->lock, flags);
3753         skb = __skb_dequeue(q);
3754         if (skb && (skb_next = skb_peek(q)))
3755                 icmp_next = is_icmp_err_skb(skb_next);
3756         spin_unlock_irqrestore(&q->lock, flags);
3757 
3758         if (is_icmp_err_skb(skb) && !icmp_next)
3759                 sk->sk_err = 0;
3760 
3761         if (skb_next)
3762                 sk->sk_error_report(sk);
3763 
3764         return skb;
3765 }
3766 EXPORT_SYMBOL(sock_dequeue_err_skb);
3767 
3768 /**
3769  * skb_clone_sk - create clone of skb, and take reference to socket
3770  * @skb: the skb to clone
3771  *
3772  * This function creates a clone of a buffer that holds a reference on
3773  * sk_refcnt.  Buffers created via this function are meant to be
3774  * returned using sock_queue_err_skb, or free via kfree_skb.
3775  *
3776  * When passing buffers allocated with this function to sock_queue_err_skb
3777  * it is necessary to wrap the call with sock_hold/sock_put in order to
3778  * prevent the socket from being released prior to being enqueued on
3779  * the sk_error_queue.
3780  */
3781 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
3782 {
3783         struct sock *sk = skb->sk;
3784         struct sk_buff *clone;
3785 
3786         if (!sk || !atomic_inc_not_zero(&sk->sk_refcnt))
3787                 return NULL;
3788 
3789         clone = skb_clone(skb, GFP_ATOMIC);
3790         if (!clone) {
3791                 sock_put(sk);
3792                 return NULL;
3793         }
3794 
3795         clone->sk = sk;
3796         clone->destructor = sock_efree;
3797 
3798         return clone;
3799 }
3800 EXPORT_SYMBOL(skb_clone_sk);
3801 
3802 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
3803                                         struct sock *sk,
3804                                         int tstype,
3805                                         bool opt_stats)
3806 {
3807         struct sock_exterr_skb *serr;
3808         int err;
3809 
3810         BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
3811 
3812         serr = SKB_EXT_ERR(skb);
3813         memset(serr, 0, sizeof(*serr));
3814         serr->ee.ee_errno = ENOMSG;
3815         serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3816         serr->ee.ee_info = tstype;
3817         serr->opt_stats = opt_stats;
3818         serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
3819         if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
3820                 serr->ee.ee_data = skb_shinfo(skb)->tskey;
3821                 if (sk->sk_protocol == IPPROTO_TCP &&
3822                     sk->sk_type == SOCK_STREAM)
3823                         serr->ee.ee_data -= sk->sk_tskey;
3824         }
3825 
3826         err = sock_queue_err_skb(sk, skb);
3827 
3828         if (err)
3829                 kfree_skb(skb);
3830 }
3831 
3832 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
3833 {
3834         bool ret;
3835 
3836         if (likely(sysctl_tstamp_allow_data || tsonly))
3837                 return true;
3838 
3839         read_lock_bh(&sk->sk_callback_lock);
3840         ret = sk->sk_socket && sk->sk_socket->file &&
3841               file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
3842         read_unlock_bh(&sk->sk_callback_lock);
3843         return ret;
3844 }
3845 
3846 void skb_complete_tx_timestamp(struct sk_buff *skb,
3847                                struct skb_shared_hwtstamps *hwtstamps)
3848 {
3849         struct sock *sk = skb->sk;
3850 
3851         if (!skb_may_tx_timestamp(sk, false))
3852                 return;
3853 
3854         /* Take a reference to prevent skb_orphan() from freeing the socket,
3855          * but only if the socket refcount is not zero.
3856          */
3857         if (likely(atomic_inc_not_zero(&sk->sk_refcnt))) {
3858                 *skb_hwtstamps(skb) = *hwtstamps;
3859                 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
3860                 sock_put(sk);
3861         }
3862 }
3863 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
3864 
3865 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3866                      struct skb_shared_hwtstamps *hwtstamps,
3867                      struct sock *sk, int tstype)
3868 {
3869         struct sk_buff *skb;
3870         bool tsonly, opt_stats = false;
3871 
3872         if (!sk)
3873                 return;
3874 
3875         tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
3876         if (!skb_may_tx_timestamp(sk, tsonly))
3877                 return;
3878 
3879         if (tsonly) {
3880 #ifdef CONFIG_INET
3881                 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
3882                     sk->sk_protocol == IPPROTO_TCP &&
3883                     sk->sk_type == SOCK_STREAM) {
3884                         skb = tcp_get_timestamping_opt_stats(sk);
3885                         opt_stats = true;
3886                 } else
3887 #endif
3888                         skb = alloc_skb(0, GFP_ATOMIC);
3889         } else {
3890                 skb = skb_clone(orig_skb, GFP_ATOMIC);
3891         }
3892         if (!skb)
3893                 return;
3894 
3895         if (tsonly) {
3896                 skb_shinfo(skb)->tx_flags = skb_shinfo(orig_skb)->tx_flags;
3897                 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
3898         }
3899 
3900         if (hwtstamps)
3901                 *skb_hwtstamps(skb) = *hwtstamps;
3902         else
3903                 skb->tstamp = ktime_get_real();
3904 
3905         __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
3906 }
3907 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
3908 
3909 void skb_tstamp_tx(struct sk_buff *orig_skb,
3910                    struct skb_shared_hwtstamps *hwtstamps)
3911 {
3912         return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
3913                                SCM_TSTAMP_SND);
3914 }
3915 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3916 
3917 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3918 {
3919         struct sock *sk = skb->sk;
3920         struct sock_exterr_skb *serr;
3921         int err = 1;
3922 
3923         skb->wifi_acked_valid = 1;
3924         skb->wifi_acked = acked;
3925 
3926         serr = SKB_EXT_ERR(skb);
3927         memset(serr, 0, sizeof(*serr));
3928         serr->ee.ee_errno = ENOMSG;
3929         serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3930 
3931         /* Take a reference to prevent skb_orphan() from freeing the socket,
3932          * but only if the socket refcount is not zero.
3933          */
3934         if (likely(atomic_inc_not_zero(&sk->sk_refcnt))) {
3935                 err = sock_queue_err_skb(sk, skb);
3936                 sock_put(sk);
3937         }
3938         if (err)
3939                 kfree_skb(skb);
3940 }
3941 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3942 
3943 /**
3944  * skb_partial_csum_set - set up and verify partial csum values for packet
3945  * @skb: the skb to set
3946  * @start: the number of bytes after skb->data to start checksumming.
3947  * @off: the offset from start to place the checksum.
3948  *
3949  * For untrusted partially-checksummed packets, we need to make sure the values
3950  * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3951  *
3952  * This function checks and sets those values and skb->ip_summed: if this
3953  * returns false you should drop the packet.
3954  */
3955 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3956 {
3957         if (unlikely(start > skb_headlen(skb)) ||
3958             unlikely((int)start + off > skb_headlen(skb) - 2)) {
3959                 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3960                                      start, off, skb_headlen(skb));
3961                 return false;
3962         }
3963         skb->ip_summed = CHECKSUM_PARTIAL;
3964         skb->csum_start = skb_headroom(skb) + start;
3965         skb->csum_offset = off;
3966         skb_set_transport_header(skb, start);
3967         return true;
3968 }
3969 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3970 
3971 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
3972                                unsigned int max)
3973 {
3974         if (skb_headlen(skb) >= len)
3975                 return 0;
3976 
3977         /* If we need to pullup then pullup to the max, so we
3978          * won't need to do it again.
3979          */
3980         if (max > skb->len)
3981                 max = skb->len;
3982 
3983         if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
3984                 return -ENOMEM;
3985 
3986         if (skb_headlen(skb) < len)
3987                 return -EPROTO;
3988 
3989         return 0;
3990 }
3991 
3992 #define MAX_TCP_HDR_LEN (15 * 4)
3993 
3994 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
3995                                       typeof(IPPROTO_IP) proto,
3996                                       unsigned int off)
3997 {
3998         switch (proto) {
3999                 int err;
4000 
4001         case IPPROTO_TCP:
4002                 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
4003                                           off + MAX_TCP_HDR_LEN);
4004                 if (!err && !skb_partial_csum_set(skb, off,
4005                                                   offsetof(struct tcphdr,
4006                                                            check)))
4007                         err = -EPROTO;
4008                 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
4009 
4010         case IPPROTO_UDP:
4011                 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
4012                                           off + sizeof(struct udphdr));
4013                 if (!err && !skb_partial_csum_set(skb, off,
4014                                                   offsetof(struct udphdr,
4015                                                            check)))
4016                         err = -EPROTO;
4017                 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
4018         }
4019 
4020         return ERR_PTR(-EPROTO);
4021 }
4022 
4023 /* This value should be large enough to cover a tagged ethernet header plus
4024  * maximally sized IP and TCP or UDP headers.
4025  */
4026 #define MAX_IP_HDR_LEN 128
4027 
4028 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
4029 {
4030         unsigned int off;
4031         bool fragment;
4032         __sum16 *csum;
4033         int err;
4034 
4035         fragment = false;
4036 
4037         err = skb_maybe_pull_tail(skb,
4038                                   sizeof(struct iphdr),
4039                                   MAX_IP_HDR_LEN);
4040         if (err < 0)
4041                 goto out;
4042 
4043         if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
4044                 fragment = true;
4045 
4046         off = ip_hdrlen(skb);
4047 
4048         err = -EPROTO;
4049 
4050         if (fragment)
4051                 goto out;
4052 
4053         csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
4054         if (IS_ERR(csum))
4055                 return PTR_ERR(csum);
4056 
4057         if (recalculate)
4058                 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
4059                                            ip_hdr(skb)->daddr,
4060                                            skb->len - off,
4061                                            ip_hdr(skb)->protocol, 0);
4062         err = 0;
4063 
4064 out:
4065         return err;
4066 }
4067 
4068 /* This value should be large enough to cover a tagged ethernet header plus
4069  * an IPv6 header, all options, and a maximal TCP or UDP header.
4070  */
4071 #define MAX_IPV6_HDR_LEN 256
4072 
4073 #define OPT_HDR(type, skb, off) \
4074         (type *)(skb_network_header(skb) + (off))
4075 
4076 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
4077 {
4078         int err;
4079         u8 nexthdr;
4080         unsigned int off;
4081         unsigned int len;
4082         bool fragment;
4083         bool done;
4084         __sum16 *csum;
4085 
4086         fragment = false;
4087         done = false;
4088 
4089         off = sizeof(struct ipv6hdr);
4090 
4091         err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4092         if (err < 0)
4093                 goto out;
4094 
4095         nexthdr = ipv6_hdr(skb)->nexthdr;
4096 
4097         len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4098         while (off <= len && !done) {
4099                 switch (nexthdr) {
4100                 case IPPROTO_DSTOPTS:
4101                 case IPPROTO_HOPOPTS:
4102                 case IPPROTO_ROUTING: {
4103                         struct ipv6_opt_hdr *hp;
4104 
4105                         err = skb_maybe_pull_tail(skb,
4106                                                   off +
4107                                                   sizeof(struct ipv6_opt_hdr),
4108                                                   MAX_IPV6_HDR_LEN);
4109                         if (err < 0)
4110                                 goto out;
4111 
4112                         hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
4113                         nexthdr = hp->nexthdr;
4114                         off += ipv6_optlen(hp);
4115                         break;
4116                 }
4117                 case IPPROTO_AH: {
4118                         struct ip_auth_hdr *hp;
4119 
4120                         err = skb_maybe_pull_tail(skb,
4121                                                   off +
4122                                                   sizeof(struct ip_auth_hdr),
4123                                                   MAX_IPV6_HDR_LEN);
4124                         if (err < 0)
4125                                 goto out;
4126 
4127                         hp = OPT_HDR(struct ip_auth_hdr, skb, off);
4128                         nexthdr = hp->nexthdr;
4129                         off += ipv6_authlen(hp);
4130                         break;
4131                 }
4132                 case IPPROTO_FRAGMENT: {
4133                         struct frag_hdr *hp;
4134 
4135                         err = skb_maybe_pull_tail(skb,
4136                                                   off +
4137                                                   sizeof(struct frag_hdr),
4138                                                   MAX_IPV6_HDR_LEN);
4139                         if (err < 0)
4140                                 goto out;
4141 
4142                         hp = OPT_HDR(struct frag_hdr, skb, off);
4143 
4144                         if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
4145                                 fragment = true;
4146 
4147                         nexthdr = hp->nexthdr;
4148                         off += sizeof(struct frag_hdr);
4149                         break;
4150                 }
4151                 default:
4152                         done = true;
4153                         break;
4154                 }
4155         }
4156 
4157         err = -EPROTO;
4158 
4159         if (!done || fragment)
4160                 goto out;
4161 
4162         csum = skb_checksum_setup_ip(skb, nexthdr, off);
4163         if (IS_ERR(csum))
4164                 return PTR_ERR(csum);
4165 
4166         if (recalculate)
4167                 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4168                                          &ipv6_hdr(skb)->daddr,
4169                                          skb->len - off, nexthdr, 0);
4170         err = 0;
4171 
4172 out:
4173         return err;
4174 }
4175 
4176 /**
4177  * skb_checksum_setup - set up partial checksum offset
4178  * @skb: the skb to set up
4179  * @recalculate: if true the pseudo-header checksum will be recalculated
4180  */
4181 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
4182 {
4183         int err;
4184 
4185         switch (skb->protocol) {
4186         case htons(ETH_P_IP):
4187                 err = skb_checksum_setup_ipv4(skb, recalculate);
4188                 break;
4189 
4190         case htons(ETH_P_IPV6):
4191                 err = skb_checksum_setup_ipv6(skb, recalculate);
4192                 break;
4193 
4194         default:
4195                 err = -EPROTO;
4196                 break;
4197         }
4198 
4199         return err;
4200 }
4201 EXPORT_SYMBOL(skb_checksum_setup);
4202 
4203 /**
4204  * skb_checksum_maybe_trim - maybe trims the given skb
4205  * @skb: the skb to check
4206  * @transport_len: the data length beyond the network header
4207  *
4208  * Checks whether the given skb has data beyond the given transport length.
4209  * If so, returns a cloned skb trimmed to this transport length.
4210  * Otherwise returns the provided skb. Returns NULL in error cases
4211  * (e.g. transport_len exceeds skb length or out-of-memory).
4212  *
4213  * Caller needs to set the skb transport header and free any returned skb if it
4214  * differs from the provided skb.
4215  */
4216 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
4217                                                unsigned int transport_len)
4218 {
4219         struct sk_buff *skb_chk;
4220         unsigned int len = skb_transport_offset(skb) + transport_len;
4221         int ret;
4222 
4223         if (skb->len < len)
4224                 return NULL;
4225         else if (skb->len == len)
4226                 return skb;
4227 
4228         skb_chk = skb_clone(skb, GFP_ATOMIC);
4229         if (!skb_chk)
4230                 return NULL;
4231 
4232         ret = pskb_trim_rcsum(skb_chk, len);
4233         if (ret) {
4234                 kfree_skb(skb_chk);
4235                 return NULL;
4236         }
4237 
4238         return skb_chk;
4239 }
4240 
4241 /**
4242  * skb_checksum_trimmed - validate checksum of an skb
4243  * @skb: the skb to check
4244  * @transport_len: the data length beyond the network header
4245  * @skb_chkf: checksum function to use
4246  *
4247  * Applies the given checksum function skb_chkf to the provided skb.
4248  * Returns a checked and maybe trimmed skb. Returns NULL on error.
4249  *
4250  * If the skb has data beyond the given transport length, then a
4251  * trimmed & cloned skb is checked and returned.
4252  *
4253  * Caller needs to set the skb transport header and free any returned skb if it
4254  * differs from the provided skb.
4255  */
4256 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4257                                      unsigned int transport_len,
4258                                      __sum16(*skb_chkf)(struct sk_buff *skb))
4259 {
4260         struct sk_buff *skb_chk;
4261         unsigned int offset = skb_transport_offset(skb);
4262         __sum16 ret;
4263 
4264         skb_chk = skb_checksum_maybe_trim(skb, transport_len);
4265         if (!skb_chk)
4266                 goto err;
4267 
4268         if (!pskb_may_pull(skb_chk, offset))
4269                 goto err;
4270 
4271         skb_pull_rcsum(skb_chk, offset);
4272         ret = skb_chkf(skb_chk);
4273         skb_push_rcsum(skb_chk, offset);
4274 
4275         if (ret)
4276                 goto err;
4277 
4278         return skb_chk;
4279 
4280 err:
4281         if (skb_chk && skb_chk != skb)
4282                 kfree_skb(skb_chk);
4283 
4284         return NULL;
4285 
4286 }
4287 EXPORT_SYMBOL(skb_checksum_trimmed);
4288 
4289 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
4290 {
4291         net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
4292                              skb->dev->name);
4293 }
4294 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
4295 
4296 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
4297 {
4298         if (head_stolen) {
4299                 skb_release_head_state(skb);
4300                 kmem_cache_free(skbuff_head_cache, skb);
4301         } else {
4302                 __kfree_skb(skb);
4303         }
4304 }
4305 EXPORT_SYMBOL(kfree_skb_partial);
4306 
4307 /**
4308  * skb_try_coalesce - try to merge skb to prior one
4309  * @to: prior buffer
4310  * @from: buffer to add
4311  * @fragstolen: pointer to boolean
4312  * @delta_truesize: how much more was allocated than was requested
4313  */
4314 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
4315                       bool *fragstolen, int *delta_truesize)
4316 {
4317         int i, delta, len = from->len;
4318 
4319         *fragstolen = false;
4320 
4321         if (skb_cloned(to))
4322                 return false;
4323 
4324         if (len <= skb_tailroom(to)) {
4325                 if (len)
4326                         BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
4327                 *delta_truesize = 0;
4328                 return true;
4329         }
4330 
4331         if (skb_has_frag_list(to) || skb_has_frag_list(from))
4332                 return false;
4333 
4334         if (skb_headlen(from) != 0) {
4335                 struct page *page;
4336                 unsigned int offset;
4337 
4338                 if (skb_shinfo(to)->nr_frags +
4339                     skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
4340                         return false;
4341 
4342                 if (skb_head_is_locked(from))
4343                         return false;
4344 
4345                 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4346 
4347                 page = virt_to_head_page(from->head);
4348                 offset = from->data - (unsigned char *)page_address(page);
4349 
4350                 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
4351                                    page, offset, skb_headlen(from));
4352                 *fragstolen = true;
4353         } else {
4354                 if (skb_shinfo(to)->nr_frags +
4355                     skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
4356                         return false;
4357 
4358                 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
4359         }
4360 
4361         WARN_ON_ONCE(delta < len);
4362 
4363         memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
4364                skb_shinfo(from)->frags,
4365                skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
4366         skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
4367 
4368         if (!skb_cloned(from))
4369                 skb_shinfo(from)->nr_frags = 0;
4370 
4371         /* if the skb is not cloned this does nothing
4372          * since we set nr_frags to 0.
4373          */
4374         for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
4375                 skb_frag_ref(from, i);
4376 
4377         to->truesize += delta;
4378         to->len += len;
4379         to->data_len += len;
4380 
4381         *delta_truesize = delta;
4382         return true;
4383 }
4384 EXPORT_SYMBOL(skb_try_coalesce);
4385 
4386 /**
4387  * skb_scrub_packet - scrub an skb
4388  *
4389  * @skb: buffer to clean
4390  * @xnet: packet is crossing netns
4391  *
4392  * skb_scrub_packet can be used after encapsulating or decapsulting a packet
4393  * into/from a tunnel. Some information have to be cleared during these
4394  * operations.
4395  * skb_scrub_packet can also be used to clean a skb before injecting it in
4396  * another namespace (@xnet == true). We have to clear all information in the
4397  * skb that could impact namespace isolation.
4398  */
4399 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
4400 {
4401         skb->tstamp = 0;
4402         skb->pkt_type = PACKET_HOST;
4403         skb->skb_iif = 0;
4404         skb->ignore_df = 0;
4405         skb_dst_drop(skb);
4406         secpath_reset(skb);
4407         nf_reset(skb);
4408         nf_reset_trace(skb);
4409 
4410         if (!xnet)
4411                 return;
4412 
4413         skb_orphan(skb);
4414         skb->mark = 0;
4415 }
4416 EXPORT_SYMBOL_GPL(skb_scrub_packet);
4417 
4418 /**
4419  * skb_gso_transport_seglen - Return length of individual segments of a gso packet
4420  *
4421  * @skb: GSO skb
4422  *
4423  * skb_gso_transport_seglen is used to determine the real size of the
4424  * individual segments, including Layer4 headers (TCP/UDP).
4425  *
4426  * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
4427  */
4428 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
4429 {
4430         const struct skb_shared_info *shinfo = skb_shinfo(skb);
4431         unsigned int thlen = 0;
4432 
4433         if (skb->encapsulation) {
4434                 thlen = skb_inner_transport_header(skb) -
4435                         skb_transport_header(skb);
4436 
4437                 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
4438                         thlen += inner_tcp_hdrlen(skb);
4439         } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
4440                 thlen = tcp_hdrlen(skb);
4441         } else if (unlikely(shinfo->gso_type & SKB_GSO_SCTP)) {
4442                 thlen = sizeof(struct sctphdr);
4443         }
4444         /* UFO sets gso_size to the size of the fragmentation
4445          * payload, i.e. the size of the L4 (UDP) header is already
4446          * accounted for.
4447          */
4448         return thlen + shinfo->gso_size;
4449 }
4450 EXPORT_SYMBOL_GPL(skb_gso_transport_seglen);
4451 
4452 /**
4453  * skb_gso_validate_mtu - Return in case such skb fits a given MTU
4454  *
4455  * @skb: GSO skb
4456  * @mtu: MTU to validate against
4457  *
4458  * skb_gso_validate_mtu validates if a given skb will fit a wanted MTU
4459  * once split.
4460  */
4461 bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu)
4462 {
4463         const struct skb_shared_info *shinfo = skb_shinfo(skb);
4464         const struct sk_buff *iter;
4465         unsigned int hlen;
4466 
4467         hlen = skb_gso_network_seglen(skb);
4468 
4469         if (shinfo->gso_size != GSO_BY_FRAGS)
4470                 return hlen <= mtu;
4471 
4472         /* Undo this so we can re-use header sizes */
4473         hlen -= GSO_BY_FRAGS;
4474 
4475         skb_walk_frags(skb, iter) {
4476                 if (hlen + skb_headlen(iter) > mtu)
4477                         return false;
4478         }
4479 
4480         return true;
4481 }
4482 EXPORT_SYMBOL_GPL(skb_gso_validate_mtu);
4483 
4484 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
4485 {
4486         if (skb_cow(skb, skb_headroom(skb)) < 0) {
4487                 kfree_skb(skb);
4488                 return NULL;
4489         }
4490 
4491         memmove(skb->data - ETH_HLEN, skb->data - skb->mac_len - VLAN_HLEN,
4492                 2 * ETH_ALEN);
4493         skb->mac_header += VLAN_HLEN;
4494         return skb;
4495 }
4496 
4497 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
4498 {
4499         struct vlan_hdr *vhdr;
4500         u16 vlan_tci;
4501 
4502         if (unlikely(skb_vlan_tag_present(skb))) {
4503                 /* vlan_tci is already set-up so leave this for another time */
4504                 return skb;
4505         }
4506 
4507         skb = skb_share_check(skb, GFP_ATOMIC);
4508         if (unlikely(!skb))
4509                 goto err_free;
4510 
4511         if (unlikely(!pskb_may_pull(skb, VLAN_HLEN)))
4512                 goto err_free;
4513 
4514         vhdr = (struct vlan_hdr *)skb->data;
4515         vlan_tci = ntohs(vhdr->h_vlan_TCI);
4516         __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
4517 
4518         skb_pull_rcsum(skb, VLAN_HLEN);
4519         vlan_set_encap_proto(skb, vhdr);
4520 
4521         skb = skb_reorder_vlan_header(skb);
4522         if (unlikely(!skb))
4523                 goto err_free;
4524 
4525         skb_reset_network_header(skb);
4526         skb_reset_transport_header(skb);
4527         skb_reset_mac_len(skb);
4528 
4529         return skb;
4530 
4531 err_free:
4532         kfree_skb(skb);
4533         return NULL;
4534 }
4535 EXPORT_SYMBOL(skb_vlan_untag);
4536 
4537 int skb_ensure_writable(struct sk_buff *skb, int write_len)
4538 {
4539         if (!pskb_may_pull(skb, write_len))
4540                 return -ENOMEM;
4541 
4542         if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
4543                 return 0;
4544 
4545         return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4546 }
4547 EXPORT_SYMBOL(skb_ensure_writable);
4548 
4549 /* remove VLAN header from packet and update csum accordingly.
4550  * expects a non skb_vlan_tag_present skb with a vlan tag payload
4551  */
4552 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
4553 {
4554         struct vlan_hdr *vhdr;
4555         int offset = skb->data - skb_mac_header(skb);
4556         int err;
4557 
4558         if (WARN_ONCE(offset,
4559                       "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
4560                       offset)) {
4561                 return -EINVAL;
4562         }
4563 
4564         err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
4565         if (unlikely(err))
4566                 return err;
4567 
4568         skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
4569 
4570         vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
4571         *vlan_tci = ntohs(vhdr->h_vlan_TCI);
4572 
4573         memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
4574         __skb_pull(skb, VLAN_HLEN);
4575 
4576         vlan_set_encap_proto(skb, vhdr);
4577         skb->mac_header += VLAN_HLEN;
4578 
4579         if (skb_network_offset(skb) < ETH_HLEN)
4580                 skb_set_network_header(skb, ETH_HLEN);
4581 
4582         skb_reset_mac_len(skb);
4583 
4584         return err;
4585 }
4586 EXPORT_SYMBOL(__skb_vlan_pop);
4587 
4588 /* Pop a vlan tag either from hwaccel or from payload.
4589  * Expects skb->data at mac header.
4590  */
4591 int skb_vlan_pop(struct sk_buff *skb)
4592 {
4593         u16 vlan_tci;
4594         __be16 vlan_proto;
4595         int err;
4596 
4597         if (likely(skb_vlan_tag_present(skb))) {
4598                 skb->vlan_tci = 0;
4599         } else {
4600                 if (unlikely(!eth_type_vlan(skb->protocol)))
4601                         return 0;
4602 
4603                 err = __skb_vlan_pop(skb, &vlan_tci);
4604                 if (err)
4605                         return err;
4606         }
4607         /* move next vlan tag to hw accel tag */
4608         if (likely(!eth_type_vlan(skb->protocol)))
4609                 return 0;
4610 
4611         vlan_proto = skb->protocol;
4612         err = __skb_vlan_pop(skb, &vlan_tci);
4613         if (unlikely(err))
4614                 return err;
4615 
4616         __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
4617         return 0;
4618 }
4619 EXPORT_SYMBOL(skb_vlan_pop);
4620 
4621 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
4622  * Expects skb->data at mac header.
4623  */
4624 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
4625 {
4626         if (skb_vlan_tag_present(skb)) {
4627                 int offset = skb->data - skb_mac_header(skb);
4628                 int err;
4629 
4630                 if (WARN_ONCE(offset,
4631                               "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
4632                               offset)) {
4633                         return -EINVAL;
4634                 }
4635 
4636                 err = __vlan_insert_tag(skb, skb->vlan_proto,
4637                                         skb_vlan_tag_get(skb));
4638                 if (err)
4639                         return err;
4640 
4641                 skb->protocol = skb->vlan_proto;
4642                 skb->mac_len += VLAN_HLEN;
4643 
4644                 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
4645         }
4646         __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
4647         return 0;
4648 }
4649 EXPORT_SYMBOL(skb_vlan_push);
4650 
4651 /**
4652  * alloc_skb_with_frags - allocate skb with page frags
4653  *
4654  * @header_len: size of linear part
4655  * @data_len: needed length in frags
4656  * @max_page_order: max page order desired.
4657  * @errcode: pointer to error code if any
4658  * @gfp_mask: allocation mask
4659  *
4660  * This can be used to allocate a paged skb, given a maximal order for frags.
4661  */
4662 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
4663                                      unsigned long data_len,
4664                                      int max_page_order,
4665                                      int *errcode,
4666                                      gfp_t gfp_mask)
4667 {
4668         int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
4669         unsigned long chunk;
4670         struct sk_buff *skb;
4671         struct page *page;
4672         gfp_t gfp_head;
4673         int i;
4674 
4675         *errcode = -EMSGSIZE;
4676         /* Note this test could be relaxed, if we succeed to allocate
4677          * high order pages...
4678          */
4679         if (npages > MAX_SKB_FRAGS)
4680                 return NULL;
4681 
4682         gfp_head = gfp_mask;
4683         if (gfp_head & __GFP_DIRECT_RECLAIM)
4684                 gfp_head |= __GFP_REPEAT;
4685 
4686         *errcode = -ENOBUFS;
4687         skb = alloc_skb(header_len, gfp_head);
4688         if (!skb)
4689                 return NULL;
4690 
4691         skb->truesize += npages << PAGE_SHIFT;
4692 
4693         for (i = 0; npages > 0; i++) {
4694                 int order = max_page_order;
4695 
4696                 while (order) {
4697                         if (npages >= 1 << order) {
4698                                 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
4699                                                    __GFP_COMP |
4700                                                    __GFP_NOWARN |
4701                                                    __GFP_NORETRY,
4702                                                    order);
4703                                 if (page)
4704                                         goto fill_page;
4705                                 /* Do not retry other high order allocations */
4706                                 order = 1;
4707                                 max_page_order = 0;
4708                         }
4709                         order--;
4710                 }
4711                 page = alloc_page(gfp_mask);
4712                 if (!page)
4713                         goto failure;
4714 fill_page:
4715                 chunk = min_t(unsigned long, data_len,
4716                               PAGE_SIZE << order);
4717                 skb_fill_page_desc(skb, i, page, 0, chunk);
4718                 data_len -= chunk;
4719                 npages -= 1 << order;
4720         }
4721         return skb;
4722 
4723 failure:
4724         kfree_skb(skb);
4725         return NULL;
4726 }
4727 EXPORT_SYMBOL(alloc_skb_with_frags);
4728 
4729 /* carve out the first off bytes from skb when off < headlen */
4730 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
4731                                     const int headlen, gfp_t gfp_mask)
4732 {
4733         int i;
4734         int size = skb_end_offset(skb);
4735         int new_hlen = headlen - off;
4736         u8 *data;
4737 
4738         size = SKB_DATA_ALIGN(size);
4739 
4740         if (skb_pfmemalloc(skb))
4741                 gfp_mask |= __GFP_MEMALLOC;
4742         data = kmalloc_reserve(size +
4743                                SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
4744                                gfp_mask, NUMA_NO_NODE, NULL);
4745         if (!data)
4746                 return -ENOMEM;
4747 
4748         size = SKB_WITH_OVERHEAD(ksize(data));
4749 
4750         /* Copy real data, and all frags */
4751         skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
4752         skb->len -= off;
4753 
4754         memcpy((struct skb_shared_info *)(data + size),
4755                skb_shinfo(skb),
4756                offsetof(struct skb_shared_info,
4757                         frags[skb_shinfo(skb)->nr_frags]));
4758         if (skb_cloned(skb)) {
4759                 /* drop the old head gracefully */
4760                 if (skb_orphan_frags(skb, gfp_mask)) {
4761                         kfree(data);
4762                         return -ENOMEM;
4763                 }
4764                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
4765                         skb_frag_ref(skb, i);
4766                 if (skb_has_frag_list(skb))
4767                         skb_clone_fraglist(skb);
4768                 skb_release_data(skb);
4769         } else {
4770                 /* we can reuse existing recount- all we did was
4771                  * relocate values
4772                  */
4773                 skb_free_head(skb);
4774         }
4775 
4776         skb->head = data;
4777         skb->data = data;
4778         skb->head_frag = 0;
4779 #ifdef NET_SKBUFF_DATA_USES_OFFSET
4780         skb->end = size;
4781 #else
4782         skb->end = skb->head + size;
4783 #endif
4784         skb_set_tail_pointer(skb, skb_headlen(skb));
4785         skb_headers_offset_update(skb, 0);
4786         skb->cloned = 0;
4787         skb->hdr_len = 0;
4788         skb->nohdr = 0;
4789         atomic_set(&skb_shinfo(skb)->dataref, 1);
4790 
4791         return 0;
4792 }
4793 
4794 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
4795 
4796 /* carve out the first eat bytes from skb's frag_list. May recurse into
4797  * pskb_carve()
4798  */
4799 static int pskb_carve_frag_list(struct sk_buff *skb,
4800                                 struct skb_shared_info *shinfo, int eat,
4801                                 gfp_t gfp_mask)
4802 {
4803         struct sk_buff *list = shinfo->frag_list;
4804         struct sk_buff *clone = NULL;
4805         struct sk_buff *insp = NULL;
4806 
4807         do {
4808                 if (!list) {
4809                         pr_err("Not enough bytes to eat. Want %d\n", eat);
4810                         return -EFAULT;
4811                 }
4812                 if (list->len <= eat) {
4813                         /* Eaten as whole. */
4814                         eat -= list->len;
4815                         list = list->next;
4816                         insp = list;
4817                 } else {
4818                         /* Eaten partially. */
4819                         if (skb_shared(list)) {
4820                                 clone = skb_clone(list, gfp_mask);
4821                                 if (!clone)
4822                                         return -ENOMEM;
4823                                 insp = list->next;
4824                                 list = clone;
4825                         } else {
4826                                 /* This may be pulled without problems. */
4827                                 insp = list;
4828                         }
4829                         if (pskb_carve(list, eat, gfp_mask) < 0) {
4830                                 kfree_skb(clone);
4831                                 return -ENOMEM;
4832                         }
4833                         break;
4834                 }
4835         } while (eat);
4836 
4837         /* Free pulled out fragments. */
4838         while ((list = shinfo->frag_list) != insp) {
4839                 shinfo->frag_list = list->next;
4840                 kfree_skb(list);
4841         }
4842         /* And insert new clone at head. */
4843         if (clone) {
4844                 clone->next = list;
4845                 shinfo->frag_list = clone;
4846         }
4847         return 0;
4848 }
4849 
4850 /* carve off first len bytes from skb. Split line (off) is in the
4851  * non-linear part of skb
4852  */
4853 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
4854                                        int pos, gfp_t gfp_mask)
4855 {
4856         int i, k = 0;
4857         int size = skb_end_offset(skb);
4858         u8 *data;
4859         const int nfrags = skb_shinfo(skb)->nr_frags;
4860         struct skb_shared_info *shinfo;
4861 
4862         size = SKB_DATA_ALIGN(size);
4863 
4864         if (skb_pfmemalloc(skb))
4865                 gfp_mask |= __GFP_MEMALLOC;
4866         data = kmalloc_reserve(size +
4867                                SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
4868                                gfp_mask, NUMA_NO_NODE, NULL);
4869         if (!data)
4870                 return -ENOMEM;
4871 
4872         size = SKB_WITH_OVERHEAD(ksize(data));
4873 
4874         memcpy((struct skb_shared_info *)(data + size),
4875                skb_shinfo(skb), offsetof(struct skb_shared_info,
4876                                          frags[skb_shinfo(skb)->nr_frags]));
4877         if (skb_orphan_frags(skb, gfp_mask)) {
4878                 kfree(data);
4879                 return -ENOMEM;
4880         }
4881         shinfo = (struct skb_shared_info *)(data + size);
4882         for (i = 0; i < nfrags; i++) {
4883                 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
4884 
4885                 if (pos + fsize > off) {
4886                         shinfo->frags[k] = skb_shinfo(skb)->frags[i];
4887 
4888                         if (pos < off) {
4889                                 /* Split frag.
4890                                  * We have two variants in this case:
4891                                  * 1. Move all the frag to the second
4892                                  *    part, if it is possible. F.e.
4893                                  *    this approach is mandatory for TUX,
4894                                  *    where splitting is expensive.
4895                                  * 2. Split is accurately. We make this.
4896                                  */
4897                                 shinfo->frags[0].page_offset += off - pos;
4898                                 skb_frag_size_sub(&shinfo->frags[0], off - pos);
4899                         }
4900                         skb_frag_ref(skb, i);
4901                         k++;
4902                 }
4903                 pos += fsize;
4904         }
4905         shinfo->nr_frags = k;
4906         if (skb_has_frag_list(skb))
4907                 skb_clone_fraglist(skb);
4908 
4909         if (k == 0) {
4910                 /* split line is in frag list */
4911                 pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask);
4912         }
4913         skb_release_data(skb);
4914 
4915         skb->head = data;
4916         skb->head_frag = 0;
4917         skb->data = data;
4918 #ifdef NET_SKBUFF_DATA_USES_OFFSET
4919         skb->end = size;
4920 #else
4921         skb->end = skb->head + size;
4922 #endif
4923         skb_reset_tail_pointer(skb);
4924         skb_headers_offset_update(skb, 0);
4925         skb->cloned   = 0;
4926         skb->hdr_len  = 0;
4927         skb->nohdr    = 0;
4928         skb->len -= off;
4929         skb->data_len = skb->len;
4930         atomic_set(&skb_shinfo(skb)->dataref, 1);
4931         return 0;
4932 }
4933 
4934 /* remove len bytes from the beginning of the skb */
4935 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
4936 {
4937         int headlen = skb_headlen(skb);
4938 
4939         if (len < headlen)
4940                 return pskb_carve_inside_header(skb, len, headlen, gfp);
4941         else
4942                 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
4943 }
4944 
4945 /* Extract to_copy bytes starting at off from skb, and return this in
4946  * a new skb
4947  */
4948 struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
4949                              int to_copy, gfp_t gfp)
4950 {
4951         struct sk_buff  *clone = skb_clone(skb, gfp);
4952 
4953         if (!clone)
4954                 return NULL;
4955 
4956         if (pskb_carve(clone, off, gfp) < 0 ||
4957             pskb_trim(clone, to_copy)) {
4958                 kfree_skb(clone);
4959                 return NULL;
4960         }
4961         return clone;
4962 }
4963 EXPORT_SYMBOL(pskb_extract);
4964 
4965 /**
4966  * skb_condense - try to get rid of fragments/frag_list if possible
4967  * @skb: buffer
4968  *
4969  * Can be used to save memory before skb is added to a busy queue.
4970  * If packet has bytes in frags and enough tail room in skb->head,
4971  * pull all of them, so that we can free the frags right now and adjust
4972  * truesize.
4973  * Notes:
4974  *      We do not reallocate skb->head thus can not fail.
4975  *      Caller must re-evaluate skb->truesize if needed.
4976  */
4977 void skb_condense(struct sk_buff *skb)
4978 {
4979         if (skb->data_len) {
4980                 if (skb->data_len > skb->end - skb->tail ||
4981                     skb_cloned(skb))
4982                         return;
4983 
4984                 /* Nice, we can free page frag(s) right now */
4985                 __pskb_pull_tail(skb, skb->data_len);
4986         }
4987         /* At this point, skb->truesize might be over estimated,
4988          * because skb had a fragment, and fragments do not tell
4989          * their truesize.
4990          * When we pulled its content into skb->head, fragment
4991          * was freed, but __pskb_pull_tail() could not possibly
4992          * adjust skb->truesize, not knowing the frag truesize.
4993          */
4994         skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4995 }
4996 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp