Version:  2.0.40 2.2.26 2.4.37 3.5 3.6 3.7 3.8 3.9 3.10 3.11 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1

Linux/net/core/skbuff.c

  1 /*
  2  *      Routines having to do with the 'struct sk_buff' memory handlers.
  3  *
  4  *      Authors:        Alan Cox <alan@lxorguk.ukuu.org.uk>
  5  *                      Florian La Roche <rzsfl@rz.uni-sb.de>
  6  *
  7  *      Fixes:
  8  *              Alan Cox        :       Fixed the worst of the load
  9  *                                      balancer bugs.
 10  *              Dave Platt      :       Interrupt stacking fix.
 11  *      Richard Kooijman        :       Timestamp fixes.
 12  *              Alan Cox        :       Changed buffer format.
 13  *              Alan Cox        :       destructor hook for AF_UNIX etc.
 14  *              Linus Torvalds  :       Better skb_clone.
 15  *              Alan Cox        :       Added skb_copy.
 16  *              Alan Cox        :       Added all the changed routines Linus
 17  *                                      only put in the headers
 18  *              Ray VanTassle   :       Fixed --skb->lock in free
 19  *              Alan Cox        :       skb_copy copy arp field
 20  *              Andi Kleen      :       slabified it.
 21  *              Robert Olsson   :       Removed skb_head_pool
 22  *
 23  *      NOTE:
 24  *              The __skb_ routines should be called with interrupts
 25  *      disabled, or you better be *real* sure that the operation is atomic
 26  *      with respect to whatever list is being frobbed (e.g. via lock_sock()
 27  *      or via disabling bottom half handlers, etc).
 28  *
 29  *      This program is free software; you can redistribute it and/or
 30  *      modify it under the terms of the GNU General Public License
 31  *      as published by the Free Software Foundation; either version
 32  *      2 of the License, or (at your option) any later version.
 33  */
 34 
 35 /*
 36  *      The functions in this file will not compile correctly with gcc 2.4.x
 37  */
 38 
 39 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 40 
 41 #include <linux/module.h>
 42 #include <linux/types.h>
 43 #include <linux/kernel.h>
 44 #include <linux/kmemcheck.h>
 45 #include <linux/mm.h>
 46 #include <linux/interrupt.h>
 47 #include <linux/in.h>
 48 #include <linux/inet.h>
 49 #include <linux/slab.h>
 50 #include <linux/tcp.h>
 51 #include <linux/udp.h>
 52 #include <linux/netdevice.h>
 53 #ifdef CONFIG_NET_CLS_ACT
 54 #include <net/pkt_sched.h>
 55 #endif
 56 #include <linux/string.h>
 57 #include <linux/skbuff.h>
 58 #include <linux/splice.h>
 59 #include <linux/cache.h>
 60 #include <linux/rtnetlink.h>
 61 #include <linux/init.h>
 62 #include <linux/scatterlist.h>
 63 #include <linux/errqueue.h>
 64 #include <linux/prefetch.h>
 65 #include <linux/if_vlan.h>
 66 
 67 #include <net/protocol.h>
 68 #include <net/dst.h>
 69 #include <net/sock.h>
 70 #include <net/checksum.h>
 71 #include <net/ip6_checksum.h>
 72 #include <net/xfrm.h>
 73 
 74 #include <asm/uaccess.h>
 75 #include <trace/events/skb.h>
 76 #include <linux/highmem.h>
 77 #include <linux/capability.h>
 78 #include <linux/user_namespace.h>
 79 
 80 struct kmem_cache *skbuff_head_cache __read_mostly;
 81 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
 82 
 83 /**
 84  *      skb_panic - private function for out-of-line support
 85  *      @skb:   buffer
 86  *      @sz:    size
 87  *      @addr:  address
 88  *      @msg:   skb_over_panic or skb_under_panic
 89  *
 90  *      Out-of-line support for skb_put() and skb_push().
 91  *      Called via the wrapper skb_over_panic() or skb_under_panic().
 92  *      Keep out of line to prevent kernel bloat.
 93  *      __builtin_return_address is not used because it is not always reliable.
 94  */
 95 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
 96                       const char msg[])
 97 {
 98         pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
 99                  msg, addr, skb->len, sz, skb->head, skb->data,
100                  (unsigned long)skb->tail, (unsigned long)skb->end,
101                  skb->dev ? skb->dev->name : "<NULL>");
102         BUG();
103 }
104 
105 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
106 {
107         skb_panic(skb, sz, addr, __func__);
108 }
109 
110 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
111 {
112         skb_panic(skb, sz, addr, __func__);
113 }
114 
115 /*
116  * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
117  * the caller if emergency pfmemalloc reserves are being used. If it is and
118  * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
119  * may be used. Otherwise, the packet data may be discarded until enough
120  * memory is free
121  */
122 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
123          __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
124 
125 static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
126                                unsigned long ip, bool *pfmemalloc)
127 {
128         void *obj;
129         bool ret_pfmemalloc = false;
130 
131         /*
132          * Try a regular allocation, when that fails and we're not entitled
133          * to the reserves, fail.
134          */
135         obj = kmalloc_node_track_caller(size,
136                                         flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
137                                         node);
138         if (obj || !(gfp_pfmemalloc_allowed(flags)))
139                 goto out;
140 
141         /* Try again but now we are using pfmemalloc reserves */
142         ret_pfmemalloc = true;
143         obj = kmalloc_node_track_caller(size, flags, node);
144 
145 out:
146         if (pfmemalloc)
147                 *pfmemalloc = ret_pfmemalloc;
148 
149         return obj;
150 }
151 
152 /*      Allocate a new skbuff. We do this ourselves so we can fill in a few
153  *      'private' fields and also do memory statistics to find all the
154  *      [BEEP] leaks.
155  *
156  */
157 
158 struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node)
159 {
160         struct sk_buff *skb;
161 
162         /* Get the HEAD */
163         skb = kmem_cache_alloc_node(skbuff_head_cache,
164                                     gfp_mask & ~__GFP_DMA, node);
165         if (!skb)
166                 goto out;
167 
168         /*
169          * Only clear those fields we need to clear, not those that we will
170          * actually initialise below. Hence, don't put any more fields after
171          * the tail pointer in struct sk_buff!
172          */
173         memset(skb, 0, offsetof(struct sk_buff, tail));
174         skb->head = NULL;
175         skb->truesize = sizeof(struct sk_buff);
176         atomic_set(&skb->users, 1);
177 
178         skb->mac_header = (typeof(skb->mac_header))~0U;
179 out:
180         return skb;
181 }
182 
183 /**
184  *      __alloc_skb     -       allocate a network buffer
185  *      @size: size to allocate
186  *      @gfp_mask: allocation mask
187  *      @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
188  *              instead of head cache and allocate a cloned (child) skb.
189  *              If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
190  *              allocations in case the data is required for writeback
191  *      @node: numa node to allocate memory on
192  *
193  *      Allocate a new &sk_buff. The returned buffer has no headroom and a
194  *      tail room of at least size bytes. The object has a reference count
195  *      of one. The return is the buffer. On a failure the return is %NULL.
196  *
197  *      Buffers may only be allocated from interrupts using a @gfp_mask of
198  *      %GFP_ATOMIC.
199  */
200 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
201                             int flags, int node)
202 {
203         struct kmem_cache *cache;
204         struct skb_shared_info *shinfo;
205         struct sk_buff *skb;
206         u8 *data;
207         bool pfmemalloc;
208 
209         cache = (flags & SKB_ALLOC_FCLONE)
210                 ? skbuff_fclone_cache : skbuff_head_cache;
211 
212         if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
213                 gfp_mask |= __GFP_MEMALLOC;
214 
215         /* Get the HEAD */
216         skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
217         if (!skb)
218                 goto out;
219         prefetchw(skb);
220 
221         /* We do our best to align skb_shared_info on a separate cache
222          * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
223          * aligned memory blocks, unless SLUB/SLAB debug is enabled.
224          * Both skb->head and skb_shared_info are cache line aligned.
225          */
226         size = SKB_DATA_ALIGN(size);
227         size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
228         data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
229         if (!data)
230                 goto nodata;
231         /* kmalloc(size) might give us more room than requested.
232          * Put skb_shared_info exactly at the end of allocated zone,
233          * to allow max possible filling before reallocation.
234          */
235         size = SKB_WITH_OVERHEAD(ksize(data));
236         prefetchw(data + size);
237 
238         /*
239          * Only clear those fields we need to clear, not those that we will
240          * actually initialise below. Hence, don't put any more fields after
241          * the tail pointer in struct sk_buff!
242          */
243         memset(skb, 0, offsetof(struct sk_buff, tail));
244         /* Account for allocated memory : skb + skb->head */
245         skb->truesize = SKB_TRUESIZE(size);
246         skb->pfmemalloc = pfmemalloc;
247         atomic_set(&skb->users, 1);
248         skb->head = data;
249         skb->data = data;
250         skb_reset_tail_pointer(skb);
251         skb->end = skb->tail + size;
252         skb->mac_header = (typeof(skb->mac_header))~0U;
253         skb->transport_header = (typeof(skb->transport_header))~0U;
254 
255         /* make sure we initialize shinfo sequentially */
256         shinfo = skb_shinfo(skb);
257         memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
258         atomic_set(&shinfo->dataref, 1);
259         kmemcheck_annotate_variable(shinfo->destructor_arg);
260 
261         if (flags & SKB_ALLOC_FCLONE) {
262                 struct sk_buff_fclones *fclones;
263 
264                 fclones = container_of(skb, struct sk_buff_fclones, skb1);
265 
266                 kmemcheck_annotate_bitfield(&fclones->skb2, flags1);
267                 skb->fclone = SKB_FCLONE_ORIG;
268                 atomic_set(&fclones->fclone_ref, 1);
269 
270                 fclones->skb2.fclone = SKB_FCLONE_CLONE;
271                 fclones->skb2.pfmemalloc = pfmemalloc;
272         }
273 out:
274         return skb;
275 nodata:
276         kmem_cache_free(cache, skb);
277         skb = NULL;
278         goto out;
279 }
280 EXPORT_SYMBOL(__alloc_skb);
281 
282 /**
283  * __build_skb - build a network buffer
284  * @data: data buffer provided by caller
285  * @frag_size: size of data, or 0 if head was kmalloced
286  *
287  * Allocate a new &sk_buff. Caller provides space holding head and
288  * skb_shared_info. @data must have been allocated by kmalloc() only if
289  * @frag_size is 0, otherwise data should come from the page allocator
290  *  or vmalloc()
291  * The return is the new skb buffer.
292  * On a failure the return is %NULL, and @data is not freed.
293  * Notes :
294  *  Before IO, driver allocates only data buffer where NIC put incoming frame
295  *  Driver should add room at head (NET_SKB_PAD) and
296  *  MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
297  *  After IO, driver calls build_skb(), to allocate sk_buff and populate it
298  *  before giving packet to stack.
299  *  RX rings only contains data buffers, not full skbs.
300  */
301 struct sk_buff *__build_skb(void *data, unsigned int frag_size)
302 {
303         struct skb_shared_info *shinfo;
304         struct sk_buff *skb;
305         unsigned int size = frag_size ? : ksize(data);
306 
307         skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
308         if (!skb)
309                 return NULL;
310 
311         size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
312 
313         memset(skb, 0, offsetof(struct sk_buff, tail));
314         skb->truesize = SKB_TRUESIZE(size);
315         atomic_set(&skb->users, 1);
316         skb->head = data;
317         skb->data = data;
318         skb_reset_tail_pointer(skb);
319         skb->end = skb->tail + size;
320         skb->mac_header = (typeof(skb->mac_header))~0U;
321         skb->transport_header = (typeof(skb->transport_header))~0U;
322 
323         /* make sure we initialize shinfo sequentially */
324         shinfo = skb_shinfo(skb);
325         memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
326         atomic_set(&shinfo->dataref, 1);
327         kmemcheck_annotate_variable(shinfo->destructor_arg);
328 
329         return skb;
330 }
331 
332 /* build_skb() is wrapper over __build_skb(), that specifically
333  * takes care of skb->head and skb->pfmemalloc
334  * This means that if @frag_size is not zero, then @data must be backed
335  * by a page fragment, not kmalloc() or vmalloc()
336  */
337 struct sk_buff *build_skb(void *data, unsigned int frag_size)
338 {
339         struct sk_buff *skb = __build_skb(data, frag_size);
340 
341         if (skb && frag_size) {
342                 skb->head_frag = 1;
343                 if (virt_to_head_page(data)->pfmemalloc)
344                         skb->pfmemalloc = 1;
345         }
346         return skb;
347 }
348 EXPORT_SYMBOL(build_skb);
349 
350 struct netdev_alloc_cache {
351         struct page_frag        frag;
352         /* we maintain a pagecount bias, so that we dont dirty cache line
353          * containing page->_count every time we allocate a fragment.
354          */
355         unsigned int            pagecnt_bias;
356 };
357 static DEFINE_PER_CPU(struct netdev_alloc_cache, netdev_alloc_cache);
358 static DEFINE_PER_CPU(struct netdev_alloc_cache, napi_alloc_cache);
359 
360 static struct page *__page_frag_refill(struct netdev_alloc_cache *nc,
361                                        gfp_t gfp_mask)
362 {
363         const unsigned int order = NETDEV_FRAG_PAGE_MAX_ORDER;
364         struct page *page = NULL;
365         gfp_t gfp = gfp_mask;
366 
367         if (order) {
368                 gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
369                             __GFP_NOMEMALLOC;
370                 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
371                 nc->frag.size = PAGE_SIZE << (page ? order : 0);
372         }
373 
374         if (unlikely(!page))
375                 page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
376 
377         nc->frag.page = page;
378 
379         return page;
380 }
381 
382 static void *__alloc_page_frag(struct netdev_alloc_cache __percpu *cache,
383                                unsigned int fragsz, gfp_t gfp_mask)
384 {
385         struct netdev_alloc_cache *nc = this_cpu_ptr(cache);
386         struct page *page = nc->frag.page;
387         unsigned int size;
388         int offset;
389 
390         if (unlikely(!page)) {
391 refill:
392                 page = __page_frag_refill(nc, gfp_mask);
393                 if (!page)
394                         return NULL;
395 
396                 /* if size can vary use frag.size else just use PAGE_SIZE */
397                 size = NETDEV_FRAG_PAGE_MAX_ORDER ? nc->frag.size : PAGE_SIZE;
398 
399                 /* Even if we own the page, we do not use atomic_set().
400                  * This would break get_page_unless_zero() users.
401                  */
402                 atomic_add(size - 1, &page->_count);
403 
404                 /* reset page count bias and offset to start of new frag */
405                 nc->pagecnt_bias = size;
406                 nc->frag.offset = size;
407         }
408 
409         offset = nc->frag.offset - fragsz;
410         if (unlikely(offset < 0)) {
411                 if (!atomic_sub_and_test(nc->pagecnt_bias, &page->_count))
412                         goto refill;
413 
414                 /* if size can vary use frag.size else just use PAGE_SIZE */
415                 size = NETDEV_FRAG_PAGE_MAX_ORDER ? nc->frag.size : PAGE_SIZE;
416 
417                 /* OK, page count is 0, we can safely set it */
418                 atomic_set(&page->_count, size);
419 
420                 /* reset page count bias and offset to start of new frag */
421                 nc->pagecnt_bias = size;
422                 offset = size - fragsz;
423         }
424 
425         nc->pagecnt_bias--;
426         nc->frag.offset = offset;
427 
428         return page_address(page) + offset;
429 }
430 
431 static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
432 {
433         unsigned long flags;
434         void *data;
435 
436         local_irq_save(flags);
437         data = __alloc_page_frag(&netdev_alloc_cache, fragsz, gfp_mask);
438         local_irq_restore(flags);
439         return data;
440 }
441 
442 /**
443  * netdev_alloc_frag - allocate a page fragment
444  * @fragsz: fragment size
445  *
446  * Allocates a frag from a page for receive buffer.
447  * Uses GFP_ATOMIC allocations.
448  */
449 void *netdev_alloc_frag(unsigned int fragsz)
450 {
451         return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
452 }
453 EXPORT_SYMBOL(netdev_alloc_frag);
454 
455 static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
456 {
457         return __alloc_page_frag(&napi_alloc_cache, fragsz, gfp_mask);
458 }
459 
460 void *napi_alloc_frag(unsigned int fragsz)
461 {
462         return __napi_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
463 }
464 EXPORT_SYMBOL(napi_alloc_frag);
465 
466 /**
467  *      __alloc_rx_skb - allocate an skbuff for rx
468  *      @length: length to allocate
469  *      @gfp_mask: get_free_pages mask, passed to alloc_skb
470  *      @flags: If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
471  *              allocations in case we have to fallback to __alloc_skb()
472  *              If SKB_ALLOC_NAPI is set, page fragment will be allocated
473  *              from napi_cache instead of netdev_cache.
474  *
475  *      Allocate a new &sk_buff and assign it a usage count of one. The
476  *      buffer has unspecified headroom built in. Users should allocate
477  *      the headroom they think they need without accounting for the
478  *      built in space. The built in space is used for optimisations.
479  *
480  *      %NULL is returned if there is no free memory.
481  */
482 static struct sk_buff *__alloc_rx_skb(unsigned int length, gfp_t gfp_mask,
483                                       int flags)
484 {
485         struct sk_buff *skb = NULL;
486         unsigned int fragsz = SKB_DATA_ALIGN(length) +
487                               SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
488 
489         if (fragsz <= PAGE_SIZE && !(gfp_mask & (__GFP_WAIT | GFP_DMA))) {
490                 void *data;
491 
492                 if (sk_memalloc_socks())
493                         gfp_mask |= __GFP_MEMALLOC;
494 
495                 data = (flags & SKB_ALLOC_NAPI) ?
496                         __napi_alloc_frag(fragsz, gfp_mask) :
497                         __netdev_alloc_frag(fragsz, gfp_mask);
498 
499                 if (likely(data)) {
500                         skb = build_skb(data, fragsz);
501                         if (unlikely(!skb))
502                                 put_page(virt_to_head_page(data));
503                 }
504         } else {
505                 skb = __alloc_skb(length, gfp_mask,
506                                   SKB_ALLOC_RX, NUMA_NO_NODE);
507         }
508         return skb;
509 }
510 
511 /**
512  *      __netdev_alloc_skb - allocate an skbuff for rx on a specific device
513  *      @dev: network device to receive on
514  *      @length: length to allocate
515  *      @gfp_mask: get_free_pages mask, passed to alloc_skb
516  *
517  *      Allocate a new &sk_buff and assign it a usage count of one. The
518  *      buffer has NET_SKB_PAD headroom built in. Users should allocate
519  *      the headroom they think they need without accounting for the
520  *      built in space. The built in space is used for optimisations.
521  *
522  *      %NULL is returned if there is no free memory.
523  */
524 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
525                                    unsigned int length, gfp_t gfp_mask)
526 {
527         struct sk_buff *skb;
528 
529         length += NET_SKB_PAD;
530         skb = __alloc_rx_skb(length, gfp_mask, 0);
531 
532         if (likely(skb)) {
533                 skb_reserve(skb, NET_SKB_PAD);
534                 skb->dev = dev;
535         }
536 
537         return skb;
538 }
539 EXPORT_SYMBOL(__netdev_alloc_skb);
540 
541 /**
542  *      __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
543  *      @napi: napi instance this buffer was allocated for
544  *      @length: length to allocate
545  *      @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
546  *
547  *      Allocate a new sk_buff for use in NAPI receive.  This buffer will
548  *      attempt to allocate the head from a special reserved region used
549  *      only for NAPI Rx allocation.  By doing this we can save several
550  *      CPU cycles by avoiding having to disable and re-enable IRQs.
551  *
552  *      %NULL is returned if there is no free memory.
553  */
554 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
555                                  unsigned int length, gfp_t gfp_mask)
556 {
557         struct sk_buff *skb;
558 
559         length += NET_SKB_PAD + NET_IP_ALIGN;
560         skb = __alloc_rx_skb(length, gfp_mask, SKB_ALLOC_NAPI);
561 
562         if (likely(skb)) {
563                 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
564                 skb->dev = napi->dev;
565         }
566 
567         return skb;
568 }
569 EXPORT_SYMBOL(__napi_alloc_skb);
570 
571 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
572                      int size, unsigned int truesize)
573 {
574         skb_fill_page_desc(skb, i, page, off, size);
575         skb->len += size;
576         skb->data_len += size;
577         skb->truesize += truesize;
578 }
579 EXPORT_SYMBOL(skb_add_rx_frag);
580 
581 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
582                           unsigned int truesize)
583 {
584         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
585 
586         skb_frag_size_add(frag, size);
587         skb->len += size;
588         skb->data_len += size;
589         skb->truesize += truesize;
590 }
591 EXPORT_SYMBOL(skb_coalesce_rx_frag);
592 
593 static void skb_drop_list(struct sk_buff **listp)
594 {
595         kfree_skb_list(*listp);
596         *listp = NULL;
597 }
598 
599 static inline void skb_drop_fraglist(struct sk_buff *skb)
600 {
601         skb_drop_list(&skb_shinfo(skb)->frag_list);
602 }
603 
604 static void skb_clone_fraglist(struct sk_buff *skb)
605 {
606         struct sk_buff *list;
607 
608         skb_walk_frags(skb, list)
609                 skb_get(list);
610 }
611 
612 static void skb_free_head(struct sk_buff *skb)
613 {
614         if (skb->head_frag)
615                 put_page(virt_to_head_page(skb->head));
616         else
617                 kfree(skb->head);
618 }
619 
620 static void skb_release_data(struct sk_buff *skb)
621 {
622         struct skb_shared_info *shinfo = skb_shinfo(skb);
623         int i;
624 
625         if (skb->cloned &&
626             atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
627                               &shinfo->dataref))
628                 return;
629 
630         for (i = 0; i < shinfo->nr_frags; i++)
631                 __skb_frag_unref(&shinfo->frags[i]);
632 
633         /*
634          * If skb buf is from userspace, we need to notify the caller
635          * the lower device DMA has done;
636          */
637         if (shinfo->tx_flags & SKBTX_DEV_ZEROCOPY) {
638                 struct ubuf_info *uarg;
639 
640                 uarg = shinfo->destructor_arg;
641                 if (uarg->callback)
642                         uarg->callback(uarg, true);
643         }
644 
645         if (shinfo->frag_list)
646                 kfree_skb_list(shinfo->frag_list);
647 
648         skb_free_head(skb);
649 }
650 
651 /*
652  *      Free an skbuff by memory without cleaning the state.
653  */
654 static void kfree_skbmem(struct sk_buff *skb)
655 {
656         struct sk_buff_fclones *fclones;
657 
658         switch (skb->fclone) {
659         case SKB_FCLONE_UNAVAILABLE:
660                 kmem_cache_free(skbuff_head_cache, skb);
661                 return;
662 
663         case SKB_FCLONE_ORIG:
664                 fclones = container_of(skb, struct sk_buff_fclones, skb1);
665 
666                 /* We usually free the clone (TX completion) before original skb
667                  * This test would have no chance to be true for the clone,
668                  * while here, branch prediction will be good.
669                  */
670                 if (atomic_read(&fclones->fclone_ref) == 1)
671                         goto fastpath;
672                 break;
673 
674         default: /* SKB_FCLONE_CLONE */
675                 fclones = container_of(skb, struct sk_buff_fclones, skb2);
676                 break;
677         }
678         if (!atomic_dec_and_test(&fclones->fclone_ref))
679                 return;
680 fastpath:
681         kmem_cache_free(skbuff_fclone_cache, fclones);
682 }
683 
684 static void skb_release_head_state(struct sk_buff *skb)
685 {
686         skb_dst_drop(skb);
687 #ifdef CONFIG_XFRM
688         secpath_put(skb->sp);
689 #endif
690         if (skb->destructor) {
691                 WARN_ON(in_irq());
692                 skb->destructor(skb);
693         }
694 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
695         nf_conntrack_put(skb->nfct);
696 #endif
697 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
698         nf_bridge_put(skb->nf_bridge);
699 #endif
700 }
701 
702 /* Free everything but the sk_buff shell. */
703 static void skb_release_all(struct sk_buff *skb)
704 {
705         skb_release_head_state(skb);
706         if (likely(skb->head))
707                 skb_release_data(skb);
708 }
709 
710 /**
711  *      __kfree_skb - private function
712  *      @skb: buffer
713  *
714  *      Free an sk_buff. Release anything attached to the buffer.
715  *      Clean the state. This is an internal helper function. Users should
716  *      always call kfree_skb
717  */
718 
719 void __kfree_skb(struct sk_buff *skb)
720 {
721         skb_release_all(skb);
722         kfree_skbmem(skb);
723 }
724 EXPORT_SYMBOL(__kfree_skb);
725 
726 /**
727  *      kfree_skb - free an sk_buff
728  *      @skb: buffer to free
729  *
730  *      Drop a reference to the buffer and free it if the usage count has
731  *      hit zero.
732  */
733 void kfree_skb(struct sk_buff *skb)
734 {
735         if (unlikely(!skb))
736                 return;
737         if (likely(atomic_read(&skb->users) == 1))
738                 smp_rmb();
739         else if (likely(!atomic_dec_and_test(&skb->users)))
740                 return;
741         trace_kfree_skb(skb, __builtin_return_address(0));
742         __kfree_skb(skb);
743 }
744 EXPORT_SYMBOL(kfree_skb);
745 
746 void kfree_skb_list(struct sk_buff *segs)
747 {
748         while (segs) {
749                 struct sk_buff *next = segs->next;
750 
751                 kfree_skb(segs);
752                 segs = next;
753         }
754 }
755 EXPORT_SYMBOL(kfree_skb_list);
756 
757 /**
758  *      skb_tx_error - report an sk_buff xmit error
759  *      @skb: buffer that triggered an error
760  *
761  *      Report xmit error if a device callback is tracking this skb.
762  *      skb must be freed afterwards.
763  */
764 void skb_tx_error(struct sk_buff *skb)
765 {
766         if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
767                 struct ubuf_info *uarg;
768 
769                 uarg = skb_shinfo(skb)->destructor_arg;
770                 if (uarg->callback)
771                         uarg->callback(uarg, false);
772                 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
773         }
774 }
775 EXPORT_SYMBOL(skb_tx_error);
776 
777 /**
778  *      consume_skb - free an skbuff
779  *      @skb: buffer to free
780  *
781  *      Drop a ref to the buffer and free it if the usage count has hit zero
782  *      Functions identically to kfree_skb, but kfree_skb assumes that the frame
783  *      is being dropped after a failure and notes that
784  */
785 void consume_skb(struct sk_buff *skb)
786 {
787         if (unlikely(!skb))
788                 return;
789         if (likely(atomic_read(&skb->users) == 1))
790                 smp_rmb();
791         else if (likely(!atomic_dec_and_test(&skb->users)))
792                 return;
793         trace_consume_skb(skb);
794         __kfree_skb(skb);
795 }
796 EXPORT_SYMBOL(consume_skb);
797 
798 /* Make sure a field is enclosed inside headers_start/headers_end section */
799 #define CHECK_SKB_FIELD(field) \
800         BUILD_BUG_ON(offsetof(struct sk_buff, field) <          \
801                      offsetof(struct sk_buff, headers_start));  \
802         BUILD_BUG_ON(offsetof(struct sk_buff, field) >          \
803                      offsetof(struct sk_buff, headers_end));    \
804 
805 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
806 {
807         new->tstamp             = old->tstamp;
808         /* We do not copy old->sk */
809         new->dev                = old->dev;
810         memcpy(new->cb, old->cb, sizeof(old->cb));
811         skb_dst_copy(new, old);
812 #ifdef CONFIG_XFRM
813         new->sp                 = secpath_get(old->sp);
814 #endif
815         __nf_copy(new, old, false);
816 
817         /* Note : this field could be in headers_start/headers_end section
818          * It is not yet because we do not want to have a 16 bit hole
819          */
820         new->queue_mapping = old->queue_mapping;
821 
822         memcpy(&new->headers_start, &old->headers_start,
823                offsetof(struct sk_buff, headers_end) -
824                offsetof(struct sk_buff, headers_start));
825         CHECK_SKB_FIELD(protocol);
826         CHECK_SKB_FIELD(csum);
827         CHECK_SKB_FIELD(hash);
828         CHECK_SKB_FIELD(priority);
829         CHECK_SKB_FIELD(skb_iif);
830         CHECK_SKB_FIELD(vlan_proto);
831         CHECK_SKB_FIELD(vlan_tci);
832         CHECK_SKB_FIELD(transport_header);
833         CHECK_SKB_FIELD(network_header);
834         CHECK_SKB_FIELD(mac_header);
835         CHECK_SKB_FIELD(inner_protocol);
836         CHECK_SKB_FIELD(inner_transport_header);
837         CHECK_SKB_FIELD(inner_network_header);
838         CHECK_SKB_FIELD(inner_mac_header);
839         CHECK_SKB_FIELD(mark);
840 #ifdef CONFIG_NETWORK_SECMARK
841         CHECK_SKB_FIELD(secmark);
842 #endif
843 #ifdef CONFIG_NET_RX_BUSY_POLL
844         CHECK_SKB_FIELD(napi_id);
845 #endif
846 #ifdef CONFIG_XPS
847         CHECK_SKB_FIELD(sender_cpu);
848 #endif
849 #ifdef CONFIG_NET_SCHED
850         CHECK_SKB_FIELD(tc_index);
851 #ifdef CONFIG_NET_CLS_ACT
852         CHECK_SKB_FIELD(tc_verd);
853 #endif
854 #endif
855 
856 }
857 
858 /*
859  * You should not add any new code to this function.  Add it to
860  * __copy_skb_header above instead.
861  */
862 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
863 {
864 #define C(x) n->x = skb->x
865 
866         n->next = n->prev = NULL;
867         n->sk = NULL;
868         __copy_skb_header(n, skb);
869 
870         C(len);
871         C(data_len);
872         C(mac_len);
873         n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
874         n->cloned = 1;
875         n->nohdr = 0;
876         n->destructor = NULL;
877         C(tail);
878         C(end);
879         C(head);
880         C(head_frag);
881         C(data);
882         C(truesize);
883         atomic_set(&n->users, 1);
884 
885         atomic_inc(&(skb_shinfo(skb)->dataref));
886         skb->cloned = 1;
887 
888         return n;
889 #undef C
890 }
891 
892 /**
893  *      skb_morph       -       morph one skb into another
894  *      @dst: the skb to receive the contents
895  *      @src: the skb to supply the contents
896  *
897  *      This is identical to skb_clone except that the target skb is
898  *      supplied by the user.
899  *
900  *      The target skb is returned upon exit.
901  */
902 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
903 {
904         skb_release_all(dst);
905         return __skb_clone(dst, src);
906 }
907 EXPORT_SYMBOL_GPL(skb_morph);
908 
909 /**
910  *      skb_copy_ubufs  -       copy userspace skb frags buffers to kernel
911  *      @skb: the skb to modify
912  *      @gfp_mask: allocation priority
913  *
914  *      This must be called on SKBTX_DEV_ZEROCOPY skb.
915  *      It will copy all frags into kernel and drop the reference
916  *      to userspace pages.
917  *
918  *      If this function is called from an interrupt gfp_mask() must be
919  *      %GFP_ATOMIC.
920  *
921  *      Returns 0 on success or a negative error code on failure
922  *      to allocate kernel memory to copy to.
923  */
924 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
925 {
926         int i;
927         int num_frags = skb_shinfo(skb)->nr_frags;
928         struct page *page, *head = NULL;
929         struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
930 
931         for (i = 0; i < num_frags; i++) {
932                 u8 *vaddr;
933                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
934 
935                 page = alloc_page(gfp_mask);
936                 if (!page) {
937                         while (head) {
938                                 struct page *next = (struct page *)page_private(head);
939                                 put_page(head);
940                                 head = next;
941                         }
942                         return -ENOMEM;
943                 }
944                 vaddr = kmap_atomic(skb_frag_page(f));
945                 memcpy(page_address(page),
946                        vaddr + f->page_offset, skb_frag_size(f));
947                 kunmap_atomic(vaddr);
948                 set_page_private(page, (unsigned long)head);
949                 head = page;
950         }
951 
952         /* skb frags release userspace buffers */
953         for (i = 0; i < num_frags; i++)
954                 skb_frag_unref(skb, i);
955 
956         uarg->callback(uarg, false);
957 
958         /* skb frags point to kernel buffers */
959         for (i = num_frags - 1; i >= 0; i--) {
960                 __skb_fill_page_desc(skb, i, head, 0,
961                                      skb_shinfo(skb)->frags[i].size);
962                 head = (struct page *)page_private(head);
963         }
964 
965         skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
966         return 0;
967 }
968 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
969 
970 /**
971  *      skb_clone       -       duplicate an sk_buff
972  *      @skb: buffer to clone
973  *      @gfp_mask: allocation priority
974  *
975  *      Duplicate an &sk_buff. The new one is not owned by a socket. Both
976  *      copies share the same packet data but not structure. The new
977  *      buffer has a reference count of 1. If the allocation fails the
978  *      function returns %NULL otherwise the new buffer is returned.
979  *
980  *      If this function is called from an interrupt gfp_mask() must be
981  *      %GFP_ATOMIC.
982  */
983 
984 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
985 {
986         struct sk_buff_fclones *fclones = container_of(skb,
987                                                        struct sk_buff_fclones,
988                                                        skb1);
989         struct sk_buff *n;
990 
991         if (skb_orphan_frags(skb, gfp_mask))
992                 return NULL;
993 
994         if (skb->fclone == SKB_FCLONE_ORIG &&
995             atomic_read(&fclones->fclone_ref) == 1) {
996                 n = &fclones->skb2;
997                 atomic_set(&fclones->fclone_ref, 2);
998         } else {
999                 if (skb_pfmemalloc(skb))
1000                         gfp_mask |= __GFP_MEMALLOC;
1001 
1002                 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1003                 if (!n)
1004                         return NULL;
1005 
1006                 kmemcheck_annotate_bitfield(n, flags1);
1007                 n->fclone = SKB_FCLONE_UNAVAILABLE;
1008         }
1009 
1010         return __skb_clone(n, skb);
1011 }
1012 EXPORT_SYMBOL(skb_clone);
1013 
1014 static void skb_headers_offset_update(struct sk_buff *skb, int off)
1015 {
1016         /* Only adjust this if it actually is csum_start rather than csum */
1017         if (skb->ip_summed == CHECKSUM_PARTIAL)
1018                 skb->csum_start += off;
1019         /* {transport,network,mac}_header and tail are relative to skb->head */
1020         skb->transport_header += off;
1021         skb->network_header   += off;
1022         if (skb_mac_header_was_set(skb))
1023                 skb->mac_header += off;
1024         skb->inner_transport_header += off;
1025         skb->inner_network_header += off;
1026         skb->inner_mac_header += off;
1027 }
1028 
1029 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
1030 {
1031         __copy_skb_header(new, old);
1032 
1033         skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1034         skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1035         skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1036 }
1037 
1038 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1039 {
1040         if (skb_pfmemalloc(skb))
1041                 return SKB_ALLOC_RX;
1042         return 0;
1043 }
1044 
1045 /**
1046  *      skb_copy        -       create private copy of an sk_buff
1047  *      @skb: buffer to copy
1048  *      @gfp_mask: allocation priority
1049  *
1050  *      Make a copy of both an &sk_buff and its data. This is used when the
1051  *      caller wishes to modify the data and needs a private copy of the
1052  *      data to alter. Returns %NULL on failure or the pointer to the buffer
1053  *      on success. The returned buffer has a reference count of 1.
1054  *
1055  *      As by-product this function converts non-linear &sk_buff to linear
1056  *      one, so that &sk_buff becomes completely private and caller is allowed
1057  *      to modify all the data of returned buffer. This means that this
1058  *      function is not recommended for use in circumstances when only
1059  *      header is going to be modified. Use pskb_copy() instead.
1060  */
1061 
1062 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1063 {
1064         int headerlen = skb_headroom(skb);
1065         unsigned int size = skb_end_offset(skb) + skb->data_len;
1066         struct sk_buff *n = __alloc_skb(size, gfp_mask,
1067                                         skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1068 
1069         if (!n)
1070                 return NULL;
1071 
1072         /* Set the data pointer */
1073         skb_reserve(n, headerlen);
1074         /* Set the tail pointer and length */
1075         skb_put(n, skb->len);
1076 
1077         if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
1078                 BUG();
1079 
1080         copy_skb_header(n, skb);
1081         return n;
1082 }
1083 EXPORT_SYMBOL(skb_copy);
1084 
1085 /**
1086  *      __pskb_copy_fclone      -  create copy of an sk_buff with private head.
1087  *      @skb: buffer to copy
1088  *      @headroom: headroom of new skb
1089  *      @gfp_mask: allocation priority
1090  *      @fclone: if true allocate the copy of the skb from the fclone
1091  *      cache instead of the head cache; it is recommended to set this
1092  *      to true for the cases where the copy will likely be cloned
1093  *
1094  *      Make a copy of both an &sk_buff and part of its data, located
1095  *      in header. Fragmented data remain shared. This is used when
1096  *      the caller wishes to modify only header of &sk_buff and needs
1097  *      private copy of the header to alter. Returns %NULL on failure
1098  *      or the pointer to the buffer on success.
1099  *      The returned buffer has a reference count of 1.
1100  */
1101 
1102 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1103                                    gfp_t gfp_mask, bool fclone)
1104 {
1105         unsigned int size = skb_headlen(skb) + headroom;
1106         int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1107         struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1108 
1109         if (!n)
1110                 goto out;
1111 
1112         /* Set the data pointer */
1113         skb_reserve(n, headroom);
1114         /* Set the tail pointer and length */
1115         skb_put(n, skb_headlen(skb));
1116         /* Copy the bytes */
1117         skb_copy_from_linear_data(skb, n->data, n->len);
1118 
1119         n->truesize += skb->data_len;
1120         n->data_len  = skb->data_len;
1121         n->len       = skb->len;
1122 
1123         if (skb_shinfo(skb)->nr_frags) {
1124                 int i;
1125 
1126                 if (skb_orphan_frags(skb, gfp_mask)) {
1127                         kfree_skb(n);
1128                         n = NULL;
1129                         goto out;
1130                 }
1131                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1132                         skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1133                         skb_frag_ref(skb, i);
1134                 }
1135                 skb_shinfo(n)->nr_frags = i;
1136         }
1137 
1138         if (skb_has_frag_list(skb)) {
1139                 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1140                 skb_clone_fraglist(n);
1141         }
1142 
1143         copy_skb_header(n, skb);
1144 out:
1145         return n;
1146 }
1147 EXPORT_SYMBOL(__pskb_copy_fclone);
1148 
1149 /**
1150  *      pskb_expand_head - reallocate header of &sk_buff
1151  *      @skb: buffer to reallocate
1152  *      @nhead: room to add at head
1153  *      @ntail: room to add at tail
1154  *      @gfp_mask: allocation priority
1155  *
1156  *      Expands (or creates identical copy, if @nhead and @ntail are zero)
1157  *      header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1158  *      reference count of 1. Returns zero in the case of success or error,
1159  *      if expansion failed. In the last case, &sk_buff is not changed.
1160  *
1161  *      All the pointers pointing into skb header may change and must be
1162  *      reloaded after call to this function.
1163  */
1164 
1165 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1166                      gfp_t gfp_mask)
1167 {
1168         int i;
1169         u8 *data;
1170         int size = nhead + skb_end_offset(skb) + ntail;
1171         long off;
1172 
1173         BUG_ON(nhead < 0);
1174 
1175         if (skb_shared(skb))
1176                 BUG();
1177 
1178         size = SKB_DATA_ALIGN(size);
1179 
1180         if (skb_pfmemalloc(skb))
1181                 gfp_mask |= __GFP_MEMALLOC;
1182         data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1183                                gfp_mask, NUMA_NO_NODE, NULL);
1184         if (!data)
1185                 goto nodata;
1186         size = SKB_WITH_OVERHEAD(ksize(data));
1187 
1188         /* Copy only real data... and, alas, header. This should be
1189          * optimized for the cases when header is void.
1190          */
1191         memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1192 
1193         memcpy((struct skb_shared_info *)(data + size),
1194                skb_shinfo(skb),
1195                offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1196 
1197         /*
1198          * if shinfo is shared we must drop the old head gracefully, but if it
1199          * is not we can just drop the old head and let the existing refcount
1200          * be since all we did is relocate the values
1201          */
1202         if (skb_cloned(skb)) {
1203                 /* copy this zero copy skb frags */
1204                 if (skb_orphan_frags(skb, gfp_mask))
1205                         goto nofrags;
1206                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1207                         skb_frag_ref(skb, i);
1208 
1209                 if (skb_has_frag_list(skb))
1210                         skb_clone_fraglist(skb);
1211 
1212                 skb_release_data(skb);
1213         } else {
1214                 skb_free_head(skb);
1215         }
1216         off = (data + nhead) - skb->head;
1217 
1218         skb->head     = data;
1219         skb->head_frag = 0;
1220         skb->data    += off;
1221 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1222         skb->end      = size;
1223         off           = nhead;
1224 #else
1225         skb->end      = skb->head + size;
1226 #endif
1227         skb->tail             += off;
1228         skb_headers_offset_update(skb, nhead);
1229         skb->cloned   = 0;
1230         skb->hdr_len  = 0;
1231         skb->nohdr    = 0;
1232         atomic_set(&skb_shinfo(skb)->dataref, 1);
1233         return 0;
1234 
1235 nofrags:
1236         kfree(data);
1237 nodata:
1238         return -ENOMEM;
1239 }
1240 EXPORT_SYMBOL(pskb_expand_head);
1241 
1242 /* Make private copy of skb with writable head and some headroom */
1243 
1244 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1245 {
1246         struct sk_buff *skb2;
1247         int delta = headroom - skb_headroom(skb);
1248 
1249         if (delta <= 0)
1250                 skb2 = pskb_copy(skb, GFP_ATOMIC);
1251         else {
1252                 skb2 = skb_clone(skb, GFP_ATOMIC);
1253                 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1254                                              GFP_ATOMIC)) {
1255                         kfree_skb(skb2);
1256                         skb2 = NULL;
1257                 }
1258         }
1259         return skb2;
1260 }
1261 EXPORT_SYMBOL(skb_realloc_headroom);
1262 
1263 /**
1264  *      skb_copy_expand -       copy and expand sk_buff
1265  *      @skb: buffer to copy
1266  *      @newheadroom: new free bytes at head
1267  *      @newtailroom: new free bytes at tail
1268  *      @gfp_mask: allocation priority
1269  *
1270  *      Make a copy of both an &sk_buff and its data and while doing so
1271  *      allocate additional space.
1272  *
1273  *      This is used when the caller wishes to modify the data and needs a
1274  *      private copy of the data to alter as well as more space for new fields.
1275  *      Returns %NULL on failure or the pointer to the buffer
1276  *      on success. The returned buffer has a reference count of 1.
1277  *
1278  *      You must pass %GFP_ATOMIC as the allocation priority if this function
1279  *      is called from an interrupt.
1280  */
1281 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1282                                 int newheadroom, int newtailroom,
1283                                 gfp_t gfp_mask)
1284 {
1285         /*
1286          *      Allocate the copy buffer
1287          */
1288         struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1289                                         gfp_mask, skb_alloc_rx_flag(skb),
1290                                         NUMA_NO_NODE);
1291         int oldheadroom = skb_headroom(skb);
1292         int head_copy_len, head_copy_off;
1293 
1294         if (!n)
1295                 return NULL;
1296 
1297         skb_reserve(n, newheadroom);
1298 
1299         /* Set the tail pointer and length */
1300         skb_put(n, skb->len);
1301 
1302         head_copy_len = oldheadroom;
1303         head_copy_off = 0;
1304         if (newheadroom <= head_copy_len)
1305                 head_copy_len = newheadroom;
1306         else
1307                 head_copy_off = newheadroom - head_copy_len;
1308 
1309         /* Copy the linear header and data. */
1310         if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1311                           skb->len + head_copy_len))
1312                 BUG();
1313 
1314         copy_skb_header(n, skb);
1315 
1316         skb_headers_offset_update(n, newheadroom - oldheadroom);
1317 
1318         return n;
1319 }
1320 EXPORT_SYMBOL(skb_copy_expand);
1321 
1322 /**
1323  *      skb_pad                 -       zero pad the tail of an skb
1324  *      @skb: buffer to pad
1325  *      @pad: space to pad
1326  *
1327  *      Ensure that a buffer is followed by a padding area that is zero
1328  *      filled. Used by network drivers which may DMA or transfer data
1329  *      beyond the buffer end onto the wire.
1330  *
1331  *      May return error in out of memory cases. The skb is freed on error.
1332  */
1333 
1334 int skb_pad(struct sk_buff *skb, int pad)
1335 {
1336         int err;
1337         int ntail;
1338 
1339         /* If the skbuff is non linear tailroom is always zero.. */
1340         if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1341                 memset(skb->data+skb->len, 0, pad);
1342                 return 0;
1343         }
1344 
1345         ntail = skb->data_len + pad - (skb->end - skb->tail);
1346         if (likely(skb_cloned(skb) || ntail > 0)) {
1347                 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1348                 if (unlikely(err))
1349                         goto free_skb;
1350         }
1351 
1352         /* FIXME: The use of this function with non-linear skb's really needs
1353          * to be audited.
1354          */
1355         err = skb_linearize(skb);
1356         if (unlikely(err))
1357                 goto free_skb;
1358 
1359         memset(skb->data + skb->len, 0, pad);
1360         return 0;
1361 
1362 free_skb:
1363         kfree_skb(skb);
1364         return err;
1365 }
1366 EXPORT_SYMBOL(skb_pad);
1367 
1368 /**
1369  *      pskb_put - add data to the tail of a potentially fragmented buffer
1370  *      @skb: start of the buffer to use
1371  *      @tail: tail fragment of the buffer to use
1372  *      @len: amount of data to add
1373  *
1374  *      This function extends the used data area of the potentially
1375  *      fragmented buffer. @tail must be the last fragment of @skb -- or
1376  *      @skb itself. If this would exceed the total buffer size the kernel
1377  *      will panic. A pointer to the first byte of the extra data is
1378  *      returned.
1379  */
1380 
1381 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1382 {
1383         if (tail != skb) {
1384                 skb->data_len += len;
1385                 skb->len += len;
1386         }
1387         return skb_put(tail, len);
1388 }
1389 EXPORT_SYMBOL_GPL(pskb_put);
1390 
1391 /**
1392  *      skb_put - add data to a buffer
1393  *      @skb: buffer to use
1394  *      @len: amount of data to add
1395  *
1396  *      This function extends the used data area of the buffer. If this would
1397  *      exceed the total buffer size the kernel will panic. A pointer to the
1398  *      first byte of the extra data is returned.
1399  */
1400 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1401 {
1402         unsigned char *tmp = skb_tail_pointer(skb);
1403         SKB_LINEAR_ASSERT(skb);
1404         skb->tail += len;
1405         skb->len  += len;
1406         if (unlikely(skb->tail > skb->end))
1407                 skb_over_panic(skb, len, __builtin_return_address(0));
1408         return tmp;
1409 }
1410 EXPORT_SYMBOL(skb_put);
1411 
1412 /**
1413  *      skb_push - add data to the start of a buffer
1414  *      @skb: buffer to use
1415  *      @len: amount of data to add
1416  *
1417  *      This function extends the used data area of the buffer at the buffer
1418  *      start. If this would exceed the total buffer headroom the kernel will
1419  *      panic. A pointer to the first byte of the extra data is returned.
1420  */
1421 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1422 {
1423         skb->data -= len;
1424         skb->len  += len;
1425         if (unlikely(skb->data<skb->head))
1426                 skb_under_panic(skb, len, __builtin_return_address(0));
1427         return skb->data;
1428 }
1429 EXPORT_SYMBOL(skb_push);
1430 
1431 /**
1432  *      skb_pull - remove data from the start of a buffer
1433  *      @skb: buffer to use
1434  *      @len: amount of data to remove
1435  *
1436  *      This function removes data from the start of a buffer, returning
1437  *      the memory to the headroom. A pointer to the next data in the buffer
1438  *      is returned. Once the data has been pulled future pushes will overwrite
1439  *      the old data.
1440  */
1441 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1442 {
1443         return skb_pull_inline(skb, len);
1444 }
1445 EXPORT_SYMBOL(skb_pull);
1446 
1447 /**
1448  *      skb_trim - remove end from a buffer
1449  *      @skb: buffer to alter
1450  *      @len: new length
1451  *
1452  *      Cut the length of a buffer down by removing data from the tail. If
1453  *      the buffer is already under the length specified it is not modified.
1454  *      The skb must be linear.
1455  */
1456 void skb_trim(struct sk_buff *skb, unsigned int len)
1457 {
1458         if (skb->len > len)
1459                 __skb_trim(skb, len);
1460 }
1461 EXPORT_SYMBOL(skb_trim);
1462 
1463 /* Trims skb to length len. It can change skb pointers.
1464  */
1465 
1466 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1467 {
1468         struct sk_buff **fragp;
1469         struct sk_buff *frag;
1470         int offset = skb_headlen(skb);
1471         int nfrags = skb_shinfo(skb)->nr_frags;
1472         int i;
1473         int err;
1474 
1475         if (skb_cloned(skb) &&
1476             unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1477                 return err;
1478 
1479         i = 0;
1480         if (offset >= len)
1481                 goto drop_pages;
1482 
1483         for (; i < nfrags; i++) {
1484                 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1485 
1486                 if (end < len) {
1487                         offset = end;
1488                         continue;
1489                 }
1490 
1491                 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1492 
1493 drop_pages:
1494                 skb_shinfo(skb)->nr_frags = i;
1495 
1496                 for (; i < nfrags; i++)
1497                         skb_frag_unref(skb, i);
1498 
1499                 if (skb_has_frag_list(skb))
1500                         skb_drop_fraglist(skb);
1501                 goto done;
1502         }
1503 
1504         for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1505              fragp = &frag->next) {
1506                 int end = offset + frag->len;
1507 
1508                 if (skb_shared(frag)) {
1509                         struct sk_buff *nfrag;
1510 
1511                         nfrag = skb_clone(frag, GFP_ATOMIC);
1512                         if (unlikely(!nfrag))
1513                                 return -ENOMEM;
1514 
1515                         nfrag->next = frag->next;
1516                         consume_skb(frag);
1517                         frag = nfrag;
1518                         *fragp = frag;
1519                 }
1520 
1521                 if (end < len) {
1522                         offset = end;
1523                         continue;
1524                 }
1525 
1526                 if (end > len &&
1527                     unlikely((err = pskb_trim(frag, len - offset))))
1528                         return err;
1529 
1530                 if (frag->next)
1531                         skb_drop_list(&frag->next);
1532                 break;
1533         }
1534 
1535 done:
1536         if (len > skb_headlen(skb)) {
1537                 skb->data_len -= skb->len - len;
1538                 skb->len       = len;
1539         } else {
1540                 skb->len       = len;
1541                 skb->data_len  = 0;
1542                 skb_set_tail_pointer(skb, len);
1543         }
1544 
1545         return 0;
1546 }
1547 EXPORT_SYMBOL(___pskb_trim);
1548 
1549 /**
1550  *      __pskb_pull_tail - advance tail of skb header
1551  *      @skb: buffer to reallocate
1552  *      @delta: number of bytes to advance tail
1553  *
1554  *      The function makes a sense only on a fragmented &sk_buff,
1555  *      it expands header moving its tail forward and copying necessary
1556  *      data from fragmented part.
1557  *
1558  *      &sk_buff MUST have reference count of 1.
1559  *
1560  *      Returns %NULL (and &sk_buff does not change) if pull failed
1561  *      or value of new tail of skb in the case of success.
1562  *
1563  *      All the pointers pointing into skb header may change and must be
1564  *      reloaded after call to this function.
1565  */
1566 
1567 /* Moves tail of skb head forward, copying data from fragmented part,
1568  * when it is necessary.
1569  * 1. It may fail due to malloc failure.
1570  * 2. It may change skb pointers.
1571  *
1572  * It is pretty complicated. Luckily, it is called only in exceptional cases.
1573  */
1574 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1575 {
1576         /* If skb has not enough free space at tail, get new one
1577          * plus 128 bytes for future expansions. If we have enough
1578          * room at tail, reallocate without expansion only if skb is cloned.
1579          */
1580         int i, k, eat = (skb->tail + delta) - skb->end;
1581 
1582         if (eat > 0 || skb_cloned(skb)) {
1583                 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1584                                      GFP_ATOMIC))
1585                         return NULL;
1586         }
1587 
1588         if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1589                 BUG();
1590 
1591         /* Optimization: no fragments, no reasons to preestimate
1592          * size of pulled pages. Superb.
1593          */
1594         if (!skb_has_frag_list(skb))
1595                 goto pull_pages;
1596 
1597         /* Estimate size of pulled pages. */
1598         eat = delta;
1599         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1600                 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1601 
1602                 if (size >= eat)
1603                         goto pull_pages;
1604                 eat -= size;
1605         }
1606 
1607         /* If we need update frag list, we are in troubles.
1608          * Certainly, it possible to add an offset to skb data,
1609          * but taking into account that pulling is expected to
1610          * be very rare operation, it is worth to fight against
1611          * further bloating skb head and crucify ourselves here instead.
1612          * Pure masohism, indeed. 8)8)
1613          */
1614         if (eat) {
1615                 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1616                 struct sk_buff *clone = NULL;
1617                 struct sk_buff *insp = NULL;
1618 
1619                 do {
1620                         BUG_ON(!list);
1621 
1622                         if (list->len <= eat) {
1623                                 /* Eaten as whole. */
1624                                 eat -= list->len;
1625                                 list = list->next;
1626                                 insp = list;
1627                         } else {
1628                                 /* Eaten partially. */
1629 
1630                                 if (skb_shared(list)) {
1631                                         /* Sucks! We need to fork list. :-( */
1632                                         clone = skb_clone(list, GFP_ATOMIC);
1633                                         if (!clone)
1634                                                 return NULL;
1635                                         insp = list->next;
1636                                         list = clone;
1637                                 } else {
1638                                         /* This may be pulled without
1639                                          * problems. */
1640                                         insp = list;
1641                                 }
1642                                 if (!pskb_pull(list, eat)) {
1643                                         kfree_skb(clone);
1644                                         return NULL;
1645                                 }
1646                                 break;
1647                         }
1648                 } while (eat);
1649 
1650                 /* Free pulled out fragments. */
1651                 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1652                         skb_shinfo(skb)->frag_list = list->next;
1653                         kfree_skb(list);
1654                 }
1655                 /* And insert new clone at head. */
1656                 if (clone) {
1657                         clone->next = list;
1658                         skb_shinfo(skb)->frag_list = clone;
1659                 }
1660         }
1661         /* Success! Now we may commit changes to skb data. */
1662 
1663 pull_pages:
1664         eat = delta;
1665         k = 0;
1666         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1667                 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1668 
1669                 if (size <= eat) {
1670                         skb_frag_unref(skb, i);
1671                         eat -= size;
1672                 } else {
1673                         skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1674                         if (eat) {
1675                                 skb_shinfo(skb)->frags[k].page_offset += eat;
1676                                 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1677                                 eat = 0;
1678                         }
1679                         k++;
1680                 }
1681         }
1682         skb_shinfo(skb)->nr_frags = k;
1683 
1684         skb->tail     += delta;
1685         skb->data_len -= delta;
1686 
1687         return skb_tail_pointer(skb);
1688 }
1689 EXPORT_SYMBOL(__pskb_pull_tail);
1690 
1691 /**
1692  *      skb_copy_bits - copy bits from skb to kernel buffer
1693  *      @skb: source skb
1694  *      @offset: offset in source
1695  *      @to: destination buffer
1696  *      @len: number of bytes to copy
1697  *
1698  *      Copy the specified number of bytes from the source skb to the
1699  *      destination buffer.
1700  *
1701  *      CAUTION ! :
1702  *              If its prototype is ever changed,
1703  *              check arch/{*}/net/{*}.S files,
1704  *              since it is called from BPF assembly code.
1705  */
1706 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1707 {
1708         int start = skb_headlen(skb);
1709         struct sk_buff *frag_iter;
1710         int i, copy;
1711 
1712         if (offset > (int)skb->len - len)
1713                 goto fault;
1714 
1715         /* Copy header. */
1716         if ((copy = start - offset) > 0) {
1717                 if (copy > len)
1718                         copy = len;
1719                 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1720                 if ((len -= copy) == 0)
1721                         return 0;
1722                 offset += copy;
1723                 to     += copy;
1724         }
1725 
1726         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1727                 int end;
1728                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1729 
1730                 WARN_ON(start > offset + len);
1731 
1732                 end = start + skb_frag_size(f);
1733                 if ((copy = end - offset) > 0) {
1734                         u8 *vaddr;
1735 
1736                         if (copy > len)
1737                                 copy = len;
1738 
1739                         vaddr = kmap_atomic(skb_frag_page(f));
1740                         memcpy(to,
1741                                vaddr + f->page_offset + offset - start,
1742                                copy);
1743                         kunmap_atomic(vaddr);
1744 
1745                         if ((len -= copy) == 0)
1746                                 return 0;
1747                         offset += copy;
1748                         to     += copy;
1749                 }
1750                 start = end;
1751         }
1752 
1753         skb_walk_frags(skb, frag_iter) {
1754                 int end;
1755 
1756                 WARN_ON(start > offset + len);
1757 
1758                 end = start + frag_iter->len;
1759                 if ((copy = end - offset) > 0) {
1760                         if (copy > len)
1761                                 copy = len;
1762                         if (skb_copy_bits(frag_iter, offset - start, to, copy))
1763                                 goto fault;
1764                         if ((len -= copy) == 0)
1765                                 return 0;
1766                         offset += copy;
1767                         to     += copy;
1768                 }
1769                 start = end;
1770         }
1771 
1772         if (!len)
1773                 return 0;
1774 
1775 fault:
1776         return -EFAULT;
1777 }
1778 EXPORT_SYMBOL(skb_copy_bits);
1779 
1780 /*
1781  * Callback from splice_to_pipe(), if we need to release some pages
1782  * at the end of the spd in case we error'ed out in filling the pipe.
1783  */
1784 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1785 {
1786         put_page(spd->pages[i]);
1787 }
1788 
1789 static struct page *linear_to_page(struct page *page, unsigned int *len,
1790                                    unsigned int *offset,
1791                                    struct sock *sk)
1792 {
1793         struct page_frag *pfrag = sk_page_frag(sk);
1794 
1795         if (!sk_page_frag_refill(sk, pfrag))
1796                 return NULL;
1797 
1798         *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
1799 
1800         memcpy(page_address(pfrag->page) + pfrag->offset,
1801                page_address(page) + *offset, *len);
1802         *offset = pfrag->offset;
1803         pfrag->offset += *len;
1804 
1805         return pfrag->page;
1806 }
1807 
1808 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1809                              struct page *page,
1810                              unsigned int offset)
1811 {
1812         return  spd->nr_pages &&
1813                 spd->pages[spd->nr_pages - 1] == page &&
1814                 (spd->partial[spd->nr_pages - 1].offset +
1815                  spd->partial[spd->nr_pages - 1].len == offset);
1816 }
1817 
1818 /*
1819  * Fill page/offset/length into spd, if it can hold more pages.
1820  */
1821 static bool spd_fill_page(struct splice_pipe_desc *spd,
1822                           struct pipe_inode_info *pipe, struct page *page,
1823                           unsigned int *len, unsigned int offset,
1824                           bool linear,
1825                           struct sock *sk)
1826 {
1827         if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1828                 return true;
1829 
1830         if (linear) {
1831                 page = linear_to_page(page, len, &offset, sk);
1832                 if (!page)
1833                         return true;
1834         }
1835         if (spd_can_coalesce(spd, page, offset)) {
1836                 spd->partial[spd->nr_pages - 1].len += *len;
1837                 return false;
1838         }
1839         get_page(page);
1840         spd->pages[spd->nr_pages] = page;
1841         spd->partial[spd->nr_pages].len = *len;
1842         spd->partial[spd->nr_pages].offset = offset;
1843         spd->nr_pages++;
1844 
1845         return false;
1846 }
1847 
1848 static bool __splice_segment(struct page *page, unsigned int poff,
1849                              unsigned int plen, unsigned int *off,
1850                              unsigned int *len,
1851                              struct splice_pipe_desc *spd, bool linear,
1852                              struct sock *sk,
1853                              struct pipe_inode_info *pipe)
1854 {
1855         if (!*len)
1856                 return true;
1857 
1858         /* skip this segment if already processed */
1859         if (*off >= plen) {
1860                 *off -= plen;
1861                 return false;
1862         }
1863 
1864         /* ignore any bits we already processed */
1865         poff += *off;
1866         plen -= *off;
1867         *off = 0;
1868 
1869         do {
1870                 unsigned int flen = min(*len, plen);
1871 
1872                 if (spd_fill_page(spd, pipe, page, &flen, poff,
1873                                   linear, sk))
1874                         return true;
1875                 poff += flen;
1876                 plen -= flen;
1877                 *len -= flen;
1878         } while (*len && plen);
1879 
1880         return false;
1881 }
1882 
1883 /*
1884  * Map linear and fragment data from the skb to spd. It reports true if the
1885  * pipe is full or if we already spliced the requested length.
1886  */
1887 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1888                               unsigned int *offset, unsigned int *len,
1889                               struct splice_pipe_desc *spd, struct sock *sk)
1890 {
1891         int seg;
1892 
1893         /* map the linear part :
1894          * If skb->head_frag is set, this 'linear' part is backed by a
1895          * fragment, and if the head is not shared with any clones then
1896          * we can avoid a copy since we own the head portion of this page.
1897          */
1898         if (__splice_segment(virt_to_page(skb->data),
1899                              (unsigned long) skb->data & (PAGE_SIZE - 1),
1900                              skb_headlen(skb),
1901                              offset, len, spd,
1902                              skb_head_is_locked(skb),
1903                              sk, pipe))
1904                 return true;
1905 
1906         /*
1907          * then map the fragments
1908          */
1909         for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1910                 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1911 
1912                 if (__splice_segment(skb_frag_page(f),
1913                                      f->page_offset, skb_frag_size(f),
1914                                      offset, len, spd, false, sk, pipe))
1915                         return true;
1916         }
1917 
1918         return false;
1919 }
1920 
1921 /*
1922  * Map data from the skb to a pipe. Should handle both the linear part,
1923  * the fragments, and the frag list. It does NOT handle frag lists within
1924  * the frag list, if such a thing exists. We'd probably need to recurse to
1925  * handle that cleanly.
1926  */
1927 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1928                     struct pipe_inode_info *pipe, unsigned int tlen,
1929                     unsigned int flags)
1930 {
1931         struct partial_page partial[MAX_SKB_FRAGS];
1932         struct page *pages[MAX_SKB_FRAGS];
1933         struct splice_pipe_desc spd = {
1934                 .pages = pages,
1935                 .partial = partial,
1936                 .nr_pages_max = MAX_SKB_FRAGS,
1937                 .flags = flags,
1938                 .ops = &nosteal_pipe_buf_ops,
1939                 .spd_release = sock_spd_release,
1940         };
1941         struct sk_buff *frag_iter;
1942         struct sock *sk = skb->sk;
1943         int ret = 0;
1944 
1945         /*
1946          * __skb_splice_bits() only fails if the output has no room left,
1947          * so no point in going over the frag_list for the error case.
1948          */
1949         if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1950                 goto done;
1951         else if (!tlen)
1952                 goto done;
1953 
1954         /*
1955          * now see if we have a frag_list to map
1956          */
1957         skb_walk_frags(skb, frag_iter) {
1958                 if (!tlen)
1959                         break;
1960                 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1961                         break;
1962         }
1963 
1964 done:
1965         if (spd.nr_pages) {
1966                 /*
1967                  * Drop the socket lock, otherwise we have reverse
1968                  * locking dependencies between sk_lock and i_mutex
1969                  * here as compared to sendfile(). We enter here
1970                  * with the socket lock held, and splice_to_pipe() will
1971                  * grab the pipe inode lock. For sendfile() emulation,
1972                  * we call into ->sendpage() with the i_mutex lock held
1973                  * and networking will grab the socket lock.
1974                  */
1975                 release_sock(sk);
1976                 ret = splice_to_pipe(pipe, &spd);
1977                 lock_sock(sk);
1978         }
1979 
1980         return ret;
1981 }
1982 
1983 /**
1984  *      skb_store_bits - store bits from kernel buffer to skb
1985  *      @skb: destination buffer
1986  *      @offset: offset in destination
1987  *      @from: source buffer
1988  *      @len: number of bytes to copy
1989  *
1990  *      Copy the specified number of bytes from the source buffer to the
1991  *      destination skb.  This function handles all the messy bits of
1992  *      traversing fragment lists and such.
1993  */
1994 
1995 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1996 {
1997         int start = skb_headlen(skb);
1998         struct sk_buff *frag_iter;
1999         int i, copy;
2000 
2001         if (offset > (int)skb->len - len)
2002                 goto fault;
2003 
2004         if ((copy = start - offset) > 0) {
2005                 if (copy > len)
2006                         copy = len;
2007                 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2008                 if ((len -= copy) == 0)
2009                         return 0;
2010                 offset += copy;
2011                 from += copy;
2012         }
2013 
2014         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2015                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2016                 int end;
2017 
2018                 WARN_ON(start > offset + len);
2019 
2020                 end = start + skb_frag_size(frag);
2021                 if ((copy = end - offset) > 0) {
2022                         u8 *vaddr;
2023 
2024                         if (copy > len)
2025                                 copy = len;
2026 
2027                         vaddr = kmap_atomic(skb_frag_page(frag));
2028                         memcpy(vaddr + frag->page_offset + offset - start,
2029                                from, copy);
2030                         kunmap_atomic(vaddr);
2031 
2032                         if ((len -= copy) == 0)
2033                                 return 0;
2034                         offset += copy;
2035                         from += copy;
2036                 }
2037                 start = end;
2038         }
2039 
2040         skb_walk_frags(skb, frag_iter) {
2041                 int end;
2042 
2043                 WARN_ON(start > offset + len);
2044 
2045                 end = start + frag_iter->len;
2046                 if ((copy = end - offset) > 0) {
2047                         if (copy > len)
2048                                 copy = len;
2049                         if (skb_store_bits(frag_iter, offset - start,
2050                                            from, copy))
2051                                 goto fault;
2052                         if ((len -= copy) == 0)
2053                                 return 0;
2054                         offset += copy;
2055                         from += copy;
2056                 }
2057                 start = end;
2058         }
2059         if (!len)
2060                 return 0;
2061 
2062 fault:
2063         return -EFAULT;
2064 }
2065 EXPORT_SYMBOL(skb_store_bits);
2066 
2067 /* Checksum skb data. */
2068 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2069                       __wsum csum, const struct skb_checksum_ops *ops)
2070 {
2071         int start = skb_headlen(skb);
2072         int i, copy = start - offset;
2073         struct sk_buff *frag_iter;
2074         int pos = 0;
2075 
2076         /* Checksum header. */
2077         if (copy > 0) {
2078                 if (copy > len)
2079                         copy = len;
2080                 csum = ops->update(skb->data + offset, copy, csum);
2081                 if ((len -= copy) == 0)
2082                         return csum;
2083                 offset += copy;
2084                 pos     = copy;
2085         }
2086 
2087         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2088                 int end;
2089                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2090 
2091                 WARN_ON(start > offset + len);
2092 
2093                 end = start + skb_frag_size(frag);
2094                 if ((copy = end - offset) > 0) {
2095                         __wsum csum2;
2096                         u8 *vaddr;
2097 
2098                         if (copy > len)
2099                                 copy = len;
2100                         vaddr = kmap_atomic(skb_frag_page(frag));
2101                         csum2 = ops->update(vaddr + frag->page_offset +
2102                                             offset - start, copy, 0);
2103                         kunmap_atomic(vaddr);
2104                         csum = ops->combine(csum, csum2, pos, copy);
2105                         if (!(len -= copy))
2106                                 return csum;
2107                         offset += copy;
2108                         pos    += copy;
2109                 }
2110                 start = end;
2111         }
2112 
2113         skb_walk_frags(skb, frag_iter) {
2114                 int end;
2115 
2116                 WARN_ON(start > offset + len);
2117 
2118                 end = start + frag_iter->len;
2119                 if ((copy = end - offset) > 0) {
2120                         __wsum csum2;
2121                         if (copy > len)
2122                                 copy = len;
2123                         csum2 = __skb_checksum(frag_iter, offset - start,
2124                                                copy, 0, ops);
2125                         csum = ops->combine(csum, csum2, pos, copy);
2126                         if ((len -= copy) == 0)
2127                                 return csum;
2128                         offset += copy;
2129                         pos    += copy;
2130                 }
2131                 start = end;
2132         }
2133         BUG_ON(len);
2134 
2135         return csum;
2136 }
2137 EXPORT_SYMBOL(__skb_checksum);
2138 
2139 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2140                     int len, __wsum csum)
2141 {
2142         const struct skb_checksum_ops ops = {
2143                 .update  = csum_partial_ext,
2144                 .combine = csum_block_add_ext,
2145         };
2146 
2147         return __skb_checksum(skb, offset, len, csum, &ops);
2148 }
2149 EXPORT_SYMBOL(skb_checksum);
2150 
2151 /* Both of above in one bottle. */
2152 
2153 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2154                                     u8 *to, int len, __wsum csum)
2155 {
2156         int start = skb_headlen(skb);
2157         int i, copy = start - offset;
2158         struct sk_buff *frag_iter;
2159         int pos = 0;
2160 
2161         /* Copy header. */
2162         if (copy > 0) {
2163                 if (copy > len)
2164                         copy = len;
2165                 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2166                                                  copy, csum);
2167                 if ((len -= copy) == 0)
2168                         return csum;
2169                 offset += copy;
2170                 to     += copy;
2171                 pos     = copy;
2172         }
2173 
2174         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2175                 int end;
2176 
2177                 WARN_ON(start > offset + len);
2178 
2179                 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2180                 if ((copy = end - offset) > 0) {
2181                         __wsum csum2;
2182                         u8 *vaddr;
2183                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2184 
2185                         if (copy > len)
2186                                 copy = len;
2187                         vaddr = kmap_atomic(skb_frag_page(frag));
2188                         csum2 = csum_partial_copy_nocheck(vaddr +
2189                                                           frag->page_offset +
2190                                                           offset - start, to,
2191                                                           copy, 0);
2192                         kunmap_atomic(vaddr);
2193                         csum = csum_block_add(csum, csum2, pos);
2194                         if (!(len -= copy))
2195                                 return csum;
2196                         offset += copy;
2197                         to     += copy;
2198                         pos    += copy;
2199                 }
2200                 start = end;
2201         }
2202 
2203         skb_walk_frags(skb, frag_iter) {
2204                 __wsum csum2;
2205                 int end;
2206 
2207                 WARN_ON(start > offset + len);
2208 
2209                 end = start + frag_iter->len;
2210                 if ((copy = end - offset) > 0) {
2211                         if (copy > len)
2212                                 copy = len;
2213                         csum2 = skb_copy_and_csum_bits(frag_iter,
2214                                                        offset - start,
2215                                                        to, copy, 0);
2216                         csum = csum_block_add(csum, csum2, pos);
2217                         if ((len -= copy) == 0)
2218                                 return csum;
2219                         offset += copy;
2220                         to     += copy;
2221                         pos    += copy;
2222                 }
2223                 start = end;
2224         }
2225         BUG_ON(len);
2226         return csum;
2227 }
2228 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2229 
2230  /**
2231  *      skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2232  *      @from: source buffer
2233  *
2234  *      Calculates the amount of linear headroom needed in the 'to' skb passed
2235  *      into skb_zerocopy().
2236  */
2237 unsigned int
2238 skb_zerocopy_headlen(const struct sk_buff *from)
2239 {
2240         unsigned int hlen = 0;
2241 
2242         if (!from->head_frag ||
2243             skb_headlen(from) < L1_CACHE_BYTES ||
2244             skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2245                 hlen = skb_headlen(from);
2246 
2247         if (skb_has_frag_list(from))
2248                 hlen = from->len;
2249 
2250         return hlen;
2251 }
2252 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2253 
2254 /**
2255  *      skb_zerocopy - Zero copy skb to skb
2256  *      @to: destination buffer
2257  *      @from: source buffer
2258  *      @len: number of bytes to copy from source buffer
2259  *      @hlen: size of linear headroom in destination buffer
2260  *
2261  *      Copies up to `len` bytes from `from` to `to` by creating references
2262  *      to the frags in the source buffer.
2263  *
2264  *      The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2265  *      headroom in the `to` buffer.
2266  *
2267  *      Return value:
2268  *      0: everything is OK
2269  *      -ENOMEM: couldn't orphan frags of @from due to lack of memory
2270  *      -EFAULT: skb_copy_bits() found some problem with skb geometry
2271  */
2272 int
2273 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2274 {
2275         int i, j = 0;
2276         int plen = 0; /* length of skb->head fragment */
2277         int ret;
2278         struct page *page;
2279         unsigned int offset;
2280 
2281         BUG_ON(!from->head_frag && !hlen);
2282 
2283         /* dont bother with small payloads */
2284         if (len <= skb_tailroom(to))
2285                 return skb_copy_bits(from, 0, skb_put(to, len), len);
2286 
2287         if (hlen) {
2288                 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2289                 if (unlikely(ret))
2290                         return ret;
2291                 len -= hlen;
2292         } else {
2293                 plen = min_t(int, skb_headlen(from), len);
2294                 if (plen) {
2295                         page = virt_to_head_page(from->head);
2296                         offset = from->data - (unsigned char *)page_address(page);
2297                         __skb_fill_page_desc(to, 0, page, offset, plen);
2298                         get_page(page);
2299                         j = 1;
2300                         len -= plen;
2301                 }
2302         }
2303 
2304         to->truesize += len + plen;
2305         to->len += len + plen;
2306         to->data_len += len + plen;
2307 
2308         if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2309                 skb_tx_error(from);
2310                 return -ENOMEM;
2311         }
2312 
2313         for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2314                 if (!len)
2315                         break;
2316                 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2317                 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
2318                 len -= skb_shinfo(to)->frags[j].size;
2319                 skb_frag_ref(to, j);
2320                 j++;
2321         }
2322         skb_shinfo(to)->nr_frags = j;
2323 
2324         return 0;
2325 }
2326 EXPORT_SYMBOL_GPL(skb_zerocopy);
2327 
2328 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2329 {
2330         __wsum csum;
2331         long csstart;
2332 
2333         if (skb->ip_summed == CHECKSUM_PARTIAL)
2334                 csstart = skb_checksum_start_offset(skb);
2335         else
2336                 csstart = skb_headlen(skb);
2337 
2338         BUG_ON(csstart > skb_headlen(skb));
2339 
2340         skb_copy_from_linear_data(skb, to, csstart);
2341 
2342         csum = 0;
2343         if (csstart != skb->len)
2344                 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2345                                               skb->len - csstart, 0);
2346 
2347         if (skb->ip_summed == CHECKSUM_PARTIAL) {
2348                 long csstuff = csstart + skb->csum_offset;
2349 
2350                 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2351         }
2352 }
2353 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2354 
2355 /**
2356  *      skb_dequeue - remove from the head of the queue
2357  *      @list: list to dequeue from
2358  *
2359  *      Remove the head of the list. The list lock is taken so the function
2360  *      may be used safely with other locking list functions. The head item is
2361  *      returned or %NULL if the list is empty.
2362  */
2363 
2364 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2365 {
2366         unsigned long flags;
2367         struct sk_buff *result;
2368 
2369         spin_lock_irqsave(&list->lock, flags);
2370         result = __skb_dequeue(list);
2371         spin_unlock_irqrestore(&list->lock, flags);
2372         return result;
2373 }
2374 EXPORT_SYMBOL(skb_dequeue);
2375 
2376 /**
2377  *      skb_dequeue_tail - remove from the tail of the queue
2378  *      @list: list to dequeue from
2379  *
2380  *      Remove the tail of the list. The list lock is taken so the function
2381  *      may be used safely with other locking list functions. The tail item is
2382  *      returned or %NULL if the list is empty.
2383  */
2384 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2385 {
2386         unsigned long flags;
2387         struct sk_buff *result;
2388 
2389         spin_lock_irqsave(&list->lock, flags);
2390         result = __skb_dequeue_tail(list);
2391         spin_unlock_irqrestore(&list->lock, flags);
2392         return result;
2393 }
2394 EXPORT_SYMBOL(skb_dequeue_tail);
2395 
2396 /**
2397  *      skb_queue_purge - empty a list
2398  *      @list: list to empty
2399  *
2400  *      Delete all buffers on an &sk_buff list. Each buffer is removed from
2401  *      the list and one reference dropped. This function takes the list
2402  *      lock and is atomic with respect to other list locking functions.
2403  */
2404 void skb_queue_purge(struct sk_buff_head *list)
2405 {
2406         struct sk_buff *skb;
2407         while ((skb = skb_dequeue(list)) != NULL)
2408                 kfree_skb(skb);
2409 }
2410 EXPORT_SYMBOL(skb_queue_purge);
2411 
2412 /**
2413  *      skb_queue_head - queue a buffer at the list head
2414  *      @list: list to use
2415  *      @newsk: buffer to queue
2416  *
2417  *      Queue a buffer at the start of the list. This function takes the
2418  *      list lock and can be used safely with other locking &sk_buff functions
2419  *      safely.
2420  *
2421  *      A buffer cannot be placed on two lists at the same time.
2422  */
2423 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2424 {
2425         unsigned long flags;
2426 
2427         spin_lock_irqsave(&list->lock, flags);
2428         __skb_queue_head(list, newsk);
2429         spin_unlock_irqrestore(&list->lock, flags);
2430 }
2431 EXPORT_SYMBOL(skb_queue_head);
2432 
2433 /**
2434  *      skb_queue_tail - queue a buffer at the list tail
2435  *      @list: list to use
2436  *      @newsk: buffer to queue
2437  *
2438  *      Queue a buffer at the tail of the list. This function takes the
2439  *      list lock and can be used safely with other locking &sk_buff functions
2440  *      safely.
2441  *
2442  *      A buffer cannot be placed on two lists at the same time.
2443  */
2444 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2445 {
2446         unsigned long flags;
2447 
2448         spin_lock_irqsave(&list->lock, flags);
2449         __skb_queue_tail(list, newsk);
2450         spin_unlock_irqrestore(&list->lock, flags);
2451 }
2452 EXPORT_SYMBOL(skb_queue_tail);
2453 
2454 /**
2455  *      skb_unlink      -       remove a buffer from a list
2456  *      @skb: buffer to remove
2457  *      @list: list to use
2458  *
2459  *      Remove a packet from a list. The list locks are taken and this
2460  *      function is atomic with respect to other list locked calls
2461  *
2462  *      You must know what list the SKB is on.
2463  */
2464 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2465 {
2466         unsigned long flags;
2467 
2468         spin_lock_irqsave(&list->lock, flags);
2469         __skb_unlink(skb, list);
2470         spin_unlock_irqrestore(&list->lock, flags);
2471 }
2472 EXPORT_SYMBOL(skb_unlink);
2473 
2474 /**
2475  *      skb_append      -       append a buffer
2476  *      @old: buffer to insert after
2477  *      @newsk: buffer to insert
2478  *      @list: list to use
2479  *
2480  *      Place a packet after a given packet in a list. The list locks are taken
2481  *      and this function is atomic with respect to other list locked calls.
2482  *      A buffer cannot be placed on two lists at the same time.
2483  */
2484 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2485 {
2486         unsigned long flags;
2487 
2488         spin_lock_irqsave(&list->lock, flags);
2489         __skb_queue_after(list, old, newsk);
2490         spin_unlock_irqrestore(&list->lock, flags);
2491 }
2492 EXPORT_SYMBOL(skb_append);
2493 
2494 /**
2495  *      skb_insert      -       insert a buffer
2496  *      @old: buffer to insert before
2497  *      @newsk: buffer to insert
2498  *      @list: list to use
2499  *
2500  *      Place a packet before a given packet in a list. The list locks are
2501  *      taken and this function is atomic with respect to other list locked
2502  *      calls.
2503  *
2504  *      A buffer cannot be placed on two lists at the same time.
2505  */
2506 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2507 {
2508         unsigned long flags;
2509 
2510         spin_lock_irqsave(&list->lock, flags);
2511         __skb_insert(newsk, old->prev, old, list);
2512         spin_unlock_irqrestore(&list->lock, flags);
2513 }
2514 EXPORT_SYMBOL(skb_insert);
2515 
2516 static inline void skb_split_inside_header(struct sk_buff *skb,
2517                                            struct sk_buff* skb1,
2518                                            const u32 len, const int pos)
2519 {
2520         int i;
2521 
2522         skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2523                                          pos - len);
2524         /* And move data appendix as is. */
2525         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2526                 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2527 
2528         skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2529         skb_shinfo(skb)->nr_frags  = 0;
2530         skb1->data_len             = skb->data_len;
2531         skb1->len                  += skb1->data_len;
2532         skb->data_len              = 0;
2533         skb->len                   = len;
2534         skb_set_tail_pointer(skb, len);
2535 }
2536 
2537 static inline void skb_split_no_header(struct sk_buff *skb,
2538                                        struct sk_buff* skb1,
2539                                        const u32 len, int pos)
2540 {
2541         int i, k = 0;
2542         const int nfrags = skb_shinfo(skb)->nr_frags;
2543 
2544         skb_shinfo(skb)->nr_frags = 0;
2545         skb1->len                 = skb1->data_len = skb->len - len;
2546         skb->len                  = len;
2547         skb->data_len             = len - pos;
2548 
2549         for (i = 0; i < nfrags; i++) {
2550                 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2551 
2552                 if (pos + size > len) {
2553                         skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2554 
2555                         if (pos < len) {
2556                                 /* Split frag.
2557                                  * We have two variants in this case:
2558                                  * 1. Move all the frag to the second
2559                                  *    part, if it is possible. F.e.
2560                                  *    this approach is mandatory for TUX,
2561                                  *    where splitting is expensive.
2562                                  * 2. Split is accurately. We make this.
2563                                  */
2564                                 skb_frag_ref(skb, i);
2565                                 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2566                                 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2567                                 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2568                                 skb_shinfo(skb)->nr_frags++;
2569                         }
2570                         k++;
2571                 } else
2572                         skb_shinfo(skb)->nr_frags++;
2573                 pos += size;
2574         }
2575         skb_shinfo(skb1)->nr_frags = k;
2576 }
2577 
2578 /**
2579  * skb_split - Split fragmented skb to two parts at length len.
2580  * @skb: the buffer to split
2581  * @skb1: the buffer to receive the second part
2582  * @len: new length for skb
2583  */
2584 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2585 {
2586         int pos = skb_headlen(skb);
2587 
2588         skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2589         if (len < pos)  /* Split line is inside header. */
2590                 skb_split_inside_header(skb, skb1, len, pos);
2591         else            /* Second chunk has no header, nothing to copy. */
2592                 skb_split_no_header(skb, skb1, len, pos);
2593 }
2594 EXPORT_SYMBOL(skb_split);
2595 
2596 /* Shifting from/to a cloned skb is a no-go.
2597  *
2598  * Caller cannot keep skb_shinfo related pointers past calling here!
2599  */
2600 static int skb_prepare_for_shift(struct sk_buff *skb)
2601 {
2602         return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2603 }
2604 
2605 /**
2606  * skb_shift - Shifts paged data partially from skb to another
2607  * @tgt: buffer into which tail data gets added
2608  * @skb: buffer from which the paged data comes from
2609  * @shiftlen: shift up to this many bytes
2610  *
2611  * Attempts to shift up to shiftlen worth of bytes, which may be less than
2612  * the length of the skb, from skb to tgt. Returns number bytes shifted.
2613  * It's up to caller to free skb if everything was shifted.
2614  *
2615  * If @tgt runs out of frags, the whole operation is aborted.
2616  *
2617  * Skb cannot include anything else but paged data while tgt is allowed
2618  * to have non-paged data as well.
2619  *
2620  * TODO: full sized shift could be optimized but that would need
2621  * specialized skb free'er to handle frags without up-to-date nr_frags.
2622  */
2623 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2624 {
2625         int from, to, merge, todo;
2626         struct skb_frag_struct *fragfrom, *fragto;
2627 
2628         BUG_ON(shiftlen > skb->len);
2629         BUG_ON(skb_headlen(skb));       /* Would corrupt stream */
2630 
2631         todo = shiftlen;
2632         from = 0;
2633         to = skb_shinfo(tgt)->nr_frags;
2634         fragfrom = &skb_shinfo(skb)->frags[from];
2635 
2636         /* Actual merge is delayed until the point when we know we can
2637          * commit all, so that we don't have to undo partial changes
2638          */
2639         if (!to ||
2640             !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2641                               fragfrom->page_offset)) {
2642                 merge = -1;
2643         } else {
2644                 merge = to - 1;
2645 
2646                 todo -= skb_frag_size(fragfrom);
2647                 if (todo < 0) {
2648                         if (skb_prepare_for_shift(skb) ||
2649                             skb_prepare_for_shift(tgt))
2650                                 return 0;
2651 
2652                         /* All previous frag pointers might be stale! */
2653                         fragfrom = &skb_shinfo(skb)->frags[from];
2654                         fragto = &skb_shinfo(tgt)->frags[merge];
2655 
2656                         skb_frag_size_add(fragto, shiftlen);
2657                         skb_frag_size_sub(fragfrom, shiftlen);
2658                         fragfrom->page_offset += shiftlen;
2659 
2660                         goto onlymerged;
2661                 }
2662 
2663                 from++;
2664         }
2665 
2666         /* Skip full, not-fitting skb to avoid expensive operations */
2667         if ((shiftlen == skb->len) &&
2668             (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2669                 return 0;
2670 
2671         if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2672                 return 0;
2673 
2674         while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2675                 if (to == MAX_SKB_FRAGS)
2676                         return 0;
2677 
2678                 fragfrom = &skb_shinfo(skb)->frags[from];
2679                 fragto = &skb_shinfo(tgt)->frags[to];
2680 
2681                 if (todo >= skb_frag_size(fragfrom)) {
2682                         *fragto = *fragfrom;
2683                         todo -= skb_frag_size(fragfrom);
2684                         from++;
2685                         to++;
2686 
2687                 } else {
2688                         __skb_frag_ref(fragfrom);
2689                         fragto->page = fragfrom->page;
2690                         fragto->page_offset = fragfrom->page_offset;
2691                         skb_frag_size_set(fragto, todo);
2692 
2693                         fragfrom->page_offset += todo;
2694                         skb_frag_size_sub(fragfrom, todo);
2695                         todo = 0;
2696 
2697                         to++;
2698                         break;
2699                 }
2700         }
2701 
2702         /* Ready to "commit" this state change to tgt */
2703         skb_shinfo(tgt)->nr_frags = to;
2704 
2705         if (merge >= 0) {
2706                 fragfrom = &skb_shinfo(skb)->frags[0];
2707                 fragto = &skb_shinfo(tgt)->frags[merge];
2708 
2709                 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2710                 __skb_frag_unref(fragfrom);
2711         }
2712 
2713         /* Reposition in the original skb */
2714         to = 0;
2715         while (from < skb_shinfo(skb)->nr_frags)
2716                 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2717         skb_shinfo(skb)->nr_frags = to;
2718 
2719         BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2720 
2721 onlymerged:
2722         /* Most likely the tgt won't ever need its checksum anymore, skb on
2723          * the other hand might need it if it needs to be resent
2724          */
2725         tgt->ip_summed = CHECKSUM_PARTIAL;
2726         skb->ip_summed = CHECKSUM_PARTIAL;
2727 
2728         /* Yak, is it really working this way? Some helper please? */
2729         skb->len -= shiftlen;
2730         skb->data_len -= shiftlen;
2731         skb->truesize -= shiftlen;
2732         tgt->len += shiftlen;
2733         tgt->data_len += shiftlen;
2734         tgt->truesize += shiftlen;
2735 
2736         return shiftlen;
2737 }
2738 
2739 /**
2740  * skb_prepare_seq_read - Prepare a sequential read of skb data
2741  * @skb: the buffer to read
2742  * @from: lower offset of data to be read
2743  * @to: upper offset of data to be read
2744  * @st: state variable
2745  *
2746  * Initializes the specified state variable. Must be called before
2747  * invoking skb_seq_read() for the first time.
2748  */
2749 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2750                           unsigned int to, struct skb_seq_state *st)
2751 {
2752         st->lower_offset = from;
2753         st->upper_offset = to;
2754         st->root_skb = st->cur_skb = skb;
2755         st->frag_idx = st->stepped_offset = 0;
2756         st->frag_data = NULL;
2757 }
2758 EXPORT_SYMBOL(skb_prepare_seq_read);
2759 
2760 /**
2761  * skb_seq_read - Sequentially read skb data
2762  * @consumed: number of bytes consumed by the caller so far
2763  * @data: destination pointer for data to be returned
2764  * @st: state variable
2765  *
2766  * Reads a block of skb data at @consumed relative to the
2767  * lower offset specified to skb_prepare_seq_read(). Assigns
2768  * the head of the data block to @data and returns the length
2769  * of the block or 0 if the end of the skb data or the upper
2770  * offset has been reached.
2771  *
2772  * The caller is not required to consume all of the data
2773  * returned, i.e. @consumed is typically set to the number
2774  * of bytes already consumed and the next call to
2775  * skb_seq_read() will return the remaining part of the block.
2776  *
2777  * Note 1: The size of each block of data returned can be arbitrary,
2778  *       this limitation is the cost for zerocopy sequential
2779  *       reads of potentially non linear data.
2780  *
2781  * Note 2: Fragment lists within fragments are not implemented
2782  *       at the moment, state->root_skb could be replaced with
2783  *       a stack for this purpose.
2784  */
2785 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2786                           struct skb_seq_state *st)
2787 {
2788         unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2789         skb_frag_t *frag;
2790 
2791         if (unlikely(abs_offset >= st->upper_offset)) {
2792                 if (st->frag_data) {
2793                         kunmap_atomic(st->frag_data);
2794                         st->frag_data = NULL;
2795                 }
2796                 return 0;
2797         }
2798 
2799 next_skb:
2800         block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2801 
2802         if (abs_offset < block_limit && !st->frag_data) {
2803                 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2804                 return block_limit - abs_offset;
2805         }
2806 
2807         if (st->frag_idx == 0 && !st->frag_data)
2808                 st->stepped_offset += skb_headlen(st->cur_skb);
2809 
2810         while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2811                 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2812                 block_limit = skb_frag_size(frag) + st->stepped_offset;
2813 
2814                 if (abs_offset < block_limit) {
2815                         if (!st->frag_data)
2816                                 st->frag_data = kmap_atomic(skb_frag_page(frag));
2817 
2818                         *data = (u8 *) st->frag_data + frag->page_offset +
2819                                 (abs_offset - st->stepped_offset);
2820 
2821                         return block_limit - abs_offset;
2822                 }
2823 
2824                 if (st->frag_data) {
2825                         kunmap_atomic(st->frag_data);
2826                         st->frag_data = NULL;
2827                 }
2828 
2829                 st->frag_idx++;
2830                 st->stepped_offset += skb_frag_size(frag);
2831         }
2832 
2833         if (st->frag_data) {
2834                 kunmap_atomic(st->frag_data);
2835                 st->frag_data = NULL;
2836         }
2837 
2838         if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2839                 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2840                 st->frag_idx = 0;
2841                 goto next_skb;
2842         } else if (st->cur_skb->next) {
2843                 st->cur_skb = st->cur_skb->next;
2844                 st->frag_idx = 0;
2845                 goto next_skb;
2846         }
2847 
2848         return 0;
2849 }
2850 EXPORT_SYMBOL(skb_seq_read);
2851 
2852 /**
2853  * skb_abort_seq_read - Abort a sequential read of skb data
2854  * @st: state variable
2855  *
2856  * Must be called if skb_seq_read() was not called until it
2857  * returned 0.
2858  */
2859 void skb_abort_seq_read(struct skb_seq_state *st)
2860 {
2861         if (st->frag_data)
2862                 kunmap_atomic(st->frag_data);
2863 }
2864 EXPORT_SYMBOL(skb_abort_seq_read);
2865 
2866 #define TS_SKB_CB(state)        ((struct skb_seq_state *) &((state)->cb))
2867 
2868 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2869                                           struct ts_config *conf,
2870                                           struct ts_state *state)
2871 {
2872         return skb_seq_read(offset, text, TS_SKB_CB(state));
2873 }
2874 
2875 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2876 {
2877         skb_abort_seq_read(TS_SKB_CB(state));
2878 }
2879 
2880 /**
2881  * skb_find_text - Find a text pattern in skb data
2882  * @skb: the buffer to look in
2883  * @from: search offset
2884  * @to: search limit
2885  * @config: textsearch configuration
2886  *
2887  * Finds a pattern in the skb data according to the specified
2888  * textsearch configuration. Use textsearch_next() to retrieve
2889  * subsequent occurrences of the pattern. Returns the offset
2890  * to the first occurrence or UINT_MAX if no match was found.
2891  */
2892 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2893                            unsigned int to, struct ts_config *config)
2894 {
2895         struct ts_state state;
2896         unsigned int ret;
2897 
2898         config->get_next_block = skb_ts_get_next_block;
2899         config->finish = skb_ts_finish;
2900 
2901         skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
2902 
2903         ret = textsearch_find(config, &state);
2904         return (ret <= to - from ? ret : UINT_MAX);
2905 }
2906 EXPORT_SYMBOL(skb_find_text);
2907 
2908 /**
2909  * skb_append_datato_frags - append the user data to a skb
2910  * @sk: sock  structure
2911  * @skb: skb structure to be appended with user data.
2912  * @getfrag: call back function to be used for getting the user data
2913  * @from: pointer to user message iov
2914  * @length: length of the iov message
2915  *
2916  * Description: This procedure append the user data in the fragment part
2917  * of the skb if any page alloc fails user this procedure returns  -ENOMEM
2918  */
2919 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2920                         int (*getfrag)(void *from, char *to, int offset,
2921                                         int len, int odd, struct sk_buff *skb),
2922                         void *from, int length)
2923 {
2924         int frg_cnt = skb_shinfo(skb)->nr_frags;
2925         int copy;
2926         int offset = 0;
2927         int ret;
2928         struct page_frag *pfrag = &current->task_frag;
2929 
2930         do {
2931                 /* Return error if we don't have space for new frag */
2932                 if (frg_cnt >= MAX_SKB_FRAGS)
2933                         return -EMSGSIZE;
2934 
2935                 if (!sk_page_frag_refill(sk, pfrag))
2936                         return -ENOMEM;
2937 
2938                 /* copy the user data to page */
2939                 copy = min_t(int, length, pfrag->size - pfrag->offset);
2940 
2941                 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
2942                               offset, copy, 0, skb);
2943                 if (ret < 0)
2944                         return -EFAULT;
2945 
2946                 /* copy was successful so update the size parameters */
2947                 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
2948                                    copy);
2949                 frg_cnt++;
2950                 pfrag->offset += copy;
2951                 get_page(pfrag->page);
2952 
2953                 skb->truesize += copy;
2954                 atomic_add(copy, &sk->sk_wmem_alloc);
2955                 skb->len += copy;
2956                 skb->data_len += copy;
2957                 offset += copy;
2958                 length -= copy;
2959 
2960         } while (length > 0);
2961 
2962         return 0;
2963 }
2964 EXPORT_SYMBOL(skb_append_datato_frags);
2965 
2966 /**
2967  *      skb_pull_rcsum - pull skb and update receive checksum
2968  *      @skb: buffer to update
2969  *      @len: length of data pulled
2970  *
2971  *      This function performs an skb_pull on the packet and updates
2972  *      the CHECKSUM_COMPLETE checksum.  It should be used on
2973  *      receive path processing instead of skb_pull unless you know
2974  *      that the checksum difference is zero (e.g., a valid IP header)
2975  *      or you are setting ip_summed to CHECKSUM_NONE.
2976  */
2977 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2978 {
2979         BUG_ON(len > skb->len);
2980         skb->len -= len;
2981         BUG_ON(skb->len < skb->data_len);
2982         skb_postpull_rcsum(skb, skb->data, len);
2983         return skb->data += len;
2984 }
2985 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2986 
2987 /**
2988  *      skb_segment - Perform protocol segmentation on skb.
2989  *      @head_skb: buffer to segment
2990  *      @features: features for the output path (see dev->features)
2991  *
2992  *      This function performs segmentation on the given skb.  It returns
2993  *      a pointer to the first in a list of new skbs for the segments.
2994  *      In case of error it returns ERR_PTR(err).
2995  */
2996 struct sk_buff *skb_segment(struct sk_buff *head_skb,
2997                             netdev_features_t features)
2998 {
2999         struct sk_buff *segs = NULL;
3000         struct sk_buff *tail = NULL;
3001         struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3002         skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3003         unsigned int mss = skb_shinfo(head_skb)->gso_size;
3004         unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3005         struct sk_buff *frag_skb = head_skb;
3006         unsigned int offset = doffset;
3007         unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3008         unsigned int headroom;
3009         unsigned int len;
3010         __be16 proto;
3011         bool csum;
3012         int sg = !!(features & NETIF_F_SG);
3013         int nfrags = skb_shinfo(head_skb)->nr_frags;
3014         int err = -ENOMEM;
3015         int i = 0;
3016         int pos;
3017         int dummy;
3018 
3019         __skb_push(head_skb, doffset);
3020         proto = skb_network_protocol(head_skb, &dummy);
3021         if (unlikely(!proto))
3022                 return ERR_PTR(-EINVAL);
3023 
3024         csum = !head_skb->encap_hdr_csum &&
3025             !!can_checksum_protocol(features, proto);
3026 
3027         headroom = skb_headroom(head_skb);
3028         pos = skb_headlen(head_skb);
3029 
3030         do {
3031                 struct sk_buff *nskb;
3032                 skb_frag_t *nskb_frag;
3033                 int hsize;
3034                 int size;
3035 
3036                 len = head_skb->len - offset;
3037                 if (len > mss)
3038                         len = mss;
3039 
3040                 hsize = skb_headlen(head_skb) - offset;
3041                 if (hsize < 0)
3042                         hsize = 0;
3043                 if (hsize > len || !sg)
3044                         hsize = len;
3045 
3046                 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3047                     (skb_headlen(list_skb) == len || sg)) {
3048                         BUG_ON(skb_headlen(list_skb) > len);
3049 
3050                         i = 0;
3051                         nfrags = skb_shinfo(list_skb)->nr_frags;
3052                         frag = skb_shinfo(list_skb)->frags;
3053                         frag_skb = list_skb;
3054                         pos += skb_headlen(list_skb);
3055 
3056                         while (pos < offset + len) {
3057                                 BUG_ON(i >= nfrags);
3058 
3059                                 size = skb_frag_size(frag);
3060                                 if (pos + size > offset + len)
3061                                         break;
3062 
3063                                 i++;
3064                                 pos += size;
3065                                 frag++;
3066                         }
3067 
3068                         nskb = skb_clone(list_skb, GFP_ATOMIC);
3069                         list_skb = list_skb->next;
3070 
3071                         if (unlikely(!nskb))
3072                                 goto err;
3073 
3074                         if (unlikely(pskb_trim(nskb, len))) {
3075                                 kfree_skb(nskb);
3076                                 goto err;
3077                         }
3078 
3079                         hsize = skb_end_offset(nskb);
3080                         if (skb_cow_head(nskb, doffset + headroom)) {
3081                                 kfree_skb(nskb);
3082                                 goto err;
3083                         }
3084 
3085                         nskb->truesize += skb_end_offset(nskb) - hsize;
3086                         skb_release_head_state(nskb);
3087                         __skb_push(nskb, doffset);
3088                 } else {
3089                         nskb = __alloc_skb(hsize + doffset + headroom,
3090                                            GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3091                                            NUMA_NO_NODE);
3092 
3093                         if (unlikely(!nskb))
3094                                 goto err;
3095 
3096                         skb_reserve(nskb, headroom);
3097                         __skb_put(nskb, doffset);
3098                 }
3099 
3100                 if (segs)
3101                         tail->next = nskb;
3102                 else
3103                         segs = nskb;
3104                 tail = nskb;
3105 
3106                 __copy_skb_header(nskb, head_skb);
3107 
3108                 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3109                 skb_reset_mac_len(nskb);
3110 
3111                 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3112                                                  nskb->data - tnl_hlen,
3113                                                  doffset + tnl_hlen);
3114 
3115                 if (nskb->len == len + doffset)
3116                         goto perform_csum_check;
3117 
3118                 if (!sg && !nskb->remcsum_offload) {
3119                         nskb->ip_summed = CHECKSUM_NONE;
3120                         nskb->csum = skb_copy_and_csum_bits(head_skb, offset,
3121                                                             skb_put(nskb, len),
3122                                                             len, 0);
3123                         SKB_GSO_CB(nskb)->csum_start =
3124                             skb_headroom(nskb) + doffset;
3125                         continue;
3126                 }
3127 
3128                 nskb_frag = skb_shinfo(nskb)->frags;
3129 
3130                 skb_copy_from_linear_data_offset(head_skb, offset,
3131                                                  skb_put(nskb, hsize), hsize);
3132 
3133                 skb_shinfo(nskb)->tx_flags = skb_shinfo(head_skb)->tx_flags &
3134                         SKBTX_SHARED_FRAG;
3135 
3136                 while (pos < offset + len) {
3137                         if (i >= nfrags) {
3138                                 BUG_ON(skb_headlen(list_skb));
3139 
3140                                 i = 0;
3141                                 nfrags = skb_shinfo(list_skb)->nr_frags;
3142                                 frag = skb_shinfo(list_skb)->frags;
3143                                 frag_skb = list_skb;
3144 
3145                                 BUG_ON(!nfrags);
3146 
3147                                 list_skb = list_skb->next;
3148                         }
3149 
3150                         if (unlikely(skb_shinfo(nskb)->nr_frags >=
3151                                      MAX_SKB_FRAGS)) {
3152                                 net_warn_ratelimited(
3153                                         "skb_segment: too many frags: %u %u\n",
3154                                         pos, mss);
3155                                 goto err;
3156                         }
3157 
3158                         if (unlikely(skb_orphan_frags(frag_skb, GFP_ATOMIC)))
3159                                 goto err;
3160 
3161                         *nskb_frag = *frag;
3162                         __skb_frag_ref(nskb_frag);
3163                         size = skb_frag_size(nskb_frag);
3164 
3165                         if (pos < offset) {
3166                                 nskb_frag->page_offset += offset - pos;
3167                                 skb_frag_size_sub(nskb_frag, offset - pos);
3168                         }
3169 
3170                         skb_shinfo(nskb)->nr_frags++;
3171 
3172                         if (pos + size <= offset + len) {
3173                                 i++;
3174                                 frag++;
3175                                 pos += size;
3176                         } else {
3177                                 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
3178                                 goto skip_fraglist;
3179                         }
3180 
3181                         nskb_frag++;
3182                 }
3183 
3184 skip_fraglist:
3185                 nskb->data_len = len - hsize;
3186                 nskb->len += nskb->data_len;
3187                 nskb->truesize += nskb->data_len;
3188 
3189 perform_csum_check:
3190                 if (!csum && !nskb->remcsum_offload) {
3191                         nskb->csum = skb_checksum(nskb, doffset,
3192                                                   nskb->len - doffset, 0);
3193                         nskb->ip_summed = CHECKSUM_NONE;
3194                         SKB_GSO_CB(nskb)->csum_start =
3195                             skb_headroom(nskb) + doffset;
3196                 }
3197         } while ((offset += len) < head_skb->len);
3198 
3199         /* Some callers want to get the end of the list.
3200          * Put it in segs->prev to avoid walking the list.
3201          * (see validate_xmit_skb_list() for example)
3202          */
3203         segs->prev = tail;
3204 
3205         /* Following permits correct backpressure, for protocols
3206          * using skb_set_owner_w().
3207          * Idea is to tranfert ownership from head_skb to last segment.
3208          */
3209         if (head_skb->destructor == sock_wfree) {
3210                 swap(tail->truesize, head_skb->truesize);
3211                 swap(tail->destructor, head_skb->destructor);
3212                 swap(tail->sk, head_skb->sk);
3213         }
3214         return segs;
3215 
3216 err:
3217         kfree_skb_list(segs);
3218         return ERR_PTR(err);
3219 }
3220 EXPORT_SYMBOL_GPL(skb_segment);
3221 
3222 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
3223 {
3224         struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
3225         unsigned int offset = skb_gro_offset(skb);
3226         unsigned int headlen = skb_headlen(skb);
3227         unsigned int len = skb_gro_len(skb);
3228         struct sk_buff *lp, *p = *head;
3229         unsigned int delta_truesize;
3230 
3231         if (unlikely(p->len + len >= 65536))
3232                 return -E2BIG;
3233 
3234         lp = NAPI_GRO_CB(p)->last;
3235         pinfo = skb_shinfo(lp);
3236 
3237         if (headlen <= offset) {
3238                 skb_frag_t *frag;
3239                 skb_frag_t *frag2;
3240                 int i = skbinfo->nr_frags;
3241                 int nr_frags = pinfo->nr_frags + i;
3242 
3243                 if (nr_frags > MAX_SKB_FRAGS)
3244                         goto merge;
3245 
3246                 offset -= headlen;
3247                 pinfo->nr_frags = nr_frags;
3248                 skbinfo->nr_frags = 0;
3249 
3250                 frag = pinfo->frags + nr_frags;
3251                 frag2 = skbinfo->frags + i;
3252                 do {
3253                         *--frag = *--frag2;
3254                 } while (--i);
3255 
3256                 frag->page_offset += offset;
3257                 skb_frag_size_sub(frag, offset);
3258 
3259                 /* all fragments truesize : remove (head size + sk_buff) */
3260                 delta_truesize = skb->truesize -
3261                                  SKB_TRUESIZE(skb_end_offset(skb));
3262 
3263                 skb->truesize -= skb->data_len;
3264                 skb->len -= skb->data_len;
3265                 skb->data_len = 0;
3266 
3267                 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
3268                 goto done;
3269         } else if (skb->head_frag) {
3270                 int nr_frags = pinfo->nr_frags;
3271                 skb_frag_t *frag = pinfo->frags + nr_frags;
3272                 struct page *page = virt_to_head_page(skb->head);
3273                 unsigned int first_size = headlen - offset;
3274                 unsigned int first_offset;
3275 
3276                 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
3277                         goto merge;
3278 
3279                 first_offset = skb->data -
3280                                (unsigned char *)page_address(page) +
3281                                offset;
3282 
3283                 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3284 
3285                 frag->page.p      = page;
3286                 frag->page_offset = first_offset;
3287                 skb_frag_size_set(frag, first_size);
3288 
3289                 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3290                 /* We dont need to clear skbinfo->nr_frags here */
3291 
3292                 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3293                 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3294                 goto done;
3295         }
3296 
3297 merge:
3298         delta_truesize = skb->truesize;
3299         if (offset > headlen) {
3300                 unsigned int eat = offset - headlen;
3301 
3302                 skbinfo->frags[0].page_offset += eat;
3303                 skb_frag_size_sub(&skbinfo->frags[0], eat);
3304                 skb->data_len -= eat;
3305                 skb->len -= eat;
3306                 offset = headlen;
3307         }
3308 
3309         __skb_pull(skb, offset);
3310 
3311         if (NAPI_GRO_CB(p)->last == p)
3312                 skb_shinfo(p)->frag_list = skb;
3313         else
3314                 NAPI_GRO_CB(p)->last->next = skb;
3315         NAPI_GRO_CB(p)->last = skb;
3316         __skb_header_release(skb);
3317         lp = p;
3318 
3319 done:
3320         NAPI_GRO_CB(p)->count++;
3321         p->data_len += len;
3322         p->truesize += delta_truesize;
3323         p->len += len;
3324         if (lp != p) {
3325                 lp->data_len += len;
3326                 lp->truesize += delta_truesize;
3327                 lp->len += len;
3328         }
3329         NAPI_GRO_CB(skb)->same_flow = 1;
3330         return 0;
3331 }
3332 
3333 void __init skb_init(void)
3334 {
3335         skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3336                                               sizeof(struct sk_buff),
3337                                               0,
3338                                               SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3339                                               NULL);
3340         skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3341                                                 sizeof(struct sk_buff_fclones),
3342                                                 0,
3343                                                 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3344                                                 NULL);
3345 }
3346 
3347 /**
3348  *      skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3349  *      @skb: Socket buffer containing the buffers to be mapped
3350  *      @sg: The scatter-gather list to map into
3351  *      @offset: The offset into the buffer's contents to start mapping
3352  *      @len: Length of buffer space to be mapped
3353  *
3354  *      Fill the specified scatter-gather list with mappings/pointers into a
3355  *      region of the buffer space attached to a socket buffer.
3356  */
3357 static int
3358 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3359 {
3360         int start = skb_headlen(skb);
3361         int i, copy = start - offset;
3362         struct sk_buff *frag_iter;
3363         int elt = 0;
3364 
3365         if (copy > 0) {
3366                 if (copy > len)
3367                         copy = len;
3368                 sg_set_buf(sg, skb->data + offset, copy);
3369                 elt++;
3370                 if ((len -= copy) == 0)
3371                         return elt;
3372                 offset += copy;
3373         }
3374 
3375         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3376                 int end;
3377 
3378                 WARN_ON(start > offset + len);
3379 
3380                 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3381                 if ((copy = end - offset) > 0) {
3382                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3383 
3384                         if (copy > len)
3385                                 copy = len;
3386                         sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3387                                         frag->page_offset+offset-start);
3388                         elt++;
3389                         if (!(len -= copy))
3390                                 return elt;
3391                         offset += copy;
3392                 }
3393                 start = end;
3394         }
3395 
3396         skb_walk_frags(skb, frag_iter) {
3397                 int end;
3398 
3399                 WARN_ON(start > offset + len);
3400 
3401                 end = start + frag_iter->len;
3402                 if ((copy = end - offset) > 0) {
3403                         if (copy > len)
3404                                 copy = len;
3405                         elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3406                                               copy);
3407                         if ((len -= copy) == 0)
3408                                 return elt;
3409                         offset += copy;
3410                 }
3411                 start = end;
3412         }
3413         BUG_ON(len);
3414         return elt;
3415 }
3416 
3417 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
3418  * sglist without mark the sg which contain last skb data as the end.
3419  * So the caller can mannipulate sg list as will when padding new data after
3420  * the first call without calling sg_unmark_end to expend sg list.
3421  *
3422  * Scenario to use skb_to_sgvec_nomark:
3423  * 1. sg_init_table
3424  * 2. skb_to_sgvec_nomark(payload1)
3425  * 3. skb_to_sgvec_nomark(payload2)
3426  *
3427  * This is equivalent to:
3428  * 1. sg_init_table
3429  * 2. skb_to_sgvec(payload1)
3430  * 3. sg_unmark_end
3431  * 4. skb_to_sgvec(payload2)
3432  *
3433  * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
3434  * is more preferable.
3435  */
3436 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
3437                         int offset, int len)
3438 {
3439         return __skb_to_sgvec(skb, sg, offset, len);
3440 }
3441 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
3442 
3443 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3444 {
3445         int nsg = __skb_to_sgvec(skb, sg, offset, len);
3446 
3447         sg_mark_end(&sg[nsg - 1]);
3448 
3449         return nsg;
3450 }
3451 EXPORT_SYMBOL_GPL(skb_to_sgvec);
3452 
3453 /**
3454  *      skb_cow_data - Check that a socket buffer's data buffers are writable
3455  *      @skb: The socket buffer to check.
3456  *      @tailbits: Amount of trailing space to be added
3457  *      @trailer: Returned pointer to the skb where the @tailbits space begins
3458  *
3459  *      Make sure that the data buffers attached to a socket buffer are
3460  *      writable. If they are not, private copies are made of the data buffers
3461  *      and the socket buffer is set to use these instead.
3462  *
3463  *      If @tailbits is given, make sure that there is space to write @tailbits
3464  *      bytes of data beyond current end of socket buffer.  @trailer will be
3465  *      set to point to the skb in which this space begins.
3466  *
3467  *      The number of scatterlist elements required to completely map the
3468  *      COW'd and extended socket buffer will be returned.
3469  */
3470 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3471 {
3472         int copyflag;
3473         int elt;
3474         struct sk_buff *skb1, **skb_p;
3475 
3476         /* If skb is cloned or its head is paged, reallocate
3477          * head pulling out all the pages (pages are considered not writable
3478          * at the moment even if they are anonymous).
3479          */
3480         if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3481             __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3482                 return -ENOMEM;
3483 
3484         /* Easy case. Most of packets will go this way. */
3485         if (!skb_has_frag_list(skb)) {
3486                 /* A little of trouble, not enough of space for trailer.
3487                  * This should not happen, when stack is tuned to generate
3488                  * good frames. OK, on miss we reallocate and reserve even more
3489                  * space, 128 bytes is fair. */
3490 
3491                 if (skb_tailroom(skb) < tailbits &&
3492                     pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3493                         return -ENOMEM;
3494 
3495                 /* Voila! */
3496                 *trailer = skb;
3497                 return 1;
3498         }
3499 
3500         /* Misery. We are in troubles, going to mincer fragments... */
3501 
3502         elt = 1;
3503         skb_p = &skb_shinfo(skb)->frag_list;
3504         copyflag = 0;
3505 
3506         while ((skb1 = *skb_p) != NULL) {
3507                 int ntail = 0;
3508 
3509                 /* The fragment is partially pulled by someone,
3510                  * this can happen on input. Copy it and everything
3511                  * after it. */
3512 
3513                 if (skb_shared(skb1))
3514                         copyflag = 1;
3515 
3516                 /* If the skb is the last, worry about trailer. */
3517 
3518                 if (skb1->next == NULL && tailbits) {
3519                         if (skb_shinfo(skb1)->nr_frags ||
3520                             skb_has_frag_list(skb1) ||
3521                             skb_tailroom(skb1) < tailbits)
3522                                 ntail = tailbits + 128;
3523                 }
3524 
3525                 if (copyflag ||
3526                     skb_cloned(skb1) ||
3527                     ntail ||
3528                     skb_shinfo(skb1)->nr_frags ||
3529                     skb_has_frag_list(skb1)) {
3530                         struct sk_buff *skb2;
3531 
3532                         /* Fuck, we are miserable poor guys... */
3533                         if (ntail == 0)
3534                                 skb2 = skb_copy(skb1, GFP_ATOMIC);
3535                         else
3536                                 skb2 = skb_copy_expand(skb1,
3537                                                        skb_headroom(skb1),
3538                                                        ntail,
3539                                                        GFP_ATOMIC);
3540                         if (unlikely(skb2 == NULL))
3541                                 return -ENOMEM;
3542 
3543                         if (skb1->sk)
3544                                 skb_set_owner_w(skb2, skb1->sk);
3545 
3546                         /* Looking around. Are we still alive?
3547                          * OK, link new skb, drop old one */
3548 
3549                         skb2->next = skb1->next;
3550                         *skb_p = skb2;
3551                         kfree_skb(skb1);
3552                         skb1 = skb2;
3553                 }
3554                 elt++;
3555                 *trailer = skb1;
3556                 skb_p = &skb1->next;
3557         }
3558 
3559         return elt;
3560 }
3561 EXPORT_SYMBOL_GPL(skb_cow_data);
3562 
3563 static void sock_rmem_free(struct sk_buff *skb)
3564 {
3565         struct sock *sk = skb->sk;
3566 
3567         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3568 }
3569 
3570 /*
3571  * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3572  */
3573 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3574 {
3575         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3576             (unsigned int)sk->sk_rcvbuf)
3577                 return -ENOMEM;
3578 
3579         skb_orphan(skb);
3580         skb->sk = sk;
3581         skb->destructor = sock_rmem_free;
3582         atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3583 
3584         /* before exiting rcu section, make sure dst is refcounted */
3585         skb_dst_force(skb);
3586 
3587         skb_queue_tail(&sk->sk_error_queue, skb);
3588         if (!sock_flag(sk, SOCK_DEAD))
3589                 sk->sk_data_ready(sk);
3590         return 0;
3591 }
3592 EXPORT_SYMBOL(sock_queue_err_skb);
3593 
3594 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
3595 {
3596         struct sk_buff_head *q = &sk->sk_error_queue;
3597         struct sk_buff *skb, *skb_next;
3598         unsigned long flags;
3599         int err = 0;
3600 
3601         spin_lock_irqsave(&q->lock, flags);
3602         skb = __skb_dequeue(q);
3603         if (skb && (skb_next = skb_peek(q)))
3604                 err = SKB_EXT_ERR(skb_next)->ee.ee_errno;
3605         spin_unlock_irqrestore(&q->lock, flags);
3606 
3607         sk->sk_err = err;
3608         if (err)
3609                 sk->sk_error_report(sk);
3610 
3611         return skb;
3612 }
3613 EXPORT_SYMBOL(sock_dequeue_err_skb);
3614 
3615 /**
3616  * skb_clone_sk - create clone of skb, and take reference to socket
3617  * @skb: the skb to clone
3618  *
3619  * This function creates a clone of a buffer that holds a reference on
3620  * sk_refcnt.  Buffers created via this function are meant to be
3621  * returned using sock_queue_err_skb, or free via kfree_skb.
3622  *
3623  * When passing buffers allocated with this function to sock_queue_err_skb
3624  * it is necessary to wrap the call with sock_hold/sock_put in order to
3625  * prevent the socket from being released prior to being enqueued on
3626  * the sk_error_queue.
3627  */
3628 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
3629 {
3630         struct sock *sk = skb->sk;
3631         struct sk_buff *clone;
3632 
3633         if (!sk || !atomic_inc_not_zero(&sk->sk_refcnt))
3634                 return NULL;
3635 
3636         clone = skb_clone(skb, GFP_ATOMIC);
3637         if (!clone) {
3638                 sock_put(sk);
3639                 return NULL;
3640         }
3641 
3642         clone->sk = sk;
3643         clone->destructor = sock_efree;
3644 
3645         return clone;
3646 }
3647 EXPORT_SYMBOL(skb_clone_sk);
3648 
3649 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
3650                                         struct sock *sk,
3651                                         int tstype)
3652 {
3653         struct sock_exterr_skb *serr;
3654         int err;
3655 
3656         serr = SKB_EXT_ERR(skb);
3657         memset(serr, 0, sizeof(*serr));
3658         serr->ee.ee_errno = ENOMSG;
3659         serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3660         serr->ee.ee_info = tstype;
3661         if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
3662                 serr->ee.ee_data = skb_shinfo(skb)->tskey;
3663                 if (sk->sk_protocol == IPPROTO_TCP)
3664                         serr->ee.ee_data -= sk->sk_tskey;
3665         }
3666 
3667         err = sock_queue_err_skb(sk, skb);
3668 
3669         if (err)
3670                 kfree_skb(skb);
3671 }
3672 
3673 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
3674 {
3675         bool ret;
3676 
3677         if (likely(sysctl_tstamp_allow_data || tsonly))
3678                 return true;
3679 
3680         read_lock_bh(&sk->sk_callback_lock);
3681         ret = sk->sk_socket && sk->sk_socket->file &&
3682               file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
3683         read_unlock_bh(&sk->sk_callback_lock);
3684         return ret;
3685 }
3686 
3687 void skb_complete_tx_timestamp(struct sk_buff *skb,
3688                                struct skb_shared_hwtstamps *hwtstamps)
3689 {
3690         struct sock *sk = skb->sk;
3691 
3692         if (!skb_may_tx_timestamp(sk, false))
3693                 return;
3694 
3695         /* take a reference to prevent skb_orphan() from freeing the socket */
3696         sock_hold(sk);
3697 
3698         *skb_hwtstamps(skb) = *hwtstamps;
3699         __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND);
3700 
3701         sock_put(sk);
3702 }
3703 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
3704 
3705 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3706                      struct skb_shared_hwtstamps *hwtstamps,
3707                      struct sock *sk, int tstype)
3708 {
3709         struct sk_buff *skb;
3710         bool tsonly;
3711 
3712         if (!sk)
3713                 return;
3714 
3715         tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
3716         if (!skb_may_tx_timestamp(sk, tsonly))
3717                 return;
3718 
3719         if (tsonly)
3720                 skb = alloc_skb(0, GFP_ATOMIC);
3721         else
3722                 skb = skb_clone(orig_skb, GFP_ATOMIC);
3723         if (!skb)
3724                 return;
3725 
3726         if (tsonly) {
3727                 skb_shinfo(skb)->tx_flags = skb_shinfo(orig_skb)->tx_flags;
3728                 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
3729         }
3730 
3731         if (hwtstamps)
3732                 *skb_hwtstamps(skb) = *hwtstamps;
3733         else
3734                 skb->tstamp = ktime_get_real();
3735 
3736         __skb_complete_tx_timestamp(skb, sk, tstype);
3737 }
3738 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
3739 
3740 void skb_tstamp_tx(struct sk_buff *orig_skb,
3741                    struct skb_shared_hwtstamps *hwtstamps)
3742 {
3743         return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
3744                                SCM_TSTAMP_SND);
3745 }
3746 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3747 
3748 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3749 {
3750         struct sock *sk = skb->sk;
3751         struct sock_exterr_skb *serr;
3752         int err;
3753 
3754         skb->wifi_acked_valid = 1;
3755         skb->wifi_acked = acked;
3756 
3757         serr = SKB_EXT_ERR(skb);
3758         memset(serr, 0, sizeof(*serr));
3759         serr->ee.ee_errno = ENOMSG;
3760         serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3761 
3762         /* take a reference to prevent skb_orphan() from freeing the socket */
3763         sock_hold(sk);
3764 
3765         err = sock_queue_err_skb(sk, skb);
3766         if (err)
3767                 kfree_skb(skb);
3768 
3769         sock_put(sk);
3770 }
3771 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3772 
3773 /**
3774  * skb_partial_csum_set - set up and verify partial csum values for packet
3775  * @skb: the skb to set
3776  * @start: the number of bytes after skb->data to start checksumming.
3777  * @off: the offset from start to place the checksum.
3778  *
3779  * For untrusted partially-checksummed packets, we need to make sure the values
3780  * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3781  *
3782  * This function checks and sets those values and skb->ip_summed: if this
3783  * returns false you should drop the packet.
3784  */
3785 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3786 {
3787         if (unlikely(start > skb_headlen(skb)) ||
3788             unlikely((int)start + off > skb_headlen(skb) - 2)) {
3789                 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3790                                      start, off, skb_headlen(skb));
3791                 return false;
3792         }
3793         skb->ip_summed = CHECKSUM_PARTIAL;
3794         skb->csum_start = skb_headroom(skb) + start;
3795         skb->csum_offset = off;
3796         skb_set_transport_header(skb, start);
3797         return true;
3798 }
3799 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3800 
3801 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
3802                                unsigned int max)
3803 {
3804         if (skb_headlen(skb) >= len)
3805                 return 0;
3806 
3807         /* If we need to pullup then pullup to the max, so we
3808          * won't need to do it again.
3809          */
3810         if (max > skb->len)
3811                 max = skb->len;
3812 
3813         if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
3814                 return -ENOMEM;
3815 
3816         if (skb_headlen(skb) < len)
3817                 return -EPROTO;
3818 
3819         return 0;
3820 }
3821 
3822 #define MAX_TCP_HDR_LEN (15 * 4)
3823 
3824 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
3825                                       typeof(IPPROTO_IP) proto,
3826                                       unsigned int off)
3827 {
3828         switch (proto) {
3829                 int err;
3830 
3831         case IPPROTO_TCP:
3832                 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
3833                                           off + MAX_TCP_HDR_LEN);
3834                 if (!err && !skb_partial_csum_set(skb, off,
3835                                                   offsetof(struct tcphdr,
3836                                                            check)))
3837                         err = -EPROTO;
3838                 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
3839 
3840         case IPPROTO_UDP:
3841                 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
3842                                           off + sizeof(struct udphdr));
3843                 if (!err && !skb_partial_csum_set(skb, off,
3844                                                   offsetof(struct udphdr,
3845                                                            check)))
3846                         err = -EPROTO;
3847                 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
3848         }
3849 
3850         return ERR_PTR(-EPROTO);
3851 }
3852 
3853 /* This value should be large enough to cover a tagged ethernet header plus
3854  * maximally sized IP and TCP or UDP headers.
3855  */
3856 #define MAX_IP_HDR_LEN 128
3857 
3858 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
3859 {
3860         unsigned int off;
3861         bool fragment;
3862         __sum16 *csum;
3863         int err;
3864 
3865         fragment = false;
3866 
3867         err = skb_maybe_pull_tail(skb,
3868                                   sizeof(struct iphdr),
3869                                   MAX_IP_HDR_LEN);
3870         if (err < 0)
3871                 goto out;
3872 
3873         if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
3874                 fragment = true;
3875 
3876         off = ip_hdrlen(skb);
3877 
3878         err = -EPROTO;
3879 
3880         if (fragment)
3881                 goto out;
3882 
3883         csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
3884         if (IS_ERR(csum))
3885                 return PTR_ERR(csum);
3886 
3887         if (recalculate)
3888                 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3889                                            ip_hdr(skb)->daddr,
3890                                            skb->len - off,
3891                                            ip_hdr(skb)->protocol, 0);
3892         err = 0;
3893 
3894 out:
3895         return err;
3896 }
3897 
3898 /* This value should be large enough to cover a tagged ethernet header plus
3899  * an IPv6 header, all options, and a maximal TCP or UDP header.
3900  */
3901 #define MAX_IPV6_HDR_LEN 256
3902 
3903 #define OPT_HDR(type, skb, off) \
3904         (type *)(skb_network_header(skb) + (off))
3905 
3906 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
3907 {
3908         int err;
3909         u8 nexthdr;
3910         unsigned int off;
3911         unsigned int len;
3912         bool fragment;
3913         bool done;
3914         __sum16 *csum;
3915 
3916         fragment = false;
3917         done = false;
3918 
3919         off = sizeof(struct ipv6hdr);
3920 
3921         err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
3922         if (err < 0)
3923                 goto out;
3924 
3925         nexthdr = ipv6_hdr(skb)->nexthdr;
3926 
3927         len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
3928         while (off <= len && !done) {
3929                 switch (nexthdr) {
3930                 case IPPROTO_DSTOPTS:
3931                 case IPPROTO_HOPOPTS:
3932                 case IPPROTO_ROUTING: {
3933                         struct ipv6_opt_hdr *hp;
3934 
3935                         err = skb_maybe_pull_tail(skb,
3936                                                   off +
3937                                                   sizeof(struct ipv6_opt_hdr),
3938                                                   MAX_IPV6_HDR_LEN);
3939                         if (err < 0)
3940                                 goto out;
3941 
3942                         hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
3943                         nexthdr = hp->nexthdr;
3944                         off += ipv6_optlen(hp);
3945                         break;
3946                 }
3947                 case IPPROTO_AH: {
3948                         struct ip_auth_hdr *hp;
3949 
3950                         err = skb_maybe_pull_tail(skb,
3951                                                   off +
3952                                                   sizeof(struct ip_auth_hdr),
3953                                                   MAX_IPV6_HDR_LEN);
3954                         if (err < 0)
3955                                 goto out;
3956 
3957                         hp = OPT_HDR(struct ip_auth_hdr, skb, off);
3958                         nexthdr = hp->nexthdr;
3959                         off += ipv6_authlen(hp);
3960                         break;
3961                 }
3962                 case IPPROTO_FRAGMENT: {
3963                         struct frag_hdr *hp;
3964 
3965                         err = skb_maybe_pull_tail(skb,
3966                                                   off +
3967                                                   sizeof(struct frag_hdr),
3968                                                   MAX_IPV6_HDR_LEN);
3969                         if (err < 0)
3970                                 goto out;
3971 
3972                         hp = OPT_HDR(struct frag_hdr, skb, off);
3973 
3974                         if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
3975                                 fragment = true;
3976 
3977                         nexthdr = hp->nexthdr;
3978                         off += sizeof(struct frag_hdr);
3979                         break;
3980                 }
3981                 default:
3982                         done = true;
3983                         break;
3984                 }
3985         }
3986 
3987         err = -EPROTO;
3988 
3989         if (!done || fragment)
3990                 goto out;
3991 
3992         csum = skb_checksum_setup_ip(skb, nexthdr, off);
3993         if (IS_ERR(csum))
3994                 return PTR_ERR(csum);
3995 
3996         if (recalculate)
3997                 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3998                                          &ipv6_hdr(skb)->daddr,
3999                                          skb->len - off, nexthdr, 0);
4000         err = 0;
4001 
4002 out:
4003         return err;
4004 }
4005 
4006 /**
4007  * skb_checksum_setup - set up partial checksum offset
4008  * @skb: the skb to set up
4009  * @recalculate: if true the pseudo-header checksum will be recalculated
4010  */
4011 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
4012 {
4013         int err;
4014 
4015         switch (skb->protocol) {
4016         case htons(ETH_P_IP):
4017                 err = skb_checksum_setup_ipv4(skb, recalculate);
4018                 break;
4019 
4020         case htons(ETH_P_IPV6):
4021                 err = skb_checksum_setup_ipv6(skb, recalculate);
4022                 break;
4023 
4024         default:
4025                 err = -EPROTO;
4026                 break;
4027         }
4028 
4029         return err;
4030 }
4031 EXPORT_SYMBOL(skb_checksum_setup);
4032 
4033 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
4034 {
4035         net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
4036                              skb->dev->name);
4037 }
4038 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
4039 
4040 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
4041 {
4042         if (head_stolen) {
4043                 skb_release_head_state(skb);
4044                 kmem_cache_free(skbuff_head_cache, skb);
4045         } else {
4046                 __kfree_skb(skb);
4047         }
4048 }
4049 EXPORT_SYMBOL(kfree_skb_partial);
4050 
4051 /**
4052  * skb_try_coalesce - try to merge skb to prior one
4053  * @to: prior buffer
4054  * @from: buffer to add
4055  * @fragstolen: pointer to boolean
4056  * @delta_truesize: how much more was allocated than was requested
4057  */
4058 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
4059                       bool *fragstolen, int *delta_truesize)
4060 {
4061         int i, delta, len = from->len;
4062 
4063         *fragstolen = false;
4064 
4065         if (skb_cloned(to))
4066                 return false;
4067 
4068         if (len <= skb_tailroom(to)) {
4069                 if (len)
4070                         BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
4071                 *delta_truesize = 0;
4072                 return true;
4073         }
4074 
4075         if (skb_has_frag_list(to) || skb_has_frag_list(from))
4076                 return false;
4077 
4078         if (skb_headlen(from) != 0) {
4079                 struct page *page;
4080                 unsigned int offset;
4081 
4082                 if (skb_shinfo(to)->nr_frags +
4083                     skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
4084                         return false;
4085 
4086                 if (skb_head_is_locked(from))
4087                         return false;
4088 
4089                 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4090 
4091                 page = virt_to_head_page(from->head);
4092                 offset = from->data - (unsigned char *)page_address(page);
4093 
4094                 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
4095                                    page, offset, skb_headlen(from));
4096                 *fragstolen = true;
4097         } else {
4098                 if (skb_shinfo(to)->nr_frags +
4099                     skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
4100                         return false;
4101 
4102                 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
4103         }
4104 
4105         WARN_ON_ONCE(delta < len);
4106 
4107         memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
4108                skb_shinfo(from)->frags,
4109                skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
4110         skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
4111 
4112         if (!skb_cloned(from))
4113                 skb_shinfo(from)->nr_frags = 0;
4114 
4115         /* if the skb is not cloned this does nothing
4116          * since we set nr_frags to 0.
4117          */
4118         for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
4119                 skb_frag_ref(from, i);
4120 
4121         to->truesize += delta;
4122         to->len += len;
4123         to->data_len += len;
4124 
4125         *delta_truesize = delta;
4126         return true;
4127 }
4128 EXPORT_SYMBOL(skb_try_coalesce);
4129 
4130 /**
4131  * skb_scrub_packet - scrub an skb
4132  *
4133  * @skb: buffer to clean
4134  * @xnet: packet is crossing netns
4135  *
4136  * skb_scrub_packet can be used after encapsulating or decapsulting a packet
4137  * into/from a tunnel. Some information have to be cleared during these
4138  * operations.
4139  * skb_scrub_packet can also be used to clean a skb before injecting it in
4140  * another namespace (@xnet == true). We have to clear all information in the
4141  * skb that could impact namespace isolation.
4142  */
4143 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
4144 {
4145         skb->tstamp.tv64 = 0;
4146         skb->pkt_type = PACKET_HOST;
4147         skb->skb_iif = 0;
4148         skb->ignore_df = 0;
4149         skb_dst_drop(skb);
4150         skb_sender_cpu_clear(skb);
4151         secpath_reset(skb);
4152         nf_reset(skb);
4153         nf_reset_trace(skb);
4154 
4155         if (!xnet)
4156                 return;
4157 
4158         skb_orphan(skb);
4159         skb->mark = 0;
4160 }
4161 EXPORT_SYMBOL_GPL(skb_scrub_packet);
4162 
4163 /**
4164  * skb_gso_transport_seglen - Return length of individual segments of a gso packet
4165  *
4166  * @skb: GSO skb
4167  *
4168  * skb_gso_transport_seglen is used to determine the real size of the
4169  * individual segments, including Layer4 headers (TCP/UDP).
4170  *
4171  * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
4172  */
4173 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
4174 {
4175         const struct skb_shared_info *shinfo = skb_shinfo(skb);
4176         unsigned int thlen = 0;
4177 
4178         if (skb->encapsulation) {
4179                 thlen = skb_inner_transport_header(skb) -
4180                         skb_transport_header(skb);
4181 
4182                 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
4183                         thlen += inner_tcp_hdrlen(skb);
4184         } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
4185                 thlen = tcp_hdrlen(skb);
4186         }
4187         /* UFO sets gso_size to the size of the fragmentation
4188          * payload, i.e. the size of the L4 (UDP) header is already
4189          * accounted for.
4190          */
4191         return thlen + shinfo->gso_size;
4192 }
4193 EXPORT_SYMBOL_GPL(skb_gso_transport_seglen);
4194 
4195 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
4196 {
4197         if (skb_cow(skb, skb_headroom(skb)) < 0) {
4198                 kfree_skb(skb);
4199                 return NULL;
4200         }
4201 
4202         memmove(skb->data - ETH_HLEN, skb->data - VLAN_ETH_HLEN, 2 * ETH_ALEN);
4203         skb->mac_header += VLAN_HLEN;
4204         return skb;
4205 }
4206 
4207 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
4208 {
4209         struct vlan_hdr *vhdr;
4210         u16 vlan_tci;
4211 
4212         if (unlikely(skb_vlan_tag_present(skb))) {
4213                 /* vlan_tci is already set-up so leave this for another time */
4214                 return skb;
4215         }
4216 
4217         skb = skb_share_check(skb, GFP_ATOMIC);
4218         if (unlikely(!skb))
4219                 goto err_free;
4220 
4221         if (unlikely(!pskb_may_pull(skb, VLAN_HLEN)))
4222                 goto err_free;
4223 
4224         vhdr = (struct vlan_hdr *)skb->data;
4225         vlan_tci = ntohs(vhdr->h_vlan_TCI);
4226         __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
4227 
4228         skb_pull_rcsum(skb, VLAN_HLEN);
4229         vlan_set_encap_proto(skb, vhdr);
4230 
4231         skb = skb_reorder_vlan_header(skb);
4232         if (unlikely(!skb))
4233                 goto err_free;
4234 
4235         skb_reset_network_header(skb);
4236         skb_reset_transport_header(skb);
4237         skb_reset_mac_len(skb);
4238 
4239         return skb;
4240 
4241 err_free:
4242         kfree_skb(skb);
4243         return NULL;
4244 }
4245 EXPORT_SYMBOL(skb_vlan_untag);
4246 
4247 int skb_ensure_writable(struct sk_buff *skb, int write_len)
4248 {
4249         if (!pskb_may_pull(skb, write_len))
4250                 return -ENOMEM;
4251 
4252         if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
4253                 return 0;
4254 
4255         return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4256 }
4257 EXPORT_SYMBOL(skb_ensure_writable);
4258 
4259 /* remove VLAN header from packet and update csum accordingly. */
4260 static int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
4261 {
4262         struct vlan_hdr *vhdr;
4263         unsigned int offset = skb->data - skb_mac_header(skb);
4264         int err;
4265 
4266         __skb_push(skb, offset);
4267         err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
4268         if (unlikely(err))
4269                 goto pull;
4270 
4271         skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
4272 
4273         vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
4274         *vlan_tci = ntohs(vhdr->h_vlan_TCI);
4275 
4276         memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
4277         __skb_pull(skb, VLAN_HLEN);
4278 
4279         vlan_set_encap_proto(skb, vhdr);
4280         skb->mac_header += VLAN_HLEN;
4281 
4282         if (skb_network_offset(skb) < ETH_HLEN)
4283                 skb_set_network_header(skb, ETH_HLEN);
4284 
4285         skb_reset_mac_len(skb);
4286 pull:
4287         __skb_pull(skb, offset);
4288 
4289         return err;
4290 }
4291 
4292 int skb_vlan_pop(struct sk_buff *skb)
4293 {
4294         u16 vlan_tci;
4295         __be16 vlan_proto;
4296         int err;
4297 
4298         if (likely(skb_vlan_tag_present(skb))) {
4299                 skb->vlan_tci = 0;
4300         } else {
4301                 if (unlikely((skb->protocol != htons(ETH_P_8021Q) &&
4302                               skb->protocol != htons(ETH_P_8021AD)) ||
4303                              skb->len < VLAN_ETH_HLEN))
4304                         return 0;
4305 
4306                 err = __skb_vlan_pop(skb, &vlan_tci);
4307                 if (err)
4308                         return err;
4309         }
4310         /* move next vlan tag to hw accel tag */
4311         if (likely((skb->protocol != htons(ETH_P_8021Q) &&
4312                     skb->protocol != htons(ETH_P_8021AD)) ||
4313                    skb->len < VLAN_ETH_HLEN))
4314                 return 0;
4315 
4316         vlan_proto = skb->protocol;
4317         err = __skb_vlan_pop(skb, &vlan_tci);
4318         if (unlikely(err))
4319                 return err;
4320 
4321         __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
4322         return 0;
4323 }
4324 EXPORT_SYMBOL(skb_vlan_pop);
4325 
4326 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
4327 {
4328         if (skb_vlan_tag_present(skb)) {
4329                 unsigned int offset = skb->data - skb_mac_header(skb);
4330                 int err;
4331 
4332                 /* __vlan_insert_tag expect skb->data pointing to mac header.
4333                  * So change skb->data before calling it and change back to
4334                  * original position later
4335                  */
4336                 __skb_push(skb, offset);
4337                 err = __vlan_insert_tag(skb, skb->vlan_proto,
4338                                         skb_vlan_tag_get(skb));
4339                 if (err)
4340                         return err;
4341                 skb->protocol = skb->vlan_proto;
4342                 skb->mac_len += VLAN_HLEN;
4343                 __skb_pull(skb, offset);
4344 
4345                 if (skb->ip_summed == CHECKSUM_COMPLETE)
4346                         skb->csum = csum_add(skb->csum, csum_partial(skb->data
4347                                         + (2 * ETH_ALEN), VLAN_HLEN, 0));
4348         }
4349         __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
4350         return 0;
4351 }
4352 EXPORT_SYMBOL(skb_vlan_push);
4353 
4354 /**
4355  * alloc_skb_with_frags - allocate skb with page frags
4356  *
4357  * @header_len: size of linear part
4358  * @data_len: needed length in frags
4359  * @max_page_order: max page order desired.
4360  * @errcode: pointer to error code if any
4361  * @gfp_mask: allocation mask
4362  *
4363  * This can be used to allocate a paged skb, given a maximal order for frags.
4364  */
4365 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
4366                                      unsigned long data_len,
4367                                      int max_page_order,
4368                                      int *errcode,
4369                                      gfp_t gfp_mask)
4370 {
4371         int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
4372         unsigned long chunk;
4373         struct sk_buff *skb;
4374         struct page *page;
4375         gfp_t gfp_head;
4376         int i;
4377 
4378         *errcode = -EMSGSIZE;
4379         /* Note this test could be relaxed, if we succeed to allocate
4380          * high order pages...
4381          */
4382         if (npages > MAX_SKB_FRAGS)
4383                 return NULL;
4384 
4385         gfp_head = gfp_mask;
4386         if (gfp_head & __GFP_WAIT)
4387                 gfp_head |= __GFP_REPEAT;
4388 
4389         *errcode = -ENOBUFS;
4390         skb = alloc_skb(header_len, gfp_head);
4391         if (!skb)
4392                 return NULL;
4393 
4394         skb->truesize += npages << PAGE_SHIFT;
4395 
4396         for (i = 0; npages > 0; i++) {
4397                 int order = max_page_order;
4398 
4399                 while (order) {
4400                         if (npages >= 1 << order) {
4401                                 page = alloc_pages((gfp_mask & ~__GFP_WAIT) |
4402                                                    __GFP_COMP |
4403                                                    __GFP_NOWARN |
4404                                                    __GFP_NORETRY,
4405                                                    order);
4406                                 if (page)
4407                                         goto fill_page;
4408                                 /* Do not retry other high order allocations */
4409                                 order = 1;
4410                                 max_page_order = 0;
4411                         }
4412                         order--;
4413                 }
4414                 page = alloc_page(gfp_mask);
4415                 if (!page)
4416                         goto failure;
4417 fill_page:
4418                 chunk = min_t(unsigned long, data_len,
4419                               PAGE_SIZE << order);
4420                 skb_fill_page_desc(skb, i, page, 0, chunk);
4421                 data_len -= chunk;
4422                 npages -= 1 << order;
4423         }
4424         return skb;
4425 
4426 failure:
4427         kfree_skb(skb);
4428         return NULL;
4429 }
4430 EXPORT_SYMBOL(alloc_skb_with_frags);
4431 

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