Version:  2.0.40 2.2.26 2.4.37 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 4.10

Linux/net/core/skbuff.c

  1 /*
  2  *      Routines having to do with the 'struct sk_buff' memory handlers.
  3  *
  4  *      Authors:        Alan Cox <alan@lxorguk.ukuu.org.uk>
  5  *                      Florian La Roche <rzsfl@rz.uni-sb.de>
  6  *
  7  *      Fixes:
  8  *              Alan Cox        :       Fixed the worst of the load
  9  *                                      balancer bugs.
 10  *              Dave Platt      :       Interrupt stacking fix.
 11  *      Richard Kooijman        :       Timestamp fixes.
 12  *              Alan Cox        :       Changed buffer format.
 13  *              Alan Cox        :       destructor hook for AF_UNIX etc.
 14  *              Linus Torvalds  :       Better skb_clone.
 15  *              Alan Cox        :       Added skb_copy.
 16  *              Alan Cox        :       Added all the changed routines Linus
 17  *                                      only put in the headers
 18  *              Ray VanTassle   :       Fixed --skb->lock in free
 19  *              Alan Cox        :       skb_copy copy arp field
 20  *              Andi Kleen      :       slabified it.
 21  *              Robert Olsson   :       Removed skb_head_pool
 22  *
 23  *      NOTE:
 24  *              The __skb_ routines should be called with interrupts
 25  *      disabled, or you better be *real* sure that the operation is atomic
 26  *      with respect to whatever list is being frobbed (e.g. via lock_sock()
 27  *      or via disabling bottom half handlers, etc).
 28  *
 29  *      This program is free software; you can redistribute it and/or
 30  *      modify it under the terms of the GNU General Public License
 31  *      as published by the Free Software Foundation; either version
 32  *      2 of the License, or (at your option) any later version.
 33  */
 34 
 35 /*
 36  *      The functions in this file will not compile correctly with gcc 2.4.x
 37  */
 38 
 39 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 40 
 41 #include <linux/module.h>
 42 #include <linux/types.h>
 43 #include <linux/kernel.h>
 44 #include <linux/kmemcheck.h>
 45 #include <linux/mm.h>
 46 #include <linux/interrupt.h>
 47 #include <linux/in.h>
 48 #include <linux/inet.h>
 49 #include <linux/slab.h>
 50 #include <linux/tcp.h>
 51 #include <linux/udp.h>
 52 #include <linux/sctp.h>
 53 #include <linux/netdevice.h>
 54 #ifdef CONFIG_NET_CLS_ACT
 55 #include <net/pkt_sched.h>
 56 #endif
 57 #include <linux/string.h>
 58 #include <linux/skbuff.h>
 59 #include <linux/splice.h>
 60 #include <linux/cache.h>
 61 #include <linux/rtnetlink.h>
 62 #include <linux/init.h>
 63 #include <linux/scatterlist.h>
 64 #include <linux/errqueue.h>
 65 #include <linux/prefetch.h>
 66 #include <linux/if_vlan.h>
 67 
 68 #include <net/protocol.h>
 69 #include <net/dst.h>
 70 #include <net/sock.h>
 71 #include <net/checksum.h>
 72 #include <net/ip6_checksum.h>
 73 #include <net/xfrm.h>
 74 
 75 #include <linux/uaccess.h>
 76 #include <trace/events/skb.h>
 77 #include <linux/highmem.h>
 78 #include <linux/capability.h>
 79 #include <linux/user_namespace.h>
 80 
 81 struct kmem_cache *skbuff_head_cache __read_mostly;
 82 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
 83 int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
 84 EXPORT_SYMBOL(sysctl_max_skb_frags);
 85 
 86 /**
 87  *      skb_panic - private function for out-of-line support
 88  *      @skb:   buffer
 89  *      @sz:    size
 90  *      @addr:  address
 91  *      @msg:   skb_over_panic or skb_under_panic
 92  *
 93  *      Out-of-line support for skb_put() and skb_push().
 94  *      Called via the wrapper skb_over_panic() or skb_under_panic().
 95  *      Keep out of line to prevent kernel bloat.
 96  *      __builtin_return_address is not used because it is not always reliable.
 97  */
 98 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
 99                       const char msg[])
100 {
101         pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
102                  msg, addr, skb->len, sz, skb->head, skb->data,
103                  (unsigned long)skb->tail, (unsigned long)skb->end,
104                  skb->dev ? skb->dev->name : "<NULL>");
105         BUG();
106 }
107 
108 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
109 {
110         skb_panic(skb, sz, addr, __func__);
111 }
112 
113 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
114 {
115         skb_panic(skb, sz, addr, __func__);
116 }
117 
118 /*
119  * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
120  * the caller if emergency pfmemalloc reserves are being used. If it is and
121  * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
122  * may be used. Otherwise, the packet data may be discarded until enough
123  * memory is free
124  */
125 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
126          __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
127 
128 static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
129                                unsigned long ip, bool *pfmemalloc)
130 {
131         void *obj;
132         bool ret_pfmemalloc = false;
133 
134         /*
135          * Try a regular allocation, when that fails and we're not entitled
136          * to the reserves, fail.
137          */
138         obj = kmalloc_node_track_caller(size,
139                                         flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
140                                         node);
141         if (obj || !(gfp_pfmemalloc_allowed(flags)))
142                 goto out;
143 
144         /* Try again but now we are using pfmemalloc reserves */
145         ret_pfmemalloc = true;
146         obj = kmalloc_node_track_caller(size, flags, node);
147 
148 out:
149         if (pfmemalloc)
150                 *pfmemalloc = ret_pfmemalloc;
151 
152         return obj;
153 }
154 
155 /*      Allocate a new skbuff. We do this ourselves so we can fill in a few
156  *      'private' fields and also do memory statistics to find all the
157  *      [BEEP] leaks.
158  *
159  */
160 
161 struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node)
162 {
163         struct sk_buff *skb;
164 
165         /* Get the HEAD */
166         skb = kmem_cache_alloc_node(skbuff_head_cache,
167                                     gfp_mask & ~__GFP_DMA, node);
168         if (!skb)
169                 goto out;
170 
171         /*
172          * Only clear those fields we need to clear, not those that we will
173          * actually initialise below. Hence, don't put any more fields after
174          * the tail pointer in struct sk_buff!
175          */
176         memset(skb, 0, offsetof(struct sk_buff, tail));
177         skb->head = NULL;
178         skb->truesize = sizeof(struct sk_buff);
179         atomic_set(&skb->users, 1);
180 
181         skb->mac_header = (typeof(skb->mac_header))~0U;
182 out:
183         return skb;
184 }
185 
186 /**
187  *      __alloc_skb     -       allocate a network buffer
188  *      @size: size to allocate
189  *      @gfp_mask: allocation mask
190  *      @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
191  *              instead of head cache and allocate a cloned (child) skb.
192  *              If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
193  *              allocations in case the data is required for writeback
194  *      @node: numa node to allocate memory on
195  *
196  *      Allocate a new &sk_buff. The returned buffer has no headroom and a
197  *      tail room of at least size bytes. The object has a reference count
198  *      of one. The return is the buffer. On a failure the return is %NULL.
199  *
200  *      Buffers may only be allocated from interrupts using a @gfp_mask of
201  *      %GFP_ATOMIC.
202  */
203 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
204                             int flags, int node)
205 {
206         struct kmem_cache *cache;
207         struct skb_shared_info *shinfo;
208         struct sk_buff *skb;
209         u8 *data;
210         bool pfmemalloc;
211 
212         cache = (flags & SKB_ALLOC_FCLONE)
213                 ? skbuff_fclone_cache : skbuff_head_cache;
214 
215         if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
216                 gfp_mask |= __GFP_MEMALLOC;
217 
218         /* Get the HEAD */
219         skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
220         if (!skb)
221                 goto out;
222         prefetchw(skb);
223 
224         /* We do our best to align skb_shared_info on a separate cache
225          * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
226          * aligned memory blocks, unless SLUB/SLAB debug is enabled.
227          * Both skb->head and skb_shared_info are cache line aligned.
228          */
229         size = SKB_DATA_ALIGN(size);
230         size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
231         data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
232         if (!data)
233                 goto nodata;
234         /* kmalloc(size) might give us more room than requested.
235          * Put skb_shared_info exactly at the end of allocated zone,
236          * to allow max possible filling before reallocation.
237          */
238         size = SKB_WITH_OVERHEAD(ksize(data));
239         prefetchw(data + size);
240 
241         /*
242          * Only clear those fields we need to clear, not those that we will
243          * actually initialise below. Hence, don't put any more fields after
244          * the tail pointer in struct sk_buff!
245          */
246         memset(skb, 0, offsetof(struct sk_buff, tail));
247         /* Account for allocated memory : skb + skb->head */
248         skb->truesize = SKB_TRUESIZE(size);
249         skb->pfmemalloc = pfmemalloc;
250         atomic_set(&skb->users, 1);
251         skb->head = data;
252         skb->data = data;
253         skb_reset_tail_pointer(skb);
254         skb->end = skb->tail + size;
255         skb->mac_header = (typeof(skb->mac_header))~0U;
256         skb->transport_header = (typeof(skb->transport_header))~0U;
257 
258         /* make sure we initialize shinfo sequentially */
259         shinfo = skb_shinfo(skb);
260         memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
261         atomic_set(&shinfo->dataref, 1);
262         kmemcheck_annotate_variable(shinfo->destructor_arg);
263 
264         if (flags & SKB_ALLOC_FCLONE) {
265                 struct sk_buff_fclones *fclones;
266 
267                 fclones = container_of(skb, struct sk_buff_fclones, skb1);
268 
269                 kmemcheck_annotate_bitfield(&fclones->skb2, flags1);
270                 skb->fclone = SKB_FCLONE_ORIG;
271                 atomic_set(&fclones->fclone_ref, 1);
272 
273                 fclones->skb2.fclone = SKB_FCLONE_CLONE;
274                 fclones->skb2.pfmemalloc = pfmemalloc;
275         }
276 out:
277         return skb;
278 nodata:
279         kmem_cache_free(cache, skb);
280         skb = NULL;
281         goto out;
282 }
283 EXPORT_SYMBOL(__alloc_skb);
284 
285 /**
286  * __build_skb - build a network buffer
287  * @data: data buffer provided by caller
288  * @frag_size: size of data, or 0 if head was kmalloced
289  *
290  * Allocate a new &sk_buff. Caller provides space holding head and
291  * skb_shared_info. @data must have been allocated by kmalloc() only if
292  * @frag_size is 0, otherwise data should come from the page allocator
293  *  or vmalloc()
294  * The return is the new skb buffer.
295  * On a failure the return is %NULL, and @data is not freed.
296  * Notes :
297  *  Before IO, driver allocates only data buffer where NIC put incoming frame
298  *  Driver should add room at head (NET_SKB_PAD) and
299  *  MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
300  *  After IO, driver calls build_skb(), to allocate sk_buff and populate it
301  *  before giving packet to stack.
302  *  RX rings only contains data buffers, not full skbs.
303  */
304 struct sk_buff *__build_skb(void *data, unsigned int frag_size)
305 {
306         struct skb_shared_info *shinfo;
307         struct sk_buff *skb;
308         unsigned int size = frag_size ? : ksize(data);
309 
310         skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
311         if (!skb)
312                 return NULL;
313 
314         size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
315 
316         memset(skb, 0, offsetof(struct sk_buff, tail));
317         skb->truesize = SKB_TRUESIZE(size);
318         atomic_set(&skb->users, 1);
319         skb->head = data;
320         skb->data = data;
321         skb_reset_tail_pointer(skb);
322         skb->end = skb->tail + size;
323         skb->mac_header = (typeof(skb->mac_header))~0U;
324         skb->transport_header = (typeof(skb->transport_header))~0U;
325 
326         /* make sure we initialize shinfo sequentially */
327         shinfo = skb_shinfo(skb);
328         memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
329         atomic_set(&shinfo->dataref, 1);
330         kmemcheck_annotate_variable(shinfo->destructor_arg);
331 
332         return skb;
333 }
334 
335 /* build_skb() is wrapper over __build_skb(), that specifically
336  * takes care of skb->head and skb->pfmemalloc
337  * This means that if @frag_size is not zero, then @data must be backed
338  * by a page fragment, not kmalloc() or vmalloc()
339  */
340 struct sk_buff *build_skb(void *data, unsigned int frag_size)
341 {
342         struct sk_buff *skb = __build_skb(data, frag_size);
343 
344         if (skb && frag_size) {
345                 skb->head_frag = 1;
346                 if (page_is_pfmemalloc(virt_to_head_page(data)))
347                         skb->pfmemalloc = 1;
348         }
349         return skb;
350 }
351 EXPORT_SYMBOL(build_skb);
352 
353 #define NAPI_SKB_CACHE_SIZE     64
354 
355 struct napi_alloc_cache {
356         struct page_frag_cache page;
357         unsigned int skb_count;
358         void *skb_cache[NAPI_SKB_CACHE_SIZE];
359 };
360 
361 static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
362 static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
363 
364 static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
365 {
366         struct page_frag_cache *nc;
367         unsigned long flags;
368         void *data;
369 
370         local_irq_save(flags);
371         nc = this_cpu_ptr(&netdev_alloc_cache);
372         data = page_frag_alloc(nc, fragsz, gfp_mask);
373         local_irq_restore(flags);
374         return data;
375 }
376 
377 /**
378  * netdev_alloc_frag - allocate a page fragment
379  * @fragsz: fragment size
380  *
381  * Allocates a frag from a page for receive buffer.
382  * Uses GFP_ATOMIC allocations.
383  */
384 void *netdev_alloc_frag(unsigned int fragsz)
385 {
386         return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
387 }
388 EXPORT_SYMBOL(netdev_alloc_frag);
389 
390 static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
391 {
392         struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
393 
394         return page_frag_alloc(&nc->page, fragsz, gfp_mask);
395 }
396 
397 void *napi_alloc_frag(unsigned int fragsz)
398 {
399         return __napi_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
400 }
401 EXPORT_SYMBOL(napi_alloc_frag);
402 
403 /**
404  *      __netdev_alloc_skb - allocate an skbuff for rx on a specific device
405  *      @dev: network device to receive on
406  *      @len: length to allocate
407  *      @gfp_mask: get_free_pages mask, passed to alloc_skb
408  *
409  *      Allocate a new &sk_buff and assign it a usage count of one. The
410  *      buffer has NET_SKB_PAD headroom built in. Users should allocate
411  *      the headroom they think they need without accounting for the
412  *      built in space. The built in space is used for optimisations.
413  *
414  *      %NULL is returned if there is no free memory.
415  */
416 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
417                                    gfp_t gfp_mask)
418 {
419         struct page_frag_cache *nc;
420         unsigned long flags;
421         struct sk_buff *skb;
422         bool pfmemalloc;
423         void *data;
424 
425         len += NET_SKB_PAD;
426 
427         if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
428             (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
429                 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
430                 if (!skb)
431                         goto skb_fail;
432                 goto skb_success;
433         }
434 
435         len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
436         len = SKB_DATA_ALIGN(len);
437 
438         if (sk_memalloc_socks())
439                 gfp_mask |= __GFP_MEMALLOC;
440 
441         local_irq_save(flags);
442 
443         nc = this_cpu_ptr(&netdev_alloc_cache);
444         data = page_frag_alloc(nc, len, gfp_mask);
445         pfmemalloc = nc->pfmemalloc;
446 
447         local_irq_restore(flags);
448 
449         if (unlikely(!data))
450                 return NULL;
451 
452         skb = __build_skb(data, len);
453         if (unlikely(!skb)) {
454                 skb_free_frag(data);
455                 return NULL;
456         }
457 
458         /* use OR instead of assignment to avoid clearing of bits in mask */
459         if (pfmemalloc)
460                 skb->pfmemalloc = 1;
461         skb->head_frag = 1;
462 
463 skb_success:
464         skb_reserve(skb, NET_SKB_PAD);
465         skb->dev = dev;
466 
467 skb_fail:
468         return skb;
469 }
470 EXPORT_SYMBOL(__netdev_alloc_skb);
471 
472 /**
473  *      __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
474  *      @napi: napi instance this buffer was allocated for
475  *      @len: length to allocate
476  *      @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
477  *
478  *      Allocate a new sk_buff for use in NAPI receive.  This buffer will
479  *      attempt to allocate the head from a special reserved region used
480  *      only for NAPI Rx allocation.  By doing this we can save several
481  *      CPU cycles by avoiding having to disable and re-enable IRQs.
482  *
483  *      %NULL is returned if there is no free memory.
484  */
485 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
486                                  gfp_t gfp_mask)
487 {
488         struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
489         struct sk_buff *skb;
490         void *data;
491 
492         len += NET_SKB_PAD + NET_IP_ALIGN;
493 
494         if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
495             (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
496                 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
497                 if (!skb)
498                         goto skb_fail;
499                 goto skb_success;
500         }
501 
502         len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
503         len = SKB_DATA_ALIGN(len);
504 
505         if (sk_memalloc_socks())
506                 gfp_mask |= __GFP_MEMALLOC;
507 
508         data = page_frag_alloc(&nc->page, len, gfp_mask);
509         if (unlikely(!data))
510                 return NULL;
511 
512         skb = __build_skb(data, len);
513         if (unlikely(!skb)) {
514                 skb_free_frag(data);
515                 return NULL;
516         }
517 
518         /* use OR instead of assignment to avoid clearing of bits in mask */
519         if (nc->page.pfmemalloc)
520                 skb->pfmemalloc = 1;
521         skb->head_frag = 1;
522 
523 skb_success:
524         skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
525         skb->dev = napi->dev;
526 
527 skb_fail:
528         return skb;
529 }
530 EXPORT_SYMBOL(__napi_alloc_skb);
531 
532 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
533                      int size, unsigned int truesize)
534 {
535         skb_fill_page_desc(skb, i, page, off, size);
536         skb->len += size;
537         skb->data_len += size;
538         skb->truesize += truesize;
539 }
540 EXPORT_SYMBOL(skb_add_rx_frag);
541 
542 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
543                           unsigned int truesize)
544 {
545         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
546 
547         skb_frag_size_add(frag, size);
548         skb->len += size;
549         skb->data_len += size;
550         skb->truesize += truesize;
551 }
552 EXPORT_SYMBOL(skb_coalesce_rx_frag);
553 
554 static void skb_drop_list(struct sk_buff **listp)
555 {
556         kfree_skb_list(*listp);
557         *listp = NULL;
558 }
559 
560 static inline void skb_drop_fraglist(struct sk_buff *skb)
561 {
562         skb_drop_list(&skb_shinfo(skb)->frag_list);
563 }
564 
565 static void skb_clone_fraglist(struct sk_buff *skb)
566 {
567         struct sk_buff *list;
568 
569         skb_walk_frags(skb, list)
570                 skb_get(list);
571 }
572 
573 static void skb_free_head(struct sk_buff *skb)
574 {
575         unsigned char *head = skb->head;
576 
577         if (skb->head_frag)
578                 skb_free_frag(head);
579         else
580                 kfree(head);
581 }
582 
583 static void skb_release_data(struct sk_buff *skb)
584 {
585         struct skb_shared_info *shinfo = skb_shinfo(skb);
586         int i;
587 
588         if (skb->cloned &&
589             atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
590                               &shinfo->dataref))
591                 return;
592 
593         for (i = 0; i < shinfo->nr_frags; i++)
594                 __skb_frag_unref(&shinfo->frags[i]);
595 
596         /*
597          * If skb buf is from userspace, we need to notify the caller
598          * the lower device DMA has done;
599          */
600         if (shinfo->tx_flags & SKBTX_DEV_ZEROCOPY) {
601                 struct ubuf_info *uarg;
602 
603                 uarg = shinfo->destructor_arg;
604                 if (uarg->callback)
605                         uarg->callback(uarg, true);
606         }
607 
608         if (shinfo->frag_list)
609                 kfree_skb_list(shinfo->frag_list);
610 
611         skb_free_head(skb);
612 }
613 
614 /*
615  *      Free an skbuff by memory without cleaning the state.
616  */
617 static void kfree_skbmem(struct sk_buff *skb)
618 {
619         struct sk_buff_fclones *fclones;
620 
621         switch (skb->fclone) {
622         case SKB_FCLONE_UNAVAILABLE:
623                 kmem_cache_free(skbuff_head_cache, skb);
624                 return;
625 
626         case SKB_FCLONE_ORIG:
627                 fclones = container_of(skb, struct sk_buff_fclones, skb1);
628 
629                 /* We usually free the clone (TX completion) before original skb
630                  * This test would have no chance to be true for the clone,
631                  * while here, branch prediction will be good.
632                  */
633                 if (atomic_read(&fclones->fclone_ref) == 1)
634                         goto fastpath;
635                 break;
636 
637         default: /* SKB_FCLONE_CLONE */
638                 fclones = container_of(skb, struct sk_buff_fclones, skb2);
639                 break;
640         }
641         if (!atomic_dec_and_test(&fclones->fclone_ref))
642                 return;
643 fastpath:
644         kmem_cache_free(skbuff_fclone_cache, fclones);
645 }
646 
647 static void skb_release_head_state(struct sk_buff *skb)
648 {
649         skb_dst_drop(skb);
650 #ifdef CONFIG_XFRM
651         secpath_put(skb->sp);
652 #endif
653         if (skb->destructor) {
654                 WARN_ON(in_irq());
655                 skb->destructor(skb);
656         }
657 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
658         nf_conntrack_put(skb->nfct);
659 #endif
660 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
661         nf_bridge_put(skb->nf_bridge);
662 #endif
663 }
664 
665 /* Free everything but the sk_buff shell. */
666 static void skb_release_all(struct sk_buff *skb)
667 {
668         skb_release_head_state(skb);
669         if (likely(skb->head))
670                 skb_release_data(skb);
671 }
672 
673 /**
674  *      __kfree_skb - private function
675  *      @skb: buffer
676  *
677  *      Free an sk_buff. Release anything attached to the buffer.
678  *      Clean the state. This is an internal helper function. Users should
679  *      always call kfree_skb
680  */
681 
682 void __kfree_skb(struct sk_buff *skb)
683 {
684         skb_release_all(skb);
685         kfree_skbmem(skb);
686 }
687 EXPORT_SYMBOL(__kfree_skb);
688 
689 /**
690  *      kfree_skb - free an sk_buff
691  *      @skb: buffer to free
692  *
693  *      Drop a reference to the buffer and free it if the usage count has
694  *      hit zero.
695  */
696 void kfree_skb(struct sk_buff *skb)
697 {
698         if (unlikely(!skb))
699                 return;
700         if (likely(atomic_read(&skb->users) == 1))
701                 smp_rmb();
702         else if (likely(!atomic_dec_and_test(&skb->users)))
703                 return;
704         trace_kfree_skb(skb, __builtin_return_address(0));
705         __kfree_skb(skb);
706 }
707 EXPORT_SYMBOL(kfree_skb);
708 
709 void kfree_skb_list(struct sk_buff *segs)
710 {
711         while (segs) {
712                 struct sk_buff *next = segs->next;
713 
714                 kfree_skb(segs);
715                 segs = next;
716         }
717 }
718 EXPORT_SYMBOL(kfree_skb_list);
719 
720 /**
721  *      skb_tx_error - report an sk_buff xmit error
722  *      @skb: buffer that triggered an error
723  *
724  *      Report xmit error if a device callback is tracking this skb.
725  *      skb must be freed afterwards.
726  */
727 void skb_tx_error(struct sk_buff *skb)
728 {
729         if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
730                 struct ubuf_info *uarg;
731 
732                 uarg = skb_shinfo(skb)->destructor_arg;
733                 if (uarg->callback)
734                         uarg->callback(uarg, false);
735                 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
736         }
737 }
738 EXPORT_SYMBOL(skb_tx_error);
739 
740 /**
741  *      consume_skb - free an skbuff
742  *      @skb: buffer to free
743  *
744  *      Drop a ref to the buffer and free it if the usage count has hit zero
745  *      Functions identically to kfree_skb, but kfree_skb assumes that the frame
746  *      is being dropped after a failure and notes that
747  */
748 void consume_skb(struct sk_buff *skb)
749 {
750         if (unlikely(!skb))
751                 return;
752         if (likely(atomic_read(&skb->users) == 1))
753                 smp_rmb();
754         else if (likely(!atomic_dec_and_test(&skb->users)))
755                 return;
756         trace_consume_skb(skb);
757         __kfree_skb(skb);
758 }
759 EXPORT_SYMBOL(consume_skb);
760 
761 void __kfree_skb_flush(void)
762 {
763         struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
764 
765         /* flush skb_cache if containing objects */
766         if (nc->skb_count) {
767                 kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count,
768                                      nc->skb_cache);
769                 nc->skb_count = 0;
770         }
771 }
772 
773 static inline void _kfree_skb_defer(struct sk_buff *skb)
774 {
775         struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
776 
777         /* drop skb->head and call any destructors for packet */
778         skb_release_all(skb);
779 
780         /* record skb to CPU local list */
781         nc->skb_cache[nc->skb_count++] = skb;
782 
783 #ifdef CONFIG_SLUB
784         /* SLUB writes into objects when freeing */
785         prefetchw(skb);
786 #endif
787 
788         /* flush skb_cache if it is filled */
789         if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
790                 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE,
791                                      nc->skb_cache);
792                 nc->skb_count = 0;
793         }
794 }
795 void __kfree_skb_defer(struct sk_buff *skb)
796 {
797         _kfree_skb_defer(skb);
798 }
799 
800 void napi_consume_skb(struct sk_buff *skb, int budget)
801 {
802         if (unlikely(!skb))
803                 return;
804 
805         /* Zero budget indicate non-NAPI context called us, like netpoll */
806         if (unlikely(!budget)) {
807                 dev_consume_skb_any(skb);
808                 return;
809         }
810 
811         if (likely(atomic_read(&skb->users) == 1))
812                 smp_rmb();
813         else if (likely(!atomic_dec_and_test(&skb->users)))
814                 return;
815         /* if reaching here SKB is ready to free */
816         trace_consume_skb(skb);
817 
818         /* if SKB is a clone, don't handle this case */
819         if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
820                 __kfree_skb(skb);
821                 return;
822         }
823 
824         _kfree_skb_defer(skb);
825 }
826 EXPORT_SYMBOL(napi_consume_skb);
827 
828 /* Make sure a field is enclosed inside headers_start/headers_end section */
829 #define CHECK_SKB_FIELD(field) \
830         BUILD_BUG_ON(offsetof(struct sk_buff, field) <          \
831                      offsetof(struct sk_buff, headers_start));  \
832         BUILD_BUG_ON(offsetof(struct sk_buff, field) >          \
833                      offsetof(struct sk_buff, headers_end));    \
834 
835 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
836 {
837         new->tstamp             = old->tstamp;
838         /* We do not copy old->sk */
839         new->dev                = old->dev;
840         memcpy(new->cb, old->cb, sizeof(old->cb));
841         skb_dst_copy(new, old);
842 #ifdef CONFIG_XFRM
843         new->sp                 = secpath_get(old->sp);
844 #endif
845         __nf_copy(new, old, false);
846 
847         /* Note : this field could be in headers_start/headers_end section
848          * It is not yet because we do not want to have a 16 bit hole
849          */
850         new->queue_mapping = old->queue_mapping;
851 
852         memcpy(&new->headers_start, &old->headers_start,
853                offsetof(struct sk_buff, headers_end) -
854                offsetof(struct sk_buff, headers_start));
855         CHECK_SKB_FIELD(protocol);
856         CHECK_SKB_FIELD(csum);
857         CHECK_SKB_FIELD(hash);
858         CHECK_SKB_FIELD(priority);
859         CHECK_SKB_FIELD(skb_iif);
860         CHECK_SKB_FIELD(vlan_proto);
861         CHECK_SKB_FIELD(vlan_tci);
862         CHECK_SKB_FIELD(transport_header);
863         CHECK_SKB_FIELD(network_header);
864         CHECK_SKB_FIELD(mac_header);
865         CHECK_SKB_FIELD(inner_protocol);
866         CHECK_SKB_FIELD(inner_transport_header);
867         CHECK_SKB_FIELD(inner_network_header);
868         CHECK_SKB_FIELD(inner_mac_header);
869         CHECK_SKB_FIELD(mark);
870 #ifdef CONFIG_NETWORK_SECMARK
871         CHECK_SKB_FIELD(secmark);
872 #endif
873 #ifdef CONFIG_NET_RX_BUSY_POLL
874         CHECK_SKB_FIELD(napi_id);
875 #endif
876 #ifdef CONFIG_XPS
877         CHECK_SKB_FIELD(sender_cpu);
878 #endif
879 #ifdef CONFIG_NET_SCHED
880         CHECK_SKB_FIELD(tc_index);
881 #ifdef CONFIG_NET_CLS_ACT
882         CHECK_SKB_FIELD(tc_verd);
883 #endif
884 #endif
885 
886 }
887 
888 /*
889  * You should not add any new code to this function.  Add it to
890  * __copy_skb_header above instead.
891  */
892 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
893 {
894 #define C(x) n->x = skb->x
895 
896         n->next = n->prev = NULL;
897         n->sk = NULL;
898         __copy_skb_header(n, skb);
899 
900         C(len);
901         C(data_len);
902         C(mac_len);
903         n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
904         n->cloned = 1;
905         n->nohdr = 0;
906         n->destructor = NULL;
907         C(tail);
908         C(end);
909         C(head);
910         C(head_frag);
911         C(data);
912         C(truesize);
913         atomic_set(&n->users, 1);
914 
915         atomic_inc(&(skb_shinfo(skb)->dataref));
916         skb->cloned = 1;
917 
918         return n;
919 #undef C
920 }
921 
922 /**
923  *      skb_morph       -       morph one skb into another
924  *      @dst: the skb to receive the contents
925  *      @src: the skb to supply the contents
926  *
927  *      This is identical to skb_clone except that the target skb is
928  *      supplied by the user.
929  *
930  *      The target skb is returned upon exit.
931  */
932 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
933 {
934         skb_release_all(dst);
935         return __skb_clone(dst, src);
936 }
937 EXPORT_SYMBOL_GPL(skb_morph);
938 
939 /**
940  *      skb_copy_ubufs  -       copy userspace skb frags buffers to kernel
941  *      @skb: the skb to modify
942  *      @gfp_mask: allocation priority
943  *
944  *      This must be called on SKBTX_DEV_ZEROCOPY skb.
945  *      It will copy all frags into kernel and drop the reference
946  *      to userspace pages.
947  *
948  *      If this function is called from an interrupt gfp_mask() must be
949  *      %GFP_ATOMIC.
950  *
951  *      Returns 0 on success or a negative error code on failure
952  *      to allocate kernel memory to copy to.
953  */
954 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
955 {
956         int i;
957         int num_frags = skb_shinfo(skb)->nr_frags;
958         struct page *page, *head = NULL;
959         struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
960 
961         for (i = 0; i < num_frags; i++) {
962                 u8 *vaddr;
963                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
964 
965                 page = alloc_page(gfp_mask);
966                 if (!page) {
967                         while (head) {
968                                 struct page *next = (struct page *)page_private(head);
969                                 put_page(head);
970                                 head = next;
971                         }
972                         return -ENOMEM;
973                 }
974                 vaddr = kmap_atomic(skb_frag_page(f));
975                 memcpy(page_address(page),
976                        vaddr + f->page_offset, skb_frag_size(f));
977                 kunmap_atomic(vaddr);
978                 set_page_private(page, (unsigned long)head);
979                 head = page;
980         }
981 
982         /* skb frags release userspace buffers */
983         for (i = 0; i < num_frags; i++)
984                 skb_frag_unref(skb, i);
985 
986         uarg->callback(uarg, false);
987 
988         /* skb frags point to kernel buffers */
989         for (i = num_frags - 1; i >= 0; i--) {
990                 __skb_fill_page_desc(skb, i, head, 0,
991                                      skb_shinfo(skb)->frags[i].size);
992                 head = (struct page *)page_private(head);
993         }
994 
995         skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
996         return 0;
997 }
998 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
999 
1000 /**
1001  *      skb_clone       -       duplicate an sk_buff
1002  *      @skb: buffer to clone
1003  *      @gfp_mask: allocation priority
1004  *
1005  *      Duplicate an &sk_buff. The new one is not owned by a socket. Both
1006  *      copies share the same packet data but not structure. The new
1007  *      buffer has a reference count of 1. If the allocation fails the
1008  *      function returns %NULL otherwise the new buffer is returned.
1009  *
1010  *      If this function is called from an interrupt gfp_mask() must be
1011  *      %GFP_ATOMIC.
1012  */
1013 
1014 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1015 {
1016         struct sk_buff_fclones *fclones = container_of(skb,
1017                                                        struct sk_buff_fclones,
1018                                                        skb1);
1019         struct sk_buff *n;
1020 
1021         if (skb_orphan_frags(skb, gfp_mask))
1022                 return NULL;
1023 
1024         if (skb->fclone == SKB_FCLONE_ORIG &&
1025             atomic_read(&fclones->fclone_ref) == 1) {
1026                 n = &fclones->skb2;
1027                 atomic_set(&fclones->fclone_ref, 2);
1028         } else {
1029                 if (skb_pfmemalloc(skb))
1030                         gfp_mask |= __GFP_MEMALLOC;
1031 
1032                 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1033                 if (!n)
1034                         return NULL;
1035 
1036                 kmemcheck_annotate_bitfield(n, flags1);
1037                 n->fclone = SKB_FCLONE_UNAVAILABLE;
1038         }
1039 
1040         return __skb_clone(n, skb);
1041 }
1042 EXPORT_SYMBOL(skb_clone);
1043 
1044 static void skb_headers_offset_update(struct sk_buff *skb, int off)
1045 {
1046         /* Only adjust this if it actually is csum_start rather than csum */
1047         if (skb->ip_summed == CHECKSUM_PARTIAL)
1048                 skb->csum_start += off;
1049         /* {transport,network,mac}_header and tail are relative to skb->head */
1050         skb->transport_header += off;
1051         skb->network_header   += off;
1052         if (skb_mac_header_was_set(skb))
1053                 skb->mac_header += off;
1054         skb->inner_transport_header += off;
1055         skb->inner_network_header += off;
1056         skb->inner_mac_header += off;
1057 }
1058 
1059 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
1060 {
1061         __copy_skb_header(new, old);
1062 
1063         skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1064         skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1065         skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1066 }
1067 
1068 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1069 {
1070         if (skb_pfmemalloc(skb))
1071                 return SKB_ALLOC_RX;
1072         return 0;
1073 }
1074 
1075 /**
1076  *      skb_copy        -       create private copy of an sk_buff
1077  *      @skb: buffer to copy
1078  *      @gfp_mask: allocation priority
1079  *
1080  *      Make a copy of both an &sk_buff and its data. This is used when the
1081  *      caller wishes to modify the data and needs a private copy of the
1082  *      data to alter. Returns %NULL on failure or the pointer to the buffer
1083  *      on success. The returned buffer has a reference count of 1.
1084  *
1085  *      As by-product this function converts non-linear &sk_buff to linear
1086  *      one, so that &sk_buff becomes completely private and caller is allowed
1087  *      to modify all the data of returned buffer. This means that this
1088  *      function is not recommended for use in circumstances when only
1089  *      header is going to be modified. Use pskb_copy() instead.
1090  */
1091 
1092 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1093 {
1094         int headerlen = skb_headroom(skb);
1095         unsigned int size = skb_end_offset(skb) + skb->data_len;
1096         struct sk_buff *n = __alloc_skb(size, gfp_mask,
1097                                         skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1098 
1099         if (!n)
1100                 return NULL;
1101 
1102         /* Set the data pointer */
1103         skb_reserve(n, headerlen);
1104         /* Set the tail pointer and length */
1105         skb_put(n, skb->len);
1106 
1107         if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
1108                 BUG();
1109 
1110         copy_skb_header(n, skb);
1111         return n;
1112 }
1113 EXPORT_SYMBOL(skb_copy);
1114 
1115 /**
1116  *      __pskb_copy_fclone      -  create copy of an sk_buff with private head.
1117  *      @skb: buffer to copy
1118  *      @headroom: headroom of new skb
1119  *      @gfp_mask: allocation priority
1120  *      @fclone: if true allocate the copy of the skb from the fclone
1121  *      cache instead of the head cache; it is recommended to set this
1122  *      to true for the cases where the copy will likely be cloned
1123  *
1124  *      Make a copy of both an &sk_buff and part of its data, located
1125  *      in header. Fragmented data remain shared. This is used when
1126  *      the caller wishes to modify only header of &sk_buff and needs
1127  *      private copy of the header to alter. Returns %NULL on failure
1128  *      or the pointer to the buffer on success.
1129  *      The returned buffer has a reference count of 1.
1130  */
1131 
1132 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1133                                    gfp_t gfp_mask, bool fclone)
1134 {
1135         unsigned int size = skb_headlen(skb) + headroom;
1136         int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1137         struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1138 
1139         if (!n)
1140                 goto out;
1141 
1142         /* Set the data pointer */
1143         skb_reserve(n, headroom);
1144         /* Set the tail pointer and length */
1145         skb_put(n, skb_headlen(skb));
1146         /* Copy the bytes */
1147         skb_copy_from_linear_data(skb, n->data, n->len);
1148 
1149         n->truesize += skb->data_len;
1150         n->data_len  = skb->data_len;
1151         n->len       = skb->len;
1152 
1153         if (skb_shinfo(skb)->nr_frags) {
1154                 int i;
1155 
1156                 if (skb_orphan_frags(skb, gfp_mask)) {
1157                         kfree_skb(n);
1158                         n = NULL;
1159                         goto out;
1160                 }
1161                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1162                         skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1163                         skb_frag_ref(skb, i);
1164                 }
1165                 skb_shinfo(n)->nr_frags = i;
1166         }
1167 
1168         if (skb_has_frag_list(skb)) {
1169                 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1170                 skb_clone_fraglist(n);
1171         }
1172 
1173         copy_skb_header(n, skb);
1174 out:
1175         return n;
1176 }
1177 EXPORT_SYMBOL(__pskb_copy_fclone);
1178 
1179 /**
1180  *      pskb_expand_head - reallocate header of &sk_buff
1181  *      @skb: buffer to reallocate
1182  *      @nhead: room to add at head
1183  *      @ntail: room to add at tail
1184  *      @gfp_mask: allocation priority
1185  *
1186  *      Expands (or creates identical copy, if @nhead and @ntail are zero)
1187  *      header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1188  *      reference count of 1. Returns zero in the case of success or error,
1189  *      if expansion failed. In the last case, &sk_buff is not changed.
1190  *
1191  *      All the pointers pointing into skb header may change and must be
1192  *      reloaded after call to this function.
1193  */
1194 
1195 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1196                      gfp_t gfp_mask)
1197 {
1198         int i;
1199         u8 *data;
1200         int size = nhead + skb_end_offset(skb) + ntail;
1201         long off;
1202 
1203         BUG_ON(nhead < 0);
1204 
1205         if (skb_shared(skb))
1206                 BUG();
1207 
1208         size = SKB_DATA_ALIGN(size);
1209 
1210         if (skb_pfmemalloc(skb))
1211                 gfp_mask |= __GFP_MEMALLOC;
1212         data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1213                                gfp_mask, NUMA_NO_NODE, NULL);
1214         if (!data)
1215                 goto nodata;
1216         size = SKB_WITH_OVERHEAD(ksize(data));
1217 
1218         /* Copy only real data... and, alas, header. This should be
1219          * optimized for the cases when header is void.
1220          */
1221         memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1222 
1223         memcpy((struct skb_shared_info *)(data + size),
1224                skb_shinfo(skb),
1225                offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1226 
1227         /*
1228          * if shinfo is shared we must drop the old head gracefully, but if it
1229          * is not we can just drop the old head and let the existing refcount
1230          * be since all we did is relocate the values
1231          */
1232         if (skb_cloned(skb)) {
1233                 /* copy this zero copy skb frags */
1234                 if (skb_orphan_frags(skb, gfp_mask))
1235                         goto nofrags;
1236                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1237                         skb_frag_ref(skb, i);
1238 
1239                 if (skb_has_frag_list(skb))
1240                         skb_clone_fraglist(skb);
1241 
1242                 skb_release_data(skb);
1243         } else {
1244                 skb_free_head(skb);
1245         }
1246         off = (data + nhead) - skb->head;
1247 
1248         skb->head     = data;
1249         skb->head_frag = 0;
1250         skb->data    += off;
1251 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1252         skb->end      = size;
1253         off           = nhead;
1254 #else
1255         skb->end      = skb->head + size;
1256 #endif
1257         skb->tail             += off;
1258         skb_headers_offset_update(skb, nhead);
1259         skb->cloned   = 0;
1260         skb->hdr_len  = 0;
1261         skb->nohdr    = 0;
1262         atomic_set(&skb_shinfo(skb)->dataref, 1);
1263         return 0;
1264 
1265 nofrags:
1266         kfree(data);
1267 nodata:
1268         return -ENOMEM;
1269 }
1270 EXPORT_SYMBOL(pskb_expand_head);
1271 
1272 /* Make private copy of skb with writable head and some headroom */
1273 
1274 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1275 {
1276         struct sk_buff *skb2;
1277         int delta = headroom - skb_headroom(skb);
1278 
1279         if (delta <= 0)
1280                 skb2 = pskb_copy(skb, GFP_ATOMIC);
1281         else {
1282                 skb2 = skb_clone(skb, GFP_ATOMIC);
1283                 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1284                                              GFP_ATOMIC)) {
1285                         kfree_skb(skb2);
1286                         skb2 = NULL;
1287                 }
1288         }
1289         return skb2;
1290 }
1291 EXPORT_SYMBOL(skb_realloc_headroom);
1292 
1293 /**
1294  *      skb_copy_expand -       copy and expand sk_buff
1295  *      @skb: buffer to copy
1296  *      @newheadroom: new free bytes at head
1297  *      @newtailroom: new free bytes at tail
1298  *      @gfp_mask: allocation priority
1299  *
1300  *      Make a copy of both an &sk_buff and its data and while doing so
1301  *      allocate additional space.
1302  *
1303  *      This is used when the caller wishes to modify the data and needs a
1304  *      private copy of the data to alter as well as more space for new fields.
1305  *      Returns %NULL on failure or the pointer to the buffer
1306  *      on success. The returned buffer has a reference count of 1.
1307  *
1308  *      You must pass %GFP_ATOMIC as the allocation priority if this function
1309  *      is called from an interrupt.
1310  */
1311 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1312                                 int newheadroom, int newtailroom,
1313                                 gfp_t gfp_mask)
1314 {
1315         /*
1316          *      Allocate the copy buffer
1317          */
1318         struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1319                                         gfp_mask, skb_alloc_rx_flag(skb),
1320                                         NUMA_NO_NODE);
1321         int oldheadroom = skb_headroom(skb);
1322         int head_copy_len, head_copy_off;
1323 
1324         if (!n)
1325                 return NULL;
1326 
1327         skb_reserve(n, newheadroom);
1328 
1329         /* Set the tail pointer and length */
1330         skb_put(n, skb->len);
1331 
1332         head_copy_len = oldheadroom;
1333         head_copy_off = 0;
1334         if (newheadroom <= head_copy_len)
1335                 head_copy_len = newheadroom;
1336         else
1337                 head_copy_off = newheadroom - head_copy_len;
1338 
1339         /* Copy the linear header and data. */
1340         if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1341                           skb->len + head_copy_len))
1342                 BUG();
1343 
1344         copy_skb_header(n, skb);
1345 
1346         skb_headers_offset_update(n, newheadroom - oldheadroom);
1347 
1348         return n;
1349 }
1350 EXPORT_SYMBOL(skb_copy_expand);
1351 
1352 /**
1353  *      skb_pad                 -       zero pad the tail of an skb
1354  *      @skb: buffer to pad
1355  *      @pad: space to pad
1356  *
1357  *      Ensure that a buffer is followed by a padding area that is zero
1358  *      filled. Used by network drivers which may DMA or transfer data
1359  *      beyond the buffer end onto the wire.
1360  *
1361  *      May return error in out of memory cases. The skb is freed on error.
1362  */
1363 
1364 int skb_pad(struct sk_buff *skb, int pad)
1365 {
1366         int err;
1367         int ntail;
1368 
1369         /* If the skbuff is non linear tailroom is always zero.. */
1370         if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1371                 memset(skb->data+skb->len, 0, pad);
1372                 return 0;
1373         }
1374 
1375         ntail = skb->data_len + pad - (skb->end - skb->tail);
1376         if (likely(skb_cloned(skb) || ntail > 0)) {
1377                 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1378                 if (unlikely(err))
1379                         goto free_skb;
1380         }
1381 
1382         /* FIXME: The use of this function with non-linear skb's really needs
1383          * to be audited.
1384          */
1385         err = skb_linearize(skb);
1386         if (unlikely(err))
1387                 goto free_skb;
1388 
1389         memset(skb->data + skb->len, 0, pad);
1390         return 0;
1391 
1392 free_skb:
1393         kfree_skb(skb);
1394         return err;
1395 }
1396 EXPORT_SYMBOL(skb_pad);
1397 
1398 /**
1399  *      pskb_put - add data to the tail of a potentially fragmented buffer
1400  *      @skb: start of the buffer to use
1401  *      @tail: tail fragment of the buffer to use
1402  *      @len: amount of data to add
1403  *
1404  *      This function extends the used data area of the potentially
1405  *      fragmented buffer. @tail must be the last fragment of @skb -- or
1406  *      @skb itself. If this would exceed the total buffer size the kernel
1407  *      will panic. A pointer to the first byte of the extra data is
1408  *      returned.
1409  */
1410 
1411 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1412 {
1413         if (tail != skb) {
1414                 skb->data_len += len;
1415                 skb->len += len;
1416         }
1417         return skb_put(tail, len);
1418 }
1419 EXPORT_SYMBOL_GPL(pskb_put);
1420 
1421 /**
1422  *      skb_put - add data to a buffer
1423  *      @skb: buffer to use
1424  *      @len: amount of data to add
1425  *
1426  *      This function extends the used data area of the buffer. If this would
1427  *      exceed the total buffer size the kernel will panic. A pointer to the
1428  *      first byte of the extra data is returned.
1429  */
1430 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1431 {
1432         unsigned char *tmp = skb_tail_pointer(skb);
1433         SKB_LINEAR_ASSERT(skb);
1434         skb->tail += len;
1435         skb->len  += len;
1436         if (unlikely(skb->tail > skb->end))
1437                 skb_over_panic(skb, len, __builtin_return_address(0));
1438         return tmp;
1439 }
1440 EXPORT_SYMBOL(skb_put);
1441 
1442 /**
1443  *      skb_push - add data to the start of a buffer
1444  *      @skb: buffer to use
1445  *      @len: amount of data to add
1446  *
1447  *      This function extends the used data area of the buffer at the buffer
1448  *      start. If this would exceed the total buffer headroom the kernel will
1449  *      panic. A pointer to the first byte of the extra data is returned.
1450  */
1451 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1452 {
1453         skb->data -= len;
1454         skb->len  += len;
1455         if (unlikely(skb->data<skb->head))
1456                 skb_under_panic(skb, len, __builtin_return_address(0));
1457         return skb->data;
1458 }
1459 EXPORT_SYMBOL(skb_push);
1460 
1461 /**
1462  *      skb_pull - remove data from the start of a buffer
1463  *      @skb: buffer to use
1464  *      @len: amount of data to remove
1465  *
1466  *      This function removes data from the start of a buffer, returning
1467  *      the memory to the headroom. A pointer to the next data in the buffer
1468  *      is returned. Once the data has been pulled future pushes will overwrite
1469  *      the old data.
1470  */
1471 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1472 {
1473         return skb_pull_inline(skb, len);
1474 }
1475 EXPORT_SYMBOL(skb_pull);
1476 
1477 /**
1478  *      skb_trim - remove end from a buffer
1479  *      @skb: buffer to alter
1480  *      @len: new length
1481  *
1482  *      Cut the length of a buffer down by removing data from the tail. If
1483  *      the buffer is already under the length specified it is not modified.
1484  *      The skb must be linear.
1485  */
1486 void skb_trim(struct sk_buff *skb, unsigned int len)
1487 {
1488         if (skb->len > len)
1489                 __skb_trim(skb, len);
1490 }
1491 EXPORT_SYMBOL(skb_trim);
1492 
1493 /* Trims skb to length len. It can change skb pointers.
1494  */
1495 
1496 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1497 {
1498         struct sk_buff **fragp;
1499         struct sk_buff *frag;
1500         int offset = skb_headlen(skb);
1501         int nfrags = skb_shinfo(skb)->nr_frags;
1502         int i;
1503         int err;
1504 
1505         if (skb_cloned(skb) &&
1506             unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1507                 return err;
1508 
1509         i = 0;
1510         if (offset >= len)
1511                 goto drop_pages;
1512 
1513         for (; i < nfrags; i++) {
1514                 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1515 
1516                 if (end < len) {
1517                         offset = end;
1518                         continue;
1519                 }
1520 
1521                 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1522 
1523 drop_pages:
1524                 skb_shinfo(skb)->nr_frags = i;
1525 
1526                 for (; i < nfrags; i++)
1527                         skb_frag_unref(skb, i);
1528 
1529                 if (skb_has_frag_list(skb))
1530                         skb_drop_fraglist(skb);
1531                 goto done;
1532         }
1533 
1534         for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1535              fragp = &frag->next) {
1536                 int end = offset + frag->len;
1537 
1538                 if (skb_shared(frag)) {
1539                         struct sk_buff *nfrag;
1540 
1541                         nfrag = skb_clone(frag, GFP_ATOMIC);
1542                         if (unlikely(!nfrag))
1543                                 return -ENOMEM;
1544 
1545                         nfrag->next = frag->next;
1546                         consume_skb(frag);
1547                         frag = nfrag;
1548                         *fragp = frag;
1549                 }
1550 
1551                 if (end < len) {
1552                         offset = end;
1553                         continue;
1554                 }
1555 
1556                 if (end > len &&
1557                     unlikely((err = pskb_trim(frag, len - offset))))
1558                         return err;
1559 
1560                 if (frag->next)
1561                         skb_drop_list(&frag->next);
1562                 break;
1563         }
1564 
1565 done:
1566         if (len > skb_headlen(skb)) {
1567                 skb->data_len -= skb->len - len;
1568                 skb->len       = len;
1569         } else {
1570                 skb->len       = len;
1571                 skb->data_len  = 0;
1572                 skb_set_tail_pointer(skb, len);
1573         }
1574 
1575         return 0;
1576 }
1577 EXPORT_SYMBOL(___pskb_trim);
1578 
1579 /**
1580  *      __pskb_pull_tail - advance tail of skb header
1581  *      @skb: buffer to reallocate
1582  *      @delta: number of bytes to advance tail
1583  *
1584  *      The function makes a sense only on a fragmented &sk_buff,
1585  *      it expands header moving its tail forward and copying necessary
1586  *      data from fragmented part.
1587  *
1588  *      &sk_buff MUST have reference count of 1.
1589  *
1590  *      Returns %NULL (and &sk_buff does not change) if pull failed
1591  *      or value of new tail of skb in the case of success.
1592  *
1593  *      All the pointers pointing into skb header may change and must be
1594  *      reloaded after call to this function.
1595  */
1596 
1597 /* Moves tail of skb head forward, copying data from fragmented part,
1598  * when it is necessary.
1599  * 1. It may fail due to malloc failure.
1600  * 2. It may change skb pointers.
1601  *
1602  * It is pretty complicated. Luckily, it is called only in exceptional cases.
1603  */
1604 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1605 {
1606         /* If skb has not enough free space at tail, get new one
1607          * plus 128 bytes for future expansions. If we have enough
1608          * room at tail, reallocate without expansion only if skb is cloned.
1609          */
1610         int i, k, eat = (skb->tail + delta) - skb->end;
1611 
1612         if (eat > 0 || skb_cloned(skb)) {
1613                 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1614                                      GFP_ATOMIC))
1615                         return NULL;
1616         }
1617 
1618         if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1619                 BUG();
1620 
1621         /* Optimization: no fragments, no reasons to preestimate
1622          * size of pulled pages. Superb.
1623          */
1624         if (!skb_has_frag_list(skb))
1625                 goto pull_pages;
1626 
1627         /* Estimate size of pulled pages. */
1628         eat = delta;
1629         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1630                 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1631 
1632                 if (size >= eat)
1633                         goto pull_pages;
1634                 eat -= size;
1635         }
1636 
1637         /* If we need update frag list, we are in troubles.
1638          * Certainly, it possible to add an offset to skb data,
1639          * but taking into account that pulling is expected to
1640          * be very rare operation, it is worth to fight against
1641          * further bloating skb head and crucify ourselves here instead.
1642          * Pure masohism, indeed. 8)8)
1643          */
1644         if (eat) {
1645                 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1646                 struct sk_buff *clone = NULL;
1647                 struct sk_buff *insp = NULL;
1648 
1649                 do {
1650                         BUG_ON(!list);
1651 
1652                         if (list->len <= eat) {
1653                                 /* Eaten as whole. */
1654                                 eat -= list->len;
1655                                 list = list->next;
1656                                 insp = list;
1657                         } else {
1658                                 /* Eaten partially. */
1659 
1660                                 if (skb_shared(list)) {
1661                                         /* Sucks! We need to fork list. :-( */
1662                                         clone = skb_clone(list, GFP_ATOMIC);
1663                                         if (!clone)
1664                                                 return NULL;
1665                                         insp = list->next;
1666                                         list = clone;
1667                                 } else {
1668                                         /* This may be pulled without
1669                                          * problems. */
1670                                         insp = list;
1671                                 }
1672                                 if (!pskb_pull(list, eat)) {
1673                                         kfree_skb(clone);
1674                                         return NULL;
1675                                 }
1676                                 break;
1677                         }
1678                 } while (eat);
1679 
1680                 /* Free pulled out fragments. */
1681                 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1682                         skb_shinfo(skb)->frag_list = list->next;
1683                         kfree_skb(list);
1684                 }
1685                 /* And insert new clone at head. */
1686                 if (clone) {
1687                         clone->next = list;
1688                         skb_shinfo(skb)->frag_list = clone;
1689                 }
1690         }
1691         /* Success! Now we may commit changes to skb data. */
1692 
1693 pull_pages:
1694         eat = delta;
1695         k = 0;
1696         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1697                 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1698 
1699                 if (size <= eat) {
1700                         skb_frag_unref(skb, i);
1701                         eat -= size;
1702                 } else {
1703                         skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1704                         if (eat) {
1705                                 skb_shinfo(skb)->frags[k].page_offset += eat;
1706                                 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1707                                 eat = 0;
1708                         }
1709                         k++;
1710                 }
1711         }
1712         skb_shinfo(skb)->nr_frags = k;
1713 
1714         skb->tail     += delta;
1715         skb->data_len -= delta;
1716 
1717         return skb_tail_pointer(skb);
1718 }
1719 EXPORT_SYMBOL(__pskb_pull_tail);
1720 
1721 /**
1722  *      skb_copy_bits - copy bits from skb to kernel buffer
1723  *      @skb: source skb
1724  *      @offset: offset in source
1725  *      @to: destination buffer
1726  *      @len: number of bytes to copy
1727  *
1728  *      Copy the specified number of bytes from the source skb to the
1729  *      destination buffer.
1730  *
1731  *      CAUTION ! :
1732  *              If its prototype is ever changed,
1733  *              check arch/{*}/net/{*}.S files,
1734  *              since it is called from BPF assembly code.
1735  */
1736 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1737 {
1738         int start = skb_headlen(skb);
1739         struct sk_buff *frag_iter;
1740         int i, copy;
1741 
1742         if (offset > (int)skb->len - len)
1743                 goto fault;
1744 
1745         /* Copy header. */
1746         if ((copy = start - offset) > 0) {
1747                 if (copy > len)
1748                         copy = len;
1749                 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1750                 if ((len -= copy) == 0)
1751                         return 0;
1752                 offset += copy;
1753                 to     += copy;
1754         }
1755 
1756         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1757                 int end;
1758                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1759 
1760                 WARN_ON(start > offset + len);
1761 
1762                 end = start + skb_frag_size(f);
1763                 if ((copy = end - offset) > 0) {
1764                         u8 *vaddr;
1765 
1766                         if (copy > len)
1767                                 copy = len;
1768 
1769                         vaddr = kmap_atomic(skb_frag_page(f));
1770                         memcpy(to,
1771                                vaddr + f->page_offset + offset - start,
1772                                copy);
1773                         kunmap_atomic(vaddr);
1774 
1775                         if ((len -= copy) == 0)
1776                                 return 0;
1777                         offset += copy;
1778                         to     += copy;
1779                 }
1780                 start = end;
1781         }
1782 
1783         skb_walk_frags(skb, frag_iter) {
1784                 int end;
1785 
1786                 WARN_ON(start > offset + len);
1787 
1788                 end = start + frag_iter->len;
1789                 if ((copy = end - offset) > 0) {
1790                         if (copy > len)
1791                                 copy = len;
1792                         if (skb_copy_bits(frag_iter, offset - start, to, copy))
1793                                 goto fault;
1794                         if ((len -= copy) == 0)
1795                                 return 0;
1796                         offset += copy;
1797                         to     += copy;
1798                 }
1799                 start = end;
1800         }
1801 
1802         if (!len)
1803                 return 0;
1804 
1805 fault:
1806         return -EFAULT;
1807 }
1808 EXPORT_SYMBOL(skb_copy_bits);
1809 
1810 /*
1811  * Callback from splice_to_pipe(), if we need to release some pages
1812  * at the end of the spd in case we error'ed out in filling the pipe.
1813  */
1814 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1815 {
1816         put_page(spd->pages[i]);
1817 }
1818 
1819 static struct page *linear_to_page(struct page *page, unsigned int *len,
1820                                    unsigned int *offset,
1821                                    struct sock *sk)
1822 {
1823         struct page_frag *pfrag = sk_page_frag(sk);
1824 
1825         if (!sk_page_frag_refill(sk, pfrag))
1826                 return NULL;
1827 
1828         *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
1829 
1830         memcpy(page_address(pfrag->page) + pfrag->offset,
1831                page_address(page) + *offset, *len);
1832         *offset = pfrag->offset;
1833         pfrag->offset += *len;
1834 
1835         return pfrag->page;
1836 }
1837 
1838 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1839                              struct page *page,
1840                              unsigned int offset)
1841 {
1842         return  spd->nr_pages &&
1843                 spd->pages[spd->nr_pages - 1] == page &&
1844                 (spd->partial[spd->nr_pages - 1].offset +
1845                  spd->partial[spd->nr_pages - 1].len == offset);
1846 }
1847 
1848 /*
1849  * Fill page/offset/length into spd, if it can hold more pages.
1850  */
1851 static bool spd_fill_page(struct splice_pipe_desc *spd,
1852                           struct pipe_inode_info *pipe, struct page *page,
1853                           unsigned int *len, unsigned int offset,
1854                           bool linear,
1855                           struct sock *sk)
1856 {
1857         if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1858                 return true;
1859 
1860         if (linear) {
1861                 page = linear_to_page(page, len, &offset, sk);
1862                 if (!page)
1863                         return true;
1864         }
1865         if (spd_can_coalesce(spd, page, offset)) {
1866                 spd->partial[spd->nr_pages - 1].len += *len;
1867                 return false;
1868         }
1869         get_page(page);
1870         spd->pages[spd->nr_pages] = page;
1871         spd->partial[spd->nr_pages].len = *len;
1872         spd->partial[spd->nr_pages].offset = offset;
1873         spd->nr_pages++;
1874 
1875         return false;
1876 }
1877 
1878 static bool __splice_segment(struct page *page, unsigned int poff,
1879                              unsigned int plen, unsigned int *off,
1880                              unsigned int *len,
1881                              struct splice_pipe_desc *spd, bool linear,
1882                              struct sock *sk,
1883                              struct pipe_inode_info *pipe)
1884 {
1885         if (!*len)
1886                 return true;
1887 
1888         /* skip this segment if already processed */
1889         if (*off >= plen) {
1890                 *off -= plen;
1891                 return false;
1892         }
1893 
1894         /* ignore any bits we already processed */
1895         poff += *off;
1896         plen -= *off;
1897         *off = 0;
1898 
1899         do {
1900                 unsigned int flen = min(*len, plen);
1901 
1902                 if (spd_fill_page(spd, pipe, page, &flen, poff,
1903                                   linear, sk))
1904                         return true;
1905                 poff += flen;
1906                 plen -= flen;
1907                 *len -= flen;
1908         } while (*len && plen);
1909 
1910         return false;
1911 }
1912 
1913 /*
1914  * Map linear and fragment data from the skb to spd. It reports true if the
1915  * pipe is full or if we already spliced the requested length.
1916  */
1917 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1918                               unsigned int *offset, unsigned int *len,
1919                               struct splice_pipe_desc *spd, struct sock *sk)
1920 {
1921         int seg;
1922         struct sk_buff *iter;
1923 
1924         /* map the linear part :
1925          * If skb->head_frag is set, this 'linear' part is backed by a
1926          * fragment, and if the head is not shared with any clones then
1927          * we can avoid a copy since we own the head portion of this page.
1928          */
1929         if (__splice_segment(virt_to_page(skb->data),
1930                              (unsigned long) skb->data & (PAGE_SIZE - 1),
1931                              skb_headlen(skb),
1932                              offset, len, spd,
1933                              skb_head_is_locked(skb),
1934                              sk, pipe))
1935                 return true;
1936 
1937         /*
1938          * then map the fragments
1939          */
1940         for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1941                 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1942 
1943                 if (__splice_segment(skb_frag_page(f),
1944                                      f->page_offset, skb_frag_size(f),
1945                                      offset, len, spd, false, sk, pipe))
1946                         return true;
1947         }
1948 
1949         skb_walk_frags(skb, iter) {
1950                 if (*offset >= iter->len) {
1951                         *offset -= iter->len;
1952                         continue;
1953                 }
1954                 /* __skb_splice_bits() only fails if the output has no room
1955                  * left, so no point in going over the frag_list for the error
1956                  * case.
1957                  */
1958                 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
1959                         return true;
1960         }
1961 
1962         return false;
1963 }
1964 
1965 /*
1966  * Map data from the skb to a pipe. Should handle both the linear part,
1967  * the fragments, and the frag list.
1968  */
1969 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
1970                     struct pipe_inode_info *pipe, unsigned int tlen,
1971                     unsigned int flags)
1972 {
1973         struct partial_page partial[MAX_SKB_FRAGS];
1974         struct page *pages[MAX_SKB_FRAGS];
1975         struct splice_pipe_desc spd = {
1976                 .pages = pages,
1977                 .partial = partial,
1978                 .nr_pages_max = MAX_SKB_FRAGS,
1979                 .flags = flags,
1980                 .ops = &nosteal_pipe_buf_ops,
1981                 .spd_release = sock_spd_release,
1982         };
1983         int ret = 0;
1984 
1985         __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
1986 
1987         if (spd.nr_pages)
1988                 ret = splice_to_pipe(pipe, &spd);
1989 
1990         return ret;
1991 }
1992 EXPORT_SYMBOL_GPL(skb_splice_bits);
1993 
1994 /**
1995  *      skb_store_bits - store bits from kernel buffer to skb
1996  *      @skb: destination buffer
1997  *      @offset: offset in destination
1998  *      @from: source buffer
1999  *      @len: number of bytes to copy
2000  *
2001  *      Copy the specified number of bytes from the source buffer to the
2002  *      destination skb.  This function handles all the messy bits of
2003  *      traversing fragment lists and such.
2004  */
2005 
2006 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2007 {
2008         int start = skb_headlen(skb);
2009         struct sk_buff *frag_iter;
2010         int i, copy;
2011 
2012         if (offset > (int)skb->len - len)
2013                 goto fault;
2014 
2015         if ((copy = start - offset) > 0) {
2016                 if (copy > len)
2017                         copy = len;
2018                 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2019                 if ((len -= copy) == 0)
2020                         return 0;
2021                 offset += copy;
2022                 from += copy;
2023         }
2024 
2025         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2026                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2027                 int end;
2028 
2029                 WARN_ON(start > offset + len);
2030 
2031                 end = start + skb_frag_size(frag);
2032                 if ((copy = end - offset) > 0) {
2033                         u8 *vaddr;
2034 
2035                         if (copy > len)
2036                                 copy = len;
2037 
2038                         vaddr = kmap_atomic(skb_frag_page(frag));
2039                         memcpy(vaddr + frag->page_offset + offset - start,
2040                                from, copy);
2041                         kunmap_atomic(vaddr);
2042 
2043                         if ((len -= copy) == 0)
2044                                 return 0;
2045                         offset += copy;
2046                         from += copy;
2047                 }
2048                 start = end;
2049         }
2050 
2051         skb_walk_frags(skb, frag_iter) {
2052                 int end;
2053 
2054                 WARN_ON(start > offset + len);
2055 
2056                 end = start + frag_iter->len;
2057                 if ((copy = end - offset) > 0) {
2058                         if (copy > len)
2059                                 copy = len;
2060                         if (skb_store_bits(frag_iter, offset - start,
2061                                            from, copy))
2062                                 goto fault;
2063                         if ((len -= copy) == 0)
2064                                 return 0;
2065                         offset += copy;
2066                         from += copy;
2067                 }
2068                 start = end;
2069         }
2070         if (!len)
2071                 return 0;
2072 
2073 fault:
2074         return -EFAULT;
2075 }
2076 EXPORT_SYMBOL(skb_store_bits);
2077 
2078 /* Checksum skb data. */
2079 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2080                       __wsum csum, const struct skb_checksum_ops *ops)
2081 {
2082         int start = skb_headlen(skb);
2083         int i, copy = start - offset;
2084         struct sk_buff *frag_iter;
2085         int pos = 0;
2086 
2087         /* Checksum header. */
2088         if (copy > 0) {
2089                 if (copy > len)
2090                         copy = len;
2091                 csum = ops->update(skb->data + offset, copy, csum);
2092                 if ((len -= copy) == 0)
2093                         return csum;
2094                 offset += copy;
2095                 pos     = copy;
2096         }
2097 
2098         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2099                 int end;
2100                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2101 
2102                 WARN_ON(start > offset + len);
2103 
2104                 end = start + skb_frag_size(frag);
2105                 if ((copy = end - offset) > 0) {
2106                         __wsum csum2;
2107                         u8 *vaddr;
2108 
2109                         if (copy > len)
2110                                 copy = len;
2111                         vaddr = kmap_atomic(skb_frag_page(frag));
2112                         csum2 = ops->update(vaddr + frag->page_offset +
2113                                             offset - start, copy, 0);
2114                         kunmap_atomic(vaddr);
2115                         csum = ops->combine(csum, csum2, pos, copy);
2116                         if (!(len -= copy))
2117                                 return csum;
2118                         offset += copy;
2119                         pos    += copy;
2120                 }
2121                 start = end;
2122         }
2123 
2124         skb_walk_frags(skb, frag_iter) {
2125                 int end;
2126 
2127                 WARN_ON(start > offset + len);
2128 
2129                 end = start + frag_iter->len;
2130                 if ((copy = end - offset) > 0) {
2131                         __wsum csum2;
2132                         if (copy > len)
2133                                 copy = len;
2134                         csum2 = __skb_checksum(frag_iter, offset - start,
2135                                                copy, 0, ops);
2136                         csum = ops->combine(csum, csum2, pos, copy);
2137                         if ((len -= copy) == 0)
2138                                 return csum;
2139                         offset += copy;
2140                         pos    += copy;
2141                 }
2142                 start = end;
2143         }
2144         BUG_ON(len);
2145 
2146         return csum;
2147 }
2148 EXPORT_SYMBOL(__skb_checksum);
2149 
2150 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2151                     int len, __wsum csum)
2152 {
2153         const struct skb_checksum_ops ops = {
2154                 .update  = csum_partial_ext,
2155                 .combine = csum_block_add_ext,
2156         };
2157 
2158         return __skb_checksum(skb, offset, len, csum, &ops);
2159 }
2160 EXPORT_SYMBOL(skb_checksum);
2161 
2162 /* Both of above in one bottle. */
2163 
2164 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2165                                     u8 *to, int len, __wsum csum)
2166 {
2167         int start = skb_headlen(skb);
2168         int i, copy = start - offset;
2169         struct sk_buff *frag_iter;
2170         int pos = 0;
2171 
2172         /* Copy header. */
2173         if (copy > 0) {
2174                 if (copy > len)
2175                         copy = len;
2176                 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2177                                                  copy, csum);
2178                 if ((len -= copy) == 0)
2179                         return csum;
2180                 offset += copy;
2181                 to     += copy;
2182                 pos     = copy;
2183         }
2184 
2185         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2186                 int end;
2187 
2188                 WARN_ON(start > offset + len);
2189 
2190                 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2191                 if ((copy = end - offset) > 0) {
2192                         __wsum csum2;
2193                         u8 *vaddr;
2194                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2195 
2196                         if (copy > len)
2197                                 copy = len;
2198                         vaddr = kmap_atomic(skb_frag_page(frag));
2199                         csum2 = csum_partial_copy_nocheck(vaddr +
2200                                                           frag->page_offset +
2201                                                           offset - start, to,
2202                                                           copy, 0);
2203                         kunmap_atomic(vaddr);
2204                         csum = csum_block_add(csum, csum2, pos);
2205                         if (!(len -= copy))
2206                                 return csum;
2207                         offset += copy;
2208                         to     += copy;
2209                         pos    += copy;
2210                 }
2211                 start = end;
2212         }
2213 
2214         skb_walk_frags(skb, frag_iter) {
2215                 __wsum csum2;
2216                 int end;
2217 
2218                 WARN_ON(start > offset + len);
2219 
2220                 end = start + frag_iter->len;
2221                 if ((copy = end - offset) > 0) {
2222                         if (copy > len)
2223                                 copy = len;
2224                         csum2 = skb_copy_and_csum_bits(frag_iter,
2225                                                        offset - start,
2226                                                        to, copy, 0);
2227                         csum = csum_block_add(csum, csum2, pos);
2228                         if ((len -= copy) == 0)
2229                                 return csum;
2230                         offset += copy;
2231                         to     += copy;
2232                         pos    += copy;
2233                 }
2234                 start = end;
2235         }
2236         BUG_ON(len);
2237         return csum;
2238 }
2239 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2240 
2241  /**
2242  *      skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2243  *      @from: source buffer
2244  *
2245  *      Calculates the amount of linear headroom needed in the 'to' skb passed
2246  *      into skb_zerocopy().
2247  */
2248 unsigned int
2249 skb_zerocopy_headlen(const struct sk_buff *from)
2250 {
2251         unsigned int hlen = 0;
2252 
2253         if (!from->head_frag ||
2254             skb_headlen(from) < L1_CACHE_BYTES ||
2255             skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2256                 hlen = skb_headlen(from);
2257 
2258         if (skb_has_frag_list(from))
2259                 hlen = from->len;
2260 
2261         return hlen;
2262 }
2263 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2264 
2265 /**
2266  *      skb_zerocopy - Zero copy skb to skb
2267  *      @to: destination buffer
2268  *      @from: source buffer
2269  *      @len: number of bytes to copy from source buffer
2270  *      @hlen: size of linear headroom in destination buffer
2271  *
2272  *      Copies up to `len` bytes from `from` to `to` by creating references
2273  *      to the frags in the source buffer.
2274  *
2275  *      The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2276  *      headroom in the `to` buffer.
2277  *
2278  *      Return value:
2279  *      0: everything is OK
2280  *      -ENOMEM: couldn't orphan frags of @from due to lack of memory
2281  *      -EFAULT: skb_copy_bits() found some problem with skb geometry
2282  */
2283 int
2284 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2285 {
2286         int i, j = 0;
2287         int plen = 0; /* length of skb->head fragment */
2288         int ret;
2289         struct page *page;
2290         unsigned int offset;
2291 
2292         BUG_ON(!from->head_frag && !hlen);
2293 
2294         /* dont bother with small payloads */
2295         if (len <= skb_tailroom(to))
2296                 return skb_copy_bits(from, 0, skb_put(to, len), len);
2297 
2298         if (hlen) {
2299                 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2300                 if (unlikely(ret))
2301                         return ret;
2302                 len -= hlen;
2303         } else {
2304                 plen = min_t(int, skb_headlen(from), len);
2305                 if (plen) {
2306                         page = virt_to_head_page(from->head);
2307                         offset = from->data - (unsigned char *)page_address(page);
2308                         __skb_fill_page_desc(to, 0, page, offset, plen);
2309                         get_page(page);
2310                         j = 1;
2311                         len -= plen;
2312                 }
2313         }
2314 
2315         to->truesize += len + plen;
2316         to->len += len + plen;
2317         to->data_len += len + plen;
2318 
2319         if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2320                 skb_tx_error(from);
2321                 return -ENOMEM;
2322         }
2323 
2324         for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2325                 if (!len)
2326                         break;
2327                 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2328                 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
2329                 len -= skb_shinfo(to)->frags[j].size;
2330                 skb_frag_ref(to, j);
2331                 j++;
2332         }
2333         skb_shinfo(to)->nr_frags = j;
2334 
2335         return 0;
2336 }
2337 EXPORT_SYMBOL_GPL(skb_zerocopy);
2338 
2339 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2340 {
2341         __wsum csum;
2342         long csstart;
2343 
2344         if (skb->ip_summed == CHECKSUM_PARTIAL)
2345                 csstart = skb_checksum_start_offset(skb);
2346         else
2347                 csstart = skb_headlen(skb);
2348 
2349         BUG_ON(csstart > skb_headlen(skb));
2350 
2351         skb_copy_from_linear_data(skb, to, csstart);
2352 
2353         csum = 0;
2354         if (csstart != skb->len)
2355                 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2356                                               skb->len - csstart, 0);
2357 
2358         if (skb->ip_summed == CHECKSUM_PARTIAL) {
2359                 long csstuff = csstart + skb->csum_offset;
2360 
2361                 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2362         }
2363 }
2364 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2365 
2366 /**
2367  *      skb_dequeue - remove from the head of the queue
2368  *      @list: list to dequeue from
2369  *
2370  *      Remove the head of the list. The list lock is taken so the function
2371  *      may be used safely with other locking list functions. The head item is
2372  *      returned or %NULL if the list is empty.
2373  */
2374 
2375 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2376 {
2377         unsigned long flags;
2378         struct sk_buff *result;
2379 
2380         spin_lock_irqsave(&list->lock, flags);
2381         result = __skb_dequeue(list);
2382         spin_unlock_irqrestore(&list->lock, flags);
2383         return result;
2384 }
2385 EXPORT_SYMBOL(skb_dequeue);
2386 
2387 /**
2388  *      skb_dequeue_tail - remove from the tail of the queue
2389  *      @list: list to dequeue from
2390  *
2391  *      Remove the tail of the list. The list lock is taken so the function
2392  *      may be used safely with other locking list functions. The tail item is
2393  *      returned or %NULL if the list is empty.
2394  */
2395 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2396 {
2397         unsigned long flags;
2398         struct sk_buff *result;
2399 
2400         spin_lock_irqsave(&list->lock, flags);
2401         result = __skb_dequeue_tail(list);
2402         spin_unlock_irqrestore(&list->lock, flags);
2403         return result;
2404 }
2405 EXPORT_SYMBOL(skb_dequeue_tail);
2406 
2407 /**
2408  *      skb_queue_purge - empty a list
2409  *      @list: list to empty
2410  *
2411  *      Delete all buffers on an &sk_buff list. Each buffer is removed from
2412  *      the list and one reference dropped. This function takes the list
2413  *      lock and is atomic with respect to other list locking functions.
2414  */
2415 void skb_queue_purge(struct sk_buff_head *list)
2416 {
2417         struct sk_buff *skb;
2418         while ((skb = skb_dequeue(list)) != NULL)
2419                 kfree_skb(skb);
2420 }
2421 EXPORT_SYMBOL(skb_queue_purge);
2422 
2423 /**
2424  *      skb_rbtree_purge - empty a skb rbtree
2425  *      @root: root of the rbtree to empty
2426  *
2427  *      Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
2428  *      the list and one reference dropped. This function does not take
2429  *      any lock. Synchronization should be handled by the caller (e.g., TCP
2430  *      out-of-order queue is protected by the socket lock).
2431  */
2432 void skb_rbtree_purge(struct rb_root *root)
2433 {
2434         struct sk_buff *skb, *next;
2435 
2436         rbtree_postorder_for_each_entry_safe(skb, next, root, rbnode)
2437                 kfree_skb(skb);
2438 
2439         *root = RB_ROOT;
2440 }
2441 
2442 /**
2443  *      skb_queue_head - queue a buffer at the list head
2444  *      @list: list to use
2445  *      @newsk: buffer to queue
2446  *
2447  *      Queue a buffer at the start of the list. This function takes the
2448  *      list lock and can be used safely with other locking &sk_buff functions
2449  *      safely.
2450  *
2451  *      A buffer cannot be placed on two lists at the same time.
2452  */
2453 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2454 {
2455         unsigned long flags;
2456 
2457         spin_lock_irqsave(&list->lock, flags);
2458         __skb_queue_head(list, newsk);
2459         spin_unlock_irqrestore(&list->lock, flags);
2460 }
2461 EXPORT_SYMBOL(skb_queue_head);
2462 
2463 /**
2464  *      skb_queue_tail - queue a buffer at the list tail
2465  *      @list: list to use
2466  *      @newsk: buffer to queue
2467  *
2468  *      Queue a buffer at the tail of the list. This function takes the
2469  *      list lock and can be used safely with other locking &sk_buff functions
2470  *      safely.
2471  *
2472  *      A buffer cannot be placed on two lists at the same time.
2473  */
2474 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2475 {
2476         unsigned long flags;
2477 
2478         spin_lock_irqsave(&list->lock, flags);
2479         __skb_queue_tail(list, newsk);
2480         spin_unlock_irqrestore(&list->lock, flags);
2481 }
2482 EXPORT_SYMBOL(skb_queue_tail);
2483 
2484 /**
2485  *      skb_unlink      -       remove a buffer from a list
2486  *      @skb: buffer to remove
2487  *      @list: list to use
2488  *
2489  *      Remove a packet from a list. The list locks are taken and this
2490  *      function is atomic with respect to other list locked calls
2491  *
2492  *      You must know what list the SKB is on.
2493  */
2494 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2495 {
2496         unsigned long flags;
2497 
2498         spin_lock_irqsave(&list->lock, flags);
2499         __skb_unlink(skb, list);
2500         spin_unlock_irqrestore(&list->lock, flags);
2501 }
2502 EXPORT_SYMBOL(skb_unlink);
2503 
2504 /**
2505  *      skb_append      -       append a buffer
2506  *      @old: buffer to insert after
2507  *      @newsk: buffer to insert
2508  *      @list: list to use
2509  *
2510  *      Place a packet after a given packet in a list. The list locks are taken
2511  *      and this function is atomic with respect to other list locked calls.
2512  *      A buffer cannot be placed on two lists at the same time.
2513  */
2514 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2515 {
2516         unsigned long flags;
2517 
2518         spin_lock_irqsave(&list->lock, flags);
2519         __skb_queue_after(list, old, newsk);
2520         spin_unlock_irqrestore(&list->lock, flags);
2521 }
2522 EXPORT_SYMBOL(skb_append);
2523 
2524 /**
2525  *      skb_insert      -       insert a buffer
2526  *      @old: buffer to insert before
2527  *      @newsk: buffer to insert
2528  *      @list: list to use
2529  *
2530  *      Place a packet before a given packet in a list. The list locks are
2531  *      taken and this function is atomic with respect to other list locked
2532  *      calls.
2533  *
2534  *      A buffer cannot be placed on two lists at the same time.
2535  */
2536 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2537 {
2538         unsigned long flags;
2539 
2540         spin_lock_irqsave(&list->lock, flags);
2541         __skb_insert(newsk, old->prev, old, list);
2542         spin_unlock_irqrestore(&list->lock, flags);
2543 }
2544 EXPORT_SYMBOL(skb_insert);
2545 
2546 static inline void skb_split_inside_header(struct sk_buff *skb,
2547                                            struct sk_buff* skb1,
2548                                            const u32 len, const int pos)
2549 {
2550         int i;
2551 
2552         skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2553                                          pos - len);
2554         /* And move data appendix as is. */
2555         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2556                 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2557 
2558         skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2559         skb_shinfo(skb)->nr_frags  = 0;
2560         skb1->data_len             = skb->data_len;
2561         skb1->len                  += skb1->data_len;
2562         skb->data_len              = 0;
2563         skb->len                   = len;
2564         skb_set_tail_pointer(skb, len);
2565 }
2566 
2567 static inline void skb_split_no_header(struct sk_buff *skb,
2568                                        struct sk_buff* skb1,
2569                                        const u32 len, int pos)
2570 {
2571         int i, k = 0;
2572         const int nfrags = skb_shinfo(skb)->nr_frags;
2573 
2574         skb_shinfo(skb)->nr_frags = 0;
2575         skb1->len                 = skb1->data_len = skb->len - len;
2576         skb->len                  = len;
2577         skb->data_len             = len - pos;
2578 
2579         for (i = 0; i < nfrags; i++) {
2580                 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2581 
2582                 if (pos + size > len) {
2583                         skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2584 
2585                         if (pos < len) {
2586                                 /* Split frag.
2587                                  * We have two variants in this case:
2588                                  * 1. Move all the frag to the second
2589                                  *    part, if it is possible. F.e.
2590                                  *    this approach is mandatory for TUX,
2591                                  *    where splitting is expensive.
2592                                  * 2. Split is accurately. We make this.
2593                                  */
2594                                 skb_frag_ref(skb, i);
2595                                 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2596                                 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2597                                 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2598                                 skb_shinfo(skb)->nr_frags++;
2599                         }
2600                         k++;
2601                 } else
2602                         skb_shinfo(skb)->nr_frags++;
2603                 pos += size;
2604         }
2605         skb_shinfo(skb1)->nr_frags = k;
2606 }
2607 
2608 /**
2609  * skb_split - Split fragmented skb to two parts at length len.
2610  * @skb: the buffer to split
2611  * @skb1: the buffer to receive the second part
2612  * @len: new length for skb
2613  */
2614 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2615 {
2616         int pos = skb_headlen(skb);
2617 
2618         skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2619         if (len < pos)  /* Split line is inside header. */
2620                 skb_split_inside_header(skb, skb1, len, pos);
2621         else            /* Second chunk has no header, nothing to copy. */
2622                 skb_split_no_header(skb, skb1, len, pos);
2623 }
2624 EXPORT_SYMBOL(skb_split);
2625 
2626 /* Shifting from/to a cloned skb is a no-go.
2627  *
2628  * Caller cannot keep skb_shinfo related pointers past calling here!
2629  */
2630 static int skb_prepare_for_shift(struct sk_buff *skb)
2631 {
2632         return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2633 }
2634 
2635 /**
2636  * skb_shift - Shifts paged data partially from skb to another
2637  * @tgt: buffer into which tail data gets added
2638  * @skb: buffer from which the paged data comes from
2639  * @shiftlen: shift up to this many bytes
2640  *
2641  * Attempts to shift up to shiftlen worth of bytes, which may be less than
2642  * the length of the skb, from skb to tgt. Returns number bytes shifted.
2643  * It's up to caller to free skb if everything was shifted.
2644  *
2645  * If @tgt runs out of frags, the whole operation is aborted.
2646  *
2647  * Skb cannot include anything else but paged data while tgt is allowed
2648  * to have non-paged data as well.
2649  *
2650  * TODO: full sized shift could be optimized but that would need
2651  * specialized skb free'er to handle frags without up-to-date nr_frags.
2652  */
2653 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2654 {
2655         int from, to, merge, todo;
2656         struct skb_frag_struct *fragfrom, *fragto;
2657 
2658         BUG_ON(shiftlen > skb->len);
2659 
2660         if (skb_headlen(skb))
2661                 return 0;
2662 
2663         todo = shiftlen;
2664         from = 0;
2665         to = skb_shinfo(tgt)->nr_frags;
2666         fragfrom = &skb_shinfo(skb)->frags[from];
2667 
2668         /* Actual merge is delayed until the point when we know we can
2669          * commit all, so that we don't have to undo partial changes
2670          */
2671         if (!to ||
2672             !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2673                               fragfrom->page_offset)) {
2674                 merge = -1;
2675         } else {
2676                 merge = to - 1;
2677 
2678                 todo -= skb_frag_size(fragfrom);
2679                 if (todo < 0) {
2680                         if (skb_prepare_for_shift(skb) ||
2681                             skb_prepare_for_shift(tgt))
2682                                 return 0;
2683 
2684                         /* All previous frag pointers might be stale! */
2685                         fragfrom = &skb_shinfo(skb)->frags[from];
2686                         fragto = &skb_shinfo(tgt)->frags[merge];
2687 
2688                         skb_frag_size_add(fragto, shiftlen);
2689                         skb_frag_size_sub(fragfrom, shiftlen);
2690                         fragfrom->page_offset += shiftlen;
2691 
2692                         goto onlymerged;
2693                 }
2694 
2695                 from++;
2696         }
2697 
2698         /* Skip full, not-fitting skb to avoid expensive operations */
2699         if ((shiftlen == skb->len) &&
2700             (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2701                 return 0;
2702 
2703         if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2704                 return 0;
2705 
2706         while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2707                 if (to == MAX_SKB_FRAGS)
2708                         return 0;
2709 
2710                 fragfrom = &skb_shinfo(skb)->frags[from];
2711                 fragto = &skb_shinfo(tgt)->frags[to];
2712 
2713                 if (todo >= skb_frag_size(fragfrom)) {
2714                         *fragto = *fragfrom;
2715                         todo -= skb_frag_size(fragfrom);
2716                         from++;
2717                         to++;
2718 
2719                 } else {
2720                         __skb_frag_ref(fragfrom);
2721                         fragto->page = fragfrom->page;
2722                         fragto->page_offset = fragfrom->page_offset;
2723                         skb_frag_size_set(fragto, todo);
2724 
2725                         fragfrom->page_offset += todo;
2726                         skb_frag_size_sub(fragfrom, todo);
2727                         todo = 0;
2728 
2729                         to++;
2730                         break;
2731                 }
2732         }
2733 
2734         /* Ready to "commit" this state change to tgt */
2735         skb_shinfo(tgt)->nr_frags = to;
2736 
2737         if (merge >= 0) {
2738                 fragfrom = &skb_shinfo(skb)->frags[0];
2739                 fragto = &skb_shinfo(tgt)->frags[merge];
2740 
2741                 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2742                 __skb_frag_unref(fragfrom);
2743         }
2744 
2745         /* Reposition in the original skb */
2746         to = 0;
2747         while (from < skb_shinfo(skb)->nr_frags)
2748                 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2749         skb_shinfo(skb)->nr_frags = to;
2750 
2751         BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2752 
2753 onlymerged:
2754         /* Most likely the tgt won't ever need its checksum anymore, skb on
2755          * the other hand might need it if it needs to be resent
2756          */
2757         tgt->ip_summed = CHECKSUM_PARTIAL;
2758         skb->ip_summed = CHECKSUM_PARTIAL;
2759 
2760         /* Yak, is it really working this way? Some helper please? */
2761         skb->len -= shiftlen;
2762         skb->data_len -= shiftlen;
2763         skb->truesize -= shiftlen;
2764         tgt->len += shiftlen;
2765         tgt->data_len += shiftlen;
2766         tgt->truesize += shiftlen;
2767 
2768         return shiftlen;
2769 }
2770 
2771 /**
2772  * skb_prepare_seq_read - Prepare a sequential read of skb data
2773  * @skb: the buffer to read
2774  * @from: lower offset of data to be read
2775  * @to: upper offset of data to be read
2776  * @st: state variable
2777  *
2778  * Initializes the specified state variable. Must be called before
2779  * invoking skb_seq_read() for the first time.
2780  */
2781 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2782                           unsigned int to, struct skb_seq_state *st)
2783 {
2784         st->lower_offset = from;
2785         st->upper_offset = to;
2786         st->root_skb = st->cur_skb = skb;
2787         st->frag_idx = st->stepped_offset = 0;
2788         st->frag_data = NULL;
2789 }
2790 EXPORT_SYMBOL(skb_prepare_seq_read);
2791 
2792 /**
2793  * skb_seq_read - Sequentially read skb data
2794  * @consumed: number of bytes consumed by the caller so far
2795  * @data: destination pointer for data to be returned
2796  * @st: state variable
2797  *
2798  * Reads a block of skb data at @consumed relative to the
2799  * lower offset specified to skb_prepare_seq_read(). Assigns
2800  * the head of the data block to @data and returns the length
2801  * of the block or 0 if the end of the skb data or the upper
2802  * offset has been reached.
2803  *
2804  * The caller is not required to consume all of the data
2805  * returned, i.e. @consumed is typically set to the number
2806  * of bytes already consumed and the next call to
2807  * skb_seq_read() will return the remaining part of the block.
2808  *
2809  * Note 1: The size of each block of data returned can be arbitrary,
2810  *       this limitation is the cost for zerocopy sequential
2811  *       reads of potentially non linear data.
2812  *
2813  * Note 2: Fragment lists within fragments are not implemented
2814  *       at the moment, state->root_skb could be replaced with
2815  *       a stack for this purpose.
2816  */
2817 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2818                           struct skb_seq_state *st)
2819 {
2820         unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2821         skb_frag_t *frag;
2822 
2823         if (unlikely(abs_offset >= st->upper_offset)) {
2824                 if (st->frag_data) {
2825                         kunmap_atomic(st->frag_data);
2826                         st->frag_data = NULL;
2827                 }
2828                 return 0;
2829         }
2830 
2831 next_skb:
2832         block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2833 
2834         if (abs_offset < block_limit && !st->frag_data) {
2835                 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2836                 return block_limit - abs_offset;
2837         }
2838 
2839         if (st->frag_idx == 0 && !st->frag_data)
2840                 st->stepped_offset += skb_headlen(st->cur_skb);
2841 
2842         while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2843                 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2844                 block_limit = skb_frag_size(frag) + st->stepped_offset;
2845 
2846                 if (abs_offset < block_limit) {
2847                         if (!st->frag_data)
2848                                 st->frag_data = kmap_atomic(skb_frag_page(frag));
2849 
2850                         *data = (u8 *) st->frag_data + frag->page_offset +
2851                                 (abs_offset - st->stepped_offset);
2852 
2853                         return block_limit - abs_offset;
2854                 }
2855 
2856                 if (st->frag_data) {
2857                         kunmap_atomic(st->frag_data);
2858                         st->frag_data = NULL;
2859                 }
2860 
2861                 st->frag_idx++;
2862                 st->stepped_offset += skb_frag_size(frag);
2863         }
2864 
2865         if (st->frag_data) {
2866                 kunmap_atomic(st->frag_data);
2867                 st->frag_data = NULL;
2868         }
2869 
2870         if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2871                 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2872                 st->frag_idx = 0;
2873                 goto next_skb;
2874         } else if (st->cur_skb->next) {
2875                 st->cur_skb = st->cur_skb->next;
2876                 st->frag_idx = 0;
2877                 goto next_skb;
2878         }
2879 
2880         return 0;
2881 }
2882 EXPORT_SYMBOL(skb_seq_read);
2883 
2884 /**
2885  * skb_abort_seq_read - Abort a sequential read of skb data
2886  * @st: state variable
2887  *
2888  * Must be called if skb_seq_read() was not called until it
2889  * returned 0.
2890  */
2891 void skb_abort_seq_read(struct skb_seq_state *st)
2892 {
2893         if (st->frag_data)
2894                 kunmap_atomic(st->frag_data);
2895 }
2896 EXPORT_SYMBOL(skb_abort_seq_read);
2897 
2898 #define TS_SKB_CB(state)        ((struct skb_seq_state *) &((state)->cb))
2899 
2900 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2901                                           struct ts_config *conf,
2902                                           struct ts_state *state)
2903 {
2904         return skb_seq_read(offset, text, TS_SKB_CB(state));
2905 }
2906 
2907 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2908 {
2909         skb_abort_seq_read(TS_SKB_CB(state));
2910 }
2911 
2912 /**
2913  * skb_find_text - Find a text pattern in skb data
2914  * @skb: the buffer to look in
2915  * @from: search offset
2916  * @to: search limit
2917  * @config: textsearch configuration
2918  *
2919  * Finds a pattern in the skb data according to the specified
2920  * textsearch configuration. Use textsearch_next() to retrieve
2921  * subsequent occurrences of the pattern. Returns the offset
2922  * to the first occurrence or UINT_MAX if no match was found.
2923  */
2924 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2925                            unsigned int to, struct ts_config *config)
2926 {
2927         struct ts_state state;
2928         unsigned int ret;
2929 
2930         config->get_next_block = skb_ts_get_next_block;
2931         config->finish = skb_ts_finish;
2932 
2933         skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
2934 
2935         ret = textsearch_find(config, &state);
2936         return (ret <= to - from ? ret : UINT_MAX);
2937 }
2938 EXPORT_SYMBOL(skb_find_text);
2939 
2940 /**
2941  * skb_append_datato_frags - append the user data to a skb
2942  * @sk: sock  structure
2943  * @skb: skb structure to be appended with user data.
2944  * @getfrag: call back function to be used for getting the user data
2945  * @from: pointer to user message iov
2946  * @length: length of the iov message
2947  *
2948  * Description: This procedure append the user data in the fragment part
2949  * of the skb if any page alloc fails user this procedure returns  -ENOMEM
2950  */
2951 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2952                         int (*getfrag)(void *from, char *to, int offset,
2953                                         int len, int odd, struct sk_buff *skb),
2954                         void *from, int length)
2955 {
2956         int frg_cnt = skb_shinfo(skb)->nr_frags;
2957         int copy;
2958         int offset = 0;
2959         int ret;
2960         struct page_frag *pfrag = &current->task_frag;
2961 
2962         do {
2963                 /* Return error if we don't have space for new frag */
2964                 if (frg_cnt >= MAX_SKB_FRAGS)
2965                         return -EMSGSIZE;
2966 
2967                 if (!sk_page_frag_refill(sk, pfrag))
2968                         return -ENOMEM;
2969 
2970                 /* copy the user data to page */
2971                 copy = min_t(int, length, pfrag->size - pfrag->offset);
2972 
2973                 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
2974                               offset, copy, 0, skb);
2975                 if (ret < 0)
2976                         return -EFAULT;
2977 
2978                 /* copy was successful so update the size parameters */
2979                 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
2980                                    copy);
2981                 frg_cnt++;
2982                 pfrag->offset += copy;
2983                 get_page(pfrag->page);
2984 
2985                 skb->truesize += copy;
2986                 atomic_add(copy, &sk->sk_wmem_alloc);
2987                 skb->len += copy;
2988                 skb->data_len += copy;
2989                 offset += copy;
2990                 length -= copy;
2991 
2992         } while (length > 0);
2993 
2994         return 0;
2995 }
2996 EXPORT_SYMBOL(skb_append_datato_frags);
2997 
2998 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
2999                          int offset, size_t size)
3000 {
3001         int i = skb_shinfo(skb)->nr_frags;
3002 
3003         if (skb_can_coalesce(skb, i, page, offset)) {
3004                 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3005         } else if (i < MAX_SKB_FRAGS) {
3006                 get_page(page);
3007                 skb_fill_page_desc(skb, i, page, offset, size);
3008         } else {
3009                 return -EMSGSIZE;
3010         }
3011 
3012         return 0;
3013 }
3014 EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3015 
3016 /**
3017  *      skb_pull_rcsum - pull skb and update receive checksum
3018  *      @skb: buffer to update
3019  *      @len: length of data pulled
3020  *
3021  *      This function performs an skb_pull on the packet and updates
3022  *      the CHECKSUM_COMPLETE checksum.  It should be used on
3023  *      receive path processing instead of skb_pull unless you know
3024  *      that the checksum difference is zero (e.g., a valid IP header)
3025  *      or you are setting ip_summed to CHECKSUM_NONE.
3026  */
3027 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3028 {
3029         unsigned char *data = skb->data;
3030 
3031         BUG_ON(len > skb->len);
3032         __skb_pull(skb, len);
3033         skb_postpull_rcsum(skb, data, len);
3034         return skb->data;
3035 }
3036 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3037 
3038 /**
3039  *      skb_segment - Perform protocol segmentation on skb.
3040  *      @head_skb: buffer to segment
3041  *      @features: features for the output path (see dev->features)
3042  *
3043  *      This function performs segmentation on the given skb.  It returns
3044  *      a pointer to the first in a list of new skbs for the segments.
3045  *      In case of error it returns ERR_PTR(err).
3046  */
3047 struct sk_buff *skb_segment(struct sk_buff *head_skb,
3048                             netdev_features_t features)
3049 {
3050         struct sk_buff *segs = NULL;
3051         struct sk_buff *tail = NULL;
3052         struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3053         skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3054         unsigned int mss = skb_shinfo(head_skb)->gso_size;
3055         unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3056         struct sk_buff *frag_skb = head_skb;
3057         unsigned int offset = doffset;
3058         unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3059         unsigned int partial_segs = 0;
3060         unsigned int headroom;
3061         unsigned int len = head_skb->len;
3062         __be16 proto;
3063         bool csum, sg;
3064         int nfrags = skb_shinfo(head_skb)->nr_frags;
3065         int err = -ENOMEM;
3066         int i = 0;
3067         int pos;
3068         int dummy;
3069 
3070         __skb_push(head_skb, doffset);
3071         proto = skb_network_protocol(head_skb, &dummy);
3072         if (unlikely(!proto))
3073                 return ERR_PTR(-EINVAL);
3074 
3075         sg = !!(features & NETIF_F_SG);
3076         csum = !!can_checksum_protocol(features, proto);
3077 
3078         if (sg && csum && (mss != GSO_BY_FRAGS))  {
3079                 if (!(features & NETIF_F_GSO_PARTIAL)) {
3080                         struct sk_buff *iter;
3081 
3082                         if (!list_skb ||
3083                             !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
3084                                 goto normal;
3085 
3086                         /* Split the buffer at the frag_list pointer.
3087                          * This is based on the assumption that all
3088                          * buffers in the chain excluding the last
3089                          * containing the same amount of data.
3090                          */
3091                         skb_walk_frags(head_skb, iter) {
3092                                 if (skb_headlen(iter))
3093                                         goto normal;
3094 
3095                                 len -= iter->len;
3096                         }
3097                 }
3098 
3099                 /* GSO partial only requires that we trim off any excess that
3100                  * doesn't fit into an MSS sized block, so take care of that
3101                  * now.
3102                  */
3103                 partial_segs = len / mss;
3104                 if (partial_segs > 1)
3105                         mss *= partial_segs;
3106                 else
3107                         partial_segs = 0;
3108         }
3109 
3110 normal:
3111         headroom = skb_headroom(head_skb);
3112         pos = skb_headlen(head_skb);
3113 
3114         do {
3115                 struct sk_buff *nskb;
3116                 skb_frag_t *nskb_frag;
3117                 int hsize;
3118                 int size;
3119 
3120                 if (unlikely(mss == GSO_BY_FRAGS)) {
3121                         len = list_skb->len;
3122                 } else {
3123                         len = head_skb->len - offset;
3124                         if (len > mss)
3125                                 len = mss;
3126                 }
3127 
3128                 hsize = skb_headlen(head_skb) - offset;
3129                 if (hsize < 0)
3130                         hsize = 0;
3131                 if (hsize > len || !sg)
3132                         hsize = len;
3133 
3134                 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3135                     (skb_headlen(list_skb) == len || sg)) {
3136                         BUG_ON(skb_headlen(list_skb) > len);
3137 
3138                         i = 0;
3139                         nfrags = skb_shinfo(list_skb)->nr_frags;
3140                         frag = skb_shinfo(list_skb)->frags;
3141                         frag_skb = list_skb;
3142                         pos += skb_headlen(list_skb);
3143 
3144                         while (pos < offset + len) {
3145                                 BUG_ON(i >= nfrags);
3146 
3147                                 size = skb_frag_size(frag);
3148                                 if (pos + size > offset + len)
3149                                         break;
3150 
3151                                 i++;
3152                                 pos += size;
3153                                 frag++;
3154                         }
3155 
3156                         nskb = skb_clone(list_skb, GFP_ATOMIC);
3157                         list_skb = list_skb->next;
3158 
3159                         if (unlikely(!nskb))
3160                                 goto err;
3161 
3162                         if (unlikely(pskb_trim(nskb, len))) {
3163                                 kfree_skb(nskb);
3164                                 goto err;
3165                         }
3166 
3167                         hsize = skb_end_offset(nskb);
3168                         if (skb_cow_head(nskb, doffset + headroom)) {
3169                                 kfree_skb(nskb);
3170                                 goto err;
3171                         }
3172 
3173                         nskb->truesize += skb_end_offset(nskb) - hsize;
3174                         skb_release_head_state(nskb);
3175                         __skb_push(nskb, doffset);
3176                 } else {
3177                         nskb = __alloc_skb(hsize + doffset + headroom,
3178                                            GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3179                                            NUMA_NO_NODE);
3180 
3181                         if (unlikely(!nskb))
3182                                 goto err;
3183 
3184                         skb_reserve(nskb, headroom);
3185                         __skb_put(nskb, doffset);
3186                 }
3187 
3188                 if (segs)
3189                         tail->next = nskb;
3190                 else
3191                         segs = nskb;
3192                 tail = nskb;
3193 
3194                 __copy_skb_header(nskb, head_skb);
3195 
3196                 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3197                 skb_reset_mac_len(nskb);
3198 
3199                 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3200                                                  nskb->data - tnl_hlen,
3201                                                  doffset + tnl_hlen);
3202 
3203                 if (nskb->len == len + doffset)
3204                         goto perform_csum_check;
3205 
3206                 if (!sg) {
3207                         if (!nskb->remcsum_offload)
3208                                 nskb->ip_summed = CHECKSUM_NONE;
3209                         SKB_GSO_CB(nskb)->csum =
3210                                 skb_copy_and_csum_bits(head_skb, offset,
3211                                                        skb_put(nskb, len),
3212                                                        len, 0);
3213                         SKB_GSO_CB(nskb)->csum_start =
3214                                 skb_headroom(nskb) + doffset;
3215                         continue;
3216                 }
3217 
3218                 nskb_frag = skb_shinfo(nskb)->frags;
3219 
3220                 skb_copy_from_linear_data_offset(head_skb, offset,
3221                                                  skb_put(nskb, hsize), hsize);
3222 
3223                 skb_shinfo(nskb)->tx_flags = skb_shinfo(head_skb)->tx_flags &
3224                         SKBTX_SHARED_FRAG;
3225 
3226                 while (pos < offset + len) {
3227                         if (i >= nfrags) {
3228                                 BUG_ON(skb_headlen(list_skb));
3229 
3230                                 i = 0;
3231                                 nfrags = skb_shinfo(list_skb)->nr_frags;
3232                                 frag = skb_shinfo(list_skb)->frags;
3233                                 frag_skb = list_skb;
3234 
3235                                 BUG_ON(!nfrags);
3236 
3237                                 list_skb = list_skb->next;
3238                         }
3239 
3240                         if (unlikely(skb_shinfo(nskb)->nr_frags >=
3241                                      MAX_SKB_FRAGS)) {
3242                                 net_warn_ratelimited(
3243                                         "skb_segment: too many frags: %u %u\n",
3244                                         pos, mss);
3245                                 goto err;
3246                         }
3247 
3248                         if (unlikely(skb_orphan_frags(frag_skb, GFP_ATOMIC)))
3249                                 goto err;
3250 
3251                         *nskb_frag = *frag;
3252                         __skb_frag_ref(nskb_frag);
3253                         size = skb_frag_size(nskb_frag);
3254 
3255                         if (pos < offset) {
3256                                 nskb_frag->page_offset += offset - pos;
3257                                 skb_frag_size_sub(nskb_frag, offset - pos);
3258                         }
3259 
3260                         skb_shinfo(nskb)->nr_frags++;
3261 
3262                         if (pos + size <= offset + len) {
3263                                 i++;
3264                                 frag++;
3265                                 pos += size;
3266                         } else {
3267                                 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
3268                                 goto skip_fraglist;
3269                         }
3270 
3271                         nskb_frag++;
3272                 }
3273 
3274 skip_fraglist:
3275                 nskb->data_len = len - hsize;
3276                 nskb->len += nskb->data_len;
3277                 nskb->truesize += nskb->data_len;
3278 
3279 perform_csum_check:
3280                 if (!csum) {
3281                         if (skb_has_shared_frag(nskb)) {
3282                                 err = __skb_linearize(nskb);
3283                                 if (err)
3284                                         goto err;
3285                         }
3286                         if (!nskb->remcsum_offload)
3287                                 nskb->ip_summed = CHECKSUM_NONE;
3288                         SKB_GSO_CB(nskb)->csum =
3289                                 skb_checksum(nskb, doffset,
3290                                              nskb->len - doffset, 0);
3291                         SKB_GSO_CB(nskb)->csum_start =
3292                                 skb_headroom(nskb) + doffset;
3293                 }
3294         } while ((offset += len) < head_skb->len);
3295 
3296         /* Some callers want to get the end of the list.
3297          * Put it in segs->prev to avoid walking the list.
3298          * (see validate_xmit_skb_list() for example)
3299          */
3300         segs->prev = tail;
3301 
3302         if (partial_segs) {
3303                 struct sk_buff *iter;
3304                 int type = skb_shinfo(head_skb)->gso_type;
3305                 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
3306 
3307                 /* Update type to add partial and then remove dodgy if set */
3308                 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
3309                 type &= ~SKB_GSO_DODGY;
3310 
3311                 /* Update GSO info and prepare to start updating headers on
3312                  * our way back down the stack of protocols.
3313                  */
3314                 for (iter = segs; iter; iter = iter->next) {
3315                         skb_shinfo(iter)->gso_size = gso_size;
3316                         skb_shinfo(iter)->gso_segs = partial_segs;
3317                         skb_shinfo(iter)->gso_type = type;
3318                         SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
3319                 }
3320 
3321                 if (tail->len - doffset <= gso_size)
3322                         skb_shinfo(tail)->gso_size = 0;
3323                 else if (tail != segs)
3324                         skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
3325         }
3326 
3327         /* Following permits correct backpressure, for protocols
3328          * using skb_set_owner_w().
3329          * Idea is to tranfert ownership from head_skb to last segment.
3330          */
3331         if (head_skb->destructor == sock_wfree) {
3332                 swap(tail->truesize, head_skb->truesize);
3333                 swap(tail->destructor, head_skb->destructor);
3334                 swap(tail->sk, head_skb->sk);
3335         }
3336         return segs;
3337 
3338 err:
3339         kfree_skb_list(segs);
3340         return ERR_PTR(err);
3341 }
3342 EXPORT_SYMBOL_GPL(skb_segment);
3343 
3344 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
3345 {
3346         struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
3347         unsigned int offset = skb_gro_offset(skb);
3348         unsigned int headlen = skb_headlen(skb);
3349         unsigned int len = skb_gro_len(skb);
3350         struct sk_buff *lp, *p = *head;
3351         unsigned int delta_truesize;
3352 
3353         if (unlikely(p->len + len >= 65536))
3354                 return -E2BIG;
3355 
3356         lp = NAPI_GRO_CB(p)->last;
3357         pinfo = skb_shinfo(lp);
3358 
3359         if (headlen <= offset) {
3360                 skb_frag_t *frag;
3361                 skb_frag_t *frag2;
3362                 int i = skbinfo->nr_frags;
3363                 int nr_frags = pinfo->nr_frags + i;
3364 
3365                 if (nr_frags > MAX_SKB_FRAGS)
3366                         goto merge;
3367 
3368                 offset -= headlen;
3369                 pinfo->nr_frags = nr_frags;
3370                 skbinfo->nr_frags = 0;
3371 
3372                 frag = pinfo->frags + nr_frags;
3373                 frag2 = skbinfo->frags + i;
3374                 do {
3375                         *--frag = *--frag2;
3376                 } while (--i);
3377 
3378                 frag->page_offset += offset;
3379                 skb_frag_size_sub(frag, offset);
3380 
3381                 /* all fragments truesize : remove (head size + sk_buff) */
3382                 delta_truesize = skb->truesize -
3383                                  SKB_TRUESIZE(skb_end_offset(skb));
3384 
3385                 skb->truesize -= skb->data_len;
3386                 skb->len -= skb->data_len;
3387                 skb->data_len = 0;
3388 
3389                 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
3390                 goto done;
3391         } else if (skb->head_frag) {
3392                 int nr_frags = pinfo->nr_frags;
3393                 skb_frag_t *frag = pinfo->frags + nr_frags;
3394                 struct page *page = virt_to_head_page(skb->head);
3395                 unsigned int first_size = headlen - offset;
3396                 unsigned int first_offset;
3397 
3398                 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
3399                         goto merge;
3400 
3401                 first_offset = skb->data -
3402                                (unsigned char *)page_address(page) +
3403                                offset;
3404 
3405                 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3406 
3407                 frag->page.p      = page;
3408                 frag->page_offset = first_offset;
3409                 skb_frag_size_set(frag, first_size);
3410 
3411                 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3412                 /* We dont need to clear skbinfo->nr_frags here */
3413 
3414                 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3415                 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3416                 goto done;
3417         }
3418 
3419 merge:
3420         delta_truesize = skb->truesize;
3421         if (offset > headlen) {
3422                 unsigned int eat = offset - headlen;
3423 
3424                 skbinfo->frags[0].page_offset += eat;
3425                 skb_frag_size_sub(&skbinfo->frags[0], eat);
3426                 skb->data_len -= eat;
3427                 skb->len -= eat;
3428                 offset = headlen;
3429         }
3430 
3431         __skb_pull(skb, offset);
3432 
3433         if (NAPI_GRO_CB(p)->last == p)
3434                 skb_shinfo(p)->frag_list = skb;
3435         else
3436                 NAPI_GRO_CB(p)->last->next = skb;
3437         NAPI_GRO_CB(p)->last = skb;
3438         __skb_header_release(skb);
3439         lp = p;
3440 
3441 done:
3442         NAPI_GRO_CB(p)->count++;
3443         p->data_len += len;
3444         p->truesize += delta_truesize;
3445         p->len += len;
3446         if (lp != p) {
3447                 lp->data_len += len;
3448                 lp->truesize += delta_truesize;
3449                 lp->len += len;
3450         }
3451         NAPI_GRO_CB(skb)->same_flow = 1;
3452         return 0;
3453 }
3454 EXPORT_SYMBOL_GPL(skb_gro_receive);
3455 
3456 void __init skb_init(void)
3457 {
3458         skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3459                                               sizeof(struct sk_buff),
3460                                               0,
3461                                               SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3462                                               NULL);
3463         skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3464                                                 sizeof(struct sk_buff_fclones),
3465                                                 0,
3466                                                 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3467                                                 NULL);
3468 }
3469 
3470 /**
3471  *      skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3472  *      @skb: Socket buffer containing the buffers to be mapped
3473  *      @sg: The scatter-gather list to map into
3474  *      @offset: The offset into the buffer's contents to start mapping
3475  *      @len: Length of buffer space to be mapped
3476  *
3477  *      Fill the specified scatter-gather list with mappings/pointers into a
3478  *      region of the buffer space attached to a socket buffer.
3479  */
3480 static int
3481 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3482 {
3483         int start = skb_headlen(skb);
3484         int i, copy = start - offset;
3485         struct sk_buff *frag_iter;
3486         int elt = 0;
3487 
3488         if (copy > 0) {
3489                 if (copy > len)
3490                         copy = len;
3491                 sg_set_buf(sg, skb->data + offset, copy);
3492                 elt++;
3493                 if ((len -= copy) == 0)
3494                         return elt;
3495                 offset += copy;
3496         }
3497 
3498         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3499                 int end;
3500 
3501                 WARN_ON(start > offset + len);
3502 
3503                 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3504                 if ((copy = end - offset) > 0) {
3505                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3506 
3507                         if (copy > len)
3508                                 copy = len;
3509                         sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3510                                         frag->page_offset+offset-start);
3511                         elt++;
3512                         if (!(len -= copy))
3513                                 return elt;
3514                         offset += copy;
3515                 }
3516                 start = end;
3517         }
3518 
3519         skb_walk_frags(skb, frag_iter) {
3520                 int end;
3521 
3522                 WARN_ON(start > offset + len);
3523 
3524                 end = start + frag_iter->len;
3525                 if ((copy = end - offset) > 0) {
3526                         if (copy > len)
3527                                 copy = len;
3528                         elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3529                                               copy);
3530                         if ((len -= copy) == 0)
3531                                 return elt;
3532                         offset += copy;
3533                 }
3534                 start = end;
3535         }
3536         BUG_ON(len);
3537         return elt;
3538 }
3539 
3540 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
3541  * sglist without mark the sg which contain last skb data as the end.
3542  * So the caller can mannipulate sg list as will when padding new data after
3543  * the first call without calling sg_unmark_end to expend sg list.
3544  *
3545  * Scenario to use skb_to_sgvec_nomark:
3546  * 1. sg_init_table
3547  * 2. skb_to_sgvec_nomark(payload1)
3548  * 3. skb_to_sgvec_nomark(payload2)
3549  *
3550  * This is equivalent to:
3551  * 1. sg_init_table
3552  * 2. skb_to_sgvec(payload1)
3553  * 3. sg_unmark_end
3554  * 4. skb_to_sgvec(payload2)
3555  *
3556  * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
3557  * is more preferable.
3558  */
3559 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
3560                         int offset, int len)
3561 {
3562         return __skb_to_sgvec(skb, sg, offset, len);
3563 }
3564 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
3565 
3566 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3567 {
3568         int nsg = __skb_to_sgvec(skb, sg, offset, len);
3569 
3570         sg_mark_end(&sg[nsg - 1]);
3571 
3572         return nsg;
3573 }
3574 EXPORT_SYMBOL_GPL(skb_to_sgvec);
3575 
3576 /**
3577  *      skb_cow_data - Check that a socket buffer's data buffers are writable
3578  *      @skb: The socket buffer to check.
3579  *      @tailbits: Amount of trailing space to be added
3580  *      @trailer: Returned pointer to the skb where the @tailbits space begins
3581  *
3582  *      Make sure that the data buffers attached to a socket buffer are
3583  *      writable. If they are not, private copies are made of the data buffers
3584  *      and the socket buffer is set to use these instead.
3585  *
3586  *      If @tailbits is given, make sure that there is space to write @tailbits
3587  *      bytes of data beyond current end of socket buffer.  @trailer will be
3588  *      set to point to the skb in which this space begins.
3589  *
3590  *      The number of scatterlist elements required to completely map the
3591  *      COW'd and extended socket buffer will be returned.
3592  */
3593 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3594 {
3595         int copyflag;
3596         int elt;
3597         struct sk_buff *skb1, **skb_p;
3598 
3599         /* If skb is cloned or its head is paged, reallocate
3600          * head pulling out all the pages (pages are considered not writable
3601          * at the moment even if they are anonymous).
3602          */
3603         if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3604             __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3605                 return -ENOMEM;
3606 
3607         /* Easy case. Most of packets will go this way. */
3608         if (!skb_has_frag_list(skb)) {
3609                 /* A little of trouble, not enough of space for trailer.
3610                  * This should not happen, when stack is tuned to generate
3611                  * good frames. OK, on miss we reallocate and reserve even more
3612                  * space, 128 bytes is fair. */
3613 
3614                 if (skb_tailroom(skb) < tailbits &&
3615                     pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3616                         return -ENOMEM;
3617 
3618                 /* Voila! */
3619                 *trailer = skb;
3620                 return 1;
3621         }
3622 
3623         /* Misery. We are in troubles, going to mincer fragments... */
3624 
3625         elt = 1;
3626         skb_p = &skb_shinfo(skb)->frag_list;
3627         copyflag = 0;
3628 
3629         while ((skb1 = *skb_p) != NULL) {
3630                 int ntail = 0;
3631 
3632                 /* The fragment is partially pulled by someone,
3633                  * this can happen on input. Copy it and everything
3634                  * after it. */
3635 
3636                 if (skb_shared(skb1))
3637                         copyflag = 1;
3638 
3639                 /* If the skb is the last, worry about trailer. */
3640 
3641                 if (skb1->next == NULL && tailbits) {
3642                         if (skb_shinfo(skb1)->nr_frags ||
3643                             skb_has_frag_list(skb1) ||
3644                             skb_tailroom(skb1) < tailbits)
3645                                 ntail = tailbits + 128;
3646                 }
3647 
3648                 if (copyflag ||
3649                     skb_cloned(skb1) ||
3650                     ntail ||
3651                     skb_shinfo(skb1)->nr_frags ||
3652                     skb_has_frag_list(skb1)) {
3653                         struct sk_buff *skb2;
3654 
3655                         /* Fuck, we are miserable poor guys... */
3656                         if (ntail == 0)
3657                                 skb2 = skb_copy(skb1, GFP_ATOMIC);
3658                         else
3659                                 skb2 = skb_copy_expand(skb1,
3660                                                        skb_headroom(skb1),
3661                                                        ntail,
3662                                                        GFP_ATOMIC);
3663                         if (unlikely(skb2 == NULL))
3664                                 return -ENOMEM;
3665 
3666                         if (skb1->sk)
3667                                 skb_set_owner_w(skb2, skb1->sk);
3668 
3669                         /* Looking around. Are we still alive?
3670                          * OK, link new skb, drop old one */
3671 
3672                         skb2->next = skb1->next;
3673                         *skb_p = skb2;
3674                         kfree_skb(skb1);
3675                         skb1 = skb2;
3676                 }
3677                 elt++;
3678                 *trailer = skb1;
3679                 skb_p = &skb1->next;
3680         }
3681 
3682         return elt;
3683 }
3684 EXPORT_SYMBOL_GPL(skb_cow_data);
3685 
3686 static void sock_rmem_free(struct sk_buff *skb)
3687 {
3688         struct sock *sk = skb->sk;
3689 
3690         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3691 }
3692 
3693 /*
3694  * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3695  */
3696 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3697 {
3698         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3699             (unsigned int)sk->sk_rcvbuf)
3700                 return -ENOMEM;
3701 
3702         skb_orphan(skb);
3703         skb->sk = sk;
3704         skb->destructor = sock_rmem_free;
3705         atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3706 
3707         /* before exiting rcu section, make sure dst is refcounted */
3708         skb_dst_force(skb);
3709 
3710         skb_queue_tail(&sk->sk_error_queue, skb);
3711         if (!sock_flag(sk, SOCK_DEAD))
3712                 sk->sk_data_ready(sk);
3713         return 0;
3714 }
3715 EXPORT_SYMBOL(sock_queue_err_skb);
3716 
3717 static bool is_icmp_err_skb(const struct sk_buff *skb)
3718 {
3719         return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
3720                        SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
3721 }
3722 
3723 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
3724 {
3725         struct sk_buff_head *q = &sk->sk_error_queue;
3726         struct sk_buff *skb, *skb_next = NULL;
3727         bool icmp_next = false;
3728         unsigned long flags;
3729 
3730         spin_lock_irqsave(&q->lock, flags);
3731         skb = __skb_dequeue(q);
3732         if (skb && (skb_next = skb_peek(q)))
3733                 icmp_next = is_icmp_err_skb(skb_next);
3734         spin_unlock_irqrestore(&q->lock, flags);
3735 
3736         if (is_icmp_err_skb(skb) && !icmp_next)
3737                 sk->sk_err = 0;
3738 
3739         if (skb_next)
3740                 sk->sk_error_report(sk);
3741 
3742         return skb;
3743 }
3744 EXPORT_SYMBOL(sock_dequeue_err_skb);
3745 
3746 /**
3747  * skb_clone_sk - create clone of skb, and take reference to socket
3748  * @skb: the skb to clone
3749  *
3750  * This function creates a clone of a buffer that holds a reference on
3751  * sk_refcnt.  Buffers created via this function are meant to be
3752  * returned using sock_queue_err_skb, or free via kfree_skb.
3753  *
3754  * When passing buffers allocated with this function to sock_queue_err_skb
3755  * it is necessary to wrap the call with sock_hold/sock_put in order to
3756  * prevent the socket from being released prior to being enqueued on
3757  * the sk_error_queue.
3758  */
3759 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
3760 {
3761         struct sock *sk = skb->sk;
3762         struct sk_buff *clone;
3763 
3764         if (!sk || !atomic_inc_not_zero(&sk->sk_refcnt))
3765                 return NULL;
3766 
3767         clone = skb_clone(skb, GFP_ATOMIC);
3768         if (!clone) {
3769                 sock_put(sk);
3770                 return NULL;
3771         }
3772 
3773         clone->sk = sk;
3774         clone->destructor = sock_efree;
3775 
3776         return clone;
3777 }
3778 EXPORT_SYMBOL(skb_clone_sk);
3779 
3780 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
3781                                         struct sock *sk,
3782                                         int tstype)
3783 {
3784         struct sock_exterr_skb *serr;
3785         int err;
3786 
3787         serr = SKB_EXT_ERR(skb);
3788         memset(serr, 0, sizeof(*serr));
3789         serr->ee.ee_errno = ENOMSG;
3790         serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3791         serr->ee.ee_info = tstype;
3792         if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
3793                 serr->ee.ee_data = skb_shinfo(skb)->tskey;
3794                 if (sk->sk_protocol == IPPROTO_TCP &&
3795                     sk->sk_type == SOCK_STREAM)
3796                         serr->ee.ee_data -= sk->sk_tskey;
3797         }
3798 
3799         err = sock_queue_err_skb(sk, skb);
3800 
3801         if (err)
3802                 kfree_skb(skb);
3803 }
3804 
3805 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
3806 {
3807         bool ret;
3808 
3809         if (likely(sysctl_tstamp_allow_data || tsonly))
3810                 return true;
3811 
3812         read_lock_bh(&sk->sk_callback_lock);
3813         ret = sk->sk_socket && sk->sk_socket->file &&
3814               file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
3815         read_unlock_bh(&sk->sk_callback_lock);
3816         return ret;
3817 }
3818 
3819 void skb_complete_tx_timestamp(struct sk_buff *skb,
3820                                struct skb_shared_hwtstamps *hwtstamps)
3821 {
3822         struct sock *sk = skb->sk;
3823 
3824         if (!skb_may_tx_timestamp(sk, false))
3825                 return;
3826 
3827         /* take a reference to prevent skb_orphan() from freeing the socket */
3828         sock_hold(sk);
3829 
3830         *skb_hwtstamps(skb) = *hwtstamps;
3831         __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND);
3832 
3833         sock_put(sk);
3834 }
3835 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
3836 
3837 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3838                      struct skb_shared_hwtstamps *hwtstamps,
3839                      struct sock *sk, int tstype)
3840 {
3841         struct sk_buff *skb;
3842         bool tsonly;
3843 
3844         if (!sk)
3845                 return;
3846 
3847         tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
3848         if (!skb_may_tx_timestamp(sk, tsonly))
3849                 return;
3850 
3851         if (tsonly) {
3852 #ifdef CONFIG_INET
3853                 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
3854                     sk->sk_protocol == IPPROTO_TCP &&
3855                     sk->sk_type == SOCK_STREAM)
3856                         skb = tcp_get_timestamping_opt_stats(sk);
3857                 else
3858 #endif
3859                         skb = alloc_skb(0, GFP_ATOMIC);
3860         } else {
3861                 skb = skb_clone(orig_skb, GFP_ATOMIC);
3862         }
3863         if (!skb)
3864                 return;
3865 
3866         if (tsonly) {
3867                 skb_shinfo(skb)->tx_flags = skb_shinfo(orig_skb)->tx_flags;
3868                 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
3869         }
3870 
3871         if (hwtstamps)
3872                 *skb_hwtstamps(skb) = *hwtstamps;
3873         else
3874                 skb->tstamp = ktime_get_real();
3875 
3876         __skb_complete_tx_timestamp(skb, sk, tstype);
3877 }
3878 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
3879 
3880 void skb_tstamp_tx(struct sk_buff *orig_skb,
3881                    struct skb_shared_hwtstamps *hwtstamps)
3882 {
3883         return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
3884                                SCM_TSTAMP_SND);
3885 }
3886 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3887 
3888 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3889 {
3890         struct sock *sk = skb->sk;
3891         struct sock_exterr_skb *serr;
3892         int err;
3893 
3894         skb->wifi_acked_valid = 1;
3895         skb->wifi_acked = acked;
3896 
3897         serr = SKB_EXT_ERR(skb);
3898         memset(serr, 0, sizeof(*serr));
3899         serr->ee.ee_errno = ENOMSG;
3900         serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3901 
3902         /* take a reference to prevent skb_orphan() from freeing the socket */
3903         sock_hold(sk);
3904 
3905         err = sock_queue_err_skb(sk, skb);
3906         if (err)
3907                 kfree_skb(skb);
3908 
3909         sock_put(sk);
3910 }
3911 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3912 
3913 /**
3914  * skb_partial_csum_set - set up and verify partial csum values for packet
3915  * @skb: the skb to set
3916  * @start: the number of bytes after skb->data to start checksumming.
3917  * @off: the offset from start to place the checksum.
3918  *
3919  * For untrusted partially-checksummed packets, we need to make sure the values
3920  * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3921  *
3922  * This function checks and sets those values and skb->ip_summed: if this
3923  * returns false you should drop the packet.
3924  */
3925 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3926 {
3927         if (unlikely(start > skb_headlen(skb)) ||
3928             unlikely((int)start + off > skb_headlen(skb) - 2)) {
3929                 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3930                                      start, off, skb_headlen(skb));
3931                 return false;
3932         }
3933         skb->ip_summed = CHECKSUM_PARTIAL;
3934         skb->csum_start = skb_headroom(skb) + start;
3935         skb->csum_offset = off;
3936         skb_set_transport_header(skb, start);
3937         return true;
3938 }
3939 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3940 
3941 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
3942                                unsigned int max)
3943 {
3944         if (skb_headlen(skb) >= len)
3945                 return 0;
3946 
3947         /* If we need to pullup then pullup to the max, so we
3948          * won't need to do it again.
3949          */
3950         if (max > skb->len)
3951                 max = skb->len;
3952 
3953         if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
3954                 return -ENOMEM;
3955 
3956         if (skb_headlen(skb) < len)
3957                 return -EPROTO;
3958 
3959         return 0;
3960 }
3961 
3962 #define MAX_TCP_HDR_LEN (15 * 4)
3963 
3964 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
3965                                       typeof(IPPROTO_IP) proto,
3966                                       unsigned int off)
3967 {
3968         switch (proto) {
3969                 int err;
3970 
3971         case IPPROTO_TCP:
3972                 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
3973                                           off + MAX_TCP_HDR_LEN);
3974                 if (!err && !skb_partial_csum_set(skb, off,
3975                                                   offsetof(struct tcphdr,
3976                                                            check)))
3977                         err = -EPROTO;
3978                 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
3979 
3980         case IPPROTO_UDP:
3981                 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
3982                                           off + sizeof(struct udphdr));
3983                 if (!err && !skb_partial_csum_set(skb, off,
3984                                                   offsetof(struct udphdr,
3985                                                            check)))
3986                         err = -EPROTO;
3987                 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
3988         }
3989 
3990         return ERR_PTR(-EPROTO);
3991 }
3992 
3993 /* This value should be large enough to cover a tagged ethernet header plus
3994  * maximally sized IP and TCP or UDP headers.
3995  */
3996 #define MAX_IP_HDR_LEN 128
3997 
3998 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
3999 {
4000         unsigned int off;
4001         bool fragment;
4002         __sum16 *csum;
4003         int err;
4004 
4005         fragment = false;
4006 
4007         err = skb_maybe_pull_tail(skb,
4008                                   sizeof(struct iphdr),
4009                                   MAX_IP_HDR_LEN);
4010         if (err < 0)
4011                 goto out;
4012 
4013         if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
4014                 fragment = true;
4015 
4016         off = ip_hdrlen(skb);
4017 
4018         err = -EPROTO;
4019 
4020         if (fragment)
4021                 goto out;
4022 
4023         csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
4024         if (IS_ERR(csum))
4025                 return PTR_ERR(csum);
4026 
4027         if (recalculate)
4028                 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
4029                                            ip_hdr(skb)->daddr,
4030                                            skb->len - off,
4031                                            ip_hdr(skb)->protocol, 0);
4032         err = 0;
4033 
4034 out:
4035         return err;
4036 }
4037 
4038 /* This value should be large enough to cover a tagged ethernet header plus
4039  * an IPv6 header, all options, and a maximal TCP or UDP header.
4040  */
4041 #define MAX_IPV6_HDR_LEN 256
4042 
4043 #define OPT_HDR(type, skb, off) \
4044         (type *)(skb_network_header(skb) + (off))
4045 
4046 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
4047 {
4048         int err;
4049         u8 nexthdr;
4050         unsigned int off;
4051         unsigned int len;
4052         bool fragment;
4053         bool done;
4054         __sum16 *csum;
4055 
4056         fragment = false;
4057         done = false;
4058 
4059         off = sizeof(struct ipv6hdr);
4060 
4061         err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4062         if (err < 0)
4063                 goto out;
4064 
4065         nexthdr = ipv6_hdr(skb)->nexthdr;
4066 
4067         len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4068         while (off <= len && !done) {
4069                 switch (nexthdr) {
4070                 case IPPROTO_DSTOPTS:
4071                 case IPPROTO_HOPOPTS:
4072                 case IPPROTO_ROUTING: {
4073                         struct ipv6_opt_hdr *hp;
4074 
4075                         err = skb_maybe_pull_tail(skb,
4076                                                   off +
4077                                                   sizeof(struct ipv6_opt_hdr),
4078                                                   MAX_IPV6_HDR_LEN);
4079                         if (err < 0)
4080                                 goto out;
4081 
4082                         hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
4083                         nexthdr = hp->nexthdr;
4084                         off += ipv6_optlen(hp);
4085                         break;
4086                 }
4087                 case IPPROTO_AH: {
4088                         struct ip_auth_hdr *hp;
4089 
4090                         err = skb_maybe_pull_tail(skb,
4091                                                   off +
4092                                                   sizeof(struct ip_auth_hdr),
4093                                                   MAX_IPV6_HDR_LEN);
4094                         if (err < 0)
4095                                 goto out;
4096 
4097                         hp = OPT_HDR(struct ip_auth_hdr, skb, off);
4098                         nexthdr = hp->nexthdr;
4099                         off += ipv6_authlen(hp);
4100                         break;
4101                 }
4102                 case IPPROTO_FRAGMENT: {
4103                         struct frag_hdr *hp;
4104 
4105                         err = skb_maybe_pull_tail(skb,
4106                                                   off +
4107                                                   sizeof(struct frag_hdr),
4108                                                   MAX_IPV6_HDR_LEN);
4109                         if (err < 0)
4110                                 goto out;
4111 
4112                         hp = OPT_HDR(struct frag_hdr, skb, off);
4113 
4114                         if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
4115                                 fragment = true;
4116 
4117                         nexthdr = hp->nexthdr;
4118                         off += sizeof(struct frag_hdr);
4119                         break;
4120                 }
4121                 default:
4122                         done = true;
4123                         break;
4124                 }
4125         }
4126 
4127         err = -EPROTO;
4128 
4129         if (!done || fragment)
4130                 goto out;
4131 
4132         csum = skb_checksum_setup_ip(skb, nexthdr, off);
4133         if (IS_ERR(csum))
4134                 return PTR_ERR(csum);
4135 
4136         if (recalculate)
4137                 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4138                                          &ipv6_hdr(skb)->daddr,
4139                                          skb->len - off, nexthdr, 0);
4140         err = 0;
4141 
4142 out:
4143         return err;
4144 }
4145 
4146 /**
4147  * skb_checksum_setup - set up partial checksum offset
4148  * @skb: the skb to set up
4149  * @recalculate: if true the pseudo-header checksum will be recalculated
4150  */
4151 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
4152 {
4153         int err;
4154 
4155         switch (skb->protocol) {
4156         case htons(ETH_P_IP):
4157                 err = skb_checksum_setup_ipv4(skb, recalculate);
4158                 break;
4159 
4160         case htons(ETH_P_IPV6):
4161                 err = skb_checksum_setup_ipv6(skb, recalculate);
4162                 break;
4163 
4164         default:
4165                 err = -EPROTO;
4166                 break;
4167         }
4168 
4169         return err;
4170 }
4171 EXPORT_SYMBOL(skb_checksum_setup);
4172 
4173 /**
4174  * skb_checksum_maybe_trim - maybe trims the given skb
4175  * @skb: the skb to check
4176  * @transport_len: the data length beyond the network header
4177  *
4178  * Checks whether the given skb has data beyond the given transport length.
4179  * If so, returns a cloned skb trimmed to this transport length.
4180  * Otherwise returns the provided skb. Returns NULL in error cases
4181  * (e.g. transport_len exceeds skb length or out-of-memory).
4182  *
4183  * Caller needs to set the skb transport header and free any returned skb if it
4184  * differs from the provided skb.
4185  */
4186 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
4187                                                unsigned int transport_len)
4188 {
4189         struct sk_buff *skb_chk;
4190         unsigned int len = skb_transport_offset(skb) + transport_len;
4191         int ret;
4192 
4193         if (skb->len < len)
4194                 return NULL;
4195         else if (skb->len == len)
4196                 return skb;
4197 
4198         skb_chk = skb_clone(skb, GFP_ATOMIC);
4199         if (!skb_chk)
4200                 return NULL;
4201 
4202         ret = pskb_trim_rcsum(skb_chk, len);
4203         if (ret) {
4204                 kfree_skb(skb_chk);
4205                 return NULL;
4206         }
4207 
4208         return skb_chk;
4209 }
4210 
4211 /**
4212  * skb_checksum_trimmed - validate checksum of an skb
4213  * @skb: the skb to check
4214  * @transport_len: the data length beyond the network header
4215  * @skb_chkf: checksum function to use
4216  *
4217  * Applies the given checksum function skb_chkf to the provided skb.
4218  * Returns a checked and maybe trimmed skb. Returns NULL on error.
4219  *
4220  * If the skb has data beyond the given transport length, then a
4221  * trimmed & cloned skb is checked and returned.
4222  *
4223  * Caller needs to set the skb transport header and free any returned skb if it
4224  * differs from the provided skb.
4225  */
4226 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4227                                      unsigned int transport_len,
4228                                      __sum16(*skb_chkf)(struct sk_buff *skb))
4229 {
4230         struct sk_buff *skb_chk;
4231         unsigned int offset = skb_transport_offset(skb);
4232         __sum16 ret;
4233 
4234         skb_chk = skb_checksum_maybe_trim(skb, transport_len);
4235         if (!skb_chk)
4236                 goto err;
4237 
4238         if (!pskb_may_pull(skb_chk, offset))
4239                 goto err;
4240 
4241         skb_pull_rcsum(skb_chk, offset);
4242         ret = skb_chkf(skb_chk);
4243         skb_push_rcsum(skb_chk, offset);
4244 
4245         if (ret)
4246                 goto err;
4247 
4248         return skb_chk;
4249 
4250 err:
4251         if (skb_chk && skb_chk != skb)
4252                 kfree_skb(skb_chk);
4253 
4254         return NULL;
4255 
4256 }
4257 EXPORT_SYMBOL(skb_checksum_trimmed);
4258 
4259 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
4260 {
4261         net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
4262                              skb->dev->name);
4263 }
4264 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
4265 
4266 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
4267 {
4268         if (head_stolen) {
4269                 skb_release_head_state(skb);
4270                 kmem_cache_free(skbuff_head_cache, skb);
4271         } else {
4272                 __kfree_skb(skb);
4273         }
4274 }
4275 EXPORT_SYMBOL(kfree_skb_partial);
4276 
4277 /**
4278  * skb_try_coalesce - try to merge skb to prior one
4279  * @to: prior buffer
4280  * @from: buffer to add
4281  * @fragstolen: pointer to boolean
4282  * @delta_truesize: how much more was allocated than was requested
4283  */
4284 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
4285                       bool *fragstolen, int *delta_truesize)
4286 {
4287         int i, delta, len = from->len;
4288 
4289         *fragstolen = false;
4290 
4291         if (skb_cloned(to))
4292                 return false;
4293 
4294         if (len <= skb_tailroom(to)) {
4295                 if (len)
4296                         BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
4297                 *delta_truesize = 0;
4298                 return true;
4299         }
4300 
4301         if (skb_has_frag_list(to) || skb_has_frag_list(from))
4302                 return false;
4303 
4304         if (skb_headlen(from) != 0) {
4305                 struct page *page;
4306                 unsigned int offset;
4307 
4308                 if (skb_shinfo(to)->nr_frags +
4309                     skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
4310                         return false;
4311 
4312                 if (skb_head_is_locked(from))
4313                         return false;
4314 
4315                 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4316 
4317                 page = virt_to_head_page(from->head);
4318                 offset = from->data - (unsigned char *)page_address(page);
4319 
4320                 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
4321                                    page, offset, skb_headlen(from));
4322                 *fragstolen = true;
4323         } else {
4324                 if (skb_shinfo(to)->nr_frags +
4325                     skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
4326                         return false;
4327 
4328                 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
4329         }
4330 
4331         WARN_ON_ONCE(delta < len);
4332 
4333         memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
4334                skb_shinfo(from)->frags,
4335                skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
4336         skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
4337 
4338         if (!skb_cloned(from))
4339                 skb_shinfo(from)->nr_frags = 0;
4340 
4341         /* if the skb is not cloned this does nothing
4342          * since we set nr_frags to 0.
4343          */
4344         for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
4345                 skb_frag_ref(from, i);
4346 
4347         to->truesize += delta;
4348         to->len += len;
4349         to->data_len += len;
4350 
4351         *delta_truesize = delta;
4352         return true;
4353 }
4354 EXPORT_SYMBOL(skb_try_coalesce);
4355 
4356 /**
4357  * skb_scrub_packet - scrub an skb
4358  *
4359  * @skb: buffer to clean
4360  * @xnet: packet is crossing netns
4361  *
4362  * skb_scrub_packet can be used after encapsulating or decapsulting a packet
4363  * into/from a tunnel. Some information have to be cleared during these
4364  * operations.
4365  * skb_scrub_packet can also be used to clean a skb before injecting it in
4366  * another namespace (@xnet == true). We have to clear all information in the
4367  * skb that could impact namespace isolation.
4368  */
4369 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
4370 {
4371         skb->tstamp = 0;
4372         skb->pkt_type = PACKET_HOST;
4373         skb->skb_iif = 0;
4374         skb->ignore_df = 0;
4375         skb_dst_drop(skb);
4376         secpath_reset(skb);
4377         nf_reset(skb);
4378         nf_reset_trace(skb);
4379 
4380         if (!xnet)
4381                 return;
4382 
4383         skb_orphan(skb);
4384         skb->mark = 0;
4385 }
4386 EXPORT_SYMBOL_GPL(skb_scrub_packet);
4387 
4388 /**
4389  * skb_gso_transport_seglen - Return length of individual segments of a gso packet
4390  *
4391  * @skb: GSO skb
4392  *
4393  * skb_gso_transport_seglen is used to determine the real size of the
4394  * individual segments, including Layer4 headers (TCP/UDP).
4395  *
4396  * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
4397  */
4398 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
4399 {
4400         const struct skb_shared_info *shinfo = skb_shinfo(skb);
4401         unsigned int thlen = 0;
4402 
4403         if (skb->encapsulation) {
4404                 thlen = skb_inner_transport_header(skb) -
4405                         skb_transport_header(skb);
4406 
4407                 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
4408                         thlen += inner_tcp_hdrlen(skb);
4409         } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
4410                 thlen = tcp_hdrlen(skb);
4411         } else if (unlikely(shinfo->gso_type & SKB_GSO_SCTP)) {
4412                 thlen = sizeof(struct sctphdr);
4413         }
4414         /* UFO sets gso_size to the size of the fragmentation
4415          * payload, i.e. the size of the L4 (UDP) header is already
4416          * accounted for.
4417          */
4418         return thlen + shinfo->gso_size;
4419 }
4420 EXPORT_SYMBOL_GPL(skb_gso_transport_seglen);
4421 
4422 /**
4423  * skb_gso_validate_mtu - Return in case such skb fits a given MTU
4424  *
4425  * @skb: GSO skb
4426  * @mtu: MTU to validate against
4427  *
4428  * skb_gso_validate_mtu validates if a given skb will fit a wanted MTU
4429  * once split.
4430  */
4431 bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu)
4432 {
4433         const struct skb_shared_info *shinfo = skb_shinfo(skb);
4434         const struct sk_buff *iter;
4435         unsigned int hlen;
4436 
4437         hlen = skb_gso_network_seglen(skb);
4438 
4439         if (shinfo->gso_size != GSO_BY_FRAGS)
4440                 return hlen <= mtu;
4441 
4442         /* Undo this so we can re-use header sizes */
4443         hlen -= GSO_BY_FRAGS;
4444 
4445         skb_walk_frags(skb, iter) {
4446                 if (hlen + skb_headlen(iter) > mtu)
4447                         return false;
4448         }
4449 
4450         return true;
4451 }
4452 EXPORT_SYMBOL_GPL(skb_gso_validate_mtu);
4453 
4454 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
4455 {
4456         if (skb_cow(skb, skb_headroom(skb)) < 0) {
4457                 kfree_skb(skb);
4458                 return NULL;
4459         }
4460 
4461         memmove(skb->data - ETH_HLEN, skb->data - skb->mac_len - VLAN_HLEN,
4462                 2 * ETH_ALEN);
4463         skb->mac_header += VLAN_HLEN;
4464         return skb;
4465 }
4466 
4467 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
4468 {
4469         struct vlan_hdr *vhdr;
4470         u16 vlan_tci;
4471 
4472         if (unlikely(skb_vlan_tag_present(skb))) {
4473                 /* vlan_tci is already set-up so leave this for another time */
4474                 return skb;
4475         }
4476 
4477         skb = skb_share_check(skb, GFP_ATOMIC);
4478         if (unlikely(!skb))
4479                 goto err_free;
4480 
4481         if (unlikely(!pskb_may_pull(skb, VLAN_HLEN)))
4482                 goto err_free;
4483 
4484         vhdr = (struct vlan_hdr *)skb->data;
4485         vlan_tci = ntohs(vhdr->h_vlan_TCI);
4486         __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
4487 
4488         skb_pull_rcsum(skb, VLAN_HLEN);
4489         vlan_set_encap_proto(skb, vhdr);
4490 
4491         skb = skb_reorder_vlan_header(skb);
4492         if (unlikely(!skb))
4493                 goto err_free;
4494 
4495         skb_reset_network_header(skb);
4496         skb_reset_transport_header(skb);
4497         skb_reset_mac_len(skb);
4498 
4499         return skb;
4500 
4501 err_free:
4502         kfree_skb(skb);
4503         return NULL;
4504 }
4505 EXPORT_SYMBOL(skb_vlan_untag);
4506 
4507 int skb_ensure_writable(struct sk_buff *skb, int write_len)
4508 {
4509         if (!pskb_may_pull(skb, write_len))
4510                 return -ENOMEM;
4511 
4512         if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
4513                 return 0;
4514 
4515         return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4516 }
4517 EXPORT_SYMBOL(skb_ensure_writable);
4518 
4519 /* remove VLAN header from packet and update csum accordingly.
4520  * expects a non skb_vlan_tag_present skb with a vlan tag payload
4521  */
4522 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
4523 {
4524         struct vlan_hdr *vhdr;
4525         int offset = skb->data - skb_mac_header(skb);
4526         int err;
4527 
4528         if (WARN_ONCE(offset,
4529                       "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
4530                       offset)) {
4531                 return -EINVAL;
4532         }
4533 
4534         err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
4535         if (unlikely(err))
4536                 return err;
4537 
4538         skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
4539 
4540         vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
4541         *vlan_tci = ntohs(vhdr->h_vlan_TCI);
4542 
4543         memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
4544         __skb_pull(skb, VLAN_HLEN);
4545 
4546         vlan_set_encap_proto(skb, vhdr);
4547         skb->mac_header += VLAN_HLEN;
4548 
4549         if (skb_network_offset(skb) < ETH_HLEN)
4550                 skb_set_network_header(skb, ETH_HLEN);
4551 
4552         skb_reset_mac_len(skb);
4553 
4554         return err;
4555 }
4556 EXPORT_SYMBOL(__skb_vlan_pop);
4557 
4558 /* Pop a vlan tag either from hwaccel or from payload.
4559  * Expects skb->data at mac header.
4560  */
4561 int skb_vlan_pop(struct sk_buff *skb)
4562 {
4563         u16 vlan_tci;
4564         __be16 vlan_proto;
4565         int err;
4566 
4567         if (likely(skb_vlan_tag_present(skb))) {
4568                 skb->vlan_tci = 0;
4569         } else {
4570                 if (unlikely(!eth_type_vlan(skb->protocol)))
4571                         return 0;
4572 
4573                 err = __skb_vlan_pop(skb, &vlan_tci);
4574                 if (err)
4575                         return err;
4576         }
4577         /* move next vlan tag to hw accel tag */
4578         if (likely(!eth_type_vlan(skb->protocol)))
4579                 return 0;
4580 
4581         vlan_proto = skb->protocol;
4582         err = __skb_vlan_pop(skb, &vlan_tci);
4583         if (unlikely(err))
4584                 return err;
4585 
4586         __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
4587         return 0;
4588 }
4589 EXPORT_SYMBOL(skb_vlan_pop);
4590 
4591 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
4592  * Expects skb->data at mac header.
4593  */
4594 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
4595 {
4596         if (skb_vlan_tag_present(skb)) {
4597                 int offset = skb->data - skb_mac_header(skb);
4598                 int err;
4599 
4600                 if (WARN_ONCE(offset,
4601                               "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
4602                               offset)) {
4603                         return -EINVAL;
4604                 }
4605 
4606                 err = __vlan_insert_tag(skb, skb->vlan_proto,
4607                                         skb_vlan_tag_get(skb));
4608                 if (err)
4609                         return err;
4610 
4611                 skb->protocol = skb->vlan_proto;
4612                 skb->mac_len += VLAN_HLEN;
4613 
4614                 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
4615         }
4616         __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
4617         return 0;
4618 }
4619 EXPORT_SYMBOL(skb_vlan_push);
4620 
4621 /**
4622  * alloc_skb_with_frags - allocate skb with page frags
4623  *
4624  * @header_len: size of linear part
4625  * @data_len: needed length in frags
4626  * @max_page_order: max page order desired.
4627  * @errcode: pointer to error code if any
4628  * @gfp_mask: allocation mask
4629  *
4630  * This can be used to allocate a paged skb, given a maximal order for frags.
4631  */
4632 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
4633                                      unsigned long data_len,
4634                                      int max_page_order,
4635                                      int *errcode,
4636                                      gfp_t gfp_mask)
4637 {
4638         int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
4639         unsigned long chunk;
4640         struct sk_buff *skb;
4641         struct page *page;
4642         gfp_t gfp_head;
4643         int i;
4644 
4645         *errcode = -EMSGSIZE;
4646         /* Note this test could be relaxed, if we succeed to allocate
4647          * high order pages...
4648          */
4649         if (npages > MAX_SKB_FRAGS)
4650                 return NULL;
4651 
4652         gfp_head = gfp_mask;
4653         if (gfp_head & __GFP_DIRECT_RECLAIM)
4654                 gfp_head |= __GFP_REPEAT;
4655 
4656         *errcode = -ENOBUFS;
4657         skb = alloc_skb(header_len, gfp_head);
4658         if (!skb)
4659                 return NULL;
4660 
4661         skb->truesize += npages << PAGE_SHIFT;
4662 
4663         for (i = 0; npages > 0; i++) {
4664                 int order = max_page_order;
4665 
4666                 while (order) {
4667                         if (npages >= 1 << order) {
4668                                 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
4669                                                    __GFP_COMP |
4670                                                    __GFP_NOWARN |
4671                                                    __GFP_NORETRY,
4672                                                    order);
4673                                 if (page)
4674                                         goto fill_page;
4675                                 /* Do not retry other high order allocations */
4676                                 order = 1;
4677                                 max_page_order = 0;
4678                         }
4679                         order--;
4680                 }
4681                 page = alloc_page(gfp_mask);
4682                 if (!page)
4683                         goto failure;
4684 fill_page:
4685                 chunk = min_t(unsigned long, data_len,
4686                               PAGE_SIZE << order);
4687                 skb_fill_page_desc(skb, i, page, 0, chunk);
4688                 data_len -= chunk;
4689                 npages -= 1 << order;
4690         }
4691         return skb;
4692 
4693 failure:
4694         kfree_skb(skb);
4695         return NULL;
4696 }
4697 EXPORT_SYMBOL(alloc_skb_with_frags);
4698 
4699 /* carve out the first off bytes from skb when off < headlen */
4700 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
4701                                     const int headlen, gfp_t gfp_mask)
4702 {
4703         int i;
4704         int size = skb_end_offset(skb);
4705         int new_hlen = headlen - off;
4706         u8 *data;
4707 
4708         size = SKB_DATA_ALIGN(size);
4709 
4710         if (skb_pfmemalloc(skb))
4711                 gfp_mask |= __GFP_MEMALLOC;
4712         data = kmalloc_reserve(size +
4713                                SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
4714                                gfp_mask, NUMA_NO_NODE, NULL);
4715         if (!data)
4716                 return -ENOMEM;
4717 
4718         size = SKB_WITH_OVERHEAD(ksize(data));
4719 
4720         /* Copy real data, and all frags */
4721         skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
4722         skb->len -= off;
4723 
4724         memcpy((struct skb_shared_info *)(data + size),
4725                skb_shinfo(skb),
4726                offsetof(struct skb_shared_info,
4727                         frags[skb_shinfo(skb)->nr_frags]));
4728         if (skb_cloned(skb)) {
4729                 /* drop the old head gracefully */
4730                 if (skb_orphan_frags(skb, gfp_mask)) {
4731                         kfree(data);
4732                         return -ENOMEM;
4733                 }
4734                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
4735                         skb_frag_ref(skb, i);
4736                 if (skb_has_frag_list(skb))
4737                         skb_clone_fraglist(skb);
4738                 skb_release_data(skb);
4739         } else {
4740                 /* we can reuse existing recount- all we did was
4741                  * relocate values
4742                  */
4743                 skb_free_head(skb);
4744         }
4745 
4746         skb->head = data;
4747         skb->data = data;
4748         skb->head_frag = 0;
4749 #ifdef NET_SKBUFF_DATA_USES_OFFSET
4750         skb->end = size;
4751 #else
4752         skb->end = skb->head + size;
4753 #endif
4754         skb_set_tail_pointer(skb, skb_headlen(skb));
4755         skb_headers_offset_update(skb, 0);
4756         skb->cloned = 0;
4757         skb->hdr_len = 0;
4758         skb->nohdr = 0;
4759         atomic_set(&skb_shinfo(skb)->dataref, 1);
4760 
4761         return 0;
4762 }
4763 
4764 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
4765 
4766 /* carve out the first eat bytes from skb's frag_list. May recurse into
4767  * pskb_carve()
4768  */
4769 static int pskb_carve_frag_list(struct sk_buff *skb,
4770                                 struct skb_shared_info *shinfo, int eat,
4771                                 gfp_t gfp_mask)
4772 {
4773         struct sk_buff *list = shinfo->frag_list;
4774         struct sk_buff *clone = NULL;
4775         struct sk_buff *insp = NULL;
4776 
4777         do {
4778                 if (!list) {
4779                         pr_err("Not enough bytes to eat. Want %d\n", eat);
4780                         return -EFAULT;
4781                 }
4782                 if (list->len <= eat) {
4783                         /* Eaten as whole. */
4784                         eat -= list->len;
4785                         list = list->next;
4786                         insp = list;
4787                 } else {
4788                         /* Eaten partially. */
4789                         if (skb_shared(list)) {
4790                                 clone = skb_clone(list, gfp_mask);
4791                                 if (!clone)
4792                                         return -ENOMEM;
4793                                 insp = list->next;
4794                                 list = clone;
4795                         } else {
4796                                 /* This may be pulled without problems. */
4797                                 insp = list;
4798                         }
4799                         if (pskb_carve(list, eat, gfp_mask) < 0) {
4800                                 kfree_skb(clone);
4801                                 return -ENOMEM;
4802                         }
4803                         break;
4804                 }
4805         } while (eat);
4806 
4807         /* Free pulled out fragments. */
4808         while ((list = shinfo->frag_list) != insp) {
4809                 shinfo->frag_list = list->next;
4810                 kfree_skb(list);
4811         }
4812         /* And insert new clone at head. */
4813         if (clone) {
4814                 clone->next = list;
4815                 shinfo->frag_list = clone;
4816         }
4817         return 0;
4818 }
4819 
4820 /* carve off first len bytes from skb. Split line (off) is in the
4821  * non-linear part of skb
4822  */
4823 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
4824                                        int pos, gfp_t gfp_mask)
4825 {
4826         int i, k = 0;
4827         int size = skb_end_offset(skb);
4828         u8 *data;
4829         const int nfrags = skb_shinfo(skb)->nr_frags;
4830         struct skb_shared_info *shinfo;
4831 
4832         size = SKB_DATA_ALIGN(size);
4833 
4834         if (skb_pfmemalloc(skb))
4835                 gfp_mask |= __GFP_MEMALLOC;
4836         data = kmalloc_reserve(size +
4837                                SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
4838                                gfp_mask, NUMA_NO_NODE, NULL);
4839         if (!data)
4840                 return -ENOMEM;
4841 
4842         size = SKB_WITH_OVERHEAD(ksize(data));
4843 
4844         memcpy((struct skb_shared_info *)(data + size),
4845                skb_shinfo(skb), offsetof(struct skb_shared_info,
4846                                          frags[skb_shinfo(skb)->nr_frags]));
4847         if (skb_orphan_frags(skb, gfp_mask)) {
4848                 kfree(data);
4849                 return -ENOMEM;
4850         }
4851         shinfo = (struct skb_shared_info *)(data + size);
4852         for (i = 0; i < nfrags; i++) {
4853                 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
4854 
4855                 if (pos + fsize > off) {
4856                         shinfo->frags[k] = skb_shinfo(skb)->frags[i];
4857 
4858                         if (pos < off) {
4859                                 /* Split frag.
4860                                  * We have two variants in this case:
4861                                  * 1. Move all the frag to the second
4862                                  *    part, if it is possible. F.e.
4863                                  *    this approach is mandatory for TUX,
4864                                  *    where splitting is expensive.
4865                                  * 2. Split is accurately. We make this.
4866                                  */
4867                                 shinfo->frags[0].page_offset += off - pos;
4868                                 skb_frag_size_sub(&shinfo->frags[0], off - pos);
4869                         }
4870                         skb_frag_ref(skb, i);
4871                         k++;
4872                 }
4873                 pos += fsize;
4874         }
4875         shinfo->nr_frags = k;
4876         if (skb_has_frag_list(skb))
4877                 skb_clone_fraglist(skb);
4878 
4879         if (k == 0) {
4880                 /* split line is in frag list */
4881                 pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask);
4882         }
4883         skb_release_data(skb);
4884 
4885         skb->head = data;
4886         skb->head_frag = 0;
4887         skb->data = data;
4888 #ifdef NET_SKBUFF_DATA_USES_OFFSET
4889         skb->end = size;
4890 #else
4891         skb->end = skb->head + size;
4892 #endif
4893         skb_reset_tail_pointer(skb);
4894         skb_headers_offset_update(skb, 0);
4895         skb->cloned   = 0;
4896         skb->hdr_len  = 0;
4897         skb->nohdr    = 0;
4898         skb->len -= off;
4899         skb->data_len = skb->len;
4900         atomic_set(&skb_shinfo(skb)->dataref, 1);
4901         return 0;
4902 }
4903 
4904 /* remove len bytes from the beginning of the skb */
4905 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
4906 {
4907         int headlen = skb_headlen(skb);
4908 
4909         if (len < headlen)
4910                 return pskb_carve_inside_header(skb, len, headlen, gfp);
4911         else
4912                 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
4913 }
4914 
4915 /* Extract to_copy bytes starting at off from skb, and return this in
4916  * a new skb
4917  */
4918 struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
4919                              int to_copy, gfp_t gfp)
4920 {
4921         struct sk_buff  *clone = skb_clone(skb, gfp);
4922 
4923         if (!clone)
4924                 return NULL;
4925 
4926         if (pskb_carve(clone, off, gfp) < 0 ||
4927             pskb_trim(clone, to_copy)) {
4928                 kfree_skb(clone);
4929                 return NULL;
4930         }
4931         return clone;
4932 }
4933 EXPORT_SYMBOL(pskb_extract);
4934 
4935 /**
4936  * skb_condense - try to get rid of fragments/frag_list if possible
4937  * @skb: buffer
4938  *
4939  * Can be used to save memory before skb is added to a busy queue.
4940  * If packet has bytes in frags and enough tail room in skb->head,
4941  * pull all of them, so that we can free the frags right now and adjust
4942  * truesize.
4943  * Notes:
4944  *      We do not reallocate skb->head thus can not fail.
4945  *      Caller must re-evaluate skb->truesize if needed.
4946  */
4947 void skb_condense(struct sk_buff *skb)
4948 {
4949         if (skb->data_len) {
4950                 if (skb->data_len > skb->end - skb->tail ||
4951                     skb_cloned(skb))
4952                         return;
4953 
4954                 /* Nice, we can free page frag(s) right now */
4955                 __pskb_pull_tail(skb, skb->data_len);
4956         }
4957         /* At this point, skb->truesize might be over estimated,
4958          * because skb had a fragment, and fragments do not tell
4959          * their truesize.
4960          * When we pulled its content into skb->head, fragment
4961          * was freed, but __pskb_pull_tail() could not possibly
4962          * adjust skb->truesize, not knowing the frag truesize.
4963          */
4964         skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4965 }
4966 

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