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

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

Version: ~ [ linux-5.8 ] ~ [ linux-5.7.12 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.55 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.136 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.191 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.232 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.232 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0-or-later
  2 /*
  3  *      Routines having to do with the 'struct sk_buff' memory handlers.
  4  *
  5  *      Authors:        Alan Cox <alan@lxorguk.ukuu.org.uk>
  6  *                      Florian La Roche <rzsfl@rz.uni-sb.de>
  7  *
  8  *      Fixes:
  9  *              Alan Cox        :       Fixed the worst of the load
 10  *                                      balancer bugs.
 11  *              Dave Platt      :       Interrupt stacking fix.
 12  *      Richard Kooijman        :       Timestamp fixes.
 13  *              Alan Cox        :       Changed buffer format.
 14  *              Alan Cox        :       destructor hook for AF_UNIX etc.
 15  *              Linus Torvalds  :       Better skb_clone.
 16  *              Alan Cox        :       Added skb_copy.
 17  *              Alan Cox        :       Added all the changed routines Linus
 18  *                                      only put in the headers
 19  *              Ray VanTassle   :       Fixed --skb->lock in free
 20  *              Alan Cox        :       skb_copy copy arp field
 21  *              Andi Kleen      :       slabified it.
 22  *              Robert Olsson   :       Removed skb_head_pool
 23  *
 24  *      NOTE:
 25  *              The __skb_ routines should be called with interrupts
 26  *      disabled, or you better be *real* sure that the operation is atomic
 27  *      with respect to whatever list is being frobbed (e.g. via lock_sock()
 28  *      or via disabling bottom half handlers, etc).
 29  */
 30 
 31 /*
 32  *      The functions in this file will not compile correctly with gcc 2.4.x
 33  */
 34 
 35 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 36 
 37 #include <linux/module.h>
 38 #include <linux/types.h>
 39 #include <linux/kernel.h>
 40 #include <linux/mm.h>
 41 #include <linux/interrupt.h>
 42 #include <linux/in.h>
 43 #include <linux/inet.h>
 44 #include <linux/slab.h>
 45 #include <linux/tcp.h>
 46 #include <linux/udp.h>
 47 #include <linux/sctp.h>
 48 #include <linux/netdevice.h>
 49 #ifdef CONFIG_NET_CLS_ACT
 50 #include <net/pkt_sched.h>
 51 #endif
 52 #include <linux/string.h>
 53 #include <linux/skbuff.h>
 54 #include <linux/splice.h>
 55 #include <linux/cache.h>
 56 #include <linux/rtnetlink.h>
 57 #include <linux/init.h>
 58 #include <linux/scatterlist.h>
 59 #include <linux/errqueue.h>
 60 #include <linux/prefetch.h>
 61 #include <linux/if_vlan.h>
 62 #include <linux/mpls.h>
 63 
 64 #include <net/protocol.h>
 65 #include <net/dst.h>
 66 #include <net/sock.h>
 67 #include <net/checksum.h>
 68 #include <net/ip6_checksum.h>
 69 #include <net/xfrm.h>
 70 #include <net/mpls.h>
 71 #include <net/mptcp.h>
 72 
 73 #include <linux/uaccess.h>
 74 #include <trace/events/skb.h>
 75 #include <linux/highmem.h>
 76 #include <linux/capability.h>
 77 #include <linux/user_namespace.h>
 78 #include <linux/indirect_call_wrapper.h>
 79 
 80 #include "datagram.h"
 81 
 82 struct kmem_cache *skbuff_head_cache __ro_after_init;
 83 static struct kmem_cache *skbuff_fclone_cache __ro_after_init;
 84 #ifdef CONFIG_SKB_EXTENSIONS
 85 static struct kmem_cache *skbuff_ext_cache __ro_after_init;
 86 #endif
 87 int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
 88 EXPORT_SYMBOL(sysctl_max_skb_frags);
 89 
 90 /**
 91  *      skb_panic - private function for out-of-line support
 92  *      @skb:   buffer
 93  *      @sz:    size
 94  *      @addr:  address
 95  *      @msg:   skb_over_panic or skb_under_panic
 96  *
 97  *      Out-of-line support for skb_put() and skb_push().
 98  *      Called via the wrapper skb_over_panic() or skb_under_panic().
 99  *      Keep out of line to prevent kernel bloat.
100  *      __builtin_return_address is not used because it is not always reliable.
101  */
102 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
103                       const char msg[])
104 {
105         pr_emerg("%s: text:%px len:%d put:%d head:%px data:%px tail:%#lx end:%#lx dev:%s\n",
106                  msg, addr, skb->len, sz, skb->head, skb->data,
107                  (unsigned long)skb->tail, (unsigned long)skb->end,
108                  skb->dev ? skb->dev->name : "<NULL>");
109         BUG();
110 }
111 
112 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
113 {
114         skb_panic(skb, sz, addr, __func__);
115 }
116 
117 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
118 {
119         skb_panic(skb, sz, addr, __func__);
120 }
121 
122 /*
123  * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
124  * the caller if emergency pfmemalloc reserves are being used. If it is and
125  * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
126  * may be used. Otherwise, the packet data may be discarded until enough
127  * memory is free
128  */
129 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
130          __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
131 
132 static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
133                                unsigned long ip, bool *pfmemalloc)
134 {
135         void *obj;
136         bool ret_pfmemalloc = false;
137 
138         /*
139          * Try a regular allocation, when that fails and we're not entitled
140          * to the reserves, fail.
141          */
142         obj = kmalloc_node_track_caller(size,
143                                         flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
144                                         node);
145         if (obj || !(gfp_pfmemalloc_allowed(flags)))
146                 goto out;
147 
148         /* Try again but now we are using pfmemalloc reserves */
149         ret_pfmemalloc = true;
150         obj = kmalloc_node_track_caller(size, flags, node);
151 
152 out:
153         if (pfmemalloc)
154                 *pfmemalloc = ret_pfmemalloc;
155 
156         return obj;
157 }
158 
159 /*      Allocate a new skbuff. We do this ourselves so we can fill in a few
160  *      'private' fields and also do memory statistics to find all the
161  *      [BEEP] leaks.
162  *
163  */
164 
165 /**
166  *      __alloc_skb     -       allocate a network buffer
167  *      @size: size to allocate
168  *      @gfp_mask: allocation mask
169  *      @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
170  *              instead of head cache and allocate a cloned (child) skb.
171  *              If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
172  *              allocations in case the data is required for writeback
173  *      @node: numa node to allocate memory on
174  *
175  *      Allocate a new &sk_buff. The returned buffer has no headroom and a
176  *      tail room of at least size bytes. The object has a reference count
177  *      of one. The return is the buffer. On a failure the return is %NULL.
178  *
179  *      Buffers may only be allocated from interrupts using a @gfp_mask of
180  *      %GFP_ATOMIC.
181  */
182 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
183                             int flags, int node)
184 {
185         struct kmem_cache *cache;
186         struct skb_shared_info *shinfo;
187         struct sk_buff *skb;
188         u8 *data;
189         bool pfmemalloc;
190 
191         cache = (flags & SKB_ALLOC_FCLONE)
192                 ? skbuff_fclone_cache : skbuff_head_cache;
193 
194         if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
195                 gfp_mask |= __GFP_MEMALLOC;
196 
197         /* Get the HEAD */
198         skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
199         if (!skb)
200                 goto out;
201         prefetchw(skb);
202 
203         /* We do our best to align skb_shared_info on a separate cache
204          * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
205          * aligned memory blocks, unless SLUB/SLAB debug is enabled.
206          * Both skb->head and skb_shared_info are cache line aligned.
207          */
208         size = SKB_DATA_ALIGN(size);
209         size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
210         data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
211         if (!data)
212                 goto nodata;
213         /* kmalloc(size) might give us more room than requested.
214          * Put skb_shared_info exactly at the end of allocated zone,
215          * to allow max possible filling before reallocation.
216          */
217         size = SKB_WITH_OVERHEAD(ksize(data));
218         prefetchw(data + size);
219 
220         /*
221          * Only clear those fields we need to clear, not those that we will
222          * actually initialise below. Hence, don't put any more fields after
223          * the tail pointer in struct sk_buff!
224          */
225         memset(skb, 0, offsetof(struct sk_buff, tail));
226         /* Account for allocated memory : skb + skb->head */
227         skb->truesize = SKB_TRUESIZE(size);
228         skb->pfmemalloc = pfmemalloc;
229         refcount_set(&skb->users, 1);
230         skb->head = data;
231         skb->data = data;
232         skb_reset_tail_pointer(skb);
233         skb->end = skb->tail + size;
234         skb->mac_header = (typeof(skb->mac_header))~0U;
235         skb->transport_header = (typeof(skb->transport_header))~0U;
236 
237         /* make sure we initialize shinfo sequentially */
238         shinfo = skb_shinfo(skb);
239         memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
240         atomic_set(&shinfo->dataref, 1);
241 
242         if (flags & SKB_ALLOC_FCLONE) {
243                 struct sk_buff_fclones *fclones;
244 
245                 fclones = container_of(skb, struct sk_buff_fclones, skb1);
246 
247                 skb->fclone = SKB_FCLONE_ORIG;
248                 refcount_set(&fclones->fclone_ref, 1);
249 
250                 fclones->skb2.fclone = SKB_FCLONE_CLONE;
251         }
252 out:
253         return skb;
254 nodata:
255         kmem_cache_free(cache, skb);
256         skb = NULL;
257         goto out;
258 }
259 EXPORT_SYMBOL(__alloc_skb);
260 
261 /* Caller must provide SKB that is memset cleared */
262 static struct sk_buff *__build_skb_around(struct sk_buff *skb,
263                                           void *data, unsigned int frag_size)
264 {
265         struct skb_shared_info *shinfo;
266         unsigned int size = frag_size ? : ksize(data);
267 
268         size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
269 
270         /* Assumes caller memset cleared SKB */
271         skb->truesize = SKB_TRUESIZE(size);
272         refcount_set(&skb->users, 1);
273         skb->head = data;
274         skb->data = data;
275         skb_reset_tail_pointer(skb);
276         skb->end = skb->tail + size;
277         skb->mac_header = (typeof(skb->mac_header))~0U;
278         skb->transport_header = (typeof(skb->transport_header))~0U;
279 
280         /* make sure we initialize shinfo sequentially */
281         shinfo = skb_shinfo(skb);
282         memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
283         atomic_set(&shinfo->dataref, 1);
284 
285         return skb;
286 }
287 
288 /**
289  * __build_skb - build a network buffer
290  * @data: data buffer provided by caller
291  * @frag_size: size of data, or 0 if head was kmalloced
292  *
293  * Allocate a new &sk_buff. Caller provides space holding head and
294  * skb_shared_info. @data must have been allocated by kmalloc() only if
295  * @frag_size is 0, otherwise data should come from the page allocator
296  *  or vmalloc()
297  * The return is the new skb buffer.
298  * On a failure the return is %NULL, and @data is not freed.
299  * Notes :
300  *  Before IO, driver allocates only data buffer where NIC put incoming frame
301  *  Driver should add room at head (NET_SKB_PAD) and
302  *  MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
303  *  After IO, driver calls build_skb(), to allocate sk_buff and populate it
304  *  before giving packet to stack.
305  *  RX rings only contains data buffers, not full skbs.
306  */
307 struct sk_buff *__build_skb(void *data, unsigned int frag_size)
308 {
309         struct sk_buff *skb;
310 
311         skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
312         if (unlikely(!skb))
313                 return NULL;
314 
315         memset(skb, 0, offsetof(struct sk_buff, tail));
316 
317         return __build_skb_around(skb, data, frag_size);
318 }
319 
320 /* build_skb() is wrapper over __build_skb(), that specifically
321  * takes care of skb->head and skb->pfmemalloc
322  * This means that if @frag_size is not zero, then @data must be backed
323  * by a page fragment, not kmalloc() or vmalloc()
324  */
325 struct sk_buff *build_skb(void *data, unsigned int frag_size)
326 {
327         struct sk_buff *skb = __build_skb(data, frag_size);
328 
329         if (skb && frag_size) {
330                 skb->head_frag = 1;
331                 if (page_is_pfmemalloc(virt_to_head_page(data)))
332                         skb->pfmemalloc = 1;
333         }
334         return skb;
335 }
336 EXPORT_SYMBOL(build_skb);
337 
338 /**
339  * build_skb_around - build a network buffer around provided skb
340  * @skb: sk_buff provide by caller, must be memset cleared
341  * @data: data buffer provided by caller
342  * @frag_size: size of data, or 0 if head was kmalloced
343  */
344 struct sk_buff *build_skb_around(struct sk_buff *skb,
345                                  void *data, unsigned int frag_size)
346 {
347         if (unlikely(!skb))
348                 return NULL;
349 
350         skb = __build_skb_around(skb, data, frag_size);
351 
352         if (skb && frag_size) {
353                 skb->head_frag = 1;
354                 if (page_is_pfmemalloc(virt_to_head_page(data)))
355                         skb->pfmemalloc = 1;
356         }
357         return skb;
358 }
359 EXPORT_SYMBOL(build_skb_around);
360 
361 #define NAPI_SKB_CACHE_SIZE     64
362 
363 struct napi_alloc_cache {
364         struct page_frag_cache page;
365         unsigned int skb_count;
366         void *skb_cache[NAPI_SKB_CACHE_SIZE];
367 };
368 
369 static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
370 static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
371 
372 static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
373 {
374         struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
375 
376         return page_frag_alloc(&nc->page, fragsz, gfp_mask);
377 }
378 
379 void *napi_alloc_frag(unsigned int fragsz)
380 {
381         fragsz = SKB_DATA_ALIGN(fragsz);
382 
383         return __napi_alloc_frag(fragsz, GFP_ATOMIC);
384 }
385 EXPORT_SYMBOL(napi_alloc_frag);
386 
387 /**
388  * netdev_alloc_frag - allocate a page fragment
389  * @fragsz: fragment size
390  *
391  * Allocates a frag from a page for receive buffer.
392  * Uses GFP_ATOMIC allocations.
393  */
394 void *netdev_alloc_frag(unsigned int fragsz)
395 {
396         struct page_frag_cache *nc;
397         void *data;
398 
399         fragsz = SKB_DATA_ALIGN(fragsz);
400         if (in_irq() || irqs_disabled()) {
401                 nc = this_cpu_ptr(&netdev_alloc_cache);
402                 data = page_frag_alloc(nc, fragsz, GFP_ATOMIC);
403         } else {
404                 local_bh_disable();
405                 data = __napi_alloc_frag(fragsz, GFP_ATOMIC);
406                 local_bh_enable();
407         }
408         return data;
409 }
410 EXPORT_SYMBOL(netdev_alloc_frag);
411 
412 /**
413  *      __netdev_alloc_skb - allocate an skbuff for rx on a specific device
414  *      @dev: network device to receive on
415  *      @len: length to allocate
416  *      @gfp_mask: get_free_pages mask, passed to alloc_skb
417  *
418  *      Allocate a new &sk_buff and assign it a usage count of one. The
419  *      buffer has NET_SKB_PAD headroom built in. Users should allocate
420  *      the headroom they think they need without accounting for the
421  *      built in space. The built in space is used for optimisations.
422  *
423  *      %NULL is returned if there is no free memory.
424  */
425 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
426                                    gfp_t gfp_mask)
427 {
428         struct page_frag_cache *nc;
429         struct sk_buff *skb;
430         bool pfmemalloc;
431         void *data;
432 
433         len += NET_SKB_PAD;
434 
435         if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
436             (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
437                 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
438                 if (!skb)
439                         goto skb_fail;
440                 goto skb_success;
441         }
442 
443         len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
444         len = SKB_DATA_ALIGN(len);
445 
446         if (sk_memalloc_socks())
447                 gfp_mask |= __GFP_MEMALLOC;
448 
449         if (in_irq() || irqs_disabled()) {
450                 nc = this_cpu_ptr(&netdev_alloc_cache);
451                 data = page_frag_alloc(nc, len, gfp_mask);
452                 pfmemalloc = nc->pfmemalloc;
453         } else {
454                 local_bh_disable();
455                 nc = this_cpu_ptr(&napi_alloc_cache.page);
456                 data = page_frag_alloc(nc, len, gfp_mask);
457                 pfmemalloc = nc->pfmemalloc;
458                 local_bh_enable();
459         }
460 
461         if (unlikely(!data))
462                 return NULL;
463 
464         skb = __build_skb(data, len);
465         if (unlikely(!skb)) {
466                 skb_free_frag(data);
467                 return NULL;
468         }
469 
470         if (pfmemalloc)
471                 skb->pfmemalloc = 1;
472         skb->head_frag = 1;
473 
474 skb_success:
475         skb_reserve(skb, NET_SKB_PAD);
476         skb->dev = dev;
477 
478 skb_fail:
479         return skb;
480 }
481 EXPORT_SYMBOL(__netdev_alloc_skb);
482 
483 /**
484  *      __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
485  *      @napi: napi instance this buffer was allocated for
486  *      @len: length to allocate
487  *      @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
488  *
489  *      Allocate a new sk_buff for use in NAPI receive.  This buffer will
490  *      attempt to allocate the head from a special reserved region used
491  *      only for NAPI Rx allocation.  By doing this we can save several
492  *      CPU cycles by avoiding having to disable and re-enable IRQs.
493  *
494  *      %NULL is returned if there is no free memory.
495  */
496 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
497                                  gfp_t gfp_mask)
498 {
499         struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
500         struct sk_buff *skb;
501         void *data;
502 
503         len += NET_SKB_PAD + NET_IP_ALIGN;
504 
505         if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
506             (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
507                 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
508                 if (!skb)
509                         goto skb_fail;
510                 goto skb_success;
511         }
512 
513         len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
514         len = SKB_DATA_ALIGN(len);
515 
516         if (sk_memalloc_socks())
517                 gfp_mask |= __GFP_MEMALLOC;
518 
519         data = page_frag_alloc(&nc->page, len, gfp_mask);
520         if (unlikely(!data))
521                 return NULL;
522 
523         skb = __build_skb(data, len);
524         if (unlikely(!skb)) {
525                 skb_free_frag(data);
526                 return NULL;
527         }
528 
529         if (nc->page.pfmemalloc)
530                 skb->pfmemalloc = 1;
531         skb->head_frag = 1;
532 
533 skb_success:
534         skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
535         skb->dev = napi->dev;
536 
537 skb_fail:
538         return skb;
539 }
540 EXPORT_SYMBOL(__napi_alloc_skb);
541 
542 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
543                      int size, unsigned int truesize)
544 {
545         skb_fill_page_desc(skb, i, page, off, size);
546         skb->len += size;
547         skb->data_len += size;
548         skb->truesize += truesize;
549 }
550 EXPORT_SYMBOL(skb_add_rx_frag);
551 
552 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
553                           unsigned int truesize)
554 {
555         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
556 
557         skb_frag_size_add(frag, size);
558         skb->len += size;
559         skb->data_len += size;
560         skb->truesize += truesize;
561 }
562 EXPORT_SYMBOL(skb_coalesce_rx_frag);
563 
564 static void skb_drop_list(struct sk_buff **listp)
565 {
566         kfree_skb_list(*listp);
567         *listp = NULL;
568 }
569 
570 static inline void skb_drop_fraglist(struct sk_buff *skb)
571 {
572         skb_drop_list(&skb_shinfo(skb)->frag_list);
573 }
574 
575 static void skb_clone_fraglist(struct sk_buff *skb)
576 {
577         struct sk_buff *list;
578 
579         skb_walk_frags(skb, list)
580                 skb_get(list);
581 }
582 
583 static void skb_free_head(struct sk_buff *skb)
584 {
585         unsigned char *head = skb->head;
586 
587         if (skb->head_frag)
588                 skb_free_frag(head);
589         else
590                 kfree(head);
591 }
592 
593 static void skb_release_data(struct sk_buff *skb)
594 {
595         struct skb_shared_info *shinfo = skb_shinfo(skb);
596         int i;
597 
598         if (skb->cloned &&
599             atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
600                               &shinfo->dataref))
601                 return;
602 
603         for (i = 0; i < shinfo->nr_frags; i++)
604                 __skb_frag_unref(&shinfo->frags[i]);
605 
606         if (shinfo->frag_list)
607                 kfree_skb_list(shinfo->frag_list);
608 
609         skb_zcopy_clear(skb, true);
610         skb_free_head(skb);
611 }
612 
613 /*
614  *      Free an skbuff by memory without cleaning the state.
615  */
616 static void kfree_skbmem(struct sk_buff *skb)
617 {
618         struct sk_buff_fclones *fclones;
619 
620         switch (skb->fclone) {
621         case SKB_FCLONE_UNAVAILABLE:
622                 kmem_cache_free(skbuff_head_cache, skb);
623                 return;
624 
625         case SKB_FCLONE_ORIG:
626                 fclones = container_of(skb, struct sk_buff_fclones, skb1);
627 
628                 /* We usually free the clone (TX completion) before original skb
629                  * This test would have no chance to be true for the clone,
630                  * while here, branch prediction will be good.
631                  */
632                 if (refcount_read(&fclones->fclone_ref) == 1)
633                         goto fastpath;
634                 break;
635 
636         default: /* SKB_FCLONE_CLONE */
637                 fclones = container_of(skb, struct sk_buff_fclones, skb2);
638                 break;
639         }
640         if (!refcount_dec_and_test(&fclones->fclone_ref))
641                 return;
642 fastpath:
643         kmem_cache_free(skbuff_fclone_cache, fclones);
644 }
645 
646 void skb_release_head_state(struct sk_buff *skb)
647 {
648         skb_dst_drop(skb);
649         if (skb->destructor) {
650                 WARN_ON(in_irq());
651                 skb->destructor(skb);
652         }
653 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
654         nf_conntrack_put(skb_nfct(skb));
655 #endif
656         skb_ext_put(skb);
657 }
658 
659 /* Free everything but the sk_buff shell. */
660 static void skb_release_all(struct sk_buff *skb)
661 {
662         skb_release_head_state(skb);
663         if (likely(skb->head))
664                 skb_release_data(skb);
665 }
666 
667 /**
668  *      __kfree_skb - private function
669  *      @skb: buffer
670  *
671  *      Free an sk_buff. Release anything attached to the buffer.
672  *      Clean the state. This is an internal helper function. Users should
673  *      always call kfree_skb
674  */
675 
676 void __kfree_skb(struct sk_buff *skb)
677 {
678         skb_release_all(skb);
679         kfree_skbmem(skb);
680 }
681 EXPORT_SYMBOL(__kfree_skb);
682 
683 /**
684  *      kfree_skb - free an sk_buff
685  *      @skb: buffer to free
686  *
687  *      Drop a reference to the buffer and free it if the usage count has
688  *      hit zero.
689  */
690 void kfree_skb(struct sk_buff *skb)
691 {
692         if (!skb_unref(skb))
693                 return;
694 
695         trace_kfree_skb(skb, __builtin_return_address(0));
696         __kfree_skb(skb);
697 }
698 EXPORT_SYMBOL(kfree_skb);
699 
700 void kfree_skb_list(struct sk_buff *segs)
701 {
702         while (segs) {
703                 struct sk_buff *next = segs->next;
704 
705                 kfree_skb(segs);
706                 segs = next;
707         }
708 }
709 EXPORT_SYMBOL(kfree_skb_list);
710 
711 /* Dump skb information and contents.
712  *
713  * Must only be called from net_ratelimit()-ed paths.
714  *
715  * Dumps up to can_dump_full whole packets if full_pkt, headers otherwise.
716  */
717 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt)
718 {
719         static atomic_t can_dump_full = ATOMIC_INIT(5);
720         struct skb_shared_info *sh = skb_shinfo(skb);
721         struct net_device *dev = skb->dev;
722         struct sock *sk = skb->sk;
723         struct sk_buff *list_skb;
724         bool has_mac, has_trans;
725         int headroom, tailroom;
726         int i, len, seg_len;
727 
728         if (full_pkt)
729                 full_pkt = atomic_dec_if_positive(&can_dump_full) >= 0;
730 
731         if (full_pkt)
732                 len = skb->len;
733         else
734                 len = min_t(int, skb->len, MAX_HEADER + 128);
735 
736         headroom = skb_headroom(skb);
737         tailroom = skb_tailroom(skb);
738 
739         has_mac = skb_mac_header_was_set(skb);
740         has_trans = skb_transport_header_was_set(skb);
741 
742         printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n"
743                "mac=(%d,%d) net=(%d,%d) trans=%d\n"
744                "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n"
745                "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n"
746                "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n",
747                level, skb->len, headroom, skb_headlen(skb), tailroom,
748                has_mac ? skb->mac_header : -1,
749                has_mac ? skb_mac_header_len(skb) : -1,
750                skb->network_header,
751                has_trans ? skb_network_header_len(skb) : -1,
752                has_trans ? skb->transport_header : -1,
753                sh->tx_flags, sh->nr_frags,
754                sh->gso_size, sh->gso_type, sh->gso_segs,
755                skb->csum, skb->ip_summed, skb->csum_complete_sw,
756                skb->csum_valid, skb->csum_level,
757                skb->hash, skb->sw_hash, skb->l4_hash,
758                ntohs(skb->protocol), skb->pkt_type, skb->skb_iif);
759 
760         if (dev)
761                 printk("%sdev name=%s feat=0x%pNF\n",
762                        level, dev->name, &dev->features);
763         if (sk)
764                 printk("%ssk family=%hu type=%u proto=%u\n",
765                        level, sk->sk_family, sk->sk_type, sk->sk_protocol);
766 
767         if (full_pkt && headroom)
768                 print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET,
769                                16, 1, skb->head, headroom, false);
770 
771         seg_len = min_t(int, skb_headlen(skb), len);
772         if (seg_len)
773                 print_hex_dump(level, "skb linear:   ", DUMP_PREFIX_OFFSET,
774                                16, 1, skb->data, seg_len, false);
775         len -= seg_len;
776 
777         if (full_pkt && tailroom)
778                 print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET,
779                                16, 1, skb_tail_pointer(skb), tailroom, false);
780 
781         for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) {
782                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
783                 u32 p_off, p_len, copied;
784                 struct page *p;
785                 u8 *vaddr;
786 
787                 skb_frag_foreach_page(frag, skb_frag_off(frag),
788                                       skb_frag_size(frag), p, p_off, p_len,
789                                       copied) {
790                         seg_len = min_t(int, p_len, len);
791                         vaddr = kmap_atomic(p);
792                         print_hex_dump(level, "skb frag:     ",
793                                        DUMP_PREFIX_OFFSET,
794                                        16, 1, vaddr + p_off, seg_len, false);
795                         kunmap_atomic(vaddr);
796                         len -= seg_len;
797                         if (!len)
798                                 break;
799                 }
800         }
801 
802         if (full_pkt && skb_has_frag_list(skb)) {
803                 printk("skb fraglist:\n");
804                 skb_walk_frags(skb, list_skb)
805                         skb_dump(level, list_skb, true);
806         }
807 }
808 EXPORT_SYMBOL(skb_dump);
809 
810 /**
811  *      skb_tx_error - report an sk_buff xmit error
812  *      @skb: buffer that triggered an error
813  *
814  *      Report xmit error if a device callback is tracking this skb.
815  *      skb must be freed afterwards.
816  */
817 void skb_tx_error(struct sk_buff *skb)
818 {
819         skb_zcopy_clear(skb, true);
820 }
821 EXPORT_SYMBOL(skb_tx_error);
822 
823 /**
824  *      consume_skb - free an skbuff
825  *      @skb: buffer to free
826  *
827  *      Drop a ref to the buffer and free it if the usage count has hit zero
828  *      Functions identically to kfree_skb, but kfree_skb assumes that the frame
829  *      is being dropped after a failure and notes that
830  */
831 void consume_skb(struct sk_buff *skb)
832 {
833         if (!skb_unref(skb))
834                 return;
835 
836         trace_consume_skb(skb);
837         __kfree_skb(skb);
838 }
839 EXPORT_SYMBOL(consume_skb);
840 
841 /**
842  *      consume_stateless_skb - free an skbuff, assuming it is stateless
843  *      @skb: buffer to free
844  *
845  *      Alike consume_skb(), but this variant assumes that this is the last
846  *      skb reference and all the head states have been already dropped
847  */
848 void __consume_stateless_skb(struct sk_buff *skb)
849 {
850         trace_consume_skb(skb);
851         skb_release_data(skb);
852         kfree_skbmem(skb);
853 }
854 
855 void __kfree_skb_flush(void)
856 {
857         struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
858 
859         /* flush skb_cache if containing objects */
860         if (nc->skb_count) {
861                 kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count,
862                                      nc->skb_cache);
863                 nc->skb_count = 0;
864         }
865 }
866 
867 static inline void _kfree_skb_defer(struct sk_buff *skb)
868 {
869         struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
870 
871         /* drop skb->head and call any destructors for packet */
872         skb_release_all(skb);
873 
874         /* record skb to CPU local list */
875         nc->skb_cache[nc->skb_count++] = skb;
876 
877 #ifdef CONFIG_SLUB
878         /* SLUB writes into objects when freeing */
879         prefetchw(skb);
880 #endif
881 
882         /* flush skb_cache if it is filled */
883         if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
884                 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE,
885                                      nc->skb_cache);
886                 nc->skb_count = 0;
887         }
888 }
889 void __kfree_skb_defer(struct sk_buff *skb)
890 {
891         _kfree_skb_defer(skb);
892 }
893 
894 void napi_consume_skb(struct sk_buff *skb, int budget)
895 {
896         if (unlikely(!skb))
897                 return;
898 
899         /* Zero budget indicate non-NAPI context called us, like netpoll */
900         if (unlikely(!budget)) {
901                 dev_consume_skb_any(skb);
902                 return;
903         }
904 
905         if (!skb_unref(skb))
906                 return;
907 
908         /* if reaching here SKB is ready to free */
909         trace_consume_skb(skb);
910 
911         /* if SKB is a clone, don't handle this case */
912         if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
913                 __kfree_skb(skb);
914                 return;
915         }
916 
917         _kfree_skb_defer(skb);
918 }
919 EXPORT_SYMBOL(napi_consume_skb);
920 
921 /* Make sure a field is enclosed inside headers_start/headers_end section */
922 #define CHECK_SKB_FIELD(field) \
923         BUILD_BUG_ON(offsetof(struct sk_buff, field) <          \
924                      offsetof(struct sk_buff, headers_start));  \
925         BUILD_BUG_ON(offsetof(struct sk_buff, field) >          \
926                      offsetof(struct sk_buff, headers_end));    \
927 
928 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
929 {
930         new->tstamp             = old->tstamp;
931         /* We do not copy old->sk */
932         new->dev                = old->dev;
933         memcpy(new->cb, old->cb, sizeof(old->cb));
934         skb_dst_copy(new, old);
935         __skb_ext_copy(new, old);
936         __nf_copy(new, old, false);
937 
938         /* Note : this field could be in headers_start/headers_end section
939          * It is not yet because we do not want to have a 16 bit hole
940          */
941         new->queue_mapping = old->queue_mapping;
942 
943         memcpy(&new->headers_start, &old->headers_start,
944                offsetof(struct sk_buff, headers_end) -
945                offsetof(struct sk_buff, headers_start));
946         CHECK_SKB_FIELD(protocol);
947         CHECK_SKB_FIELD(csum);
948         CHECK_SKB_FIELD(hash);
949         CHECK_SKB_FIELD(priority);
950         CHECK_SKB_FIELD(skb_iif);
951         CHECK_SKB_FIELD(vlan_proto);
952         CHECK_SKB_FIELD(vlan_tci);
953         CHECK_SKB_FIELD(transport_header);
954         CHECK_SKB_FIELD(network_header);
955         CHECK_SKB_FIELD(mac_header);
956         CHECK_SKB_FIELD(inner_protocol);
957         CHECK_SKB_FIELD(inner_transport_header);
958         CHECK_SKB_FIELD(inner_network_header);
959         CHECK_SKB_FIELD(inner_mac_header);
960         CHECK_SKB_FIELD(mark);
961 #ifdef CONFIG_NETWORK_SECMARK
962         CHECK_SKB_FIELD(secmark);
963 #endif
964 #ifdef CONFIG_NET_RX_BUSY_POLL
965         CHECK_SKB_FIELD(napi_id);
966 #endif
967 #ifdef CONFIG_XPS
968         CHECK_SKB_FIELD(sender_cpu);
969 #endif
970 #ifdef CONFIG_NET_SCHED
971         CHECK_SKB_FIELD(tc_index);
972 #endif
973 
974 }
975 
976 /*
977  * You should not add any new code to this function.  Add it to
978  * __copy_skb_header above instead.
979  */
980 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
981 {
982 #define C(x) n->x = skb->x
983 
984         n->next = n->prev = NULL;
985         n->sk = NULL;
986         __copy_skb_header(n, skb);
987 
988         C(len);
989         C(data_len);
990         C(mac_len);
991         n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
992         n->cloned = 1;
993         n->nohdr = 0;
994         n->peeked = 0;
995         C(pfmemalloc);
996         n->destructor = NULL;
997         C(tail);
998         C(end);
999         C(head);
1000         C(head_frag);
1001         C(data);
1002         C(truesize);
1003         refcount_set(&n->users, 1);
1004 
1005         atomic_inc(&(skb_shinfo(skb)->dataref));
1006         skb->cloned = 1;
1007 
1008         return n;
1009 #undef C
1010 }
1011 
1012 /**
1013  * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg
1014  * @first: first sk_buff of the msg
1015  */
1016 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first)
1017 {
1018         struct sk_buff *n;
1019 
1020         n = alloc_skb(0, GFP_ATOMIC);
1021         if (!n)
1022                 return NULL;
1023 
1024         n->len = first->len;
1025         n->data_len = first->len;
1026         n->truesize = first->truesize;
1027 
1028         skb_shinfo(n)->frag_list = first;
1029 
1030         __copy_skb_header(n, first);
1031         n->destructor = NULL;
1032 
1033         return n;
1034 }
1035 EXPORT_SYMBOL_GPL(alloc_skb_for_msg);
1036 
1037 /**
1038  *      skb_morph       -       morph one skb into another
1039  *      @dst: the skb to receive the contents
1040  *      @src: the skb to supply the contents
1041  *
1042  *      This is identical to skb_clone except that the target skb is
1043  *      supplied by the user.
1044  *
1045  *      The target skb is returned upon exit.
1046  */
1047 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
1048 {
1049         skb_release_all(dst);
1050         return __skb_clone(dst, src);
1051 }
1052 EXPORT_SYMBOL_GPL(skb_morph);
1053 
1054 int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
1055 {
1056         unsigned long max_pg, num_pg, new_pg, old_pg;
1057         struct user_struct *user;
1058 
1059         if (capable(CAP_IPC_LOCK) || !size)
1060                 return 0;
1061 
1062         num_pg = (size >> PAGE_SHIFT) + 2;      /* worst case */
1063         max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1064         user = mmp->user ? : current_user();
1065 
1066         do {
1067                 old_pg = atomic_long_read(&user->locked_vm);
1068                 new_pg = old_pg + num_pg;
1069                 if (new_pg > max_pg)
1070                         return -ENOBUFS;
1071         } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) !=
1072                  old_pg);
1073 
1074         if (!mmp->user) {
1075                 mmp->user = get_uid(user);
1076                 mmp->num_pg = num_pg;
1077         } else {
1078                 mmp->num_pg += num_pg;
1079         }
1080 
1081         return 0;
1082 }
1083 EXPORT_SYMBOL_GPL(mm_account_pinned_pages);
1084 
1085 void mm_unaccount_pinned_pages(struct mmpin *mmp)
1086 {
1087         if (mmp->user) {
1088                 atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
1089                 free_uid(mmp->user);
1090         }
1091 }
1092 EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages);
1093 
1094 struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size)
1095 {
1096         struct ubuf_info *uarg;
1097         struct sk_buff *skb;
1098 
1099         WARN_ON_ONCE(!in_task());
1100 
1101         skb = sock_omalloc(sk, 0, GFP_KERNEL);
1102         if (!skb)
1103                 return NULL;
1104 
1105         BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
1106         uarg = (void *)skb->cb;
1107         uarg->mmp.user = NULL;
1108 
1109         if (mm_account_pinned_pages(&uarg->mmp, size)) {
1110                 kfree_skb(skb);
1111                 return NULL;
1112         }
1113 
1114         uarg->callback = sock_zerocopy_callback;
1115         uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
1116         uarg->len = 1;
1117         uarg->bytelen = size;
1118         uarg->zerocopy = 1;
1119         refcount_set(&uarg->refcnt, 1);
1120         sock_hold(sk);
1121 
1122         return uarg;
1123 }
1124 EXPORT_SYMBOL_GPL(sock_zerocopy_alloc);
1125 
1126 static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg)
1127 {
1128         return container_of((void *)uarg, struct sk_buff, cb);
1129 }
1130 
1131 struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
1132                                         struct ubuf_info *uarg)
1133 {
1134         if (uarg) {
1135                 const u32 byte_limit = 1 << 19;         /* limit to a few TSO */
1136                 u32 bytelen, next;
1137 
1138                 /* realloc only when socket is locked (TCP, UDP cork),
1139                  * so uarg->len and sk_zckey access is serialized
1140                  */
1141                 if (!sock_owned_by_user(sk)) {
1142                         WARN_ON_ONCE(1);
1143                         return NULL;
1144                 }
1145 
1146                 bytelen = uarg->bytelen + size;
1147                 if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) {
1148                         /* TCP can create new skb to attach new uarg */
1149                         if (sk->sk_type == SOCK_STREAM)
1150                                 goto new_alloc;
1151                         return NULL;
1152                 }
1153 
1154                 next = (u32)atomic_read(&sk->sk_zckey);
1155                 if ((u32)(uarg->id + uarg->len) == next) {
1156                         if (mm_account_pinned_pages(&uarg->mmp, size))
1157                                 return NULL;
1158                         uarg->len++;
1159                         uarg->bytelen = bytelen;
1160                         atomic_set(&sk->sk_zckey, ++next);
1161 
1162                         /* no extra ref when appending to datagram (MSG_MORE) */
1163                         if (sk->sk_type == SOCK_STREAM)
1164                                 sock_zerocopy_get(uarg);
1165 
1166                         return uarg;
1167                 }
1168         }
1169 
1170 new_alloc:
1171         return sock_zerocopy_alloc(sk, size);
1172 }
1173 EXPORT_SYMBOL_GPL(sock_zerocopy_realloc);
1174 
1175 static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1176 {
1177         struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1178         u32 old_lo, old_hi;
1179         u64 sum_len;
1180 
1181         old_lo = serr->ee.ee_info;
1182         old_hi = serr->ee.ee_data;
1183         sum_len = old_hi - old_lo + 1ULL + len;
1184 
1185         if (sum_len >= (1ULL << 32))
1186                 return false;
1187 
1188         if (lo != old_hi + 1)
1189                 return false;
1190 
1191         serr->ee.ee_data += len;
1192         return true;
1193 }
1194 
1195 void sock_zerocopy_callback(struct ubuf_info *uarg, bool success)
1196 {
1197         struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1198         struct sock_exterr_skb *serr;
1199         struct sock *sk = skb->sk;
1200         struct sk_buff_head *q;
1201         unsigned long flags;
1202         u32 lo, hi;
1203         u16 len;
1204 
1205         mm_unaccount_pinned_pages(&uarg->mmp);
1206 
1207         /* if !len, there was only 1 call, and it was aborted
1208          * so do not queue a completion notification
1209          */
1210         if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1211                 goto release;
1212 
1213         len = uarg->len;
1214         lo = uarg->id;
1215         hi = uarg->id + len - 1;
1216 
1217         serr = SKB_EXT_ERR(skb);
1218         memset(serr, 0, sizeof(*serr));
1219         serr->ee.ee_errno = 0;
1220         serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1221         serr->ee.ee_data = hi;
1222         serr->ee.ee_info = lo;
1223         if (!success)
1224                 serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1225 
1226         q = &sk->sk_error_queue;
1227         spin_lock_irqsave(&q->lock, flags);
1228         tail = skb_peek_tail(q);
1229         if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1230             !skb_zerocopy_notify_extend(tail, lo, len)) {
1231                 __skb_queue_tail(q, skb);
1232                 skb = NULL;
1233         }
1234         spin_unlock_irqrestore(&q->lock, flags);
1235 
1236         sk->sk_error_report(sk);
1237 
1238 release:
1239         consume_skb(skb);
1240         sock_put(sk);
1241 }
1242 EXPORT_SYMBOL_GPL(sock_zerocopy_callback);
1243 
1244 void sock_zerocopy_put(struct ubuf_info *uarg)
1245 {
1246         if (uarg && refcount_dec_and_test(&uarg->refcnt)) {
1247                 if (uarg->callback)
1248                         uarg->callback(uarg, uarg->zerocopy);
1249                 else
1250                         consume_skb(skb_from_uarg(uarg));
1251         }
1252 }
1253 EXPORT_SYMBOL_GPL(sock_zerocopy_put);
1254 
1255 void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1256 {
1257         if (uarg) {
1258                 struct sock *sk = skb_from_uarg(uarg)->sk;
1259 
1260                 atomic_dec(&sk->sk_zckey);
1261                 uarg->len--;
1262 
1263                 if (have_uref)
1264                         sock_zerocopy_put(uarg);
1265         }
1266 }
1267 EXPORT_SYMBOL_GPL(sock_zerocopy_put_abort);
1268 
1269 int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len)
1270 {
1271         return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len);
1272 }
1273 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram);
1274 
1275 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1276                              struct msghdr *msg, int len,
1277                              struct ubuf_info *uarg)
1278 {
1279         struct ubuf_info *orig_uarg = skb_zcopy(skb);
1280         struct iov_iter orig_iter = msg->msg_iter;
1281         int err, orig_len = skb->len;
1282 
1283         /* An skb can only point to one uarg. This edge case happens when
1284          * TCP appends to an skb, but zerocopy_realloc triggered a new alloc.
1285          */
1286         if (orig_uarg && uarg != orig_uarg)
1287                 return -EEXIST;
1288 
1289         err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len);
1290         if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1291                 struct sock *save_sk = skb->sk;
1292 
1293                 /* Streams do not free skb on error. Reset to prev state. */
1294                 msg->msg_iter = orig_iter;
1295                 skb->sk = sk;
1296                 ___pskb_trim(skb, orig_len);
1297                 skb->sk = save_sk;
1298                 return err;
1299         }
1300 
1301         skb_zcopy_set(skb, uarg, NULL);
1302         return skb->len - orig_len;
1303 }
1304 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1305 
1306 static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1307                               gfp_t gfp_mask)
1308 {
1309         if (skb_zcopy(orig)) {
1310                 if (skb_zcopy(nskb)) {
1311                         /* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1312                         if (!gfp_mask) {
1313                                 WARN_ON_ONCE(1);
1314                                 return -ENOMEM;
1315                         }
1316                         if (skb_uarg(nskb) == skb_uarg(orig))
1317                                 return 0;
1318                         if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1319                                 return -EIO;
1320                 }
1321                 skb_zcopy_set(nskb, skb_uarg(orig), NULL);
1322         }
1323         return 0;
1324 }
1325 
1326 /**
1327  *      skb_copy_ubufs  -       copy userspace skb frags buffers to kernel
1328  *      @skb: the skb to modify
1329  *      @gfp_mask: allocation priority
1330  *
1331  *      This must be called on SKBTX_DEV_ZEROCOPY skb.
1332  *      It will copy all frags into kernel and drop the reference
1333  *      to userspace pages.
1334  *
1335  *      If this function is called from an interrupt gfp_mask() must be
1336  *      %GFP_ATOMIC.
1337  *
1338  *      Returns 0 on success or a negative error code on failure
1339  *      to allocate kernel memory to copy to.
1340  */
1341 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1342 {
1343         int num_frags = skb_shinfo(skb)->nr_frags;
1344         struct page *page, *head = NULL;
1345         int i, new_frags;
1346         u32 d_off;
1347 
1348         if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1349                 return -EINVAL;
1350 
1351         if (!num_frags)
1352                 goto release;
1353 
1354         new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1355         for (i = 0; i < new_frags; i++) {
1356                 page = alloc_page(gfp_mask);
1357                 if (!page) {
1358                         while (head) {
1359                                 struct page *next = (struct page *)page_private(head);
1360                                 put_page(head);
1361                                 head = next;
1362                         }
1363                         return -ENOMEM;
1364                 }
1365                 set_page_private(page, (unsigned long)head);
1366                 head = page;
1367         }
1368 
1369         page = head;
1370         d_off = 0;
1371         for (i = 0; i < num_frags; i++) {
1372                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1373                 u32 p_off, p_len, copied;
1374                 struct page *p;
1375                 u8 *vaddr;
1376 
1377                 skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f),
1378                                       p, p_off, p_len, copied) {
1379                         u32 copy, done = 0;
1380                         vaddr = kmap_atomic(p);
1381 
1382                         while (done < p_len) {
1383                                 if (d_off == PAGE_SIZE) {
1384                                         d_off = 0;
1385                                         page = (struct page *)page_private(page);
1386                                 }
1387                                 copy = min_t(u32, PAGE_SIZE - d_off, p_len - done);
1388                                 memcpy(page_address(page) + d_off,
1389                                        vaddr + p_off + done, copy);
1390                                 done += copy;
1391                                 d_off += copy;
1392                         }
1393                         kunmap_atomic(vaddr);
1394                 }
1395         }
1396 
1397         /* skb frags release userspace buffers */
1398         for (i = 0; i < num_frags; i++)
1399                 skb_frag_unref(skb, i);
1400 
1401         /* skb frags point to kernel buffers */
1402         for (i = 0; i < new_frags - 1; i++) {
1403                 __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE);
1404                 head = (struct page *)page_private(head);
1405         }
1406         __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off);
1407         skb_shinfo(skb)->nr_frags = new_frags;
1408 
1409 release:
1410         skb_zcopy_clear(skb, false);
1411         return 0;
1412 }
1413 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
1414 
1415 /**
1416  *      skb_clone       -       duplicate an sk_buff
1417  *      @skb: buffer to clone
1418  *      @gfp_mask: allocation priority
1419  *
1420  *      Duplicate an &sk_buff. The new one is not owned by a socket. Both
1421  *      copies share the same packet data but not structure. The new
1422  *      buffer has a reference count of 1. If the allocation fails the
1423  *      function returns %NULL otherwise the new buffer is returned.
1424  *
1425  *      If this function is called from an interrupt gfp_mask() must be
1426  *      %GFP_ATOMIC.
1427  */
1428 
1429 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1430 {
1431         struct sk_buff_fclones *fclones = container_of(skb,
1432                                                        struct sk_buff_fclones,
1433                                                        skb1);
1434         struct sk_buff *n;
1435 
1436         if (skb_orphan_frags(skb, gfp_mask))
1437                 return NULL;
1438 
1439         if (skb->fclone == SKB_FCLONE_ORIG &&
1440             refcount_read(&fclones->fclone_ref) == 1) {
1441                 n = &fclones->skb2;
1442                 refcount_set(&fclones->fclone_ref, 2);
1443         } else {
1444                 if (skb_pfmemalloc(skb))
1445                         gfp_mask |= __GFP_MEMALLOC;
1446 
1447                 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1448                 if (!n)
1449                         return NULL;
1450 
1451                 n->fclone = SKB_FCLONE_UNAVAILABLE;
1452         }
1453 
1454         return __skb_clone(n, skb);
1455 }
1456 EXPORT_SYMBOL(skb_clone);
1457 
1458 void skb_headers_offset_update(struct sk_buff *skb, int off)
1459 {
1460         /* Only adjust this if it actually is csum_start rather than csum */
1461         if (skb->ip_summed == CHECKSUM_PARTIAL)
1462                 skb->csum_start += off;
1463         /* {transport,network,mac}_header and tail are relative to skb->head */
1464         skb->transport_header += off;
1465         skb->network_header   += off;
1466         if (skb_mac_header_was_set(skb))
1467                 skb->mac_header += off;
1468         skb->inner_transport_header += off;
1469         skb->inner_network_header += off;
1470         skb->inner_mac_header += off;
1471 }
1472 EXPORT_SYMBOL(skb_headers_offset_update);
1473 
1474 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old)
1475 {
1476         __copy_skb_header(new, old);
1477 
1478         skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1479         skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1480         skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1481 }
1482 EXPORT_SYMBOL(skb_copy_header);
1483 
1484 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1485 {
1486         if (skb_pfmemalloc(skb))
1487                 return SKB_ALLOC_RX;
1488         return 0;
1489 }
1490 
1491 /**
1492  *      skb_copy        -       create private copy of an sk_buff
1493  *      @skb: buffer to copy
1494  *      @gfp_mask: allocation priority
1495  *
1496  *      Make a copy of both an &sk_buff and its data. This is used when the
1497  *      caller wishes to modify the data and needs a private copy of the
1498  *      data to alter. Returns %NULL on failure or the pointer to the buffer
1499  *      on success. The returned buffer has a reference count of 1.
1500  *
1501  *      As by-product this function converts non-linear &sk_buff to linear
1502  *      one, so that &sk_buff becomes completely private and caller is allowed
1503  *      to modify all the data of returned buffer. This means that this
1504  *      function is not recommended for use in circumstances when only
1505  *      header is going to be modified. Use pskb_copy() instead.
1506  */
1507 
1508 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1509 {
1510         int headerlen = skb_headroom(skb);
1511         unsigned int size = skb_end_offset(skb) + skb->data_len;
1512         struct sk_buff *n = __alloc_skb(size, gfp_mask,
1513                                         skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1514 
1515         if (!n)
1516                 return NULL;
1517 
1518         /* Set the data pointer */
1519         skb_reserve(n, headerlen);
1520         /* Set the tail pointer and length */
1521         skb_put(n, skb->len);
1522 
1523         BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
1524 
1525         skb_copy_header(n, skb);
1526         return n;
1527 }
1528 EXPORT_SYMBOL(skb_copy);
1529 
1530 /**
1531  *      __pskb_copy_fclone      -  create copy of an sk_buff with private head.
1532  *      @skb: buffer to copy
1533  *      @headroom: headroom of new skb
1534  *      @gfp_mask: allocation priority
1535  *      @fclone: if true allocate the copy of the skb from the fclone
1536  *      cache instead of the head cache; it is recommended to set this
1537  *      to true for the cases where the copy will likely be cloned
1538  *
1539  *      Make a copy of both an &sk_buff and part of its data, located
1540  *      in header. Fragmented data remain shared. This is used when
1541  *      the caller wishes to modify only header of &sk_buff and needs
1542  *      private copy of the header to alter. Returns %NULL on failure
1543  *      or the pointer to the buffer on success.
1544  *      The returned buffer has a reference count of 1.
1545  */
1546 
1547 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1548                                    gfp_t gfp_mask, bool fclone)
1549 {
1550         unsigned int size = skb_headlen(skb) + headroom;
1551         int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1552         struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1553 
1554         if (!n)
1555                 goto out;
1556 
1557         /* Set the data pointer */
1558         skb_reserve(n, headroom);
1559         /* Set the tail pointer and length */
1560         skb_put(n, skb_headlen(skb));
1561         /* Copy the bytes */
1562         skb_copy_from_linear_data(skb, n->data, n->len);
1563 
1564         n->truesize += skb->data_len;
1565         n->data_len  = skb->data_len;
1566         n->len       = skb->len;
1567 
1568         if (skb_shinfo(skb)->nr_frags) {
1569                 int i;
1570 
1571                 if (skb_orphan_frags(skb, gfp_mask) ||
1572                     skb_zerocopy_clone(n, skb, gfp_mask)) {
1573                         kfree_skb(n);
1574                         n = NULL;
1575                         goto out;
1576                 }
1577                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1578                         skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1579                         skb_frag_ref(skb, i);
1580                 }
1581                 skb_shinfo(n)->nr_frags = i;
1582         }
1583 
1584         if (skb_has_frag_list(skb)) {
1585                 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1586                 skb_clone_fraglist(n);
1587         }
1588 
1589         skb_copy_header(n, skb);
1590 out:
1591         return n;
1592 }
1593 EXPORT_SYMBOL(__pskb_copy_fclone);
1594 
1595 /**
1596  *      pskb_expand_head - reallocate header of &sk_buff
1597  *      @skb: buffer to reallocate
1598  *      @nhead: room to add at head
1599  *      @ntail: room to add at tail
1600  *      @gfp_mask: allocation priority
1601  *
1602  *      Expands (or creates identical copy, if @nhead and @ntail are zero)
1603  *      header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1604  *      reference count of 1. Returns zero in the case of success or error,
1605  *      if expansion failed. In the last case, &sk_buff is not changed.
1606  *
1607  *      All the pointers pointing into skb header may change and must be
1608  *      reloaded after call to this function.
1609  */
1610 
1611 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1612                      gfp_t gfp_mask)
1613 {
1614         int i, osize = skb_end_offset(skb);
1615         int size = osize + nhead + ntail;
1616         long off;
1617         u8 *data;
1618 
1619         BUG_ON(nhead < 0);
1620 
1621         BUG_ON(skb_shared(skb));
1622 
1623         size = SKB_DATA_ALIGN(size);
1624 
1625         if (skb_pfmemalloc(skb))
1626                 gfp_mask |= __GFP_MEMALLOC;
1627         data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1628                                gfp_mask, NUMA_NO_NODE, NULL);
1629         if (!data)
1630                 goto nodata;
1631         size = SKB_WITH_OVERHEAD(ksize(data));
1632 
1633         /* Copy only real data... and, alas, header. This should be
1634          * optimized for the cases when header is void.
1635          */
1636         memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1637 
1638         memcpy((struct skb_shared_info *)(data + size),
1639                skb_shinfo(skb),
1640                offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1641 
1642         /*
1643          * if shinfo is shared we must drop the old head gracefully, but if it
1644          * is not we can just drop the old head and let the existing refcount
1645          * be since all we did is relocate the values
1646          */
1647         if (skb_cloned(skb)) {
1648                 if (skb_orphan_frags(skb, gfp_mask))
1649                         goto nofrags;
1650                 if (skb_zcopy(skb))
1651                         refcount_inc(&skb_uarg(skb)->refcnt);
1652                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1653                         skb_frag_ref(skb, i);
1654 
1655                 if (skb_has_frag_list(skb))
1656                         skb_clone_fraglist(skb);
1657 
1658                 skb_release_data(skb);
1659         } else {
1660                 skb_free_head(skb);
1661         }
1662         off = (data + nhead) - skb->head;
1663 
1664         skb->head     = data;
1665         skb->head_frag = 0;
1666         skb->data    += off;
1667 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1668         skb->end      = size;
1669         off           = nhead;
1670 #else
1671         skb->end      = skb->head + size;
1672 #endif
1673         skb->tail             += off;
1674         skb_headers_offset_update(skb, nhead);
1675         skb->cloned   = 0;
1676         skb->hdr_len  = 0;
1677         skb->nohdr    = 0;
1678         atomic_set(&skb_shinfo(skb)->dataref, 1);
1679 
1680         skb_metadata_clear(skb);
1681 
1682         /* It is not generally safe to change skb->truesize.
1683          * For the moment, we really care of rx path, or
1684          * when skb is orphaned (not attached to a socket).
1685          */
1686         if (!skb->sk || skb->destructor == sock_edemux)
1687                 skb->truesize += size - osize;
1688 
1689         return 0;
1690 
1691 nofrags:
1692         kfree(data);
1693 nodata:
1694         return -ENOMEM;
1695 }
1696 EXPORT_SYMBOL(pskb_expand_head);
1697 
1698 /* Make private copy of skb with writable head and some headroom */
1699 
1700 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1701 {
1702         struct sk_buff *skb2;
1703         int delta = headroom - skb_headroom(skb);
1704 
1705         if (delta <= 0)
1706                 skb2 = pskb_copy(skb, GFP_ATOMIC);
1707         else {
1708                 skb2 = skb_clone(skb, GFP_ATOMIC);
1709                 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1710                                              GFP_ATOMIC)) {
1711                         kfree_skb(skb2);
1712                         skb2 = NULL;
1713                 }
1714         }
1715         return skb2;
1716 }
1717 EXPORT_SYMBOL(skb_realloc_headroom);
1718 
1719 /**
1720  *      skb_copy_expand -       copy and expand sk_buff
1721  *      @skb: buffer to copy
1722  *      @newheadroom: new free bytes at head
1723  *      @newtailroom: new free bytes at tail
1724  *      @gfp_mask: allocation priority
1725  *
1726  *      Make a copy of both an &sk_buff and its data and while doing so
1727  *      allocate additional space.
1728  *
1729  *      This is used when the caller wishes to modify the data and needs a
1730  *      private copy of the data to alter as well as more space for new fields.
1731  *      Returns %NULL on failure or the pointer to the buffer
1732  *      on success. The returned buffer has a reference count of 1.
1733  *
1734  *      You must pass %GFP_ATOMIC as the allocation priority if this function
1735  *      is called from an interrupt.
1736  */
1737 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1738                                 int newheadroom, int newtailroom,
1739                                 gfp_t gfp_mask)
1740 {
1741         /*
1742          *      Allocate the copy buffer
1743          */
1744         struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1745                                         gfp_mask, skb_alloc_rx_flag(skb),
1746                                         NUMA_NO_NODE);
1747         int oldheadroom = skb_headroom(skb);
1748         int head_copy_len, head_copy_off;
1749 
1750         if (!n)
1751                 return NULL;
1752 
1753         skb_reserve(n, newheadroom);
1754 
1755         /* Set the tail pointer and length */
1756         skb_put(n, skb->len);
1757 
1758         head_copy_len = oldheadroom;
1759         head_copy_off = 0;
1760         if (newheadroom <= head_copy_len)
1761                 head_copy_len = newheadroom;
1762         else
1763                 head_copy_off = newheadroom - head_copy_len;
1764 
1765         /* Copy the linear header and data. */
1766         BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1767                              skb->len + head_copy_len));
1768 
1769         skb_copy_header(n, skb);
1770 
1771         skb_headers_offset_update(n, newheadroom - oldheadroom);
1772 
1773         return n;
1774 }
1775 EXPORT_SYMBOL(skb_copy_expand);
1776 
1777 /**
1778  *      __skb_pad               -       zero pad the tail of an skb
1779  *      @skb: buffer to pad
1780  *      @pad: space to pad
1781  *      @free_on_error: free buffer on error
1782  *
1783  *      Ensure that a buffer is followed by a padding area that is zero
1784  *      filled. Used by network drivers which may DMA or transfer data
1785  *      beyond the buffer end onto the wire.
1786  *
1787  *      May return error in out of memory cases. The skb is freed on error
1788  *      if @free_on_error is true.
1789  */
1790 
1791 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error)
1792 {
1793         int err;
1794         int ntail;
1795 
1796         /* If the skbuff is non linear tailroom is always zero.. */
1797         if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1798                 memset(skb->data+skb->len, 0, pad);
1799                 return 0;
1800         }
1801 
1802         ntail = skb->data_len + pad - (skb->end - skb->tail);
1803         if (likely(skb_cloned(skb) || ntail > 0)) {
1804                 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1805                 if (unlikely(err))
1806                         goto free_skb;
1807         }
1808 
1809         /* FIXME: The use of this function with non-linear skb's really needs
1810          * to be audited.
1811          */
1812         err = skb_linearize(skb);
1813         if (unlikely(err))
1814                 goto free_skb;
1815 
1816         memset(skb->data + skb->len, 0, pad);
1817         return 0;
1818 
1819 free_skb:
1820         if (free_on_error)
1821                 kfree_skb(skb);
1822         return err;
1823 }
1824 EXPORT_SYMBOL(__skb_pad);
1825 
1826 /**
1827  *      pskb_put - add data to the tail of a potentially fragmented buffer
1828  *      @skb: start of the buffer to use
1829  *      @tail: tail fragment of the buffer to use
1830  *      @len: amount of data to add
1831  *
1832  *      This function extends the used data area of the potentially
1833  *      fragmented buffer. @tail must be the last fragment of @skb -- or
1834  *      @skb itself. If this would exceed the total buffer size the kernel
1835  *      will panic. A pointer to the first byte of the extra data is
1836  *      returned.
1837  */
1838 
1839 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1840 {
1841         if (tail != skb) {
1842                 skb->data_len += len;
1843                 skb->len += len;
1844         }
1845         return skb_put(tail, len);
1846 }
1847 EXPORT_SYMBOL_GPL(pskb_put);
1848 
1849 /**
1850  *      skb_put - add data to a buffer
1851  *      @skb: buffer to use
1852  *      @len: amount of data to add
1853  *
1854  *      This function extends the used data area of the buffer. If this would
1855  *      exceed the total buffer size the kernel will panic. A pointer to the
1856  *      first byte of the extra data is returned.
1857  */
1858 void *skb_put(struct sk_buff *skb, unsigned int len)
1859 {
1860         void *tmp = skb_tail_pointer(skb);
1861         SKB_LINEAR_ASSERT(skb);
1862         skb->tail += len;
1863         skb->len  += len;
1864         if (unlikely(skb->tail > skb->end))
1865                 skb_over_panic(skb, len, __builtin_return_address(0));
1866         return tmp;
1867 }
1868 EXPORT_SYMBOL(skb_put);
1869 
1870 /**
1871  *      skb_push - add data to the start of a buffer
1872  *      @skb: buffer to use
1873  *      @len: amount of data to add
1874  *
1875  *      This function extends the used data area of the buffer at the buffer
1876  *      start. If this would exceed the total buffer headroom the kernel will
1877  *      panic. A pointer to the first byte of the extra data is returned.
1878  */
1879 void *skb_push(struct sk_buff *skb, unsigned int len)
1880 {
1881         skb->data -= len;
1882         skb->len  += len;
1883         if (unlikely(skb->data < skb->head))
1884                 skb_under_panic(skb, len, __builtin_return_address(0));
1885         return skb->data;
1886 }
1887 EXPORT_SYMBOL(skb_push);
1888 
1889 /**
1890  *      skb_pull - remove data from the start of a buffer
1891  *      @skb: buffer to use
1892  *      @len: amount of data to remove
1893  *
1894  *      This function removes data from the start of a buffer, returning
1895  *      the memory to the headroom. A pointer to the next data in the buffer
1896  *      is returned. Once the data has been pulled future pushes will overwrite
1897  *      the old data.
1898  */
1899 void *skb_pull(struct sk_buff *skb, unsigned int len)
1900 {
1901         return skb_pull_inline(skb, len);
1902 }
1903 EXPORT_SYMBOL(skb_pull);
1904 
1905 /**
1906  *      skb_trim - remove end from a buffer
1907  *      @skb: buffer to alter
1908  *      @len: new length
1909  *
1910  *      Cut the length of a buffer down by removing data from the tail. If
1911  *      the buffer is already under the length specified it is not modified.
1912  *      The skb must be linear.
1913  */
1914 void skb_trim(struct sk_buff *skb, unsigned int len)
1915 {
1916         if (skb->len > len)
1917                 __skb_trim(skb, len);
1918 }
1919 EXPORT_SYMBOL(skb_trim);
1920 
1921 /* Trims skb to length len. It can change skb pointers.
1922  */
1923 
1924 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1925 {
1926         struct sk_buff **fragp;
1927         struct sk_buff *frag;
1928         int offset = skb_headlen(skb);
1929         int nfrags = skb_shinfo(skb)->nr_frags;
1930         int i;
1931         int err;
1932 
1933         if (skb_cloned(skb) &&
1934             unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1935                 return err;
1936 
1937         i = 0;
1938         if (offset >= len)
1939                 goto drop_pages;
1940 
1941         for (; i < nfrags; i++) {
1942                 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1943 
1944                 if (end < len) {
1945                         offset = end;
1946                         continue;
1947                 }
1948 
1949                 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1950 
1951 drop_pages:
1952                 skb_shinfo(skb)->nr_frags = i;
1953 
1954                 for (; i < nfrags; i++)
1955                         skb_frag_unref(skb, i);
1956 
1957                 if (skb_has_frag_list(skb))
1958                         skb_drop_fraglist(skb);
1959                 goto done;
1960         }
1961 
1962         for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1963              fragp = &frag->next) {
1964                 int end = offset + frag->len;
1965 
1966                 if (skb_shared(frag)) {
1967                         struct sk_buff *nfrag;
1968 
1969                         nfrag = skb_clone(frag, GFP_ATOMIC);
1970                         if (unlikely(!nfrag))
1971                                 return -ENOMEM;
1972 
1973                         nfrag->next = frag->next;
1974                         consume_skb(frag);
1975                         frag = nfrag;
1976                         *fragp = frag;
1977                 }
1978 
1979                 if (end < len) {
1980                         offset = end;
1981                         continue;
1982                 }
1983 
1984                 if (end > len &&
1985                     unlikely((err = pskb_trim(frag, len - offset))))
1986                         return err;
1987 
1988                 if (frag->next)
1989                         skb_drop_list(&frag->next);
1990                 break;
1991         }
1992 
1993 done:
1994         if (len > skb_headlen(skb)) {
1995                 skb->data_len -= skb->len - len;
1996                 skb->len       = len;
1997         } else {
1998                 skb->len       = len;
1999                 skb->data_len  = 0;
2000                 skb_set_tail_pointer(skb, len);
2001         }
2002 
2003         if (!skb->sk || skb->destructor == sock_edemux)
2004                 skb_condense(skb);
2005         return 0;
2006 }
2007 EXPORT_SYMBOL(___pskb_trim);
2008 
2009 /* Note : use pskb_trim_rcsum() instead of calling this directly
2010  */
2011 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len)
2012 {
2013         if (skb->ip_summed == CHECKSUM_COMPLETE) {
2014                 int delta = skb->len - len;
2015 
2016                 skb->csum = csum_block_sub(skb->csum,
2017                                            skb_checksum(skb, len, delta, 0),
2018                                            len);
2019         }
2020         return __pskb_trim(skb, len);
2021 }
2022 EXPORT_SYMBOL(pskb_trim_rcsum_slow);
2023 
2024 /**
2025  *      __pskb_pull_tail - advance tail of skb header
2026  *      @skb: buffer to reallocate
2027  *      @delta: number of bytes to advance tail
2028  *
2029  *      The function makes a sense only on a fragmented &sk_buff,
2030  *      it expands header moving its tail forward and copying necessary
2031  *      data from fragmented part.
2032  *
2033  *      &sk_buff MUST have reference count of 1.
2034  *
2035  *      Returns %NULL (and &sk_buff does not change) if pull failed
2036  *      or value of new tail of skb in the case of success.
2037  *
2038  *      All the pointers pointing into skb header may change and must be
2039  *      reloaded after call to this function.
2040  */
2041 
2042 /* Moves tail of skb head forward, copying data from fragmented part,
2043  * when it is necessary.
2044  * 1. It may fail due to malloc failure.
2045  * 2. It may change skb pointers.
2046  *
2047  * It is pretty complicated. Luckily, it is called only in exceptional cases.
2048  */
2049 void *__pskb_pull_tail(struct sk_buff *skb, int delta)
2050 {
2051         /* If skb has not enough free space at tail, get new one
2052          * plus 128 bytes for future expansions. If we have enough
2053          * room at tail, reallocate without expansion only if skb is cloned.
2054          */
2055         int i, k, eat = (skb->tail + delta) - skb->end;
2056 
2057         if (eat > 0 || skb_cloned(skb)) {
2058                 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
2059                                      GFP_ATOMIC))
2060                         return NULL;
2061         }
2062 
2063         BUG_ON(skb_copy_bits(skb, skb_headlen(skb),
2064                              skb_tail_pointer(skb), delta));
2065 
2066         /* Optimization: no fragments, no reasons to preestimate
2067          * size of pulled pages. Superb.
2068          */
2069         if (!skb_has_frag_list(skb))
2070                 goto pull_pages;
2071 
2072         /* Estimate size of pulled pages. */
2073         eat = delta;
2074         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2075                 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2076 
2077                 if (size >= eat)
2078                         goto pull_pages;
2079                 eat -= size;
2080         }
2081 
2082         /* If we need update frag list, we are in troubles.
2083          * Certainly, it is possible to add an offset to skb data,
2084          * but taking into account that pulling is expected to
2085          * be very rare operation, it is worth to fight against
2086          * further bloating skb head and crucify ourselves here instead.
2087          * Pure masohism, indeed. 8)8)
2088          */
2089         if (eat) {
2090                 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2091                 struct sk_buff *clone = NULL;
2092                 struct sk_buff *insp = NULL;
2093 
2094                 do {
2095                         if (list->len <= eat) {
2096                                 /* Eaten as whole. */
2097                                 eat -= list->len;
2098                                 list = list->next;
2099                                 insp = list;
2100                         } else {
2101                                 /* Eaten partially. */
2102 
2103                                 if (skb_shared(list)) {
2104                                         /* Sucks! We need to fork list. :-( */
2105                                         clone = skb_clone(list, GFP_ATOMIC);
2106                                         if (!clone)
2107                                                 return NULL;
2108                                         insp = list->next;
2109                                         list = clone;
2110                                 } else {
2111                                         /* This may be pulled without
2112                                          * problems. */
2113                                         insp = list;
2114                                 }
2115                                 if (!pskb_pull(list, eat)) {
2116                                         kfree_skb(clone);
2117                                         return NULL;
2118                                 }
2119                                 break;
2120                         }
2121                 } while (eat);
2122 
2123                 /* Free pulled out fragments. */
2124                 while ((list = skb_shinfo(skb)->frag_list) != insp) {
2125                         skb_shinfo(skb)->frag_list = list->next;
2126                         kfree_skb(list);
2127                 }
2128                 /* And insert new clone at head. */
2129                 if (clone) {
2130                         clone->next = list;
2131                         skb_shinfo(skb)->frag_list = clone;
2132                 }
2133         }
2134         /* Success! Now we may commit changes to skb data. */
2135 
2136 pull_pages:
2137         eat = delta;
2138         k = 0;
2139         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2140                 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2141 
2142                 if (size <= eat) {
2143                         skb_frag_unref(skb, i);
2144                         eat -= size;
2145                 } else {
2146                         skb_frag_t *frag = &skb_shinfo(skb)->frags[k];
2147 
2148                         *frag = skb_shinfo(skb)->frags[i];
2149                         if (eat) {
2150                                 skb_frag_off_add(frag, eat);
2151                                 skb_frag_size_sub(frag, eat);
2152                                 if (!i)
2153                                         goto end;
2154                                 eat = 0;
2155                         }
2156                         k++;
2157                 }
2158         }
2159         skb_shinfo(skb)->nr_frags = k;
2160 
2161 end:
2162         skb->tail     += delta;
2163         skb->data_len -= delta;
2164 
2165         if (!skb->data_len)
2166                 skb_zcopy_clear(skb, false);
2167 
2168         return skb_tail_pointer(skb);
2169 }
2170 EXPORT_SYMBOL(__pskb_pull_tail);
2171 
2172 /**
2173  *      skb_copy_bits - copy bits from skb to kernel buffer
2174  *      @skb: source skb
2175  *      @offset: offset in source
2176  *      @to: destination buffer
2177  *      @len: number of bytes to copy
2178  *
2179  *      Copy the specified number of bytes from the source skb to the
2180  *      destination buffer.
2181  *
2182  *      CAUTION ! :
2183  *              If its prototype is ever changed,
2184  *              check arch/{*}/net/{*}.S files,
2185  *              since it is called from BPF assembly code.
2186  */
2187 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2188 {
2189         int start = skb_headlen(skb);
2190         struct sk_buff *frag_iter;
2191         int i, copy;
2192 
2193         if (offset > (int)skb->len - len)
2194                 goto fault;
2195 
2196         /* Copy header. */
2197         if ((copy = start - offset) > 0) {
2198                 if (copy > len)
2199                         copy = len;
2200                 skb_copy_from_linear_data_offset(skb, offset, to, copy);
2201                 if ((len -= copy) == 0)
2202                         return 0;
2203                 offset += copy;
2204                 to     += copy;
2205         }
2206 
2207         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2208                 int end;
2209                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2210 
2211                 WARN_ON(start > offset + len);
2212 
2213                 end = start + skb_frag_size(f);
2214                 if ((copy = end - offset) > 0) {
2215                         u32 p_off, p_len, copied;
2216                         struct page *p;
2217                         u8 *vaddr;
2218 
2219                         if (copy > len)
2220                                 copy = len;
2221 
2222                         skb_frag_foreach_page(f,
2223                                               skb_frag_off(f) + offset - start,
2224                                               copy, p, p_off, p_len, copied) {
2225                                 vaddr = kmap_atomic(p);
2226                                 memcpy(to + copied, vaddr + p_off, p_len);
2227                                 kunmap_atomic(vaddr);
2228                         }
2229 
2230                         if ((len -= copy) == 0)
2231                                 return 0;
2232                         offset += copy;
2233                         to     += copy;
2234                 }
2235                 start = end;
2236         }
2237 
2238         skb_walk_frags(skb, frag_iter) {
2239                 int end;
2240 
2241                 WARN_ON(start > offset + len);
2242 
2243                 end = start + frag_iter->len;
2244                 if ((copy = end - offset) > 0) {
2245                         if (copy > len)
2246                                 copy = len;
2247                         if (skb_copy_bits(frag_iter, offset - start, to, copy))
2248                                 goto fault;
2249                         if ((len -= copy) == 0)
2250                                 return 0;
2251                         offset += copy;
2252                         to     += copy;
2253                 }
2254                 start = end;
2255         }
2256 
2257         if (!len)
2258                 return 0;
2259 
2260 fault:
2261         return -EFAULT;
2262 }
2263 EXPORT_SYMBOL(skb_copy_bits);
2264 
2265 /*
2266  * Callback from splice_to_pipe(), if we need to release some pages
2267  * at the end of the spd in case we error'ed out in filling the pipe.
2268  */
2269 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
2270 {
2271         put_page(spd->pages[i]);
2272 }
2273 
2274 static struct page *linear_to_page(struct page *page, unsigned int *len,
2275                                    unsigned int *offset,
2276                                    struct sock *sk)
2277 {
2278         struct page_frag *pfrag = sk_page_frag(sk);
2279 
2280         if (!sk_page_frag_refill(sk, pfrag))
2281                 return NULL;
2282 
2283         *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
2284 
2285         memcpy(page_address(pfrag->page) + pfrag->offset,
2286                page_address(page) + *offset, *len);
2287         *offset = pfrag->offset;
2288         pfrag->offset += *len;
2289 
2290         return pfrag->page;
2291 }
2292 
2293 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
2294                              struct page *page,
2295                              unsigned int offset)
2296 {
2297         return  spd->nr_pages &&
2298                 spd->pages[spd->nr_pages - 1] == page &&
2299                 (spd->partial[spd->nr_pages - 1].offset +
2300                  spd->partial[spd->nr_pages - 1].len == offset);
2301 }
2302 
2303 /*
2304  * Fill page/offset/length into spd, if it can hold more pages.
2305  */
2306 static bool spd_fill_page(struct splice_pipe_desc *spd,
2307                           struct pipe_inode_info *pipe, struct page *page,
2308                           unsigned int *len, unsigned int offset,
2309                           bool linear,
2310                           struct sock *sk)
2311 {
2312         if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
2313                 return true;
2314 
2315         if (linear) {
2316                 page = linear_to_page(page, len, &offset, sk);
2317                 if (!page)
2318                         return true;
2319         }
2320         if (spd_can_coalesce(spd, page, offset)) {
2321                 spd->partial[spd->nr_pages - 1].len += *len;
2322                 return false;
2323         }
2324         get_page(page);
2325         spd->pages[spd->nr_pages] = page;
2326         spd->partial[spd->nr_pages].len = *len;
2327         spd->partial[spd->nr_pages].offset = offset;
2328         spd->nr_pages++;
2329 
2330         return false;
2331 }
2332 
2333 static bool __splice_segment(struct page *page, unsigned int poff,
2334                              unsigned int plen, unsigned int *off,
2335                              unsigned int *len,
2336                              struct splice_pipe_desc *spd, bool linear,
2337                              struct sock *sk,
2338                              struct pipe_inode_info *pipe)
2339 {
2340         if (!*len)
2341                 return true;
2342 
2343         /* skip this segment if already processed */
2344         if (*off >= plen) {
2345                 *off -= plen;
2346                 return false;
2347         }
2348 
2349         /* ignore any bits we already processed */
2350         poff += *off;
2351         plen -= *off;
2352         *off = 0;
2353 
2354         do {
2355                 unsigned int flen = min(*len, plen);
2356 
2357                 if (spd_fill_page(spd, pipe, page, &flen, poff,
2358                                   linear, sk))
2359                         return true;
2360                 poff += flen;
2361                 plen -= flen;
2362                 *len -= flen;
2363         } while (*len && plen);
2364 
2365         return false;
2366 }
2367 
2368 /*
2369  * Map linear and fragment data from the skb to spd. It reports true if the
2370  * pipe is full or if we already spliced the requested length.
2371  */
2372 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
2373                               unsigned int *offset, unsigned int *len,
2374                               struct splice_pipe_desc *spd, struct sock *sk)
2375 {
2376         int seg;
2377         struct sk_buff *iter;
2378 
2379         /* map the linear part :
2380          * If skb->head_frag is set, this 'linear' part is backed by a
2381          * fragment, and if the head is not shared with any clones then
2382          * we can avoid a copy since we own the head portion of this page.
2383          */
2384         if (__splice_segment(virt_to_page(skb->data),
2385                              (unsigned long) skb->data & (PAGE_SIZE - 1),
2386                              skb_headlen(skb),
2387                              offset, len, spd,
2388                              skb_head_is_locked(skb),
2389                              sk, pipe))
2390                 return true;
2391 
2392         /*
2393          * then map the fragments
2394          */
2395         for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
2396                 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
2397 
2398                 if (__splice_segment(skb_frag_page(f),
2399                                      skb_frag_off(f), skb_frag_size(f),
2400                                      offset, len, spd, false, sk, pipe))
2401                         return true;
2402         }
2403 
2404         skb_walk_frags(skb, iter) {
2405                 if (*offset >= iter->len) {
2406                         *offset -= iter->len;
2407                         continue;
2408                 }
2409                 /* __skb_splice_bits() only fails if the output has no room
2410                  * left, so no point in going over the frag_list for the error
2411                  * case.
2412                  */
2413                 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
2414                         return true;
2415         }
2416 
2417         return false;
2418 }
2419 
2420 /*
2421  * Map data from the skb to a pipe. Should handle both the linear part,
2422  * the fragments, and the frag list.
2423  */
2424 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2425                     struct pipe_inode_info *pipe, unsigned int tlen,
2426                     unsigned int flags)
2427 {
2428         struct partial_page partial[MAX_SKB_FRAGS];
2429         struct page *pages[MAX_SKB_FRAGS];
2430         struct splice_pipe_desc spd = {
2431                 .pages = pages,
2432                 .partial = partial,
2433                 .nr_pages_max = MAX_SKB_FRAGS,
2434                 .ops = &nosteal_pipe_buf_ops,
2435                 .spd_release = sock_spd_release,
2436         };
2437         int ret = 0;
2438 
2439         __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2440 
2441         if (spd.nr_pages)
2442                 ret = splice_to_pipe(pipe, &spd);
2443 
2444         return ret;
2445 }
2446 EXPORT_SYMBOL_GPL(skb_splice_bits);
2447 
2448 /* Send skb data on a socket. Socket must be locked. */
2449 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
2450                          int len)
2451 {
2452         unsigned int orig_len = len;
2453         struct sk_buff *head = skb;
2454         unsigned short fragidx;
2455         int slen, ret;
2456 
2457 do_frag_list:
2458 
2459         /* Deal with head data */
2460         while (offset < skb_headlen(skb) && len) {
2461                 struct kvec kv;
2462                 struct msghdr msg;
2463 
2464                 slen = min_t(int, len, skb_headlen(skb) - offset);
2465                 kv.iov_base = skb->data + offset;
2466                 kv.iov_len = slen;
2467                 memset(&msg, 0, sizeof(msg));
2468                 msg.msg_flags = MSG_DONTWAIT;
2469 
2470                 ret = kernel_sendmsg_locked(sk, &msg, &kv, 1, slen);
2471                 if (ret <= 0)
2472                         goto error;
2473 
2474                 offset += ret;
2475                 len -= ret;
2476         }
2477 
2478         /* All the data was skb head? */
2479         if (!len)
2480                 goto out;
2481 
2482         /* Make offset relative to start of frags */
2483         offset -= skb_headlen(skb);
2484 
2485         /* Find where we are in frag list */
2486         for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2487                 skb_frag_t *frag  = &skb_shinfo(skb)->frags[fragidx];
2488 
2489                 if (offset < skb_frag_size(frag))
2490                         break;
2491 
2492                 offset -= skb_frag_size(frag);
2493         }
2494 
2495         for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2496                 skb_frag_t *frag  = &skb_shinfo(skb)->frags[fragidx];
2497 
2498                 slen = min_t(size_t, len, skb_frag_size(frag) - offset);
2499 
2500                 while (slen) {
2501                         ret = kernel_sendpage_locked(sk, skb_frag_page(frag),
2502                                                      skb_frag_off(frag) + offset,
2503                                                      slen, MSG_DONTWAIT);
2504                         if (ret <= 0)
2505                                 goto error;
2506 
2507                         len -= ret;
2508                         offset += ret;
2509                         slen -= ret;
2510                 }
2511 
2512                 offset = 0;
2513         }
2514 
2515         if (len) {
2516                 /* Process any frag lists */
2517 
2518                 if (skb == head) {
2519                         if (skb_has_frag_list(skb)) {
2520                                 skb = skb_shinfo(skb)->frag_list;
2521                                 goto do_frag_list;
2522                         }
2523                 } else if (skb->next) {
2524                         skb = skb->next;
2525                         goto do_frag_list;
2526                 }
2527         }
2528 
2529 out:
2530         return orig_len - len;
2531 
2532 error:
2533         return orig_len == len ? ret : orig_len - len;
2534 }
2535 EXPORT_SYMBOL_GPL(skb_send_sock_locked);
2536 
2537 /**
2538  *      skb_store_bits - store bits from kernel buffer to skb
2539  *      @skb: destination buffer
2540  *      @offset: offset in destination
2541  *      @from: source buffer
2542  *      @len: number of bytes to copy
2543  *
2544  *      Copy the specified number of bytes from the source buffer to the
2545  *      destination skb.  This function handles all the messy bits of
2546  *      traversing fragment lists and such.
2547  */
2548 
2549 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2550 {
2551         int start = skb_headlen(skb);
2552         struct sk_buff *frag_iter;
2553         int i, copy;
2554 
2555         if (offset > (int)skb->len - len)
2556                 goto fault;
2557 
2558         if ((copy = start - offset) > 0) {
2559                 if (copy > len)
2560                         copy = len;
2561                 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2562                 if ((len -= copy) == 0)
2563                         return 0;
2564                 offset += copy;
2565                 from += copy;
2566         }
2567 
2568         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2569                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2570                 int end;
2571 
2572                 WARN_ON(start > offset + len);
2573 
2574                 end = start + skb_frag_size(frag);
2575                 if ((copy = end - offset) > 0) {
2576                         u32 p_off, p_len, copied;
2577                         struct page *p;
2578                         u8 *vaddr;
2579 
2580                         if (copy > len)
2581                                 copy = len;
2582 
2583                         skb_frag_foreach_page(frag,
2584                                               skb_frag_off(frag) + offset - start,
2585                                               copy, p, p_off, p_len, copied) {
2586                                 vaddr = kmap_atomic(p);
2587                                 memcpy(vaddr + p_off, from + copied, p_len);
2588                                 kunmap_atomic(vaddr);
2589                         }
2590 
2591                         if ((len -= copy) == 0)
2592                                 return 0;
2593                         offset += copy;
2594                         from += copy;
2595                 }
2596                 start = end;
2597         }
2598 
2599         skb_walk_frags(skb, frag_iter) {
2600                 int end;
2601 
2602                 WARN_ON(start > offset + len);
2603 
2604                 end = start + frag_iter->len;
2605                 if ((copy = end - offset) > 0) {
2606                         if (copy > len)
2607                                 copy = len;
2608                         if (skb_store_bits(frag_iter, offset - start,
2609                                            from, copy))
2610                                 goto fault;
2611                         if ((len -= copy) == 0)
2612                                 return 0;
2613                         offset += copy;
2614                         from += copy;
2615                 }
2616                 start = end;
2617         }
2618         if (!len)
2619                 return 0;
2620 
2621 fault:
2622         return -EFAULT;
2623 }
2624 EXPORT_SYMBOL(skb_store_bits);
2625 
2626 /* Checksum skb data. */
2627 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2628                       __wsum csum, const struct skb_checksum_ops *ops)
2629 {
2630         int start = skb_headlen(skb);
2631         int i, copy = start - offset;
2632         struct sk_buff *frag_iter;
2633         int pos = 0;
2634 
2635         /* Checksum header. */
2636         if (copy > 0) {
2637                 if (copy > len)
2638                         copy = len;
2639                 csum = INDIRECT_CALL_1(ops->update, csum_partial_ext,
2640                                        skb->data + offset, copy, csum);
2641                 if ((len -= copy) == 0)
2642                         return csum;
2643                 offset += copy;
2644                 pos     = copy;
2645         }
2646 
2647         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2648                 int end;
2649                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2650 
2651                 WARN_ON(start > offset + len);
2652 
2653                 end = start + skb_frag_size(frag);
2654                 if ((copy = end - offset) > 0) {
2655                         u32 p_off, p_len, copied;
2656                         struct page *p;
2657                         __wsum csum2;
2658                         u8 *vaddr;
2659 
2660                         if (copy > len)
2661                                 copy = len;
2662 
2663                         skb_frag_foreach_page(frag,
2664                                               skb_frag_off(frag) + offset - start,
2665                                               copy, p, p_off, p_len, copied) {
2666                                 vaddr = kmap_atomic(p);
2667                                 csum2 = INDIRECT_CALL_1(ops->update,
2668                                                         csum_partial_ext,
2669                                                         vaddr + p_off, p_len, 0);
2670                                 kunmap_atomic(vaddr);
2671                                 csum = INDIRECT_CALL_1(ops->combine,
2672                                                        csum_block_add_ext, csum,
2673                                                        csum2, pos, p_len);
2674                                 pos += p_len;
2675                         }
2676 
2677                         if (!(len -= copy))
2678                                 return csum;
2679                         offset += copy;
2680                 }
2681                 start = end;
2682         }
2683 
2684         skb_walk_frags(skb, frag_iter) {
2685                 int end;
2686 
2687                 WARN_ON(start > offset + len);
2688 
2689                 end = start + frag_iter->len;
2690                 if ((copy = end - offset) > 0) {
2691                         __wsum csum2;
2692                         if (copy > len)
2693                                 copy = len;
2694                         csum2 = __skb_checksum(frag_iter, offset - start,
2695                                                copy, 0, ops);
2696                         csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext,
2697                                                csum, csum2, pos, copy);
2698                         if ((len -= copy) == 0)
2699                                 return csum;
2700                         offset += copy;
2701                         pos    += copy;
2702                 }
2703                 start = end;
2704         }
2705         BUG_ON(len);
2706 
2707         return csum;
2708 }
2709 EXPORT_SYMBOL(__skb_checksum);
2710 
2711 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2712                     int len, __wsum csum)
2713 {
2714         const struct skb_checksum_ops ops = {
2715                 .update  = csum_partial_ext,
2716                 .combine = csum_block_add_ext,
2717         };
2718 
2719         return __skb_checksum(skb, offset, len, csum, &ops);
2720 }
2721 EXPORT_SYMBOL(skb_checksum);
2722 
2723 /* Both of above in one bottle. */
2724 
2725 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2726                                     u8 *to, int len, __wsum csum)
2727 {
2728         int start = skb_headlen(skb);
2729         int i, copy = start - offset;
2730         struct sk_buff *frag_iter;
2731         int pos = 0;
2732 
2733         /* Copy header. */
2734         if (copy > 0) {
2735                 if (copy > len)
2736                         copy = len;
2737                 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2738                                                  copy, csum);
2739                 if ((len -= copy) == 0)
2740                         return csum;
2741                 offset += copy;
2742                 to     += copy;
2743                 pos     = copy;
2744         }
2745 
2746         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2747                 int end;
2748 
2749                 WARN_ON(start > offset + len);
2750 
2751                 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2752                 if ((copy = end - offset) > 0) {
2753                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2754                         u32 p_off, p_len, copied;
2755                         struct page *p;
2756                         __wsum csum2;
2757                         u8 *vaddr;
2758 
2759                         if (copy > len)
2760                                 copy = len;
2761 
2762                         skb_frag_foreach_page(frag,
2763                                               skb_frag_off(frag) + offset - start,
2764                                               copy, p, p_off, p_len, copied) {
2765                                 vaddr = kmap_atomic(p);
2766                                 csum2 = csum_partial_copy_nocheck(vaddr + p_off,
2767                                                                   to + copied,
2768                                                                   p_len, 0);
2769                                 kunmap_atomic(vaddr);
2770                                 csum = csum_block_add(csum, csum2, pos);
2771                                 pos += p_len;
2772                         }
2773 
2774                         if (!(len -= copy))
2775                                 return csum;
2776                         offset += copy;
2777                         to     += copy;
2778                 }
2779                 start = end;
2780         }
2781 
2782         skb_walk_frags(skb, frag_iter) {
2783                 __wsum csum2;
2784                 int end;
2785 
2786                 WARN_ON(start > offset + len);
2787 
2788                 end = start + frag_iter->len;
2789                 if ((copy = end - offset) > 0) {
2790                         if (copy > len)
2791                                 copy = len;
2792                         csum2 = skb_copy_and_csum_bits(frag_iter,
2793                                                        offset - start,
2794                                                        to, copy, 0);
2795                         csum = csum_block_add(csum, csum2, pos);
2796                         if ((len -= copy) == 0)
2797                                 return csum;
2798                         offset += copy;
2799                         to     += copy;
2800                         pos    += copy;
2801                 }
2802                 start = end;
2803         }
2804         BUG_ON(len);
2805         return csum;
2806 }
2807 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2808 
2809 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
2810 {
2811         __sum16 sum;
2812 
2813         sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
2814         /* See comments in __skb_checksum_complete(). */
2815         if (likely(!sum)) {
2816                 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2817                     !skb->csum_complete_sw)
2818                         netdev_rx_csum_fault(skb->dev, skb);
2819         }
2820         if (!skb_shared(skb))
2821                 skb->csum_valid = !sum;
2822         return sum;
2823 }
2824 EXPORT_SYMBOL(__skb_checksum_complete_head);
2825 
2826 /* This function assumes skb->csum already holds pseudo header's checksum,
2827  * which has been changed from the hardware checksum, for example, by
2828  * __skb_checksum_validate_complete(). And, the original skb->csum must
2829  * have been validated unsuccessfully for CHECKSUM_COMPLETE case.
2830  *
2831  * It returns non-zero if the recomputed checksum is still invalid, otherwise
2832  * zero. The new checksum is stored back into skb->csum unless the skb is
2833  * shared.
2834  */
2835 __sum16 __skb_checksum_complete(struct sk_buff *skb)
2836 {
2837         __wsum csum;
2838         __sum16 sum;
2839 
2840         csum = skb_checksum(skb, 0, skb->len, 0);
2841 
2842         sum = csum_fold(csum_add(skb->csum, csum));
2843         /* This check is inverted, because we already knew the hardware
2844          * checksum is invalid before calling this function. So, if the
2845          * re-computed checksum is valid instead, then we have a mismatch
2846          * between the original skb->csum and skb_checksum(). This means either
2847          * the original hardware checksum is incorrect or we screw up skb->csum
2848          * when moving skb->data around.
2849          */
2850         if (likely(!sum)) {
2851                 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2852                     !skb->csum_complete_sw)
2853                         netdev_rx_csum_fault(skb->dev, skb);
2854         }
2855 
2856         if (!skb_shared(skb)) {
2857                 /* Save full packet checksum */
2858                 skb->csum = csum;
2859                 skb->ip_summed = CHECKSUM_COMPLETE;
2860                 skb->csum_complete_sw = 1;
2861                 skb->csum_valid = !sum;
2862         }
2863 
2864         return sum;
2865 }
2866 EXPORT_SYMBOL(__skb_checksum_complete);
2867 
2868 static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
2869 {
2870         net_warn_ratelimited(
2871                 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2872                 __func__);
2873         return 0;
2874 }
2875 
2876 static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
2877                                        int offset, int len)
2878 {
2879         net_warn_ratelimited(
2880                 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2881                 __func__);
2882         return 0;
2883 }
2884 
2885 static const struct skb_checksum_ops default_crc32c_ops = {
2886         .update  = warn_crc32c_csum_update,
2887         .combine = warn_crc32c_csum_combine,
2888 };
2889 
2890 const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
2891         &default_crc32c_ops;
2892 EXPORT_SYMBOL(crc32c_csum_stub);
2893 
2894  /**
2895  *      skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2896  *      @from: source buffer
2897  *
2898  *      Calculates the amount of linear headroom needed in the 'to' skb passed
2899  *      into skb_zerocopy().
2900  */
2901 unsigned int
2902 skb_zerocopy_headlen(const struct sk_buff *from)
2903 {
2904         unsigned int hlen = 0;
2905 
2906         if (!from->head_frag ||
2907             skb_headlen(from) < L1_CACHE_BYTES ||
2908             skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2909                 hlen = skb_headlen(from);
2910 
2911         if (skb_has_frag_list(from))
2912                 hlen = from->len;
2913 
2914         return hlen;
2915 }
2916 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2917 
2918 /**
2919  *      skb_zerocopy - Zero copy skb to skb
2920  *      @to: destination buffer
2921  *      @from: source buffer
2922  *      @len: number of bytes to copy from source buffer
2923  *      @hlen: size of linear headroom in destination buffer
2924  *
2925  *      Copies up to `len` bytes from `from` to `to` by creating references
2926  *      to the frags in the source buffer.
2927  *
2928  *      The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2929  *      headroom in the `to` buffer.
2930  *
2931  *      Return value:
2932  *      0: everything is OK
2933  *      -ENOMEM: couldn't orphan frags of @from due to lack of memory
2934  *      -EFAULT: skb_copy_bits() found some problem with skb geometry
2935  */
2936 int
2937 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2938 {
2939         int i, j = 0;
2940         int plen = 0; /* length of skb->head fragment */
2941         int ret;
2942         struct page *page;
2943         unsigned int offset;
2944 
2945         BUG_ON(!from->head_frag && !hlen);
2946 
2947         /* dont bother with small payloads */
2948         if (len <= skb_tailroom(to))
2949                 return skb_copy_bits(from, 0, skb_put(to, len), len);
2950 
2951         if (hlen) {
2952                 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2953                 if (unlikely(ret))
2954                         return ret;
2955                 len -= hlen;
2956         } else {
2957                 plen = min_t(int, skb_headlen(from), len);
2958                 if (plen) {
2959                         page = virt_to_head_page(from->head);
2960                         offset = from->data - (unsigned char *)page_address(page);
2961                         __skb_fill_page_desc(to, 0, page, offset, plen);
2962                         get_page(page);
2963                         j = 1;
2964                         len -= plen;
2965                 }
2966         }
2967 
2968         to->truesize += len + plen;
2969         to->len += len + plen;
2970         to->data_len += len + plen;
2971 
2972         if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2973                 skb_tx_error(from);
2974                 return -ENOMEM;
2975         }
2976         skb_zerocopy_clone(to, from, GFP_ATOMIC);
2977 
2978         for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2979                 int size;
2980 
2981                 if (!len)
2982                         break;
2983                 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2984                 size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]),
2985                                         len);
2986                 skb_frag_size_set(&skb_shinfo(to)->frags[j], size);
2987                 len -= size;
2988                 skb_frag_ref(to, j);
2989                 j++;
2990         }
2991         skb_shinfo(to)->nr_frags = j;
2992 
2993         return 0;
2994 }
2995 EXPORT_SYMBOL_GPL(skb_zerocopy);
2996 
2997 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2998 {
2999         __wsum csum;
3000         long csstart;
3001 
3002         if (skb->ip_summed == CHECKSUM_PARTIAL)
3003                 csstart = skb_checksum_start_offset(skb);
3004         else
3005                 csstart = skb_headlen(skb);
3006 
3007         BUG_ON(csstart > skb_headlen(skb));
3008 
3009         skb_copy_from_linear_data(skb, to, csstart);
3010 
3011         csum = 0;
3012         if (csstart != skb->len)
3013                 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
3014                                               skb->len - csstart, 0);
3015 
3016         if (skb->ip_summed == CHECKSUM_PARTIAL) {
3017                 long csstuff = csstart + skb->csum_offset;
3018 
3019                 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
3020         }
3021 }
3022 EXPORT_SYMBOL(skb_copy_and_csum_dev);
3023 
3024 /**
3025  *      skb_dequeue - remove from the head of the queue
3026  *      @list: list to dequeue from
3027  *
3028  *      Remove the head of the list. The list lock is taken so the function
3029  *      may be used safely with other locking list functions. The head item is
3030  *      returned or %NULL if the list is empty.
3031  */
3032 
3033 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
3034 {
3035         unsigned long flags;
3036         struct sk_buff *result;
3037 
3038         spin_lock_irqsave(&list->lock, flags);
3039         result = __skb_dequeue(list);
3040         spin_unlock_irqrestore(&list->lock, flags);
3041         return result;
3042 }
3043 EXPORT_SYMBOL(skb_dequeue);
3044 
3045 /**
3046  *      skb_dequeue_tail - remove from the tail of the queue
3047  *      @list: list to dequeue from
3048  *
3049  *      Remove the tail of the list. The list lock is taken so the function
3050  *      may be used safely with other locking list functions. The tail item is
3051  *      returned or %NULL if the list is empty.
3052  */
3053 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
3054 {
3055         unsigned long flags;
3056         struct sk_buff *result;
3057 
3058         spin_lock_irqsave(&list->lock, flags);
3059         result = __skb_dequeue_tail(list);
3060         spin_unlock_irqrestore(&list->lock, flags);
3061         return result;
3062 }
3063 EXPORT_SYMBOL(skb_dequeue_tail);
3064 
3065 /**
3066  *      skb_queue_purge - empty a list
3067  *      @list: list to empty
3068  *
3069  *      Delete all buffers on an &sk_buff list. Each buffer is removed from
3070  *      the list and one reference dropped. This function takes the list
3071  *      lock and is atomic with respect to other list locking functions.
3072  */
3073 void skb_queue_purge(struct sk_buff_head *list)
3074 {
3075         struct sk_buff *skb;
3076         while ((skb = skb_dequeue(list)) != NULL)
3077                 kfree_skb(skb);
3078 }
3079 EXPORT_SYMBOL(skb_queue_purge);
3080 
3081 /**
3082  *      skb_rbtree_purge - empty a skb rbtree
3083  *      @root: root of the rbtree to empty
3084  *      Return value: the sum of truesizes of all purged skbs.
3085  *
3086  *      Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
3087  *      the list and one reference dropped. This function does not take
3088  *      any lock. Synchronization should be handled by the caller (e.g., TCP
3089  *      out-of-order queue is protected by the socket lock).
3090  */
3091 unsigned int skb_rbtree_purge(struct rb_root *root)
3092 {
3093         struct rb_node *p = rb_first(root);
3094         unsigned int sum = 0;
3095 
3096         while (p) {
3097                 struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
3098 
3099                 p = rb_next(p);
3100                 rb_erase(&skb->rbnode, root);
3101                 sum += skb->truesize;
3102                 kfree_skb(skb);
3103         }
3104         return sum;
3105 }
3106 
3107 /**
3108  *      skb_queue_head - queue a buffer at the list head
3109  *      @list: list to use
3110  *      @newsk: buffer to queue
3111  *
3112  *      Queue a buffer at the start of the list. This function takes the
3113  *      list lock and can be used safely with other locking &sk_buff functions
3114  *      safely.
3115  *
3116  *      A buffer cannot be placed on two lists at the same time.
3117  */
3118 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
3119 {
3120         unsigned long flags;
3121 
3122         spin_lock_irqsave(&list->lock, flags);
3123         __skb_queue_head(list, newsk);
3124         spin_unlock_irqrestore(&list->lock, flags);
3125 }
3126 EXPORT_SYMBOL(skb_queue_head);
3127 
3128 /**
3129  *      skb_queue_tail - queue a buffer at the list tail
3130  *      @list: list to use
3131  *      @newsk: buffer to queue
3132  *
3133  *      Queue a buffer at the tail of the list. This function takes the
3134  *      list lock and can be used safely with other locking &sk_buff functions
3135  *      safely.
3136  *
3137  *      A buffer cannot be placed on two lists at the same time.
3138  */
3139 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
3140 {
3141         unsigned long flags;
3142 
3143         spin_lock_irqsave(&list->lock, flags);
3144         __skb_queue_tail(list, newsk);
3145         spin_unlock_irqrestore(&list->lock, flags);
3146 }
3147 EXPORT_SYMBOL(skb_queue_tail);
3148 
3149 /**
3150  *      skb_unlink      -       remove a buffer from a list
3151  *      @skb: buffer to remove
3152  *      @list: list to use
3153  *
3154  *      Remove a packet from a list. The list locks are taken and this
3155  *      function is atomic with respect to other list locked calls
3156  *
3157  *      You must know what list the SKB is on.
3158  */
3159 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
3160 {
3161         unsigned long flags;
3162 
3163         spin_lock_irqsave(&list->lock, flags);
3164         __skb_unlink(skb, list);
3165         spin_unlock_irqrestore(&list->lock, flags);
3166 }
3167 EXPORT_SYMBOL(skb_unlink);
3168 
3169 /**
3170  *      skb_append      -       append a buffer
3171  *      @old: buffer to insert after
3172  *      @newsk: buffer to insert
3173  *      @list: list to use
3174  *
3175  *      Place a packet after a given packet in a list. The list locks are taken
3176  *      and this function is atomic with respect to other list locked calls.
3177  *      A buffer cannot be placed on two lists at the same time.
3178  */
3179 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
3180 {
3181         unsigned long flags;
3182 
3183         spin_lock_irqsave(&list->lock, flags);
3184         __skb_queue_after(list, old, newsk);
3185         spin_unlock_irqrestore(&list->lock, flags);
3186 }
3187 EXPORT_SYMBOL(skb_append);
3188 
3189 static inline void skb_split_inside_header(struct sk_buff *skb,
3190                                            struct sk_buff* skb1,
3191                                            const u32 len, const int pos)
3192 {
3193         int i;
3194 
3195         skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
3196                                          pos - len);
3197         /* And move data appendix as is. */
3198         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
3199                 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
3200 
3201         skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
3202         skb_shinfo(skb)->nr_frags  = 0;
3203         skb1->data_len             = skb->data_len;
3204         skb1->len                  += skb1->data_len;
3205         skb->data_len              = 0;
3206         skb->len                   = len;
3207         skb_set_tail_pointer(skb, len);
3208 }
3209 
3210 static inline void skb_split_no_header(struct sk_buff *skb,
3211                                        struct sk_buff* skb1,
3212                                        const u32 len, int pos)
3213 {
3214         int i, k = 0;
3215         const int nfrags = skb_shinfo(skb)->nr_frags;
3216 
3217         skb_shinfo(skb)->nr_frags = 0;
3218         skb1->len                 = skb1->data_len = skb->len - len;
3219         skb->len                  = len;
3220         skb->data_len             = len - pos;
3221 
3222         for (i = 0; i < nfrags; i++) {
3223                 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
3224 
3225                 if (pos + size > len) {
3226                         skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
3227 
3228                         if (pos < len) {
3229                                 /* Split frag.
3230                                  * We have two variants in this case:
3231                                  * 1. Move all the frag to the second
3232                                  *    part, if it is possible. F.e.
3233                                  *    this approach is mandatory for TUX,
3234                                  *    where splitting is expensive.
3235                                  * 2. Split is accurately. We make this.
3236                                  */
3237                                 skb_frag_ref(skb, i);
3238                                 skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos);
3239                                 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
3240                                 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
3241                                 skb_shinfo(skb)->nr_frags++;
3242                         }
3243                         k++;
3244                 } else
3245                         skb_shinfo(skb)->nr_frags++;
3246                 pos += size;
3247         }
3248         skb_shinfo(skb1)->nr_frags = k;
3249 }
3250 
3251 /**
3252  * skb_split - Split fragmented skb to two parts at length len.
3253  * @skb: the buffer to split
3254  * @skb1: the buffer to receive the second part
3255  * @len: new length for skb
3256  */
3257 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
3258 {
3259         int pos = skb_headlen(skb);
3260 
3261         skb_shinfo(skb1)->tx_flags |= skb_shinfo(skb)->tx_flags &
3262                                       SKBTX_SHARED_FRAG;
3263         skb_zerocopy_clone(skb1, skb, 0);
3264         if (len < pos)  /* Split line is inside header. */
3265                 skb_split_inside_header(skb, skb1, len, pos);
3266         else            /* Second chunk has no header, nothing to copy. */
3267                 skb_split_no_header(skb, skb1, len, pos);
3268 }
3269 EXPORT_SYMBOL(skb_split);
3270 
3271 /* Shifting from/to a cloned skb is a no-go.
3272  *
3273  * Caller cannot keep skb_shinfo related pointers past calling here!
3274  */
3275 static int skb_prepare_for_shift(struct sk_buff *skb)
3276 {
3277         return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3278 }
3279 
3280 /**
3281  * skb_shift - Shifts paged data partially from skb to another
3282  * @tgt: buffer into which tail data gets added
3283  * @skb: buffer from which the paged data comes from
3284  * @shiftlen: shift up to this many bytes
3285  *
3286  * Attempts to shift up to shiftlen worth of bytes, which may be less than
3287  * the length of the skb, from skb to tgt. Returns number bytes shifted.
3288  * It's up to caller to free skb if everything was shifted.
3289  *
3290  * If @tgt runs out of frags, the whole operation is aborted.
3291  *
3292  * Skb cannot include anything else but paged data while tgt is allowed
3293  * to have non-paged data as well.
3294  *
3295  * TODO: full sized shift could be optimized but that would need
3296  * specialized skb free'er to handle frags without up-to-date nr_frags.
3297  */
3298 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
3299 {
3300         int from, to, merge, todo;
3301         skb_frag_t *fragfrom, *fragto;
3302 
3303         BUG_ON(shiftlen > skb->len);
3304 
3305         if (skb_headlen(skb))
3306                 return 0;
3307         if (skb_zcopy(tgt) || skb_zcopy(skb))
3308                 return 0;
3309 
3310         todo = shiftlen;
3311         from = 0;
3312         to = skb_shinfo(tgt)->nr_frags;
3313         fragfrom = &skb_shinfo(skb)->frags[from];
3314 
3315         /* Actual merge is delayed until the point when we know we can
3316          * commit all, so that we don't have to undo partial changes
3317          */
3318         if (!to ||
3319             !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
3320                               skb_frag_off(fragfrom))) {
3321                 merge = -1;
3322         } else {
3323                 merge = to - 1;
3324 
3325                 todo -= skb_frag_size(fragfrom);
3326                 if (todo < 0) {
3327                         if (skb_prepare_for_shift(skb) ||
3328                             skb_prepare_for_shift(tgt))
3329                                 return 0;
3330 
3331                         /* All previous frag pointers might be stale! */
3332                         fragfrom = &skb_shinfo(skb)->frags[from];
3333                         fragto = &skb_shinfo(tgt)->frags[merge];
3334 
3335                         skb_frag_size_add(fragto, shiftlen);
3336                         skb_frag_size_sub(fragfrom, shiftlen);
3337                         skb_frag_off_add(fragfrom, shiftlen);
3338 
3339                         goto onlymerged;
3340                 }
3341 
3342                 from++;
3343         }
3344 
3345         /* Skip full, not-fitting skb to avoid expensive operations */
3346         if ((shiftlen == skb->len) &&
3347             (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
3348                 return 0;
3349 
3350         if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
3351                 return 0;
3352 
3353         while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
3354                 if (to == MAX_SKB_FRAGS)
3355                         return 0;
3356 
3357                 fragfrom = &skb_shinfo(skb)->frags[from];
3358                 fragto = &skb_shinfo(tgt)->frags[to];
3359 
3360                 if (todo >= skb_frag_size(fragfrom)) {
3361                         *fragto = *fragfrom;
3362                         todo -= skb_frag_size(fragfrom);
3363                         from++;
3364                         to++;
3365 
3366                 } else {
3367                         __skb_frag_ref(fragfrom);
3368                         skb_frag_page_copy(fragto, fragfrom);
3369                         skb_frag_off_copy(fragto, fragfrom);
3370                         skb_frag_size_set(fragto, todo);
3371 
3372                         skb_frag_off_add(fragfrom, todo);
3373                         skb_frag_size_sub(fragfrom, todo);
3374                         todo = 0;
3375 
3376                         to++;
3377                         break;
3378                 }
3379         }
3380 
3381         /* Ready to "commit" this state change to tgt */
3382         skb_shinfo(tgt)->nr_frags = to;
3383 
3384         if (merge >= 0) {
3385                 fragfrom = &skb_shinfo(skb)->frags[0];
3386                 fragto = &skb_shinfo(tgt)->frags[merge];
3387 
3388                 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
3389                 __skb_frag_unref(fragfrom);
3390         }
3391 
3392         /* Reposition in the original skb */
3393         to = 0;
3394         while (from < skb_shinfo(skb)->nr_frags)
3395                 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
3396         skb_shinfo(skb)->nr_frags = to;
3397 
3398         BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
3399 
3400 onlymerged:
3401         /* Most likely the tgt won't ever need its checksum anymore, skb on
3402          * the other hand might need it if it needs to be resent
3403          */
3404         tgt->ip_summed = CHECKSUM_PARTIAL;
3405         skb->ip_summed = CHECKSUM_PARTIAL;
3406 
3407         /* Yak, is it really working this way? Some helper please? */
3408         skb->len -= shiftlen;
3409         skb->data_len -= shiftlen;
3410         skb->truesize -= shiftlen;
3411         tgt->len += shiftlen;
3412         tgt->data_len += shiftlen;
3413         tgt->truesize += shiftlen;
3414 
3415         return shiftlen;
3416 }
3417 
3418 /**
3419  * skb_prepare_seq_read - Prepare a sequential read of skb data
3420  * @skb: the buffer to read
3421  * @from: lower offset of data to be read
3422  * @to: upper offset of data to be read
3423  * @st: state variable
3424  *
3425  * Initializes the specified state variable. Must be called before
3426  * invoking skb_seq_read() for the first time.
3427  */
3428 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
3429                           unsigned int to, struct skb_seq_state *st)
3430 {
3431         st->lower_offset = from;
3432         st->upper_offset = to;
3433         st->root_skb = st->cur_skb = skb;
3434         st->frag_idx = st->stepped_offset = 0;
3435         st->frag_data = NULL;
3436 }
3437 EXPORT_SYMBOL(skb_prepare_seq_read);
3438 
3439 /**
3440  * skb_seq_read - Sequentially read skb data
3441  * @consumed: number of bytes consumed by the caller so far
3442  * @data: destination pointer for data to be returned
3443  * @st: state variable
3444  *
3445  * Reads a block of skb data at @consumed relative to the
3446  * lower offset specified to skb_prepare_seq_read(). Assigns
3447  * the head of the data block to @data and returns the length
3448  * of the block or 0 if the end of the skb data or the upper
3449  * offset has been reached.
3450  *
3451  * The caller is not required to consume all of the data
3452  * returned, i.e. @consumed is typically set to the number
3453  * of bytes already consumed and the next call to
3454  * skb_seq_read() will return the remaining part of the block.
3455  *
3456  * Note 1: The size of each block of data returned can be arbitrary,
3457  *       this limitation is the cost for zerocopy sequential
3458  *       reads of potentially non linear data.
3459  *
3460  * Note 2: Fragment lists within fragments are not implemented
3461  *       at the moment, state->root_skb could be replaced with
3462  *       a stack for this purpose.
3463  */
3464 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
3465                           struct skb_seq_state *st)
3466 {
3467         unsigned int block_limit, abs_offset = consumed + st->lower_offset;
3468         skb_frag_t *frag;
3469 
3470         if (unlikely(abs_offset >= st->upper_offset)) {
3471                 if (st->frag_data) {
3472                         kunmap_atomic(st->frag_data);
3473                         st->frag_data = NULL;
3474                 }
3475                 return 0;
3476         }
3477 
3478 next_skb:
3479         block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
3480 
3481         if (abs_offset < block_limit && !st->frag_data) {
3482                 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
3483                 return block_limit - abs_offset;
3484         }
3485 
3486         if (st->frag_idx == 0 && !st->frag_data)
3487                 st->stepped_offset += skb_headlen(st->cur_skb);
3488 
3489         while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
3490                 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
3491                 block_limit = skb_frag_size(frag) + st->stepped_offset;
3492 
3493                 if (abs_offset < block_limit) {
3494                         if (!st->frag_data)
3495                                 st->frag_data = kmap_atomic(skb_frag_page(frag));
3496 
3497                         *data = (u8 *) st->frag_data + skb_frag_off(frag) +
3498                                 (abs_offset - st->stepped_offset);
3499 
3500                         return block_limit - abs_offset;
3501                 }
3502 
3503                 if (st->frag_data) {
3504                         kunmap_atomic(st->frag_data);
3505                         st->frag_data = NULL;
3506                 }
3507 
3508                 st->frag_idx++;
3509                 st->stepped_offset += skb_frag_size(frag);
3510         }
3511 
3512         if (st->frag_data) {
3513                 kunmap_atomic(st->frag_data);
3514                 st->frag_data = NULL;
3515         }
3516 
3517         if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
3518                 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
3519                 st->frag_idx = 0;
3520                 goto next_skb;
3521         } else if (st->cur_skb->next) {
3522                 st->cur_skb = st->cur_skb->next;
3523                 st->frag_idx = 0;
3524                 goto next_skb;
3525         }
3526 
3527         return 0;
3528 }
3529 EXPORT_SYMBOL(skb_seq_read);
3530 
3531 /**
3532  * skb_abort_seq_read - Abort a sequential read of skb data
3533  * @st: state variable
3534  *
3535  * Must be called if skb_seq_read() was not called until it
3536  * returned 0.
3537  */
3538 void skb_abort_seq_read(struct skb_seq_state *st)
3539 {
3540         if (st->frag_data)
3541                 kunmap_atomic(st->frag_data);
3542 }
3543 EXPORT_SYMBOL(skb_abort_seq_read);
3544 
3545 #define TS_SKB_CB(state)        ((struct skb_seq_state *) &((state)->cb))
3546 
3547 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
3548                                           struct ts_config *conf,
3549                                           struct ts_state *state)
3550 {
3551         return skb_seq_read(offset, text, TS_SKB_CB(state));
3552 }
3553 
3554 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
3555 {
3556         skb_abort_seq_read(TS_SKB_CB(state));
3557 }
3558 
3559 /**
3560  * skb_find_text - Find a text pattern in skb data
3561  * @skb: the buffer to look in
3562  * @from: search offset
3563  * @to: search limit
3564  * @config: textsearch configuration
3565  *
3566  * Finds a pattern in the skb data according to the specified
3567  * textsearch configuration. Use textsearch_next() to retrieve
3568  * subsequent occurrences of the pattern. Returns the offset
3569  * to the first occurrence or UINT_MAX if no match was found.
3570  */
3571 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
3572                            unsigned int to, struct ts_config *config)
3573 {
3574         struct ts_state state;
3575         unsigned int ret;
3576 
3577         config->get_next_block = skb_ts_get_next_block;
3578         config->finish = skb_ts_finish;
3579 
3580         skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
3581 
3582         ret = textsearch_find(config, &state);
3583         return (ret <= to - from ? ret : UINT_MAX);
3584 }
3585 EXPORT_SYMBOL(skb_find_text);
3586 
3587 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3588                          int offset, size_t size)
3589 {
3590         int i = skb_shinfo(skb)->nr_frags;
3591 
3592         if (skb_can_coalesce(skb, i, page, offset)) {
3593                 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3594         } else if (i < MAX_SKB_FRAGS) {
3595                 get_page(page);
3596                 skb_fill_page_desc(skb, i, page, offset, size);
3597         } else {
3598                 return -EMSGSIZE;
3599         }
3600 
3601         return 0;
3602 }
3603 EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3604 
3605 /**
3606  *      skb_pull_rcsum - pull skb and update receive checksum
3607  *      @skb: buffer to update
3608  *      @len: length of data pulled
3609  *
3610  *      This function performs an skb_pull on the packet and updates
3611  *      the CHECKSUM_COMPLETE checksum.  It should be used on
3612  *      receive path processing instead of skb_pull unless you know
3613  *      that the checksum difference is zero (e.g., a valid IP header)
3614  *      or you are setting ip_summed to CHECKSUM_NONE.
3615  */
3616 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3617 {
3618         unsigned char *data = skb->data;
3619 
3620         BUG_ON(len > skb->len);
3621         __skb_pull(skb, len);
3622         skb_postpull_rcsum(skb, data, len);
3623         return skb->data;
3624 }
3625 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3626 
3627 static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb)
3628 {
3629         skb_frag_t head_frag;
3630         struct page *page;
3631 
3632         page = virt_to_head_page(frag_skb->head);
3633         __skb_frag_set_page(&head_frag, page);
3634         skb_frag_off_set(&head_frag, frag_skb->data -
3635                          (unsigned char *)page_address(page));
3636         skb_frag_size_set(&head_frag, skb_headlen(frag_skb));
3637         return head_frag;
3638 }
3639 
3640 struct sk_buff *skb_segment_list(struct sk_buff *skb,
3641                                  netdev_features_t features,
3642                                  unsigned int offset)
3643 {
3644         struct sk_buff *list_skb = skb_shinfo(skb)->frag_list;
3645         unsigned int tnl_hlen = skb_tnl_header_len(skb);
3646         unsigned int delta_truesize = 0;
3647         unsigned int delta_len = 0;
3648         struct sk_buff *tail = NULL;
3649         struct sk_buff *nskb;
3650 
3651         skb_push(skb, -skb_network_offset(skb) + offset);
3652 
3653         skb_shinfo(skb)->frag_list = NULL;
3654 
3655         do {
3656                 nskb = list_skb;
3657                 list_skb = list_skb->next;
3658 
3659                 if (!tail)
3660                         skb->next = nskb;
3661                 else
3662                         tail->next = nskb;
3663 
3664                 tail = nskb;
3665 
3666                 delta_len += nskb->len;
3667                 delta_truesize += nskb->truesize;
3668 
3669                 skb_push(nskb, -skb_network_offset(nskb) + offset);
3670 
3671                 skb_release_head_state(nskb);
3672                  __copy_skb_header(nskb, skb);
3673 
3674                 skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb));
3675                 skb_copy_from_linear_data_offset(skb, -tnl_hlen,
3676                                                  nskb->data - tnl_hlen,
3677                                                  offset + tnl_hlen);
3678 
3679                 if (skb_needs_linearize(nskb, features) &&
3680                     __skb_linearize(nskb))
3681                         goto err_linearize;
3682 
3683         } while (list_skb);
3684 
3685         skb->truesize = skb->truesize - delta_truesize;
3686         skb->data_len = skb->data_len - delta_len;
3687         skb->len = skb->len - delta_len;
3688 
3689         skb_gso_reset(skb);
3690 
3691         skb->prev = tail;
3692 
3693         if (skb_needs_linearize(skb, features) &&
3694             __skb_linearize(skb))
3695                 goto err_linearize;
3696 
3697         skb_get(skb);
3698 
3699         return skb;
3700 
3701 err_linearize:
3702         kfree_skb_list(skb->next);
3703         skb->next = NULL;
3704         return ERR_PTR(-ENOMEM);
3705 }
3706 EXPORT_SYMBOL_GPL(skb_segment_list);
3707 
3708 int skb_gro_receive_list(struct sk_buff *p, struct sk_buff *skb)
3709 {
3710         if (unlikely(p->len + skb->len >= 65536))
3711                 return -E2BIG;
3712 
3713         if (NAPI_GRO_CB(p)->last == p)
3714                 skb_shinfo(p)->frag_list = skb;
3715         else
3716                 NAPI_GRO_CB(p)->last->next = skb;
3717 
3718         skb_pull(skb, skb_gro_offset(skb));
3719 
3720         NAPI_GRO_CB(p)->last = skb;
3721         NAPI_GRO_CB(p)->count++;
3722         p->data_len += skb->len;
3723         p->truesize += skb->truesize;
3724         p->len += skb->len;
3725 
3726         NAPI_GRO_CB(skb)->same_flow = 1;
3727 
3728         return 0;
3729 }
3730 
3731 /**
3732  *      skb_segment - Perform protocol segmentation on skb.
3733  *      @head_skb: buffer to segment
3734  *      @features: features for the output path (see dev->features)
3735  *
3736  *      This function performs segmentation on the given skb.  It returns
3737  *      a pointer to the first in a list of new skbs for the segments.
3738  *      In case of error it returns ERR_PTR(err).
3739  */
3740 struct sk_buff *skb_segment(struct sk_buff *head_skb,
3741                             netdev_features_t features)
3742 {
3743         struct sk_buff *segs = NULL;
3744         struct sk_buff *tail = NULL;
3745         struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3746         skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3747         unsigned int mss = skb_shinfo(head_skb)->gso_size;
3748         unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3749         struct sk_buff *frag_skb = head_skb;
3750         unsigned int offset = doffset;
3751         unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3752         unsigned int partial_segs = 0;
3753         unsigned int headroom;
3754         unsigned int len = head_skb->len;
3755         __be16 proto;
3756         bool csum, sg;
3757         int nfrags = skb_shinfo(head_skb)->nr_frags;
3758         int err = -ENOMEM;
3759         int i = 0;
3760         int pos;
3761         int dummy;
3762 
3763         if (list_skb && !list_skb->head_frag && skb_headlen(list_skb) &&
3764             (skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY)) {
3765                 /* gso_size is untrusted, and we have a frag_list with a linear
3766                  * non head_frag head.
3767                  *
3768                  * (we assume checking the first list_skb member suffices;
3769                  * i.e if either of the list_skb members have non head_frag
3770                  * head, then the first one has too).
3771                  *
3772                  * If head_skb's headlen does not fit requested gso_size, it
3773                  * means that the frag_list members do NOT terminate on exact
3774                  * gso_size boundaries. Hence we cannot perform skb_frag_t page
3775                  * sharing. Therefore we must fallback to copying the frag_list
3776                  * skbs; we do so by disabling SG.
3777                  */
3778                 if (mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb))
3779                         features &= ~NETIF_F_SG;
3780         }
3781 
3782         __skb_push(head_skb, doffset);
3783         proto = skb_network_protocol(head_skb, &dummy);
3784         if (unlikely(!proto))
3785                 return ERR_PTR(-EINVAL);
3786 
3787         sg = !!(features & NETIF_F_SG);
3788         csum = !!can_checksum_protocol(features, proto);
3789 
3790         if (sg && csum && (mss != GSO_BY_FRAGS))  {
3791                 if (!(features & NETIF_F_GSO_PARTIAL)) {
3792                         struct sk_buff *iter;
3793                         unsigned int frag_len;
3794 
3795                         if (!list_skb ||
3796                             !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
3797                                 goto normal;
3798 
3799                         /* If we get here then all the required
3800                          * GSO features except frag_list are supported.
3801                          * Try to split the SKB to multiple GSO SKBs
3802                          * with no frag_list.
3803                          * Currently we can do that only when the buffers don't
3804                          * have a linear part and all the buffers except
3805                          * the last are of the same length.
3806                          */
3807                         frag_len = list_skb->len;
3808                         skb_walk_frags(head_skb, iter) {
3809                                 if (frag_len != iter->len && iter->next)
3810                                         goto normal;
3811                                 if (skb_headlen(iter) && !iter->head_frag)
3812                                         goto normal;
3813 
3814                                 len -= iter->len;
3815                         }
3816 
3817                         if (len != frag_len)
3818                                 goto normal;
3819                 }
3820 
3821                 /* GSO partial only requires that we trim off any excess that
3822                  * doesn't fit into an MSS sized block, so take care of that
3823                  * now.
3824                  */
3825                 partial_segs = len / mss;
3826                 if (partial_segs > 1)
3827                         mss *= partial_segs;
3828                 else
3829                         partial_segs = 0;
3830         }
3831 
3832 normal:
3833         headroom = skb_headroom(head_skb);
3834         pos = skb_headlen(head_skb);
3835 
3836         do {
3837                 struct sk_buff *nskb;
3838                 skb_frag_t *nskb_frag;
3839                 int hsize;
3840                 int size;
3841 
3842                 if (unlikely(mss == GSO_BY_FRAGS)) {
3843                         len = list_skb->len;
3844                 } else {
3845                         len = head_skb->len - offset;
3846                         if (len > mss)
3847                                 len = mss;
3848                 }
3849 
3850                 hsize = skb_headlen(head_skb) - offset;
3851                 if (hsize < 0)
3852                         hsize = 0;
3853                 if (hsize > len || !sg)
3854                         hsize = len;
3855 
3856                 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3857                     (skb_headlen(list_skb) == len || sg)) {
3858                         BUG_ON(skb_headlen(list_skb) > len);
3859 
3860                         i = 0;
3861                         nfrags = skb_shinfo(list_skb)->nr_frags;
3862                         frag = skb_shinfo(list_skb)->frags;
3863                         frag_skb = list_skb;
3864                         pos += skb_headlen(list_skb);
3865 
3866                         while (pos < offset + len) {
3867                                 BUG_ON(i >= nfrags);
3868 
3869                                 size = skb_frag_size(frag);
3870                                 if (pos + size > offset + len)
3871                                         break;
3872 
3873                                 i++;
3874                                 pos += size;
3875                                 frag++;
3876                         }
3877 
3878                         nskb = skb_clone(list_skb, GFP_ATOMIC);
3879                         list_skb = list_skb->next;
3880 
3881                         if (unlikely(!nskb))
3882                                 goto err;
3883 
3884                         if (unlikely(pskb_trim(nskb, len))) {
3885                                 kfree_skb(nskb);
3886                                 goto err;
3887                         }
3888 
3889                         hsize = skb_end_offset(nskb);
3890                         if (skb_cow_head(nskb, doffset + headroom)) {
3891                                 kfree_skb(nskb);
3892                                 goto err;
3893                         }
3894 
3895                         nskb->truesize += skb_end_offset(nskb) - hsize;
3896                         skb_release_head_state(nskb);
3897                         __skb_push(nskb, doffset);
3898                 } else {
3899                         nskb = __alloc_skb(hsize + doffset + headroom,
3900                                            GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3901                                            NUMA_NO_NODE);
3902 
3903                         if (unlikely(!nskb))
3904                                 goto err;
3905 
3906                         skb_reserve(nskb, headroom);
3907                         __skb_put(nskb, doffset);
3908                 }
3909 
3910                 if (segs)
3911                         tail->next = nskb;
3912                 else
3913                         segs = nskb;
3914                 tail = nskb;
3915 
3916                 __copy_skb_header(nskb, head_skb);
3917 
3918                 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3919                 skb_reset_mac_len(nskb);
3920 
3921                 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3922                                                  nskb->data - tnl_hlen,
3923                                                  doffset + tnl_hlen);
3924 
3925                 if (nskb->len == len + doffset)
3926                         goto perform_csum_check;
3927 
3928                 if (!sg) {
3929                         if (!csum) {
3930                                 if (!nskb->remcsum_offload)
3931                                         nskb->ip_summed = CHECKSUM_NONE;
3932                                 SKB_GSO_CB(nskb)->csum =
3933                                         skb_copy_and_csum_bits(head_skb, offset,
3934                                                                skb_put(nskb,
3935                                                                        len),
3936                                                                len, 0);
3937                                 SKB_GSO_CB(nskb)->csum_start =
3938                                         skb_headroom(nskb) + doffset;
3939                         } else {
3940                                 skb_copy_bits(head_skb, offset,
3941                                               skb_put(nskb, len),
3942                                               len);
3943                         }
3944                         continue;
3945                 }
3946 
3947                 nskb_frag = skb_shinfo(nskb)->frags;
3948 
3949                 skb_copy_from_linear_data_offset(head_skb, offset,
3950                                                  skb_put(nskb, hsize), hsize);
3951 
3952                 skb_shinfo(nskb)->tx_flags |= skb_shinfo(head_skb)->tx_flags &
3953                                               SKBTX_SHARED_FRAG;
3954 
3955                 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
3956                     skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
3957                         goto err;
3958 
3959                 while (pos < offset + len) {
3960                         if (i >= nfrags) {
3961                                 i = 0;
3962                                 nfrags = skb_shinfo(list_skb)->nr_frags;
3963                                 frag = skb_shinfo(list_skb)->frags;
3964                                 frag_skb = list_skb;
3965                                 if (!skb_headlen(list_skb)) {
3966                                         BUG_ON(!nfrags);
3967                                 } else {
3968                                         BUG_ON(!list_skb->head_frag);
3969 
3970                                         /* to make room for head_frag. */
3971                                         i--;
3972                                         frag--;
3973                                 }
3974                                 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
3975                                     skb_zerocopy_clone(nskb, frag_skb,
3976                                                        GFP_ATOMIC))
3977                                         goto err;
3978 
3979                                 list_skb = list_skb->next;
3980                         }
3981 
3982                         if (unlikely(skb_shinfo(nskb)->nr_frags >=
3983                                      MAX_SKB_FRAGS)) {
3984                                 net_warn_ratelimited(
3985                                         "skb_segment: too many frags: %u %u\n",
3986                                         pos, mss);
3987                                 err = -EINVAL;
3988                                 goto err;
3989                         }
3990 
3991                         *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag;
3992                         __skb_frag_ref(nskb_frag);
3993                         size = skb_frag_size(nskb_frag);
3994 
3995                         if (pos < offset) {
3996                                 skb_frag_off_add(nskb_frag, offset - pos);
3997                                 skb_frag_size_sub(nskb_frag, offset - pos);
3998                         }
3999 
4000                         skb_shinfo(nskb)->nr_frags++;
4001 
4002                         if (pos + size <= offset + len) {
4003                                 i++;
4004                                 frag++;
4005                                 pos += size;
4006                         } else {
4007                                 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
4008                                 goto skip_fraglist;
4009                         }
4010 
4011                         nskb_frag++;
4012                 }
4013 
4014 skip_fraglist:
4015                 nskb->data_len = len - hsize;
4016                 nskb->len += nskb->data_len;
4017                 nskb->truesize += nskb->data_len;
4018 
4019 perform_csum_check:
4020                 if (!csum) {
4021                         if (skb_has_shared_frag(nskb) &&
4022                             __skb_linearize(nskb))
4023                                 goto err;
4024 
4025                         if (!nskb->remcsum_offload)
4026                                 nskb->ip_summed = CHECKSUM_NONE;
4027                         SKB_GSO_CB(nskb)->csum =
4028                                 skb_checksum(nskb, doffset,
4029                                              nskb->len - doffset, 0);
4030                         SKB_GSO_CB(nskb)->csum_start =
4031                                 skb_headroom(nskb) + doffset;
4032                 }
4033         } while ((offset += len) < head_skb->len);
4034 
4035         /* Some callers want to get the end of the list.
4036          * Put it in segs->prev to avoid walking the list.
4037          * (see validate_xmit_skb_list() for example)
4038          */
4039         segs->prev = tail;
4040 
4041         if (partial_segs) {
4042                 struct sk_buff *iter;
4043                 int type = skb_shinfo(head_skb)->gso_type;
4044                 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
4045 
4046                 /* Update type to add partial and then remove dodgy if set */
4047                 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
4048                 type &= ~SKB_GSO_DODGY;
4049 
4050                 /* Update GSO info and prepare to start updating headers on
4051                  * our way back down the stack of protocols.
4052                  */
4053                 for (iter = segs; iter; iter = iter->next) {
4054                         skb_shinfo(iter)->gso_size = gso_size;
4055                         skb_shinfo(iter)->gso_segs = partial_segs;
4056                         skb_shinfo(iter)->gso_type = type;
4057                         SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
4058                 }
4059 
4060                 if (tail->len - doffset <= gso_size)
4061                         skb_shinfo(tail)->gso_size = 0;
4062                 else if (tail != segs)
4063                         skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
4064         }
4065 
4066         /* Following permits correct backpressure, for protocols
4067          * using skb_set_owner_w().
4068          * Idea is to tranfert ownership from head_skb to last segment.
4069          */
4070         if (head_skb->destructor == sock_wfree) {
4071                 swap(tail->truesize, head_skb->truesize);
4072                 swap(tail->destructor, head_skb->destructor);
4073                 swap(tail->sk, head_skb->sk);
4074         }
4075         return segs;
4076 
4077 err:
4078         kfree_skb_list(segs);
4079         return ERR_PTR(err);
4080 }
4081 EXPORT_SYMBOL_GPL(skb_segment);
4082 
4083 int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb)
4084 {
4085         struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
4086         unsigned int offset = skb_gro_offset(skb);
4087         unsigned int headlen = skb_headlen(skb);
4088         unsigned int len = skb_gro_len(skb);
4089         unsigned int delta_truesize;
4090         struct sk_buff *lp;
4091 
4092         if (unlikely(p->len + len >= 65536 || NAPI_GRO_CB(skb)->flush))
4093                 return -E2BIG;
4094 
4095         lp = NAPI_GRO_CB(p)->last;
4096         pinfo = skb_shinfo(lp);
4097 
4098         if (headlen <= offset) {
4099                 skb_frag_t *frag;
4100                 skb_frag_t *frag2;
4101                 int i = skbinfo->nr_frags;
4102                 int nr_frags = pinfo->nr_frags + i;
4103 
4104                 if (nr_frags > MAX_SKB_FRAGS)
4105                         goto merge;
4106 
4107                 offset -= headlen;
4108                 pinfo->nr_frags = nr_frags;
4109                 skbinfo->nr_frags = 0;
4110 
4111                 frag = pinfo->frags + nr_frags;
4112                 frag2 = skbinfo->frags + i;
4113                 do {
4114                         *--frag = *--frag2;
4115                 } while (--i);
4116 
4117                 skb_frag_off_add(frag, offset);
4118                 skb_frag_size_sub(frag, offset);
4119 
4120                 /* all fragments truesize : remove (head size + sk_buff) */
4121                 delta_truesize = skb->truesize -
4122                                  SKB_TRUESIZE(skb_end_offset(skb));
4123 
4124                 skb->truesize -= skb->data_len;
4125                 skb->len -= skb->data_len;
4126                 skb->data_len = 0;
4127 
4128                 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
4129                 goto done;
4130         } else if (skb->head_frag) {
4131                 int nr_frags = pinfo->nr_frags;
4132                 skb_frag_t *frag = pinfo->frags + nr_frags;
4133                 struct page *page = virt_to_head_page(skb->head);
4134                 unsigned int first_size = headlen - offset;
4135                 unsigned int first_offset;
4136 
4137                 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
4138                         goto merge;
4139 
4140                 first_offset = skb->data -
4141                                (unsigned char *)page_address(page) +
4142                                offset;
4143 
4144                 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
4145 
4146                 __skb_frag_set_page(frag, page);
4147                 skb_frag_off_set(frag, first_offset);
4148                 skb_frag_size_set(frag, first_size);
4149 
4150                 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
4151                 /* We dont need to clear skbinfo->nr_frags here */
4152 
4153                 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4154                 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
4155                 goto done;
4156         }
4157 
4158 merge:
4159         delta_truesize = skb->truesize;
4160         if (offset > headlen) {
4161                 unsigned int eat = offset - headlen;
4162 
4163                 skb_frag_off_add(&skbinfo->frags[0], eat);
4164                 skb_frag_size_sub(&skbinfo->frags[0], eat);
4165                 skb->data_len -= eat;
4166                 skb->len -= eat;
4167                 offset = headlen;
4168         }
4169 
4170         __skb_pull(skb, offset);
4171 
4172         if (NAPI_GRO_CB(p)->last == p)
4173                 skb_shinfo(p)->frag_list = skb;
4174         else
4175                 NAPI_GRO_CB(p)->last->next = skb;
4176         NAPI_GRO_CB(p)->last = skb;
4177         __skb_header_release(skb);
4178         lp = p;
4179 
4180 done:
4181         NAPI_GRO_CB(p)->count++;
4182         p->data_len += len;
4183         p->truesize += delta_truesize;
4184         p->len += len;
4185         if (lp != p) {
4186                 lp->data_len += len;
4187                 lp->truesize += delta_truesize;
4188                 lp->len += len;
4189         }
4190         NAPI_GRO_CB(skb)->same_flow = 1;
4191         return 0;
4192 }
4193 
4194 #ifdef CONFIG_SKB_EXTENSIONS
4195 #define SKB_EXT_ALIGN_VALUE     8
4196 #define SKB_EXT_CHUNKSIZEOF(x)  (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE)
4197 
4198 static const u8 skb_ext_type_len[] = {
4199 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4200         [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info),
4201 #endif
4202 #ifdef CONFIG_XFRM
4203         [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path),
4204 #endif
4205 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4206         [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext),
4207 #endif
4208 #if IS_ENABLED(CONFIG_MPTCP)
4209         [SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext),
4210 #endif
4211 };
4212 
4213 static __always_inline unsigned int skb_ext_total_length(void)
4214 {
4215         return SKB_EXT_CHUNKSIZEOF(struct skb_ext) +
4216 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4217                 skb_ext_type_len[SKB_EXT_BRIDGE_NF] +
4218 #endif
4219 #ifdef CONFIG_XFRM
4220                 skb_ext_type_len[SKB_EXT_SEC_PATH] +
4221 #endif
4222 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4223                 skb_ext_type_len[TC_SKB_EXT] +
4224 #endif
4225 #if IS_ENABLED(CONFIG_MPTCP)
4226                 skb_ext_type_len[SKB_EXT_MPTCP] +
4227 #endif
4228                 0;
4229 }
4230 
4231 static void skb_extensions_init(void)
4232 {
4233         BUILD_BUG_ON(SKB_EXT_NUM >= 8);
4234         BUILD_BUG_ON(skb_ext_total_length() > 255);
4235 
4236         skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache",
4237                                              SKB_EXT_ALIGN_VALUE * skb_ext_total_length(),
4238                                              0,
4239                                              SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4240                                              NULL);
4241 }
4242 #else
4243 static void skb_extensions_init(void) {}
4244 #endif
4245 
4246 void __init skb_init(void)
4247 {
4248         skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache",
4249                                               sizeof(struct sk_buff),
4250                                               0,
4251                                               SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4252                                               offsetof(struct sk_buff, cb),
4253                                               sizeof_field(struct sk_buff, cb),
4254                                               NULL);
4255         skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
4256                                                 sizeof(struct sk_buff_fclones),
4257                                                 0,
4258                                                 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4259                                                 NULL);
4260         skb_extensions_init();
4261 }
4262 
4263 static int
4264 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
4265                unsigned int recursion_level)
4266 {
4267         int start = skb_headlen(skb);
4268         int i, copy = start - offset;
4269         struct sk_buff *frag_iter;
4270         int elt = 0;
4271 
4272         if (unlikely(recursion_level >= 24))
4273                 return -EMSGSIZE;
4274 
4275         if (copy > 0) {
4276                 if (copy > len)
4277                         copy = len;
4278                 sg_set_buf(sg, skb->data + offset, copy);
4279                 elt++;
4280                 if ((len -= copy) == 0)
4281                         return elt;
4282                 offset += copy;
4283         }
4284 
4285         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4286                 int end;
4287 
4288                 WARN_ON(start > offset + len);
4289 
4290                 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
4291                 if ((copy = end - offset) > 0) {
4292                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4293                         if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4294                                 return -EMSGSIZE;
4295 
4296                         if (copy > len)
4297                                 copy = len;
4298                         sg_set_page(&sg[elt], skb_frag_page(frag), copy,
4299                                     skb_frag_off(frag) + offset - start);
4300                         elt++;
4301                         if (!(len -= copy))
4302                                 return elt;
4303                         offset += copy;
4304                 }
4305                 start = end;
4306         }
4307 
4308         skb_walk_frags(skb, frag_iter) {
4309                 int end, ret;
4310 
4311                 WARN_ON(start > offset + len);
4312 
4313                 end = start + frag_iter->len;
4314                 if ((copy = end - offset) > 0) {
4315                         if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4316                                 return -EMSGSIZE;
4317 
4318                         if (copy > len)
4319                                 copy = len;
4320                         ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
4321                                               copy, recursion_level + 1);
4322                         if (unlikely(ret < 0))
4323                                 return ret;
4324                         elt += ret;
4325                         if ((len -= copy) == 0)
4326                                 return elt;
4327                         offset += copy;
4328                 }
4329                 start = end;
4330         }
4331         BUG_ON(len);
4332         return elt;
4333 }
4334 
4335 /**
4336  *      skb_to_sgvec - Fill a scatter-gather list from a socket buffer
4337  *      @skb: Socket buffer containing the buffers to be mapped
4338  *      @sg: The scatter-gather list to map into
4339  *      @offset: The offset into the buffer's contents to start mapping
4340  *      @len: Length of buffer space to be mapped
4341  *
4342  *      Fill the specified scatter-gather list with mappings/pointers into a
4343  *      region of the buffer space attached to a socket buffer. Returns either
4344  *      the number of scatterlist items used, or -EMSGSIZE if the contents
4345  *      could not fit.
4346  */
4347 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
4348 {
4349         int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
4350 
4351         if (nsg <= 0)
4352                 return nsg;
4353 
4354         sg_mark_end(&sg[nsg - 1]);
4355 
4356         return nsg;
4357 }
4358 EXPORT_SYMBOL_GPL(skb_to_sgvec);
4359 
4360 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
4361  * sglist without mark the sg which contain last skb data as the end.
4362  * So the caller can mannipulate sg list as will when padding new data after
4363  * the first call without calling sg_unmark_end to expend sg list.
4364  *
4365  * Scenario to use skb_to_sgvec_nomark:
4366  * 1. sg_init_table
4367  * 2. skb_to_sgvec_nomark(payload1)
4368  * 3. skb_to_sgvec_nomark(payload2)
4369  *
4370  * This is equivalent to:
4371  * 1. sg_init_table
4372  * 2. skb_to_sgvec(payload1)
4373  * 3. sg_unmark_end
4374  * 4. skb_to_sgvec(payload2)
4375  *
4376  * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
4377  * is more preferable.
4378  */
4379 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
4380                         int offset, int len)
4381 {
4382         return __skb_to_sgvec(skb, sg, offset, len, 0);
4383 }
4384 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
4385 
4386 
4387 
4388 /**
4389  *      skb_cow_data - Check that a socket buffer's data buffers are writable
4390  *      @skb: The socket buffer to check.
4391  *      @tailbits: Amount of trailing space to be added
4392  *      @trailer: Returned pointer to the skb where the @tailbits space begins
4393  *
4394  *      Make sure that the data buffers attached to a socket buffer are
4395  *      writable. If they are not, private copies are made of the data buffers
4396  *      and the socket buffer is set to use these instead.
4397  *
4398  *      If @tailbits is given, make sure that there is space to write @tailbits
4399  *      bytes of data beyond current end of socket buffer.  @trailer will be
4400  *      set to point to the skb in which this space begins.
4401  *
4402  *      The number of scatterlist elements required to completely map the
4403  *      COW'd and extended socket buffer will be returned.
4404  */
4405 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
4406 {
4407         int copyflag;
4408         int elt;
4409         struct sk_buff *skb1, **skb_p;
4410 
4411         /* If skb is cloned or its head is paged, reallocate
4412          * head pulling out all the pages (pages are considered not writable
4413          * at the moment even if they are anonymous).
4414          */
4415         if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
4416             __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
4417                 return -ENOMEM;
4418 
4419         /* Easy case. Most of packets will go this way. */
4420         if (!skb_has_frag_list(skb)) {
4421                 /* A little of trouble, not enough of space for trailer.
4422                  * This should not happen, when stack is tuned to generate
4423                  * good frames. OK, on miss we reallocate and reserve even more
4424                  * space, 128 bytes is fair. */
4425 
4426                 if (skb_tailroom(skb) < tailbits &&
4427                     pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
4428                         return -ENOMEM;
4429 
4430                 /* Voila! */
4431                 *trailer = skb;
4432                 return 1;
4433         }
4434 
4435         /* Misery. We are in troubles, going to mincer fragments... */
4436 
4437         elt = 1;
4438         skb_p = &skb_shinfo(skb)->frag_list;
4439         copyflag = 0;
4440 
4441         while ((skb1 = *skb_p) != NULL) {
4442                 int ntail = 0;
4443 
4444                 /* The fragment is partially pulled by someone,
4445                  * this can happen on input. Copy it and everything
4446                  * after it. */
4447 
4448                 if (skb_shared(skb1))
4449                         copyflag = 1;
4450 
4451                 /* If the skb is the last, worry about trailer. */
4452 
4453                 if (skb1->next == NULL && tailbits) {
4454                         if (skb_shinfo(skb1)->nr_frags ||
4455                             skb_has_frag_list(skb1) ||
4456                             skb_tailroom(skb1) < tailbits)
4457                                 ntail = tailbits + 128;
4458                 }
4459 
4460                 if (copyflag ||
4461                     skb_cloned(skb1) ||
4462                     ntail ||
4463                     skb_shinfo(skb1)->nr_frags ||
4464                     skb_has_frag_list(skb1)) {
4465                         struct sk_buff *skb2;
4466 
4467                         /* Fuck, we are miserable poor guys... */
4468                         if (ntail == 0)
4469                                 skb2 = skb_copy(skb1, GFP_ATOMIC);
4470                         else
4471                                 skb2 = skb_copy_expand(skb1,
4472                                                        skb_headroom(skb1),
4473                                                        ntail,
4474                                                        GFP_ATOMIC);
4475                         if (unlikely(skb2 == NULL))
4476                                 return -ENOMEM;
4477 
4478                         if (skb1->sk)
4479                                 skb_set_owner_w(skb2, skb1->sk);
4480 
4481                         /* Looking around. Are we still alive?
4482                          * OK, link new skb, drop old one */
4483 
4484                         skb2->next = skb1->next;
4485                         *skb_p = skb2;
4486                         kfree_skb(skb1);
4487                         skb1 = skb2;
4488                 }
4489                 elt++;
4490                 *trailer = skb1;
4491                 skb_p = &skb1->next;
4492         }
4493 
4494         return elt;
4495 }
4496 EXPORT_SYMBOL_GPL(skb_cow_data);
4497 
4498 static void sock_rmem_free(struct sk_buff *skb)
4499 {
4500         struct sock *sk = skb->sk;
4501 
4502         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
4503 }
4504 
4505 static void skb_set_err_queue(struct sk_buff *skb)
4506 {
4507         /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
4508          * So, it is safe to (mis)use it to mark skbs on the error queue.
4509          */
4510         skb->pkt_type = PACKET_OUTGOING;
4511         BUILD_BUG_ON(PACKET_OUTGOING == 0);
4512 }
4513 
4514 /*
4515  * Note: We dont mem charge error packets (no sk_forward_alloc changes)
4516  */
4517 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
4518 {
4519         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
4520             (unsigned int)READ_ONCE(sk->sk_rcvbuf))
4521                 return -ENOMEM;
4522 
4523         skb_orphan(skb);
4524         skb->sk = sk;
4525         skb->destructor = sock_rmem_free;
4526         atomic_add(skb->truesize, &sk->sk_rmem_alloc);
4527         skb_set_err_queue(skb);
4528 
4529         /* before exiting rcu section, make sure dst is refcounted */
4530         skb_dst_force(skb);
4531 
4532         skb_queue_tail(&sk->sk_error_queue, skb);
4533         if (!sock_flag(sk, SOCK_DEAD))
4534                 sk->sk_error_report(sk);
4535         return 0;
4536 }
4537 EXPORT_SYMBOL(sock_queue_err_skb);
4538 
4539 static bool is_icmp_err_skb(const struct sk_buff *skb)
4540 {
4541         return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
4542                        SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
4543 }
4544 
4545 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
4546 {
4547         struct sk_buff_head *q = &sk->sk_error_queue;
4548         struct sk_buff *skb, *skb_next = NULL;
4549         bool icmp_next = false;
4550         unsigned long flags;
4551 
4552         spin_lock_irqsave(&q->lock, flags);
4553         skb = __skb_dequeue(q);
4554         if (skb && (skb_next = skb_peek(q))) {
4555                 icmp_next = is_icmp_err_skb(skb_next);
4556                 if (icmp_next)
4557                         sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_origin;
4558         }
4559         spin_unlock_irqrestore(&q->lock, flags);
4560 
4561         if (is_icmp_err_skb(skb) && !icmp_next)
4562                 sk->sk_err = 0;
4563 
4564         if (skb_next)
4565                 sk->sk_error_report(sk);
4566 
4567         return skb;
4568 }
4569 EXPORT_SYMBOL(sock_dequeue_err_skb);
4570 
4571 /**
4572  * skb_clone_sk - create clone of skb, and take reference to socket
4573  * @skb: the skb to clone
4574  *
4575  * This function creates a clone of a buffer that holds a reference on
4576  * sk_refcnt.  Buffers created via this function are meant to be
4577  * returned using sock_queue_err_skb, or free via kfree_skb.
4578  *
4579  * When passing buffers allocated with this function to sock_queue_err_skb
4580  * it is necessary to wrap the call with sock_hold/sock_put in order to
4581  * prevent the socket from being released prior to being enqueued on
4582  * the sk_error_queue.
4583  */
4584 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
4585 {
4586         struct sock *sk = skb->sk;
4587         struct sk_buff *clone;
4588 
4589         if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
4590                 return NULL;
4591 
4592         clone = skb_clone(skb, GFP_ATOMIC);
4593         if (!clone) {
4594                 sock_put(sk);
4595                 return NULL;
4596         }
4597 
4598         clone->sk = sk;
4599         clone->destructor = sock_efree;
4600 
4601         return clone;
4602 }
4603 EXPORT_SYMBOL(skb_clone_sk);
4604 
4605 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
4606                                         struct sock *sk,
4607                                         int tstype,
4608                                         bool opt_stats)
4609 {
4610         struct sock_exterr_skb *serr;
4611         int err;
4612 
4613         BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
4614 
4615         serr = SKB_EXT_ERR(skb);
4616         memset(serr, 0, sizeof(*serr));
4617         serr->ee.ee_errno = ENOMSG;
4618         serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
4619         serr->ee.ee_info = tstype;
4620         serr->opt_stats = opt_stats;
4621         serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
4622         if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
4623                 serr->ee.ee_data = skb_shinfo(skb)->tskey;
4624                 if (sk->sk_protocol == IPPROTO_TCP &&
4625                     sk->sk_type == SOCK_STREAM)
4626                         serr->ee.ee_data -= sk->sk_tskey;
4627         }
4628 
4629         err = sock_queue_err_skb(sk, skb);
4630 
4631         if (err)
4632                 kfree_skb(skb);
4633 }
4634 
4635 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
4636 {
4637         bool ret;
4638 
4639         if (likely(sysctl_tstamp_allow_data || tsonly))
4640                 return true;
4641 
4642         read_lock_bh(&sk->sk_callback_lock);
4643         ret = sk->sk_socket && sk->sk_socket->file &&
4644               file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
4645         read_unlock_bh(&sk->sk_callback_lock);
4646         return ret;
4647 }
4648 
4649 void skb_complete_tx_timestamp(struct sk_buff *skb,
4650                                struct skb_shared_hwtstamps *hwtstamps)
4651 {
4652         struct sock *sk = skb->sk;
4653 
4654         if (!skb_may_tx_timestamp(sk, false))
4655                 goto err;
4656 
4657         /* Take a reference to prevent skb_orphan() from freeing the socket,
4658          * but only if the socket refcount is not zero.
4659          */
4660         if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4661                 *skb_hwtstamps(skb) = *hwtstamps;
4662                 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
4663                 sock_put(sk);
4664                 return;
4665         }
4666 
4667 err:
4668         kfree_skb(skb);
4669 }
4670 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
4671 
4672 void __skb_tstamp_tx(struct sk_buff *orig_skb,
4673                      struct skb_shared_hwtstamps *hwtstamps,
4674                      struct sock *sk, int tstype)
4675 {
4676         struct sk_buff *skb;
4677         bool tsonly, opt_stats = false;
4678 
4679         if (!sk)
4680                 return;
4681 
4682         if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
4683             skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
4684                 return;
4685 
4686         tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
4687         if (!skb_may_tx_timestamp(sk, tsonly))
4688                 return;
4689 
4690         if (tsonly) {
4691 #ifdef CONFIG_INET
4692                 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
4693                     sk->sk_protocol == IPPROTO_TCP &&
4694                     sk->sk_type == SOCK_STREAM) {
4695                         skb = tcp_get_timestamping_opt_stats(sk);
4696                         opt_stats = true;
4697                 } else
4698 #endif
4699                         skb = alloc_skb(0, GFP_ATOMIC);
4700         } else {
4701                 skb = skb_clone(orig_skb, GFP_ATOMIC);
4702         }
4703         if (!skb)
4704                 return;
4705 
4706         if (tsonly) {
4707                 skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
4708                                              SKBTX_ANY_TSTAMP;
4709                 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
4710         }
4711 
4712         if (hwtstamps)
4713                 *skb_hwtstamps(skb) = *hwtstamps;
4714         else
4715                 skb->tstamp = ktime_get_real();
4716 
4717         __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
4718 }
4719 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
4720 
4721 void skb_tstamp_tx(struct sk_buff *orig_skb,
4722                    struct skb_shared_hwtstamps *hwtstamps)
4723 {
4724         return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
4725                                SCM_TSTAMP_SND);
4726 }
4727 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
4728 
4729 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
4730 {
4731         struct sock *sk = skb->sk;
4732         struct sock_exterr_skb *serr;
4733         int err = 1;
4734 
4735         skb->wifi_acked_valid = 1;
4736         skb->wifi_acked = acked;
4737 
4738         serr = SKB_EXT_ERR(skb);
4739         memset(serr, 0, sizeof(*serr));
4740         serr->ee.ee_errno = ENOMSG;
4741         serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
4742 
4743         /* Take a reference to prevent skb_orphan() from freeing the socket,
4744          * but only if the socket refcount is not zero.
4745          */
4746         if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4747                 err = sock_queue_err_skb(sk, skb);
4748                 sock_put(sk);
4749         }
4750         if (err)
4751                 kfree_skb(skb);
4752 }
4753 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
4754 
4755 /**
4756  * skb_partial_csum_set - set up and verify partial csum values for packet
4757  * @skb: the skb to set
4758  * @start: the number of bytes after skb->data to start checksumming.
4759  * @off: the offset from start to place the checksum.
4760  *
4761  * For untrusted partially-checksummed packets, we need to make sure the values
4762  * for skb->csum_start and skb->csum_offset are valid so we don't oops.
4763  *
4764  * This function checks and sets those values and skb->ip_summed: if this
4765  * returns false you should drop the packet.
4766  */
4767 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
4768 {
4769         u32 csum_end = (u32)start + (u32)off + sizeof(__sum16);
4770         u32 csum_start = skb_headroom(skb) + (u32)start;
4771 
4772         if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) {
4773                 net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n",
4774                                      start, off, skb_headroom(skb), skb_headlen(skb));
4775                 return false;
4776         }
4777         skb->ip_summed = CHECKSUM_PARTIAL;
4778         skb->csum_start = csum_start;
4779         skb->csum_offset = off;
4780         skb_set_transport_header(skb, start);
4781         return true;
4782 }
4783 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
4784 
4785 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
4786                                unsigned int max)
4787 {
4788         if (skb_headlen(skb) >= len)
4789                 return 0;
4790 
4791         /* If we need to pullup then pullup to the max, so we
4792          * won't need to do it again.
4793          */
4794         if (max > skb->len)
4795                 max = skb->len;
4796 
4797         if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
4798                 return -ENOMEM;
4799 
4800         if (skb_headlen(skb) < len)
4801                 return -EPROTO;
4802 
4803         return 0;
4804 }
4805 
4806 #define MAX_TCP_HDR_LEN (15 * 4)
4807 
4808 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
4809                                       typeof(IPPROTO_IP) proto,
4810                                       unsigned int off)
4811 {
4812         int err;
4813 
4814         switch (proto) {
4815         case IPPROTO_TCP:
4816                 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
4817                                           off + MAX_TCP_HDR_LEN);
4818                 if (!err && !skb_partial_csum_set(skb, off,
4819                                                   offsetof(struct tcphdr,
4820                                                            check)))
4821                         err = -EPROTO;
4822                 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
4823 
4824         case IPPROTO_UDP:
4825                 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
4826                                           off + sizeof(struct udphdr));
4827                 if (!err && !skb_partial_csum_set(skb, off,
4828                                                   offsetof(struct udphdr,
4829                                                            check)))
4830                         err = -EPROTO;
4831                 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
4832         }
4833 
4834         return ERR_PTR(-EPROTO);
4835 }
4836 
4837 /* This value should be large enough to cover a tagged ethernet header plus
4838  * maximally sized IP and TCP or UDP headers.
4839  */
4840 #define MAX_IP_HDR_LEN 128
4841 
4842 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
4843 {
4844         unsigned int off;
4845         bool fragment;
4846         __sum16 *csum;
4847         int err;
4848 
4849         fragment = false;
4850 
4851         err = skb_maybe_pull_tail(skb,
4852                                   sizeof(struct iphdr),
4853                                   MAX_IP_HDR_LEN);
4854         if (err < 0)
4855                 goto out;
4856 
4857         if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
4858                 fragment = true;
4859 
4860         off = ip_hdrlen(skb);
4861 
4862         err = -EPROTO;
4863 
4864         if (fragment)
4865                 goto out;
4866 
4867         csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
4868         if (IS_ERR(csum))
4869                 return PTR_ERR(csum);
4870 
4871         if (recalculate)
4872                 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
4873                                            ip_hdr(skb)->daddr,
4874                                            skb->len - off,
4875                                            ip_hdr(skb)->protocol, 0);
4876         err = 0;
4877 
4878 out:
4879         return err;
4880 }
4881 
4882 /* This value should be large enough to cover a tagged ethernet header plus
4883  * an IPv6 header, all options, and a maximal TCP or UDP header.
4884  */
4885 #define MAX_IPV6_HDR_LEN 256
4886 
4887 #define OPT_HDR(type, skb, off) \
4888         (type *)(skb_network_header(skb) + (off))
4889 
4890 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
4891 {
4892         int err;
4893         u8 nexthdr;
4894         unsigned int off;
4895         unsigned int len;
4896         bool fragment;
4897         bool done;
4898         __sum16 *csum;
4899 
4900         fragment = false;
4901         done = false;
4902 
4903         off = sizeof(struct ipv6hdr);
4904 
4905         err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4906         if (err < 0)
4907                 goto out;
4908 
4909         nexthdr = ipv6_hdr(skb)->nexthdr;
4910 
4911         len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4912         while (off <= len && !done) {
4913                 switch (nexthdr) {
4914                 case IPPROTO_DSTOPTS:
4915                 case IPPROTO_HOPOPTS:
4916                 case IPPROTO_ROUTING: {
4917                         struct ipv6_opt_hdr *hp;
4918 
4919                         err = skb_maybe_pull_tail(skb,
4920                                                   off +
4921                                                   sizeof(struct ipv6_opt_hdr),
4922                                                   MAX_IPV6_HDR_LEN);
4923                         if (err < 0)
4924                                 goto out;
4925 
4926                         hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
4927                         nexthdr = hp->nexthdr;
4928                         off += ipv6_optlen(hp);
4929                         break;
4930                 }
4931                 case IPPROTO_AH: {
4932                         struct ip_auth_hdr *hp;
4933 
4934                         err = skb_maybe_pull_tail(skb,
4935                                                   off +
4936                                                   sizeof(struct ip_auth_hdr),
4937                                                   MAX_IPV6_HDR_LEN);
4938                         if (err < 0)
4939                                 goto out;
4940 
4941                         hp = OPT_HDR(struct ip_auth_hdr, skb, off);
4942                         nexthdr = hp->nexthdr;
4943                         off += ipv6_authlen(hp);
4944                         break;
4945                 }
4946                 case IPPROTO_FRAGMENT: {
4947                         struct frag_hdr *hp;
4948 
4949                         err = skb_maybe_pull_tail(skb,
4950                                                   off +
4951                                                   sizeof(struct frag_hdr),
4952                                                   MAX_IPV6_HDR_LEN);
4953                         if (err < 0)
4954                                 goto out;
4955 
4956                         hp = OPT_HDR(struct frag_hdr, skb, off);
4957 
4958                         if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
4959                                 fragment = true;
4960 
4961                         nexthdr = hp->nexthdr;
4962                         off += sizeof(struct frag_hdr);
4963                         break;
4964                 }
4965                 default:
4966                         done = true;
4967                         break;
4968                 }
4969         }
4970 
4971         err = -EPROTO;
4972 
4973         if (!done || fragment)
4974                 goto out;
4975 
4976         csum = skb_checksum_setup_ip(skb, nexthdr, off);
4977         if (IS_ERR(csum))
4978                 return PTR_ERR(csum);
4979 
4980         if (recalculate)
4981                 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4982                                          &ipv6_hdr(skb)->daddr,
4983                                          skb->len - off, nexthdr, 0);
4984         err = 0;
4985 
4986 out:
4987         return err;
4988 }
4989 
4990 /**
4991  * skb_checksum_setup - set up partial checksum offset
4992  * @skb: the skb to set up
4993  * @recalculate: if true the pseudo-header checksum will be recalculated
4994  */
4995 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
4996 {
4997         int err;
4998 
4999         switch (skb->protocol) {
5000         case htons(ETH_P_IP):
5001                 err = skb_checksum_setup_ipv4(skb, recalculate);
5002                 break;
5003 
5004         case htons(ETH_P_IPV6):
5005                 err = skb_checksum_setup_ipv6(skb, recalculate);
5006                 break;
5007 
5008         default:
5009                 err = -EPROTO;
5010                 break;
5011         }
5012 
5013         return err;
5014 }
5015 EXPORT_SYMBOL(skb_checksum_setup);
5016 
5017 /**
5018  * skb_checksum_maybe_trim - maybe trims the given skb
5019  * @skb: the skb to check
5020  * @transport_len: the data length beyond the network header
5021  *
5022  * Checks whether the given skb has data beyond the given transport length.
5023  * If so, returns a cloned skb trimmed to this transport length.
5024  * Otherwise returns the provided skb. Returns NULL in error cases
5025  * (e.g. transport_len exceeds skb length or out-of-memory).
5026  *
5027  * Caller needs to set the skb transport header and free any returned skb if it
5028  * differs from the provided skb.
5029  */
5030 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
5031                                                unsigned int transport_len)
5032 {
5033         struct sk_buff *skb_chk;
5034         unsigned int len = skb_transport_offset(skb) + transport_len;
5035         int ret;
5036 
5037         if (skb->len < len)
5038                 return NULL;
5039         else if (skb->len == len)
5040                 return skb;
5041 
5042         skb_chk = skb_clone(skb, GFP_ATOMIC);
5043         if (!skb_chk)
5044                 return NULL;
5045 
5046         ret = pskb_trim_rcsum(skb_chk, len);
5047         if (ret) {
5048                 kfree_skb(skb_chk);
5049                 return NULL;
5050         }
5051 
5052         return skb_chk;
5053 }
5054 
5055 /**
5056  * skb_checksum_trimmed - validate checksum of an skb
5057  * @skb: the skb to check
5058  * @transport_len: the data length beyond the network header
5059  * @skb_chkf: checksum function to use
5060  *
5061  * Applies the given checksum function skb_chkf to the provided skb.
5062  * Returns a checked and maybe trimmed skb. Returns NULL on error.
5063  *
5064  * If the skb has data beyond the given transport length, then a
5065  * trimmed & cloned skb is checked and returned.
5066  *
5067  * Caller needs to set the skb transport header and free any returned skb if it
5068  * differs from the provided skb.
5069  */
5070 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5071                                      unsigned int transport_len,
5072                                      __sum16(*skb_chkf)(struct sk_buff *skb))
5073 {
5074         struct sk_buff *skb_chk;
5075         unsigned int offset = skb_transport_offset(skb);
5076         __sum16 ret;
5077 
5078         skb_chk = skb_checksum_maybe_trim(skb, transport_len);
5079         if (!skb_chk)
5080                 goto err;
5081 
5082         if (!pskb_may_pull(skb_chk, offset))
5083                 goto err;
5084 
5085         skb_pull_rcsum(skb_chk, offset);
5086         ret = skb_chkf(skb_chk);
5087         skb_push_rcsum(skb_chk, offset);
5088 
5089         if (ret)
5090                 goto err;
5091 
5092         return skb_chk;
5093 
5094 err:
5095         if (skb_chk && skb_chk != skb)
5096                 kfree_skb(skb_chk);
5097 
5098         return NULL;
5099 
5100 }
5101 EXPORT_SYMBOL(skb_checksum_trimmed);
5102 
5103 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
5104 {
5105         net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
5106                              skb->dev->name);
5107 }
5108 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
5109 
5110 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
5111 {
5112         if (head_stolen) {
5113                 skb_release_head_state(skb);
5114                 kmem_cache_free(skbuff_head_cache, skb);
5115         } else {
5116                 __kfree_skb(skb);
5117         }
5118 }
5119 EXPORT_SYMBOL(kfree_skb_partial);
5120 
5121 /**
5122  * skb_try_coalesce - try to merge skb to prior one
5123  * @to: prior buffer
5124  * @from: buffer to add
5125  * @fragstolen: pointer to boolean
5126  * @delta_truesize: how much more was allocated than was requested
5127  */
5128 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
5129                       bool *fragstolen, int *delta_truesize)
5130 {
5131         struct skb_shared_info *to_shinfo, *from_shinfo;
5132         int i, delta, len = from->len;
5133 
5134         *fragstolen = false;
5135 
5136         if (skb_cloned(to))
5137                 return false;
5138 
5139         if (len <= skb_tailroom(to)) {
5140                 if (len)
5141                         BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
5142                 *delta_truesize = 0;
5143                 return true;
5144         }
5145 
5146         to_shinfo = skb_shinfo(to);
5147         from_shinfo = skb_shinfo(from);
5148         if (to_shinfo->frag_list || from_shinfo->frag_list)
5149                 return false;
5150         if (skb_zcopy(to) || skb_zcopy(from))
5151                 return false;
5152 
5153         if (skb_headlen(from) != 0) {
5154                 struct page *page;
5155                 unsigned int offset;
5156 
5157                 if (to_shinfo->nr_frags +
5158                     from_shinfo->nr_frags >= MAX_SKB_FRAGS)
5159                         return false;
5160 
5161                 if (skb_head_is_locked(from))
5162                         return false;
5163 
5164                 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
5165 
5166                 page = virt_to_head_page(from->head);
5167                 offset = from->data - (unsigned char *)page_address(page);
5168 
5169                 skb_fill_page_desc(to, to_shinfo->nr_frags,
5170                                    page, offset, skb_headlen(from));
5171                 *fragstolen = true;
5172         } else {
5173                 if (to_shinfo->nr_frags +
5174                     from_shinfo->nr_frags > MAX_SKB_FRAGS)
5175                         return false;
5176 
5177                 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
5178         }
5179 
5180         WARN_ON_ONCE(delta < len);
5181 
5182         memcpy(to_shinfo->frags + to_shinfo->nr_frags,
5183                from_shinfo->frags,
5184                from_shinfo->nr_frags * sizeof(skb_frag_t));
5185         to_shinfo->nr_frags += from_shinfo->nr_frags;
5186 
5187         if (!skb_cloned(from))
5188                 from_shinfo->nr_frags = 0;
5189 
5190         /* if the skb is not cloned this does nothing
5191          * since we set nr_frags to 0.
5192          */
5193         for (i = 0; i < from_shinfo->nr_frags; i++)
5194                 __skb_frag_ref(&from_shinfo->frags[i]);
5195 
5196         to->truesize += delta;
5197         to->len += len;
5198         to->data_len += len;
5199 
5200         *delta_truesize = delta;
5201         return true;
5202 }
5203 EXPORT_SYMBOL(skb_try_coalesce);
5204 
5205 /**
5206  * skb_scrub_packet - scrub an skb
5207  *
5208  * @skb: buffer to clean
5209  * @xnet: packet is crossing netns
5210  *
5211  * skb_scrub_packet can be used after encapsulating or decapsulting a packet
5212  * into/from a tunnel. Some information have to be cleared during these
5213  * operations.
5214  * skb_scrub_packet can also be used to clean a skb before injecting it in
5215  * another namespace (@xnet == true). We have to clear all information in the
5216  * skb that could impact namespace isolation.
5217  */
5218 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
5219 {
5220         skb->pkt_type = PACKET_HOST;
5221         skb->skb_iif = 0;
5222         skb->ignore_df = 0;
5223         skb_dst_drop(skb);
5224         skb_ext_reset(skb);
5225         nf_reset_ct(skb);
5226         nf_reset_trace(skb);
5227 
5228 #ifdef CONFIG_NET_SWITCHDEV
5229         skb->offload_fwd_mark = 0;
5230         skb->offload_l3_fwd_mark = 0;
5231 #endif
5232 
5233         if (!xnet)
5234                 return;
5235 
5236         ipvs_reset(skb);
5237         skb->mark = 0;
5238         skb->tstamp = 0;
5239 }
5240 EXPORT_SYMBOL_GPL(skb_scrub_packet);
5241 
5242 /**
5243  * skb_gso_transport_seglen - Return length of individual segments of a gso packet
5244  *
5245  * @skb: GSO skb
5246  *
5247  * skb_gso_transport_seglen is used to determine the real size of the
5248  * individual segments, including Layer4 headers (TCP/UDP).
5249  *
5250  * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
5251  */
5252 static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
5253 {
5254         const struct skb_shared_info *shinfo = skb_shinfo(skb);
5255         unsigned int thlen = 0;
5256 
5257         if (skb->encapsulation) {
5258                 thlen = skb_inner_transport_header(skb) -
5259                         skb_transport_header(skb);
5260 
5261                 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
5262                         thlen += inner_tcp_hdrlen(skb);
5263         } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
5264                 thlen = tcp_hdrlen(skb);
5265         } else if (unlikely(skb_is_gso_sctp(skb))) {
5266                 thlen = sizeof(struct sctphdr);
5267         } else if (shinfo->gso_type & SKB_GSO_UDP_L4) {
5268                 thlen = sizeof(struct udphdr);
5269         }
5270         /* UFO sets gso_size to the size of the fragmentation
5271          * payload, i.e. the size of the L4 (UDP) header is already
5272          * accounted for.
5273          */
5274         return thlen + shinfo->gso_size;
5275 }
5276 
5277 /**
5278  * skb_gso_network_seglen - Return length of individual segments of a gso packet
5279  *
5280  * @skb: GSO skb
5281  *
5282  * skb_gso_network_seglen is used to determine the real size of the
5283  * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
5284  *
5285  * The MAC/L2 header is not accounted for.
5286  */
5287 static unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
5288 {
5289         unsigned int hdr_len = skb_transport_header(skb) -
5290                                skb_network_header(skb);
5291 
5292         return hdr_len + skb_gso_transport_seglen(skb);
5293 }
5294 
5295 /**
5296  * skb_gso_mac_seglen - Return length of individual segments of a gso packet
5297  *
5298  * @skb: GSO skb
5299  *
5300  * skb_gso_mac_seglen is used to determine the real size of the
5301  * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4
5302  * headers (TCP/UDP).
5303  */
5304 static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb)
5305 {
5306         unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
5307 
5308         return hdr_len + skb_gso_transport_seglen(skb);
5309 }
5310 
5311 /**
5312  * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS
5313  *
5314  * There are a couple of instances where we have a GSO skb, and we
5315  * want to determine what size it would be after it is segmented.
5316  *
5317  * We might want to check:
5318  * -    L3+L4+payload size (e.g. IP forwarding)
5319  * - L2+L3+L4+payload size (e.g. sanity check before passing to driver)
5320  *
5321  * This is a helper to do that correctly considering GSO_BY_FRAGS.
5322  *
5323  * @skb: GSO skb
5324  *
5325  * @seg_len: The segmented length (from skb_gso_*_seglen). In the
5326  *           GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS].
5327  *
5328  * @max_len: The maximum permissible length.
5329  *
5330  * Returns true if the segmented length <= max length.
5331  */
5332 static inline bool skb_gso_size_check(const struct sk_buff *skb,
5333                                       unsigned int seg_len,
5334                                       unsigned int max_len) {
5335         const struct skb_shared_info *shinfo = skb_shinfo(skb);
5336         const struct sk_buff *iter;
5337 
5338         if (shinfo->gso_size != GSO_BY_FRAGS)
5339                 return seg_len <= max_len;
5340 
5341         /* Undo this so we can re-use header sizes */
5342         seg_len -= GSO_BY_FRAGS;
5343 
5344         skb_walk_frags(skb, iter) {
5345                 if (seg_len + skb_headlen(iter) > max_len)
5346                         return false;
5347         }
5348 
5349         return true;
5350 }
5351 
5352 /**
5353  * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU?
5354  *
5355  * @skb: GSO skb
5356  * @mtu: MTU to validate against
5357  *
5358  * skb_gso_validate_network_len validates if a given skb will fit a
5359  * wanted MTU once split. It considers L3 headers, L4 headers, and the
5360  * payload.
5361  */
5362 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu)
5363 {
5364         return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu);
5365 }
5366 EXPORT_SYMBOL_GPL(skb_gso_validate_network_len);
5367 
5368 /**
5369  * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length?
5370  *
5371  * @skb: GSO skb
5372  * @len: length to validate against
5373  *
5374  * skb_gso_validate_mac_len validates if a given skb will fit a wanted
5375  * length once split, including L2, L3 and L4 headers and the payload.
5376  */
5377 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len)
5378 {
5379         return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len);
5380 }
5381 EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len);
5382 
5383 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
5384 {
5385         int mac_len, meta_len;
5386         void *meta;
5387 
5388         if (skb_cow(skb, skb_headroom(skb)) < 0) {
5389                 kfree_skb(skb);
5390                 return NULL;
5391         }
5392 
5393         mac_len = skb->data - skb_mac_header(skb);
5394         if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) {
5395                 memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb),
5396                         mac_len - VLAN_HLEN - ETH_TLEN);
5397         }
5398 
5399         meta_len = skb_metadata_len(skb);
5400         if (meta_len) {
5401                 meta = skb_metadata_end(skb) - meta_len;
5402                 memmove(meta + VLAN_HLEN, meta, meta_len);
5403         }
5404 
5405         skb->mac_header += VLAN_HLEN;
5406         return skb;
5407 }
5408 
5409 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
5410 {
5411         struct vlan_hdr *vhdr;
5412         u16 vlan_tci;
5413 
5414         if (unlikely(skb_vlan_tag_present(skb))) {
5415                 /* vlan_tci is already set-up so leave this for another time */
5416                 return skb;
5417         }
5418 
5419         skb = skb_share_check(skb, GFP_ATOMIC);
5420         if (unlikely(!skb))
5421                 goto err_free;
5422 
5423         if (unlikely(!pskb_may_pull(skb, VLAN_HLEN)))
5424                 goto err_free;
5425 
5426         vhdr = (struct vlan_hdr *)skb->data;
5427         vlan_tci = ntohs(vhdr->h_vlan_TCI);
5428         __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
5429 
5430         skb_pull_rcsum(skb, VLAN_HLEN);
5431         vlan_set_encap_proto(skb, vhdr);
5432 
5433         skb = skb_reorder_vlan_header(skb);
5434         if (unlikely(!skb))
5435                 goto err_free;
5436 
5437         skb_reset_network_header(skb);
5438         skb_reset_transport_header(skb);
5439         skb_reset_mac_len(skb);
5440 
5441         return skb;
5442 
5443 err_free:
5444         kfree_skb(skb);
5445         return NULL;
5446 }
5447 EXPORT_SYMBOL(skb_vlan_untag);
5448 
5449 int skb_ensure_writable(struct sk_buff *skb, int write_len)
5450 {
5451         if (!pskb_may_pull(skb, write_len))
5452                 return -ENOMEM;
5453 
5454         if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
5455                 return 0;
5456 
5457         return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5458 }
5459 EXPORT_SYMBOL(skb_ensure_writable);
5460 
5461 /* remove VLAN header from packet and update csum accordingly.
5462  * expects a non skb_vlan_tag_present skb with a vlan tag payload
5463  */
5464 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
5465 {
5466         struct vlan_hdr *vhdr;
5467         int offset = skb->data - skb_mac_header(skb);
5468         int err;
5469 
5470         if (WARN_ONCE(offset,
5471                       "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
5472                       offset)) {
5473                 return -EINVAL;
5474         }
5475 
5476         err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
5477         if (unlikely(err))
5478                 return err;
5479 
5480         skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5481 
5482         vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
5483         *vlan_tci = ntohs(vhdr->h_vlan_TCI);
5484 
5485         memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
5486         __skb_pull(skb, VLAN_HLEN);
5487 
5488         vlan_set_encap_proto(skb, vhdr);
5489         skb->mac_header += VLAN_HLEN;
5490 
5491         if (skb_network_offset(skb) < ETH_HLEN)
5492                 skb_set_network_header(skb, ETH_HLEN);
5493 
5494         skb_reset_mac_len(skb);
5495 
5496         return err;
5497 }
5498 EXPORT_SYMBOL(__skb_vlan_pop);
5499 
5500 /* Pop a vlan tag either from hwaccel or from payload.
5501  * Expects skb->data at mac header.
5502  */
5503 int skb_vlan_pop(struct sk_buff *skb)
5504 {
5505         u16 vlan_tci;
5506         __be16 vlan_proto;
5507         int err;
5508 
5509         if (likely(skb_vlan_tag_present(skb))) {
5510                 __vlan_hwaccel_clear_tag(skb);
5511         } else {
5512                 if (unlikely(!eth_type_vlan(skb->protocol)))
5513                         return 0;
5514 
5515                 err = __skb_vlan_pop(skb, &vlan_tci);
5516                 if (err)
5517                         return err;
5518         }
5519         /* move next vlan tag to hw accel tag */
5520         if (likely(!eth_type_vlan(skb->protocol)))
5521                 return 0;
5522 
5523         vlan_proto = skb->protocol;
5524         err = __skb_vlan_pop(skb, &vlan_tci);
5525         if (unlikely(err))
5526                 return err;
5527 
5528         __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5529         return 0;
5530 }
5531 EXPORT_SYMBOL(skb_vlan_pop);
5532 
5533 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
5534  * Expects skb->data at mac header.
5535  */
5536 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
5537 {
5538         if (skb_vlan_tag_present(skb)) {
5539                 int offset = skb->data - skb_mac_header(skb);
5540                 int err;
5541 
5542                 if (WARN_ONCE(offset,
5543                               "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
5544                               offset)) {
5545                         return -EINVAL;
5546                 }
5547 
5548                 err = __vlan_insert_tag(skb, skb->vlan_proto,
5549                                         skb_vlan_tag_get(skb));
5550                 if (err)
5551                         return err;
5552 
5553                 skb->protocol = skb->vlan_proto;
5554                 skb->mac_len += VLAN_HLEN;
5555 
5556                 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5557         }
5558         __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5559         return 0;
5560 }
5561 EXPORT_SYMBOL(skb_vlan_push);
5562 
5563 /* Update the ethertype of hdr and the skb csum value if required. */
5564 static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr,
5565                              __be16 ethertype)
5566 {
5567         if (skb->ip_summed == CHECKSUM_COMPLETE) {
5568                 __be16 diff[] = { ~hdr->h_proto, ethertype };
5569 
5570                 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5571         }
5572 
5573         hdr->h_proto = ethertype;
5574 }
5575 
5576 /**
5577  * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of
5578  *                   the packet
5579  *
5580  * @skb: buffer
5581  * @mpls_lse: MPLS label stack entry to push
5582  * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848)
5583  * @mac_len: length of the MAC header
5584  * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is
5585  *            ethernet
5586  *
5587  * Expects skb->data at mac header.
5588  *
5589  * Returns 0 on success, -errno otherwise.
5590  */
5591 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
5592                   int mac_len, bool ethernet)
5593 {
5594         struct mpls_shim_hdr *lse;
5595         int err;
5596 
5597         if (unlikely(!eth_p_mpls(mpls_proto)))
5598                 return -EINVAL;
5599 
5600         /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */
5601         if (skb->encapsulation)
5602                 return -EINVAL;
5603 
5604         err = skb_cow_head(skb, MPLS_HLEN);
5605         if (unlikely(err))
5606                 return err;
5607 
5608         if (!skb->inner_protocol) {
5609                 skb_set_inner_network_header(skb, skb_network_offset(skb));
5610                 skb_set_inner_protocol(skb, skb->protocol);
5611         }
5612 
5613         skb_push(skb, MPLS_HLEN);
5614         memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
5615                 mac_len);
5616         skb_reset_mac_header(skb);
5617         skb_set_network_header(skb, mac_len);
5618         skb_reset_mac_len(skb);
5619 
5620         lse = mpls_hdr(skb);
5621         lse->label_stack_entry = mpls_lse;
5622         skb_postpush_rcsum(skb, lse, MPLS_HLEN);
5623 
5624         if (ethernet)
5625                 skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto);
5626         skb->protocol = mpls_proto;
5627 
5628         return 0;
5629 }
5630 EXPORT_SYMBOL_GPL(skb_mpls_push);
5631 
5632 /**
5633  * skb_mpls_pop() - pop the outermost MPLS header
5634  *
5635  * @skb: buffer
5636  * @next_proto: ethertype of header after popped MPLS header
5637  * @mac_len: length of the MAC header
5638  * @ethernet: flag to indicate if the packet is ethernet
5639  *
5640  * Expects skb->data at mac header.
5641  *
5642  * Returns 0 on success, -errno otherwise.
5643  */
5644 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
5645                  bool ethernet)
5646 {
5647         int err;
5648 
5649         if (unlikely(!eth_p_mpls(skb->protocol)))
5650                 return 0;
5651 
5652         err = skb_ensure_writable(skb, mac_len + MPLS_HLEN);
5653         if (unlikely(err))
5654                 return err;
5655 
5656         skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
5657         memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
5658                 mac_len);
5659 
5660         __skb_pull(skb, MPLS_HLEN);
5661         skb_reset_mac_header(skb);
5662         skb_set_network_header(skb, mac_len);
5663 
5664         if (ethernet) {
5665                 struct ethhdr *hdr;
5666 
5667                 /* use mpls_hdr() to get ethertype to account for VLANs. */
5668                 hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
5669                 skb_mod_eth_type(skb, hdr, next_proto);
5670         }
5671         skb->protocol = next_proto;
5672 
5673         return 0;
5674 }
5675 EXPORT_SYMBOL_GPL(skb_mpls_pop);
5676 
5677 /**
5678  * skb_mpls_update_lse() - modify outermost MPLS header and update csum
5679  *
5680  * @skb: buffer
5681  * @mpls_lse: new MPLS label stack entry to update to
5682  *
5683  * Expects skb->data at mac header.
5684  *
5685  * Returns 0 on success, -errno otherwise.
5686  */
5687 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse)
5688 {
5689         int err;
5690 
5691         if (unlikely(!eth_p_mpls(skb->protocol)))
5692                 return -EINVAL;
5693 
5694         err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
5695         if (unlikely(err))
5696                 return err;
5697 
5698         if (skb->ip_summed == CHECKSUM_COMPLETE) {
5699                 __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse };
5700 
5701                 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5702         }
5703 
5704         mpls_hdr(skb)->label_stack_entry = mpls_lse;
5705 
5706         return 0;
5707 }
5708 EXPORT_SYMBOL_GPL(skb_mpls_update_lse);
5709 
5710 /**
5711  * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header
5712  *
5713  * @skb: buffer
5714  *
5715  * Expects skb->data at mac header.
5716  *
5717  * Returns 0 on success, -errno otherwise.
5718  */
5719 int skb_mpls_dec_ttl(struct sk_buff *skb)
5720 {
5721         u32 lse;
5722         u8 ttl;
5723 
5724         if (unlikely(!eth_p_mpls(skb->protocol)))
5725                 return -EINVAL;
5726 
5727         lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry);
5728         ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT;
5729         if (!--ttl)
5730                 return -EINVAL;
5731 
5732         lse &= ~MPLS_LS_TTL_MASK;
5733         lse |= ttl << MPLS_LS_TTL_SHIFT;
5734 
5735         return skb_mpls_update_lse(skb, cpu_to_be32(lse));
5736 }
5737 EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl);
5738 
5739 /**
5740  * alloc_skb_with_frags - allocate skb with page frags
5741  *
5742  * @header_len: size of linear part
5743  * @data_len: needed length in frags
5744  * @max_page_order: max page order desired.
5745  * @errcode: pointer to error code if any
5746  * @gfp_mask: allocation mask
5747  *
5748  * This can be used to allocate a paged skb, given a maximal order for frags.
5749  */
5750 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
5751                                      unsigned long data_len,
5752                                      int max_page_order,
5753                                      int *errcode,
5754                                      gfp_t gfp_mask)
5755 {
5756         int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
5757         unsigned long chunk;
5758         struct sk_buff *skb;
5759         struct page *page;
5760         int i;
5761 
5762         *errcode = -EMSGSIZE;
5763         /* Note this test could be relaxed, if we succeed to allocate
5764          * high order pages...
5765          */
5766         if (npages > MAX_SKB_FRAGS)
5767                 return NULL;
5768 
5769         *errcode = -ENOBUFS;
5770         skb = alloc_skb(header_len, gfp_mask);
5771         if (!skb)
5772                 return NULL;
5773 
5774         skb->truesize += npages << PAGE_SHIFT;
5775 
5776         for (i = 0; npages > 0; i++) {
5777                 int order = max_page_order;
5778 
5779                 while (order) {
5780                         if (npages >= 1 << order) {
5781                                 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
5782                                                    __GFP_COMP |
5783                                                    __GFP_NOWARN,
5784                                                    order);
5785                                 if (page)
5786                                         goto fill_page;
5787                                 /* Do not retry other high order allocations */
5788                                 order = 1;
5789                                 max_page_order = 0;
5790                         }
5791                         order--;
5792                 }
5793                 page = alloc_page(gfp_mask);
5794                 if (!page)
5795                         goto failure;
5796 fill_page:
5797                 chunk = min_t(unsigned long, data_len,
5798                               PAGE_SIZE << order);
5799                 skb_fill_page_desc(skb, i, page, 0, chunk);
5800                 data_len -= chunk;
5801                 npages -= 1 << order;
5802         }
5803         return skb;
5804 
5805 failure:
5806         kfree_skb(skb);
5807         return NULL;
5808 }
5809 EXPORT_SYMBOL(alloc_skb_with_frags);
5810 
5811 /* carve out the first off bytes from skb when off < headlen */
5812 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
5813                                     const int headlen, gfp_t gfp_mask)
5814 {
5815         int i;
5816         int size = skb_end_offset(skb);
5817         int new_hlen = headlen - off;
5818         u8 *data;
5819 
5820         size = SKB_DATA_ALIGN(size);
5821 
5822         if (skb_pfmemalloc(skb))
5823                 gfp_mask |= __GFP_MEMALLOC;
5824         data = kmalloc_reserve(size +
5825                                SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5826                                gfp_mask, NUMA_NO_NODE, NULL);
5827         if (!data)
5828                 return -ENOMEM;
5829 
5830         size = SKB_WITH_OVERHEAD(ksize(data));
5831 
5832         /* Copy real data, and all frags */
5833         skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
5834         skb->len -= off;
5835 
5836         memcpy((struct skb_shared_info *)(data + size),
5837                skb_shinfo(skb),
5838                offsetof(struct skb_shared_info,
5839                         frags[skb_shinfo(skb)->nr_frags]));
5840         if (skb_cloned(skb)) {
5841                 /* drop the old head gracefully */
5842                 if (skb_orphan_frags(skb, gfp_mask)) {
5843                         kfree(data);
5844                         return -ENOMEM;
5845                 }
5846                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
5847                         skb_frag_ref(skb, i);
5848                 if (skb_has_frag_list(skb))
5849                         skb_clone_fraglist(skb);
5850                 skb_release_data(skb);
5851         } else {
5852                 /* we can reuse existing recount- all we did was
5853                  * relocate values
5854                  */
5855                 skb_free_head(skb);
5856         }
5857 
5858         skb->head = data;
5859         skb->data = data;
5860         skb->head_frag = 0;
5861 #ifdef NET_SKBUFF_DATA_USES_OFFSET
5862         skb->end = size;
5863 #else
5864         skb->end = skb->head + size;
5865 #endif
5866         skb_set_tail_pointer(skb, skb_headlen(skb));
5867         skb_headers_offset_update(skb, 0);
5868         skb->cloned = 0;
5869         skb->hdr_len = 0;
5870         skb->nohdr = 0;
5871         atomic_set(&skb_shinfo(skb)->dataref, 1);
5872 
5873         return 0;
5874 }
5875 
5876 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
5877 
5878 /* carve out the first eat bytes from skb's frag_list. May recurse into
5879  * pskb_carve()
5880  */
5881 static int pskb_carve_frag_list(struct sk_buff *skb,
5882                                 struct skb_shared_info *shinfo, int eat,
5883                                 gfp_t gfp_mask)
5884 {
5885         struct sk_buff *list = shinfo->frag_list;
5886         struct sk_buff *clone = NULL;
5887         struct sk_buff *insp = NULL;
5888 
5889         do {
5890                 if (!list) {
5891                         pr_err("Not enough bytes to eat. Want %d\n", eat);
5892                         return -EFAULT;
5893                 }
5894                 if (list->len <= eat) {
5895                         /* Eaten as whole. */
5896                         eat -= list->len;
5897                         list = list->next;
5898                         insp = list;
5899                 } else {
5900                         /* Eaten partially. */
5901                         if (skb_shared(list)) {
5902                                 clone = skb_clone(list, gfp_mask);
5903                                 if (!clone)
5904                                         return -ENOMEM;
5905                                 insp = list->next;
5906                                 list = clone;
5907                         } else {
5908                                 /* This may be pulled without problems. */
5909                                 insp = list;
5910                         }
5911                         if (pskb_carve(list, eat, gfp_mask) < 0) {
5912                                 kfree_skb(clone);
5913                                 return -ENOMEM;
5914                         }
5915                         break;
5916                 }
5917         } while (eat);
5918 
5919         /* Free pulled out fragments. */
5920         while ((list = shinfo->frag_list) != insp) {
5921                 shinfo->frag_list = list->next;
5922                 kfree_skb(list);
5923         }
5924         /* And insert new clone at head. */
5925         if (clone) {
5926                 clone->next = list;
5927                 shinfo->frag_list = clone;
5928         }
5929         return 0;
5930 }
5931 
5932 /* carve off first len bytes from skb. Split line (off) is in the
5933  * non-linear part of skb
5934  */
5935 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
5936                                        int pos, gfp_t gfp_mask)
5937 {
5938         int i, k = 0;
5939         int size = skb_end_offset(skb);
5940         u8 *data;
5941         const int nfrags = skb_shinfo(skb)->nr_frags;
5942         struct skb_shared_info *shinfo;
5943 
5944         size = SKB_DATA_ALIGN(size);
5945 
5946         if (skb_pfmemalloc(skb))
5947                 gfp_mask |= __GFP_MEMALLOC;
5948         data = kmalloc_reserve(size +
5949                                SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5950                                gfp_mask, NUMA_NO_NODE, NULL);
5951         if (!data)
5952                 return -ENOMEM;
5953 
5954         size = SKB_WITH_OVERHEAD(ksize(data));
5955 
5956         memcpy((struct skb_shared_info *)(data + size),
5957                skb_shinfo(skb), offsetof(struct skb_shared_info,
5958                                          frags[skb_shinfo(skb)->nr_frags]));
5959         if (skb_orphan_frags(skb, gfp_mask)) {
5960                 kfree(data);
5961                 return -ENOMEM;
5962         }
5963         shinfo = (struct skb_shared_info *)(data + size);
5964         for (i = 0; i < nfrags; i++) {
5965                 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
5966 
5967                 if (pos + fsize > off) {
5968                         shinfo->frags[k] = skb_shinfo(skb)->frags[i];
5969 
5970                         if (pos < off) {
5971                                 /* Split frag.
5972                                  * We have two variants in this case:
5973                                  * 1. Move all the frag to the second
5974                                  *    part, if it is possible. F.e.
5975                                  *    this approach is mandatory for TUX,
5976                                  *    where splitting is expensive.
5977                                  * 2. Split is accurately. We make this.
5978                                  */
5979                                 skb_frag_off_add(&shinfo->frags[0], off - pos);
5980                                 skb_frag_size_sub(&shinfo->frags[0], off - pos);
5981                         }
5982                         skb_frag_ref(skb, i);
5983                         k++;
5984                 }
5985                 pos += fsize;
5986         }
5987         shinfo->nr_frags = k;
5988         if (skb_has_frag_list(skb))
5989                 skb_clone_fraglist(skb);
5990 
5991         if (k == 0) {
5992                 /* split line is in frag list */
5993                 pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask);
5994         }
5995         skb_release_data(skb);
5996 
5997         skb->head = data;
5998         skb->head_frag = 0;
5999         skb->data = data;
6000 #ifdef NET_SKBUFF_DATA_USES_OFFSET
6001         skb->end = size;
6002 #else
6003         skb->end = skb->head + size;
6004 #endif
6005         skb_reset_tail_pointer(skb);
6006         skb_headers_offset_update(skb, 0);
6007         skb->cloned   = 0;
6008         skb->hdr_len  = 0;
6009         skb->nohdr    = 0;
6010         skb->len -= off;
6011         skb->data_len = skb->len;
6012         atomic_set(&skb_shinfo(skb)->dataref, 1);
6013         return 0;
6014 }
6015 
6016 /* remove len bytes from the beginning of the skb */
6017 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
6018 {
6019         int headlen = skb_headlen(skb);
6020 
6021         if (len < headlen)
6022                 return pskb_carve_inside_header(skb, len, headlen, gfp);
6023         else
6024                 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
6025 }
6026 
6027 /* Extract to_copy bytes starting at off from skb, and return this in
6028  * a new skb
6029  */
6030 struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
6031                              int to_copy, gfp_t gfp)
6032 {
6033         struct sk_buff  *clone = skb_clone(skb, gfp);
6034 
6035         if (!clone)
6036                 return NULL;
6037 
6038         if (pskb_carve(clone, off, gfp) < 0 ||
6039             pskb_trim(clone, to_copy)) {
6040                 kfree_skb(clone);
6041                 return NULL;
6042         }
6043         return clone;
6044 }
6045 EXPORT_SYMBOL(pskb_extract);
6046 
6047 /**
6048  * skb_condense - try to get rid of fragments/frag_list if possible
6049  * @skb: buffer
6050  *
6051  * Can be used to save memory before skb is added to a busy queue.
6052  * If packet has bytes in frags and enough tail room in skb->head,
6053  * pull all of them, so that we can free the frags right now and adjust
6054  * truesize.
6055  * Notes:
6056  *      We do not reallocate skb->head thus can not fail.
6057  *      Caller must re-evaluate skb->truesize if needed.
6058  */
6059 void skb_condense(struct sk_buff *skb)
6060 {
6061         if (skb->data_len) {
6062                 if (skb->data_len > skb->end - skb->tail ||
6063                     skb_cloned(skb))
6064                         return;
6065 
6066                 /* Nice, we can free page frag(s) right now */
6067                 __pskb_pull_tail(skb, skb->data_len);
6068         }
6069         /* At this point, skb->truesize might be over estimated,
6070          * because skb had a fragment, and fragments do not tell
6071          * their truesize.
6072          * When we pulled its content into skb->head, fragment
6073          * was freed, but __pskb_pull_tail() could not possibly
6074          * adjust skb->truesize, not knowing the frag truesize.
6075          */
6076         skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6077 }
6078 
6079 #ifdef CONFIG_SKB_EXTENSIONS
6080 static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id)
6081 {
6082         return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE);
6083 }
6084 
6085 /**
6086  * __skb_ext_alloc - allocate a new skb extensions storage
6087  *
6088  * @flags: See kmalloc().
6089  *
6090  * Returns the newly allocated pointer. The pointer can later attached to a
6091  * skb via __skb_ext_set().
6092  * Note: caller must handle the skb_ext as an opaque data.
6093  */
6094 struct skb_ext *__skb_ext_alloc(gfp_t flags)
6095 {
6096         struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags);
6097 
6098         if (new) {
6099                 memset(new->offset, 0, sizeof(new->offset));
6100                 refcount_set(&new->refcnt, 1);
6101         }
6102 
6103         return new;
6104 }
6105 
6106 static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old,
6107                                          unsigned int old_active)
6108 {
6109         struct skb_ext *new;
6110 
6111         if (refcount_read(&old->refcnt) == 1)
6112                 return old;
6113 
6114         new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
6115         if (!new)
6116                 return NULL;
6117 
6118         memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE);
6119         refcount_set(&new->refcnt, 1);
6120 
6121 #ifdef CONFIG_XFRM
6122         if (old_active & (1 << SKB_EXT_SEC_PATH)) {
6123                 struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH);
6124                 unsigned int i;
6125 
6126                 for (i = 0; i < sp->len; i++)
6127                         xfrm_state_hold(sp->xvec[i]);
6128         }
6129 #endif
6130         __skb_ext_put(old);
6131         return new;
6132 }
6133 
6134 /**
6135  * __skb_ext_set - attach the specified extension storage to this skb
6136  * @skb: buffer
6137  * @id: extension id
6138  * @ext: extension storage previously allocated via __skb_ext_alloc()
6139  *
6140  * Existing extensions, if any, are cleared.
6141  *
6142  * Returns the pointer to the extension.
6143  */
6144 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
6145                     struct skb_ext *ext)
6146 {
6147         unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext);
6148 
6149         skb_ext_put(skb);
6150         newlen = newoff + skb_ext_type_len[id];
6151         ext->chunks = newlen;
6152         ext->offset[id] = newoff;
6153         skb->extensions = ext;
6154         skb->active_extensions = 1 << id;
6155         return skb_ext_get_ptr(ext, id);
6156 }
6157 
6158 /**
6159  * skb_ext_add - allocate space for given extension, COW if needed
6160  * @skb: buffer
6161  * @id: extension to allocate space for
6162  *
6163  * Allocates enough space for the given extension.
6164  * If the extension is already present, a pointer to that extension
6165  * is returned.
6166  *
6167  * If the skb was cloned, COW applies and the returned memory can be
6168  * modified without changing the extension space of clones buffers.
6169  *
6170  * Returns pointer to the extension or NULL on allocation failure.
6171  */
6172 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id)
6173 {
6174         struct skb_ext *new, *old = NULL;
6175         unsigned int newlen, newoff;
6176 
6177         if (skb->active_extensions) {
6178                 old = skb->extensions;
6179 
6180                 new = skb_ext_maybe_cow(old, skb->active_extensions);
6181                 if (!new)
6182                         return NULL;
6183 
6184                 if (__skb_ext_exist(new, id))
6185                         goto set_active;
6186 
6187                 newoff = new->chunks;
6188         } else {
6189                 newoff = SKB_EXT_CHUNKSIZEOF(*new);
6190 
6191                 new = __skb_ext_alloc(GFP_ATOMIC);
6192                 if (!new)
6193                         return NULL;
6194         }
6195 
6196         newlen = newoff + skb_ext_type_len[id];
6197         new->chunks = newlen;
6198         new->offset[id] = newoff;
6199 set_active:
6200         skb->extensions = new;
6201         skb->active_extensions |= 1 << id;
6202         return skb_ext_get_ptr(new, id);
6203 }
6204 EXPORT_SYMBOL(skb_ext_add);
6205 
6206 #ifdef CONFIG_XFRM
6207 static void skb_ext_put_sp(struct sec_path *sp)
6208 {
6209         unsigned int i;
6210 
6211         for (i = 0; i < sp->len; i++)
6212                 xfrm_state_put(sp->xvec[i]);
6213 }
6214 #endif
6215 
6216 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
6217 {
6218         struct skb_ext *ext = skb->extensions;
6219 
6220         skb->active_extensions &= ~(1 << id);
6221         if (skb->active_extensions == 0) {
6222                 skb->extensions = NULL;
6223                 __skb_ext_put(ext);
6224 #ifdef CONFIG_XFRM
6225         } else if (id == SKB_EXT_SEC_PATH &&
6226                    refcount_read(&ext->refcnt) == 1) {
6227                 struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH);
6228 
6229                 skb_ext_put_sp(sp);
6230                 sp->len = 0;
6231 #endif
6232         }
6233 }
6234 EXPORT_SYMBOL(__skb_ext_del);
6235 
6236 void __skb_ext_put(struct skb_ext *ext)
6237 {
6238         /* If this is last clone, nothing can increment
6239          * it after check passes.  Avoids one atomic op.
6240          */
6241         if (refcount_read(&ext->refcnt) == 1)
6242                 goto free_now;
6243 
6244         if (!refcount_dec_and_test(&ext->refcnt))
6245                 return;
6246 free_now:
6247 #ifdef CONFIG_XFRM
6248         if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH))
6249                 skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH));
6250 #endif
6251 
6252         kmem_cache_free(skbuff_ext_cache, ext);
6253 }
6254 EXPORT_SYMBOL(__skb_ext_put);
6255 #endif /* CONFIG_SKB_EXTENSIONS */
6256 

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

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

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

osdn.jp