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

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

Version: ~ [ linux-5.14-rc3 ] ~ [ linux-5.13.5 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.53 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.135 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.198 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.240 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.276 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.276 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ 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 /*
  2  *      Routines having to do with the 'struct sk_buff' memory handlers.
  3  *
  4  *      Authors:        Alan Cox <alan@lxorguk.ukuu.org.uk>
  5  *                      Florian La Roche <rzsfl@rz.uni-sb.de>
  6  *
  7  *      Fixes:
  8  *              Alan Cox        :       Fixed the worst of the load
  9  *                                      balancer bugs.
 10  *              Dave Platt      :       Interrupt stacking fix.
 11  *      Richard Kooijman        :       Timestamp fixes.
 12  *              Alan Cox        :       Changed buffer format.
 13  *              Alan Cox        :       destructor hook for AF_UNIX etc.
 14  *              Linus Torvalds  :       Better skb_clone.
 15  *              Alan Cox        :       Added skb_copy.
 16  *              Alan Cox        :       Added all the changed routines Linus
 17  *                                      only put in the headers
 18  *              Ray VanTassle   :       Fixed --skb->lock in free
 19  *              Alan Cox        :       skb_copy copy arp field
 20  *              Andi Kleen      :       slabified it.
 21  *              Robert Olsson   :       Removed skb_head_pool
 22  *
 23  *      NOTE:
 24  *              The __skb_ routines should be called with interrupts
 25  *      disabled, or you better be *real* sure that the operation is atomic
 26  *      with respect to whatever list is being frobbed (e.g. via lock_sock()
 27  *      or via disabling bottom half handlers, etc).
 28  *
 29  *      This program is free software; you can redistribute it and/or
 30  *      modify it under the terms of the GNU General Public License
 31  *      as published by the Free Software Foundation; either version
 32  *      2 of the License, or (at your option) any later version.
 33  */
 34 
 35 /*
 36  *      The functions in this file will not compile correctly with gcc 2.4.x
 37  */
 38 
 39 #include <linux/module.h>
 40 #include <linux/types.h>
 41 #include <linux/kernel.h>
 42 #include <linux/kmemcheck.h>
 43 #include <linux/mm.h>
 44 #include <linux/interrupt.h>
 45 #include <linux/in.h>
 46 #include <linux/inet.h>
 47 #include <linux/slab.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 
 61 #include <net/protocol.h>
 62 #include <net/dst.h>
 63 #include <net/sock.h>
 64 #include <net/checksum.h>
 65 #include <net/xfrm.h>
 66 
 67 #include <asm/uaccess.h>
 68 #include <asm/system.h>
 69 #include <trace/events/skb.h>
 70 
 71 #include "kmap_skb.h"
 72 
 73 static struct kmem_cache *skbuff_head_cache __read_mostly;
 74 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
 75 
 76 static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
 77                                   struct pipe_buffer *buf)
 78 {
 79         put_page(buf->page);
 80 }
 81 
 82 static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
 83                                 struct pipe_buffer *buf)
 84 {
 85         get_page(buf->page);
 86 }
 87 
 88 static int sock_pipe_buf_steal(struct pipe_inode_info *pipe,
 89                                struct pipe_buffer *buf)
 90 {
 91         return 1;
 92 }
 93 
 94 
 95 /* Pipe buffer operations for a socket. */
 96 static struct pipe_buf_operations sock_pipe_buf_ops = {
 97         .can_merge = 0,
 98         .map = generic_pipe_buf_map,
 99         .unmap = generic_pipe_buf_unmap,
100         .confirm = generic_pipe_buf_confirm,
101         .release = sock_pipe_buf_release,
102         .steal = sock_pipe_buf_steal,
103         .get = sock_pipe_buf_get,
104 };
105 
106 /*
107  *      Keep out-of-line to prevent kernel bloat.
108  *      __builtin_return_address is not used because it is not always
109  *      reliable.
110  */
111 
112 /**
113  *      skb_over_panic  -       private function
114  *      @skb: buffer
115  *      @sz: size
116  *      @here: address
117  *
118  *      Out of line support code for skb_put(). Not user callable.
119  */
120 void skb_over_panic(struct sk_buff *skb, int sz, void *here)
121 {
122         printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
123                           "data:%p tail:%#lx end:%#lx dev:%s\n",
124                here, skb->len, sz, skb->head, skb->data,
125                (unsigned long)skb->tail, (unsigned long)skb->end,
126                skb->dev ? skb->dev->name : "<NULL>");
127         BUG();
128 }
129 EXPORT_SYMBOL(skb_over_panic);
130 
131 /**
132  *      skb_under_panic -       private function
133  *      @skb: buffer
134  *      @sz: size
135  *      @here: address
136  *
137  *      Out of line support code for skb_push(). Not user callable.
138  */
139 
140 void skb_under_panic(struct sk_buff *skb, int sz, void *here)
141 {
142         printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
143                           "data:%p tail:%#lx end:%#lx dev:%s\n",
144                here, skb->len, sz, skb->head, skb->data,
145                (unsigned long)skb->tail, (unsigned long)skb->end,
146                skb->dev ? skb->dev->name : "<NULL>");
147         BUG();
148 }
149 EXPORT_SYMBOL(skb_under_panic);
150 
151 /*      Allocate a new skbuff. We do this ourselves so we can fill in a few
152  *      'private' fields and also do memory statistics to find all the
153  *      [BEEP] leaks.
154  *
155  */
156 
157 /**
158  *      __alloc_skb     -       allocate a network buffer
159  *      @size: size to allocate
160  *      @gfp_mask: allocation mask
161  *      @fclone: allocate from fclone cache instead of head cache
162  *              and allocate a cloned (child) skb
163  *      @node: numa node to allocate memory on
164  *
165  *      Allocate a new &sk_buff. The returned buffer has no headroom and a
166  *      tail room of size bytes. The object has a reference count of one.
167  *      The return is the buffer. On a failure the return is %NULL.
168  *
169  *      Buffers may only be allocated from interrupts using a @gfp_mask of
170  *      %GFP_ATOMIC.
171  */
172 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
173                             int fclone, int node)
174 {
175         struct kmem_cache *cache;
176         struct skb_shared_info *shinfo;
177         struct sk_buff *skb;
178         u8 *data;
179 
180         cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
181 
182         /* Get the HEAD */
183         skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
184         if (!skb)
185                 goto out;
186 
187         size = SKB_DATA_ALIGN(size);
188         data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info),
189                         gfp_mask, node);
190         if (!data)
191                 goto nodata;
192 
193         /*
194          * Only clear those fields we need to clear, not those that we will
195          * actually initialise below. Hence, don't put any more fields after
196          * the tail pointer in struct sk_buff!
197          */
198         memset(skb, 0, offsetof(struct sk_buff, tail));
199         skb->truesize = size + sizeof(struct sk_buff);
200         atomic_set(&skb->users, 1);
201         skb->head = data;
202         skb->data = data;
203         skb_reset_tail_pointer(skb);
204         skb->end = skb->tail + size;
205         kmemcheck_annotate_bitfield(skb, flags1);
206         kmemcheck_annotate_bitfield(skb, flags2);
207 #ifdef NET_SKBUFF_DATA_USES_OFFSET
208         skb->mac_header = ~0U;
209 #endif
210 
211         /* make sure we initialize shinfo sequentially */
212         shinfo = skb_shinfo(skb);
213         atomic_set(&shinfo->dataref, 1);
214         shinfo->nr_frags  = 0;
215         shinfo->gso_size = 0;
216         shinfo->gso_segs = 0;
217         shinfo->gso_type = 0;
218         shinfo->ip6_frag_id = 0;
219         shinfo->tx_flags.flags = 0;
220         skb_frag_list_init(skb);
221         memset(&shinfo->hwtstamps, 0, sizeof(shinfo->hwtstamps));
222 
223         if (fclone) {
224                 struct sk_buff *child = skb + 1;
225                 atomic_t *fclone_ref = (atomic_t *) (child + 1);
226 
227                 kmemcheck_annotate_bitfield(child, flags1);
228                 kmemcheck_annotate_bitfield(child, flags2);
229                 skb->fclone = SKB_FCLONE_ORIG;
230                 atomic_set(fclone_ref, 1);
231 
232                 child->fclone = SKB_FCLONE_UNAVAILABLE;
233         }
234 out:
235         return skb;
236 nodata:
237         kmem_cache_free(cache, skb);
238         skb = NULL;
239         goto out;
240 }
241 EXPORT_SYMBOL(__alloc_skb);
242 
243 /**
244  *      __netdev_alloc_skb - allocate an skbuff for rx on a specific device
245  *      @dev: network device to receive on
246  *      @length: length to allocate
247  *      @gfp_mask: get_free_pages mask, passed to alloc_skb
248  *
249  *      Allocate a new &sk_buff and assign it a usage count of one. The
250  *      buffer has unspecified headroom built in. Users should allocate
251  *      the headroom they think they need without accounting for the
252  *      built in space. The built in space is used for optimisations.
253  *
254  *      %NULL is returned if there is no free memory.
255  */
256 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
257                 unsigned int length, gfp_t gfp_mask)
258 {
259         int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1;
260         struct sk_buff *skb;
261 
262         skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, node);
263         if (likely(skb)) {
264                 skb_reserve(skb, NET_SKB_PAD);
265                 skb->dev = dev;
266         }
267         return skb;
268 }
269 EXPORT_SYMBOL(__netdev_alloc_skb);
270 
271 struct page *__netdev_alloc_page(struct net_device *dev, gfp_t gfp_mask)
272 {
273         int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1;
274         struct page *page;
275 
276         page = alloc_pages_node(node, gfp_mask, 0);
277         return page;
278 }
279 EXPORT_SYMBOL(__netdev_alloc_page);
280 
281 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
282                 int size)
283 {
284         skb_fill_page_desc(skb, i, page, off, size);
285         skb->len += size;
286         skb->data_len += size;
287         skb->truesize += size;
288 }
289 EXPORT_SYMBOL(skb_add_rx_frag);
290 
291 /**
292  *      dev_alloc_skb - allocate an skbuff for receiving
293  *      @length: length to allocate
294  *
295  *      Allocate a new &sk_buff and assign it a usage count of one. The
296  *      buffer has unspecified headroom built in. Users should allocate
297  *      the headroom they think they need without accounting for the
298  *      built in space. The built in space is used for optimisations.
299  *
300  *      %NULL is returned if there is no free memory. Although this function
301  *      allocates memory it can be called from an interrupt.
302  */
303 struct sk_buff *dev_alloc_skb(unsigned int length)
304 {
305         /*
306          * There is more code here than it seems:
307          * __dev_alloc_skb is an inline
308          */
309         return __dev_alloc_skb(length, GFP_ATOMIC);
310 }
311 EXPORT_SYMBOL(dev_alloc_skb);
312 
313 static void skb_drop_list(struct sk_buff **listp)
314 {
315         struct sk_buff *list = *listp;
316 
317         *listp = NULL;
318 
319         do {
320                 struct sk_buff *this = list;
321                 list = list->next;
322                 kfree_skb(this);
323         } while (list);
324 }
325 
326 static inline void skb_drop_fraglist(struct sk_buff *skb)
327 {
328         skb_drop_list(&skb_shinfo(skb)->frag_list);
329 }
330 
331 static void skb_clone_fraglist(struct sk_buff *skb)
332 {
333         struct sk_buff *list;
334 
335         skb_walk_frags(skb, list)
336                 skb_get(list);
337 }
338 
339 static void skb_release_data(struct sk_buff *skb)
340 {
341         if (!skb->cloned ||
342             !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
343                                &skb_shinfo(skb)->dataref)) {
344                 if (skb_shinfo(skb)->nr_frags) {
345                         int i;
346                         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
347                                 put_page(skb_shinfo(skb)->frags[i].page);
348                 }
349 
350                 if (skb_has_frags(skb))
351                         skb_drop_fraglist(skb);
352 
353                 kfree(skb->head);
354         }
355 }
356 
357 /*
358  *      Free an skbuff by memory without cleaning the state.
359  */
360 static void kfree_skbmem(struct sk_buff *skb)
361 {
362         struct sk_buff *other;
363         atomic_t *fclone_ref;
364 
365         switch (skb->fclone) {
366         case SKB_FCLONE_UNAVAILABLE:
367                 kmem_cache_free(skbuff_head_cache, skb);
368                 break;
369 
370         case SKB_FCLONE_ORIG:
371                 fclone_ref = (atomic_t *) (skb + 2);
372                 if (atomic_dec_and_test(fclone_ref))
373                         kmem_cache_free(skbuff_fclone_cache, skb);
374                 break;
375 
376         case SKB_FCLONE_CLONE:
377                 fclone_ref = (atomic_t *) (skb + 1);
378                 other = skb - 1;
379 
380                 /* The clone portion is available for
381                  * fast-cloning again.
382                  */
383                 skb->fclone = SKB_FCLONE_UNAVAILABLE;
384 
385                 if (atomic_dec_and_test(fclone_ref))
386                         kmem_cache_free(skbuff_fclone_cache, other);
387                 break;
388         }
389 }
390 
391 static void skb_release_head_state(struct sk_buff *skb)
392 {
393         skb_dst_drop(skb);
394 #ifdef CONFIG_XFRM
395         secpath_put(skb->sp);
396 #endif
397         if (skb->destructor) {
398                 WARN_ON(in_irq());
399                 skb->destructor(skb);
400         }
401 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
402         nf_conntrack_put(skb->nfct);
403         nf_conntrack_put_reasm(skb->nfct_reasm);
404 #endif
405 #ifdef CONFIG_BRIDGE_NETFILTER
406         nf_bridge_put(skb->nf_bridge);
407 #endif
408 /* XXX: IS this still necessary? - JHS */
409 #ifdef CONFIG_NET_SCHED
410         skb->tc_index = 0;
411 #ifdef CONFIG_NET_CLS_ACT
412         skb->tc_verd = 0;
413 #endif
414 #endif
415 }
416 
417 /* Free everything but the sk_buff shell. */
418 static void skb_release_all(struct sk_buff *skb)
419 {
420         skb_release_head_state(skb);
421         skb_release_data(skb);
422 }
423 
424 /**
425  *      __kfree_skb - private function
426  *      @skb: buffer
427  *
428  *      Free an sk_buff. Release anything attached to the buffer.
429  *      Clean the state. This is an internal helper function. Users should
430  *      always call kfree_skb
431  */
432 
433 void __kfree_skb(struct sk_buff *skb)
434 {
435         skb_release_all(skb);
436         kfree_skbmem(skb);
437 }
438 EXPORT_SYMBOL(__kfree_skb);
439 
440 /**
441  *      kfree_skb - free an sk_buff
442  *      @skb: buffer to free
443  *
444  *      Drop a reference to the buffer and free it if the usage count has
445  *      hit zero.
446  */
447 void kfree_skb(struct sk_buff *skb)
448 {
449         if (unlikely(!skb))
450                 return;
451         if (likely(atomic_read(&skb->users) == 1))
452                 smp_rmb();
453         else if (likely(!atomic_dec_and_test(&skb->users)))
454                 return;
455         trace_kfree_skb(skb, __builtin_return_address(0));
456         __kfree_skb(skb);
457 }
458 EXPORT_SYMBOL(kfree_skb);
459 
460 /**
461  *      consume_skb - free an skbuff
462  *      @skb: buffer to free
463  *
464  *      Drop a ref to the buffer and free it if the usage count has hit zero
465  *      Functions identically to kfree_skb, but kfree_skb assumes that the frame
466  *      is being dropped after a failure and notes that
467  */
468 void consume_skb(struct sk_buff *skb)
469 {
470         if (unlikely(!skb))
471                 return;
472         if (likely(atomic_read(&skb->users) == 1))
473                 smp_rmb();
474         else if (likely(!atomic_dec_and_test(&skb->users)))
475                 return;
476         __kfree_skb(skb);
477 }
478 EXPORT_SYMBOL(consume_skb);
479 
480 /**
481  *      skb_recycle_check - check if skb can be reused for receive
482  *      @skb: buffer
483  *      @skb_size: minimum receive buffer size
484  *
485  *      Checks that the skb passed in is not shared or cloned, and
486  *      that it is linear and its head portion at least as large as
487  *      skb_size so that it can be recycled as a receive buffer.
488  *      If these conditions are met, this function does any necessary
489  *      reference count dropping and cleans up the skbuff as if it
490  *      just came from __alloc_skb().
491  */
492 int skb_recycle_check(struct sk_buff *skb, int skb_size)
493 {
494         struct skb_shared_info *shinfo;
495 
496         if (skb_is_nonlinear(skb) || skb->fclone != SKB_FCLONE_UNAVAILABLE)
497                 return 0;
498 
499         skb_size = SKB_DATA_ALIGN(skb_size + NET_SKB_PAD);
500         if (skb_end_pointer(skb) - skb->head < skb_size)
501                 return 0;
502 
503         if (skb_shared(skb) || skb_cloned(skb))
504                 return 0;
505 
506         skb_release_head_state(skb);
507         shinfo = skb_shinfo(skb);
508         atomic_set(&shinfo->dataref, 1);
509         shinfo->nr_frags = 0;
510         shinfo->gso_size = 0;
511         shinfo->gso_segs = 0;
512         shinfo->gso_type = 0;
513         shinfo->ip6_frag_id = 0;
514         shinfo->tx_flags.flags = 0;
515         skb_frag_list_init(skb);
516         memset(&shinfo->hwtstamps, 0, sizeof(shinfo->hwtstamps));
517 
518         memset(skb, 0, offsetof(struct sk_buff, tail));
519         skb->data = skb->head + NET_SKB_PAD;
520         skb_reset_tail_pointer(skb);
521 
522         return 1;
523 }
524 EXPORT_SYMBOL(skb_recycle_check);
525 
526 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
527 {
528         new->tstamp             = old->tstamp;
529         new->dev                = old->dev;
530         new->transport_header   = old->transport_header;
531         new->network_header     = old->network_header;
532         new->mac_header         = old->mac_header;
533         skb_dst_set(new, dst_clone(skb_dst(old)));
534 #ifdef CONFIG_XFRM
535         new->sp                 = secpath_get(old->sp);
536 #endif
537         memcpy(new->cb, old->cb, sizeof(old->cb));
538         new->csum               = old->csum;
539         new->local_df           = old->local_df;
540         new->pkt_type           = old->pkt_type;
541         new->ip_summed          = old->ip_summed;
542         skb_copy_queue_mapping(new, old);
543         new->priority           = old->priority;
544 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
545         new->ipvs_property      = old->ipvs_property;
546 #endif
547         new->protocol           = old->protocol;
548         new->mark               = old->mark;
549         new->iif                = old->iif;
550         __nf_copy(new, old);
551 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
552     defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
553         new->nf_trace           = old->nf_trace;
554 #endif
555 #ifdef CONFIG_NET_SCHED
556         new->tc_index           = old->tc_index;
557 #ifdef CONFIG_NET_CLS_ACT
558         new->tc_verd            = old->tc_verd;
559 #endif
560 #endif
561         new->vlan_tci           = old->vlan_tci;
562 
563         skb_copy_secmark(new, old);
564 }
565 
566 /*
567  * You should not add any new code to this function.  Add it to
568  * __copy_skb_header above instead.
569  */
570 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
571 {
572 #define C(x) n->x = skb->x
573 
574         n->next = n->prev = NULL;
575         n->sk = NULL;
576         __copy_skb_header(n, skb);
577 
578         C(len);
579         C(data_len);
580         C(mac_len);
581         n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
582         n->cloned = 1;
583         n->nohdr = 0;
584         n->destructor = NULL;
585         C(tail);
586         C(end);
587         C(head);
588         C(data);
589         C(truesize);
590         atomic_set(&n->users, 1);
591 
592         atomic_inc(&(skb_shinfo(skb)->dataref));
593         skb->cloned = 1;
594 
595         return n;
596 #undef C
597 }
598 
599 /**
600  *      skb_morph       -       morph one skb into another
601  *      @dst: the skb to receive the contents
602  *      @src: the skb to supply the contents
603  *
604  *      This is identical to skb_clone except that the target skb is
605  *      supplied by the user.
606  *
607  *      The target skb is returned upon exit.
608  */
609 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
610 {
611         skb_release_all(dst);
612         return __skb_clone(dst, src);
613 }
614 EXPORT_SYMBOL_GPL(skb_morph);
615 
616 /**
617  *      skb_clone       -       duplicate an sk_buff
618  *      @skb: buffer to clone
619  *      @gfp_mask: allocation priority
620  *
621  *      Duplicate an &sk_buff. The new one is not owned by a socket. Both
622  *      copies share the same packet data but not structure. The new
623  *      buffer has a reference count of 1. If the allocation fails the
624  *      function returns %NULL otherwise the new buffer is returned.
625  *
626  *      If this function is called from an interrupt gfp_mask() must be
627  *      %GFP_ATOMIC.
628  */
629 
630 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
631 {
632         struct sk_buff *n;
633 
634         n = skb + 1;
635         if (skb->fclone == SKB_FCLONE_ORIG &&
636             n->fclone == SKB_FCLONE_UNAVAILABLE) {
637                 atomic_t *fclone_ref = (atomic_t *) (n + 1);
638                 n->fclone = SKB_FCLONE_CLONE;
639                 atomic_inc(fclone_ref);
640         } else {
641                 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
642                 if (!n)
643                         return NULL;
644 
645                 kmemcheck_annotate_bitfield(n, flags1);
646                 kmemcheck_annotate_bitfield(n, flags2);
647                 n->fclone = SKB_FCLONE_UNAVAILABLE;
648         }
649 
650         return __skb_clone(n, skb);
651 }
652 EXPORT_SYMBOL(skb_clone);
653 
654 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
655 {
656 #ifndef NET_SKBUFF_DATA_USES_OFFSET
657         /*
658          *      Shift between the two data areas in bytes
659          */
660         unsigned long offset = new->data - old->data;
661 #endif
662 
663         __copy_skb_header(new, old);
664 
665 #ifndef NET_SKBUFF_DATA_USES_OFFSET
666         /* {transport,network,mac}_header are relative to skb->head */
667         new->transport_header += offset;
668         new->network_header   += offset;
669         if (skb_mac_header_was_set(new))
670                 new->mac_header       += offset;
671 #endif
672         skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
673         skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
674         skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
675 }
676 
677 /**
678  *      skb_copy        -       create private copy of an sk_buff
679  *      @skb: buffer to copy
680  *      @gfp_mask: allocation priority
681  *
682  *      Make a copy of both an &sk_buff and its data. This is used when the
683  *      caller wishes to modify the data and needs a private copy of the
684  *      data to alter. Returns %NULL on failure or the pointer to the buffer
685  *      on success. The returned buffer has a reference count of 1.
686  *
687  *      As by-product this function converts non-linear &sk_buff to linear
688  *      one, so that &sk_buff becomes completely private and caller is allowed
689  *      to modify all the data of returned buffer. This means that this
690  *      function is not recommended for use in circumstances when only
691  *      header is going to be modified. Use pskb_copy() instead.
692  */
693 
694 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
695 {
696         int headerlen = skb->data - skb->head;
697         /*
698          *      Allocate the copy buffer
699          */
700         struct sk_buff *n;
701 #ifdef NET_SKBUFF_DATA_USES_OFFSET
702         n = alloc_skb(skb->end + skb->data_len, gfp_mask);
703 #else
704         n = alloc_skb(skb->end - skb->head + skb->data_len, gfp_mask);
705 #endif
706         if (!n)
707                 return NULL;
708 
709         /* Set the data pointer */
710         skb_reserve(n, headerlen);
711         /* Set the tail pointer and length */
712         skb_put(n, skb->len);
713 
714         if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
715                 BUG();
716 
717         copy_skb_header(n, skb);
718         return n;
719 }
720 EXPORT_SYMBOL(skb_copy);
721 
722 /**
723  *      pskb_copy       -       create copy of an sk_buff with private head.
724  *      @skb: buffer to copy
725  *      @gfp_mask: allocation priority
726  *
727  *      Make a copy of both an &sk_buff and part of its data, located
728  *      in header. Fragmented data remain shared. This is used when
729  *      the caller wishes to modify only header of &sk_buff and needs
730  *      private copy of the header to alter. Returns %NULL on failure
731  *      or the pointer to the buffer on success.
732  *      The returned buffer has a reference count of 1.
733  */
734 
735 struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
736 {
737         /*
738          *      Allocate the copy buffer
739          */
740         struct sk_buff *n;
741 #ifdef NET_SKBUFF_DATA_USES_OFFSET
742         n = alloc_skb(skb->end, gfp_mask);
743 #else
744         n = alloc_skb(skb->end - skb->head, gfp_mask);
745 #endif
746         if (!n)
747                 goto out;
748 
749         /* Set the data pointer */
750         skb_reserve(n, skb->data - skb->head);
751         /* Set the tail pointer and length */
752         skb_put(n, skb_headlen(skb));
753         /* Copy the bytes */
754         skb_copy_from_linear_data(skb, n->data, n->len);
755 
756         n->truesize += skb->data_len;
757         n->data_len  = skb->data_len;
758         n->len       = skb->len;
759 
760         if (skb_shinfo(skb)->nr_frags) {
761                 int i;
762 
763                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
764                         skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
765                         get_page(skb_shinfo(n)->frags[i].page);
766                 }
767                 skb_shinfo(n)->nr_frags = i;
768         }
769 
770         if (skb_has_frags(skb)) {
771                 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
772                 skb_clone_fraglist(n);
773         }
774 
775         copy_skb_header(n, skb);
776 out:
777         return n;
778 }
779 EXPORT_SYMBOL(pskb_copy);
780 
781 /**
782  *      pskb_expand_head - reallocate header of &sk_buff
783  *      @skb: buffer to reallocate
784  *      @nhead: room to add at head
785  *      @ntail: room to add at tail
786  *      @gfp_mask: allocation priority
787  *
788  *      Expands (or creates identical copy, if &nhead and &ntail are zero)
789  *      header of skb. &sk_buff itself is not changed. &sk_buff MUST have
790  *      reference count of 1. Returns zero in the case of success or error,
791  *      if expansion failed. In the last case, &sk_buff is not changed.
792  *
793  *      All the pointers pointing into skb header may change and must be
794  *      reloaded after call to this function.
795  */
796 
797 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
798                      gfp_t gfp_mask)
799 {
800         int i;
801         u8 *data;
802 #ifdef NET_SKBUFF_DATA_USES_OFFSET
803         int size = nhead + skb->end + ntail;
804 #else
805         int size = nhead + (skb->end - skb->head) + ntail;
806 #endif
807         long off;
808 
809         BUG_ON(nhead < 0);
810 
811         if (skb_shared(skb))
812                 BUG();
813 
814         size = SKB_DATA_ALIGN(size);
815 
816         data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
817         if (!data)
818                 goto nodata;
819 
820         /* Copy only real data... and, alas, header. This should be
821          * optimized for the cases when header is void. */
822 #ifdef NET_SKBUFF_DATA_USES_OFFSET
823         memcpy(data + nhead, skb->head, skb->tail);
824 #else
825         memcpy(data + nhead, skb->head, skb->tail - skb->head);
826 #endif
827         memcpy(data + size, skb_end_pointer(skb),
828                sizeof(struct skb_shared_info));
829 
830         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
831                 get_page(skb_shinfo(skb)->frags[i].page);
832 
833         if (skb_has_frags(skb))
834                 skb_clone_fraglist(skb);
835 
836         skb_release_data(skb);
837 
838         off = (data + nhead) - skb->head;
839 
840         skb->head     = data;
841         skb->data    += off;
842 #ifdef NET_SKBUFF_DATA_USES_OFFSET
843         skb->end      = size;
844         off           = nhead;
845 #else
846         skb->end      = skb->head + size;
847 #endif
848         /* {transport,network,mac}_header and tail are relative to skb->head */
849         skb->tail             += off;
850         skb->transport_header += off;
851         skb->network_header   += off;
852         if (skb_mac_header_was_set(skb))
853                 skb->mac_header += off;
854         skb->csum_start       += nhead;
855         skb->cloned   = 0;
856         skb->hdr_len  = 0;
857         skb->nohdr    = 0;
858         atomic_set(&skb_shinfo(skb)->dataref, 1);
859         return 0;
860 
861 nodata:
862         return -ENOMEM;
863 }
864 EXPORT_SYMBOL(pskb_expand_head);
865 
866 /* Make private copy of skb with writable head and some headroom */
867 
868 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
869 {
870         struct sk_buff *skb2;
871         int delta = headroom - skb_headroom(skb);
872 
873         if (delta <= 0)
874                 skb2 = pskb_copy(skb, GFP_ATOMIC);
875         else {
876                 skb2 = skb_clone(skb, GFP_ATOMIC);
877                 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
878                                              GFP_ATOMIC)) {
879                         kfree_skb(skb2);
880                         skb2 = NULL;
881                 }
882         }
883         return skb2;
884 }
885 EXPORT_SYMBOL(skb_realloc_headroom);
886 
887 /**
888  *      skb_copy_expand -       copy and expand sk_buff
889  *      @skb: buffer to copy
890  *      @newheadroom: new free bytes at head
891  *      @newtailroom: new free bytes at tail
892  *      @gfp_mask: allocation priority
893  *
894  *      Make a copy of both an &sk_buff and its data and while doing so
895  *      allocate additional space.
896  *
897  *      This is used when the caller wishes to modify the data and needs a
898  *      private copy of the data to alter as well as more space for new fields.
899  *      Returns %NULL on failure or the pointer to the buffer
900  *      on success. The returned buffer has a reference count of 1.
901  *
902  *      You must pass %GFP_ATOMIC as the allocation priority if this function
903  *      is called from an interrupt.
904  */
905 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
906                                 int newheadroom, int newtailroom,
907                                 gfp_t gfp_mask)
908 {
909         /*
910          *      Allocate the copy buffer
911          */
912         struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
913                                       gfp_mask);
914         int oldheadroom = skb_headroom(skb);
915         int head_copy_len, head_copy_off;
916         int off;
917 
918         if (!n)
919                 return NULL;
920 
921         skb_reserve(n, newheadroom);
922 
923         /* Set the tail pointer and length */
924         skb_put(n, skb->len);
925 
926         head_copy_len = oldheadroom;
927         head_copy_off = 0;
928         if (newheadroom <= head_copy_len)
929                 head_copy_len = newheadroom;
930         else
931                 head_copy_off = newheadroom - head_copy_len;
932 
933         /* Copy the linear header and data. */
934         if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
935                           skb->len + head_copy_len))
936                 BUG();
937 
938         copy_skb_header(n, skb);
939 
940         off                  = newheadroom - oldheadroom;
941         n->csum_start       += off;
942 #ifdef NET_SKBUFF_DATA_USES_OFFSET
943         n->transport_header += off;
944         n->network_header   += off;
945         if (skb_mac_header_was_set(skb))
946                 n->mac_header += off;
947 #endif
948 
949         return n;
950 }
951 EXPORT_SYMBOL(skb_copy_expand);
952 
953 /**
954  *      skb_pad                 -       zero pad the tail of an skb
955  *      @skb: buffer to pad
956  *      @pad: space to pad
957  *
958  *      Ensure that a buffer is followed by a padding area that is zero
959  *      filled. Used by network drivers which may DMA or transfer data
960  *      beyond the buffer end onto the wire.
961  *
962  *      May return error in out of memory cases. The skb is freed on error.
963  */
964 
965 int skb_pad(struct sk_buff *skb, int pad)
966 {
967         int err;
968         int ntail;
969 
970         /* If the skbuff is non linear tailroom is always zero.. */
971         if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
972                 memset(skb->data+skb->len, 0, pad);
973                 return 0;
974         }
975 
976         ntail = skb->data_len + pad - (skb->end - skb->tail);
977         if (likely(skb_cloned(skb) || ntail > 0)) {
978                 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
979                 if (unlikely(err))
980                         goto free_skb;
981         }
982 
983         /* FIXME: The use of this function with non-linear skb's really needs
984          * to be audited.
985          */
986         err = skb_linearize(skb);
987         if (unlikely(err))
988                 goto free_skb;
989 
990         memset(skb->data + skb->len, 0, pad);
991         return 0;
992 
993 free_skb:
994         kfree_skb(skb);
995         return err;
996 }
997 EXPORT_SYMBOL(skb_pad);
998 
999 /**
1000  *      skb_put - add data to a buffer
1001  *      @skb: buffer to use
1002  *      @len: amount of data to add
1003  *
1004  *      This function extends the used data area of the buffer. If this would
1005  *      exceed the total buffer size the kernel will panic. A pointer to the
1006  *      first byte of the extra data is returned.
1007  */
1008 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1009 {
1010         unsigned char *tmp = skb_tail_pointer(skb);
1011         SKB_LINEAR_ASSERT(skb);
1012         skb->tail += len;
1013         skb->len  += len;
1014         if (unlikely(skb->tail > skb->end))
1015                 skb_over_panic(skb, len, __builtin_return_address(0));
1016         return tmp;
1017 }
1018 EXPORT_SYMBOL(skb_put);
1019 
1020 /**
1021  *      skb_push - add data to the start of a buffer
1022  *      @skb: buffer to use
1023  *      @len: amount of data to add
1024  *
1025  *      This function extends the used data area of the buffer at the buffer
1026  *      start. If this would exceed the total buffer headroom the kernel will
1027  *      panic. A pointer to the first byte of the extra data is returned.
1028  */
1029 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1030 {
1031         skb->data -= len;
1032         skb->len  += len;
1033         if (unlikely(skb->data<skb->head))
1034                 skb_under_panic(skb, len, __builtin_return_address(0));
1035         return skb->data;
1036 }
1037 EXPORT_SYMBOL(skb_push);
1038 
1039 /**
1040  *      skb_pull - remove data from the start of a buffer
1041  *      @skb: buffer to use
1042  *      @len: amount of data to remove
1043  *
1044  *      This function removes data from the start of a buffer, returning
1045  *      the memory to the headroom. A pointer to the next data in the buffer
1046  *      is returned. Once the data has been pulled future pushes will overwrite
1047  *      the old data.
1048  */
1049 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1050 {
1051         return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1052 }
1053 EXPORT_SYMBOL(skb_pull);
1054 
1055 /**
1056  *      skb_trim - remove end from a buffer
1057  *      @skb: buffer to alter
1058  *      @len: new length
1059  *
1060  *      Cut the length of a buffer down by removing data from the tail. If
1061  *      the buffer is already under the length specified it is not modified.
1062  *      The skb must be linear.
1063  */
1064 void skb_trim(struct sk_buff *skb, unsigned int len)
1065 {
1066         if (skb->len > len)
1067                 __skb_trim(skb, len);
1068 }
1069 EXPORT_SYMBOL(skb_trim);
1070 
1071 /* Trims skb to length len. It can change skb pointers.
1072  */
1073 
1074 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1075 {
1076         struct sk_buff **fragp;
1077         struct sk_buff *frag;
1078         int offset = skb_headlen(skb);
1079         int nfrags = skb_shinfo(skb)->nr_frags;
1080         int i;
1081         int err;
1082 
1083         if (skb_cloned(skb) &&
1084             unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1085                 return err;
1086 
1087         i = 0;
1088         if (offset >= len)
1089                 goto drop_pages;
1090 
1091         for (; i < nfrags; i++) {
1092                 int end = offset + skb_shinfo(skb)->frags[i].size;
1093 
1094                 if (end < len) {
1095                         offset = end;
1096                         continue;
1097                 }
1098 
1099                 skb_shinfo(skb)->frags[i++].size = len - offset;
1100 
1101 drop_pages:
1102                 skb_shinfo(skb)->nr_frags = i;
1103 
1104                 for (; i < nfrags; i++)
1105                         put_page(skb_shinfo(skb)->frags[i].page);
1106 
1107                 if (skb_has_frags(skb))
1108                         skb_drop_fraglist(skb);
1109                 goto done;
1110         }
1111 
1112         for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1113              fragp = &frag->next) {
1114                 int end = offset + frag->len;
1115 
1116                 if (skb_shared(frag)) {
1117                         struct sk_buff *nfrag;
1118 
1119                         nfrag = skb_clone(frag, GFP_ATOMIC);
1120                         if (unlikely(!nfrag))
1121                                 return -ENOMEM;
1122 
1123                         nfrag->next = frag->next;
1124                         kfree_skb(frag);
1125                         frag = nfrag;
1126                         *fragp = frag;
1127                 }
1128 
1129                 if (end < len) {
1130                         offset = end;
1131                         continue;
1132                 }
1133 
1134                 if (end > len &&
1135                     unlikely((err = pskb_trim(frag, len - offset))))
1136                         return err;
1137 
1138                 if (frag->next)
1139                         skb_drop_list(&frag->next);
1140                 break;
1141         }
1142 
1143 done:
1144         if (len > skb_headlen(skb)) {
1145                 skb->data_len -= skb->len - len;
1146                 skb->len       = len;
1147         } else {
1148                 skb->len       = len;
1149                 skb->data_len  = 0;
1150                 skb_set_tail_pointer(skb, len);
1151         }
1152 
1153         return 0;
1154 }
1155 EXPORT_SYMBOL(___pskb_trim);
1156 
1157 /**
1158  *      __pskb_pull_tail - advance tail of skb header
1159  *      @skb: buffer to reallocate
1160  *      @delta: number of bytes to advance tail
1161  *
1162  *      The function makes a sense only on a fragmented &sk_buff,
1163  *      it expands header moving its tail forward and copying necessary
1164  *      data from fragmented part.
1165  *
1166  *      &sk_buff MUST have reference count of 1.
1167  *
1168  *      Returns %NULL (and &sk_buff does not change) if pull failed
1169  *      or value of new tail of skb in the case of success.
1170  *
1171  *      All the pointers pointing into skb header may change and must be
1172  *      reloaded after call to this function.
1173  */
1174 
1175 /* Moves tail of skb head forward, copying data from fragmented part,
1176  * when it is necessary.
1177  * 1. It may fail due to malloc failure.
1178  * 2. It may change skb pointers.
1179  *
1180  * It is pretty complicated. Luckily, it is called only in exceptional cases.
1181  */
1182 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1183 {
1184         /* If skb has not enough free space at tail, get new one
1185          * plus 128 bytes for future expansions. If we have enough
1186          * room at tail, reallocate without expansion only if skb is cloned.
1187          */
1188         int i, k, eat = (skb->tail + delta) - skb->end;
1189 
1190         if (eat > 0 || skb_cloned(skb)) {
1191                 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1192                                      GFP_ATOMIC))
1193                         return NULL;
1194         }
1195 
1196         if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1197                 BUG();
1198 
1199         /* Optimization: no fragments, no reasons to preestimate
1200          * size of pulled pages. Superb.
1201          */
1202         if (!skb_has_frags(skb))
1203                 goto pull_pages;
1204 
1205         /* Estimate size of pulled pages. */
1206         eat = delta;
1207         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1208                 if (skb_shinfo(skb)->frags[i].size >= eat)
1209                         goto pull_pages;
1210                 eat -= skb_shinfo(skb)->frags[i].size;
1211         }
1212 
1213         /* If we need update frag list, we are in troubles.
1214          * Certainly, it possible to add an offset to skb data,
1215          * but taking into account that pulling is expected to
1216          * be very rare operation, it is worth to fight against
1217          * further bloating skb head and crucify ourselves here instead.
1218          * Pure masohism, indeed. 8)8)
1219          */
1220         if (eat) {
1221                 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1222                 struct sk_buff *clone = NULL;
1223                 struct sk_buff *insp = NULL;
1224 
1225                 do {
1226                         BUG_ON(!list);
1227 
1228                         if (list->len <= eat) {
1229                                 /* Eaten as whole. */
1230                                 eat -= list->len;
1231                                 list = list->next;
1232                                 insp = list;
1233                         } else {
1234                                 /* Eaten partially. */
1235 
1236                                 if (skb_shared(list)) {
1237                                         /* Sucks! We need to fork list. :-( */
1238                                         clone = skb_clone(list, GFP_ATOMIC);
1239                                         if (!clone)
1240                                                 return NULL;
1241                                         insp = list->next;
1242                                         list = clone;
1243                                 } else {
1244                                         /* This may be pulled without
1245                                          * problems. */
1246                                         insp = list;
1247                                 }
1248                                 if (!pskb_pull(list, eat)) {
1249                                         kfree_skb(clone);
1250                                         return NULL;
1251                                 }
1252                                 break;
1253                         }
1254                 } while (eat);
1255 
1256                 /* Free pulled out fragments. */
1257                 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1258                         skb_shinfo(skb)->frag_list = list->next;
1259                         kfree_skb(list);
1260                 }
1261                 /* And insert new clone at head. */
1262                 if (clone) {
1263                         clone->next = list;
1264                         skb_shinfo(skb)->frag_list = clone;
1265                 }
1266         }
1267         /* Success! Now we may commit changes to skb data. */
1268 
1269 pull_pages:
1270         eat = delta;
1271         k = 0;
1272         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1273                 if (skb_shinfo(skb)->frags[i].size <= eat) {
1274                         put_page(skb_shinfo(skb)->frags[i].page);
1275                         eat -= skb_shinfo(skb)->frags[i].size;
1276                 } else {
1277                         skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1278                         if (eat) {
1279                                 skb_shinfo(skb)->frags[k].page_offset += eat;
1280                                 skb_shinfo(skb)->frags[k].size -= eat;
1281                                 eat = 0;
1282                         }
1283                         k++;
1284                 }
1285         }
1286         skb_shinfo(skb)->nr_frags = k;
1287 
1288         skb->tail     += delta;
1289         skb->data_len -= delta;
1290 
1291         return skb_tail_pointer(skb);
1292 }
1293 EXPORT_SYMBOL(__pskb_pull_tail);
1294 
1295 /* Copy some data bits from skb to kernel buffer. */
1296 
1297 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1298 {
1299         int start = skb_headlen(skb);
1300         struct sk_buff *frag_iter;
1301         int i, copy;
1302 
1303         if (offset > (int)skb->len - len)
1304                 goto fault;
1305 
1306         /* Copy header. */
1307         if ((copy = start - offset) > 0) {
1308                 if (copy > len)
1309                         copy = len;
1310                 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1311                 if ((len -= copy) == 0)
1312                         return 0;
1313                 offset += copy;
1314                 to     += copy;
1315         }
1316 
1317         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1318                 int end;
1319 
1320                 WARN_ON(start > offset + len);
1321 
1322                 end = start + skb_shinfo(skb)->frags[i].size;
1323                 if ((copy = end - offset) > 0) {
1324                         u8 *vaddr;
1325 
1326                         if (copy > len)
1327                                 copy = len;
1328 
1329                         vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1330                         memcpy(to,
1331                                vaddr + skb_shinfo(skb)->frags[i].page_offset+
1332                                offset - start, copy);
1333                         kunmap_skb_frag(vaddr);
1334 
1335                         if ((len -= copy) == 0)
1336                                 return 0;
1337                         offset += copy;
1338                         to     += copy;
1339                 }
1340                 start = end;
1341         }
1342 
1343         skb_walk_frags(skb, frag_iter) {
1344                 int end;
1345 
1346                 WARN_ON(start > offset + len);
1347 
1348                 end = start + frag_iter->len;
1349                 if ((copy = end - offset) > 0) {
1350                         if (copy > len)
1351                                 copy = len;
1352                         if (skb_copy_bits(frag_iter, offset - start, to, copy))
1353                                 goto fault;
1354                         if ((len -= copy) == 0)
1355                                 return 0;
1356                         offset += copy;
1357                         to     += copy;
1358                 }
1359                 start = end;
1360         }
1361         if (!len)
1362                 return 0;
1363 
1364 fault:
1365         return -EFAULT;
1366 }
1367 EXPORT_SYMBOL(skb_copy_bits);
1368 
1369 /*
1370  * Callback from splice_to_pipe(), if we need to release some pages
1371  * at the end of the spd in case we error'ed out in filling the pipe.
1372  */
1373 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1374 {
1375         put_page(spd->pages[i]);
1376 }
1377 
1378 static inline struct page *linear_to_page(struct page *page, unsigned int *len,
1379                                           unsigned int *offset,
1380                                           struct sk_buff *skb, struct sock *sk)
1381 {
1382         struct page *p = sk->sk_sndmsg_page;
1383         unsigned int off;
1384 
1385         if (!p) {
1386 new_page:
1387                 p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1388                 if (!p)
1389                         return NULL;
1390 
1391                 off = sk->sk_sndmsg_off = 0;
1392                 /* hold one ref to this page until it's full */
1393         } else {
1394                 unsigned int mlen;
1395 
1396                 off = sk->sk_sndmsg_off;
1397                 mlen = PAGE_SIZE - off;
1398                 if (mlen < 64 && mlen < *len) {
1399                         put_page(p);
1400                         goto new_page;
1401                 }
1402 
1403                 *len = min_t(unsigned int, *len, mlen);
1404         }
1405 
1406         memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1407         sk->sk_sndmsg_off += *len;
1408         *offset = off;
1409         get_page(p);
1410 
1411         return p;
1412 }
1413 
1414 /*
1415  * Fill page/offset/length into spd, if it can hold more pages.
1416  */
1417 static inline int spd_fill_page(struct splice_pipe_desc *spd, struct page *page,
1418                                 unsigned int *len, unsigned int offset,
1419                                 struct sk_buff *skb, int linear,
1420                                 struct sock *sk)
1421 {
1422         if (unlikely(spd->nr_pages == PIPE_BUFFERS))
1423                 return 1;
1424 
1425         if (linear) {
1426                 page = linear_to_page(page, len, &offset, skb, sk);
1427                 if (!page)
1428                         return 1;
1429         } else
1430                 get_page(page);
1431 
1432         spd->pages[spd->nr_pages] = page;
1433         spd->partial[spd->nr_pages].len = *len;
1434         spd->partial[spd->nr_pages].offset = offset;
1435         spd->nr_pages++;
1436 
1437         return 0;
1438 }
1439 
1440 static inline void __segment_seek(struct page **page, unsigned int *poff,
1441                                   unsigned int *plen, unsigned int off)
1442 {
1443         unsigned long n;
1444 
1445         *poff += off;
1446         n = *poff / PAGE_SIZE;
1447         if (n)
1448                 *page = nth_page(*page, n);
1449 
1450         *poff = *poff % PAGE_SIZE;
1451         *plen -= off;
1452 }
1453 
1454 static inline int __splice_segment(struct page *page, unsigned int poff,
1455                                    unsigned int plen, unsigned int *off,
1456                                    unsigned int *len, struct sk_buff *skb,
1457                                    struct splice_pipe_desc *spd, int linear,
1458                                    struct sock *sk)
1459 {
1460         if (!*len)
1461                 return 1;
1462 
1463         /* skip this segment if already processed */
1464         if (*off >= plen) {
1465                 *off -= plen;
1466                 return 0;
1467         }
1468 
1469         /* ignore any bits we already processed */
1470         if (*off) {
1471                 __segment_seek(&page, &poff, &plen, *off);
1472                 *off = 0;
1473         }
1474 
1475         do {
1476                 unsigned int flen = min(*len, plen);
1477 
1478                 /* the linear region may spread across several pages  */
1479                 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1480 
1481                 if (spd_fill_page(spd, page, &flen, poff, skb, linear, sk))
1482                         return 1;
1483 
1484                 __segment_seek(&page, &poff, &plen, flen);
1485                 *len -= flen;
1486 
1487         } while (*len && plen);
1488 
1489         return 0;
1490 }
1491 
1492 /*
1493  * Map linear and fragment data from the skb to spd. It reports failure if the
1494  * pipe is full or if we already spliced the requested length.
1495  */
1496 static int __skb_splice_bits(struct sk_buff *skb, unsigned int *offset,
1497                              unsigned int *len, struct splice_pipe_desc *spd,
1498                              struct sock *sk)
1499 {
1500         int seg;
1501 
1502         /*
1503          * map the linear part
1504          */
1505         if (__splice_segment(virt_to_page(skb->data),
1506                              (unsigned long) skb->data & (PAGE_SIZE - 1),
1507                              skb_headlen(skb),
1508                              offset, len, skb, spd, 1, sk))
1509                 return 1;
1510 
1511         /*
1512          * then map the fragments
1513          */
1514         for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1515                 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1516 
1517                 if (__splice_segment(f->page, f->page_offset, f->size,
1518                                      offset, len, skb, spd, 0, sk))
1519                         return 1;
1520         }
1521 
1522         return 0;
1523 }
1524 
1525 /*
1526  * Map data from the skb to a pipe. Should handle both the linear part,
1527  * the fragments, and the frag list. It does NOT handle frag lists within
1528  * the frag list, if such a thing exists. We'd probably need to recurse to
1529  * handle that cleanly.
1530  */
1531 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1532                     struct pipe_inode_info *pipe, unsigned int tlen,
1533                     unsigned int flags)
1534 {
1535         struct partial_page partial[PIPE_BUFFERS];
1536         struct page *pages[PIPE_BUFFERS];
1537         struct splice_pipe_desc spd = {
1538                 .pages = pages,
1539                 .partial = partial,
1540                 .flags = flags,
1541                 .ops = &sock_pipe_buf_ops,
1542                 .spd_release = sock_spd_release,
1543         };
1544         struct sk_buff *frag_iter;
1545         struct sock *sk = skb->sk;
1546 
1547         /*
1548          * __skb_splice_bits() only fails if the output has no room left,
1549          * so no point in going over the frag_list for the error case.
1550          */
1551         if (__skb_splice_bits(skb, &offset, &tlen, &spd, sk))
1552                 goto done;
1553         else if (!tlen)
1554                 goto done;
1555 
1556         /*
1557          * now see if we have a frag_list to map
1558          */
1559         skb_walk_frags(skb, frag_iter) {
1560                 if (!tlen)
1561                         break;
1562                 if (__skb_splice_bits(frag_iter, &offset, &tlen, &spd, sk))
1563                         break;
1564         }
1565 
1566 done:
1567         if (spd.nr_pages) {
1568                 int ret;
1569 
1570                 /*
1571                  * Drop the socket lock, otherwise we have reverse
1572                  * locking dependencies between sk_lock and i_mutex
1573                  * here as compared to sendfile(). We enter here
1574                  * with the socket lock held, and splice_to_pipe() will
1575                  * grab the pipe inode lock. For sendfile() emulation,
1576                  * we call into ->sendpage() with the i_mutex lock held
1577                  * and networking will grab the socket lock.
1578                  */
1579                 release_sock(sk);
1580                 ret = splice_to_pipe(pipe, &spd);
1581                 lock_sock(sk);
1582                 return ret;
1583         }
1584 
1585         return 0;
1586 }
1587 
1588 /**
1589  *      skb_store_bits - store bits from kernel buffer to skb
1590  *      @skb: destination buffer
1591  *      @offset: offset in destination
1592  *      @from: source buffer
1593  *      @len: number of bytes to copy
1594  *
1595  *      Copy the specified number of bytes from the source buffer to the
1596  *      destination skb.  This function handles all the messy bits of
1597  *      traversing fragment lists and such.
1598  */
1599 
1600 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1601 {
1602         int start = skb_headlen(skb);
1603         struct sk_buff *frag_iter;
1604         int i, copy;
1605 
1606         if (offset > (int)skb->len - len)
1607                 goto fault;
1608 
1609         if ((copy = start - offset) > 0) {
1610                 if (copy > len)
1611                         copy = len;
1612                 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1613                 if ((len -= copy) == 0)
1614                         return 0;
1615                 offset += copy;
1616                 from += copy;
1617         }
1618 
1619         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1620                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1621                 int end;
1622 
1623                 WARN_ON(start > offset + len);
1624 
1625                 end = start + frag->size;
1626                 if ((copy = end - offset) > 0) {
1627                         u8 *vaddr;
1628 
1629                         if (copy > len)
1630                                 copy = len;
1631 
1632                         vaddr = kmap_skb_frag(frag);
1633                         memcpy(vaddr + frag->page_offset + offset - start,
1634                                from, copy);
1635                         kunmap_skb_frag(vaddr);
1636 
1637                         if ((len -= copy) == 0)
1638                                 return 0;
1639                         offset += copy;
1640                         from += copy;
1641                 }
1642                 start = end;
1643         }
1644 
1645         skb_walk_frags(skb, frag_iter) {
1646                 int end;
1647 
1648                 WARN_ON(start > offset + len);
1649 
1650                 end = start + frag_iter->len;
1651                 if ((copy = end - offset) > 0) {
1652                         if (copy > len)
1653                                 copy = len;
1654                         if (skb_store_bits(frag_iter, offset - start,
1655                                            from, copy))
1656                                 goto fault;
1657                         if ((len -= copy) == 0)
1658                                 return 0;
1659                         offset += copy;
1660                         from += copy;
1661                 }
1662                 start = end;
1663         }
1664         if (!len)
1665                 return 0;
1666 
1667 fault:
1668         return -EFAULT;
1669 }
1670 EXPORT_SYMBOL(skb_store_bits);
1671 
1672 /* Checksum skb data. */
1673 
1674 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1675                           int len, __wsum csum)
1676 {
1677         int start = skb_headlen(skb);
1678         int i, copy = start - offset;
1679         struct sk_buff *frag_iter;
1680         int pos = 0;
1681 
1682         /* Checksum header. */
1683         if (copy > 0) {
1684                 if (copy > len)
1685                         copy = len;
1686                 csum = csum_partial(skb->data + offset, copy, csum);
1687                 if ((len -= copy) == 0)
1688                         return csum;
1689                 offset += copy;
1690                 pos     = copy;
1691         }
1692 
1693         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1694                 int end;
1695 
1696                 WARN_ON(start > offset + len);
1697 
1698                 end = start + skb_shinfo(skb)->frags[i].size;
1699                 if ((copy = end - offset) > 0) {
1700                         __wsum csum2;
1701                         u8 *vaddr;
1702                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1703 
1704                         if (copy > len)
1705                                 copy = len;
1706                         vaddr = kmap_skb_frag(frag);
1707                         csum2 = csum_partial(vaddr + frag->page_offset +
1708                                              offset - start, copy, 0);
1709                         kunmap_skb_frag(vaddr);
1710                         csum = csum_block_add(csum, csum2, pos);
1711                         if (!(len -= copy))
1712                                 return csum;
1713                         offset += copy;
1714                         pos    += copy;
1715                 }
1716                 start = end;
1717         }
1718 
1719         skb_walk_frags(skb, frag_iter) {
1720                 int end;
1721 
1722                 WARN_ON(start > offset + len);
1723 
1724                 end = start + frag_iter->len;
1725                 if ((copy = end - offset) > 0) {
1726                         __wsum csum2;
1727                         if (copy > len)
1728                                 copy = len;
1729                         csum2 = skb_checksum(frag_iter, offset - start,
1730                                              copy, 0);
1731                         csum = csum_block_add(csum, csum2, pos);
1732                         if ((len -= copy) == 0)
1733                                 return csum;
1734                         offset += copy;
1735                         pos    += copy;
1736                 }
1737                 start = end;
1738         }
1739         BUG_ON(len);
1740 
1741         return csum;
1742 }
1743 EXPORT_SYMBOL(skb_checksum);
1744 
1745 /* Both of above in one bottle. */
1746 
1747 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1748                                     u8 *to, int len, __wsum csum)
1749 {
1750         int start = skb_headlen(skb);
1751         int i, copy = start - offset;
1752         struct sk_buff *frag_iter;
1753         int pos = 0;
1754 
1755         /* Copy header. */
1756         if (copy > 0) {
1757                 if (copy > len)
1758                         copy = len;
1759                 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1760                                                  copy, csum);
1761                 if ((len -= copy) == 0)
1762                         return csum;
1763                 offset += copy;
1764                 to     += copy;
1765                 pos     = copy;
1766         }
1767 
1768         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1769                 int end;
1770 
1771                 WARN_ON(start > offset + len);
1772 
1773                 end = start + skb_shinfo(skb)->frags[i].size;
1774                 if ((copy = end - offset) > 0) {
1775                         __wsum csum2;
1776                         u8 *vaddr;
1777                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1778 
1779                         if (copy > len)
1780                                 copy = len;
1781                         vaddr = kmap_skb_frag(frag);
1782                         csum2 = csum_partial_copy_nocheck(vaddr +
1783                                                           frag->page_offset +
1784                                                           offset - start, to,
1785                                                           copy, 0);
1786                         kunmap_skb_frag(vaddr);
1787                         csum = csum_block_add(csum, csum2, pos);
1788                         if (!(len -= copy))
1789                                 return csum;
1790                         offset += copy;
1791                         to     += copy;
1792                         pos    += copy;
1793                 }
1794                 start = end;
1795         }
1796 
1797         skb_walk_frags(skb, frag_iter) {
1798                 __wsum csum2;
1799                 int end;
1800 
1801                 WARN_ON(start > offset + len);
1802 
1803                 end = start + frag_iter->len;
1804                 if ((copy = end - offset) > 0) {
1805                         if (copy > len)
1806                                 copy = len;
1807                         csum2 = skb_copy_and_csum_bits(frag_iter,
1808                                                        offset - start,
1809                                                        to, copy, 0);
1810                         csum = csum_block_add(csum, csum2, pos);
1811                         if ((len -= copy) == 0)
1812                                 return csum;
1813                         offset += copy;
1814                         to     += copy;
1815                         pos    += copy;
1816                 }
1817                 start = end;
1818         }
1819         BUG_ON(len);
1820         return csum;
1821 }
1822 EXPORT_SYMBOL(skb_copy_and_csum_bits);
1823 
1824 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1825 {
1826         __wsum csum;
1827         long csstart;
1828 
1829         if (skb->ip_summed == CHECKSUM_PARTIAL)
1830                 csstart = skb->csum_start - skb_headroom(skb);
1831         else
1832                 csstart = skb_headlen(skb);
1833 
1834         BUG_ON(csstart > skb_headlen(skb));
1835 
1836         skb_copy_from_linear_data(skb, to, csstart);
1837 
1838         csum = 0;
1839         if (csstart != skb->len)
1840                 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1841                                               skb->len - csstart, 0);
1842 
1843         if (skb->ip_summed == CHECKSUM_PARTIAL) {
1844                 long csstuff = csstart + skb->csum_offset;
1845 
1846                 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
1847         }
1848 }
1849 EXPORT_SYMBOL(skb_copy_and_csum_dev);
1850 
1851 /**
1852  *      skb_dequeue - remove from the head of the queue
1853  *      @list: list to dequeue from
1854  *
1855  *      Remove the head of the list. The list lock is taken so the function
1856  *      may be used safely with other locking list functions. The head item is
1857  *      returned or %NULL if the list is empty.
1858  */
1859 
1860 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1861 {
1862         unsigned long flags;
1863         struct sk_buff *result;
1864 
1865         spin_lock_irqsave(&list->lock, flags);
1866         result = __skb_dequeue(list);
1867         spin_unlock_irqrestore(&list->lock, flags);
1868         return result;
1869 }
1870 EXPORT_SYMBOL(skb_dequeue);
1871 
1872 /**
1873  *      skb_dequeue_tail - remove from the tail of the queue
1874  *      @list: list to dequeue from
1875  *
1876  *      Remove the tail of the list. The list lock is taken so the function
1877  *      may be used safely with other locking list functions. The tail item is
1878  *      returned or %NULL if the list is empty.
1879  */
1880 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1881 {
1882         unsigned long flags;
1883         struct sk_buff *result;
1884 
1885         spin_lock_irqsave(&list->lock, flags);
1886         result = __skb_dequeue_tail(list);
1887         spin_unlock_irqrestore(&list->lock, flags);
1888         return result;
1889 }
1890 EXPORT_SYMBOL(skb_dequeue_tail);
1891 
1892 /**
1893  *      skb_queue_purge - empty a list
1894  *      @list: list to empty
1895  *
1896  *      Delete all buffers on an &sk_buff list. Each buffer is removed from
1897  *      the list and one reference dropped. This function takes the list
1898  *      lock and is atomic with respect to other list locking functions.
1899  */
1900 void skb_queue_purge(struct sk_buff_head *list)
1901 {
1902         struct sk_buff *skb;
1903         while ((skb = skb_dequeue(list)) != NULL)
1904                 kfree_skb(skb);
1905 }
1906 EXPORT_SYMBOL(skb_queue_purge);
1907 
1908 /**
1909  *      skb_queue_head - queue a buffer at the list head
1910  *      @list: list to use
1911  *      @newsk: buffer to queue
1912  *
1913  *      Queue a buffer at the start of the list. This function takes the
1914  *      list lock and can be used safely with other locking &sk_buff functions
1915  *      safely.
1916  *
1917  *      A buffer cannot be placed on two lists at the same time.
1918  */
1919 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1920 {
1921         unsigned long flags;
1922 
1923         spin_lock_irqsave(&list->lock, flags);
1924         __skb_queue_head(list, newsk);
1925         spin_unlock_irqrestore(&list->lock, flags);
1926 }
1927 EXPORT_SYMBOL(skb_queue_head);
1928 
1929 /**
1930  *      skb_queue_tail - queue a buffer at the list tail
1931  *      @list: list to use
1932  *      @newsk: buffer to queue
1933  *
1934  *      Queue a buffer at the tail of the list. This function takes the
1935  *      list lock and can be used safely with other locking &sk_buff functions
1936  *      safely.
1937  *
1938  *      A buffer cannot be placed on two lists at the same time.
1939  */
1940 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1941 {
1942         unsigned long flags;
1943 
1944         spin_lock_irqsave(&list->lock, flags);
1945         __skb_queue_tail(list, newsk);
1946         spin_unlock_irqrestore(&list->lock, flags);
1947 }
1948 EXPORT_SYMBOL(skb_queue_tail);
1949 
1950 /**
1951  *      skb_unlink      -       remove a buffer from a list
1952  *      @skb: buffer to remove
1953  *      @list: list to use
1954  *
1955  *      Remove a packet from a list. The list locks are taken and this
1956  *      function is atomic with respect to other list locked calls
1957  *
1958  *      You must know what list the SKB is on.
1959  */
1960 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1961 {
1962         unsigned long flags;
1963 
1964         spin_lock_irqsave(&list->lock, flags);
1965         __skb_unlink(skb, list);
1966         spin_unlock_irqrestore(&list->lock, flags);
1967 }
1968 EXPORT_SYMBOL(skb_unlink);
1969 
1970 /**
1971  *      skb_append      -       append a buffer
1972  *      @old: buffer to insert after
1973  *      @newsk: buffer to insert
1974  *      @list: list to use
1975  *
1976  *      Place a packet after a given packet in a list. The list locks are taken
1977  *      and this function is atomic with respect to other list locked calls.
1978  *      A buffer cannot be placed on two lists at the same time.
1979  */
1980 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1981 {
1982         unsigned long flags;
1983 
1984         spin_lock_irqsave(&list->lock, flags);
1985         __skb_queue_after(list, old, newsk);
1986         spin_unlock_irqrestore(&list->lock, flags);
1987 }
1988 EXPORT_SYMBOL(skb_append);
1989 
1990 /**
1991  *      skb_insert      -       insert a buffer
1992  *      @old: buffer to insert before
1993  *      @newsk: buffer to insert
1994  *      @list: list to use
1995  *
1996  *      Place a packet before a given packet in a list. The list locks are
1997  *      taken and this function is atomic with respect to other list locked
1998  *      calls.
1999  *
2000  *      A buffer cannot be placed on two lists at the same time.
2001  */
2002 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2003 {
2004         unsigned long flags;
2005 
2006         spin_lock_irqsave(&list->lock, flags);
2007         __skb_insert(newsk, old->prev, old, list);
2008         spin_unlock_irqrestore(&list->lock, flags);
2009 }
2010 EXPORT_SYMBOL(skb_insert);
2011 
2012 static inline void skb_split_inside_header(struct sk_buff *skb,
2013                                            struct sk_buff* skb1,
2014                                            const u32 len, const int pos)
2015 {
2016         int i;
2017 
2018         skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2019                                          pos - len);
2020         /* And move data appendix as is. */
2021         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2022                 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2023 
2024         skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2025         skb_shinfo(skb)->nr_frags  = 0;
2026         skb1->data_len             = skb->data_len;
2027         skb1->len                  += skb1->data_len;
2028         skb->data_len              = 0;
2029         skb->len                   = len;
2030         skb_set_tail_pointer(skb, len);
2031 }
2032 
2033 static inline void skb_split_no_header(struct sk_buff *skb,
2034                                        struct sk_buff* skb1,
2035                                        const u32 len, int pos)
2036 {
2037         int i, k = 0;
2038         const int nfrags = skb_shinfo(skb)->nr_frags;
2039 
2040         skb_shinfo(skb)->nr_frags = 0;
2041         skb1->len                 = skb1->data_len = skb->len - len;
2042         skb->len                  = len;
2043         skb->data_len             = len - pos;
2044 
2045         for (i = 0; i < nfrags; i++) {
2046                 int size = skb_shinfo(skb)->frags[i].size;
2047 
2048                 if (pos + size > len) {
2049                         skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2050 
2051                         if (pos < len) {
2052                                 /* Split frag.
2053                                  * We have two variants in this case:
2054                                  * 1. Move all the frag to the second
2055                                  *    part, if it is possible. F.e.
2056                                  *    this approach is mandatory for TUX,
2057                                  *    where splitting is expensive.
2058                                  * 2. Split is accurately. We make this.
2059                                  */
2060                                 get_page(skb_shinfo(skb)->frags[i].page);
2061                                 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2062                                 skb_shinfo(skb1)->frags[0].size -= len - pos;
2063                                 skb_shinfo(skb)->frags[i].size  = len - pos;
2064                                 skb_shinfo(skb)->nr_frags++;
2065                         }
2066                         k++;
2067                 } else
2068                         skb_shinfo(skb)->nr_frags++;
2069                 pos += size;
2070         }
2071         skb_shinfo(skb1)->nr_frags = k;
2072 }
2073 
2074 /**
2075  * skb_split - Split fragmented skb to two parts at length len.
2076  * @skb: the buffer to split
2077  * @skb1: the buffer to receive the second part
2078  * @len: new length for skb
2079  */
2080 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2081 {
2082         int pos = skb_headlen(skb);
2083 
2084         if (len < pos)  /* Split line is inside header. */
2085                 skb_split_inside_header(skb, skb1, len, pos);
2086         else            /* Second chunk has no header, nothing to copy. */
2087                 skb_split_no_header(skb, skb1, len, pos);
2088 }
2089 EXPORT_SYMBOL(skb_split);
2090 
2091 /* Shifting from/to a cloned skb is a no-go.
2092  *
2093  * Caller cannot keep skb_shinfo related pointers past calling here!
2094  */
2095 static int skb_prepare_for_shift(struct sk_buff *skb)
2096 {
2097         return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2098 }
2099 
2100 /**
2101  * skb_shift - Shifts paged data partially from skb to another
2102  * @tgt: buffer into which tail data gets added
2103  * @skb: buffer from which the paged data comes from
2104  * @shiftlen: shift up to this many bytes
2105  *
2106  * Attempts to shift up to shiftlen worth of bytes, which may be less than
2107  * the length of the skb, from tgt to skb. Returns number bytes shifted.
2108  * It's up to caller to free skb if everything was shifted.
2109  *
2110  * If @tgt runs out of frags, the whole operation is aborted.
2111  *
2112  * Skb cannot include anything else but paged data while tgt is allowed
2113  * to have non-paged data as well.
2114  *
2115  * TODO: full sized shift could be optimized but that would need
2116  * specialized skb free'er to handle frags without up-to-date nr_frags.
2117  */
2118 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2119 {
2120         int from, to, merge, todo;
2121         struct skb_frag_struct *fragfrom, *fragto;
2122 
2123         BUG_ON(shiftlen > skb->len);
2124         BUG_ON(skb_headlen(skb));       /* Would corrupt stream */
2125 
2126         todo = shiftlen;
2127         from = 0;
2128         to = skb_shinfo(tgt)->nr_frags;
2129         fragfrom = &skb_shinfo(skb)->frags[from];
2130 
2131         /* Actual merge is delayed until the point when we know we can
2132          * commit all, so that we don't have to undo partial changes
2133          */
2134         if (!to ||
2135             !skb_can_coalesce(tgt, to, fragfrom->page, fragfrom->page_offset)) {
2136                 merge = -1;
2137         } else {
2138                 merge = to - 1;
2139 
2140                 todo -= fragfrom->size;
2141                 if (todo < 0) {
2142                         if (skb_prepare_for_shift(skb) ||
2143                             skb_prepare_for_shift(tgt))
2144                                 return 0;
2145 
2146                         /* All previous frag pointers might be stale! */
2147                         fragfrom = &skb_shinfo(skb)->frags[from];
2148                         fragto = &skb_shinfo(tgt)->frags[merge];
2149 
2150                         fragto->size += shiftlen;
2151                         fragfrom->size -= shiftlen;
2152                         fragfrom->page_offset += shiftlen;
2153 
2154                         goto onlymerged;
2155                 }
2156 
2157                 from++;
2158         }
2159 
2160         /* Skip full, not-fitting skb to avoid expensive operations */
2161         if ((shiftlen == skb->len) &&
2162             (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2163                 return 0;
2164 
2165         if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2166                 return 0;
2167 
2168         while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2169                 if (to == MAX_SKB_FRAGS)
2170                         return 0;
2171 
2172                 fragfrom = &skb_shinfo(skb)->frags[from];
2173                 fragto = &skb_shinfo(tgt)->frags[to];
2174 
2175                 if (todo >= fragfrom->size) {
2176                         *fragto = *fragfrom;
2177                         todo -= fragfrom->size;
2178                         from++;
2179                         to++;
2180 
2181                 } else {
2182                         get_page(fragfrom->page);
2183                         fragto->page = fragfrom->page;
2184                         fragto->page_offset = fragfrom->page_offset;
2185                         fragto->size = todo;
2186 
2187                         fragfrom->page_offset += todo;
2188                         fragfrom->size -= todo;
2189                         todo = 0;
2190 
2191                         to++;
2192                         break;
2193                 }
2194         }
2195 
2196         /* Ready to "commit" this state change to tgt */
2197         skb_shinfo(tgt)->nr_frags = to;
2198 
2199         if (merge >= 0) {
2200                 fragfrom = &skb_shinfo(skb)->frags[0];
2201                 fragto = &skb_shinfo(tgt)->frags[merge];
2202 
2203                 fragto->size += fragfrom->size;
2204                 put_page(fragfrom->page);
2205         }
2206 
2207         /* Reposition in the original skb */
2208         to = 0;
2209         while (from < skb_shinfo(skb)->nr_frags)
2210                 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2211         skb_shinfo(skb)->nr_frags = to;
2212 
2213         BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2214 
2215 onlymerged:
2216         /* Most likely the tgt won't ever need its checksum anymore, skb on
2217          * the other hand might need it if it needs to be resent
2218          */
2219         tgt->ip_summed = CHECKSUM_PARTIAL;
2220         skb->ip_summed = CHECKSUM_PARTIAL;
2221 
2222         /* Yak, is it really working this way? Some helper please? */
2223         skb->len -= shiftlen;
2224         skb->data_len -= shiftlen;
2225         skb->truesize -= shiftlen;
2226         tgt->len += shiftlen;
2227         tgt->data_len += shiftlen;
2228         tgt->truesize += shiftlen;
2229 
2230         return shiftlen;
2231 }
2232 
2233 /**
2234  * skb_prepare_seq_read - Prepare a sequential read of skb data
2235  * @skb: the buffer to read
2236  * @from: lower offset of data to be read
2237  * @to: upper offset of data to be read
2238  * @st: state variable
2239  *
2240  * Initializes the specified state variable. Must be called before
2241  * invoking skb_seq_read() for the first time.
2242  */
2243 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2244                           unsigned int to, struct skb_seq_state *st)
2245 {
2246         st->lower_offset = from;
2247         st->upper_offset = to;
2248         st->root_skb = st->cur_skb = skb;
2249         st->frag_idx = st->stepped_offset = 0;
2250         st->frag_data = NULL;
2251 }
2252 EXPORT_SYMBOL(skb_prepare_seq_read);
2253 
2254 /**
2255  * skb_seq_read - Sequentially read skb data
2256  * @consumed: number of bytes consumed by the caller so far
2257  * @data: destination pointer for data to be returned
2258  * @st: state variable
2259  *
2260  * Reads a block of skb data at &consumed relative to the
2261  * lower offset specified to skb_prepare_seq_read(). Assigns
2262  * the head of the data block to &data and returns the length
2263  * of the block or 0 if the end of the skb data or the upper
2264  * offset has been reached.
2265  *
2266  * The caller is not required to consume all of the data
2267  * returned, i.e. &consumed is typically set to the number
2268  * of bytes already consumed and the next call to
2269  * skb_seq_read() will return the remaining part of the block.
2270  *
2271  * Note 1: The size of each block of data returned can be arbitary,
2272  *       this limitation is the cost for zerocopy seqeuental
2273  *       reads of potentially non linear data.
2274  *
2275  * Note 2: Fragment lists within fragments are not implemented
2276  *       at the moment, state->root_skb could be replaced with
2277  *       a stack for this purpose.
2278  */
2279 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2280                           struct skb_seq_state *st)
2281 {
2282         unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2283         skb_frag_t *frag;
2284 
2285         if (unlikely(abs_offset >= st->upper_offset))
2286                 return 0;
2287 
2288 next_skb:
2289         block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2290 
2291         if (abs_offset < block_limit && !st->frag_data) {
2292                 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2293                 return block_limit - abs_offset;
2294         }
2295 
2296         if (st->frag_idx == 0 && !st->frag_data)
2297                 st->stepped_offset += skb_headlen(st->cur_skb);
2298 
2299         while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2300                 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2301                 block_limit = frag->size + st->stepped_offset;
2302 
2303                 if (abs_offset < block_limit) {
2304                         if (!st->frag_data)
2305                                 st->frag_data = kmap_skb_frag(frag);
2306 
2307                         *data = (u8 *) st->frag_data + frag->page_offset +
2308                                 (abs_offset - st->stepped_offset);
2309 
2310                         return block_limit - abs_offset;
2311                 }
2312 
2313                 if (st->frag_data) {
2314                         kunmap_skb_frag(st->frag_data);
2315                         st->frag_data = NULL;
2316                 }
2317 
2318                 st->frag_idx++;
2319                 st->stepped_offset += frag->size;
2320         }
2321 
2322         if (st->frag_data) {
2323                 kunmap_skb_frag(st->frag_data);
2324                 st->frag_data = NULL;
2325         }
2326 
2327         if (st->root_skb == st->cur_skb && skb_has_frags(st->root_skb)) {
2328                 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2329                 st->frag_idx = 0;
2330                 goto next_skb;
2331         } else if (st->cur_skb->next) {
2332                 st->cur_skb = st->cur_skb->next;
2333                 st->frag_idx = 0;
2334                 goto next_skb;
2335         }
2336 
2337         return 0;
2338 }
2339 EXPORT_SYMBOL(skb_seq_read);
2340 
2341 /**
2342  * skb_abort_seq_read - Abort a sequential read of skb data
2343  * @st: state variable
2344  *
2345  * Must be called if skb_seq_read() was not called until it
2346  * returned 0.
2347  */
2348 void skb_abort_seq_read(struct skb_seq_state *st)
2349 {
2350         if (st->frag_data)
2351                 kunmap_skb_frag(st->frag_data);
2352 }
2353 EXPORT_SYMBOL(skb_abort_seq_read);
2354 
2355 #define TS_SKB_CB(state)        ((struct skb_seq_state *) &((state)->cb))
2356 
2357 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2358                                           struct ts_config *conf,
2359                                           struct ts_state *state)
2360 {
2361         return skb_seq_read(offset, text, TS_SKB_CB(state));
2362 }
2363 
2364 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2365 {
2366         skb_abort_seq_read(TS_SKB_CB(state));
2367 }
2368 
2369 /**
2370  * skb_find_text - Find a text pattern in skb data
2371  * @skb: the buffer to look in
2372  * @from: search offset
2373  * @to: search limit
2374  * @config: textsearch configuration
2375  * @state: uninitialized textsearch state variable
2376  *
2377  * Finds a pattern in the skb data according to the specified
2378  * textsearch configuration. Use textsearch_next() to retrieve
2379  * subsequent occurrences of the pattern. Returns the offset
2380  * to the first occurrence or UINT_MAX if no match was found.
2381  */
2382 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2383                            unsigned int to, struct ts_config *config,
2384                            struct ts_state *state)
2385 {
2386         unsigned int ret;
2387 
2388         config->get_next_block = skb_ts_get_next_block;
2389         config->finish = skb_ts_finish;
2390 
2391         skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2392 
2393         ret = textsearch_find(config, state);
2394         return (ret <= to - from ? ret : UINT_MAX);
2395 }
2396 EXPORT_SYMBOL(skb_find_text);
2397 
2398 /**
2399  * skb_append_datato_frags: - append the user data to a skb
2400  * @sk: sock  structure
2401  * @skb: skb structure to be appened with user data.
2402  * @getfrag: call back function to be used for getting the user data
2403  * @from: pointer to user message iov
2404  * @length: length of the iov message
2405  *
2406  * Description: This procedure append the user data in the fragment part
2407  * of the skb if any page alloc fails user this procedure returns  -ENOMEM
2408  */
2409 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2410                         int (*getfrag)(void *from, char *to, int offset,
2411                                         int len, int odd, struct sk_buff *skb),
2412                         void *from, int length)
2413 {
2414         int frg_cnt = 0;
2415         skb_frag_t *frag = NULL;
2416         struct page *page = NULL;
2417         int copy, left;
2418         int offset = 0;
2419         int ret;
2420 
2421         do {
2422                 /* Return error if we don't have space for new frag */
2423                 frg_cnt = skb_shinfo(skb)->nr_frags;
2424                 if (frg_cnt >= MAX_SKB_FRAGS)
2425                         return -EFAULT;
2426 
2427                 /* allocate a new page for next frag */
2428                 page = alloc_pages(sk->sk_allocation, 0);
2429 
2430                 /* If alloc_page fails just return failure and caller will
2431                  * free previous allocated pages by doing kfree_skb()
2432                  */
2433                 if (page == NULL)
2434                         return -ENOMEM;
2435 
2436                 /* initialize the next frag */
2437                 sk->sk_sndmsg_page = page;
2438                 sk->sk_sndmsg_off = 0;
2439                 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2440                 skb->truesize += PAGE_SIZE;
2441                 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2442 
2443                 /* get the new initialized frag */
2444                 frg_cnt = skb_shinfo(skb)->nr_frags;
2445                 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2446 
2447                 /* copy the user data to page */
2448                 left = PAGE_SIZE - frag->page_offset;
2449                 copy = (length > left)? left : length;
2450 
2451                 ret = getfrag(from, (page_address(frag->page) +
2452                             frag->page_offset + frag->size),
2453                             offset, copy, 0, skb);
2454                 if (ret < 0)
2455                         return -EFAULT;
2456 
2457                 /* copy was successful so update the size parameters */
2458                 sk->sk_sndmsg_off += copy;
2459                 frag->size += copy;
2460                 skb->len += copy;
2461                 skb->data_len += copy;
2462                 offset += copy;
2463                 length -= copy;
2464 
2465         } while (length > 0);
2466 
2467         return 0;
2468 }
2469 EXPORT_SYMBOL(skb_append_datato_frags);
2470 
2471 /**
2472  *      skb_pull_rcsum - pull skb and update receive checksum
2473  *      @skb: buffer to update
2474  *      @len: length of data pulled
2475  *
2476  *      This function performs an skb_pull on the packet and updates
2477  *      the CHECKSUM_COMPLETE checksum.  It should be used on
2478  *      receive path processing instead of skb_pull unless you know
2479  *      that the checksum difference is zero (e.g., a valid IP header)
2480  *      or you are setting ip_summed to CHECKSUM_NONE.
2481  */
2482 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2483 {
2484         BUG_ON(len > skb->len);
2485         skb->len -= len;
2486         BUG_ON(skb->len < skb->data_len);
2487         skb_postpull_rcsum(skb, skb->data, len);
2488         return skb->data += len;
2489 }
2490 
2491 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2492 
2493 /**
2494  *      skb_segment - Perform protocol segmentation on skb.
2495  *      @skb: buffer to segment
2496  *      @features: features for the output path (see dev->features)
2497  *
2498  *      This function performs segmentation on the given skb.  It returns
2499  *      a pointer to the first in a list of new skbs for the segments.
2500  *      In case of error it returns ERR_PTR(err).
2501  */
2502 struct sk_buff *skb_segment(struct sk_buff *skb, int features)
2503 {
2504         struct sk_buff *segs = NULL;
2505         struct sk_buff *tail = NULL;
2506         struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2507         unsigned int mss = skb_shinfo(skb)->gso_size;
2508         unsigned int doffset = skb->data - skb_mac_header(skb);
2509         unsigned int offset = doffset;
2510         unsigned int headroom;
2511         unsigned int len;
2512         int sg = features & NETIF_F_SG;
2513         int nfrags = skb_shinfo(skb)->nr_frags;
2514         int err = -ENOMEM;
2515         int i = 0;
2516         int pos;
2517 
2518         __skb_push(skb, doffset);
2519         headroom = skb_headroom(skb);
2520         pos = skb_headlen(skb);
2521 
2522         do {
2523                 struct sk_buff *nskb;
2524                 skb_frag_t *frag;
2525                 int hsize;
2526                 int size;
2527 
2528                 len = skb->len - offset;
2529                 if (len > mss)
2530                         len = mss;
2531 
2532                 hsize = skb_headlen(skb) - offset;
2533                 if (hsize < 0)
2534                         hsize = 0;
2535                 if (hsize > len || !sg)
2536                         hsize = len;
2537 
2538                 if (!hsize && i >= nfrags) {
2539                         BUG_ON(fskb->len != len);
2540 
2541                         pos += len;
2542                         nskb = skb_clone(fskb, GFP_ATOMIC);
2543                         fskb = fskb->next;
2544 
2545                         if (unlikely(!nskb))
2546                                 goto err;
2547 
2548                         hsize = skb_end_pointer(nskb) - nskb->head;
2549                         if (skb_cow_head(nskb, doffset + headroom)) {
2550                                 kfree_skb(nskb);
2551                                 goto err;
2552                         }
2553 
2554                         nskb->truesize += skb_end_pointer(nskb) - nskb->head -
2555                                           hsize;
2556                         skb_release_head_state(nskb);
2557                         __skb_push(nskb, doffset);
2558                 } else {
2559                         nskb = alloc_skb(hsize + doffset + headroom,
2560                                          GFP_ATOMIC);
2561 
2562                         if (unlikely(!nskb))
2563                                 goto err;
2564 
2565                         skb_reserve(nskb, headroom);
2566                         __skb_put(nskb, doffset);
2567                 }
2568 
2569                 if (segs)
2570                         tail->next = nskb;
2571                 else
2572                         segs = nskb;
2573                 tail = nskb;
2574 
2575                 __copy_skb_header(nskb, skb);
2576 
2577                 /* nskb and skb might have different headroom */
2578                 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2579                         nskb->csum_start += skb_headroom(nskb) - headroom;
2580 
2581                 skb_reset_mac_header(nskb);
2582                 skb_set_network_header(nskb, skb->mac_len);
2583                 nskb->transport_header = (nskb->network_header +
2584                                           skb_network_header_len(skb));
2585                 nskb->mac_len = nskb->network_header - nskb->mac_header;
2586                 skb_copy_from_linear_data(skb, nskb->data, doffset);
2587 
2588                 if (fskb != skb_shinfo(skb)->frag_list)
2589                         continue;
2590 
2591                 if (!sg) {
2592                         nskb->ip_summed = CHECKSUM_NONE;
2593                         nskb->csum = skb_copy_and_csum_bits(skb, offset,
2594                                                             skb_put(nskb, len),
2595                                                             len, 0);
2596                         continue;
2597                 }
2598 
2599                 frag = skb_shinfo(nskb)->frags;
2600 
2601                 skb_copy_from_linear_data_offset(skb, offset,
2602                                                  skb_put(nskb, hsize), hsize);
2603 
2604                 while (pos < offset + len && i < nfrags) {
2605                         *frag = skb_shinfo(skb)->frags[i];
2606                         get_page(frag->page);
2607                         size = frag->size;
2608 
2609                         if (pos < offset) {
2610                                 frag->page_offset += offset - pos;
2611                                 frag->size -= offset - pos;
2612                         }
2613 
2614                         skb_shinfo(nskb)->nr_frags++;
2615 
2616                         if (pos + size <= offset + len) {
2617                                 i++;
2618                                 pos += size;
2619                         } else {
2620                                 frag->size -= pos + size - (offset + len);
2621                                 goto skip_fraglist;
2622                         }
2623 
2624                         frag++;
2625                 }
2626 
2627                 if (pos < offset + len) {
2628                         struct sk_buff *fskb2 = fskb;
2629 
2630                         BUG_ON(pos + fskb->len != offset + len);
2631 
2632                         pos += fskb->len;
2633                         fskb = fskb->next;
2634 
2635                         if (fskb2->next) {
2636                                 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2637                                 if (!fskb2)
2638                                         goto err;
2639                         } else
2640                                 skb_get(fskb2);
2641 
2642                         SKB_FRAG_ASSERT(nskb);
2643                         skb_shinfo(nskb)->frag_list = fskb2;
2644                 }
2645 
2646 skip_fraglist:
2647                 nskb->data_len = len - hsize;
2648                 nskb->len += nskb->data_len;
2649                 nskb->truesize += nskb->data_len;
2650         } while ((offset += len) < skb->len);
2651 
2652         return segs;
2653 
2654 err:
2655         while ((skb = segs)) {
2656                 segs = skb->next;
2657                 kfree_skb(skb);
2658         }
2659         return ERR_PTR(err);
2660 }
2661 EXPORT_SYMBOL_GPL(skb_segment);
2662 
2663 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2664 {
2665         struct sk_buff *p = *head;
2666         struct sk_buff *nskb;
2667         struct skb_shared_info *skbinfo = skb_shinfo(skb);
2668         struct skb_shared_info *pinfo = skb_shinfo(p);
2669         unsigned int headroom;
2670         unsigned int len = skb_gro_len(skb);
2671         unsigned int offset = skb_gro_offset(skb);
2672         unsigned int headlen = skb_headlen(skb);
2673 
2674         if (p->len + len >= 65536)
2675                 return -E2BIG;
2676 
2677         if (pinfo->frag_list)
2678                 goto merge;
2679         else if (headlen <= offset) {
2680                 skb_frag_t *frag;
2681                 skb_frag_t *frag2;
2682                 int i = skbinfo->nr_frags;
2683                 int nr_frags = pinfo->nr_frags + i;
2684 
2685                 offset -= headlen;
2686 
2687                 if (nr_frags > MAX_SKB_FRAGS)
2688                         return -E2BIG;
2689 
2690                 pinfo->nr_frags = nr_frags;
2691                 skbinfo->nr_frags = 0;
2692 
2693                 frag = pinfo->frags + nr_frags;
2694                 frag2 = skbinfo->frags + i;
2695                 do {
2696                         *--frag = *--frag2;
2697                 } while (--i);
2698 
2699                 frag->page_offset += offset;
2700                 frag->size -= offset;
2701 
2702                 skb->truesize -= skb->data_len;
2703                 skb->len -= skb->data_len;
2704                 skb->data_len = 0;
2705 
2706                 NAPI_GRO_CB(skb)->free = 1;
2707                 goto done;
2708         } else if (skb_gro_len(p) != pinfo->gso_size)
2709                 return -E2BIG;
2710 
2711         headroom = skb_headroom(p);
2712         nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
2713         if (unlikely(!nskb))
2714                 return -ENOMEM;
2715 
2716         __copy_skb_header(nskb, p);
2717         nskb->mac_len = p->mac_len;
2718 
2719         skb_reserve(nskb, headroom);
2720         __skb_put(nskb, skb_gro_offset(p));
2721 
2722         skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
2723         skb_set_network_header(nskb, skb_network_offset(p));
2724         skb_set_transport_header(nskb, skb_transport_offset(p));
2725 
2726         __skb_pull(p, skb_gro_offset(p));
2727         memcpy(skb_mac_header(nskb), skb_mac_header(p),
2728                p->data - skb_mac_header(p));
2729 
2730         *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2731         skb_shinfo(nskb)->frag_list = p;
2732         skb_shinfo(nskb)->gso_size = pinfo->gso_size;
2733         pinfo->gso_size = 0;
2734         skb_header_release(p);
2735         nskb->prev = p;
2736 
2737         nskb->data_len += p->len;
2738         nskb->truesize += p->len;
2739         nskb->len += p->len;
2740 
2741         *head = nskb;
2742         nskb->next = p->next;
2743         p->next = NULL;
2744 
2745         p = nskb;
2746 
2747 merge:
2748         if (offset > headlen) {
2749                 unsigned int eat = offset - headlen;
2750 
2751                 skbinfo->frags[0].page_offset += eat;
2752                 skbinfo->frags[0].size -= eat;
2753                 skb->data_len -= eat;
2754                 skb->len -= eat;
2755                 offset = headlen;
2756         }
2757 
2758         __skb_pull(skb, offset);
2759 
2760         p->prev->next = skb;
2761         p->prev = skb;
2762         skb_header_release(skb);
2763 
2764 done:
2765         NAPI_GRO_CB(p)->count++;
2766         p->data_len += len;
2767         p->truesize += len;
2768         p->len += len;
2769 
2770         NAPI_GRO_CB(skb)->same_flow = 1;
2771         return 0;
2772 }
2773 EXPORT_SYMBOL_GPL(skb_gro_receive);
2774 
2775 void __init skb_init(void)
2776 {
2777         skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2778                                               sizeof(struct sk_buff),
2779                                               0,
2780                                               SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2781                                               NULL);
2782         skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2783                                                 (2*sizeof(struct sk_buff)) +
2784                                                 sizeof(atomic_t),
2785                                                 0,
2786                                                 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2787                                                 NULL);
2788 }
2789 
2790 /**
2791  *      skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2792  *      @skb: Socket buffer containing the buffers to be mapped
2793  *      @sg: The scatter-gather list to map into
2794  *      @offset: The offset into the buffer's contents to start mapping
2795  *      @len: Length of buffer space to be mapped
2796  *
2797  *      Fill the specified scatter-gather list with mappings/pointers into a
2798  *      region of the buffer space attached to a socket buffer.
2799  */
2800 static int
2801 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2802 {
2803         int start = skb_headlen(skb);
2804         int i, copy = start - offset;
2805         struct sk_buff *frag_iter;
2806         int elt = 0;
2807 
2808         if (copy > 0) {
2809                 if (copy > len)
2810                         copy = len;
2811                 sg_set_buf(sg, skb->data + offset, copy);
2812                 elt++;
2813                 if ((len -= copy) == 0)
2814                         return elt;
2815                 offset += copy;
2816         }
2817 
2818         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2819                 int end;
2820 
2821                 WARN_ON(start > offset + len);
2822 
2823                 end = start + skb_shinfo(skb)->frags[i].size;
2824                 if ((copy = end - offset) > 0) {
2825                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2826 
2827                         if (copy > len)
2828                                 copy = len;
2829                         sg_set_page(&sg[elt], frag->page, copy,
2830                                         frag->page_offset+offset-start);
2831                         elt++;
2832                         if (!(len -= copy))
2833                                 return elt;
2834                         offset += copy;
2835                 }
2836                 start = end;
2837         }
2838 
2839         skb_walk_frags(skb, frag_iter) {
2840                 int end;
2841 
2842                 WARN_ON(start > offset + len);
2843 
2844                 end = start + frag_iter->len;
2845                 if ((copy = end - offset) > 0) {
2846                         if (copy > len)
2847                                 copy = len;
2848                         elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
2849                                               copy);
2850                         if ((len -= copy) == 0)
2851                                 return elt;
2852                         offset += copy;
2853                 }
2854                 start = end;
2855         }
2856         BUG_ON(len);
2857         return elt;
2858 }
2859 
2860 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2861 {
2862         int nsg = __skb_to_sgvec(skb, sg, offset, len);
2863 
2864         sg_mark_end(&sg[nsg - 1]);
2865 
2866         return nsg;
2867 }
2868 EXPORT_SYMBOL_GPL(skb_to_sgvec);
2869 
2870 /**
2871  *      skb_cow_data - Check that a socket buffer's data buffers are writable
2872  *      @skb: The socket buffer to check.
2873  *      @tailbits: Amount of trailing space to be added
2874  *      @trailer: Returned pointer to the skb where the @tailbits space begins
2875  *
2876  *      Make sure that the data buffers attached to a socket buffer are
2877  *      writable. If they are not, private copies are made of the data buffers
2878  *      and the socket buffer is set to use these instead.
2879  *
2880  *      If @tailbits is given, make sure that there is space to write @tailbits
2881  *      bytes of data beyond current end of socket buffer.  @trailer will be
2882  *      set to point to the skb in which this space begins.
2883  *
2884  *      The number of scatterlist elements required to completely map the
2885  *      COW'd and extended socket buffer will be returned.
2886  */
2887 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
2888 {
2889         int copyflag;
2890         int elt;
2891         struct sk_buff *skb1, **skb_p;
2892 
2893         /* If skb is cloned or its head is paged, reallocate
2894          * head pulling out all the pages (pages are considered not writable
2895          * at the moment even if they are anonymous).
2896          */
2897         if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
2898             __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
2899                 return -ENOMEM;
2900 
2901         /* Easy case. Most of packets will go this way. */
2902         if (!skb_has_frags(skb)) {
2903                 /* A little of trouble, not enough of space for trailer.
2904                  * This should not happen, when stack is tuned to generate
2905                  * good frames. OK, on miss we reallocate and reserve even more
2906                  * space, 128 bytes is fair. */
2907 
2908                 if (skb_tailroom(skb) < tailbits &&
2909                     pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
2910                         return -ENOMEM;
2911 
2912                 /* Voila! */
2913                 *trailer = skb;
2914                 return 1;
2915         }
2916 
2917         /* Misery. We are in troubles, going to mincer fragments... */
2918 
2919         elt = 1;
2920         skb_p = &skb_shinfo(skb)->frag_list;
2921         copyflag = 0;
2922 
2923         while ((skb1 = *skb_p) != NULL) {
2924                 int ntail = 0;
2925 
2926                 /* The fragment is partially pulled by someone,
2927                  * this can happen on input. Copy it and everything
2928                  * after it. */
2929 
2930                 if (skb_shared(skb1))
2931                         copyflag = 1;
2932 
2933                 /* If the skb is the last, worry about trailer. */
2934 
2935                 if (skb1->next == NULL && tailbits) {
2936                         if (skb_shinfo(skb1)->nr_frags ||
2937                             skb_has_frags(skb1) ||
2938                             skb_tailroom(skb1) < tailbits)
2939                                 ntail = tailbits + 128;
2940                 }
2941 
2942                 if (copyflag ||
2943                     skb_cloned(skb1) ||
2944                     ntail ||
2945                     skb_shinfo(skb1)->nr_frags ||
2946                     skb_has_frags(skb1)) {
2947                         struct sk_buff *skb2;
2948 
2949                         /* Fuck, we are miserable poor guys... */
2950                         if (ntail == 0)
2951                                 skb2 = skb_copy(skb1, GFP_ATOMIC);
2952                         else
2953                                 skb2 = skb_copy_expand(skb1,
2954                                                        skb_headroom(skb1),
2955                                                        ntail,
2956                                                        GFP_ATOMIC);
2957                         if (unlikely(skb2 == NULL))
2958                                 return -ENOMEM;
2959 
2960                         if (skb1->sk)
2961                                 skb_set_owner_w(skb2, skb1->sk);
2962 
2963                         /* Looking around. Are we still alive?
2964                          * OK, link new skb, drop old one */
2965 
2966                         skb2->next = skb1->next;
2967                         *skb_p = skb2;
2968                         kfree_skb(skb1);
2969                         skb1 = skb2;
2970                 }
2971                 elt++;
2972                 *trailer = skb1;
2973                 skb_p = &skb1->next;
2974         }
2975 
2976         return elt;
2977 }
2978 EXPORT_SYMBOL_GPL(skb_cow_data);
2979 
2980 static void sock_rmem_free(struct sk_buff *skb)
2981 {
2982         struct sock *sk = skb->sk;
2983 
2984         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
2985 }
2986 
2987 /*
2988  * Note: We dont mem charge error packets (no sk_forward_alloc changes)
2989  */
2990 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
2991 {
2992         int len = skb->len;
2993 
2994         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
2995             (unsigned)sk->sk_rcvbuf)
2996                 return -ENOMEM;
2997 
2998         skb_orphan(skb);
2999         skb->sk = sk;
3000         skb->destructor = sock_rmem_free;
3001         atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3002 
3003         skb_queue_tail(&sk->sk_error_queue, skb);
3004         if (!sock_flag(sk, SOCK_DEAD))
3005                 sk->sk_data_ready(sk, len);
3006         return 0;
3007 }
3008 EXPORT_SYMBOL(sock_queue_err_skb);
3009 
3010 void skb_tstamp_tx(struct sk_buff *orig_skb,
3011                 struct skb_shared_hwtstamps *hwtstamps)
3012 {
3013         struct sock *sk = orig_skb->sk;
3014         struct sock_exterr_skb *serr;
3015         struct sk_buff *skb;
3016         int err;
3017 
3018         if (!sk)
3019                 return;
3020 
3021         skb = skb_clone(orig_skb, GFP_ATOMIC);
3022         if (!skb)
3023                 return;
3024 
3025         if (hwtstamps) {
3026                 *skb_hwtstamps(skb) =
3027                         *hwtstamps;
3028         } else {
3029                 /*
3030                  * no hardware time stamps available,
3031                  * so keep the skb_shared_tx and only
3032                  * store software time stamp
3033                  */
3034                 skb->tstamp = ktime_get_real();
3035         }
3036 
3037         serr = SKB_EXT_ERR(skb);
3038         memset(serr, 0, sizeof(*serr));
3039         serr->ee.ee_errno = ENOMSG;
3040         serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3041 
3042         err = sock_queue_err_skb(sk, skb);
3043 
3044         if (err)
3045                 kfree_skb(skb);
3046 }
3047 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3048 
3049 
3050 /**
3051  * skb_partial_csum_set - set up and verify partial csum values for packet
3052  * @skb: the skb to set
3053  * @start: the number of bytes after skb->data to start checksumming.
3054  * @off: the offset from start to place the checksum.
3055  *
3056  * For untrusted partially-checksummed packets, we need to make sure the values
3057  * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3058  *
3059  * This function checks and sets those values and skb->ip_summed: if this
3060  * returns false you should drop the packet.
3061  */
3062 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3063 {
3064         if (unlikely(start > skb_headlen(skb)) ||
3065             unlikely((int)start + off > skb_headlen(skb) - 2)) {
3066                 if (net_ratelimit())
3067                         printk(KERN_WARNING
3068                                "bad partial csum: csum=%u/%u len=%u\n",
3069                                start, off, skb_headlen(skb));
3070                 return false;
3071         }
3072         skb->ip_summed = CHECKSUM_PARTIAL;
3073         skb->csum_start = skb_headroom(skb) + start;
3074         skb->csum_offset = off;
3075         return true;
3076 }
3077 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3078 
3079 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3080 {
3081         if (net_ratelimit())
3082                 pr_warning("%s: received packets cannot be forwarded"
3083                            " while LRO is enabled\n", skb->dev->name);
3084 }
3085 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
3086 

~ [ 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