2 * Routines having to do with the 'struct sk_buff' memory handlers.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
8 * Alan Cox : Fixed the worst of the load
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
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).
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.
36 * The functions in this file will not compile correctly with gcc 2.4.x
39 #include <linux/module.h>
40 #include <linux/types.h>
41 #include <linux/kernel.h>
43 #include <linux/interrupt.h>
45 #include <linux/inet.h>
46 #include <linux/slab.h>
47 #include <linux/netdevice.h>
48 #ifdef CONFIG_NET_CLS_ACT
49 #include <net/pkt_sched.h>
51 #include <linux/string.h>
52 #include <linux/skbuff.h>
53 #include <linux/splice.h>
54 #include <linux/cache.h>
55 #include <linux/rtnetlink.h>
56 #include <linux/init.h>
57 #include <linux/scatterlist.h>
58 #include <linux/errqueue.h>
60 #include <net/protocol.h>
63 #include <net/checksum.h>
66 #include <asm/uaccess.h>
67 #include <asm/system.h>
71 static struct kmem_cache *skbuff_head_cache __read_mostly;
72 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
74 static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
75 struct pipe_buffer *buf)
80 static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
81 struct pipe_buffer *buf)
86 static int sock_pipe_buf_steal(struct pipe_inode_info *pipe,
87 struct pipe_buffer *buf)
93 /* Pipe buffer operations for a socket. */
94 static struct pipe_buf_operations sock_pipe_buf_ops = {
96 .map = generic_pipe_buf_map,
97 .unmap = generic_pipe_buf_unmap,
98 .confirm = generic_pipe_buf_confirm,
99 .release = sock_pipe_buf_release,
100 .steal = sock_pipe_buf_steal,
101 .get = sock_pipe_buf_get,
105 * Keep out-of-line to prevent kernel bloat.
106 * __builtin_return_address is not used because it is not always
111 * skb_over_panic - private function
116 * Out of line support code for skb_put(). Not user callable.
118 void skb_over_panic(struct sk_buff *skb, int sz, void *here)
120 printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
121 "data:%p tail:%#lx end:%#lx dev:%s\n",
122 here, skb->len, sz, skb->head, skb->data,
123 (unsigned long)skb->tail, (unsigned long)skb->end,
124 skb->dev ? skb->dev->name : "<NULL>");
127 EXPORT_SYMBOL(skb_over_panic);
130 * skb_under_panic - private function
135 * Out of line support code for skb_push(). Not user callable.
138 void skb_under_panic(struct sk_buff *skb, int sz, void *here)
140 printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
141 "data:%p tail:%#lx end:%#lx dev:%s\n",
142 here, skb->len, sz, skb->head, skb->data,
143 (unsigned long)skb->tail, (unsigned long)skb->end,
144 skb->dev ? skb->dev->name : "<NULL>");
147 EXPORT_SYMBOL(skb_under_panic);
149 void skb_truesize_bug(struct sk_buff *skb)
151 WARN(net_ratelimit(), KERN_ERR "SKB BUG: Invalid truesize (%u) "
152 "len=%u, sizeof(sk_buff)=%Zd\n",
153 skb->truesize, skb->len, sizeof(struct sk_buff));
155 EXPORT_SYMBOL(skb_truesize_bug);
157 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
158 * 'private' fields and also do memory statistics to find all the
164 * __alloc_skb - allocate a network buffer
165 * @size: size to allocate
166 * @gfp_mask: allocation mask
167 * @fclone: allocate from fclone cache instead of head cache
168 * and allocate a cloned (child) skb
169 * @node: numa node to allocate memory on
171 * Allocate a new &sk_buff. The returned buffer has no headroom and a
172 * tail room of size bytes. The object has a reference count of one.
173 * The return is the buffer. On a failure the return is %NULL.
175 * Buffers may only be allocated from interrupts using a @gfp_mask of
178 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
179 int fclone, int node)
181 struct kmem_cache *cache;
182 struct skb_shared_info *shinfo;
186 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
189 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
193 size = SKB_DATA_ALIGN(size);
194 data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info),
200 * Only clear those fields we need to clear, not those that we will
201 * actually initialise below. Hence, don't put any more fields after
202 * the tail pointer in struct sk_buff!
204 memset(skb, 0, offsetof(struct sk_buff, tail));
205 skb->truesize = size + sizeof(struct sk_buff);
206 atomic_set(&skb->users, 1);
209 skb_reset_tail_pointer(skb);
210 skb->end = skb->tail + size;
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 shinfo->frag_list = NULL;
221 memset(&shinfo->hwtstamps, 0, sizeof(shinfo->hwtstamps));
224 struct sk_buff *child = skb + 1;
225 atomic_t *fclone_ref = (atomic_t *) (child + 1);
227 skb->fclone = SKB_FCLONE_ORIG;
228 atomic_set(fclone_ref, 1);
230 child->fclone = SKB_FCLONE_UNAVAILABLE;
235 kmem_cache_free(cache, skb);
239 EXPORT_SYMBOL(__alloc_skb);
242 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
243 * @dev: network device to receive on
244 * @length: length to allocate
245 * @gfp_mask: get_free_pages mask, passed to alloc_skb
247 * Allocate a new &sk_buff and assign it a usage count of one. The
248 * buffer has unspecified headroom built in. Users should allocate
249 * the headroom they think they need without accounting for the
250 * built in space. The built in space is used for optimisations.
252 * %NULL is returned if there is no free memory.
254 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
255 unsigned int length, gfp_t gfp_mask)
257 int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1;
260 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, node);
262 skb_reserve(skb, NET_SKB_PAD);
267 EXPORT_SYMBOL(__netdev_alloc_skb);
269 struct page *__netdev_alloc_page(struct net_device *dev, gfp_t gfp_mask)
271 int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1;
274 page = alloc_pages_node(node, gfp_mask, 0);
277 EXPORT_SYMBOL(__netdev_alloc_page);
279 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
282 skb_fill_page_desc(skb, i, page, off, size);
284 skb->data_len += size;
285 skb->truesize += size;
287 EXPORT_SYMBOL(skb_add_rx_frag);
290 * dev_alloc_skb - allocate an skbuff for receiving
291 * @length: length to allocate
293 * Allocate a new &sk_buff and assign it a usage count of one. The
294 * buffer has unspecified headroom built in. Users should allocate
295 * the headroom they think they need without accounting for the
296 * built in space. The built in space is used for optimisations.
298 * %NULL is returned if there is no free memory. Although this function
299 * allocates memory it can be called from an interrupt.
301 struct sk_buff *dev_alloc_skb(unsigned int length)
304 * There is more code here than it seems:
305 * __dev_alloc_skb is an inline
307 return __dev_alloc_skb(length, GFP_ATOMIC);
309 EXPORT_SYMBOL(dev_alloc_skb);
311 static void skb_drop_list(struct sk_buff **listp)
313 struct sk_buff *list = *listp;
318 struct sk_buff *this = list;
324 static inline void skb_drop_fraglist(struct sk_buff *skb)
326 skb_drop_list(&skb_shinfo(skb)->frag_list);
329 static void skb_clone_fraglist(struct sk_buff *skb)
331 struct sk_buff *list;
333 for (list = skb_shinfo(skb)->frag_list; list; list = list->next)
337 static void skb_release_data(struct sk_buff *skb)
340 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
341 &skb_shinfo(skb)->dataref)) {
342 if (skb_shinfo(skb)->nr_frags) {
344 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
345 put_page(skb_shinfo(skb)->frags[i].page);
348 if (skb_shinfo(skb)->frag_list)
349 skb_drop_fraglist(skb);
356 * Free an skbuff by memory without cleaning the state.
358 static void kfree_skbmem(struct sk_buff *skb)
360 struct sk_buff *other;
361 atomic_t *fclone_ref;
363 switch (skb->fclone) {
364 case SKB_FCLONE_UNAVAILABLE:
365 kmem_cache_free(skbuff_head_cache, skb);
368 case SKB_FCLONE_ORIG:
369 fclone_ref = (atomic_t *) (skb + 2);
370 if (atomic_dec_and_test(fclone_ref))
371 kmem_cache_free(skbuff_fclone_cache, skb);
374 case SKB_FCLONE_CLONE:
375 fclone_ref = (atomic_t *) (skb + 1);
378 /* The clone portion is available for
379 * fast-cloning again.
381 skb->fclone = SKB_FCLONE_UNAVAILABLE;
383 if (atomic_dec_and_test(fclone_ref))
384 kmem_cache_free(skbuff_fclone_cache, other);
389 static void skb_release_head_state(struct sk_buff *skb)
391 dst_release(skb->dst);
393 secpath_put(skb->sp);
395 if (skb->destructor) {
397 skb->destructor(skb);
399 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
400 nf_conntrack_put(skb->nfct);
401 nf_conntrack_put_reasm(skb->nfct_reasm);
403 #ifdef CONFIG_BRIDGE_NETFILTER
404 nf_bridge_put(skb->nf_bridge);
406 /* XXX: IS this still necessary? - JHS */
407 #ifdef CONFIG_NET_SCHED
409 #ifdef CONFIG_NET_CLS_ACT
415 /* Free everything but the sk_buff shell. */
416 static void skb_release_all(struct sk_buff *skb)
418 skb_release_head_state(skb);
419 skb_release_data(skb);
423 * __kfree_skb - private function
426 * Free an sk_buff. Release anything attached to the buffer.
427 * Clean the state. This is an internal helper function. Users should
428 * always call kfree_skb
431 void __kfree_skb(struct sk_buff *skb)
433 skb_release_all(skb);
436 EXPORT_SYMBOL(__kfree_skb);
439 * kfree_skb - free an sk_buff
440 * @skb: buffer to free
442 * Drop a reference to the buffer and free it if the usage count has
445 void kfree_skb(struct sk_buff *skb)
449 if (likely(atomic_read(&skb->users) == 1))
451 else if (likely(!atomic_dec_and_test(&skb->users)))
455 EXPORT_SYMBOL(kfree_skb);
458 * skb_recycle_check - check if skb can be reused for receive
460 * @skb_size: minimum receive buffer size
462 * Checks that the skb passed in is not shared or cloned, and
463 * that it is linear and its head portion at least as large as
464 * skb_size so that it can be recycled as a receive buffer.
465 * If these conditions are met, this function does any necessary
466 * reference count dropping and cleans up the skbuff as if it
467 * just came from __alloc_skb().
469 int skb_recycle_check(struct sk_buff *skb, int skb_size)
471 struct skb_shared_info *shinfo;
473 if (skb_is_nonlinear(skb) || skb->fclone != SKB_FCLONE_UNAVAILABLE)
476 skb_size = SKB_DATA_ALIGN(skb_size + NET_SKB_PAD);
477 if (skb_end_pointer(skb) - skb->head < skb_size)
480 if (skb_shared(skb) || skb_cloned(skb))
483 skb_release_head_state(skb);
484 shinfo = skb_shinfo(skb);
485 atomic_set(&shinfo->dataref, 1);
486 shinfo->nr_frags = 0;
487 shinfo->gso_size = 0;
488 shinfo->gso_segs = 0;
489 shinfo->gso_type = 0;
490 shinfo->ip6_frag_id = 0;
491 shinfo->frag_list = NULL;
493 memset(skb, 0, offsetof(struct sk_buff, tail));
494 skb->data = skb->head + NET_SKB_PAD;
495 skb_reset_tail_pointer(skb);
499 EXPORT_SYMBOL(skb_recycle_check);
501 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
503 new->tstamp = old->tstamp;
505 new->transport_header = old->transport_header;
506 new->network_header = old->network_header;
507 new->mac_header = old->mac_header;
508 new->dst = dst_clone(old->dst);
510 new->sp = secpath_get(old->sp);
512 memcpy(new->cb, old->cb, sizeof(old->cb));
513 new->csum_start = old->csum_start;
514 new->csum_offset = old->csum_offset;
515 new->local_df = old->local_df;
516 new->pkt_type = old->pkt_type;
517 new->ip_summed = old->ip_summed;
518 skb_copy_queue_mapping(new, old);
519 new->priority = old->priority;
520 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
521 new->ipvs_property = old->ipvs_property;
523 new->protocol = old->protocol;
524 new->mark = old->mark;
526 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
527 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
528 new->nf_trace = old->nf_trace;
530 #ifdef CONFIG_NET_SCHED
531 new->tc_index = old->tc_index;
532 #ifdef CONFIG_NET_CLS_ACT
533 new->tc_verd = old->tc_verd;
536 new->vlan_tci = old->vlan_tci;
538 skb_copy_secmark(new, old);
541 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
543 #define C(x) n->x = skb->x
545 n->next = n->prev = NULL;
547 __copy_skb_header(n, skb);
552 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
555 n->destructor = NULL;
562 #if defined(CONFIG_MAC80211) || defined(CONFIG_MAC80211_MODULE)
566 atomic_set(&n->users, 1);
568 atomic_inc(&(skb_shinfo(skb)->dataref));
576 * skb_morph - morph one skb into another
577 * @dst: the skb to receive the contents
578 * @src: the skb to supply the contents
580 * This is identical to skb_clone except that the target skb is
581 * supplied by the user.
583 * The target skb is returned upon exit.
585 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
587 skb_release_all(dst);
588 return __skb_clone(dst, src);
590 EXPORT_SYMBOL_GPL(skb_morph);
593 * skb_clone - duplicate an sk_buff
594 * @skb: buffer to clone
595 * @gfp_mask: allocation priority
597 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
598 * copies share the same packet data but not structure. The new
599 * buffer has a reference count of 1. If the allocation fails the
600 * function returns %NULL otherwise the new buffer is returned.
602 * If this function is called from an interrupt gfp_mask() must be
606 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
611 if (skb->fclone == SKB_FCLONE_ORIG &&
612 n->fclone == SKB_FCLONE_UNAVAILABLE) {
613 atomic_t *fclone_ref = (atomic_t *) (n + 1);
614 n->fclone = SKB_FCLONE_CLONE;
615 atomic_inc(fclone_ref);
617 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
620 n->fclone = SKB_FCLONE_UNAVAILABLE;
623 return __skb_clone(n, skb);
625 EXPORT_SYMBOL(skb_clone);
627 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
629 #ifndef NET_SKBUFF_DATA_USES_OFFSET
631 * Shift between the two data areas in bytes
633 unsigned long offset = new->data - old->data;
636 __copy_skb_header(new, old);
638 #ifndef NET_SKBUFF_DATA_USES_OFFSET
639 /* {transport,network,mac}_header are relative to skb->head */
640 new->transport_header += offset;
641 new->network_header += offset;
642 new->mac_header += offset;
644 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
645 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
646 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
650 * skb_copy - create private copy of an sk_buff
651 * @skb: buffer to copy
652 * @gfp_mask: allocation priority
654 * Make a copy of both an &sk_buff and its data. This is used when the
655 * caller wishes to modify the data and needs a private copy of the
656 * data to alter. Returns %NULL on failure or the pointer to the buffer
657 * on success. The returned buffer has a reference count of 1.
659 * As by-product this function converts non-linear &sk_buff to linear
660 * one, so that &sk_buff becomes completely private and caller is allowed
661 * to modify all the data of returned buffer. This means that this
662 * function is not recommended for use in circumstances when only
663 * header is going to be modified. Use pskb_copy() instead.
666 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
668 int headerlen = skb->data - skb->head;
670 * Allocate the copy buffer
673 #ifdef NET_SKBUFF_DATA_USES_OFFSET
674 n = alloc_skb(skb->end + skb->data_len, gfp_mask);
676 n = alloc_skb(skb->end - skb->head + skb->data_len, gfp_mask);
681 /* Set the data pointer */
682 skb_reserve(n, headerlen);
683 /* Set the tail pointer and length */
684 skb_put(n, skb->len);
686 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
689 copy_skb_header(n, skb);
692 EXPORT_SYMBOL(skb_copy);
695 * pskb_copy - create copy of an sk_buff with private head.
696 * @skb: buffer to copy
697 * @gfp_mask: allocation priority
699 * Make a copy of both an &sk_buff and part of its data, located
700 * in header. Fragmented data remain shared. This is used when
701 * the caller wishes to modify only header of &sk_buff and needs
702 * private copy of the header to alter. Returns %NULL on failure
703 * or the pointer to the buffer on success.
704 * The returned buffer has a reference count of 1.
707 struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
710 * Allocate the copy buffer
713 #ifdef NET_SKBUFF_DATA_USES_OFFSET
714 n = alloc_skb(skb->end, gfp_mask);
716 n = alloc_skb(skb->end - skb->head, gfp_mask);
721 /* Set the data pointer */
722 skb_reserve(n, skb->data - skb->head);
723 /* Set the tail pointer and length */
724 skb_put(n, skb_headlen(skb));
726 skb_copy_from_linear_data(skb, n->data, n->len);
728 n->truesize += skb->data_len;
729 n->data_len = skb->data_len;
732 if (skb_shinfo(skb)->nr_frags) {
735 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
736 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
737 get_page(skb_shinfo(n)->frags[i].page);
739 skb_shinfo(n)->nr_frags = i;
742 if (skb_shinfo(skb)->frag_list) {
743 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
744 skb_clone_fraglist(n);
747 copy_skb_header(n, skb);
751 EXPORT_SYMBOL(pskb_copy);
754 * pskb_expand_head - reallocate header of &sk_buff
755 * @skb: buffer to reallocate
756 * @nhead: room to add at head
757 * @ntail: room to add at tail
758 * @gfp_mask: allocation priority
760 * Expands (or creates identical copy, if &nhead and &ntail are zero)
761 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
762 * reference count of 1. Returns zero in the case of success or error,
763 * if expansion failed. In the last case, &sk_buff is not changed.
765 * All the pointers pointing into skb header may change and must be
766 * reloaded after call to this function.
769 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
774 #ifdef NET_SKBUFF_DATA_USES_OFFSET
775 int size = nhead + skb->end + ntail;
777 int size = nhead + (skb->end - skb->head) + ntail;
786 size = SKB_DATA_ALIGN(size);
788 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
792 /* Copy only real data... and, alas, header. This should be
793 * optimized for the cases when header is void. */
794 #ifdef NET_SKBUFF_DATA_USES_OFFSET
795 memcpy(data + nhead, skb->head, skb->tail);
797 memcpy(data + nhead, skb->head, skb->tail - skb->head);
799 memcpy(data + size, skb_end_pointer(skb),
800 sizeof(struct skb_shared_info));
802 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
803 get_page(skb_shinfo(skb)->frags[i].page);
805 if (skb_shinfo(skb)->frag_list)
806 skb_clone_fraglist(skb);
808 skb_release_data(skb);
810 off = (data + nhead) - skb->head;
814 #ifdef NET_SKBUFF_DATA_USES_OFFSET
818 skb->end = skb->head + size;
820 /* {transport,network,mac}_header and tail are relative to skb->head */
822 skb->transport_header += off;
823 skb->network_header += off;
824 skb->mac_header += off;
825 skb->csum_start += nhead;
829 atomic_set(&skb_shinfo(skb)->dataref, 1);
835 EXPORT_SYMBOL(pskb_expand_head);
837 /* Make private copy of skb with writable head and some headroom */
839 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
841 struct sk_buff *skb2;
842 int delta = headroom - skb_headroom(skb);
845 skb2 = pskb_copy(skb, GFP_ATOMIC);
847 skb2 = skb_clone(skb, GFP_ATOMIC);
848 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
856 EXPORT_SYMBOL(skb_realloc_headroom);
859 * skb_copy_expand - copy and expand sk_buff
860 * @skb: buffer to copy
861 * @newheadroom: new free bytes at head
862 * @newtailroom: new free bytes at tail
863 * @gfp_mask: allocation priority
865 * Make a copy of both an &sk_buff and its data and while doing so
866 * allocate additional space.
868 * This is used when the caller wishes to modify the data and needs a
869 * private copy of the data to alter as well as more space for new fields.
870 * Returns %NULL on failure or the pointer to the buffer
871 * on success. The returned buffer has a reference count of 1.
873 * You must pass %GFP_ATOMIC as the allocation priority if this function
874 * is called from an interrupt.
876 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
877 int newheadroom, int newtailroom,
881 * Allocate the copy buffer
883 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
885 int oldheadroom = skb_headroom(skb);
886 int head_copy_len, head_copy_off;
892 skb_reserve(n, newheadroom);
894 /* Set the tail pointer and length */
895 skb_put(n, skb->len);
897 head_copy_len = oldheadroom;
899 if (newheadroom <= head_copy_len)
900 head_copy_len = newheadroom;
902 head_copy_off = newheadroom - head_copy_len;
904 /* Copy the linear header and data. */
905 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
906 skb->len + head_copy_len))
909 copy_skb_header(n, skb);
911 off = newheadroom - oldheadroom;
912 n->csum_start += off;
913 #ifdef NET_SKBUFF_DATA_USES_OFFSET
914 n->transport_header += off;
915 n->network_header += off;
916 n->mac_header += off;
921 EXPORT_SYMBOL(skb_copy_expand);
924 * skb_pad - zero pad the tail of an skb
925 * @skb: buffer to pad
928 * Ensure that a buffer is followed by a padding area that is zero
929 * filled. Used by network drivers which may DMA or transfer data
930 * beyond the buffer end onto the wire.
932 * May return error in out of memory cases. The skb is freed on error.
935 int skb_pad(struct sk_buff *skb, int pad)
940 /* If the skbuff is non linear tailroom is always zero.. */
941 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
942 memset(skb->data+skb->len, 0, pad);
946 ntail = skb->data_len + pad - (skb->end - skb->tail);
947 if (likely(skb_cloned(skb) || ntail > 0)) {
948 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
953 /* FIXME: The use of this function with non-linear skb's really needs
956 err = skb_linearize(skb);
960 memset(skb->data + skb->len, 0, pad);
967 EXPORT_SYMBOL(skb_pad);
970 * skb_put - add data to a buffer
971 * @skb: buffer to use
972 * @len: amount of data to add
974 * This function extends the used data area of the buffer. If this would
975 * exceed the total buffer size the kernel will panic. A pointer to the
976 * first byte of the extra data is returned.
978 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
980 unsigned char *tmp = skb_tail_pointer(skb);
981 SKB_LINEAR_ASSERT(skb);
984 if (unlikely(skb->tail > skb->end))
985 skb_over_panic(skb, len, __builtin_return_address(0));
988 EXPORT_SYMBOL(skb_put);
991 * skb_push - add data to the start of a buffer
992 * @skb: buffer to use
993 * @len: amount of data to add
995 * This function extends the used data area of the buffer at the buffer
996 * start. If this would exceed the total buffer headroom the kernel will
997 * panic. A pointer to the first byte of the extra data is returned.
999 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1003 if (unlikely(skb->data<skb->head))
1004 skb_under_panic(skb, len, __builtin_return_address(0));
1007 EXPORT_SYMBOL(skb_push);
1010 * skb_pull - remove data from the start of a buffer
1011 * @skb: buffer to use
1012 * @len: amount of data to remove
1014 * This function removes data from the start of a buffer, returning
1015 * the memory to the headroom. A pointer to the next data in the buffer
1016 * is returned. Once the data has been pulled future pushes will overwrite
1019 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1021 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1023 EXPORT_SYMBOL(skb_pull);
1026 * skb_trim - remove end from a buffer
1027 * @skb: buffer to alter
1030 * Cut the length of a buffer down by removing data from the tail. If
1031 * the buffer is already under the length specified it is not modified.
1032 * The skb must be linear.
1034 void skb_trim(struct sk_buff *skb, unsigned int len)
1037 __skb_trim(skb, len);
1039 EXPORT_SYMBOL(skb_trim);
1041 /* Trims skb to length len. It can change skb pointers.
1044 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1046 struct sk_buff **fragp;
1047 struct sk_buff *frag;
1048 int offset = skb_headlen(skb);
1049 int nfrags = skb_shinfo(skb)->nr_frags;
1053 if (skb_cloned(skb) &&
1054 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1061 for (; i < nfrags; i++) {
1062 int end = offset + skb_shinfo(skb)->frags[i].size;
1069 skb_shinfo(skb)->frags[i++].size = len - offset;
1072 skb_shinfo(skb)->nr_frags = i;
1074 for (; i < nfrags; i++)
1075 put_page(skb_shinfo(skb)->frags[i].page);
1077 if (skb_shinfo(skb)->frag_list)
1078 skb_drop_fraglist(skb);
1082 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1083 fragp = &frag->next) {
1084 int end = offset + frag->len;
1086 if (skb_shared(frag)) {
1087 struct sk_buff *nfrag;
1089 nfrag = skb_clone(frag, GFP_ATOMIC);
1090 if (unlikely(!nfrag))
1093 nfrag->next = frag->next;
1105 unlikely((err = pskb_trim(frag, len - offset))))
1109 skb_drop_list(&frag->next);
1114 if (len > skb_headlen(skb)) {
1115 skb->data_len -= skb->len - len;
1120 skb_set_tail_pointer(skb, len);
1125 EXPORT_SYMBOL(___pskb_trim);
1128 * __pskb_pull_tail - advance tail of skb header
1129 * @skb: buffer to reallocate
1130 * @delta: number of bytes to advance tail
1132 * The function makes a sense only on a fragmented &sk_buff,
1133 * it expands header moving its tail forward and copying necessary
1134 * data from fragmented part.
1136 * &sk_buff MUST have reference count of 1.
1138 * Returns %NULL (and &sk_buff does not change) if pull failed
1139 * or value of new tail of skb in the case of success.
1141 * All the pointers pointing into skb header may change and must be
1142 * reloaded after call to this function.
1145 /* Moves tail of skb head forward, copying data from fragmented part,
1146 * when it is necessary.
1147 * 1. It may fail due to malloc failure.
1148 * 2. It may change skb pointers.
1150 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1152 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1154 /* If skb has not enough free space at tail, get new one
1155 * plus 128 bytes for future expansions. If we have enough
1156 * room at tail, reallocate without expansion only if skb is cloned.
1158 int i, k, eat = (skb->tail + delta) - skb->end;
1160 if (eat > 0 || skb_cloned(skb)) {
1161 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1166 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1169 /* Optimization: no fragments, no reasons to preestimate
1170 * size of pulled pages. Superb.
1172 if (!skb_shinfo(skb)->frag_list)
1175 /* Estimate size of pulled pages. */
1177 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1178 if (skb_shinfo(skb)->frags[i].size >= eat)
1180 eat -= skb_shinfo(skb)->frags[i].size;
1183 /* If we need update frag list, we are in troubles.
1184 * Certainly, it possible to add an offset to skb data,
1185 * but taking into account that pulling is expected to
1186 * be very rare operation, it is worth to fight against
1187 * further bloating skb head and crucify ourselves here instead.
1188 * Pure masohism, indeed. 8)8)
1191 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1192 struct sk_buff *clone = NULL;
1193 struct sk_buff *insp = NULL;
1198 if (list->len <= eat) {
1199 /* Eaten as whole. */
1204 /* Eaten partially. */
1206 if (skb_shared(list)) {
1207 /* Sucks! We need to fork list. :-( */
1208 clone = skb_clone(list, GFP_ATOMIC);
1214 /* This may be pulled without
1218 if (!pskb_pull(list, eat)) {
1227 /* Free pulled out fragments. */
1228 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1229 skb_shinfo(skb)->frag_list = list->next;
1232 /* And insert new clone at head. */
1235 skb_shinfo(skb)->frag_list = clone;
1238 /* Success! Now we may commit changes to skb data. */
1243 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1244 if (skb_shinfo(skb)->frags[i].size <= eat) {
1245 put_page(skb_shinfo(skb)->frags[i].page);
1246 eat -= skb_shinfo(skb)->frags[i].size;
1248 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1250 skb_shinfo(skb)->frags[k].page_offset += eat;
1251 skb_shinfo(skb)->frags[k].size -= eat;
1257 skb_shinfo(skb)->nr_frags = k;
1260 skb->data_len -= delta;
1262 return skb_tail_pointer(skb);
1264 EXPORT_SYMBOL(__pskb_pull_tail);
1266 /* Copy some data bits from skb to kernel buffer. */
1268 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1271 int start = skb_headlen(skb);
1273 if (offset > (int)skb->len - len)
1277 if ((copy = start - offset) > 0) {
1280 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1281 if ((len -= copy) == 0)
1287 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1290 WARN_ON(start > offset + len);
1292 end = start + skb_shinfo(skb)->frags[i].size;
1293 if ((copy = end - offset) > 0) {
1299 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1301 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1302 offset - start, copy);
1303 kunmap_skb_frag(vaddr);
1305 if ((len -= copy) == 0)
1313 if (skb_shinfo(skb)->frag_list) {
1314 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1316 for (; list; list = list->next) {
1319 WARN_ON(start > offset + len);
1321 end = start + list->len;
1322 if ((copy = end - offset) > 0) {
1325 if (skb_copy_bits(list, offset - start,
1328 if ((len -= copy) == 0)
1342 EXPORT_SYMBOL(skb_copy_bits);
1345 * Callback from splice_to_pipe(), if we need to release some pages
1346 * at the end of the spd in case we error'ed out in filling the pipe.
1348 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1350 put_page(spd->pages[i]);
1353 static inline struct page *linear_to_page(struct page *page, unsigned int *len,
1354 unsigned int *offset,
1355 struct sk_buff *skb)
1357 struct sock *sk = skb->sk;
1358 struct page *p = sk->sk_sndmsg_page;
1363 p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1367 off = sk->sk_sndmsg_off = 0;
1368 /* hold one ref to this page until it's full */
1372 off = sk->sk_sndmsg_off;
1373 mlen = PAGE_SIZE - off;
1374 if (mlen < 64 && mlen < *len) {
1379 *len = min_t(unsigned int, *len, mlen);
1382 memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1383 sk->sk_sndmsg_off += *len;
1391 * Fill page/offset/length into spd, if it can hold more pages.
1393 static inline int spd_fill_page(struct splice_pipe_desc *spd, struct page *page,
1394 unsigned int *len, unsigned int offset,
1395 struct sk_buff *skb, int linear)
1397 if (unlikely(spd->nr_pages == PIPE_BUFFERS))
1401 page = linear_to_page(page, len, &offset, skb);
1407 spd->pages[spd->nr_pages] = page;
1408 spd->partial[spd->nr_pages].len = *len;
1409 spd->partial[spd->nr_pages].offset = offset;
1415 static inline void __segment_seek(struct page **page, unsigned int *poff,
1416 unsigned int *plen, unsigned int off)
1421 n = *poff / PAGE_SIZE;
1423 *page = nth_page(*page, n);
1425 *poff = *poff % PAGE_SIZE;
1429 static inline int __splice_segment(struct page *page, unsigned int poff,
1430 unsigned int plen, unsigned int *off,
1431 unsigned int *len, struct sk_buff *skb,
1432 struct splice_pipe_desc *spd, int linear)
1437 /* skip this segment if already processed */
1443 /* ignore any bits we already processed */
1445 __segment_seek(&page, &poff, &plen, *off);
1450 unsigned int flen = min(*len, plen);
1452 /* the linear region may spread across several pages */
1453 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1455 if (spd_fill_page(spd, page, &flen, poff, skb, linear))
1458 __segment_seek(&page, &poff, &plen, flen);
1461 } while (*len && plen);
1467 * Map linear and fragment data from the skb to spd. It reports failure if the
1468 * pipe is full or if we already spliced the requested length.
1470 static int __skb_splice_bits(struct sk_buff *skb, unsigned int *offset,
1472 struct splice_pipe_desc *spd)
1477 * map the linear part
1479 if (__splice_segment(virt_to_page(skb->data),
1480 (unsigned long) skb->data & (PAGE_SIZE - 1),
1482 offset, len, skb, spd, 1))
1486 * then map the fragments
1488 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1489 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1491 if (__splice_segment(f->page, f->page_offset, f->size,
1492 offset, len, skb, spd, 0))
1500 * Map data from the skb to a pipe. Should handle both the linear part,
1501 * the fragments, and the frag list. It does NOT handle frag lists within
1502 * the frag list, if such a thing exists. We'd probably need to recurse to
1503 * handle that cleanly.
1505 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1506 struct pipe_inode_info *pipe, unsigned int tlen,
1509 struct partial_page partial[PIPE_BUFFERS];
1510 struct page *pages[PIPE_BUFFERS];
1511 struct splice_pipe_desc spd = {
1515 .ops = &sock_pipe_buf_ops,
1516 .spd_release = sock_spd_release,
1520 * __skb_splice_bits() only fails if the output has no room left,
1521 * so no point in going over the frag_list for the error case.
1523 if (__skb_splice_bits(skb, &offset, &tlen, &spd))
1529 * now see if we have a frag_list to map
1531 if (skb_shinfo(skb)->frag_list) {
1532 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1534 for (; list && tlen; list = list->next) {
1535 if (__skb_splice_bits(list, &offset, &tlen, &spd))
1542 struct sock *sk = skb->sk;
1546 * Drop the socket lock, otherwise we have reverse
1547 * locking dependencies between sk_lock and i_mutex
1548 * here as compared to sendfile(). We enter here
1549 * with the socket lock held, and splice_to_pipe() will
1550 * grab the pipe inode lock. For sendfile() emulation,
1551 * we call into ->sendpage() with the i_mutex lock held
1552 * and networking will grab the socket lock.
1555 ret = splice_to_pipe(pipe, &spd);
1564 * skb_store_bits - store bits from kernel buffer to skb
1565 * @skb: destination buffer
1566 * @offset: offset in destination
1567 * @from: source buffer
1568 * @len: number of bytes to copy
1570 * Copy the specified number of bytes from the source buffer to the
1571 * destination skb. This function handles all the messy bits of
1572 * traversing fragment lists and such.
1575 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1578 int start = skb_headlen(skb);
1580 if (offset > (int)skb->len - len)
1583 if ((copy = start - offset) > 0) {
1586 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1587 if ((len -= copy) == 0)
1593 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1594 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1597 WARN_ON(start > offset + len);
1599 end = start + frag->size;
1600 if ((copy = end - offset) > 0) {
1606 vaddr = kmap_skb_frag(frag);
1607 memcpy(vaddr + frag->page_offset + offset - start,
1609 kunmap_skb_frag(vaddr);
1611 if ((len -= copy) == 0)
1619 if (skb_shinfo(skb)->frag_list) {
1620 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1622 for (; list; list = list->next) {
1625 WARN_ON(start > offset + len);
1627 end = start + list->len;
1628 if ((copy = end - offset) > 0) {
1631 if (skb_store_bits(list, offset - start,
1634 if ((len -= copy) == 0)
1648 EXPORT_SYMBOL(skb_store_bits);
1650 /* Checksum skb data. */
1652 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1653 int len, __wsum csum)
1655 int start = skb_headlen(skb);
1656 int i, copy = start - offset;
1659 /* Checksum header. */
1663 csum = csum_partial(skb->data + offset, copy, csum);
1664 if ((len -= copy) == 0)
1670 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1673 WARN_ON(start > offset + len);
1675 end = start + skb_shinfo(skb)->frags[i].size;
1676 if ((copy = end - offset) > 0) {
1679 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1683 vaddr = kmap_skb_frag(frag);
1684 csum2 = csum_partial(vaddr + frag->page_offset +
1685 offset - start, copy, 0);
1686 kunmap_skb_frag(vaddr);
1687 csum = csum_block_add(csum, csum2, pos);
1696 if (skb_shinfo(skb)->frag_list) {
1697 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1699 for (; list; list = list->next) {
1702 WARN_ON(start > offset + len);
1704 end = start + list->len;
1705 if ((copy = end - offset) > 0) {
1709 csum2 = skb_checksum(list, offset - start,
1711 csum = csum_block_add(csum, csum2, pos);
1712 if ((len -= copy) == 0)
1724 EXPORT_SYMBOL(skb_checksum);
1726 /* Both of above in one bottle. */
1728 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1729 u8 *to, int len, __wsum csum)
1731 int start = skb_headlen(skb);
1732 int i, copy = start - offset;
1739 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1741 if ((len -= copy) == 0)
1748 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1751 WARN_ON(start > offset + len);
1753 end = start + skb_shinfo(skb)->frags[i].size;
1754 if ((copy = end - offset) > 0) {
1757 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1761 vaddr = kmap_skb_frag(frag);
1762 csum2 = csum_partial_copy_nocheck(vaddr +
1766 kunmap_skb_frag(vaddr);
1767 csum = csum_block_add(csum, csum2, pos);
1777 if (skb_shinfo(skb)->frag_list) {
1778 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1780 for (; list; list = list->next) {
1784 WARN_ON(start > offset + len);
1786 end = start + list->len;
1787 if ((copy = end - offset) > 0) {
1790 csum2 = skb_copy_and_csum_bits(list,
1793 csum = csum_block_add(csum, csum2, pos);
1794 if ((len -= copy) == 0)
1806 EXPORT_SYMBOL(skb_copy_and_csum_bits);
1808 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1813 if (skb->ip_summed == CHECKSUM_PARTIAL)
1814 csstart = skb->csum_start - skb_headroom(skb);
1816 csstart = skb_headlen(skb);
1818 BUG_ON(csstart > skb_headlen(skb));
1820 skb_copy_from_linear_data(skb, to, csstart);
1823 if (csstart != skb->len)
1824 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1825 skb->len - csstart, 0);
1827 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1828 long csstuff = csstart + skb->csum_offset;
1830 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
1833 EXPORT_SYMBOL(skb_copy_and_csum_dev);
1836 * skb_dequeue - remove from the head of the queue
1837 * @list: list to dequeue from
1839 * Remove the head of the list. The list lock is taken so the function
1840 * may be used safely with other locking list functions. The head item is
1841 * returned or %NULL if the list is empty.
1844 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1846 unsigned long flags;
1847 struct sk_buff *result;
1849 spin_lock_irqsave(&list->lock, flags);
1850 result = __skb_dequeue(list);
1851 spin_unlock_irqrestore(&list->lock, flags);
1854 EXPORT_SYMBOL(skb_dequeue);
1857 * skb_dequeue_tail - remove from the tail of the queue
1858 * @list: list to dequeue from
1860 * Remove the tail of the list. The list lock is taken so the function
1861 * may be used safely with other locking list functions. The tail item is
1862 * returned or %NULL if the list is empty.
1864 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1866 unsigned long flags;
1867 struct sk_buff *result;
1869 spin_lock_irqsave(&list->lock, flags);
1870 result = __skb_dequeue_tail(list);
1871 spin_unlock_irqrestore(&list->lock, flags);
1874 EXPORT_SYMBOL(skb_dequeue_tail);
1877 * skb_queue_purge - empty a list
1878 * @list: list to empty
1880 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1881 * the list and one reference dropped. This function takes the list
1882 * lock and is atomic with respect to other list locking functions.
1884 void skb_queue_purge(struct sk_buff_head *list)
1886 struct sk_buff *skb;
1887 while ((skb = skb_dequeue(list)) != NULL)
1890 EXPORT_SYMBOL(skb_queue_purge);
1893 * skb_queue_head - queue a buffer at the list head
1894 * @list: list to use
1895 * @newsk: buffer to queue
1897 * Queue a buffer at the start of the list. This function takes the
1898 * list lock and can be used safely with other locking &sk_buff functions
1901 * A buffer cannot be placed on two lists at the same time.
1903 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1905 unsigned long flags;
1907 spin_lock_irqsave(&list->lock, flags);
1908 __skb_queue_head(list, newsk);
1909 spin_unlock_irqrestore(&list->lock, flags);
1911 EXPORT_SYMBOL(skb_queue_head);
1914 * skb_queue_tail - queue a buffer at the list tail
1915 * @list: list to use
1916 * @newsk: buffer to queue
1918 * Queue a buffer at the tail of the list. This function takes the
1919 * list lock and can be used safely with other locking &sk_buff functions
1922 * A buffer cannot be placed on two lists at the same time.
1924 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1926 unsigned long flags;
1928 spin_lock_irqsave(&list->lock, flags);
1929 __skb_queue_tail(list, newsk);
1930 spin_unlock_irqrestore(&list->lock, flags);
1932 EXPORT_SYMBOL(skb_queue_tail);
1935 * skb_unlink - remove a buffer from a list
1936 * @skb: buffer to remove
1937 * @list: list to use
1939 * Remove a packet from a list. The list locks are taken and this
1940 * function is atomic with respect to other list locked calls
1942 * You must know what list the SKB is on.
1944 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1946 unsigned long flags;
1948 spin_lock_irqsave(&list->lock, flags);
1949 __skb_unlink(skb, list);
1950 spin_unlock_irqrestore(&list->lock, flags);
1952 EXPORT_SYMBOL(skb_unlink);
1955 * skb_append - append a buffer
1956 * @old: buffer to insert after
1957 * @newsk: buffer to insert
1958 * @list: list to use
1960 * Place a packet after a given packet in a list. The list locks are taken
1961 * and this function is atomic with respect to other list locked calls.
1962 * A buffer cannot be placed on two lists at the same time.
1964 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1966 unsigned long flags;
1968 spin_lock_irqsave(&list->lock, flags);
1969 __skb_queue_after(list, old, newsk);
1970 spin_unlock_irqrestore(&list->lock, flags);
1972 EXPORT_SYMBOL(skb_append);
1975 * skb_insert - insert a buffer
1976 * @old: buffer to insert before
1977 * @newsk: buffer to insert
1978 * @list: list to use
1980 * Place a packet before a given packet in a list. The list locks are
1981 * taken and this function is atomic with respect to other list locked
1984 * A buffer cannot be placed on two lists at the same time.
1986 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1988 unsigned long flags;
1990 spin_lock_irqsave(&list->lock, flags);
1991 __skb_insert(newsk, old->prev, old, list);
1992 spin_unlock_irqrestore(&list->lock, flags);
1994 EXPORT_SYMBOL(skb_insert);
1996 static inline void skb_split_inside_header(struct sk_buff *skb,
1997 struct sk_buff* skb1,
1998 const u32 len, const int pos)
2002 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2004 /* And move data appendix as is. */
2005 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2006 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2008 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2009 skb_shinfo(skb)->nr_frags = 0;
2010 skb1->data_len = skb->data_len;
2011 skb1->len += skb1->data_len;
2014 skb_set_tail_pointer(skb, len);
2017 static inline void skb_split_no_header(struct sk_buff *skb,
2018 struct sk_buff* skb1,
2019 const u32 len, int pos)
2022 const int nfrags = skb_shinfo(skb)->nr_frags;
2024 skb_shinfo(skb)->nr_frags = 0;
2025 skb1->len = skb1->data_len = skb->len - len;
2027 skb->data_len = len - pos;
2029 for (i = 0; i < nfrags; i++) {
2030 int size = skb_shinfo(skb)->frags[i].size;
2032 if (pos + size > len) {
2033 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2037 * We have two variants in this case:
2038 * 1. Move all the frag to the second
2039 * part, if it is possible. F.e.
2040 * this approach is mandatory for TUX,
2041 * where splitting is expensive.
2042 * 2. Split is accurately. We make this.
2044 get_page(skb_shinfo(skb)->frags[i].page);
2045 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2046 skb_shinfo(skb1)->frags[0].size -= len - pos;
2047 skb_shinfo(skb)->frags[i].size = len - pos;
2048 skb_shinfo(skb)->nr_frags++;
2052 skb_shinfo(skb)->nr_frags++;
2055 skb_shinfo(skb1)->nr_frags = k;
2059 * skb_split - Split fragmented skb to two parts at length len.
2060 * @skb: the buffer to split
2061 * @skb1: the buffer to receive the second part
2062 * @len: new length for skb
2064 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2066 int pos = skb_headlen(skb);
2068 if (len < pos) /* Split line is inside header. */
2069 skb_split_inside_header(skb, skb1, len, pos);
2070 else /* Second chunk has no header, nothing to copy. */
2071 skb_split_no_header(skb, skb1, len, pos);
2073 EXPORT_SYMBOL(skb_split);
2075 /* Shifting from/to a cloned skb is a no-go.
2077 * Caller cannot keep skb_shinfo related pointers past calling here!
2079 static int skb_prepare_for_shift(struct sk_buff *skb)
2081 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2085 * skb_shift - Shifts paged data partially from skb to another
2086 * @tgt: buffer into which tail data gets added
2087 * @skb: buffer from which the paged data comes from
2088 * @shiftlen: shift up to this many bytes
2090 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2091 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2092 * It's up to caller to free skb if everything was shifted.
2094 * If @tgt runs out of frags, the whole operation is aborted.
2096 * Skb cannot include anything else but paged data while tgt is allowed
2097 * to have non-paged data as well.
2099 * TODO: full sized shift could be optimized but that would need
2100 * specialized skb free'er to handle frags without up-to-date nr_frags.
2102 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2104 int from, to, merge, todo;
2105 struct skb_frag_struct *fragfrom, *fragto;
2107 BUG_ON(shiftlen > skb->len);
2108 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2112 to = skb_shinfo(tgt)->nr_frags;
2113 fragfrom = &skb_shinfo(skb)->frags[from];
2115 /* Actual merge is delayed until the point when we know we can
2116 * commit all, so that we don't have to undo partial changes
2119 !skb_can_coalesce(tgt, to, fragfrom->page, fragfrom->page_offset)) {
2124 todo -= fragfrom->size;
2126 if (skb_prepare_for_shift(skb) ||
2127 skb_prepare_for_shift(tgt))
2130 /* All previous frag pointers might be stale! */
2131 fragfrom = &skb_shinfo(skb)->frags[from];
2132 fragto = &skb_shinfo(tgt)->frags[merge];
2134 fragto->size += shiftlen;
2135 fragfrom->size -= shiftlen;
2136 fragfrom->page_offset += shiftlen;
2144 /* Skip full, not-fitting skb to avoid expensive operations */
2145 if ((shiftlen == skb->len) &&
2146 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2149 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2152 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2153 if (to == MAX_SKB_FRAGS)
2156 fragfrom = &skb_shinfo(skb)->frags[from];
2157 fragto = &skb_shinfo(tgt)->frags[to];
2159 if (todo >= fragfrom->size) {
2160 *fragto = *fragfrom;
2161 todo -= fragfrom->size;
2166 get_page(fragfrom->page);
2167 fragto->page = fragfrom->page;
2168 fragto->page_offset = fragfrom->page_offset;
2169 fragto->size = todo;
2171 fragfrom->page_offset += todo;
2172 fragfrom->size -= todo;
2180 /* Ready to "commit" this state change to tgt */
2181 skb_shinfo(tgt)->nr_frags = to;
2184 fragfrom = &skb_shinfo(skb)->frags[0];
2185 fragto = &skb_shinfo(tgt)->frags[merge];
2187 fragto->size += fragfrom->size;
2188 put_page(fragfrom->page);
2191 /* Reposition in the original skb */
2193 while (from < skb_shinfo(skb)->nr_frags)
2194 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2195 skb_shinfo(skb)->nr_frags = to;
2197 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2200 /* Most likely the tgt won't ever need its checksum anymore, skb on
2201 * the other hand might need it if it needs to be resent
2203 tgt->ip_summed = CHECKSUM_PARTIAL;
2204 skb->ip_summed = CHECKSUM_PARTIAL;
2206 /* Yak, is it really working this way? Some helper please? */
2207 skb->len -= shiftlen;
2208 skb->data_len -= shiftlen;
2209 skb->truesize -= shiftlen;
2210 tgt->len += shiftlen;
2211 tgt->data_len += shiftlen;
2212 tgt->truesize += shiftlen;
2218 * skb_prepare_seq_read - Prepare a sequential read of skb data
2219 * @skb: the buffer to read
2220 * @from: lower offset of data to be read
2221 * @to: upper offset of data to be read
2222 * @st: state variable
2224 * Initializes the specified state variable. Must be called before
2225 * invoking skb_seq_read() for the first time.
2227 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2228 unsigned int to, struct skb_seq_state *st)
2230 st->lower_offset = from;
2231 st->upper_offset = to;
2232 st->root_skb = st->cur_skb = skb;
2233 st->frag_idx = st->stepped_offset = 0;
2234 st->frag_data = NULL;
2236 EXPORT_SYMBOL(skb_prepare_seq_read);
2239 * skb_seq_read - Sequentially read skb data
2240 * @consumed: number of bytes consumed by the caller so far
2241 * @data: destination pointer for data to be returned
2242 * @st: state variable
2244 * Reads a block of skb data at &consumed relative to the
2245 * lower offset specified to skb_prepare_seq_read(). Assigns
2246 * the head of the data block to &data and returns the length
2247 * of the block or 0 if the end of the skb data or the upper
2248 * offset has been reached.
2250 * The caller is not required to consume all of the data
2251 * returned, i.e. &consumed is typically set to the number
2252 * of bytes already consumed and the next call to
2253 * skb_seq_read() will return the remaining part of the block.
2255 * Note 1: The size of each block of data returned can be arbitary,
2256 * this limitation is the cost for zerocopy seqeuental
2257 * reads of potentially non linear data.
2259 * Note 2: Fragment lists within fragments are not implemented
2260 * at the moment, state->root_skb could be replaced with
2261 * a stack for this purpose.
2263 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2264 struct skb_seq_state *st)
2266 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2269 if (unlikely(abs_offset >= st->upper_offset))
2273 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2275 if (abs_offset < block_limit) {
2276 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2277 return block_limit - abs_offset;
2280 if (st->frag_idx == 0 && !st->frag_data)
2281 st->stepped_offset += skb_headlen(st->cur_skb);
2283 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2284 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2285 block_limit = frag->size + st->stepped_offset;
2287 if (abs_offset < block_limit) {
2289 st->frag_data = kmap_skb_frag(frag);
2291 *data = (u8 *) st->frag_data + frag->page_offset +
2292 (abs_offset - st->stepped_offset);
2294 return block_limit - abs_offset;
2297 if (st->frag_data) {
2298 kunmap_skb_frag(st->frag_data);
2299 st->frag_data = NULL;
2303 st->stepped_offset += frag->size;
2306 if (st->frag_data) {
2307 kunmap_skb_frag(st->frag_data);
2308 st->frag_data = NULL;
2311 if (st->root_skb == st->cur_skb &&
2312 skb_shinfo(st->root_skb)->frag_list) {
2313 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2316 } else if (st->cur_skb->next) {
2317 st->cur_skb = st->cur_skb->next;
2324 EXPORT_SYMBOL(skb_seq_read);
2327 * skb_abort_seq_read - Abort a sequential read of skb data
2328 * @st: state variable
2330 * Must be called if skb_seq_read() was not called until it
2333 void skb_abort_seq_read(struct skb_seq_state *st)
2336 kunmap_skb_frag(st->frag_data);
2338 EXPORT_SYMBOL(skb_abort_seq_read);
2340 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2342 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2343 struct ts_config *conf,
2344 struct ts_state *state)
2346 return skb_seq_read(offset, text, TS_SKB_CB(state));
2349 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2351 skb_abort_seq_read(TS_SKB_CB(state));
2355 * skb_find_text - Find a text pattern in skb data
2356 * @skb: the buffer to look in
2357 * @from: search offset
2359 * @config: textsearch configuration
2360 * @state: uninitialized textsearch state variable
2362 * Finds a pattern in the skb data according to the specified
2363 * textsearch configuration. Use textsearch_next() to retrieve
2364 * subsequent occurrences of the pattern. Returns the offset
2365 * to the first occurrence or UINT_MAX if no match was found.
2367 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2368 unsigned int to, struct ts_config *config,
2369 struct ts_state *state)
2373 config->get_next_block = skb_ts_get_next_block;
2374 config->finish = skb_ts_finish;
2376 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2378 ret = textsearch_find(config, state);
2379 return (ret <= to - from ? ret : UINT_MAX);
2381 EXPORT_SYMBOL(skb_find_text);
2384 * skb_append_datato_frags: - append the user data to a skb
2385 * @sk: sock structure
2386 * @skb: skb structure to be appened with user data.
2387 * @getfrag: call back function to be used for getting the user data
2388 * @from: pointer to user message iov
2389 * @length: length of the iov message
2391 * Description: This procedure append the user data in the fragment part
2392 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2394 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2395 int (*getfrag)(void *from, char *to, int offset,
2396 int len, int odd, struct sk_buff *skb),
2397 void *from, int length)
2400 skb_frag_t *frag = NULL;
2401 struct page *page = NULL;
2407 /* Return error if we don't have space for new frag */
2408 frg_cnt = skb_shinfo(skb)->nr_frags;
2409 if (frg_cnt >= MAX_SKB_FRAGS)
2412 /* allocate a new page for next frag */
2413 page = alloc_pages(sk->sk_allocation, 0);
2415 /* If alloc_page fails just return failure and caller will
2416 * free previous allocated pages by doing kfree_skb()
2421 /* initialize the next frag */
2422 sk->sk_sndmsg_page = page;
2423 sk->sk_sndmsg_off = 0;
2424 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2425 skb->truesize += PAGE_SIZE;
2426 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2428 /* get the new initialized frag */
2429 frg_cnt = skb_shinfo(skb)->nr_frags;
2430 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2432 /* copy the user data to page */
2433 left = PAGE_SIZE - frag->page_offset;
2434 copy = (length > left)? left : length;
2436 ret = getfrag(from, (page_address(frag->page) +
2437 frag->page_offset + frag->size),
2438 offset, copy, 0, skb);
2442 /* copy was successful so update the size parameters */
2443 sk->sk_sndmsg_off += copy;
2446 skb->data_len += copy;
2450 } while (length > 0);
2454 EXPORT_SYMBOL(skb_append_datato_frags);
2457 * skb_pull_rcsum - pull skb and update receive checksum
2458 * @skb: buffer to update
2459 * @len: length of data pulled
2461 * This function performs an skb_pull on the packet and updates
2462 * the CHECKSUM_COMPLETE checksum. It should be used on
2463 * receive path processing instead of skb_pull unless you know
2464 * that the checksum difference is zero (e.g., a valid IP header)
2465 * or you are setting ip_summed to CHECKSUM_NONE.
2467 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2469 BUG_ON(len > skb->len);
2471 BUG_ON(skb->len < skb->data_len);
2472 skb_postpull_rcsum(skb, skb->data, len);
2473 return skb->data += len;
2476 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2479 * skb_segment - Perform protocol segmentation on skb.
2480 * @skb: buffer to segment
2481 * @features: features for the output path (see dev->features)
2483 * This function performs segmentation on the given skb. It returns
2484 * a pointer to the first in a list of new skbs for the segments.
2485 * In case of error it returns ERR_PTR(err).
2487 struct sk_buff *skb_segment(struct sk_buff *skb, int features)
2489 struct sk_buff *segs = NULL;
2490 struct sk_buff *tail = NULL;
2491 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2492 unsigned int mss = skb_shinfo(skb)->gso_size;
2493 unsigned int doffset = skb->data - skb_mac_header(skb);
2494 unsigned int offset = doffset;
2495 unsigned int headroom;
2497 int sg = features & NETIF_F_SG;
2498 int nfrags = skb_shinfo(skb)->nr_frags;
2503 __skb_push(skb, doffset);
2504 headroom = skb_headroom(skb);
2505 pos = skb_headlen(skb);
2508 struct sk_buff *nskb;
2513 len = skb->len - offset;
2517 hsize = skb_headlen(skb) - offset;
2520 if (hsize > len || !sg)
2523 if (!hsize && i >= nfrags) {
2524 BUG_ON(fskb->len != len);
2527 nskb = skb_clone(fskb, GFP_ATOMIC);
2530 if (unlikely(!nskb))
2533 hsize = skb_end_pointer(nskb) - nskb->head;
2534 if (skb_cow_head(nskb, doffset + headroom)) {
2539 nskb->truesize += skb_end_pointer(nskb) - nskb->head -
2541 skb_release_head_state(nskb);
2542 __skb_push(nskb, doffset);
2544 nskb = alloc_skb(hsize + doffset + headroom,
2547 if (unlikely(!nskb))
2550 skb_reserve(nskb, headroom);
2551 __skb_put(nskb, doffset);
2560 __copy_skb_header(nskb, skb);
2561 nskb->mac_len = skb->mac_len;
2563 skb_reset_mac_header(nskb);
2564 skb_set_network_header(nskb, skb->mac_len);
2565 nskb->transport_header = (nskb->network_header +
2566 skb_network_header_len(skb));
2567 skb_copy_from_linear_data(skb, nskb->data, doffset);
2569 if (pos >= offset + len)
2573 nskb->ip_summed = CHECKSUM_NONE;
2574 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2580 frag = skb_shinfo(nskb)->frags;
2582 skb_copy_from_linear_data_offset(skb, offset,
2583 skb_put(nskb, hsize), hsize);
2585 while (pos < offset + len && i < nfrags) {
2586 *frag = skb_shinfo(skb)->frags[i];
2587 get_page(frag->page);
2591 frag->page_offset += offset - pos;
2592 frag->size -= offset - pos;
2595 skb_shinfo(nskb)->nr_frags++;
2597 if (pos + size <= offset + len) {
2601 frag->size -= pos + size - (offset + len);
2608 if (pos < offset + len) {
2609 struct sk_buff *fskb2 = fskb;
2611 BUG_ON(pos + fskb->len != offset + len);
2617 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2623 BUG_ON(skb_shinfo(nskb)->frag_list);
2624 skb_shinfo(nskb)->frag_list = fskb2;
2628 nskb->data_len = len - hsize;
2629 nskb->len += nskb->data_len;
2630 nskb->truesize += nskb->data_len;
2631 } while ((offset += len) < skb->len);
2636 while ((skb = segs)) {
2640 return ERR_PTR(err);
2642 EXPORT_SYMBOL_GPL(skb_segment);
2644 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2646 struct sk_buff *p = *head;
2647 struct sk_buff *nskb;
2648 unsigned int headroom;
2649 unsigned int len = skb_gro_len(skb);
2651 if (p->len + len >= 65536)
2654 if (skb_shinfo(p)->frag_list)
2656 else if (skb_headlen(skb) <= skb_gro_offset(skb)) {
2657 if (skb_shinfo(p)->nr_frags + skb_shinfo(skb)->nr_frags >
2661 skb_shinfo(skb)->frags[0].page_offset +=
2662 skb_gro_offset(skb) - skb_headlen(skb);
2663 skb_shinfo(skb)->frags[0].size -=
2664 skb_gro_offset(skb) - skb_headlen(skb);
2666 memcpy(skb_shinfo(p)->frags + skb_shinfo(p)->nr_frags,
2667 skb_shinfo(skb)->frags,
2668 skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
2670 skb_shinfo(p)->nr_frags += skb_shinfo(skb)->nr_frags;
2671 skb_shinfo(skb)->nr_frags = 0;
2673 skb->truesize -= skb->data_len;
2674 skb->len -= skb->data_len;
2677 NAPI_GRO_CB(skb)->free = 1;
2681 headroom = skb_headroom(p);
2682 nskb = netdev_alloc_skb(p->dev, headroom + skb_gro_offset(p));
2683 if (unlikely(!nskb))
2686 __copy_skb_header(nskb, p);
2687 nskb->mac_len = p->mac_len;
2689 skb_reserve(nskb, headroom);
2690 __skb_put(nskb, skb_gro_offset(p));
2692 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
2693 skb_set_network_header(nskb, skb_network_offset(p));
2694 skb_set_transport_header(nskb, skb_transport_offset(p));
2696 __skb_pull(p, skb_gro_offset(p));
2697 memcpy(skb_mac_header(nskb), skb_mac_header(p),
2698 p->data - skb_mac_header(p));
2700 *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2701 skb_shinfo(nskb)->frag_list = p;
2702 skb_shinfo(nskb)->gso_size = skb_shinfo(p)->gso_size;
2703 skb_header_release(p);
2706 nskb->data_len += p->len;
2707 nskb->truesize += p->len;
2708 nskb->len += p->len;
2711 nskb->next = p->next;
2717 if (skb_gro_offset(skb) > skb_headlen(skb)) {
2718 skb_shinfo(skb)->frags[0].page_offset +=
2719 skb_gro_offset(skb) - skb_headlen(skb);
2720 skb_shinfo(skb)->frags[0].size -=
2721 skb_gro_offset(skb) - skb_headlen(skb);
2722 skb_gro_reset_offset(skb);
2723 skb_gro_pull(skb, skb_headlen(skb));
2726 __skb_pull(skb, skb_gro_offset(skb));
2728 p->prev->next = skb;
2730 skb_header_release(skb);
2733 NAPI_GRO_CB(p)->count++;
2738 NAPI_GRO_CB(skb)->same_flow = 1;
2741 EXPORT_SYMBOL_GPL(skb_gro_receive);
2743 void __init skb_init(void)
2745 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2746 sizeof(struct sk_buff),
2748 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2750 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2751 (2*sizeof(struct sk_buff)) +
2754 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2759 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2760 * @skb: Socket buffer containing the buffers to be mapped
2761 * @sg: The scatter-gather list to map into
2762 * @offset: The offset into the buffer's contents to start mapping
2763 * @len: Length of buffer space to be mapped
2765 * Fill the specified scatter-gather list with mappings/pointers into a
2766 * region of the buffer space attached to a socket buffer.
2769 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2771 int start = skb_headlen(skb);
2772 int i, copy = start - offset;
2778 sg_set_buf(sg, skb->data + offset, copy);
2780 if ((len -= copy) == 0)
2785 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2788 WARN_ON(start > offset + len);
2790 end = start + skb_shinfo(skb)->frags[i].size;
2791 if ((copy = end - offset) > 0) {
2792 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2796 sg_set_page(&sg[elt], frag->page, copy,
2797 frag->page_offset+offset-start);
2806 if (skb_shinfo(skb)->frag_list) {
2807 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2809 for (; list; list = list->next) {
2812 WARN_ON(start > offset + len);
2814 end = start + list->len;
2815 if ((copy = end - offset) > 0) {
2818 elt += __skb_to_sgvec(list, sg+elt, offset - start,
2820 if ((len -= copy) == 0)
2831 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2833 int nsg = __skb_to_sgvec(skb, sg, offset, len);
2835 sg_mark_end(&sg[nsg - 1]);
2839 EXPORT_SYMBOL_GPL(skb_to_sgvec);
2842 * skb_cow_data - Check that a socket buffer's data buffers are writable
2843 * @skb: The socket buffer to check.
2844 * @tailbits: Amount of trailing space to be added
2845 * @trailer: Returned pointer to the skb where the @tailbits space begins
2847 * Make sure that the data buffers attached to a socket buffer are
2848 * writable. If they are not, private copies are made of the data buffers
2849 * and the socket buffer is set to use these instead.
2851 * If @tailbits is given, make sure that there is space to write @tailbits
2852 * bytes of data beyond current end of socket buffer. @trailer will be
2853 * set to point to the skb in which this space begins.
2855 * The number of scatterlist elements required to completely map the
2856 * COW'd and extended socket buffer will be returned.
2858 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
2862 struct sk_buff *skb1, **skb_p;
2864 /* If skb is cloned or its head is paged, reallocate
2865 * head pulling out all the pages (pages are considered not writable
2866 * at the moment even if they are anonymous).
2868 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
2869 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
2872 /* Easy case. Most of packets will go this way. */
2873 if (!skb_shinfo(skb)->frag_list) {
2874 /* A little of trouble, not enough of space for trailer.
2875 * This should not happen, when stack is tuned to generate
2876 * good frames. OK, on miss we reallocate and reserve even more
2877 * space, 128 bytes is fair. */
2879 if (skb_tailroom(skb) < tailbits &&
2880 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
2888 /* Misery. We are in troubles, going to mincer fragments... */
2891 skb_p = &skb_shinfo(skb)->frag_list;
2894 while ((skb1 = *skb_p) != NULL) {
2897 /* The fragment is partially pulled by someone,
2898 * this can happen on input. Copy it and everything
2901 if (skb_shared(skb1))
2904 /* If the skb is the last, worry about trailer. */
2906 if (skb1->next == NULL && tailbits) {
2907 if (skb_shinfo(skb1)->nr_frags ||
2908 skb_shinfo(skb1)->frag_list ||
2909 skb_tailroom(skb1) < tailbits)
2910 ntail = tailbits + 128;
2916 skb_shinfo(skb1)->nr_frags ||
2917 skb_shinfo(skb1)->frag_list) {
2918 struct sk_buff *skb2;
2920 /* Fuck, we are miserable poor guys... */
2922 skb2 = skb_copy(skb1, GFP_ATOMIC);
2924 skb2 = skb_copy_expand(skb1,
2928 if (unlikely(skb2 == NULL))
2932 skb_set_owner_w(skb2, skb1->sk);
2934 /* Looking around. Are we still alive?
2935 * OK, link new skb, drop old one */
2937 skb2->next = skb1->next;
2944 skb_p = &skb1->next;
2949 EXPORT_SYMBOL_GPL(skb_cow_data);
2951 void skb_tstamp_tx(struct sk_buff *orig_skb,
2952 struct skb_shared_hwtstamps *hwtstamps)
2954 struct sock *sk = orig_skb->sk;
2955 struct sock_exterr_skb *serr;
2956 struct sk_buff *skb;
2962 skb = skb_clone(orig_skb, GFP_ATOMIC);
2967 *skb_hwtstamps(skb) =
2971 * no hardware time stamps available,
2972 * so keep the skb_shared_tx and only
2973 * store software time stamp
2975 skb->tstamp = ktime_get_real();
2978 serr = SKB_EXT_ERR(skb);
2979 memset(serr, 0, sizeof(*serr));
2980 serr->ee.ee_errno = ENOMSG;
2981 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
2982 err = sock_queue_err_skb(sk, skb);
2986 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
2990 * skb_partial_csum_set - set up and verify partial csum values for packet
2991 * @skb: the skb to set
2992 * @start: the number of bytes after skb->data to start checksumming.
2993 * @off: the offset from start to place the checksum.
2995 * For untrusted partially-checksummed packets, we need to make sure the values
2996 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
2998 * This function checks and sets those values and skb->ip_summed: if this
2999 * returns false you should drop the packet.
3001 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3003 if (unlikely(start > skb->len - 2) ||
3004 unlikely((int)start + off > skb->len - 2)) {
3005 if (net_ratelimit())
3007 "bad partial csum: csum=%u/%u len=%u\n",
3008 start, off, skb->len);
3011 skb->ip_summed = CHECKSUM_PARTIAL;
3012 skb->csum_start = skb_headroom(skb) + start;
3013 skb->csum_offset = off;
3016 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3018 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3020 if (net_ratelimit())
3021 pr_warning("%s: received packets cannot be forwarded"
3022 " while LRO is enabled\n", skb->dev->name);
3024 EXPORT_SYMBOL(__skb_warn_lro_forwarding);