2 * Routines having to do with the 'struct sk_buff' memory handlers.
4 * Authors: Alan Cox <iiitac@pyr.swan.ac.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
7 * Version: $Id: skbuff.c,v 1.90 2001/11/07 05:56:19 davem Exp $
10 * Alan Cox : Fixed the worst of the load
12 * Dave Platt : Interrupt stacking fix.
13 * Richard Kooijman : Timestamp fixes.
14 * Alan Cox : Changed buffer format.
15 * Alan Cox : destructor hook for AF_UNIX etc.
16 * Linus Torvalds : Better skb_clone.
17 * Alan Cox : Added skb_copy.
18 * Alan Cox : Added all the changed routines Linus
19 * only put in the headers
20 * Ray VanTassle : Fixed --skb->lock in free
21 * Alan Cox : skb_copy copy arp field
22 * Andi Kleen : slabified it.
23 * Robert Olsson : Removed skb_head_pool
26 * The __skb_ routines should be called with interrupts
27 * disabled, or you better be *real* sure that the operation is atomic
28 * with respect to whatever list is being frobbed (e.g. via lock_sock()
29 * or via disabling bottom half handlers, etc).
31 * This program is free software; you can redistribute it and/or
32 * modify it under the terms of the GNU General Public License
33 * as published by the Free Software Foundation; either version
34 * 2 of the License, or (at your option) any later version.
38 * The functions in this file will not compile correctly with gcc 2.4.x
41 #include <linux/module.h>
42 #include <linux/types.h>
43 #include <linux/kernel.h>
45 #include <linux/interrupt.h>
47 #include <linux/inet.h>
48 #include <linux/slab.h>
49 #include <linux/netdevice.h>
50 #ifdef CONFIG_NET_CLS_ACT
51 #include <net/pkt_sched.h>
53 #include <linux/string.h>
54 #include <linux/skbuff.h>
55 #include <linux/cache.h>
56 #include <linux/rtnetlink.h>
57 #include <linux/init.h>
58 #include <linux/scatterlist.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;
75 * Keep out-of-line to prevent kernel bloat.
76 * __builtin_return_address is not used because it is not always
81 * skb_over_panic - private function
86 * Out of line support code for skb_put(). Not user callable.
88 void skb_over_panic(struct sk_buff *skb, int sz, void *here)
90 printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
91 "data:%p tail:%#lx end:%#lx dev:%s\n",
92 here, skb->len, sz, skb->head, skb->data,
93 (unsigned long)skb->tail, (unsigned long)skb->end,
94 skb->dev ? skb->dev->name : "<NULL>");
99 * skb_under_panic - private function
104 * Out of line support code for skb_push(). Not user callable.
107 void skb_under_panic(struct sk_buff *skb, int sz, void *here)
109 printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
110 "data:%p tail:%#lx end:%#lx dev:%s\n",
111 here, skb->len, sz, skb->head, skb->data,
112 (unsigned long)skb->tail, (unsigned long)skb->end,
113 skb->dev ? skb->dev->name : "<NULL>");
117 void skb_truesize_bug(struct sk_buff *skb)
119 printk(KERN_ERR "SKB BUG: Invalid truesize (%u) "
120 "len=%u, sizeof(sk_buff)=%Zd\n",
121 skb->truesize, skb->len, sizeof(struct sk_buff));
123 EXPORT_SYMBOL(skb_truesize_bug);
125 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
126 * 'private' fields and also do memory statistics to find all the
132 * __alloc_skb - allocate a network buffer
133 * @size: size to allocate
134 * @gfp_mask: allocation mask
135 * @fclone: allocate from fclone cache instead of head cache
136 * and allocate a cloned (child) skb
137 * @node: numa node to allocate memory on
139 * Allocate a new &sk_buff. The returned buffer has no headroom and a
140 * tail room of size bytes. The object has a reference count of one.
141 * The return is the buffer. On a failure the return is %NULL.
143 * Buffers may only be allocated from interrupts using a @gfp_mask of
146 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
147 int fclone, int node)
149 struct kmem_cache *cache;
150 struct skb_shared_info *shinfo;
154 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
157 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
161 size = SKB_DATA_ALIGN(size);
162 data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info),
168 * See comment in sk_buff definition, just before the 'tail' member
170 memset(skb, 0, offsetof(struct sk_buff, tail));
171 skb->truesize = size + sizeof(struct sk_buff);
172 atomic_set(&skb->users, 1);
175 skb_reset_tail_pointer(skb);
176 skb->end = skb->tail + size;
177 /* make sure we initialize shinfo sequentially */
178 shinfo = skb_shinfo(skb);
179 atomic_set(&shinfo->dataref, 1);
180 shinfo->nr_frags = 0;
181 shinfo->gso_size = 0;
182 shinfo->gso_segs = 0;
183 shinfo->gso_type = 0;
184 shinfo->ip6_frag_id = 0;
185 shinfo->frag_list = NULL;
188 struct sk_buff *child = skb + 1;
189 atomic_t *fclone_ref = (atomic_t *) (child + 1);
191 skb->fclone = SKB_FCLONE_ORIG;
192 atomic_set(fclone_ref, 1);
194 child->fclone = SKB_FCLONE_UNAVAILABLE;
199 kmem_cache_free(cache, skb);
205 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
206 * @dev: network device to receive on
207 * @length: length to allocate
208 * @gfp_mask: get_free_pages mask, passed to alloc_skb
210 * Allocate a new &sk_buff and assign it a usage count of one. The
211 * buffer has unspecified headroom built in. Users should allocate
212 * the headroom they think they need without accounting for the
213 * built in space. The built in space is used for optimisations.
215 * %NULL is returned if there is no free memory.
217 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
218 unsigned int length, gfp_t gfp_mask)
220 int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1;
223 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, node);
225 skb_reserve(skb, NET_SKB_PAD);
231 static void skb_drop_list(struct sk_buff **listp)
233 struct sk_buff *list = *listp;
238 struct sk_buff *this = list;
244 static inline void skb_drop_fraglist(struct sk_buff *skb)
246 skb_drop_list(&skb_shinfo(skb)->frag_list);
249 static void skb_clone_fraglist(struct sk_buff *skb)
251 struct sk_buff *list;
253 for (list = skb_shinfo(skb)->frag_list; list; list = list->next)
257 static void skb_release_data(struct sk_buff *skb)
260 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
261 &skb_shinfo(skb)->dataref)) {
262 if (skb_shinfo(skb)->nr_frags) {
264 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
265 put_page(skb_shinfo(skb)->frags[i].page);
268 if (skb_shinfo(skb)->frag_list)
269 skb_drop_fraglist(skb);
276 * Free an skbuff by memory without cleaning the state.
278 void kfree_skbmem(struct sk_buff *skb)
280 struct sk_buff *other;
281 atomic_t *fclone_ref;
283 skb_release_data(skb);
284 switch (skb->fclone) {
285 case SKB_FCLONE_UNAVAILABLE:
286 kmem_cache_free(skbuff_head_cache, skb);
289 case SKB_FCLONE_ORIG:
290 fclone_ref = (atomic_t *) (skb + 2);
291 if (atomic_dec_and_test(fclone_ref))
292 kmem_cache_free(skbuff_fclone_cache, skb);
295 case SKB_FCLONE_CLONE:
296 fclone_ref = (atomic_t *) (skb + 1);
299 /* The clone portion is available for
300 * fast-cloning again.
302 skb->fclone = SKB_FCLONE_UNAVAILABLE;
304 if (atomic_dec_and_test(fclone_ref))
305 kmem_cache_free(skbuff_fclone_cache, other);
311 * __kfree_skb - private function
314 * Free an sk_buff. Release anything attached to the buffer.
315 * Clean the state. This is an internal helper function. Users should
316 * always call kfree_skb
319 void __kfree_skb(struct sk_buff *skb)
321 dst_release(skb->dst);
323 secpath_put(skb->sp);
325 if (skb->destructor) {
327 skb->destructor(skb);
329 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
330 nf_conntrack_put(skb->nfct);
331 nf_conntrack_put_reasm(skb->nfct_reasm);
333 #ifdef CONFIG_BRIDGE_NETFILTER
334 nf_bridge_put(skb->nf_bridge);
336 /* XXX: IS this still necessary? - JHS */
337 #ifdef CONFIG_NET_SCHED
339 #ifdef CONFIG_NET_CLS_ACT
348 * kfree_skb - free an sk_buff
349 * @skb: buffer to free
351 * Drop a reference to the buffer and free it if the usage count has
354 void kfree_skb(struct sk_buff *skb)
358 if (likely(atomic_read(&skb->users) == 1))
360 else if (likely(!atomic_dec_and_test(&skb->users)))
366 * skb_clone - duplicate an sk_buff
367 * @skb: buffer to clone
368 * @gfp_mask: allocation priority
370 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
371 * copies share the same packet data but not structure. The new
372 * buffer has a reference count of 1. If the allocation fails the
373 * function returns %NULL otherwise the new buffer is returned.
375 * If this function is called from an interrupt gfp_mask() must be
379 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
384 if (skb->fclone == SKB_FCLONE_ORIG &&
385 n->fclone == SKB_FCLONE_UNAVAILABLE) {
386 atomic_t *fclone_ref = (atomic_t *) (n + 1);
387 n->fclone = SKB_FCLONE_CLONE;
388 atomic_inc(fclone_ref);
390 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
393 n->fclone = SKB_FCLONE_UNAVAILABLE;
396 #define C(x) n->x = skb->x
398 n->next = n->prev = NULL;
409 secpath_get(skb->sp);
411 memcpy(n->cb, skb->cb, sizeof(skb->cb));
418 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
422 skb_copy_queue_mapping(n, skb);
424 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
428 n->destructor = NULL;
431 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
432 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
435 #ifdef CONFIG_NET_SCHED
437 #ifdef CONFIG_NET_CLS_ACT
438 n->tc_verd = SET_TC_VERD(skb->tc_verd,0);
439 n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd);
440 n->tc_verd = CLR_TC_MUNGED(n->tc_verd);
444 skb_copy_secmark(n, skb);
446 atomic_set(&n->users, 1);
452 atomic_inc(&(skb_shinfo(skb)->dataref));
458 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
460 #ifndef NET_SKBUFF_DATA_USES_OFFSET
462 * Shift between the two data areas in bytes
464 unsigned long offset = new->data - old->data;
468 skb_copy_queue_mapping(new, old);
469 new->priority = old->priority;
470 new->protocol = old->protocol;
471 new->dst = dst_clone(old->dst);
473 new->sp = secpath_get(old->sp);
475 new->transport_header = old->transport_header;
476 new->network_header = old->network_header;
477 new->mac_header = old->mac_header;
478 #ifndef NET_SKBUFF_DATA_USES_OFFSET
479 /* {transport,network,mac}_header are relative to skb->head */
480 new->transport_header += offset;
481 new->network_header += offset;
482 new->mac_header += offset;
484 memcpy(new->cb, old->cb, sizeof(old->cb));
485 new->local_df = old->local_df;
486 new->fclone = SKB_FCLONE_UNAVAILABLE;
487 new->pkt_type = old->pkt_type;
488 new->tstamp = old->tstamp;
489 new->destructor = NULL;
490 new->mark = old->mark;
492 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
493 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
494 new->nf_trace = old->nf_trace;
496 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
497 new->ipvs_property = old->ipvs_property;
499 #ifdef CONFIG_NET_SCHED
500 #ifdef CONFIG_NET_CLS_ACT
501 new->tc_verd = old->tc_verd;
503 new->tc_index = old->tc_index;
505 skb_copy_secmark(new, old);
506 atomic_set(&new->users, 1);
507 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
508 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
509 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
513 * skb_copy - create private copy of an sk_buff
514 * @skb: buffer to copy
515 * @gfp_mask: allocation priority
517 * Make a copy of both an &sk_buff and its data. This is used when the
518 * caller wishes to modify the data and needs a private copy of the
519 * data to alter. Returns %NULL on failure or the pointer to the buffer
520 * on success. The returned buffer has a reference count of 1.
522 * As by-product this function converts non-linear &sk_buff to linear
523 * one, so that &sk_buff becomes completely private and caller is allowed
524 * to modify all the data of returned buffer. This means that this
525 * function is not recommended for use in circumstances when only
526 * header is going to be modified. Use pskb_copy() instead.
529 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
531 int headerlen = skb->data - skb->head;
533 * Allocate the copy buffer
536 #ifdef NET_SKBUFF_DATA_USES_OFFSET
537 n = alloc_skb(skb->end + skb->data_len, gfp_mask);
539 n = alloc_skb(skb->end - skb->head + skb->data_len, gfp_mask);
544 /* Set the data pointer */
545 skb_reserve(n, headerlen);
546 /* Set the tail pointer and length */
547 skb_put(n, skb->len);
549 n->ip_summed = skb->ip_summed;
551 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
554 copy_skb_header(n, skb);
560 * pskb_copy - create copy of an sk_buff with private head.
561 * @skb: buffer to copy
562 * @gfp_mask: allocation priority
564 * Make a copy of both an &sk_buff and part of its data, located
565 * in header. Fragmented data remain shared. This is used when
566 * the caller wishes to modify only header of &sk_buff and needs
567 * private copy of the header to alter. Returns %NULL on failure
568 * or the pointer to the buffer on success.
569 * The returned buffer has a reference count of 1.
572 struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
575 * Allocate the copy buffer
578 #ifdef NET_SKBUFF_DATA_USES_OFFSET
579 n = alloc_skb(skb->end, gfp_mask);
581 n = alloc_skb(skb->end - skb->head, gfp_mask);
586 /* Set the data pointer */
587 skb_reserve(n, skb->data - skb->head);
588 /* Set the tail pointer and length */
589 skb_put(n, skb_headlen(skb));
591 skb_copy_from_linear_data(skb, n->data, n->len);
593 n->ip_summed = skb->ip_summed;
595 n->truesize += skb->data_len;
596 n->data_len = skb->data_len;
599 if (skb_shinfo(skb)->nr_frags) {
602 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
603 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
604 get_page(skb_shinfo(n)->frags[i].page);
606 skb_shinfo(n)->nr_frags = i;
609 if (skb_shinfo(skb)->frag_list) {
610 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
611 skb_clone_fraglist(n);
614 copy_skb_header(n, skb);
620 * pskb_expand_head - reallocate header of &sk_buff
621 * @skb: buffer to reallocate
622 * @nhead: room to add at head
623 * @ntail: room to add at tail
624 * @gfp_mask: allocation priority
626 * Expands (or creates identical copy, if &nhead and &ntail are zero)
627 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
628 * reference count of 1. Returns zero in the case of success or error,
629 * if expansion failed. In the last case, &sk_buff is not changed.
631 * All the pointers pointing into skb header may change and must be
632 * reloaded after call to this function.
635 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
640 #ifdef NET_SKBUFF_DATA_USES_OFFSET
641 int size = nhead + skb->end + ntail;
643 int size = nhead + (skb->end - skb->head) + ntail;
650 size = SKB_DATA_ALIGN(size);
652 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
656 /* Copy only real data... and, alas, header. This should be
657 * optimized for the cases when header is void. */
658 #ifdef NET_SKBUFF_DATA_USES_OFFSET
659 memcpy(data + nhead, skb->head, skb->tail);
661 memcpy(data + nhead, skb->head, skb->tail - skb->head);
663 memcpy(data + size, skb_end_pointer(skb),
664 sizeof(struct skb_shared_info));
666 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
667 get_page(skb_shinfo(skb)->frags[i].page);
669 if (skb_shinfo(skb)->frag_list)
670 skb_clone_fraglist(skb);
672 skb_release_data(skb);
674 off = (data + nhead) - skb->head;
678 #ifdef NET_SKBUFF_DATA_USES_OFFSET
682 skb->end = skb->head + size;
684 /* {transport,network,mac}_header and tail are relative to skb->head */
686 skb->transport_header += off;
687 skb->network_header += off;
688 skb->mac_header += off;
692 atomic_set(&skb_shinfo(skb)->dataref, 1);
699 /* Make private copy of skb with writable head and some headroom */
701 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
703 struct sk_buff *skb2;
704 int delta = headroom - skb_headroom(skb);
707 skb2 = pskb_copy(skb, GFP_ATOMIC);
709 skb2 = skb_clone(skb, GFP_ATOMIC);
710 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
721 * skb_copy_expand - copy and expand sk_buff
722 * @skb: buffer to copy
723 * @newheadroom: new free bytes at head
724 * @newtailroom: new free bytes at tail
725 * @gfp_mask: allocation priority
727 * Make a copy of both an &sk_buff and its data and while doing so
728 * allocate additional space.
730 * This is used when the caller wishes to modify the data and needs a
731 * private copy of the data to alter as well as more space for new fields.
732 * Returns %NULL on failure or the pointer to the buffer
733 * on success. The returned buffer has a reference count of 1.
735 * You must pass %GFP_ATOMIC as the allocation priority if this function
736 * is called from an interrupt.
738 * BUG ALERT: ip_summed is not copied. Why does this work? Is it used
739 * only by netfilter in the cases when checksum is recalculated? --ANK
741 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
742 int newheadroom, int newtailroom,
746 * Allocate the copy buffer
748 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
750 int oldheadroom = skb_headroom(skb);
751 int head_copy_len, head_copy_off;
757 skb_reserve(n, newheadroom);
759 /* Set the tail pointer and length */
760 skb_put(n, skb->len);
762 head_copy_len = oldheadroom;
764 if (newheadroom <= head_copy_len)
765 head_copy_len = newheadroom;
767 head_copy_off = newheadroom - head_copy_len;
769 /* Copy the linear header and data. */
770 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
771 skb->len + head_copy_len))
774 copy_skb_header(n, skb);
776 #ifdef NET_SKBUFF_DATA_USES_OFFSET
777 off = newheadroom - oldheadroom;
779 n->transport_header += off;
780 n->network_header += off;
781 n->mac_header += off;
787 * skb_pad - zero pad the tail of an skb
788 * @skb: buffer to pad
791 * Ensure that a buffer is followed by a padding area that is zero
792 * filled. Used by network drivers which may DMA or transfer data
793 * beyond the buffer end onto the wire.
795 * May return error in out of memory cases. The skb is freed on error.
798 int skb_pad(struct sk_buff *skb, int pad)
803 /* If the skbuff is non linear tailroom is always zero.. */
804 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
805 memset(skb->data+skb->len, 0, pad);
809 ntail = skb->data_len + pad - (skb->end - skb->tail);
810 if (likely(skb_cloned(skb) || ntail > 0)) {
811 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
816 /* FIXME: The use of this function with non-linear skb's really needs
819 err = skb_linearize(skb);
823 memset(skb->data + skb->len, 0, pad);
831 /* Trims skb to length len. It can change skb pointers.
834 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
836 struct sk_buff **fragp;
837 struct sk_buff *frag;
838 int offset = skb_headlen(skb);
839 int nfrags = skb_shinfo(skb)->nr_frags;
843 if (skb_cloned(skb) &&
844 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
851 for (; i < nfrags; i++) {
852 int end = offset + skb_shinfo(skb)->frags[i].size;
859 skb_shinfo(skb)->frags[i++].size = len - offset;
862 skb_shinfo(skb)->nr_frags = i;
864 for (; i < nfrags; i++)
865 put_page(skb_shinfo(skb)->frags[i].page);
867 if (skb_shinfo(skb)->frag_list)
868 skb_drop_fraglist(skb);
872 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
873 fragp = &frag->next) {
874 int end = offset + frag->len;
876 if (skb_shared(frag)) {
877 struct sk_buff *nfrag;
879 nfrag = skb_clone(frag, GFP_ATOMIC);
880 if (unlikely(!nfrag))
883 nfrag->next = frag->next;
895 unlikely((err = pskb_trim(frag, len - offset))))
899 skb_drop_list(&frag->next);
904 if (len > skb_headlen(skb)) {
905 skb->data_len -= skb->len - len;
910 skb_set_tail_pointer(skb, len);
917 * __pskb_pull_tail - advance tail of skb header
918 * @skb: buffer to reallocate
919 * @delta: number of bytes to advance tail
921 * The function makes a sense only on a fragmented &sk_buff,
922 * it expands header moving its tail forward and copying necessary
923 * data from fragmented part.
925 * &sk_buff MUST have reference count of 1.
927 * Returns %NULL (and &sk_buff does not change) if pull failed
928 * or value of new tail of skb in the case of success.
930 * All the pointers pointing into skb header may change and must be
931 * reloaded after call to this function.
934 /* Moves tail of skb head forward, copying data from fragmented part,
935 * when it is necessary.
936 * 1. It may fail due to malloc failure.
937 * 2. It may change skb pointers.
939 * It is pretty complicated. Luckily, it is called only in exceptional cases.
941 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
943 /* If skb has not enough free space at tail, get new one
944 * plus 128 bytes for future expansions. If we have enough
945 * room at tail, reallocate without expansion only if skb is cloned.
947 int i, k, eat = (skb->tail + delta) - skb->end;
949 if (eat > 0 || skb_cloned(skb)) {
950 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
955 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
958 /* Optimization: no fragments, no reasons to preestimate
959 * size of pulled pages. Superb.
961 if (!skb_shinfo(skb)->frag_list)
964 /* Estimate size of pulled pages. */
966 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
967 if (skb_shinfo(skb)->frags[i].size >= eat)
969 eat -= skb_shinfo(skb)->frags[i].size;
972 /* If we need update frag list, we are in troubles.
973 * Certainly, it possible to add an offset to skb data,
974 * but taking into account that pulling is expected to
975 * be very rare operation, it is worth to fight against
976 * further bloating skb head and crucify ourselves here instead.
977 * Pure masohism, indeed. 8)8)
980 struct sk_buff *list = skb_shinfo(skb)->frag_list;
981 struct sk_buff *clone = NULL;
982 struct sk_buff *insp = NULL;
987 if (list->len <= eat) {
988 /* Eaten as whole. */
993 /* Eaten partially. */
995 if (skb_shared(list)) {
996 /* Sucks! We need to fork list. :-( */
997 clone = skb_clone(list, GFP_ATOMIC);
1003 /* This may be pulled without
1007 if (!pskb_pull(list, eat)) {
1016 /* Free pulled out fragments. */
1017 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1018 skb_shinfo(skb)->frag_list = list->next;
1021 /* And insert new clone at head. */
1024 skb_shinfo(skb)->frag_list = clone;
1027 /* Success! Now we may commit changes to skb data. */
1032 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1033 if (skb_shinfo(skb)->frags[i].size <= eat) {
1034 put_page(skb_shinfo(skb)->frags[i].page);
1035 eat -= skb_shinfo(skb)->frags[i].size;
1037 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1039 skb_shinfo(skb)->frags[k].page_offset += eat;
1040 skb_shinfo(skb)->frags[k].size -= eat;
1046 skb_shinfo(skb)->nr_frags = k;
1049 skb->data_len -= delta;
1051 return skb_tail_pointer(skb);
1054 /* Copy some data bits from skb to kernel buffer. */
1056 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1059 int start = skb_headlen(skb);
1061 if (offset > (int)skb->len - len)
1065 if ((copy = start - offset) > 0) {
1068 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1069 if ((len -= copy) == 0)
1075 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1078 BUG_TRAP(start <= offset + len);
1080 end = start + skb_shinfo(skb)->frags[i].size;
1081 if ((copy = end - offset) > 0) {
1087 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1089 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1090 offset - start, copy);
1091 kunmap_skb_frag(vaddr);
1093 if ((len -= copy) == 0)
1101 if (skb_shinfo(skb)->frag_list) {
1102 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1104 for (; list; list = list->next) {
1107 BUG_TRAP(start <= offset + len);
1109 end = start + list->len;
1110 if ((copy = end - offset) > 0) {
1113 if (skb_copy_bits(list, offset - start,
1116 if ((len -= copy) == 0)
1132 * skb_store_bits - store bits from kernel buffer to skb
1133 * @skb: destination buffer
1134 * @offset: offset in destination
1135 * @from: source buffer
1136 * @len: number of bytes to copy
1138 * Copy the specified number of bytes from the source buffer to the
1139 * destination skb. This function handles all the messy bits of
1140 * traversing fragment lists and such.
1143 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1146 int start = skb_headlen(skb);
1148 if (offset > (int)skb->len - len)
1151 if ((copy = start - offset) > 0) {
1154 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1155 if ((len -= copy) == 0)
1161 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1162 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1165 BUG_TRAP(start <= offset + len);
1167 end = start + frag->size;
1168 if ((copy = end - offset) > 0) {
1174 vaddr = kmap_skb_frag(frag);
1175 memcpy(vaddr + frag->page_offset + offset - start,
1177 kunmap_skb_frag(vaddr);
1179 if ((len -= copy) == 0)
1187 if (skb_shinfo(skb)->frag_list) {
1188 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1190 for (; list; list = list->next) {
1193 BUG_TRAP(start <= offset + len);
1195 end = start + list->len;
1196 if ((copy = end - offset) > 0) {
1199 if (skb_store_bits(list, offset - start,
1202 if ((len -= copy) == 0)
1217 EXPORT_SYMBOL(skb_store_bits);
1219 /* Checksum skb data. */
1221 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1222 int len, __wsum csum)
1224 int start = skb_headlen(skb);
1225 int i, copy = start - offset;
1228 /* Checksum header. */
1232 csum = csum_partial(skb->data + offset, copy, csum);
1233 if ((len -= copy) == 0)
1239 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1242 BUG_TRAP(start <= offset + len);
1244 end = start + skb_shinfo(skb)->frags[i].size;
1245 if ((copy = end - offset) > 0) {
1248 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1252 vaddr = kmap_skb_frag(frag);
1253 csum2 = csum_partial(vaddr + frag->page_offset +
1254 offset - start, copy, 0);
1255 kunmap_skb_frag(vaddr);
1256 csum = csum_block_add(csum, csum2, pos);
1265 if (skb_shinfo(skb)->frag_list) {
1266 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1268 for (; list; list = list->next) {
1271 BUG_TRAP(start <= offset + len);
1273 end = start + list->len;
1274 if ((copy = end - offset) > 0) {
1278 csum2 = skb_checksum(list, offset - start,
1280 csum = csum_block_add(csum, csum2, pos);
1281 if ((len -= copy) == 0)
1294 /* Both of above in one bottle. */
1296 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1297 u8 *to, int len, __wsum csum)
1299 int start = skb_headlen(skb);
1300 int i, copy = start - offset;
1307 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1309 if ((len -= copy) == 0)
1316 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1319 BUG_TRAP(start <= offset + len);
1321 end = start + skb_shinfo(skb)->frags[i].size;
1322 if ((copy = end - offset) > 0) {
1325 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1329 vaddr = kmap_skb_frag(frag);
1330 csum2 = csum_partial_copy_nocheck(vaddr +
1334 kunmap_skb_frag(vaddr);
1335 csum = csum_block_add(csum, csum2, pos);
1345 if (skb_shinfo(skb)->frag_list) {
1346 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1348 for (; list; list = list->next) {
1352 BUG_TRAP(start <= offset + len);
1354 end = start + list->len;
1355 if ((copy = end - offset) > 0) {
1358 csum2 = skb_copy_and_csum_bits(list,
1361 csum = csum_block_add(csum, csum2, pos);
1362 if ((len -= copy) == 0)
1375 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1380 if (skb->ip_summed == CHECKSUM_PARTIAL)
1381 csstart = skb->csum_start - skb_headroom(skb);
1383 csstart = skb_headlen(skb);
1385 BUG_ON(csstart > skb_headlen(skb));
1387 skb_copy_from_linear_data(skb, to, csstart);
1390 if (csstart != skb->len)
1391 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1392 skb->len - csstart, 0);
1394 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1395 long csstuff = csstart + skb->csum_offset;
1397 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
1402 * skb_dequeue - remove from the head of the queue
1403 * @list: list to dequeue from
1405 * Remove the head of the list. The list lock is taken so the function
1406 * may be used safely with other locking list functions. The head item is
1407 * returned or %NULL if the list is empty.
1410 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1412 unsigned long flags;
1413 struct sk_buff *result;
1415 spin_lock_irqsave(&list->lock, flags);
1416 result = __skb_dequeue(list);
1417 spin_unlock_irqrestore(&list->lock, flags);
1422 * skb_dequeue_tail - remove from the tail of the queue
1423 * @list: list to dequeue from
1425 * Remove the tail of the list. The list lock is taken so the function
1426 * may be used safely with other locking list functions. The tail item is
1427 * returned or %NULL if the list is empty.
1429 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1431 unsigned long flags;
1432 struct sk_buff *result;
1434 spin_lock_irqsave(&list->lock, flags);
1435 result = __skb_dequeue_tail(list);
1436 spin_unlock_irqrestore(&list->lock, flags);
1441 * skb_queue_purge - empty a list
1442 * @list: list to empty
1444 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1445 * the list and one reference dropped. This function takes the list
1446 * lock and is atomic with respect to other list locking functions.
1448 void skb_queue_purge(struct sk_buff_head *list)
1450 struct sk_buff *skb;
1451 while ((skb = skb_dequeue(list)) != NULL)
1456 * skb_queue_head - queue a buffer at the list head
1457 * @list: list to use
1458 * @newsk: buffer to queue
1460 * Queue a buffer at the start of the list. This function takes the
1461 * list lock and can be used safely with other locking &sk_buff functions
1464 * A buffer cannot be placed on two lists at the same time.
1466 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1468 unsigned long flags;
1470 spin_lock_irqsave(&list->lock, flags);
1471 __skb_queue_head(list, newsk);
1472 spin_unlock_irqrestore(&list->lock, flags);
1476 * skb_queue_tail - queue a buffer at the list tail
1477 * @list: list to use
1478 * @newsk: buffer to queue
1480 * Queue a buffer at the tail of the list. This function takes the
1481 * list lock and can be used safely with other locking &sk_buff functions
1484 * A buffer cannot be placed on two lists at the same time.
1486 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1488 unsigned long flags;
1490 spin_lock_irqsave(&list->lock, flags);
1491 __skb_queue_tail(list, newsk);
1492 spin_unlock_irqrestore(&list->lock, flags);
1496 * skb_unlink - remove a buffer from a list
1497 * @skb: buffer to remove
1498 * @list: list to use
1500 * Remove a packet from a list. The list locks are taken and this
1501 * function is atomic with respect to other list locked calls
1503 * You must know what list the SKB is on.
1505 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1507 unsigned long flags;
1509 spin_lock_irqsave(&list->lock, flags);
1510 __skb_unlink(skb, list);
1511 spin_unlock_irqrestore(&list->lock, flags);
1515 * skb_append - append a buffer
1516 * @old: buffer to insert after
1517 * @newsk: buffer to insert
1518 * @list: list to use
1520 * Place a packet after a given packet in a list. The list locks are taken
1521 * and this function is atomic with respect to other list locked calls.
1522 * A buffer cannot be placed on two lists at the same time.
1524 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1526 unsigned long flags;
1528 spin_lock_irqsave(&list->lock, flags);
1529 __skb_append(old, newsk, list);
1530 spin_unlock_irqrestore(&list->lock, flags);
1535 * skb_insert - insert a buffer
1536 * @old: buffer to insert before
1537 * @newsk: buffer to insert
1538 * @list: list to use
1540 * Place a packet before a given packet in a list. The list locks are
1541 * taken and this function is atomic with respect to other list locked
1544 * A buffer cannot be placed on two lists at the same time.
1546 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1548 unsigned long flags;
1550 spin_lock_irqsave(&list->lock, flags);
1551 __skb_insert(newsk, old->prev, old, list);
1552 spin_unlock_irqrestore(&list->lock, flags);
1555 static inline void skb_split_inside_header(struct sk_buff *skb,
1556 struct sk_buff* skb1,
1557 const u32 len, const int pos)
1561 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
1563 /* And move data appendix as is. */
1564 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1565 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
1567 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
1568 skb_shinfo(skb)->nr_frags = 0;
1569 skb1->data_len = skb->data_len;
1570 skb1->len += skb1->data_len;
1573 skb_set_tail_pointer(skb, len);
1576 static inline void skb_split_no_header(struct sk_buff *skb,
1577 struct sk_buff* skb1,
1578 const u32 len, int pos)
1581 const int nfrags = skb_shinfo(skb)->nr_frags;
1583 skb_shinfo(skb)->nr_frags = 0;
1584 skb1->len = skb1->data_len = skb->len - len;
1586 skb->data_len = len - pos;
1588 for (i = 0; i < nfrags; i++) {
1589 int size = skb_shinfo(skb)->frags[i].size;
1591 if (pos + size > len) {
1592 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
1596 * We have two variants in this case:
1597 * 1. Move all the frag to the second
1598 * part, if it is possible. F.e.
1599 * this approach is mandatory for TUX,
1600 * where splitting is expensive.
1601 * 2. Split is accurately. We make this.
1603 get_page(skb_shinfo(skb)->frags[i].page);
1604 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
1605 skb_shinfo(skb1)->frags[0].size -= len - pos;
1606 skb_shinfo(skb)->frags[i].size = len - pos;
1607 skb_shinfo(skb)->nr_frags++;
1611 skb_shinfo(skb)->nr_frags++;
1614 skb_shinfo(skb1)->nr_frags = k;
1618 * skb_split - Split fragmented skb to two parts at length len.
1619 * @skb: the buffer to split
1620 * @skb1: the buffer to receive the second part
1621 * @len: new length for skb
1623 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
1625 int pos = skb_headlen(skb);
1627 if (len < pos) /* Split line is inside header. */
1628 skb_split_inside_header(skb, skb1, len, pos);
1629 else /* Second chunk has no header, nothing to copy. */
1630 skb_split_no_header(skb, skb1, len, pos);
1634 * skb_prepare_seq_read - Prepare a sequential read of skb data
1635 * @skb: the buffer to read
1636 * @from: lower offset of data to be read
1637 * @to: upper offset of data to be read
1638 * @st: state variable
1640 * Initializes the specified state variable. Must be called before
1641 * invoking skb_seq_read() for the first time.
1643 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1644 unsigned int to, struct skb_seq_state *st)
1646 st->lower_offset = from;
1647 st->upper_offset = to;
1648 st->root_skb = st->cur_skb = skb;
1649 st->frag_idx = st->stepped_offset = 0;
1650 st->frag_data = NULL;
1654 * skb_seq_read - Sequentially read skb data
1655 * @consumed: number of bytes consumed by the caller so far
1656 * @data: destination pointer for data to be returned
1657 * @st: state variable
1659 * Reads a block of skb data at &consumed relative to the
1660 * lower offset specified to skb_prepare_seq_read(). Assigns
1661 * the head of the data block to &data and returns the length
1662 * of the block or 0 if the end of the skb data or the upper
1663 * offset has been reached.
1665 * The caller is not required to consume all of the data
1666 * returned, i.e. &consumed is typically set to the number
1667 * of bytes already consumed and the next call to
1668 * skb_seq_read() will return the remaining part of the block.
1670 * Note: The size of each block of data returned can be arbitary,
1671 * this limitation is the cost for zerocopy seqeuental
1672 * reads of potentially non linear data.
1674 * Note: Fragment lists within fragments are not implemented
1675 * at the moment, state->root_skb could be replaced with
1676 * a stack for this purpose.
1678 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1679 struct skb_seq_state *st)
1681 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
1684 if (unlikely(abs_offset >= st->upper_offset))
1688 block_limit = skb_headlen(st->cur_skb);
1690 if (abs_offset < block_limit) {
1691 *data = st->cur_skb->data + abs_offset;
1692 return block_limit - abs_offset;
1695 if (st->frag_idx == 0 && !st->frag_data)
1696 st->stepped_offset += skb_headlen(st->cur_skb);
1698 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
1699 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
1700 block_limit = frag->size + st->stepped_offset;
1702 if (abs_offset < block_limit) {
1704 st->frag_data = kmap_skb_frag(frag);
1706 *data = (u8 *) st->frag_data + frag->page_offset +
1707 (abs_offset - st->stepped_offset);
1709 return block_limit - abs_offset;
1712 if (st->frag_data) {
1713 kunmap_skb_frag(st->frag_data);
1714 st->frag_data = NULL;
1718 st->stepped_offset += frag->size;
1721 if (st->frag_data) {
1722 kunmap_skb_frag(st->frag_data);
1723 st->frag_data = NULL;
1726 if (st->cur_skb->next) {
1727 st->cur_skb = st->cur_skb->next;
1730 } else if (st->root_skb == st->cur_skb &&
1731 skb_shinfo(st->root_skb)->frag_list) {
1732 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
1740 * skb_abort_seq_read - Abort a sequential read of skb data
1741 * @st: state variable
1743 * Must be called if skb_seq_read() was not called until it
1746 void skb_abort_seq_read(struct skb_seq_state *st)
1749 kunmap_skb_frag(st->frag_data);
1752 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
1754 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
1755 struct ts_config *conf,
1756 struct ts_state *state)
1758 return skb_seq_read(offset, text, TS_SKB_CB(state));
1761 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
1763 skb_abort_seq_read(TS_SKB_CB(state));
1767 * skb_find_text - Find a text pattern in skb data
1768 * @skb: the buffer to look in
1769 * @from: search offset
1771 * @config: textsearch configuration
1772 * @state: uninitialized textsearch state variable
1774 * Finds a pattern in the skb data according to the specified
1775 * textsearch configuration. Use textsearch_next() to retrieve
1776 * subsequent occurrences of the pattern. Returns the offset
1777 * to the first occurrence or UINT_MAX if no match was found.
1779 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1780 unsigned int to, struct ts_config *config,
1781 struct ts_state *state)
1785 config->get_next_block = skb_ts_get_next_block;
1786 config->finish = skb_ts_finish;
1788 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
1790 ret = textsearch_find(config, state);
1791 return (ret <= to - from ? ret : UINT_MAX);
1795 * skb_append_datato_frags: - append the user data to a skb
1796 * @sk: sock structure
1797 * @skb: skb structure to be appened with user data.
1798 * @getfrag: call back function to be used for getting the user data
1799 * @from: pointer to user message iov
1800 * @length: length of the iov message
1802 * Description: This procedure append the user data in the fragment part
1803 * of the skb if any page alloc fails user this procedure returns -ENOMEM
1805 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
1806 int (*getfrag)(void *from, char *to, int offset,
1807 int len, int odd, struct sk_buff *skb),
1808 void *from, int length)
1811 skb_frag_t *frag = NULL;
1812 struct page *page = NULL;
1818 /* Return error if we don't have space for new frag */
1819 frg_cnt = skb_shinfo(skb)->nr_frags;
1820 if (frg_cnt >= MAX_SKB_FRAGS)
1823 /* allocate a new page for next frag */
1824 page = alloc_pages(sk->sk_allocation, 0);
1826 /* If alloc_page fails just return failure and caller will
1827 * free previous allocated pages by doing kfree_skb()
1832 /* initialize the next frag */
1833 sk->sk_sndmsg_page = page;
1834 sk->sk_sndmsg_off = 0;
1835 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
1836 skb->truesize += PAGE_SIZE;
1837 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1839 /* get the new initialized frag */
1840 frg_cnt = skb_shinfo(skb)->nr_frags;
1841 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
1843 /* copy the user data to page */
1844 left = PAGE_SIZE - frag->page_offset;
1845 copy = (length > left)? left : length;
1847 ret = getfrag(from, (page_address(frag->page) +
1848 frag->page_offset + frag->size),
1849 offset, copy, 0, skb);
1853 /* copy was successful so update the size parameters */
1854 sk->sk_sndmsg_off += copy;
1857 skb->data_len += copy;
1861 } while (length > 0);
1867 * skb_pull_rcsum - pull skb and update receive checksum
1868 * @skb: buffer to update
1869 * @start: start of data before pull
1870 * @len: length of data pulled
1872 * This function performs an skb_pull on the packet and updates
1873 * update the CHECKSUM_COMPLETE checksum. It should be used on
1874 * receive path processing instead of skb_pull unless you know
1875 * that the checksum difference is zero (e.g., a valid IP header)
1876 * or you are setting ip_summed to CHECKSUM_NONE.
1878 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
1880 BUG_ON(len > skb->len);
1882 BUG_ON(skb->len < skb->data_len);
1883 skb_postpull_rcsum(skb, skb->data, len);
1884 return skb->data += len;
1887 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
1890 * skb_segment - Perform protocol segmentation on skb.
1891 * @skb: buffer to segment
1892 * @features: features for the output path (see dev->features)
1894 * This function performs segmentation on the given skb. It returns
1895 * the segment at the given position. It returns NULL if there are
1896 * no more segments to generate, or when an error is encountered.
1898 struct sk_buff *skb_segment(struct sk_buff *skb, int features)
1900 struct sk_buff *segs = NULL;
1901 struct sk_buff *tail = NULL;
1902 unsigned int mss = skb_shinfo(skb)->gso_size;
1903 unsigned int doffset = skb->data - skb_mac_header(skb);
1904 unsigned int offset = doffset;
1905 unsigned int headroom;
1907 int sg = features & NETIF_F_SG;
1908 int nfrags = skb_shinfo(skb)->nr_frags;
1913 __skb_push(skb, doffset);
1914 headroom = skb_headroom(skb);
1915 pos = skb_headlen(skb);
1918 struct sk_buff *nskb;
1924 len = skb->len - offset;
1928 hsize = skb_headlen(skb) - offset;
1931 if (hsize > len || !sg)
1934 nskb = alloc_skb(hsize + doffset + headroom, GFP_ATOMIC);
1935 if (unlikely(!nskb))
1944 nskb->dev = skb->dev;
1945 skb_copy_queue_mapping(nskb, skb);
1946 nskb->priority = skb->priority;
1947 nskb->protocol = skb->protocol;
1948 nskb->dst = dst_clone(skb->dst);
1949 memcpy(nskb->cb, skb->cb, sizeof(skb->cb));
1950 nskb->pkt_type = skb->pkt_type;
1951 nskb->mac_len = skb->mac_len;
1953 skb_reserve(nskb, headroom);
1954 skb_reset_mac_header(nskb);
1955 skb_set_network_header(nskb, skb->mac_len);
1956 nskb->transport_header = (nskb->network_header +
1957 skb_network_header_len(skb));
1958 skb_copy_from_linear_data(skb, skb_put(nskb, doffset),
1961 nskb->csum = skb_copy_and_csum_bits(skb, offset,
1967 frag = skb_shinfo(nskb)->frags;
1970 nskb->ip_summed = CHECKSUM_PARTIAL;
1971 nskb->csum = skb->csum;
1972 skb_copy_from_linear_data_offset(skb, offset,
1973 skb_put(nskb, hsize), hsize);
1975 while (pos < offset + len) {
1976 BUG_ON(i >= nfrags);
1978 *frag = skb_shinfo(skb)->frags[i];
1979 get_page(frag->page);
1983 frag->page_offset += offset - pos;
1984 frag->size -= offset - pos;
1989 if (pos + size <= offset + len) {
1993 frag->size -= pos + size - (offset + len);
2000 skb_shinfo(nskb)->nr_frags = k;
2001 nskb->data_len = len - hsize;
2002 nskb->len += nskb->data_len;
2003 nskb->truesize += nskb->data_len;
2004 } while ((offset += len) < skb->len);
2009 while ((skb = segs)) {
2013 return ERR_PTR(err);
2016 EXPORT_SYMBOL_GPL(skb_segment);
2018 void __init skb_init(void)
2020 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2021 sizeof(struct sk_buff),
2023 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2025 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2026 (2*sizeof(struct sk_buff)) +
2029 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2034 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2035 * @skb: Socket buffer containing the buffers to be mapped
2036 * @sg: The scatter-gather list to map into
2037 * @offset: The offset into the buffer's contents to start mapping
2038 * @len: Length of buffer space to be mapped
2040 * Fill the specified scatter-gather list with mappings/pointers into a
2041 * region of the buffer space attached to a socket buffer.
2044 skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2046 int start = skb_headlen(skb);
2047 int i, copy = start - offset;
2053 sg[elt].page = virt_to_page(skb->data + offset);
2054 sg[elt].offset = (unsigned long)(skb->data + offset) % PAGE_SIZE;
2055 sg[elt].length = copy;
2057 if ((len -= copy) == 0)
2062 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2065 BUG_TRAP(start <= offset + len);
2067 end = start + skb_shinfo(skb)->frags[i].size;
2068 if ((copy = end - offset) > 0) {
2069 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2073 sg[elt].page = frag->page;
2074 sg[elt].offset = frag->page_offset+offset-start;
2075 sg[elt].length = copy;
2084 if (skb_shinfo(skb)->frag_list) {
2085 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2087 for (; list; list = list->next) {
2090 BUG_TRAP(start <= offset + len);
2092 end = start + list->len;
2093 if ((copy = end - offset) > 0) {
2096 elt += skb_to_sgvec(list, sg+elt, offset - start, copy);
2097 if ((len -= copy) == 0)
2109 * skb_cow_data - Check that a socket buffer's data buffers are writable
2110 * @skb: The socket buffer to check.
2111 * @tailbits: Amount of trailing space to be added
2112 * @trailer: Returned pointer to the skb where the @tailbits space begins
2114 * Make sure that the data buffers attached to a socket buffer are
2115 * writable. If they are not, private copies are made of the data buffers
2116 * and the socket buffer is set to use these instead.
2118 * If @tailbits is given, make sure that there is space to write @tailbits
2119 * bytes of data beyond current end of socket buffer. @trailer will be
2120 * set to point to the skb in which this space begins.
2122 * The number of scatterlist elements required to completely map the
2123 * COW'd and extended socket buffer will be returned.
2125 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
2129 struct sk_buff *skb1, **skb_p;
2131 /* If skb is cloned or its head is paged, reallocate
2132 * head pulling out all the pages (pages are considered not writable
2133 * at the moment even if they are anonymous).
2135 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
2136 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
2139 /* Easy case. Most of packets will go this way. */
2140 if (!skb_shinfo(skb)->frag_list) {
2141 /* A little of trouble, not enough of space for trailer.
2142 * This should not happen, when stack is tuned to generate
2143 * good frames. OK, on miss we reallocate and reserve even more
2144 * space, 128 bytes is fair. */
2146 if (skb_tailroom(skb) < tailbits &&
2147 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
2155 /* Misery. We are in troubles, going to mincer fragments... */
2158 skb_p = &skb_shinfo(skb)->frag_list;
2161 while ((skb1 = *skb_p) != NULL) {
2164 /* The fragment is partially pulled by someone,
2165 * this can happen on input. Copy it and everything
2168 if (skb_shared(skb1))
2171 /* If the skb is the last, worry about trailer. */
2173 if (skb1->next == NULL && tailbits) {
2174 if (skb_shinfo(skb1)->nr_frags ||
2175 skb_shinfo(skb1)->frag_list ||
2176 skb_tailroom(skb1) < tailbits)
2177 ntail = tailbits + 128;
2183 skb_shinfo(skb1)->nr_frags ||
2184 skb_shinfo(skb1)->frag_list) {
2185 struct sk_buff *skb2;
2187 /* Fuck, we are miserable poor guys... */
2189 skb2 = skb_copy(skb1, GFP_ATOMIC);
2191 skb2 = skb_copy_expand(skb1,
2195 if (unlikely(skb2 == NULL))
2199 skb_set_owner_w(skb2, skb1->sk);
2201 /* Looking around. Are we still alive?
2202 * OK, link new skb, drop old one */
2204 skb2->next = skb1->next;
2211 skb_p = &skb1->next;
2217 EXPORT_SYMBOL(___pskb_trim);
2218 EXPORT_SYMBOL(__kfree_skb);
2219 EXPORT_SYMBOL(kfree_skb);
2220 EXPORT_SYMBOL(__pskb_pull_tail);
2221 EXPORT_SYMBOL(__alloc_skb);
2222 EXPORT_SYMBOL(__netdev_alloc_skb);
2223 EXPORT_SYMBOL(pskb_copy);
2224 EXPORT_SYMBOL(pskb_expand_head);
2225 EXPORT_SYMBOL(skb_checksum);
2226 EXPORT_SYMBOL(skb_clone);
2227 EXPORT_SYMBOL(skb_copy);
2228 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2229 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2230 EXPORT_SYMBOL(skb_copy_bits);
2231 EXPORT_SYMBOL(skb_copy_expand);
2232 EXPORT_SYMBOL(skb_over_panic);
2233 EXPORT_SYMBOL(skb_pad);
2234 EXPORT_SYMBOL(skb_realloc_headroom);
2235 EXPORT_SYMBOL(skb_under_panic);
2236 EXPORT_SYMBOL(skb_dequeue);
2237 EXPORT_SYMBOL(skb_dequeue_tail);
2238 EXPORT_SYMBOL(skb_insert);
2239 EXPORT_SYMBOL(skb_queue_purge);
2240 EXPORT_SYMBOL(skb_queue_head);
2241 EXPORT_SYMBOL(skb_queue_tail);
2242 EXPORT_SYMBOL(skb_unlink);
2243 EXPORT_SYMBOL(skb_append);
2244 EXPORT_SYMBOL(skb_split);
2245 EXPORT_SYMBOL(skb_prepare_seq_read);
2246 EXPORT_SYMBOL(skb_seq_read);
2247 EXPORT_SYMBOL(skb_abort_seq_read);
2248 EXPORT_SYMBOL(skb_find_text);
2249 EXPORT_SYMBOL(skb_append_datato_frags);
2251 EXPORT_SYMBOL_GPL(skb_to_sgvec);
2252 EXPORT_SYMBOL_GPL(skb_cow_data);