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/config.h>
42 #include <linux/module.h>
43 #include <linux/types.h>
44 #include <linux/kernel.h>
45 #include <linux/sched.h>
47 #include <linux/interrupt.h>
49 #include <linux/inet.h>
50 #include <linux/slab.h>
51 #include <linux/netdevice.h>
52 #ifdef CONFIG_NET_CLS_ACT
53 #include <net/pkt_sched.h>
55 #include <linux/string.h>
56 #include <linux/skbuff.h>
57 #include <linux/cache.h>
58 #include <linux/rtnetlink.h>
59 #include <linux/init.h>
60 #include <linux/highmem.h>
62 #include <net/protocol.h>
65 #include <net/checksum.h>
68 #include <asm/uaccess.h>
69 #include <asm/system.h>
71 static kmem_cache_t *skbuff_head_cache __read_mostly;
72 static kmem_cache_t *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:%p end:%p dev:%s\n",
92 here, skb->len, sz, skb->head, skb->data, skb->tail, skb->end,
93 skb->dev ? skb->dev->name : "<NULL>");
98 * skb_under_panic - private function
103 * Out of line support code for skb_push(). Not user callable.
106 void skb_under_panic(struct sk_buff *skb, int sz, void *here)
108 printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
109 "data:%p tail:%p end:%p dev:%s\n",
110 here, skb->len, sz, skb->head, skb->data, skb->tail, skb->end,
111 skb->dev ? skb->dev->name : "<NULL>");
115 void skb_truesize_bug(struct sk_buff *skb)
117 printk(KERN_ERR "SKB BUG: Invalid truesize (%u) "
118 "len=%u, sizeof(sk_buff)=%Zd\n",
119 skb->truesize, skb->len, sizeof(struct sk_buff));
121 EXPORT_SYMBOL(skb_truesize_bug);
123 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
124 * 'private' fields and also do memory statistics to find all the
130 * __alloc_skb - allocate a network buffer
131 * @size: size to allocate
132 * @gfp_mask: allocation mask
133 * @fclone: allocate from fclone cache instead of head cache
134 * and allocate a cloned (child) skb
136 * Allocate a new &sk_buff. The returned buffer has no headroom and a
137 * tail room of size bytes. The object has a reference count of one.
138 * The return is the buffer. On a failure the return is %NULL.
140 * Buffers may only be allocated from interrupts using a @gfp_mask of
143 #ifndef CONFIG_HAVE_ARCH_ALLOC_SKB
144 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
148 struct skb_shared_info *shinfo;
152 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
155 skb = kmem_cache_alloc(cache, gfp_mask & ~__GFP_DMA);
159 /* Get the DATA. Size must match skb_add_mtu(). */
160 size = SKB_DATA_ALIGN(size);
161 data = ____kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
165 memset(skb, 0, offsetof(struct sk_buff, truesize));
166 skb->truesize = size + sizeof(struct sk_buff);
167 atomic_set(&skb->users, 1);
171 skb->end = data + size;
172 /* make sure we initialize shinfo sequentially */
173 shinfo = skb_shinfo(skb);
174 atomic_set(&shinfo->dataref, 1);
175 shinfo->nr_frags = 0;
176 shinfo->gso_size = 0;
177 shinfo->gso_segs = 0;
178 shinfo->gso_type = 0;
179 shinfo->ip6_frag_id = 0;
180 shinfo->frag_list = NULL;
183 struct sk_buff *child = skb + 1;
184 atomic_t *fclone_ref = (atomic_t *) (child + 1);
186 skb->fclone = SKB_FCLONE_ORIG;
187 atomic_set(fclone_ref, 1);
189 child->fclone = SKB_FCLONE_UNAVAILABLE;
194 kmem_cache_free(cache, skb);
198 #endif /* !CONFIG_HAVE_ARCH_ALLOC_SKB */
201 * alloc_skb_from_cache - allocate a network buffer
202 * @cp: kmem_cache from which to allocate the data area
203 * (object size must be big enough for @size bytes + skb overheads)
204 * @size: size to allocate
205 * @gfp_mask: allocation mask
207 * Allocate a new &sk_buff. The returned buffer has no headroom and
208 * tail room of size bytes. The object has a reference count of one.
209 * The return is the buffer. On a failure the return is %NULL.
211 * Buffers may only be allocated from interrupts using a @gfp_mask of
214 struct sk_buff *alloc_skb_from_cache(kmem_cache_t *cp,
220 struct skb_shared_info *shinfo;
224 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
227 skb = kmem_cache_alloc(cache, gfp_mask & ~__GFP_DMA);
232 size = SKB_DATA_ALIGN(size);
233 data = kmem_cache_alloc(cp, gfp_mask);
237 memset(skb, 0, offsetof(struct sk_buff, truesize));
238 skb->truesize = size + sizeof(struct sk_buff);
239 atomic_set(&skb->users, 1);
243 skb->end = data + size;
244 /* make sure we initialize shinfo sequentially */
245 shinfo = skb_shinfo(skb);
246 atomic_set(&shinfo->dataref, 1);
247 shinfo->nr_frags = 0;
248 shinfo->gso_size = 0;
249 shinfo->gso_segs = 0;
250 shinfo->gso_type = 0;
251 shinfo->ip6_frag_id = 0;
252 shinfo->frag_list = NULL;
255 struct sk_buff *child = skb + 1;
256 atomic_t *fclone_ref = (atomic_t *) (child + 1);
258 skb->fclone = SKB_FCLONE_ORIG;
259 atomic_set(fclone_ref, 1);
261 child->fclone = SKB_FCLONE_UNAVAILABLE;
266 kmem_cache_free(cache, skb);
272 static void skb_drop_list(struct sk_buff **listp)
274 struct sk_buff *list = *listp;
279 struct sk_buff *this = list;
285 static inline void skb_drop_fraglist(struct sk_buff *skb)
287 skb_drop_list(&skb_shinfo(skb)->frag_list);
290 static void skb_clone_fraglist(struct sk_buff *skb)
292 struct sk_buff *list;
294 for (list = skb_shinfo(skb)->frag_list; list; list = list->next)
298 void skb_release_data(struct sk_buff *skb)
301 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
302 &skb_shinfo(skb)->dataref)) {
303 if (skb_shinfo(skb)->nr_frags) {
305 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
306 put_page(skb_shinfo(skb)->frags[i].page);
309 if (skb_shinfo(skb)->frag_list)
310 skb_drop_fraglist(skb);
317 * Free an skbuff by memory without cleaning the state.
319 void kfree_skbmem(struct sk_buff *skb)
321 struct sk_buff *other;
322 atomic_t *fclone_ref;
324 skb_release_data(skb);
325 switch (skb->fclone) {
326 case SKB_FCLONE_UNAVAILABLE:
327 kmem_cache_free(skbuff_head_cache, skb);
330 case SKB_FCLONE_ORIG:
331 fclone_ref = (atomic_t *) (skb + 2);
332 if (atomic_dec_and_test(fclone_ref))
333 kmem_cache_free(skbuff_fclone_cache, skb);
336 case SKB_FCLONE_CLONE:
337 fclone_ref = (atomic_t *) (skb + 1);
340 /* The clone portion is available for
341 * fast-cloning again.
343 skb->fclone = SKB_FCLONE_UNAVAILABLE;
345 if (atomic_dec_and_test(fclone_ref))
346 kmem_cache_free(skbuff_fclone_cache, other);
352 * __kfree_skb - private function
355 * Free an sk_buff. Release anything attached to the buffer.
356 * Clean the state. This is an internal helper function. Users should
357 * always call kfree_skb
360 void __kfree_skb(struct sk_buff *skb)
362 dst_release(skb->dst);
364 secpath_put(skb->sp);
366 if (skb->destructor) {
368 skb->destructor(skb);
370 #ifdef CONFIG_NETFILTER
371 nf_conntrack_put(skb->nfct);
372 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
373 nf_conntrack_put_reasm(skb->nfct_reasm);
375 #ifdef CONFIG_BRIDGE_NETFILTER
376 nf_bridge_put(skb->nf_bridge);
379 /* XXX: IS this still necessary? - JHS */
380 #ifdef CONFIG_NET_SCHED
382 #ifdef CONFIG_NET_CLS_ACT
391 * kfree_skb - free an sk_buff
392 * @skb: buffer to free
394 * Drop a reference to the buffer and free it if the usage count has
397 void kfree_skb(struct sk_buff *skb)
401 if (likely(atomic_read(&skb->users) == 1))
403 else if (likely(!atomic_dec_and_test(&skb->users)))
409 * skb_clone - duplicate an sk_buff
410 * @skb: buffer to clone
411 * @gfp_mask: allocation priority
413 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
414 * copies share the same packet data but not structure. The new
415 * buffer has a reference count of 1. If the allocation fails the
416 * function returns %NULL otherwise the new buffer is returned.
418 * If this function is called from an interrupt gfp_mask() must be
422 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
427 if (skb->fclone == SKB_FCLONE_ORIG &&
428 n->fclone == SKB_FCLONE_UNAVAILABLE) {
429 atomic_t *fclone_ref = (atomic_t *) (n + 1);
430 n->fclone = SKB_FCLONE_CLONE;
431 atomic_inc(fclone_ref);
433 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
436 n->fclone = SKB_FCLONE_UNAVAILABLE;
439 #define C(x) n->x = skb->x
441 n->next = n->prev = NULL;
452 secpath_get(skb->sp);
454 memcpy(n->cb, skb->cb, sizeof(skb->cb));
468 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
472 n->destructor = NULL;
473 #ifdef CONFIG_NETFILTER
476 nf_conntrack_get(skb->nfct);
478 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
480 nf_conntrack_get_reasm(skb->nfct_reasm);
482 #ifdef CONFIG_BRIDGE_NETFILTER
484 nf_bridge_get(skb->nf_bridge);
486 #endif /*CONFIG_NETFILTER*/
487 #ifdef CONFIG_NET_SCHED
489 #ifdef CONFIG_NET_CLS_ACT
490 n->tc_verd = SET_TC_VERD(skb->tc_verd,0);
491 n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd);
492 n->tc_verd = CLR_TC_MUNGED(n->tc_verd);
498 atomic_set(&n->users, 1);
504 atomic_inc(&(skb_shinfo(skb)->dataref));
510 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
513 * Shift between the two data areas in bytes
515 unsigned long offset = new->data - old->data;
519 new->priority = old->priority;
520 new->protocol = old->protocol;
521 new->dst = dst_clone(old->dst);
523 new->sp = secpath_get(old->sp);
525 new->h.raw = old->h.raw + offset;
526 new->nh.raw = old->nh.raw + offset;
527 new->mac.raw = old->mac.raw + offset;
528 memcpy(new->cb, old->cb, sizeof(old->cb));
529 new->local_df = old->local_df;
530 new->fclone = SKB_FCLONE_UNAVAILABLE;
531 new->pkt_type = old->pkt_type;
532 new->tstamp = old->tstamp;
533 new->destructor = NULL;
534 #ifdef CONFIG_NETFILTER
535 new->nfmark = old->nfmark;
536 new->nfct = old->nfct;
537 nf_conntrack_get(old->nfct);
538 new->nfctinfo = old->nfctinfo;
539 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
540 new->nfct_reasm = old->nfct_reasm;
541 nf_conntrack_get_reasm(old->nfct_reasm);
543 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
544 new->ipvs_property = old->ipvs_property;
546 #ifdef CONFIG_BRIDGE_NETFILTER
547 new->nf_bridge = old->nf_bridge;
548 nf_bridge_get(old->nf_bridge);
551 #ifdef CONFIG_NET_SCHED
552 #ifdef CONFIG_NET_CLS_ACT
553 new->tc_verd = old->tc_verd;
555 new->tc_index = old->tc_index;
557 atomic_set(&new->users, 1);
558 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
559 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
560 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
564 * skb_copy - create private copy of an sk_buff
565 * @skb: buffer to copy
566 * @gfp_mask: allocation priority
568 * Make a copy of both an &sk_buff and its data. This is used when the
569 * caller wishes to modify the data and needs a private copy of the
570 * data to alter. Returns %NULL on failure or the pointer to the buffer
571 * on success. The returned buffer has a reference count of 1.
573 * As by-product this function converts non-linear &sk_buff to linear
574 * one, so that &sk_buff becomes completely private and caller is allowed
575 * to modify all the data of returned buffer. This means that this
576 * function is not recommended for use in circumstances when only
577 * header is going to be modified. Use pskb_copy() instead.
580 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
582 int headerlen = skb->data - skb->head;
584 * Allocate the copy buffer
586 struct sk_buff *n = alloc_skb(skb->end - skb->head + skb->data_len,
591 /* Set the data pointer */
592 skb_reserve(n, headerlen);
593 /* Set the tail pointer and length */
594 skb_put(n, skb->len);
596 n->ip_summed = skb->ip_summed;
598 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
601 copy_skb_header(n, skb);
607 * pskb_copy - create copy of an sk_buff with private head.
608 * @skb: buffer to copy
609 * @gfp_mask: allocation priority
611 * Make a copy of both an &sk_buff and part of its data, located
612 * in header. Fragmented data remain shared. This is used when
613 * the caller wishes to modify only header of &sk_buff and needs
614 * private copy of the header to alter. Returns %NULL on failure
615 * or the pointer to the buffer on success.
616 * The returned buffer has a reference count of 1.
619 struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
622 * Allocate the copy buffer
624 struct sk_buff *n = alloc_skb(skb->end - skb->head, gfp_mask);
629 /* Set the data pointer */
630 skb_reserve(n, skb->data - skb->head);
631 /* Set the tail pointer and length */
632 skb_put(n, skb_headlen(skb));
634 memcpy(n->data, skb->data, n->len);
636 n->ip_summed = skb->ip_summed;
638 n->data_len = skb->data_len;
641 if (skb_shinfo(skb)->nr_frags) {
644 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
645 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
646 get_page(skb_shinfo(n)->frags[i].page);
648 skb_shinfo(n)->nr_frags = i;
651 if (skb_shinfo(skb)->frag_list) {
652 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
653 skb_clone_fraglist(n);
656 copy_skb_header(n, skb);
662 * pskb_expand_head - reallocate header of &sk_buff
663 * @skb: buffer to reallocate
664 * @nhead: room to add at head
665 * @ntail: room to add at tail
666 * @gfp_mask: allocation priority
668 * Expands (or creates identical copy, if &nhead and &ntail are zero)
669 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
670 * reference count of 1. Returns zero in the case of success or error,
671 * if expansion failed. In the last case, &sk_buff is not changed.
673 * All the pointers pointing into skb header may change and must be
674 * reloaded after call to this function.
677 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
682 int size = nhead + (skb->end - skb->head) + ntail;
688 size = SKB_DATA_ALIGN(size);
690 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
694 /* Copy only real data... and, alas, header. This should be
695 * optimized for the cases when header is void. */
696 memcpy(data + nhead, skb->head, skb->tail - skb->head);
697 memcpy(data + size, skb->end, sizeof(struct skb_shared_info));
699 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
700 get_page(skb_shinfo(skb)->frags[i].page);
702 if (skb_shinfo(skb)->frag_list)
703 skb_clone_fraglist(skb);
705 skb_release_data(skb);
707 off = (data + nhead) - skb->head;
710 skb->end = data + size;
718 atomic_set(&skb_shinfo(skb)->dataref, 1);
725 /* Make private copy of skb with writable head and some headroom */
727 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
729 struct sk_buff *skb2;
730 int delta = headroom - skb_headroom(skb);
733 skb2 = pskb_copy(skb, GFP_ATOMIC);
735 skb2 = skb_clone(skb, GFP_ATOMIC);
736 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
747 * skb_copy_expand - copy and expand sk_buff
748 * @skb: buffer to copy
749 * @newheadroom: new free bytes at head
750 * @newtailroom: new free bytes at tail
751 * @gfp_mask: allocation priority
753 * Make a copy of both an &sk_buff and its data and while doing so
754 * allocate additional space.
756 * This is used when the caller wishes to modify the data and needs a
757 * private copy of the data to alter as well as more space for new fields.
758 * Returns %NULL on failure or the pointer to the buffer
759 * on success. The returned buffer has a reference count of 1.
761 * You must pass %GFP_ATOMIC as the allocation priority if this function
762 * is called from an interrupt.
764 * BUG ALERT: ip_summed is not copied. Why does this work? Is it used
765 * only by netfilter in the cases when checksum is recalculated? --ANK
767 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
768 int newheadroom, int newtailroom,
772 * Allocate the copy buffer
774 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
776 int head_copy_len, head_copy_off;
781 skb_reserve(n, newheadroom);
783 /* Set the tail pointer and length */
784 skb_put(n, skb->len);
786 head_copy_len = skb_headroom(skb);
788 if (newheadroom <= head_copy_len)
789 head_copy_len = newheadroom;
791 head_copy_off = newheadroom - head_copy_len;
793 /* Copy the linear header and data. */
794 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
795 skb->len + head_copy_len))
798 copy_skb_header(n, skb);
804 * skb_pad - zero pad the tail of an skb
805 * @skb: buffer to pad
808 * Ensure that a buffer is followed by a padding area that is zero
809 * filled. Used by network drivers which may DMA or transfer data
810 * beyond the buffer end onto the wire.
812 * May return NULL in out of memory cases.
815 struct sk_buff *skb_pad(struct sk_buff *skb, int pad)
817 struct sk_buff *nskb;
819 /* If the skbuff is non linear tailroom is always zero.. */
820 if (skb_tailroom(skb) >= pad) {
821 memset(skb->data+skb->len, 0, pad);
825 nskb = skb_copy_expand(skb, skb_headroom(skb), skb_tailroom(skb) + pad, GFP_ATOMIC);
828 memset(nskb->data+nskb->len, 0, pad);
832 /* Trims skb to length len. It can change skb pointers.
835 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
837 struct sk_buff **fragp;
838 struct sk_buff *frag;
839 int offset = skb_headlen(skb);
840 int nfrags = skb_shinfo(skb)->nr_frags;
844 if (skb_cloned(skb) &&
845 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
852 for (; i < nfrags; i++) {
853 int end = offset + skb_shinfo(skb)->frags[i].size;
860 skb_shinfo(skb)->frags[i++].size = len - offset;
863 skb_shinfo(skb)->nr_frags = i;
865 for (; i < nfrags; i++)
866 put_page(skb_shinfo(skb)->frags[i].page);
868 if (skb_shinfo(skb)->frag_list)
869 skb_drop_fraglist(skb);
873 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
874 fragp = &frag->next) {
875 int end = offset + frag->len;
877 if (skb_shared(frag)) {
878 struct sk_buff *nfrag;
880 nfrag = skb_clone(frag, GFP_ATOMIC);
881 if (unlikely(!nfrag))
884 nfrag->next = frag->next;
896 unlikely((err = pskb_trim(frag, len - offset))))
900 skb_drop_list(&frag->next);
905 if (len > skb_headlen(skb)) {
906 skb->data_len -= skb->len - len;
911 skb->tail = skb->data + len;
918 * __pskb_pull_tail - advance tail of skb header
919 * @skb: buffer to reallocate
920 * @delta: number of bytes to advance tail
922 * The function makes a sense only on a fragmented &sk_buff,
923 * it expands header moving its tail forward and copying necessary
924 * data from fragmented part.
926 * &sk_buff MUST have reference count of 1.
928 * Returns %NULL (and &sk_buff does not change) if pull failed
929 * or value of new tail of skb in the case of success.
931 * All the pointers pointing into skb header may change and must be
932 * reloaded after call to this function.
935 /* Moves tail of skb head forward, copying data from fragmented part,
936 * when it is necessary.
937 * 1. It may fail due to malloc failure.
938 * 2. It may change skb pointers.
940 * It is pretty complicated. Luckily, it is called only in exceptional cases.
942 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
944 /* If skb has not enough free space at tail, get new one
945 * plus 128 bytes for future expansions. If we have enough
946 * room at tail, reallocate without expansion only if skb is cloned.
948 int i, k, eat = (skb->tail + delta) - skb->end;
950 if (eat > 0 || skb_cloned(skb)) {
951 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
956 if (skb_copy_bits(skb, skb_headlen(skb), skb->tail, delta))
959 /* Optimization: no fragments, no reasons to preestimate
960 * size of pulled pages. Superb.
962 if (!skb_shinfo(skb)->frag_list)
965 /* Estimate size of pulled pages. */
967 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
968 if (skb_shinfo(skb)->frags[i].size >= eat)
970 eat -= skb_shinfo(skb)->frags[i].size;
973 /* If we need update frag list, we are in troubles.
974 * Certainly, it possible to add an offset to skb data,
975 * but taking into account that pulling is expected to
976 * be very rare operation, it is worth to fight against
977 * further bloating skb head and crucify ourselves here instead.
978 * Pure masohism, indeed. 8)8)
981 struct sk_buff *list = skb_shinfo(skb)->frag_list;
982 struct sk_buff *clone = NULL;
983 struct sk_buff *insp = NULL;
988 if (list->len <= eat) {
989 /* Eaten as whole. */
994 /* Eaten partially. */
996 if (skb_shared(list)) {
997 /* Sucks! We need to fork list. :-( */
998 clone = skb_clone(list, GFP_ATOMIC);
1004 /* This may be pulled without
1008 if (!pskb_pull(list, eat)) {
1017 /* Free pulled out fragments. */
1018 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1019 skb_shinfo(skb)->frag_list = list->next;
1022 /* And insert new clone at head. */
1025 skb_shinfo(skb)->frag_list = clone;
1028 /* Success! Now we may commit changes to skb data. */
1033 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1034 if (skb_shinfo(skb)->frags[i].size <= eat) {
1035 put_page(skb_shinfo(skb)->frags[i].page);
1036 eat -= skb_shinfo(skb)->frags[i].size;
1038 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1040 skb_shinfo(skb)->frags[k].page_offset += eat;
1041 skb_shinfo(skb)->frags[k].size -= eat;
1047 skb_shinfo(skb)->nr_frags = k;
1050 skb->data_len -= delta;
1055 /* Copy some data bits from skb to kernel buffer. */
1057 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1060 int start = skb_headlen(skb);
1062 if (offset > (int)skb->len - len)
1066 if ((copy = start - offset) > 0) {
1069 memcpy(to, skb->data + offset, copy);
1070 if ((len -= copy) == 0)
1076 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1079 BUG_TRAP(start <= offset + len);
1081 end = start + skb_shinfo(skb)->frags[i].size;
1082 if ((copy = end - offset) > 0) {
1088 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1090 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1091 offset - start, copy);
1092 kunmap_skb_frag(vaddr);
1094 if ((len -= copy) == 0)
1102 if (skb_shinfo(skb)->frag_list) {
1103 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1105 for (; list; list = list->next) {
1108 BUG_TRAP(start <= offset + len);
1110 end = start + list->len;
1111 if ((copy = end - offset) > 0) {
1114 if (skb_copy_bits(list, offset - start,
1117 if ((len -= copy) == 0)
1133 * skb_store_bits - store bits from kernel buffer to skb
1134 * @skb: destination buffer
1135 * @offset: offset in destination
1136 * @from: source buffer
1137 * @len: number of bytes to copy
1139 * Copy the specified number of bytes from the source buffer to the
1140 * destination skb. This function handles all the messy bits of
1141 * traversing fragment lists and such.
1144 int skb_store_bits(const struct sk_buff *skb, int offset, void *from, int len)
1147 int start = skb_headlen(skb);
1149 if (offset > (int)skb->len - len)
1152 if ((copy = start - offset) > 0) {
1155 memcpy(skb->data + offset, from, copy);
1156 if ((len -= copy) == 0)
1162 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1163 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1166 BUG_TRAP(start <= offset + len);
1168 end = start + frag->size;
1169 if ((copy = end - offset) > 0) {
1175 vaddr = kmap_skb_frag(frag);
1176 memcpy(vaddr + frag->page_offset + offset - start,
1178 kunmap_skb_frag(vaddr);
1180 if ((len -= copy) == 0)
1188 if (skb_shinfo(skb)->frag_list) {
1189 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1191 for (; list; list = list->next) {
1194 BUG_TRAP(start <= offset + len);
1196 end = start + list->len;
1197 if ((copy = end - offset) > 0) {
1200 if (skb_store_bits(list, offset - start,
1203 if ((len -= copy) == 0)
1218 EXPORT_SYMBOL(skb_store_bits);
1220 /* Checksum skb data. */
1222 unsigned int skb_checksum(const struct sk_buff *skb, int offset,
1223 int len, unsigned int csum)
1225 int start = skb_headlen(skb);
1226 int i, copy = start - offset;
1229 /* Checksum header. */
1233 csum = csum_partial(skb->data + offset, copy, csum);
1234 if ((len -= copy) == 0)
1240 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1243 BUG_TRAP(start <= offset + len);
1245 end = start + skb_shinfo(skb)->frags[i].size;
1246 if ((copy = end - offset) > 0) {
1249 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1253 vaddr = kmap_skb_frag(frag);
1254 csum2 = csum_partial(vaddr + frag->page_offset +
1255 offset - start, copy, 0);
1256 kunmap_skb_frag(vaddr);
1257 csum = csum_block_add(csum, csum2, pos);
1266 if (skb_shinfo(skb)->frag_list) {
1267 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1269 for (; list; list = list->next) {
1272 BUG_TRAP(start <= offset + len);
1274 end = start + list->len;
1275 if ((copy = end - offset) > 0) {
1279 csum2 = skb_checksum(list, offset - start,
1281 csum = csum_block_add(csum, csum2, pos);
1282 if ((len -= copy) == 0)
1295 /* Both of above in one bottle. */
1297 unsigned int skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1298 u8 *to, int len, unsigned int csum)
1300 int start = skb_headlen(skb);
1301 int i, copy = start - offset;
1308 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1310 if ((len -= copy) == 0)
1317 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1320 BUG_TRAP(start <= offset + len);
1322 end = start + skb_shinfo(skb)->frags[i].size;
1323 if ((copy = end - offset) > 0) {
1326 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1330 vaddr = kmap_skb_frag(frag);
1331 csum2 = csum_partial_copy_nocheck(vaddr +
1335 kunmap_skb_frag(vaddr);
1336 csum = csum_block_add(csum, csum2, pos);
1346 if (skb_shinfo(skb)->frag_list) {
1347 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1349 for (; list; list = list->next) {
1353 BUG_TRAP(start <= offset + len);
1355 end = start + list->len;
1356 if ((copy = end - offset) > 0) {
1359 csum2 = skb_copy_and_csum_bits(list,
1362 csum = csum_block_add(csum, csum2, pos);
1363 if ((len -= copy) == 0)
1376 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1381 if (skb->ip_summed == CHECKSUM_HW)
1382 csstart = skb->h.raw - skb->data;
1384 csstart = skb_headlen(skb);
1386 BUG_ON(csstart > skb_headlen(skb));
1388 memcpy(to, skb->data, csstart);
1391 if (csstart != skb->len)
1392 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1393 skb->len - csstart, 0);
1395 if (skb->ip_summed == CHECKSUM_HW) {
1396 long csstuff = csstart + skb->csum;
1398 *((unsigned short *)(to + csstuff)) = csum_fold(csum);
1403 * skb_dequeue - remove from the head of the queue
1404 * @list: list to dequeue from
1406 * Remove the head of the list. The list lock is taken so the function
1407 * may be used safely with other locking list functions. The head item is
1408 * returned or %NULL if the list is empty.
1411 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1413 unsigned long flags;
1414 struct sk_buff *result;
1416 spin_lock_irqsave(&list->lock, flags);
1417 result = __skb_dequeue(list);
1418 spin_unlock_irqrestore(&list->lock, flags);
1423 * skb_dequeue_tail - remove from the tail of the queue
1424 * @list: list to dequeue from
1426 * Remove the tail of the list. The list lock is taken so the function
1427 * may be used safely with other locking list functions. The tail item is
1428 * returned or %NULL if the list is empty.
1430 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1432 unsigned long flags;
1433 struct sk_buff *result;
1435 spin_lock_irqsave(&list->lock, flags);
1436 result = __skb_dequeue_tail(list);
1437 spin_unlock_irqrestore(&list->lock, flags);
1442 * skb_queue_purge - empty a list
1443 * @list: list to empty
1445 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1446 * the list and one reference dropped. This function takes the list
1447 * lock and is atomic with respect to other list locking functions.
1449 void skb_queue_purge(struct sk_buff_head *list)
1451 struct sk_buff *skb;
1452 while ((skb = skb_dequeue(list)) != NULL)
1457 * skb_queue_head - queue a buffer at the list head
1458 * @list: list to use
1459 * @newsk: buffer to queue
1461 * Queue a buffer at the start of the list. This function takes the
1462 * list lock and can be used safely with other locking &sk_buff functions
1465 * A buffer cannot be placed on two lists at the same time.
1467 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1469 unsigned long flags;
1471 spin_lock_irqsave(&list->lock, flags);
1472 __skb_queue_head(list, newsk);
1473 spin_unlock_irqrestore(&list->lock, flags);
1477 * skb_queue_tail - queue a buffer at the list tail
1478 * @list: list to use
1479 * @newsk: buffer to queue
1481 * Queue a buffer at the tail of the list. This function takes the
1482 * list lock and can be used safely with other locking &sk_buff functions
1485 * A buffer cannot be placed on two lists at the same time.
1487 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1489 unsigned long flags;
1491 spin_lock_irqsave(&list->lock, flags);
1492 __skb_queue_tail(list, newsk);
1493 spin_unlock_irqrestore(&list->lock, flags);
1497 * skb_unlink - remove a buffer from a list
1498 * @skb: buffer to remove
1499 * @list: list to use
1501 * Remove a packet from a list. The list locks are taken and this
1502 * function is atomic with respect to other list locked calls
1504 * You must know what list the SKB is on.
1506 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1508 unsigned long flags;
1510 spin_lock_irqsave(&list->lock, flags);
1511 __skb_unlink(skb, list);
1512 spin_unlock_irqrestore(&list->lock, flags);
1516 * skb_append - append a buffer
1517 * @old: buffer to insert after
1518 * @newsk: buffer to insert
1519 * @list: list to use
1521 * Place a packet after a given packet in a list. The list locks are taken
1522 * and this function is atomic with respect to other list locked calls.
1523 * A buffer cannot be placed on two lists at the same time.
1525 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1527 unsigned long flags;
1529 spin_lock_irqsave(&list->lock, flags);
1530 __skb_append(old, newsk, list);
1531 spin_unlock_irqrestore(&list->lock, flags);
1536 * skb_insert - insert a buffer
1537 * @old: buffer to insert before
1538 * @newsk: buffer to insert
1539 * @list: list to use
1541 * Place a packet before a given packet in a list. The list locks are
1542 * taken and this function is atomic with respect to other list locked
1545 * A buffer cannot be placed on two lists at the same time.
1547 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1549 unsigned long flags;
1551 spin_lock_irqsave(&list->lock, flags);
1552 __skb_insert(newsk, old->prev, old, list);
1553 spin_unlock_irqrestore(&list->lock, flags);
1558 * Tune the memory allocator for a new MTU size.
1560 void skb_add_mtu(int mtu)
1562 /* Must match allocation in alloc_skb */
1563 mtu = SKB_DATA_ALIGN(mtu) + sizeof(struct skb_shared_info);
1565 kmem_add_cache_size(mtu);
1569 static inline void skb_split_inside_header(struct sk_buff *skb,
1570 struct sk_buff* skb1,
1571 const u32 len, const int pos)
1575 memcpy(skb_put(skb1, pos - len), skb->data + len, pos - len);
1577 /* And move data appendix as is. */
1578 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1579 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
1581 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
1582 skb_shinfo(skb)->nr_frags = 0;
1583 skb1->data_len = skb->data_len;
1584 skb1->len += skb1->data_len;
1587 skb->tail = skb->data + len;
1590 static inline void skb_split_no_header(struct sk_buff *skb,
1591 struct sk_buff* skb1,
1592 const u32 len, int pos)
1595 const int nfrags = skb_shinfo(skb)->nr_frags;
1597 skb_shinfo(skb)->nr_frags = 0;
1598 skb1->len = skb1->data_len = skb->len - len;
1600 skb->data_len = len - pos;
1602 for (i = 0; i < nfrags; i++) {
1603 int size = skb_shinfo(skb)->frags[i].size;
1605 if (pos + size > len) {
1606 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
1610 * We have two variants in this case:
1611 * 1. Move all the frag to the second
1612 * part, if it is possible. F.e.
1613 * this approach is mandatory for TUX,
1614 * where splitting is expensive.
1615 * 2. Split is accurately. We make this.
1617 get_page(skb_shinfo(skb)->frags[i].page);
1618 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
1619 skb_shinfo(skb1)->frags[0].size -= len - pos;
1620 skb_shinfo(skb)->frags[i].size = len - pos;
1621 skb_shinfo(skb)->nr_frags++;
1625 skb_shinfo(skb)->nr_frags++;
1628 skb_shinfo(skb1)->nr_frags = k;
1632 * skb_split - Split fragmented skb to two parts at length len.
1633 * @skb: the buffer to split
1634 * @skb1: the buffer to receive the second part
1635 * @len: new length for skb
1637 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
1639 int pos = skb_headlen(skb);
1641 if (len < pos) /* Split line is inside header. */
1642 skb_split_inside_header(skb, skb1, len, pos);
1643 else /* Second chunk has no header, nothing to copy. */
1644 skb_split_no_header(skb, skb1, len, pos);
1648 * skb_prepare_seq_read - Prepare a sequential read of skb data
1649 * @skb: the buffer to read
1650 * @from: lower offset of data to be read
1651 * @to: upper offset of data to be read
1652 * @st: state variable
1654 * Initializes the specified state variable. Must be called before
1655 * invoking skb_seq_read() for the first time.
1657 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1658 unsigned int to, struct skb_seq_state *st)
1660 st->lower_offset = from;
1661 st->upper_offset = to;
1662 st->root_skb = st->cur_skb = skb;
1663 st->frag_idx = st->stepped_offset = 0;
1664 st->frag_data = NULL;
1668 * skb_seq_read - Sequentially read skb data
1669 * @consumed: number of bytes consumed by the caller so far
1670 * @data: destination pointer for data to be returned
1671 * @st: state variable
1673 * Reads a block of skb data at &consumed relative to the
1674 * lower offset specified to skb_prepare_seq_read(). Assigns
1675 * the head of the data block to &data and returns the length
1676 * of the block or 0 if the end of the skb data or the upper
1677 * offset has been reached.
1679 * The caller is not required to consume all of the data
1680 * returned, i.e. &consumed is typically set to the number
1681 * of bytes already consumed and the next call to
1682 * skb_seq_read() will return the remaining part of the block.
1684 * Note: The size of each block of data returned can be arbitary,
1685 * this limitation is the cost for zerocopy seqeuental
1686 * reads of potentially non linear data.
1688 * Note: Fragment lists within fragments are not implemented
1689 * at the moment, state->root_skb could be replaced with
1690 * a stack for this purpose.
1692 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1693 struct skb_seq_state *st)
1695 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
1698 if (unlikely(abs_offset >= st->upper_offset))
1702 block_limit = skb_headlen(st->cur_skb);
1704 if (abs_offset < block_limit) {
1705 *data = st->cur_skb->data + abs_offset;
1706 return block_limit - abs_offset;
1709 if (st->frag_idx == 0 && !st->frag_data)
1710 st->stepped_offset += skb_headlen(st->cur_skb);
1712 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
1713 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
1714 block_limit = frag->size + st->stepped_offset;
1716 if (abs_offset < block_limit) {
1718 st->frag_data = kmap_skb_frag(frag);
1720 *data = (u8 *) st->frag_data + frag->page_offset +
1721 (abs_offset - st->stepped_offset);
1723 return block_limit - abs_offset;
1726 if (st->frag_data) {
1727 kunmap_skb_frag(st->frag_data);
1728 st->frag_data = NULL;
1732 st->stepped_offset += frag->size;
1735 if (st->cur_skb->next) {
1736 st->cur_skb = st->cur_skb->next;
1739 } else if (st->root_skb == st->cur_skb &&
1740 skb_shinfo(st->root_skb)->frag_list) {
1741 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
1749 * skb_abort_seq_read - Abort a sequential read of skb data
1750 * @st: state variable
1752 * Must be called if skb_seq_read() was not called until it
1755 void skb_abort_seq_read(struct skb_seq_state *st)
1758 kunmap_skb_frag(st->frag_data);
1761 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
1763 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
1764 struct ts_config *conf,
1765 struct ts_state *state)
1767 return skb_seq_read(offset, text, TS_SKB_CB(state));
1770 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
1772 skb_abort_seq_read(TS_SKB_CB(state));
1776 * skb_find_text - Find a text pattern in skb data
1777 * @skb: the buffer to look in
1778 * @from: search offset
1780 * @config: textsearch configuration
1781 * @state: uninitialized textsearch state variable
1783 * Finds a pattern in the skb data according to the specified
1784 * textsearch configuration. Use textsearch_next() to retrieve
1785 * subsequent occurrences of the pattern. Returns the offset
1786 * to the first occurrence or UINT_MAX if no match was found.
1788 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1789 unsigned int to, struct ts_config *config,
1790 struct ts_state *state)
1792 config->get_next_block = skb_ts_get_next_block;
1793 config->finish = skb_ts_finish;
1795 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
1797 return textsearch_find(config, state);
1801 * skb_append_datato_frags: - append the user data to a skb
1802 * @sk: sock structure
1803 * @skb: skb structure to be appened with user data.
1804 * @getfrag: call back function to be used for getting the user data
1805 * @from: pointer to user message iov
1806 * @length: length of the iov message
1808 * Description: This procedure append the user data in the fragment part
1809 * of the skb if any page alloc fails user this procedure returns -ENOMEM
1811 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
1812 int (*getfrag)(void *from, char *to, int offset,
1813 int len, int odd, struct sk_buff *skb),
1814 void *from, int length)
1817 skb_frag_t *frag = NULL;
1818 struct page *page = NULL;
1824 /* Return error if we don't have space for new frag */
1825 frg_cnt = skb_shinfo(skb)->nr_frags;
1826 if (frg_cnt >= MAX_SKB_FRAGS)
1829 /* allocate a new page for next frag */
1830 page = alloc_pages(sk->sk_allocation, 0);
1832 /* If alloc_page fails just return failure and caller will
1833 * free previous allocated pages by doing kfree_skb()
1838 /* initialize the next frag */
1839 sk->sk_sndmsg_page = page;
1840 sk->sk_sndmsg_off = 0;
1841 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
1842 skb->truesize += PAGE_SIZE;
1843 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1845 /* get the new initialized frag */
1846 frg_cnt = skb_shinfo(skb)->nr_frags;
1847 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
1849 /* copy the user data to page */
1850 left = PAGE_SIZE - frag->page_offset;
1851 copy = (length > left)? left : length;
1853 ret = getfrag(from, (page_address(frag->page) +
1854 frag->page_offset + frag->size),
1855 offset, copy, 0, skb);
1859 /* copy was successful so update the size parameters */
1860 sk->sk_sndmsg_off += copy;
1863 skb->data_len += copy;
1867 } while (length > 0);
1873 * skb_pull_rcsum - pull skb and update receive checksum
1874 * @skb: buffer to update
1875 * @start: start of data before pull
1876 * @len: length of data pulled
1878 * This function performs an skb_pull on the packet and updates
1879 * update the CHECKSUM_HW checksum. It should be used on receive
1880 * path processing instead of skb_pull unless you know that the
1881 * checksum difference is zero (e.g., a valid IP header) or you
1882 * are setting ip_summed to CHECKSUM_NONE.
1884 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
1886 BUG_ON(len > skb->len);
1888 BUG_ON(skb->len < skb->data_len);
1889 skb_postpull_rcsum(skb, skb->data, len);
1890 return skb->data += len;
1893 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
1896 * skb_segment - Perform protocol segmentation on skb.
1897 * @skb: buffer to segment
1898 * @features: features for the output path (see dev->features)
1900 * This function performs segmentation on the given skb. It returns
1901 * the segment at the given position. It returns NULL if there are
1902 * no more segments to generate, or when an error is encountered.
1904 struct sk_buff *skb_segment(struct sk_buff *skb, int features)
1906 struct sk_buff *segs = NULL;
1907 struct sk_buff *tail = NULL;
1908 unsigned int mss = skb_shinfo(skb)->gso_size;
1909 unsigned int doffset = skb->data - skb->mac.raw;
1910 unsigned int offset = doffset;
1911 unsigned int headroom;
1913 int sg = features & NETIF_F_SG;
1914 int nfrags = skb_shinfo(skb)->nr_frags;
1919 __skb_push(skb, doffset);
1920 headroom = skb_headroom(skb);
1921 pos = skb_headlen(skb);
1924 struct sk_buff *nskb;
1930 len = skb->len - offset;
1934 hsize = skb_headlen(skb) - offset;
1937 nsize = hsize + doffset;
1938 if (nsize > len + doffset || !sg)
1939 nsize = len + doffset;
1941 nskb = alloc_skb(nsize + headroom, GFP_ATOMIC);
1942 if (unlikely(!nskb))
1951 nskb->dev = skb->dev;
1952 nskb->priority = skb->priority;
1953 nskb->protocol = skb->protocol;
1954 nskb->dst = dst_clone(skb->dst);
1955 memcpy(nskb->cb, skb->cb, sizeof(skb->cb));
1956 nskb->pkt_type = skb->pkt_type;
1957 nskb->mac_len = skb->mac_len;
1959 skb_reserve(nskb, headroom);
1960 nskb->mac.raw = nskb->data;
1961 nskb->nh.raw = nskb->data + skb->mac_len;
1962 nskb->h.raw = nskb->nh.raw + (skb->h.raw - skb->nh.raw);
1963 memcpy(skb_put(nskb, doffset), skb->data, doffset);
1966 nskb->csum = skb_copy_and_csum_bits(skb, offset,
1972 frag = skb_shinfo(nskb)->frags;
1975 nskb->ip_summed = CHECKSUM_HW;
1976 nskb->csum = skb->csum;
1977 memcpy(skb_put(nskb, hsize), skb->data + offset, hsize);
1979 while (pos < offset + len) {
1980 BUG_ON(i >= nfrags);
1982 *frag = skb_shinfo(skb)->frags[i];
1983 get_page(frag->page);
1987 frag->page_offset += offset - pos;
1988 frag->size -= offset - pos;
1993 if (pos + size <= offset + len) {
1997 frag->size -= pos + size - (offset + len);
2004 skb_shinfo(nskb)->nr_frags = k;
2005 nskb->data_len = len - hsize;
2006 nskb->len += nskb->data_len;
2007 nskb->truesize += nskb->data_len;
2008 } while ((offset += len) < skb->len);
2013 while ((skb = segs)) {
2017 return ERR_PTR(err);
2020 EXPORT_SYMBOL_GPL(skb_segment);
2022 void __init skb_init(void)
2024 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2025 sizeof(struct sk_buff),
2029 if (!skbuff_head_cache)
2030 panic("cannot create skbuff cache");
2032 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2033 (2*sizeof(struct sk_buff)) +
2038 if (!skbuff_fclone_cache)
2039 panic("cannot create skbuff cache");
2042 EXPORT_SYMBOL(___pskb_trim);
2043 EXPORT_SYMBOL(__kfree_skb);
2044 EXPORT_SYMBOL(kfree_skb);
2045 EXPORT_SYMBOL(__pskb_pull_tail);
2046 EXPORT_SYMBOL(__alloc_skb);
2047 EXPORT_SYMBOL(pskb_copy);
2048 EXPORT_SYMBOL(pskb_expand_head);
2049 EXPORT_SYMBOL(skb_checksum);
2050 EXPORT_SYMBOL(skb_clone);
2051 EXPORT_SYMBOL(skb_clone_fraglist);
2052 EXPORT_SYMBOL(skb_copy);
2053 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2054 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2055 EXPORT_SYMBOL(skb_copy_bits);
2056 EXPORT_SYMBOL(skb_copy_expand);
2057 EXPORT_SYMBOL(skb_over_panic);
2058 EXPORT_SYMBOL(skb_pad);
2059 EXPORT_SYMBOL(skb_realloc_headroom);
2060 EXPORT_SYMBOL(skb_under_panic);
2061 EXPORT_SYMBOL(skb_dequeue);
2062 EXPORT_SYMBOL(skb_dequeue_tail);
2063 EXPORT_SYMBOL(skb_insert);
2064 EXPORT_SYMBOL(skb_queue_purge);
2065 EXPORT_SYMBOL(skb_queue_head);
2066 EXPORT_SYMBOL(skb_queue_tail);
2067 EXPORT_SYMBOL(skb_unlink);
2068 EXPORT_SYMBOL(skb_append);
2069 EXPORT_SYMBOL(skb_split);
2070 EXPORT_SYMBOL(skb_prepare_seq_read);
2071 EXPORT_SYMBOL(skb_seq_read);
2072 EXPORT_SYMBOL(skb_abort_seq_read);
2073 EXPORT_SYMBOL(skb_find_text);
2074 EXPORT_SYMBOL(skb_append_datato_frags);