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);
497 #if defined(CONFIG_VNET) || defined(CONFIG_VNET_MODULE)
500 #if defined(CONFIG_VNET) || defined(CONFIG_VNET_MODULE)
504 atomic_set(&n->users, 1);
510 atomic_inc(&(skb_shinfo(skb)->dataref));
516 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
519 * Shift between the two data areas in bytes
521 unsigned long offset = new->data - old->data;
525 new->priority = old->priority;
526 new->protocol = old->protocol;
527 new->dst = dst_clone(old->dst);
529 new->sp = secpath_get(old->sp);
531 new->h.raw = old->h.raw + offset;
532 new->nh.raw = old->nh.raw + offset;
533 new->mac.raw = old->mac.raw + offset;
534 memcpy(new->cb, old->cb, sizeof(old->cb));
535 new->local_df = old->local_df;
536 new->fclone = SKB_FCLONE_UNAVAILABLE;
537 new->pkt_type = old->pkt_type;
538 new->tstamp = old->tstamp;
539 new->destructor = NULL;
540 #ifdef CONFIG_NETFILTER
541 new->nfmark = old->nfmark;
542 new->nfct = old->nfct;
543 nf_conntrack_get(old->nfct);
544 new->nfctinfo = old->nfctinfo;
545 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
546 new->nfct_reasm = old->nfct_reasm;
547 nf_conntrack_get_reasm(old->nfct_reasm);
549 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
550 new->ipvs_property = old->ipvs_property;
552 #ifdef CONFIG_BRIDGE_NETFILTER
553 new->nf_bridge = old->nf_bridge;
554 nf_bridge_get(old->nf_bridge);
557 #ifdef CONFIG_NET_SCHED
558 #ifdef CONFIG_NET_CLS_ACT
559 new->tc_verd = old->tc_verd;
561 new->tc_index = old->tc_index;
563 #if defined(CONFIG_VNET) || defined(CONFIG_VNET_MODULE)
566 atomic_set(&new->users, 1);
567 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
568 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
569 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
573 * skb_copy - create private copy of an sk_buff
574 * @skb: buffer to copy
575 * @gfp_mask: allocation priority
577 * Make a copy of both an &sk_buff and its data. This is used when the
578 * caller wishes to modify the data and needs a private copy of the
579 * data to alter. Returns %NULL on failure or the pointer to the buffer
580 * on success. The returned buffer has a reference count of 1.
582 * As by-product this function converts non-linear &sk_buff to linear
583 * one, so that &sk_buff becomes completely private and caller is allowed
584 * to modify all the data of returned buffer. This means that this
585 * function is not recommended for use in circumstances when only
586 * header is going to be modified. Use pskb_copy() instead.
589 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
591 int headerlen = skb->data - skb->head;
593 * Allocate the copy buffer
595 struct sk_buff *n = alloc_skb(skb->end - skb->head + skb->data_len,
600 /* Set the data pointer */
601 skb_reserve(n, headerlen);
602 /* Set the tail pointer and length */
603 skb_put(n, skb->len);
605 n->ip_summed = skb->ip_summed;
607 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
610 copy_skb_header(n, skb);
616 * pskb_copy - create copy of an sk_buff with private head.
617 * @skb: buffer to copy
618 * @gfp_mask: allocation priority
620 * Make a copy of both an &sk_buff and part of its data, located
621 * in header. Fragmented data remain shared. This is used when
622 * the caller wishes to modify only header of &sk_buff and needs
623 * private copy of the header to alter. Returns %NULL on failure
624 * or the pointer to the buffer on success.
625 * The returned buffer has a reference count of 1.
628 struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
631 * Allocate the copy buffer
633 struct sk_buff *n = alloc_skb(skb->end - skb->head, gfp_mask);
638 /* Set the data pointer */
639 skb_reserve(n, skb->data - skb->head);
640 /* Set the tail pointer and length */
641 skb_put(n, skb_headlen(skb));
643 memcpy(n->data, skb->data, n->len);
645 n->ip_summed = skb->ip_summed;
647 n->data_len = skb->data_len;
650 if (skb_shinfo(skb)->nr_frags) {
653 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
654 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
655 get_page(skb_shinfo(n)->frags[i].page);
657 skb_shinfo(n)->nr_frags = i;
660 if (skb_shinfo(skb)->frag_list) {
661 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
662 skb_clone_fraglist(n);
665 copy_skb_header(n, skb);
671 * pskb_expand_head - reallocate header of &sk_buff
672 * @skb: buffer to reallocate
673 * @nhead: room to add at head
674 * @ntail: room to add at tail
675 * @gfp_mask: allocation priority
677 * Expands (or creates identical copy, if &nhead and &ntail are zero)
678 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
679 * reference count of 1. Returns zero in the case of success or error,
680 * if expansion failed. In the last case, &sk_buff is not changed.
682 * All the pointers pointing into skb header may change and must be
683 * reloaded after call to this function.
686 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
691 int size = nhead + (skb->end - skb->head) + ntail;
697 size = SKB_DATA_ALIGN(size);
699 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
703 /* Copy only real data... and, alas, header. This should be
704 * optimized for the cases when header is void. */
705 memcpy(data + nhead, skb->head, skb->tail - skb->head);
706 memcpy(data + size, skb->end, sizeof(struct skb_shared_info));
708 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
709 get_page(skb_shinfo(skb)->frags[i].page);
711 if (skb_shinfo(skb)->frag_list)
712 skb_clone_fraglist(skb);
714 skb_release_data(skb);
716 off = (data + nhead) - skb->head;
719 skb->end = data + size;
727 atomic_set(&skb_shinfo(skb)->dataref, 1);
734 /* Make private copy of skb with writable head and some headroom */
736 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
738 struct sk_buff *skb2;
739 int delta = headroom - skb_headroom(skb);
742 skb2 = pskb_copy(skb, GFP_ATOMIC);
744 skb2 = skb_clone(skb, GFP_ATOMIC);
745 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
756 * skb_copy_expand - copy and expand sk_buff
757 * @skb: buffer to copy
758 * @newheadroom: new free bytes at head
759 * @newtailroom: new free bytes at tail
760 * @gfp_mask: allocation priority
762 * Make a copy of both an &sk_buff and its data and while doing so
763 * allocate additional space.
765 * This is used when the caller wishes to modify the data and needs a
766 * private copy of the data to alter as well as more space for new fields.
767 * Returns %NULL on failure or the pointer to the buffer
768 * on success. The returned buffer has a reference count of 1.
770 * You must pass %GFP_ATOMIC as the allocation priority if this function
771 * is called from an interrupt.
773 * BUG ALERT: ip_summed is not copied. Why does this work? Is it used
774 * only by netfilter in the cases when checksum is recalculated? --ANK
776 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
777 int newheadroom, int newtailroom,
781 * Allocate the copy buffer
783 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
785 int head_copy_len, head_copy_off;
790 skb_reserve(n, newheadroom);
792 /* Set the tail pointer and length */
793 skb_put(n, skb->len);
795 head_copy_len = skb_headroom(skb);
797 if (newheadroom <= head_copy_len)
798 head_copy_len = newheadroom;
800 head_copy_off = newheadroom - head_copy_len;
802 /* Copy the linear header and data. */
803 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
804 skb->len + head_copy_len))
807 copy_skb_header(n, skb);
813 * skb_pad - zero pad the tail of an skb
814 * @skb: buffer to pad
817 * Ensure that a buffer is followed by a padding area that is zero
818 * filled. Used by network drivers which may DMA or transfer data
819 * beyond the buffer end onto the wire.
821 * May return NULL in out of memory cases.
824 struct sk_buff *skb_pad(struct sk_buff *skb, int pad)
826 struct sk_buff *nskb;
828 /* If the skbuff is non linear tailroom is always zero.. */
829 if (skb_tailroom(skb) >= pad) {
830 memset(skb->data+skb->len, 0, pad);
834 nskb = skb_copy_expand(skb, skb_headroom(skb), skb_tailroom(skb) + pad, GFP_ATOMIC);
837 memset(nskb->data+nskb->len, 0, pad);
841 /* Trims skb to length len. It can change skb pointers.
844 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
846 struct sk_buff **fragp;
847 struct sk_buff *frag;
848 int offset = skb_headlen(skb);
849 int nfrags = skb_shinfo(skb)->nr_frags;
853 if (skb_cloned(skb) &&
854 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
861 for (; i < nfrags; i++) {
862 int end = offset + skb_shinfo(skb)->frags[i].size;
869 skb_shinfo(skb)->frags[i++].size = len - offset;
872 skb_shinfo(skb)->nr_frags = i;
874 for (; i < nfrags; i++)
875 put_page(skb_shinfo(skb)->frags[i].page);
877 if (skb_shinfo(skb)->frag_list)
878 skb_drop_fraglist(skb);
882 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
883 fragp = &frag->next) {
884 int end = offset + frag->len;
886 if (skb_shared(frag)) {
887 struct sk_buff *nfrag;
889 nfrag = skb_clone(frag, GFP_ATOMIC);
890 if (unlikely(!nfrag))
893 nfrag->next = frag->next;
905 unlikely((err = pskb_trim(frag, len - offset))))
909 skb_drop_list(&frag->next);
914 if (len > skb_headlen(skb)) {
915 skb->data_len -= skb->len - len;
920 skb->tail = skb->data + len;
927 * __pskb_pull_tail - advance tail of skb header
928 * @skb: buffer to reallocate
929 * @delta: number of bytes to advance tail
931 * The function makes a sense only on a fragmented &sk_buff,
932 * it expands header moving its tail forward and copying necessary
933 * data from fragmented part.
935 * &sk_buff MUST have reference count of 1.
937 * Returns %NULL (and &sk_buff does not change) if pull failed
938 * or value of new tail of skb in the case of success.
940 * All the pointers pointing into skb header may change and must be
941 * reloaded after call to this function.
944 /* Moves tail of skb head forward, copying data from fragmented part,
945 * when it is necessary.
946 * 1. It may fail due to malloc failure.
947 * 2. It may change skb pointers.
949 * It is pretty complicated. Luckily, it is called only in exceptional cases.
951 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
953 /* If skb has not enough free space at tail, get new one
954 * plus 128 bytes for future expansions. If we have enough
955 * room at tail, reallocate without expansion only if skb is cloned.
957 int i, k, eat = (skb->tail + delta) - skb->end;
959 if (eat > 0 || skb_cloned(skb)) {
960 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
965 if (skb_copy_bits(skb, skb_headlen(skb), skb->tail, delta))
968 /* Optimization: no fragments, no reasons to preestimate
969 * size of pulled pages. Superb.
971 if (!skb_shinfo(skb)->frag_list)
974 /* Estimate size of pulled pages. */
976 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
977 if (skb_shinfo(skb)->frags[i].size >= eat)
979 eat -= skb_shinfo(skb)->frags[i].size;
982 /* If we need update frag list, we are in troubles.
983 * Certainly, it possible to add an offset to skb data,
984 * but taking into account that pulling is expected to
985 * be very rare operation, it is worth to fight against
986 * further bloating skb head and crucify ourselves here instead.
987 * Pure masohism, indeed. 8)8)
990 struct sk_buff *list = skb_shinfo(skb)->frag_list;
991 struct sk_buff *clone = NULL;
992 struct sk_buff *insp = NULL;
997 if (list->len <= eat) {
998 /* Eaten as whole. */
1003 /* Eaten partially. */
1005 if (skb_shared(list)) {
1006 /* Sucks! We need to fork list. :-( */
1007 clone = skb_clone(list, GFP_ATOMIC);
1013 /* This may be pulled without
1017 if (!pskb_pull(list, eat)) {
1026 /* Free pulled out fragments. */
1027 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1028 skb_shinfo(skb)->frag_list = list->next;
1031 /* And insert new clone at head. */
1034 skb_shinfo(skb)->frag_list = clone;
1037 /* Success! Now we may commit changes to skb data. */
1042 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1043 if (skb_shinfo(skb)->frags[i].size <= eat) {
1044 put_page(skb_shinfo(skb)->frags[i].page);
1045 eat -= skb_shinfo(skb)->frags[i].size;
1047 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1049 skb_shinfo(skb)->frags[k].page_offset += eat;
1050 skb_shinfo(skb)->frags[k].size -= eat;
1056 skb_shinfo(skb)->nr_frags = k;
1059 skb->data_len -= delta;
1064 /* Copy some data bits from skb to kernel buffer. */
1066 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1069 int start = skb_headlen(skb);
1071 if (offset > (int)skb->len - len)
1075 if ((copy = start - offset) > 0) {
1078 memcpy(to, skb->data + offset, copy);
1079 if ((len -= copy) == 0)
1085 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1088 BUG_TRAP(start <= offset + len);
1090 end = start + skb_shinfo(skb)->frags[i].size;
1091 if ((copy = end - offset) > 0) {
1097 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1099 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1100 offset - start, copy);
1101 kunmap_skb_frag(vaddr);
1103 if ((len -= copy) == 0)
1111 if (skb_shinfo(skb)->frag_list) {
1112 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1114 for (; list; list = list->next) {
1117 BUG_TRAP(start <= offset + len);
1119 end = start + list->len;
1120 if ((copy = end - offset) > 0) {
1123 if (skb_copy_bits(list, offset - start,
1126 if ((len -= copy) == 0)
1142 * skb_store_bits - store bits from kernel buffer to skb
1143 * @skb: destination buffer
1144 * @offset: offset in destination
1145 * @from: source buffer
1146 * @len: number of bytes to copy
1148 * Copy the specified number of bytes from the source buffer to the
1149 * destination skb. This function handles all the messy bits of
1150 * traversing fragment lists and such.
1153 int skb_store_bits(const struct sk_buff *skb, int offset, void *from, int len)
1156 int start = skb_headlen(skb);
1158 if (offset > (int)skb->len - len)
1161 if ((copy = start - offset) > 0) {
1164 memcpy(skb->data + offset, from, copy);
1165 if ((len -= copy) == 0)
1171 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1172 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1175 BUG_TRAP(start <= offset + len);
1177 end = start + frag->size;
1178 if ((copy = end - offset) > 0) {
1184 vaddr = kmap_skb_frag(frag);
1185 memcpy(vaddr + frag->page_offset + offset - start,
1187 kunmap_skb_frag(vaddr);
1189 if ((len -= copy) == 0)
1197 if (skb_shinfo(skb)->frag_list) {
1198 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1200 for (; list; list = list->next) {
1203 BUG_TRAP(start <= offset + len);
1205 end = start + list->len;
1206 if ((copy = end - offset) > 0) {
1209 if (skb_store_bits(list, offset - start,
1212 if ((len -= copy) == 0)
1227 EXPORT_SYMBOL(skb_store_bits);
1229 /* Checksum skb data. */
1231 unsigned int skb_checksum(const struct sk_buff *skb, int offset,
1232 int len, unsigned int csum)
1234 int start = skb_headlen(skb);
1235 int i, copy = start - offset;
1238 /* Checksum header. */
1242 csum = csum_partial(skb->data + offset, copy, csum);
1243 if ((len -= copy) == 0)
1249 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1252 BUG_TRAP(start <= offset + len);
1254 end = start + skb_shinfo(skb)->frags[i].size;
1255 if ((copy = end - offset) > 0) {
1258 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1262 vaddr = kmap_skb_frag(frag);
1263 csum2 = csum_partial(vaddr + frag->page_offset +
1264 offset - start, copy, 0);
1265 kunmap_skb_frag(vaddr);
1266 csum = csum_block_add(csum, csum2, pos);
1275 if (skb_shinfo(skb)->frag_list) {
1276 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1278 for (; list; list = list->next) {
1281 BUG_TRAP(start <= offset + len);
1283 end = start + list->len;
1284 if ((copy = end - offset) > 0) {
1288 csum2 = skb_checksum(list, offset - start,
1290 csum = csum_block_add(csum, csum2, pos);
1291 if ((len -= copy) == 0)
1304 /* Both of above in one bottle. */
1306 unsigned int skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1307 u8 *to, int len, unsigned int csum)
1309 int start = skb_headlen(skb);
1310 int i, copy = start - offset;
1317 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1319 if ((len -= copy) == 0)
1326 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1329 BUG_TRAP(start <= offset + len);
1331 end = start + skb_shinfo(skb)->frags[i].size;
1332 if ((copy = end - offset) > 0) {
1335 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1339 vaddr = kmap_skb_frag(frag);
1340 csum2 = csum_partial_copy_nocheck(vaddr +
1344 kunmap_skb_frag(vaddr);
1345 csum = csum_block_add(csum, csum2, pos);
1355 if (skb_shinfo(skb)->frag_list) {
1356 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1358 for (; list; list = list->next) {
1362 BUG_TRAP(start <= offset + len);
1364 end = start + list->len;
1365 if ((copy = end - offset) > 0) {
1368 csum2 = skb_copy_and_csum_bits(list,
1371 csum = csum_block_add(csum, csum2, pos);
1372 if ((len -= copy) == 0)
1385 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1390 if (skb->ip_summed == CHECKSUM_HW)
1391 csstart = skb->h.raw - skb->data;
1393 csstart = skb_headlen(skb);
1395 BUG_ON(csstart > skb_headlen(skb));
1397 memcpy(to, skb->data, csstart);
1400 if (csstart != skb->len)
1401 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1402 skb->len - csstart, 0);
1404 if (skb->ip_summed == CHECKSUM_HW) {
1405 long csstuff = csstart + skb->csum;
1407 *((unsigned short *)(to + csstuff)) = csum_fold(csum);
1412 * skb_dequeue - remove from the head of the queue
1413 * @list: list to dequeue from
1415 * Remove the head of the list. The list lock is taken so the function
1416 * may be used safely with other locking list functions. The head item is
1417 * returned or %NULL if the list is empty.
1420 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1422 unsigned long flags;
1423 struct sk_buff *result;
1425 spin_lock_irqsave(&list->lock, flags);
1426 result = __skb_dequeue(list);
1427 spin_unlock_irqrestore(&list->lock, flags);
1432 * skb_dequeue_tail - remove from the tail of the queue
1433 * @list: list to dequeue from
1435 * Remove the tail of the list. The list lock is taken so the function
1436 * may be used safely with other locking list functions. The tail item is
1437 * returned or %NULL if the list is empty.
1439 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1441 unsigned long flags;
1442 struct sk_buff *result;
1444 spin_lock_irqsave(&list->lock, flags);
1445 result = __skb_dequeue_tail(list);
1446 spin_unlock_irqrestore(&list->lock, flags);
1451 * skb_queue_purge - empty a list
1452 * @list: list to empty
1454 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1455 * the list and one reference dropped. This function takes the list
1456 * lock and is atomic with respect to other list locking functions.
1458 void skb_queue_purge(struct sk_buff_head *list)
1460 struct sk_buff *skb;
1461 while ((skb = skb_dequeue(list)) != NULL)
1466 * skb_queue_head - queue a buffer at the list head
1467 * @list: list to use
1468 * @newsk: buffer to queue
1470 * Queue a buffer at the start of the list. This function takes the
1471 * list lock and can be used safely with other locking &sk_buff functions
1474 * A buffer cannot be placed on two lists at the same time.
1476 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1478 unsigned long flags;
1480 spin_lock_irqsave(&list->lock, flags);
1481 __skb_queue_head(list, newsk);
1482 spin_unlock_irqrestore(&list->lock, flags);
1486 * skb_queue_tail - queue a buffer at the list tail
1487 * @list: list to use
1488 * @newsk: buffer to queue
1490 * Queue a buffer at the tail of the list. This function takes the
1491 * list lock and can be used safely with other locking &sk_buff functions
1494 * A buffer cannot be placed on two lists at the same time.
1496 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1498 unsigned long flags;
1500 spin_lock_irqsave(&list->lock, flags);
1501 __skb_queue_tail(list, newsk);
1502 spin_unlock_irqrestore(&list->lock, flags);
1506 * skb_unlink - remove a buffer from a list
1507 * @skb: buffer to remove
1508 * @list: list to use
1510 * Remove a packet from a list. The list locks are taken and this
1511 * function is atomic with respect to other list locked calls
1513 * You must know what list the SKB is on.
1515 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1517 unsigned long flags;
1519 spin_lock_irqsave(&list->lock, flags);
1520 __skb_unlink(skb, list);
1521 spin_unlock_irqrestore(&list->lock, flags);
1525 * skb_append - append a buffer
1526 * @old: buffer to insert after
1527 * @newsk: buffer to insert
1528 * @list: list to use
1530 * Place a packet after a given packet in a list. The list locks are taken
1531 * and this function is atomic with respect to other list locked calls.
1532 * A buffer cannot be placed on two lists at the same time.
1534 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1536 unsigned long flags;
1538 spin_lock_irqsave(&list->lock, flags);
1539 __skb_append(old, newsk, list);
1540 spin_unlock_irqrestore(&list->lock, flags);
1545 * skb_insert - insert a buffer
1546 * @old: buffer to insert before
1547 * @newsk: buffer to insert
1548 * @list: list to use
1550 * Place a packet before a given packet in a list. The list locks are
1551 * taken and this function is atomic with respect to other list locked
1554 * A buffer cannot be placed on two lists at the same time.
1556 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1558 unsigned long flags;
1560 spin_lock_irqsave(&list->lock, flags);
1561 __skb_insert(newsk, old->prev, old, list);
1562 spin_unlock_irqrestore(&list->lock, flags);
1567 * Tune the memory allocator for a new MTU size.
1569 void skb_add_mtu(int mtu)
1571 /* Must match allocation in alloc_skb */
1572 mtu = SKB_DATA_ALIGN(mtu) + sizeof(struct skb_shared_info);
1574 kmem_add_cache_size(mtu);
1578 static inline void skb_split_inside_header(struct sk_buff *skb,
1579 struct sk_buff* skb1,
1580 const u32 len, const int pos)
1584 memcpy(skb_put(skb1, pos - len), skb->data + len, pos - len);
1586 /* And move data appendix as is. */
1587 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1588 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
1590 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
1591 skb_shinfo(skb)->nr_frags = 0;
1592 skb1->data_len = skb->data_len;
1593 skb1->len += skb1->data_len;
1596 skb->tail = skb->data + len;
1599 static inline void skb_split_no_header(struct sk_buff *skb,
1600 struct sk_buff* skb1,
1601 const u32 len, int pos)
1604 const int nfrags = skb_shinfo(skb)->nr_frags;
1606 skb_shinfo(skb)->nr_frags = 0;
1607 skb1->len = skb1->data_len = skb->len - len;
1609 skb->data_len = len - pos;
1611 for (i = 0; i < nfrags; i++) {
1612 int size = skb_shinfo(skb)->frags[i].size;
1614 if (pos + size > len) {
1615 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
1619 * We have two variants in this case:
1620 * 1. Move all the frag to the second
1621 * part, if it is possible. F.e.
1622 * this approach is mandatory for TUX,
1623 * where splitting is expensive.
1624 * 2. Split is accurately. We make this.
1626 get_page(skb_shinfo(skb)->frags[i].page);
1627 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
1628 skb_shinfo(skb1)->frags[0].size -= len - pos;
1629 skb_shinfo(skb)->frags[i].size = len - pos;
1630 skb_shinfo(skb)->nr_frags++;
1634 skb_shinfo(skb)->nr_frags++;
1637 skb_shinfo(skb1)->nr_frags = k;
1641 * skb_split - Split fragmented skb to two parts at length len.
1642 * @skb: the buffer to split
1643 * @skb1: the buffer to receive the second part
1644 * @len: new length for skb
1646 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
1648 int pos = skb_headlen(skb);
1650 if (len < pos) /* Split line is inside header. */
1651 skb_split_inside_header(skb, skb1, len, pos);
1652 else /* Second chunk has no header, nothing to copy. */
1653 skb_split_no_header(skb, skb1, len, pos);
1657 * skb_prepare_seq_read - Prepare a sequential read of skb data
1658 * @skb: the buffer to read
1659 * @from: lower offset of data to be read
1660 * @to: upper offset of data to be read
1661 * @st: state variable
1663 * Initializes the specified state variable. Must be called before
1664 * invoking skb_seq_read() for the first time.
1666 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1667 unsigned int to, struct skb_seq_state *st)
1669 st->lower_offset = from;
1670 st->upper_offset = to;
1671 st->root_skb = st->cur_skb = skb;
1672 st->frag_idx = st->stepped_offset = 0;
1673 st->frag_data = NULL;
1677 * skb_seq_read - Sequentially read skb data
1678 * @consumed: number of bytes consumed by the caller so far
1679 * @data: destination pointer for data to be returned
1680 * @st: state variable
1682 * Reads a block of skb data at &consumed relative to the
1683 * lower offset specified to skb_prepare_seq_read(). Assigns
1684 * the head of the data block to &data and returns the length
1685 * of the block or 0 if the end of the skb data or the upper
1686 * offset has been reached.
1688 * The caller is not required to consume all of the data
1689 * returned, i.e. &consumed is typically set to the number
1690 * of bytes already consumed and the next call to
1691 * skb_seq_read() will return the remaining part of the block.
1693 * Note: The size of each block of data returned can be arbitary,
1694 * this limitation is the cost for zerocopy seqeuental
1695 * reads of potentially non linear data.
1697 * Note: Fragment lists within fragments are not implemented
1698 * at the moment, state->root_skb could be replaced with
1699 * a stack for this purpose.
1701 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1702 struct skb_seq_state *st)
1704 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
1707 if (unlikely(abs_offset >= st->upper_offset))
1711 block_limit = skb_headlen(st->cur_skb);
1713 if (abs_offset < block_limit) {
1714 *data = st->cur_skb->data + abs_offset;
1715 return block_limit - abs_offset;
1718 if (st->frag_idx == 0 && !st->frag_data)
1719 st->stepped_offset += skb_headlen(st->cur_skb);
1721 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
1722 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
1723 block_limit = frag->size + st->stepped_offset;
1725 if (abs_offset < block_limit) {
1727 st->frag_data = kmap_skb_frag(frag);
1729 *data = (u8 *) st->frag_data + frag->page_offset +
1730 (abs_offset - st->stepped_offset);
1732 return block_limit - abs_offset;
1735 if (st->frag_data) {
1736 kunmap_skb_frag(st->frag_data);
1737 st->frag_data = NULL;
1741 st->stepped_offset += frag->size;
1744 if (st->cur_skb->next) {
1745 st->cur_skb = st->cur_skb->next;
1748 } else if (st->root_skb == st->cur_skb &&
1749 skb_shinfo(st->root_skb)->frag_list) {
1750 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
1758 * skb_abort_seq_read - Abort a sequential read of skb data
1759 * @st: state variable
1761 * Must be called if skb_seq_read() was not called until it
1764 void skb_abort_seq_read(struct skb_seq_state *st)
1767 kunmap_skb_frag(st->frag_data);
1770 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
1772 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
1773 struct ts_config *conf,
1774 struct ts_state *state)
1776 return skb_seq_read(offset, text, TS_SKB_CB(state));
1779 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
1781 skb_abort_seq_read(TS_SKB_CB(state));
1785 * skb_find_text - Find a text pattern in skb data
1786 * @skb: the buffer to look in
1787 * @from: search offset
1789 * @config: textsearch configuration
1790 * @state: uninitialized textsearch state variable
1792 * Finds a pattern in the skb data according to the specified
1793 * textsearch configuration. Use textsearch_next() to retrieve
1794 * subsequent occurrences of the pattern. Returns the offset
1795 * to the first occurrence or UINT_MAX if no match was found.
1797 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1798 unsigned int to, struct ts_config *config,
1799 struct ts_state *state)
1801 config->get_next_block = skb_ts_get_next_block;
1802 config->finish = skb_ts_finish;
1804 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
1806 return textsearch_find(config, state);
1810 * skb_append_datato_frags: - append the user data to a skb
1811 * @sk: sock structure
1812 * @skb: skb structure to be appened with user data.
1813 * @getfrag: call back function to be used for getting the user data
1814 * @from: pointer to user message iov
1815 * @length: length of the iov message
1817 * Description: This procedure append the user data in the fragment part
1818 * of the skb if any page alloc fails user this procedure returns -ENOMEM
1820 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
1821 int (*getfrag)(void *from, char *to, int offset,
1822 int len, int odd, struct sk_buff *skb),
1823 void *from, int length)
1826 skb_frag_t *frag = NULL;
1827 struct page *page = NULL;
1833 /* Return error if we don't have space for new frag */
1834 frg_cnt = skb_shinfo(skb)->nr_frags;
1835 if (frg_cnt >= MAX_SKB_FRAGS)
1838 /* allocate a new page for next frag */
1839 page = alloc_pages(sk->sk_allocation, 0);
1841 /* If alloc_page fails just return failure and caller will
1842 * free previous allocated pages by doing kfree_skb()
1847 /* initialize the next frag */
1848 sk->sk_sndmsg_page = page;
1849 sk->sk_sndmsg_off = 0;
1850 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
1851 skb->truesize += PAGE_SIZE;
1852 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1854 /* get the new initialized frag */
1855 frg_cnt = skb_shinfo(skb)->nr_frags;
1856 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
1858 /* copy the user data to page */
1859 left = PAGE_SIZE - frag->page_offset;
1860 copy = (length > left)? left : length;
1862 ret = getfrag(from, (page_address(frag->page) +
1863 frag->page_offset + frag->size),
1864 offset, copy, 0, skb);
1868 /* copy was successful so update the size parameters */
1869 sk->sk_sndmsg_off += copy;
1872 skb->data_len += copy;
1876 } while (length > 0);
1882 * skb_pull_rcsum - pull skb and update receive checksum
1883 * @skb: buffer to update
1884 * @start: start of data before pull
1885 * @len: length of data pulled
1887 * This function performs an skb_pull on the packet and updates
1888 * update the CHECKSUM_HW checksum. It should be used on receive
1889 * path processing instead of skb_pull unless you know that the
1890 * checksum difference is zero (e.g., a valid IP header) or you
1891 * are setting ip_summed to CHECKSUM_NONE.
1893 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
1895 BUG_ON(len > skb->len);
1897 BUG_ON(skb->len < skb->data_len);
1898 skb_postpull_rcsum(skb, skb->data, len);
1899 return skb->data += len;
1902 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
1905 * skb_segment - Perform protocol segmentation on skb.
1906 * @skb: buffer to segment
1907 * @features: features for the output path (see dev->features)
1909 * This function performs segmentation on the given skb. It returns
1910 * the segment at the given position. It returns NULL if there are
1911 * no more segments to generate, or when an error is encountered.
1913 struct sk_buff *skb_segment(struct sk_buff *skb, int features)
1915 struct sk_buff *segs = NULL;
1916 struct sk_buff *tail = NULL;
1917 unsigned int mss = skb_shinfo(skb)->gso_size;
1918 unsigned int doffset = skb->data - skb->mac.raw;
1919 unsigned int offset = doffset;
1920 unsigned int headroom;
1922 int sg = features & NETIF_F_SG;
1923 int nfrags = skb_shinfo(skb)->nr_frags;
1928 __skb_push(skb, doffset);
1929 headroom = skb_headroom(skb);
1930 pos = skb_headlen(skb);
1933 struct sk_buff *nskb;
1939 len = skb->len - offset;
1943 hsize = skb_headlen(skb) - offset;
1946 nsize = hsize + doffset;
1947 if (nsize > len + doffset || !sg)
1948 nsize = len + doffset;
1950 nskb = alloc_skb(nsize + headroom, GFP_ATOMIC);
1951 if (unlikely(!nskb))
1960 nskb->dev = skb->dev;
1961 nskb->priority = skb->priority;
1962 nskb->protocol = skb->protocol;
1963 nskb->dst = dst_clone(skb->dst);
1964 memcpy(nskb->cb, skb->cb, sizeof(skb->cb));
1965 nskb->pkt_type = skb->pkt_type;
1966 nskb->mac_len = skb->mac_len;
1968 skb_reserve(nskb, headroom);
1969 nskb->mac.raw = nskb->data;
1970 nskb->nh.raw = nskb->data + skb->mac_len;
1971 nskb->h.raw = nskb->nh.raw + (skb->h.raw - skb->nh.raw);
1972 memcpy(skb_put(nskb, doffset), skb->data, doffset);
1975 nskb->csum = skb_copy_and_csum_bits(skb, offset,
1981 frag = skb_shinfo(nskb)->frags;
1984 nskb->ip_summed = CHECKSUM_HW;
1985 nskb->csum = skb->csum;
1986 memcpy(skb_put(nskb, hsize), skb->data + offset, hsize);
1988 while (pos < offset + len) {
1989 BUG_ON(i >= nfrags);
1991 *frag = skb_shinfo(skb)->frags[i];
1992 get_page(frag->page);
1996 frag->page_offset += offset - pos;
1997 frag->size -= offset - pos;
2002 if (pos + size <= offset + len) {
2006 frag->size -= pos + size - (offset + len);
2013 skb_shinfo(nskb)->nr_frags = k;
2014 nskb->data_len = len - hsize;
2015 nskb->len += nskb->data_len;
2016 nskb->truesize += nskb->data_len;
2017 } while ((offset += len) < skb->len);
2022 while ((skb = segs)) {
2026 return ERR_PTR(err);
2029 EXPORT_SYMBOL_GPL(skb_segment);
2031 void __init skb_init(void)
2033 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2034 sizeof(struct sk_buff),
2038 if (!skbuff_head_cache)
2039 panic("cannot create skbuff cache");
2041 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2042 (2*sizeof(struct sk_buff)) +
2047 if (!skbuff_fclone_cache)
2048 panic("cannot create skbuff cache");
2051 EXPORT_SYMBOL(___pskb_trim);
2052 EXPORT_SYMBOL(__kfree_skb);
2053 EXPORT_SYMBOL(kfree_skb);
2054 EXPORT_SYMBOL(__pskb_pull_tail);
2055 EXPORT_SYMBOL(__alloc_skb);
2056 EXPORT_SYMBOL(pskb_copy);
2057 EXPORT_SYMBOL(pskb_expand_head);
2058 EXPORT_SYMBOL(skb_checksum);
2059 EXPORT_SYMBOL(skb_clone);
2060 EXPORT_SYMBOL(skb_clone_fraglist);
2061 EXPORT_SYMBOL(skb_copy);
2062 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2063 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2064 EXPORT_SYMBOL(skb_copy_bits);
2065 EXPORT_SYMBOL(skb_copy_expand);
2066 EXPORT_SYMBOL(skb_over_panic);
2067 EXPORT_SYMBOL(skb_pad);
2068 EXPORT_SYMBOL(skb_realloc_headroom);
2069 EXPORT_SYMBOL(skb_under_panic);
2070 EXPORT_SYMBOL(skb_dequeue);
2071 EXPORT_SYMBOL(skb_dequeue_tail);
2072 EXPORT_SYMBOL(skb_insert);
2073 EXPORT_SYMBOL(skb_queue_purge);
2074 EXPORT_SYMBOL(skb_queue_head);
2075 EXPORT_SYMBOL(skb_queue_tail);
2076 EXPORT_SYMBOL(skb_unlink);
2077 EXPORT_SYMBOL(skb_append);
2078 EXPORT_SYMBOL(skb_split);
2079 EXPORT_SYMBOL(skb_prepare_seq_read);
2080 EXPORT_SYMBOL(skb_seq_read);
2081 EXPORT_SYMBOL(skb_abort_seq_read);
2082 EXPORT_SYMBOL(skb_find_text);
2083 EXPORT_SYMBOL(skb_append_datato_frags);