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));
422 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
426 n->destructor = NULL;
429 #ifdef CONFIG_NET_SCHED
431 #ifdef CONFIG_NET_CLS_ACT
432 n->tc_verd = SET_TC_VERD(skb->tc_verd,0);
433 n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd);
434 n->tc_verd = CLR_TC_MUNGED(n->tc_verd);
437 skb_copy_secmark(n, skb);
440 atomic_set(&n->users, 1);
446 atomic_inc(&(skb_shinfo(skb)->dataref));
452 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
454 #ifndef NET_SKBUFF_DATA_USES_OFFSET
456 * Shift between the two data areas in bytes
458 unsigned long offset = new->data - old->data;
462 new->priority = old->priority;
463 new->protocol = old->protocol;
464 new->dst = dst_clone(old->dst);
466 new->sp = secpath_get(old->sp);
468 new->transport_header = old->transport_header;
469 new->network_header = old->network_header;
470 new->mac_header = old->mac_header;
471 #ifndef NET_SKBUFF_DATA_USES_OFFSET
472 /* {transport,network,mac}_header are relative to skb->head */
473 new->transport_header += offset;
474 new->network_header += offset;
475 new->mac_header += offset;
477 memcpy(new->cb, old->cb, sizeof(old->cb));
478 new->local_df = old->local_df;
479 new->fclone = SKB_FCLONE_UNAVAILABLE;
480 new->pkt_type = old->pkt_type;
481 new->tstamp = old->tstamp;
482 new->destructor = NULL;
483 new->mark = old->mark;
485 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
486 new->ipvs_property = old->ipvs_property;
488 #ifdef CONFIG_NET_SCHED
489 #ifdef CONFIG_NET_CLS_ACT
490 new->tc_verd = old->tc_verd;
492 new->tc_index = old->tc_index;
494 skb_copy_secmark(new, old);
495 atomic_set(&new->users, 1);
496 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
497 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
498 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
502 * skb_copy - create private copy of an sk_buff
503 * @skb: buffer to copy
504 * @gfp_mask: allocation priority
506 * Make a copy of both an &sk_buff and its data. This is used when the
507 * caller wishes to modify the data and needs a private copy of the
508 * data to alter. Returns %NULL on failure or the pointer to the buffer
509 * on success. The returned buffer has a reference count of 1.
511 * As by-product this function converts non-linear &sk_buff to linear
512 * one, so that &sk_buff becomes completely private and caller is allowed
513 * to modify all the data of returned buffer. This means that this
514 * function is not recommended for use in circumstances when only
515 * header is going to be modified. Use pskb_copy() instead.
518 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
520 int headerlen = skb->data - skb->head;
522 * Allocate the copy buffer
525 #ifdef NET_SKBUFF_DATA_USES_OFFSET
526 n = alloc_skb(skb->end + skb->data_len, gfp_mask);
528 n = alloc_skb(skb->end - skb->head + skb->data_len, gfp_mask);
533 /* Set the data pointer */
534 skb_reserve(n, headerlen);
535 /* Set the tail pointer and length */
536 skb_put(n, skb->len);
538 n->ip_summed = skb->ip_summed;
540 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
543 copy_skb_header(n, skb);
549 * pskb_copy - create copy of an sk_buff with private head.
550 * @skb: buffer to copy
551 * @gfp_mask: allocation priority
553 * Make a copy of both an &sk_buff and part of its data, located
554 * in header. Fragmented data remain shared. This is used when
555 * the caller wishes to modify only header of &sk_buff and needs
556 * private copy of the header to alter. Returns %NULL on failure
557 * or the pointer to the buffer on success.
558 * The returned buffer has a reference count of 1.
561 struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
564 * Allocate the copy buffer
567 #ifdef NET_SKBUFF_DATA_USES_OFFSET
568 n = alloc_skb(skb->end, gfp_mask);
570 n = alloc_skb(skb->end - skb->head, gfp_mask);
575 /* Set the data pointer */
576 skb_reserve(n, skb->data - skb->head);
577 /* Set the tail pointer and length */
578 skb_put(n, skb_headlen(skb));
580 skb_copy_from_linear_data(skb, n->data, n->len);
582 n->ip_summed = skb->ip_summed;
584 n->truesize += skb->data_len;
585 n->data_len = skb->data_len;
588 if (skb_shinfo(skb)->nr_frags) {
591 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
592 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
593 get_page(skb_shinfo(n)->frags[i].page);
595 skb_shinfo(n)->nr_frags = i;
598 if (skb_shinfo(skb)->frag_list) {
599 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
600 skb_clone_fraglist(n);
603 copy_skb_header(n, skb);
609 * pskb_expand_head - reallocate header of &sk_buff
610 * @skb: buffer to reallocate
611 * @nhead: room to add at head
612 * @ntail: room to add at tail
613 * @gfp_mask: allocation priority
615 * Expands (or creates identical copy, if &nhead and &ntail are zero)
616 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
617 * reference count of 1. Returns zero in the case of success or error,
618 * if expansion failed. In the last case, &sk_buff is not changed.
620 * All the pointers pointing into skb header may change and must be
621 * reloaded after call to this function.
624 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
629 #ifdef NET_SKBUFF_DATA_USES_OFFSET
630 int size = nhead + skb->end + ntail;
632 int size = nhead + (skb->end - skb->head) + ntail;
639 size = SKB_DATA_ALIGN(size);
641 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
645 /* Copy only real data... and, alas, header. This should be
646 * optimized for the cases when header is void. */
647 memcpy(data + nhead, skb->head,
648 #ifdef NET_SKBUFF_DATA_USES_OFFSET
651 skb->tail - skb->head);
653 memcpy(data + size, skb_end_pointer(skb),
654 sizeof(struct skb_shared_info));
656 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
657 get_page(skb_shinfo(skb)->frags[i].page);
659 if (skb_shinfo(skb)->frag_list)
660 skb_clone_fraglist(skb);
662 skb_release_data(skb);
664 off = (data + nhead) - skb->head;
668 #ifdef NET_SKBUFF_DATA_USES_OFFSET
672 skb->end = skb->head + size;
674 /* {transport,network,mac}_header and tail are relative to skb->head */
676 skb->transport_header += off;
677 skb->network_header += off;
678 skb->mac_header += off;
681 atomic_set(&skb_shinfo(skb)->dataref, 1);
688 /* Make private copy of skb with writable head and some headroom */
690 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
692 struct sk_buff *skb2;
693 int delta = headroom - skb_headroom(skb);
696 skb2 = pskb_copy(skb, GFP_ATOMIC);
698 skb2 = skb_clone(skb, GFP_ATOMIC);
699 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
710 * skb_copy_expand - copy and expand sk_buff
711 * @skb: buffer to copy
712 * @newheadroom: new free bytes at head
713 * @newtailroom: new free bytes at tail
714 * @gfp_mask: allocation priority
716 * Make a copy of both an &sk_buff and its data and while doing so
717 * allocate additional space.
719 * This is used when the caller wishes to modify the data and needs a
720 * private copy of the data to alter as well as more space for new fields.
721 * Returns %NULL on failure or the pointer to the buffer
722 * on success. The returned buffer has a reference count of 1.
724 * You must pass %GFP_ATOMIC as the allocation priority if this function
725 * is called from an interrupt.
727 * BUG ALERT: ip_summed is not copied. Why does this work? Is it used
728 * only by netfilter in the cases when checksum is recalculated? --ANK
730 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
731 int newheadroom, int newtailroom,
735 * Allocate the copy buffer
737 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
739 int oldheadroom = skb_headroom(skb);
740 int head_copy_len, head_copy_off;
746 skb_reserve(n, newheadroom);
748 /* Set the tail pointer and length */
749 skb_put(n, skb->len);
751 head_copy_len = oldheadroom;
753 if (newheadroom <= head_copy_len)
754 head_copy_len = newheadroom;
756 head_copy_off = newheadroom - head_copy_len;
758 /* Copy the linear header and data. */
759 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
760 skb->len + head_copy_len))
763 copy_skb_header(n, skb);
765 #ifdef NET_SKBUFF_DATA_USES_OFFSET
766 off = newheadroom - oldheadroom;
768 n->transport_header += off;
769 n->network_header += off;
770 n->mac_header += off;
776 * skb_pad - zero pad the tail of an skb
777 * @skb: buffer to pad
780 * Ensure that a buffer is followed by a padding area that is zero
781 * filled. Used by network drivers which may DMA or transfer data
782 * beyond the buffer end onto the wire.
784 * May return error in out of memory cases. The skb is freed on error.
787 int skb_pad(struct sk_buff *skb, int pad)
792 /* If the skbuff is non linear tailroom is always zero.. */
793 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
794 memset(skb->data+skb->len, 0, pad);
798 ntail = skb->data_len + pad - (skb->end - skb->tail);
799 if (likely(skb_cloned(skb) || ntail > 0)) {
800 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
805 /* FIXME: The use of this function with non-linear skb's really needs
808 err = skb_linearize(skb);
812 memset(skb->data + skb->len, 0, pad);
820 /* Trims skb to length len. It can change skb pointers.
823 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
825 struct sk_buff **fragp;
826 struct sk_buff *frag;
827 int offset = skb_headlen(skb);
828 int nfrags = skb_shinfo(skb)->nr_frags;
832 if (skb_cloned(skb) &&
833 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
840 for (; i < nfrags; i++) {
841 int end = offset + skb_shinfo(skb)->frags[i].size;
848 skb_shinfo(skb)->frags[i++].size = len - offset;
851 skb_shinfo(skb)->nr_frags = i;
853 for (; i < nfrags; i++)
854 put_page(skb_shinfo(skb)->frags[i].page);
856 if (skb_shinfo(skb)->frag_list)
857 skb_drop_fraglist(skb);
861 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
862 fragp = &frag->next) {
863 int end = offset + frag->len;
865 if (skb_shared(frag)) {
866 struct sk_buff *nfrag;
868 nfrag = skb_clone(frag, GFP_ATOMIC);
869 if (unlikely(!nfrag))
872 nfrag->next = frag->next;
884 unlikely((err = pskb_trim(frag, len - offset))))
888 skb_drop_list(&frag->next);
893 if (len > skb_headlen(skb)) {
894 skb->data_len -= skb->len - len;
899 skb_set_tail_pointer(skb, len);
906 * __pskb_pull_tail - advance tail of skb header
907 * @skb: buffer to reallocate
908 * @delta: number of bytes to advance tail
910 * The function makes a sense only on a fragmented &sk_buff,
911 * it expands header moving its tail forward and copying necessary
912 * data from fragmented part.
914 * &sk_buff MUST have reference count of 1.
916 * Returns %NULL (and &sk_buff does not change) if pull failed
917 * or value of new tail of skb in the case of success.
919 * All the pointers pointing into skb header may change and must be
920 * reloaded after call to this function.
923 /* Moves tail of skb head forward, copying data from fragmented part,
924 * when it is necessary.
925 * 1. It may fail due to malloc failure.
926 * 2. It may change skb pointers.
928 * It is pretty complicated. Luckily, it is called only in exceptional cases.
930 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
932 /* If skb has not enough free space at tail, get new one
933 * plus 128 bytes for future expansions. If we have enough
934 * room at tail, reallocate without expansion only if skb is cloned.
936 int i, k, eat = (skb->tail + delta) - skb->end;
938 if (eat > 0 || skb_cloned(skb)) {
939 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
944 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
947 /* Optimization: no fragments, no reasons to preestimate
948 * size of pulled pages. Superb.
950 if (!skb_shinfo(skb)->frag_list)
953 /* Estimate size of pulled pages. */
955 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
956 if (skb_shinfo(skb)->frags[i].size >= eat)
958 eat -= skb_shinfo(skb)->frags[i].size;
961 /* If we need update frag list, we are in troubles.
962 * Certainly, it possible to add an offset to skb data,
963 * but taking into account that pulling is expected to
964 * be very rare operation, it is worth to fight against
965 * further bloating skb head and crucify ourselves here instead.
966 * Pure masohism, indeed. 8)8)
969 struct sk_buff *list = skb_shinfo(skb)->frag_list;
970 struct sk_buff *clone = NULL;
971 struct sk_buff *insp = NULL;
976 if (list->len <= eat) {
977 /* Eaten as whole. */
982 /* Eaten partially. */
984 if (skb_shared(list)) {
985 /* Sucks! We need to fork list. :-( */
986 clone = skb_clone(list, GFP_ATOMIC);
992 /* This may be pulled without
996 if (!pskb_pull(list, eat)) {
1005 /* Free pulled out fragments. */
1006 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1007 skb_shinfo(skb)->frag_list = list->next;
1010 /* And insert new clone at head. */
1013 skb_shinfo(skb)->frag_list = clone;
1016 /* Success! Now we may commit changes to skb data. */
1021 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1022 if (skb_shinfo(skb)->frags[i].size <= eat) {
1023 put_page(skb_shinfo(skb)->frags[i].page);
1024 eat -= skb_shinfo(skb)->frags[i].size;
1026 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1028 skb_shinfo(skb)->frags[k].page_offset += eat;
1029 skb_shinfo(skb)->frags[k].size -= eat;
1035 skb_shinfo(skb)->nr_frags = k;
1038 skb->data_len -= delta;
1040 return skb_tail_pointer(skb);
1043 /* Copy some data bits from skb to kernel buffer. */
1045 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1048 int end = skb_headlen(skb);
1050 if (offset > (int)skb->len - len)
1054 if ((copy = end - offset) > 0) {
1057 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1058 if ((len -= copy) == 0)
1064 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1067 end = offset + skb_shinfo(skb)->frags[i].size;
1068 if ((copy = end - offset) > 0) {
1074 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1076 vaddr + skb_shinfo(skb)->frags[i].page_offset,
1078 kunmap_skb_frag(vaddr);
1080 if ((len -= copy) == 0)
1087 if (skb_shinfo(skb)->frag_list) {
1088 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1090 for (; list; list = list->next) {
1093 end = offset + list->len;
1094 if ((copy = end - offset) > 0) {
1097 if (skb_copy_bits(list, 0, to, copy))
1099 if ((len -= copy) == 0)
1114 * skb_store_bits - store bits from kernel buffer to skb
1115 * @skb: destination buffer
1116 * @offset: offset in destination
1117 * @from: source buffer
1118 * @len: number of bytes to copy
1120 * Copy the specified number of bytes from the source buffer to the
1121 * destination skb. This function handles all the messy bits of
1122 * traversing fragment lists and such.
1125 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1128 int end = skb_headlen(skb);
1130 if (offset > (int)skb->len - len)
1133 if ((copy = end - offset) > 0) {
1136 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1137 if ((len -= copy) == 0)
1143 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1144 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1147 end = offset + frag->size;
1148 if ((copy = end - offset) > 0) {
1154 vaddr = kmap_skb_frag(frag);
1155 memcpy(vaddr + frag->page_offset, from, copy);
1156 kunmap_skb_frag(vaddr);
1158 if ((len -= copy) == 0)
1165 if (skb_shinfo(skb)->frag_list) {
1166 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1168 for (; list; list = list->next) {
1171 end = offset + list->len;
1172 if ((copy = end - offset) > 0) {
1175 if (skb_store_bits(list, 0, from, copy))
1177 if ((len -= copy) == 0)
1191 EXPORT_SYMBOL(skb_store_bits);
1193 /* Checksum skb data. */
1195 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1196 int len, __wsum csum)
1198 int end = skb_headlen(skb);
1199 int i, copy = end - offset;
1202 /* Checksum header. */
1206 csum = csum_partial(skb->data + offset, copy, csum);
1207 if ((len -= copy) == 0)
1213 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1216 end = offset + skb_shinfo(skb)->frags[i].size;
1217 if ((copy = end - offset) > 0) {
1220 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1224 vaddr = kmap_skb_frag(frag);
1225 csum2 = csum_partial(vaddr + frag->page_offset,
1227 kunmap_skb_frag(vaddr);
1228 csum = csum_block_add(csum, csum2, pos);
1236 if (skb_shinfo(skb)->frag_list) {
1237 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1239 for (; list; list = list->next) {
1242 end = offset + list->len;
1243 if ((copy = end - offset) > 0) {
1247 csum2 = skb_checksum(list, 0, copy, 0);
1248 csum = csum_block_add(csum, csum2, pos);
1249 if ((len -= copy) == 0)
1261 /* Both of above in one bottle. */
1263 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1264 u8 *to, int len, __wsum csum)
1266 int end = skb_headlen(skb);
1267 int i, copy = end - offset;
1274 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1276 if ((len -= copy) == 0)
1283 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1286 end = offset + skb_shinfo(skb)->frags[i].size;
1287 if ((copy = end - offset) > 0) {
1290 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1294 vaddr = kmap_skb_frag(frag);
1295 csum2 = csum_partial_copy_nocheck(vaddr +
1298 kunmap_skb_frag(vaddr);
1299 csum = csum_block_add(csum, csum2, pos);
1308 if (skb_shinfo(skb)->frag_list) {
1309 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1311 for (; list; list = list->next) {
1315 end = offset + list->len;
1316 if ((copy = end - offset) > 0) {
1319 csum2 = skb_copy_and_csum_bits(list, 0,
1321 csum = csum_block_add(csum, csum2, pos);
1322 if ((len -= copy) == 0)
1334 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1339 if (skb->ip_summed == CHECKSUM_PARTIAL)
1340 csstart = skb->csum_start - skb_headroom(skb);
1342 csstart = skb_headlen(skb);
1344 BUG_ON(csstart > skb_headlen(skb));
1346 skb_copy_from_linear_data(skb, to, csstart);
1349 if (csstart != skb->len)
1350 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1351 skb->len - csstart, 0);
1353 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1354 long csstuff = csstart + skb->csum_offset;
1356 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
1361 * skb_dequeue - remove from the head of the queue
1362 * @list: list to dequeue from
1364 * Remove the head of the list. The list lock is taken so the function
1365 * may be used safely with other locking list functions. The head item is
1366 * returned or %NULL if the list is empty.
1369 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1371 unsigned long flags;
1372 struct sk_buff *result;
1374 spin_lock_irqsave(&list->lock, flags);
1375 result = __skb_dequeue(list);
1376 spin_unlock_irqrestore(&list->lock, flags);
1381 * skb_dequeue_tail - remove from the tail of the queue
1382 * @list: list to dequeue from
1384 * Remove the tail of the list. The list lock is taken so the function
1385 * may be used safely with other locking list functions. The tail item is
1386 * returned or %NULL if the list is empty.
1388 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1390 unsigned long flags;
1391 struct sk_buff *result;
1393 spin_lock_irqsave(&list->lock, flags);
1394 result = __skb_dequeue_tail(list);
1395 spin_unlock_irqrestore(&list->lock, flags);
1400 * skb_queue_purge - empty a list
1401 * @list: list to empty
1403 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1404 * the list and one reference dropped. This function takes the list
1405 * lock and is atomic with respect to other list locking functions.
1407 void skb_queue_purge(struct sk_buff_head *list)
1409 struct sk_buff *skb;
1410 while ((skb = skb_dequeue(list)) != NULL)
1415 * skb_queue_head - queue a buffer at the list head
1416 * @list: list to use
1417 * @newsk: buffer to queue
1419 * Queue a buffer at the start of the list. This function takes the
1420 * list lock and can be used safely with other locking &sk_buff functions
1423 * A buffer cannot be placed on two lists at the same time.
1425 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1427 unsigned long flags;
1429 spin_lock_irqsave(&list->lock, flags);
1430 __skb_queue_head(list, newsk);
1431 spin_unlock_irqrestore(&list->lock, flags);
1435 * skb_queue_tail - queue a buffer at the list tail
1436 * @list: list to use
1437 * @newsk: buffer to queue
1439 * Queue a buffer at the tail of the list. This function takes the
1440 * list lock and can be used safely with other locking &sk_buff functions
1443 * A buffer cannot be placed on two lists at the same time.
1445 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1447 unsigned long flags;
1449 spin_lock_irqsave(&list->lock, flags);
1450 __skb_queue_tail(list, newsk);
1451 spin_unlock_irqrestore(&list->lock, flags);
1455 * skb_unlink - remove a buffer from a list
1456 * @skb: buffer to remove
1457 * @list: list to use
1459 * Remove a packet from a list. The list locks are taken and this
1460 * function is atomic with respect to other list locked calls
1462 * You must know what list the SKB is on.
1464 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1466 unsigned long flags;
1468 spin_lock_irqsave(&list->lock, flags);
1469 __skb_unlink(skb, list);
1470 spin_unlock_irqrestore(&list->lock, flags);
1474 * skb_append - append a buffer
1475 * @old: buffer to insert after
1476 * @newsk: buffer to insert
1477 * @list: list to use
1479 * Place a packet after a given packet in a list. The list locks are taken
1480 * and this function is atomic with respect to other list locked calls.
1481 * A buffer cannot be placed on two lists at the same time.
1483 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1485 unsigned long flags;
1487 spin_lock_irqsave(&list->lock, flags);
1488 __skb_append(old, newsk, list);
1489 spin_unlock_irqrestore(&list->lock, flags);
1494 * skb_insert - insert a buffer
1495 * @old: buffer to insert before
1496 * @newsk: buffer to insert
1497 * @list: list to use
1499 * Place a packet before a given packet in a list. The list locks are
1500 * taken and this function is atomic with respect to other list locked
1503 * A buffer cannot be placed on two lists at the same time.
1505 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1507 unsigned long flags;
1509 spin_lock_irqsave(&list->lock, flags);
1510 __skb_insert(newsk, old->prev, old, list);
1511 spin_unlock_irqrestore(&list->lock, flags);
1514 static inline void skb_split_inside_header(struct sk_buff *skb,
1515 struct sk_buff* skb1,
1516 const u32 len, const int pos)
1520 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
1522 /* And move data appendix as is. */
1523 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1524 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
1526 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
1527 skb_shinfo(skb)->nr_frags = 0;
1528 skb1->data_len = skb->data_len;
1529 skb1->len += skb1->data_len;
1532 skb_set_tail_pointer(skb, len);
1535 static inline void skb_split_no_header(struct sk_buff *skb,
1536 struct sk_buff* skb1,
1537 const u32 len, int pos)
1540 const int nfrags = skb_shinfo(skb)->nr_frags;
1542 skb_shinfo(skb)->nr_frags = 0;
1543 skb1->len = skb1->data_len = skb->len - len;
1545 skb->data_len = len - pos;
1547 for (i = 0; i < nfrags; i++) {
1548 int size = skb_shinfo(skb)->frags[i].size;
1550 if (pos + size > len) {
1551 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
1555 * We have two variants in this case:
1556 * 1. Move all the frag to the second
1557 * part, if it is possible. F.e.
1558 * this approach is mandatory for TUX,
1559 * where splitting is expensive.
1560 * 2. Split is accurately. We make this.
1562 get_page(skb_shinfo(skb)->frags[i].page);
1563 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
1564 skb_shinfo(skb1)->frags[0].size -= len - pos;
1565 skb_shinfo(skb)->frags[i].size = len - pos;
1566 skb_shinfo(skb)->nr_frags++;
1570 skb_shinfo(skb)->nr_frags++;
1573 skb_shinfo(skb1)->nr_frags = k;
1577 * skb_split - Split fragmented skb to two parts at length len.
1578 * @skb: the buffer to split
1579 * @skb1: the buffer to receive the second part
1580 * @len: new length for skb
1582 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
1584 int pos = skb_headlen(skb);
1586 if (len < pos) /* Split line is inside header. */
1587 skb_split_inside_header(skb, skb1, len, pos);
1588 else /* Second chunk has no header, nothing to copy. */
1589 skb_split_no_header(skb, skb1, len, pos);
1593 * skb_prepare_seq_read - Prepare a sequential read of skb data
1594 * @skb: the buffer to read
1595 * @from: lower offset of data to be read
1596 * @to: upper offset of data to be read
1597 * @st: state variable
1599 * Initializes the specified state variable. Must be called before
1600 * invoking skb_seq_read() for the first time.
1602 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1603 unsigned int to, struct skb_seq_state *st)
1605 st->lower_offset = from;
1606 st->upper_offset = to;
1607 st->root_skb = st->cur_skb = skb;
1608 st->frag_idx = st->stepped_offset = 0;
1609 st->frag_data = NULL;
1613 * skb_seq_read - Sequentially read skb data
1614 * @consumed: number of bytes consumed by the caller so far
1615 * @data: destination pointer for data to be returned
1616 * @st: state variable
1618 * Reads a block of skb data at &consumed relative to the
1619 * lower offset specified to skb_prepare_seq_read(). Assigns
1620 * the head of the data block to &data and returns the length
1621 * of the block or 0 if the end of the skb data or the upper
1622 * offset has been reached.
1624 * The caller is not required to consume all of the data
1625 * returned, i.e. &consumed is typically set to the number
1626 * of bytes already consumed and the next call to
1627 * skb_seq_read() will return the remaining part of the block.
1629 * Note: The size of each block of data returned can be arbitary,
1630 * this limitation is the cost for zerocopy seqeuental
1631 * reads of potentially non linear data.
1633 * Note: Fragment lists within fragments are not implemented
1634 * at the moment, state->root_skb could be replaced with
1635 * a stack for this purpose.
1637 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1638 struct skb_seq_state *st)
1640 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
1643 if (unlikely(abs_offset >= st->upper_offset))
1647 block_limit = skb_headlen(st->cur_skb);
1649 if (abs_offset < block_limit) {
1650 *data = st->cur_skb->data + abs_offset;
1651 return block_limit - abs_offset;
1654 if (st->frag_idx == 0 && !st->frag_data)
1655 st->stepped_offset += skb_headlen(st->cur_skb);
1657 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
1658 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
1659 block_limit = frag->size + st->stepped_offset;
1661 if (abs_offset < block_limit) {
1663 st->frag_data = kmap_skb_frag(frag);
1665 *data = (u8 *) st->frag_data + frag->page_offset +
1666 (abs_offset - st->stepped_offset);
1668 return block_limit - abs_offset;
1671 if (st->frag_data) {
1672 kunmap_skb_frag(st->frag_data);
1673 st->frag_data = NULL;
1677 st->stepped_offset += frag->size;
1680 if (st->cur_skb->next) {
1681 st->cur_skb = st->cur_skb->next;
1684 } else if (st->root_skb == st->cur_skb &&
1685 skb_shinfo(st->root_skb)->frag_list) {
1686 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
1694 * skb_abort_seq_read - Abort a sequential read of skb data
1695 * @st: state variable
1697 * Must be called if skb_seq_read() was not called until it
1700 void skb_abort_seq_read(struct skb_seq_state *st)
1703 kunmap_skb_frag(st->frag_data);
1706 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
1708 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
1709 struct ts_config *conf,
1710 struct ts_state *state)
1712 return skb_seq_read(offset, text, TS_SKB_CB(state));
1715 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
1717 skb_abort_seq_read(TS_SKB_CB(state));
1721 * skb_find_text - Find a text pattern in skb data
1722 * @skb: the buffer to look in
1723 * @from: search offset
1725 * @config: textsearch configuration
1726 * @state: uninitialized textsearch state variable
1728 * Finds a pattern in the skb data according to the specified
1729 * textsearch configuration. Use textsearch_next() to retrieve
1730 * subsequent occurrences of the pattern. Returns the offset
1731 * to the first occurrence or UINT_MAX if no match was found.
1733 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1734 unsigned int to, struct ts_config *config,
1735 struct ts_state *state)
1739 config->get_next_block = skb_ts_get_next_block;
1740 config->finish = skb_ts_finish;
1742 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
1744 ret = textsearch_find(config, state);
1745 return (ret <= to - from ? ret : UINT_MAX);
1749 * skb_append_datato_frags: - append the user data to a skb
1750 * @sk: sock structure
1751 * @skb: skb structure to be appened with user data.
1752 * @getfrag: call back function to be used for getting the user data
1753 * @from: pointer to user message iov
1754 * @length: length of the iov message
1756 * Description: This procedure append the user data in the fragment part
1757 * of the skb if any page alloc fails user this procedure returns -ENOMEM
1759 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
1760 int (*getfrag)(void *from, char *to, int offset,
1761 int len, int odd, struct sk_buff *skb),
1762 void *from, int length)
1765 skb_frag_t *frag = NULL;
1766 struct page *page = NULL;
1772 /* Return error if we don't have space for new frag */
1773 frg_cnt = skb_shinfo(skb)->nr_frags;
1774 if (frg_cnt >= MAX_SKB_FRAGS)
1777 /* allocate a new page for next frag */
1778 page = alloc_pages(sk->sk_allocation, 0);
1780 /* If alloc_page fails just return failure and caller will
1781 * free previous allocated pages by doing kfree_skb()
1786 /* initialize the next frag */
1787 sk->sk_sndmsg_page = page;
1788 sk->sk_sndmsg_off = 0;
1789 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
1790 skb->truesize += PAGE_SIZE;
1791 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1793 /* get the new initialized frag */
1794 frg_cnt = skb_shinfo(skb)->nr_frags;
1795 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
1797 /* copy the user data to page */
1798 left = PAGE_SIZE - frag->page_offset;
1799 copy = (length > left)? left : length;
1801 ret = getfrag(from, (page_address(frag->page) +
1802 frag->page_offset + frag->size),
1803 offset, copy, 0, skb);
1807 /* copy was successful so update the size parameters */
1808 sk->sk_sndmsg_off += copy;
1811 skb->data_len += copy;
1815 } while (length > 0);
1821 * skb_pull_rcsum - pull skb and update receive checksum
1822 * @skb: buffer to update
1823 * @start: start of data before pull
1824 * @len: length of data pulled
1826 * This function performs an skb_pull on the packet and updates
1827 * update the CHECKSUM_COMPLETE checksum. It should be used on
1828 * receive path processing instead of skb_pull unless you know
1829 * that the checksum difference is zero (e.g., a valid IP header)
1830 * or you are setting ip_summed to CHECKSUM_NONE.
1832 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
1834 BUG_ON(len > skb->len);
1836 BUG_ON(skb->len < skb->data_len);
1837 skb_postpull_rcsum(skb, skb->data, len);
1838 return skb->data += len;
1841 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
1844 * skb_segment - Perform protocol segmentation on skb.
1845 * @skb: buffer to segment
1846 * @features: features for the output path (see dev->features)
1848 * This function performs segmentation on the given skb. It returns
1849 * the segment at the given position. It returns NULL if there are
1850 * no more segments to generate, or when an error is encountered.
1852 struct sk_buff *skb_segment(struct sk_buff *skb, int features)
1854 struct sk_buff *segs = NULL;
1855 struct sk_buff *tail = NULL;
1856 unsigned int mss = skb_shinfo(skb)->gso_size;
1857 unsigned int doffset = skb->data - skb_mac_header(skb);
1858 unsigned int offset = doffset;
1859 unsigned int headroom;
1861 int sg = features & NETIF_F_SG;
1862 int nfrags = skb_shinfo(skb)->nr_frags;
1867 __skb_push(skb, doffset);
1868 headroom = skb_headroom(skb);
1869 pos = skb_headlen(skb);
1872 struct sk_buff *nskb;
1878 len = skb->len - offset;
1882 hsize = skb_headlen(skb) - offset;
1885 if (hsize > len || !sg)
1888 nskb = alloc_skb(hsize + doffset + headroom, GFP_ATOMIC);
1889 if (unlikely(!nskb))
1898 nskb->dev = skb->dev;
1899 nskb->priority = skb->priority;
1900 nskb->protocol = skb->protocol;
1901 nskb->dst = dst_clone(skb->dst);
1902 memcpy(nskb->cb, skb->cb, sizeof(skb->cb));
1903 nskb->pkt_type = skb->pkt_type;
1904 nskb->mac_len = skb->mac_len;
1906 skb_reserve(nskb, headroom);
1907 skb_reset_mac_header(nskb);
1908 skb_set_network_header(nskb, skb->mac_len);
1909 nskb->transport_header = (nskb->network_header +
1910 skb_network_header_len(skb));
1911 skb_copy_from_linear_data(skb, skb_put(nskb, doffset),
1914 nskb->csum = skb_copy_and_csum_bits(skb, offset,
1920 frag = skb_shinfo(nskb)->frags;
1923 nskb->ip_summed = CHECKSUM_PARTIAL;
1924 nskb->csum = skb->csum;
1925 skb_copy_from_linear_data_offset(skb, offset,
1926 skb_put(nskb, hsize), hsize);
1928 while (pos < offset + len) {
1929 BUG_ON(i >= nfrags);
1931 *frag = skb_shinfo(skb)->frags[i];
1932 get_page(frag->page);
1936 frag->page_offset += offset - pos;
1937 frag->size -= offset - pos;
1942 if (pos + size <= offset + len) {
1946 frag->size -= pos + size - (offset + len);
1953 skb_shinfo(nskb)->nr_frags = k;
1954 nskb->data_len = len - hsize;
1955 nskb->len += nskb->data_len;
1956 nskb->truesize += nskb->data_len;
1957 } while ((offset += len) < skb->len);
1962 while ((skb = segs)) {
1966 return ERR_PTR(err);
1969 EXPORT_SYMBOL_GPL(skb_segment);
1971 void __init skb_init(void)
1973 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
1974 sizeof(struct sk_buff),
1976 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1978 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
1979 (2*sizeof(struct sk_buff)) +
1982 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1987 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
1988 * @skb: Socket buffer containing the buffers to be mapped
1989 * @sg: The scatter-gather list to map into
1990 * @offset: The offset into the buffer's contents to start mapping
1991 * @len: Length of buffer space to be mapped
1993 * Fill the specified scatter-gather list with mappings/pointers into a
1994 * region of the buffer space attached to a socket buffer.
1997 skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
1999 int end = skb_headlen(skb);
2000 int i, copy = end - offset;
2006 sg[elt].page = virt_to_page(skb->data + offset);
2007 sg[elt].offset = (unsigned long)(skb->data + offset) % PAGE_SIZE;
2008 sg[elt].length = copy;
2010 if ((len -= copy) == 0)
2015 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2018 end = offset + skb_shinfo(skb)->frags[i].size;
2019 if ((copy = end - offset) > 0) {
2020 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2024 sg[elt].page = frag->page;
2025 sg[elt].offset = frag->page_offset;
2026 sg[elt].length = copy;
2034 if (skb_shinfo(skb)->frag_list) {
2035 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2037 for (; list; list = list->next) {
2040 end = offset + list->len;
2041 if ((copy = end - offset) > 0) {
2044 elt += skb_to_sgvec(list, sg+elt, 0, copy);
2045 if ((len -= copy) == 0)
2056 * skb_cow_data - Check that a socket buffer's data buffers are writable
2057 * @skb: The socket buffer to check.
2058 * @tailbits: Amount of trailing space to be added
2059 * @trailer: Returned pointer to the skb where the @tailbits space begins
2061 * Make sure that the data buffers attached to a socket buffer are
2062 * writable. If they are not, private copies are made of the data buffers
2063 * and the socket buffer is set to use these instead.
2065 * If @tailbits is given, make sure that there is space to write @tailbits
2066 * bytes of data beyond current end of socket buffer. @trailer will be
2067 * set to point to the skb in which this space begins.
2069 * The number of scatterlist elements required to completely map the
2070 * COW'd and extended socket buffer will be returned.
2072 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
2076 struct sk_buff *skb1, **skb_p;
2078 /* If skb is cloned or its head is paged, reallocate
2079 * head pulling out all the pages (pages are considered not writable
2080 * at the moment even if they are anonymous).
2082 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
2083 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
2086 /* Easy case. Most of packets will go this way. */
2087 if (!skb_shinfo(skb)->frag_list) {
2088 /* A little of trouble, not enough of space for trailer.
2089 * This should not happen, when stack is tuned to generate
2090 * good frames. OK, on miss we reallocate and reserve even more
2091 * space, 128 bytes is fair. */
2093 if (skb_tailroom(skb) < tailbits &&
2094 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
2102 /* Misery. We are in troubles, going to mincer fragments... */
2105 skb_p = &skb_shinfo(skb)->frag_list;
2108 while ((skb1 = *skb_p) != NULL) {
2111 /* The fragment is partially pulled by someone,
2112 * this can happen on input. Copy it and everything
2115 if (skb_shared(skb1))
2118 /* If the skb is the last, worry about trailer. */
2120 if (skb1->next == NULL && tailbits) {
2121 if (skb_shinfo(skb1)->nr_frags ||
2122 skb_shinfo(skb1)->frag_list ||
2123 skb_tailroom(skb1) < tailbits)
2124 ntail = tailbits + 128;
2130 skb_shinfo(skb1)->nr_frags ||
2131 skb_shinfo(skb1)->frag_list) {
2132 struct sk_buff *skb2;
2134 /* Fuck, we are miserable poor guys... */
2136 skb2 = skb_copy(skb1, GFP_ATOMIC);
2138 skb2 = skb_copy_expand(skb1,
2142 if (unlikely(skb2 == NULL))
2146 skb_set_owner_w(skb2, skb1->sk);
2148 /* Looking around. Are we still alive?
2149 * OK, link new skb, drop old one */
2151 skb2->next = skb1->next;
2158 skb_p = &skb1->next;
2164 EXPORT_SYMBOL(___pskb_trim);
2165 EXPORT_SYMBOL(__kfree_skb);
2166 EXPORT_SYMBOL(kfree_skb);
2167 EXPORT_SYMBOL(__pskb_pull_tail);
2168 EXPORT_SYMBOL(__alloc_skb);
2169 EXPORT_SYMBOL(__netdev_alloc_skb);
2170 EXPORT_SYMBOL(pskb_copy);
2171 EXPORT_SYMBOL(pskb_expand_head);
2172 EXPORT_SYMBOL(skb_checksum);
2173 EXPORT_SYMBOL(skb_clone);
2174 EXPORT_SYMBOL(skb_clone_fraglist);
2175 EXPORT_SYMBOL(skb_copy);
2176 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2177 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2178 EXPORT_SYMBOL(skb_copy_bits);
2179 EXPORT_SYMBOL(skb_copy_expand);
2180 EXPORT_SYMBOL(skb_over_panic);
2181 EXPORT_SYMBOL(skb_pad);
2182 EXPORT_SYMBOL(skb_realloc_headroom);
2183 EXPORT_SYMBOL(skb_under_panic);
2184 EXPORT_SYMBOL(skb_dequeue);
2185 EXPORT_SYMBOL(skb_dequeue_tail);
2186 EXPORT_SYMBOL(skb_insert);
2187 EXPORT_SYMBOL(skb_queue_purge);
2188 EXPORT_SYMBOL(skb_queue_head);
2189 EXPORT_SYMBOL(skb_queue_tail);
2190 EXPORT_SYMBOL(skb_unlink);
2191 EXPORT_SYMBOL(skb_append);
2192 EXPORT_SYMBOL(skb_split);
2193 EXPORT_SYMBOL(skb_prepare_seq_read);
2194 EXPORT_SYMBOL(skb_seq_read);
2195 EXPORT_SYMBOL(skb_abort_seq_read);
2196 EXPORT_SYMBOL(skb_find_text);
2197 EXPORT_SYMBOL(skb_append_datato_frags);
2199 EXPORT_SYMBOL_GPL(skb_to_sgvec);
2200 EXPORT_SYMBOL_GPL(skb_cow_data);