2 * Routines having to do with the 'struct sk_buff' memory handlers.
4 * Authors: Alan Cox <iiitac@pyr.swan.ac.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
7 * Version: $Id: skbuff.c,v 1.90 2001/11/07 05:56:19 davem Exp $
10 * Alan Cox : Fixed the worst of the load
12 * Dave Platt : Interrupt stacking fix.
13 * Richard Kooijman : Timestamp fixes.
14 * Alan Cox : Changed buffer format.
15 * Alan Cox : destructor hook for AF_UNIX etc.
16 * Linus Torvalds : Better skb_clone.
17 * Alan Cox : Added skb_copy.
18 * Alan Cox : Added all the changed routines Linus
19 * only put in the headers
20 * Ray VanTassle : Fixed --skb->lock in free
21 * Alan Cox : skb_copy copy arp field
22 * Andi Kleen : slabified it.
23 * Robert Olsson : Removed skb_head_pool
26 * The __skb_ routines should be called with interrupts
27 * disabled, or you better be *real* sure that the operation is atomic
28 * with respect to whatever list is being frobbed (e.g. via lock_sock()
29 * or via disabling bottom half handlers, etc).
31 * This program is free software; you can redistribute it and/or
32 * modify it under the terms of the GNU General Public License
33 * as published by the Free Software Foundation; either version
34 * 2 of the License, or (at your option) any later version.
38 * The functions in this file will not compile correctly with gcc 2.4.x
41 #include <linux/module.h>
42 #include <linux/types.h>
43 #include <linux/kernel.h>
45 #include <linux/interrupt.h>
47 #include <linux/inet.h>
48 #include <linux/slab.h>
49 #include <linux/netdevice.h>
50 #ifdef CONFIG_NET_CLS_ACT
51 #include <net/pkt_sched.h>
53 #include <linux/string.h>
54 #include <linux/skbuff.h>
55 #include <linux/cache.h>
56 #include <linux/rtnetlink.h>
57 #include <linux/init.h>
58 #include <linux/scatterlist.h>
60 #include <net/protocol.h>
63 #include <net/checksum.h>
66 #include <asm/uaccess.h>
67 #include <asm/system.h>
71 static struct kmem_cache *skbuff_head_cache __read_mostly;
72 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
75 * Keep out-of-line to prevent kernel bloat.
76 * __builtin_return_address is not used because it is not always
81 * skb_over_panic - private function
86 * Out of line support code for skb_put(). Not user callable.
88 void skb_over_panic(struct sk_buff *skb, int sz, void *here)
90 printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
91 "data:%p tail:%#lx end:%#lx dev:%s\n",
92 here, skb->len, sz, skb->head, skb->data,
93 (unsigned long)skb->tail, (unsigned long)skb->end,
94 skb->dev ? skb->dev->name : "<NULL>");
99 * skb_under_panic - private function
104 * Out of line support code for skb_push(). Not user callable.
107 void skb_under_panic(struct sk_buff *skb, int sz, void *here)
109 printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
110 "data:%p tail:%#lx end:%#lx dev:%s\n",
111 here, skb->len, sz, skb->head, skb->data,
112 (unsigned long)skb->tail, (unsigned long)skb->end,
113 skb->dev ? skb->dev->name : "<NULL>");
117 void skb_truesize_bug(struct sk_buff *skb)
119 printk(KERN_ERR "SKB BUG: Invalid truesize (%u) "
120 "len=%u, sizeof(sk_buff)=%Zd\n",
121 skb->truesize, skb->len, sizeof(struct sk_buff));
123 EXPORT_SYMBOL(skb_truesize_bug);
125 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
126 * 'private' fields and also do memory statistics to find all the
132 * __alloc_skb - allocate a network buffer
133 * @size: size to allocate
134 * @gfp_mask: allocation mask
135 * @fclone: allocate from fclone cache instead of head cache
136 * and allocate a cloned (child) skb
137 * @node: numa node to allocate memory on
139 * Allocate a new &sk_buff. The returned buffer has no headroom and a
140 * tail room of size bytes. The object has a reference count of one.
141 * The return is the buffer. On a failure the return is %NULL.
143 * Buffers may only be allocated from interrupts using a @gfp_mask of
146 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
147 int fclone, int node)
149 struct kmem_cache *cache;
150 struct skb_shared_info *shinfo;
154 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
157 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
161 size = SKB_DATA_ALIGN(size);
162 data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info),
168 * See comment in sk_buff definition, just before the 'tail' member
170 memset(skb, 0, offsetof(struct sk_buff, tail));
171 skb->truesize = size + sizeof(struct sk_buff);
172 atomic_set(&skb->users, 1);
175 skb_reset_tail_pointer(skb);
176 skb->end = skb->tail + size;
177 /* make sure we initialize shinfo sequentially */
178 shinfo = skb_shinfo(skb);
179 atomic_set(&shinfo->dataref, 1);
180 shinfo->nr_frags = 0;
181 shinfo->gso_size = 0;
182 shinfo->gso_segs = 0;
183 shinfo->gso_type = 0;
184 shinfo->ip6_frag_id = 0;
185 shinfo->frag_list = NULL;
188 struct sk_buff *child = skb + 1;
189 atomic_t *fclone_ref = (atomic_t *) (child + 1);
191 skb->fclone = SKB_FCLONE_ORIG;
192 atomic_set(fclone_ref, 1);
194 child->fclone = SKB_FCLONE_UNAVAILABLE;
199 kmem_cache_free(cache, skb);
205 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
206 * @dev: network device to receive on
207 * @length: length to allocate
208 * @gfp_mask: get_free_pages mask, passed to alloc_skb
210 * Allocate a new &sk_buff and assign it a usage count of one. The
211 * buffer has unspecified headroom built in. Users should allocate
212 * the headroom they think they need without accounting for the
213 * built in space. The built in space is used for optimisations.
215 * %NULL is returned if there is no free memory.
217 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
218 unsigned int length, gfp_t gfp_mask)
220 int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1;
223 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, node);
225 skb_reserve(skb, NET_SKB_PAD);
231 static void skb_drop_list(struct sk_buff **listp)
233 struct sk_buff *list = *listp;
238 struct sk_buff *this = list;
244 static inline void skb_drop_fraglist(struct sk_buff *skb)
246 skb_drop_list(&skb_shinfo(skb)->frag_list);
249 static void skb_clone_fraglist(struct sk_buff *skb)
251 struct sk_buff *list;
253 for (list = skb_shinfo(skb)->frag_list; list; list = list->next)
257 static void skb_release_data(struct sk_buff *skb)
260 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
261 &skb_shinfo(skb)->dataref)) {
262 if (skb_shinfo(skb)->nr_frags) {
264 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
265 put_page(skb_shinfo(skb)->frags[i].page);
268 if (skb_shinfo(skb)->frag_list)
269 skb_drop_fraglist(skb);
276 * Free an skbuff by memory without cleaning the state.
278 void kfree_skbmem(struct sk_buff *skb)
280 struct sk_buff *other;
281 atomic_t *fclone_ref;
283 skb_release_data(skb);
284 switch (skb->fclone) {
285 case SKB_FCLONE_UNAVAILABLE:
286 kmem_cache_free(skbuff_head_cache, skb);
289 case SKB_FCLONE_ORIG:
290 fclone_ref = (atomic_t *) (skb + 2);
291 if (atomic_dec_and_test(fclone_ref))
292 kmem_cache_free(skbuff_fclone_cache, skb);
295 case SKB_FCLONE_CLONE:
296 fclone_ref = (atomic_t *) (skb + 1);
299 /* The clone portion is available for
300 * fast-cloning again.
302 skb->fclone = SKB_FCLONE_UNAVAILABLE;
304 if (atomic_dec_and_test(fclone_ref))
305 kmem_cache_free(skbuff_fclone_cache, other);
311 * __kfree_skb - private function
314 * Free an sk_buff. Release anything attached to the buffer.
315 * Clean the state. This is an internal helper function. Users should
316 * always call kfree_skb
319 void __kfree_skb(struct sk_buff *skb)
321 dst_release(skb->dst);
323 secpath_put(skb->sp);
325 if (skb->destructor) {
327 skb->destructor(skb);
329 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
330 nf_conntrack_put(skb->nfct);
331 nf_conntrack_put_reasm(skb->nfct_reasm);
333 #ifdef CONFIG_BRIDGE_NETFILTER
334 nf_bridge_put(skb->nf_bridge);
336 /* XXX: IS this still necessary? - JHS */
337 #ifdef CONFIG_NET_SCHED
339 #ifdef CONFIG_NET_CLS_ACT
348 * kfree_skb - free an sk_buff
349 * @skb: buffer to free
351 * Drop a reference to the buffer and free it if the usage count has
354 void kfree_skb(struct sk_buff *skb)
358 if (likely(atomic_read(&skb->users) == 1))
360 else if (likely(!atomic_dec_and_test(&skb->users)))
366 * skb_clone - duplicate an sk_buff
367 * @skb: buffer to clone
368 * @gfp_mask: allocation priority
370 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
371 * copies share the same packet data but not structure. The new
372 * buffer has a reference count of 1. If the allocation fails the
373 * function returns %NULL otherwise the new buffer is returned.
375 * If this function is called from an interrupt gfp_mask() must be
379 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
384 if (skb->fclone == SKB_FCLONE_ORIG &&
385 n->fclone == SKB_FCLONE_UNAVAILABLE) {
386 atomic_t *fclone_ref = (atomic_t *) (n + 1);
387 n->fclone = SKB_FCLONE_CLONE;
388 atomic_inc(fclone_ref);
390 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
393 n->fclone = SKB_FCLONE_UNAVAILABLE;
396 #define C(x) n->x = skb->x
398 n->next = n->prev = NULL;
409 secpath_get(skb->sp);
411 memcpy(n->cb, skb->cb, sizeof(skb->cb));
418 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
422 skb_copy_queue_mapping(n, skb);
424 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
428 n->destructor = NULL;
431 #ifdef CONFIG_NET_SCHED
433 #ifdef CONFIG_NET_CLS_ACT
434 n->tc_verd = SET_TC_VERD(skb->tc_verd,0);
435 n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd);
436 n->tc_verd = CLR_TC_MUNGED(n->tc_verd);
440 skb_copy_secmark(n, skb);
442 atomic_set(&n->users, 1);
448 atomic_inc(&(skb_shinfo(skb)->dataref));
454 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
456 #ifndef NET_SKBUFF_DATA_USES_OFFSET
458 * Shift between the two data areas in bytes
460 unsigned long offset = new->data - old->data;
464 skb_copy_queue_mapping(new, old);
465 new->priority = old->priority;
466 new->protocol = old->protocol;
467 new->dst = dst_clone(old->dst);
469 new->sp = secpath_get(old->sp);
471 new->transport_header = old->transport_header;
472 new->network_header = old->network_header;
473 new->mac_header = old->mac_header;
474 #ifndef NET_SKBUFF_DATA_USES_OFFSET
475 /* {transport,network,mac}_header are relative to skb->head */
476 new->transport_header += offset;
477 new->network_header += offset;
478 new->mac_header += offset;
480 memcpy(new->cb, old->cb, sizeof(old->cb));
481 new->local_df = old->local_df;
482 new->fclone = SKB_FCLONE_UNAVAILABLE;
483 new->pkt_type = old->pkt_type;
484 new->tstamp = old->tstamp;
485 new->destructor = NULL;
486 new->mark = old->mark;
488 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
489 new->ipvs_property = old->ipvs_property;
491 #ifdef CONFIG_NET_SCHED
492 #ifdef CONFIG_NET_CLS_ACT
493 new->tc_verd = old->tc_verd;
495 new->tc_index = old->tc_index;
497 skb_copy_secmark(new, old);
498 atomic_set(&new->users, 1);
499 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
500 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
501 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
505 * skb_copy - create private copy of an sk_buff
506 * @skb: buffer to copy
507 * @gfp_mask: allocation priority
509 * Make a copy of both an &sk_buff and its data. This is used when the
510 * caller wishes to modify the data and needs a private copy of the
511 * data to alter. Returns %NULL on failure or the pointer to the buffer
512 * on success. The returned buffer has a reference count of 1.
514 * As by-product this function converts non-linear &sk_buff to linear
515 * one, so that &sk_buff becomes completely private and caller is allowed
516 * to modify all the data of returned buffer. This means that this
517 * function is not recommended for use in circumstances when only
518 * header is going to be modified. Use pskb_copy() instead.
521 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
523 int headerlen = skb->data - skb->head;
525 * Allocate the copy buffer
528 #ifdef NET_SKBUFF_DATA_USES_OFFSET
529 n = alloc_skb(skb->end + skb->data_len, gfp_mask);
531 n = alloc_skb(skb->end - skb->head + skb->data_len, gfp_mask);
536 /* Set the data pointer */
537 skb_reserve(n, headerlen);
538 /* Set the tail pointer and length */
539 skb_put(n, skb->len);
541 n->ip_summed = skb->ip_summed;
543 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
546 copy_skb_header(n, skb);
552 * pskb_copy - create copy of an sk_buff with private head.
553 * @skb: buffer to copy
554 * @gfp_mask: allocation priority
556 * Make a copy of both an &sk_buff and part of its data, located
557 * in header. Fragmented data remain shared. This is used when
558 * the caller wishes to modify only header of &sk_buff and needs
559 * private copy of the header to alter. Returns %NULL on failure
560 * or the pointer to the buffer on success.
561 * The returned buffer has a reference count of 1.
564 struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
567 * Allocate the copy buffer
570 #ifdef NET_SKBUFF_DATA_USES_OFFSET
571 n = alloc_skb(skb->end, gfp_mask);
573 n = alloc_skb(skb->end - skb->head, gfp_mask);
578 /* Set the data pointer */
579 skb_reserve(n, skb->data - skb->head);
580 /* Set the tail pointer and length */
581 skb_put(n, skb_headlen(skb));
583 skb_copy_from_linear_data(skb, n->data, n->len);
585 n->ip_summed = skb->ip_summed;
587 n->truesize += skb->data_len;
588 n->data_len = skb->data_len;
591 if (skb_shinfo(skb)->nr_frags) {
594 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
595 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
596 get_page(skb_shinfo(n)->frags[i].page);
598 skb_shinfo(n)->nr_frags = i;
601 if (skb_shinfo(skb)->frag_list) {
602 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
603 skb_clone_fraglist(n);
606 copy_skb_header(n, skb);
612 * pskb_expand_head - reallocate header of &sk_buff
613 * @skb: buffer to reallocate
614 * @nhead: room to add at head
615 * @ntail: room to add at tail
616 * @gfp_mask: allocation priority
618 * Expands (or creates identical copy, if &nhead and &ntail are zero)
619 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
620 * reference count of 1. Returns zero in the case of success or error,
621 * if expansion failed. In the last case, &sk_buff is not changed.
623 * All the pointers pointing into skb header may change and must be
624 * reloaded after call to this function.
627 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
632 #ifdef NET_SKBUFF_DATA_USES_OFFSET
633 int size = nhead + skb->end + ntail;
635 int size = nhead + (skb->end - skb->head) + ntail;
642 size = SKB_DATA_ALIGN(size);
644 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
648 /* Copy only real data... and, alas, header. This should be
649 * optimized for the cases when header is void. */
650 #ifdef NET_SKBUFF_DATA_USES_OFFSET
651 memcpy(data + nhead, skb->head, skb->tail);
653 memcpy(data + nhead, skb->head, skb->tail - skb->head);
655 memcpy(data + size, skb_end_pointer(skb),
656 sizeof(struct skb_shared_info));
658 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
659 get_page(skb_shinfo(skb)->frags[i].page);
661 if (skb_shinfo(skb)->frag_list)
662 skb_clone_fraglist(skb);
664 skb_release_data(skb);
666 off = (data + nhead) - skb->head;
670 #ifdef NET_SKBUFF_DATA_USES_OFFSET
674 skb->end = skb->head + size;
676 /* {transport,network,mac}_header and tail are relative to skb->head */
678 skb->transport_header += off;
679 skb->network_header += off;
680 skb->mac_header += off;
684 atomic_set(&skb_shinfo(skb)->dataref, 1);
691 /* Make private copy of skb with writable head and some headroom */
693 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
695 struct sk_buff *skb2;
696 int delta = headroom - skb_headroom(skb);
699 skb2 = pskb_copy(skb, GFP_ATOMIC);
701 skb2 = skb_clone(skb, GFP_ATOMIC);
702 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
713 * skb_copy_expand - copy and expand sk_buff
714 * @skb: buffer to copy
715 * @newheadroom: new free bytes at head
716 * @newtailroom: new free bytes at tail
717 * @gfp_mask: allocation priority
719 * Make a copy of both an &sk_buff and its data and while doing so
720 * allocate additional space.
722 * This is used when the caller wishes to modify the data and needs a
723 * private copy of the data to alter as well as more space for new fields.
724 * Returns %NULL on failure or the pointer to the buffer
725 * on success. The returned buffer has a reference count of 1.
727 * You must pass %GFP_ATOMIC as the allocation priority if this function
728 * is called from an interrupt.
730 * BUG ALERT: ip_summed is not copied. Why does this work? Is it used
731 * only by netfilter in the cases when checksum is recalculated? --ANK
733 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
734 int newheadroom, int newtailroom,
738 * Allocate the copy buffer
740 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
742 int oldheadroom = skb_headroom(skb);
743 int head_copy_len, head_copy_off;
749 skb_reserve(n, newheadroom);
751 /* Set the tail pointer and length */
752 skb_put(n, skb->len);
754 head_copy_len = oldheadroom;
756 if (newheadroom <= head_copy_len)
757 head_copy_len = newheadroom;
759 head_copy_off = newheadroom - head_copy_len;
761 /* Copy the linear header and data. */
762 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
763 skb->len + head_copy_len))
766 copy_skb_header(n, skb);
768 #ifdef NET_SKBUFF_DATA_USES_OFFSET
769 off = newheadroom - oldheadroom;
771 n->transport_header += off;
772 n->network_header += off;
773 n->mac_header += off;
779 * skb_pad - zero pad the tail of an skb
780 * @skb: buffer to pad
783 * Ensure that a buffer is followed by a padding area that is zero
784 * filled. Used by network drivers which may DMA or transfer data
785 * beyond the buffer end onto the wire.
787 * May return error in out of memory cases. The skb is freed on error.
790 int skb_pad(struct sk_buff *skb, int pad)
795 /* If the skbuff is non linear tailroom is always zero.. */
796 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
797 memset(skb->data+skb->len, 0, pad);
801 ntail = skb->data_len + pad - (skb->end - skb->tail);
802 if (likely(skb_cloned(skb) || ntail > 0)) {
803 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
808 /* FIXME: The use of this function with non-linear skb's really needs
811 err = skb_linearize(skb);
815 memset(skb->data + skb->len, 0, pad);
823 /* Trims skb to length len. It can change skb pointers.
826 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
828 struct sk_buff **fragp;
829 struct sk_buff *frag;
830 int offset = skb_headlen(skb);
831 int nfrags = skb_shinfo(skb)->nr_frags;
835 if (skb_cloned(skb) &&
836 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
843 for (; i < nfrags; i++) {
844 int end = offset + skb_shinfo(skb)->frags[i].size;
851 skb_shinfo(skb)->frags[i++].size = len - offset;
854 skb_shinfo(skb)->nr_frags = i;
856 for (; i < nfrags; i++)
857 put_page(skb_shinfo(skb)->frags[i].page);
859 if (skb_shinfo(skb)->frag_list)
860 skb_drop_fraglist(skb);
864 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
865 fragp = &frag->next) {
866 int end = offset + frag->len;
868 if (skb_shared(frag)) {
869 struct sk_buff *nfrag;
871 nfrag = skb_clone(frag, GFP_ATOMIC);
872 if (unlikely(!nfrag))
875 nfrag->next = frag->next;
887 unlikely((err = pskb_trim(frag, len - offset))))
891 skb_drop_list(&frag->next);
896 if (len > skb_headlen(skb)) {
897 skb->data_len -= skb->len - len;
902 skb_set_tail_pointer(skb, len);
909 * __pskb_pull_tail - advance tail of skb header
910 * @skb: buffer to reallocate
911 * @delta: number of bytes to advance tail
913 * The function makes a sense only on a fragmented &sk_buff,
914 * it expands header moving its tail forward and copying necessary
915 * data from fragmented part.
917 * &sk_buff MUST have reference count of 1.
919 * Returns %NULL (and &sk_buff does not change) if pull failed
920 * or value of new tail of skb in the case of success.
922 * All the pointers pointing into skb header may change and must be
923 * reloaded after call to this function.
926 /* Moves tail of skb head forward, copying data from fragmented part,
927 * when it is necessary.
928 * 1. It may fail due to malloc failure.
929 * 2. It may change skb pointers.
931 * It is pretty complicated. Luckily, it is called only in exceptional cases.
933 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
935 /* If skb has not enough free space at tail, get new one
936 * plus 128 bytes for future expansions. If we have enough
937 * room at tail, reallocate without expansion only if skb is cloned.
939 int i, k, eat = (skb->tail + delta) - skb->end;
941 if (eat > 0 || skb_cloned(skb)) {
942 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
947 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
950 /* Optimization: no fragments, no reasons to preestimate
951 * size of pulled pages. Superb.
953 if (!skb_shinfo(skb)->frag_list)
956 /* Estimate size of pulled pages. */
958 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
959 if (skb_shinfo(skb)->frags[i].size >= eat)
961 eat -= skb_shinfo(skb)->frags[i].size;
964 /* If we need update frag list, we are in troubles.
965 * Certainly, it possible to add an offset to skb data,
966 * but taking into account that pulling is expected to
967 * be very rare operation, it is worth to fight against
968 * further bloating skb head and crucify ourselves here instead.
969 * Pure masohism, indeed. 8)8)
972 struct sk_buff *list = skb_shinfo(skb)->frag_list;
973 struct sk_buff *clone = NULL;
974 struct sk_buff *insp = NULL;
979 if (list->len <= eat) {
980 /* Eaten as whole. */
985 /* Eaten partially. */
987 if (skb_shared(list)) {
988 /* Sucks! We need to fork list. :-( */
989 clone = skb_clone(list, GFP_ATOMIC);
995 /* This may be pulled without
999 if (!pskb_pull(list, eat)) {
1008 /* Free pulled out fragments. */
1009 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1010 skb_shinfo(skb)->frag_list = list->next;
1013 /* And insert new clone at head. */
1016 skb_shinfo(skb)->frag_list = clone;
1019 /* Success! Now we may commit changes to skb data. */
1024 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1025 if (skb_shinfo(skb)->frags[i].size <= eat) {
1026 put_page(skb_shinfo(skb)->frags[i].page);
1027 eat -= skb_shinfo(skb)->frags[i].size;
1029 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1031 skb_shinfo(skb)->frags[k].page_offset += eat;
1032 skb_shinfo(skb)->frags[k].size -= eat;
1038 skb_shinfo(skb)->nr_frags = k;
1041 skb->data_len -= delta;
1043 return skb_tail_pointer(skb);
1046 /* Copy some data bits from skb to kernel buffer. */
1048 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1051 int start = skb_headlen(skb);
1053 if (offset > (int)skb->len - len)
1057 if ((copy = start - offset) > 0) {
1060 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1061 if ((len -= copy) == 0)
1067 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1070 BUG_TRAP(start <= offset + len);
1072 end = start + skb_shinfo(skb)->frags[i].size;
1073 if ((copy = end - offset) > 0) {
1079 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1081 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1082 offset - start, copy);
1083 kunmap_skb_frag(vaddr);
1085 if ((len -= copy) == 0)
1093 if (skb_shinfo(skb)->frag_list) {
1094 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1096 for (; list; list = list->next) {
1099 BUG_TRAP(start <= offset + len);
1101 end = start + list->len;
1102 if ((copy = end - offset) > 0) {
1105 if (skb_copy_bits(list, offset - start,
1108 if ((len -= copy) == 0)
1124 * skb_store_bits - store bits from kernel buffer to skb
1125 * @skb: destination buffer
1126 * @offset: offset in destination
1127 * @from: source buffer
1128 * @len: number of bytes to copy
1130 * Copy the specified number of bytes from the source buffer to the
1131 * destination skb. This function handles all the messy bits of
1132 * traversing fragment lists and such.
1135 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1138 int start = skb_headlen(skb);
1140 if (offset > (int)skb->len - len)
1143 if ((copy = start - offset) > 0) {
1146 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1147 if ((len -= copy) == 0)
1153 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1154 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1157 BUG_TRAP(start <= offset + len);
1159 end = start + frag->size;
1160 if ((copy = end - offset) > 0) {
1166 vaddr = kmap_skb_frag(frag);
1167 memcpy(vaddr + frag->page_offset + offset - start,
1169 kunmap_skb_frag(vaddr);
1171 if ((len -= copy) == 0)
1179 if (skb_shinfo(skb)->frag_list) {
1180 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1182 for (; list; list = list->next) {
1185 BUG_TRAP(start <= offset + len);
1187 end = start + list->len;
1188 if ((copy = end - offset) > 0) {
1191 if (skb_store_bits(list, offset - start,
1194 if ((len -= copy) == 0)
1209 EXPORT_SYMBOL(skb_store_bits);
1211 /* Checksum skb data. */
1213 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1214 int len, __wsum csum)
1216 int start = skb_headlen(skb);
1217 int i, copy = start - offset;
1220 /* Checksum header. */
1224 csum = csum_partial(skb->data + offset, copy, csum);
1225 if ((len -= copy) == 0)
1231 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1234 BUG_TRAP(start <= offset + len);
1236 end = start + skb_shinfo(skb)->frags[i].size;
1237 if ((copy = end - offset) > 0) {
1240 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1244 vaddr = kmap_skb_frag(frag);
1245 csum2 = csum_partial(vaddr + frag->page_offset +
1246 offset - start, copy, 0);
1247 kunmap_skb_frag(vaddr);
1248 csum = csum_block_add(csum, csum2, pos);
1257 if (skb_shinfo(skb)->frag_list) {
1258 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1260 for (; list; list = list->next) {
1263 BUG_TRAP(start <= offset + len);
1265 end = start + list->len;
1266 if ((copy = end - offset) > 0) {
1270 csum2 = skb_checksum(list, offset - start,
1272 csum = csum_block_add(csum, csum2, pos);
1273 if ((len -= copy) == 0)
1286 /* Both of above in one bottle. */
1288 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1289 u8 *to, int len, __wsum csum)
1291 int start = skb_headlen(skb);
1292 int i, copy = start - offset;
1299 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1301 if ((len -= copy) == 0)
1308 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1311 BUG_TRAP(start <= offset + len);
1313 end = start + skb_shinfo(skb)->frags[i].size;
1314 if ((copy = end - offset) > 0) {
1317 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1321 vaddr = kmap_skb_frag(frag);
1322 csum2 = csum_partial_copy_nocheck(vaddr +
1326 kunmap_skb_frag(vaddr);
1327 csum = csum_block_add(csum, csum2, pos);
1337 if (skb_shinfo(skb)->frag_list) {
1338 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1340 for (; list; list = list->next) {
1344 BUG_TRAP(start <= offset + len);
1346 end = start + list->len;
1347 if ((copy = end - offset) > 0) {
1350 csum2 = skb_copy_and_csum_bits(list,
1353 csum = csum_block_add(csum, csum2, pos);
1354 if ((len -= copy) == 0)
1367 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1372 if (skb->ip_summed == CHECKSUM_PARTIAL)
1373 csstart = skb->csum_start - skb_headroom(skb);
1375 csstart = skb_headlen(skb);
1377 BUG_ON(csstart > skb_headlen(skb));
1379 skb_copy_from_linear_data(skb, to, csstart);
1382 if (csstart != skb->len)
1383 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1384 skb->len - csstart, 0);
1386 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1387 long csstuff = csstart + skb->csum_offset;
1389 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
1394 * skb_dequeue - remove from the head of the queue
1395 * @list: list to dequeue from
1397 * Remove the head of the list. The list lock is taken so the function
1398 * may be used safely with other locking list functions. The head item is
1399 * returned or %NULL if the list is empty.
1402 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1404 unsigned long flags;
1405 struct sk_buff *result;
1407 spin_lock_irqsave(&list->lock, flags);
1408 result = __skb_dequeue(list);
1409 spin_unlock_irqrestore(&list->lock, flags);
1414 * skb_dequeue_tail - remove from the tail of the queue
1415 * @list: list to dequeue from
1417 * Remove the tail of the list. The list lock is taken so the function
1418 * may be used safely with other locking list functions. The tail item is
1419 * returned or %NULL if the list is empty.
1421 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1423 unsigned long flags;
1424 struct sk_buff *result;
1426 spin_lock_irqsave(&list->lock, flags);
1427 result = __skb_dequeue_tail(list);
1428 spin_unlock_irqrestore(&list->lock, flags);
1433 * skb_queue_purge - empty a list
1434 * @list: list to empty
1436 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1437 * the list and one reference dropped. This function takes the list
1438 * lock and is atomic with respect to other list locking functions.
1440 void skb_queue_purge(struct sk_buff_head *list)
1442 struct sk_buff *skb;
1443 while ((skb = skb_dequeue(list)) != NULL)
1448 * skb_queue_head - queue a buffer at the list head
1449 * @list: list to use
1450 * @newsk: buffer to queue
1452 * Queue a buffer at the start of the list. This function takes the
1453 * list lock and can be used safely with other locking &sk_buff functions
1456 * A buffer cannot be placed on two lists at the same time.
1458 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1460 unsigned long flags;
1462 spin_lock_irqsave(&list->lock, flags);
1463 __skb_queue_head(list, newsk);
1464 spin_unlock_irqrestore(&list->lock, flags);
1468 * skb_queue_tail - queue a buffer at the list tail
1469 * @list: list to use
1470 * @newsk: buffer to queue
1472 * Queue a buffer at the tail of the list. This function takes the
1473 * list lock and can be used safely with other locking &sk_buff functions
1476 * A buffer cannot be placed on two lists at the same time.
1478 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1480 unsigned long flags;
1482 spin_lock_irqsave(&list->lock, flags);
1483 __skb_queue_tail(list, newsk);
1484 spin_unlock_irqrestore(&list->lock, flags);
1488 * skb_unlink - remove a buffer from a list
1489 * @skb: buffer to remove
1490 * @list: list to use
1492 * Remove a packet from a list. The list locks are taken and this
1493 * function is atomic with respect to other list locked calls
1495 * You must know what list the SKB is on.
1497 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1499 unsigned long flags;
1501 spin_lock_irqsave(&list->lock, flags);
1502 __skb_unlink(skb, list);
1503 spin_unlock_irqrestore(&list->lock, flags);
1507 * skb_append - append a buffer
1508 * @old: buffer to insert after
1509 * @newsk: buffer to insert
1510 * @list: list to use
1512 * Place a packet after a given packet in a list. The list locks are taken
1513 * and this function is atomic with respect to other list locked calls.
1514 * A buffer cannot be placed on two lists at the same time.
1516 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1518 unsigned long flags;
1520 spin_lock_irqsave(&list->lock, flags);
1521 __skb_append(old, newsk, list);
1522 spin_unlock_irqrestore(&list->lock, flags);
1527 * skb_insert - insert a buffer
1528 * @old: buffer to insert before
1529 * @newsk: buffer to insert
1530 * @list: list to use
1532 * Place a packet before a given packet in a list. The list locks are
1533 * taken and this function is atomic with respect to other list locked
1536 * A buffer cannot be placed on two lists at the same time.
1538 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1540 unsigned long flags;
1542 spin_lock_irqsave(&list->lock, flags);
1543 __skb_insert(newsk, old->prev, old, list);
1544 spin_unlock_irqrestore(&list->lock, flags);
1547 static inline void skb_split_inside_header(struct sk_buff *skb,
1548 struct sk_buff* skb1,
1549 const u32 len, const int pos)
1553 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
1555 /* And move data appendix as is. */
1556 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1557 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
1559 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
1560 skb_shinfo(skb)->nr_frags = 0;
1561 skb1->data_len = skb->data_len;
1562 skb1->len += skb1->data_len;
1565 skb_set_tail_pointer(skb, len);
1568 static inline void skb_split_no_header(struct sk_buff *skb,
1569 struct sk_buff* skb1,
1570 const u32 len, int pos)
1573 const int nfrags = skb_shinfo(skb)->nr_frags;
1575 skb_shinfo(skb)->nr_frags = 0;
1576 skb1->len = skb1->data_len = skb->len - len;
1578 skb->data_len = len - pos;
1580 for (i = 0; i < nfrags; i++) {
1581 int size = skb_shinfo(skb)->frags[i].size;
1583 if (pos + size > len) {
1584 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
1588 * We have two variants in this case:
1589 * 1. Move all the frag to the second
1590 * part, if it is possible. F.e.
1591 * this approach is mandatory for TUX,
1592 * where splitting is expensive.
1593 * 2. Split is accurately. We make this.
1595 get_page(skb_shinfo(skb)->frags[i].page);
1596 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
1597 skb_shinfo(skb1)->frags[0].size -= len - pos;
1598 skb_shinfo(skb)->frags[i].size = len - pos;
1599 skb_shinfo(skb)->nr_frags++;
1603 skb_shinfo(skb)->nr_frags++;
1606 skb_shinfo(skb1)->nr_frags = k;
1610 * skb_split - Split fragmented skb to two parts at length len.
1611 * @skb: the buffer to split
1612 * @skb1: the buffer to receive the second part
1613 * @len: new length for skb
1615 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
1617 int pos = skb_headlen(skb);
1619 if (len < pos) /* Split line is inside header. */
1620 skb_split_inside_header(skb, skb1, len, pos);
1621 else /* Second chunk has no header, nothing to copy. */
1622 skb_split_no_header(skb, skb1, len, pos);
1626 * skb_prepare_seq_read - Prepare a sequential read of skb data
1627 * @skb: the buffer to read
1628 * @from: lower offset of data to be read
1629 * @to: upper offset of data to be read
1630 * @st: state variable
1632 * Initializes the specified state variable. Must be called before
1633 * invoking skb_seq_read() for the first time.
1635 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1636 unsigned int to, struct skb_seq_state *st)
1638 st->lower_offset = from;
1639 st->upper_offset = to;
1640 st->root_skb = st->cur_skb = skb;
1641 st->frag_idx = st->stepped_offset = 0;
1642 st->frag_data = NULL;
1646 * skb_seq_read - Sequentially read skb data
1647 * @consumed: number of bytes consumed by the caller so far
1648 * @data: destination pointer for data to be returned
1649 * @st: state variable
1651 * Reads a block of skb data at &consumed relative to the
1652 * lower offset specified to skb_prepare_seq_read(). Assigns
1653 * the head of the data block to &data and returns the length
1654 * of the block or 0 if the end of the skb data or the upper
1655 * offset has been reached.
1657 * The caller is not required to consume all of the data
1658 * returned, i.e. &consumed is typically set to the number
1659 * of bytes already consumed and the next call to
1660 * skb_seq_read() will return the remaining part of the block.
1662 * Note: The size of each block of data returned can be arbitary,
1663 * this limitation is the cost for zerocopy seqeuental
1664 * reads of potentially non linear data.
1666 * Note: Fragment lists within fragments are not implemented
1667 * at the moment, state->root_skb could be replaced with
1668 * a stack for this purpose.
1670 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1671 struct skb_seq_state *st)
1673 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
1676 if (unlikely(abs_offset >= st->upper_offset))
1680 block_limit = skb_headlen(st->cur_skb);
1682 if (abs_offset < block_limit) {
1683 *data = st->cur_skb->data + abs_offset;
1684 return block_limit - abs_offset;
1687 if (st->frag_idx == 0 && !st->frag_data)
1688 st->stepped_offset += skb_headlen(st->cur_skb);
1690 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
1691 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
1692 block_limit = frag->size + st->stepped_offset;
1694 if (abs_offset < block_limit) {
1696 st->frag_data = kmap_skb_frag(frag);
1698 *data = (u8 *) st->frag_data + frag->page_offset +
1699 (abs_offset - st->stepped_offset);
1701 return block_limit - abs_offset;
1704 if (st->frag_data) {
1705 kunmap_skb_frag(st->frag_data);
1706 st->frag_data = NULL;
1710 st->stepped_offset += frag->size;
1713 if (st->frag_data) {
1714 kunmap_skb_frag(st->frag_data);
1715 st->frag_data = NULL;
1718 if (st->cur_skb->next) {
1719 st->cur_skb = st->cur_skb->next;
1722 } else if (st->root_skb == st->cur_skb &&
1723 skb_shinfo(st->root_skb)->frag_list) {
1724 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
1732 * skb_abort_seq_read - Abort a sequential read of skb data
1733 * @st: state variable
1735 * Must be called if skb_seq_read() was not called until it
1738 void skb_abort_seq_read(struct skb_seq_state *st)
1741 kunmap_skb_frag(st->frag_data);
1744 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
1746 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
1747 struct ts_config *conf,
1748 struct ts_state *state)
1750 return skb_seq_read(offset, text, TS_SKB_CB(state));
1753 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
1755 skb_abort_seq_read(TS_SKB_CB(state));
1759 * skb_find_text - Find a text pattern in skb data
1760 * @skb: the buffer to look in
1761 * @from: search offset
1763 * @config: textsearch configuration
1764 * @state: uninitialized textsearch state variable
1766 * Finds a pattern in the skb data according to the specified
1767 * textsearch configuration. Use textsearch_next() to retrieve
1768 * subsequent occurrences of the pattern. Returns the offset
1769 * to the first occurrence or UINT_MAX if no match was found.
1771 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1772 unsigned int to, struct ts_config *config,
1773 struct ts_state *state)
1777 config->get_next_block = skb_ts_get_next_block;
1778 config->finish = skb_ts_finish;
1780 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
1782 ret = textsearch_find(config, state);
1783 return (ret <= to - from ? ret : UINT_MAX);
1787 * skb_append_datato_frags: - append the user data to a skb
1788 * @sk: sock structure
1789 * @skb: skb structure to be appened with user data.
1790 * @getfrag: call back function to be used for getting the user data
1791 * @from: pointer to user message iov
1792 * @length: length of the iov message
1794 * Description: This procedure append the user data in the fragment part
1795 * of the skb if any page alloc fails user this procedure returns -ENOMEM
1797 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
1798 int (*getfrag)(void *from, char *to, int offset,
1799 int len, int odd, struct sk_buff *skb),
1800 void *from, int length)
1803 skb_frag_t *frag = NULL;
1804 struct page *page = NULL;
1810 /* Return error if we don't have space for new frag */
1811 frg_cnt = skb_shinfo(skb)->nr_frags;
1812 if (frg_cnt >= MAX_SKB_FRAGS)
1815 /* allocate a new page for next frag */
1816 page = alloc_pages(sk->sk_allocation, 0);
1818 /* If alloc_page fails just return failure and caller will
1819 * free previous allocated pages by doing kfree_skb()
1824 /* initialize the next frag */
1825 sk->sk_sndmsg_page = page;
1826 sk->sk_sndmsg_off = 0;
1827 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
1828 skb->truesize += PAGE_SIZE;
1829 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1831 /* get the new initialized frag */
1832 frg_cnt = skb_shinfo(skb)->nr_frags;
1833 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
1835 /* copy the user data to page */
1836 left = PAGE_SIZE - frag->page_offset;
1837 copy = (length > left)? left : length;
1839 ret = getfrag(from, (page_address(frag->page) +
1840 frag->page_offset + frag->size),
1841 offset, copy, 0, skb);
1845 /* copy was successful so update the size parameters */
1846 sk->sk_sndmsg_off += copy;
1849 skb->data_len += copy;
1853 } while (length > 0);
1859 * skb_pull_rcsum - pull skb and update receive checksum
1860 * @skb: buffer to update
1861 * @start: start of data before pull
1862 * @len: length of data pulled
1864 * This function performs an skb_pull on the packet and updates
1865 * update the CHECKSUM_COMPLETE checksum. It should be used on
1866 * receive path processing instead of skb_pull unless you know
1867 * that the checksum difference is zero (e.g., a valid IP header)
1868 * or you are setting ip_summed to CHECKSUM_NONE.
1870 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
1872 BUG_ON(len > skb->len);
1874 BUG_ON(skb->len < skb->data_len);
1875 skb_postpull_rcsum(skb, skb->data, len);
1876 return skb->data += len;
1879 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
1882 * skb_segment - Perform protocol segmentation on skb.
1883 * @skb: buffer to segment
1884 * @features: features for the output path (see dev->features)
1886 * This function performs segmentation on the given skb. It returns
1887 * the segment at the given position. It returns NULL if there are
1888 * no more segments to generate, or when an error is encountered.
1890 struct sk_buff *skb_segment(struct sk_buff *skb, int features)
1892 struct sk_buff *segs = NULL;
1893 struct sk_buff *tail = NULL;
1894 unsigned int mss = skb_shinfo(skb)->gso_size;
1895 unsigned int doffset = skb->data - skb_mac_header(skb);
1896 unsigned int offset = doffset;
1897 unsigned int headroom;
1899 int sg = features & NETIF_F_SG;
1900 int nfrags = skb_shinfo(skb)->nr_frags;
1905 __skb_push(skb, doffset);
1906 headroom = skb_headroom(skb);
1907 pos = skb_headlen(skb);
1910 struct sk_buff *nskb;
1916 len = skb->len - offset;
1920 hsize = skb_headlen(skb) - offset;
1923 if (hsize > len || !sg)
1926 nskb = alloc_skb(hsize + doffset + headroom, GFP_ATOMIC);
1927 if (unlikely(!nskb))
1936 nskb->dev = skb->dev;
1937 skb_copy_queue_mapping(nskb, skb);
1938 nskb->priority = skb->priority;
1939 nskb->protocol = skb->protocol;
1940 nskb->dst = dst_clone(skb->dst);
1941 memcpy(nskb->cb, skb->cb, sizeof(skb->cb));
1942 nskb->pkt_type = skb->pkt_type;
1943 nskb->mac_len = skb->mac_len;
1945 skb_reserve(nskb, headroom);
1946 skb_reset_mac_header(nskb);
1947 skb_set_network_header(nskb, skb->mac_len);
1948 nskb->transport_header = (nskb->network_header +
1949 skb_network_header_len(skb));
1950 skb_copy_from_linear_data(skb, skb_put(nskb, doffset),
1953 nskb->csum = skb_copy_and_csum_bits(skb, offset,
1959 frag = skb_shinfo(nskb)->frags;
1962 nskb->ip_summed = CHECKSUM_PARTIAL;
1963 nskb->csum = skb->csum;
1964 skb_copy_from_linear_data_offset(skb, offset,
1965 skb_put(nskb, hsize), hsize);
1967 while (pos < offset + len) {
1968 BUG_ON(i >= nfrags);
1970 *frag = skb_shinfo(skb)->frags[i];
1971 get_page(frag->page);
1975 frag->page_offset += offset - pos;
1976 frag->size -= offset - pos;
1981 if (pos + size <= offset + len) {
1985 frag->size -= pos + size - (offset + len);
1992 skb_shinfo(nskb)->nr_frags = k;
1993 nskb->data_len = len - hsize;
1994 nskb->len += nskb->data_len;
1995 nskb->truesize += nskb->data_len;
1996 } while ((offset += len) < skb->len);
2001 while ((skb = segs)) {
2005 return ERR_PTR(err);
2008 EXPORT_SYMBOL_GPL(skb_segment);
2010 void __init skb_init(void)
2012 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2013 sizeof(struct sk_buff),
2015 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2017 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2018 (2*sizeof(struct sk_buff)) +
2021 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2026 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2027 * @skb: Socket buffer containing the buffers to be mapped
2028 * @sg: The scatter-gather list to map into
2029 * @offset: The offset into the buffer's contents to start mapping
2030 * @len: Length of buffer space to be mapped
2032 * Fill the specified scatter-gather list with mappings/pointers into a
2033 * region of the buffer space attached to a socket buffer.
2036 skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2038 int start = skb_headlen(skb);
2039 int i, copy = start - offset;
2045 sg[elt].page = virt_to_page(skb->data + offset);
2046 sg[elt].offset = (unsigned long)(skb->data + offset) % PAGE_SIZE;
2047 sg[elt].length = copy;
2049 if ((len -= copy) == 0)
2054 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2057 BUG_TRAP(start <= offset + len);
2059 end = start + skb_shinfo(skb)->frags[i].size;
2060 if ((copy = end - offset) > 0) {
2061 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2065 sg[elt].page = frag->page;
2066 sg[elt].offset = frag->page_offset+offset-start;
2067 sg[elt].length = copy;
2076 if (skb_shinfo(skb)->frag_list) {
2077 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2079 for (; list; list = list->next) {
2082 BUG_TRAP(start <= offset + len);
2084 end = start + list->len;
2085 if ((copy = end - offset) > 0) {
2088 elt += skb_to_sgvec(list, sg+elt, offset - start, copy);
2089 if ((len -= copy) == 0)
2101 * skb_cow_data - Check that a socket buffer's data buffers are writable
2102 * @skb: The socket buffer to check.
2103 * @tailbits: Amount of trailing space to be added
2104 * @trailer: Returned pointer to the skb where the @tailbits space begins
2106 * Make sure that the data buffers attached to a socket buffer are
2107 * writable. If they are not, private copies are made of the data buffers
2108 * and the socket buffer is set to use these instead.
2110 * If @tailbits is given, make sure that there is space to write @tailbits
2111 * bytes of data beyond current end of socket buffer. @trailer will be
2112 * set to point to the skb in which this space begins.
2114 * The number of scatterlist elements required to completely map the
2115 * COW'd and extended socket buffer will be returned.
2117 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
2121 struct sk_buff *skb1, **skb_p;
2123 /* If skb is cloned or its head is paged, reallocate
2124 * head pulling out all the pages (pages are considered not writable
2125 * at the moment even if they are anonymous).
2127 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
2128 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
2131 /* Easy case. Most of packets will go this way. */
2132 if (!skb_shinfo(skb)->frag_list) {
2133 /* A little of trouble, not enough of space for trailer.
2134 * This should not happen, when stack is tuned to generate
2135 * good frames. OK, on miss we reallocate and reserve even more
2136 * space, 128 bytes is fair. */
2138 if (skb_tailroom(skb) < tailbits &&
2139 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
2147 /* Misery. We are in troubles, going to mincer fragments... */
2150 skb_p = &skb_shinfo(skb)->frag_list;
2153 while ((skb1 = *skb_p) != NULL) {
2156 /* The fragment is partially pulled by someone,
2157 * this can happen on input. Copy it and everything
2160 if (skb_shared(skb1))
2163 /* If the skb is the last, worry about trailer. */
2165 if (skb1->next == NULL && tailbits) {
2166 if (skb_shinfo(skb1)->nr_frags ||
2167 skb_shinfo(skb1)->frag_list ||
2168 skb_tailroom(skb1) < tailbits)
2169 ntail = tailbits + 128;
2175 skb_shinfo(skb1)->nr_frags ||
2176 skb_shinfo(skb1)->frag_list) {
2177 struct sk_buff *skb2;
2179 /* Fuck, we are miserable poor guys... */
2181 skb2 = skb_copy(skb1, GFP_ATOMIC);
2183 skb2 = skb_copy_expand(skb1,
2187 if (unlikely(skb2 == NULL))
2191 skb_set_owner_w(skb2, skb1->sk);
2193 /* Looking around. Are we still alive?
2194 * OK, link new skb, drop old one */
2196 skb2->next = skb1->next;
2203 skb_p = &skb1->next;
2209 EXPORT_SYMBOL(___pskb_trim);
2210 EXPORT_SYMBOL(__kfree_skb);
2211 EXPORT_SYMBOL(kfree_skb);
2212 EXPORT_SYMBOL(__pskb_pull_tail);
2213 EXPORT_SYMBOL(__alloc_skb);
2214 EXPORT_SYMBOL(__netdev_alloc_skb);
2215 EXPORT_SYMBOL(pskb_copy);
2216 EXPORT_SYMBOL(pskb_expand_head);
2217 EXPORT_SYMBOL(skb_checksum);
2218 EXPORT_SYMBOL(skb_clone);
2219 EXPORT_SYMBOL(skb_copy);
2220 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2221 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2222 EXPORT_SYMBOL(skb_copy_bits);
2223 EXPORT_SYMBOL(skb_copy_expand);
2224 EXPORT_SYMBOL(skb_over_panic);
2225 EXPORT_SYMBOL(skb_pad);
2226 EXPORT_SYMBOL(skb_realloc_headroom);
2227 EXPORT_SYMBOL(skb_under_panic);
2228 EXPORT_SYMBOL(skb_dequeue);
2229 EXPORT_SYMBOL(skb_dequeue_tail);
2230 EXPORT_SYMBOL(skb_insert);
2231 EXPORT_SYMBOL(skb_queue_purge);
2232 EXPORT_SYMBOL(skb_queue_head);
2233 EXPORT_SYMBOL(skb_queue_tail);
2234 EXPORT_SYMBOL(skb_unlink);
2235 EXPORT_SYMBOL(skb_append);
2236 EXPORT_SYMBOL(skb_split);
2237 EXPORT_SYMBOL(skb_prepare_seq_read);
2238 EXPORT_SYMBOL(skb_seq_read);
2239 EXPORT_SYMBOL(skb_abort_seq_read);
2240 EXPORT_SYMBOL(skb_find_text);
2241 EXPORT_SYMBOL(skb_append_datato_frags);
2243 EXPORT_SYMBOL_GPL(skb_to_sgvec);
2244 EXPORT_SYMBOL_GPL(skb_cow_data);