2 * Routines having to do with the 'struct sk_buff' memory handlers.
4 * Authors: Alan Cox <iiitac@pyr.swan.ac.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
7 * Version: $Id: skbuff.c,v 1.90 2001/11/07 05:56:19 davem Exp $
10 * Alan Cox : Fixed the worst of the load
12 * Dave Platt : Interrupt stacking fix.
13 * Richard Kooijman : Timestamp fixes.
14 * Alan Cox : Changed buffer format.
15 * Alan Cox : destructor hook for AF_UNIX etc.
16 * Linus Torvalds : Better skb_clone.
17 * Alan Cox : Added skb_copy.
18 * Alan Cox : Added all the changed routines Linus
19 * only put in the headers
20 * Ray VanTassle : Fixed --skb->lock in free
21 * Alan Cox : skb_copy copy arp field
22 * Andi Kleen : slabified it.
23 * Robert Olsson : Removed skb_head_pool
26 * The __skb_ routines should be called with interrupts
27 * disabled, or you better be *real* sure that the operation is atomic
28 * with respect to whatever list is being frobbed (e.g. via lock_sock()
29 * or via disabling bottom half handlers, etc).
31 * This program is free software; you can redistribute it and/or
32 * modify it under the terms of the GNU General Public License
33 * as published by the Free Software Foundation; either version
34 * 2 of the License, or (at your option) any later version.
38 * The functions in this file will not compile correctly with gcc 2.4.x
41 #include <linux/config.h>
42 #include <linux/module.h>
43 #include <linux/types.h>
44 #include <linux/kernel.h>
45 #include <linux/sched.h>
47 #include <linux/interrupt.h>
49 #include <linux/inet.h>
50 #include <linux/slab.h>
51 #include <linux/netdevice.h>
52 #ifdef CONFIG_NET_CLS_ACT
53 #include <net/pkt_sched.h>
55 #include <linux/string.h>
56 #include <linux/skbuff.h>
57 #include <linux/cache.h>
58 #include <linux/rtnetlink.h>
59 #include <linux/init.h>
60 #include <linux/highmem.h>
62 #include <net/protocol.h>
65 #include <net/checksum.h>
68 #include <asm/uaccess.h>
69 #include <asm/system.h>
71 static kmem_cache_t
*skbuff_head_cache __read_mostly
;
72 static kmem_cache_t
*skbuff_fclone_cache __read_mostly
;
75 * Keep out-of-line to prevent kernel bloat.
76 * __builtin_return_address is not used because it is not always
81 * skb_over_panic - private function
86 * Out of line support code for skb_put(). Not user callable.
88 void skb_over_panic(struct sk_buff
*skb
, int sz
, void *here
)
90 printk(KERN_EMERG
"skb_over_panic: text:%p len:%d put:%d head:%p "
91 "data:%p tail:%p end:%p dev:%s\n",
92 here
, skb
->len
, sz
, skb
->head
, skb
->data
, skb
->tail
, skb
->end
,
93 skb
->dev
? skb
->dev
->name
: "<NULL>");
98 * skb_under_panic - private function
103 * Out of line support code for skb_push(). Not user callable.
106 void skb_under_panic(struct sk_buff
*skb
, int sz
, void *here
)
108 printk(KERN_EMERG
"skb_under_panic: text:%p len:%d put:%d head:%p "
109 "data:%p tail:%p end:%p dev:%s\n",
110 here
, skb
->len
, sz
, skb
->head
, skb
->data
, skb
->tail
, skb
->end
,
111 skb
->dev
? skb
->dev
->name
: "<NULL>");
115 void skb_truesize_bug(struct sk_buff
*skb
)
117 printk(KERN_ERR
"SKB BUG: Invalid truesize (%u) "
118 "len=%u, sizeof(sk_buff)=%Zd\n",
119 skb
->truesize
, skb
->len
, sizeof(struct sk_buff
));
121 EXPORT_SYMBOL(skb_truesize_bug
);
123 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
124 * 'private' fields and also do memory statistics to find all the
130 * __alloc_skb - allocate a network buffer
131 * @size: size to allocate
132 * @gfp_mask: allocation mask
133 * @fclone: allocate from fclone cache instead of head cache
134 * and allocate a cloned (child) skb
136 * Allocate a new &sk_buff. The returned buffer has no headroom and a
137 * tail room of size bytes. The object has a reference count of one.
138 * The return is the buffer. On a failure the return is %NULL.
140 * Buffers may only be allocated from interrupts using a @gfp_mask of
143 struct sk_buff
*__alloc_skb(unsigned int size
, gfp_t gfp_mask
,
147 struct skb_shared_info
*shinfo
;
151 cache
= fclone
? skbuff_fclone_cache
: skbuff_head_cache
;
154 skb
= kmem_cache_alloc(cache
, gfp_mask
& ~__GFP_DMA
);
158 /* Get the DATA. Size must match skb_add_mtu(). */
159 size
= SKB_DATA_ALIGN(size
);
160 data
= ____kmalloc(size
+ sizeof(struct skb_shared_info
), gfp_mask
);
164 memset(skb
, 0, offsetof(struct sk_buff
, truesize
));
165 skb
->truesize
= size
+ sizeof(struct sk_buff
);
166 atomic_set(&skb
->users
, 1);
170 skb
->end
= data
+ size
;
171 /* make sure we initialize shinfo sequentially */
172 shinfo
= skb_shinfo(skb
);
173 atomic_set(&shinfo
->dataref
, 1);
174 shinfo
->nr_frags
= 0;
175 shinfo
->tso_size
= 0;
176 shinfo
->tso_segs
= 0;
177 shinfo
->ufo_size
= 0;
178 shinfo
->ip6_frag_id
= 0;
179 shinfo
->frag_list
= NULL
;
182 struct sk_buff
*child
= skb
+ 1;
183 atomic_t
*fclone_ref
= (atomic_t
*) (child
+ 1);
185 skb
->fclone
= SKB_FCLONE_ORIG
;
186 atomic_set(fclone_ref
, 1);
188 child
->fclone
= SKB_FCLONE_UNAVAILABLE
;
193 kmem_cache_free(cache
, skb
);
199 * alloc_skb_from_cache - allocate a network buffer
200 * @cp: kmem_cache from which to allocate the data area
201 * (object size must be big enough for @size bytes + skb overheads)
202 * @size: size to allocate
203 * @gfp_mask: allocation mask
205 * Allocate a new &sk_buff. The returned buffer has no headroom and
206 * tail room of size bytes. The object has a reference count of one.
207 * The return is the buffer. On a failure the return is %NULL.
209 * Buffers may only be allocated from interrupts using a @gfp_mask of
212 struct sk_buff
*alloc_skb_from_cache(kmem_cache_t
*cp
,
220 skb
= kmem_cache_alloc(skbuff_head_cache
,
221 gfp_mask
& ~__GFP_DMA
);
226 size
= SKB_DATA_ALIGN(size
);
227 data
= kmem_cache_alloc(cp
, gfp_mask
);
231 memset(skb
, 0, offsetof(struct sk_buff
, truesize
));
232 skb
->truesize
= size
+ sizeof(struct sk_buff
);
233 atomic_set(&skb
->users
, 1);
237 skb
->end
= data
+ size
;
239 atomic_set(&(skb_shinfo(skb
)->dataref
), 1);
240 skb_shinfo(skb
)->nr_frags
= 0;
241 skb_shinfo(skb
)->tso_size
= 0;
242 skb_shinfo(skb
)->tso_segs
= 0;
243 skb_shinfo(skb
)->frag_list
= NULL
;
247 kmem_cache_free(skbuff_head_cache
, skb
);
253 static void skb_drop_fraglist(struct sk_buff
*skb
)
255 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
257 skb_shinfo(skb
)->frag_list
= NULL
;
260 struct sk_buff
*this = list
;
266 static void skb_clone_fraglist(struct sk_buff
*skb
)
268 struct sk_buff
*list
;
270 for (list
= skb_shinfo(skb
)->frag_list
; list
; list
= list
->next
)
274 void skb_release_data(struct sk_buff
*skb
)
277 !atomic_sub_return(skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1,
278 &skb_shinfo(skb
)->dataref
)) {
279 if (skb_shinfo(skb
)->nr_frags
) {
281 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
282 put_page(skb_shinfo(skb
)->frags
[i
].page
);
285 if (skb_shinfo(skb
)->frag_list
)
286 skb_drop_fraglist(skb
);
293 * Free an skbuff by memory without cleaning the state.
295 void kfree_skbmem(struct sk_buff
*skb
)
297 struct sk_buff
*other
;
298 atomic_t
*fclone_ref
;
300 skb_release_data(skb
);
301 switch (skb
->fclone
) {
302 case SKB_FCLONE_UNAVAILABLE
:
303 kmem_cache_free(skbuff_head_cache
, skb
);
306 case SKB_FCLONE_ORIG
:
307 fclone_ref
= (atomic_t
*) (skb
+ 2);
308 if (atomic_dec_and_test(fclone_ref
))
309 kmem_cache_free(skbuff_fclone_cache
, skb
);
312 case SKB_FCLONE_CLONE
:
313 fclone_ref
= (atomic_t
*) (skb
+ 1);
316 /* The clone portion is available for
317 * fast-cloning again.
319 skb
->fclone
= SKB_FCLONE_UNAVAILABLE
;
321 if (atomic_dec_and_test(fclone_ref
))
322 kmem_cache_free(skbuff_fclone_cache
, other
);
328 * __kfree_skb - private function
331 * Free an sk_buff. Release anything attached to the buffer.
332 * Clean the state. This is an internal helper function. Users should
333 * always call kfree_skb
336 void __kfree_skb(struct sk_buff
*skb
)
338 dst_release(skb
->dst
);
340 secpath_put(skb
->sp
);
342 if (skb
->destructor
) {
344 skb
->destructor(skb
);
346 #ifdef CONFIG_NETFILTER
347 nf_conntrack_put(skb
->nfct
);
348 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
349 nf_conntrack_put_reasm(skb
->nfct_reasm
);
351 #ifdef CONFIG_BRIDGE_NETFILTER
352 nf_bridge_put(skb
->nf_bridge
);
355 /* XXX: IS this still necessary? - JHS */
356 #ifdef CONFIG_NET_SCHED
358 #ifdef CONFIG_NET_CLS_ACT
367 * kfree_skb - free an sk_buff
368 * @skb: buffer to free
370 * Drop a reference to the buffer and free it if the usage count has
373 void kfree_skb(struct sk_buff
*skb
)
377 if (likely(atomic_read(&skb
->users
) == 1))
379 else if (likely(!atomic_dec_and_test(&skb
->users
)))
385 * skb_clone - duplicate an sk_buff
386 * @skb: buffer to clone
387 * @gfp_mask: allocation priority
389 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
390 * copies share the same packet data but not structure. The new
391 * buffer has a reference count of 1. If the allocation fails the
392 * function returns %NULL otherwise the new buffer is returned.
394 * If this function is called from an interrupt gfp_mask() must be
398 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
403 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
404 n
->fclone
== SKB_FCLONE_UNAVAILABLE
) {
405 atomic_t
*fclone_ref
= (atomic_t
*) (n
+ 1);
406 n
->fclone
= SKB_FCLONE_CLONE
;
407 atomic_inc(fclone_ref
);
409 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
412 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
415 #define C(x) n->x = skb->x
417 n
->next
= n
->prev
= NULL
;
428 secpath_get(skb
->sp
);
430 memcpy(n
->cb
, skb
->cb
, sizeof(skb
->cb
));
440 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
444 n
->destructor
= NULL
;
445 #ifdef CONFIG_NETFILTER
448 nf_conntrack_get(skb
->nfct
);
450 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
452 nf_conntrack_get_reasm(skb
->nfct_reasm
);
454 #ifdef CONFIG_BRIDGE_NETFILTER
456 nf_bridge_get(skb
->nf_bridge
);
458 #endif /*CONFIG_NETFILTER*/
459 #ifdef CONFIG_NET_SCHED
461 #ifdef CONFIG_NET_CLS_ACT
462 n
->tc_verd
= SET_TC_VERD(skb
->tc_verd
,0);
463 n
->tc_verd
= CLR_TC_OK2MUNGE(n
->tc_verd
);
464 n
->tc_verd
= CLR_TC_MUNGED(n
->tc_verd
);
467 skb_copy_secmark(n
, skb
);
470 atomic_set(&n
->users
, 1);
476 atomic_inc(&(skb_shinfo(skb
)->dataref
));
482 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
485 * Shift between the two data areas in bytes
487 unsigned long offset
= new->data
- old
->data
;
491 new->priority
= old
->priority
;
492 new->protocol
= old
->protocol
;
493 new->dst
= dst_clone(old
->dst
);
495 new->sp
= secpath_get(old
->sp
);
497 new->h
.raw
= old
->h
.raw
+ offset
;
498 new->nh
.raw
= old
->nh
.raw
+ offset
;
499 new->mac
.raw
= old
->mac
.raw
+ offset
;
500 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
501 new->local_df
= old
->local_df
;
502 new->fclone
= SKB_FCLONE_UNAVAILABLE
;
503 new->pkt_type
= old
->pkt_type
;
504 new->tstamp
= old
->tstamp
;
505 new->destructor
= NULL
;
506 #ifdef CONFIG_NETFILTER
507 new->nfmark
= old
->nfmark
;
508 new->nfct
= old
->nfct
;
509 nf_conntrack_get(old
->nfct
);
510 new->nfctinfo
= old
->nfctinfo
;
511 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
512 new->nfct_reasm
= old
->nfct_reasm
;
513 nf_conntrack_get_reasm(old
->nfct_reasm
);
515 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
516 new->ipvs_property
= old
->ipvs_property
;
518 #ifdef CONFIG_BRIDGE_NETFILTER
519 new->nf_bridge
= old
->nf_bridge
;
520 nf_bridge_get(old
->nf_bridge
);
523 #ifdef CONFIG_NET_SCHED
524 #ifdef CONFIG_NET_CLS_ACT
525 new->tc_verd
= old
->tc_verd
;
527 new->tc_index
= old
->tc_index
;
529 skb_copy_secmark(new, old
);
530 atomic_set(&new->users
, 1);
531 skb_shinfo(new)->tso_size
= skb_shinfo(old
)->tso_size
;
532 skb_shinfo(new)->tso_segs
= skb_shinfo(old
)->tso_segs
;
536 * skb_copy - create private copy of an sk_buff
537 * @skb: buffer to copy
538 * @gfp_mask: allocation priority
540 * Make a copy of both an &sk_buff and its data. This is used when the
541 * caller wishes to modify the data and needs a private copy of the
542 * data to alter. Returns %NULL on failure or the pointer to the buffer
543 * on success. The returned buffer has a reference count of 1.
545 * As by-product this function converts non-linear &sk_buff to linear
546 * one, so that &sk_buff becomes completely private and caller is allowed
547 * to modify all the data of returned buffer. This means that this
548 * function is not recommended for use in circumstances when only
549 * header is going to be modified. Use pskb_copy() instead.
552 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
554 int headerlen
= skb
->data
- skb
->head
;
556 * Allocate the copy buffer
558 struct sk_buff
*n
= alloc_skb(skb
->end
- skb
->head
+ skb
->data_len
,
563 /* Set the data pointer */
564 skb_reserve(n
, headerlen
);
565 /* Set the tail pointer and length */
566 skb_put(n
, skb
->len
);
568 n
->ip_summed
= skb
->ip_summed
;
570 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
573 copy_skb_header(n
, skb
);
579 * pskb_copy - create copy of an sk_buff with private head.
580 * @skb: buffer to copy
581 * @gfp_mask: allocation priority
583 * Make a copy of both an &sk_buff and part of its data, located
584 * in header. Fragmented data remain shared. This is used when
585 * the caller wishes to modify only header of &sk_buff and needs
586 * private copy of the header to alter. Returns %NULL on failure
587 * or the pointer to the buffer on success.
588 * The returned buffer has a reference count of 1.
591 struct sk_buff
*pskb_copy(struct sk_buff
*skb
, gfp_t gfp_mask
)
594 * Allocate the copy buffer
596 struct sk_buff
*n
= alloc_skb(skb
->end
- skb
->head
, gfp_mask
);
601 /* Set the data pointer */
602 skb_reserve(n
, skb
->data
- skb
->head
);
603 /* Set the tail pointer and length */
604 skb_put(n
, skb_headlen(skb
));
606 memcpy(n
->data
, skb
->data
, n
->len
);
608 n
->ip_summed
= skb
->ip_summed
;
610 n
->data_len
= skb
->data_len
;
613 if (skb_shinfo(skb
)->nr_frags
) {
616 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
617 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
618 get_page(skb_shinfo(n
)->frags
[i
].page
);
620 skb_shinfo(n
)->nr_frags
= i
;
623 if (skb_shinfo(skb
)->frag_list
) {
624 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
625 skb_clone_fraglist(n
);
628 copy_skb_header(n
, skb
);
634 * pskb_expand_head - reallocate header of &sk_buff
635 * @skb: buffer to reallocate
636 * @nhead: room to add at head
637 * @ntail: room to add at tail
638 * @gfp_mask: allocation priority
640 * Expands (or creates identical copy, if &nhead and &ntail are zero)
641 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
642 * reference count of 1. Returns zero in the case of success or error,
643 * if expansion failed. In the last case, &sk_buff is not changed.
645 * All the pointers pointing into skb header may change and must be
646 * reloaded after call to this function.
649 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
654 int size
= nhead
+ (skb
->end
- skb
->head
) + ntail
;
660 size
= SKB_DATA_ALIGN(size
);
662 data
= kmalloc(size
+ sizeof(struct skb_shared_info
), gfp_mask
);
666 /* Copy only real data... and, alas, header. This should be
667 * optimized for the cases when header is void. */
668 memcpy(data
+ nhead
, skb
->head
, skb
->tail
- skb
->head
);
669 memcpy(data
+ size
, skb
->end
, sizeof(struct skb_shared_info
));
671 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
672 get_page(skb_shinfo(skb
)->frags
[i
].page
);
674 if (skb_shinfo(skb
)->frag_list
)
675 skb_clone_fraglist(skb
);
677 skb_release_data(skb
);
679 off
= (data
+ nhead
) - skb
->head
;
682 skb
->end
= data
+ size
;
690 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
697 /* Make private copy of skb with writable head and some headroom */
699 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
701 struct sk_buff
*skb2
;
702 int delta
= headroom
- skb_headroom(skb
);
705 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
707 skb2
= skb_clone(skb
, GFP_ATOMIC
);
708 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
719 * skb_copy_expand - copy and expand sk_buff
720 * @skb: buffer to copy
721 * @newheadroom: new free bytes at head
722 * @newtailroom: new free bytes at tail
723 * @gfp_mask: allocation priority
725 * Make a copy of both an &sk_buff and its data and while doing so
726 * allocate additional space.
728 * This is used when the caller wishes to modify the data and needs a
729 * private copy of the data to alter as well as more space for new fields.
730 * Returns %NULL on failure or the pointer to the buffer
731 * on success. The returned buffer has a reference count of 1.
733 * You must pass %GFP_ATOMIC as the allocation priority if this function
734 * is called from an interrupt.
736 * BUG ALERT: ip_summed is not copied. Why does this work? Is it used
737 * only by netfilter in the cases when checksum is recalculated? --ANK
739 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
740 int newheadroom
, int newtailroom
,
744 * Allocate the copy buffer
746 struct sk_buff
*n
= alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
748 int head_copy_len
, head_copy_off
;
753 skb_reserve(n
, newheadroom
);
755 /* Set the tail pointer and length */
756 skb_put(n
, skb
->len
);
758 head_copy_len
= skb_headroom(skb
);
760 if (newheadroom
<= head_copy_len
)
761 head_copy_len
= newheadroom
;
763 head_copy_off
= newheadroom
- head_copy_len
;
765 /* Copy the linear header and data. */
766 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
767 skb
->len
+ head_copy_len
))
770 copy_skb_header(n
, skb
);
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 NULL in out of memory cases.
787 struct sk_buff
*skb_pad(struct sk_buff
*skb
, int pad
)
789 struct sk_buff
*nskb
;
791 /* If the skbuff is non linear tailroom is always zero.. */
792 if (skb_tailroom(skb
) >= pad
) {
793 memset(skb
->data
+skb
->len
, 0, pad
);
797 nskb
= skb_copy_expand(skb
, skb_headroom(skb
), skb_tailroom(skb
) + pad
, GFP_ATOMIC
);
800 memset(nskb
->data
+nskb
->len
, 0, pad
);
804 /* Trims skb to length len. It can change skb pointers, if "realloc" is 1.
805 * If realloc==0 and trimming is impossible without change of data,
809 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
, int realloc
)
811 int offset
= skb_headlen(skb
);
812 int nfrags
= skb_shinfo(skb
)->nr_frags
;
815 for (i
= 0; i
< nfrags
; i
++) {
816 int end
= offset
+ skb_shinfo(skb
)->frags
[i
].size
;
818 if (skb_cloned(skb
)) {
820 if (pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))
824 put_page(skb_shinfo(skb
)->frags
[i
].page
);
825 skb_shinfo(skb
)->nr_frags
--;
827 skb_shinfo(skb
)->frags
[i
].size
= len
- offset
;
834 skb
->data_len
-= skb
->len
- len
;
837 if (len
<= skb_headlen(skb
)) {
840 skb
->tail
= skb
->data
+ len
;
841 if (skb_shinfo(skb
)->frag_list
&& !skb_cloned(skb
))
842 skb_drop_fraglist(skb
);
844 skb
->data_len
-= skb
->len
- len
;
853 * __pskb_pull_tail - advance tail of skb header
854 * @skb: buffer to reallocate
855 * @delta: number of bytes to advance tail
857 * The function makes a sense only on a fragmented &sk_buff,
858 * it expands header moving its tail forward and copying necessary
859 * data from fragmented part.
861 * &sk_buff MUST have reference count of 1.
863 * Returns %NULL (and &sk_buff does not change) if pull failed
864 * or value of new tail of skb in the case of success.
866 * All the pointers pointing into skb header may change and must be
867 * reloaded after call to this function.
870 /* Moves tail of skb head forward, copying data from fragmented part,
871 * when it is necessary.
872 * 1. It may fail due to malloc failure.
873 * 2. It may change skb pointers.
875 * It is pretty complicated. Luckily, it is called only in exceptional cases.
877 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
879 /* If skb has not enough free space at tail, get new one
880 * plus 128 bytes for future expansions. If we have enough
881 * room at tail, reallocate without expansion only if skb is cloned.
883 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
885 if (eat
> 0 || skb_cloned(skb
)) {
886 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
891 if (skb_copy_bits(skb
, skb_headlen(skb
), skb
->tail
, delta
))
894 /* Optimization: no fragments, no reasons to preestimate
895 * size of pulled pages. Superb.
897 if (!skb_shinfo(skb
)->frag_list
)
900 /* Estimate size of pulled pages. */
902 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
903 if (skb_shinfo(skb
)->frags
[i
].size
>= eat
)
905 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
908 /* If we need update frag list, we are in troubles.
909 * Certainly, it possible to add an offset to skb data,
910 * but taking into account that pulling is expected to
911 * be very rare operation, it is worth to fight against
912 * further bloating skb head and crucify ourselves here instead.
913 * Pure masohism, indeed. 8)8)
916 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
917 struct sk_buff
*clone
= NULL
;
918 struct sk_buff
*insp
= NULL
;
923 if (list
->len
<= eat
) {
924 /* Eaten as whole. */
929 /* Eaten partially. */
931 if (skb_shared(list
)) {
932 /* Sucks! We need to fork list. :-( */
933 clone
= skb_clone(list
, GFP_ATOMIC
);
939 /* This may be pulled without
943 if (!pskb_pull(list
, eat
)) {
952 /* Free pulled out fragments. */
953 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
954 skb_shinfo(skb
)->frag_list
= list
->next
;
957 /* And insert new clone at head. */
960 skb_shinfo(skb
)->frag_list
= clone
;
963 /* Success! Now we may commit changes to skb data. */
968 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
969 if (skb_shinfo(skb
)->frags
[i
].size
<= eat
) {
970 put_page(skb_shinfo(skb
)->frags
[i
].page
);
971 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
973 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
975 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
976 skb_shinfo(skb
)->frags
[k
].size
-= eat
;
982 skb_shinfo(skb
)->nr_frags
= k
;
985 skb
->data_len
-= delta
;
990 /* Copy some data bits from skb to kernel buffer. */
992 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
995 int start
= skb_headlen(skb
);
997 if (offset
> (int)skb
->len
- len
)
1001 if ((copy
= start
- offset
) > 0) {
1004 memcpy(to
, skb
->data
+ offset
, copy
);
1005 if ((len
-= copy
) == 0)
1011 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1014 BUG_TRAP(start
<= offset
+ len
);
1016 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1017 if ((copy
= end
- offset
) > 0) {
1023 vaddr
= kmap_skb_frag(&skb_shinfo(skb
)->frags
[i
]);
1025 vaddr
+ skb_shinfo(skb
)->frags
[i
].page_offset
+
1026 offset
- start
, copy
);
1027 kunmap_skb_frag(vaddr
);
1029 if ((len
-= copy
) == 0)
1037 if (skb_shinfo(skb
)->frag_list
) {
1038 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1040 for (; list
; list
= list
->next
) {
1043 BUG_TRAP(start
<= offset
+ len
);
1045 end
= start
+ list
->len
;
1046 if ((copy
= end
- offset
) > 0) {
1049 if (skb_copy_bits(list
, offset
- start
,
1052 if ((len
-= copy
) == 0)
1068 * skb_store_bits - store bits from kernel buffer to skb
1069 * @skb: destination buffer
1070 * @offset: offset in destination
1071 * @from: source buffer
1072 * @len: number of bytes to copy
1074 * Copy the specified number of bytes from the source buffer to the
1075 * destination skb. This function handles all the messy bits of
1076 * traversing fragment lists and such.
1079 int skb_store_bits(const struct sk_buff
*skb
, int offset
, void *from
, int len
)
1082 int start
= skb_headlen(skb
);
1084 if (offset
> (int)skb
->len
- len
)
1087 if ((copy
= start
- offset
) > 0) {
1090 memcpy(skb
->data
+ offset
, from
, copy
);
1091 if ((len
-= copy
) == 0)
1097 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1098 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1101 BUG_TRAP(start
<= offset
+ len
);
1103 end
= start
+ frag
->size
;
1104 if ((copy
= end
- offset
) > 0) {
1110 vaddr
= kmap_skb_frag(frag
);
1111 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1113 kunmap_skb_frag(vaddr
);
1115 if ((len
-= copy
) == 0)
1123 if (skb_shinfo(skb
)->frag_list
) {
1124 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1126 for (; list
; list
= list
->next
) {
1129 BUG_TRAP(start
<= offset
+ len
);
1131 end
= start
+ list
->len
;
1132 if ((copy
= end
- offset
) > 0) {
1135 if (skb_store_bits(list
, offset
- start
,
1138 if ((len
-= copy
) == 0)
1153 EXPORT_SYMBOL(skb_store_bits
);
1155 /* Checksum skb data. */
1157 unsigned int skb_checksum(const struct sk_buff
*skb
, int offset
,
1158 int len
, unsigned int csum
)
1160 int start
= skb_headlen(skb
);
1161 int i
, copy
= start
- offset
;
1164 /* Checksum header. */
1168 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1169 if ((len
-= copy
) == 0)
1175 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1178 BUG_TRAP(start
<= offset
+ len
);
1180 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1181 if ((copy
= end
- offset
) > 0) {
1184 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1188 vaddr
= kmap_skb_frag(frag
);
1189 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
1190 offset
- start
, copy
, 0);
1191 kunmap_skb_frag(vaddr
);
1192 csum
= csum_block_add(csum
, csum2
, pos
);
1201 if (skb_shinfo(skb
)->frag_list
) {
1202 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1204 for (; list
; list
= list
->next
) {
1207 BUG_TRAP(start
<= offset
+ len
);
1209 end
= start
+ list
->len
;
1210 if ((copy
= end
- offset
) > 0) {
1214 csum2
= skb_checksum(list
, offset
- start
,
1216 csum
= csum_block_add(csum
, csum2
, pos
);
1217 if ((len
-= copy
) == 0)
1230 /* Both of above in one bottle. */
1232 unsigned int skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
1233 u8
*to
, int len
, unsigned int csum
)
1235 int start
= skb_headlen(skb
);
1236 int i
, copy
= start
- offset
;
1243 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
1245 if ((len
-= copy
) == 0)
1252 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1255 BUG_TRAP(start
<= offset
+ len
);
1257 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1258 if ((copy
= end
- offset
) > 0) {
1261 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1265 vaddr
= kmap_skb_frag(frag
);
1266 csum2
= csum_partial_copy_nocheck(vaddr
+
1270 kunmap_skb_frag(vaddr
);
1271 csum
= csum_block_add(csum
, csum2
, pos
);
1281 if (skb_shinfo(skb
)->frag_list
) {
1282 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1284 for (; list
; list
= list
->next
) {
1288 BUG_TRAP(start
<= offset
+ len
);
1290 end
= start
+ list
->len
;
1291 if ((copy
= end
- offset
) > 0) {
1294 csum2
= skb_copy_and_csum_bits(list
,
1297 csum
= csum_block_add(csum
, csum2
, pos
);
1298 if ((len
-= copy
) == 0)
1311 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
1316 if (skb
->ip_summed
== CHECKSUM_HW
)
1317 csstart
= skb
->h
.raw
- skb
->data
;
1319 csstart
= skb_headlen(skb
);
1321 BUG_ON(csstart
> skb_headlen(skb
));
1323 memcpy(to
, skb
->data
, csstart
);
1326 if (csstart
!= skb
->len
)
1327 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
1328 skb
->len
- csstart
, 0);
1330 if (skb
->ip_summed
== CHECKSUM_HW
) {
1331 long csstuff
= csstart
+ skb
->csum
;
1333 *((unsigned short *)(to
+ csstuff
)) = csum_fold(csum
);
1338 * skb_dequeue - remove from the head of the queue
1339 * @list: list to dequeue from
1341 * Remove the head of the list. The list lock is taken so the function
1342 * may be used safely with other locking list functions. The head item is
1343 * returned or %NULL if the list is empty.
1346 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
1348 unsigned long flags
;
1349 struct sk_buff
*result
;
1351 spin_lock_irqsave(&list
->lock
, flags
);
1352 result
= __skb_dequeue(list
);
1353 spin_unlock_irqrestore(&list
->lock
, flags
);
1358 * skb_dequeue_tail - remove from the tail of the queue
1359 * @list: list to dequeue from
1361 * Remove the tail of the list. The list lock is taken so the function
1362 * may be used safely with other locking list functions. The tail item is
1363 * returned or %NULL if the list is empty.
1365 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
1367 unsigned long flags
;
1368 struct sk_buff
*result
;
1370 spin_lock_irqsave(&list
->lock
, flags
);
1371 result
= __skb_dequeue_tail(list
);
1372 spin_unlock_irqrestore(&list
->lock
, flags
);
1377 * skb_queue_purge - empty a list
1378 * @list: list to empty
1380 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1381 * the list and one reference dropped. This function takes the list
1382 * lock and is atomic with respect to other list locking functions.
1384 void skb_queue_purge(struct sk_buff_head
*list
)
1386 struct sk_buff
*skb
;
1387 while ((skb
= skb_dequeue(list
)) != NULL
)
1392 * skb_queue_head - queue a buffer at the list head
1393 * @list: list to use
1394 * @newsk: buffer to queue
1396 * Queue a buffer at the start of the list. This function takes the
1397 * list lock and can be used safely with other locking &sk_buff functions
1400 * A buffer cannot be placed on two lists at the same time.
1402 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1404 unsigned long flags
;
1406 spin_lock_irqsave(&list
->lock
, flags
);
1407 __skb_queue_head(list
, newsk
);
1408 spin_unlock_irqrestore(&list
->lock
, flags
);
1412 * skb_queue_tail - queue a buffer at the list tail
1413 * @list: list to use
1414 * @newsk: buffer to queue
1416 * Queue a buffer at the tail of the list. This function takes the
1417 * list lock and can be used safely with other locking &sk_buff functions
1420 * A buffer cannot be placed on two lists at the same time.
1422 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1424 unsigned long flags
;
1426 spin_lock_irqsave(&list
->lock
, flags
);
1427 __skb_queue_tail(list
, newsk
);
1428 spin_unlock_irqrestore(&list
->lock
, flags
);
1432 * skb_unlink - remove a buffer from a list
1433 * @skb: buffer to remove
1434 * @list: list to use
1436 * Remove a packet from a list. The list locks are taken and this
1437 * function is atomic with respect to other list locked calls
1439 * You must know what list the SKB is on.
1441 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
1443 unsigned long flags
;
1445 spin_lock_irqsave(&list
->lock
, flags
);
1446 __skb_unlink(skb
, list
);
1447 spin_unlock_irqrestore(&list
->lock
, flags
);
1451 * skb_append - append a buffer
1452 * @old: buffer to insert after
1453 * @newsk: buffer to insert
1454 * @list: list to use
1456 * Place a packet after a given packet in a list. The list locks are taken
1457 * and this function is atomic with respect to other list locked calls.
1458 * A buffer cannot be placed on two lists at the same time.
1460 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
1462 unsigned long flags
;
1464 spin_lock_irqsave(&list
->lock
, flags
);
1465 __skb_append(old
, newsk
, list
);
1466 spin_unlock_irqrestore(&list
->lock
, flags
);
1471 * skb_insert - insert a buffer
1472 * @old: buffer to insert before
1473 * @newsk: buffer to insert
1474 * @list: list to use
1476 * Place a packet before a given packet in a list. The list locks are
1477 * taken and this function is atomic with respect to other list locked
1480 * A buffer cannot be placed on two lists at the same time.
1482 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
1484 unsigned long flags
;
1486 spin_lock_irqsave(&list
->lock
, flags
);
1487 __skb_insert(newsk
, old
->prev
, old
, list
);
1488 spin_unlock_irqrestore(&list
->lock
, flags
);
1493 * Tune the memory allocator for a new MTU size.
1495 void skb_add_mtu(int mtu
)
1497 /* Must match allocation in alloc_skb */
1498 mtu
= SKB_DATA_ALIGN(mtu
) + sizeof(struct skb_shared_info
);
1500 kmem_add_cache_size(mtu
);
1504 static inline void skb_split_inside_header(struct sk_buff
*skb
,
1505 struct sk_buff
* skb1
,
1506 const u32 len
, const int pos
)
1510 memcpy(skb_put(skb1
, pos
- len
), skb
->data
+ len
, pos
- len
);
1512 /* And move data appendix as is. */
1513 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1514 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
1516 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
1517 skb_shinfo(skb
)->nr_frags
= 0;
1518 skb1
->data_len
= skb
->data_len
;
1519 skb1
->len
+= skb1
->data_len
;
1522 skb
->tail
= skb
->data
+ len
;
1525 static inline void skb_split_no_header(struct sk_buff
*skb
,
1526 struct sk_buff
* skb1
,
1527 const u32 len
, int pos
)
1530 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
1532 skb_shinfo(skb
)->nr_frags
= 0;
1533 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
1535 skb
->data_len
= len
- pos
;
1537 for (i
= 0; i
< nfrags
; i
++) {
1538 int size
= skb_shinfo(skb
)->frags
[i
].size
;
1540 if (pos
+ size
> len
) {
1541 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1545 * We have two variants in this case:
1546 * 1. Move all the frag to the second
1547 * part, if it is possible. F.e.
1548 * this approach is mandatory for TUX,
1549 * where splitting is expensive.
1550 * 2. Split is accurately. We make this.
1552 get_page(skb_shinfo(skb
)->frags
[i
].page
);
1553 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
1554 skb_shinfo(skb1
)->frags
[0].size
-= len
- pos
;
1555 skb_shinfo(skb
)->frags
[i
].size
= len
- pos
;
1556 skb_shinfo(skb
)->nr_frags
++;
1560 skb_shinfo(skb
)->nr_frags
++;
1563 skb_shinfo(skb1
)->nr_frags
= k
;
1567 * skb_split - Split fragmented skb to two parts at length len.
1568 * @skb: the buffer to split
1569 * @skb1: the buffer to receive the second part
1570 * @len: new length for skb
1572 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
1574 int pos
= skb_headlen(skb
);
1576 if (len
< pos
) /* Split line is inside header. */
1577 skb_split_inside_header(skb
, skb1
, len
, pos
);
1578 else /* Second chunk has no header, nothing to copy. */
1579 skb_split_no_header(skb
, skb1
, len
, pos
);
1583 * skb_prepare_seq_read - Prepare a sequential read of skb data
1584 * @skb: the buffer to read
1585 * @from: lower offset of data to be read
1586 * @to: upper offset of data to be read
1587 * @st: state variable
1589 * Initializes the specified state variable. Must be called before
1590 * invoking skb_seq_read() for the first time.
1592 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
1593 unsigned int to
, struct skb_seq_state
*st
)
1595 st
->lower_offset
= from
;
1596 st
->upper_offset
= to
;
1597 st
->root_skb
= st
->cur_skb
= skb
;
1598 st
->frag_idx
= st
->stepped_offset
= 0;
1599 st
->frag_data
= NULL
;
1603 * skb_seq_read - Sequentially read skb data
1604 * @consumed: number of bytes consumed by the caller so far
1605 * @data: destination pointer for data to be returned
1606 * @st: state variable
1608 * Reads a block of skb data at &consumed relative to the
1609 * lower offset specified to skb_prepare_seq_read(). Assigns
1610 * the head of the data block to &data and returns the length
1611 * of the block or 0 if the end of the skb data or the upper
1612 * offset has been reached.
1614 * The caller is not required to consume all of the data
1615 * returned, i.e. &consumed is typically set to the number
1616 * of bytes already consumed and the next call to
1617 * skb_seq_read() will return the remaining part of the block.
1619 * Note: The size of each block of data returned can be arbitary,
1620 * this limitation is the cost for zerocopy seqeuental
1621 * reads of potentially non linear data.
1623 * Note: Fragment lists within fragments are not implemented
1624 * at the moment, state->root_skb could be replaced with
1625 * a stack for this purpose.
1627 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
1628 struct skb_seq_state
*st
)
1630 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
1633 if (unlikely(abs_offset
>= st
->upper_offset
))
1637 block_limit
= skb_headlen(st
->cur_skb
);
1639 if (abs_offset
< block_limit
) {
1640 *data
= st
->cur_skb
->data
+ abs_offset
;
1641 return block_limit
- abs_offset
;
1644 if (st
->frag_idx
== 0 && !st
->frag_data
)
1645 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
1647 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
1648 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
1649 block_limit
= frag
->size
+ st
->stepped_offset
;
1651 if (abs_offset
< block_limit
) {
1653 st
->frag_data
= kmap_skb_frag(frag
);
1655 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
1656 (abs_offset
- st
->stepped_offset
);
1658 return block_limit
- abs_offset
;
1661 if (st
->frag_data
) {
1662 kunmap_skb_frag(st
->frag_data
);
1663 st
->frag_data
= NULL
;
1667 st
->stepped_offset
+= frag
->size
;
1670 if (st
->cur_skb
->next
) {
1671 st
->cur_skb
= st
->cur_skb
->next
;
1674 } else if (st
->root_skb
== st
->cur_skb
&&
1675 skb_shinfo(st
->root_skb
)->frag_list
) {
1676 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
1684 * skb_abort_seq_read - Abort a sequential read of skb data
1685 * @st: state variable
1687 * Must be called if skb_seq_read() was not called until it
1690 void skb_abort_seq_read(struct skb_seq_state
*st
)
1693 kunmap_skb_frag(st
->frag_data
);
1696 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
1698 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
1699 struct ts_config
*conf
,
1700 struct ts_state
*state
)
1702 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
1705 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
1707 skb_abort_seq_read(TS_SKB_CB(state
));
1711 * skb_find_text - Find a text pattern in skb data
1712 * @skb: the buffer to look in
1713 * @from: search offset
1715 * @config: textsearch configuration
1716 * @state: uninitialized textsearch state variable
1718 * Finds a pattern in the skb data according to the specified
1719 * textsearch configuration. Use textsearch_next() to retrieve
1720 * subsequent occurrences of the pattern. Returns the offset
1721 * to the first occurrence or UINT_MAX if no match was found.
1723 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
1724 unsigned int to
, struct ts_config
*config
,
1725 struct ts_state
*state
)
1727 config
->get_next_block
= skb_ts_get_next_block
;
1728 config
->finish
= skb_ts_finish
;
1730 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
1732 return textsearch_find(config
, state
);
1736 * skb_append_datato_frags: - append the user data to a skb
1737 * @sk: sock structure
1738 * @skb: skb structure to be appened with user data.
1739 * @getfrag: call back function to be used for getting the user data
1740 * @from: pointer to user message iov
1741 * @length: length of the iov message
1743 * Description: This procedure append the user data in the fragment part
1744 * of the skb if any page alloc fails user this procedure returns -ENOMEM
1746 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
1747 int (*getfrag
)(void *from
, char *to
, int offset
,
1748 int len
, int odd
, struct sk_buff
*skb
),
1749 void *from
, int length
)
1752 skb_frag_t
*frag
= NULL
;
1753 struct page
*page
= NULL
;
1759 /* Return error if we don't have space for new frag */
1760 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
1761 if (frg_cnt
>= MAX_SKB_FRAGS
)
1764 /* allocate a new page for next frag */
1765 page
= alloc_pages(sk
->sk_allocation
, 0);
1767 /* If alloc_page fails just return failure and caller will
1768 * free previous allocated pages by doing kfree_skb()
1773 /* initialize the next frag */
1774 sk
->sk_sndmsg_page
= page
;
1775 sk
->sk_sndmsg_off
= 0;
1776 skb_fill_page_desc(skb
, frg_cnt
, page
, 0, 0);
1777 skb
->truesize
+= PAGE_SIZE
;
1778 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
1780 /* get the new initialized frag */
1781 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
1782 frag
= &skb_shinfo(skb
)->frags
[frg_cnt
- 1];
1784 /* copy the user data to page */
1785 left
= PAGE_SIZE
- frag
->page_offset
;
1786 copy
= (length
> left
)? left
: length
;
1788 ret
= getfrag(from
, (page_address(frag
->page
) +
1789 frag
->page_offset
+ frag
->size
),
1790 offset
, copy
, 0, skb
);
1794 /* copy was successful so update the size parameters */
1795 sk
->sk_sndmsg_off
+= copy
;
1798 skb
->data_len
+= copy
;
1802 } while (length
> 0);
1808 * skb_pull_rcsum - pull skb and update receive checksum
1809 * @skb: buffer to update
1810 * @start: start of data before pull
1811 * @len: length of data pulled
1813 * This function performs an skb_pull on the packet and updates
1814 * update the CHECKSUM_HW checksum. It should be used on receive
1815 * path processing instead of skb_pull unless you know that the
1816 * checksum difference is zero (e.g., a valid IP header) or you
1817 * are setting ip_summed to CHECKSUM_NONE.
1819 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
1821 BUG_ON(len
> skb
->len
);
1823 BUG_ON(skb
->len
< skb
->data_len
);
1824 skb_postpull_rcsum(skb
, skb
->data
, len
);
1825 return skb
->data
+= len
;
1828 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
1830 void __init
skb_init(void)
1832 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
1833 sizeof(struct sk_buff
),
1837 if (!skbuff_head_cache
)
1838 panic("cannot create skbuff cache");
1840 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
1841 (2*sizeof(struct sk_buff
)) +
1846 if (!skbuff_fclone_cache
)
1847 panic("cannot create skbuff cache");
1850 EXPORT_SYMBOL(___pskb_trim
);
1851 EXPORT_SYMBOL(__kfree_skb
);
1852 EXPORT_SYMBOL(kfree_skb
);
1853 EXPORT_SYMBOL(__pskb_pull_tail
);
1854 EXPORT_SYMBOL(__alloc_skb
);
1855 EXPORT_SYMBOL(pskb_copy
);
1856 EXPORT_SYMBOL(pskb_expand_head
);
1857 EXPORT_SYMBOL(skb_checksum
);
1858 EXPORT_SYMBOL(skb_clone
);
1859 EXPORT_SYMBOL(skb_clone_fraglist
);
1860 EXPORT_SYMBOL(skb_copy
);
1861 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
1862 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
1863 EXPORT_SYMBOL(skb_copy_bits
);
1864 EXPORT_SYMBOL(skb_copy_expand
);
1865 EXPORT_SYMBOL(skb_over_panic
);
1866 EXPORT_SYMBOL(skb_pad
);
1867 EXPORT_SYMBOL(skb_realloc_headroom
);
1868 EXPORT_SYMBOL(skb_under_panic
);
1869 EXPORT_SYMBOL(skb_dequeue
);
1870 EXPORT_SYMBOL(skb_dequeue_tail
);
1871 EXPORT_SYMBOL(skb_insert
);
1872 EXPORT_SYMBOL(skb_queue_purge
);
1873 EXPORT_SYMBOL(skb_queue_head
);
1874 EXPORT_SYMBOL(skb_queue_tail
);
1875 EXPORT_SYMBOL(skb_unlink
);
1876 EXPORT_SYMBOL(skb_append
);
1877 EXPORT_SYMBOL(skb_split
);
1878 EXPORT_SYMBOL(skb_prepare_seq_read
);
1879 EXPORT_SYMBOL(skb_seq_read
);
1880 EXPORT_SYMBOL(skb_abort_seq_read
);
1881 EXPORT_SYMBOL(skb_find_text
);
1882 EXPORT_SYMBOL(skb_append_datato_frags
);