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 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
116 * 'private' fields and also do memory statistics to find all the
122 * __alloc_skb - allocate a network buffer
123 * @size: size to allocate
124 * @gfp_mask: allocation mask
125 * @fclone: allocate from fclone cache instead of head cache
126 * and allocate a cloned (child) skb
128 * Allocate a new &sk_buff. The returned buffer has no headroom and a
129 * tail room of size bytes. The object has a reference count of one.
130 * The return is the buffer. On a failure the return is %NULL.
132 * Buffers may only be allocated from interrupts using a @gfp_mask of
135 struct sk_buff
*__alloc_skb(unsigned int size
, gfp_t gfp_mask
,
139 struct skb_shared_info
*shinfo
;
143 cache
= fclone
? skbuff_fclone_cache
: skbuff_head_cache
;
146 skb
= kmem_cache_alloc(cache
, gfp_mask
& ~__GFP_DMA
);
150 /* Get the DATA. Size must match skb_add_mtu(). */
151 size
= SKB_DATA_ALIGN(size
);
152 data
= kmalloc(size
+ sizeof(struct skb_shared_info
), gfp_mask
);
156 memset(skb
, 0, offsetof(struct sk_buff
, truesize
));
157 skb
->truesize
= size
+ sizeof(struct sk_buff
);
158 atomic_set(&skb
->users
, 1);
162 skb
->end
= data
+ size
;
163 /* make sure we initialize shinfo sequentially */
164 shinfo
= skb_shinfo(skb
);
165 atomic_set(&shinfo
->dataref
, 1);
166 shinfo
->nr_frags
= 0;
167 shinfo
->tso_size
= 0;
168 shinfo
->tso_segs
= 0;
169 shinfo
->ufo_size
= 0;
170 shinfo
->ip6_frag_id
= 0;
171 shinfo
->frag_list
= NULL
;
174 struct sk_buff
*child
= skb
+ 1;
175 atomic_t
*fclone_ref
= (atomic_t
*) (child
+ 1);
177 skb
->fclone
= SKB_FCLONE_ORIG
;
178 atomic_set(fclone_ref
, 1);
180 child
->fclone
= SKB_FCLONE_UNAVAILABLE
;
185 kmem_cache_free(cache
, skb
);
191 * alloc_skb_from_cache - allocate a network buffer
192 * @cp: kmem_cache from which to allocate the data area
193 * (object size must be big enough for @size bytes + skb overheads)
194 * @size: size to allocate
195 * @gfp_mask: allocation mask
197 * Allocate a new &sk_buff. The returned buffer has no headroom and
198 * tail room of size bytes. The object has a reference count of one.
199 * The return is the buffer. On a failure the return is %NULL.
201 * Buffers may only be allocated from interrupts using a @gfp_mask of
204 struct sk_buff
*alloc_skb_from_cache(kmem_cache_t
*cp
,
212 skb
= kmem_cache_alloc(skbuff_head_cache
,
213 gfp_mask
& ~__GFP_DMA
);
218 size
= SKB_DATA_ALIGN(size
);
219 data
= kmem_cache_alloc(cp
, gfp_mask
);
223 memset(skb
, 0, offsetof(struct sk_buff
, truesize
));
224 skb
->truesize
= size
+ sizeof(struct sk_buff
);
225 atomic_set(&skb
->users
, 1);
229 skb
->end
= data
+ size
;
231 atomic_set(&(skb_shinfo(skb
)->dataref
), 1);
232 skb_shinfo(skb
)->nr_frags
= 0;
233 skb_shinfo(skb
)->tso_size
= 0;
234 skb_shinfo(skb
)->tso_segs
= 0;
235 skb_shinfo(skb
)->ufo_size
= 0;
236 skb_shinfo(skb
)->frag_list
= NULL
;
240 kmem_cache_free(skbuff_head_cache
, skb
);
246 static void skb_drop_list(struct sk_buff
**listp
)
248 struct sk_buff
*list
= *listp
;
253 struct sk_buff
*this = list
;
259 static inline void skb_drop_fraglist(struct sk_buff
*skb
)
261 skb_drop_list(&skb_shinfo(skb
)->frag_list
);
264 static void skb_clone_fraglist(struct sk_buff
*skb
)
266 struct sk_buff
*list
;
268 for (list
= skb_shinfo(skb
)->frag_list
; list
; list
= list
->next
)
272 void skb_release_data(struct sk_buff
*skb
)
275 !atomic_sub_return(skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1,
276 &skb_shinfo(skb
)->dataref
)) {
277 if (skb_shinfo(skb
)->nr_frags
) {
279 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
280 put_page(skb_shinfo(skb
)->frags
[i
].page
);
283 if (skb_shinfo(skb
)->frag_list
)
284 skb_drop_fraglist(skb
);
291 * Free an skbuff by memory without cleaning the state.
293 void kfree_skbmem(struct sk_buff
*skb
)
295 struct sk_buff
*other
;
296 atomic_t
*fclone_ref
;
298 skb_release_data(skb
);
299 switch (skb
->fclone
) {
300 case SKB_FCLONE_UNAVAILABLE
:
301 kmem_cache_free(skbuff_head_cache
, skb
);
304 case SKB_FCLONE_ORIG
:
305 fclone_ref
= (atomic_t
*) (skb
+ 2);
306 if (atomic_dec_and_test(fclone_ref
))
307 kmem_cache_free(skbuff_fclone_cache
, skb
);
310 case SKB_FCLONE_CLONE
:
311 fclone_ref
= (atomic_t
*) (skb
+ 1);
314 /* The clone portion is available for
315 * fast-cloning again.
317 skb
->fclone
= SKB_FCLONE_UNAVAILABLE
;
319 if (atomic_dec_and_test(fclone_ref
))
320 kmem_cache_free(skbuff_fclone_cache
, other
);
326 * __kfree_skb - private function
329 * Free an sk_buff. Release anything attached to the buffer.
330 * Clean the state. This is an internal helper function. Users should
331 * always call kfree_skb
334 void __kfree_skb(struct sk_buff
*skb
)
336 dst_release(skb
->dst
);
338 secpath_put(skb
->sp
);
340 if (skb
->destructor
) {
342 skb
->destructor(skb
);
344 #ifdef CONFIG_NETFILTER
345 nf_conntrack_put(skb
->nfct
);
346 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
347 nf_conntrack_put_reasm(skb
->nfct_reasm
);
349 #ifdef CONFIG_BRIDGE_NETFILTER
350 nf_bridge_put(skb
->nf_bridge
);
353 /* XXX: IS this still necessary? - JHS */
354 #ifdef CONFIG_NET_SCHED
356 #ifdef CONFIG_NET_CLS_ACT
365 * skb_clone - duplicate an sk_buff
366 * @skb: buffer to clone
367 * @gfp_mask: allocation priority
369 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
370 * copies share the same packet data but not structure. The new
371 * buffer has a reference count of 1. If the allocation fails the
372 * function returns %NULL otherwise the new buffer is returned.
374 * If this function is called from an interrupt gfp_mask() must be
378 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
383 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
384 n
->fclone
== SKB_FCLONE_UNAVAILABLE
) {
385 atomic_t
*fclone_ref
= (atomic_t
*) (n
+ 1);
386 n
->fclone
= SKB_FCLONE_CLONE
;
387 atomic_inc(fclone_ref
);
389 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
392 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
395 #define C(x) n->x = skb->x
397 n
->next
= n
->prev
= NULL
;
408 secpath_get(skb
->sp
);
410 memcpy(n
->cb
, skb
->cb
, sizeof(skb
->cb
));
421 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
425 n
->destructor
= NULL
;
426 #ifdef CONFIG_NETFILTER
429 nf_conntrack_get(skb
->nfct
);
431 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
433 nf_conntrack_get_reasm(skb
->nfct_reasm
);
435 #ifdef CONFIG_BRIDGE_NETFILTER
437 nf_bridge_get(skb
->nf_bridge
);
439 #endif /*CONFIG_NETFILTER*/
440 #ifdef CONFIG_NET_SCHED
442 #ifdef CONFIG_NET_CLS_ACT
443 n
->tc_verd
= SET_TC_VERD(skb
->tc_verd
,0);
444 n
->tc_verd
= CLR_TC_OK2MUNGE(n
->tc_verd
);
445 n
->tc_verd
= CLR_TC_MUNGED(n
->tc_verd
);
451 atomic_set(&n
->users
, 1);
457 atomic_inc(&(skb_shinfo(skb
)->dataref
));
463 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
466 * Shift between the two data areas in bytes
468 unsigned long offset
= new->data
- old
->data
;
472 new->priority
= old
->priority
;
473 new->protocol
= old
->protocol
;
474 new->dst
= dst_clone(old
->dst
);
476 new->sp
= secpath_get(old
->sp
);
478 new->h
.raw
= old
->h
.raw
+ offset
;
479 new->nh
.raw
= old
->nh
.raw
+ offset
;
480 new->mac
.raw
= old
->mac
.raw
+ offset
;
481 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
482 new->local_df
= old
->local_df
;
483 new->fclone
= SKB_FCLONE_UNAVAILABLE
;
484 new->pkt_type
= old
->pkt_type
;
485 new->tstamp
= old
->tstamp
;
486 new->destructor
= NULL
;
487 #ifdef CONFIG_NETFILTER
488 new->nfmark
= old
->nfmark
;
489 new->nfct
= old
->nfct
;
490 nf_conntrack_get(old
->nfct
);
491 new->nfctinfo
= old
->nfctinfo
;
492 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
493 new->nfct_reasm
= old
->nfct_reasm
;
494 nf_conntrack_get_reasm(old
->nfct_reasm
);
496 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
497 new->ipvs_property
= old
->ipvs_property
;
499 #ifdef CONFIG_BRIDGE_NETFILTER
500 new->nf_bridge
= old
->nf_bridge
;
501 nf_bridge_get(old
->nf_bridge
);
504 #ifdef CONFIG_NET_SCHED
505 #ifdef CONFIG_NET_CLS_ACT
506 new->tc_verd
= old
->tc_verd
;
508 new->tc_index
= old
->tc_index
;
510 atomic_set(&new->users
, 1);
511 skb_shinfo(new)->tso_size
= skb_shinfo(old
)->tso_size
;
512 skb_shinfo(new)->tso_segs
= skb_shinfo(old
)->tso_segs
;
513 skb_shinfo(new)->ufo_size
= skb_shinfo(old
)->ufo_size
;
517 * skb_copy - create private copy of an sk_buff
518 * @skb: buffer to copy
519 * @gfp_mask: allocation priority
521 * Make a copy of both an &sk_buff and its data. This is used when the
522 * caller wishes to modify the data and needs a private copy of the
523 * data to alter. Returns %NULL on failure or the pointer to the buffer
524 * on success. The returned buffer has a reference count of 1.
526 * As by-product this function converts non-linear &sk_buff to linear
527 * one, so that &sk_buff becomes completely private and caller is allowed
528 * to modify all the data of returned buffer. This means that this
529 * function is not recommended for use in circumstances when only
530 * header is going to be modified. Use pskb_copy() instead.
533 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
535 int headerlen
= skb
->data
- skb
->head
;
537 * Allocate the copy buffer
539 struct sk_buff
*n
= alloc_skb(skb
->end
- skb
->head
+ skb
->data_len
,
544 /* Set the data pointer */
545 skb_reserve(n
, headerlen
);
546 /* Set the tail pointer and length */
547 skb_put(n
, skb
->len
);
549 n
->ip_summed
= skb
->ip_summed
;
551 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
554 copy_skb_header(n
, skb
);
560 * pskb_copy - create copy of an sk_buff with private head.
561 * @skb: buffer to copy
562 * @gfp_mask: allocation priority
564 * Make a copy of both an &sk_buff and part of its data, located
565 * in header. Fragmented data remain shared. This is used when
566 * the caller wishes to modify only header of &sk_buff and needs
567 * private copy of the header to alter. Returns %NULL on failure
568 * or the pointer to the buffer on success.
569 * The returned buffer has a reference count of 1.
572 struct sk_buff
*pskb_copy(struct sk_buff
*skb
, gfp_t gfp_mask
)
575 * Allocate the copy buffer
577 struct sk_buff
*n
= alloc_skb(skb
->end
- skb
->head
, gfp_mask
);
582 /* Set the data pointer */
583 skb_reserve(n
, skb
->data
- skb
->head
);
584 /* Set the tail pointer and length */
585 skb_put(n
, skb_headlen(skb
));
587 memcpy(n
->data
, skb
->data
, n
->len
);
589 n
->ip_summed
= skb
->ip_summed
;
591 n
->truesize
+= skb
->data_len
;
592 n
->data_len
= skb
->data_len
;
595 if (skb_shinfo(skb
)->nr_frags
) {
598 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
599 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
600 get_page(skb_shinfo(n
)->frags
[i
].page
);
602 skb_shinfo(n
)->nr_frags
= i
;
605 if (skb_shinfo(skb
)->frag_list
) {
606 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
607 skb_clone_fraglist(n
);
610 copy_skb_header(n
, skb
);
616 * pskb_expand_head - reallocate header of &sk_buff
617 * @skb: buffer to reallocate
618 * @nhead: room to add at head
619 * @ntail: room to add at tail
620 * @gfp_mask: allocation priority
622 * Expands (or creates identical copy, if &nhead and &ntail are zero)
623 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
624 * reference count of 1. Returns zero in the case of success or error,
625 * if expansion failed. In the last case, &sk_buff is not changed.
627 * All the pointers pointing into skb header may change and must be
628 * reloaded after call to this function.
631 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
636 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 memcpy(data
+ nhead
, skb
->head
, skb
->tail
- skb
->head
);
651 memcpy(data
+ size
, skb
->end
, sizeof(struct skb_shared_info
));
653 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
654 get_page(skb_shinfo(skb
)->frags
[i
].page
);
656 if (skb_shinfo(skb
)->frag_list
)
657 skb_clone_fraglist(skb
);
659 skb_release_data(skb
);
661 off
= (data
+ nhead
) - skb
->head
;
664 skb
->end
= data
+ size
;
672 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
679 /* Make private copy of skb with writable head and some headroom */
681 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
683 struct sk_buff
*skb2
;
684 int delta
= headroom
- skb_headroom(skb
);
687 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
689 skb2
= skb_clone(skb
, GFP_ATOMIC
);
690 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
701 * skb_copy_expand - copy and expand sk_buff
702 * @skb: buffer to copy
703 * @newheadroom: new free bytes at head
704 * @newtailroom: new free bytes at tail
705 * @gfp_mask: allocation priority
707 * Make a copy of both an &sk_buff and its data and while doing so
708 * allocate additional space.
710 * This is used when the caller wishes to modify the data and needs a
711 * private copy of the data to alter as well as more space for new fields.
712 * Returns %NULL on failure or the pointer to the buffer
713 * on success. The returned buffer has a reference count of 1.
715 * You must pass %GFP_ATOMIC as the allocation priority if this function
716 * is called from an interrupt.
718 * BUG ALERT: ip_summed is not copied. Why does this work? Is it used
719 * only by netfilter in the cases when checksum is recalculated? --ANK
721 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
722 int newheadroom
, int newtailroom
,
726 * Allocate the copy buffer
728 struct sk_buff
*n
= alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
730 int head_copy_len
, head_copy_off
;
735 skb_reserve(n
, newheadroom
);
737 /* Set the tail pointer and length */
738 skb_put(n
, skb
->len
);
740 head_copy_len
= skb_headroom(skb
);
742 if (newheadroom
<= head_copy_len
)
743 head_copy_len
= newheadroom
;
745 head_copy_off
= newheadroom
- head_copy_len
;
747 /* Copy the linear header and data. */
748 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
749 skb
->len
+ head_copy_len
))
752 copy_skb_header(n
, skb
);
758 * skb_pad - zero pad the tail of an skb
759 * @skb: buffer to pad
762 * Ensure that a buffer is followed by a padding area that is zero
763 * filled. Used by network drivers which may DMA or transfer data
764 * beyond the buffer end onto the wire.
766 * May return NULL in out of memory cases.
769 struct sk_buff
*skb_pad(struct sk_buff
*skb
, int pad
)
771 struct sk_buff
*nskb
;
773 /* If the skbuff is non linear tailroom is always zero.. */
774 if (skb_tailroom(skb
) >= pad
) {
775 memset(skb
->data
+skb
->len
, 0, pad
);
779 nskb
= skb_copy_expand(skb
, skb_headroom(skb
), skb_tailroom(skb
) + pad
, GFP_ATOMIC
);
782 memset(nskb
->data
+nskb
->len
, 0, pad
);
786 /* Trims skb to length len. It can change skb pointers.
789 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
791 struct sk_buff
**fragp
;
792 struct sk_buff
*frag
;
793 int offset
= skb_headlen(skb
);
794 int nfrags
= skb_shinfo(skb
)->nr_frags
;
798 if (skb_cloned(skb
) &&
799 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
806 for (; i
< nfrags
; i
++) {
807 int end
= offset
+ skb_shinfo(skb
)->frags
[i
].size
;
814 skb_shinfo(skb
)->frags
[i
++].size
= len
- offset
;
817 skb_shinfo(skb
)->nr_frags
= i
;
819 for (; i
< nfrags
; i
++)
820 put_page(skb_shinfo(skb
)->frags
[i
].page
);
822 if (skb_shinfo(skb
)->frag_list
)
823 skb_drop_fraglist(skb
);
827 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
828 fragp
= &frag
->next
) {
829 int end
= offset
+ frag
->len
;
831 if (skb_shared(frag
)) {
832 struct sk_buff
*nfrag
;
834 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
835 if (unlikely(!nfrag
))
838 nfrag
->next
= frag
->next
;
850 unlikely((err
= pskb_trim(frag
, len
- offset
))))
854 skb_drop_list(&frag
->next
);
859 if (len
> skb_headlen(skb
)) {
860 skb
->data_len
-= skb
->len
- len
;
865 skb
->tail
= skb
->data
+ len
;
872 * __pskb_pull_tail - advance tail of skb header
873 * @skb: buffer to reallocate
874 * @delta: number of bytes to advance tail
876 * The function makes a sense only on a fragmented &sk_buff,
877 * it expands header moving its tail forward and copying necessary
878 * data from fragmented part.
880 * &sk_buff MUST have reference count of 1.
882 * Returns %NULL (and &sk_buff does not change) if pull failed
883 * or value of new tail of skb in the case of success.
885 * All the pointers pointing into skb header may change and must be
886 * reloaded after call to this function.
889 /* Moves tail of skb head forward, copying data from fragmented part,
890 * when it is necessary.
891 * 1. It may fail due to malloc failure.
892 * 2. It may change skb pointers.
894 * It is pretty complicated. Luckily, it is called only in exceptional cases.
896 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
898 /* If skb has not enough free space at tail, get new one
899 * plus 128 bytes for future expansions. If we have enough
900 * room at tail, reallocate without expansion only if skb is cloned.
902 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
904 if (eat
> 0 || skb_cloned(skb
)) {
905 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
910 if (skb_copy_bits(skb
, skb_headlen(skb
), skb
->tail
, delta
))
913 /* Optimization: no fragments, no reasons to preestimate
914 * size of pulled pages. Superb.
916 if (!skb_shinfo(skb
)->frag_list
)
919 /* Estimate size of pulled pages. */
921 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
922 if (skb_shinfo(skb
)->frags
[i
].size
>= eat
)
924 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
927 /* If we need update frag list, we are in troubles.
928 * Certainly, it possible to add an offset to skb data,
929 * but taking into account that pulling is expected to
930 * be very rare operation, it is worth to fight against
931 * further bloating skb head and crucify ourselves here instead.
932 * Pure masohism, indeed. 8)8)
935 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
936 struct sk_buff
*clone
= NULL
;
937 struct sk_buff
*insp
= NULL
;
942 if (list
->len
<= eat
) {
943 /* Eaten as whole. */
948 /* Eaten partially. */
950 if (skb_shared(list
)) {
951 /* Sucks! We need to fork list. :-( */
952 clone
= skb_clone(list
, GFP_ATOMIC
);
958 /* This may be pulled without
962 if (!pskb_pull(list
, eat
)) {
971 /* Free pulled out fragments. */
972 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
973 skb_shinfo(skb
)->frag_list
= list
->next
;
976 /* And insert new clone at head. */
979 skb_shinfo(skb
)->frag_list
= clone
;
982 /* Success! Now we may commit changes to skb data. */
987 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
988 if (skb_shinfo(skb
)->frags
[i
].size
<= eat
) {
989 put_page(skb_shinfo(skb
)->frags
[i
].page
);
990 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
992 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
994 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
995 skb_shinfo(skb
)->frags
[k
].size
-= eat
;
1001 skb_shinfo(skb
)->nr_frags
= k
;
1004 skb
->data_len
-= delta
;
1009 /* Copy some data bits from skb to kernel buffer. */
1011 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1014 int start
= skb_headlen(skb
);
1016 if (offset
> (int)skb
->len
- len
)
1020 if ((copy
= start
- offset
) > 0) {
1023 memcpy(to
, skb
->data
+ offset
, copy
);
1024 if ((len
-= copy
) == 0)
1030 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1033 BUG_TRAP(start
<= offset
+ len
);
1035 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1036 if ((copy
= end
- offset
) > 0) {
1042 vaddr
= kmap_skb_frag(&skb_shinfo(skb
)->frags
[i
]);
1044 vaddr
+ skb_shinfo(skb
)->frags
[i
].page_offset
+
1045 offset
- start
, copy
);
1046 kunmap_skb_frag(vaddr
);
1048 if ((len
-= copy
) == 0)
1056 if (skb_shinfo(skb
)->frag_list
) {
1057 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1059 for (; list
; list
= list
->next
) {
1062 BUG_TRAP(start
<= offset
+ len
);
1064 end
= start
+ list
->len
;
1065 if ((copy
= end
- offset
) > 0) {
1068 if (skb_copy_bits(list
, offset
- start
,
1071 if ((len
-= copy
) == 0)
1087 * skb_store_bits - store bits from kernel buffer to skb
1088 * @skb: destination buffer
1089 * @offset: offset in destination
1090 * @from: source buffer
1091 * @len: number of bytes to copy
1093 * Copy the specified number of bytes from the source buffer to the
1094 * destination skb. This function handles all the messy bits of
1095 * traversing fragment lists and such.
1098 int skb_store_bits(const struct sk_buff
*skb
, int offset
, void *from
, int len
)
1101 int start
= skb_headlen(skb
);
1103 if (offset
> (int)skb
->len
- len
)
1106 if ((copy
= start
- offset
) > 0) {
1109 memcpy(skb
->data
+ offset
, from
, copy
);
1110 if ((len
-= copy
) == 0)
1116 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1117 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1120 BUG_TRAP(start
<= offset
+ len
);
1122 end
= start
+ frag
->size
;
1123 if ((copy
= end
- offset
) > 0) {
1129 vaddr
= kmap_skb_frag(frag
);
1130 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1132 kunmap_skb_frag(vaddr
);
1134 if ((len
-= copy
) == 0)
1142 if (skb_shinfo(skb
)->frag_list
) {
1143 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1145 for (; list
; list
= list
->next
) {
1148 BUG_TRAP(start
<= offset
+ len
);
1150 end
= start
+ list
->len
;
1151 if ((copy
= end
- offset
) > 0) {
1154 if (skb_store_bits(list
, offset
- start
,
1157 if ((len
-= copy
) == 0)
1172 EXPORT_SYMBOL(skb_store_bits
);
1174 /* Checksum skb data. */
1176 unsigned int skb_checksum(const struct sk_buff
*skb
, int offset
,
1177 int len
, unsigned int csum
)
1179 int start
= skb_headlen(skb
);
1180 int i
, copy
= start
- offset
;
1183 /* Checksum header. */
1187 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1188 if ((len
-= copy
) == 0)
1194 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1197 BUG_TRAP(start
<= offset
+ len
);
1199 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1200 if ((copy
= end
- offset
) > 0) {
1203 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1207 vaddr
= kmap_skb_frag(frag
);
1208 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
1209 offset
- start
, copy
, 0);
1210 kunmap_skb_frag(vaddr
);
1211 csum
= csum_block_add(csum
, csum2
, pos
);
1220 if (skb_shinfo(skb
)->frag_list
) {
1221 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1223 for (; list
; list
= list
->next
) {
1226 BUG_TRAP(start
<= offset
+ len
);
1228 end
= start
+ list
->len
;
1229 if ((copy
= end
- offset
) > 0) {
1233 csum2
= skb_checksum(list
, offset
- start
,
1235 csum
= csum_block_add(csum
, csum2
, pos
);
1236 if ((len
-= copy
) == 0)
1249 /* Both of above in one bottle. */
1251 unsigned int skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
1252 u8
*to
, int len
, unsigned int csum
)
1254 int start
= skb_headlen(skb
);
1255 int i
, copy
= start
- offset
;
1262 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
1264 if ((len
-= copy
) == 0)
1271 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1274 BUG_TRAP(start
<= offset
+ len
);
1276 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1277 if ((copy
= end
- offset
) > 0) {
1280 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1284 vaddr
= kmap_skb_frag(frag
);
1285 csum2
= csum_partial_copy_nocheck(vaddr
+
1289 kunmap_skb_frag(vaddr
);
1290 csum
= csum_block_add(csum
, csum2
, pos
);
1300 if (skb_shinfo(skb
)->frag_list
) {
1301 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1303 for (; list
; list
= list
->next
) {
1307 BUG_TRAP(start
<= offset
+ len
);
1309 end
= start
+ list
->len
;
1310 if ((copy
= end
- offset
) > 0) {
1313 csum2
= skb_copy_and_csum_bits(list
,
1316 csum
= csum_block_add(csum
, csum2
, pos
);
1317 if ((len
-= copy
) == 0)
1330 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
1335 if (skb
->ip_summed
== CHECKSUM_HW
)
1336 csstart
= skb
->h
.raw
- skb
->data
;
1338 csstart
= skb_headlen(skb
);
1340 BUG_ON(csstart
> skb_headlen(skb
));
1342 memcpy(to
, skb
->data
, csstart
);
1345 if (csstart
!= skb
->len
)
1346 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
1347 skb
->len
- csstart
, 0);
1349 if (skb
->ip_summed
== CHECKSUM_HW
) {
1350 long csstuff
= csstart
+ skb
->csum
;
1352 *((unsigned short *)(to
+ csstuff
)) = csum_fold(csum
);
1357 * skb_dequeue - remove from the head of the queue
1358 * @list: list to dequeue from
1360 * Remove the head of the list. The list lock is taken so the function
1361 * may be used safely with other locking list functions. The head item is
1362 * returned or %NULL if the list is empty.
1365 struct sk_buff
*skb_dequeue(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(list
);
1372 spin_unlock_irqrestore(&list
->lock
, flags
);
1377 * skb_dequeue_tail - remove from the tail of the queue
1378 * @list: list to dequeue from
1380 * Remove the tail of the list. The list lock is taken so the function
1381 * may be used safely with other locking list functions. The tail item is
1382 * returned or %NULL if the list is empty.
1384 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
1386 unsigned long flags
;
1387 struct sk_buff
*result
;
1389 spin_lock_irqsave(&list
->lock
, flags
);
1390 result
= __skb_dequeue_tail(list
);
1391 spin_unlock_irqrestore(&list
->lock
, flags
);
1396 * skb_queue_purge - empty a list
1397 * @list: list to empty
1399 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1400 * the list and one reference dropped. This function takes the list
1401 * lock and is atomic with respect to other list locking functions.
1403 void skb_queue_purge(struct sk_buff_head
*list
)
1405 struct sk_buff
*skb
;
1406 while ((skb
= skb_dequeue(list
)) != NULL
)
1411 * skb_queue_head - queue a buffer at the list head
1412 * @list: list to use
1413 * @newsk: buffer to queue
1415 * Queue a buffer at the start of the list. This function takes the
1416 * list lock and can be used safely with other locking &sk_buff functions
1419 * A buffer cannot be placed on two lists at the same time.
1421 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1423 unsigned long flags
;
1425 spin_lock_irqsave(&list
->lock
, flags
);
1426 __skb_queue_head(list
, newsk
);
1427 spin_unlock_irqrestore(&list
->lock
, flags
);
1431 * skb_queue_tail - queue a buffer at the list tail
1432 * @list: list to use
1433 * @newsk: buffer to queue
1435 * Queue a buffer at the tail of the list. This function takes the
1436 * list lock and can be used safely with other locking &sk_buff functions
1439 * A buffer cannot be placed on two lists at the same time.
1441 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1443 unsigned long flags
;
1445 spin_lock_irqsave(&list
->lock
, flags
);
1446 __skb_queue_tail(list
, newsk
);
1447 spin_unlock_irqrestore(&list
->lock
, flags
);
1451 * skb_unlink - remove a buffer from a list
1452 * @skb: buffer to remove
1453 * @list: list to use
1455 * Remove a packet from a list. The list locks are taken and this
1456 * function is atomic with respect to other list locked calls
1458 * You must know what list the SKB is on.
1460 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
1462 unsigned long flags
;
1464 spin_lock_irqsave(&list
->lock
, flags
);
1465 __skb_unlink(skb
, list
);
1466 spin_unlock_irqrestore(&list
->lock
, flags
);
1470 * skb_append - append a buffer
1471 * @old: buffer to insert after
1472 * @newsk: buffer to insert
1473 * @list: list to use
1475 * Place a packet after a given packet in a list. The list locks are taken
1476 * and this function is atomic with respect to other list locked calls.
1477 * A buffer cannot be placed on two lists at the same time.
1479 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
1481 unsigned long flags
;
1483 spin_lock_irqsave(&list
->lock
, flags
);
1484 __skb_append(old
, newsk
, list
);
1485 spin_unlock_irqrestore(&list
->lock
, flags
);
1490 * skb_insert - insert a buffer
1491 * @old: buffer to insert before
1492 * @newsk: buffer to insert
1493 * @list: list to use
1495 * Place a packet before a given packet in a list. The list locks are
1496 * taken and this function is atomic with respect to other list locked
1499 * A buffer cannot be placed on two lists at the same time.
1501 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
1503 unsigned long flags
;
1505 spin_lock_irqsave(&list
->lock
, flags
);
1506 __skb_insert(newsk
, old
->prev
, old
, list
);
1507 spin_unlock_irqrestore(&list
->lock
, flags
);
1512 * Tune the memory allocator for a new MTU size.
1514 void skb_add_mtu(int mtu
)
1516 /* Must match allocation in alloc_skb */
1517 mtu
= SKB_DATA_ALIGN(mtu
) + sizeof(struct skb_shared_info
);
1519 kmem_add_cache_size(mtu
);
1523 static inline void skb_split_inside_header(struct sk_buff
*skb
,
1524 struct sk_buff
* skb1
,
1525 const u32 len
, const int pos
)
1529 memcpy(skb_put(skb1
, pos
- len
), skb
->data
+ len
, pos
- len
);
1531 /* And move data appendix as is. */
1532 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1533 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
1535 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
1536 skb_shinfo(skb
)->nr_frags
= 0;
1537 skb1
->data_len
= skb
->data_len
;
1538 skb1
->len
+= skb1
->data_len
;
1541 skb
->tail
= skb
->data
+ len
;
1544 static inline void skb_split_no_header(struct sk_buff
*skb
,
1545 struct sk_buff
* skb1
,
1546 const u32 len
, int pos
)
1549 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
1551 skb_shinfo(skb
)->nr_frags
= 0;
1552 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
1554 skb
->data_len
= len
- pos
;
1556 for (i
= 0; i
< nfrags
; i
++) {
1557 int size
= skb_shinfo(skb
)->frags
[i
].size
;
1559 if (pos
+ size
> len
) {
1560 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1564 * We have two variants in this case:
1565 * 1. Move all the frag to the second
1566 * part, if it is possible. F.e.
1567 * this approach is mandatory for TUX,
1568 * where splitting is expensive.
1569 * 2. Split is accurately. We make this.
1571 get_page(skb_shinfo(skb
)->frags
[i
].page
);
1572 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
1573 skb_shinfo(skb1
)->frags
[0].size
-= len
- pos
;
1574 skb_shinfo(skb
)->frags
[i
].size
= len
- pos
;
1575 skb_shinfo(skb
)->nr_frags
++;
1579 skb_shinfo(skb
)->nr_frags
++;
1582 skb_shinfo(skb1
)->nr_frags
= k
;
1586 * skb_split - Split fragmented skb to two parts at length len.
1587 * @skb: the buffer to split
1588 * @skb1: the buffer to receive the second part
1589 * @len: new length for skb
1591 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
1593 int pos
= skb_headlen(skb
);
1595 if (len
< pos
) /* Split line is inside header. */
1596 skb_split_inside_header(skb
, skb1
, len
, pos
);
1597 else /* Second chunk has no header, nothing to copy. */
1598 skb_split_no_header(skb
, skb1
, len
, pos
);
1602 * skb_prepare_seq_read - Prepare a sequential read of skb data
1603 * @skb: the buffer to read
1604 * @from: lower offset of data to be read
1605 * @to: upper offset of data to be read
1606 * @st: state variable
1608 * Initializes the specified state variable. Must be called before
1609 * invoking skb_seq_read() for the first time.
1611 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
1612 unsigned int to
, struct skb_seq_state
*st
)
1614 st
->lower_offset
= from
;
1615 st
->upper_offset
= to
;
1616 st
->root_skb
= st
->cur_skb
= skb
;
1617 st
->frag_idx
= st
->stepped_offset
= 0;
1618 st
->frag_data
= NULL
;
1622 * skb_seq_read - Sequentially read skb data
1623 * @consumed: number of bytes consumed by the caller so far
1624 * @data: destination pointer for data to be returned
1625 * @st: state variable
1627 * Reads a block of skb data at &consumed relative to the
1628 * lower offset specified to skb_prepare_seq_read(). Assigns
1629 * the head of the data block to &data and returns the length
1630 * of the block or 0 if the end of the skb data or the upper
1631 * offset has been reached.
1633 * The caller is not required to consume all of the data
1634 * returned, i.e. &consumed is typically set to the number
1635 * of bytes already consumed and the next call to
1636 * skb_seq_read() will return the remaining part of the block.
1638 * Note: The size of each block of data returned can be arbitary,
1639 * this limitation is the cost for zerocopy seqeuental
1640 * reads of potentially non linear data.
1642 * Note: Fragment lists within fragments are not implemented
1643 * at the moment, state->root_skb could be replaced with
1644 * a stack for this purpose.
1646 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
1647 struct skb_seq_state
*st
)
1649 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
1652 if (unlikely(abs_offset
>= st
->upper_offset
))
1656 block_limit
= skb_headlen(st
->cur_skb
);
1658 if (abs_offset
< block_limit
) {
1659 *data
= st
->cur_skb
->data
+ abs_offset
;
1660 return block_limit
- abs_offset
;
1663 if (st
->frag_idx
== 0 && !st
->frag_data
)
1664 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
1666 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
1667 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
1668 block_limit
= frag
->size
+ st
->stepped_offset
;
1670 if (abs_offset
< block_limit
) {
1672 st
->frag_data
= kmap_skb_frag(frag
);
1674 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
1675 (abs_offset
- st
->stepped_offset
);
1677 return block_limit
- abs_offset
;
1680 if (st
->frag_data
) {
1681 kunmap_skb_frag(st
->frag_data
);
1682 st
->frag_data
= NULL
;
1686 st
->stepped_offset
+= frag
->size
;
1689 if (st
->cur_skb
->next
) {
1690 st
->cur_skb
= st
->cur_skb
->next
;
1693 } else if (st
->root_skb
== st
->cur_skb
&&
1694 skb_shinfo(st
->root_skb
)->frag_list
) {
1695 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
1703 * skb_abort_seq_read - Abort a sequential read of skb data
1704 * @st: state variable
1706 * Must be called if skb_seq_read() was not called until it
1709 void skb_abort_seq_read(struct skb_seq_state
*st
)
1712 kunmap_skb_frag(st
->frag_data
);
1715 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
1717 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
1718 struct ts_config
*conf
,
1719 struct ts_state
*state
)
1721 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
1724 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
1726 skb_abort_seq_read(TS_SKB_CB(state
));
1730 * skb_find_text - Find a text pattern in skb data
1731 * @skb: the buffer to look in
1732 * @from: search offset
1734 * @config: textsearch configuration
1735 * @state: uninitialized textsearch state variable
1737 * Finds a pattern in the skb data according to the specified
1738 * textsearch configuration. Use textsearch_next() to retrieve
1739 * subsequent occurrences of the pattern. Returns the offset
1740 * to the first occurrence or UINT_MAX if no match was found.
1742 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
1743 unsigned int to
, struct ts_config
*config
,
1744 struct ts_state
*state
)
1746 config
->get_next_block
= skb_ts_get_next_block
;
1747 config
->finish
= skb_ts_finish
;
1749 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
1751 return textsearch_find(config
, state
);
1755 * skb_append_datato_frags: - append the user data to a skb
1756 * @sk: sock structure
1757 * @skb: skb structure to be appened with user data.
1758 * @getfrag: call back function to be used for getting the user data
1759 * @from: pointer to user message iov
1760 * @length: length of the iov message
1762 * Description: This procedure append the user data in the fragment part
1763 * of the skb if any page alloc fails user this procedure returns -ENOMEM
1765 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
1766 int (*getfrag
)(void *from
, char *to
, int offset
,
1767 int len
, int odd
, struct sk_buff
*skb
),
1768 void *from
, int length
)
1771 skb_frag_t
*frag
= NULL
;
1772 struct page
*page
= NULL
;
1778 /* Return error if we don't have space for new frag */
1779 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
1780 if (frg_cnt
>= MAX_SKB_FRAGS
)
1783 /* allocate a new page for next frag */
1784 page
= alloc_pages(sk
->sk_allocation
, 0);
1786 /* If alloc_page fails just return failure and caller will
1787 * free previous allocated pages by doing kfree_skb()
1792 /* initialize the next frag */
1793 sk
->sk_sndmsg_page
= page
;
1794 sk
->sk_sndmsg_off
= 0;
1795 skb_fill_page_desc(skb
, frg_cnt
, page
, 0, 0);
1796 skb
->truesize
+= PAGE_SIZE
;
1797 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
1799 /* get the new initialized frag */
1800 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
1801 frag
= &skb_shinfo(skb
)->frags
[frg_cnt
- 1];
1803 /* copy the user data to page */
1804 left
= PAGE_SIZE
- frag
->page_offset
;
1805 copy
= (length
> left
)? left
: length
;
1807 ret
= getfrag(from
, (page_address(frag
->page
) +
1808 frag
->page_offset
+ frag
->size
),
1809 offset
, copy
, 0, skb
);
1813 /* copy was successful so update the size parameters */
1814 sk
->sk_sndmsg_off
+= copy
;
1817 skb
->data_len
+= copy
;
1821 } while (length
> 0);
1826 void __init
skb_init(void)
1828 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
1829 sizeof(struct sk_buff
),
1833 if (!skbuff_head_cache
)
1834 panic("cannot create skbuff cache");
1836 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
1837 (2*sizeof(struct sk_buff
)) +
1842 if (!skbuff_fclone_cache
)
1843 panic("cannot create skbuff cache");
1846 EXPORT_SYMBOL(___pskb_trim
);
1847 EXPORT_SYMBOL(__kfree_skb
);
1848 EXPORT_SYMBOL(__pskb_pull_tail
);
1849 EXPORT_SYMBOL(__alloc_skb
);
1850 EXPORT_SYMBOL(pskb_copy
);
1851 EXPORT_SYMBOL(pskb_expand_head
);
1852 EXPORT_SYMBOL(skb_checksum
);
1853 EXPORT_SYMBOL(skb_clone
);
1854 EXPORT_SYMBOL(skb_clone_fraglist
);
1855 EXPORT_SYMBOL(skb_copy
);
1856 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
1857 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
1858 EXPORT_SYMBOL(skb_copy_bits
);
1859 EXPORT_SYMBOL(skb_copy_expand
);
1860 EXPORT_SYMBOL(skb_over_panic
);
1861 EXPORT_SYMBOL(skb_pad
);
1862 EXPORT_SYMBOL(skb_realloc_headroom
);
1863 EXPORT_SYMBOL(skb_under_panic
);
1864 EXPORT_SYMBOL(skb_dequeue
);
1865 EXPORT_SYMBOL(skb_dequeue_tail
);
1866 EXPORT_SYMBOL(skb_insert
);
1867 EXPORT_SYMBOL(skb_queue_purge
);
1868 EXPORT_SYMBOL(skb_queue_head
);
1869 EXPORT_SYMBOL(skb_queue_tail
);
1870 EXPORT_SYMBOL(skb_unlink
);
1871 EXPORT_SYMBOL(skb_append
);
1872 EXPORT_SYMBOL(skb_split
);
1873 EXPORT_SYMBOL(skb_prepare_seq_read
);
1874 EXPORT_SYMBOL(skb_seq_read
);
1875 EXPORT_SYMBOL(skb_abort_seq_read
);
1876 EXPORT_SYMBOL(skb_find_text
);
1877 EXPORT_SYMBOL(skb_append_datato_frags
);