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
126 * Allocate a new &sk_buff. The returned buffer has no headroom and a
127 * tail room of size bytes. The object has a reference count of one.
128 * The return is the buffer. On a failure the return is %NULL.
130 * Buffers may only be allocated from interrupts using a @gfp_mask of
133 struct sk_buff
*__alloc_skb(unsigned int size
, gfp_t gfp_mask
,
141 skb
= kmem_cache_alloc(skbuff_fclone_cache
,
142 gfp_mask
& ~__GFP_DMA
);
144 skb
= kmem_cache_alloc(skbuff_head_cache
,
145 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
;
164 struct sk_buff
*child
= skb
+ 1;
165 atomic_t
*fclone_ref
= (atomic_t
*) (child
+ 1);
167 skb
->fclone
= SKB_FCLONE_ORIG
;
168 atomic_set(fclone_ref
, 1);
170 child
->fclone
= SKB_FCLONE_UNAVAILABLE
;
172 atomic_set(&(skb_shinfo(skb
)->dataref
), 1);
173 skb_shinfo(skb
)->nr_frags
= 0;
174 skb_shinfo(skb
)->tso_size
= 0;
175 skb_shinfo(skb
)->tso_segs
= 0;
176 skb_shinfo(skb
)->frag_list
= NULL
;
180 kmem_cache_free(skbuff_head_cache
, skb
);
186 * alloc_skb_from_cache - allocate a network buffer
187 * @cp: kmem_cache from which to allocate the data area
188 * (object size must be big enough for @size bytes + skb overheads)
189 * @size: size to allocate
190 * @gfp_mask: allocation mask
192 * Allocate a new &sk_buff. The returned buffer has no headroom and
193 * tail room of size bytes. The object has a reference count of one.
194 * The return is the buffer. On a failure the return is %NULL.
196 * Buffers may only be allocated from interrupts using a @gfp_mask of
199 struct sk_buff
*alloc_skb_from_cache(kmem_cache_t
*cp
,
207 skb
= kmem_cache_alloc(skbuff_head_cache
,
208 gfp_mask
& ~__GFP_DMA
);
213 size
= SKB_DATA_ALIGN(size
);
214 data
= kmem_cache_alloc(cp
, gfp_mask
);
218 memset(skb
, 0, offsetof(struct sk_buff
, truesize
));
219 skb
->truesize
= size
+ sizeof(struct sk_buff
);
220 atomic_set(&skb
->users
, 1);
224 skb
->end
= data
+ size
;
226 atomic_set(&(skb_shinfo(skb
)->dataref
), 1);
227 skb_shinfo(skb
)->nr_frags
= 0;
228 skb_shinfo(skb
)->tso_size
= 0;
229 skb_shinfo(skb
)->tso_segs
= 0;
230 skb_shinfo(skb
)->frag_list
= NULL
;
234 kmem_cache_free(skbuff_head_cache
, skb
);
240 static void skb_drop_fraglist(struct sk_buff
*skb
)
242 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
244 skb_shinfo(skb
)->frag_list
= NULL
;
247 struct sk_buff
*this = list
;
253 static void skb_clone_fraglist(struct sk_buff
*skb
)
255 struct sk_buff
*list
;
257 for (list
= skb_shinfo(skb
)->frag_list
; list
; list
= list
->next
)
261 void skb_release_data(struct sk_buff
*skb
)
264 !atomic_sub_return(skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1,
265 &skb_shinfo(skb
)->dataref
)) {
266 if (skb_shinfo(skb
)->nr_frags
) {
268 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
269 put_page(skb_shinfo(skb
)->frags
[i
].page
);
272 if (skb_shinfo(skb
)->frag_list
)
273 skb_drop_fraglist(skb
);
280 * Free an skbuff by memory without cleaning the state.
282 void kfree_skbmem(struct sk_buff
*skb
)
284 struct sk_buff
*other
;
285 atomic_t
*fclone_ref
;
287 skb_release_data(skb
);
288 switch (skb
->fclone
) {
289 case SKB_FCLONE_UNAVAILABLE
:
290 kmem_cache_free(skbuff_head_cache
, skb
);
293 case SKB_FCLONE_ORIG
:
294 fclone_ref
= (atomic_t
*) (skb
+ 2);
295 if (atomic_dec_and_test(fclone_ref
))
296 kmem_cache_free(skbuff_fclone_cache
, skb
);
299 case SKB_FCLONE_CLONE
:
300 fclone_ref
= (atomic_t
*) (skb
+ 1);
303 /* The clone portion is available for
304 * fast-cloning again.
306 skb
->fclone
= SKB_FCLONE_UNAVAILABLE
;
308 if (atomic_dec_and_test(fclone_ref
))
309 kmem_cache_free(skbuff_fclone_cache
, other
);
315 * __kfree_skb - private function
318 * Free an sk_buff. Release anything attached to the buffer.
319 * Clean the state. This is an internal helper function. Users should
320 * always call kfree_skb
323 void __kfree_skb(struct sk_buff
*skb
)
325 dst_release(skb
->dst
);
327 secpath_put(skb
->sp
);
329 if (skb
->destructor
) {
331 skb
->destructor(skb
);
333 #ifdef CONFIG_NETFILTER
334 nf_conntrack_put(skb
->nfct
);
335 #ifdef CONFIG_BRIDGE_NETFILTER
336 nf_bridge_put(skb
->nf_bridge
);
339 /* XXX: IS this still necessary? - JHS */
340 #ifdef CONFIG_NET_SCHED
342 #ifdef CONFIG_NET_CLS_ACT
351 * skb_clone - duplicate an sk_buff
352 * @skb: buffer to clone
353 * @gfp_mask: allocation priority
355 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
356 * copies share the same packet data but not structure. The new
357 * buffer has a reference count of 1. If the allocation fails the
358 * function returns %NULL otherwise the new buffer is returned.
360 * If this function is called from an interrupt gfp_mask() must be
364 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
369 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
370 n
->fclone
== SKB_FCLONE_UNAVAILABLE
) {
371 atomic_t
*fclone_ref
= (atomic_t
*) (n
+ 1);
372 n
->fclone
= SKB_FCLONE_CLONE
;
373 atomic_inc(fclone_ref
);
375 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
378 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
381 #define C(x) n->x = skb->x
383 n
->next
= n
->prev
= NULL
;
394 secpath_get(skb
->sp
);
396 memcpy(n
->cb
, skb
->cb
, sizeof(skb
->cb
));
407 n
->destructor
= NULL
;
408 #ifdef CONFIG_NETFILTER
411 nf_conntrack_get(skb
->nfct
);
413 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
416 #ifdef CONFIG_BRIDGE_NETFILTER
418 nf_bridge_get(skb
->nf_bridge
);
420 #endif /*CONFIG_NETFILTER*/
421 #ifdef CONFIG_NET_SCHED
423 #ifdef CONFIG_NET_CLS_ACT
424 n
->tc_verd
= SET_TC_VERD(skb
->tc_verd
,0);
425 n
->tc_verd
= CLR_TC_OK2MUNGE(n
->tc_verd
);
426 n
->tc_verd
= CLR_TC_MUNGED(n
->tc_verd
);
432 atomic_set(&n
->users
, 1);
438 atomic_inc(&(skb_shinfo(skb
)->dataref
));
444 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
447 * Shift between the two data areas in bytes
449 unsigned long offset
= new->data
- old
->data
;
453 new->priority
= old
->priority
;
454 new->protocol
= old
->protocol
;
455 new->dst
= dst_clone(old
->dst
);
457 new->sp
= secpath_get(old
->sp
);
459 new->h
.raw
= old
->h
.raw
+ offset
;
460 new->nh
.raw
= old
->nh
.raw
+ offset
;
461 new->mac
.raw
= old
->mac
.raw
+ offset
;
462 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
463 new->local_df
= old
->local_df
;
464 new->fclone
= SKB_FCLONE_UNAVAILABLE
;
465 new->pkt_type
= old
->pkt_type
;
466 new->tstamp
= old
->tstamp
;
467 new->destructor
= NULL
;
468 #ifdef CONFIG_NETFILTER
469 new->nfmark
= old
->nfmark
;
470 new->nfct
= old
->nfct
;
471 nf_conntrack_get(old
->nfct
);
472 new->nfctinfo
= old
->nfctinfo
;
473 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
474 new->ipvs_property
= old
->ipvs_property
;
476 #ifdef CONFIG_BRIDGE_NETFILTER
477 new->nf_bridge
= old
->nf_bridge
;
478 nf_bridge_get(old
->nf_bridge
);
481 #ifdef CONFIG_NET_SCHED
482 #ifdef CONFIG_NET_CLS_ACT
483 new->tc_verd
= old
->tc_verd
;
485 new->tc_index
= old
->tc_index
;
487 atomic_set(&new->users
, 1);
488 skb_shinfo(new)->tso_size
= skb_shinfo(old
)->tso_size
;
489 skb_shinfo(new)->tso_segs
= skb_shinfo(old
)->tso_segs
;
493 * skb_copy - create private copy of an sk_buff
494 * @skb: buffer to copy
495 * @gfp_mask: allocation priority
497 * Make a copy of both an &sk_buff and its data. This is used when the
498 * caller wishes to modify the data and needs a private copy of the
499 * data to alter. Returns %NULL on failure or the pointer to the buffer
500 * on success. The returned buffer has a reference count of 1.
502 * As by-product this function converts non-linear &sk_buff to linear
503 * one, so that &sk_buff becomes completely private and caller is allowed
504 * to modify all the data of returned buffer. This means that this
505 * function is not recommended for use in circumstances when only
506 * header is going to be modified. Use pskb_copy() instead.
509 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
511 int headerlen
= skb
->data
- skb
->head
;
513 * Allocate the copy buffer
515 struct sk_buff
*n
= alloc_skb(skb
->end
- skb
->head
+ skb
->data_len
,
520 /* Set the data pointer */
521 skb_reserve(n
, headerlen
);
522 /* Set the tail pointer and length */
523 skb_put(n
, skb
->len
);
525 n
->ip_summed
= skb
->ip_summed
;
527 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
530 copy_skb_header(n
, skb
);
536 * pskb_copy - create copy of an sk_buff with private head.
537 * @skb: buffer to copy
538 * @gfp_mask: allocation priority
540 * Make a copy of both an &sk_buff and part of its data, located
541 * in header. Fragmented data remain shared. This is used when
542 * the caller wishes to modify only header of &sk_buff and needs
543 * private copy of the header to alter. Returns %NULL on failure
544 * or the pointer to the buffer on success.
545 * The returned buffer has a reference count of 1.
548 struct sk_buff
*pskb_copy(struct sk_buff
*skb
, gfp_t gfp_mask
)
551 * Allocate the copy buffer
553 struct sk_buff
*n
= alloc_skb(skb
->end
- skb
->head
, gfp_mask
);
558 /* Set the data pointer */
559 skb_reserve(n
, skb
->data
- skb
->head
);
560 /* Set the tail pointer and length */
561 skb_put(n
, skb_headlen(skb
));
563 memcpy(n
->data
, skb
->data
, n
->len
);
565 n
->ip_summed
= skb
->ip_summed
;
567 n
->data_len
= skb
->data_len
;
570 if (skb_shinfo(skb
)->nr_frags
) {
573 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
574 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
575 get_page(skb_shinfo(n
)->frags
[i
].page
);
577 skb_shinfo(n
)->nr_frags
= i
;
580 if (skb_shinfo(skb
)->frag_list
) {
581 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
582 skb_clone_fraglist(n
);
585 copy_skb_header(n
, skb
);
591 * pskb_expand_head - reallocate header of &sk_buff
592 * @skb: buffer to reallocate
593 * @nhead: room to add at head
594 * @ntail: room to add at tail
595 * @gfp_mask: allocation priority
597 * Expands (or creates identical copy, if &nhead and &ntail are zero)
598 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
599 * reference count of 1. Returns zero in the case of success or error,
600 * if expansion failed. In the last case, &sk_buff is not changed.
602 * All the pointers pointing into skb header may change and must be
603 * reloaded after call to this function.
606 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
611 int size
= nhead
+ (skb
->end
- skb
->head
) + ntail
;
617 size
= SKB_DATA_ALIGN(size
);
619 data
= kmalloc(size
+ sizeof(struct skb_shared_info
), gfp_mask
);
623 /* Copy only real data... and, alas, header. This should be
624 * optimized for the cases when header is void. */
625 memcpy(data
+ nhead
, skb
->head
, skb
->tail
- skb
->head
);
626 memcpy(data
+ size
, skb
->end
, sizeof(struct skb_shared_info
));
628 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
629 get_page(skb_shinfo(skb
)->frags
[i
].page
);
631 if (skb_shinfo(skb
)->frag_list
)
632 skb_clone_fraglist(skb
);
634 skb_release_data(skb
);
636 off
= (data
+ nhead
) - skb
->head
;
639 skb
->end
= data
+ size
;
647 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
654 /* Make private copy of skb with writable head and some headroom */
656 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
658 struct sk_buff
*skb2
;
659 int delta
= headroom
- skb_headroom(skb
);
662 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
664 skb2
= skb_clone(skb
, GFP_ATOMIC
);
665 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
676 * skb_copy_expand - copy and expand sk_buff
677 * @skb: buffer to copy
678 * @newheadroom: new free bytes at head
679 * @newtailroom: new free bytes at tail
680 * @gfp_mask: allocation priority
682 * Make a copy of both an &sk_buff and its data and while doing so
683 * allocate additional space.
685 * This is used when the caller wishes to modify the data and needs a
686 * private copy of the data to alter as well as more space for new fields.
687 * Returns %NULL on failure or the pointer to the buffer
688 * on success. The returned buffer has a reference count of 1.
690 * You must pass %GFP_ATOMIC as the allocation priority if this function
691 * is called from an interrupt.
693 * BUG ALERT: ip_summed is not copied. Why does this work? Is it used
694 * only by netfilter in the cases when checksum is recalculated? --ANK
696 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
697 int newheadroom
, int newtailroom
,
701 * Allocate the copy buffer
703 struct sk_buff
*n
= alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
705 int head_copy_len
, head_copy_off
;
710 skb_reserve(n
, newheadroom
);
712 /* Set the tail pointer and length */
713 skb_put(n
, skb
->len
);
715 head_copy_len
= skb_headroom(skb
);
717 if (newheadroom
<= head_copy_len
)
718 head_copy_len
= newheadroom
;
720 head_copy_off
= newheadroom
- head_copy_len
;
722 /* Copy the linear header and data. */
723 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
724 skb
->len
+ head_copy_len
))
727 copy_skb_header(n
, skb
);
733 * skb_pad - zero pad the tail of an skb
734 * @skb: buffer to pad
737 * Ensure that a buffer is followed by a padding area that is zero
738 * filled. Used by network drivers which may DMA or transfer data
739 * beyond the buffer end onto the wire.
741 * May return NULL in out of memory cases.
744 struct sk_buff
*skb_pad(struct sk_buff
*skb
, int pad
)
746 struct sk_buff
*nskb
;
748 /* If the skbuff is non linear tailroom is always zero.. */
749 if (skb_tailroom(skb
) >= pad
) {
750 memset(skb
->data
+skb
->len
, 0, pad
);
754 nskb
= skb_copy_expand(skb
, skb_headroom(skb
), skb_tailroom(skb
) + pad
, GFP_ATOMIC
);
757 memset(nskb
->data
+nskb
->len
, 0, pad
);
761 /* Trims skb to length len. It can change skb pointers, if "realloc" is 1.
762 * If realloc==0 and trimming is impossible without change of data,
766 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
, int realloc
)
768 int offset
= skb_headlen(skb
);
769 int nfrags
= skb_shinfo(skb
)->nr_frags
;
772 for (i
= 0; i
< nfrags
; i
++) {
773 int end
= offset
+ skb_shinfo(skb
)->frags
[i
].size
;
775 if (skb_cloned(skb
)) {
778 if (pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))
782 put_page(skb_shinfo(skb
)->frags
[i
].page
);
783 skb_shinfo(skb
)->nr_frags
--;
785 skb_shinfo(skb
)->frags
[i
].size
= len
- offset
;
792 skb
->data_len
-= skb
->len
- len
;
795 if (len
<= skb_headlen(skb
)) {
798 skb
->tail
= skb
->data
+ len
;
799 if (skb_shinfo(skb
)->frag_list
&& !skb_cloned(skb
))
800 skb_drop_fraglist(skb
);
802 skb
->data_len
-= skb
->len
- len
;
811 * __pskb_pull_tail - advance tail of skb header
812 * @skb: buffer to reallocate
813 * @delta: number of bytes to advance tail
815 * The function makes a sense only on a fragmented &sk_buff,
816 * it expands header moving its tail forward and copying necessary
817 * data from fragmented part.
819 * &sk_buff MUST have reference count of 1.
821 * Returns %NULL (and &sk_buff does not change) if pull failed
822 * or value of new tail of skb in the case of success.
824 * All the pointers pointing into skb header may change and must be
825 * reloaded after call to this function.
828 /* Moves tail of skb head forward, copying data from fragmented part,
829 * when it is necessary.
830 * 1. It may fail due to malloc failure.
831 * 2. It may change skb pointers.
833 * It is pretty complicated. Luckily, it is called only in exceptional cases.
835 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
837 /* If skb has not enough free space at tail, get new one
838 * plus 128 bytes for future expansions. If we have enough
839 * room at tail, reallocate without expansion only if skb is cloned.
841 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
843 if (eat
> 0 || skb_cloned(skb
)) {
844 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
849 if (skb_copy_bits(skb
, skb_headlen(skb
), skb
->tail
, delta
))
852 /* Optimization: no fragments, no reasons to preestimate
853 * size of pulled pages. Superb.
855 if (!skb_shinfo(skb
)->frag_list
)
858 /* Estimate size of pulled pages. */
860 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
861 if (skb_shinfo(skb
)->frags
[i
].size
>= eat
)
863 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
866 /* If we need update frag list, we are in troubles.
867 * Certainly, it possible to add an offset to skb data,
868 * but taking into account that pulling is expected to
869 * be very rare operation, it is worth to fight against
870 * further bloating skb head and crucify ourselves here instead.
871 * Pure masohism, indeed. 8)8)
874 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
875 struct sk_buff
*clone
= NULL
;
876 struct sk_buff
*insp
= NULL
;
882 if (list
->len
<= eat
) {
883 /* Eaten as whole. */
888 /* Eaten partially. */
890 if (skb_shared(list
)) {
891 /* Sucks! We need to fork list. :-( */
892 clone
= skb_clone(list
, GFP_ATOMIC
);
898 /* This may be pulled without
902 if (!pskb_pull(list
, eat
)) {
911 /* Free pulled out fragments. */
912 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
913 skb_shinfo(skb
)->frag_list
= list
->next
;
916 /* And insert new clone at head. */
919 skb_shinfo(skb
)->frag_list
= clone
;
922 /* Success! Now we may commit changes to skb data. */
927 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
928 if (skb_shinfo(skb
)->frags
[i
].size
<= eat
) {
929 put_page(skb_shinfo(skb
)->frags
[i
].page
);
930 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
932 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
934 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
935 skb_shinfo(skb
)->frags
[k
].size
-= eat
;
941 skb_shinfo(skb
)->nr_frags
= k
;
944 skb
->data_len
-= delta
;
949 /* Copy some data bits from skb to kernel buffer. */
951 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
954 int start
= skb_headlen(skb
);
956 if (offset
> (int)skb
->len
- len
)
960 if ((copy
= start
- offset
) > 0) {
963 memcpy(to
, skb
->data
+ offset
, copy
);
964 if ((len
-= copy
) == 0)
970 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
973 BUG_TRAP(start
<= offset
+ len
);
975 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
976 if ((copy
= end
- offset
) > 0) {
982 vaddr
= kmap_skb_frag(&skb_shinfo(skb
)->frags
[i
]);
984 vaddr
+ skb_shinfo(skb
)->frags
[i
].page_offset
+
985 offset
- start
, copy
);
986 kunmap_skb_frag(vaddr
);
988 if ((len
-= copy
) == 0)
996 if (skb_shinfo(skb
)->frag_list
) {
997 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
999 for (; list
; list
= list
->next
) {
1002 BUG_TRAP(start
<= offset
+ len
);
1004 end
= start
+ list
->len
;
1005 if ((copy
= end
- offset
) > 0) {
1008 if (skb_copy_bits(list
, offset
- start
,
1011 if ((len
-= copy
) == 0)
1027 * skb_store_bits - store bits from kernel buffer to skb
1028 * @skb: destination buffer
1029 * @offset: offset in destination
1030 * @from: source buffer
1031 * @len: number of bytes to copy
1033 * Copy the specified number of bytes from the source buffer to the
1034 * destination skb. This function handles all the messy bits of
1035 * traversing fragment lists and such.
1038 int skb_store_bits(const struct sk_buff
*skb
, int offset
, void *from
, int len
)
1041 int start
= skb_headlen(skb
);
1043 if (offset
> (int)skb
->len
- len
)
1046 if ((copy
= start
- offset
) > 0) {
1049 memcpy(skb
->data
+ offset
, from
, copy
);
1050 if ((len
-= copy
) == 0)
1056 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1057 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1060 BUG_TRAP(start
<= offset
+ len
);
1062 end
= start
+ frag
->size
;
1063 if ((copy
= end
- offset
) > 0) {
1069 vaddr
= kmap_skb_frag(frag
);
1070 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1072 kunmap_skb_frag(vaddr
);
1074 if ((len
-= copy
) == 0)
1082 if (skb_shinfo(skb
)->frag_list
) {
1083 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1085 for (; list
; list
= list
->next
) {
1088 BUG_TRAP(start
<= offset
+ len
);
1090 end
= start
+ list
->len
;
1091 if ((copy
= end
- offset
) > 0) {
1094 if (skb_store_bits(list
, offset
- start
,
1097 if ((len
-= copy
) == 0)
1112 EXPORT_SYMBOL(skb_store_bits
);
1114 /* Checksum skb data. */
1116 unsigned int skb_checksum(const struct sk_buff
*skb
, int offset
,
1117 int len
, unsigned int csum
)
1119 int start
= skb_headlen(skb
);
1120 int i
, copy
= start
- offset
;
1123 /* Checksum header. */
1127 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1128 if ((len
-= copy
) == 0)
1134 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1137 BUG_TRAP(start
<= offset
+ len
);
1139 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1140 if ((copy
= end
- offset
) > 0) {
1143 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1147 vaddr
= kmap_skb_frag(frag
);
1148 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
1149 offset
- start
, copy
, 0);
1150 kunmap_skb_frag(vaddr
);
1151 csum
= csum_block_add(csum
, csum2
, pos
);
1160 if (skb_shinfo(skb
)->frag_list
) {
1161 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1163 for (; list
; list
= list
->next
) {
1166 BUG_TRAP(start
<= offset
+ len
);
1168 end
= start
+ list
->len
;
1169 if ((copy
= end
- offset
) > 0) {
1173 csum2
= skb_checksum(list
, offset
- start
,
1175 csum
= csum_block_add(csum
, csum2
, pos
);
1176 if ((len
-= copy
) == 0)
1190 /* Both of above in one bottle. */
1192 unsigned int skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
1193 u8
*to
, int len
, unsigned int csum
)
1195 int start
= skb_headlen(skb
);
1196 int i
, copy
= start
- offset
;
1203 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
1205 if ((len
-= copy
) == 0)
1212 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1215 BUG_TRAP(start
<= offset
+ len
);
1217 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1218 if ((copy
= end
- offset
) > 0) {
1221 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1225 vaddr
= kmap_skb_frag(frag
);
1226 csum2
= csum_partial_copy_nocheck(vaddr
+
1230 kunmap_skb_frag(vaddr
);
1231 csum
= csum_block_add(csum
, csum2
, pos
);
1241 if (skb_shinfo(skb
)->frag_list
) {
1242 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1244 for (; list
; list
= list
->next
) {
1248 BUG_TRAP(start
<= offset
+ len
);
1250 end
= start
+ list
->len
;
1251 if ((copy
= end
- offset
) > 0) {
1254 csum2
= skb_copy_and_csum_bits(list
,
1257 csum
= csum_block_add(csum
, csum2
, pos
);
1258 if ((len
-= copy
) == 0)
1272 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
1277 if (skb
->ip_summed
== CHECKSUM_HW
)
1278 csstart
= skb
->h
.raw
- skb
->data
;
1280 csstart
= skb_headlen(skb
);
1282 if (csstart
> skb_headlen(skb
))
1285 memcpy(to
, skb
->data
, csstart
);
1288 if (csstart
!= skb
->len
)
1289 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
1290 skb
->len
- csstart
, 0);
1292 if (skb
->ip_summed
== CHECKSUM_HW
) {
1293 long csstuff
= csstart
+ skb
->csum
;
1295 *((unsigned short *)(to
+ csstuff
)) = csum_fold(csum
);
1300 * skb_dequeue - remove from the head of the queue
1301 * @list: list to dequeue from
1303 * Remove the head of the list. The list lock is taken so the function
1304 * may be used safely with other locking list functions. The head item is
1305 * returned or %NULL if the list is empty.
1308 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
1310 unsigned long flags
;
1311 struct sk_buff
*result
;
1313 spin_lock_irqsave(&list
->lock
, flags
);
1314 result
= __skb_dequeue(list
);
1315 spin_unlock_irqrestore(&list
->lock
, flags
);
1320 * skb_dequeue_tail - remove from the tail of the queue
1321 * @list: list to dequeue from
1323 * Remove the tail of the list. The list lock is taken so the function
1324 * may be used safely with other locking list functions. The tail item is
1325 * returned or %NULL if the list is empty.
1327 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
1329 unsigned long flags
;
1330 struct sk_buff
*result
;
1332 spin_lock_irqsave(&list
->lock
, flags
);
1333 result
= __skb_dequeue_tail(list
);
1334 spin_unlock_irqrestore(&list
->lock
, flags
);
1339 * skb_queue_purge - empty a list
1340 * @list: list to empty
1342 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1343 * the list and one reference dropped. This function takes the list
1344 * lock and is atomic with respect to other list locking functions.
1346 void skb_queue_purge(struct sk_buff_head
*list
)
1348 struct sk_buff
*skb
;
1349 while ((skb
= skb_dequeue(list
)) != NULL
)
1354 * skb_queue_head - queue a buffer at the list head
1355 * @list: list to use
1356 * @newsk: buffer to queue
1358 * Queue a buffer at the start of the list. This function takes the
1359 * list lock and can be used safely with other locking &sk_buff functions
1362 * A buffer cannot be placed on two lists at the same time.
1364 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1366 unsigned long flags
;
1368 spin_lock_irqsave(&list
->lock
, flags
);
1369 __skb_queue_head(list
, newsk
);
1370 spin_unlock_irqrestore(&list
->lock
, flags
);
1374 * skb_queue_tail - queue a buffer at the list tail
1375 * @list: list to use
1376 * @newsk: buffer to queue
1378 * Queue a buffer at the tail of the list. This function takes the
1379 * list lock and can be used safely with other locking &sk_buff functions
1382 * A buffer cannot be placed on two lists at the same time.
1384 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1386 unsigned long flags
;
1388 spin_lock_irqsave(&list
->lock
, flags
);
1389 __skb_queue_tail(list
, newsk
);
1390 spin_unlock_irqrestore(&list
->lock
, flags
);
1394 * skb_unlink - remove a buffer from a list
1395 * @skb: buffer to remove
1396 * @list: list to use
1398 * Remove a packet from a list. The list locks are taken and this
1399 * function is atomic with respect to other list locked calls
1401 * You must know what list the SKB is on.
1403 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
1405 unsigned long flags
;
1407 spin_lock_irqsave(&list
->lock
, flags
);
1408 __skb_unlink(skb
, list
);
1409 spin_unlock_irqrestore(&list
->lock
, flags
);
1413 * skb_append - append a buffer
1414 * @old: buffer to insert after
1415 * @newsk: buffer to insert
1416 * @list: list to use
1418 * Place a packet after a given packet in a list. The list locks are taken
1419 * and this function is atomic with respect to other list locked calls.
1420 * A buffer cannot be placed on two lists at the same time.
1422 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
1424 unsigned long flags
;
1426 spin_lock_irqsave(&list
->lock
, flags
);
1427 __skb_append(old
, newsk
, list
);
1428 spin_unlock_irqrestore(&list
->lock
, flags
);
1433 * skb_insert - insert a buffer
1434 * @old: buffer to insert before
1435 * @newsk: buffer to insert
1436 * @list: list to use
1438 * Place a packet before a given packet in a list. The list locks are
1439 * taken and this function is atomic with respect to other list locked
1442 * A buffer cannot be placed on two lists at the same time.
1444 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
1446 unsigned long flags
;
1448 spin_lock_irqsave(&list
->lock
, flags
);
1449 __skb_insert(newsk
, old
->prev
, old
, list
);
1450 spin_unlock_irqrestore(&list
->lock
, flags
);
1455 * Tune the memory allocator for a new MTU size.
1457 void skb_add_mtu(int mtu
)
1459 /* Must match allocation in alloc_skb */
1460 mtu
= SKB_DATA_ALIGN(mtu
) + sizeof(struct skb_shared_info
);
1462 kmem_add_cache_size(mtu
);
1466 static inline void skb_split_inside_header(struct sk_buff
*skb
,
1467 struct sk_buff
* skb1
,
1468 const u32 len
, const int pos
)
1472 memcpy(skb_put(skb1
, pos
- len
), skb
->data
+ len
, pos
- len
);
1474 /* And move data appendix as is. */
1475 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1476 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
1478 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
1479 skb_shinfo(skb
)->nr_frags
= 0;
1480 skb1
->data_len
= skb
->data_len
;
1481 skb1
->len
+= skb1
->data_len
;
1484 skb
->tail
= skb
->data
+ len
;
1487 static inline void skb_split_no_header(struct sk_buff
*skb
,
1488 struct sk_buff
* skb1
,
1489 const u32 len
, int pos
)
1492 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
1494 skb_shinfo(skb
)->nr_frags
= 0;
1495 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
1497 skb
->data_len
= len
- pos
;
1499 for (i
= 0; i
< nfrags
; i
++) {
1500 int size
= skb_shinfo(skb
)->frags
[i
].size
;
1502 if (pos
+ size
> len
) {
1503 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1507 * We have two variants in this case:
1508 * 1. Move all the frag to the second
1509 * part, if it is possible. F.e.
1510 * this approach is mandatory for TUX,
1511 * where splitting is expensive.
1512 * 2. Split is accurately. We make this.
1514 get_page(skb_shinfo(skb
)->frags
[i
].page
);
1515 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
1516 skb_shinfo(skb1
)->frags
[0].size
-= len
- pos
;
1517 skb_shinfo(skb
)->frags
[i
].size
= len
- pos
;
1518 skb_shinfo(skb
)->nr_frags
++;
1522 skb_shinfo(skb
)->nr_frags
++;
1525 skb_shinfo(skb1
)->nr_frags
= k
;
1529 * skb_split - Split fragmented skb to two parts at length len.
1530 * @skb: the buffer to split
1531 * @skb1: the buffer to receive the second part
1532 * @len: new length for skb
1534 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
1536 int pos
= skb_headlen(skb
);
1538 if (len
< pos
) /* Split line is inside header. */
1539 skb_split_inside_header(skb
, skb1
, len
, pos
);
1540 else /* Second chunk has no header, nothing to copy. */
1541 skb_split_no_header(skb
, skb1
, len
, pos
);
1545 * skb_prepare_seq_read - Prepare a sequential read of skb data
1546 * @skb: the buffer to read
1547 * @from: lower offset of data to be read
1548 * @to: upper offset of data to be read
1549 * @st: state variable
1551 * Initializes the specified state variable. Must be called before
1552 * invoking skb_seq_read() for the first time.
1554 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
1555 unsigned int to
, struct skb_seq_state
*st
)
1557 st
->lower_offset
= from
;
1558 st
->upper_offset
= to
;
1559 st
->root_skb
= st
->cur_skb
= skb
;
1560 st
->frag_idx
= st
->stepped_offset
= 0;
1561 st
->frag_data
= NULL
;
1565 * skb_seq_read - Sequentially read skb data
1566 * @consumed: number of bytes consumed by the caller so far
1567 * @data: destination pointer for data to be returned
1568 * @st: state variable
1570 * Reads a block of skb data at &consumed relative to the
1571 * lower offset specified to skb_prepare_seq_read(). Assigns
1572 * the head of the data block to &data and returns the length
1573 * of the block or 0 if the end of the skb data or the upper
1574 * offset has been reached.
1576 * The caller is not required to consume all of the data
1577 * returned, i.e. &consumed is typically set to the number
1578 * of bytes already consumed and the next call to
1579 * skb_seq_read() will return the remaining part of the block.
1581 * Note: The size of each block of data returned can be arbitary,
1582 * this limitation is the cost for zerocopy seqeuental
1583 * reads of potentially non linear data.
1585 * Note: Fragment lists within fragments are not implemented
1586 * at the moment, state->root_skb could be replaced with
1587 * a stack for this purpose.
1589 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
1590 struct skb_seq_state
*st
)
1592 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
1595 if (unlikely(abs_offset
>= st
->upper_offset
))
1599 block_limit
= skb_headlen(st
->cur_skb
);
1601 if (abs_offset
< block_limit
) {
1602 *data
= st
->cur_skb
->data
+ abs_offset
;
1603 return block_limit
- abs_offset
;
1606 if (st
->frag_idx
== 0 && !st
->frag_data
)
1607 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
1609 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
1610 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
1611 block_limit
= frag
->size
+ st
->stepped_offset
;
1613 if (abs_offset
< block_limit
) {
1615 st
->frag_data
= kmap_skb_frag(frag
);
1617 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
1618 (abs_offset
- st
->stepped_offset
);
1620 return block_limit
- abs_offset
;
1623 if (st
->frag_data
) {
1624 kunmap_skb_frag(st
->frag_data
);
1625 st
->frag_data
= NULL
;
1629 st
->stepped_offset
+= frag
->size
;
1632 if (st
->cur_skb
->next
) {
1633 st
->cur_skb
= st
->cur_skb
->next
;
1636 } else if (st
->root_skb
== st
->cur_skb
&&
1637 skb_shinfo(st
->root_skb
)->frag_list
) {
1638 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
1646 * skb_abort_seq_read - Abort a sequential read of skb data
1647 * @st: state variable
1649 * Must be called if skb_seq_read() was not called until it
1652 void skb_abort_seq_read(struct skb_seq_state
*st
)
1655 kunmap_skb_frag(st
->frag_data
);
1658 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
1660 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
1661 struct ts_config
*conf
,
1662 struct ts_state
*state
)
1664 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
1667 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
1669 skb_abort_seq_read(TS_SKB_CB(state
));
1673 * skb_find_text - Find a text pattern in skb data
1674 * @skb: the buffer to look in
1675 * @from: search offset
1677 * @config: textsearch configuration
1678 * @state: uninitialized textsearch state variable
1680 * Finds a pattern in the skb data according to the specified
1681 * textsearch configuration. Use textsearch_next() to retrieve
1682 * subsequent occurrences of the pattern. Returns the offset
1683 * to the first occurrence or UINT_MAX if no match was found.
1685 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
1686 unsigned int to
, struct ts_config
*config
,
1687 struct ts_state
*state
)
1689 config
->get_next_block
= skb_ts_get_next_block
;
1690 config
->finish
= skb_ts_finish
;
1692 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
1694 return textsearch_find(config
, state
);
1697 void __init
skb_init(void)
1699 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
1700 sizeof(struct sk_buff
),
1704 if (!skbuff_head_cache
)
1705 panic("cannot create skbuff cache");
1707 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
1708 (2*sizeof(struct sk_buff
)) +
1713 if (!skbuff_fclone_cache
)
1714 panic("cannot create skbuff cache");
1717 EXPORT_SYMBOL(___pskb_trim
);
1718 EXPORT_SYMBOL(__kfree_skb
);
1719 EXPORT_SYMBOL(__pskb_pull_tail
);
1720 EXPORT_SYMBOL(__alloc_skb
);
1721 EXPORT_SYMBOL(pskb_copy
);
1722 EXPORT_SYMBOL(pskb_expand_head
);
1723 EXPORT_SYMBOL(skb_checksum
);
1724 EXPORT_SYMBOL(skb_clone
);
1725 EXPORT_SYMBOL(skb_clone_fraglist
);
1726 EXPORT_SYMBOL(skb_copy
);
1727 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
1728 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
1729 EXPORT_SYMBOL(skb_copy_bits
);
1730 EXPORT_SYMBOL(skb_copy_expand
);
1731 EXPORT_SYMBOL(skb_over_panic
);
1732 EXPORT_SYMBOL(skb_pad
);
1733 EXPORT_SYMBOL(skb_realloc_headroom
);
1734 EXPORT_SYMBOL(skb_under_panic
);
1735 EXPORT_SYMBOL(skb_dequeue
);
1736 EXPORT_SYMBOL(skb_dequeue_tail
);
1737 EXPORT_SYMBOL(skb_insert
);
1738 EXPORT_SYMBOL(skb_queue_purge
);
1739 EXPORT_SYMBOL(skb_queue_head
);
1740 EXPORT_SYMBOL(skb_queue_tail
);
1741 EXPORT_SYMBOL(skb_unlink
);
1742 EXPORT_SYMBOL(skb_append
);
1743 EXPORT_SYMBOL(skb_split
);
1744 EXPORT_SYMBOL(skb_prepare_seq_read
);
1745 EXPORT_SYMBOL(skb_seq_read
);
1746 EXPORT_SYMBOL(skb_abort_seq_read
);
1747 EXPORT_SYMBOL(skb_find_text
);