2 * Routines having to do with the 'struct sk_buff' memory handlers.
4 * Authors: Alan Cox <iiitac@pyr.swan.ac.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
7 * Version: $Id: skbuff.c,v 1.90 2001/11/07 05:56:19 davem Exp $
10 * Alan Cox : Fixed the worst of the load
12 * Dave Platt : Interrupt stacking fix.
13 * Richard Kooijman : Timestamp fixes.
14 * Alan Cox : Changed buffer format.
15 * Alan Cox : destructor hook for AF_UNIX etc.
16 * Linus Torvalds : Better skb_clone.
17 * Alan Cox : Added skb_copy.
18 * Alan Cox : Added all the changed routines Linus
19 * only put in the headers
20 * Ray VanTassle : Fixed --skb->lock in free
21 * Alan Cox : skb_copy copy arp field
22 * Andi Kleen : slabified it.
23 * Robert Olsson : Removed skb_head_pool
26 * The __skb_ routines should be called with interrupts
27 * disabled, or you better be *real* sure that the operation is atomic
28 * with respect to whatever list is being frobbed (e.g. via lock_sock()
29 * or via disabling bottom half handlers, etc).
31 * This program is free software; you can redistribute it and/or
32 * modify it under the terms of the GNU General Public License
33 * as published by the Free Software Foundation; either version
34 * 2 of the License, or (at your option) any later version.
38 * The functions in this file will not compile correctly with gcc 2.4.x
41 #include <linux/module.h>
42 #include <linux/types.h>
43 #include <linux/kernel.h>
45 #include <linux/interrupt.h>
47 #include <linux/inet.h>
48 #include <linux/slab.h>
49 #include <linux/netdevice.h>
50 #ifdef CONFIG_NET_CLS_ACT
51 #include <net/pkt_sched.h>
53 #include <linux/string.h>
54 #include <linux/skbuff.h>
55 #include <linux/cache.h>
56 #include <linux/rtnetlink.h>
57 #include <linux/init.h>
59 #include <net/protocol.h>
62 #include <net/checksum.h>
65 #include <asm/uaccess.h>
66 #include <asm/system.h>
70 static struct kmem_cache
*skbuff_head_cache __read_mostly
;
71 static struct kmem_cache
*skbuff_fclone_cache __read_mostly
;
74 * Keep out-of-line to prevent kernel bloat.
75 * __builtin_return_address is not used because it is not always
80 * skb_over_panic - private function
85 * Out of line support code for skb_put(). Not user callable.
87 void skb_over_panic(struct sk_buff
*skb
, int sz
, void *here
)
89 printk(KERN_EMERG
"skb_over_panic: text:%p len:%d put:%d head:%p "
90 "data:%p tail:%#lx end:%p dev:%s\n",
91 here
, skb
->len
, sz
, skb
->head
, skb
->data
,
92 (unsigned long)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:%#lx end:%p dev:%s\n",
110 here
, skb
->len
, sz
, skb
->head
, skb
->data
,
111 (unsigned long)skb
->tail
, skb
->end
,
112 skb
->dev
? skb
->dev
->name
: "<NULL>");
116 void skb_truesize_bug(struct sk_buff
*skb
)
118 printk(KERN_ERR
"SKB BUG: Invalid truesize (%u) "
119 "len=%u, sizeof(sk_buff)=%Zd\n",
120 skb
->truesize
, skb
->len
, sizeof(struct sk_buff
));
122 EXPORT_SYMBOL(skb_truesize_bug
);
124 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
125 * 'private' fields and also do memory statistics to find all the
131 * __alloc_skb - allocate a network buffer
132 * @size: size to allocate
133 * @gfp_mask: allocation mask
134 * @fclone: allocate from fclone cache instead of head cache
135 * and allocate a cloned (child) skb
136 * @node: numa node to allocate memory on
138 * Allocate a new &sk_buff. The returned buffer has no headroom and a
139 * tail room of size bytes. The object has a reference count of one.
140 * The return is the buffer. On a failure the return is %NULL.
142 * Buffers may only be allocated from interrupts using a @gfp_mask of
145 struct sk_buff
*__alloc_skb(unsigned int size
, gfp_t gfp_mask
,
146 int fclone
, int node
)
148 struct kmem_cache
*cache
;
149 struct skb_shared_info
*shinfo
;
153 cache
= fclone
? skbuff_fclone_cache
: skbuff_head_cache
;
156 skb
= kmem_cache_alloc_node(cache
, gfp_mask
& ~__GFP_DMA
, node
);
160 /* Get the DATA. Size must match skb_add_mtu(). */
161 size
= SKB_DATA_ALIGN(size
);
162 data
= kmalloc_node_track_caller(size
+ sizeof(struct skb_shared_info
),
167 memset(skb
, 0, offsetof(struct sk_buff
, truesize
));
168 skb
->truesize
= size
+ sizeof(struct sk_buff
);
169 atomic_set(&skb
->users
, 1);
172 skb_reset_tail_pointer(skb
);
173 skb
->end
= data
+ size
;
174 /* make sure we initialize shinfo sequentially */
175 shinfo
= skb_shinfo(skb
);
176 atomic_set(&shinfo
->dataref
, 1);
177 shinfo
->nr_frags
= 0;
178 shinfo
->gso_size
= 0;
179 shinfo
->gso_segs
= 0;
180 shinfo
->gso_type
= 0;
181 shinfo
->ip6_frag_id
= 0;
182 shinfo
->frag_list
= NULL
;
185 struct sk_buff
*child
= skb
+ 1;
186 atomic_t
*fclone_ref
= (atomic_t
*) (child
+ 1);
188 skb
->fclone
= SKB_FCLONE_ORIG
;
189 atomic_set(fclone_ref
, 1);
191 child
->fclone
= SKB_FCLONE_UNAVAILABLE
;
196 kmem_cache_free(cache
, skb
);
202 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
203 * @dev: network device to receive on
204 * @length: length to allocate
205 * @gfp_mask: get_free_pages mask, passed to alloc_skb
207 * Allocate a new &sk_buff and assign it a usage count of one. The
208 * buffer has unspecified headroom built in. Users should allocate
209 * the headroom they think they need without accounting for the
210 * built in space. The built in space is used for optimisations.
212 * %NULL is returned if there is no free memory.
214 struct sk_buff
*__netdev_alloc_skb(struct net_device
*dev
,
215 unsigned int length
, gfp_t gfp_mask
)
217 int node
= dev
->dev
.parent
? dev_to_node(dev
->dev
.parent
) : -1;
220 skb
= __alloc_skb(length
+ NET_SKB_PAD
, gfp_mask
, 0, node
);
222 skb_reserve(skb
, NET_SKB_PAD
);
228 static void skb_drop_list(struct sk_buff
**listp
)
230 struct sk_buff
*list
= *listp
;
235 struct sk_buff
*this = list
;
241 static inline void skb_drop_fraglist(struct sk_buff
*skb
)
243 skb_drop_list(&skb_shinfo(skb
)->frag_list
);
246 static void skb_clone_fraglist(struct sk_buff
*skb
)
248 struct sk_buff
*list
;
250 for (list
= skb_shinfo(skb
)->frag_list
; list
; list
= list
->next
)
254 static void skb_release_data(struct sk_buff
*skb
)
257 !atomic_sub_return(skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1,
258 &skb_shinfo(skb
)->dataref
)) {
259 if (skb_shinfo(skb
)->nr_frags
) {
261 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
262 put_page(skb_shinfo(skb
)->frags
[i
].page
);
265 if (skb_shinfo(skb
)->frag_list
)
266 skb_drop_fraglist(skb
);
273 * Free an skbuff by memory without cleaning the state.
275 void kfree_skbmem(struct sk_buff
*skb
)
277 struct sk_buff
*other
;
278 atomic_t
*fclone_ref
;
280 skb_release_data(skb
);
281 switch (skb
->fclone
) {
282 case SKB_FCLONE_UNAVAILABLE
:
283 kmem_cache_free(skbuff_head_cache
, skb
);
286 case SKB_FCLONE_ORIG
:
287 fclone_ref
= (atomic_t
*) (skb
+ 2);
288 if (atomic_dec_and_test(fclone_ref
))
289 kmem_cache_free(skbuff_fclone_cache
, skb
);
292 case SKB_FCLONE_CLONE
:
293 fclone_ref
= (atomic_t
*) (skb
+ 1);
296 /* The clone portion is available for
297 * fast-cloning again.
299 skb
->fclone
= SKB_FCLONE_UNAVAILABLE
;
301 if (atomic_dec_and_test(fclone_ref
))
302 kmem_cache_free(skbuff_fclone_cache
, other
);
308 * __kfree_skb - private function
311 * Free an sk_buff. Release anything attached to the buffer.
312 * Clean the state. This is an internal helper function. Users should
313 * always call kfree_skb
316 void __kfree_skb(struct sk_buff
*skb
)
318 dst_release(skb
->dst
);
320 secpath_put(skb
->sp
);
322 if (skb
->destructor
) {
324 skb
->destructor(skb
);
326 #ifdef CONFIG_NETFILTER
327 nf_conntrack_put(skb
->nfct
);
328 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
329 nf_conntrack_put_reasm(skb
->nfct_reasm
);
331 #ifdef CONFIG_BRIDGE_NETFILTER
332 nf_bridge_put(skb
->nf_bridge
);
335 /* XXX: IS this still necessary? - JHS */
336 #ifdef CONFIG_NET_SCHED
338 #ifdef CONFIG_NET_CLS_ACT
347 * kfree_skb - free an sk_buff
348 * @skb: buffer to free
350 * Drop a reference to the buffer and free it if the usage count has
353 void kfree_skb(struct sk_buff
*skb
)
357 if (likely(atomic_read(&skb
->users
) == 1))
359 else if (likely(!atomic_dec_and_test(&skb
->users
)))
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
;
428 #ifdef CONFIG_NET_SCHED
430 #ifdef CONFIG_NET_CLS_ACT
431 n
->tc_verd
= SET_TC_VERD(skb
->tc_verd
,0);
432 n
->tc_verd
= CLR_TC_OK2MUNGE(n
->tc_verd
);
433 n
->tc_verd
= CLR_TC_MUNGED(n
->tc_verd
);
436 skb_copy_secmark(n
, skb
);
439 atomic_set(&n
->users
, 1);
445 atomic_inc(&(skb_shinfo(skb
)->dataref
));
451 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
453 #ifndef NET_SKBUFF_DATA_USES_OFFSET
455 * Shift between the two data areas in bytes
457 unsigned long offset
= new->data
- old
->data
;
461 new->priority
= old
->priority
;
462 new->protocol
= old
->protocol
;
463 new->dst
= dst_clone(old
->dst
);
465 new->sp
= secpath_get(old
->sp
);
467 new->transport_header
= old
->transport_header
;
468 new->network_header
= old
->network_header
;
469 new->mac_header
= old
->mac_header
;
470 #ifndef NET_SKBUFF_DATA_USES_OFFSET
471 /* {transport,network,mac}_header are relative to skb->head */
472 new->transport_header
+= offset
;
473 new->network_header
+= offset
;
474 new->mac_header
+= offset
;
476 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
477 new->local_df
= old
->local_df
;
478 new->fclone
= SKB_FCLONE_UNAVAILABLE
;
479 new->pkt_type
= old
->pkt_type
;
480 new->tstamp
= old
->tstamp
;
481 new->destructor
= NULL
;
482 new->mark
= old
->mark
;
484 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
485 new->ipvs_property
= old
->ipvs_property
;
487 #ifdef CONFIG_NET_SCHED
488 #ifdef CONFIG_NET_CLS_ACT
489 new->tc_verd
= old
->tc_verd
;
491 new->tc_index
= old
->tc_index
;
493 skb_copy_secmark(new, old
);
494 atomic_set(&new->users
, 1);
495 skb_shinfo(new)->gso_size
= skb_shinfo(old
)->gso_size
;
496 skb_shinfo(new)->gso_segs
= skb_shinfo(old
)->gso_segs
;
497 skb_shinfo(new)->gso_type
= skb_shinfo(old
)->gso_type
;
501 * skb_copy - create private copy of an sk_buff
502 * @skb: buffer to copy
503 * @gfp_mask: allocation priority
505 * Make a copy of both an &sk_buff and its data. This is used when the
506 * caller wishes to modify the data and needs a private copy of the
507 * data to alter. Returns %NULL on failure or the pointer to the buffer
508 * on success. The returned buffer has a reference count of 1.
510 * As by-product this function converts non-linear &sk_buff to linear
511 * one, so that &sk_buff becomes completely private and caller is allowed
512 * to modify all the data of returned buffer. This means that this
513 * function is not recommended for use in circumstances when only
514 * header is going to be modified. Use pskb_copy() instead.
517 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
519 int headerlen
= skb
->data
- skb
->head
;
521 * Allocate the copy buffer
523 struct sk_buff
*n
= alloc_skb(skb
->end
- skb
->head
+ skb
->data_len
,
528 /* Set the data pointer */
529 skb_reserve(n
, headerlen
);
530 /* Set the tail pointer and length */
531 skb_put(n
, skb
->len
);
533 n
->ip_summed
= skb
->ip_summed
;
535 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
538 copy_skb_header(n
, skb
);
544 * pskb_copy - create copy of an sk_buff with private head.
545 * @skb: buffer to copy
546 * @gfp_mask: allocation priority
548 * Make a copy of both an &sk_buff and part of its data, located
549 * in header. Fragmented data remain shared. This is used when
550 * the caller wishes to modify only header of &sk_buff and needs
551 * private copy of the header to alter. Returns %NULL on failure
552 * or the pointer to the buffer on success.
553 * The returned buffer has a reference count of 1.
556 struct sk_buff
*pskb_copy(struct sk_buff
*skb
, gfp_t gfp_mask
)
559 * Allocate the copy buffer
561 struct sk_buff
*n
= alloc_skb(skb
->end
- skb
->head
, gfp_mask
);
566 /* Set the data pointer */
567 skb_reserve(n
, skb
->data
- skb
->head
);
568 /* Set the tail pointer and length */
569 skb_put(n
, skb_headlen(skb
));
571 memcpy(n
->data
, skb
->data
, n
->len
);
573 n
->ip_summed
= skb
->ip_summed
;
575 n
->truesize
+= skb
->data_len
;
576 n
->data_len
= skb
->data_len
;
579 if (skb_shinfo(skb
)->nr_frags
) {
582 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
583 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
584 get_page(skb_shinfo(n
)->frags
[i
].page
);
586 skb_shinfo(n
)->nr_frags
= i
;
589 if (skb_shinfo(skb
)->frag_list
) {
590 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
591 skb_clone_fraglist(n
);
594 copy_skb_header(n
, skb
);
600 * pskb_expand_head - reallocate header of &sk_buff
601 * @skb: buffer to reallocate
602 * @nhead: room to add at head
603 * @ntail: room to add at tail
604 * @gfp_mask: allocation priority
606 * Expands (or creates identical copy, if &nhead and &ntail are zero)
607 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
608 * reference count of 1. Returns zero in the case of success or error,
609 * if expansion failed. In the last case, &sk_buff is not changed.
611 * All the pointers pointing into skb header may change and must be
612 * reloaded after call to this function.
615 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
620 int size
= nhead
+ (skb
->end
- skb
->head
) + ntail
;
626 size
= SKB_DATA_ALIGN(size
);
628 data
= kmalloc(size
+ sizeof(struct skb_shared_info
), gfp_mask
);
632 /* Copy only real data... and, alas, header. This should be
633 * optimized for the cases when header is void. */
634 memcpy(data
+ nhead
, skb
->head
,
636 #ifndef NET_SKBUFF_DATA_USES_OFFSET
640 memcpy(data
+ size
, skb
->end
, sizeof(struct skb_shared_info
));
642 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
643 get_page(skb_shinfo(skb
)->frags
[i
].page
);
645 if (skb_shinfo(skb
)->frag_list
)
646 skb_clone_fraglist(skb
);
648 skb_release_data(skb
);
650 off
= (data
+ nhead
) - skb
->head
;
653 skb
->end
= data
+ size
;
655 #ifndef NET_SKBUFF_DATA_USES_OFFSET
656 /* {transport,network,mac}_header and tail are relative to skb->head */
658 skb
->transport_header
+= off
;
659 skb
->network_header
+= off
;
660 skb
->mac_header
+= off
;
664 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
671 /* Make private copy of skb with writable head and some headroom */
673 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
675 struct sk_buff
*skb2
;
676 int delta
= headroom
- skb_headroom(skb
);
679 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
681 skb2
= skb_clone(skb
, GFP_ATOMIC
);
682 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
693 * skb_copy_expand - copy and expand sk_buff
694 * @skb: buffer to copy
695 * @newheadroom: new free bytes at head
696 * @newtailroom: new free bytes at tail
697 * @gfp_mask: allocation priority
699 * Make a copy of both an &sk_buff and its data and while doing so
700 * allocate additional space.
702 * This is used when the caller wishes to modify the data and needs a
703 * private copy of the data to alter as well as more space for new fields.
704 * Returns %NULL on failure or the pointer to the buffer
705 * on success. The returned buffer has a reference count of 1.
707 * You must pass %GFP_ATOMIC as the allocation priority if this function
708 * is called from an interrupt.
710 * BUG ALERT: ip_summed is not copied. Why does this work? Is it used
711 * only by netfilter in the cases when checksum is recalculated? --ANK
713 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
714 int newheadroom
, int newtailroom
,
718 * Allocate the copy buffer
720 struct sk_buff
*n
= alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
722 int head_copy_len
, head_copy_off
;
727 skb_reserve(n
, newheadroom
);
729 /* Set the tail pointer and length */
730 skb_put(n
, skb
->len
);
732 head_copy_len
= skb_headroom(skb
);
734 if (newheadroom
<= head_copy_len
)
735 head_copy_len
= newheadroom
;
737 head_copy_off
= newheadroom
- head_copy_len
;
739 /* Copy the linear header and data. */
740 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
741 skb
->len
+ head_copy_len
))
744 copy_skb_header(n
, skb
);
750 * skb_pad - zero pad the tail of an skb
751 * @skb: buffer to pad
754 * Ensure that a buffer is followed by a padding area that is zero
755 * filled. Used by network drivers which may DMA or transfer data
756 * beyond the buffer end onto the wire.
758 * May return error in out of memory cases. The skb is freed on error.
761 int skb_pad(struct sk_buff
*skb
, int pad
)
766 /* If the skbuff is non linear tailroom is always zero.. */
767 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
768 memset(skb
->data
+skb
->len
, 0, pad
);
772 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb_tail_pointer(skb
));
773 if (likely(skb_cloned(skb
) || ntail
> 0)) {
774 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
779 /* FIXME: The use of this function with non-linear skb's really needs
782 err
= skb_linearize(skb
);
786 memset(skb
->data
+ skb
->len
, 0, pad
);
794 /* Trims skb to length len. It can change skb pointers.
797 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
799 struct sk_buff
**fragp
;
800 struct sk_buff
*frag
;
801 int offset
= skb_headlen(skb
);
802 int nfrags
= skb_shinfo(skb
)->nr_frags
;
806 if (skb_cloned(skb
) &&
807 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
814 for (; i
< nfrags
; i
++) {
815 int end
= offset
+ skb_shinfo(skb
)->frags
[i
].size
;
822 skb_shinfo(skb
)->frags
[i
++].size
= len
- offset
;
825 skb_shinfo(skb
)->nr_frags
= i
;
827 for (; i
< nfrags
; i
++)
828 put_page(skb_shinfo(skb
)->frags
[i
].page
);
830 if (skb_shinfo(skb
)->frag_list
)
831 skb_drop_fraglist(skb
);
835 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
836 fragp
= &frag
->next
) {
837 int end
= offset
+ frag
->len
;
839 if (skb_shared(frag
)) {
840 struct sk_buff
*nfrag
;
842 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
843 if (unlikely(!nfrag
))
846 nfrag
->next
= frag
->next
;
858 unlikely((err
= pskb_trim(frag
, len
- offset
))))
862 skb_drop_list(&frag
->next
);
867 if (len
> skb_headlen(skb
)) {
868 skb
->data_len
-= skb
->len
- len
;
873 skb_set_tail_pointer(skb
, len
);
880 * __pskb_pull_tail - advance tail of skb header
881 * @skb: buffer to reallocate
882 * @delta: number of bytes to advance tail
884 * The function makes a sense only on a fragmented &sk_buff,
885 * it expands header moving its tail forward and copying necessary
886 * data from fragmented part.
888 * &sk_buff MUST have reference count of 1.
890 * Returns %NULL (and &sk_buff does not change) if pull failed
891 * or value of new tail of skb in the case of success.
893 * All the pointers pointing into skb header may change and must be
894 * reloaded after call to this function.
897 /* Moves tail of skb head forward, copying data from fragmented part,
898 * when it is necessary.
899 * 1. It may fail due to malloc failure.
900 * 2. It may change skb pointers.
902 * It is pretty complicated. Luckily, it is called only in exceptional cases.
904 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
906 /* If skb has not enough free space at tail, get new one
907 * plus 128 bytes for future expansions. If we have enough
908 * room at tail, reallocate without expansion only if skb is cloned.
910 int i
, k
, eat
= (skb_tail_pointer(skb
) + delta
) - skb
->end
;
912 if (eat
> 0 || skb_cloned(skb
)) {
913 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
918 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
921 /* Optimization: no fragments, no reasons to preestimate
922 * size of pulled pages. Superb.
924 if (!skb_shinfo(skb
)->frag_list
)
927 /* Estimate size of pulled pages. */
929 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
930 if (skb_shinfo(skb
)->frags
[i
].size
>= eat
)
932 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
935 /* If we need update frag list, we are in troubles.
936 * Certainly, it possible to add an offset to skb data,
937 * but taking into account that pulling is expected to
938 * be very rare operation, it is worth to fight against
939 * further bloating skb head and crucify ourselves here instead.
940 * Pure masohism, indeed. 8)8)
943 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
944 struct sk_buff
*clone
= NULL
;
945 struct sk_buff
*insp
= NULL
;
950 if (list
->len
<= eat
) {
951 /* Eaten as whole. */
956 /* Eaten partially. */
958 if (skb_shared(list
)) {
959 /* Sucks! We need to fork list. :-( */
960 clone
= skb_clone(list
, GFP_ATOMIC
);
966 /* This may be pulled without
970 if (!pskb_pull(list
, eat
)) {
979 /* Free pulled out fragments. */
980 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
981 skb_shinfo(skb
)->frag_list
= list
->next
;
984 /* And insert new clone at head. */
987 skb_shinfo(skb
)->frag_list
= clone
;
990 /* Success! Now we may commit changes to skb data. */
995 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
996 if (skb_shinfo(skb
)->frags
[i
].size
<= eat
) {
997 put_page(skb_shinfo(skb
)->frags
[i
].page
);
998 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1000 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1002 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1003 skb_shinfo(skb
)->frags
[k
].size
-= eat
;
1009 skb_shinfo(skb
)->nr_frags
= k
;
1012 skb
->data_len
-= delta
;
1014 return skb_tail_pointer(skb
);
1017 /* Copy some data bits from skb to kernel buffer. */
1019 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1022 int start
= skb_headlen(skb
);
1024 if (offset
> (int)skb
->len
- len
)
1028 if ((copy
= start
- offset
) > 0) {
1031 memcpy(to
, skb
->data
+ offset
, copy
);
1032 if ((len
-= copy
) == 0)
1038 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1041 BUG_TRAP(start
<= offset
+ len
);
1043 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1044 if ((copy
= end
- offset
) > 0) {
1050 vaddr
= kmap_skb_frag(&skb_shinfo(skb
)->frags
[i
]);
1052 vaddr
+ skb_shinfo(skb
)->frags
[i
].page_offset
+
1053 offset
- start
, copy
);
1054 kunmap_skb_frag(vaddr
);
1056 if ((len
-= copy
) == 0)
1064 if (skb_shinfo(skb
)->frag_list
) {
1065 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1067 for (; list
; list
= list
->next
) {
1070 BUG_TRAP(start
<= offset
+ len
);
1072 end
= start
+ list
->len
;
1073 if ((copy
= end
- offset
) > 0) {
1076 if (skb_copy_bits(list
, offset
- start
,
1079 if ((len
-= copy
) == 0)
1095 * skb_store_bits - store bits from kernel buffer to skb
1096 * @skb: destination buffer
1097 * @offset: offset in destination
1098 * @from: source buffer
1099 * @len: number of bytes to copy
1101 * Copy the specified number of bytes from the source buffer to the
1102 * destination skb. This function handles all the messy bits of
1103 * traversing fragment lists and such.
1106 int skb_store_bits(const struct sk_buff
*skb
, int offset
, void *from
, int len
)
1109 int start
= skb_headlen(skb
);
1111 if (offset
> (int)skb
->len
- len
)
1114 if ((copy
= start
- offset
) > 0) {
1117 memcpy(skb
->data
+ offset
, from
, copy
);
1118 if ((len
-= copy
) == 0)
1124 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1125 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1128 BUG_TRAP(start
<= offset
+ len
);
1130 end
= start
+ frag
->size
;
1131 if ((copy
= end
- offset
) > 0) {
1137 vaddr
= kmap_skb_frag(frag
);
1138 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1140 kunmap_skb_frag(vaddr
);
1142 if ((len
-= copy
) == 0)
1150 if (skb_shinfo(skb
)->frag_list
) {
1151 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1153 for (; list
; list
= list
->next
) {
1156 BUG_TRAP(start
<= offset
+ len
);
1158 end
= start
+ list
->len
;
1159 if ((copy
= end
- offset
) > 0) {
1162 if (skb_store_bits(list
, offset
- start
,
1165 if ((len
-= copy
) == 0)
1180 EXPORT_SYMBOL(skb_store_bits
);
1182 /* Checksum skb data. */
1184 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
1185 int len
, __wsum csum
)
1187 int start
= skb_headlen(skb
);
1188 int i
, copy
= start
- offset
;
1191 /* Checksum header. */
1195 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1196 if ((len
-= copy
) == 0)
1202 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1205 BUG_TRAP(start
<= offset
+ len
);
1207 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1208 if ((copy
= end
- offset
) > 0) {
1211 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1215 vaddr
= kmap_skb_frag(frag
);
1216 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
1217 offset
- start
, copy
, 0);
1218 kunmap_skb_frag(vaddr
);
1219 csum
= csum_block_add(csum
, csum2
, pos
);
1228 if (skb_shinfo(skb
)->frag_list
) {
1229 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1231 for (; list
; list
= list
->next
) {
1234 BUG_TRAP(start
<= offset
+ len
);
1236 end
= start
+ list
->len
;
1237 if ((copy
= end
- offset
) > 0) {
1241 csum2
= skb_checksum(list
, offset
- start
,
1243 csum
= csum_block_add(csum
, csum2
, pos
);
1244 if ((len
-= copy
) == 0)
1257 /* Both of above in one bottle. */
1259 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
1260 u8
*to
, int len
, __wsum csum
)
1262 int start
= skb_headlen(skb
);
1263 int i
, copy
= start
- offset
;
1270 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
1272 if ((len
-= copy
) == 0)
1279 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1282 BUG_TRAP(start
<= offset
+ len
);
1284 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1285 if ((copy
= end
- offset
) > 0) {
1288 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1292 vaddr
= kmap_skb_frag(frag
);
1293 csum2
= csum_partial_copy_nocheck(vaddr
+
1297 kunmap_skb_frag(vaddr
);
1298 csum
= csum_block_add(csum
, csum2
, pos
);
1308 if (skb_shinfo(skb
)->frag_list
) {
1309 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1311 for (; list
; list
= list
->next
) {
1315 BUG_TRAP(start
<= offset
+ len
);
1317 end
= start
+ list
->len
;
1318 if ((copy
= end
- offset
) > 0) {
1321 csum2
= skb_copy_and_csum_bits(list
,
1324 csum
= csum_block_add(csum
, csum2
, pos
);
1325 if ((len
-= copy
) == 0)
1338 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
1343 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1344 csstart
= skb_transport_offset(skb
);
1346 csstart
= skb_headlen(skb
);
1348 BUG_ON(csstart
> skb_headlen(skb
));
1350 memcpy(to
, skb
->data
, csstart
);
1353 if (csstart
!= skb
->len
)
1354 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
1355 skb
->len
- csstart
, 0);
1357 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1358 long csstuff
= csstart
+ skb
->csum_offset
;
1360 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
1365 * skb_dequeue - remove from the head of the queue
1366 * @list: list to dequeue from
1368 * Remove the head of the list. The list lock is taken so the function
1369 * may be used safely with other locking list functions. The head item is
1370 * returned or %NULL if the list is empty.
1373 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
1375 unsigned long flags
;
1376 struct sk_buff
*result
;
1378 spin_lock_irqsave(&list
->lock
, flags
);
1379 result
= __skb_dequeue(list
);
1380 spin_unlock_irqrestore(&list
->lock
, flags
);
1385 * skb_dequeue_tail - remove from the tail of the queue
1386 * @list: list to dequeue from
1388 * Remove the tail of the list. The list lock is taken so the function
1389 * may be used safely with other locking list functions. The tail item is
1390 * returned or %NULL if the list is empty.
1392 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
1394 unsigned long flags
;
1395 struct sk_buff
*result
;
1397 spin_lock_irqsave(&list
->lock
, flags
);
1398 result
= __skb_dequeue_tail(list
);
1399 spin_unlock_irqrestore(&list
->lock
, flags
);
1404 * skb_queue_purge - empty a list
1405 * @list: list to empty
1407 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1408 * the list and one reference dropped. This function takes the list
1409 * lock and is atomic with respect to other list locking functions.
1411 void skb_queue_purge(struct sk_buff_head
*list
)
1413 struct sk_buff
*skb
;
1414 while ((skb
= skb_dequeue(list
)) != NULL
)
1419 * skb_queue_head - queue a buffer at the list head
1420 * @list: list to use
1421 * @newsk: buffer to queue
1423 * Queue a buffer at the start of the list. This function takes the
1424 * list lock and can be used safely with other locking &sk_buff functions
1427 * A buffer cannot be placed on two lists at the same time.
1429 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1431 unsigned long flags
;
1433 spin_lock_irqsave(&list
->lock
, flags
);
1434 __skb_queue_head(list
, newsk
);
1435 spin_unlock_irqrestore(&list
->lock
, flags
);
1439 * skb_queue_tail - queue a buffer at the list tail
1440 * @list: list to use
1441 * @newsk: buffer to queue
1443 * Queue a buffer at the tail of the list. This function takes the
1444 * list lock and can be used safely with other locking &sk_buff functions
1447 * A buffer cannot be placed on two lists at the same time.
1449 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1451 unsigned long flags
;
1453 spin_lock_irqsave(&list
->lock
, flags
);
1454 __skb_queue_tail(list
, newsk
);
1455 spin_unlock_irqrestore(&list
->lock
, flags
);
1459 * skb_unlink - remove a buffer from a list
1460 * @skb: buffer to remove
1461 * @list: list to use
1463 * Remove a packet from a list. The list locks are taken and this
1464 * function is atomic with respect to other list locked calls
1466 * You must know what list the SKB is on.
1468 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
1470 unsigned long flags
;
1472 spin_lock_irqsave(&list
->lock
, flags
);
1473 __skb_unlink(skb
, list
);
1474 spin_unlock_irqrestore(&list
->lock
, flags
);
1478 * skb_append - append a buffer
1479 * @old: buffer to insert after
1480 * @newsk: buffer to insert
1481 * @list: list to use
1483 * Place a packet after a given packet in a list. The list locks are taken
1484 * and this function is atomic with respect to other list locked calls.
1485 * A buffer cannot be placed on two lists at the same time.
1487 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
1489 unsigned long flags
;
1491 spin_lock_irqsave(&list
->lock
, flags
);
1492 __skb_append(old
, newsk
, list
);
1493 spin_unlock_irqrestore(&list
->lock
, flags
);
1498 * skb_insert - insert a buffer
1499 * @old: buffer to insert before
1500 * @newsk: buffer to insert
1501 * @list: list to use
1503 * Place a packet before a given packet in a list. The list locks are
1504 * taken and this function is atomic with respect to other list locked
1507 * A buffer cannot be placed on two lists at the same time.
1509 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
1511 unsigned long flags
;
1513 spin_lock_irqsave(&list
->lock
, flags
);
1514 __skb_insert(newsk
, old
->prev
, old
, list
);
1515 spin_unlock_irqrestore(&list
->lock
, flags
);
1520 * Tune the memory allocator for a new MTU size.
1522 void skb_add_mtu(int mtu
)
1524 /* Must match allocation in alloc_skb */
1525 mtu
= SKB_DATA_ALIGN(mtu
) + sizeof(struct skb_shared_info
);
1527 kmem_add_cache_size(mtu
);
1531 static inline void skb_split_inside_header(struct sk_buff
*skb
,
1532 struct sk_buff
* skb1
,
1533 const u32 len
, const int pos
)
1537 memcpy(skb_put(skb1
, pos
- len
), skb
->data
+ len
, pos
- len
);
1539 /* And move data appendix as is. */
1540 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1541 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
1543 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
1544 skb_shinfo(skb
)->nr_frags
= 0;
1545 skb1
->data_len
= skb
->data_len
;
1546 skb1
->len
+= skb1
->data_len
;
1549 skb_set_tail_pointer(skb
, len
);
1552 static inline void skb_split_no_header(struct sk_buff
*skb
,
1553 struct sk_buff
* skb1
,
1554 const u32 len
, int pos
)
1557 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
1559 skb_shinfo(skb
)->nr_frags
= 0;
1560 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
1562 skb
->data_len
= len
- pos
;
1564 for (i
= 0; i
< nfrags
; i
++) {
1565 int size
= skb_shinfo(skb
)->frags
[i
].size
;
1567 if (pos
+ size
> len
) {
1568 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1572 * We have two variants in this case:
1573 * 1. Move all the frag to the second
1574 * part, if it is possible. F.e.
1575 * this approach is mandatory for TUX,
1576 * where splitting is expensive.
1577 * 2. Split is accurately. We make this.
1579 get_page(skb_shinfo(skb
)->frags
[i
].page
);
1580 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
1581 skb_shinfo(skb1
)->frags
[0].size
-= len
- pos
;
1582 skb_shinfo(skb
)->frags
[i
].size
= len
- pos
;
1583 skb_shinfo(skb
)->nr_frags
++;
1587 skb_shinfo(skb
)->nr_frags
++;
1590 skb_shinfo(skb1
)->nr_frags
= k
;
1594 * skb_split - Split fragmented skb to two parts at length len.
1595 * @skb: the buffer to split
1596 * @skb1: the buffer to receive the second part
1597 * @len: new length for skb
1599 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
1601 int pos
= skb_headlen(skb
);
1603 if (len
< pos
) /* Split line is inside header. */
1604 skb_split_inside_header(skb
, skb1
, len
, pos
);
1605 else /* Second chunk has no header, nothing to copy. */
1606 skb_split_no_header(skb
, skb1
, len
, pos
);
1610 * skb_prepare_seq_read - Prepare a sequential read of skb data
1611 * @skb: the buffer to read
1612 * @from: lower offset of data to be read
1613 * @to: upper offset of data to be read
1614 * @st: state variable
1616 * Initializes the specified state variable. Must be called before
1617 * invoking skb_seq_read() for the first time.
1619 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
1620 unsigned int to
, struct skb_seq_state
*st
)
1622 st
->lower_offset
= from
;
1623 st
->upper_offset
= to
;
1624 st
->root_skb
= st
->cur_skb
= skb
;
1625 st
->frag_idx
= st
->stepped_offset
= 0;
1626 st
->frag_data
= NULL
;
1630 * skb_seq_read - Sequentially read skb data
1631 * @consumed: number of bytes consumed by the caller so far
1632 * @data: destination pointer for data to be returned
1633 * @st: state variable
1635 * Reads a block of skb data at &consumed relative to the
1636 * lower offset specified to skb_prepare_seq_read(). Assigns
1637 * the head of the data block to &data and returns the length
1638 * of the block or 0 if the end of the skb data or the upper
1639 * offset has been reached.
1641 * The caller is not required to consume all of the data
1642 * returned, i.e. &consumed is typically set to the number
1643 * of bytes already consumed and the next call to
1644 * skb_seq_read() will return the remaining part of the block.
1646 * Note: The size of each block of data returned can be arbitary,
1647 * this limitation is the cost for zerocopy seqeuental
1648 * reads of potentially non linear data.
1650 * Note: Fragment lists within fragments are not implemented
1651 * at the moment, state->root_skb could be replaced with
1652 * a stack for this purpose.
1654 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
1655 struct skb_seq_state
*st
)
1657 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
1660 if (unlikely(abs_offset
>= st
->upper_offset
))
1664 block_limit
= skb_headlen(st
->cur_skb
);
1666 if (abs_offset
< block_limit
) {
1667 *data
= st
->cur_skb
->data
+ abs_offset
;
1668 return block_limit
- abs_offset
;
1671 if (st
->frag_idx
== 0 && !st
->frag_data
)
1672 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
1674 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
1675 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
1676 block_limit
= frag
->size
+ st
->stepped_offset
;
1678 if (abs_offset
< block_limit
) {
1680 st
->frag_data
= kmap_skb_frag(frag
);
1682 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
1683 (abs_offset
- st
->stepped_offset
);
1685 return block_limit
- abs_offset
;
1688 if (st
->frag_data
) {
1689 kunmap_skb_frag(st
->frag_data
);
1690 st
->frag_data
= NULL
;
1694 st
->stepped_offset
+= frag
->size
;
1697 if (st
->cur_skb
->next
) {
1698 st
->cur_skb
= st
->cur_skb
->next
;
1701 } else if (st
->root_skb
== st
->cur_skb
&&
1702 skb_shinfo(st
->root_skb
)->frag_list
) {
1703 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
1711 * skb_abort_seq_read - Abort a sequential read of skb data
1712 * @st: state variable
1714 * Must be called if skb_seq_read() was not called until it
1717 void skb_abort_seq_read(struct skb_seq_state
*st
)
1720 kunmap_skb_frag(st
->frag_data
);
1723 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
1725 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
1726 struct ts_config
*conf
,
1727 struct ts_state
*state
)
1729 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
1732 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
1734 skb_abort_seq_read(TS_SKB_CB(state
));
1738 * skb_find_text - Find a text pattern in skb data
1739 * @skb: the buffer to look in
1740 * @from: search offset
1742 * @config: textsearch configuration
1743 * @state: uninitialized textsearch state variable
1745 * Finds a pattern in the skb data according to the specified
1746 * textsearch configuration. Use textsearch_next() to retrieve
1747 * subsequent occurrences of the pattern. Returns the offset
1748 * to the first occurrence or UINT_MAX if no match was found.
1750 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
1751 unsigned int to
, struct ts_config
*config
,
1752 struct ts_state
*state
)
1756 config
->get_next_block
= skb_ts_get_next_block
;
1757 config
->finish
= skb_ts_finish
;
1759 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
1761 ret
= textsearch_find(config
, state
);
1762 return (ret
<= to
- from
? ret
: UINT_MAX
);
1766 * skb_append_datato_frags: - append the user data to a skb
1767 * @sk: sock structure
1768 * @skb: skb structure to be appened with user data.
1769 * @getfrag: call back function to be used for getting the user data
1770 * @from: pointer to user message iov
1771 * @length: length of the iov message
1773 * Description: This procedure append the user data in the fragment part
1774 * of the skb if any page alloc fails user this procedure returns -ENOMEM
1776 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
1777 int (*getfrag
)(void *from
, char *to
, int offset
,
1778 int len
, int odd
, struct sk_buff
*skb
),
1779 void *from
, int length
)
1782 skb_frag_t
*frag
= NULL
;
1783 struct page
*page
= NULL
;
1789 /* Return error if we don't have space for new frag */
1790 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
1791 if (frg_cnt
>= MAX_SKB_FRAGS
)
1794 /* allocate a new page for next frag */
1795 page
= alloc_pages(sk
->sk_allocation
, 0);
1797 /* If alloc_page fails just return failure and caller will
1798 * free previous allocated pages by doing kfree_skb()
1803 /* initialize the next frag */
1804 sk
->sk_sndmsg_page
= page
;
1805 sk
->sk_sndmsg_off
= 0;
1806 skb_fill_page_desc(skb
, frg_cnt
, page
, 0, 0);
1807 skb
->truesize
+= PAGE_SIZE
;
1808 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
1810 /* get the new initialized frag */
1811 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
1812 frag
= &skb_shinfo(skb
)->frags
[frg_cnt
- 1];
1814 /* copy the user data to page */
1815 left
= PAGE_SIZE
- frag
->page_offset
;
1816 copy
= (length
> left
)? left
: length
;
1818 ret
= getfrag(from
, (page_address(frag
->page
) +
1819 frag
->page_offset
+ frag
->size
),
1820 offset
, copy
, 0, skb
);
1824 /* copy was successful so update the size parameters */
1825 sk
->sk_sndmsg_off
+= copy
;
1828 skb
->data_len
+= copy
;
1832 } while (length
> 0);
1838 * skb_pull_rcsum - pull skb and update receive checksum
1839 * @skb: buffer to update
1840 * @start: start of data before pull
1841 * @len: length of data pulled
1843 * This function performs an skb_pull on the packet and updates
1844 * update the CHECKSUM_COMPLETE checksum. It should be used on
1845 * receive path processing instead of skb_pull unless you know
1846 * that the checksum difference is zero (e.g., a valid IP header)
1847 * or you are setting ip_summed to CHECKSUM_NONE.
1849 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
1851 BUG_ON(len
> skb
->len
);
1853 BUG_ON(skb
->len
< skb
->data_len
);
1854 skb_postpull_rcsum(skb
, skb
->data
, len
);
1855 return skb
->data
+= len
;
1858 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
1861 * skb_segment - Perform protocol segmentation on skb.
1862 * @skb: buffer to segment
1863 * @features: features for the output path (see dev->features)
1865 * This function performs segmentation on the given skb. It returns
1866 * the segment at the given position. It returns NULL if there are
1867 * no more segments to generate, or when an error is encountered.
1869 struct sk_buff
*skb_segment(struct sk_buff
*skb
, int features
)
1871 struct sk_buff
*segs
= NULL
;
1872 struct sk_buff
*tail
= NULL
;
1873 unsigned int mss
= skb_shinfo(skb
)->gso_size
;
1874 unsigned int doffset
= skb
->data
- skb_mac_header(skb
);
1875 unsigned int offset
= doffset
;
1876 unsigned int headroom
;
1878 int sg
= features
& NETIF_F_SG
;
1879 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1884 __skb_push(skb
, doffset
);
1885 headroom
= skb_headroom(skb
);
1886 pos
= skb_headlen(skb
);
1889 struct sk_buff
*nskb
;
1895 len
= skb
->len
- offset
;
1899 hsize
= skb_headlen(skb
) - offset
;
1902 if (hsize
> len
|| !sg
)
1905 nskb
= alloc_skb(hsize
+ doffset
+ headroom
, GFP_ATOMIC
);
1906 if (unlikely(!nskb
))
1915 nskb
->dev
= skb
->dev
;
1916 nskb
->priority
= skb
->priority
;
1917 nskb
->protocol
= skb
->protocol
;
1918 nskb
->dst
= dst_clone(skb
->dst
);
1919 memcpy(nskb
->cb
, skb
->cb
, sizeof(skb
->cb
));
1920 nskb
->pkt_type
= skb
->pkt_type
;
1921 nskb
->mac_len
= skb
->mac_len
;
1923 skb_reserve(nskb
, headroom
);
1924 skb_reset_mac_header(nskb
);
1925 skb_set_network_header(nskb
, skb
->mac_len
);
1926 nskb
->transport_header
= (nskb
->network_header
+
1927 skb_network_header_len(skb
));
1928 memcpy(skb_put(nskb
, doffset
), skb
->data
, doffset
);
1931 nskb
->csum
= skb_copy_and_csum_bits(skb
, offset
,
1937 frag
= skb_shinfo(nskb
)->frags
;
1940 nskb
->ip_summed
= CHECKSUM_PARTIAL
;
1941 nskb
->csum
= skb
->csum
;
1942 memcpy(skb_put(nskb
, hsize
), skb
->data
+ offset
, hsize
);
1944 while (pos
< offset
+ len
) {
1945 BUG_ON(i
>= nfrags
);
1947 *frag
= skb_shinfo(skb
)->frags
[i
];
1948 get_page(frag
->page
);
1952 frag
->page_offset
+= offset
- pos
;
1953 frag
->size
-= offset
- pos
;
1958 if (pos
+ size
<= offset
+ len
) {
1962 frag
->size
-= pos
+ size
- (offset
+ len
);
1969 skb_shinfo(nskb
)->nr_frags
= k
;
1970 nskb
->data_len
= len
- hsize
;
1971 nskb
->len
+= nskb
->data_len
;
1972 nskb
->truesize
+= nskb
->data_len
;
1973 } while ((offset
+= len
) < skb
->len
);
1978 while ((skb
= segs
)) {
1982 return ERR_PTR(err
);
1985 EXPORT_SYMBOL_GPL(skb_segment
);
1987 void __init
skb_init(void)
1989 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
1990 sizeof(struct sk_buff
),
1992 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
1994 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
1995 (2*sizeof(struct sk_buff
)) +
1998 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2002 EXPORT_SYMBOL(___pskb_trim
);
2003 EXPORT_SYMBOL(__kfree_skb
);
2004 EXPORT_SYMBOL(kfree_skb
);
2005 EXPORT_SYMBOL(__pskb_pull_tail
);
2006 EXPORT_SYMBOL(__alloc_skb
);
2007 EXPORT_SYMBOL(__netdev_alloc_skb
);
2008 EXPORT_SYMBOL(pskb_copy
);
2009 EXPORT_SYMBOL(pskb_expand_head
);
2010 EXPORT_SYMBOL(skb_checksum
);
2011 EXPORT_SYMBOL(skb_clone
);
2012 EXPORT_SYMBOL(skb_clone_fraglist
);
2013 EXPORT_SYMBOL(skb_copy
);
2014 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
2015 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
2016 EXPORT_SYMBOL(skb_copy_bits
);
2017 EXPORT_SYMBOL(skb_copy_expand
);
2018 EXPORT_SYMBOL(skb_over_panic
);
2019 EXPORT_SYMBOL(skb_pad
);
2020 EXPORT_SYMBOL(skb_realloc_headroom
);
2021 EXPORT_SYMBOL(skb_under_panic
);
2022 EXPORT_SYMBOL(skb_dequeue
);
2023 EXPORT_SYMBOL(skb_dequeue_tail
);
2024 EXPORT_SYMBOL(skb_insert
);
2025 EXPORT_SYMBOL(skb_queue_purge
);
2026 EXPORT_SYMBOL(skb_queue_head
);
2027 EXPORT_SYMBOL(skb_queue_tail
);
2028 EXPORT_SYMBOL(skb_unlink
);
2029 EXPORT_SYMBOL(skb_append
);
2030 EXPORT_SYMBOL(skb_split
);
2031 EXPORT_SYMBOL(skb_prepare_seq_read
);
2032 EXPORT_SYMBOL(skb_seq_read
);
2033 EXPORT_SYMBOL(skb_abort_seq_read
);
2034 EXPORT_SYMBOL(skb_find_text
);
2035 EXPORT_SYMBOL(skb_append_datato_frags
);