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:%#lx dev:%s\n",
91 here
, skb
->len
, sz
, skb
->head
, skb
->data
,
92 (unsigned long)skb
->tail
, (unsigned long)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:%#lx dev:%s\n",
110 here
, skb
->len
, sz
, skb
->head
, skb
->data
,
111 (unsigned long)skb
->tail
, (unsigned long)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
= skb
->tail
+ 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
524 #ifdef NET_SKBUFF_DATA_USES_OFFSET
525 n
= alloc_skb(skb
->end
+ skb
->data_len
, gfp_mask
);
527 n
= alloc_skb(skb
->end
- skb
->head
+ skb
->data_len
, gfp_mask
);
532 /* Set the data pointer */
533 skb_reserve(n
, headerlen
);
534 /* Set the tail pointer and length */
535 skb_put(n
, skb
->len
);
537 n
->ip_summed
= skb
->ip_summed
;
539 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
542 copy_skb_header(n
, skb
);
548 * pskb_copy - create copy of an sk_buff with private head.
549 * @skb: buffer to copy
550 * @gfp_mask: allocation priority
552 * Make a copy of both an &sk_buff and part of its data, located
553 * in header. Fragmented data remain shared. This is used when
554 * the caller wishes to modify only header of &sk_buff and needs
555 * private copy of the header to alter. Returns %NULL on failure
556 * or the pointer to the buffer on success.
557 * The returned buffer has a reference count of 1.
560 struct sk_buff
*pskb_copy(struct sk_buff
*skb
, gfp_t gfp_mask
)
563 * Allocate the copy buffer
566 #ifdef NET_SKBUFF_DATA_USES_OFFSET
567 n
= alloc_skb(skb
->end
, gfp_mask
);
569 n
= alloc_skb(skb
->end
- skb
->head
, gfp_mask
);
574 /* Set the data pointer */
575 skb_reserve(n
, skb
->data
- skb
->head
);
576 /* Set the tail pointer and length */
577 skb_put(n
, skb_headlen(skb
));
579 memcpy(n
->data
, skb
->data
, n
->len
);
581 n
->ip_summed
= skb
->ip_summed
;
583 n
->truesize
+= skb
->data_len
;
584 n
->data_len
= skb
->data_len
;
587 if (skb_shinfo(skb
)->nr_frags
) {
590 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
591 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
592 get_page(skb_shinfo(n
)->frags
[i
].page
);
594 skb_shinfo(n
)->nr_frags
= i
;
597 if (skb_shinfo(skb
)->frag_list
) {
598 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
599 skb_clone_fraglist(n
);
602 copy_skb_header(n
, skb
);
608 * pskb_expand_head - reallocate header of &sk_buff
609 * @skb: buffer to reallocate
610 * @nhead: room to add at head
611 * @ntail: room to add at tail
612 * @gfp_mask: allocation priority
614 * Expands (or creates identical copy, if &nhead and &ntail are zero)
615 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
616 * reference count of 1. Returns zero in the case of success or error,
617 * if expansion failed. In the last case, &sk_buff is not changed.
619 * All the pointers pointing into skb header may change and must be
620 * reloaded after call to this function.
623 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
628 #ifdef NET_SKBUFF_DATA_USES_OFFSET
629 int size
= nhead
+ skb
->end
+ ntail
;
631 int size
= nhead
+ (skb
->end
- skb
->head
) + ntail
;
638 size
= SKB_DATA_ALIGN(size
);
640 data
= kmalloc(size
+ sizeof(struct skb_shared_info
), gfp_mask
);
644 /* Copy only real data... and, alas, header. This should be
645 * optimized for the cases when header is void. */
646 memcpy(data
+ nhead
, skb
->head
,
647 #ifdef NET_SKBUFF_DATA_USES_OFFSET
650 skb
->tail
- skb
->head
);
652 memcpy(data
+ size
, skb_end_pointer(skb
),
653 sizeof(struct skb_shared_info
));
655 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
656 get_page(skb_shinfo(skb
)->frags
[i
].page
);
658 if (skb_shinfo(skb
)->frag_list
)
659 skb_clone_fraglist(skb
);
661 skb_release_data(skb
);
663 off
= (data
+ nhead
) - skb
->head
;
667 #ifdef NET_SKBUFF_DATA_USES_OFFSET
670 skb
->end
= skb
->head
+ size
;
671 /* {transport,network,mac}_header and tail are relative to skb->head */
673 skb
->transport_header
+= off
;
674 skb
->network_header
+= off
;
675 skb
->mac_header
+= off
;
679 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
686 /* Make private copy of skb with writable head and some headroom */
688 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
690 struct sk_buff
*skb2
;
691 int delta
= headroom
- skb_headroom(skb
);
694 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
696 skb2
= skb_clone(skb
, GFP_ATOMIC
);
697 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
708 * skb_copy_expand - copy and expand sk_buff
709 * @skb: buffer to copy
710 * @newheadroom: new free bytes at head
711 * @newtailroom: new free bytes at tail
712 * @gfp_mask: allocation priority
714 * Make a copy of both an &sk_buff and its data and while doing so
715 * allocate additional space.
717 * This is used when the caller wishes to modify the data and needs a
718 * private copy of the data to alter as well as more space for new fields.
719 * Returns %NULL on failure or the pointer to the buffer
720 * on success. The returned buffer has a reference count of 1.
722 * You must pass %GFP_ATOMIC as the allocation priority if this function
723 * is called from an interrupt.
725 * BUG ALERT: ip_summed is not copied. Why does this work? Is it used
726 * only by netfilter in the cases when checksum is recalculated? --ANK
728 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
729 int newheadroom
, int newtailroom
,
733 * Allocate the copy buffer
735 struct sk_buff
*n
= alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
737 int head_copy_len
, head_copy_off
;
742 skb_reserve(n
, newheadroom
);
744 /* Set the tail pointer and length */
745 skb_put(n
, skb
->len
);
747 head_copy_len
= skb_headroom(skb
);
749 if (newheadroom
<= head_copy_len
)
750 head_copy_len
= newheadroom
;
752 head_copy_off
= newheadroom
- head_copy_len
;
754 /* Copy the linear header and data. */
755 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
756 skb
->len
+ head_copy_len
))
759 copy_skb_header(n
, skb
);
765 * skb_pad - zero pad the tail of an skb
766 * @skb: buffer to pad
769 * Ensure that a buffer is followed by a padding area that is zero
770 * filled. Used by network drivers which may DMA or transfer data
771 * beyond the buffer end onto the wire.
773 * May return error in out of memory cases. The skb is freed on error.
776 int skb_pad(struct sk_buff
*skb
, int pad
)
781 /* If the skbuff is non linear tailroom is always zero.. */
782 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
783 memset(skb
->data
+skb
->len
, 0, pad
);
787 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
788 if (likely(skb_cloned(skb
) || ntail
> 0)) {
789 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
794 /* FIXME: The use of this function with non-linear skb's really needs
797 err
= skb_linearize(skb
);
801 memset(skb
->data
+ skb
->len
, 0, pad
);
809 /* Trims skb to length len. It can change skb pointers.
812 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
814 struct sk_buff
**fragp
;
815 struct sk_buff
*frag
;
816 int offset
= skb_headlen(skb
);
817 int nfrags
= skb_shinfo(skb
)->nr_frags
;
821 if (skb_cloned(skb
) &&
822 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
829 for (; i
< nfrags
; i
++) {
830 int end
= offset
+ skb_shinfo(skb
)->frags
[i
].size
;
837 skb_shinfo(skb
)->frags
[i
++].size
= len
- offset
;
840 skb_shinfo(skb
)->nr_frags
= i
;
842 for (; i
< nfrags
; i
++)
843 put_page(skb_shinfo(skb
)->frags
[i
].page
);
845 if (skb_shinfo(skb
)->frag_list
)
846 skb_drop_fraglist(skb
);
850 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
851 fragp
= &frag
->next
) {
852 int end
= offset
+ frag
->len
;
854 if (skb_shared(frag
)) {
855 struct sk_buff
*nfrag
;
857 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
858 if (unlikely(!nfrag
))
861 nfrag
->next
= frag
->next
;
873 unlikely((err
= pskb_trim(frag
, len
- offset
))))
877 skb_drop_list(&frag
->next
);
882 if (len
> skb_headlen(skb
)) {
883 skb
->data_len
-= skb
->len
- len
;
888 skb_set_tail_pointer(skb
, len
);
895 * __pskb_pull_tail - advance tail of skb header
896 * @skb: buffer to reallocate
897 * @delta: number of bytes to advance tail
899 * The function makes a sense only on a fragmented &sk_buff,
900 * it expands header moving its tail forward and copying necessary
901 * data from fragmented part.
903 * &sk_buff MUST have reference count of 1.
905 * Returns %NULL (and &sk_buff does not change) if pull failed
906 * or value of new tail of skb in the case of success.
908 * All the pointers pointing into skb header may change and must be
909 * reloaded after call to this function.
912 /* Moves tail of skb head forward, copying data from fragmented part,
913 * when it is necessary.
914 * 1. It may fail due to malloc failure.
915 * 2. It may change skb pointers.
917 * It is pretty complicated. Luckily, it is called only in exceptional cases.
919 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
921 /* If skb has not enough free space at tail, get new one
922 * plus 128 bytes for future expansions. If we have enough
923 * room at tail, reallocate without expansion only if skb is cloned.
925 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
927 if (eat
> 0 || skb_cloned(skb
)) {
928 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
933 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
936 /* Optimization: no fragments, no reasons to preestimate
937 * size of pulled pages. Superb.
939 if (!skb_shinfo(skb
)->frag_list
)
942 /* Estimate size of pulled pages. */
944 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
945 if (skb_shinfo(skb
)->frags
[i
].size
>= eat
)
947 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
950 /* If we need update frag list, we are in troubles.
951 * Certainly, it possible to add an offset to skb data,
952 * but taking into account that pulling is expected to
953 * be very rare operation, it is worth to fight against
954 * further bloating skb head and crucify ourselves here instead.
955 * Pure masohism, indeed. 8)8)
958 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
959 struct sk_buff
*clone
= NULL
;
960 struct sk_buff
*insp
= NULL
;
965 if (list
->len
<= eat
) {
966 /* Eaten as whole. */
971 /* Eaten partially. */
973 if (skb_shared(list
)) {
974 /* Sucks! We need to fork list. :-( */
975 clone
= skb_clone(list
, GFP_ATOMIC
);
981 /* This may be pulled without
985 if (!pskb_pull(list
, eat
)) {
994 /* Free pulled out fragments. */
995 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
996 skb_shinfo(skb
)->frag_list
= list
->next
;
999 /* And insert new clone at head. */
1002 skb_shinfo(skb
)->frag_list
= clone
;
1005 /* Success! Now we may commit changes to skb data. */
1010 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1011 if (skb_shinfo(skb
)->frags
[i
].size
<= eat
) {
1012 put_page(skb_shinfo(skb
)->frags
[i
].page
);
1013 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1015 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1017 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1018 skb_shinfo(skb
)->frags
[k
].size
-= eat
;
1024 skb_shinfo(skb
)->nr_frags
= k
;
1027 skb
->data_len
-= delta
;
1029 return skb_tail_pointer(skb
);
1032 /* Copy some data bits from skb to kernel buffer. */
1034 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1037 int start
= skb_headlen(skb
);
1039 if (offset
> (int)skb
->len
- len
)
1043 if ((copy
= start
- offset
) > 0) {
1046 memcpy(to
, skb
->data
+ offset
, copy
);
1047 if ((len
-= copy
) == 0)
1053 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1056 BUG_TRAP(start
<= offset
+ len
);
1058 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1059 if ((copy
= end
- offset
) > 0) {
1065 vaddr
= kmap_skb_frag(&skb_shinfo(skb
)->frags
[i
]);
1067 vaddr
+ skb_shinfo(skb
)->frags
[i
].page_offset
+
1068 offset
- start
, copy
);
1069 kunmap_skb_frag(vaddr
);
1071 if ((len
-= copy
) == 0)
1079 if (skb_shinfo(skb
)->frag_list
) {
1080 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1082 for (; list
; list
= list
->next
) {
1085 BUG_TRAP(start
<= offset
+ len
);
1087 end
= start
+ list
->len
;
1088 if ((copy
= end
- offset
) > 0) {
1091 if (skb_copy_bits(list
, offset
- start
,
1094 if ((len
-= copy
) == 0)
1110 * skb_store_bits - store bits from kernel buffer to skb
1111 * @skb: destination buffer
1112 * @offset: offset in destination
1113 * @from: source buffer
1114 * @len: number of bytes to copy
1116 * Copy the specified number of bytes from the source buffer to the
1117 * destination skb. This function handles all the messy bits of
1118 * traversing fragment lists and such.
1121 int skb_store_bits(const struct sk_buff
*skb
, int offset
, void *from
, int len
)
1124 int start
= skb_headlen(skb
);
1126 if (offset
> (int)skb
->len
- len
)
1129 if ((copy
= start
- offset
) > 0) {
1132 memcpy(skb
->data
+ offset
, from
, copy
);
1133 if ((len
-= copy
) == 0)
1139 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1140 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1143 BUG_TRAP(start
<= offset
+ len
);
1145 end
= start
+ frag
->size
;
1146 if ((copy
= end
- offset
) > 0) {
1152 vaddr
= kmap_skb_frag(frag
);
1153 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1155 kunmap_skb_frag(vaddr
);
1157 if ((len
-= copy
) == 0)
1165 if (skb_shinfo(skb
)->frag_list
) {
1166 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1168 for (; list
; list
= list
->next
) {
1171 BUG_TRAP(start
<= offset
+ len
);
1173 end
= start
+ list
->len
;
1174 if ((copy
= end
- offset
) > 0) {
1177 if (skb_store_bits(list
, offset
- start
,
1180 if ((len
-= copy
) == 0)
1195 EXPORT_SYMBOL(skb_store_bits
);
1197 /* Checksum skb data. */
1199 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
1200 int len
, __wsum csum
)
1202 int start
= skb_headlen(skb
);
1203 int i
, copy
= start
- offset
;
1206 /* Checksum header. */
1210 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1211 if ((len
-= copy
) == 0)
1217 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1220 BUG_TRAP(start
<= offset
+ len
);
1222 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1223 if ((copy
= end
- offset
) > 0) {
1226 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1230 vaddr
= kmap_skb_frag(frag
);
1231 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
1232 offset
- start
, copy
, 0);
1233 kunmap_skb_frag(vaddr
);
1234 csum
= csum_block_add(csum
, csum2
, pos
);
1243 if (skb_shinfo(skb
)->frag_list
) {
1244 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1246 for (; list
; list
= list
->next
) {
1249 BUG_TRAP(start
<= offset
+ len
);
1251 end
= start
+ list
->len
;
1252 if ((copy
= end
- offset
) > 0) {
1256 csum2
= skb_checksum(list
, offset
- start
,
1258 csum
= csum_block_add(csum
, csum2
, pos
);
1259 if ((len
-= copy
) == 0)
1272 /* Both of above in one bottle. */
1274 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
1275 u8
*to
, int len
, __wsum csum
)
1277 int start
= skb_headlen(skb
);
1278 int i
, copy
= start
- offset
;
1285 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
1287 if ((len
-= copy
) == 0)
1294 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1297 BUG_TRAP(start
<= offset
+ len
);
1299 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1300 if ((copy
= end
- offset
) > 0) {
1303 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1307 vaddr
= kmap_skb_frag(frag
);
1308 csum2
= csum_partial_copy_nocheck(vaddr
+
1312 kunmap_skb_frag(vaddr
);
1313 csum
= csum_block_add(csum
, csum2
, pos
);
1323 if (skb_shinfo(skb
)->frag_list
) {
1324 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1326 for (; list
; list
= list
->next
) {
1330 BUG_TRAP(start
<= offset
+ len
);
1332 end
= start
+ list
->len
;
1333 if ((copy
= end
- offset
) > 0) {
1336 csum2
= skb_copy_and_csum_bits(list
,
1339 csum
= csum_block_add(csum
, csum2
, pos
);
1340 if ((len
-= copy
) == 0)
1353 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
1358 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1359 csstart
= skb_transport_offset(skb
);
1361 csstart
= skb_headlen(skb
);
1363 BUG_ON(csstart
> skb_headlen(skb
));
1365 memcpy(to
, skb
->data
, csstart
);
1368 if (csstart
!= skb
->len
)
1369 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
1370 skb
->len
- csstart
, 0);
1372 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1373 long csstuff
= csstart
+ skb
->csum_offset
;
1375 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
1380 * skb_dequeue - remove from the head of the queue
1381 * @list: list to dequeue from
1383 * Remove the head of the list. The list lock is taken so the function
1384 * may be used safely with other locking list functions. The head item is
1385 * returned or %NULL if the list is empty.
1388 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
1390 unsigned long flags
;
1391 struct sk_buff
*result
;
1393 spin_lock_irqsave(&list
->lock
, flags
);
1394 result
= __skb_dequeue(list
);
1395 spin_unlock_irqrestore(&list
->lock
, flags
);
1400 * skb_dequeue_tail - remove from the tail of the queue
1401 * @list: list to dequeue from
1403 * Remove the tail of the list. The list lock is taken so the function
1404 * may be used safely with other locking list functions. The tail item is
1405 * returned or %NULL if the list is empty.
1407 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
1409 unsigned long flags
;
1410 struct sk_buff
*result
;
1412 spin_lock_irqsave(&list
->lock
, flags
);
1413 result
= __skb_dequeue_tail(list
);
1414 spin_unlock_irqrestore(&list
->lock
, flags
);
1419 * skb_queue_purge - empty a list
1420 * @list: list to empty
1422 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1423 * the list and one reference dropped. This function takes the list
1424 * lock and is atomic with respect to other list locking functions.
1426 void skb_queue_purge(struct sk_buff_head
*list
)
1428 struct sk_buff
*skb
;
1429 while ((skb
= skb_dequeue(list
)) != NULL
)
1434 * skb_queue_head - queue a buffer at the list head
1435 * @list: list to use
1436 * @newsk: buffer to queue
1438 * Queue a buffer at the start of the list. This function takes the
1439 * list lock and can be used safely with other locking &sk_buff functions
1442 * A buffer cannot be placed on two lists at the same time.
1444 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1446 unsigned long flags
;
1448 spin_lock_irqsave(&list
->lock
, flags
);
1449 __skb_queue_head(list
, newsk
);
1450 spin_unlock_irqrestore(&list
->lock
, flags
);
1454 * skb_queue_tail - queue a buffer at the list tail
1455 * @list: list to use
1456 * @newsk: buffer to queue
1458 * Queue a buffer at the tail of the list. This function takes the
1459 * list lock and can be used safely with other locking &sk_buff functions
1462 * A buffer cannot be placed on two lists at the same time.
1464 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1466 unsigned long flags
;
1468 spin_lock_irqsave(&list
->lock
, flags
);
1469 __skb_queue_tail(list
, newsk
);
1470 spin_unlock_irqrestore(&list
->lock
, flags
);
1474 * skb_unlink - remove a buffer from a list
1475 * @skb: buffer to remove
1476 * @list: list to use
1478 * Remove a packet from a list. The list locks are taken and this
1479 * function is atomic with respect to other list locked calls
1481 * You must know what list the SKB is on.
1483 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
1485 unsigned long flags
;
1487 spin_lock_irqsave(&list
->lock
, flags
);
1488 __skb_unlink(skb
, list
);
1489 spin_unlock_irqrestore(&list
->lock
, flags
);
1493 * skb_append - append a buffer
1494 * @old: buffer to insert after
1495 * @newsk: buffer to insert
1496 * @list: list to use
1498 * Place a packet after a given packet in a list. The list locks are taken
1499 * and this function is atomic with respect to other list locked calls.
1500 * A buffer cannot be placed on two lists at the same time.
1502 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
1504 unsigned long flags
;
1506 spin_lock_irqsave(&list
->lock
, flags
);
1507 __skb_append(old
, newsk
, list
);
1508 spin_unlock_irqrestore(&list
->lock
, flags
);
1513 * skb_insert - insert a buffer
1514 * @old: buffer to insert before
1515 * @newsk: buffer to insert
1516 * @list: list to use
1518 * Place a packet before a given packet in a list. The list locks are
1519 * taken and this function is atomic with respect to other list locked
1522 * A buffer cannot be placed on two lists at the same time.
1524 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
1526 unsigned long flags
;
1528 spin_lock_irqsave(&list
->lock
, flags
);
1529 __skb_insert(newsk
, old
->prev
, old
, list
);
1530 spin_unlock_irqrestore(&list
->lock
, flags
);
1535 * Tune the memory allocator for a new MTU size.
1537 void skb_add_mtu(int mtu
)
1539 /* Must match allocation in alloc_skb */
1540 mtu
= SKB_DATA_ALIGN(mtu
) + sizeof(struct skb_shared_info
);
1542 kmem_add_cache_size(mtu
);
1546 static inline void skb_split_inside_header(struct sk_buff
*skb
,
1547 struct sk_buff
* skb1
,
1548 const u32 len
, const int pos
)
1552 memcpy(skb_put(skb1
, pos
- len
), skb
->data
+ len
, pos
- len
);
1554 /* And move data appendix as is. */
1555 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1556 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
1558 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
1559 skb_shinfo(skb
)->nr_frags
= 0;
1560 skb1
->data_len
= skb
->data_len
;
1561 skb1
->len
+= skb1
->data_len
;
1564 skb_set_tail_pointer(skb
, len
);
1567 static inline void skb_split_no_header(struct sk_buff
*skb
,
1568 struct sk_buff
* skb1
,
1569 const u32 len
, int pos
)
1572 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
1574 skb_shinfo(skb
)->nr_frags
= 0;
1575 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
1577 skb
->data_len
= len
- pos
;
1579 for (i
= 0; i
< nfrags
; i
++) {
1580 int size
= skb_shinfo(skb
)->frags
[i
].size
;
1582 if (pos
+ size
> len
) {
1583 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1587 * We have two variants in this case:
1588 * 1. Move all the frag to the second
1589 * part, if it is possible. F.e.
1590 * this approach is mandatory for TUX,
1591 * where splitting is expensive.
1592 * 2. Split is accurately. We make this.
1594 get_page(skb_shinfo(skb
)->frags
[i
].page
);
1595 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
1596 skb_shinfo(skb1
)->frags
[0].size
-= len
- pos
;
1597 skb_shinfo(skb
)->frags
[i
].size
= len
- pos
;
1598 skb_shinfo(skb
)->nr_frags
++;
1602 skb_shinfo(skb
)->nr_frags
++;
1605 skb_shinfo(skb1
)->nr_frags
= k
;
1609 * skb_split - Split fragmented skb to two parts at length len.
1610 * @skb: the buffer to split
1611 * @skb1: the buffer to receive the second part
1612 * @len: new length for skb
1614 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
1616 int pos
= skb_headlen(skb
);
1618 if (len
< pos
) /* Split line is inside header. */
1619 skb_split_inside_header(skb
, skb1
, len
, pos
);
1620 else /* Second chunk has no header, nothing to copy. */
1621 skb_split_no_header(skb
, skb1
, len
, pos
);
1625 * skb_prepare_seq_read - Prepare a sequential read of skb data
1626 * @skb: the buffer to read
1627 * @from: lower offset of data to be read
1628 * @to: upper offset of data to be read
1629 * @st: state variable
1631 * Initializes the specified state variable. Must be called before
1632 * invoking skb_seq_read() for the first time.
1634 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
1635 unsigned int to
, struct skb_seq_state
*st
)
1637 st
->lower_offset
= from
;
1638 st
->upper_offset
= to
;
1639 st
->root_skb
= st
->cur_skb
= skb
;
1640 st
->frag_idx
= st
->stepped_offset
= 0;
1641 st
->frag_data
= NULL
;
1645 * skb_seq_read - Sequentially read skb data
1646 * @consumed: number of bytes consumed by the caller so far
1647 * @data: destination pointer for data to be returned
1648 * @st: state variable
1650 * Reads a block of skb data at &consumed relative to the
1651 * lower offset specified to skb_prepare_seq_read(). Assigns
1652 * the head of the data block to &data and returns the length
1653 * of the block or 0 if the end of the skb data or the upper
1654 * offset has been reached.
1656 * The caller is not required to consume all of the data
1657 * returned, i.e. &consumed is typically set to the number
1658 * of bytes already consumed and the next call to
1659 * skb_seq_read() will return the remaining part of the block.
1661 * Note: The size of each block of data returned can be arbitary,
1662 * this limitation is the cost for zerocopy seqeuental
1663 * reads of potentially non linear data.
1665 * Note: Fragment lists within fragments are not implemented
1666 * at the moment, state->root_skb could be replaced with
1667 * a stack for this purpose.
1669 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
1670 struct skb_seq_state
*st
)
1672 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
1675 if (unlikely(abs_offset
>= st
->upper_offset
))
1679 block_limit
= skb_headlen(st
->cur_skb
);
1681 if (abs_offset
< block_limit
) {
1682 *data
= st
->cur_skb
->data
+ abs_offset
;
1683 return block_limit
- abs_offset
;
1686 if (st
->frag_idx
== 0 && !st
->frag_data
)
1687 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
1689 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
1690 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
1691 block_limit
= frag
->size
+ st
->stepped_offset
;
1693 if (abs_offset
< block_limit
) {
1695 st
->frag_data
= kmap_skb_frag(frag
);
1697 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
1698 (abs_offset
- st
->stepped_offset
);
1700 return block_limit
- abs_offset
;
1703 if (st
->frag_data
) {
1704 kunmap_skb_frag(st
->frag_data
);
1705 st
->frag_data
= NULL
;
1709 st
->stepped_offset
+= frag
->size
;
1712 if (st
->cur_skb
->next
) {
1713 st
->cur_skb
= st
->cur_skb
->next
;
1716 } else if (st
->root_skb
== st
->cur_skb
&&
1717 skb_shinfo(st
->root_skb
)->frag_list
) {
1718 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
1726 * skb_abort_seq_read - Abort a sequential read of skb data
1727 * @st: state variable
1729 * Must be called if skb_seq_read() was not called until it
1732 void skb_abort_seq_read(struct skb_seq_state
*st
)
1735 kunmap_skb_frag(st
->frag_data
);
1738 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
1740 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
1741 struct ts_config
*conf
,
1742 struct ts_state
*state
)
1744 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
1747 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
1749 skb_abort_seq_read(TS_SKB_CB(state
));
1753 * skb_find_text - Find a text pattern in skb data
1754 * @skb: the buffer to look in
1755 * @from: search offset
1757 * @config: textsearch configuration
1758 * @state: uninitialized textsearch state variable
1760 * Finds a pattern in the skb data according to the specified
1761 * textsearch configuration. Use textsearch_next() to retrieve
1762 * subsequent occurrences of the pattern. Returns the offset
1763 * to the first occurrence or UINT_MAX if no match was found.
1765 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
1766 unsigned int to
, struct ts_config
*config
,
1767 struct ts_state
*state
)
1771 config
->get_next_block
= skb_ts_get_next_block
;
1772 config
->finish
= skb_ts_finish
;
1774 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
1776 ret
= textsearch_find(config
, state
);
1777 return (ret
<= to
- from
? ret
: UINT_MAX
);
1781 * skb_append_datato_frags: - append the user data to a skb
1782 * @sk: sock structure
1783 * @skb: skb structure to be appened with user data.
1784 * @getfrag: call back function to be used for getting the user data
1785 * @from: pointer to user message iov
1786 * @length: length of the iov message
1788 * Description: This procedure append the user data in the fragment part
1789 * of the skb if any page alloc fails user this procedure returns -ENOMEM
1791 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
1792 int (*getfrag
)(void *from
, char *to
, int offset
,
1793 int len
, int odd
, struct sk_buff
*skb
),
1794 void *from
, int length
)
1797 skb_frag_t
*frag
= NULL
;
1798 struct page
*page
= NULL
;
1804 /* Return error if we don't have space for new frag */
1805 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
1806 if (frg_cnt
>= MAX_SKB_FRAGS
)
1809 /* allocate a new page for next frag */
1810 page
= alloc_pages(sk
->sk_allocation
, 0);
1812 /* If alloc_page fails just return failure and caller will
1813 * free previous allocated pages by doing kfree_skb()
1818 /* initialize the next frag */
1819 sk
->sk_sndmsg_page
= page
;
1820 sk
->sk_sndmsg_off
= 0;
1821 skb_fill_page_desc(skb
, frg_cnt
, page
, 0, 0);
1822 skb
->truesize
+= PAGE_SIZE
;
1823 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
1825 /* get the new initialized frag */
1826 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
1827 frag
= &skb_shinfo(skb
)->frags
[frg_cnt
- 1];
1829 /* copy the user data to page */
1830 left
= PAGE_SIZE
- frag
->page_offset
;
1831 copy
= (length
> left
)? left
: length
;
1833 ret
= getfrag(from
, (page_address(frag
->page
) +
1834 frag
->page_offset
+ frag
->size
),
1835 offset
, copy
, 0, skb
);
1839 /* copy was successful so update the size parameters */
1840 sk
->sk_sndmsg_off
+= copy
;
1843 skb
->data_len
+= copy
;
1847 } while (length
> 0);
1853 * skb_pull_rcsum - pull skb and update receive checksum
1854 * @skb: buffer to update
1855 * @start: start of data before pull
1856 * @len: length of data pulled
1858 * This function performs an skb_pull on the packet and updates
1859 * update the CHECKSUM_COMPLETE checksum. It should be used on
1860 * receive path processing instead of skb_pull unless you know
1861 * that the checksum difference is zero (e.g., a valid IP header)
1862 * or you are setting ip_summed to CHECKSUM_NONE.
1864 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
1866 BUG_ON(len
> skb
->len
);
1868 BUG_ON(skb
->len
< skb
->data_len
);
1869 skb_postpull_rcsum(skb
, skb
->data
, len
);
1870 return skb
->data
+= len
;
1873 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
1876 * skb_segment - Perform protocol segmentation on skb.
1877 * @skb: buffer to segment
1878 * @features: features for the output path (see dev->features)
1880 * This function performs segmentation on the given skb. It returns
1881 * the segment at the given position. It returns NULL if there are
1882 * no more segments to generate, or when an error is encountered.
1884 struct sk_buff
*skb_segment(struct sk_buff
*skb
, int features
)
1886 struct sk_buff
*segs
= NULL
;
1887 struct sk_buff
*tail
= NULL
;
1888 unsigned int mss
= skb_shinfo(skb
)->gso_size
;
1889 unsigned int doffset
= skb
->data
- skb_mac_header(skb
);
1890 unsigned int offset
= doffset
;
1891 unsigned int headroom
;
1893 int sg
= features
& NETIF_F_SG
;
1894 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1899 __skb_push(skb
, doffset
);
1900 headroom
= skb_headroom(skb
);
1901 pos
= skb_headlen(skb
);
1904 struct sk_buff
*nskb
;
1910 len
= skb
->len
- offset
;
1914 hsize
= skb_headlen(skb
) - offset
;
1917 if (hsize
> len
|| !sg
)
1920 nskb
= alloc_skb(hsize
+ doffset
+ headroom
, GFP_ATOMIC
);
1921 if (unlikely(!nskb
))
1930 nskb
->dev
= skb
->dev
;
1931 nskb
->priority
= skb
->priority
;
1932 nskb
->protocol
= skb
->protocol
;
1933 nskb
->dst
= dst_clone(skb
->dst
);
1934 memcpy(nskb
->cb
, skb
->cb
, sizeof(skb
->cb
));
1935 nskb
->pkt_type
= skb
->pkt_type
;
1936 nskb
->mac_len
= skb
->mac_len
;
1938 skb_reserve(nskb
, headroom
);
1939 skb_reset_mac_header(nskb
);
1940 skb_set_network_header(nskb
, skb
->mac_len
);
1941 nskb
->transport_header
= (nskb
->network_header
+
1942 skb_network_header_len(skb
));
1943 memcpy(skb_put(nskb
, doffset
), skb
->data
, doffset
);
1946 nskb
->csum
= skb_copy_and_csum_bits(skb
, offset
,
1952 frag
= skb_shinfo(nskb
)->frags
;
1955 nskb
->ip_summed
= CHECKSUM_PARTIAL
;
1956 nskb
->csum
= skb
->csum
;
1957 memcpy(skb_put(nskb
, hsize
), skb
->data
+ offset
, hsize
);
1959 while (pos
< offset
+ len
) {
1960 BUG_ON(i
>= nfrags
);
1962 *frag
= skb_shinfo(skb
)->frags
[i
];
1963 get_page(frag
->page
);
1967 frag
->page_offset
+= offset
- pos
;
1968 frag
->size
-= offset
- pos
;
1973 if (pos
+ size
<= offset
+ len
) {
1977 frag
->size
-= pos
+ size
- (offset
+ len
);
1984 skb_shinfo(nskb
)->nr_frags
= k
;
1985 nskb
->data_len
= len
- hsize
;
1986 nskb
->len
+= nskb
->data_len
;
1987 nskb
->truesize
+= nskb
->data_len
;
1988 } while ((offset
+= len
) < skb
->len
);
1993 while ((skb
= segs
)) {
1997 return ERR_PTR(err
);
2000 EXPORT_SYMBOL_GPL(skb_segment
);
2002 void __init
skb_init(void)
2004 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
2005 sizeof(struct sk_buff
),
2007 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2009 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
2010 (2*sizeof(struct sk_buff
)) +
2013 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2017 EXPORT_SYMBOL(___pskb_trim
);
2018 EXPORT_SYMBOL(__kfree_skb
);
2019 EXPORT_SYMBOL(kfree_skb
);
2020 EXPORT_SYMBOL(__pskb_pull_tail
);
2021 EXPORT_SYMBOL(__alloc_skb
);
2022 EXPORT_SYMBOL(__netdev_alloc_skb
);
2023 EXPORT_SYMBOL(pskb_copy
);
2024 EXPORT_SYMBOL(pskb_expand_head
);
2025 EXPORT_SYMBOL(skb_checksum
);
2026 EXPORT_SYMBOL(skb_clone
);
2027 EXPORT_SYMBOL(skb_clone_fraglist
);
2028 EXPORT_SYMBOL(skb_copy
);
2029 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
2030 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
2031 EXPORT_SYMBOL(skb_copy_bits
);
2032 EXPORT_SYMBOL(skb_copy_expand
);
2033 EXPORT_SYMBOL(skb_over_panic
);
2034 EXPORT_SYMBOL(skb_pad
);
2035 EXPORT_SYMBOL(skb_realloc_headroom
);
2036 EXPORT_SYMBOL(skb_under_panic
);
2037 EXPORT_SYMBOL(skb_dequeue
);
2038 EXPORT_SYMBOL(skb_dequeue_tail
);
2039 EXPORT_SYMBOL(skb_insert
);
2040 EXPORT_SYMBOL(skb_queue_purge
);
2041 EXPORT_SYMBOL(skb_queue_head
);
2042 EXPORT_SYMBOL(skb_queue_tail
);
2043 EXPORT_SYMBOL(skb_unlink
);
2044 EXPORT_SYMBOL(skb_append
);
2045 EXPORT_SYMBOL(skb_split
);
2046 EXPORT_SYMBOL(skb_prepare_seq_read
);
2047 EXPORT_SYMBOL(skb_seq_read
);
2048 EXPORT_SYMBOL(skb_abort_seq_read
);
2049 EXPORT_SYMBOL(skb_find_text
);
2050 EXPORT_SYMBOL(skb_append_datato_frags
);