2 * Routines having to do with the 'struct sk_buff' memory handlers.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
8 * Alan Cox : Fixed the worst of the load
10 * Dave Platt : Interrupt stacking fix.
11 * Richard Kooijman : Timestamp fixes.
12 * Alan Cox : Changed buffer format.
13 * Alan Cox : destructor hook for AF_UNIX etc.
14 * Linus Torvalds : Better skb_clone.
15 * Alan Cox : Added skb_copy.
16 * Alan Cox : Added all the changed routines Linus
17 * only put in the headers
18 * Ray VanTassle : Fixed --skb->lock in free
19 * Alan Cox : skb_copy copy arp field
20 * Andi Kleen : slabified it.
21 * Robert Olsson : Removed skb_head_pool
24 * The __skb_ routines should be called with interrupts
25 * disabled, or you better be *real* sure that the operation is atomic
26 * with respect to whatever list is being frobbed (e.g. via lock_sock()
27 * or via disabling bottom half handlers, etc).
29 * This program is free software; you can redistribute it and/or
30 * modify it under the terms of the GNU General Public License
31 * as published by the Free Software Foundation; either version
32 * 2 of the License, or (at your option) any later version.
36 * The functions in this file will not compile correctly with gcc 2.4.x
39 #include <linux/module.h>
40 #include <linux/types.h>
41 #include <linux/kernel.h>
43 #include <linux/interrupt.h>
45 #include <linux/inet.h>
46 #include <linux/slab.h>
47 #include <linux/netdevice.h>
48 #ifdef CONFIG_NET_CLS_ACT
49 #include <net/pkt_sched.h>
51 #include <linux/string.h>
52 #include <linux/skbuff.h>
53 #include <linux/splice.h>
54 #include <linux/cache.h>
55 #include <linux/rtnetlink.h>
56 #include <linux/init.h>
57 #include <linux/scatterlist.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
;
73 static void sock_pipe_buf_release(struct pipe_inode_info
*pipe
,
74 struct pipe_buffer
*buf
)
79 static void sock_pipe_buf_get(struct pipe_inode_info
*pipe
,
80 struct pipe_buffer
*buf
)
85 static int sock_pipe_buf_steal(struct pipe_inode_info
*pipe
,
86 struct pipe_buffer
*buf
)
92 /* Pipe buffer operations for a socket. */
93 static struct pipe_buf_operations sock_pipe_buf_ops
= {
95 .map
= generic_pipe_buf_map
,
96 .unmap
= generic_pipe_buf_unmap
,
97 .confirm
= generic_pipe_buf_confirm
,
98 .release
= sock_pipe_buf_release
,
99 .steal
= sock_pipe_buf_steal
,
100 .get
= sock_pipe_buf_get
,
104 * Keep out-of-line to prevent kernel bloat.
105 * __builtin_return_address is not used because it is not always
110 * skb_over_panic - private function
115 * Out of line support code for skb_put(). Not user callable.
117 void skb_over_panic(struct sk_buff
*skb
, int sz
, void *here
)
119 printk(KERN_EMERG
"skb_over_panic: text:%p len:%d put:%d head:%p "
120 "data:%p tail:%#lx end:%#lx dev:%s\n",
121 here
, skb
->len
, sz
, skb
->head
, skb
->data
,
122 (unsigned long)skb
->tail
, (unsigned long)skb
->end
,
123 skb
->dev
? skb
->dev
->name
: "<NULL>");
128 * skb_under_panic - private function
133 * Out of line support code for skb_push(). Not user callable.
136 void skb_under_panic(struct sk_buff
*skb
, int sz
, void *here
)
138 printk(KERN_EMERG
"skb_under_panic: text:%p len:%d put:%d head:%p "
139 "data:%p tail:%#lx end:%#lx dev:%s\n",
140 here
, skb
->len
, sz
, skb
->head
, skb
->data
,
141 (unsigned long)skb
->tail
, (unsigned long)skb
->end
,
142 skb
->dev
? skb
->dev
->name
: "<NULL>");
146 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
147 * 'private' fields and also do memory statistics to find all the
153 * __alloc_skb - allocate a network buffer
154 * @size: size to allocate
155 * @gfp_mask: allocation mask
156 * @fclone: allocate from fclone cache instead of head cache
157 * and allocate a cloned (child) skb
158 * @node: numa node to allocate memory on
160 * Allocate a new &sk_buff. The returned buffer has no headroom and a
161 * tail room of size bytes. The object has a reference count of one.
162 * The return is the buffer. On a failure the return is %NULL.
164 * Buffers may only be allocated from interrupts using a @gfp_mask of
167 struct sk_buff
*__alloc_skb(unsigned int size
, gfp_t gfp_mask
,
168 int fclone
, int node
)
170 struct kmem_cache
*cache
;
171 struct skb_shared_info
*shinfo
;
175 cache
= fclone
? skbuff_fclone_cache
: skbuff_head_cache
;
178 skb
= kmem_cache_alloc_node(cache
, gfp_mask
& ~__GFP_DMA
, node
);
182 size
= SKB_DATA_ALIGN(size
);
183 data
= kmalloc_node_track_caller(size
+ sizeof(struct skb_shared_info
),
189 * Only clear those fields we need to clear, not those that we will
190 * actually initialise below. Hence, don't put any more fields after
191 * the tail pointer in struct sk_buff!
193 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
194 skb
->truesize
= size
+ sizeof(struct sk_buff
);
195 atomic_set(&skb
->users
, 1);
198 skb_reset_tail_pointer(skb
);
199 skb
->end
= skb
->tail
+ size
;
200 /* make sure we initialize shinfo sequentially */
201 shinfo
= skb_shinfo(skb
);
202 atomic_set(&shinfo
->dataref
, 1);
203 shinfo
->nr_frags
= 0;
204 shinfo
->gso_size
= 0;
205 shinfo
->gso_segs
= 0;
206 shinfo
->gso_type
= 0;
207 shinfo
->ip6_frag_id
= 0;
208 shinfo
->frag_list
= NULL
;
211 struct sk_buff
*child
= skb
+ 1;
212 atomic_t
*fclone_ref
= (atomic_t
*) (child
+ 1);
214 skb
->fclone
= SKB_FCLONE_ORIG
;
215 atomic_set(fclone_ref
, 1);
217 child
->fclone
= SKB_FCLONE_UNAVAILABLE
;
222 kmem_cache_free(cache
, skb
);
228 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
229 * @dev: network device to receive on
230 * @length: length to allocate
231 * @gfp_mask: get_free_pages mask, passed to alloc_skb
233 * Allocate a new &sk_buff and assign it a usage count of one. The
234 * buffer has unspecified headroom built in. Users should allocate
235 * the headroom they think they need without accounting for the
236 * built in space. The built in space is used for optimisations.
238 * %NULL is returned if there is no free memory.
240 struct sk_buff
*__netdev_alloc_skb(struct net_device
*dev
,
241 unsigned int length
, gfp_t gfp_mask
)
243 int node
= dev
->dev
.parent
? dev_to_node(dev
->dev
.parent
) : -1;
246 skb
= __alloc_skb(length
+ NET_SKB_PAD
, gfp_mask
, 0, node
);
248 skb_reserve(skb
, NET_SKB_PAD
);
254 struct page
*__netdev_alloc_page(struct net_device
*dev
, gfp_t gfp_mask
)
256 int node
= dev
->dev
.parent
? dev_to_node(dev
->dev
.parent
) : -1;
259 page
= alloc_pages_node(node
, gfp_mask
, 0);
262 EXPORT_SYMBOL(__netdev_alloc_page
);
264 void skb_add_rx_frag(struct sk_buff
*skb
, int i
, struct page
*page
, int off
,
267 skb_fill_page_desc(skb
, i
, page
, off
, size
);
269 skb
->data_len
+= size
;
270 skb
->truesize
+= size
;
272 EXPORT_SYMBOL(skb_add_rx_frag
);
275 * dev_alloc_skb - allocate an skbuff for receiving
276 * @length: length to allocate
278 * Allocate a new &sk_buff and assign it a usage count of one. The
279 * buffer has unspecified headroom built in. Users should allocate
280 * the headroom they think they need without accounting for the
281 * built in space. The built in space is used for optimisations.
283 * %NULL is returned if there is no free memory. Although this function
284 * allocates memory it can be called from an interrupt.
286 struct sk_buff
*dev_alloc_skb(unsigned int length
)
289 * There is more code here than it seems:
290 * __dev_alloc_skb is an inline
292 return __dev_alloc_skb(length
, GFP_ATOMIC
);
294 EXPORT_SYMBOL(dev_alloc_skb
);
296 static void skb_drop_list(struct sk_buff
**listp
)
298 struct sk_buff
*list
= *listp
;
303 struct sk_buff
*this = list
;
309 static inline void skb_drop_fraglist(struct sk_buff
*skb
)
311 skb_drop_list(&skb_shinfo(skb
)->frag_list
);
314 static void skb_clone_fraglist(struct sk_buff
*skb
)
316 struct sk_buff
*list
;
318 for (list
= skb_shinfo(skb
)->frag_list
; list
; list
= list
->next
)
322 static void skb_release_data(struct sk_buff
*skb
)
325 !atomic_sub_return(skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1,
326 &skb_shinfo(skb
)->dataref
)) {
327 if (skb_shinfo(skb
)->nr_frags
) {
329 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
330 put_page(skb_shinfo(skb
)->frags
[i
].page
);
333 if (skb_shinfo(skb
)->frag_list
)
334 skb_drop_fraglist(skb
);
341 * Free an skbuff by memory without cleaning the state.
343 static void kfree_skbmem(struct sk_buff
*skb
)
345 struct sk_buff
*other
;
346 atomic_t
*fclone_ref
;
348 switch (skb
->fclone
) {
349 case SKB_FCLONE_UNAVAILABLE
:
350 kmem_cache_free(skbuff_head_cache
, skb
);
353 case SKB_FCLONE_ORIG
:
354 fclone_ref
= (atomic_t
*) (skb
+ 2);
355 if (atomic_dec_and_test(fclone_ref
))
356 kmem_cache_free(skbuff_fclone_cache
, skb
);
359 case SKB_FCLONE_CLONE
:
360 fclone_ref
= (atomic_t
*) (skb
+ 1);
363 /* The clone portion is available for
364 * fast-cloning again.
366 skb
->fclone
= SKB_FCLONE_UNAVAILABLE
;
368 if (atomic_dec_and_test(fclone_ref
))
369 kmem_cache_free(skbuff_fclone_cache
, other
);
374 static void skb_release_head_state(struct sk_buff
*skb
)
376 dst_release(skb
->dst
);
378 secpath_put(skb
->sp
);
380 if (skb
->destructor
) {
382 skb
->destructor(skb
);
384 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
385 nf_conntrack_put(skb
->nfct
);
386 nf_conntrack_put_reasm(skb
->nfct_reasm
);
388 #ifdef CONFIG_BRIDGE_NETFILTER
389 nf_bridge_put(skb
->nf_bridge
);
391 /* XXX: IS this still necessary? - JHS */
392 #ifdef CONFIG_NET_SCHED
394 #ifdef CONFIG_NET_CLS_ACT
400 /* Free everything but the sk_buff shell. */
401 static void skb_release_all(struct sk_buff
*skb
)
403 skb_release_head_state(skb
);
404 skb_release_data(skb
);
408 * __kfree_skb - private function
411 * Free an sk_buff. Release anything attached to the buffer.
412 * Clean the state. This is an internal helper function. Users should
413 * always call kfree_skb
416 void __kfree_skb(struct sk_buff
*skb
)
418 skb_release_all(skb
);
423 * kfree_skb - free an sk_buff
424 * @skb: buffer to free
426 * Drop a reference to the buffer and free it if the usage count has
429 void kfree_skb(struct sk_buff
*skb
)
433 if (likely(atomic_read(&skb
->users
) == 1))
435 else if (likely(!atomic_dec_and_test(&skb
->users
)))
441 * skb_recycle_check - check if skb can be reused for receive
443 * @skb_size: minimum receive buffer size
445 * Checks that the skb passed in is not shared or cloned, and
446 * that it is linear and its head portion at least as large as
447 * skb_size so that it can be recycled as a receive buffer.
448 * If these conditions are met, this function does any necessary
449 * reference count dropping and cleans up the skbuff as if it
450 * just came from __alloc_skb().
452 int skb_recycle_check(struct sk_buff
*skb
, int skb_size
)
454 struct skb_shared_info
*shinfo
;
456 if (skb_is_nonlinear(skb
) || skb
->fclone
!= SKB_FCLONE_UNAVAILABLE
)
459 skb_size
= SKB_DATA_ALIGN(skb_size
+ NET_SKB_PAD
);
460 if (skb_end_pointer(skb
) - skb
->head
< skb_size
)
463 if (skb_shared(skb
) || skb_cloned(skb
))
466 skb_release_head_state(skb
);
467 shinfo
= skb_shinfo(skb
);
468 atomic_set(&shinfo
->dataref
, 1);
469 shinfo
->nr_frags
= 0;
470 shinfo
->gso_size
= 0;
471 shinfo
->gso_segs
= 0;
472 shinfo
->gso_type
= 0;
473 shinfo
->ip6_frag_id
= 0;
474 shinfo
->frag_list
= NULL
;
476 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
477 skb
->data
= skb
->head
+ NET_SKB_PAD
;
478 skb_reset_tail_pointer(skb
);
482 EXPORT_SYMBOL(skb_recycle_check
);
484 static void __copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
486 new->tstamp
= old
->tstamp
;
488 new->transport_header
= old
->transport_header
;
489 new->network_header
= old
->network_header
;
490 new->mac_header
= old
->mac_header
;
491 new->dst
= dst_clone(old
->dst
);
493 new->sp
= secpath_get(old
->sp
);
495 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
496 new->csum_start
= old
->csum_start
;
497 new->csum_offset
= old
->csum_offset
;
498 new->local_df
= old
->local_df
;
499 new->pkt_type
= old
->pkt_type
;
500 new->ip_summed
= old
->ip_summed
;
501 skb_copy_queue_mapping(new, old
);
502 new->priority
= old
->priority
;
503 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
504 new->ipvs_property
= old
->ipvs_property
;
506 new->protocol
= old
->protocol
;
507 new->mark
= old
->mark
;
509 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
510 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
511 new->nf_trace
= old
->nf_trace
;
513 #ifdef CONFIG_NET_SCHED
514 new->tc_index
= old
->tc_index
;
515 #ifdef CONFIG_NET_CLS_ACT
516 new->tc_verd
= old
->tc_verd
;
519 new->vlan_tci
= old
->vlan_tci
;
521 skb_copy_secmark(new, old
);
524 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
526 #define C(x) n->x = skb->x
528 n
->next
= n
->prev
= NULL
;
530 __copy_skb_header(n
, skb
);
535 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
538 n
->destructor
= NULL
;
545 #if defined(CONFIG_MAC80211) || defined(CONFIG_MAC80211_MODULE)
549 atomic_set(&n
->users
, 1);
551 atomic_inc(&(skb_shinfo(skb
)->dataref
));
559 * skb_morph - morph one skb into another
560 * @dst: the skb to receive the contents
561 * @src: the skb to supply the contents
563 * This is identical to skb_clone except that the target skb is
564 * supplied by the user.
566 * The target skb is returned upon exit.
568 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
570 skb_release_all(dst
);
571 return __skb_clone(dst
, src
);
573 EXPORT_SYMBOL_GPL(skb_morph
);
576 * skb_clone - duplicate an sk_buff
577 * @skb: buffer to clone
578 * @gfp_mask: allocation priority
580 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
581 * copies share the same packet data but not structure. The new
582 * buffer has a reference count of 1. If the allocation fails the
583 * function returns %NULL otherwise the new buffer is returned.
585 * If this function is called from an interrupt gfp_mask() must be
589 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
594 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
595 n
->fclone
== SKB_FCLONE_UNAVAILABLE
) {
596 atomic_t
*fclone_ref
= (atomic_t
*) (n
+ 1);
597 n
->fclone
= SKB_FCLONE_CLONE
;
598 atomic_inc(fclone_ref
);
600 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
603 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
606 return __skb_clone(n
, skb
);
609 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
611 #ifndef NET_SKBUFF_DATA_USES_OFFSET
613 * Shift between the two data areas in bytes
615 unsigned long offset
= new->data
- old
->data
;
618 __copy_skb_header(new, old
);
620 #ifndef NET_SKBUFF_DATA_USES_OFFSET
621 /* {transport,network,mac}_header are relative to skb->head */
622 new->transport_header
+= offset
;
623 new->network_header
+= offset
;
624 new->mac_header
+= offset
;
626 skb_shinfo(new)->gso_size
= skb_shinfo(old
)->gso_size
;
627 skb_shinfo(new)->gso_segs
= skb_shinfo(old
)->gso_segs
;
628 skb_shinfo(new)->gso_type
= skb_shinfo(old
)->gso_type
;
632 * skb_copy - create private copy of an sk_buff
633 * @skb: buffer to copy
634 * @gfp_mask: allocation priority
636 * Make a copy of both an &sk_buff and its data. This is used when the
637 * caller wishes to modify the data and needs a private copy of the
638 * data to alter. Returns %NULL on failure or the pointer to the buffer
639 * on success. The returned buffer has a reference count of 1.
641 * As by-product this function converts non-linear &sk_buff to linear
642 * one, so that &sk_buff becomes completely private and caller is allowed
643 * to modify all the data of returned buffer. This means that this
644 * function is not recommended for use in circumstances when only
645 * header is going to be modified. Use pskb_copy() instead.
648 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
650 int headerlen
= skb
->data
- skb
->head
;
652 * Allocate the copy buffer
655 #ifdef NET_SKBUFF_DATA_USES_OFFSET
656 n
= alloc_skb(skb
->end
+ skb
->data_len
, gfp_mask
);
658 n
= alloc_skb(skb
->end
- skb
->head
+ skb
->data_len
, gfp_mask
);
663 /* Set the data pointer */
664 skb_reserve(n
, headerlen
);
665 /* Set the tail pointer and length */
666 skb_put(n
, skb
->len
);
668 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
671 copy_skb_header(n
, skb
);
677 * pskb_copy - create copy of an sk_buff with private head.
678 * @skb: buffer to copy
679 * @gfp_mask: allocation priority
681 * Make a copy of both an &sk_buff and part of its data, located
682 * in header. Fragmented data remain shared. This is used when
683 * the caller wishes to modify only header of &sk_buff and needs
684 * private copy of the header to alter. Returns %NULL on failure
685 * or the pointer to the buffer on success.
686 * The returned buffer has a reference count of 1.
689 struct sk_buff
*pskb_copy(struct sk_buff
*skb
, gfp_t gfp_mask
)
692 * Allocate the copy buffer
695 #ifdef NET_SKBUFF_DATA_USES_OFFSET
696 n
= alloc_skb(skb
->end
, gfp_mask
);
698 n
= alloc_skb(skb
->end
- skb
->head
, gfp_mask
);
703 /* Set the data pointer */
704 skb_reserve(n
, skb
->data
- skb
->head
);
705 /* Set the tail pointer and length */
706 skb_put(n
, skb_headlen(skb
));
708 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
710 n
->truesize
+= skb
->data_len
;
711 n
->data_len
= skb
->data_len
;
714 if (skb_shinfo(skb
)->nr_frags
) {
717 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
718 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
719 get_page(skb_shinfo(n
)->frags
[i
].page
);
721 skb_shinfo(n
)->nr_frags
= i
;
724 if (skb_shinfo(skb
)->frag_list
) {
725 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
726 skb_clone_fraglist(n
);
729 copy_skb_header(n
, skb
);
735 * pskb_expand_head - reallocate header of &sk_buff
736 * @skb: buffer to reallocate
737 * @nhead: room to add at head
738 * @ntail: room to add at tail
739 * @gfp_mask: allocation priority
741 * Expands (or creates identical copy, if &nhead and &ntail are zero)
742 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
743 * reference count of 1. Returns zero in the case of success or error,
744 * if expansion failed. In the last case, &sk_buff is not changed.
746 * All the pointers pointing into skb header may change and must be
747 * reloaded after call to this function.
750 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
755 #ifdef NET_SKBUFF_DATA_USES_OFFSET
756 int size
= nhead
+ skb
->end
+ ntail
;
758 int size
= nhead
+ (skb
->end
- skb
->head
) + ntail
;
767 size
= SKB_DATA_ALIGN(size
);
769 data
= kmalloc(size
+ sizeof(struct skb_shared_info
), gfp_mask
);
773 /* Copy only real data... and, alas, header. This should be
774 * optimized for the cases when header is void. */
775 #ifdef NET_SKBUFF_DATA_USES_OFFSET
776 memcpy(data
+ nhead
, skb
->head
, skb
->tail
);
778 memcpy(data
+ nhead
, skb
->head
, skb
->tail
- skb
->head
);
780 memcpy(data
+ size
, skb_end_pointer(skb
),
781 sizeof(struct skb_shared_info
));
783 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
784 get_page(skb_shinfo(skb
)->frags
[i
].page
);
786 if (skb_shinfo(skb
)->frag_list
)
787 skb_clone_fraglist(skb
);
789 skb_release_data(skb
);
791 off
= (data
+ nhead
) - skb
->head
;
795 #ifdef NET_SKBUFF_DATA_USES_OFFSET
799 skb
->end
= skb
->head
+ size
;
801 /* {transport,network,mac}_header and tail are relative to skb->head */
803 skb
->transport_header
+= off
;
804 skb
->network_header
+= off
;
805 skb
->mac_header
+= off
;
806 skb
->csum_start
+= nhead
;
810 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
817 /* Make private copy of skb with writable head and some headroom */
819 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
821 struct sk_buff
*skb2
;
822 int delta
= headroom
- skb_headroom(skb
);
825 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
827 skb2
= skb_clone(skb
, GFP_ATOMIC
);
828 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
839 * skb_copy_expand - copy and expand sk_buff
840 * @skb: buffer to copy
841 * @newheadroom: new free bytes at head
842 * @newtailroom: new free bytes at tail
843 * @gfp_mask: allocation priority
845 * Make a copy of both an &sk_buff and its data and while doing so
846 * allocate additional space.
848 * This is used when the caller wishes to modify the data and needs a
849 * private copy of the data to alter as well as more space for new fields.
850 * Returns %NULL on failure or the pointer to the buffer
851 * on success. The returned buffer has a reference count of 1.
853 * You must pass %GFP_ATOMIC as the allocation priority if this function
854 * is called from an interrupt.
856 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
857 int newheadroom
, int newtailroom
,
861 * Allocate the copy buffer
863 struct sk_buff
*n
= alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
865 int oldheadroom
= skb_headroom(skb
);
866 int head_copy_len
, head_copy_off
;
872 skb_reserve(n
, newheadroom
);
874 /* Set the tail pointer and length */
875 skb_put(n
, skb
->len
);
877 head_copy_len
= oldheadroom
;
879 if (newheadroom
<= head_copy_len
)
880 head_copy_len
= newheadroom
;
882 head_copy_off
= newheadroom
- head_copy_len
;
884 /* Copy the linear header and data. */
885 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
886 skb
->len
+ head_copy_len
))
889 copy_skb_header(n
, skb
);
891 off
= newheadroom
- oldheadroom
;
892 n
->csum_start
+= off
;
893 #ifdef NET_SKBUFF_DATA_USES_OFFSET
894 n
->transport_header
+= off
;
895 n
->network_header
+= off
;
896 n
->mac_header
+= off
;
903 * skb_pad - zero pad the tail of an skb
904 * @skb: buffer to pad
907 * Ensure that a buffer is followed by a padding area that is zero
908 * filled. Used by network drivers which may DMA or transfer data
909 * beyond the buffer end onto the wire.
911 * May return error in out of memory cases. The skb is freed on error.
914 int skb_pad(struct sk_buff
*skb
, int pad
)
919 /* If the skbuff is non linear tailroom is always zero.. */
920 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
921 memset(skb
->data
+skb
->len
, 0, pad
);
925 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
926 if (likely(skb_cloned(skb
) || ntail
> 0)) {
927 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
932 /* FIXME: The use of this function with non-linear skb's really needs
935 err
= skb_linearize(skb
);
939 memset(skb
->data
+ skb
->len
, 0, pad
);
948 * skb_put - add data to a buffer
949 * @skb: buffer to use
950 * @len: amount of data to add
952 * This function extends the used data area of the buffer. If this would
953 * exceed the total buffer size the kernel will panic. A pointer to the
954 * first byte of the extra data is returned.
956 unsigned char *skb_put(struct sk_buff
*skb
, unsigned int len
)
958 unsigned char *tmp
= skb_tail_pointer(skb
);
959 SKB_LINEAR_ASSERT(skb
);
962 if (unlikely(skb
->tail
> skb
->end
))
963 skb_over_panic(skb
, len
, __builtin_return_address(0));
966 EXPORT_SYMBOL(skb_put
);
969 * skb_push - add data to the start of a buffer
970 * @skb: buffer to use
971 * @len: amount of data to add
973 * This function extends the used data area of the buffer at the buffer
974 * start. If this would exceed the total buffer headroom the kernel will
975 * panic. A pointer to the first byte of the extra data is returned.
977 unsigned char *skb_push(struct sk_buff
*skb
, unsigned int len
)
981 if (unlikely(skb
->data
<skb
->head
))
982 skb_under_panic(skb
, len
, __builtin_return_address(0));
985 EXPORT_SYMBOL(skb_push
);
988 * skb_pull - remove data from the start of a buffer
989 * @skb: buffer to use
990 * @len: amount of data to remove
992 * This function removes data from the start of a buffer, returning
993 * the memory to the headroom. A pointer to the next data in the buffer
994 * is returned. Once the data has been pulled future pushes will overwrite
997 unsigned char *skb_pull(struct sk_buff
*skb
, unsigned int len
)
999 return unlikely(len
> skb
->len
) ? NULL
: __skb_pull(skb
, len
);
1001 EXPORT_SYMBOL(skb_pull
);
1004 * skb_trim - remove end from a buffer
1005 * @skb: buffer to alter
1008 * Cut the length of a buffer down by removing data from the tail. If
1009 * the buffer is already under the length specified it is not modified.
1010 * The skb must be linear.
1012 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1015 __skb_trim(skb
, len
);
1017 EXPORT_SYMBOL(skb_trim
);
1019 /* Trims skb to length len. It can change skb pointers.
1022 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1024 struct sk_buff
**fragp
;
1025 struct sk_buff
*frag
;
1026 int offset
= skb_headlen(skb
);
1027 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1031 if (skb_cloned(skb
) &&
1032 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1039 for (; i
< nfrags
; i
++) {
1040 int end
= offset
+ skb_shinfo(skb
)->frags
[i
].size
;
1047 skb_shinfo(skb
)->frags
[i
++].size
= len
- offset
;
1050 skb_shinfo(skb
)->nr_frags
= i
;
1052 for (; i
< nfrags
; i
++)
1053 put_page(skb_shinfo(skb
)->frags
[i
].page
);
1055 if (skb_shinfo(skb
)->frag_list
)
1056 skb_drop_fraglist(skb
);
1060 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1061 fragp
= &frag
->next
) {
1062 int end
= offset
+ frag
->len
;
1064 if (skb_shared(frag
)) {
1065 struct sk_buff
*nfrag
;
1067 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1068 if (unlikely(!nfrag
))
1071 nfrag
->next
= frag
->next
;
1083 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1087 skb_drop_list(&frag
->next
);
1092 if (len
> skb_headlen(skb
)) {
1093 skb
->data_len
-= skb
->len
- len
;
1098 skb_set_tail_pointer(skb
, len
);
1105 * __pskb_pull_tail - advance tail of skb header
1106 * @skb: buffer to reallocate
1107 * @delta: number of bytes to advance tail
1109 * The function makes a sense only on a fragmented &sk_buff,
1110 * it expands header moving its tail forward and copying necessary
1111 * data from fragmented part.
1113 * &sk_buff MUST have reference count of 1.
1115 * Returns %NULL (and &sk_buff does not change) if pull failed
1116 * or value of new tail of skb in the case of success.
1118 * All the pointers pointing into skb header may change and must be
1119 * reloaded after call to this function.
1122 /* Moves tail of skb head forward, copying data from fragmented part,
1123 * when it is necessary.
1124 * 1. It may fail due to malloc failure.
1125 * 2. It may change skb pointers.
1127 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1129 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1131 /* If skb has not enough free space at tail, get new one
1132 * plus 128 bytes for future expansions. If we have enough
1133 * room at tail, reallocate without expansion only if skb is cloned.
1135 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1137 if (eat
> 0 || skb_cloned(skb
)) {
1138 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1143 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
1146 /* Optimization: no fragments, no reasons to preestimate
1147 * size of pulled pages. Superb.
1149 if (!skb_shinfo(skb
)->frag_list
)
1152 /* Estimate size of pulled pages. */
1154 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1155 if (skb_shinfo(skb
)->frags
[i
].size
>= eat
)
1157 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1160 /* If we need update frag list, we are in troubles.
1161 * Certainly, it possible to add an offset to skb data,
1162 * but taking into account that pulling is expected to
1163 * be very rare operation, it is worth to fight against
1164 * further bloating skb head and crucify ourselves here instead.
1165 * Pure masohism, indeed. 8)8)
1168 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1169 struct sk_buff
*clone
= NULL
;
1170 struct sk_buff
*insp
= NULL
;
1175 if (list
->len
<= eat
) {
1176 /* Eaten as whole. */
1181 /* Eaten partially. */
1183 if (skb_shared(list
)) {
1184 /* Sucks! We need to fork list. :-( */
1185 clone
= skb_clone(list
, GFP_ATOMIC
);
1191 /* This may be pulled without
1195 if (!pskb_pull(list
, eat
)) {
1204 /* Free pulled out fragments. */
1205 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1206 skb_shinfo(skb
)->frag_list
= list
->next
;
1209 /* And insert new clone at head. */
1212 skb_shinfo(skb
)->frag_list
= clone
;
1215 /* Success! Now we may commit changes to skb data. */
1220 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1221 if (skb_shinfo(skb
)->frags
[i
].size
<= eat
) {
1222 put_page(skb_shinfo(skb
)->frags
[i
].page
);
1223 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1225 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1227 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1228 skb_shinfo(skb
)->frags
[k
].size
-= eat
;
1234 skb_shinfo(skb
)->nr_frags
= k
;
1237 skb
->data_len
-= delta
;
1239 return skb_tail_pointer(skb
);
1242 /* Copy some data bits from skb to kernel buffer. */
1244 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1247 int start
= skb_headlen(skb
);
1249 if (offset
> (int)skb
->len
- len
)
1253 if ((copy
= start
- offset
) > 0) {
1256 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
1257 if ((len
-= copy
) == 0)
1263 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1266 WARN_ON(start
> offset
+ len
);
1268 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1269 if ((copy
= end
- offset
) > 0) {
1275 vaddr
= kmap_skb_frag(&skb_shinfo(skb
)->frags
[i
]);
1277 vaddr
+ skb_shinfo(skb
)->frags
[i
].page_offset
+
1278 offset
- start
, copy
);
1279 kunmap_skb_frag(vaddr
);
1281 if ((len
-= copy
) == 0)
1289 if (skb_shinfo(skb
)->frag_list
) {
1290 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1292 for (; list
; list
= list
->next
) {
1295 WARN_ON(start
> offset
+ len
);
1297 end
= start
+ list
->len
;
1298 if ((copy
= end
- offset
) > 0) {
1301 if (skb_copy_bits(list
, offset
- start
,
1304 if ((len
-= copy
) == 0)
1320 * Callback from splice_to_pipe(), if we need to release some pages
1321 * at the end of the spd in case we error'ed out in filling the pipe.
1323 static void sock_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
1325 put_page(spd
->pages
[i
]);
1328 static inline struct page
*linear_to_page(struct page
*page
, unsigned int len
,
1329 unsigned int offset
)
1331 struct page
*p
= alloc_pages(GFP_KERNEL
, 0);
1335 memcpy(page_address(p
) + offset
, page_address(page
) + offset
, len
);
1341 * Fill page/offset/length into spd, if it can hold more pages.
1343 static inline int spd_fill_page(struct splice_pipe_desc
*spd
, struct page
*page
,
1344 unsigned int len
, unsigned int offset
,
1345 struct sk_buff
*skb
, int linear
)
1347 if (unlikely(spd
->nr_pages
== PIPE_BUFFERS
))
1351 page
= linear_to_page(page
, len
, offset
);
1357 spd
->pages
[spd
->nr_pages
] = page
;
1358 spd
->partial
[spd
->nr_pages
].len
= len
;
1359 spd
->partial
[spd
->nr_pages
].offset
= offset
;
1365 static inline void __segment_seek(struct page
**page
, unsigned int *poff
,
1366 unsigned int *plen
, unsigned int off
)
1369 *page
+= *poff
/ PAGE_SIZE
;
1370 *poff
= *poff
% PAGE_SIZE
;
1374 static inline int __splice_segment(struct page
*page
, unsigned int poff
,
1375 unsigned int plen
, unsigned int *off
,
1376 unsigned int *len
, struct sk_buff
*skb
,
1377 struct splice_pipe_desc
*spd
, int linear
)
1382 /* skip this segment if already processed */
1388 /* ignore any bits we already processed */
1390 __segment_seek(&page
, &poff
, &plen
, *off
);
1395 unsigned int flen
= min(*len
, plen
);
1397 /* the linear region may spread across several pages */
1398 flen
= min_t(unsigned int, flen
, PAGE_SIZE
- poff
);
1400 if (spd_fill_page(spd
, page
, flen
, poff
, skb
, linear
))
1403 __segment_seek(&page
, &poff
, &plen
, flen
);
1406 } while (*len
&& plen
);
1412 * Map linear and fragment data from the skb to spd. It reports failure if the
1413 * pipe is full or if we already spliced the requested length.
1415 static int __skb_splice_bits(struct sk_buff
*skb
, unsigned int *offset
,
1417 struct splice_pipe_desc
*spd
)
1422 * map the linear part
1424 if (__splice_segment(virt_to_page(skb
->data
),
1425 (unsigned long) skb
->data
& (PAGE_SIZE
- 1),
1427 offset
, len
, skb
, spd
, 1))
1431 * then map the fragments
1433 for (seg
= 0; seg
< skb_shinfo(skb
)->nr_frags
; seg
++) {
1434 const skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[seg
];
1436 if (__splice_segment(f
->page
, f
->page_offset
, f
->size
,
1437 offset
, len
, skb
, spd
, 0))
1445 * Map data from the skb to a pipe. Should handle both the linear part,
1446 * the fragments, and the frag list. It does NOT handle frag lists within
1447 * the frag list, if such a thing exists. We'd probably need to recurse to
1448 * handle that cleanly.
1450 int skb_splice_bits(struct sk_buff
*skb
, unsigned int offset
,
1451 struct pipe_inode_info
*pipe
, unsigned int tlen
,
1454 struct partial_page partial
[PIPE_BUFFERS
];
1455 struct page
*pages
[PIPE_BUFFERS
];
1456 struct splice_pipe_desc spd
= {
1460 .ops
= &sock_pipe_buf_ops
,
1461 .spd_release
= sock_spd_release
,
1465 * __skb_splice_bits() only fails if the output has no room left,
1466 * so no point in going over the frag_list for the error case.
1468 if (__skb_splice_bits(skb
, &offset
, &tlen
, &spd
))
1474 * now see if we have a frag_list to map
1476 if (skb_shinfo(skb
)->frag_list
) {
1477 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1479 for (; list
&& tlen
; list
= list
->next
) {
1480 if (__skb_splice_bits(list
, &offset
, &tlen
, &spd
))
1487 struct sock
*sk
= skb
->sk
;
1491 * Drop the socket lock, otherwise we have reverse
1492 * locking dependencies between sk_lock and i_mutex
1493 * here as compared to sendfile(). We enter here
1494 * with the socket lock held, and splice_to_pipe() will
1495 * grab the pipe inode lock. For sendfile() emulation,
1496 * we call into ->sendpage() with the i_mutex lock held
1497 * and networking will grab the socket lock.
1500 ret
= splice_to_pipe(pipe
, &spd
);
1509 * skb_store_bits - store bits from kernel buffer to skb
1510 * @skb: destination buffer
1511 * @offset: offset in destination
1512 * @from: source buffer
1513 * @len: number of bytes to copy
1515 * Copy the specified number of bytes from the source buffer to the
1516 * destination skb. This function handles all the messy bits of
1517 * traversing fragment lists and such.
1520 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
1523 int start
= skb_headlen(skb
);
1525 if (offset
> (int)skb
->len
- len
)
1528 if ((copy
= start
- offset
) > 0) {
1531 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
1532 if ((len
-= copy
) == 0)
1538 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1539 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1542 WARN_ON(start
> offset
+ len
);
1544 end
= start
+ frag
->size
;
1545 if ((copy
= end
- offset
) > 0) {
1551 vaddr
= kmap_skb_frag(frag
);
1552 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1554 kunmap_skb_frag(vaddr
);
1556 if ((len
-= copy
) == 0)
1564 if (skb_shinfo(skb
)->frag_list
) {
1565 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1567 for (; list
; list
= list
->next
) {
1570 WARN_ON(start
> offset
+ len
);
1572 end
= start
+ list
->len
;
1573 if ((copy
= end
- offset
) > 0) {
1576 if (skb_store_bits(list
, offset
- start
,
1579 if ((len
-= copy
) == 0)
1594 EXPORT_SYMBOL(skb_store_bits
);
1596 /* Checksum skb data. */
1598 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
1599 int len
, __wsum csum
)
1601 int start
= skb_headlen(skb
);
1602 int i
, copy
= start
- offset
;
1605 /* Checksum header. */
1609 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1610 if ((len
-= copy
) == 0)
1616 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1619 WARN_ON(start
> offset
+ len
);
1621 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1622 if ((copy
= end
- offset
) > 0) {
1625 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1629 vaddr
= kmap_skb_frag(frag
);
1630 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
1631 offset
- start
, copy
, 0);
1632 kunmap_skb_frag(vaddr
);
1633 csum
= csum_block_add(csum
, csum2
, pos
);
1642 if (skb_shinfo(skb
)->frag_list
) {
1643 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1645 for (; list
; list
= list
->next
) {
1648 WARN_ON(start
> offset
+ len
);
1650 end
= start
+ list
->len
;
1651 if ((copy
= end
- offset
) > 0) {
1655 csum2
= skb_checksum(list
, offset
- start
,
1657 csum
= csum_block_add(csum
, csum2
, pos
);
1658 if ((len
-= copy
) == 0)
1671 /* Both of above in one bottle. */
1673 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
1674 u8
*to
, int len
, __wsum csum
)
1676 int start
= skb_headlen(skb
);
1677 int i
, copy
= start
- offset
;
1684 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
1686 if ((len
-= copy
) == 0)
1693 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1696 WARN_ON(start
> offset
+ len
);
1698 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1699 if ((copy
= end
- offset
) > 0) {
1702 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1706 vaddr
= kmap_skb_frag(frag
);
1707 csum2
= csum_partial_copy_nocheck(vaddr
+
1711 kunmap_skb_frag(vaddr
);
1712 csum
= csum_block_add(csum
, csum2
, pos
);
1722 if (skb_shinfo(skb
)->frag_list
) {
1723 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1725 for (; list
; list
= list
->next
) {
1729 WARN_ON(start
> offset
+ len
);
1731 end
= start
+ list
->len
;
1732 if ((copy
= end
- offset
) > 0) {
1735 csum2
= skb_copy_and_csum_bits(list
,
1738 csum
= csum_block_add(csum
, csum2
, pos
);
1739 if ((len
-= copy
) == 0)
1752 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
1757 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1758 csstart
= skb
->csum_start
- skb_headroom(skb
);
1760 csstart
= skb_headlen(skb
);
1762 BUG_ON(csstart
> skb_headlen(skb
));
1764 skb_copy_from_linear_data(skb
, to
, csstart
);
1767 if (csstart
!= skb
->len
)
1768 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
1769 skb
->len
- csstart
, 0);
1771 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1772 long csstuff
= csstart
+ skb
->csum_offset
;
1774 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
1779 * skb_dequeue - remove from the head of the queue
1780 * @list: list to dequeue from
1782 * Remove the head of the list. The list lock is taken so the function
1783 * may be used safely with other locking list functions. The head item is
1784 * returned or %NULL if the list is empty.
1787 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
1789 unsigned long flags
;
1790 struct sk_buff
*result
;
1792 spin_lock_irqsave(&list
->lock
, flags
);
1793 result
= __skb_dequeue(list
);
1794 spin_unlock_irqrestore(&list
->lock
, flags
);
1799 * skb_dequeue_tail - remove from the tail of the queue
1800 * @list: list to dequeue from
1802 * Remove the tail of the list. The list lock is taken so the function
1803 * may be used safely with other locking list functions. The tail item is
1804 * returned or %NULL if the list is empty.
1806 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
1808 unsigned long flags
;
1809 struct sk_buff
*result
;
1811 spin_lock_irqsave(&list
->lock
, flags
);
1812 result
= __skb_dequeue_tail(list
);
1813 spin_unlock_irqrestore(&list
->lock
, flags
);
1818 * skb_queue_purge - empty a list
1819 * @list: list to empty
1821 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1822 * the list and one reference dropped. This function takes the list
1823 * lock and is atomic with respect to other list locking functions.
1825 void skb_queue_purge(struct sk_buff_head
*list
)
1827 struct sk_buff
*skb
;
1828 while ((skb
= skb_dequeue(list
)) != NULL
)
1833 * skb_queue_head - queue a buffer at the list head
1834 * @list: list to use
1835 * @newsk: buffer to queue
1837 * Queue a buffer at the start of the list. This function takes the
1838 * list lock and can be used safely with other locking &sk_buff functions
1841 * A buffer cannot be placed on two lists at the same time.
1843 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1845 unsigned long flags
;
1847 spin_lock_irqsave(&list
->lock
, flags
);
1848 __skb_queue_head(list
, newsk
);
1849 spin_unlock_irqrestore(&list
->lock
, flags
);
1853 * skb_queue_tail - queue a buffer at the list tail
1854 * @list: list to use
1855 * @newsk: buffer to queue
1857 * Queue a buffer at the tail of the list. This function takes the
1858 * list lock and can be used safely with other locking &sk_buff functions
1861 * A buffer cannot be placed on two lists at the same time.
1863 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1865 unsigned long flags
;
1867 spin_lock_irqsave(&list
->lock
, flags
);
1868 __skb_queue_tail(list
, newsk
);
1869 spin_unlock_irqrestore(&list
->lock
, flags
);
1873 * skb_unlink - remove a buffer from a list
1874 * @skb: buffer to remove
1875 * @list: list to use
1877 * Remove a packet from a list. The list locks are taken and this
1878 * function is atomic with respect to other list locked calls
1880 * You must know what list the SKB is on.
1882 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
1884 unsigned long flags
;
1886 spin_lock_irqsave(&list
->lock
, flags
);
1887 __skb_unlink(skb
, list
);
1888 spin_unlock_irqrestore(&list
->lock
, flags
);
1892 * skb_append - append a buffer
1893 * @old: buffer to insert after
1894 * @newsk: buffer to insert
1895 * @list: list to use
1897 * Place a packet after a given packet in a list. The list locks are taken
1898 * and this function is atomic with respect to other list locked calls.
1899 * A buffer cannot be placed on two lists at the same time.
1901 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
1903 unsigned long flags
;
1905 spin_lock_irqsave(&list
->lock
, flags
);
1906 __skb_queue_after(list
, old
, newsk
);
1907 spin_unlock_irqrestore(&list
->lock
, flags
);
1912 * skb_insert - insert a buffer
1913 * @old: buffer to insert before
1914 * @newsk: buffer to insert
1915 * @list: list to use
1917 * Place a packet before a given packet in a list. The list locks are
1918 * taken and this function is atomic with respect to other list locked
1921 * A buffer cannot be placed on two lists at the same time.
1923 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
1925 unsigned long flags
;
1927 spin_lock_irqsave(&list
->lock
, flags
);
1928 __skb_insert(newsk
, old
->prev
, old
, list
);
1929 spin_unlock_irqrestore(&list
->lock
, flags
);
1932 static inline void skb_split_inside_header(struct sk_buff
*skb
,
1933 struct sk_buff
* skb1
,
1934 const u32 len
, const int pos
)
1938 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
1940 /* And move data appendix as is. */
1941 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1942 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
1944 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
1945 skb_shinfo(skb
)->nr_frags
= 0;
1946 skb1
->data_len
= skb
->data_len
;
1947 skb1
->len
+= skb1
->data_len
;
1950 skb_set_tail_pointer(skb
, len
);
1953 static inline void skb_split_no_header(struct sk_buff
*skb
,
1954 struct sk_buff
* skb1
,
1955 const u32 len
, int pos
)
1958 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
1960 skb_shinfo(skb
)->nr_frags
= 0;
1961 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
1963 skb
->data_len
= len
- pos
;
1965 for (i
= 0; i
< nfrags
; i
++) {
1966 int size
= skb_shinfo(skb
)->frags
[i
].size
;
1968 if (pos
+ size
> len
) {
1969 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1973 * We have two variants in this case:
1974 * 1. Move all the frag to the second
1975 * part, if it is possible. F.e.
1976 * this approach is mandatory for TUX,
1977 * where splitting is expensive.
1978 * 2. Split is accurately. We make this.
1980 get_page(skb_shinfo(skb
)->frags
[i
].page
);
1981 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
1982 skb_shinfo(skb1
)->frags
[0].size
-= len
- pos
;
1983 skb_shinfo(skb
)->frags
[i
].size
= len
- pos
;
1984 skb_shinfo(skb
)->nr_frags
++;
1988 skb_shinfo(skb
)->nr_frags
++;
1991 skb_shinfo(skb1
)->nr_frags
= k
;
1995 * skb_split - Split fragmented skb to two parts at length len.
1996 * @skb: the buffer to split
1997 * @skb1: the buffer to receive the second part
1998 * @len: new length for skb
2000 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
2002 int pos
= skb_headlen(skb
);
2004 if (len
< pos
) /* Split line is inside header. */
2005 skb_split_inside_header(skb
, skb1
, len
, pos
);
2006 else /* Second chunk has no header, nothing to copy. */
2007 skb_split_no_header(skb
, skb1
, len
, pos
);
2010 /* Shifting from/to a cloned skb is a no-go.
2012 * Caller cannot keep skb_shinfo related pointers past calling here!
2014 static int skb_prepare_for_shift(struct sk_buff
*skb
)
2016 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2020 * skb_shift - Shifts paged data partially from skb to another
2021 * @tgt: buffer into which tail data gets added
2022 * @skb: buffer from which the paged data comes from
2023 * @shiftlen: shift up to this many bytes
2025 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2026 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2027 * It's up to caller to free skb if everything was shifted.
2029 * If @tgt runs out of frags, the whole operation is aborted.
2031 * Skb cannot include anything else but paged data while tgt is allowed
2032 * to have non-paged data as well.
2034 * TODO: full sized shift could be optimized but that would need
2035 * specialized skb free'er to handle frags without up-to-date nr_frags.
2037 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
2039 int from
, to
, merge
, todo
;
2040 struct skb_frag_struct
*fragfrom
, *fragto
;
2042 BUG_ON(shiftlen
> skb
->len
);
2043 BUG_ON(skb_headlen(skb
)); /* Would corrupt stream */
2047 to
= skb_shinfo(tgt
)->nr_frags
;
2048 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2050 /* Actual merge is delayed until the point when we know we can
2051 * commit all, so that we don't have to undo partial changes
2054 !skb_can_coalesce(tgt
, to
, fragfrom
->page
, fragfrom
->page_offset
)) {
2059 todo
-= fragfrom
->size
;
2061 if (skb_prepare_for_shift(skb
) ||
2062 skb_prepare_for_shift(tgt
))
2065 /* All previous frag pointers might be stale! */
2066 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2067 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2069 fragto
->size
+= shiftlen
;
2070 fragfrom
->size
-= shiftlen
;
2071 fragfrom
->page_offset
+= shiftlen
;
2079 /* Skip full, not-fitting skb to avoid expensive operations */
2080 if ((shiftlen
== skb
->len
) &&
2081 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
2084 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
2087 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
2088 if (to
== MAX_SKB_FRAGS
)
2091 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2092 fragto
= &skb_shinfo(tgt
)->frags
[to
];
2094 if (todo
>= fragfrom
->size
) {
2095 *fragto
= *fragfrom
;
2096 todo
-= fragfrom
->size
;
2101 get_page(fragfrom
->page
);
2102 fragto
->page
= fragfrom
->page
;
2103 fragto
->page_offset
= fragfrom
->page_offset
;
2104 fragto
->size
= todo
;
2106 fragfrom
->page_offset
+= todo
;
2107 fragfrom
->size
-= todo
;
2115 /* Ready to "commit" this state change to tgt */
2116 skb_shinfo(tgt
)->nr_frags
= to
;
2119 fragfrom
= &skb_shinfo(skb
)->frags
[0];
2120 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2122 fragto
->size
+= fragfrom
->size
;
2123 put_page(fragfrom
->page
);
2126 /* Reposition in the original skb */
2128 while (from
< skb_shinfo(skb
)->nr_frags
)
2129 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
2130 skb_shinfo(skb
)->nr_frags
= to
;
2132 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
2135 /* Most likely the tgt won't ever need its checksum anymore, skb on
2136 * the other hand might need it if it needs to be resent
2138 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
2139 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2141 /* Yak, is it really working this way? Some helper please? */
2142 skb
->len
-= shiftlen
;
2143 skb
->data_len
-= shiftlen
;
2144 skb
->truesize
-= shiftlen
;
2145 tgt
->len
+= shiftlen
;
2146 tgt
->data_len
+= shiftlen
;
2147 tgt
->truesize
+= shiftlen
;
2153 * skb_prepare_seq_read - Prepare a sequential read of skb data
2154 * @skb: the buffer to read
2155 * @from: lower offset of data to be read
2156 * @to: upper offset of data to be read
2157 * @st: state variable
2159 * Initializes the specified state variable. Must be called before
2160 * invoking skb_seq_read() for the first time.
2162 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
2163 unsigned int to
, struct skb_seq_state
*st
)
2165 st
->lower_offset
= from
;
2166 st
->upper_offset
= to
;
2167 st
->root_skb
= st
->cur_skb
= skb
;
2168 st
->frag_idx
= st
->stepped_offset
= 0;
2169 st
->frag_data
= NULL
;
2173 * skb_seq_read - Sequentially read skb data
2174 * @consumed: number of bytes consumed by the caller so far
2175 * @data: destination pointer for data to be returned
2176 * @st: state variable
2178 * Reads a block of skb data at &consumed relative to the
2179 * lower offset specified to skb_prepare_seq_read(). Assigns
2180 * the head of the data block to &data and returns the length
2181 * of the block or 0 if the end of the skb data or the upper
2182 * offset has been reached.
2184 * The caller is not required to consume all of the data
2185 * returned, i.e. &consumed is typically set to the number
2186 * of bytes already consumed and the next call to
2187 * skb_seq_read() will return the remaining part of the block.
2189 * Note 1: The size of each block of data returned can be arbitary,
2190 * this limitation is the cost for zerocopy seqeuental
2191 * reads of potentially non linear data.
2193 * Note 2: Fragment lists within fragments are not implemented
2194 * at the moment, state->root_skb could be replaced with
2195 * a stack for this purpose.
2197 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
2198 struct skb_seq_state
*st
)
2200 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
2203 if (unlikely(abs_offset
>= st
->upper_offset
))
2207 block_limit
= skb_headlen(st
->cur_skb
) + st
->stepped_offset
;
2209 if (abs_offset
< block_limit
) {
2210 *data
= st
->cur_skb
->data
+ (abs_offset
- st
->stepped_offset
);
2211 return block_limit
- abs_offset
;
2214 if (st
->frag_idx
== 0 && !st
->frag_data
)
2215 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
2217 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
2218 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
2219 block_limit
= frag
->size
+ st
->stepped_offset
;
2221 if (abs_offset
< block_limit
) {
2223 st
->frag_data
= kmap_skb_frag(frag
);
2225 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
2226 (abs_offset
- st
->stepped_offset
);
2228 return block_limit
- abs_offset
;
2231 if (st
->frag_data
) {
2232 kunmap_skb_frag(st
->frag_data
);
2233 st
->frag_data
= NULL
;
2237 st
->stepped_offset
+= frag
->size
;
2240 if (st
->frag_data
) {
2241 kunmap_skb_frag(st
->frag_data
);
2242 st
->frag_data
= NULL
;
2245 if (st
->root_skb
== st
->cur_skb
&&
2246 skb_shinfo(st
->root_skb
)->frag_list
) {
2247 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
2250 } else if (st
->cur_skb
->next
) {
2251 st
->cur_skb
= st
->cur_skb
->next
;
2260 * skb_abort_seq_read - Abort a sequential read of skb data
2261 * @st: state variable
2263 * Must be called if skb_seq_read() was not called until it
2266 void skb_abort_seq_read(struct skb_seq_state
*st
)
2269 kunmap_skb_frag(st
->frag_data
);
2272 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2274 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
2275 struct ts_config
*conf
,
2276 struct ts_state
*state
)
2278 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
2281 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
2283 skb_abort_seq_read(TS_SKB_CB(state
));
2287 * skb_find_text - Find a text pattern in skb data
2288 * @skb: the buffer to look in
2289 * @from: search offset
2291 * @config: textsearch configuration
2292 * @state: uninitialized textsearch state variable
2294 * Finds a pattern in the skb data according to the specified
2295 * textsearch configuration. Use textsearch_next() to retrieve
2296 * subsequent occurrences of the pattern. Returns the offset
2297 * to the first occurrence or UINT_MAX if no match was found.
2299 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
2300 unsigned int to
, struct ts_config
*config
,
2301 struct ts_state
*state
)
2305 config
->get_next_block
= skb_ts_get_next_block
;
2306 config
->finish
= skb_ts_finish
;
2308 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
2310 ret
= textsearch_find(config
, state
);
2311 return (ret
<= to
- from
? ret
: UINT_MAX
);
2315 * skb_append_datato_frags: - append the user data to a skb
2316 * @sk: sock structure
2317 * @skb: skb structure to be appened with user data.
2318 * @getfrag: call back function to be used for getting the user data
2319 * @from: pointer to user message iov
2320 * @length: length of the iov message
2322 * Description: This procedure append the user data in the fragment part
2323 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2325 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
2326 int (*getfrag
)(void *from
, char *to
, int offset
,
2327 int len
, int odd
, struct sk_buff
*skb
),
2328 void *from
, int length
)
2331 skb_frag_t
*frag
= NULL
;
2332 struct page
*page
= NULL
;
2338 /* Return error if we don't have space for new frag */
2339 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2340 if (frg_cnt
>= MAX_SKB_FRAGS
)
2343 /* allocate a new page for next frag */
2344 page
= alloc_pages(sk
->sk_allocation
, 0);
2346 /* If alloc_page fails just return failure and caller will
2347 * free previous allocated pages by doing kfree_skb()
2352 /* initialize the next frag */
2353 sk
->sk_sndmsg_page
= page
;
2354 sk
->sk_sndmsg_off
= 0;
2355 skb_fill_page_desc(skb
, frg_cnt
, page
, 0, 0);
2356 skb
->truesize
+= PAGE_SIZE
;
2357 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
2359 /* get the new initialized frag */
2360 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2361 frag
= &skb_shinfo(skb
)->frags
[frg_cnt
- 1];
2363 /* copy the user data to page */
2364 left
= PAGE_SIZE
- frag
->page_offset
;
2365 copy
= (length
> left
)? left
: length
;
2367 ret
= getfrag(from
, (page_address(frag
->page
) +
2368 frag
->page_offset
+ frag
->size
),
2369 offset
, copy
, 0, skb
);
2373 /* copy was successful so update the size parameters */
2374 sk
->sk_sndmsg_off
+= copy
;
2377 skb
->data_len
+= copy
;
2381 } while (length
> 0);
2387 * skb_pull_rcsum - pull skb and update receive checksum
2388 * @skb: buffer to update
2389 * @len: length of data pulled
2391 * This function performs an skb_pull on the packet and updates
2392 * the CHECKSUM_COMPLETE checksum. It should be used on
2393 * receive path processing instead of skb_pull unless you know
2394 * that the checksum difference is zero (e.g., a valid IP header)
2395 * or you are setting ip_summed to CHECKSUM_NONE.
2397 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
2399 BUG_ON(len
> skb
->len
);
2401 BUG_ON(skb
->len
< skb
->data_len
);
2402 skb_postpull_rcsum(skb
, skb
->data
, len
);
2403 return skb
->data
+= len
;
2406 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
2409 * skb_segment - Perform protocol segmentation on skb.
2410 * @skb: buffer to segment
2411 * @features: features for the output path (see dev->features)
2413 * This function performs segmentation on the given skb. It returns
2414 * a pointer to the first in a list of new skbs for the segments.
2415 * In case of error it returns ERR_PTR(err).
2417 struct sk_buff
*skb_segment(struct sk_buff
*skb
, int features
)
2419 struct sk_buff
*segs
= NULL
;
2420 struct sk_buff
*tail
= NULL
;
2421 struct sk_buff
*fskb
= skb_shinfo(skb
)->frag_list
;
2422 unsigned int mss
= skb_shinfo(skb
)->gso_size
;
2423 unsigned int doffset
= skb
->data
- skb_mac_header(skb
);
2424 unsigned int offset
= doffset
;
2425 unsigned int headroom
;
2427 int sg
= features
& NETIF_F_SG
;
2428 int nfrags
= skb_shinfo(skb
)->nr_frags
;
2433 __skb_push(skb
, doffset
);
2434 headroom
= skb_headroom(skb
);
2435 pos
= skb_headlen(skb
);
2438 struct sk_buff
*nskb
;
2443 len
= skb
->len
- offset
;
2447 hsize
= skb_headlen(skb
) - offset
;
2450 if (hsize
> len
|| !sg
)
2453 if (!hsize
&& i
>= nfrags
) {
2454 BUG_ON(fskb
->len
!= len
);
2457 nskb
= skb_clone(fskb
, GFP_ATOMIC
);
2460 if (unlikely(!nskb
))
2463 hsize
= skb_end_pointer(nskb
) - nskb
->head
;
2464 if (skb_cow_head(nskb
, doffset
+ headroom
)) {
2469 nskb
->truesize
+= skb_end_pointer(nskb
) - nskb
->head
-
2471 skb_release_head_state(nskb
);
2472 __skb_push(nskb
, doffset
);
2474 nskb
= alloc_skb(hsize
+ doffset
+ headroom
,
2477 if (unlikely(!nskb
))
2480 skb_reserve(nskb
, headroom
);
2481 __skb_put(nskb
, doffset
);
2490 __copy_skb_header(nskb
, skb
);
2491 nskb
->mac_len
= skb
->mac_len
;
2493 skb_reset_mac_header(nskb
);
2494 skb_set_network_header(nskb
, skb
->mac_len
);
2495 nskb
->transport_header
= (nskb
->network_header
+
2496 skb_network_header_len(skb
));
2497 skb_copy_from_linear_data(skb
, nskb
->data
, doffset
);
2499 if (pos
>= offset
+ len
)
2503 nskb
->ip_summed
= CHECKSUM_NONE
;
2504 nskb
->csum
= skb_copy_and_csum_bits(skb
, offset
,
2510 frag
= skb_shinfo(nskb
)->frags
;
2512 skb_copy_from_linear_data_offset(skb
, offset
,
2513 skb_put(nskb
, hsize
), hsize
);
2515 while (pos
< offset
+ len
&& i
< nfrags
) {
2516 *frag
= skb_shinfo(skb
)->frags
[i
];
2517 get_page(frag
->page
);
2521 frag
->page_offset
+= offset
- pos
;
2522 frag
->size
-= offset
- pos
;
2525 skb_shinfo(nskb
)->nr_frags
++;
2527 if (pos
+ size
<= offset
+ len
) {
2531 frag
->size
-= pos
+ size
- (offset
+ len
);
2538 if (pos
< offset
+ len
) {
2539 struct sk_buff
*fskb2
= fskb
;
2541 BUG_ON(pos
+ fskb
->len
!= offset
+ len
);
2547 fskb2
= skb_clone(fskb2
, GFP_ATOMIC
);
2553 BUG_ON(skb_shinfo(nskb
)->frag_list
);
2554 skb_shinfo(nskb
)->frag_list
= fskb2
;
2558 nskb
->data_len
= len
- hsize
;
2559 nskb
->len
+= nskb
->data_len
;
2560 nskb
->truesize
+= nskb
->data_len
;
2561 } while ((offset
+= len
) < skb
->len
);
2566 while ((skb
= segs
)) {
2570 return ERR_PTR(err
);
2573 EXPORT_SYMBOL_GPL(skb_segment
);
2575 int skb_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2577 struct sk_buff
*p
= *head
;
2578 struct sk_buff
*nskb
;
2579 unsigned int headroom
;
2580 unsigned int hlen
= p
->data
- skb_mac_header(p
);
2581 unsigned int len
= skb
->len
;
2583 if (hlen
+ p
->len
+ len
>= 65536)
2586 if (skb_shinfo(p
)->frag_list
)
2588 else if (!skb_headlen(p
) && !skb_headlen(skb
) &&
2589 skb_shinfo(p
)->nr_frags
+ skb_shinfo(skb
)->nr_frags
<
2591 memcpy(skb_shinfo(p
)->frags
+ skb_shinfo(p
)->nr_frags
,
2592 skb_shinfo(skb
)->frags
,
2593 skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
2595 skb_shinfo(p
)->nr_frags
+= skb_shinfo(skb
)->nr_frags
;
2596 skb_shinfo(skb
)->nr_frags
= 0;
2598 skb
->truesize
-= skb
->data_len
;
2599 skb
->len
-= skb
->data_len
;
2602 NAPI_GRO_CB(skb
)->free
= 1;
2606 headroom
= skb_headroom(p
);
2607 nskb
= netdev_alloc_skb(p
->dev
, headroom
);
2608 if (unlikely(!nskb
))
2611 __copy_skb_header(nskb
, p
);
2612 nskb
->mac_len
= p
->mac_len
;
2614 skb_reserve(nskb
, headroom
);
2616 skb_set_mac_header(nskb
, -hlen
);
2617 skb_set_network_header(nskb
, skb_network_offset(p
));
2618 skb_set_transport_header(nskb
, skb_transport_offset(p
));
2620 memcpy(skb_mac_header(nskb
), skb_mac_header(p
), hlen
);
2622 *NAPI_GRO_CB(nskb
) = *NAPI_GRO_CB(p
);
2623 skb_shinfo(nskb
)->frag_list
= p
;
2624 skb_shinfo(nskb
)->gso_size
= skb_shinfo(p
)->gso_size
;
2625 skb_header_release(p
);
2628 nskb
->data_len
+= p
->len
;
2629 nskb
->truesize
+= p
->len
;
2630 nskb
->len
+= p
->len
;
2633 nskb
->next
= p
->next
;
2639 p
->prev
->next
= skb
;
2641 skb_header_release(skb
);
2644 NAPI_GRO_CB(p
)->count
++;
2649 NAPI_GRO_CB(skb
)->same_flow
= 1;
2652 EXPORT_SYMBOL_GPL(skb_gro_receive
);
2654 void __init
skb_init(void)
2656 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
2657 sizeof(struct sk_buff
),
2659 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2661 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
2662 (2*sizeof(struct sk_buff
)) +
2665 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2670 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2671 * @skb: Socket buffer containing the buffers to be mapped
2672 * @sg: The scatter-gather list to map into
2673 * @offset: The offset into the buffer's contents to start mapping
2674 * @len: Length of buffer space to be mapped
2676 * Fill the specified scatter-gather list with mappings/pointers into a
2677 * region of the buffer space attached to a socket buffer.
2680 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2682 int start
= skb_headlen(skb
);
2683 int i
, copy
= start
- offset
;
2689 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
2691 if ((len
-= copy
) == 0)
2696 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2699 WARN_ON(start
> offset
+ len
);
2701 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
2702 if ((copy
= end
- offset
) > 0) {
2703 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2707 sg_set_page(&sg
[elt
], frag
->page
, copy
,
2708 frag
->page_offset
+offset
-start
);
2717 if (skb_shinfo(skb
)->frag_list
) {
2718 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
2720 for (; list
; list
= list
->next
) {
2723 WARN_ON(start
> offset
+ len
);
2725 end
= start
+ list
->len
;
2726 if ((copy
= end
- offset
) > 0) {
2729 elt
+= __skb_to_sgvec(list
, sg
+elt
, offset
- start
,
2731 if ((len
-= copy
) == 0)
2742 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2744 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
);
2746 sg_mark_end(&sg
[nsg
- 1]);
2752 * skb_cow_data - Check that a socket buffer's data buffers are writable
2753 * @skb: The socket buffer to check.
2754 * @tailbits: Amount of trailing space to be added
2755 * @trailer: Returned pointer to the skb where the @tailbits space begins
2757 * Make sure that the data buffers attached to a socket buffer are
2758 * writable. If they are not, private copies are made of the data buffers
2759 * and the socket buffer is set to use these instead.
2761 * If @tailbits is given, make sure that there is space to write @tailbits
2762 * bytes of data beyond current end of socket buffer. @trailer will be
2763 * set to point to the skb in which this space begins.
2765 * The number of scatterlist elements required to completely map the
2766 * COW'd and extended socket buffer will be returned.
2768 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
2772 struct sk_buff
*skb1
, **skb_p
;
2774 /* If skb is cloned or its head is paged, reallocate
2775 * head pulling out all the pages (pages are considered not writable
2776 * at the moment even if they are anonymous).
2778 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
2779 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
2782 /* Easy case. Most of packets will go this way. */
2783 if (!skb_shinfo(skb
)->frag_list
) {
2784 /* A little of trouble, not enough of space for trailer.
2785 * This should not happen, when stack is tuned to generate
2786 * good frames. OK, on miss we reallocate and reserve even more
2787 * space, 128 bytes is fair. */
2789 if (skb_tailroom(skb
) < tailbits
&&
2790 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
2798 /* Misery. We are in troubles, going to mincer fragments... */
2801 skb_p
= &skb_shinfo(skb
)->frag_list
;
2804 while ((skb1
= *skb_p
) != NULL
) {
2807 /* The fragment is partially pulled by someone,
2808 * this can happen on input. Copy it and everything
2811 if (skb_shared(skb1
))
2814 /* If the skb is the last, worry about trailer. */
2816 if (skb1
->next
== NULL
&& tailbits
) {
2817 if (skb_shinfo(skb1
)->nr_frags
||
2818 skb_shinfo(skb1
)->frag_list
||
2819 skb_tailroom(skb1
) < tailbits
)
2820 ntail
= tailbits
+ 128;
2826 skb_shinfo(skb1
)->nr_frags
||
2827 skb_shinfo(skb1
)->frag_list
) {
2828 struct sk_buff
*skb2
;
2830 /* Fuck, we are miserable poor guys... */
2832 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
2834 skb2
= skb_copy_expand(skb1
,
2838 if (unlikely(skb2
== NULL
))
2842 skb_set_owner_w(skb2
, skb1
->sk
);
2844 /* Looking around. Are we still alive?
2845 * OK, link new skb, drop old one */
2847 skb2
->next
= skb1
->next
;
2854 skb_p
= &skb1
->next
;
2861 * skb_partial_csum_set - set up and verify partial csum values for packet
2862 * @skb: the skb to set
2863 * @start: the number of bytes after skb->data to start checksumming.
2864 * @off: the offset from start to place the checksum.
2866 * For untrusted partially-checksummed packets, we need to make sure the values
2867 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
2869 * This function checks and sets those values and skb->ip_summed: if this
2870 * returns false you should drop the packet.
2872 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
2874 if (unlikely(start
> skb
->len
- 2) ||
2875 unlikely((int)start
+ off
> skb
->len
- 2)) {
2876 if (net_ratelimit())
2878 "bad partial csum: csum=%u/%u len=%u\n",
2879 start
, off
, skb
->len
);
2882 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2883 skb
->csum_start
= skb_headroom(skb
) + start
;
2884 skb
->csum_offset
= off
;
2888 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
2890 if (net_ratelimit())
2891 pr_warning("%s: received packets cannot be forwarded"
2892 " while LRO is enabled\n", skb
->dev
->name
);
2895 EXPORT_SYMBOL(___pskb_trim
);
2896 EXPORT_SYMBOL(__kfree_skb
);
2897 EXPORT_SYMBOL(kfree_skb
);
2898 EXPORT_SYMBOL(__pskb_pull_tail
);
2899 EXPORT_SYMBOL(__alloc_skb
);
2900 EXPORT_SYMBOL(__netdev_alloc_skb
);
2901 EXPORT_SYMBOL(pskb_copy
);
2902 EXPORT_SYMBOL(pskb_expand_head
);
2903 EXPORT_SYMBOL(skb_checksum
);
2904 EXPORT_SYMBOL(skb_clone
);
2905 EXPORT_SYMBOL(skb_copy
);
2906 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
2907 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
2908 EXPORT_SYMBOL(skb_copy_bits
);
2909 EXPORT_SYMBOL(skb_copy_expand
);
2910 EXPORT_SYMBOL(skb_over_panic
);
2911 EXPORT_SYMBOL(skb_pad
);
2912 EXPORT_SYMBOL(skb_realloc_headroom
);
2913 EXPORT_SYMBOL(skb_under_panic
);
2914 EXPORT_SYMBOL(skb_dequeue
);
2915 EXPORT_SYMBOL(skb_dequeue_tail
);
2916 EXPORT_SYMBOL(skb_insert
);
2917 EXPORT_SYMBOL(skb_queue_purge
);
2918 EXPORT_SYMBOL(skb_queue_head
);
2919 EXPORT_SYMBOL(skb_queue_tail
);
2920 EXPORT_SYMBOL(skb_unlink
);
2921 EXPORT_SYMBOL(skb_append
);
2922 EXPORT_SYMBOL(skb_split
);
2923 EXPORT_SYMBOL(skb_prepare_seq_read
);
2924 EXPORT_SYMBOL(skb_seq_read
);
2925 EXPORT_SYMBOL(skb_abort_seq_read
);
2926 EXPORT_SYMBOL(skb_find_text
);
2927 EXPORT_SYMBOL(skb_append_datato_frags
);
2928 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);
2930 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
2931 EXPORT_SYMBOL_GPL(skb_cow_data
);
2932 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);