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
)
76 struct sk_buff
*skb
= (struct sk_buff
*) buf
->private;
81 static void sock_pipe_buf_get(struct pipe_inode_info
*pipe
,
82 struct pipe_buffer
*buf
)
84 struct sk_buff
*skb
= (struct sk_buff
*) buf
->private;
89 static int sock_pipe_buf_steal(struct pipe_inode_info
*pipe
,
90 struct pipe_buffer
*buf
)
96 /* Pipe buffer operations for a socket. */
97 static struct pipe_buf_operations sock_pipe_buf_ops
= {
99 .map
= generic_pipe_buf_map
,
100 .unmap
= generic_pipe_buf_unmap
,
101 .confirm
= generic_pipe_buf_confirm
,
102 .release
= sock_pipe_buf_release
,
103 .steal
= sock_pipe_buf_steal
,
104 .get
= sock_pipe_buf_get
,
108 * Keep out-of-line to prevent kernel bloat.
109 * __builtin_return_address is not used because it is not always
114 * skb_over_panic - private function
119 * Out of line support code for skb_put(). Not user callable.
121 void skb_over_panic(struct sk_buff
*skb
, int sz
, void *here
)
123 printk(KERN_EMERG
"skb_over_panic: text:%p len:%d put:%d head:%p "
124 "data:%p tail:%#lx end:%#lx dev:%s\n",
125 here
, skb
->len
, sz
, skb
->head
, skb
->data
,
126 (unsigned long)skb
->tail
, (unsigned long)skb
->end
,
127 skb
->dev
? skb
->dev
->name
: "<NULL>");
132 * skb_under_panic - private function
137 * Out of line support code for skb_push(). Not user callable.
140 void skb_under_panic(struct sk_buff
*skb
, int sz
, void *here
)
142 printk(KERN_EMERG
"skb_under_panic: text:%p len:%d put:%d head:%p "
143 "data:%p tail:%#lx end:%#lx dev:%s\n",
144 here
, skb
->len
, sz
, skb
->head
, skb
->data
,
145 (unsigned long)skb
->tail
, (unsigned long)skb
->end
,
146 skb
->dev
? skb
->dev
->name
: "<NULL>");
150 void skb_truesize_bug(struct sk_buff
*skb
)
152 WARN(net_ratelimit(), KERN_ERR
"SKB BUG: Invalid truesize (%u) "
153 "len=%u, sizeof(sk_buff)=%Zd\n",
154 skb
->truesize
, skb
->len
, sizeof(struct sk_buff
));
156 EXPORT_SYMBOL(skb_truesize_bug
);
158 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
159 * 'private' fields and also do memory statistics to find all the
165 * __alloc_skb - allocate a network buffer
166 * @size: size to allocate
167 * @gfp_mask: allocation mask
168 * @fclone: allocate from fclone cache instead of head cache
169 * and allocate a cloned (child) skb
170 * @node: numa node to allocate memory on
172 * Allocate a new &sk_buff. The returned buffer has no headroom and a
173 * tail room of size bytes. The object has a reference count of one.
174 * The return is the buffer. On a failure the return is %NULL.
176 * Buffers may only be allocated from interrupts using a @gfp_mask of
179 struct sk_buff
*__alloc_skb(unsigned int size
, gfp_t gfp_mask
,
180 int fclone
, int node
)
182 struct kmem_cache
*cache
;
183 struct skb_shared_info
*shinfo
;
187 cache
= fclone
? skbuff_fclone_cache
: skbuff_head_cache
;
190 skb
= kmem_cache_alloc_node(cache
, gfp_mask
& ~__GFP_DMA
, node
);
194 size
= SKB_DATA_ALIGN(size
);
195 data
= kmalloc_node_track_caller(size
+ sizeof(struct skb_shared_info
),
201 * Only clear those fields we need to clear, not those that we will
202 * actually initialise below. Hence, don't put any more fields after
203 * the tail pointer in struct sk_buff!
205 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
206 skb
->truesize
= size
+ sizeof(struct sk_buff
);
207 atomic_set(&skb
->users
, 1);
210 skb_reset_tail_pointer(skb
);
211 skb
->end
= skb
->tail
+ size
;
212 /* make sure we initialize shinfo sequentially */
213 shinfo
= skb_shinfo(skb
);
214 atomic_set(&shinfo
->dataref
, 1);
215 shinfo
->nr_frags
= 0;
216 shinfo
->gso_size
= 0;
217 shinfo
->gso_segs
= 0;
218 shinfo
->gso_type
= 0;
219 shinfo
->ip6_frag_id
= 0;
220 shinfo
->frag_list
= NULL
;
223 struct sk_buff
*child
= skb
+ 1;
224 atomic_t
*fclone_ref
= (atomic_t
*) (child
+ 1);
226 skb
->fclone
= SKB_FCLONE_ORIG
;
227 atomic_set(fclone_ref
, 1);
229 child
->fclone
= SKB_FCLONE_UNAVAILABLE
;
234 kmem_cache_free(cache
, skb
);
240 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
241 * @dev: network device to receive on
242 * @length: length to allocate
243 * @gfp_mask: get_free_pages mask, passed to alloc_skb
245 * Allocate a new &sk_buff and assign it a usage count of one. The
246 * buffer has unspecified headroom built in. Users should allocate
247 * the headroom they think they need without accounting for the
248 * built in space. The built in space is used for optimisations.
250 * %NULL is returned if there is no free memory.
252 struct sk_buff
*__netdev_alloc_skb(struct net_device
*dev
,
253 unsigned int length
, gfp_t gfp_mask
)
255 int node
= dev
->dev
.parent
? dev_to_node(dev
->dev
.parent
) : -1;
258 skb
= __alloc_skb(length
+ NET_SKB_PAD
, gfp_mask
, 0, node
);
260 skb_reserve(skb
, NET_SKB_PAD
);
266 struct page
*__netdev_alloc_page(struct net_device
*dev
, gfp_t gfp_mask
)
268 int node
= dev
->dev
.parent
? dev_to_node(dev
->dev
.parent
) : -1;
271 page
= alloc_pages_node(node
, gfp_mask
, 0);
274 EXPORT_SYMBOL(__netdev_alloc_page
);
276 void skb_add_rx_frag(struct sk_buff
*skb
, int i
, struct page
*page
, int off
,
279 skb_fill_page_desc(skb
, i
, page
, off
, size
);
281 skb
->data_len
+= size
;
282 skb
->truesize
+= size
;
284 EXPORT_SYMBOL(skb_add_rx_frag
);
287 * dev_alloc_skb - allocate an skbuff for receiving
288 * @length: length to allocate
290 * Allocate a new &sk_buff and assign it a usage count of one. The
291 * buffer has unspecified headroom built in. Users should allocate
292 * the headroom they think they need without accounting for the
293 * built in space. The built in space is used for optimisations.
295 * %NULL is returned if there is no free memory. Although this function
296 * allocates memory it can be called from an interrupt.
298 struct sk_buff
*dev_alloc_skb(unsigned int length
)
301 * There is more code here than it seems:
302 * __dev_alloc_skb is an inline
304 return __dev_alloc_skb(length
, GFP_ATOMIC
);
306 EXPORT_SYMBOL(dev_alloc_skb
);
308 static void skb_drop_list(struct sk_buff
**listp
)
310 struct sk_buff
*list
= *listp
;
315 struct sk_buff
*this = list
;
321 static inline void skb_drop_fraglist(struct sk_buff
*skb
)
323 skb_drop_list(&skb_shinfo(skb
)->frag_list
);
326 static void skb_clone_fraglist(struct sk_buff
*skb
)
328 struct sk_buff
*list
;
330 for (list
= skb_shinfo(skb
)->frag_list
; list
; list
= list
->next
)
334 static void skb_release_data(struct sk_buff
*skb
)
337 !atomic_sub_return(skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1,
338 &skb_shinfo(skb
)->dataref
)) {
339 if (skb_shinfo(skb
)->nr_frags
) {
341 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
342 put_page(skb_shinfo(skb
)->frags
[i
].page
);
345 if (skb_shinfo(skb
)->frag_list
)
346 skb_drop_fraglist(skb
);
353 * Free an skbuff by memory without cleaning the state.
355 static void kfree_skbmem(struct sk_buff
*skb
)
357 struct sk_buff
*other
;
358 atomic_t
*fclone_ref
;
360 switch (skb
->fclone
) {
361 case SKB_FCLONE_UNAVAILABLE
:
362 kmem_cache_free(skbuff_head_cache
, skb
);
365 case SKB_FCLONE_ORIG
:
366 fclone_ref
= (atomic_t
*) (skb
+ 2);
367 if (atomic_dec_and_test(fclone_ref
))
368 kmem_cache_free(skbuff_fclone_cache
, skb
);
371 case SKB_FCLONE_CLONE
:
372 fclone_ref
= (atomic_t
*) (skb
+ 1);
375 /* The clone portion is available for
376 * fast-cloning again.
378 skb
->fclone
= SKB_FCLONE_UNAVAILABLE
;
380 if (atomic_dec_and_test(fclone_ref
))
381 kmem_cache_free(skbuff_fclone_cache
, other
);
386 static void skb_release_head_state(struct sk_buff
*skb
)
388 dst_release(skb
->dst
);
390 secpath_put(skb
->sp
);
392 if (skb
->destructor
) {
394 skb
->destructor(skb
);
396 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
397 nf_conntrack_put(skb
->nfct
);
398 nf_conntrack_put_reasm(skb
->nfct_reasm
);
400 #ifdef CONFIG_BRIDGE_NETFILTER
401 nf_bridge_put(skb
->nf_bridge
);
403 /* XXX: IS this still necessary? - JHS */
404 #ifdef CONFIG_NET_SCHED
406 #ifdef CONFIG_NET_CLS_ACT
412 /* Free everything but the sk_buff shell. */
413 static void skb_release_all(struct sk_buff
*skb
)
415 skb_release_head_state(skb
);
416 skb_release_data(skb
);
420 * __kfree_skb - private function
423 * Free an sk_buff. Release anything attached to the buffer.
424 * Clean the state. This is an internal helper function. Users should
425 * always call kfree_skb
428 void __kfree_skb(struct sk_buff
*skb
)
430 skb_release_all(skb
);
435 * kfree_skb - free an sk_buff
436 * @skb: buffer to free
438 * Drop a reference to the buffer and free it if the usage count has
441 void kfree_skb(struct sk_buff
*skb
)
445 if (likely(atomic_read(&skb
->users
) == 1))
447 else if (likely(!atomic_dec_and_test(&skb
->users
)))
453 * skb_recycle_check - check if skb can be reused for receive
455 * @skb_size: minimum receive buffer size
457 * Checks that the skb passed in is not shared or cloned, and
458 * that it is linear and its head portion at least as large as
459 * skb_size so that it can be recycled as a receive buffer.
460 * If these conditions are met, this function does any necessary
461 * reference count dropping and cleans up the skbuff as if it
462 * just came from __alloc_skb().
464 int skb_recycle_check(struct sk_buff
*skb
, int skb_size
)
466 struct skb_shared_info
*shinfo
;
468 if (skb_is_nonlinear(skb
) || skb
->fclone
!= SKB_FCLONE_UNAVAILABLE
)
471 skb_size
= SKB_DATA_ALIGN(skb_size
+ NET_SKB_PAD
);
472 if (skb_end_pointer(skb
) - skb
->head
< skb_size
)
475 if (skb_shared(skb
) || skb_cloned(skb
))
478 skb_release_head_state(skb
);
479 shinfo
= skb_shinfo(skb
);
480 atomic_set(&shinfo
->dataref
, 1);
481 shinfo
->nr_frags
= 0;
482 shinfo
->gso_size
= 0;
483 shinfo
->gso_segs
= 0;
484 shinfo
->gso_type
= 0;
485 shinfo
->ip6_frag_id
= 0;
486 shinfo
->frag_list
= NULL
;
488 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
489 skb
->data
= skb
->head
+ NET_SKB_PAD
;
490 skb_reset_tail_pointer(skb
);
494 EXPORT_SYMBOL(skb_recycle_check
);
496 static void __copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
498 new->tstamp
= old
->tstamp
;
500 new->transport_header
= old
->transport_header
;
501 new->network_header
= old
->network_header
;
502 new->mac_header
= old
->mac_header
;
503 new->dst
= dst_clone(old
->dst
);
505 new->sp
= secpath_get(old
->sp
);
507 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
508 new->csum_start
= old
->csum_start
;
509 new->csum_offset
= old
->csum_offset
;
510 new->local_df
= old
->local_df
;
511 new->pkt_type
= old
->pkt_type
;
512 new->ip_summed
= old
->ip_summed
;
513 skb_copy_queue_mapping(new, old
);
514 new->priority
= old
->priority
;
515 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
516 new->ipvs_property
= old
->ipvs_property
;
518 new->protocol
= old
->protocol
;
519 new->mark
= old
->mark
;
521 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
522 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
523 new->nf_trace
= old
->nf_trace
;
525 #ifdef CONFIG_NET_SCHED
526 new->tc_index
= old
->tc_index
;
527 #ifdef CONFIG_NET_CLS_ACT
528 new->tc_verd
= old
->tc_verd
;
531 new->vlan_tci
= old
->vlan_tci
;
533 skb_copy_secmark(new, old
);
536 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
538 #define C(x) n->x = skb->x
540 n
->next
= n
->prev
= NULL
;
542 __copy_skb_header(n
, skb
);
547 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
550 n
->destructor
= NULL
;
557 #if defined(CONFIG_MAC80211) || defined(CONFIG_MAC80211_MODULE)
561 atomic_set(&n
->users
, 1);
563 atomic_inc(&(skb_shinfo(skb
)->dataref
));
571 * skb_morph - morph one skb into another
572 * @dst: the skb to receive the contents
573 * @src: the skb to supply the contents
575 * This is identical to skb_clone except that the target skb is
576 * supplied by the user.
578 * The target skb is returned upon exit.
580 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
582 skb_release_all(dst
);
583 return __skb_clone(dst
, src
);
585 EXPORT_SYMBOL_GPL(skb_morph
);
588 * skb_clone - duplicate an sk_buff
589 * @skb: buffer to clone
590 * @gfp_mask: allocation priority
592 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
593 * copies share the same packet data but not structure. The new
594 * buffer has a reference count of 1. If the allocation fails the
595 * function returns %NULL otherwise the new buffer is returned.
597 * If this function is called from an interrupt gfp_mask() must be
601 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
606 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
607 n
->fclone
== SKB_FCLONE_UNAVAILABLE
) {
608 atomic_t
*fclone_ref
= (atomic_t
*) (n
+ 1);
609 n
->fclone
= SKB_FCLONE_CLONE
;
610 atomic_inc(fclone_ref
);
612 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
615 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
618 return __skb_clone(n
, skb
);
621 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
623 #ifndef NET_SKBUFF_DATA_USES_OFFSET
625 * Shift between the two data areas in bytes
627 unsigned long offset
= new->data
- old
->data
;
630 __copy_skb_header(new, old
);
632 #ifndef NET_SKBUFF_DATA_USES_OFFSET
633 /* {transport,network,mac}_header are relative to skb->head */
634 new->transport_header
+= offset
;
635 new->network_header
+= offset
;
636 new->mac_header
+= offset
;
638 skb_shinfo(new)->gso_size
= skb_shinfo(old
)->gso_size
;
639 skb_shinfo(new)->gso_segs
= skb_shinfo(old
)->gso_segs
;
640 skb_shinfo(new)->gso_type
= skb_shinfo(old
)->gso_type
;
644 * skb_copy - create private copy of an sk_buff
645 * @skb: buffer to copy
646 * @gfp_mask: allocation priority
648 * Make a copy of both an &sk_buff and its data. This is used when the
649 * caller wishes to modify the data and needs a private copy of the
650 * data to alter. Returns %NULL on failure or the pointer to the buffer
651 * on success. The returned buffer has a reference count of 1.
653 * As by-product this function converts non-linear &sk_buff to linear
654 * one, so that &sk_buff becomes completely private and caller is allowed
655 * to modify all the data of returned buffer. This means that this
656 * function is not recommended for use in circumstances when only
657 * header is going to be modified. Use pskb_copy() instead.
660 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
662 int headerlen
= skb
->data
- skb
->head
;
664 * Allocate the copy buffer
667 #ifdef NET_SKBUFF_DATA_USES_OFFSET
668 n
= alloc_skb(skb
->end
+ skb
->data_len
, gfp_mask
);
670 n
= alloc_skb(skb
->end
- skb
->head
+ skb
->data_len
, gfp_mask
);
675 /* Set the data pointer */
676 skb_reserve(n
, headerlen
);
677 /* Set the tail pointer and length */
678 skb_put(n
, skb
->len
);
680 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
683 copy_skb_header(n
, skb
);
689 * pskb_copy - create copy of an sk_buff with private head.
690 * @skb: buffer to copy
691 * @gfp_mask: allocation priority
693 * Make a copy of both an &sk_buff and part of its data, located
694 * in header. Fragmented data remain shared. This is used when
695 * the caller wishes to modify only header of &sk_buff and needs
696 * private copy of the header to alter. Returns %NULL on failure
697 * or the pointer to the buffer on success.
698 * The returned buffer has a reference count of 1.
701 struct sk_buff
*pskb_copy(struct sk_buff
*skb
, gfp_t gfp_mask
)
704 * Allocate the copy buffer
707 #ifdef NET_SKBUFF_DATA_USES_OFFSET
708 n
= alloc_skb(skb
->end
, gfp_mask
);
710 n
= alloc_skb(skb
->end
- skb
->head
, gfp_mask
);
715 /* Set the data pointer */
716 skb_reserve(n
, skb
->data
- skb
->head
);
717 /* Set the tail pointer and length */
718 skb_put(n
, skb_headlen(skb
));
720 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
722 n
->truesize
+= skb
->data_len
;
723 n
->data_len
= skb
->data_len
;
726 if (skb_shinfo(skb
)->nr_frags
) {
729 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
730 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
731 get_page(skb_shinfo(n
)->frags
[i
].page
);
733 skb_shinfo(n
)->nr_frags
= i
;
736 if (skb_shinfo(skb
)->frag_list
) {
737 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
738 skb_clone_fraglist(n
);
741 copy_skb_header(n
, skb
);
747 * pskb_expand_head - reallocate header of &sk_buff
748 * @skb: buffer to reallocate
749 * @nhead: room to add at head
750 * @ntail: room to add at tail
751 * @gfp_mask: allocation priority
753 * Expands (or creates identical copy, if &nhead and &ntail are zero)
754 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
755 * reference count of 1. Returns zero in the case of success or error,
756 * if expansion failed. In the last case, &sk_buff is not changed.
758 * All the pointers pointing into skb header may change and must be
759 * reloaded after call to this function.
762 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
767 #ifdef NET_SKBUFF_DATA_USES_OFFSET
768 int size
= nhead
+ skb
->end
+ ntail
;
770 int size
= nhead
+ (skb
->end
- skb
->head
) + ntail
;
779 size
= SKB_DATA_ALIGN(size
);
781 data
= kmalloc(size
+ sizeof(struct skb_shared_info
), gfp_mask
);
785 /* Copy only real data... and, alas, header. This should be
786 * optimized for the cases when header is void. */
787 #ifdef NET_SKBUFF_DATA_USES_OFFSET
788 memcpy(data
+ nhead
, skb
->head
, skb
->tail
);
790 memcpy(data
+ nhead
, skb
->head
, skb
->tail
- skb
->head
);
792 memcpy(data
+ size
, skb_end_pointer(skb
),
793 sizeof(struct skb_shared_info
));
795 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
796 get_page(skb_shinfo(skb
)->frags
[i
].page
);
798 if (skb_shinfo(skb
)->frag_list
)
799 skb_clone_fraglist(skb
);
801 skb_release_data(skb
);
803 off
= (data
+ nhead
) - skb
->head
;
807 #ifdef NET_SKBUFF_DATA_USES_OFFSET
811 skb
->end
= skb
->head
+ size
;
813 /* {transport,network,mac}_header and tail are relative to skb->head */
815 skb
->transport_header
+= off
;
816 skb
->network_header
+= off
;
817 skb
->mac_header
+= off
;
818 skb
->csum_start
+= nhead
;
822 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
829 /* Make private copy of skb with writable head and some headroom */
831 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
833 struct sk_buff
*skb2
;
834 int delta
= headroom
- skb_headroom(skb
);
837 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
839 skb2
= skb_clone(skb
, GFP_ATOMIC
);
840 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
851 * skb_copy_expand - copy and expand sk_buff
852 * @skb: buffer to copy
853 * @newheadroom: new free bytes at head
854 * @newtailroom: new free bytes at tail
855 * @gfp_mask: allocation priority
857 * Make a copy of both an &sk_buff and its data and while doing so
858 * allocate additional space.
860 * This is used when the caller wishes to modify the data and needs a
861 * private copy of the data to alter as well as more space for new fields.
862 * Returns %NULL on failure or the pointer to the buffer
863 * on success. The returned buffer has a reference count of 1.
865 * You must pass %GFP_ATOMIC as the allocation priority if this function
866 * is called from an interrupt.
868 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
869 int newheadroom
, int newtailroom
,
873 * Allocate the copy buffer
875 struct sk_buff
*n
= alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
877 int oldheadroom
= skb_headroom(skb
);
878 int head_copy_len
, head_copy_off
;
884 skb_reserve(n
, newheadroom
);
886 /* Set the tail pointer and length */
887 skb_put(n
, skb
->len
);
889 head_copy_len
= oldheadroom
;
891 if (newheadroom
<= head_copy_len
)
892 head_copy_len
= newheadroom
;
894 head_copy_off
= newheadroom
- head_copy_len
;
896 /* Copy the linear header and data. */
897 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
898 skb
->len
+ head_copy_len
))
901 copy_skb_header(n
, skb
);
903 off
= newheadroom
- oldheadroom
;
904 n
->csum_start
+= off
;
905 #ifdef NET_SKBUFF_DATA_USES_OFFSET
906 n
->transport_header
+= off
;
907 n
->network_header
+= off
;
908 n
->mac_header
+= off
;
915 * skb_pad - zero pad the tail of an skb
916 * @skb: buffer to pad
919 * Ensure that a buffer is followed by a padding area that is zero
920 * filled. Used by network drivers which may DMA or transfer data
921 * beyond the buffer end onto the wire.
923 * May return error in out of memory cases. The skb is freed on error.
926 int skb_pad(struct sk_buff
*skb
, int pad
)
931 /* If the skbuff is non linear tailroom is always zero.. */
932 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
933 memset(skb
->data
+skb
->len
, 0, pad
);
937 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
938 if (likely(skb_cloned(skb
) || ntail
> 0)) {
939 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
944 /* FIXME: The use of this function with non-linear skb's really needs
947 err
= skb_linearize(skb
);
951 memset(skb
->data
+ skb
->len
, 0, pad
);
960 * skb_put - add data to a buffer
961 * @skb: buffer to use
962 * @len: amount of data to add
964 * This function extends the used data area of the buffer. If this would
965 * exceed the total buffer size the kernel will panic. A pointer to the
966 * first byte of the extra data is returned.
968 unsigned char *skb_put(struct sk_buff
*skb
, unsigned int len
)
970 unsigned char *tmp
= skb_tail_pointer(skb
);
971 SKB_LINEAR_ASSERT(skb
);
974 if (unlikely(skb
->tail
> skb
->end
))
975 skb_over_panic(skb
, len
, __builtin_return_address(0));
978 EXPORT_SYMBOL(skb_put
);
981 * skb_push - add data to the start of a buffer
982 * @skb: buffer to use
983 * @len: amount of data to add
985 * This function extends the used data area of the buffer at the buffer
986 * start. If this would exceed the total buffer headroom the kernel will
987 * panic. A pointer to the first byte of the extra data is returned.
989 unsigned char *skb_push(struct sk_buff
*skb
, unsigned int len
)
993 if (unlikely(skb
->data
<skb
->head
))
994 skb_under_panic(skb
, len
, __builtin_return_address(0));
997 EXPORT_SYMBOL(skb_push
);
1000 * skb_pull - remove data from the start of a buffer
1001 * @skb: buffer to use
1002 * @len: amount of data to remove
1004 * This function removes data from the start of a buffer, returning
1005 * the memory to the headroom. A pointer to the next data in the buffer
1006 * is returned. Once the data has been pulled future pushes will overwrite
1009 unsigned char *skb_pull(struct sk_buff
*skb
, unsigned int len
)
1011 return unlikely(len
> skb
->len
) ? NULL
: __skb_pull(skb
, len
);
1013 EXPORT_SYMBOL(skb_pull
);
1016 * skb_trim - remove end from a buffer
1017 * @skb: buffer to alter
1020 * Cut the length of a buffer down by removing data from the tail. If
1021 * the buffer is already under the length specified it is not modified.
1022 * The skb must be linear.
1024 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1027 __skb_trim(skb
, len
);
1029 EXPORT_SYMBOL(skb_trim
);
1031 /* Trims skb to length len. It can change skb pointers.
1034 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1036 struct sk_buff
**fragp
;
1037 struct sk_buff
*frag
;
1038 int offset
= skb_headlen(skb
);
1039 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1043 if (skb_cloned(skb
) &&
1044 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1051 for (; i
< nfrags
; i
++) {
1052 int end
= offset
+ skb_shinfo(skb
)->frags
[i
].size
;
1059 skb_shinfo(skb
)->frags
[i
++].size
= len
- offset
;
1062 skb_shinfo(skb
)->nr_frags
= i
;
1064 for (; i
< nfrags
; i
++)
1065 put_page(skb_shinfo(skb
)->frags
[i
].page
);
1067 if (skb_shinfo(skb
)->frag_list
)
1068 skb_drop_fraglist(skb
);
1072 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1073 fragp
= &frag
->next
) {
1074 int end
= offset
+ frag
->len
;
1076 if (skb_shared(frag
)) {
1077 struct sk_buff
*nfrag
;
1079 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1080 if (unlikely(!nfrag
))
1083 nfrag
->next
= frag
->next
;
1095 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1099 skb_drop_list(&frag
->next
);
1104 if (len
> skb_headlen(skb
)) {
1105 skb
->data_len
-= skb
->len
- len
;
1110 skb_set_tail_pointer(skb
, len
);
1117 * __pskb_pull_tail - advance tail of skb header
1118 * @skb: buffer to reallocate
1119 * @delta: number of bytes to advance tail
1121 * The function makes a sense only on a fragmented &sk_buff,
1122 * it expands header moving its tail forward and copying necessary
1123 * data from fragmented part.
1125 * &sk_buff MUST have reference count of 1.
1127 * Returns %NULL (and &sk_buff does not change) if pull failed
1128 * or value of new tail of skb in the case of success.
1130 * All the pointers pointing into skb header may change and must be
1131 * reloaded after call to this function.
1134 /* Moves tail of skb head forward, copying data from fragmented part,
1135 * when it is necessary.
1136 * 1. It may fail due to malloc failure.
1137 * 2. It may change skb pointers.
1139 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1141 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1143 /* If skb has not enough free space at tail, get new one
1144 * plus 128 bytes for future expansions. If we have enough
1145 * room at tail, reallocate without expansion only if skb is cloned.
1147 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1149 if (eat
> 0 || skb_cloned(skb
)) {
1150 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1155 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
1158 /* Optimization: no fragments, no reasons to preestimate
1159 * size of pulled pages. Superb.
1161 if (!skb_shinfo(skb
)->frag_list
)
1164 /* Estimate size of pulled pages. */
1166 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1167 if (skb_shinfo(skb
)->frags
[i
].size
>= eat
)
1169 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1172 /* If we need update frag list, we are in troubles.
1173 * Certainly, it possible to add an offset to skb data,
1174 * but taking into account that pulling is expected to
1175 * be very rare operation, it is worth to fight against
1176 * further bloating skb head and crucify ourselves here instead.
1177 * Pure masohism, indeed. 8)8)
1180 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1181 struct sk_buff
*clone
= NULL
;
1182 struct sk_buff
*insp
= NULL
;
1187 if (list
->len
<= eat
) {
1188 /* Eaten as whole. */
1193 /* Eaten partially. */
1195 if (skb_shared(list
)) {
1196 /* Sucks! We need to fork list. :-( */
1197 clone
= skb_clone(list
, GFP_ATOMIC
);
1203 /* This may be pulled without
1207 if (!pskb_pull(list
, eat
)) {
1216 /* Free pulled out fragments. */
1217 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1218 skb_shinfo(skb
)->frag_list
= list
->next
;
1221 /* And insert new clone at head. */
1224 skb_shinfo(skb
)->frag_list
= clone
;
1227 /* Success! Now we may commit changes to skb data. */
1232 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1233 if (skb_shinfo(skb
)->frags
[i
].size
<= eat
) {
1234 put_page(skb_shinfo(skb
)->frags
[i
].page
);
1235 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1237 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1239 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1240 skb_shinfo(skb
)->frags
[k
].size
-= eat
;
1246 skb_shinfo(skb
)->nr_frags
= k
;
1249 skb
->data_len
-= delta
;
1251 return skb_tail_pointer(skb
);
1254 /* Copy some data bits from skb to kernel buffer. */
1256 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1259 int start
= skb_headlen(skb
);
1261 if (offset
> (int)skb
->len
- len
)
1265 if ((copy
= start
- offset
) > 0) {
1268 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
1269 if ((len
-= copy
) == 0)
1275 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1278 WARN_ON(start
> offset
+ len
);
1280 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1281 if ((copy
= end
- offset
) > 0) {
1287 vaddr
= kmap_skb_frag(&skb_shinfo(skb
)->frags
[i
]);
1289 vaddr
+ skb_shinfo(skb
)->frags
[i
].page_offset
+
1290 offset
- start
, copy
);
1291 kunmap_skb_frag(vaddr
);
1293 if ((len
-= copy
) == 0)
1301 if (skb_shinfo(skb
)->frag_list
) {
1302 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1304 for (; list
; list
= list
->next
) {
1307 WARN_ON(start
> offset
+ len
);
1309 end
= start
+ list
->len
;
1310 if ((copy
= end
- offset
) > 0) {
1313 if (skb_copy_bits(list
, offset
- start
,
1316 if ((len
-= copy
) == 0)
1332 * Callback from splice_to_pipe(), if we need to release some pages
1333 * at the end of the spd in case we error'ed out in filling the pipe.
1335 static void sock_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
1337 struct sk_buff
*skb
= (struct sk_buff
*) spd
->partial
[i
].private;
1343 * Fill page/offset/length into spd, if it can hold more pages.
1345 static inline int spd_fill_page(struct splice_pipe_desc
*spd
, struct page
*page
,
1346 unsigned int len
, unsigned int offset
,
1347 struct sk_buff
*skb
)
1349 if (unlikely(spd
->nr_pages
== PIPE_BUFFERS
))
1352 spd
->pages
[spd
->nr_pages
] = page
;
1353 spd
->partial
[spd
->nr_pages
].len
= len
;
1354 spd
->partial
[spd
->nr_pages
].offset
= offset
;
1355 spd
->partial
[spd
->nr_pages
].private = (unsigned long) skb_get(skb
);
1360 static inline void __segment_seek(struct page
**page
, unsigned int *poff
,
1361 unsigned int *plen
, unsigned int off
)
1364 *page
+= *poff
/ PAGE_SIZE
;
1365 *poff
= *poff
% PAGE_SIZE
;
1369 static inline int __splice_segment(struct page
*page
, unsigned int poff
,
1370 unsigned int plen
, unsigned int *off
,
1371 unsigned int *len
, struct sk_buff
*skb
,
1372 struct splice_pipe_desc
*spd
)
1377 /* skip this segment if already processed */
1383 /* ignore any bits we already processed */
1385 __segment_seek(&page
, &poff
, &plen
, *off
);
1390 unsigned int flen
= min(*len
, plen
);
1392 /* the linear region may spread across several pages */
1393 flen
= min_t(unsigned int, flen
, PAGE_SIZE
- poff
);
1395 if (spd_fill_page(spd
, page
, flen
, poff
, skb
))
1398 __segment_seek(&page
, &poff
, &plen
, flen
);
1401 } while (*len
&& plen
);
1407 * Map linear and fragment data from the skb to spd. It reports failure if the
1408 * pipe is full or if we already spliced the requested length.
1410 static int __skb_splice_bits(struct sk_buff
*skb
, unsigned int *offset
,
1412 struct splice_pipe_desc
*spd
)
1417 * map the linear part
1419 if (__splice_segment(virt_to_page(skb
->data
),
1420 (unsigned long) skb
->data
& (PAGE_SIZE
- 1),
1422 offset
, len
, skb
, spd
))
1426 * then map the fragments
1428 for (seg
= 0; seg
< skb_shinfo(skb
)->nr_frags
; seg
++) {
1429 const skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[seg
];
1431 if (__splice_segment(f
->page
, f
->page_offset
, f
->size
,
1432 offset
, len
, skb
, spd
))
1440 * Map data from the skb to a pipe. Should handle both the linear part,
1441 * the fragments, and the frag list. It does NOT handle frag lists within
1442 * the frag list, if such a thing exists. We'd probably need to recurse to
1443 * handle that cleanly.
1445 int skb_splice_bits(struct sk_buff
*__skb
, unsigned int offset
,
1446 struct pipe_inode_info
*pipe
, unsigned int tlen
,
1449 struct partial_page partial
[PIPE_BUFFERS
];
1450 struct page
*pages
[PIPE_BUFFERS
];
1451 struct splice_pipe_desc spd
= {
1455 .ops
= &sock_pipe_buf_ops
,
1456 .spd_release
= sock_spd_release
,
1458 struct sk_buff
*skb
;
1461 * I'd love to avoid the clone here, but tcp_read_sock()
1462 * ignores reference counts and unconditonally kills the sk_buff
1463 * on return from the actor.
1465 skb
= skb_clone(__skb
, GFP_KERNEL
);
1470 * __skb_splice_bits() only fails if the output has no room left,
1471 * so no point in going over the frag_list for the error case.
1473 if (__skb_splice_bits(skb
, &offset
, &tlen
, &spd
))
1479 * now see if we have a frag_list to map
1481 if (skb_shinfo(skb
)->frag_list
) {
1482 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1484 for (; list
&& tlen
; list
= list
->next
) {
1485 if (__skb_splice_bits(list
, &offset
, &tlen
, &spd
))
1492 * drop our reference to the clone, the pipe consumption will
1499 struct sock
*sk
= __skb
->sk
;
1502 * Drop the socket lock, otherwise we have reverse
1503 * locking dependencies between sk_lock and i_mutex
1504 * here as compared to sendfile(). We enter here
1505 * with the socket lock held, and splice_to_pipe() will
1506 * grab the pipe inode lock. For sendfile() emulation,
1507 * we call into ->sendpage() with the i_mutex lock held
1508 * and networking will grab the socket lock.
1511 ret
= splice_to_pipe(pipe
, &spd
);
1520 * skb_store_bits - store bits from kernel buffer to skb
1521 * @skb: destination buffer
1522 * @offset: offset in destination
1523 * @from: source buffer
1524 * @len: number of bytes to copy
1526 * Copy the specified number of bytes from the source buffer to the
1527 * destination skb. This function handles all the messy bits of
1528 * traversing fragment lists and such.
1531 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
1534 int start
= skb_headlen(skb
);
1536 if (offset
> (int)skb
->len
- len
)
1539 if ((copy
= start
- offset
) > 0) {
1542 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
1543 if ((len
-= copy
) == 0)
1549 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1550 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1553 WARN_ON(start
> offset
+ len
);
1555 end
= start
+ frag
->size
;
1556 if ((copy
= end
- offset
) > 0) {
1562 vaddr
= kmap_skb_frag(frag
);
1563 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1565 kunmap_skb_frag(vaddr
);
1567 if ((len
-= copy
) == 0)
1575 if (skb_shinfo(skb
)->frag_list
) {
1576 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1578 for (; list
; list
= list
->next
) {
1581 WARN_ON(start
> offset
+ len
);
1583 end
= start
+ list
->len
;
1584 if ((copy
= end
- offset
) > 0) {
1587 if (skb_store_bits(list
, offset
- start
,
1590 if ((len
-= copy
) == 0)
1605 EXPORT_SYMBOL(skb_store_bits
);
1607 /* Checksum skb data. */
1609 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
1610 int len
, __wsum csum
)
1612 int start
= skb_headlen(skb
);
1613 int i
, copy
= start
- offset
;
1616 /* Checksum header. */
1620 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1621 if ((len
-= copy
) == 0)
1627 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1630 WARN_ON(start
> offset
+ len
);
1632 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1633 if ((copy
= end
- offset
) > 0) {
1636 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1640 vaddr
= kmap_skb_frag(frag
);
1641 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
1642 offset
- start
, copy
, 0);
1643 kunmap_skb_frag(vaddr
);
1644 csum
= csum_block_add(csum
, csum2
, pos
);
1653 if (skb_shinfo(skb
)->frag_list
) {
1654 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1656 for (; list
; list
= list
->next
) {
1659 WARN_ON(start
> offset
+ len
);
1661 end
= start
+ list
->len
;
1662 if ((copy
= end
- offset
) > 0) {
1666 csum2
= skb_checksum(list
, offset
- start
,
1668 csum
= csum_block_add(csum
, csum2
, pos
);
1669 if ((len
-= copy
) == 0)
1682 /* Both of above in one bottle. */
1684 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
1685 u8
*to
, int len
, __wsum csum
)
1687 int start
= skb_headlen(skb
);
1688 int i
, copy
= start
- offset
;
1695 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
1697 if ((len
-= copy
) == 0)
1704 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1707 WARN_ON(start
> offset
+ len
);
1709 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1710 if ((copy
= end
- offset
) > 0) {
1713 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1717 vaddr
= kmap_skb_frag(frag
);
1718 csum2
= csum_partial_copy_nocheck(vaddr
+
1722 kunmap_skb_frag(vaddr
);
1723 csum
= csum_block_add(csum
, csum2
, pos
);
1733 if (skb_shinfo(skb
)->frag_list
) {
1734 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1736 for (; list
; list
= list
->next
) {
1740 WARN_ON(start
> offset
+ len
);
1742 end
= start
+ list
->len
;
1743 if ((copy
= end
- offset
) > 0) {
1746 csum2
= skb_copy_and_csum_bits(list
,
1749 csum
= csum_block_add(csum
, csum2
, pos
);
1750 if ((len
-= copy
) == 0)
1763 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
1768 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1769 csstart
= skb
->csum_start
- skb_headroom(skb
);
1771 csstart
= skb_headlen(skb
);
1773 BUG_ON(csstart
> skb_headlen(skb
));
1775 skb_copy_from_linear_data(skb
, to
, csstart
);
1778 if (csstart
!= skb
->len
)
1779 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
1780 skb
->len
- csstart
, 0);
1782 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1783 long csstuff
= csstart
+ skb
->csum_offset
;
1785 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
1790 * skb_dequeue - remove from the head of the queue
1791 * @list: list to dequeue from
1793 * Remove the head of the list. The list lock is taken so the function
1794 * may be used safely with other locking list functions. The head item is
1795 * returned or %NULL if the list is empty.
1798 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
1800 unsigned long flags
;
1801 struct sk_buff
*result
;
1803 spin_lock_irqsave(&list
->lock
, flags
);
1804 result
= __skb_dequeue(list
);
1805 spin_unlock_irqrestore(&list
->lock
, flags
);
1810 * skb_dequeue_tail - remove from the tail of the queue
1811 * @list: list to dequeue from
1813 * Remove the tail of the list. The list lock is taken so the function
1814 * may be used safely with other locking list functions. The tail item is
1815 * returned or %NULL if the list is empty.
1817 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
1819 unsigned long flags
;
1820 struct sk_buff
*result
;
1822 spin_lock_irqsave(&list
->lock
, flags
);
1823 result
= __skb_dequeue_tail(list
);
1824 spin_unlock_irqrestore(&list
->lock
, flags
);
1829 * skb_queue_purge - empty a list
1830 * @list: list to empty
1832 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1833 * the list and one reference dropped. This function takes the list
1834 * lock and is atomic with respect to other list locking functions.
1836 void skb_queue_purge(struct sk_buff_head
*list
)
1838 struct sk_buff
*skb
;
1839 while ((skb
= skb_dequeue(list
)) != NULL
)
1844 * skb_queue_head - queue a buffer at the list head
1845 * @list: list to use
1846 * @newsk: buffer to queue
1848 * Queue a buffer at the start of the list. This function takes the
1849 * list lock and can be used safely with other locking &sk_buff functions
1852 * A buffer cannot be placed on two lists at the same time.
1854 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1856 unsigned long flags
;
1858 spin_lock_irqsave(&list
->lock
, flags
);
1859 __skb_queue_head(list
, newsk
);
1860 spin_unlock_irqrestore(&list
->lock
, flags
);
1864 * skb_queue_tail - queue a buffer at the list tail
1865 * @list: list to use
1866 * @newsk: buffer to queue
1868 * Queue a buffer at the tail of the list. This function takes the
1869 * list lock and can be used safely with other locking &sk_buff functions
1872 * A buffer cannot be placed on two lists at the same time.
1874 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1876 unsigned long flags
;
1878 spin_lock_irqsave(&list
->lock
, flags
);
1879 __skb_queue_tail(list
, newsk
);
1880 spin_unlock_irqrestore(&list
->lock
, flags
);
1884 * skb_unlink - remove a buffer from a list
1885 * @skb: buffer to remove
1886 * @list: list to use
1888 * Remove a packet from a list. The list locks are taken and this
1889 * function is atomic with respect to other list locked calls
1891 * You must know what list the SKB is on.
1893 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
1895 unsigned long flags
;
1897 spin_lock_irqsave(&list
->lock
, flags
);
1898 __skb_unlink(skb
, list
);
1899 spin_unlock_irqrestore(&list
->lock
, flags
);
1903 * skb_append - append a buffer
1904 * @old: buffer to insert after
1905 * @newsk: buffer to insert
1906 * @list: list to use
1908 * Place a packet after a given packet in a list. The list locks are taken
1909 * and this function is atomic with respect to other list locked calls.
1910 * A buffer cannot be placed on two lists at the same time.
1912 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
1914 unsigned long flags
;
1916 spin_lock_irqsave(&list
->lock
, flags
);
1917 __skb_queue_after(list
, old
, newsk
);
1918 spin_unlock_irqrestore(&list
->lock
, flags
);
1923 * skb_insert - insert a buffer
1924 * @old: buffer to insert before
1925 * @newsk: buffer to insert
1926 * @list: list to use
1928 * Place a packet before a given packet in a list. The list locks are
1929 * taken and this function is atomic with respect to other list locked
1932 * A buffer cannot be placed on two lists at the same time.
1934 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
1936 unsigned long flags
;
1938 spin_lock_irqsave(&list
->lock
, flags
);
1939 __skb_insert(newsk
, old
->prev
, old
, list
);
1940 spin_unlock_irqrestore(&list
->lock
, flags
);
1943 static inline void skb_split_inside_header(struct sk_buff
*skb
,
1944 struct sk_buff
* skb1
,
1945 const u32 len
, const int pos
)
1949 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
1951 /* And move data appendix as is. */
1952 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1953 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
1955 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
1956 skb_shinfo(skb
)->nr_frags
= 0;
1957 skb1
->data_len
= skb
->data_len
;
1958 skb1
->len
+= skb1
->data_len
;
1961 skb_set_tail_pointer(skb
, len
);
1964 static inline void skb_split_no_header(struct sk_buff
*skb
,
1965 struct sk_buff
* skb1
,
1966 const u32 len
, int pos
)
1969 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
1971 skb_shinfo(skb
)->nr_frags
= 0;
1972 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
1974 skb
->data_len
= len
- pos
;
1976 for (i
= 0; i
< nfrags
; i
++) {
1977 int size
= skb_shinfo(skb
)->frags
[i
].size
;
1979 if (pos
+ size
> len
) {
1980 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1984 * We have two variants in this case:
1985 * 1. Move all the frag to the second
1986 * part, if it is possible. F.e.
1987 * this approach is mandatory for TUX,
1988 * where splitting is expensive.
1989 * 2. Split is accurately. We make this.
1991 get_page(skb_shinfo(skb
)->frags
[i
].page
);
1992 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
1993 skb_shinfo(skb1
)->frags
[0].size
-= len
- pos
;
1994 skb_shinfo(skb
)->frags
[i
].size
= len
- pos
;
1995 skb_shinfo(skb
)->nr_frags
++;
1999 skb_shinfo(skb
)->nr_frags
++;
2002 skb_shinfo(skb1
)->nr_frags
= k
;
2006 * skb_split - Split fragmented skb to two parts at length len.
2007 * @skb: the buffer to split
2008 * @skb1: the buffer to receive the second part
2009 * @len: new length for skb
2011 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
2013 int pos
= skb_headlen(skb
);
2015 if (len
< pos
) /* Split line is inside header. */
2016 skb_split_inside_header(skb
, skb1
, len
, pos
);
2017 else /* Second chunk has no header, nothing to copy. */
2018 skb_split_no_header(skb
, skb1
, len
, pos
);
2021 /* Shifting from/to a cloned skb is a no-go.
2023 * Caller cannot keep skb_shinfo related pointers past calling here!
2025 static int skb_prepare_for_shift(struct sk_buff
*skb
)
2027 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2031 * skb_shift - Shifts paged data partially from skb to another
2032 * @tgt: buffer into which tail data gets added
2033 * @skb: buffer from which the paged data comes from
2034 * @shiftlen: shift up to this many bytes
2036 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2037 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2038 * It's up to caller to free skb if everything was shifted.
2040 * If @tgt runs out of frags, the whole operation is aborted.
2042 * Skb cannot include anything else but paged data while tgt is allowed
2043 * to have non-paged data as well.
2045 * TODO: full sized shift could be optimized but that would need
2046 * specialized skb free'er to handle frags without up-to-date nr_frags.
2048 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
2050 int from
, to
, merge
, todo
;
2051 struct skb_frag_struct
*fragfrom
, *fragto
;
2053 BUG_ON(shiftlen
> skb
->len
);
2054 BUG_ON(skb_headlen(skb
)); /* Would corrupt stream */
2058 to
= skb_shinfo(tgt
)->nr_frags
;
2059 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2061 /* Actual merge is delayed until the point when we know we can
2062 * commit all, so that we don't have to undo partial changes
2065 !skb_can_coalesce(tgt
, to
, fragfrom
->page
, fragfrom
->page_offset
)) {
2070 todo
-= fragfrom
->size
;
2072 if (skb_prepare_for_shift(skb
) ||
2073 skb_prepare_for_shift(tgt
))
2076 /* All previous frag pointers might be stale! */
2077 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2078 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2080 fragto
->size
+= shiftlen
;
2081 fragfrom
->size
-= shiftlen
;
2082 fragfrom
->page_offset
+= shiftlen
;
2090 /* Skip full, not-fitting skb to avoid expensive operations */
2091 if ((shiftlen
== skb
->len
) &&
2092 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
2095 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
2098 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
2099 if (to
== MAX_SKB_FRAGS
)
2102 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2103 fragto
= &skb_shinfo(tgt
)->frags
[to
];
2105 if (todo
>= fragfrom
->size
) {
2106 *fragto
= *fragfrom
;
2107 todo
-= fragfrom
->size
;
2112 get_page(fragfrom
->page
);
2113 fragto
->page
= fragfrom
->page
;
2114 fragto
->page_offset
= fragfrom
->page_offset
;
2115 fragto
->size
= todo
;
2117 fragfrom
->page_offset
+= todo
;
2118 fragfrom
->size
-= todo
;
2126 /* Ready to "commit" this state change to tgt */
2127 skb_shinfo(tgt
)->nr_frags
= to
;
2130 fragfrom
= &skb_shinfo(skb
)->frags
[0];
2131 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2133 fragto
->size
+= fragfrom
->size
;
2134 put_page(fragfrom
->page
);
2137 /* Reposition in the original skb */
2139 while (from
< skb_shinfo(skb
)->nr_frags
)
2140 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
2141 skb_shinfo(skb
)->nr_frags
= to
;
2143 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
2146 /* Most likely the tgt won't ever need its checksum anymore, skb on
2147 * the other hand might need it if it needs to be resent
2149 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
2150 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2152 /* Yak, is it really working this way? Some helper please? */
2153 skb
->len
-= shiftlen
;
2154 skb
->data_len
-= shiftlen
;
2155 skb
->truesize
-= shiftlen
;
2156 tgt
->len
+= shiftlen
;
2157 tgt
->data_len
+= shiftlen
;
2158 tgt
->truesize
+= shiftlen
;
2164 * skb_prepare_seq_read - Prepare a sequential read of skb data
2165 * @skb: the buffer to read
2166 * @from: lower offset of data to be read
2167 * @to: upper offset of data to be read
2168 * @st: state variable
2170 * Initializes the specified state variable. Must be called before
2171 * invoking skb_seq_read() for the first time.
2173 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
2174 unsigned int to
, struct skb_seq_state
*st
)
2176 st
->lower_offset
= from
;
2177 st
->upper_offset
= to
;
2178 st
->root_skb
= st
->cur_skb
= skb
;
2179 st
->frag_idx
= st
->stepped_offset
= 0;
2180 st
->frag_data
= NULL
;
2184 * skb_seq_read - Sequentially read skb data
2185 * @consumed: number of bytes consumed by the caller so far
2186 * @data: destination pointer for data to be returned
2187 * @st: state variable
2189 * Reads a block of skb data at &consumed relative to the
2190 * lower offset specified to skb_prepare_seq_read(). Assigns
2191 * the head of the data block to &data and returns the length
2192 * of the block or 0 if the end of the skb data or the upper
2193 * offset has been reached.
2195 * The caller is not required to consume all of the data
2196 * returned, i.e. &consumed is typically set to the number
2197 * of bytes already consumed and the next call to
2198 * skb_seq_read() will return the remaining part of the block.
2200 * Note 1: The size of each block of data returned can be arbitary,
2201 * this limitation is the cost for zerocopy seqeuental
2202 * reads of potentially non linear data.
2204 * Note 2: Fragment lists within fragments are not implemented
2205 * at the moment, state->root_skb could be replaced with
2206 * a stack for this purpose.
2208 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
2209 struct skb_seq_state
*st
)
2211 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
2214 if (unlikely(abs_offset
>= st
->upper_offset
))
2218 block_limit
= skb_headlen(st
->cur_skb
);
2220 if (abs_offset
< block_limit
) {
2221 *data
= st
->cur_skb
->data
+ abs_offset
;
2222 return block_limit
- abs_offset
;
2225 if (st
->frag_idx
== 0 && !st
->frag_data
)
2226 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
2228 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
2229 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
2230 block_limit
= frag
->size
+ st
->stepped_offset
;
2232 if (abs_offset
< block_limit
) {
2234 st
->frag_data
= kmap_skb_frag(frag
);
2236 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
2237 (abs_offset
- st
->stepped_offset
);
2239 return block_limit
- abs_offset
;
2242 if (st
->frag_data
) {
2243 kunmap_skb_frag(st
->frag_data
);
2244 st
->frag_data
= NULL
;
2248 st
->stepped_offset
+= frag
->size
;
2251 if (st
->frag_data
) {
2252 kunmap_skb_frag(st
->frag_data
);
2253 st
->frag_data
= NULL
;
2256 if (st
->cur_skb
->next
) {
2257 st
->cur_skb
= st
->cur_skb
->next
;
2260 } else if (st
->root_skb
== st
->cur_skb
&&
2261 skb_shinfo(st
->root_skb
)->frag_list
) {
2262 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
2270 * skb_abort_seq_read - Abort a sequential read of skb data
2271 * @st: state variable
2273 * Must be called if skb_seq_read() was not called until it
2276 void skb_abort_seq_read(struct skb_seq_state
*st
)
2279 kunmap_skb_frag(st
->frag_data
);
2282 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2284 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
2285 struct ts_config
*conf
,
2286 struct ts_state
*state
)
2288 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
2291 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
2293 skb_abort_seq_read(TS_SKB_CB(state
));
2297 * skb_find_text - Find a text pattern in skb data
2298 * @skb: the buffer to look in
2299 * @from: search offset
2301 * @config: textsearch configuration
2302 * @state: uninitialized textsearch state variable
2304 * Finds a pattern in the skb data according to the specified
2305 * textsearch configuration. Use textsearch_next() to retrieve
2306 * subsequent occurrences of the pattern. Returns the offset
2307 * to the first occurrence or UINT_MAX if no match was found.
2309 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
2310 unsigned int to
, struct ts_config
*config
,
2311 struct ts_state
*state
)
2315 config
->get_next_block
= skb_ts_get_next_block
;
2316 config
->finish
= skb_ts_finish
;
2318 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
2320 ret
= textsearch_find(config
, state
);
2321 return (ret
<= to
- from
? ret
: UINT_MAX
);
2325 * skb_append_datato_frags: - append the user data to a skb
2326 * @sk: sock structure
2327 * @skb: skb structure to be appened with user data.
2328 * @getfrag: call back function to be used for getting the user data
2329 * @from: pointer to user message iov
2330 * @length: length of the iov message
2332 * Description: This procedure append the user data in the fragment part
2333 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2335 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
2336 int (*getfrag
)(void *from
, char *to
, int offset
,
2337 int len
, int odd
, struct sk_buff
*skb
),
2338 void *from
, int length
)
2341 skb_frag_t
*frag
= NULL
;
2342 struct page
*page
= NULL
;
2348 /* Return error if we don't have space for new frag */
2349 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2350 if (frg_cnt
>= MAX_SKB_FRAGS
)
2353 /* allocate a new page for next frag */
2354 page
= alloc_pages(sk
->sk_allocation
, 0);
2356 /* If alloc_page fails just return failure and caller will
2357 * free previous allocated pages by doing kfree_skb()
2362 /* initialize the next frag */
2363 sk
->sk_sndmsg_page
= page
;
2364 sk
->sk_sndmsg_off
= 0;
2365 skb_fill_page_desc(skb
, frg_cnt
, page
, 0, 0);
2366 skb
->truesize
+= PAGE_SIZE
;
2367 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
2369 /* get the new initialized frag */
2370 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2371 frag
= &skb_shinfo(skb
)->frags
[frg_cnt
- 1];
2373 /* copy the user data to page */
2374 left
= PAGE_SIZE
- frag
->page_offset
;
2375 copy
= (length
> left
)? left
: length
;
2377 ret
= getfrag(from
, (page_address(frag
->page
) +
2378 frag
->page_offset
+ frag
->size
),
2379 offset
, copy
, 0, skb
);
2383 /* copy was successful so update the size parameters */
2384 sk
->sk_sndmsg_off
+= copy
;
2387 skb
->data_len
+= copy
;
2391 } while (length
> 0);
2397 * skb_pull_rcsum - pull skb and update receive checksum
2398 * @skb: buffer to update
2399 * @len: length of data pulled
2401 * This function performs an skb_pull on the packet and updates
2402 * the CHECKSUM_COMPLETE checksum. It should be used on
2403 * receive path processing instead of skb_pull unless you know
2404 * that the checksum difference is zero (e.g., a valid IP header)
2405 * or you are setting ip_summed to CHECKSUM_NONE.
2407 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
2409 BUG_ON(len
> skb
->len
);
2411 BUG_ON(skb
->len
< skb
->data_len
);
2412 skb_postpull_rcsum(skb
, skb
->data
, len
);
2413 return skb
->data
+= len
;
2416 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
2419 * skb_segment - Perform protocol segmentation on skb.
2420 * @skb: buffer to segment
2421 * @features: features for the output path (see dev->features)
2423 * This function performs segmentation on the given skb. It returns
2424 * a pointer to the first in a list of new skbs for the segments.
2425 * In case of error it returns ERR_PTR(err).
2427 struct sk_buff
*skb_segment(struct sk_buff
*skb
, int features
)
2429 struct sk_buff
*segs
= NULL
;
2430 struct sk_buff
*tail
= NULL
;
2431 unsigned int mss
= skb_shinfo(skb
)->gso_size
;
2432 unsigned int doffset
= skb
->data
- skb_mac_header(skb
);
2433 unsigned int offset
= doffset
;
2434 unsigned int headroom
;
2436 int sg
= features
& NETIF_F_SG
;
2437 int nfrags
= skb_shinfo(skb
)->nr_frags
;
2442 __skb_push(skb
, doffset
);
2443 headroom
= skb_headroom(skb
);
2444 pos
= skb_headlen(skb
);
2447 struct sk_buff
*nskb
;
2453 len
= skb
->len
- offset
;
2457 hsize
= skb_headlen(skb
) - offset
;
2460 if (hsize
> len
|| !sg
)
2463 nskb
= alloc_skb(hsize
+ doffset
+ headroom
, GFP_ATOMIC
);
2464 if (unlikely(!nskb
))
2473 __copy_skb_header(nskb
, skb
);
2474 nskb
->mac_len
= skb
->mac_len
;
2476 skb_reserve(nskb
, headroom
);
2477 skb_reset_mac_header(nskb
);
2478 skb_set_network_header(nskb
, skb
->mac_len
);
2479 nskb
->transport_header
= (nskb
->network_header
+
2480 skb_network_header_len(skb
));
2481 skb_copy_from_linear_data(skb
, skb_put(nskb
, doffset
),
2484 nskb
->ip_summed
= CHECKSUM_NONE
;
2485 nskb
->csum
= skb_copy_and_csum_bits(skb
, offset
,
2491 frag
= skb_shinfo(nskb
)->frags
;
2494 skb_copy_from_linear_data_offset(skb
, offset
,
2495 skb_put(nskb
, hsize
), hsize
);
2497 while (pos
< offset
+ len
) {
2498 BUG_ON(i
>= nfrags
);
2500 *frag
= skb_shinfo(skb
)->frags
[i
];
2501 get_page(frag
->page
);
2505 frag
->page_offset
+= offset
- pos
;
2506 frag
->size
-= offset
- pos
;
2511 if (pos
+ size
<= offset
+ len
) {
2515 frag
->size
-= pos
+ size
- (offset
+ len
);
2522 skb_shinfo(nskb
)->nr_frags
= k
;
2523 nskb
->data_len
= len
- hsize
;
2524 nskb
->len
+= nskb
->data_len
;
2525 nskb
->truesize
+= nskb
->data_len
;
2526 } while ((offset
+= len
) < skb
->len
);
2531 while ((skb
= segs
)) {
2535 return ERR_PTR(err
);
2538 EXPORT_SYMBOL_GPL(skb_segment
);
2540 void __init
skb_init(void)
2542 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
2543 sizeof(struct sk_buff
),
2545 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2547 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
2548 (2*sizeof(struct sk_buff
)) +
2551 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2556 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2557 * @skb: Socket buffer containing the buffers to be mapped
2558 * @sg: The scatter-gather list to map into
2559 * @offset: The offset into the buffer's contents to start mapping
2560 * @len: Length of buffer space to be mapped
2562 * Fill the specified scatter-gather list with mappings/pointers into a
2563 * region of the buffer space attached to a socket buffer.
2566 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2568 int start
= skb_headlen(skb
);
2569 int i
, copy
= start
- offset
;
2575 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
2577 if ((len
-= copy
) == 0)
2582 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2585 WARN_ON(start
> offset
+ len
);
2587 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
2588 if ((copy
= end
- offset
) > 0) {
2589 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2593 sg_set_page(&sg
[elt
], frag
->page
, copy
,
2594 frag
->page_offset
+offset
-start
);
2603 if (skb_shinfo(skb
)->frag_list
) {
2604 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
2606 for (; list
; list
= list
->next
) {
2609 WARN_ON(start
> offset
+ len
);
2611 end
= start
+ list
->len
;
2612 if ((copy
= end
- offset
) > 0) {
2615 elt
+= __skb_to_sgvec(list
, sg
+elt
, offset
- start
,
2617 if ((len
-= copy
) == 0)
2628 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2630 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
);
2632 sg_mark_end(&sg
[nsg
- 1]);
2638 * skb_cow_data - Check that a socket buffer's data buffers are writable
2639 * @skb: The socket buffer to check.
2640 * @tailbits: Amount of trailing space to be added
2641 * @trailer: Returned pointer to the skb where the @tailbits space begins
2643 * Make sure that the data buffers attached to a socket buffer are
2644 * writable. If they are not, private copies are made of the data buffers
2645 * and the socket buffer is set to use these instead.
2647 * If @tailbits is given, make sure that there is space to write @tailbits
2648 * bytes of data beyond current end of socket buffer. @trailer will be
2649 * set to point to the skb in which this space begins.
2651 * The number of scatterlist elements required to completely map the
2652 * COW'd and extended socket buffer will be returned.
2654 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
2658 struct sk_buff
*skb1
, **skb_p
;
2660 /* If skb is cloned or its head is paged, reallocate
2661 * head pulling out all the pages (pages are considered not writable
2662 * at the moment even if they are anonymous).
2664 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
2665 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
2668 /* Easy case. Most of packets will go this way. */
2669 if (!skb_shinfo(skb
)->frag_list
) {
2670 /* A little of trouble, not enough of space for trailer.
2671 * This should not happen, when stack is tuned to generate
2672 * good frames. OK, on miss we reallocate and reserve even more
2673 * space, 128 bytes is fair. */
2675 if (skb_tailroom(skb
) < tailbits
&&
2676 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
2684 /* Misery. We are in troubles, going to mincer fragments... */
2687 skb_p
= &skb_shinfo(skb
)->frag_list
;
2690 while ((skb1
= *skb_p
) != NULL
) {
2693 /* The fragment is partially pulled by someone,
2694 * this can happen on input. Copy it and everything
2697 if (skb_shared(skb1
))
2700 /* If the skb is the last, worry about trailer. */
2702 if (skb1
->next
== NULL
&& tailbits
) {
2703 if (skb_shinfo(skb1
)->nr_frags
||
2704 skb_shinfo(skb1
)->frag_list
||
2705 skb_tailroom(skb1
) < tailbits
)
2706 ntail
= tailbits
+ 128;
2712 skb_shinfo(skb1
)->nr_frags
||
2713 skb_shinfo(skb1
)->frag_list
) {
2714 struct sk_buff
*skb2
;
2716 /* Fuck, we are miserable poor guys... */
2718 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
2720 skb2
= skb_copy_expand(skb1
,
2724 if (unlikely(skb2
== NULL
))
2728 skb_set_owner_w(skb2
, skb1
->sk
);
2730 /* Looking around. Are we still alive?
2731 * OK, link new skb, drop old one */
2733 skb2
->next
= skb1
->next
;
2740 skb_p
= &skb1
->next
;
2747 * skb_partial_csum_set - set up and verify partial csum values for packet
2748 * @skb: the skb to set
2749 * @start: the number of bytes after skb->data to start checksumming.
2750 * @off: the offset from start to place the checksum.
2752 * For untrusted partially-checksummed packets, we need to make sure the values
2753 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
2755 * This function checks and sets those values and skb->ip_summed: if this
2756 * returns false you should drop the packet.
2758 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
2760 if (unlikely(start
> skb
->len
- 2) ||
2761 unlikely((int)start
+ off
> skb
->len
- 2)) {
2762 if (net_ratelimit())
2764 "bad partial csum: csum=%u/%u len=%u\n",
2765 start
, off
, skb
->len
);
2768 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2769 skb
->csum_start
= skb_headroom(skb
) + start
;
2770 skb
->csum_offset
= off
;
2774 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
2776 if (net_ratelimit())
2777 pr_warning("%s: received packets cannot be forwarded"
2778 " while LRO is enabled\n", skb
->dev
->name
);
2781 EXPORT_SYMBOL(___pskb_trim
);
2782 EXPORT_SYMBOL(__kfree_skb
);
2783 EXPORT_SYMBOL(kfree_skb
);
2784 EXPORT_SYMBOL(__pskb_pull_tail
);
2785 EXPORT_SYMBOL(__alloc_skb
);
2786 EXPORT_SYMBOL(__netdev_alloc_skb
);
2787 EXPORT_SYMBOL(pskb_copy
);
2788 EXPORT_SYMBOL(pskb_expand_head
);
2789 EXPORT_SYMBOL(skb_checksum
);
2790 EXPORT_SYMBOL(skb_clone
);
2791 EXPORT_SYMBOL(skb_copy
);
2792 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
2793 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
2794 EXPORT_SYMBOL(skb_copy_bits
);
2795 EXPORT_SYMBOL(skb_copy_expand
);
2796 EXPORT_SYMBOL(skb_over_panic
);
2797 EXPORT_SYMBOL(skb_pad
);
2798 EXPORT_SYMBOL(skb_realloc_headroom
);
2799 EXPORT_SYMBOL(skb_under_panic
);
2800 EXPORT_SYMBOL(skb_dequeue
);
2801 EXPORT_SYMBOL(skb_dequeue_tail
);
2802 EXPORT_SYMBOL(skb_insert
);
2803 EXPORT_SYMBOL(skb_queue_purge
);
2804 EXPORT_SYMBOL(skb_queue_head
);
2805 EXPORT_SYMBOL(skb_queue_tail
);
2806 EXPORT_SYMBOL(skb_unlink
);
2807 EXPORT_SYMBOL(skb_append
);
2808 EXPORT_SYMBOL(skb_split
);
2809 EXPORT_SYMBOL(skb_prepare_seq_read
);
2810 EXPORT_SYMBOL(skb_seq_read
);
2811 EXPORT_SYMBOL(skb_abort_seq_read
);
2812 EXPORT_SYMBOL(skb_find_text
);
2813 EXPORT_SYMBOL(skb_append_datato_frags
);
2814 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);
2816 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
2817 EXPORT_SYMBOL_GPL(skb_cow_data
);
2818 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);