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>
58 #include <linux/errqueue.h>
60 #include <net/protocol.h>
63 #include <net/checksum.h>
66 #include <asm/uaccess.h>
67 #include <asm/system.h>
68 #include <trace/skb.h>
72 static struct kmem_cache
*skbuff_head_cache __read_mostly
;
73 static struct kmem_cache
*skbuff_fclone_cache __read_mostly
;
75 static void sock_pipe_buf_release(struct pipe_inode_info
*pipe
,
76 struct pipe_buffer
*buf
)
81 static void sock_pipe_buf_get(struct pipe_inode_info
*pipe
,
82 struct pipe_buffer
*buf
)
87 static int sock_pipe_buf_steal(struct pipe_inode_info
*pipe
,
88 struct pipe_buffer
*buf
)
94 /* Pipe buffer operations for a socket. */
95 static struct pipe_buf_operations sock_pipe_buf_ops
= {
97 .map
= generic_pipe_buf_map
,
98 .unmap
= generic_pipe_buf_unmap
,
99 .confirm
= generic_pipe_buf_confirm
,
100 .release
= sock_pipe_buf_release
,
101 .steal
= sock_pipe_buf_steal
,
102 .get
= sock_pipe_buf_get
,
106 * Keep out-of-line to prevent kernel bloat.
107 * __builtin_return_address is not used because it is not always
112 * skb_over_panic - private function
117 * Out of line support code for skb_put(). Not user callable.
119 void skb_over_panic(struct sk_buff
*skb
, int sz
, void *here
)
121 printk(KERN_EMERG
"skb_over_panic: text:%p len:%d put:%d head:%p "
122 "data:%p tail:%#lx end:%#lx dev:%s\n",
123 here
, skb
->len
, sz
, skb
->head
, skb
->data
,
124 (unsigned long)skb
->tail
, (unsigned long)skb
->end
,
125 skb
->dev
? skb
->dev
->name
: "<NULL>");
128 EXPORT_SYMBOL(skb_over_panic
);
131 * skb_under_panic - private function
136 * Out of line support code for skb_push(). Not user callable.
139 void skb_under_panic(struct sk_buff
*skb
, int sz
, void *here
)
141 printk(KERN_EMERG
"skb_under_panic: text:%p len:%d put:%d head:%p "
142 "data:%p tail:%#lx end:%#lx dev:%s\n",
143 here
, skb
->len
, sz
, skb
->head
, skb
->data
,
144 (unsigned long)skb
->tail
, (unsigned long)skb
->end
,
145 skb
->dev
? skb
->dev
->name
: "<NULL>");
148 EXPORT_SYMBOL(skb_under_panic
);
150 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
151 * 'private' fields and also do memory statistics to find all the
157 * __alloc_skb - allocate a network buffer
158 * @size: size to allocate
159 * @gfp_mask: allocation mask
160 * @fclone: allocate from fclone cache instead of head cache
161 * and allocate a cloned (child) skb
162 * @node: numa node to allocate memory on
164 * Allocate a new &sk_buff. The returned buffer has no headroom and a
165 * tail room of size bytes. The object has a reference count of one.
166 * The return is the buffer. On a failure the return is %NULL.
168 * Buffers may only be allocated from interrupts using a @gfp_mask of
171 struct sk_buff
*__alloc_skb(unsigned int size
, gfp_t gfp_mask
,
172 int fclone
, int node
)
174 struct kmem_cache
*cache
;
175 struct skb_shared_info
*shinfo
;
179 cache
= fclone
? skbuff_fclone_cache
: skbuff_head_cache
;
182 skb
= kmem_cache_alloc_node(cache
, gfp_mask
& ~__GFP_DMA
, node
);
186 size
= SKB_DATA_ALIGN(size
);
187 data
= kmalloc_node_track_caller(size
+ sizeof(struct skb_shared_info
),
193 * Only clear those fields we need to clear, not those that we will
194 * actually initialise below. Hence, don't put any more fields after
195 * the tail pointer in struct sk_buff!
197 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
198 skb
->truesize
= size
+ sizeof(struct sk_buff
);
199 atomic_set(&skb
->users
, 1);
202 skb_reset_tail_pointer(skb
);
203 skb
->end
= skb
->tail
+ size
;
204 /* make sure we initialize shinfo sequentially */
205 shinfo
= skb_shinfo(skb
);
206 atomic_set(&shinfo
->dataref
, 1);
207 shinfo
->nr_frags
= 0;
208 shinfo
->gso_size
= 0;
209 shinfo
->gso_segs
= 0;
210 shinfo
->gso_type
= 0;
211 shinfo
->ip6_frag_id
= 0;
212 shinfo
->tx_flags
.flags
= 0;
213 shinfo
->frag_list
= NULL
;
214 memset(&shinfo
->hwtstamps
, 0, sizeof(shinfo
->hwtstamps
));
217 struct sk_buff
*child
= skb
+ 1;
218 atomic_t
*fclone_ref
= (atomic_t
*) (child
+ 1);
220 skb
->fclone
= SKB_FCLONE_ORIG
;
221 atomic_set(fclone_ref
, 1);
223 child
->fclone
= SKB_FCLONE_UNAVAILABLE
;
228 kmem_cache_free(cache
, skb
);
232 EXPORT_SYMBOL(__alloc_skb
);
235 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
236 * @dev: network device to receive on
237 * @length: length to allocate
238 * @gfp_mask: get_free_pages mask, passed to alloc_skb
240 * Allocate a new &sk_buff and assign it a usage count of one. The
241 * buffer has unspecified headroom built in. Users should allocate
242 * the headroom they think they need without accounting for the
243 * built in space. The built in space is used for optimisations.
245 * %NULL is returned if there is no free memory.
247 struct sk_buff
*__netdev_alloc_skb(struct net_device
*dev
,
248 unsigned int length
, gfp_t gfp_mask
)
250 int node
= dev
->dev
.parent
? dev_to_node(dev
->dev
.parent
) : -1;
253 skb
= __alloc_skb(length
+ NET_SKB_PAD
, gfp_mask
, 0, node
);
255 skb_reserve(skb
, NET_SKB_PAD
);
260 EXPORT_SYMBOL(__netdev_alloc_skb
);
262 struct page
*__netdev_alloc_page(struct net_device
*dev
, gfp_t gfp_mask
)
264 int node
= dev
->dev
.parent
? dev_to_node(dev
->dev
.parent
) : -1;
267 page
= alloc_pages_node(node
, gfp_mask
, 0);
270 EXPORT_SYMBOL(__netdev_alloc_page
);
272 void skb_add_rx_frag(struct sk_buff
*skb
, int i
, struct page
*page
, int off
,
275 skb_fill_page_desc(skb
, i
, page
, off
, size
);
277 skb
->data_len
+= size
;
278 skb
->truesize
+= size
;
280 EXPORT_SYMBOL(skb_add_rx_frag
);
283 * dev_alloc_skb - allocate an skbuff for receiving
284 * @length: length to allocate
286 * Allocate a new &sk_buff and assign it a usage count of one. The
287 * buffer has unspecified headroom built in. Users should allocate
288 * the headroom they think they need without accounting for the
289 * built in space. The built in space is used for optimisations.
291 * %NULL is returned if there is no free memory. Although this function
292 * allocates memory it can be called from an interrupt.
294 struct sk_buff
*dev_alloc_skb(unsigned int length
)
297 * There is more code here than it seems:
298 * __dev_alloc_skb is an inline
300 return __dev_alloc_skb(length
, GFP_ATOMIC
);
302 EXPORT_SYMBOL(dev_alloc_skb
);
304 static void skb_drop_list(struct sk_buff
**listp
)
306 struct sk_buff
*list
= *listp
;
311 struct sk_buff
*this = list
;
317 static inline void skb_drop_fraglist(struct sk_buff
*skb
)
319 skb_drop_list(&skb_shinfo(skb
)->frag_list
);
322 static void skb_clone_fraglist(struct sk_buff
*skb
)
324 struct sk_buff
*list
;
326 for (list
= skb_shinfo(skb
)->frag_list
; list
; list
= list
->next
)
330 static void skb_release_data(struct sk_buff
*skb
)
333 !atomic_sub_return(skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1,
334 &skb_shinfo(skb
)->dataref
)) {
335 if (skb_shinfo(skb
)->nr_frags
) {
337 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
338 put_page(skb_shinfo(skb
)->frags
[i
].page
);
341 if (skb_shinfo(skb
)->frag_list
)
342 skb_drop_fraglist(skb
);
349 * Free an skbuff by memory without cleaning the state.
351 static void kfree_skbmem(struct sk_buff
*skb
)
353 struct sk_buff
*other
;
354 atomic_t
*fclone_ref
;
356 switch (skb
->fclone
) {
357 case SKB_FCLONE_UNAVAILABLE
:
358 kmem_cache_free(skbuff_head_cache
, skb
);
361 case SKB_FCLONE_ORIG
:
362 fclone_ref
= (atomic_t
*) (skb
+ 2);
363 if (atomic_dec_and_test(fclone_ref
))
364 kmem_cache_free(skbuff_fclone_cache
, skb
);
367 case SKB_FCLONE_CLONE
:
368 fclone_ref
= (atomic_t
*) (skb
+ 1);
371 /* The clone portion is available for
372 * fast-cloning again.
374 skb
->fclone
= SKB_FCLONE_UNAVAILABLE
;
376 if (atomic_dec_and_test(fclone_ref
))
377 kmem_cache_free(skbuff_fclone_cache
, other
);
382 static void skb_release_head_state(struct sk_buff
*skb
)
384 dst_release(skb
->dst
);
386 secpath_put(skb
->sp
);
388 if (skb
->destructor
) {
390 skb
->destructor(skb
);
392 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
393 nf_conntrack_put(skb
->nfct
);
394 nf_conntrack_put_reasm(skb
->nfct_reasm
);
396 #ifdef CONFIG_BRIDGE_NETFILTER
397 nf_bridge_put(skb
->nf_bridge
);
399 /* XXX: IS this still necessary? - JHS */
400 #ifdef CONFIG_NET_SCHED
402 #ifdef CONFIG_NET_CLS_ACT
408 /* Free everything but the sk_buff shell. */
409 static void skb_release_all(struct sk_buff
*skb
)
411 skb_release_head_state(skb
);
412 skb_release_data(skb
);
416 * __kfree_skb - private function
419 * Free an sk_buff. Release anything attached to the buffer.
420 * Clean the state. This is an internal helper function. Users should
421 * always call kfree_skb
424 void __kfree_skb(struct sk_buff
*skb
)
426 skb_release_all(skb
);
429 EXPORT_SYMBOL(__kfree_skb
);
432 * kfree_skb - free an sk_buff
433 * @skb: buffer to free
435 * Drop a reference to the buffer and free it if the usage count has
438 void kfree_skb(struct sk_buff
*skb
)
442 if (likely(atomic_read(&skb
->users
) == 1))
444 else if (likely(!atomic_dec_and_test(&skb
->users
)))
446 trace_kfree_skb(skb
, __builtin_return_address(0));
449 EXPORT_SYMBOL(kfree_skb
);
452 * consume_skb - free an skbuff
453 * @skb: buffer to free
455 * Drop a ref to the buffer and free it if the usage count has hit zero
456 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
457 * is being dropped after a failure and notes that
459 void consume_skb(struct sk_buff
*skb
)
463 if (likely(atomic_read(&skb
->users
) == 1))
465 else if (likely(!atomic_dec_and_test(&skb
->users
)))
469 EXPORT_SYMBOL(consume_skb
);
472 * skb_recycle_check - check if skb can be reused for receive
474 * @skb_size: minimum receive buffer size
476 * Checks that the skb passed in is not shared or cloned, and
477 * that it is linear and its head portion at least as large as
478 * skb_size so that it can be recycled as a receive buffer.
479 * If these conditions are met, this function does any necessary
480 * reference count dropping and cleans up the skbuff as if it
481 * just came from __alloc_skb().
483 int skb_recycle_check(struct sk_buff
*skb
, int skb_size
)
485 struct skb_shared_info
*shinfo
;
487 if (skb_is_nonlinear(skb
) || skb
->fclone
!= SKB_FCLONE_UNAVAILABLE
)
490 skb_size
= SKB_DATA_ALIGN(skb_size
+ NET_SKB_PAD
);
491 if (skb_end_pointer(skb
) - skb
->head
< skb_size
)
494 if (skb_shared(skb
) || skb_cloned(skb
))
497 skb_release_head_state(skb
);
498 shinfo
= skb_shinfo(skb
);
499 atomic_set(&shinfo
->dataref
, 1);
500 shinfo
->nr_frags
= 0;
501 shinfo
->gso_size
= 0;
502 shinfo
->gso_segs
= 0;
503 shinfo
->gso_type
= 0;
504 shinfo
->ip6_frag_id
= 0;
505 shinfo
->tx_flags
.flags
= 0;
506 shinfo
->frag_list
= NULL
;
507 memset(&shinfo
->hwtstamps
, 0, sizeof(shinfo
->hwtstamps
));
509 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
510 skb
->data
= skb
->head
+ NET_SKB_PAD
;
511 skb_reset_tail_pointer(skb
);
515 EXPORT_SYMBOL(skb_recycle_check
);
517 static void __copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
519 new->tstamp
= old
->tstamp
;
521 new->transport_header
= old
->transport_header
;
522 new->network_header
= old
->network_header
;
523 new->mac_header
= old
->mac_header
;
524 new->dst
= dst_clone(old
->dst
);
526 new->sp
= secpath_get(old
->sp
);
528 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
529 new->csum_start
= old
->csum_start
;
530 new->csum_offset
= old
->csum_offset
;
531 new->local_df
= old
->local_df
;
532 new->pkt_type
= old
->pkt_type
;
533 new->ip_summed
= old
->ip_summed
;
534 skb_copy_queue_mapping(new, old
);
535 new->priority
= old
->priority
;
536 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
537 new->ipvs_property
= old
->ipvs_property
;
539 new->protocol
= old
->protocol
;
540 new->mark
= old
->mark
;
543 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
544 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
545 new->nf_trace
= old
->nf_trace
;
547 #ifdef CONFIG_NET_SCHED
548 new->tc_index
= old
->tc_index
;
549 #ifdef CONFIG_NET_CLS_ACT
550 new->tc_verd
= old
->tc_verd
;
553 new->vlan_tci
= old
->vlan_tci
;
554 #if defined(CONFIG_MAC80211) || defined(CONFIG_MAC80211_MODULE)
555 new->do_not_encrypt
= old
->do_not_encrypt
;
556 new->requeue
= old
->requeue
;
559 skb_copy_secmark(new, old
);
563 * You should not add any new code to this function. Add it to
564 * __copy_skb_header above instead.
566 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
568 #define C(x) n->x = skb->x
570 n
->next
= n
->prev
= NULL
;
572 __copy_skb_header(n
, skb
);
577 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
580 n
->destructor
= NULL
;
586 atomic_set(&n
->users
, 1);
588 atomic_inc(&(skb_shinfo(skb
)->dataref
));
596 * skb_morph - morph one skb into another
597 * @dst: the skb to receive the contents
598 * @src: the skb to supply the contents
600 * This is identical to skb_clone except that the target skb is
601 * supplied by the user.
603 * The target skb is returned upon exit.
605 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
607 skb_release_all(dst
);
608 return __skb_clone(dst
, src
);
610 EXPORT_SYMBOL_GPL(skb_morph
);
613 * skb_clone - duplicate an sk_buff
614 * @skb: buffer to clone
615 * @gfp_mask: allocation priority
617 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
618 * copies share the same packet data but not structure. The new
619 * buffer has a reference count of 1. If the allocation fails the
620 * function returns %NULL otherwise the new buffer is returned.
622 * If this function is called from an interrupt gfp_mask() must be
626 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
631 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
632 n
->fclone
== SKB_FCLONE_UNAVAILABLE
) {
633 atomic_t
*fclone_ref
= (atomic_t
*) (n
+ 1);
634 n
->fclone
= SKB_FCLONE_CLONE
;
635 atomic_inc(fclone_ref
);
637 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
640 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
643 return __skb_clone(n
, skb
);
645 EXPORT_SYMBOL(skb_clone
);
647 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
649 #ifndef NET_SKBUFF_DATA_USES_OFFSET
651 * Shift between the two data areas in bytes
653 unsigned long offset
= new->data
- old
->data
;
656 __copy_skb_header(new, old
);
658 #ifndef NET_SKBUFF_DATA_USES_OFFSET
659 /* {transport,network,mac}_header are relative to skb->head */
660 new->transport_header
+= offset
;
661 new->network_header
+= offset
;
662 new->mac_header
+= offset
;
664 skb_shinfo(new)->gso_size
= skb_shinfo(old
)->gso_size
;
665 skb_shinfo(new)->gso_segs
= skb_shinfo(old
)->gso_segs
;
666 skb_shinfo(new)->gso_type
= skb_shinfo(old
)->gso_type
;
670 * skb_copy - create private copy of an sk_buff
671 * @skb: buffer to copy
672 * @gfp_mask: allocation priority
674 * Make a copy of both an &sk_buff and its data. This is used when the
675 * caller wishes to modify the data and needs a private copy of the
676 * data to alter. Returns %NULL on failure or the pointer to the buffer
677 * on success. The returned buffer has a reference count of 1.
679 * As by-product this function converts non-linear &sk_buff to linear
680 * one, so that &sk_buff becomes completely private and caller is allowed
681 * to modify all the data of returned buffer. This means that this
682 * function is not recommended for use in circumstances when only
683 * header is going to be modified. Use pskb_copy() instead.
686 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
688 int headerlen
= skb
->data
- skb
->head
;
690 * Allocate the copy buffer
693 #ifdef NET_SKBUFF_DATA_USES_OFFSET
694 n
= alloc_skb(skb
->end
+ skb
->data_len
, gfp_mask
);
696 n
= alloc_skb(skb
->end
- skb
->head
+ skb
->data_len
, gfp_mask
);
701 /* Set the data pointer */
702 skb_reserve(n
, headerlen
);
703 /* Set the tail pointer and length */
704 skb_put(n
, skb
->len
);
706 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
709 copy_skb_header(n
, skb
);
712 EXPORT_SYMBOL(skb_copy
);
715 * pskb_copy - create copy of an sk_buff with private head.
716 * @skb: buffer to copy
717 * @gfp_mask: allocation priority
719 * Make a copy of both an &sk_buff and part of its data, located
720 * in header. Fragmented data remain shared. This is used when
721 * the caller wishes to modify only header of &sk_buff and needs
722 * private copy of the header to alter. Returns %NULL on failure
723 * or the pointer to the buffer on success.
724 * The returned buffer has a reference count of 1.
727 struct sk_buff
*pskb_copy(struct sk_buff
*skb
, gfp_t gfp_mask
)
730 * Allocate the copy buffer
733 #ifdef NET_SKBUFF_DATA_USES_OFFSET
734 n
= alloc_skb(skb
->end
, gfp_mask
);
736 n
= alloc_skb(skb
->end
- skb
->head
, gfp_mask
);
741 /* Set the data pointer */
742 skb_reserve(n
, skb
->data
- skb
->head
);
743 /* Set the tail pointer and length */
744 skb_put(n
, skb_headlen(skb
));
746 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
748 n
->truesize
+= skb
->data_len
;
749 n
->data_len
= skb
->data_len
;
752 if (skb_shinfo(skb
)->nr_frags
) {
755 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
756 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
757 get_page(skb_shinfo(n
)->frags
[i
].page
);
759 skb_shinfo(n
)->nr_frags
= i
;
762 if (skb_shinfo(skb
)->frag_list
) {
763 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
764 skb_clone_fraglist(n
);
767 copy_skb_header(n
, skb
);
771 EXPORT_SYMBOL(pskb_copy
);
774 * pskb_expand_head - reallocate header of &sk_buff
775 * @skb: buffer to reallocate
776 * @nhead: room to add at head
777 * @ntail: room to add at tail
778 * @gfp_mask: allocation priority
780 * Expands (or creates identical copy, if &nhead and &ntail are zero)
781 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
782 * reference count of 1. Returns zero in the case of success or error,
783 * if expansion failed. In the last case, &sk_buff is not changed.
785 * All the pointers pointing into skb header may change and must be
786 * reloaded after call to this function.
789 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
794 #ifdef NET_SKBUFF_DATA_USES_OFFSET
795 int size
= nhead
+ skb
->end
+ ntail
;
797 int size
= nhead
+ (skb
->end
- skb
->head
) + ntail
;
806 size
= SKB_DATA_ALIGN(size
);
808 data
= kmalloc(size
+ sizeof(struct skb_shared_info
), gfp_mask
);
812 /* Copy only real data... and, alas, header. This should be
813 * optimized for the cases when header is void. */
814 #ifdef NET_SKBUFF_DATA_USES_OFFSET
815 memcpy(data
+ nhead
, skb
->head
, skb
->tail
);
817 memcpy(data
+ nhead
, skb
->head
, skb
->tail
- skb
->head
);
819 memcpy(data
+ size
, skb_end_pointer(skb
),
820 sizeof(struct skb_shared_info
));
822 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
823 get_page(skb_shinfo(skb
)->frags
[i
].page
);
825 if (skb_shinfo(skb
)->frag_list
)
826 skb_clone_fraglist(skb
);
828 skb_release_data(skb
);
830 off
= (data
+ nhead
) - skb
->head
;
834 #ifdef NET_SKBUFF_DATA_USES_OFFSET
838 skb
->end
= skb
->head
+ size
;
840 /* {transport,network,mac}_header and tail are relative to skb->head */
842 skb
->transport_header
+= off
;
843 skb
->network_header
+= off
;
844 skb
->mac_header
+= off
;
845 skb
->csum_start
+= nhead
;
849 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
855 EXPORT_SYMBOL(pskb_expand_head
);
857 /* Make private copy of skb with writable head and some headroom */
859 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
861 struct sk_buff
*skb2
;
862 int delta
= headroom
- skb_headroom(skb
);
865 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
867 skb2
= skb_clone(skb
, GFP_ATOMIC
);
868 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
876 EXPORT_SYMBOL(skb_realloc_headroom
);
879 * skb_copy_expand - copy and expand sk_buff
880 * @skb: buffer to copy
881 * @newheadroom: new free bytes at head
882 * @newtailroom: new free bytes at tail
883 * @gfp_mask: allocation priority
885 * Make a copy of both an &sk_buff and its data and while doing so
886 * allocate additional space.
888 * This is used when the caller wishes to modify the data and needs a
889 * private copy of the data to alter as well as more space for new fields.
890 * Returns %NULL on failure or the pointer to the buffer
891 * on success. The returned buffer has a reference count of 1.
893 * You must pass %GFP_ATOMIC as the allocation priority if this function
894 * is called from an interrupt.
896 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
897 int newheadroom
, int newtailroom
,
901 * Allocate the copy buffer
903 struct sk_buff
*n
= alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
905 int oldheadroom
= skb_headroom(skb
);
906 int head_copy_len
, head_copy_off
;
912 skb_reserve(n
, newheadroom
);
914 /* Set the tail pointer and length */
915 skb_put(n
, skb
->len
);
917 head_copy_len
= oldheadroom
;
919 if (newheadroom
<= head_copy_len
)
920 head_copy_len
= newheadroom
;
922 head_copy_off
= newheadroom
- head_copy_len
;
924 /* Copy the linear header and data. */
925 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
926 skb
->len
+ head_copy_len
))
929 copy_skb_header(n
, skb
);
931 off
= newheadroom
- oldheadroom
;
932 n
->csum_start
+= off
;
933 #ifdef NET_SKBUFF_DATA_USES_OFFSET
934 n
->transport_header
+= off
;
935 n
->network_header
+= off
;
936 n
->mac_header
+= off
;
941 EXPORT_SYMBOL(skb_copy_expand
);
944 * skb_pad - zero pad the tail of an skb
945 * @skb: buffer to pad
948 * Ensure that a buffer is followed by a padding area that is zero
949 * filled. Used by network drivers which may DMA or transfer data
950 * beyond the buffer end onto the wire.
952 * May return error in out of memory cases. The skb is freed on error.
955 int skb_pad(struct sk_buff
*skb
, int pad
)
960 /* If the skbuff is non linear tailroom is always zero.. */
961 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
962 memset(skb
->data
+skb
->len
, 0, pad
);
966 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
967 if (likely(skb_cloned(skb
) || ntail
> 0)) {
968 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
973 /* FIXME: The use of this function with non-linear skb's really needs
976 err
= skb_linearize(skb
);
980 memset(skb
->data
+ skb
->len
, 0, pad
);
987 EXPORT_SYMBOL(skb_pad
);
990 * skb_put - add data to a buffer
991 * @skb: buffer to use
992 * @len: amount of data to add
994 * This function extends the used data area of the buffer. If this would
995 * exceed the total buffer size the kernel will panic. A pointer to the
996 * first byte of the extra data is returned.
998 unsigned char *skb_put(struct sk_buff
*skb
, unsigned int len
)
1000 unsigned char *tmp
= skb_tail_pointer(skb
);
1001 SKB_LINEAR_ASSERT(skb
);
1004 if (unlikely(skb
->tail
> skb
->end
))
1005 skb_over_panic(skb
, len
, __builtin_return_address(0));
1008 EXPORT_SYMBOL(skb_put
);
1011 * skb_push - add data to the start of a buffer
1012 * @skb: buffer to use
1013 * @len: amount of data to add
1015 * This function extends the used data area of the buffer at the buffer
1016 * start. If this would exceed the total buffer headroom the kernel will
1017 * panic. A pointer to the first byte of the extra data is returned.
1019 unsigned char *skb_push(struct sk_buff
*skb
, unsigned int len
)
1023 if (unlikely(skb
->data
<skb
->head
))
1024 skb_under_panic(skb
, len
, __builtin_return_address(0));
1027 EXPORT_SYMBOL(skb_push
);
1030 * skb_pull - remove data from the start of a buffer
1031 * @skb: buffer to use
1032 * @len: amount of data to remove
1034 * This function removes data from the start of a buffer, returning
1035 * the memory to the headroom. A pointer to the next data in the buffer
1036 * is returned. Once the data has been pulled future pushes will overwrite
1039 unsigned char *skb_pull(struct sk_buff
*skb
, unsigned int len
)
1041 return unlikely(len
> skb
->len
) ? NULL
: __skb_pull(skb
, len
);
1043 EXPORT_SYMBOL(skb_pull
);
1046 * skb_trim - remove end from a buffer
1047 * @skb: buffer to alter
1050 * Cut the length of a buffer down by removing data from the tail. If
1051 * the buffer is already under the length specified it is not modified.
1052 * The skb must be linear.
1054 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1057 __skb_trim(skb
, len
);
1059 EXPORT_SYMBOL(skb_trim
);
1061 /* Trims skb to length len. It can change skb pointers.
1064 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1066 struct sk_buff
**fragp
;
1067 struct sk_buff
*frag
;
1068 int offset
= skb_headlen(skb
);
1069 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1073 if (skb_cloned(skb
) &&
1074 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1081 for (; i
< nfrags
; i
++) {
1082 int end
= offset
+ skb_shinfo(skb
)->frags
[i
].size
;
1089 skb_shinfo(skb
)->frags
[i
++].size
= len
- offset
;
1092 skb_shinfo(skb
)->nr_frags
= i
;
1094 for (; i
< nfrags
; i
++)
1095 put_page(skb_shinfo(skb
)->frags
[i
].page
);
1097 if (skb_shinfo(skb
)->frag_list
)
1098 skb_drop_fraglist(skb
);
1102 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1103 fragp
= &frag
->next
) {
1104 int end
= offset
+ frag
->len
;
1106 if (skb_shared(frag
)) {
1107 struct sk_buff
*nfrag
;
1109 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1110 if (unlikely(!nfrag
))
1113 nfrag
->next
= frag
->next
;
1125 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1129 skb_drop_list(&frag
->next
);
1134 if (len
> skb_headlen(skb
)) {
1135 skb
->data_len
-= skb
->len
- len
;
1140 skb_set_tail_pointer(skb
, len
);
1145 EXPORT_SYMBOL(___pskb_trim
);
1148 * __pskb_pull_tail - advance tail of skb header
1149 * @skb: buffer to reallocate
1150 * @delta: number of bytes to advance tail
1152 * The function makes a sense only on a fragmented &sk_buff,
1153 * it expands header moving its tail forward and copying necessary
1154 * data from fragmented part.
1156 * &sk_buff MUST have reference count of 1.
1158 * Returns %NULL (and &sk_buff does not change) if pull failed
1159 * or value of new tail of skb in the case of success.
1161 * All the pointers pointing into skb header may change and must be
1162 * reloaded after call to this function.
1165 /* Moves tail of skb head forward, copying data from fragmented part,
1166 * when it is necessary.
1167 * 1. It may fail due to malloc failure.
1168 * 2. It may change skb pointers.
1170 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1172 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1174 /* If skb has not enough free space at tail, get new one
1175 * plus 128 bytes for future expansions. If we have enough
1176 * room at tail, reallocate without expansion only if skb is cloned.
1178 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1180 if (eat
> 0 || skb_cloned(skb
)) {
1181 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1186 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
1189 /* Optimization: no fragments, no reasons to preestimate
1190 * size of pulled pages. Superb.
1192 if (!skb_shinfo(skb
)->frag_list
)
1195 /* Estimate size of pulled pages. */
1197 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1198 if (skb_shinfo(skb
)->frags
[i
].size
>= eat
)
1200 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1203 /* If we need update frag list, we are in troubles.
1204 * Certainly, it possible to add an offset to skb data,
1205 * but taking into account that pulling is expected to
1206 * be very rare operation, it is worth to fight against
1207 * further bloating skb head and crucify ourselves here instead.
1208 * Pure masohism, indeed. 8)8)
1211 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1212 struct sk_buff
*clone
= NULL
;
1213 struct sk_buff
*insp
= NULL
;
1218 if (list
->len
<= eat
) {
1219 /* Eaten as whole. */
1224 /* Eaten partially. */
1226 if (skb_shared(list
)) {
1227 /* Sucks! We need to fork list. :-( */
1228 clone
= skb_clone(list
, GFP_ATOMIC
);
1234 /* This may be pulled without
1238 if (!pskb_pull(list
, eat
)) {
1246 /* Free pulled out fragments. */
1247 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1248 skb_shinfo(skb
)->frag_list
= list
->next
;
1251 /* And insert new clone at head. */
1254 skb_shinfo(skb
)->frag_list
= clone
;
1257 /* Success! Now we may commit changes to skb data. */
1262 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1263 if (skb_shinfo(skb
)->frags
[i
].size
<= eat
) {
1264 put_page(skb_shinfo(skb
)->frags
[i
].page
);
1265 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1267 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1269 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1270 skb_shinfo(skb
)->frags
[k
].size
-= eat
;
1276 skb_shinfo(skb
)->nr_frags
= k
;
1279 skb
->data_len
-= delta
;
1281 return skb_tail_pointer(skb
);
1283 EXPORT_SYMBOL(__pskb_pull_tail
);
1285 /* Copy some data bits from skb to kernel buffer. */
1287 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1290 int start
= skb_headlen(skb
);
1292 if (offset
> (int)skb
->len
- len
)
1296 if ((copy
= start
- offset
) > 0) {
1299 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
1300 if ((len
-= copy
) == 0)
1306 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1309 WARN_ON(start
> offset
+ len
);
1311 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1312 if ((copy
= end
- offset
) > 0) {
1318 vaddr
= kmap_skb_frag(&skb_shinfo(skb
)->frags
[i
]);
1320 vaddr
+ skb_shinfo(skb
)->frags
[i
].page_offset
+
1321 offset
- start
, copy
);
1322 kunmap_skb_frag(vaddr
);
1324 if ((len
-= copy
) == 0)
1332 if (skb_shinfo(skb
)->frag_list
) {
1333 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1335 for (; list
; list
= list
->next
) {
1338 WARN_ON(start
> offset
+ len
);
1340 end
= start
+ list
->len
;
1341 if ((copy
= end
- offset
) > 0) {
1344 if (skb_copy_bits(list
, offset
- start
,
1347 if ((len
-= copy
) == 0)
1361 EXPORT_SYMBOL(skb_copy_bits
);
1364 * Callback from splice_to_pipe(), if we need to release some pages
1365 * at the end of the spd in case we error'ed out in filling the pipe.
1367 static void sock_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
1369 put_page(spd
->pages
[i
]);
1372 static inline struct page
*linear_to_page(struct page
*page
, unsigned int *len
,
1373 unsigned int *offset
,
1374 struct sk_buff
*skb
, struct sock
*sk
)
1376 struct page
*p
= sk
->sk_sndmsg_page
;
1381 p
= sk
->sk_sndmsg_page
= alloc_pages(sk
->sk_allocation
, 0);
1385 off
= sk
->sk_sndmsg_off
= 0;
1386 /* hold one ref to this page until it's full */
1390 off
= sk
->sk_sndmsg_off
;
1391 mlen
= PAGE_SIZE
- off
;
1392 if (mlen
< 64 && mlen
< *len
) {
1397 *len
= min_t(unsigned int, *len
, mlen
);
1400 memcpy(page_address(p
) + off
, page_address(page
) + *offset
, *len
);
1401 sk
->sk_sndmsg_off
+= *len
;
1409 * Fill page/offset/length into spd, if it can hold more pages.
1411 static inline int spd_fill_page(struct splice_pipe_desc
*spd
, struct page
*page
,
1412 unsigned int *len
, unsigned int offset
,
1413 struct sk_buff
*skb
, int linear
,
1416 if (unlikely(spd
->nr_pages
== PIPE_BUFFERS
))
1420 page
= linear_to_page(page
, len
, &offset
, skb
, sk
);
1426 spd
->pages
[spd
->nr_pages
] = page
;
1427 spd
->partial
[spd
->nr_pages
].len
= *len
;
1428 spd
->partial
[spd
->nr_pages
].offset
= offset
;
1434 static inline void __segment_seek(struct page
**page
, unsigned int *poff
,
1435 unsigned int *plen
, unsigned int off
)
1440 n
= *poff
/ PAGE_SIZE
;
1442 *page
= nth_page(*page
, n
);
1444 *poff
= *poff
% PAGE_SIZE
;
1448 static inline int __splice_segment(struct page
*page
, unsigned int poff
,
1449 unsigned int plen
, unsigned int *off
,
1450 unsigned int *len
, struct sk_buff
*skb
,
1451 struct splice_pipe_desc
*spd
, int linear
,
1457 /* skip this segment if already processed */
1463 /* ignore any bits we already processed */
1465 __segment_seek(&page
, &poff
, &plen
, *off
);
1470 unsigned int flen
= min(*len
, plen
);
1472 /* the linear region may spread across several pages */
1473 flen
= min_t(unsigned int, flen
, PAGE_SIZE
- poff
);
1475 if (spd_fill_page(spd
, page
, &flen
, poff
, skb
, linear
, sk
))
1478 __segment_seek(&page
, &poff
, &plen
, flen
);
1481 } while (*len
&& plen
);
1487 * Map linear and fragment data from the skb to spd. It reports failure if the
1488 * pipe is full or if we already spliced the requested length.
1490 static int __skb_splice_bits(struct sk_buff
*skb
, unsigned int *offset
,
1491 unsigned int *len
, struct splice_pipe_desc
*spd
,
1497 * map the linear part
1499 if (__splice_segment(virt_to_page(skb
->data
),
1500 (unsigned long) skb
->data
& (PAGE_SIZE
- 1),
1502 offset
, len
, skb
, spd
, 1, sk
))
1506 * then map the fragments
1508 for (seg
= 0; seg
< skb_shinfo(skb
)->nr_frags
; seg
++) {
1509 const skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[seg
];
1511 if (__splice_segment(f
->page
, f
->page_offset
, f
->size
,
1512 offset
, len
, skb
, spd
, 0, sk
))
1520 * Map data from the skb to a pipe. Should handle both the linear part,
1521 * the fragments, and the frag list. It does NOT handle frag lists within
1522 * the frag list, if such a thing exists. We'd probably need to recurse to
1523 * handle that cleanly.
1525 int skb_splice_bits(struct sk_buff
*skb
, unsigned int offset
,
1526 struct pipe_inode_info
*pipe
, unsigned int tlen
,
1529 struct partial_page partial
[PIPE_BUFFERS
];
1530 struct page
*pages
[PIPE_BUFFERS
];
1531 struct splice_pipe_desc spd
= {
1535 .ops
= &sock_pipe_buf_ops
,
1536 .spd_release
= sock_spd_release
,
1538 struct sock
*sk
= skb
->sk
;
1541 * __skb_splice_bits() only fails if the output has no room left,
1542 * so no point in going over the frag_list for the error case.
1544 if (__skb_splice_bits(skb
, &offset
, &tlen
, &spd
, sk
))
1550 * now see if we have a frag_list to map
1552 if (skb_shinfo(skb
)->frag_list
) {
1553 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1555 for (; list
&& tlen
; list
= list
->next
) {
1556 if (__skb_splice_bits(list
, &offset
, &tlen
, &spd
, sk
))
1566 * Drop the socket lock, otherwise we have reverse
1567 * locking dependencies between sk_lock and i_mutex
1568 * here as compared to sendfile(). We enter here
1569 * with the socket lock held, and splice_to_pipe() will
1570 * grab the pipe inode lock. For sendfile() emulation,
1571 * we call into ->sendpage() with the i_mutex lock held
1572 * and networking will grab the socket lock.
1575 ret
= splice_to_pipe(pipe
, &spd
);
1584 * skb_store_bits - store bits from kernel buffer to skb
1585 * @skb: destination buffer
1586 * @offset: offset in destination
1587 * @from: source buffer
1588 * @len: number of bytes to copy
1590 * Copy the specified number of bytes from the source buffer to the
1591 * destination skb. This function handles all the messy bits of
1592 * traversing fragment lists and such.
1595 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
1598 int start
= skb_headlen(skb
);
1600 if (offset
> (int)skb
->len
- len
)
1603 if ((copy
= start
- offset
) > 0) {
1606 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
1607 if ((len
-= copy
) == 0)
1613 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1614 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1617 WARN_ON(start
> offset
+ len
);
1619 end
= start
+ frag
->size
;
1620 if ((copy
= end
- offset
) > 0) {
1626 vaddr
= kmap_skb_frag(frag
);
1627 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1629 kunmap_skb_frag(vaddr
);
1631 if ((len
-= copy
) == 0)
1639 if (skb_shinfo(skb
)->frag_list
) {
1640 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1642 for (; list
; list
= list
->next
) {
1645 WARN_ON(start
> offset
+ len
);
1647 end
= start
+ list
->len
;
1648 if ((copy
= end
- offset
) > 0) {
1651 if (skb_store_bits(list
, offset
- start
,
1654 if ((len
-= copy
) == 0)
1668 EXPORT_SYMBOL(skb_store_bits
);
1670 /* Checksum skb data. */
1672 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
1673 int len
, __wsum csum
)
1675 int start
= skb_headlen(skb
);
1676 int i
, copy
= start
- offset
;
1679 /* Checksum header. */
1683 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1684 if ((len
-= copy
) == 0)
1690 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1693 WARN_ON(start
> offset
+ len
);
1695 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1696 if ((copy
= end
- offset
) > 0) {
1699 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1703 vaddr
= kmap_skb_frag(frag
);
1704 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
1705 offset
- start
, copy
, 0);
1706 kunmap_skb_frag(vaddr
);
1707 csum
= csum_block_add(csum
, csum2
, pos
);
1716 if (skb_shinfo(skb
)->frag_list
) {
1717 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1719 for (; list
; list
= list
->next
) {
1722 WARN_ON(start
> offset
+ len
);
1724 end
= start
+ list
->len
;
1725 if ((copy
= end
- offset
) > 0) {
1729 csum2
= skb_checksum(list
, offset
- start
,
1731 csum
= csum_block_add(csum
, csum2
, pos
);
1732 if ((len
-= copy
) == 0)
1744 EXPORT_SYMBOL(skb_checksum
);
1746 /* Both of above in one bottle. */
1748 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
1749 u8
*to
, int len
, __wsum csum
)
1751 int start
= skb_headlen(skb
);
1752 int i
, copy
= start
- offset
;
1759 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
1761 if ((len
-= copy
) == 0)
1768 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1771 WARN_ON(start
> offset
+ len
);
1773 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1774 if ((copy
= end
- offset
) > 0) {
1777 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1781 vaddr
= kmap_skb_frag(frag
);
1782 csum2
= csum_partial_copy_nocheck(vaddr
+
1786 kunmap_skb_frag(vaddr
);
1787 csum
= csum_block_add(csum
, csum2
, pos
);
1797 if (skb_shinfo(skb
)->frag_list
) {
1798 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1800 for (; list
; list
= list
->next
) {
1804 WARN_ON(start
> offset
+ len
);
1806 end
= start
+ list
->len
;
1807 if ((copy
= end
- offset
) > 0) {
1810 csum2
= skb_copy_and_csum_bits(list
,
1813 csum
= csum_block_add(csum
, csum2
, pos
);
1814 if ((len
-= copy
) == 0)
1826 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
1828 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
1833 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1834 csstart
= skb
->csum_start
- skb_headroom(skb
);
1836 csstart
= skb_headlen(skb
);
1838 BUG_ON(csstart
> skb_headlen(skb
));
1840 skb_copy_from_linear_data(skb
, to
, csstart
);
1843 if (csstart
!= skb
->len
)
1844 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
1845 skb
->len
- csstart
, 0);
1847 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1848 long csstuff
= csstart
+ skb
->csum_offset
;
1850 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
1853 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
1856 * skb_dequeue - remove from the head of the queue
1857 * @list: list to dequeue from
1859 * Remove the head of the list. The list lock is taken so the function
1860 * may be used safely with other locking list functions. The head item is
1861 * returned or %NULL if the list is empty.
1864 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
1866 unsigned long flags
;
1867 struct sk_buff
*result
;
1869 spin_lock_irqsave(&list
->lock
, flags
);
1870 result
= __skb_dequeue(list
);
1871 spin_unlock_irqrestore(&list
->lock
, flags
);
1874 EXPORT_SYMBOL(skb_dequeue
);
1877 * skb_dequeue_tail - remove from the tail of the queue
1878 * @list: list to dequeue from
1880 * Remove the tail of the list. The list lock is taken so the function
1881 * may be used safely with other locking list functions. The tail item is
1882 * returned or %NULL if the list is empty.
1884 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
1886 unsigned long flags
;
1887 struct sk_buff
*result
;
1889 spin_lock_irqsave(&list
->lock
, flags
);
1890 result
= __skb_dequeue_tail(list
);
1891 spin_unlock_irqrestore(&list
->lock
, flags
);
1894 EXPORT_SYMBOL(skb_dequeue_tail
);
1897 * skb_queue_purge - empty a list
1898 * @list: list to empty
1900 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1901 * the list and one reference dropped. This function takes the list
1902 * lock and is atomic with respect to other list locking functions.
1904 void skb_queue_purge(struct sk_buff_head
*list
)
1906 struct sk_buff
*skb
;
1907 while ((skb
= skb_dequeue(list
)) != NULL
)
1910 EXPORT_SYMBOL(skb_queue_purge
);
1913 * skb_queue_head - queue a buffer at the list head
1914 * @list: list to use
1915 * @newsk: buffer to queue
1917 * Queue a buffer at the start of the list. This function takes the
1918 * list lock and can be used safely with other locking &sk_buff functions
1921 * A buffer cannot be placed on two lists at the same time.
1923 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1925 unsigned long flags
;
1927 spin_lock_irqsave(&list
->lock
, flags
);
1928 __skb_queue_head(list
, newsk
);
1929 spin_unlock_irqrestore(&list
->lock
, flags
);
1931 EXPORT_SYMBOL(skb_queue_head
);
1934 * skb_queue_tail - queue a buffer at the list tail
1935 * @list: list to use
1936 * @newsk: buffer to queue
1938 * Queue a buffer at the tail of the list. This function takes the
1939 * list lock and can be used safely with other locking &sk_buff functions
1942 * A buffer cannot be placed on two lists at the same time.
1944 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1946 unsigned long flags
;
1948 spin_lock_irqsave(&list
->lock
, flags
);
1949 __skb_queue_tail(list
, newsk
);
1950 spin_unlock_irqrestore(&list
->lock
, flags
);
1952 EXPORT_SYMBOL(skb_queue_tail
);
1955 * skb_unlink - remove a buffer from a list
1956 * @skb: buffer to remove
1957 * @list: list to use
1959 * Remove a packet from a list. The list locks are taken and this
1960 * function is atomic with respect to other list locked calls
1962 * You must know what list the SKB is on.
1964 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
1966 unsigned long flags
;
1968 spin_lock_irqsave(&list
->lock
, flags
);
1969 __skb_unlink(skb
, list
);
1970 spin_unlock_irqrestore(&list
->lock
, flags
);
1972 EXPORT_SYMBOL(skb_unlink
);
1975 * skb_append - append a buffer
1976 * @old: buffer to insert after
1977 * @newsk: buffer to insert
1978 * @list: list to use
1980 * Place a packet after a given packet in a list. The list locks are taken
1981 * and this function is atomic with respect to other list locked calls.
1982 * A buffer cannot be placed on two lists at the same time.
1984 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
1986 unsigned long flags
;
1988 spin_lock_irqsave(&list
->lock
, flags
);
1989 __skb_queue_after(list
, old
, newsk
);
1990 spin_unlock_irqrestore(&list
->lock
, flags
);
1992 EXPORT_SYMBOL(skb_append
);
1995 * skb_insert - insert a buffer
1996 * @old: buffer to insert before
1997 * @newsk: buffer to insert
1998 * @list: list to use
2000 * Place a packet before a given packet in a list. The list locks are
2001 * taken and this function is atomic with respect to other list locked
2004 * A buffer cannot be placed on two lists at the same time.
2006 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2008 unsigned long flags
;
2010 spin_lock_irqsave(&list
->lock
, flags
);
2011 __skb_insert(newsk
, old
->prev
, old
, list
);
2012 spin_unlock_irqrestore(&list
->lock
, flags
);
2014 EXPORT_SYMBOL(skb_insert
);
2016 static inline void skb_split_inside_header(struct sk_buff
*skb
,
2017 struct sk_buff
* skb1
,
2018 const u32 len
, const int pos
)
2022 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
2024 /* And move data appendix as is. */
2025 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
2026 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
2028 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
2029 skb_shinfo(skb
)->nr_frags
= 0;
2030 skb1
->data_len
= skb
->data_len
;
2031 skb1
->len
+= skb1
->data_len
;
2034 skb_set_tail_pointer(skb
, len
);
2037 static inline void skb_split_no_header(struct sk_buff
*skb
,
2038 struct sk_buff
* skb1
,
2039 const u32 len
, int pos
)
2042 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
2044 skb_shinfo(skb
)->nr_frags
= 0;
2045 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
2047 skb
->data_len
= len
- pos
;
2049 for (i
= 0; i
< nfrags
; i
++) {
2050 int size
= skb_shinfo(skb
)->frags
[i
].size
;
2052 if (pos
+ size
> len
) {
2053 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
2057 * We have two variants in this case:
2058 * 1. Move all the frag to the second
2059 * part, if it is possible. F.e.
2060 * this approach is mandatory for TUX,
2061 * where splitting is expensive.
2062 * 2. Split is accurately. We make this.
2064 get_page(skb_shinfo(skb
)->frags
[i
].page
);
2065 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
2066 skb_shinfo(skb1
)->frags
[0].size
-= len
- pos
;
2067 skb_shinfo(skb
)->frags
[i
].size
= len
- pos
;
2068 skb_shinfo(skb
)->nr_frags
++;
2072 skb_shinfo(skb
)->nr_frags
++;
2075 skb_shinfo(skb1
)->nr_frags
= k
;
2079 * skb_split - Split fragmented skb to two parts at length len.
2080 * @skb: the buffer to split
2081 * @skb1: the buffer to receive the second part
2082 * @len: new length for skb
2084 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
2086 int pos
= skb_headlen(skb
);
2088 if (len
< pos
) /* Split line is inside header. */
2089 skb_split_inside_header(skb
, skb1
, len
, pos
);
2090 else /* Second chunk has no header, nothing to copy. */
2091 skb_split_no_header(skb
, skb1
, len
, pos
);
2093 EXPORT_SYMBOL(skb_split
);
2095 /* Shifting from/to a cloned skb is a no-go.
2097 * Caller cannot keep skb_shinfo related pointers past calling here!
2099 static int skb_prepare_for_shift(struct sk_buff
*skb
)
2101 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2105 * skb_shift - Shifts paged data partially from skb to another
2106 * @tgt: buffer into which tail data gets added
2107 * @skb: buffer from which the paged data comes from
2108 * @shiftlen: shift up to this many bytes
2110 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2111 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2112 * It's up to caller to free skb if everything was shifted.
2114 * If @tgt runs out of frags, the whole operation is aborted.
2116 * Skb cannot include anything else but paged data while tgt is allowed
2117 * to have non-paged data as well.
2119 * TODO: full sized shift could be optimized but that would need
2120 * specialized skb free'er to handle frags without up-to-date nr_frags.
2122 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
2124 int from
, to
, merge
, todo
;
2125 struct skb_frag_struct
*fragfrom
, *fragto
;
2127 BUG_ON(shiftlen
> skb
->len
);
2128 BUG_ON(skb_headlen(skb
)); /* Would corrupt stream */
2132 to
= skb_shinfo(tgt
)->nr_frags
;
2133 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2135 /* Actual merge is delayed until the point when we know we can
2136 * commit all, so that we don't have to undo partial changes
2139 !skb_can_coalesce(tgt
, to
, fragfrom
->page
, fragfrom
->page_offset
)) {
2144 todo
-= fragfrom
->size
;
2146 if (skb_prepare_for_shift(skb
) ||
2147 skb_prepare_for_shift(tgt
))
2150 /* All previous frag pointers might be stale! */
2151 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2152 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2154 fragto
->size
+= shiftlen
;
2155 fragfrom
->size
-= shiftlen
;
2156 fragfrom
->page_offset
+= shiftlen
;
2164 /* Skip full, not-fitting skb to avoid expensive operations */
2165 if ((shiftlen
== skb
->len
) &&
2166 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
2169 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
2172 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
2173 if (to
== MAX_SKB_FRAGS
)
2176 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2177 fragto
= &skb_shinfo(tgt
)->frags
[to
];
2179 if (todo
>= fragfrom
->size
) {
2180 *fragto
= *fragfrom
;
2181 todo
-= fragfrom
->size
;
2186 get_page(fragfrom
->page
);
2187 fragto
->page
= fragfrom
->page
;
2188 fragto
->page_offset
= fragfrom
->page_offset
;
2189 fragto
->size
= todo
;
2191 fragfrom
->page_offset
+= todo
;
2192 fragfrom
->size
-= todo
;
2200 /* Ready to "commit" this state change to tgt */
2201 skb_shinfo(tgt
)->nr_frags
= to
;
2204 fragfrom
= &skb_shinfo(skb
)->frags
[0];
2205 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2207 fragto
->size
+= fragfrom
->size
;
2208 put_page(fragfrom
->page
);
2211 /* Reposition in the original skb */
2213 while (from
< skb_shinfo(skb
)->nr_frags
)
2214 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
2215 skb_shinfo(skb
)->nr_frags
= to
;
2217 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
2220 /* Most likely the tgt won't ever need its checksum anymore, skb on
2221 * the other hand might need it if it needs to be resent
2223 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
2224 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2226 /* Yak, is it really working this way? Some helper please? */
2227 skb
->len
-= shiftlen
;
2228 skb
->data_len
-= shiftlen
;
2229 skb
->truesize
-= shiftlen
;
2230 tgt
->len
+= shiftlen
;
2231 tgt
->data_len
+= shiftlen
;
2232 tgt
->truesize
+= shiftlen
;
2238 * skb_prepare_seq_read - Prepare a sequential read of skb data
2239 * @skb: the buffer to read
2240 * @from: lower offset of data to be read
2241 * @to: upper offset of data to be read
2242 * @st: state variable
2244 * Initializes the specified state variable. Must be called before
2245 * invoking skb_seq_read() for the first time.
2247 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
2248 unsigned int to
, struct skb_seq_state
*st
)
2250 st
->lower_offset
= from
;
2251 st
->upper_offset
= to
;
2252 st
->root_skb
= st
->cur_skb
= skb
;
2253 st
->frag_idx
= st
->stepped_offset
= 0;
2254 st
->frag_data
= NULL
;
2256 EXPORT_SYMBOL(skb_prepare_seq_read
);
2259 * skb_seq_read - Sequentially read skb data
2260 * @consumed: number of bytes consumed by the caller so far
2261 * @data: destination pointer for data to be returned
2262 * @st: state variable
2264 * Reads a block of skb data at &consumed relative to the
2265 * lower offset specified to skb_prepare_seq_read(). Assigns
2266 * the head of the data block to &data and returns the length
2267 * of the block or 0 if the end of the skb data or the upper
2268 * offset has been reached.
2270 * The caller is not required to consume all of the data
2271 * returned, i.e. &consumed is typically set to the number
2272 * of bytes already consumed and the next call to
2273 * skb_seq_read() will return the remaining part of the block.
2275 * Note 1: The size of each block of data returned can be arbitary,
2276 * this limitation is the cost for zerocopy seqeuental
2277 * reads of potentially non linear data.
2279 * Note 2: Fragment lists within fragments are not implemented
2280 * at the moment, state->root_skb could be replaced with
2281 * a stack for this purpose.
2283 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
2284 struct skb_seq_state
*st
)
2286 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
2289 if (unlikely(abs_offset
>= st
->upper_offset
))
2293 block_limit
= skb_headlen(st
->cur_skb
) + st
->stepped_offset
;
2295 if (abs_offset
< block_limit
) {
2296 *data
= st
->cur_skb
->data
+ (abs_offset
- st
->stepped_offset
);
2297 return block_limit
- abs_offset
;
2300 if (st
->frag_idx
== 0 && !st
->frag_data
)
2301 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
2303 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
2304 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
2305 block_limit
= frag
->size
+ st
->stepped_offset
;
2307 if (abs_offset
< block_limit
) {
2309 st
->frag_data
= kmap_skb_frag(frag
);
2311 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
2312 (abs_offset
- st
->stepped_offset
);
2314 return block_limit
- abs_offset
;
2317 if (st
->frag_data
) {
2318 kunmap_skb_frag(st
->frag_data
);
2319 st
->frag_data
= NULL
;
2323 st
->stepped_offset
+= frag
->size
;
2326 if (st
->frag_data
) {
2327 kunmap_skb_frag(st
->frag_data
);
2328 st
->frag_data
= NULL
;
2331 if (st
->root_skb
== st
->cur_skb
&&
2332 skb_shinfo(st
->root_skb
)->frag_list
) {
2333 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
2336 } else if (st
->cur_skb
->next
) {
2337 st
->cur_skb
= st
->cur_skb
->next
;
2344 EXPORT_SYMBOL(skb_seq_read
);
2347 * skb_abort_seq_read - Abort a sequential read of skb data
2348 * @st: state variable
2350 * Must be called if skb_seq_read() was not called until it
2353 void skb_abort_seq_read(struct skb_seq_state
*st
)
2356 kunmap_skb_frag(st
->frag_data
);
2358 EXPORT_SYMBOL(skb_abort_seq_read
);
2360 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2362 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
2363 struct ts_config
*conf
,
2364 struct ts_state
*state
)
2366 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
2369 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
2371 skb_abort_seq_read(TS_SKB_CB(state
));
2375 * skb_find_text - Find a text pattern in skb data
2376 * @skb: the buffer to look in
2377 * @from: search offset
2379 * @config: textsearch configuration
2380 * @state: uninitialized textsearch state variable
2382 * Finds a pattern in the skb data according to the specified
2383 * textsearch configuration. Use textsearch_next() to retrieve
2384 * subsequent occurrences of the pattern. Returns the offset
2385 * to the first occurrence or UINT_MAX if no match was found.
2387 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
2388 unsigned int to
, struct ts_config
*config
,
2389 struct ts_state
*state
)
2393 config
->get_next_block
= skb_ts_get_next_block
;
2394 config
->finish
= skb_ts_finish
;
2396 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
2398 ret
= textsearch_find(config
, state
);
2399 return (ret
<= to
- from
? ret
: UINT_MAX
);
2401 EXPORT_SYMBOL(skb_find_text
);
2404 * skb_append_datato_frags: - append the user data to a skb
2405 * @sk: sock structure
2406 * @skb: skb structure to be appened with user data.
2407 * @getfrag: call back function to be used for getting the user data
2408 * @from: pointer to user message iov
2409 * @length: length of the iov message
2411 * Description: This procedure append the user data in the fragment part
2412 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2414 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
2415 int (*getfrag
)(void *from
, char *to
, int offset
,
2416 int len
, int odd
, struct sk_buff
*skb
),
2417 void *from
, int length
)
2420 skb_frag_t
*frag
= NULL
;
2421 struct page
*page
= NULL
;
2427 /* Return error if we don't have space for new frag */
2428 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2429 if (frg_cnt
>= MAX_SKB_FRAGS
)
2432 /* allocate a new page for next frag */
2433 page
= alloc_pages(sk
->sk_allocation
, 0);
2435 /* If alloc_page fails just return failure and caller will
2436 * free previous allocated pages by doing kfree_skb()
2441 /* initialize the next frag */
2442 sk
->sk_sndmsg_page
= page
;
2443 sk
->sk_sndmsg_off
= 0;
2444 skb_fill_page_desc(skb
, frg_cnt
, page
, 0, 0);
2445 skb
->truesize
+= PAGE_SIZE
;
2446 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
2448 /* get the new initialized frag */
2449 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2450 frag
= &skb_shinfo(skb
)->frags
[frg_cnt
- 1];
2452 /* copy the user data to page */
2453 left
= PAGE_SIZE
- frag
->page_offset
;
2454 copy
= (length
> left
)? left
: length
;
2456 ret
= getfrag(from
, (page_address(frag
->page
) +
2457 frag
->page_offset
+ frag
->size
),
2458 offset
, copy
, 0, skb
);
2462 /* copy was successful so update the size parameters */
2463 sk
->sk_sndmsg_off
+= copy
;
2466 skb
->data_len
+= copy
;
2470 } while (length
> 0);
2474 EXPORT_SYMBOL(skb_append_datato_frags
);
2477 * skb_pull_rcsum - pull skb and update receive checksum
2478 * @skb: buffer to update
2479 * @len: length of data pulled
2481 * This function performs an skb_pull on the packet and updates
2482 * the CHECKSUM_COMPLETE checksum. It should be used on
2483 * receive path processing instead of skb_pull unless you know
2484 * that the checksum difference is zero (e.g., a valid IP header)
2485 * or you are setting ip_summed to CHECKSUM_NONE.
2487 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
2489 BUG_ON(len
> skb
->len
);
2491 BUG_ON(skb
->len
< skb
->data_len
);
2492 skb_postpull_rcsum(skb
, skb
->data
, len
);
2493 return skb
->data
+= len
;
2496 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
2499 * skb_segment - Perform protocol segmentation on skb.
2500 * @skb: buffer to segment
2501 * @features: features for the output path (see dev->features)
2503 * This function performs segmentation on the given skb. It returns
2504 * a pointer to the first in a list of new skbs for the segments.
2505 * In case of error it returns ERR_PTR(err).
2507 struct sk_buff
*skb_segment(struct sk_buff
*skb
, int features
)
2509 struct sk_buff
*segs
= NULL
;
2510 struct sk_buff
*tail
= NULL
;
2511 struct sk_buff
*fskb
= skb_shinfo(skb
)->frag_list
;
2512 unsigned int mss
= skb_shinfo(skb
)->gso_size
;
2513 unsigned int doffset
= skb
->data
- skb_mac_header(skb
);
2514 unsigned int offset
= doffset
;
2515 unsigned int headroom
;
2517 int sg
= features
& NETIF_F_SG
;
2518 int nfrags
= skb_shinfo(skb
)->nr_frags
;
2523 __skb_push(skb
, doffset
);
2524 headroom
= skb_headroom(skb
);
2525 pos
= skb_headlen(skb
);
2528 struct sk_buff
*nskb
;
2533 len
= skb
->len
- offset
;
2537 hsize
= skb_headlen(skb
) - offset
;
2540 if (hsize
> len
|| !sg
)
2543 if (!hsize
&& i
>= nfrags
) {
2544 BUG_ON(fskb
->len
!= len
);
2547 nskb
= skb_clone(fskb
, GFP_ATOMIC
);
2550 if (unlikely(!nskb
))
2553 hsize
= skb_end_pointer(nskb
) - nskb
->head
;
2554 if (skb_cow_head(nskb
, doffset
+ headroom
)) {
2559 nskb
->truesize
+= skb_end_pointer(nskb
) - nskb
->head
-
2561 skb_release_head_state(nskb
);
2562 __skb_push(nskb
, doffset
);
2564 nskb
= alloc_skb(hsize
+ doffset
+ headroom
,
2567 if (unlikely(!nskb
))
2570 skb_reserve(nskb
, headroom
);
2571 __skb_put(nskb
, doffset
);
2580 __copy_skb_header(nskb
, skb
);
2581 nskb
->mac_len
= skb
->mac_len
;
2583 skb_reset_mac_header(nskb
);
2584 skb_set_network_header(nskb
, skb
->mac_len
);
2585 nskb
->transport_header
= (nskb
->network_header
+
2586 skb_network_header_len(skb
));
2587 skb_copy_from_linear_data(skb
, nskb
->data
, doffset
);
2589 if (fskb
!= skb_shinfo(skb
)->frag_list
)
2593 nskb
->ip_summed
= CHECKSUM_NONE
;
2594 nskb
->csum
= skb_copy_and_csum_bits(skb
, offset
,
2600 frag
= skb_shinfo(nskb
)->frags
;
2602 skb_copy_from_linear_data_offset(skb
, offset
,
2603 skb_put(nskb
, hsize
), hsize
);
2605 while (pos
< offset
+ len
&& i
< nfrags
) {
2606 *frag
= skb_shinfo(skb
)->frags
[i
];
2607 get_page(frag
->page
);
2611 frag
->page_offset
+= offset
- pos
;
2612 frag
->size
-= offset
- pos
;
2615 skb_shinfo(nskb
)->nr_frags
++;
2617 if (pos
+ size
<= offset
+ len
) {
2621 frag
->size
-= pos
+ size
- (offset
+ len
);
2628 if (pos
< offset
+ len
) {
2629 struct sk_buff
*fskb2
= fskb
;
2631 BUG_ON(pos
+ fskb
->len
!= offset
+ len
);
2637 fskb2
= skb_clone(fskb2
, GFP_ATOMIC
);
2643 BUG_ON(skb_shinfo(nskb
)->frag_list
);
2644 skb_shinfo(nskb
)->frag_list
= fskb2
;
2648 nskb
->data_len
= len
- hsize
;
2649 nskb
->len
+= nskb
->data_len
;
2650 nskb
->truesize
+= nskb
->data_len
;
2651 } while ((offset
+= len
) < skb
->len
);
2656 while ((skb
= segs
)) {
2660 return ERR_PTR(err
);
2662 EXPORT_SYMBOL_GPL(skb_segment
);
2664 int skb_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2666 struct sk_buff
*p
= *head
;
2667 struct sk_buff
*nskb
;
2668 unsigned int headroom
;
2669 unsigned int len
= skb_gro_len(skb
);
2671 if (p
->len
+ len
>= 65536)
2674 if (skb_shinfo(p
)->frag_list
)
2676 else if (skb_headlen(skb
) <= skb_gro_offset(skb
)) {
2677 if (skb_shinfo(p
)->nr_frags
+ skb_shinfo(skb
)->nr_frags
>
2681 skb_shinfo(skb
)->frags
[0].page_offset
+=
2682 skb_gro_offset(skb
) - skb_headlen(skb
);
2683 skb_shinfo(skb
)->frags
[0].size
-=
2684 skb_gro_offset(skb
) - skb_headlen(skb
);
2686 memcpy(skb_shinfo(p
)->frags
+ skb_shinfo(p
)->nr_frags
,
2687 skb_shinfo(skb
)->frags
,
2688 skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
2690 skb_shinfo(p
)->nr_frags
+= skb_shinfo(skb
)->nr_frags
;
2691 skb_shinfo(skb
)->nr_frags
= 0;
2693 skb
->truesize
-= skb
->data_len
;
2694 skb
->len
-= skb
->data_len
;
2697 NAPI_GRO_CB(skb
)->free
= 1;
2701 headroom
= skb_headroom(p
);
2702 nskb
= netdev_alloc_skb(p
->dev
, headroom
+ skb_gro_offset(p
));
2703 if (unlikely(!nskb
))
2706 __copy_skb_header(nskb
, p
);
2707 nskb
->mac_len
= p
->mac_len
;
2709 skb_reserve(nskb
, headroom
);
2710 __skb_put(nskb
, skb_gro_offset(p
));
2712 skb_set_mac_header(nskb
, skb_mac_header(p
) - p
->data
);
2713 skb_set_network_header(nskb
, skb_network_offset(p
));
2714 skb_set_transport_header(nskb
, skb_transport_offset(p
));
2716 __skb_pull(p
, skb_gro_offset(p
));
2717 memcpy(skb_mac_header(nskb
), skb_mac_header(p
),
2718 p
->data
- skb_mac_header(p
));
2720 *NAPI_GRO_CB(nskb
) = *NAPI_GRO_CB(p
);
2721 skb_shinfo(nskb
)->frag_list
= p
;
2722 skb_shinfo(nskb
)->gso_size
= skb_shinfo(p
)->gso_size
;
2723 skb_header_release(p
);
2726 nskb
->data_len
+= p
->len
;
2727 nskb
->truesize
+= p
->len
;
2728 nskb
->len
+= p
->len
;
2731 nskb
->next
= p
->next
;
2737 if (skb_gro_offset(skb
) > skb_headlen(skb
)) {
2738 skb_shinfo(skb
)->frags
[0].page_offset
+=
2739 skb_gro_offset(skb
) - skb_headlen(skb
);
2740 skb_shinfo(skb
)->frags
[0].size
-=
2741 skb_gro_offset(skb
) - skb_headlen(skb
);
2742 skb_gro_reset_offset(skb
);
2743 skb_gro_pull(skb
, skb_headlen(skb
));
2746 __skb_pull(skb
, skb_gro_offset(skb
));
2748 p
->prev
->next
= skb
;
2750 skb_header_release(skb
);
2753 NAPI_GRO_CB(p
)->count
++;
2758 NAPI_GRO_CB(skb
)->same_flow
= 1;
2761 EXPORT_SYMBOL_GPL(skb_gro_receive
);
2763 void __init
skb_init(void)
2765 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
2766 sizeof(struct sk_buff
),
2768 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2770 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
2771 (2*sizeof(struct sk_buff
)) +
2774 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2779 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2780 * @skb: Socket buffer containing the buffers to be mapped
2781 * @sg: The scatter-gather list to map into
2782 * @offset: The offset into the buffer's contents to start mapping
2783 * @len: Length of buffer space to be mapped
2785 * Fill the specified scatter-gather list with mappings/pointers into a
2786 * region of the buffer space attached to a socket buffer.
2789 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2791 int start
= skb_headlen(skb
);
2792 int i
, copy
= start
- offset
;
2798 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
2800 if ((len
-= copy
) == 0)
2805 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2808 WARN_ON(start
> offset
+ len
);
2810 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
2811 if ((copy
= end
- offset
) > 0) {
2812 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2816 sg_set_page(&sg
[elt
], frag
->page
, copy
,
2817 frag
->page_offset
+offset
-start
);
2826 if (skb_shinfo(skb
)->frag_list
) {
2827 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
2829 for (; list
; list
= list
->next
) {
2832 WARN_ON(start
> offset
+ len
);
2834 end
= start
+ list
->len
;
2835 if ((copy
= end
- offset
) > 0) {
2838 elt
+= __skb_to_sgvec(list
, sg
+elt
, offset
- start
,
2840 if ((len
-= copy
) == 0)
2851 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2853 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
);
2855 sg_mark_end(&sg
[nsg
- 1]);
2859 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
2862 * skb_cow_data - Check that a socket buffer's data buffers are writable
2863 * @skb: The socket buffer to check.
2864 * @tailbits: Amount of trailing space to be added
2865 * @trailer: Returned pointer to the skb where the @tailbits space begins
2867 * Make sure that the data buffers attached to a socket buffer are
2868 * writable. If they are not, private copies are made of the data buffers
2869 * and the socket buffer is set to use these instead.
2871 * If @tailbits is given, make sure that there is space to write @tailbits
2872 * bytes of data beyond current end of socket buffer. @trailer will be
2873 * set to point to the skb in which this space begins.
2875 * The number of scatterlist elements required to completely map the
2876 * COW'd and extended socket buffer will be returned.
2878 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
2882 struct sk_buff
*skb1
, **skb_p
;
2884 /* If skb is cloned or its head is paged, reallocate
2885 * head pulling out all the pages (pages are considered not writable
2886 * at the moment even if they are anonymous).
2888 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
2889 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
2892 /* Easy case. Most of packets will go this way. */
2893 if (!skb_shinfo(skb
)->frag_list
) {
2894 /* A little of trouble, not enough of space for trailer.
2895 * This should not happen, when stack is tuned to generate
2896 * good frames. OK, on miss we reallocate and reserve even more
2897 * space, 128 bytes is fair. */
2899 if (skb_tailroom(skb
) < tailbits
&&
2900 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
2908 /* Misery. We are in troubles, going to mincer fragments... */
2911 skb_p
= &skb_shinfo(skb
)->frag_list
;
2914 while ((skb1
= *skb_p
) != NULL
) {
2917 /* The fragment is partially pulled by someone,
2918 * this can happen on input. Copy it and everything
2921 if (skb_shared(skb1
))
2924 /* If the skb is the last, worry about trailer. */
2926 if (skb1
->next
== NULL
&& tailbits
) {
2927 if (skb_shinfo(skb1
)->nr_frags
||
2928 skb_shinfo(skb1
)->frag_list
||
2929 skb_tailroom(skb1
) < tailbits
)
2930 ntail
= tailbits
+ 128;
2936 skb_shinfo(skb1
)->nr_frags
||
2937 skb_shinfo(skb1
)->frag_list
) {
2938 struct sk_buff
*skb2
;
2940 /* Fuck, we are miserable poor guys... */
2942 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
2944 skb2
= skb_copy_expand(skb1
,
2948 if (unlikely(skb2
== NULL
))
2952 skb_set_owner_w(skb2
, skb1
->sk
);
2954 /* Looking around. Are we still alive?
2955 * OK, link new skb, drop old one */
2957 skb2
->next
= skb1
->next
;
2964 skb_p
= &skb1
->next
;
2969 EXPORT_SYMBOL_GPL(skb_cow_data
);
2971 void skb_tstamp_tx(struct sk_buff
*orig_skb
,
2972 struct skb_shared_hwtstamps
*hwtstamps
)
2974 struct sock
*sk
= orig_skb
->sk
;
2975 struct sock_exterr_skb
*serr
;
2976 struct sk_buff
*skb
;
2982 skb
= skb_clone(orig_skb
, GFP_ATOMIC
);
2987 *skb_hwtstamps(skb
) =
2991 * no hardware time stamps available,
2992 * so keep the skb_shared_tx and only
2993 * store software time stamp
2995 skb
->tstamp
= ktime_get_real();
2998 serr
= SKB_EXT_ERR(skb
);
2999 memset(serr
, 0, sizeof(*serr
));
3000 serr
->ee
.ee_errno
= ENOMSG
;
3001 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TIMESTAMPING
;
3002 err
= sock_queue_err_skb(sk
, skb
);
3006 EXPORT_SYMBOL_GPL(skb_tstamp_tx
);
3010 * skb_partial_csum_set - set up and verify partial csum values for packet
3011 * @skb: the skb to set
3012 * @start: the number of bytes after skb->data to start checksumming.
3013 * @off: the offset from start to place the checksum.
3015 * For untrusted partially-checksummed packets, we need to make sure the values
3016 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3018 * This function checks and sets those values and skb->ip_summed: if this
3019 * returns false you should drop the packet.
3021 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
3023 if (unlikely(start
> skb
->len
- 2) ||
3024 unlikely((int)start
+ off
> skb
->len
- 2)) {
3025 if (net_ratelimit())
3027 "bad partial csum: csum=%u/%u len=%u\n",
3028 start
, off
, skb
->len
);
3031 skb
->ip_summed
= CHECKSUM_PARTIAL
;
3032 skb
->csum_start
= skb_headroom(skb
) + start
;
3033 skb
->csum_offset
= off
;
3036 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);
3038 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
3040 if (net_ratelimit())
3041 pr_warning("%s: received packets cannot be forwarded"
3042 " while LRO is enabled\n", skb
->dev
->name
);
3044 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);