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>
42 #include <linux/kmemcheck.h>
44 #include <linux/interrupt.h>
46 #include <linux/inet.h>
47 #include <linux/slab.h>
48 #include <linux/netdevice.h>
49 #ifdef CONFIG_NET_CLS_ACT
50 #include <net/pkt_sched.h>
52 #include <linux/string.h>
53 #include <linux/skbuff.h>
54 #include <linux/splice.h>
55 #include <linux/cache.h>
56 #include <linux/rtnetlink.h>
57 #include <linux/init.h>
58 #include <linux/scatterlist.h>
59 #include <linux/errqueue.h>
61 #include <net/protocol.h>
64 #include <net/checksum.h>
67 #include <asm/uaccess.h>
68 #include <asm/system.h>
69 #include <trace/events/skb.h>
73 static struct kmem_cache
*skbuff_head_cache __read_mostly
;
74 static struct kmem_cache
*skbuff_fclone_cache __read_mostly
;
76 static void sock_pipe_buf_release(struct pipe_inode_info
*pipe
,
77 struct pipe_buffer
*buf
)
82 static void sock_pipe_buf_get(struct pipe_inode_info
*pipe
,
83 struct pipe_buffer
*buf
)
88 static int sock_pipe_buf_steal(struct pipe_inode_info
*pipe
,
89 struct pipe_buffer
*buf
)
95 /* Pipe buffer operations for a socket. */
96 static struct pipe_buf_operations sock_pipe_buf_ops
= {
98 .map
= generic_pipe_buf_map
,
99 .unmap
= generic_pipe_buf_unmap
,
100 .confirm
= generic_pipe_buf_confirm
,
101 .release
= sock_pipe_buf_release
,
102 .steal
= sock_pipe_buf_steal
,
103 .get
= sock_pipe_buf_get
,
107 * Keep out-of-line to prevent kernel bloat.
108 * __builtin_return_address is not used because it is not always
113 * skb_over_panic - private function
118 * Out of line support code for skb_put(). Not user callable.
120 void skb_over_panic(struct sk_buff
*skb
, int sz
, void *here
)
122 printk(KERN_EMERG
"skb_over_panic: text:%p len:%d put:%d head:%p "
123 "data:%p tail:%#lx end:%#lx dev:%s\n",
124 here
, skb
->len
, sz
, skb
->head
, skb
->data
,
125 (unsigned long)skb
->tail
, (unsigned long)skb
->end
,
126 skb
->dev
? skb
->dev
->name
: "<NULL>");
129 EXPORT_SYMBOL(skb_over_panic
);
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>");
149 EXPORT_SYMBOL(skb_under_panic
);
151 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
152 * 'private' fields and also do memory statistics to find all the
158 * __alloc_skb - allocate a network buffer
159 * @size: size to allocate
160 * @gfp_mask: allocation mask
161 * @fclone: allocate from fclone cache instead of head cache
162 * and allocate a cloned (child) skb
163 * @node: numa node to allocate memory on
165 * Allocate a new &sk_buff. The returned buffer has no headroom and a
166 * tail room of size bytes. The object has a reference count of one.
167 * The return is the buffer. On a failure the return is %NULL.
169 * Buffers may only be allocated from interrupts using a @gfp_mask of
172 struct sk_buff
*__alloc_skb(unsigned int size
, gfp_t gfp_mask
,
173 int fclone
, int node
)
175 struct kmem_cache
*cache
;
176 struct skb_shared_info
*shinfo
;
180 cache
= fclone
? skbuff_fclone_cache
: skbuff_head_cache
;
183 skb
= kmem_cache_alloc_node(cache
, gfp_mask
& ~__GFP_DMA
, node
);
187 size
= SKB_DATA_ALIGN(size
);
188 data
= kmalloc_node_track_caller(size
+ sizeof(struct skb_shared_info
),
194 * Only clear those fields we need to clear, not those that we will
195 * actually initialise below. Hence, don't put any more fields after
196 * the tail pointer in struct sk_buff!
198 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
199 skb
->truesize
= size
+ sizeof(struct sk_buff
);
200 atomic_set(&skb
->users
, 1);
203 skb_reset_tail_pointer(skb
);
204 skb
->end
= skb
->tail
+ size
;
205 kmemcheck_annotate_bitfield(skb
, flags1
);
206 kmemcheck_annotate_bitfield(skb
, flags2
);
207 /* make sure we initialize shinfo sequentially */
208 shinfo
= skb_shinfo(skb
);
209 atomic_set(&shinfo
->dataref
, 1);
210 shinfo
->nr_frags
= 0;
211 shinfo
->gso_size
= 0;
212 shinfo
->gso_segs
= 0;
213 shinfo
->gso_type
= 0;
214 shinfo
->ip6_frag_id
= 0;
215 shinfo
->tx_flags
.flags
= 0;
216 skb_frag_list_init(skb
);
217 memset(&shinfo
->hwtstamps
, 0, sizeof(shinfo
->hwtstamps
));
220 struct sk_buff
*child
= skb
+ 1;
221 atomic_t
*fclone_ref
= (atomic_t
*) (child
+ 1);
223 kmemcheck_annotate_bitfield(child
, flags1
);
224 kmemcheck_annotate_bitfield(child
, flags2
);
225 skb
->fclone
= SKB_FCLONE_ORIG
;
226 atomic_set(fclone_ref
, 1);
228 child
->fclone
= SKB_FCLONE_UNAVAILABLE
;
233 kmem_cache_free(cache
, skb
);
237 EXPORT_SYMBOL(__alloc_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
);
265 EXPORT_SYMBOL(__netdev_alloc_skb
);
267 struct page
*__netdev_alloc_page(struct net_device
*dev
, gfp_t gfp_mask
)
269 int node
= dev
->dev
.parent
? dev_to_node(dev
->dev
.parent
) : -1;
272 page
= alloc_pages_node(node
, gfp_mask
, 0);
275 EXPORT_SYMBOL(__netdev_alloc_page
);
277 void skb_add_rx_frag(struct sk_buff
*skb
, int i
, struct page
*page
, int off
,
280 skb_fill_page_desc(skb
, i
, page
, off
, size
);
282 skb
->data_len
+= size
;
283 skb
->truesize
+= size
;
285 EXPORT_SYMBOL(skb_add_rx_frag
);
288 * dev_alloc_skb - allocate an skbuff for receiving
289 * @length: length to allocate
291 * Allocate a new &sk_buff and assign it a usage count of one. The
292 * buffer has unspecified headroom built in. Users should allocate
293 * the headroom they think they need without accounting for the
294 * built in space. The built in space is used for optimisations.
296 * %NULL is returned if there is no free memory. Although this function
297 * allocates memory it can be called from an interrupt.
299 struct sk_buff
*dev_alloc_skb(unsigned int length
)
302 * There is more code here than it seems:
303 * __dev_alloc_skb is an inline
305 return __dev_alloc_skb(length
, GFP_ATOMIC
);
307 EXPORT_SYMBOL(dev_alloc_skb
);
309 static void skb_drop_list(struct sk_buff
**listp
)
311 struct sk_buff
*list
= *listp
;
316 struct sk_buff
*this = list
;
322 static inline void skb_drop_fraglist(struct sk_buff
*skb
)
324 skb_drop_list(&skb_shinfo(skb
)->frag_list
);
327 static void skb_clone_fraglist(struct sk_buff
*skb
)
329 struct sk_buff
*list
;
331 skb_walk_frags(skb
, list
)
335 static void skb_release_data(struct sk_buff
*skb
)
338 !atomic_sub_return(skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1,
339 &skb_shinfo(skb
)->dataref
)) {
340 if (skb_shinfo(skb
)->nr_frags
) {
342 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
343 put_page(skb_shinfo(skb
)->frags
[i
].page
);
346 if (skb_has_frags(skb
))
347 skb_drop_fraglist(skb
);
354 * Free an skbuff by memory without cleaning the state.
356 static void kfree_skbmem(struct sk_buff
*skb
)
358 struct sk_buff
*other
;
359 atomic_t
*fclone_ref
;
361 switch (skb
->fclone
) {
362 case SKB_FCLONE_UNAVAILABLE
:
363 kmem_cache_free(skbuff_head_cache
, skb
);
366 case SKB_FCLONE_ORIG
:
367 fclone_ref
= (atomic_t
*) (skb
+ 2);
368 if (atomic_dec_and_test(fclone_ref
))
369 kmem_cache_free(skbuff_fclone_cache
, skb
);
372 case SKB_FCLONE_CLONE
:
373 fclone_ref
= (atomic_t
*) (skb
+ 1);
376 /* The clone portion is available for
377 * fast-cloning again.
379 skb
->fclone
= SKB_FCLONE_UNAVAILABLE
;
381 if (atomic_dec_and_test(fclone_ref
))
382 kmem_cache_free(skbuff_fclone_cache
, other
);
387 static void skb_release_head_state(struct sk_buff
*skb
)
391 secpath_put(skb
->sp
);
393 if (skb
->destructor
) {
395 skb
->destructor(skb
);
397 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
398 nf_conntrack_put(skb
->nfct
);
399 nf_conntrack_put_reasm(skb
->nfct_reasm
);
401 #ifdef CONFIG_BRIDGE_NETFILTER
402 nf_bridge_put(skb
->nf_bridge
);
404 /* XXX: IS this still necessary? - JHS */
405 #ifdef CONFIG_NET_SCHED
407 #ifdef CONFIG_NET_CLS_ACT
413 /* Free everything but the sk_buff shell. */
414 static void skb_release_all(struct sk_buff
*skb
)
416 skb_release_head_state(skb
);
417 skb_release_data(skb
);
421 * __kfree_skb - private function
424 * Free an sk_buff. Release anything attached to the buffer.
425 * Clean the state. This is an internal helper function. Users should
426 * always call kfree_skb
429 void __kfree_skb(struct sk_buff
*skb
)
431 skb_release_all(skb
);
434 EXPORT_SYMBOL(__kfree_skb
);
437 * kfree_skb - free an sk_buff
438 * @skb: buffer to free
440 * Drop a reference to the buffer and free it if the usage count has
443 void kfree_skb(struct sk_buff
*skb
)
447 if (likely(atomic_read(&skb
->users
) == 1))
449 else if (likely(!atomic_dec_and_test(&skb
->users
)))
451 trace_kfree_skb(skb
, __builtin_return_address(0));
454 EXPORT_SYMBOL(kfree_skb
);
457 * consume_skb - free an skbuff
458 * @skb: buffer to free
460 * Drop a ref to the buffer and free it if the usage count has hit zero
461 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
462 * is being dropped after a failure and notes that
464 void consume_skb(struct sk_buff
*skb
)
468 if (likely(atomic_read(&skb
->users
) == 1))
470 else if (likely(!atomic_dec_and_test(&skb
->users
)))
474 EXPORT_SYMBOL(consume_skb
);
477 * skb_recycle_check - check if skb can be reused for receive
479 * @skb_size: minimum receive buffer size
481 * Checks that the skb passed in is not shared or cloned, and
482 * that it is linear and its head portion at least as large as
483 * skb_size so that it can be recycled as a receive buffer.
484 * If these conditions are met, this function does any necessary
485 * reference count dropping and cleans up the skbuff as if it
486 * just came from __alloc_skb().
488 int skb_recycle_check(struct sk_buff
*skb
, int skb_size
)
490 struct skb_shared_info
*shinfo
;
492 if (skb_is_nonlinear(skb
) || skb
->fclone
!= SKB_FCLONE_UNAVAILABLE
)
495 skb_size
= SKB_DATA_ALIGN(skb_size
+ NET_SKB_PAD
);
496 if (skb_end_pointer(skb
) - skb
->head
< skb_size
)
499 if (skb_shared(skb
) || skb_cloned(skb
))
502 skb_release_head_state(skb
);
503 shinfo
= skb_shinfo(skb
);
504 atomic_set(&shinfo
->dataref
, 1);
505 shinfo
->nr_frags
= 0;
506 shinfo
->gso_size
= 0;
507 shinfo
->gso_segs
= 0;
508 shinfo
->gso_type
= 0;
509 shinfo
->ip6_frag_id
= 0;
510 shinfo
->tx_flags
.flags
= 0;
511 skb_frag_list_init(skb
);
512 memset(&shinfo
->hwtstamps
, 0, sizeof(shinfo
->hwtstamps
));
514 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
515 skb
->data
= skb
->head
+ NET_SKB_PAD
;
516 skb_reset_tail_pointer(skb
);
520 EXPORT_SYMBOL(skb_recycle_check
);
522 static void __copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
524 new->tstamp
= old
->tstamp
;
526 new->transport_header
= old
->transport_header
;
527 new->network_header
= old
->network_header
;
528 new->mac_header
= old
->mac_header
;
529 skb_dst_set(new, dst_clone(skb_dst(old
)));
531 new->sp
= secpath_get(old
->sp
);
533 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
534 new->csum
= old
->csum
;
535 new->local_df
= old
->local_df
;
536 new->pkt_type
= old
->pkt_type
;
537 new->ip_summed
= old
->ip_summed
;
538 skb_copy_queue_mapping(new, old
);
539 new->priority
= old
->priority
;
540 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
541 new->ipvs_property
= old
->ipvs_property
;
543 new->protocol
= old
->protocol
;
544 new->mark
= old
->mark
;
547 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
548 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
549 new->nf_trace
= old
->nf_trace
;
551 #ifdef CONFIG_NET_SCHED
552 new->tc_index
= old
->tc_index
;
553 #ifdef CONFIG_NET_CLS_ACT
554 new->tc_verd
= old
->tc_verd
;
557 new->vlan_tci
= old
->vlan_tci
;
558 #if defined(CONFIG_MAC80211) || defined(CONFIG_MAC80211_MODULE)
559 new->do_not_encrypt
= old
->do_not_encrypt
;
562 skb_copy_secmark(new, old
);
566 * You should not add any new code to this function. Add it to
567 * __copy_skb_header above instead.
569 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
571 #define C(x) n->x = skb->x
573 n
->next
= n
->prev
= NULL
;
575 __copy_skb_header(n
, skb
);
580 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
583 n
->destructor
= NULL
;
589 atomic_set(&n
->users
, 1);
591 atomic_inc(&(skb_shinfo(skb
)->dataref
));
599 * skb_morph - morph one skb into another
600 * @dst: the skb to receive the contents
601 * @src: the skb to supply the contents
603 * This is identical to skb_clone except that the target skb is
604 * supplied by the user.
606 * The target skb is returned upon exit.
608 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
610 skb_release_all(dst
);
611 return __skb_clone(dst
, src
);
613 EXPORT_SYMBOL_GPL(skb_morph
);
616 * skb_clone - duplicate an sk_buff
617 * @skb: buffer to clone
618 * @gfp_mask: allocation priority
620 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
621 * copies share the same packet data but not structure. The new
622 * buffer has a reference count of 1. If the allocation fails the
623 * function returns %NULL otherwise the new buffer is returned.
625 * If this function is called from an interrupt gfp_mask() must be
629 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
634 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
635 n
->fclone
== SKB_FCLONE_UNAVAILABLE
) {
636 atomic_t
*fclone_ref
= (atomic_t
*) (n
+ 1);
637 n
->fclone
= SKB_FCLONE_CLONE
;
638 atomic_inc(fclone_ref
);
640 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
644 kmemcheck_annotate_bitfield(n
, flags1
);
645 kmemcheck_annotate_bitfield(n
, flags2
);
646 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
649 return __skb_clone(n
, skb
);
651 EXPORT_SYMBOL(skb_clone
);
653 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
655 #ifndef NET_SKBUFF_DATA_USES_OFFSET
657 * Shift between the two data areas in bytes
659 unsigned long offset
= new->data
- old
->data
;
662 __copy_skb_header(new, old
);
664 #ifndef NET_SKBUFF_DATA_USES_OFFSET
665 /* {transport,network,mac}_header are relative to skb->head */
666 new->transport_header
+= offset
;
667 new->network_header
+= offset
;
668 new->mac_header
+= offset
;
670 skb_shinfo(new)->gso_size
= skb_shinfo(old
)->gso_size
;
671 skb_shinfo(new)->gso_segs
= skb_shinfo(old
)->gso_segs
;
672 skb_shinfo(new)->gso_type
= skb_shinfo(old
)->gso_type
;
676 * skb_copy - create private copy of an sk_buff
677 * @skb: buffer to copy
678 * @gfp_mask: allocation priority
680 * Make a copy of both an &sk_buff and its data. This is used when the
681 * caller wishes to modify the data and needs a private copy of the
682 * data to alter. Returns %NULL on failure or the pointer to the buffer
683 * on success. The returned buffer has a reference count of 1.
685 * As by-product this function converts non-linear &sk_buff to linear
686 * one, so that &sk_buff becomes completely private and caller is allowed
687 * to modify all the data of returned buffer. This means that this
688 * function is not recommended for use in circumstances when only
689 * header is going to be modified. Use pskb_copy() instead.
692 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
694 int headerlen
= skb
->data
- skb
->head
;
696 * Allocate the copy buffer
699 #ifdef NET_SKBUFF_DATA_USES_OFFSET
700 n
= alloc_skb(skb
->end
+ skb
->data_len
, gfp_mask
);
702 n
= alloc_skb(skb
->end
- skb
->head
+ skb
->data_len
, gfp_mask
);
707 /* Set the data pointer */
708 skb_reserve(n
, headerlen
);
709 /* Set the tail pointer and length */
710 skb_put(n
, skb
->len
);
712 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
715 copy_skb_header(n
, skb
);
718 EXPORT_SYMBOL(skb_copy
);
721 * pskb_copy - create copy of an sk_buff with private head.
722 * @skb: buffer to copy
723 * @gfp_mask: allocation priority
725 * Make a copy of both an &sk_buff and part of its data, located
726 * in header. Fragmented data remain shared. This is used when
727 * the caller wishes to modify only header of &sk_buff and needs
728 * private copy of the header to alter. Returns %NULL on failure
729 * or the pointer to the buffer on success.
730 * The returned buffer has a reference count of 1.
733 struct sk_buff
*pskb_copy(struct sk_buff
*skb
, gfp_t gfp_mask
)
736 * Allocate the copy buffer
739 #ifdef NET_SKBUFF_DATA_USES_OFFSET
740 n
= alloc_skb(skb
->end
, gfp_mask
);
742 n
= alloc_skb(skb
->end
- skb
->head
, gfp_mask
);
747 /* Set the data pointer */
748 skb_reserve(n
, skb
->data
- skb
->head
);
749 /* Set the tail pointer and length */
750 skb_put(n
, skb_headlen(skb
));
752 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
754 n
->truesize
+= skb
->data_len
;
755 n
->data_len
= skb
->data_len
;
758 if (skb_shinfo(skb
)->nr_frags
) {
761 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
762 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
763 get_page(skb_shinfo(n
)->frags
[i
].page
);
765 skb_shinfo(n
)->nr_frags
= i
;
768 if (skb_has_frags(skb
)) {
769 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
770 skb_clone_fraglist(n
);
773 copy_skb_header(n
, skb
);
777 EXPORT_SYMBOL(pskb_copy
);
780 * pskb_expand_head - reallocate header of &sk_buff
781 * @skb: buffer to reallocate
782 * @nhead: room to add at head
783 * @ntail: room to add at tail
784 * @gfp_mask: allocation priority
786 * Expands (or creates identical copy, if &nhead and &ntail are zero)
787 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
788 * reference count of 1. Returns zero in the case of success or error,
789 * if expansion failed. In the last case, &sk_buff is not changed.
791 * All the pointers pointing into skb header may change and must be
792 * reloaded after call to this function.
795 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
800 #ifdef NET_SKBUFF_DATA_USES_OFFSET
801 int size
= nhead
+ skb
->end
+ ntail
;
803 int size
= nhead
+ (skb
->end
- skb
->head
) + ntail
;
812 size
= SKB_DATA_ALIGN(size
);
814 data
= kmalloc(size
+ sizeof(struct skb_shared_info
), gfp_mask
);
818 /* Copy only real data... and, alas, header. This should be
819 * optimized for the cases when header is void. */
820 #ifdef NET_SKBUFF_DATA_USES_OFFSET
821 memcpy(data
+ nhead
, skb
->head
, skb
->tail
);
823 memcpy(data
+ nhead
, skb
->head
, skb
->tail
- skb
->head
);
825 memcpy(data
+ size
, skb_end_pointer(skb
),
826 sizeof(struct skb_shared_info
));
828 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
829 get_page(skb_shinfo(skb
)->frags
[i
].page
);
831 if (skb_has_frags(skb
))
832 skb_clone_fraglist(skb
);
834 skb_release_data(skb
);
836 off
= (data
+ nhead
) - skb
->head
;
840 #ifdef NET_SKBUFF_DATA_USES_OFFSET
844 skb
->end
= skb
->head
+ size
;
846 /* {transport,network,mac}_header and tail are relative to skb->head */
848 skb
->transport_header
+= off
;
849 skb
->network_header
+= off
;
850 skb
->mac_header
+= off
;
851 skb
->csum_start
+= nhead
;
855 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
861 EXPORT_SYMBOL(pskb_expand_head
);
863 /* Make private copy of skb with writable head and some headroom */
865 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
867 struct sk_buff
*skb2
;
868 int delta
= headroom
- skb_headroom(skb
);
871 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
873 skb2
= skb_clone(skb
, GFP_ATOMIC
);
874 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
882 EXPORT_SYMBOL(skb_realloc_headroom
);
885 * skb_copy_expand - copy and expand sk_buff
886 * @skb: buffer to copy
887 * @newheadroom: new free bytes at head
888 * @newtailroom: new free bytes at tail
889 * @gfp_mask: allocation priority
891 * Make a copy of both an &sk_buff and its data and while doing so
892 * allocate additional space.
894 * This is used when the caller wishes to modify the data and needs a
895 * private copy of the data to alter as well as more space for new fields.
896 * Returns %NULL on failure or the pointer to the buffer
897 * on success. The returned buffer has a reference count of 1.
899 * You must pass %GFP_ATOMIC as the allocation priority if this function
900 * is called from an interrupt.
902 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
903 int newheadroom
, int newtailroom
,
907 * Allocate the copy buffer
909 struct sk_buff
*n
= alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
911 int oldheadroom
= skb_headroom(skb
);
912 int head_copy_len
, head_copy_off
;
918 skb_reserve(n
, newheadroom
);
920 /* Set the tail pointer and length */
921 skb_put(n
, skb
->len
);
923 head_copy_len
= oldheadroom
;
925 if (newheadroom
<= head_copy_len
)
926 head_copy_len
= newheadroom
;
928 head_copy_off
= newheadroom
- head_copy_len
;
930 /* Copy the linear header and data. */
931 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
932 skb
->len
+ head_copy_len
))
935 copy_skb_header(n
, skb
);
937 off
= newheadroom
- oldheadroom
;
938 n
->csum_start
+= off
;
939 #ifdef NET_SKBUFF_DATA_USES_OFFSET
940 n
->transport_header
+= off
;
941 n
->network_header
+= off
;
942 n
->mac_header
+= off
;
947 EXPORT_SYMBOL(skb_copy_expand
);
950 * skb_pad - zero pad the tail of an skb
951 * @skb: buffer to pad
954 * Ensure that a buffer is followed by a padding area that is zero
955 * filled. Used by network drivers which may DMA or transfer data
956 * beyond the buffer end onto the wire.
958 * May return error in out of memory cases. The skb is freed on error.
961 int skb_pad(struct sk_buff
*skb
, int pad
)
966 /* If the skbuff is non linear tailroom is always zero.. */
967 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
968 memset(skb
->data
+skb
->len
, 0, pad
);
972 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
973 if (likely(skb_cloned(skb
) || ntail
> 0)) {
974 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
979 /* FIXME: The use of this function with non-linear skb's really needs
982 err
= skb_linearize(skb
);
986 memset(skb
->data
+ skb
->len
, 0, pad
);
993 EXPORT_SYMBOL(skb_pad
);
996 * skb_put - add data to a buffer
997 * @skb: buffer to use
998 * @len: amount of data to add
1000 * This function extends the used data area of the buffer. If this would
1001 * exceed the total buffer size the kernel will panic. A pointer to the
1002 * first byte of the extra data is returned.
1004 unsigned char *skb_put(struct sk_buff
*skb
, unsigned int len
)
1006 unsigned char *tmp
= skb_tail_pointer(skb
);
1007 SKB_LINEAR_ASSERT(skb
);
1010 if (unlikely(skb
->tail
> skb
->end
))
1011 skb_over_panic(skb
, len
, __builtin_return_address(0));
1014 EXPORT_SYMBOL(skb_put
);
1017 * skb_push - add data to the start of a buffer
1018 * @skb: buffer to use
1019 * @len: amount of data to add
1021 * This function extends the used data area of the buffer at the buffer
1022 * start. If this would exceed the total buffer headroom the kernel will
1023 * panic. A pointer to the first byte of the extra data is returned.
1025 unsigned char *skb_push(struct sk_buff
*skb
, unsigned int len
)
1029 if (unlikely(skb
->data
<skb
->head
))
1030 skb_under_panic(skb
, len
, __builtin_return_address(0));
1033 EXPORT_SYMBOL(skb_push
);
1036 * skb_pull - remove data from the start of a buffer
1037 * @skb: buffer to use
1038 * @len: amount of data to remove
1040 * This function removes data from the start of a buffer, returning
1041 * the memory to the headroom. A pointer to the next data in the buffer
1042 * is returned. Once the data has been pulled future pushes will overwrite
1045 unsigned char *skb_pull(struct sk_buff
*skb
, unsigned int len
)
1047 return unlikely(len
> skb
->len
) ? NULL
: __skb_pull(skb
, len
);
1049 EXPORT_SYMBOL(skb_pull
);
1052 * skb_trim - remove end from a buffer
1053 * @skb: buffer to alter
1056 * Cut the length of a buffer down by removing data from the tail. If
1057 * the buffer is already under the length specified it is not modified.
1058 * The skb must be linear.
1060 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1063 __skb_trim(skb
, len
);
1065 EXPORT_SYMBOL(skb_trim
);
1067 /* Trims skb to length len. It can change skb pointers.
1070 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1072 struct sk_buff
**fragp
;
1073 struct sk_buff
*frag
;
1074 int offset
= skb_headlen(skb
);
1075 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1079 if (skb_cloned(skb
) &&
1080 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1087 for (; i
< nfrags
; i
++) {
1088 int end
= offset
+ skb_shinfo(skb
)->frags
[i
].size
;
1095 skb_shinfo(skb
)->frags
[i
++].size
= len
- offset
;
1098 skb_shinfo(skb
)->nr_frags
= i
;
1100 for (; i
< nfrags
; i
++)
1101 put_page(skb_shinfo(skb
)->frags
[i
].page
);
1103 if (skb_has_frags(skb
))
1104 skb_drop_fraglist(skb
);
1108 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1109 fragp
= &frag
->next
) {
1110 int end
= offset
+ frag
->len
;
1112 if (skb_shared(frag
)) {
1113 struct sk_buff
*nfrag
;
1115 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1116 if (unlikely(!nfrag
))
1119 nfrag
->next
= frag
->next
;
1131 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1135 skb_drop_list(&frag
->next
);
1140 if (len
> skb_headlen(skb
)) {
1141 skb
->data_len
-= skb
->len
- len
;
1146 skb_set_tail_pointer(skb
, len
);
1151 EXPORT_SYMBOL(___pskb_trim
);
1154 * __pskb_pull_tail - advance tail of skb header
1155 * @skb: buffer to reallocate
1156 * @delta: number of bytes to advance tail
1158 * The function makes a sense only on a fragmented &sk_buff,
1159 * it expands header moving its tail forward and copying necessary
1160 * data from fragmented part.
1162 * &sk_buff MUST have reference count of 1.
1164 * Returns %NULL (and &sk_buff does not change) if pull failed
1165 * or value of new tail of skb in the case of success.
1167 * All the pointers pointing into skb header may change and must be
1168 * reloaded after call to this function.
1171 /* Moves tail of skb head forward, copying data from fragmented part,
1172 * when it is necessary.
1173 * 1. It may fail due to malloc failure.
1174 * 2. It may change skb pointers.
1176 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1178 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1180 /* If skb has not enough free space at tail, get new one
1181 * plus 128 bytes for future expansions. If we have enough
1182 * room at tail, reallocate without expansion only if skb is cloned.
1184 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1186 if (eat
> 0 || skb_cloned(skb
)) {
1187 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1192 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
1195 /* Optimization: no fragments, no reasons to preestimate
1196 * size of pulled pages. Superb.
1198 if (!skb_has_frags(skb
))
1201 /* Estimate size of pulled pages. */
1203 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1204 if (skb_shinfo(skb
)->frags
[i
].size
>= eat
)
1206 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1209 /* If we need update frag list, we are in troubles.
1210 * Certainly, it possible to add an offset to skb data,
1211 * but taking into account that pulling is expected to
1212 * be very rare operation, it is worth to fight against
1213 * further bloating skb head and crucify ourselves here instead.
1214 * Pure masohism, indeed. 8)8)
1217 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1218 struct sk_buff
*clone
= NULL
;
1219 struct sk_buff
*insp
= NULL
;
1224 if (list
->len
<= eat
) {
1225 /* Eaten as whole. */
1230 /* Eaten partially. */
1232 if (skb_shared(list
)) {
1233 /* Sucks! We need to fork list. :-( */
1234 clone
= skb_clone(list
, GFP_ATOMIC
);
1240 /* This may be pulled without
1244 if (!pskb_pull(list
, eat
)) {
1252 /* Free pulled out fragments. */
1253 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1254 skb_shinfo(skb
)->frag_list
= list
->next
;
1257 /* And insert new clone at head. */
1260 skb_shinfo(skb
)->frag_list
= clone
;
1263 /* Success! Now we may commit changes to skb data. */
1268 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1269 if (skb_shinfo(skb
)->frags
[i
].size
<= eat
) {
1270 put_page(skb_shinfo(skb
)->frags
[i
].page
);
1271 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1273 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1275 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1276 skb_shinfo(skb
)->frags
[k
].size
-= eat
;
1282 skb_shinfo(skb
)->nr_frags
= k
;
1285 skb
->data_len
-= delta
;
1287 return skb_tail_pointer(skb
);
1289 EXPORT_SYMBOL(__pskb_pull_tail
);
1291 /* Copy some data bits from skb to kernel buffer. */
1293 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1295 int start
= skb_headlen(skb
);
1296 struct sk_buff
*frag_iter
;
1299 if (offset
> (int)skb
->len
- len
)
1303 if ((copy
= start
- offset
) > 0) {
1306 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
1307 if ((len
-= copy
) == 0)
1313 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1316 WARN_ON(start
> offset
+ len
);
1318 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1319 if ((copy
= end
- offset
) > 0) {
1325 vaddr
= kmap_skb_frag(&skb_shinfo(skb
)->frags
[i
]);
1327 vaddr
+ skb_shinfo(skb
)->frags
[i
].page_offset
+
1328 offset
- start
, copy
);
1329 kunmap_skb_frag(vaddr
);
1331 if ((len
-= copy
) == 0)
1339 skb_walk_frags(skb
, frag_iter
) {
1342 WARN_ON(start
> offset
+ len
);
1344 end
= start
+ frag_iter
->len
;
1345 if ((copy
= end
- offset
) > 0) {
1348 if (skb_copy_bits(frag_iter
, offset
- start
, to
, copy
))
1350 if ((len
-= copy
) == 0)
1363 EXPORT_SYMBOL(skb_copy_bits
);
1366 * Callback from splice_to_pipe(), if we need to release some pages
1367 * at the end of the spd in case we error'ed out in filling the pipe.
1369 static void sock_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
1371 put_page(spd
->pages
[i
]);
1374 static inline struct page
*linear_to_page(struct page
*page
, unsigned int *len
,
1375 unsigned int *offset
,
1376 struct sk_buff
*skb
, struct sock
*sk
)
1378 struct page
*p
= sk
->sk_sndmsg_page
;
1383 p
= sk
->sk_sndmsg_page
= alloc_pages(sk
->sk_allocation
, 0);
1387 off
= sk
->sk_sndmsg_off
= 0;
1388 /* hold one ref to this page until it's full */
1392 off
= sk
->sk_sndmsg_off
;
1393 mlen
= PAGE_SIZE
- off
;
1394 if (mlen
< 64 && mlen
< *len
) {
1399 *len
= min_t(unsigned int, *len
, mlen
);
1402 memcpy(page_address(p
) + off
, page_address(page
) + *offset
, *len
);
1403 sk
->sk_sndmsg_off
+= *len
;
1411 * Fill page/offset/length into spd, if it can hold more pages.
1413 static inline int spd_fill_page(struct splice_pipe_desc
*spd
, struct page
*page
,
1414 unsigned int *len
, unsigned int offset
,
1415 struct sk_buff
*skb
, int linear
,
1418 if (unlikely(spd
->nr_pages
== PIPE_BUFFERS
))
1422 page
= linear_to_page(page
, len
, &offset
, skb
, sk
);
1428 spd
->pages
[spd
->nr_pages
] = page
;
1429 spd
->partial
[spd
->nr_pages
].len
= *len
;
1430 spd
->partial
[spd
->nr_pages
].offset
= offset
;
1436 static inline void __segment_seek(struct page
**page
, unsigned int *poff
,
1437 unsigned int *plen
, unsigned int off
)
1442 n
= *poff
/ PAGE_SIZE
;
1444 *page
= nth_page(*page
, n
);
1446 *poff
= *poff
% PAGE_SIZE
;
1450 static inline int __splice_segment(struct page
*page
, unsigned int poff
,
1451 unsigned int plen
, unsigned int *off
,
1452 unsigned int *len
, struct sk_buff
*skb
,
1453 struct splice_pipe_desc
*spd
, int linear
,
1459 /* skip this segment if already processed */
1465 /* ignore any bits we already processed */
1467 __segment_seek(&page
, &poff
, &plen
, *off
);
1472 unsigned int flen
= min(*len
, plen
);
1474 /* the linear region may spread across several pages */
1475 flen
= min_t(unsigned int, flen
, PAGE_SIZE
- poff
);
1477 if (spd_fill_page(spd
, page
, &flen
, poff
, skb
, linear
, sk
))
1480 __segment_seek(&page
, &poff
, &plen
, flen
);
1483 } while (*len
&& plen
);
1489 * Map linear and fragment data from the skb to spd. It reports failure if the
1490 * pipe is full or if we already spliced the requested length.
1492 static int __skb_splice_bits(struct sk_buff
*skb
, unsigned int *offset
,
1493 unsigned int *len
, struct splice_pipe_desc
*spd
,
1499 * map the linear part
1501 if (__splice_segment(virt_to_page(skb
->data
),
1502 (unsigned long) skb
->data
& (PAGE_SIZE
- 1),
1504 offset
, len
, skb
, spd
, 1, sk
))
1508 * then map the fragments
1510 for (seg
= 0; seg
< skb_shinfo(skb
)->nr_frags
; seg
++) {
1511 const skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[seg
];
1513 if (__splice_segment(f
->page
, f
->page_offset
, f
->size
,
1514 offset
, len
, skb
, spd
, 0, sk
))
1522 * Map data from the skb to a pipe. Should handle both the linear part,
1523 * the fragments, and the frag list. It does NOT handle frag lists within
1524 * the frag list, if such a thing exists. We'd probably need to recurse to
1525 * handle that cleanly.
1527 int skb_splice_bits(struct sk_buff
*skb
, unsigned int offset
,
1528 struct pipe_inode_info
*pipe
, unsigned int tlen
,
1531 struct partial_page partial
[PIPE_BUFFERS
];
1532 struct page
*pages
[PIPE_BUFFERS
];
1533 struct splice_pipe_desc spd
= {
1537 .ops
= &sock_pipe_buf_ops
,
1538 .spd_release
= sock_spd_release
,
1540 struct sk_buff
*frag_iter
;
1541 struct sock
*sk
= skb
->sk
;
1544 * __skb_splice_bits() only fails if the output has no room left,
1545 * so no point in going over the frag_list for the error case.
1547 if (__skb_splice_bits(skb
, &offset
, &tlen
, &spd
, sk
))
1553 * now see if we have a frag_list to map
1555 skb_walk_frags(skb
, frag_iter
) {
1558 if (__skb_splice_bits(frag_iter
, &offset
, &tlen
, &spd
, sk
))
1567 * Drop the socket lock, otherwise we have reverse
1568 * locking dependencies between sk_lock and i_mutex
1569 * here as compared to sendfile(). We enter here
1570 * with the socket lock held, and splice_to_pipe() will
1571 * grab the pipe inode lock. For sendfile() emulation,
1572 * we call into ->sendpage() with the i_mutex lock held
1573 * and networking will grab the socket lock.
1576 ret
= splice_to_pipe(pipe
, &spd
);
1585 * skb_store_bits - store bits from kernel buffer to skb
1586 * @skb: destination buffer
1587 * @offset: offset in destination
1588 * @from: source buffer
1589 * @len: number of bytes to copy
1591 * Copy the specified number of bytes from the source buffer to the
1592 * destination skb. This function handles all the messy bits of
1593 * traversing fragment lists and such.
1596 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
1598 int start
= skb_headlen(skb
);
1599 struct sk_buff
*frag_iter
;
1602 if (offset
> (int)skb
->len
- len
)
1605 if ((copy
= start
- offset
) > 0) {
1608 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
1609 if ((len
-= copy
) == 0)
1615 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1616 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1619 WARN_ON(start
> offset
+ len
);
1621 end
= start
+ frag
->size
;
1622 if ((copy
= end
- offset
) > 0) {
1628 vaddr
= kmap_skb_frag(frag
);
1629 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1631 kunmap_skb_frag(vaddr
);
1633 if ((len
-= copy
) == 0)
1641 skb_walk_frags(skb
, frag_iter
) {
1644 WARN_ON(start
> offset
+ len
);
1646 end
= start
+ frag_iter
->len
;
1647 if ((copy
= end
- offset
) > 0) {
1650 if (skb_store_bits(frag_iter
, offset
- start
,
1653 if ((len
-= copy
) == 0)
1666 EXPORT_SYMBOL(skb_store_bits
);
1668 /* Checksum skb data. */
1670 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
1671 int len
, __wsum csum
)
1673 int start
= skb_headlen(skb
);
1674 int i
, copy
= start
- offset
;
1675 struct sk_buff
*frag_iter
;
1678 /* Checksum header. */
1682 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1683 if ((len
-= copy
) == 0)
1689 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1692 WARN_ON(start
> offset
+ len
);
1694 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1695 if ((copy
= end
- offset
) > 0) {
1698 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1702 vaddr
= kmap_skb_frag(frag
);
1703 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
1704 offset
- start
, copy
, 0);
1705 kunmap_skb_frag(vaddr
);
1706 csum
= csum_block_add(csum
, csum2
, pos
);
1715 skb_walk_frags(skb
, frag_iter
) {
1718 WARN_ON(start
> offset
+ len
);
1720 end
= start
+ frag_iter
->len
;
1721 if ((copy
= end
- offset
) > 0) {
1725 csum2
= skb_checksum(frag_iter
, offset
- start
,
1727 csum
= csum_block_add(csum
, csum2
, pos
);
1728 if ((len
-= copy
) == 0)
1739 EXPORT_SYMBOL(skb_checksum
);
1741 /* Both of above in one bottle. */
1743 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
1744 u8
*to
, int len
, __wsum csum
)
1746 int start
= skb_headlen(skb
);
1747 int i
, copy
= start
- offset
;
1748 struct sk_buff
*frag_iter
;
1755 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
1757 if ((len
-= copy
) == 0)
1764 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1767 WARN_ON(start
> offset
+ len
);
1769 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1770 if ((copy
= end
- offset
) > 0) {
1773 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1777 vaddr
= kmap_skb_frag(frag
);
1778 csum2
= csum_partial_copy_nocheck(vaddr
+
1782 kunmap_skb_frag(vaddr
);
1783 csum
= csum_block_add(csum
, csum2
, pos
);
1793 skb_walk_frags(skb
, frag_iter
) {
1797 WARN_ON(start
> offset
+ len
);
1799 end
= start
+ frag_iter
->len
;
1800 if ((copy
= end
- offset
) > 0) {
1803 csum2
= skb_copy_and_csum_bits(frag_iter
,
1806 csum
= csum_block_add(csum
, csum2
, pos
);
1807 if ((len
-= copy
) == 0)
1818 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
1820 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
1825 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1826 csstart
= skb
->csum_start
- skb_headroom(skb
);
1828 csstart
= skb_headlen(skb
);
1830 BUG_ON(csstart
> skb_headlen(skb
));
1832 skb_copy_from_linear_data(skb
, to
, csstart
);
1835 if (csstart
!= skb
->len
)
1836 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
1837 skb
->len
- csstart
, 0);
1839 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1840 long csstuff
= csstart
+ skb
->csum_offset
;
1842 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
1845 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
1848 * skb_dequeue - remove from the head of the queue
1849 * @list: list to dequeue from
1851 * Remove the head of the list. The list lock is taken so the function
1852 * may be used safely with other locking list functions. The head item is
1853 * returned or %NULL if the list is empty.
1856 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
1858 unsigned long flags
;
1859 struct sk_buff
*result
;
1861 spin_lock_irqsave(&list
->lock
, flags
);
1862 result
= __skb_dequeue(list
);
1863 spin_unlock_irqrestore(&list
->lock
, flags
);
1866 EXPORT_SYMBOL(skb_dequeue
);
1869 * skb_dequeue_tail - remove from the tail of the queue
1870 * @list: list to dequeue from
1872 * Remove the tail of the list. The list lock is taken so the function
1873 * may be used safely with other locking list functions. The tail item is
1874 * returned or %NULL if the list is empty.
1876 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
1878 unsigned long flags
;
1879 struct sk_buff
*result
;
1881 spin_lock_irqsave(&list
->lock
, flags
);
1882 result
= __skb_dequeue_tail(list
);
1883 spin_unlock_irqrestore(&list
->lock
, flags
);
1886 EXPORT_SYMBOL(skb_dequeue_tail
);
1889 * skb_queue_purge - empty a list
1890 * @list: list to empty
1892 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1893 * the list and one reference dropped. This function takes the list
1894 * lock and is atomic with respect to other list locking functions.
1896 void skb_queue_purge(struct sk_buff_head
*list
)
1898 struct sk_buff
*skb
;
1899 while ((skb
= skb_dequeue(list
)) != NULL
)
1902 EXPORT_SYMBOL(skb_queue_purge
);
1905 * skb_queue_head - queue a buffer at the list head
1906 * @list: list to use
1907 * @newsk: buffer to queue
1909 * Queue a buffer at the start of the list. This function takes the
1910 * list lock and can be used safely with other locking &sk_buff functions
1913 * A buffer cannot be placed on two lists at the same time.
1915 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1917 unsigned long flags
;
1919 spin_lock_irqsave(&list
->lock
, flags
);
1920 __skb_queue_head(list
, newsk
);
1921 spin_unlock_irqrestore(&list
->lock
, flags
);
1923 EXPORT_SYMBOL(skb_queue_head
);
1926 * skb_queue_tail - queue a buffer at the list tail
1927 * @list: list to use
1928 * @newsk: buffer to queue
1930 * Queue a buffer at the tail of the list. This function takes the
1931 * list lock and can be used safely with other locking &sk_buff functions
1934 * A buffer cannot be placed on two lists at the same time.
1936 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1938 unsigned long flags
;
1940 spin_lock_irqsave(&list
->lock
, flags
);
1941 __skb_queue_tail(list
, newsk
);
1942 spin_unlock_irqrestore(&list
->lock
, flags
);
1944 EXPORT_SYMBOL(skb_queue_tail
);
1947 * skb_unlink - remove a buffer from a list
1948 * @skb: buffer to remove
1949 * @list: list to use
1951 * Remove a packet from a list. The list locks are taken and this
1952 * function is atomic with respect to other list locked calls
1954 * You must know what list the SKB is on.
1956 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
1958 unsigned long flags
;
1960 spin_lock_irqsave(&list
->lock
, flags
);
1961 __skb_unlink(skb
, list
);
1962 spin_unlock_irqrestore(&list
->lock
, flags
);
1964 EXPORT_SYMBOL(skb_unlink
);
1967 * skb_append - append a buffer
1968 * @old: buffer to insert after
1969 * @newsk: buffer to insert
1970 * @list: list to use
1972 * Place a packet after a given packet in a list. The list locks are taken
1973 * and this function is atomic with respect to other list locked calls.
1974 * A buffer cannot be placed on two lists at the same time.
1976 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
1978 unsigned long flags
;
1980 spin_lock_irqsave(&list
->lock
, flags
);
1981 __skb_queue_after(list
, old
, newsk
);
1982 spin_unlock_irqrestore(&list
->lock
, flags
);
1984 EXPORT_SYMBOL(skb_append
);
1987 * skb_insert - insert a buffer
1988 * @old: buffer to insert before
1989 * @newsk: buffer to insert
1990 * @list: list to use
1992 * Place a packet before a given packet in a list. The list locks are
1993 * taken and this function is atomic with respect to other list locked
1996 * A buffer cannot be placed on two lists at the same time.
1998 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2000 unsigned long flags
;
2002 spin_lock_irqsave(&list
->lock
, flags
);
2003 __skb_insert(newsk
, old
->prev
, old
, list
);
2004 spin_unlock_irqrestore(&list
->lock
, flags
);
2006 EXPORT_SYMBOL(skb_insert
);
2008 static inline void skb_split_inside_header(struct sk_buff
*skb
,
2009 struct sk_buff
* skb1
,
2010 const u32 len
, const int pos
)
2014 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
2016 /* And move data appendix as is. */
2017 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
2018 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
2020 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
2021 skb_shinfo(skb
)->nr_frags
= 0;
2022 skb1
->data_len
= skb
->data_len
;
2023 skb1
->len
+= skb1
->data_len
;
2026 skb_set_tail_pointer(skb
, len
);
2029 static inline void skb_split_no_header(struct sk_buff
*skb
,
2030 struct sk_buff
* skb1
,
2031 const u32 len
, int pos
)
2034 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
2036 skb_shinfo(skb
)->nr_frags
= 0;
2037 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
2039 skb
->data_len
= len
- pos
;
2041 for (i
= 0; i
< nfrags
; i
++) {
2042 int size
= skb_shinfo(skb
)->frags
[i
].size
;
2044 if (pos
+ size
> len
) {
2045 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
2049 * We have two variants in this case:
2050 * 1. Move all the frag to the second
2051 * part, if it is possible. F.e.
2052 * this approach is mandatory for TUX,
2053 * where splitting is expensive.
2054 * 2. Split is accurately. We make this.
2056 get_page(skb_shinfo(skb
)->frags
[i
].page
);
2057 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
2058 skb_shinfo(skb1
)->frags
[0].size
-= len
- pos
;
2059 skb_shinfo(skb
)->frags
[i
].size
= len
- pos
;
2060 skb_shinfo(skb
)->nr_frags
++;
2064 skb_shinfo(skb
)->nr_frags
++;
2067 skb_shinfo(skb1
)->nr_frags
= k
;
2071 * skb_split - Split fragmented skb to two parts at length len.
2072 * @skb: the buffer to split
2073 * @skb1: the buffer to receive the second part
2074 * @len: new length for skb
2076 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
2078 int pos
= skb_headlen(skb
);
2080 if (len
< pos
) /* Split line is inside header. */
2081 skb_split_inside_header(skb
, skb1
, len
, pos
);
2082 else /* Second chunk has no header, nothing to copy. */
2083 skb_split_no_header(skb
, skb1
, len
, pos
);
2085 EXPORT_SYMBOL(skb_split
);
2087 /* Shifting from/to a cloned skb is a no-go.
2089 * Caller cannot keep skb_shinfo related pointers past calling here!
2091 static int skb_prepare_for_shift(struct sk_buff
*skb
)
2093 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2097 * skb_shift - Shifts paged data partially from skb to another
2098 * @tgt: buffer into which tail data gets added
2099 * @skb: buffer from which the paged data comes from
2100 * @shiftlen: shift up to this many bytes
2102 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2103 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2104 * It's up to caller to free skb if everything was shifted.
2106 * If @tgt runs out of frags, the whole operation is aborted.
2108 * Skb cannot include anything else but paged data while tgt is allowed
2109 * to have non-paged data as well.
2111 * TODO: full sized shift could be optimized but that would need
2112 * specialized skb free'er to handle frags without up-to-date nr_frags.
2114 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
2116 int from
, to
, merge
, todo
;
2117 struct skb_frag_struct
*fragfrom
, *fragto
;
2119 BUG_ON(shiftlen
> skb
->len
);
2120 BUG_ON(skb_headlen(skb
)); /* Would corrupt stream */
2124 to
= skb_shinfo(tgt
)->nr_frags
;
2125 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2127 /* Actual merge is delayed until the point when we know we can
2128 * commit all, so that we don't have to undo partial changes
2131 !skb_can_coalesce(tgt
, to
, fragfrom
->page
, fragfrom
->page_offset
)) {
2136 todo
-= fragfrom
->size
;
2138 if (skb_prepare_for_shift(skb
) ||
2139 skb_prepare_for_shift(tgt
))
2142 /* All previous frag pointers might be stale! */
2143 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2144 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2146 fragto
->size
+= shiftlen
;
2147 fragfrom
->size
-= shiftlen
;
2148 fragfrom
->page_offset
+= shiftlen
;
2156 /* Skip full, not-fitting skb to avoid expensive operations */
2157 if ((shiftlen
== skb
->len
) &&
2158 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
2161 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
2164 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
2165 if (to
== MAX_SKB_FRAGS
)
2168 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2169 fragto
= &skb_shinfo(tgt
)->frags
[to
];
2171 if (todo
>= fragfrom
->size
) {
2172 *fragto
= *fragfrom
;
2173 todo
-= fragfrom
->size
;
2178 get_page(fragfrom
->page
);
2179 fragto
->page
= fragfrom
->page
;
2180 fragto
->page_offset
= fragfrom
->page_offset
;
2181 fragto
->size
= todo
;
2183 fragfrom
->page_offset
+= todo
;
2184 fragfrom
->size
-= todo
;
2192 /* Ready to "commit" this state change to tgt */
2193 skb_shinfo(tgt
)->nr_frags
= to
;
2196 fragfrom
= &skb_shinfo(skb
)->frags
[0];
2197 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2199 fragto
->size
+= fragfrom
->size
;
2200 put_page(fragfrom
->page
);
2203 /* Reposition in the original skb */
2205 while (from
< skb_shinfo(skb
)->nr_frags
)
2206 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
2207 skb_shinfo(skb
)->nr_frags
= to
;
2209 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
2212 /* Most likely the tgt won't ever need its checksum anymore, skb on
2213 * the other hand might need it if it needs to be resent
2215 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
2216 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2218 /* Yak, is it really working this way? Some helper please? */
2219 skb
->len
-= shiftlen
;
2220 skb
->data_len
-= shiftlen
;
2221 skb
->truesize
-= shiftlen
;
2222 tgt
->len
+= shiftlen
;
2223 tgt
->data_len
+= shiftlen
;
2224 tgt
->truesize
+= shiftlen
;
2230 * skb_prepare_seq_read - Prepare a sequential read of skb data
2231 * @skb: the buffer to read
2232 * @from: lower offset of data to be read
2233 * @to: upper offset of data to be read
2234 * @st: state variable
2236 * Initializes the specified state variable. Must be called before
2237 * invoking skb_seq_read() for the first time.
2239 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
2240 unsigned int to
, struct skb_seq_state
*st
)
2242 st
->lower_offset
= from
;
2243 st
->upper_offset
= to
;
2244 st
->root_skb
= st
->cur_skb
= skb
;
2245 st
->frag_idx
= st
->stepped_offset
= 0;
2246 st
->frag_data
= NULL
;
2248 EXPORT_SYMBOL(skb_prepare_seq_read
);
2251 * skb_seq_read - Sequentially read skb data
2252 * @consumed: number of bytes consumed by the caller so far
2253 * @data: destination pointer for data to be returned
2254 * @st: state variable
2256 * Reads a block of skb data at &consumed relative to the
2257 * lower offset specified to skb_prepare_seq_read(). Assigns
2258 * the head of the data block to &data and returns the length
2259 * of the block or 0 if the end of the skb data or the upper
2260 * offset has been reached.
2262 * The caller is not required to consume all of the data
2263 * returned, i.e. &consumed is typically set to the number
2264 * of bytes already consumed and the next call to
2265 * skb_seq_read() will return the remaining part of the block.
2267 * Note 1: The size of each block of data returned can be arbitary,
2268 * this limitation is the cost for zerocopy seqeuental
2269 * reads of potentially non linear data.
2271 * Note 2: Fragment lists within fragments are not implemented
2272 * at the moment, state->root_skb could be replaced with
2273 * a stack for this purpose.
2275 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
2276 struct skb_seq_state
*st
)
2278 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
2281 if (unlikely(abs_offset
>= st
->upper_offset
))
2285 block_limit
= skb_headlen(st
->cur_skb
) + st
->stepped_offset
;
2287 if (abs_offset
< block_limit
&& !st
->frag_data
) {
2288 *data
= st
->cur_skb
->data
+ (abs_offset
- st
->stepped_offset
);
2289 return block_limit
- abs_offset
;
2292 if (st
->frag_idx
== 0 && !st
->frag_data
)
2293 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
2295 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
2296 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
2297 block_limit
= frag
->size
+ st
->stepped_offset
;
2299 if (abs_offset
< block_limit
) {
2301 st
->frag_data
= kmap_skb_frag(frag
);
2303 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
2304 (abs_offset
- st
->stepped_offset
);
2306 return block_limit
- abs_offset
;
2309 if (st
->frag_data
) {
2310 kunmap_skb_frag(st
->frag_data
);
2311 st
->frag_data
= NULL
;
2315 st
->stepped_offset
+= frag
->size
;
2318 if (st
->frag_data
) {
2319 kunmap_skb_frag(st
->frag_data
);
2320 st
->frag_data
= NULL
;
2323 if (st
->root_skb
== st
->cur_skb
&& skb_has_frags(st
->root_skb
)) {
2324 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
2327 } else if (st
->cur_skb
->next
) {
2328 st
->cur_skb
= st
->cur_skb
->next
;
2335 EXPORT_SYMBOL(skb_seq_read
);
2338 * skb_abort_seq_read - Abort a sequential read of skb data
2339 * @st: state variable
2341 * Must be called if skb_seq_read() was not called until it
2344 void skb_abort_seq_read(struct skb_seq_state
*st
)
2347 kunmap_skb_frag(st
->frag_data
);
2349 EXPORT_SYMBOL(skb_abort_seq_read
);
2351 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2353 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
2354 struct ts_config
*conf
,
2355 struct ts_state
*state
)
2357 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
2360 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
2362 skb_abort_seq_read(TS_SKB_CB(state
));
2366 * skb_find_text - Find a text pattern in skb data
2367 * @skb: the buffer to look in
2368 * @from: search offset
2370 * @config: textsearch configuration
2371 * @state: uninitialized textsearch state variable
2373 * Finds a pattern in the skb data according to the specified
2374 * textsearch configuration. Use textsearch_next() to retrieve
2375 * subsequent occurrences of the pattern. Returns the offset
2376 * to the first occurrence or UINT_MAX if no match was found.
2378 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
2379 unsigned int to
, struct ts_config
*config
,
2380 struct ts_state
*state
)
2384 config
->get_next_block
= skb_ts_get_next_block
;
2385 config
->finish
= skb_ts_finish
;
2387 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
2389 ret
= textsearch_find(config
, state
);
2390 return (ret
<= to
- from
? ret
: UINT_MAX
);
2392 EXPORT_SYMBOL(skb_find_text
);
2395 * skb_append_datato_frags: - append the user data to a skb
2396 * @sk: sock structure
2397 * @skb: skb structure to be appened with user data.
2398 * @getfrag: call back function to be used for getting the user data
2399 * @from: pointer to user message iov
2400 * @length: length of the iov message
2402 * Description: This procedure append the user data in the fragment part
2403 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2405 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
2406 int (*getfrag
)(void *from
, char *to
, int offset
,
2407 int len
, int odd
, struct sk_buff
*skb
),
2408 void *from
, int length
)
2411 skb_frag_t
*frag
= NULL
;
2412 struct page
*page
= NULL
;
2418 /* Return error if we don't have space for new frag */
2419 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2420 if (frg_cnt
>= MAX_SKB_FRAGS
)
2423 /* allocate a new page for next frag */
2424 page
= alloc_pages(sk
->sk_allocation
, 0);
2426 /* If alloc_page fails just return failure and caller will
2427 * free previous allocated pages by doing kfree_skb()
2432 /* initialize the next frag */
2433 sk
->sk_sndmsg_page
= page
;
2434 sk
->sk_sndmsg_off
= 0;
2435 skb_fill_page_desc(skb
, frg_cnt
, page
, 0, 0);
2436 skb
->truesize
+= PAGE_SIZE
;
2437 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
2439 /* get the new initialized frag */
2440 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2441 frag
= &skb_shinfo(skb
)->frags
[frg_cnt
- 1];
2443 /* copy the user data to page */
2444 left
= PAGE_SIZE
- frag
->page_offset
;
2445 copy
= (length
> left
)? left
: length
;
2447 ret
= getfrag(from
, (page_address(frag
->page
) +
2448 frag
->page_offset
+ frag
->size
),
2449 offset
, copy
, 0, skb
);
2453 /* copy was successful so update the size parameters */
2454 sk
->sk_sndmsg_off
+= copy
;
2457 skb
->data_len
+= copy
;
2461 } while (length
> 0);
2465 EXPORT_SYMBOL(skb_append_datato_frags
);
2468 * skb_pull_rcsum - pull skb and update receive checksum
2469 * @skb: buffer to update
2470 * @len: length of data pulled
2472 * This function performs an skb_pull on the packet and updates
2473 * the CHECKSUM_COMPLETE checksum. It should be used on
2474 * receive path processing instead of skb_pull unless you know
2475 * that the checksum difference is zero (e.g., a valid IP header)
2476 * or you are setting ip_summed to CHECKSUM_NONE.
2478 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
2480 BUG_ON(len
> skb
->len
);
2482 BUG_ON(skb
->len
< skb
->data_len
);
2483 skb_postpull_rcsum(skb
, skb
->data
, len
);
2484 return skb
->data
+= len
;
2487 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
2490 * skb_segment - Perform protocol segmentation on skb.
2491 * @skb: buffer to segment
2492 * @features: features for the output path (see dev->features)
2494 * This function performs segmentation on the given skb. It returns
2495 * a pointer to the first in a list of new skbs for the segments.
2496 * In case of error it returns ERR_PTR(err).
2498 struct sk_buff
*skb_segment(struct sk_buff
*skb
, int features
)
2500 struct sk_buff
*segs
= NULL
;
2501 struct sk_buff
*tail
= NULL
;
2502 struct sk_buff
*fskb
= skb_shinfo(skb
)->frag_list
;
2503 unsigned int mss
= skb_shinfo(skb
)->gso_size
;
2504 unsigned int doffset
= skb
->data
- skb_mac_header(skb
);
2505 unsigned int offset
= doffset
;
2506 unsigned int headroom
;
2508 int sg
= features
& NETIF_F_SG
;
2509 int nfrags
= skb_shinfo(skb
)->nr_frags
;
2514 __skb_push(skb
, doffset
);
2515 headroom
= skb_headroom(skb
);
2516 pos
= skb_headlen(skb
);
2519 struct sk_buff
*nskb
;
2524 len
= skb
->len
- offset
;
2528 hsize
= skb_headlen(skb
) - offset
;
2531 if (hsize
> len
|| !sg
)
2534 if (!hsize
&& i
>= nfrags
) {
2535 BUG_ON(fskb
->len
!= len
);
2538 nskb
= skb_clone(fskb
, GFP_ATOMIC
);
2541 if (unlikely(!nskb
))
2544 hsize
= skb_end_pointer(nskb
) - nskb
->head
;
2545 if (skb_cow_head(nskb
, doffset
+ headroom
)) {
2550 nskb
->truesize
+= skb_end_pointer(nskb
) - nskb
->head
-
2552 skb_release_head_state(nskb
);
2553 __skb_push(nskb
, doffset
);
2555 nskb
= alloc_skb(hsize
+ doffset
+ headroom
,
2558 if (unlikely(!nskb
))
2561 skb_reserve(nskb
, headroom
);
2562 __skb_put(nskb
, doffset
);
2571 __copy_skb_header(nskb
, skb
);
2572 nskb
->mac_len
= skb
->mac_len
;
2574 skb_reset_mac_header(nskb
);
2575 skb_set_network_header(nskb
, skb
->mac_len
);
2576 nskb
->transport_header
= (nskb
->network_header
+
2577 skb_network_header_len(skb
));
2578 skb_copy_from_linear_data(skb
, nskb
->data
, doffset
);
2580 if (fskb
!= skb_shinfo(skb
)->frag_list
)
2584 nskb
->ip_summed
= CHECKSUM_NONE
;
2585 nskb
->csum
= skb_copy_and_csum_bits(skb
, offset
,
2591 frag
= skb_shinfo(nskb
)->frags
;
2593 skb_copy_from_linear_data_offset(skb
, offset
,
2594 skb_put(nskb
, hsize
), hsize
);
2596 while (pos
< offset
+ len
&& i
< nfrags
) {
2597 *frag
= skb_shinfo(skb
)->frags
[i
];
2598 get_page(frag
->page
);
2602 frag
->page_offset
+= offset
- pos
;
2603 frag
->size
-= offset
- pos
;
2606 skb_shinfo(nskb
)->nr_frags
++;
2608 if (pos
+ size
<= offset
+ len
) {
2612 frag
->size
-= pos
+ size
- (offset
+ len
);
2619 if (pos
< offset
+ len
) {
2620 struct sk_buff
*fskb2
= fskb
;
2622 BUG_ON(pos
+ fskb
->len
!= offset
+ len
);
2628 fskb2
= skb_clone(fskb2
, GFP_ATOMIC
);
2634 SKB_FRAG_ASSERT(nskb
);
2635 skb_shinfo(nskb
)->frag_list
= fskb2
;
2639 nskb
->data_len
= len
- hsize
;
2640 nskb
->len
+= nskb
->data_len
;
2641 nskb
->truesize
+= nskb
->data_len
;
2642 } while ((offset
+= len
) < skb
->len
);
2647 while ((skb
= segs
)) {
2651 return ERR_PTR(err
);
2653 EXPORT_SYMBOL_GPL(skb_segment
);
2655 int skb_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2657 struct sk_buff
*p
= *head
;
2658 struct sk_buff
*nskb
;
2659 struct skb_shared_info
*skbinfo
= skb_shinfo(skb
);
2660 struct skb_shared_info
*pinfo
= skb_shinfo(p
);
2661 unsigned int headroom
;
2662 unsigned int len
= skb_gro_len(skb
);
2663 unsigned int offset
= skb_gro_offset(skb
);
2664 unsigned int headlen
= skb_headlen(skb
);
2666 if (p
->len
+ len
>= 65536)
2669 if (pinfo
->frag_list
)
2671 else if (headlen
<= offset
) {
2674 int i
= skbinfo
->nr_frags
;
2675 int nr_frags
= pinfo
->nr_frags
+ i
;
2679 if (nr_frags
> MAX_SKB_FRAGS
)
2682 pinfo
->nr_frags
= nr_frags
;
2683 skbinfo
->nr_frags
= 0;
2685 frag
= pinfo
->frags
+ nr_frags
;
2686 frag2
= skbinfo
->frags
+ i
;
2691 frag
->page_offset
+= offset
;
2692 frag
->size
-= offset
;
2694 skb
->truesize
-= skb
->data_len
;
2695 skb
->len
-= skb
->data_len
;
2698 NAPI_GRO_CB(skb
)->free
= 1;
2702 headroom
= skb_headroom(p
);
2703 nskb
= netdev_alloc_skb(p
->dev
, headroom
+ skb_gro_offset(p
));
2704 if (unlikely(!nskb
))
2707 __copy_skb_header(nskb
, p
);
2708 nskb
->mac_len
= p
->mac_len
;
2710 skb_reserve(nskb
, headroom
);
2711 __skb_put(nskb
, skb_gro_offset(p
));
2713 skb_set_mac_header(nskb
, skb_mac_header(p
) - p
->data
);
2714 skb_set_network_header(nskb
, skb_network_offset(p
));
2715 skb_set_transport_header(nskb
, skb_transport_offset(p
));
2717 __skb_pull(p
, skb_gro_offset(p
));
2718 memcpy(skb_mac_header(nskb
), skb_mac_header(p
),
2719 p
->data
- skb_mac_header(p
));
2721 *NAPI_GRO_CB(nskb
) = *NAPI_GRO_CB(p
);
2722 skb_shinfo(nskb
)->frag_list
= p
;
2723 skb_shinfo(nskb
)->gso_size
= pinfo
->gso_size
;
2724 skb_header_release(p
);
2727 nskb
->data_len
+= p
->len
;
2728 nskb
->truesize
+= p
->len
;
2729 nskb
->len
+= p
->len
;
2732 nskb
->next
= p
->next
;
2738 if (offset
> headlen
) {
2739 skbinfo
->frags
[0].page_offset
+= offset
- headlen
;
2740 skbinfo
->frags
[0].size
-= offset
- headlen
;
2744 __skb_pull(skb
, offset
);
2746 p
->prev
->next
= skb
;
2748 skb_header_release(skb
);
2751 NAPI_GRO_CB(p
)->count
++;
2756 NAPI_GRO_CB(skb
)->same_flow
= 1;
2759 EXPORT_SYMBOL_GPL(skb_gro_receive
);
2761 void __init
skb_init(void)
2763 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
2764 sizeof(struct sk_buff
),
2766 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2768 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
2769 (2*sizeof(struct sk_buff
)) +
2772 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2777 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2778 * @skb: Socket buffer containing the buffers to be mapped
2779 * @sg: The scatter-gather list to map into
2780 * @offset: The offset into the buffer's contents to start mapping
2781 * @len: Length of buffer space to be mapped
2783 * Fill the specified scatter-gather list with mappings/pointers into a
2784 * region of the buffer space attached to a socket buffer.
2787 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2789 int start
= skb_headlen(skb
);
2790 int i
, copy
= start
- offset
;
2791 struct sk_buff
*frag_iter
;
2797 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
2799 if ((len
-= copy
) == 0)
2804 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2807 WARN_ON(start
> offset
+ len
);
2809 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
2810 if ((copy
= end
- offset
) > 0) {
2811 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2815 sg_set_page(&sg
[elt
], frag
->page
, copy
,
2816 frag
->page_offset
+offset
-start
);
2825 skb_walk_frags(skb
, frag_iter
) {
2828 WARN_ON(start
> offset
+ len
);
2830 end
= start
+ frag_iter
->len
;
2831 if ((copy
= end
- offset
) > 0) {
2834 elt
+= __skb_to_sgvec(frag_iter
, sg
+elt
, offset
- start
,
2836 if ((len
-= copy
) == 0)
2846 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2848 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
);
2850 sg_mark_end(&sg
[nsg
- 1]);
2854 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
2857 * skb_cow_data - Check that a socket buffer's data buffers are writable
2858 * @skb: The socket buffer to check.
2859 * @tailbits: Amount of trailing space to be added
2860 * @trailer: Returned pointer to the skb where the @tailbits space begins
2862 * Make sure that the data buffers attached to a socket buffer are
2863 * writable. If they are not, private copies are made of the data buffers
2864 * and the socket buffer is set to use these instead.
2866 * If @tailbits is given, make sure that there is space to write @tailbits
2867 * bytes of data beyond current end of socket buffer. @trailer will be
2868 * set to point to the skb in which this space begins.
2870 * The number of scatterlist elements required to completely map the
2871 * COW'd and extended socket buffer will be returned.
2873 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
2877 struct sk_buff
*skb1
, **skb_p
;
2879 /* If skb is cloned or its head is paged, reallocate
2880 * head pulling out all the pages (pages are considered not writable
2881 * at the moment even if they are anonymous).
2883 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
2884 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
2887 /* Easy case. Most of packets will go this way. */
2888 if (!skb_has_frags(skb
)) {
2889 /* A little of trouble, not enough of space for trailer.
2890 * This should not happen, when stack is tuned to generate
2891 * good frames. OK, on miss we reallocate and reserve even more
2892 * space, 128 bytes is fair. */
2894 if (skb_tailroom(skb
) < tailbits
&&
2895 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
2903 /* Misery. We are in troubles, going to mincer fragments... */
2906 skb_p
= &skb_shinfo(skb
)->frag_list
;
2909 while ((skb1
= *skb_p
) != NULL
) {
2912 /* The fragment is partially pulled by someone,
2913 * this can happen on input. Copy it and everything
2916 if (skb_shared(skb1
))
2919 /* If the skb is the last, worry about trailer. */
2921 if (skb1
->next
== NULL
&& tailbits
) {
2922 if (skb_shinfo(skb1
)->nr_frags
||
2923 skb_has_frags(skb1
) ||
2924 skb_tailroom(skb1
) < tailbits
)
2925 ntail
= tailbits
+ 128;
2931 skb_shinfo(skb1
)->nr_frags
||
2932 skb_has_frags(skb1
)) {
2933 struct sk_buff
*skb2
;
2935 /* Fuck, we are miserable poor guys... */
2937 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
2939 skb2
= skb_copy_expand(skb1
,
2943 if (unlikely(skb2
== NULL
))
2947 skb_set_owner_w(skb2
, skb1
->sk
);
2949 /* Looking around. Are we still alive?
2950 * OK, link new skb, drop old one */
2952 skb2
->next
= skb1
->next
;
2959 skb_p
= &skb1
->next
;
2964 EXPORT_SYMBOL_GPL(skb_cow_data
);
2966 void skb_tstamp_tx(struct sk_buff
*orig_skb
,
2967 struct skb_shared_hwtstamps
*hwtstamps
)
2969 struct sock
*sk
= orig_skb
->sk
;
2970 struct sock_exterr_skb
*serr
;
2971 struct sk_buff
*skb
;
2977 skb
= skb_clone(orig_skb
, GFP_ATOMIC
);
2982 *skb_hwtstamps(skb
) =
2986 * no hardware time stamps available,
2987 * so keep the skb_shared_tx and only
2988 * store software time stamp
2990 skb
->tstamp
= ktime_get_real();
2993 serr
= SKB_EXT_ERR(skb
);
2994 memset(serr
, 0, sizeof(*serr
));
2995 serr
->ee
.ee_errno
= ENOMSG
;
2996 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TIMESTAMPING
;
2997 err
= sock_queue_err_skb(sk
, skb
);
3001 EXPORT_SYMBOL_GPL(skb_tstamp_tx
);
3005 * skb_partial_csum_set - set up and verify partial csum values for packet
3006 * @skb: the skb to set
3007 * @start: the number of bytes after skb->data to start checksumming.
3008 * @off: the offset from start to place the checksum.
3010 * For untrusted partially-checksummed packets, we need to make sure the values
3011 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3013 * This function checks and sets those values and skb->ip_summed: if this
3014 * returns false you should drop the packet.
3016 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
3018 if (unlikely(start
> skb_headlen(skb
)) ||
3019 unlikely((int)start
+ off
> skb_headlen(skb
) - 2)) {
3020 if (net_ratelimit())
3022 "bad partial csum: csum=%u/%u len=%u\n",
3023 start
, off
, skb_headlen(skb
));
3026 skb
->ip_summed
= CHECKSUM_PARTIAL
;
3027 skb
->csum_start
= skb_headroom(skb
) + start
;
3028 skb
->csum_offset
= off
;
3031 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);
3033 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
3035 if (net_ratelimit())
3036 pr_warning("%s: received packets cannot be forwarded"
3037 " while LRO is enabled\n", skb
->dev
->name
);
3039 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);