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 const 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 #ifdef NET_SKBUFF_DATA_USES_OFFSET
208 skb
->mac_header
= ~0U;
211 /* make sure we initialize shinfo sequentially */
212 shinfo
= skb_shinfo(skb
);
213 atomic_set(&shinfo
->dataref
, 1);
214 shinfo
->nr_frags
= 0;
215 shinfo
->gso_size
= 0;
216 shinfo
->gso_segs
= 0;
217 shinfo
->gso_type
= 0;
218 shinfo
->ip6_frag_id
= 0;
219 shinfo
->tx_flags
.flags
= 0;
220 skb_frag_list_init(skb
);
221 memset(&shinfo
->hwtstamps
, 0, sizeof(shinfo
->hwtstamps
));
224 struct sk_buff
*child
= skb
+ 1;
225 atomic_t
*fclone_ref
= (atomic_t
*) (child
+ 1);
227 kmemcheck_annotate_bitfield(child
, flags1
);
228 kmemcheck_annotate_bitfield(child
, flags2
);
229 skb
->fclone
= SKB_FCLONE_ORIG
;
230 atomic_set(fclone_ref
, 1);
232 child
->fclone
= SKB_FCLONE_UNAVAILABLE
;
237 kmem_cache_free(cache
, skb
);
241 EXPORT_SYMBOL(__alloc_skb
);
244 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
245 * @dev: network device to receive on
246 * @length: length to allocate
247 * @gfp_mask: get_free_pages mask, passed to alloc_skb
249 * Allocate a new &sk_buff and assign it a usage count of one. The
250 * buffer has unspecified headroom built in. Users should allocate
251 * the headroom they think they need without accounting for the
252 * built in space. The built in space is used for optimisations.
254 * %NULL is returned if there is no free memory.
256 struct sk_buff
*__netdev_alloc_skb(struct net_device
*dev
,
257 unsigned int length
, gfp_t gfp_mask
)
259 int node
= dev
->dev
.parent
? dev_to_node(dev
->dev
.parent
) : -1;
262 skb
= __alloc_skb(length
+ NET_SKB_PAD
, gfp_mask
, 0, node
);
264 skb_reserve(skb
, NET_SKB_PAD
);
269 EXPORT_SYMBOL(__netdev_alloc_skb
);
271 struct page
*__netdev_alloc_page(struct net_device
*dev
, gfp_t gfp_mask
)
273 int node
= dev
->dev
.parent
? dev_to_node(dev
->dev
.parent
) : -1;
276 page
= alloc_pages_node(node
, gfp_mask
, 0);
279 EXPORT_SYMBOL(__netdev_alloc_page
);
281 void skb_add_rx_frag(struct sk_buff
*skb
, int i
, struct page
*page
, int off
,
284 skb_fill_page_desc(skb
, i
, page
, off
, size
);
286 skb
->data_len
+= size
;
287 skb
->truesize
+= size
;
289 EXPORT_SYMBOL(skb_add_rx_frag
);
292 * dev_alloc_skb - allocate an skbuff for receiving
293 * @length: length to allocate
295 * Allocate a new &sk_buff and assign it a usage count of one. The
296 * buffer has unspecified headroom built in. Users should allocate
297 * the headroom they think they need without accounting for the
298 * built in space. The built in space is used for optimisations.
300 * %NULL is returned if there is no free memory. Although this function
301 * allocates memory it can be called from an interrupt.
303 struct sk_buff
*dev_alloc_skb(unsigned int length
)
306 * There is more code here than it seems:
307 * __dev_alloc_skb is an inline
309 return __dev_alloc_skb(length
, GFP_ATOMIC
);
311 EXPORT_SYMBOL(dev_alloc_skb
);
313 static void skb_drop_list(struct sk_buff
**listp
)
315 struct sk_buff
*list
= *listp
;
320 struct sk_buff
*this = list
;
326 static inline void skb_drop_fraglist(struct sk_buff
*skb
)
328 skb_drop_list(&skb_shinfo(skb
)->frag_list
);
331 static void skb_clone_fraglist(struct sk_buff
*skb
)
333 struct sk_buff
*list
;
335 skb_walk_frags(skb
, list
)
339 static void skb_release_data(struct sk_buff
*skb
)
342 !atomic_sub_return(skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1,
343 &skb_shinfo(skb
)->dataref
)) {
344 if (skb_shinfo(skb
)->nr_frags
) {
346 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
347 put_page(skb_shinfo(skb
)->frags
[i
].page
);
350 if (skb_has_frags(skb
))
351 skb_drop_fraglist(skb
);
358 * Free an skbuff by memory without cleaning the state.
360 static void kfree_skbmem(struct sk_buff
*skb
)
362 struct sk_buff
*other
;
363 atomic_t
*fclone_ref
;
365 switch (skb
->fclone
) {
366 case SKB_FCLONE_UNAVAILABLE
:
367 kmem_cache_free(skbuff_head_cache
, skb
);
370 case SKB_FCLONE_ORIG
:
371 fclone_ref
= (atomic_t
*) (skb
+ 2);
372 if (atomic_dec_and_test(fclone_ref
))
373 kmem_cache_free(skbuff_fclone_cache
, skb
);
376 case SKB_FCLONE_CLONE
:
377 fclone_ref
= (atomic_t
*) (skb
+ 1);
380 /* The clone portion is available for
381 * fast-cloning again.
383 skb
->fclone
= SKB_FCLONE_UNAVAILABLE
;
385 if (atomic_dec_and_test(fclone_ref
))
386 kmem_cache_free(skbuff_fclone_cache
, other
);
391 static void skb_release_head_state(struct sk_buff
*skb
)
395 secpath_put(skb
->sp
);
397 if (skb
->destructor
) {
399 skb
->destructor(skb
);
401 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
402 nf_conntrack_put(skb
->nfct
);
403 nf_conntrack_put_reasm(skb
->nfct_reasm
);
405 #ifdef CONFIG_BRIDGE_NETFILTER
406 nf_bridge_put(skb
->nf_bridge
);
408 /* XXX: IS this still necessary? - JHS */
409 #ifdef CONFIG_NET_SCHED
411 #ifdef CONFIG_NET_CLS_ACT
417 /* Free everything but the sk_buff shell. */
418 static void skb_release_all(struct sk_buff
*skb
)
420 skb_release_head_state(skb
);
421 skb_release_data(skb
);
425 * __kfree_skb - private function
428 * Free an sk_buff. Release anything attached to the buffer.
429 * Clean the state. This is an internal helper function. Users should
430 * always call kfree_skb
433 void __kfree_skb(struct sk_buff
*skb
)
435 skb_release_all(skb
);
438 EXPORT_SYMBOL(__kfree_skb
);
441 * kfree_skb - free an sk_buff
442 * @skb: buffer to free
444 * Drop a reference to the buffer and free it if the usage count has
447 void kfree_skb(struct sk_buff
*skb
)
451 if (likely(atomic_read(&skb
->users
) == 1))
453 else if (likely(!atomic_dec_and_test(&skb
->users
)))
455 trace_kfree_skb(skb
, __builtin_return_address(0));
458 EXPORT_SYMBOL(kfree_skb
);
461 * consume_skb - free an skbuff
462 * @skb: buffer to free
464 * Drop a ref to the buffer and free it if the usage count has hit zero
465 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
466 * is being dropped after a failure and notes that
468 void consume_skb(struct sk_buff
*skb
)
472 if (likely(atomic_read(&skb
->users
) == 1))
474 else if (likely(!atomic_dec_and_test(&skb
->users
)))
478 EXPORT_SYMBOL(consume_skb
);
481 * skb_recycle_check - check if skb can be reused for receive
483 * @skb_size: minimum receive buffer size
485 * Checks that the skb passed in is not shared or cloned, and
486 * that it is linear and its head portion at least as large as
487 * skb_size so that it can be recycled as a receive buffer.
488 * If these conditions are met, this function does any necessary
489 * reference count dropping and cleans up the skbuff as if it
490 * just came from __alloc_skb().
492 int skb_recycle_check(struct sk_buff
*skb
, int skb_size
)
494 struct skb_shared_info
*shinfo
;
499 if (skb_is_nonlinear(skb
) || skb
->fclone
!= SKB_FCLONE_UNAVAILABLE
)
502 skb_size
= SKB_DATA_ALIGN(skb_size
+ NET_SKB_PAD
);
503 if (skb_end_pointer(skb
) - skb
->head
< skb_size
)
506 if (skb_shared(skb
) || skb_cloned(skb
))
509 skb_release_head_state(skb
);
510 shinfo
= skb_shinfo(skb
);
511 atomic_set(&shinfo
->dataref
, 1);
512 shinfo
->nr_frags
= 0;
513 shinfo
->gso_size
= 0;
514 shinfo
->gso_segs
= 0;
515 shinfo
->gso_type
= 0;
516 shinfo
->ip6_frag_id
= 0;
517 shinfo
->tx_flags
.flags
= 0;
518 skb_frag_list_init(skb
);
519 memset(&shinfo
->hwtstamps
, 0, sizeof(shinfo
->hwtstamps
));
521 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
522 skb
->data
= skb
->head
+ NET_SKB_PAD
;
523 skb_reset_tail_pointer(skb
);
527 EXPORT_SYMBOL(skb_recycle_check
);
529 static void __copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
531 new->tstamp
= old
->tstamp
;
533 new->transport_header
= old
->transport_header
;
534 new->network_header
= old
->network_header
;
535 new->mac_header
= old
->mac_header
;
536 skb_dst_set(new, dst_clone(skb_dst(old
)));
537 new->rxhash
= old
->rxhash
;
539 new->sp
= secpath_get(old
->sp
);
541 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
542 new->csum
= old
->csum
;
543 new->local_df
= old
->local_df
;
544 new->pkt_type
= old
->pkt_type
;
545 new->ip_summed
= old
->ip_summed
;
546 skb_copy_queue_mapping(new, old
);
547 new->priority
= old
->priority
;
548 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
549 new->ipvs_property
= old
->ipvs_property
;
551 new->protocol
= old
->protocol
;
552 new->mark
= old
->mark
;
553 new->skb_iif
= old
->skb_iif
;
555 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
556 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
557 new->nf_trace
= old
->nf_trace
;
559 #ifdef CONFIG_NET_SCHED
560 new->tc_index
= old
->tc_index
;
561 #ifdef CONFIG_NET_CLS_ACT
562 new->tc_verd
= old
->tc_verd
;
565 new->vlan_tci
= old
->vlan_tci
;
567 skb_copy_secmark(new, old
);
571 * You should not add any new code to this function. Add it to
572 * __copy_skb_header above instead.
574 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
576 #define C(x) n->x = skb->x
578 n
->next
= n
->prev
= NULL
;
580 __copy_skb_header(n
, skb
);
586 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
589 n
->destructor
= NULL
;
595 atomic_set(&n
->users
, 1);
597 atomic_inc(&(skb_shinfo(skb
)->dataref
));
605 * skb_morph - morph one skb into another
606 * @dst: the skb to receive the contents
607 * @src: the skb to supply the contents
609 * This is identical to skb_clone except that the target skb is
610 * supplied by the user.
612 * The target skb is returned upon exit.
614 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
616 skb_release_all(dst
);
617 return __skb_clone(dst
, src
);
619 EXPORT_SYMBOL_GPL(skb_morph
);
622 * skb_clone - duplicate an sk_buff
623 * @skb: buffer to clone
624 * @gfp_mask: allocation priority
626 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
627 * copies share the same packet data but not structure. The new
628 * buffer has a reference count of 1. If the allocation fails the
629 * function returns %NULL otherwise the new buffer is returned.
631 * If this function is called from an interrupt gfp_mask() must be
635 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
640 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
641 n
->fclone
== SKB_FCLONE_UNAVAILABLE
) {
642 atomic_t
*fclone_ref
= (atomic_t
*) (n
+ 1);
643 n
->fclone
= SKB_FCLONE_CLONE
;
644 atomic_inc(fclone_ref
);
646 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
650 kmemcheck_annotate_bitfield(n
, flags1
);
651 kmemcheck_annotate_bitfield(n
, flags2
);
652 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
655 return __skb_clone(n
, skb
);
657 EXPORT_SYMBOL(skb_clone
);
659 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
661 #ifndef NET_SKBUFF_DATA_USES_OFFSET
663 * Shift between the two data areas in bytes
665 unsigned long offset
= new->data
- old
->data
;
668 __copy_skb_header(new, old
);
670 #ifndef NET_SKBUFF_DATA_USES_OFFSET
671 /* {transport,network,mac}_header are relative to skb->head */
672 new->transport_header
+= offset
;
673 new->network_header
+= offset
;
674 if (skb_mac_header_was_set(new))
675 new->mac_header
+= offset
;
677 skb_shinfo(new)->gso_size
= skb_shinfo(old
)->gso_size
;
678 skb_shinfo(new)->gso_segs
= skb_shinfo(old
)->gso_segs
;
679 skb_shinfo(new)->gso_type
= skb_shinfo(old
)->gso_type
;
683 * skb_copy - create private copy of an sk_buff
684 * @skb: buffer to copy
685 * @gfp_mask: allocation priority
687 * Make a copy of both an &sk_buff and its data. This is used when the
688 * caller wishes to modify the data and needs a private copy of the
689 * data to alter. Returns %NULL on failure or the pointer to the buffer
690 * on success. The returned buffer has a reference count of 1.
692 * As by-product this function converts non-linear &sk_buff to linear
693 * one, so that &sk_buff becomes completely private and caller is allowed
694 * to modify all the data of returned buffer. This means that this
695 * function is not recommended for use in circumstances when only
696 * header is going to be modified. Use pskb_copy() instead.
699 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
701 int headerlen
= skb
->data
- skb
->head
;
703 * Allocate the copy buffer
706 #ifdef NET_SKBUFF_DATA_USES_OFFSET
707 n
= alloc_skb(skb
->end
+ skb
->data_len
, gfp_mask
);
709 n
= alloc_skb(skb
->end
- skb
->head
+ skb
->data_len
, gfp_mask
);
714 /* Set the data pointer */
715 skb_reserve(n
, headerlen
);
716 /* Set the tail pointer and length */
717 skb_put(n
, skb
->len
);
719 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
722 copy_skb_header(n
, skb
);
725 EXPORT_SYMBOL(skb_copy
);
728 * pskb_copy - create copy of an sk_buff with private head.
729 * @skb: buffer to copy
730 * @gfp_mask: allocation priority
732 * Make a copy of both an &sk_buff and part of its data, located
733 * in header. Fragmented data remain shared. This is used when
734 * the caller wishes to modify only header of &sk_buff and needs
735 * private copy of the header to alter. Returns %NULL on failure
736 * or the pointer to the buffer on success.
737 * The returned buffer has a reference count of 1.
740 struct sk_buff
*pskb_copy(struct sk_buff
*skb
, gfp_t gfp_mask
)
743 * Allocate the copy buffer
746 #ifdef NET_SKBUFF_DATA_USES_OFFSET
747 n
= alloc_skb(skb
->end
, gfp_mask
);
749 n
= alloc_skb(skb
->end
- skb
->head
, gfp_mask
);
754 /* Set the data pointer */
755 skb_reserve(n
, skb
->data
- skb
->head
);
756 /* Set the tail pointer and length */
757 skb_put(n
, skb_headlen(skb
));
759 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
761 n
->truesize
+= skb
->data_len
;
762 n
->data_len
= skb
->data_len
;
765 if (skb_shinfo(skb
)->nr_frags
) {
768 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
769 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
770 get_page(skb_shinfo(n
)->frags
[i
].page
);
772 skb_shinfo(n
)->nr_frags
= i
;
775 if (skb_has_frags(skb
)) {
776 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
777 skb_clone_fraglist(n
);
780 copy_skb_header(n
, skb
);
784 EXPORT_SYMBOL(pskb_copy
);
787 * pskb_expand_head - reallocate header of &sk_buff
788 * @skb: buffer to reallocate
789 * @nhead: room to add at head
790 * @ntail: room to add at tail
791 * @gfp_mask: allocation priority
793 * Expands (or creates identical copy, if &nhead and &ntail are zero)
794 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
795 * reference count of 1. Returns zero in the case of success or error,
796 * if expansion failed. In the last case, &sk_buff is not changed.
798 * All the pointers pointing into skb header may change and must be
799 * reloaded after call to this function.
802 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
807 #ifdef NET_SKBUFF_DATA_USES_OFFSET
808 int size
= nhead
+ skb
->end
+ ntail
;
810 int size
= nhead
+ (skb
->end
- skb
->head
) + ntail
;
819 size
= SKB_DATA_ALIGN(size
);
821 data
= kmalloc(size
+ sizeof(struct skb_shared_info
), gfp_mask
);
825 /* Copy only real data... and, alas, header. This should be
826 * optimized for the cases when header is void. */
827 #ifdef NET_SKBUFF_DATA_USES_OFFSET
828 memcpy(data
+ nhead
, skb
->head
, skb
->tail
);
830 memcpy(data
+ nhead
, skb
->head
, skb
->tail
- skb
->head
);
832 memcpy(data
+ size
, skb_end_pointer(skb
),
833 sizeof(struct skb_shared_info
));
835 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
836 get_page(skb_shinfo(skb
)->frags
[i
].page
);
838 if (skb_has_frags(skb
))
839 skb_clone_fraglist(skb
);
841 skb_release_data(skb
);
843 off
= (data
+ nhead
) - skb
->head
;
847 #ifdef NET_SKBUFF_DATA_USES_OFFSET
851 skb
->end
= skb
->head
+ size
;
853 /* {transport,network,mac}_header and tail are relative to skb->head */
855 skb
->transport_header
+= off
;
856 skb
->network_header
+= off
;
857 if (skb_mac_header_was_set(skb
))
858 skb
->mac_header
+= off
;
859 skb
->csum_start
+= nhead
;
863 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
869 EXPORT_SYMBOL(pskb_expand_head
);
871 /* Make private copy of skb with writable head and some headroom */
873 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
875 struct sk_buff
*skb2
;
876 int delta
= headroom
- skb_headroom(skb
);
879 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
881 skb2
= skb_clone(skb
, GFP_ATOMIC
);
882 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
890 EXPORT_SYMBOL(skb_realloc_headroom
);
893 * skb_copy_expand - copy and expand sk_buff
894 * @skb: buffer to copy
895 * @newheadroom: new free bytes at head
896 * @newtailroom: new free bytes at tail
897 * @gfp_mask: allocation priority
899 * Make a copy of both an &sk_buff and its data and while doing so
900 * allocate additional space.
902 * This is used when the caller wishes to modify the data and needs a
903 * private copy of the data to alter as well as more space for new fields.
904 * Returns %NULL on failure or the pointer to the buffer
905 * on success. The returned buffer has a reference count of 1.
907 * You must pass %GFP_ATOMIC as the allocation priority if this function
908 * is called from an interrupt.
910 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
911 int newheadroom
, int newtailroom
,
915 * Allocate the copy buffer
917 struct sk_buff
*n
= alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
919 int oldheadroom
= skb_headroom(skb
);
920 int head_copy_len
, head_copy_off
;
926 skb_reserve(n
, newheadroom
);
928 /* Set the tail pointer and length */
929 skb_put(n
, skb
->len
);
931 head_copy_len
= oldheadroom
;
933 if (newheadroom
<= head_copy_len
)
934 head_copy_len
= newheadroom
;
936 head_copy_off
= newheadroom
- head_copy_len
;
938 /* Copy the linear header and data. */
939 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
940 skb
->len
+ head_copy_len
))
943 copy_skb_header(n
, skb
);
945 off
= newheadroom
- oldheadroom
;
946 n
->csum_start
+= off
;
947 #ifdef NET_SKBUFF_DATA_USES_OFFSET
948 n
->transport_header
+= off
;
949 n
->network_header
+= off
;
950 if (skb_mac_header_was_set(skb
))
951 n
->mac_header
+= off
;
956 EXPORT_SYMBOL(skb_copy_expand
);
959 * skb_pad - zero pad the tail of an skb
960 * @skb: buffer to pad
963 * Ensure that a buffer is followed by a padding area that is zero
964 * filled. Used by network drivers which may DMA or transfer data
965 * beyond the buffer end onto the wire.
967 * May return error in out of memory cases. The skb is freed on error.
970 int skb_pad(struct sk_buff
*skb
, int pad
)
975 /* If the skbuff is non linear tailroom is always zero.. */
976 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
977 memset(skb
->data
+skb
->len
, 0, pad
);
981 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
982 if (likely(skb_cloned(skb
) || ntail
> 0)) {
983 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
988 /* FIXME: The use of this function with non-linear skb's really needs
991 err
= skb_linearize(skb
);
995 memset(skb
->data
+ skb
->len
, 0, pad
);
1002 EXPORT_SYMBOL(skb_pad
);
1005 * skb_put - add data to a buffer
1006 * @skb: buffer to use
1007 * @len: amount of data to add
1009 * This function extends the used data area of the buffer. If this would
1010 * exceed the total buffer size the kernel will panic. A pointer to the
1011 * first byte of the extra data is returned.
1013 unsigned char *skb_put(struct sk_buff
*skb
, unsigned int len
)
1015 unsigned char *tmp
= skb_tail_pointer(skb
);
1016 SKB_LINEAR_ASSERT(skb
);
1019 if (unlikely(skb
->tail
> skb
->end
))
1020 skb_over_panic(skb
, len
, __builtin_return_address(0));
1023 EXPORT_SYMBOL(skb_put
);
1026 * skb_push - add data to the start of a buffer
1027 * @skb: buffer to use
1028 * @len: amount of data to add
1030 * This function extends the used data area of the buffer at the buffer
1031 * start. If this would exceed the total buffer headroom the kernel will
1032 * panic. A pointer to the first byte of the extra data is returned.
1034 unsigned char *skb_push(struct sk_buff
*skb
, unsigned int len
)
1038 if (unlikely(skb
->data
<skb
->head
))
1039 skb_under_panic(skb
, len
, __builtin_return_address(0));
1042 EXPORT_SYMBOL(skb_push
);
1045 * skb_pull - remove data from the start of a buffer
1046 * @skb: buffer to use
1047 * @len: amount of data to remove
1049 * This function removes data from the start of a buffer, returning
1050 * the memory to the headroom. A pointer to the next data in the buffer
1051 * is returned. Once the data has been pulled future pushes will overwrite
1054 unsigned char *skb_pull(struct sk_buff
*skb
, unsigned int len
)
1056 return unlikely(len
> skb
->len
) ? NULL
: __skb_pull(skb
, len
);
1058 EXPORT_SYMBOL(skb_pull
);
1061 * skb_trim - remove end from a buffer
1062 * @skb: buffer to alter
1065 * Cut the length of a buffer down by removing data from the tail. If
1066 * the buffer is already under the length specified it is not modified.
1067 * The skb must be linear.
1069 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1072 __skb_trim(skb
, len
);
1074 EXPORT_SYMBOL(skb_trim
);
1076 /* Trims skb to length len. It can change skb pointers.
1079 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1081 struct sk_buff
**fragp
;
1082 struct sk_buff
*frag
;
1083 int offset
= skb_headlen(skb
);
1084 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1088 if (skb_cloned(skb
) &&
1089 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1096 for (; i
< nfrags
; i
++) {
1097 int end
= offset
+ skb_shinfo(skb
)->frags
[i
].size
;
1104 skb_shinfo(skb
)->frags
[i
++].size
= len
- offset
;
1107 skb_shinfo(skb
)->nr_frags
= i
;
1109 for (; i
< nfrags
; i
++)
1110 put_page(skb_shinfo(skb
)->frags
[i
].page
);
1112 if (skb_has_frags(skb
))
1113 skb_drop_fraglist(skb
);
1117 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1118 fragp
= &frag
->next
) {
1119 int end
= offset
+ frag
->len
;
1121 if (skb_shared(frag
)) {
1122 struct sk_buff
*nfrag
;
1124 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1125 if (unlikely(!nfrag
))
1128 nfrag
->next
= frag
->next
;
1140 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1144 skb_drop_list(&frag
->next
);
1149 if (len
> skb_headlen(skb
)) {
1150 skb
->data_len
-= skb
->len
- len
;
1155 skb_set_tail_pointer(skb
, len
);
1160 EXPORT_SYMBOL(___pskb_trim
);
1163 * __pskb_pull_tail - advance tail of skb header
1164 * @skb: buffer to reallocate
1165 * @delta: number of bytes to advance tail
1167 * The function makes a sense only on a fragmented &sk_buff,
1168 * it expands header moving its tail forward and copying necessary
1169 * data from fragmented part.
1171 * &sk_buff MUST have reference count of 1.
1173 * Returns %NULL (and &sk_buff does not change) if pull failed
1174 * or value of new tail of skb in the case of success.
1176 * All the pointers pointing into skb header may change and must be
1177 * reloaded after call to this function.
1180 /* Moves tail of skb head forward, copying data from fragmented part,
1181 * when it is necessary.
1182 * 1. It may fail due to malloc failure.
1183 * 2. It may change skb pointers.
1185 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1187 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1189 /* If skb has not enough free space at tail, get new one
1190 * plus 128 bytes for future expansions. If we have enough
1191 * room at tail, reallocate without expansion only if skb is cloned.
1193 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1195 if (eat
> 0 || skb_cloned(skb
)) {
1196 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1201 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
1204 /* Optimization: no fragments, no reasons to preestimate
1205 * size of pulled pages. Superb.
1207 if (!skb_has_frags(skb
))
1210 /* Estimate size of pulled pages. */
1212 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1213 if (skb_shinfo(skb
)->frags
[i
].size
>= eat
)
1215 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1218 /* If we need update frag list, we are in troubles.
1219 * Certainly, it possible to add an offset to skb data,
1220 * but taking into account that pulling is expected to
1221 * be very rare operation, it is worth to fight against
1222 * further bloating skb head and crucify ourselves here instead.
1223 * Pure masohism, indeed. 8)8)
1226 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1227 struct sk_buff
*clone
= NULL
;
1228 struct sk_buff
*insp
= NULL
;
1233 if (list
->len
<= eat
) {
1234 /* Eaten as whole. */
1239 /* Eaten partially. */
1241 if (skb_shared(list
)) {
1242 /* Sucks! We need to fork list. :-( */
1243 clone
= skb_clone(list
, GFP_ATOMIC
);
1249 /* This may be pulled without
1253 if (!pskb_pull(list
, eat
)) {
1261 /* Free pulled out fragments. */
1262 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1263 skb_shinfo(skb
)->frag_list
= list
->next
;
1266 /* And insert new clone at head. */
1269 skb_shinfo(skb
)->frag_list
= clone
;
1272 /* Success! Now we may commit changes to skb data. */
1277 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1278 if (skb_shinfo(skb
)->frags
[i
].size
<= eat
) {
1279 put_page(skb_shinfo(skb
)->frags
[i
].page
);
1280 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1282 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1284 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1285 skb_shinfo(skb
)->frags
[k
].size
-= eat
;
1291 skb_shinfo(skb
)->nr_frags
= k
;
1294 skb
->data_len
-= delta
;
1296 return skb_tail_pointer(skb
);
1298 EXPORT_SYMBOL(__pskb_pull_tail
);
1300 /* Copy some data bits from skb to kernel buffer. */
1302 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1304 int start
= skb_headlen(skb
);
1305 struct sk_buff
*frag_iter
;
1308 if (offset
> (int)skb
->len
- len
)
1312 if ((copy
= start
- offset
) > 0) {
1315 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
1316 if ((len
-= copy
) == 0)
1322 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1325 WARN_ON(start
> offset
+ len
);
1327 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1328 if ((copy
= end
- offset
) > 0) {
1334 vaddr
= kmap_skb_frag(&skb_shinfo(skb
)->frags
[i
]);
1336 vaddr
+ skb_shinfo(skb
)->frags
[i
].page_offset
+
1337 offset
- start
, copy
);
1338 kunmap_skb_frag(vaddr
);
1340 if ((len
-= copy
) == 0)
1348 skb_walk_frags(skb
, frag_iter
) {
1351 WARN_ON(start
> offset
+ len
);
1353 end
= start
+ frag_iter
->len
;
1354 if ((copy
= end
- offset
) > 0) {
1357 if (skb_copy_bits(frag_iter
, offset
- start
, to
, copy
))
1359 if ((len
-= copy
) == 0)
1372 EXPORT_SYMBOL(skb_copy_bits
);
1375 * Callback from splice_to_pipe(), if we need to release some pages
1376 * at the end of the spd in case we error'ed out in filling the pipe.
1378 static void sock_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
1380 put_page(spd
->pages
[i
]);
1383 static inline struct page
*linear_to_page(struct page
*page
, unsigned int *len
,
1384 unsigned int *offset
,
1385 struct sk_buff
*skb
, struct sock
*sk
)
1387 struct page
*p
= sk
->sk_sndmsg_page
;
1392 p
= sk
->sk_sndmsg_page
= alloc_pages(sk
->sk_allocation
, 0);
1396 off
= sk
->sk_sndmsg_off
= 0;
1397 /* hold one ref to this page until it's full */
1401 off
= sk
->sk_sndmsg_off
;
1402 mlen
= PAGE_SIZE
- off
;
1403 if (mlen
< 64 && mlen
< *len
) {
1408 *len
= min_t(unsigned int, *len
, mlen
);
1411 memcpy(page_address(p
) + off
, page_address(page
) + *offset
, *len
);
1412 sk
->sk_sndmsg_off
+= *len
;
1420 * Fill page/offset/length into spd, if it can hold more pages.
1422 static inline int spd_fill_page(struct splice_pipe_desc
*spd
, struct page
*page
,
1423 unsigned int *len
, unsigned int offset
,
1424 struct sk_buff
*skb
, int linear
,
1427 if (unlikely(spd
->nr_pages
== PIPE_BUFFERS
))
1431 page
= linear_to_page(page
, len
, &offset
, skb
, sk
);
1437 spd
->pages
[spd
->nr_pages
] = page
;
1438 spd
->partial
[spd
->nr_pages
].len
= *len
;
1439 spd
->partial
[spd
->nr_pages
].offset
= offset
;
1445 static inline void __segment_seek(struct page
**page
, unsigned int *poff
,
1446 unsigned int *plen
, unsigned int off
)
1451 n
= *poff
/ PAGE_SIZE
;
1453 *page
= nth_page(*page
, n
);
1455 *poff
= *poff
% PAGE_SIZE
;
1459 static inline int __splice_segment(struct page
*page
, unsigned int poff
,
1460 unsigned int plen
, unsigned int *off
,
1461 unsigned int *len
, struct sk_buff
*skb
,
1462 struct splice_pipe_desc
*spd
, int linear
,
1468 /* skip this segment if already processed */
1474 /* ignore any bits we already processed */
1476 __segment_seek(&page
, &poff
, &plen
, *off
);
1481 unsigned int flen
= min(*len
, plen
);
1483 /* the linear region may spread across several pages */
1484 flen
= min_t(unsigned int, flen
, PAGE_SIZE
- poff
);
1486 if (spd_fill_page(spd
, page
, &flen
, poff
, skb
, linear
, sk
))
1489 __segment_seek(&page
, &poff
, &plen
, flen
);
1492 } while (*len
&& plen
);
1498 * Map linear and fragment data from the skb to spd. It reports failure if the
1499 * pipe is full or if we already spliced the requested length.
1501 static int __skb_splice_bits(struct sk_buff
*skb
, unsigned int *offset
,
1502 unsigned int *len
, struct splice_pipe_desc
*spd
,
1508 * map the linear part
1510 if (__splice_segment(virt_to_page(skb
->data
),
1511 (unsigned long) skb
->data
& (PAGE_SIZE
- 1),
1513 offset
, len
, skb
, spd
, 1, sk
))
1517 * then map the fragments
1519 for (seg
= 0; seg
< skb_shinfo(skb
)->nr_frags
; seg
++) {
1520 const skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[seg
];
1522 if (__splice_segment(f
->page
, f
->page_offset
, f
->size
,
1523 offset
, len
, skb
, spd
, 0, sk
))
1531 * Map data from the skb to a pipe. Should handle both the linear part,
1532 * the fragments, and the frag list. It does NOT handle frag lists within
1533 * the frag list, if such a thing exists. We'd probably need to recurse to
1534 * handle that cleanly.
1536 int skb_splice_bits(struct sk_buff
*skb
, unsigned int offset
,
1537 struct pipe_inode_info
*pipe
, unsigned int tlen
,
1540 struct partial_page partial
[PIPE_BUFFERS
];
1541 struct page
*pages
[PIPE_BUFFERS
];
1542 struct splice_pipe_desc spd
= {
1546 .ops
= &sock_pipe_buf_ops
,
1547 .spd_release
= sock_spd_release
,
1549 struct sk_buff
*frag_iter
;
1550 struct sock
*sk
= skb
->sk
;
1553 * __skb_splice_bits() only fails if the output has no room left,
1554 * so no point in going over the frag_list for the error case.
1556 if (__skb_splice_bits(skb
, &offset
, &tlen
, &spd
, sk
))
1562 * now see if we have a frag_list to map
1564 skb_walk_frags(skb
, frag_iter
) {
1567 if (__skb_splice_bits(frag_iter
, &offset
, &tlen
, &spd
, sk
))
1576 * Drop the socket lock, otherwise we have reverse
1577 * locking dependencies between sk_lock and i_mutex
1578 * here as compared to sendfile(). We enter here
1579 * with the socket lock held, and splice_to_pipe() will
1580 * grab the pipe inode lock. For sendfile() emulation,
1581 * we call into ->sendpage() with the i_mutex lock held
1582 * and networking will grab the socket lock.
1585 ret
= splice_to_pipe(pipe
, &spd
);
1594 * skb_store_bits - store bits from kernel buffer to skb
1595 * @skb: destination buffer
1596 * @offset: offset in destination
1597 * @from: source buffer
1598 * @len: number of bytes to copy
1600 * Copy the specified number of bytes from the source buffer to the
1601 * destination skb. This function handles all the messy bits of
1602 * traversing fragment lists and such.
1605 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
1607 int start
= skb_headlen(skb
);
1608 struct sk_buff
*frag_iter
;
1611 if (offset
> (int)skb
->len
- len
)
1614 if ((copy
= start
- offset
) > 0) {
1617 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
1618 if ((len
-= copy
) == 0)
1624 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1625 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1628 WARN_ON(start
> offset
+ len
);
1630 end
= start
+ frag
->size
;
1631 if ((copy
= end
- offset
) > 0) {
1637 vaddr
= kmap_skb_frag(frag
);
1638 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1640 kunmap_skb_frag(vaddr
);
1642 if ((len
-= copy
) == 0)
1650 skb_walk_frags(skb
, frag_iter
) {
1653 WARN_ON(start
> offset
+ len
);
1655 end
= start
+ frag_iter
->len
;
1656 if ((copy
= end
- offset
) > 0) {
1659 if (skb_store_bits(frag_iter
, offset
- start
,
1662 if ((len
-= copy
) == 0)
1675 EXPORT_SYMBOL(skb_store_bits
);
1677 /* Checksum skb data. */
1679 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
1680 int len
, __wsum csum
)
1682 int start
= skb_headlen(skb
);
1683 int i
, copy
= start
- offset
;
1684 struct sk_buff
*frag_iter
;
1687 /* Checksum header. */
1691 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1692 if ((len
-= copy
) == 0)
1698 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1701 WARN_ON(start
> offset
+ len
);
1703 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1704 if ((copy
= end
- offset
) > 0) {
1707 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1711 vaddr
= kmap_skb_frag(frag
);
1712 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
1713 offset
- start
, copy
, 0);
1714 kunmap_skb_frag(vaddr
);
1715 csum
= csum_block_add(csum
, csum2
, pos
);
1724 skb_walk_frags(skb
, frag_iter
) {
1727 WARN_ON(start
> offset
+ len
);
1729 end
= start
+ frag_iter
->len
;
1730 if ((copy
= end
- offset
) > 0) {
1734 csum2
= skb_checksum(frag_iter
, offset
- start
,
1736 csum
= csum_block_add(csum
, csum2
, pos
);
1737 if ((len
-= copy
) == 0)
1748 EXPORT_SYMBOL(skb_checksum
);
1750 /* Both of above in one bottle. */
1752 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
1753 u8
*to
, int len
, __wsum csum
)
1755 int start
= skb_headlen(skb
);
1756 int i
, copy
= start
- offset
;
1757 struct sk_buff
*frag_iter
;
1764 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
1766 if ((len
-= copy
) == 0)
1773 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1776 WARN_ON(start
> offset
+ len
);
1778 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1779 if ((copy
= end
- offset
) > 0) {
1782 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1786 vaddr
= kmap_skb_frag(frag
);
1787 csum2
= csum_partial_copy_nocheck(vaddr
+
1791 kunmap_skb_frag(vaddr
);
1792 csum
= csum_block_add(csum
, csum2
, pos
);
1802 skb_walk_frags(skb
, frag_iter
) {
1806 WARN_ON(start
> offset
+ len
);
1808 end
= start
+ frag_iter
->len
;
1809 if ((copy
= end
- offset
) > 0) {
1812 csum2
= skb_copy_and_csum_bits(frag_iter
,
1815 csum
= csum_block_add(csum
, csum2
, pos
);
1816 if ((len
-= copy
) == 0)
1827 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
1829 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
1834 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1835 csstart
= skb
->csum_start
- skb_headroom(skb
);
1837 csstart
= skb_headlen(skb
);
1839 BUG_ON(csstart
> skb_headlen(skb
));
1841 skb_copy_from_linear_data(skb
, to
, csstart
);
1844 if (csstart
!= skb
->len
)
1845 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
1846 skb
->len
- csstart
, 0);
1848 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1849 long csstuff
= csstart
+ skb
->csum_offset
;
1851 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
1854 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
1857 * skb_dequeue - remove from the head of the queue
1858 * @list: list to dequeue from
1860 * Remove the head of the list. The list lock is taken so the function
1861 * may be used safely with other locking list functions. The head item is
1862 * returned or %NULL if the list is empty.
1865 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
1867 unsigned long flags
;
1868 struct sk_buff
*result
;
1870 spin_lock_irqsave(&list
->lock
, flags
);
1871 result
= __skb_dequeue(list
);
1872 spin_unlock_irqrestore(&list
->lock
, flags
);
1875 EXPORT_SYMBOL(skb_dequeue
);
1878 * skb_dequeue_tail - remove from the tail of the queue
1879 * @list: list to dequeue from
1881 * Remove the tail of the list. The list lock is taken so the function
1882 * may be used safely with other locking list functions. The tail item is
1883 * returned or %NULL if the list is empty.
1885 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
1887 unsigned long flags
;
1888 struct sk_buff
*result
;
1890 spin_lock_irqsave(&list
->lock
, flags
);
1891 result
= __skb_dequeue_tail(list
);
1892 spin_unlock_irqrestore(&list
->lock
, flags
);
1895 EXPORT_SYMBOL(skb_dequeue_tail
);
1898 * skb_queue_purge - empty a list
1899 * @list: list to empty
1901 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1902 * the list and one reference dropped. This function takes the list
1903 * lock and is atomic with respect to other list locking functions.
1905 void skb_queue_purge(struct sk_buff_head
*list
)
1907 struct sk_buff
*skb
;
1908 while ((skb
= skb_dequeue(list
)) != NULL
)
1911 EXPORT_SYMBOL(skb_queue_purge
);
1914 * skb_queue_head - queue a buffer at the list head
1915 * @list: list to use
1916 * @newsk: buffer to queue
1918 * Queue a buffer at the start of the list. This function takes the
1919 * list lock and can be used safely with other locking &sk_buff functions
1922 * A buffer cannot be placed on two lists at the same time.
1924 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1926 unsigned long flags
;
1928 spin_lock_irqsave(&list
->lock
, flags
);
1929 __skb_queue_head(list
, newsk
);
1930 spin_unlock_irqrestore(&list
->lock
, flags
);
1932 EXPORT_SYMBOL(skb_queue_head
);
1935 * skb_queue_tail - queue a buffer at the list tail
1936 * @list: list to use
1937 * @newsk: buffer to queue
1939 * Queue a buffer at the tail of the list. This function takes the
1940 * list lock and can be used safely with other locking &sk_buff functions
1943 * A buffer cannot be placed on two lists at the same time.
1945 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1947 unsigned long flags
;
1949 spin_lock_irqsave(&list
->lock
, flags
);
1950 __skb_queue_tail(list
, newsk
);
1951 spin_unlock_irqrestore(&list
->lock
, flags
);
1953 EXPORT_SYMBOL(skb_queue_tail
);
1956 * skb_unlink - remove a buffer from a list
1957 * @skb: buffer to remove
1958 * @list: list to use
1960 * Remove a packet from a list. The list locks are taken and this
1961 * function is atomic with respect to other list locked calls
1963 * You must know what list the SKB is on.
1965 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
1967 unsigned long flags
;
1969 spin_lock_irqsave(&list
->lock
, flags
);
1970 __skb_unlink(skb
, list
);
1971 spin_unlock_irqrestore(&list
->lock
, flags
);
1973 EXPORT_SYMBOL(skb_unlink
);
1976 * skb_append - append a buffer
1977 * @old: buffer to insert after
1978 * @newsk: buffer to insert
1979 * @list: list to use
1981 * Place a packet after a given packet in a list. The list locks are taken
1982 * and this function is atomic with respect to other list locked calls.
1983 * A buffer cannot be placed on two lists at the same time.
1985 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
1987 unsigned long flags
;
1989 spin_lock_irqsave(&list
->lock
, flags
);
1990 __skb_queue_after(list
, old
, newsk
);
1991 spin_unlock_irqrestore(&list
->lock
, flags
);
1993 EXPORT_SYMBOL(skb_append
);
1996 * skb_insert - insert a buffer
1997 * @old: buffer to insert before
1998 * @newsk: buffer to insert
1999 * @list: list to use
2001 * Place a packet before a given packet in a list. The list locks are
2002 * taken and this function is atomic with respect to other list locked
2005 * A buffer cannot be placed on two lists at the same time.
2007 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2009 unsigned long flags
;
2011 spin_lock_irqsave(&list
->lock
, flags
);
2012 __skb_insert(newsk
, old
->prev
, old
, list
);
2013 spin_unlock_irqrestore(&list
->lock
, flags
);
2015 EXPORT_SYMBOL(skb_insert
);
2017 static inline void skb_split_inside_header(struct sk_buff
*skb
,
2018 struct sk_buff
* skb1
,
2019 const u32 len
, const int pos
)
2023 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
2025 /* And move data appendix as is. */
2026 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
2027 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
2029 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
2030 skb_shinfo(skb
)->nr_frags
= 0;
2031 skb1
->data_len
= skb
->data_len
;
2032 skb1
->len
+= skb1
->data_len
;
2035 skb_set_tail_pointer(skb
, len
);
2038 static inline void skb_split_no_header(struct sk_buff
*skb
,
2039 struct sk_buff
* skb1
,
2040 const u32 len
, int pos
)
2043 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
2045 skb_shinfo(skb
)->nr_frags
= 0;
2046 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
2048 skb
->data_len
= len
- pos
;
2050 for (i
= 0; i
< nfrags
; i
++) {
2051 int size
= skb_shinfo(skb
)->frags
[i
].size
;
2053 if (pos
+ size
> len
) {
2054 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
2058 * We have two variants in this case:
2059 * 1. Move all the frag to the second
2060 * part, if it is possible. F.e.
2061 * this approach is mandatory for TUX,
2062 * where splitting is expensive.
2063 * 2. Split is accurately. We make this.
2065 get_page(skb_shinfo(skb
)->frags
[i
].page
);
2066 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
2067 skb_shinfo(skb1
)->frags
[0].size
-= len
- pos
;
2068 skb_shinfo(skb
)->frags
[i
].size
= len
- pos
;
2069 skb_shinfo(skb
)->nr_frags
++;
2073 skb_shinfo(skb
)->nr_frags
++;
2076 skb_shinfo(skb1
)->nr_frags
= k
;
2080 * skb_split - Split fragmented skb to two parts at length len.
2081 * @skb: the buffer to split
2082 * @skb1: the buffer to receive the second part
2083 * @len: new length for skb
2085 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
2087 int pos
= skb_headlen(skb
);
2089 if (len
< pos
) /* Split line is inside header. */
2090 skb_split_inside_header(skb
, skb1
, len
, pos
);
2091 else /* Second chunk has no header, nothing to copy. */
2092 skb_split_no_header(skb
, skb1
, len
, pos
);
2094 EXPORT_SYMBOL(skb_split
);
2096 /* Shifting from/to a cloned skb is a no-go.
2098 * Caller cannot keep skb_shinfo related pointers past calling here!
2100 static int skb_prepare_for_shift(struct sk_buff
*skb
)
2102 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2106 * skb_shift - Shifts paged data partially from skb to another
2107 * @tgt: buffer into which tail data gets added
2108 * @skb: buffer from which the paged data comes from
2109 * @shiftlen: shift up to this many bytes
2111 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2112 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2113 * It's up to caller to free skb if everything was shifted.
2115 * If @tgt runs out of frags, the whole operation is aborted.
2117 * Skb cannot include anything else but paged data while tgt is allowed
2118 * to have non-paged data as well.
2120 * TODO: full sized shift could be optimized but that would need
2121 * specialized skb free'er to handle frags without up-to-date nr_frags.
2123 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
2125 int from
, to
, merge
, todo
;
2126 struct skb_frag_struct
*fragfrom
, *fragto
;
2128 BUG_ON(shiftlen
> skb
->len
);
2129 BUG_ON(skb_headlen(skb
)); /* Would corrupt stream */
2133 to
= skb_shinfo(tgt
)->nr_frags
;
2134 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2136 /* Actual merge is delayed until the point when we know we can
2137 * commit all, so that we don't have to undo partial changes
2140 !skb_can_coalesce(tgt
, to
, fragfrom
->page
, fragfrom
->page_offset
)) {
2145 todo
-= fragfrom
->size
;
2147 if (skb_prepare_for_shift(skb
) ||
2148 skb_prepare_for_shift(tgt
))
2151 /* All previous frag pointers might be stale! */
2152 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2153 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2155 fragto
->size
+= shiftlen
;
2156 fragfrom
->size
-= shiftlen
;
2157 fragfrom
->page_offset
+= shiftlen
;
2165 /* Skip full, not-fitting skb to avoid expensive operations */
2166 if ((shiftlen
== skb
->len
) &&
2167 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
2170 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
2173 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
2174 if (to
== MAX_SKB_FRAGS
)
2177 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2178 fragto
= &skb_shinfo(tgt
)->frags
[to
];
2180 if (todo
>= fragfrom
->size
) {
2181 *fragto
= *fragfrom
;
2182 todo
-= fragfrom
->size
;
2187 get_page(fragfrom
->page
);
2188 fragto
->page
= fragfrom
->page
;
2189 fragto
->page_offset
= fragfrom
->page_offset
;
2190 fragto
->size
= todo
;
2192 fragfrom
->page_offset
+= todo
;
2193 fragfrom
->size
-= todo
;
2201 /* Ready to "commit" this state change to tgt */
2202 skb_shinfo(tgt
)->nr_frags
= to
;
2205 fragfrom
= &skb_shinfo(skb
)->frags
[0];
2206 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2208 fragto
->size
+= fragfrom
->size
;
2209 put_page(fragfrom
->page
);
2212 /* Reposition in the original skb */
2214 while (from
< skb_shinfo(skb
)->nr_frags
)
2215 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
2216 skb_shinfo(skb
)->nr_frags
= to
;
2218 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
2221 /* Most likely the tgt won't ever need its checksum anymore, skb on
2222 * the other hand might need it if it needs to be resent
2224 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
2225 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2227 /* Yak, is it really working this way? Some helper please? */
2228 skb
->len
-= shiftlen
;
2229 skb
->data_len
-= shiftlen
;
2230 skb
->truesize
-= shiftlen
;
2231 tgt
->len
+= shiftlen
;
2232 tgt
->data_len
+= shiftlen
;
2233 tgt
->truesize
+= shiftlen
;
2239 * skb_prepare_seq_read - Prepare a sequential read of skb data
2240 * @skb: the buffer to read
2241 * @from: lower offset of data to be read
2242 * @to: upper offset of data to be read
2243 * @st: state variable
2245 * Initializes the specified state variable. Must be called before
2246 * invoking skb_seq_read() for the first time.
2248 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
2249 unsigned int to
, struct skb_seq_state
*st
)
2251 st
->lower_offset
= from
;
2252 st
->upper_offset
= to
;
2253 st
->root_skb
= st
->cur_skb
= skb
;
2254 st
->frag_idx
= st
->stepped_offset
= 0;
2255 st
->frag_data
= NULL
;
2257 EXPORT_SYMBOL(skb_prepare_seq_read
);
2260 * skb_seq_read - Sequentially read skb data
2261 * @consumed: number of bytes consumed by the caller so far
2262 * @data: destination pointer for data to be returned
2263 * @st: state variable
2265 * Reads a block of skb data at &consumed relative to the
2266 * lower offset specified to skb_prepare_seq_read(). Assigns
2267 * the head of the data block to &data and returns the length
2268 * of the block or 0 if the end of the skb data or the upper
2269 * offset has been reached.
2271 * The caller is not required to consume all of the data
2272 * returned, i.e. &consumed is typically set to the number
2273 * of bytes already consumed and the next call to
2274 * skb_seq_read() will return the remaining part of the block.
2276 * Note 1: The size of each block of data returned can be arbitary,
2277 * this limitation is the cost for zerocopy seqeuental
2278 * reads of potentially non linear data.
2280 * Note 2: Fragment lists within fragments are not implemented
2281 * at the moment, state->root_skb could be replaced with
2282 * a stack for this purpose.
2284 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
2285 struct skb_seq_state
*st
)
2287 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
2290 if (unlikely(abs_offset
>= st
->upper_offset
))
2294 block_limit
= skb_headlen(st
->cur_skb
) + st
->stepped_offset
;
2296 if (abs_offset
< block_limit
&& !st
->frag_data
) {
2297 *data
= st
->cur_skb
->data
+ (abs_offset
- st
->stepped_offset
);
2298 return block_limit
- abs_offset
;
2301 if (st
->frag_idx
== 0 && !st
->frag_data
)
2302 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
2304 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
2305 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
2306 block_limit
= frag
->size
+ st
->stepped_offset
;
2308 if (abs_offset
< block_limit
) {
2310 st
->frag_data
= kmap_skb_frag(frag
);
2312 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
2313 (abs_offset
- st
->stepped_offset
);
2315 return block_limit
- abs_offset
;
2318 if (st
->frag_data
) {
2319 kunmap_skb_frag(st
->frag_data
);
2320 st
->frag_data
= NULL
;
2324 st
->stepped_offset
+= frag
->size
;
2327 if (st
->frag_data
) {
2328 kunmap_skb_frag(st
->frag_data
);
2329 st
->frag_data
= NULL
;
2332 if (st
->root_skb
== st
->cur_skb
&& skb_has_frags(st
->root_skb
)) {
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 SKB_FRAG_ASSERT(nskb
);
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 struct skb_shared_info
*skbinfo
= skb_shinfo(skb
);
2669 struct skb_shared_info
*pinfo
= skb_shinfo(p
);
2670 unsigned int headroom
;
2671 unsigned int len
= skb_gro_len(skb
);
2672 unsigned int offset
= skb_gro_offset(skb
);
2673 unsigned int headlen
= skb_headlen(skb
);
2675 if (p
->len
+ len
>= 65536)
2678 if (pinfo
->frag_list
)
2680 else if (headlen
<= offset
) {
2683 int i
= skbinfo
->nr_frags
;
2684 int nr_frags
= pinfo
->nr_frags
+ i
;
2688 if (nr_frags
> MAX_SKB_FRAGS
)
2691 pinfo
->nr_frags
= nr_frags
;
2692 skbinfo
->nr_frags
= 0;
2694 frag
= pinfo
->frags
+ nr_frags
;
2695 frag2
= skbinfo
->frags
+ i
;
2700 frag
->page_offset
+= offset
;
2701 frag
->size
-= offset
;
2703 skb
->truesize
-= skb
->data_len
;
2704 skb
->len
-= skb
->data_len
;
2707 NAPI_GRO_CB(skb
)->free
= 1;
2709 } else if (skb_gro_len(p
) != pinfo
->gso_size
)
2712 headroom
= skb_headroom(p
);
2713 nskb
= netdev_alloc_skb(p
->dev
, headroom
+ skb_gro_offset(p
));
2714 if (unlikely(!nskb
))
2717 __copy_skb_header(nskb
, p
);
2718 nskb
->mac_len
= p
->mac_len
;
2720 skb_reserve(nskb
, headroom
);
2721 __skb_put(nskb
, skb_gro_offset(p
));
2723 skb_set_mac_header(nskb
, skb_mac_header(p
) - p
->data
);
2724 skb_set_network_header(nskb
, skb_network_offset(p
));
2725 skb_set_transport_header(nskb
, skb_transport_offset(p
));
2727 __skb_pull(p
, skb_gro_offset(p
));
2728 memcpy(skb_mac_header(nskb
), skb_mac_header(p
),
2729 p
->data
- skb_mac_header(p
));
2731 *NAPI_GRO_CB(nskb
) = *NAPI_GRO_CB(p
);
2732 skb_shinfo(nskb
)->frag_list
= p
;
2733 skb_shinfo(nskb
)->gso_size
= pinfo
->gso_size
;
2734 skb_header_release(p
);
2737 nskb
->data_len
+= p
->len
;
2738 nskb
->truesize
+= p
->len
;
2739 nskb
->len
+= p
->len
;
2742 nskb
->next
= p
->next
;
2748 if (offset
> headlen
) {
2749 skbinfo
->frags
[0].page_offset
+= offset
- headlen
;
2750 skbinfo
->frags
[0].size
-= offset
- headlen
;
2754 __skb_pull(skb
, offset
);
2756 p
->prev
->next
= skb
;
2758 skb_header_release(skb
);
2761 NAPI_GRO_CB(p
)->count
++;
2766 NAPI_GRO_CB(skb
)->same_flow
= 1;
2769 EXPORT_SYMBOL_GPL(skb_gro_receive
);
2771 void __init
skb_init(void)
2773 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
2774 sizeof(struct sk_buff
),
2776 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2778 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
2779 (2*sizeof(struct sk_buff
)) +
2782 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2787 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2788 * @skb: Socket buffer containing the buffers to be mapped
2789 * @sg: The scatter-gather list to map into
2790 * @offset: The offset into the buffer's contents to start mapping
2791 * @len: Length of buffer space to be mapped
2793 * Fill the specified scatter-gather list with mappings/pointers into a
2794 * region of the buffer space attached to a socket buffer.
2797 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2799 int start
= skb_headlen(skb
);
2800 int i
, copy
= start
- offset
;
2801 struct sk_buff
*frag_iter
;
2807 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
2809 if ((len
-= copy
) == 0)
2814 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2817 WARN_ON(start
> offset
+ len
);
2819 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
2820 if ((copy
= end
- offset
) > 0) {
2821 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2825 sg_set_page(&sg
[elt
], frag
->page
, copy
,
2826 frag
->page_offset
+offset
-start
);
2835 skb_walk_frags(skb
, frag_iter
) {
2838 WARN_ON(start
> offset
+ len
);
2840 end
= start
+ frag_iter
->len
;
2841 if ((copy
= end
- offset
) > 0) {
2844 elt
+= __skb_to_sgvec(frag_iter
, sg
+elt
, offset
- start
,
2846 if ((len
-= copy
) == 0)
2856 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2858 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
);
2860 sg_mark_end(&sg
[nsg
- 1]);
2864 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
2867 * skb_cow_data - Check that a socket buffer's data buffers are writable
2868 * @skb: The socket buffer to check.
2869 * @tailbits: Amount of trailing space to be added
2870 * @trailer: Returned pointer to the skb where the @tailbits space begins
2872 * Make sure that the data buffers attached to a socket buffer are
2873 * writable. If they are not, private copies are made of the data buffers
2874 * and the socket buffer is set to use these instead.
2876 * If @tailbits is given, make sure that there is space to write @tailbits
2877 * bytes of data beyond current end of socket buffer. @trailer will be
2878 * set to point to the skb in which this space begins.
2880 * The number of scatterlist elements required to completely map the
2881 * COW'd and extended socket buffer will be returned.
2883 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
2887 struct sk_buff
*skb1
, **skb_p
;
2889 /* If skb is cloned or its head is paged, reallocate
2890 * head pulling out all the pages (pages are considered not writable
2891 * at the moment even if they are anonymous).
2893 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
2894 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
2897 /* Easy case. Most of packets will go this way. */
2898 if (!skb_has_frags(skb
)) {
2899 /* A little of trouble, not enough of space for trailer.
2900 * This should not happen, when stack is tuned to generate
2901 * good frames. OK, on miss we reallocate and reserve even more
2902 * space, 128 bytes is fair. */
2904 if (skb_tailroom(skb
) < tailbits
&&
2905 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
2913 /* Misery. We are in troubles, going to mincer fragments... */
2916 skb_p
= &skb_shinfo(skb
)->frag_list
;
2919 while ((skb1
= *skb_p
) != NULL
) {
2922 /* The fragment is partially pulled by someone,
2923 * this can happen on input. Copy it and everything
2926 if (skb_shared(skb1
))
2929 /* If the skb is the last, worry about trailer. */
2931 if (skb1
->next
== NULL
&& tailbits
) {
2932 if (skb_shinfo(skb1
)->nr_frags
||
2933 skb_has_frags(skb1
) ||
2934 skb_tailroom(skb1
) < tailbits
)
2935 ntail
= tailbits
+ 128;
2941 skb_shinfo(skb1
)->nr_frags
||
2942 skb_has_frags(skb1
)) {
2943 struct sk_buff
*skb2
;
2945 /* Fuck, we are miserable poor guys... */
2947 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
2949 skb2
= skb_copy_expand(skb1
,
2953 if (unlikely(skb2
== NULL
))
2957 skb_set_owner_w(skb2
, skb1
->sk
);
2959 /* Looking around. Are we still alive?
2960 * OK, link new skb, drop old one */
2962 skb2
->next
= skb1
->next
;
2969 skb_p
= &skb1
->next
;
2974 EXPORT_SYMBOL_GPL(skb_cow_data
);
2976 void skb_tstamp_tx(struct sk_buff
*orig_skb
,
2977 struct skb_shared_hwtstamps
*hwtstamps
)
2979 struct sock
*sk
= orig_skb
->sk
;
2980 struct sock_exterr_skb
*serr
;
2981 struct sk_buff
*skb
;
2987 skb
= skb_clone(orig_skb
, GFP_ATOMIC
);
2992 *skb_hwtstamps(skb
) =
2996 * no hardware time stamps available,
2997 * so keep the skb_shared_tx and only
2998 * store software time stamp
3000 skb
->tstamp
= ktime_get_real();
3003 serr
= SKB_EXT_ERR(skb
);
3004 memset(serr
, 0, sizeof(*serr
));
3005 serr
->ee
.ee_errno
= ENOMSG
;
3006 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TIMESTAMPING
;
3007 err
= sock_queue_err_skb(sk
, skb
);
3011 EXPORT_SYMBOL_GPL(skb_tstamp_tx
);
3015 * skb_partial_csum_set - set up and verify partial csum values for packet
3016 * @skb: the skb to set
3017 * @start: the number of bytes after skb->data to start checksumming.
3018 * @off: the offset from start to place the checksum.
3020 * For untrusted partially-checksummed packets, we need to make sure the values
3021 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3023 * This function checks and sets those values and skb->ip_summed: if this
3024 * returns false you should drop the packet.
3026 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
3028 if (unlikely(start
> skb_headlen(skb
)) ||
3029 unlikely((int)start
+ off
> skb_headlen(skb
) - 2)) {
3030 if (net_ratelimit())
3032 "bad partial csum: csum=%u/%u len=%u\n",
3033 start
, off
, skb_headlen(skb
));
3036 skb
->ip_summed
= CHECKSUM_PARTIAL
;
3037 skb
->csum_start
= skb_headroom(skb
) + start
;
3038 skb
->csum_offset
= off
;
3041 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);
3043 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
3045 if (net_ratelimit())
3046 pr_warning("%s: received packets cannot be forwarded"
3047 " while LRO is enabled\n", skb
->dev
->name
);
3049 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);