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
)));
538 new->sp
= secpath_get(old
->sp
);
540 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
541 new->csum
= old
->csum
;
542 new->local_df
= old
->local_df
;
543 new->pkt_type
= old
->pkt_type
;
544 new->ip_summed
= old
->ip_summed
;
545 skb_copy_queue_mapping(new, old
);
546 new->priority
= old
->priority
;
547 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
548 new->ipvs_property
= old
->ipvs_property
;
550 new->protocol
= old
->protocol
;
551 new->mark
= old
->mark
;
552 new->skb_iif
= old
->skb_iif
;
554 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
555 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
556 new->nf_trace
= old
->nf_trace
;
558 #ifdef CONFIG_NET_SCHED
559 new->tc_index
= old
->tc_index
;
560 #ifdef CONFIG_NET_CLS_ACT
561 new->tc_verd
= old
->tc_verd
;
564 new->vlan_tci
= old
->vlan_tci
;
566 skb_copy_secmark(new, old
);
570 * You should not add any new code to this function. Add it to
571 * __copy_skb_header above instead.
573 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
575 #define C(x) n->x = skb->x
577 n
->next
= n
->prev
= NULL
;
579 __copy_skb_header(n
, skb
);
584 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
587 n
->destructor
= NULL
;
593 atomic_set(&n
->users
, 1);
595 atomic_inc(&(skb_shinfo(skb
)->dataref
));
603 * skb_morph - morph one skb into another
604 * @dst: the skb to receive the contents
605 * @src: the skb to supply the contents
607 * This is identical to skb_clone except that the target skb is
608 * supplied by the user.
610 * The target skb is returned upon exit.
612 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
614 skb_release_all(dst
);
615 return __skb_clone(dst
, src
);
617 EXPORT_SYMBOL_GPL(skb_morph
);
620 * skb_clone - duplicate an sk_buff
621 * @skb: buffer to clone
622 * @gfp_mask: allocation priority
624 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
625 * copies share the same packet data but not structure. The new
626 * buffer has a reference count of 1. If the allocation fails the
627 * function returns %NULL otherwise the new buffer is returned.
629 * If this function is called from an interrupt gfp_mask() must be
633 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
638 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
639 n
->fclone
== SKB_FCLONE_UNAVAILABLE
) {
640 atomic_t
*fclone_ref
= (atomic_t
*) (n
+ 1);
641 n
->fclone
= SKB_FCLONE_CLONE
;
642 atomic_inc(fclone_ref
);
644 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
648 kmemcheck_annotate_bitfield(n
, flags1
);
649 kmemcheck_annotate_bitfield(n
, flags2
);
650 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
653 return __skb_clone(n
, skb
);
655 EXPORT_SYMBOL(skb_clone
);
657 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
659 #ifndef NET_SKBUFF_DATA_USES_OFFSET
661 * Shift between the two data areas in bytes
663 unsigned long offset
= new->data
- old
->data
;
666 __copy_skb_header(new, old
);
668 #ifndef NET_SKBUFF_DATA_USES_OFFSET
669 /* {transport,network,mac}_header are relative to skb->head */
670 new->transport_header
+= offset
;
671 new->network_header
+= offset
;
672 if (skb_mac_header_was_set(new))
673 new->mac_header
+= offset
;
675 skb_shinfo(new)->gso_size
= skb_shinfo(old
)->gso_size
;
676 skb_shinfo(new)->gso_segs
= skb_shinfo(old
)->gso_segs
;
677 skb_shinfo(new)->gso_type
= skb_shinfo(old
)->gso_type
;
681 * skb_copy - create private copy of an sk_buff
682 * @skb: buffer to copy
683 * @gfp_mask: allocation priority
685 * Make a copy of both an &sk_buff and its data. This is used when the
686 * caller wishes to modify the data and needs a private copy of the
687 * data to alter. Returns %NULL on failure or the pointer to the buffer
688 * on success. The returned buffer has a reference count of 1.
690 * As by-product this function converts non-linear &sk_buff to linear
691 * one, so that &sk_buff becomes completely private and caller is allowed
692 * to modify all the data of returned buffer. This means that this
693 * function is not recommended for use in circumstances when only
694 * header is going to be modified. Use pskb_copy() instead.
697 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
699 int headerlen
= skb
->data
- skb
->head
;
701 * Allocate the copy buffer
704 #ifdef NET_SKBUFF_DATA_USES_OFFSET
705 n
= alloc_skb(skb
->end
+ skb
->data_len
, gfp_mask
);
707 n
= alloc_skb(skb
->end
- skb
->head
+ skb
->data_len
, gfp_mask
);
712 /* Set the data pointer */
713 skb_reserve(n
, headerlen
);
714 /* Set the tail pointer and length */
715 skb_put(n
, skb
->len
);
717 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
720 copy_skb_header(n
, skb
);
723 EXPORT_SYMBOL(skb_copy
);
726 * pskb_copy - create copy of an sk_buff with private head.
727 * @skb: buffer to copy
728 * @gfp_mask: allocation priority
730 * Make a copy of both an &sk_buff and part of its data, located
731 * in header. Fragmented data remain shared. This is used when
732 * the caller wishes to modify only header of &sk_buff and needs
733 * private copy of the header to alter. Returns %NULL on failure
734 * or the pointer to the buffer on success.
735 * The returned buffer has a reference count of 1.
738 struct sk_buff
*pskb_copy(struct sk_buff
*skb
, gfp_t gfp_mask
)
741 * Allocate the copy buffer
744 #ifdef NET_SKBUFF_DATA_USES_OFFSET
745 n
= alloc_skb(skb
->end
, gfp_mask
);
747 n
= alloc_skb(skb
->end
- skb
->head
, gfp_mask
);
752 /* Set the data pointer */
753 skb_reserve(n
, skb
->data
- skb
->head
);
754 /* Set the tail pointer and length */
755 skb_put(n
, skb_headlen(skb
));
757 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
759 n
->truesize
+= skb
->data_len
;
760 n
->data_len
= skb
->data_len
;
763 if (skb_shinfo(skb
)->nr_frags
) {
766 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
767 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
768 get_page(skb_shinfo(n
)->frags
[i
].page
);
770 skb_shinfo(n
)->nr_frags
= i
;
773 if (skb_has_frags(skb
)) {
774 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
775 skb_clone_fraglist(n
);
778 copy_skb_header(n
, skb
);
782 EXPORT_SYMBOL(pskb_copy
);
785 * pskb_expand_head - reallocate header of &sk_buff
786 * @skb: buffer to reallocate
787 * @nhead: room to add at head
788 * @ntail: room to add at tail
789 * @gfp_mask: allocation priority
791 * Expands (or creates identical copy, if &nhead and &ntail are zero)
792 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
793 * reference count of 1. Returns zero in the case of success or error,
794 * if expansion failed. In the last case, &sk_buff is not changed.
796 * All the pointers pointing into skb header may change and must be
797 * reloaded after call to this function.
800 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
805 #ifdef NET_SKBUFF_DATA_USES_OFFSET
806 int size
= nhead
+ skb
->end
+ ntail
;
808 int size
= nhead
+ (skb
->end
- skb
->head
) + ntail
;
817 size
= SKB_DATA_ALIGN(size
);
819 data
= kmalloc(size
+ sizeof(struct skb_shared_info
), gfp_mask
);
823 /* Copy only real data... and, alas, header. This should be
824 * optimized for the cases when header is void. */
825 #ifdef NET_SKBUFF_DATA_USES_OFFSET
826 memcpy(data
+ nhead
, skb
->head
, skb
->tail
);
828 memcpy(data
+ nhead
, skb
->head
, skb
->tail
- skb
->head
);
830 memcpy(data
+ size
, skb_end_pointer(skb
),
831 sizeof(struct skb_shared_info
));
833 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
834 get_page(skb_shinfo(skb
)->frags
[i
].page
);
836 if (skb_has_frags(skb
))
837 skb_clone_fraglist(skb
);
839 skb_release_data(skb
);
841 off
= (data
+ nhead
) - skb
->head
;
845 #ifdef NET_SKBUFF_DATA_USES_OFFSET
849 skb
->end
= skb
->head
+ size
;
851 /* {transport,network,mac}_header and tail are relative to skb->head */
853 skb
->transport_header
+= off
;
854 skb
->network_header
+= off
;
855 if (skb_mac_header_was_set(skb
))
856 skb
->mac_header
+= off
;
857 skb
->csum_start
+= nhead
;
861 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
867 EXPORT_SYMBOL(pskb_expand_head
);
869 /* Make private copy of skb with writable head and some headroom */
871 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
873 struct sk_buff
*skb2
;
874 int delta
= headroom
- skb_headroom(skb
);
877 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
879 skb2
= skb_clone(skb
, GFP_ATOMIC
);
880 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
888 EXPORT_SYMBOL(skb_realloc_headroom
);
891 * skb_copy_expand - copy and expand sk_buff
892 * @skb: buffer to copy
893 * @newheadroom: new free bytes at head
894 * @newtailroom: new free bytes at tail
895 * @gfp_mask: allocation priority
897 * Make a copy of both an &sk_buff and its data and while doing so
898 * allocate additional space.
900 * This is used when the caller wishes to modify the data and needs a
901 * private copy of the data to alter as well as more space for new fields.
902 * Returns %NULL on failure or the pointer to the buffer
903 * on success. The returned buffer has a reference count of 1.
905 * You must pass %GFP_ATOMIC as the allocation priority if this function
906 * is called from an interrupt.
908 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
909 int newheadroom
, int newtailroom
,
913 * Allocate the copy buffer
915 struct sk_buff
*n
= alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
917 int oldheadroom
= skb_headroom(skb
);
918 int head_copy_len
, head_copy_off
;
924 skb_reserve(n
, newheadroom
);
926 /* Set the tail pointer and length */
927 skb_put(n
, skb
->len
);
929 head_copy_len
= oldheadroom
;
931 if (newheadroom
<= head_copy_len
)
932 head_copy_len
= newheadroom
;
934 head_copy_off
= newheadroom
- head_copy_len
;
936 /* Copy the linear header and data. */
937 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
938 skb
->len
+ head_copy_len
))
941 copy_skb_header(n
, skb
);
943 off
= newheadroom
- oldheadroom
;
944 n
->csum_start
+= off
;
945 #ifdef NET_SKBUFF_DATA_USES_OFFSET
946 n
->transport_header
+= off
;
947 n
->network_header
+= off
;
948 if (skb_mac_header_was_set(skb
))
949 n
->mac_header
+= off
;
954 EXPORT_SYMBOL(skb_copy_expand
);
957 * skb_pad - zero pad the tail of an skb
958 * @skb: buffer to pad
961 * Ensure that a buffer is followed by a padding area that is zero
962 * filled. Used by network drivers which may DMA or transfer data
963 * beyond the buffer end onto the wire.
965 * May return error in out of memory cases. The skb is freed on error.
968 int skb_pad(struct sk_buff
*skb
, int pad
)
973 /* If the skbuff is non linear tailroom is always zero.. */
974 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
975 memset(skb
->data
+skb
->len
, 0, pad
);
979 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
980 if (likely(skb_cloned(skb
) || ntail
> 0)) {
981 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
986 /* FIXME: The use of this function with non-linear skb's really needs
989 err
= skb_linearize(skb
);
993 memset(skb
->data
+ skb
->len
, 0, pad
);
1000 EXPORT_SYMBOL(skb_pad
);
1003 * skb_put - add data to a buffer
1004 * @skb: buffer to use
1005 * @len: amount of data to add
1007 * This function extends the used data area of the buffer. If this would
1008 * exceed the total buffer size the kernel will panic. A pointer to the
1009 * first byte of the extra data is returned.
1011 unsigned char *skb_put(struct sk_buff
*skb
, unsigned int len
)
1013 unsigned char *tmp
= skb_tail_pointer(skb
);
1014 SKB_LINEAR_ASSERT(skb
);
1017 if (unlikely(skb
->tail
> skb
->end
))
1018 skb_over_panic(skb
, len
, __builtin_return_address(0));
1021 EXPORT_SYMBOL(skb_put
);
1024 * skb_push - add data to the start of a buffer
1025 * @skb: buffer to use
1026 * @len: amount of data to add
1028 * This function extends the used data area of the buffer at the buffer
1029 * start. If this would exceed the total buffer headroom the kernel will
1030 * panic. A pointer to the first byte of the extra data is returned.
1032 unsigned char *skb_push(struct sk_buff
*skb
, unsigned int len
)
1036 if (unlikely(skb
->data
<skb
->head
))
1037 skb_under_panic(skb
, len
, __builtin_return_address(0));
1040 EXPORT_SYMBOL(skb_push
);
1043 * skb_pull - remove data from the start of a buffer
1044 * @skb: buffer to use
1045 * @len: amount of data to remove
1047 * This function removes data from the start of a buffer, returning
1048 * the memory to the headroom. A pointer to the next data in the buffer
1049 * is returned. Once the data has been pulled future pushes will overwrite
1052 unsigned char *skb_pull(struct sk_buff
*skb
, unsigned int len
)
1054 return unlikely(len
> skb
->len
) ? NULL
: __skb_pull(skb
, len
);
1056 EXPORT_SYMBOL(skb_pull
);
1059 * skb_trim - remove end from a buffer
1060 * @skb: buffer to alter
1063 * Cut the length of a buffer down by removing data from the tail. If
1064 * the buffer is already under the length specified it is not modified.
1065 * The skb must be linear.
1067 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1070 __skb_trim(skb
, len
);
1072 EXPORT_SYMBOL(skb_trim
);
1074 /* Trims skb to length len. It can change skb pointers.
1077 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1079 struct sk_buff
**fragp
;
1080 struct sk_buff
*frag
;
1081 int offset
= skb_headlen(skb
);
1082 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1086 if (skb_cloned(skb
) &&
1087 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1094 for (; i
< nfrags
; i
++) {
1095 int end
= offset
+ skb_shinfo(skb
)->frags
[i
].size
;
1102 skb_shinfo(skb
)->frags
[i
++].size
= len
- offset
;
1105 skb_shinfo(skb
)->nr_frags
= i
;
1107 for (; i
< nfrags
; i
++)
1108 put_page(skb_shinfo(skb
)->frags
[i
].page
);
1110 if (skb_has_frags(skb
))
1111 skb_drop_fraglist(skb
);
1115 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1116 fragp
= &frag
->next
) {
1117 int end
= offset
+ frag
->len
;
1119 if (skb_shared(frag
)) {
1120 struct sk_buff
*nfrag
;
1122 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1123 if (unlikely(!nfrag
))
1126 nfrag
->next
= frag
->next
;
1138 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1142 skb_drop_list(&frag
->next
);
1147 if (len
> skb_headlen(skb
)) {
1148 skb
->data_len
-= skb
->len
- len
;
1153 skb_set_tail_pointer(skb
, len
);
1158 EXPORT_SYMBOL(___pskb_trim
);
1161 * __pskb_pull_tail - advance tail of skb header
1162 * @skb: buffer to reallocate
1163 * @delta: number of bytes to advance tail
1165 * The function makes a sense only on a fragmented &sk_buff,
1166 * it expands header moving its tail forward and copying necessary
1167 * data from fragmented part.
1169 * &sk_buff MUST have reference count of 1.
1171 * Returns %NULL (and &sk_buff does not change) if pull failed
1172 * or value of new tail of skb in the case of success.
1174 * All the pointers pointing into skb header may change and must be
1175 * reloaded after call to this function.
1178 /* Moves tail of skb head forward, copying data from fragmented part,
1179 * when it is necessary.
1180 * 1. It may fail due to malloc failure.
1181 * 2. It may change skb pointers.
1183 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1185 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1187 /* If skb has not enough free space at tail, get new one
1188 * plus 128 bytes for future expansions. If we have enough
1189 * room at tail, reallocate without expansion only if skb is cloned.
1191 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1193 if (eat
> 0 || skb_cloned(skb
)) {
1194 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1199 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
1202 /* Optimization: no fragments, no reasons to preestimate
1203 * size of pulled pages. Superb.
1205 if (!skb_has_frags(skb
))
1208 /* Estimate size of pulled pages. */
1210 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1211 if (skb_shinfo(skb
)->frags
[i
].size
>= eat
)
1213 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1216 /* If we need update frag list, we are in troubles.
1217 * Certainly, it possible to add an offset to skb data,
1218 * but taking into account that pulling is expected to
1219 * be very rare operation, it is worth to fight against
1220 * further bloating skb head and crucify ourselves here instead.
1221 * Pure masohism, indeed. 8)8)
1224 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1225 struct sk_buff
*clone
= NULL
;
1226 struct sk_buff
*insp
= NULL
;
1231 if (list
->len
<= eat
) {
1232 /* Eaten as whole. */
1237 /* Eaten partially. */
1239 if (skb_shared(list
)) {
1240 /* Sucks! We need to fork list. :-( */
1241 clone
= skb_clone(list
, GFP_ATOMIC
);
1247 /* This may be pulled without
1251 if (!pskb_pull(list
, eat
)) {
1259 /* Free pulled out fragments. */
1260 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1261 skb_shinfo(skb
)->frag_list
= list
->next
;
1264 /* And insert new clone at head. */
1267 skb_shinfo(skb
)->frag_list
= clone
;
1270 /* Success! Now we may commit changes to skb data. */
1275 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1276 if (skb_shinfo(skb
)->frags
[i
].size
<= eat
) {
1277 put_page(skb_shinfo(skb
)->frags
[i
].page
);
1278 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1280 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1282 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1283 skb_shinfo(skb
)->frags
[k
].size
-= eat
;
1289 skb_shinfo(skb
)->nr_frags
= k
;
1292 skb
->data_len
-= delta
;
1294 return skb_tail_pointer(skb
);
1296 EXPORT_SYMBOL(__pskb_pull_tail
);
1298 /* Copy some data bits from skb to kernel buffer. */
1300 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1302 int start
= skb_headlen(skb
);
1303 struct sk_buff
*frag_iter
;
1306 if (offset
> (int)skb
->len
- len
)
1310 if ((copy
= start
- offset
) > 0) {
1313 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
1314 if ((len
-= copy
) == 0)
1320 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1323 WARN_ON(start
> offset
+ len
);
1325 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1326 if ((copy
= end
- offset
) > 0) {
1332 vaddr
= kmap_skb_frag(&skb_shinfo(skb
)->frags
[i
]);
1334 vaddr
+ skb_shinfo(skb
)->frags
[i
].page_offset
+
1335 offset
- start
, copy
);
1336 kunmap_skb_frag(vaddr
);
1338 if ((len
-= copy
) == 0)
1346 skb_walk_frags(skb
, frag_iter
) {
1349 WARN_ON(start
> offset
+ len
);
1351 end
= start
+ frag_iter
->len
;
1352 if ((copy
= end
- offset
) > 0) {
1355 if (skb_copy_bits(frag_iter
, offset
- start
, to
, copy
))
1357 if ((len
-= copy
) == 0)
1370 EXPORT_SYMBOL(skb_copy_bits
);
1373 * Callback from splice_to_pipe(), if we need to release some pages
1374 * at the end of the spd in case we error'ed out in filling the pipe.
1376 static void sock_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
1378 put_page(spd
->pages
[i
]);
1381 static inline struct page
*linear_to_page(struct page
*page
, unsigned int *len
,
1382 unsigned int *offset
,
1383 struct sk_buff
*skb
, struct sock
*sk
)
1385 struct page
*p
= sk
->sk_sndmsg_page
;
1390 p
= sk
->sk_sndmsg_page
= alloc_pages(sk
->sk_allocation
, 0);
1394 off
= sk
->sk_sndmsg_off
= 0;
1395 /* hold one ref to this page until it's full */
1399 off
= sk
->sk_sndmsg_off
;
1400 mlen
= PAGE_SIZE
- off
;
1401 if (mlen
< 64 && mlen
< *len
) {
1406 *len
= min_t(unsigned int, *len
, mlen
);
1409 memcpy(page_address(p
) + off
, page_address(page
) + *offset
, *len
);
1410 sk
->sk_sndmsg_off
+= *len
;
1418 * Fill page/offset/length into spd, if it can hold more pages.
1420 static inline int spd_fill_page(struct splice_pipe_desc
*spd
,
1421 struct pipe_inode_info
*pipe
, struct page
*page
,
1422 unsigned int *len
, unsigned int offset
,
1423 struct sk_buff
*skb
, int linear
,
1426 if (unlikely(spd
->nr_pages
== pipe
->buffers
))
1430 page
= linear_to_page(page
, len
, &offset
, skb
, sk
);
1436 spd
->pages
[spd
->nr_pages
] = page
;
1437 spd
->partial
[spd
->nr_pages
].len
= *len
;
1438 spd
->partial
[spd
->nr_pages
].offset
= offset
;
1444 static inline void __segment_seek(struct page
**page
, unsigned int *poff
,
1445 unsigned int *plen
, unsigned int off
)
1450 n
= *poff
/ PAGE_SIZE
;
1452 *page
= nth_page(*page
, n
);
1454 *poff
= *poff
% PAGE_SIZE
;
1458 static inline int __splice_segment(struct page
*page
, unsigned int poff
,
1459 unsigned int plen
, unsigned int *off
,
1460 unsigned int *len
, struct sk_buff
*skb
,
1461 struct splice_pipe_desc
*spd
, int linear
,
1463 struct pipe_inode_info
*pipe
)
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
, pipe
, 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
, struct pipe_inode_info
*pipe
,
1502 unsigned int *offset
, unsigned int *len
,
1503 struct splice_pipe_desc
*spd
, struct sock
*sk
)
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
, pipe
))
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
, pipe
))
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_DEF_BUFFERS
];
1541 struct page
*pages
[PIPE_DEF_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 if (splice_grow_spd(pipe
, &spd
))
1557 * __skb_splice_bits() only fails if the output has no room left,
1558 * so no point in going over the frag_list for the error case.
1560 if (__skb_splice_bits(skb
, pipe
, &offset
, &tlen
, &spd
, sk
))
1566 * now see if we have a frag_list to map
1568 skb_walk_frags(skb
, frag_iter
) {
1571 if (__skb_splice_bits(frag_iter
, pipe
, &offset
, &tlen
, &spd
, sk
))
1578 * Drop the socket lock, otherwise we have reverse
1579 * locking dependencies between sk_lock and i_mutex
1580 * here as compared to sendfile(). We enter here
1581 * with the socket lock held, and splice_to_pipe() will
1582 * grab the pipe inode lock. For sendfile() emulation,
1583 * we call into ->sendpage() with the i_mutex lock held
1584 * and networking will grab the socket lock.
1587 ret
= splice_to_pipe(pipe
, &spd
);
1591 splice_shrink_spd(pipe
, &spd
);
1596 * skb_store_bits - store bits from kernel buffer to skb
1597 * @skb: destination buffer
1598 * @offset: offset in destination
1599 * @from: source buffer
1600 * @len: number of bytes to copy
1602 * Copy the specified number of bytes from the source buffer to the
1603 * destination skb. This function handles all the messy bits of
1604 * traversing fragment lists and such.
1607 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
1609 int start
= skb_headlen(skb
);
1610 struct sk_buff
*frag_iter
;
1613 if (offset
> (int)skb
->len
- len
)
1616 if ((copy
= start
- offset
) > 0) {
1619 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
1620 if ((len
-= copy
) == 0)
1626 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1627 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1630 WARN_ON(start
> offset
+ len
);
1632 end
= start
+ frag
->size
;
1633 if ((copy
= end
- offset
) > 0) {
1639 vaddr
= kmap_skb_frag(frag
);
1640 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1642 kunmap_skb_frag(vaddr
);
1644 if ((len
-= copy
) == 0)
1652 skb_walk_frags(skb
, frag_iter
) {
1655 WARN_ON(start
> offset
+ len
);
1657 end
= start
+ frag_iter
->len
;
1658 if ((copy
= end
- offset
) > 0) {
1661 if (skb_store_bits(frag_iter
, offset
- start
,
1664 if ((len
-= copy
) == 0)
1677 EXPORT_SYMBOL(skb_store_bits
);
1679 /* Checksum skb data. */
1681 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
1682 int len
, __wsum csum
)
1684 int start
= skb_headlen(skb
);
1685 int i
, copy
= start
- offset
;
1686 struct sk_buff
*frag_iter
;
1689 /* Checksum header. */
1693 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1694 if ((len
-= copy
) == 0)
1700 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1703 WARN_ON(start
> offset
+ len
);
1705 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1706 if ((copy
= end
- offset
) > 0) {
1709 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1713 vaddr
= kmap_skb_frag(frag
);
1714 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
1715 offset
- start
, copy
, 0);
1716 kunmap_skb_frag(vaddr
);
1717 csum
= csum_block_add(csum
, csum2
, pos
);
1726 skb_walk_frags(skb
, frag_iter
) {
1729 WARN_ON(start
> offset
+ len
);
1731 end
= start
+ frag_iter
->len
;
1732 if ((copy
= end
- offset
) > 0) {
1736 csum2
= skb_checksum(frag_iter
, offset
- start
,
1738 csum
= csum_block_add(csum
, csum2
, pos
);
1739 if ((len
-= copy
) == 0)
1750 EXPORT_SYMBOL(skb_checksum
);
1752 /* Both of above in one bottle. */
1754 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
1755 u8
*to
, int len
, __wsum csum
)
1757 int start
= skb_headlen(skb
);
1758 int i
, copy
= start
- offset
;
1759 struct sk_buff
*frag_iter
;
1766 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
1768 if ((len
-= copy
) == 0)
1775 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1778 WARN_ON(start
> offset
+ len
);
1780 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1781 if ((copy
= end
- offset
) > 0) {
1784 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1788 vaddr
= kmap_skb_frag(frag
);
1789 csum2
= csum_partial_copy_nocheck(vaddr
+
1793 kunmap_skb_frag(vaddr
);
1794 csum
= csum_block_add(csum
, csum2
, pos
);
1804 skb_walk_frags(skb
, frag_iter
) {
1808 WARN_ON(start
> offset
+ len
);
1810 end
= start
+ frag_iter
->len
;
1811 if ((copy
= end
- offset
) > 0) {
1814 csum2
= skb_copy_and_csum_bits(frag_iter
,
1817 csum
= csum_block_add(csum
, csum2
, pos
);
1818 if ((len
-= copy
) == 0)
1829 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
1831 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
1836 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1837 csstart
= skb
->csum_start
- skb_headroom(skb
);
1839 csstart
= skb_headlen(skb
);
1841 BUG_ON(csstart
> skb_headlen(skb
));
1843 skb_copy_from_linear_data(skb
, to
, csstart
);
1846 if (csstart
!= skb
->len
)
1847 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
1848 skb
->len
- csstart
, 0);
1850 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1851 long csstuff
= csstart
+ skb
->csum_offset
;
1853 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
1856 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
1859 * skb_dequeue - remove from the head of the queue
1860 * @list: list to dequeue from
1862 * Remove the head of the list. The list lock is taken so the function
1863 * may be used safely with other locking list functions. The head item is
1864 * returned or %NULL if the list is empty.
1867 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
1869 unsigned long flags
;
1870 struct sk_buff
*result
;
1872 spin_lock_irqsave(&list
->lock
, flags
);
1873 result
= __skb_dequeue(list
);
1874 spin_unlock_irqrestore(&list
->lock
, flags
);
1877 EXPORT_SYMBOL(skb_dequeue
);
1880 * skb_dequeue_tail - remove from the tail of the queue
1881 * @list: list to dequeue from
1883 * Remove the tail of the list. The list lock is taken so the function
1884 * may be used safely with other locking list functions. The tail item is
1885 * returned or %NULL if the list is empty.
1887 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
1889 unsigned long flags
;
1890 struct sk_buff
*result
;
1892 spin_lock_irqsave(&list
->lock
, flags
);
1893 result
= __skb_dequeue_tail(list
);
1894 spin_unlock_irqrestore(&list
->lock
, flags
);
1897 EXPORT_SYMBOL(skb_dequeue_tail
);
1900 * skb_queue_purge - empty a list
1901 * @list: list to empty
1903 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1904 * the list and one reference dropped. This function takes the list
1905 * lock and is atomic with respect to other list locking functions.
1907 void skb_queue_purge(struct sk_buff_head
*list
)
1909 struct sk_buff
*skb
;
1910 while ((skb
= skb_dequeue(list
)) != NULL
)
1913 EXPORT_SYMBOL(skb_queue_purge
);
1916 * skb_queue_head - queue a buffer at the list head
1917 * @list: list to use
1918 * @newsk: buffer to queue
1920 * Queue a buffer at the start of the list. This function takes the
1921 * list lock and can be used safely with other locking &sk_buff functions
1924 * A buffer cannot be placed on two lists at the same time.
1926 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1928 unsigned long flags
;
1930 spin_lock_irqsave(&list
->lock
, flags
);
1931 __skb_queue_head(list
, newsk
);
1932 spin_unlock_irqrestore(&list
->lock
, flags
);
1934 EXPORT_SYMBOL(skb_queue_head
);
1937 * skb_queue_tail - queue a buffer at the list tail
1938 * @list: list to use
1939 * @newsk: buffer to queue
1941 * Queue a buffer at the tail of the list. This function takes the
1942 * list lock and can be used safely with other locking &sk_buff functions
1945 * A buffer cannot be placed on two lists at the same time.
1947 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1949 unsigned long flags
;
1951 spin_lock_irqsave(&list
->lock
, flags
);
1952 __skb_queue_tail(list
, newsk
);
1953 spin_unlock_irqrestore(&list
->lock
, flags
);
1955 EXPORT_SYMBOL(skb_queue_tail
);
1958 * skb_unlink - remove a buffer from a list
1959 * @skb: buffer to remove
1960 * @list: list to use
1962 * Remove a packet from a list. The list locks are taken and this
1963 * function is atomic with respect to other list locked calls
1965 * You must know what list the SKB is on.
1967 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
1969 unsigned long flags
;
1971 spin_lock_irqsave(&list
->lock
, flags
);
1972 __skb_unlink(skb
, list
);
1973 spin_unlock_irqrestore(&list
->lock
, flags
);
1975 EXPORT_SYMBOL(skb_unlink
);
1978 * skb_append - append a buffer
1979 * @old: buffer to insert after
1980 * @newsk: buffer to insert
1981 * @list: list to use
1983 * Place a packet after a given packet in a list. The list locks are taken
1984 * and this function is atomic with respect to other list locked calls.
1985 * A buffer cannot be placed on two lists at the same time.
1987 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
1989 unsigned long flags
;
1991 spin_lock_irqsave(&list
->lock
, flags
);
1992 __skb_queue_after(list
, old
, newsk
);
1993 spin_unlock_irqrestore(&list
->lock
, flags
);
1995 EXPORT_SYMBOL(skb_append
);
1998 * skb_insert - insert a buffer
1999 * @old: buffer to insert before
2000 * @newsk: buffer to insert
2001 * @list: list to use
2003 * Place a packet before a given packet in a list. The list locks are
2004 * taken and this function is atomic with respect to other list locked
2007 * A buffer cannot be placed on two lists at the same time.
2009 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2011 unsigned long flags
;
2013 spin_lock_irqsave(&list
->lock
, flags
);
2014 __skb_insert(newsk
, old
->prev
, old
, list
);
2015 spin_unlock_irqrestore(&list
->lock
, flags
);
2017 EXPORT_SYMBOL(skb_insert
);
2019 static inline void skb_split_inside_header(struct sk_buff
*skb
,
2020 struct sk_buff
* skb1
,
2021 const u32 len
, const int pos
)
2025 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
2027 /* And move data appendix as is. */
2028 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
2029 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
2031 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
2032 skb_shinfo(skb
)->nr_frags
= 0;
2033 skb1
->data_len
= skb
->data_len
;
2034 skb1
->len
+= skb1
->data_len
;
2037 skb_set_tail_pointer(skb
, len
);
2040 static inline void skb_split_no_header(struct sk_buff
*skb
,
2041 struct sk_buff
* skb1
,
2042 const u32 len
, int pos
)
2045 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
2047 skb_shinfo(skb
)->nr_frags
= 0;
2048 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
2050 skb
->data_len
= len
- pos
;
2052 for (i
= 0; i
< nfrags
; i
++) {
2053 int size
= skb_shinfo(skb
)->frags
[i
].size
;
2055 if (pos
+ size
> len
) {
2056 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
2060 * We have two variants in this case:
2061 * 1. Move all the frag to the second
2062 * part, if it is possible. F.e.
2063 * this approach is mandatory for TUX,
2064 * where splitting is expensive.
2065 * 2. Split is accurately. We make this.
2067 get_page(skb_shinfo(skb
)->frags
[i
].page
);
2068 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
2069 skb_shinfo(skb1
)->frags
[0].size
-= len
- pos
;
2070 skb_shinfo(skb
)->frags
[i
].size
= len
- pos
;
2071 skb_shinfo(skb
)->nr_frags
++;
2075 skb_shinfo(skb
)->nr_frags
++;
2078 skb_shinfo(skb1
)->nr_frags
= k
;
2082 * skb_split - Split fragmented skb to two parts at length len.
2083 * @skb: the buffer to split
2084 * @skb1: the buffer to receive the second part
2085 * @len: new length for skb
2087 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
2089 int pos
= skb_headlen(skb
);
2091 if (len
< pos
) /* Split line is inside header. */
2092 skb_split_inside_header(skb
, skb1
, len
, pos
);
2093 else /* Second chunk has no header, nothing to copy. */
2094 skb_split_no_header(skb
, skb1
, len
, pos
);
2096 EXPORT_SYMBOL(skb_split
);
2098 /* Shifting from/to a cloned skb is a no-go.
2100 * Caller cannot keep skb_shinfo related pointers past calling here!
2102 static int skb_prepare_for_shift(struct sk_buff
*skb
)
2104 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2108 * skb_shift - Shifts paged data partially from skb to another
2109 * @tgt: buffer into which tail data gets added
2110 * @skb: buffer from which the paged data comes from
2111 * @shiftlen: shift up to this many bytes
2113 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2114 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2115 * It's up to caller to free skb if everything was shifted.
2117 * If @tgt runs out of frags, the whole operation is aborted.
2119 * Skb cannot include anything else but paged data while tgt is allowed
2120 * to have non-paged data as well.
2122 * TODO: full sized shift could be optimized but that would need
2123 * specialized skb free'er to handle frags without up-to-date nr_frags.
2125 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
2127 int from
, to
, merge
, todo
;
2128 struct skb_frag_struct
*fragfrom
, *fragto
;
2130 BUG_ON(shiftlen
> skb
->len
);
2131 BUG_ON(skb_headlen(skb
)); /* Would corrupt stream */
2135 to
= skb_shinfo(tgt
)->nr_frags
;
2136 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2138 /* Actual merge is delayed until the point when we know we can
2139 * commit all, so that we don't have to undo partial changes
2142 !skb_can_coalesce(tgt
, to
, fragfrom
->page
, fragfrom
->page_offset
)) {
2147 todo
-= fragfrom
->size
;
2149 if (skb_prepare_for_shift(skb
) ||
2150 skb_prepare_for_shift(tgt
))
2153 /* All previous frag pointers might be stale! */
2154 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2155 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2157 fragto
->size
+= shiftlen
;
2158 fragfrom
->size
-= shiftlen
;
2159 fragfrom
->page_offset
+= shiftlen
;
2167 /* Skip full, not-fitting skb to avoid expensive operations */
2168 if ((shiftlen
== skb
->len
) &&
2169 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
2172 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
2175 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
2176 if (to
== MAX_SKB_FRAGS
)
2179 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2180 fragto
= &skb_shinfo(tgt
)->frags
[to
];
2182 if (todo
>= fragfrom
->size
) {
2183 *fragto
= *fragfrom
;
2184 todo
-= fragfrom
->size
;
2189 get_page(fragfrom
->page
);
2190 fragto
->page
= fragfrom
->page
;
2191 fragto
->page_offset
= fragfrom
->page_offset
;
2192 fragto
->size
= todo
;
2194 fragfrom
->page_offset
+= todo
;
2195 fragfrom
->size
-= todo
;
2203 /* Ready to "commit" this state change to tgt */
2204 skb_shinfo(tgt
)->nr_frags
= to
;
2207 fragfrom
= &skb_shinfo(skb
)->frags
[0];
2208 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2210 fragto
->size
+= fragfrom
->size
;
2211 put_page(fragfrom
->page
);
2214 /* Reposition in the original skb */
2216 while (from
< skb_shinfo(skb
)->nr_frags
)
2217 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
2218 skb_shinfo(skb
)->nr_frags
= to
;
2220 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
2223 /* Most likely the tgt won't ever need its checksum anymore, skb on
2224 * the other hand might need it if it needs to be resent
2226 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
2227 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2229 /* Yak, is it really working this way? Some helper please? */
2230 skb
->len
-= shiftlen
;
2231 skb
->data_len
-= shiftlen
;
2232 skb
->truesize
-= shiftlen
;
2233 tgt
->len
+= shiftlen
;
2234 tgt
->data_len
+= shiftlen
;
2235 tgt
->truesize
+= shiftlen
;
2241 * skb_prepare_seq_read - Prepare a sequential read of skb data
2242 * @skb: the buffer to read
2243 * @from: lower offset of data to be read
2244 * @to: upper offset of data to be read
2245 * @st: state variable
2247 * Initializes the specified state variable. Must be called before
2248 * invoking skb_seq_read() for the first time.
2250 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
2251 unsigned int to
, struct skb_seq_state
*st
)
2253 st
->lower_offset
= from
;
2254 st
->upper_offset
= to
;
2255 st
->root_skb
= st
->cur_skb
= skb
;
2256 st
->frag_idx
= st
->stepped_offset
= 0;
2257 st
->frag_data
= NULL
;
2259 EXPORT_SYMBOL(skb_prepare_seq_read
);
2262 * skb_seq_read - Sequentially read skb data
2263 * @consumed: number of bytes consumed by the caller so far
2264 * @data: destination pointer for data to be returned
2265 * @st: state variable
2267 * Reads a block of skb data at &consumed relative to the
2268 * lower offset specified to skb_prepare_seq_read(). Assigns
2269 * the head of the data block to &data and returns the length
2270 * of the block or 0 if the end of the skb data or the upper
2271 * offset has been reached.
2273 * The caller is not required to consume all of the data
2274 * returned, i.e. &consumed is typically set to the number
2275 * of bytes already consumed and the next call to
2276 * skb_seq_read() will return the remaining part of the block.
2278 * Note 1: The size of each block of data returned can be arbitary,
2279 * this limitation is the cost for zerocopy seqeuental
2280 * reads of potentially non linear data.
2282 * Note 2: Fragment lists within fragments are not implemented
2283 * at the moment, state->root_skb could be replaced with
2284 * a stack for this purpose.
2286 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
2287 struct skb_seq_state
*st
)
2289 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
2292 if (unlikely(abs_offset
>= st
->upper_offset
))
2296 block_limit
= skb_headlen(st
->cur_skb
) + st
->stepped_offset
;
2298 if (abs_offset
< block_limit
&& !st
->frag_data
) {
2299 *data
= st
->cur_skb
->data
+ (abs_offset
- st
->stepped_offset
);
2300 return block_limit
- abs_offset
;
2303 if (st
->frag_idx
== 0 && !st
->frag_data
)
2304 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
2306 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
2307 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
2308 block_limit
= frag
->size
+ st
->stepped_offset
;
2310 if (abs_offset
< block_limit
) {
2312 st
->frag_data
= kmap_skb_frag(frag
);
2314 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
2315 (abs_offset
- st
->stepped_offset
);
2317 return block_limit
- abs_offset
;
2320 if (st
->frag_data
) {
2321 kunmap_skb_frag(st
->frag_data
);
2322 st
->frag_data
= NULL
;
2326 st
->stepped_offset
+= frag
->size
;
2329 if (st
->frag_data
) {
2330 kunmap_skb_frag(st
->frag_data
);
2331 st
->frag_data
= NULL
;
2334 if (st
->root_skb
== st
->cur_skb
&& skb_has_frags(st
->root_skb
)) {
2335 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
2338 } else if (st
->cur_skb
->next
) {
2339 st
->cur_skb
= st
->cur_skb
->next
;
2346 EXPORT_SYMBOL(skb_seq_read
);
2349 * skb_abort_seq_read - Abort a sequential read of skb data
2350 * @st: state variable
2352 * Must be called if skb_seq_read() was not called until it
2355 void skb_abort_seq_read(struct skb_seq_state
*st
)
2358 kunmap_skb_frag(st
->frag_data
);
2360 EXPORT_SYMBOL(skb_abort_seq_read
);
2362 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2364 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
2365 struct ts_config
*conf
,
2366 struct ts_state
*state
)
2368 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
2371 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
2373 skb_abort_seq_read(TS_SKB_CB(state
));
2377 * skb_find_text - Find a text pattern in skb data
2378 * @skb: the buffer to look in
2379 * @from: search offset
2381 * @config: textsearch configuration
2382 * @state: uninitialized textsearch state variable
2384 * Finds a pattern in the skb data according to the specified
2385 * textsearch configuration. Use textsearch_next() to retrieve
2386 * subsequent occurrences of the pattern. Returns the offset
2387 * to the first occurrence or UINT_MAX if no match was found.
2389 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
2390 unsigned int to
, struct ts_config
*config
,
2391 struct ts_state
*state
)
2395 config
->get_next_block
= skb_ts_get_next_block
;
2396 config
->finish
= skb_ts_finish
;
2398 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
2400 ret
= textsearch_find(config
, state
);
2401 return (ret
<= to
- from
? ret
: UINT_MAX
);
2403 EXPORT_SYMBOL(skb_find_text
);
2406 * skb_append_datato_frags: - append the user data to a skb
2407 * @sk: sock structure
2408 * @skb: skb structure to be appened with user data.
2409 * @getfrag: call back function to be used for getting the user data
2410 * @from: pointer to user message iov
2411 * @length: length of the iov message
2413 * Description: This procedure append the user data in the fragment part
2414 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2416 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
2417 int (*getfrag
)(void *from
, char *to
, int offset
,
2418 int len
, int odd
, struct sk_buff
*skb
),
2419 void *from
, int length
)
2422 skb_frag_t
*frag
= NULL
;
2423 struct page
*page
= NULL
;
2429 /* Return error if we don't have space for new frag */
2430 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2431 if (frg_cnt
>= MAX_SKB_FRAGS
)
2434 /* allocate a new page for next frag */
2435 page
= alloc_pages(sk
->sk_allocation
, 0);
2437 /* If alloc_page fails just return failure and caller will
2438 * free previous allocated pages by doing kfree_skb()
2443 /* initialize the next frag */
2444 sk
->sk_sndmsg_page
= page
;
2445 sk
->sk_sndmsg_off
= 0;
2446 skb_fill_page_desc(skb
, frg_cnt
, page
, 0, 0);
2447 skb
->truesize
+= PAGE_SIZE
;
2448 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
2450 /* get the new initialized frag */
2451 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2452 frag
= &skb_shinfo(skb
)->frags
[frg_cnt
- 1];
2454 /* copy the user data to page */
2455 left
= PAGE_SIZE
- frag
->page_offset
;
2456 copy
= (length
> left
)? left
: length
;
2458 ret
= getfrag(from
, (page_address(frag
->page
) +
2459 frag
->page_offset
+ frag
->size
),
2460 offset
, copy
, 0, skb
);
2464 /* copy was successful so update the size parameters */
2465 sk
->sk_sndmsg_off
+= copy
;
2468 skb
->data_len
+= copy
;
2472 } while (length
> 0);
2476 EXPORT_SYMBOL(skb_append_datato_frags
);
2479 * skb_pull_rcsum - pull skb and update receive checksum
2480 * @skb: buffer to update
2481 * @len: length of data pulled
2483 * This function performs an skb_pull on the packet and updates
2484 * the CHECKSUM_COMPLETE checksum. It should be used on
2485 * receive path processing instead of skb_pull unless you know
2486 * that the checksum difference is zero (e.g., a valid IP header)
2487 * or you are setting ip_summed to CHECKSUM_NONE.
2489 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
2491 BUG_ON(len
> skb
->len
);
2493 BUG_ON(skb
->len
< skb
->data_len
);
2494 skb_postpull_rcsum(skb
, skb
->data
, len
);
2495 return skb
->data
+= len
;
2498 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
2501 * skb_segment - Perform protocol segmentation on skb.
2502 * @skb: buffer to segment
2503 * @features: features for the output path (see dev->features)
2505 * This function performs segmentation on the given skb. It returns
2506 * a pointer to the first in a list of new skbs for the segments.
2507 * In case of error it returns ERR_PTR(err).
2509 struct sk_buff
*skb_segment(struct sk_buff
*skb
, int features
)
2511 struct sk_buff
*segs
= NULL
;
2512 struct sk_buff
*tail
= NULL
;
2513 struct sk_buff
*fskb
= skb_shinfo(skb
)->frag_list
;
2514 unsigned int mss
= skb_shinfo(skb
)->gso_size
;
2515 unsigned int doffset
= skb
->data
- skb_mac_header(skb
);
2516 unsigned int offset
= doffset
;
2517 unsigned int headroom
;
2519 int sg
= features
& NETIF_F_SG
;
2520 int nfrags
= skb_shinfo(skb
)->nr_frags
;
2525 __skb_push(skb
, doffset
);
2526 headroom
= skb_headroom(skb
);
2527 pos
= skb_headlen(skb
);
2530 struct sk_buff
*nskb
;
2535 len
= skb
->len
- offset
;
2539 hsize
= skb_headlen(skb
) - offset
;
2542 if (hsize
> len
|| !sg
)
2545 if (!hsize
&& i
>= nfrags
) {
2546 BUG_ON(fskb
->len
!= len
);
2549 nskb
= skb_clone(fskb
, GFP_ATOMIC
);
2552 if (unlikely(!nskb
))
2555 hsize
= skb_end_pointer(nskb
) - nskb
->head
;
2556 if (skb_cow_head(nskb
, doffset
+ headroom
)) {
2561 nskb
->truesize
+= skb_end_pointer(nskb
) - nskb
->head
-
2563 skb_release_head_state(nskb
);
2564 __skb_push(nskb
, doffset
);
2566 nskb
= alloc_skb(hsize
+ doffset
+ headroom
,
2569 if (unlikely(!nskb
))
2572 skb_reserve(nskb
, headroom
);
2573 __skb_put(nskb
, doffset
);
2582 __copy_skb_header(nskb
, skb
);
2583 nskb
->mac_len
= skb
->mac_len
;
2585 skb_reset_mac_header(nskb
);
2586 skb_set_network_header(nskb
, skb
->mac_len
);
2587 nskb
->transport_header
= (nskb
->network_header
+
2588 skb_network_header_len(skb
));
2589 skb_copy_from_linear_data(skb
, nskb
->data
, doffset
);
2591 if (fskb
!= skb_shinfo(skb
)->frag_list
)
2595 nskb
->ip_summed
= CHECKSUM_NONE
;
2596 nskb
->csum
= skb_copy_and_csum_bits(skb
, offset
,
2602 frag
= skb_shinfo(nskb
)->frags
;
2604 skb_copy_from_linear_data_offset(skb
, offset
,
2605 skb_put(nskb
, hsize
), hsize
);
2607 while (pos
< offset
+ len
&& i
< nfrags
) {
2608 *frag
= skb_shinfo(skb
)->frags
[i
];
2609 get_page(frag
->page
);
2613 frag
->page_offset
+= offset
- pos
;
2614 frag
->size
-= offset
- pos
;
2617 skb_shinfo(nskb
)->nr_frags
++;
2619 if (pos
+ size
<= offset
+ len
) {
2623 frag
->size
-= pos
+ size
- (offset
+ len
);
2630 if (pos
< offset
+ len
) {
2631 struct sk_buff
*fskb2
= fskb
;
2633 BUG_ON(pos
+ fskb
->len
!= offset
+ len
);
2639 fskb2
= skb_clone(fskb2
, GFP_ATOMIC
);
2645 SKB_FRAG_ASSERT(nskb
);
2646 skb_shinfo(nskb
)->frag_list
= fskb2
;
2650 nskb
->data_len
= len
- hsize
;
2651 nskb
->len
+= nskb
->data_len
;
2652 nskb
->truesize
+= nskb
->data_len
;
2653 } while ((offset
+= len
) < skb
->len
);
2658 while ((skb
= segs
)) {
2662 return ERR_PTR(err
);
2664 EXPORT_SYMBOL_GPL(skb_segment
);
2666 int skb_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2668 struct sk_buff
*p
= *head
;
2669 struct sk_buff
*nskb
;
2670 struct skb_shared_info
*skbinfo
= skb_shinfo(skb
);
2671 struct skb_shared_info
*pinfo
= skb_shinfo(p
);
2672 unsigned int headroom
;
2673 unsigned int len
= skb_gro_len(skb
);
2674 unsigned int offset
= skb_gro_offset(skb
);
2675 unsigned int headlen
= skb_headlen(skb
);
2677 if (p
->len
+ len
>= 65536)
2680 if (pinfo
->frag_list
)
2682 else if (headlen
<= offset
) {
2685 int i
= skbinfo
->nr_frags
;
2686 int nr_frags
= pinfo
->nr_frags
+ i
;
2690 if (nr_frags
> MAX_SKB_FRAGS
)
2693 pinfo
->nr_frags
= nr_frags
;
2694 skbinfo
->nr_frags
= 0;
2696 frag
= pinfo
->frags
+ nr_frags
;
2697 frag2
= skbinfo
->frags
+ i
;
2702 frag
->page_offset
+= offset
;
2703 frag
->size
-= offset
;
2705 skb
->truesize
-= skb
->data_len
;
2706 skb
->len
-= skb
->data_len
;
2709 NAPI_GRO_CB(skb
)->free
= 1;
2711 } else if (skb_gro_len(p
) != pinfo
->gso_size
)
2714 headroom
= skb_headroom(p
);
2715 nskb
= netdev_alloc_skb(p
->dev
, headroom
+ skb_gro_offset(p
));
2716 if (unlikely(!nskb
))
2719 __copy_skb_header(nskb
, p
);
2720 nskb
->mac_len
= p
->mac_len
;
2722 skb_reserve(nskb
, headroom
);
2723 __skb_put(nskb
, skb_gro_offset(p
));
2725 skb_set_mac_header(nskb
, skb_mac_header(p
) - p
->data
);
2726 skb_set_network_header(nskb
, skb_network_offset(p
));
2727 skb_set_transport_header(nskb
, skb_transport_offset(p
));
2729 __skb_pull(p
, skb_gro_offset(p
));
2730 memcpy(skb_mac_header(nskb
), skb_mac_header(p
),
2731 p
->data
- skb_mac_header(p
));
2733 *NAPI_GRO_CB(nskb
) = *NAPI_GRO_CB(p
);
2734 skb_shinfo(nskb
)->frag_list
= p
;
2735 skb_shinfo(nskb
)->gso_size
= pinfo
->gso_size
;
2736 skb_header_release(p
);
2739 nskb
->data_len
+= p
->len
;
2740 nskb
->truesize
+= p
->len
;
2741 nskb
->len
+= p
->len
;
2744 nskb
->next
= p
->next
;
2750 if (offset
> headlen
) {
2751 skbinfo
->frags
[0].page_offset
+= offset
- headlen
;
2752 skbinfo
->frags
[0].size
-= offset
- headlen
;
2756 __skb_pull(skb
, offset
);
2758 p
->prev
->next
= skb
;
2760 skb_header_release(skb
);
2763 NAPI_GRO_CB(p
)->count
++;
2768 NAPI_GRO_CB(skb
)->same_flow
= 1;
2771 EXPORT_SYMBOL_GPL(skb_gro_receive
);
2773 void __init
skb_init(void)
2775 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
2776 sizeof(struct sk_buff
),
2778 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2780 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
2781 (2*sizeof(struct sk_buff
)) +
2784 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2789 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2790 * @skb: Socket buffer containing the buffers to be mapped
2791 * @sg: The scatter-gather list to map into
2792 * @offset: The offset into the buffer's contents to start mapping
2793 * @len: Length of buffer space to be mapped
2795 * Fill the specified scatter-gather list with mappings/pointers into a
2796 * region of the buffer space attached to a socket buffer.
2799 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2801 int start
= skb_headlen(skb
);
2802 int i
, copy
= start
- offset
;
2803 struct sk_buff
*frag_iter
;
2809 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
2811 if ((len
-= copy
) == 0)
2816 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2819 WARN_ON(start
> offset
+ len
);
2821 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
2822 if ((copy
= end
- offset
) > 0) {
2823 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2827 sg_set_page(&sg
[elt
], frag
->page
, copy
,
2828 frag
->page_offset
+offset
-start
);
2837 skb_walk_frags(skb
, frag_iter
) {
2840 WARN_ON(start
> offset
+ len
);
2842 end
= start
+ frag_iter
->len
;
2843 if ((copy
= end
- offset
) > 0) {
2846 elt
+= __skb_to_sgvec(frag_iter
, sg
+elt
, offset
- start
,
2848 if ((len
-= copy
) == 0)
2858 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2860 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
);
2862 sg_mark_end(&sg
[nsg
- 1]);
2866 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
2869 * skb_cow_data - Check that a socket buffer's data buffers are writable
2870 * @skb: The socket buffer to check.
2871 * @tailbits: Amount of trailing space to be added
2872 * @trailer: Returned pointer to the skb where the @tailbits space begins
2874 * Make sure that the data buffers attached to a socket buffer are
2875 * writable. If they are not, private copies are made of the data buffers
2876 * and the socket buffer is set to use these instead.
2878 * If @tailbits is given, make sure that there is space to write @tailbits
2879 * bytes of data beyond current end of socket buffer. @trailer will be
2880 * set to point to the skb in which this space begins.
2882 * The number of scatterlist elements required to completely map the
2883 * COW'd and extended socket buffer will be returned.
2885 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
2889 struct sk_buff
*skb1
, **skb_p
;
2891 /* If skb is cloned or its head is paged, reallocate
2892 * head pulling out all the pages (pages are considered not writable
2893 * at the moment even if they are anonymous).
2895 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
2896 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
2899 /* Easy case. Most of packets will go this way. */
2900 if (!skb_has_frags(skb
)) {
2901 /* A little of trouble, not enough of space for trailer.
2902 * This should not happen, when stack is tuned to generate
2903 * good frames. OK, on miss we reallocate and reserve even more
2904 * space, 128 bytes is fair. */
2906 if (skb_tailroom(skb
) < tailbits
&&
2907 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
2915 /* Misery. We are in troubles, going to mincer fragments... */
2918 skb_p
= &skb_shinfo(skb
)->frag_list
;
2921 while ((skb1
= *skb_p
) != NULL
) {
2924 /* The fragment is partially pulled by someone,
2925 * this can happen on input. Copy it and everything
2928 if (skb_shared(skb1
))
2931 /* If the skb is the last, worry about trailer. */
2933 if (skb1
->next
== NULL
&& tailbits
) {
2934 if (skb_shinfo(skb1
)->nr_frags
||
2935 skb_has_frags(skb1
) ||
2936 skb_tailroom(skb1
) < tailbits
)
2937 ntail
= tailbits
+ 128;
2943 skb_shinfo(skb1
)->nr_frags
||
2944 skb_has_frags(skb1
)) {
2945 struct sk_buff
*skb2
;
2947 /* Fuck, we are miserable poor guys... */
2949 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
2951 skb2
= skb_copy_expand(skb1
,
2955 if (unlikely(skb2
== NULL
))
2959 skb_set_owner_w(skb2
, skb1
->sk
);
2961 /* Looking around. Are we still alive?
2962 * OK, link new skb, drop old one */
2964 skb2
->next
= skb1
->next
;
2971 skb_p
= &skb1
->next
;
2976 EXPORT_SYMBOL_GPL(skb_cow_data
);
2978 void skb_tstamp_tx(struct sk_buff
*orig_skb
,
2979 struct skb_shared_hwtstamps
*hwtstamps
)
2981 struct sock
*sk
= orig_skb
->sk
;
2982 struct sock_exterr_skb
*serr
;
2983 struct sk_buff
*skb
;
2989 skb
= skb_clone(orig_skb
, GFP_ATOMIC
);
2994 *skb_hwtstamps(skb
) =
2998 * no hardware time stamps available,
2999 * so keep the skb_shared_tx and only
3000 * store software time stamp
3002 skb
->tstamp
= ktime_get_real();
3005 serr
= SKB_EXT_ERR(skb
);
3006 memset(serr
, 0, sizeof(*serr
));
3007 serr
->ee
.ee_errno
= ENOMSG
;
3008 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TIMESTAMPING
;
3009 err
= sock_queue_err_skb(sk
, skb
);
3013 EXPORT_SYMBOL_GPL(skb_tstamp_tx
);
3017 * skb_partial_csum_set - set up and verify partial csum values for packet
3018 * @skb: the skb to set
3019 * @start: the number of bytes after skb->data to start checksumming.
3020 * @off: the offset from start to place the checksum.
3022 * For untrusted partially-checksummed packets, we need to make sure the values
3023 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3025 * This function checks and sets those values and skb->ip_summed: if this
3026 * returns false you should drop the packet.
3028 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
3030 if (unlikely(start
> skb_headlen(skb
)) ||
3031 unlikely((int)start
+ off
> skb_headlen(skb
) - 2)) {
3032 if (net_ratelimit())
3034 "bad partial csum: csum=%u/%u len=%u\n",
3035 start
, off
, skb_headlen(skb
));
3038 skb
->ip_summed
= CHECKSUM_PARTIAL
;
3039 skb
->csum_start
= skb_headroom(skb
) + start
;
3040 skb
->csum_offset
= off
;
3043 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);
3045 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
3047 if (net_ratelimit())
3048 pr_warning("%s: received packets cannot be forwarded"
3049 " while LRO is enabled\n", skb
->dev
->name
);
3051 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);