2 * Routines having to do with the 'struct sk_buff' memory handlers.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
8 * Alan Cox : Fixed the worst of the load
10 * Dave Platt : Interrupt stacking fix.
11 * Richard Kooijman : Timestamp fixes.
12 * Alan Cox : Changed buffer format.
13 * Alan Cox : destructor hook for AF_UNIX etc.
14 * Linus Torvalds : Better skb_clone.
15 * Alan Cox : Added skb_copy.
16 * Alan Cox : Added all the changed routines Linus
17 * only put in the headers
18 * Ray VanTassle : Fixed --skb->lock in free
19 * Alan Cox : skb_copy copy arp field
20 * Andi Kleen : slabified it.
21 * Robert Olsson : Removed skb_head_pool
24 * The __skb_ routines should be called with interrupts
25 * disabled, or you better be *real* sure that the operation is atomic
26 * with respect to whatever list is being frobbed (e.g. via lock_sock()
27 * or via disabling bottom half handlers, etc).
29 * This program is free software; you can redistribute it and/or
30 * modify it under the terms of the GNU General Public License
31 * as published by the Free Software Foundation; either version
32 * 2 of the License, or (at your option) any later version.
36 * The functions in this file will not compile correctly with gcc 2.4.x
39 #include <linux/module.h>
40 #include <linux/types.h>
41 #include <linux/kernel.h>
43 #include <linux/interrupt.h>
45 #include <linux/inet.h>
46 #include <linux/slab.h>
47 #include <linux/netdevice.h>
48 #ifdef CONFIG_NET_CLS_ACT
49 #include <net/pkt_sched.h>
51 #include <linux/string.h>
52 #include <linux/skbuff.h>
53 #include <linux/splice.h>
54 #include <linux/cache.h>
55 #include <linux/rtnetlink.h>
56 #include <linux/init.h>
57 #include <linux/scatterlist.h>
58 #include <linux/errqueue.h>
60 #include <net/protocol.h>
63 #include <net/checksum.h>
66 #include <asm/uaccess.h>
67 #include <asm/system.h>
68 #include <trace/skb.h>
72 static struct kmem_cache
*skbuff_head_cache __read_mostly
;
73 static struct kmem_cache
*skbuff_fclone_cache __read_mostly
;
75 static void sock_pipe_buf_release(struct pipe_inode_info
*pipe
,
76 struct pipe_buffer
*buf
)
81 static void sock_pipe_buf_get(struct pipe_inode_info
*pipe
,
82 struct pipe_buffer
*buf
)
87 static int sock_pipe_buf_steal(struct pipe_inode_info
*pipe
,
88 struct pipe_buffer
*buf
)
94 /* Pipe buffer operations for a socket. */
95 static struct pipe_buf_operations sock_pipe_buf_ops
= {
97 .map
= generic_pipe_buf_map
,
98 .unmap
= generic_pipe_buf_unmap
,
99 .confirm
= generic_pipe_buf_confirm
,
100 .release
= sock_pipe_buf_release
,
101 .steal
= sock_pipe_buf_steal
,
102 .get
= sock_pipe_buf_get
,
106 * Keep out-of-line to prevent kernel bloat.
107 * __builtin_return_address is not used because it is not always
112 * skb_over_panic - private function
117 * Out of line support code for skb_put(). Not user callable.
119 void skb_over_panic(struct sk_buff
*skb
, int sz
, void *here
)
121 printk(KERN_EMERG
"skb_over_panic: text:%p len:%d put:%d head:%p "
122 "data:%p tail:%#lx end:%#lx dev:%s\n",
123 here
, skb
->len
, sz
, skb
->head
, skb
->data
,
124 (unsigned long)skb
->tail
, (unsigned long)skb
->end
,
125 skb
->dev
? skb
->dev
->name
: "<NULL>");
128 EXPORT_SYMBOL(skb_over_panic
);
131 * skb_under_panic - private function
136 * Out of line support code for skb_push(). Not user callable.
139 void skb_under_panic(struct sk_buff
*skb
, int sz
, void *here
)
141 printk(KERN_EMERG
"skb_under_panic: text:%p len:%d put:%d head:%p "
142 "data:%p tail:%#lx end:%#lx dev:%s\n",
143 here
, skb
->len
, sz
, skb
->head
, skb
->data
,
144 (unsigned long)skb
->tail
, (unsigned long)skb
->end
,
145 skb
->dev
? skb
->dev
->name
: "<NULL>");
148 EXPORT_SYMBOL(skb_under_panic
);
150 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
151 * 'private' fields and also do memory statistics to find all the
157 * __alloc_skb - allocate a network buffer
158 * @size: size to allocate
159 * @gfp_mask: allocation mask
160 * @fclone: allocate from fclone cache instead of head cache
161 * and allocate a cloned (child) skb
162 * @node: numa node to allocate memory on
164 * Allocate a new &sk_buff. The returned buffer has no headroom and a
165 * tail room of size bytes. The object has a reference count of one.
166 * The return is the buffer. On a failure the return is %NULL.
168 * Buffers may only be allocated from interrupts using a @gfp_mask of
171 struct sk_buff
*__alloc_skb(unsigned int size
, gfp_t gfp_mask
,
172 int fclone
, int node
)
174 struct kmem_cache
*cache
;
175 struct skb_shared_info
*shinfo
;
179 cache
= fclone
? skbuff_fclone_cache
: skbuff_head_cache
;
182 skb
= kmem_cache_alloc_node(cache
, gfp_mask
& ~__GFP_DMA
, node
);
186 size
= SKB_DATA_ALIGN(size
);
187 data
= kmalloc_node_track_caller(size
+ sizeof(struct skb_shared_info
),
193 * Only clear those fields we need to clear, not those that we will
194 * actually initialise below. Hence, don't put any more fields after
195 * the tail pointer in struct sk_buff!
197 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
198 skb
->truesize
= size
+ sizeof(struct sk_buff
);
199 atomic_set(&skb
->users
, 1);
202 skb_reset_tail_pointer(skb
);
203 skb
->end
= skb
->tail
+ size
;
204 /* make sure we initialize shinfo sequentially */
205 shinfo
= skb_shinfo(skb
);
206 atomic_set(&shinfo
->dataref
, 1);
207 shinfo
->nr_frags
= 0;
208 shinfo
->gso_size
= 0;
209 shinfo
->gso_segs
= 0;
210 shinfo
->gso_type
= 0;
211 shinfo
->ip6_frag_id
= 0;
212 shinfo
->tx_flags
.flags
= 0;
213 skb_frag_list_init(skb
);
214 memset(&shinfo
->hwtstamps
, 0, sizeof(shinfo
->hwtstamps
));
217 struct sk_buff
*child
= skb
+ 1;
218 atomic_t
*fclone_ref
= (atomic_t
*) (child
+ 1);
220 skb
->fclone
= SKB_FCLONE_ORIG
;
221 atomic_set(fclone_ref
, 1);
223 child
->fclone
= SKB_FCLONE_UNAVAILABLE
;
228 kmem_cache_free(cache
, skb
);
232 EXPORT_SYMBOL(__alloc_skb
);
235 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
236 * @dev: network device to receive on
237 * @length: length to allocate
238 * @gfp_mask: get_free_pages mask, passed to alloc_skb
240 * Allocate a new &sk_buff and assign it a usage count of one. The
241 * buffer has unspecified headroom built in. Users should allocate
242 * the headroom they think they need without accounting for the
243 * built in space. The built in space is used for optimisations.
245 * %NULL is returned if there is no free memory.
247 struct sk_buff
*__netdev_alloc_skb(struct net_device
*dev
,
248 unsigned int length
, gfp_t gfp_mask
)
250 int node
= dev
->dev
.parent
? dev_to_node(dev
->dev
.parent
) : -1;
253 skb
= __alloc_skb(length
+ NET_SKB_PAD
, gfp_mask
, 0, node
);
255 skb_reserve(skb
, NET_SKB_PAD
);
260 EXPORT_SYMBOL(__netdev_alloc_skb
);
262 struct page
*__netdev_alloc_page(struct net_device
*dev
, gfp_t gfp_mask
)
264 int node
= dev
->dev
.parent
? dev_to_node(dev
->dev
.parent
) : -1;
267 page
= alloc_pages_node(node
, gfp_mask
, 0);
270 EXPORT_SYMBOL(__netdev_alloc_page
);
272 void skb_add_rx_frag(struct sk_buff
*skb
, int i
, struct page
*page
, int off
,
275 skb_fill_page_desc(skb
, i
, page
, off
, size
);
277 skb
->data_len
+= size
;
278 skb
->truesize
+= size
;
280 EXPORT_SYMBOL(skb_add_rx_frag
);
283 * dev_alloc_skb - allocate an skbuff for receiving
284 * @length: length to allocate
286 * Allocate a new &sk_buff and assign it a usage count of one. The
287 * buffer has unspecified headroom built in. Users should allocate
288 * the headroom they think they need without accounting for the
289 * built in space. The built in space is used for optimisations.
291 * %NULL is returned if there is no free memory. Although this function
292 * allocates memory it can be called from an interrupt.
294 struct sk_buff
*dev_alloc_skb(unsigned int length
)
297 * There is more code here than it seems:
298 * __dev_alloc_skb is an inline
300 return __dev_alloc_skb(length
, GFP_ATOMIC
);
302 EXPORT_SYMBOL(dev_alloc_skb
);
304 static void skb_drop_list(struct sk_buff
**listp
)
306 struct sk_buff
*list
= *listp
;
311 struct sk_buff
*this = list
;
317 static inline void skb_drop_fraglist(struct sk_buff
*skb
)
319 skb_drop_list(&skb_shinfo(skb
)->frag_list
);
322 static void skb_clone_fraglist(struct sk_buff
*skb
)
324 struct sk_buff
*list
;
326 skb_walk_frags(skb
, list
)
330 static void skb_release_data(struct sk_buff
*skb
)
333 !atomic_sub_return(skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1,
334 &skb_shinfo(skb
)->dataref
)) {
335 if (skb_shinfo(skb
)->nr_frags
) {
337 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
338 put_page(skb_shinfo(skb
)->frags
[i
].page
);
341 if (skb_has_frags(skb
))
342 skb_drop_fraglist(skb
);
349 * Free an skbuff by memory without cleaning the state.
351 static void kfree_skbmem(struct sk_buff
*skb
)
353 struct sk_buff
*other
;
354 atomic_t
*fclone_ref
;
356 switch (skb
->fclone
) {
357 case SKB_FCLONE_UNAVAILABLE
:
358 kmem_cache_free(skbuff_head_cache
, skb
);
361 case SKB_FCLONE_ORIG
:
362 fclone_ref
= (atomic_t
*) (skb
+ 2);
363 if (atomic_dec_and_test(fclone_ref
))
364 kmem_cache_free(skbuff_fclone_cache
, skb
);
367 case SKB_FCLONE_CLONE
:
368 fclone_ref
= (atomic_t
*) (skb
+ 1);
371 /* The clone portion is available for
372 * fast-cloning again.
374 skb
->fclone
= SKB_FCLONE_UNAVAILABLE
;
376 if (atomic_dec_and_test(fclone_ref
))
377 kmem_cache_free(skbuff_fclone_cache
, other
);
382 static void skb_release_head_state(struct sk_buff
*skb
)
386 secpath_put(skb
->sp
);
388 if (skb
->destructor
) {
390 skb
->destructor(skb
);
392 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
393 nf_conntrack_put(skb
->nfct
);
394 nf_conntrack_put_reasm(skb
->nfct_reasm
);
396 #ifdef CONFIG_BRIDGE_NETFILTER
397 nf_bridge_put(skb
->nf_bridge
);
399 /* XXX: IS this still necessary? - JHS */
400 #ifdef CONFIG_NET_SCHED
402 #ifdef CONFIG_NET_CLS_ACT
408 /* Free everything but the sk_buff shell. */
409 static void skb_release_all(struct sk_buff
*skb
)
411 skb_release_head_state(skb
);
412 skb_release_data(skb
);
416 * __kfree_skb - private function
419 * Free an sk_buff. Release anything attached to the buffer.
420 * Clean the state. This is an internal helper function. Users should
421 * always call kfree_skb
424 void __kfree_skb(struct sk_buff
*skb
)
426 skb_release_all(skb
);
429 EXPORT_SYMBOL(__kfree_skb
);
432 * kfree_skb - free an sk_buff
433 * @skb: buffer to free
435 * Drop a reference to the buffer and free it if the usage count has
438 void kfree_skb(struct sk_buff
*skb
)
442 if (likely(atomic_read(&skb
->users
) == 1))
444 else if (likely(!atomic_dec_and_test(&skb
->users
)))
446 trace_kfree_skb(skb
, __builtin_return_address(0));
449 EXPORT_SYMBOL(kfree_skb
);
452 * consume_skb - free an skbuff
453 * @skb: buffer to free
455 * Drop a ref to the buffer and free it if the usage count has hit zero
456 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
457 * is being dropped after a failure and notes that
459 void consume_skb(struct sk_buff
*skb
)
463 if (likely(atomic_read(&skb
->users
) == 1))
465 else if (likely(!atomic_dec_and_test(&skb
->users
)))
469 EXPORT_SYMBOL(consume_skb
);
472 * skb_recycle_check - check if skb can be reused for receive
474 * @skb_size: minimum receive buffer size
476 * Checks that the skb passed in is not shared or cloned, and
477 * that it is linear and its head portion at least as large as
478 * skb_size so that it can be recycled as a receive buffer.
479 * If these conditions are met, this function does any necessary
480 * reference count dropping and cleans up the skbuff as if it
481 * just came from __alloc_skb().
483 int skb_recycle_check(struct sk_buff
*skb
, int skb_size
)
485 struct skb_shared_info
*shinfo
;
487 if (skb_is_nonlinear(skb
) || skb
->fclone
!= SKB_FCLONE_UNAVAILABLE
)
490 skb_size
= SKB_DATA_ALIGN(skb_size
+ NET_SKB_PAD
);
491 if (skb_end_pointer(skb
) - skb
->head
< skb_size
)
494 if (skb_shared(skb
) || skb_cloned(skb
))
497 skb_release_head_state(skb
);
498 shinfo
= skb_shinfo(skb
);
499 atomic_set(&shinfo
->dataref
, 1);
500 shinfo
->nr_frags
= 0;
501 shinfo
->gso_size
= 0;
502 shinfo
->gso_segs
= 0;
503 shinfo
->gso_type
= 0;
504 shinfo
->ip6_frag_id
= 0;
505 shinfo
->tx_flags
.flags
= 0;
506 skb_frag_list_init(skb
);
507 memset(&shinfo
->hwtstamps
, 0, sizeof(shinfo
->hwtstamps
));
509 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
510 skb
->data
= skb
->head
+ NET_SKB_PAD
;
511 skb_reset_tail_pointer(skb
);
515 EXPORT_SYMBOL(skb_recycle_check
);
517 static void __copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
519 new->tstamp
= old
->tstamp
;
521 new->transport_header
= old
->transport_header
;
522 new->network_header
= old
->network_header
;
523 new->mac_header
= old
->mac_header
;
524 skb_dst_set(new, dst_clone(skb_dst(old
)));
526 new->sp
= secpath_get(old
->sp
);
528 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
529 new->csum
= old
->csum
;
530 new->local_df
= old
->local_df
;
531 new->pkt_type
= old
->pkt_type
;
532 new->ip_summed
= old
->ip_summed
;
533 skb_copy_queue_mapping(new, old
);
534 new->priority
= old
->priority
;
535 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
536 new->ipvs_property
= old
->ipvs_property
;
538 new->protocol
= old
->protocol
;
539 new->mark
= old
->mark
;
542 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
543 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
544 new->nf_trace
= old
->nf_trace
;
546 #ifdef CONFIG_NET_SCHED
547 new->tc_index
= old
->tc_index
;
548 #ifdef CONFIG_NET_CLS_ACT
549 new->tc_verd
= old
->tc_verd
;
552 new->vlan_tci
= old
->vlan_tci
;
553 #if defined(CONFIG_MAC80211) || defined(CONFIG_MAC80211_MODULE)
554 new->do_not_encrypt
= old
->do_not_encrypt
;
557 skb_copy_secmark(new, old
);
561 * You should not add any new code to this function. Add it to
562 * __copy_skb_header above instead.
564 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
566 #define C(x) n->x = skb->x
568 n
->next
= n
->prev
= NULL
;
570 __copy_skb_header(n
, skb
);
575 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
578 n
->destructor
= NULL
;
584 atomic_set(&n
->users
, 1);
586 atomic_inc(&(skb_shinfo(skb
)->dataref
));
594 * skb_morph - morph one skb into another
595 * @dst: the skb to receive the contents
596 * @src: the skb to supply the contents
598 * This is identical to skb_clone except that the target skb is
599 * supplied by the user.
601 * The target skb is returned upon exit.
603 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
605 skb_release_all(dst
);
606 return __skb_clone(dst
, src
);
608 EXPORT_SYMBOL_GPL(skb_morph
);
611 * skb_clone - duplicate an sk_buff
612 * @skb: buffer to clone
613 * @gfp_mask: allocation priority
615 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
616 * copies share the same packet data but not structure. The new
617 * buffer has a reference count of 1. If the allocation fails the
618 * function returns %NULL otherwise the new buffer is returned.
620 * If this function is called from an interrupt gfp_mask() must be
624 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
629 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
630 n
->fclone
== SKB_FCLONE_UNAVAILABLE
) {
631 atomic_t
*fclone_ref
= (atomic_t
*) (n
+ 1);
632 n
->fclone
= SKB_FCLONE_CLONE
;
633 atomic_inc(fclone_ref
);
635 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
638 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
641 return __skb_clone(n
, skb
);
643 EXPORT_SYMBOL(skb_clone
);
645 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
647 #ifndef NET_SKBUFF_DATA_USES_OFFSET
649 * Shift between the two data areas in bytes
651 unsigned long offset
= new->data
- old
->data
;
654 __copy_skb_header(new, old
);
656 #ifndef NET_SKBUFF_DATA_USES_OFFSET
657 /* {transport,network,mac}_header are relative to skb->head */
658 new->transport_header
+= offset
;
659 new->network_header
+= offset
;
660 new->mac_header
+= offset
;
662 skb_shinfo(new)->gso_size
= skb_shinfo(old
)->gso_size
;
663 skb_shinfo(new)->gso_segs
= skb_shinfo(old
)->gso_segs
;
664 skb_shinfo(new)->gso_type
= skb_shinfo(old
)->gso_type
;
668 * skb_copy - create private copy of an sk_buff
669 * @skb: buffer to copy
670 * @gfp_mask: allocation priority
672 * Make a copy of both an &sk_buff and its data. This is used when the
673 * caller wishes to modify the data and needs a private copy of the
674 * data to alter. Returns %NULL on failure or the pointer to the buffer
675 * on success. The returned buffer has a reference count of 1.
677 * As by-product this function converts non-linear &sk_buff to linear
678 * one, so that &sk_buff becomes completely private and caller is allowed
679 * to modify all the data of returned buffer. This means that this
680 * function is not recommended for use in circumstances when only
681 * header is going to be modified. Use pskb_copy() instead.
684 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
686 int headerlen
= skb
->data
- skb
->head
;
688 * Allocate the copy buffer
691 #ifdef NET_SKBUFF_DATA_USES_OFFSET
692 n
= alloc_skb(skb
->end
+ skb
->data_len
, gfp_mask
);
694 n
= alloc_skb(skb
->end
- skb
->head
+ skb
->data_len
, gfp_mask
);
699 /* Set the data pointer */
700 skb_reserve(n
, headerlen
);
701 /* Set the tail pointer and length */
702 skb_put(n
, skb
->len
);
704 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
707 copy_skb_header(n
, skb
);
710 EXPORT_SYMBOL(skb_copy
);
713 * pskb_copy - create copy of an sk_buff with private head.
714 * @skb: buffer to copy
715 * @gfp_mask: allocation priority
717 * Make a copy of both an &sk_buff and part of its data, located
718 * in header. Fragmented data remain shared. This is used when
719 * the caller wishes to modify only header of &sk_buff and needs
720 * private copy of the header to alter. Returns %NULL on failure
721 * or the pointer to the buffer on success.
722 * The returned buffer has a reference count of 1.
725 struct sk_buff
*pskb_copy(struct sk_buff
*skb
, gfp_t gfp_mask
)
728 * Allocate the copy buffer
731 #ifdef NET_SKBUFF_DATA_USES_OFFSET
732 n
= alloc_skb(skb
->end
, gfp_mask
);
734 n
= alloc_skb(skb
->end
- skb
->head
, gfp_mask
);
739 /* Set the data pointer */
740 skb_reserve(n
, skb
->data
- skb
->head
);
741 /* Set the tail pointer and length */
742 skb_put(n
, skb_headlen(skb
));
744 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
746 n
->truesize
+= skb
->data_len
;
747 n
->data_len
= skb
->data_len
;
750 if (skb_shinfo(skb
)->nr_frags
) {
753 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
754 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
755 get_page(skb_shinfo(n
)->frags
[i
].page
);
757 skb_shinfo(n
)->nr_frags
= i
;
760 if (skb_has_frags(skb
)) {
761 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
762 skb_clone_fraglist(n
);
765 copy_skb_header(n
, skb
);
769 EXPORT_SYMBOL(pskb_copy
);
772 * pskb_expand_head - reallocate header of &sk_buff
773 * @skb: buffer to reallocate
774 * @nhead: room to add at head
775 * @ntail: room to add at tail
776 * @gfp_mask: allocation priority
778 * Expands (or creates identical copy, if &nhead and &ntail are zero)
779 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
780 * reference count of 1. Returns zero in the case of success or error,
781 * if expansion failed. In the last case, &sk_buff is not changed.
783 * All the pointers pointing into skb header may change and must be
784 * reloaded after call to this function.
787 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
792 #ifdef NET_SKBUFF_DATA_USES_OFFSET
793 int size
= nhead
+ skb
->end
+ ntail
;
795 int size
= nhead
+ (skb
->end
- skb
->head
) + ntail
;
804 size
= SKB_DATA_ALIGN(size
);
806 data
= kmalloc(size
+ sizeof(struct skb_shared_info
), gfp_mask
);
810 /* Copy only real data... and, alas, header. This should be
811 * optimized for the cases when header is void. */
812 #ifdef NET_SKBUFF_DATA_USES_OFFSET
813 memcpy(data
+ nhead
, skb
->head
, skb
->tail
);
815 memcpy(data
+ nhead
, skb
->head
, skb
->tail
- skb
->head
);
817 memcpy(data
+ size
, skb_end_pointer(skb
),
818 sizeof(struct skb_shared_info
));
820 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
821 get_page(skb_shinfo(skb
)->frags
[i
].page
);
823 if (skb_has_frags(skb
))
824 skb_clone_fraglist(skb
);
826 skb_release_data(skb
);
828 off
= (data
+ nhead
) - skb
->head
;
832 #ifdef NET_SKBUFF_DATA_USES_OFFSET
836 skb
->end
= skb
->head
+ size
;
838 /* {transport,network,mac}_header and tail are relative to skb->head */
840 skb
->transport_header
+= off
;
841 skb
->network_header
+= off
;
842 skb
->mac_header
+= off
;
843 skb
->csum_start
+= nhead
;
847 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
853 EXPORT_SYMBOL(pskb_expand_head
);
855 /* Make private copy of skb with writable head and some headroom */
857 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
859 struct sk_buff
*skb2
;
860 int delta
= headroom
- skb_headroom(skb
);
863 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
865 skb2
= skb_clone(skb
, GFP_ATOMIC
);
866 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
874 EXPORT_SYMBOL(skb_realloc_headroom
);
877 * skb_copy_expand - copy and expand sk_buff
878 * @skb: buffer to copy
879 * @newheadroom: new free bytes at head
880 * @newtailroom: new free bytes at tail
881 * @gfp_mask: allocation priority
883 * Make a copy of both an &sk_buff and its data and while doing so
884 * allocate additional space.
886 * This is used when the caller wishes to modify the data and needs a
887 * private copy of the data to alter as well as more space for new fields.
888 * Returns %NULL on failure or the pointer to the buffer
889 * on success. The returned buffer has a reference count of 1.
891 * You must pass %GFP_ATOMIC as the allocation priority if this function
892 * is called from an interrupt.
894 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
895 int newheadroom
, int newtailroom
,
899 * Allocate the copy buffer
901 struct sk_buff
*n
= alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
903 int oldheadroom
= skb_headroom(skb
);
904 int head_copy_len
, head_copy_off
;
910 skb_reserve(n
, newheadroom
);
912 /* Set the tail pointer and length */
913 skb_put(n
, skb
->len
);
915 head_copy_len
= oldheadroom
;
917 if (newheadroom
<= head_copy_len
)
918 head_copy_len
= newheadroom
;
920 head_copy_off
= newheadroom
- head_copy_len
;
922 /* Copy the linear header and data. */
923 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
924 skb
->len
+ head_copy_len
))
927 copy_skb_header(n
, skb
);
929 off
= newheadroom
- oldheadroom
;
930 n
->csum_start
+= off
;
931 #ifdef NET_SKBUFF_DATA_USES_OFFSET
932 n
->transport_header
+= off
;
933 n
->network_header
+= off
;
934 n
->mac_header
+= off
;
939 EXPORT_SYMBOL(skb_copy_expand
);
942 * skb_pad - zero pad the tail of an skb
943 * @skb: buffer to pad
946 * Ensure that a buffer is followed by a padding area that is zero
947 * filled. Used by network drivers which may DMA or transfer data
948 * beyond the buffer end onto the wire.
950 * May return error in out of memory cases. The skb is freed on error.
953 int skb_pad(struct sk_buff
*skb
, int pad
)
958 /* If the skbuff is non linear tailroom is always zero.. */
959 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
960 memset(skb
->data
+skb
->len
, 0, pad
);
964 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
965 if (likely(skb_cloned(skb
) || ntail
> 0)) {
966 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
971 /* FIXME: The use of this function with non-linear skb's really needs
974 err
= skb_linearize(skb
);
978 memset(skb
->data
+ skb
->len
, 0, pad
);
985 EXPORT_SYMBOL(skb_pad
);
988 * skb_put - add data to a buffer
989 * @skb: buffer to use
990 * @len: amount of data to add
992 * This function extends the used data area of the buffer. If this would
993 * exceed the total buffer size the kernel will panic. A pointer to the
994 * first byte of the extra data is returned.
996 unsigned char *skb_put(struct sk_buff
*skb
, unsigned int len
)
998 unsigned char *tmp
= skb_tail_pointer(skb
);
999 SKB_LINEAR_ASSERT(skb
);
1002 if (unlikely(skb
->tail
> skb
->end
))
1003 skb_over_panic(skb
, len
, __builtin_return_address(0));
1006 EXPORT_SYMBOL(skb_put
);
1009 * skb_push - add data to the start of a buffer
1010 * @skb: buffer to use
1011 * @len: amount of data to add
1013 * This function extends the used data area of the buffer at the buffer
1014 * start. If this would exceed the total buffer headroom the kernel will
1015 * panic. A pointer to the first byte of the extra data is returned.
1017 unsigned char *skb_push(struct sk_buff
*skb
, unsigned int len
)
1021 if (unlikely(skb
->data
<skb
->head
))
1022 skb_under_panic(skb
, len
, __builtin_return_address(0));
1025 EXPORT_SYMBOL(skb_push
);
1028 * skb_pull - remove data from the start of a buffer
1029 * @skb: buffer to use
1030 * @len: amount of data to remove
1032 * This function removes data from the start of a buffer, returning
1033 * the memory to the headroom. A pointer to the next data in the buffer
1034 * is returned. Once the data has been pulled future pushes will overwrite
1037 unsigned char *skb_pull(struct sk_buff
*skb
, unsigned int len
)
1039 return unlikely(len
> skb
->len
) ? NULL
: __skb_pull(skb
, len
);
1041 EXPORT_SYMBOL(skb_pull
);
1044 * skb_trim - remove end from a buffer
1045 * @skb: buffer to alter
1048 * Cut the length of a buffer down by removing data from the tail. If
1049 * the buffer is already under the length specified it is not modified.
1050 * The skb must be linear.
1052 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1055 __skb_trim(skb
, len
);
1057 EXPORT_SYMBOL(skb_trim
);
1059 /* Trims skb to length len. It can change skb pointers.
1062 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1064 struct sk_buff
**fragp
;
1065 struct sk_buff
*frag
;
1066 int offset
= skb_headlen(skb
);
1067 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1071 if (skb_cloned(skb
) &&
1072 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1079 for (; i
< nfrags
; i
++) {
1080 int end
= offset
+ skb_shinfo(skb
)->frags
[i
].size
;
1087 skb_shinfo(skb
)->frags
[i
++].size
= len
- offset
;
1090 skb_shinfo(skb
)->nr_frags
= i
;
1092 for (; i
< nfrags
; i
++)
1093 put_page(skb_shinfo(skb
)->frags
[i
].page
);
1095 if (skb_has_frags(skb
))
1096 skb_drop_fraglist(skb
);
1100 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1101 fragp
= &frag
->next
) {
1102 int end
= offset
+ frag
->len
;
1104 if (skb_shared(frag
)) {
1105 struct sk_buff
*nfrag
;
1107 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1108 if (unlikely(!nfrag
))
1111 nfrag
->next
= frag
->next
;
1123 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1127 skb_drop_list(&frag
->next
);
1132 if (len
> skb_headlen(skb
)) {
1133 skb
->data_len
-= skb
->len
- len
;
1138 skb_set_tail_pointer(skb
, len
);
1143 EXPORT_SYMBOL(___pskb_trim
);
1146 * __pskb_pull_tail - advance tail of skb header
1147 * @skb: buffer to reallocate
1148 * @delta: number of bytes to advance tail
1150 * The function makes a sense only on a fragmented &sk_buff,
1151 * it expands header moving its tail forward and copying necessary
1152 * data from fragmented part.
1154 * &sk_buff MUST have reference count of 1.
1156 * Returns %NULL (and &sk_buff does not change) if pull failed
1157 * or value of new tail of skb in the case of success.
1159 * All the pointers pointing into skb header may change and must be
1160 * reloaded after call to this function.
1163 /* Moves tail of skb head forward, copying data from fragmented part,
1164 * when it is necessary.
1165 * 1. It may fail due to malloc failure.
1166 * 2. It may change skb pointers.
1168 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1170 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1172 /* If skb has not enough free space at tail, get new one
1173 * plus 128 bytes for future expansions. If we have enough
1174 * room at tail, reallocate without expansion only if skb is cloned.
1176 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1178 if (eat
> 0 || skb_cloned(skb
)) {
1179 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1184 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
1187 /* Optimization: no fragments, no reasons to preestimate
1188 * size of pulled pages. Superb.
1190 if (!skb_has_frags(skb
))
1193 /* Estimate size of pulled pages. */
1195 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1196 if (skb_shinfo(skb
)->frags
[i
].size
>= eat
)
1198 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1201 /* If we need update frag list, we are in troubles.
1202 * Certainly, it possible to add an offset to skb data,
1203 * but taking into account that pulling is expected to
1204 * be very rare operation, it is worth to fight against
1205 * further bloating skb head and crucify ourselves here instead.
1206 * Pure masohism, indeed. 8)8)
1209 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1210 struct sk_buff
*clone
= NULL
;
1211 struct sk_buff
*insp
= NULL
;
1216 if (list
->len
<= eat
) {
1217 /* Eaten as whole. */
1222 /* Eaten partially. */
1224 if (skb_shared(list
)) {
1225 /* Sucks! We need to fork list. :-( */
1226 clone
= skb_clone(list
, GFP_ATOMIC
);
1232 /* This may be pulled without
1236 if (!pskb_pull(list
, eat
)) {
1244 /* Free pulled out fragments. */
1245 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1246 skb_shinfo(skb
)->frag_list
= list
->next
;
1249 /* And insert new clone at head. */
1252 skb_shinfo(skb
)->frag_list
= clone
;
1255 /* Success! Now we may commit changes to skb data. */
1260 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1261 if (skb_shinfo(skb
)->frags
[i
].size
<= eat
) {
1262 put_page(skb_shinfo(skb
)->frags
[i
].page
);
1263 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1265 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1267 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1268 skb_shinfo(skb
)->frags
[k
].size
-= eat
;
1274 skb_shinfo(skb
)->nr_frags
= k
;
1277 skb
->data_len
-= delta
;
1279 return skb_tail_pointer(skb
);
1281 EXPORT_SYMBOL(__pskb_pull_tail
);
1283 /* Copy some data bits from skb to kernel buffer. */
1285 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1287 int start
= skb_headlen(skb
);
1288 struct sk_buff
*frag_iter
;
1291 if (offset
> (int)skb
->len
- len
)
1295 if ((copy
= start
- offset
) > 0) {
1298 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
1299 if ((len
-= copy
) == 0)
1305 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1308 WARN_ON(start
> offset
+ len
);
1310 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1311 if ((copy
= end
- offset
) > 0) {
1317 vaddr
= kmap_skb_frag(&skb_shinfo(skb
)->frags
[i
]);
1319 vaddr
+ skb_shinfo(skb
)->frags
[i
].page_offset
+
1320 offset
- start
, copy
);
1321 kunmap_skb_frag(vaddr
);
1323 if ((len
-= copy
) == 0)
1331 skb_walk_frags(skb
, frag_iter
) {
1334 WARN_ON(start
> offset
+ len
);
1336 end
= start
+ frag_iter
->len
;
1337 if ((copy
= end
- offset
) > 0) {
1340 if (skb_copy_bits(frag_iter
, offset
- start
, to
, copy
))
1342 if ((len
-= copy
) == 0)
1355 EXPORT_SYMBOL(skb_copy_bits
);
1358 * Callback from splice_to_pipe(), if we need to release some pages
1359 * at the end of the spd in case we error'ed out in filling the pipe.
1361 static void sock_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
1363 put_page(spd
->pages
[i
]);
1366 static inline struct page
*linear_to_page(struct page
*page
, unsigned int *len
,
1367 unsigned int *offset
,
1368 struct sk_buff
*skb
, struct sock
*sk
)
1370 struct page
*p
= sk
->sk_sndmsg_page
;
1375 p
= sk
->sk_sndmsg_page
= alloc_pages(sk
->sk_allocation
, 0);
1379 off
= sk
->sk_sndmsg_off
= 0;
1380 /* hold one ref to this page until it's full */
1384 off
= sk
->sk_sndmsg_off
;
1385 mlen
= PAGE_SIZE
- off
;
1386 if (mlen
< 64 && mlen
< *len
) {
1391 *len
= min_t(unsigned int, *len
, mlen
);
1394 memcpy(page_address(p
) + off
, page_address(page
) + *offset
, *len
);
1395 sk
->sk_sndmsg_off
+= *len
;
1403 * Fill page/offset/length into spd, if it can hold more pages.
1405 static inline int spd_fill_page(struct splice_pipe_desc
*spd
, struct page
*page
,
1406 unsigned int *len
, unsigned int offset
,
1407 struct sk_buff
*skb
, int linear
,
1410 if (unlikely(spd
->nr_pages
== PIPE_BUFFERS
))
1414 page
= linear_to_page(page
, len
, &offset
, skb
, sk
);
1420 spd
->pages
[spd
->nr_pages
] = page
;
1421 spd
->partial
[spd
->nr_pages
].len
= *len
;
1422 spd
->partial
[spd
->nr_pages
].offset
= offset
;
1428 static inline void __segment_seek(struct page
**page
, unsigned int *poff
,
1429 unsigned int *plen
, unsigned int off
)
1434 n
= *poff
/ PAGE_SIZE
;
1436 *page
= nth_page(*page
, n
);
1438 *poff
= *poff
% PAGE_SIZE
;
1442 static inline int __splice_segment(struct page
*page
, unsigned int poff
,
1443 unsigned int plen
, unsigned int *off
,
1444 unsigned int *len
, struct sk_buff
*skb
,
1445 struct splice_pipe_desc
*spd
, int linear
,
1451 /* skip this segment if already processed */
1457 /* ignore any bits we already processed */
1459 __segment_seek(&page
, &poff
, &plen
, *off
);
1464 unsigned int flen
= min(*len
, plen
);
1466 /* the linear region may spread across several pages */
1467 flen
= min_t(unsigned int, flen
, PAGE_SIZE
- poff
);
1469 if (spd_fill_page(spd
, page
, &flen
, poff
, skb
, linear
, sk
))
1472 __segment_seek(&page
, &poff
, &plen
, flen
);
1475 } while (*len
&& plen
);
1481 * Map linear and fragment data from the skb to spd. It reports failure if the
1482 * pipe is full or if we already spliced the requested length.
1484 static int __skb_splice_bits(struct sk_buff
*skb
, unsigned int *offset
,
1485 unsigned int *len
, struct splice_pipe_desc
*spd
,
1491 * map the linear part
1493 if (__splice_segment(virt_to_page(skb
->data
),
1494 (unsigned long) skb
->data
& (PAGE_SIZE
- 1),
1496 offset
, len
, skb
, spd
, 1, sk
))
1500 * then map the fragments
1502 for (seg
= 0; seg
< skb_shinfo(skb
)->nr_frags
; seg
++) {
1503 const skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[seg
];
1505 if (__splice_segment(f
->page
, f
->page_offset
, f
->size
,
1506 offset
, len
, skb
, spd
, 0, sk
))
1514 * Map data from the skb to a pipe. Should handle both the linear part,
1515 * the fragments, and the frag list. It does NOT handle frag lists within
1516 * the frag list, if such a thing exists. We'd probably need to recurse to
1517 * handle that cleanly.
1519 int skb_splice_bits(struct sk_buff
*skb
, unsigned int offset
,
1520 struct pipe_inode_info
*pipe
, unsigned int tlen
,
1523 struct partial_page partial
[PIPE_BUFFERS
];
1524 struct page
*pages
[PIPE_BUFFERS
];
1525 struct splice_pipe_desc spd
= {
1529 .ops
= &sock_pipe_buf_ops
,
1530 .spd_release
= sock_spd_release
,
1532 struct sk_buff
*frag_iter
;
1533 struct sock
*sk
= skb
->sk
;
1536 * __skb_splice_bits() only fails if the output has no room left,
1537 * so no point in going over the frag_list for the error case.
1539 if (__skb_splice_bits(skb
, &offset
, &tlen
, &spd
, sk
))
1545 * now see if we have a frag_list to map
1547 skb_walk_frags(skb
, frag_iter
) {
1550 if (__skb_splice_bits(frag_iter
, &offset
, &tlen
, &spd
, sk
))
1559 * Drop the socket lock, otherwise we have reverse
1560 * locking dependencies between sk_lock and i_mutex
1561 * here as compared to sendfile(). We enter here
1562 * with the socket lock held, and splice_to_pipe() will
1563 * grab the pipe inode lock. For sendfile() emulation,
1564 * we call into ->sendpage() with the i_mutex lock held
1565 * and networking will grab the socket lock.
1568 ret
= splice_to_pipe(pipe
, &spd
);
1577 * skb_store_bits - store bits from kernel buffer to skb
1578 * @skb: destination buffer
1579 * @offset: offset in destination
1580 * @from: source buffer
1581 * @len: number of bytes to copy
1583 * Copy the specified number of bytes from the source buffer to the
1584 * destination skb. This function handles all the messy bits of
1585 * traversing fragment lists and such.
1588 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
1590 int start
= skb_headlen(skb
);
1591 struct sk_buff
*frag_iter
;
1594 if (offset
> (int)skb
->len
- len
)
1597 if ((copy
= start
- offset
) > 0) {
1600 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
1601 if ((len
-= copy
) == 0)
1607 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1608 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1611 WARN_ON(start
> offset
+ len
);
1613 end
= start
+ frag
->size
;
1614 if ((copy
= end
- offset
) > 0) {
1620 vaddr
= kmap_skb_frag(frag
);
1621 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1623 kunmap_skb_frag(vaddr
);
1625 if ((len
-= copy
) == 0)
1633 skb_walk_frags(skb
, frag_iter
) {
1636 WARN_ON(start
> offset
+ len
);
1638 end
= start
+ frag_iter
->len
;
1639 if ((copy
= end
- offset
) > 0) {
1642 if (skb_store_bits(frag_iter
, offset
- start
,
1645 if ((len
-= copy
) == 0)
1658 EXPORT_SYMBOL(skb_store_bits
);
1660 /* Checksum skb data. */
1662 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
1663 int len
, __wsum csum
)
1665 int start
= skb_headlen(skb
);
1666 int i
, copy
= start
- offset
;
1667 struct sk_buff
*frag_iter
;
1670 /* Checksum header. */
1674 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1675 if ((len
-= copy
) == 0)
1681 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1684 WARN_ON(start
> offset
+ len
);
1686 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1687 if ((copy
= end
- offset
) > 0) {
1690 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1694 vaddr
= kmap_skb_frag(frag
);
1695 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
1696 offset
- start
, copy
, 0);
1697 kunmap_skb_frag(vaddr
);
1698 csum
= csum_block_add(csum
, csum2
, pos
);
1707 skb_walk_frags(skb
, frag_iter
) {
1710 WARN_ON(start
> offset
+ len
);
1712 end
= start
+ frag_iter
->len
;
1713 if ((copy
= end
- offset
) > 0) {
1717 csum2
= skb_checksum(frag_iter
, offset
- start
,
1719 csum
= csum_block_add(csum
, csum2
, pos
);
1720 if ((len
-= copy
) == 0)
1731 EXPORT_SYMBOL(skb_checksum
);
1733 /* Both of above in one bottle. */
1735 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
1736 u8
*to
, int len
, __wsum csum
)
1738 int start
= skb_headlen(skb
);
1739 int i
, copy
= start
- offset
;
1740 struct sk_buff
*frag_iter
;
1747 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
1749 if ((len
-= copy
) == 0)
1756 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1759 WARN_ON(start
> offset
+ len
);
1761 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1762 if ((copy
= end
- offset
) > 0) {
1765 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1769 vaddr
= kmap_skb_frag(frag
);
1770 csum2
= csum_partial_copy_nocheck(vaddr
+
1774 kunmap_skb_frag(vaddr
);
1775 csum
= csum_block_add(csum
, csum2
, pos
);
1785 skb_walk_frags(skb
, frag_iter
) {
1789 WARN_ON(start
> offset
+ len
);
1791 end
= start
+ frag_iter
->len
;
1792 if ((copy
= end
- offset
) > 0) {
1795 csum2
= skb_copy_and_csum_bits(frag_iter
,
1798 csum
= csum_block_add(csum
, csum2
, pos
);
1799 if ((len
-= copy
) == 0)
1810 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
1812 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
1817 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1818 csstart
= skb
->csum_start
- skb_headroom(skb
);
1820 csstart
= skb_headlen(skb
);
1822 BUG_ON(csstart
> skb_headlen(skb
));
1824 skb_copy_from_linear_data(skb
, to
, csstart
);
1827 if (csstart
!= skb
->len
)
1828 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
1829 skb
->len
- csstart
, 0);
1831 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1832 long csstuff
= csstart
+ skb
->csum_offset
;
1834 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
1837 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
1840 * skb_dequeue - remove from the head of the queue
1841 * @list: list to dequeue from
1843 * Remove the head of the list. The list lock is taken so the function
1844 * may be used safely with other locking list functions. The head item is
1845 * returned or %NULL if the list is empty.
1848 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
1850 unsigned long flags
;
1851 struct sk_buff
*result
;
1853 spin_lock_irqsave(&list
->lock
, flags
);
1854 result
= __skb_dequeue(list
);
1855 spin_unlock_irqrestore(&list
->lock
, flags
);
1858 EXPORT_SYMBOL(skb_dequeue
);
1861 * skb_dequeue_tail - remove from the tail of the queue
1862 * @list: list to dequeue from
1864 * Remove the tail of the list. The list lock is taken so the function
1865 * may be used safely with other locking list functions. The tail item is
1866 * returned or %NULL if the list is empty.
1868 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
1870 unsigned long flags
;
1871 struct sk_buff
*result
;
1873 spin_lock_irqsave(&list
->lock
, flags
);
1874 result
= __skb_dequeue_tail(list
);
1875 spin_unlock_irqrestore(&list
->lock
, flags
);
1878 EXPORT_SYMBOL(skb_dequeue_tail
);
1881 * skb_queue_purge - empty a list
1882 * @list: list to empty
1884 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1885 * the list and one reference dropped. This function takes the list
1886 * lock and is atomic with respect to other list locking functions.
1888 void skb_queue_purge(struct sk_buff_head
*list
)
1890 struct sk_buff
*skb
;
1891 while ((skb
= skb_dequeue(list
)) != NULL
)
1894 EXPORT_SYMBOL(skb_queue_purge
);
1897 * skb_queue_head - queue a buffer at the list head
1898 * @list: list to use
1899 * @newsk: buffer to queue
1901 * Queue a buffer at the start of the list. This function takes the
1902 * list lock and can be used safely with other locking &sk_buff functions
1905 * A buffer cannot be placed on two lists at the same time.
1907 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1909 unsigned long flags
;
1911 spin_lock_irqsave(&list
->lock
, flags
);
1912 __skb_queue_head(list
, newsk
);
1913 spin_unlock_irqrestore(&list
->lock
, flags
);
1915 EXPORT_SYMBOL(skb_queue_head
);
1918 * skb_queue_tail - queue a buffer at the list tail
1919 * @list: list to use
1920 * @newsk: buffer to queue
1922 * Queue a buffer at the tail of the list. This function takes the
1923 * list lock and can be used safely with other locking &sk_buff functions
1926 * A buffer cannot be placed on two lists at the same time.
1928 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1930 unsigned long flags
;
1932 spin_lock_irqsave(&list
->lock
, flags
);
1933 __skb_queue_tail(list
, newsk
);
1934 spin_unlock_irqrestore(&list
->lock
, flags
);
1936 EXPORT_SYMBOL(skb_queue_tail
);
1939 * skb_unlink - remove a buffer from a list
1940 * @skb: buffer to remove
1941 * @list: list to use
1943 * Remove a packet from a list. The list locks are taken and this
1944 * function is atomic with respect to other list locked calls
1946 * You must know what list the SKB is on.
1948 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
1950 unsigned long flags
;
1952 spin_lock_irqsave(&list
->lock
, flags
);
1953 __skb_unlink(skb
, list
);
1954 spin_unlock_irqrestore(&list
->lock
, flags
);
1956 EXPORT_SYMBOL(skb_unlink
);
1959 * skb_append - append a buffer
1960 * @old: buffer to insert after
1961 * @newsk: buffer to insert
1962 * @list: list to use
1964 * Place a packet after a given packet in a list. The list locks are taken
1965 * and this function is atomic with respect to other list locked calls.
1966 * A buffer cannot be placed on two lists at the same time.
1968 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
1970 unsigned long flags
;
1972 spin_lock_irqsave(&list
->lock
, flags
);
1973 __skb_queue_after(list
, old
, newsk
);
1974 spin_unlock_irqrestore(&list
->lock
, flags
);
1976 EXPORT_SYMBOL(skb_append
);
1979 * skb_insert - insert a buffer
1980 * @old: buffer to insert before
1981 * @newsk: buffer to insert
1982 * @list: list to use
1984 * Place a packet before a given packet in a list. The list locks are
1985 * taken and this function is atomic with respect to other list locked
1988 * A buffer cannot be placed on two lists at the same time.
1990 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
1992 unsigned long flags
;
1994 spin_lock_irqsave(&list
->lock
, flags
);
1995 __skb_insert(newsk
, old
->prev
, old
, list
);
1996 spin_unlock_irqrestore(&list
->lock
, flags
);
1998 EXPORT_SYMBOL(skb_insert
);
2000 static inline void skb_split_inside_header(struct sk_buff
*skb
,
2001 struct sk_buff
* skb1
,
2002 const u32 len
, const int pos
)
2006 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
2008 /* And move data appendix as is. */
2009 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
2010 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
2012 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
2013 skb_shinfo(skb
)->nr_frags
= 0;
2014 skb1
->data_len
= skb
->data_len
;
2015 skb1
->len
+= skb1
->data_len
;
2018 skb_set_tail_pointer(skb
, len
);
2021 static inline void skb_split_no_header(struct sk_buff
*skb
,
2022 struct sk_buff
* skb1
,
2023 const u32 len
, int pos
)
2026 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
2028 skb_shinfo(skb
)->nr_frags
= 0;
2029 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
2031 skb
->data_len
= len
- pos
;
2033 for (i
= 0; i
< nfrags
; i
++) {
2034 int size
= skb_shinfo(skb
)->frags
[i
].size
;
2036 if (pos
+ size
> len
) {
2037 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
2041 * We have two variants in this case:
2042 * 1. Move all the frag to the second
2043 * part, if it is possible. F.e.
2044 * this approach is mandatory for TUX,
2045 * where splitting is expensive.
2046 * 2. Split is accurately. We make this.
2048 get_page(skb_shinfo(skb
)->frags
[i
].page
);
2049 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
2050 skb_shinfo(skb1
)->frags
[0].size
-= len
- pos
;
2051 skb_shinfo(skb
)->frags
[i
].size
= len
- pos
;
2052 skb_shinfo(skb
)->nr_frags
++;
2056 skb_shinfo(skb
)->nr_frags
++;
2059 skb_shinfo(skb1
)->nr_frags
= k
;
2063 * skb_split - Split fragmented skb to two parts at length len.
2064 * @skb: the buffer to split
2065 * @skb1: the buffer to receive the second part
2066 * @len: new length for skb
2068 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
2070 int pos
= skb_headlen(skb
);
2072 if (len
< pos
) /* Split line is inside header. */
2073 skb_split_inside_header(skb
, skb1
, len
, pos
);
2074 else /* Second chunk has no header, nothing to copy. */
2075 skb_split_no_header(skb
, skb1
, len
, pos
);
2077 EXPORT_SYMBOL(skb_split
);
2079 /* Shifting from/to a cloned skb is a no-go.
2081 * Caller cannot keep skb_shinfo related pointers past calling here!
2083 static int skb_prepare_for_shift(struct sk_buff
*skb
)
2085 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2089 * skb_shift - Shifts paged data partially from skb to another
2090 * @tgt: buffer into which tail data gets added
2091 * @skb: buffer from which the paged data comes from
2092 * @shiftlen: shift up to this many bytes
2094 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2095 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2096 * It's up to caller to free skb if everything was shifted.
2098 * If @tgt runs out of frags, the whole operation is aborted.
2100 * Skb cannot include anything else but paged data while tgt is allowed
2101 * to have non-paged data as well.
2103 * TODO: full sized shift could be optimized but that would need
2104 * specialized skb free'er to handle frags without up-to-date nr_frags.
2106 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
2108 int from
, to
, merge
, todo
;
2109 struct skb_frag_struct
*fragfrom
, *fragto
;
2111 BUG_ON(shiftlen
> skb
->len
);
2112 BUG_ON(skb_headlen(skb
)); /* Would corrupt stream */
2116 to
= skb_shinfo(tgt
)->nr_frags
;
2117 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2119 /* Actual merge is delayed until the point when we know we can
2120 * commit all, so that we don't have to undo partial changes
2123 !skb_can_coalesce(tgt
, to
, fragfrom
->page
, fragfrom
->page_offset
)) {
2128 todo
-= fragfrom
->size
;
2130 if (skb_prepare_for_shift(skb
) ||
2131 skb_prepare_for_shift(tgt
))
2134 /* All previous frag pointers might be stale! */
2135 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2136 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2138 fragto
->size
+= shiftlen
;
2139 fragfrom
->size
-= shiftlen
;
2140 fragfrom
->page_offset
+= shiftlen
;
2148 /* Skip full, not-fitting skb to avoid expensive operations */
2149 if ((shiftlen
== skb
->len
) &&
2150 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
2153 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
2156 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
2157 if (to
== MAX_SKB_FRAGS
)
2160 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2161 fragto
= &skb_shinfo(tgt
)->frags
[to
];
2163 if (todo
>= fragfrom
->size
) {
2164 *fragto
= *fragfrom
;
2165 todo
-= fragfrom
->size
;
2170 get_page(fragfrom
->page
);
2171 fragto
->page
= fragfrom
->page
;
2172 fragto
->page_offset
= fragfrom
->page_offset
;
2173 fragto
->size
= todo
;
2175 fragfrom
->page_offset
+= todo
;
2176 fragfrom
->size
-= todo
;
2184 /* Ready to "commit" this state change to tgt */
2185 skb_shinfo(tgt
)->nr_frags
= to
;
2188 fragfrom
= &skb_shinfo(skb
)->frags
[0];
2189 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2191 fragto
->size
+= fragfrom
->size
;
2192 put_page(fragfrom
->page
);
2195 /* Reposition in the original skb */
2197 while (from
< skb_shinfo(skb
)->nr_frags
)
2198 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
2199 skb_shinfo(skb
)->nr_frags
= to
;
2201 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
2204 /* Most likely the tgt won't ever need its checksum anymore, skb on
2205 * the other hand might need it if it needs to be resent
2207 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
2208 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2210 /* Yak, is it really working this way? Some helper please? */
2211 skb
->len
-= shiftlen
;
2212 skb
->data_len
-= shiftlen
;
2213 skb
->truesize
-= shiftlen
;
2214 tgt
->len
+= shiftlen
;
2215 tgt
->data_len
+= shiftlen
;
2216 tgt
->truesize
+= shiftlen
;
2222 * skb_prepare_seq_read - Prepare a sequential read of skb data
2223 * @skb: the buffer to read
2224 * @from: lower offset of data to be read
2225 * @to: upper offset of data to be read
2226 * @st: state variable
2228 * Initializes the specified state variable. Must be called before
2229 * invoking skb_seq_read() for the first time.
2231 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
2232 unsigned int to
, struct skb_seq_state
*st
)
2234 st
->lower_offset
= from
;
2235 st
->upper_offset
= to
;
2236 st
->root_skb
= st
->cur_skb
= skb
;
2237 st
->frag_idx
= st
->stepped_offset
= 0;
2238 st
->frag_data
= NULL
;
2240 EXPORT_SYMBOL(skb_prepare_seq_read
);
2243 * skb_seq_read - Sequentially read skb data
2244 * @consumed: number of bytes consumed by the caller so far
2245 * @data: destination pointer for data to be returned
2246 * @st: state variable
2248 * Reads a block of skb data at &consumed relative to the
2249 * lower offset specified to skb_prepare_seq_read(). Assigns
2250 * the head of the data block to &data and returns the length
2251 * of the block or 0 if the end of the skb data or the upper
2252 * offset has been reached.
2254 * The caller is not required to consume all of the data
2255 * returned, i.e. &consumed is typically set to the number
2256 * of bytes already consumed and the next call to
2257 * skb_seq_read() will return the remaining part of the block.
2259 * Note 1: The size of each block of data returned can be arbitary,
2260 * this limitation is the cost for zerocopy seqeuental
2261 * reads of potentially non linear data.
2263 * Note 2: Fragment lists within fragments are not implemented
2264 * at the moment, state->root_skb could be replaced with
2265 * a stack for this purpose.
2267 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
2268 struct skb_seq_state
*st
)
2270 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
2273 if (unlikely(abs_offset
>= st
->upper_offset
))
2277 block_limit
= skb_headlen(st
->cur_skb
) + st
->stepped_offset
;
2279 if (abs_offset
< block_limit
&& !st
->frag_data
) {
2280 *data
= st
->cur_skb
->data
+ (abs_offset
- st
->stepped_offset
);
2281 return block_limit
- abs_offset
;
2284 if (st
->frag_idx
== 0 && !st
->frag_data
)
2285 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
2287 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
2288 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
2289 block_limit
= frag
->size
+ st
->stepped_offset
;
2291 if (abs_offset
< block_limit
) {
2293 st
->frag_data
= kmap_skb_frag(frag
);
2295 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
2296 (abs_offset
- st
->stepped_offset
);
2298 return block_limit
- abs_offset
;
2301 if (st
->frag_data
) {
2302 kunmap_skb_frag(st
->frag_data
);
2303 st
->frag_data
= NULL
;
2307 st
->stepped_offset
+= frag
->size
;
2310 if (st
->frag_data
) {
2311 kunmap_skb_frag(st
->frag_data
);
2312 st
->frag_data
= NULL
;
2315 if (st
->root_skb
== st
->cur_skb
&& skb_has_frags(st
->root_skb
)) {
2316 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
2319 } else if (st
->cur_skb
->next
) {
2320 st
->cur_skb
= st
->cur_skb
->next
;
2327 EXPORT_SYMBOL(skb_seq_read
);
2330 * skb_abort_seq_read - Abort a sequential read of skb data
2331 * @st: state variable
2333 * Must be called if skb_seq_read() was not called until it
2336 void skb_abort_seq_read(struct skb_seq_state
*st
)
2339 kunmap_skb_frag(st
->frag_data
);
2341 EXPORT_SYMBOL(skb_abort_seq_read
);
2343 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2345 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
2346 struct ts_config
*conf
,
2347 struct ts_state
*state
)
2349 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
2352 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
2354 skb_abort_seq_read(TS_SKB_CB(state
));
2358 * skb_find_text - Find a text pattern in skb data
2359 * @skb: the buffer to look in
2360 * @from: search offset
2362 * @config: textsearch configuration
2363 * @state: uninitialized textsearch state variable
2365 * Finds a pattern in the skb data according to the specified
2366 * textsearch configuration. Use textsearch_next() to retrieve
2367 * subsequent occurrences of the pattern. Returns the offset
2368 * to the first occurrence or UINT_MAX if no match was found.
2370 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
2371 unsigned int to
, struct ts_config
*config
,
2372 struct ts_state
*state
)
2376 config
->get_next_block
= skb_ts_get_next_block
;
2377 config
->finish
= skb_ts_finish
;
2379 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
2381 ret
= textsearch_find(config
, state
);
2382 return (ret
<= to
- from
? ret
: UINT_MAX
);
2384 EXPORT_SYMBOL(skb_find_text
);
2387 * skb_append_datato_frags: - append the user data to a skb
2388 * @sk: sock structure
2389 * @skb: skb structure to be appened with user data.
2390 * @getfrag: call back function to be used for getting the user data
2391 * @from: pointer to user message iov
2392 * @length: length of the iov message
2394 * Description: This procedure append the user data in the fragment part
2395 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2397 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
2398 int (*getfrag
)(void *from
, char *to
, int offset
,
2399 int len
, int odd
, struct sk_buff
*skb
),
2400 void *from
, int length
)
2403 skb_frag_t
*frag
= NULL
;
2404 struct page
*page
= NULL
;
2410 /* Return error if we don't have space for new frag */
2411 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2412 if (frg_cnt
>= MAX_SKB_FRAGS
)
2415 /* allocate a new page for next frag */
2416 page
= alloc_pages(sk
->sk_allocation
, 0);
2418 /* If alloc_page fails just return failure and caller will
2419 * free previous allocated pages by doing kfree_skb()
2424 /* initialize the next frag */
2425 sk
->sk_sndmsg_page
= page
;
2426 sk
->sk_sndmsg_off
= 0;
2427 skb_fill_page_desc(skb
, frg_cnt
, page
, 0, 0);
2428 skb
->truesize
+= PAGE_SIZE
;
2429 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
2431 /* get the new initialized frag */
2432 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2433 frag
= &skb_shinfo(skb
)->frags
[frg_cnt
- 1];
2435 /* copy the user data to page */
2436 left
= PAGE_SIZE
- frag
->page_offset
;
2437 copy
= (length
> left
)? left
: length
;
2439 ret
= getfrag(from
, (page_address(frag
->page
) +
2440 frag
->page_offset
+ frag
->size
),
2441 offset
, copy
, 0, skb
);
2445 /* copy was successful so update the size parameters */
2446 sk
->sk_sndmsg_off
+= copy
;
2449 skb
->data_len
+= copy
;
2453 } while (length
> 0);
2457 EXPORT_SYMBOL(skb_append_datato_frags
);
2460 * skb_pull_rcsum - pull skb and update receive checksum
2461 * @skb: buffer to update
2462 * @len: length of data pulled
2464 * This function performs an skb_pull on the packet and updates
2465 * the CHECKSUM_COMPLETE checksum. It should be used on
2466 * receive path processing instead of skb_pull unless you know
2467 * that the checksum difference is zero (e.g., a valid IP header)
2468 * or you are setting ip_summed to CHECKSUM_NONE.
2470 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
2472 BUG_ON(len
> skb
->len
);
2474 BUG_ON(skb
->len
< skb
->data_len
);
2475 skb_postpull_rcsum(skb
, skb
->data
, len
);
2476 return skb
->data
+= len
;
2479 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
2482 * skb_segment - Perform protocol segmentation on skb.
2483 * @skb: buffer to segment
2484 * @features: features for the output path (see dev->features)
2486 * This function performs segmentation on the given skb. It returns
2487 * a pointer to the first in a list of new skbs for the segments.
2488 * In case of error it returns ERR_PTR(err).
2490 struct sk_buff
*skb_segment(struct sk_buff
*skb
, int features
)
2492 struct sk_buff
*segs
= NULL
;
2493 struct sk_buff
*tail
= NULL
;
2494 struct sk_buff
*fskb
= skb_shinfo(skb
)->frag_list
;
2495 unsigned int mss
= skb_shinfo(skb
)->gso_size
;
2496 unsigned int doffset
= skb
->data
- skb_mac_header(skb
);
2497 unsigned int offset
= doffset
;
2498 unsigned int headroom
;
2500 int sg
= features
& NETIF_F_SG
;
2501 int nfrags
= skb_shinfo(skb
)->nr_frags
;
2506 __skb_push(skb
, doffset
);
2507 headroom
= skb_headroom(skb
);
2508 pos
= skb_headlen(skb
);
2511 struct sk_buff
*nskb
;
2516 len
= skb
->len
- offset
;
2520 hsize
= skb_headlen(skb
) - offset
;
2523 if (hsize
> len
|| !sg
)
2526 if (!hsize
&& i
>= nfrags
) {
2527 BUG_ON(fskb
->len
!= len
);
2530 nskb
= skb_clone(fskb
, GFP_ATOMIC
);
2533 if (unlikely(!nskb
))
2536 hsize
= skb_end_pointer(nskb
) - nskb
->head
;
2537 if (skb_cow_head(nskb
, doffset
+ headroom
)) {
2542 nskb
->truesize
+= skb_end_pointer(nskb
) - nskb
->head
-
2544 skb_release_head_state(nskb
);
2545 __skb_push(nskb
, doffset
);
2547 nskb
= alloc_skb(hsize
+ doffset
+ headroom
,
2550 if (unlikely(!nskb
))
2553 skb_reserve(nskb
, headroom
);
2554 __skb_put(nskb
, doffset
);
2563 __copy_skb_header(nskb
, skb
);
2564 nskb
->mac_len
= skb
->mac_len
;
2566 skb_reset_mac_header(nskb
);
2567 skb_set_network_header(nskb
, skb
->mac_len
);
2568 nskb
->transport_header
= (nskb
->network_header
+
2569 skb_network_header_len(skb
));
2570 skb_copy_from_linear_data(skb
, nskb
->data
, doffset
);
2572 if (fskb
!= skb_shinfo(skb
)->frag_list
)
2576 nskb
->ip_summed
= CHECKSUM_NONE
;
2577 nskb
->csum
= skb_copy_and_csum_bits(skb
, offset
,
2583 frag
= skb_shinfo(nskb
)->frags
;
2585 skb_copy_from_linear_data_offset(skb
, offset
,
2586 skb_put(nskb
, hsize
), hsize
);
2588 while (pos
< offset
+ len
&& i
< nfrags
) {
2589 *frag
= skb_shinfo(skb
)->frags
[i
];
2590 get_page(frag
->page
);
2594 frag
->page_offset
+= offset
- pos
;
2595 frag
->size
-= offset
- pos
;
2598 skb_shinfo(nskb
)->nr_frags
++;
2600 if (pos
+ size
<= offset
+ len
) {
2604 frag
->size
-= pos
+ size
- (offset
+ len
);
2611 if (pos
< offset
+ len
) {
2612 struct sk_buff
*fskb2
= fskb
;
2614 BUG_ON(pos
+ fskb
->len
!= offset
+ len
);
2620 fskb2
= skb_clone(fskb2
, GFP_ATOMIC
);
2626 SKB_FRAG_ASSERT(nskb
);
2627 skb_shinfo(nskb
)->frag_list
= fskb2
;
2631 nskb
->data_len
= len
- hsize
;
2632 nskb
->len
+= nskb
->data_len
;
2633 nskb
->truesize
+= nskb
->data_len
;
2634 } while ((offset
+= len
) < skb
->len
);
2639 while ((skb
= segs
)) {
2643 return ERR_PTR(err
);
2645 EXPORT_SYMBOL_GPL(skb_segment
);
2647 int skb_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2649 struct sk_buff
*p
= *head
;
2650 struct sk_buff
*nskb
;
2651 struct skb_shared_info
*skbinfo
= skb_shinfo(skb
);
2652 struct skb_shared_info
*pinfo
= skb_shinfo(p
);
2653 unsigned int headroom
;
2654 unsigned int len
= skb_gro_len(skb
);
2655 unsigned int offset
= skb_gro_offset(skb
);
2656 unsigned int headlen
= skb_headlen(skb
);
2658 if (p
->len
+ len
>= 65536)
2661 if (pinfo
->frag_list
)
2663 else if (headlen
<= offset
) {
2666 int i
= skbinfo
->nr_frags
;
2667 int nr_frags
= pinfo
->nr_frags
+ i
;
2671 if (nr_frags
> MAX_SKB_FRAGS
)
2674 pinfo
->nr_frags
= nr_frags
;
2675 skbinfo
->nr_frags
= 0;
2677 frag
= pinfo
->frags
+ nr_frags
;
2678 frag2
= skbinfo
->frags
+ i
;
2683 frag
->page_offset
+= offset
;
2684 frag
->size
-= offset
;
2686 skb
->truesize
-= skb
->data_len
;
2687 skb
->len
-= skb
->data_len
;
2690 NAPI_GRO_CB(skb
)->free
= 1;
2694 headroom
= skb_headroom(p
);
2695 nskb
= netdev_alloc_skb(p
->dev
, headroom
+ skb_gro_offset(p
));
2696 if (unlikely(!nskb
))
2699 __copy_skb_header(nskb
, p
);
2700 nskb
->mac_len
= p
->mac_len
;
2702 skb_reserve(nskb
, headroom
);
2703 __skb_put(nskb
, skb_gro_offset(p
));
2705 skb_set_mac_header(nskb
, skb_mac_header(p
) - p
->data
);
2706 skb_set_network_header(nskb
, skb_network_offset(p
));
2707 skb_set_transport_header(nskb
, skb_transport_offset(p
));
2709 __skb_pull(p
, skb_gro_offset(p
));
2710 memcpy(skb_mac_header(nskb
), skb_mac_header(p
),
2711 p
->data
- skb_mac_header(p
));
2713 *NAPI_GRO_CB(nskb
) = *NAPI_GRO_CB(p
);
2714 skb_shinfo(nskb
)->frag_list
= p
;
2715 skb_shinfo(nskb
)->gso_size
= pinfo
->gso_size
;
2716 skb_header_release(p
);
2719 nskb
->data_len
+= p
->len
;
2720 nskb
->truesize
+= p
->len
;
2721 nskb
->len
+= p
->len
;
2724 nskb
->next
= p
->next
;
2730 if (offset
> headlen
) {
2731 skbinfo
->frags
[0].page_offset
+= offset
- headlen
;
2732 skbinfo
->frags
[0].size
-= offset
- headlen
;
2736 __skb_pull(skb
, offset
);
2738 p
->prev
->next
= skb
;
2740 skb_header_release(skb
);
2743 NAPI_GRO_CB(p
)->count
++;
2748 NAPI_GRO_CB(skb
)->same_flow
= 1;
2751 EXPORT_SYMBOL_GPL(skb_gro_receive
);
2753 void __init
skb_init(void)
2755 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
2756 sizeof(struct sk_buff
),
2758 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2760 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
2761 (2*sizeof(struct sk_buff
)) +
2764 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2769 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2770 * @skb: Socket buffer containing the buffers to be mapped
2771 * @sg: The scatter-gather list to map into
2772 * @offset: The offset into the buffer's contents to start mapping
2773 * @len: Length of buffer space to be mapped
2775 * Fill the specified scatter-gather list with mappings/pointers into a
2776 * region of the buffer space attached to a socket buffer.
2779 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2781 int start
= skb_headlen(skb
);
2782 int i
, copy
= start
- offset
;
2783 struct sk_buff
*frag_iter
;
2789 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
2791 if ((len
-= copy
) == 0)
2796 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2799 WARN_ON(start
> offset
+ len
);
2801 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
2802 if ((copy
= end
- offset
) > 0) {
2803 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2807 sg_set_page(&sg
[elt
], frag
->page
, copy
,
2808 frag
->page_offset
+offset
-start
);
2817 skb_walk_frags(skb
, frag_iter
) {
2820 WARN_ON(start
> offset
+ len
);
2822 end
= start
+ frag_iter
->len
;
2823 if ((copy
= end
- offset
) > 0) {
2826 elt
+= __skb_to_sgvec(frag_iter
, sg
+elt
, offset
- start
,
2828 if ((len
-= copy
) == 0)
2838 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2840 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
);
2842 sg_mark_end(&sg
[nsg
- 1]);
2846 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
2849 * skb_cow_data - Check that a socket buffer's data buffers are writable
2850 * @skb: The socket buffer to check.
2851 * @tailbits: Amount of trailing space to be added
2852 * @trailer: Returned pointer to the skb where the @tailbits space begins
2854 * Make sure that the data buffers attached to a socket buffer are
2855 * writable. If they are not, private copies are made of the data buffers
2856 * and the socket buffer is set to use these instead.
2858 * If @tailbits is given, make sure that there is space to write @tailbits
2859 * bytes of data beyond current end of socket buffer. @trailer will be
2860 * set to point to the skb in which this space begins.
2862 * The number of scatterlist elements required to completely map the
2863 * COW'd and extended socket buffer will be returned.
2865 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
2869 struct sk_buff
*skb1
, **skb_p
;
2871 /* If skb is cloned or its head is paged, reallocate
2872 * head pulling out all the pages (pages are considered not writable
2873 * at the moment even if they are anonymous).
2875 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
2876 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
2879 /* Easy case. Most of packets will go this way. */
2880 if (!skb_has_frags(skb
)) {
2881 /* A little of trouble, not enough of space for trailer.
2882 * This should not happen, when stack is tuned to generate
2883 * good frames. OK, on miss we reallocate and reserve even more
2884 * space, 128 bytes is fair. */
2886 if (skb_tailroom(skb
) < tailbits
&&
2887 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
2895 /* Misery. We are in troubles, going to mincer fragments... */
2898 skb_p
= &skb_shinfo(skb
)->frag_list
;
2901 while ((skb1
= *skb_p
) != NULL
) {
2904 /* The fragment is partially pulled by someone,
2905 * this can happen on input. Copy it and everything
2908 if (skb_shared(skb1
))
2911 /* If the skb is the last, worry about trailer. */
2913 if (skb1
->next
== NULL
&& tailbits
) {
2914 if (skb_shinfo(skb1
)->nr_frags
||
2915 skb_has_frags(skb1
) ||
2916 skb_tailroom(skb1
) < tailbits
)
2917 ntail
= tailbits
+ 128;
2923 skb_shinfo(skb1
)->nr_frags
||
2924 skb_has_frags(skb1
)) {
2925 struct sk_buff
*skb2
;
2927 /* Fuck, we are miserable poor guys... */
2929 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
2931 skb2
= skb_copy_expand(skb1
,
2935 if (unlikely(skb2
== NULL
))
2939 skb_set_owner_w(skb2
, skb1
->sk
);
2941 /* Looking around. Are we still alive?
2942 * OK, link new skb, drop old one */
2944 skb2
->next
= skb1
->next
;
2951 skb_p
= &skb1
->next
;
2956 EXPORT_SYMBOL_GPL(skb_cow_data
);
2958 void skb_tstamp_tx(struct sk_buff
*orig_skb
,
2959 struct skb_shared_hwtstamps
*hwtstamps
)
2961 struct sock
*sk
= orig_skb
->sk
;
2962 struct sock_exterr_skb
*serr
;
2963 struct sk_buff
*skb
;
2969 skb
= skb_clone(orig_skb
, GFP_ATOMIC
);
2974 *skb_hwtstamps(skb
) =
2978 * no hardware time stamps available,
2979 * so keep the skb_shared_tx and only
2980 * store software time stamp
2982 skb
->tstamp
= ktime_get_real();
2985 serr
= SKB_EXT_ERR(skb
);
2986 memset(serr
, 0, sizeof(*serr
));
2987 serr
->ee
.ee_errno
= ENOMSG
;
2988 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TIMESTAMPING
;
2989 err
= sock_queue_err_skb(sk
, skb
);
2993 EXPORT_SYMBOL_GPL(skb_tstamp_tx
);
2997 * skb_partial_csum_set - set up and verify partial csum values for packet
2998 * @skb: the skb to set
2999 * @start: the number of bytes after skb->data to start checksumming.
3000 * @off: the offset from start to place the checksum.
3002 * For untrusted partially-checksummed packets, we need to make sure the values
3003 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3005 * This function checks and sets those values and skb->ip_summed: if this
3006 * returns false you should drop the packet.
3008 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
3010 if (unlikely(start
> skb_headlen(skb
)) ||
3011 unlikely((int)start
+ off
> skb_headlen(skb
) - 2)) {
3012 if (net_ratelimit())
3014 "bad partial csum: csum=%u/%u len=%u\n",
3015 start
, off
, skb_headlen(skb
));
3018 skb
->ip_summed
= CHECKSUM_PARTIAL
;
3019 skb
->csum_start
= skb_headroom(skb
) + start
;
3020 skb
->csum_offset
= off
;
3023 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);
3025 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
3027 if (net_ratelimit())
3028 pr_warning("%s: received packets cannot be forwarded"
3029 " while LRO is enabled\n", skb
->dev
->name
);
3031 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);