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 static 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>");
131 * skb_under_panic - private function
136 * Out of line support code for skb_push(). Not user callable.
139 static 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>");
149 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
150 * 'private' fields and also do memory statistics to find all the
156 * __alloc_skb - allocate a network buffer
157 * @size: size to allocate
158 * @gfp_mask: allocation mask
159 * @fclone: allocate from fclone cache instead of head cache
160 * and allocate a cloned (child) skb
161 * @node: numa node to allocate memory on
163 * Allocate a new &sk_buff. The returned buffer has no headroom and a
164 * tail room of size bytes. The object has a reference count of one.
165 * The return is the buffer. On a failure the return is %NULL.
167 * Buffers may only be allocated from interrupts using a @gfp_mask of
170 struct sk_buff
*__alloc_skb(unsigned int size
, gfp_t gfp_mask
,
171 int fclone
, int node
)
173 struct kmem_cache
*cache
;
174 struct skb_shared_info
*shinfo
;
178 cache
= fclone
? skbuff_fclone_cache
: skbuff_head_cache
;
181 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
),
191 prefetchw(data
+ size
);
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 #ifdef NET_SKBUFF_DATA_USES_OFFSET
206 skb
->mac_header
= ~0U;
209 /* make sure we initialize shinfo sequentially */
210 shinfo
= skb_shinfo(skb
);
211 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
212 atomic_set(&shinfo
->dataref
, 1);
215 struct sk_buff
*child
= skb
+ 1;
216 atomic_t
*fclone_ref
= (atomic_t
*) (child
+ 1);
218 kmemcheck_annotate_bitfield(child
, flags1
);
219 kmemcheck_annotate_bitfield(child
, flags2
);
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
)
252 skb
= __alloc_skb(length
+ NET_SKB_PAD
, gfp_mask
, 0, NUMA_NO_NODE
);
254 skb_reserve(skb
, NET_SKB_PAD
);
259 EXPORT_SYMBOL(__netdev_alloc_skb
);
261 void skb_add_rx_frag(struct sk_buff
*skb
, int i
, struct page
*page
, int off
,
264 skb_fill_page_desc(skb
, i
, page
, off
, size
);
266 skb
->data_len
+= size
;
267 skb
->truesize
+= size
;
269 EXPORT_SYMBOL(skb_add_rx_frag
);
272 * dev_alloc_skb - allocate an skbuff for receiving
273 * @length: length to allocate
275 * Allocate a new &sk_buff and assign it a usage count of one. The
276 * buffer has unspecified headroom built in. Users should allocate
277 * the headroom they think they need without accounting for the
278 * built in space. The built in space is used for optimisations.
280 * %NULL is returned if there is no free memory. Although this function
281 * allocates memory it can be called from an interrupt.
283 struct sk_buff
*dev_alloc_skb(unsigned int length
)
286 * There is more code here than it seems:
287 * __dev_alloc_skb is an inline
289 return __dev_alloc_skb(length
, GFP_ATOMIC
);
291 EXPORT_SYMBOL(dev_alloc_skb
);
293 static void skb_drop_list(struct sk_buff
**listp
)
295 struct sk_buff
*list
= *listp
;
300 struct sk_buff
*this = list
;
306 static inline void skb_drop_fraglist(struct sk_buff
*skb
)
308 skb_drop_list(&skb_shinfo(skb
)->frag_list
);
311 static void skb_clone_fraglist(struct sk_buff
*skb
)
313 struct sk_buff
*list
;
315 skb_walk_frags(skb
, list
)
319 static void skb_release_data(struct sk_buff
*skb
)
322 !atomic_sub_return(skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1,
323 &skb_shinfo(skb
)->dataref
)) {
324 if (skb_shinfo(skb
)->nr_frags
) {
326 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
327 put_page(skb_shinfo(skb
)->frags
[i
].page
);
330 if (skb_has_frag_list(skb
))
331 skb_drop_fraglist(skb
);
338 * Free an skbuff by memory without cleaning the state.
340 static void kfree_skbmem(struct sk_buff
*skb
)
342 struct sk_buff
*other
;
343 atomic_t
*fclone_ref
;
345 switch (skb
->fclone
) {
346 case SKB_FCLONE_UNAVAILABLE
:
347 kmem_cache_free(skbuff_head_cache
, skb
);
350 case SKB_FCLONE_ORIG
:
351 fclone_ref
= (atomic_t
*) (skb
+ 2);
352 if (atomic_dec_and_test(fclone_ref
))
353 kmem_cache_free(skbuff_fclone_cache
, skb
);
356 case SKB_FCLONE_CLONE
:
357 fclone_ref
= (atomic_t
*) (skb
+ 1);
360 /* The clone portion is available for
361 * fast-cloning again.
363 skb
->fclone
= SKB_FCLONE_UNAVAILABLE
;
365 if (atomic_dec_and_test(fclone_ref
))
366 kmem_cache_free(skbuff_fclone_cache
, other
);
371 static void skb_release_head_state(struct sk_buff
*skb
)
375 secpath_put(skb
->sp
);
377 if (skb
->destructor
) {
379 skb
->destructor(skb
);
381 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
382 nf_conntrack_put(skb
->nfct
);
383 nf_conntrack_put_reasm(skb
->nfct_reasm
);
385 #ifdef CONFIG_BRIDGE_NETFILTER
386 nf_bridge_put(skb
->nf_bridge
);
388 /* XXX: IS this still necessary? - JHS */
389 #ifdef CONFIG_NET_SCHED
391 #ifdef CONFIG_NET_CLS_ACT
397 /* Free everything but the sk_buff shell. */
398 static void skb_release_all(struct sk_buff
*skb
)
400 skb_release_head_state(skb
);
401 skb_release_data(skb
);
405 * __kfree_skb - private function
408 * Free an sk_buff. Release anything attached to the buffer.
409 * Clean the state. This is an internal helper function. Users should
410 * always call kfree_skb
413 void __kfree_skb(struct sk_buff
*skb
)
415 skb_release_all(skb
);
418 EXPORT_SYMBOL(__kfree_skb
);
421 * kfree_skb - free an sk_buff
422 * @skb: buffer to free
424 * Drop a reference to the buffer and free it if the usage count has
427 void kfree_skb(struct sk_buff
*skb
)
431 if (likely(atomic_read(&skb
->users
) == 1))
433 else if (likely(!atomic_dec_and_test(&skb
->users
)))
435 trace_kfree_skb(skb
, __builtin_return_address(0));
438 EXPORT_SYMBOL(kfree_skb
);
441 * consume_skb - free an skbuff
442 * @skb: buffer to free
444 * Drop a ref to the buffer and free it if the usage count has hit zero
445 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
446 * is being dropped after a failure and notes that
448 void consume_skb(struct sk_buff
*skb
)
452 if (likely(atomic_read(&skb
->users
) == 1))
454 else if (likely(!atomic_dec_and_test(&skb
->users
)))
456 trace_consume_skb(skb
);
459 EXPORT_SYMBOL(consume_skb
);
462 * skb_recycle_check - check if skb can be reused for receive
464 * @skb_size: minimum receive buffer size
466 * Checks that the skb passed in is not shared or cloned, and
467 * that it is linear and its head portion at least as large as
468 * skb_size so that it can be recycled as a receive buffer.
469 * If these conditions are met, this function does any necessary
470 * reference count dropping and cleans up the skbuff as if it
471 * just came from __alloc_skb().
473 bool skb_recycle_check(struct sk_buff
*skb
, int skb_size
)
475 struct skb_shared_info
*shinfo
;
480 if (skb_is_nonlinear(skb
) || skb
->fclone
!= SKB_FCLONE_UNAVAILABLE
)
483 skb_size
= SKB_DATA_ALIGN(skb_size
+ NET_SKB_PAD
);
484 if (skb_end_pointer(skb
) - skb
->head
< skb_size
)
487 if (skb_shared(skb
) || skb_cloned(skb
))
490 skb_release_head_state(skb
);
492 shinfo
= skb_shinfo(skb
);
493 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
494 atomic_set(&shinfo
->dataref
, 1);
496 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
497 skb
->data
= skb
->head
+ NET_SKB_PAD
;
498 skb_reset_tail_pointer(skb
);
502 EXPORT_SYMBOL(skb_recycle_check
);
504 static void __copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
506 new->tstamp
= old
->tstamp
;
508 new->transport_header
= old
->transport_header
;
509 new->network_header
= old
->network_header
;
510 new->mac_header
= old
->mac_header
;
511 skb_dst_copy(new, old
);
512 new->rxhash
= old
->rxhash
;
514 new->sp
= secpath_get(old
->sp
);
516 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
517 new->csum
= old
->csum
;
518 new->local_df
= old
->local_df
;
519 new->pkt_type
= old
->pkt_type
;
520 new->ip_summed
= old
->ip_summed
;
521 skb_copy_queue_mapping(new, old
);
522 new->priority
= old
->priority
;
523 new->deliver_no_wcard
= old
->deliver_no_wcard
;
524 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
525 new->ipvs_property
= old
->ipvs_property
;
527 new->protocol
= old
->protocol
;
528 new->mark
= old
->mark
;
529 new->skb_iif
= old
->skb_iif
;
531 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
532 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
533 new->nf_trace
= old
->nf_trace
;
535 #ifdef CONFIG_NET_SCHED
536 new->tc_index
= old
->tc_index
;
537 #ifdef CONFIG_NET_CLS_ACT
538 new->tc_verd
= old
->tc_verd
;
541 new->vlan_tci
= old
->vlan_tci
;
543 skb_copy_secmark(new, old
);
547 * You should not add any new code to this function. Add it to
548 * __copy_skb_header above instead.
550 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
552 #define C(x) n->x = skb->x
554 n
->next
= n
->prev
= NULL
;
556 __copy_skb_header(n
, skb
);
561 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
564 n
->destructor
= NULL
;
570 atomic_set(&n
->users
, 1);
572 atomic_inc(&(skb_shinfo(skb
)->dataref
));
580 * skb_morph - morph one skb into another
581 * @dst: the skb to receive the contents
582 * @src: the skb to supply the contents
584 * This is identical to skb_clone except that the target skb is
585 * supplied by the user.
587 * The target skb is returned upon exit.
589 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
591 skb_release_all(dst
);
592 return __skb_clone(dst
, src
);
594 EXPORT_SYMBOL_GPL(skb_morph
);
597 * skb_clone - duplicate an sk_buff
598 * @skb: buffer to clone
599 * @gfp_mask: allocation priority
601 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
602 * copies share the same packet data but not structure. The new
603 * buffer has a reference count of 1. If the allocation fails the
604 * function returns %NULL otherwise the new buffer is returned.
606 * If this function is called from an interrupt gfp_mask() must be
610 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
615 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
616 n
->fclone
== SKB_FCLONE_UNAVAILABLE
) {
617 atomic_t
*fclone_ref
= (atomic_t
*) (n
+ 1);
618 n
->fclone
= SKB_FCLONE_CLONE
;
619 atomic_inc(fclone_ref
);
621 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
625 kmemcheck_annotate_bitfield(n
, flags1
);
626 kmemcheck_annotate_bitfield(n
, flags2
);
627 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
630 return __skb_clone(n
, skb
);
632 EXPORT_SYMBOL(skb_clone
);
634 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
636 #ifndef NET_SKBUFF_DATA_USES_OFFSET
638 * Shift between the two data areas in bytes
640 unsigned long offset
= new->data
- old
->data
;
643 __copy_skb_header(new, old
);
645 #ifndef NET_SKBUFF_DATA_USES_OFFSET
646 /* {transport,network,mac}_header are relative to skb->head */
647 new->transport_header
+= offset
;
648 new->network_header
+= offset
;
649 if (skb_mac_header_was_set(new))
650 new->mac_header
+= offset
;
652 skb_shinfo(new)->gso_size
= skb_shinfo(old
)->gso_size
;
653 skb_shinfo(new)->gso_segs
= skb_shinfo(old
)->gso_segs
;
654 skb_shinfo(new)->gso_type
= skb_shinfo(old
)->gso_type
;
658 * skb_copy - create private copy of an sk_buff
659 * @skb: buffer to copy
660 * @gfp_mask: allocation priority
662 * Make a copy of both an &sk_buff and its data. This is used when the
663 * caller wishes to modify the data and needs a private copy of the
664 * data to alter. Returns %NULL on failure or the pointer to the buffer
665 * on success. The returned buffer has a reference count of 1.
667 * As by-product this function converts non-linear &sk_buff to linear
668 * one, so that &sk_buff becomes completely private and caller is allowed
669 * to modify all the data of returned buffer. This means that this
670 * function is not recommended for use in circumstances when only
671 * header is going to be modified. Use pskb_copy() instead.
674 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
676 int headerlen
= skb_headroom(skb
);
677 unsigned int size
= (skb_end_pointer(skb
) - skb
->head
) + skb
->data_len
;
678 struct sk_buff
*n
= alloc_skb(size
, gfp_mask
);
683 /* Set the data pointer */
684 skb_reserve(n
, headerlen
);
685 /* Set the tail pointer and length */
686 skb_put(n
, skb
->len
);
688 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
691 copy_skb_header(n
, skb
);
694 EXPORT_SYMBOL(skb_copy
);
697 * pskb_copy - create copy of an sk_buff with private head.
698 * @skb: buffer to copy
699 * @gfp_mask: allocation priority
701 * Make a copy of both an &sk_buff and part of its data, located
702 * in header. Fragmented data remain shared. This is used when
703 * the caller wishes to modify only header of &sk_buff and needs
704 * private copy of the header to alter. Returns %NULL on failure
705 * or the pointer to the buffer on success.
706 * The returned buffer has a reference count of 1.
709 struct sk_buff
*pskb_copy(struct sk_buff
*skb
, gfp_t gfp_mask
)
711 unsigned int size
= skb_end_pointer(skb
) - skb
->head
;
712 struct sk_buff
*n
= alloc_skb(size
, gfp_mask
);
717 /* Set the data pointer */
718 skb_reserve(n
, skb_headroom(skb
));
719 /* Set the tail pointer and length */
720 skb_put(n
, skb_headlen(skb
));
722 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
724 n
->truesize
+= skb
->data_len
;
725 n
->data_len
= skb
->data_len
;
728 if (skb_shinfo(skb
)->nr_frags
) {
731 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
732 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
733 get_page(skb_shinfo(n
)->frags
[i
].page
);
735 skb_shinfo(n
)->nr_frags
= i
;
738 if (skb_has_frag_list(skb
)) {
739 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
740 skb_clone_fraglist(n
);
743 copy_skb_header(n
, skb
);
747 EXPORT_SYMBOL(pskb_copy
);
750 * pskb_expand_head - reallocate header of &sk_buff
751 * @skb: buffer to reallocate
752 * @nhead: room to add at head
753 * @ntail: room to add at tail
754 * @gfp_mask: allocation priority
756 * Expands (or creates identical copy, if &nhead and &ntail are zero)
757 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
758 * reference count of 1. Returns zero in the case of success or error,
759 * if expansion failed. In the last case, &sk_buff is not changed.
761 * All the pointers pointing into skb header may change and must be
762 * reloaded after call to this function.
765 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
770 int size
= nhead
+ (skb_end_pointer(skb
) - skb
->head
) + ntail
;
779 size
= SKB_DATA_ALIGN(size
);
781 /* Check if we can avoid taking references on fragments if we own
782 * the last reference on skb->head. (see skb_release_data())
787 int delta
= skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1;
789 fastpath
= atomic_read(&skb_shinfo(skb
)->dataref
) == delta
;
793 size
+ sizeof(struct skb_shared_info
) <= ksize(skb
->head
)) {
794 memmove(skb
->head
+ size
, skb_shinfo(skb
),
795 offsetof(struct skb_shared_info
,
796 frags
[skb_shinfo(skb
)->nr_frags
]));
797 memmove(skb
->head
+ nhead
, skb
->head
,
798 skb_tail_pointer(skb
) - skb
->head
);
803 data
= kmalloc(size
+ sizeof(struct skb_shared_info
), gfp_mask
);
807 /* Copy only real data... and, alas, header. This should be
808 * optimized for the cases when header is void.
810 memcpy(data
+ nhead
, skb
->head
, skb_tail_pointer(skb
) - skb
->head
);
812 memcpy((struct skb_shared_info
*)(data
+ size
),
814 offsetof(struct skb_shared_info
, frags
[skb_shinfo(skb
)->nr_frags
]));
819 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
820 get_page(skb_shinfo(skb
)->frags
[i
].page
);
822 if (skb_has_frag_list(skb
))
823 skb_clone_fraglist(skb
);
825 skb_release_data(skb
);
827 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 if (skb_mac_header_was_set(skb
))
843 skb
->mac_header
+= off
;
844 /* Only adjust this if it actually is csum_start rather than csum */
845 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
846 skb
->csum_start
+= nhead
;
850 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
856 EXPORT_SYMBOL(pskb_expand_head
);
858 /* Make private copy of skb with writable head and some headroom */
860 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
862 struct sk_buff
*skb2
;
863 int delta
= headroom
- skb_headroom(skb
);
866 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
868 skb2
= skb_clone(skb
, GFP_ATOMIC
);
869 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
877 EXPORT_SYMBOL(skb_realloc_headroom
);
880 * skb_copy_expand - copy and expand sk_buff
881 * @skb: buffer to copy
882 * @newheadroom: new free bytes at head
883 * @newtailroom: new free bytes at tail
884 * @gfp_mask: allocation priority
886 * Make a copy of both an &sk_buff and its data and while doing so
887 * allocate additional space.
889 * This is used when the caller wishes to modify the data and needs a
890 * private copy of the data to alter as well as more space for new fields.
891 * Returns %NULL on failure or the pointer to the buffer
892 * on success. The returned buffer has a reference count of 1.
894 * You must pass %GFP_ATOMIC as the allocation priority if this function
895 * is called from an interrupt.
897 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
898 int newheadroom
, int newtailroom
,
902 * Allocate the copy buffer
904 struct sk_buff
*n
= alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
906 int oldheadroom
= skb_headroom(skb
);
907 int head_copy_len
, head_copy_off
;
913 skb_reserve(n
, newheadroom
);
915 /* Set the tail pointer and length */
916 skb_put(n
, skb
->len
);
918 head_copy_len
= oldheadroom
;
920 if (newheadroom
<= head_copy_len
)
921 head_copy_len
= newheadroom
;
923 head_copy_off
= newheadroom
- head_copy_len
;
925 /* Copy the linear header and data. */
926 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
927 skb
->len
+ head_copy_len
))
930 copy_skb_header(n
, skb
);
932 off
= newheadroom
- oldheadroom
;
933 if (n
->ip_summed
== CHECKSUM_PARTIAL
)
934 n
->csum_start
+= off
;
935 #ifdef NET_SKBUFF_DATA_USES_OFFSET
936 n
->transport_header
+= off
;
937 n
->network_header
+= off
;
938 if (skb_mac_header_was_set(skb
))
939 n
->mac_header
+= off
;
944 EXPORT_SYMBOL(skb_copy_expand
);
947 * skb_pad - zero pad the tail of an skb
948 * @skb: buffer to pad
951 * Ensure that a buffer is followed by a padding area that is zero
952 * filled. Used by network drivers which may DMA or transfer data
953 * beyond the buffer end onto the wire.
955 * May return error in out of memory cases. The skb is freed on error.
958 int skb_pad(struct sk_buff
*skb
, int pad
)
963 /* If the skbuff is non linear tailroom is always zero.. */
964 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
965 memset(skb
->data
+skb
->len
, 0, pad
);
969 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
970 if (likely(skb_cloned(skb
) || ntail
> 0)) {
971 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
976 /* FIXME: The use of this function with non-linear skb's really needs
979 err
= skb_linearize(skb
);
983 memset(skb
->data
+ skb
->len
, 0, pad
);
990 EXPORT_SYMBOL(skb_pad
);
993 * skb_put - add data to a buffer
994 * @skb: buffer to use
995 * @len: amount of data to add
997 * This function extends the used data area of the buffer. If this would
998 * exceed the total buffer size the kernel will panic. A pointer to the
999 * first byte of the extra data is returned.
1001 unsigned char *skb_put(struct sk_buff
*skb
, unsigned int len
)
1003 unsigned char *tmp
= skb_tail_pointer(skb
);
1004 SKB_LINEAR_ASSERT(skb
);
1007 if (unlikely(skb
->tail
> skb
->end
))
1008 skb_over_panic(skb
, len
, __builtin_return_address(0));
1011 EXPORT_SYMBOL(skb_put
);
1014 * skb_push - add data to the start of a buffer
1015 * @skb: buffer to use
1016 * @len: amount of data to add
1018 * This function extends the used data area of the buffer at the buffer
1019 * start. If this would exceed the total buffer headroom the kernel will
1020 * panic. A pointer to the first byte of the extra data is returned.
1022 unsigned char *skb_push(struct sk_buff
*skb
, unsigned int len
)
1026 if (unlikely(skb
->data
<skb
->head
))
1027 skb_under_panic(skb
, len
, __builtin_return_address(0));
1030 EXPORT_SYMBOL(skb_push
);
1033 * skb_pull - remove data from the start of a buffer
1034 * @skb: buffer to use
1035 * @len: amount of data to remove
1037 * This function removes data from the start of a buffer, returning
1038 * the memory to the headroom. A pointer to the next data in the buffer
1039 * is returned. Once the data has been pulled future pushes will overwrite
1042 unsigned char *skb_pull(struct sk_buff
*skb
, unsigned int len
)
1044 return skb_pull_inline(skb
, len
);
1046 EXPORT_SYMBOL(skb_pull
);
1049 * skb_trim - remove end from a buffer
1050 * @skb: buffer to alter
1053 * Cut the length of a buffer down by removing data from the tail. If
1054 * the buffer is already under the length specified it is not modified.
1055 * The skb must be linear.
1057 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1060 __skb_trim(skb
, len
);
1062 EXPORT_SYMBOL(skb_trim
);
1064 /* Trims skb to length len. It can change skb pointers.
1067 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1069 struct sk_buff
**fragp
;
1070 struct sk_buff
*frag
;
1071 int offset
= skb_headlen(skb
);
1072 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1076 if (skb_cloned(skb
) &&
1077 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1084 for (; i
< nfrags
; i
++) {
1085 int end
= offset
+ skb_shinfo(skb
)->frags
[i
].size
;
1092 skb_shinfo(skb
)->frags
[i
++].size
= len
- offset
;
1095 skb_shinfo(skb
)->nr_frags
= i
;
1097 for (; i
< nfrags
; i
++)
1098 put_page(skb_shinfo(skb
)->frags
[i
].page
);
1100 if (skb_has_frag_list(skb
))
1101 skb_drop_fraglist(skb
);
1105 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1106 fragp
= &frag
->next
) {
1107 int end
= offset
+ frag
->len
;
1109 if (skb_shared(frag
)) {
1110 struct sk_buff
*nfrag
;
1112 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1113 if (unlikely(!nfrag
))
1116 nfrag
->next
= frag
->next
;
1128 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1132 skb_drop_list(&frag
->next
);
1137 if (len
> skb_headlen(skb
)) {
1138 skb
->data_len
-= skb
->len
- len
;
1143 skb_set_tail_pointer(skb
, len
);
1148 EXPORT_SYMBOL(___pskb_trim
);
1151 * __pskb_pull_tail - advance tail of skb header
1152 * @skb: buffer to reallocate
1153 * @delta: number of bytes to advance tail
1155 * The function makes a sense only on a fragmented &sk_buff,
1156 * it expands header moving its tail forward and copying necessary
1157 * data from fragmented part.
1159 * &sk_buff MUST have reference count of 1.
1161 * Returns %NULL (and &sk_buff does not change) if pull failed
1162 * or value of new tail of skb in the case of success.
1164 * All the pointers pointing into skb header may change and must be
1165 * reloaded after call to this function.
1168 /* Moves tail of skb head forward, copying data from fragmented part,
1169 * when it is necessary.
1170 * 1. It may fail due to malloc failure.
1171 * 2. It may change skb pointers.
1173 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1175 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1177 /* If skb has not enough free space at tail, get new one
1178 * plus 128 bytes for future expansions. If we have enough
1179 * room at tail, reallocate without expansion only if skb is cloned.
1181 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1183 if (eat
> 0 || skb_cloned(skb
)) {
1184 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1189 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
1192 /* Optimization: no fragments, no reasons to preestimate
1193 * size of pulled pages. Superb.
1195 if (!skb_has_frag_list(skb
))
1198 /* Estimate size of pulled pages. */
1200 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1201 if (skb_shinfo(skb
)->frags
[i
].size
>= eat
)
1203 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1206 /* If we need update frag list, we are in troubles.
1207 * Certainly, it possible to add an offset to skb data,
1208 * but taking into account that pulling is expected to
1209 * be very rare operation, it is worth to fight against
1210 * further bloating skb head and crucify ourselves here instead.
1211 * Pure masohism, indeed. 8)8)
1214 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1215 struct sk_buff
*clone
= NULL
;
1216 struct sk_buff
*insp
= NULL
;
1221 if (list
->len
<= eat
) {
1222 /* Eaten as whole. */
1227 /* Eaten partially. */
1229 if (skb_shared(list
)) {
1230 /* Sucks! We need to fork list. :-( */
1231 clone
= skb_clone(list
, GFP_ATOMIC
);
1237 /* This may be pulled without
1241 if (!pskb_pull(list
, eat
)) {
1249 /* Free pulled out fragments. */
1250 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1251 skb_shinfo(skb
)->frag_list
= list
->next
;
1254 /* And insert new clone at head. */
1257 skb_shinfo(skb
)->frag_list
= clone
;
1260 /* Success! Now we may commit changes to skb data. */
1265 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1266 if (skb_shinfo(skb
)->frags
[i
].size
<= eat
) {
1267 put_page(skb_shinfo(skb
)->frags
[i
].page
);
1268 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1270 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1272 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1273 skb_shinfo(skb
)->frags
[k
].size
-= eat
;
1279 skb_shinfo(skb
)->nr_frags
= k
;
1282 skb
->data_len
-= delta
;
1284 return skb_tail_pointer(skb
);
1286 EXPORT_SYMBOL(__pskb_pull_tail
);
1288 /* Copy some data bits from skb to kernel buffer. */
1290 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1292 int start
= skb_headlen(skb
);
1293 struct sk_buff
*frag_iter
;
1296 if (offset
> (int)skb
->len
- len
)
1300 if ((copy
= start
- offset
) > 0) {
1303 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
1304 if ((len
-= copy
) == 0)
1310 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1313 WARN_ON(start
> offset
+ len
);
1315 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1316 if ((copy
= end
- offset
) > 0) {
1322 vaddr
= kmap_skb_frag(&skb_shinfo(skb
)->frags
[i
]);
1324 vaddr
+ skb_shinfo(skb
)->frags
[i
].page_offset
+
1325 offset
- start
, copy
);
1326 kunmap_skb_frag(vaddr
);
1328 if ((len
-= copy
) == 0)
1336 skb_walk_frags(skb
, frag_iter
) {
1339 WARN_ON(start
> offset
+ len
);
1341 end
= start
+ frag_iter
->len
;
1342 if ((copy
= end
- offset
) > 0) {
1345 if (skb_copy_bits(frag_iter
, offset
- start
, to
, copy
))
1347 if ((len
-= copy
) == 0)
1360 EXPORT_SYMBOL(skb_copy_bits
);
1363 * Callback from splice_to_pipe(), if we need to release some pages
1364 * at the end of the spd in case we error'ed out in filling the pipe.
1366 static void sock_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
1368 put_page(spd
->pages
[i
]);
1371 static inline struct page
*linear_to_page(struct page
*page
, unsigned int *len
,
1372 unsigned int *offset
,
1373 struct sk_buff
*skb
, struct sock
*sk
)
1375 struct page
*p
= sk
->sk_sndmsg_page
;
1380 p
= sk
->sk_sndmsg_page
= alloc_pages(sk
->sk_allocation
, 0);
1384 off
= sk
->sk_sndmsg_off
= 0;
1385 /* hold one ref to this page until it's full */
1389 off
= sk
->sk_sndmsg_off
;
1390 mlen
= PAGE_SIZE
- off
;
1391 if (mlen
< 64 && mlen
< *len
) {
1396 *len
= min_t(unsigned int, *len
, mlen
);
1399 memcpy(page_address(p
) + off
, page_address(page
) + *offset
, *len
);
1400 sk
->sk_sndmsg_off
+= *len
;
1408 * Fill page/offset/length into spd, if it can hold more pages.
1410 static inline int spd_fill_page(struct splice_pipe_desc
*spd
,
1411 struct pipe_inode_info
*pipe
, struct page
*page
,
1412 unsigned int *len
, unsigned int offset
,
1413 struct sk_buff
*skb
, int linear
,
1416 if (unlikely(spd
->nr_pages
== pipe
->buffers
))
1420 page
= linear_to_page(page
, len
, &offset
, skb
, sk
);
1426 spd
->pages
[spd
->nr_pages
] = page
;
1427 spd
->partial
[spd
->nr_pages
].len
= *len
;
1428 spd
->partial
[spd
->nr_pages
].offset
= offset
;
1434 static inline void __segment_seek(struct page
**page
, unsigned int *poff
,
1435 unsigned int *plen
, unsigned int off
)
1440 n
= *poff
/ PAGE_SIZE
;
1442 *page
= nth_page(*page
, n
);
1444 *poff
= *poff
% PAGE_SIZE
;
1448 static inline int __splice_segment(struct page
*page
, unsigned int poff
,
1449 unsigned int plen
, unsigned int *off
,
1450 unsigned int *len
, struct sk_buff
*skb
,
1451 struct splice_pipe_desc
*spd
, int linear
,
1453 struct pipe_inode_info
*pipe
)
1458 /* skip this segment if already processed */
1464 /* ignore any bits we already processed */
1466 __segment_seek(&page
, &poff
, &plen
, *off
);
1471 unsigned int flen
= min(*len
, plen
);
1473 /* the linear region may spread across several pages */
1474 flen
= min_t(unsigned int, flen
, PAGE_SIZE
- poff
);
1476 if (spd_fill_page(spd
, pipe
, page
, &flen
, poff
, skb
, linear
, sk
))
1479 __segment_seek(&page
, &poff
, &plen
, flen
);
1482 } while (*len
&& plen
);
1488 * Map linear and fragment data from the skb to spd. It reports failure if the
1489 * pipe is full or if we already spliced the requested length.
1491 static int __skb_splice_bits(struct sk_buff
*skb
, struct pipe_inode_info
*pipe
,
1492 unsigned int *offset
, unsigned int *len
,
1493 struct splice_pipe_desc
*spd
, struct sock
*sk
)
1498 * map the linear part
1500 if (__splice_segment(virt_to_page(skb
->data
),
1501 (unsigned long) skb
->data
& (PAGE_SIZE
- 1),
1503 offset
, len
, skb
, spd
, 1, sk
, pipe
))
1507 * then map the fragments
1509 for (seg
= 0; seg
< skb_shinfo(skb
)->nr_frags
; seg
++) {
1510 const skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[seg
];
1512 if (__splice_segment(f
->page
, f
->page_offset
, f
->size
,
1513 offset
, len
, skb
, spd
, 0, sk
, pipe
))
1521 * Map data from the skb to a pipe. Should handle both the linear part,
1522 * the fragments, and the frag list. It does NOT handle frag lists within
1523 * the frag list, if such a thing exists. We'd probably need to recurse to
1524 * handle that cleanly.
1526 int skb_splice_bits(struct sk_buff
*skb
, unsigned int offset
,
1527 struct pipe_inode_info
*pipe
, unsigned int tlen
,
1530 struct partial_page partial
[PIPE_DEF_BUFFERS
];
1531 struct page
*pages
[PIPE_DEF_BUFFERS
];
1532 struct splice_pipe_desc spd
= {
1536 .ops
= &sock_pipe_buf_ops
,
1537 .spd_release
= sock_spd_release
,
1539 struct sk_buff
*frag_iter
;
1540 struct sock
*sk
= skb
->sk
;
1543 if (splice_grow_spd(pipe
, &spd
))
1547 * __skb_splice_bits() only fails if the output has no room left,
1548 * so no point in going over the frag_list for the error case.
1550 if (__skb_splice_bits(skb
, pipe
, &offset
, &tlen
, &spd
, sk
))
1556 * now see if we have a frag_list to map
1558 skb_walk_frags(skb
, frag_iter
) {
1561 if (__skb_splice_bits(frag_iter
, pipe
, &offset
, &tlen
, &spd
, sk
))
1568 * Drop the socket lock, otherwise we have reverse
1569 * locking dependencies between sk_lock and i_mutex
1570 * here as compared to sendfile(). We enter here
1571 * with the socket lock held, and splice_to_pipe() will
1572 * grab the pipe inode lock. For sendfile() emulation,
1573 * we call into ->sendpage() with the i_mutex lock held
1574 * and networking will grab the socket lock.
1577 ret
= splice_to_pipe(pipe
, &spd
);
1581 splice_shrink_spd(pipe
, &spd
);
1586 * skb_store_bits - store bits from kernel buffer to skb
1587 * @skb: destination buffer
1588 * @offset: offset in destination
1589 * @from: source buffer
1590 * @len: number of bytes to copy
1592 * Copy the specified number of bytes from the source buffer to the
1593 * destination skb. This function handles all the messy bits of
1594 * traversing fragment lists and such.
1597 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
1599 int start
= skb_headlen(skb
);
1600 struct sk_buff
*frag_iter
;
1603 if (offset
> (int)skb
->len
- len
)
1606 if ((copy
= start
- offset
) > 0) {
1609 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
1610 if ((len
-= copy
) == 0)
1616 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1617 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1620 WARN_ON(start
> offset
+ len
);
1622 end
= start
+ frag
->size
;
1623 if ((copy
= end
- offset
) > 0) {
1629 vaddr
= kmap_skb_frag(frag
);
1630 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1632 kunmap_skb_frag(vaddr
);
1634 if ((len
-= copy
) == 0)
1642 skb_walk_frags(skb
, frag_iter
) {
1645 WARN_ON(start
> offset
+ len
);
1647 end
= start
+ frag_iter
->len
;
1648 if ((copy
= end
- offset
) > 0) {
1651 if (skb_store_bits(frag_iter
, offset
- start
,
1654 if ((len
-= copy
) == 0)
1667 EXPORT_SYMBOL(skb_store_bits
);
1669 /* Checksum skb data. */
1671 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
1672 int len
, __wsum csum
)
1674 int start
= skb_headlen(skb
);
1675 int i
, copy
= start
- offset
;
1676 struct sk_buff
*frag_iter
;
1679 /* Checksum header. */
1683 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1684 if ((len
-= copy
) == 0)
1690 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1693 WARN_ON(start
> offset
+ len
);
1695 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1696 if ((copy
= end
- offset
) > 0) {
1699 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1703 vaddr
= kmap_skb_frag(frag
);
1704 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
1705 offset
- start
, copy
, 0);
1706 kunmap_skb_frag(vaddr
);
1707 csum
= csum_block_add(csum
, csum2
, pos
);
1716 skb_walk_frags(skb
, frag_iter
) {
1719 WARN_ON(start
> offset
+ len
);
1721 end
= start
+ frag_iter
->len
;
1722 if ((copy
= end
- offset
) > 0) {
1726 csum2
= skb_checksum(frag_iter
, offset
- start
,
1728 csum
= csum_block_add(csum
, csum2
, pos
);
1729 if ((len
-= copy
) == 0)
1740 EXPORT_SYMBOL(skb_checksum
);
1742 /* Both of above in one bottle. */
1744 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
1745 u8
*to
, int len
, __wsum csum
)
1747 int start
= skb_headlen(skb
);
1748 int i
, copy
= start
- offset
;
1749 struct sk_buff
*frag_iter
;
1756 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
1758 if ((len
-= copy
) == 0)
1765 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1768 WARN_ON(start
> offset
+ len
);
1770 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1771 if ((copy
= end
- offset
) > 0) {
1774 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1778 vaddr
= kmap_skb_frag(frag
);
1779 csum2
= csum_partial_copy_nocheck(vaddr
+
1783 kunmap_skb_frag(vaddr
);
1784 csum
= csum_block_add(csum
, csum2
, pos
);
1794 skb_walk_frags(skb
, frag_iter
) {
1798 WARN_ON(start
> offset
+ len
);
1800 end
= start
+ frag_iter
->len
;
1801 if ((copy
= end
- offset
) > 0) {
1804 csum2
= skb_copy_and_csum_bits(frag_iter
,
1807 csum
= csum_block_add(csum
, csum2
, pos
);
1808 if ((len
-= copy
) == 0)
1819 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
1821 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
1826 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1827 csstart
= skb_checksum_start_offset(skb
);
1829 csstart
= skb_headlen(skb
);
1831 BUG_ON(csstart
> skb_headlen(skb
));
1833 skb_copy_from_linear_data(skb
, to
, csstart
);
1836 if (csstart
!= skb
->len
)
1837 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
1838 skb
->len
- csstart
, 0);
1840 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1841 long csstuff
= csstart
+ skb
->csum_offset
;
1843 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
1846 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
1849 * skb_dequeue - remove from the head of the queue
1850 * @list: list to dequeue from
1852 * Remove the head of the list. The list lock is taken so the function
1853 * may be used safely with other locking list functions. The head item is
1854 * returned or %NULL if the list is empty.
1857 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
1859 unsigned long flags
;
1860 struct sk_buff
*result
;
1862 spin_lock_irqsave(&list
->lock
, flags
);
1863 result
= __skb_dequeue(list
);
1864 spin_unlock_irqrestore(&list
->lock
, flags
);
1867 EXPORT_SYMBOL(skb_dequeue
);
1870 * skb_dequeue_tail - remove from the tail of the queue
1871 * @list: list to dequeue from
1873 * Remove the tail of the list. The list lock is taken so the function
1874 * may be used safely with other locking list functions. The tail item is
1875 * returned or %NULL if the list is empty.
1877 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
1879 unsigned long flags
;
1880 struct sk_buff
*result
;
1882 spin_lock_irqsave(&list
->lock
, flags
);
1883 result
= __skb_dequeue_tail(list
);
1884 spin_unlock_irqrestore(&list
->lock
, flags
);
1887 EXPORT_SYMBOL(skb_dequeue_tail
);
1890 * skb_queue_purge - empty a list
1891 * @list: list to empty
1893 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1894 * the list and one reference dropped. This function takes the list
1895 * lock and is atomic with respect to other list locking functions.
1897 void skb_queue_purge(struct sk_buff_head
*list
)
1899 struct sk_buff
*skb
;
1900 while ((skb
= skb_dequeue(list
)) != NULL
)
1903 EXPORT_SYMBOL(skb_queue_purge
);
1906 * skb_queue_head - queue a buffer at the list head
1907 * @list: list to use
1908 * @newsk: buffer to queue
1910 * Queue a buffer at the start of the list. This function takes the
1911 * list lock and can be used safely with other locking &sk_buff functions
1914 * A buffer cannot be placed on two lists at the same time.
1916 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1918 unsigned long flags
;
1920 spin_lock_irqsave(&list
->lock
, flags
);
1921 __skb_queue_head(list
, newsk
);
1922 spin_unlock_irqrestore(&list
->lock
, flags
);
1924 EXPORT_SYMBOL(skb_queue_head
);
1927 * skb_queue_tail - queue a buffer at the list tail
1928 * @list: list to use
1929 * @newsk: buffer to queue
1931 * Queue a buffer at the tail of the list. This function takes the
1932 * list lock and can be used safely with other locking &sk_buff functions
1935 * A buffer cannot be placed on two lists at the same time.
1937 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1939 unsigned long flags
;
1941 spin_lock_irqsave(&list
->lock
, flags
);
1942 __skb_queue_tail(list
, newsk
);
1943 spin_unlock_irqrestore(&list
->lock
, flags
);
1945 EXPORT_SYMBOL(skb_queue_tail
);
1948 * skb_unlink - remove a buffer from a list
1949 * @skb: buffer to remove
1950 * @list: list to use
1952 * Remove a packet from a list. The list locks are taken and this
1953 * function is atomic with respect to other list locked calls
1955 * You must know what list the SKB is on.
1957 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
1959 unsigned long flags
;
1961 spin_lock_irqsave(&list
->lock
, flags
);
1962 __skb_unlink(skb
, list
);
1963 spin_unlock_irqrestore(&list
->lock
, flags
);
1965 EXPORT_SYMBOL(skb_unlink
);
1968 * skb_append - append a buffer
1969 * @old: buffer to insert after
1970 * @newsk: buffer to insert
1971 * @list: list to use
1973 * Place a packet after a given packet in a list. The list locks are taken
1974 * and this function is atomic with respect to other list locked calls.
1975 * A buffer cannot be placed on two lists at the same time.
1977 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
1979 unsigned long flags
;
1981 spin_lock_irqsave(&list
->lock
, flags
);
1982 __skb_queue_after(list
, old
, newsk
);
1983 spin_unlock_irqrestore(&list
->lock
, flags
);
1985 EXPORT_SYMBOL(skb_append
);
1988 * skb_insert - insert a buffer
1989 * @old: buffer to insert before
1990 * @newsk: buffer to insert
1991 * @list: list to use
1993 * Place a packet before a given packet in a list. The list locks are
1994 * taken and this function is atomic with respect to other list locked
1997 * A buffer cannot be placed on two lists at the same time.
1999 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2001 unsigned long flags
;
2003 spin_lock_irqsave(&list
->lock
, flags
);
2004 __skb_insert(newsk
, old
->prev
, old
, list
);
2005 spin_unlock_irqrestore(&list
->lock
, flags
);
2007 EXPORT_SYMBOL(skb_insert
);
2009 static inline void skb_split_inside_header(struct sk_buff
*skb
,
2010 struct sk_buff
* skb1
,
2011 const u32 len
, const int pos
)
2015 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
2017 /* And move data appendix as is. */
2018 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
2019 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
2021 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
2022 skb_shinfo(skb
)->nr_frags
= 0;
2023 skb1
->data_len
= skb
->data_len
;
2024 skb1
->len
+= skb1
->data_len
;
2027 skb_set_tail_pointer(skb
, len
);
2030 static inline void skb_split_no_header(struct sk_buff
*skb
,
2031 struct sk_buff
* skb1
,
2032 const u32 len
, int pos
)
2035 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
2037 skb_shinfo(skb
)->nr_frags
= 0;
2038 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
2040 skb
->data_len
= len
- pos
;
2042 for (i
= 0; i
< nfrags
; i
++) {
2043 int size
= skb_shinfo(skb
)->frags
[i
].size
;
2045 if (pos
+ size
> len
) {
2046 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
2050 * We have two variants in this case:
2051 * 1. Move all the frag to the second
2052 * part, if it is possible. F.e.
2053 * this approach is mandatory for TUX,
2054 * where splitting is expensive.
2055 * 2. Split is accurately. We make this.
2057 get_page(skb_shinfo(skb
)->frags
[i
].page
);
2058 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
2059 skb_shinfo(skb1
)->frags
[0].size
-= len
- pos
;
2060 skb_shinfo(skb
)->frags
[i
].size
= len
- pos
;
2061 skb_shinfo(skb
)->nr_frags
++;
2065 skb_shinfo(skb
)->nr_frags
++;
2068 skb_shinfo(skb1
)->nr_frags
= k
;
2072 * skb_split - Split fragmented skb to two parts at length len.
2073 * @skb: the buffer to split
2074 * @skb1: the buffer to receive the second part
2075 * @len: new length for skb
2077 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
2079 int pos
= skb_headlen(skb
);
2081 if (len
< pos
) /* Split line is inside header. */
2082 skb_split_inside_header(skb
, skb1
, len
, pos
);
2083 else /* Second chunk has no header, nothing to copy. */
2084 skb_split_no_header(skb
, skb1
, len
, pos
);
2086 EXPORT_SYMBOL(skb_split
);
2088 /* Shifting from/to a cloned skb is a no-go.
2090 * Caller cannot keep skb_shinfo related pointers past calling here!
2092 static int skb_prepare_for_shift(struct sk_buff
*skb
)
2094 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2098 * skb_shift - Shifts paged data partially from skb to another
2099 * @tgt: buffer into which tail data gets added
2100 * @skb: buffer from which the paged data comes from
2101 * @shiftlen: shift up to this many bytes
2103 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2104 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2105 * It's up to caller to free skb if everything was shifted.
2107 * If @tgt runs out of frags, the whole operation is aborted.
2109 * Skb cannot include anything else but paged data while tgt is allowed
2110 * to have non-paged data as well.
2112 * TODO: full sized shift could be optimized but that would need
2113 * specialized skb free'er to handle frags without up-to-date nr_frags.
2115 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
2117 int from
, to
, merge
, todo
;
2118 struct skb_frag_struct
*fragfrom
, *fragto
;
2120 BUG_ON(shiftlen
> skb
->len
);
2121 BUG_ON(skb_headlen(skb
)); /* Would corrupt stream */
2125 to
= skb_shinfo(tgt
)->nr_frags
;
2126 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2128 /* Actual merge is delayed until the point when we know we can
2129 * commit all, so that we don't have to undo partial changes
2132 !skb_can_coalesce(tgt
, to
, fragfrom
->page
, fragfrom
->page_offset
)) {
2137 todo
-= fragfrom
->size
;
2139 if (skb_prepare_for_shift(skb
) ||
2140 skb_prepare_for_shift(tgt
))
2143 /* All previous frag pointers might be stale! */
2144 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2145 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2147 fragto
->size
+= shiftlen
;
2148 fragfrom
->size
-= shiftlen
;
2149 fragfrom
->page_offset
+= shiftlen
;
2157 /* Skip full, not-fitting skb to avoid expensive operations */
2158 if ((shiftlen
== skb
->len
) &&
2159 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
2162 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
2165 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
2166 if (to
== MAX_SKB_FRAGS
)
2169 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2170 fragto
= &skb_shinfo(tgt
)->frags
[to
];
2172 if (todo
>= fragfrom
->size
) {
2173 *fragto
= *fragfrom
;
2174 todo
-= fragfrom
->size
;
2179 get_page(fragfrom
->page
);
2180 fragto
->page
= fragfrom
->page
;
2181 fragto
->page_offset
= fragfrom
->page_offset
;
2182 fragto
->size
= todo
;
2184 fragfrom
->page_offset
+= todo
;
2185 fragfrom
->size
-= todo
;
2193 /* Ready to "commit" this state change to tgt */
2194 skb_shinfo(tgt
)->nr_frags
= to
;
2197 fragfrom
= &skb_shinfo(skb
)->frags
[0];
2198 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2200 fragto
->size
+= fragfrom
->size
;
2201 put_page(fragfrom
->page
);
2204 /* Reposition in the original skb */
2206 while (from
< skb_shinfo(skb
)->nr_frags
)
2207 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
2208 skb_shinfo(skb
)->nr_frags
= to
;
2210 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
2213 /* Most likely the tgt won't ever need its checksum anymore, skb on
2214 * the other hand might need it if it needs to be resent
2216 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
2217 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2219 /* Yak, is it really working this way? Some helper please? */
2220 skb
->len
-= shiftlen
;
2221 skb
->data_len
-= shiftlen
;
2222 skb
->truesize
-= shiftlen
;
2223 tgt
->len
+= shiftlen
;
2224 tgt
->data_len
+= shiftlen
;
2225 tgt
->truesize
+= shiftlen
;
2231 * skb_prepare_seq_read - Prepare a sequential read of skb data
2232 * @skb: the buffer to read
2233 * @from: lower offset of data to be read
2234 * @to: upper offset of data to be read
2235 * @st: state variable
2237 * Initializes the specified state variable. Must be called before
2238 * invoking skb_seq_read() for the first time.
2240 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
2241 unsigned int to
, struct skb_seq_state
*st
)
2243 st
->lower_offset
= from
;
2244 st
->upper_offset
= to
;
2245 st
->root_skb
= st
->cur_skb
= skb
;
2246 st
->frag_idx
= st
->stepped_offset
= 0;
2247 st
->frag_data
= NULL
;
2249 EXPORT_SYMBOL(skb_prepare_seq_read
);
2252 * skb_seq_read - Sequentially read skb data
2253 * @consumed: number of bytes consumed by the caller so far
2254 * @data: destination pointer for data to be returned
2255 * @st: state variable
2257 * Reads a block of skb data at &consumed relative to the
2258 * lower offset specified to skb_prepare_seq_read(). Assigns
2259 * the head of the data block to &data and returns the length
2260 * of the block or 0 if the end of the skb data or the upper
2261 * offset has been reached.
2263 * The caller is not required to consume all of the data
2264 * returned, i.e. &consumed is typically set to the number
2265 * of bytes already consumed and the next call to
2266 * skb_seq_read() will return the remaining part of the block.
2268 * Note 1: The size of each block of data returned can be arbitary,
2269 * this limitation is the cost for zerocopy seqeuental
2270 * reads of potentially non linear data.
2272 * Note 2: Fragment lists within fragments are not implemented
2273 * at the moment, state->root_skb could be replaced with
2274 * a stack for this purpose.
2276 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
2277 struct skb_seq_state
*st
)
2279 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
2282 if (unlikely(abs_offset
>= st
->upper_offset
))
2286 block_limit
= skb_headlen(st
->cur_skb
) + st
->stepped_offset
;
2288 if (abs_offset
< block_limit
&& !st
->frag_data
) {
2289 *data
= st
->cur_skb
->data
+ (abs_offset
- st
->stepped_offset
);
2290 return block_limit
- abs_offset
;
2293 if (st
->frag_idx
== 0 && !st
->frag_data
)
2294 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
2296 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
2297 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
2298 block_limit
= frag
->size
+ st
->stepped_offset
;
2300 if (abs_offset
< block_limit
) {
2302 st
->frag_data
= kmap_skb_frag(frag
);
2304 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
2305 (abs_offset
- st
->stepped_offset
);
2307 return block_limit
- abs_offset
;
2310 if (st
->frag_data
) {
2311 kunmap_skb_frag(st
->frag_data
);
2312 st
->frag_data
= NULL
;
2316 st
->stepped_offset
+= frag
->size
;
2319 if (st
->frag_data
) {
2320 kunmap_skb_frag(st
->frag_data
);
2321 st
->frag_data
= NULL
;
2324 if (st
->root_skb
== st
->cur_skb
&& skb_has_frag_list(st
->root_skb
)) {
2325 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
2328 } else if (st
->cur_skb
->next
) {
2329 st
->cur_skb
= st
->cur_skb
->next
;
2336 EXPORT_SYMBOL(skb_seq_read
);
2339 * skb_abort_seq_read - Abort a sequential read of skb data
2340 * @st: state variable
2342 * Must be called if skb_seq_read() was not called until it
2345 void skb_abort_seq_read(struct skb_seq_state
*st
)
2348 kunmap_skb_frag(st
->frag_data
);
2350 EXPORT_SYMBOL(skb_abort_seq_read
);
2352 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2354 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
2355 struct ts_config
*conf
,
2356 struct ts_state
*state
)
2358 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
2361 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
2363 skb_abort_seq_read(TS_SKB_CB(state
));
2367 * skb_find_text - Find a text pattern in skb data
2368 * @skb: the buffer to look in
2369 * @from: search offset
2371 * @config: textsearch configuration
2372 * @state: uninitialized textsearch state variable
2374 * Finds a pattern in the skb data according to the specified
2375 * textsearch configuration. Use textsearch_next() to retrieve
2376 * subsequent occurrences of the pattern. Returns the offset
2377 * to the first occurrence or UINT_MAX if no match was found.
2379 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
2380 unsigned int to
, struct ts_config
*config
,
2381 struct ts_state
*state
)
2385 config
->get_next_block
= skb_ts_get_next_block
;
2386 config
->finish
= skb_ts_finish
;
2388 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
2390 ret
= textsearch_find(config
, state
);
2391 return (ret
<= to
- from
? ret
: UINT_MAX
);
2393 EXPORT_SYMBOL(skb_find_text
);
2396 * skb_append_datato_frags: - append the user data to a skb
2397 * @sk: sock structure
2398 * @skb: skb structure to be appened with user data.
2399 * @getfrag: call back function to be used for getting the user data
2400 * @from: pointer to user message iov
2401 * @length: length of the iov message
2403 * Description: This procedure append the user data in the fragment part
2404 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2406 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
2407 int (*getfrag
)(void *from
, char *to
, int offset
,
2408 int len
, int odd
, struct sk_buff
*skb
),
2409 void *from
, int length
)
2412 skb_frag_t
*frag
= NULL
;
2413 struct page
*page
= NULL
;
2419 /* Return error if we don't have space for new frag */
2420 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2421 if (frg_cnt
>= MAX_SKB_FRAGS
)
2424 /* allocate a new page for next frag */
2425 page
= alloc_pages(sk
->sk_allocation
, 0);
2427 /* If alloc_page fails just return failure and caller will
2428 * free previous allocated pages by doing kfree_skb()
2433 /* initialize the next frag */
2434 sk
->sk_sndmsg_page
= page
;
2435 sk
->sk_sndmsg_off
= 0;
2436 skb_fill_page_desc(skb
, frg_cnt
, page
, 0, 0);
2437 skb
->truesize
+= PAGE_SIZE
;
2438 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
2440 /* get the new initialized frag */
2441 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2442 frag
= &skb_shinfo(skb
)->frags
[frg_cnt
- 1];
2444 /* copy the user data to page */
2445 left
= PAGE_SIZE
- frag
->page_offset
;
2446 copy
= (length
> left
)? left
: length
;
2448 ret
= getfrag(from
, (page_address(frag
->page
) +
2449 frag
->page_offset
+ frag
->size
),
2450 offset
, copy
, 0, skb
);
2454 /* copy was successful so update the size parameters */
2455 sk
->sk_sndmsg_off
+= copy
;
2458 skb
->data_len
+= copy
;
2462 } while (length
> 0);
2466 EXPORT_SYMBOL(skb_append_datato_frags
);
2469 * skb_pull_rcsum - pull skb and update receive checksum
2470 * @skb: buffer to update
2471 * @len: length of data pulled
2473 * This function performs an skb_pull on the packet and updates
2474 * the CHECKSUM_COMPLETE checksum. It should be used on
2475 * receive path processing instead of skb_pull unless you know
2476 * that the checksum difference is zero (e.g., a valid IP header)
2477 * or you are setting ip_summed to CHECKSUM_NONE.
2479 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
2481 BUG_ON(len
> skb
->len
);
2483 BUG_ON(skb
->len
< skb
->data_len
);
2484 skb_postpull_rcsum(skb
, skb
->data
, len
);
2485 return skb
->data
+= len
;
2487 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
2490 * skb_segment - Perform protocol segmentation on skb.
2491 * @skb: buffer to segment
2492 * @features: features for the output path (see dev->features)
2494 * This function performs segmentation on the given skb. It returns
2495 * a pointer to the first in a list of new skbs for the segments.
2496 * In case of error it returns ERR_PTR(err).
2498 struct sk_buff
*skb_segment(struct sk_buff
*skb
, int features
)
2500 struct sk_buff
*segs
= NULL
;
2501 struct sk_buff
*tail
= NULL
;
2502 struct sk_buff
*fskb
= skb_shinfo(skb
)->frag_list
;
2503 unsigned int mss
= skb_shinfo(skb
)->gso_size
;
2504 unsigned int doffset
= skb
->data
- skb_mac_header(skb
);
2505 unsigned int offset
= doffset
;
2506 unsigned int headroom
;
2508 int sg
= features
& NETIF_F_SG
;
2509 int nfrags
= skb_shinfo(skb
)->nr_frags
;
2514 __skb_push(skb
, doffset
);
2515 headroom
= skb_headroom(skb
);
2516 pos
= skb_headlen(skb
);
2519 struct sk_buff
*nskb
;
2524 len
= skb
->len
- offset
;
2528 hsize
= skb_headlen(skb
) - offset
;
2531 if (hsize
> len
|| !sg
)
2534 if (!hsize
&& i
>= nfrags
) {
2535 BUG_ON(fskb
->len
!= len
);
2538 nskb
= skb_clone(fskb
, GFP_ATOMIC
);
2541 if (unlikely(!nskb
))
2544 hsize
= skb_end_pointer(nskb
) - nskb
->head
;
2545 if (skb_cow_head(nskb
, doffset
+ headroom
)) {
2550 nskb
->truesize
+= skb_end_pointer(nskb
) - nskb
->head
-
2552 skb_release_head_state(nskb
);
2553 __skb_push(nskb
, doffset
);
2555 nskb
= alloc_skb(hsize
+ doffset
+ headroom
,
2558 if (unlikely(!nskb
))
2561 skb_reserve(nskb
, headroom
);
2562 __skb_put(nskb
, doffset
);
2571 __copy_skb_header(nskb
, skb
);
2572 nskb
->mac_len
= skb
->mac_len
;
2574 /* nskb and skb might have different headroom */
2575 if (nskb
->ip_summed
== CHECKSUM_PARTIAL
)
2576 nskb
->csum_start
+= skb_headroom(nskb
) - headroom
;
2578 skb_reset_mac_header(nskb
);
2579 skb_set_network_header(nskb
, skb
->mac_len
);
2580 nskb
->transport_header
= (nskb
->network_header
+
2581 skb_network_header_len(skb
));
2582 skb_copy_from_linear_data(skb
, nskb
->data
, doffset
);
2584 if (fskb
!= skb_shinfo(skb
)->frag_list
)
2588 nskb
->ip_summed
= CHECKSUM_NONE
;
2589 nskb
->csum
= skb_copy_and_csum_bits(skb
, offset
,
2595 frag
= skb_shinfo(nskb
)->frags
;
2597 skb_copy_from_linear_data_offset(skb
, offset
,
2598 skb_put(nskb
, hsize
), hsize
);
2600 while (pos
< offset
+ len
&& i
< nfrags
) {
2601 *frag
= skb_shinfo(skb
)->frags
[i
];
2602 get_page(frag
->page
);
2606 frag
->page_offset
+= offset
- pos
;
2607 frag
->size
-= offset
- pos
;
2610 skb_shinfo(nskb
)->nr_frags
++;
2612 if (pos
+ size
<= offset
+ len
) {
2616 frag
->size
-= pos
+ size
- (offset
+ len
);
2623 if (pos
< offset
+ len
) {
2624 struct sk_buff
*fskb2
= fskb
;
2626 BUG_ON(pos
+ fskb
->len
!= offset
+ len
);
2632 fskb2
= skb_clone(fskb2
, GFP_ATOMIC
);
2638 SKB_FRAG_ASSERT(nskb
);
2639 skb_shinfo(nskb
)->frag_list
= fskb2
;
2643 nskb
->data_len
= len
- hsize
;
2644 nskb
->len
+= nskb
->data_len
;
2645 nskb
->truesize
+= nskb
->data_len
;
2646 } while ((offset
+= len
) < skb
->len
);
2651 while ((skb
= segs
)) {
2655 return ERR_PTR(err
);
2657 EXPORT_SYMBOL_GPL(skb_segment
);
2659 int skb_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2661 struct sk_buff
*p
= *head
;
2662 struct sk_buff
*nskb
;
2663 struct skb_shared_info
*skbinfo
= skb_shinfo(skb
);
2664 struct skb_shared_info
*pinfo
= skb_shinfo(p
);
2665 unsigned int headroom
;
2666 unsigned int len
= skb_gro_len(skb
);
2667 unsigned int offset
= skb_gro_offset(skb
);
2668 unsigned int headlen
= skb_headlen(skb
);
2670 if (p
->len
+ len
>= 65536)
2673 if (pinfo
->frag_list
)
2675 else if (headlen
<= offset
) {
2678 int i
= skbinfo
->nr_frags
;
2679 int nr_frags
= pinfo
->nr_frags
+ i
;
2683 if (nr_frags
> MAX_SKB_FRAGS
)
2686 pinfo
->nr_frags
= nr_frags
;
2687 skbinfo
->nr_frags
= 0;
2689 frag
= pinfo
->frags
+ nr_frags
;
2690 frag2
= skbinfo
->frags
+ i
;
2695 frag
->page_offset
+= offset
;
2696 frag
->size
-= offset
;
2698 skb
->truesize
-= skb
->data_len
;
2699 skb
->len
-= skb
->data_len
;
2702 NAPI_GRO_CB(skb
)->free
= 1;
2704 } else if (skb_gro_len(p
) != pinfo
->gso_size
)
2707 headroom
= skb_headroom(p
);
2708 nskb
= alloc_skb(headroom
+ skb_gro_offset(p
), GFP_ATOMIC
);
2709 if (unlikely(!nskb
))
2712 __copy_skb_header(nskb
, p
);
2713 nskb
->mac_len
= p
->mac_len
;
2715 skb_reserve(nskb
, headroom
);
2716 __skb_put(nskb
, skb_gro_offset(p
));
2718 skb_set_mac_header(nskb
, skb_mac_header(p
) - p
->data
);
2719 skb_set_network_header(nskb
, skb_network_offset(p
));
2720 skb_set_transport_header(nskb
, skb_transport_offset(p
));
2722 __skb_pull(p
, skb_gro_offset(p
));
2723 memcpy(skb_mac_header(nskb
), skb_mac_header(p
),
2724 p
->data
- skb_mac_header(p
));
2726 *NAPI_GRO_CB(nskb
) = *NAPI_GRO_CB(p
);
2727 skb_shinfo(nskb
)->frag_list
= p
;
2728 skb_shinfo(nskb
)->gso_size
= pinfo
->gso_size
;
2729 pinfo
->gso_size
= 0;
2730 skb_header_release(p
);
2733 nskb
->data_len
+= p
->len
;
2734 nskb
->truesize
+= p
->len
;
2735 nskb
->len
+= p
->len
;
2738 nskb
->next
= p
->next
;
2744 if (offset
> headlen
) {
2745 skbinfo
->frags
[0].page_offset
+= offset
- headlen
;
2746 skbinfo
->frags
[0].size
-= offset
- headlen
;
2750 __skb_pull(skb
, offset
);
2752 p
->prev
->next
= skb
;
2754 skb_header_release(skb
);
2757 NAPI_GRO_CB(p
)->count
++;
2762 NAPI_GRO_CB(skb
)->same_flow
= 1;
2765 EXPORT_SYMBOL_GPL(skb_gro_receive
);
2767 void __init
skb_init(void)
2769 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
2770 sizeof(struct sk_buff
),
2772 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2774 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
2775 (2*sizeof(struct sk_buff
)) +
2778 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2783 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2784 * @skb: Socket buffer containing the buffers to be mapped
2785 * @sg: The scatter-gather list to map into
2786 * @offset: The offset into the buffer's contents to start mapping
2787 * @len: Length of buffer space to be mapped
2789 * Fill the specified scatter-gather list with mappings/pointers into a
2790 * region of the buffer space attached to a socket buffer.
2793 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2795 int start
= skb_headlen(skb
);
2796 int i
, copy
= start
- offset
;
2797 struct sk_buff
*frag_iter
;
2803 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
2805 if ((len
-= copy
) == 0)
2810 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2813 WARN_ON(start
> offset
+ len
);
2815 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
2816 if ((copy
= end
- offset
) > 0) {
2817 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2821 sg_set_page(&sg
[elt
], frag
->page
, copy
,
2822 frag
->page_offset
+offset
-start
);
2831 skb_walk_frags(skb
, frag_iter
) {
2834 WARN_ON(start
> offset
+ len
);
2836 end
= start
+ frag_iter
->len
;
2837 if ((copy
= end
- offset
) > 0) {
2840 elt
+= __skb_to_sgvec(frag_iter
, sg
+elt
, offset
- start
,
2842 if ((len
-= copy
) == 0)
2852 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2854 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
);
2856 sg_mark_end(&sg
[nsg
- 1]);
2860 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
2863 * skb_cow_data - Check that a socket buffer's data buffers are writable
2864 * @skb: The socket buffer to check.
2865 * @tailbits: Amount of trailing space to be added
2866 * @trailer: Returned pointer to the skb where the @tailbits space begins
2868 * Make sure that the data buffers attached to a socket buffer are
2869 * writable. If they are not, private copies are made of the data buffers
2870 * and the socket buffer is set to use these instead.
2872 * If @tailbits is given, make sure that there is space to write @tailbits
2873 * bytes of data beyond current end of socket buffer. @trailer will be
2874 * set to point to the skb in which this space begins.
2876 * The number of scatterlist elements required to completely map the
2877 * COW'd and extended socket buffer will be returned.
2879 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
2883 struct sk_buff
*skb1
, **skb_p
;
2885 /* If skb is cloned or its head is paged, reallocate
2886 * head pulling out all the pages (pages are considered not writable
2887 * at the moment even if they are anonymous).
2889 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
2890 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
2893 /* Easy case. Most of packets will go this way. */
2894 if (!skb_has_frag_list(skb
)) {
2895 /* A little of trouble, not enough of space for trailer.
2896 * This should not happen, when stack is tuned to generate
2897 * good frames. OK, on miss we reallocate and reserve even more
2898 * space, 128 bytes is fair. */
2900 if (skb_tailroom(skb
) < tailbits
&&
2901 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
2909 /* Misery. We are in troubles, going to mincer fragments... */
2912 skb_p
= &skb_shinfo(skb
)->frag_list
;
2915 while ((skb1
= *skb_p
) != NULL
) {
2918 /* The fragment is partially pulled by someone,
2919 * this can happen on input. Copy it and everything
2922 if (skb_shared(skb1
))
2925 /* If the skb is the last, worry about trailer. */
2927 if (skb1
->next
== NULL
&& tailbits
) {
2928 if (skb_shinfo(skb1
)->nr_frags
||
2929 skb_has_frag_list(skb1
) ||
2930 skb_tailroom(skb1
) < tailbits
)
2931 ntail
= tailbits
+ 128;
2937 skb_shinfo(skb1
)->nr_frags
||
2938 skb_has_frag_list(skb1
)) {
2939 struct sk_buff
*skb2
;
2941 /* Fuck, we are miserable poor guys... */
2943 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
2945 skb2
= skb_copy_expand(skb1
,
2949 if (unlikely(skb2
== NULL
))
2953 skb_set_owner_w(skb2
, skb1
->sk
);
2955 /* Looking around. Are we still alive?
2956 * OK, link new skb, drop old one */
2958 skb2
->next
= skb1
->next
;
2965 skb_p
= &skb1
->next
;
2970 EXPORT_SYMBOL_GPL(skb_cow_data
);
2972 static void sock_rmem_free(struct sk_buff
*skb
)
2974 struct sock
*sk
= skb
->sk
;
2976 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
2980 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
2982 int sock_queue_err_skb(struct sock
*sk
, struct sk_buff
*skb
)
2984 if (atomic_read(&sk
->sk_rmem_alloc
) + skb
->truesize
>=
2985 (unsigned)sk
->sk_rcvbuf
)
2990 skb
->destructor
= sock_rmem_free
;
2991 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
2993 skb_queue_tail(&sk
->sk_error_queue
, skb
);
2994 if (!sock_flag(sk
, SOCK_DEAD
))
2995 sk
->sk_data_ready(sk
, skb
->len
);
2998 EXPORT_SYMBOL(sock_queue_err_skb
);
3000 void skb_tstamp_tx(struct sk_buff
*orig_skb
,
3001 struct skb_shared_hwtstamps
*hwtstamps
)
3003 struct sock
*sk
= orig_skb
->sk
;
3004 struct sock_exterr_skb
*serr
;
3005 struct sk_buff
*skb
;
3011 skb
= skb_clone(orig_skb
, GFP_ATOMIC
);
3016 *skb_hwtstamps(skb
) =
3020 * no hardware time stamps available,
3021 * so keep the shared tx_flags and only
3022 * store software time stamp
3024 skb
->tstamp
= ktime_get_real();
3027 serr
= SKB_EXT_ERR(skb
);
3028 memset(serr
, 0, sizeof(*serr
));
3029 serr
->ee
.ee_errno
= ENOMSG
;
3030 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TIMESTAMPING
;
3032 err
= sock_queue_err_skb(sk
, skb
);
3037 EXPORT_SYMBOL_GPL(skb_tstamp_tx
);
3041 * skb_partial_csum_set - set up and verify partial csum values for packet
3042 * @skb: the skb to set
3043 * @start: the number of bytes after skb->data to start checksumming.
3044 * @off: the offset from start to place the checksum.
3046 * For untrusted partially-checksummed packets, we need to make sure the values
3047 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3049 * This function checks and sets those values and skb->ip_summed: if this
3050 * returns false you should drop the packet.
3052 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
3054 if (unlikely(start
> skb_headlen(skb
)) ||
3055 unlikely((int)start
+ off
> skb_headlen(skb
) - 2)) {
3056 if (net_ratelimit())
3058 "bad partial csum: csum=%u/%u len=%u\n",
3059 start
, off
, skb_headlen(skb
));
3062 skb
->ip_summed
= CHECKSUM_PARTIAL
;
3063 skb
->csum_start
= skb_headroom(skb
) + start
;
3064 skb
->csum_offset
= off
;
3067 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);
3069 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
3071 if (net_ratelimit())
3072 pr_warning("%s: received packets cannot be forwarded"
3073 " while LRO is enabled\n", skb
->dev
->name
);
3075 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);