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
)
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_frag_list(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 bool skb_recycle_check(struct sk_buff
*skb
, int skb_size
)
485 struct skb_shared_info
*shinfo
;
490 if (skb_is_nonlinear(skb
) || skb
->fclone
!= SKB_FCLONE_UNAVAILABLE
)
493 skb_size
= SKB_DATA_ALIGN(skb_size
+ NET_SKB_PAD
);
494 if (skb_end_pointer(skb
) - skb
->head
< skb_size
)
497 if (skb_shared(skb
) || skb_cloned(skb
))
500 skb_release_head_state(skb
);
502 shinfo
= skb_shinfo(skb
);
503 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
504 atomic_set(&shinfo
->dataref
, 1);
506 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
507 skb
->data
= skb
->head
+ NET_SKB_PAD
;
508 skb_reset_tail_pointer(skb
);
512 EXPORT_SYMBOL(skb_recycle_check
);
514 static void __copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
516 new->tstamp
= old
->tstamp
;
518 new->transport_header
= old
->transport_header
;
519 new->network_header
= old
->network_header
;
520 new->mac_header
= old
->mac_header
;
521 skb_dst_copy(new, old
);
522 new->rxhash
= old
->rxhash
;
524 new->sp
= secpath_get(old
->sp
);
526 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
527 new->csum
= old
->csum
;
528 new->local_df
= old
->local_df
;
529 new->pkt_type
= old
->pkt_type
;
530 new->ip_summed
= old
->ip_summed
;
531 skb_copy_queue_mapping(new, old
);
532 new->priority
= old
->priority
;
533 new->deliver_no_wcard
= old
->deliver_no_wcard
;
534 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
535 new->ipvs_property
= old
->ipvs_property
;
537 new->protocol
= old
->protocol
;
538 new->mark
= old
->mark
;
539 new->skb_iif
= old
->skb_iif
;
541 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
542 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
543 new->nf_trace
= old
->nf_trace
;
545 #ifdef CONFIG_NET_SCHED
546 new->tc_index
= old
->tc_index
;
547 #ifdef CONFIG_NET_CLS_ACT
548 new->tc_verd
= old
->tc_verd
;
551 new->vlan_tci
= old
->vlan_tci
;
553 skb_copy_secmark(new, old
);
557 * You should not add any new code to this function. Add it to
558 * __copy_skb_header above instead.
560 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
562 #define C(x) n->x = skb->x
564 n
->next
= n
->prev
= NULL
;
566 __copy_skb_header(n
, skb
);
571 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
574 n
->destructor
= NULL
;
580 atomic_set(&n
->users
, 1);
582 atomic_inc(&(skb_shinfo(skb
)->dataref
));
590 * skb_morph - morph one skb into another
591 * @dst: the skb to receive the contents
592 * @src: the skb to supply the contents
594 * This is identical to skb_clone except that the target skb is
595 * supplied by the user.
597 * The target skb is returned upon exit.
599 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
601 skb_release_all(dst
);
602 return __skb_clone(dst
, src
);
604 EXPORT_SYMBOL_GPL(skb_morph
);
607 * skb_clone - duplicate an sk_buff
608 * @skb: buffer to clone
609 * @gfp_mask: allocation priority
611 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
612 * copies share the same packet data but not structure. The new
613 * buffer has a reference count of 1. If the allocation fails the
614 * function returns %NULL otherwise the new buffer is returned.
616 * If this function is called from an interrupt gfp_mask() must be
620 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
625 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
626 n
->fclone
== SKB_FCLONE_UNAVAILABLE
) {
627 atomic_t
*fclone_ref
= (atomic_t
*) (n
+ 1);
628 n
->fclone
= SKB_FCLONE_CLONE
;
629 atomic_inc(fclone_ref
);
631 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
635 kmemcheck_annotate_bitfield(n
, flags1
);
636 kmemcheck_annotate_bitfield(n
, flags2
);
637 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
640 return __skb_clone(n
, skb
);
642 EXPORT_SYMBOL(skb_clone
);
644 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
646 #ifndef NET_SKBUFF_DATA_USES_OFFSET
648 * Shift between the two data areas in bytes
650 unsigned long offset
= new->data
- old
->data
;
653 __copy_skb_header(new, old
);
655 #ifndef NET_SKBUFF_DATA_USES_OFFSET
656 /* {transport,network,mac}_header are relative to skb->head */
657 new->transport_header
+= offset
;
658 new->network_header
+= offset
;
659 if (skb_mac_header_was_set(new))
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_headroom(skb
);
687 unsigned int size
= (skb_end_pointer(skb
) - skb
->head
) + skb
->data_len
;
688 struct sk_buff
*n
= alloc_skb(size
, gfp_mask
);
693 /* Set the data pointer */
694 skb_reserve(n
, headerlen
);
695 /* Set the tail pointer and length */
696 skb_put(n
, skb
->len
);
698 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
701 copy_skb_header(n
, skb
);
704 EXPORT_SYMBOL(skb_copy
);
707 * pskb_copy - create copy of an sk_buff with private head.
708 * @skb: buffer to copy
709 * @gfp_mask: allocation priority
711 * Make a copy of both an &sk_buff and part of its data, located
712 * in header. Fragmented data remain shared. This is used when
713 * the caller wishes to modify only header of &sk_buff and needs
714 * private copy of the header to alter. Returns %NULL on failure
715 * or the pointer to the buffer on success.
716 * The returned buffer has a reference count of 1.
719 struct sk_buff
*pskb_copy(struct sk_buff
*skb
, gfp_t gfp_mask
)
721 unsigned int size
= skb_end_pointer(skb
) - skb
->head
;
722 struct sk_buff
*n
= alloc_skb(size
, gfp_mask
);
727 /* Set the data pointer */
728 skb_reserve(n
, skb_headroom(skb
));
729 /* Set the tail pointer and length */
730 skb_put(n
, skb_headlen(skb
));
732 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
734 n
->truesize
+= skb
->data_len
;
735 n
->data_len
= skb
->data_len
;
738 if (skb_shinfo(skb
)->nr_frags
) {
741 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
742 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
743 get_page(skb_shinfo(n
)->frags
[i
].page
);
745 skb_shinfo(n
)->nr_frags
= i
;
748 if (skb_has_frag_list(skb
)) {
749 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
750 skb_clone_fraglist(n
);
753 copy_skb_header(n
, skb
);
757 EXPORT_SYMBOL(pskb_copy
);
760 * pskb_expand_head - reallocate header of &sk_buff
761 * @skb: buffer to reallocate
762 * @nhead: room to add at head
763 * @ntail: room to add at tail
764 * @gfp_mask: allocation priority
766 * Expands (or creates identical copy, if &nhead and &ntail are zero)
767 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
768 * reference count of 1. Returns zero in the case of success or error,
769 * if expansion failed. In the last case, &sk_buff is not changed.
771 * All the pointers pointing into skb header may change and must be
772 * reloaded after call to this function.
775 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
780 int size
= nhead
+ (skb_end_pointer(skb
) - skb
->head
) + ntail
;
789 size
= SKB_DATA_ALIGN(size
);
791 data
= kmalloc(size
+ sizeof(struct skb_shared_info
), gfp_mask
);
795 /* Copy only real data... and, alas, header. This should be
796 * optimized for the cases when header is void.
798 memcpy(data
+ nhead
, skb
->head
, skb_tail_pointer(skb
) - skb
->head
);
800 memcpy((struct skb_shared_info
*)(data
+ size
),
802 offsetof(struct skb_shared_info
, frags
[skb_shinfo(skb
)->nr_frags
]));
804 /* Check if we can avoid taking references on fragments if we own
805 * the last reference on skb->head. (see skb_release_data())
810 int delta
= skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1;
812 fastpath
= atomic_read(&skb_shinfo(skb
)->dataref
) == delta
;
818 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
819 get_page(skb_shinfo(skb
)->frags
[i
].page
);
821 if (skb_has_frag_list(skb
))
822 skb_clone_fraglist(skb
);
824 skb_release_data(skb
);
826 off
= (data
+ nhead
) - skb
->head
;
830 #ifdef NET_SKBUFF_DATA_USES_OFFSET
834 skb
->end
= skb
->head
+ size
;
836 /* {transport,network,mac}_header and tail are relative to skb->head */
838 skb
->transport_header
+= off
;
839 skb
->network_header
+= off
;
840 if (skb_mac_header_was_set(skb
))
841 skb
->mac_header
+= off
;
842 /* Only adjust this if it actually is csum_start rather than csum */
843 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
844 skb
->csum_start
+= nhead
;
848 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
854 EXPORT_SYMBOL(pskb_expand_head
);
856 /* Make private copy of skb with writable head and some headroom */
858 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
860 struct sk_buff
*skb2
;
861 int delta
= headroom
- skb_headroom(skb
);
864 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
866 skb2
= skb_clone(skb
, GFP_ATOMIC
);
867 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
875 EXPORT_SYMBOL(skb_realloc_headroom
);
878 * skb_copy_expand - copy and expand sk_buff
879 * @skb: buffer to copy
880 * @newheadroom: new free bytes at head
881 * @newtailroom: new free bytes at tail
882 * @gfp_mask: allocation priority
884 * Make a copy of both an &sk_buff and its data and while doing so
885 * allocate additional space.
887 * This is used when the caller wishes to modify the data and needs a
888 * private copy of the data to alter as well as more space for new fields.
889 * Returns %NULL on failure or the pointer to the buffer
890 * on success. The returned buffer has a reference count of 1.
892 * You must pass %GFP_ATOMIC as the allocation priority if this function
893 * is called from an interrupt.
895 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
896 int newheadroom
, int newtailroom
,
900 * Allocate the copy buffer
902 struct sk_buff
*n
= alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
904 int oldheadroom
= skb_headroom(skb
);
905 int head_copy_len
, head_copy_off
;
911 skb_reserve(n
, newheadroom
);
913 /* Set the tail pointer and length */
914 skb_put(n
, skb
->len
);
916 head_copy_len
= oldheadroom
;
918 if (newheadroom
<= head_copy_len
)
919 head_copy_len
= newheadroom
;
921 head_copy_off
= newheadroom
- head_copy_len
;
923 /* Copy the linear header and data. */
924 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
925 skb
->len
+ head_copy_len
))
928 copy_skb_header(n
, skb
);
930 off
= newheadroom
- oldheadroom
;
931 if (n
->ip_summed
== CHECKSUM_PARTIAL
)
932 n
->csum_start
+= off
;
933 #ifdef NET_SKBUFF_DATA_USES_OFFSET
934 n
->transport_header
+= off
;
935 n
->network_header
+= off
;
936 if (skb_mac_header_was_set(skb
))
937 n
->mac_header
+= off
;
942 EXPORT_SYMBOL(skb_copy_expand
);
945 * skb_pad - zero pad the tail of an skb
946 * @skb: buffer to pad
949 * Ensure that a buffer is followed by a padding area that is zero
950 * filled. Used by network drivers which may DMA or transfer data
951 * beyond the buffer end onto the wire.
953 * May return error in out of memory cases. The skb is freed on error.
956 int skb_pad(struct sk_buff
*skb
, int pad
)
961 /* If the skbuff is non linear tailroom is always zero.. */
962 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
963 memset(skb
->data
+skb
->len
, 0, pad
);
967 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
968 if (likely(skb_cloned(skb
) || ntail
> 0)) {
969 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
974 /* FIXME: The use of this function with non-linear skb's really needs
977 err
= skb_linearize(skb
);
981 memset(skb
->data
+ skb
->len
, 0, pad
);
988 EXPORT_SYMBOL(skb_pad
);
991 * skb_put - add data to a buffer
992 * @skb: buffer to use
993 * @len: amount of data to add
995 * This function extends the used data area of the buffer. If this would
996 * exceed the total buffer size the kernel will panic. A pointer to the
997 * first byte of the extra data is returned.
999 unsigned char *skb_put(struct sk_buff
*skb
, unsigned int len
)
1001 unsigned char *tmp
= skb_tail_pointer(skb
);
1002 SKB_LINEAR_ASSERT(skb
);
1005 if (unlikely(skb
->tail
> skb
->end
))
1006 skb_over_panic(skb
, len
, __builtin_return_address(0));
1009 EXPORT_SYMBOL(skb_put
);
1012 * skb_push - add data to the start of a buffer
1013 * @skb: buffer to use
1014 * @len: amount of data to add
1016 * This function extends the used data area of the buffer at the buffer
1017 * start. If this would exceed the total buffer headroom the kernel will
1018 * panic. A pointer to the first byte of the extra data is returned.
1020 unsigned char *skb_push(struct sk_buff
*skb
, unsigned int len
)
1024 if (unlikely(skb
->data
<skb
->head
))
1025 skb_under_panic(skb
, len
, __builtin_return_address(0));
1028 EXPORT_SYMBOL(skb_push
);
1031 * skb_pull - remove data from the start of a buffer
1032 * @skb: buffer to use
1033 * @len: amount of data to remove
1035 * This function removes data from the start of a buffer, returning
1036 * the memory to the headroom. A pointer to the next data in the buffer
1037 * is returned. Once the data has been pulled future pushes will overwrite
1040 unsigned char *skb_pull(struct sk_buff
*skb
, unsigned int len
)
1042 return skb_pull_inline(skb
, len
);
1044 EXPORT_SYMBOL(skb_pull
);
1047 * skb_trim - remove end from a buffer
1048 * @skb: buffer to alter
1051 * Cut the length of a buffer down by removing data from the tail. If
1052 * the buffer is already under the length specified it is not modified.
1053 * The skb must be linear.
1055 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1058 __skb_trim(skb
, len
);
1060 EXPORT_SYMBOL(skb_trim
);
1062 /* Trims skb to length len. It can change skb pointers.
1065 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1067 struct sk_buff
**fragp
;
1068 struct sk_buff
*frag
;
1069 int offset
= skb_headlen(skb
);
1070 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1074 if (skb_cloned(skb
) &&
1075 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1082 for (; i
< nfrags
; i
++) {
1083 int end
= offset
+ skb_shinfo(skb
)->frags
[i
].size
;
1090 skb_shinfo(skb
)->frags
[i
++].size
= len
- offset
;
1093 skb_shinfo(skb
)->nr_frags
= i
;
1095 for (; i
< nfrags
; i
++)
1096 put_page(skb_shinfo(skb
)->frags
[i
].page
);
1098 if (skb_has_frag_list(skb
))
1099 skb_drop_fraglist(skb
);
1103 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1104 fragp
= &frag
->next
) {
1105 int end
= offset
+ frag
->len
;
1107 if (skb_shared(frag
)) {
1108 struct sk_buff
*nfrag
;
1110 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1111 if (unlikely(!nfrag
))
1114 nfrag
->next
= frag
->next
;
1126 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1130 skb_drop_list(&frag
->next
);
1135 if (len
> skb_headlen(skb
)) {
1136 skb
->data_len
-= skb
->len
- len
;
1141 skb_set_tail_pointer(skb
, len
);
1146 EXPORT_SYMBOL(___pskb_trim
);
1149 * __pskb_pull_tail - advance tail of skb header
1150 * @skb: buffer to reallocate
1151 * @delta: number of bytes to advance tail
1153 * The function makes a sense only on a fragmented &sk_buff,
1154 * it expands header moving its tail forward and copying necessary
1155 * data from fragmented part.
1157 * &sk_buff MUST have reference count of 1.
1159 * Returns %NULL (and &sk_buff does not change) if pull failed
1160 * or value of new tail of skb in the case of success.
1162 * All the pointers pointing into skb header may change and must be
1163 * reloaded after call to this function.
1166 /* Moves tail of skb head forward, copying data from fragmented part,
1167 * when it is necessary.
1168 * 1. It may fail due to malloc failure.
1169 * 2. It may change skb pointers.
1171 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1173 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1175 /* If skb has not enough free space at tail, get new one
1176 * plus 128 bytes for future expansions. If we have enough
1177 * room at tail, reallocate without expansion only if skb is cloned.
1179 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1181 if (eat
> 0 || skb_cloned(skb
)) {
1182 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1187 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
1190 /* Optimization: no fragments, no reasons to preestimate
1191 * size of pulled pages. Superb.
1193 if (!skb_has_frag_list(skb
))
1196 /* Estimate size of pulled pages. */
1198 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1199 if (skb_shinfo(skb
)->frags
[i
].size
>= eat
)
1201 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1204 /* If we need update frag list, we are in troubles.
1205 * Certainly, it possible to add an offset to skb data,
1206 * but taking into account that pulling is expected to
1207 * be very rare operation, it is worth to fight against
1208 * further bloating skb head and crucify ourselves here instead.
1209 * Pure masohism, indeed. 8)8)
1212 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1213 struct sk_buff
*clone
= NULL
;
1214 struct sk_buff
*insp
= NULL
;
1219 if (list
->len
<= eat
) {
1220 /* Eaten as whole. */
1225 /* Eaten partially. */
1227 if (skb_shared(list
)) {
1228 /* Sucks! We need to fork list. :-( */
1229 clone
= skb_clone(list
, GFP_ATOMIC
);
1235 /* This may be pulled without
1239 if (!pskb_pull(list
, eat
)) {
1247 /* Free pulled out fragments. */
1248 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1249 skb_shinfo(skb
)->frag_list
= list
->next
;
1252 /* And insert new clone at head. */
1255 skb_shinfo(skb
)->frag_list
= clone
;
1258 /* Success! Now we may commit changes to skb data. */
1263 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1264 if (skb_shinfo(skb
)->frags
[i
].size
<= eat
) {
1265 put_page(skb_shinfo(skb
)->frags
[i
].page
);
1266 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1268 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1270 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1271 skb_shinfo(skb
)->frags
[k
].size
-= eat
;
1277 skb_shinfo(skb
)->nr_frags
= k
;
1280 skb
->data_len
-= delta
;
1282 return skb_tail_pointer(skb
);
1284 EXPORT_SYMBOL(__pskb_pull_tail
);
1286 /* Copy some data bits from skb to kernel buffer. */
1288 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1290 int start
= skb_headlen(skb
);
1291 struct sk_buff
*frag_iter
;
1294 if (offset
> (int)skb
->len
- len
)
1298 if ((copy
= start
- offset
) > 0) {
1301 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
1302 if ((len
-= copy
) == 0)
1308 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1311 WARN_ON(start
> offset
+ len
);
1313 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1314 if ((copy
= end
- offset
) > 0) {
1320 vaddr
= kmap_skb_frag(&skb_shinfo(skb
)->frags
[i
]);
1322 vaddr
+ skb_shinfo(skb
)->frags
[i
].page_offset
+
1323 offset
- start
, copy
);
1324 kunmap_skb_frag(vaddr
);
1326 if ((len
-= copy
) == 0)
1334 skb_walk_frags(skb
, frag_iter
) {
1337 WARN_ON(start
> offset
+ len
);
1339 end
= start
+ frag_iter
->len
;
1340 if ((copy
= end
- offset
) > 0) {
1343 if (skb_copy_bits(frag_iter
, offset
- start
, to
, copy
))
1345 if ((len
-= copy
) == 0)
1358 EXPORT_SYMBOL(skb_copy_bits
);
1361 * Callback from splice_to_pipe(), if we need to release some pages
1362 * at the end of the spd in case we error'ed out in filling the pipe.
1364 static void sock_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
1366 put_page(spd
->pages
[i
]);
1369 static inline struct page
*linear_to_page(struct page
*page
, unsigned int *len
,
1370 unsigned int *offset
,
1371 struct sk_buff
*skb
, struct sock
*sk
)
1373 struct page
*p
= sk
->sk_sndmsg_page
;
1378 p
= sk
->sk_sndmsg_page
= alloc_pages(sk
->sk_allocation
, 0);
1382 off
= sk
->sk_sndmsg_off
= 0;
1383 /* hold one ref to this page until it's full */
1387 off
= sk
->sk_sndmsg_off
;
1388 mlen
= PAGE_SIZE
- off
;
1389 if (mlen
< 64 && mlen
< *len
) {
1394 *len
= min_t(unsigned int, *len
, mlen
);
1397 memcpy(page_address(p
) + off
, page_address(page
) + *offset
, *len
);
1398 sk
->sk_sndmsg_off
+= *len
;
1406 * Fill page/offset/length into spd, if it can hold more pages.
1408 static inline int spd_fill_page(struct splice_pipe_desc
*spd
,
1409 struct pipe_inode_info
*pipe
, struct page
*page
,
1410 unsigned int *len
, unsigned int offset
,
1411 struct sk_buff
*skb
, int linear
,
1414 if (unlikely(spd
->nr_pages
== pipe
->buffers
))
1418 page
= linear_to_page(page
, len
, &offset
, skb
, sk
);
1424 spd
->pages
[spd
->nr_pages
] = page
;
1425 spd
->partial
[spd
->nr_pages
].len
= *len
;
1426 spd
->partial
[spd
->nr_pages
].offset
= offset
;
1432 static inline void __segment_seek(struct page
**page
, unsigned int *poff
,
1433 unsigned int *plen
, unsigned int off
)
1438 n
= *poff
/ PAGE_SIZE
;
1440 *page
= nth_page(*page
, n
);
1442 *poff
= *poff
% PAGE_SIZE
;
1446 static inline int __splice_segment(struct page
*page
, unsigned int poff
,
1447 unsigned int plen
, unsigned int *off
,
1448 unsigned int *len
, struct sk_buff
*skb
,
1449 struct splice_pipe_desc
*spd
, int linear
,
1451 struct pipe_inode_info
*pipe
)
1456 /* skip this segment if already processed */
1462 /* ignore any bits we already processed */
1464 __segment_seek(&page
, &poff
, &plen
, *off
);
1469 unsigned int flen
= min(*len
, plen
);
1471 /* the linear region may spread across several pages */
1472 flen
= min_t(unsigned int, flen
, PAGE_SIZE
- poff
);
1474 if (spd_fill_page(spd
, pipe
, page
, &flen
, poff
, skb
, linear
, sk
))
1477 __segment_seek(&page
, &poff
, &plen
, flen
);
1480 } while (*len
&& plen
);
1486 * Map linear and fragment data from the skb to spd. It reports failure if the
1487 * pipe is full or if we already spliced the requested length.
1489 static int __skb_splice_bits(struct sk_buff
*skb
, struct pipe_inode_info
*pipe
,
1490 unsigned int *offset
, unsigned int *len
,
1491 struct splice_pipe_desc
*spd
, struct sock
*sk
)
1496 * map the linear part
1498 if (__splice_segment(virt_to_page(skb
->data
),
1499 (unsigned long) skb
->data
& (PAGE_SIZE
- 1),
1501 offset
, len
, skb
, spd
, 1, sk
, pipe
))
1505 * then map the fragments
1507 for (seg
= 0; seg
< skb_shinfo(skb
)->nr_frags
; seg
++) {
1508 const skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[seg
];
1510 if (__splice_segment(f
->page
, f
->page_offset
, f
->size
,
1511 offset
, len
, skb
, spd
, 0, sk
, pipe
))
1519 * Map data from the skb to a pipe. Should handle both the linear part,
1520 * the fragments, and the frag list. It does NOT handle frag lists within
1521 * the frag list, if such a thing exists. We'd probably need to recurse to
1522 * handle that cleanly.
1524 int skb_splice_bits(struct sk_buff
*skb
, unsigned int offset
,
1525 struct pipe_inode_info
*pipe
, unsigned int tlen
,
1528 struct partial_page partial
[PIPE_DEF_BUFFERS
];
1529 struct page
*pages
[PIPE_DEF_BUFFERS
];
1530 struct splice_pipe_desc spd
= {
1534 .ops
= &sock_pipe_buf_ops
,
1535 .spd_release
= sock_spd_release
,
1537 struct sk_buff
*frag_iter
;
1538 struct sock
*sk
= skb
->sk
;
1541 if (splice_grow_spd(pipe
, &spd
))
1545 * __skb_splice_bits() only fails if the output has no room left,
1546 * so no point in going over the frag_list for the error case.
1548 if (__skb_splice_bits(skb
, pipe
, &offset
, &tlen
, &spd
, sk
))
1554 * now see if we have a frag_list to map
1556 skb_walk_frags(skb
, frag_iter
) {
1559 if (__skb_splice_bits(frag_iter
, pipe
, &offset
, &tlen
, &spd
, sk
))
1566 * Drop the socket lock, otherwise we have reverse
1567 * locking dependencies between sk_lock and i_mutex
1568 * here as compared to sendfile(). We enter here
1569 * with the socket lock held, and splice_to_pipe() will
1570 * grab the pipe inode lock. For sendfile() emulation,
1571 * we call into ->sendpage() with the i_mutex lock held
1572 * and networking will grab the socket lock.
1575 ret
= splice_to_pipe(pipe
, &spd
);
1579 splice_shrink_spd(pipe
, &spd
);
1584 * skb_store_bits - store bits from kernel buffer to skb
1585 * @skb: destination buffer
1586 * @offset: offset in destination
1587 * @from: source buffer
1588 * @len: number of bytes to copy
1590 * Copy the specified number of bytes from the source buffer to the
1591 * destination skb. This function handles all the messy bits of
1592 * traversing fragment lists and such.
1595 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
1597 int start
= skb_headlen(skb
);
1598 struct sk_buff
*frag_iter
;
1601 if (offset
> (int)skb
->len
- len
)
1604 if ((copy
= start
- offset
) > 0) {
1607 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
1608 if ((len
-= copy
) == 0)
1614 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1615 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1618 WARN_ON(start
> offset
+ len
);
1620 end
= start
+ frag
->size
;
1621 if ((copy
= end
- offset
) > 0) {
1627 vaddr
= kmap_skb_frag(frag
);
1628 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1630 kunmap_skb_frag(vaddr
);
1632 if ((len
-= copy
) == 0)
1640 skb_walk_frags(skb
, frag_iter
) {
1643 WARN_ON(start
> offset
+ len
);
1645 end
= start
+ frag_iter
->len
;
1646 if ((copy
= end
- offset
) > 0) {
1649 if (skb_store_bits(frag_iter
, offset
- start
,
1652 if ((len
-= copy
) == 0)
1665 EXPORT_SYMBOL(skb_store_bits
);
1667 /* Checksum skb data. */
1669 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
1670 int len
, __wsum csum
)
1672 int start
= skb_headlen(skb
);
1673 int i
, copy
= start
- offset
;
1674 struct sk_buff
*frag_iter
;
1677 /* Checksum header. */
1681 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1682 if ((len
-= copy
) == 0)
1688 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1691 WARN_ON(start
> offset
+ len
);
1693 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1694 if ((copy
= end
- offset
) > 0) {
1697 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1701 vaddr
= kmap_skb_frag(frag
);
1702 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
1703 offset
- start
, copy
, 0);
1704 kunmap_skb_frag(vaddr
);
1705 csum
= csum_block_add(csum
, csum2
, pos
);
1714 skb_walk_frags(skb
, frag_iter
) {
1717 WARN_ON(start
> offset
+ len
);
1719 end
= start
+ frag_iter
->len
;
1720 if ((copy
= end
- offset
) > 0) {
1724 csum2
= skb_checksum(frag_iter
, offset
- start
,
1726 csum
= csum_block_add(csum
, csum2
, pos
);
1727 if ((len
-= copy
) == 0)
1738 EXPORT_SYMBOL(skb_checksum
);
1740 /* Both of above in one bottle. */
1742 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
1743 u8
*to
, int len
, __wsum csum
)
1745 int start
= skb_headlen(skb
);
1746 int i
, copy
= start
- offset
;
1747 struct sk_buff
*frag_iter
;
1754 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
1756 if ((len
-= copy
) == 0)
1763 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1766 WARN_ON(start
> offset
+ len
);
1768 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1769 if ((copy
= end
- offset
) > 0) {
1772 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1776 vaddr
= kmap_skb_frag(frag
);
1777 csum2
= csum_partial_copy_nocheck(vaddr
+
1781 kunmap_skb_frag(vaddr
);
1782 csum
= csum_block_add(csum
, csum2
, pos
);
1792 skb_walk_frags(skb
, frag_iter
) {
1796 WARN_ON(start
> offset
+ len
);
1798 end
= start
+ frag_iter
->len
;
1799 if ((copy
= end
- offset
) > 0) {
1802 csum2
= skb_copy_and_csum_bits(frag_iter
,
1805 csum
= csum_block_add(csum
, csum2
, pos
);
1806 if ((len
-= copy
) == 0)
1817 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
1819 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
1824 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1825 csstart
= skb
->csum_start
- skb_headroom(skb
);
1827 csstart
= skb_headlen(skb
);
1829 BUG_ON(csstart
> skb_headlen(skb
));
1831 skb_copy_from_linear_data(skb
, to
, csstart
);
1834 if (csstart
!= skb
->len
)
1835 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
1836 skb
->len
- csstart
, 0);
1838 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1839 long csstuff
= csstart
+ skb
->csum_offset
;
1841 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
1844 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
1847 * skb_dequeue - remove from the head of the queue
1848 * @list: list to dequeue from
1850 * Remove the head of the list. The list lock is taken so the function
1851 * may be used safely with other locking list functions. The head item is
1852 * returned or %NULL if the list is empty.
1855 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
1857 unsigned long flags
;
1858 struct sk_buff
*result
;
1860 spin_lock_irqsave(&list
->lock
, flags
);
1861 result
= __skb_dequeue(list
);
1862 spin_unlock_irqrestore(&list
->lock
, flags
);
1865 EXPORT_SYMBOL(skb_dequeue
);
1868 * skb_dequeue_tail - remove from the tail of the queue
1869 * @list: list to dequeue from
1871 * Remove the tail of the list. The list lock is taken so the function
1872 * may be used safely with other locking list functions. The tail item is
1873 * returned or %NULL if the list is empty.
1875 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
1877 unsigned long flags
;
1878 struct sk_buff
*result
;
1880 spin_lock_irqsave(&list
->lock
, flags
);
1881 result
= __skb_dequeue_tail(list
);
1882 spin_unlock_irqrestore(&list
->lock
, flags
);
1885 EXPORT_SYMBOL(skb_dequeue_tail
);
1888 * skb_queue_purge - empty a list
1889 * @list: list to empty
1891 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1892 * the list and one reference dropped. This function takes the list
1893 * lock and is atomic with respect to other list locking functions.
1895 void skb_queue_purge(struct sk_buff_head
*list
)
1897 struct sk_buff
*skb
;
1898 while ((skb
= skb_dequeue(list
)) != NULL
)
1901 EXPORT_SYMBOL(skb_queue_purge
);
1904 * skb_queue_head - queue a buffer at the list head
1905 * @list: list to use
1906 * @newsk: buffer to queue
1908 * Queue a buffer at the start of the list. This function takes the
1909 * list lock and can be used safely with other locking &sk_buff functions
1912 * A buffer cannot be placed on two lists at the same time.
1914 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1916 unsigned long flags
;
1918 spin_lock_irqsave(&list
->lock
, flags
);
1919 __skb_queue_head(list
, newsk
);
1920 spin_unlock_irqrestore(&list
->lock
, flags
);
1922 EXPORT_SYMBOL(skb_queue_head
);
1925 * skb_queue_tail - queue a buffer at the list tail
1926 * @list: list to use
1927 * @newsk: buffer to queue
1929 * Queue a buffer at the tail of the list. This function takes the
1930 * list lock and can be used safely with other locking &sk_buff functions
1933 * A buffer cannot be placed on two lists at the same time.
1935 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1937 unsigned long flags
;
1939 spin_lock_irqsave(&list
->lock
, flags
);
1940 __skb_queue_tail(list
, newsk
);
1941 spin_unlock_irqrestore(&list
->lock
, flags
);
1943 EXPORT_SYMBOL(skb_queue_tail
);
1946 * skb_unlink - remove a buffer from a list
1947 * @skb: buffer to remove
1948 * @list: list to use
1950 * Remove a packet from a list. The list locks are taken and this
1951 * function is atomic with respect to other list locked calls
1953 * You must know what list the SKB is on.
1955 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
1957 unsigned long flags
;
1959 spin_lock_irqsave(&list
->lock
, flags
);
1960 __skb_unlink(skb
, list
);
1961 spin_unlock_irqrestore(&list
->lock
, flags
);
1963 EXPORT_SYMBOL(skb_unlink
);
1966 * skb_append - append a buffer
1967 * @old: buffer to insert after
1968 * @newsk: buffer to insert
1969 * @list: list to use
1971 * Place a packet after a given packet in a list. The list locks are taken
1972 * and this function is atomic with respect to other list locked calls.
1973 * A buffer cannot be placed on two lists at the same time.
1975 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
1977 unsigned long flags
;
1979 spin_lock_irqsave(&list
->lock
, flags
);
1980 __skb_queue_after(list
, old
, newsk
);
1981 spin_unlock_irqrestore(&list
->lock
, flags
);
1983 EXPORT_SYMBOL(skb_append
);
1986 * skb_insert - insert a buffer
1987 * @old: buffer to insert before
1988 * @newsk: buffer to insert
1989 * @list: list to use
1991 * Place a packet before a given packet in a list. The list locks are
1992 * taken and this function is atomic with respect to other list locked
1995 * A buffer cannot be placed on two lists at the same time.
1997 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
1999 unsigned long flags
;
2001 spin_lock_irqsave(&list
->lock
, flags
);
2002 __skb_insert(newsk
, old
->prev
, old
, list
);
2003 spin_unlock_irqrestore(&list
->lock
, flags
);
2005 EXPORT_SYMBOL(skb_insert
);
2007 static inline void skb_split_inside_header(struct sk_buff
*skb
,
2008 struct sk_buff
* skb1
,
2009 const u32 len
, const int pos
)
2013 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
2015 /* And move data appendix as is. */
2016 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
2017 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
2019 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
2020 skb_shinfo(skb
)->nr_frags
= 0;
2021 skb1
->data_len
= skb
->data_len
;
2022 skb1
->len
+= skb1
->data_len
;
2025 skb_set_tail_pointer(skb
, len
);
2028 static inline void skb_split_no_header(struct sk_buff
*skb
,
2029 struct sk_buff
* skb1
,
2030 const u32 len
, int pos
)
2033 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
2035 skb_shinfo(skb
)->nr_frags
= 0;
2036 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
2038 skb
->data_len
= len
- pos
;
2040 for (i
= 0; i
< nfrags
; i
++) {
2041 int size
= skb_shinfo(skb
)->frags
[i
].size
;
2043 if (pos
+ size
> len
) {
2044 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
2048 * We have two variants in this case:
2049 * 1. Move all the frag to the second
2050 * part, if it is possible. F.e.
2051 * this approach is mandatory for TUX,
2052 * where splitting is expensive.
2053 * 2. Split is accurately. We make this.
2055 get_page(skb_shinfo(skb
)->frags
[i
].page
);
2056 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
2057 skb_shinfo(skb1
)->frags
[0].size
-= len
- pos
;
2058 skb_shinfo(skb
)->frags
[i
].size
= len
- pos
;
2059 skb_shinfo(skb
)->nr_frags
++;
2063 skb_shinfo(skb
)->nr_frags
++;
2066 skb_shinfo(skb1
)->nr_frags
= k
;
2070 * skb_split - Split fragmented skb to two parts at length len.
2071 * @skb: the buffer to split
2072 * @skb1: the buffer to receive the second part
2073 * @len: new length for skb
2075 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
2077 int pos
= skb_headlen(skb
);
2079 if (len
< pos
) /* Split line is inside header. */
2080 skb_split_inside_header(skb
, skb1
, len
, pos
);
2081 else /* Second chunk has no header, nothing to copy. */
2082 skb_split_no_header(skb
, skb1
, len
, pos
);
2084 EXPORT_SYMBOL(skb_split
);
2086 /* Shifting from/to a cloned skb is a no-go.
2088 * Caller cannot keep skb_shinfo related pointers past calling here!
2090 static int skb_prepare_for_shift(struct sk_buff
*skb
)
2092 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2096 * skb_shift - Shifts paged data partially from skb to another
2097 * @tgt: buffer into which tail data gets added
2098 * @skb: buffer from which the paged data comes from
2099 * @shiftlen: shift up to this many bytes
2101 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2102 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2103 * It's up to caller to free skb if everything was shifted.
2105 * If @tgt runs out of frags, the whole operation is aborted.
2107 * Skb cannot include anything else but paged data while tgt is allowed
2108 * to have non-paged data as well.
2110 * TODO: full sized shift could be optimized but that would need
2111 * specialized skb free'er to handle frags without up-to-date nr_frags.
2113 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
2115 int from
, to
, merge
, todo
;
2116 struct skb_frag_struct
*fragfrom
, *fragto
;
2118 BUG_ON(shiftlen
> skb
->len
);
2119 BUG_ON(skb_headlen(skb
)); /* Would corrupt stream */
2123 to
= skb_shinfo(tgt
)->nr_frags
;
2124 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2126 /* Actual merge is delayed until the point when we know we can
2127 * commit all, so that we don't have to undo partial changes
2130 !skb_can_coalesce(tgt
, to
, fragfrom
->page
, fragfrom
->page_offset
)) {
2135 todo
-= fragfrom
->size
;
2137 if (skb_prepare_for_shift(skb
) ||
2138 skb_prepare_for_shift(tgt
))
2141 /* All previous frag pointers might be stale! */
2142 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2143 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2145 fragto
->size
+= shiftlen
;
2146 fragfrom
->size
-= shiftlen
;
2147 fragfrom
->page_offset
+= shiftlen
;
2155 /* Skip full, not-fitting skb to avoid expensive operations */
2156 if ((shiftlen
== skb
->len
) &&
2157 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
2160 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
2163 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
2164 if (to
== MAX_SKB_FRAGS
)
2167 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2168 fragto
= &skb_shinfo(tgt
)->frags
[to
];
2170 if (todo
>= fragfrom
->size
) {
2171 *fragto
= *fragfrom
;
2172 todo
-= fragfrom
->size
;
2177 get_page(fragfrom
->page
);
2178 fragto
->page
= fragfrom
->page
;
2179 fragto
->page_offset
= fragfrom
->page_offset
;
2180 fragto
->size
= todo
;
2182 fragfrom
->page_offset
+= todo
;
2183 fragfrom
->size
-= todo
;
2191 /* Ready to "commit" this state change to tgt */
2192 skb_shinfo(tgt
)->nr_frags
= to
;
2195 fragfrom
= &skb_shinfo(skb
)->frags
[0];
2196 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2198 fragto
->size
+= fragfrom
->size
;
2199 put_page(fragfrom
->page
);
2202 /* Reposition in the original skb */
2204 while (from
< skb_shinfo(skb
)->nr_frags
)
2205 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
2206 skb_shinfo(skb
)->nr_frags
= to
;
2208 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
2211 /* Most likely the tgt won't ever need its checksum anymore, skb on
2212 * the other hand might need it if it needs to be resent
2214 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
2215 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2217 /* Yak, is it really working this way? Some helper please? */
2218 skb
->len
-= shiftlen
;
2219 skb
->data_len
-= shiftlen
;
2220 skb
->truesize
-= shiftlen
;
2221 tgt
->len
+= shiftlen
;
2222 tgt
->data_len
+= shiftlen
;
2223 tgt
->truesize
+= shiftlen
;
2229 * skb_prepare_seq_read - Prepare a sequential read of skb data
2230 * @skb: the buffer to read
2231 * @from: lower offset of data to be read
2232 * @to: upper offset of data to be read
2233 * @st: state variable
2235 * Initializes the specified state variable. Must be called before
2236 * invoking skb_seq_read() for the first time.
2238 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
2239 unsigned int to
, struct skb_seq_state
*st
)
2241 st
->lower_offset
= from
;
2242 st
->upper_offset
= to
;
2243 st
->root_skb
= st
->cur_skb
= skb
;
2244 st
->frag_idx
= st
->stepped_offset
= 0;
2245 st
->frag_data
= NULL
;
2247 EXPORT_SYMBOL(skb_prepare_seq_read
);
2250 * skb_seq_read - Sequentially read skb data
2251 * @consumed: number of bytes consumed by the caller so far
2252 * @data: destination pointer for data to be returned
2253 * @st: state variable
2255 * Reads a block of skb data at &consumed relative to the
2256 * lower offset specified to skb_prepare_seq_read(). Assigns
2257 * the head of the data block to &data and returns the length
2258 * of the block or 0 if the end of the skb data or the upper
2259 * offset has been reached.
2261 * The caller is not required to consume all of the data
2262 * returned, i.e. &consumed is typically set to the number
2263 * of bytes already consumed and the next call to
2264 * skb_seq_read() will return the remaining part of the block.
2266 * Note 1: The size of each block of data returned can be arbitary,
2267 * this limitation is the cost for zerocopy seqeuental
2268 * reads of potentially non linear data.
2270 * Note 2: Fragment lists within fragments are not implemented
2271 * at the moment, state->root_skb could be replaced with
2272 * a stack for this purpose.
2274 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
2275 struct skb_seq_state
*st
)
2277 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
2280 if (unlikely(abs_offset
>= st
->upper_offset
))
2284 block_limit
= skb_headlen(st
->cur_skb
) + st
->stepped_offset
;
2286 if (abs_offset
< block_limit
&& !st
->frag_data
) {
2287 *data
= st
->cur_skb
->data
+ (abs_offset
- st
->stepped_offset
);
2288 return block_limit
- abs_offset
;
2291 if (st
->frag_idx
== 0 && !st
->frag_data
)
2292 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
2294 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
2295 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
2296 block_limit
= frag
->size
+ st
->stepped_offset
;
2298 if (abs_offset
< block_limit
) {
2300 st
->frag_data
= kmap_skb_frag(frag
);
2302 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
2303 (abs_offset
- st
->stepped_offset
);
2305 return block_limit
- abs_offset
;
2308 if (st
->frag_data
) {
2309 kunmap_skb_frag(st
->frag_data
);
2310 st
->frag_data
= NULL
;
2314 st
->stepped_offset
+= frag
->size
;
2317 if (st
->frag_data
) {
2318 kunmap_skb_frag(st
->frag_data
);
2319 st
->frag_data
= NULL
;
2322 if (st
->root_skb
== st
->cur_skb
&& skb_has_frag_list(st
->root_skb
)) {
2323 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
2326 } else if (st
->cur_skb
->next
) {
2327 st
->cur_skb
= st
->cur_skb
->next
;
2334 EXPORT_SYMBOL(skb_seq_read
);
2337 * skb_abort_seq_read - Abort a sequential read of skb data
2338 * @st: state variable
2340 * Must be called if skb_seq_read() was not called until it
2343 void skb_abort_seq_read(struct skb_seq_state
*st
)
2346 kunmap_skb_frag(st
->frag_data
);
2348 EXPORT_SYMBOL(skb_abort_seq_read
);
2350 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2352 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
2353 struct ts_config
*conf
,
2354 struct ts_state
*state
)
2356 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
2359 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
2361 skb_abort_seq_read(TS_SKB_CB(state
));
2365 * skb_find_text - Find a text pattern in skb data
2366 * @skb: the buffer to look in
2367 * @from: search offset
2369 * @config: textsearch configuration
2370 * @state: uninitialized textsearch state variable
2372 * Finds a pattern in the skb data according to the specified
2373 * textsearch configuration. Use textsearch_next() to retrieve
2374 * subsequent occurrences of the pattern. Returns the offset
2375 * to the first occurrence or UINT_MAX if no match was found.
2377 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
2378 unsigned int to
, struct ts_config
*config
,
2379 struct ts_state
*state
)
2383 config
->get_next_block
= skb_ts_get_next_block
;
2384 config
->finish
= skb_ts_finish
;
2386 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
2388 ret
= textsearch_find(config
, state
);
2389 return (ret
<= to
- from
? ret
: UINT_MAX
);
2391 EXPORT_SYMBOL(skb_find_text
);
2394 * skb_append_datato_frags: - append the user data to a skb
2395 * @sk: sock structure
2396 * @skb: skb structure to be appened with user data.
2397 * @getfrag: call back function to be used for getting the user data
2398 * @from: pointer to user message iov
2399 * @length: length of the iov message
2401 * Description: This procedure append the user data in the fragment part
2402 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2404 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
2405 int (*getfrag
)(void *from
, char *to
, int offset
,
2406 int len
, int odd
, struct sk_buff
*skb
),
2407 void *from
, int length
)
2410 skb_frag_t
*frag
= NULL
;
2411 struct page
*page
= NULL
;
2417 /* Return error if we don't have space for new frag */
2418 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2419 if (frg_cnt
>= MAX_SKB_FRAGS
)
2422 /* allocate a new page for next frag */
2423 page
= alloc_pages(sk
->sk_allocation
, 0);
2425 /* If alloc_page fails just return failure and caller will
2426 * free previous allocated pages by doing kfree_skb()
2431 /* initialize the next frag */
2432 sk
->sk_sndmsg_page
= page
;
2433 sk
->sk_sndmsg_off
= 0;
2434 skb_fill_page_desc(skb
, frg_cnt
, page
, 0, 0);
2435 skb
->truesize
+= PAGE_SIZE
;
2436 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
2438 /* get the new initialized frag */
2439 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2440 frag
= &skb_shinfo(skb
)->frags
[frg_cnt
- 1];
2442 /* copy the user data to page */
2443 left
= PAGE_SIZE
- frag
->page_offset
;
2444 copy
= (length
> left
)? left
: length
;
2446 ret
= getfrag(from
, (page_address(frag
->page
) +
2447 frag
->page_offset
+ frag
->size
),
2448 offset
, copy
, 0, skb
);
2452 /* copy was successful so update the size parameters */
2453 sk
->sk_sndmsg_off
+= copy
;
2456 skb
->data_len
+= copy
;
2460 } while (length
> 0);
2464 EXPORT_SYMBOL(skb_append_datato_frags
);
2467 * skb_pull_rcsum - pull skb and update receive checksum
2468 * @skb: buffer to update
2469 * @len: length of data pulled
2471 * This function performs an skb_pull on the packet and updates
2472 * the CHECKSUM_COMPLETE checksum. It should be used on
2473 * receive path processing instead of skb_pull unless you know
2474 * that the checksum difference is zero (e.g., a valid IP header)
2475 * or you are setting ip_summed to CHECKSUM_NONE.
2477 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
2479 BUG_ON(len
> skb
->len
);
2481 BUG_ON(skb
->len
< skb
->data_len
);
2482 skb_postpull_rcsum(skb
, skb
->data
, len
);
2483 return skb
->data
+= len
;
2485 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
2488 * skb_segment - Perform protocol segmentation on skb.
2489 * @skb: buffer to segment
2490 * @features: features for the output path (see dev->features)
2492 * This function performs segmentation on the given skb. It returns
2493 * a pointer to the first in a list of new skbs for the segments.
2494 * In case of error it returns ERR_PTR(err).
2496 struct sk_buff
*skb_segment(struct sk_buff
*skb
, int features
)
2498 struct sk_buff
*segs
= NULL
;
2499 struct sk_buff
*tail
= NULL
;
2500 struct sk_buff
*fskb
= skb_shinfo(skb
)->frag_list
;
2501 unsigned int mss
= skb_shinfo(skb
)->gso_size
;
2502 unsigned int doffset
= skb
->data
- skb_mac_header(skb
);
2503 unsigned int offset
= doffset
;
2504 unsigned int headroom
;
2506 int sg
= features
& NETIF_F_SG
;
2507 int nfrags
= skb_shinfo(skb
)->nr_frags
;
2512 __skb_push(skb
, doffset
);
2513 headroom
= skb_headroom(skb
);
2514 pos
= skb_headlen(skb
);
2517 struct sk_buff
*nskb
;
2522 len
= skb
->len
- offset
;
2526 hsize
= skb_headlen(skb
) - offset
;
2529 if (hsize
> len
|| !sg
)
2532 if (!hsize
&& i
>= nfrags
) {
2533 BUG_ON(fskb
->len
!= len
);
2536 nskb
= skb_clone(fskb
, GFP_ATOMIC
);
2539 if (unlikely(!nskb
))
2542 hsize
= skb_end_pointer(nskb
) - nskb
->head
;
2543 if (skb_cow_head(nskb
, doffset
+ headroom
)) {
2548 nskb
->truesize
+= skb_end_pointer(nskb
) - nskb
->head
-
2550 skb_release_head_state(nskb
);
2551 __skb_push(nskb
, doffset
);
2553 nskb
= alloc_skb(hsize
+ doffset
+ headroom
,
2556 if (unlikely(!nskb
))
2559 skb_reserve(nskb
, headroom
);
2560 __skb_put(nskb
, doffset
);
2569 __copy_skb_header(nskb
, skb
);
2570 nskb
->mac_len
= skb
->mac_len
;
2572 /* nskb and skb might have different headroom */
2573 if (nskb
->ip_summed
== CHECKSUM_PARTIAL
)
2574 nskb
->csum_start
+= skb_headroom(nskb
) - headroom
;
2576 skb_reset_mac_header(nskb
);
2577 skb_set_network_header(nskb
, skb
->mac_len
);
2578 nskb
->transport_header
= (nskb
->network_header
+
2579 skb_network_header_len(skb
));
2580 skb_copy_from_linear_data(skb
, nskb
->data
, doffset
);
2582 if (fskb
!= skb_shinfo(skb
)->frag_list
)
2586 nskb
->ip_summed
= CHECKSUM_NONE
;
2587 nskb
->csum
= skb_copy_and_csum_bits(skb
, offset
,
2593 frag
= skb_shinfo(nskb
)->frags
;
2595 skb_copy_from_linear_data_offset(skb
, offset
,
2596 skb_put(nskb
, hsize
), hsize
);
2598 while (pos
< offset
+ len
&& i
< nfrags
) {
2599 *frag
= skb_shinfo(skb
)->frags
[i
];
2600 get_page(frag
->page
);
2604 frag
->page_offset
+= offset
- pos
;
2605 frag
->size
-= offset
- pos
;
2608 skb_shinfo(nskb
)->nr_frags
++;
2610 if (pos
+ size
<= offset
+ len
) {
2614 frag
->size
-= pos
+ size
- (offset
+ len
);
2621 if (pos
< offset
+ len
) {
2622 struct sk_buff
*fskb2
= fskb
;
2624 BUG_ON(pos
+ fskb
->len
!= offset
+ len
);
2630 fskb2
= skb_clone(fskb2
, GFP_ATOMIC
);
2636 SKB_FRAG_ASSERT(nskb
);
2637 skb_shinfo(nskb
)->frag_list
= fskb2
;
2641 nskb
->data_len
= len
- hsize
;
2642 nskb
->len
+= nskb
->data_len
;
2643 nskb
->truesize
+= nskb
->data_len
;
2644 } while ((offset
+= len
) < skb
->len
);
2649 while ((skb
= segs
)) {
2653 return ERR_PTR(err
);
2655 EXPORT_SYMBOL_GPL(skb_segment
);
2657 int skb_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2659 struct sk_buff
*p
= *head
;
2660 struct sk_buff
*nskb
;
2661 struct skb_shared_info
*skbinfo
= skb_shinfo(skb
);
2662 struct skb_shared_info
*pinfo
= skb_shinfo(p
);
2663 unsigned int headroom
;
2664 unsigned int len
= skb_gro_len(skb
);
2665 unsigned int offset
= skb_gro_offset(skb
);
2666 unsigned int headlen
= skb_headlen(skb
);
2668 if (p
->len
+ len
>= 65536)
2671 if (pinfo
->frag_list
)
2673 else if (headlen
<= offset
) {
2676 int i
= skbinfo
->nr_frags
;
2677 int nr_frags
= pinfo
->nr_frags
+ i
;
2681 if (nr_frags
> MAX_SKB_FRAGS
)
2684 pinfo
->nr_frags
= nr_frags
;
2685 skbinfo
->nr_frags
= 0;
2687 frag
= pinfo
->frags
+ nr_frags
;
2688 frag2
= skbinfo
->frags
+ i
;
2693 frag
->page_offset
+= offset
;
2694 frag
->size
-= offset
;
2696 skb
->truesize
-= skb
->data_len
;
2697 skb
->len
-= skb
->data_len
;
2700 NAPI_GRO_CB(skb
)->free
= 1;
2702 } else if (skb_gro_len(p
) != pinfo
->gso_size
)
2705 headroom
= skb_headroom(p
);
2706 nskb
= alloc_skb(headroom
+ skb_gro_offset(p
), GFP_ATOMIC
);
2707 if (unlikely(!nskb
))
2710 __copy_skb_header(nskb
, p
);
2711 nskb
->mac_len
= p
->mac_len
;
2713 skb_reserve(nskb
, headroom
);
2714 __skb_put(nskb
, skb_gro_offset(p
));
2716 skb_set_mac_header(nskb
, skb_mac_header(p
) - p
->data
);
2717 skb_set_network_header(nskb
, skb_network_offset(p
));
2718 skb_set_transport_header(nskb
, skb_transport_offset(p
));
2720 __skb_pull(p
, skb_gro_offset(p
));
2721 memcpy(skb_mac_header(nskb
), skb_mac_header(p
),
2722 p
->data
- skb_mac_header(p
));
2724 *NAPI_GRO_CB(nskb
) = *NAPI_GRO_CB(p
);
2725 skb_shinfo(nskb
)->frag_list
= p
;
2726 skb_shinfo(nskb
)->gso_size
= pinfo
->gso_size
;
2727 pinfo
->gso_size
= 0;
2728 skb_header_release(p
);
2731 nskb
->data_len
+= p
->len
;
2732 nskb
->truesize
+= p
->len
;
2733 nskb
->len
+= p
->len
;
2736 nskb
->next
= p
->next
;
2742 if (offset
> headlen
) {
2743 skbinfo
->frags
[0].page_offset
+= offset
- headlen
;
2744 skbinfo
->frags
[0].size
-= offset
- headlen
;
2748 __skb_pull(skb
, offset
);
2750 p
->prev
->next
= skb
;
2752 skb_header_release(skb
);
2755 NAPI_GRO_CB(p
)->count
++;
2760 NAPI_GRO_CB(skb
)->same_flow
= 1;
2763 EXPORT_SYMBOL_GPL(skb_gro_receive
);
2765 void __init
skb_init(void)
2767 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
2768 sizeof(struct sk_buff
),
2770 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2772 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
2773 (2*sizeof(struct sk_buff
)) +
2776 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2781 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2782 * @skb: Socket buffer containing the buffers to be mapped
2783 * @sg: The scatter-gather list to map into
2784 * @offset: The offset into the buffer's contents to start mapping
2785 * @len: Length of buffer space to be mapped
2787 * Fill the specified scatter-gather list with mappings/pointers into a
2788 * region of the buffer space attached to a socket buffer.
2791 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2793 int start
= skb_headlen(skb
);
2794 int i
, copy
= start
- offset
;
2795 struct sk_buff
*frag_iter
;
2801 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
2803 if ((len
-= copy
) == 0)
2808 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2811 WARN_ON(start
> offset
+ len
);
2813 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
2814 if ((copy
= end
- offset
) > 0) {
2815 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2819 sg_set_page(&sg
[elt
], frag
->page
, copy
,
2820 frag
->page_offset
+offset
-start
);
2829 skb_walk_frags(skb
, frag_iter
) {
2832 WARN_ON(start
> offset
+ len
);
2834 end
= start
+ frag_iter
->len
;
2835 if ((copy
= end
- offset
) > 0) {
2838 elt
+= __skb_to_sgvec(frag_iter
, sg
+elt
, offset
- start
,
2840 if ((len
-= copy
) == 0)
2850 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2852 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
);
2854 sg_mark_end(&sg
[nsg
- 1]);
2858 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
2861 * skb_cow_data - Check that a socket buffer's data buffers are writable
2862 * @skb: The socket buffer to check.
2863 * @tailbits: Amount of trailing space to be added
2864 * @trailer: Returned pointer to the skb where the @tailbits space begins
2866 * Make sure that the data buffers attached to a socket buffer are
2867 * writable. If they are not, private copies are made of the data buffers
2868 * and the socket buffer is set to use these instead.
2870 * If @tailbits is given, make sure that there is space to write @tailbits
2871 * bytes of data beyond current end of socket buffer. @trailer will be
2872 * set to point to the skb in which this space begins.
2874 * The number of scatterlist elements required to completely map the
2875 * COW'd and extended socket buffer will be returned.
2877 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
2881 struct sk_buff
*skb1
, **skb_p
;
2883 /* If skb is cloned or its head is paged, reallocate
2884 * head pulling out all the pages (pages are considered not writable
2885 * at the moment even if they are anonymous).
2887 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
2888 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
2891 /* Easy case. Most of packets will go this way. */
2892 if (!skb_has_frag_list(skb
)) {
2893 /* A little of trouble, not enough of space for trailer.
2894 * This should not happen, when stack is tuned to generate
2895 * good frames. OK, on miss we reallocate and reserve even more
2896 * space, 128 bytes is fair. */
2898 if (skb_tailroom(skb
) < tailbits
&&
2899 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
2907 /* Misery. We are in troubles, going to mincer fragments... */
2910 skb_p
= &skb_shinfo(skb
)->frag_list
;
2913 while ((skb1
= *skb_p
) != NULL
) {
2916 /* The fragment is partially pulled by someone,
2917 * this can happen on input. Copy it and everything
2920 if (skb_shared(skb1
))
2923 /* If the skb is the last, worry about trailer. */
2925 if (skb1
->next
== NULL
&& tailbits
) {
2926 if (skb_shinfo(skb1
)->nr_frags
||
2927 skb_has_frag_list(skb1
) ||
2928 skb_tailroom(skb1
) < tailbits
)
2929 ntail
= tailbits
+ 128;
2935 skb_shinfo(skb1
)->nr_frags
||
2936 skb_has_frag_list(skb1
)) {
2937 struct sk_buff
*skb2
;
2939 /* Fuck, we are miserable poor guys... */
2941 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
2943 skb2
= skb_copy_expand(skb1
,
2947 if (unlikely(skb2
== NULL
))
2951 skb_set_owner_w(skb2
, skb1
->sk
);
2953 /* Looking around. Are we still alive?
2954 * OK, link new skb, drop old one */
2956 skb2
->next
= skb1
->next
;
2963 skb_p
= &skb1
->next
;
2968 EXPORT_SYMBOL_GPL(skb_cow_data
);
2970 static void sock_rmem_free(struct sk_buff
*skb
)
2972 struct sock
*sk
= skb
->sk
;
2974 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
2978 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
2980 int sock_queue_err_skb(struct sock
*sk
, struct sk_buff
*skb
)
2982 if (atomic_read(&sk
->sk_rmem_alloc
) + skb
->truesize
>=
2983 (unsigned)sk
->sk_rcvbuf
)
2988 skb
->destructor
= sock_rmem_free
;
2989 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
2991 skb_queue_tail(&sk
->sk_error_queue
, skb
);
2992 if (!sock_flag(sk
, SOCK_DEAD
))
2993 sk
->sk_data_ready(sk
, skb
->len
);
2996 EXPORT_SYMBOL(sock_queue_err_skb
);
2998 void skb_tstamp_tx(struct sk_buff
*orig_skb
,
2999 struct skb_shared_hwtstamps
*hwtstamps
)
3001 struct sock
*sk
= orig_skb
->sk
;
3002 struct sock_exterr_skb
*serr
;
3003 struct sk_buff
*skb
;
3009 skb
= skb_clone(orig_skb
, GFP_ATOMIC
);
3014 *skb_hwtstamps(skb
) =
3018 * no hardware time stamps available,
3019 * so keep the shared tx_flags and only
3020 * store software time stamp
3022 skb
->tstamp
= ktime_get_real();
3025 serr
= SKB_EXT_ERR(skb
);
3026 memset(serr
, 0, sizeof(*serr
));
3027 serr
->ee
.ee_errno
= ENOMSG
;
3028 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TIMESTAMPING
;
3030 err
= sock_queue_err_skb(sk
, skb
);
3035 EXPORT_SYMBOL_GPL(skb_tstamp_tx
);
3039 * skb_partial_csum_set - set up and verify partial csum values for packet
3040 * @skb: the skb to set
3041 * @start: the number of bytes after skb->data to start checksumming.
3042 * @off: the offset from start to place the checksum.
3044 * For untrusted partially-checksummed packets, we need to make sure the values
3045 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3047 * This function checks and sets those values and skb->ip_summed: if this
3048 * returns false you should drop the packet.
3050 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
3052 if (unlikely(start
> skb_headlen(skb
)) ||
3053 unlikely((int)start
+ off
> skb_headlen(skb
) - 2)) {
3054 if (net_ratelimit())
3056 "bad partial csum: csum=%u/%u len=%u\n",
3057 start
, off
, skb_headlen(skb
));
3060 skb
->ip_summed
= CHECKSUM_PARTIAL
;
3061 skb
->csum_start
= skb_headroom(skb
) + start
;
3062 skb
->csum_offset
= off
;
3065 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);
3067 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
3069 if (net_ratelimit())
3070 pr_warning("%s: received packets cannot be forwarded"
3071 " while LRO is enabled\n", skb
->dev
->name
);
3073 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);