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>
60 #include <linux/prefetch.h>
62 #include <net/protocol.h>
65 #include <net/checksum.h>
68 #include <asm/uaccess.h>
69 #include <asm/system.h>
70 #include <trace/events/skb.h>
74 static struct kmem_cache
*skbuff_head_cache __read_mostly
;
75 static struct kmem_cache
*skbuff_fclone_cache __read_mostly
;
77 static void sock_pipe_buf_release(struct pipe_inode_info
*pipe
,
78 struct pipe_buffer
*buf
)
83 static void sock_pipe_buf_get(struct pipe_inode_info
*pipe
,
84 struct pipe_buffer
*buf
)
89 static int sock_pipe_buf_steal(struct pipe_inode_info
*pipe
,
90 struct pipe_buffer
*buf
)
96 /* Pipe buffer operations for a socket. */
97 static const struct pipe_buf_operations sock_pipe_buf_ops
= {
99 .map
= generic_pipe_buf_map
,
100 .unmap
= generic_pipe_buf_unmap
,
101 .confirm
= generic_pipe_buf_confirm
,
102 .release
= sock_pipe_buf_release
,
103 .steal
= sock_pipe_buf_steal
,
104 .get
= sock_pipe_buf_get
,
108 * Keep out-of-line to prevent kernel bloat.
109 * __builtin_return_address is not used because it is not always
114 * skb_over_panic - private function
119 * Out of line support code for skb_put(). Not user callable.
121 static void skb_over_panic(struct sk_buff
*skb
, int sz
, void *here
)
123 printk(KERN_EMERG
"skb_over_panic: text:%p len:%d put:%d head:%p "
124 "data:%p tail:%#lx end:%#lx dev:%s\n",
125 here
, skb
->len
, sz
, skb
->head
, skb
->data
,
126 (unsigned long)skb
->tail
, (unsigned long)skb
->end
,
127 skb
->dev
? skb
->dev
->name
: "<NULL>");
132 * skb_under_panic - private function
137 * Out of line support code for skb_push(). Not user callable.
140 static void skb_under_panic(struct sk_buff
*skb
, int sz
, void *here
)
142 printk(KERN_EMERG
"skb_under_panic: text:%p len:%d put:%d head:%p "
143 "data:%p tail:%#lx end:%#lx dev:%s\n",
144 here
, skb
->len
, sz
, skb
->head
, skb
->data
,
145 (unsigned long)skb
->tail
, (unsigned long)skb
->end
,
146 skb
->dev
? skb
->dev
->name
: "<NULL>");
150 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
151 * 'private' fields and also do memory statistics to find all the
157 * __alloc_skb - allocate a network buffer
158 * @size: size to allocate
159 * @gfp_mask: allocation mask
160 * @fclone: allocate from fclone cache instead of head cache
161 * and allocate a cloned (child) skb
162 * @node: numa node to allocate memory on
164 * Allocate a new &sk_buff. The returned buffer has no headroom and a
165 * tail room of size bytes. The object has a reference count of one.
166 * The return is the buffer. On a failure the return is %NULL.
168 * Buffers may only be allocated from interrupts using a @gfp_mask of
171 struct sk_buff
*__alloc_skb(unsigned int size
, gfp_t gfp_mask
,
172 int fclone
, int node
)
174 struct kmem_cache
*cache
;
175 struct skb_shared_info
*shinfo
;
179 cache
= fclone
? skbuff_fclone_cache
: skbuff_head_cache
;
182 skb
= kmem_cache_alloc_node(cache
, gfp_mask
& ~__GFP_DMA
, node
);
187 size
= SKB_DATA_ALIGN(size
);
188 data
= kmalloc_node_track_caller(size
+ sizeof(struct skb_shared_info
),
192 prefetchw(data
+ size
);
195 * Only clear those fields we need to clear, not those that we will
196 * actually initialise below. Hence, don't put any more fields after
197 * the tail pointer in struct sk_buff!
199 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
200 skb
->truesize
= size
+ sizeof(struct sk_buff
);
201 atomic_set(&skb
->users
, 1);
204 skb_reset_tail_pointer(skb
);
205 skb
->end
= skb
->tail
+ size
;
206 #ifdef NET_SKBUFF_DATA_USES_OFFSET
207 skb
->mac_header
= ~0U;
210 /* make sure we initialize shinfo sequentially */
211 shinfo
= skb_shinfo(skb
);
212 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
213 atomic_set(&shinfo
->dataref
, 1);
214 kmemcheck_annotate_variable(shinfo
->destructor_arg
);
217 struct sk_buff
*child
= skb
+ 1;
218 atomic_t
*fclone_ref
= (atomic_t
*) (child
+ 1);
220 kmemcheck_annotate_bitfield(child
, flags1
);
221 kmemcheck_annotate_bitfield(child
, flags2
);
222 skb
->fclone
= SKB_FCLONE_ORIG
;
223 atomic_set(fclone_ref
, 1);
225 child
->fclone
= SKB_FCLONE_UNAVAILABLE
;
230 kmem_cache_free(cache
, skb
);
234 EXPORT_SYMBOL(__alloc_skb
);
237 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
238 * @dev: network device to receive on
239 * @length: length to allocate
240 * @gfp_mask: get_free_pages mask, passed to alloc_skb
242 * Allocate a new &sk_buff and assign it a usage count of one. The
243 * buffer has unspecified headroom built in. Users should allocate
244 * the headroom they think they need without accounting for the
245 * built in space. The built in space is used for optimisations.
247 * %NULL is returned if there is no free memory.
249 struct sk_buff
*__netdev_alloc_skb(struct net_device
*dev
,
250 unsigned int length
, gfp_t gfp_mask
)
254 skb
= __alloc_skb(length
+ NET_SKB_PAD
, gfp_mask
, 0, NUMA_NO_NODE
);
256 skb_reserve(skb
, NET_SKB_PAD
);
261 EXPORT_SYMBOL(__netdev_alloc_skb
);
263 void skb_add_rx_frag(struct sk_buff
*skb
, int i
, struct page
*page
, int off
,
266 skb_fill_page_desc(skb
, i
, page
, off
, size
);
268 skb
->data_len
+= size
;
269 skb
->truesize
+= size
;
271 EXPORT_SYMBOL(skb_add_rx_frag
);
274 * dev_alloc_skb - allocate an skbuff for receiving
275 * @length: length to allocate
277 * Allocate a new &sk_buff and assign it a usage count of one. The
278 * buffer has unspecified headroom built in. Users should allocate
279 * the headroom they think they need without accounting for the
280 * built in space. The built in space is used for optimisations.
282 * %NULL is returned if there is no free memory. Although this function
283 * allocates memory it can be called from an interrupt.
285 struct sk_buff
*dev_alloc_skb(unsigned int length
)
288 * There is more code here than it seems:
289 * __dev_alloc_skb is an inline
291 return __dev_alloc_skb(length
, GFP_ATOMIC
);
293 EXPORT_SYMBOL(dev_alloc_skb
);
295 static void skb_drop_list(struct sk_buff
**listp
)
297 struct sk_buff
*list
= *listp
;
302 struct sk_buff
*this = list
;
308 static inline void skb_drop_fraglist(struct sk_buff
*skb
)
310 skb_drop_list(&skb_shinfo(skb
)->frag_list
);
313 static void skb_clone_fraglist(struct sk_buff
*skb
)
315 struct sk_buff
*list
;
317 skb_walk_frags(skb
, list
)
321 static void skb_release_data(struct sk_buff
*skb
)
324 !atomic_sub_return(skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1,
325 &skb_shinfo(skb
)->dataref
)) {
326 if (skb_shinfo(skb
)->nr_frags
) {
328 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
329 put_page(skb_shinfo(skb
)->frags
[i
].page
);
333 * If skb buf is from userspace, we need to notify the caller
334 * the lower device DMA has done;
336 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
337 struct ubuf_info
*uarg
;
339 uarg
= skb_shinfo(skb
)->destructor_arg
;
341 uarg
->callback(uarg
);
344 if (skb_has_frag_list(skb
))
345 skb_drop_fraglist(skb
);
352 * Free an skbuff by memory without cleaning the state.
354 static void kfree_skbmem(struct sk_buff
*skb
)
356 struct sk_buff
*other
;
357 atomic_t
*fclone_ref
;
359 switch (skb
->fclone
) {
360 case SKB_FCLONE_UNAVAILABLE
:
361 kmem_cache_free(skbuff_head_cache
, skb
);
364 case SKB_FCLONE_ORIG
:
365 fclone_ref
= (atomic_t
*) (skb
+ 2);
366 if (atomic_dec_and_test(fclone_ref
))
367 kmem_cache_free(skbuff_fclone_cache
, skb
);
370 case SKB_FCLONE_CLONE
:
371 fclone_ref
= (atomic_t
*) (skb
+ 1);
374 /* The clone portion is available for
375 * fast-cloning again.
377 skb
->fclone
= SKB_FCLONE_UNAVAILABLE
;
379 if (atomic_dec_and_test(fclone_ref
))
380 kmem_cache_free(skbuff_fclone_cache
, other
);
385 static void skb_release_head_state(struct sk_buff
*skb
)
389 secpath_put(skb
->sp
);
391 if (skb
->destructor
) {
393 skb
->destructor(skb
);
395 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
396 nf_conntrack_put(skb
->nfct
);
398 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
399 nf_conntrack_put_reasm(skb
->nfct_reasm
);
401 #ifdef CONFIG_BRIDGE_NETFILTER
402 nf_bridge_put(skb
->nf_bridge
);
404 /* XXX: IS this still necessary? - JHS */
405 #ifdef CONFIG_NET_SCHED
407 #ifdef CONFIG_NET_CLS_ACT
413 /* Free everything but the sk_buff shell. */
414 static void skb_release_all(struct sk_buff
*skb
)
416 skb_release_head_state(skb
);
417 skb_release_data(skb
);
421 * __kfree_skb - private function
424 * Free an sk_buff. Release anything attached to the buffer.
425 * Clean the state. This is an internal helper function. Users should
426 * always call kfree_skb
429 void __kfree_skb(struct sk_buff
*skb
)
431 skb_release_all(skb
);
434 EXPORT_SYMBOL(__kfree_skb
);
437 * kfree_skb - free an sk_buff
438 * @skb: buffer to free
440 * Drop a reference to the buffer and free it if the usage count has
443 void kfree_skb(struct sk_buff
*skb
)
447 if (likely(atomic_read(&skb
->users
) == 1))
449 else if (likely(!atomic_dec_and_test(&skb
->users
)))
451 trace_kfree_skb(skb
, __builtin_return_address(0));
454 EXPORT_SYMBOL(kfree_skb
);
457 * consume_skb - free an skbuff
458 * @skb: buffer to free
460 * Drop a ref to the buffer and free it if the usage count has hit zero
461 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
462 * is being dropped after a failure and notes that
464 void consume_skb(struct sk_buff
*skb
)
468 if (likely(atomic_read(&skb
->users
) == 1))
470 else if (likely(!atomic_dec_and_test(&skb
->users
)))
472 trace_consume_skb(skb
);
475 EXPORT_SYMBOL(consume_skb
);
478 * skb_recycle_check - check if skb can be reused for receive
480 * @skb_size: minimum receive buffer size
482 * Checks that the skb passed in is not shared or cloned, and
483 * that it is linear and its head portion at least as large as
484 * skb_size so that it can be recycled as a receive buffer.
485 * If these conditions are met, this function does any necessary
486 * reference count dropping and cleans up the skbuff as if it
487 * just came from __alloc_skb().
489 bool skb_recycle_check(struct sk_buff
*skb
, int skb_size
)
491 struct skb_shared_info
*shinfo
;
496 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
)
499 if (skb_is_nonlinear(skb
) || skb
->fclone
!= SKB_FCLONE_UNAVAILABLE
)
502 skb_size
= SKB_DATA_ALIGN(skb_size
+ NET_SKB_PAD
);
503 if (skb_end_pointer(skb
) - skb
->head
< skb_size
)
506 if (skb_shared(skb
) || skb_cloned(skb
))
509 skb_release_head_state(skb
);
511 shinfo
= skb_shinfo(skb
);
512 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
513 atomic_set(&shinfo
->dataref
, 1);
515 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
516 skb
->data
= skb
->head
+ NET_SKB_PAD
;
517 skb_reset_tail_pointer(skb
);
521 EXPORT_SYMBOL(skb_recycle_check
);
523 static void __copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
525 new->tstamp
= old
->tstamp
;
527 new->transport_header
= old
->transport_header
;
528 new->network_header
= old
->network_header
;
529 new->mac_header
= old
->mac_header
;
530 skb_dst_copy(new, old
);
531 new->rxhash
= old
->rxhash
;
533 new->sp
= secpath_get(old
->sp
);
535 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
536 new->csum
= old
->csum
;
537 new->local_df
= old
->local_df
;
538 new->pkt_type
= old
->pkt_type
;
539 new->ip_summed
= old
->ip_summed
;
540 skb_copy_queue_mapping(new, old
);
541 new->priority
= old
->priority
;
542 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
543 new->ipvs_property
= old
->ipvs_property
;
545 new->protocol
= old
->protocol
;
546 new->mark
= old
->mark
;
547 new->skb_iif
= old
->skb_iif
;
549 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
550 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
551 new->nf_trace
= old
->nf_trace
;
553 #ifdef CONFIG_NET_SCHED
554 new->tc_index
= old
->tc_index
;
555 #ifdef CONFIG_NET_CLS_ACT
556 new->tc_verd
= old
->tc_verd
;
559 new->vlan_tci
= old
->vlan_tci
;
561 skb_copy_secmark(new, old
);
565 * You should not add any new code to this function. Add it to
566 * __copy_skb_header above instead.
568 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
570 #define C(x) n->x = skb->x
572 n
->next
= n
->prev
= NULL
;
574 __copy_skb_header(n
, skb
);
579 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
582 n
->destructor
= NULL
;
588 atomic_set(&n
->users
, 1);
590 atomic_inc(&(skb_shinfo(skb
)->dataref
));
598 * skb_morph - morph one skb into another
599 * @dst: the skb to receive the contents
600 * @src: the skb to supply the contents
602 * This is identical to skb_clone except that the target skb is
603 * supplied by the user.
605 * The target skb is returned upon exit.
607 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
609 skb_release_all(dst
);
610 return __skb_clone(dst
, src
);
612 EXPORT_SYMBOL_GPL(skb_morph
);
614 /* skb frags copy userspace buffers to kernel */
615 static int skb_copy_ubufs(struct sk_buff
*skb
, gfp_t gfp_mask
)
618 int num_frags
= skb_shinfo(skb
)->nr_frags
;
619 struct page
*page
, *head
= NULL
;
620 struct ubuf_info
*uarg
= skb_shinfo(skb
)->destructor_arg
;
622 for (i
= 0; i
< num_frags
; i
++) {
624 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
626 page
= alloc_page(GFP_ATOMIC
);
629 struct page
*next
= (struct page
*)head
->private;
635 vaddr
= kmap_skb_frag(&skb_shinfo(skb
)->frags
[i
]);
636 memcpy(page_address(page
),
637 vaddr
+ f
->page_offset
, f
->size
);
638 kunmap_skb_frag(vaddr
);
639 page
->private = (unsigned long)head
;
643 /* skb frags release userspace buffers */
644 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
645 put_page(skb_shinfo(skb
)->frags
[i
].page
);
647 uarg
->callback(uarg
);
649 /* skb frags point to kernel buffers */
650 for (i
= skb_shinfo(skb
)->nr_frags
; i
> 0; i
--) {
651 skb_shinfo(skb
)->frags
[i
- 1].page_offset
= 0;
652 skb_shinfo(skb
)->frags
[i
- 1].page
= head
;
653 head
= (struct page
*)head
->private;
660 * skb_clone - duplicate an sk_buff
661 * @skb: buffer to clone
662 * @gfp_mask: allocation priority
664 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
665 * copies share the same packet data but not structure. The new
666 * buffer has a reference count of 1. If the allocation fails the
667 * function returns %NULL otherwise the new buffer is returned.
669 * If this function is called from an interrupt gfp_mask() must be
673 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
677 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
678 if (skb_copy_ubufs(skb
, gfp_mask
))
680 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
684 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
685 n
->fclone
== SKB_FCLONE_UNAVAILABLE
) {
686 atomic_t
*fclone_ref
= (atomic_t
*) (n
+ 1);
687 n
->fclone
= SKB_FCLONE_CLONE
;
688 atomic_inc(fclone_ref
);
690 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
694 kmemcheck_annotate_bitfield(n
, flags1
);
695 kmemcheck_annotate_bitfield(n
, flags2
);
696 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
699 return __skb_clone(n
, skb
);
701 EXPORT_SYMBOL(skb_clone
);
703 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
705 #ifndef NET_SKBUFF_DATA_USES_OFFSET
707 * Shift between the two data areas in bytes
709 unsigned long offset
= new->data
- old
->data
;
712 __copy_skb_header(new, old
);
714 #ifndef NET_SKBUFF_DATA_USES_OFFSET
715 /* {transport,network,mac}_header are relative to skb->head */
716 new->transport_header
+= offset
;
717 new->network_header
+= offset
;
718 if (skb_mac_header_was_set(new))
719 new->mac_header
+= offset
;
721 skb_shinfo(new)->gso_size
= skb_shinfo(old
)->gso_size
;
722 skb_shinfo(new)->gso_segs
= skb_shinfo(old
)->gso_segs
;
723 skb_shinfo(new)->gso_type
= skb_shinfo(old
)->gso_type
;
727 * skb_copy - create private copy of an sk_buff
728 * @skb: buffer to copy
729 * @gfp_mask: allocation priority
731 * Make a copy of both an &sk_buff and its data. This is used when the
732 * caller wishes to modify the data and needs a private copy of the
733 * data to alter. Returns %NULL on failure or the pointer to the buffer
734 * on success. The returned buffer has a reference count of 1.
736 * As by-product this function converts non-linear &sk_buff to linear
737 * one, so that &sk_buff becomes completely private and caller is allowed
738 * to modify all the data of returned buffer. This means that this
739 * function is not recommended for use in circumstances when only
740 * header is going to be modified. Use pskb_copy() instead.
743 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
745 int headerlen
= skb_headroom(skb
);
746 unsigned int size
= (skb_end_pointer(skb
) - skb
->head
) + skb
->data_len
;
747 struct sk_buff
*n
= alloc_skb(size
, gfp_mask
);
752 /* Set the data pointer */
753 skb_reserve(n
, headerlen
);
754 /* Set the tail pointer and length */
755 skb_put(n
, skb
->len
);
757 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
760 copy_skb_header(n
, skb
);
763 EXPORT_SYMBOL(skb_copy
);
766 * pskb_copy - create copy of an sk_buff with private head.
767 * @skb: buffer to copy
768 * @gfp_mask: allocation priority
770 * Make a copy of both an &sk_buff and part of its data, located
771 * in header. Fragmented data remain shared. This is used when
772 * the caller wishes to modify only header of &sk_buff and needs
773 * private copy of the header to alter. Returns %NULL on failure
774 * or the pointer to the buffer on success.
775 * The returned buffer has a reference count of 1.
778 struct sk_buff
*pskb_copy(struct sk_buff
*skb
, gfp_t gfp_mask
)
780 unsigned int size
= skb_end_pointer(skb
) - skb
->head
;
781 struct sk_buff
*n
= alloc_skb(size
, gfp_mask
);
786 /* Set the data pointer */
787 skb_reserve(n
, skb_headroom(skb
));
788 /* Set the tail pointer and length */
789 skb_put(n
, skb_headlen(skb
));
791 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
793 n
->truesize
+= skb
->data_len
;
794 n
->data_len
= skb
->data_len
;
797 if (skb_shinfo(skb
)->nr_frags
) {
800 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
801 if (skb_copy_ubufs(skb
, gfp_mask
)) {
805 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
807 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
808 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
809 get_page(skb_shinfo(n
)->frags
[i
].page
);
811 skb_shinfo(n
)->nr_frags
= i
;
814 if (skb_has_frag_list(skb
)) {
815 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
816 skb_clone_fraglist(n
);
819 copy_skb_header(n
, skb
);
823 EXPORT_SYMBOL(pskb_copy
);
826 * pskb_expand_head - reallocate header of &sk_buff
827 * @skb: buffer to reallocate
828 * @nhead: room to add at head
829 * @ntail: room to add at tail
830 * @gfp_mask: allocation priority
832 * Expands (or creates identical copy, if &nhead and &ntail are zero)
833 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
834 * reference count of 1. Returns zero in the case of success or error,
835 * if expansion failed. In the last case, &sk_buff is not changed.
837 * All the pointers pointing into skb header may change and must be
838 * reloaded after call to this function.
841 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
846 int size
= nhead
+ (skb_end_pointer(skb
) - skb
->head
) + ntail
;
855 size
= SKB_DATA_ALIGN(size
);
857 /* Check if we can avoid taking references on fragments if we own
858 * the last reference on skb->head. (see skb_release_data())
863 int delta
= skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1;
864 fastpath
= atomic_read(&skb_shinfo(skb
)->dataref
) == delta
;
868 size
+ sizeof(struct skb_shared_info
) <= ksize(skb
->head
)) {
869 memmove(skb
->head
+ size
, skb_shinfo(skb
),
870 offsetof(struct skb_shared_info
,
871 frags
[skb_shinfo(skb
)->nr_frags
]));
872 memmove(skb
->head
+ nhead
, skb
->head
,
873 skb_tail_pointer(skb
) - skb
->head
);
878 data
= kmalloc(size
+ sizeof(struct skb_shared_info
), gfp_mask
);
882 /* Copy only real data... and, alas, header. This should be
883 * optimized for the cases when header is void.
885 memcpy(data
+ nhead
, skb
->head
, skb_tail_pointer(skb
) - skb
->head
);
887 memcpy((struct skb_shared_info
*)(data
+ size
),
889 offsetof(struct skb_shared_info
, frags
[skb_shinfo(skb
)->nr_frags
]));
894 /* copy this zero copy skb frags */
895 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
896 if (skb_copy_ubufs(skb
, gfp_mask
))
898 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
900 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
901 get_page(skb_shinfo(skb
)->frags
[i
].page
);
903 if (skb_has_frag_list(skb
))
904 skb_clone_fraglist(skb
);
906 skb_release_data(skb
);
908 off
= (data
+ nhead
) - skb
->head
;
913 #ifdef NET_SKBUFF_DATA_USES_OFFSET
917 skb
->end
= skb
->head
+ size
;
919 /* {transport,network,mac}_header and tail are relative to skb->head */
921 skb
->transport_header
+= off
;
922 skb
->network_header
+= off
;
923 if (skb_mac_header_was_set(skb
))
924 skb
->mac_header
+= off
;
925 /* Only adjust this if it actually is csum_start rather than csum */
926 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
927 skb
->csum_start
+= nhead
;
931 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
939 EXPORT_SYMBOL(pskb_expand_head
);
941 /* Make private copy of skb with writable head and some headroom */
943 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
945 struct sk_buff
*skb2
;
946 int delta
= headroom
- skb_headroom(skb
);
949 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
951 skb2
= skb_clone(skb
, GFP_ATOMIC
);
952 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
960 EXPORT_SYMBOL(skb_realloc_headroom
);
963 * skb_copy_expand - copy and expand sk_buff
964 * @skb: buffer to copy
965 * @newheadroom: new free bytes at head
966 * @newtailroom: new free bytes at tail
967 * @gfp_mask: allocation priority
969 * Make a copy of both an &sk_buff and its data and while doing so
970 * allocate additional space.
972 * This is used when the caller wishes to modify the data and needs a
973 * private copy of the data to alter as well as more space for new fields.
974 * Returns %NULL on failure or the pointer to the buffer
975 * on success. The returned buffer has a reference count of 1.
977 * You must pass %GFP_ATOMIC as the allocation priority if this function
978 * is called from an interrupt.
980 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
981 int newheadroom
, int newtailroom
,
985 * Allocate the copy buffer
987 struct sk_buff
*n
= alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
989 int oldheadroom
= skb_headroom(skb
);
990 int head_copy_len
, head_copy_off
;
996 skb_reserve(n
, newheadroom
);
998 /* Set the tail pointer and length */
999 skb_put(n
, skb
->len
);
1001 head_copy_len
= oldheadroom
;
1003 if (newheadroom
<= head_copy_len
)
1004 head_copy_len
= newheadroom
;
1006 head_copy_off
= newheadroom
- head_copy_len
;
1008 /* Copy the linear header and data. */
1009 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
1010 skb
->len
+ head_copy_len
))
1013 copy_skb_header(n
, skb
);
1015 off
= newheadroom
- oldheadroom
;
1016 if (n
->ip_summed
== CHECKSUM_PARTIAL
)
1017 n
->csum_start
+= off
;
1018 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1019 n
->transport_header
+= off
;
1020 n
->network_header
+= off
;
1021 if (skb_mac_header_was_set(skb
))
1022 n
->mac_header
+= off
;
1027 EXPORT_SYMBOL(skb_copy_expand
);
1030 * skb_pad - zero pad the tail of an skb
1031 * @skb: buffer to pad
1032 * @pad: space to pad
1034 * Ensure that a buffer is followed by a padding area that is zero
1035 * filled. Used by network drivers which may DMA or transfer data
1036 * beyond the buffer end onto the wire.
1038 * May return error in out of memory cases. The skb is freed on error.
1041 int skb_pad(struct sk_buff
*skb
, int pad
)
1046 /* If the skbuff is non linear tailroom is always zero.. */
1047 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
1048 memset(skb
->data
+skb
->len
, 0, pad
);
1052 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
1053 if (likely(skb_cloned(skb
) || ntail
> 0)) {
1054 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
1059 /* FIXME: The use of this function with non-linear skb's really needs
1062 err
= skb_linearize(skb
);
1066 memset(skb
->data
+ skb
->len
, 0, pad
);
1073 EXPORT_SYMBOL(skb_pad
);
1076 * skb_put - add data to a buffer
1077 * @skb: buffer to use
1078 * @len: amount of data to add
1080 * This function extends the used data area of the buffer. If this would
1081 * exceed the total buffer size the kernel will panic. A pointer to the
1082 * first byte of the extra data is returned.
1084 unsigned char *skb_put(struct sk_buff
*skb
, unsigned int len
)
1086 unsigned char *tmp
= skb_tail_pointer(skb
);
1087 SKB_LINEAR_ASSERT(skb
);
1090 if (unlikely(skb
->tail
> skb
->end
))
1091 skb_over_panic(skb
, len
, __builtin_return_address(0));
1094 EXPORT_SYMBOL(skb_put
);
1097 * skb_push - add data to the start of a buffer
1098 * @skb: buffer to use
1099 * @len: amount of data to add
1101 * This function extends the used data area of the buffer at the buffer
1102 * start. If this would exceed the total buffer headroom the kernel will
1103 * panic. A pointer to the first byte of the extra data is returned.
1105 unsigned char *skb_push(struct sk_buff
*skb
, unsigned int len
)
1109 if (unlikely(skb
->data
<skb
->head
))
1110 skb_under_panic(skb
, len
, __builtin_return_address(0));
1113 EXPORT_SYMBOL(skb_push
);
1116 * skb_pull - remove data from the start of a buffer
1117 * @skb: buffer to use
1118 * @len: amount of data to remove
1120 * This function removes data from the start of a buffer, returning
1121 * the memory to the headroom. A pointer to the next data in the buffer
1122 * is returned. Once the data has been pulled future pushes will overwrite
1125 unsigned char *skb_pull(struct sk_buff
*skb
, unsigned int len
)
1127 return skb_pull_inline(skb
, len
);
1129 EXPORT_SYMBOL(skb_pull
);
1132 * skb_trim - remove end from a buffer
1133 * @skb: buffer to alter
1136 * Cut the length of a buffer down by removing data from the tail. If
1137 * the buffer is already under the length specified it is not modified.
1138 * The skb must be linear.
1140 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1143 __skb_trim(skb
, len
);
1145 EXPORT_SYMBOL(skb_trim
);
1147 /* Trims skb to length len. It can change skb pointers.
1150 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1152 struct sk_buff
**fragp
;
1153 struct sk_buff
*frag
;
1154 int offset
= skb_headlen(skb
);
1155 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1159 if (skb_cloned(skb
) &&
1160 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1167 for (; i
< nfrags
; i
++) {
1168 int end
= offset
+ skb_shinfo(skb
)->frags
[i
].size
;
1175 skb_shinfo(skb
)->frags
[i
++].size
= len
- offset
;
1178 skb_shinfo(skb
)->nr_frags
= i
;
1180 for (; i
< nfrags
; i
++)
1181 put_page(skb_shinfo(skb
)->frags
[i
].page
);
1183 if (skb_has_frag_list(skb
))
1184 skb_drop_fraglist(skb
);
1188 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1189 fragp
= &frag
->next
) {
1190 int end
= offset
+ frag
->len
;
1192 if (skb_shared(frag
)) {
1193 struct sk_buff
*nfrag
;
1195 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1196 if (unlikely(!nfrag
))
1199 nfrag
->next
= frag
->next
;
1211 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1215 skb_drop_list(&frag
->next
);
1220 if (len
> skb_headlen(skb
)) {
1221 skb
->data_len
-= skb
->len
- len
;
1226 skb_set_tail_pointer(skb
, len
);
1231 EXPORT_SYMBOL(___pskb_trim
);
1234 * __pskb_pull_tail - advance tail of skb header
1235 * @skb: buffer to reallocate
1236 * @delta: number of bytes to advance tail
1238 * The function makes a sense only on a fragmented &sk_buff,
1239 * it expands header moving its tail forward and copying necessary
1240 * data from fragmented part.
1242 * &sk_buff MUST have reference count of 1.
1244 * Returns %NULL (and &sk_buff does not change) if pull failed
1245 * or value of new tail of skb in the case of success.
1247 * All the pointers pointing into skb header may change and must be
1248 * reloaded after call to this function.
1251 /* Moves tail of skb head forward, copying data from fragmented part,
1252 * when it is necessary.
1253 * 1. It may fail due to malloc failure.
1254 * 2. It may change skb pointers.
1256 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1258 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1260 /* If skb has not enough free space at tail, get new one
1261 * plus 128 bytes for future expansions. If we have enough
1262 * room at tail, reallocate without expansion only if skb is cloned.
1264 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1266 if (eat
> 0 || skb_cloned(skb
)) {
1267 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1272 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
1275 /* Optimization: no fragments, no reasons to preestimate
1276 * size of pulled pages. Superb.
1278 if (!skb_has_frag_list(skb
))
1281 /* Estimate size of pulled pages. */
1283 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1284 if (skb_shinfo(skb
)->frags
[i
].size
>= eat
)
1286 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1289 /* If we need update frag list, we are in troubles.
1290 * Certainly, it possible to add an offset to skb data,
1291 * but taking into account that pulling is expected to
1292 * be very rare operation, it is worth to fight against
1293 * further bloating skb head and crucify ourselves here instead.
1294 * Pure masohism, indeed. 8)8)
1297 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1298 struct sk_buff
*clone
= NULL
;
1299 struct sk_buff
*insp
= NULL
;
1304 if (list
->len
<= eat
) {
1305 /* Eaten as whole. */
1310 /* Eaten partially. */
1312 if (skb_shared(list
)) {
1313 /* Sucks! We need to fork list. :-( */
1314 clone
= skb_clone(list
, GFP_ATOMIC
);
1320 /* This may be pulled without
1324 if (!pskb_pull(list
, eat
)) {
1332 /* Free pulled out fragments. */
1333 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1334 skb_shinfo(skb
)->frag_list
= list
->next
;
1337 /* And insert new clone at head. */
1340 skb_shinfo(skb
)->frag_list
= clone
;
1343 /* Success! Now we may commit changes to skb data. */
1348 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1349 if (skb_shinfo(skb
)->frags
[i
].size
<= eat
) {
1350 put_page(skb_shinfo(skb
)->frags
[i
].page
);
1351 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1353 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1355 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1356 skb_shinfo(skb
)->frags
[k
].size
-= eat
;
1362 skb_shinfo(skb
)->nr_frags
= k
;
1365 skb
->data_len
-= delta
;
1367 return skb_tail_pointer(skb
);
1369 EXPORT_SYMBOL(__pskb_pull_tail
);
1371 /* Copy some data bits from skb to kernel buffer. */
1373 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1375 int start
= skb_headlen(skb
);
1376 struct sk_buff
*frag_iter
;
1379 if (offset
> (int)skb
->len
- len
)
1383 if ((copy
= start
- offset
) > 0) {
1386 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
1387 if ((len
-= copy
) == 0)
1393 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1396 WARN_ON(start
> offset
+ len
);
1398 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1399 if ((copy
= end
- offset
) > 0) {
1405 vaddr
= kmap_skb_frag(&skb_shinfo(skb
)->frags
[i
]);
1407 vaddr
+ skb_shinfo(skb
)->frags
[i
].page_offset
+
1408 offset
- start
, copy
);
1409 kunmap_skb_frag(vaddr
);
1411 if ((len
-= copy
) == 0)
1419 skb_walk_frags(skb
, frag_iter
) {
1422 WARN_ON(start
> offset
+ len
);
1424 end
= start
+ frag_iter
->len
;
1425 if ((copy
= end
- offset
) > 0) {
1428 if (skb_copy_bits(frag_iter
, offset
- start
, to
, copy
))
1430 if ((len
-= copy
) == 0)
1444 EXPORT_SYMBOL(skb_copy_bits
);
1447 * Callback from splice_to_pipe(), if we need to release some pages
1448 * at the end of the spd in case we error'ed out in filling the pipe.
1450 static void sock_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
1452 put_page(spd
->pages
[i
]);
1455 static inline struct page
*linear_to_page(struct page
*page
, unsigned int *len
,
1456 unsigned int *offset
,
1457 struct sk_buff
*skb
, struct sock
*sk
)
1459 struct page
*p
= sk
->sk_sndmsg_page
;
1464 p
= sk
->sk_sndmsg_page
= alloc_pages(sk
->sk_allocation
, 0);
1468 off
= sk
->sk_sndmsg_off
= 0;
1469 /* hold one ref to this page until it's full */
1473 off
= sk
->sk_sndmsg_off
;
1474 mlen
= PAGE_SIZE
- off
;
1475 if (mlen
< 64 && mlen
< *len
) {
1480 *len
= min_t(unsigned int, *len
, mlen
);
1483 memcpy(page_address(p
) + off
, page_address(page
) + *offset
, *len
);
1484 sk
->sk_sndmsg_off
+= *len
;
1492 * Fill page/offset/length into spd, if it can hold more pages.
1494 static inline int spd_fill_page(struct splice_pipe_desc
*spd
,
1495 struct pipe_inode_info
*pipe
, struct page
*page
,
1496 unsigned int *len
, unsigned int offset
,
1497 struct sk_buff
*skb
, int linear
,
1500 if (unlikely(spd
->nr_pages
== pipe
->buffers
))
1504 page
= linear_to_page(page
, len
, &offset
, skb
, sk
);
1510 spd
->pages
[spd
->nr_pages
] = page
;
1511 spd
->partial
[spd
->nr_pages
].len
= *len
;
1512 spd
->partial
[spd
->nr_pages
].offset
= offset
;
1518 static inline void __segment_seek(struct page
**page
, unsigned int *poff
,
1519 unsigned int *plen
, unsigned int off
)
1524 n
= *poff
/ PAGE_SIZE
;
1526 *page
= nth_page(*page
, n
);
1528 *poff
= *poff
% PAGE_SIZE
;
1532 static inline int __splice_segment(struct page
*page
, unsigned int poff
,
1533 unsigned int plen
, unsigned int *off
,
1534 unsigned int *len
, struct sk_buff
*skb
,
1535 struct splice_pipe_desc
*spd
, int linear
,
1537 struct pipe_inode_info
*pipe
)
1542 /* skip this segment if already processed */
1548 /* ignore any bits we already processed */
1550 __segment_seek(&page
, &poff
, &plen
, *off
);
1555 unsigned int flen
= min(*len
, plen
);
1557 /* the linear region may spread across several pages */
1558 flen
= min_t(unsigned int, flen
, PAGE_SIZE
- poff
);
1560 if (spd_fill_page(spd
, pipe
, page
, &flen
, poff
, skb
, linear
, sk
))
1563 __segment_seek(&page
, &poff
, &plen
, flen
);
1566 } while (*len
&& plen
);
1572 * Map linear and fragment data from the skb to spd. It reports failure if the
1573 * pipe is full or if we already spliced the requested length.
1575 static int __skb_splice_bits(struct sk_buff
*skb
, struct pipe_inode_info
*pipe
,
1576 unsigned int *offset
, unsigned int *len
,
1577 struct splice_pipe_desc
*spd
, struct sock
*sk
)
1582 * map the linear part
1584 if (__splice_segment(virt_to_page(skb
->data
),
1585 (unsigned long) skb
->data
& (PAGE_SIZE
- 1),
1587 offset
, len
, skb
, spd
, 1, sk
, pipe
))
1591 * then map the fragments
1593 for (seg
= 0; seg
< skb_shinfo(skb
)->nr_frags
; seg
++) {
1594 const skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[seg
];
1596 if (__splice_segment(f
->page
, f
->page_offset
, f
->size
,
1597 offset
, len
, skb
, spd
, 0, sk
, pipe
))
1605 * Map data from the skb to a pipe. Should handle both the linear part,
1606 * the fragments, and the frag list. It does NOT handle frag lists within
1607 * the frag list, if such a thing exists. We'd probably need to recurse to
1608 * handle that cleanly.
1610 int skb_splice_bits(struct sk_buff
*skb
, unsigned int offset
,
1611 struct pipe_inode_info
*pipe
, unsigned int tlen
,
1614 struct partial_page partial
[PIPE_DEF_BUFFERS
];
1615 struct page
*pages
[PIPE_DEF_BUFFERS
];
1616 struct splice_pipe_desc spd
= {
1620 .ops
= &sock_pipe_buf_ops
,
1621 .spd_release
= sock_spd_release
,
1623 struct sk_buff
*frag_iter
;
1624 struct sock
*sk
= skb
->sk
;
1627 if (splice_grow_spd(pipe
, &spd
))
1631 * __skb_splice_bits() only fails if the output has no room left,
1632 * so no point in going over the frag_list for the error case.
1634 if (__skb_splice_bits(skb
, pipe
, &offset
, &tlen
, &spd
, sk
))
1640 * now see if we have a frag_list to map
1642 skb_walk_frags(skb
, frag_iter
) {
1645 if (__skb_splice_bits(frag_iter
, pipe
, &offset
, &tlen
, &spd
, sk
))
1652 * Drop the socket lock, otherwise we have reverse
1653 * locking dependencies between sk_lock and i_mutex
1654 * here as compared to sendfile(). We enter here
1655 * with the socket lock held, and splice_to_pipe() will
1656 * grab the pipe inode lock. For sendfile() emulation,
1657 * we call into ->sendpage() with the i_mutex lock held
1658 * and networking will grab the socket lock.
1661 ret
= splice_to_pipe(pipe
, &spd
);
1665 splice_shrink_spd(pipe
, &spd
);
1670 * skb_store_bits - store bits from kernel buffer to skb
1671 * @skb: destination buffer
1672 * @offset: offset in destination
1673 * @from: source buffer
1674 * @len: number of bytes to copy
1676 * Copy the specified number of bytes from the source buffer to the
1677 * destination skb. This function handles all the messy bits of
1678 * traversing fragment lists and such.
1681 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
1683 int start
= skb_headlen(skb
);
1684 struct sk_buff
*frag_iter
;
1687 if (offset
> (int)skb
->len
- len
)
1690 if ((copy
= start
- offset
) > 0) {
1693 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
1694 if ((len
-= copy
) == 0)
1700 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1701 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1704 WARN_ON(start
> offset
+ len
);
1706 end
= start
+ frag
->size
;
1707 if ((copy
= end
- offset
) > 0) {
1713 vaddr
= kmap_skb_frag(frag
);
1714 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1716 kunmap_skb_frag(vaddr
);
1718 if ((len
-= copy
) == 0)
1726 skb_walk_frags(skb
, frag_iter
) {
1729 WARN_ON(start
> offset
+ len
);
1731 end
= start
+ frag_iter
->len
;
1732 if ((copy
= end
- offset
) > 0) {
1735 if (skb_store_bits(frag_iter
, offset
- start
,
1738 if ((len
-= copy
) == 0)
1751 EXPORT_SYMBOL(skb_store_bits
);
1753 /* Checksum skb data. */
1755 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
1756 int len
, __wsum csum
)
1758 int start
= skb_headlen(skb
);
1759 int i
, copy
= start
- offset
;
1760 struct sk_buff
*frag_iter
;
1763 /* Checksum header. */
1767 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1768 if ((len
-= copy
) == 0)
1774 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1777 WARN_ON(start
> offset
+ len
);
1779 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1780 if ((copy
= end
- offset
) > 0) {
1783 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1787 vaddr
= kmap_skb_frag(frag
);
1788 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
1789 offset
- start
, copy
, 0);
1790 kunmap_skb_frag(vaddr
);
1791 csum
= csum_block_add(csum
, csum2
, pos
);
1800 skb_walk_frags(skb
, frag_iter
) {
1803 WARN_ON(start
> offset
+ len
);
1805 end
= start
+ frag_iter
->len
;
1806 if ((copy
= end
- offset
) > 0) {
1810 csum2
= skb_checksum(frag_iter
, offset
- start
,
1812 csum
= csum_block_add(csum
, csum2
, pos
);
1813 if ((len
-= copy
) == 0)
1824 EXPORT_SYMBOL(skb_checksum
);
1826 /* Both of above in one bottle. */
1828 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
1829 u8
*to
, int len
, __wsum csum
)
1831 int start
= skb_headlen(skb
);
1832 int i
, copy
= start
- offset
;
1833 struct sk_buff
*frag_iter
;
1840 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
1842 if ((len
-= copy
) == 0)
1849 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1852 WARN_ON(start
> offset
+ len
);
1854 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1855 if ((copy
= end
- offset
) > 0) {
1858 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1862 vaddr
= kmap_skb_frag(frag
);
1863 csum2
= csum_partial_copy_nocheck(vaddr
+
1867 kunmap_skb_frag(vaddr
);
1868 csum
= csum_block_add(csum
, csum2
, pos
);
1878 skb_walk_frags(skb
, frag_iter
) {
1882 WARN_ON(start
> offset
+ len
);
1884 end
= start
+ frag_iter
->len
;
1885 if ((copy
= end
- offset
) > 0) {
1888 csum2
= skb_copy_and_csum_bits(frag_iter
,
1891 csum
= csum_block_add(csum
, csum2
, pos
);
1892 if ((len
-= copy
) == 0)
1903 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
1905 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
1910 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1911 csstart
= skb_checksum_start_offset(skb
);
1913 csstart
= skb_headlen(skb
);
1915 BUG_ON(csstart
> skb_headlen(skb
));
1917 skb_copy_from_linear_data(skb
, to
, csstart
);
1920 if (csstart
!= skb
->len
)
1921 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
1922 skb
->len
- csstart
, 0);
1924 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1925 long csstuff
= csstart
+ skb
->csum_offset
;
1927 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
1930 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
1933 * skb_dequeue - remove from the head of the queue
1934 * @list: list to dequeue from
1936 * Remove the head of the list. The list lock is taken so the function
1937 * may be used safely with other locking list functions. The head item is
1938 * returned or %NULL if the list is empty.
1941 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
1943 unsigned long flags
;
1944 struct sk_buff
*result
;
1946 spin_lock_irqsave(&list
->lock
, flags
);
1947 result
= __skb_dequeue(list
);
1948 spin_unlock_irqrestore(&list
->lock
, flags
);
1951 EXPORT_SYMBOL(skb_dequeue
);
1954 * skb_dequeue_tail - remove from the tail of the queue
1955 * @list: list to dequeue from
1957 * Remove the tail of the list. The list lock is taken so the function
1958 * may be used safely with other locking list functions. The tail item is
1959 * returned or %NULL if the list is empty.
1961 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
1963 unsigned long flags
;
1964 struct sk_buff
*result
;
1966 spin_lock_irqsave(&list
->lock
, flags
);
1967 result
= __skb_dequeue_tail(list
);
1968 spin_unlock_irqrestore(&list
->lock
, flags
);
1971 EXPORT_SYMBOL(skb_dequeue_tail
);
1974 * skb_queue_purge - empty a list
1975 * @list: list to empty
1977 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1978 * the list and one reference dropped. This function takes the list
1979 * lock and is atomic with respect to other list locking functions.
1981 void skb_queue_purge(struct sk_buff_head
*list
)
1983 struct sk_buff
*skb
;
1984 while ((skb
= skb_dequeue(list
)) != NULL
)
1987 EXPORT_SYMBOL(skb_queue_purge
);
1990 * skb_queue_head - queue a buffer at the list head
1991 * @list: list to use
1992 * @newsk: buffer to queue
1994 * Queue a buffer at the start of the list. This function takes the
1995 * list lock and can be used safely with other locking &sk_buff functions
1998 * A buffer cannot be placed on two lists at the same time.
2000 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2002 unsigned long flags
;
2004 spin_lock_irqsave(&list
->lock
, flags
);
2005 __skb_queue_head(list
, newsk
);
2006 spin_unlock_irqrestore(&list
->lock
, flags
);
2008 EXPORT_SYMBOL(skb_queue_head
);
2011 * skb_queue_tail - queue a buffer at the list tail
2012 * @list: list to use
2013 * @newsk: buffer to queue
2015 * Queue a buffer at the tail of the list. This function takes the
2016 * list lock and can be used safely with other locking &sk_buff functions
2019 * A buffer cannot be placed on two lists at the same time.
2021 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2023 unsigned long flags
;
2025 spin_lock_irqsave(&list
->lock
, flags
);
2026 __skb_queue_tail(list
, newsk
);
2027 spin_unlock_irqrestore(&list
->lock
, flags
);
2029 EXPORT_SYMBOL(skb_queue_tail
);
2032 * skb_unlink - remove a buffer from a list
2033 * @skb: buffer to remove
2034 * @list: list to use
2036 * Remove a packet from a list. The list locks are taken and this
2037 * function is atomic with respect to other list locked calls
2039 * You must know what list the SKB is on.
2041 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
2043 unsigned long flags
;
2045 spin_lock_irqsave(&list
->lock
, flags
);
2046 __skb_unlink(skb
, list
);
2047 spin_unlock_irqrestore(&list
->lock
, flags
);
2049 EXPORT_SYMBOL(skb_unlink
);
2052 * skb_append - append a buffer
2053 * @old: buffer to insert after
2054 * @newsk: buffer to insert
2055 * @list: list to use
2057 * Place a packet after a given packet in a list. The list locks are taken
2058 * and this function is atomic with respect to other list locked calls.
2059 * A buffer cannot be placed on two lists at the same time.
2061 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2063 unsigned long flags
;
2065 spin_lock_irqsave(&list
->lock
, flags
);
2066 __skb_queue_after(list
, old
, newsk
);
2067 spin_unlock_irqrestore(&list
->lock
, flags
);
2069 EXPORT_SYMBOL(skb_append
);
2072 * skb_insert - insert a buffer
2073 * @old: buffer to insert before
2074 * @newsk: buffer to insert
2075 * @list: list to use
2077 * Place a packet before a given packet in a list. The list locks are
2078 * taken and this function is atomic with respect to other list locked
2081 * A buffer cannot be placed on two lists at the same time.
2083 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2085 unsigned long flags
;
2087 spin_lock_irqsave(&list
->lock
, flags
);
2088 __skb_insert(newsk
, old
->prev
, old
, list
);
2089 spin_unlock_irqrestore(&list
->lock
, flags
);
2091 EXPORT_SYMBOL(skb_insert
);
2093 static inline void skb_split_inside_header(struct sk_buff
*skb
,
2094 struct sk_buff
* skb1
,
2095 const u32 len
, const int pos
)
2099 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
2101 /* And move data appendix as is. */
2102 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
2103 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
2105 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
2106 skb_shinfo(skb
)->nr_frags
= 0;
2107 skb1
->data_len
= skb
->data_len
;
2108 skb1
->len
+= skb1
->data_len
;
2111 skb_set_tail_pointer(skb
, len
);
2114 static inline void skb_split_no_header(struct sk_buff
*skb
,
2115 struct sk_buff
* skb1
,
2116 const u32 len
, int pos
)
2119 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
2121 skb_shinfo(skb
)->nr_frags
= 0;
2122 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
2124 skb
->data_len
= len
- pos
;
2126 for (i
= 0; i
< nfrags
; i
++) {
2127 int size
= skb_shinfo(skb
)->frags
[i
].size
;
2129 if (pos
+ size
> len
) {
2130 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
2134 * We have two variants in this case:
2135 * 1. Move all the frag to the second
2136 * part, if it is possible. F.e.
2137 * this approach is mandatory for TUX,
2138 * where splitting is expensive.
2139 * 2. Split is accurately. We make this.
2141 get_page(skb_shinfo(skb
)->frags
[i
].page
);
2142 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
2143 skb_shinfo(skb1
)->frags
[0].size
-= len
- pos
;
2144 skb_shinfo(skb
)->frags
[i
].size
= len
- pos
;
2145 skb_shinfo(skb
)->nr_frags
++;
2149 skb_shinfo(skb
)->nr_frags
++;
2152 skb_shinfo(skb1
)->nr_frags
= k
;
2156 * skb_split - Split fragmented skb to two parts at length len.
2157 * @skb: the buffer to split
2158 * @skb1: the buffer to receive the second part
2159 * @len: new length for skb
2161 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
2163 int pos
= skb_headlen(skb
);
2165 if (len
< pos
) /* Split line is inside header. */
2166 skb_split_inside_header(skb
, skb1
, len
, pos
);
2167 else /* Second chunk has no header, nothing to copy. */
2168 skb_split_no_header(skb
, skb1
, len
, pos
);
2170 EXPORT_SYMBOL(skb_split
);
2172 /* Shifting from/to a cloned skb is a no-go.
2174 * Caller cannot keep skb_shinfo related pointers past calling here!
2176 static int skb_prepare_for_shift(struct sk_buff
*skb
)
2178 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2182 * skb_shift - Shifts paged data partially from skb to another
2183 * @tgt: buffer into which tail data gets added
2184 * @skb: buffer from which the paged data comes from
2185 * @shiftlen: shift up to this many bytes
2187 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2188 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2189 * It's up to caller to free skb if everything was shifted.
2191 * If @tgt runs out of frags, the whole operation is aborted.
2193 * Skb cannot include anything else but paged data while tgt is allowed
2194 * to have non-paged data as well.
2196 * TODO: full sized shift could be optimized but that would need
2197 * specialized skb free'er to handle frags without up-to-date nr_frags.
2199 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
2201 int from
, to
, merge
, todo
;
2202 struct skb_frag_struct
*fragfrom
, *fragto
;
2204 BUG_ON(shiftlen
> skb
->len
);
2205 BUG_ON(skb_headlen(skb
)); /* Would corrupt stream */
2209 to
= skb_shinfo(tgt
)->nr_frags
;
2210 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2212 /* Actual merge is delayed until the point when we know we can
2213 * commit all, so that we don't have to undo partial changes
2216 !skb_can_coalesce(tgt
, to
, fragfrom
->page
, fragfrom
->page_offset
)) {
2221 todo
-= fragfrom
->size
;
2223 if (skb_prepare_for_shift(skb
) ||
2224 skb_prepare_for_shift(tgt
))
2227 /* All previous frag pointers might be stale! */
2228 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2229 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2231 fragto
->size
+= shiftlen
;
2232 fragfrom
->size
-= shiftlen
;
2233 fragfrom
->page_offset
+= shiftlen
;
2241 /* Skip full, not-fitting skb to avoid expensive operations */
2242 if ((shiftlen
== skb
->len
) &&
2243 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
2246 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
2249 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
2250 if (to
== MAX_SKB_FRAGS
)
2253 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2254 fragto
= &skb_shinfo(tgt
)->frags
[to
];
2256 if (todo
>= fragfrom
->size
) {
2257 *fragto
= *fragfrom
;
2258 todo
-= fragfrom
->size
;
2263 get_page(fragfrom
->page
);
2264 fragto
->page
= fragfrom
->page
;
2265 fragto
->page_offset
= fragfrom
->page_offset
;
2266 fragto
->size
= todo
;
2268 fragfrom
->page_offset
+= todo
;
2269 fragfrom
->size
-= todo
;
2277 /* Ready to "commit" this state change to tgt */
2278 skb_shinfo(tgt
)->nr_frags
= to
;
2281 fragfrom
= &skb_shinfo(skb
)->frags
[0];
2282 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2284 fragto
->size
+= fragfrom
->size
;
2285 put_page(fragfrom
->page
);
2288 /* Reposition in the original skb */
2290 while (from
< skb_shinfo(skb
)->nr_frags
)
2291 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
2292 skb_shinfo(skb
)->nr_frags
= to
;
2294 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
2297 /* Most likely the tgt won't ever need its checksum anymore, skb on
2298 * the other hand might need it if it needs to be resent
2300 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
2301 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2303 /* Yak, is it really working this way? Some helper please? */
2304 skb
->len
-= shiftlen
;
2305 skb
->data_len
-= shiftlen
;
2306 skb
->truesize
-= shiftlen
;
2307 tgt
->len
+= shiftlen
;
2308 tgt
->data_len
+= shiftlen
;
2309 tgt
->truesize
+= shiftlen
;
2315 * skb_prepare_seq_read - Prepare a sequential read of skb data
2316 * @skb: the buffer to read
2317 * @from: lower offset of data to be read
2318 * @to: upper offset of data to be read
2319 * @st: state variable
2321 * Initializes the specified state variable. Must be called before
2322 * invoking skb_seq_read() for the first time.
2324 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
2325 unsigned int to
, struct skb_seq_state
*st
)
2327 st
->lower_offset
= from
;
2328 st
->upper_offset
= to
;
2329 st
->root_skb
= st
->cur_skb
= skb
;
2330 st
->frag_idx
= st
->stepped_offset
= 0;
2331 st
->frag_data
= NULL
;
2333 EXPORT_SYMBOL(skb_prepare_seq_read
);
2336 * skb_seq_read - Sequentially read skb data
2337 * @consumed: number of bytes consumed by the caller so far
2338 * @data: destination pointer for data to be returned
2339 * @st: state variable
2341 * Reads a block of skb data at &consumed relative to the
2342 * lower offset specified to skb_prepare_seq_read(). Assigns
2343 * the head of the data block to &data and returns the length
2344 * of the block or 0 if the end of the skb data or the upper
2345 * offset has been reached.
2347 * The caller is not required to consume all of the data
2348 * returned, i.e. &consumed is typically set to the number
2349 * of bytes already consumed and the next call to
2350 * skb_seq_read() will return the remaining part of the block.
2352 * Note 1: The size of each block of data returned can be arbitrary,
2353 * this limitation is the cost for zerocopy seqeuental
2354 * reads of potentially non linear data.
2356 * Note 2: Fragment lists within fragments are not implemented
2357 * at the moment, state->root_skb could be replaced with
2358 * a stack for this purpose.
2360 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
2361 struct skb_seq_state
*st
)
2363 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
2366 if (unlikely(abs_offset
>= st
->upper_offset
))
2370 block_limit
= skb_headlen(st
->cur_skb
) + st
->stepped_offset
;
2372 if (abs_offset
< block_limit
&& !st
->frag_data
) {
2373 *data
= st
->cur_skb
->data
+ (abs_offset
- st
->stepped_offset
);
2374 return block_limit
- abs_offset
;
2377 if (st
->frag_idx
== 0 && !st
->frag_data
)
2378 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
2380 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
2381 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
2382 block_limit
= frag
->size
+ st
->stepped_offset
;
2384 if (abs_offset
< block_limit
) {
2386 st
->frag_data
= kmap_skb_frag(frag
);
2388 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
2389 (abs_offset
- st
->stepped_offset
);
2391 return block_limit
- abs_offset
;
2394 if (st
->frag_data
) {
2395 kunmap_skb_frag(st
->frag_data
);
2396 st
->frag_data
= NULL
;
2400 st
->stepped_offset
+= frag
->size
;
2403 if (st
->frag_data
) {
2404 kunmap_skb_frag(st
->frag_data
);
2405 st
->frag_data
= NULL
;
2408 if (st
->root_skb
== st
->cur_skb
&& skb_has_frag_list(st
->root_skb
)) {
2409 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
2412 } else if (st
->cur_skb
->next
) {
2413 st
->cur_skb
= st
->cur_skb
->next
;
2420 EXPORT_SYMBOL(skb_seq_read
);
2423 * skb_abort_seq_read - Abort a sequential read of skb data
2424 * @st: state variable
2426 * Must be called if skb_seq_read() was not called until it
2429 void skb_abort_seq_read(struct skb_seq_state
*st
)
2432 kunmap_skb_frag(st
->frag_data
);
2434 EXPORT_SYMBOL(skb_abort_seq_read
);
2436 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2438 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
2439 struct ts_config
*conf
,
2440 struct ts_state
*state
)
2442 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
2445 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
2447 skb_abort_seq_read(TS_SKB_CB(state
));
2451 * skb_find_text - Find a text pattern in skb data
2452 * @skb: the buffer to look in
2453 * @from: search offset
2455 * @config: textsearch configuration
2456 * @state: uninitialized textsearch state variable
2458 * Finds a pattern in the skb data according to the specified
2459 * textsearch configuration. Use textsearch_next() to retrieve
2460 * subsequent occurrences of the pattern. Returns the offset
2461 * to the first occurrence or UINT_MAX if no match was found.
2463 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
2464 unsigned int to
, struct ts_config
*config
,
2465 struct ts_state
*state
)
2469 config
->get_next_block
= skb_ts_get_next_block
;
2470 config
->finish
= skb_ts_finish
;
2472 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
2474 ret
= textsearch_find(config
, state
);
2475 return (ret
<= to
- from
? ret
: UINT_MAX
);
2477 EXPORT_SYMBOL(skb_find_text
);
2480 * skb_append_datato_frags: - append the user data to a skb
2481 * @sk: sock structure
2482 * @skb: skb structure to be appened with user data.
2483 * @getfrag: call back function to be used for getting the user data
2484 * @from: pointer to user message iov
2485 * @length: length of the iov message
2487 * Description: This procedure append the user data in the fragment part
2488 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2490 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
2491 int (*getfrag
)(void *from
, char *to
, int offset
,
2492 int len
, int odd
, struct sk_buff
*skb
),
2493 void *from
, int length
)
2496 skb_frag_t
*frag
= NULL
;
2497 struct page
*page
= NULL
;
2503 /* Return error if we don't have space for new frag */
2504 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2505 if (frg_cnt
>= MAX_SKB_FRAGS
)
2508 /* allocate a new page for next frag */
2509 page
= alloc_pages(sk
->sk_allocation
, 0);
2511 /* If alloc_page fails just return failure and caller will
2512 * free previous allocated pages by doing kfree_skb()
2517 /* initialize the next frag */
2518 skb_fill_page_desc(skb
, frg_cnt
, page
, 0, 0);
2519 skb
->truesize
+= PAGE_SIZE
;
2520 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
2522 /* get the new initialized frag */
2523 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2524 frag
= &skb_shinfo(skb
)->frags
[frg_cnt
- 1];
2526 /* copy the user data to page */
2527 left
= PAGE_SIZE
- frag
->page_offset
;
2528 copy
= (length
> left
)? left
: length
;
2530 ret
= getfrag(from
, (page_address(frag
->page
) +
2531 frag
->page_offset
+ frag
->size
),
2532 offset
, copy
, 0, skb
);
2536 /* copy was successful so update the size parameters */
2539 skb
->data_len
+= copy
;
2543 } while (length
> 0);
2547 EXPORT_SYMBOL(skb_append_datato_frags
);
2550 * skb_pull_rcsum - pull skb and update receive checksum
2551 * @skb: buffer to update
2552 * @len: length of data pulled
2554 * This function performs an skb_pull on the packet and updates
2555 * the CHECKSUM_COMPLETE checksum. It should be used on
2556 * receive path processing instead of skb_pull unless you know
2557 * that the checksum difference is zero (e.g., a valid IP header)
2558 * or you are setting ip_summed to CHECKSUM_NONE.
2560 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
2562 BUG_ON(len
> skb
->len
);
2564 BUG_ON(skb
->len
< skb
->data_len
);
2565 skb_postpull_rcsum(skb
, skb
->data
, len
);
2566 return skb
->data
+= len
;
2568 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
2571 * skb_segment - Perform protocol segmentation on skb.
2572 * @skb: buffer to segment
2573 * @features: features for the output path (see dev->features)
2575 * This function performs segmentation on the given skb. It returns
2576 * a pointer to the first in a list of new skbs for the segments.
2577 * In case of error it returns ERR_PTR(err).
2579 struct sk_buff
*skb_segment(struct sk_buff
*skb
, u32 features
)
2581 struct sk_buff
*segs
= NULL
;
2582 struct sk_buff
*tail
= NULL
;
2583 struct sk_buff
*fskb
= skb_shinfo(skb
)->frag_list
;
2584 unsigned int mss
= skb_shinfo(skb
)->gso_size
;
2585 unsigned int doffset
= skb
->data
- skb_mac_header(skb
);
2586 unsigned int offset
= doffset
;
2587 unsigned int headroom
;
2589 int sg
= !!(features
& NETIF_F_SG
);
2590 int nfrags
= skb_shinfo(skb
)->nr_frags
;
2595 __skb_push(skb
, doffset
);
2596 headroom
= skb_headroom(skb
);
2597 pos
= skb_headlen(skb
);
2600 struct sk_buff
*nskb
;
2605 len
= skb
->len
- offset
;
2609 hsize
= skb_headlen(skb
) - offset
;
2612 if (hsize
> len
|| !sg
)
2615 if (!hsize
&& i
>= nfrags
) {
2616 BUG_ON(fskb
->len
!= len
);
2619 nskb
= skb_clone(fskb
, GFP_ATOMIC
);
2622 if (unlikely(!nskb
))
2625 hsize
= skb_end_pointer(nskb
) - nskb
->head
;
2626 if (skb_cow_head(nskb
, doffset
+ headroom
)) {
2631 nskb
->truesize
+= skb_end_pointer(nskb
) - nskb
->head
-
2633 skb_release_head_state(nskb
);
2634 __skb_push(nskb
, doffset
);
2636 nskb
= alloc_skb(hsize
+ doffset
+ headroom
,
2639 if (unlikely(!nskb
))
2642 skb_reserve(nskb
, headroom
);
2643 __skb_put(nskb
, doffset
);
2652 __copy_skb_header(nskb
, skb
);
2653 nskb
->mac_len
= skb
->mac_len
;
2655 /* nskb and skb might have different headroom */
2656 if (nskb
->ip_summed
== CHECKSUM_PARTIAL
)
2657 nskb
->csum_start
+= skb_headroom(nskb
) - headroom
;
2659 skb_reset_mac_header(nskb
);
2660 skb_set_network_header(nskb
, skb
->mac_len
);
2661 nskb
->transport_header
= (nskb
->network_header
+
2662 skb_network_header_len(skb
));
2663 skb_copy_from_linear_data(skb
, nskb
->data
, doffset
);
2665 if (fskb
!= skb_shinfo(skb
)->frag_list
)
2669 nskb
->ip_summed
= CHECKSUM_NONE
;
2670 nskb
->csum
= skb_copy_and_csum_bits(skb
, offset
,
2676 frag
= skb_shinfo(nskb
)->frags
;
2678 skb_copy_from_linear_data_offset(skb
, offset
,
2679 skb_put(nskb
, hsize
), hsize
);
2681 while (pos
< offset
+ len
&& i
< nfrags
) {
2682 *frag
= skb_shinfo(skb
)->frags
[i
];
2683 get_page(frag
->page
);
2687 frag
->page_offset
+= offset
- pos
;
2688 frag
->size
-= offset
- pos
;
2691 skb_shinfo(nskb
)->nr_frags
++;
2693 if (pos
+ size
<= offset
+ len
) {
2697 frag
->size
-= pos
+ size
- (offset
+ len
);
2704 if (pos
< offset
+ len
) {
2705 struct sk_buff
*fskb2
= fskb
;
2707 BUG_ON(pos
+ fskb
->len
!= offset
+ len
);
2713 fskb2
= skb_clone(fskb2
, GFP_ATOMIC
);
2719 SKB_FRAG_ASSERT(nskb
);
2720 skb_shinfo(nskb
)->frag_list
= fskb2
;
2724 nskb
->data_len
= len
- hsize
;
2725 nskb
->len
+= nskb
->data_len
;
2726 nskb
->truesize
+= nskb
->data_len
;
2727 } while ((offset
+= len
) < skb
->len
);
2732 while ((skb
= segs
)) {
2736 return ERR_PTR(err
);
2738 EXPORT_SYMBOL_GPL(skb_segment
);
2740 int skb_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2742 struct sk_buff
*p
= *head
;
2743 struct sk_buff
*nskb
;
2744 struct skb_shared_info
*skbinfo
= skb_shinfo(skb
);
2745 struct skb_shared_info
*pinfo
= skb_shinfo(p
);
2746 unsigned int headroom
;
2747 unsigned int len
= skb_gro_len(skb
);
2748 unsigned int offset
= skb_gro_offset(skb
);
2749 unsigned int headlen
= skb_headlen(skb
);
2751 if (p
->len
+ len
>= 65536)
2754 if (pinfo
->frag_list
)
2756 else if (headlen
<= offset
) {
2759 int i
= skbinfo
->nr_frags
;
2760 int nr_frags
= pinfo
->nr_frags
+ i
;
2764 if (nr_frags
> MAX_SKB_FRAGS
)
2767 pinfo
->nr_frags
= nr_frags
;
2768 skbinfo
->nr_frags
= 0;
2770 frag
= pinfo
->frags
+ nr_frags
;
2771 frag2
= skbinfo
->frags
+ i
;
2776 frag
->page_offset
+= offset
;
2777 frag
->size
-= offset
;
2779 skb
->truesize
-= skb
->data_len
;
2780 skb
->len
-= skb
->data_len
;
2783 NAPI_GRO_CB(skb
)->free
= 1;
2785 } else if (skb_gro_len(p
) != pinfo
->gso_size
)
2788 headroom
= skb_headroom(p
);
2789 nskb
= alloc_skb(headroom
+ skb_gro_offset(p
), GFP_ATOMIC
);
2790 if (unlikely(!nskb
))
2793 __copy_skb_header(nskb
, p
);
2794 nskb
->mac_len
= p
->mac_len
;
2796 skb_reserve(nskb
, headroom
);
2797 __skb_put(nskb
, skb_gro_offset(p
));
2799 skb_set_mac_header(nskb
, skb_mac_header(p
) - p
->data
);
2800 skb_set_network_header(nskb
, skb_network_offset(p
));
2801 skb_set_transport_header(nskb
, skb_transport_offset(p
));
2803 __skb_pull(p
, skb_gro_offset(p
));
2804 memcpy(skb_mac_header(nskb
), skb_mac_header(p
),
2805 p
->data
- skb_mac_header(p
));
2807 *NAPI_GRO_CB(nskb
) = *NAPI_GRO_CB(p
);
2808 skb_shinfo(nskb
)->frag_list
= p
;
2809 skb_shinfo(nskb
)->gso_size
= pinfo
->gso_size
;
2810 pinfo
->gso_size
= 0;
2811 skb_header_release(p
);
2814 nskb
->data_len
+= p
->len
;
2815 nskb
->truesize
+= p
->len
;
2816 nskb
->len
+= p
->len
;
2819 nskb
->next
= p
->next
;
2825 if (offset
> headlen
) {
2826 unsigned int eat
= offset
- headlen
;
2828 skbinfo
->frags
[0].page_offset
+= eat
;
2829 skbinfo
->frags
[0].size
-= eat
;
2830 skb
->data_len
-= eat
;
2835 __skb_pull(skb
, offset
);
2837 p
->prev
->next
= skb
;
2839 skb_header_release(skb
);
2842 NAPI_GRO_CB(p
)->count
++;
2847 NAPI_GRO_CB(skb
)->same_flow
= 1;
2850 EXPORT_SYMBOL_GPL(skb_gro_receive
);
2852 void __init
skb_init(void)
2854 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
2855 sizeof(struct sk_buff
),
2857 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2859 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
2860 (2*sizeof(struct sk_buff
)) +
2863 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2868 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2869 * @skb: Socket buffer containing the buffers to be mapped
2870 * @sg: The scatter-gather list to map into
2871 * @offset: The offset into the buffer's contents to start mapping
2872 * @len: Length of buffer space to be mapped
2874 * Fill the specified scatter-gather list with mappings/pointers into a
2875 * region of the buffer space attached to a socket buffer.
2878 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2880 int start
= skb_headlen(skb
);
2881 int i
, copy
= start
- offset
;
2882 struct sk_buff
*frag_iter
;
2888 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
2890 if ((len
-= copy
) == 0)
2895 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2898 WARN_ON(start
> offset
+ len
);
2900 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
2901 if ((copy
= end
- offset
) > 0) {
2902 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2906 sg_set_page(&sg
[elt
], frag
->page
, copy
,
2907 frag
->page_offset
+offset
-start
);
2916 skb_walk_frags(skb
, frag_iter
) {
2919 WARN_ON(start
> offset
+ len
);
2921 end
= start
+ frag_iter
->len
;
2922 if ((copy
= end
- offset
) > 0) {
2925 elt
+= __skb_to_sgvec(frag_iter
, sg
+elt
, offset
- start
,
2927 if ((len
-= copy
) == 0)
2937 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2939 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
);
2941 sg_mark_end(&sg
[nsg
- 1]);
2945 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
2948 * skb_cow_data - Check that a socket buffer's data buffers are writable
2949 * @skb: The socket buffer to check.
2950 * @tailbits: Amount of trailing space to be added
2951 * @trailer: Returned pointer to the skb where the @tailbits space begins
2953 * Make sure that the data buffers attached to a socket buffer are
2954 * writable. If they are not, private copies are made of the data buffers
2955 * and the socket buffer is set to use these instead.
2957 * If @tailbits is given, make sure that there is space to write @tailbits
2958 * bytes of data beyond current end of socket buffer. @trailer will be
2959 * set to point to the skb in which this space begins.
2961 * The number of scatterlist elements required to completely map the
2962 * COW'd and extended socket buffer will be returned.
2964 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
2968 struct sk_buff
*skb1
, **skb_p
;
2970 /* If skb is cloned or its head is paged, reallocate
2971 * head pulling out all the pages (pages are considered not writable
2972 * at the moment even if they are anonymous).
2974 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
2975 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
2978 /* Easy case. Most of packets will go this way. */
2979 if (!skb_has_frag_list(skb
)) {
2980 /* A little of trouble, not enough of space for trailer.
2981 * This should not happen, when stack is tuned to generate
2982 * good frames. OK, on miss we reallocate and reserve even more
2983 * space, 128 bytes is fair. */
2985 if (skb_tailroom(skb
) < tailbits
&&
2986 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
2994 /* Misery. We are in troubles, going to mincer fragments... */
2997 skb_p
= &skb_shinfo(skb
)->frag_list
;
3000 while ((skb1
= *skb_p
) != NULL
) {
3003 /* The fragment is partially pulled by someone,
3004 * this can happen on input. Copy it and everything
3007 if (skb_shared(skb1
))
3010 /* If the skb is the last, worry about trailer. */
3012 if (skb1
->next
== NULL
&& tailbits
) {
3013 if (skb_shinfo(skb1
)->nr_frags
||
3014 skb_has_frag_list(skb1
) ||
3015 skb_tailroom(skb1
) < tailbits
)
3016 ntail
= tailbits
+ 128;
3022 skb_shinfo(skb1
)->nr_frags
||
3023 skb_has_frag_list(skb1
)) {
3024 struct sk_buff
*skb2
;
3026 /* Fuck, we are miserable poor guys... */
3028 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
3030 skb2
= skb_copy_expand(skb1
,
3034 if (unlikely(skb2
== NULL
))
3038 skb_set_owner_w(skb2
, skb1
->sk
);
3040 /* Looking around. Are we still alive?
3041 * OK, link new skb, drop old one */
3043 skb2
->next
= skb1
->next
;
3050 skb_p
= &skb1
->next
;
3055 EXPORT_SYMBOL_GPL(skb_cow_data
);
3057 static void sock_rmem_free(struct sk_buff
*skb
)
3059 struct sock
*sk
= skb
->sk
;
3061 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
3065 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3067 int sock_queue_err_skb(struct sock
*sk
, struct sk_buff
*skb
)
3069 if (atomic_read(&sk
->sk_rmem_alloc
) + skb
->truesize
>=
3070 (unsigned)sk
->sk_rcvbuf
)
3075 skb
->destructor
= sock_rmem_free
;
3076 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
3078 /* before exiting rcu section, make sure dst is refcounted */
3081 skb_queue_tail(&sk
->sk_error_queue
, skb
);
3082 if (!sock_flag(sk
, SOCK_DEAD
))
3083 sk
->sk_data_ready(sk
, skb
->len
);
3086 EXPORT_SYMBOL(sock_queue_err_skb
);
3088 void skb_tstamp_tx(struct sk_buff
*orig_skb
,
3089 struct skb_shared_hwtstamps
*hwtstamps
)
3091 struct sock
*sk
= orig_skb
->sk
;
3092 struct sock_exterr_skb
*serr
;
3093 struct sk_buff
*skb
;
3099 skb
= skb_clone(orig_skb
, GFP_ATOMIC
);
3104 *skb_hwtstamps(skb
) =
3108 * no hardware time stamps available,
3109 * so keep the shared tx_flags and only
3110 * store software time stamp
3112 skb
->tstamp
= ktime_get_real();
3115 serr
= SKB_EXT_ERR(skb
);
3116 memset(serr
, 0, sizeof(*serr
));
3117 serr
->ee
.ee_errno
= ENOMSG
;
3118 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TIMESTAMPING
;
3120 err
= sock_queue_err_skb(sk
, skb
);
3125 EXPORT_SYMBOL_GPL(skb_tstamp_tx
);
3129 * skb_partial_csum_set - set up and verify partial csum values for packet
3130 * @skb: the skb to set
3131 * @start: the number of bytes after skb->data to start checksumming.
3132 * @off: the offset from start to place the checksum.
3134 * For untrusted partially-checksummed packets, we need to make sure the values
3135 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3137 * This function checks and sets those values and skb->ip_summed: if this
3138 * returns false you should drop the packet.
3140 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
3142 if (unlikely(start
> skb_headlen(skb
)) ||
3143 unlikely((int)start
+ off
> skb_headlen(skb
) - 2)) {
3144 if (net_ratelimit())
3146 "bad partial csum: csum=%u/%u len=%u\n",
3147 start
, off
, skb_headlen(skb
));
3150 skb
->ip_summed
= CHECKSUM_PARTIAL
;
3151 skb
->csum_start
= skb_headroom(skb
) + start
;
3152 skb
->csum_offset
= off
;
3155 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);
3157 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
3159 if (net_ratelimit())
3160 pr_warning("%s: received packets cannot be forwarded"
3161 " while LRO is enabled\n", skb
->dev
->name
);
3163 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);