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
;
532 new->l4_rxhash
= old
->l4_rxhash
;
534 new->sp
= secpath_get(old
->sp
);
536 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
537 new->csum
= old
->csum
;
538 new->local_df
= old
->local_df
;
539 new->pkt_type
= old
->pkt_type
;
540 new->ip_summed
= old
->ip_summed
;
541 skb_copy_queue_mapping(new, old
);
542 new->priority
= old
->priority
;
543 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
544 new->ipvs_property
= old
->ipvs_property
;
546 new->protocol
= old
->protocol
;
547 new->mark
= old
->mark
;
548 new->skb_iif
= old
->skb_iif
;
550 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
551 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
552 new->nf_trace
= old
->nf_trace
;
554 #ifdef CONFIG_NET_SCHED
555 new->tc_index
= old
->tc_index
;
556 #ifdef CONFIG_NET_CLS_ACT
557 new->tc_verd
= old
->tc_verd
;
560 new->vlan_tci
= old
->vlan_tci
;
562 skb_copy_secmark(new, old
);
566 * You should not add any new code to this function. Add it to
567 * __copy_skb_header above instead.
569 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
571 #define C(x) n->x = skb->x
573 n
->next
= n
->prev
= NULL
;
575 __copy_skb_header(n
, skb
);
580 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
583 n
->destructor
= NULL
;
589 atomic_set(&n
->users
, 1);
591 atomic_inc(&(skb_shinfo(skb
)->dataref
));
599 * skb_morph - morph one skb into another
600 * @dst: the skb to receive the contents
601 * @src: the skb to supply the contents
603 * This is identical to skb_clone except that the target skb is
604 * supplied by the user.
606 * The target skb is returned upon exit.
608 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
610 skb_release_all(dst
);
611 return __skb_clone(dst
, src
);
613 EXPORT_SYMBOL_GPL(skb_morph
);
615 /* skb frags copy userspace buffers to kernel */
616 static int skb_copy_ubufs(struct sk_buff
*skb
, gfp_t gfp_mask
)
619 int num_frags
= skb_shinfo(skb
)->nr_frags
;
620 struct page
*page
, *head
= NULL
;
621 struct ubuf_info
*uarg
= skb_shinfo(skb
)->destructor_arg
;
623 for (i
= 0; i
< num_frags
; i
++) {
625 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
627 page
= alloc_page(GFP_ATOMIC
);
630 struct page
*next
= (struct page
*)head
->private;
636 vaddr
= kmap_skb_frag(&skb_shinfo(skb
)->frags
[i
]);
637 memcpy(page_address(page
),
638 vaddr
+ f
->page_offset
, f
->size
);
639 kunmap_skb_frag(vaddr
);
640 page
->private = (unsigned long)head
;
644 /* skb frags release userspace buffers */
645 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
646 put_page(skb_shinfo(skb
)->frags
[i
].page
);
648 uarg
->callback(uarg
);
650 /* skb frags point to kernel buffers */
651 for (i
= skb_shinfo(skb
)->nr_frags
; i
> 0; i
--) {
652 skb_shinfo(skb
)->frags
[i
- 1].page_offset
= 0;
653 skb_shinfo(skb
)->frags
[i
- 1].page
= head
;
654 head
= (struct page
*)head
->private;
661 * skb_clone - duplicate an sk_buff
662 * @skb: buffer to clone
663 * @gfp_mask: allocation priority
665 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
666 * copies share the same packet data but not structure. The new
667 * buffer has a reference count of 1. If the allocation fails the
668 * function returns %NULL otherwise the new buffer is returned.
670 * If this function is called from an interrupt gfp_mask() must be
674 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
678 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
679 if (skb_copy_ubufs(skb
, gfp_mask
))
681 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
685 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
686 n
->fclone
== SKB_FCLONE_UNAVAILABLE
) {
687 atomic_t
*fclone_ref
= (atomic_t
*) (n
+ 1);
688 n
->fclone
= SKB_FCLONE_CLONE
;
689 atomic_inc(fclone_ref
);
691 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
695 kmemcheck_annotate_bitfield(n
, flags1
);
696 kmemcheck_annotate_bitfield(n
, flags2
);
697 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
700 return __skb_clone(n
, skb
);
702 EXPORT_SYMBOL(skb_clone
);
704 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
706 #ifndef NET_SKBUFF_DATA_USES_OFFSET
708 * Shift between the two data areas in bytes
710 unsigned long offset
= new->data
- old
->data
;
713 __copy_skb_header(new, old
);
715 #ifndef NET_SKBUFF_DATA_USES_OFFSET
716 /* {transport,network,mac}_header are relative to skb->head */
717 new->transport_header
+= offset
;
718 new->network_header
+= offset
;
719 if (skb_mac_header_was_set(new))
720 new->mac_header
+= offset
;
722 skb_shinfo(new)->gso_size
= skb_shinfo(old
)->gso_size
;
723 skb_shinfo(new)->gso_segs
= skb_shinfo(old
)->gso_segs
;
724 skb_shinfo(new)->gso_type
= skb_shinfo(old
)->gso_type
;
728 * skb_copy - create private copy of an sk_buff
729 * @skb: buffer to copy
730 * @gfp_mask: allocation priority
732 * Make a copy of both an &sk_buff and its data. This is used when the
733 * caller wishes to modify the data and needs a private copy of the
734 * data to alter. Returns %NULL on failure or the pointer to the buffer
735 * on success. The returned buffer has a reference count of 1.
737 * As by-product this function converts non-linear &sk_buff to linear
738 * one, so that &sk_buff becomes completely private and caller is allowed
739 * to modify all the data of returned buffer. This means that this
740 * function is not recommended for use in circumstances when only
741 * header is going to be modified. Use pskb_copy() instead.
744 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
746 int headerlen
= skb_headroom(skb
);
747 unsigned int size
= (skb_end_pointer(skb
) - skb
->head
) + skb
->data_len
;
748 struct sk_buff
*n
= alloc_skb(size
, gfp_mask
);
753 /* Set the data pointer */
754 skb_reserve(n
, headerlen
);
755 /* Set the tail pointer and length */
756 skb_put(n
, skb
->len
);
758 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
761 copy_skb_header(n
, skb
);
764 EXPORT_SYMBOL(skb_copy
);
767 * pskb_copy - create copy of an sk_buff with private head.
768 * @skb: buffer to copy
769 * @gfp_mask: allocation priority
771 * Make a copy of both an &sk_buff and part of its data, located
772 * in header. Fragmented data remain shared. This is used when
773 * the caller wishes to modify only header of &sk_buff and needs
774 * private copy of the header to alter. Returns %NULL on failure
775 * or the pointer to the buffer on success.
776 * The returned buffer has a reference count of 1.
779 struct sk_buff
*pskb_copy(struct sk_buff
*skb
, gfp_t gfp_mask
)
781 unsigned int size
= skb_end_pointer(skb
) - skb
->head
;
782 struct sk_buff
*n
= alloc_skb(size
, gfp_mask
);
787 /* Set the data pointer */
788 skb_reserve(n
, skb_headroom(skb
));
789 /* Set the tail pointer and length */
790 skb_put(n
, skb_headlen(skb
));
792 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
794 n
->truesize
+= skb
->data_len
;
795 n
->data_len
= skb
->data_len
;
798 if (skb_shinfo(skb
)->nr_frags
) {
801 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
802 if (skb_copy_ubufs(skb
, gfp_mask
)) {
807 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
809 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
810 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
811 get_page(skb_shinfo(n
)->frags
[i
].page
);
813 skb_shinfo(n
)->nr_frags
= i
;
816 if (skb_has_frag_list(skb
)) {
817 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
818 skb_clone_fraglist(n
);
821 copy_skb_header(n
, skb
);
825 EXPORT_SYMBOL(pskb_copy
);
828 * pskb_expand_head - reallocate header of &sk_buff
829 * @skb: buffer to reallocate
830 * @nhead: room to add at head
831 * @ntail: room to add at tail
832 * @gfp_mask: allocation priority
834 * Expands (or creates identical copy, if &nhead and &ntail are zero)
835 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
836 * reference count of 1. Returns zero in the case of success or error,
837 * if expansion failed. In the last case, &sk_buff is not changed.
839 * All the pointers pointing into skb header may change and must be
840 * reloaded after call to this function.
843 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
848 int size
= nhead
+ (skb_end_pointer(skb
) - skb
->head
) + ntail
;
857 size
= SKB_DATA_ALIGN(size
);
859 /* Check if we can avoid taking references on fragments if we own
860 * the last reference on skb->head. (see skb_release_data())
865 int delta
= skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1;
866 fastpath
= atomic_read(&skb_shinfo(skb
)->dataref
) == delta
;
870 size
+ sizeof(struct skb_shared_info
) <= ksize(skb
->head
)) {
871 memmove(skb
->head
+ size
, skb_shinfo(skb
),
872 offsetof(struct skb_shared_info
,
873 frags
[skb_shinfo(skb
)->nr_frags
]));
874 memmove(skb
->head
+ nhead
, skb
->head
,
875 skb_tail_pointer(skb
) - skb
->head
);
880 data
= kmalloc(size
+ sizeof(struct skb_shared_info
), gfp_mask
);
884 /* Copy only real data... and, alas, header. This should be
885 * optimized for the cases when header is void.
887 memcpy(data
+ nhead
, skb
->head
, skb_tail_pointer(skb
) - skb
->head
);
889 memcpy((struct skb_shared_info
*)(data
+ size
),
891 offsetof(struct skb_shared_info
, frags
[skb_shinfo(skb
)->nr_frags
]));
896 /* copy this zero copy skb frags */
897 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
898 if (skb_copy_ubufs(skb
, gfp_mask
))
900 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
902 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
903 get_page(skb_shinfo(skb
)->frags
[i
].page
);
905 if (skb_has_frag_list(skb
))
906 skb_clone_fraglist(skb
);
908 skb_release_data(skb
);
910 off
= (data
+ nhead
) - skb
->head
;
915 #ifdef NET_SKBUFF_DATA_USES_OFFSET
919 skb
->end
= skb
->head
+ size
;
921 /* {transport,network,mac}_header and tail are relative to skb->head */
923 skb
->transport_header
+= off
;
924 skb
->network_header
+= off
;
925 if (skb_mac_header_was_set(skb
))
926 skb
->mac_header
+= off
;
927 /* Only adjust this if it actually is csum_start rather than csum */
928 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
929 skb
->csum_start
+= nhead
;
933 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
941 EXPORT_SYMBOL(pskb_expand_head
);
943 /* Make private copy of skb with writable head and some headroom */
945 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
947 struct sk_buff
*skb2
;
948 int delta
= headroom
- skb_headroom(skb
);
951 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
953 skb2
= skb_clone(skb
, GFP_ATOMIC
);
954 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
962 EXPORT_SYMBOL(skb_realloc_headroom
);
965 * skb_copy_expand - copy and expand sk_buff
966 * @skb: buffer to copy
967 * @newheadroom: new free bytes at head
968 * @newtailroom: new free bytes at tail
969 * @gfp_mask: allocation priority
971 * Make a copy of both an &sk_buff and its data and while doing so
972 * allocate additional space.
974 * This is used when the caller wishes to modify the data and needs a
975 * private copy of the data to alter as well as more space for new fields.
976 * Returns %NULL on failure or the pointer to the buffer
977 * on success. The returned buffer has a reference count of 1.
979 * You must pass %GFP_ATOMIC as the allocation priority if this function
980 * is called from an interrupt.
982 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
983 int newheadroom
, int newtailroom
,
987 * Allocate the copy buffer
989 struct sk_buff
*n
= alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
991 int oldheadroom
= skb_headroom(skb
);
992 int head_copy_len
, head_copy_off
;
998 skb_reserve(n
, newheadroom
);
1000 /* Set the tail pointer and length */
1001 skb_put(n
, skb
->len
);
1003 head_copy_len
= oldheadroom
;
1005 if (newheadroom
<= head_copy_len
)
1006 head_copy_len
= newheadroom
;
1008 head_copy_off
= newheadroom
- head_copy_len
;
1010 /* Copy the linear header and data. */
1011 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
1012 skb
->len
+ head_copy_len
))
1015 copy_skb_header(n
, skb
);
1017 off
= newheadroom
- oldheadroom
;
1018 if (n
->ip_summed
== CHECKSUM_PARTIAL
)
1019 n
->csum_start
+= off
;
1020 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1021 n
->transport_header
+= off
;
1022 n
->network_header
+= off
;
1023 if (skb_mac_header_was_set(skb
))
1024 n
->mac_header
+= off
;
1029 EXPORT_SYMBOL(skb_copy_expand
);
1032 * skb_pad - zero pad the tail of an skb
1033 * @skb: buffer to pad
1034 * @pad: space to pad
1036 * Ensure that a buffer is followed by a padding area that is zero
1037 * filled. Used by network drivers which may DMA or transfer data
1038 * beyond the buffer end onto the wire.
1040 * May return error in out of memory cases. The skb is freed on error.
1043 int skb_pad(struct sk_buff
*skb
, int pad
)
1048 /* If the skbuff is non linear tailroom is always zero.. */
1049 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
1050 memset(skb
->data
+skb
->len
, 0, pad
);
1054 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
1055 if (likely(skb_cloned(skb
) || ntail
> 0)) {
1056 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
1061 /* FIXME: The use of this function with non-linear skb's really needs
1064 err
= skb_linearize(skb
);
1068 memset(skb
->data
+ skb
->len
, 0, pad
);
1075 EXPORT_SYMBOL(skb_pad
);
1078 * skb_put - add data to a buffer
1079 * @skb: buffer to use
1080 * @len: amount of data to add
1082 * This function extends the used data area of the buffer. If this would
1083 * exceed the total buffer size the kernel will panic. A pointer to the
1084 * first byte of the extra data is returned.
1086 unsigned char *skb_put(struct sk_buff
*skb
, unsigned int len
)
1088 unsigned char *tmp
= skb_tail_pointer(skb
);
1089 SKB_LINEAR_ASSERT(skb
);
1092 if (unlikely(skb
->tail
> skb
->end
))
1093 skb_over_panic(skb
, len
, __builtin_return_address(0));
1096 EXPORT_SYMBOL(skb_put
);
1099 * skb_push - add data to the start of a buffer
1100 * @skb: buffer to use
1101 * @len: amount of data to add
1103 * This function extends the used data area of the buffer at the buffer
1104 * start. If this would exceed the total buffer headroom the kernel will
1105 * panic. A pointer to the first byte of the extra data is returned.
1107 unsigned char *skb_push(struct sk_buff
*skb
, unsigned int len
)
1111 if (unlikely(skb
->data
<skb
->head
))
1112 skb_under_panic(skb
, len
, __builtin_return_address(0));
1115 EXPORT_SYMBOL(skb_push
);
1118 * skb_pull - remove data from the start of a buffer
1119 * @skb: buffer to use
1120 * @len: amount of data to remove
1122 * This function removes data from the start of a buffer, returning
1123 * the memory to the headroom. A pointer to the next data in the buffer
1124 * is returned. Once the data has been pulled future pushes will overwrite
1127 unsigned char *skb_pull(struct sk_buff
*skb
, unsigned int len
)
1129 return skb_pull_inline(skb
, len
);
1131 EXPORT_SYMBOL(skb_pull
);
1134 * skb_trim - remove end from a buffer
1135 * @skb: buffer to alter
1138 * Cut the length of a buffer down by removing data from the tail. If
1139 * the buffer is already under the length specified it is not modified.
1140 * The skb must be linear.
1142 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1145 __skb_trim(skb
, len
);
1147 EXPORT_SYMBOL(skb_trim
);
1149 /* Trims skb to length len. It can change skb pointers.
1152 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1154 struct sk_buff
**fragp
;
1155 struct sk_buff
*frag
;
1156 int offset
= skb_headlen(skb
);
1157 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1161 if (skb_cloned(skb
) &&
1162 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1169 for (; i
< nfrags
; i
++) {
1170 int end
= offset
+ skb_shinfo(skb
)->frags
[i
].size
;
1177 skb_shinfo(skb
)->frags
[i
++].size
= len
- offset
;
1180 skb_shinfo(skb
)->nr_frags
= i
;
1182 for (; i
< nfrags
; i
++)
1183 put_page(skb_shinfo(skb
)->frags
[i
].page
);
1185 if (skb_has_frag_list(skb
))
1186 skb_drop_fraglist(skb
);
1190 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1191 fragp
= &frag
->next
) {
1192 int end
= offset
+ frag
->len
;
1194 if (skb_shared(frag
)) {
1195 struct sk_buff
*nfrag
;
1197 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1198 if (unlikely(!nfrag
))
1201 nfrag
->next
= frag
->next
;
1213 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1217 skb_drop_list(&frag
->next
);
1222 if (len
> skb_headlen(skb
)) {
1223 skb
->data_len
-= skb
->len
- len
;
1228 skb_set_tail_pointer(skb
, len
);
1233 EXPORT_SYMBOL(___pskb_trim
);
1236 * __pskb_pull_tail - advance tail of skb header
1237 * @skb: buffer to reallocate
1238 * @delta: number of bytes to advance tail
1240 * The function makes a sense only on a fragmented &sk_buff,
1241 * it expands header moving its tail forward and copying necessary
1242 * data from fragmented part.
1244 * &sk_buff MUST have reference count of 1.
1246 * Returns %NULL (and &sk_buff does not change) if pull failed
1247 * or value of new tail of skb in the case of success.
1249 * All the pointers pointing into skb header may change and must be
1250 * reloaded after call to this function.
1253 /* Moves tail of skb head forward, copying data from fragmented part,
1254 * when it is necessary.
1255 * 1. It may fail due to malloc failure.
1256 * 2. It may change skb pointers.
1258 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1260 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1262 /* If skb has not enough free space at tail, get new one
1263 * plus 128 bytes for future expansions. If we have enough
1264 * room at tail, reallocate without expansion only if skb is cloned.
1266 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1268 if (eat
> 0 || skb_cloned(skb
)) {
1269 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1274 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
1277 /* Optimization: no fragments, no reasons to preestimate
1278 * size of pulled pages. Superb.
1280 if (!skb_has_frag_list(skb
))
1283 /* Estimate size of pulled pages. */
1285 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1286 if (skb_shinfo(skb
)->frags
[i
].size
>= eat
)
1288 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1291 /* If we need update frag list, we are in troubles.
1292 * Certainly, it possible to add an offset to skb data,
1293 * but taking into account that pulling is expected to
1294 * be very rare operation, it is worth to fight against
1295 * further bloating skb head and crucify ourselves here instead.
1296 * Pure masohism, indeed. 8)8)
1299 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1300 struct sk_buff
*clone
= NULL
;
1301 struct sk_buff
*insp
= NULL
;
1306 if (list
->len
<= eat
) {
1307 /* Eaten as whole. */
1312 /* Eaten partially. */
1314 if (skb_shared(list
)) {
1315 /* Sucks! We need to fork list. :-( */
1316 clone
= skb_clone(list
, GFP_ATOMIC
);
1322 /* This may be pulled without
1326 if (!pskb_pull(list
, eat
)) {
1334 /* Free pulled out fragments. */
1335 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1336 skb_shinfo(skb
)->frag_list
= list
->next
;
1339 /* And insert new clone at head. */
1342 skb_shinfo(skb
)->frag_list
= clone
;
1345 /* Success! Now we may commit changes to skb data. */
1350 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1351 if (skb_shinfo(skb
)->frags
[i
].size
<= eat
) {
1352 put_page(skb_shinfo(skb
)->frags
[i
].page
);
1353 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1355 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1357 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1358 skb_shinfo(skb
)->frags
[k
].size
-= eat
;
1364 skb_shinfo(skb
)->nr_frags
= k
;
1367 skb
->data_len
-= delta
;
1369 return skb_tail_pointer(skb
);
1371 EXPORT_SYMBOL(__pskb_pull_tail
);
1374 * skb_copy_bits - copy bits from skb to kernel buffer
1376 * @offset: offset in source
1377 * @to: destination buffer
1378 * @len: number of bytes to copy
1380 * Copy the specified number of bytes from the source skb to the
1381 * destination buffer.
1384 * If its prototype is ever changed,
1385 * check arch/{*}/net/{*}.S files,
1386 * since it is called from BPF assembly code.
1388 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1390 int start
= skb_headlen(skb
);
1391 struct sk_buff
*frag_iter
;
1394 if (offset
> (int)skb
->len
- len
)
1398 if ((copy
= start
- offset
) > 0) {
1401 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
1402 if ((len
-= copy
) == 0)
1408 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1411 WARN_ON(start
> offset
+ len
);
1413 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1414 if ((copy
= end
- offset
) > 0) {
1420 vaddr
= kmap_skb_frag(&skb_shinfo(skb
)->frags
[i
]);
1422 vaddr
+ skb_shinfo(skb
)->frags
[i
].page_offset
+
1423 offset
- start
, copy
);
1424 kunmap_skb_frag(vaddr
);
1426 if ((len
-= copy
) == 0)
1434 skb_walk_frags(skb
, frag_iter
) {
1437 WARN_ON(start
> offset
+ len
);
1439 end
= start
+ frag_iter
->len
;
1440 if ((copy
= end
- offset
) > 0) {
1443 if (skb_copy_bits(frag_iter
, offset
- start
, to
, copy
))
1445 if ((len
-= copy
) == 0)
1459 EXPORT_SYMBOL(skb_copy_bits
);
1462 * Callback from splice_to_pipe(), if we need to release some pages
1463 * at the end of the spd in case we error'ed out in filling the pipe.
1465 static void sock_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
1467 put_page(spd
->pages
[i
]);
1470 static inline struct page
*linear_to_page(struct page
*page
, unsigned int *len
,
1471 unsigned int *offset
,
1472 struct sk_buff
*skb
, struct sock
*sk
)
1474 struct page
*p
= sk
->sk_sndmsg_page
;
1479 p
= sk
->sk_sndmsg_page
= alloc_pages(sk
->sk_allocation
, 0);
1483 off
= sk
->sk_sndmsg_off
= 0;
1484 /* hold one ref to this page until it's full */
1488 off
= sk
->sk_sndmsg_off
;
1489 mlen
= PAGE_SIZE
- off
;
1490 if (mlen
< 64 && mlen
< *len
) {
1495 *len
= min_t(unsigned int, *len
, mlen
);
1498 memcpy(page_address(p
) + off
, page_address(page
) + *offset
, *len
);
1499 sk
->sk_sndmsg_off
+= *len
;
1507 * Fill page/offset/length into spd, if it can hold more pages.
1509 static inline int spd_fill_page(struct splice_pipe_desc
*spd
,
1510 struct pipe_inode_info
*pipe
, struct page
*page
,
1511 unsigned int *len
, unsigned int offset
,
1512 struct sk_buff
*skb
, int linear
,
1515 if (unlikely(spd
->nr_pages
== pipe
->buffers
))
1519 page
= linear_to_page(page
, len
, &offset
, skb
, sk
);
1525 spd
->pages
[spd
->nr_pages
] = page
;
1526 spd
->partial
[spd
->nr_pages
].len
= *len
;
1527 spd
->partial
[spd
->nr_pages
].offset
= offset
;
1533 static inline void __segment_seek(struct page
**page
, unsigned int *poff
,
1534 unsigned int *plen
, unsigned int off
)
1539 n
= *poff
/ PAGE_SIZE
;
1541 *page
= nth_page(*page
, n
);
1543 *poff
= *poff
% PAGE_SIZE
;
1547 static inline int __splice_segment(struct page
*page
, unsigned int poff
,
1548 unsigned int plen
, unsigned int *off
,
1549 unsigned int *len
, struct sk_buff
*skb
,
1550 struct splice_pipe_desc
*spd
, int linear
,
1552 struct pipe_inode_info
*pipe
)
1557 /* skip this segment if already processed */
1563 /* ignore any bits we already processed */
1565 __segment_seek(&page
, &poff
, &plen
, *off
);
1570 unsigned int flen
= min(*len
, plen
);
1572 /* the linear region may spread across several pages */
1573 flen
= min_t(unsigned int, flen
, PAGE_SIZE
- poff
);
1575 if (spd_fill_page(spd
, pipe
, page
, &flen
, poff
, skb
, linear
, sk
))
1578 __segment_seek(&page
, &poff
, &plen
, flen
);
1581 } while (*len
&& plen
);
1587 * Map linear and fragment data from the skb to spd. It reports failure if the
1588 * pipe is full or if we already spliced the requested length.
1590 static int __skb_splice_bits(struct sk_buff
*skb
, struct pipe_inode_info
*pipe
,
1591 unsigned int *offset
, unsigned int *len
,
1592 struct splice_pipe_desc
*spd
, struct sock
*sk
)
1597 * map the linear part
1599 if (__splice_segment(virt_to_page(skb
->data
),
1600 (unsigned long) skb
->data
& (PAGE_SIZE
- 1),
1602 offset
, len
, skb
, spd
, 1, sk
, pipe
))
1606 * then map the fragments
1608 for (seg
= 0; seg
< skb_shinfo(skb
)->nr_frags
; seg
++) {
1609 const skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[seg
];
1611 if (__splice_segment(f
->page
, f
->page_offset
, f
->size
,
1612 offset
, len
, skb
, spd
, 0, sk
, pipe
))
1620 * Map data from the skb to a pipe. Should handle both the linear part,
1621 * the fragments, and the frag list. It does NOT handle frag lists within
1622 * the frag list, if such a thing exists. We'd probably need to recurse to
1623 * handle that cleanly.
1625 int skb_splice_bits(struct sk_buff
*skb
, unsigned int offset
,
1626 struct pipe_inode_info
*pipe
, unsigned int tlen
,
1629 struct partial_page partial
[PIPE_DEF_BUFFERS
];
1630 struct page
*pages
[PIPE_DEF_BUFFERS
];
1631 struct splice_pipe_desc spd
= {
1635 .ops
= &sock_pipe_buf_ops
,
1636 .spd_release
= sock_spd_release
,
1638 struct sk_buff
*frag_iter
;
1639 struct sock
*sk
= skb
->sk
;
1642 if (splice_grow_spd(pipe
, &spd
))
1646 * __skb_splice_bits() only fails if the output has no room left,
1647 * so no point in going over the frag_list for the error case.
1649 if (__skb_splice_bits(skb
, pipe
, &offset
, &tlen
, &spd
, sk
))
1655 * now see if we have a frag_list to map
1657 skb_walk_frags(skb
, frag_iter
) {
1660 if (__skb_splice_bits(frag_iter
, pipe
, &offset
, &tlen
, &spd
, sk
))
1667 * Drop the socket lock, otherwise we have reverse
1668 * locking dependencies between sk_lock and i_mutex
1669 * here as compared to sendfile(). We enter here
1670 * with the socket lock held, and splice_to_pipe() will
1671 * grab the pipe inode lock. For sendfile() emulation,
1672 * we call into ->sendpage() with the i_mutex lock held
1673 * and networking will grab the socket lock.
1676 ret
= splice_to_pipe(pipe
, &spd
);
1680 splice_shrink_spd(pipe
, &spd
);
1685 * skb_store_bits - store bits from kernel buffer to skb
1686 * @skb: destination buffer
1687 * @offset: offset in destination
1688 * @from: source buffer
1689 * @len: number of bytes to copy
1691 * Copy the specified number of bytes from the source buffer to the
1692 * destination skb. This function handles all the messy bits of
1693 * traversing fragment lists and such.
1696 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
1698 int start
= skb_headlen(skb
);
1699 struct sk_buff
*frag_iter
;
1702 if (offset
> (int)skb
->len
- len
)
1705 if ((copy
= start
- offset
) > 0) {
1708 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
1709 if ((len
-= copy
) == 0)
1715 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1716 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1719 WARN_ON(start
> offset
+ len
);
1721 end
= start
+ frag
->size
;
1722 if ((copy
= end
- offset
) > 0) {
1728 vaddr
= kmap_skb_frag(frag
);
1729 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1731 kunmap_skb_frag(vaddr
);
1733 if ((len
-= copy
) == 0)
1741 skb_walk_frags(skb
, frag_iter
) {
1744 WARN_ON(start
> offset
+ len
);
1746 end
= start
+ frag_iter
->len
;
1747 if ((copy
= end
- offset
) > 0) {
1750 if (skb_store_bits(frag_iter
, offset
- start
,
1753 if ((len
-= copy
) == 0)
1766 EXPORT_SYMBOL(skb_store_bits
);
1768 /* Checksum skb data. */
1770 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
1771 int len
, __wsum csum
)
1773 int start
= skb_headlen(skb
);
1774 int i
, copy
= start
- offset
;
1775 struct sk_buff
*frag_iter
;
1778 /* Checksum header. */
1782 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1783 if ((len
-= copy
) == 0)
1789 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1792 WARN_ON(start
> offset
+ len
);
1794 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1795 if ((copy
= end
- offset
) > 0) {
1798 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1802 vaddr
= kmap_skb_frag(frag
);
1803 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
1804 offset
- start
, copy
, 0);
1805 kunmap_skb_frag(vaddr
);
1806 csum
= csum_block_add(csum
, csum2
, pos
);
1815 skb_walk_frags(skb
, frag_iter
) {
1818 WARN_ON(start
> offset
+ len
);
1820 end
= start
+ frag_iter
->len
;
1821 if ((copy
= end
- offset
) > 0) {
1825 csum2
= skb_checksum(frag_iter
, offset
- start
,
1827 csum
= csum_block_add(csum
, csum2
, pos
);
1828 if ((len
-= copy
) == 0)
1839 EXPORT_SYMBOL(skb_checksum
);
1841 /* Both of above in one bottle. */
1843 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
1844 u8
*to
, int len
, __wsum csum
)
1846 int start
= skb_headlen(skb
);
1847 int i
, copy
= start
- offset
;
1848 struct sk_buff
*frag_iter
;
1855 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
1857 if ((len
-= copy
) == 0)
1864 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1867 WARN_ON(start
> offset
+ len
);
1869 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1870 if ((copy
= end
- offset
) > 0) {
1873 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1877 vaddr
= kmap_skb_frag(frag
);
1878 csum2
= csum_partial_copy_nocheck(vaddr
+
1882 kunmap_skb_frag(vaddr
);
1883 csum
= csum_block_add(csum
, csum2
, pos
);
1893 skb_walk_frags(skb
, frag_iter
) {
1897 WARN_ON(start
> offset
+ len
);
1899 end
= start
+ frag_iter
->len
;
1900 if ((copy
= end
- offset
) > 0) {
1903 csum2
= skb_copy_and_csum_bits(frag_iter
,
1906 csum
= csum_block_add(csum
, csum2
, pos
);
1907 if ((len
-= copy
) == 0)
1918 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
1920 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
1925 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1926 csstart
= skb_checksum_start_offset(skb
);
1928 csstart
= skb_headlen(skb
);
1930 BUG_ON(csstart
> skb_headlen(skb
));
1932 skb_copy_from_linear_data(skb
, to
, csstart
);
1935 if (csstart
!= skb
->len
)
1936 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
1937 skb
->len
- csstart
, 0);
1939 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1940 long csstuff
= csstart
+ skb
->csum_offset
;
1942 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
1945 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
1948 * skb_dequeue - remove from the head of the queue
1949 * @list: list to dequeue from
1951 * Remove the head of the list. The list lock is taken so the function
1952 * may be used safely with other locking list functions. The head item is
1953 * returned or %NULL if the list is empty.
1956 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
1958 unsigned long flags
;
1959 struct sk_buff
*result
;
1961 spin_lock_irqsave(&list
->lock
, flags
);
1962 result
= __skb_dequeue(list
);
1963 spin_unlock_irqrestore(&list
->lock
, flags
);
1966 EXPORT_SYMBOL(skb_dequeue
);
1969 * skb_dequeue_tail - remove from the tail of the queue
1970 * @list: list to dequeue from
1972 * Remove the tail of the list. The list lock is taken so the function
1973 * may be used safely with other locking list functions. The tail item is
1974 * returned or %NULL if the list is empty.
1976 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
1978 unsigned long flags
;
1979 struct sk_buff
*result
;
1981 spin_lock_irqsave(&list
->lock
, flags
);
1982 result
= __skb_dequeue_tail(list
);
1983 spin_unlock_irqrestore(&list
->lock
, flags
);
1986 EXPORT_SYMBOL(skb_dequeue_tail
);
1989 * skb_queue_purge - empty a list
1990 * @list: list to empty
1992 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1993 * the list and one reference dropped. This function takes the list
1994 * lock and is atomic with respect to other list locking functions.
1996 void skb_queue_purge(struct sk_buff_head
*list
)
1998 struct sk_buff
*skb
;
1999 while ((skb
= skb_dequeue(list
)) != NULL
)
2002 EXPORT_SYMBOL(skb_queue_purge
);
2005 * skb_queue_head - queue a buffer at the list head
2006 * @list: list to use
2007 * @newsk: buffer to queue
2009 * Queue a buffer at the start of the list. This function takes the
2010 * list lock and can be used safely with other locking &sk_buff functions
2013 * A buffer cannot be placed on two lists at the same time.
2015 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2017 unsigned long flags
;
2019 spin_lock_irqsave(&list
->lock
, flags
);
2020 __skb_queue_head(list
, newsk
);
2021 spin_unlock_irqrestore(&list
->lock
, flags
);
2023 EXPORT_SYMBOL(skb_queue_head
);
2026 * skb_queue_tail - queue a buffer at the list tail
2027 * @list: list to use
2028 * @newsk: buffer to queue
2030 * Queue a buffer at the tail of the list. This function takes the
2031 * list lock and can be used safely with other locking &sk_buff functions
2034 * A buffer cannot be placed on two lists at the same time.
2036 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2038 unsigned long flags
;
2040 spin_lock_irqsave(&list
->lock
, flags
);
2041 __skb_queue_tail(list
, newsk
);
2042 spin_unlock_irqrestore(&list
->lock
, flags
);
2044 EXPORT_SYMBOL(skb_queue_tail
);
2047 * skb_unlink - remove a buffer from a list
2048 * @skb: buffer to remove
2049 * @list: list to use
2051 * Remove a packet from a list. The list locks are taken and this
2052 * function is atomic with respect to other list locked calls
2054 * You must know what list the SKB is on.
2056 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
2058 unsigned long flags
;
2060 spin_lock_irqsave(&list
->lock
, flags
);
2061 __skb_unlink(skb
, list
);
2062 spin_unlock_irqrestore(&list
->lock
, flags
);
2064 EXPORT_SYMBOL(skb_unlink
);
2067 * skb_append - append a buffer
2068 * @old: buffer to insert after
2069 * @newsk: buffer to insert
2070 * @list: list to use
2072 * Place a packet after a given packet in a list. The list locks are taken
2073 * and this function is atomic with respect to other list locked calls.
2074 * A buffer cannot be placed on two lists at the same time.
2076 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2078 unsigned long flags
;
2080 spin_lock_irqsave(&list
->lock
, flags
);
2081 __skb_queue_after(list
, old
, newsk
);
2082 spin_unlock_irqrestore(&list
->lock
, flags
);
2084 EXPORT_SYMBOL(skb_append
);
2087 * skb_insert - insert a buffer
2088 * @old: buffer to insert before
2089 * @newsk: buffer to insert
2090 * @list: list to use
2092 * Place a packet before a given packet in a list. The list locks are
2093 * taken and this function is atomic with respect to other list locked
2096 * A buffer cannot be placed on two lists at the same time.
2098 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2100 unsigned long flags
;
2102 spin_lock_irqsave(&list
->lock
, flags
);
2103 __skb_insert(newsk
, old
->prev
, old
, list
);
2104 spin_unlock_irqrestore(&list
->lock
, flags
);
2106 EXPORT_SYMBOL(skb_insert
);
2108 static inline void skb_split_inside_header(struct sk_buff
*skb
,
2109 struct sk_buff
* skb1
,
2110 const u32 len
, const int pos
)
2114 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
2116 /* And move data appendix as is. */
2117 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
2118 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
2120 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
2121 skb_shinfo(skb
)->nr_frags
= 0;
2122 skb1
->data_len
= skb
->data_len
;
2123 skb1
->len
+= skb1
->data_len
;
2126 skb_set_tail_pointer(skb
, len
);
2129 static inline void skb_split_no_header(struct sk_buff
*skb
,
2130 struct sk_buff
* skb1
,
2131 const u32 len
, int pos
)
2134 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
2136 skb_shinfo(skb
)->nr_frags
= 0;
2137 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
2139 skb
->data_len
= len
- pos
;
2141 for (i
= 0; i
< nfrags
; i
++) {
2142 int size
= skb_shinfo(skb
)->frags
[i
].size
;
2144 if (pos
+ size
> len
) {
2145 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
2149 * We have two variants in this case:
2150 * 1. Move all the frag to the second
2151 * part, if it is possible. F.e.
2152 * this approach is mandatory for TUX,
2153 * where splitting is expensive.
2154 * 2. Split is accurately. We make this.
2156 get_page(skb_shinfo(skb
)->frags
[i
].page
);
2157 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
2158 skb_shinfo(skb1
)->frags
[0].size
-= len
- pos
;
2159 skb_shinfo(skb
)->frags
[i
].size
= len
- pos
;
2160 skb_shinfo(skb
)->nr_frags
++;
2164 skb_shinfo(skb
)->nr_frags
++;
2167 skb_shinfo(skb1
)->nr_frags
= k
;
2171 * skb_split - Split fragmented skb to two parts at length len.
2172 * @skb: the buffer to split
2173 * @skb1: the buffer to receive the second part
2174 * @len: new length for skb
2176 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
2178 int pos
= skb_headlen(skb
);
2180 if (len
< pos
) /* Split line is inside header. */
2181 skb_split_inside_header(skb
, skb1
, len
, pos
);
2182 else /* Second chunk has no header, nothing to copy. */
2183 skb_split_no_header(skb
, skb1
, len
, pos
);
2185 EXPORT_SYMBOL(skb_split
);
2187 /* Shifting from/to a cloned skb is a no-go.
2189 * Caller cannot keep skb_shinfo related pointers past calling here!
2191 static int skb_prepare_for_shift(struct sk_buff
*skb
)
2193 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2197 * skb_shift - Shifts paged data partially from skb to another
2198 * @tgt: buffer into which tail data gets added
2199 * @skb: buffer from which the paged data comes from
2200 * @shiftlen: shift up to this many bytes
2202 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2203 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2204 * It's up to caller to free skb if everything was shifted.
2206 * If @tgt runs out of frags, the whole operation is aborted.
2208 * Skb cannot include anything else but paged data while tgt is allowed
2209 * to have non-paged data as well.
2211 * TODO: full sized shift could be optimized but that would need
2212 * specialized skb free'er to handle frags without up-to-date nr_frags.
2214 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
2216 int from
, to
, merge
, todo
;
2217 struct skb_frag_struct
*fragfrom
, *fragto
;
2219 BUG_ON(shiftlen
> skb
->len
);
2220 BUG_ON(skb_headlen(skb
)); /* Would corrupt stream */
2224 to
= skb_shinfo(tgt
)->nr_frags
;
2225 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2227 /* Actual merge is delayed until the point when we know we can
2228 * commit all, so that we don't have to undo partial changes
2231 !skb_can_coalesce(tgt
, to
, fragfrom
->page
, fragfrom
->page_offset
)) {
2236 todo
-= fragfrom
->size
;
2238 if (skb_prepare_for_shift(skb
) ||
2239 skb_prepare_for_shift(tgt
))
2242 /* All previous frag pointers might be stale! */
2243 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2244 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2246 fragto
->size
+= shiftlen
;
2247 fragfrom
->size
-= shiftlen
;
2248 fragfrom
->page_offset
+= shiftlen
;
2256 /* Skip full, not-fitting skb to avoid expensive operations */
2257 if ((shiftlen
== skb
->len
) &&
2258 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
2261 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
2264 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
2265 if (to
== MAX_SKB_FRAGS
)
2268 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2269 fragto
= &skb_shinfo(tgt
)->frags
[to
];
2271 if (todo
>= fragfrom
->size
) {
2272 *fragto
= *fragfrom
;
2273 todo
-= fragfrom
->size
;
2278 get_page(fragfrom
->page
);
2279 fragto
->page
= fragfrom
->page
;
2280 fragto
->page_offset
= fragfrom
->page_offset
;
2281 fragto
->size
= todo
;
2283 fragfrom
->page_offset
+= todo
;
2284 fragfrom
->size
-= todo
;
2292 /* Ready to "commit" this state change to tgt */
2293 skb_shinfo(tgt
)->nr_frags
= to
;
2296 fragfrom
= &skb_shinfo(skb
)->frags
[0];
2297 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2299 fragto
->size
+= fragfrom
->size
;
2300 put_page(fragfrom
->page
);
2303 /* Reposition in the original skb */
2305 while (from
< skb_shinfo(skb
)->nr_frags
)
2306 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
2307 skb_shinfo(skb
)->nr_frags
= to
;
2309 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
2312 /* Most likely the tgt won't ever need its checksum anymore, skb on
2313 * the other hand might need it if it needs to be resent
2315 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
2316 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2318 /* Yak, is it really working this way? Some helper please? */
2319 skb
->len
-= shiftlen
;
2320 skb
->data_len
-= shiftlen
;
2321 skb
->truesize
-= shiftlen
;
2322 tgt
->len
+= shiftlen
;
2323 tgt
->data_len
+= shiftlen
;
2324 tgt
->truesize
+= shiftlen
;
2330 * skb_prepare_seq_read - Prepare a sequential read of skb data
2331 * @skb: the buffer to read
2332 * @from: lower offset of data to be read
2333 * @to: upper offset of data to be read
2334 * @st: state variable
2336 * Initializes the specified state variable. Must be called before
2337 * invoking skb_seq_read() for the first time.
2339 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
2340 unsigned int to
, struct skb_seq_state
*st
)
2342 st
->lower_offset
= from
;
2343 st
->upper_offset
= to
;
2344 st
->root_skb
= st
->cur_skb
= skb
;
2345 st
->frag_idx
= st
->stepped_offset
= 0;
2346 st
->frag_data
= NULL
;
2348 EXPORT_SYMBOL(skb_prepare_seq_read
);
2351 * skb_seq_read - Sequentially read skb data
2352 * @consumed: number of bytes consumed by the caller so far
2353 * @data: destination pointer for data to be returned
2354 * @st: state variable
2356 * Reads a block of skb data at &consumed relative to the
2357 * lower offset specified to skb_prepare_seq_read(). Assigns
2358 * the head of the data block to &data and returns the length
2359 * of the block or 0 if the end of the skb data or the upper
2360 * offset has been reached.
2362 * The caller is not required to consume all of the data
2363 * returned, i.e. &consumed is typically set to the number
2364 * of bytes already consumed and the next call to
2365 * skb_seq_read() will return the remaining part of the block.
2367 * Note 1: The size of each block of data returned can be arbitrary,
2368 * this limitation is the cost for zerocopy seqeuental
2369 * reads of potentially non linear data.
2371 * Note 2: Fragment lists within fragments are not implemented
2372 * at the moment, state->root_skb could be replaced with
2373 * a stack for this purpose.
2375 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
2376 struct skb_seq_state
*st
)
2378 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
2381 if (unlikely(abs_offset
>= st
->upper_offset
))
2385 block_limit
= skb_headlen(st
->cur_skb
) + st
->stepped_offset
;
2387 if (abs_offset
< block_limit
&& !st
->frag_data
) {
2388 *data
= st
->cur_skb
->data
+ (abs_offset
- st
->stepped_offset
);
2389 return block_limit
- abs_offset
;
2392 if (st
->frag_idx
== 0 && !st
->frag_data
)
2393 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
2395 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
2396 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
2397 block_limit
= frag
->size
+ st
->stepped_offset
;
2399 if (abs_offset
< block_limit
) {
2401 st
->frag_data
= kmap_skb_frag(frag
);
2403 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
2404 (abs_offset
- st
->stepped_offset
);
2406 return block_limit
- abs_offset
;
2409 if (st
->frag_data
) {
2410 kunmap_skb_frag(st
->frag_data
);
2411 st
->frag_data
= NULL
;
2415 st
->stepped_offset
+= frag
->size
;
2418 if (st
->frag_data
) {
2419 kunmap_skb_frag(st
->frag_data
);
2420 st
->frag_data
= NULL
;
2423 if (st
->root_skb
== st
->cur_skb
&& skb_has_frag_list(st
->root_skb
)) {
2424 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
2427 } else if (st
->cur_skb
->next
) {
2428 st
->cur_skb
= st
->cur_skb
->next
;
2435 EXPORT_SYMBOL(skb_seq_read
);
2438 * skb_abort_seq_read - Abort a sequential read of skb data
2439 * @st: state variable
2441 * Must be called if skb_seq_read() was not called until it
2444 void skb_abort_seq_read(struct skb_seq_state
*st
)
2447 kunmap_skb_frag(st
->frag_data
);
2449 EXPORT_SYMBOL(skb_abort_seq_read
);
2451 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2453 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
2454 struct ts_config
*conf
,
2455 struct ts_state
*state
)
2457 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
2460 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
2462 skb_abort_seq_read(TS_SKB_CB(state
));
2466 * skb_find_text - Find a text pattern in skb data
2467 * @skb: the buffer to look in
2468 * @from: search offset
2470 * @config: textsearch configuration
2471 * @state: uninitialized textsearch state variable
2473 * Finds a pattern in the skb data according to the specified
2474 * textsearch configuration. Use textsearch_next() to retrieve
2475 * subsequent occurrences of the pattern. Returns the offset
2476 * to the first occurrence or UINT_MAX if no match was found.
2478 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
2479 unsigned int to
, struct ts_config
*config
,
2480 struct ts_state
*state
)
2484 config
->get_next_block
= skb_ts_get_next_block
;
2485 config
->finish
= skb_ts_finish
;
2487 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
2489 ret
= textsearch_find(config
, state
);
2490 return (ret
<= to
- from
? ret
: UINT_MAX
);
2492 EXPORT_SYMBOL(skb_find_text
);
2495 * skb_append_datato_frags: - append the user data to a skb
2496 * @sk: sock structure
2497 * @skb: skb structure to be appened with user data.
2498 * @getfrag: call back function to be used for getting the user data
2499 * @from: pointer to user message iov
2500 * @length: length of the iov message
2502 * Description: This procedure append the user data in the fragment part
2503 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2505 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
2506 int (*getfrag
)(void *from
, char *to
, int offset
,
2507 int len
, int odd
, struct sk_buff
*skb
),
2508 void *from
, int length
)
2511 skb_frag_t
*frag
= NULL
;
2512 struct page
*page
= NULL
;
2518 /* Return error if we don't have space for new frag */
2519 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2520 if (frg_cnt
>= MAX_SKB_FRAGS
)
2523 /* allocate a new page for next frag */
2524 page
= alloc_pages(sk
->sk_allocation
, 0);
2526 /* If alloc_page fails just return failure and caller will
2527 * free previous allocated pages by doing kfree_skb()
2532 /* initialize the next frag */
2533 skb_fill_page_desc(skb
, frg_cnt
, page
, 0, 0);
2534 skb
->truesize
+= PAGE_SIZE
;
2535 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
2537 /* get the new initialized frag */
2538 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2539 frag
= &skb_shinfo(skb
)->frags
[frg_cnt
- 1];
2541 /* copy the user data to page */
2542 left
= PAGE_SIZE
- frag
->page_offset
;
2543 copy
= (length
> left
)? left
: length
;
2545 ret
= getfrag(from
, (page_address(frag
->page
) +
2546 frag
->page_offset
+ frag
->size
),
2547 offset
, copy
, 0, skb
);
2551 /* copy was successful so update the size parameters */
2554 skb
->data_len
+= copy
;
2558 } while (length
> 0);
2562 EXPORT_SYMBOL(skb_append_datato_frags
);
2565 * skb_pull_rcsum - pull skb and update receive checksum
2566 * @skb: buffer to update
2567 * @len: length of data pulled
2569 * This function performs an skb_pull on the packet and updates
2570 * the CHECKSUM_COMPLETE checksum. It should be used on
2571 * receive path processing instead of skb_pull unless you know
2572 * that the checksum difference is zero (e.g., a valid IP header)
2573 * or you are setting ip_summed to CHECKSUM_NONE.
2575 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
2577 BUG_ON(len
> skb
->len
);
2579 BUG_ON(skb
->len
< skb
->data_len
);
2580 skb_postpull_rcsum(skb
, skb
->data
, len
);
2581 return skb
->data
+= len
;
2583 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
2586 * skb_segment - Perform protocol segmentation on skb.
2587 * @skb: buffer to segment
2588 * @features: features for the output path (see dev->features)
2590 * This function performs segmentation on the given skb. It returns
2591 * a pointer to the first in a list of new skbs for the segments.
2592 * In case of error it returns ERR_PTR(err).
2594 struct sk_buff
*skb_segment(struct sk_buff
*skb
, u32 features
)
2596 struct sk_buff
*segs
= NULL
;
2597 struct sk_buff
*tail
= NULL
;
2598 struct sk_buff
*fskb
= skb_shinfo(skb
)->frag_list
;
2599 unsigned int mss
= skb_shinfo(skb
)->gso_size
;
2600 unsigned int doffset
= skb
->data
- skb_mac_header(skb
);
2601 unsigned int offset
= doffset
;
2602 unsigned int headroom
;
2604 int sg
= !!(features
& NETIF_F_SG
);
2605 int nfrags
= skb_shinfo(skb
)->nr_frags
;
2610 __skb_push(skb
, doffset
);
2611 headroom
= skb_headroom(skb
);
2612 pos
= skb_headlen(skb
);
2615 struct sk_buff
*nskb
;
2620 len
= skb
->len
- offset
;
2624 hsize
= skb_headlen(skb
) - offset
;
2627 if (hsize
> len
|| !sg
)
2630 if (!hsize
&& i
>= nfrags
) {
2631 BUG_ON(fskb
->len
!= len
);
2634 nskb
= skb_clone(fskb
, GFP_ATOMIC
);
2637 if (unlikely(!nskb
))
2640 hsize
= skb_end_pointer(nskb
) - nskb
->head
;
2641 if (skb_cow_head(nskb
, doffset
+ headroom
)) {
2646 nskb
->truesize
+= skb_end_pointer(nskb
) - nskb
->head
-
2648 skb_release_head_state(nskb
);
2649 __skb_push(nskb
, doffset
);
2651 nskb
= alloc_skb(hsize
+ doffset
+ headroom
,
2654 if (unlikely(!nskb
))
2657 skb_reserve(nskb
, headroom
);
2658 __skb_put(nskb
, doffset
);
2667 __copy_skb_header(nskb
, skb
);
2668 nskb
->mac_len
= skb
->mac_len
;
2670 /* nskb and skb might have different headroom */
2671 if (nskb
->ip_summed
== CHECKSUM_PARTIAL
)
2672 nskb
->csum_start
+= skb_headroom(nskb
) - headroom
;
2674 skb_reset_mac_header(nskb
);
2675 skb_set_network_header(nskb
, skb
->mac_len
);
2676 nskb
->transport_header
= (nskb
->network_header
+
2677 skb_network_header_len(skb
));
2678 skb_copy_from_linear_data(skb
, nskb
->data
, doffset
);
2680 if (fskb
!= skb_shinfo(skb
)->frag_list
)
2684 nskb
->ip_summed
= CHECKSUM_NONE
;
2685 nskb
->csum
= skb_copy_and_csum_bits(skb
, offset
,
2691 frag
= skb_shinfo(nskb
)->frags
;
2693 skb_copy_from_linear_data_offset(skb
, offset
,
2694 skb_put(nskb
, hsize
), hsize
);
2696 while (pos
< offset
+ len
&& i
< nfrags
) {
2697 *frag
= skb_shinfo(skb
)->frags
[i
];
2698 get_page(frag
->page
);
2702 frag
->page_offset
+= offset
- pos
;
2703 frag
->size
-= offset
- pos
;
2706 skb_shinfo(nskb
)->nr_frags
++;
2708 if (pos
+ size
<= offset
+ len
) {
2712 frag
->size
-= pos
+ size
- (offset
+ len
);
2719 if (pos
< offset
+ len
) {
2720 struct sk_buff
*fskb2
= fskb
;
2722 BUG_ON(pos
+ fskb
->len
!= offset
+ len
);
2728 fskb2
= skb_clone(fskb2
, GFP_ATOMIC
);
2734 SKB_FRAG_ASSERT(nskb
);
2735 skb_shinfo(nskb
)->frag_list
= fskb2
;
2739 nskb
->data_len
= len
- hsize
;
2740 nskb
->len
+= nskb
->data_len
;
2741 nskb
->truesize
+= nskb
->data_len
;
2742 } while ((offset
+= len
) < skb
->len
);
2747 while ((skb
= segs
)) {
2751 return ERR_PTR(err
);
2753 EXPORT_SYMBOL_GPL(skb_segment
);
2755 int skb_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2757 struct sk_buff
*p
= *head
;
2758 struct sk_buff
*nskb
;
2759 struct skb_shared_info
*skbinfo
= skb_shinfo(skb
);
2760 struct skb_shared_info
*pinfo
= skb_shinfo(p
);
2761 unsigned int headroom
;
2762 unsigned int len
= skb_gro_len(skb
);
2763 unsigned int offset
= skb_gro_offset(skb
);
2764 unsigned int headlen
= skb_headlen(skb
);
2766 if (p
->len
+ len
>= 65536)
2769 if (pinfo
->frag_list
)
2771 else if (headlen
<= offset
) {
2774 int i
= skbinfo
->nr_frags
;
2775 int nr_frags
= pinfo
->nr_frags
+ i
;
2779 if (nr_frags
> MAX_SKB_FRAGS
)
2782 pinfo
->nr_frags
= nr_frags
;
2783 skbinfo
->nr_frags
= 0;
2785 frag
= pinfo
->frags
+ nr_frags
;
2786 frag2
= skbinfo
->frags
+ i
;
2791 frag
->page_offset
+= offset
;
2792 frag
->size
-= offset
;
2794 skb
->truesize
-= skb
->data_len
;
2795 skb
->len
-= skb
->data_len
;
2798 NAPI_GRO_CB(skb
)->free
= 1;
2800 } else if (skb_gro_len(p
) != pinfo
->gso_size
)
2803 headroom
= skb_headroom(p
);
2804 nskb
= alloc_skb(headroom
+ skb_gro_offset(p
), GFP_ATOMIC
);
2805 if (unlikely(!nskb
))
2808 __copy_skb_header(nskb
, p
);
2809 nskb
->mac_len
= p
->mac_len
;
2811 skb_reserve(nskb
, headroom
);
2812 __skb_put(nskb
, skb_gro_offset(p
));
2814 skb_set_mac_header(nskb
, skb_mac_header(p
) - p
->data
);
2815 skb_set_network_header(nskb
, skb_network_offset(p
));
2816 skb_set_transport_header(nskb
, skb_transport_offset(p
));
2818 __skb_pull(p
, skb_gro_offset(p
));
2819 memcpy(skb_mac_header(nskb
), skb_mac_header(p
),
2820 p
->data
- skb_mac_header(p
));
2822 *NAPI_GRO_CB(nskb
) = *NAPI_GRO_CB(p
);
2823 skb_shinfo(nskb
)->frag_list
= p
;
2824 skb_shinfo(nskb
)->gso_size
= pinfo
->gso_size
;
2825 pinfo
->gso_size
= 0;
2826 skb_header_release(p
);
2829 nskb
->data_len
+= p
->len
;
2830 nskb
->truesize
+= p
->len
;
2831 nskb
->len
+= p
->len
;
2834 nskb
->next
= p
->next
;
2840 if (offset
> headlen
) {
2841 unsigned int eat
= offset
- headlen
;
2843 skbinfo
->frags
[0].page_offset
+= eat
;
2844 skbinfo
->frags
[0].size
-= eat
;
2845 skb
->data_len
-= eat
;
2850 __skb_pull(skb
, offset
);
2852 p
->prev
->next
= skb
;
2854 skb_header_release(skb
);
2857 NAPI_GRO_CB(p
)->count
++;
2862 NAPI_GRO_CB(skb
)->same_flow
= 1;
2865 EXPORT_SYMBOL_GPL(skb_gro_receive
);
2867 void __init
skb_init(void)
2869 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
2870 sizeof(struct sk_buff
),
2872 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2874 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
2875 (2*sizeof(struct sk_buff
)) +
2878 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2883 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2884 * @skb: Socket buffer containing the buffers to be mapped
2885 * @sg: The scatter-gather list to map into
2886 * @offset: The offset into the buffer's contents to start mapping
2887 * @len: Length of buffer space to be mapped
2889 * Fill the specified scatter-gather list with mappings/pointers into a
2890 * region of the buffer space attached to a socket buffer.
2893 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2895 int start
= skb_headlen(skb
);
2896 int i
, copy
= start
- offset
;
2897 struct sk_buff
*frag_iter
;
2903 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
2905 if ((len
-= copy
) == 0)
2910 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2913 WARN_ON(start
> offset
+ len
);
2915 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
2916 if ((copy
= end
- offset
) > 0) {
2917 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2921 sg_set_page(&sg
[elt
], frag
->page
, copy
,
2922 frag
->page_offset
+offset
-start
);
2931 skb_walk_frags(skb
, frag_iter
) {
2934 WARN_ON(start
> offset
+ len
);
2936 end
= start
+ frag_iter
->len
;
2937 if ((copy
= end
- offset
) > 0) {
2940 elt
+= __skb_to_sgvec(frag_iter
, sg
+elt
, offset
- start
,
2942 if ((len
-= copy
) == 0)
2952 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2954 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
);
2956 sg_mark_end(&sg
[nsg
- 1]);
2960 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
2963 * skb_cow_data - Check that a socket buffer's data buffers are writable
2964 * @skb: The socket buffer to check.
2965 * @tailbits: Amount of trailing space to be added
2966 * @trailer: Returned pointer to the skb where the @tailbits space begins
2968 * Make sure that the data buffers attached to a socket buffer are
2969 * writable. If they are not, private copies are made of the data buffers
2970 * and the socket buffer is set to use these instead.
2972 * If @tailbits is given, make sure that there is space to write @tailbits
2973 * bytes of data beyond current end of socket buffer. @trailer will be
2974 * set to point to the skb in which this space begins.
2976 * The number of scatterlist elements required to completely map the
2977 * COW'd and extended socket buffer will be returned.
2979 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
2983 struct sk_buff
*skb1
, **skb_p
;
2985 /* If skb is cloned or its head is paged, reallocate
2986 * head pulling out all the pages (pages are considered not writable
2987 * at the moment even if they are anonymous).
2989 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
2990 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
2993 /* Easy case. Most of packets will go this way. */
2994 if (!skb_has_frag_list(skb
)) {
2995 /* A little of trouble, not enough of space for trailer.
2996 * This should not happen, when stack is tuned to generate
2997 * good frames. OK, on miss we reallocate and reserve even more
2998 * space, 128 bytes is fair. */
3000 if (skb_tailroom(skb
) < tailbits
&&
3001 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
3009 /* Misery. We are in troubles, going to mincer fragments... */
3012 skb_p
= &skb_shinfo(skb
)->frag_list
;
3015 while ((skb1
= *skb_p
) != NULL
) {
3018 /* The fragment is partially pulled by someone,
3019 * this can happen on input. Copy it and everything
3022 if (skb_shared(skb1
))
3025 /* If the skb is the last, worry about trailer. */
3027 if (skb1
->next
== NULL
&& tailbits
) {
3028 if (skb_shinfo(skb1
)->nr_frags
||
3029 skb_has_frag_list(skb1
) ||
3030 skb_tailroom(skb1
) < tailbits
)
3031 ntail
= tailbits
+ 128;
3037 skb_shinfo(skb1
)->nr_frags
||
3038 skb_has_frag_list(skb1
)) {
3039 struct sk_buff
*skb2
;
3041 /* Fuck, we are miserable poor guys... */
3043 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
3045 skb2
= skb_copy_expand(skb1
,
3049 if (unlikely(skb2
== NULL
))
3053 skb_set_owner_w(skb2
, skb1
->sk
);
3055 /* Looking around. Are we still alive?
3056 * OK, link new skb, drop old one */
3058 skb2
->next
= skb1
->next
;
3065 skb_p
= &skb1
->next
;
3070 EXPORT_SYMBOL_GPL(skb_cow_data
);
3072 static void sock_rmem_free(struct sk_buff
*skb
)
3074 struct sock
*sk
= skb
->sk
;
3076 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
3080 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3082 int sock_queue_err_skb(struct sock
*sk
, struct sk_buff
*skb
)
3084 if (atomic_read(&sk
->sk_rmem_alloc
) + skb
->truesize
>=
3085 (unsigned)sk
->sk_rcvbuf
)
3090 skb
->destructor
= sock_rmem_free
;
3091 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
3093 /* before exiting rcu section, make sure dst is refcounted */
3096 skb_queue_tail(&sk
->sk_error_queue
, skb
);
3097 if (!sock_flag(sk
, SOCK_DEAD
))
3098 sk
->sk_data_ready(sk
, skb
->len
);
3101 EXPORT_SYMBOL(sock_queue_err_skb
);
3103 void skb_tstamp_tx(struct sk_buff
*orig_skb
,
3104 struct skb_shared_hwtstamps
*hwtstamps
)
3106 struct sock
*sk
= orig_skb
->sk
;
3107 struct sock_exterr_skb
*serr
;
3108 struct sk_buff
*skb
;
3114 skb
= skb_clone(orig_skb
, GFP_ATOMIC
);
3119 *skb_hwtstamps(skb
) =
3123 * no hardware time stamps available,
3124 * so keep the shared tx_flags and only
3125 * store software time stamp
3127 skb
->tstamp
= ktime_get_real();
3130 serr
= SKB_EXT_ERR(skb
);
3131 memset(serr
, 0, sizeof(*serr
));
3132 serr
->ee
.ee_errno
= ENOMSG
;
3133 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TIMESTAMPING
;
3135 err
= sock_queue_err_skb(sk
, skb
);
3140 EXPORT_SYMBOL_GPL(skb_tstamp_tx
);
3144 * skb_partial_csum_set - set up and verify partial csum values for packet
3145 * @skb: the skb to set
3146 * @start: the number of bytes after skb->data to start checksumming.
3147 * @off: the offset from start to place the checksum.
3149 * For untrusted partially-checksummed packets, we need to make sure the values
3150 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3152 * This function checks and sets those values and skb->ip_summed: if this
3153 * returns false you should drop the packet.
3155 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
3157 if (unlikely(start
> skb_headlen(skb
)) ||
3158 unlikely((int)start
+ off
> skb_headlen(skb
) - 2)) {
3159 if (net_ratelimit())
3161 "bad partial csum: csum=%u/%u len=%u\n",
3162 start
, off
, skb_headlen(skb
));
3165 skb
->ip_summed
= CHECKSUM_PARTIAL
;
3166 skb
->csum_start
= skb_headroom(skb
) + start
;
3167 skb
->csum_offset
= off
;
3170 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);
3172 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
3174 if (net_ratelimit())
3175 pr_warning("%s: received packets cannot be forwarded"
3176 " while LRO is enabled\n", skb
->dev
->name
);
3178 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);