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->ooo_okay
= old
->ooo_okay
;
533 new->l4_rxhash
= old
->l4_rxhash
;
535 new->sp
= secpath_get(old
->sp
);
537 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
538 new->csum
= old
->csum
;
539 new->local_df
= old
->local_df
;
540 new->pkt_type
= old
->pkt_type
;
541 new->ip_summed
= old
->ip_summed
;
542 skb_copy_queue_mapping(new, old
);
543 new->priority
= old
->priority
;
544 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
545 new->ipvs_property
= old
->ipvs_property
;
547 new->protocol
= old
->protocol
;
548 new->mark
= old
->mark
;
549 new->skb_iif
= old
->skb_iif
;
551 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
552 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
553 new->nf_trace
= old
->nf_trace
;
555 #ifdef CONFIG_NET_SCHED
556 new->tc_index
= old
->tc_index
;
557 #ifdef CONFIG_NET_CLS_ACT
558 new->tc_verd
= old
->tc_verd
;
561 new->vlan_tci
= old
->vlan_tci
;
563 skb_copy_secmark(new, old
);
567 * You should not add any new code to this function. Add it to
568 * __copy_skb_header above instead.
570 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
572 #define C(x) n->x = skb->x
574 n
->next
= n
->prev
= NULL
;
576 __copy_skb_header(n
, skb
);
581 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
584 n
->destructor
= NULL
;
590 atomic_set(&n
->users
, 1);
592 atomic_inc(&(skb_shinfo(skb
)->dataref
));
600 * skb_morph - morph one skb into another
601 * @dst: the skb to receive the contents
602 * @src: the skb to supply the contents
604 * This is identical to skb_clone except that the target skb is
605 * supplied by the user.
607 * The target skb is returned upon exit.
609 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
611 skb_release_all(dst
);
612 return __skb_clone(dst
, src
);
614 EXPORT_SYMBOL_GPL(skb_morph
);
616 /* skb frags copy userspace buffers to kernel */
617 static int skb_copy_ubufs(struct sk_buff
*skb
, gfp_t gfp_mask
)
620 int num_frags
= skb_shinfo(skb
)->nr_frags
;
621 struct page
*page
, *head
= NULL
;
622 struct ubuf_info
*uarg
= skb_shinfo(skb
)->destructor_arg
;
624 for (i
= 0; i
< num_frags
; i
++) {
626 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
628 page
= alloc_page(GFP_ATOMIC
);
631 struct page
*next
= (struct page
*)head
->private;
637 vaddr
= kmap_skb_frag(&skb_shinfo(skb
)->frags
[i
]);
638 memcpy(page_address(page
),
639 vaddr
+ f
->page_offset
, f
->size
);
640 kunmap_skb_frag(vaddr
);
641 page
->private = (unsigned long)head
;
645 /* skb frags release userspace buffers */
646 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
647 put_page(skb_shinfo(skb
)->frags
[i
].page
);
649 uarg
->callback(uarg
);
651 /* skb frags point to kernel buffers */
652 for (i
= skb_shinfo(skb
)->nr_frags
; i
> 0; i
--) {
653 skb_shinfo(skb
)->frags
[i
- 1].page_offset
= 0;
654 skb_shinfo(skb
)->frags
[i
- 1].page
= head
;
655 head
= (struct page
*)head
->private;
662 * skb_clone - duplicate an sk_buff
663 * @skb: buffer to clone
664 * @gfp_mask: allocation priority
666 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
667 * copies share the same packet data but not structure. The new
668 * buffer has a reference count of 1. If the allocation fails the
669 * function returns %NULL otherwise the new buffer is returned.
671 * If this function is called from an interrupt gfp_mask() must be
675 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
679 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
680 if (skb_copy_ubufs(skb
, gfp_mask
))
682 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
686 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
687 n
->fclone
== SKB_FCLONE_UNAVAILABLE
) {
688 atomic_t
*fclone_ref
= (atomic_t
*) (n
+ 1);
689 n
->fclone
= SKB_FCLONE_CLONE
;
690 atomic_inc(fclone_ref
);
692 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
696 kmemcheck_annotate_bitfield(n
, flags1
);
697 kmemcheck_annotate_bitfield(n
, flags2
);
698 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
701 return __skb_clone(n
, skb
);
703 EXPORT_SYMBOL(skb_clone
);
705 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
707 #ifndef NET_SKBUFF_DATA_USES_OFFSET
709 * Shift between the two data areas in bytes
711 unsigned long offset
= new->data
- old
->data
;
714 __copy_skb_header(new, old
);
716 #ifndef NET_SKBUFF_DATA_USES_OFFSET
717 /* {transport,network,mac}_header are relative to skb->head */
718 new->transport_header
+= offset
;
719 new->network_header
+= offset
;
720 if (skb_mac_header_was_set(new))
721 new->mac_header
+= offset
;
723 skb_shinfo(new)->gso_size
= skb_shinfo(old
)->gso_size
;
724 skb_shinfo(new)->gso_segs
= skb_shinfo(old
)->gso_segs
;
725 skb_shinfo(new)->gso_type
= skb_shinfo(old
)->gso_type
;
729 * skb_copy - create private copy of an sk_buff
730 * @skb: buffer to copy
731 * @gfp_mask: allocation priority
733 * Make a copy of both an &sk_buff and its data. This is used when the
734 * caller wishes to modify the data and needs a private copy of the
735 * data to alter. Returns %NULL on failure or the pointer to the buffer
736 * on success. The returned buffer has a reference count of 1.
738 * As by-product this function converts non-linear &sk_buff to linear
739 * one, so that &sk_buff becomes completely private and caller is allowed
740 * to modify all the data of returned buffer. This means that this
741 * function is not recommended for use in circumstances when only
742 * header is going to be modified. Use pskb_copy() instead.
745 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
747 int headerlen
= skb_headroom(skb
);
748 unsigned int size
= (skb_end_pointer(skb
) - skb
->head
) + skb
->data_len
;
749 struct sk_buff
*n
= alloc_skb(size
, gfp_mask
);
754 /* Set the data pointer */
755 skb_reserve(n
, headerlen
);
756 /* Set the tail pointer and length */
757 skb_put(n
, skb
->len
);
759 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
762 copy_skb_header(n
, skb
);
765 EXPORT_SYMBOL(skb_copy
);
768 * pskb_copy - create copy of an sk_buff with private head.
769 * @skb: buffer to copy
770 * @gfp_mask: allocation priority
772 * Make a copy of both an &sk_buff and part of its data, located
773 * in header. Fragmented data remain shared. This is used when
774 * the caller wishes to modify only header of &sk_buff and needs
775 * private copy of the header to alter. Returns %NULL on failure
776 * or the pointer to the buffer on success.
777 * The returned buffer has a reference count of 1.
780 struct sk_buff
*pskb_copy(struct sk_buff
*skb
, gfp_t gfp_mask
)
782 unsigned int size
= skb_end_pointer(skb
) - skb
->head
;
783 struct sk_buff
*n
= alloc_skb(size
, gfp_mask
);
788 /* Set the data pointer */
789 skb_reserve(n
, skb_headroom(skb
));
790 /* Set the tail pointer and length */
791 skb_put(n
, skb_headlen(skb
));
793 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
795 n
->truesize
+= skb
->data_len
;
796 n
->data_len
= skb
->data_len
;
799 if (skb_shinfo(skb
)->nr_frags
) {
802 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
803 if (skb_copy_ubufs(skb
, gfp_mask
)) {
808 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
810 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
811 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
812 get_page(skb_shinfo(n
)->frags
[i
].page
);
814 skb_shinfo(n
)->nr_frags
= i
;
817 if (skb_has_frag_list(skb
)) {
818 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
819 skb_clone_fraglist(n
);
822 copy_skb_header(n
, skb
);
826 EXPORT_SYMBOL(pskb_copy
);
829 * pskb_expand_head - reallocate header of &sk_buff
830 * @skb: buffer to reallocate
831 * @nhead: room to add at head
832 * @ntail: room to add at tail
833 * @gfp_mask: allocation priority
835 * Expands (or creates identical copy, if &nhead and &ntail are zero)
836 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
837 * reference count of 1. Returns zero in the case of success or error,
838 * if expansion failed. In the last case, &sk_buff is not changed.
840 * All the pointers pointing into skb header may change and must be
841 * reloaded after call to this function.
844 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
849 int size
= nhead
+ (skb_end_pointer(skb
) - skb
->head
) + ntail
;
858 size
= SKB_DATA_ALIGN(size
);
860 /* Check if we can avoid taking references on fragments if we own
861 * the last reference on skb->head. (see skb_release_data())
866 int delta
= skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1;
867 fastpath
= atomic_read(&skb_shinfo(skb
)->dataref
) == delta
;
871 size
+ sizeof(struct skb_shared_info
) <= ksize(skb
->head
)) {
872 memmove(skb
->head
+ size
, skb_shinfo(skb
),
873 offsetof(struct skb_shared_info
,
874 frags
[skb_shinfo(skb
)->nr_frags
]));
875 memmove(skb
->head
+ nhead
, skb
->head
,
876 skb_tail_pointer(skb
) - skb
->head
);
881 data
= kmalloc(size
+ sizeof(struct skb_shared_info
), gfp_mask
);
885 /* Copy only real data... and, alas, header. This should be
886 * optimized for the cases when header is void.
888 memcpy(data
+ nhead
, skb
->head
, skb_tail_pointer(skb
) - skb
->head
);
890 memcpy((struct skb_shared_info
*)(data
+ size
),
892 offsetof(struct skb_shared_info
, frags
[skb_shinfo(skb
)->nr_frags
]));
897 /* copy this zero copy skb frags */
898 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
899 if (skb_copy_ubufs(skb
, gfp_mask
))
901 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
903 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
904 get_page(skb_shinfo(skb
)->frags
[i
].page
);
906 if (skb_has_frag_list(skb
))
907 skb_clone_fraglist(skb
);
909 skb_release_data(skb
);
911 off
= (data
+ nhead
) - skb
->head
;
916 #ifdef NET_SKBUFF_DATA_USES_OFFSET
920 skb
->end
= skb
->head
+ size
;
922 /* {transport,network,mac}_header and tail are relative to skb->head */
924 skb
->transport_header
+= off
;
925 skb
->network_header
+= off
;
926 if (skb_mac_header_was_set(skb
))
927 skb
->mac_header
+= off
;
928 /* Only adjust this if it actually is csum_start rather than csum */
929 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
930 skb
->csum_start
+= nhead
;
934 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
942 EXPORT_SYMBOL(pskb_expand_head
);
944 /* Make private copy of skb with writable head and some headroom */
946 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
948 struct sk_buff
*skb2
;
949 int delta
= headroom
- skb_headroom(skb
);
952 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
954 skb2
= skb_clone(skb
, GFP_ATOMIC
);
955 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
963 EXPORT_SYMBOL(skb_realloc_headroom
);
966 * skb_copy_expand - copy and expand sk_buff
967 * @skb: buffer to copy
968 * @newheadroom: new free bytes at head
969 * @newtailroom: new free bytes at tail
970 * @gfp_mask: allocation priority
972 * Make a copy of both an &sk_buff and its data and while doing so
973 * allocate additional space.
975 * This is used when the caller wishes to modify the data and needs a
976 * private copy of the data to alter as well as more space for new fields.
977 * Returns %NULL on failure or the pointer to the buffer
978 * on success. The returned buffer has a reference count of 1.
980 * You must pass %GFP_ATOMIC as the allocation priority if this function
981 * is called from an interrupt.
983 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
984 int newheadroom
, int newtailroom
,
988 * Allocate the copy buffer
990 struct sk_buff
*n
= alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
992 int oldheadroom
= skb_headroom(skb
);
993 int head_copy_len
, head_copy_off
;
999 skb_reserve(n
, newheadroom
);
1001 /* Set the tail pointer and length */
1002 skb_put(n
, skb
->len
);
1004 head_copy_len
= oldheadroom
;
1006 if (newheadroom
<= head_copy_len
)
1007 head_copy_len
= newheadroom
;
1009 head_copy_off
= newheadroom
- head_copy_len
;
1011 /* Copy the linear header and data. */
1012 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
1013 skb
->len
+ head_copy_len
))
1016 copy_skb_header(n
, skb
);
1018 off
= newheadroom
- oldheadroom
;
1019 if (n
->ip_summed
== CHECKSUM_PARTIAL
)
1020 n
->csum_start
+= off
;
1021 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1022 n
->transport_header
+= off
;
1023 n
->network_header
+= off
;
1024 if (skb_mac_header_was_set(skb
))
1025 n
->mac_header
+= off
;
1030 EXPORT_SYMBOL(skb_copy_expand
);
1033 * skb_pad - zero pad the tail of an skb
1034 * @skb: buffer to pad
1035 * @pad: space to pad
1037 * Ensure that a buffer is followed by a padding area that is zero
1038 * filled. Used by network drivers which may DMA or transfer data
1039 * beyond the buffer end onto the wire.
1041 * May return error in out of memory cases. The skb is freed on error.
1044 int skb_pad(struct sk_buff
*skb
, int pad
)
1049 /* If the skbuff is non linear tailroom is always zero.. */
1050 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
1051 memset(skb
->data
+skb
->len
, 0, pad
);
1055 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
1056 if (likely(skb_cloned(skb
) || ntail
> 0)) {
1057 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
1062 /* FIXME: The use of this function with non-linear skb's really needs
1065 err
= skb_linearize(skb
);
1069 memset(skb
->data
+ skb
->len
, 0, pad
);
1076 EXPORT_SYMBOL(skb_pad
);
1079 * skb_put - add data to a buffer
1080 * @skb: buffer to use
1081 * @len: amount of data to add
1083 * This function extends the used data area of the buffer. If this would
1084 * exceed the total buffer size the kernel will panic. A pointer to the
1085 * first byte of the extra data is returned.
1087 unsigned char *skb_put(struct sk_buff
*skb
, unsigned int len
)
1089 unsigned char *tmp
= skb_tail_pointer(skb
);
1090 SKB_LINEAR_ASSERT(skb
);
1093 if (unlikely(skb
->tail
> skb
->end
))
1094 skb_over_panic(skb
, len
, __builtin_return_address(0));
1097 EXPORT_SYMBOL(skb_put
);
1100 * skb_push - add data to the start of a buffer
1101 * @skb: buffer to use
1102 * @len: amount of data to add
1104 * This function extends the used data area of the buffer at the buffer
1105 * start. If this would exceed the total buffer headroom the kernel will
1106 * panic. A pointer to the first byte of the extra data is returned.
1108 unsigned char *skb_push(struct sk_buff
*skb
, unsigned int len
)
1112 if (unlikely(skb
->data
<skb
->head
))
1113 skb_under_panic(skb
, len
, __builtin_return_address(0));
1116 EXPORT_SYMBOL(skb_push
);
1119 * skb_pull - remove data from the start of a buffer
1120 * @skb: buffer to use
1121 * @len: amount of data to remove
1123 * This function removes data from the start of a buffer, returning
1124 * the memory to the headroom. A pointer to the next data in the buffer
1125 * is returned. Once the data has been pulled future pushes will overwrite
1128 unsigned char *skb_pull(struct sk_buff
*skb
, unsigned int len
)
1130 return skb_pull_inline(skb
, len
);
1132 EXPORT_SYMBOL(skb_pull
);
1135 * skb_trim - remove end from a buffer
1136 * @skb: buffer to alter
1139 * Cut the length of a buffer down by removing data from the tail. If
1140 * the buffer is already under the length specified it is not modified.
1141 * The skb must be linear.
1143 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1146 __skb_trim(skb
, len
);
1148 EXPORT_SYMBOL(skb_trim
);
1150 /* Trims skb to length len. It can change skb pointers.
1153 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1155 struct sk_buff
**fragp
;
1156 struct sk_buff
*frag
;
1157 int offset
= skb_headlen(skb
);
1158 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1162 if (skb_cloned(skb
) &&
1163 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1170 for (; i
< nfrags
; i
++) {
1171 int end
= offset
+ skb_shinfo(skb
)->frags
[i
].size
;
1178 skb_shinfo(skb
)->frags
[i
++].size
= len
- offset
;
1181 skb_shinfo(skb
)->nr_frags
= i
;
1183 for (; i
< nfrags
; i
++)
1184 put_page(skb_shinfo(skb
)->frags
[i
].page
);
1186 if (skb_has_frag_list(skb
))
1187 skb_drop_fraglist(skb
);
1191 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1192 fragp
= &frag
->next
) {
1193 int end
= offset
+ frag
->len
;
1195 if (skb_shared(frag
)) {
1196 struct sk_buff
*nfrag
;
1198 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1199 if (unlikely(!nfrag
))
1202 nfrag
->next
= frag
->next
;
1214 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1218 skb_drop_list(&frag
->next
);
1223 if (len
> skb_headlen(skb
)) {
1224 skb
->data_len
-= skb
->len
- len
;
1229 skb_set_tail_pointer(skb
, len
);
1234 EXPORT_SYMBOL(___pskb_trim
);
1237 * __pskb_pull_tail - advance tail of skb header
1238 * @skb: buffer to reallocate
1239 * @delta: number of bytes to advance tail
1241 * The function makes a sense only on a fragmented &sk_buff,
1242 * it expands header moving its tail forward and copying necessary
1243 * data from fragmented part.
1245 * &sk_buff MUST have reference count of 1.
1247 * Returns %NULL (and &sk_buff does not change) if pull failed
1248 * or value of new tail of skb in the case of success.
1250 * All the pointers pointing into skb header may change and must be
1251 * reloaded after call to this function.
1254 /* Moves tail of skb head forward, copying data from fragmented part,
1255 * when it is necessary.
1256 * 1. It may fail due to malloc failure.
1257 * 2. It may change skb pointers.
1259 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1261 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1263 /* If skb has not enough free space at tail, get new one
1264 * plus 128 bytes for future expansions. If we have enough
1265 * room at tail, reallocate without expansion only if skb is cloned.
1267 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1269 if (eat
> 0 || skb_cloned(skb
)) {
1270 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1275 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
1278 /* Optimization: no fragments, no reasons to preestimate
1279 * size of pulled pages. Superb.
1281 if (!skb_has_frag_list(skb
))
1284 /* Estimate size of pulled pages. */
1286 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1287 if (skb_shinfo(skb
)->frags
[i
].size
>= eat
)
1289 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1292 /* If we need update frag list, we are in troubles.
1293 * Certainly, it possible to add an offset to skb data,
1294 * but taking into account that pulling is expected to
1295 * be very rare operation, it is worth to fight against
1296 * further bloating skb head and crucify ourselves here instead.
1297 * Pure masohism, indeed. 8)8)
1300 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1301 struct sk_buff
*clone
= NULL
;
1302 struct sk_buff
*insp
= NULL
;
1307 if (list
->len
<= eat
) {
1308 /* Eaten as whole. */
1313 /* Eaten partially. */
1315 if (skb_shared(list
)) {
1316 /* Sucks! We need to fork list. :-( */
1317 clone
= skb_clone(list
, GFP_ATOMIC
);
1323 /* This may be pulled without
1327 if (!pskb_pull(list
, eat
)) {
1335 /* Free pulled out fragments. */
1336 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1337 skb_shinfo(skb
)->frag_list
= list
->next
;
1340 /* And insert new clone at head. */
1343 skb_shinfo(skb
)->frag_list
= clone
;
1346 /* Success! Now we may commit changes to skb data. */
1351 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1352 if (skb_shinfo(skb
)->frags
[i
].size
<= eat
) {
1353 put_page(skb_shinfo(skb
)->frags
[i
].page
);
1354 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1356 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1358 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1359 skb_shinfo(skb
)->frags
[k
].size
-= eat
;
1365 skb_shinfo(skb
)->nr_frags
= k
;
1368 skb
->data_len
-= delta
;
1370 return skb_tail_pointer(skb
);
1372 EXPORT_SYMBOL(__pskb_pull_tail
);
1375 * skb_copy_bits - copy bits from skb to kernel buffer
1377 * @offset: offset in source
1378 * @to: destination buffer
1379 * @len: number of bytes to copy
1381 * Copy the specified number of bytes from the source skb to the
1382 * destination buffer.
1385 * If its prototype is ever changed,
1386 * check arch/{*}/net/{*}.S files,
1387 * since it is called from BPF assembly code.
1389 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1391 int start
= skb_headlen(skb
);
1392 struct sk_buff
*frag_iter
;
1395 if (offset
> (int)skb
->len
- len
)
1399 if ((copy
= start
- offset
) > 0) {
1402 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
1403 if ((len
-= copy
) == 0)
1409 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1412 WARN_ON(start
> offset
+ len
);
1414 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1415 if ((copy
= end
- offset
) > 0) {
1421 vaddr
= kmap_skb_frag(&skb_shinfo(skb
)->frags
[i
]);
1423 vaddr
+ skb_shinfo(skb
)->frags
[i
].page_offset
+
1424 offset
- start
, copy
);
1425 kunmap_skb_frag(vaddr
);
1427 if ((len
-= copy
) == 0)
1435 skb_walk_frags(skb
, frag_iter
) {
1438 WARN_ON(start
> offset
+ len
);
1440 end
= start
+ frag_iter
->len
;
1441 if ((copy
= end
- offset
) > 0) {
1444 if (skb_copy_bits(frag_iter
, offset
- start
, to
, copy
))
1446 if ((len
-= copy
) == 0)
1460 EXPORT_SYMBOL(skb_copy_bits
);
1463 * Callback from splice_to_pipe(), if we need to release some pages
1464 * at the end of the spd in case we error'ed out in filling the pipe.
1466 static void sock_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
1468 put_page(spd
->pages
[i
]);
1471 static inline struct page
*linear_to_page(struct page
*page
, unsigned int *len
,
1472 unsigned int *offset
,
1473 struct sk_buff
*skb
, struct sock
*sk
)
1475 struct page
*p
= sk
->sk_sndmsg_page
;
1480 p
= sk
->sk_sndmsg_page
= alloc_pages(sk
->sk_allocation
, 0);
1484 off
= sk
->sk_sndmsg_off
= 0;
1485 /* hold one ref to this page until it's full */
1489 off
= sk
->sk_sndmsg_off
;
1490 mlen
= PAGE_SIZE
- off
;
1491 if (mlen
< 64 && mlen
< *len
) {
1496 *len
= min_t(unsigned int, *len
, mlen
);
1499 memcpy(page_address(p
) + off
, page_address(page
) + *offset
, *len
);
1500 sk
->sk_sndmsg_off
+= *len
;
1508 * Fill page/offset/length into spd, if it can hold more pages.
1510 static inline int spd_fill_page(struct splice_pipe_desc
*spd
,
1511 struct pipe_inode_info
*pipe
, struct page
*page
,
1512 unsigned int *len
, unsigned int offset
,
1513 struct sk_buff
*skb
, int linear
,
1516 if (unlikely(spd
->nr_pages
== pipe
->buffers
))
1520 page
= linear_to_page(page
, len
, &offset
, skb
, sk
);
1526 spd
->pages
[spd
->nr_pages
] = page
;
1527 spd
->partial
[spd
->nr_pages
].len
= *len
;
1528 spd
->partial
[spd
->nr_pages
].offset
= offset
;
1534 static inline void __segment_seek(struct page
**page
, unsigned int *poff
,
1535 unsigned int *plen
, unsigned int off
)
1540 n
= *poff
/ PAGE_SIZE
;
1542 *page
= nth_page(*page
, n
);
1544 *poff
= *poff
% PAGE_SIZE
;
1548 static inline int __splice_segment(struct page
*page
, unsigned int poff
,
1549 unsigned int plen
, unsigned int *off
,
1550 unsigned int *len
, struct sk_buff
*skb
,
1551 struct splice_pipe_desc
*spd
, int linear
,
1553 struct pipe_inode_info
*pipe
)
1558 /* skip this segment if already processed */
1564 /* ignore any bits we already processed */
1566 __segment_seek(&page
, &poff
, &plen
, *off
);
1571 unsigned int flen
= min(*len
, plen
);
1573 /* the linear region may spread across several pages */
1574 flen
= min_t(unsigned int, flen
, PAGE_SIZE
- poff
);
1576 if (spd_fill_page(spd
, pipe
, page
, &flen
, poff
, skb
, linear
, sk
))
1579 __segment_seek(&page
, &poff
, &plen
, flen
);
1582 } while (*len
&& plen
);
1588 * Map linear and fragment data from the skb to spd. It reports failure if the
1589 * pipe is full or if we already spliced the requested length.
1591 static int __skb_splice_bits(struct sk_buff
*skb
, struct pipe_inode_info
*pipe
,
1592 unsigned int *offset
, unsigned int *len
,
1593 struct splice_pipe_desc
*spd
, struct sock
*sk
)
1598 * map the linear part
1600 if (__splice_segment(virt_to_page(skb
->data
),
1601 (unsigned long) skb
->data
& (PAGE_SIZE
- 1),
1603 offset
, len
, skb
, spd
, 1, sk
, pipe
))
1607 * then map the fragments
1609 for (seg
= 0; seg
< skb_shinfo(skb
)->nr_frags
; seg
++) {
1610 const skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[seg
];
1612 if (__splice_segment(f
->page
, f
->page_offset
, f
->size
,
1613 offset
, len
, skb
, spd
, 0, sk
, pipe
))
1621 * Map data from the skb to a pipe. Should handle both the linear part,
1622 * the fragments, and the frag list. It does NOT handle frag lists within
1623 * the frag list, if such a thing exists. We'd probably need to recurse to
1624 * handle that cleanly.
1626 int skb_splice_bits(struct sk_buff
*skb
, unsigned int offset
,
1627 struct pipe_inode_info
*pipe
, unsigned int tlen
,
1630 struct partial_page partial
[PIPE_DEF_BUFFERS
];
1631 struct page
*pages
[PIPE_DEF_BUFFERS
];
1632 struct splice_pipe_desc spd
= {
1636 .ops
= &sock_pipe_buf_ops
,
1637 .spd_release
= sock_spd_release
,
1639 struct sk_buff
*frag_iter
;
1640 struct sock
*sk
= skb
->sk
;
1643 if (splice_grow_spd(pipe
, &spd
))
1647 * __skb_splice_bits() only fails if the output has no room left,
1648 * so no point in going over the frag_list for the error case.
1650 if (__skb_splice_bits(skb
, pipe
, &offset
, &tlen
, &spd
, sk
))
1656 * now see if we have a frag_list to map
1658 skb_walk_frags(skb
, frag_iter
) {
1661 if (__skb_splice_bits(frag_iter
, pipe
, &offset
, &tlen
, &spd
, sk
))
1668 * Drop the socket lock, otherwise we have reverse
1669 * locking dependencies between sk_lock and i_mutex
1670 * here as compared to sendfile(). We enter here
1671 * with the socket lock held, and splice_to_pipe() will
1672 * grab the pipe inode lock. For sendfile() emulation,
1673 * we call into ->sendpage() with the i_mutex lock held
1674 * and networking will grab the socket lock.
1677 ret
= splice_to_pipe(pipe
, &spd
);
1681 splice_shrink_spd(pipe
, &spd
);
1686 * skb_store_bits - store bits from kernel buffer to skb
1687 * @skb: destination buffer
1688 * @offset: offset in destination
1689 * @from: source buffer
1690 * @len: number of bytes to copy
1692 * Copy the specified number of bytes from the source buffer to the
1693 * destination skb. This function handles all the messy bits of
1694 * traversing fragment lists and such.
1697 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
1699 int start
= skb_headlen(skb
);
1700 struct sk_buff
*frag_iter
;
1703 if (offset
> (int)skb
->len
- len
)
1706 if ((copy
= start
- offset
) > 0) {
1709 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
1710 if ((len
-= copy
) == 0)
1716 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1717 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1720 WARN_ON(start
> offset
+ len
);
1722 end
= start
+ frag
->size
;
1723 if ((copy
= end
- offset
) > 0) {
1729 vaddr
= kmap_skb_frag(frag
);
1730 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1732 kunmap_skb_frag(vaddr
);
1734 if ((len
-= copy
) == 0)
1742 skb_walk_frags(skb
, frag_iter
) {
1745 WARN_ON(start
> offset
+ len
);
1747 end
= start
+ frag_iter
->len
;
1748 if ((copy
= end
- offset
) > 0) {
1751 if (skb_store_bits(frag_iter
, offset
- start
,
1754 if ((len
-= copy
) == 0)
1767 EXPORT_SYMBOL(skb_store_bits
);
1769 /* Checksum skb data. */
1771 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
1772 int len
, __wsum csum
)
1774 int start
= skb_headlen(skb
);
1775 int i
, copy
= start
- offset
;
1776 struct sk_buff
*frag_iter
;
1779 /* Checksum header. */
1783 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1784 if ((len
-= copy
) == 0)
1790 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1793 WARN_ON(start
> offset
+ len
);
1795 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1796 if ((copy
= end
- offset
) > 0) {
1799 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1803 vaddr
= kmap_skb_frag(frag
);
1804 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
1805 offset
- start
, copy
, 0);
1806 kunmap_skb_frag(vaddr
);
1807 csum
= csum_block_add(csum
, csum2
, pos
);
1816 skb_walk_frags(skb
, frag_iter
) {
1819 WARN_ON(start
> offset
+ len
);
1821 end
= start
+ frag_iter
->len
;
1822 if ((copy
= end
- offset
) > 0) {
1826 csum2
= skb_checksum(frag_iter
, offset
- start
,
1828 csum
= csum_block_add(csum
, csum2
, pos
);
1829 if ((len
-= copy
) == 0)
1840 EXPORT_SYMBOL(skb_checksum
);
1842 /* Both of above in one bottle. */
1844 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
1845 u8
*to
, int len
, __wsum csum
)
1847 int start
= skb_headlen(skb
);
1848 int i
, copy
= start
- offset
;
1849 struct sk_buff
*frag_iter
;
1856 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
1858 if ((len
-= copy
) == 0)
1865 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1868 WARN_ON(start
> offset
+ len
);
1870 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1871 if ((copy
= end
- offset
) > 0) {
1874 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1878 vaddr
= kmap_skb_frag(frag
);
1879 csum2
= csum_partial_copy_nocheck(vaddr
+
1883 kunmap_skb_frag(vaddr
);
1884 csum
= csum_block_add(csum
, csum2
, pos
);
1894 skb_walk_frags(skb
, frag_iter
) {
1898 WARN_ON(start
> offset
+ len
);
1900 end
= start
+ frag_iter
->len
;
1901 if ((copy
= end
- offset
) > 0) {
1904 csum2
= skb_copy_and_csum_bits(frag_iter
,
1907 csum
= csum_block_add(csum
, csum2
, pos
);
1908 if ((len
-= copy
) == 0)
1919 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
1921 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
1926 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1927 csstart
= skb_checksum_start_offset(skb
);
1929 csstart
= skb_headlen(skb
);
1931 BUG_ON(csstart
> skb_headlen(skb
));
1933 skb_copy_from_linear_data(skb
, to
, csstart
);
1936 if (csstart
!= skb
->len
)
1937 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
1938 skb
->len
- csstart
, 0);
1940 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1941 long csstuff
= csstart
+ skb
->csum_offset
;
1943 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
1946 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
1949 * skb_dequeue - remove from the head of the queue
1950 * @list: list to dequeue from
1952 * Remove the head of the list. The list lock is taken so the function
1953 * may be used safely with other locking list functions. The head item is
1954 * returned or %NULL if the list is empty.
1957 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
1959 unsigned long flags
;
1960 struct sk_buff
*result
;
1962 spin_lock_irqsave(&list
->lock
, flags
);
1963 result
= __skb_dequeue(list
);
1964 spin_unlock_irqrestore(&list
->lock
, flags
);
1967 EXPORT_SYMBOL(skb_dequeue
);
1970 * skb_dequeue_tail - remove from the tail of the queue
1971 * @list: list to dequeue from
1973 * Remove the tail of the list. The list lock is taken so the function
1974 * may be used safely with other locking list functions. The tail item is
1975 * returned or %NULL if the list is empty.
1977 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
1979 unsigned long flags
;
1980 struct sk_buff
*result
;
1982 spin_lock_irqsave(&list
->lock
, flags
);
1983 result
= __skb_dequeue_tail(list
);
1984 spin_unlock_irqrestore(&list
->lock
, flags
);
1987 EXPORT_SYMBOL(skb_dequeue_tail
);
1990 * skb_queue_purge - empty a list
1991 * @list: list to empty
1993 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1994 * the list and one reference dropped. This function takes the list
1995 * lock and is atomic with respect to other list locking functions.
1997 void skb_queue_purge(struct sk_buff_head
*list
)
1999 struct sk_buff
*skb
;
2000 while ((skb
= skb_dequeue(list
)) != NULL
)
2003 EXPORT_SYMBOL(skb_queue_purge
);
2006 * skb_queue_head - queue a buffer at the list head
2007 * @list: list to use
2008 * @newsk: buffer to queue
2010 * Queue a buffer at the start of the list. This function takes the
2011 * list lock and can be used safely with other locking &sk_buff functions
2014 * A buffer cannot be placed on two lists at the same time.
2016 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2018 unsigned long flags
;
2020 spin_lock_irqsave(&list
->lock
, flags
);
2021 __skb_queue_head(list
, newsk
);
2022 spin_unlock_irqrestore(&list
->lock
, flags
);
2024 EXPORT_SYMBOL(skb_queue_head
);
2027 * skb_queue_tail - queue a buffer at the list tail
2028 * @list: list to use
2029 * @newsk: buffer to queue
2031 * Queue a buffer at the tail of the list. This function takes the
2032 * list lock and can be used safely with other locking &sk_buff functions
2035 * A buffer cannot be placed on two lists at the same time.
2037 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2039 unsigned long flags
;
2041 spin_lock_irqsave(&list
->lock
, flags
);
2042 __skb_queue_tail(list
, newsk
);
2043 spin_unlock_irqrestore(&list
->lock
, flags
);
2045 EXPORT_SYMBOL(skb_queue_tail
);
2048 * skb_unlink - remove a buffer from a list
2049 * @skb: buffer to remove
2050 * @list: list to use
2052 * Remove a packet from a list. The list locks are taken and this
2053 * function is atomic with respect to other list locked calls
2055 * You must know what list the SKB is on.
2057 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
2059 unsigned long flags
;
2061 spin_lock_irqsave(&list
->lock
, flags
);
2062 __skb_unlink(skb
, list
);
2063 spin_unlock_irqrestore(&list
->lock
, flags
);
2065 EXPORT_SYMBOL(skb_unlink
);
2068 * skb_append - append a buffer
2069 * @old: buffer to insert after
2070 * @newsk: buffer to insert
2071 * @list: list to use
2073 * Place a packet after a given packet in a list. The list locks are taken
2074 * and this function is atomic with respect to other list locked calls.
2075 * A buffer cannot be placed on two lists at the same time.
2077 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2079 unsigned long flags
;
2081 spin_lock_irqsave(&list
->lock
, flags
);
2082 __skb_queue_after(list
, old
, newsk
);
2083 spin_unlock_irqrestore(&list
->lock
, flags
);
2085 EXPORT_SYMBOL(skb_append
);
2088 * skb_insert - insert a buffer
2089 * @old: buffer to insert before
2090 * @newsk: buffer to insert
2091 * @list: list to use
2093 * Place a packet before a given packet in a list. The list locks are
2094 * taken and this function is atomic with respect to other list locked
2097 * A buffer cannot be placed on two lists at the same time.
2099 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2101 unsigned long flags
;
2103 spin_lock_irqsave(&list
->lock
, flags
);
2104 __skb_insert(newsk
, old
->prev
, old
, list
);
2105 spin_unlock_irqrestore(&list
->lock
, flags
);
2107 EXPORT_SYMBOL(skb_insert
);
2109 static inline void skb_split_inside_header(struct sk_buff
*skb
,
2110 struct sk_buff
* skb1
,
2111 const u32 len
, const int pos
)
2115 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
2117 /* And move data appendix as is. */
2118 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
2119 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
2121 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
2122 skb_shinfo(skb
)->nr_frags
= 0;
2123 skb1
->data_len
= skb
->data_len
;
2124 skb1
->len
+= skb1
->data_len
;
2127 skb_set_tail_pointer(skb
, len
);
2130 static inline void skb_split_no_header(struct sk_buff
*skb
,
2131 struct sk_buff
* skb1
,
2132 const u32 len
, int pos
)
2135 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
2137 skb_shinfo(skb
)->nr_frags
= 0;
2138 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
2140 skb
->data_len
= len
- pos
;
2142 for (i
= 0; i
< nfrags
; i
++) {
2143 int size
= skb_shinfo(skb
)->frags
[i
].size
;
2145 if (pos
+ size
> len
) {
2146 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
2150 * We have two variants in this case:
2151 * 1. Move all the frag to the second
2152 * part, if it is possible. F.e.
2153 * this approach is mandatory for TUX,
2154 * where splitting is expensive.
2155 * 2. Split is accurately. We make this.
2157 get_page(skb_shinfo(skb
)->frags
[i
].page
);
2158 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
2159 skb_shinfo(skb1
)->frags
[0].size
-= len
- pos
;
2160 skb_shinfo(skb
)->frags
[i
].size
= len
- pos
;
2161 skb_shinfo(skb
)->nr_frags
++;
2165 skb_shinfo(skb
)->nr_frags
++;
2168 skb_shinfo(skb1
)->nr_frags
= k
;
2172 * skb_split - Split fragmented skb to two parts at length len.
2173 * @skb: the buffer to split
2174 * @skb1: the buffer to receive the second part
2175 * @len: new length for skb
2177 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
2179 int pos
= skb_headlen(skb
);
2181 if (len
< pos
) /* Split line is inside header. */
2182 skb_split_inside_header(skb
, skb1
, len
, pos
);
2183 else /* Second chunk has no header, nothing to copy. */
2184 skb_split_no_header(skb
, skb1
, len
, pos
);
2186 EXPORT_SYMBOL(skb_split
);
2188 /* Shifting from/to a cloned skb is a no-go.
2190 * Caller cannot keep skb_shinfo related pointers past calling here!
2192 static int skb_prepare_for_shift(struct sk_buff
*skb
)
2194 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2198 * skb_shift - Shifts paged data partially from skb to another
2199 * @tgt: buffer into which tail data gets added
2200 * @skb: buffer from which the paged data comes from
2201 * @shiftlen: shift up to this many bytes
2203 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2204 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2205 * It's up to caller to free skb if everything was shifted.
2207 * If @tgt runs out of frags, the whole operation is aborted.
2209 * Skb cannot include anything else but paged data while tgt is allowed
2210 * to have non-paged data as well.
2212 * TODO: full sized shift could be optimized but that would need
2213 * specialized skb free'er to handle frags without up-to-date nr_frags.
2215 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
2217 int from
, to
, merge
, todo
;
2218 struct skb_frag_struct
*fragfrom
, *fragto
;
2220 BUG_ON(shiftlen
> skb
->len
);
2221 BUG_ON(skb_headlen(skb
)); /* Would corrupt stream */
2225 to
= skb_shinfo(tgt
)->nr_frags
;
2226 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2228 /* Actual merge is delayed until the point when we know we can
2229 * commit all, so that we don't have to undo partial changes
2232 !skb_can_coalesce(tgt
, to
, fragfrom
->page
, fragfrom
->page_offset
)) {
2237 todo
-= fragfrom
->size
;
2239 if (skb_prepare_for_shift(skb
) ||
2240 skb_prepare_for_shift(tgt
))
2243 /* All previous frag pointers might be stale! */
2244 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2245 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2247 fragto
->size
+= shiftlen
;
2248 fragfrom
->size
-= shiftlen
;
2249 fragfrom
->page_offset
+= shiftlen
;
2257 /* Skip full, not-fitting skb to avoid expensive operations */
2258 if ((shiftlen
== skb
->len
) &&
2259 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
2262 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
2265 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
2266 if (to
== MAX_SKB_FRAGS
)
2269 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2270 fragto
= &skb_shinfo(tgt
)->frags
[to
];
2272 if (todo
>= fragfrom
->size
) {
2273 *fragto
= *fragfrom
;
2274 todo
-= fragfrom
->size
;
2279 get_page(fragfrom
->page
);
2280 fragto
->page
= fragfrom
->page
;
2281 fragto
->page_offset
= fragfrom
->page_offset
;
2282 fragto
->size
= todo
;
2284 fragfrom
->page_offset
+= todo
;
2285 fragfrom
->size
-= todo
;
2293 /* Ready to "commit" this state change to tgt */
2294 skb_shinfo(tgt
)->nr_frags
= to
;
2297 fragfrom
= &skb_shinfo(skb
)->frags
[0];
2298 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2300 fragto
->size
+= fragfrom
->size
;
2301 put_page(fragfrom
->page
);
2304 /* Reposition in the original skb */
2306 while (from
< skb_shinfo(skb
)->nr_frags
)
2307 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
2308 skb_shinfo(skb
)->nr_frags
= to
;
2310 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
2313 /* Most likely the tgt won't ever need its checksum anymore, skb on
2314 * the other hand might need it if it needs to be resent
2316 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
2317 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2319 /* Yak, is it really working this way? Some helper please? */
2320 skb
->len
-= shiftlen
;
2321 skb
->data_len
-= shiftlen
;
2322 skb
->truesize
-= shiftlen
;
2323 tgt
->len
+= shiftlen
;
2324 tgt
->data_len
+= shiftlen
;
2325 tgt
->truesize
+= shiftlen
;
2331 * skb_prepare_seq_read - Prepare a sequential read of skb data
2332 * @skb: the buffer to read
2333 * @from: lower offset of data to be read
2334 * @to: upper offset of data to be read
2335 * @st: state variable
2337 * Initializes the specified state variable. Must be called before
2338 * invoking skb_seq_read() for the first time.
2340 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
2341 unsigned int to
, struct skb_seq_state
*st
)
2343 st
->lower_offset
= from
;
2344 st
->upper_offset
= to
;
2345 st
->root_skb
= st
->cur_skb
= skb
;
2346 st
->frag_idx
= st
->stepped_offset
= 0;
2347 st
->frag_data
= NULL
;
2349 EXPORT_SYMBOL(skb_prepare_seq_read
);
2352 * skb_seq_read - Sequentially read skb data
2353 * @consumed: number of bytes consumed by the caller so far
2354 * @data: destination pointer for data to be returned
2355 * @st: state variable
2357 * Reads a block of skb data at &consumed relative to the
2358 * lower offset specified to skb_prepare_seq_read(). Assigns
2359 * the head of the data block to &data and returns the length
2360 * of the block or 0 if the end of the skb data or the upper
2361 * offset has been reached.
2363 * The caller is not required to consume all of the data
2364 * returned, i.e. &consumed is typically set to the number
2365 * of bytes already consumed and the next call to
2366 * skb_seq_read() will return the remaining part of the block.
2368 * Note 1: The size of each block of data returned can be arbitrary,
2369 * this limitation is the cost for zerocopy seqeuental
2370 * reads of potentially non linear data.
2372 * Note 2: Fragment lists within fragments are not implemented
2373 * at the moment, state->root_skb could be replaced with
2374 * a stack for this purpose.
2376 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
2377 struct skb_seq_state
*st
)
2379 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
2382 if (unlikely(abs_offset
>= st
->upper_offset
))
2386 block_limit
= skb_headlen(st
->cur_skb
) + st
->stepped_offset
;
2388 if (abs_offset
< block_limit
&& !st
->frag_data
) {
2389 *data
= st
->cur_skb
->data
+ (abs_offset
- st
->stepped_offset
);
2390 return block_limit
- abs_offset
;
2393 if (st
->frag_idx
== 0 && !st
->frag_data
)
2394 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
2396 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
2397 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
2398 block_limit
= frag
->size
+ st
->stepped_offset
;
2400 if (abs_offset
< block_limit
) {
2402 st
->frag_data
= kmap_skb_frag(frag
);
2404 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
2405 (abs_offset
- st
->stepped_offset
);
2407 return block_limit
- abs_offset
;
2410 if (st
->frag_data
) {
2411 kunmap_skb_frag(st
->frag_data
);
2412 st
->frag_data
= NULL
;
2416 st
->stepped_offset
+= frag
->size
;
2419 if (st
->frag_data
) {
2420 kunmap_skb_frag(st
->frag_data
);
2421 st
->frag_data
= NULL
;
2424 if (st
->root_skb
== st
->cur_skb
&& skb_has_frag_list(st
->root_skb
)) {
2425 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
2428 } else if (st
->cur_skb
->next
) {
2429 st
->cur_skb
= st
->cur_skb
->next
;
2436 EXPORT_SYMBOL(skb_seq_read
);
2439 * skb_abort_seq_read - Abort a sequential read of skb data
2440 * @st: state variable
2442 * Must be called if skb_seq_read() was not called until it
2445 void skb_abort_seq_read(struct skb_seq_state
*st
)
2448 kunmap_skb_frag(st
->frag_data
);
2450 EXPORT_SYMBOL(skb_abort_seq_read
);
2452 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2454 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
2455 struct ts_config
*conf
,
2456 struct ts_state
*state
)
2458 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
2461 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
2463 skb_abort_seq_read(TS_SKB_CB(state
));
2467 * skb_find_text - Find a text pattern in skb data
2468 * @skb: the buffer to look in
2469 * @from: search offset
2471 * @config: textsearch configuration
2472 * @state: uninitialized textsearch state variable
2474 * Finds a pattern in the skb data according to the specified
2475 * textsearch configuration. Use textsearch_next() to retrieve
2476 * subsequent occurrences of the pattern. Returns the offset
2477 * to the first occurrence or UINT_MAX if no match was found.
2479 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
2480 unsigned int to
, struct ts_config
*config
,
2481 struct ts_state
*state
)
2485 config
->get_next_block
= skb_ts_get_next_block
;
2486 config
->finish
= skb_ts_finish
;
2488 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
2490 ret
= textsearch_find(config
, state
);
2491 return (ret
<= to
- from
? ret
: UINT_MAX
);
2493 EXPORT_SYMBOL(skb_find_text
);
2496 * skb_append_datato_frags: - append the user data to a skb
2497 * @sk: sock structure
2498 * @skb: skb structure to be appened with user data.
2499 * @getfrag: call back function to be used for getting the user data
2500 * @from: pointer to user message iov
2501 * @length: length of the iov message
2503 * Description: This procedure append the user data in the fragment part
2504 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2506 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
2507 int (*getfrag
)(void *from
, char *to
, int offset
,
2508 int len
, int odd
, struct sk_buff
*skb
),
2509 void *from
, int length
)
2512 skb_frag_t
*frag
= NULL
;
2513 struct page
*page
= NULL
;
2519 /* Return error if we don't have space for new frag */
2520 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2521 if (frg_cnt
>= MAX_SKB_FRAGS
)
2524 /* allocate a new page for next frag */
2525 page
= alloc_pages(sk
->sk_allocation
, 0);
2527 /* If alloc_page fails just return failure and caller will
2528 * free previous allocated pages by doing kfree_skb()
2533 /* initialize the next frag */
2534 skb_fill_page_desc(skb
, frg_cnt
, page
, 0, 0);
2535 skb
->truesize
+= PAGE_SIZE
;
2536 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
2538 /* get the new initialized frag */
2539 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2540 frag
= &skb_shinfo(skb
)->frags
[frg_cnt
- 1];
2542 /* copy the user data to page */
2543 left
= PAGE_SIZE
- frag
->page_offset
;
2544 copy
= (length
> left
)? left
: length
;
2546 ret
= getfrag(from
, (page_address(frag
->page
) +
2547 frag
->page_offset
+ frag
->size
),
2548 offset
, copy
, 0, skb
);
2552 /* copy was successful so update the size parameters */
2555 skb
->data_len
+= copy
;
2559 } while (length
> 0);
2563 EXPORT_SYMBOL(skb_append_datato_frags
);
2566 * skb_pull_rcsum - pull skb and update receive checksum
2567 * @skb: buffer to update
2568 * @len: length of data pulled
2570 * This function performs an skb_pull on the packet and updates
2571 * the CHECKSUM_COMPLETE checksum. It should be used on
2572 * receive path processing instead of skb_pull unless you know
2573 * that the checksum difference is zero (e.g., a valid IP header)
2574 * or you are setting ip_summed to CHECKSUM_NONE.
2576 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
2578 BUG_ON(len
> skb
->len
);
2580 BUG_ON(skb
->len
< skb
->data_len
);
2581 skb_postpull_rcsum(skb
, skb
->data
, len
);
2582 return skb
->data
+= len
;
2584 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
2587 * skb_segment - Perform protocol segmentation on skb.
2588 * @skb: buffer to segment
2589 * @features: features for the output path (see dev->features)
2591 * This function performs segmentation on the given skb. It returns
2592 * a pointer to the first in a list of new skbs for the segments.
2593 * In case of error it returns ERR_PTR(err).
2595 struct sk_buff
*skb_segment(struct sk_buff
*skb
, u32 features
)
2597 struct sk_buff
*segs
= NULL
;
2598 struct sk_buff
*tail
= NULL
;
2599 struct sk_buff
*fskb
= skb_shinfo(skb
)->frag_list
;
2600 unsigned int mss
= skb_shinfo(skb
)->gso_size
;
2601 unsigned int doffset
= skb
->data
- skb_mac_header(skb
);
2602 unsigned int offset
= doffset
;
2603 unsigned int headroom
;
2605 int sg
= !!(features
& NETIF_F_SG
);
2606 int nfrags
= skb_shinfo(skb
)->nr_frags
;
2611 __skb_push(skb
, doffset
);
2612 headroom
= skb_headroom(skb
);
2613 pos
= skb_headlen(skb
);
2616 struct sk_buff
*nskb
;
2621 len
= skb
->len
- offset
;
2625 hsize
= skb_headlen(skb
) - offset
;
2628 if (hsize
> len
|| !sg
)
2631 if (!hsize
&& i
>= nfrags
) {
2632 BUG_ON(fskb
->len
!= len
);
2635 nskb
= skb_clone(fskb
, GFP_ATOMIC
);
2638 if (unlikely(!nskb
))
2641 hsize
= skb_end_pointer(nskb
) - nskb
->head
;
2642 if (skb_cow_head(nskb
, doffset
+ headroom
)) {
2647 nskb
->truesize
+= skb_end_pointer(nskb
) - nskb
->head
-
2649 skb_release_head_state(nskb
);
2650 __skb_push(nskb
, doffset
);
2652 nskb
= alloc_skb(hsize
+ doffset
+ headroom
,
2655 if (unlikely(!nskb
))
2658 skb_reserve(nskb
, headroom
);
2659 __skb_put(nskb
, doffset
);
2668 __copy_skb_header(nskb
, skb
);
2669 nskb
->mac_len
= skb
->mac_len
;
2671 /* nskb and skb might have different headroom */
2672 if (nskb
->ip_summed
== CHECKSUM_PARTIAL
)
2673 nskb
->csum_start
+= skb_headroom(nskb
) - headroom
;
2675 skb_reset_mac_header(nskb
);
2676 skb_set_network_header(nskb
, skb
->mac_len
);
2677 nskb
->transport_header
= (nskb
->network_header
+
2678 skb_network_header_len(skb
));
2679 skb_copy_from_linear_data(skb
, nskb
->data
, doffset
);
2681 if (fskb
!= skb_shinfo(skb
)->frag_list
)
2685 nskb
->ip_summed
= CHECKSUM_NONE
;
2686 nskb
->csum
= skb_copy_and_csum_bits(skb
, offset
,
2692 frag
= skb_shinfo(nskb
)->frags
;
2694 skb_copy_from_linear_data_offset(skb
, offset
,
2695 skb_put(nskb
, hsize
), hsize
);
2697 while (pos
< offset
+ len
&& i
< nfrags
) {
2698 *frag
= skb_shinfo(skb
)->frags
[i
];
2699 get_page(frag
->page
);
2703 frag
->page_offset
+= offset
- pos
;
2704 frag
->size
-= offset
- pos
;
2707 skb_shinfo(nskb
)->nr_frags
++;
2709 if (pos
+ size
<= offset
+ len
) {
2713 frag
->size
-= pos
+ size
- (offset
+ len
);
2720 if (pos
< offset
+ len
) {
2721 struct sk_buff
*fskb2
= fskb
;
2723 BUG_ON(pos
+ fskb
->len
!= offset
+ len
);
2729 fskb2
= skb_clone(fskb2
, GFP_ATOMIC
);
2735 SKB_FRAG_ASSERT(nskb
);
2736 skb_shinfo(nskb
)->frag_list
= fskb2
;
2740 nskb
->data_len
= len
- hsize
;
2741 nskb
->len
+= nskb
->data_len
;
2742 nskb
->truesize
+= nskb
->data_len
;
2743 } while ((offset
+= len
) < skb
->len
);
2748 while ((skb
= segs
)) {
2752 return ERR_PTR(err
);
2754 EXPORT_SYMBOL_GPL(skb_segment
);
2756 int skb_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2758 struct sk_buff
*p
= *head
;
2759 struct sk_buff
*nskb
;
2760 struct skb_shared_info
*skbinfo
= skb_shinfo(skb
);
2761 struct skb_shared_info
*pinfo
= skb_shinfo(p
);
2762 unsigned int headroom
;
2763 unsigned int len
= skb_gro_len(skb
);
2764 unsigned int offset
= skb_gro_offset(skb
);
2765 unsigned int headlen
= skb_headlen(skb
);
2767 if (p
->len
+ len
>= 65536)
2770 if (pinfo
->frag_list
)
2772 else if (headlen
<= offset
) {
2775 int i
= skbinfo
->nr_frags
;
2776 int nr_frags
= pinfo
->nr_frags
+ i
;
2780 if (nr_frags
> MAX_SKB_FRAGS
)
2783 pinfo
->nr_frags
= nr_frags
;
2784 skbinfo
->nr_frags
= 0;
2786 frag
= pinfo
->frags
+ nr_frags
;
2787 frag2
= skbinfo
->frags
+ i
;
2792 frag
->page_offset
+= offset
;
2793 frag
->size
-= offset
;
2795 skb
->truesize
-= skb
->data_len
;
2796 skb
->len
-= skb
->data_len
;
2799 NAPI_GRO_CB(skb
)->free
= 1;
2801 } else if (skb_gro_len(p
) != pinfo
->gso_size
)
2804 headroom
= skb_headroom(p
);
2805 nskb
= alloc_skb(headroom
+ skb_gro_offset(p
), GFP_ATOMIC
);
2806 if (unlikely(!nskb
))
2809 __copy_skb_header(nskb
, p
);
2810 nskb
->mac_len
= p
->mac_len
;
2812 skb_reserve(nskb
, headroom
);
2813 __skb_put(nskb
, skb_gro_offset(p
));
2815 skb_set_mac_header(nskb
, skb_mac_header(p
) - p
->data
);
2816 skb_set_network_header(nskb
, skb_network_offset(p
));
2817 skb_set_transport_header(nskb
, skb_transport_offset(p
));
2819 __skb_pull(p
, skb_gro_offset(p
));
2820 memcpy(skb_mac_header(nskb
), skb_mac_header(p
),
2821 p
->data
- skb_mac_header(p
));
2823 *NAPI_GRO_CB(nskb
) = *NAPI_GRO_CB(p
);
2824 skb_shinfo(nskb
)->frag_list
= p
;
2825 skb_shinfo(nskb
)->gso_size
= pinfo
->gso_size
;
2826 pinfo
->gso_size
= 0;
2827 skb_header_release(p
);
2830 nskb
->data_len
+= p
->len
;
2831 nskb
->truesize
+= p
->len
;
2832 nskb
->len
+= p
->len
;
2835 nskb
->next
= p
->next
;
2841 if (offset
> headlen
) {
2842 unsigned int eat
= offset
- headlen
;
2844 skbinfo
->frags
[0].page_offset
+= eat
;
2845 skbinfo
->frags
[0].size
-= eat
;
2846 skb
->data_len
-= eat
;
2851 __skb_pull(skb
, offset
);
2853 p
->prev
->next
= skb
;
2855 skb_header_release(skb
);
2858 NAPI_GRO_CB(p
)->count
++;
2863 NAPI_GRO_CB(skb
)->same_flow
= 1;
2866 EXPORT_SYMBOL_GPL(skb_gro_receive
);
2868 void __init
skb_init(void)
2870 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
2871 sizeof(struct sk_buff
),
2873 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2875 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
2876 (2*sizeof(struct sk_buff
)) +
2879 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2884 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2885 * @skb: Socket buffer containing the buffers to be mapped
2886 * @sg: The scatter-gather list to map into
2887 * @offset: The offset into the buffer's contents to start mapping
2888 * @len: Length of buffer space to be mapped
2890 * Fill the specified scatter-gather list with mappings/pointers into a
2891 * region of the buffer space attached to a socket buffer.
2894 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2896 int start
= skb_headlen(skb
);
2897 int i
, copy
= start
- offset
;
2898 struct sk_buff
*frag_iter
;
2904 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
2906 if ((len
-= copy
) == 0)
2911 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2914 WARN_ON(start
> offset
+ len
);
2916 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
2917 if ((copy
= end
- offset
) > 0) {
2918 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2922 sg_set_page(&sg
[elt
], frag
->page
, copy
,
2923 frag
->page_offset
+offset
-start
);
2932 skb_walk_frags(skb
, frag_iter
) {
2935 WARN_ON(start
> offset
+ len
);
2937 end
= start
+ frag_iter
->len
;
2938 if ((copy
= end
- offset
) > 0) {
2941 elt
+= __skb_to_sgvec(frag_iter
, sg
+elt
, offset
- start
,
2943 if ((len
-= copy
) == 0)
2953 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2955 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
);
2957 sg_mark_end(&sg
[nsg
- 1]);
2961 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
2964 * skb_cow_data - Check that a socket buffer's data buffers are writable
2965 * @skb: The socket buffer to check.
2966 * @tailbits: Amount of trailing space to be added
2967 * @trailer: Returned pointer to the skb where the @tailbits space begins
2969 * Make sure that the data buffers attached to a socket buffer are
2970 * writable. If they are not, private copies are made of the data buffers
2971 * and the socket buffer is set to use these instead.
2973 * If @tailbits is given, make sure that there is space to write @tailbits
2974 * bytes of data beyond current end of socket buffer. @trailer will be
2975 * set to point to the skb in which this space begins.
2977 * The number of scatterlist elements required to completely map the
2978 * COW'd and extended socket buffer will be returned.
2980 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
2984 struct sk_buff
*skb1
, **skb_p
;
2986 /* If skb is cloned or its head is paged, reallocate
2987 * head pulling out all the pages (pages are considered not writable
2988 * at the moment even if they are anonymous).
2990 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
2991 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
2994 /* Easy case. Most of packets will go this way. */
2995 if (!skb_has_frag_list(skb
)) {
2996 /* A little of trouble, not enough of space for trailer.
2997 * This should not happen, when stack is tuned to generate
2998 * good frames. OK, on miss we reallocate and reserve even more
2999 * space, 128 bytes is fair. */
3001 if (skb_tailroom(skb
) < tailbits
&&
3002 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
3010 /* Misery. We are in troubles, going to mincer fragments... */
3013 skb_p
= &skb_shinfo(skb
)->frag_list
;
3016 while ((skb1
= *skb_p
) != NULL
) {
3019 /* The fragment is partially pulled by someone,
3020 * this can happen on input. Copy it and everything
3023 if (skb_shared(skb1
))
3026 /* If the skb is the last, worry about trailer. */
3028 if (skb1
->next
== NULL
&& tailbits
) {
3029 if (skb_shinfo(skb1
)->nr_frags
||
3030 skb_has_frag_list(skb1
) ||
3031 skb_tailroom(skb1
) < tailbits
)
3032 ntail
= tailbits
+ 128;
3038 skb_shinfo(skb1
)->nr_frags
||
3039 skb_has_frag_list(skb1
)) {
3040 struct sk_buff
*skb2
;
3042 /* Fuck, we are miserable poor guys... */
3044 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
3046 skb2
= skb_copy_expand(skb1
,
3050 if (unlikely(skb2
== NULL
))
3054 skb_set_owner_w(skb2
, skb1
->sk
);
3056 /* Looking around. Are we still alive?
3057 * OK, link new skb, drop old one */
3059 skb2
->next
= skb1
->next
;
3066 skb_p
= &skb1
->next
;
3071 EXPORT_SYMBOL_GPL(skb_cow_data
);
3073 static void sock_rmem_free(struct sk_buff
*skb
)
3075 struct sock
*sk
= skb
->sk
;
3077 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
3081 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3083 int sock_queue_err_skb(struct sock
*sk
, struct sk_buff
*skb
)
3085 if (atomic_read(&sk
->sk_rmem_alloc
) + skb
->truesize
>=
3086 (unsigned)sk
->sk_rcvbuf
)
3091 skb
->destructor
= sock_rmem_free
;
3092 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
3094 /* before exiting rcu section, make sure dst is refcounted */
3097 skb_queue_tail(&sk
->sk_error_queue
, skb
);
3098 if (!sock_flag(sk
, SOCK_DEAD
))
3099 sk
->sk_data_ready(sk
, skb
->len
);
3102 EXPORT_SYMBOL(sock_queue_err_skb
);
3104 void skb_tstamp_tx(struct sk_buff
*orig_skb
,
3105 struct skb_shared_hwtstamps
*hwtstamps
)
3107 struct sock
*sk
= orig_skb
->sk
;
3108 struct sock_exterr_skb
*serr
;
3109 struct sk_buff
*skb
;
3115 skb
= skb_clone(orig_skb
, GFP_ATOMIC
);
3120 *skb_hwtstamps(skb
) =
3124 * no hardware time stamps available,
3125 * so keep the shared tx_flags and only
3126 * store software time stamp
3128 skb
->tstamp
= ktime_get_real();
3131 serr
= SKB_EXT_ERR(skb
);
3132 memset(serr
, 0, sizeof(*serr
));
3133 serr
->ee
.ee_errno
= ENOMSG
;
3134 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TIMESTAMPING
;
3136 err
= sock_queue_err_skb(sk
, skb
);
3141 EXPORT_SYMBOL_GPL(skb_tstamp_tx
);
3145 * skb_partial_csum_set - set up and verify partial csum values for packet
3146 * @skb: the skb to set
3147 * @start: the number of bytes after skb->data to start checksumming.
3148 * @off: the offset from start to place the checksum.
3150 * For untrusted partially-checksummed packets, we need to make sure the values
3151 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3153 * This function checks and sets those values and skb->ip_summed: if this
3154 * returns false you should drop the packet.
3156 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
3158 if (unlikely(start
> skb_headlen(skb
)) ||
3159 unlikely((int)start
+ off
> skb_headlen(skb
) - 2)) {
3160 if (net_ratelimit())
3162 "bad partial csum: csum=%u/%u len=%u\n",
3163 start
, off
, skb_headlen(skb
));
3166 skb
->ip_summed
= CHECKSUM_PARTIAL
;
3167 skb
->csum_start
= skb_headroom(skb
) + start
;
3168 skb
->csum_offset
= off
;
3171 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);
3173 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
3175 if (net_ratelimit())
3176 pr_warning("%s: received packets cannot be forwarded"
3177 " while LRO is enabled\n", skb
->dev
->name
);
3179 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);