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
);
332 if (skb_has_frag_list(skb
))
333 skb_drop_fraglist(skb
);
340 * Free an skbuff by memory without cleaning the state.
342 static void kfree_skbmem(struct sk_buff
*skb
)
344 struct sk_buff
*other
;
345 atomic_t
*fclone_ref
;
347 switch (skb
->fclone
) {
348 case SKB_FCLONE_UNAVAILABLE
:
349 kmem_cache_free(skbuff_head_cache
, skb
);
352 case SKB_FCLONE_ORIG
:
353 fclone_ref
= (atomic_t
*) (skb
+ 2);
354 if (atomic_dec_and_test(fclone_ref
))
355 kmem_cache_free(skbuff_fclone_cache
, skb
);
358 case SKB_FCLONE_CLONE
:
359 fclone_ref
= (atomic_t
*) (skb
+ 1);
362 /* The clone portion is available for
363 * fast-cloning again.
365 skb
->fclone
= SKB_FCLONE_UNAVAILABLE
;
367 if (atomic_dec_and_test(fclone_ref
))
368 kmem_cache_free(skbuff_fclone_cache
, other
);
373 static void skb_release_head_state(struct sk_buff
*skb
)
377 secpath_put(skb
->sp
);
379 if (skb
->destructor
) {
381 skb
->destructor(skb
);
383 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
384 nf_conntrack_put(skb
->nfct
);
386 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
387 nf_conntrack_put_reasm(skb
->nfct_reasm
);
389 #ifdef CONFIG_BRIDGE_NETFILTER
390 nf_bridge_put(skb
->nf_bridge
);
392 /* XXX: IS this still necessary? - JHS */
393 #ifdef CONFIG_NET_SCHED
395 #ifdef CONFIG_NET_CLS_ACT
401 /* Free everything but the sk_buff shell. */
402 static void skb_release_all(struct sk_buff
*skb
)
404 skb_release_head_state(skb
);
405 skb_release_data(skb
);
409 * __kfree_skb - private function
412 * Free an sk_buff. Release anything attached to the buffer.
413 * Clean the state. This is an internal helper function. Users should
414 * always call kfree_skb
417 void __kfree_skb(struct sk_buff
*skb
)
419 skb_release_all(skb
);
422 EXPORT_SYMBOL(__kfree_skb
);
425 * kfree_skb - free an sk_buff
426 * @skb: buffer to free
428 * Drop a reference to the buffer and free it if the usage count has
431 void kfree_skb(struct sk_buff
*skb
)
435 if (likely(atomic_read(&skb
->users
) == 1))
437 else if (likely(!atomic_dec_and_test(&skb
->users
)))
439 trace_kfree_skb(skb
, __builtin_return_address(0));
442 EXPORT_SYMBOL(kfree_skb
);
445 * consume_skb - free an skbuff
446 * @skb: buffer to free
448 * Drop a ref to the buffer and free it if the usage count has hit zero
449 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
450 * is being dropped after a failure and notes that
452 void consume_skb(struct sk_buff
*skb
)
456 if (likely(atomic_read(&skb
->users
) == 1))
458 else if (likely(!atomic_dec_and_test(&skb
->users
)))
460 trace_consume_skb(skb
);
463 EXPORT_SYMBOL(consume_skb
);
466 * skb_recycle_check - check if skb can be reused for receive
468 * @skb_size: minimum receive buffer size
470 * Checks that the skb passed in is not shared or cloned, and
471 * that it is linear and its head portion at least as large as
472 * skb_size so that it can be recycled as a receive buffer.
473 * If these conditions are met, this function does any necessary
474 * reference count dropping and cleans up the skbuff as if it
475 * just came from __alloc_skb().
477 bool skb_recycle_check(struct sk_buff
*skb
, int skb_size
)
479 struct skb_shared_info
*shinfo
;
484 if (skb_is_nonlinear(skb
) || skb
->fclone
!= SKB_FCLONE_UNAVAILABLE
)
487 skb_size
= SKB_DATA_ALIGN(skb_size
+ NET_SKB_PAD
);
488 if (skb_end_pointer(skb
) - skb
->head
< skb_size
)
491 if (skb_shared(skb
) || skb_cloned(skb
))
494 skb_release_head_state(skb
);
496 shinfo
= skb_shinfo(skb
);
497 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
498 atomic_set(&shinfo
->dataref
, 1);
500 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
501 skb
->data
= skb
->head
+ NET_SKB_PAD
;
502 skb_reset_tail_pointer(skb
);
506 EXPORT_SYMBOL(skb_recycle_check
);
508 static void __copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
510 new->tstamp
= old
->tstamp
;
512 new->transport_header
= old
->transport_header
;
513 new->network_header
= old
->network_header
;
514 new->mac_header
= old
->mac_header
;
515 skb_dst_copy(new, old
);
516 new->rxhash
= old
->rxhash
;
518 new->sp
= secpath_get(old
->sp
);
520 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
521 new->csum
= old
->csum
;
522 new->local_df
= old
->local_df
;
523 new->pkt_type
= old
->pkt_type
;
524 new->ip_summed
= old
->ip_summed
;
525 skb_copy_queue_mapping(new, old
);
526 new->priority
= old
->priority
;
527 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
528 new->ipvs_property
= old
->ipvs_property
;
530 new->protocol
= old
->protocol
;
531 new->mark
= old
->mark
;
532 new->skb_iif
= old
->skb_iif
;
534 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
535 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
536 new->nf_trace
= old
->nf_trace
;
538 #ifdef CONFIG_NET_SCHED
539 new->tc_index
= old
->tc_index
;
540 #ifdef CONFIG_NET_CLS_ACT
541 new->tc_verd
= old
->tc_verd
;
544 new->vlan_tci
= old
->vlan_tci
;
546 skb_copy_secmark(new, old
);
550 * You should not add any new code to this function. Add it to
551 * __copy_skb_header above instead.
553 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
555 #define C(x) n->x = skb->x
557 n
->next
= n
->prev
= NULL
;
559 __copy_skb_header(n
, skb
);
564 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
567 n
->destructor
= NULL
;
573 atomic_set(&n
->users
, 1);
575 atomic_inc(&(skb_shinfo(skb
)->dataref
));
583 * skb_morph - morph one skb into another
584 * @dst: the skb to receive the contents
585 * @src: the skb to supply the contents
587 * This is identical to skb_clone except that the target skb is
588 * supplied by the user.
590 * The target skb is returned upon exit.
592 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
594 skb_release_all(dst
);
595 return __skb_clone(dst
, src
);
597 EXPORT_SYMBOL_GPL(skb_morph
);
600 * skb_clone - duplicate an sk_buff
601 * @skb: buffer to clone
602 * @gfp_mask: allocation priority
604 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
605 * copies share the same packet data but not structure. The new
606 * buffer has a reference count of 1. If the allocation fails the
607 * function returns %NULL otherwise the new buffer is returned.
609 * If this function is called from an interrupt gfp_mask() must be
613 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
618 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
619 n
->fclone
== SKB_FCLONE_UNAVAILABLE
) {
620 atomic_t
*fclone_ref
= (atomic_t
*) (n
+ 1);
621 n
->fclone
= SKB_FCLONE_CLONE
;
622 atomic_inc(fclone_ref
);
624 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
628 kmemcheck_annotate_bitfield(n
, flags1
);
629 kmemcheck_annotate_bitfield(n
, flags2
);
630 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
633 return __skb_clone(n
, skb
);
635 EXPORT_SYMBOL(skb_clone
);
637 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
639 #ifndef NET_SKBUFF_DATA_USES_OFFSET
641 * Shift between the two data areas in bytes
643 unsigned long offset
= new->data
- old
->data
;
646 __copy_skb_header(new, old
);
648 #ifndef NET_SKBUFF_DATA_USES_OFFSET
649 /* {transport,network,mac}_header are relative to skb->head */
650 new->transport_header
+= offset
;
651 new->network_header
+= offset
;
652 if (skb_mac_header_was_set(new))
653 new->mac_header
+= offset
;
655 skb_shinfo(new)->gso_size
= skb_shinfo(old
)->gso_size
;
656 skb_shinfo(new)->gso_segs
= skb_shinfo(old
)->gso_segs
;
657 skb_shinfo(new)->gso_type
= skb_shinfo(old
)->gso_type
;
661 * skb_copy - create private copy of an sk_buff
662 * @skb: buffer to copy
663 * @gfp_mask: allocation priority
665 * Make a copy of both an &sk_buff and its data. This is used when the
666 * caller wishes to modify the data and needs a private copy of the
667 * data to alter. Returns %NULL on failure or the pointer to the buffer
668 * on success. The returned buffer has a reference count of 1.
670 * As by-product this function converts non-linear &sk_buff to linear
671 * one, so that &sk_buff becomes completely private and caller is allowed
672 * to modify all the data of returned buffer. This means that this
673 * function is not recommended for use in circumstances when only
674 * header is going to be modified. Use pskb_copy() instead.
677 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
679 int headerlen
= skb_headroom(skb
);
680 unsigned int size
= (skb_end_pointer(skb
) - skb
->head
) + skb
->data_len
;
681 struct sk_buff
*n
= alloc_skb(size
, gfp_mask
);
686 /* Set the data pointer */
687 skb_reserve(n
, headerlen
);
688 /* Set the tail pointer and length */
689 skb_put(n
, skb
->len
);
691 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
694 copy_skb_header(n
, skb
);
697 EXPORT_SYMBOL(skb_copy
);
700 * pskb_copy - create copy of an sk_buff with private head.
701 * @skb: buffer to copy
702 * @gfp_mask: allocation priority
704 * Make a copy of both an &sk_buff and part of its data, located
705 * in header. Fragmented data remain shared. This is used when
706 * the caller wishes to modify only header of &sk_buff and needs
707 * private copy of the header to alter. Returns %NULL on failure
708 * or the pointer to the buffer on success.
709 * The returned buffer has a reference count of 1.
712 struct sk_buff
*pskb_copy(struct sk_buff
*skb
, gfp_t gfp_mask
)
714 unsigned int size
= skb_end_pointer(skb
) - skb
->head
;
715 struct sk_buff
*n
= alloc_skb(size
, gfp_mask
);
720 /* Set the data pointer */
721 skb_reserve(n
, skb_headroom(skb
));
722 /* Set the tail pointer and length */
723 skb_put(n
, skb_headlen(skb
));
725 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
727 n
->truesize
+= skb
->data_len
;
728 n
->data_len
= skb
->data_len
;
731 if (skb_shinfo(skb
)->nr_frags
) {
734 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
735 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
736 get_page(skb_shinfo(n
)->frags
[i
].page
);
738 skb_shinfo(n
)->nr_frags
= i
;
741 if (skb_has_frag_list(skb
)) {
742 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
743 skb_clone_fraglist(n
);
746 copy_skb_header(n
, skb
);
750 EXPORT_SYMBOL(pskb_copy
);
753 * pskb_expand_head - reallocate header of &sk_buff
754 * @skb: buffer to reallocate
755 * @nhead: room to add at head
756 * @ntail: room to add at tail
757 * @gfp_mask: allocation priority
759 * Expands (or creates identical copy, if &nhead and &ntail are zero)
760 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
761 * reference count of 1. Returns zero in the case of success or error,
762 * if expansion failed. In the last case, &sk_buff is not changed.
764 * All the pointers pointing into skb header may change and must be
765 * reloaded after call to this function.
768 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
773 int size
= nhead
+ (skb_end_pointer(skb
) - skb
->head
) + ntail
;
782 size
= SKB_DATA_ALIGN(size
);
784 /* Check if we can avoid taking references on fragments if we own
785 * the last reference on skb->head. (see skb_release_data())
790 int delta
= skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1;
792 fastpath
= atomic_read(&skb_shinfo(skb
)->dataref
) == delta
;
796 size
+ sizeof(struct skb_shared_info
) <= ksize(skb
->head
)) {
797 memmove(skb
->head
+ size
, skb_shinfo(skb
),
798 offsetof(struct skb_shared_info
,
799 frags
[skb_shinfo(skb
)->nr_frags
]));
800 memmove(skb
->head
+ nhead
, skb
->head
,
801 skb_tail_pointer(skb
) - skb
->head
);
806 data
= kmalloc(size
+ sizeof(struct skb_shared_info
), gfp_mask
);
810 /* Copy only real data... and, alas, header. This should be
811 * optimized for the cases when header is void.
813 memcpy(data
+ nhead
, skb
->head
, skb_tail_pointer(skb
) - skb
->head
);
815 memcpy((struct skb_shared_info
*)(data
+ size
),
817 offsetof(struct skb_shared_info
, frags
[skb_shinfo(skb
)->nr_frags
]));
822 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
823 get_page(skb_shinfo(skb
)->frags
[i
].page
);
825 if (skb_has_frag_list(skb
))
826 skb_clone_fraglist(skb
);
828 skb_release_data(skb
);
830 off
= (data
+ nhead
) - skb
->head
;
835 #ifdef NET_SKBUFF_DATA_USES_OFFSET
839 skb
->end
= skb
->head
+ size
;
841 /* {transport,network,mac}_header and tail are relative to skb->head */
843 skb
->transport_header
+= off
;
844 skb
->network_header
+= off
;
845 if (skb_mac_header_was_set(skb
))
846 skb
->mac_header
+= off
;
847 /* Only adjust this if it actually is csum_start rather than csum */
848 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
849 skb
->csum_start
+= nhead
;
853 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
859 EXPORT_SYMBOL(pskb_expand_head
);
861 /* Make private copy of skb with writable head and some headroom */
863 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
865 struct sk_buff
*skb2
;
866 int delta
= headroom
- skb_headroom(skb
);
869 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
871 skb2
= skb_clone(skb
, GFP_ATOMIC
);
872 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
880 EXPORT_SYMBOL(skb_realloc_headroom
);
883 * skb_copy_expand - copy and expand sk_buff
884 * @skb: buffer to copy
885 * @newheadroom: new free bytes at head
886 * @newtailroom: new free bytes at tail
887 * @gfp_mask: allocation priority
889 * Make a copy of both an &sk_buff and its data and while doing so
890 * allocate additional space.
892 * This is used when the caller wishes to modify the data and needs a
893 * private copy of the data to alter as well as more space for new fields.
894 * Returns %NULL on failure or the pointer to the buffer
895 * on success. The returned buffer has a reference count of 1.
897 * You must pass %GFP_ATOMIC as the allocation priority if this function
898 * is called from an interrupt.
900 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
901 int newheadroom
, int newtailroom
,
905 * Allocate the copy buffer
907 struct sk_buff
*n
= alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
909 int oldheadroom
= skb_headroom(skb
);
910 int head_copy_len
, head_copy_off
;
916 skb_reserve(n
, newheadroom
);
918 /* Set the tail pointer and length */
919 skb_put(n
, skb
->len
);
921 head_copy_len
= oldheadroom
;
923 if (newheadroom
<= head_copy_len
)
924 head_copy_len
= newheadroom
;
926 head_copy_off
= newheadroom
- head_copy_len
;
928 /* Copy the linear header and data. */
929 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
930 skb
->len
+ head_copy_len
))
933 copy_skb_header(n
, skb
);
935 off
= newheadroom
- oldheadroom
;
936 if (n
->ip_summed
== CHECKSUM_PARTIAL
)
937 n
->csum_start
+= off
;
938 #ifdef NET_SKBUFF_DATA_USES_OFFSET
939 n
->transport_header
+= off
;
940 n
->network_header
+= off
;
941 if (skb_mac_header_was_set(skb
))
942 n
->mac_header
+= off
;
947 EXPORT_SYMBOL(skb_copy_expand
);
950 * skb_pad - zero pad the tail of an skb
951 * @skb: buffer to pad
954 * Ensure that a buffer is followed by a padding area that is zero
955 * filled. Used by network drivers which may DMA or transfer data
956 * beyond the buffer end onto the wire.
958 * May return error in out of memory cases. The skb is freed on error.
961 int skb_pad(struct sk_buff
*skb
, int pad
)
966 /* If the skbuff is non linear tailroom is always zero.. */
967 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
968 memset(skb
->data
+skb
->len
, 0, pad
);
972 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
973 if (likely(skb_cloned(skb
) || ntail
> 0)) {
974 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
979 /* FIXME: The use of this function with non-linear skb's really needs
982 err
= skb_linearize(skb
);
986 memset(skb
->data
+ skb
->len
, 0, pad
);
993 EXPORT_SYMBOL(skb_pad
);
996 * skb_put - add data to a buffer
997 * @skb: buffer to use
998 * @len: amount of data to add
1000 * This function extends the used data area of the buffer. If this would
1001 * exceed the total buffer size the kernel will panic. A pointer to the
1002 * first byte of the extra data is returned.
1004 unsigned char *skb_put(struct sk_buff
*skb
, unsigned int len
)
1006 unsigned char *tmp
= skb_tail_pointer(skb
);
1007 SKB_LINEAR_ASSERT(skb
);
1010 if (unlikely(skb
->tail
> skb
->end
))
1011 skb_over_panic(skb
, len
, __builtin_return_address(0));
1014 EXPORT_SYMBOL(skb_put
);
1017 * skb_push - add data to the start of a buffer
1018 * @skb: buffer to use
1019 * @len: amount of data to add
1021 * This function extends the used data area of the buffer at the buffer
1022 * start. If this would exceed the total buffer headroom the kernel will
1023 * panic. A pointer to the first byte of the extra data is returned.
1025 unsigned char *skb_push(struct sk_buff
*skb
, unsigned int len
)
1029 if (unlikely(skb
->data
<skb
->head
))
1030 skb_under_panic(skb
, len
, __builtin_return_address(0));
1033 EXPORT_SYMBOL(skb_push
);
1036 * skb_pull - remove data from the start of a buffer
1037 * @skb: buffer to use
1038 * @len: amount of data to remove
1040 * This function removes data from the start of a buffer, returning
1041 * the memory to the headroom. A pointer to the next data in the buffer
1042 * is returned. Once the data has been pulled future pushes will overwrite
1045 unsigned char *skb_pull(struct sk_buff
*skb
, unsigned int len
)
1047 return skb_pull_inline(skb
, len
);
1049 EXPORT_SYMBOL(skb_pull
);
1052 * skb_trim - remove end from a buffer
1053 * @skb: buffer to alter
1056 * Cut the length of a buffer down by removing data from the tail. If
1057 * the buffer is already under the length specified it is not modified.
1058 * The skb must be linear.
1060 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1063 __skb_trim(skb
, len
);
1065 EXPORT_SYMBOL(skb_trim
);
1067 /* Trims skb to length len. It can change skb pointers.
1070 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1072 struct sk_buff
**fragp
;
1073 struct sk_buff
*frag
;
1074 int offset
= skb_headlen(skb
);
1075 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1079 if (skb_cloned(skb
) &&
1080 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1087 for (; i
< nfrags
; i
++) {
1088 int end
= offset
+ skb_shinfo(skb
)->frags
[i
].size
;
1095 skb_shinfo(skb
)->frags
[i
++].size
= len
- offset
;
1098 skb_shinfo(skb
)->nr_frags
= i
;
1100 for (; i
< nfrags
; i
++)
1101 put_page(skb_shinfo(skb
)->frags
[i
].page
);
1103 if (skb_has_frag_list(skb
))
1104 skb_drop_fraglist(skb
);
1108 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1109 fragp
= &frag
->next
) {
1110 int end
= offset
+ frag
->len
;
1112 if (skb_shared(frag
)) {
1113 struct sk_buff
*nfrag
;
1115 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1116 if (unlikely(!nfrag
))
1119 nfrag
->next
= frag
->next
;
1131 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1135 skb_drop_list(&frag
->next
);
1140 if (len
> skb_headlen(skb
)) {
1141 skb
->data_len
-= skb
->len
- len
;
1146 skb_set_tail_pointer(skb
, len
);
1151 EXPORT_SYMBOL(___pskb_trim
);
1154 * __pskb_pull_tail - advance tail of skb header
1155 * @skb: buffer to reallocate
1156 * @delta: number of bytes to advance tail
1158 * The function makes a sense only on a fragmented &sk_buff,
1159 * it expands header moving its tail forward and copying necessary
1160 * data from fragmented part.
1162 * &sk_buff MUST have reference count of 1.
1164 * Returns %NULL (and &sk_buff does not change) if pull failed
1165 * or value of new tail of skb in the case of success.
1167 * All the pointers pointing into skb header may change and must be
1168 * reloaded after call to this function.
1171 /* Moves tail of skb head forward, copying data from fragmented part,
1172 * when it is necessary.
1173 * 1. It may fail due to malloc failure.
1174 * 2. It may change skb pointers.
1176 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1178 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1180 /* If skb has not enough free space at tail, get new one
1181 * plus 128 bytes for future expansions. If we have enough
1182 * room at tail, reallocate without expansion only if skb is cloned.
1184 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1186 if (eat
> 0 || skb_cloned(skb
)) {
1187 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1192 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
1195 /* Optimization: no fragments, no reasons to preestimate
1196 * size of pulled pages. Superb.
1198 if (!skb_has_frag_list(skb
))
1201 /* Estimate size of pulled pages. */
1203 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1204 if (skb_shinfo(skb
)->frags
[i
].size
>= eat
)
1206 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1209 /* If we need update frag list, we are in troubles.
1210 * Certainly, it possible to add an offset to skb data,
1211 * but taking into account that pulling is expected to
1212 * be very rare operation, it is worth to fight against
1213 * further bloating skb head and crucify ourselves here instead.
1214 * Pure masohism, indeed. 8)8)
1217 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1218 struct sk_buff
*clone
= NULL
;
1219 struct sk_buff
*insp
= NULL
;
1224 if (list
->len
<= eat
) {
1225 /* Eaten as whole. */
1230 /* Eaten partially. */
1232 if (skb_shared(list
)) {
1233 /* Sucks! We need to fork list. :-( */
1234 clone
= skb_clone(list
, GFP_ATOMIC
);
1240 /* This may be pulled without
1244 if (!pskb_pull(list
, eat
)) {
1252 /* Free pulled out fragments. */
1253 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1254 skb_shinfo(skb
)->frag_list
= list
->next
;
1257 /* And insert new clone at head. */
1260 skb_shinfo(skb
)->frag_list
= clone
;
1263 /* Success! Now we may commit changes to skb data. */
1268 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1269 if (skb_shinfo(skb
)->frags
[i
].size
<= eat
) {
1270 put_page(skb_shinfo(skb
)->frags
[i
].page
);
1271 eat
-= skb_shinfo(skb
)->frags
[i
].size
;
1273 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1275 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1276 skb_shinfo(skb
)->frags
[k
].size
-= eat
;
1282 skb_shinfo(skb
)->nr_frags
= k
;
1285 skb
->data_len
-= delta
;
1287 return skb_tail_pointer(skb
);
1289 EXPORT_SYMBOL(__pskb_pull_tail
);
1291 /* Copy some data bits from skb to kernel buffer. */
1293 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1295 int start
= skb_headlen(skb
);
1296 struct sk_buff
*frag_iter
;
1299 if (offset
> (int)skb
->len
- len
)
1303 if ((copy
= start
- offset
) > 0) {
1306 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
1307 if ((len
-= copy
) == 0)
1313 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1316 WARN_ON(start
> offset
+ len
);
1318 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1319 if ((copy
= end
- offset
) > 0) {
1325 vaddr
= kmap_skb_frag(&skb_shinfo(skb
)->frags
[i
]);
1327 vaddr
+ skb_shinfo(skb
)->frags
[i
].page_offset
+
1328 offset
- start
, copy
);
1329 kunmap_skb_frag(vaddr
);
1331 if ((len
-= copy
) == 0)
1339 skb_walk_frags(skb
, frag_iter
) {
1342 WARN_ON(start
> offset
+ len
);
1344 end
= start
+ frag_iter
->len
;
1345 if ((copy
= end
- offset
) > 0) {
1348 if (skb_copy_bits(frag_iter
, offset
- start
, to
, copy
))
1350 if ((len
-= copy
) == 0)
1363 EXPORT_SYMBOL(skb_copy_bits
);
1366 * Callback from splice_to_pipe(), if we need to release some pages
1367 * at the end of the spd in case we error'ed out in filling the pipe.
1369 static void sock_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
1371 put_page(spd
->pages
[i
]);
1374 static inline struct page
*linear_to_page(struct page
*page
, unsigned int *len
,
1375 unsigned int *offset
,
1376 struct sk_buff
*skb
, struct sock
*sk
)
1378 struct page
*p
= sk
->sk_sndmsg_page
;
1383 p
= sk
->sk_sndmsg_page
= alloc_pages(sk
->sk_allocation
, 0);
1387 off
= sk
->sk_sndmsg_off
= 0;
1388 /* hold one ref to this page until it's full */
1392 off
= sk
->sk_sndmsg_off
;
1393 mlen
= PAGE_SIZE
- off
;
1394 if (mlen
< 64 && mlen
< *len
) {
1399 *len
= min_t(unsigned int, *len
, mlen
);
1402 memcpy(page_address(p
) + off
, page_address(page
) + *offset
, *len
);
1403 sk
->sk_sndmsg_off
+= *len
;
1411 * Fill page/offset/length into spd, if it can hold more pages.
1413 static inline int spd_fill_page(struct splice_pipe_desc
*spd
,
1414 struct pipe_inode_info
*pipe
, struct page
*page
,
1415 unsigned int *len
, unsigned int offset
,
1416 struct sk_buff
*skb
, int linear
,
1419 if (unlikely(spd
->nr_pages
== pipe
->buffers
))
1423 page
= linear_to_page(page
, len
, &offset
, skb
, sk
);
1429 spd
->pages
[spd
->nr_pages
] = page
;
1430 spd
->partial
[spd
->nr_pages
].len
= *len
;
1431 spd
->partial
[spd
->nr_pages
].offset
= offset
;
1437 static inline void __segment_seek(struct page
**page
, unsigned int *poff
,
1438 unsigned int *plen
, unsigned int off
)
1443 n
= *poff
/ PAGE_SIZE
;
1445 *page
= nth_page(*page
, n
);
1447 *poff
= *poff
% PAGE_SIZE
;
1451 static inline int __splice_segment(struct page
*page
, unsigned int poff
,
1452 unsigned int plen
, unsigned int *off
,
1453 unsigned int *len
, struct sk_buff
*skb
,
1454 struct splice_pipe_desc
*spd
, int linear
,
1456 struct pipe_inode_info
*pipe
)
1461 /* skip this segment if already processed */
1467 /* ignore any bits we already processed */
1469 __segment_seek(&page
, &poff
, &plen
, *off
);
1474 unsigned int flen
= min(*len
, plen
);
1476 /* the linear region may spread across several pages */
1477 flen
= min_t(unsigned int, flen
, PAGE_SIZE
- poff
);
1479 if (spd_fill_page(spd
, pipe
, page
, &flen
, poff
, skb
, linear
, sk
))
1482 __segment_seek(&page
, &poff
, &plen
, flen
);
1485 } while (*len
&& plen
);
1491 * Map linear and fragment data from the skb to spd. It reports failure if the
1492 * pipe is full or if we already spliced the requested length.
1494 static int __skb_splice_bits(struct sk_buff
*skb
, struct pipe_inode_info
*pipe
,
1495 unsigned int *offset
, unsigned int *len
,
1496 struct splice_pipe_desc
*spd
, struct sock
*sk
)
1501 * map the linear part
1503 if (__splice_segment(virt_to_page(skb
->data
),
1504 (unsigned long) skb
->data
& (PAGE_SIZE
- 1),
1506 offset
, len
, skb
, spd
, 1, sk
, pipe
))
1510 * then map the fragments
1512 for (seg
= 0; seg
< skb_shinfo(skb
)->nr_frags
; seg
++) {
1513 const skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[seg
];
1515 if (__splice_segment(f
->page
, f
->page_offset
, f
->size
,
1516 offset
, len
, skb
, spd
, 0, sk
, pipe
))
1524 * Map data from the skb to a pipe. Should handle both the linear part,
1525 * the fragments, and the frag list. It does NOT handle frag lists within
1526 * the frag list, if such a thing exists. We'd probably need to recurse to
1527 * handle that cleanly.
1529 int skb_splice_bits(struct sk_buff
*skb
, unsigned int offset
,
1530 struct pipe_inode_info
*pipe
, unsigned int tlen
,
1533 struct partial_page partial
[PIPE_DEF_BUFFERS
];
1534 struct page
*pages
[PIPE_DEF_BUFFERS
];
1535 struct splice_pipe_desc spd
= {
1539 .ops
= &sock_pipe_buf_ops
,
1540 .spd_release
= sock_spd_release
,
1542 struct sk_buff
*frag_iter
;
1543 struct sock
*sk
= skb
->sk
;
1546 if (splice_grow_spd(pipe
, &spd
))
1550 * __skb_splice_bits() only fails if the output has no room left,
1551 * so no point in going over the frag_list for the error case.
1553 if (__skb_splice_bits(skb
, pipe
, &offset
, &tlen
, &spd
, sk
))
1559 * now see if we have a frag_list to map
1561 skb_walk_frags(skb
, frag_iter
) {
1564 if (__skb_splice_bits(frag_iter
, pipe
, &offset
, &tlen
, &spd
, sk
))
1571 * Drop the socket lock, otherwise we have reverse
1572 * locking dependencies between sk_lock and i_mutex
1573 * here as compared to sendfile(). We enter here
1574 * with the socket lock held, and splice_to_pipe() will
1575 * grab the pipe inode lock. For sendfile() emulation,
1576 * we call into ->sendpage() with the i_mutex lock held
1577 * and networking will grab the socket lock.
1580 ret
= splice_to_pipe(pipe
, &spd
);
1584 splice_shrink_spd(pipe
, &spd
);
1589 * skb_store_bits - store bits from kernel buffer to skb
1590 * @skb: destination buffer
1591 * @offset: offset in destination
1592 * @from: source buffer
1593 * @len: number of bytes to copy
1595 * Copy the specified number of bytes from the source buffer to the
1596 * destination skb. This function handles all the messy bits of
1597 * traversing fragment lists and such.
1600 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
1602 int start
= skb_headlen(skb
);
1603 struct sk_buff
*frag_iter
;
1606 if (offset
> (int)skb
->len
- len
)
1609 if ((copy
= start
- offset
) > 0) {
1612 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
1613 if ((len
-= copy
) == 0)
1619 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1620 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1623 WARN_ON(start
> offset
+ len
);
1625 end
= start
+ frag
->size
;
1626 if ((copy
= end
- offset
) > 0) {
1632 vaddr
= kmap_skb_frag(frag
);
1633 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1635 kunmap_skb_frag(vaddr
);
1637 if ((len
-= copy
) == 0)
1645 skb_walk_frags(skb
, frag_iter
) {
1648 WARN_ON(start
> offset
+ len
);
1650 end
= start
+ frag_iter
->len
;
1651 if ((copy
= end
- offset
) > 0) {
1654 if (skb_store_bits(frag_iter
, offset
- start
,
1657 if ((len
-= copy
) == 0)
1670 EXPORT_SYMBOL(skb_store_bits
);
1672 /* Checksum skb data. */
1674 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
1675 int len
, __wsum csum
)
1677 int start
= skb_headlen(skb
);
1678 int i
, copy
= start
- offset
;
1679 struct sk_buff
*frag_iter
;
1682 /* Checksum header. */
1686 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1687 if ((len
-= copy
) == 0)
1693 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1696 WARN_ON(start
> offset
+ len
);
1698 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1699 if ((copy
= end
- offset
) > 0) {
1702 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1706 vaddr
= kmap_skb_frag(frag
);
1707 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
1708 offset
- start
, copy
, 0);
1709 kunmap_skb_frag(vaddr
);
1710 csum
= csum_block_add(csum
, csum2
, pos
);
1719 skb_walk_frags(skb
, frag_iter
) {
1722 WARN_ON(start
> offset
+ len
);
1724 end
= start
+ frag_iter
->len
;
1725 if ((copy
= end
- offset
) > 0) {
1729 csum2
= skb_checksum(frag_iter
, offset
- start
,
1731 csum
= csum_block_add(csum
, csum2
, pos
);
1732 if ((len
-= copy
) == 0)
1743 EXPORT_SYMBOL(skb_checksum
);
1745 /* Both of above in one bottle. */
1747 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
1748 u8
*to
, int len
, __wsum csum
)
1750 int start
= skb_headlen(skb
);
1751 int i
, copy
= start
- offset
;
1752 struct sk_buff
*frag_iter
;
1759 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
1761 if ((len
-= copy
) == 0)
1768 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1771 WARN_ON(start
> offset
+ len
);
1773 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
1774 if ((copy
= end
- offset
) > 0) {
1777 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1781 vaddr
= kmap_skb_frag(frag
);
1782 csum2
= csum_partial_copy_nocheck(vaddr
+
1786 kunmap_skb_frag(vaddr
);
1787 csum
= csum_block_add(csum
, csum2
, pos
);
1797 skb_walk_frags(skb
, frag_iter
) {
1801 WARN_ON(start
> offset
+ len
);
1803 end
= start
+ frag_iter
->len
;
1804 if ((copy
= end
- offset
) > 0) {
1807 csum2
= skb_copy_and_csum_bits(frag_iter
,
1810 csum
= csum_block_add(csum
, csum2
, pos
);
1811 if ((len
-= copy
) == 0)
1822 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
1824 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
1829 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1830 csstart
= skb_checksum_start_offset(skb
);
1832 csstart
= skb_headlen(skb
);
1834 BUG_ON(csstart
> skb_headlen(skb
));
1836 skb_copy_from_linear_data(skb
, to
, csstart
);
1839 if (csstart
!= skb
->len
)
1840 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
1841 skb
->len
- csstart
, 0);
1843 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1844 long csstuff
= csstart
+ skb
->csum_offset
;
1846 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
1849 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
1852 * skb_dequeue - remove from the head of the queue
1853 * @list: list to dequeue from
1855 * Remove the head of the list. The list lock is taken so the function
1856 * may be used safely with other locking list functions. The head item is
1857 * returned or %NULL if the list is empty.
1860 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
1862 unsigned long flags
;
1863 struct sk_buff
*result
;
1865 spin_lock_irqsave(&list
->lock
, flags
);
1866 result
= __skb_dequeue(list
);
1867 spin_unlock_irqrestore(&list
->lock
, flags
);
1870 EXPORT_SYMBOL(skb_dequeue
);
1873 * skb_dequeue_tail - remove from the tail of the queue
1874 * @list: list to dequeue from
1876 * Remove the tail of the list. The list lock is taken so the function
1877 * may be used safely with other locking list functions. The tail item is
1878 * returned or %NULL if the list is empty.
1880 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
1882 unsigned long flags
;
1883 struct sk_buff
*result
;
1885 spin_lock_irqsave(&list
->lock
, flags
);
1886 result
= __skb_dequeue_tail(list
);
1887 spin_unlock_irqrestore(&list
->lock
, flags
);
1890 EXPORT_SYMBOL(skb_dequeue_tail
);
1893 * skb_queue_purge - empty a list
1894 * @list: list to empty
1896 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1897 * the list and one reference dropped. This function takes the list
1898 * lock and is atomic with respect to other list locking functions.
1900 void skb_queue_purge(struct sk_buff_head
*list
)
1902 struct sk_buff
*skb
;
1903 while ((skb
= skb_dequeue(list
)) != NULL
)
1906 EXPORT_SYMBOL(skb_queue_purge
);
1909 * skb_queue_head - queue a buffer at the list head
1910 * @list: list to use
1911 * @newsk: buffer to queue
1913 * Queue a buffer at the start of the list. This function takes the
1914 * list lock and can be used safely with other locking &sk_buff functions
1917 * A buffer cannot be placed on two lists at the same time.
1919 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1921 unsigned long flags
;
1923 spin_lock_irqsave(&list
->lock
, flags
);
1924 __skb_queue_head(list
, newsk
);
1925 spin_unlock_irqrestore(&list
->lock
, flags
);
1927 EXPORT_SYMBOL(skb_queue_head
);
1930 * skb_queue_tail - queue a buffer at the list tail
1931 * @list: list to use
1932 * @newsk: buffer to queue
1934 * Queue a buffer at the tail of the list. This function takes the
1935 * list lock and can be used safely with other locking &sk_buff functions
1938 * A buffer cannot be placed on two lists at the same time.
1940 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
1942 unsigned long flags
;
1944 spin_lock_irqsave(&list
->lock
, flags
);
1945 __skb_queue_tail(list
, newsk
);
1946 spin_unlock_irqrestore(&list
->lock
, flags
);
1948 EXPORT_SYMBOL(skb_queue_tail
);
1951 * skb_unlink - remove a buffer from a list
1952 * @skb: buffer to remove
1953 * @list: list to use
1955 * Remove a packet from a list. The list locks are taken and this
1956 * function is atomic with respect to other list locked calls
1958 * You must know what list the SKB is on.
1960 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
1962 unsigned long flags
;
1964 spin_lock_irqsave(&list
->lock
, flags
);
1965 __skb_unlink(skb
, list
);
1966 spin_unlock_irqrestore(&list
->lock
, flags
);
1968 EXPORT_SYMBOL(skb_unlink
);
1971 * skb_append - append a buffer
1972 * @old: buffer to insert after
1973 * @newsk: buffer to insert
1974 * @list: list to use
1976 * Place a packet after a given packet in a list. The list locks are taken
1977 * and this function is atomic with respect to other list locked calls.
1978 * A buffer cannot be placed on two lists at the same time.
1980 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
1982 unsigned long flags
;
1984 spin_lock_irqsave(&list
->lock
, flags
);
1985 __skb_queue_after(list
, old
, newsk
);
1986 spin_unlock_irqrestore(&list
->lock
, flags
);
1988 EXPORT_SYMBOL(skb_append
);
1991 * skb_insert - insert a buffer
1992 * @old: buffer to insert before
1993 * @newsk: buffer to insert
1994 * @list: list to use
1996 * Place a packet before a given packet in a list. The list locks are
1997 * taken and this function is atomic with respect to other list locked
2000 * A buffer cannot be placed on two lists at the same time.
2002 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2004 unsigned long flags
;
2006 spin_lock_irqsave(&list
->lock
, flags
);
2007 __skb_insert(newsk
, old
->prev
, old
, list
);
2008 spin_unlock_irqrestore(&list
->lock
, flags
);
2010 EXPORT_SYMBOL(skb_insert
);
2012 static inline void skb_split_inside_header(struct sk_buff
*skb
,
2013 struct sk_buff
* skb1
,
2014 const u32 len
, const int pos
)
2018 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
2020 /* And move data appendix as is. */
2021 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
2022 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
2024 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
2025 skb_shinfo(skb
)->nr_frags
= 0;
2026 skb1
->data_len
= skb
->data_len
;
2027 skb1
->len
+= skb1
->data_len
;
2030 skb_set_tail_pointer(skb
, len
);
2033 static inline void skb_split_no_header(struct sk_buff
*skb
,
2034 struct sk_buff
* skb1
,
2035 const u32 len
, int pos
)
2038 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
2040 skb_shinfo(skb
)->nr_frags
= 0;
2041 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
2043 skb
->data_len
= len
- pos
;
2045 for (i
= 0; i
< nfrags
; i
++) {
2046 int size
= skb_shinfo(skb
)->frags
[i
].size
;
2048 if (pos
+ size
> len
) {
2049 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
2053 * We have two variants in this case:
2054 * 1. Move all the frag to the second
2055 * part, if it is possible. F.e.
2056 * this approach is mandatory for TUX,
2057 * where splitting is expensive.
2058 * 2. Split is accurately. We make this.
2060 get_page(skb_shinfo(skb
)->frags
[i
].page
);
2061 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
2062 skb_shinfo(skb1
)->frags
[0].size
-= len
- pos
;
2063 skb_shinfo(skb
)->frags
[i
].size
= len
- pos
;
2064 skb_shinfo(skb
)->nr_frags
++;
2068 skb_shinfo(skb
)->nr_frags
++;
2071 skb_shinfo(skb1
)->nr_frags
= k
;
2075 * skb_split - Split fragmented skb to two parts at length len.
2076 * @skb: the buffer to split
2077 * @skb1: the buffer to receive the second part
2078 * @len: new length for skb
2080 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
2082 int pos
= skb_headlen(skb
);
2084 if (len
< pos
) /* Split line is inside header. */
2085 skb_split_inside_header(skb
, skb1
, len
, pos
);
2086 else /* Second chunk has no header, nothing to copy. */
2087 skb_split_no_header(skb
, skb1
, len
, pos
);
2089 EXPORT_SYMBOL(skb_split
);
2091 /* Shifting from/to a cloned skb is a no-go.
2093 * Caller cannot keep skb_shinfo related pointers past calling here!
2095 static int skb_prepare_for_shift(struct sk_buff
*skb
)
2097 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2101 * skb_shift - Shifts paged data partially from skb to another
2102 * @tgt: buffer into which tail data gets added
2103 * @skb: buffer from which the paged data comes from
2104 * @shiftlen: shift up to this many bytes
2106 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2107 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2108 * It's up to caller to free skb if everything was shifted.
2110 * If @tgt runs out of frags, the whole operation is aborted.
2112 * Skb cannot include anything else but paged data while tgt is allowed
2113 * to have non-paged data as well.
2115 * TODO: full sized shift could be optimized but that would need
2116 * specialized skb free'er to handle frags without up-to-date nr_frags.
2118 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
2120 int from
, to
, merge
, todo
;
2121 struct skb_frag_struct
*fragfrom
, *fragto
;
2123 BUG_ON(shiftlen
> skb
->len
);
2124 BUG_ON(skb_headlen(skb
)); /* Would corrupt stream */
2128 to
= skb_shinfo(tgt
)->nr_frags
;
2129 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2131 /* Actual merge is delayed until the point when we know we can
2132 * commit all, so that we don't have to undo partial changes
2135 !skb_can_coalesce(tgt
, to
, fragfrom
->page
, fragfrom
->page_offset
)) {
2140 todo
-= fragfrom
->size
;
2142 if (skb_prepare_for_shift(skb
) ||
2143 skb_prepare_for_shift(tgt
))
2146 /* All previous frag pointers might be stale! */
2147 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2148 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2150 fragto
->size
+= shiftlen
;
2151 fragfrom
->size
-= shiftlen
;
2152 fragfrom
->page_offset
+= shiftlen
;
2160 /* Skip full, not-fitting skb to avoid expensive operations */
2161 if ((shiftlen
== skb
->len
) &&
2162 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
2165 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
2168 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
2169 if (to
== MAX_SKB_FRAGS
)
2172 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2173 fragto
= &skb_shinfo(tgt
)->frags
[to
];
2175 if (todo
>= fragfrom
->size
) {
2176 *fragto
= *fragfrom
;
2177 todo
-= fragfrom
->size
;
2182 get_page(fragfrom
->page
);
2183 fragto
->page
= fragfrom
->page
;
2184 fragto
->page_offset
= fragfrom
->page_offset
;
2185 fragto
->size
= todo
;
2187 fragfrom
->page_offset
+= todo
;
2188 fragfrom
->size
-= todo
;
2196 /* Ready to "commit" this state change to tgt */
2197 skb_shinfo(tgt
)->nr_frags
= to
;
2200 fragfrom
= &skb_shinfo(skb
)->frags
[0];
2201 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2203 fragto
->size
+= fragfrom
->size
;
2204 put_page(fragfrom
->page
);
2207 /* Reposition in the original skb */
2209 while (from
< skb_shinfo(skb
)->nr_frags
)
2210 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
2211 skb_shinfo(skb
)->nr_frags
= to
;
2213 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
2216 /* Most likely the tgt won't ever need its checksum anymore, skb on
2217 * the other hand might need it if it needs to be resent
2219 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
2220 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2222 /* Yak, is it really working this way? Some helper please? */
2223 skb
->len
-= shiftlen
;
2224 skb
->data_len
-= shiftlen
;
2225 skb
->truesize
-= shiftlen
;
2226 tgt
->len
+= shiftlen
;
2227 tgt
->data_len
+= shiftlen
;
2228 tgt
->truesize
+= shiftlen
;
2234 * skb_prepare_seq_read - Prepare a sequential read of skb data
2235 * @skb: the buffer to read
2236 * @from: lower offset of data to be read
2237 * @to: upper offset of data to be read
2238 * @st: state variable
2240 * Initializes the specified state variable. Must be called before
2241 * invoking skb_seq_read() for the first time.
2243 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
2244 unsigned int to
, struct skb_seq_state
*st
)
2246 st
->lower_offset
= from
;
2247 st
->upper_offset
= to
;
2248 st
->root_skb
= st
->cur_skb
= skb
;
2249 st
->frag_idx
= st
->stepped_offset
= 0;
2250 st
->frag_data
= NULL
;
2252 EXPORT_SYMBOL(skb_prepare_seq_read
);
2255 * skb_seq_read - Sequentially read skb data
2256 * @consumed: number of bytes consumed by the caller so far
2257 * @data: destination pointer for data to be returned
2258 * @st: state variable
2260 * Reads a block of skb data at &consumed relative to the
2261 * lower offset specified to skb_prepare_seq_read(). Assigns
2262 * the head of the data block to &data and returns the length
2263 * of the block or 0 if the end of the skb data or the upper
2264 * offset has been reached.
2266 * The caller is not required to consume all of the data
2267 * returned, i.e. &consumed is typically set to the number
2268 * of bytes already consumed and the next call to
2269 * skb_seq_read() will return the remaining part of the block.
2271 * Note 1: The size of each block of data returned can be arbitrary,
2272 * this limitation is the cost for zerocopy seqeuental
2273 * reads of potentially non linear data.
2275 * Note 2: Fragment lists within fragments are not implemented
2276 * at the moment, state->root_skb could be replaced with
2277 * a stack for this purpose.
2279 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
2280 struct skb_seq_state
*st
)
2282 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
2285 if (unlikely(abs_offset
>= st
->upper_offset
))
2289 block_limit
= skb_headlen(st
->cur_skb
) + st
->stepped_offset
;
2291 if (abs_offset
< block_limit
&& !st
->frag_data
) {
2292 *data
= st
->cur_skb
->data
+ (abs_offset
- st
->stepped_offset
);
2293 return block_limit
- abs_offset
;
2296 if (st
->frag_idx
== 0 && !st
->frag_data
)
2297 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
2299 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
2300 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
2301 block_limit
= frag
->size
+ st
->stepped_offset
;
2303 if (abs_offset
< block_limit
) {
2305 st
->frag_data
= kmap_skb_frag(frag
);
2307 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
2308 (abs_offset
- st
->stepped_offset
);
2310 return block_limit
- abs_offset
;
2313 if (st
->frag_data
) {
2314 kunmap_skb_frag(st
->frag_data
);
2315 st
->frag_data
= NULL
;
2319 st
->stepped_offset
+= frag
->size
;
2322 if (st
->frag_data
) {
2323 kunmap_skb_frag(st
->frag_data
);
2324 st
->frag_data
= NULL
;
2327 if (st
->root_skb
== st
->cur_skb
&& skb_has_frag_list(st
->root_skb
)) {
2328 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
2331 } else if (st
->cur_skb
->next
) {
2332 st
->cur_skb
= st
->cur_skb
->next
;
2339 EXPORT_SYMBOL(skb_seq_read
);
2342 * skb_abort_seq_read - Abort a sequential read of skb data
2343 * @st: state variable
2345 * Must be called if skb_seq_read() was not called until it
2348 void skb_abort_seq_read(struct skb_seq_state
*st
)
2351 kunmap_skb_frag(st
->frag_data
);
2353 EXPORT_SYMBOL(skb_abort_seq_read
);
2355 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2357 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
2358 struct ts_config
*conf
,
2359 struct ts_state
*state
)
2361 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
2364 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
2366 skb_abort_seq_read(TS_SKB_CB(state
));
2370 * skb_find_text - Find a text pattern in skb data
2371 * @skb: the buffer to look in
2372 * @from: search offset
2374 * @config: textsearch configuration
2375 * @state: uninitialized textsearch state variable
2377 * Finds a pattern in the skb data according to the specified
2378 * textsearch configuration. Use textsearch_next() to retrieve
2379 * subsequent occurrences of the pattern. Returns the offset
2380 * to the first occurrence or UINT_MAX if no match was found.
2382 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
2383 unsigned int to
, struct ts_config
*config
,
2384 struct ts_state
*state
)
2388 config
->get_next_block
= skb_ts_get_next_block
;
2389 config
->finish
= skb_ts_finish
;
2391 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
2393 ret
= textsearch_find(config
, state
);
2394 return (ret
<= to
- from
? ret
: UINT_MAX
);
2396 EXPORT_SYMBOL(skb_find_text
);
2399 * skb_append_datato_frags: - append the user data to a skb
2400 * @sk: sock structure
2401 * @skb: skb structure to be appened with user data.
2402 * @getfrag: call back function to be used for getting the user data
2403 * @from: pointer to user message iov
2404 * @length: length of the iov message
2406 * Description: This procedure append the user data in the fragment part
2407 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2409 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
2410 int (*getfrag
)(void *from
, char *to
, int offset
,
2411 int len
, int odd
, struct sk_buff
*skb
),
2412 void *from
, int length
)
2415 skb_frag_t
*frag
= NULL
;
2416 struct page
*page
= NULL
;
2422 /* Return error if we don't have space for new frag */
2423 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2424 if (frg_cnt
>= MAX_SKB_FRAGS
)
2427 /* allocate a new page for next frag */
2428 page
= alloc_pages(sk
->sk_allocation
, 0);
2430 /* If alloc_page fails just return failure and caller will
2431 * free previous allocated pages by doing kfree_skb()
2436 /* initialize the next frag */
2437 skb_fill_page_desc(skb
, frg_cnt
, page
, 0, 0);
2438 skb
->truesize
+= PAGE_SIZE
;
2439 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
2441 /* get the new initialized frag */
2442 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2443 frag
= &skb_shinfo(skb
)->frags
[frg_cnt
- 1];
2445 /* copy the user data to page */
2446 left
= PAGE_SIZE
- frag
->page_offset
;
2447 copy
= (length
> left
)? left
: length
;
2449 ret
= getfrag(from
, (page_address(frag
->page
) +
2450 frag
->page_offset
+ frag
->size
),
2451 offset
, copy
, 0, skb
);
2455 /* copy was successful so update the size parameters */
2458 skb
->data_len
+= copy
;
2462 } while (length
> 0);
2466 EXPORT_SYMBOL(skb_append_datato_frags
);
2469 * skb_pull_rcsum - pull skb and update receive checksum
2470 * @skb: buffer to update
2471 * @len: length of data pulled
2473 * This function performs an skb_pull on the packet and updates
2474 * the CHECKSUM_COMPLETE checksum. It should be used on
2475 * receive path processing instead of skb_pull unless you know
2476 * that the checksum difference is zero (e.g., a valid IP header)
2477 * or you are setting ip_summed to CHECKSUM_NONE.
2479 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
2481 BUG_ON(len
> skb
->len
);
2483 BUG_ON(skb
->len
< skb
->data_len
);
2484 skb_postpull_rcsum(skb
, skb
->data
, len
);
2485 return skb
->data
+= len
;
2487 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
2490 * skb_segment - Perform protocol segmentation on skb.
2491 * @skb: buffer to segment
2492 * @features: features for the output path (see dev->features)
2494 * This function performs segmentation on the given skb. It returns
2495 * a pointer to the first in a list of new skbs for the segments.
2496 * In case of error it returns ERR_PTR(err).
2498 struct sk_buff
*skb_segment(struct sk_buff
*skb
, u32 features
)
2500 struct sk_buff
*segs
= NULL
;
2501 struct sk_buff
*tail
= NULL
;
2502 struct sk_buff
*fskb
= skb_shinfo(skb
)->frag_list
;
2503 unsigned int mss
= skb_shinfo(skb
)->gso_size
;
2504 unsigned int doffset
= skb
->data
- skb_mac_header(skb
);
2505 unsigned int offset
= doffset
;
2506 unsigned int headroom
;
2508 int sg
= !!(features
& NETIF_F_SG
);
2509 int nfrags
= skb_shinfo(skb
)->nr_frags
;
2514 __skb_push(skb
, doffset
);
2515 headroom
= skb_headroom(skb
);
2516 pos
= skb_headlen(skb
);
2519 struct sk_buff
*nskb
;
2524 len
= skb
->len
- offset
;
2528 hsize
= skb_headlen(skb
) - offset
;
2531 if (hsize
> len
|| !sg
)
2534 if (!hsize
&& i
>= nfrags
) {
2535 BUG_ON(fskb
->len
!= len
);
2538 nskb
= skb_clone(fskb
, GFP_ATOMIC
);
2541 if (unlikely(!nskb
))
2544 hsize
= skb_end_pointer(nskb
) - nskb
->head
;
2545 if (skb_cow_head(nskb
, doffset
+ headroom
)) {
2550 nskb
->truesize
+= skb_end_pointer(nskb
) - nskb
->head
-
2552 skb_release_head_state(nskb
);
2553 __skb_push(nskb
, doffset
);
2555 nskb
= alloc_skb(hsize
+ doffset
+ headroom
,
2558 if (unlikely(!nskb
))
2561 skb_reserve(nskb
, headroom
);
2562 __skb_put(nskb
, doffset
);
2571 __copy_skb_header(nskb
, skb
);
2572 nskb
->mac_len
= skb
->mac_len
;
2574 /* nskb and skb might have different headroom */
2575 if (nskb
->ip_summed
== CHECKSUM_PARTIAL
)
2576 nskb
->csum_start
+= skb_headroom(nskb
) - headroom
;
2578 skb_reset_mac_header(nskb
);
2579 skb_set_network_header(nskb
, skb
->mac_len
);
2580 nskb
->transport_header
= (nskb
->network_header
+
2581 skb_network_header_len(skb
));
2582 skb_copy_from_linear_data(skb
, nskb
->data
, doffset
);
2584 if (fskb
!= skb_shinfo(skb
)->frag_list
)
2588 nskb
->ip_summed
= CHECKSUM_NONE
;
2589 nskb
->csum
= skb_copy_and_csum_bits(skb
, offset
,
2595 frag
= skb_shinfo(nskb
)->frags
;
2597 skb_copy_from_linear_data_offset(skb
, offset
,
2598 skb_put(nskb
, hsize
), hsize
);
2600 while (pos
< offset
+ len
&& i
< nfrags
) {
2601 *frag
= skb_shinfo(skb
)->frags
[i
];
2602 get_page(frag
->page
);
2606 frag
->page_offset
+= offset
- pos
;
2607 frag
->size
-= offset
- pos
;
2610 skb_shinfo(nskb
)->nr_frags
++;
2612 if (pos
+ size
<= offset
+ len
) {
2616 frag
->size
-= pos
+ size
- (offset
+ len
);
2623 if (pos
< offset
+ len
) {
2624 struct sk_buff
*fskb2
= fskb
;
2626 BUG_ON(pos
+ fskb
->len
!= offset
+ len
);
2632 fskb2
= skb_clone(fskb2
, GFP_ATOMIC
);
2638 SKB_FRAG_ASSERT(nskb
);
2639 skb_shinfo(nskb
)->frag_list
= fskb2
;
2643 nskb
->data_len
= len
- hsize
;
2644 nskb
->len
+= nskb
->data_len
;
2645 nskb
->truesize
+= nskb
->data_len
;
2646 } while ((offset
+= len
) < skb
->len
);
2651 while ((skb
= segs
)) {
2655 return ERR_PTR(err
);
2657 EXPORT_SYMBOL_GPL(skb_segment
);
2659 int skb_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2661 struct sk_buff
*p
= *head
;
2662 struct sk_buff
*nskb
;
2663 struct skb_shared_info
*skbinfo
= skb_shinfo(skb
);
2664 struct skb_shared_info
*pinfo
= skb_shinfo(p
);
2665 unsigned int headroom
;
2666 unsigned int len
= skb_gro_len(skb
);
2667 unsigned int offset
= skb_gro_offset(skb
);
2668 unsigned int headlen
= skb_headlen(skb
);
2670 if (p
->len
+ len
>= 65536)
2673 if (pinfo
->frag_list
)
2675 else if (headlen
<= offset
) {
2678 int i
= skbinfo
->nr_frags
;
2679 int nr_frags
= pinfo
->nr_frags
+ i
;
2683 if (nr_frags
> MAX_SKB_FRAGS
)
2686 pinfo
->nr_frags
= nr_frags
;
2687 skbinfo
->nr_frags
= 0;
2689 frag
= pinfo
->frags
+ nr_frags
;
2690 frag2
= skbinfo
->frags
+ i
;
2695 frag
->page_offset
+= offset
;
2696 frag
->size
-= offset
;
2698 skb
->truesize
-= skb
->data_len
;
2699 skb
->len
-= skb
->data_len
;
2702 NAPI_GRO_CB(skb
)->free
= 1;
2704 } else if (skb_gro_len(p
) != pinfo
->gso_size
)
2707 headroom
= skb_headroom(p
);
2708 nskb
= alloc_skb(headroom
+ skb_gro_offset(p
), GFP_ATOMIC
);
2709 if (unlikely(!nskb
))
2712 __copy_skb_header(nskb
, p
);
2713 nskb
->mac_len
= p
->mac_len
;
2715 skb_reserve(nskb
, headroom
);
2716 __skb_put(nskb
, skb_gro_offset(p
));
2718 skb_set_mac_header(nskb
, skb_mac_header(p
) - p
->data
);
2719 skb_set_network_header(nskb
, skb_network_offset(p
));
2720 skb_set_transport_header(nskb
, skb_transport_offset(p
));
2722 __skb_pull(p
, skb_gro_offset(p
));
2723 memcpy(skb_mac_header(nskb
), skb_mac_header(p
),
2724 p
->data
- skb_mac_header(p
));
2726 *NAPI_GRO_CB(nskb
) = *NAPI_GRO_CB(p
);
2727 skb_shinfo(nskb
)->frag_list
= p
;
2728 skb_shinfo(nskb
)->gso_size
= pinfo
->gso_size
;
2729 pinfo
->gso_size
= 0;
2730 skb_header_release(p
);
2733 nskb
->data_len
+= p
->len
;
2734 nskb
->truesize
+= p
->len
;
2735 nskb
->len
+= p
->len
;
2738 nskb
->next
= p
->next
;
2744 if (offset
> headlen
) {
2745 unsigned int eat
= offset
- headlen
;
2747 skbinfo
->frags
[0].page_offset
+= eat
;
2748 skbinfo
->frags
[0].size
-= eat
;
2749 skb
->data_len
-= eat
;
2754 __skb_pull(skb
, offset
);
2756 p
->prev
->next
= skb
;
2758 skb_header_release(skb
);
2761 NAPI_GRO_CB(p
)->count
++;
2766 NAPI_GRO_CB(skb
)->same_flow
= 1;
2769 EXPORT_SYMBOL_GPL(skb_gro_receive
);
2771 void __init
skb_init(void)
2773 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
2774 sizeof(struct sk_buff
),
2776 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2778 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
2779 (2*sizeof(struct sk_buff
)) +
2782 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2787 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2788 * @skb: Socket buffer containing the buffers to be mapped
2789 * @sg: The scatter-gather list to map into
2790 * @offset: The offset into the buffer's contents to start mapping
2791 * @len: Length of buffer space to be mapped
2793 * Fill the specified scatter-gather list with mappings/pointers into a
2794 * region of the buffer space attached to a socket buffer.
2797 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2799 int start
= skb_headlen(skb
);
2800 int i
, copy
= start
- offset
;
2801 struct sk_buff
*frag_iter
;
2807 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
2809 if ((len
-= copy
) == 0)
2814 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2817 WARN_ON(start
> offset
+ len
);
2819 end
= start
+ skb_shinfo(skb
)->frags
[i
].size
;
2820 if ((copy
= end
- offset
) > 0) {
2821 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2825 sg_set_page(&sg
[elt
], frag
->page
, copy
,
2826 frag
->page_offset
+offset
-start
);
2835 skb_walk_frags(skb
, frag_iter
) {
2838 WARN_ON(start
> offset
+ len
);
2840 end
= start
+ frag_iter
->len
;
2841 if ((copy
= end
- offset
) > 0) {
2844 elt
+= __skb_to_sgvec(frag_iter
, sg
+elt
, offset
- start
,
2846 if ((len
-= copy
) == 0)
2856 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2858 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
);
2860 sg_mark_end(&sg
[nsg
- 1]);
2864 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
2867 * skb_cow_data - Check that a socket buffer's data buffers are writable
2868 * @skb: The socket buffer to check.
2869 * @tailbits: Amount of trailing space to be added
2870 * @trailer: Returned pointer to the skb where the @tailbits space begins
2872 * Make sure that the data buffers attached to a socket buffer are
2873 * writable. If they are not, private copies are made of the data buffers
2874 * and the socket buffer is set to use these instead.
2876 * If @tailbits is given, make sure that there is space to write @tailbits
2877 * bytes of data beyond current end of socket buffer. @trailer will be
2878 * set to point to the skb in which this space begins.
2880 * The number of scatterlist elements required to completely map the
2881 * COW'd and extended socket buffer will be returned.
2883 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
2887 struct sk_buff
*skb1
, **skb_p
;
2889 /* If skb is cloned or its head is paged, reallocate
2890 * head pulling out all the pages (pages are considered not writable
2891 * at the moment even if they are anonymous).
2893 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
2894 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
2897 /* Easy case. Most of packets will go this way. */
2898 if (!skb_has_frag_list(skb
)) {
2899 /* A little of trouble, not enough of space for trailer.
2900 * This should not happen, when stack is tuned to generate
2901 * good frames. OK, on miss we reallocate and reserve even more
2902 * space, 128 bytes is fair. */
2904 if (skb_tailroom(skb
) < tailbits
&&
2905 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
2913 /* Misery. We are in troubles, going to mincer fragments... */
2916 skb_p
= &skb_shinfo(skb
)->frag_list
;
2919 while ((skb1
= *skb_p
) != NULL
) {
2922 /* The fragment is partially pulled by someone,
2923 * this can happen on input. Copy it and everything
2926 if (skb_shared(skb1
))
2929 /* If the skb is the last, worry about trailer. */
2931 if (skb1
->next
== NULL
&& tailbits
) {
2932 if (skb_shinfo(skb1
)->nr_frags
||
2933 skb_has_frag_list(skb1
) ||
2934 skb_tailroom(skb1
) < tailbits
)
2935 ntail
= tailbits
+ 128;
2941 skb_shinfo(skb1
)->nr_frags
||
2942 skb_has_frag_list(skb1
)) {
2943 struct sk_buff
*skb2
;
2945 /* Fuck, we are miserable poor guys... */
2947 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
2949 skb2
= skb_copy_expand(skb1
,
2953 if (unlikely(skb2
== NULL
))
2957 skb_set_owner_w(skb2
, skb1
->sk
);
2959 /* Looking around. Are we still alive?
2960 * OK, link new skb, drop old one */
2962 skb2
->next
= skb1
->next
;
2969 skb_p
= &skb1
->next
;
2974 EXPORT_SYMBOL_GPL(skb_cow_data
);
2976 static void sock_rmem_free(struct sk_buff
*skb
)
2978 struct sock
*sk
= skb
->sk
;
2980 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
2984 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
2986 int sock_queue_err_skb(struct sock
*sk
, struct sk_buff
*skb
)
2988 if (atomic_read(&sk
->sk_rmem_alloc
) + skb
->truesize
>=
2989 (unsigned)sk
->sk_rcvbuf
)
2994 skb
->destructor
= sock_rmem_free
;
2995 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
2997 /* before exiting rcu section, make sure dst is refcounted */
3000 skb_queue_tail(&sk
->sk_error_queue
, skb
);
3001 if (!sock_flag(sk
, SOCK_DEAD
))
3002 sk
->sk_data_ready(sk
, skb
->len
);
3005 EXPORT_SYMBOL(sock_queue_err_skb
);
3007 void skb_tstamp_tx(struct sk_buff
*orig_skb
,
3008 struct skb_shared_hwtstamps
*hwtstamps
)
3010 struct sock
*sk
= orig_skb
->sk
;
3011 struct sock_exterr_skb
*serr
;
3012 struct sk_buff
*skb
;
3018 skb
= skb_clone(orig_skb
, GFP_ATOMIC
);
3023 *skb_hwtstamps(skb
) =
3027 * no hardware time stamps available,
3028 * so keep the shared tx_flags and only
3029 * store software time stamp
3031 skb
->tstamp
= ktime_get_real();
3034 serr
= SKB_EXT_ERR(skb
);
3035 memset(serr
, 0, sizeof(*serr
));
3036 serr
->ee
.ee_errno
= ENOMSG
;
3037 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TIMESTAMPING
;
3039 err
= sock_queue_err_skb(sk
, skb
);
3044 EXPORT_SYMBOL_GPL(skb_tstamp_tx
);
3048 * skb_partial_csum_set - set up and verify partial csum values for packet
3049 * @skb: the skb to set
3050 * @start: the number of bytes after skb->data to start checksumming.
3051 * @off: the offset from start to place the checksum.
3053 * For untrusted partially-checksummed packets, we need to make sure the values
3054 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3056 * This function checks and sets those values and skb->ip_summed: if this
3057 * returns false you should drop the packet.
3059 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
3061 if (unlikely(start
> skb_headlen(skb
)) ||
3062 unlikely((int)start
+ off
> skb_headlen(skb
) - 2)) {
3063 if (net_ratelimit())
3065 "bad partial csum: csum=%u/%u len=%u\n",
3066 start
, off
, skb_headlen(skb
));
3069 skb
->ip_summed
= CHECKSUM_PARTIAL
;
3070 skb
->csum_start
= skb_headroom(skb
) + start
;
3071 skb
->csum_offset
= off
;
3074 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);
3076 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
3078 if (net_ratelimit())
3079 pr_warning("%s: received packets cannot be forwarded"
3080 " while LRO is enabled\n", skb
->dev
->name
);
3082 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);