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 <trace/events/skb.h>
70 #include <linux/highmem.h>
72 static struct kmem_cache
*skbuff_head_cache __read_mostly
;
73 static struct kmem_cache
*skbuff_fclone_cache __read_mostly
;
75 static void sock_pipe_buf_release(struct pipe_inode_info
*pipe
,
76 struct pipe_buffer
*buf
)
81 static void sock_pipe_buf_get(struct pipe_inode_info
*pipe
,
82 struct pipe_buffer
*buf
)
87 static int sock_pipe_buf_steal(struct pipe_inode_info
*pipe
,
88 struct pipe_buffer
*buf
)
94 /* Pipe buffer operations for a socket. */
95 static const struct pipe_buf_operations sock_pipe_buf_ops
= {
97 .map
= generic_pipe_buf_map
,
98 .unmap
= generic_pipe_buf_unmap
,
99 .confirm
= generic_pipe_buf_confirm
,
100 .release
= sock_pipe_buf_release
,
101 .steal
= sock_pipe_buf_steal
,
102 .get
= sock_pipe_buf_get
,
106 * Keep out-of-line to prevent kernel bloat.
107 * __builtin_return_address is not used because it is not always
112 * skb_over_panic - private function
117 * Out of line support code for skb_put(). Not user callable.
119 static void skb_over_panic(struct sk_buff
*skb
, int sz
, void *here
)
121 printk(KERN_EMERG
"skb_over_panic: text:%p len:%d put:%d head:%p "
122 "data:%p tail:%#lx end:%#lx dev:%s\n",
123 here
, skb
->len
, sz
, skb
->head
, skb
->data
,
124 (unsigned long)skb
->tail
, (unsigned long)skb
->end
,
125 skb
->dev
? skb
->dev
->name
: "<NULL>");
130 * skb_under_panic - private function
135 * Out of line support code for skb_push(). Not user callable.
138 static void skb_under_panic(struct sk_buff
*skb
, int sz
, void *here
)
140 printk(KERN_EMERG
"skb_under_panic: text:%p len:%d put:%d head:%p "
141 "data:%p tail:%#lx end:%#lx dev:%s\n",
142 here
, skb
->len
, sz
, skb
->head
, skb
->data
,
143 (unsigned long)skb
->tail
, (unsigned long)skb
->end
,
144 skb
->dev
? skb
->dev
->name
: "<NULL>");
148 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
149 * 'private' fields and also do memory statistics to find all the
155 * __alloc_skb - allocate a network buffer
156 * @size: size to allocate
157 * @gfp_mask: allocation mask
158 * @fclone: allocate from fclone cache instead of head cache
159 * and allocate a cloned (child) skb
160 * @node: numa node to allocate memory on
162 * Allocate a new &sk_buff. The returned buffer has no headroom and a
163 * tail room of size bytes. The object has a reference count of one.
164 * The return is the buffer. On a failure the return is %NULL.
166 * Buffers may only be allocated from interrupts using a @gfp_mask of
169 struct sk_buff
*__alloc_skb(unsigned int size
, gfp_t gfp_mask
,
170 int fclone
, int node
)
172 struct kmem_cache
*cache
;
173 struct skb_shared_info
*shinfo
;
177 cache
= fclone
? skbuff_fclone_cache
: skbuff_head_cache
;
180 skb
= kmem_cache_alloc_node(cache
, gfp_mask
& ~__GFP_DMA
, node
);
185 /* We do our best to align skb_shared_info on a separate cache
186 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
187 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
188 * Both skb->head and skb_shared_info are cache line aligned.
190 size
= SKB_DATA_ALIGN(size
);
191 size
+= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
192 data
= kmalloc_node_track_caller(size
, gfp_mask
, node
);
195 /* kmalloc(size) might give us more room than requested.
196 * Put skb_shared_info exactly at the end of allocated zone,
197 * to allow max possible filling before reallocation.
199 size
= SKB_WITH_OVERHEAD(ksize(data
));
200 prefetchw(data
+ size
);
203 * Only clear those fields we need to clear, not those that we will
204 * actually initialise below. Hence, don't put any more fields after
205 * the tail pointer in struct sk_buff!
207 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
208 /* Account for allocated memory : skb + skb->head */
209 skb
->truesize
= SKB_TRUESIZE(size
);
210 atomic_set(&skb
->users
, 1);
213 skb_reset_tail_pointer(skb
);
214 skb
->end
= skb
->tail
+ size
;
215 #ifdef NET_SKBUFF_DATA_USES_OFFSET
216 skb
->mac_header
= ~0U;
219 /* make sure we initialize shinfo sequentially */
220 shinfo
= skb_shinfo(skb
);
221 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
222 atomic_set(&shinfo
->dataref
, 1);
223 kmemcheck_annotate_variable(shinfo
->destructor_arg
);
226 struct sk_buff
*child
= skb
+ 1;
227 atomic_t
*fclone_ref
= (atomic_t
*) (child
+ 1);
229 kmemcheck_annotate_bitfield(child
, flags1
);
230 kmemcheck_annotate_bitfield(child
, flags2
);
231 skb
->fclone
= SKB_FCLONE_ORIG
;
232 atomic_set(fclone_ref
, 1);
234 child
->fclone
= SKB_FCLONE_UNAVAILABLE
;
239 kmem_cache_free(cache
, skb
);
243 EXPORT_SYMBOL(__alloc_skb
);
246 * build_skb - build a network buffer
247 * @data: data buffer provided by caller
249 * Allocate a new &sk_buff. Caller provides space holding head and
250 * skb_shared_info. @data must have been allocated by kmalloc()
251 * The return is the new skb buffer.
252 * On a failure the return is %NULL, and @data is not freed.
254 * Before IO, driver allocates only data buffer where NIC put incoming frame
255 * Driver should add room at head (NET_SKB_PAD) and
256 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
257 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
258 * before giving packet to stack.
259 * RX rings only contains data buffers, not full skbs.
261 struct sk_buff
*build_skb(void *data
)
263 struct skb_shared_info
*shinfo
;
267 skb
= kmem_cache_alloc(skbuff_head_cache
, GFP_ATOMIC
);
271 size
= ksize(data
) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
273 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
274 skb
->truesize
= SKB_TRUESIZE(size
);
275 atomic_set(&skb
->users
, 1);
278 skb_reset_tail_pointer(skb
);
279 skb
->end
= skb
->tail
+ size
;
280 #ifdef NET_SKBUFF_DATA_USES_OFFSET
281 skb
->mac_header
= ~0U;
284 /* make sure we initialize shinfo sequentially */
285 shinfo
= skb_shinfo(skb
);
286 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
287 atomic_set(&shinfo
->dataref
, 1);
288 kmemcheck_annotate_variable(shinfo
->destructor_arg
);
292 EXPORT_SYMBOL(build_skb
);
295 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
296 * @dev: network device to receive on
297 * @length: length to allocate
298 * @gfp_mask: get_free_pages mask, passed to alloc_skb
300 * Allocate a new &sk_buff and assign it a usage count of one. The
301 * buffer has unspecified headroom built in. Users should allocate
302 * the headroom they think they need without accounting for the
303 * built in space. The built in space is used for optimisations.
305 * %NULL is returned if there is no free memory.
307 struct sk_buff
*__netdev_alloc_skb(struct net_device
*dev
,
308 unsigned int length
, gfp_t gfp_mask
)
312 skb
= __alloc_skb(length
+ NET_SKB_PAD
, gfp_mask
, 0, NUMA_NO_NODE
);
314 skb_reserve(skb
, NET_SKB_PAD
);
319 EXPORT_SYMBOL(__netdev_alloc_skb
);
321 void skb_add_rx_frag(struct sk_buff
*skb
, int i
, struct page
*page
, int off
,
322 int size
, unsigned int truesize
)
324 skb_fill_page_desc(skb
, i
, page
, off
, size
);
326 skb
->data_len
+= size
;
327 skb
->truesize
+= truesize
;
329 EXPORT_SYMBOL(skb_add_rx_frag
);
332 * dev_alloc_skb - allocate an skbuff for receiving
333 * @length: length to allocate
335 * Allocate a new &sk_buff and assign it a usage count of one. The
336 * buffer has unspecified headroom built in. Users should allocate
337 * the headroom they think they need without accounting for the
338 * built in space. The built in space is used for optimisations.
340 * %NULL is returned if there is no free memory. Although this function
341 * allocates memory it can be called from an interrupt.
343 struct sk_buff
*dev_alloc_skb(unsigned int length
)
346 * There is more code here than it seems:
347 * __dev_alloc_skb is an inline
349 return __dev_alloc_skb(length
, GFP_ATOMIC
);
351 EXPORT_SYMBOL(dev_alloc_skb
);
353 static void skb_drop_list(struct sk_buff
**listp
)
355 struct sk_buff
*list
= *listp
;
360 struct sk_buff
*this = list
;
366 static inline void skb_drop_fraglist(struct sk_buff
*skb
)
368 skb_drop_list(&skb_shinfo(skb
)->frag_list
);
371 static void skb_clone_fraglist(struct sk_buff
*skb
)
373 struct sk_buff
*list
;
375 skb_walk_frags(skb
, list
)
379 static void skb_release_data(struct sk_buff
*skb
)
382 !atomic_sub_return(skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1,
383 &skb_shinfo(skb
)->dataref
)) {
384 if (skb_shinfo(skb
)->nr_frags
) {
386 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
387 skb_frag_unref(skb
, i
);
391 * If skb buf is from userspace, we need to notify the caller
392 * the lower device DMA has done;
394 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
395 struct ubuf_info
*uarg
;
397 uarg
= skb_shinfo(skb
)->destructor_arg
;
399 uarg
->callback(uarg
);
402 if (skb_has_frag_list(skb
))
403 skb_drop_fraglist(skb
);
410 * Free an skbuff by memory without cleaning the state.
412 static void kfree_skbmem(struct sk_buff
*skb
)
414 struct sk_buff
*other
;
415 atomic_t
*fclone_ref
;
417 switch (skb
->fclone
) {
418 case SKB_FCLONE_UNAVAILABLE
:
419 kmem_cache_free(skbuff_head_cache
, skb
);
422 case SKB_FCLONE_ORIG
:
423 fclone_ref
= (atomic_t
*) (skb
+ 2);
424 if (atomic_dec_and_test(fclone_ref
))
425 kmem_cache_free(skbuff_fclone_cache
, skb
);
428 case SKB_FCLONE_CLONE
:
429 fclone_ref
= (atomic_t
*) (skb
+ 1);
432 /* The clone portion is available for
433 * fast-cloning again.
435 skb
->fclone
= SKB_FCLONE_UNAVAILABLE
;
437 if (atomic_dec_and_test(fclone_ref
))
438 kmem_cache_free(skbuff_fclone_cache
, other
);
443 static void skb_release_head_state(struct sk_buff
*skb
)
447 secpath_put(skb
->sp
);
449 if (skb
->destructor
) {
451 skb
->destructor(skb
);
453 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
454 nf_conntrack_put(skb
->nfct
);
456 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
457 nf_conntrack_put_reasm(skb
->nfct_reasm
);
459 #ifdef CONFIG_BRIDGE_NETFILTER
460 nf_bridge_put(skb
->nf_bridge
);
462 /* XXX: IS this still necessary? - JHS */
463 #ifdef CONFIG_NET_SCHED
465 #ifdef CONFIG_NET_CLS_ACT
471 /* Free everything but the sk_buff shell. */
472 static void skb_release_all(struct sk_buff
*skb
)
474 skb_release_head_state(skb
);
475 skb_release_data(skb
);
479 * __kfree_skb - private function
482 * Free an sk_buff. Release anything attached to the buffer.
483 * Clean the state. This is an internal helper function. Users should
484 * always call kfree_skb
487 void __kfree_skb(struct sk_buff
*skb
)
489 skb_release_all(skb
);
492 EXPORT_SYMBOL(__kfree_skb
);
495 * kfree_skb - free an sk_buff
496 * @skb: buffer to free
498 * Drop a reference to the buffer and free it if the usage count has
501 void kfree_skb(struct sk_buff
*skb
)
505 if (likely(atomic_read(&skb
->users
) == 1))
507 else if (likely(!atomic_dec_and_test(&skb
->users
)))
509 trace_kfree_skb(skb
, __builtin_return_address(0));
512 EXPORT_SYMBOL(kfree_skb
);
515 * consume_skb - free an skbuff
516 * @skb: buffer to free
518 * Drop a ref to the buffer and free it if the usage count has hit zero
519 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
520 * is being dropped after a failure and notes that
522 void consume_skb(struct sk_buff
*skb
)
526 if (likely(atomic_read(&skb
->users
) == 1))
528 else if (likely(!atomic_dec_and_test(&skb
->users
)))
530 trace_consume_skb(skb
);
533 EXPORT_SYMBOL(consume_skb
);
536 * skb_recycle - clean up an skb for reuse
539 * Recycles the skb to be reused as a receive buffer. This
540 * function does any necessary reference count dropping, and
541 * cleans up the skbuff as if it just came from __alloc_skb().
543 void skb_recycle(struct sk_buff
*skb
)
545 struct skb_shared_info
*shinfo
;
547 skb_release_head_state(skb
);
549 shinfo
= skb_shinfo(skb
);
550 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
551 atomic_set(&shinfo
->dataref
, 1);
553 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
554 skb
->data
= skb
->head
+ NET_SKB_PAD
;
555 skb_reset_tail_pointer(skb
);
557 EXPORT_SYMBOL(skb_recycle
);
560 * skb_recycle_check - check if skb can be reused for receive
562 * @skb_size: minimum receive buffer size
564 * Checks that the skb passed in is not shared or cloned, and
565 * that it is linear and its head portion at least as large as
566 * skb_size so that it can be recycled as a receive buffer.
567 * If these conditions are met, this function does any necessary
568 * reference count dropping and cleans up the skbuff as if it
569 * just came from __alloc_skb().
571 bool skb_recycle_check(struct sk_buff
*skb
, int skb_size
)
573 if (!skb_is_recycleable(skb
, skb_size
))
580 EXPORT_SYMBOL(skb_recycle_check
);
582 static void __copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
584 new->tstamp
= old
->tstamp
;
586 new->transport_header
= old
->transport_header
;
587 new->network_header
= old
->network_header
;
588 new->mac_header
= old
->mac_header
;
589 skb_dst_copy(new, old
);
590 new->rxhash
= old
->rxhash
;
591 new->ooo_okay
= old
->ooo_okay
;
592 new->l4_rxhash
= old
->l4_rxhash
;
593 new->no_fcs
= old
->no_fcs
;
595 new->sp
= secpath_get(old
->sp
);
597 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
598 new->csum
= old
->csum
;
599 new->local_df
= old
->local_df
;
600 new->pkt_type
= old
->pkt_type
;
601 new->ip_summed
= old
->ip_summed
;
602 skb_copy_queue_mapping(new, old
);
603 new->priority
= old
->priority
;
604 #if IS_ENABLED(CONFIG_IP_VS)
605 new->ipvs_property
= old
->ipvs_property
;
607 new->protocol
= old
->protocol
;
608 new->mark
= old
->mark
;
609 new->skb_iif
= old
->skb_iif
;
611 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
612 new->nf_trace
= old
->nf_trace
;
614 #ifdef CONFIG_NET_SCHED
615 new->tc_index
= old
->tc_index
;
616 #ifdef CONFIG_NET_CLS_ACT
617 new->tc_verd
= old
->tc_verd
;
620 new->vlan_tci
= old
->vlan_tci
;
622 skb_copy_secmark(new, old
);
626 * You should not add any new code to this function. Add it to
627 * __copy_skb_header above instead.
629 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
631 #define C(x) n->x = skb->x
633 n
->next
= n
->prev
= NULL
;
635 __copy_skb_header(n
, skb
);
640 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
643 n
->destructor
= NULL
;
649 atomic_set(&n
->users
, 1);
651 atomic_inc(&(skb_shinfo(skb
)->dataref
));
659 * skb_morph - morph one skb into another
660 * @dst: the skb to receive the contents
661 * @src: the skb to supply the contents
663 * This is identical to skb_clone except that the target skb is
664 * supplied by the user.
666 * The target skb is returned upon exit.
668 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
670 skb_release_all(dst
);
671 return __skb_clone(dst
, src
);
673 EXPORT_SYMBOL_GPL(skb_morph
);
675 /* skb_copy_ubufs - copy userspace skb frags buffers to kernel
676 * @skb: the skb to modify
677 * @gfp_mask: allocation priority
679 * This must be called on SKBTX_DEV_ZEROCOPY skb.
680 * It will copy all frags into kernel and drop the reference
681 * to userspace pages.
683 * If this function is called from an interrupt gfp_mask() must be
686 * Returns 0 on success or a negative error code on failure
687 * to allocate kernel memory to copy to.
689 int skb_copy_ubufs(struct sk_buff
*skb
, gfp_t gfp_mask
)
692 int num_frags
= skb_shinfo(skb
)->nr_frags
;
693 struct page
*page
, *head
= NULL
;
694 struct ubuf_info
*uarg
= skb_shinfo(skb
)->destructor_arg
;
696 for (i
= 0; i
< num_frags
; i
++) {
698 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
700 page
= alloc_page(GFP_ATOMIC
);
703 struct page
*next
= (struct page
*)head
->private;
709 vaddr
= kmap_atomic(skb_frag_page(f
));
710 memcpy(page_address(page
),
711 vaddr
+ f
->page_offset
, skb_frag_size(f
));
712 kunmap_atomic(vaddr
);
713 page
->private = (unsigned long)head
;
717 /* skb frags release userspace buffers */
718 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
719 skb_frag_unref(skb
, i
);
721 uarg
->callback(uarg
);
723 /* skb frags point to kernel buffers */
724 for (i
= skb_shinfo(skb
)->nr_frags
; i
> 0; i
--) {
725 __skb_fill_page_desc(skb
, i
-1, head
, 0,
726 skb_shinfo(skb
)->frags
[i
- 1].size
);
727 head
= (struct page
*)head
->private;
730 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
736 * skb_clone - duplicate an sk_buff
737 * @skb: buffer to clone
738 * @gfp_mask: allocation priority
740 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
741 * copies share the same packet data but not structure. The new
742 * buffer has a reference count of 1. If the allocation fails the
743 * function returns %NULL otherwise the new buffer is returned.
745 * If this function is called from an interrupt gfp_mask() must be
749 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
753 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
754 if (skb_copy_ubufs(skb
, gfp_mask
))
759 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
760 n
->fclone
== SKB_FCLONE_UNAVAILABLE
) {
761 atomic_t
*fclone_ref
= (atomic_t
*) (n
+ 1);
762 n
->fclone
= SKB_FCLONE_CLONE
;
763 atomic_inc(fclone_ref
);
765 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
769 kmemcheck_annotate_bitfield(n
, flags1
);
770 kmemcheck_annotate_bitfield(n
, flags2
);
771 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
774 return __skb_clone(n
, skb
);
776 EXPORT_SYMBOL(skb_clone
);
778 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
780 #ifndef NET_SKBUFF_DATA_USES_OFFSET
782 * Shift between the two data areas in bytes
784 unsigned long offset
= new->data
- old
->data
;
787 __copy_skb_header(new, old
);
789 #ifndef NET_SKBUFF_DATA_USES_OFFSET
790 /* {transport,network,mac}_header are relative to skb->head */
791 new->transport_header
+= offset
;
792 new->network_header
+= offset
;
793 if (skb_mac_header_was_set(new))
794 new->mac_header
+= offset
;
796 skb_shinfo(new)->gso_size
= skb_shinfo(old
)->gso_size
;
797 skb_shinfo(new)->gso_segs
= skb_shinfo(old
)->gso_segs
;
798 skb_shinfo(new)->gso_type
= skb_shinfo(old
)->gso_type
;
802 * skb_copy - create private copy of an sk_buff
803 * @skb: buffer to copy
804 * @gfp_mask: allocation priority
806 * Make a copy of both an &sk_buff and its data. This is used when the
807 * caller wishes to modify the data and needs a private copy of the
808 * data to alter. Returns %NULL on failure or the pointer to the buffer
809 * on success. The returned buffer has a reference count of 1.
811 * As by-product this function converts non-linear &sk_buff to linear
812 * one, so that &sk_buff becomes completely private and caller is allowed
813 * to modify all the data of returned buffer. This means that this
814 * function is not recommended for use in circumstances when only
815 * header is going to be modified. Use pskb_copy() instead.
818 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
820 int headerlen
= skb_headroom(skb
);
821 unsigned int size
= (skb_end_pointer(skb
) - skb
->head
) + skb
->data_len
;
822 struct sk_buff
*n
= alloc_skb(size
, gfp_mask
);
827 /* Set the data pointer */
828 skb_reserve(n
, headerlen
);
829 /* Set the tail pointer and length */
830 skb_put(n
, skb
->len
);
832 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
835 copy_skb_header(n
, skb
);
838 EXPORT_SYMBOL(skb_copy
);
841 * __pskb_copy - create copy of an sk_buff with private head.
842 * @skb: buffer to copy
843 * @headroom: headroom of new skb
844 * @gfp_mask: allocation priority
846 * Make a copy of both an &sk_buff and part of its data, located
847 * in header. Fragmented data remain shared. This is used when
848 * the caller wishes to modify only header of &sk_buff and needs
849 * private copy of the header to alter. Returns %NULL on failure
850 * or the pointer to the buffer on success.
851 * The returned buffer has a reference count of 1.
854 struct sk_buff
*__pskb_copy(struct sk_buff
*skb
, int headroom
, gfp_t gfp_mask
)
856 unsigned int size
= skb_headlen(skb
) + headroom
;
857 struct sk_buff
*n
= alloc_skb(size
, gfp_mask
);
862 /* Set the data pointer */
863 skb_reserve(n
, headroom
);
864 /* Set the tail pointer and length */
865 skb_put(n
, skb_headlen(skb
));
867 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
869 n
->truesize
+= skb
->data_len
;
870 n
->data_len
= skb
->data_len
;
873 if (skb_shinfo(skb
)->nr_frags
) {
876 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
877 if (skb_copy_ubufs(skb
, gfp_mask
)) {
883 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
884 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
885 skb_frag_ref(skb
, i
);
887 skb_shinfo(n
)->nr_frags
= i
;
890 if (skb_has_frag_list(skb
)) {
891 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
892 skb_clone_fraglist(n
);
895 copy_skb_header(n
, skb
);
899 EXPORT_SYMBOL(__pskb_copy
);
902 * pskb_expand_head - reallocate header of &sk_buff
903 * @skb: buffer to reallocate
904 * @nhead: room to add at head
905 * @ntail: room to add at tail
906 * @gfp_mask: allocation priority
908 * Expands (or creates identical copy, if &nhead and &ntail are zero)
909 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
910 * reference count of 1. Returns zero in the case of success or error,
911 * if expansion failed. In the last case, &sk_buff is not changed.
913 * All the pointers pointing into skb header may change and must be
914 * reloaded after call to this function.
917 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
922 int size
= nhead
+ (skb_end_pointer(skb
) - skb
->head
) + ntail
;
931 size
= SKB_DATA_ALIGN(size
);
933 /* Check if we can avoid taking references on fragments if we own
934 * the last reference on skb->head. (see skb_release_data())
939 int delta
= skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1;
940 fastpath
= atomic_read(&skb_shinfo(skb
)->dataref
) == delta
;
944 size
+ sizeof(struct skb_shared_info
) <= ksize(skb
->head
)) {
945 memmove(skb
->head
+ size
, skb_shinfo(skb
),
946 offsetof(struct skb_shared_info
,
947 frags
[skb_shinfo(skb
)->nr_frags
]));
948 memmove(skb
->head
+ nhead
, skb
->head
,
949 skb_tail_pointer(skb
) - skb
->head
);
954 data
= kmalloc(size
+ SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)),
958 size
= SKB_WITH_OVERHEAD(ksize(data
));
960 /* Copy only real data... and, alas, header. This should be
961 * optimized for the cases when header is void.
963 memcpy(data
+ nhead
, skb
->head
, skb_tail_pointer(skb
) - skb
->head
);
965 memcpy((struct skb_shared_info
*)(data
+ size
),
967 offsetof(struct skb_shared_info
, frags
[skb_shinfo(skb
)->nr_frags
]));
972 /* copy this zero copy skb frags */
973 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
974 if (skb_copy_ubufs(skb
, gfp_mask
))
977 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
978 skb_frag_ref(skb
, i
);
980 if (skb_has_frag_list(skb
))
981 skb_clone_fraglist(skb
);
983 skb_release_data(skb
);
985 off
= (data
+ nhead
) - skb
->head
;
990 #ifdef NET_SKBUFF_DATA_USES_OFFSET
994 skb
->end
= skb
->head
+ size
;
996 /* {transport,network,mac}_header and tail are relative to skb->head */
998 skb
->transport_header
+= off
;
999 skb
->network_header
+= off
;
1000 if (skb_mac_header_was_set(skb
))
1001 skb
->mac_header
+= off
;
1002 /* Only adjust this if it actually is csum_start rather than csum */
1003 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1004 skb
->csum_start
+= nhead
;
1008 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
1016 EXPORT_SYMBOL(pskb_expand_head
);
1018 /* Make private copy of skb with writable head and some headroom */
1020 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
1022 struct sk_buff
*skb2
;
1023 int delta
= headroom
- skb_headroom(skb
);
1026 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
1028 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1029 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
1037 EXPORT_SYMBOL(skb_realloc_headroom
);
1040 * skb_copy_expand - copy and expand sk_buff
1041 * @skb: buffer to copy
1042 * @newheadroom: new free bytes at head
1043 * @newtailroom: new free bytes at tail
1044 * @gfp_mask: allocation priority
1046 * Make a copy of both an &sk_buff and its data and while doing so
1047 * allocate additional space.
1049 * This is used when the caller wishes to modify the data and needs a
1050 * private copy of the data to alter as well as more space for new fields.
1051 * Returns %NULL on failure or the pointer to the buffer
1052 * on success. The returned buffer has a reference count of 1.
1054 * You must pass %GFP_ATOMIC as the allocation priority if this function
1055 * is called from an interrupt.
1057 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
1058 int newheadroom
, int newtailroom
,
1062 * Allocate the copy buffer
1064 struct sk_buff
*n
= alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
1066 int oldheadroom
= skb_headroom(skb
);
1067 int head_copy_len
, head_copy_off
;
1073 skb_reserve(n
, newheadroom
);
1075 /* Set the tail pointer and length */
1076 skb_put(n
, skb
->len
);
1078 head_copy_len
= oldheadroom
;
1080 if (newheadroom
<= head_copy_len
)
1081 head_copy_len
= newheadroom
;
1083 head_copy_off
= newheadroom
- head_copy_len
;
1085 /* Copy the linear header and data. */
1086 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
1087 skb
->len
+ head_copy_len
))
1090 copy_skb_header(n
, skb
);
1092 off
= newheadroom
- oldheadroom
;
1093 if (n
->ip_summed
== CHECKSUM_PARTIAL
)
1094 n
->csum_start
+= off
;
1095 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1096 n
->transport_header
+= off
;
1097 n
->network_header
+= off
;
1098 if (skb_mac_header_was_set(skb
))
1099 n
->mac_header
+= off
;
1104 EXPORT_SYMBOL(skb_copy_expand
);
1107 * skb_pad - zero pad the tail of an skb
1108 * @skb: buffer to pad
1109 * @pad: space to pad
1111 * Ensure that a buffer is followed by a padding area that is zero
1112 * filled. Used by network drivers which may DMA or transfer data
1113 * beyond the buffer end onto the wire.
1115 * May return error in out of memory cases. The skb is freed on error.
1118 int skb_pad(struct sk_buff
*skb
, int pad
)
1123 /* If the skbuff is non linear tailroom is always zero.. */
1124 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
1125 memset(skb
->data
+skb
->len
, 0, pad
);
1129 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
1130 if (likely(skb_cloned(skb
) || ntail
> 0)) {
1131 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
1136 /* FIXME: The use of this function with non-linear skb's really needs
1139 err
= skb_linearize(skb
);
1143 memset(skb
->data
+ skb
->len
, 0, pad
);
1150 EXPORT_SYMBOL(skb_pad
);
1153 * skb_put - add data to a buffer
1154 * @skb: buffer to use
1155 * @len: amount of data to add
1157 * This function extends the used data area of the buffer. If this would
1158 * exceed the total buffer size the kernel will panic. A pointer to the
1159 * first byte of the extra data is returned.
1161 unsigned char *skb_put(struct sk_buff
*skb
, unsigned int len
)
1163 unsigned char *tmp
= skb_tail_pointer(skb
);
1164 SKB_LINEAR_ASSERT(skb
);
1167 if (unlikely(skb
->tail
> skb
->end
))
1168 skb_over_panic(skb
, len
, __builtin_return_address(0));
1171 EXPORT_SYMBOL(skb_put
);
1174 * skb_push - add data to the start of a buffer
1175 * @skb: buffer to use
1176 * @len: amount of data to add
1178 * This function extends the used data area of the buffer at the buffer
1179 * start. If this would exceed the total buffer headroom the kernel will
1180 * panic. A pointer to the first byte of the extra data is returned.
1182 unsigned char *skb_push(struct sk_buff
*skb
, unsigned int len
)
1186 if (unlikely(skb
->data
<skb
->head
))
1187 skb_under_panic(skb
, len
, __builtin_return_address(0));
1190 EXPORT_SYMBOL(skb_push
);
1193 * skb_pull - remove data from the start of a buffer
1194 * @skb: buffer to use
1195 * @len: amount of data to remove
1197 * This function removes data from the start of a buffer, returning
1198 * the memory to the headroom. A pointer to the next data in the buffer
1199 * is returned. Once the data has been pulled future pushes will overwrite
1202 unsigned char *skb_pull(struct sk_buff
*skb
, unsigned int len
)
1204 return skb_pull_inline(skb
, len
);
1206 EXPORT_SYMBOL(skb_pull
);
1209 * skb_trim - remove end from a buffer
1210 * @skb: buffer to alter
1213 * Cut the length of a buffer down by removing data from the tail. If
1214 * the buffer is already under the length specified it is not modified.
1215 * The skb must be linear.
1217 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1220 __skb_trim(skb
, len
);
1222 EXPORT_SYMBOL(skb_trim
);
1224 /* Trims skb to length len. It can change skb pointers.
1227 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1229 struct sk_buff
**fragp
;
1230 struct sk_buff
*frag
;
1231 int offset
= skb_headlen(skb
);
1232 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1236 if (skb_cloned(skb
) &&
1237 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1244 for (; i
< nfrags
; i
++) {
1245 int end
= offset
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1252 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
++], len
- offset
);
1255 skb_shinfo(skb
)->nr_frags
= i
;
1257 for (; i
< nfrags
; i
++)
1258 skb_frag_unref(skb
, i
);
1260 if (skb_has_frag_list(skb
))
1261 skb_drop_fraglist(skb
);
1265 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1266 fragp
= &frag
->next
) {
1267 int end
= offset
+ frag
->len
;
1269 if (skb_shared(frag
)) {
1270 struct sk_buff
*nfrag
;
1272 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1273 if (unlikely(!nfrag
))
1276 nfrag
->next
= frag
->next
;
1288 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1292 skb_drop_list(&frag
->next
);
1297 if (len
> skb_headlen(skb
)) {
1298 skb
->data_len
-= skb
->len
- len
;
1303 skb_set_tail_pointer(skb
, len
);
1308 EXPORT_SYMBOL(___pskb_trim
);
1311 * __pskb_pull_tail - advance tail of skb header
1312 * @skb: buffer to reallocate
1313 * @delta: number of bytes to advance tail
1315 * The function makes a sense only on a fragmented &sk_buff,
1316 * it expands header moving its tail forward and copying necessary
1317 * data from fragmented part.
1319 * &sk_buff MUST have reference count of 1.
1321 * Returns %NULL (and &sk_buff does not change) if pull failed
1322 * or value of new tail of skb in the case of success.
1324 * All the pointers pointing into skb header may change and must be
1325 * reloaded after call to this function.
1328 /* Moves tail of skb head forward, copying data from fragmented part,
1329 * when it is necessary.
1330 * 1. It may fail due to malloc failure.
1331 * 2. It may change skb pointers.
1333 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1335 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1337 /* If skb has not enough free space at tail, get new one
1338 * plus 128 bytes for future expansions. If we have enough
1339 * room at tail, reallocate without expansion only if skb is cloned.
1341 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1343 if (eat
> 0 || skb_cloned(skb
)) {
1344 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1349 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
1352 /* Optimization: no fragments, no reasons to preestimate
1353 * size of pulled pages. Superb.
1355 if (!skb_has_frag_list(skb
))
1358 /* Estimate size of pulled pages. */
1360 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1361 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1368 /* If we need update frag list, we are in troubles.
1369 * Certainly, it possible to add an offset to skb data,
1370 * but taking into account that pulling is expected to
1371 * be very rare operation, it is worth to fight against
1372 * further bloating skb head and crucify ourselves here instead.
1373 * Pure masohism, indeed. 8)8)
1376 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1377 struct sk_buff
*clone
= NULL
;
1378 struct sk_buff
*insp
= NULL
;
1383 if (list
->len
<= eat
) {
1384 /* Eaten as whole. */
1389 /* Eaten partially. */
1391 if (skb_shared(list
)) {
1392 /* Sucks! We need to fork list. :-( */
1393 clone
= skb_clone(list
, GFP_ATOMIC
);
1399 /* This may be pulled without
1403 if (!pskb_pull(list
, eat
)) {
1411 /* Free pulled out fragments. */
1412 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1413 skb_shinfo(skb
)->frag_list
= list
->next
;
1416 /* And insert new clone at head. */
1419 skb_shinfo(skb
)->frag_list
= clone
;
1422 /* Success! Now we may commit changes to skb data. */
1427 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1428 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1431 skb_frag_unref(skb
, i
);
1434 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1436 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1437 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[k
], eat
);
1443 skb_shinfo(skb
)->nr_frags
= k
;
1446 skb
->data_len
-= delta
;
1448 return skb_tail_pointer(skb
);
1450 EXPORT_SYMBOL(__pskb_pull_tail
);
1453 * skb_copy_bits - copy bits from skb to kernel buffer
1455 * @offset: offset in source
1456 * @to: destination buffer
1457 * @len: number of bytes to copy
1459 * Copy the specified number of bytes from the source skb to the
1460 * destination buffer.
1463 * If its prototype is ever changed,
1464 * check arch/{*}/net/{*}.S files,
1465 * since it is called from BPF assembly code.
1467 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1469 int start
= skb_headlen(skb
);
1470 struct sk_buff
*frag_iter
;
1473 if (offset
> (int)skb
->len
- len
)
1477 if ((copy
= start
- offset
) > 0) {
1480 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
1481 if ((len
-= copy
) == 0)
1487 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1489 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
1491 WARN_ON(start
> offset
+ len
);
1493 end
= start
+ skb_frag_size(f
);
1494 if ((copy
= end
- offset
) > 0) {
1500 vaddr
= kmap_atomic(skb_frag_page(f
));
1502 vaddr
+ f
->page_offset
+ offset
- start
,
1504 kunmap_atomic(vaddr
);
1506 if ((len
-= copy
) == 0)
1514 skb_walk_frags(skb
, frag_iter
) {
1517 WARN_ON(start
> offset
+ len
);
1519 end
= start
+ frag_iter
->len
;
1520 if ((copy
= end
- offset
) > 0) {
1523 if (skb_copy_bits(frag_iter
, offset
- start
, to
, copy
))
1525 if ((len
-= copy
) == 0)
1539 EXPORT_SYMBOL(skb_copy_bits
);
1542 * Callback from splice_to_pipe(), if we need to release some pages
1543 * at the end of the spd in case we error'ed out in filling the pipe.
1545 static void sock_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
1547 put_page(spd
->pages
[i
]);
1550 static struct page
*linear_to_page(struct page
*page
, unsigned int *len
,
1551 unsigned int *offset
,
1552 struct sk_buff
*skb
, struct sock
*sk
)
1554 struct page
*p
= sk
->sk_sndmsg_page
;
1559 p
= sk
->sk_sndmsg_page
= alloc_pages(sk
->sk_allocation
, 0);
1563 off
= sk
->sk_sndmsg_off
= 0;
1564 /* hold one ref to this page until it's full */
1568 /* If we are the only user of the page, we can reset offset */
1569 if (page_count(p
) == 1)
1570 sk
->sk_sndmsg_off
= 0;
1571 off
= sk
->sk_sndmsg_off
;
1572 mlen
= PAGE_SIZE
- off
;
1573 if (mlen
< 64 && mlen
< *len
) {
1578 *len
= min_t(unsigned int, *len
, mlen
);
1581 memcpy(page_address(p
) + off
, page_address(page
) + *offset
, *len
);
1582 sk
->sk_sndmsg_off
+= *len
;
1588 static bool spd_can_coalesce(const struct splice_pipe_desc
*spd
,
1590 unsigned int offset
)
1592 return spd
->nr_pages
&&
1593 spd
->pages
[spd
->nr_pages
- 1] == page
&&
1594 (spd
->partial
[spd
->nr_pages
- 1].offset
+
1595 spd
->partial
[spd
->nr_pages
- 1].len
== offset
);
1599 * Fill page/offset/length into spd, if it can hold more pages.
1601 static bool spd_fill_page(struct splice_pipe_desc
*spd
,
1602 struct pipe_inode_info
*pipe
, struct page
*page
,
1603 unsigned int *len
, unsigned int offset
,
1604 struct sk_buff
*skb
, bool linear
,
1607 if (unlikely(spd
->nr_pages
== MAX_SKB_FRAGS
))
1611 page
= linear_to_page(page
, len
, &offset
, skb
, sk
);
1615 if (spd_can_coalesce(spd
, page
, offset
)) {
1616 spd
->partial
[spd
->nr_pages
- 1].len
+= *len
;
1620 spd
->pages
[spd
->nr_pages
] = page
;
1621 spd
->partial
[spd
->nr_pages
].len
= *len
;
1622 spd
->partial
[spd
->nr_pages
].offset
= offset
;
1628 static inline void __segment_seek(struct page
**page
, unsigned int *poff
,
1629 unsigned int *plen
, unsigned int off
)
1634 n
= *poff
/ PAGE_SIZE
;
1636 *page
= nth_page(*page
, n
);
1638 *poff
= *poff
% PAGE_SIZE
;
1642 static bool __splice_segment(struct page
*page
, unsigned int poff
,
1643 unsigned int plen
, unsigned int *off
,
1644 unsigned int *len
, struct sk_buff
*skb
,
1645 struct splice_pipe_desc
*spd
, bool linear
,
1647 struct pipe_inode_info
*pipe
)
1652 /* skip this segment if already processed */
1658 /* ignore any bits we already processed */
1660 __segment_seek(&page
, &poff
, &plen
, *off
);
1665 unsigned int flen
= min(*len
, plen
);
1667 /* the linear region may spread across several pages */
1668 flen
= min_t(unsigned int, flen
, PAGE_SIZE
- poff
);
1670 if (spd_fill_page(spd
, pipe
, page
, &flen
, poff
, skb
, linear
, sk
))
1673 __segment_seek(&page
, &poff
, &plen
, flen
);
1676 } while (*len
&& plen
);
1682 * Map linear and fragment data from the skb to spd. It reports true if the
1683 * pipe is full or if we already spliced the requested length.
1685 static bool __skb_splice_bits(struct sk_buff
*skb
, struct pipe_inode_info
*pipe
,
1686 unsigned int *offset
, unsigned int *len
,
1687 struct splice_pipe_desc
*spd
, struct sock
*sk
)
1692 * map the linear part
1694 if (__splice_segment(virt_to_page(skb
->data
),
1695 (unsigned long) skb
->data
& (PAGE_SIZE
- 1),
1697 offset
, len
, skb
, spd
, true, sk
, pipe
))
1701 * then map the fragments
1703 for (seg
= 0; seg
< skb_shinfo(skb
)->nr_frags
; seg
++) {
1704 const skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[seg
];
1706 if (__splice_segment(skb_frag_page(f
),
1707 f
->page_offset
, skb_frag_size(f
),
1708 offset
, len
, skb
, spd
, false, sk
, pipe
))
1716 * Map data from the skb to a pipe. Should handle both the linear part,
1717 * the fragments, and the frag list. It does NOT handle frag lists within
1718 * the frag list, if such a thing exists. We'd probably need to recurse to
1719 * handle that cleanly.
1721 int skb_splice_bits(struct sk_buff
*skb
, unsigned int offset
,
1722 struct pipe_inode_info
*pipe
, unsigned int tlen
,
1725 struct partial_page partial
[MAX_SKB_FRAGS
];
1726 struct page
*pages
[MAX_SKB_FRAGS
];
1727 struct splice_pipe_desc spd
= {
1731 .ops
= &sock_pipe_buf_ops
,
1732 .spd_release
= sock_spd_release
,
1734 struct sk_buff
*frag_iter
;
1735 struct sock
*sk
= skb
->sk
;
1739 * __skb_splice_bits() only fails if the output has no room left,
1740 * so no point in going over the frag_list for the error case.
1742 if (__skb_splice_bits(skb
, pipe
, &offset
, &tlen
, &spd
, sk
))
1748 * now see if we have a frag_list to map
1750 skb_walk_frags(skb
, frag_iter
) {
1753 if (__skb_splice_bits(frag_iter
, pipe
, &offset
, &tlen
, &spd
, sk
))
1760 * Drop the socket lock, otherwise we have reverse
1761 * locking dependencies between sk_lock and i_mutex
1762 * here as compared to sendfile(). We enter here
1763 * with the socket lock held, and splice_to_pipe() will
1764 * grab the pipe inode lock. For sendfile() emulation,
1765 * we call into ->sendpage() with the i_mutex lock held
1766 * and networking will grab the socket lock.
1769 ret
= splice_to_pipe(pipe
, &spd
);
1777 * skb_store_bits - store bits from kernel buffer to skb
1778 * @skb: destination buffer
1779 * @offset: offset in destination
1780 * @from: source buffer
1781 * @len: number of bytes to copy
1783 * Copy the specified number of bytes from the source buffer to the
1784 * destination skb. This function handles all the messy bits of
1785 * traversing fragment lists and such.
1788 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
1790 int start
= skb_headlen(skb
);
1791 struct sk_buff
*frag_iter
;
1794 if (offset
> (int)skb
->len
- len
)
1797 if ((copy
= start
- offset
) > 0) {
1800 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
1801 if ((len
-= copy
) == 0)
1807 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1808 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1811 WARN_ON(start
> offset
+ len
);
1813 end
= start
+ skb_frag_size(frag
);
1814 if ((copy
= end
- offset
) > 0) {
1820 vaddr
= kmap_atomic(skb_frag_page(frag
));
1821 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1823 kunmap_atomic(vaddr
);
1825 if ((len
-= copy
) == 0)
1833 skb_walk_frags(skb
, frag_iter
) {
1836 WARN_ON(start
> offset
+ len
);
1838 end
= start
+ frag_iter
->len
;
1839 if ((copy
= end
- offset
) > 0) {
1842 if (skb_store_bits(frag_iter
, offset
- start
,
1845 if ((len
-= copy
) == 0)
1858 EXPORT_SYMBOL(skb_store_bits
);
1860 /* Checksum skb data. */
1862 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
1863 int len
, __wsum csum
)
1865 int start
= skb_headlen(skb
);
1866 int i
, copy
= start
- offset
;
1867 struct sk_buff
*frag_iter
;
1870 /* Checksum header. */
1874 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1875 if ((len
-= copy
) == 0)
1881 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1883 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1885 WARN_ON(start
> offset
+ len
);
1887 end
= start
+ skb_frag_size(frag
);
1888 if ((copy
= end
- offset
) > 0) {
1894 vaddr
= kmap_atomic(skb_frag_page(frag
));
1895 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
1896 offset
- start
, copy
, 0);
1897 kunmap_atomic(vaddr
);
1898 csum
= csum_block_add(csum
, csum2
, pos
);
1907 skb_walk_frags(skb
, frag_iter
) {
1910 WARN_ON(start
> offset
+ len
);
1912 end
= start
+ frag_iter
->len
;
1913 if ((copy
= end
- offset
) > 0) {
1917 csum2
= skb_checksum(frag_iter
, offset
- start
,
1919 csum
= csum_block_add(csum
, csum2
, pos
);
1920 if ((len
-= copy
) == 0)
1931 EXPORT_SYMBOL(skb_checksum
);
1933 /* Both of above in one bottle. */
1935 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
1936 u8
*to
, int len
, __wsum csum
)
1938 int start
= skb_headlen(skb
);
1939 int i
, copy
= start
- offset
;
1940 struct sk_buff
*frag_iter
;
1947 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
1949 if ((len
-= copy
) == 0)
1956 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1959 WARN_ON(start
> offset
+ len
);
1961 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1962 if ((copy
= end
- offset
) > 0) {
1965 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1969 vaddr
= kmap_atomic(skb_frag_page(frag
));
1970 csum2
= csum_partial_copy_nocheck(vaddr
+
1974 kunmap_atomic(vaddr
);
1975 csum
= csum_block_add(csum
, csum2
, pos
);
1985 skb_walk_frags(skb
, frag_iter
) {
1989 WARN_ON(start
> offset
+ len
);
1991 end
= start
+ frag_iter
->len
;
1992 if ((copy
= end
- offset
) > 0) {
1995 csum2
= skb_copy_and_csum_bits(frag_iter
,
1998 csum
= csum_block_add(csum
, csum2
, pos
);
1999 if ((len
-= copy
) == 0)
2010 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
2012 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
2017 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
2018 csstart
= skb_checksum_start_offset(skb
);
2020 csstart
= skb_headlen(skb
);
2022 BUG_ON(csstart
> skb_headlen(skb
));
2024 skb_copy_from_linear_data(skb
, to
, csstart
);
2027 if (csstart
!= skb
->len
)
2028 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
2029 skb
->len
- csstart
, 0);
2031 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2032 long csstuff
= csstart
+ skb
->csum_offset
;
2034 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
2037 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
2040 * skb_dequeue - remove from the head of the queue
2041 * @list: list to dequeue from
2043 * Remove the head of the list. The list lock is taken so the function
2044 * may be used safely with other locking list functions. The head item is
2045 * returned or %NULL if the list is empty.
2048 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
2050 unsigned long flags
;
2051 struct sk_buff
*result
;
2053 spin_lock_irqsave(&list
->lock
, flags
);
2054 result
= __skb_dequeue(list
);
2055 spin_unlock_irqrestore(&list
->lock
, flags
);
2058 EXPORT_SYMBOL(skb_dequeue
);
2061 * skb_dequeue_tail - remove from the tail of the queue
2062 * @list: list to dequeue from
2064 * Remove the tail of the list. The list lock is taken so the function
2065 * may be used safely with other locking list functions. The tail item is
2066 * returned or %NULL if the list is empty.
2068 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
2070 unsigned long flags
;
2071 struct sk_buff
*result
;
2073 spin_lock_irqsave(&list
->lock
, flags
);
2074 result
= __skb_dequeue_tail(list
);
2075 spin_unlock_irqrestore(&list
->lock
, flags
);
2078 EXPORT_SYMBOL(skb_dequeue_tail
);
2081 * skb_queue_purge - empty a list
2082 * @list: list to empty
2084 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2085 * the list and one reference dropped. This function takes the list
2086 * lock and is atomic with respect to other list locking functions.
2088 void skb_queue_purge(struct sk_buff_head
*list
)
2090 struct sk_buff
*skb
;
2091 while ((skb
= skb_dequeue(list
)) != NULL
)
2094 EXPORT_SYMBOL(skb_queue_purge
);
2097 * skb_queue_head - queue a buffer at the list head
2098 * @list: list to use
2099 * @newsk: buffer to queue
2101 * Queue a buffer at the start of the list. This function takes the
2102 * list lock and can be used safely with other locking &sk_buff functions
2105 * A buffer cannot be placed on two lists at the same time.
2107 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2109 unsigned long flags
;
2111 spin_lock_irqsave(&list
->lock
, flags
);
2112 __skb_queue_head(list
, newsk
);
2113 spin_unlock_irqrestore(&list
->lock
, flags
);
2115 EXPORT_SYMBOL(skb_queue_head
);
2118 * skb_queue_tail - queue a buffer at the list tail
2119 * @list: list to use
2120 * @newsk: buffer to queue
2122 * Queue a buffer at the tail of the list. This function takes the
2123 * list lock and can be used safely with other locking &sk_buff functions
2126 * A buffer cannot be placed on two lists at the same time.
2128 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2130 unsigned long flags
;
2132 spin_lock_irqsave(&list
->lock
, flags
);
2133 __skb_queue_tail(list
, newsk
);
2134 spin_unlock_irqrestore(&list
->lock
, flags
);
2136 EXPORT_SYMBOL(skb_queue_tail
);
2139 * skb_unlink - remove a buffer from a list
2140 * @skb: buffer to remove
2141 * @list: list to use
2143 * Remove a packet from a list. The list locks are taken and this
2144 * function is atomic with respect to other list locked calls
2146 * You must know what list the SKB is on.
2148 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
2150 unsigned long flags
;
2152 spin_lock_irqsave(&list
->lock
, flags
);
2153 __skb_unlink(skb
, list
);
2154 spin_unlock_irqrestore(&list
->lock
, flags
);
2156 EXPORT_SYMBOL(skb_unlink
);
2159 * skb_append - append a buffer
2160 * @old: buffer to insert after
2161 * @newsk: buffer to insert
2162 * @list: list to use
2164 * Place a packet after a given packet in a list. The list locks are taken
2165 * and this function is atomic with respect to other list locked calls.
2166 * A buffer cannot be placed on two lists at the same time.
2168 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2170 unsigned long flags
;
2172 spin_lock_irqsave(&list
->lock
, flags
);
2173 __skb_queue_after(list
, old
, newsk
);
2174 spin_unlock_irqrestore(&list
->lock
, flags
);
2176 EXPORT_SYMBOL(skb_append
);
2179 * skb_insert - insert a buffer
2180 * @old: buffer to insert before
2181 * @newsk: buffer to insert
2182 * @list: list to use
2184 * Place a packet before a given packet in a list. The list locks are
2185 * taken and this function is atomic with respect to other list locked
2188 * A buffer cannot be placed on two lists at the same time.
2190 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2192 unsigned long flags
;
2194 spin_lock_irqsave(&list
->lock
, flags
);
2195 __skb_insert(newsk
, old
->prev
, old
, list
);
2196 spin_unlock_irqrestore(&list
->lock
, flags
);
2198 EXPORT_SYMBOL(skb_insert
);
2200 static inline void skb_split_inside_header(struct sk_buff
*skb
,
2201 struct sk_buff
* skb1
,
2202 const u32 len
, const int pos
)
2206 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
2208 /* And move data appendix as is. */
2209 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
2210 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
2212 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
2213 skb_shinfo(skb
)->nr_frags
= 0;
2214 skb1
->data_len
= skb
->data_len
;
2215 skb1
->len
+= skb1
->data_len
;
2218 skb_set_tail_pointer(skb
, len
);
2221 static inline void skb_split_no_header(struct sk_buff
*skb
,
2222 struct sk_buff
* skb1
,
2223 const u32 len
, int pos
)
2226 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
2228 skb_shinfo(skb
)->nr_frags
= 0;
2229 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
2231 skb
->data_len
= len
- pos
;
2233 for (i
= 0; i
< nfrags
; i
++) {
2234 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
2236 if (pos
+ size
> len
) {
2237 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
2241 * We have two variants in this case:
2242 * 1. Move all the frag to the second
2243 * part, if it is possible. F.e.
2244 * this approach is mandatory for TUX,
2245 * where splitting is expensive.
2246 * 2. Split is accurately. We make this.
2248 skb_frag_ref(skb
, i
);
2249 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
2250 skb_frag_size_sub(&skb_shinfo(skb1
)->frags
[0], len
- pos
);
2251 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
], len
- pos
);
2252 skb_shinfo(skb
)->nr_frags
++;
2256 skb_shinfo(skb
)->nr_frags
++;
2259 skb_shinfo(skb1
)->nr_frags
= k
;
2263 * skb_split - Split fragmented skb to two parts at length len.
2264 * @skb: the buffer to split
2265 * @skb1: the buffer to receive the second part
2266 * @len: new length for skb
2268 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
2270 int pos
= skb_headlen(skb
);
2272 if (len
< pos
) /* Split line is inside header. */
2273 skb_split_inside_header(skb
, skb1
, len
, pos
);
2274 else /* Second chunk has no header, nothing to copy. */
2275 skb_split_no_header(skb
, skb1
, len
, pos
);
2277 EXPORT_SYMBOL(skb_split
);
2279 /* Shifting from/to a cloned skb is a no-go.
2281 * Caller cannot keep skb_shinfo related pointers past calling here!
2283 static int skb_prepare_for_shift(struct sk_buff
*skb
)
2285 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2289 * skb_shift - Shifts paged data partially from skb to another
2290 * @tgt: buffer into which tail data gets added
2291 * @skb: buffer from which the paged data comes from
2292 * @shiftlen: shift up to this many bytes
2294 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2295 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2296 * It's up to caller to free skb if everything was shifted.
2298 * If @tgt runs out of frags, the whole operation is aborted.
2300 * Skb cannot include anything else but paged data while tgt is allowed
2301 * to have non-paged data as well.
2303 * TODO: full sized shift could be optimized but that would need
2304 * specialized skb free'er to handle frags without up-to-date nr_frags.
2306 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
2308 int from
, to
, merge
, todo
;
2309 struct skb_frag_struct
*fragfrom
, *fragto
;
2311 BUG_ON(shiftlen
> skb
->len
);
2312 BUG_ON(skb_headlen(skb
)); /* Would corrupt stream */
2316 to
= skb_shinfo(tgt
)->nr_frags
;
2317 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2319 /* Actual merge is delayed until the point when we know we can
2320 * commit all, so that we don't have to undo partial changes
2323 !skb_can_coalesce(tgt
, to
, skb_frag_page(fragfrom
),
2324 fragfrom
->page_offset
)) {
2329 todo
-= skb_frag_size(fragfrom
);
2331 if (skb_prepare_for_shift(skb
) ||
2332 skb_prepare_for_shift(tgt
))
2335 /* All previous frag pointers might be stale! */
2336 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2337 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2339 skb_frag_size_add(fragto
, shiftlen
);
2340 skb_frag_size_sub(fragfrom
, shiftlen
);
2341 fragfrom
->page_offset
+= shiftlen
;
2349 /* Skip full, not-fitting skb to avoid expensive operations */
2350 if ((shiftlen
== skb
->len
) &&
2351 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
2354 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
2357 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
2358 if (to
== MAX_SKB_FRAGS
)
2361 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2362 fragto
= &skb_shinfo(tgt
)->frags
[to
];
2364 if (todo
>= skb_frag_size(fragfrom
)) {
2365 *fragto
= *fragfrom
;
2366 todo
-= skb_frag_size(fragfrom
);
2371 __skb_frag_ref(fragfrom
);
2372 fragto
->page
= fragfrom
->page
;
2373 fragto
->page_offset
= fragfrom
->page_offset
;
2374 skb_frag_size_set(fragto
, todo
);
2376 fragfrom
->page_offset
+= todo
;
2377 skb_frag_size_sub(fragfrom
, todo
);
2385 /* Ready to "commit" this state change to tgt */
2386 skb_shinfo(tgt
)->nr_frags
= to
;
2389 fragfrom
= &skb_shinfo(skb
)->frags
[0];
2390 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2392 skb_frag_size_add(fragto
, skb_frag_size(fragfrom
));
2393 __skb_frag_unref(fragfrom
);
2396 /* Reposition in the original skb */
2398 while (from
< skb_shinfo(skb
)->nr_frags
)
2399 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
2400 skb_shinfo(skb
)->nr_frags
= to
;
2402 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
2405 /* Most likely the tgt won't ever need its checksum anymore, skb on
2406 * the other hand might need it if it needs to be resent
2408 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
2409 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2411 /* Yak, is it really working this way? Some helper please? */
2412 skb
->len
-= shiftlen
;
2413 skb
->data_len
-= shiftlen
;
2414 skb
->truesize
-= shiftlen
;
2415 tgt
->len
+= shiftlen
;
2416 tgt
->data_len
+= shiftlen
;
2417 tgt
->truesize
+= shiftlen
;
2423 * skb_prepare_seq_read - Prepare a sequential read of skb data
2424 * @skb: the buffer to read
2425 * @from: lower offset of data to be read
2426 * @to: upper offset of data to be read
2427 * @st: state variable
2429 * Initializes the specified state variable. Must be called before
2430 * invoking skb_seq_read() for the first time.
2432 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
2433 unsigned int to
, struct skb_seq_state
*st
)
2435 st
->lower_offset
= from
;
2436 st
->upper_offset
= to
;
2437 st
->root_skb
= st
->cur_skb
= skb
;
2438 st
->frag_idx
= st
->stepped_offset
= 0;
2439 st
->frag_data
= NULL
;
2441 EXPORT_SYMBOL(skb_prepare_seq_read
);
2444 * skb_seq_read - Sequentially read skb data
2445 * @consumed: number of bytes consumed by the caller so far
2446 * @data: destination pointer for data to be returned
2447 * @st: state variable
2449 * Reads a block of skb data at &consumed relative to the
2450 * lower offset specified to skb_prepare_seq_read(). Assigns
2451 * the head of the data block to &data and returns the length
2452 * of the block or 0 if the end of the skb data or the upper
2453 * offset has been reached.
2455 * The caller is not required to consume all of the data
2456 * returned, i.e. &consumed is typically set to the number
2457 * of bytes already consumed and the next call to
2458 * skb_seq_read() will return the remaining part of the block.
2460 * Note 1: The size of each block of data returned can be arbitrary,
2461 * this limitation is the cost for zerocopy seqeuental
2462 * reads of potentially non linear data.
2464 * Note 2: Fragment lists within fragments are not implemented
2465 * at the moment, state->root_skb could be replaced with
2466 * a stack for this purpose.
2468 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
2469 struct skb_seq_state
*st
)
2471 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
2474 if (unlikely(abs_offset
>= st
->upper_offset
))
2478 block_limit
= skb_headlen(st
->cur_skb
) + st
->stepped_offset
;
2480 if (abs_offset
< block_limit
&& !st
->frag_data
) {
2481 *data
= st
->cur_skb
->data
+ (abs_offset
- st
->stepped_offset
);
2482 return block_limit
- abs_offset
;
2485 if (st
->frag_idx
== 0 && !st
->frag_data
)
2486 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
2488 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
2489 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
2490 block_limit
= skb_frag_size(frag
) + st
->stepped_offset
;
2492 if (abs_offset
< block_limit
) {
2494 st
->frag_data
= kmap_atomic(skb_frag_page(frag
));
2496 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
2497 (abs_offset
- st
->stepped_offset
);
2499 return block_limit
- abs_offset
;
2502 if (st
->frag_data
) {
2503 kunmap_atomic(st
->frag_data
);
2504 st
->frag_data
= NULL
;
2508 st
->stepped_offset
+= skb_frag_size(frag
);
2511 if (st
->frag_data
) {
2512 kunmap_atomic(st
->frag_data
);
2513 st
->frag_data
= NULL
;
2516 if (st
->root_skb
== st
->cur_skb
&& skb_has_frag_list(st
->root_skb
)) {
2517 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
2520 } else if (st
->cur_skb
->next
) {
2521 st
->cur_skb
= st
->cur_skb
->next
;
2528 EXPORT_SYMBOL(skb_seq_read
);
2531 * skb_abort_seq_read - Abort a sequential read of skb data
2532 * @st: state variable
2534 * Must be called if skb_seq_read() was not called until it
2537 void skb_abort_seq_read(struct skb_seq_state
*st
)
2540 kunmap_atomic(st
->frag_data
);
2542 EXPORT_SYMBOL(skb_abort_seq_read
);
2544 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2546 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
2547 struct ts_config
*conf
,
2548 struct ts_state
*state
)
2550 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
2553 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
2555 skb_abort_seq_read(TS_SKB_CB(state
));
2559 * skb_find_text - Find a text pattern in skb data
2560 * @skb: the buffer to look in
2561 * @from: search offset
2563 * @config: textsearch configuration
2564 * @state: uninitialized textsearch state variable
2566 * Finds a pattern in the skb data according to the specified
2567 * textsearch configuration. Use textsearch_next() to retrieve
2568 * subsequent occurrences of the pattern. Returns the offset
2569 * to the first occurrence or UINT_MAX if no match was found.
2571 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
2572 unsigned int to
, struct ts_config
*config
,
2573 struct ts_state
*state
)
2577 config
->get_next_block
= skb_ts_get_next_block
;
2578 config
->finish
= skb_ts_finish
;
2580 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
2582 ret
= textsearch_find(config
, state
);
2583 return (ret
<= to
- from
? ret
: UINT_MAX
);
2585 EXPORT_SYMBOL(skb_find_text
);
2588 * skb_append_datato_frags: - append the user data to a skb
2589 * @sk: sock structure
2590 * @skb: skb structure to be appened with user data.
2591 * @getfrag: call back function to be used for getting the user data
2592 * @from: pointer to user message iov
2593 * @length: length of the iov message
2595 * Description: This procedure append the user data in the fragment part
2596 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2598 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
2599 int (*getfrag
)(void *from
, char *to
, int offset
,
2600 int len
, int odd
, struct sk_buff
*skb
),
2601 void *from
, int length
)
2604 skb_frag_t
*frag
= NULL
;
2605 struct page
*page
= NULL
;
2611 /* Return error if we don't have space for new frag */
2612 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2613 if (frg_cnt
>= MAX_SKB_FRAGS
)
2616 /* allocate a new page for next frag */
2617 page
= alloc_pages(sk
->sk_allocation
, 0);
2619 /* If alloc_page fails just return failure and caller will
2620 * free previous allocated pages by doing kfree_skb()
2625 /* initialize the next frag */
2626 skb_fill_page_desc(skb
, frg_cnt
, page
, 0, 0);
2627 skb
->truesize
+= PAGE_SIZE
;
2628 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
2630 /* get the new initialized frag */
2631 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2632 frag
= &skb_shinfo(skb
)->frags
[frg_cnt
- 1];
2634 /* copy the user data to page */
2635 left
= PAGE_SIZE
- frag
->page_offset
;
2636 copy
= (length
> left
)? left
: length
;
2638 ret
= getfrag(from
, skb_frag_address(frag
) + skb_frag_size(frag
),
2639 offset
, copy
, 0, skb
);
2643 /* copy was successful so update the size parameters */
2644 skb_frag_size_add(frag
, copy
);
2646 skb
->data_len
+= copy
;
2650 } while (length
> 0);
2654 EXPORT_SYMBOL(skb_append_datato_frags
);
2657 * skb_pull_rcsum - pull skb and update receive checksum
2658 * @skb: buffer to update
2659 * @len: length of data pulled
2661 * This function performs an skb_pull on the packet and updates
2662 * the CHECKSUM_COMPLETE checksum. It should be used on
2663 * receive path processing instead of skb_pull unless you know
2664 * that the checksum difference is zero (e.g., a valid IP header)
2665 * or you are setting ip_summed to CHECKSUM_NONE.
2667 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
2669 BUG_ON(len
> skb
->len
);
2671 BUG_ON(skb
->len
< skb
->data_len
);
2672 skb_postpull_rcsum(skb
, skb
->data
, len
);
2673 return skb
->data
+= len
;
2675 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
2678 * skb_segment - Perform protocol segmentation on skb.
2679 * @skb: buffer to segment
2680 * @features: features for the output path (see dev->features)
2682 * This function performs segmentation on the given skb. It returns
2683 * a pointer to the first in a list of new skbs for the segments.
2684 * In case of error it returns ERR_PTR(err).
2686 struct sk_buff
*skb_segment(struct sk_buff
*skb
, netdev_features_t features
)
2688 struct sk_buff
*segs
= NULL
;
2689 struct sk_buff
*tail
= NULL
;
2690 struct sk_buff
*fskb
= skb_shinfo(skb
)->frag_list
;
2691 unsigned int mss
= skb_shinfo(skb
)->gso_size
;
2692 unsigned int doffset
= skb
->data
- skb_mac_header(skb
);
2693 unsigned int offset
= doffset
;
2694 unsigned int headroom
;
2696 int sg
= !!(features
& NETIF_F_SG
);
2697 int nfrags
= skb_shinfo(skb
)->nr_frags
;
2702 __skb_push(skb
, doffset
);
2703 headroom
= skb_headroom(skb
);
2704 pos
= skb_headlen(skb
);
2707 struct sk_buff
*nskb
;
2712 len
= skb
->len
- offset
;
2716 hsize
= skb_headlen(skb
) - offset
;
2719 if (hsize
> len
|| !sg
)
2722 if (!hsize
&& i
>= nfrags
) {
2723 BUG_ON(fskb
->len
!= len
);
2726 nskb
= skb_clone(fskb
, GFP_ATOMIC
);
2729 if (unlikely(!nskb
))
2732 hsize
= skb_end_pointer(nskb
) - nskb
->head
;
2733 if (skb_cow_head(nskb
, doffset
+ headroom
)) {
2738 nskb
->truesize
+= skb_end_pointer(nskb
) - nskb
->head
-
2740 skb_release_head_state(nskb
);
2741 __skb_push(nskb
, doffset
);
2743 nskb
= alloc_skb(hsize
+ doffset
+ headroom
,
2746 if (unlikely(!nskb
))
2749 skb_reserve(nskb
, headroom
);
2750 __skb_put(nskb
, doffset
);
2759 __copy_skb_header(nskb
, skb
);
2760 nskb
->mac_len
= skb
->mac_len
;
2762 /* nskb and skb might have different headroom */
2763 if (nskb
->ip_summed
== CHECKSUM_PARTIAL
)
2764 nskb
->csum_start
+= skb_headroom(nskb
) - headroom
;
2766 skb_reset_mac_header(nskb
);
2767 skb_set_network_header(nskb
, skb
->mac_len
);
2768 nskb
->transport_header
= (nskb
->network_header
+
2769 skb_network_header_len(skb
));
2770 skb_copy_from_linear_data(skb
, nskb
->data
, doffset
);
2772 if (fskb
!= skb_shinfo(skb
)->frag_list
)
2776 nskb
->ip_summed
= CHECKSUM_NONE
;
2777 nskb
->csum
= skb_copy_and_csum_bits(skb
, offset
,
2783 frag
= skb_shinfo(nskb
)->frags
;
2785 skb_copy_from_linear_data_offset(skb
, offset
,
2786 skb_put(nskb
, hsize
), hsize
);
2788 while (pos
< offset
+ len
&& i
< nfrags
) {
2789 *frag
= skb_shinfo(skb
)->frags
[i
];
2790 __skb_frag_ref(frag
);
2791 size
= skb_frag_size(frag
);
2794 frag
->page_offset
+= offset
- pos
;
2795 skb_frag_size_sub(frag
, offset
- pos
);
2798 skb_shinfo(nskb
)->nr_frags
++;
2800 if (pos
+ size
<= offset
+ len
) {
2804 skb_frag_size_sub(frag
, pos
+ size
- (offset
+ len
));
2811 if (pos
< offset
+ len
) {
2812 struct sk_buff
*fskb2
= fskb
;
2814 BUG_ON(pos
+ fskb
->len
!= offset
+ len
);
2820 fskb2
= skb_clone(fskb2
, GFP_ATOMIC
);
2826 SKB_FRAG_ASSERT(nskb
);
2827 skb_shinfo(nskb
)->frag_list
= fskb2
;
2831 nskb
->data_len
= len
- hsize
;
2832 nskb
->len
+= nskb
->data_len
;
2833 nskb
->truesize
+= nskb
->data_len
;
2834 } while ((offset
+= len
) < skb
->len
);
2839 while ((skb
= segs
)) {
2843 return ERR_PTR(err
);
2845 EXPORT_SYMBOL_GPL(skb_segment
);
2847 int skb_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2849 struct sk_buff
*p
= *head
;
2850 struct sk_buff
*nskb
;
2851 struct skb_shared_info
*skbinfo
= skb_shinfo(skb
);
2852 struct skb_shared_info
*pinfo
= skb_shinfo(p
);
2853 unsigned int headroom
;
2854 unsigned int len
= skb_gro_len(skb
);
2855 unsigned int offset
= skb_gro_offset(skb
);
2856 unsigned int headlen
= skb_headlen(skb
);
2858 if (p
->len
+ len
>= 65536)
2861 if (pinfo
->frag_list
)
2863 else if (headlen
<= offset
) {
2866 int i
= skbinfo
->nr_frags
;
2867 int nr_frags
= pinfo
->nr_frags
+ i
;
2871 if (nr_frags
> MAX_SKB_FRAGS
)
2874 pinfo
->nr_frags
= nr_frags
;
2875 skbinfo
->nr_frags
= 0;
2877 frag
= pinfo
->frags
+ nr_frags
;
2878 frag2
= skbinfo
->frags
+ i
;
2883 frag
->page_offset
+= offset
;
2884 skb_frag_size_sub(frag
, offset
);
2886 skb
->truesize
-= skb
->data_len
;
2887 skb
->len
-= skb
->data_len
;
2890 NAPI_GRO_CB(skb
)->free
= 1;
2892 } else if (skb_gro_len(p
) != pinfo
->gso_size
)
2895 headroom
= skb_headroom(p
);
2896 nskb
= alloc_skb(headroom
+ skb_gro_offset(p
), GFP_ATOMIC
);
2897 if (unlikely(!nskb
))
2900 __copy_skb_header(nskb
, p
);
2901 nskb
->mac_len
= p
->mac_len
;
2903 skb_reserve(nskb
, headroom
);
2904 __skb_put(nskb
, skb_gro_offset(p
));
2906 skb_set_mac_header(nskb
, skb_mac_header(p
) - p
->data
);
2907 skb_set_network_header(nskb
, skb_network_offset(p
));
2908 skb_set_transport_header(nskb
, skb_transport_offset(p
));
2910 __skb_pull(p
, skb_gro_offset(p
));
2911 memcpy(skb_mac_header(nskb
), skb_mac_header(p
),
2912 p
->data
- skb_mac_header(p
));
2914 *NAPI_GRO_CB(nskb
) = *NAPI_GRO_CB(p
);
2915 skb_shinfo(nskb
)->frag_list
= p
;
2916 skb_shinfo(nskb
)->gso_size
= pinfo
->gso_size
;
2917 pinfo
->gso_size
= 0;
2918 skb_header_release(p
);
2921 nskb
->data_len
+= p
->len
;
2922 nskb
->truesize
+= p
->truesize
;
2923 nskb
->len
+= p
->len
;
2926 nskb
->next
= p
->next
;
2932 p
->truesize
+= skb
->truesize
- len
;
2933 if (offset
> headlen
) {
2934 unsigned int eat
= offset
- headlen
;
2936 skbinfo
->frags
[0].page_offset
+= eat
;
2937 skb_frag_size_sub(&skbinfo
->frags
[0], eat
);
2938 skb
->data_len
-= eat
;
2943 __skb_pull(skb
, offset
);
2945 p
->prev
->next
= skb
;
2947 skb_header_release(skb
);
2950 NAPI_GRO_CB(p
)->count
++;
2955 NAPI_GRO_CB(skb
)->same_flow
= 1;
2958 EXPORT_SYMBOL_GPL(skb_gro_receive
);
2960 void __init
skb_init(void)
2962 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
2963 sizeof(struct sk_buff
),
2965 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2967 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
2968 (2*sizeof(struct sk_buff
)) +
2971 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2976 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2977 * @skb: Socket buffer containing the buffers to be mapped
2978 * @sg: The scatter-gather list to map into
2979 * @offset: The offset into the buffer's contents to start mapping
2980 * @len: Length of buffer space to be mapped
2982 * Fill the specified scatter-gather list with mappings/pointers into a
2983 * region of the buffer space attached to a socket buffer.
2986 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2988 int start
= skb_headlen(skb
);
2989 int i
, copy
= start
- offset
;
2990 struct sk_buff
*frag_iter
;
2996 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
2998 if ((len
-= copy
) == 0)
3003 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
3006 WARN_ON(start
> offset
+ len
);
3008 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
3009 if ((copy
= end
- offset
) > 0) {
3010 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
3014 sg_set_page(&sg
[elt
], skb_frag_page(frag
), copy
,
3015 frag
->page_offset
+offset
-start
);
3024 skb_walk_frags(skb
, frag_iter
) {
3027 WARN_ON(start
> offset
+ len
);
3029 end
= start
+ frag_iter
->len
;
3030 if ((copy
= end
- offset
) > 0) {
3033 elt
+= __skb_to_sgvec(frag_iter
, sg
+elt
, offset
- start
,
3035 if ((len
-= copy
) == 0)
3045 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
3047 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
);
3049 sg_mark_end(&sg
[nsg
- 1]);
3053 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
3056 * skb_cow_data - Check that a socket buffer's data buffers are writable
3057 * @skb: The socket buffer to check.
3058 * @tailbits: Amount of trailing space to be added
3059 * @trailer: Returned pointer to the skb where the @tailbits space begins
3061 * Make sure that the data buffers attached to a socket buffer are
3062 * writable. If they are not, private copies are made of the data buffers
3063 * and the socket buffer is set to use these instead.
3065 * If @tailbits is given, make sure that there is space to write @tailbits
3066 * bytes of data beyond current end of socket buffer. @trailer will be
3067 * set to point to the skb in which this space begins.
3069 * The number of scatterlist elements required to completely map the
3070 * COW'd and extended socket buffer will be returned.
3072 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
3076 struct sk_buff
*skb1
, **skb_p
;
3078 /* If skb is cloned or its head is paged, reallocate
3079 * head pulling out all the pages (pages are considered not writable
3080 * at the moment even if they are anonymous).
3082 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
3083 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
3086 /* Easy case. Most of packets will go this way. */
3087 if (!skb_has_frag_list(skb
)) {
3088 /* A little of trouble, not enough of space for trailer.
3089 * This should not happen, when stack is tuned to generate
3090 * good frames. OK, on miss we reallocate and reserve even more
3091 * space, 128 bytes is fair. */
3093 if (skb_tailroom(skb
) < tailbits
&&
3094 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
3102 /* Misery. We are in troubles, going to mincer fragments... */
3105 skb_p
= &skb_shinfo(skb
)->frag_list
;
3108 while ((skb1
= *skb_p
) != NULL
) {
3111 /* The fragment is partially pulled by someone,
3112 * this can happen on input. Copy it and everything
3115 if (skb_shared(skb1
))
3118 /* If the skb is the last, worry about trailer. */
3120 if (skb1
->next
== NULL
&& tailbits
) {
3121 if (skb_shinfo(skb1
)->nr_frags
||
3122 skb_has_frag_list(skb1
) ||
3123 skb_tailroom(skb1
) < tailbits
)
3124 ntail
= tailbits
+ 128;
3130 skb_shinfo(skb1
)->nr_frags
||
3131 skb_has_frag_list(skb1
)) {
3132 struct sk_buff
*skb2
;
3134 /* Fuck, we are miserable poor guys... */
3136 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
3138 skb2
= skb_copy_expand(skb1
,
3142 if (unlikely(skb2
== NULL
))
3146 skb_set_owner_w(skb2
, skb1
->sk
);
3148 /* Looking around. Are we still alive?
3149 * OK, link new skb, drop old one */
3151 skb2
->next
= skb1
->next
;
3158 skb_p
= &skb1
->next
;
3163 EXPORT_SYMBOL_GPL(skb_cow_data
);
3165 static void sock_rmem_free(struct sk_buff
*skb
)
3167 struct sock
*sk
= skb
->sk
;
3169 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
3173 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3175 int sock_queue_err_skb(struct sock
*sk
, struct sk_buff
*skb
)
3179 if (atomic_read(&sk
->sk_rmem_alloc
) + skb
->truesize
>=
3180 (unsigned int)sk
->sk_rcvbuf
)
3185 skb
->destructor
= sock_rmem_free
;
3186 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
3188 /* before exiting rcu section, make sure dst is refcounted */
3191 skb_queue_tail(&sk
->sk_error_queue
, skb
);
3192 if (!sock_flag(sk
, SOCK_DEAD
))
3193 sk
->sk_data_ready(sk
, len
);
3196 EXPORT_SYMBOL(sock_queue_err_skb
);
3198 void skb_tstamp_tx(struct sk_buff
*orig_skb
,
3199 struct skb_shared_hwtstamps
*hwtstamps
)
3201 struct sock
*sk
= orig_skb
->sk
;
3202 struct sock_exterr_skb
*serr
;
3203 struct sk_buff
*skb
;
3209 skb
= skb_clone(orig_skb
, GFP_ATOMIC
);
3214 *skb_hwtstamps(skb
) =
3218 * no hardware time stamps available,
3219 * so keep the shared tx_flags and only
3220 * store software time stamp
3222 skb
->tstamp
= ktime_get_real();
3225 serr
= SKB_EXT_ERR(skb
);
3226 memset(serr
, 0, sizeof(*serr
));
3227 serr
->ee
.ee_errno
= ENOMSG
;
3228 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TIMESTAMPING
;
3230 err
= sock_queue_err_skb(sk
, skb
);
3235 EXPORT_SYMBOL_GPL(skb_tstamp_tx
);
3237 void skb_complete_wifi_ack(struct sk_buff
*skb
, bool acked
)
3239 struct sock
*sk
= skb
->sk
;
3240 struct sock_exterr_skb
*serr
;
3243 skb
->wifi_acked_valid
= 1;
3244 skb
->wifi_acked
= acked
;
3246 serr
= SKB_EXT_ERR(skb
);
3247 memset(serr
, 0, sizeof(*serr
));
3248 serr
->ee
.ee_errno
= ENOMSG
;
3249 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TXSTATUS
;
3251 err
= sock_queue_err_skb(sk
, skb
);
3255 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack
);
3259 * skb_partial_csum_set - set up and verify partial csum values for packet
3260 * @skb: the skb to set
3261 * @start: the number of bytes after skb->data to start checksumming.
3262 * @off: the offset from start to place the checksum.
3264 * For untrusted partially-checksummed packets, we need to make sure the values
3265 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3267 * This function checks and sets those values and skb->ip_summed: if this
3268 * returns false you should drop the packet.
3270 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
3272 if (unlikely(start
> skb_headlen(skb
)) ||
3273 unlikely((int)start
+ off
> skb_headlen(skb
) - 2)) {
3274 if (net_ratelimit())
3276 "bad partial csum: csum=%u/%u len=%u\n",
3277 start
, off
, skb_headlen(skb
));
3280 skb
->ip_summed
= CHECKSUM_PARTIAL
;
3281 skb
->csum_start
= skb_headroom(skb
) + start
;
3282 skb
->csum_offset
= off
;
3285 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);
3287 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
3289 if (net_ratelimit())
3290 pr_warning("%s: received packets cannot be forwarded"
3291 " while LRO is enabled\n", skb
->dev
->name
);
3293 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);