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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
41 #include <linux/module.h>
42 #include <linux/types.h>
43 #include <linux/kernel.h>
44 #include <linux/kmemcheck.h>
46 #include <linux/interrupt.h>
48 #include <linux/inet.h>
49 #include <linux/slab.h>
50 #include <linux/netdevice.h>
51 #ifdef CONFIG_NET_CLS_ACT
52 #include <net/pkt_sched.h>
54 #include <linux/string.h>
55 #include <linux/skbuff.h>
56 #include <linux/splice.h>
57 #include <linux/cache.h>
58 #include <linux/rtnetlink.h>
59 #include <linux/init.h>
60 #include <linux/scatterlist.h>
61 #include <linux/errqueue.h>
62 #include <linux/prefetch.h>
64 #include <net/protocol.h>
67 #include <net/checksum.h>
70 #include <asm/uaccess.h>
71 #include <trace/events/skb.h>
72 #include <linux/highmem.h>
74 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 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
124 __func__
, here
, skb
->len
, sz
, skb
->head
, skb
->data
,
125 (unsigned long)skb
->tail
, (unsigned long)skb
->end
,
126 skb
->dev
? skb
->dev
->name
: "<NULL>");
131 * skb_under_panic - private function
136 * Out of line support code for skb_push(). Not user callable.
139 static void skb_under_panic(struct sk_buff
*skb
, int sz
, void *here
)
141 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
142 __func__
, 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
248 * @frag_size: size of fragment, or 0 if head was kmalloced
250 * Allocate a new &sk_buff. Caller provides space holding head and
251 * skb_shared_info. @data must have been allocated by kmalloc()
252 * The return is the new skb buffer.
253 * On a failure the return is %NULL, and @data is not freed.
255 * Before IO, driver allocates only data buffer where NIC put incoming frame
256 * Driver should add room at head (NET_SKB_PAD) and
257 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
258 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
259 * before giving packet to stack.
260 * RX rings only contains data buffers, not full skbs.
262 struct sk_buff
*build_skb(void *data
, unsigned int frag_size
)
264 struct skb_shared_info
*shinfo
;
266 unsigned int size
= frag_size
? : ksize(data
);
268 skb
= kmem_cache_alloc(skbuff_head_cache
, GFP_ATOMIC
);
272 size
-= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
274 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
275 skb
->truesize
= SKB_TRUESIZE(size
);
276 skb
->head_frag
= frag_size
!= 0;
277 atomic_set(&skb
->users
, 1);
280 skb_reset_tail_pointer(skb
);
281 skb
->end
= skb
->tail
+ size
;
282 #ifdef NET_SKBUFF_DATA_USES_OFFSET
283 skb
->mac_header
= ~0U;
286 /* make sure we initialize shinfo sequentially */
287 shinfo
= skb_shinfo(skb
);
288 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
289 atomic_set(&shinfo
->dataref
, 1);
290 kmemcheck_annotate_variable(shinfo
->destructor_arg
);
294 EXPORT_SYMBOL(build_skb
);
296 struct netdev_alloc_cache
{
300 static DEFINE_PER_CPU(struct netdev_alloc_cache
, netdev_alloc_cache
);
303 * netdev_alloc_frag - allocate a page fragment
304 * @fragsz: fragment size
306 * Allocates a frag from a page for receive buffer.
307 * Uses GFP_ATOMIC allocations.
309 void *netdev_alloc_frag(unsigned int fragsz
)
311 struct netdev_alloc_cache
*nc
;
315 local_irq_save(flags
);
316 nc
= &__get_cpu_var(netdev_alloc_cache
);
317 if (unlikely(!nc
->page
)) {
319 nc
->page
= alloc_page(GFP_ATOMIC
| __GFP_COLD
);
322 if (likely(nc
->page
)) {
323 if (nc
->offset
+ fragsz
> PAGE_SIZE
) {
327 data
= page_address(nc
->page
) + nc
->offset
;
328 nc
->offset
+= fragsz
;
331 local_irq_restore(flags
);
334 EXPORT_SYMBOL(netdev_alloc_frag
);
337 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
338 * @dev: network device to receive on
339 * @length: length to allocate
340 * @gfp_mask: get_free_pages mask, passed to alloc_skb
342 * Allocate a new &sk_buff and assign it a usage count of one. The
343 * buffer has unspecified headroom built in. Users should allocate
344 * the headroom they think they need without accounting for the
345 * built in space. The built in space is used for optimisations.
347 * %NULL is returned if there is no free memory.
349 struct sk_buff
*__netdev_alloc_skb(struct net_device
*dev
,
350 unsigned int length
, gfp_t gfp_mask
)
352 struct sk_buff
*skb
= NULL
;
353 unsigned int fragsz
= SKB_DATA_ALIGN(length
+ NET_SKB_PAD
) +
354 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
356 if (fragsz
<= PAGE_SIZE
&& !(gfp_mask
& __GFP_WAIT
)) {
357 void *data
= netdev_alloc_frag(fragsz
);
360 skb
= build_skb(data
, fragsz
);
362 put_page(virt_to_head_page(data
));
365 skb
= __alloc_skb(length
+ NET_SKB_PAD
, gfp_mask
, 0, NUMA_NO_NODE
);
368 skb_reserve(skb
, NET_SKB_PAD
);
373 EXPORT_SYMBOL(__netdev_alloc_skb
);
375 void skb_add_rx_frag(struct sk_buff
*skb
, int i
, struct page
*page
, int off
,
376 int size
, unsigned int truesize
)
378 skb_fill_page_desc(skb
, i
, page
, off
, size
);
380 skb
->data_len
+= size
;
381 skb
->truesize
+= truesize
;
383 EXPORT_SYMBOL(skb_add_rx_frag
);
385 static void skb_drop_list(struct sk_buff
**listp
)
387 struct sk_buff
*list
= *listp
;
392 struct sk_buff
*this = list
;
398 static inline void skb_drop_fraglist(struct sk_buff
*skb
)
400 skb_drop_list(&skb_shinfo(skb
)->frag_list
);
403 static void skb_clone_fraglist(struct sk_buff
*skb
)
405 struct sk_buff
*list
;
407 skb_walk_frags(skb
, list
)
411 static void skb_free_head(struct sk_buff
*skb
)
414 put_page(virt_to_head_page(skb
->head
));
419 static void skb_release_data(struct sk_buff
*skb
)
422 !atomic_sub_return(skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1,
423 &skb_shinfo(skb
)->dataref
)) {
424 if (skb_shinfo(skb
)->nr_frags
) {
426 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
427 skb_frag_unref(skb
, i
);
431 * If skb buf is from userspace, we need to notify the caller
432 * the lower device DMA has done;
434 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
435 struct ubuf_info
*uarg
;
437 uarg
= skb_shinfo(skb
)->destructor_arg
;
439 uarg
->callback(uarg
);
442 if (skb_has_frag_list(skb
))
443 skb_drop_fraglist(skb
);
450 * Free an skbuff by memory without cleaning the state.
452 static void kfree_skbmem(struct sk_buff
*skb
)
454 struct sk_buff
*other
;
455 atomic_t
*fclone_ref
;
457 switch (skb
->fclone
) {
458 case SKB_FCLONE_UNAVAILABLE
:
459 kmem_cache_free(skbuff_head_cache
, skb
);
462 case SKB_FCLONE_ORIG
:
463 fclone_ref
= (atomic_t
*) (skb
+ 2);
464 if (atomic_dec_and_test(fclone_ref
))
465 kmem_cache_free(skbuff_fclone_cache
, skb
);
468 case SKB_FCLONE_CLONE
:
469 fclone_ref
= (atomic_t
*) (skb
+ 1);
472 /* The clone portion is available for
473 * fast-cloning again.
475 skb
->fclone
= SKB_FCLONE_UNAVAILABLE
;
477 if (atomic_dec_and_test(fclone_ref
))
478 kmem_cache_free(skbuff_fclone_cache
, other
);
483 static void skb_release_head_state(struct sk_buff
*skb
)
487 secpath_put(skb
->sp
);
489 if (skb
->destructor
) {
491 skb
->destructor(skb
);
493 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
494 nf_conntrack_put(skb
->nfct
);
496 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
497 nf_conntrack_put_reasm(skb
->nfct_reasm
);
499 #ifdef CONFIG_BRIDGE_NETFILTER
500 nf_bridge_put(skb
->nf_bridge
);
502 /* XXX: IS this still necessary? - JHS */
503 #ifdef CONFIG_NET_SCHED
505 #ifdef CONFIG_NET_CLS_ACT
511 /* Free everything but the sk_buff shell. */
512 static void skb_release_all(struct sk_buff
*skb
)
514 skb_release_head_state(skb
);
515 skb_release_data(skb
);
519 * __kfree_skb - private function
522 * Free an sk_buff. Release anything attached to the buffer.
523 * Clean the state. This is an internal helper function. Users should
524 * always call kfree_skb
527 void __kfree_skb(struct sk_buff
*skb
)
529 skb_release_all(skb
);
532 EXPORT_SYMBOL(__kfree_skb
);
535 * kfree_skb - free an sk_buff
536 * @skb: buffer to free
538 * Drop a reference to the buffer and free it if the usage count has
541 void kfree_skb(struct sk_buff
*skb
)
545 if (likely(atomic_read(&skb
->users
) == 1))
547 else if (likely(!atomic_dec_and_test(&skb
->users
)))
549 trace_kfree_skb(skb
, __builtin_return_address(0));
552 EXPORT_SYMBOL(kfree_skb
);
555 * consume_skb - free an skbuff
556 * @skb: buffer to free
558 * Drop a ref to the buffer and free it if the usage count has hit zero
559 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
560 * is being dropped after a failure and notes that
562 void consume_skb(struct sk_buff
*skb
)
566 if (likely(atomic_read(&skb
->users
) == 1))
568 else if (likely(!atomic_dec_and_test(&skb
->users
)))
570 trace_consume_skb(skb
);
573 EXPORT_SYMBOL(consume_skb
);
576 * skb_recycle - clean up an skb for reuse
579 * Recycles the skb to be reused as a receive buffer. This
580 * function does any necessary reference count dropping, and
581 * cleans up the skbuff as if it just came from __alloc_skb().
583 void skb_recycle(struct sk_buff
*skb
)
585 struct skb_shared_info
*shinfo
;
587 skb_release_head_state(skb
);
589 shinfo
= skb_shinfo(skb
);
590 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
591 atomic_set(&shinfo
->dataref
, 1);
593 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
594 skb
->data
= skb
->head
+ NET_SKB_PAD
;
595 skb_reset_tail_pointer(skb
);
597 EXPORT_SYMBOL(skb_recycle
);
600 * skb_recycle_check - check if skb can be reused for receive
602 * @skb_size: minimum receive buffer size
604 * Checks that the skb passed in is not shared or cloned, and
605 * that it is linear and its head portion at least as large as
606 * skb_size so that it can be recycled as a receive buffer.
607 * If these conditions are met, this function does any necessary
608 * reference count dropping and cleans up the skbuff as if it
609 * just came from __alloc_skb().
611 bool skb_recycle_check(struct sk_buff
*skb
, int skb_size
)
613 if (!skb_is_recycleable(skb
, skb_size
))
620 EXPORT_SYMBOL(skb_recycle_check
);
622 static void __copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
624 new->tstamp
= old
->tstamp
;
626 new->transport_header
= old
->transport_header
;
627 new->network_header
= old
->network_header
;
628 new->mac_header
= old
->mac_header
;
629 skb_dst_copy(new, old
);
630 new->rxhash
= old
->rxhash
;
631 new->ooo_okay
= old
->ooo_okay
;
632 new->l4_rxhash
= old
->l4_rxhash
;
633 new->no_fcs
= old
->no_fcs
;
635 new->sp
= secpath_get(old
->sp
);
637 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
638 new->csum
= old
->csum
;
639 new->local_df
= old
->local_df
;
640 new->pkt_type
= old
->pkt_type
;
641 new->ip_summed
= old
->ip_summed
;
642 skb_copy_queue_mapping(new, old
);
643 new->priority
= old
->priority
;
644 #if IS_ENABLED(CONFIG_IP_VS)
645 new->ipvs_property
= old
->ipvs_property
;
647 new->protocol
= old
->protocol
;
648 new->mark
= old
->mark
;
649 new->skb_iif
= old
->skb_iif
;
651 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
652 new->nf_trace
= old
->nf_trace
;
654 #ifdef CONFIG_NET_SCHED
655 new->tc_index
= old
->tc_index
;
656 #ifdef CONFIG_NET_CLS_ACT
657 new->tc_verd
= old
->tc_verd
;
660 new->vlan_tci
= old
->vlan_tci
;
662 skb_copy_secmark(new, old
);
666 * You should not add any new code to this function. Add it to
667 * __copy_skb_header above instead.
669 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
671 #define C(x) n->x = skb->x
673 n
->next
= n
->prev
= NULL
;
675 __copy_skb_header(n
, skb
);
680 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
683 n
->destructor
= NULL
;
690 atomic_set(&n
->users
, 1);
692 atomic_inc(&(skb_shinfo(skb
)->dataref
));
700 * skb_morph - morph one skb into another
701 * @dst: the skb to receive the contents
702 * @src: the skb to supply the contents
704 * This is identical to skb_clone except that the target skb is
705 * supplied by the user.
707 * The target skb is returned upon exit.
709 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
711 skb_release_all(dst
);
712 return __skb_clone(dst
, src
);
714 EXPORT_SYMBOL_GPL(skb_morph
);
716 /* skb_copy_ubufs - copy userspace skb frags buffers to kernel
717 * @skb: the skb to modify
718 * @gfp_mask: allocation priority
720 * This must be called on SKBTX_DEV_ZEROCOPY skb.
721 * It will copy all frags into kernel and drop the reference
722 * to userspace pages.
724 * If this function is called from an interrupt gfp_mask() must be
727 * Returns 0 on success or a negative error code on failure
728 * to allocate kernel memory to copy to.
730 int skb_copy_ubufs(struct sk_buff
*skb
, gfp_t gfp_mask
)
733 int num_frags
= skb_shinfo(skb
)->nr_frags
;
734 struct page
*page
, *head
= NULL
;
735 struct ubuf_info
*uarg
= skb_shinfo(skb
)->destructor_arg
;
737 for (i
= 0; i
< num_frags
; i
++) {
739 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
741 page
= alloc_page(GFP_ATOMIC
);
744 struct page
*next
= (struct page
*)head
->private;
750 vaddr
= kmap_atomic(skb_frag_page(f
));
751 memcpy(page_address(page
),
752 vaddr
+ f
->page_offset
, skb_frag_size(f
));
753 kunmap_atomic(vaddr
);
754 page
->private = (unsigned long)head
;
758 /* skb frags release userspace buffers */
759 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
760 skb_frag_unref(skb
, i
);
762 uarg
->callback(uarg
);
764 /* skb frags point to kernel buffers */
765 for (i
= skb_shinfo(skb
)->nr_frags
; i
> 0; i
--) {
766 __skb_fill_page_desc(skb
, i
-1, head
, 0,
767 skb_shinfo(skb
)->frags
[i
- 1].size
);
768 head
= (struct page
*)head
->private;
771 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
777 * skb_clone - duplicate an sk_buff
778 * @skb: buffer to clone
779 * @gfp_mask: allocation priority
781 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
782 * copies share the same packet data but not structure. The new
783 * buffer has a reference count of 1. If the allocation fails the
784 * function returns %NULL otherwise the new buffer is returned.
786 * If this function is called from an interrupt gfp_mask() must be
790 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
794 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
795 if (skb_copy_ubufs(skb
, gfp_mask
))
800 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
801 n
->fclone
== SKB_FCLONE_UNAVAILABLE
) {
802 atomic_t
*fclone_ref
= (atomic_t
*) (n
+ 1);
803 n
->fclone
= SKB_FCLONE_CLONE
;
804 atomic_inc(fclone_ref
);
806 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
810 kmemcheck_annotate_bitfield(n
, flags1
);
811 kmemcheck_annotate_bitfield(n
, flags2
);
812 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
815 return __skb_clone(n
, skb
);
817 EXPORT_SYMBOL(skb_clone
);
819 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
821 #ifndef NET_SKBUFF_DATA_USES_OFFSET
823 * Shift between the two data areas in bytes
825 unsigned long offset
= new->data
- old
->data
;
828 __copy_skb_header(new, old
);
830 #ifndef NET_SKBUFF_DATA_USES_OFFSET
831 /* {transport,network,mac}_header are relative to skb->head */
832 new->transport_header
+= offset
;
833 new->network_header
+= offset
;
834 if (skb_mac_header_was_set(new))
835 new->mac_header
+= offset
;
837 skb_shinfo(new)->gso_size
= skb_shinfo(old
)->gso_size
;
838 skb_shinfo(new)->gso_segs
= skb_shinfo(old
)->gso_segs
;
839 skb_shinfo(new)->gso_type
= skb_shinfo(old
)->gso_type
;
843 * skb_copy - create private copy of an sk_buff
844 * @skb: buffer to copy
845 * @gfp_mask: allocation priority
847 * Make a copy of both an &sk_buff and its data. This is used when the
848 * caller wishes to modify the data and needs a private copy of the
849 * data to alter. Returns %NULL on failure or the pointer to the buffer
850 * on success. The returned buffer has a reference count of 1.
852 * As by-product this function converts non-linear &sk_buff to linear
853 * one, so that &sk_buff becomes completely private and caller is allowed
854 * to modify all the data of returned buffer. This means that this
855 * function is not recommended for use in circumstances when only
856 * header is going to be modified. Use pskb_copy() instead.
859 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
861 int headerlen
= skb_headroom(skb
);
862 unsigned int size
= skb_end_offset(skb
) + skb
->data_len
;
863 struct sk_buff
*n
= alloc_skb(size
, gfp_mask
);
868 /* Set the data pointer */
869 skb_reserve(n
, headerlen
);
870 /* Set the tail pointer and length */
871 skb_put(n
, skb
->len
);
873 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
876 copy_skb_header(n
, skb
);
879 EXPORT_SYMBOL(skb_copy
);
882 * __pskb_copy - create copy of an sk_buff with private head.
883 * @skb: buffer to copy
884 * @headroom: headroom of new skb
885 * @gfp_mask: allocation priority
887 * Make a copy of both an &sk_buff and part of its data, located
888 * in header. Fragmented data remain shared. This is used when
889 * the caller wishes to modify only header of &sk_buff and needs
890 * private copy of the header to alter. Returns %NULL on failure
891 * or the pointer to the buffer on success.
892 * The returned buffer has a reference count of 1.
895 struct sk_buff
*__pskb_copy(struct sk_buff
*skb
, int headroom
, gfp_t gfp_mask
)
897 unsigned int size
= skb_headlen(skb
) + headroom
;
898 struct sk_buff
*n
= alloc_skb(size
, gfp_mask
);
903 /* Set the data pointer */
904 skb_reserve(n
, headroom
);
905 /* Set the tail pointer and length */
906 skb_put(n
, skb_headlen(skb
));
908 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
910 n
->truesize
+= skb
->data_len
;
911 n
->data_len
= skb
->data_len
;
914 if (skb_shinfo(skb
)->nr_frags
) {
917 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
918 if (skb_copy_ubufs(skb
, gfp_mask
)) {
924 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
925 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
926 skb_frag_ref(skb
, i
);
928 skb_shinfo(n
)->nr_frags
= i
;
931 if (skb_has_frag_list(skb
)) {
932 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
933 skb_clone_fraglist(n
);
936 copy_skb_header(n
, skb
);
940 EXPORT_SYMBOL(__pskb_copy
);
943 * pskb_expand_head - reallocate header of &sk_buff
944 * @skb: buffer to reallocate
945 * @nhead: room to add at head
946 * @ntail: room to add at tail
947 * @gfp_mask: allocation priority
949 * Expands (or creates identical copy, if &nhead and &ntail are zero)
950 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
951 * reference count of 1. Returns zero in the case of success or error,
952 * if expansion failed. In the last case, &sk_buff is not changed.
954 * All the pointers pointing into skb header may change and must be
955 * reloaded after call to this function.
958 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
963 int size
= nhead
+ skb_end_offset(skb
) + ntail
;
971 size
= SKB_DATA_ALIGN(size
);
973 data
= kmalloc(size
+ SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)),
977 size
= SKB_WITH_OVERHEAD(ksize(data
));
979 /* Copy only real data... and, alas, header. This should be
980 * optimized for the cases when header is void.
982 memcpy(data
+ nhead
, skb
->head
, skb_tail_pointer(skb
) - skb
->head
);
984 memcpy((struct skb_shared_info
*)(data
+ size
),
986 offsetof(struct skb_shared_info
, frags
[skb_shinfo(skb
)->nr_frags
]));
989 * if shinfo is shared we must drop the old head gracefully, but if it
990 * is not we can just drop the old head and let the existing refcount
991 * be since all we did is relocate the values
993 if (skb_cloned(skb
)) {
994 /* copy this zero copy skb frags */
995 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
996 if (skb_copy_ubufs(skb
, gfp_mask
))
999 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1000 skb_frag_ref(skb
, i
);
1002 if (skb_has_frag_list(skb
))
1003 skb_clone_fraglist(skb
);
1005 skb_release_data(skb
);
1009 off
= (data
+ nhead
) - skb
->head
;
1014 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1018 skb
->end
= skb
->head
+ size
;
1020 /* {transport,network,mac}_header and tail are relative to skb->head */
1022 skb
->transport_header
+= off
;
1023 skb
->network_header
+= off
;
1024 if (skb_mac_header_was_set(skb
))
1025 skb
->mac_header
+= off
;
1026 /* Only adjust this if it actually is csum_start rather than csum */
1027 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1028 skb
->csum_start
+= nhead
;
1032 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
1040 EXPORT_SYMBOL(pskb_expand_head
);
1042 /* Make private copy of skb with writable head and some headroom */
1044 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
1046 struct sk_buff
*skb2
;
1047 int delta
= headroom
- skb_headroom(skb
);
1050 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
1052 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1053 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
1061 EXPORT_SYMBOL(skb_realloc_headroom
);
1064 * skb_copy_expand - copy and expand sk_buff
1065 * @skb: buffer to copy
1066 * @newheadroom: new free bytes at head
1067 * @newtailroom: new free bytes at tail
1068 * @gfp_mask: allocation priority
1070 * Make a copy of both an &sk_buff and its data and while doing so
1071 * allocate additional space.
1073 * This is used when the caller wishes to modify the data and needs a
1074 * private copy of the data to alter as well as more space for new fields.
1075 * Returns %NULL on failure or the pointer to the buffer
1076 * on success. The returned buffer has a reference count of 1.
1078 * You must pass %GFP_ATOMIC as the allocation priority if this function
1079 * is called from an interrupt.
1081 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
1082 int newheadroom
, int newtailroom
,
1086 * Allocate the copy buffer
1088 struct sk_buff
*n
= alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
1090 int oldheadroom
= skb_headroom(skb
);
1091 int head_copy_len
, head_copy_off
;
1097 skb_reserve(n
, newheadroom
);
1099 /* Set the tail pointer and length */
1100 skb_put(n
, skb
->len
);
1102 head_copy_len
= oldheadroom
;
1104 if (newheadroom
<= head_copy_len
)
1105 head_copy_len
= newheadroom
;
1107 head_copy_off
= newheadroom
- head_copy_len
;
1109 /* Copy the linear header and data. */
1110 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
1111 skb
->len
+ head_copy_len
))
1114 copy_skb_header(n
, skb
);
1116 off
= newheadroom
- oldheadroom
;
1117 if (n
->ip_summed
== CHECKSUM_PARTIAL
)
1118 n
->csum_start
+= off
;
1119 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1120 n
->transport_header
+= off
;
1121 n
->network_header
+= off
;
1122 if (skb_mac_header_was_set(skb
))
1123 n
->mac_header
+= off
;
1128 EXPORT_SYMBOL(skb_copy_expand
);
1131 * skb_pad - zero pad the tail of an skb
1132 * @skb: buffer to pad
1133 * @pad: space to pad
1135 * Ensure that a buffer is followed by a padding area that is zero
1136 * filled. Used by network drivers which may DMA or transfer data
1137 * beyond the buffer end onto the wire.
1139 * May return error in out of memory cases. The skb is freed on error.
1142 int skb_pad(struct sk_buff
*skb
, int pad
)
1147 /* If the skbuff is non linear tailroom is always zero.. */
1148 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
1149 memset(skb
->data
+skb
->len
, 0, pad
);
1153 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
1154 if (likely(skb_cloned(skb
) || ntail
> 0)) {
1155 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
1160 /* FIXME: The use of this function with non-linear skb's really needs
1163 err
= skb_linearize(skb
);
1167 memset(skb
->data
+ skb
->len
, 0, pad
);
1174 EXPORT_SYMBOL(skb_pad
);
1177 * skb_put - add data to a buffer
1178 * @skb: buffer to use
1179 * @len: amount of data to add
1181 * This function extends the used data area of the buffer. If this would
1182 * exceed the total buffer size the kernel will panic. A pointer to the
1183 * first byte of the extra data is returned.
1185 unsigned char *skb_put(struct sk_buff
*skb
, unsigned int len
)
1187 unsigned char *tmp
= skb_tail_pointer(skb
);
1188 SKB_LINEAR_ASSERT(skb
);
1191 if (unlikely(skb
->tail
> skb
->end
))
1192 skb_over_panic(skb
, len
, __builtin_return_address(0));
1195 EXPORT_SYMBOL(skb_put
);
1198 * skb_push - add data to the start of a buffer
1199 * @skb: buffer to use
1200 * @len: amount of data to add
1202 * This function extends the used data area of the buffer at the buffer
1203 * start. If this would exceed the total buffer headroom the kernel will
1204 * panic. A pointer to the first byte of the extra data is returned.
1206 unsigned char *skb_push(struct sk_buff
*skb
, unsigned int len
)
1210 if (unlikely(skb
->data
<skb
->head
))
1211 skb_under_panic(skb
, len
, __builtin_return_address(0));
1214 EXPORT_SYMBOL(skb_push
);
1217 * skb_pull - remove data from the start of a buffer
1218 * @skb: buffer to use
1219 * @len: amount of data to remove
1221 * This function removes data from the start of a buffer, returning
1222 * the memory to the headroom. A pointer to the next data in the buffer
1223 * is returned. Once the data has been pulled future pushes will overwrite
1226 unsigned char *skb_pull(struct sk_buff
*skb
, unsigned int len
)
1228 return skb_pull_inline(skb
, len
);
1230 EXPORT_SYMBOL(skb_pull
);
1233 * skb_trim - remove end from a buffer
1234 * @skb: buffer to alter
1237 * Cut the length of a buffer down by removing data from the tail. If
1238 * the buffer is already under the length specified it is not modified.
1239 * The skb must be linear.
1241 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1244 __skb_trim(skb
, len
);
1246 EXPORT_SYMBOL(skb_trim
);
1248 /* Trims skb to length len. It can change skb pointers.
1251 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1253 struct sk_buff
**fragp
;
1254 struct sk_buff
*frag
;
1255 int offset
= skb_headlen(skb
);
1256 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1260 if (skb_cloned(skb
) &&
1261 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1268 for (; i
< nfrags
; i
++) {
1269 int end
= offset
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1276 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
++], len
- offset
);
1279 skb_shinfo(skb
)->nr_frags
= i
;
1281 for (; i
< nfrags
; i
++)
1282 skb_frag_unref(skb
, i
);
1284 if (skb_has_frag_list(skb
))
1285 skb_drop_fraglist(skb
);
1289 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1290 fragp
= &frag
->next
) {
1291 int end
= offset
+ frag
->len
;
1293 if (skb_shared(frag
)) {
1294 struct sk_buff
*nfrag
;
1296 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1297 if (unlikely(!nfrag
))
1300 nfrag
->next
= frag
->next
;
1312 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1316 skb_drop_list(&frag
->next
);
1321 if (len
> skb_headlen(skb
)) {
1322 skb
->data_len
-= skb
->len
- len
;
1327 skb_set_tail_pointer(skb
, len
);
1332 EXPORT_SYMBOL(___pskb_trim
);
1335 * __pskb_pull_tail - advance tail of skb header
1336 * @skb: buffer to reallocate
1337 * @delta: number of bytes to advance tail
1339 * The function makes a sense only on a fragmented &sk_buff,
1340 * it expands header moving its tail forward and copying necessary
1341 * data from fragmented part.
1343 * &sk_buff MUST have reference count of 1.
1345 * Returns %NULL (and &sk_buff does not change) if pull failed
1346 * or value of new tail of skb in the case of success.
1348 * All the pointers pointing into skb header may change and must be
1349 * reloaded after call to this function.
1352 /* Moves tail of skb head forward, copying data from fragmented part,
1353 * when it is necessary.
1354 * 1. It may fail due to malloc failure.
1355 * 2. It may change skb pointers.
1357 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1359 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1361 /* If skb has not enough free space at tail, get new one
1362 * plus 128 bytes for future expansions. If we have enough
1363 * room at tail, reallocate without expansion only if skb is cloned.
1365 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1367 if (eat
> 0 || skb_cloned(skb
)) {
1368 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1373 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
1376 /* Optimization: no fragments, no reasons to preestimate
1377 * size of pulled pages. Superb.
1379 if (!skb_has_frag_list(skb
))
1382 /* Estimate size of pulled pages. */
1384 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1385 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1392 /* If we need update frag list, we are in troubles.
1393 * Certainly, it possible to add an offset to skb data,
1394 * but taking into account that pulling is expected to
1395 * be very rare operation, it is worth to fight against
1396 * further bloating skb head and crucify ourselves here instead.
1397 * Pure masohism, indeed. 8)8)
1400 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1401 struct sk_buff
*clone
= NULL
;
1402 struct sk_buff
*insp
= NULL
;
1407 if (list
->len
<= eat
) {
1408 /* Eaten as whole. */
1413 /* Eaten partially. */
1415 if (skb_shared(list
)) {
1416 /* Sucks! We need to fork list. :-( */
1417 clone
= skb_clone(list
, GFP_ATOMIC
);
1423 /* This may be pulled without
1427 if (!pskb_pull(list
, eat
)) {
1435 /* Free pulled out fragments. */
1436 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1437 skb_shinfo(skb
)->frag_list
= list
->next
;
1440 /* And insert new clone at head. */
1443 skb_shinfo(skb
)->frag_list
= clone
;
1446 /* Success! Now we may commit changes to skb data. */
1451 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1452 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1455 skb_frag_unref(skb
, i
);
1458 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1460 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1461 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[k
], eat
);
1467 skb_shinfo(skb
)->nr_frags
= k
;
1470 skb
->data_len
-= delta
;
1472 return skb_tail_pointer(skb
);
1474 EXPORT_SYMBOL(__pskb_pull_tail
);
1477 * skb_copy_bits - copy bits from skb to kernel buffer
1479 * @offset: offset in source
1480 * @to: destination buffer
1481 * @len: number of bytes to copy
1483 * Copy the specified number of bytes from the source skb to the
1484 * destination buffer.
1487 * If its prototype is ever changed,
1488 * check arch/{*}/net/{*}.S files,
1489 * since it is called from BPF assembly code.
1491 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1493 int start
= skb_headlen(skb
);
1494 struct sk_buff
*frag_iter
;
1497 if (offset
> (int)skb
->len
- len
)
1501 if ((copy
= start
- offset
) > 0) {
1504 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
1505 if ((len
-= copy
) == 0)
1511 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1513 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
1515 WARN_ON(start
> offset
+ len
);
1517 end
= start
+ skb_frag_size(f
);
1518 if ((copy
= end
- offset
) > 0) {
1524 vaddr
= kmap_atomic(skb_frag_page(f
));
1526 vaddr
+ f
->page_offset
+ offset
- start
,
1528 kunmap_atomic(vaddr
);
1530 if ((len
-= copy
) == 0)
1538 skb_walk_frags(skb
, frag_iter
) {
1541 WARN_ON(start
> offset
+ len
);
1543 end
= start
+ frag_iter
->len
;
1544 if ((copy
= end
- offset
) > 0) {
1547 if (skb_copy_bits(frag_iter
, offset
- start
, to
, copy
))
1549 if ((len
-= copy
) == 0)
1563 EXPORT_SYMBOL(skb_copy_bits
);
1566 * Callback from splice_to_pipe(), if we need to release some pages
1567 * at the end of the spd in case we error'ed out in filling the pipe.
1569 static void sock_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
1571 put_page(spd
->pages
[i
]);
1574 static struct page
*linear_to_page(struct page
*page
, unsigned int *len
,
1575 unsigned int *offset
,
1576 struct sk_buff
*skb
, struct sock
*sk
)
1578 struct page
*p
= sk
->sk_sndmsg_page
;
1583 p
= sk
->sk_sndmsg_page
= alloc_pages(sk
->sk_allocation
, 0);
1587 off
= sk
->sk_sndmsg_off
= 0;
1588 /* hold one ref to this page until it's full */
1592 /* If we are the only user of the page, we can reset offset */
1593 if (page_count(p
) == 1)
1594 sk
->sk_sndmsg_off
= 0;
1595 off
= sk
->sk_sndmsg_off
;
1596 mlen
= PAGE_SIZE
- off
;
1597 if (mlen
< 64 && mlen
< *len
) {
1602 *len
= min_t(unsigned int, *len
, mlen
);
1605 memcpy(page_address(p
) + off
, page_address(page
) + *offset
, *len
);
1606 sk
->sk_sndmsg_off
+= *len
;
1612 static bool spd_can_coalesce(const struct splice_pipe_desc
*spd
,
1614 unsigned int offset
)
1616 return spd
->nr_pages
&&
1617 spd
->pages
[spd
->nr_pages
- 1] == page
&&
1618 (spd
->partial
[spd
->nr_pages
- 1].offset
+
1619 spd
->partial
[spd
->nr_pages
- 1].len
== offset
);
1623 * Fill page/offset/length into spd, if it can hold more pages.
1625 static bool spd_fill_page(struct splice_pipe_desc
*spd
,
1626 struct pipe_inode_info
*pipe
, struct page
*page
,
1627 unsigned int *len
, unsigned int offset
,
1628 struct sk_buff
*skb
, bool linear
,
1631 if (unlikely(spd
->nr_pages
== MAX_SKB_FRAGS
))
1635 page
= linear_to_page(page
, len
, &offset
, skb
, sk
);
1639 if (spd_can_coalesce(spd
, page
, offset
)) {
1640 spd
->partial
[spd
->nr_pages
- 1].len
+= *len
;
1644 spd
->pages
[spd
->nr_pages
] = page
;
1645 spd
->partial
[spd
->nr_pages
].len
= *len
;
1646 spd
->partial
[spd
->nr_pages
].offset
= offset
;
1652 static inline void __segment_seek(struct page
**page
, unsigned int *poff
,
1653 unsigned int *plen
, unsigned int off
)
1658 n
= *poff
/ PAGE_SIZE
;
1660 *page
= nth_page(*page
, n
);
1662 *poff
= *poff
% PAGE_SIZE
;
1666 static bool __splice_segment(struct page
*page
, unsigned int poff
,
1667 unsigned int plen
, unsigned int *off
,
1668 unsigned int *len
, struct sk_buff
*skb
,
1669 struct splice_pipe_desc
*spd
, bool linear
,
1671 struct pipe_inode_info
*pipe
)
1676 /* skip this segment if already processed */
1682 /* ignore any bits we already processed */
1684 __segment_seek(&page
, &poff
, &plen
, *off
);
1689 unsigned int flen
= min(*len
, plen
);
1691 /* the linear region may spread across several pages */
1692 flen
= min_t(unsigned int, flen
, PAGE_SIZE
- poff
);
1694 if (spd_fill_page(spd
, pipe
, page
, &flen
, poff
, skb
, linear
, sk
))
1697 __segment_seek(&page
, &poff
, &plen
, flen
);
1700 } while (*len
&& plen
);
1706 * Map linear and fragment data from the skb to spd. It reports true if the
1707 * pipe is full or if we already spliced the requested length.
1709 static bool __skb_splice_bits(struct sk_buff
*skb
, struct pipe_inode_info
*pipe
,
1710 unsigned int *offset
, unsigned int *len
,
1711 struct splice_pipe_desc
*spd
, struct sock
*sk
)
1715 /* map the linear part :
1716 * If skb->head_frag is set, this 'linear' part is backed by a
1717 * fragment, and if the head is not shared with any clones then
1718 * we can avoid a copy since we own the head portion of this page.
1720 if (__splice_segment(virt_to_page(skb
->data
),
1721 (unsigned long) skb
->data
& (PAGE_SIZE
- 1),
1723 offset
, len
, skb
, spd
,
1724 skb_head_is_locked(skb
),
1729 * then map the fragments
1731 for (seg
= 0; seg
< skb_shinfo(skb
)->nr_frags
; seg
++) {
1732 const skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[seg
];
1734 if (__splice_segment(skb_frag_page(f
),
1735 f
->page_offset
, skb_frag_size(f
),
1736 offset
, len
, skb
, spd
, false, sk
, pipe
))
1744 * Map data from the skb to a pipe. Should handle both the linear part,
1745 * the fragments, and the frag list. It does NOT handle frag lists within
1746 * the frag list, if such a thing exists. We'd probably need to recurse to
1747 * handle that cleanly.
1749 int skb_splice_bits(struct sk_buff
*skb
, unsigned int offset
,
1750 struct pipe_inode_info
*pipe
, unsigned int tlen
,
1753 struct partial_page partial
[MAX_SKB_FRAGS
];
1754 struct page
*pages
[MAX_SKB_FRAGS
];
1755 struct splice_pipe_desc spd
= {
1759 .ops
= &sock_pipe_buf_ops
,
1760 .spd_release
= sock_spd_release
,
1762 struct sk_buff
*frag_iter
;
1763 struct sock
*sk
= skb
->sk
;
1767 * __skb_splice_bits() only fails if the output has no room left,
1768 * so no point in going over the frag_list for the error case.
1770 if (__skb_splice_bits(skb
, pipe
, &offset
, &tlen
, &spd
, sk
))
1776 * now see if we have a frag_list to map
1778 skb_walk_frags(skb
, frag_iter
) {
1781 if (__skb_splice_bits(frag_iter
, pipe
, &offset
, &tlen
, &spd
, sk
))
1788 * Drop the socket lock, otherwise we have reverse
1789 * locking dependencies between sk_lock and i_mutex
1790 * here as compared to sendfile(). We enter here
1791 * with the socket lock held, and splice_to_pipe() will
1792 * grab the pipe inode lock. For sendfile() emulation,
1793 * we call into ->sendpage() with the i_mutex lock held
1794 * and networking will grab the socket lock.
1797 ret
= splice_to_pipe(pipe
, &spd
);
1805 * skb_store_bits - store bits from kernel buffer to skb
1806 * @skb: destination buffer
1807 * @offset: offset in destination
1808 * @from: source buffer
1809 * @len: number of bytes to copy
1811 * Copy the specified number of bytes from the source buffer to the
1812 * destination skb. This function handles all the messy bits of
1813 * traversing fragment lists and such.
1816 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
1818 int start
= skb_headlen(skb
);
1819 struct sk_buff
*frag_iter
;
1822 if (offset
> (int)skb
->len
- len
)
1825 if ((copy
= start
- offset
) > 0) {
1828 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
1829 if ((len
-= copy
) == 0)
1835 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1836 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1839 WARN_ON(start
> offset
+ len
);
1841 end
= start
+ skb_frag_size(frag
);
1842 if ((copy
= end
- offset
) > 0) {
1848 vaddr
= kmap_atomic(skb_frag_page(frag
));
1849 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1851 kunmap_atomic(vaddr
);
1853 if ((len
-= copy
) == 0)
1861 skb_walk_frags(skb
, frag_iter
) {
1864 WARN_ON(start
> offset
+ len
);
1866 end
= start
+ frag_iter
->len
;
1867 if ((copy
= end
- offset
) > 0) {
1870 if (skb_store_bits(frag_iter
, offset
- start
,
1873 if ((len
-= copy
) == 0)
1886 EXPORT_SYMBOL(skb_store_bits
);
1888 /* Checksum skb data. */
1890 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
1891 int len
, __wsum csum
)
1893 int start
= skb_headlen(skb
);
1894 int i
, copy
= start
- offset
;
1895 struct sk_buff
*frag_iter
;
1898 /* Checksum header. */
1902 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1903 if ((len
-= copy
) == 0)
1909 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1911 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1913 WARN_ON(start
> offset
+ len
);
1915 end
= start
+ skb_frag_size(frag
);
1916 if ((copy
= end
- offset
) > 0) {
1922 vaddr
= kmap_atomic(skb_frag_page(frag
));
1923 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
1924 offset
- start
, copy
, 0);
1925 kunmap_atomic(vaddr
);
1926 csum
= csum_block_add(csum
, csum2
, pos
);
1935 skb_walk_frags(skb
, frag_iter
) {
1938 WARN_ON(start
> offset
+ len
);
1940 end
= start
+ frag_iter
->len
;
1941 if ((copy
= end
- offset
) > 0) {
1945 csum2
= skb_checksum(frag_iter
, offset
- start
,
1947 csum
= csum_block_add(csum
, csum2
, pos
);
1948 if ((len
-= copy
) == 0)
1959 EXPORT_SYMBOL(skb_checksum
);
1961 /* Both of above in one bottle. */
1963 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
1964 u8
*to
, int len
, __wsum csum
)
1966 int start
= skb_headlen(skb
);
1967 int i
, copy
= start
- offset
;
1968 struct sk_buff
*frag_iter
;
1975 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
1977 if ((len
-= copy
) == 0)
1984 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1987 WARN_ON(start
> offset
+ len
);
1989 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1990 if ((copy
= end
- offset
) > 0) {
1993 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1997 vaddr
= kmap_atomic(skb_frag_page(frag
));
1998 csum2
= csum_partial_copy_nocheck(vaddr
+
2002 kunmap_atomic(vaddr
);
2003 csum
= csum_block_add(csum
, csum2
, pos
);
2013 skb_walk_frags(skb
, frag_iter
) {
2017 WARN_ON(start
> offset
+ len
);
2019 end
= start
+ frag_iter
->len
;
2020 if ((copy
= end
- offset
) > 0) {
2023 csum2
= skb_copy_and_csum_bits(frag_iter
,
2026 csum
= csum_block_add(csum
, csum2
, pos
);
2027 if ((len
-= copy
) == 0)
2038 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
2040 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
2045 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
2046 csstart
= skb_checksum_start_offset(skb
);
2048 csstart
= skb_headlen(skb
);
2050 BUG_ON(csstart
> skb_headlen(skb
));
2052 skb_copy_from_linear_data(skb
, to
, csstart
);
2055 if (csstart
!= skb
->len
)
2056 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
2057 skb
->len
- csstart
, 0);
2059 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2060 long csstuff
= csstart
+ skb
->csum_offset
;
2062 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
2065 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
2068 * skb_dequeue - remove from the head of the queue
2069 * @list: list to dequeue from
2071 * Remove the head of the list. The list lock is taken so the function
2072 * may be used safely with other locking list functions. The head item is
2073 * returned or %NULL if the list is empty.
2076 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
2078 unsigned long flags
;
2079 struct sk_buff
*result
;
2081 spin_lock_irqsave(&list
->lock
, flags
);
2082 result
= __skb_dequeue(list
);
2083 spin_unlock_irqrestore(&list
->lock
, flags
);
2086 EXPORT_SYMBOL(skb_dequeue
);
2089 * skb_dequeue_tail - remove from the tail of the queue
2090 * @list: list to dequeue from
2092 * Remove the tail of the list. The list lock is taken so the function
2093 * may be used safely with other locking list functions. The tail item is
2094 * returned or %NULL if the list is empty.
2096 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
2098 unsigned long flags
;
2099 struct sk_buff
*result
;
2101 spin_lock_irqsave(&list
->lock
, flags
);
2102 result
= __skb_dequeue_tail(list
);
2103 spin_unlock_irqrestore(&list
->lock
, flags
);
2106 EXPORT_SYMBOL(skb_dequeue_tail
);
2109 * skb_queue_purge - empty a list
2110 * @list: list to empty
2112 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2113 * the list and one reference dropped. This function takes the list
2114 * lock and is atomic with respect to other list locking functions.
2116 void skb_queue_purge(struct sk_buff_head
*list
)
2118 struct sk_buff
*skb
;
2119 while ((skb
= skb_dequeue(list
)) != NULL
)
2122 EXPORT_SYMBOL(skb_queue_purge
);
2125 * skb_queue_head - queue a buffer at the list head
2126 * @list: list to use
2127 * @newsk: buffer to queue
2129 * Queue a buffer at the start of the list. This function takes the
2130 * list lock and can be used safely with other locking &sk_buff functions
2133 * A buffer cannot be placed on two lists at the same time.
2135 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2137 unsigned long flags
;
2139 spin_lock_irqsave(&list
->lock
, flags
);
2140 __skb_queue_head(list
, newsk
);
2141 spin_unlock_irqrestore(&list
->lock
, flags
);
2143 EXPORT_SYMBOL(skb_queue_head
);
2146 * skb_queue_tail - queue a buffer at the list tail
2147 * @list: list to use
2148 * @newsk: buffer to queue
2150 * Queue a buffer at the tail of the list. This function takes the
2151 * list lock and can be used safely with other locking &sk_buff functions
2154 * A buffer cannot be placed on two lists at the same time.
2156 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2158 unsigned long flags
;
2160 spin_lock_irqsave(&list
->lock
, flags
);
2161 __skb_queue_tail(list
, newsk
);
2162 spin_unlock_irqrestore(&list
->lock
, flags
);
2164 EXPORT_SYMBOL(skb_queue_tail
);
2167 * skb_unlink - remove a buffer from a list
2168 * @skb: buffer to remove
2169 * @list: list to use
2171 * Remove a packet from a list. The list locks are taken and this
2172 * function is atomic with respect to other list locked calls
2174 * You must know what list the SKB is on.
2176 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
2178 unsigned long flags
;
2180 spin_lock_irqsave(&list
->lock
, flags
);
2181 __skb_unlink(skb
, list
);
2182 spin_unlock_irqrestore(&list
->lock
, flags
);
2184 EXPORT_SYMBOL(skb_unlink
);
2187 * skb_append - append a buffer
2188 * @old: buffer to insert after
2189 * @newsk: buffer to insert
2190 * @list: list to use
2192 * Place a packet after a given packet in a list. The list locks are taken
2193 * and this function is atomic with respect to other list locked calls.
2194 * A buffer cannot be placed on two lists at the same time.
2196 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2198 unsigned long flags
;
2200 spin_lock_irqsave(&list
->lock
, flags
);
2201 __skb_queue_after(list
, old
, newsk
);
2202 spin_unlock_irqrestore(&list
->lock
, flags
);
2204 EXPORT_SYMBOL(skb_append
);
2207 * skb_insert - insert a buffer
2208 * @old: buffer to insert before
2209 * @newsk: buffer to insert
2210 * @list: list to use
2212 * Place a packet before a given packet in a list. The list locks are
2213 * taken and this function is atomic with respect to other list locked
2216 * A buffer cannot be placed on two lists at the same time.
2218 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2220 unsigned long flags
;
2222 spin_lock_irqsave(&list
->lock
, flags
);
2223 __skb_insert(newsk
, old
->prev
, old
, list
);
2224 spin_unlock_irqrestore(&list
->lock
, flags
);
2226 EXPORT_SYMBOL(skb_insert
);
2228 static inline void skb_split_inside_header(struct sk_buff
*skb
,
2229 struct sk_buff
* skb1
,
2230 const u32 len
, const int pos
)
2234 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
2236 /* And move data appendix as is. */
2237 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
2238 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
2240 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
2241 skb_shinfo(skb
)->nr_frags
= 0;
2242 skb1
->data_len
= skb
->data_len
;
2243 skb1
->len
+= skb1
->data_len
;
2246 skb_set_tail_pointer(skb
, len
);
2249 static inline void skb_split_no_header(struct sk_buff
*skb
,
2250 struct sk_buff
* skb1
,
2251 const u32 len
, int pos
)
2254 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
2256 skb_shinfo(skb
)->nr_frags
= 0;
2257 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
2259 skb
->data_len
= len
- pos
;
2261 for (i
= 0; i
< nfrags
; i
++) {
2262 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
2264 if (pos
+ size
> len
) {
2265 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
2269 * We have two variants in this case:
2270 * 1. Move all the frag to the second
2271 * part, if it is possible. F.e.
2272 * this approach is mandatory for TUX,
2273 * where splitting is expensive.
2274 * 2. Split is accurately. We make this.
2276 skb_frag_ref(skb
, i
);
2277 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
2278 skb_frag_size_sub(&skb_shinfo(skb1
)->frags
[0], len
- pos
);
2279 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
], len
- pos
);
2280 skb_shinfo(skb
)->nr_frags
++;
2284 skb_shinfo(skb
)->nr_frags
++;
2287 skb_shinfo(skb1
)->nr_frags
= k
;
2291 * skb_split - Split fragmented skb to two parts at length len.
2292 * @skb: the buffer to split
2293 * @skb1: the buffer to receive the second part
2294 * @len: new length for skb
2296 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
2298 int pos
= skb_headlen(skb
);
2300 if (len
< pos
) /* Split line is inside header. */
2301 skb_split_inside_header(skb
, skb1
, len
, pos
);
2302 else /* Second chunk has no header, nothing to copy. */
2303 skb_split_no_header(skb
, skb1
, len
, pos
);
2305 EXPORT_SYMBOL(skb_split
);
2307 /* Shifting from/to a cloned skb is a no-go.
2309 * Caller cannot keep skb_shinfo related pointers past calling here!
2311 static int skb_prepare_for_shift(struct sk_buff
*skb
)
2313 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2317 * skb_shift - Shifts paged data partially from skb to another
2318 * @tgt: buffer into which tail data gets added
2319 * @skb: buffer from which the paged data comes from
2320 * @shiftlen: shift up to this many bytes
2322 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2323 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2324 * It's up to caller to free skb if everything was shifted.
2326 * If @tgt runs out of frags, the whole operation is aborted.
2328 * Skb cannot include anything else but paged data while tgt is allowed
2329 * to have non-paged data as well.
2331 * TODO: full sized shift could be optimized but that would need
2332 * specialized skb free'er to handle frags without up-to-date nr_frags.
2334 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
2336 int from
, to
, merge
, todo
;
2337 struct skb_frag_struct
*fragfrom
, *fragto
;
2339 BUG_ON(shiftlen
> skb
->len
);
2340 BUG_ON(skb_headlen(skb
)); /* Would corrupt stream */
2344 to
= skb_shinfo(tgt
)->nr_frags
;
2345 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2347 /* Actual merge is delayed until the point when we know we can
2348 * commit all, so that we don't have to undo partial changes
2351 !skb_can_coalesce(tgt
, to
, skb_frag_page(fragfrom
),
2352 fragfrom
->page_offset
)) {
2357 todo
-= skb_frag_size(fragfrom
);
2359 if (skb_prepare_for_shift(skb
) ||
2360 skb_prepare_for_shift(tgt
))
2363 /* All previous frag pointers might be stale! */
2364 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2365 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2367 skb_frag_size_add(fragto
, shiftlen
);
2368 skb_frag_size_sub(fragfrom
, shiftlen
);
2369 fragfrom
->page_offset
+= shiftlen
;
2377 /* Skip full, not-fitting skb to avoid expensive operations */
2378 if ((shiftlen
== skb
->len
) &&
2379 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
2382 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
2385 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
2386 if (to
== MAX_SKB_FRAGS
)
2389 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2390 fragto
= &skb_shinfo(tgt
)->frags
[to
];
2392 if (todo
>= skb_frag_size(fragfrom
)) {
2393 *fragto
= *fragfrom
;
2394 todo
-= skb_frag_size(fragfrom
);
2399 __skb_frag_ref(fragfrom
);
2400 fragto
->page
= fragfrom
->page
;
2401 fragto
->page_offset
= fragfrom
->page_offset
;
2402 skb_frag_size_set(fragto
, todo
);
2404 fragfrom
->page_offset
+= todo
;
2405 skb_frag_size_sub(fragfrom
, todo
);
2413 /* Ready to "commit" this state change to tgt */
2414 skb_shinfo(tgt
)->nr_frags
= to
;
2417 fragfrom
= &skb_shinfo(skb
)->frags
[0];
2418 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2420 skb_frag_size_add(fragto
, skb_frag_size(fragfrom
));
2421 __skb_frag_unref(fragfrom
);
2424 /* Reposition in the original skb */
2426 while (from
< skb_shinfo(skb
)->nr_frags
)
2427 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
2428 skb_shinfo(skb
)->nr_frags
= to
;
2430 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
2433 /* Most likely the tgt won't ever need its checksum anymore, skb on
2434 * the other hand might need it if it needs to be resent
2436 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
2437 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2439 /* Yak, is it really working this way? Some helper please? */
2440 skb
->len
-= shiftlen
;
2441 skb
->data_len
-= shiftlen
;
2442 skb
->truesize
-= shiftlen
;
2443 tgt
->len
+= shiftlen
;
2444 tgt
->data_len
+= shiftlen
;
2445 tgt
->truesize
+= shiftlen
;
2451 * skb_prepare_seq_read - Prepare a sequential read of skb data
2452 * @skb: the buffer to read
2453 * @from: lower offset of data to be read
2454 * @to: upper offset of data to be read
2455 * @st: state variable
2457 * Initializes the specified state variable. Must be called before
2458 * invoking skb_seq_read() for the first time.
2460 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
2461 unsigned int to
, struct skb_seq_state
*st
)
2463 st
->lower_offset
= from
;
2464 st
->upper_offset
= to
;
2465 st
->root_skb
= st
->cur_skb
= skb
;
2466 st
->frag_idx
= st
->stepped_offset
= 0;
2467 st
->frag_data
= NULL
;
2469 EXPORT_SYMBOL(skb_prepare_seq_read
);
2472 * skb_seq_read - Sequentially read skb data
2473 * @consumed: number of bytes consumed by the caller so far
2474 * @data: destination pointer for data to be returned
2475 * @st: state variable
2477 * Reads a block of skb data at &consumed relative to the
2478 * lower offset specified to skb_prepare_seq_read(). Assigns
2479 * the head of the data block to &data and returns the length
2480 * of the block or 0 if the end of the skb data or the upper
2481 * offset has been reached.
2483 * The caller is not required to consume all of the data
2484 * returned, i.e. &consumed is typically set to the number
2485 * of bytes already consumed and the next call to
2486 * skb_seq_read() will return the remaining part of the block.
2488 * Note 1: The size of each block of data returned can be arbitrary,
2489 * this limitation is the cost for zerocopy seqeuental
2490 * reads of potentially non linear data.
2492 * Note 2: Fragment lists within fragments are not implemented
2493 * at the moment, state->root_skb could be replaced with
2494 * a stack for this purpose.
2496 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
2497 struct skb_seq_state
*st
)
2499 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
2502 if (unlikely(abs_offset
>= st
->upper_offset
))
2506 block_limit
= skb_headlen(st
->cur_skb
) + st
->stepped_offset
;
2508 if (abs_offset
< block_limit
&& !st
->frag_data
) {
2509 *data
= st
->cur_skb
->data
+ (abs_offset
- st
->stepped_offset
);
2510 return block_limit
- abs_offset
;
2513 if (st
->frag_idx
== 0 && !st
->frag_data
)
2514 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
2516 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
2517 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
2518 block_limit
= skb_frag_size(frag
) + st
->stepped_offset
;
2520 if (abs_offset
< block_limit
) {
2522 st
->frag_data
= kmap_atomic(skb_frag_page(frag
));
2524 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
2525 (abs_offset
- st
->stepped_offset
);
2527 return block_limit
- abs_offset
;
2530 if (st
->frag_data
) {
2531 kunmap_atomic(st
->frag_data
);
2532 st
->frag_data
= NULL
;
2536 st
->stepped_offset
+= skb_frag_size(frag
);
2539 if (st
->frag_data
) {
2540 kunmap_atomic(st
->frag_data
);
2541 st
->frag_data
= NULL
;
2544 if (st
->root_skb
== st
->cur_skb
&& skb_has_frag_list(st
->root_skb
)) {
2545 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
2548 } else if (st
->cur_skb
->next
) {
2549 st
->cur_skb
= st
->cur_skb
->next
;
2556 EXPORT_SYMBOL(skb_seq_read
);
2559 * skb_abort_seq_read - Abort a sequential read of skb data
2560 * @st: state variable
2562 * Must be called if skb_seq_read() was not called until it
2565 void skb_abort_seq_read(struct skb_seq_state
*st
)
2568 kunmap_atomic(st
->frag_data
);
2570 EXPORT_SYMBOL(skb_abort_seq_read
);
2572 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2574 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
2575 struct ts_config
*conf
,
2576 struct ts_state
*state
)
2578 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
2581 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
2583 skb_abort_seq_read(TS_SKB_CB(state
));
2587 * skb_find_text - Find a text pattern in skb data
2588 * @skb: the buffer to look in
2589 * @from: search offset
2591 * @config: textsearch configuration
2592 * @state: uninitialized textsearch state variable
2594 * Finds a pattern in the skb data according to the specified
2595 * textsearch configuration. Use textsearch_next() to retrieve
2596 * subsequent occurrences of the pattern. Returns the offset
2597 * to the first occurrence or UINT_MAX if no match was found.
2599 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
2600 unsigned int to
, struct ts_config
*config
,
2601 struct ts_state
*state
)
2605 config
->get_next_block
= skb_ts_get_next_block
;
2606 config
->finish
= skb_ts_finish
;
2608 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
2610 ret
= textsearch_find(config
, state
);
2611 return (ret
<= to
- from
? ret
: UINT_MAX
);
2613 EXPORT_SYMBOL(skb_find_text
);
2616 * skb_append_datato_frags: - append the user data to a skb
2617 * @sk: sock structure
2618 * @skb: skb structure to be appened with user data.
2619 * @getfrag: call back function to be used for getting the user data
2620 * @from: pointer to user message iov
2621 * @length: length of the iov message
2623 * Description: This procedure append the user data in the fragment part
2624 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2626 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
2627 int (*getfrag
)(void *from
, char *to
, int offset
,
2628 int len
, int odd
, struct sk_buff
*skb
),
2629 void *from
, int length
)
2632 skb_frag_t
*frag
= NULL
;
2633 struct page
*page
= NULL
;
2639 /* Return error if we don't have space for new frag */
2640 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2641 if (frg_cnt
>= MAX_SKB_FRAGS
)
2644 /* allocate a new page for next frag */
2645 page
= alloc_pages(sk
->sk_allocation
, 0);
2647 /* If alloc_page fails just return failure and caller will
2648 * free previous allocated pages by doing kfree_skb()
2653 /* initialize the next frag */
2654 skb_fill_page_desc(skb
, frg_cnt
, page
, 0, 0);
2655 skb
->truesize
+= PAGE_SIZE
;
2656 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
2658 /* get the new initialized frag */
2659 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2660 frag
= &skb_shinfo(skb
)->frags
[frg_cnt
- 1];
2662 /* copy the user data to page */
2663 left
= PAGE_SIZE
- frag
->page_offset
;
2664 copy
= (length
> left
)? left
: length
;
2666 ret
= getfrag(from
, skb_frag_address(frag
) + skb_frag_size(frag
),
2667 offset
, copy
, 0, skb
);
2671 /* copy was successful so update the size parameters */
2672 skb_frag_size_add(frag
, copy
);
2674 skb
->data_len
+= copy
;
2678 } while (length
> 0);
2682 EXPORT_SYMBOL(skb_append_datato_frags
);
2685 * skb_pull_rcsum - pull skb and update receive checksum
2686 * @skb: buffer to update
2687 * @len: length of data pulled
2689 * This function performs an skb_pull on the packet and updates
2690 * the CHECKSUM_COMPLETE checksum. It should be used on
2691 * receive path processing instead of skb_pull unless you know
2692 * that the checksum difference is zero (e.g., a valid IP header)
2693 * or you are setting ip_summed to CHECKSUM_NONE.
2695 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
2697 BUG_ON(len
> skb
->len
);
2699 BUG_ON(skb
->len
< skb
->data_len
);
2700 skb_postpull_rcsum(skb
, skb
->data
, len
);
2701 return skb
->data
+= len
;
2703 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
2706 * skb_segment - Perform protocol segmentation on skb.
2707 * @skb: buffer to segment
2708 * @features: features for the output path (see dev->features)
2710 * This function performs segmentation on the given skb. It returns
2711 * a pointer to the first in a list of new skbs for the segments.
2712 * In case of error it returns ERR_PTR(err).
2714 struct sk_buff
*skb_segment(struct sk_buff
*skb
, netdev_features_t features
)
2716 struct sk_buff
*segs
= NULL
;
2717 struct sk_buff
*tail
= NULL
;
2718 struct sk_buff
*fskb
= skb_shinfo(skb
)->frag_list
;
2719 unsigned int mss
= skb_shinfo(skb
)->gso_size
;
2720 unsigned int doffset
= skb
->data
- skb_mac_header(skb
);
2721 unsigned int offset
= doffset
;
2722 unsigned int headroom
;
2724 int sg
= !!(features
& NETIF_F_SG
);
2725 int nfrags
= skb_shinfo(skb
)->nr_frags
;
2730 __skb_push(skb
, doffset
);
2731 headroom
= skb_headroom(skb
);
2732 pos
= skb_headlen(skb
);
2735 struct sk_buff
*nskb
;
2740 len
= skb
->len
- offset
;
2744 hsize
= skb_headlen(skb
) - offset
;
2747 if (hsize
> len
|| !sg
)
2750 if (!hsize
&& i
>= nfrags
) {
2751 BUG_ON(fskb
->len
!= len
);
2754 nskb
= skb_clone(fskb
, GFP_ATOMIC
);
2757 if (unlikely(!nskb
))
2760 hsize
= skb_end_offset(nskb
);
2761 if (skb_cow_head(nskb
, doffset
+ headroom
)) {
2766 nskb
->truesize
+= skb_end_offset(nskb
) - hsize
;
2767 skb_release_head_state(nskb
);
2768 __skb_push(nskb
, doffset
);
2770 nskb
= alloc_skb(hsize
+ doffset
+ headroom
,
2773 if (unlikely(!nskb
))
2776 skb_reserve(nskb
, headroom
);
2777 __skb_put(nskb
, doffset
);
2786 __copy_skb_header(nskb
, skb
);
2787 nskb
->mac_len
= skb
->mac_len
;
2789 /* nskb and skb might have different headroom */
2790 if (nskb
->ip_summed
== CHECKSUM_PARTIAL
)
2791 nskb
->csum_start
+= skb_headroom(nskb
) - headroom
;
2793 skb_reset_mac_header(nskb
);
2794 skb_set_network_header(nskb
, skb
->mac_len
);
2795 nskb
->transport_header
= (nskb
->network_header
+
2796 skb_network_header_len(skb
));
2797 skb_copy_from_linear_data(skb
, nskb
->data
, doffset
);
2799 if (fskb
!= skb_shinfo(skb
)->frag_list
)
2803 nskb
->ip_summed
= CHECKSUM_NONE
;
2804 nskb
->csum
= skb_copy_and_csum_bits(skb
, offset
,
2810 frag
= skb_shinfo(nskb
)->frags
;
2812 skb_copy_from_linear_data_offset(skb
, offset
,
2813 skb_put(nskb
, hsize
), hsize
);
2815 while (pos
< offset
+ len
&& i
< nfrags
) {
2816 *frag
= skb_shinfo(skb
)->frags
[i
];
2817 __skb_frag_ref(frag
);
2818 size
= skb_frag_size(frag
);
2821 frag
->page_offset
+= offset
- pos
;
2822 skb_frag_size_sub(frag
, offset
- pos
);
2825 skb_shinfo(nskb
)->nr_frags
++;
2827 if (pos
+ size
<= offset
+ len
) {
2831 skb_frag_size_sub(frag
, pos
+ size
- (offset
+ len
));
2838 if (pos
< offset
+ len
) {
2839 struct sk_buff
*fskb2
= fskb
;
2841 BUG_ON(pos
+ fskb
->len
!= offset
+ len
);
2847 fskb2
= skb_clone(fskb2
, GFP_ATOMIC
);
2853 SKB_FRAG_ASSERT(nskb
);
2854 skb_shinfo(nskb
)->frag_list
= fskb2
;
2858 nskb
->data_len
= len
- hsize
;
2859 nskb
->len
+= nskb
->data_len
;
2860 nskb
->truesize
+= nskb
->data_len
;
2861 } while ((offset
+= len
) < skb
->len
);
2866 while ((skb
= segs
)) {
2870 return ERR_PTR(err
);
2872 EXPORT_SYMBOL_GPL(skb_segment
);
2874 int skb_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2876 struct sk_buff
*p
= *head
;
2877 struct sk_buff
*nskb
;
2878 struct skb_shared_info
*skbinfo
= skb_shinfo(skb
);
2879 struct skb_shared_info
*pinfo
= skb_shinfo(p
);
2880 unsigned int headroom
;
2881 unsigned int len
= skb_gro_len(skb
);
2882 unsigned int offset
= skb_gro_offset(skb
);
2883 unsigned int headlen
= skb_headlen(skb
);
2884 unsigned int delta_truesize
;
2886 if (p
->len
+ len
>= 65536)
2889 if (pinfo
->frag_list
)
2891 else if (headlen
<= offset
) {
2894 int i
= skbinfo
->nr_frags
;
2895 int nr_frags
= pinfo
->nr_frags
+ i
;
2899 if (nr_frags
> MAX_SKB_FRAGS
)
2902 pinfo
->nr_frags
= nr_frags
;
2903 skbinfo
->nr_frags
= 0;
2905 frag
= pinfo
->frags
+ nr_frags
;
2906 frag2
= skbinfo
->frags
+ i
;
2911 frag
->page_offset
+= offset
;
2912 skb_frag_size_sub(frag
, offset
);
2914 /* all fragments truesize : remove (head size + sk_buff) */
2915 delta_truesize
= skb
->truesize
-
2916 SKB_TRUESIZE(skb_end_offset(skb
));
2918 skb
->truesize
-= skb
->data_len
;
2919 skb
->len
-= skb
->data_len
;
2922 NAPI_GRO_CB(skb
)->free
= NAPI_GRO_FREE
;
2924 } else if (skb
->head_frag
) {
2925 int nr_frags
= pinfo
->nr_frags
;
2926 skb_frag_t
*frag
= pinfo
->frags
+ nr_frags
;
2927 struct page
*page
= virt_to_head_page(skb
->head
);
2928 unsigned int first_size
= headlen
- offset
;
2929 unsigned int first_offset
;
2931 if (nr_frags
+ 1 + skbinfo
->nr_frags
> MAX_SKB_FRAGS
)
2934 first_offset
= skb
->data
-
2935 (unsigned char *)page_address(page
) +
2938 pinfo
->nr_frags
= nr_frags
+ 1 + skbinfo
->nr_frags
;
2940 frag
->page
.p
= page
;
2941 frag
->page_offset
= first_offset
;
2942 skb_frag_size_set(frag
, first_size
);
2944 memcpy(frag
+ 1, skbinfo
->frags
, sizeof(*frag
) * skbinfo
->nr_frags
);
2945 /* We dont need to clear skbinfo->nr_frags here */
2947 delta_truesize
= skb
->truesize
- SKB_DATA_ALIGN(sizeof(struct sk_buff
));
2948 NAPI_GRO_CB(skb
)->free
= NAPI_GRO_FREE_STOLEN_HEAD
;
2950 } else if (skb_gro_len(p
) != pinfo
->gso_size
)
2953 headroom
= skb_headroom(p
);
2954 nskb
= alloc_skb(headroom
+ skb_gro_offset(p
), GFP_ATOMIC
);
2955 if (unlikely(!nskb
))
2958 __copy_skb_header(nskb
, p
);
2959 nskb
->mac_len
= p
->mac_len
;
2961 skb_reserve(nskb
, headroom
);
2962 __skb_put(nskb
, skb_gro_offset(p
));
2964 skb_set_mac_header(nskb
, skb_mac_header(p
) - p
->data
);
2965 skb_set_network_header(nskb
, skb_network_offset(p
));
2966 skb_set_transport_header(nskb
, skb_transport_offset(p
));
2968 __skb_pull(p
, skb_gro_offset(p
));
2969 memcpy(skb_mac_header(nskb
), skb_mac_header(p
),
2970 p
->data
- skb_mac_header(p
));
2972 *NAPI_GRO_CB(nskb
) = *NAPI_GRO_CB(p
);
2973 skb_shinfo(nskb
)->frag_list
= p
;
2974 skb_shinfo(nskb
)->gso_size
= pinfo
->gso_size
;
2975 pinfo
->gso_size
= 0;
2976 skb_header_release(p
);
2979 nskb
->data_len
+= p
->len
;
2980 nskb
->truesize
+= p
->truesize
;
2981 nskb
->len
+= p
->len
;
2984 nskb
->next
= p
->next
;
2990 delta_truesize
= skb
->truesize
;
2991 if (offset
> headlen
) {
2992 unsigned int eat
= offset
- headlen
;
2994 skbinfo
->frags
[0].page_offset
+= eat
;
2995 skb_frag_size_sub(&skbinfo
->frags
[0], eat
);
2996 skb
->data_len
-= eat
;
3001 __skb_pull(skb
, offset
);
3003 p
->prev
->next
= skb
;
3005 skb_header_release(skb
);
3008 NAPI_GRO_CB(p
)->count
++;
3010 p
->truesize
+= delta_truesize
;
3013 NAPI_GRO_CB(skb
)->same_flow
= 1;
3016 EXPORT_SYMBOL_GPL(skb_gro_receive
);
3018 void __init
skb_init(void)
3020 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
3021 sizeof(struct sk_buff
),
3023 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
3025 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
3026 (2*sizeof(struct sk_buff
)) +
3029 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
3034 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3035 * @skb: Socket buffer containing the buffers to be mapped
3036 * @sg: The scatter-gather list to map into
3037 * @offset: The offset into the buffer's contents to start mapping
3038 * @len: Length of buffer space to be mapped
3040 * Fill the specified scatter-gather list with mappings/pointers into a
3041 * region of the buffer space attached to a socket buffer.
3044 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
3046 int start
= skb_headlen(skb
);
3047 int i
, copy
= start
- offset
;
3048 struct sk_buff
*frag_iter
;
3054 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
3056 if ((len
-= copy
) == 0)
3061 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
3064 WARN_ON(start
> offset
+ len
);
3066 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
3067 if ((copy
= end
- offset
) > 0) {
3068 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
3072 sg_set_page(&sg
[elt
], skb_frag_page(frag
), copy
,
3073 frag
->page_offset
+offset
-start
);
3082 skb_walk_frags(skb
, frag_iter
) {
3085 WARN_ON(start
> offset
+ len
);
3087 end
= start
+ frag_iter
->len
;
3088 if ((copy
= end
- offset
) > 0) {
3091 elt
+= __skb_to_sgvec(frag_iter
, sg
+elt
, offset
- start
,
3093 if ((len
-= copy
) == 0)
3103 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
3105 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
);
3107 sg_mark_end(&sg
[nsg
- 1]);
3111 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
3114 * skb_cow_data - Check that a socket buffer's data buffers are writable
3115 * @skb: The socket buffer to check.
3116 * @tailbits: Amount of trailing space to be added
3117 * @trailer: Returned pointer to the skb where the @tailbits space begins
3119 * Make sure that the data buffers attached to a socket buffer are
3120 * writable. If they are not, private copies are made of the data buffers
3121 * and the socket buffer is set to use these instead.
3123 * If @tailbits is given, make sure that there is space to write @tailbits
3124 * bytes of data beyond current end of socket buffer. @trailer will be
3125 * set to point to the skb in which this space begins.
3127 * The number of scatterlist elements required to completely map the
3128 * COW'd and extended socket buffer will be returned.
3130 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
3134 struct sk_buff
*skb1
, **skb_p
;
3136 /* If skb is cloned or its head is paged, reallocate
3137 * head pulling out all the pages (pages are considered not writable
3138 * at the moment even if they are anonymous).
3140 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
3141 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
3144 /* Easy case. Most of packets will go this way. */
3145 if (!skb_has_frag_list(skb
)) {
3146 /* A little of trouble, not enough of space for trailer.
3147 * This should not happen, when stack is tuned to generate
3148 * good frames. OK, on miss we reallocate and reserve even more
3149 * space, 128 bytes is fair. */
3151 if (skb_tailroom(skb
) < tailbits
&&
3152 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
3160 /* Misery. We are in troubles, going to mincer fragments... */
3163 skb_p
= &skb_shinfo(skb
)->frag_list
;
3166 while ((skb1
= *skb_p
) != NULL
) {
3169 /* The fragment is partially pulled by someone,
3170 * this can happen on input. Copy it and everything
3173 if (skb_shared(skb1
))
3176 /* If the skb is the last, worry about trailer. */
3178 if (skb1
->next
== NULL
&& tailbits
) {
3179 if (skb_shinfo(skb1
)->nr_frags
||
3180 skb_has_frag_list(skb1
) ||
3181 skb_tailroom(skb1
) < tailbits
)
3182 ntail
= tailbits
+ 128;
3188 skb_shinfo(skb1
)->nr_frags
||
3189 skb_has_frag_list(skb1
)) {
3190 struct sk_buff
*skb2
;
3192 /* Fuck, we are miserable poor guys... */
3194 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
3196 skb2
= skb_copy_expand(skb1
,
3200 if (unlikely(skb2
== NULL
))
3204 skb_set_owner_w(skb2
, skb1
->sk
);
3206 /* Looking around. Are we still alive?
3207 * OK, link new skb, drop old one */
3209 skb2
->next
= skb1
->next
;
3216 skb_p
= &skb1
->next
;
3221 EXPORT_SYMBOL_GPL(skb_cow_data
);
3223 static void sock_rmem_free(struct sk_buff
*skb
)
3225 struct sock
*sk
= skb
->sk
;
3227 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
3231 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3233 int sock_queue_err_skb(struct sock
*sk
, struct sk_buff
*skb
)
3237 if (atomic_read(&sk
->sk_rmem_alloc
) + skb
->truesize
>=
3238 (unsigned int)sk
->sk_rcvbuf
)
3243 skb
->destructor
= sock_rmem_free
;
3244 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
3246 /* before exiting rcu section, make sure dst is refcounted */
3249 skb_queue_tail(&sk
->sk_error_queue
, skb
);
3250 if (!sock_flag(sk
, SOCK_DEAD
))
3251 sk
->sk_data_ready(sk
, len
);
3254 EXPORT_SYMBOL(sock_queue_err_skb
);
3256 void skb_tstamp_tx(struct sk_buff
*orig_skb
,
3257 struct skb_shared_hwtstamps
*hwtstamps
)
3259 struct sock
*sk
= orig_skb
->sk
;
3260 struct sock_exterr_skb
*serr
;
3261 struct sk_buff
*skb
;
3267 skb
= skb_clone(orig_skb
, GFP_ATOMIC
);
3272 *skb_hwtstamps(skb
) =
3276 * no hardware time stamps available,
3277 * so keep the shared tx_flags and only
3278 * store software time stamp
3280 skb
->tstamp
= ktime_get_real();
3283 serr
= SKB_EXT_ERR(skb
);
3284 memset(serr
, 0, sizeof(*serr
));
3285 serr
->ee
.ee_errno
= ENOMSG
;
3286 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TIMESTAMPING
;
3288 err
= sock_queue_err_skb(sk
, skb
);
3293 EXPORT_SYMBOL_GPL(skb_tstamp_tx
);
3295 void skb_complete_wifi_ack(struct sk_buff
*skb
, bool acked
)
3297 struct sock
*sk
= skb
->sk
;
3298 struct sock_exterr_skb
*serr
;
3301 skb
->wifi_acked_valid
= 1;
3302 skb
->wifi_acked
= acked
;
3304 serr
= SKB_EXT_ERR(skb
);
3305 memset(serr
, 0, sizeof(*serr
));
3306 serr
->ee
.ee_errno
= ENOMSG
;
3307 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TXSTATUS
;
3309 err
= sock_queue_err_skb(sk
, skb
);
3313 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack
);
3317 * skb_partial_csum_set - set up and verify partial csum values for packet
3318 * @skb: the skb to set
3319 * @start: the number of bytes after skb->data to start checksumming.
3320 * @off: the offset from start to place the checksum.
3322 * For untrusted partially-checksummed packets, we need to make sure the values
3323 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3325 * This function checks and sets those values and skb->ip_summed: if this
3326 * returns false you should drop the packet.
3328 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
3330 if (unlikely(start
> skb_headlen(skb
)) ||
3331 unlikely((int)start
+ off
> skb_headlen(skb
) - 2)) {
3332 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3333 start
, off
, skb_headlen(skb
));
3336 skb
->ip_summed
= CHECKSUM_PARTIAL
;
3337 skb
->csum_start
= skb_headroom(skb
) + start
;
3338 skb
->csum_offset
= off
;
3341 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);
3343 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
3345 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
3348 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);
3350 void kfree_skb_partial(struct sk_buff
*skb
, bool head_stolen
)
3353 kmem_cache_free(skbuff_head_cache
, skb
);
3357 EXPORT_SYMBOL(kfree_skb_partial
);
3360 * skb_try_coalesce - try to merge skb to prior one
3362 * @from: buffer to add
3363 * @fragstolen: pointer to boolean
3366 bool skb_try_coalesce(struct sk_buff
*to
, struct sk_buff
*from
,
3367 bool *fragstolen
, int *delta_truesize
)
3369 int i
, delta
, len
= from
->len
;
3371 *fragstolen
= false;
3376 if (len
<= skb_tailroom(to
)) {
3377 BUG_ON(skb_copy_bits(from
, 0, skb_put(to
, len
), len
));
3378 *delta_truesize
= 0;
3382 if (skb_has_frag_list(to
) || skb_has_frag_list(from
))
3385 if (skb_headlen(from
) != 0) {
3387 unsigned int offset
;
3389 if (skb_shinfo(to
)->nr_frags
+
3390 skb_shinfo(from
)->nr_frags
>= MAX_SKB_FRAGS
)
3393 if (skb_head_is_locked(from
))
3396 delta
= from
->truesize
- SKB_DATA_ALIGN(sizeof(struct sk_buff
));
3398 page
= virt_to_head_page(from
->head
);
3399 offset
= from
->data
- (unsigned char *)page_address(page
);
3401 skb_fill_page_desc(to
, skb_shinfo(to
)->nr_frags
,
3402 page
, offset
, skb_headlen(from
));
3405 if (skb_shinfo(to
)->nr_frags
+
3406 skb_shinfo(from
)->nr_frags
> MAX_SKB_FRAGS
)
3409 delta
= from
->truesize
-
3410 SKB_TRUESIZE(skb_end_pointer(from
) - from
->head
);
3413 WARN_ON_ONCE(delta
< len
);
3415 memcpy(skb_shinfo(to
)->frags
+ skb_shinfo(to
)->nr_frags
,
3416 skb_shinfo(from
)->frags
,
3417 skb_shinfo(from
)->nr_frags
* sizeof(skb_frag_t
));
3418 skb_shinfo(to
)->nr_frags
+= skb_shinfo(from
)->nr_frags
;
3420 if (!skb_cloned(from
))
3421 skb_shinfo(from
)->nr_frags
= 0;
3423 /* if the skb is cloned this does nothing since we set nr_frags to 0 */
3424 for (i
= 0; i
< skb_shinfo(from
)->nr_frags
; i
++)
3425 skb_frag_ref(from
, i
);
3427 to
->truesize
+= delta
;
3429 to
->data_len
+= len
;
3431 *delta_truesize
= delta
;
3434 EXPORT_SYMBOL(skb_try_coalesce
);