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>
73 static struct kmem_cache
*skbuff_head_cache __read_mostly
;
74 static struct kmem_cache
*skbuff_fclone_cache __read_mostly
;
76 static void sock_pipe_buf_release(struct pipe_inode_info
*pipe
,
77 struct pipe_buffer
*buf
)
82 static void sock_pipe_buf_get(struct pipe_inode_info
*pipe
,
83 struct pipe_buffer
*buf
)
88 static int sock_pipe_buf_steal(struct pipe_inode_info
*pipe
,
89 struct pipe_buffer
*buf
)
95 /* Pipe buffer operations for a socket. */
96 static const struct pipe_buf_operations sock_pipe_buf_ops
= {
98 .map
= generic_pipe_buf_map
,
99 .unmap
= generic_pipe_buf_unmap
,
100 .confirm
= generic_pipe_buf_confirm
,
101 .release
= sock_pipe_buf_release
,
102 .steal
= sock_pipe_buf_steal
,
103 .get
= sock_pipe_buf_get
,
107 * Keep out-of-line to prevent kernel bloat.
108 * __builtin_return_address is not used because it is not always
113 * skb_over_panic - private function
118 * Out of line support code for skb_put(). Not user callable.
120 static void skb_over_panic(struct sk_buff
*skb
, int sz
, void *here
)
122 printk(KERN_EMERG
"skb_over_panic: text:%p len:%d put:%d head:%p "
123 "data:%p tail:%#lx end:%#lx dev:%s\n",
124 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 printk(KERN_EMERG
"skb_under_panic: text:%p len:%d put:%d head:%p "
142 "data:%p tail:%#lx end:%#lx dev:%s\n",
143 here
, skb
->len
, sz
, skb
->head
, skb
->data
,
144 (unsigned long)skb
->tail
, (unsigned long)skb
->end
,
145 skb
->dev
? skb
->dev
->name
: "<NULL>");
149 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
150 * 'private' fields and also do memory statistics to find all the
156 * __alloc_skb - allocate a network buffer
157 * @size: size to allocate
158 * @gfp_mask: allocation mask
159 * @fclone: allocate from fclone cache instead of head cache
160 * and allocate a cloned (child) skb
161 * @node: numa node to allocate memory on
163 * Allocate a new &sk_buff. The returned buffer has no headroom and a
164 * tail room of size bytes. The object has a reference count of one.
165 * The return is the buffer. On a failure the return is %NULL.
167 * Buffers may only be allocated from interrupts using a @gfp_mask of
170 struct sk_buff
*__alloc_skb(unsigned int size
, gfp_t gfp_mask
,
171 int fclone
, int node
)
173 struct kmem_cache
*cache
;
174 struct skb_shared_info
*shinfo
;
178 cache
= fclone
? skbuff_fclone_cache
: skbuff_head_cache
;
181 skb
= kmem_cache_alloc_node(cache
, gfp_mask
& ~__GFP_DMA
, node
);
186 /* We do our best to align skb_shared_info on a separate cache
187 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
188 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
189 * Both skb->head and skb_shared_info are cache line aligned.
191 size
= SKB_DATA_ALIGN(size
);
192 size
+= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
193 data
= kmalloc_node_track_caller(size
, gfp_mask
, node
);
196 /* kmalloc(size) might give us more room than requested.
197 * Put skb_shared_info exactly at the end of allocated zone,
198 * to allow max possible filling before reallocation.
200 size
= SKB_WITH_OVERHEAD(ksize(data
));
201 prefetchw(data
+ size
);
204 * Only clear those fields we need to clear, not those that we will
205 * actually initialise below. Hence, don't put any more fields after
206 * the tail pointer in struct sk_buff!
208 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
209 /* Account for allocated memory : skb + skb->head */
210 skb
->truesize
= SKB_TRUESIZE(size
);
211 atomic_set(&skb
->users
, 1);
214 skb_reset_tail_pointer(skb
);
215 skb
->end
= skb
->tail
+ size
;
216 #ifdef NET_SKBUFF_DATA_USES_OFFSET
217 skb
->mac_header
= ~0U;
220 /* make sure we initialize shinfo sequentially */
221 shinfo
= skb_shinfo(skb
);
222 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
223 atomic_set(&shinfo
->dataref
, 1);
224 kmemcheck_annotate_variable(shinfo
->destructor_arg
);
227 struct sk_buff
*child
= skb
+ 1;
228 atomic_t
*fclone_ref
= (atomic_t
*) (child
+ 1);
230 kmemcheck_annotate_bitfield(child
, flags1
);
231 kmemcheck_annotate_bitfield(child
, flags2
);
232 skb
->fclone
= SKB_FCLONE_ORIG
;
233 atomic_set(fclone_ref
, 1);
235 child
->fclone
= SKB_FCLONE_UNAVAILABLE
;
240 kmem_cache_free(cache
, skb
);
244 EXPORT_SYMBOL(__alloc_skb
);
247 * build_skb - build a network buffer
248 * @data: data buffer provided by caller
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
)
264 struct skb_shared_info
*shinfo
;
268 skb
= kmem_cache_alloc(skbuff_head_cache
, GFP_ATOMIC
);
272 size
= ksize(data
) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
274 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
275 skb
->truesize
= SKB_TRUESIZE(size
);
276 atomic_set(&skb
->users
, 1);
279 skb_reset_tail_pointer(skb
);
280 skb
->end
= skb
->tail
+ size
;
281 #ifdef NET_SKBUFF_DATA_USES_OFFSET
282 skb
->mac_header
= ~0U;
285 /* make sure we initialize shinfo sequentially */
286 shinfo
= skb_shinfo(skb
);
287 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
288 atomic_set(&shinfo
->dataref
, 1);
289 kmemcheck_annotate_variable(shinfo
->destructor_arg
);
293 EXPORT_SYMBOL(build_skb
);
296 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
297 * @dev: network device to receive on
298 * @length: length to allocate
299 * @gfp_mask: get_free_pages mask, passed to alloc_skb
301 * Allocate a new &sk_buff and assign it a usage count of one. The
302 * buffer has unspecified headroom built in. Users should allocate
303 * the headroom they think they need without accounting for the
304 * built in space. The built in space is used for optimisations.
306 * %NULL is returned if there is no free memory.
308 struct sk_buff
*__netdev_alloc_skb(struct net_device
*dev
,
309 unsigned int length
, gfp_t gfp_mask
)
313 skb
= __alloc_skb(length
+ NET_SKB_PAD
, gfp_mask
, 0, NUMA_NO_NODE
);
315 skb_reserve(skb
, NET_SKB_PAD
);
320 EXPORT_SYMBOL(__netdev_alloc_skb
);
322 void skb_add_rx_frag(struct sk_buff
*skb
, int i
, struct page
*page
, int off
,
323 int size
, unsigned int truesize
)
325 skb_fill_page_desc(skb
, i
, page
, off
, size
);
327 skb
->data_len
+= size
;
328 skb
->truesize
+= truesize
;
330 EXPORT_SYMBOL(skb_add_rx_frag
);
333 * dev_alloc_skb - allocate an skbuff for receiving
334 * @length: length to allocate
336 * Allocate a new &sk_buff and assign it a usage count of one. The
337 * buffer has unspecified headroom built in. Users should allocate
338 * the headroom they think they need without accounting for the
339 * built in space. The built in space is used for optimisations.
341 * %NULL is returned if there is no free memory. Although this function
342 * allocates memory it can be called from an interrupt.
344 struct sk_buff
*dev_alloc_skb(unsigned int length
)
347 * There is more code here than it seems:
348 * __dev_alloc_skb is an inline
350 return __dev_alloc_skb(length
, GFP_ATOMIC
);
352 EXPORT_SYMBOL(dev_alloc_skb
);
354 static void skb_drop_list(struct sk_buff
**listp
)
356 struct sk_buff
*list
= *listp
;
361 struct sk_buff
*this = list
;
367 static inline void skb_drop_fraglist(struct sk_buff
*skb
)
369 skb_drop_list(&skb_shinfo(skb
)->frag_list
);
372 static void skb_clone_fraglist(struct sk_buff
*skb
)
374 struct sk_buff
*list
;
376 skb_walk_frags(skb
, list
)
380 static void skb_release_data(struct sk_buff
*skb
)
383 !atomic_sub_return(skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1,
384 &skb_shinfo(skb
)->dataref
)) {
385 if (skb_shinfo(skb
)->nr_frags
) {
387 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
388 skb_frag_unref(skb
, i
);
392 * If skb buf is from userspace, we need to notify the caller
393 * the lower device DMA has done;
395 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
396 struct ubuf_info
*uarg
;
398 uarg
= skb_shinfo(skb
)->destructor_arg
;
400 uarg
->callback(uarg
);
403 if (skb_has_frag_list(skb
))
404 skb_drop_fraglist(skb
);
411 * Free an skbuff by memory without cleaning the state.
413 static void kfree_skbmem(struct sk_buff
*skb
)
415 struct sk_buff
*other
;
416 atomic_t
*fclone_ref
;
418 switch (skb
->fclone
) {
419 case SKB_FCLONE_UNAVAILABLE
:
420 kmem_cache_free(skbuff_head_cache
, skb
);
423 case SKB_FCLONE_ORIG
:
424 fclone_ref
= (atomic_t
*) (skb
+ 2);
425 if (atomic_dec_and_test(fclone_ref
))
426 kmem_cache_free(skbuff_fclone_cache
, skb
);
429 case SKB_FCLONE_CLONE
:
430 fclone_ref
= (atomic_t
*) (skb
+ 1);
433 /* The clone portion is available for
434 * fast-cloning again.
436 skb
->fclone
= SKB_FCLONE_UNAVAILABLE
;
438 if (atomic_dec_and_test(fclone_ref
))
439 kmem_cache_free(skbuff_fclone_cache
, other
);
444 static void skb_release_head_state(struct sk_buff
*skb
)
448 secpath_put(skb
->sp
);
450 if (skb
->destructor
) {
452 skb
->destructor(skb
);
454 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
455 nf_conntrack_put(skb
->nfct
);
457 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
458 nf_conntrack_put_reasm(skb
->nfct_reasm
);
460 #ifdef CONFIG_BRIDGE_NETFILTER
461 nf_bridge_put(skb
->nf_bridge
);
463 /* XXX: IS this still necessary? - JHS */
464 #ifdef CONFIG_NET_SCHED
466 #ifdef CONFIG_NET_CLS_ACT
472 /* Free everything but the sk_buff shell. */
473 static void skb_release_all(struct sk_buff
*skb
)
475 skb_release_head_state(skb
);
476 skb_release_data(skb
);
480 * __kfree_skb - private function
483 * Free an sk_buff. Release anything attached to the buffer.
484 * Clean the state. This is an internal helper function. Users should
485 * always call kfree_skb
488 void __kfree_skb(struct sk_buff
*skb
)
490 skb_release_all(skb
);
493 EXPORT_SYMBOL(__kfree_skb
);
496 * kfree_skb - free an sk_buff
497 * @skb: buffer to free
499 * Drop a reference to the buffer and free it if the usage count has
502 void kfree_skb(struct sk_buff
*skb
)
506 if (likely(atomic_read(&skb
->users
) == 1))
508 else if (likely(!atomic_dec_and_test(&skb
->users
)))
510 trace_kfree_skb(skb
, __builtin_return_address(0));
513 EXPORT_SYMBOL(kfree_skb
);
516 * consume_skb - free an skbuff
517 * @skb: buffer to free
519 * Drop a ref to the buffer and free it if the usage count has hit zero
520 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
521 * is being dropped after a failure and notes that
523 void consume_skb(struct sk_buff
*skb
)
527 if (likely(atomic_read(&skb
->users
) == 1))
529 else if (likely(!atomic_dec_and_test(&skb
->users
)))
531 trace_consume_skb(skb
);
534 EXPORT_SYMBOL(consume_skb
);
537 * skb_recycle - clean up an skb for reuse
540 * Recycles the skb to be reused as a receive buffer. This
541 * function does any necessary reference count dropping, and
542 * cleans up the skbuff as if it just came from __alloc_skb().
544 void skb_recycle(struct sk_buff
*skb
)
546 struct skb_shared_info
*shinfo
;
548 skb_release_head_state(skb
);
550 shinfo
= skb_shinfo(skb
);
551 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
552 atomic_set(&shinfo
->dataref
, 1);
554 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
555 skb
->data
= skb
->head
+ NET_SKB_PAD
;
556 skb_reset_tail_pointer(skb
);
558 EXPORT_SYMBOL(skb_recycle
);
561 * skb_recycle_check - check if skb can be reused for receive
563 * @skb_size: minimum receive buffer size
565 * Checks that the skb passed in is not shared or cloned, and
566 * that it is linear and its head portion at least as large as
567 * skb_size so that it can be recycled as a receive buffer.
568 * If these conditions are met, this function does any necessary
569 * reference count dropping and cleans up the skbuff as if it
570 * just came from __alloc_skb().
572 bool skb_recycle_check(struct sk_buff
*skb
, int skb_size
)
574 if (!skb_is_recycleable(skb
, skb_size
))
581 EXPORT_SYMBOL(skb_recycle_check
);
583 static void __copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
585 new->tstamp
= old
->tstamp
;
587 new->transport_header
= old
->transport_header
;
588 new->network_header
= old
->network_header
;
589 new->mac_header
= old
->mac_header
;
590 skb_dst_copy(new, old
);
591 new->rxhash
= old
->rxhash
;
592 new->ooo_okay
= old
->ooo_okay
;
593 new->l4_rxhash
= old
->l4_rxhash
;
594 new->no_fcs
= old
->no_fcs
;
596 new->sp
= secpath_get(old
->sp
);
598 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
599 new->csum
= old
->csum
;
600 new->local_df
= old
->local_df
;
601 new->pkt_type
= old
->pkt_type
;
602 new->ip_summed
= old
->ip_summed
;
603 skb_copy_queue_mapping(new, old
);
604 new->priority
= old
->priority
;
605 #if IS_ENABLED(CONFIG_IP_VS)
606 new->ipvs_property
= old
->ipvs_property
;
608 new->protocol
= old
->protocol
;
609 new->mark
= old
->mark
;
610 new->skb_iif
= old
->skb_iif
;
612 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
613 new->nf_trace
= old
->nf_trace
;
615 #ifdef CONFIG_NET_SCHED
616 new->tc_index
= old
->tc_index
;
617 #ifdef CONFIG_NET_CLS_ACT
618 new->tc_verd
= old
->tc_verd
;
621 new->vlan_tci
= old
->vlan_tci
;
623 skb_copy_secmark(new, old
);
627 * You should not add any new code to this function. Add it to
628 * __copy_skb_header above instead.
630 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
632 #define C(x) n->x = skb->x
634 n
->next
= n
->prev
= NULL
;
636 __copy_skb_header(n
, skb
);
641 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
644 n
->destructor
= NULL
;
650 atomic_set(&n
->users
, 1);
652 atomic_inc(&(skb_shinfo(skb
)->dataref
));
660 * skb_morph - morph one skb into another
661 * @dst: the skb to receive the contents
662 * @src: the skb to supply the contents
664 * This is identical to skb_clone except that the target skb is
665 * supplied by the user.
667 * The target skb is returned upon exit.
669 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
671 skb_release_all(dst
);
672 return __skb_clone(dst
, src
);
674 EXPORT_SYMBOL_GPL(skb_morph
);
676 /* skb_copy_ubufs - copy userspace skb frags buffers to kernel
677 * @skb: the skb to modify
678 * @gfp_mask: allocation priority
680 * This must be called on SKBTX_DEV_ZEROCOPY skb.
681 * It will copy all frags into kernel and drop the reference
682 * to userspace pages.
684 * If this function is called from an interrupt gfp_mask() must be
687 * Returns 0 on success or a negative error code on failure
688 * to allocate kernel memory to copy to.
690 int skb_copy_ubufs(struct sk_buff
*skb
, gfp_t gfp_mask
)
693 int num_frags
= skb_shinfo(skb
)->nr_frags
;
694 struct page
*page
, *head
= NULL
;
695 struct ubuf_info
*uarg
= skb_shinfo(skb
)->destructor_arg
;
697 for (i
= 0; i
< num_frags
; i
++) {
699 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
701 page
= alloc_page(GFP_ATOMIC
);
704 struct page
*next
= (struct page
*)head
->private;
710 vaddr
= kmap_skb_frag(&skb_shinfo(skb
)->frags
[i
]);
711 memcpy(page_address(page
),
712 vaddr
+ f
->page_offset
, skb_frag_size(f
));
713 kunmap_skb_frag(vaddr
);
714 page
->private = (unsigned long)head
;
718 /* skb frags release userspace buffers */
719 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
720 skb_frag_unref(skb
, i
);
722 uarg
->callback(uarg
);
724 /* skb frags point to kernel buffers */
725 for (i
= skb_shinfo(skb
)->nr_frags
; i
> 0; i
--) {
726 __skb_fill_page_desc(skb
, i
-1, head
, 0,
727 skb_shinfo(skb
)->frags
[i
- 1].size
);
728 head
= (struct page
*)head
->private;
731 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
737 * skb_clone - duplicate an sk_buff
738 * @skb: buffer to clone
739 * @gfp_mask: allocation priority
741 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
742 * copies share the same packet data but not structure. The new
743 * buffer has a reference count of 1. If the allocation fails the
744 * function returns %NULL otherwise the new buffer is returned.
746 * If this function is called from an interrupt gfp_mask() must be
750 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
754 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
755 if (skb_copy_ubufs(skb
, gfp_mask
))
760 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
761 n
->fclone
== SKB_FCLONE_UNAVAILABLE
) {
762 atomic_t
*fclone_ref
= (atomic_t
*) (n
+ 1);
763 n
->fclone
= SKB_FCLONE_CLONE
;
764 atomic_inc(fclone_ref
);
766 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
770 kmemcheck_annotate_bitfield(n
, flags1
);
771 kmemcheck_annotate_bitfield(n
, flags2
);
772 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
775 return __skb_clone(n
, skb
);
777 EXPORT_SYMBOL(skb_clone
);
779 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
781 #ifndef NET_SKBUFF_DATA_USES_OFFSET
783 * Shift between the two data areas in bytes
785 unsigned long offset
= new->data
- old
->data
;
788 __copy_skb_header(new, old
);
790 #ifndef NET_SKBUFF_DATA_USES_OFFSET
791 /* {transport,network,mac}_header are relative to skb->head */
792 new->transport_header
+= offset
;
793 new->network_header
+= offset
;
794 if (skb_mac_header_was_set(new))
795 new->mac_header
+= offset
;
797 skb_shinfo(new)->gso_size
= skb_shinfo(old
)->gso_size
;
798 skb_shinfo(new)->gso_segs
= skb_shinfo(old
)->gso_segs
;
799 skb_shinfo(new)->gso_type
= skb_shinfo(old
)->gso_type
;
803 * skb_copy - create private copy of an sk_buff
804 * @skb: buffer to copy
805 * @gfp_mask: allocation priority
807 * Make a copy of both an &sk_buff and its data. This is used when the
808 * caller wishes to modify the data and needs a private copy of the
809 * data to alter. Returns %NULL on failure or the pointer to the buffer
810 * on success. The returned buffer has a reference count of 1.
812 * As by-product this function converts non-linear &sk_buff to linear
813 * one, so that &sk_buff becomes completely private and caller is allowed
814 * to modify all the data of returned buffer. This means that this
815 * function is not recommended for use in circumstances when only
816 * header is going to be modified. Use pskb_copy() instead.
819 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
821 int headerlen
= skb_headroom(skb
);
822 unsigned int size
= (skb_end_pointer(skb
) - skb
->head
) + skb
->data_len
;
823 struct sk_buff
*n
= alloc_skb(size
, gfp_mask
);
828 /* Set the data pointer */
829 skb_reserve(n
, headerlen
);
830 /* Set the tail pointer and length */
831 skb_put(n
, skb
->len
);
833 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
836 copy_skb_header(n
, skb
);
839 EXPORT_SYMBOL(skb_copy
);
842 * __pskb_copy - create copy of an sk_buff with private head.
843 * @skb: buffer to copy
844 * @headroom: headroom of new skb
845 * @gfp_mask: allocation priority
847 * Make a copy of both an &sk_buff and part of its data, located
848 * in header. Fragmented data remain shared. This is used when
849 * the caller wishes to modify only header of &sk_buff and needs
850 * private copy of the header to alter. Returns %NULL on failure
851 * or the pointer to the buffer on success.
852 * The returned buffer has a reference count of 1.
855 struct sk_buff
*__pskb_copy(struct sk_buff
*skb
, int headroom
, gfp_t gfp_mask
)
857 unsigned int size
= skb_headlen(skb
) + headroom
;
858 struct sk_buff
*n
= alloc_skb(size
, gfp_mask
);
863 /* Set the data pointer */
864 skb_reserve(n
, headroom
);
865 /* Set the tail pointer and length */
866 skb_put(n
, skb_headlen(skb
));
868 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
870 n
->truesize
+= skb
->data_len
;
871 n
->data_len
= skb
->data_len
;
874 if (skb_shinfo(skb
)->nr_frags
) {
877 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
878 if (skb_copy_ubufs(skb
, gfp_mask
)) {
884 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
885 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
886 skb_frag_ref(skb
, i
);
888 skb_shinfo(n
)->nr_frags
= i
;
891 if (skb_has_frag_list(skb
)) {
892 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
893 skb_clone_fraglist(n
);
896 copy_skb_header(n
, skb
);
900 EXPORT_SYMBOL(__pskb_copy
);
903 * pskb_expand_head - reallocate header of &sk_buff
904 * @skb: buffer to reallocate
905 * @nhead: room to add at head
906 * @ntail: room to add at tail
907 * @gfp_mask: allocation priority
909 * Expands (or creates identical copy, if &nhead and &ntail are zero)
910 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
911 * reference count of 1. Returns zero in the case of success or error,
912 * if expansion failed. In the last case, &sk_buff is not changed.
914 * All the pointers pointing into skb header may change and must be
915 * reloaded after call to this function.
918 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
923 int size
= nhead
+ (skb_end_pointer(skb
) - skb
->head
) + ntail
;
932 size
= SKB_DATA_ALIGN(size
);
934 /* Check if we can avoid taking references on fragments if we own
935 * the last reference on skb->head. (see skb_release_data())
940 int delta
= skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1;
941 fastpath
= atomic_read(&skb_shinfo(skb
)->dataref
) == delta
;
945 size
+ sizeof(struct skb_shared_info
) <= ksize(skb
->head
)) {
946 memmove(skb
->head
+ size
, skb_shinfo(skb
),
947 offsetof(struct skb_shared_info
,
948 frags
[skb_shinfo(skb
)->nr_frags
]));
949 memmove(skb
->head
+ nhead
, skb
->head
,
950 skb_tail_pointer(skb
) - skb
->head
);
955 data
= kmalloc(size
+ SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)),
959 size
= SKB_WITH_OVERHEAD(ksize(data
));
961 /* Copy only real data... and, alas, header. This should be
962 * optimized for the cases when header is void.
964 memcpy(data
+ nhead
, skb
->head
, skb_tail_pointer(skb
) - skb
->head
);
966 memcpy((struct skb_shared_info
*)(data
+ size
),
968 offsetof(struct skb_shared_info
, frags
[skb_shinfo(skb
)->nr_frags
]));
973 /* copy this zero copy skb frags */
974 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
975 if (skb_copy_ubufs(skb
, gfp_mask
))
978 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
979 skb_frag_ref(skb
, i
);
981 if (skb_has_frag_list(skb
))
982 skb_clone_fraglist(skb
);
984 skb_release_data(skb
);
986 off
= (data
+ nhead
) - skb
->head
;
991 #ifdef NET_SKBUFF_DATA_USES_OFFSET
995 skb
->end
= skb
->head
+ size
;
997 /* {transport,network,mac}_header and tail are relative to skb->head */
999 skb
->transport_header
+= off
;
1000 skb
->network_header
+= off
;
1001 if (skb_mac_header_was_set(skb
))
1002 skb
->mac_header
+= off
;
1003 /* Only adjust this if it actually is csum_start rather than csum */
1004 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1005 skb
->csum_start
+= nhead
;
1009 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
1017 EXPORT_SYMBOL(pskb_expand_head
);
1019 /* Make private copy of skb with writable head and some headroom */
1021 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
1023 struct sk_buff
*skb2
;
1024 int delta
= headroom
- skb_headroom(skb
);
1027 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
1029 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1030 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
1038 EXPORT_SYMBOL(skb_realloc_headroom
);
1041 * skb_copy_expand - copy and expand sk_buff
1042 * @skb: buffer to copy
1043 * @newheadroom: new free bytes at head
1044 * @newtailroom: new free bytes at tail
1045 * @gfp_mask: allocation priority
1047 * Make a copy of both an &sk_buff and its data and while doing so
1048 * allocate additional space.
1050 * This is used when the caller wishes to modify the data and needs a
1051 * private copy of the data to alter as well as more space for new fields.
1052 * Returns %NULL on failure or the pointer to the buffer
1053 * on success. The returned buffer has a reference count of 1.
1055 * You must pass %GFP_ATOMIC as the allocation priority if this function
1056 * is called from an interrupt.
1058 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
1059 int newheadroom
, int newtailroom
,
1063 * Allocate the copy buffer
1065 struct sk_buff
*n
= alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
1067 int oldheadroom
= skb_headroom(skb
);
1068 int head_copy_len
, head_copy_off
;
1074 skb_reserve(n
, newheadroom
);
1076 /* Set the tail pointer and length */
1077 skb_put(n
, skb
->len
);
1079 head_copy_len
= oldheadroom
;
1081 if (newheadroom
<= head_copy_len
)
1082 head_copy_len
= newheadroom
;
1084 head_copy_off
= newheadroom
- head_copy_len
;
1086 /* Copy the linear header and data. */
1087 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
1088 skb
->len
+ head_copy_len
))
1091 copy_skb_header(n
, skb
);
1093 off
= newheadroom
- oldheadroom
;
1094 if (n
->ip_summed
== CHECKSUM_PARTIAL
)
1095 n
->csum_start
+= off
;
1096 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1097 n
->transport_header
+= off
;
1098 n
->network_header
+= off
;
1099 if (skb_mac_header_was_set(skb
))
1100 n
->mac_header
+= off
;
1105 EXPORT_SYMBOL(skb_copy_expand
);
1108 * skb_pad - zero pad the tail of an skb
1109 * @skb: buffer to pad
1110 * @pad: space to pad
1112 * Ensure that a buffer is followed by a padding area that is zero
1113 * filled. Used by network drivers which may DMA or transfer data
1114 * beyond the buffer end onto the wire.
1116 * May return error in out of memory cases. The skb is freed on error.
1119 int skb_pad(struct sk_buff
*skb
, int pad
)
1124 /* If the skbuff is non linear tailroom is always zero.. */
1125 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
1126 memset(skb
->data
+skb
->len
, 0, pad
);
1130 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
1131 if (likely(skb_cloned(skb
) || ntail
> 0)) {
1132 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
1137 /* FIXME: The use of this function with non-linear skb's really needs
1140 err
= skb_linearize(skb
);
1144 memset(skb
->data
+ skb
->len
, 0, pad
);
1151 EXPORT_SYMBOL(skb_pad
);
1154 * skb_put - add data to a buffer
1155 * @skb: buffer to use
1156 * @len: amount of data to add
1158 * This function extends the used data area of the buffer. If this would
1159 * exceed the total buffer size the kernel will panic. A pointer to the
1160 * first byte of the extra data is returned.
1162 unsigned char *skb_put(struct sk_buff
*skb
, unsigned int len
)
1164 unsigned char *tmp
= skb_tail_pointer(skb
);
1165 SKB_LINEAR_ASSERT(skb
);
1168 if (unlikely(skb
->tail
> skb
->end
))
1169 skb_over_panic(skb
, len
, __builtin_return_address(0));
1172 EXPORT_SYMBOL(skb_put
);
1175 * skb_push - add data to the start of a buffer
1176 * @skb: buffer to use
1177 * @len: amount of data to add
1179 * This function extends the used data area of the buffer at the buffer
1180 * start. If this would exceed the total buffer headroom the kernel will
1181 * panic. A pointer to the first byte of the extra data is returned.
1183 unsigned char *skb_push(struct sk_buff
*skb
, unsigned int len
)
1187 if (unlikely(skb
->data
<skb
->head
))
1188 skb_under_panic(skb
, len
, __builtin_return_address(0));
1191 EXPORT_SYMBOL(skb_push
);
1194 * skb_pull - remove data from the start of a buffer
1195 * @skb: buffer to use
1196 * @len: amount of data to remove
1198 * This function removes data from the start of a buffer, returning
1199 * the memory to the headroom. A pointer to the next data in the buffer
1200 * is returned. Once the data has been pulled future pushes will overwrite
1203 unsigned char *skb_pull(struct sk_buff
*skb
, unsigned int len
)
1205 return skb_pull_inline(skb
, len
);
1207 EXPORT_SYMBOL(skb_pull
);
1210 * skb_trim - remove end from a buffer
1211 * @skb: buffer to alter
1214 * Cut the length of a buffer down by removing data from the tail. If
1215 * the buffer is already under the length specified it is not modified.
1216 * The skb must be linear.
1218 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1221 __skb_trim(skb
, len
);
1223 EXPORT_SYMBOL(skb_trim
);
1225 /* Trims skb to length len. It can change skb pointers.
1228 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1230 struct sk_buff
**fragp
;
1231 struct sk_buff
*frag
;
1232 int offset
= skb_headlen(skb
);
1233 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1237 if (skb_cloned(skb
) &&
1238 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1245 for (; i
< nfrags
; i
++) {
1246 int end
= offset
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1253 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
++], len
- offset
);
1256 skb_shinfo(skb
)->nr_frags
= i
;
1258 for (; i
< nfrags
; i
++)
1259 skb_frag_unref(skb
, i
);
1261 if (skb_has_frag_list(skb
))
1262 skb_drop_fraglist(skb
);
1266 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1267 fragp
= &frag
->next
) {
1268 int end
= offset
+ frag
->len
;
1270 if (skb_shared(frag
)) {
1271 struct sk_buff
*nfrag
;
1273 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1274 if (unlikely(!nfrag
))
1277 nfrag
->next
= frag
->next
;
1289 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1293 skb_drop_list(&frag
->next
);
1298 if (len
> skb_headlen(skb
)) {
1299 skb
->data_len
-= skb
->len
- len
;
1304 skb_set_tail_pointer(skb
, len
);
1309 EXPORT_SYMBOL(___pskb_trim
);
1312 * __pskb_pull_tail - advance tail of skb header
1313 * @skb: buffer to reallocate
1314 * @delta: number of bytes to advance tail
1316 * The function makes a sense only on a fragmented &sk_buff,
1317 * it expands header moving its tail forward and copying necessary
1318 * data from fragmented part.
1320 * &sk_buff MUST have reference count of 1.
1322 * Returns %NULL (and &sk_buff does not change) if pull failed
1323 * or value of new tail of skb in the case of success.
1325 * All the pointers pointing into skb header may change and must be
1326 * reloaded after call to this function.
1329 /* Moves tail of skb head forward, copying data from fragmented part,
1330 * when it is necessary.
1331 * 1. It may fail due to malloc failure.
1332 * 2. It may change skb pointers.
1334 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1336 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1338 /* If skb has not enough free space at tail, get new one
1339 * plus 128 bytes for future expansions. If we have enough
1340 * room at tail, reallocate without expansion only if skb is cloned.
1342 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1344 if (eat
> 0 || skb_cloned(skb
)) {
1345 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1350 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
1353 /* Optimization: no fragments, no reasons to preestimate
1354 * size of pulled pages. Superb.
1356 if (!skb_has_frag_list(skb
))
1359 /* Estimate size of pulled pages. */
1361 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1362 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1369 /* If we need update frag list, we are in troubles.
1370 * Certainly, it possible to add an offset to skb data,
1371 * but taking into account that pulling is expected to
1372 * be very rare operation, it is worth to fight against
1373 * further bloating skb head and crucify ourselves here instead.
1374 * Pure masohism, indeed. 8)8)
1377 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1378 struct sk_buff
*clone
= NULL
;
1379 struct sk_buff
*insp
= NULL
;
1384 if (list
->len
<= eat
) {
1385 /* Eaten as whole. */
1390 /* Eaten partially. */
1392 if (skb_shared(list
)) {
1393 /* Sucks! We need to fork list. :-( */
1394 clone
= skb_clone(list
, GFP_ATOMIC
);
1400 /* This may be pulled without
1404 if (!pskb_pull(list
, eat
)) {
1412 /* Free pulled out fragments. */
1413 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1414 skb_shinfo(skb
)->frag_list
= list
->next
;
1417 /* And insert new clone at head. */
1420 skb_shinfo(skb
)->frag_list
= clone
;
1423 /* Success! Now we may commit changes to skb data. */
1428 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1429 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1432 skb_frag_unref(skb
, i
);
1435 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1437 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1438 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[k
], eat
);
1444 skb_shinfo(skb
)->nr_frags
= k
;
1447 skb
->data_len
-= delta
;
1449 return skb_tail_pointer(skb
);
1451 EXPORT_SYMBOL(__pskb_pull_tail
);
1454 * skb_copy_bits - copy bits from skb to kernel buffer
1456 * @offset: offset in source
1457 * @to: destination buffer
1458 * @len: number of bytes to copy
1460 * Copy the specified number of bytes from the source skb to the
1461 * destination buffer.
1464 * If its prototype is ever changed,
1465 * check arch/{*}/net/{*}.S files,
1466 * since it is called from BPF assembly code.
1468 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1470 int start
= skb_headlen(skb
);
1471 struct sk_buff
*frag_iter
;
1474 if (offset
> (int)skb
->len
- len
)
1478 if ((copy
= start
- offset
) > 0) {
1481 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
1482 if ((len
-= copy
) == 0)
1488 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1491 WARN_ON(start
> offset
+ len
);
1493 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1494 if ((copy
= end
- offset
) > 0) {
1500 vaddr
= kmap_skb_frag(&skb_shinfo(skb
)->frags
[i
]);
1502 vaddr
+ skb_shinfo(skb
)->frags
[i
].page_offset
+
1503 offset
- start
, copy
);
1504 kunmap_skb_frag(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 inline 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 off
= sk
->sk_sndmsg_off
;
1569 mlen
= PAGE_SIZE
- off
;
1570 if (mlen
< 64 && mlen
< *len
) {
1575 *len
= min_t(unsigned int, *len
, mlen
);
1578 memcpy(page_address(p
) + off
, page_address(page
) + *offset
, *len
);
1579 sk
->sk_sndmsg_off
+= *len
;
1587 * Fill page/offset/length into spd, if it can hold more pages.
1589 static inline int spd_fill_page(struct splice_pipe_desc
*spd
,
1590 struct pipe_inode_info
*pipe
, struct page
*page
,
1591 unsigned int *len
, unsigned int offset
,
1592 struct sk_buff
*skb
, int linear
,
1595 if (unlikely(spd
->nr_pages
== pipe
->buffers
))
1599 page
= linear_to_page(page
, len
, &offset
, skb
, sk
);
1605 spd
->pages
[spd
->nr_pages
] = page
;
1606 spd
->partial
[spd
->nr_pages
].len
= *len
;
1607 spd
->partial
[spd
->nr_pages
].offset
= offset
;
1613 static inline void __segment_seek(struct page
**page
, unsigned int *poff
,
1614 unsigned int *plen
, unsigned int off
)
1619 n
= *poff
/ PAGE_SIZE
;
1621 *page
= nth_page(*page
, n
);
1623 *poff
= *poff
% PAGE_SIZE
;
1627 static inline int __splice_segment(struct page
*page
, unsigned int poff
,
1628 unsigned int plen
, unsigned int *off
,
1629 unsigned int *len
, struct sk_buff
*skb
,
1630 struct splice_pipe_desc
*spd
, int linear
,
1632 struct pipe_inode_info
*pipe
)
1637 /* skip this segment if already processed */
1643 /* ignore any bits we already processed */
1645 __segment_seek(&page
, &poff
, &plen
, *off
);
1650 unsigned int flen
= min(*len
, plen
);
1652 /* the linear region may spread across several pages */
1653 flen
= min_t(unsigned int, flen
, PAGE_SIZE
- poff
);
1655 if (spd_fill_page(spd
, pipe
, page
, &flen
, poff
, skb
, linear
, sk
))
1658 __segment_seek(&page
, &poff
, &plen
, flen
);
1661 } while (*len
&& plen
);
1667 * Map linear and fragment data from the skb to spd. It reports failure if the
1668 * pipe is full or if we already spliced the requested length.
1670 static int __skb_splice_bits(struct sk_buff
*skb
, struct pipe_inode_info
*pipe
,
1671 unsigned int *offset
, unsigned int *len
,
1672 struct splice_pipe_desc
*spd
, struct sock
*sk
)
1677 * map the linear part
1679 if (__splice_segment(virt_to_page(skb
->data
),
1680 (unsigned long) skb
->data
& (PAGE_SIZE
- 1),
1682 offset
, len
, skb
, spd
, 1, sk
, pipe
))
1686 * then map the fragments
1688 for (seg
= 0; seg
< skb_shinfo(skb
)->nr_frags
; seg
++) {
1689 const skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[seg
];
1691 if (__splice_segment(skb_frag_page(f
),
1692 f
->page_offset
, skb_frag_size(f
),
1693 offset
, len
, skb
, spd
, 0, sk
, pipe
))
1701 * Map data from the skb to a pipe. Should handle both the linear part,
1702 * the fragments, and the frag list. It does NOT handle frag lists within
1703 * the frag list, if such a thing exists. We'd probably need to recurse to
1704 * handle that cleanly.
1706 int skb_splice_bits(struct sk_buff
*skb
, unsigned int offset
,
1707 struct pipe_inode_info
*pipe
, unsigned int tlen
,
1710 struct partial_page partial
[PIPE_DEF_BUFFERS
];
1711 struct page
*pages
[PIPE_DEF_BUFFERS
];
1712 struct splice_pipe_desc spd
= {
1716 .ops
= &sock_pipe_buf_ops
,
1717 .spd_release
= sock_spd_release
,
1719 struct sk_buff
*frag_iter
;
1720 struct sock
*sk
= skb
->sk
;
1723 if (splice_grow_spd(pipe
, &spd
))
1727 * __skb_splice_bits() only fails if the output has no room left,
1728 * so no point in going over the frag_list for the error case.
1730 if (__skb_splice_bits(skb
, pipe
, &offset
, &tlen
, &spd
, sk
))
1736 * now see if we have a frag_list to map
1738 skb_walk_frags(skb
, frag_iter
) {
1741 if (__skb_splice_bits(frag_iter
, pipe
, &offset
, &tlen
, &spd
, sk
))
1748 * Drop the socket lock, otherwise we have reverse
1749 * locking dependencies between sk_lock and i_mutex
1750 * here as compared to sendfile(). We enter here
1751 * with the socket lock held, and splice_to_pipe() will
1752 * grab the pipe inode lock. For sendfile() emulation,
1753 * we call into ->sendpage() with the i_mutex lock held
1754 * and networking will grab the socket lock.
1757 ret
= splice_to_pipe(pipe
, &spd
);
1761 splice_shrink_spd(pipe
, &spd
);
1766 * skb_store_bits - store bits from kernel buffer to skb
1767 * @skb: destination buffer
1768 * @offset: offset in destination
1769 * @from: source buffer
1770 * @len: number of bytes to copy
1772 * Copy the specified number of bytes from the source buffer to the
1773 * destination skb. This function handles all the messy bits of
1774 * traversing fragment lists and such.
1777 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
1779 int start
= skb_headlen(skb
);
1780 struct sk_buff
*frag_iter
;
1783 if (offset
> (int)skb
->len
- len
)
1786 if ((copy
= start
- offset
) > 0) {
1789 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
1790 if ((len
-= copy
) == 0)
1796 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1797 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1800 WARN_ON(start
> offset
+ len
);
1802 end
= start
+ skb_frag_size(frag
);
1803 if ((copy
= end
- offset
) > 0) {
1809 vaddr
= kmap_skb_frag(frag
);
1810 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1812 kunmap_skb_frag(vaddr
);
1814 if ((len
-= copy
) == 0)
1822 skb_walk_frags(skb
, frag_iter
) {
1825 WARN_ON(start
> offset
+ len
);
1827 end
= start
+ frag_iter
->len
;
1828 if ((copy
= end
- offset
) > 0) {
1831 if (skb_store_bits(frag_iter
, offset
- start
,
1834 if ((len
-= copy
) == 0)
1847 EXPORT_SYMBOL(skb_store_bits
);
1849 /* Checksum skb data. */
1851 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
1852 int len
, __wsum csum
)
1854 int start
= skb_headlen(skb
);
1855 int i
, copy
= start
- offset
;
1856 struct sk_buff
*frag_iter
;
1859 /* Checksum header. */
1863 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1864 if ((len
-= copy
) == 0)
1870 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1873 WARN_ON(start
> offset
+ len
);
1875 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1876 if ((copy
= end
- offset
) > 0) {
1879 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1883 vaddr
= kmap_skb_frag(frag
);
1884 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
1885 offset
- start
, copy
, 0);
1886 kunmap_skb_frag(vaddr
);
1887 csum
= csum_block_add(csum
, csum2
, pos
);
1896 skb_walk_frags(skb
, frag_iter
) {
1899 WARN_ON(start
> offset
+ len
);
1901 end
= start
+ frag_iter
->len
;
1902 if ((copy
= end
- offset
) > 0) {
1906 csum2
= skb_checksum(frag_iter
, offset
- start
,
1908 csum
= csum_block_add(csum
, csum2
, pos
);
1909 if ((len
-= copy
) == 0)
1920 EXPORT_SYMBOL(skb_checksum
);
1922 /* Both of above in one bottle. */
1924 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
1925 u8
*to
, int len
, __wsum csum
)
1927 int start
= skb_headlen(skb
);
1928 int i
, copy
= start
- offset
;
1929 struct sk_buff
*frag_iter
;
1936 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
1938 if ((len
-= copy
) == 0)
1945 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1948 WARN_ON(start
> offset
+ len
);
1950 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1951 if ((copy
= end
- offset
) > 0) {
1954 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1958 vaddr
= kmap_skb_frag(frag
);
1959 csum2
= csum_partial_copy_nocheck(vaddr
+
1963 kunmap_skb_frag(vaddr
);
1964 csum
= csum_block_add(csum
, csum2
, pos
);
1974 skb_walk_frags(skb
, frag_iter
) {
1978 WARN_ON(start
> offset
+ len
);
1980 end
= start
+ frag_iter
->len
;
1981 if ((copy
= end
- offset
) > 0) {
1984 csum2
= skb_copy_and_csum_bits(frag_iter
,
1987 csum
= csum_block_add(csum
, csum2
, pos
);
1988 if ((len
-= copy
) == 0)
1999 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
2001 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
2006 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
2007 csstart
= skb_checksum_start_offset(skb
);
2009 csstart
= skb_headlen(skb
);
2011 BUG_ON(csstart
> skb_headlen(skb
));
2013 skb_copy_from_linear_data(skb
, to
, csstart
);
2016 if (csstart
!= skb
->len
)
2017 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
2018 skb
->len
- csstart
, 0);
2020 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2021 long csstuff
= csstart
+ skb
->csum_offset
;
2023 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
2026 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
2029 * skb_dequeue - remove from the head of the queue
2030 * @list: list to dequeue from
2032 * Remove the head of the list. The list lock is taken so the function
2033 * may be used safely with other locking list functions. The head item is
2034 * returned or %NULL if the list is empty.
2037 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
2039 unsigned long flags
;
2040 struct sk_buff
*result
;
2042 spin_lock_irqsave(&list
->lock
, flags
);
2043 result
= __skb_dequeue(list
);
2044 spin_unlock_irqrestore(&list
->lock
, flags
);
2047 EXPORT_SYMBOL(skb_dequeue
);
2050 * skb_dequeue_tail - remove from the tail of the queue
2051 * @list: list to dequeue from
2053 * Remove the tail of the list. The list lock is taken so the function
2054 * may be used safely with other locking list functions. The tail item is
2055 * returned or %NULL if the list is empty.
2057 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
2059 unsigned long flags
;
2060 struct sk_buff
*result
;
2062 spin_lock_irqsave(&list
->lock
, flags
);
2063 result
= __skb_dequeue_tail(list
);
2064 spin_unlock_irqrestore(&list
->lock
, flags
);
2067 EXPORT_SYMBOL(skb_dequeue_tail
);
2070 * skb_queue_purge - empty a list
2071 * @list: list to empty
2073 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2074 * the list and one reference dropped. This function takes the list
2075 * lock and is atomic with respect to other list locking functions.
2077 void skb_queue_purge(struct sk_buff_head
*list
)
2079 struct sk_buff
*skb
;
2080 while ((skb
= skb_dequeue(list
)) != NULL
)
2083 EXPORT_SYMBOL(skb_queue_purge
);
2086 * skb_queue_head - queue a buffer at the list head
2087 * @list: list to use
2088 * @newsk: buffer to queue
2090 * Queue a buffer at the start of the list. This function takes the
2091 * list lock and can be used safely with other locking &sk_buff functions
2094 * A buffer cannot be placed on two lists at the same time.
2096 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2098 unsigned long flags
;
2100 spin_lock_irqsave(&list
->lock
, flags
);
2101 __skb_queue_head(list
, newsk
);
2102 spin_unlock_irqrestore(&list
->lock
, flags
);
2104 EXPORT_SYMBOL(skb_queue_head
);
2107 * skb_queue_tail - queue a buffer at the list tail
2108 * @list: list to use
2109 * @newsk: buffer to queue
2111 * Queue a buffer at the tail of the list. This function takes the
2112 * list lock and can be used safely with other locking &sk_buff functions
2115 * A buffer cannot be placed on two lists at the same time.
2117 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2119 unsigned long flags
;
2121 spin_lock_irqsave(&list
->lock
, flags
);
2122 __skb_queue_tail(list
, newsk
);
2123 spin_unlock_irqrestore(&list
->lock
, flags
);
2125 EXPORT_SYMBOL(skb_queue_tail
);
2128 * skb_unlink - remove a buffer from a list
2129 * @skb: buffer to remove
2130 * @list: list to use
2132 * Remove a packet from a list. The list locks are taken and this
2133 * function is atomic with respect to other list locked calls
2135 * You must know what list the SKB is on.
2137 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
2139 unsigned long flags
;
2141 spin_lock_irqsave(&list
->lock
, flags
);
2142 __skb_unlink(skb
, list
);
2143 spin_unlock_irqrestore(&list
->lock
, flags
);
2145 EXPORT_SYMBOL(skb_unlink
);
2148 * skb_append - append a buffer
2149 * @old: buffer to insert after
2150 * @newsk: buffer to insert
2151 * @list: list to use
2153 * Place a packet after a given packet in a list. The list locks are taken
2154 * and this function is atomic with respect to other list locked calls.
2155 * A buffer cannot be placed on two lists at the same time.
2157 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2159 unsigned long flags
;
2161 spin_lock_irqsave(&list
->lock
, flags
);
2162 __skb_queue_after(list
, old
, newsk
);
2163 spin_unlock_irqrestore(&list
->lock
, flags
);
2165 EXPORT_SYMBOL(skb_append
);
2168 * skb_insert - insert a buffer
2169 * @old: buffer to insert before
2170 * @newsk: buffer to insert
2171 * @list: list to use
2173 * Place a packet before a given packet in a list. The list locks are
2174 * taken and this function is atomic with respect to other list locked
2177 * A buffer cannot be placed on two lists at the same time.
2179 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2181 unsigned long flags
;
2183 spin_lock_irqsave(&list
->lock
, flags
);
2184 __skb_insert(newsk
, old
->prev
, old
, list
);
2185 spin_unlock_irqrestore(&list
->lock
, flags
);
2187 EXPORT_SYMBOL(skb_insert
);
2189 static inline void skb_split_inside_header(struct sk_buff
*skb
,
2190 struct sk_buff
* skb1
,
2191 const u32 len
, const int pos
)
2195 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
2197 /* And move data appendix as is. */
2198 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
2199 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
2201 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
2202 skb_shinfo(skb
)->nr_frags
= 0;
2203 skb1
->data_len
= skb
->data_len
;
2204 skb1
->len
+= skb1
->data_len
;
2207 skb_set_tail_pointer(skb
, len
);
2210 static inline void skb_split_no_header(struct sk_buff
*skb
,
2211 struct sk_buff
* skb1
,
2212 const u32 len
, int pos
)
2215 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
2217 skb_shinfo(skb
)->nr_frags
= 0;
2218 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
2220 skb
->data_len
= len
- pos
;
2222 for (i
= 0; i
< nfrags
; i
++) {
2223 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
2225 if (pos
+ size
> len
) {
2226 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
2230 * We have two variants in this case:
2231 * 1. Move all the frag to the second
2232 * part, if it is possible. F.e.
2233 * this approach is mandatory for TUX,
2234 * where splitting is expensive.
2235 * 2. Split is accurately. We make this.
2237 skb_frag_ref(skb
, i
);
2238 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
2239 skb_frag_size_sub(&skb_shinfo(skb1
)->frags
[0], len
- pos
);
2240 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
], len
- pos
);
2241 skb_shinfo(skb
)->nr_frags
++;
2245 skb_shinfo(skb
)->nr_frags
++;
2248 skb_shinfo(skb1
)->nr_frags
= k
;
2252 * skb_split - Split fragmented skb to two parts at length len.
2253 * @skb: the buffer to split
2254 * @skb1: the buffer to receive the second part
2255 * @len: new length for skb
2257 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
2259 int pos
= skb_headlen(skb
);
2261 if (len
< pos
) /* Split line is inside header. */
2262 skb_split_inside_header(skb
, skb1
, len
, pos
);
2263 else /* Second chunk has no header, nothing to copy. */
2264 skb_split_no_header(skb
, skb1
, len
, pos
);
2266 EXPORT_SYMBOL(skb_split
);
2268 /* Shifting from/to a cloned skb is a no-go.
2270 * Caller cannot keep skb_shinfo related pointers past calling here!
2272 static int skb_prepare_for_shift(struct sk_buff
*skb
)
2274 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2278 * skb_shift - Shifts paged data partially from skb to another
2279 * @tgt: buffer into which tail data gets added
2280 * @skb: buffer from which the paged data comes from
2281 * @shiftlen: shift up to this many bytes
2283 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2284 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2285 * It's up to caller to free skb if everything was shifted.
2287 * If @tgt runs out of frags, the whole operation is aborted.
2289 * Skb cannot include anything else but paged data while tgt is allowed
2290 * to have non-paged data as well.
2292 * TODO: full sized shift could be optimized but that would need
2293 * specialized skb free'er to handle frags without up-to-date nr_frags.
2295 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
2297 int from
, to
, merge
, todo
;
2298 struct skb_frag_struct
*fragfrom
, *fragto
;
2300 BUG_ON(shiftlen
> skb
->len
);
2301 BUG_ON(skb_headlen(skb
)); /* Would corrupt stream */
2305 to
= skb_shinfo(tgt
)->nr_frags
;
2306 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2308 /* Actual merge is delayed until the point when we know we can
2309 * commit all, so that we don't have to undo partial changes
2312 !skb_can_coalesce(tgt
, to
, skb_frag_page(fragfrom
),
2313 fragfrom
->page_offset
)) {
2318 todo
-= skb_frag_size(fragfrom
);
2320 if (skb_prepare_for_shift(skb
) ||
2321 skb_prepare_for_shift(tgt
))
2324 /* All previous frag pointers might be stale! */
2325 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2326 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2328 skb_frag_size_add(fragto
, shiftlen
);
2329 skb_frag_size_sub(fragfrom
, shiftlen
);
2330 fragfrom
->page_offset
+= shiftlen
;
2338 /* Skip full, not-fitting skb to avoid expensive operations */
2339 if ((shiftlen
== skb
->len
) &&
2340 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
2343 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
2346 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
2347 if (to
== MAX_SKB_FRAGS
)
2350 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2351 fragto
= &skb_shinfo(tgt
)->frags
[to
];
2353 if (todo
>= skb_frag_size(fragfrom
)) {
2354 *fragto
= *fragfrom
;
2355 todo
-= skb_frag_size(fragfrom
);
2360 __skb_frag_ref(fragfrom
);
2361 fragto
->page
= fragfrom
->page
;
2362 fragto
->page_offset
= fragfrom
->page_offset
;
2363 skb_frag_size_set(fragto
, todo
);
2365 fragfrom
->page_offset
+= todo
;
2366 skb_frag_size_sub(fragfrom
, todo
);
2374 /* Ready to "commit" this state change to tgt */
2375 skb_shinfo(tgt
)->nr_frags
= to
;
2378 fragfrom
= &skb_shinfo(skb
)->frags
[0];
2379 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2381 skb_frag_size_add(fragto
, skb_frag_size(fragfrom
));
2382 __skb_frag_unref(fragfrom
);
2385 /* Reposition in the original skb */
2387 while (from
< skb_shinfo(skb
)->nr_frags
)
2388 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
2389 skb_shinfo(skb
)->nr_frags
= to
;
2391 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
2394 /* Most likely the tgt won't ever need its checksum anymore, skb on
2395 * the other hand might need it if it needs to be resent
2397 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
2398 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2400 /* Yak, is it really working this way? Some helper please? */
2401 skb
->len
-= shiftlen
;
2402 skb
->data_len
-= shiftlen
;
2403 skb
->truesize
-= shiftlen
;
2404 tgt
->len
+= shiftlen
;
2405 tgt
->data_len
+= shiftlen
;
2406 tgt
->truesize
+= shiftlen
;
2412 * skb_prepare_seq_read - Prepare a sequential read of skb data
2413 * @skb: the buffer to read
2414 * @from: lower offset of data to be read
2415 * @to: upper offset of data to be read
2416 * @st: state variable
2418 * Initializes the specified state variable. Must be called before
2419 * invoking skb_seq_read() for the first time.
2421 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
2422 unsigned int to
, struct skb_seq_state
*st
)
2424 st
->lower_offset
= from
;
2425 st
->upper_offset
= to
;
2426 st
->root_skb
= st
->cur_skb
= skb
;
2427 st
->frag_idx
= st
->stepped_offset
= 0;
2428 st
->frag_data
= NULL
;
2430 EXPORT_SYMBOL(skb_prepare_seq_read
);
2433 * skb_seq_read - Sequentially read skb data
2434 * @consumed: number of bytes consumed by the caller so far
2435 * @data: destination pointer for data to be returned
2436 * @st: state variable
2438 * Reads a block of skb data at &consumed relative to the
2439 * lower offset specified to skb_prepare_seq_read(). Assigns
2440 * the head of the data block to &data and returns the length
2441 * of the block or 0 if the end of the skb data or the upper
2442 * offset has been reached.
2444 * The caller is not required to consume all of the data
2445 * returned, i.e. &consumed is typically set to the number
2446 * of bytes already consumed and the next call to
2447 * skb_seq_read() will return the remaining part of the block.
2449 * Note 1: The size of each block of data returned can be arbitrary,
2450 * this limitation is the cost for zerocopy seqeuental
2451 * reads of potentially non linear data.
2453 * Note 2: Fragment lists within fragments are not implemented
2454 * at the moment, state->root_skb could be replaced with
2455 * a stack for this purpose.
2457 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
2458 struct skb_seq_state
*st
)
2460 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
2463 if (unlikely(abs_offset
>= st
->upper_offset
))
2467 block_limit
= skb_headlen(st
->cur_skb
) + st
->stepped_offset
;
2469 if (abs_offset
< block_limit
&& !st
->frag_data
) {
2470 *data
= st
->cur_skb
->data
+ (abs_offset
- st
->stepped_offset
);
2471 return block_limit
- abs_offset
;
2474 if (st
->frag_idx
== 0 && !st
->frag_data
)
2475 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
2477 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
2478 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
2479 block_limit
= skb_frag_size(frag
) + st
->stepped_offset
;
2481 if (abs_offset
< block_limit
) {
2483 st
->frag_data
= kmap_skb_frag(frag
);
2485 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
2486 (abs_offset
- st
->stepped_offset
);
2488 return block_limit
- abs_offset
;
2491 if (st
->frag_data
) {
2492 kunmap_skb_frag(st
->frag_data
);
2493 st
->frag_data
= NULL
;
2497 st
->stepped_offset
+= skb_frag_size(frag
);
2500 if (st
->frag_data
) {
2501 kunmap_skb_frag(st
->frag_data
);
2502 st
->frag_data
= NULL
;
2505 if (st
->root_skb
== st
->cur_skb
&& skb_has_frag_list(st
->root_skb
)) {
2506 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
2509 } else if (st
->cur_skb
->next
) {
2510 st
->cur_skb
= st
->cur_skb
->next
;
2517 EXPORT_SYMBOL(skb_seq_read
);
2520 * skb_abort_seq_read - Abort a sequential read of skb data
2521 * @st: state variable
2523 * Must be called if skb_seq_read() was not called until it
2526 void skb_abort_seq_read(struct skb_seq_state
*st
)
2529 kunmap_skb_frag(st
->frag_data
);
2531 EXPORT_SYMBOL(skb_abort_seq_read
);
2533 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2535 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
2536 struct ts_config
*conf
,
2537 struct ts_state
*state
)
2539 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
2542 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
2544 skb_abort_seq_read(TS_SKB_CB(state
));
2548 * skb_find_text - Find a text pattern in skb data
2549 * @skb: the buffer to look in
2550 * @from: search offset
2552 * @config: textsearch configuration
2553 * @state: uninitialized textsearch state variable
2555 * Finds a pattern in the skb data according to the specified
2556 * textsearch configuration. Use textsearch_next() to retrieve
2557 * subsequent occurrences of the pattern. Returns the offset
2558 * to the first occurrence or UINT_MAX if no match was found.
2560 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
2561 unsigned int to
, struct ts_config
*config
,
2562 struct ts_state
*state
)
2566 config
->get_next_block
= skb_ts_get_next_block
;
2567 config
->finish
= skb_ts_finish
;
2569 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
2571 ret
= textsearch_find(config
, state
);
2572 return (ret
<= to
- from
? ret
: UINT_MAX
);
2574 EXPORT_SYMBOL(skb_find_text
);
2577 * skb_append_datato_frags: - append the user data to a skb
2578 * @sk: sock structure
2579 * @skb: skb structure to be appened with user data.
2580 * @getfrag: call back function to be used for getting the user data
2581 * @from: pointer to user message iov
2582 * @length: length of the iov message
2584 * Description: This procedure append the user data in the fragment part
2585 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2587 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
2588 int (*getfrag
)(void *from
, char *to
, int offset
,
2589 int len
, int odd
, struct sk_buff
*skb
),
2590 void *from
, int length
)
2593 skb_frag_t
*frag
= NULL
;
2594 struct page
*page
= NULL
;
2600 /* Return error if we don't have space for new frag */
2601 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2602 if (frg_cnt
>= MAX_SKB_FRAGS
)
2605 /* allocate a new page for next frag */
2606 page
= alloc_pages(sk
->sk_allocation
, 0);
2608 /* If alloc_page fails just return failure and caller will
2609 * free previous allocated pages by doing kfree_skb()
2614 /* initialize the next frag */
2615 skb_fill_page_desc(skb
, frg_cnt
, page
, 0, 0);
2616 skb
->truesize
+= PAGE_SIZE
;
2617 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
2619 /* get the new initialized frag */
2620 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2621 frag
= &skb_shinfo(skb
)->frags
[frg_cnt
- 1];
2623 /* copy the user data to page */
2624 left
= PAGE_SIZE
- frag
->page_offset
;
2625 copy
= (length
> left
)? left
: length
;
2627 ret
= getfrag(from
, skb_frag_address(frag
) + skb_frag_size(frag
),
2628 offset
, copy
, 0, skb
);
2632 /* copy was successful so update the size parameters */
2633 skb_frag_size_add(frag
, copy
);
2635 skb
->data_len
+= copy
;
2639 } while (length
> 0);
2643 EXPORT_SYMBOL(skb_append_datato_frags
);
2646 * skb_pull_rcsum - pull skb and update receive checksum
2647 * @skb: buffer to update
2648 * @len: length of data pulled
2650 * This function performs an skb_pull on the packet and updates
2651 * the CHECKSUM_COMPLETE checksum. It should be used on
2652 * receive path processing instead of skb_pull unless you know
2653 * that the checksum difference is zero (e.g., a valid IP header)
2654 * or you are setting ip_summed to CHECKSUM_NONE.
2656 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
2658 BUG_ON(len
> skb
->len
);
2660 BUG_ON(skb
->len
< skb
->data_len
);
2661 skb_postpull_rcsum(skb
, skb
->data
, len
);
2662 return skb
->data
+= len
;
2664 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
2667 * skb_segment - Perform protocol segmentation on skb.
2668 * @skb: buffer to segment
2669 * @features: features for the output path (see dev->features)
2671 * This function performs segmentation on the given skb. It returns
2672 * a pointer to the first in a list of new skbs for the segments.
2673 * In case of error it returns ERR_PTR(err).
2675 struct sk_buff
*skb_segment(struct sk_buff
*skb
, netdev_features_t features
)
2677 struct sk_buff
*segs
= NULL
;
2678 struct sk_buff
*tail
= NULL
;
2679 struct sk_buff
*fskb
= skb_shinfo(skb
)->frag_list
;
2680 unsigned int mss
= skb_shinfo(skb
)->gso_size
;
2681 unsigned int doffset
= skb
->data
- skb_mac_header(skb
);
2682 unsigned int offset
= doffset
;
2683 unsigned int headroom
;
2685 int sg
= !!(features
& NETIF_F_SG
);
2686 int nfrags
= skb_shinfo(skb
)->nr_frags
;
2691 __skb_push(skb
, doffset
);
2692 headroom
= skb_headroom(skb
);
2693 pos
= skb_headlen(skb
);
2696 struct sk_buff
*nskb
;
2701 len
= skb
->len
- offset
;
2705 hsize
= skb_headlen(skb
) - offset
;
2708 if (hsize
> len
|| !sg
)
2711 if (!hsize
&& i
>= nfrags
) {
2712 BUG_ON(fskb
->len
!= len
);
2715 nskb
= skb_clone(fskb
, GFP_ATOMIC
);
2718 if (unlikely(!nskb
))
2721 hsize
= skb_end_pointer(nskb
) - nskb
->head
;
2722 if (skb_cow_head(nskb
, doffset
+ headroom
)) {
2727 nskb
->truesize
+= skb_end_pointer(nskb
) - nskb
->head
-
2729 skb_release_head_state(nskb
);
2730 __skb_push(nskb
, doffset
);
2732 nskb
= alloc_skb(hsize
+ doffset
+ headroom
,
2735 if (unlikely(!nskb
))
2738 skb_reserve(nskb
, headroom
);
2739 __skb_put(nskb
, doffset
);
2748 __copy_skb_header(nskb
, skb
);
2749 nskb
->mac_len
= skb
->mac_len
;
2751 /* nskb and skb might have different headroom */
2752 if (nskb
->ip_summed
== CHECKSUM_PARTIAL
)
2753 nskb
->csum_start
+= skb_headroom(nskb
) - headroom
;
2755 skb_reset_mac_header(nskb
);
2756 skb_set_network_header(nskb
, skb
->mac_len
);
2757 nskb
->transport_header
= (nskb
->network_header
+
2758 skb_network_header_len(skb
));
2759 skb_copy_from_linear_data(skb
, nskb
->data
, doffset
);
2761 if (fskb
!= skb_shinfo(skb
)->frag_list
)
2765 nskb
->ip_summed
= CHECKSUM_NONE
;
2766 nskb
->csum
= skb_copy_and_csum_bits(skb
, offset
,
2772 frag
= skb_shinfo(nskb
)->frags
;
2774 skb_copy_from_linear_data_offset(skb
, offset
,
2775 skb_put(nskb
, hsize
), hsize
);
2777 while (pos
< offset
+ len
&& i
< nfrags
) {
2778 *frag
= skb_shinfo(skb
)->frags
[i
];
2779 __skb_frag_ref(frag
);
2780 size
= skb_frag_size(frag
);
2783 frag
->page_offset
+= offset
- pos
;
2784 skb_frag_size_sub(frag
, offset
- pos
);
2787 skb_shinfo(nskb
)->nr_frags
++;
2789 if (pos
+ size
<= offset
+ len
) {
2793 skb_frag_size_sub(frag
, pos
+ size
- (offset
+ len
));
2800 if (pos
< offset
+ len
) {
2801 struct sk_buff
*fskb2
= fskb
;
2803 BUG_ON(pos
+ fskb
->len
!= offset
+ len
);
2809 fskb2
= skb_clone(fskb2
, GFP_ATOMIC
);
2815 SKB_FRAG_ASSERT(nskb
);
2816 skb_shinfo(nskb
)->frag_list
= fskb2
;
2820 nskb
->data_len
= len
- hsize
;
2821 nskb
->len
+= nskb
->data_len
;
2822 nskb
->truesize
+= nskb
->data_len
;
2823 } while ((offset
+= len
) < skb
->len
);
2828 while ((skb
= segs
)) {
2832 return ERR_PTR(err
);
2834 EXPORT_SYMBOL_GPL(skb_segment
);
2836 int skb_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2838 struct sk_buff
*p
= *head
;
2839 struct sk_buff
*nskb
;
2840 struct skb_shared_info
*skbinfo
= skb_shinfo(skb
);
2841 struct skb_shared_info
*pinfo
= skb_shinfo(p
);
2842 unsigned int headroom
;
2843 unsigned int len
= skb_gro_len(skb
);
2844 unsigned int offset
= skb_gro_offset(skb
);
2845 unsigned int headlen
= skb_headlen(skb
);
2847 if (p
->len
+ len
>= 65536)
2850 if (pinfo
->frag_list
)
2852 else if (headlen
<= offset
) {
2855 int i
= skbinfo
->nr_frags
;
2856 int nr_frags
= pinfo
->nr_frags
+ i
;
2860 if (nr_frags
> MAX_SKB_FRAGS
)
2863 pinfo
->nr_frags
= nr_frags
;
2864 skbinfo
->nr_frags
= 0;
2866 frag
= pinfo
->frags
+ nr_frags
;
2867 frag2
= skbinfo
->frags
+ i
;
2872 frag
->page_offset
+= offset
;
2873 skb_frag_size_sub(frag
, offset
);
2875 skb
->truesize
-= skb
->data_len
;
2876 skb
->len
-= skb
->data_len
;
2879 NAPI_GRO_CB(skb
)->free
= 1;
2881 } else if (skb_gro_len(p
) != pinfo
->gso_size
)
2884 headroom
= skb_headroom(p
);
2885 nskb
= alloc_skb(headroom
+ skb_gro_offset(p
), GFP_ATOMIC
);
2886 if (unlikely(!nskb
))
2889 __copy_skb_header(nskb
, p
);
2890 nskb
->mac_len
= p
->mac_len
;
2892 skb_reserve(nskb
, headroom
);
2893 __skb_put(nskb
, skb_gro_offset(p
));
2895 skb_set_mac_header(nskb
, skb_mac_header(p
) - p
->data
);
2896 skb_set_network_header(nskb
, skb_network_offset(p
));
2897 skb_set_transport_header(nskb
, skb_transport_offset(p
));
2899 __skb_pull(p
, skb_gro_offset(p
));
2900 memcpy(skb_mac_header(nskb
), skb_mac_header(p
),
2901 p
->data
- skb_mac_header(p
));
2903 *NAPI_GRO_CB(nskb
) = *NAPI_GRO_CB(p
);
2904 skb_shinfo(nskb
)->frag_list
= p
;
2905 skb_shinfo(nskb
)->gso_size
= pinfo
->gso_size
;
2906 pinfo
->gso_size
= 0;
2907 skb_header_release(p
);
2910 nskb
->data_len
+= p
->len
;
2911 nskb
->truesize
+= p
->truesize
;
2912 nskb
->len
+= p
->len
;
2915 nskb
->next
= p
->next
;
2921 p
->truesize
+= skb
->truesize
- len
;
2922 if (offset
> headlen
) {
2923 unsigned int eat
= offset
- headlen
;
2925 skbinfo
->frags
[0].page_offset
+= eat
;
2926 skb_frag_size_sub(&skbinfo
->frags
[0], eat
);
2927 skb
->data_len
-= eat
;
2932 __skb_pull(skb
, offset
);
2934 p
->prev
->next
= skb
;
2936 skb_header_release(skb
);
2939 NAPI_GRO_CB(p
)->count
++;
2944 NAPI_GRO_CB(skb
)->same_flow
= 1;
2947 EXPORT_SYMBOL_GPL(skb_gro_receive
);
2949 void __init
skb_init(void)
2951 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
2952 sizeof(struct sk_buff
),
2954 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2956 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
2957 (2*sizeof(struct sk_buff
)) +
2960 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2965 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2966 * @skb: Socket buffer containing the buffers to be mapped
2967 * @sg: The scatter-gather list to map into
2968 * @offset: The offset into the buffer's contents to start mapping
2969 * @len: Length of buffer space to be mapped
2971 * Fill the specified scatter-gather list with mappings/pointers into a
2972 * region of the buffer space attached to a socket buffer.
2975 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
2977 int start
= skb_headlen(skb
);
2978 int i
, copy
= start
- offset
;
2979 struct sk_buff
*frag_iter
;
2985 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
2987 if ((len
-= copy
) == 0)
2992 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2995 WARN_ON(start
> offset
+ len
);
2997 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
2998 if ((copy
= end
- offset
) > 0) {
2999 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
3003 sg_set_page(&sg
[elt
], skb_frag_page(frag
), copy
,
3004 frag
->page_offset
+offset
-start
);
3013 skb_walk_frags(skb
, frag_iter
) {
3016 WARN_ON(start
> offset
+ len
);
3018 end
= start
+ frag_iter
->len
;
3019 if ((copy
= end
- offset
) > 0) {
3022 elt
+= __skb_to_sgvec(frag_iter
, sg
+elt
, offset
- start
,
3024 if ((len
-= copy
) == 0)
3034 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
3036 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
);
3038 sg_mark_end(&sg
[nsg
- 1]);
3042 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
3045 * skb_cow_data - Check that a socket buffer's data buffers are writable
3046 * @skb: The socket buffer to check.
3047 * @tailbits: Amount of trailing space to be added
3048 * @trailer: Returned pointer to the skb where the @tailbits space begins
3050 * Make sure that the data buffers attached to a socket buffer are
3051 * writable. If they are not, private copies are made of the data buffers
3052 * and the socket buffer is set to use these instead.
3054 * If @tailbits is given, make sure that there is space to write @tailbits
3055 * bytes of data beyond current end of socket buffer. @trailer will be
3056 * set to point to the skb in which this space begins.
3058 * The number of scatterlist elements required to completely map the
3059 * COW'd and extended socket buffer will be returned.
3061 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
3065 struct sk_buff
*skb1
, **skb_p
;
3067 /* If skb is cloned or its head is paged, reallocate
3068 * head pulling out all the pages (pages are considered not writable
3069 * at the moment even if they are anonymous).
3071 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
3072 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
3075 /* Easy case. Most of packets will go this way. */
3076 if (!skb_has_frag_list(skb
)) {
3077 /* A little of trouble, not enough of space for trailer.
3078 * This should not happen, when stack is tuned to generate
3079 * good frames. OK, on miss we reallocate and reserve even more
3080 * space, 128 bytes is fair. */
3082 if (skb_tailroom(skb
) < tailbits
&&
3083 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
3091 /* Misery. We are in troubles, going to mincer fragments... */
3094 skb_p
= &skb_shinfo(skb
)->frag_list
;
3097 while ((skb1
= *skb_p
) != NULL
) {
3100 /* The fragment is partially pulled by someone,
3101 * this can happen on input. Copy it and everything
3104 if (skb_shared(skb1
))
3107 /* If the skb is the last, worry about trailer. */
3109 if (skb1
->next
== NULL
&& tailbits
) {
3110 if (skb_shinfo(skb1
)->nr_frags
||
3111 skb_has_frag_list(skb1
) ||
3112 skb_tailroom(skb1
) < tailbits
)
3113 ntail
= tailbits
+ 128;
3119 skb_shinfo(skb1
)->nr_frags
||
3120 skb_has_frag_list(skb1
)) {
3121 struct sk_buff
*skb2
;
3123 /* Fuck, we are miserable poor guys... */
3125 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
3127 skb2
= skb_copy_expand(skb1
,
3131 if (unlikely(skb2
== NULL
))
3135 skb_set_owner_w(skb2
, skb1
->sk
);
3137 /* Looking around. Are we still alive?
3138 * OK, link new skb, drop old one */
3140 skb2
->next
= skb1
->next
;
3147 skb_p
= &skb1
->next
;
3152 EXPORT_SYMBOL_GPL(skb_cow_data
);
3154 static void sock_rmem_free(struct sk_buff
*skb
)
3156 struct sock
*sk
= skb
->sk
;
3158 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
3162 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3164 int sock_queue_err_skb(struct sock
*sk
, struct sk_buff
*skb
)
3168 if (atomic_read(&sk
->sk_rmem_alloc
) + skb
->truesize
>=
3169 (unsigned)sk
->sk_rcvbuf
)
3174 skb
->destructor
= sock_rmem_free
;
3175 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
3177 /* before exiting rcu section, make sure dst is refcounted */
3180 skb_queue_tail(&sk
->sk_error_queue
, skb
);
3181 if (!sock_flag(sk
, SOCK_DEAD
))
3182 sk
->sk_data_ready(sk
, len
);
3185 EXPORT_SYMBOL(sock_queue_err_skb
);
3187 void skb_tstamp_tx(struct sk_buff
*orig_skb
,
3188 struct skb_shared_hwtstamps
*hwtstamps
)
3190 struct sock
*sk
= orig_skb
->sk
;
3191 struct sock_exterr_skb
*serr
;
3192 struct sk_buff
*skb
;
3198 skb
= skb_clone(orig_skb
, GFP_ATOMIC
);
3203 *skb_hwtstamps(skb
) =
3207 * no hardware time stamps available,
3208 * so keep the shared tx_flags and only
3209 * store software time stamp
3211 skb
->tstamp
= ktime_get_real();
3214 serr
= SKB_EXT_ERR(skb
);
3215 memset(serr
, 0, sizeof(*serr
));
3216 serr
->ee
.ee_errno
= ENOMSG
;
3217 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TIMESTAMPING
;
3219 err
= sock_queue_err_skb(sk
, skb
);
3224 EXPORT_SYMBOL_GPL(skb_tstamp_tx
);
3226 void skb_complete_wifi_ack(struct sk_buff
*skb
, bool acked
)
3228 struct sock
*sk
= skb
->sk
;
3229 struct sock_exterr_skb
*serr
;
3232 skb
->wifi_acked_valid
= 1;
3233 skb
->wifi_acked
= acked
;
3235 serr
= SKB_EXT_ERR(skb
);
3236 memset(serr
, 0, sizeof(*serr
));
3237 serr
->ee
.ee_errno
= ENOMSG
;
3238 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TXSTATUS
;
3240 err
= sock_queue_err_skb(sk
, skb
);
3244 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack
);
3248 * skb_partial_csum_set - set up and verify partial csum values for packet
3249 * @skb: the skb to set
3250 * @start: the number of bytes after skb->data to start checksumming.
3251 * @off: the offset from start to place the checksum.
3253 * For untrusted partially-checksummed packets, we need to make sure the values
3254 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3256 * This function checks and sets those values and skb->ip_summed: if this
3257 * returns false you should drop the packet.
3259 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
3261 if (unlikely(start
> skb_headlen(skb
)) ||
3262 unlikely((int)start
+ off
> skb_headlen(skb
) - 2)) {
3263 if (net_ratelimit())
3265 "bad partial csum: csum=%u/%u len=%u\n",
3266 start
, off
, skb_headlen(skb
));
3269 skb
->ip_summed
= CHECKSUM_PARTIAL
;
3270 skb
->csum_start
= skb_headroom(skb
) + start
;
3271 skb
->csum_offset
= off
;
3274 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);
3276 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
3278 if (net_ratelimit())
3279 pr_warning("%s: received packets cannot be forwarded"
3280 " while LRO is enabled\n", skb
->dev
->name
);
3282 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);