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 * skb_panic - private function for out-of-line support
112 * @msg: skb_over_panic or skb_under_panic
114 * Out-of-line support for skb_put() and skb_push().
115 * Called via the wrapper skb_over_panic() or skb_under_panic().
116 * Keep out of line to prevent kernel bloat.
117 * __builtin_return_address is not used because it is not always reliable.
119 static void skb_panic(struct sk_buff
*skb
, unsigned int sz
, void *addr
,
122 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
123 msg
, addr
, skb
->len
, sz
, skb
->head
, skb
->data
,
124 (unsigned long)skb
->tail
, (unsigned long)skb
->end
,
125 skb
->dev
? skb
->dev
->name
: "<NULL>");
129 static void skb_over_panic(struct sk_buff
*skb
, unsigned int sz
, void *addr
)
131 skb_panic(skb
, sz
, addr
, __func__
);
134 static void skb_under_panic(struct sk_buff
*skb
, unsigned int sz
, void *addr
)
136 skb_panic(skb
, sz
, addr
, __func__
);
140 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
141 * the caller if emergency pfmemalloc reserves are being used. If it is and
142 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
143 * may be used. Otherwise, the packet data may be discarded until enough
146 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
147 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
149 static void *__kmalloc_reserve(size_t size
, gfp_t flags
, int node
,
150 unsigned long ip
, bool *pfmemalloc
)
153 bool ret_pfmemalloc
= false;
156 * Try a regular allocation, when that fails and we're not entitled
157 * to the reserves, fail.
159 obj
= kmalloc_node_track_caller(size
,
160 flags
| __GFP_NOMEMALLOC
| __GFP_NOWARN
,
162 if (obj
|| !(gfp_pfmemalloc_allowed(flags
)))
165 /* Try again but now we are using pfmemalloc reserves */
166 ret_pfmemalloc
= true;
167 obj
= kmalloc_node_track_caller(size
, flags
, node
);
171 *pfmemalloc
= ret_pfmemalloc
;
176 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
177 * 'private' fields and also do memory statistics to find all the
182 struct sk_buff
*__alloc_skb_head(gfp_t gfp_mask
, int node
)
187 skb
= kmem_cache_alloc_node(skbuff_head_cache
,
188 gfp_mask
& ~__GFP_DMA
, node
);
193 * Only clear those fields we need to clear, not those that we will
194 * actually initialise below. Hence, don't put any more fields after
195 * the tail pointer in struct sk_buff!
197 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
199 skb
->truesize
= sizeof(struct sk_buff
);
200 atomic_set(&skb
->users
, 1);
202 #ifdef NET_SKBUFF_DATA_USES_OFFSET
203 skb
->mac_header
= ~0U;
210 * __alloc_skb - allocate a network buffer
211 * @size: size to allocate
212 * @gfp_mask: allocation mask
213 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
214 * instead of head cache and allocate a cloned (child) skb.
215 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
216 * allocations in case the data is required for writeback
217 * @node: numa node to allocate memory on
219 * Allocate a new &sk_buff. The returned buffer has no headroom and a
220 * tail room of at least size bytes. The object has a reference count
221 * of one. The return is the buffer. On a failure the return is %NULL.
223 * Buffers may only be allocated from interrupts using a @gfp_mask of
226 struct sk_buff
*__alloc_skb(unsigned int size
, gfp_t gfp_mask
,
229 struct kmem_cache
*cache
;
230 struct skb_shared_info
*shinfo
;
235 cache
= (flags
& SKB_ALLOC_FCLONE
)
236 ? skbuff_fclone_cache
: skbuff_head_cache
;
238 if (sk_memalloc_socks() && (flags
& SKB_ALLOC_RX
))
239 gfp_mask
|= __GFP_MEMALLOC
;
242 skb
= kmem_cache_alloc_node(cache
, gfp_mask
& ~__GFP_DMA
, node
);
247 /* We do our best to align skb_shared_info on a separate cache
248 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
249 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
250 * Both skb->head and skb_shared_info are cache line aligned.
252 size
= SKB_DATA_ALIGN(size
);
253 size
+= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
254 data
= kmalloc_reserve(size
, gfp_mask
, node
, &pfmemalloc
);
257 /* kmalloc(size) might give us more room than requested.
258 * Put skb_shared_info exactly at the end of allocated zone,
259 * to allow max possible filling before reallocation.
261 size
= SKB_WITH_OVERHEAD(ksize(data
));
262 prefetchw(data
+ size
);
265 * Only clear those fields we need to clear, not those that we will
266 * actually initialise below. Hence, don't put any more fields after
267 * the tail pointer in struct sk_buff!
269 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
270 /* Account for allocated memory : skb + skb->head */
271 skb
->truesize
= SKB_TRUESIZE(size
);
272 skb
->pfmemalloc
= pfmemalloc
;
273 atomic_set(&skb
->users
, 1);
276 skb_reset_tail_pointer(skb
);
277 skb
->end
= skb
->tail
+ size
;
278 #ifdef NET_SKBUFF_DATA_USES_OFFSET
279 skb
->mac_header
= ~0U;
280 skb
->transport_header
= ~0U;
283 /* make sure we initialize shinfo sequentially */
284 shinfo
= skb_shinfo(skb
);
285 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
286 atomic_set(&shinfo
->dataref
, 1);
287 kmemcheck_annotate_variable(shinfo
->destructor_arg
);
289 if (flags
& SKB_ALLOC_FCLONE
) {
290 struct sk_buff
*child
= skb
+ 1;
291 atomic_t
*fclone_ref
= (atomic_t
*) (child
+ 1);
293 kmemcheck_annotate_bitfield(child
, flags1
);
294 kmemcheck_annotate_bitfield(child
, flags2
);
295 skb
->fclone
= SKB_FCLONE_ORIG
;
296 atomic_set(fclone_ref
, 1);
298 child
->fclone
= SKB_FCLONE_UNAVAILABLE
;
299 child
->pfmemalloc
= pfmemalloc
;
304 kmem_cache_free(cache
, skb
);
308 EXPORT_SYMBOL(__alloc_skb
);
311 * build_skb - build a network buffer
312 * @data: data buffer provided by caller
313 * @frag_size: size of fragment, or 0 if head was kmalloced
315 * Allocate a new &sk_buff. Caller provides space holding head and
316 * skb_shared_info. @data must have been allocated by kmalloc()
317 * The return is the new skb buffer.
318 * On a failure the return is %NULL, and @data is not freed.
320 * Before IO, driver allocates only data buffer where NIC put incoming frame
321 * Driver should add room at head (NET_SKB_PAD) and
322 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
323 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
324 * before giving packet to stack.
325 * RX rings only contains data buffers, not full skbs.
327 struct sk_buff
*build_skb(void *data
, unsigned int frag_size
)
329 struct skb_shared_info
*shinfo
;
331 unsigned int size
= frag_size
? : ksize(data
);
333 skb
= kmem_cache_alloc(skbuff_head_cache
, GFP_ATOMIC
);
337 size
-= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
339 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
340 skb
->truesize
= SKB_TRUESIZE(size
);
341 skb
->head_frag
= frag_size
!= 0;
342 atomic_set(&skb
->users
, 1);
345 skb_reset_tail_pointer(skb
);
346 skb
->end
= skb
->tail
+ size
;
347 #ifdef NET_SKBUFF_DATA_USES_OFFSET
348 skb
->mac_header
= ~0U;
349 skb
->transport_header
= ~0U;
352 /* make sure we initialize shinfo sequentially */
353 shinfo
= skb_shinfo(skb
);
354 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
355 atomic_set(&shinfo
->dataref
, 1);
356 kmemcheck_annotate_variable(shinfo
->destructor_arg
);
360 EXPORT_SYMBOL(build_skb
);
362 struct netdev_alloc_cache
{
363 struct page_frag frag
;
364 /* we maintain a pagecount bias, so that we dont dirty cache line
365 * containing page->_count every time we allocate a fragment.
367 unsigned int pagecnt_bias
;
369 static DEFINE_PER_CPU(struct netdev_alloc_cache
, netdev_alloc_cache
);
371 static void *__netdev_alloc_frag(unsigned int fragsz
, gfp_t gfp_mask
)
373 struct netdev_alloc_cache
*nc
;
378 local_irq_save(flags
);
379 nc
= &__get_cpu_var(netdev_alloc_cache
);
380 if (unlikely(!nc
->frag
.page
)) {
382 for (order
= NETDEV_FRAG_PAGE_MAX_ORDER
; ;) {
383 gfp_t gfp
= gfp_mask
;
386 gfp
|= __GFP_COMP
| __GFP_NOWARN
;
387 nc
->frag
.page
= alloc_pages(gfp
, order
);
388 if (likely(nc
->frag
.page
))
393 nc
->frag
.size
= PAGE_SIZE
<< order
;
395 atomic_set(&nc
->frag
.page
->_count
, NETDEV_PAGECNT_MAX_BIAS
);
396 nc
->pagecnt_bias
= NETDEV_PAGECNT_MAX_BIAS
;
400 if (nc
->frag
.offset
+ fragsz
> nc
->frag
.size
) {
401 /* avoid unnecessary locked operations if possible */
402 if ((atomic_read(&nc
->frag
.page
->_count
) == nc
->pagecnt_bias
) ||
403 atomic_sub_and_test(nc
->pagecnt_bias
, &nc
->frag
.page
->_count
))
408 data
= page_address(nc
->frag
.page
) + nc
->frag
.offset
;
409 nc
->frag
.offset
+= fragsz
;
412 local_irq_restore(flags
);
417 * netdev_alloc_frag - allocate a page fragment
418 * @fragsz: fragment size
420 * Allocates a frag from a page for receive buffer.
421 * Uses GFP_ATOMIC allocations.
423 void *netdev_alloc_frag(unsigned int fragsz
)
425 return __netdev_alloc_frag(fragsz
, GFP_ATOMIC
| __GFP_COLD
);
427 EXPORT_SYMBOL(netdev_alloc_frag
);
430 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
431 * @dev: network device to receive on
432 * @length: length to allocate
433 * @gfp_mask: get_free_pages mask, passed to alloc_skb
435 * Allocate a new &sk_buff and assign it a usage count of one. The
436 * buffer has unspecified headroom built in. Users should allocate
437 * the headroom they think they need without accounting for the
438 * built in space. The built in space is used for optimisations.
440 * %NULL is returned if there is no free memory.
442 struct sk_buff
*__netdev_alloc_skb(struct net_device
*dev
,
443 unsigned int length
, gfp_t gfp_mask
)
445 struct sk_buff
*skb
= NULL
;
446 unsigned int fragsz
= SKB_DATA_ALIGN(length
+ NET_SKB_PAD
) +
447 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
449 if (fragsz
<= PAGE_SIZE
&& !(gfp_mask
& (__GFP_WAIT
| GFP_DMA
))) {
452 if (sk_memalloc_socks())
453 gfp_mask
|= __GFP_MEMALLOC
;
455 data
= __netdev_alloc_frag(fragsz
, gfp_mask
);
458 skb
= build_skb(data
, fragsz
);
460 put_page(virt_to_head_page(data
));
463 skb
= __alloc_skb(length
+ NET_SKB_PAD
, gfp_mask
,
464 SKB_ALLOC_RX
, NUMA_NO_NODE
);
467 skb_reserve(skb
, NET_SKB_PAD
);
472 EXPORT_SYMBOL(__netdev_alloc_skb
);
474 void skb_add_rx_frag(struct sk_buff
*skb
, int i
, struct page
*page
, int off
,
475 int size
, unsigned int truesize
)
477 skb_fill_page_desc(skb
, i
, page
, off
, size
);
479 skb
->data_len
+= size
;
480 skb
->truesize
+= truesize
;
482 EXPORT_SYMBOL(skb_add_rx_frag
);
484 static void skb_drop_list(struct sk_buff
**listp
)
486 struct sk_buff
*list
= *listp
;
491 struct sk_buff
*this = list
;
497 static inline void skb_drop_fraglist(struct sk_buff
*skb
)
499 skb_drop_list(&skb_shinfo(skb
)->frag_list
);
502 static void skb_clone_fraglist(struct sk_buff
*skb
)
504 struct sk_buff
*list
;
506 skb_walk_frags(skb
, list
)
510 static void skb_free_head(struct sk_buff
*skb
)
513 put_page(virt_to_head_page(skb
->head
));
518 static void skb_release_data(struct sk_buff
*skb
)
521 !atomic_sub_return(skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1,
522 &skb_shinfo(skb
)->dataref
)) {
523 if (skb_shinfo(skb
)->nr_frags
) {
525 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
526 skb_frag_unref(skb
, i
);
530 * If skb buf is from userspace, we need to notify the caller
531 * the lower device DMA has done;
533 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
534 struct ubuf_info
*uarg
;
536 uarg
= skb_shinfo(skb
)->destructor_arg
;
538 uarg
->callback(uarg
, true);
541 if (skb_has_frag_list(skb
))
542 skb_drop_fraglist(skb
);
549 * Free an skbuff by memory without cleaning the state.
551 static void kfree_skbmem(struct sk_buff
*skb
)
553 struct sk_buff
*other
;
554 atomic_t
*fclone_ref
;
556 switch (skb
->fclone
) {
557 case SKB_FCLONE_UNAVAILABLE
:
558 kmem_cache_free(skbuff_head_cache
, skb
);
561 case SKB_FCLONE_ORIG
:
562 fclone_ref
= (atomic_t
*) (skb
+ 2);
563 if (atomic_dec_and_test(fclone_ref
))
564 kmem_cache_free(skbuff_fclone_cache
, skb
);
567 case SKB_FCLONE_CLONE
:
568 fclone_ref
= (atomic_t
*) (skb
+ 1);
571 /* The clone portion is available for
572 * fast-cloning again.
574 skb
->fclone
= SKB_FCLONE_UNAVAILABLE
;
576 if (atomic_dec_and_test(fclone_ref
))
577 kmem_cache_free(skbuff_fclone_cache
, other
);
582 static void skb_release_head_state(struct sk_buff
*skb
)
586 secpath_put(skb
->sp
);
588 if (skb
->destructor
) {
590 skb
->destructor(skb
);
592 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
593 nf_conntrack_put(skb
->nfct
);
595 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
596 nf_conntrack_put_reasm(skb
->nfct_reasm
);
598 #ifdef CONFIG_BRIDGE_NETFILTER
599 nf_bridge_put(skb
->nf_bridge
);
601 /* XXX: IS this still necessary? - JHS */
602 #ifdef CONFIG_NET_SCHED
604 #ifdef CONFIG_NET_CLS_ACT
610 /* Free everything but the sk_buff shell. */
611 static void skb_release_all(struct sk_buff
*skb
)
613 skb_release_head_state(skb
);
614 if (likely(skb
->data
))
615 skb_release_data(skb
);
619 * __kfree_skb - private function
622 * Free an sk_buff. Release anything attached to the buffer.
623 * Clean the state. This is an internal helper function. Users should
624 * always call kfree_skb
627 void __kfree_skb(struct sk_buff
*skb
)
629 skb_release_all(skb
);
632 EXPORT_SYMBOL(__kfree_skb
);
635 * kfree_skb - free an sk_buff
636 * @skb: buffer to free
638 * Drop a reference to the buffer and free it if the usage count has
641 void kfree_skb(struct sk_buff
*skb
)
645 if (likely(atomic_read(&skb
->users
) == 1))
647 else if (likely(!atomic_dec_and_test(&skb
->users
)))
649 trace_kfree_skb(skb
, __builtin_return_address(0));
652 EXPORT_SYMBOL(kfree_skb
);
655 * skb_tx_error - report an sk_buff xmit error
656 * @skb: buffer that triggered an error
658 * Report xmit error if a device callback is tracking this skb.
659 * skb must be freed afterwards.
661 void skb_tx_error(struct sk_buff
*skb
)
663 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
664 struct ubuf_info
*uarg
;
666 uarg
= skb_shinfo(skb
)->destructor_arg
;
668 uarg
->callback(uarg
, false);
669 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
672 EXPORT_SYMBOL(skb_tx_error
);
675 * consume_skb - free an skbuff
676 * @skb: buffer to free
678 * Drop a ref to the buffer and free it if the usage count has hit zero
679 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
680 * is being dropped after a failure and notes that
682 void consume_skb(struct sk_buff
*skb
)
686 if (likely(atomic_read(&skb
->users
) == 1))
688 else if (likely(!atomic_dec_and_test(&skb
->users
)))
690 trace_consume_skb(skb
);
693 EXPORT_SYMBOL(consume_skb
);
695 static void __copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
697 new->tstamp
= old
->tstamp
;
699 new->transport_header
= old
->transport_header
;
700 new->network_header
= old
->network_header
;
701 new->mac_header
= old
->mac_header
;
702 new->inner_transport_header
= old
->inner_transport_header
;
703 new->inner_network_header
= old
->inner_network_header
;
704 new->inner_mac_header
= old
->inner_mac_header
;
705 skb_dst_copy(new, old
);
706 new->rxhash
= old
->rxhash
;
707 new->ooo_okay
= old
->ooo_okay
;
708 new->l4_rxhash
= old
->l4_rxhash
;
709 new->no_fcs
= old
->no_fcs
;
710 new->encapsulation
= old
->encapsulation
;
712 new->sp
= secpath_get(old
->sp
);
714 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
715 new->csum
= old
->csum
;
716 new->local_df
= old
->local_df
;
717 new->pkt_type
= old
->pkt_type
;
718 new->ip_summed
= old
->ip_summed
;
719 skb_copy_queue_mapping(new, old
);
720 new->priority
= old
->priority
;
721 #if IS_ENABLED(CONFIG_IP_VS)
722 new->ipvs_property
= old
->ipvs_property
;
724 new->pfmemalloc
= old
->pfmemalloc
;
725 new->protocol
= old
->protocol
;
726 new->mark
= old
->mark
;
727 new->skb_iif
= old
->skb_iif
;
729 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
730 new->nf_trace
= old
->nf_trace
;
732 #ifdef CONFIG_NET_SCHED
733 new->tc_index
= old
->tc_index
;
734 #ifdef CONFIG_NET_CLS_ACT
735 new->tc_verd
= old
->tc_verd
;
738 new->vlan_proto
= old
->vlan_proto
;
739 new->vlan_tci
= old
->vlan_tci
;
741 skb_copy_secmark(new, old
);
745 * You should not add any new code to this function. Add it to
746 * __copy_skb_header above instead.
748 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
750 #define C(x) n->x = skb->x
752 n
->next
= n
->prev
= NULL
;
754 __copy_skb_header(n
, skb
);
759 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
762 n
->destructor
= NULL
;
769 atomic_set(&n
->users
, 1);
771 atomic_inc(&(skb_shinfo(skb
)->dataref
));
779 * skb_morph - morph one skb into another
780 * @dst: the skb to receive the contents
781 * @src: the skb to supply the contents
783 * This is identical to skb_clone except that the target skb is
784 * supplied by the user.
786 * The target skb is returned upon exit.
788 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
790 skb_release_all(dst
);
791 return __skb_clone(dst
, src
);
793 EXPORT_SYMBOL_GPL(skb_morph
);
796 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
797 * @skb: the skb to modify
798 * @gfp_mask: allocation priority
800 * This must be called on SKBTX_DEV_ZEROCOPY skb.
801 * It will copy all frags into kernel and drop the reference
802 * to userspace pages.
804 * If this function is called from an interrupt gfp_mask() must be
807 * Returns 0 on success or a negative error code on failure
808 * to allocate kernel memory to copy to.
810 int skb_copy_ubufs(struct sk_buff
*skb
, gfp_t gfp_mask
)
813 int num_frags
= skb_shinfo(skb
)->nr_frags
;
814 struct page
*page
, *head
= NULL
;
815 struct ubuf_info
*uarg
= skb_shinfo(skb
)->destructor_arg
;
817 for (i
= 0; i
< num_frags
; i
++) {
819 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
821 page
= alloc_page(gfp_mask
);
824 struct page
*next
= (struct page
*)head
->private;
830 vaddr
= kmap_atomic(skb_frag_page(f
));
831 memcpy(page_address(page
),
832 vaddr
+ f
->page_offset
, skb_frag_size(f
));
833 kunmap_atomic(vaddr
);
834 page
->private = (unsigned long)head
;
838 /* skb frags release userspace buffers */
839 for (i
= 0; i
< num_frags
; i
++)
840 skb_frag_unref(skb
, i
);
842 uarg
->callback(uarg
, false);
844 /* skb frags point to kernel buffers */
845 for (i
= num_frags
- 1; i
>= 0; i
--) {
846 __skb_fill_page_desc(skb
, i
, head
, 0,
847 skb_shinfo(skb
)->frags
[i
].size
);
848 head
= (struct page
*)head
->private;
851 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
854 EXPORT_SYMBOL_GPL(skb_copy_ubufs
);
857 * skb_clone - duplicate an sk_buff
858 * @skb: buffer to clone
859 * @gfp_mask: allocation priority
861 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
862 * copies share the same packet data but not structure. The new
863 * buffer has a reference count of 1. If the allocation fails the
864 * function returns %NULL otherwise the new buffer is returned.
866 * If this function is called from an interrupt gfp_mask() must be
870 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
874 if (skb_orphan_frags(skb
, gfp_mask
))
878 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
879 n
->fclone
== SKB_FCLONE_UNAVAILABLE
) {
880 atomic_t
*fclone_ref
= (atomic_t
*) (n
+ 1);
881 n
->fclone
= SKB_FCLONE_CLONE
;
882 atomic_inc(fclone_ref
);
884 if (skb_pfmemalloc(skb
))
885 gfp_mask
|= __GFP_MEMALLOC
;
887 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
891 kmemcheck_annotate_bitfield(n
, flags1
);
892 kmemcheck_annotate_bitfield(n
, flags2
);
893 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
896 return __skb_clone(n
, skb
);
898 EXPORT_SYMBOL(skb_clone
);
900 static void skb_headers_offset_update(struct sk_buff
*skb
, int off
)
902 /* {transport,network,mac}_header and tail are relative to skb->head */
903 skb
->transport_header
+= off
;
904 skb
->network_header
+= off
;
905 if (skb_mac_header_was_set(skb
))
906 skb
->mac_header
+= off
;
907 skb
->inner_transport_header
+= off
;
908 skb
->inner_network_header
+= off
;
909 skb
->inner_mac_header
+= off
;
912 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
914 #ifndef NET_SKBUFF_DATA_USES_OFFSET
916 * Shift between the two data areas in bytes
918 unsigned long offset
= new->data
- old
->data
;
921 __copy_skb_header(new, old
);
923 #ifndef NET_SKBUFF_DATA_USES_OFFSET
924 skb_headers_offset_update(new, offset
);
926 skb_shinfo(new)->gso_size
= skb_shinfo(old
)->gso_size
;
927 skb_shinfo(new)->gso_segs
= skb_shinfo(old
)->gso_segs
;
928 skb_shinfo(new)->gso_type
= skb_shinfo(old
)->gso_type
;
931 static inline int skb_alloc_rx_flag(const struct sk_buff
*skb
)
933 if (skb_pfmemalloc(skb
))
939 * skb_copy - create private copy of an sk_buff
940 * @skb: buffer to copy
941 * @gfp_mask: allocation priority
943 * Make a copy of both an &sk_buff and its data. This is used when the
944 * caller wishes to modify the data and needs a private copy of the
945 * data to alter. Returns %NULL on failure or the pointer to the buffer
946 * on success. The returned buffer has a reference count of 1.
948 * As by-product this function converts non-linear &sk_buff to linear
949 * one, so that &sk_buff becomes completely private and caller is allowed
950 * to modify all the data of returned buffer. This means that this
951 * function is not recommended for use in circumstances when only
952 * header is going to be modified. Use pskb_copy() instead.
955 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
957 int headerlen
= skb_headroom(skb
);
958 unsigned int size
= skb_end_offset(skb
) + skb
->data_len
;
959 struct sk_buff
*n
= __alloc_skb(size
, gfp_mask
,
960 skb_alloc_rx_flag(skb
), NUMA_NO_NODE
);
965 /* Set the data pointer */
966 skb_reserve(n
, headerlen
);
967 /* Set the tail pointer and length */
968 skb_put(n
, skb
->len
);
970 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
973 copy_skb_header(n
, skb
);
976 EXPORT_SYMBOL(skb_copy
);
979 * __pskb_copy - create copy of an sk_buff with private head.
980 * @skb: buffer to copy
981 * @headroom: headroom of new skb
982 * @gfp_mask: allocation priority
984 * Make a copy of both an &sk_buff and part of its data, located
985 * in header. Fragmented data remain shared. This is used when
986 * the caller wishes to modify only header of &sk_buff and needs
987 * private copy of the header to alter. Returns %NULL on failure
988 * or the pointer to the buffer on success.
989 * The returned buffer has a reference count of 1.
992 struct sk_buff
*__pskb_copy(struct sk_buff
*skb
, int headroom
, gfp_t gfp_mask
)
994 unsigned int size
= skb_headlen(skb
) + headroom
;
995 struct sk_buff
*n
= __alloc_skb(size
, gfp_mask
,
996 skb_alloc_rx_flag(skb
), NUMA_NO_NODE
);
1001 /* Set the data pointer */
1002 skb_reserve(n
, headroom
);
1003 /* Set the tail pointer and length */
1004 skb_put(n
, skb_headlen(skb
));
1005 /* Copy the bytes */
1006 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
1008 n
->truesize
+= skb
->data_len
;
1009 n
->data_len
= skb
->data_len
;
1012 if (skb_shinfo(skb
)->nr_frags
) {
1015 if (skb_orphan_frags(skb
, gfp_mask
)) {
1020 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1021 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
1022 skb_frag_ref(skb
, i
);
1024 skb_shinfo(n
)->nr_frags
= i
;
1027 if (skb_has_frag_list(skb
)) {
1028 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
1029 skb_clone_fraglist(n
);
1032 copy_skb_header(n
, skb
);
1036 EXPORT_SYMBOL(__pskb_copy
);
1039 * pskb_expand_head - reallocate header of &sk_buff
1040 * @skb: buffer to reallocate
1041 * @nhead: room to add at head
1042 * @ntail: room to add at tail
1043 * @gfp_mask: allocation priority
1045 * Expands (or creates identical copy, if &nhead and &ntail are zero)
1046 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
1047 * reference count of 1. Returns zero in the case of success or error,
1048 * if expansion failed. In the last case, &sk_buff is not changed.
1050 * All the pointers pointing into skb header may change and must be
1051 * reloaded after call to this function.
1054 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
1059 int size
= nhead
+ skb_end_offset(skb
) + ntail
;
1064 if (skb_shared(skb
))
1067 size
= SKB_DATA_ALIGN(size
);
1069 if (skb_pfmemalloc(skb
))
1070 gfp_mask
|= __GFP_MEMALLOC
;
1071 data
= kmalloc_reserve(size
+ SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)),
1072 gfp_mask
, NUMA_NO_NODE
, NULL
);
1075 size
= SKB_WITH_OVERHEAD(ksize(data
));
1077 /* Copy only real data... and, alas, header. This should be
1078 * optimized for the cases when header is void.
1080 memcpy(data
+ nhead
, skb
->head
, skb_tail_pointer(skb
) - skb
->head
);
1082 memcpy((struct skb_shared_info
*)(data
+ size
),
1084 offsetof(struct skb_shared_info
, frags
[skb_shinfo(skb
)->nr_frags
]));
1087 * if shinfo is shared we must drop the old head gracefully, but if it
1088 * is not we can just drop the old head and let the existing refcount
1089 * be since all we did is relocate the values
1091 if (skb_cloned(skb
)) {
1092 /* copy this zero copy skb frags */
1093 if (skb_orphan_frags(skb
, gfp_mask
))
1095 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1096 skb_frag_ref(skb
, i
);
1098 if (skb_has_frag_list(skb
))
1099 skb_clone_fraglist(skb
);
1101 skb_release_data(skb
);
1105 off
= (data
+ nhead
) - skb
->head
;
1110 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1114 skb
->end
= skb
->head
+ size
;
1117 skb_headers_offset_update(skb
, off
);
1118 /* Only adjust this if it actually is csum_start rather than csum */
1119 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1120 skb
->csum_start
+= nhead
;
1124 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
1132 EXPORT_SYMBOL(pskb_expand_head
);
1134 /* Make private copy of skb with writable head and some headroom */
1136 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
1138 struct sk_buff
*skb2
;
1139 int delta
= headroom
- skb_headroom(skb
);
1142 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
1144 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1145 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
1153 EXPORT_SYMBOL(skb_realloc_headroom
);
1156 * skb_copy_expand - copy and expand sk_buff
1157 * @skb: buffer to copy
1158 * @newheadroom: new free bytes at head
1159 * @newtailroom: new free bytes at tail
1160 * @gfp_mask: allocation priority
1162 * Make a copy of both an &sk_buff and its data and while doing so
1163 * allocate additional space.
1165 * This is used when the caller wishes to modify the data and needs a
1166 * private copy of the data to alter as well as more space for new fields.
1167 * Returns %NULL on failure or the pointer to the buffer
1168 * on success. The returned buffer has a reference count of 1.
1170 * You must pass %GFP_ATOMIC as the allocation priority if this function
1171 * is called from an interrupt.
1173 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
1174 int newheadroom
, int newtailroom
,
1178 * Allocate the copy buffer
1180 struct sk_buff
*n
= __alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
1181 gfp_mask
, skb_alloc_rx_flag(skb
),
1183 int oldheadroom
= skb_headroom(skb
);
1184 int head_copy_len
, head_copy_off
;
1190 skb_reserve(n
, newheadroom
);
1192 /* Set the tail pointer and length */
1193 skb_put(n
, skb
->len
);
1195 head_copy_len
= oldheadroom
;
1197 if (newheadroom
<= head_copy_len
)
1198 head_copy_len
= newheadroom
;
1200 head_copy_off
= newheadroom
- head_copy_len
;
1202 /* Copy the linear header and data. */
1203 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
1204 skb
->len
+ head_copy_len
))
1207 copy_skb_header(n
, skb
);
1209 off
= newheadroom
- oldheadroom
;
1210 if (n
->ip_summed
== CHECKSUM_PARTIAL
)
1211 n
->csum_start
+= off
;
1212 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1213 skb_headers_offset_update(n
, off
);
1218 EXPORT_SYMBOL(skb_copy_expand
);
1221 * skb_pad - zero pad the tail of an skb
1222 * @skb: buffer to pad
1223 * @pad: space to pad
1225 * Ensure that a buffer is followed by a padding area that is zero
1226 * filled. Used by network drivers which may DMA or transfer data
1227 * beyond the buffer end onto the wire.
1229 * May return error in out of memory cases. The skb is freed on error.
1232 int skb_pad(struct sk_buff
*skb
, int pad
)
1237 /* If the skbuff is non linear tailroom is always zero.. */
1238 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
1239 memset(skb
->data
+skb
->len
, 0, pad
);
1243 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
1244 if (likely(skb_cloned(skb
) || ntail
> 0)) {
1245 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
1250 /* FIXME: The use of this function with non-linear skb's really needs
1253 err
= skb_linearize(skb
);
1257 memset(skb
->data
+ skb
->len
, 0, pad
);
1264 EXPORT_SYMBOL(skb_pad
);
1267 * skb_put - add data to a buffer
1268 * @skb: buffer to use
1269 * @len: amount of data to add
1271 * This function extends the used data area of the buffer. If this would
1272 * exceed the total buffer size the kernel will panic. A pointer to the
1273 * first byte of the extra data is returned.
1275 unsigned char *skb_put(struct sk_buff
*skb
, unsigned int len
)
1277 unsigned char *tmp
= skb_tail_pointer(skb
);
1278 SKB_LINEAR_ASSERT(skb
);
1281 if (unlikely(skb
->tail
> skb
->end
))
1282 skb_over_panic(skb
, len
, __builtin_return_address(0));
1285 EXPORT_SYMBOL(skb_put
);
1288 * skb_push - add data to the start of a buffer
1289 * @skb: buffer to use
1290 * @len: amount of data to add
1292 * This function extends the used data area of the buffer at the buffer
1293 * start. If this would exceed the total buffer headroom the kernel will
1294 * panic. A pointer to the first byte of the extra data is returned.
1296 unsigned char *skb_push(struct sk_buff
*skb
, unsigned int len
)
1300 if (unlikely(skb
->data
<skb
->head
))
1301 skb_under_panic(skb
, len
, __builtin_return_address(0));
1304 EXPORT_SYMBOL(skb_push
);
1307 * skb_pull - remove data from the start of a buffer
1308 * @skb: buffer to use
1309 * @len: amount of data to remove
1311 * This function removes data from the start of a buffer, returning
1312 * the memory to the headroom. A pointer to the next data in the buffer
1313 * is returned. Once the data has been pulled future pushes will overwrite
1316 unsigned char *skb_pull(struct sk_buff
*skb
, unsigned int len
)
1318 return skb_pull_inline(skb
, len
);
1320 EXPORT_SYMBOL(skb_pull
);
1323 * skb_trim - remove end from a buffer
1324 * @skb: buffer to alter
1327 * Cut the length of a buffer down by removing data from the tail. If
1328 * the buffer is already under the length specified it is not modified.
1329 * The skb must be linear.
1331 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1334 __skb_trim(skb
, len
);
1336 EXPORT_SYMBOL(skb_trim
);
1338 /* Trims skb to length len. It can change skb pointers.
1341 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1343 struct sk_buff
**fragp
;
1344 struct sk_buff
*frag
;
1345 int offset
= skb_headlen(skb
);
1346 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1350 if (skb_cloned(skb
) &&
1351 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1358 for (; i
< nfrags
; i
++) {
1359 int end
= offset
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1366 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
++], len
- offset
);
1369 skb_shinfo(skb
)->nr_frags
= i
;
1371 for (; i
< nfrags
; i
++)
1372 skb_frag_unref(skb
, i
);
1374 if (skb_has_frag_list(skb
))
1375 skb_drop_fraglist(skb
);
1379 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1380 fragp
= &frag
->next
) {
1381 int end
= offset
+ frag
->len
;
1383 if (skb_shared(frag
)) {
1384 struct sk_buff
*nfrag
;
1386 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1387 if (unlikely(!nfrag
))
1390 nfrag
->next
= frag
->next
;
1402 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1406 skb_drop_list(&frag
->next
);
1411 if (len
> skb_headlen(skb
)) {
1412 skb
->data_len
-= skb
->len
- len
;
1417 skb_set_tail_pointer(skb
, len
);
1422 EXPORT_SYMBOL(___pskb_trim
);
1425 * __pskb_pull_tail - advance tail of skb header
1426 * @skb: buffer to reallocate
1427 * @delta: number of bytes to advance tail
1429 * The function makes a sense only on a fragmented &sk_buff,
1430 * it expands header moving its tail forward and copying necessary
1431 * data from fragmented part.
1433 * &sk_buff MUST have reference count of 1.
1435 * Returns %NULL (and &sk_buff does not change) if pull failed
1436 * or value of new tail of skb in the case of success.
1438 * All the pointers pointing into skb header may change and must be
1439 * reloaded after call to this function.
1442 /* Moves tail of skb head forward, copying data from fragmented part,
1443 * when it is necessary.
1444 * 1. It may fail due to malloc failure.
1445 * 2. It may change skb pointers.
1447 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1449 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1451 /* If skb has not enough free space at tail, get new one
1452 * plus 128 bytes for future expansions. If we have enough
1453 * room at tail, reallocate without expansion only if skb is cloned.
1455 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1457 if (eat
> 0 || skb_cloned(skb
)) {
1458 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1463 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
1466 /* Optimization: no fragments, no reasons to preestimate
1467 * size of pulled pages. Superb.
1469 if (!skb_has_frag_list(skb
))
1472 /* Estimate size of pulled pages. */
1474 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1475 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1482 /* If we need update frag list, we are in troubles.
1483 * Certainly, it possible to add an offset to skb data,
1484 * but taking into account that pulling is expected to
1485 * be very rare operation, it is worth to fight against
1486 * further bloating skb head and crucify ourselves here instead.
1487 * Pure masohism, indeed. 8)8)
1490 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1491 struct sk_buff
*clone
= NULL
;
1492 struct sk_buff
*insp
= NULL
;
1497 if (list
->len
<= eat
) {
1498 /* Eaten as whole. */
1503 /* Eaten partially. */
1505 if (skb_shared(list
)) {
1506 /* Sucks! We need to fork list. :-( */
1507 clone
= skb_clone(list
, GFP_ATOMIC
);
1513 /* This may be pulled without
1517 if (!pskb_pull(list
, eat
)) {
1525 /* Free pulled out fragments. */
1526 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1527 skb_shinfo(skb
)->frag_list
= list
->next
;
1530 /* And insert new clone at head. */
1533 skb_shinfo(skb
)->frag_list
= clone
;
1536 /* Success! Now we may commit changes to skb data. */
1541 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1542 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1545 skb_frag_unref(skb
, i
);
1548 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1550 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1551 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[k
], eat
);
1557 skb_shinfo(skb
)->nr_frags
= k
;
1560 skb
->data_len
-= delta
;
1562 return skb_tail_pointer(skb
);
1564 EXPORT_SYMBOL(__pskb_pull_tail
);
1567 * skb_copy_bits - copy bits from skb to kernel buffer
1569 * @offset: offset in source
1570 * @to: destination buffer
1571 * @len: number of bytes to copy
1573 * Copy the specified number of bytes from the source skb to the
1574 * destination buffer.
1577 * If its prototype is ever changed,
1578 * check arch/{*}/net/{*}.S files,
1579 * since it is called from BPF assembly code.
1581 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1583 int start
= skb_headlen(skb
);
1584 struct sk_buff
*frag_iter
;
1587 if (offset
> (int)skb
->len
- len
)
1591 if ((copy
= start
- offset
) > 0) {
1594 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
1595 if ((len
-= copy
) == 0)
1601 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1603 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
1605 WARN_ON(start
> offset
+ len
);
1607 end
= start
+ skb_frag_size(f
);
1608 if ((copy
= end
- offset
) > 0) {
1614 vaddr
= kmap_atomic(skb_frag_page(f
));
1616 vaddr
+ f
->page_offset
+ offset
- start
,
1618 kunmap_atomic(vaddr
);
1620 if ((len
-= copy
) == 0)
1628 skb_walk_frags(skb
, frag_iter
) {
1631 WARN_ON(start
> offset
+ len
);
1633 end
= start
+ frag_iter
->len
;
1634 if ((copy
= end
- offset
) > 0) {
1637 if (skb_copy_bits(frag_iter
, offset
- start
, to
, copy
))
1639 if ((len
-= copy
) == 0)
1653 EXPORT_SYMBOL(skb_copy_bits
);
1656 * Callback from splice_to_pipe(), if we need to release some pages
1657 * at the end of the spd in case we error'ed out in filling the pipe.
1659 static void sock_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
1661 put_page(spd
->pages
[i
]);
1664 static struct page
*linear_to_page(struct page
*page
, unsigned int *len
,
1665 unsigned int *offset
,
1668 struct page_frag
*pfrag
= sk_page_frag(sk
);
1670 if (!sk_page_frag_refill(sk
, pfrag
))
1673 *len
= min_t(unsigned int, *len
, pfrag
->size
- pfrag
->offset
);
1675 memcpy(page_address(pfrag
->page
) + pfrag
->offset
,
1676 page_address(page
) + *offset
, *len
);
1677 *offset
= pfrag
->offset
;
1678 pfrag
->offset
+= *len
;
1683 static bool spd_can_coalesce(const struct splice_pipe_desc
*spd
,
1685 unsigned int offset
)
1687 return spd
->nr_pages
&&
1688 spd
->pages
[spd
->nr_pages
- 1] == page
&&
1689 (spd
->partial
[spd
->nr_pages
- 1].offset
+
1690 spd
->partial
[spd
->nr_pages
- 1].len
== offset
);
1694 * Fill page/offset/length into spd, if it can hold more pages.
1696 static bool spd_fill_page(struct splice_pipe_desc
*spd
,
1697 struct pipe_inode_info
*pipe
, struct page
*page
,
1698 unsigned int *len
, unsigned int offset
,
1702 if (unlikely(spd
->nr_pages
== MAX_SKB_FRAGS
))
1706 page
= linear_to_page(page
, len
, &offset
, sk
);
1710 if (spd_can_coalesce(spd
, page
, offset
)) {
1711 spd
->partial
[spd
->nr_pages
- 1].len
+= *len
;
1715 spd
->pages
[spd
->nr_pages
] = page
;
1716 spd
->partial
[spd
->nr_pages
].len
= *len
;
1717 spd
->partial
[spd
->nr_pages
].offset
= offset
;
1723 static bool __splice_segment(struct page
*page
, unsigned int poff
,
1724 unsigned int plen
, unsigned int *off
,
1726 struct splice_pipe_desc
*spd
, bool linear
,
1728 struct pipe_inode_info
*pipe
)
1733 /* skip this segment if already processed */
1739 /* ignore any bits we already processed */
1745 unsigned int flen
= min(*len
, plen
);
1747 if (spd_fill_page(spd
, pipe
, page
, &flen
, poff
,
1753 } while (*len
&& plen
);
1759 * Map linear and fragment data from the skb to spd. It reports true if the
1760 * pipe is full or if we already spliced the requested length.
1762 static bool __skb_splice_bits(struct sk_buff
*skb
, struct pipe_inode_info
*pipe
,
1763 unsigned int *offset
, unsigned int *len
,
1764 struct splice_pipe_desc
*spd
, struct sock
*sk
)
1768 /* map the linear part :
1769 * If skb->head_frag is set, this 'linear' part is backed by a
1770 * fragment, and if the head is not shared with any clones then
1771 * we can avoid a copy since we own the head portion of this page.
1773 if (__splice_segment(virt_to_page(skb
->data
),
1774 (unsigned long) skb
->data
& (PAGE_SIZE
- 1),
1777 skb_head_is_locked(skb
),
1782 * then map the fragments
1784 for (seg
= 0; seg
< skb_shinfo(skb
)->nr_frags
; seg
++) {
1785 const skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[seg
];
1787 if (__splice_segment(skb_frag_page(f
),
1788 f
->page_offset
, skb_frag_size(f
),
1789 offset
, len
, spd
, false, sk
, pipe
))
1797 * Map data from the skb to a pipe. Should handle both the linear part,
1798 * the fragments, and the frag list. It does NOT handle frag lists within
1799 * the frag list, if such a thing exists. We'd probably need to recurse to
1800 * handle that cleanly.
1802 int skb_splice_bits(struct sk_buff
*skb
, unsigned int offset
,
1803 struct pipe_inode_info
*pipe
, unsigned int tlen
,
1806 struct partial_page partial
[MAX_SKB_FRAGS
];
1807 struct page
*pages
[MAX_SKB_FRAGS
];
1808 struct splice_pipe_desc spd
= {
1811 .nr_pages_max
= MAX_SKB_FRAGS
,
1813 .ops
= &sock_pipe_buf_ops
,
1814 .spd_release
= sock_spd_release
,
1816 struct sk_buff
*frag_iter
;
1817 struct sock
*sk
= skb
->sk
;
1821 * __skb_splice_bits() only fails if the output has no room left,
1822 * so no point in going over the frag_list for the error case.
1824 if (__skb_splice_bits(skb
, pipe
, &offset
, &tlen
, &spd
, sk
))
1830 * now see if we have a frag_list to map
1832 skb_walk_frags(skb
, frag_iter
) {
1835 if (__skb_splice_bits(frag_iter
, pipe
, &offset
, &tlen
, &spd
, sk
))
1842 * Drop the socket lock, otherwise we have reverse
1843 * locking dependencies between sk_lock and i_mutex
1844 * here as compared to sendfile(). We enter here
1845 * with the socket lock held, and splice_to_pipe() will
1846 * grab the pipe inode lock. For sendfile() emulation,
1847 * we call into ->sendpage() with the i_mutex lock held
1848 * and networking will grab the socket lock.
1851 ret
= splice_to_pipe(pipe
, &spd
);
1859 * skb_store_bits - store bits from kernel buffer to skb
1860 * @skb: destination buffer
1861 * @offset: offset in destination
1862 * @from: source buffer
1863 * @len: number of bytes to copy
1865 * Copy the specified number of bytes from the source buffer to the
1866 * destination skb. This function handles all the messy bits of
1867 * traversing fragment lists and such.
1870 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
1872 int start
= skb_headlen(skb
);
1873 struct sk_buff
*frag_iter
;
1876 if (offset
> (int)skb
->len
- len
)
1879 if ((copy
= start
- offset
) > 0) {
1882 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
1883 if ((len
-= copy
) == 0)
1889 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1890 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1893 WARN_ON(start
> offset
+ len
);
1895 end
= start
+ skb_frag_size(frag
);
1896 if ((copy
= end
- offset
) > 0) {
1902 vaddr
= kmap_atomic(skb_frag_page(frag
));
1903 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1905 kunmap_atomic(vaddr
);
1907 if ((len
-= copy
) == 0)
1915 skb_walk_frags(skb
, frag_iter
) {
1918 WARN_ON(start
> offset
+ len
);
1920 end
= start
+ frag_iter
->len
;
1921 if ((copy
= end
- offset
) > 0) {
1924 if (skb_store_bits(frag_iter
, offset
- start
,
1927 if ((len
-= copy
) == 0)
1940 EXPORT_SYMBOL(skb_store_bits
);
1942 /* Checksum skb data. */
1944 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
1945 int len
, __wsum csum
)
1947 int start
= skb_headlen(skb
);
1948 int i
, copy
= start
- offset
;
1949 struct sk_buff
*frag_iter
;
1952 /* Checksum header. */
1956 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1957 if ((len
-= copy
) == 0)
1963 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1965 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1967 WARN_ON(start
> offset
+ len
);
1969 end
= start
+ skb_frag_size(frag
);
1970 if ((copy
= end
- offset
) > 0) {
1976 vaddr
= kmap_atomic(skb_frag_page(frag
));
1977 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
1978 offset
- start
, copy
, 0);
1979 kunmap_atomic(vaddr
);
1980 csum
= csum_block_add(csum
, csum2
, pos
);
1989 skb_walk_frags(skb
, frag_iter
) {
1992 WARN_ON(start
> offset
+ len
);
1994 end
= start
+ frag_iter
->len
;
1995 if ((copy
= end
- offset
) > 0) {
1999 csum2
= skb_checksum(frag_iter
, offset
- start
,
2001 csum
= csum_block_add(csum
, csum2
, pos
);
2002 if ((len
-= copy
) == 0)
2013 EXPORT_SYMBOL(skb_checksum
);
2015 /* Both of above in one bottle. */
2017 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
2018 u8
*to
, int len
, __wsum csum
)
2020 int start
= skb_headlen(skb
);
2021 int i
, copy
= start
- offset
;
2022 struct sk_buff
*frag_iter
;
2029 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
2031 if ((len
-= copy
) == 0)
2038 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2041 WARN_ON(start
> offset
+ len
);
2043 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
2044 if ((copy
= end
- offset
) > 0) {
2047 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2051 vaddr
= kmap_atomic(skb_frag_page(frag
));
2052 csum2
= csum_partial_copy_nocheck(vaddr
+
2056 kunmap_atomic(vaddr
);
2057 csum
= csum_block_add(csum
, csum2
, pos
);
2067 skb_walk_frags(skb
, frag_iter
) {
2071 WARN_ON(start
> offset
+ len
);
2073 end
= start
+ frag_iter
->len
;
2074 if ((copy
= end
- offset
) > 0) {
2077 csum2
= skb_copy_and_csum_bits(frag_iter
,
2080 csum
= csum_block_add(csum
, csum2
, pos
);
2081 if ((len
-= copy
) == 0)
2092 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
2094 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
2099 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
2100 csstart
= skb_checksum_start_offset(skb
);
2102 csstart
= skb_headlen(skb
);
2104 BUG_ON(csstart
> skb_headlen(skb
));
2106 skb_copy_from_linear_data(skb
, to
, csstart
);
2109 if (csstart
!= skb
->len
)
2110 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
2111 skb
->len
- csstart
, 0);
2113 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2114 long csstuff
= csstart
+ skb
->csum_offset
;
2116 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
2119 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
2122 * skb_dequeue - remove from the head of the queue
2123 * @list: list to dequeue from
2125 * Remove the head of the list. The list lock is taken so the function
2126 * may be used safely with other locking list functions. The head item is
2127 * returned or %NULL if the list is empty.
2130 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
2132 unsigned long flags
;
2133 struct sk_buff
*result
;
2135 spin_lock_irqsave(&list
->lock
, flags
);
2136 result
= __skb_dequeue(list
);
2137 spin_unlock_irqrestore(&list
->lock
, flags
);
2140 EXPORT_SYMBOL(skb_dequeue
);
2143 * skb_dequeue_tail - remove from the tail of the queue
2144 * @list: list to dequeue from
2146 * Remove the tail of the list. The list lock is taken so the function
2147 * may be used safely with other locking list functions. The tail item is
2148 * returned or %NULL if the list is empty.
2150 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
2152 unsigned long flags
;
2153 struct sk_buff
*result
;
2155 spin_lock_irqsave(&list
->lock
, flags
);
2156 result
= __skb_dequeue_tail(list
);
2157 spin_unlock_irqrestore(&list
->lock
, flags
);
2160 EXPORT_SYMBOL(skb_dequeue_tail
);
2163 * skb_queue_purge - empty a list
2164 * @list: list to empty
2166 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2167 * the list and one reference dropped. This function takes the list
2168 * lock and is atomic with respect to other list locking functions.
2170 void skb_queue_purge(struct sk_buff_head
*list
)
2172 struct sk_buff
*skb
;
2173 while ((skb
= skb_dequeue(list
)) != NULL
)
2176 EXPORT_SYMBOL(skb_queue_purge
);
2179 * skb_queue_head - queue a buffer at the list head
2180 * @list: list to use
2181 * @newsk: buffer to queue
2183 * Queue a buffer at the start of the list. This function takes the
2184 * list lock and can be used safely with other locking &sk_buff functions
2187 * A buffer cannot be placed on two lists at the same time.
2189 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2191 unsigned long flags
;
2193 spin_lock_irqsave(&list
->lock
, flags
);
2194 __skb_queue_head(list
, newsk
);
2195 spin_unlock_irqrestore(&list
->lock
, flags
);
2197 EXPORT_SYMBOL(skb_queue_head
);
2200 * skb_queue_tail - queue a buffer at the list tail
2201 * @list: list to use
2202 * @newsk: buffer to queue
2204 * Queue a buffer at the tail of the list. This function takes the
2205 * list lock and can be used safely with other locking &sk_buff functions
2208 * A buffer cannot be placed on two lists at the same time.
2210 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2212 unsigned long flags
;
2214 spin_lock_irqsave(&list
->lock
, flags
);
2215 __skb_queue_tail(list
, newsk
);
2216 spin_unlock_irqrestore(&list
->lock
, flags
);
2218 EXPORT_SYMBOL(skb_queue_tail
);
2221 * skb_unlink - remove a buffer from a list
2222 * @skb: buffer to remove
2223 * @list: list to use
2225 * Remove a packet from a list. The list locks are taken and this
2226 * function is atomic with respect to other list locked calls
2228 * You must know what list the SKB is on.
2230 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
2232 unsigned long flags
;
2234 spin_lock_irqsave(&list
->lock
, flags
);
2235 __skb_unlink(skb
, list
);
2236 spin_unlock_irqrestore(&list
->lock
, flags
);
2238 EXPORT_SYMBOL(skb_unlink
);
2241 * skb_append - append a buffer
2242 * @old: buffer to insert after
2243 * @newsk: buffer to insert
2244 * @list: list to use
2246 * Place a packet after a given packet in a list. The list locks are taken
2247 * and this function is atomic with respect to other list locked calls.
2248 * A buffer cannot be placed on two lists at the same time.
2250 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2252 unsigned long flags
;
2254 spin_lock_irqsave(&list
->lock
, flags
);
2255 __skb_queue_after(list
, old
, newsk
);
2256 spin_unlock_irqrestore(&list
->lock
, flags
);
2258 EXPORT_SYMBOL(skb_append
);
2261 * skb_insert - insert a buffer
2262 * @old: buffer to insert before
2263 * @newsk: buffer to insert
2264 * @list: list to use
2266 * Place a packet before a given packet in a list. The list locks are
2267 * taken and this function is atomic with respect to other list locked
2270 * A buffer cannot be placed on two lists at the same time.
2272 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2274 unsigned long flags
;
2276 spin_lock_irqsave(&list
->lock
, flags
);
2277 __skb_insert(newsk
, old
->prev
, old
, list
);
2278 spin_unlock_irqrestore(&list
->lock
, flags
);
2280 EXPORT_SYMBOL(skb_insert
);
2282 static inline void skb_split_inside_header(struct sk_buff
*skb
,
2283 struct sk_buff
* skb1
,
2284 const u32 len
, const int pos
)
2288 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
2290 /* And move data appendix as is. */
2291 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
2292 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
2294 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
2295 skb_shinfo(skb
)->nr_frags
= 0;
2296 skb1
->data_len
= skb
->data_len
;
2297 skb1
->len
+= skb1
->data_len
;
2300 skb_set_tail_pointer(skb
, len
);
2303 static inline void skb_split_no_header(struct sk_buff
*skb
,
2304 struct sk_buff
* skb1
,
2305 const u32 len
, int pos
)
2308 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
2310 skb_shinfo(skb
)->nr_frags
= 0;
2311 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
2313 skb
->data_len
= len
- pos
;
2315 for (i
= 0; i
< nfrags
; i
++) {
2316 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
2318 if (pos
+ size
> len
) {
2319 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
2323 * We have two variants in this case:
2324 * 1. Move all the frag to the second
2325 * part, if it is possible. F.e.
2326 * this approach is mandatory for TUX,
2327 * where splitting is expensive.
2328 * 2. Split is accurately. We make this.
2330 skb_frag_ref(skb
, i
);
2331 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
2332 skb_frag_size_sub(&skb_shinfo(skb1
)->frags
[0], len
- pos
);
2333 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
], len
- pos
);
2334 skb_shinfo(skb
)->nr_frags
++;
2338 skb_shinfo(skb
)->nr_frags
++;
2341 skb_shinfo(skb1
)->nr_frags
= k
;
2345 * skb_split - Split fragmented skb to two parts at length len.
2346 * @skb: the buffer to split
2347 * @skb1: the buffer to receive the second part
2348 * @len: new length for skb
2350 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
2352 int pos
= skb_headlen(skb
);
2354 skb_shinfo(skb1
)->tx_flags
= skb_shinfo(skb
)->tx_flags
& SKBTX_SHARED_FRAG
;
2355 if (len
< pos
) /* Split line is inside header. */
2356 skb_split_inside_header(skb
, skb1
, len
, pos
);
2357 else /* Second chunk has no header, nothing to copy. */
2358 skb_split_no_header(skb
, skb1
, len
, pos
);
2360 EXPORT_SYMBOL(skb_split
);
2362 /* Shifting from/to a cloned skb is a no-go.
2364 * Caller cannot keep skb_shinfo related pointers past calling here!
2366 static int skb_prepare_for_shift(struct sk_buff
*skb
)
2368 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2372 * skb_shift - Shifts paged data partially from skb to another
2373 * @tgt: buffer into which tail data gets added
2374 * @skb: buffer from which the paged data comes from
2375 * @shiftlen: shift up to this many bytes
2377 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2378 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2379 * It's up to caller to free skb if everything was shifted.
2381 * If @tgt runs out of frags, the whole operation is aborted.
2383 * Skb cannot include anything else but paged data while tgt is allowed
2384 * to have non-paged data as well.
2386 * TODO: full sized shift could be optimized but that would need
2387 * specialized skb free'er to handle frags without up-to-date nr_frags.
2389 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
2391 int from
, to
, merge
, todo
;
2392 struct skb_frag_struct
*fragfrom
, *fragto
;
2394 BUG_ON(shiftlen
> skb
->len
);
2395 BUG_ON(skb_headlen(skb
)); /* Would corrupt stream */
2399 to
= skb_shinfo(tgt
)->nr_frags
;
2400 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2402 /* Actual merge is delayed until the point when we know we can
2403 * commit all, so that we don't have to undo partial changes
2406 !skb_can_coalesce(tgt
, to
, skb_frag_page(fragfrom
),
2407 fragfrom
->page_offset
)) {
2412 todo
-= skb_frag_size(fragfrom
);
2414 if (skb_prepare_for_shift(skb
) ||
2415 skb_prepare_for_shift(tgt
))
2418 /* All previous frag pointers might be stale! */
2419 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2420 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2422 skb_frag_size_add(fragto
, shiftlen
);
2423 skb_frag_size_sub(fragfrom
, shiftlen
);
2424 fragfrom
->page_offset
+= shiftlen
;
2432 /* Skip full, not-fitting skb to avoid expensive operations */
2433 if ((shiftlen
== skb
->len
) &&
2434 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
2437 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
2440 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
2441 if (to
== MAX_SKB_FRAGS
)
2444 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2445 fragto
= &skb_shinfo(tgt
)->frags
[to
];
2447 if (todo
>= skb_frag_size(fragfrom
)) {
2448 *fragto
= *fragfrom
;
2449 todo
-= skb_frag_size(fragfrom
);
2454 __skb_frag_ref(fragfrom
);
2455 fragto
->page
= fragfrom
->page
;
2456 fragto
->page_offset
= fragfrom
->page_offset
;
2457 skb_frag_size_set(fragto
, todo
);
2459 fragfrom
->page_offset
+= todo
;
2460 skb_frag_size_sub(fragfrom
, todo
);
2468 /* Ready to "commit" this state change to tgt */
2469 skb_shinfo(tgt
)->nr_frags
= to
;
2472 fragfrom
= &skb_shinfo(skb
)->frags
[0];
2473 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2475 skb_frag_size_add(fragto
, skb_frag_size(fragfrom
));
2476 __skb_frag_unref(fragfrom
);
2479 /* Reposition in the original skb */
2481 while (from
< skb_shinfo(skb
)->nr_frags
)
2482 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
2483 skb_shinfo(skb
)->nr_frags
= to
;
2485 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
2488 /* Most likely the tgt won't ever need its checksum anymore, skb on
2489 * the other hand might need it if it needs to be resent
2491 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
2492 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2494 /* Yak, is it really working this way? Some helper please? */
2495 skb
->len
-= shiftlen
;
2496 skb
->data_len
-= shiftlen
;
2497 skb
->truesize
-= shiftlen
;
2498 tgt
->len
+= shiftlen
;
2499 tgt
->data_len
+= shiftlen
;
2500 tgt
->truesize
+= shiftlen
;
2506 * skb_prepare_seq_read - Prepare a sequential read of skb data
2507 * @skb: the buffer to read
2508 * @from: lower offset of data to be read
2509 * @to: upper offset of data to be read
2510 * @st: state variable
2512 * Initializes the specified state variable. Must be called before
2513 * invoking skb_seq_read() for the first time.
2515 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
2516 unsigned int to
, struct skb_seq_state
*st
)
2518 st
->lower_offset
= from
;
2519 st
->upper_offset
= to
;
2520 st
->root_skb
= st
->cur_skb
= skb
;
2521 st
->frag_idx
= st
->stepped_offset
= 0;
2522 st
->frag_data
= NULL
;
2524 EXPORT_SYMBOL(skb_prepare_seq_read
);
2527 * skb_seq_read - Sequentially read skb data
2528 * @consumed: number of bytes consumed by the caller so far
2529 * @data: destination pointer for data to be returned
2530 * @st: state variable
2532 * Reads a block of skb data at &consumed relative to the
2533 * lower offset specified to skb_prepare_seq_read(). Assigns
2534 * the head of the data block to &data and returns the length
2535 * of the block or 0 if the end of the skb data or the upper
2536 * offset has been reached.
2538 * The caller is not required to consume all of the data
2539 * returned, i.e. &consumed is typically set to the number
2540 * of bytes already consumed and the next call to
2541 * skb_seq_read() will return the remaining part of the block.
2543 * Note 1: The size of each block of data returned can be arbitrary,
2544 * this limitation is the cost for zerocopy seqeuental
2545 * reads of potentially non linear data.
2547 * Note 2: Fragment lists within fragments are not implemented
2548 * at the moment, state->root_skb could be replaced with
2549 * a stack for this purpose.
2551 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
2552 struct skb_seq_state
*st
)
2554 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
2557 if (unlikely(abs_offset
>= st
->upper_offset
))
2561 block_limit
= skb_headlen(st
->cur_skb
) + st
->stepped_offset
;
2563 if (abs_offset
< block_limit
&& !st
->frag_data
) {
2564 *data
= st
->cur_skb
->data
+ (abs_offset
- st
->stepped_offset
);
2565 return block_limit
- abs_offset
;
2568 if (st
->frag_idx
== 0 && !st
->frag_data
)
2569 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
2571 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
2572 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
2573 block_limit
= skb_frag_size(frag
) + st
->stepped_offset
;
2575 if (abs_offset
< block_limit
) {
2577 st
->frag_data
= kmap_atomic(skb_frag_page(frag
));
2579 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
2580 (abs_offset
- st
->stepped_offset
);
2582 return block_limit
- abs_offset
;
2585 if (st
->frag_data
) {
2586 kunmap_atomic(st
->frag_data
);
2587 st
->frag_data
= NULL
;
2591 st
->stepped_offset
+= skb_frag_size(frag
);
2594 if (st
->frag_data
) {
2595 kunmap_atomic(st
->frag_data
);
2596 st
->frag_data
= NULL
;
2599 if (st
->root_skb
== st
->cur_skb
&& skb_has_frag_list(st
->root_skb
)) {
2600 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
2603 } else if (st
->cur_skb
->next
) {
2604 st
->cur_skb
= st
->cur_skb
->next
;
2611 EXPORT_SYMBOL(skb_seq_read
);
2614 * skb_abort_seq_read - Abort a sequential read of skb data
2615 * @st: state variable
2617 * Must be called if skb_seq_read() was not called until it
2620 void skb_abort_seq_read(struct skb_seq_state
*st
)
2623 kunmap_atomic(st
->frag_data
);
2625 EXPORT_SYMBOL(skb_abort_seq_read
);
2627 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2629 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
2630 struct ts_config
*conf
,
2631 struct ts_state
*state
)
2633 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
2636 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
2638 skb_abort_seq_read(TS_SKB_CB(state
));
2642 * skb_find_text - Find a text pattern in skb data
2643 * @skb: the buffer to look in
2644 * @from: search offset
2646 * @config: textsearch configuration
2647 * @state: uninitialized textsearch state variable
2649 * Finds a pattern in the skb data according to the specified
2650 * textsearch configuration. Use textsearch_next() to retrieve
2651 * subsequent occurrences of the pattern. Returns the offset
2652 * to the first occurrence or UINT_MAX if no match was found.
2654 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
2655 unsigned int to
, struct ts_config
*config
,
2656 struct ts_state
*state
)
2660 config
->get_next_block
= skb_ts_get_next_block
;
2661 config
->finish
= skb_ts_finish
;
2663 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
2665 ret
= textsearch_find(config
, state
);
2666 return (ret
<= to
- from
? ret
: UINT_MAX
);
2668 EXPORT_SYMBOL(skb_find_text
);
2671 * skb_append_datato_frags - append the user data to a skb
2672 * @sk: sock structure
2673 * @skb: skb structure to be appened with user data.
2674 * @getfrag: call back function to be used for getting the user data
2675 * @from: pointer to user message iov
2676 * @length: length of the iov message
2678 * Description: This procedure append the user data in the fragment part
2679 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2681 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
2682 int (*getfrag
)(void *from
, char *to
, int offset
,
2683 int len
, int odd
, struct sk_buff
*skb
),
2684 void *from
, int length
)
2686 int frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2690 struct page_frag
*pfrag
= ¤t
->task_frag
;
2693 /* Return error if we don't have space for new frag */
2694 if (frg_cnt
>= MAX_SKB_FRAGS
)
2697 if (!sk_page_frag_refill(sk
, pfrag
))
2700 /* copy the user data to page */
2701 copy
= min_t(int, length
, pfrag
->size
- pfrag
->offset
);
2703 ret
= getfrag(from
, page_address(pfrag
->page
) + pfrag
->offset
,
2704 offset
, copy
, 0, skb
);
2708 /* copy was successful so update the size parameters */
2709 skb_fill_page_desc(skb
, frg_cnt
, pfrag
->page
, pfrag
->offset
,
2712 pfrag
->offset
+= copy
;
2713 get_page(pfrag
->page
);
2715 skb
->truesize
+= copy
;
2716 atomic_add(copy
, &sk
->sk_wmem_alloc
);
2718 skb
->data_len
+= copy
;
2722 } while (length
> 0);
2726 EXPORT_SYMBOL(skb_append_datato_frags
);
2729 * skb_pull_rcsum - pull skb and update receive checksum
2730 * @skb: buffer to update
2731 * @len: length of data pulled
2733 * This function performs an skb_pull on the packet and updates
2734 * the CHECKSUM_COMPLETE checksum. It should be used on
2735 * receive path processing instead of skb_pull unless you know
2736 * that the checksum difference is zero (e.g., a valid IP header)
2737 * or you are setting ip_summed to CHECKSUM_NONE.
2739 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
2741 BUG_ON(len
> skb
->len
);
2743 BUG_ON(skb
->len
< skb
->data_len
);
2744 skb_postpull_rcsum(skb
, skb
->data
, len
);
2745 return skb
->data
+= len
;
2747 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
2750 * skb_segment - Perform protocol segmentation on skb.
2751 * @skb: buffer to segment
2752 * @features: features for the output path (see dev->features)
2754 * This function performs segmentation on the given skb. It returns
2755 * a pointer to the first in a list of new skbs for the segments.
2756 * In case of error it returns ERR_PTR(err).
2758 struct sk_buff
*skb_segment(struct sk_buff
*skb
, netdev_features_t features
)
2760 struct sk_buff
*segs
= NULL
;
2761 struct sk_buff
*tail
= NULL
;
2762 struct sk_buff
*fskb
= skb_shinfo(skb
)->frag_list
;
2763 unsigned int mss
= skb_shinfo(skb
)->gso_size
;
2764 unsigned int doffset
= skb
->data
- skb_mac_header(skb
);
2765 unsigned int offset
= doffset
;
2766 unsigned int tnl_hlen
= skb_tnl_header_len(skb
);
2767 unsigned int headroom
;
2771 int sg
= !!(features
& NETIF_F_SG
);
2772 int nfrags
= skb_shinfo(skb
)->nr_frags
;
2777 proto
= skb_network_protocol(skb
);
2778 if (unlikely(!proto
))
2779 return ERR_PTR(-EINVAL
);
2781 csum
= !!can_checksum_protocol(features
, proto
);
2782 __skb_push(skb
, doffset
);
2783 headroom
= skb_headroom(skb
);
2784 pos
= skb_headlen(skb
);
2787 struct sk_buff
*nskb
;
2792 len
= skb
->len
- offset
;
2796 hsize
= skb_headlen(skb
) - offset
;
2799 if (hsize
> len
|| !sg
)
2802 if (!hsize
&& i
>= nfrags
) {
2803 BUG_ON(fskb
->len
!= len
);
2806 nskb
= skb_clone(fskb
, GFP_ATOMIC
);
2809 if (unlikely(!nskb
))
2812 hsize
= skb_end_offset(nskb
);
2813 if (skb_cow_head(nskb
, doffset
+ headroom
)) {
2818 nskb
->truesize
+= skb_end_offset(nskb
) - hsize
;
2819 skb_release_head_state(nskb
);
2820 __skb_push(nskb
, doffset
);
2822 nskb
= __alloc_skb(hsize
+ doffset
+ headroom
,
2823 GFP_ATOMIC
, skb_alloc_rx_flag(skb
),
2826 if (unlikely(!nskb
))
2829 skb_reserve(nskb
, headroom
);
2830 __skb_put(nskb
, doffset
);
2839 __copy_skb_header(nskb
, skb
);
2840 nskb
->mac_len
= skb
->mac_len
;
2842 /* nskb and skb might have different headroom */
2843 if (nskb
->ip_summed
== CHECKSUM_PARTIAL
)
2844 nskb
->csum_start
+= skb_headroom(nskb
) - headroom
;
2846 skb_reset_mac_header(nskb
);
2847 skb_set_network_header(nskb
, skb
->mac_len
);
2848 nskb
->transport_header
= (nskb
->network_header
+
2849 skb_network_header_len(skb
));
2851 skb_copy_from_linear_data_offset(skb
, -tnl_hlen
,
2852 nskb
->data
- tnl_hlen
,
2853 doffset
+ tnl_hlen
);
2855 if (fskb
!= skb_shinfo(skb
)->frag_list
)
2856 goto perform_csum_check
;
2859 nskb
->ip_summed
= CHECKSUM_NONE
;
2860 nskb
->csum
= skb_copy_and_csum_bits(skb
, offset
,
2866 frag
= skb_shinfo(nskb
)->frags
;
2868 skb_copy_from_linear_data_offset(skb
, offset
,
2869 skb_put(nskb
, hsize
), hsize
);
2871 skb_shinfo(nskb
)->tx_flags
= skb_shinfo(skb
)->tx_flags
& SKBTX_SHARED_FRAG
;
2873 while (pos
< offset
+ len
&& i
< nfrags
) {
2874 *frag
= skb_shinfo(skb
)->frags
[i
];
2875 __skb_frag_ref(frag
);
2876 size
= skb_frag_size(frag
);
2879 frag
->page_offset
+= offset
- pos
;
2880 skb_frag_size_sub(frag
, offset
- pos
);
2883 skb_shinfo(nskb
)->nr_frags
++;
2885 if (pos
+ size
<= offset
+ len
) {
2889 skb_frag_size_sub(frag
, pos
+ size
- (offset
+ len
));
2896 if (pos
< offset
+ len
) {
2897 struct sk_buff
*fskb2
= fskb
;
2899 BUG_ON(pos
+ fskb
->len
!= offset
+ len
);
2905 fskb2
= skb_clone(fskb2
, GFP_ATOMIC
);
2911 SKB_FRAG_ASSERT(nskb
);
2912 skb_shinfo(nskb
)->frag_list
= fskb2
;
2916 nskb
->data_len
= len
- hsize
;
2917 nskb
->len
+= nskb
->data_len
;
2918 nskb
->truesize
+= nskb
->data_len
;
2922 nskb
->csum
= skb_checksum(nskb
, doffset
,
2923 nskb
->len
- doffset
, 0);
2924 nskb
->ip_summed
= CHECKSUM_NONE
;
2926 } while ((offset
+= len
) < skb
->len
);
2931 while ((skb
= segs
)) {
2935 return ERR_PTR(err
);
2937 EXPORT_SYMBOL_GPL(skb_segment
);
2939 int skb_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2941 struct sk_buff
*p
= *head
;
2942 struct sk_buff
*nskb
;
2943 struct skb_shared_info
*skbinfo
= skb_shinfo(skb
);
2944 struct skb_shared_info
*pinfo
= skb_shinfo(p
);
2945 unsigned int headroom
;
2946 unsigned int len
= skb_gro_len(skb
);
2947 unsigned int offset
= skb_gro_offset(skb
);
2948 unsigned int headlen
= skb_headlen(skb
);
2949 unsigned int delta_truesize
;
2951 if (p
->len
+ len
>= 65536)
2954 if (pinfo
->frag_list
)
2956 else if (headlen
<= offset
) {
2959 int i
= skbinfo
->nr_frags
;
2960 int nr_frags
= pinfo
->nr_frags
+ i
;
2964 if (nr_frags
> MAX_SKB_FRAGS
)
2967 pinfo
->nr_frags
= nr_frags
;
2968 skbinfo
->nr_frags
= 0;
2970 frag
= pinfo
->frags
+ nr_frags
;
2971 frag2
= skbinfo
->frags
+ i
;
2976 frag
->page_offset
+= offset
;
2977 skb_frag_size_sub(frag
, offset
);
2979 /* all fragments truesize : remove (head size + sk_buff) */
2980 delta_truesize
= skb
->truesize
-
2981 SKB_TRUESIZE(skb_end_offset(skb
));
2983 skb
->truesize
-= skb
->data_len
;
2984 skb
->len
-= skb
->data_len
;
2987 NAPI_GRO_CB(skb
)->free
= NAPI_GRO_FREE
;
2989 } else if (skb
->head_frag
) {
2990 int nr_frags
= pinfo
->nr_frags
;
2991 skb_frag_t
*frag
= pinfo
->frags
+ nr_frags
;
2992 struct page
*page
= virt_to_head_page(skb
->head
);
2993 unsigned int first_size
= headlen
- offset
;
2994 unsigned int first_offset
;
2996 if (nr_frags
+ 1 + skbinfo
->nr_frags
> MAX_SKB_FRAGS
)
2999 first_offset
= skb
->data
-
3000 (unsigned char *)page_address(page
) +
3003 pinfo
->nr_frags
= nr_frags
+ 1 + skbinfo
->nr_frags
;
3005 frag
->page
.p
= page
;
3006 frag
->page_offset
= first_offset
;
3007 skb_frag_size_set(frag
, first_size
);
3009 memcpy(frag
+ 1, skbinfo
->frags
, sizeof(*frag
) * skbinfo
->nr_frags
);
3010 /* We dont need to clear skbinfo->nr_frags here */
3012 delta_truesize
= skb
->truesize
- SKB_DATA_ALIGN(sizeof(struct sk_buff
));
3013 NAPI_GRO_CB(skb
)->free
= NAPI_GRO_FREE_STOLEN_HEAD
;
3015 } else if (skb_gro_len(p
) != pinfo
->gso_size
)
3018 headroom
= skb_headroom(p
);
3019 nskb
= alloc_skb(headroom
+ skb_gro_offset(p
), GFP_ATOMIC
);
3020 if (unlikely(!nskb
))
3023 __copy_skb_header(nskb
, p
);
3024 nskb
->mac_len
= p
->mac_len
;
3026 skb_reserve(nskb
, headroom
);
3027 __skb_put(nskb
, skb_gro_offset(p
));
3029 skb_set_mac_header(nskb
, skb_mac_header(p
) - p
->data
);
3030 skb_set_network_header(nskb
, skb_network_offset(p
));
3031 skb_set_transport_header(nskb
, skb_transport_offset(p
));
3033 __skb_pull(p
, skb_gro_offset(p
));
3034 memcpy(skb_mac_header(nskb
), skb_mac_header(p
),
3035 p
->data
- skb_mac_header(p
));
3037 skb_shinfo(nskb
)->frag_list
= p
;
3038 skb_shinfo(nskb
)->gso_size
= pinfo
->gso_size
;
3039 pinfo
->gso_size
= 0;
3040 skb_header_release(p
);
3041 NAPI_GRO_CB(nskb
)->last
= p
;
3043 nskb
->data_len
+= p
->len
;
3044 nskb
->truesize
+= p
->truesize
;
3045 nskb
->len
+= p
->len
;
3048 nskb
->next
= p
->next
;
3054 delta_truesize
= skb
->truesize
;
3055 if (offset
> headlen
) {
3056 unsigned int eat
= offset
- headlen
;
3058 skbinfo
->frags
[0].page_offset
+= eat
;
3059 skb_frag_size_sub(&skbinfo
->frags
[0], eat
);
3060 skb
->data_len
-= eat
;
3065 __skb_pull(skb
, offset
);
3067 NAPI_GRO_CB(p
)->last
->next
= skb
;
3068 NAPI_GRO_CB(p
)->last
= skb
;
3069 skb_header_release(skb
);
3072 NAPI_GRO_CB(p
)->count
++;
3074 p
->truesize
+= delta_truesize
;
3077 NAPI_GRO_CB(skb
)->same_flow
= 1;
3080 EXPORT_SYMBOL_GPL(skb_gro_receive
);
3082 void __init
skb_init(void)
3084 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
3085 sizeof(struct sk_buff
),
3087 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
3089 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
3090 (2*sizeof(struct sk_buff
)) +
3093 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
3098 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3099 * @skb: Socket buffer containing the buffers to be mapped
3100 * @sg: The scatter-gather list to map into
3101 * @offset: The offset into the buffer's contents to start mapping
3102 * @len: Length of buffer space to be mapped
3104 * Fill the specified scatter-gather list with mappings/pointers into a
3105 * region of the buffer space attached to a socket buffer.
3108 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
3110 int start
= skb_headlen(skb
);
3111 int i
, copy
= start
- offset
;
3112 struct sk_buff
*frag_iter
;
3118 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
3120 if ((len
-= copy
) == 0)
3125 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
3128 WARN_ON(start
> offset
+ len
);
3130 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
3131 if ((copy
= end
- offset
) > 0) {
3132 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
3136 sg_set_page(&sg
[elt
], skb_frag_page(frag
), copy
,
3137 frag
->page_offset
+offset
-start
);
3146 skb_walk_frags(skb
, frag_iter
) {
3149 WARN_ON(start
> offset
+ len
);
3151 end
= start
+ frag_iter
->len
;
3152 if ((copy
= end
- offset
) > 0) {
3155 elt
+= __skb_to_sgvec(frag_iter
, sg
+elt
, offset
- start
,
3157 if ((len
-= copy
) == 0)
3167 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
3169 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
);
3171 sg_mark_end(&sg
[nsg
- 1]);
3175 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
3178 * skb_cow_data - Check that a socket buffer's data buffers are writable
3179 * @skb: The socket buffer to check.
3180 * @tailbits: Amount of trailing space to be added
3181 * @trailer: Returned pointer to the skb where the @tailbits space begins
3183 * Make sure that the data buffers attached to a socket buffer are
3184 * writable. If they are not, private copies are made of the data buffers
3185 * and the socket buffer is set to use these instead.
3187 * If @tailbits is given, make sure that there is space to write @tailbits
3188 * bytes of data beyond current end of socket buffer. @trailer will be
3189 * set to point to the skb in which this space begins.
3191 * The number of scatterlist elements required to completely map the
3192 * COW'd and extended socket buffer will be returned.
3194 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
3198 struct sk_buff
*skb1
, **skb_p
;
3200 /* If skb is cloned or its head is paged, reallocate
3201 * head pulling out all the pages (pages are considered not writable
3202 * at the moment even if they are anonymous).
3204 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
3205 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
3208 /* Easy case. Most of packets will go this way. */
3209 if (!skb_has_frag_list(skb
)) {
3210 /* A little of trouble, not enough of space for trailer.
3211 * This should not happen, when stack is tuned to generate
3212 * good frames. OK, on miss we reallocate and reserve even more
3213 * space, 128 bytes is fair. */
3215 if (skb_tailroom(skb
) < tailbits
&&
3216 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
3224 /* Misery. We are in troubles, going to mincer fragments... */
3227 skb_p
= &skb_shinfo(skb
)->frag_list
;
3230 while ((skb1
= *skb_p
) != NULL
) {
3233 /* The fragment is partially pulled by someone,
3234 * this can happen on input. Copy it and everything
3237 if (skb_shared(skb1
))
3240 /* If the skb is the last, worry about trailer. */
3242 if (skb1
->next
== NULL
&& tailbits
) {
3243 if (skb_shinfo(skb1
)->nr_frags
||
3244 skb_has_frag_list(skb1
) ||
3245 skb_tailroom(skb1
) < tailbits
)
3246 ntail
= tailbits
+ 128;
3252 skb_shinfo(skb1
)->nr_frags
||
3253 skb_has_frag_list(skb1
)) {
3254 struct sk_buff
*skb2
;
3256 /* Fuck, we are miserable poor guys... */
3258 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
3260 skb2
= skb_copy_expand(skb1
,
3264 if (unlikely(skb2
== NULL
))
3268 skb_set_owner_w(skb2
, skb1
->sk
);
3270 /* Looking around. Are we still alive?
3271 * OK, link new skb, drop old one */
3273 skb2
->next
= skb1
->next
;
3280 skb_p
= &skb1
->next
;
3285 EXPORT_SYMBOL_GPL(skb_cow_data
);
3287 static void sock_rmem_free(struct sk_buff
*skb
)
3289 struct sock
*sk
= skb
->sk
;
3291 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
3295 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3297 int sock_queue_err_skb(struct sock
*sk
, struct sk_buff
*skb
)
3301 if (atomic_read(&sk
->sk_rmem_alloc
) + skb
->truesize
>=
3302 (unsigned int)sk
->sk_rcvbuf
)
3307 skb
->destructor
= sock_rmem_free
;
3308 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
3310 /* before exiting rcu section, make sure dst is refcounted */
3313 skb_queue_tail(&sk
->sk_error_queue
, skb
);
3314 if (!sock_flag(sk
, SOCK_DEAD
))
3315 sk
->sk_data_ready(sk
, len
);
3318 EXPORT_SYMBOL(sock_queue_err_skb
);
3320 void skb_tstamp_tx(struct sk_buff
*orig_skb
,
3321 struct skb_shared_hwtstamps
*hwtstamps
)
3323 struct sock
*sk
= orig_skb
->sk
;
3324 struct sock_exterr_skb
*serr
;
3325 struct sk_buff
*skb
;
3332 *skb_hwtstamps(orig_skb
) =
3336 * no hardware time stamps available,
3337 * so keep the shared tx_flags and only
3338 * store software time stamp
3340 orig_skb
->tstamp
= ktime_get_real();
3343 skb
= skb_clone(orig_skb
, GFP_ATOMIC
);
3347 serr
= SKB_EXT_ERR(skb
);
3348 memset(serr
, 0, sizeof(*serr
));
3349 serr
->ee
.ee_errno
= ENOMSG
;
3350 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TIMESTAMPING
;
3352 err
= sock_queue_err_skb(sk
, skb
);
3357 EXPORT_SYMBOL_GPL(skb_tstamp_tx
);
3359 void skb_complete_wifi_ack(struct sk_buff
*skb
, bool acked
)
3361 struct sock
*sk
= skb
->sk
;
3362 struct sock_exterr_skb
*serr
;
3365 skb
->wifi_acked_valid
= 1;
3366 skb
->wifi_acked
= acked
;
3368 serr
= SKB_EXT_ERR(skb
);
3369 memset(serr
, 0, sizeof(*serr
));
3370 serr
->ee
.ee_errno
= ENOMSG
;
3371 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TXSTATUS
;
3373 err
= sock_queue_err_skb(sk
, skb
);
3377 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack
);
3381 * skb_partial_csum_set - set up and verify partial csum values for packet
3382 * @skb: the skb to set
3383 * @start: the number of bytes after skb->data to start checksumming.
3384 * @off: the offset from start to place the checksum.
3386 * For untrusted partially-checksummed packets, we need to make sure the values
3387 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3389 * This function checks and sets those values and skb->ip_summed: if this
3390 * returns false you should drop the packet.
3392 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
3394 if (unlikely(start
> skb_headlen(skb
)) ||
3395 unlikely((int)start
+ off
> skb_headlen(skb
) - 2)) {
3396 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3397 start
, off
, skb_headlen(skb
));
3400 skb
->ip_summed
= CHECKSUM_PARTIAL
;
3401 skb
->csum_start
= skb_headroom(skb
) + start
;
3402 skb
->csum_offset
= off
;
3403 skb_set_transport_header(skb
, start
);
3406 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);
3408 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
3410 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
3413 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);
3415 void kfree_skb_partial(struct sk_buff
*skb
, bool head_stolen
)
3418 skb_release_head_state(skb
);
3419 kmem_cache_free(skbuff_head_cache
, skb
);
3424 EXPORT_SYMBOL(kfree_skb_partial
);
3427 * skb_try_coalesce - try to merge skb to prior one
3429 * @from: buffer to add
3430 * @fragstolen: pointer to boolean
3431 * @delta_truesize: how much more was allocated than was requested
3433 bool skb_try_coalesce(struct sk_buff
*to
, struct sk_buff
*from
,
3434 bool *fragstolen
, int *delta_truesize
)
3436 int i
, delta
, len
= from
->len
;
3438 *fragstolen
= false;
3443 if (len
<= skb_tailroom(to
)) {
3444 BUG_ON(skb_copy_bits(from
, 0, skb_put(to
, len
), len
));
3445 *delta_truesize
= 0;
3449 if (skb_has_frag_list(to
) || skb_has_frag_list(from
))
3452 if (skb_headlen(from
) != 0) {
3454 unsigned int offset
;
3456 if (skb_shinfo(to
)->nr_frags
+
3457 skb_shinfo(from
)->nr_frags
>= MAX_SKB_FRAGS
)
3460 if (skb_head_is_locked(from
))
3463 delta
= from
->truesize
- SKB_DATA_ALIGN(sizeof(struct sk_buff
));
3465 page
= virt_to_head_page(from
->head
);
3466 offset
= from
->data
- (unsigned char *)page_address(page
);
3468 skb_fill_page_desc(to
, skb_shinfo(to
)->nr_frags
,
3469 page
, offset
, skb_headlen(from
));
3472 if (skb_shinfo(to
)->nr_frags
+
3473 skb_shinfo(from
)->nr_frags
> MAX_SKB_FRAGS
)
3476 delta
= from
->truesize
- SKB_TRUESIZE(skb_end_offset(from
));
3479 WARN_ON_ONCE(delta
< len
);
3481 memcpy(skb_shinfo(to
)->frags
+ skb_shinfo(to
)->nr_frags
,
3482 skb_shinfo(from
)->frags
,
3483 skb_shinfo(from
)->nr_frags
* sizeof(skb_frag_t
));
3484 skb_shinfo(to
)->nr_frags
+= skb_shinfo(from
)->nr_frags
;
3486 if (!skb_cloned(from
))
3487 skb_shinfo(from
)->nr_frags
= 0;
3489 /* if the skb is not cloned this does nothing
3490 * since we set nr_frags to 0.
3492 for (i
= 0; i
< skb_shinfo(from
)->nr_frags
; i
++)
3493 skb_frag_ref(from
, i
);
3495 to
->truesize
+= delta
;
3497 to
->data_len
+= len
;
3499 *delta_truesize
= delta
;
3502 EXPORT_SYMBOL(skb_try_coalesce
);