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 skb
->mac_header
= (typeof(skb
->mac_header
))~0U;
208 * __alloc_skb - allocate a network buffer
209 * @size: size to allocate
210 * @gfp_mask: allocation mask
211 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
212 * instead of head cache and allocate a cloned (child) skb.
213 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
214 * allocations in case the data is required for writeback
215 * @node: numa node to allocate memory on
217 * Allocate a new &sk_buff. The returned buffer has no headroom and a
218 * tail room of at least size bytes. The object has a reference count
219 * of one. The return is the buffer. On a failure the return is %NULL.
221 * Buffers may only be allocated from interrupts using a @gfp_mask of
224 struct sk_buff
*__alloc_skb(unsigned int size
, gfp_t gfp_mask
,
227 struct kmem_cache
*cache
;
228 struct skb_shared_info
*shinfo
;
233 cache
= (flags
& SKB_ALLOC_FCLONE
)
234 ? skbuff_fclone_cache
: skbuff_head_cache
;
236 if (sk_memalloc_socks() && (flags
& SKB_ALLOC_RX
))
237 gfp_mask
|= __GFP_MEMALLOC
;
240 skb
= kmem_cache_alloc_node(cache
, gfp_mask
& ~__GFP_DMA
, node
);
245 /* We do our best to align skb_shared_info on a separate cache
246 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
247 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
248 * Both skb->head and skb_shared_info are cache line aligned.
250 size
= SKB_DATA_ALIGN(size
);
251 size
+= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
252 data
= kmalloc_reserve(size
, gfp_mask
, node
, &pfmemalloc
);
255 /* kmalloc(size) might give us more room than requested.
256 * Put skb_shared_info exactly at the end of allocated zone,
257 * to allow max possible filling before reallocation.
259 size
= SKB_WITH_OVERHEAD(ksize(data
));
260 prefetchw(data
+ size
);
263 * Only clear those fields we need to clear, not those that we will
264 * actually initialise below. Hence, don't put any more fields after
265 * the tail pointer in struct sk_buff!
267 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
268 /* Account for allocated memory : skb + skb->head */
269 skb
->truesize
= SKB_TRUESIZE(size
);
270 skb
->pfmemalloc
= pfmemalloc
;
271 atomic_set(&skb
->users
, 1);
274 skb_reset_tail_pointer(skb
);
275 skb
->end
= skb
->tail
+ size
;
276 skb
->mac_header
= (typeof(skb
->mac_header
))~0U;
277 skb
->transport_header
= (typeof(skb
->transport_header
))~0U;
279 /* make sure we initialize shinfo sequentially */
280 shinfo
= skb_shinfo(skb
);
281 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
282 atomic_set(&shinfo
->dataref
, 1);
283 kmemcheck_annotate_variable(shinfo
->destructor_arg
);
285 if (flags
& SKB_ALLOC_FCLONE
) {
286 struct sk_buff
*child
= skb
+ 1;
287 atomic_t
*fclone_ref
= (atomic_t
*) (child
+ 1);
289 kmemcheck_annotate_bitfield(child
, flags1
);
290 kmemcheck_annotate_bitfield(child
, flags2
);
291 skb
->fclone
= SKB_FCLONE_ORIG
;
292 atomic_set(fclone_ref
, 1);
294 child
->fclone
= SKB_FCLONE_UNAVAILABLE
;
295 child
->pfmemalloc
= pfmemalloc
;
300 kmem_cache_free(cache
, skb
);
304 EXPORT_SYMBOL(__alloc_skb
);
307 * build_skb - build a network buffer
308 * @data: data buffer provided by caller
309 * @frag_size: size of fragment, or 0 if head was kmalloced
311 * Allocate a new &sk_buff. Caller provides space holding head and
312 * skb_shared_info. @data must have been allocated by kmalloc() only if
313 * @frag_size is 0, otherwise data should come from the page allocator.
314 * The return is the new skb buffer.
315 * On a failure the return is %NULL, and @data is not freed.
317 * Before IO, driver allocates only data buffer where NIC put incoming frame
318 * Driver should add room at head (NET_SKB_PAD) and
319 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
320 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
321 * before giving packet to stack.
322 * RX rings only contains data buffers, not full skbs.
324 struct sk_buff
*build_skb(void *data
, unsigned int frag_size
)
326 struct skb_shared_info
*shinfo
;
328 unsigned int size
= frag_size
? : ksize(data
);
330 skb
= kmem_cache_alloc(skbuff_head_cache
, GFP_ATOMIC
);
334 size
-= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
336 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
337 skb
->truesize
= SKB_TRUESIZE(size
);
338 skb
->head_frag
= frag_size
!= 0;
339 atomic_set(&skb
->users
, 1);
342 skb_reset_tail_pointer(skb
);
343 skb
->end
= skb
->tail
+ size
;
344 skb
->mac_header
= (typeof(skb
->mac_header
))~0U;
345 skb
->transport_header
= (typeof(skb
->transport_header
))~0U;
347 /* make sure we initialize shinfo sequentially */
348 shinfo
= skb_shinfo(skb
);
349 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
350 atomic_set(&shinfo
->dataref
, 1);
351 kmemcheck_annotate_variable(shinfo
->destructor_arg
);
355 EXPORT_SYMBOL(build_skb
);
357 struct netdev_alloc_cache
{
358 struct page_frag frag
;
359 /* we maintain a pagecount bias, so that we dont dirty cache line
360 * containing page->_count every time we allocate a fragment.
362 unsigned int pagecnt_bias
;
364 static DEFINE_PER_CPU(struct netdev_alloc_cache
, netdev_alloc_cache
);
366 static void *__netdev_alloc_frag(unsigned int fragsz
, gfp_t gfp_mask
)
368 struct netdev_alloc_cache
*nc
;
373 local_irq_save(flags
);
374 nc
= &__get_cpu_var(netdev_alloc_cache
);
375 if (unlikely(!nc
->frag
.page
)) {
377 for (order
= NETDEV_FRAG_PAGE_MAX_ORDER
; ;) {
378 gfp_t gfp
= gfp_mask
;
381 gfp
|= __GFP_COMP
| __GFP_NOWARN
;
382 nc
->frag
.page
= alloc_pages(gfp
, order
);
383 if (likely(nc
->frag
.page
))
388 nc
->frag
.size
= PAGE_SIZE
<< order
;
390 atomic_set(&nc
->frag
.page
->_count
, NETDEV_PAGECNT_MAX_BIAS
);
391 nc
->pagecnt_bias
= NETDEV_PAGECNT_MAX_BIAS
;
395 if (nc
->frag
.offset
+ fragsz
> nc
->frag
.size
) {
396 /* avoid unnecessary locked operations if possible */
397 if ((atomic_read(&nc
->frag
.page
->_count
) == nc
->pagecnt_bias
) ||
398 atomic_sub_and_test(nc
->pagecnt_bias
, &nc
->frag
.page
->_count
))
403 data
= page_address(nc
->frag
.page
) + nc
->frag
.offset
;
404 nc
->frag
.offset
+= fragsz
;
407 local_irq_restore(flags
);
412 * netdev_alloc_frag - allocate a page fragment
413 * @fragsz: fragment size
415 * Allocates a frag from a page for receive buffer.
416 * Uses GFP_ATOMIC allocations.
418 void *netdev_alloc_frag(unsigned int fragsz
)
420 return __netdev_alloc_frag(fragsz
, GFP_ATOMIC
| __GFP_COLD
);
422 EXPORT_SYMBOL(netdev_alloc_frag
);
425 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
426 * @dev: network device to receive on
427 * @length: length to allocate
428 * @gfp_mask: get_free_pages mask, passed to alloc_skb
430 * Allocate a new &sk_buff and assign it a usage count of one. The
431 * buffer has unspecified headroom built in. Users should allocate
432 * the headroom they think they need without accounting for the
433 * built in space. The built in space is used for optimisations.
435 * %NULL is returned if there is no free memory.
437 struct sk_buff
*__netdev_alloc_skb(struct net_device
*dev
,
438 unsigned int length
, gfp_t gfp_mask
)
440 struct sk_buff
*skb
= NULL
;
441 unsigned int fragsz
= SKB_DATA_ALIGN(length
+ NET_SKB_PAD
) +
442 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
444 if (fragsz
<= PAGE_SIZE
&& !(gfp_mask
& (__GFP_WAIT
| GFP_DMA
))) {
447 if (sk_memalloc_socks())
448 gfp_mask
|= __GFP_MEMALLOC
;
450 data
= __netdev_alloc_frag(fragsz
, gfp_mask
);
453 skb
= build_skb(data
, fragsz
);
455 put_page(virt_to_head_page(data
));
458 skb
= __alloc_skb(length
+ NET_SKB_PAD
, gfp_mask
,
459 SKB_ALLOC_RX
, NUMA_NO_NODE
);
462 skb_reserve(skb
, NET_SKB_PAD
);
467 EXPORT_SYMBOL(__netdev_alloc_skb
);
469 void skb_add_rx_frag(struct sk_buff
*skb
, int i
, struct page
*page
, int off
,
470 int size
, unsigned int truesize
)
472 skb_fill_page_desc(skb
, i
, page
, off
, size
);
474 skb
->data_len
+= size
;
475 skb
->truesize
+= truesize
;
477 EXPORT_SYMBOL(skb_add_rx_frag
);
479 static void skb_drop_list(struct sk_buff
**listp
)
481 kfree_skb_list(*listp
);
485 static inline void skb_drop_fraglist(struct sk_buff
*skb
)
487 skb_drop_list(&skb_shinfo(skb
)->frag_list
);
490 static void skb_clone_fraglist(struct sk_buff
*skb
)
492 struct sk_buff
*list
;
494 skb_walk_frags(skb
, list
)
498 static void skb_free_head(struct sk_buff
*skb
)
501 put_page(virt_to_head_page(skb
->head
));
506 static void skb_release_data(struct sk_buff
*skb
)
509 !atomic_sub_return(skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1,
510 &skb_shinfo(skb
)->dataref
)) {
511 if (skb_shinfo(skb
)->nr_frags
) {
513 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
514 skb_frag_unref(skb
, i
);
518 * If skb buf is from userspace, we need to notify the caller
519 * the lower device DMA has done;
521 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
522 struct ubuf_info
*uarg
;
524 uarg
= skb_shinfo(skb
)->destructor_arg
;
526 uarg
->callback(uarg
, true);
529 if (skb_has_frag_list(skb
))
530 skb_drop_fraglist(skb
);
537 * Free an skbuff by memory without cleaning the state.
539 static void kfree_skbmem(struct sk_buff
*skb
)
541 struct sk_buff
*other
;
542 atomic_t
*fclone_ref
;
544 switch (skb
->fclone
) {
545 case SKB_FCLONE_UNAVAILABLE
:
546 kmem_cache_free(skbuff_head_cache
, skb
);
549 case SKB_FCLONE_ORIG
:
550 fclone_ref
= (atomic_t
*) (skb
+ 2);
551 if (atomic_dec_and_test(fclone_ref
))
552 kmem_cache_free(skbuff_fclone_cache
, skb
);
555 case SKB_FCLONE_CLONE
:
556 fclone_ref
= (atomic_t
*) (skb
+ 1);
559 /* The clone portion is available for
560 * fast-cloning again.
562 skb
->fclone
= SKB_FCLONE_UNAVAILABLE
;
564 if (atomic_dec_and_test(fclone_ref
))
565 kmem_cache_free(skbuff_fclone_cache
, other
);
570 static void skb_release_head_state(struct sk_buff
*skb
)
574 secpath_put(skb
->sp
);
576 if (skb
->destructor
) {
578 skb
->destructor(skb
);
580 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
581 nf_conntrack_put(skb
->nfct
);
583 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
584 nf_conntrack_put_reasm(skb
->nfct_reasm
);
586 #ifdef CONFIG_BRIDGE_NETFILTER
587 nf_bridge_put(skb
->nf_bridge
);
589 /* XXX: IS this still necessary? - JHS */
590 #ifdef CONFIG_NET_SCHED
592 #ifdef CONFIG_NET_CLS_ACT
598 /* Free everything but the sk_buff shell. */
599 static void skb_release_all(struct sk_buff
*skb
)
601 skb_release_head_state(skb
);
602 if (likely(skb
->head
))
603 skb_release_data(skb
);
607 * __kfree_skb - private function
610 * Free an sk_buff. Release anything attached to the buffer.
611 * Clean the state. This is an internal helper function. Users should
612 * always call kfree_skb
615 void __kfree_skb(struct sk_buff
*skb
)
617 skb_release_all(skb
);
620 EXPORT_SYMBOL(__kfree_skb
);
623 * kfree_skb - free an sk_buff
624 * @skb: buffer to free
626 * Drop a reference to the buffer and free it if the usage count has
629 void kfree_skb(struct sk_buff
*skb
)
633 if (likely(atomic_read(&skb
->users
) == 1))
635 else if (likely(!atomic_dec_and_test(&skb
->users
)))
637 trace_kfree_skb(skb
, __builtin_return_address(0));
640 EXPORT_SYMBOL(kfree_skb
);
642 void kfree_skb_list(struct sk_buff
*segs
)
645 struct sk_buff
*next
= segs
->next
;
651 EXPORT_SYMBOL(kfree_skb_list
);
654 * skb_tx_error - report an sk_buff xmit error
655 * @skb: buffer that triggered an error
657 * Report xmit error if a device callback is tracking this skb.
658 * skb must be freed afterwards.
660 void skb_tx_error(struct sk_buff
*skb
)
662 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
663 struct ubuf_info
*uarg
;
665 uarg
= skb_shinfo(skb
)->destructor_arg
;
667 uarg
->callback(uarg
, false);
668 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
671 EXPORT_SYMBOL(skb_tx_error
);
674 * consume_skb - free an skbuff
675 * @skb: buffer to free
677 * Drop a ref to the buffer and free it if the usage count has hit zero
678 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
679 * is being dropped after a failure and notes that
681 void consume_skb(struct sk_buff
*skb
)
685 if (likely(atomic_read(&skb
->users
) == 1))
687 else if (likely(!atomic_dec_and_test(&skb
->users
)))
689 trace_consume_skb(skb
);
692 EXPORT_SYMBOL(consume_skb
);
694 static void __copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
696 new->tstamp
= old
->tstamp
;
698 new->transport_header
= old
->transport_header
;
699 new->network_header
= old
->network_header
;
700 new->mac_header
= old
->mac_header
;
701 new->inner_protocol
= old
->inner_protocol
;
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
);
743 #ifdef CONFIG_NET_RX_BUSY_POLL
744 new->napi_id
= old
->napi_id
;
749 * You should not add any new code to this function. Add it to
750 * __copy_skb_header above instead.
752 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
754 #define C(x) n->x = skb->x
756 n
->next
= n
->prev
= NULL
;
758 __copy_skb_header(n
, skb
);
763 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
766 n
->destructor
= NULL
;
773 atomic_set(&n
->users
, 1);
775 atomic_inc(&(skb_shinfo(skb
)->dataref
));
783 * skb_morph - morph one skb into another
784 * @dst: the skb to receive the contents
785 * @src: the skb to supply the contents
787 * This is identical to skb_clone except that the target skb is
788 * supplied by the user.
790 * The target skb is returned upon exit.
792 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
794 skb_release_all(dst
);
795 return __skb_clone(dst
, src
);
797 EXPORT_SYMBOL_GPL(skb_morph
);
800 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
801 * @skb: the skb to modify
802 * @gfp_mask: allocation priority
804 * This must be called on SKBTX_DEV_ZEROCOPY skb.
805 * It will copy all frags into kernel and drop the reference
806 * to userspace pages.
808 * If this function is called from an interrupt gfp_mask() must be
811 * Returns 0 on success or a negative error code on failure
812 * to allocate kernel memory to copy to.
814 int skb_copy_ubufs(struct sk_buff
*skb
, gfp_t gfp_mask
)
817 int num_frags
= skb_shinfo(skb
)->nr_frags
;
818 struct page
*page
, *head
= NULL
;
819 struct ubuf_info
*uarg
= skb_shinfo(skb
)->destructor_arg
;
821 for (i
= 0; i
< num_frags
; i
++) {
823 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
825 page
= alloc_page(gfp_mask
);
828 struct page
*next
= (struct page
*)page_private(head
);
834 vaddr
= kmap_atomic(skb_frag_page(f
));
835 memcpy(page_address(page
),
836 vaddr
+ f
->page_offset
, skb_frag_size(f
));
837 kunmap_atomic(vaddr
);
838 set_page_private(page
, (unsigned long)head
);
842 /* skb frags release userspace buffers */
843 for (i
= 0; i
< num_frags
; i
++)
844 skb_frag_unref(skb
, i
);
846 uarg
->callback(uarg
, false);
848 /* skb frags point to kernel buffers */
849 for (i
= num_frags
- 1; i
>= 0; i
--) {
850 __skb_fill_page_desc(skb
, i
, head
, 0,
851 skb_shinfo(skb
)->frags
[i
].size
);
852 head
= (struct page
*)page_private(head
);
855 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
858 EXPORT_SYMBOL_GPL(skb_copy_ubufs
);
861 * skb_clone - duplicate an sk_buff
862 * @skb: buffer to clone
863 * @gfp_mask: allocation priority
865 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
866 * copies share the same packet data but not structure. The new
867 * buffer has a reference count of 1. If the allocation fails the
868 * function returns %NULL otherwise the new buffer is returned.
870 * If this function is called from an interrupt gfp_mask() must be
874 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
878 if (skb_orphan_frags(skb
, gfp_mask
))
882 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
883 n
->fclone
== SKB_FCLONE_UNAVAILABLE
) {
884 atomic_t
*fclone_ref
= (atomic_t
*) (n
+ 1);
885 n
->fclone
= SKB_FCLONE_CLONE
;
886 atomic_inc(fclone_ref
);
888 if (skb_pfmemalloc(skb
))
889 gfp_mask
|= __GFP_MEMALLOC
;
891 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
895 kmemcheck_annotate_bitfield(n
, flags1
);
896 kmemcheck_annotate_bitfield(n
, flags2
);
897 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
900 return __skb_clone(n
, skb
);
902 EXPORT_SYMBOL(skb_clone
);
904 static void skb_headers_offset_update(struct sk_buff
*skb
, int off
)
906 /* {transport,network,mac}_header and tail are relative to skb->head */
907 skb
->transport_header
+= off
;
908 skb
->network_header
+= off
;
909 if (skb_mac_header_was_set(skb
))
910 skb
->mac_header
+= off
;
911 skb
->inner_transport_header
+= off
;
912 skb
->inner_network_header
+= off
;
913 skb
->inner_mac_header
+= off
;
916 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
918 __copy_skb_header(new, old
);
920 skb_shinfo(new)->gso_size
= skb_shinfo(old
)->gso_size
;
921 skb_shinfo(new)->gso_segs
= skb_shinfo(old
)->gso_segs
;
922 skb_shinfo(new)->gso_type
= skb_shinfo(old
)->gso_type
;
925 static inline int skb_alloc_rx_flag(const struct sk_buff
*skb
)
927 if (skb_pfmemalloc(skb
))
933 * skb_copy - create private copy of an sk_buff
934 * @skb: buffer to copy
935 * @gfp_mask: allocation priority
937 * Make a copy of both an &sk_buff and its data. This is used when the
938 * caller wishes to modify the data and needs a private copy of the
939 * data to alter. Returns %NULL on failure or the pointer to the buffer
940 * on success. The returned buffer has a reference count of 1.
942 * As by-product this function converts non-linear &sk_buff to linear
943 * one, so that &sk_buff becomes completely private and caller is allowed
944 * to modify all the data of returned buffer. This means that this
945 * function is not recommended for use in circumstances when only
946 * header is going to be modified. Use pskb_copy() instead.
949 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
951 int headerlen
= skb_headroom(skb
);
952 unsigned int size
= skb_end_offset(skb
) + skb
->data_len
;
953 struct sk_buff
*n
= __alloc_skb(size
, gfp_mask
,
954 skb_alloc_rx_flag(skb
), NUMA_NO_NODE
);
959 /* Set the data pointer */
960 skb_reserve(n
, headerlen
);
961 /* Set the tail pointer and length */
962 skb_put(n
, skb
->len
);
964 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
967 copy_skb_header(n
, skb
);
970 EXPORT_SYMBOL(skb_copy
);
973 * __pskb_copy - create copy of an sk_buff with private head.
974 * @skb: buffer to copy
975 * @headroom: headroom of new skb
976 * @gfp_mask: allocation priority
978 * Make a copy of both an &sk_buff and part of its data, located
979 * in header. Fragmented data remain shared. This is used when
980 * the caller wishes to modify only header of &sk_buff and needs
981 * private copy of the header to alter. Returns %NULL on failure
982 * or the pointer to the buffer on success.
983 * The returned buffer has a reference count of 1.
986 struct sk_buff
*__pskb_copy(struct sk_buff
*skb
, int headroom
, gfp_t gfp_mask
)
988 unsigned int size
= skb_headlen(skb
) + headroom
;
989 struct sk_buff
*n
= __alloc_skb(size
, gfp_mask
,
990 skb_alloc_rx_flag(skb
), NUMA_NO_NODE
);
995 /* Set the data pointer */
996 skb_reserve(n
, headroom
);
997 /* Set the tail pointer and length */
998 skb_put(n
, skb_headlen(skb
));
1000 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
1002 n
->truesize
+= skb
->data_len
;
1003 n
->data_len
= skb
->data_len
;
1006 if (skb_shinfo(skb
)->nr_frags
) {
1009 if (skb_orphan_frags(skb
, gfp_mask
)) {
1014 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1015 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
1016 skb_frag_ref(skb
, i
);
1018 skb_shinfo(n
)->nr_frags
= i
;
1021 if (skb_has_frag_list(skb
)) {
1022 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
1023 skb_clone_fraglist(n
);
1026 copy_skb_header(n
, skb
);
1030 EXPORT_SYMBOL(__pskb_copy
);
1033 * pskb_expand_head - reallocate header of &sk_buff
1034 * @skb: buffer to reallocate
1035 * @nhead: room to add at head
1036 * @ntail: room to add at tail
1037 * @gfp_mask: allocation priority
1039 * Expands (or creates identical copy, if &nhead and &ntail are zero)
1040 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
1041 * reference count of 1. Returns zero in the case of success or error,
1042 * if expansion failed. In the last case, &sk_buff is not changed.
1044 * All the pointers pointing into skb header may change and must be
1045 * reloaded after call to this function.
1048 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
1053 int size
= nhead
+ skb_end_offset(skb
) + ntail
;
1058 if (skb_shared(skb
))
1061 size
= SKB_DATA_ALIGN(size
);
1063 if (skb_pfmemalloc(skb
))
1064 gfp_mask
|= __GFP_MEMALLOC
;
1065 data
= kmalloc_reserve(size
+ SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)),
1066 gfp_mask
, NUMA_NO_NODE
, NULL
);
1069 size
= SKB_WITH_OVERHEAD(ksize(data
));
1071 /* Copy only real data... and, alas, header. This should be
1072 * optimized for the cases when header is void.
1074 memcpy(data
+ nhead
, skb
->head
, skb_tail_pointer(skb
) - skb
->head
);
1076 memcpy((struct skb_shared_info
*)(data
+ size
),
1078 offsetof(struct skb_shared_info
, frags
[skb_shinfo(skb
)->nr_frags
]));
1081 * if shinfo is shared we must drop the old head gracefully, but if it
1082 * is not we can just drop the old head and let the existing refcount
1083 * be since all we did is relocate the values
1085 if (skb_cloned(skb
)) {
1086 /* copy this zero copy skb frags */
1087 if (skb_orphan_frags(skb
, gfp_mask
))
1089 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1090 skb_frag_ref(skb
, i
);
1092 if (skb_has_frag_list(skb
))
1093 skb_clone_fraglist(skb
);
1095 skb_release_data(skb
);
1099 off
= (data
+ nhead
) - skb
->head
;
1104 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1108 skb
->end
= skb
->head
+ size
;
1111 skb_headers_offset_update(skb
, nhead
);
1112 /* Only adjust this if it actually is csum_start rather than csum */
1113 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1114 skb
->csum_start
+= nhead
;
1118 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
1126 EXPORT_SYMBOL(pskb_expand_head
);
1128 /* Make private copy of skb with writable head and some headroom */
1130 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
1132 struct sk_buff
*skb2
;
1133 int delta
= headroom
- skb_headroom(skb
);
1136 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
1138 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1139 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
1147 EXPORT_SYMBOL(skb_realloc_headroom
);
1150 * skb_copy_expand - copy and expand sk_buff
1151 * @skb: buffer to copy
1152 * @newheadroom: new free bytes at head
1153 * @newtailroom: new free bytes at tail
1154 * @gfp_mask: allocation priority
1156 * Make a copy of both an &sk_buff and its data and while doing so
1157 * allocate additional space.
1159 * This is used when the caller wishes to modify the data and needs a
1160 * private copy of the data to alter as well as more space for new fields.
1161 * Returns %NULL on failure or the pointer to the buffer
1162 * on success. The returned buffer has a reference count of 1.
1164 * You must pass %GFP_ATOMIC as the allocation priority if this function
1165 * is called from an interrupt.
1167 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
1168 int newheadroom
, int newtailroom
,
1172 * Allocate the copy buffer
1174 struct sk_buff
*n
= __alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
1175 gfp_mask
, skb_alloc_rx_flag(skb
),
1177 int oldheadroom
= skb_headroom(skb
);
1178 int head_copy_len
, head_copy_off
;
1184 skb_reserve(n
, newheadroom
);
1186 /* Set the tail pointer and length */
1187 skb_put(n
, skb
->len
);
1189 head_copy_len
= oldheadroom
;
1191 if (newheadroom
<= head_copy_len
)
1192 head_copy_len
= newheadroom
;
1194 head_copy_off
= newheadroom
- head_copy_len
;
1196 /* Copy the linear header and data. */
1197 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
1198 skb
->len
+ head_copy_len
))
1201 copy_skb_header(n
, skb
);
1203 off
= newheadroom
- oldheadroom
;
1204 if (n
->ip_summed
== CHECKSUM_PARTIAL
)
1205 n
->csum_start
+= off
;
1207 skb_headers_offset_update(n
, off
);
1211 EXPORT_SYMBOL(skb_copy_expand
);
1214 * skb_pad - zero pad the tail of an skb
1215 * @skb: buffer to pad
1216 * @pad: space to pad
1218 * Ensure that a buffer is followed by a padding area that is zero
1219 * filled. Used by network drivers which may DMA or transfer data
1220 * beyond the buffer end onto the wire.
1222 * May return error in out of memory cases. The skb is freed on error.
1225 int skb_pad(struct sk_buff
*skb
, int pad
)
1230 /* If the skbuff is non linear tailroom is always zero.. */
1231 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
1232 memset(skb
->data
+skb
->len
, 0, pad
);
1236 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
1237 if (likely(skb_cloned(skb
) || ntail
> 0)) {
1238 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
1243 /* FIXME: The use of this function with non-linear skb's really needs
1246 err
= skb_linearize(skb
);
1250 memset(skb
->data
+ skb
->len
, 0, pad
);
1257 EXPORT_SYMBOL(skb_pad
);
1260 * skb_put - add data to a buffer
1261 * @skb: buffer to use
1262 * @len: amount of data to add
1264 * This function extends the used data area of the buffer. If this would
1265 * exceed the total buffer size the kernel will panic. A pointer to the
1266 * first byte of the extra data is returned.
1268 unsigned char *skb_put(struct sk_buff
*skb
, unsigned int len
)
1270 unsigned char *tmp
= skb_tail_pointer(skb
);
1271 SKB_LINEAR_ASSERT(skb
);
1274 if (unlikely(skb
->tail
> skb
->end
))
1275 skb_over_panic(skb
, len
, __builtin_return_address(0));
1278 EXPORT_SYMBOL(skb_put
);
1281 * skb_push - add data to the start of a buffer
1282 * @skb: buffer to use
1283 * @len: amount of data to add
1285 * This function extends the used data area of the buffer at the buffer
1286 * start. If this would exceed the total buffer headroom the kernel will
1287 * panic. A pointer to the first byte of the extra data is returned.
1289 unsigned char *skb_push(struct sk_buff
*skb
, unsigned int len
)
1293 if (unlikely(skb
->data
<skb
->head
))
1294 skb_under_panic(skb
, len
, __builtin_return_address(0));
1297 EXPORT_SYMBOL(skb_push
);
1300 * skb_pull - remove data from the start of a buffer
1301 * @skb: buffer to use
1302 * @len: amount of data to remove
1304 * This function removes data from the start of a buffer, returning
1305 * the memory to the headroom. A pointer to the next data in the buffer
1306 * is returned. Once the data has been pulled future pushes will overwrite
1309 unsigned char *skb_pull(struct sk_buff
*skb
, unsigned int len
)
1311 return skb_pull_inline(skb
, len
);
1313 EXPORT_SYMBOL(skb_pull
);
1316 * skb_trim - remove end from a buffer
1317 * @skb: buffer to alter
1320 * Cut the length of a buffer down by removing data from the tail. If
1321 * the buffer is already under the length specified it is not modified.
1322 * The skb must be linear.
1324 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1327 __skb_trim(skb
, len
);
1329 EXPORT_SYMBOL(skb_trim
);
1331 /* Trims skb to length len. It can change skb pointers.
1334 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1336 struct sk_buff
**fragp
;
1337 struct sk_buff
*frag
;
1338 int offset
= skb_headlen(skb
);
1339 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1343 if (skb_cloned(skb
) &&
1344 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1351 for (; i
< nfrags
; i
++) {
1352 int end
= offset
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1359 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
++], len
- offset
);
1362 skb_shinfo(skb
)->nr_frags
= i
;
1364 for (; i
< nfrags
; i
++)
1365 skb_frag_unref(skb
, i
);
1367 if (skb_has_frag_list(skb
))
1368 skb_drop_fraglist(skb
);
1372 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1373 fragp
= &frag
->next
) {
1374 int end
= offset
+ frag
->len
;
1376 if (skb_shared(frag
)) {
1377 struct sk_buff
*nfrag
;
1379 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1380 if (unlikely(!nfrag
))
1383 nfrag
->next
= frag
->next
;
1395 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1399 skb_drop_list(&frag
->next
);
1404 if (len
> skb_headlen(skb
)) {
1405 skb
->data_len
-= skb
->len
- len
;
1410 skb_set_tail_pointer(skb
, len
);
1415 EXPORT_SYMBOL(___pskb_trim
);
1418 * __pskb_pull_tail - advance tail of skb header
1419 * @skb: buffer to reallocate
1420 * @delta: number of bytes to advance tail
1422 * The function makes a sense only on a fragmented &sk_buff,
1423 * it expands header moving its tail forward and copying necessary
1424 * data from fragmented part.
1426 * &sk_buff MUST have reference count of 1.
1428 * Returns %NULL (and &sk_buff does not change) if pull failed
1429 * or value of new tail of skb in the case of success.
1431 * All the pointers pointing into skb header may change and must be
1432 * reloaded after call to this function.
1435 /* Moves tail of skb head forward, copying data from fragmented part,
1436 * when it is necessary.
1437 * 1. It may fail due to malloc failure.
1438 * 2. It may change skb pointers.
1440 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1442 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1444 /* If skb has not enough free space at tail, get new one
1445 * plus 128 bytes for future expansions. If we have enough
1446 * room at tail, reallocate without expansion only if skb is cloned.
1448 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1450 if (eat
> 0 || skb_cloned(skb
)) {
1451 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1456 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
1459 /* Optimization: no fragments, no reasons to preestimate
1460 * size of pulled pages. Superb.
1462 if (!skb_has_frag_list(skb
))
1465 /* Estimate size of pulled pages. */
1467 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1468 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1475 /* If we need update frag list, we are in troubles.
1476 * Certainly, it possible to add an offset to skb data,
1477 * but taking into account that pulling is expected to
1478 * be very rare operation, it is worth to fight against
1479 * further bloating skb head and crucify ourselves here instead.
1480 * Pure masohism, indeed. 8)8)
1483 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1484 struct sk_buff
*clone
= NULL
;
1485 struct sk_buff
*insp
= NULL
;
1490 if (list
->len
<= eat
) {
1491 /* Eaten as whole. */
1496 /* Eaten partially. */
1498 if (skb_shared(list
)) {
1499 /* Sucks! We need to fork list. :-( */
1500 clone
= skb_clone(list
, GFP_ATOMIC
);
1506 /* This may be pulled without
1510 if (!pskb_pull(list
, eat
)) {
1518 /* Free pulled out fragments. */
1519 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1520 skb_shinfo(skb
)->frag_list
= list
->next
;
1523 /* And insert new clone at head. */
1526 skb_shinfo(skb
)->frag_list
= clone
;
1529 /* Success! Now we may commit changes to skb data. */
1534 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1535 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1538 skb_frag_unref(skb
, i
);
1541 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1543 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1544 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[k
], eat
);
1550 skb_shinfo(skb
)->nr_frags
= k
;
1553 skb
->data_len
-= delta
;
1555 return skb_tail_pointer(skb
);
1557 EXPORT_SYMBOL(__pskb_pull_tail
);
1560 * skb_copy_bits - copy bits from skb to kernel buffer
1562 * @offset: offset in source
1563 * @to: destination buffer
1564 * @len: number of bytes to copy
1566 * Copy the specified number of bytes from the source skb to the
1567 * destination buffer.
1570 * If its prototype is ever changed,
1571 * check arch/{*}/net/{*}.S files,
1572 * since it is called from BPF assembly code.
1574 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1576 int start
= skb_headlen(skb
);
1577 struct sk_buff
*frag_iter
;
1580 if (offset
> (int)skb
->len
- len
)
1584 if ((copy
= start
- offset
) > 0) {
1587 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
1588 if ((len
-= copy
) == 0)
1594 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1596 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
1598 WARN_ON(start
> offset
+ len
);
1600 end
= start
+ skb_frag_size(f
);
1601 if ((copy
= end
- offset
) > 0) {
1607 vaddr
= kmap_atomic(skb_frag_page(f
));
1609 vaddr
+ f
->page_offset
+ offset
- start
,
1611 kunmap_atomic(vaddr
);
1613 if ((len
-= copy
) == 0)
1621 skb_walk_frags(skb
, frag_iter
) {
1624 WARN_ON(start
> offset
+ len
);
1626 end
= start
+ frag_iter
->len
;
1627 if ((copy
= end
- offset
) > 0) {
1630 if (skb_copy_bits(frag_iter
, offset
- start
, to
, copy
))
1632 if ((len
-= copy
) == 0)
1646 EXPORT_SYMBOL(skb_copy_bits
);
1649 * Callback from splice_to_pipe(), if we need to release some pages
1650 * at the end of the spd in case we error'ed out in filling the pipe.
1652 static void sock_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
1654 put_page(spd
->pages
[i
]);
1657 static struct page
*linear_to_page(struct page
*page
, unsigned int *len
,
1658 unsigned int *offset
,
1661 struct page_frag
*pfrag
= sk_page_frag(sk
);
1663 if (!sk_page_frag_refill(sk
, pfrag
))
1666 *len
= min_t(unsigned int, *len
, pfrag
->size
- pfrag
->offset
);
1668 memcpy(page_address(pfrag
->page
) + pfrag
->offset
,
1669 page_address(page
) + *offset
, *len
);
1670 *offset
= pfrag
->offset
;
1671 pfrag
->offset
+= *len
;
1676 static bool spd_can_coalesce(const struct splice_pipe_desc
*spd
,
1678 unsigned int offset
)
1680 return spd
->nr_pages
&&
1681 spd
->pages
[spd
->nr_pages
- 1] == page
&&
1682 (spd
->partial
[spd
->nr_pages
- 1].offset
+
1683 spd
->partial
[spd
->nr_pages
- 1].len
== offset
);
1687 * Fill page/offset/length into spd, if it can hold more pages.
1689 static bool spd_fill_page(struct splice_pipe_desc
*spd
,
1690 struct pipe_inode_info
*pipe
, struct page
*page
,
1691 unsigned int *len
, unsigned int offset
,
1695 if (unlikely(spd
->nr_pages
== MAX_SKB_FRAGS
))
1699 page
= linear_to_page(page
, len
, &offset
, sk
);
1703 if (spd_can_coalesce(spd
, page
, offset
)) {
1704 spd
->partial
[spd
->nr_pages
- 1].len
+= *len
;
1708 spd
->pages
[spd
->nr_pages
] = page
;
1709 spd
->partial
[spd
->nr_pages
].len
= *len
;
1710 spd
->partial
[spd
->nr_pages
].offset
= offset
;
1716 static bool __splice_segment(struct page
*page
, unsigned int poff
,
1717 unsigned int plen
, unsigned int *off
,
1719 struct splice_pipe_desc
*spd
, bool linear
,
1721 struct pipe_inode_info
*pipe
)
1726 /* skip this segment if already processed */
1732 /* ignore any bits we already processed */
1738 unsigned int flen
= min(*len
, plen
);
1740 if (spd_fill_page(spd
, pipe
, page
, &flen
, poff
,
1746 } while (*len
&& plen
);
1752 * Map linear and fragment data from the skb to spd. It reports true if the
1753 * pipe is full or if we already spliced the requested length.
1755 static bool __skb_splice_bits(struct sk_buff
*skb
, struct pipe_inode_info
*pipe
,
1756 unsigned int *offset
, unsigned int *len
,
1757 struct splice_pipe_desc
*spd
, struct sock
*sk
)
1761 /* map the linear part :
1762 * If skb->head_frag is set, this 'linear' part is backed by a
1763 * fragment, and if the head is not shared with any clones then
1764 * we can avoid a copy since we own the head portion of this page.
1766 if (__splice_segment(virt_to_page(skb
->data
),
1767 (unsigned long) skb
->data
& (PAGE_SIZE
- 1),
1770 skb_head_is_locked(skb
),
1775 * then map the fragments
1777 for (seg
= 0; seg
< skb_shinfo(skb
)->nr_frags
; seg
++) {
1778 const skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[seg
];
1780 if (__splice_segment(skb_frag_page(f
),
1781 f
->page_offset
, skb_frag_size(f
),
1782 offset
, len
, spd
, false, sk
, pipe
))
1790 * Map data from the skb to a pipe. Should handle both the linear part,
1791 * the fragments, and the frag list. It does NOT handle frag lists within
1792 * the frag list, if such a thing exists. We'd probably need to recurse to
1793 * handle that cleanly.
1795 int skb_splice_bits(struct sk_buff
*skb
, unsigned int offset
,
1796 struct pipe_inode_info
*pipe
, unsigned int tlen
,
1799 struct partial_page partial
[MAX_SKB_FRAGS
];
1800 struct page
*pages
[MAX_SKB_FRAGS
];
1801 struct splice_pipe_desc spd
= {
1804 .nr_pages_max
= MAX_SKB_FRAGS
,
1806 .ops
= &sock_pipe_buf_ops
,
1807 .spd_release
= sock_spd_release
,
1809 struct sk_buff
*frag_iter
;
1810 struct sock
*sk
= skb
->sk
;
1814 * __skb_splice_bits() only fails if the output has no room left,
1815 * so no point in going over the frag_list for the error case.
1817 if (__skb_splice_bits(skb
, pipe
, &offset
, &tlen
, &spd
, sk
))
1823 * now see if we have a frag_list to map
1825 skb_walk_frags(skb
, frag_iter
) {
1828 if (__skb_splice_bits(frag_iter
, pipe
, &offset
, &tlen
, &spd
, sk
))
1835 * Drop the socket lock, otherwise we have reverse
1836 * locking dependencies between sk_lock and i_mutex
1837 * here as compared to sendfile(). We enter here
1838 * with the socket lock held, and splice_to_pipe() will
1839 * grab the pipe inode lock. For sendfile() emulation,
1840 * we call into ->sendpage() with the i_mutex lock held
1841 * and networking will grab the socket lock.
1844 ret
= splice_to_pipe(pipe
, &spd
);
1852 * skb_store_bits - store bits from kernel buffer to skb
1853 * @skb: destination buffer
1854 * @offset: offset in destination
1855 * @from: source buffer
1856 * @len: number of bytes to copy
1858 * Copy the specified number of bytes from the source buffer to the
1859 * destination skb. This function handles all the messy bits of
1860 * traversing fragment lists and such.
1863 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
1865 int start
= skb_headlen(skb
);
1866 struct sk_buff
*frag_iter
;
1869 if (offset
> (int)skb
->len
- len
)
1872 if ((copy
= start
- offset
) > 0) {
1875 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
1876 if ((len
-= copy
) == 0)
1882 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1883 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1886 WARN_ON(start
> offset
+ len
);
1888 end
= start
+ skb_frag_size(frag
);
1889 if ((copy
= end
- offset
) > 0) {
1895 vaddr
= kmap_atomic(skb_frag_page(frag
));
1896 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1898 kunmap_atomic(vaddr
);
1900 if ((len
-= copy
) == 0)
1908 skb_walk_frags(skb
, frag_iter
) {
1911 WARN_ON(start
> offset
+ len
);
1913 end
= start
+ frag_iter
->len
;
1914 if ((copy
= end
- offset
) > 0) {
1917 if (skb_store_bits(frag_iter
, offset
- start
,
1920 if ((len
-= copy
) == 0)
1933 EXPORT_SYMBOL(skb_store_bits
);
1935 /* Checksum skb data. */
1937 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
1938 int len
, __wsum csum
)
1940 int start
= skb_headlen(skb
);
1941 int i
, copy
= start
- offset
;
1942 struct sk_buff
*frag_iter
;
1945 /* Checksum header. */
1949 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1950 if ((len
-= copy
) == 0)
1956 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1958 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1960 WARN_ON(start
> offset
+ len
);
1962 end
= start
+ skb_frag_size(frag
);
1963 if ((copy
= end
- offset
) > 0) {
1969 vaddr
= kmap_atomic(skb_frag_page(frag
));
1970 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
1971 offset
- start
, copy
, 0);
1972 kunmap_atomic(vaddr
);
1973 csum
= csum_block_add(csum
, csum2
, pos
);
1982 skb_walk_frags(skb
, frag_iter
) {
1985 WARN_ON(start
> offset
+ len
);
1987 end
= start
+ frag_iter
->len
;
1988 if ((copy
= end
- offset
) > 0) {
1992 csum2
= skb_checksum(frag_iter
, offset
- start
,
1994 csum
= csum_block_add(csum
, csum2
, pos
);
1995 if ((len
-= copy
) == 0)
2006 EXPORT_SYMBOL(skb_checksum
);
2008 /* Both of above in one bottle. */
2010 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
2011 u8
*to
, int len
, __wsum csum
)
2013 int start
= skb_headlen(skb
);
2014 int i
, copy
= start
- offset
;
2015 struct sk_buff
*frag_iter
;
2022 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
2024 if ((len
-= copy
) == 0)
2031 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2034 WARN_ON(start
> offset
+ len
);
2036 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
2037 if ((copy
= end
- offset
) > 0) {
2040 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2044 vaddr
= kmap_atomic(skb_frag_page(frag
));
2045 csum2
= csum_partial_copy_nocheck(vaddr
+
2049 kunmap_atomic(vaddr
);
2050 csum
= csum_block_add(csum
, csum2
, pos
);
2060 skb_walk_frags(skb
, frag_iter
) {
2064 WARN_ON(start
> offset
+ len
);
2066 end
= start
+ frag_iter
->len
;
2067 if ((copy
= end
- offset
) > 0) {
2070 csum2
= skb_copy_and_csum_bits(frag_iter
,
2073 csum
= csum_block_add(csum
, csum2
, pos
);
2074 if ((len
-= copy
) == 0)
2085 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
2087 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
2092 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
2093 csstart
= skb_checksum_start_offset(skb
);
2095 csstart
= skb_headlen(skb
);
2097 BUG_ON(csstart
> skb_headlen(skb
));
2099 skb_copy_from_linear_data(skb
, to
, csstart
);
2102 if (csstart
!= skb
->len
)
2103 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
2104 skb
->len
- csstart
, 0);
2106 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2107 long csstuff
= csstart
+ skb
->csum_offset
;
2109 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
2112 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
2115 * skb_dequeue - remove from the head of the queue
2116 * @list: list to dequeue from
2118 * Remove the head of the list. The list lock is taken so the function
2119 * may be used safely with other locking list functions. The head item is
2120 * returned or %NULL if the list is empty.
2123 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
2125 unsigned long flags
;
2126 struct sk_buff
*result
;
2128 spin_lock_irqsave(&list
->lock
, flags
);
2129 result
= __skb_dequeue(list
);
2130 spin_unlock_irqrestore(&list
->lock
, flags
);
2133 EXPORT_SYMBOL(skb_dequeue
);
2136 * skb_dequeue_tail - remove from the tail of the queue
2137 * @list: list to dequeue from
2139 * Remove the tail of the list. The list lock is taken so the function
2140 * may be used safely with other locking list functions. The tail item is
2141 * returned or %NULL if the list is empty.
2143 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
2145 unsigned long flags
;
2146 struct sk_buff
*result
;
2148 spin_lock_irqsave(&list
->lock
, flags
);
2149 result
= __skb_dequeue_tail(list
);
2150 spin_unlock_irqrestore(&list
->lock
, flags
);
2153 EXPORT_SYMBOL(skb_dequeue_tail
);
2156 * skb_queue_purge - empty a list
2157 * @list: list to empty
2159 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2160 * the list and one reference dropped. This function takes the list
2161 * lock and is atomic with respect to other list locking functions.
2163 void skb_queue_purge(struct sk_buff_head
*list
)
2165 struct sk_buff
*skb
;
2166 while ((skb
= skb_dequeue(list
)) != NULL
)
2169 EXPORT_SYMBOL(skb_queue_purge
);
2172 * skb_queue_head - queue a buffer at the list head
2173 * @list: list to use
2174 * @newsk: buffer to queue
2176 * Queue a buffer at the start of the list. This function takes the
2177 * list lock and can be used safely with other locking &sk_buff functions
2180 * A buffer cannot be placed on two lists at the same time.
2182 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2184 unsigned long flags
;
2186 spin_lock_irqsave(&list
->lock
, flags
);
2187 __skb_queue_head(list
, newsk
);
2188 spin_unlock_irqrestore(&list
->lock
, flags
);
2190 EXPORT_SYMBOL(skb_queue_head
);
2193 * skb_queue_tail - queue a buffer at the list tail
2194 * @list: list to use
2195 * @newsk: buffer to queue
2197 * Queue a buffer at the tail of the list. This function takes the
2198 * list lock and can be used safely with other locking &sk_buff functions
2201 * A buffer cannot be placed on two lists at the same time.
2203 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2205 unsigned long flags
;
2207 spin_lock_irqsave(&list
->lock
, flags
);
2208 __skb_queue_tail(list
, newsk
);
2209 spin_unlock_irqrestore(&list
->lock
, flags
);
2211 EXPORT_SYMBOL(skb_queue_tail
);
2214 * skb_unlink - remove a buffer from a list
2215 * @skb: buffer to remove
2216 * @list: list to use
2218 * Remove a packet from a list. The list locks are taken and this
2219 * function is atomic with respect to other list locked calls
2221 * You must know what list the SKB is on.
2223 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
2225 unsigned long flags
;
2227 spin_lock_irqsave(&list
->lock
, flags
);
2228 __skb_unlink(skb
, list
);
2229 spin_unlock_irqrestore(&list
->lock
, flags
);
2231 EXPORT_SYMBOL(skb_unlink
);
2234 * skb_append - append a buffer
2235 * @old: buffer to insert after
2236 * @newsk: buffer to insert
2237 * @list: list to use
2239 * Place a packet after a given packet in a list. The list locks are taken
2240 * and this function is atomic with respect to other list locked calls.
2241 * A buffer cannot be placed on two lists at the same time.
2243 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2245 unsigned long flags
;
2247 spin_lock_irqsave(&list
->lock
, flags
);
2248 __skb_queue_after(list
, old
, newsk
);
2249 spin_unlock_irqrestore(&list
->lock
, flags
);
2251 EXPORT_SYMBOL(skb_append
);
2254 * skb_insert - insert a buffer
2255 * @old: buffer to insert before
2256 * @newsk: buffer to insert
2257 * @list: list to use
2259 * Place a packet before a given packet in a list. The list locks are
2260 * taken and this function is atomic with respect to other list locked
2263 * A buffer cannot be placed on two lists at the same time.
2265 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2267 unsigned long flags
;
2269 spin_lock_irqsave(&list
->lock
, flags
);
2270 __skb_insert(newsk
, old
->prev
, old
, list
);
2271 spin_unlock_irqrestore(&list
->lock
, flags
);
2273 EXPORT_SYMBOL(skb_insert
);
2275 static inline void skb_split_inside_header(struct sk_buff
*skb
,
2276 struct sk_buff
* skb1
,
2277 const u32 len
, const int pos
)
2281 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
2283 /* And move data appendix as is. */
2284 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
2285 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
2287 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
2288 skb_shinfo(skb
)->nr_frags
= 0;
2289 skb1
->data_len
= skb
->data_len
;
2290 skb1
->len
+= skb1
->data_len
;
2293 skb_set_tail_pointer(skb
, len
);
2296 static inline void skb_split_no_header(struct sk_buff
*skb
,
2297 struct sk_buff
* skb1
,
2298 const u32 len
, int pos
)
2301 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
2303 skb_shinfo(skb
)->nr_frags
= 0;
2304 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
2306 skb
->data_len
= len
- pos
;
2308 for (i
= 0; i
< nfrags
; i
++) {
2309 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
2311 if (pos
+ size
> len
) {
2312 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
2316 * We have two variants in this case:
2317 * 1. Move all the frag to the second
2318 * part, if it is possible. F.e.
2319 * this approach is mandatory for TUX,
2320 * where splitting is expensive.
2321 * 2. Split is accurately. We make this.
2323 skb_frag_ref(skb
, i
);
2324 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
2325 skb_frag_size_sub(&skb_shinfo(skb1
)->frags
[0], len
- pos
);
2326 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
], len
- pos
);
2327 skb_shinfo(skb
)->nr_frags
++;
2331 skb_shinfo(skb
)->nr_frags
++;
2334 skb_shinfo(skb1
)->nr_frags
= k
;
2338 * skb_split - Split fragmented skb to two parts at length len.
2339 * @skb: the buffer to split
2340 * @skb1: the buffer to receive the second part
2341 * @len: new length for skb
2343 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
2345 int pos
= skb_headlen(skb
);
2347 skb_shinfo(skb1
)->tx_flags
= skb_shinfo(skb
)->tx_flags
& SKBTX_SHARED_FRAG
;
2348 if (len
< pos
) /* Split line is inside header. */
2349 skb_split_inside_header(skb
, skb1
, len
, pos
);
2350 else /* Second chunk has no header, nothing to copy. */
2351 skb_split_no_header(skb
, skb1
, len
, pos
);
2353 EXPORT_SYMBOL(skb_split
);
2355 /* Shifting from/to a cloned skb is a no-go.
2357 * Caller cannot keep skb_shinfo related pointers past calling here!
2359 static int skb_prepare_for_shift(struct sk_buff
*skb
)
2361 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2365 * skb_shift - Shifts paged data partially from skb to another
2366 * @tgt: buffer into which tail data gets added
2367 * @skb: buffer from which the paged data comes from
2368 * @shiftlen: shift up to this many bytes
2370 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2371 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2372 * It's up to caller to free skb if everything was shifted.
2374 * If @tgt runs out of frags, the whole operation is aborted.
2376 * Skb cannot include anything else but paged data while tgt is allowed
2377 * to have non-paged data as well.
2379 * TODO: full sized shift could be optimized but that would need
2380 * specialized skb free'er to handle frags without up-to-date nr_frags.
2382 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
2384 int from
, to
, merge
, todo
;
2385 struct skb_frag_struct
*fragfrom
, *fragto
;
2387 BUG_ON(shiftlen
> skb
->len
);
2388 BUG_ON(skb_headlen(skb
)); /* Would corrupt stream */
2392 to
= skb_shinfo(tgt
)->nr_frags
;
2393 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2395 /* Actual merge is delayed until the point when we know we can
2396 * commit all, so that we don't have to undo partial changes
2399 !skb_can_coalesce(tgt
, to
, skb_frag_page(fragfrom
),
2400 fragfrom
->page_offset
)) {
2405 todo
-= skb_frag_size(fragfrom
);
2407 if (skb_prepare_for_shift(skb
) ||
2408 skb_prepare_for_shift(tgt
))
2411 /* All previous frag pointers might be stale! */
2412 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2413 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2415 skb_frag_size_add(fragto
, shiftlen
);
2416 skb_frag_size_sub(fragfrom
, shiftlen
);
2417 fragfrom
->page_offset
+= shiftlen
;
2425 /* Skip full, not-fitting skb to avoid expensive operations */
2426 if ((shiftlen
== skb
->len
) &&
2427 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
2430 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
2433 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
2434 if (to
== MAX_SKB_FRAGS
)
2437 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2438 fragto
= &skb_shinfo(tgt
)->frags
[to
];
2440 if (todo
>= skb_frag_size(fragfrom
)) {
2441 *fragto
= *fragfrom
;
2442 todo
-= skb_frag_size(fragfrom
);
2447 __skb_frag_ref(fragfrom
);
2448 fragto
->page
= fragfrom
->page
;
2449 fragto
->page_offset
= fragfrom
->page_offset
;
2450 skb_frag_size_set(fragto
, todo
);
2452 fragfrom
->page_offset
+= todo
;
2453 skb_frag_size_sub(fragfrom
, todo
);
2461 /* Ready to "commit" this state change to tgt */
2462 skb_shinfo(tgt
)->nr_frags
= to
;
2465 fragfrom
= &skb_shinfo(skb
)->frags
[0];
2466 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2468 skb_frag_size_add(fragto
, skb_frag_size(fragfrom
));
2469 __skb_frag_unref(fragfrom
);
2472 /* Reposition in the original skb */
2474 while (from
< skb_shinfo(skb
)->nr_frags
)
2475 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
2476 skb_shinfo(skb
)->nr_frags
= to
;
2478 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
2481 /* Most likely the tgt won't ever need its checksum anymore, skb on
2482 * the other hand might need it if it needs to be resent
2484 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
2485 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2487 /* Yak, is it really working this way? Some helper please? */
2488 skb
->len
-= shiftlen
;
2489 skb
->data_len
-= shiftlen
;
2490 skb
->truesize
-= shiftlen
;
2491 tgt
->len
+= shiftlen
;
2492 tgt
->data_len
+= shiftlen
;
2493 tgt
->truesize
+= shiftlen
;
2499 * skb_prepare_seq_read - Prepare a sequential read of skb data
2500 * @skb: the buffer to read
2501 * @from: lower offset of data to be read
2502 * @to: upper offset of data to be read
2503 * @st: state variable
2505 * Initializes the specified state variable. Must be called before
2506 * invoking skb_seq_read() for the first time.
2508 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
2509 unsigned int to
, struct skb_seq_state
*st
)
2511 st
->lower_offset
= from
;
2512 st
->upper_offset
= to
;
2513 st
->root_skb
= st
->cur_skb
= skb
;
2514 st
->frag_idx
= st
->stepped_offset
= 0;
2515 st
->frag_data
= NULL
;
2517 EXPORT_SYMBOL(skb_prepare_seq_read
);
2520 * skb_seq_read - Sequentially read skb data
2521 * @consumed: number of bytes consumed by the caller so far
2522 * @data: destination pointer for data to be returned
2523 * @st: state variable
2525 * Reads a block of skb data at &consumed relative to the
2526 * lower offset specified to skb_prepare_seq_read(). Assigns
2527 * the head of the data block to &data and returns the length
2528 * of the block or 0 if the end of the skb data or the upper
2529 * offset has been reached.
2531 * The caller is not required to consume all of the data
2532 * returned, i.e. &consumed is typically set to the number
2533 * of bytes already consumed and the next call to
2534 * skb_seq_read() will return the remaining part of the block.
2536 * Note 1: The size of each block of data returned can be arbitrary,
2537 * this limitation is the cost for zerocopy seqeuental
2538 * reads of potentially non linear data.
2540 * Note 2: Fragment lists within fragments are not implemented
2541 * at the moment, state->root_skb could be replaced with
2542 * a stack for this purpose.
2544 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
2545 struct skb_seq_state
*st
)
2547 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
2550 if (unlikely(abs_offset
>= st
->upper_offset
)) {
2551 if (st
->frag_data
) {
2552 kunmap_atomic(st
->frag_data
);
2553 st
->frag_data
= NULL
;
2559 block_limit
= skb_headlen(st
->cur_skb
) + st
->stepped_offset
;
2561 if (abs_offset
< block_limit
&& !st
->frag_data
) {
2562 *data
= st
->cur_skb
->data
+ (abs_offset
- st
->stepped_offset
);
2563 return block_limit
- abs_offset
;
2566 if (st
->frag_idx
== 0 && !st
->frag_data
)
2567 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
2569 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
2570 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
2571 block_limit
= skb_frag_size(frag
) + st
->stepped_offset
;
2573 if (abs_offset
< block_limit
) {
2575 st
->frag_data
= kmap_atomic(skb_frag_page(frag
));
2577 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
2578 (abs_offset
- st
->stepped_offset
);
2580 return block_limit
- abs_offset
;
2583 if (st
->frag_data
) {
2584 kunmap_atomic(st
->frag_data
);
2585 st
->frag_data
= NULL
;
2589 st
->stepped_offset
+= skb_frag_size(frag
);
2592 if (st
->frag_data
) {
2593 kunmap_atomic(st
->frag_data
);
2594 st
->frag_data
= NULL
;
2597 if (st
->root_skb
== st
->cur_skb
&& skb_has_frag_list(st
->root_skb
)) {
2598 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
2601 } else if (st
->cur_skb
->next
) {
2602 st
->cur_skb
= st
->cur_skb
->next
;
2609 EXPORT_SYMBOL(skb_seq_read
);
2612 * skb_abort_seq_read - Abort a sequential read of skb data
2613 * @st: state variable
2615 * Must be called if skb_seq_read() was not called until it
2618 void skb_abort_seq_read(struct skb_seq_state
*st
)
2621 kunmap_atomic(st
->frag_data
);
2623 EXPORT_SYMBOL(skb_abort_seq_read
);
2625 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2627 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
2628 struct ts_config
*conf
,
2629 struct ts_state
*state
)
2631 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
2634 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
2636 skb_abort_seq_read(TS_SKB_CB(state
));
2640 * skb_find_text - Find a text pattern in skb data
2641 * @skb: the buffer to look in
2642 * @from: search offset
2644 * @config: textsearch configuration
2645 * @state: uninitialized textsearch state variable
2647 * Finds a pattern in the skb data according to the specified
2648 * textsearch configuration. Use textsearch_next() to retrieve
2649 * subsequent occurrences of the pattern. Returns the offset
2650 * to the first occurrence or UINT_MAX if no match was found.
2652 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
2653 unsigned int to
, struct ts_config
*config
,
2654 struct ts_state
*state
)
2658 config
->get_next_block
= skb_ts_get_next_block
;
2659 config
->finish
= skb_ts_finish
;
2661 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
2663 ret
= textsearch_find(config
, state
);
2664 return (ret
<= to
- from
? ret
: UINT_MAX
);
2666 EXPORT_SYMBOL(skb_find_text
);
2669 * skb_append_datato_frags - append the user data to a skb
2670 * @sk: sock structure
2671 * @skb: skb structure to be appened with user data.
2672 * @getfrag: call back function to be used for getting the user data
2673 * @from: pointer to user message iov
2674 * @length: length of the iov message
2676 * Description: This procedure append the user data in the fragment part
2677 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2679 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
2680 int (*getfrag
)(void *from
, char *to
, int offset
,
2681 int len
, int odd
, struct sk_buff
*skb
),
2682 void *from
, int length
)
2684 int frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2688 struct page_frag
*pfrag
= ¤t
->task_frag
;
2691 /* Return error if we don't have space for new frag */
2692 if (frg_cnt
>= MAX_SKB_FRAGS
)
2695 if (!sk_page_frag_refill(sk
, pfrag
))
2698 /* copy the user data to page */
2699 copy
= min_t(int, length
, pfrag
->size
- pfrag
->offset
);
2701 ret
= getfrag(from
, page_address(pfrag
->page
) + pfrag
->offset
,
2702 offset
, copy
, 0, skb
);
2706 /* copy was successful so update the size parameters */
2707 skb_fill_page_desc(skb
, frg_cnt
, pfrag
->page
, pfrag
->offset
,
2710 pfrag
->offset
+= copy
;
2711 get_page(pfrag
->page
);
2713 skb
->truesize
+= copy
;
2714 atomic_add(copy
, &sk
->sk_wmem_alloc
);
2716 skb
->data_len
+= copy
;
2720 } while (length
> 0);
2724 EXPORT_SYMBOL(skb_append_datato_frags
);
2727 * skb_pull_rcsum - pull skb and update receive checksum
2728 * @skb: buffer to update
2729 * @len: length of data pulled
2731 * This function performs an skb_pull on the packet and updates
2732 * the CHECKSUM_COMPLETE checksum. It should be used on
2733 * receive path processing instead of skb_pull unless you know
2734 * that the checksum difference is zero (e.g., a valid IP header)
2735 * or you are setting ip_summed to CHECKSUM_NONE.
2737 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
2739 BUG_ON(len
> skb
->len
);
2741 BUG_ON(skb
->len
< skb
->data_len
);
2742 skb_postpull_rcsum(skb
, skb
->data
, len
);
2743 return skb
->data
+= len
;
2745 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
2748 * skb_segment - Perform protocol segmentation on skb.
2749 * @skb: buffer to segment
2750 * @features: features for the output path (see dev->features)
2752 * This function performs segmentation on the given skb. It returns
2753 * a pointer to the first in a list of new skbs for the segments.
2754 * In case of error it returns ERR_PTR(err).
2756 struct sk_buff
*skb_segment(struct sk_buff
*skb
, netdev_features_t features
)
2758 struct sk_buff
*segs
= NULL
;
2759 struct sk_buff
*tail
= NULL
;
2760 struct sk_buff
*fskb
= skb_shinfo(skb
)->frag_list
;
2761 unsigned int mss
= skb_shinfo(skb
)->gso_size
;
2762 unsigned int doffset
= skb
->data
- skb_mac_header(skb
);
2763 unsigned int offset
= doffset
;
2764 unsigned int tnl_hlen
= skb_tnl_header_len(skb
);
2765 unsigned int headroom
;
2769 int sg
= !!(features
& NETIF_F_SG
);
2770 int nfrags
= skb_shinfo(skb
)->nr_frags
;
2775 proto
= skb_network_protocol(skb
);
2776 if (unlikely(!proto
))
2777 return ERR_PTR(-EINVAL
);
2779 csum
= !!can_checksum_protocol(features
, proto
);
2780 __skb_push(skb
, doffset
);
2781 headroom
= skb_headroom(skb
);
2782 pos
= skb_headlen(skb
);
2785 struct sk_buff
*nskb
;
2790 len
= skb
->len
- offset
;
2794 hsize
= skb_headlen(skb
) - offset
;
2797 if (hsize
> len
|| !sg
)
2800 if (!hsize
&& i
>= nfrags
) {
2801 BUG_ON(fskb
->len
!= len
);
2804 nskb
= skb_clone(fskb
, GFP_ATOMIC
);
2807 if (unlikely(!nskb
))
2810 hsize
= skb_end_offset(nskb
);
2811 if (skb_cow_head(nskb
, doffset
+ headroom
)) {
2816 nskb
->truesize
+= skb_end_offset(nskb
) - hsize
;
2817 skb_release_head_state(nskb
);
2818 __skb_push(nskb
, doffset
);
2820 nskb
= __alloc_skb(hsize
+ doffset
+ headroom
,
2821 GFP_ATOMIC
, skb_alloc_rx_flag(skb
),
2824 if (unlikely(!nskb
))
2827 skb_reserve(nskb
, headroom
);
2828 __skb_put(nskb
, doffset
);
2837 __copy_skb_header(nskb
, skb
);
2838 nskb
->mac_len
= skb
->mac_len
;
2840 /* nskb and skb might have different headroom */
2841 if (nskb
->ip_summed
== CHECKSUM_PARTIAL
)
2842 nskb
->csum_start
+= skb_headroom(nskb
) - headroom
;
2844 skb_reset_mac_header(nskb
);
2845 skb_set_network_header(nskb
, skb
->mac_len
);
2846 nskb
->transport_header
= (nskb
->network_header
+
2847 skb_network_header_len(skb
));
2849 skb_copy_from_linear_data_offset(skb
, -tnl_hlen
,
2850 nskb
->data
- tnl_hlen
,
2851 doffset
+ tnl_hlen
);
2853 if (fskb
!= skb_shinfo(skb
)->frag_list
)
2854 goto perform_csum_check
;
2857 nskb
->ip_summed
= CHECKSUM_NONE
;
2858 nskb
->csum
= skb_copy_and_csum_bits(skb
, offset
,
2864 frag
= skb_shinfo(nskb
)->frags
;
2866 skb_copy_from_linear_data_offset(skb
, offset
,
2867 skb_put(nskb
, hsize
), hsize
);
2869 skb_shinfo(nskb
)->tx_flags
= skb_shinfo(skb
)->tx_flags
& SKBTX_SHARED_FRAG
;
2871 while (pos
< offset
+ len
&& i
< nfrags
) {
2872 *frag
= skb_shinfo(skb
)->frags
[i
];
2873 __skb_frag_ref(frag
);
2874 size
= skb_frag_size(frag
);
2877 frag
->page_offset
+= offset
- pos
;
2878 skb_frag_size_sub(frag
, offset
- pos
);
2881 skb_shinfo(nskb
)->nr_frags
++;
2883 if (pos
+ size
<= offset
+ len
) {
2887 skb_frag_size_sub(frag
, pos
+ size
- (offset
+ len
));
2894 if (pos
< offset
+ len
) {
2895 struct sk_buff
*fskb2
= fskb
;
2897 BUG_ON(pos
+ fskb
->len
!= offset
+ len
);
2903 fskb2
= skb_clone(fskb2
, GFP_ATOMIC
);
2909 SKB_FRAG_ASSERT(nskb
);
2910 skb_shinfo(nskb
)->frag_list
= fskb2
;
2914 nskb
->data_len
= len
- hsize
;
2915 nskb
->len
+= nskb
->data_len
;
2916 nskb
->truesize
+= nskb
->data_len
;
2920 nskb
->csum
= skb_checksum(nskb
, doffset
,
2921 nskb
->len
- doffset
, 0);
2922 nskb
->ip_summed
= CHECKSUM_NONE
;
2924 } while ((offset
+= len
) < skb
->len
);
2929 while ((skb
= segs
)) {
2933 return ERR_PTR(err
);
2935 EXPORT_SYMBOL_GPL(skb_segment
);
2937 int skb_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2939 struct sk_buff
*p
= *head
;
2940 struct sk_buff
*nskb
;
2941 struct skb_shared_info
*skbinfo
= skb_shinfo(skb
);
2942 struct skb_shared_info
*pinfo
= skb_shinfo(p
);
2943 unsigned int headroom
;
2944 unsigned int len
= skb_gro_len(skb
);
2945 unsigned int offset
= skb_gro_offset(skb
);
2946 unsigned int headlen
= skb_headlen(skb
);
2947 unsigned int delta_truesize
;
2949 if (p
->len
+ len
>= 65536)
2952 if (pinfo
->frag_list
)
2954 else if (headlen
<= offset
) {
2957 int i
= skbinfo
->nr_frags
;
2958 int nr_frags
= pinfo
->nr_frags
+ i
;
2962 if (nr_frags
> MAX_SKB_FRAGS
)
2965 pinfo
->nr_frags
= nr_frags
;
2966 skbinfo
->nr_frags
= 0;
2968 frag
= pinfo
->frags
+ nr_frags
;
2969 frag2
= skbinfo
->frags
+ i
;
2974 frag
->page_offset
+= offset
;
2975 skb_frag_size_sub(frag
, offset
);
2977 /* all fragments truesize : remove (head size + sk_buff) */
2978 delta_truesize
= skb
->truesize
-
2979 SKB_TRUESIZE(skb_end_offset(skb
));
2981 skb
->truesize
-= skb
->data_len
;
2982 skb
->len
-= skb
->data_len
;
2985 NAPI_GRO_CB(skb
)->free
= NAPI_GRO_FREE
;
2987 } else if (skb
->head_frag
) {
2988 int nr_frags
= pinfo
->nr_frags
;
2989 skb_frag_t
*frag
= pinfo
->frags
+ nr_frags
;
2990 struct page
*page
= virt_to_head_page(skb
->head
);
2991 unsigned int first_size
= headlen
- offset
;
2992 unsigned int first_offset
;
2994 if (nr_frags
+ 1 + skbinfo
->nr_frags
> MAX_SKB_FRAGS
)
2997 first_offset
= skb
->data
-
2998 (unsigned char *)page_address(page
) +
3001 pinfo
->nr_frags
= nr_frags
+ 1 + skbinfo
->nr_frags
;
3003 frag
->page
.p
= page
;
3004 frag
->page_offset
= first_offset
;
3005 skb_frag_size_set(frag
, first_size
);
3007 memcpy(frag
+ 1, skbinfo
->frags
, sizeof(*frag
) * skbinfo
->nr_frags
);
3008 /* We dont need to clear skbinfo->nr_frags here */
3010 delta_truesize
= skb
->truesize
- SKB_DATA_ALIGN(sizeof(struct sk_buff
));
3011 NAPI_GRO_CB(skb
)->free
= NAPI_GRO_FREE_STOLEN_HEAD
;
3013 } else if (skb_gro_len(p
) != pinfo
->gso_size
)
3016 headroom
= skb_headroom(p
);
3017 nskb
= alloc_skb(headroom
+ skb_gro_offset(p
), GFP_ATOMIC
);
3018 if (unlikely(!nskb
))
3021 __copy_skb_header(nskb
, p
);
3022 nskb
->mac_len
= p
->mac_len
;
3024 skb_reserve(nskb
, headroom
);
3025 __skb_put(nskb
, skb_gro_offset(p
));
3027 skb_set_mac_header(nskb
, skb_mac_header(p
) - p
->data
);
3028 skb_set_network_header(nskb
, skb_network_offset(p
));
3029 skb_set_transport_header(nskb
, skb_transport_offset(p
));
3031 __skb_pull(p
, skb_gro_offset(p
));
3032 memcpy(skb_mac_header(nskb
), skb_mac_header(p
),
3033 p
->data
- skb_mac_header(p
));
3035 skb_shinfo(nskb
)->frag_list
= p
;
3036 skb_shinfo(nskb
)->gso_size
= pinfo
->gso_size
;
3037 pinfo
->gso_size
= 0;
3038 skb_header_release(p
);
3039 NAPI_GRO_CB(nskb
)->last
= p
;
3041 nskb
->data_len
+= p
->len
;
3042 nskb
->truesize
+= p
->truesize
;
3043 nskb
->len
+= p
->len
;
3046 nskb
->next
= p
->next
;
3052 delta_truesize
= skb
->truesize
;
3053 if (offset
> headlen
) {
3054 unsigned int eat
= offset
- headlen
;
3056 skbinfo
->frags
[0].page_offset
+= eat
;
3057 skb_frag_size_sub(&skbinfo
->frags
[0], eat
);
3058 skb
->data_len
-= eat
;
3063 __skb_pull(skb
, offset
);
3065 NAPI_GRO_CB(p
)->last
->next
= skb
;
3066 NAPI_GRO_CB(p
)->last
= skb
;
3067 skb_header_release(skb
);
3070 NAPI_GRO_CB(p
)->count
++;
3072 p
->truesize
+= delta_truesize
;
3075 NAPI_GRO_CB(skb
)->same_flow
= 1;
3078 EXPORT_SYMBOL_GPL(skb_gro_receive
);
3080 void __init
skb_init(void)
3082 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
3083 sizeof(struct sk_buff
),
3085 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
3087 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
3088 (2*sizeof(struct sk_buff
)) +
3091 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
3096 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3097 * @skb: Socket buffer containing the buffers to be mapped
3098 * @sg: The scatter-gather list to map into
3099 * @offset: The offset into the buffer's contents to start mapping
3100 * @len: Length of buffer space to be mapped
3102 * Fill the specified scatter-gather list with mappings/pointers into a
3103 * region of the buffer space attached to a socket buffer.
3106 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
3108 int start
= skb_headlen(skb
);
3109 int i
, copy
= start
- offset
;
3110 struct sk_buff
*frag_iter
;
3116 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
3118 if ((len
-= copy
) == 0)
3123 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
3126 WARN_ON(start
> offset
+ len
);
3128 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
3129 if ((copy
= end
- offset
) > 0) {
3130 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
3134 sg_set_page(&sg
[elt
], skb_frag_page(frag
), copy
,
3135 frag
->page_offset
+offset
-start
);
3144 skb_walk_frags(skb
, frag_iter
) {
3147 WARN_ON(start
> offset
+ len
);
3149 end
= start
+ frag_iter
->len
;
3150 if ((copy
= end
- offset
) > 0) {
3153 elt
+= __skb_to_sgvec(frag_iter
, sg
+elt
, offset
- start
,
3155 if ((len
-= copy
) == 0)
3165 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
3167 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
);
3169 sg_mark_end(&sg
[nsg
- 1]);
3173 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
3176 * skb_cow_data - Check that a socket buffer's data buffers are writable
3177 * @skb: The socket buffer to check.
3178 * @tailbits: Amount of trailing space to be added
3179 * @trailer: Returned pointer to the skb where the @tailbits space begins
3181 * Make sure that the data buffers attached to a socket buffer are
3182 * writable. If they are not, private copies are made of the data buffers
3183 * and the socket buffer is set to use these instead.
3185 * If @tailbits is given, make sure that there is space to write @tailbits
3186 * bytes of data beyond current end of socket buffer. @trailer will be
3187 * set to point to the skb in which this space begins.
3189 * The number of scatterlist elements required to completely map the
3190 * COW'd and extended socket buffer will be returned.
3192 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
3196 struct sk_buff
*skb1
, **skb_p
;
3198 /* If skb is cloned or its head is paged, reallocate
3199 * head pulling out all the pages (pages are considered not writable
3200 * at the moment even if they are anonymous).
3202 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
3203 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
3206 /* Easy case. Most of packets will go this way. */
3207 if (!skb_has_frag_list(skb
)) {
3208 /* A little of trouble, not enough of space for trailer.
3209 * This should not happen, when stack is tuned to generate
3210 * good frames. OK, on miss we reallocate and reserve even more
3211 * space, 128 bytes is fair. */
3213 if (skb_tailroom(skb
) < tailbits
&&
3214 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
3222 /* Misery. We are in troubles, going to mincer fragments... */
3225 skb_p
= &skb_shinfo(skb
)->frag_list
;
3228 while ((skb1
= *skb_p
) != NULL
) {
3231 /* The fragment is partially pulled by someone,
3232 * this can happen on input. Copy it and everything
3235 if (skb_shared(skb1
))
3238 /* If the skb is the last, worry about trailer. */
3240 if (skb1
->next
== NULL
&& tailbits
) {
3241 if (skb_shinfo(skb1
)->nr_frags
||
3242 skb_has_frag_list(skb1
) ||
3243 skb_tailroom(skb1
) < tailbits
)
3244 ntail
= tailbits
+ 128;
3250 skb_shinfo(skb1
)->nr_frags
||
3251 skb_has_frag_list(skb1
)) {
3252 struct sk_buff
*skb2
;
3254 /* Fuck, we are miserable poor guys... */
3256 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
3258 skb2
= skb_copy_expand(skb1
,
3262 if (unlikely(skb2
== NULL
))
3266 skb_set_owner_w(skb2
, skb1
->sk
);
3268 /* Looking around. Are we still alive?
3269 * OK, link new skb, drop old one */
3271 skb2
->next
= skb1
->next
;
3278 skb_p
= &skb1
->next
;
3283 EXPORT_SYMBOL_GPL(skb_cow_data
);
3285 static void sock_rmem_free(struct sk_buff
*skb
)
3287 struct sock
*sk
= skb
->sk
;
3289 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
3293 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3295 int sock_queue_err_skb(struct sock
*sk
, struct sk_buff
*skb
)
3299 if (atomic_read(&sk
->sk_rmem_alloc
) + skb
->truesize
>=
3300 (unsigned int)sk
->sk_rcvbuf
)
3305 skb
->destructor
= sock_rmem_free
;
3306 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
3308 /* before exiting rcu section, make sure dst is refcounted */
3311 skb_queue_tail(&sk
->sk_error_queue
, skb
);
3312 if (!sock_flag(sk
, SOCK_DEAD
))
3313 sk
->sk_data_ready(sk
, len
);
3316 EXPORT_SYMBOL(sock_queue_err_skb
);
3318 void skb_tstamp_tx(struct sk_buff
*orig_skb
,
3319 struct skb_shared_hwtstamps
*hwtstamps
)
3321 struct sock
*sk
= orig_skb
->sk
;
3322 struct sock_exterr_skb
*serr
;
3323 struct sk_buff
*skb
;
3330 *skb_hwtstamps(orig_skb
) =
3334 * no hardware time stamps available,
3335 * so keep the shared tx_flags and only
3336 * store software time stamp
3338 orig_skb
->tstamp
= ktime_get_real();
3341 skb
= skb_clone(orig_skb
, GFP_ATOMIC
);
3345 serr
= SKB_EXT_ERR(skb
);
3346 memset(serr
, 0, sizeof(*serr
));
3347 serr
->ee
.ee_errno
= ENOMSG
;
3348 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TIMESTAMPING
;
3350 err
= sock_queue_err_skb(sk
, skb
);
3355 EXPORT_SYMBOL_GPL(skb_tstamp_tx
);
3357 void skb_complete_wifi_ack(struct sk_buff
*skb
, bool acked
)
3359 struct sock
*sk
= skb
->sk
;
3360 struct sock_exterr_skb
*serr
;
3363 skb
->wifi_acked_valid
= 1;
3364 skb
->wifi_acked
= acked
;
3366 serr
= SKB_EXT_ERR(skb
);
3367 memset(serr
, 0, sizeof(*serr
));
3368 serr
->ee
.ee_errno
= ENOMSG
;
3369 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TXSTATUS
;
3371 err
= sock_queue_err_skb(sk
, skb
);
3375 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack
);
3379 * skb_partial_csum_set - set up and verify partial csum values for packet
3380 * @skb: the skb to set
3381 * @start: the number of bytes after skb->data to start checksumming.
3382 * @off: the offset from start to place the checksum.
3384 * For untrusted partially-checksummed packets, we need to make sure the values
3385 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3387 * This function checks and sets those values and skb->ip_summed: if this
3388 * returns false you should drop the packet.
3390 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
3392 if (unlikely(start
> skb_headlen(skb
)) ||
3393 unlikely((int)start
+ off
> skb_headlen(skb
) - 2)) {
3394 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3395 start
, off
, skb_headlen(skb
));
3398 skb
->ip_summed
= CHECKSUM_PARTIAL
;
3399 skb
->csum_start
= skb_headroom(skb
) + start
;
3400 skb
->csum_offset
= off
;
3401 skb_set_transport_header(skb
, start
);
3404 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);
3406 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
3408 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
3411 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);
3413 void kfree_skb_partial(struct sk_buff
*skb
, bool head_stolen
)
3416 skb_release_head_state(skb
);
3417 kmem_cache_free(skbuff_head_cache
, skb
);
3422 EXPORT_SYMBOL(kfree_skb_partial
);
3425 * skb_try_coalesce - try to merge skb to prior one
3427 * @from: buffer to add
3428 * @fragstolen: pointer to boolean
3429 * @delta_truesize: how much more was allocated than was requested
3431 bool skb_try_coalesce(struct sk_buff
*to
, struct sk_buff
*from
,
3432 bool *fragstolen
, int *delta_truesize
)
3434 int i
, delta
, len
= from
->len
;
3436 *fragstolen
= false;
3441 if (len
<= skb_tailroom(to
)) {
3442 BUG_ON(skb_copy_bits(from
, 0, skb_put(to
, len
), len
));
3443 *delta_truesize
= 0;
3447 if (skb_has_frag_list(to
) || skb_has_frag_list(from
))
3450 if (skb_headlen(from
) != 0) {
3452 unsigned int offset
;
3454 if (skb_shinfo(to
)->nr_frags
+
3455 skb_shinfo(from
)->nr_frags
>= MAX_SKB_FRAGS
)
3458 if (skb_head_is_locked(from
))
3461 delta
= from
->truesize
- SKB_DATA_ALIGN(sizeof(struct sk_buff
));
3463 page
= virt_to_head_page(from
->head
);
3464 offset
= from
->data
- (unsigned char *)page_address(page
);
3466 skb_fill_page_desc(to
, skb_shinfo(to
)->nr_frags
,
3467 page
, offset
, skb_headlen(from
));
3470 if (skb_shinfo(to
)->nr_frags
+
3471 skb_shinfo(from
)->nr_frags
> MAX_SKB_FRAGS
)
3474 delta
= from
->truesize
- SKB_TRUESIZE(skb_end_offset(from
));
3477 WARN_ON_ONCE(delta
< len
);
3479 memcpy(skb_shinfo(to
)->frags
+ skb_shinfo(to
)->nr_frags
,
3480 skb_shinfo(from
)->frags
,
3481 skb_shinfo(from
)->nr_frags
* sizeof(skb_frag_t
));
3482 skb_shinfo(to
)->nr_frags
+= skb_shinfo(from
)->nr_frags
;
3484 if (!skb_cloned(from
))
3485 skb_shinfo(from
)->nr_frags
= 0;
3487 /* if the skb is not cloned this does nothing
3488 * since we set nr_frags to 0.
3490 for (i
= 0; i
< skb_shinfo(from
)->nr_frags
; i
++)
3491 skb_frag_ref(from
, i
);
3493 to
->truesize
+= delta
;
3495 to
->data_len
+= len
;
3497 *delta_truesize
= delta
;
3500 EXPORT_SYMBOL(skb_try_coalesce
);
3503 * skb_scrub_packet - scrub an skb before sending it to another netns
3505 * @skb: buffer to clean
3507 * skb_scrub_packet can be used to clean an skb before injecting it in
3508 * another namespace. We have to clear all information in the skb that
3509 * could impact namespace isolation.
3511 void skb_scrub_packet(struct sk_buff
*skb
)
3514 skb
->tstamp
.tv64
= 0;
3515 skb
->pkt_type
= PACKET_HOST
;
3521 nf_reset_trace(skb
);
3523 EXPORT_SYMBOL_GPL(skb_scrub_packet
);