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>
45 #include <linux/interrupt.h>
47 #include <linux/inet.h>
48 #include <linux/slab.h>
49 #include <linux/tcp.h>
50 #include <linux/udp.h>
51 #include <linux/sctp.h>
52 #include <linux/netdevice.h>
53 #ifdef CONFIG_NET_CLS_ACT
54 #include <net/pkt_sched.h>
56 #include <linux/string.h>
57 #include <linux/skbuff.h>
58 #include <linux/splice.h>
59 #include <linux/cache.h>
60 #include <linux/rtnetlink.h>
61 #include <linux/init.h>
62 #include <linux/scatterlist.h>
63 #include <linux/errqueue.h>
64 #include <linux/prefetch.h>
65 #include <linux/if_vlan.h>
67 #include <net/protocol.h>
70 #include <net/checksum.h>
71 #include <net/ip6_checksum.h>
74 #include <linux/uaccess.h>
75 #include <trace/events/skb.h>
76 #include <linux/highmem.h>
77 #include <linux/capability.h>
78 #include <linux/user_namespace.h>
80 struct kmem_cache
*skbuff_head_cache __ro_after_init
;
81 static struct kmem_cache
*skbuff_fclone_cache __ro_after_init
;
82 #ifdef CONFIG_SKB_EXTENSIONS
83 static struct kmem_cache
*skbuff_ext_cache __ro_after_init
;
85 int sysctl_max_skb_frags __read_mostly
= MAX_SKB_FRAGS
;
86 EXPORT_SYMBOL(sysctl_max_skb_frags
);
89 * skb_panic - private function for out-of-line support
93 * @msg: skb_over_panic or skb_under_panic
95 * Out-of-line support for skb_put() and skb_push().
96 * Called via the wrapper skb_over_panic() or skb_under_panic().
97 * Keep out of line to prevent kernel bloat.
98 * __builtin_return_address is not used because it is not always reliable.
100 static void skb_panic(struct sk_buff
*skb
, unsigned int sz
, void *addr
,
103 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
104 msg
, addr
, skb
->len
, sz
, skb
->head
, skb
->data
,
105 (unsigned long)skb
->tail
, (unsigned long)skb
->end
,
106 skb
->dev
? skb
->dev
->name
: "<NULL>");
110 static void skb_over_panic(struct sk_buff
*skb
, unsigned int sz
, void *addr
)
112 skb_panic(skb
, sz
, addr
, __func__
);
115 static void skb_under_panic(struct sk_buff
*skb
, unsigned int sz
, void *addr
)
117 skb_panic(skb
, sz
, addr
, __func__
);
121 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
122 * the caller if emergency pfmemalloc reserves are being used. If it is and
123 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
124 * may be used. Otherwise, the packet data may be discarded until enough
127 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
128 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
130 static void *__kmalloc_reserve(size_t size
, gfp_t flags
, int node
,
131 unsigned long ip
, bool *pfmemalloc
)
134 bool ret_pfmemalloc
= false;
137 * Try a regular allocation, when that fails and we're not entitled
138 * to the reserves, fail.
140 obj
= kmalloc_node_track_caller(size
,
141 flags
| __GFP_NOMEMALLOC
| __GFP_NOWARN
,
143 if (obj
|| !(gfp_pfmemalloc_allowed(flags
)))
146 /* Try again but now we are using pfmemalloc reserves */
147 ret_pfmemalloc
= true;
148 obj
= kmalloc_node_track_caller(size
, flags
, node
);
152 *pfmemalloc
= ret_pfmemalloc
;
157 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
158 * 'private' fields and also do memory statistics to find all the
164 * __alloc_skb - allocate a network buffer
165 * @size: size to allocate
166 * @gfp_mask: allocation mask
167 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
168 * instead of head cache and allocate a cloned (child) skb.
169 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
170 * allocations in case the data is required for writeback
171 * @node: numa node to allocate memory on
173 * Allocate a new &sk_buff. The returned buffer has no headroom and a
174 * tail room of at least size bytes. The object has a reference count
175 * of one. The return is the buffer. On a failure the return is %NULL.
177 * Buffers may only be allocated from interrupts using a @gfp_mask of
180 struct sk_buff
*__alloc_skb(unsigned int size
, gfp_t gfp_mask
,
183 struct kmem_cache
*cache
;
184 struct skb_shared_info
*shinfo
;
189 cache
= (flags
& SKB_ALLOC_FCLONE
)
190 ? skbuff_fclone_cache
: skbuff_head_cache
;
192 if (sk_memalloc_socks() && (flags
& SKB_ALLOC_RX
))
193 gfp_mask
|= __GFP_MEMALLOC
;
196 skb
= kmem_cache_alloc_node(cache
, gfp_mask
& ~__GFP_DMA
, node
);
201 /* We do our best to align skb_shared_info on a separate cache
202 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
203 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
204 * Both skb->head and skb_shared_info are cache line aligned.
206 size
= SKB_DATA_ALIGN(size
);
207 size
+= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
208 data
= kmalloc_reserve(size
, gfp_mask
, node
, &pfmemalloc
);
211 /* kmalloc(size) might give us more room than requested.
212 * Put skb_shared_info exactly at the end of allocated zone,
213 * to allow max possible filling before reallocation.
215 size
= SKB_WITH_OVERHEAD(ksize(data
));
216 prefetchw(data
+ size
);
219 * Only clear those fields we need to clear, not those that we will
220 * actually initialise below. Hence, don't put any more fields after
221 * the tail pointer in struct sk_buff!
223 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
224 /* Account for allocated memory : skb + skb->head */
225 skb
->truesize
= SKB_TRUESIZE(size
);
226 skb
->pfmemalloc
= pfmemalloc
;
227 refcount_set(&skb
->users
, 1);
230 skb_reset_tail_pointer(skb
);
231 skb
->end
= skb
->tail
+ size
;
232 skb
->mac_header
= (typeof(skb
->mac_header
))~0U;
233 skb
->transport_header
= (typeof(skb
->transport_header
))~0U;
235 /* make sure we initialize shinfo sequentially */
236 shinfo
= skb_shinfo(skb
);
237 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
238 atomic_set(&shinfo
->dataref
, 1);
240 if (flags
& SKB_ALLOC_FCLONE
) {
241 struct sk_buff_fclones
*fclones
;
243 fclones
= container_of(skb
, struct sk_buff_fclones
, skb1
);
245 skb
->fclone
= SKB_FCLONE_ORIG
;
246 refcount_set(&fclones
->fclone_ref
, 1);
248 fclones
->skb2
.fclone
= SKB_FCLONE_CLONE
;
253 kmem_cache_free(cache
, skb
);
257 EXPORT_SYMBOL(__alloc_skb
);
260 * __build_skb - build a network buffer
261 * @data: data buffer provided by caller
262 * @frag_size: size of data, or 0 if head was kmalloced
264 * Allocate a new &sk_buff. Caller provides space holding head and
265 * skb_shared_info. @data must have been allocated by kmalloc() only if
266 * @frag_size is 0, otherwise data should come from the page allocator
268 * The return is the new skb buffer.
269 * On a failure the return is %NULL, and @data is not freed.
271 * Before IO, driver allocates only data buffer where NIC put incoming frame
272 * Driver should add room at head (NET_SKB_PAD) and
273 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
274 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
275 * before giving packet to stack.
276 * RX rings only contains data buffers, not full skbs.
278 struct sk_buff
*__build_skb(void *data
, unsigned int frag_size
)
280 struct skb_shared_info
*shinfo
;
282 unsigned int size
= frag_size
? : ksize(data
);
284 skb
= kmem_cache_alloc(skbuff_head_cache
, GFP_ATOMIC
);
288 size
-= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
290 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
291 skb
->truesize
= SKB_TRUESIZE(size
);
292 refcount_set(&skb
->users
, 1);
295 skb_reset_tail_pointer(skb
);
296 skb
->end
= skb
->tail
+ size
;
297 skb
->mac_header
= (typeof(skb
->mac_header
))~0U;
298 skb
->transport_header
= (typeof(skb
->transport_header
))~0U;
300 /* make sure we initialize shinfo sequentially */
301 shinfo
= skb_shinfo(skb
);
302 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
303 atomic_set(&shinfo
->dataref
, 1);
308 /* build_skb() is wrapper over __build_skb(), that specifically
309 * takes care of skb->head and skb->pfmemalloc
310 * This means that if @frag_size is not zero, then @data must be backed
311 * by a page fragment, not kmalloc() or vmalloc()
313 struct sk_buff
*build_skb(void *data
, unsigned int frag_size
)
315 struct sk_buff
*skb
= __build_skb(data
, frag_size
);
317 if (skb
&& frag_size
) {
319 if (page_is_pfmemalloc(virt_to_head_page(data
)))
324 EXPORT_SYMBOL(build_skb
);
326 #define NAPI_SKB_CACHE_SIZE 64
328 struct napi_alloc_cache
{
329 struct page_frag_cache page
;
330 unsigned int skb_count
;
331 void *skb_cache
[NAPI_SKB_CACHE_SIZE
];
334 static DEFINE_PER_CPU(struct page_frag_cache
, netdev_alloc_cache
);
335 static DEFINE_PER_CPU(struct napi_alloc_cache
, napi_alloc_cache
);
337 static void *__netdev_alloc_frag(unsigned int fragsz
, gfp_t gfp_mask
)
339 struct page_frag_cache
*nc
;
343 local_irq_save(flags
);
344 nc
= this_cpu_ptr(&netdev_alloc_cache
);
345 data
= page_frag_alloc(nc
, fragsz
, gfp_mask
);
346 local_irq_restore(flags
);
351 * netdev_alloc_frag - allocate a page fragment
352 * @fragsz: fragment size
354 * Allocates a frag from a page for receive buffer.
355 * Uses GFP_ATOMIC allocations.
357 void *netdev_alloc_frag(unsigned int fragsz
)
359 return __netdev_alloc_frag(fragsz
, GFP_ATOMIC
);
361 EXPORT_SYMBOL(netdev_alloc_frag
);
363 static void *__napi_alloc_frag(unsigned int fragsz
, gfp_t gfp_mask
)
365 struct napi_alloc_cache
*nc
= this_cpu_ptr(&napi_alloc_cache
);
367 return page_frag_alloc(&nc
->page
, fragsz
, gfp_mask
);
370 void *napi_alloc_frag(unsigned int fragsz
)
372 return __napi_alloc_frag(fragsz
, GFP_ATOMIC
);
374 EXPORT_SYMBOL(napi_alloc_frag
);
377 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
378 * @dev: network device to receive on
379 * @len: length to allocate
380 * @gfp_mask: get_free_pages mask, passed to alloc_skb
382 * Allocate a new &sk_buff and assign it a usage count of one. The
383 * buffer has NET_SKB_PAD headroom built in. Users should allocate
384 * the headroom they think they need without accounting for the
385 * built in space. The built in space is used for optimisations.
387 * %NULL is returned if there is no free memory.
389 struct sk_buff
*__netdev_alloc_skb(struct net_device
*dev
, unsigned int len
,
392 struct page_frag_cache
*nc
;
400 if ((len
> SKB_WITH_OVERHEAD(PAGE_SIZE
)) ||
401 (gfp_mask
& (__GFP_DIRECT_RECLAIM
| GFP_DMA
))) {
402 skb
= __alloc_skb(len
, gfp_mask
, SKB_ALLOC_RX
, NUMA_NO_NODE
);
408 len
+= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
409 len
= SKB_DATA_ALIGN(len
);
411 if (sk_memalloc_socks())
412 gfp_mask
|= __GFP_MEMALLOC
;
414 local_irq_save(flags
);
416 nc
= this_cpu_ptr(&netdev_alloc_cache
);
417 data
= page_frag_alloc(nc
, len
, gfp_mask
);
418 pfmemalloc
= nc
->pfmemalloc
;
420 local_irq_restore(flags
);
425 skb
= __build_skb(data
, len
);
426 if (unlikely(!skb
)) {
431 /* use OR instead of assignment to avoid clearing of bits in mask */
437 skb_reserve(skb
, NET_SKB_PAD
);
443 EXPORT_SYMBOL(__netdev_alloc_skb
);
446 * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
447 * @napi: napi instance this buffer was allocated for
448 * @len: length to allocate
449 * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
451 * Allocate a new sk_buff for use in NAPI receive. This buffer will
452 * attempt to allocate the head from a special reserved region used
453 * only for NAPI Rx allocation. By doing this we can save several
454 * CPU cycles by avoiding having to disable and re-enable IRQs.
456 * %NULL is returned if there is no free memory.
458 struct sk_buff
*__napi_alloc_skb(struct napi_struct
*napi
, unsigned int len
,
461 struct napi_alloc_cache
*nc
= this_cpu_ptr(&napi_alloc_cache
);
465 len
+= NET_SKB_PAD
+ NET_IP_ALIGN
;
467 if ((len
> SKB_WITH_OVERHEAD(PAGE_SIZE
)) ||
468 (gfp_mask
& (__GFP_DIRECT_RECLAIM
| GFP_DMA
))) {
469 skb
= __alloc_skb(len
, gfp_mask
, SKB_ALLOC_RX
, NUMA_NO_NODE
);
475 len
+= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
476 len
= SKB_DATA_ALIGN(len
);
478 if (sk_memalloc_socks())
479 gfp_mask
|= __GFP_MEMALLOC
;
481 data
= page_frag_alloc(&nc
->page
, len
, gfp_mask
);
485 skb
= __build_skb(data
, len
);
486 if (unlikely(!skb
)) {
491 /* use OR instead of assignment to avoid clearing of bits in mask */
492 if (nc
->page
.pfmemalloc
)
497 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
);
498 skb
->dev
= napi
->dev
;
503 EXPORT_SYMBOL(__napi_alloc_skb
);
505 void skb_add_rx_frag(struct sk_buff
*skb
, int i
, struct page
*page
, int off
,
506 int size
, unsigned int truesize
)
508 skb_fill_page_desc(skb
, i
, page
, off
, size
);
510 skb
->data_len
+= size
;
511 skb
->truesize
+= truesize
;
513 EXPORT_SYMBOL(skb_add_rx_frag
);
515 void skb_coalesce_rx_frag(struct sk_buff
*skb
, int i
, int size
,
516 unsigned int truesize
)
518 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
520 skb_frag_size_add(frag
, size
);
522 skb
->data_len
+= size
;
523 skb
->truesize
+= truesize
;
525 EXPORT_SYMBOL(skb_coalesce_rx_frag
);
527 static void skb_drop_list(struct sk_buff
**listp
)
529 kfree_skb_list(*listp
);
533 static inline void skb_drop_fraglist(struct sk_buff
*skb
)
535 skb_drop_list(&skb_shinfo(skb
)->frag_list
);
538 static void skb_clone_fraglist(struct sk_buff
*skb
)
540 struct sk_buff
*list
;
542 skb_walk_frags(skb
, list
)
546 static void skb_free_head(struct sk_buff
*skb
)
548 unsigned char *head
= skb
->head
;
556 static void skb_release_data(struct sk_buff
*skb
)
558 struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
562 atomic_sub_return(skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1,
566 for (i
= 0; i
< shinfo
->nr_frags
; i
++)
567 __skb_frag_unref(&shinfo
->frags
[i
]);
569 if (shinfo
->frag_list
)
570 kfree_skb_list(shinfo
->frag_list
);
572 skb_zcopy_clear(skb
, true);
577 * Free an skbuff by memory without cleaning the state.
579 static void kfree_skbmem(struct sk_buff
*skb
)
581 struct sk_buff_fclones
*fclones
;
583 switch (skb
->fclone
) {
584 case SKB_FCLONE_UNAVAILABLE
:
585 kmem_cache_free(skbuff_head_cache
, skb
);
588 case SKB_FCLONE_ORIG
:
589 fclones
= container_of(skb
, struct sk_buff_fclones
, skb1
);
591 /* We usually free the clone (TX completion) before original skb
592 * This test would have no chance to be true for the clone,
593 * while here, branch prediction will be good.
595 if (refcount_read(&fclones
->fclone_ref
) == 1)
599 default: /* SKB_FCLONE_CLONE */
600 fclones
= container_of(skb
, struct sk_buff_fclones
, skb2
);
603 if (!refcount_dec_and_test(&fclones
->fclone_ref
))
606 kmem_cache_free(skbuff_fclone_cache
, fclones
);
609 void skb_release_head_state(struct sk_buff
*skb
)
612 if (skb
->destructor
) {
614 skb
->destructor(skb
);
616 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
617 nf_conntrack_put(skb_nfct(skb
));
622 /* Free everything but the sk_buff shell. */
623 static void skb_release_all(struct sk_buff
*skb
)
625 skb_release_head_state(skb
);
626 if (likely(skb
->head
))
627 skb_release_data(skb
);
631 * __kfree_skb - private function
634 * Free an sk_buff. Release anything attached to the buffer.
635 * Clean the state. This is an internal helper function. Users should
636 * always call kfree_skb
639 void __kfree_skb(struct sk_buff
*skb
)
641 skb_release_all(skb
);
644 EXPORT_SYMBOL(__kfree_skb
);
647 * kfree_skb - free an sk_buff
648 * @skb: buffer to free
650 * Drop a reference to the buffer and free it if the usage count has
653 void kfree_skb(struct sk_buff
*skb
)
658 trace_kfree_skb(skb
, __builtin_return_address(0));
661 EXPORT_SYMBOL(kfree_skb
);
663 void kfree_skb_list(struct sk_buff
*segs
)
666 struct sk_buff
*next
= segs
->next
;
672 EXPORT_SYMBOL(kfree_skb_list
);
675 * skb_tx_error - report an sk_buff xmit error
676 * @skb: buffer that triggered an error
678 * Report xmit error if a device callback is tracking this skb.
679 * skb must be freed afterwards.
681 void skb_tx_error(struct sk_buff
*skb
)
683 skb_zcopy_clear(skb
, true);
685 EXPORT_SYMBOL(skb_tx_error
);
688 * consume_skb - free an skbuff
689 * @skb: buffer to free
691 * Drop a ref to the buffer and free it if the usage count has hit zero
692 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
693 * is being dropped after a failure and notes that
695 void consume_skb(struct sk_buff
*skb
)
700 trace_consume_skb(skb
);
703 EXPORT_SYMBOL(consume_skb
);
706 * consume_stateless_skb - free an skbuff, assuming it is stateless
707 * @skb: buffer to free
709 * Alike consume_skb(), but this variant assumes that this is the last
710 * skb reference and all the head states have been already dropped
712 void __consume_stateless_skb(struct sk_buff
*skb
)
714 trace_consume_skb(skb
);
715 skb_release_data(skb
);
719 void __kfree_skb_flush(void)
721 struct napi_alloc_cache
*nc
= this_cpu_ptr(&napi_alloc_cache
);
723 /* flush skb_cache if containing objects */
725 kmem_cache_free_bulk(skbuff_head_cache
, nc
->skb_count
,
731 static inline void _kfree_skb_defer(struct sk_buff
*skb
)
733 struct napi_alloc_cache
*nc
= this_cpu_ptr(&napi_alloc_cache
);
735 /* drop skb->head and call any destructors for packet */
736 skb_release_all(skb
);
738 /* record skb to CPU local list */
739 nc
->skb_cache
[nc
->skb_count
++] = skb
;
742 /* SLUB writes into objects when freeing */
746 /* flush skb_cache if it is filled */
747 if (unlikely(nc
->skb_count
== NAPI_SKB_CACHE_SIZE
)) {
748 kmem_cache_free_bulk(skbuff_head_cache
, NAPI_SKB_CACHE_SIZE
,
753 void __kfree_skb_defer(struct sk_buff
*skb
)
755 _kfree_skb_defer(skb
);
758 void napi_consume_skb(struct sk_buff
*skb
, int budget
)
763 /* Zero budget indicate non-NAPI context called us, like netpoll */
764 if (unlikely(!budget
)) {
765 dev_consume_skb_any(skb
);
772 /* if reaching here SKB is ready to free */
773 trace_consume_skb(skb
);
775 /* if SKB is a clone, don't handle this case */
776 if (skb
->fclone
!= SKB_FCLONE_UNAVAILABLE
) {
781 _kfree_skb_defer(skb
);
783 EXPORT_SYMBOL(napi_consume_skb
);
785 /* Make sure a field is enclosed inside headers_start/headers_end section */
786 #define CHECK_SKB_FIELD(field) \
787 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
788 offsetof(struct sk_buff, headers_start)); \
789 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
790 offsetof(struct sk_buff, headers_end)); \
792 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
794 new->tstamp
= old
->tstamp
;
795 /* We do not copy old->sk */
797 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
798 skb_dst_copy(new, old
);
799 __skb_ext_copy(new, old
);
800 __nf_copy(new, old
, false);
802 /* Note : this field could be in headers_start/headers_end section
803 * It is not yet because we do not want to have a 16 bit hole
805 new->queue_mapping
= old
->queue_mapping
;
807 memcpy(&new->headers_start
, &old
->headers_start
,
808 offsetof(struct sk_buff
, headers_end
) -
809 offsetof(struct sk_buff
, headers_start
));
810 CHECK_SKB_FIELD(protocol
);
811 CHECK_SKB_FIELD(csum
);
812 CHECK_SKB_FIELD(hash
);
813 CHECK_SKB_FIELD(priority
);
814 CHECK_SKB_FIELD(skb_iif
);
815 CHECK_SKB_FIELD(vlan_proto
);
816 CHECK_SKB_FIELD(vlan_tci
);
817 CHECK_SKB_FIELD(transport_header
);
818 CHECK_SKB_FIELD(network_header
);
819 CHECK_SKB_FIELD(mac_header
);
820 CHECK_SKB_FIELD(inner_protocol
);
821 CHECK_SKB_FIELD(inner_transport_header
);
822 CHECK_SKB_FIELD(inner_network_header
);
823 CHECK_SKB_FIELD(inner_mac_header
);
824 CHECK_SKB_FIELD(mark
);
825 #ifdef CONFIG_NETWORK_SECMARK
826 CHECK_SKB_FIELD(secmark
);
828 #ifdef CONFIG_NET_RX_BUSY_POLL
829 CHECK_SKB_FIELD(napi_id
);
832 CHECK_SKB_FIELD(sender_cpu
);
834 #ifdef CONFIG_NET_SCHED
835 CHECK_SKB_FIELD(tc_index
);
841 * You should not add any new code to this function. Add it to
842 * __copy_skb_header above instead.
844 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
846 #define C(x) n->x = skb->x
848 n
->next
= n
->prev
= NULL
;
850 __copy_skb_header(n
, skb
);
855 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
860 n
->destructor
= NULL
;
867 refcount_set(&n
->users
, 1);
869 atomic_inc(&(skb_shinfo(skb
)->dataref
));
877 * skb_morph - morph one skb into another
878 * @dst: the skb to receive the contents
879 * @src: the skb to supply the contents
881 * This is identical to skb_clone except that the target skb is
882 * supplied by the user.
884 * The target skb is returned upon exit.
886 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
888 skb_release_all(dst
);
889 return __skb_clone(dst
, src
);
891 EXPORT_SYMBOL_GPL(skb_morph
);
893 int mm_account_pinned_pages(struct mmpin
*mmp
, size_t size
)
895 unsigned long max_pg
, num_pg
, new_pg
, old_pg
;
896 struct user_struct
*user
;
898 if (capable(CAP_IPC_LOCK
) || !size
)
901 num_pg
= (size
>> PAGE_SHIFT
) + 2; /* worst case */
902 max_pg
= rlimit(RLIMIT_MEMLOCK
) >> PAGE_SHIFT
;
903 user
= mmp
->user
? : current_user();
906 old_pg
= atomic_long_read(&user
->locked_vm
);
907 new_pg
= old_pg
+ num_pg
;
910 } while (atomic_long_cmpxchg(&user
->locked_vm
, old_pg
, new_pg
) !=
914 mmp
->user
= get_uid(user
);
915 mmp
->num_pg
= num_pg
;
917 mmp
->num_pg
+= num_pg
;
922 EXPORT_SYMBOL_GPL(mm_account_pinned_pages
);
924 void mm_unaccount_pinned_pages(struct mmpin
*mmp
)
927 atomic_long_sub(mmp
->num_pg
, &mmp
->user
->locked_vm
);
931 EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages
);
933 struct ubuf_info
*sock_zerocopy_alloc(struct sock
*sk
, size_t size
)
935 struct ubuf_info
*uarg
;
938 WARN_ON_ONCE(!in_task());
940 skb
= sock_omalloc(sk
, 0, GFP_KERNEL
);
944 BUILD_BUG_ON(sizeof(*uarg
) > sizeof(skb
->cb
));
945 uarg
= (void *)skb
->cb
;
946 uarg
->mmp
.user
= NULL
;
948 if (mm_account_pinned_pages(&uarg
->mmp
, size
)) {
953 uarg
->callback
= sock_zerocopy_callback
;
954 uarg
->id
= ((u32
)atomic_inc_return(&sk
->sk_zckey
)) - 1;
956 uarg
->bytelen
= size
;
958 refcount_set(&uarg
->refcnt
, 1);
963 EXPORT_SYMBOL_GPL(sock_zerocopy_alloc
);
965 static inline struct sk_buff
*skb_from_uarg(struct ubuf_info
*uarg
)
967 return container_of((void *)uarg
, struct sk_buff
, cb
);
970 struct ubuf_info
*sock_zerocopy_realloc(struct sock
*sk
, size_t size
,
971 struct ubuf_info
*uarg
)
974 const u32 byte_limit
= 1 << 19; /* limit to a few TSO */
977 /* realloc only when socket is locked (TCP, UDP cork),
978 * so uarg->len and sk_zckey access is serialized
980 if (!sock_owned_by_user(sk
)) {
985 bytelen
= uarg
->bytelen
+ size
;
986 if (uarg
->len
== USHRT_MAX
- 1 || bytelen
> byte_limit
) {
987 /* TCP can create new skb to attach new uarg */
988 if (sk
->sk_type
== SOCK_STREAM
)
993 next
= (u32
)atomic_read(&sk
->sk_zckey
);
994 if ((u32
)(uarg
->id
+ uarg
->len
) == next
) {
995 if (mm_account_pinned_pages(&uarg
->mmp
, size
))
998 uarg
->bytelen
= bytelen
;
999 atomic_set(&sk
->sk_zckey
, ++next
);
1000 sock_zerocopy_get(uarg
);
1006 return sock_zerocopy_alloc(sk
, size
);
1008 EXPORT_SYMBOL_GPL(sock_zerocopy_realloc
);
1010 static bool skb_zerocopy_notify_extend(struct sk_buff
*skb
, u32 lo
, u16 len
)
1012 struct sock_exterr_skb
*serr
= SKB_EXT_ERR(skb
);
1016 old_lo
= serr
->ee
.ee_info
;
1017 old_hi
= serr
->ee
.ee_data
;
1018 sum_len
= old_hi
- old_lo
+ 1ULL + len
;
1020 if (sum_len
>= (1ULL << 32))
1023 if (lo
!= old_hi
+ 1)
1026 serr
->ee
.ee_data
+= len
;
1030 void sock_zerocopy_callback(struct ubuf_info
*uarg
, bool success
)
1032 struct sk_buff
*tail
, *skb
= skb_from_uarg(uarg
);
1033 struct sock_exterr_skb
*serr
;
1034 struct sock
*sk
= skb
->sk
;
1035 struct sk_buff_head
*q
;
1036 unsigned long flags
;
1040 mm_unaccount_pinned_pages(&uarg
->mmp
);
1042 /* if !len, there was only 1 call, and it was aborted
1043 * so do not queue a completion notification
1045 if (!uarg
->len
|| sock_flag(sk
, SOCK_DEAD
))
1050 hi
= uarg
->id
+ len
- 1;
1052 serr
= SKB_EXT_ERR(skb
);
1053 memset(serr
, 0, sizeof(*serr
));
1054 serr
->ee
.ee_errno
= 0;
1055 serr
->ee
.ee_origin
= SO_EE_ORIGIN_ZEROCOPY
;
1056 serr
->ee
.ee_data
= hi
;
1057 serr
->ee
.ee_info
= lo
;
1059 serr
->ee
.ee_code
|= SO_EE_CODE_ZEROCOPY_COPIED
;
1061 q
= &sk
->sk_error_queue
;
1062 spin_lock_irqsave(&q
->lock
, flags
);
1063 tail
= skb_peek_tail(q
);
1064 if (!tail
|| SKB_EXT_ERR(tail
)->ee
.ee_origin
!= SO_EE_ORIGIN_ZEROCOPY
||
1065 !skb_zerocopy_notify_extend(tail
, lo
, len
)) {
1066 __skb_queue_tail(q
, skb
);
1069 spin_unlock_irqrestore(&q
->lock
, flags
);
1071 sk
->sk_error_report(sk
);
1077 EXPORT_SYMBOL_GPL(sock_zerocopy_callback
);
1079 void sock_zerocopy_put(struct ubuf_info
*uarg
)
1081 if (uarg
&& refcount_dec_and_test(&uarg
->refcnt
)) {
1083 uarg
->callback(uarg
, uarg
->zerocopy
);
1085 consume_skb(skb_from_uarg(uarg
));
1088 EXPORT_SYMBOL_GPL(sock_zerocopy_put
);
1090 void sock_zerocopy_put_abort(struct ubuf_info
*uarg
, bool have_uref
)
1093 struct sock
*sk
= skb_from_uarg(uarg
)->sk
;
1095 atomic_dec(&sk
->sk_zckey
);
1099 sock_zerocopy_put(uarg
);
1102 EXPORT_SYMBOL_GPL(sock_zerocopy_put_abort
);
1104 extern int __zerocopy_sg_from_iter(struct sock
*sk
, struct sk_buff
*skb
,
1105 struct iov_iter
*from
, size_t length
);
1107 int skb_zerocopy_iter_dgram(struct sk_buff
*skb
, struct msghdr
*msg
, int len
)
1109 return __zerocopy_sg_from_iter(skb
->sk
, skb
, &msg
->msg_iter
, len
);
1111 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram
);
1113 int skb_zerocopy_iter_stream(struct sock
*sk
, struct sk_buff
*skb
,
1114 struct msghdr
*msg
, int len
,
1115 struct ubuf_info
*uarg
)
1117 struct ubuf_info
*orig_uarg
= skb_zcopy(skb
);
1118 struct iov_iter orig_iter
= msg
->msg_iter
;
1119 int err
, orig_len
= skb
->len
;
1121 /* An skb can only point to one uarg. This edge case happens when
1122 * TCP appends to an skb, but zerocopy_realloc triggered a new alloc.
1124 if (orig_uarg
&& uarg
!= orig_uarg
)
1127 err
= __zerocopy_sg_from_iter(sk
, skb
, &msg
->msg_iter
, len
);
1128 if (err
== -EFAULT
|| (err
== -EMSGSIZE
&& skb
->len
== orig_len
)) {
1129 struct sock
*save_sk
= skb
->sk
;
1131 /* Streams do not free skb on error. Reset to prev state. */
1132 msg
->msg_iter
= orig_iter
;
1134 ___pskb_trim(skb
, orig_len
);
1139 skb_zcopy_set(skb
, uarg
, NULL
);
1140 return skb
->len
- orig_len
;
1142 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream
);
1144 static int skb_zerocopy_clone(struct sk_buff
*nskb
, struct sk_buff
*orig
,
1147 if (skb_zcopy(orig
)) {
1148 if (skb_zcopy(nskb
)) {
1149 /* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1154 if (skb_uarg(nskb
) == skb_uarg(orig
))
1156 if (skb_copy_ubufs(nskb
, GFP_ATOMIC
))
1159 skb_zcopy_set(nskb
, skb_uarg(orig
), NULL
);
1165 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
1166 * @skb: the skb to modify
1167 * @gfp_mask: allocation priority
1169 * This must be called on SKBTX_DEV_ZEROCOPY skb.
1170 * It will copy all frags into kernel and drop the reference
1171 * to userspace pages.
1173 * If this function is called from an interrupt gfp_mask() must be
1176 * Returns 0 on success or a negative error code on failure
1177 * to allocate kernel memory to copy to.
1179 int skb_copy_ubufs(struct sk_buff
*skb
, gfp_t gfp_mask
)
1181 int num_frags
= skb_shinfo(skb
)->nr_frags
;
1182 struct page
*page
, *head
= NULL
;
1186 if (skb_shared(skb
) || skb_unclone(skb
, gfp_mask
))
1192 new_frags
= (__skb_pagelen(skb
) + PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1193 for (i
= 0; i
< new_frags
; i
++) {
1194 page
= alloc_page(gfp_mask
);
1197 struct page
*next
= (struct page
*)page_private(head
);
1203 set_page_private(page
, (unsigned long)head
);
1209 for (i
= 0; i
< num_frags
; i
++) {
1210 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
1211 u32 p_off
, p_len
, copied
;
1215 skb_frag_foreach_page(f
, f
->page_offset
, skb_frag_size(f
),
1216 p
, p_off
, p_len
, copied
) {
1218 vaddr
= kmap_atomic(p
);
1220 while (done
< p_len
) {
1221 if (d_off
== PAGE_SIZE
) {
1223 page
= (struct page
*)page_private(page
);
1225 copy
= min_t(u32
, PAGE_SIZE
- d_off
, p_len
- done
);
1226 memcpy(page_address(page
) + d_off
,
1227 vaddr
+ p_off
+ done
, copy
);
1231 kunmap_atomic(vaddr
);
1235 /* skb frags release userspace buffers */
1236 for (i
= 0; i
< num_frags
; i
++)
1237 skb_frag_unref(skb
, i
);
1239 /* skb frags point to kernel buffers */
1240 for (i
= 0; i
< new_frags
- 1; i
++) {
1241 __skb_fill_page_desc(skb
, i
, head
, 0, PAGE_SIZE
);
1242 head
= (struct page
*)page_private(head
);
1244 __skb_fill_page_desc(skb
, new_frags
- 1, head
, 0, d_off
);
1245 skb_shinfo(skb
)->nr_frags
= new_frags
;
1248 skb_zcopy_clear(skb
, false);
1251 EXPORT_SYMBOL_GPL(skb_copy_ubufs
);
1254 * skb_clone - duplicate an sk_buff
1255 * @skb: buffer to clone
1256 * @gfp_mask: allocation priority
1258 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1259 * copies share the same packet data but not structure. The new
1260 * buffer has a reference count of 1. If the allocation fails the
1261 * function returns %NULL otherwise the new buffer is returned.
1263 * If this function is called from an interrupt gfp_mask() must be
1267 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
1269 struct sk_buff_fclones
*fclones
= container_of(skb
,
1270 struct sk_buff_fclones
,
1274 if (skb_orphan_frags(skb
, gfp_mask
))
1277 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
1278 refcount_read(&fclones
->fclone_ref
) == 1) {
1280 refcount_set(&fclones
->fclone_ref
, 2);
1282 if (skb_pfmemalloc(skb
))
1283 gfp_mask
|= __GFP_MEMALLOC
;
1285 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
1289 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
1292 return __skb_clone(n
, skb
);
1294 EXPORT_SYMBOL(skb_clone
);
1296 void skb_headers_offset_update(struct sk_buff
*skb
, int off
)
1298 /* Only adjust this if it actually is csum_start rather than csum */
1299 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1300 skb
->csum_start
+= off
;
1301 /* {transport,network,mac}_header and tail are relative to skb->head */
1302 skb
->transport_header
+= off
;
1303 skb
->network_header
+= off
;
1304 if (skb_mac_header_was_set(skb
))
1305 skb
->mac_header
+= off
;
1306 skb
->inner_transport_header
+= off
;
1307 skb
->inner_network_header
+= off
;
1308 skb
->inner_mac_header
+= off
;
1310 EXPORT_SYMBOL(skb_headers_offset_update
);
1312 void skb_copy_header(struct sk_buff
*new, const struct sk_buff
*old
)
1314 __copy_skb_header(new, old
);
1316 skb_shinfo(new)->gso_size
= skb_shinfo(old
)->gso_size
;
1317 skb_shinfo(new)->gso_segs
= skb_shinfo(old
)->gso_segs
;
1318 skb_shinfo(new)->gso_type
= skb_shinfo(old
)->gso_type
;
1320 EXPORT_SYMBOL(skb_copy_header
);
1322 static inline int skb_alloc_rx_flag(const struct sk_buff
*skb
)
1324 if (skb_pfmemalloc(skb
))
1325 return SKB_ALLOC_RX
;
1330 * skb_copy - create private copy of an sk_buff
1331 * @skb: buffer to copy
1332 * @gfp_mask: allocation priority
1334 * Make a copy of both an &sk_buff and its data. This is used when the
1335 * caller wishes to modify the data and needs a private copy of the
1336 * data to alter. Returns %NULL on failure or the pointer to the buffer
1337 * on success. The returned buffer has a reference count of 1.
1339 * As by-product this function converts non-linear &sk_buff to linear
1340 * one, so that &sk_buff becomes completely private and caller is allowed
1341 * to modify all the data of returned buffer. This means that this
1342 * function is not recommended for use in circumstances when only
1343 * header is going to be modified. Use pskb_copy() instead.
1346 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
1348 int headerlen
= skb_headroom(skb
);
1349 unsigned int size
= skb_end_offset(skb
) + skb
->data_len
;
1350 struct sk_buff
*n
= __alloc_skb(size
, gfp_mask
,
1351 skb_alloc_rx_flag(skb
), NUMA_NO_NODE
);
1356 /* Set the data pointer */
1357 skb_reserve(n
, headerlen
);
1358 /* Set the tail pointer and length */
1359 skb_put(n
, skb
->len
);
1361 BUG_ON(skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
));
1363 skb_copy_header(n
, skb
);
1366 EXPORT_SYMBOL(skb_copy
);
1369 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1370 * @skb: buffer to copy
1371 * @headroom: headroom of new skb
1372 * @gfp_mask: allocation priority
1373 * @fclone: if true allocate the copy of the skb from the fclone
1374 * cache instead of the head cache; it is recommended to set this
1375 * to true for the cases where the copy will likely be cloned
1377 * Make a copy of both an &sk_buff and part of its data, located
1378 * in header. Fragmented data remain shared. This is used when
1379 * the caller wishes to modify only header of &sk_buff and needs
1380 * private copy of the header to alter. Returns %NULL on failure
1381 * or the pointer to the buffer on success.
1382 * The returned buffer has a reference count of 1.
1385 struct sk_buff
*__pskb_copy_fclone(struct sk_buff
*skb
, int headroom
,
1386 gfp_t gfp_mask
, bool fclone
)
1388 unsigned int size
= skb_headlen(skb
) + headroom
;
1389 int flags
= skb_alloc_rx_flag(skb
) | (fclone
? SKB_ALLOC_FCLONE
: 0);
1390 struct sk_buff
*n
= __alloc_skb(size
, gfp_mask
, flags
, NUMA_NO_NODE
);
1395 /* Set the data pointer */
1396 skb_reserve(n
, headroom
);
1397 /* Set the tail pointer and length */
1398 skb_put(n
, skb_headlen(skb
));
1399 /* Copy the bytes */
1400 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
1402 n
->truesize
+= skb
->data_len
;
1403 n
->data_len
= skb
->data_len
;
1406 if (skb_shinfo(skb
)->nr_frags
) {
1409 if (skb_orphan_frags(skb
, gfp_mask
) ||
1410 skb_zerocopy_clone(n
, skb
, gfp_mask
)) {
1415 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1416 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
1417 skb_frag_ref(skb
, i
);
1419 skb_shinfo(n
)->nr_frags
= i
;
1422 if (skb_has_frag_list(skb
)) {
1423 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
1424 skb_clone_fraglist(n
);
1427 skb_copy_header(n
, skb
);
1431 EXPORT_SYMBOL(__pskb_copy_fclone
);
1434 * pskb_expand_head - reallocate header of &sk_buff
1435 * @skb: buffer to reallocate
1436 * @nhead: room to add at head
1437 * @ntail: room to add at tail
1438 * @gfp_mask: allocation priority
1440 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1441 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1442 * reference count of 1. Returns zero in the case of success or error,
1443 * if expansion failed. In the last case, &sk_buff is not changed.
1445 * All the pointers pointing into skb header may change and must be
1446 * reloaded after call to this function.
1449 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
1452 int i
, osize
= skb_end_offset(skb
);
1453 int size
= osize
+ nhead
+ ntail
;
1459 BUG_ON(skb_shared(skb
));
1461 size
= SKB_DATA_ALIGN(size
);
1463 if (skb_pfmemalloc(skb
))
1464 gfp_mask
|= __GFP_MEMALLOC
;
1465 data
= kmalloc_reserve(size
+ SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)),
1466 gfp_mask
, NUMA_NO_NODE
, NULL
);
1469 size
= SKB_WITH_OVERHEAD(ksize(data
));
1471 /* Copy only real data... and, alas, header. This should be
1472 * optimized for the cases when header is void.
1474 memcpy(data
+ nhead
, skb
->head
, skb_tail_pointer(skb
) - skb
->head
);
1476 memcpy((struct skb_shared_info
*)(data
+ size
),
1478 offsetof(struct skb_shared_info
, frags
[skb_shinfo(skb
)->nr_frags
]));
1481 * if shinfo is shared we must drop the old head gracefully, but if it
1482 * is not we can just drop the old head and let the existing refcount
1483 * be since all we did is relocate the values
1485 if (skb_cloned(skb
)) {
1486 if (skb_orphan_frags(skb
, gfp_mask
))
1489 refcount_inc(&skb_uarg(skb
)->refcnt
);
1490 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1491 skb_frag_ref(skb
, i
);
1493 if (skb_has_frag_list(skb
))
1494 skb_clone_fraglist(skb
);
1496 skb_release_data(skb
);
1500 off
= (data
+ nhead
) - skb
->head
;
1505 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1509 skb
->end
= skb
->head
+ size
;
1512 skb_headers_offset_update(skb
, nhead
);
1516 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
1518 skb_metadata_clear(skb
);
1520 /* It is not generally safe to change skb->truesize.
1521 * For the moment, we really care of rx path, or
1522 * when skb is orphaned (not attached to a socket).
1524 if (!skb
->sk
|| skb
->destructor
== sock_edemux
)
1525 skb
->truesize
+= size
- osize
;
1534 EXPORT_SYMBOL(pskb_expand_head
);
1536 /* Make private copy of skb with writable head and some headroom */
1538 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
1540 struct sk_buff
*skb2
;
1541 int delta
= headroom
- skb_headroom(skb
);
1544 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
1546 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1547 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
1555 EXPORT_SYMBOL(skb_realloc_headroom
);
1558 * skb_copy_expand - copy and expand sk_buff
1559 * @skb: buffer to copy
1560 * @newheadroom: new free bytes at head
1561 * @newtailroom: new free bytes at tail
1562 * @gfp_mask: allocation priority
1564 * Make a copy of both an &sk_buff and its data and while doing so
1565 * allocate additional space.
1567 * This is used when the caller wishes to modify the data and needs a
1568 * private copy of the data to alter as well as more space for new fields.
1569 * Returns %NULL on failure or the pointer to the buffer
1570 * on success. The returned buffer has a reference count of 1.
1572 * You must pass %GFP_ATOMIC as the allocation priority if this function
1573 * is called from an interrupt.
1575 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
1576 int newheadroom
, int newtailroom
,
1580 * Allocate the copy buffer
1582 struct sk_buff
*n
= __alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
1583 gfp_mask
, skb_alloc_rx_flag(skb
),
1585 int oldheadroom
= skb_headroom(skb
);
1586 int head_copy_len
, head_copy_off
;
1591 skb_reserve(n
, newheadroom
);
1593 /* Set the tail pointer and length */
1594 skb_put(n
, skb
->len
);
1596 head_copy_len
= oldheadroom
;
1598 if (newheadroom
<= head_copy_len
)
1599 head_copy_len
= newheadroom
;
1601 head_copy_off
= newheadroom
- head_copy_len
;
1603 /* Copy the linear header and data. */
1604 BUG_ON(skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
1605 skb
->len
+ head_copy_len
));
1607 skb_copy_header(n
, skb
);
1609 skb_headers_offset_update(n
, newheadroom
- oldheadroom
);
1613 EXPORT_SYMBOL(skb_copy_expand
);
1616 * __skb_pad - zero pad the tail of an skb
1617 * @skb: buffer to pad
1618 * @pad: space to pad
1619 * @free_on_error: free buffer on error
1621 * Ensure that a buffer is followed by a padding area that is zero
1622 * filled. Used by network drivers which may DMA or transfer data
1623 * beyond the buffer end onto the wire.
1625 * May return error in out of memory cases. The skb is freed on error
1626 * if @free_on_error is true.
1629 int __skb_pad(struct sk_buff
*skb
, int pad
, bool free_on_error
)
1634 /* If the skbuff is non linear tailroom is always zero.. */
1635 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
1636 memset(skb
->data
+skb
->len
, 0, pad
);
1640 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
1641 if (likely(skb_cloned(skb
) || ntail
> 0)) {
1642 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
1647 /* FIXME: The use of this function with non-linear skb's really needs
1650 err
= skb_linearize(skb
);
1654 memset(skb
->data
+ skb
->len
, 0, pad
);
1662 EXPORT_SYMBOL(__skb_pad
);
1665 * pskb_put - add data to the tail of a potentially fragmented buffer
1666 * @skb: start of the buffer to use
1667 * @tail: tail fragment of the buffer to use
1668 * @len: amount of data to add
1670 * This function extends the used data area of the potentially
1671 * fragmented buffer. @tail must be the last fragment of @skb -- or
1672 * @skb itself. If this would exceed the total buffer size the kernel
1673 * will panic. A pointer to the first byte of the extra data is
1677 void *pskb_put(struct sk_buff
*skb
, struct sk_buff
*tail
, int len
)
1680 skb
->data_len
+= len
;
1683 return skb_put(tail
, len
);
1685 EXPORT_SYMBOL_GPL(pskb_put
);
1688 * skb_put - add data to a buffer
1689 * @skb: buffer to use
1690 * @len: amount of data to add
1692 * This function extends the used data area of the buffer. If this would
1693 * exceed the total buffer size the kernel will panic. A pointer to the
1694 * first byte of the extra data is returned.
1696 void *skb_put(struct sk_buff
*skb
, unsigned int len
)
1698 void *tmp
= skb_tail_pointer(skb
);
1699 SKB_LINEAR_ASSERT(skb
);
1702 if (unlikely(skb
->tail
> skb
->end
))
1703 skb_over_panic(skb
, len
, __builtin_return_address(0));
1706 EXPORT_SYMBOL(skb_put
);
1709 * skb_push - add data to the start of a buffer
1710 * @skb: buffer to use
1711 * @len: amount of data to add
1713 * This function extends the used data area of the buffer at the buffer
1714 * start. If this would exceed the total buffer headroom the kernel will
1715 * panic. A pointer to the first byte of the extra data is returned.
1717 void *skb_push(struct sk_buff
*skb
, unsigned int len
)
1721 if (unlikely(skb
->data
< skb
->head
))
1722 skb_under_panic(skb
, len
, __builtin_return_address(0));
1725 EXPORT_SYMBOL(skb_push
);
1728 * skb_pull - remove data from the start of a buffer
1729 * @skb: buffer to use
1730 * @len: amount of data to remove
1732 * This function removes data from the start of a buffer, returning
1733 * the memory to the headroom. A pointer to the next data in the buffer
1734 * is returned. Once the data has been pulled future pushes will overwrite
1737 void *skb_pull(struct sk_buff
*skb
, unsigned int len
)
1739 return skb_pull_inline(skb
, len
);
1741 EXPORT_SYMBOL(skb_pull
);
1744 * skb_trim - remove end from a buffer
1745 * @skb: buffer to alter
1748 * Cut the length of a buffer down by removing data from the tail. If
1749 * the buffer is already under the length specified it is not modified.
1750 * The skb must be linear.
1752 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1755 __skb_trim(skb
, len
);
1757 EXPORT_SYMBOL(skb_trim
);
1759 /* Trims skb to length len. It can change skb pointers.
1762 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1764 struct sk_buff
**fragp
;
1765 struct sk_buff
*frag
;
1766 int offset
= skb_headlen(skb
);
1767 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1771 if (skb_cloned(skb
) &&
1772 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1779 for (; i
< nfrags
; i
++) {
1780 int end
= offset
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1787 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
++], len
- offset
);
1790 skb_shinfo(skb
)->nr_frags
= i
;
1792 for (; i
< nfrags
; i
++)
1793 skb_frag_unref(skb
, i
);
1795 if (skb_has_frag_list(skb
))
1796 skb_drop_fraglist(skb
);
1800 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1801 fragp
= &frag
->next
) {
1802 int end
= offset
+ frag
->len
;
1804 if (skb_shared(frag
)) {
1805 struct sk_buff
*nfrag
;
1807 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1808 if (unlikely(!nfrag
))
1811 nfrag
->next
= frag
->next
;
1823 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1827 skb_drop_list(&frag
->next
);
1832 if (len
> skb_headlen(skb
)) {
1833 skb
->data_len
-= skb
->len
- len
;
1838 skb_set_tail_pointer(skb
, len
);
1841 if (!skb
->sk
|| skb
->destructor
== sock_edemux
)
1845 EXPORT_SYMBOL(___pskb_trim
);
1847 /* Note : use pskb_trim_rcsum() instead of calling this directly
1849 int pskb_trim_rcsum_slow(struct sk_buff
*skb
, unsigned int len
)
1851 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
1852 int delta
= skb
->len
- len
;
1854 skb
->csum
= csum_block_sub(skb
->csum
,
1855 skb_checksum(skb
, len
, delta
, 0),
1858 return __pskb_trim(skb
, len
);
1860 EXPORT_SYMBOL(pskb_trim_rcsum_slow
);
1863 * __pskb_pull_tail - advance tail of skb header
1864 * @skb: buffer to reallocate
1865 * @delta: number of bytes to advance tail
1867 * The function makes a sense only on a fragmented &sk_buff,
1868 * it expands header moving its tail forward and copying necessary
1869 * data from fragmented part.
1871 * &sk_buff MUST have reference count of 1.
1873 * Returns %NULL (and &sk_buff does not change) if pull failed
1874 * or value of new tail of skb in the case of success.
1876 * All the pointers pointing into skb header may change and must be
1877 * reloaded after call to this function.
1880 /* Moves tail of skb head forward, copying data from fragmented part,
1881 * when it is necessary.
1882 * 1. It may fail due to malloc failure.
1883 * 2. It may change skb pointers.
1885 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1887 void *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1889 /* If skb has not enough free space at tail, get new one
1890 * plus 128 bytes for future expansions. If we have enough
1891 * room at tail, reallocate without expansion only if skb is cloned.
1893 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1895 if (eat
> 0 || skb_cloned(skb
)) {
1896 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1901 BUG_ON(skb_copy_bits(skb
, skb_headlen(skb
),
1902 skb_tail_pointer(skb
), delta
));
1904 /* Optimization: no fragments, no reasons to preestimate
1905 * size of pulled pages. Superb.
1907 if (!skb_has_frag_list(skb
))
1910 /* Estimate size of pulled pages. */
1912 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1913 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1920 /* If we need update frag list, we are in troubles.
1921 * Certainly, it is possible to add an offset to skb data,
1922 * but taking into account that pulling is expected to
1923 * be very rare operation, it is worth to fight against
1924 * further bloating skb head and crucify ourselves here instead.
1925 * Pure masohism, indeed. 8)8)
1928 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1929 struct sk_buff
*clone
= NULL
;
1930 struct sk_buff
*insp
= NULL
;
1933 if (list
->len
<= eat
) {
1934 /* Eaten as whole. */
1939 /* Eaten partially. */
1941 if (skb_shared(list
)) {
1942 /* Sucks! We need to fork list. :-( */
1943 clone
= skb_clone(list
, GFP_ATOMIC
);
1949 /* This may be pulled without
1953 if (!pskb_pull(list
, eat
)) {
1961 /* Free pulled out fragments. */
1962 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1963 skb_shinfo(skb
)->frag_list
= list
->next
;
1966 /* And insert new clone at head. */
1969 skb_shinfo(skb
)->frag_list
= clone
;
1972 /* Success! Now we may commit changes to skb data. */
1977 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1978 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1981 skb_frag_unref(skb
, i
);
1984 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1986 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1987 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[k
], eat
);
1995 skb_shinfo(skb
)->nr_frags
= k
;
1999 skb
->data_len
-= delta
;
2002 skb_zcopy_clear(skb
, false);
2004 return skb_tail_pointer(skb
);
2006 EXPORT_SYMBOL(__pskb_pull_tail
);
2009 * skb_copy_bits - copy bits from skb to kernel buffer
2011 * @offset: offset in source
2012 * @to: destination buffer
2013 * @len: number of bytes to copy
2015 * Copy the specified number of bytes from the source skb to the
2016 * destination buffer.
2019 * If its prototype is ever changed,
2020 * check arch/{*}/net/{*}.S files,
2021 * since it is called from BPF assembly code.
2023 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
2025 int start
= skb_headlen(skb
);
2026 struct sk_buff
*frag_iter
;
2029 if (offset
> (int)skb
->len
- len
)
2033 if ((copy
= start
- offset
) > 0) {
2036 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
2037 if ((len
-= copy
) == 0)
2043 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2045 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
2047 WARN_ON(start
> offset
+ len
);
2049 end
= start
+ skb_frag_size(f
);
2050 if ((copy
= end
- offset
) > 0) {
2051 u32 p_off
, p_len
, copied
;
2058 skb_frag_foreach_page(f
,
2059 f
->page_offset
+ offset
- start
,
2060 copy
, p
, p_off
, p_len
, copied
) {
2061 vaddr
= kmap_atomic(p
);
2062 memcpy(to
+ copied
, vaddr
+ p_off
, p_len
);
2063 kunmap_atomic(vaddr
);
2066 if ((len
-= copy
) == 0)
2074 skb_walk_frags(skb
, frag_iter
) {
2077 WARN_ON(start
> offset
+ len
);
2079 end
= start
+ frag_iter
->len
;
2080 if ((copy
= end
- offset
) > 0) {
2083 if (skb_copy_bits(frag_iter
, offset
- start
, to
, copy
))
2085 if ((len
-= copy
) == 0)
2099 EXPORT_SYMBOL(skb_copy_bits
);
2102 * Callback from splice_to_pipe(), if we need to release some pages
2103 * at the end of the spd in case we error'ed out in filling the pipe.
2105 static void sock_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
2107 put_page(spd
->pages
[i
]);
2110 static struct page
*linear_to_page(struct page
*page
, unsigned int *len
,
2111 unsigned int *offset
,
2114 struct page_frag
*pfrag
= sk_page_frag(sk
);
2116 if (!sk_page_frag_refill(sk
, pfrag
))
2119 *len
= min_t(unsigned int, *len
, pfrag
->size
- pfrag
->offset
);
2121 memcpy(page_address(pfrag
->page
) + pfrag
->offset
,
2122 page_address(page
) + *offset
, *len
);
2123 *offset
= pfrag
->offset
;
2124 pfrag
->offset
+= *len
;
2129 static bool spd_can_coalesce(const struct splice_pipe_desc
*spd
,
2131 unsigned int offset
)
2133 return spd
->nr_pages
&&
2134 spd
->pages
[spd
->nr_pages
- 1] == page
&&
2135 (spd
->partial
[spd
->nr_pages
- 1].offset
+
2136 spd
->partial
[spd
->nr_pages
- 1].len
== offset
);
2140 * Fill page/offset/length into spd, if it can hold more pages.
2142 static bool spd_fill_page(struct splice_pipe_desc
*spd
,
2143 struct pipe_inode_info
*pipe
, struct page
*page
,
2144 unsigned int *len
, unsigned int offset
,
2148 if (unlikely(spd
->nr_pages
== MAX_SKB_FRAGS
))
2152 page
= linear_to_page(page
, len
, &offset
, sk
);
2156 if (spd_can_coalesce(spd
, page
, offset
)) {
2157 spd
->partial
[spd
->nr_pages
- 1].len
+= *len
;
2161 spd
->pages
[spd
->nr_pages
] = page
;
2162 spd
->partial
[spd
->nr_pages
].len
= *len
;
2163 spd
->partial
[spd
->nr_pages
].offset
= offset
;
2169 static bool __splice_segment(struct page
*page
, unsigned int poff
,
2170 unsigned int plen
, unsigned int *off
,
2172 struct splice_pipe_desc
*spd
, bool linear
,
2174 struct pipe_inode_info
*pipe
)
2179 /* skip this segment if already processed */
2185 /* ignore any bits we already processed */
2191 unsigned int flen
= min(*len
, plen
);
2193 if (spd_fill_page(spd
, pipe
, page
, &flen
, poff
,
2199 } while (*len
&& plen
);
2205 * Map linear and fragment data from the skb to spd. It reports true if the
2206 * pipe is full or if we already spliced the requested length.
2208 static bool __skb_splice_bits(struct sk_buff
*skb
, struct pipe_inode_info
*pipe
,
2209 unsigned int *offset
, unsigned int *len
,
2210 struct splice_pipe_desc
*spd
, struct sock
*sk
)
2213 struct sk_buff
*iter
;
2215 /* map the linear part :
2216 * If skb->head_frag is set, this 'linear' part is backed by a
2217 * fragment, and if the head is not shared with any clones then
2218 * we can avoid a copy since we own the head portion of this page.
2220 if (__splice_segment(virt_to_page(skb
->data
),
2221 (unsigned long) skb
->data
& (PAGE_SIZE
- 1),
2224 skb_head_is_locked(skb
),
2229 * then map the fragments
2231 for (seg
= 0; seg
< skb_shinfo(skb
)->nr_frags
; seg
++) {
2232 const skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[seg
];
2234 if (__splice_segment(skb_frag_page(f
),
2235 f
->page_offset
, skb_frag_size(f
),
2236 offset
, len
, spd
, false, sk
, pipe
))
2240 skb_walk_frags(skb
, iter
) {
2241 if (*offset
>= iter
->len
) {
2242 *offset
-= iter
->len
;
2245 /* __skb_splice_bits() only fails if the output has no room
2246 * left, so no point in going over the frag_list for the error
2249 if (__skb_splice_bits(iter
, pipe
, offset
, len
, spd
, sk
))
2257 * Map data from the skb to a pipe. Should handle both the linear part,
2258 * the fragments, and the frag list.
2260 int skb_splice_bits(struct sk_buff
*skb
, struct sock
*sk
, unsigned int offset
,
2261 struct pipe_inode_info
*pipe
, unsigned int tlen
,
2264 struct partial_page partial
[MAX_SKB_FRAGS
];
2265 struct page
*pages
[MAX_SKB_FRAGS
];
2266 struct splice_pipe_desc spd
= {
2269 .nr_pages_max
= MAX_SKB_FRAGS
,
2270 .ops
= &nosteal_pipe_buf_ops
,
2271 .spd_release
= sock_spd_release
,
2275 __skb_splice_bits(skb
, pipe
, &offset
, &tlen
, &spd
, sk
);
2278 ret
= splice_to_pipe(pipe
, &spd
);
2282 EXPORT_SYMBOL_GPL(skb_splice_bits
);
2284 /* Send skb data on a socket. Socket must be locked. */
2285 int skb_send_sock_locked(struct sock
*sk
, struct sk_buff
*skb
, int offset
,
2288 unsigned int orig_len
= len
;
2289 struct sk_buff
*head
= skb
;
2290 unsigned short fragidx
;
2295 /* Deal with head data */
2296 while (offset
< skb_headlen(skb
) && len
) {
2300 slen
= min_t(int, len
, skb_headlen(skb
) - offset
);
2301 kv
.iov_base
= skb
->data
+ offset
;
2303 memset(&msg
, 0, sizeof(msg
));
2305 ret
= kernel_sendmsg_locked(sk
, &msg
, &kv
, 1, slen
);
2313 /* All the data was skb head? */
2317 /* Make offset relative to start of frags */
2318 offset
-= skb_headlen(skb
);
2320 /* Find where we are in frag list */
2321 for (fragidx
= 0; fragidx
< skb_shinfo(skb
)->nr_frags
; fragidx
++) {
2322 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[fragidx
];
2324 if (offset
< frag
->size
)
2327 offset
-= frag
->size
;
2330 for (; len
&& fragidx
< skb_shinfo(skb
)->nr_frags
; fragidx
++) {
2331 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[fragidx
];
2333 slen
= min_t(size_t, len
, frag
->size
- offset
);
2336 ret
= kernel_sendpage_locked(sk
, frag
->page
.p
,
2337 frag
->page_offset
+ offset
,
2338 slen
, MSG_DONTWAIT
);
2351 /* Process any frag lists */
2354 if (skb_has_frag_list(skb
)) {
2355 skb
= skb_shinfo(skb
)->frag_list
;
2358 } else if (skb
->next
) {
2365 return orig_len
- len
;
2368 return orig_len
== len
? ret
: orig_len
- len
;
2370 EXPORT_SYMBOL_GPL(skb_send_sock_locked
);
2373 * skb_store_bits - store bits from kernel buffer to skb
2374 * @skb: destination buffer
2375 * @offset: offset in destination
2376 * @from: source buffer
2377 * @len: number of bytes to copy
2379 * Copy the specified number of bytes from the source buffer to the
2380 * destination skb. This function handles all the messy bits of
2381 * traversing fragment lists and such.
2384 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
2386 int start
= skb_headlen(skb
);
2387 struct sk_buff
*frag_iter
;
2390 if (offset
> (int)skb
->len
- len
)
2393 if ((copy
= start
- offset
) > 0) {
2396 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
2397 if ((len
-= copy
) == 0)
2403 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2404 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2407 WARN_ON(start
> offset
+ len
);
2409 end
= start
+ skb_frag_size(frag
);
2410 if ((copy
= end
- offset
) > 0) {
2411 u32 p_off
, p_len
, copied
;
2418 skb_frag_foreach_page(frag
,
2419 frag
->page_offset
+ offset
- start
,
2420 copy
, p
, p_off
, p_len
, copied
) {
2421 vaddr
= kmap_atomic(p
);
2422 memcpy(vaddr
+ p_off
, from
+ copied
, p_len
);
2423 kunmap_atomic(vaddr
);
2426 if ((len
-= copy
) == 0)
2434 skb_walk_frags(skb
, frag_iter
) {
2437 WARN_ON(start
> offset
+ len
);
2439 end
= start
+ frag_iter
->len
;
2440 if ((copy
= end
- offset
) > 0) {
2443 if (skb_store_bits(frag_iter
, offset
- start
,
2446 if ((len
-= copy
) == 0)
2459 EXPORT_SYMBOL(skb_store_bits
);
2461 /* Checksum skb data. */
2462 __wsum
__skb_checksum(const struct sk_buff
*skb
, int offset
, int len
,
2463 __wsum csum
, const struct skb_checksum_ops
*ops
)
2465 int start
= skb_headlen(skb
);
2466 int i
, copy
= start
- offset
;
2467 struct sk_buff
*frag_iter
;
2470 /* Checksum header. */
2474 csum
= ops
->update(skb
->data
+ offset
, copy
, csum
);
2475 if ((len
-= copy
) == 0)
2481 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2483 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2485 WARN_ON(start
> offset
+ len
);
2487 end
= start
+ skb_frag_size(frag
);
2488 if ((copy
= end
- offset
) > 0) {
2489 u32 p_off
, p_len
, copied
;
2497 skb_frag_foreach_page(frag
,
2498 frag
->page_offset
+ offset
- start
,
2499 copy
, p
, p_off
, p_len
, copied
) {
2500 vaddr
= kmap_atomic(p
);
2501 csum2
= ops
->update(vaddr
+ p_off
, p_len
, 0);
2502 kunmap_atomic(vaddr
);
2503 csum
= ops
->combine(csum
, csum2
, pos
, p_len
);
2514 skb_walk_frags(skb
, frag_iter
) {
2517 WARN_ON(start
> offset
+ len
);
2519 end
= start
+ frag_iter
->len
;
2520 if ((copy
= end
- offset
) > 0) {
2524 csum2
= __skb_checksum(frag_iter
, offset
- start
,
2526 csum
= ops
->combine(csum
, csum2
, pos
, copy
);
2527 if ((len
-= copy
) == 0)
2538 EXPORT_SYMBOL(__skb_checksum
);
2540 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
2541 int len
, __wsum csum
)
2543 const struct skb_checksum_ops ops
= {
2544 .update
= csum_partial_ext
,
2545 .combine
= csum_block_add_ext
,
2548 return __skb_checksum(skb
, offset
, len
, csum
, &ops
);
2550 EXPORT_SYMBOL(skb_checksum
);
2552 /* Both of above in one bottle. */
2554 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
2555 u8
*to
, int len
, __wsum csum
)
2557 int start
= skb_headlen(skb
);
2558 int i
, copy
= start
- offset
;
2559 struct sk_buff
*frag_iter
;
2566 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
2568 if ((len
-= copy
) == 0)
2575 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2578 WARN_ON(start
> offset
+ len
);
2580 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
2581 if ((copy
= end
- offset
) > 0) {
2582 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2583 u32 p_off
, p_len
, copied
;
2591 skb_frag_foreach_page(frag
,
2592 frag
->page_offset
+ offset
- start
,
2593 copy
, p
, p_off
, p_len
, copied
) {
2594 vaddr
= kmap_atomic(p
);
2595 csum2
= csum_partial_copy_nocheck(vaddr
+ p_off
,
2598 kunmap_atomic(vaddr
);
2599 csum
= csum_block_add(csum
, csum2
, pos
);
2611 skb_walk_frags(skb
, frag_iter
) {
2615 WARN_ON(start
> offset
+ len
);
2617 end
= start
+ frag_iter
->len
;
2618 if ((copy
= end
- offset
) > 0) {
2621 csum2
= skb_copy_and_csum_bits(frag_iter
,
2624 csum
= csum_block_add(csum
, csum2
, pos
);
2625 if ((len
-= copy
) == 0)
2636 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
2638 __sum16
__skb_checksum_complete_head(struct sk_buff
*skb
, int len
)
2642 sum
= csum_fold(skb_checksum(skb
, 0, len
, skb
->csum
));
2643 /* See comments in __skb_checksum_complete(). */
2645 if (unlikely(skb
->ip_summed
== CHECKSUM_COMPLETE
) &&
2646 !skb
->csum_complete_sw
)
2647 netdev_rx_csum_fault(skb
->dev
, skb
);
2649 if (!skb_shared(skb
))
2650 skb
->csum_valid
= !sum
;
2653 EXPORT_SYMBOL(__skb_checksum_complete_head
);
2655 /* This function assumes skb->csum already holds pseudo header's checksum,
2656 * which has been changed from the hardware checksum, for example, by
2657 * __skb_checksum_validate_complete(). And, the original skb->csum must
2658 * have been validated unsuccessfully for CHECKSUM_COMPLETE case.
2660 * It returns non-zero if the recomputed checksum is still invalid, otherwise
2661 * zero. The new checksum is stored back into skb->csum unless the skb is
2664 __sum16
__skb_checksum_complete(struct sk_buff
*skb
)
2669 csum
= skb_checksum(skb
, 0, skb
->len
, 0);
2671 sum
= csum_fold(csum_add(skb
->csum
, csum
));
2672 /* This check is inverted, because we already knew the hardware
2673 * checksum is invalid before calling this function. So, if the
2674 * re-computed checksum is valid instead, then we have a mismatch
2675 * between the original skb->csum and skb_checksum(). This means either
2676 * the original hardware checksum is incorrect or we screw up skb->csum
2677 * when moving skb->data around.
2680 if (unlikely(skb
->ip_summed
== CHECKSUM_COMPLETE
) &&
2681 !skb
->csum_complete_sw
)
2682 netdev_rx_csum_fault(skb
->dev
, skb
);
2685 if (!skb_shared(skb
)) {
2686 /* Save full packet checksum */
2688 skb
->ip_summed
= CHECKSUM_COMPLETE
;
2689 skb
->csum_complete_sw
= 1;
2690 skb
->csum_valid
= !sum
;
2695 EXPORT_SYMBOL(__skb_checksum_complete
);
2697 static __wsum
warn_crc32c_csum_update(const void *buff
, int len
, __wsum sum
)
2699 net_warn_ratelimited(
2700 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2705 static __wsum
warn_crc32c_csum_combine(__wsum csum
, __wsum csum2
,
2706 int offset
, int len
)
2708 net_warn_ratelimited(
2709 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2714 static const struct skb_checksum_ops default_crc32c_ops
= {
2715 .update
= warn_crc32c_csum_update
,
2716 .combine
= warn_crc32c_csum_combine
,
2719 const struct skb_checksum_ops
*crc32c_csum_stub __read_mostly
=
2720 &default_crc32c_ops
;
2721 EXPORT_SYMBOL(crc32c_csum_stub
);
2724 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2725 * @from: source buffer
2727 * Calculates the amount of linear headroom needed in the 'to' skb passed
2728 * into skb_zerocopy().
2731 skb_zerocopy_headlen(const struct sk_buff
*from
)
2733 unsigned int hlen
= 0;
2735 if (!from
->head_frag
||
2736 skb_headlen(from
) < L1_CACHE_BYTES
||
2737 skb_shinfo(from
)->nr_frags
>= MAX_SKB_FRAGS
)
2738 hlen
= skb_headlen(from
);
2740 if (skb_has_frag_list(from
))
2745 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen
);
2748 * skb_zerocopy - Zero copy skb to skb
2749 * @to: destination buffer
2750 * @from: source buffer
2751 * @len: number of bytes to copy from source buffer
2752 * @hlen: size of linear headroom in destination buffer
2754 * Copies up to `len` bytes from `from` to `to` by creating references
2755 * to the frags in the source buffer.
2757 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2758 * headroom in the `to` buffer.
2761 * 0: everything is OK
2762 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2763 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2766 skb_zerocopy(struct sk_buff
*to
, struct sk_buff
*from
, int len
, int hlen
)
2769 int plen
= 0; /* length of skb->head fragment */
2772 unsigned int offset
;
2774 BUG_ON(!from
->head_frag
&& !hlen
);
2776 /* dont bother with small payloads */
2777 if (len
<= skb_tailroom(to
))
2778 return skb_copy_bits(from
, 0, skb_put(to
, len
), len
);
2781 ret
= skb_copy_bits(from
, 0, skb_put(to
, hlen
), hlen
);
2786 plen
= min_t(int, skb_headlen(from
), len
);
2788 page
= virt_to_head_page(from
->head
);
2789 offset
= from
->data
- (unsigned char *)page_address(page
);
2790 __skb_fill_page_desc(to
, 0, page
, offset
, plen
);
2797 to
->truesize
+= len
+ plen
;
2798 to
->len
+= len
+ plen
;
2799 to
->data_len
+= len
+ plen
;
2801 if (unlikely(skb_orphan_frags(from
, GFP_ATOMIC
))) {
2805 skb_zerocopy_clone(to
, from
, GFP_ATOMIC
);
2807 for (i
= 0; i
< skb_shinfo(from
)->nr_frags
; i
++) {
2810 skb_shinfo(to
)->frags
[j
] = skb_shinfo(from
)->frags
[i
];
2811 skb_shinfo(to
)->frags
[j
].size
= min_t(int, skb_shinfo(to
)->frags
[j
].size
, len
);
2812 len
-= skb_shinfo(to
)->frags
[j
].size
;
2813 skb_frag_ref(to
, j
);
2816 skb_shinfo(to
)->nr_frags
= j
;
2820 EXPORT_SYMBOL_GPL(skb_zerocopy
);
2822 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
2827 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
2828 csstart
= skb_checksum_start_offset(skb
);
2830 csstart
= skb_headlen(skb
);
2832 BUG_ON(csstart
> skb_headlen(skb
));
2834 skb_copy_from_linear_data(skb
, to
, csstart
);
2837 if (csstart
!= skb
->len
)
2838 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
2839 skb
->len
- csstart
, 0);
2841 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2842 long csstuff
= csstart
+ skb
->csum_offset
;
2844 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
2847 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
2850 * skb_dequeue - remove from the head of the queue
2851 * @list: list to dequeue from
2853 * Remove the head of the list. The list lock is taken so the function
2854 * may be used safely with other locking list functions. The head item is
2855 * returned or %NULL if the list is empty.
2858 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
2860 unsigned long flags
;
2861 struct sk_buff
*result
;
2863 spin_lock_irqsave(&list
->lock
, flags
);
2864 result
= __skb_dequeue(list
);
2865 spin_unlock_irqrestore(&list
->lock
, flags
);
2868 EXPORT_SYMBOL(skb_dequeue
);
2871 * skb_dequeue_tail - remove from the tail of the queue
2872 * @list: list to dequeue from
2874 * Remove the tail of the list. The list lock is taken so the function
2875 * may be used safely with other locking list functions. The tail item is
2876 * returned or %NULL if the list is empty.
2878 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
2880 unsigned long flags
;
2881 struct sk_buff
*result
;
2883 spin_lock_irqsave(&list
->lock
, flags
);
2884 result
= __skb_dequeue_tail(list
);
2885 spin_unlock_irqrestore(&list
->lock
, flags
);
2888 EXPORT_SYMBOL(skb_dequeue_tail
);
2891 * skb_queue_purge - empty a list
2892 * @list: list to empty
2894 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2895 * the list and one reference dropped. This function takes the list
2896 * lock and is atomic with respect to other list locking functions.
2898 void skb_queue_purge(struct sk_buff_head
*list
)
2900 struct sk_buff
*skb
;
2901 while ((skb
= skb_dequeue(list
)) != NULL
)
2904 EXPORT_SYMBOL(skb_queue_purge
);
2907 * skb_rbtree_purge - empty a skb rbtree
2908 * @root: root of the rbtree to empty
2909 * Return value: the sum of truesizes of all purged skbs.
2911 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
2912 * the list and one reference dropped. This function does not take
2913 * any lock. Synchronization should be handled by the caller (e.g., TCP
2914 * out-of-order queue is protected by the socket lock).
2916 unsigned int skb_rbtree_purge(struct rb_root
*root
)
2918 struct rb_node
*p
= rb_first(root
);
2919 unsigned int sum
= 0;
2922 struct sk_buff
*skb
= rb_entry(p
, struct sk_buff
, rbnode
);
2925 rb_erase(&skb
->rbnode
, root
);
2926 sum
+= skb
->truesize
;
2933 * skb_queue_head - queue a buffer at the list head
2934 * @list: list to use
2935 * @newsk: buffer to queue
2937 * Queue a buffer at the start of the list. This function takes the
2938 * list lock and can be used safely with other locking &sk_buff functions
2941 * A buffer cannot be placed on two lists at the same time.
2943 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2945 unsigned long flags
;
2947 spin_lock_irqsave(&list
->lock
, flags
);
2948 __skb_queue_head(list
, newsk
);
2949 spin_unlock_irqrestore(&list
->lock
, flags
);
2951 EXPORT_SYMBOL(skb_queue_head
);
2954 * skb_queue_tail - queue a buffer at the list tail
2955 * @list: list to use
2956 * @newsk: buffer to queue
2958 * Queue a buffer at the tail of the list. This function takes the
2959 * list lock and can be used safely with other locking &sk_buff functions
2962 * A buffer cannot be placed on two lists at the same time.
2964 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2966 unsigned long flags
;
2968 spin_lock_irqsave(&list
->lock
, flags
);
2969 __skb_queue_tail(list
, newsk
);
2970 spin_unlock_irqrestore(&list
->lock
, flags
);
2972 EXPORT_SYMBOL(skb_queue_tail
);
2975 * skb_unlink - remove a buffer from a list
2976 * @skb: buffer to remove
2977 * @list: list to use
2979 * Remove a packet from a list. The list locks are taken and this
2980 * function is atomic with respect to other list locked calls
2982 * You must know what list the SKB is on.
2984 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
2986 unsigned long flags
;
2988 spin_lock_irqsave(&list
->lock
, flags
);
2989 __skb_unlink(skb
, list
);
2990 spin_unlock_irqrestore(&list
->lock
, flags
);
2992 EXPORT_SYMBOL(skb_unlink
);
2995 * skb_append - append a buffer
2996 * @old: buffer to insert after
2997 * @newsk: buffer to insert
2998 * @list: list to use
3000 * Place a packet after a given packet in a list. The list locks are taken
3001 * and this function is atomic with respect to other list locked calls.
3002 * A buffer cannot be placed on two lists at the same time.
3004 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
3006 unsigned long flags
;
3008 spin_lock_irqsave(&list
->lock
, flags
);
3009 __skb_queue_after(list
, old
, newsk
);
3010 spin_unlock_irqrestore(&list
->lock
, flags
);
3012 EXPORT_SYMBOL(skb_append
);
3014 static inline void skb_split_inside_header(struct sk_buff
*skb
,
3015 struct sk_buff
* skb1
,
3016 const u32 len
, const int pos
)
3020 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
3022 /* And move data appendix as is. */
3023 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
3024 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
3026 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
3027 skb_shinfo(skb
)->nr_frags
= 0;
3028 skb1
->data_len
= skb
->data_len
;
3029 skb1
->len
+= skb1
->data_len
;
3032 skb_set_tail_pointer(skb
, len
);
3035 static inline void skb_split_no_header(struct sk_buff
*skb
,
3036 struct sk_buff
* skb1
,
3037 const u32 len
, int pos
)
3040 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
3042 skb_shinfo(skb
)->nr_frags
= 0;
3043 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
3045 skb
->data_len
= len
- pos
;
3047 for (i
= 0; i
< nfrags
; i
++) {
3048 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
3050 if (pos
+ size
> len
) {
3051 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
3055 * We have two variants in this case:
3056 * 1. Move all the frag to the second
3057 * part, if it is possible. F.e.
3058 * this approach is mandatory for TUX,
3059 * where splitting is expensive.
3060 * 2. Split is accurately. We make this.
3062 skb_frag_ref(skb
, i
);
3063 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
3064 skb_frag_size_sub(&skb_shinfo(skb1
)->frags
[0], len
- pos
);
3065 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
], len
- pos
);
3066 skb_shinfo(skb
)->nr_frags
++;
3070 skb_shinfo(skb
)->nr_frags
++;
3073 skb_shinfo(skb1
)->nr_frags
= k
;
3077 * skb_split - Split fragmented skb to two parts at length len.
3078 * @skb: the buffer to split
3079 * @skb1: the buffer to receive the second part
3080 * @len: new length for skb
3082 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
3084 int pos
= skb_headlen(skb
);
3086 skb_shinfo(skb1
)->tx_flags
|= skb_shinfo(skb
)->tx_flags
&
3088 skb_zerocopy_clone(skb1
, skb
, 0);
3089 if (len
< pos
) /* Split line is inside header. */
3090 skb_split_inside_header(skb
, skb1
, len
, pos
);
3091 else /* Second chunk has no header, nothing to copy. */
3092 skb_split_no_header(skb
, skb1
, len
, pos
);
3094 EXPORT_SYMBOL(skb_split
);
3096 /* Shifting from/to a cloned skb is a no-go.
3098 * Caller cannot keep skb_shinfo related pointers past calling here!
3100 static int skb_prepare_for_shift(struct sk_buff
*skb
)
3102 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
3106 * skb_shift - Shifts paged data partially from skb to another
3107 * @tgt: buffer into which tail data gets added
3108 * @skb: buffer from which the paged data comes from
3109 * @shiftlen: shift up to this many bytes
3111 * Attempts to shift up to shiftlen worth of bytes, which may be less than
3112 * the length of the skb, from skb to tgt. Returns number bytes shifted.
3113 * It's up to caller to free skb if everything was shifted.
3115 * If @tgt runs out of frags, the whole operation is aborted.
3117 * Skb cannot include anything else but paged data while tgt is allowed
3118 * to have non-paged data as well.
3120 * TODO: full sized shift could be optimized but that would need
3121 * specialized skb free'er to handle frags without up-to-date nr_frags.
3123 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
3125 int from
, to
, merge
, todo
;
3126 struct skb_frag_struct
*fragfrom
, *fragto
;
3128 BUG_ON(shiftlen
> skb
->len
);
3130 if (skb_headlen(skb
))
3132 if (skb_zcopy(tgt
) || skb_zcopy(skb
))
3137 to
= skb_shinfo(tgt
)->nr_frags
;
3138 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
3140 /* Actual merge is delayed until the point when we know we can
3141 * commit all, so that we don't have to undo partial changes
3144 !skb_can_coalesce(tgt
, to
, skb_frag_page(fragfrom
),
3145 fragfrom
->page_offset
)) {
3150 todo
-= skb_frag_size(fragfrom
);
3152 if (skb_prepare_for_shift(skb
) ||
3153 skb_prepare_for_shift(tgt
))
3156 /* All previous frag pointers might be stale! */
3157 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
3158 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
3160 skb_frag_size_add(fragto
, shiftlen
);
3161 skb_frag_size_sub(fragfrom
, shiftlen
);
3162 fragfrom
->page_offset
+= shiftlen
;
3170 /* Skip full, not-fitting skb to avoid expensive operations */
3171 if ((shiftlen
== skb
->len
) &&
3172 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
3175 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
3178 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
3179 if (to
== MAX_SKB_FRAGS
)
3182 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
3183 fragto
= &skb_shinfo(tgt
)->frags
[to
];
3185 if (todo
>= skb_frag_size(fragfrom
)) {
3186 *fragto
= *fragfrom
;
3187 todo
-= skb_frag_size(fragfrom
);
3192 __skb_frag_ref(fragfrom
);
3193 fragto
->page
= fragfrom
->page
;
3194 fragto
->page_offset
= fragfrom
->page_offset
;
3195 skb_frag_size_set(fragto
, todo
);
3197 fragfrom
->page_offset
+= todo
;
3198 skb_frag_size_sub(fragfrom
, todo
);
3206 /* Ready to "commit" this state change to tgt */
3207 skb_shinfo(tgt
)->nr_frags
= to
;
3210 fragfrom
= &skb_shinfo(skb
)->frags
[0];
3211 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
3213 skb_frag_size_add(fragto
, skb_frag_size(fragfrom
));
3214 __skb_frag_unref(fragfrom
);
3217 /* Reposition in the original skb */
3219 while (from
< skb_shinfo(skb
)->nr_frags
)
3220 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
3221 skb_shinfo(skb
)->nr_frags
= to
;
3223 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
3226 /* Most likely the tgt won't ever need its checksum anymore, skb on
3227 * the other hand might need it if it needs to be resent
3229 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
3230 skb
->ip_summed
= CHECKSUM_PARTIAL
;
3232 /* Yak, is it really working this way? Some helper please? */
3233 skb
->len
-= shiftlen
;
3234 skb
->data_len
-= shiftlen
;
3235 skb
->truesize
-= shiftlen
;
3236 tgt
->len
+= shiftlen
;
3237 tgt
->data_len
+= shiftlen
;
3238 tgt
->truesize
+= shiftlen
;
3244 * skb_prepare_seq_read - Prepare a sequential read of skb data
3245 * @skb: the buffer to read
3246 * @from: lower offset of data to be read
3247 * @to: upper offset of data to be read
3248 * @st: state variable
3250 * Initializes the specified state variable. Must be called before
3251 * invoking skb_seq_read() for the first time.
3253 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
3254 unsigned int to
, struct skb_seq_state
*st
)
3256 st
->lower_offset
= from
;
3257 st
->upper_offset
= to
;
3258 st
->root_skb
= st
->cur_skb
= skb
;
3259 st
->frag_idx
= st
->stepped_offset
= 0;
3260 st
->frag_data
= NULL
;
3262 EXPORT_SYMBOL(skb_prepare_seq_read
);
3265 * skb_seq_read - Sequentially read skb data
3266 * @consumed: number of bytes consumed by the caller so far
3267 * @data: destination pointer for data to be returned
3268 * @st: state variable
3270 * Reads a block of skb data at @consumed relative to the
3271 * lower offset specified to skb_prepare_seq_read(). Assigns
3272 * the head of the data block to @data and returns the length
3273 * of the block or 0 if the end of the skb data or the upper
3274 * offset has been reached.
3276 * The caller is not required to consume all of the data
3277 * returned, i.e. @consumed is typically set to the number
3278 * of bytes already consumed and the next call to
3279 * skb_seq_read() will return the remaining part of the block.
3281 * Note 1: The size of each block of data returned can be arbitrary,
3282 * this limitation is the cost for zerocopy sequential
3283 * reads of potentially non linear data.
3285 * Note 2: Fragment lists within fragments are not implemented
3286 * at the moment, state->root_skb could be replaced with
3287 * a stack for this purpose.
3289 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
3290 struct skb_seq_state
*st
)
3292 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
3295 if (unlikely(abs_offset
>= st
->upper_offset
)) {
3296 if (st
->frag_data
) {
3297 kunmap_atomic(st
->frag_data
);
3298 st
->frag_data
= NULL
;
3304 block_limit
= skb_headlen(st
->cur_skb
) + st
->stepped_offset
;
3306 if (abs_offset
< block_limit
&& !st
->frag_data
) {
3307 *data
= st
->cur_skb
->data
+ (abs_offset
- st
->stepped_offset
);
3308 return block_limit
- abs_offset
;
3311 if (st
->frag_idx
== 0 && !st
->frag_data
)
3312 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
3314 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
3315 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
3316 block_limit
= skb_frag_size(frag
) + st
->stepped_offset
;
3318 if (abs_offset
< block_limit
) {
3320 st
->frag_data
= kmap_atomic(skb_frag_page(frag
));
3322 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
3323 (abs_offset
- st
->stepped_offset
);
3325 return block_limit
- abs_offset
;
3328 if (st
->frag_data
) {
3329 kunmap_atomic(st
->frag_data
);
3330 st
->frag_data
= NULL
;
3334 st
->stepped_offset
+= skb_frag_size(frag
);
3337 if (st
->frag_data
) {
3338 kunmap_atomic(st
->frag_data
);
3339 st
->frag_data
= NULL
;
3342 if (st
->root_skb
== st
->cur_skb
&& skb_has_frag_list(st
->root_skb
)) {
3343 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
3346 } else if (st
->cur_skb
->next
) {
3347 st
->cur_skb
= st
->cur_skb
->next
;
3354 EXPORT_SYMBOL(skb_seq_read
);
3357 * skb_abort_seq_read - Abort a sequential read of skb data
3358 * @st: state variable
3360 * Must be called if skb_seq_read() was not called until it
3363 void skb_abort_seq_read(struct skb_seq_state
*st
)
3366 kunmap_atomic(st
->frag_data
);
3368 EXPORT_SYMBOL(skb_abort_seq_read
);
3370 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
3372 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
3373 struct ts_config
*conf
,
3374 struct ts_state
*state
)
3376 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
3379 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
3381 skb_abort_seq_read(TS_SKB_CB(state
));
3385 * skb_find_text - Find a text pattern in skb data
3386 * @skb: the buffer to look in
3387 * @from: search offset
3389 * @config: textsearch configuration
3391 * Finds a pattern in the skb data according to the specified
3392 * textsearch configuration. Use textsearch_next() to retrieve
3393 * subsequent occurrences of the pattern. Returns the offset
3394 * to the first occurrence or UINT_MAX if no match was found.
3396 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
3397 unsigned int to
, struct ts_config
*config
)
3399 struct ts_state state
;
3402 config
->get_next_block
= skb_ts_get_next_block
;
3403 config
->finish
= skb_ts_finish
;
3405 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(&state
));
3407 ret
= textsearch_find(config
, &state
);
3408 return (ret
<= to
- from
? ret
: UINT_MAX
);
3410 EXPORT_SYMBOL(skb_find_text
);
3412 int skb_append_pagefrags(struct sk_buff
*skb
, struct page
*page
,
3413 int offset
, size_t size
)
3415 int i
= skb_shinfo(skb
)->nr_frags
;
3417 if (skb_can_coalesce(skb
, i
, page
, offset
)) {
3418 skb_frag_size_add(&skb_shinfo(skb
)->frags
[i
- 1], size
);
3419 } else if (i
< MAX_SKB_FRAGS
) {
3421 skb_fill_page_desc(skb
, i
, page
, offset
, size
);
3428 EXPORT_SYMBOL_GPL(skb_append_pagefrags
);
3431 * skb_pull_rcsum - pull skb and update receive checksum
3432 * @skb: buffer to update
3433 * @len: length of data pulled
3435 * This function performs an skb_pull on the packet and updates
3436 * the CHECKSUM_COMPLETE checksum. It should be used on
3437 * receive path processing instead of skb_pull unless you know
3438 * that the checksum difference is zero (e.g., a valid IP header)
3439 * or you are setting ip_summed to CHECKSUM_NONE.
3441 void *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
3443 unsigned char *data
= skb
->data
;
3445 BUG_ON(len
> skb
->len
);
3446 __skb_pull(skb
, len
);
3447 skb_postpull_rcsum(skb
, data
, len
);
3450 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
3452 static inline skb_frag_t
skb_head_frag_to_page_desc(struct sk_buff
*frag_skb
)
3454 skb_frag_t head_frag
;
3457 page
= virt_to_head_page(frag_skb
->head
);
3458 head_frag
.page
.p
= page
;
3459 head_frag
.page_offset
= frag_skb
->data
-
3460 (unsigned char *)page_address(page
);
3461 head_frag
.size
= skb_headlen(frag_skb
);
3466 * skb_segment - Perform protocol segmentation on skb.
3467 * @head_skb: buffer to segment
3468 * @features: features for the output path (see dev->features)
3470 * This function performs segmentation on the given skb. It returns
3471 * a pointer to the first in a list of new skbs for the segments.
3472 * In case of error it returns ERR_PTR(err).
3474 struct sk_buff
*skb_segment(struct sk_buff
*head_skb
,
3475 netdev_features_t features
)
3477 struct sk_buff
*segs
= NULL
;
3478 struct sk_buff
*tail
= NULL
;
3479 struct sk_buff
*list_skb
= skb_shinfo(head_skb
)->frag_list
;
3480 skb_frag_t
*frag
= skb_shinfo(head_skb
)->frags
;
3481 unsigned int mss
= skb_shinfo(head_skb
)->gso_size
;
3482 unsigned int doffset
= head_skb
->data
- skb_mac_header(head_skb
);
3483 struct sk_buff
*frag_skb
= head_skb
;
3484 unsigned int offset
= doffset
;
3485 unsigned int tnl_hlen
= skb_tnl_header_len(head_skb
);
3486 unsigned int partial_segs
= 0;
3487 unsigned int headroom
;
3488 unsigned int len
= head_skb
->len
;
3491 int nfrags
= skb_shinfo(head_skb
)->nr_frags
;
3497 __skb_push(head_skb
, doffset
);
3498 proto
= skb_network_protocol(head_skb
, &dummy
);
3499 if (unlikely(!proto
))
3500 return ERR_PTR(-EINVAL
);
3502 sg
= !!(features
& NETIF_F_SG
);
3503 csum
= !!can_checksum_protocol(features
, proto
);
3505 if (sg
&& csum
&& (mss
!= GSO_BY_FRAGS
)) {
3506 if (!(features
& NETIF_F_GSO_PARTIAL
)) {
3507 struct sk_buff
*iter
;
3508 unsigned int frag_len
;
3511 !net_gso_ok(features
, skb_shinfo(head_skb
)->gso_type
))
3514 /* If we get here then all the required
3515 * GSO features except frag_list are supported.
3516 * Try to split the SKB to multiple GSO SKBs
3517 * with no frag_list.
3518 * Currently we can do that only when the buffers don't
3519 * have a linear part and all the buffers except
3520 * the last are of the same length.
3522 frag_len
= list_skb
->len
;
3523 skb_walk_frags(head_skb
, iter
) {
3524 if (frag_len
!= iter
->len
&& iter
->next
)
3526 if (skb_headlen(iter
) && !iter
->head_frag
)
3532 if (len
!= frag_len
)
3536 /* GSO partial only requires that we trim off any excess that
3537 * doesn't fit into an MSS sized block, so take care of that
3540 partial_segs
= len
/ mss
;
3541 if (partial_segs
> 1)
3542 mss
*= partial_segs
;
3548 headroom
= skb_headroom(head_skb
);
3549 pos
= skb_headlen(head_skb
);
3552 struct sk_buff
*nskb
;
3553 skb_frag_t
*nskb_frag
;
3557 if (unlikely(mss
== GSO_BY_FRAGS
)) {
3558 len
= list_skb
->len
;
3560 len
= head_skb
->len
- offset
;
3565 hsize
= skb_headlen(head_skb
) - offset
;
3568 if (hsize
> len
|| !sg
)
3571 if (!hsize
&& i
>= nfrags
&& skb_headlen(list_skb
) &&
3572 (skb_headlen(list_skb
) == len
|| sg
)) {
3573 BUG_ON(skb_headlen(list_skb
) > len
);
3576 nfrags
= skb_shinfo(list_skb
)->nr_frags
;
3577 frag
= skb_shinfo(list_skb
)->frags
;
3578 frag_skb
= list_skb
;
3579 pos
+= skb_headlen(list_skb
);
3581 while (pos
< offset
+ len
) {
3582 BUG_ON(i
>= nfrags
);
3584 size
= skb_frag_size(frag
);
3585 if (pos
+ size
> offset
+ len
)
3593 nskb
= skb_clone(list_skb
, GFP_ATOMIC
);
3594 list_skb
= list_skb
->next
;
3596 if (unlikely(!nskb
))
3599 if (unlikely(pskb_trim(nskb
, len
))) {
3604 hsize
= skb_end_offset(nskb
);
3605 if (skb_cow_head(nskb
, doffset
+ headroom
)) {
3610 nskb
->truesize
+= skb_end_offset(nskb
) - hsize
;
3611 skb_release_head_state(nskb
);
3612 __skb_push(nskb
, doffset
);
3614 nskb
= __alloc_skb(hsize
+ doffset
+ headroom
,
3615 GFP_ATOMIC
, skb_alloc_rx_flag(head_skb
),
3618 if (unlikely(!nskb
))
3621 skb_reserve(nskb
, headroom
);
3622 __skb_put(nskb
, doffset
);
3631 __copy_skb_header(nskb
, head_skb
);
3633 skb_headers_offset_update(nskb
, skb_headroom(nskb
) - headroom
);
3634 skb_reset_mac_len(nskb
);
3636 skb_copy_from_linear_data_offset(head_skb
, -tnl_hlen
,
3637 nskb
->data
- tnl_hlen
,
3638 doffset
+ tnl_hlen
);
3640 if (nskb
->len
== len
+ doffset
)
3641 goto perform_csum_check
;
3644 if (!nskb
->remcsum_offload
)
3645 nskb
->ip_summed
= CHECKSUM_NONE
;
3646 SKB_GSO_CB(nskb
)->csum
=
3647 skb_copy_and_csum_bits(head_skb
, offset
,
3650 SKB_GSO_CB(nskb
)->csum_start
=
3651 skb_headroom(nskb
) + doffset
;
3655 nskb_frag
= skb_shinfo(nskb
)->frags
;
3657 skb_copy_from_linear_data_offset(head_skb
, offset
,
3658 skb_put(nskb
, hsize
), hsize
);
3660 skb_shinfo(nskb
)->tx_flags
|= skb_shinfo(head_skb
)->tx_flags
&
3663 if (skb_orphan_frags(frag_skb
, GFP_ATOMIC
) ||
3664 skb_zerocopy_clone(nskb
, frag_skb
, GFP_ATOMIC
))
3667 while (pos
< offset
+ len
) {
3670 nfrags
= skb_shinfo(list_skb
)->nr_frags
;
3671 frag
= skb_shinfo(list_skb
)->frags
;
3672 frag_skb
= list_skb
;
3673 if (!skb_headlen(list_skb
)) {
3676 BUG_ON(!list_skb
->head_frag
);
3678 /* to make room for head_frag. */
3682 if (skb_orphan_frags(frag_skb
, GFP_ATOMIC
) ||
3683 skb_zerocopy_clone(nskb
, frag_skb
,
3687 list_skb
= list_skb
->next
;
3690 if (unlikely(skb_shinfo(nskb
)->nr_frags
>=
3692 net_warn_ratelimited(
3693 "skb_segment: too many frags: %u %u\n",
3699 *nskb_frag
= (i
< 0) ? skb_head_frag_to_page_desc(frag_skb
) : *frag
;
3700 __skb_frag_ref(nskb_frag
);
3701 size
= skb_frag_size(nskb_frag
);
3704 nskb_frag
->page_offset
+= offset
- pos
;
3705 skb_frag_size_sub(nskb_frag
, offset
- pos
);
3708 skb_shinfo(nskb
)->nr_frags
++;
3710 if (pos
+ size
<= offset
+ len
) {
3715 skb_frag_size_sub(nskb_frag
, pos
+ size
- (offset
+ len
));
3723 nskb
->data_len
= len
- hsize
;
3724 nskb
->len
+= nskb
->data_len
;
3725 nskb
->truesize
+= nskb
->data_len
;
3729 if (skb_has_shared_frag(nskb
) &&
3730 __skb_linearize(nskb
))
3733 if (!nskb
->remcsum_offload
)
3734 nskb
->ip_summed
= CHECKSUM_NONE
;
3735 SKB_GSO_CB(nskb
)->csum
=
3736 skb_checksum(nskb
, doffset
,
3737 nskb
->len
- doffset
, 0);
3738 SKB_GSO_CB(nskb
)->csum_start
=
3739 skb_headroom(nskb
) + doffset
;
3741 } while ((offset
+= len
) < head_skb
->len
);
3743 /* Some callers want to get the end of the list.
3744 * Put it in segs->prev to avoid walking the list.
3745 * (see validate_xmit_skb_list() for example)
3750 struct sk_buff
*iter
;
3751 int type
= skb_shinfo(head_skb
)->gso_type
;
3752 unsigned short gso_size
= skb_shinfo(head_skb
)->gso_size
;
3754 /* Update type to add partial and then remove dodgy if set */
3755 type
|= (features
& NETIF_F_GSO_PARTIAL
) / NETIF_F_GSO_PARTIAL
* SKB_GSO_PARTIAL
;
3756 type
&= ~SKB_GSO_DODGY
;
3758 /* Update GSO info and prepare to start updating headers on
3759 * our way back down the stack of protocols.
3761 for (iter
= segs
; iter
; iter
= iter
->next
) {
3762 skb_shinfo(iter
)->gso_size
= gso_size
;
3763 skb_shinfo(iter
)->gso_segs
= partial_segs
;
3764 skb_shinfo(iter
)->gso_type
= type
;
3765 SKB_GSO_CB(iter
)->data_offset
= skb_headroom(iter
) + doffset
;
3768 if (tail
->len
- doffset
<= gso_size
)
3769 skb_shinfo(tail
)->gso_size
= 0;
3770 else if (tail
!= segs
)
3771 skb_shinfo(tail
)->gso_segs
= DIV_ROUND_UP(tail
->len
- doffset
, gso_size
);
3774 /* Following permits correct backpressure, for protocols
3775 * using skb_set_owner_w().
3776 * Idea is to tranfert ownership from head_skb to last segment.
3778 if (head_skb
->destructor
== sock_wfree
) {
3779 swap(tail
->truesize
, head_skb
->truesize
);
3780 swap(tail
->destructor
, head_skb
->destructor
);
3781 swap(tail
->sk
, head_skb
->sk
);
3786 kfree_skb_list(segs
);
3787 return ERR_PTR(err
);
3789 EXPORT_SYMBOL_GPL(skb_segment
);
3791 int skb_gro_receive(struct sk_buff
*p
, struct sk_buff
*skb
)
3793 struct skb_shared_info
*pinfo
, *skbinfo
= skb_shinfo(skb
);
3794 unsigned int offset
= skb_gro_offset(skb
);
3795 unsigned int headlen
= skb_headlen(skb
);
3796 unsigned int len
= skb_gro_len(skb
);
3797 unsigned int delta_truesize
;
3800 if (unlikely(p
->len
+ len
>= 65536))
3803 lp
= NAPI_GRO_CB(p
)->last
;
3804 pinfo
= skb_shinfo(lp
);
3806 if (headlen
<= offset
) {
3809 int i
= skbinfo
->nr_frags
;
3810 int nr_frags
= pinfo
->nr_frags
+ i
;
3812 if (nr_frags
> MAX_SKB_FRAGS
)
3816 pinfo
->nr_frags
= nr_frags
;
3817 skbinfo
->nr_frags
= 0;
3819 frag
= pinfo
->frags
+ nr_frags
;
3820 frag2
= skbinfo
->frags
+ i
;
3825 frag
->page_offset
+= offset
;
3826 skb_frag_size_sub(frag
, offset
);
3828 /* all fragments truesize : remove (head size + sk_buff) */
3829 delta_truesize
= skb
->truesize
-
3830 SKB_TRUESIZE(skb_end_offset(skb
));
3832 skb
->truesize
-= skb
->data_len
;
3833 skb
->len
-= skb
->data_len
;
3836 NAPI_GRO_CB(skb
)->free
= NAPI_GRO_FREE
;
3838 } else if (skb
->head_frag
) {
3839 int nr_frags
= pinfo
->nr_frags
;
3840 skb_frag_t
*frag
= pinfo
->frags
+ nr_frags
;
3841 struct page
*page
= virt_to_head_page(skb
->head
);
3842 unsigned int first_size
= headlen
- offset
;
3843 unsigned int first_offset
;
3845 if (nr_frags
+ 1 + skbinfo
->nr_frags
> MAX_SKB_FRAGS
)
3848 first_offset
= skb
->data
-
3849 (unsigned char *)page_address(page
) +
3852 pinfo
->nr_frags
= nr_frags
+ 1 + skbinfo
->nr_frags
;
3854 frag
->page
.p
= page
;
3855 frag
->page_offset
= first_offset
;
3856 skb_frag_size_set(frag
, first_size
);
3858 memcpy(frag
+ 1, skbinfo
->frags
, sizeof(*frag
) * skbinfo
->nr_frags
);
3859 /* We dont need to clear skbinfo->nr_frags here */
3861 delta_truesize
= skb
->truesize
- SKB_DATA_ALIGN(sizeof(struct sk_buff
));
3862 NAPI_GRO_CB(skb
)->free
= NAPI_GRO_FREE_STOLEN_HEAD
;
3867 delta_truesize
= skb
->truesize
;
3868 if (offset
> headlen
) {
3869 unsigned int eat
= offset
- headlen
;
3871 skbinfo
->frags
[0].page_offset
+= eat
;
3872 skb_frag_size_sub(&skbinfo
->frags
[0], eat
);
3873 skb
->data_len
-= eat
;
3878 __skb_pull(skb
, offset
);
3880 if (NAPI_GRO_CB(p
)->last
== p
)
3881 skb_shinfo(p
)->frag_list
= skb
;
3883 NAPI_GRO_CB(p
)->last
->next
= skb
;
3884 NAPI_GRO_CB(p
)->last
= skb
;
3885 __skb_header_release(skb
);
3889 NAPI_GRO_CB(p
)->count
++;
3891 p
->truesize
+= delta_truesize
;
3894 lp
->data_len
+= len
;
3895 lp
->truesize
+= delta_truesize
;
3898 NAPI_GRO_CB(skb
)->same_flow
= 1;
3901 EXPORT_SYMBOL_GPL(skb_gro_receive
);
3903 #ifdef CONFIG_SKB_EXTENSIONS
3904 #define SKB_EXT_ALIGN_VALUE 8
3905 #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE)
3907 static const u8 skb_ext_type_len
[] = {
3908 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3909 [SKB_EXT_BRIDGE_NF
] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info
),
3912 [SKB_EXT_SEC_PATH
] = SKB_EXT_CHUNKSIZEOF(struct sec_path
),
3916 static __always_inline
unsigned int skb_ext_total_length(void)
3918 return SKB_EXT_CHUNKSIZEOF(struct skb_ext
) +
3919 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3920 skb_ext_type_len
[SKB_EXT_BRIDGE_NF
] +
3923 skb_ext_type_len
[SKB_EXT_SEC_PATH
] +
3928 static void skb_extensions_init(void)
3930 BUILD_BUG_ON(SKB_EXT_NUM
>= 8);
3931 BUILD_BUG_ON(skb_ext_total_length() > 255);
3933 skbuff_ext_cache
= kmem_cache_create("skbuff_ext_cache",
3934 SKB_EXT_ALIGN_VALUE
* skb_ext_total_length(),
3936 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
3940 static void skb_extensions_init(void) {}
3943 void __init
skb_init(void)
3945 skbuff_head_cache
= kmem_cache_create_usercopy("skbuff_head_cache",
3946 sizeof(struct sk_buff
),
3948 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
3949 offsetof(struct sk_buff
, cb
),
3950 sizeof_field(struct sk_buff
, cb
),
3952 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
3953 sizeof(struct sk_buff_fclones
),
3955 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
3957 skb_extensions_init();
3961 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
,
3962 unsigned int recursion_level
)
3964 int start
= skb_headlen(skb
);
3965 int i
, copy
= start
- offset
;
3966 struct sk_buff
*frag_iter
;
3969 if (unlikely(recursion_level
>= 24))
3975 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
3977 if ((len
-= copy
) == 0)
3982 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
3985 WARN_ON(start
> offset
+ len
);
3987 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
3988 if ((copy
= end
- offset
) > 0) {
3989 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
3990 if (unlikely(elt
&& sg_is_last(&sg
[elt
- 1])))
3995 sg_set_page(&sg
[elt
], skb_frag_page(frag
), copy
,
3996 frag
->page_offset
+offset
-start
);
4005 skb_walk_frags(skb
, frag_iter
) {
4008 WARN_ON(start
> offset
+ len
);
4010 end
= start
+ frag_iter
->len
;
4011 if ((copy
= end
- offset
) > 0) {
4012 if (unlikely(elt
&& sg_is_last(&sg
[elt
- 1])))
4017 ret
= __skb_to_sgvec(frag_iter
, sg
+elt
, offset
- start
,
4018 copy
, recursion_level
+ 1);
4019 if (unlikely(ret
< 0))
4022 if ((len
-= copy
) == 0)
4033 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
4034 * @skb: Socket buffer containing the buffers to be mapped
4035 * @sg: The scatter-gather list to map into
4036 * @offset: The offset into the buffer's contents to start mapping
4037 * @len: Length of buffer space to be mapped
4039 * Fill the specified scatter-gather list with mappings/pointers into a
4040 * region of the buffer space attached to a socket buffer. Returns either
4041 * the number of scatterlist items used, or -EMSGSIZE if the contents
4044 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
4046 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
, 0);
4051 sg_mark_end(&sg
[nsg
- 1]);
4055 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
4057 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
4058 * sglist without mark the sg which contain last skb data as the end.
4059 * So the caller can mannipulate sg list as will when padding new data after
4060 * the first call without calling sg_unmark_end to expend sg list.
4062 * Scenario to use skb_to_sgvec_nomark:
4064 * 2. skb_to_sgvec_nomark(payload1)
4065 * 3. skb_to_sgvec_nomark(payload2)
4067 * This is equivalent to:
4069 * 2. skb_to_sgvec(payload1)
4071 * 4. skb_to_sgvec(payload2)
4073 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
4074 * is more preferable.
4076 int skb_to_sgvec_nomark(struct sk_buff
*skb
, struct scatterlist
*sg
,
4077 int offset
, int len
)
4079 return __skb_to_sgvec(skb
, sg
, offset
, len
, 0);
4081 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark
);
4086 * skb_cow_data - Check that a socket buffer's data buffers are writable
4087 * @skb: The socket buffer to check.
4088 * @tailbits: Amount of trailing space to be added
4089 * @trailer: Returned pointer to the skb where the @tailbits space begins
4091 * Make sure that the data buffers attached to a socket buffer are
4092 * writable. If they are not, private copies are made of the data buffers
4093 * and the socket buffer is set to use these instead.
4095 * If @tailbits is given, make sure that there is space to write @tailbits
4096 * bytes of data beyond current end of socket buffer. @trailer will be
4097 * set to point to the skb in which this space begins.
4099 * The number of scatterlist elements required to completely map the
4100 * COW'd and extended socket buffer will be returned.
4102 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
4106 struct sk_buff
*skb1
, **skb_p
;
4108 /* If skb is cloned or its head is paged, reallocate
4109 * head pulling out all the pages (pages are considered not writable
4110 * at the moment even if they are anonymous).
4112 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
4113 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
4116 /* Easy case. Most of packets will go this way. */
4117 if (!skb_has_frag_list(skb
)) {
4118 /* A little of trouble, not enough of space for trailer.
4119 * This should not happen, when stack is tuned to generate
4120 * good frames. OK, on miss we reallocate and reserve even more
4121 * space, 128 bytes is fair. */
4123 if (skb_tailroom(skb
) < tailbits
&&
4124 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
4132 /* Misery. We are in troubles, going to mincer fragments... */
4135 skb_p
= &skb_shinfo(skb
)->frag_list
;
4138 while ((skb1
= *skb_p
) != NULL
) {
4141 /* The fragment is partially pulled by someone,
4142 * this can happen on input. Copy it and everything
4145 if (skb_shared(skb1
))
4148 /* If the skb is the last, worry about trailer. */
4150 if (skb1
->next
== NULL
&& tailbits
) {
4151 if (skb_shinfo(skb1
)->nr_frags
||
4152 skb_has_frag_list(skb1
) ||
4153 skb_tailroom(skb1
) < tailbits
)
4154 ntail
= tailbits
+ 128;
4160 skb_shinfo(skb1
)->nr_frags
||
4161 skb_has_frag_list(skb1
)) {
4162 struct sk_buff
*skb2
;
4164 /* Fuck, we are miserable poor guys... */
4166 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
4168 skb2
= skb_copy_expand(skb1
,
4172 if (unlikely(skb2
== NULL
))
4176 skb_set_owner_w(skb2
, skb1
->sk
);
4178 /* Looking around. Are we still alive?
4179 * OK, link new skb, drop old one */
4181 skb2
->next
= skb1
->next
;
4188 skb_p
= &skb1
->next
;
4193 EXPORT_SYMBOL_GPL(skb_cow_data
);
4195 static void sock_rmem_free(struct sk_buff
*skb
)
4197 struct sock
*sk
= skb
->sk
;
4199 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
4202 static void skb_set_err_queue(struct sk_buff
*skb
)
4204 /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
4205 * So, it is safe to (mis)use it to mark skbs on the error queue.
4207 skb
->pkt_type
= PACKET_OUTGOING
;
4208 BUILD_BUG_ON(PACKET_OUTGOING
== 0);
4212 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
4214 int sock_queue_err_skb(struct sock
*sk
, struct sk_buff
*skb
)
4216 if (atomic_read(&sk
->sk_rmem_alloc
) + skb
->truesize
>=
4217 (unsigned int)sk
->sk_rcvbuf
)
4222 skb
->destructor
= sock_rmem_free
;
4223 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
4224 skb_set_err_queue(skb
);
4226 /* before exiting rcu section, make sure dst is refcounted */
4229 skb_queue_tail(&sk
->sk_error_queue
, skb
);
4230 if (!sock_flag(sk
, SOCK_DEAD
))
4231 sk
->sk_error_report(sk
);
4234 EXPORT_SYMBOL(sock_queue_err_skb
);
4236 static bool is_icmp_err_skb(const struct sk_buff
*skb
)
4238 return skb
&& (SKB_EXT_ERR(skb
)->ee
.ee_origin
== SO_EE_ORIGIN_ICMP
||
4239 SKB_EXT_ERR(skb
)->ee
.ee_origin
== SO_EE_ORIGIN_ICMP6
);
4242 struct sk_buff
*sock_dequeue_err_skb(struct sock
*sk
)
4244 struct sk_buff_head
*q
= &sk
->sk_error_queue
;
4245 struct sk_buff
*skb
, *skb_next
= NULL
;
4246 bool icmp_next
= false;
4247 unsigned long flags
;
4249 spin_lock_irqsave(&q
->lock
, flags
);
4250 skb
= __skb_dequeue(q
);
4251 if (skb
&& (skb_next
= skb_peek(q
))) {
4252 icmp_next
= is_icmp_err_skb(skb_next
);
4254 sk
->sk_err
= SKB_EXT_ERR(skb_next
)->ee
.ee_origin
;
4256 spin_unlock_irqrestore(&q
->lock
, flags
);
4258 if (is_icmp_err_skb(skb
) && !icmp_next
)
4262 sk
->sk_error_report(sk
);
4266 EXPORT_SYMBOL(sock_dequeue_err_skb
);
4269 * skb_clone_sk - create clone of skb, and take reference to socket
4270 * @skb: the skb to clone
4272 * This function creates a clone of a buffer that holds a reference on
4273 * sk_refcnt. Buffers created via this function are meant to be
4274 * returned using sock_queue_err_skb, or free via kfree_skb.
4276 * When passing buffers allocated with this function to sock_queue_err_skb
4277 * it is necessary to wrap the call with sock_hold/sock_put in order to
4278 * prevent the socket from being released prior to being enqueued on
4279 * the sk_error_queue.
4281 struct sk_buff
*skb_clone_sk(struct sk_buff
*skb
)
4283 struct sock
*sk
= skb
->sk
;
4284 struct sk_buff
*clone
;
4286 if (!sk
|| !refcount_inc_not_zero(&sk
->sk_refcnt
))
4289 clone
= skb_clone(skb
, GFP_ATOMIC
);
4296 clone
->destructor
= sock_efree
;
4300 EXPORT_SYMBOL(skb_clone_sk
);
4302 static void __skb_complete_tx_timestamp(struct sk_buff
*skb
,
4307 struct sock_exterr_skb
*serr
;
4310 BUILD_BUG_ON(sizeof(struct sock_exterr_skb
) > sizeof(skb
->cb
));
4312 serr
= SKB_EXT_ERR(skb
);
4313 memset(serr
, 0, sizeof(*serr
));
4314 serr
->ee
.ee_errno
= ENOMSG
;
4315 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TIMESTAMPING
;
4316 serr
->ee
.ee_info
= tstype
;
4317 serr
->opt_stats
= opt_stats
;
4318 serr
->header
.h4
.iif
= skb
->dev
? skb
->dev
->ifindex
: 0;
4319 if (sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_ID
) {
4320 serr
->ee
.ee_data
= skb_shinfo(skb
)->tskey
;
4321 if (sk
->sk_protocol
== IPPROTO_TCP
&&
4322 sk
->sk_type
== SOCK_STREAM
)
4323 serr
->ee
.ee_data
-= sk
->sk_tskey
;
4326 err
= sock_queue_err_skb(sk
, skb
);
4332 static bool skb_may_tx_timestamp(struct sock
*sk
, bool tsonly
)
4336 if (likely(sysctl_tstamp_allow_data
|| tsonly
))
4339 read_lock_bh(&sk
->sk_callback_lock
);
4340 ret
= sk
->sk_socket
&& sk
->sk_socket
->file
&&
4341 file_ns_capable(sk
->sk_socket
->file
, &init_user_ns
, CAP_NET_RAW
);
4342 read_unlock_bh(&sk
->sk_callback_lock
);
4346 void skb_complete_tx_timestamp(struct sk_buff
*skb
,
4347 struct skb_shared_hwtstamps
*hwtstamps
)
4349 struct sock
*sk
= skb
->sk
;
4351 if (!skb_may_tx_timestamp(sk
, false))
4354 /* Take a reference to prevent skb_orphan() from freeing the socket,
4355 * but only if the socket refcount is not zero.
4357 if (likely(refcount_inc_not_zero(&sk
->sk_refcnt
))) {
4358 *skb_hwtstamps(skb
) = *hwtstamps
;
4359 __skb_complete_tx_timestamp(skb
, sk
, SCM_TSTAMP_SND
, false);
4367 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp
);
4369 void __skb_tstamp_tx(struct sk_buff
*orig_skb
,
4370 struct skb_shared_hwtstamps
*hwtstamps
,
4371 struct sock
*sk
, int tstype
)
4373 struct sk_buff
*skb
;
4374 bool tsonly
, opt_stats
= false;
4379 if (!hwtstamps
&& !(sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_TX_SWHW
) &&
4380 skb_shinfo(orig_skb
)->tx_flags
& SKBTX_IN_PROGRESS
)
4383 tsonly
= sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_TSONLY
;
4384 if (!skb_may_tx_timestamp(sk
, tsonly
))
4389 if ((sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_STATS
) &&
4390 sk
->sk_protocol
== IPPROTO_TCP
&&
4391 sk
->sk_type
== SOCK_STREAM
) {
4392 skb
= tcp_get_timestamping_opt_stats(sk
);
4396 skb
= alloc_skb(0, GFP_ATOMIC
);
4398 skb
= skb_clone(orig_skb
, GFP_ATOMIC
);
4404 skb_shinfo(skb
)->tx_flags
|= skb_shinfo(orig_skb
)->tx_flags
&
4406 skb_shinfo(skb
)->tskey
= skb_shinfo(orig_skb
)->tskey
;
4410 *skb_hwtstamps(skb
) = *hwtstamps
;
4412 skb
->tstamp
= ktime_get_real();
4414 __skb_complete_tx_timestamp(skb
, sk
, tstype
, opt_stats
);
4416 EXPORT_SYMBOL_GPL(__skb_tstamp_tx
);
4418 void skb_tstamp_tx(struct sk_buff
*orig_skb
,
4419 struct skb_shared_hwtstamps
*hwtstamps
)
4421 return __skb_tstamp_tx(orig_skb
, hwtstamps
, orig_skb
->sk
,
4424 EXPORT_SYMBOL_GPL(skb_tstamp_tx
);
4426 void skb_complete_wifi_ack(struct sk_buff
*skb
, bool acked
)
4428 struct sock
*sk
= skb
->sk
;
4429 struct sock_exterr_skb
*serr
;
4432 skb
->wifi_acked_valid
= 1;
4433 skb
->wifi_acked
= acked
;
4435 serr
= SKB_EXT_ERR(skb
);
4436 memset(serr
, 0, sizeof(*serr
));
4437 serr
->ee
.ee_errno
= ENOMSG
;
4438 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TXSTATUS
;
4440 /* Take a reference to prevent skb_orphan() from freeing the socket,
4441 * but only if the socket refcount is not zero.
4443 if (likely(refcount_inc_not_zero(&sk
->sk_refcnt
))) {
4444 err
= sock_queue_err_skb(sk
, skb
);
4450 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack
);
4453 * skb_partial_csum_set - set up and verify partial csum values for packet
4454 * @skb: the skb to set
4455 * @start: the number of bytes after skb->data to start checksumming.
4456 * @off: the offset from start to place the checksum.
4458 * For untrusted partially-checksummed packets, we need to make sure the values
4459 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
4461 * This function checks and sets those values and skb->ip_summed: if this
4462 * returns false you should drop the packet.
4464 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
4466 u32 csum_end
= (u32
)start
+ (u32
)off
+ sizeof(__sum16
);
4467 u32 csum_start
= skb_headroom(skb
) + (u32
)start
;
4469 if (unlikely(csum_start
> U16_MAX
|| csum_end
> skb_headlen(skb
))) {
4470 net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n",
4471 start
, off
, skb_headroom(skb
), skb_headlen(skb
));
4474 skb
->ip_summed
= CHECKSUM_PARTIAL
;
4475 skb
->csum_start
= csum_start
;
4476 skb
->csum_offset
= off
;
4477 skb_set_transport_header(skb
, start
);
4480 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);
4482 static int skb_maybe_pull_tail(struct sk_buff
*skb
, unsigned int len
,
4485 if (skb_headlen(skb
) >= len
)
4488 /* If we need to pullup then pullup to the max, so we
4489 * won't need to do it again.
4494 if (__pskb_pull_tail(skb
, max
- skb_headlen(skb
)) == NULL
)
4497 if (skb_headlen(skb
) < len
)
4503 #define MAX_TCP_HDR_LEN (15 * 4)
4505 static __sum16
*skb_checksum_setup_ip(struct sk_buff
*skb
,
4506 typeof(IPPROTO_IP
) proto
,
4513 err
= skb_maybe_pull_tail(skb
, off
+ sizeof(struct tcphdr
),
4514 off
+ MAX_TCP_HDR_LEN
);
4515 if (!err
&& !skb_partial_csum_set(skb
, off
,
4516 offsetof(struct tcphdr
,
4519 return err
? ERR_PTR(err
) : &tcp_hdr(skb
)->check
;
4522 err
= skb_maybe_pull_tail(skb
, off
+ sizeof(struct udphdr
),
4523 off
+ sizeof(struct udphdr
));
4524 if (!err
&& !skb_partial_csum_set(skb
, off
,
4525 offsetof(struct udphdr
,
4528 return err
? ERR_PTR(err
) : &udp_hdr(skb
)->check
;
4531 return ERR_PTR(-EPROTO
);
4534 /* This value should be large enough to cover a tagged ethernet header plus
4535 * maximally sized IP and TCP or UDP headers.
4537 #define MAX_IP_HDR_LEN 128
4539 static int skb_checksum_setup_ipv4(struct sk_buff
*skb
, bool recalculate
)
4548 err
= skb_maybe_pull_tail(skb
,
4549 sizeof(struct iphdr
),
4554 if (ip_hdr(skb
)->frag_off
& htons(IP_OFFSET
| IP_MF
))
4557 off
= ip_hdrlen(skb
);
4564 csum
= skb_checksum_setup_ip(skb
, ip_hdr(skb
)->protocol
, off
);
4566 return PTR_ERR(csum
);
4569 *csum
= ~csum_tcpudp_magic(ip_hdr(skb
)->saddr
,
4572 ip_hdr(skb
)->protocol
, 0);
4579 /* This value should be large enough to cover a tagged ethernet header plus
4580 * an IPv6 header, all options, and a maximal TCP or UDP header.
4582 #define MAX_IPV6_HDR_LEN 256
4584 #define OPT_HDR(type, skb, off) \
4585 (type *)(skb_network_header(skb) + (off))
4587 static int skb_checksum_setup_ipv6(struct sk_buff
*skb
, bool recalculate
)
4600 off
= sizeof(struct ipv6hdr
);
4602 err
= skb_maybe_pull_tail(skb
, off
, MAX_IPV6_HDR_LEN
);
4606 nexthdr
= ipv6_hdr(skb
)->nexthdr
;
4608 len
= sizeof(struct ipv6hdr
) + ntohs(ipv6_hdr(skb
)->payload_len
);
4609 while (off
<= len
&& !done
) {
4611 case IPPROTO_DSTOPTS
:
4612 case IPPROTO_HOPOPTS
:
4613 case IPPROTO_ROUTING
: {
4614 struct ipv6_opt_hdr
*hp
;
4616 err
= skb_maybe_pull_tail(skb
,
4618 sizeof(struct ipv6_opt_hdr
),
4623 hp
= OPT_HDR(struct ipv6_opt_hdr
, skb
, off
);
4624 nexthdr
= hp
->nexthdr
;
4625 off
+= ipv6_optlen(hp
);
4629 struct ip_auth_hdr
*hp
;
4631 err
= skb_maybe_pull_tail(skb
,
4633 sizeof(struct ip_auth_hdr
),
4638 hp
= OPT_HDR(struct ip_auth_hdr
, skb
, off
);
4639 nexthdr
= hp
->nexthdr
;
4640 off
+= ipv6_authlen(hp
);
4643 case IPPROTO_FRAGMENT
: {
4644 struct frag_hdr
*hp
;
4646 err
= skb_maybe_pull_tail(skb
,
4648 sizeof(struct frag_hdr
),
4653 hp
= OPT_HDR(struct frag_hdr
, skb
, off
);
4655 if (hp
->frag_off
& htons(IP6_OFFSET
| IP6_MF
))
4658 nexthdr
= hp
->nexthdr
;
4659 off
+= sizeof(struct frag_hdr
);
4670 if (!done
|| fragment
)
4673 csum
= skb_checksum_setup_ip(skb
, nexthdr
, off
);
4675 return PTR_ERR(csum
);
4678 *csum
= ~csum_ipv6_magic(&ipv6_hdr(skb
)->saddr
,
4679 &ipv6_hdr(skb
)->daddr
,
4680 skb
->len
- off
, nexthdr
, 0);
4688 * skb_checksum_setup - set up partial checksum offset
4689 * @skb: the skb to set up
4690 * @recalculate: if true the pseudo-header checksum will be recalculated
4692 int skb_checksum_setup(struct sk_buff
*skb
, bool recalculate
)
4696 switch (skb
->protocol
) {
4697 case htons(ETH_P_IP
):
4698 err
= skb_checksum_setup_ipv4(skb
, recalculate
);
4701 case htons(ETH_P_IPV6
):
4702 err
= skb_checksum_setup_ipv6(skb
, recalculate
);
4712 EXPORT_SYMBOL(skb_checksum_setup
);
4715 * skb_checksum_maybe_trim - maybe trims the given skb
4716 * @skb: the skb to check
4717 * @transport_len: the data length beyond the network header
4719 * Checks whether the given skb has data beyond the given transport length.
4720 * If so, returns a cloned skb trimmed to this transport length.
4721 * Otherwise returns the provided skb. Returns NULL in error cases
4722 * (e.g. transport_len exceeds skb length or out-of-memory).
4724 * Caller needs to set the skb transport header and free any returned skb if it
4725 * differs from the provided skb.
4727 static struct sk_buff
*skb_checksum_maybe_trim(struct sk_buff
*skb
,
4728 unsigned int transport_len
)
4730 struct sk_buff
*skb_chk
;
4731 unsigned int len
= skb_transport_offset(skb
) + transport_len
;
4736 else if (skb
->len
== len
)
4739 skb_chk
= skb_clone(skb
, GFP_ATOMIC
);
4743 ret
= pskb_trim_rcsum(skb_chk
, len
);
4753 * skb_checksum_trimmed - validate checksum of an skb
4754 * @skb: the skb to check
4755 * @transport_len: the data length beyond the network header
4756 * @skb_chkf: checksum function to use
4758 * Applies the given checksum function skb_chkf to the provided skb.
4759 * Returns a checked and maybe trimmed skb. Returns NULL on error.
4761 * If the skb has data beyond the given transport length, then a
4762 * trimmed & cloned skb is checked and returned.
4764 * Caller needs to set the skb transport header and free any returned skb if it
4765 * differs from the provided skb.
4767 struct sk_buff
*skb_checksum_trimmed(struct sk_buff
*skb
,
4768 unsigned int transport_len
,
4769 __sum16(*skb_chkf
)(struct sk_buff
*skb
))
4771 struct sk_buff
*skb_chk
;
4772 unsigned int offset
= skb_transport_offset(skb
);
4775 skb_chk
= skb_checksum_maybe_trim(skb
, transport_len
);
4779 if (!pskb_may_pull(skb_chk
, offset
))
4782 skb_pull_rcsum(skb_chk
, offset
);
4783 ret
= skb_chkf(skb_chk
);
4784 skb_push_rcsum(skb_chk
, offset
);
4792 if (skb_chk
&& skb_chk
!= skb
)
4798 EXPORT_SYMBOL(skb_checksum_trimmed
);
4800 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
4802 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
4805 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);
4807 void kfree_skb_partial(struct sk_buff
*skb
, bool head_stolen
)
4810 skb_release_head_state(skb
);
4811 kmem_cache_free(skbuff_head_cache
, skb
);
4816 EXPORT_SYMBOL(kfree_skb_partial
);
4819 * skb_try_coalesce - try to merge skb to prior one
4821 * @from: buffer to add
4822 * @fragstolen: pointer to boolean
4823 * @delta_truesize: how much more was allocated than was requested
4825 bool skb_try_coalesce(struct sk_buff
*to
, struct sk_buff
*from
,
4826 bool *fragstolen
, int *delta_truesize
)
4828 struct skb_shared_info
*to_shinfo
, *from_shinfo
;
4829 int i
, delta
, len
= from
->len
;
4831 *fragstolen
= false;
4836 if (len
<= skb_tailroom(to
)) {
4838 BUG_ON(skb_copy_bits(from
, 0, skb_put(to
, len
), len
));
4839 *delta_truesize
= 0;
4843 to_shinfo
= skb_shinfo(to
);
4844 from_shinfo
= skb_shinfo(from
);
4845 if (to_shinfo
->frag_list
|| from_shinfo
->frag_list
)
4847 if (skb_zcopy(to
) || skb_zcopy(from
))
4850 if (skb_headlen(from
) != 0) {
4852 unsigned int offset
;
4854 if (to_shinfo
->nr_frags
+
4855 from_shinfo
->nr_frags
>= MAX_SKB_FRAGS
)
4858 if (skb_head_is_locked(from
))
4861 delta
= from
->truesize
- SKB_DATA_ALIGN(sizeof(struct sk_buff
));
4863 page
= virt_to_head_page(from
->head
);
4864 offset
= from
->data
- (unsigned char *)page_address(page
);
4866 skb_fill_page_desc(to
, to_shinfo
->nr_frags
,
4867 page
, offset
, skb_headlen(from
));
4870 if (to_shinfo
->nr_frags
+
4871 from_shinfo
->nr_frags
> MAX_SKB_FRAGS
)
4874 delta
= from
->truesize
- SKB_TRUESIZE(skb_end_offset(from
));
4877 WARN_ON_ONCE(delta
< len
);
4879 memcpy(to_shinfo
->frags
+ to_shinfo
->nr_frags
,
4881 from_shinfo
->nr_frags
* sizeof(skb_frag_t
));
4882 to_shinfo
->nr_frags
+= from_shinfo
->nr_frags
;
4884 if (!skb_cloned(from
))
4885 from_shinfo
->nr_frags
= 0;
4887 /* if the skb is not cloned this does nothing
4888 * since we set nr_frags to 0.
4890 for (i
= 0; i
< from_shinfo
->nr_frags
; i
++)
4891 __skb_frag_ref(&from_shinfo
->frags
[i
]);
4893 to
->truesize
+= delta
;
4895 to
->data_len
+= len
;
4897 *delta_truesize
= delta
;
4900 EXPORT_SYMBOL(skb_try_coalesce
);
4903 * skb_scrub_packet - scrub an skb
4905 * @skb: buffer to clean
4906 * @xnet: packet is crossing netns
4908 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
4909 * into/from a tunnel. Some information have to be cleared during these
4911 * skb_scrub_packet can also be used to clean a skb before injecting it in
4912 * another namespace (@xnet == true). We have to clear all information in the
4913 * skb that could impact namespace isolation.
4915 void skb_scrub_packet(struct sk_buff
*skb
, bool xnet
)
4917 skb
->pkt_type
= PACKET_HOST
;
4923 nf_reset_trace(skb
);
4925 #ifdef CONFIG_NET_SWITCHDEV
4926 skb
->offload_fwd_mark
= 0;
4927 skb
->offload_l3_fwd_mark
= 0;
4937 EXPORT_SYMBOL_GPL(skb_scrub_packet
);
4940 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
4944 * skb_gso_transport_seglen is used to determine the real size of the
4945 * individual segments, including Layer4 headers (TCP/UDP).
4947 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
4949 static unsigned int skb_gso_transport_seglen(const struct sk_buff
*skb
)
4951 const struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
4952 unsigned int thlen
= 0;
4954 if (skb
->encapsulation
) {
4955 thlen
= skb_inner_transport_header(skb
) -
4956 skb_transport_header(skb
);
4958 if (likely(shinfo
->gso_type
& (SKB_GSO_TCPV4
| SKB_GSO_TCPV6
)))
4959 thlen
+= inner_tcp_hdrlen(skb
);
4960 } else if (likely(shinfo
->gso_type
& (SKB_GSO_TCPV4
| SKB_GSO_TCPV6
))) {
4961 thlen
= tcp_hdrlen(skb
);
4962 } else if (unlikely(skb_is_gso_sctp(skb
))) {
4963 thlen
= sizeof(struct sctphdr
);
4964 } else if (shinfo
->gso_type
& SKB_GSO_UDP_L4
) {
4965 thlen
= sizeof(struct udphdr
);
4967 /* UFO sets gso_size to the size of the fragmentation
4968 * payload, i.e. the size of the L4 (UDP) header is already
4971 return thlen
+ shinfo
->gso_size
;
4975 * skb_gso_network_seglen - Return length of individual segments of a gso packet
4979 * skb_gso_network_seglen is used to determine the real size of the
4980 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
4982 * The MAC/L2 header is not accounted for.
4984 static unsigned int skb_gso_network_seglen(const struct sk_buff
*skb
)
4986 unsigned int hdr_len
= skb_transport_header(skb
) -
4987 skb_network_header(skb
);
4989 return hdr_len
+ skb_gso_transport_seglen(skb
);
4993 * skb_gso_mac_seglen - Return length of individual segments of a gso packet
4997 * skb_gso_mac_seglen is used to determine the real size of the
4998 * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4
4999 * headers (TCP/UDP).
5001 static unsigned int skb_gso_mac_seglen(const struct sk_buff
*skb
)
5003 unsigned int hdr_len
= skb_transport_header(skb
) - skb_mac_header(skb
);
5005 return hdr_len
+ skb_gso_transport_seglen(skb
);
5009 * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS
5011 * There are a couple of instances where we have a GSO skb, and we
5012 * want to determine what size it would be after it is segmented.
5014 * We might want to check:
5015 * - L3+L4+payload size (e.g. IP forwarding)
5016 * - L2+L3+L4+payload size (e.g. sanity check before passing to driver)
5018 * This is a helper to do that correctly considering GSO_BY_FRAGS.
5022 * @seg_len: The segmented length (from skb_gso_*_seglen). In the
5023 * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS].
5025 * @max_len: The maximum permissible length.
5027 * Returns true if the segmented length <= max length.
5029 static inline bool skb_gso_size_check(const struct sk_buff
*skb
,
5030 unsigned int seg_len
,
5031 unsigned int max_len
) {
5032 const struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
5033 const struct sk_buff
*iter
;
5035 if (shinfo
->gso_size
!= GSO_BY_FRAGS
)
5036 return seg_len
<= max_len
;
5038 /* Undo this so we can re-use header sizes */
5039 seg_len
-= GSO_BY_FRAGS
;
5041 skb_walk_frags(skb
, iter
) {
5042 if (seg_len
+ skb_headlen(iter
) > max_len
)
5050 * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU?
5053 * @mtu: MTU to validate against
5055 * skb_gso_validate_network_len validates if a given skb will fit a
5056 * wanted MTU once split. It considers L3 headers, L4 headers, and the
5059 bool skb_gso_validate_network_len(const struct sk_buff
*skb
, unsigned int mtu
)
5061 return skb_gso_size_check(skb
, skb_gso_network_seglen(skb
), mtu
);
5063 EXPORT_SYMBOL_GPL(skb_gso_validate_network_len
);
5066 * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length?
5069 * @len: length to validate against
5071 * skb_gso_validate_mac_len validates if a given skb will fit a wanted
5072 * length once split, including L2, L3 and L4 headers and the payload.
5074 bool skb_gso_validate_mac_len(const struct sk_buff
*skb
, unsigned int len
)
5076 return skb_gso_size_check(skb
, skb_gso_mac_seglen(skb
), len
);
5078 EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len
);
5080 static struct sk_buff
*skb_reorder_vlan_header(struct sk_buff
*skb
)
5084 if (skb_cow(skb
, skb_headroom(skb
)) < 0) {
5089 mac_len
= skb
->data
- skb_mac_header(skb
);
5090 if (likely(mac_len
> VLAN_HLEN
+ ETH_TLEN
)) {
5091 memmove(skb_mac_header(skb
) + VLAN_HLEN
, skb_mac_header(skb
),
5092 mac_len
- VLAN_HLEN
- ETH_TLEN
);
5094 skb
->mac_header
+= VLAN_HLEN
;
5098 struct sk_buff
*skb_vlan_untag(struct sk_buff
*skb
)
5100 struct vlan_hdr
*vhdr
;
5103 if (unlikely(skb_vlan_tag_present(skb
))) {
5104 /* vlan_tci is already set-up so leave this for another time */
5108 skb
= skb_share_check(skb
, GFP_ATOMIC
);
5112 if (unlikely(!pskb_may_pull(skb
, VLAN_HLEN
)))
5115 vhdr
= (struct vlan_hdr
*)skb
->data
;
5116 vlan_tci
= ntohs(vhdr
->h_vlan_TCI
);
5117 __vlan_hwaccel_put_tag(skb
, skb
->protocol
, vlan_tci
);
5119 skb_pull_rcsum(skb
, VLAN_HLEN
);
5120 vlan_set_encap_proto(skb
, vhdr
);
5122 skb
= skb_reorder_vlan_header(skb
);
5126 skb_reset_network_header(skb
);
5127 skb_reset_transport_header(skb
);
5128 skb_reset_mac_len(skb
);
5136 EXPORT_SYMBOL(skb_vlan_untag
);
5138 int skb_ensure_writable(struct sk_buff
*skb
, int write_len
)
5140 if (!pskb_may_pull(skb
, write_len
))
5143 if (!skb_cloned(skb
) || skb_clone_writable(skb
, write_len
))
5146 return pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
5148 EXPORT_SYMBOL(skb_ensure_writable
);
5150 /* remove VLAN header from packet and update csum accordingly.
5151 * expects a non skb_vlan_tag_present skb with a vlan tag payload
5153 int __skb_vlan_pop(struct sk_buff
*skb
, u16
*vlan_tci
)
5155 struct vlan_hdr
*vhdr
;
5156 int offset
= skb
->data
- skb_mac_header(skb
);
5159 if (WARN_ONCE(offset
,
5160 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
5165 err
= skb_ensure_writable(skb
, VLAN_ETH_HLEN
);
5169 skb_postpull_rcsum(skb
, skb
->data
+ (2 * ETH_ALEN
), VLAN_HLEN
);
5171 vhdr
= (struct vlan_hdr
*)(skb
->data
+ ETH_HLEN
);
5172 *vlan_tci
= ntohs(vhdr
->h_vlan_TCI
);
5174 memmove(skb
->data
+ VLAN_HLEN
, skb
->data
, 2 * ETH_ALEN
);
5175 __skb_pull(skb
, VLAN_HLEN
);
5177 vlan_set_encap_proto(skb
, vhdr
);
5178 skb
->mac_header
+= VLAN_HLEN
;
5180 if (skb_network_offset(skb
) < ETH_HLEN
)
5181 skb_set_network_header(skb
, ETH_HLEN
);
5183 skb_reset_mac_len(skb
);
5187 EXPORT_SYMBOL(__skb_vlan_pop
);
5189 /* Pop a vlan tag either from hwaccel or from payload.
5190 * Expects skb->data at mac header.
5192 int skb_vlan_pop(struct sk_buff
*skb
)
5198 if (likely(skb_vlan_tag_present(skb
))) {
5199 __vlan_hwaccel_clear_tag(skb
);
5201 if (unlikely(!eth_type_vlan(skb
->protocol
)))
5204 err
= __skb_vlan_pop(skb
, &vlan_tci
);
5208 /* move next vlan tag to hw accel tag */
5209 if (likely(!eth_type_vlan(skb
->protocol
)))
5212 vlan_proto
= skb
->protocol
;
5213 err
= __skb_vlan_pop(skb
, &vlan_tci
);
5217 __vlan_hwaccel_put_tag(skb
, vlan_proto
, vlan_tci
);
5220 EXPORT_SYMBOL(skb_vlan_pop
);
5222 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
5223 * Expects skb->data at mac header.
5225 int skb_vlan_push(struct sk_buff
*skb
, __be16 vlan_proto
, u16 vlan_tci
)
5227 if (skb_vlan_tag_present(skb
)) {
5228 int offset
= skb
->data
- skb_mac_header(skb
);
5231 if (WARN_ONCE(offset
,
5232 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
5237 err
= __vlan_insert_tag(skb
, skb
->vlan_proto
,
5238 skb_vlan_tag_get(skb
));
5242 skb
->protocol
= skb
->vlan_proto
;
5243 skb
->mac_len
+= VLAN_HLEN
;
5245 skb_postpush_rcsum(skb
, skb
->data
+ (2 * ETH_ALEN
), VLAN_HLEN
);
5247 __vlan_hwaccel_put_tag(skb
, vlan_proto
, vlan_tci
);
5250 EXPORT_SYMBOL(skb_vlan_push
);
5253 * alloc_skb_with_frags - allocate skb with page frags
5255 * @header_len: size of linear part
5256 * @data_len: needed length in frags
5257 * @max_page_order: max page order desired.
5258 * @errcode: pointer to error code if any
5259 * @gfp_mask: allocation mask
5261 * This can be used to allocate a paged skb, given a maximal order for frags.
5263 struct sk_buff
*alloc_skb_with_frags(unsigned long header_len
,
5264 unsigned long data_len
,
5269 int npages
= (data_len
+ (PAGE_SIZE
- 1)) >> PAGE_SHIFT
;
5270 unsigned long chunk
;
5271 struct sk_buff
*skb
;
5276 *errcode
= -EMSGSIZE
;
5277 /* Note this test could be relaxed, if we succeed to allocate
5278 * high order pages...
5280 if (npages
> MAX_SKB_FRAGS
)
5283 gfp_head
= gfp_mask
;
5284 if (gfp_head
& __GFP_DIRECT_RECLAIM
)
5285 gfp_head
|= __GFP_RETRY_MAYFAIL
;
5287 *errcode
= -ENOBUFS
;
5288 skb
= alloc_skb(header_len
, gfp_head
);
5292 skb
->truesize
+= npages
<< PAGE_SHIFT
;
5294 for (i
= 0; npages
> 0; i
++) {
5295 int order
= max_page_order
;
5298 if (npages
>= 1 << order
) {
5299 page
= alloc_pages((gfp_mask
& ~__GFP_DIRECT_RECLAIM
) |
5305 /* Do not retry other high order allocations */
5311 page
= alloc_page(gfp_mask
);
5315 chunk
= min_t(unsigned long, data_len
,
5316 PAGE_SIZE
<< order
);
5317 skb_fill_page_desc(skb
, i
, page
, 0, chunk
);
5319 npages
-= 1 << order
;
5327 EXPORT_SYMBOL(alloc_skb_with_frags
);
5329 /* carve out the first off bytes from skb when off < headlen */
5330 static int pskb_carve_inside_header(struct sk_buff
*skb
, const u32 off
,
5331 const int headlen
, gfp_t gfp_mask
)
5334 int size
= skb_end_offset(skb
);
5335 int new_hlen
= headlen
- off
;
5338 size
= SKB_DATA_ALIGN(size
);
5340 if (skb_pfmemalloc(skb
))
5341 gfp_mask
|= __GFP_MEMALLOC
;
5342 data
= kmalloc_reserve(size
+
5343 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)),
5344 gfp_mask
, NUMA_NO_NODE
, NULL
);
5348 size
= SKB_WITH_OVERHEAD(ksize(data
));
5350 /* Copy real data, and all frags */
5351 skb_copy_from_linear_data_offset(skb
, off
, data
, new_hlen
);
5354 memcpy((struct skb_shared_info
*)(data
+ size
),
5356 offsetof(struct skb_shared_info
,
5357 frags
[skb_shinfo(skb
)->nr_frags
]));
5358 if (skb_cloned(skb
)) {
5359 /* drop the old head gracefully */
5360 if (skb_orphan_frags(skb
, gfp_mask
)) {
5364 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
5365 skb_frag_ref(skb
, i
);
5366 if (skb_has_frag_list(skb
))
5367 skb_clone_fraglist(skb
);
5368 skb_release_data(skb
);
5370 /* we can reuse existing recount- all we did was
5379 #ifdef NET_SKBUFF_DATA_USES_OFFSET
5382 skb
->end
= skb
->head
+ size
;
5384 skb_set_tail_pointer(skb
, skb_headlen(skb
));
5385 skb_headers_offset_update(skb
, 0);
5389 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
5394 static int pskb_carve(struct sk_buff
*skb
, const u32 off
, gfp_t gfp
);
5396 /* carve out the first eat bytes from skb's frag_list. May recurse into
5399 static int pskb_carve_frag_list(struct sk_buff
*skb
,
5400 struct skb_shared_info
*shinfo
, int eat
,
5403 struct sk_buff
*list
= shinfo
->frag_list
;
5404 struct sk_buff
*clone
= NULL
;
5405 struct sk_buff
*insp
= NULL
;
5409 pr_err("Not enough bytes to eat. Want %d\n", eat
);
5412 if (list
->len
<= eat
) {
5413 /* Eaten as whole. */
5418 /* Eaten partially. */
5419 if (skb_shared(list
)) {
5420 clone
= skb_clone(list
, gfp_mask
);
5426 /* This may be pulled without problems. */
5429 if (pskb_carve(list
, eat
, gfp_mask
) < 0) {
5437 /* Free pulled out fragments. */
5438 while ((list
= shinfo
->frag_list
) != insp
) {
5439 shinfo
->frag_list
= list
->next
;
5442 /* And insert new clone at head. */
5445 shinfo
->frag_list
= clone
;
5450 /* carve off first len bytes from skb. Split line (off) is in the
5451 * non-linear part of skb
5453 static int pskb_carve_inside_nonlinear(struct sk_buff
*skb
, const u32 off
,
5454 int pos
, gfp_t gfp_mask
)
5457 int size
= skb_end_offset(skb
);
5459 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
5460 struct skb_shared_info
*shinfo
;
5462 size
= SKB_DATA_ALIGN(size
);
5464 if (skb_pfmemalloc(skb
))
5465 gfp_mask
|= __GFP_MEMALLOC
;
5466 data
= kmalloc_reserve(size
+
5467 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)),
5468 gfp_mask
, NUMA_NO_NODE
, NULL
);
5472 size
= SKB_WITH_OVERHEAD(ksize(data
));
5474 memcpy((struct skb_shared_info
*)(data
+ size
),
5475 skb_shinfo(skb
), offsetof(struct skb_shared_info
,
5476 frags
[skb_shinfo(skb
)->nr_frags
]));
5477 if (skb_orphan_frags(skb
, gfp_mask
)) {
5481 shinfo
= (struct skb_shared_info
*)(data
+ size
);
5482 for (i
= 0; i
< nfrags
; i
++) {
5483 int fsize
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
5485 if (pos
+ fsize
> off
) {
5486 shinfo
->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
5490 * We have two variants in this case:
5491 * 1. Move all the frag to the second
5492 * part, if it is possible. F.e.
5493 * this approach is mandatory for TUX,
5494 * where splitting is expensive.
5495 * 2. Split is accurately. We make this.
5497 shinfo
->frags
[0].page_offset
+= off
- pos
;
5498 skb_frag_size_sub(&shinfo
->frags
[0], off
- pos
);
5500 skb_frag_ref(skb
, i
);
5505 shinfo
->nr_frags
= k
;
5506 if (skb_has_frag_list(skb
))
5507 skb_clone_fraglist(skb
);
5510 /* split line is in frag list */
5511 pskb_carve_frag_list(skb
, shinfo
, off
- pos
, gfp_mask
);
5513 skb_release_data(skb
);
5518 #ifdef NET_SKBUFF_DATA_USES_OFFSET
5521 skb
->end
= skb
->head
+ size
;
5523 skb_reset_tail_pointer(skb
);
5524 skb_headers_offset_update(skb
, 0);
5529 skb
->data_len
= skb
->len
;
5530 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
5534 /* remove len bytes from the beginning of the skb */
5535 static int pskb_carve(struct sk_buff
*skb
, const u32 len
, gfp_t gfp
)
5537 int headlen
= skb_headlen(skb
);
5540 return pskb_carve_inside_header(skb
, len
, headlen
, gfp
);
5542 return pskb_carve_inside_nonlinear(skb
, len
, headlen
, gfp
);
5545 /* Extract to_copy bytes starting at off from skb, and return this in
5548 struct sk_buff
*pskb_extract(struct sk_buff
*skb
, int off
,
5549 int to_copy
, gfp_t gfp
)
5551 struct sk_buff
*clone
= skb_clone(skb
, gfp
);
5556 if (pskb_carve(clone
, off
, gfp
) < 0 ||
5557 pskb_trim(clone
, to_copy
)) {
5563 EXPORT_SYMBOL(pskb_extract
);
5566 * skb_condense - try to get rid of fragments/frag_list if possible
5569 * Can be used to save memory before skb is added to a busy queue.
5570 * If packet has bytes in frags and enough tail room in skb->head,
5571 * pull all of them, so that we can free the frags right now and adjust
5574 * We do not reallocate skb->head thus can not fail.
5575 * Caller must re-evaluate skb->truesize if needed.
5577 void skb_condense(struct sk_buff
*skb
)
5579 if (skb
->data_len
) {
5580 if (skb
->data_len
> skb
->end
- skb
->tail
||
5584 /* Nice, we can free page frag(s) right now */
5585 __pskb_pull_tail(skb
, skb
->data_len
);
5587 /* At this point, skb->truesize might be over estimated,
5588 * because skb had a fragment, and fragments do not tell
5590 * When we pulled its content into skb->head, fragment
5591 * was freed, but __pskb_pull_tail() could not possibly
5592 * adjust skb->truesize, not knowing the frag truesize.
5594 skb
->truesize
= SKB_TRUESIZE(skb_end_offset(skb
));
5597 #ifdef CONFIG_SKB_EXTENSIONS
5598 static void *skb_ext_get_ptr(struct skb_ext
*ext
, enum skb_ext_id id
)
5600 return (void *)ext
+ (ext
->offset
[id
] * SKB_EXT_ALIGN_VALUE
);
5603 static struct skb_ext
*skb_ext_alloc(void)
5605 struct skb_ext
*new = kmem_cache_alloc(skbuff_ext_cache
, GFP_ATOMIC
);
5608 memset(new->offset
, 0, sizeof(new->offset
));
5609 refcount_set(&new->refcnt
, 1);
5615 static struct skb_ext
*skb_ext_maybe_cow(struct skb_ext
*old
,
5616 unsigned int old_active
)
5618 struct skb_ext
*new;
5620 if (refcount_read(&old
->refcnt
) == 1)
5623 new = kmem_cache_alloc(skbuff_ext_cache
, GFP_ATOMIC
);
5627 memcpy(new, old
, old
->chunks
* SKB_EXT_ALIGN_VALUE
);
5628 refcount_set(&new->refcnt
, 1);
5631 if (old_active
& (1 << SKB_EXT_SEC_PATH
)) {
5632 struct sec_path
*sp
= skb_ext_get_ptr(old
, SKB_EXT_SEC_PATH
);
5635 for (i
= 0; i
< sp
->len
; i
++)
5636 xfrm_state_hold(sp
->xvec
[i
]);
5644 * skb_ext_add - allocate space for given extension, COW if needed
5646 * @id: extension to allocate space for
5648 * Allocates enough space for the given extension.
5649 * If the extension is already present, a pointer to that extension
5652 * If the skb was cloned, COW applies and the returned memory can be
5653 * modified without changing the extension space of clones buffers.
5655 * Returns pointer to the extension or NULL on allocation failure.
5657 void *skb_ext_add(struct sk_buff
*skb
, enum skb_ext_id id
)
5659 struct skb_ext
*new, *old
= NULL
;
5660 unsigned int newlen
, newoff
;
5662 if (skb
->active_extensions
) {
5663 old
= skb
->extensions
;
5665 new = skb_ext_maybe_cow(old
, skb
->active_extensions
);
5669 if (__skb_ext_exist(new, id
))
5672 newoff
= new->chunks
;
5674 newoff
= SKB_EXT_CHUNKSIZEOF(*new);
5676 new = skb_ext_alloc();
5681 newlen
= newoff
+ skb_ext_type_len
[id
];
5682 new->chunks
= newlen
;
5683 new->offset
[id
] = newoff
;
5685 skb
->extensions
= new;
5686 skb
->active_extensions
|= 1 << id
;
5687 return skb_ext_get_ptr(new, id
);
5689 EXPORT_SYMBOL(skb_ext_add
);
5692 static void skb_ext_put_sp(struct sec_path
*sp
)
5696 for (i
= 0; i
< sp
->len
; i
++)
5697 xfrm_state_put(sp
->xvec
[i
]);
5701 void __skb_ext_del(struct sk_buff
*skb
, enum skb_ext_id id
)
5703 struct skb_ext
*ext
= skb
->extensions
;
5705 skb
->active_extensions
&= ~(1 << id
);
5706 if (skb
->active_extensions
== 0) {
5707 skb
->extensions
= NULL
;
5710 } else if (id
== SKB_EXT_SEC_PATH
&&
5711 refcount_read(&ext
->refcnt
) == 1) {
5712 struct sec_path
*sp
= skb_ext_get_ptr(ext
, SKB_EXT_SEC_PATH
);
5719 EXPORT_SYMBOL(__skb_ext_del
);
5721 void __skb_ext_put(struct skb_ext
*ext
)
5723 /* If this is last clone, nothing can increment
5724 * it after check passes. Avoids one atomic op.
5726 if (refcount_read(&ext
->refcnt
) == 1)
5729 if (!refcount_dec_and_test(&ext
->refcnt
))
5733 if (__skb_ext_exist(ext
, SKB_EXT_SEC_PATH
))
5734 skb_ext_put_sp(skb_ext_get_ptr(ext
, SKB_EXT_SEC_PATH
));
5737 kmem_cache_free(skbuff_ext_cache
, ext
);
5739 EXPORT_SYMBOL(__skb_ext_put
);
5740 #endif /* CONFIG_SKB_EXTENSIONS */