2 * Routines having to do with the 'struct sk_buff' memory handlers.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
8 * Alan Cox : Fixed the worst of the load
10 * Dave Platt : Interrupt stacking fix.
11 * Richard Kooijman : Timestamp fixes.
12 * Alan Cox : Changed buffer format.
13 * Alan Cox : destructor hook for AF_UNIX etc.
14 * Linus Torvalds : Better skb_clone.
15 * Alan Cox : Added skb_copy.
16 * Alan Cox : Added all the changed routines Linus
17 * only put in the headers
18 * Ray VanTassle : Fixed --skb->lock in free
19 * Alan Cox : skb_copy copy arp field
20 * Andi Kleen : slabified it.
21 * Robert Olsson : Removed skb_head_pool
24 * The __skb_ routines should be called with interrupts
25 * disabled, or you better be *real* sure that the operation is atomic
26 * with respect to whatever list is being frobbed (e.g. via lock_sock()
27 * or via disabling bottom half handlers, etc).
29 * This program is free software; you can redistribute it and/or
30 * modify it under the terms of the GNU General Public License
31 * as published by the Free Software Foundation; either version
32 * 2 of the License, or (at your option) any later version.
36 * The functions in this file will not compile correctly with gcc 2.4.x
39 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
41 #include <linux/module.h>
42 #include <linux/types.h>
43 #include <linux/kernel.h>
44 #include <linux/kmemcheck.h>
46 #include <linux/interrupt.h>
48 #include <linux/inet.h>
49 #include <linux/slab.h>
50 #include <linux/tcp.h>
51 #include <linux/udp.h>
52 #include <linux/sctp.h>
53 #include <linux/netdevice.h>
54 #ifdef CONFIG_NET_CLS_ACT
55 #include <net/pkt_sched.h>
57 #include <linux/string.h>
58 #include <linux/skbuff.h>
59 #include <linux/splice.h>
60 #include <linux/cache.h>
61 #include <linux/rtnetlink.h>
62 #include <linux/init.h>
63 #include <linux/scatterlist.h>
64 #include <linux/errqueue.h>
65 #include <linux/prefetch.h>
66 #include <linux/if_vlan.h>
68 #include <net/protocol.h>
71 #include <net/checksum.h>
72 #include <net/ip6_checksum.h>
75 #include <linux/uaccess.h>
76 #include <trace/events/skb.h>
77 #include <linux/highmem.h>
78 #include <linux/capability.h>
79 #include <linux/user_namespace.h>
81 struct kmem_cache
*skbuff_head_cache __read_mostly
;
82 static struct kmem_cache
*skbuff_fclone_cache __read_mostly
;
83 int sysctl_max_skb_frags __read_mostly
= MAX_SKB_FRAGS
;
84 EXPORT_SYMBOL(sysctl_max_skb_frags
);
87 * skb_panic - private function for out-of-line support
91 * @msg: skb_over_panic or skb_under_panic
93 * Out-of-line support for skb_put() and skb_push().
94 * Called via the wrapper skb_over_panic() or skb_under_panic().
95 * Keep out of line to prevent kernel bloat.
96 * __builtin_return_address is not used because it is not always reliable.
98 static void skb_panic(struct sk_buff
*skb
, unsigned int sz
, void *addr
,
101 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
102 msg
, addr
, skb
->len
, sz
, skb
->head
, skb
->data
,
103 (unsigned long)skb
->tail
, (unsigned long)skb
->end
,
104 skb
->dev
? skb
->dev
->name
: "<NULL>");
108 static void skb_over_panic(struct sk_buff
*skb
, unsigned int sz
, void *addr
)
110 skb_panic(skb
, sz
, addr
, __func__
);
113 static void skb_under_panic(struct sk_buff
*skb
, unsigned int sz
, void *addr
)
115 skb_panic(skb
, sz
, addr
, __func__
);
119 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
120 * the caller if emergency pfmemalloc reserves are being used. If it is and
121 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
122 * may be used. Otherwise, the packet data may be discarded until enough
125 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
126 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
128 static void *__kmalloc_reserve(size_t size
, gfp_t flags
, int node
,
129 unsigned long ip
, bool *pfmemalloc
)
132 bool ret_pfmemalloc
= false;
135 * Try a regular allocation, when that fails and we're not entitled
136 * to the reserves, fail.
138 obj
= kmalloc_node_track_caller(size
,
139 flags
| __GFP_NOMEMALLOC
| __GFP_NOWARN
,
141 if (obj
|| !(gfp_pfmemalloc_allowed(flags
)))
144 /* Try again but now we are using pfmemalloc reserves */
145 ret_pfmemalloc
= true;
146 obj
= kmalloc_node_track_caller(size
, flags
, node
);
150 *pfmemalloc
= ret_pfmemalloc
;
155 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
156 * 'private' fields and also do memory statistics to find all the
161 struct sk_buff
*__alloc_skb_head(gfp_t gfp_mask
, int node
)
166 skb
= kmem_cache_alloc_node(skbuff_head_cache
,
167 gfp_mask
& ~__GFP_DMA
, node
);
172 * Only clear those fields we need to clear, not those that we will
173 * actually initialise below. Hence, don't put any more fields after
174 * the tail pointer in struct sk_buff!
176 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
178 skb
->truesize
= sizeof(struct sk_buff
);
179 atomic_set(&skb
->users
, 1);
181 skb
->mac_header
= (typeof(skb
->mac_header
))~0U;
187 * __alloc_skb - allocate a network buffer
188 * @size: size to allocate
189 * @gfp_mask: allocation mask
190 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
191 * instead of head cache and allocate a cloned (child) skb.
192 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
193 * allocations in case the data is required for writeback
194 * @node: numa node to allocate memory on
196 * Allocate a new &sk_buff. The returned buffer has no headroom and a
197 * tail room of at least size bytes. The object has a reference count
198 * of one. The return is the buffer. On a failure the return is %NULL.
200 * Buffers may only be allocated from interrupts using a @gfp_mask of
203 struct sk_buff
*__alloc_skb(unsigned int size
, gfp_t gfp_mask
,
206 struct kmem_cache
*cache
;
207 struct skb_shared_info
*shinfo
;
212 cache
= (flags
& SKB_ALLOC_FCLONE
)
213 ? skbuff_fclone_cache
: skbuff_head_cache
;
215 if (sk_memalloc_socks() && (flags
& SKB_ALLOC_RX
))
216 gfp_mask
|= __GFP_MEMALLOC
;
219 skb
= kmem_cache_alloc_node(cache
, gfp_mask
& ~__GFP_DMA
, node
);
224 /* We do our best to align skb_shared_info on a separate cache
225 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
226 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
227 * Both skb->head and skb_shared_info are cache line aligned.
229 size
= SKB_DATA_ALIGN(size
);
230 size
+= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
231 data
= kmalloc_reserve(size
, gfp_mask
, node
, &pfmemalloc
);
234 /* kmalloc(size) might give us more room than requested.
235 * Put skb_shared_info exactly at the end of allocated zone,
236 * to allow max possible filling before reallocation.
238 size
= SKB_WITH_OVERHEAD(ksize(data
));
239 prefetchw(data
+ size
);
242 * Only clear those fields we need to clear, not those that we will
243 * actually initialise below. Hence, don't put any more fields after
244 * the tail pointer in struct sk_buff!
246 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
247 /* Account for allocated memory : skb + skb->head */
248 skb
->truesize
= SKB_TRUESIZE(size
);
249 skb
->pfmemalloc
= pfmemalloc
;
250 atomic_set(&skb
->users
, 1);
253 skb_reset_tail_pointer(skb
);
254 skb
->end
= skb
->tail
+ size
;
255 skb
->mac_header
= (typeof(skb
->mac_header
))~0U;
256 skb
->transport_header
= (typeof(skb
->transport_header
))~0U;
258 /* make sure we initialize shinfo sequentially */
259 shinfo
= skb_shinfo(skb
);
260 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
261 atomic_set(&shinfo
->dataref
, 1);
262 kmemcheck_annotate_variable(shinfo
->destructor_arg
);
264 if (flags
& SKB_ALLOC_FCLONE
) {
265 struct sk_buff_fclones
*fclones
;
267 fclones
= container_of(skb
, struct sk_buff_fclones
, skb1
);
269 kmemcheck_annotate_bitfield(&fclones
->skb2
, flags1
);
270 skb
->fclone
= SKB_FCLONE_ORIG
;
271 atomic_set(&fclones
->fclone_ref
, 1);
273 fclones
->skb2
.fclone
= SKB_FCLONE_CLONE
;
278 kmem_cache_free(cache
, skb
);
282 EXPORT_SYMBOL(__alloc_skb
);
285 * __build_skb - build a network buffer
286 * @data: data buffer provided by caller
287 * @frag_size: size of data, or 0 if head was kmalloced
289 * Allocate a new &sk_buff. Caller provides space holding head and
290 * skb_shared_info. @data must have been allocated by kmalloc() only if
291 * @frag_size is 0, otherwise data should come from the page allocator
293 * The return is the new skb buffer.
294 * On a failure the return is %NULL, and @data is not freed.
296 * Before IO, driver allocates only data buffer where NIC put incoming frame
297 * Driver should add room at head (NET_SKB_PAD) and
298 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
299 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
300 * before giving packet to stack.
301 * RX rings only contains data buffers, not full skbs.
303 struct sk_buff
*__build_skb(void *data
, unsigned int frag_size
)
305 struct skb_shared_info
*shinfo
;
307 unsigned int size
= frag_size
? : ksize(data
);
309 skb
= kmem_cache_alloc(skbuff_head_cache
, GFP_ATOMIC
);
313 size
-= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
315 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
316 skb
->truesize
= SKB_TRUESIZE(size
);
317 atomic_set(&skb
->users
, 1);
320 skb_reset_tail_pointer(skb
);
321 skb
->end
= skb
->tail
+ size
;
322 skb
->mac_header
= (typeof(skb
->mac_header
))~0U;
323 skb
->transport_header
= (typeof(skb
->transport_header
))~0U;
325 /* make sure we initialize shinfo sequentially */
326 shinfo
= skb_shinfo(skb
);
327 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
328 atomic_set(&shinfo
->dataref
, 1);
329 kmemcheck_annotate_variable(shinfo
->destructor_arg
);
334 /* build_skb() is wrapper over __build_skb(), that specifically
335 * takes care of skb->head and skb->pfmemalloc
336 * This means that if @frag_size is not zero, then @data must be backed
337 * by a page fragment, not kmalloc() or vmalloc()
339 struct sk_buff
*build_skb(void *data
, unsigned int frag_size
)
341 struct sk_buff
*skb
= __build_skb(data
, frag_size
);
343 if (skb
&& frag_size
) {
345 if (page_is_pfmemalloc(virt_to_head_page(data
)))
350 EXPORT_SYMBOL(build_skb
);
352 #define NAPI_SKB_CACHE_SIZE 64
354 struct napi_alloc_cache
{
355 struct page_frag_cache page
;
356 unsigned int skb_count
;
357 void *skb_cache
[NAPI_SKB_CACHE_SIZE
];
360 static DEFINE_PER_CPU(struct page_frag_cache
, netdev_alloc_cache
);
361 static DEFINE_PER_CPU(struct napi_alloc_cache
, napi_alloc_cache
);
363 static void *__netdev_alloc_frag(unsigned int fragsz
, gfp_t gfp_mask
)
365 struct page_frag_cache
*nc
;
369 local_irq_save(flags
);
370 nc
= this_cpu_ptr(&netdev_alloc_cache
);
371 data
= page_frag_alloc(nc
, fragsz
, gfp_mask
);
372 local_irq_restore(flags
);
377 * netdev_alloc_frag - allocate a page fragment
378 * @fragsz: fragment size
380 * Allocates a frag from a page for receive buffer.
381 * Uses GFP_ATOMIC allocations.
383 void *netdev_alloc_frag(unsigned int fragsz
)
385 return __netdev_alloc_frag(fragsz
, GFP_ATOMIC
| __GFP_COLD
);
387 EXPORT_SYMBOL(netdev_alloc_frag
);
389 static void *__napi_alloc_frag(unsigned int fragsz
, gfp_t gfp_mask
)
391 struct napi_alloc_cache
*nc
= this_cpu_ptr(&napi_alloc_cache
);
393 return page_frag_alloc(&nc
->page
, fragsz
, gfp_mask
);
396 void *napi_alloc_frag(unsigned int fragsz
)
398 return __napi_alloc_frag(fragsz
, GFP_ATOMIC
| __GFP_COLD
);
400 EXPORT_SYMBOL(napi_alloc_frag
);
403 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
404 * @dev: network device to receive on
405 * @len: length to allocate
406 * @gfp_mask: get_free_pages mask, passed to alloc_skb
408 * Allocate a new &sk_buff and assign it a usage count of one. The
409 * buffer has NET_SKB_PAD headroom built in. Users should allocate
410 * the headroom they think they need without accounting for the
411 * built in space. The built in space is used for optimisations.
413 * %NULL is returned if there is no free memory.
415 struct sk_buff
*__netdev_alloc_skb(struct net_device
*dev
, unsigned int len
,
418 struct page_frag_cache
*nc
;
426 if ((len
> SKB_WITH_OVERHEAD(PAGE_SIZE
)) ||
427 (gfp_mask
& (__GFP_DIRECT_RECLAIM
| GFP_DMA
))) {
428 skb
= __alloc_skb(len
, gfp_mask
, SKB_ALLOC_RX
, NUMA_NO_NODE
);
434 len
+= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
435 len
= SKB_DATA_ALIGN(len
);
437 if (sk_memalloc_socks())
438 gfp_mask
|= __GFP_MEMALLOC
;
440 local_irq_save(flags
);
442 nc
= this_cpu_ptr(&netdev_alloc_cache
);
443 data
= page_frag_alloc(nc
, len
, gfp_mask
);
444 pfmemalloc
= nc
->pfmemalloc
;
446 local_irq_restore(flags
);
451 skb
= __build_skb(data
, len
);
452 if (unlikely(!skb
)) {
457 /* use OR instead of assignment to avoid clearing of bits in mask */
463 skb_reserve(skb
, NET_SKB_PAD
);
469 EXPORT_SYMBOL(__netdev_alloc_skb
);
472 * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
473 * @napi: napi instance this buffer was allocated for
474 * @len: length to allocate
475 * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
477 * Allocate a new sk_buff for use in NAPI receive. This buffer will
478 * attempt to allocate the head from a special reserved region used
479 * only for NAPI Rx allocation. By doing this we can save several
480 * CPU cycles by avoiding having to disable and re-enable IRQs.
482 * %NULL is returned if there is no free memory.
484 struct sk_buff
*__napi_alloc_skb(struct napi_struct
*napi
, unsigned int len
,
487 struct napi_alloc_cache
*nc
= this_cpu_ptr(&napi_alloc_cache
);
491 len
+= NET_SKB_PAD
+ NET_IP_ALIGN
;
493 if ((len
> SKB_WITH_OVERHEAD(PAGE_SIZE
)) ||
494 (gfp_mask
& (__GFP_DIRECT_RECLAIM
| GFP_DMA
))) {
495 skb
= __alloc_skb(len
, gfp_mask
, SKB_ALLOC_RX
, NUMA_NO_NODE
);
501 len
+= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
502 len
= SKB_DATA_ALIGN(len
);
504 if (sk_memalloc_socks())
505 gfp_mask
|= __GFP_MEMALLOC
;
507 data
= page_frag_alloc(&nc
->page
, len
, gfp_mask
);
511 skb
= __build_skb(data
, len
);
512 if (unlikely(!skb
)) {
517 /* use OR instead of assignment to avoid clearing of bits in mask */
518 if (nc
->page
.pfmemalloc
)
523 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
);
524 skb
->dev
= napi
->dev
;
529 EXPORT_SYMBOL(__napi_alloc_skb
);
531 void skb_add_rx_frag(struct sk_buff
*skb
, int i
, struct page
*page
, int off
,
532 int size
, unsigned int truesize
)
534 skb_fill_page_desc(skb
, i
, page
, off
, size
);
536 skb
->data_len
+= size
;
537 skb
->truesize
+= truesize
;
539 EXPORT_SYMBOL(skb_add_rx_frag
);
541 void skb_coalesce_rx_frag(struct sk_buff
*skb
, int i
, int size
,
542 unsigned int truesize
)
544 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
546 skb_frag_size_add(frag
, size
);
548 skb
->data_len
+= size
;
549 skb
->truesize
+= truesize
;
551 EXPORT_SYMBOL(skb_coalesce_rx_frag
);
553 static void skb_drop_list(struct sk_buff
**listp
)
555 kfree_skb_list(*listp
);
559 static inline void skb_drop_fraglist(struct sk_buff
*skb
)
561 skb_drop_list(&skb_shinfo(skb
)->frag_list
);
564 static void skb_clone_fraglist(struct sk_buff
*skb
)
566 struct sk_buff
*list
;
568 skb_walk_frags(skb
, list
)
572 static void skb_free_head(struct sk_buff
*skb
)
574 unsigned char *head
= skb
->head
;
582 static void skb_release_data(struct sk_buff
*skb
)
584 struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
588 atomic_sub_return(skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1,
592 for (i
= 0; i
< shinfo
->nr_frags
; i
++)
593 __skb_frag_unref(&shinfo
->frags
[i
]);
596 * If skb buf is from userspace, we need to notify the caller
597 * the lower device DMA has done;
599 if (shinfo
->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
600 struct ubuf_info
*uarg
;
602 uarg
= shinfo
->destructor_arg
;
604 uarg
->callback(uarg
, true);
607 if (shinfo
->frag_list
)
608 kfree_skb_list(shinfo
->frag_list
);
614 * Free an skbuff by memory without cleaning the state.
616 static void kfree_skbmem(struct sk_buff
*skb
)
618 struct sk_buff_fclones
*fclones
;
620 switch (skb
->fclone
) {
621 case SKB_FCLONE_UNAVAILABLE
:
622 kmem_cache_free(skbuff_head_cache
, skb
);
625 case SKB_FCLONE_ORIG
:
626 fclones
= container_of(skb
, struct sk_buff_fclones
, skb1
);
628 /* We usually free the clone (TX completion) before original skb
629 * This test would have no chance to be true for the clone,
630 * while here, branch prediction will be good.
632 if (atomic_read(&fclones
->fclone_ref
) == 1)
636 default: /* SKB_FCLONE_CLONE */
637 fclones
= container_of(skb
, struct sk_buff_fclones
, skb2
);
640 if (!atomic_dec_and_test(&fclones
->fclone_ref
))
643 kmem_cache_free(skbuff_fclone_cache
, fclones
);
646 static void skb_release_head_state(struct sk_buff
*skb
)
650 secpath_put(skb
->sp
);
652 if (skb
->destructor
) {
654 skb
->destructor(skb
);
656 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
657 nf_conntrack_put(skb_nfct(skb
));
659 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
660 nf_bridge_put(skb
->nf_bridge
);
664 /* Free everything but the sk_buff shell. */
665 static void skb_release_all(struct sk_buff
*skb
)
667 skb_release_head_state(skb
);
668 if (likely(skb
->head
))
669 skb_release_data(skb
);
673 * __kfree_skb - private function
676 * Free an sk_buff. Release anything attached to the buffer.
677 * Clean the state. This is an internal helper function. Users should
678 * always call kfree_skb
681 void __kfree_skb(struct sk_buff
*skb
)
683 skb_release_all(skb
);
686 EXPORT_SYMBOL(__kfree_skb
);
689 * kfree_skb - free an sk_buff
690 * @skb: buffer to free
692 * Drop a reference to the buffer and free it if the usage count has
695 void kfree_skb(struct sk_buff
*skb
)
699 if (likely(atomic_read(&skb
->users
) == 1))
701 else if (likely(!atomic_dec_and_test(&skb
->users
)))
703 trace_kfree_skb(skb
, __builtin_return_address(0));
706 EXPORT_SYMBOL(kfree_skb
);
708 void kfree_skb_list(struct sk_buff
*segs
)
711 struct sk_buff
*next
= segs
->next
;
717 EXPORT_SYMBOL(kfree_skb_list
);
720 * skb_tx_error - report an sk_buff xmit error
721 * @skb: buffer that triggered an error
723 * Report xmit error if a device callback is tracking this skb.
724 * skb must be freed afterwards.
726 void skb_tx_error(struct sk_buff
*skb
)
728 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
729 struct ubuf_info
*uarg
;
731 uarg
= skb_shinfo(skb
)->destructor_arg
;
733 uarg
->callback(uarg
, false);
734 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
737 EXPORT_SYMBOL(skb_tx_error
);
740 * consume_skb - free an skbuff
741 * @skb: buffer to free
743 * Drop a ref to the buffer and free it if the usage count has hit zero
744 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
745 * is being dropped after a failure and notes that
747 void consume_skb(struct sk_buff
*skb
)
751 if (likely(atomic_read(&skb
->users
) == 1))
753 else if (likely(!atomic_dec_and_test(&skb
->users
)))
755 trace_consume_skb(skb
);
758 EXPORT_SYMBOL(consume_skb
);
760 void __kfree_skb_flush(void)
762 struct napi_alloc_cache
*nc
= this_cpu_ptr(&napi_alloc_cache
);
764 /* flush skb_cache if containing objects */
766 kmem_cache_free_bulk(skbuff_head_cache
, nc
->skb_count
,
772 static inline void _kfree_skb_defer(struct sk_buff
*skb
)
774 struct napi_alloc_cache
*nc
= this_cpu_ptr(&napi_alloc_cache
);
776 /* drop skb->head and call any destructors for packet */
777 skb_release_all(skb
);
779 /* record skb to CPU local list */
780 nc
->skb_cache
[nc
->skb_count
++] = skb
;
783 /* SLUB writes into objects when freeing */
787 /* flush skb_cache if it is filled */
788 if (unlikely(nc
->skb_count
== NAPI_SKB_CACHE_SIZE
)) {
789 kmem_cache_free_bulk(skbuff_head_cache
, NAPI_SKB_CACHE_SIZE
,
794 void __kfree_skb_defer(struct sk_buff
*skb
)
796 _kfree_skb_defer(skb
);
799 void napi_consume_skb(struct sk_buff
*skb
, int budget
)
804 /* Zero budget indicate non-NAPI context called us, like netpoll */
805 if (unlikely(!budget
)) {
806 dev_consume_skb_any(skb
);
810 if (likely(atomic_read(&skb
->users
) == 1))
812 else if (likely(!atomic_dec_and_test(&skb
->users
)))
814 /* if reaching here SKB is ready to free */
815 trace_consume_skb(skb
);
817 /* if SKB is a clone, don't handle this case */
818 if (skb
->fclone
!= SKB_FCLONE_UNAVAILABLE
) {
823 _kfree_skb_defer(skb
);
825 EXPORT_SYMBOL(napi_consume_skb
);
827 /* Make sure a field is enclosed inside headers_start/headers_end section */
828 #define CHECK_SKB_FIELD(field) \
829 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
830 offsetof(struct sk_buff, headers_start)); \
831 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
832 offsetof(struct sk_buff, headers_end)); \
834 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
836 new->tstamp
= old
->tstamp
;
837 /* We do not copy old->sk */
839 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
840 skb_dst_copy(new, old
);
842 new->sp
= secpath_get(old
->sp
);
844 __nf_copy(new, old
, false);
846 /* Note : this field could be in headers_start/headers_end section
847 * It is not yet because we do not want to have a 16 bit hole
849 new->queue_mapping
= old
->queue_mapping
;
851 memcpy(&new->headers_start
, &old
->headers_start
,
852 offsetof(struct sk_buff
, headers_end
) -
853 offsetof(struct sk_buff
, headers_start
));
854 CHECK_SKB_FIELD(protocol
);
855 CHECK_SKB_FIELD(csum
);
856 CHECK_SKB_FIELD(hash
);
857 CHECK_SKB_FIELD(priority
);
858 CHECK_SKB_FIELD(skb_iif
);
859 CHECK_SKB_FIELD(vlan_proto
);
860 CHECK_SKB_FIELD(vlan_tci
);
861 CHECK_SKB_FIELD(transport_header
);
862 CHECK_SKB_FIELD(network_header
);
863 CHECK_SKB_FIELD(mac_header
);
864 CHECK_SKB_FIELD(inner_protocol
);
865 CHECK_SKB_FIELD(inner_transport_header
);
866 CHECK_SKB_FIELD(inner_network_header
);
867 CHECK_SKB_FIELD(inner_mac_header
);
868 CHECK_SKB_FIELD(mark
);
869 #ifdef CONFIG_NETWORK_SECMARK
870 CHECK_SKB_FIELD(secmark
);
872 #ifdef CONFIG_NET_RX_BUSY_POLL
873 CHECK_SKB_FIELD(napi_id
);
876 CHECK_SKB_FIELD(sender_cpu
);
878 #ifdef CONFIG_NET_SCHED
879 CHECK_SKB_FIELD(tc_index
);
885 * You should not add any new code to this function. Add it to
886 * __copy_skb_header above instead.
888 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
890 #define C(x) n->x = skb->x
892 n
->next
= n
->prev
= NULL
;
894 __copy_skb_header(n
, skb
);
899 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
902 n
->destructor
= NULL
;
909 atomic_set(&n
->users
, 1);
911 atomic_inc(&(skb_shinfo(skb
)->dataref
));
919 * skb_morph - morph one skb into another
920 * @dst: the skb to receive the contents
921 * @src: the skb to supply the contents
923 * This is identical to skb_clone except that the target skb is
924 * supplied by the user.
926 * The target skb is returned upon exit.
928 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
930 skb_release_all(dst
);
931 return __skb_clone(dst
, src
);
933 EXPORT_SYMBOL_GPL(skb_morph
);
936 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
937 * @skb: the skb to modify
938 * @gfp_mask: allocation priority
940 * This must be called on SKBTX_DEV_ZEROCOPY skb.
941 * It will copy all frags into kernel and drop the reference
942 * to userspace pages.
944 * If this function is called from an interrupt gfp_mask() must be
947 * Returns 0 on success or a negative error code on failure
948 * to allocate kernel memory to copy to.
950 int skb_copy_ubufs(struct sk_buff
*skb
, gfp_t gfp_mask
)
953 int num_frags
= skb_shinfo(skb
)->nr_frags
;
954 struct page
*page
, *head
= NULL
;
955 struct ubuf_info
*uarg
= skb_shinfo(skb
)->destructor_arg
;
957 for (i
= 0; i
< num_frags
; i
++) {
959 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
961 page
= alloc_page(gfp_mask
);
964 struct page
*next
= (struct page
*)page_private(head
);
970 vaddr
= kmap_atomic(skb_frag_page(f
));
971 memcpy(page_address(page
),
972 vaddr
+ f
->page_offset
, skb_frag_size(f
));
973 kunmap_atomic(vaddr
);
974 set_page_private(page
, (unsigned long)head
);
978 /* skb frags release userspace buffers */
979 for (i
= 0; i
< num_frags
; i
++)
980 skb_frag_unref(skb
, i
);
982 uarg
->callback(uarg
, false);
984 /* skb frags point to kernel buffers */
985 for (i
= num_frags
- 1; i
>= 0; i
--) {
986 __skb_fill_page_desc(skb
, i
, head
, 0,
987 skb_shinfo(skb
)->frags
[i
].size
);
988 head
= (struct page
*)page_private(head
);
991 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
994 EXPORT_SYMBOL_GPL(skb_copy_ubufs
);
997 * skb_clone - duplicate an sk_buff
998 * @skb: buffer to clone
999 * @gfp_mask: allocation priority
1001 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1002 * copies share the same packet data but not structure. The new
1003 * buffer has a reference count of 1. If the allocation fails the
1004 * function returns %NULL otherwise the new buffer is returned.
1006 * If this function is called from an interrupt gfp_mask() must be
1010 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
1012 struct sk_buff_fclones
*fclones
= container_of(skb
,
1013 struct sk_buff_fclones
,
1017 if (skb_orphan_frags(skb
, gfp_mask
))
1020 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
1021 atomic_read(&fclones
->fclone_ref
) == 1) {
1023 atomic_set(&fclones
->fclone_ref
, 2);
1025 if (skb_pfmemalloc(skb
))
1026 gfp_mask
|= __GFP_MEMALLOC
;
1028 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
1032 kmemcheck_annotate_bitfield(n
, flags1
);
1033 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
1036 return __skb_clone(n
, skb
);
1038 EXPORT_SYMBOL(skb_clone
);
1040 static void skb_headers_offset_update(struct sk_buff
*skb
, int off
)
1042 /* Only adjust this if it actually is csum_start rather than csum */
1043 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1044 skb
->csum_start
+= off
;
1045 /* {transport,network,mac}_header and tail are relative to skb->head */
1046 skb
->transport_header
+= off
;
1047 skb
->network_header
+= off
;
1048 if (skb_mac_header_was_set(skb
))
1049 skb
->mac_header
+= off
;
1050 skb
->inner_transport_header
+= off
;
1051 skb
->inner_network_header
+= off
;
1052 skb
->inner_mac_header
+= off
;
1055 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
1057 __copy_skb_header(new, old
);
1059 skb_shinfo(new)->gso_size
= skb_shinfo(old
)->gso_size
;
1060 skb_shinfo(new)->gso_segs
= skb_shinfo(old
)->gso_segs
;
1061 skb_shinfo(new)->gso_type
= skb_shinfo(old
)->gso_type
;
1064 static inline int skb_alloc_rx_flag(const struct sk_buff
*skb
)
1066 if (skb_pfmemalloc(skb
))
1067 return SKB_ALLOC_RX
;
1072 * skb_copy - create private copy of an sk_buff
1073 * @skb: buffer to copy
1074 * @gfp_mask: allocation priority
1076 * Make a copy of both an &sk_buff and its data. This is used when the
1077 * caller wishes to modify the data and needs a private copy of the
1078 * data to alter. Returns %NULL on failure or the pointer to the buffer
1079 * on success. The returned buffer has a reference count of 1.
1081 * As by-product this function converts non-linear &sk_buff to linear
1082 * one, so that &sk_buff becomes completely private and caller is allowed
1083 * to modify all the data of returned buffer. This means that this
1084 * function is not recommended for use in circumstances when only
1085 * header is going to be modified. Use pskb_copy() instead.
1088 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
1090 int headerlen
= skb_headroom(skb
);
1091 unsigned int size
= skb_end_offset(skb
) + skb
->data_len
;
1092 struct sk_buff
*n
= __alloc_skb(size
, gfp_mask
,
1093 skb_alloc_rx_flag(skb
), NUMA_NO_NODE
);
1098 /* Set the data pointer */
1099 skb_reserve(n
, headerlen
);
1100 /* Set the tail pointer and length */
1101 skb_put(n
, skb
->len
);
1103 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
1106 copy_skb_header(n
, skb
);
1109 EXPORT_SYMBOL(skb_copy
);
1112 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1113 * @skb: buffer to copy
1114 * @headroom: headroom of new skb
1115 * @gfp_mask: allocation priority
1116 * @fclone: if true allocate the copy of the skb from the fclone
1117 * cache instead of the head cache; it is recommended to set this
1118 * to true for the cases where the copy will likely be cloned
1120 * Make a copy of both an &sk_buff and part of its data, located
1121 * in header. Fragmented data remain shared. This is used when
1122 * the caller wishes to modify only header of &sk_buff and needs
1123 * private copy of the header to alter. Returns %NULL on failure
1124 * or the pointer to the buffer on success.
1125 * The returned buffer has a reference count of 1.
1128 struct sk_buff
*__pskb_copy_fclone(struct sk_buff
*skb
, int headroom
,
1129 gfp_t gfp_mask
, bool fclone
)
1131 unsigned int size
= skb_headlen(skb
) + headroom
;
1132 int flags
= skb_alloc_rx_flag(skb
) | (fclone
? SKB_ALLOC_FCLONE
: 0);
1133 struct sk_buff
*n
= __alloc_skb(size
, gfp_mask
, flags
, NUMA_NO_NODE
);
1138 /* Set the data pointer */
1139 skb_reserve(n
, headroom
);
1140 /* Set the tail pointer and length */
1141 skb_put(n
, skb_headlen(skb
));
1142 /* Copy the bytes */
1143 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
1145 n
->truesize
+= skb
->data_len
;
1146 n
->data_len
= skb
->data_len
;
1149 if (skb_shinfo(skb
)->nr_frags
) {
1152 if (skb_orphan_frags(skb
, gfp_mask
)) {
1157 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1158 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
1159 skb_frag_ref(skb
, i
);
1161 skb_shinfo(n
)->nr_frags
= i
;
1164 if (skb_has_frag_list(skb
)) {
1165 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
1166 skb_clone_fraglist(n
);
1169 copy_skb_header(n
, skb
);
1173 EXPORT_SYMBOL(__pskb_copy_fclone
);
1176 * pskb_expand_head - reallocate header of &sk_buff
1177 * @skb: buffer to reallocate
1178 * @nhead: room to add at head
1179 * @ntail: room to add at tail
1180 * @gfp_mask: allocation priority
1182 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1183 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1184 * reference count of 1. Returns zero in the case of success or error,
1185 * if expansion failed. In the last case, &sk_buff is not changed.
1187 * All the pointers pointing into skb header may change and must be
1188 * reloaded after call to this function.
1191 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
1194 int i
, osize
= skb_end_offset(skb
);
1195 int size
= osize
+ nhead
+ ntail
;
1201 if (skb_shared(skb
))
1204 size
= SKB_DATA_ALIGN(size
);
1206 if (skb_pfmemalloc(skb
))
1207 gfp_mask
|= __GFP_MEMALLOC
;
1208 data
= kmalloc_reserve(size
+ SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)),
1209 gfp_mask
, NUMA_NO_NODE
, NULL
);
1212 size
= SKB_WITH_OVERHEAD(ksize(data
));
1214 /* Copy only real data... and, alas, header. This should be
1215 * optimized for the cases when header is void.
1217 memcpy(data
+ nhead
, skb
->head
, skb_tail_pointer(skb
) - skb
->head
);
1219 memcpy((struct skb_shared_info
*)(data
+ size
),
1221 offsetof(struct skb_shared_info
, frags
[skb_shinfo(skb
)->nr_frags
]));
1224 * if shinfo is shared we must drop the old head gracefully, but if it
1225 * is not we can just drop the old head and let the existing refcount
1226 * be since all we did is relocate the values
1228 if (skb_cloned(skb
)) {
1229 /* copy this zero copy skb frags */
1230 if (skb_orphan_frags(skb
, gfp_mask
))
1232 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1233 skb_frag_ref(skb
, i
);
1235 if (skb_has_frag_list(skb
))
1236 skb_clone_fraglist(skb
);
1238 skb_release_data(skb
);
1242 off
= (data
+ nhead
) - skb
->head
;
1247 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1251 skb
->end
= skb
->head
+ size
;
1254 skb_headers_offset_update(skb
, nhead
);
1258 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
1260 /* It is not generally safe to change skb->truesize.
1261 * For the moment, we really care of rx path, or
1262 * when skb is orphaned (not attached to a socket).
1264 if (!skb
->sk
|| skb
->destructor
== sock_edemux
)
1265 skb
->truesize
+= size
- osize
;
1274 EXPORT_SYMBOL(pskb_expand_head
);
1276 /* Make private copy of skb with writable head and some headroom */
1278 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
1280 struct sk_buff
*skb2
;
1281 int delta
= headroom
- skb_headroom(skb
);
1284 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
1286 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1287 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
1295 EXPORT_SYMBOL(skb_realloc_headroom
);
1298 * skb_copy_expand - copy and expand sk_buff
1299 * @skb: buffer to copy
1300 * @newheadroom: new free bytes at head
1301 * @newtailroom: new free bytes at tail
1302 * @gfp_mask: allocation priority
1304 * Make a copy of both an &sk_buff and its data and while doing so
1305 * allocate additional space.
1307 * This is used when the caller wishes to modify the data and needs a
1308 * private copy of the data to alter as well as more space for new fields.
1309 * Returns %NULL on failure or the pointer to the buffer
1310 * on success. The returned buffer has a reference count of 1.
1312 * You must pass %GFP_ATOMIC as the allocation priority if this function
1313 * is called from an interrupt.
1315 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
1316 int newheadroom
, int newtailroom
,
1320 * Allocate the copy buffer
1322 struct sk_buff
*n
= __alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
1323 gfp_mask
, skb_alloc_rx_flag(skb
),
1325 int oldheadroom
= skb_headroom(skb
);
1326 int head_copy_len
, head_copy_off
;
1331 skb_reserve(n
, newheadroom
);
1333 /* Set the tail pointer and length */
1334 skb_put(n
, skb
->len
);
1336 head_copy_len
= oldheadroom
;
1338 if (newheadroom
<= head_copy_len
)
1339 head_copy_len
= newheadroom
;
1341 head_copy_off
= newheadroom
- head_copy_len
;
1343 /* Copy the linear header and data. */
1344 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
1345 skb
->len
+ head_copy_len
))
1348 copy_skb_header(n
, skb
);
1350 skb_headers_offset_update(n
, newheadroom
- oldheadroom
);
1354 EXPORT_SYMBOL(skb_copy_expand
);
1357 * skb_pad - zero pad the tail of an skb
1358 * @skb: buffer to pad
1359 * @pad: space to pad
1361 * Ensure that a buffer is followed by a padding area that is zero
1362 * filled. Used by network drivers which may DMA or transfer data
1363 * beyond the buffer end onto the wire.
1365 * May return error in out of memory cases. The skb is freed on error.
1368 int skb_pad(struct sk_buff
*skb
, int pad
)
1373 /* If the skbuff is non linear tailroom is always zero.. */
1374 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
1375 memset(skb
->data
+skb
->len
, 0, pad
);
1379 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
1380 if (likely(skb_cloned(skb
) || ntail
> 0)) {
1381 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
1386 /* FIXME: The use of this function with non-linear skb's really needs
1389 err
= skb_linearize(skb
);
1393 memset(skb
->data
+ skb
->len
, 0, pad
);
1400 EXPORT_SYMBOL(skb_pad
);
1403 * pskb_put - add data to the tail of a potentially fragmented buffer
1404 * @skb: start of the buffer to use
1405 * @tail: tail fragment of the buffer to use
1406 * @len: amount of data to add
1408 * This function extends the used data area of the potentially
1409 * fragmented buffer. @tail must be the last fragment of @skb -- or
1410 * @skb itself. If this would exceed the total buffer size the kernel
1411 * will panic. A pointer to the first byte of the extra data is
1415 unsigned char *pskb_put(struct sk_buff
*skb
, struct sk_buff
*tail
, int len
)
1418 skb
->data_len
+= len
;
1421 return skb_put(tail
, len
);
1423 EXPORT_SYMBOL_GPL(pskb_put
);
1426 * skb_put - add data to a buffer
1427 * @skb: buffer to use
1428 * @len: amount of data to add
1430 * This function extends the used data area of the buffer. If this would
1431 * exceed the total buffer size the kernel will panic. A pointer to the
1432 * first byte of the extra data is returned.
1434 unsigned char *skb_put(struct sk_buff
*skb
, unsigned int len
)
1436 unsigned char *tmp
= skb_tail_pointer(skb
);
1437 SKB_LINEAR_ASSERT(skb
);
1440 if (unlikely(skb
->tail
> skb
->end
))
1441 skb_over_panic(skb
, len
, __builtin_return_address(0));
1444 EXPORT_SYMBOL(skb_put
);
1447 * skb_push - add data to the start of a buffer
1448 * @skb: buffer to use
1449 * @len: amount of data to add
1451 * This function extends the used data area of the buffer at the buffer
1452 * start. If this would exceed the total buffer headroom the kernel will
1453 * panic. A pointer to the first byte of the extra data is returned.
1455 unsigned char *skb_push(struct sk_buff
*skb
, unsigned int len
)
1459 if (unlikely(skb
->data
<skb
->head
))
1460 skb_under_panic(skb
, len
, __builtin_return_address(0));
1463 EXPORT_SYMBOL(skb_push
);
1466 * skb_pull - remove data from the start of a buffer
1467 * @skb: buffer to use
1468 * @len: amount of data to remove
1470 * This function removes data from the start of a buffer, returning
1471 * the memory to the headroom. A pointer to the next data in the buffer
1472 * is returned. Once the data has been pulled future pushes will overwrite
1475 unsigned char *skb_pull(struct sk_buff
*skb
, unsigned int len
)
1477 return skb_pull_inline(skb
, len
);
1479 EXPORT_SYMBOL(skb_pull
);
1482 * skb_trim - remove end from a buffer
1483 * @skb: buffer to alter
1486 * Cut the length of a buffer down by removing data from the tail. If
1487 * the buffer is already under the length specified it is not modified.
1488 * The skb must be linear.
1490 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1493 __skb_trim(skb
, len
);
1495 EXPORT_SYMBOL(skb_trim
);
1497 /* Trims skb to length len. It can change skb pointers.
1500 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1502 struct sk_buff
**fragp
;
1503 struct sk_buff
*frag
;
1504 int offset
= skb_headlen(skb
);
1505 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1509 if (skb_cloned(skb
) &&
1510 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1517 for (; i
< nfrags
; i
++) {
1518 int end
= offset
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1525 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
++], len
- offset
);
1528 skb_shinfo(skb
)->nr_frags
= i
;
1530 for (; i
< nfrags
; i
++)
1531 skb_frag_unref(skb
, i
);
1533 if (skb_has_frag_list(skb
))
1534 skb_drop_fraglist(skb
);
1538 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1539 fragp
= &frag
->next
) {
1540 int end
= offset
+ frag
->len
;
1542 if (skb_shared(frag
)) {
1543 struct sk_buff
*nfrag
;
1545 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1546 if (unlikely(!nfrag
))
1549 nfrag
->next
= frag
->next
;
1561 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1565 skb_drop_list(&frag
->next
);
1570 if (len
> skb_headlen(skb
)) {
1571 skb
->data_len
-= skb
->len
- len
;
1576 skb_set_tail_pointer(skb
, len
);
1579 if (!skb
->sk
|| skb
->destructor
== sock_edemux
)
1583 EXPORT_SYMBOL(___pskb_trim
);
1586 * __pskb_pull_tail - advance tail of skb header
1587 * @skb: buffer to reallocate
1588 * @delta: number of bytes to advance tail
1590 * The function makes a sense only on a fragmented &sk_buff,
1591 * it expands header moving its tail forward and copying necessary
1592 * data from fragmented part.
1594 * &sk_buff MUST have reference count of 1.
1596 * Returns %NULL (and &sk_buff does not change) if pull failed
1597 * or value of new tail of skb in the case of success.
1599 * All the pointers pointing into skb header may change and must be
1600 * reloaded after call to this function.
1603 /* Moves tail of skb head forward, copying data from fragmented part,
1604 * when it is necessary.
1605 * 1. It may fail due to malloc failure.
1606 * 2. It may change skb pointers.
1608 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1610 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1612 /* If skb has not enough free space at tail, get new one
1613 * plus 128 bytes for future expansions. If we have enough
1614 * room at tail, reallocate without expansion only if skb is cloned.
1616 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1618 if (eat
> 0 || skb_cloned(skb
)) {
1619 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1624 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
1627 /* Optimization: no fragments, no reasons to preestimate
1628 * size of pulled pages. Superb.
1630 if (!skb_has_frag_list(skb
))
1633 /* Estimate size of pulled pages. */
1635 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1636 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1643 /* If we need update frag list, we are in troubles.
1644 * Certainly, it possible to add an offset to skb data,
1645 * but taking into account that pulling is expected to
1646 * be very rare operation, it is worth to fight against
1647 * further bloating skb head and crucify ourselves here instead.
1648 * Pure masohism, indeed. 8)8)
1651 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1652 struct sk_buff
*clone
= NULL
;
1653 struct sk_buff
*insp
= NULL
;
1658 if (list
->len
<= eat
) {
1659 /* Eaten as whole. */
1664 /* Eaten partially. */
1666 if (skb_shared(list
)) {
1667 /* Sucks! We need to fork list. :-( */
1668 clone
= skb_clone(list
, GFP_ATOMIC
);
1674 /* This may be pulled without
1678 if (!pskb_pull(list
, eat
)) {
1686 /* Free pulled out fragments. */
1687 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1688 skb_shinfo(skb
)->frag_list
= list
->next
;
1691 /* And insert new clone at head. */
1694 skb_shinfo(skb
)->frag_list
= clone
;
1697 /* Success! Now we may commit changes to skb data. */
1702 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1703 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1706 skb_frag_unref(skb
, i
);
1709 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1711 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1712 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[k
], eat
);
1718 skb_shinfo(skb
)->nr_frags
= k
;
1721 skb
->data_len
-= delta
;
1723 return skb_tail_pointer(skb
);
1725 EXPORT_SYMBOL(__pskb_pull_tail
);
1728 * skb_copy_bits - copy bits from skb to kernel buffer
1730 * @offset: offset in source
1731 * @to: destination buffer
1732 * @len: number of bytes to copy
1734 * Copy the specified number of bytes from the source skb to the
1735 * destination buffer.
1738 * If its prototype is ever changed,
1739 * check arch/{*}/net/{*}.S files,
1740 * since it is called from BPF assembly code.
1742 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1744 int start
= skb_headlen(skb
);
1745 struct sk_buff
*frag_iter
;
1748 if (offset
> (int)skb
->len
- len
)
1752 if ((copy
= start
- offset
) > 0) {
1755 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
1756 if ((len
-= copy
) == 0)
1762 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1764 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
1766 WARN_ON(start
> offset
+ len
);
1768 end
= start
+ skb_frag_size(f
);
1769 if ((copy
= end
- offset
) > 0) {
1775 vaddr
= kmap_atomic(skb_frag_page(f
));
1777 vaddr
+ f
->page_offset
+ offset
- start
,
1779 kunmap_atomic(vaddr
);
1781 if ((len
-= copy
) == 0)
1789 skb_walk_frags(skb
, frag_iter
) {
1792 WARN_ON(start
> offset
+ len
);
1794 end
= start
+ frag_iter
->len
;
1795 if ((copy
= end
- offset
) > 0) {
1798 if (skb_copy_bits(frag_iter
, offset
- start
, to
, copy
))
1800 if ((len
-= copy
) == 0)
1814 EXPORT_SYMBOL(skb_copy_bits
);
1817 * Callback from splice_to_pipe(), if we need to release some pages
1818 * at the end of the spd in case we error'ed out in filling the pipe.
1820 static void sock_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
1822 put_page(spd
->pages
[i
]);
1825 static struct page
*linear_to_page(struct page
*page
, unsigned int *len
,
1826 unsigned int *offset
,
1829 struct page_frag
*pfrag
= sk_page_frag(sk
);
1831 if (!sk_page_frag_refill(sk
, pfrag
))
1834 *len
= min_t(unsigned int, *len
, pfrag
->size
- pfrag
->offset
);
1836 memcpy(page_address(pfrag
->page
) + pfrag
->offset
,
1837 page_address(page
) + *offset
, *len
);
1838 *offset
= pfrag
->offset
;
1839 pfrag
->offset
+= *len
;
1844 static bool spd_can_coalesce(const struct splice_pipe_desc
*spd
,
1846 unsigned int offset
)
1848 return spd
->nr_pages
&&
1849 spd
->pages
[spd
->nr_pages
- 1] == page
&&
1850 (spd
->partial
[spd
->nr_pages
- 1].offset
+
1851 spd
->partial
[spd
->nr_pages
- 1].len
== offset
);
1855 * Fill page/offset/length into spd, if it can hold more pages.
1857 static bool spd_fill_page(struct splice_pipe_desc
*spd
,
1858 struct pipe_inode_info
*pipe
, struct page
*page
,
1859 unsigned int *len
, unsigned int offset
,
1863 if (unlikely(spd
->nr_pages
== MAX_SKB_FRAGS
))
1867 page
= linear_to_page(page
, len
, &offset
, sk
);
1871 if (spd_can_coalesce(spd
, page
, offset
)) {
1872 spd
->partial
[spd
->nr_pages
- 1].len
+= *len
;
1876 spd
->pages
[spd
->nr_pages
] = page
;
1877 spd
->partial
[spd
->nr_pages
].len
= *len
;
1878 spd
->partial
[spd
->nr_pages
].offset
= offset
;
1884 static bool __splice_segment(struct page
*page
, unsigned int poff
,
1885 unsigned int plen
, unsigned int *off
,
1887 struct splice_pipe_desc
*spd
, bool linear
,
1889 struct pipe_inode_info
*pipe
)
1894 /* skip this segment if already processed */
1900 /* ignore any bits we already processed */
1906 unsigned int flen
= min(*len
, plen
);
1908 if (spd_fill_page(spd
, pipe
, page
, &flen
, poff
,
1914 } while (*len
&& plen
);
1920 * Map linear and fragment data from the skb to spd. It reports true if the
1921 * pipe is full or if we already spliced the requested length.
1923 static bool __skb_splice_bits(struct sk_buff
*skb
, struct pipe_inode_info
*pipe
,
1924 unsigned int *offset
, unsigned int *len
,
1925 struct splice_pipe_desc
*spd
, struct sock
*sk
)
1928 struct sk_buff
*iter
;
1930 /* map the linear part :
1931 * If skb->head_frag is set, this 'linear' part is backed by a
1932 * fragment, and if the head is not shared with any clones then
1933 * we can avoid a copy since we own the head portion of this page.
1935 if (__splice_segment(virt_to_page(skb
->data
),
1936 (unsigned long) skb
->data
& (PAGE_SIZE
- 1),
1939 skb_head_is_locked(skb
),
1944 * then map the fragments
1946 for (seg
= 0; seg
< skb_shinfo(skb
)->nr_frags
; seg
++) {
1947 const skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[seg
];
1949 if (__splice_segment(skb_frag_page(f
),
1950 f
->page_offset
, skb_frag_size(f
),
1951 offset
, len
, spd
, false, sk
, pipe
))
1955 skb_walk_frags(skb
, iter
) {
1956 if (*offset
>= iter
->len
) {
1957 *offset
-= iter
->len
;
1960 /* __skb_splice_bits() only fails if the output has no room
1961 * left, so no point in going over the frag_list for the error
1964 if (__skb_splice_bits(iter
, pipe
, offset
, len
, spd
, sk
))
1972 * Map data from the skb to a pipe. Should handle both the linear part,
1973 * the fragments, and the frag list.
1975 int skb_splice_bits(struct sk_buff
*skb
, struct sock
*sk
, unsigned int offset
,
1976 struct pipe_inode_info
*pipe
, unsigned int tlen
,
1979 struct partial_page partial
[MAX_SKB_FRAGS
];
1980 struct page
*pages
[MAX_SKB_FRAGS
];
1981 struct splice_pipe_desc spd
= {
1984 .nr_pages_max
= MAX_SKB_FRAGS
,
1985 .ops
= &nosteal_pipe_buf_ops
,
1986 .spd_release
= sock_spd_release
,
1990 __skb_splice_bits(skb
, pipe
, &offset
, &tlen
, &spd
, sk
);
1993 ret
= splice_to_pipe(pipe
, &spd
);
1997 EXPORT_SYMBOL_GPL(skb_splice_bits
);
2000 * skb_store_bits - store bits from kernel buffer to skb
2001 * @skb: destination buffer
2002 * @offset: offset in destination
2003 * @from: source buffer
2004 * @len: number of bytes to copy
2006 * Copy the specified number of bytes from the source buffer to the
2007 * destination skb. This function handles all the messy bits of
2008 * traversing fragment lists and such.
2011 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
2013 int start
= skb_headlen(skb
);
2014 struct sk_buff
*frag_iter
;
2017 if (offset
> (int)skb
->len
- len
)
2020 if ((copy
= start
- offset
) > 0) {
2023 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
2024 if ((len
-= copy
) == 0)
2030 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2031 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2034 WARN_ON(start
> offset
+ len
);
2036 end
= start
+ skb_frag_size(frag
);
2037 if ((copy
= end
- offset
) > 0) {
2043 vaddr
= kmap_atomic(skb_frag_page(frag
));
2044 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
2046 kunmap_atomic(vaddr
);
2048 if ((len
-= copy
) == 0)
2056 skb_walk_frags(skb
, frag_iter
) {
2059 WARN_ON(start
> offset
+ len
);
2061 end
= start
+ frag_iter
->len
;
2062 if ((copy
= end
- offset
) > 0) {
2065 if (skb_store_bits(frag_iter
, offset
- start
,
2068 if ((len
-= copy
) == 0)
2081 EXPORT_SYMBOL(skb_store_bits
);
2083 /* Checksum skb data. */
2084 __wsum
__skb_checksum(const struct sk_buff
*skb
, int offset
, int len
,
2085 __wsum csum
, const struct skb_checksum_ops
*ops
)
2087 int start
= skb_headlen(skb
);
2088 int i
, copy
= start
- offset
;
2089 struct sk_buff
*frag_iter
;
2092 /* Checksum header. */
2096 csum
= ops
->update(skb
->data
+ offset
, copy
, csum
);
2097 if ((len
-= copy
) == 0)
2103 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2105 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2107 WARN_ON(start
> offset
+ len
);
2109 end
= start
+ skb_frag_size(frag
);
2110 if ((copy
= end
- offset
) > 0) {
2116 vaddr
= kmap_atomic(skb_frag_page(frag
));
2117 csum2
= ops
->update(vaddr
+ frag
->page_offset
+
2118 offset
- start
, copy
, 0);
2119 kunmap_atomic(vaddr
);
2120 csum
= ops
->combine(csum
, csum2
, pos
, copy
);
2129 skb_walk_frags(skb
, frag_iter
) {
2132 WARN_ON(start
> offset
+ len
);
2134 end
= start
+ frag_iter
->len
;
2135 if ((copy
= end
- offset
) > 0) {
2139 csum2
= __skb_checksum(frag_iter
, offset
- start
,
2141 csum
= ops
->combine(csum
, csum2
, pos
, copy
);
2142 if ((len
-= copy
) == 0)
2153 EXPORT_SYMBOL(__skb_checksum
);
2155 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
2156 int len
, __wsum csum
)
2158 const struct skb_checksum_ops ops
= {
2159 .update
= csum_partial_ext
,
2160 .combine
= csum_block_add_ext
,
2163 return __skb_checksum(skb
, offset
, len
, csum
, &ops
);
2165 EXPORT_SYMBOL(skb_checksum
);
2167 /* Both of above in one bottle. */
2169 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
2170 u8
*to
, int len
, __wsum csum
)
2172 int start
= skb_headlen(skb
);
2173 int i
, copy
= start
- offset
;
2174 struct sk_buff
*frag_iter
;
2181 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
2183 if ((len
-= copy
) == 0)
2190 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2193 WARN_ON(start
> offset
+ len
);
2195 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
2196 if ((copy
= end
- offset
) > 0) {
2199 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2203 vaddr
= kmap_atomic(skb_frag_page(frag
));
2204 csum2
= csum_partial_copy_nocheck(vaddr
+
2208 kunmap_atomic(vaddr
);
2209 csum
= csum_block_add(csum
, csum2
, pos
);
2219 skb_walk_frags(skb
, frag_iter
) {
2223 WARN_ON(start
> offset
+ len
);
2225 end
= start
+ frag_iter
->len
;
2226 if ((copy
= end
- offset
) > 0) {
2229 csum2
= skb_copy_and_csum_bits(frag_iter
,
2232 csum
= csum_block_add(csum
, csum2
, pos
);
2233 if ((len
-= copy
) == 0)
2244 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
2247 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2248 * @from: source buffer
2250 * Calculates the amount of linear headroom needed in the 'to' skb passed
2251 * into skb_zerocopy().
2254 skb_zerocopy_headlen(const struct sk_buff
*from
)
2256 unsigned int hlen
= 0;
2258 if (!from
->head_frag
||
2259 skb_headlen(from
) < L1_CACHE_BYTES
||
2260 skb_shinfo(from
)->nr_frags
>= MAX_SKB_FRAGS
)
2261 hlen
= skb_headlen(from
);
2263 if (skb_has_frag_list(from
))
2268 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen
);
2271 * skb_zerocopy - Zero copy skb to skb
2272 * @to: destination buffer
2273 * @from: source buffer
2274 * @len: number of bytes to copy from source buffer
2275 * @hlen: size of linear headroom in destination buffer
2277 * Copies up to `len` bytes from `from` to `to` by creating references
2278 * to the frags in the source buffer.
2280 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2281 * headroom in the `to` buffer.
2284 * 0: everything is OK
2285 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2286 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2289 skb_zerocopy(struct sk_buff
*to
, struct sk_buff
*from
, int len
, int hlen
)
2292 int plen
= 0; /* length of skb->head fragment */
2295 unsigned int offset
;
2297 BUG_ON(!from
->head_frag
&& !hlen
);
2299 /* dont bother with small payloads */
2300 if (len
<= skb_tailroom(to
))
2301 return skb_copy_bits(from
, 0, skb_put(to
, len
), len
);
2304 ret
= skb_copy_bits(from
, 0, skb_put(to
, hlen
), hlen
);
2309 plen
= min_t(int, skb_headlen(from
), len
);
2311 page
= virt_to_head_page(from
->head
);
2312 offset
= from
->data
- (unsigned char *)page_address(page
);
2313 __skb_fill_page_desc(to
, 0, page
, offset
, plen
);
2320 to
->truesize
+= len
+ plen
;
2321 to
->len
+= len
+ plen
;
2322 to
->data_len
+= len
+ plen
;
2324 if (unlikely(skb_orphan_frags(from
, GFP_ATOMIC
))) {
2329 for (i
= 0; i
< skb_shinfo(from
)->nr_frags
; i
++) {
2332 skb_shinfo(to
)->frags
[j
] = skb_shinfo(from
)->frags
[i
];
2333 skb_shinfo(to
)->frags
[j
].size
= min_t(int, skb_shinfo(to
)->frags
[j
].size
, len
);
2334 len
-= skb_shinfo(to
)->frags
[j
].size
;
2335 skb_frag_ref(to
, j
);
2338 skb_shinfo(to
)->nr_frags
= j
;
2342 EXPORT_SYMBOL_GPL(skb_zerocopy
);
2344 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
2349 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
2350 csstart
= skb_checksum_start_offset(skb
);
2352 csstart
= skb_headlen(skb
);
2354 BUG_ON(csstart
> skb_headlen(skb
));
2356 skb_copy_from_linear_data(skb
, to
, csstart
);
2359 if (csstart
!= skb
->len
)
2360 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
2361 skb
->len
- csstart
, 0);
2363 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2364 long csstuff
= csstart
+ skb
->csum_offset
;
2366 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
2369 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
2372 * skb_dequeue - remove from the head of the queue
2373 * @list: list to dequeue from
2375 * Remove the head of the list. The list lock is taken so the function
2376 * may be used safely with other locking list functions. The head item is
2377 * returned or %NULL if the list is empty.
2380 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
2382 unsigned long flags
;
2383 struct sk_buff
*result
;
2385 spin_lock_irqsave(&list
->lock
, flags
);
2386 result
= __skb_dequeue(list
);
2387 spin_unlock_irqrestore(&list
->lock
, flags
);
2390 EXPORT_SYMBOL(skb_dequeue
);
2393 * skb_dequeue_tail - remove from the tail of the queue
2394 * @list: list to dequeue from
2396 * Remove the tail of the list. The list lock is taken so the function
2397 * may be used safely with other locking list functions. The tail item is
2398 * returned or %NULL if the list is empty.
2400 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
2402 unsigned long flags
;
2403 struct sk_buff
*result
;
2405 spin_lock_irqsave(&list
->lock
, flags
);
2406 result
= __skb_dequeue_tail(list
);
2407 spin_unlock_irqrestore(&list
->lock
, flags
);
2410 EXPORT_SYMBOL(skb_dequeue_tail
);
2413 * skb_queue_purge - empty a list
2414 * @list: list to empty
2416 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2417 * the list and one reference dropped. This function takes the list
2418 * lock and is atomic with respect to other list locking functions.
2420 void skb_queue_purge(struct sk_buff_head
*list
)
2422 struct sk_buff
*skb
;
2423 while ((skb
= skb_dequeue(list
)) != NULL
)
2426 EXPORT_SYMBOL(skb_queue_purge
);
2429 * skb_rbtree_purge - empty a skb rbtree
2430 * @root: root of the rbtree to empty
2432 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
2433 * the list and one reference dropped. This function does not take
2434 * any lock. Synchronization should be handled by the caller (e.g., TCP
2435 * out-of-order queue is protected by the socket lock).
2437 void skb_rbtree_purge(struct rb_root
*root
)
2439 struct sk_buff
*skb
, *next
;
2441 rbtree_postorder_for_each_entry_safe(skb
, next
, root
, rbnode
)
2448 * skb_queue_head - queue a buffer at the list head
2449 * @list: list to use
2450 * @newsk: buffer to queue
2452 * Queue a buffer at the start of the list. This function takes the
2453 * list lock and can be used safely with other locking &sk_buff functions
2456 * A buffer cannot be placed on two lists at the same time.
2458 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2460 unsigned long flags
;
2462 spin_lock_irqsave(&list
->lock
, flags
);
2463 __skb_queue_head(list
, newsk
);
2464 spin_unlock_irqrestore(&list
->lock
, flags
);
2466 EXPORT_SYMBOL(skb_queue_head
);
2469 * skb_queue_tail - queue a buffer at the list tail
2470 * @list: list to use
2471 * @newsk: buffer to queue
2473 * Queue a buffer at the tail of the list. This function takes the
2474 * list lock and can be used safely with other locking &sk_buff functions
2477 * A buffer cannot be placed on two lists at the same time.
2479 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2481 unsigned long flags
;
2483 spin_lock_irqsave(&list
->lock
, flags
);
2484 __skb_queue_tail(list
, newsk
);
2485 spin_unlock_irqrestore(&list
->lock
, flags
);
2487 EXPORT_SYMBOL(skb_queue_tail
);
2490 * skb_unlink - remove a buffer from a list
2491 * @skb: buffer to remove
2492 * @list: list to use
2494 * Remove a packet from a list. The list locks are taken and this
2495 * function is atomic with respect to other list locked calls
2497 * You must know what list the SKB is on.
2499 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
2501 unsigned long flags
;
2503 spin_lock_irqsave(&list
->lock
, flags
);
2504 __skb_unlink(skb
, list
);
2505 spin_unlock_irqrestore(&list
->lock
, flags
);
2507 EXPORT_SYMBOL(skb_unlink
);
2510 * skb_append - append a buffer
2511 * @old: buffer to insert after
2512 * @newsk: buffer to insert
2513 * @list: list to use
2515 * Place a packet after a given packet in a list. The list locks are taken
2516 * and this function is atomic with respect to other list locked calls.
2517 * A buffer cannot be placed on two lists at the same time.
2519 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2521 unsigned long flags
;
2523 spin_lock_irqsave(&list
->lock
, flags
);
2524 __skb_queue_after(list
, old
, newsk
);
2525 spin_unlock_irqrestore(&list
->lock
, flags
);
2527 EXPORT_SYMBOL(skb_append
);
2530 * skb_insert - insert a buffer
2531 * @old: buffer to insert before
2532 * @newsk: buffer to insert
2533 * @list: list to use
2535 * Place a packet before a given packet in a list. The list locks are
2536 * taken and this function is atomic with respect to other list locked
2539 * A buffer cannot be placed on two lists at the same time.
2541 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2543 unsigned long flags
;
2545 spin_lock_irqsave(&list
->lock
, flags
);
2546 __skb_insert(newsk
, old
->prev
, old
, list
);
2547 spin_unlock_irqrestore(&list
->lock
, flags
);
2549 EXPORT_SYMBOL(skb_insert
);
2551 static inline void skb_split_inside_header(struct sk_buff
*skb
,
2552 struct sk_buff
* skb1
,
2553 const u32 len
, const int pos
)
2557 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
2559 /* And move data appendix as is. */
2560 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
2561 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
2563 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
2564 skb_shinfo(skb
)->nr_frags
= 0;
2565 skb1
->data_len
= skb
->data_len
;
2566 skb1
->len
+= skb1
->data_len
;
2569 skb_set_tail_pointer(skb
, len
);
2572 static inline void skb_split_no_header(struct sk_buff
*skb
,
2573 struct sk_buff
* skb1
,
2574 const u32 len
, int pos
)
2577 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
2579 skb_shinfo(skb
)->nr_frags
= 0;
2580 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
2582 skb
->data_len
= len
- pos
;
2584 for (i
= 0; i
< nfrags
; i
++) {
2585 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
2587 if (pos
+ size
> len
) {
2588 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
2592 * We have two variants in this case:
2593 * 1. Move all the frag to the second
2594 * part, if it is possible. F.e.
2595 * this approach is mandatory for TUX,
2596 * where splitting is expensive.
2597 * 2. Split is accurately. We make this.
2599 skb_frag_ref(skb
, i
);
2600 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
2601 skb_frag_size_sub(&skb_shinfo(skb1
)->frags
[0], len
- pos
);
2602 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
], len
- pos
);
2603 skb_shinfo(skb
)->nr_frags
++;
2607 skb_shinfo(skb
)->nr_frags
++;
2610 skb_shinfo(skb1
)->nr_frags
= k
;
2614 * skb_split - Split fragmented skb to two parts at length len.
2615 * @skb: the buffer to split
2616 * @skb1: the buffer to receive the second part
2617 * @len: new length for skb
2619 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
2621 int pos
= skb_headlen(skb
);
2623 skb_shinfo(skb1
)->tx_flags
= skb_shinfo(skb
)->tx_flags
& SKBTX_SHARED_FRAG
;
2624 if (len
< pos
) /* Split line is inside header. */
2625 skb_split_inside_header(skb
, skb1
, len
, pos
);
2626 else /* Second chunk has no header, nothing to copy. */
2627 skb_split_no_header(skb
, skb1
, len
, pos
);
2629 EXPORT_SYMBOL(skb_split
);
2631 /* Shifting from/to a cloned skb is a no-go.
2633 * Caller cannot keep skb_shinfo related pointers past calling here!
2635 static int skb_prepare_for_shift(struct sk_buff
*skb
)
2637 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2641 * skb_shift - Shifts paged data partially from skb to another
2642 * @tgt: buffer into which tail data gets added
2643 * @skb: buffer from which the paged data comes from
2644 * @shiftlen: shift up to this many bytes
2646 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2647 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2648 * It's up to caller to free skb if everything was shifted.
2650 * If @tgt runs out of frags, the whole operation is aborted.
2652 * Skb cannot include anything else but paged data while tgt is allowed
2653 * to have non-paged data as well.
2655 * TODO: full sized shift could be optimized but that would need
2656 * specialized skb free'er to handle frags without up-to-date nr_frags.
2658 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
2660 int from
, to
, merge
, todo
;
2661 struct skb_frag_struct
*fragfrom
, *fragto
;
2663 BUG_ON(shiftlen
> skb
->len
);
2665 if (skb_headlen(skb
))
2670 to
= skb_shinfo(tgt
)->nr_frags
;
2671 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2673 /* Actual merge is delayed until the point when we know we can
2674 * commit all, so that we don't have to undo partial changes
2677 !skb_can_coalesce(tgt
, to
, skb_frag_page(fragfrom
),
2678 fragfrom
->page_offset
)) {
2683 todo
-= skb_frag_size(fragfrom
);
2685 if (skb_prepare_for_shift(skb
) ||
2686 skb_prepare_for_shift(tgt
))
2689 /* All previous frag pointers might be stale! */
2690 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2691 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2693 skb_frag_size_add(fragto
, shiftlen
);
2694 skb_frag_size_sub(fragfrom
, shiftlen
);
2695 fragfrom
->page_offset
+= shiftlen
;
2703 /* Skip full, not-fitting skb to avoid expensive operations */
2704 if ((shiftlen
== skb
->len
) &&
2705 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
2708 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
2711 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
2712 if (to
== MAX_SKB_FRAGS
)
2715 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2716 fragto
= &skb_shinfo(tgt
)->frags
[to
];
2718 if (todo
>= skb_frag_size(fragfrom
)) {
2719 *fragto
= *fragfrom
;
2720 todo
-= skb_frag_size(fragfrom
);
2725 __skb_frag_ref(fragfrom
);
2726 fragto
->page
= fragfrom
->page
;
2727 fragto
->page_offset
= fragfrom
->page_offset
;
2728 skb_frag_size_set(fragto
, todo
);
2730 fragfrom
->page_offset
+= todo
;
2731 skb_frag_size_sub(fragfrom
, todo
);
2739 /* Ready to "commit" this state change to tgt */
2740 skb_shinfo(tgt
)->nr_frags
= to
;
2743 fragfrom
= &skb_shinfo(skb
)->frags
[0];
2744 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2746 skb_frag_size_add(fragto
, skb_frag_size(fragfrom
));
2747 __skb_frag_unref(fragfrom
);
2750 /* Reposition in the original skb */
2752 while (from
< skb_shinfo(skb
)->nr_frags
)
2753 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
2754 skb_shinfo(skb
)->nr_frags
= to
;
2756 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
2759 /* Most likely the tgt won't ever need its checksum anymore, skb on
2760 * the other hand might need it if it needs to be resent
2762 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
2763 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2765 /* Yak, is it really working this way? Some helper please? */
2766 skb
->len
-= shiftlen
;
2767 skb
->data_len
-= shiftlen
;
2768 skb
->truesize
-= shiftlen
;
2769 tgt
->len
+= shiftlen
;
2770 tgt
->data_len
+= shiftlen
;
2771 tgt
->truesize
+= shiftlen
;
2777 * skb_prepare_seq_read - Prepare a sequential read of skb data
2778 * @skb: the buffer to read
2779 * @from: lower offset of data to be read
2780 * @to: upper offset of data to be read
2781 * @st: state variable
2783 * Initializes the specified state variable. Must be called before
2784 * invoking skb_seq_read() for the first time.
2786 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
2787 unsigned int to
, struct skb_seq_state
*st
)
2789 st
->lower_offset
= from
;
2790 st
->upper_offset
= to
;
2791 st
->root_skb
= st
->cur_skb
= skb
;
2792 st
->frag_idx
= st
->stepped_offset
= 0;
2793 st
->frag_data
= NULL
;
2795 EXPORT_SYMBOL(skb_prepare_seq_read
);
2798 * skb_seq_read - Sequentially read skb data
2799 * @consumed: number of bytes consumed by the caller so far
2800 * @data: destination pointer for data to be returned
2801 * @st: state variable
2803 * Reads a block of skb data at @consumed relative to the
2804 * lower offset specified to skb_prepare_seq_read(). Assigns
2805 * the head of the data block to @data and returns the length
2806 * of the block or 0 if the end of the skb data or the upper
2807 * offset has been reached.
2809 * The caller is not required to consume all of the data
2810 * returned, i.e. @consumed is typically set to the number
2811 * of bytes already consumed and the next call to
2812 * skb_seq_read() will return the remaining part of the block.
2814 * Note 1: The size of each block of data returned can be arbitrary,
2815 * this limitation is the cost for zerocopy sequential
2816 * reads of potentially non linear data.
2818 * Note 2: Fragment lists within fragments are not implemented
2819 * at the moment, state->root_skb could be replaced with
2820 * a stack for this purpose.
2822 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
2823 struct skb_seq_state
*st
)
2825 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
2828 if (unlikely(abs_offset
>= st
->upper_offset
)) {
2829 if (st
->frag_data
) {
2830 kunmap_atomic(st
->frag_data
);
2831 st
->frag_data
= NULL
;
2837 block_limit
= skb_headlen(st
->cur_skb
) + st
->stepped_offset
;
2839 if (abs_offset
< block_limit
&& !st
->frag_data
) {
2840 *data
= st
->cur_skb
->data
+ (abs_offset
- st
->stepped_offset
);
2841 return block_limit
- abs_offset
;
2844 if (st
->frag_idx
== 0 && !st
->frag_data
)
2845 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
2847 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
2848 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
2849 block_limit
= skb_frag_size(frag
) + st
->stepped_offset
;
2851 if (abs_offset
< block_limit
) {
2853 st
->frag_data
= kmap_atomic(skb_frag_page(frag
));
2855 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
2856 (abs_offset
- st
->stepped_offset
);
2858 return block_limit
- abs_offset
;
2861 if (st
->frag_data
) {
2862 kunmap_atomic(st
->frag_data
);
2863 st
->frag_data
= NULL
;
2867 st
->stepped_offset
+= skb_frag_size(frag
);
2870 if (st
->frag_data
) {
2871 kunmap_atomic(st
->frag_data
);
2872 st
->frag_data
= NULL
;
2875 if (st
->root_skb
== st
->cur_skb
&& skb_has_frag_list(st
->root_skb
)) {
2876 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
2879 } else if (st
->cur_skb
->next
) {
2880 st
->cur_skb
= st
->cur_skb
->next
;
2887 EXPORT_SYMBOL(skb_seq_read
);
2890 * skb_abort_seq_read - Abort a sequential read of skb data
2891 * @st: state variable
2893 * Must be called if skb_seq_read() was not called until it
2896 void skb_abort_seq_read(struct skb_seq_state
*st
)
2899 kunmap_atomic(st
->frag_data
);
2901 EXPORT_SYMBOL(skb_abort_seq_read
);
2903 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2905 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
2906 struct ts_config
*conf
,
2907 struct ts_state
*state
)
2909 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
2912 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
2914 skb_abort_seq_read(TS_SKB_CB(state
));
2918 * skb_find_text - Find a text pattern in skb data
2919 * @skb: the buffer to look in
2920 * @from: search offset
2922 * @config: textsearch configuration
2924 * Finds a pattern in the skb data according to the specified
2925 * textsearch configuration. Use textsearch_next() to retrieve
2926 * subsequent occurrences of the pattern. Returns the offset
2927 * to the first occurrence or UINT_MAX if no match was found.
2929 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
2930 unsigned int to
, struct ts_config
*config
)
2932 struct ts_state state
;
2935 config
->get_next_block
= skb_ts_get_next_block
;
2936 config
->finish
= skb_ts_finish
;
2938 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(&state
));
2940 ret
= textsearch_find(config
, &state
);
2941 return (ret
<= to
- from
? ret
: UINT_MAX
);
2943 EXPORT_SYMBOL(skb_find_text
);
2946 * skb_append_datato_frags - append the user data to a skb
2947 * @sk: sock structure
2948 * @skb: skb structure to be appended with user data.
2949 * @getfrag: call back function to be used for getting the user data
2950 * @from: pointer to user message iov
2951 * @length: length of the iov message
2953 * Description: This procedure append the user data in the fragment part
2954 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2956 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
2957 int (*getfrag
)(void *from
, char *to
, int offset
,
2958 int len
, int odd
, struct sk_buff
*skb
),
2959 void *from
, int length
)
2961 int frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2965 struct page_frag
*pfrag
= ¤t
->task_frag
;
2968 /* Return error if we don't have space for new frag */
2969 if (frg_cnt
>= MAX_SKB_FRAGS
)
2972 if (!sk_page_frag_refill(sk
, pfrag
))
2975 /* copy the user data to page */
2976 copy
= min_t(int, length
, pfrag
->size
- pfrag
->offset
);
2978 ret
= getfrag(from
, page_address(pfrag
->page
) + pfrag
->offset
,
2979 offset
, copy
, 0, skb
);
2983 /* copy was successful so update the size parameters */
2984 skb_fill_page_desc(skb
, frg_cnt
, pfrag
->page
, pfrag
->offset
,
2987 pfrag
->offset
+= copy
;
2988 get_page(pfrag
->page
);
2990 skb
->truesize
+= copy
;
2991 atomic_add(copy
, &sk
->sk_wmem_alloc
);
2993 skb
->data_len
+= copy
;
2997 } while (length
> 0);
3001 EXPORT_SYMBOL(skb_append_datato_frags
);
3003 int skb_append_pagefrags(struct sk_buff
*skb
, struct page
*page
,
3004 int offset
, size_t size
)
3006 int i
= skb_shinfo(skb
)->nr_frags
;
3008 if (skb_can_coalesce(skb
, i
, page
, offset
)) {
3009 skb_frag_size_add(&skb_shinfo(skb
)->frags
[i
- 1], size
);
3010 } else if (i
< MAX_SKB_FRAGS
) {
3012 skb_fill_page_desc(skb
, i
, page
, offset
, size
);
3019 EXPORT_SYMBOL_GPL(skb_append_pagefrags
);
3022 * skb_pull_rcsum - pull skb and update receive checksum
3023 * @skb: buffer to update
3024 * @len: length of data pulled
3026 * This function performs an skb_pull on the packet and updates
3027 * the CHECKSUM_COMPLETE checksum. It should be used on
3028 * receive path processing instead of skb_pull unless you know
3029 * that the checksum difference is zero (e.g., a valid IP header)
3030 * or you are setting ip_summed to CHECKSUM_NONE.
3032 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
3034 unsigned char *data
= skb
->data
;
3036 BUG_ON(len
> skb
->len
);
3037 __skb_pull(skb
, len
);
3038 skb_postpull_rcsum(skb
, data
, len
);
3041 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
3044 * skb_segment - Perform protocol segmentation on skb.
3045 * @head_skb: buffer to segment
3046 * @features: features for the output path (see dev->features)
3048 * This function performs segmentation on the given skb. It returns
3049 * a pointer to the first in a list of new skbs for the segments.
3050 * In case of error it returns ERR_PTR(err).
3052 struct sk_buff
*skb_segment(struct sk_buff
*head_skb
,
3053 netdev_features_t features
)
3055 struct sk_buff
*segs
= NULL
;
3056 struct sk_buff
*tail
= NULL
;
3057 struct sk_buff
*list_skb
= skb_shinfo(head_skb
)->frag_list
;
3058 skb_frag_t
*frag
= skb_shinfo(head_skb
)->frags
;
3059 unsigned int mss
= skb_shinfo(head_skb
)->gso_size
;
3060 unsigned int doffset
= head_skb
->data
- skb_mac_header(head_skb
);
3061 struct sk_buff
*frag_skb
= head_skb
;
3062 unsigned int offset
= doffset
;
3063 unsigned int tnl_hlen
= skb_tnl_header_len(head_skb
);
3064 unsigned int partial_segs
= 0;
3065 unsigned int headroom
;
3066 unsigned int len
= head_skb
->len
;
3069 int nfrags
= skb_shinfo(head_skb
)->nr_frags
;
3075 __skb_push(head_skb
, doffset
);
3076 proto
= skb_network_protocol(head_skb
, &dummy
);
3077 if (unlikely(!proto
))
3078 return ERR_PTR(-EINVAL
);
3080 sg
= !!(features
& NETIF_F_SG
);
3081 csum
= !!can_checksum_protocol(features
, proto
);
3083 if (sg
&& csum
&& (mss
!= GSO_BY_FRAGS
)) {
3084 if (!(features
& NETIF_F_GSO_PARTIAL
)) {
3085 struct sk_buff
*iter
;
3086 unsigned int frag_len
;
3089 !net_gso_ok(features
, skb_shinfo(head_skb
)->gso_type
))
3092 /* If we get here then all the required
3093 * GSO features except frag_list are supported.
3094 * Try to split the SKB to multiple GSO SKBs
3095 * with no frag_list.
3096 * Currently we can do that only when the buffers don't
3097 * have a linear part and all the buffers except
3098 * the last are of the same length.
3100 frag_len
= list_skb
->len
;
3101 skb_walk_frags(head_skb
, iter
) {
3102 if (frag_len
!= iter
->len
&& iter
->next
)
3104 if (skb_headlen(iter
) && !iter
->head_frag
)
3110 if (len
!= frag_len
)
3114 /* GSO partial only requires that we trim off any excess that
3115 * doesn't fit into an MSS sized block, so take care of that
3118 partial_segs
= len
/ mss
;
3119 if (partial_segs
> 1)
3120 mss
*= partial_segs
;
3126 headroom
= skb_headroom(head_skb
);
3127 pos
= skb_headlen(head_skb
);
3130 struct sk_buff
*nskb
;
3131 skb_frag_t
*nskb_frag
;
3135 if (unlikely(mss
== GSO_BY_FRAGS
)) {
3136 len
= list_skb
->len
;
3138 len
= head_skb
->len
- offset
;
3143 hsize
= skb_headlen(head_skb
) - offset
;
3146 if (hsize
> len
|| !sg
)
3149 if (!hsize
&& i
>= nfrags
&& skb_headlen(list_skb
) &&
3150 (skb_headlen(list_skb
) == len
|| sg
)) {
3151 BUG_ON(skb_headlen(list_skb
) > len
);
3154 nfrags
= skb_shinfo(list_skb
)->nr_frags
;
3155 frag
= skb_shinfo(list_skb
)->frags
;
3156 frag_skb
= list_skb
;
3157 pos
+= skb_headlen(list_skb
);
3159 while (pos
< offset
+ len
) {
3160 BUG_ON(i
>= nfrags
);
3162 size
= skb_frag_size(frag
);
3163 if (pos
+ size
> offset
+ len
)
3171 nskb
= skb_clone(list_skb
, GFP_ATOMIC
);
3172 list_skb
= list_skb
->next
;
3174 if (unlikely(!nskb
))
3177 if (unlikely(pskb_trim(nskb
, len
))) {
3182 hsize
= skb_end_offset(nskb
);
3183 if (skb_cow_head(nskb
, doffset
+ headroom
)) {
3188 nskb
->truesize
+= skb_end_offset(nskb
) - hsize
;
3189 skb_release_head_state(nskb
);
3190 __skb_push(nskb
, doffset
);
3192 nskb
= __alloc_skb(hsize
+ doffset
+ headroom
,
3193 GFP_ATOMIC
, skb_alloc_rx_flag(head_skb
),
3196 if (unlikely(!nskb
))
3199 skb_reserve(nskb
, headroom
);
3200 __skb_put(nskb
, doffset
);
3209 __copy_skb_header(nskb
, head_skb
);
3211 skb_headers_offset_update(nskb
, skb_headroom(nskb
) - headroom
);
3212 skb_reset_mac_len(nskb
);
3214 skb_copy_from_linear_data_offset(head_skb
, -tnl_hlen
,
3215 nskb
->data
- tnl_hlen
,
3216 doffset
+ tnl_hlen
);
3218 if (nskb
->len
== len
+ doffset
)
3219 goto perform_csum_check
;
3222 if (!nskb
->remcsum_offload
)
3223 nskb
->ip_summed
= CHECKSUM_NONE
;
3224 SKB_GSO_CB(nskb
)->csum
=
3225 skb_copy_and_csum_bits(head_skb
, offset
,
3228 SKB_GSO_CB(nskb
)->csum_start
=
3229 skb_headroom(nskb
) + doffset
;
3233 nskb_frag
= skb_shinfo(nskb
)->frags
;
3235 skb_copy_from_linear_data_offset(head_skb
, offset
,
3236 skb_put(nskb
, hsize
), hsize
);
3238 skb_shinfo(nskb
)->tx_flags
= skb_shinfo(head_skb
)->tx_flags
&
3241 while (pos
< offset
+ len
) {
3243 BUG_ON(skb_headlen(list_skb
));
3246 nfrags
= skb_shinfo(list_skb
)->nr_frags
;
3247 frag
= skb_shinfo(list_skb
)->frags
;
3248 frag_skb
= list_skb
;
3252 list_skb
= list_skb
->next
;
3255 if (unlikely(skb_shinfo(nskb
)->nr_frags
>=
3257 net_warn_ratelimited(
3258 "skb_segment: too many frags: %u %u\n",
3263 if (unlikely(skb_orphan_frags(frag_skb
, GFP_ATOMIC
)))
3267 __skb_frag_ref(nskb_frag
);
3268 size
= skb_frag_size(nskb_frag
);
3271 nskb_frag
->page_offset
+= offset
- pos
;
3272 skb_frag_size_sub(nskb_frag
, offset
- pos
);
3275 skb_shinfo(nskb
)->nr_frags
++;
3277 if (pos
+ size
<= offset
+ len
) {
3282 skb_frag_size_sub(nskb_frag
, pos
+ size
- (offset
+ len
));
3290 nskb
->data_len
= len
- hsize
;
3291 nskb
->len
+= nskb
->data_len
;
3292 nskb
->truesize
+= nskb
->data_len
;
3296 if (skb_has_shared_frag(nskb
)) {
3297 err
= __skb_linearize(nskb
);
3301 if (!nskb
->remcsum_offload
)
3302 nskb
->ip_summed
= CHECKSUM_NONE
;
3303 SKB_GSO_CB(nskb
)->csum
=
3304 skb_checksum(nskb
, doffset
,
3305 nskb
->len
- doffset
, 0);
3306 SKB_GSO_CB(nskb
)->csum_start
=
3307 skb_headroom(nskb
) + doffset
;
3309 } while ((offset
+= len
) < head_skb
->len
);
3311 /* Some callers want to get the end of the list.
3312 * Put it in segs->prev to avoid walking the list.
3313 * (see validate_xmit_skb_list() for example)
3318 struct sk_buff
*iter
;
3319 int type
= skb_shinfo(head_skb
)->gso_type
;
3320 unsigned short gso_size
= skb_shinfo(head_skb
)->gso_size
;
3322 /* Update type to add partial and then remove dodgy if set */
3323 type
|= (features
& NETIF_F_GSO_PARTIAL
) / NETIF_F_GSO_PARTIAL
* SKB_GSO_PARTIAL
;
3324 type
&= ~SKB_GSO_DODGY
;
3326 /* Update GSO info and prepare to start updating headers on
3327 * our way back down the stack of protocols.
3329 for (iter
= segs
; iter
; iter
= iter
->next
) {
3330 skb_shinfo(iter
)->gso_size
= gso_size
;
3331 skb_shinfo(iter
)->gso_segs
= partial_segs
;
3332 skb_shinfo(iter
)->gso_type
= type
;
3333 SKB_GSO_CB(iter
)->data_offset
= skb_headroom(iter
) + doffset
;
3336 if (tail
->len
- doffset
<= gso_size
)
3337 skb_shinfo(tail
)->gso_size
= 0;
3338 else if (tail
!= segs
)
3339 skb_shinfo(tail
)->gso_segs
= DIV_ROUND_UP(tail
->len
- doffset
, gso_size
);
3342 /* Following permits correct backpressure, for protocols
3343 * using skb_set_owner_w().
3344 * Idea is to tranfert ownership from head_skb to last segment.
3346 if (head_skb
->destructor
== sock_wfree
) {
3347 swap(tail
->truesize
, head_skb
->truesize
);
3348 swap(tail
->destructor
, head_skb
->destructor
);
3349 swap(tail
->sk
, head_skb
->sk
);
3354 kfree_skb_list(segs
);
3355 return ERR_PTR(err
);
3357 EXPORT_SYMBOL_GPL(skb_segment
);
3359 int skb_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
3361 struct skb_shared_info
*pinfo
, *skbinfo
= skb_shinfo(skb
);
3362 unsigned int offset
= skb_gro_offset(skb
);
3363 unsigned int headlen
= skb_headlen(skb
);
3364 unsigned int len
= skb_gro_len(skb
);
3365 struct sk_buff
*lp
, *p
= *head
;
3366 unsigned int delta_truesize
;
3368 if (unlikely(p
->len
+ len
>= 65536))
3371 lp
= NAPI_GRO_CB(p
)->last
;
3372 pinfo
= skb_shinfo(lp
);
3374 if (headlen
<= offset
) {
3377 int i
= skbinfo
->nr_frags
;
3378 int nr_frags
= pinfo
->nr_frags
+ i
;
3380 if (nr_frags
> MAX_SKB_FRAGS
)
3384 pinfo
->nr_frags
= nr_frags
;
3385 skbinfo
->nr_frags
= 0;
3387 frag
= pinfo
->frags
+ nr_frags
;
3388 frag2
= skbinfo
->frags
+ i
;
3393 frag
->page_offset
+= offset
;
3394 skb_frag_size_sub(frag
, offset
);
3396 /* all fragments truesize : remove (head size + sk_buff) */
3397 delta_truesize
= skb
->truesize
-
3398 SKB_TRUESIZE(skb_end_offset(skb
));
3400 skb
->truesize
-= skb
->data_len
;
3401 skb
->len
-= skb
->data_len
;
3404 NAPI_GRO_CB(skb
)->free
= NAPI_GRO_FREE
;
3406 } else if (skb
->head_frag
) {
3407 int nr_frags
= pinfo
->nr_frags
;
3408 skb_frag_t
*frag
= pinfo
->frags
+ nr_frags
;
3409 struct page
*page
= virt_to_head_page(skb
->head
);
3410 unsigned int first_size
= headlen
- offset
;
3411 unsigned int first_offset
;
3413 if (nr_frags
+ 1 + skbinfo
->nr_frags
> MAX_SKB_FRAGS
)
3416 first_offset
= skb
->data
-
3417 (unsigned char *)page_address(page
) +
3420 pinfo
->nr_frags
= nr_frags
+ 1 + skbinfo
->nr_frags
;
3422 frag
->page
.p
= page
;
3423 frag
->page_offset
= first_offset
;
3424 skb_frag_size_set(frag
, first_size
);
3426 memcpy(frag
+ 1, skbinfo
->frags
, sizeof(*frag
) * skbinfo
->nr_frags
);
3427 /* We dont need to clear skbinfo->nr_frags here */
3429 delta_truesize
= skb
->truesize
- SKB_DATA_ALIGN(sizeof(struct sk_buff
));
3430 NAPI_GRO_CB(skb
)->free
= NAPI_GRO_FREE_STOLEN_HEAD
;
3435 delta_truesize
= skb
->truesize
;
3436 if (offset
> headlen
) {
3437 unsigned int eat
= offset
- headlen
;
3439 skbinfo
->frags
[0].page_offset
+= eat
;
3440 skb_frag_size_sub(&skbinfo
->frags
[0], eat
);
3441 skb
->data_len
-= eat
;
3446 __skb_pull(skb
, offset
);
3448 if (NAPI_GRO_CB(p
)->last
== p
)
3449 skb_shinfo(p
)->frag_list
= skb
;
3451 NAPI_GRO_CB(p
)->last
->next
= skb
;
3452 NAPI_GRO_CB(p
)->last
= skb
;
3453 __skb_header_release(skb
);
3457 NAPI_GRO_CB(p
)->count
++;
3459 p
->truesize
+= delta_truesize
;
3462 lp
->data_len
+= len
;
3463 lp
->truesize
+= delta_truesize
;
3466 NAPI_GRO_CB(skb
)->same_flow
= 1;
3469 EXPORT_SYMBOL_GPL(skb_gro_receive
);
3471 void __init
skb_init(void)
3473 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
3474 sizeof(struct sk_buff
),
3476 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
3478 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
3479 sizeof(struct sk_buff_fclones
),
3481 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
3486 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3487 * @skb: Socket buffer containing the buffers to be mapped
3488 * @sg: The scatter-gather list to map into
3489 * @offset: The offset into the buffer's contents to start mapping
3490 * @len: Length of buffer space to be mapped
3492 * Fill the specified scatter-gather list with mappings/pointers into a
3493 * region of the buffer space attached to a socket buffer.
3496 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
3498 int start
= skb_headlen(skb
);
3499 int i
, copy
= start
- offset
;
3500 struct sk_buff
*frag_iter
;
3506 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
3508 if ((len
-= copy
) == 0)
3513 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
3516 WARN_ON(start
> offset
+ len
);
3518 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
3519 if ((copy
= end
- offset
) > 0) {
3520 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
3524 sg_set_page(&sg
[elt
], skb_frag_page(frag
), copy
,
3525 frag
->page_offset
+offset
-start
);
3534 skb_walk_frags(skb
, frag_iter
) {
3537 WARN_ON(start
> offset
+ len
);
3539 end
= start
+ frag_iter
->len
;
3540 if ((copy
= end
- offset
) > 0) {
3543 elt
+= __skb_to_sgvec(frag_iter
, sg
+elt
, offset
- start
,
3545 if ((len
-= copy
) == 0)
3555 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
3556 * sglist without mark the sg which contain last skb data as the end.
3557 * So the caller can mannipulate sg list as will when padding new data after
3558 * the first call without calling sg_unmark_end to expend sg list.
3560 * Scenario to use skb_to_sgvec_nomark:
3562 * 2. skb_to_sgvec_nomark(payload1)
3563 * 3. skb_to_sgvec_nomark(payload2)
3565 * This is equivalent to:
3567 * 2. skb_to_sgvec(payload1)
3569 * 4. skb_to_sgvec(payload2)
3571 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
3572 * is more preferable.
3574 int skb_to_sgvec_nomark(struct sk_buff
*skb
, struct scatterlist
*sg
,
3575 int offset
, int len
)
3577 return __skb_to_sgvec(skb
, sg
, offset
, len
);
3579 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark
);
3581 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
3583 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
);
3585 sg_mark_end(&sg
[nsg
- 1]);
3589 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
3592 * skb_cow_data - Check that a socket buffer's data buffers are writable
3593 * @skb: The socket buffer to check.
3594 * @tailbits: Amount of trailing space to be added
3595 * @trailer: Returned pointer to the skb where the @tailbits space begins
3597 * Make sure that the data buffers attached to a socket buffer are
3598 * writable. If they are not, private copies are made of the data buffers
3599 * and the socket buffer is set to use these instead.
3601 * If @tailbits is given, make sure that there is space to write @tailbits
3602 * bytes of data beyond current end of socket buffer. @trailer will be
3603 * set to point to the skb in which this space begins.
3605 * The number of scatterlist elements required to completely map the
3606 * COW'd and extended socket buffer will be returned.
3608 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
3612 struct sk_buff
*skb1
, **skb_p
;
3614 /* If skb is cloned or its head is paged, reallocate
3615 * head pulling out all the pages (pages are considered not writable
3616 * at the moment even if they are anonymous).
3618 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
3619 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
3622 /* Easy case. Most of packets will go this way. */
3623 if (!skb_has_frag_list(skb
)) {
3624 /* A little of trouble, not enough of space for trailer.
3625 * This should not happen, when stack is tuned to generate
3626 * good frames. OK, on miss we reallocate and reserve even more
3627 * space, 128 bytes is fair. */
3629 if (skb_tailroom(skb
) < tailbits
&&
3630 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
3638 /* Misery. We are in troubles, going to mincer fragments... */
3641 skb_p
= &skb_shinfo(skb
)->frag_list
;
3644 while ((skb1
= *skb_p
) != NULL
) {
3647 /* The fragment is partially pulled by someone,
3648 * this can happen on input. Copy it and everything
3651 if (skb_shared(skb1
))
3654 /* If the skb is the last, worry about trailer. */
3656 if (skb1
->next
== NULL
&& tailbits
) {
3657 if (skb_shinfo(skb1
)->nr_frags
||
3658 skb_has_frag_list(skb1
) ||
3659 skb_tailroom(skb1
) < tailbits
)
3660 ntail
= tailbits
+ 128;
3666 skb_shinfo(skb1
)->nr_frags
||
3667 skb_has_frag_list(skb1
)) {
3668 struct sk_buff
*skb2
;
3670 /* Fuck, we are miserable poor guys... */
3672 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
3674 skb2
= skb_copy_expand(skb1
,
3678 if (unlikely(skb2
== NULL
))
3682 skb_set_owner_w(skb2
, skb1
->sk
);
3684 /* Looking around. Are we still alive?
3685 * OK, link new skb, drop old one */
3687 skb2
->next
= skb1
->next
;
3694 skb_p
= &skb1
->next
;
3699 EXPORT_SYMBOL_GPL(skb_cow_data
);
3701 static void sock_rmem_free(struct sk_buff
*skb
)
3703 struct sock
*sk
= skb
->sk
;
3705 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
3708 static void skb_set_err_queue(struct sk_buff
*skb
)
3710 /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
3711 * So, it is safe to (mis)use it to mark skbs on the error queue.
3713 skb
->pkt_type
= PACKET_OUTGOING
;
3714 BUILD_BUG_ON(PACKET_OUTGOING
== 0);
3718 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3720 int sock_queue_err_skb(struct sock
*sk
, struct sk_buff
*skb
)
3722 if (atomic_read(&sk
->sk_rmem_alloc
) + skb
->truesize
>=
3723 (unsigned int)sk
->sk_rcvbuf
)
3728 skb
->destructor
= sock_rmem_free
;
3729 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
3730 skb_set_err_queue(skb
);
3732 /* before exiting rcu section, make sure dst is refcounted */
3735 skb_queue_tail(&sk
->sk_error_queue
, skb
);
3736 if (!sock_flag(sk
, SOCK_DEAD
))
3737 sk
->sk_data_ready(sk
);
3740 EXPORT_SYMBOL(sock_queue_err_skb
);
3742 static bool is_icmp_err_skb(const struct sk_buff
*skb
)
3744 return skb
&& (SKB_EXT_ERR(skb
)->ee
.ee_origin
== SO_EE_ORIGIN_ICMP
||
3745 SKB_EXT_ERR(skb
)->ee
.ee_origin
== SO_EE_ORIGIN_ICMP6
);
3748 struct sk_buff
*sock_dequeue_err_skb(struct sock
*sk
)
3750 struct sk_buff_head
*q
= &sk
->sk_error_queue
;
3751 struct sk_buff
*skb
, *skb_next
= NULL
;
3752 bool icmp_next
= false;
3753 unsigned long flags
;
3755 spin_lock_irqsave(&q
->lock
, flags
);
3756 skb
= __skb_dequeue(q
);
3757 if (skb
&& (skb_next
= skb_peek(q
))) {
3758 icmp_next
= is_icmp_err_skb(skb_next
);
3760 sk
->sk_err
= SKB_EXT_ERR(skb_next
)->ee
.ee_origin
;
3762 spin_unlock_irqrestore(&q
->lock
, flags
);
3764 if (is_icmp_err_skb(skb
) && !icmp_next
)
3768 sk
->sk_error_report(sk
);
3772 EXPORT_SYMBOL(sock_dequeue_err_skb
);
3775 * skb_clone_sk - create clone of skb, and take reference to socket
3776 * @skb: the skb to clone
3778 * This function creates a clone of a buffer that holds a reference on
3779 * sk_refcnt. Buffers created via this function are meant to be
3780 * returned using sock_queue_err_skb, or free via kfree_skb.
3782 * When passing buffers allocated with this function to sock_queue_err_skb
3783 * it is necessary to wrap the call with sock_hold/sock_put in order to
3784 * prevent the socket from being released prior to being enqueued on
3785 * the sk_error_queue.
3787 struct sk_buff
*skb_clone_sk(struct sk_buff
*skb
)
3789 struct sock
*sk
= skb
->sk
;
3790 struct sk_buff
*clone
;
3792 if (!sk
|| !atomic_inc_not_zero(&sk
->sk_refcnt
))
3795 clone
= skb_clone(skb
, GFP_ATOMIC
);
3802 clone
->destructor
= sock_efree
;
3806 EXPORT_SYMBOL(skb_clone_sk
);
3808 static void __skb_complete_tx_timestamp(struct sk_buff
*skb
,
3813 struct sock_exterr_skb
*serr
;
3816 BUILD_BUG_ON(sizeof(struct sock_exterr_skb
) > sizeof(skb
->cb
));
3818 serr
= SKB_EXT_ERR(skb
);
3819 memset(serr
, 0, sizeof(*serr
));
3820 serr
->ee
.ee_errno
= ENOMSG
;
3821 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TIMESTAMPING
;
3822 serr
->ee
.ee_info
= tstype
;
3823 serr
->opt_stats
= opt_stats
;
3824 serr
->header
.h4
.iif
= skb
->dev
? skb
->dev
->ifindex
: 0;
3825 if (sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_ID
) {
3826 serr
->ee
.ee_data
= skb_shinfo(skb
)->tskey
;
3827 if (sk
->sk_protocol
== IPPROTO_TCP
&&
3828 sk
->sk_type
== SOCK_STREAM
)
3829 serr
->ee
.ee_data
-= sk
->sk_tskey
;
3832 err
= sock_queue_err_skb(sk
, skb
);
3838 static bool skb_may_tx_timestamp(struct sock
*sk
, bool tsonly
)
3842 if (likely(sysctl_tstamp_allow_data
|| tsonly
))
3845 read_lock_bh(&sk
->sk_callback_lock
);
3846 ret
= sk
->sk_socket
&& sk
->sk_socket
->file
&&
3847 file_ns_capable(sk
->sk_socket
->file
, &init_user_ns
, CAP_NET_RAW
);
3848 read_unlock_bh(&sk
->sk_callback_lock
);
3852 void skb_complete_tx_timestamp(struct sk_buff
*skb
,
3853 struct skb_shared_hwtstamps
*hwtstamps
)
3855 struct sock
*sk
= skb
->sk
;
3857 if (!skb_may_tx_timestamp(sk
, false))
3860 /* Take a reference to prevent skb_orphan() from freeing the socket,
3861 * but only if the socket refcount is not zero.
3863 if (likely(atomic_inc_not_zero(&sk
->sk_refcnt
))) {
3864 *skb_hwtstamps(skb
) = *hwtstamps
;
3865 __skb_complete_tx_timestamp(skb
, sk
, SCM_TSTAMP_SND
, false);
3869 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp
);
3871 void __skb_tstamp_tx(struct sk_buff
*orig_skb
,
3872 struct skb_shared_hwtstamps
*hwtstamps
,
3873 struct sock
*sk
, int tstype
)
3875 struct sk_buff
*skb
;
3876 bool tsonly
, opt_stats
= false;
3881 tsonly
= sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_TSONLY
;
3882 if (!skb_may_tx_timestamp(sk
, tsonly
))
3887 if ((sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_STATS
) &&
3888 sk
->sk_protocol
== IPPROTO_TCP
&&
3889 sk
->sk_type
== SOCK_STREAM
) {
3890 skb
= tcp_get_timestamping_opt_stats(sk
);
3894 skb
= alloc_skb(0, GFP_ATOMIC
);
3896 skb
= skb_clone(orig_skb
, GFP_ATOMIC
);
3902 skb_shinfo(skb
)->tx_flags
= skb_shinfo(orig_skb
)->tx_flags
;
3903 skb_shinfo(skb
)->tskey
= skb_shinfo(orig_skb
)->tskey
;
3907 *skb_hwtstamps(skb
) = *hwtstamps
;
3909 skb
->tstamp
= ktime_get_real();
3911 __skb_complete_tx_timestamp(skb
, sk
, tstype
, opt_stats
);
3913 EXPORT_SYMBOL_GPL(__skb_tstamp_tx
);
3915 void skb_tstamp_tx(struct sk_buff
*orig_skb
,
3916 struct skb_shared_hwtstamps
*hwtstamps
)
3918 return __skb_tstamp_tx(orig_skb
, hwtstamps
, orig_skb
->sk
,
3921 EXPORT_SYMBOL_GPL(skb_tstamp_tx
);
3923 void skb_complete_wifi_ack(struct sk_buff
*skb
, bool acked
)
3925 struct sock
*sk
= skb
->sk
;
3926 struct sock_exterr_skb
*serr
;
3929 skb
->wifi_acked_valid
= 1;
3930 skb
->wifi_acked
= acked
;
3932 serr
= SKB_EXT_ERR(skb
);
3933 memset(serr
, 0, sizeof(*serr
));
3934 serr
->ee
.ee_errno
= ENOMSG
;
3935 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TXSTATUS
;
3937 /* Take a reference to prevent skb_orphan() from freeing the socket,
3938 * but only if the socket refcount is not zero.
3940 if (likely(atomic_inc_not_zero(&sk
->sk_refcnt
))) {
3941 err
= sock_queue_err_skb(sk
, skb
);
3947 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack
);
3950 * skb_partial_csum_set - set up and verify partial csum values for packet
3951 * @skb: the skb to set
3952 * @start: the number of bytes after skb->data to start checksumming.
3953 * @off: the offset from start to place the checksum.
3955 * For untrusted partially-checksummed packets, we need to make sure the values
3956 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3958 * This function checks and sets those values and skb->ip_summed: if this
3959 * returns false you should drop the packet.
3961 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
3963 if (unlikely(start
> skb_headlen(skb
)) ||
3964 unlikely((int)start
+ off
> skb_headlen(skb
) - 2)) {
3965 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3966 start
, off
, skb_headlen(skb
));
3969 skb
->ip_summed
= CHECKSUM_PARTIAL
;
3970 skb
->csum_start
= skb_headroom(skb
) + start
;
3971 skb
->csum_offset
= off
;
3972 skb_set_transport_header(skb
, start
);
3975 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);
3977 static int skb_maybe_pull_tail(struct sk_buff
*skb
, unsigned int len
,
3980 if (skb_headlen(skb
) >= len
)
3983 /* If we need to pullup then pullup to the max, so we
3984 * won't need to do it again.
3989 if (__pskb_pull_tail(skb
, max
- skb_headlen(skb
)) == NULL
)
3992 if (skb_headlen(skb
) < len
)
3998 #define MAX_TCP_HDR_LEN (15 * 4)
4000 static __sum16
*skb_checksum_setup_ip(struct sk_buff
*skb
,
4001 typeof(IPPROTO_IP
) proto
,
4008 err
= skb_maybe_pull_tail(skb
, off
+ sizeof(struct tcphdr
),
4009 off
+ MAX_TCP_HDR_LEN
);
4010 if (!err
&& !skb_partial_csum_set(skb
, off
,
4011 offsetof(struct tcphdr
,
4014 return err
? ERR_PTR(err
) : &tcp_hdr(skb
)->check
;
4017 err
= skb_maybe_pull_tail(skb
, off
+ sizeof(struct udphdr
),
4018 off
+ sizeof(struct udphdr
));
4019 if (!err
&& !skb_partial_csum_set(skb
, off
,
4020 offsetof(struct udphdr
,
4023 return err
? ERR_PTR(err
) : &udp_hdr(skb
)->check
;
4026 return ERR_PTR(-EPROTO
);
4029 /* This value should be large enough to cover a tagged ethernet header plus
4030 * maximally sized IP and TCP or UDP headers.
4032 #define MAX_IP_HDR_LEN 128
4034 static int skb_checksum_setup_ipv4(struct sk_buff
*skb
, bool recalculate
)
4043 err
= skb_maybe_pull_tail(skb
,
4044 sizeof(struct iphdr
),
4049 if (ip_hdr(skb
)->frag_off
& htons(IP_OFFSET
| IP_MF
))
4052 off
= ip_hdrlen(skb
);
4059 csum
= skb_checksum_setup_ip(skb
, ip_hdr(skb
)->protocol
, off
);
4061 return PTR_ERR(csum
);
4064 *csum
= ~csum_tcpudp_magic(ip_hdr(skb
)->saddr
,
4067 ip_hdr(skb
)->protocol
, 0);
4074 /* This value should be large enough to cover a tagged ethernet header plus
4075 * an IPv6 header, all options, and a maximal TCP or UDP header.
4077 #define MAX_IPV6_HDR_LEN 256
4079 #define OPT_HDR(type, skb, off) \
4080 (type *)(skb_network_header(skb) + (off))
4082 static int skb_checksum_setup_ipv6(struct sk_buff
*skb
, bool recalculate
)
4095 off
= sizeof(struct ipv6hdr
);
4097 err
= skb_maybe_pull_tail(skb
, off
, MAX_IPV6_HDR_LEN
);
4101 nexthdr
= ipv6_hdr(skb
)->nexthdr
;
4103 len
= sizeof(struct ipv6hdr
) + ntohs(ipv6_hdr(skb
)->payload_len
);
4104 while (off
<= len
&& !done
) {
4106 case IPPROTO_DSTOPTS
:
4107 case IPPROTO_HOPOPTS
:
4108 case IPPROTO_ROUTING
: {
4109 struct ipv6_opt_hdr
*hp
;
4111 err
= skb_maybe_pull_tail(skb
,
4113 sizeof(struct ipv6_opt_hdr
),
4118 hp
= OPT_HDR(struct ipv6_opt_hdr
, skb
, off
);
4119 nexthdr
= hp
->nexthdr
;
4120 off
+= ipv6_optlen(hp
);
4124 struct ip_auth_hdr
*hp
;
4126 err
= skb_maybe_pull_tail(skb
,
4128 sizeof(struct ip_auth_hdr
),
4133 hp
= OPT_HDR(struct ip_auth_hdr
, skb
, off
);
4134 nexthdr
= hp
->nexthdr
;
4135 off
+= ipv6_authlen(hp
);
4138 case IPPROTO_FRAGMENT
: {
4139 struct frag_hdr
*hp
;
4141 err
= skb_maybe_pull_tail(skb
,
4143 sizeof(struct frag_hdr
),
4148 hp
= OPT_HDR(struct frag_hdr
, skb
, off
);
4150 if (hp
->frag_off
& htons(IP6_OFFSET
| IP6_MF
))
4153 nexthdr
= hp
->nexthdr
;
4154 off
+= sizeof(struct frag_hdr
);
4165 if (!done
|| fragment
)
4168 csum
= skb_checksum_setup_ip(skb
, nexthdr
, off
);
4170 return PTR_ERR(csum
);
4173 *csum
= ~csum_ipv6_magic(&ipv6_hdr(skb
)->saddr
,
4174 &ipv6_hdr(skb
)->daddr
,
4175 skb
->len
- off
, nexthdr
, 0);
4183 * skb_checksum_setup - set up partial checksum offset
4184 * @skb: the skb to set up
4185 * @recalculate: if true the pseudo-header checksum will be recalculated
4187 int skb_checksum_setup(struct sk_buff
*skb
, bool recalculate
)
4191 switch (skb
->protocol
) {
4192 case htons(ETH_P_IP
):
4193 err
= skb_checksum_setup_ipv4(skb
, recalculate
);
4196 case htons(ETH_P_IPV6
):
4197 err
= skb_checksum_setup_ipv6(skb
, recalculate
);
4207 EXPORT_SYMBOL(skb_checksum_setup
);
4210 * skb_checksum_maybe_trim - maybe trims the given skb
4211 * @skb: the skb to check
4212 * @transport_len: the data length beyond the network header
4214 * Checks whether the given skb has data beyond the given transport length.
4215 * If so, returns a cloned skb trimmed to this transport length.
4216 * Otherwise returns the provided skb. Returns NULL in error cases
4217 * (e.g. transport_len exceeds skb length or out-of-memory).
4219 * Caller needs to set the skb transport header and free any returned skb if it
4220 * differs from the provided skb.
4222 static struct sk_buff
*skb_checksum_maybe_trim(struct sk_buff
*skb
,
4223 unsigned int transport_len
)
4225 struct sk_buff
*skb_chk
;
4226 unsigned int len
= skb_transport_offset(skb
) + transport_len
;
4231 else if (skb
->len
== len
)
4234 skb_chk
= skb_clone(skb
, GFP_ATOMIC
);
4238 ret
= pskb_trim_rcsum(skb_chk
, len
);
4248 * skb_checksum_trimmed - validate checksum of an skb
4249 * @skb: the skb to check
4250 * @transport_len: the data length beyond the network header
4251 * @skb_chkf: checksum function to use
4253 * Applies the given checksum function skb_chkf to the provided skb.
4254 * Returns a checked and maybe trimmed skb. Returns NULL on error.
4256 * If the skb has data beyond the given transport length, then a
4257 * trimmed & cloned skb is checked and returned.
4259 * Caller needs to set the skb transport header and free any returned skb if it
4260 * differs from the provided skb.
4262 struct sk_buff
*skb_checksum_trimmed(struct sk_buff
*skb
,
4263 unsigned int transport_len
,
4264 __sum16(*skb_chkf
)(struct sk_buff
*skb
))
4266 struct sk_buff
*skb_chk
;
4267 unsigned int offset
= skb_transport_offset(skb
);
4270 skb_chk
= skb_checksum_maybe_trim(skb
, transport_len
);
4274 if (!pskb_may_pull(skb_chk
, offset
))
4277 skb_pull_rcsum(skb_chk
, offset
);
4278 ret
= skb_chkf(skb_chk
);
4279 skb_push_rcsum(skb_chk
, offset
);
4287 if (skb_chk
&& skb_chk
!= skb
)
4293 EXPORT_SYMBOL(skb_checksum_trimmed
);
4295 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
4297 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
4300 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);
4302 void kfree_skb_partial(struct sk_buff
*skb
, bool head_stolen
)
4305 skb_release_head_state(skb
);
4306 kmem_cache_free(skbuff_head_cache
, skb
);
4311 EXPORT_SYMBOL(kfree_skb_partial
);
4314 * skb_try_coalesce - try to merge skb to prior one
4316 * @from: buffer to add
4317 * @fragstolen: pointer to boolean
4318 * @delta_truesize: how much more was allocated than was requested
4320 bool skb_try_coalesce(struct sk_buff
*to
, struct sk_buff
*from
,
4321 bool *fragstolen
, int *delta_truesize
)
4323 int i
, delta
, len
= from
->len
;
4325 *fragstolen
= false;
4330 if (len
<= skb_tailroom(to
)) {
4332 BUG_ON(skb_copy_bits(from
, 0, skb_put(to
, len
), len
));
4333 *delta_truesize
= 0;
4337 if (skb_has_frag_list(to
) || skb_has_frag_list(from
))
4340 if (skb_headlen(from
) != 0) {
4342 unsigned int offset
;
4344 if (skb_shinfo(to
)->nr_frags
+
4345 skb_shinfo(from
)->nr_frags
>= MAX_SKB_FRAGS
)
4348 if (skb_head_is_locked(from
))
4351 delta
= from
->truesize
- SKB_DATA_ALIGN(sizeof(struct sk_buff
));
4353 page
= virt_to_head_page(from
->head
);
4354 offset
= from
->data
- (unsigned char *)page_address(page
);
4356 skb_fill_page_desc(to
, skb_shinfo(to
)->nr_frags
,
4357 page
, offset
, skb_headlen(from
));
4360 if (skb_shinfo(to
)->nr_frags
+
4361 skb_shinfo(from
)->nr_frags
> MAX_SKB_FRAGS
)
4364 delta
= from
->truesize
- SKB_TRUESIZE(skb_end_offset(from
));
4367 WARN_ON_ONCE(delta
< len
);
4369 memcpy(skb_shinfo(to
)->frags
+ skb_shinfo(to
)->nr_frags
,
4370 skb_shinfo(from
)->frags
,
4371 skb_shinfo(from
)->nr_frags
* sizeof(skb_frag_t
));
4372 skb_shinfo(to
)->nr_frags
+= skb_shinfo(from
)->nr_frags
;
4374 if (!skb_cloned(from
))
4375 skb_shinfo(from
)->nr_frags
= 0;
4377 /* if the skb is not cloned this does nothing
4378 * since we set nr_frags to 0.
4380 for (i
= 0; i
< skb_shinfo(from
)->nr_frags
; i
++)
4381 skb_frag_ref(from
, i
);
4383 to
->truesize
+= delta
;
4385 to
->data_len
+= len
;
4387 *delta_truesize
= delta
;
4390 EXPORT_SYMBOL(skb_try_coalesce
);
4393 * skb_scrub_packet - scrub an skb
4395 * @skb: buffer to clean
4396 * @xnet: packet is crossing netns
4398 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
4399 * into/from a tunnel. Some information have to be cleared during these
4401 * skb_scrub_packet can also be used to clean a skb before injecting it in
4402 * another namespace (@xnet == true). We have to clear all information in the
4403 * skb that could impact namespace isolation.
4405 void skb_scrub_packet(struct sk_buff
*skb
, bool xnet
)
4408 skb
->pkt_type
= PACKET_HOST
;
4414 nf_reset_trace(skb
);
4422 EXPORT_SYMBOL_GPL(skb_scrub_packet
);
4425 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
4429 * skb_gso_transport_seglen is used to determine the real size of the
4430 * individual segments, including Layer4 headers (TCP/UDP).
4432 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
4434 unsigned int skb_gso_transport_seglen(const struct sk_buff
*skb
)
4436 const struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
4437 unsigned int thlen
= 0;
4439 if (skb
->encapsulation
) {
4440 thlen
= skb_inner_transport_header(skb
) -
4441 skb_transport_header(skb
);
4443 if (likely(shinfo
->gso_type
& (SKB_GSO_TCPV4
| SKB_GSO_TCPV6
)))
4444 thlen
+= inner_tcp_hdrlen(skb
);
4445 } else if (likely(shinfo
->gso_type
& (SKB_GSO_TCPV4
| SKB_GSO_TCPV6
))) {
4446 thlen
= tcp_hdrlen(skb
);
4447 } else if (unlikely(shinfo
->gso_type
& SKB_GSO_SCTP
)) {
4448 thlen
= sizeof(struct sctphdr
);
4450 /* UFO sets gso_size to the size of the fragmentation
4451 * payload, i.e. the size of the L4 (UDP) header is already
4454 return thlen
+ shinfo
->gso_size
;
4456 EXPORT_SYMBOL_GPL(skb_gso_transport_seglen
);
4459 * skb_gso_validate_mtu - Return in case such skb fits a given MTU
4462 * @mtu: MTU to validate against
4464 * skb_gso_validate_mtu validates if a given skb will fit a wanted MTU
4467 bool skb_gso_validate_mtu(const struct sk_buff
*skb
, unsigned int mtu
)
4469 const struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
4470 const struct sk_buff
*iter
;
4473 hlen
= skb_gso_network_seglen(skb
);
4475 if (shinfo
->gso_size
!= GSO_BY_FRAGS
)
4478 /* Undo this so we can re-use header sizes */
4479 hlen
-= GSO_BY_FRAGS
;
4481 skb_walk_frags(skb
, iter
) {
4482 if (hlen
+ skb_headlen(iter
) > mtu
)
4488 EXPORT_SYMBOL_GPL(skb_gso_validate_mtu
);
4490 static struct sk_buff
*skb_reorder_vlan_header(struct sk_buff
*skb
)
4492 if (skb_cow(skb
, skb_headroom(skb
)) < 0) {
4497 memmove(skb
->data
- ETH_HLEN
, skb
->data
- skb
->mac_len
- VLAN_HLEN
,
4499 skb
->mac_header
+= VLAN_HLEN
;
4503 struct sk_buff
*skb_vlan_untag(struct sk_buff
*skb
)
4505 struct vlan_hdr
*vhdr
;
4508 if (unlikely(skb_vlan_tag_present(skb
))) {
4509 /* vlan_tci is already set-up so leave this for another time */
4513 skb
= skb_share_check(skb
, GFP_ATOMIC
);
4517 if (unlikely(!pskb_may_pull(skb
, VLAN_HLEN
)))
4520 vhdr
= (struct vlan_hdr
*)skb
->data
;
4521 vlan_tci
= ntohs(vhdr
->h_vlan_TCI
);
4522 __vlan_hwaccel_put_tag(skb
, skb
->protocol
, vlan_tci
);
4524 skb_pull_rcsum(skb
, VLAN_HLEN
);
4525 vlan_set_encap_proto(skb
, vhdr
);
4527 skb
= skb_reorder_vlan_header(skb
);
4531 skb_reset_network_header(skb
);
4532 skb_reset_transport_header(skb
);
4533 skb_reset_mac_len(skb
);
4541 EXPORT_SYMBOL(skb_vlan_untag
);
4543 int skb_ensure_writable(struct sk_buff
*skb
, int write_len
)
4545 if (!pskb_may_pull(skb
, write_len
))
4548 if (!skb_cloned(skb
) || skb_clone_writable(skb
, write_len
))
4551 return pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
4553 EXPORT_SYMBOL(skb_ensure_writable
);
4555 /* remove VLAN header from packet and update csum accordingly.
4556 * expects a non skb_vlan_tag_present skb with a vlan tag payload
4558 int __skb_vlan_pop(struct sk_buff
*skb
, u16
*vlan_tci
)
4560 struct vlan_hdr
*vhdr
;
4561 int offset
= skb
->data
- skb_mac_header(skb
);
4564 if (WARN_ONCE(offset
,
4565 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
4570 err
= skb_ensure_writable(skb
, VLAN_ETH_HLEN
);
4574 skb_postpull_rcsum(skb
, skb
->data
+ (2 * ETH_ALEN
), VLAN_HLEN
);
4576 vhdr
= (struct vlan_hdr
*)(skb
->data
+ ETH_HLEN
);
4577 *vlan_tci
= ntohs(vhdr
->h_vlan_TCI
);
4579 memmove(skb
->data
+ VLAN_HLEN
, skb
->data
, 2 * ETH_ALEN
);
4580 __skb_pull(skb
, VLAN_HLEN
);
4582 vlan_set_encap_proto(skb
, vhdr
);
4583 skb
->mac_header
+= VLAN_HLEN
;
4585 if (skb_network_offset(skb
) < ETH_HLEN
)
4586 skb_set_network_header(skb
, ETH_HLEN
);
4588 skb_reset_mac_len(skb
);
4592 EXPORT_SYMBOL(__skb_vlan_pop
);
4594 /* Pop a vlan tag either from hwaccel or from payload.
4595 * Expects skb->data at mac header.
4597 int skb_vlan_pop(struct sk_buff
*skb
)
4603 if (likely(skb_vlan_tag_present(skb
))) {
4606 if (unlikely(!eth_type_vlan(skb
->protocol
)))
4609 err
= __skb_vlan_pop(skb
, &vlan_tci
);
4613 /* move next vlan tag to hw accel tag */
4614 if (likely(!eth_type_vlan(skb
->protocol
)))
4617 vlan_proto
= skb
->protocol
;
4618 err
= __skb_vlan_pop(skb
, &vlan_tci
);
4622 __vlan_hwaccel_put_tag(skb
, vlan_proto
, vlan_tci
);
4625 EXPORT_SYMBOL(skb_vlan_pop
);
4627 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
4628 * Expects skb->data at mac header.
4630 int skb_vlan_push(struct sk_buff
*skb
, __be16 vlan_proto
, u16 vlan_tci
)
4632 if (skb_vlan_tag_present(skb
)) {
4633 int offset
= skb
->data
- skb_mac_header(skb
);
4636 if (WARN_ONCE(offset
,
4637 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
4642 err
= __vlan_insert_tag(skb
, skb
->vlan_proto
,
4643 skb_vlan_tag_get(skb
));
4647 skb
->protocol
= skb
->vlan_proto
;
4648 skb
->mac_len
+= VLAN_HLEN
;
4650 skb_postpush_rcsum(skb
, skb
->data
+ (2 * ETH_ALEN
), VLAN_HLEN
);
4652 __vlan_hwaccel_put_tag(skb
, vlan_proto
, vlan_tci
);
4655 EXPORT_SYMBOL(skb_vlan_push
);
4658 * alloc_skb_with_frags - allocate skb with page frags
4660 * @header_len: size of linear part
4661 * @data_len: needed length in frags
4662 * @max_page_order: max page order desired.
4663 * @errcode: pointer to error code if any
4664 * @gfp_mask: allocation mask
4666 * This can be used to allocate a paged skb, given a maximal order for frags.
4668 struct sk_buff
*alloc_skb_with_frags(unsigned long header_len
,
4669 unsigned long data_len
,
4674 int npages
= (data_len
+ (PAGE_SIZE
- 1)) >> PAGE_SHIFT
;
4675 unsigned long chunk
;
4676 struct sk_buff
*skb
;
4681 *errcode
= -EMSGSIZE
;
4682 /* Note this test could be relaxed, if we succeed to allocate
4683 * high order pages...
4685 if (npages
> MAX_SKB_FRAGS
)
4688 gfp_head
= gfp_mask
;
4689 if (gfp_head
& __GFP_DIRECT_RECLAIM
)
4690 gfp_head
|= __GFP_REPEAT
;
4692 *errcode
= -ENOBUFS
;
4693 skb
= alloc_skb(header_len
, gfp_head
);
4697 skb
->truesize
+= npages
<< PAGE_SHIFT
;
4699 for (i
= 0; npages
> 0; i
++) {
4700 int order
= max_page_order
;
4703 if (npages
>= 1 << order
) {
4704 page
= alloc_pages((gfp_mask
& ~__GFP_DIRECT_RECLAIM
) |
4711 /* Do not retry other high order allocations */
4717 page
= alloc_page(gfp_mask
);
4721 chunk
= min_t(unsigned long, data_len
,
4722 PAGE_SIZE
<< order
);
4723 skb_fill_page_desc(skb
, i
, page
, 0, chunk
);
4725 npages
-= 1 << order
;
4733 EXPORT_SYMBOL(alloc_skb_with_frags
);
4735 /* carve out the first off bytes from skb when off < headlen */
4736 static int pskb_carve_inside_header(struct sk_buff
*skb
, const u32 off
,
4737 const int headlen
, gfp_t gfp_mask
)
4740 int size
= skb_end_offset(skb
);
4741 int new_hlen
= headlen
- off
;
4744 size
= SKB_DATA_ALIGN(size
);
4746 if (skb_pfmemalloc(skb
))
4747 gfp_mask
|= __GFP_MEMALLOC
;
4748 data
= kmalloc_reserve(size
+
4749 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)),
4750 gfp_mask
, NUMA_NO_NODE
, NULL
);
4754 size
= SKB_WITH_OVERHEAD(ksize(data
));
4756 /* Copy real data, and all frags */
4757 skb_copy_from_linear_data_offset(skb
, off
, data
, new_hlen
);
4760 memcpy((struct skb_shared_info
*)(data
+ size
),
4762 offsetof(struct skb_shared_info
,
4763 frags
[skb_shinfo(skb
)->nr_frags
]));
4764 if (skb_cloned(skb
)) {
4765 /* drop the old head gracefully */
4766 if (skb_orphan_frags(skb
, gfp_mask
)) {
4770 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
4771 skb_frag_ref(skb
, i
);
4772 if (skb_has_frag_list(skb
))
4773 skb_clone_fraglist(skb
);
4774 skb_release_data(skb
);
4776 /* we can reuse existing recount- all we did was
4785 #ifdef NET_SKBUFF_DATA_USES_OFFSET
4788 skb
->end
= skb
->head
+ size
;
4790 skb_set_tail_pointer(skb
, skb_headlen(skb
));
4791 skb_headers_offset_update(skb
, 0);
4795 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
4800 static int pskb_carve(struct sk_buff
*skb
, const u32 off
, gfp_t gfp
);
4802 /* carve out the first eat bytes from skb's frag_list. May recurse into
4805 static int pskb_carve_frag_list(struct sk_buff
*skb
,
4806 struct skb_shared_info
*shinfo
, int eat
,
4809 struct sk_buff
*list
= shinfo
->frag_list
;
4810 struct sk_buff
*clone
= NULL
;
4811 struct sk_buff
*insp
= NULL
;
4815 pr_err("Not enough bytes to eat. Want %d\n", eat
);
4818 if (list
->len
<= eat
) {
4819 /* Eaten as whole. */
4824 /* Eaten partially. */
4825 if (skb_shared(list
)) {
4826 clone
= skb_clone(list
, gfp_mask
);
4832 /* This may be pulled without problems. */
4835 if (pskb_carve(list
, eat
, gfp_mask
) < 0) {
4843 /* Free pulled out fragments. */
4844 while ((list
= shinfo
->frag_list
) != insp
) {
4845 shinfo
->frag_list
= list
->next
;
4848 /* And insert new clone at head. */
4851 shinfo
->frag_list
= clone
;
4856 /* carve off first len bytes from skb. Split line (off) is in the
4857 * non-linear part of skb
4859 static int pskb_carve_inside_nonlinear(struct sk_buff
*skb
, const u32 off
,
4860 int pos
, gfp_t gfp_mask
)
4863 int size
= skb_end_offset(skb
);
4865 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
4866 struct skb_shared_info
*shinfo
;
4868 size
= SKB_DATA_ALIGN(size
);
4870 if (skb_pfmemalloc(skb
))
4871 gfp_mask
|= __GFP_MEMALLOC
;
4872 data
= kmalloc_reserve(size
+
4873 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)),
4874 gfp_mask
, NUMA_NO_NODE
, NULL
);
4878 size
= SKB_WITH_OVERHEAD(ksize(data
));
4880 memcpy((struct skb_shared_info
*)(data
+ size
),
4881 skb_shinfo(skb
), offsetof(struct skb_shared_info
,
4882 frags
[skb_shinfo(skb
)->nr_frags
]));
4883 if (skb_orphan_frags(skb
, gfp_mask
)) {
4887 shinfo
= (struct skb_shared_info
*)(data
+ size
);
4888 for (i
= 0; i
< nfrags
; i
++) {
4889 int fsize
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
4891 if (pos
+ fsize
> off
) {
4892 shinfo
->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
4896 * We have two variants in this case:
4897 * 1. Move all the frag to the second
4898 * part, if it is possible. F.e.
4899 * this approach is mandatory for TUX,
4900 * where splitting is expensive.
4901 * 2. Split is accurately. We make this.
4903 shinfo
->frags
[0].page_offset
+= off
- pos
;
4904 skb_frag_size_sub(&shinfo
->frags
[0], off
- pos
);
4906 skb_frag_ref(skb
, i
);
4911 shinfo
->nr_frags
= k
;
4912 if (skb_has_frag_list(skb
))
4913 skb_clone_fraglist(skb
);
4916 /* split line is in frag list */
4917 pskb_carve_frag_list(skb
, shinfo
, off
- pos
, gfp_mask
);
4919 skb_release_data(skb
);
4924 #ifdef NET_SKBUFF_DATA_USES_OFFSET
4927 skb
->end
= skb
->head
+ size
;
4929 skb_reset_tail_pointer(skb
);
4930 skb_headers_offset_update(skb
, 0);
4935 skb
->data_len
= skb
->len
;
4936 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
4940 /* remove len bytes from the beginning of the skb */
4941 static int pskb_carve(struct sk_buff
*skb
, const u32 len
, gfp_t gfp
)
4943 int headlen
= skb_headlen(skb
);
4946 return pskb_carve_inside_header(skb
, len
, headlen
, gfp
);
4948 return pskb_carve_inside_nonlinear(skb
, len
, headlen
, gfp
);
4951 /* Extract to_copy bytes starting at off from skb, and return this in
4954 struct sk_buff
*pskb_extract(struct sk_buff
*skb
, int off
,
4955 int to_copy
, gfp_t gfp
)
4957 struct sk_buff
*clone
= skb_clone(skb
, gfp
);
4962 if (pskb_carve(clone
, off
, gfp
) < 0 ||
4963 pskb_trim(clone
, to_copy
)) {
4969 EXPORT_SYMBOL(pskb_extract
);
4972 * skb_condense - try to get rid of fragments/frag_list if possible
4975 * Can be used to save memory before skb is added to a busy queue.
4976 * If packet has bytes in frags and enough tail room in skb->head,
4977 * pull all of them, so that we can free the frags right now and adjust
4980 * We do not reallocate skb->head thus can not fail.
4981 * Caller must re-evaluate skb->truesize if needed.
4983 void skb_condense(struct sk_buff
*skb
)
4985 if (skb
->data_len
) {
4986 if (skb
->data_len
> skb
->end
- skb
->tail
||
4990 /* Nice, we can free page frag(s) right now */
4991 __pskb_pull_tail(skb
, skb
->data_len
);
4993 /* At this point, skb->truesize might be over estimated,
4994 * because skb had a fragment, and fragments do not tell
4996 * When we pulled its content into skb->head, fragment
4997 * was freed, but __pskb_pull_tail() could not possibly
4998 * adjust skb->truesize, not knowing the frag truesize.
5000 skb
->truesize
= SKB_TRUESIZE(skb_end_offset(skb
));