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 refcount_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 refcount_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 refcount_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 refcount_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 (refcount_read(&fclones
->fclone_ref
) == 1)
636 default: /* SKB_FCLONE_CLONE */
637 fclones
= container_of(skb
, struct sk_buff_fclones
, skb2
);
640 if (!refcount_dec_and_test(&fclones
->fclone_ref
))
643 kmem_cache_free(skbuff_fclone_cache
, fclones
);
646 void skb_release_head_state(struct sk_buff
*skb
)
650 if (skb
->destructor
) {
652 skb
->destructor(skb
);
654 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
655 nf_conntrack_put(skb_nfct(skb
));
657 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
658 nf_bridge_put(skb
->nf_bridge
);
662 /* Free everything but the sk_buff shell. */
663 static void skb_release_all(struct sk_buff
*skb
)
665 skb_release_head_state(skb
);
666 if (likely(skb
->head
))
667 skb_release_data(skb
);
671 * __kfree_skb - private function
674 * Free an sk_buff. Release anything attached to the buffer.
675 * Clean the state. This is an internal helper function. Users should
676 * always call kfree_skb
679 void __kfree_skb(struct sk_buff
*skb
)
681 skb_release_all(skb
);
684 EXPORT_SYMBOL(__kfree_skb
);
687 * kfree_skb - free an sk_buff
688 * @skb: buffer to free
690 * Drop a reference to the buffer and free it if the usage count has
693 void kfree_skb(struct sk_buff
*skb
)
698 trace_kfree_skb(skb
, __builtin_return_address(0));
701 EXPORT_SYMBOL(kfree_skb
);
703 void kfree_skb_list(struct sk_buff
*segs
)
706 struct sk_buff
*next
= segs
->next
;
712 EXPORT_SYMBOL(kfree_skb_list
);
715 * skb_tx_error - report an sk_buff xmit error
716 * @skb: buffer that triggered an error
718 * Report xmit error if a device callback is tracking this skb.
719 * skb must be freed afterwards.
721 void skb_tx_error(struct sk_buff
*skb
)
723 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
724 struct ubuf_info
*uarg
;
726 uarg
= skb_shinfo(skb
)->destructor_arg
;
728 uarg
->callback(uarg
, false);
729 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
732 EXPORT_SYMBOL(skb_tx_error
);
735 * consume_skb - free an skbuff
736 * @skb: buffer to free
738 * Drop a ref to the buffer and free it if the usage count has hit zero
739 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
740 * is being dropped after a failure and notes that
742 void consume_skb(struct sk_buff
*skb
)
747 trace_consume_skb(skb
);
750 EXPORT_SYMBOL(consume_skb
);
753 * consume_stateless_skb - free an skbuff, assuming it is stateless
754 * @skb: buffer to free
756 * Works like consume_skb(), but this variant assumes that all the head
757 * states have been already dropped.
759 void consume_stateless_skb(struct sk_buff
*skb
)
764 trace_consume_skb(skb
);
765 if (likely(skb
->head
))
766 skb_release_data(skb
);
770 void __kfree_skb_flush(void)
772 struct napi_alloc_cache
*nc
= this_cpu_ptr(&napi_alloc_cache
);
774 /* flush skb_cache if containing objects */
776 kmem_cache_free_bulk(skbuff_head_cache
, nc
->skb_count
,
782 static inline void _kfree_skb_defer(struct sk_buff
*skb
)
784 struct napi_alloc_cache
*nc
= this_cpu_ptr(&napi_alloc_cache
);
786 /* drop skb->head and call any destructors for packet */
787 skb_release_all(skb
);
789 /* record skb to CPU local list */
790 nc
->skb_cache
[nc
->skb_count
++] = skb
;
793 /* SLUB writes into objects when freeing */
797 /* flush skb_cache if it is filled */
798 if (unlikely(nc
->skb_count
== NAPI_SKB_CACHE_SIZE
)) {
799 kmem_cache_free_bulk(skbuff_head_cache
, NAPI_SKB_CACHE_SIZE
,
804 void __kfree_skb_defer(struct sk_buff
*skb
)
806 _kfree_skb_defer(skb
);
809 void napi_consume_skb(struct sk_buff
*skb
, int budget
)
814 /* Zero budget indicate non-NAPI context called us, like netpoll */
815 if (unlikely(!budget
)) {
816 dev_consume_skb_any(skb
);
823 /* if reaching here SKB is ready to free */
824 trace_consume_skb(skb
);
826 /* if SKB is a clone, don't handle this case */
827 if (skb
->fclone
!= SKB_FCLONE_UNAVAILABLE
) {
832 _kfree_skb_defer(skb
);
834 EXPORT_SYMBOL(napi_consume_skb
);
836 /* Make sure a field is enclosed inside headers_start/headers_end section */
837 #define CHECK_SKB_FIELD(field) \
838 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
839 offsetof(struct sk_buff, headers_start)); \
840 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
841 offsetof(struct sk_buff, headers_end)); \
843 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
845 new->tstamp
= old
->tstamp
;
846 /* We do not copy old->sk */
848 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
849 skb_dst_copy(new, old
);
851 new->sp
= secpath_get(old
->sp
);
853 __nf_copy(new, old
, false);
855 /* Note : this field could be in headers_start/headers_end section
856 * It is not yet because we do not want to have a 16 bit hole
858 new->queue_mapping
= old
->queue_mapping
;
860 memcpy(&new->headers_start
, &old
->headers_start
,
861 offsetof(struct sk_buff
, headers_end
) -
862 offsetof(struct sk_buff
, headers_start
));
863 CHECK_SKB_FIELD(protocol
);
864 CHECK_SKB_FIELD(csum
);
865 CHECK_SKB_FIELD(hash
);
866 CHECK_SKB_FIELD(priority
);
867 CHECK_SKB_FIELD(skb_iif
);
868 CHECK_SKB_FIELD(vlan_proto
);
869 CHECK_SKB_FIELD(vlan_tci
);
870 CHECK_SKB_FIELD(transport_header
);
871 CHECK_SKB_FIELD(network_header
);
872 CHECK_SKB_FIELD(mac_header
);
873 CHECK_SKB_FIELD(inner_protocol
);
874 CHECK_SKB_FIELD(inner_transport_header
);
875 CHECK_SKB_FIELD(inner_network_header
);
876 CHECK_SKB_FIELD(inner_mac_header
);
877 CHECK_SKB_FIELD(mark
);
878 #ifdef CONFIG_NETWORK_SECMARK
879 CHECK_SKB_FIELD(secmark
);
881 #ifdef CONFIG_NET_RX_BUSY_POLL
882 CHECK_SKB_FIELD(napi_id
);
885 CHECK_SKB_FIELD(sender_cpu
);
887 #ifdef CONFIG_NET_SCHED
888 CHECK_SKB_FIELD(tc_index
);
894 * You should not add any new code to this function. Add it to
895 * __copy_skb_header above instead.
897 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
899 #define C(x) n->x = skb->x
901 n
->next
= n
->prev
= NULL
;
903 __copy_skb_header(n
, skb
);
908 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
911 n
->destructor
= NULL
;
918 refcount_set(&n
->users
, 1);
920 atomic_inc(&(skb_shinfo(skb
)->dataref
));
928 * skb_morph - morph one skb into another
929 * @dst: the skb to receive the contents
930 * @src: the skb to supply the contents
932 * This is identical to skb_clone except that the target skb is
933 * supplied by the user.
935 * The target skb is returned upon exit.
937 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
939 skb_release_all(dst
);
940 return __skb_clone(dst
, src
);
942 EXPORT_SYMBOL_GPL(skb_morph
);
945 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
946 * @skb: the skb to modify
947 * @gfp_mask: allocation priority
949 * This must be called on SKBTX_DEV_ZEROCOPY skb.
950 * It will copy all frags into kernel and drop the reference
951 * to userspace pages.
953 * If this function is called from an interrupt gfp_mask() must be
956 * Returns 0 on success or a negative error code on failure
957 * to allocate kernel memory to copy to.
959 int skb_copy_ubufs(struct sk_buff
*skb
, gfp_t gfp_mask
)
962 int num_frags
= skb_shinfo(skb
)->nr_frags
;
963 struct page
*page
, *head
= NULL
;
964 struct ubuf_info
*uarg
= skb_shinfo(skb
)->destructor_arg
;
966 for (i
= 0; i
< num_frags
; i
++) {
968 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
970 page
= alloc_page(gfp_mask
);
973 struct page
*next
= (struct page
*)page_private(head
);
979 vaddr
= kmap_atomic(skb_frag_page(f
));
980 memcpy(page_address(page
),
981 vaddr
+ f
->page_offset
, skb_frag_size(f
));
982 kunmap_atomic(vaddr
);
983 set_page_private(page
, (unsigned long)head
);
987 /* skb frags release userspace buffers */
988 for (i
= 0; i
< num_frags
; i
++)
989 skb_frag_unref(skb
, i
);
991 uarg
->callback(uarg
, false);
993 /* skb frags point to kernel buffers */
994 for (i
= num_frags
- 1; i
>= 0; i
--) {
995 __skb_fill_page_desc(skb
, i
, head
, 0,
996 skb_shinfo(skb
)->frags
[i
].size
);
997 head
= (struct page
*)page_private(head
);
1000 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
1003 EXPORT_SYMBOL_GPL(skb_copy_ubufs
);
1006 * skb_clone - duplicate an sk_buff
1007 * @skb: buffer to clone
1008 * @gfp_mask: allocation priority
1010 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1011 * copies share the same packet data but not structure. The new
1012 * buffer has a reference count of 1. If the allocation fails the
1013 * function returns %NULL otherwise the new buffer is returned.
1015 * If this function is called from an interrupt gfp_mask() must be
1019 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
1021 struct sk_buff_fclones
*fclones
= container_of(skb
,
1022 struct sk_buff_fclones
,
1026 if (skb_orphan_frags(skb
, gfp_mask
))
1029 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
1030 refcount_read(&fclones
->fclone_ref
) == 1) {
1032 refcount_set(&fclones
->fclone_ref
, 2);
1034 if (skb_pfmemalloc(skb
))
1035 gfp_mask
|= __GFP_MEMALLOC
;
1037 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
1041 kmemcheck_annotate_bitfield(n
, flags1
);
1042 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
1045 return __skb_clone(n
, skb
);
1047 EXPORT_SYMBOL(skb_clone
);
1049 static void skb_headers_offset_update(struct sk_buff
*skb
, int off
)
1051 /* Only adjust this if it actually is csum_start rather than csum */
1052 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1053 skb
->csum_start
+= off
;
1054 /* {transport,network,mac}_header and tail are relative to skb->head */
1055 skb
->transport_header
+= off
;
1056 skb
->network_header
+= off
;
1057 if (skb_mac_header_was_set(skb
))
1058 skb
->mac_header
+= off
;
1059 skb
->inner_transport_header
+= off
;
1060 skb
->inner_network_header
+= off
;
1061 skb
->inner_mac_header
+= off
;
1064 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
1066 __copy_skb_header(new, old
);
1068 skb_shinfo(new)->gso_size
= skb_shinfo(old
)->gso_size
;
1069 skb_shinfo(new)->gso_segs
= skb_shinfo(old
)->gso_segs
;
1070 skb_shinfo(new)->gso_type
= skb_shinfo(old
)->gso_type
;
1073 static inline int skb_alloc_rx_flag(const struct sk_buff
*skb
)
1075 if (skb_pfmemalloc(skb
))
1076 return SKB_ALLOC_RX
;
1081 * skb_copy - create private copy of an sk_buff
1082 * @skb: buffer to copy
1083 * @gfp_mask: allocation priority
1085 * Make a copy of both an &sk_buff and its data. This is used when the
1086 * caller wishes to modify the data and needs a private copy of the
1087 * data to alter. Returns %NULL on failure or the pointer to the buffer
1088 * on success. The returned buffer has a reference count of 1.
1090 * As by-product this function converts non-linear &sk_buff to linear
1091 * one, so that &sk_buff becomes completely private and caller is allowed
1092 * to modify all the data of returned buffer. This means that this
1093 * function is not recommended for use in circumstances when only
1094 * header is going to be modified. Use pskb_copy() instead.
1097 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
1099 int headerlen
= skb_headroom(skb
);
1100 unsigned int size
= skb_end_offset(skb
) + skb
->data_len
;
1101 struct sk_buff
*n
= __alloc_skb(size
, gfp_mask
,
1102 skb_alloc_rx_flag(skb
), NUMA_NO_NODE
);
1107 /* Set the data pointer */
1108 skb_reserve(n
, headerlen
);
1109 /* Set the tail pointer and length */
1110 skb_put(n
, skb
->len
);
1112 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
1115 copy_skb_header(n
, skb
);
1118 EXPORT_SYMBOL(skb_copy
);
1121 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1122 * @skb: buffer to copy
1123 * @headroom: headroom of new skb
1124 * @gfp_mask: allocation priority
1125 * @fclone: if true allocate the copy of the skb from the fclone
1126 * cache instead of the head cache; it is recommended to set this
1127 * to true for the cases where the copy will likely be cloned
1129 * Make a copy of both an &sk_buff and part of its data, located
1130 * in header. Fragmented data remain shared. This is used when
1131 * the caller wishes to modify only header of &sk_buff and needs
1132 * private copy of the header to alter. Returns %NULL on failure
1133 * or the pointer to the buffer on success.
1134 * The returned buffer has a reference count of 1.
1137 struct sk_buff
*__pskb_copy_fclone(struct sk_buff
*skb
, int headroom
,
1138 gfp_t gfp_mask
, bool fclone
)
1140 unsigned int size
= skb_headlen(skb
) + headroom
;
1141 int flags
= skb_alloc_rx_flag(skb
) | (fclone
? SKB_ALLOC_FCLONE
: 0);
1142 struct sk_buff
*n
= __alloc_skb(size
, gfp_mask
, flags
, NUMA_NO_NODE
);
1147 /* Set the data pointer */
1148 skb_reserve(n
, headroom
);
1149 /* Set the tail pointer and length */
1150 skb_put(n
, skb_headlen(skb
));
1151 /* Copy the bytes */
1152 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
1154 n
->truesize
+= skb
->data_len
;
1155 n
->data_len
= skb
->data_len
;
1158 if (skb_shinfo(skb
)->nr_frags
) {
1161 if (skb_orphan_frags(skb
, gfp_mask
)) {
1166 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1167 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
1168 skb_frag_ref(skb
, i
);
1170 skb_shinfo(n
)->nr_frags
= i
;
1173 if (skb_has_frag_list(skb
)) {
1174 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
1175 skb_clone_fraglist(n
);
1178 copy_skb_header(n
, skb
);
1182 EXPORT_SYMBOL(__pskb_copy_fclone
);
1185 * pskb_expand_head - reallocate header of &sk_buff
1186 * @skb: buffer to reallocate
1187 * @nhead: room to add at head
1188 * @ntail: room to add at tail
1189 * @gfp_mask: allocation priority
1191 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1192 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1193 * reference count of 1. Returns zero in the case of success or error,
1194 * if expansion failed. In the last case, &sk_buff is not changed.
1196 * All the pointers pointing into skb header may change and must be
1197 * reloaded after call to this function.
1200 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
1203 int i
, osize
= skb_end_offset(skb
);
1204 int size
= osize
+ nhead
+ ntail
;
1210 if (skb_shared(skb
))
1213 size
= SKB_DATA_ALIGN(size
);
1215 if (skb_pfmemalloc(skb
))
1216 gfp_mask
|= __GFP_MEMALLOC
;
1217 data
= kmalloc_reserve(size
+ SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)),
1218 gfp_mask
, NUMA_NO_NODE
, NULL
);
1221 size
= SKB_WITH_OVERHEAD(ksize(data
));
1223 /* Copy only real data... and, alas, header. This should be
1224 * optimized for the cases when header is void.
1226 memcpy(data
+ nhead
, skb
->head
, skb_tail_pointer(skb
) - skb
->head
);
1228 memcpy((struct skb_shared_info
*)(data
+ size
),
1230 offsetof(struct skb_shared_info
, frags
[skb_shinfo(skb
)->nr_frags
]));
1233 * if shinfo is shared we must drop the old head gracefully, but if it
1234 * is not we can just drop the old head and let the existing refcount
1235 * be since all we did is relocate the values
1237 if (skb_cloned(skb
)) {
1238 /* copy this zero copy skb frags */
1239 if (skb_orphan_frags(skb
, gfp_mask
))
1241 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1242 skb_frag_ref(skb
, i
);
1244 if (skb_has_frag_list(skb
))
1245 skb_clone_fraglist(skb
);
1247 skb_release_data(skb
);
1251 off
= (data
+ nhead
) - skb
->head
;
1256 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1260 skb
->end
= skb
->head
+ size
;
1263 skb_headers_offset_update(skb
, nhead
);
1267 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
1269 /* It is not generally safe to change skb->truesize.
1270 * For the moment, we really care of rx path, or
1271 * when skb is orphaned (not attached to a socket).
1273 if (!skb
->sk
|| skb
->destructor
== sock_edemux
)
1274 skb
->truesize
+= size
- osize
;
1283 EXPORT_SYMBOL(pskb_expand_head
);
1285 /* Make private copy of skb with writable head and some headroom */
1287 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
1289 struct sk_buff
*skb2
;
1290 int delta
= headroom
- skb_headroom(skb
);
1293 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
1295 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1296 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
1304 EXPORT_SYMBOL(skb_realloc_headroom
);
1307 * skb_copy_expand - copy and expand sk_buff
1308 * @skb: buffer to copy
1309 * @newheadroom: new free bytes at head
1310 * @newtailroom: new free bytes at tail
1311 * @gfp_mask: allocation priority
1313 * Make a copy of both an &sk_buff and its data and while doing so
1314 * allocate additional space.
1316 * This is used when the caller wishes to modify the data and needs a
1317 * private copy of the data to alter as well as more space for new fields.
1318 * Returns %NULL on failure or the pointer to the buffer
1319 * on success. The returned buffer has a reference count of 1.
1321 * You must pass %GFP_ATOMIC as the allocation priority if this function
1322 * is called from an interrupt.
1324 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
1325 int newheadroom
, int newtailroom
,
1329 * Allocate the copy buffer
1331 struct sk_buff
*n
= __alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
1332 gfp_mask
, skb_alloc_rx_flag(skb
),
1334 int oldheadroom
= skb_headroom(skb
);
1335 int head_copy_len
, head_copy_off
;
1340 skb_reserve(n
, newheadroom
);
1342 /* Set the tail pointer and length */
1343 skb_put(n
, skb
->len
);
1345 head_copy_len
= oldheadroom
;
1347 if (newheadroom
<= head_copy_len
)
1348 head_copy_len
= newheadroom
;
1350 head_copy_off
= newheadroom
- head_copy_len
;
1352 /* Copy the linear header and data. */
1353 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
1354 skb
->len
+ head_copy_len
))
1357 copy_skb_header(n
, skb
);
1359 skb_headers_offset_update(n
, newheadroom
- oldheadroom
);
1363 EXPORT_SYMBOL(skb_copy_expand
);
1366 * __skb_pad - zero pad the tail of an skb
1367 * @skb: buffer to pad
1368 * @pad: space to pad
1369 * @free_on_error: free buffer on error
1371 * Ensure that a buffer is followed by a padding area that is zero
1372 * filled. Used by network drivers which may DMA or transfer data
1373 * beyond the buffer end onto the wire.
1375 * May return error in out of memory cases. The skb is freed on error
1376 * if @free_on_error is true.
1379 int __skb_pad(struct sk_buff
*skb
, int pad
, bool free_on_error
)
1384 /* If the skbuff is non linear tailroom is always zero.. */
1385 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
1386 memset(skb
->data
+skb
->len
, 0, pad
);
1390 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
1391 if (likely(skb_cloned(skb
) || ntail
> 0)) {
1392 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
1397 /* FIXME: The use of this function with non-linear skb's really needs
1400 err
= skb_linearize(skb
);
1404 memset(skb
->data
+ skb
->len
, 0, pad
);
1412 EXPORT_SYMBOL(__skb_pad
);
1415 * pskb_put - add data to the tail of a potentially fragmented buffer
1416 * @skb: start of the buffer to use
1417 * @tail: tail fragment of the buffer to use
1418 * @len: amount of data to add
1420 * This function extends the used data area of the potentially
1421 * fragmented buffer. @tail must be the last fragment of @skb -- or
1422 * @skb itself. If this would exceed the total buffer size the kernel
1423 * will panic. A pointer to the first byte of the extra data is
1427 void *pskb_put(struct sk_buff
*skb
, struct sk_buff
*tail
, int len
)
1430 skb
->data_len
+= len
;
1433 return skb_put(tail
, len
);
1435 EXPORT_SYMBOL_GPL(pskb_put
);
1438 * skb_put - add data to a buffer
1439 * @skb: buffer to use
1440 * @len: amount of data to add
1442 * This function extends the used data area of the buffer. If this would
1443 * exceed the total buffer size the kernel will panic. A pointer to the
1444 * first byte of the extra data is returned.
1446 void *skb_put(struct sk_buff
*skb
, unsigned int len
)
1448 void *tmp
= skb_tail_pointer(skb
);
1449 SKB_LINEAR_ASSERT(skb
);
1452 if (unlikely(skb
->tail
> skb
->end
))
1453 skb_over_panic(skb
, len
, __builtin_return_address(0));
1456 EXPORT_SYMBOL(skb_put
);
1459 * skb_push - add data to the start of a buffer
1460 * @skb: buffer to use
1461 * @len: amount of data to add
1463 * This function extends the used data area of the buffer at the buffer
1464 * start. If this would exceed the total buffer headroom the kernel will
1465 * panic. A pointer to the first byte of the extra data is returned.
1467 void *skb_push(struct sk_buff
*skb
, unsigned int len
)
1471 if (unlikely(skb
->data
<skb
->head
))
1472 skb_under_panic(skb
, len
, __builtin_return_address(0));
1475 EXPORT_SYMBOL(skb_push
);
1478 * skb_pull - remove data from the start of a buffer
1479 * @skb: buffer to use
1480 * @len: amount of data to remove
1482 * This function removes data from the start of a buffer, returning
1483 * the memory to the headroom. A pointer to the next data in the buffer
1484 * is returned. Once the data has been pulled future pushes will overwrite
1487 void *skb_pull(struct sk_buff
*skb
, unsigned int len
)
1489 return skb_pull_inline(skb
, len
);
1491 EXPORT_SYMBOL(skb_pull
);
1494 * skb_trim - remove end from a buffer
1495 * @skb: buffer to alter
1498 * Cut the length of a buffer down by removing data from the tail. If
1499 * the buffer is already under the length specified it is not modified.
1500 * The skb must be linear.
1502 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1505 __skb_trim(skb
, len
);
1507 EXPORT_SYMBOL(skb_trim
);
1509 /* Trims skb to length len. It can change skb pointers.
1512 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1514 struct sk_buff
**fragp
;
1515 struct sk_buff
*frag
;
1516 int offset
= skb_headlen(skb
);
1517 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1521 if (skb_cloned(skb
) &&
1522 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1529 for (; i
< nfrags
; i
++) {
1530 int end
= offset
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1537 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
++], len
- offset
);
1540 skb_shinfo(skb
)->nr_frags
= i
;
1542 for (; i
< nfrags
; i
++)
1543 skb_frag_unref(skb
, i
);
1545 if (skb_has_frag_list(skb
))
1546 skb_drop_fraglist(skb
);
1550 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1551 fragp
= &frag
->next
) {
1552 int end
= offset
+ frag
->len
;
1554 if (skb_shared(frag
)) {
1555 struct sk_buff
*nfrag
;
1557 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1558 if (unlikely(!nfrag
))
1561 nfrag
->next
= frag
->next
;
1573 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1577 skb_drop_list(&frag
->next
);
1582 if (len
> skb_headlen(skb
)) {
1583 skb
->data_len
-= skb
->len
- len
;
1588 skb_set_tail_pointer(skb
, len
);
1591 if (!skb
->sk
|| skb
->destructor
== sock_edemux
)
1595 EXPORT_SYMBOL(___pskb_trim
);
1598 * __pskb_pull_tail - advance tail of skb header
1599 * @skb: buffer to reallocate
1600 * @delta: number of bytes to advance tail
1602 * The function makes a sense only on a fragmented &sk_buff,
1603 * it expands header moving its tail forward and copying necessary
1604 * data from fragmented part.
1606 * &sk_buff MUST have reference count of 1.
1608 * Returns %NULL (and &sk_buff does not change) if pull failed
1609 * or value of new tail of skb in the case of success.
1611 * All the pointers pointing into skb header may change and must be
1612 * reloaded after call to this function.
1615 /* Moves tail of skb head forward, copying data from fragmented part,
1616 * when it is necessary.
1617 * 1. It may fail due to malloc failure.
1618 * 2. It may change skb pointers.
1620 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1622 void *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1624 /* If skb has not enough free space at tail, get new one
1625 * plus 128 bytes for future expansions. If we have enough
1626 * room at tail, reallocate without expansion only if skb is cloned.
1628 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1630 if (eat
> 0 || skb_cloned(skb
)) {
1631 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1636 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
1639 /* Optimization: no fragments, no reasons to preestimate
1640 * size of pulled pages. Superb.
1642 if (!skb_has_frag_list(skb
))
1645 /* Estimate size of pulled pages. */
1647 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1648 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1655 /* If we need update frag list, we are in troubles.
1656 * Certainly, it possible to add an offset to skb data,
1657 * but taking into account that pulling is expected to
1658 * be very rare operation, it is worth to fight against
1659 * further bloating skb head and crucify ourselves here instead.
1660 * Pure masohism, indeed. 8)8)
1663 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1664 struct sk_buff
*clone
= NULL
;
1665 struct sk_buff
*insp
= NULL
;
1670 if (list
->len
<= eat
) {
1671 /* Eaten as whole. */
1676 /* Eaten partially. */
1678 if (skb_shared(list
)) {
1679 /* Sucks! We need to fork list. :-( */
1680 clone
= skb_clone(list
, GFP_ATOMIC
);
1686 /* This may be pulled without
1690 if (!pskb_pull(list
, eat
)) {
1698 /* Free pulled out fragments. */
1699 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1700 skb_shinfo(skb
)->frag_list
= list
->next
;
1703 /* And insert new clone at head. */
1706 skb_shinfo(skb
)->frag_list
= clone
;
1709 /* Success! Now we may commit changes to skb data. */
1714 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1715 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1718 skb_frag_unref(skb
, i
);
1721 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1723 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1724 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[k
], eat
);
1730 skb_shinfo(skb
)->nr_frags
= k
;
1733 skb
->data_len
-= delta
;
1735 return skb_tail_pointer(skb
);
1737 EXPORT_SYMBOL(__pskb_pull_tail
);
1740 * skb_copy_bits - copy bits from skb to kernel buffer
1742 * @offset: offset in source
1743 * @to: destination buffer
1744 * @len: number of bytes to copy
1746 * Copy the specified number of bytes from the source skb to the
1747 * destination buffer.
1750 * If its prototype is ever changed,
1751 * check arch/{*}/net/{*}.S files,
1752 * since it is called from BPF assembly code.
1754 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1756 int start
= skb_headlen(skb
);
1757 struct sk_buff
*frag_iter
;
1760 if (offset
> (int)skb
->len
- len
)
1764 if ((copy
= start
- offset
) > 0) {
1767 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
1768 if ((len
-= copy
) == 0)
1774 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1776 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
1778 WARN_ON(start
> offset
+ len
);
1780 end
= start
+ skb_frag_size(f
);
1781 if ((copy
= end
- offset
) > 0) {
1787 vaddr
= kmap_atomic(skb_frag_page(f
));
1789 vaddr
+ f
->page_offset
+ offset
- start
,
1791 kunmap_atomic(vaddr
);
1793 if ((len
-= copy
) == 0)
1801 skb_walk_frags(skb
, frag_iter
) {
1804 WARN_ON(start
> offset
+ len
);
1806 end
= start
+ frag_iter
->len
;
1807 if ((copy
= end
- offset
) > 0) {
1810 if (skb_copy_bits(frag_iter
, offset
- start
, to
, copy
))
1812 if ((len
-= copy
) == 0)
1826 EXPORT_SYMBOL(skb_copy_bits
);
1829 * Callback from splice_to_pipe(), if we need to release some pages
1830 * at the end of the spd in case we error'ed out in filling the pipe.
1832 static void sock_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
1834 put_page(spd
->pages
[i
]);
1837 static struct page
*linear_to_page(struct page
*page
, unsigned int *len
,
1838 unsigned int *offset
,
1841 struct page_frag
*pfrag
= sk_page_frag(sk
);
1843 if (!sk_page_frag_refill(sk
, pfrag
))
1846 *len
= min_t(unsigned int, *len
, pfrag
->size
- pfrag
->offset
);
1848 memcpy(page_address(pfrag
->page
) + pfrag
->offset
,
1849 page_address(page
) + *offset
, *len
);
1850 *offset
= pfrag
->offset
;
1851 pfrag
->offset
+= *len
;
1856 static bool spd_can_coalesce(const struct splice_pipe_desc
*spd
,
1858 unsigned int offset
)
1860 return spd
->nr_pages
&&
1861 spd
->pages
[spd
->nr_pages
- 1] == page
&&
1862 (spd
->partial
[spd
->nr_pages
- 1].offset
+
1863 spd
->partial
[spd
->nr_pages
- 1].len
== offset
);
1867 * Fill page/offset/length into spd, if it can hold more pages.
1869 static bool spd_fill_page(struct splice_pipe_desc
*spd
,
1870 struct pipe_inode_info
*pipe
, struct page
*page
,
1871 unsigned int *len
, unsigned int offset
,
1875 if (unlikely(spd
->nr_pages
== MAX_SKB_FRAGS
))
1879 page
= linear_to_page(page
, len
, &offset
, sk
);
1883 if (spd_can_coalesce(spd
, page
, offset
)) {
1884 spd
->partial
[spd
->nr_pages
- 1].len
+= *len
;
1888 spd
->pages
[spd
->nr_pages
] = page
;
1889 spd
->partial
[spd
->nr_pages
].len
= *len
;
1890 spd
->partial
[spd
->nr_pages
].offset
= offset
;
1896 static bool __splice_segment(struct page
*page
, unsigned int poff
,
1897 unsigned int plen
, unsigned int *off
,
1899 struct splice_pipe_desc
*spd
, bool linear
,
1901 struct pipe_inode_info
*pipe
)
1906 /* skip this segment if already processed */
1912 /* ignore any bits we already processed */
1918 unsigned int flen
= min(*len
, plen
);
1920 if (spd_fill_page(spd
, pipe
, page
, &flen
, poff
,
1926 } while (*len
&& plen
);
1932 * Map linear and fragment data from the skb to spd. It reports true if the
1933 * pipe is full or if we already spliced the requested length.
1935 static bool __skb_splice_bits(struct sk_buff
*skb
, struct pipe_inode_info
*pipe
,
1936 unsigned int *offset
, unsigned int *len
,
1937 struct splice_pipe_desc
*spd
, struct sock
*sk
)
1940 struct sk_buff
*iter
;
1942 /* map the linear part :
1943 * If skb->head_frag is set, this 'linear' part is backed by a
1944 * fragment, and if the head is not shared with any clones then
1945 * we can avoid a copy since we own the head portion of this page.
1947 if (__splice_segment(virt_to_page(skb
->data
),
1948 (unsigned long) skb
->data
& (PAGE_SIZE
- 1),
1951 skb_head_is_locked(skb
),
1956 * then map the fragments
1958 for (seg
= 0; seg
< skb_shinfo(skb
)->nr_frags
; seg
++) {
1959 const skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[seg
];
1961 if (__splice_segment(skb_frag_page(f
),
1962 f
->page_offset
, skb_frag_size(f
),
1963 offset
, len
, spd
, false, sk
, pipe
))
1967 skb_walk_frags(skb
, iter
) {
1968 if (*offset
>= iter
->len
) {
1969 *offset
-= iter
->len
;
1972 /* __skb_splice_bits() only fails if the output has no room
1973 * left, so no point in going over the frag_list for the error
1976 if (__skb_splice_bits(iter
, pipe
, offset
, len
, spd
, sk
))
1984 * Map data from the skb to a pipe. Should handle both the linear part,
1985 * the fragments, and the frag list.
1987 int skb_splice_bits(struct sk_buff
*skb
, struct sock
*sk
, unsigned int offset
,
1988 struct pipe_inode_info
*pipe
, unsigned int tlen
,
1991 struct partial_page partial
[MAX_SKB_FRAGS
];
1992 struct page
*pages
[MAX_SKB_FRAGS
];
1993 struct splice_pipe_desc spd
= {
1996 .nr_pages_max
= MAX_SKB_FRAGS
,
1997 .ops
= &nosteal_pipe_buf_ops
,
1998 .spd_release
= sock_spd_release
,
2002 __skb_splice_bits(skb
, pipe
, &offset
, &tlen
, &spd
, sk
);
2005 ret
= splice_to_pipe(pipe
, &spd
);
2009 EXPORT_SYMBOL_GPL(skb_splice_bits
);
2012 * skb_store_bits - store bits from kernel buffer to skb
2013 * @skb: destination buffer
2014 * @offset: offset in destination
2015 * @from: source buffer
2016 * @len: number of bytes to copy
2018 * Copy the specified number of bytes from the source buffer to the
2019 * destination skb. This function handles all the messy bits of
2020 * traversing fragment lists and such.
2023 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
2025 int start
= skb_headlen(skb
);
2026 struct sk_buff
*frag_iter
;
2029 if (offset
> (int)skb
->len
- len
)
2032 if ((copy
= start
- offset
) > 0) {
2035 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
2036 if ((len
-= copy
) == 0)
2042 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2043 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2046 WARN_ON(start
> offset
+ len
);
2048 end
= start
+ skb_frag_size(frag
);
2049 if ((copy
= end
- offset
) > 0) {
2055 vaddr
= kmap_atomic(skb_frag_page(frag
));
2056 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
2058 kunmap_atomic(vaddr
);
2060 if ((len
-= copy
) == 0)
2068 skb_walk_frags(skb
, frag_iter
) {
2071 WARN_ON(start
> offset
+ len
);
2073 end
= start
+ frag_iter
->len
;
2074 if ((copy
= end
- offset
) > 0) {
2077 if (skb_store_bits(frag_iter
, offset
- start
,
2080 if ((len
-= copy
) == 0)
2093 EXPORT_SYMBOL(skb_store_bits
);
2095 /* Checksum skb data. */
2096 __wsum
__skb_checksum(const struct sk_buff
*skb
, int offset
, int len
,
2097 __wsum csum
, const struct skb_checksum_ops
*ops
)
2099 int start
= skb_headlen(skb
);
2100 int i
, copy
= start
- offset
;
2101 struct sk_buff
*frag_iter
;
2104 /* Checksum header. */
2108 csum
= ops
->update(skb
->data
+ offset
, copy
, csum
);
2109 if ((len
-= copy
) == 0)
2115 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2117 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2119 WARN_ON(start
> offset
+ len
);
2121 end
= start
+ skb_frag_size(frag
);
2122 if ((copy
= end
- offset
) > 0) {
2128 vaddr
= kmap_atomic(skb_frag_page(frag
));
2129 csum2
= ops
->update(vaddr
+ frag
->page_offset
+
2130 offset
- start
, copy
, 0);
2131 kunmap_atomic(vaddr
);
2132 csum
= ops
->combine(csum
, csum2
, pos
, copy
);
2141 skb_walk_frags(skb
, frag_iter
) {
2144 WARN_ON(start
> offset
+ len
);
2146 end
= start
+ frag_iter
->len
;
2147 if ((copy
= end
- offset
) > 0) {
2151 csum2
= __skb_checksum(frag_iter
, offset
- start
,
2153 csum
= ops
->combine(csum
, csum2
, pos
, copy
);
2154 if ((len
-= copy
) == 0)
2165 EXPORT_SYMBOL(__skb_checksum
);
2167 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
2168 int len
, __wsum csum
)
2170 const struct skb_checksum_ops ops
= {
2171 .update
= csum_partial_ext
,
2172 .combine
= csum_block_add_ext
,
2175 return __skb_checksum(skb
, offset
, len
, csum
, &ops
);
2177 EXPORT_SYMBOL(skb_checksum
);
2179 /* Both of above in one bottle. */
2181 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
2182 u8
*to
, int len
, __wsum csum
)
2184 int start
= skb_headlen(skb
);
2185 int i
, copy
= start
- offset
;
2186 struct sk_buff
*frag_iter
;
2193 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
2195 if ((len
-= copy
) == 0)
2202 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2205 WARN_ON(start
> offset
+ len
);
2207 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
2208 if ((copy
= end
- offset
) > 0) {
2211 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2215 vaddr
= kmap_atomic(skb_frag_page(frag
));
2216 csum2
= csum_partial_copy_nocheck(vaddr
+
2220 kunmap_atomic(vaddr
);
2221 csum
= csum_block_add(csum
, csum2
, pos
);
2231 skb_walk_frags(skb
, frag_iter
) {
2235 WARN_ON(start
> offset
+ len
);
2237 end
= start
+ frag_iter
->len
;
2238 if ((copy
= end
- offset
) > 0) {
2241 csum2
= skb_copy_and_csum_bits(frag_iter
,
2244 csum
= csum_block_add(csum
, csum2
, pos
);
2245 if ((len
-= copy
) == 0)
2256 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
2258 static __wsum
warn_crc32c_csum_update(const void *buff
, int len
, __wsum sum
)
2260 net_warn_ratelimited(
2261 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2266 static __wsum
warn_crc32c_csum_combine(__wsum csum
, __wsum csum2
,
2267 int offset
, int len
)
2269 net_warn_ratelimited(
2270 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2275 static const struct skb_checksum_ops default_crc32c_ops
= {
2276 .update
= warn_crc32c_csum_update
,
2277 .combine
= warn_crc32c_csum_combine
,
2280 const struct skb_checksum_ops
*crc32c_csum_stub __read_mostly
=
2281 &default_crc32c_ops
;
2282 EXPORT_SYMBOL(crc32c_csum_stub
);
2285 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2286 * @from: source buffer
2288 * Calculates the amount of linear headroom needed in the 'to' skb passed
2289 * into skb_zerocopy().
2292 skb_zerocopy_headlen(const struct sk_buff
*from
)
2294 unsigned int hlen
= 0;
2296 if (!from
->head_frag
||
2297 skb_headlen(from
) < L1_CACHE_BYTES
||
2298 skb_shinfo(from
)->nr_frags
>= MAX_SKB_FRAGS
)
2299 hlen
= skb_headlen(from
);
2301 if (skb_has_frag_list(from
))
2306 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen
);
2309 * skb_zerocopy - Zero copy skb to skb
2310 * @to: destination buffer
2311 * @from: source buffer
2312 * @len: number of bytes to copy from source buffer
2313 * @hlen: size of linear headroom in destination buffer
2315 * Copies up to `len` bytes from `from` to `to` by creating references
2316 * to the frags in the source buffer.
2318 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2319 * headroom in the `to` buffer.
2322 * 0: everything is OK
2323 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2324 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2327 skb_zerocopy(struct sk_buff
*to
, struct sk_buff
*from
, int len
, int hlen
)
2330 int plen
= 0; /* length of skb->head fragment */
2333 unsigned int offset
;
2335 BUG_ON(!from
->head_frag
&& !hlen
);
2337 /* dont bother with small payloads */
2338 if (len
<= skb_tailroom(to
))
2339 return skb_copy_bits(from
, 0, skb_put(to
, len
), len
);
2342 ret
= skb_copy_bits(from
, 0, skb_put(to
, hlen
), hlen
);
2347 plen
= min_t(int, skb_headlen(from
), len
);
2349 page
= virt_to_head_page(from
->head
);
2350 offset
= from
->data
- (unsigned char *)page_address(page
);
2351 __skb_fill_page_desc(to
, 0, page
, offset
, plen
);
2358 to
->truesize
+= len
+ plen
;
2359 to
->len
+= len
+ plen
;
2360 to
->data_len
+= len
+ plen
;
2362 if (unlikely(skb_orphan_frags(from
, GFP_ATOMIC
))) {
2367 for (i
= 0; i
< skb_shinfo(from
)->nr_frags
; i
++) {
2370 skb_shinfo(to
)->frags
[j
] = skb_shinfo(from
)->frags
[i
];
2371 skb_shinfo(to
)->frags
[j
].size
= min_t(int, skb_shinfo(to
)->frags
[j
].size
, len
);
2372 len
-= skb_shinfo(to
)->frags
[j
].size
;
2373 skb_frag_ref(to
, j
);
2376 skb_shinfo(to
)->nr_frags
= j
;
2380 EXPORT_SYMBOL_GPL(skb_zerocopy
);
2382 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
2387 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
2388 csstart
= skb_checksum_start_offset(skb
);
2390 csstart
= skb_headlen(skb
);
2392 BUG_ON(csstart
> skb_headlen(skb
));
2394 skb_copy_from_linear_data(skb
, to
, csstart
);
2397 if (csstart
!= skb
->len
)
2398 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
2399 skb
->len
- csstart
, 0);
2401 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2402 long csstuff
= csstart
+ skb
->csum_offset
;
2404 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
2407 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
2410 * skb_dequeue - remove from the head of the queue
2411 * @list: list to dequeue from
2413 * Remove the head of the list. The list lock is taken so the function
2414 * may be used safely with other locking list functions. The head item is
2415 * returned or %NULL if the list is empty.
2418 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
2420 unsigned long flags
;
2421 struct sk_buff
*result
;
2423 spin_lock_irqsave(&list
->lock
, flags
);
2424 result
= __skb_dequeue(list
);
2425 spin_unlock_irqrestore(&list
->lock
, flags
);
2428 EXPORT_SYMBOL(skb_dequeue
);
2431 * skb_dequeue_tail - remove from the tail of the queue
2432 * @list: list to dequeue from
2434 * Remove the tail of the list. The list lock is taken so the function
2435 * may be used safely with other locking list functions. The tail item is
2436 * returned or %NULL if the list is empty.
2438 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
2440 unsigned long flags
;
2441 struct sk_buff
*result
;
2443 spin_lock_irqsave(&list
->lock
, flags
);
2444 result
= __skb_dequeue_tail(list
);
2445 spin_unlock_irqrestore(&list
->lock
, flags
);
2448 EXPORT_SYMBOL(skb_dequeue_tail
);
2451 * skb_queue_purge - empty a list
2452 * @list: list to empty
2454 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2455 * the list and one reference dropped. This function takes the list
2456 * lock and is atomic with respect to other list locking functions.
2458 void skb_queue_purge(struct sk_buff_head
*list
)
2460 struct sk_buff
*skb
;
2461 while ((skb
= skb_dequeue(list
)) != NULL
)
2464 EXPORT_SYMBOL(skb_queue_purge
);
2467 * skb_rbtree_purge - empty a skb rbtree
2468 * @root: root of the rbtree to empty
2470 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
2471 * the list and one reference dropped. This function does not take
2472 * any lock. Synchronization should be handled by the caller (e.g., TCP
2473 * out-of-order queue is protected by the socket lock).
2475 void skb_rbtree_purge(struct rb_root
*root
)
2477 struct sk_buff
*skb
, *next
;
2479 rbtree_postorder_for_each_entry_safe(skb
, next
, root
, rbnode
)
2486 * skb_queue_head - queue a buffer at the list head
2487 * @list: list to use
2488 * @newsk: buffer to queue
2490 * Queue a buffer at the start of the list. This function takes the
2491 * list lock and can be used safely with other locking &sk_buff functions
2494 * A buffer cannot be placed on two lists at the same time.
2496 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2498 unsigned long flags
;
2500 spin_lock_irqsave(&list
->lock
, flags
);
2501 __skb_queue_head(list
, newsk
);
2502 spin_unlock_irqrestore(&list
->lock
, flags
);
2504 EXPORT_SYMBOL(skb_queue_head
);
2507 * skb_queue_tail - queue a buffer at the list tail
2508 * @list: list to use
2509 * @newsk: buffer to queue
2511 * Queue a buffer at the tail of the list. This function takes the
2512 * list lock and can be used safely with other locking &sk_buff functions
2515 * A buffer cannot be placed on two lists at the same time.
2517 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2519 unsigned long flags
;
2521 spin_lock_irqsave(&list
->lock
, flags
);
2522 __skb_queue_tail(list
, newsk
);
2523 spin_unlock_irqrestore(&list
->lock
, flags
);
2525 EXPORT_SYMBOL(skb_queue_tail
);
2528 * skb_unlink - remove a buffer from a list
2529 * @skb: buffer to remove
2530 * @list: list to use
2532 * Remove a packet from a list. The list locks are taken and this
2533 * function is atomic with respect to other list locked calls
2535 * You must know what list the SKB is on.
2537 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
2539 unsigned long flags
;
2541 spin_lock_irqsave(&list
->lock
, flags
);
2542 __skb_unlink(skb
, list
);
2543 spin_unlock_irqrestore(&list
->lock
, flags
);
2545 EXPORT_SYMBOL(skb_unlink
);
2548 * skb_append - append a buffer
2549 * @old: buffer to insert after
2550 * @newsk: buffer to insert
2551 * @list: list to use
2553 * Place a packet after a given packet in a list. The list locks are taken
2554 * and this function is atomic with respect to other list locked calls.
2555 * A buffer cannot be placed on two lists at the same time.
2557 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2559 unsigned long flags
;
2561 spin_lock_irqsave(&list
->lock
, flags
);
2562 __skb_queue_after(list
, old
, newsk
);
2563 spin_unlock_irqrestore(&list
->lock
, flags
);
2565 EXPORT_SYMBOL(skb_append
);
2568 * skb_insert - insert a buffer
2569 * @old: buffer to insert before
2570 * @newsk: buffer to insert
2571 * @list: list to use
2573 * Place a packet before a given packet in a list. The list locks are
2574 * taken and this function is atomic with respect to other list locked
2577 * A buffer cannot be placed on two lists at the same time.
2579 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2581 unsigned long flags
;
2583 spin_lock_irqsave(&list
->lock
, flags
);
2584 __skb_insert(newsk
, old
->prev
, old
, list
);
2585 spin_unlock_irqrestore(&list
->lock
, flags
);
2587 EXPORT_SYMBOL(skb_insert
);
2589 static inline void skb_split_inside_header(struct sk_buff
*skb
,
2590 struct sk_buff
* skb1
,
2591 const u32 len
, const int pos
)
2595 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
2597 /* And move data appendix as is. */
2598 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
2599 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
2601 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
2602 skb_shinfo(skb
)->nr_frags
= 0;
2603 skb1
->data_len
= skb
->data_len
;
2604 skb1
->len
+= skb1
->data_len
;
2607 skb_set_tail_pointer(skb
, len
);
2610 static inline void skb_split_no_header(struct sk_buff
*skb
,
2611 struct sk_buff
* skb1
,
2612 const u32 len
, int pos
)
2615 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
2617 skb_shinfo(skb
)->nr_frags
= 0;
2618 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
2620 skb
->data_len
= len
- pos
;
2622 for (i
= 0; i
< nfrags
; i
++) {
2623 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
2625 if (pos
+ size
> len
) {
2626 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
2630 * We have two variants in this case:
2631 * 1. Move all the frag to the second
2632 * part, if it is possible. F.e.
2633 * this approach is mandatory for TUX,
2634 * where splitting is expensive.
2635 * 2. Split is accurately. We make this.
2637 skb_frag_ref(skb
, i
);
2638 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
2639 skb_frag_size_sub(&skb_shinfo(skb1
)->frags
[0], len
- pos
);
2640 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
], len
- pos
);
2641 skb_shinfo(skb
)->nr_frags
++;
2645 skb_shinfo(skb
)->nr_frags
++;
2648 skb_shinfo(skb1
)->nr_frags
= k
;
2652 * skb_split - Split fragmented skb to two parts at length len.
2653 * @skb: the buffer to split
2654 * @skb1: the buffer to receive the second part
2655 * @len: new length for skb
2657 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
2659 int pos
= skb_headlen(skb
);
2661 skb_shinfo(skb1
)->tx_flags
|= skb_shinfo(skb
)->tx_flags
&
2663 if (len
< pos
) /* Split line is inside header. */
2664 skb_split_inside_header(skb
, skb1
, len
, pos
);
2665 else /* Second chunk has no header, nothing to copy. */
2666 skb_split_no_header(skb
, skb1
, len
, pos
);
2668 EXPORT_SYMBOL(skb_split
);
2670 /* Shifting from/to a cloned skb is a no-go.
2672 * Caller cannot keep skb_shinfo related pointers past calling here!
2674 static int skb_prepare_for_shift(struct sk_buff
*skb
)
2676 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2680 * skb_shift - Shifts paged data partially from skb to another
2681 * @tgt: buffer into which tail data gets added
2682 * @skb: buffer from which the paged data comes from
2683 * @shiftlen: shift up to this many bytes
2685 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2686 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2687 * It's up to caller to free skb if everything was shifted.
2689 * If @tgt runs out of frags, the whole operation is aborted.
2691 * Skb cannot include anything else but paged data while tgt is allowed
2692 * to have non-paged data as well.
2694 * TODO: full sized shift could be optimized but that would need
2695 * specialized skb free'er to handle frags without up-to-date nr_frags.
2697 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
2699 int from
, to
, merge
, todo
;
2700 struct skb_frag_struct
*fragfrom
, *fragto
;
2702 BUG_ON(shiftlen
> skb
->len
);
2704 if (skb_headlen(skb
))
2709 to
= skb_shinfo(tgt
)->nr_frags
;
2710 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2712 /* Actual merge is delayed until the point when we know we can
2713 * commit all, so that we don't have to undo partial changes
2716 !skb_can_coalesce(tgt
, to
, skb_frag_page(fragfrom
),
2717 fragfrom
->page_offset
)) {
2722 todo
-= skb_frag_size(fragfrom
);
2724 if (skb_prepare_for_shift(skb
) ||
2725 skb_prepare_for_shift(tgt
))
2728 /* All previous frag pointers might be stale! */
2729 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2730 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2732 skb_frag_size_add(fragto
, shiftlen
);
2733 skb_frag_size_sub(fragfrom
, shiftlen
);
2734 fragfrom
->page_offset
+= shiftlen
;
2742 /* Skip full, not-fitting skb to avoid expensive operations */
2743 if ((shiftlen
== skb
->len
) &&
2744 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
2747 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
2750 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
2751 if (to
== MAX_SKB_FRAGS
)
2754 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2755 fragto
= &skb_shinfo(tgt
)->frags
[to
];
2757 if (todo
>= skb_frag_size(fragfrom
)) {
2758 *fragto
= *fragfrom
;
2759 todo
-= skb_frag_size(fragfrom
);
2764 __skb_frag_ref(fragfrom
);
2765 fragto
->page
= fragfrom
->page
;
2766 fragto
->page_offset
= fragfrom
->page_offset
;
2767 skb_frag_size_set(fragto
, todo
);
2769 fragfrom
->page_offset
+= todo
;
2770 skb_frag_size_sub(fragfrom
, todo
);
2778 /* Ready to "commit" this state change to tgt */
2779 skb_shinfo(tgt
)->nr_frags
= to
;
2782 fragfrom
= &skb_shinfo(skb
)->frags
[0];
2783 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2785 skb_frag_size_add(fragto
, skb_frag_size(fragfrom
));
2786 __skb_frag_unref(fragfrom
);
2789 /* Reposition in the original skb */
2791 while (from
< skb_shinfo(skb
)->nr_frags
)
2792 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
2793 skb_shinfo(skb
)->nr_frags
= to
;
2795 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
2798 /* Most likely the tgt won't ever need its checksum anymore, skb on
2799 * the other hand might need it if it needs to be resent
2801 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
2802 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2804 /* Yak, is it really working this way? Some helper please? */
2805 skb
->len
-= shiftlen
;
2806 skb
->data_len
-= shiftlen
;
2807 skb
->truesize
-= shiftlen
;
2808 tgt
->len
+= shiftlen
;
2809 tgt
->data_len
+= shiftlen
;
2810 tgt
->truesize
+= shiftlen
;
2816 * skb_prepare_seq_read - Prepare a sequential read of skb data
2817 * @skb: the buffer to read
2818 * @from: lower offset of data to be read
2819 * @to: upper offset of data to be read
2820 * @st: state variable
2822 * Initializes the specified state variable. Must be called before
2823 * invoking skb_seq_read() for the first time.
2825 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
2826 unsigned int to
, struct skb_seq_state
*st
)
2828 st
->lower_offset
= from
;
2829 st
->upper_offset
= to
;
2830 st
->root_skb
= st
->cur_skb
= skb
;
2831 st
->frag_idx
= st
->stepped_offset
= 0;
2832 st
->frag_data
= NULL
;
2834 EXPORT_SYMBOL(skb_prepare_seq_read
);
2837 * skb_seq_read - Sequentially read skb data
2838 * @consumed: number of bytes consumed by the caller so far
2839 * @data: destination pointer for data to be returned
2840 * @st: state variable
2842 * Reads a block of skb data at @consumed relative to the
2843 * lower offset specified to skb_prepare_seq_read(). Assigns
2844 * the head of the data block to @data and returns the length
2845 * of the block or 0 if the end of the skb data or the upper
2846 * offset has been reached.
2848 * The caller is not required to consume all of the data
2849 * returned, i.e. @consumed is typically set to the number
2850 * of bytes already consumed and the next call to
2851 * skb_seq_read() will return the remaining part of the block.
2853 * Note 1: The size of each block of data returned can be arbitrary,
2854 * this limitation is the cost for zerocopy sequential
2855 * reads of potentially non linear data.
2857 * Note 2: Fragment lists within fragments are not implemented
2858 * at the moment, state->root_skb could be replaced with
2859 * a stack for this purpose.
2861 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
2862 struct skb_seq_state
*st
)
2864 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
2867 if (unlikely(abs_offset
>= st
->upper_offset
)) {
2868 if (st
->frag_data
) {
2869 kunmap_atomic(st
->frag_data
);
2870 st
->frag_data
= NULL
;
2876 block_limit
= skb_headlen(st
->cur_skb
) + st
->stepped_offset
;
2878 if (abs_offset
< block_limit
&& !st
->frag_data
) {
2879 *data
= st
->cur_skb
->data
+ (abs_offset
- st
->stepped_offset
);
2880 return block_limit
- abs_offset
;
2883 if (st
->frag_idx
== 0 && !st
->frag_data
)
2884 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
2886 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
2887 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
2888 block_limit
= skb_frag_size(frag
) + st
->stepped_offset
;
2890 if (abs_offset
< block_limit
) {
2892 st
->frag_data
= kmap_atomic(skb_frag_page(frag
));
2894 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
2895 (abs_offset
- st
->stepped_offset
);
2897 return block_limit
- abs_offset
;
2900 if (st
->frag_data
) {
2901 kunmap_atomic(st
->frag_data
);
2902 st
->frag_data
= NULL
;
2906 st
->stepped_offset
+= skb_frag_size(frag
);
2909 if (st
->frag_data
) {
2910 kunmap_atomic(st
->frag_data
);
2911 st
->frag_data
= NULL
;
2914 if (st
->root_skb
== st
->cur_skb
&& skb_has_frag_list(st
->root_skb
)) {
2915 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
2918 } else if (st
->cur_skb
->next
) {
2919 st
->cur_skb
= st
->cur_skb
->next
;
2926 EXPORT_SYMBOL(skb_seq_read
);
2929 * skb_abort_seq_read - Abort a sequential read of skb data
2930 * @st: state variable
2932 * Must be called if skb_seq_read() was not called until it
2935 void skb_abort_seq_read(struct skb_seq_state
*st
)
2938 kunmap_atomic(st
->frag_data
);
2940 EXPORT_SYMBOL(skb_abort_seq_read
);
2942 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2944 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
2945 struct ts_config
*conf
,
2946 struct ts_state
*state
)
2948 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
2951 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
2953 skb_abort_seq_read(TS_SKB_CB(state
));
2957 * skb_find_text - Find a text pattern in skb data
2958 * @skb: the buffer to look in
2959 * @from: search offset
2961 * @config: textsearch configuration
2963 * Finds a pattern in the skb data according to the specified
2964 * textsearch configuration. Use textsearch_next() to retrieve
2965 * subsequent occurrences of the pattern. Returns the offset
2966 * to the first occurrence or UINT_MAX if no match was found.
2968 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
2969 unsigned int to
, struct ts_config
*config
)
2971 struct ts_state state
;
2974 config
->get_next_block
= skb_ts_get_next_block
;
2975 config
->finish
= skb_ts_finish
;
2977 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(&state
));
2979 ret
= textsearch_find(config
, &state
);
2980 return (ret
<= to
- from
? ret
: UINT_MAX
);
2982 EXPORT_SYMBOL(skb_find_text
);
2985 * skb_append_datato_frags - append the user data to a skb
2986 * @sk: sock structure
2987 * @skb: skb structure to be appended with user data.
2988 * @getfrag: call back function to be used for getting the user data
2989 * @from: pointer to user message iov
2990 * @length: length of the iov message
2992 * Description: This procedure append the user data in the fragment part
2993 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2995 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
2996 int (*getfrag
)(void *from
, char *to
, int offset
,
2997 int len
, int odd
, struct sk_buff
*skb
),
2998 void *from
, int length
)
3000 int frg_cnt
= skb_shinfo(skb
)->nr_frags
;
3004 struct page_frag
*pfrag
= ¤t
->task_frag
;
3007 /* Return error if we don't have space for new frag */
3008 if (frg_cnt
>= MAX_SKB_FRAGS
)
3011 if (!sk_page_frag_refill(sk
, pfrag
))
3014 /* copy the user data to page */
3015 copy
= min_t(int, length
, pfrag
->size
- pfrag
->offset
);
3017 ret
= getfrag(from
, page_address(pfrag
->page
) + pfrag
->offset
,
3018 offset
, copy
, 0, skb
);
3022 /* copy was successful so update the size parameters */
3023 skb_fill_page_desc(skb
, frg_cnt
, pfrag
->page
, pfrag
->offset
,
3026 pfrag
->offset
+= copy
;
3027 get_page(pfrag
->page
);
3029 skb
->truesize
+= copy
;
3030 refcount_add(copy
, &sk
->sk_wmem_alloc
);
3032 skb
->data_len
+= copy
;
3036 } while (length
> 0);
3040 EXPORT_SYMBOL(skb_append_datato_frags
);
3042 int skb_append_pagefrags(struct sk_buff
*skb
, struct page
*page
,
3043 int offset
, size_t size
)
3045 int i
= skb_shinfo(skb
)->nr_frags
;
3047 if (skb_can_coalesce(skb
, i
, page
, offset
)) {
3048 skb_frag_size_add(&skb_shinfo(skb
)->frags
[i
- 1], size
);
3049 } else if (i
< MAX_SKB_FRAGS
) {
3051 skb_fill_page_desc(skb
, i
, page
, offset
, size
);
3058 EXPORT_SYMBOL_GPL(skb_append_pagefrags
);
3061 * skb_pull_rcsum - pull skb and update receive checksum
3062 * @skb: buffer to update
3063 * @len: length of data pulled
3065 * This function performs an skb_pull on the packet and updates
3066 * the CHECKSUM_COMPLETE checksum. It should be used on
3067 * receive path processing instead of skb_pull unless you know
3068 * that the checksum difference is zero (e.g., a valid IP header)
3069 * or you are setting ip_summed to CHECKSUM_NONE.
3071 void *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
3073 unsigned char *data
= skb
->data
;
3075 BUG_ON(len
> skb
->len
);
3076 __skb_pull(skb
, len
);
3077 skb_postpull_rcsum(skb
, data
, len
);
3080 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
3083 * skb_segment - Perform protocol segmentation on skb.
3084 * @head_skb: buffer to segment
3085 * @features: features for the output path (see dev->features)
3087 * This function performs segmentation on the given skb. It returns
3088 * a pointer to the first in a list of new skbs for the segments.
3089 * In case of error it returns ERR_PTR(err).
3091 struct sk_buff
*skb_segment(struct sk_buff
*head_skb
,
3092 netdev_features_t features
)
3094 struct sk_buff
*segs
= NULL
;
3095 struct sk_buff
*tail
= NULL
;
3096 struct sk_buff
*list_skb
= skb_shinfo(head_skb
)->frag_list
;
3097 skb_frag_t
*frag
= skb_shinfo(head_skb
)->frags
;
3098 unsigned int mss
= skb_shinfo(head_skb
)->gso_size
;
3099 unsigned int doffset
= head_skb
->data
- skb_mac_header(head_skb
);
3100 struct sk_buff
*frag_skb
= head_skb
;
3101 unsigned int offset
= doffset
;
3102 unsigned int tnl_hlen
= skb_tnl_header_len(head_skb
);
3103 unsigned int partial_segs
= 0;
3104 unsigned int headroom
;
3105 unsigned int len
= head_skb
->len
;
3108 int nfrags
= skb_shinfo(head_skb
)->nr_frags
;
3114 __skb_push(head_skb
, doffset
);
3115 proto
= skb_network_protocol(head_skb
, &dummy
);
3116 if (unlikely(!proto
))
3117 return ERR_PTR(-EINVAL
);
3119 sg
= !!(features
& NETIF_F_SG
);
3120 csum
= !!can_checksum_protocol(features
, proto
);
3122 if (sg
&& csum
&& (mss
!= GSO_BY_FRAGS
)) {
3123 if (!(features
& NETIF_F_GSO_PARTIAL
)) {
3124 struct sk_buff
*iter
;
3125 unsigned int frag_len
;
3128 !net_gso_ok(features
, skb_shinfo(head_skb
)->gso_type
))
3131 /* If we get here then all the required
3132 * GSO features except frag_list are supported.
3133 * Try to split the SKB to multiple GSO SKBs
3134 * with no frag_list.
3135 * Currently we can do that only when the buffers don't
3136 * have a linear part and all the buffers except
3137 * the last are of the same length.
3139 frag_len
= list_skb
->len
;
3140 skb_walk_frags(head_skb
, iter
) {
3141 if (frag_len
!= iter
->len
&& iter
->next
)
3143 if (skb_headlen(iter
) && !iter
->head_frag
)
3149 if (len
!= frag_len
)
3153 /* GSO partial only requires that we trim off any excess that
3154 * doesn't fit into an MSS sized block, so take care of that
3157 partial_segs
= len
/ mss
;
3158 if (partial_segs
> 1)
3159 mss
*= partial_segs
;
3165 headroom
= skb_headroom(head_skb
);
3166 pos
= skb_headlen(head_skb
);
3169 struct sk_buff
*nskb
;
3170 skb_frag_t
*nskb_frag
;
3174 if (unlikely(mss
== GSO_BY_FRAGS
)) {
3175 len
= list_skb
->len
;
3177 len
= head_skb
->len
- offset
;
3182 hsize
= skb_headlen(head_skb
) - offset
;
3185 if (hsize
> len
|| !sg
)
3188 if (!hsize
&& i
>= nfrags
&& skb_headlen(list_skb
) &&
3189 (skb_headlen(list_skb
) == len
|| sg
)) {
3190 BUG_ON(skb_headlen(list_skb
) > len
);
3193 nfrags
= skb_shinfo(list_skb
)->nr_frags
;
3194 frag
= skb_shinfo(list_skb
)->frags
;
3195 frag_skb
= list_skb
;
3196 pos
+= skb_headlen(list_skb
);
3198 while (pos
< offset
+ len
) {
3199 BUG_ON(i
>= nfrags
);
3201 size
= skb_frag_size(frag
);
3202 if (pos
+ size
> offset
+ len
)
3210 nskb
= skb_clone(list_skb
, GFP_ATOMIC
);
3211 list_skb
= list_skb
->next
;
3213 if (unlikely(!nskb
))
3216 if (unlikely(pskb_trim(nskb
, len
))) {
3221 hsize
= skb_end_offset(nskb
);
3222 if (skb_cow_head(nskb
, doffset
+ headroom
)) {
3227 nskb
->truesize
+= skb_end_offset(nskb
) - hsize
;
3228 skb_release_head_state(nskb
);
3229 __skb_push(nskb
, doffset
);
3231 nskb
= __alloc_skb(hsize
+ doffset
+ headroom
,
3232 GFP_ATOMIC
, skb_alloc_rx_flag(head_skb
),
3235 if (unlikely(!nskb
))
3238 skb_reserve(nskb
, headroom
);
3239 __skb_put(nskb
, doffset
);
3248 __copy_skb_header(nskb
, head_skb
);
3250 skb_headers_offset_update(nskb
, skb_headroom(nskb
) - headroom
);
3251 skb_reset_mac_len(nskb
);
3253 skb_copy_from_linear_data_offset(head_skb
, -tnl_hlen
,
3254 nskb
->data
- tnl_hlen
,
3255 doffset
+ tnl_hlen
);
3257 if (nskb
->len
== len
+ doffset
)
3258 goto perform_csum_check
;
3261 if (!nskb
->remcsum_offload
)
3262 nskb
->ip_summed
= CHECKSUM_NONE
;
3263 SKB_GSO_CB(nskb
)->csum
=
3264 skb_copy_and_csum_bits(head_skb
, offset
,
3267 SKB_GSO_CB(nskb
)->csum_start
=
3268 skb_headroom(nskb
) + doffset
;
3272 nskb_frag
= skb_shinfo(nskb
)->frags
;
3274 skb_copy_from_linear_data_offset(head_skb
, offset
,
3275 skb_put(nskb
, hsize
), hsize
);
3277 skb_shinfo(nskb
)->tx_flags
|= skb_shinfo(head_skb
)->tx_flags
&
3280 while (pos
< offset
+ len
) {
3282 BUG_ON(skb_headlen(list_skb
));
3285 nfrags
= skb_shinfo(list_skb
)->nr_frags
;
3286 frag
= skb_shinfo(list_skb
)->frags
;
3287 frag_skb
= list_skb
;
3291 list_skb
= list_skb
->next
;
3294 if (unlikely(skb_shinfo(nskb
)->nr_frags
>=
3296 net_warn_ratelimited(
3297 "skb_segment: too many frags: %u %u\n",
3302 if (unlikely(skb_orphan_frags(frag_skb
, GFP_ATOMIC
)))
3306 __skb_frag_ref(nskb_frag
);
3307 size
= skb_frag_size(nskb_frag
);
3310 nskb_frag
->page_offset
+= offset
- pos
;
3311 skb_frag_size_sub(nskb_frag
, offset
- pos
);
3314 skb_shinfo(nskb
)->nr_frags
++;
3316 if (pos
+ size
<= offset
+ len
) {
3321 skb_frag_size_sub(nskb_frag
, pos
+ size
- (offset
+ len
));
3329 nskb
->data_len
= len
- hsize
;
3330 nskb
->len
+= nskb
->data_len
;
3331 nskb
->truesize
+= nskb
->data_len
;
3335 if (skb_has_shared_frag(nskb
)) {
3336 err
= __skb_linearize(nskb
);
3340 if (!nskb
->remcsum_offload
)
3341 nskb
->ip_summed
= CHECKSUM_NONE
;
3342 SKB_GSO_CB(nskb
)->csum
=
3343 skb_checksum(nskb
, doffset
,
3344 nskb
->len
- doffset
, 0);
3345 SKB_GSO_CB(nskb
)->csum_start
=
3346 skb_headroom(nskb
) + doffset
;
3348 } while ((offset
+= len
) < head_skb
->len
);
3350 /* Some callers want to get the end of the list.
3351 * Put it in segs->prev to avoid walking the list.
3352 * (see validate_xmit_skb_list() for example)
3357 struct sk_buff
*iter
;
3358 int type
= skb_shinfo(head_skb
)->gso_type
;
3359 unsigned short gso_size
= skb_shinfo(head_skb
)->gso_size
;
3361 /* Update type to add partial and then remove dodgy if set */
3362 type
|= (features
& NETIF_F_GSO_PARTIAL
) / NETIF_F_GSO_PARTIAL
* SKB_GSO_PARTIAL
;
3363 type
&= ~SKB_GSO_DODGY
;
3365 /* Update GSO info and prepare to start updating headers on
3366 * our way back down the stack of protocols.
3368 for (iter
= segs
; iter
; iter
= iter
->next
) {
3369 skb_shinfo(iter
)->gso_size
= gso_size
;
3370 skb_shinfo(iter
)->gso_segs
= partial_segs
;
3371 skb_shinfo(iter
)->gso_type
= type
;
3372 SKB_GSO_CB(iter
)->data_offset
= skb_headroom(iter
) + doffset
;
3375 if (tail
->len
- doffset
<= gso_size
)
3376 skb_shinfo(tail
)->gso_size
= 0;
3377 else if (tail
!= segs
)
3378 skb_shinfo(tail
)->gso_segs
= DIV_ROUND_UP(tail
->len
- doffset
, gso_size
);
3381 /* Following permits correct backpressure, for protocols
3382 * using skb_set_owner_w().
3383 * Idea is to tranfert ownership from head_skb to last segment.
3385 if (head_skb
->destructor
== sock_wfree
) {
3386 swap(tail
->truesize
, head_skb
->truesize
);
3387 swap(tail
->destructor
, head_skb
->destructor
);
3388 swap(tail
->sk
, head_skb
->sk
);
3393 kfree_skb_list(segs
);
3394 return ERR_PTR(err
);
3396 EXPORT_SYMBOL_GPL(skb_segment
);
3398 int skb_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
3400 struct skb_shared_info
*pinfo
, *skbinfo
= skb_shinfo(skb
);
3401 unsigned int offset
= skb_gro_offset(skb
);
3402 unsigned int headlen
= skb_headlen(skb
);
3403 unsigned int len
= skb_gro_len(skb
);
3404 struct sk_buff
*lp
, *p
= *head
;
3405 unsigned int delta_truesize
;
3407 if (unlikely(p
->len
+ len
>= 65536))
3410 lp
= NAPI_GRO_CB(p
)->last
;
3411 pinfo
= skb_shinfo(lp
);
3413 if (headlen
<= offset
) {
3416 int i
= skbinfo
->nr_frags
;
3417 int nr_frags
= pinfo
->nr_frags
+ i
;
3419 if (nr_frags
> MAX_SKB_FRAGS
)
3423 pinfo
->nr_frags
= nr_frags
;
3424 skbinfo
->nr_frags
= 0;
3426 frag
= pinfo
->frags
+ nr_frags
;
3427 frag2
= skbinfo
->frags
+ i
;
3432 frag
->page_offset
+= offset
;
3433 skb_frag_size_sub(frag
, offset
);
3435 /* all fragments truesize : remove (head size + sk_buff) */
3436 delta_truesize
= skb
->truesize
-
3437 SKB_TRUESIZE(skb_end_offset(skb
));
3439 skb
->truesize
-= skb
->data_len
;
3440 skb
->len
-= skb
->data_len
;
3443 NAPI_GRO_CB(skb
)->free
= NAPI_GRO_FREE
;
3445 } else if (skb
->head_frag
) {
3446 int nr_frags
= pinfo
->nr_frags
;
3447 skb_frag_t
*frag
= pinfo
->frags
+ nr_frags
;
3448 struct page
*page
= virt_to_head_page(skb
->head
);
3449 unsigned int first_size
= headlen
- offset
;
3450 unsigned int first_offset
;
3452 if (nr_frags
+ 1 + skbinfo
->nr_frags
> MAX_SKB_FRAGS
)
3455 first_offset
= skb
->data
-
3456 (unsigned char *)page_address(page
) +
3459 pinfo
->nr_frags
= nr_frags
+ 1 + skbinfo
->nr_frags
;
3461 frag
->page
.p
= page
;
3462 frag
->page_offset
= first_offset
;
3463 skb_frag_size_set(frag
, first_size
);
3465 memcpy(frag
+ 1, skbinfo
->frags
, sizeof(*frag
) * skbinfo
->nr_frags
);
3466 /* We dont need to clear skbinfo->nr_frags here */
3468 delta_truesize
= skb
->truesize
- SKB_DATA_ALIGN(sizeof(struct sk_buff
));
3469 NAPI_GRO_CB(skb
)->free
= NAPI_GRO_FREE_STOLEN_HEAD
;
3474 delta_truesize
= skb
->truesize
;
3475 if (offset
> headlen
) {
3476 unsigned int eat
= offset
- headlen
;
3478 skbinfo
->frags
[0].page_offset
+= eat
;
3479 skb_frag_size_sub(&skbinfo
->frags
[0], eat
);
3480 skb
->data_len
-= eat
;
3485 __skb_pull(skb
, offset
);
3487 if (NAPI_GRO_CB(p
)->last
== p
)
3488 skb_shinfo(p
)->frag_list
= skb
;
3490 NAPI_GRO_CB(p
)->last
->next
= skb
;
3491 NAPI_GRO_CB(p
)->last
= skb
;
3492 __skb_header_release(skb
);
3496 NAPI_GRO_CB(p
)->count
++;
3498 p
->truesize
+= delta_truesize
;
3501 lp
->data_len
+= len
;
3502 lp
->truesize
+= delta_truesize
;
3505 NAPI_GRO_CB(skb
)->same_flow
= 1;
3508 EXPORT_SYMBOL_GPL(skb_gro_receive
);
3510 void __init
skb_init(void)
3512 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
3513 sizeof(struct sk_buff
),
3515 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
3517 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
3518 sizeof(struct sk_buff_fclones
),
3520 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
3525 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
,
3526 unsigned int recursion_level
)
3528 int start
= skb_headlen(skb
);
3529 int i
, copy
= start
- offset
;
3530 struct sk_buff
*frag_iter
;
3533 if (unlikely(recursion_level
>= 24))
3539 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
3541 if ((len
-= copy
) == 0)
3546 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
3549 WARN_ON(start
> offset
+ len
);
3551 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
3552 if ((copy
= end
- offset
) > 0) {
3553 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
3554 if (unlikely(elt
&& sg_is_last(&sg
[elt
- 1])))
3559 sg_set_page(&sg
[elt
], skb_frag_page(frag
), copy
,
3560 frag
->page_offset
+offset
-start
);
3569 skb_walk_frags(skb
, frag_iter
) {
3572 WARN_ON(start
> offset
+ len
);
3574 end
= start
+ frag_iter
->len
;
3575 if ((copy
= end
- offset
) > 0) {
3576 if (unlikely(elt
&& sg_is_last(&sg
[elt
- 1])))
3581 ret
= __skb_to_sgvec(frag_iter
, sg
+elt
, offset
- start
,
3582 copy
, recursion_level
+ 1);
3583 if (unlikely(ret
< 0))
3586 if ((len
-= copy
) == 0)
3597 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3598 * @skb: Socket buffer containing the buffers to be mapped
3599 * @sg: The scatter-gather list to map into
3600 * @offset: The offset into the buffer's contents to start mapping
3601 * @len: Length of buffer space to be mapped
3603 * Fill the specified scatter-gather list with mappings/pointers into a
3604 * region of the buffer space attached to a socket buffer. Returns either
3605 * the number of scatterlist items used, or -EMSGSIZE if the contents
3608 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
3610 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
, 0);
3615 sg_mark_end(&sg
[nsg
- 1]);
3619 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
3621 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
3622 * sglist without mark the sg which contain last skb data as the end.
3623 * So the caller can mannipulate sg list as will when padding new data after
3624 * the first call without calling sg_unmark_end to expend sg list.
3626 * Scenario to use skb_to_sgvec_nomark:
3628 * 2. skb_to_sgvec_nomark(payload1)
3629 * 3. skb_to_sgvec_nomark(payload2)
3631 * This is equivalent to:
3633 * 2. skb_to_sgvec(payload1)
3635 * 4. skb_to_sgvec(payload2)
3637 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
3638 * is more preferable.
3640 int skb_to_sgvec_nomark(struct sk_buff
*skb
, struct scatterlist
*sg
,
3641 int offset
, int len
)
3643 return __skb_to_sgvec(skb
, sg
, offset
, len
, 0);
3645 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark
);
3650 * skb_cow_data - Check that a socket buffer's data buffers are writable
3651 * @skb: The socket buffer to check.
3652 * @tailbits: Amount of trailing space to be added
3653 * @trailer: Returned pointer to the skb where the @tailbits space begins
3655 * Make sure that the data buffers attached to a socket buffer are
3656 * writable. If they are not, private copies are made of the data buffers
3657 * and the socket buffer is set to use these instead.
3659 * If @tailbits is given, make sure that there is space to write @tailbits
3660 * bytes of data beyond current end of socket buffer. @trailer will be
3661 * set to point to the skb in which this space begins.
3663 * The number of scatterlist elements required to completely map the
3664 * COW'd and extended socket buffer will be returned.
3666 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
3670 struct sk_buff
*skb1
, **skb_p
;
3672 /* If skb is cloned or its head is paged, reallocate
3673 * head pulling out all the pages (pages are considered not writable
3674 * at the moment even if they are anonymous).
3676 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
3677 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
3680 /* Easy case. Most of packets will go this way. */
3681 if (!skb_has_frag_list(skb
)) {
3682 /* A little of trouble, not enough of space for trailer.
3683 * This should not happen, when stack is tuned to generate
3684 * good frames. OK, on miss we reallocate and reserve even more
3685 * space, 128 bytes is fair. */
3687 if (skb_tailroom(skb
) < tailbits
&&
3688 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
3696 /* Misery. We are in troubles, going to mincer fragments... */
3699 skb_p
= &skb_shinfo(skb
)->frag_list
;
3702 while ((skb1
= *skb_p
) != NULL
) {
3705 /* The fragment is partially pulled by someone,
3706 * this can happen on input. Copy it and everything
3709 if (skb_shared(skb1
))
3712 /* If the skb is the last, worry about trailer. */
3714 if (skb1
->next
== NULL
&& tailbits
) {
3715 if (skb_shinfo(skb1
)->nr_frags
||
3716 skb_has_frag_list(skb1
) ||
3717 skb_tailroom(skb1
) < tailbits
)
3718 ntail
= tailbits
+ 128;
3724 skb_shinfo(skb1
)->nr_frags
||
3725 skb_has_frag_list(skb1
)) {
3726 struct sk_buff
*skb2
;
3728 /* Fuck, we are miserable poor guys... */
3730 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
3732 skb2
= skb_copy_expand(skb1
,
3736 if (unlikely(skb2
== NULL
))
3740 skb_set_owner_w(skb2
, skb1
->sk
);
3742 /* Looking around. Are we still alive?
3743 * OK, link new skb, drop old one */
3745 skb2
->next
= skb1
->next
;
3752 skb_p
= &skb1
->next
;
3757 EXPORT_SYMBOL_GPL(skb_cow_data
);
3759 static void sock_rmem_free(struct sk_buff
*skb
)
3761 struct sock
*sk
= skb
->sk
;
3763 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
3766 static void skb_set_err_queue(struct sk_buff
*skb
)
3768 /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
3769 * So, it is safe to (mis)use it to mark skbs on the error queue.
3771 skb
->pkt_type
= PACKET_OUTGOING
;
3772 BUILD_BUG_ON(PACKET_OUTGOING
== 0);
3776 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3778 int sock_queue_err_skb(struct sock
*sk
, struct sk_buff
*skb
)
3780 if (atomic_read(&sk
->sk_rmem_alloc
) + skb
->truesize
>=
3781 (unsigned int)sk
->sk_rcvbuf
)
3786 skb
->destructor
= sock_rmem_free
;
3787 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
3788 skb_set_err_queue(skb
);
3790 /* before exiting rcu section, make sure dst is refcounted */
3793 skb_queue_tail(&sk
->sk_error_queue
, skb
);
3794 if (!sock_flag(sk
, SOCK_DEAD
))
3795 sk
->sk_data_ready(sk
);
3798 EXPORT_SYMBOL(sock_queue_err_skb
);
3800 static bool is_icmp_err_skb(const struct sk_buff
*skb
)
3802 return skb
&& (SKB_EXT_ERR(skb
)->ee
.ee_origin
== SO_EE_ORIGIN_ICMP
||
3803 SKB_EXT_ERR(skb
)->ee
.ee_origin
== SO_EE_ORIGIN_ICMP6
);
3806 struct sk_buff
*sock_dequeue_err_skb(struct sock
*sk
)
3808 struct sk_buff_head
*q
= &sk
->sk_error_queue
;
3809 struct sk_buff
*skb
, *skb_next
= NULL
;
3810 bool icmp_next
= false;
3811 unsigned long flags
;
3813 spin_lock_irqsave(&q
->lock
, flags
);
3814 skb
= __skb_dequeue(q
);
3815 if (skb
&& (skb_next
= skb_peek(q
))) {
3816 icmp_next
= is_icmp_err_skb(skb_next
);
3818 sk
->sk_err
= SKB_EXT_ERR(skb_next
)->ee
.ee_origin
;
3820 spin_unlock_irqrestore(&q
->lock
, flags
);
3822 if (is_icmp_err_skb(skb
) && !icmp_next
)
3826 sk
->sk_error_report(sk
);
3830 EXPORT_SYMBOL(sock_dequeue_err_skb
);
3833 * skb_clone_sk - create clone of skb, and take reference to socket
3834 * @skb: the skb to clone
3836 * This function creates a clone of a buffer that holds a reference on
3837 * sk_refcnt. Buffers created via this function are meant to be
3838 * returned using sock_queue_err_skb, or free via kfree_skb.
3840 * When passing buffers allocated with this function to sock_queue_err_skb
3841 * it is necessary to wrap the call with sock_hold/sock_put in order to
3842 * prevent the socket from being released prior to being enqueued on
3843 * the sk_error_queue.
3845 struct sk_buff
*skb_clone_sk(struct sk_buff
*skb
)
3847 struct sock
*sk
= skb
->sk
;
3848 struct sk_buff
*clone
;
3850 if (!sk
|| !refcount_inc_not_zero(&sk
->sk_refcnt
))
3853 clone
= skb_clone(skb
, GFP_ATOMIC
);
3860 clone
->destructor
= sock_efree
;
3864 EXPORT_SYMBOL(skb_clone_sk
);
3866 static void __skb_complete_tx_timestamp(struct sk_buff
*skb
,
3871 struct sock_exterr_skb
*serr
;
3874 BUILD_BUG_ON(sizeof(struct sock_exterr_skb
) > sizeof(skb
->cb
));
3876 serr
= SKB_EXT_ERR(skb
);
3877 memset(serr
, 0, sizeof(*serr
));
3878 serr
->ee
.ee_errno
= ENOMSG
;
3879 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TIMESTAMPING
;
3880 serr
->ee
.ee_info
= tstype
;
3881 serr
->opt_stats
= opt_stats
;
3882 serr
->header
.h4
.iif
= skb
->dev
? skb
->dev
->ifindex
: 0;
3883 if (sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_ID
) {
3884 serr
->ee
.ee_data
= skb_shinfo(skb
)->tskey
;
3885 if (sk
->sk_protocol
== IPPROTO_TCP
&&
3886 sk
->sk_type
== SOCK_STREAM
)
3887 serr
->ee
.ee_data
-= sk
->sk_tskey
;
3890 err
= sock_queue_err_skb(sk
, skb
);
3896 static bool skb_may_tx_timestamp(struct sock
*sk
, bool tsonly
)
3900 if (likely(sysctl_tstamp_allow_data
|| tsonly
))
3903 read_lock_bh(&sk
->sk_callback_lock
);
3904 ret
= sk
->sk_socket
&& sk
->sk_socket
->file
&&
3905 file_ns_capable(sk
->sk_socket
->file
, &init_user_ns
, CAP_NET_RAW
);
3906 read_unlock_bh(&sk
->sk_callback_lock
);
3910 void skb_complete_tx_timestamp(struct sk_buff
*skb
,
3911 struct skb_shared_hwtstamps
*hwtstamps
)
3913 struct sock
*sk
= skb
->sk
;
3915 if (!skb_may_tx_timestamp(sk
, false))
3918 /* Take a reference to prevent skb_orphan() from freeing the socket,
3919 * but only if the socket refcount is not zero.
3921 if (likely(refcount_inc_not_zero(&sk
->sk_refcnt
))) {
3922 *skb_hwtstamps(skb
) = *hwtstamps
;
3923 __skb_complete_tx_timestamp(skb
, sk
, SCM_TSTAMP_SND
, false);
3927 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp
);
3929 void __skb_tstamp_tx(struct sk_buff
*orig_skb
,
3930 struct skb_shared_hwtstamps
*hwtstamps
,
3931 struct sock
*sk
, int tstype
)
3933 struct sk_buff
*skb
;
3934 bool tsonly
, opt_stats
= false;
3939 if (!hwtstamps
&& !(sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_TX_SWHW
) &&
3940 skb_shinfo(orig_skb
)->tx_flags
& SKBTX_IN_PROGRESS
)
3943 tsonly
= sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_TSONLY
;
3944 if (!skb_may_tx_timestamp(sk
, tsonly
))
3949 if ((sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_STATS
) &&
3950 sk
->sk_protocol
== IPPROTO_TCP
&&
3951 sk
->sk_type
== SOCK_STREAM
) {
3952 skb
= tcp_get_timestamping_opt_stats(sk
);
3956 skb
= alloc_skb(0, GFP_ATOMIC
);
3958 skb
= skb_clone(orig_skb
, GFP_ATOMIC
);
3964 skb_shinfo(skb
)->tx_flags
|= skb_shinfo(orig_skb
)->tx_flags
&
3966 skb_shinfo(skb
)->tskey
= skb_shinfo(orig_skb
)->tskey
;
3970 *skb_hwtstamps(skb
) = *hwtstamps
;
3972 skb
->tstamp
= ktime_get_real();
3974 __skb_complete_tx_timestamp(skb
, sk
, tstype
, opt_stats
);
3976 EXPORT_SYMBOL_GPL(__skb_tstamp_tx
);
3978 void skb_tstamp_tx(struct sk_buff
*orig_skb
,
3979 struct skb_shared_hwtstamps
*hwtstamps
)
3981 return __skb_tstamp_tx(orig_skb
, hwtstamps
, orig_skb
->sk
,
3984 EXPORT_SYMBOL_GPL(skb_tstamp_tx
);
3986 void skb_complete_wifi_ack(struct sk_buff
*skb
, bool acked
)
3988 struct sock
*sk
= skb
->sk
;
3989 struct sock_exterr_skb
*serr
;
3992 skb
->wifi_acked_valid
= 1;
3993 skb
->wifi_acked
= acked
;
3995 serr
= SKB_EXT_ERR(skb
);
3996 memset(serr
, 0, sizeof(*serr
));
3997 serr
->ee
.ee_errno
= ENOMSG
;
3998 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TXSTATUS
;
4000 /* Take a reference to prevent skb_orphan() from freeing the socket,
4001 * but only if the socket refcount is not zero.
4003 if (likely(refcount_inc_not_zero(&sk
->sk_refcnt
))) {
4004 err
= sock_queue_err_skb(sk
, skb
);
4010 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack
);
4013 * skb_partial_csum_set - set up and verify partial csum values for packet
4014 * @skb: the skb to set
4015 * @start: the number of bytes after skb->data to start checksumming.
4016 * @off: the offset from start to place the checksum.
4018 * For untrusted partially-checksummed packets, we need to make sure the values
4019 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
4021 * This function checks and sets those values and skb->ip_summed: if this
4022 * returns false you should drop the packet.
4024 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
4026 if (unlikely(start
> skb_headlen(skb
)) ||
4027 unlikely((int)start
+ off
> skb_headlen(skb
) - 2)) {
4028 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
4029 start
, off
, skb_headlen(skb
));
4032 skb
->ip_summed
= CHECKSUM_PARTIAL
;
4033 skb
->csum_start
= skb_headroom(skb
) + start
;
4034 skb
->csum_offset
= off
;
4035 skb_set_transport_header(skb
, start
);
4038 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);
4040 static int skb_maybe_pull_tail(struct sk_buff
*skb
, unsigned int len
,
4043 if (skb_headlen(skb
) >= len
)
4046 /* If we need to pullup then pullup to the max, so we
4047 * won't need to do it again.
4052 if (__pskb_pull_tail(skb
, max
- skb_headlen(skb
)) == NULL
)
4055 if (skb_headlen(skb
) < len
)
4061 #define MAX_TCP_HDR_LEN (15 * 4)
4063 static __sum16
*skb_checksum_setup_ip(struct sk_buff
*skb
,
4064 typeof(IPPROTO_IP
) proto
,
4071 err
= skb_maybe_pull_tail(skb
, off
+ sizeof(struct tcphdr
),
4072 off
+ MAX_TCP_HDR_LEN
);
4073 if (!err
&& !skb_partial_csum_set(skb
, off
,
4074 offsetof(struct tcphdr
,
4077 return err
? ERR_PTR(err
) : &tcp_hdr(skb
)->check
;
4080 err
= skb_maybe_pull_tail(skb
, off
+ sizeof(struct udphdr
),
4081 off
+ sizeof(struct udphdr
));
4082 if (!err
&& !skb_partial_csum_set(skb
, off
,
4083 offsetof(struct udphdr
,
4086 return err
? ERR_PTR(err
) : &udp_hdr(skb
)->check
;
4089 return ERR_PTR(-EPROTO
);
4092 /* This value should be large enough to cover a tagged ethernet header plus
4093 * maximally sized IP and TCP or UDP headers.
4095 #define MAX_IP_HDR_LEN 128
4097 static int skb_checksum_setup_ipv4(struct sk_buff
*skb
, bool recalculate
)
4106 err
= skb_maybe_pull_tail(skb
,
4107 sizeof(struct iphdr
),
4112 if (ip_hdr(skb
)->frag_off
& htons(IP_OFFSET
| IP_MF
))
4115 off
= ip_hdrlen(skb
);
4122 csum
= skb_checksum_setup_ip(skb
, ip_hdr(skb
)->protocol
, off
);
4124 return PTR_ERR(csum
);
4127 *csum
= ~csum_tcpudp_magic(ip_hdr(skb
)->saddr
,
4130 ip_hdr(skb
)->protocol
, 0);
4137 /* This value should be large enough to cover a tagged ethernet header plus
4138 * an IPv6 header, all options, and a maximal TCP or UDP header.
4140 #define MAX_IPV6_HDR_LEN 256
4142 #define OPT_HDR(type, skb, off) \
4143 (type *)(skb_network_header(skb) + (off))
4145 static int skb_checksum_setup_ipv6(struct sk_buff
*skb
, bool recalculate
)
4158 off
= sizeof(struct ipv6hdr
);
4160 err
= skb_maybe_pull_tail(skb
, off
, MAX_IPV6_HDR_LEN
);
4164 nexthdr
= ipv6_hdr(skb
)->nexthdr
;
4166 len
= sizeof(struct ipv6hdr
) + ntohs(ipv6_hdr(skb
)->payload_len
);
4167 while (off
<= len
&& !done
) {
4169 case IPPROTO_DSTOPTS
:
4170 case IPPROTO_HOPOPTS
:
4171 case IPPROTO_ROUTING
: {
4172 struct ipv6_opt_hdr
*hp
;
4174 err
= skb_maybe_pull_tail(skb
,
4176 sizeof(struct ipv6_opt_hdr
),
4181 hp
= OPT_HDR(struct ipv6_opt_hdr
, skb
, off
);
4182 nexthdr
= hp
->nexthdr
;
4183 off
+= ipv6_optlen(hp
);
4187 struct ip_auth_hdr
*hp
;
4189 err
= skb_maybe_pull_tail(skb
,
4191 sizeof(struct ip_auth_hdr
),
4196 hp
= OPT_HDR(struct ip_auth_hdr
, skb
, off
);
4197 nexthdr
= hp
->nexthdr
;
4198 off
+= ipv6_authlen(hp
);
4201 case IPPROTO_FRAGMENT
: {
4202 struct frag_hdr
*hp
;
4204 err
= skb_maybe_pull_tail(skb
,
4206 sizeof(struct frag_hdr
),
4211 hp
= OPT_HDR(struct frag_hdr
, skb
, off
);
4213 if (hp
->frag_off
& htons(IP6_OFFSET
| IP6_MF
))
4216 nexthdr
= hp
->nexthdr
;
4217 off
+= sizeof(struct frag_hdr
);
4228 if (!done
|| fragment
)
4231 csum
= skb_checksum_setup_ip(skb
, nexthdr
, off
);
4233 return PTR_ERR(csum
);
4236 *csum
= ~csum_ipv6_magic(&ipv6_hdr(skb
)->saddr
,
4237 &ipv6_hdr(skb
)->daddr
,
4238 skb
->len
- off
, nexthdr
, 0);
4246 * skb_checksum_setup - set up partial checksum offset
4247 * @skb: the skb to set up
4248 * @recalculate: if true the pseudo-header checksum will be recalculated
4250 int skb_checksum_setup(struct sk_buff
*skb
, bool recalculate
)
4254 switch (skb
->protocol
) {
4255 case htons(ETH_P_IP
):
4256 err
= skb_checksum_setup_ipv4(skb
, recalculate
);
4259 case htons(ETH_P_IPV6
):
4260 err
= skb_checksum_setup_ipv6(skb
, recalculate
);
4270 EXPORT_SYMBOL(skb_checksum_setup
);
4273 * skb_checksum_maybe_trim - maybe trims the given skb
4274 * @skb: the skb to check
4275 * @transport_len: the data length beyond the network header
4277 * Checks whether the given skb has data beyond the given transport length.
4278 * If so, returns a cloned skb trimmed to this transport length.
4279 * Otherwise returns the provided skb. Returns NULL in error cases
4280 * (e.g. transport_len exceeds skb length or out-of-memory).
4282 * Caller needs to set the skb transport header and free any returned skb if it
4283 * differs from the provided skb.
4285 static struct sk_buff
*skb_checksum_maybe_trim(struct sk_buff
*skb
,
4286 unsigned int transport_len
)
4288 struct sk_buff
*skb_chk
;
4289 unsigned int len
= skb_transport_offset(skb
) + transport_len
;
4294 else if (skb
->len
== len
)
4297 skb_chk
= skb_clone(skb
, GFP_ATOMIC
);
4301 ret
= pskb_trim_rcsum(skb_chk
, len
);
4311 * skb_checksum_trimmed - validate checksum of an skb
4312 * @skb: the skb to check
4313 * @transport_len: the data length beyond the network header
4314 * @skb_chkf: checksum function to use
4316 * Applies the given checksum function skb_chkf to the provided skb.
4317 * Returns a checked and maybe trimmed skb. Returns NULL on error.
4319 * If the skb has data beyond the given transport length, then a
4320 * trimmed & cloned skb is checked and returned.
4322 * Caller needs to set the skb transport header and free any returned skb if it
4323 * differs from the provided skb.
4325 struct sk_buff
*skb_checksum_trimmed(struct sk_buff
*skb
,
4326 unsigned int transport_len
,
4327 __sum16(*skb_chkf
)(struct sk_buff
*skb
))
4329 struct sk_buff
*skb_chk
;
4330 unsigned int offset
= skb_transport_offset(skb
);
4333 skb_chk
= skb_checksum_maybe_trim(skb
, transport_len
);
4337 if (!pskb_may_pull(skb_chk
, offset
))
4340 skb_pull_rcsum(skb_chk
, offset
);
4341 ret
= skb_chkf(skb_chk
);
4342 skb_push_rcsum(skb_chk
, offset
);
4350 if (skb_chk
&& skb_chk
!= skb
)
4356 EXPORT_SYMBOL(skb_checksum_trimmed
);
4358 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
4360 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
4363 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);
4365 void kfree_skb_partial(struct sk_buff
*skb
, bool head_stolen
)
4368 skb_release_head_state(skb
);
4369 kmem_cache_free(skbuff_head_cache
, skb
);
4374 EXPORT_SYMBOL(kfree_skb_partial
);
4377 * skb_try_coalesce - try to merge skb to prior one
4379 * @from: buffer to add
4380 * @fragstolen: pointer to boolean
4381 * @delta_truesize: how much more was allocated than was requested
4383 bool skb_try_coalesce(struct sk_buff
*to
, struct sk_buff
*from
,
4384 bool *fragstolen
, int *delta_truesize
)
4386 int i
, delta
, len
= from
->len
;
4388 *fragstolen
= false;
4393 if (len
<= skb_tailroom(to
)) {
4395 BUG_ON(skb_copy_bits(from
, 0, skb_put(to
, len
), len
));
4396 *delta_truesize
= 0;
4400 if (skb_has_frag_list(to
) || skb_has_frag_list(from
))
4403 if (skb_headlen(from
) != 0) {
4405 unsigned int offset
;
4407 if (skb_shinfo(to
)->nr_frags
+
4408 skb_shinfo(from
)->nr_frags
>= MAX_SKB_FRAGS
)
4411 if (skb_head_is_locked(from
))
4414 delta
= from
->truesize
- SKB_DATA_ALIGN(sizeof(struct sk_buff
));
4416 page
= virt_to_head_page(from
->head
);
4417 offset
= from
->data
- (unsigned char *)page_address(page
);
4419 skb_fill_page_desc(to
, skb_shinfo(to
)->nr_frags
,
4420 page
, offset
, skb_headlen(from
));
4423 if (skb_shinfo(to
)->nr_frags
+
4424 skb_shinfo(from
)->nr_frags
> MAX_SKB_FRAGS
)
4427 delta
= from
->truesize
- SKB_TRUESIZE(skb_end_offset(from
));
4430 WARN_ON_ONCE(delta
< len
);
4432 memcpy(skb_shinfo(to
)->frags
+ skb_shinfo(to
)->nr_frags
,
4433 skb_shinfo(from
)->frags
,
4434 skb_shinfo(from
)->nr_frags
* sizeof(skb_frag_t
));
4435 skb_shinfo(to
)->nr_frags
+= skb_shinfo(from
)->nr_frags
;
4437 if (!skb_cloned(from
))
4438 skb_shinfo(from
)->nr_frags
= 0;
4440 /* if the skb is not cloned this does nothing
4441 * since we set nr_frags to 0.
4443 for (i
= 0; i
< skb_shinfo(from
)->nr_frags
; i
++)
4444 skb_frag_ref(from
, i
);
4446 to
->truesize
+= delta
;
4448 to
->data_len
+= len
;
4450 *delta_truesize
= delta
;
4453 EXPORT_SYMBOL(skb_try_coalesce
);
4456 * skb_scrub_packet - scrub an skb
4458 * @skb: buffer to clean
4459 * @xnet: packet is crossing netns
4461 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
4462 * into/from a tunnel. Some information have to be cleared during these
4464 * skb_scrub_packet can also be used to clean a skb before injecting it in
4465 * another namespace (@xnet == true). We have to clear all information in the
4466 * skb that could impact namespace isolation.
4468 void skb_scrub_packet(struct sk_buff
*skb
, bool xnet
)
4471 skb
->pkt_type
= PACKET_HOST
;
4477 nf_reset_trace(skb
);
4485 EXPORT_SYMBOL_GPL(skb_scrub_packet
);
4488 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
4492 * skb_gso_transport_seglen is used to determine the real size of the
4493 * individual segments, including Layer4 headers (TCP/UDP).
4495 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
4497 unsigned int skb_gso_transport_seglen(const struct sk_buff
*skb
)
4499 const struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
4500 unsigned int thlen
= 0;
4502 if (skb
->encapsulation
) {
4503 thlen
= skb_inner_transport_header(skb
) -
4504 skb_transport_header(skb
);
4506 if (likely(shinfo
->gso_type
& (SKB_GSO_TCPV4
| SKB_GSO_TCPV6
)))
4507 thlen
+= inner_tcp_hdrlen(skb
);
4508 } else if (likely(shinfo
->gso_type
& (SKB_GSO_TCPV4
| SKB_GSO_TCPV6
))) {
4509 thlen
= tcp_hdrlen(skb
);
4510 } else if (unlikely(shinfo
->gso_type
& SKB_GSO_SCTP
)) {
4511 thlen
= sizeof(struct sctphdr
);
4513 /* UFO sets gso_size to the size of the fragmentation
4514 * payload, i.e. the size of the L4 (UDP) header is already
4517 return thlen
+ shinfo
->gso_size
;
4519 EXPORT_SYMBOL_GPL(skb_gso_transport_seglen
);
4522 * skb_gso_validate_mtu - Return in case such skb fits a given MTU
4525 * @mtu: MTU to validate against
4527 * skb_gso_validate_mtu validates if a given skb will fit a wanted MTU
4530 bool skb_gso_validate_mtu(const struct sk_buff
*skb
, unsigned int mtu
)
4532 const struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
4533 const struct sk_buff
*iter
;
4536 hlen
= skb_gso_network_seglen(skb
);
4538 if (shinfo
->gso_size
!= GSO_BY_FRAGS
)
4541 /* Undo this so we can re-use header sizes */
4542 hlen
-= GSO_BY_FRAGS
;
4544 skb_walk_frags(skb
, iter
) {
4545 if (hlen
+ skb_headlen(iter
) > mtu
)
4551 EXPORT_SYMBOL_GPL(skb_gso_validate_mtu
);
4553 static struct sk_buff
*skb_reorder_vlan_header(struct sk_buff
*skb
)
4555 if (skb_cow(skb
, skb_headroom(skb
)) < 0) {
4560 memmove(skb
->data
- ETH_HLEN
, skb
->data
- skb
->mac_len
- VLAN_HLEN
,
4562 skb
->mac_header
+= VLAN_HLEN
;
4566 struct sk_buff
*skb_vlan_untag(struct sk_buff
*skb
)
4568 struct vlan_hdr
*vhdr
;
4571 if (unlikely(skb_vlan_tag_present(skb
))) {
4572 /* vlan_tci is already set-up so leave this for another time */
4576 skb
= skb_share_check(skb
, GFP_ATOMIC
);
4580 if (unlikely(!pskb_may_pull(skb
, VLAN_HLEN
)))
4583 vhdr
= (struct vlan_hdr
*)skb
->data
;
4584 vlan_tci
= ntohs(vhdr
->h_vlan_TCI
);
4585 __vlan_hwaccel_put_tag(skb
, skb
->protocol
, vlan_tci
);
4587 skb_pull_rcsum(skb
, VLAN_HLEN
);
4588 vlan_set_encap_proto(skb
, vhdr
);
4590 skb
= skb_reorder_vlan_header(skb
);
4594 skb_reset_network_header(skb
);
4595 skb_reset_transport_header(skb
);
4596 skb_reset_mac_len(skb
);
4604 EXPORT_SYMBOL(skb_vlan_untag
);
4606 int skb_ensure_writable(struct sk_buff
*skb
, int write_len
)
4608 if (!pskb_may_pull(skb
, write_len
))
4611 if (!skb_cloned(skb
) || skb_clone_writable(skb
, write_len
))
4614 return pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
4616 EXPORT_SYMBOL(skb_ensure_writable
);
4618 /* remove VLAN header from packet and update csum accordingly.
4619 * expects a non skb_vlan_tag_present skb with a vlan tag payload
4621 int __skb_vlan_pop(struct sk_buff
*skb
, u16
*vlan_tci
)
4623 struct vlan_hdr
*vhdr
;
4624 int offset
= skb
->data
- skb_mac_header(skb
);
4627 if (WARN_ONCE(offset
,
4628 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
4633 err
= skb_ensure_writable(skb
, VLAN_ETH_HLEN
);
4637 skb_postpull_rcsum(skb
, skb
->data
+ (2 * ETH_ALEN
), VLAN_HLEN
);
4639 vhdr
= (struct vlan_hdr
*)(skb
->data
+ ETH_HLEN
);
4640 *vlan_tci
= ntohs(vhdr
->h_vlan_TCI
);
4642 memmove(skb
->data
+ VLAN_HLEN
, skb
->data
, 2 * ETH_ALEN
);
4643 __skb_pull(skb
, VLAN_HLEN
);
4645 vlan_set_encap_proto(skb
, vhdr
);
4646 skb
->mac_header
+= VLAN_HLEN
;
4648 if (skb_network_offset(skb
) < ETH_HLEN
)
4649 skb_set_network_header(skb
, ETH_HLEN
);
4651 skb_reset_mac_len(skb
);
4655 EXPORT_SYMBOL(__skb_vlan_pop
);
4657 /* Pop a vlan tag either from hwaccel or from payload.
4658 * Expects skb->data at mac header.
4660 int skb_vlan_pop(struct sk_buff
*skb
)
4666 if (likely(skb_vlan_tag_present(skb
))) {
4669 if (unlikely(!eth_type_vlan(skb
->protocol
)))
4672 err
= __skb_vlan_pop(skb
, &vlan_tci
);
4676 /* move next vlan tag to hw accel tag */
4677 if (likely(!eth_type_vlan(skb
->protocol
)))
4680 vlan_proto
= skb
->protocol
;
4681 err
= __skb_vlan_pop(skb
, &vlan_tci
);
4685 __vlan_hwaccel_put_tag(skb
, vlan_proto
, vlan_tci
);
4688 EXPORT_SYMBOL(skb_vlan_pop
);
4690 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
4691 * Expects skb->data at mac header.
4693 int skb_vlan_push(struct sk_buff
*skb
, __be16 vlan_proto
, u16 vlan_tci
)
4695 if (skb_vlan_tag_present(skb
)) {
4696 int offset
= skb
->data
- skb_mac_header(skb
);
4699 if (WARN_ONCE(offset
,
4700 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
4705 err
= __vlan_insert_tag(skb
, skb
->vlan_proto
,
4706 skb_vlan_tag_get(skb
));
4710 skb
->protocol
= skb
->vlan_proto
;
4711 skb
->mac_len
+= VLAN_HLEN
;
4713 skb_postpush_rcsum(skb
, skb
->data
+ (2 * ETH_ALEN
), VLAN_HLEN
);
4715 __vlan_hwaccel_put_tag(skb
, vlan_proto
, vlan_tci
);
4718 EXPORT_SYMBOL(skb_vlan_push
);
4721 * alloc_skb_with_frags - allocate skb with page frags
4723 * @header_len: size of linear part
4724 * @data_len: needed length in frags
4725 * @max_page_order: max page order desired.
4726 * @errcode: pointer to error code if any
4727 * @gfp_mask: allocation mask
4729 * This can be used to allocate a paged skb, given a maximal order for frags.
4731 struct sk_buff
*alloc_skb_with_frags(unsigned long header_len
,
4732 unsigned long data_len
,
4737 int npages
= (data_len
+ (PAGE_SIZE
- 1)) >> PAGE_SHIFT
;
4738 unsigned long chunk
;
4739 struct sk_buff
*skb
;
4744 *errcode
= -EMSGSIZE
;
4745 /* Note this test could be relaxed, if we succeed to allocate
4746 * high order pages...
4748 if (npages
> MAX_SKB_FRAGS
)
4751 gfp_head
= gfp_mask
;
4752 if (gfp_head
& __GFP_DIRECT_RECLAIM
)
4753 gfp_head
|= __GFP_RETRY_MAYFAIL
;
4755 *errcode
= -ENOBUFS
;
4756 skb
= alloc_skb(header_len
, gfp_head
);
4760 skb
->truesize
+= npages
<< PAGE_SHIFT
;
4762 for (i
= 0; npages
> 0; i
++) {
4763 int order
= max_page_order
;
4766 if (npages
>= 1 << order
) {
4767 page
= alloc_pages((gfp_mask
& ~__GFP_DIRECT_RECLAIM
) |
4774 /* Do not retry other high order allocations */
4780 page
= alloc_page(gfp_mask
);
4784 chunk
= min_t(unsigned long, data_len
,
4785 PAGE_SIZE
<< order
);
4786 skb_fill_page_desc(skb
, i
, page
, 0, chunk
);
4788 npages
-= 1 << order
;
4796 EXPORT_SYMBOL(alloc_skb_with_frags
);
4798 /* carve out the first off bytes from skb when off < headlen */
4799 static int pskb_carve_inside_header(struct sk_buff
*skb
, const u32 off
,
4800 const int headlen
, gfp_t gfp_mask
)
4803 int size
= skb_end_offset(skb
);
4804 int new_hlen
= headlen
- off
;
4807 size
= SKB_DATA_ALIGN(size
);
4809 if (skb_pfmemalloc(skb
))
4810 gfp_mask
|= __GFP_MEMALLOC
;
4811 data
= kmalloc_reserve(size
+
4812 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)),
4813 gfp_mask
, NUMA_NO_NODE
, NULL
);
4817 size
= SKB_WITH_OVERHEAD(ksize(data
));
4819 /* Copy real data, and all frags */
4820 skb_copy_from_linear_data_offset(skb
, off
, data
, new_hlen
);
4823 memcpy((struct skb_shared_info
*)(data
+ size
),
4825 offsetof(struct skb_shared_info
,
4826 frags
[skb_shinfo(skb
)->nr_frags
]));
4827 if (skb_cloned(skb
)) {
4828 /* drop the old head gracefully */
4829 if (skb_orphan_frags(skb
, gfp_mask
)) {
4833 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
4834 skb_frag_ref(skb
, i
);
4835 if (skb_has_frag_list(skb
))
4836 skb_clone_fraglist(skb
);
4837 skb_release_data(skb
);
4839 /* we can reuse existing recount- all we did was
4848 #ifdef NET_SKBUFF_DATA_USES_OFFSET
4851 skb
->end
= skb
->head
+ size
;
4853 skb_set_tail_pointer(skb
, skb_headlen(skb
));
4854 skb_headers_offset_update(skb
, 0);
4858 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
4863 static int pskb_carve(struct sk_buff
*skb
, const u32 off
, gfp_t gfp
);
4865 /* carve out the first eat bytes from skb's frag_list. May recurse into
4868 static int pskb_carve_frag_list(struct sk_buff
*skb
,
4869 struct skb_shared_info
*shinfo
, int eat
,
4872 struct sk_buff
*list
= shinfo
->frag_list
;
4873 struct sk_buff
*clone
= NULL
;
4874 struct sk_buff
*insp
= NULL
;
4878 pr_err("Not enough bytes to eat. Want %d\n", eat
);
4881 if (list
->len
<= eat
) {
4882 /* Eaten as whole. */
4887 /* Eaten partially. */
4888 if (skb_shared(list
)) {
4889 clone
= skb_clone(list
, gfp_mask
);
4895 /* This may be pulled without problems. */
4898 if (pskb_carve(list
, eat
, gfp_mask
) < 0) {
4906 /* Free pulled out fragments. */
4907 while ((list
= shinfo
->frag_list
) != insp
) {
4908 shinfo
->frag_list
= list
->next
;
4911 /* And insert new clone at head. */
4914 shinfo
->frag_list
= clone
;
4919 /* carve off first len bytes from skb. Split line (off) is in the
4920 * non-linear part of skb
4922 static int pskb_carve_inside_nonlinear(struct sk_buff
*skb
, const u32 off
,
4923 int pos
, gfp_t gfp_mask
)
4926 int size
= skb_end_offset(skb
);
4928 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
4929 struct skb_shared_info
*shinfo
;
4931 size
= SKB_DATA_ALIGN(size
);
4933 if (skb_pfmemalloc(skb
))
4934 gfp_mask
|= __GFP_MEMALLOC
;
4935 data
= kmalloc_reserve(size
+
4936 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)),
4937 gfp_mask
, NUMA_NO_NODE
, NULL
);
4941 size
= SKB_WITH_OVERHEAD(ksize(data
));
4943 memcpy((struct skb_shared_info
*)(data
+ size
),
4944 skb_shinfo(skb
), offsetof(struct skb_shared_info
,
4945 frags
[skb_shinfo(skb
)->nr_frags
]));
4946 if (skb_orphan_frags(skb
, gfp_mask
)) {
4950 shinfo
= (struct skb_shared_info
*)(data
+ size
);
4951 for (i
= 0; i
< nfrags
; i
++) {
4952 int fsize
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
4954 if (pos
+ fsize
> off
) {
4955 shinfo
->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
4959 * We have two variants in this case:
4960 * 1. Move all the frag to the second
4961 * part, if it is possible. F.e.
4962 * this approach is mandatory for TUX,
4963 * where splitting is expensive.
4964 * 2. Split is accurately. We make this.
4966 shinfo
->frags
[0].page_offset
+= off
- pos
;
4967 skb_frag_size_sub(&shinfo
->frags
[0], off
- pos
);
4969 skb_frag_ref(skb
, i
);
4974 shinfo
->nr_frags
= k
;
4975 if (skb_has_frag_list(skb
))
4976 skb_clone_fraglist(skb
);
4979 /* split line is in frag list */
4980 pskb_carve_frag_list(skb
, shinfo
, off
- pos
, gfp_mask
);
4982 skb_release_data(skb
);
4987 #ifdef NET_SKBUFF_DATA_USES_OFFSET
4990 skb
->end
= skb
->head
+ size
;
4992 skb_reset_tail_pointer(skb
);
4993 skb_headers_offset_update(skb
, 0);
4998 skb
->data_len
= skb
->len
;
4999 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
5003 /* remove len bytes from the beginning of the skb */
5004 static int pskb_carve(struct sk_buff
*skb
, const u32 len
, gfp_t gfp
)
5006 int headlen
= skb_headlen(skb
);
5009 return pskb_carve_inside_header(skb
, len
, headlen
, gfp
);
5011 return pskb_carve_inside_nonlinear(skb
, len
, headlen
, gfp
);
5014 /* Extract to_copy bytes starting at off from skb, and return this in
5017 struct sk_buff
*pskb_extract(struct sk_buff
*skb
, int off
,
5018 int to_copy
, gfp_t gfp
)
5020 struct sk_buff
*clone
= skb_clone(skb
, gfp
);
5025 if (pskb_carve(clone
, off
, gfp
) < 0 ||
5026 pskb_trim(clone
, to_copy
)) {
5032 EXPORT_SYMBOL(pskb_extract
);
5035 * skb_condense - try to get rid of fragments/frag_list if possible
5038 * Can be used to save memory before skb is added to a busy queue.
5039 * If packet has bytes in frags and enough tail room in skb->head,
5040 * pull all of them, so that we can free the frags right now and adjust
5043 * We do not reallocate skb->head thus can not fail.
5044 * Caller must re-evaluate skb->truesize if needed.
5046 void skb_condense(struct sk_buff
*skb
)
5048 if (skb
->data_len
) {
5049 if (skb
->data_len
> skb
->end
- skb
->tail
||
5053 /* Nice, we can free page frag(s) right now */
5054 __pskb_pull_tail(skb
, skb
->data_len
);
5056 /* At this point, skb->truesize might be over estimated,
5057 * because skb had a fragment, and fragments do not tell
5059 * When we pulled its content into skb->head, fragment
5060 * was freed, but __pskb_pull_tail() could not possibly
5061 * adjust skb->truesize, not knowing the frag truesize.
5063 skb
->truesize
= SKB_TRUESIZE(skb_end_offset(skb
));